Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * alloc.c | |
5 | * | |
6 | * Extent allocs and frees | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
60b11392 | 30 | #include <linux/swap.h> |
a90714c1 | 31 | #include <linux/quotaops.h> |
ccd979bd MF |
32 | |
33 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
34 | #include <cluster/masklog.h> | |
35 | ||
36 | #include "ocfs2.h" | |
37 | ||
38 | #include "alloc.h" | |
60b11392 | 39 | #include "aops.h" |
d6b32bbb | 40 | #include "blockcheck.h" |
ccd979bd MF |
41 | #include "dlmglue.h" |
42 | #include "extent_map.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
45 | #include "localalloc.h" | |
46 | #include "suballoc.h" | |
47 | #include "sysfile.h" | |
48 | #include "file.h" | |
49 | #include "super.h" | |
50 | #include "uptodate.h" | |
2a50a743 | 51 | #include "xattr.h" |
bcbbb24a | 52 | #include "refcounttree.h" |
ccd979bd MF |
53 | |
54 | #include "buffer_head_io.h" | |
55 | ||
853a3a14 TM |
56 | enum ocfs2_contig_type { |
57 | CONTIG_NONE = 0, | |
58 | CONTIG_LEFT, | |
59 | CONTIG_RIGHT, | |
60 | CONTIG_LEFTRIGHT, | |
61 | }; | |
e7d4cb6b | 62 | |
853a3a14 TM |
63 | static enum ocfs2_contig_type |
64 | ocfs2_extent_rec_contig(struct super_block *sb, | |
65 | struct ocfs2_extent_rec *ext, | |
66 | struct ocfs2_extent_rec *insert_rec); | |
1625f8ac JB |
67 | /* |
68 | * Operations for a specific extent tree type. | |
69 | * | |
70 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
71 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 72 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
73 | * for the allocation portion of the extent tree. |
74 | */ | |
e7d4cb6b | 75 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
76 | /* |
77 | * last_eb_blk is the block number of the right most leaf extent | |
78 | * block. Most on-disk structures containing an extent tree store | |
79 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
80 | * ->eo_get_last_eb_blk() operations access this value. They are | |
81 | * both required. | |
82 | */ | |
35dc0aa3 JB |
83 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
84 | u64 blkno); | |
85 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
86 | |
87 | /* | |
88 | * The on-disk structure usually keeps track of how many total | |
89 | * clusters are stored in this extent tree. This function updates | |
90 | * that value. new_clusters is the delta, and must be | |
91 | * added to the total. Required. | |
92 | */ | |
6136ca5f | 93 | void (*eo_update_clusters)(struct ocfs2_extent_tree *et, |
35dc0aa3 | 94 | u32 new_clusters); |
1625f8ac | 95 | |
92ba470c JB |
96 | /* |
97 | * If this extent tree is supported by an extent map, insert | |
98 | * a record into the map. | |
99 | */ | |
100 | void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et, | |
101 | struct ocfs2_extent_rec *rec); | |
102 | ||
4c911eef JB |
103 | /* |
104 | * If this extent tree is supported by an extent map, truncate the | |
105 | * map to clusters, | |
106 | */ | |
107 | void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et, | |
108 | u32 clusters); | |
109 | ||
1625f8ac JB |
110 | /* |
111 | * If ->eo_insert_check() exists, it is called before rec is | |
112 | * inserted into the extent tree. It is optional. | |
113 | */ | |
6136ca5f | 114 | int (*eo_insert_check)(struct ocfs2_extent_tree *et, |
1e61ee79 | 115 | struct ocfs2_extent_rec *rec); |
6136ca5f | 116 | int (*eo_sanity_check)(struct ocfs2_extent_tree *et); |
0ce1010f | 117 | |
1625f8ac JB |
118 | /* |
119 | * -------------------------------------------------------------- | |
120 | * The remaining are internal to ocfs2_extent_tree and don't have | |
121 | * accessor functions | |
122 | */ | |
123 | ||
124 | /* | |
125 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
126 | * It is required. | |
127 | */ | |
0ce1010f | 128 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
129 | |
130 | /* | |
131 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
132 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
133 | * to 0 (unlimited). Optional. | |
134 | */ | |
6136ca5f | 135 | void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et); |
853a3a14 TM |
136 | |
137 | /* | |
138 | * ->eo_extent_contig test whether the 2 ocfs2_extent_rec | |
139 | * are contiguous or not. Optional. Don't need to set it if use | |
140 | * ocfs2_extent_rec as the tree leaf. | |
141 | */ | |
142 | enum ocfs2_contig_type | |
143 | (*eo_extent_contig)(struct ocfs2_extent_tree *et, | |
144 | struct ocfs2_extent_rec *ext, | |
145 | struct ocfs2_extent_rec *insert_rec); | |
e7d4cb6b TM |
146 | }; |
147 | ||
e7d4cb6b | 148 | |
f99b9b7c JB |
149 | /* |
150 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
151 | * in the methods. | |
152 | */ | |
153 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
154 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
155 | u64 blkno); | |
6136ca5f | 156 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
f99b9b7c | 157 | u32 clusters); |
92ba470c JB |
158 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
159 | struct ocfs2_extent_rec *rec); | |
4c911eef JB |
160 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
161 | u32 clusters); | |
6136ca5f | 162 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
f99b9b7c | 163 | struct ocfs2_extent_rec *rec); |
6136ca5f | 164 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et); |
f99b9b7c JB |
165 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); |
166 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
167 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, | |
168 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
169 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
92ba470c | 170 | .eo_extent_map_insert = ocfs2_dinode_extent_map_insert, |
4c911eef | 171 | .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate, |
f99b9b7c JB |
172 | .eo_insert_check = ocfs2_dinode_insert_check, |
173 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
174 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
175 | }; | |
0ce1010f | 176 | |
e7d4cb6b TM |
177 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
178 | u64 blkno) | |
179 | { | |
ea5efa15 | 180 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 181 | |
f99b9b7c | 182 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
183 | di->i_last_eb_blk = cpu_to_le64(blkno); |
184 | } | |
185 | ||
186 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
187 | { | |
ea5efa15 | 188 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 189 | |
f99b9b7c | 190 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
191 | return le64_to_cpu(di->i_last_eb_blk); |
192 | } | |
193 | ||
6136ca5f | 194 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
e7d4cb6b TM |
195 | u32 clusters) |
196 | { | |
6136ca5f | 197 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
ea5efa15 | 198 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
199 | |
200 | le32_add_cpu(&di->i_clusters, clusters); | |
6136ca5f JB |
201 | spin_lock(&oi->ip_lock); |
202 | oi->ip_clusters = le32_to_cpu(di->i_clusters); | |
203 | spin_unlock(&oi->ip_lock); | |
e7d4cb6b TM |
204 | } |
205 | ||
92ba470c JB |
206 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
207 | struct ocfs2_extent_rec *rec) | |
208 | { | |
209 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
210 | ||
211 | ocfs2_extent_map_insert_rec(inode, rec); | |
212 | } | |
213 | ||
4c911eef JB |
214 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
215 | u32 clusters) | |
216 | { | |
217 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
218 | ||
219 | ocfs2_extent_map_trunc(inode, clusters); | |
220 | } | |
221 | ||
6136ca5f | 222 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
223 | struct ocfs2_extent_rec *rec) |
224 | { | |
6136ca5f JB |
225 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
226 | struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb); | |
1e61ee79 | 227 | |
6136ca5f | 228 | BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
1e61ee79 | 229 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && |
6136ca5f | 230 | (oi->ip_clusters != le32_to_cpu(rec->e_cpos)), |
1e61ee79 JB |
231 | "Device %s, asking for sparse allocation: inode %llu, " |
232 | "cpos %u, clusters %u\n", | |
233 | osb->dev_str, | |
6136ca5f JB |
234 | (unsigned long long)oi->ip_blkno, |
235 | rec->e_cpos, oi->ip_clusters); | |
1e61ee79 JB |
236 | |
237 | return 0; | |
238 | } | |
239 | ||
6136ca5f | 240 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 241 | { |
10995aa2 | 242 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 243 | |
f99b9b7c | 244 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 245 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 246 | |
10995aa2 | 247 | return 0; |
e7d4cb6b TM |
248 | } |
249 | ||
f99b9b7c JB |
250 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
251 | { | |
252 | struct ocfs2_dinode *di = et->et_object; | |
253 | ||
254 | et->et_root_el = &di->id2.i_list; | |
255 | } | |
256 | ||
e7d4cb6b | 257 | |
0ce1010f JB |
258 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
259 | { | |
2a50a743 | 260 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
0ce1010f | 261 | |
2a50a743 | 262 | et->et_root_el = &vb->vb_xv->xr_list; |
0ce1010f JB |
263 | } |
264 | ||
f56654c4 TM |
265 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
266 | u64 blkno) | |
267 | { | |
2a50a743 | 268 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 269 | |
2a50a743 | 270 | vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno); |
f56654c4 TM |
271 | } |
272 | ||
273 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
274 | { | |
2a50a743 | 275 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 276 | |
2a50a743 | 277 | return le64_to_cpu(vb->vb_xv->xr_last_eb_blk); |
f56654c4 TM |
278 | } |
279 | ||
6136ca5f | 280 | static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et, |
f56654c4 TM |
281 | u32 clusters) |
282 | { | |
2a50a743 | 283 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 284 | |
2a50a743 | 285 | le32_add_cpu(&vb->vb_xv->xr_clusters, clusters); |
f56654c4 TM |
286 | } |
287 | ||
1a09f556 | 288 | static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
289 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
290 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
291 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 292 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
293 | }; |
294 | ||
0ce1010f JB |
295 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
296 | { | |
297 | struct ocfs2_xattr_block *xb = et->et_object; | |
298 | ||
299 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
300 | } | |
301 | ||
6136ca5f | 302 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et) |
943cced3 | 303 | { |
6136ca5f | 304 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
943cced3 | 305 | et->et_max_leaf_clusters = |
6136ca5f | 306 | ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE); |
943cced3 JB |
307 | } |
308 | ||
ba492615 TM |
309 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
310 | u64 blkno) | |
311 | { | |
ea5efa15 | 312 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
313 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
314 | ||
315 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
316 | } | |
317 | ||
318 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
319 | { | |
ea5efa15 | 320 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
321 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
322 | ||
323 | return le64_to_cpu(xt->xt_last_eb_blk); | |
324 | } | |
325 | ||
6136ca5f | 326 | static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et, |
ba492615 TM |
327 | u32 clusters) |
328 | { | |
ea5efa15 | 329 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
330 | |
331 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
332 | } | |
333 | ||
ba492615 | 334 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
335 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
336 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
337 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 338 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 339 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
340 | }; |
341 | ||
9b7895ef MF |
342 | static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et, |
343 | u64 blkno) | |
344 | { | |
345 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
346 | ||
347 | dx_root->dr_last_eb_blk = cpu_to_le64(blkno); | |
348 | } | |
349 | ||
350 | static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
351 | { | |
352 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
353 | ||
354 | return le64_to_cpu(dx_root->dr_last_eb_blk); | |
355 | } | |
356 | ||
6136ca5f | 357 | static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et, |
9b7895ef MF |
358 | u32 clusters) |
359 | { | |
360 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
361 | ||
362 | le32_add_cpu(&dx_root->dr_clusters, clusters); | |
363 | } | |
364 | ||
6136ca5f | 365 | static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et) |
9b7895ef MF |
366 | { |
367 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
368 | ||
369 | BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root)); | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et) | |
375 | { | |
376 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
377 | ||
378 | et->et_root_el = &dx_root->dr_list; | |
379 | } | |
380 | ||
381 | static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = { | |
382 | .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk, | |
383 | .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk, | |
384 | .eo_update_clusters = ocfs2_dx_root_update_clusters, | |
385 | .eo_sanity_check = ocfs2_dx_root_sanity_check, | |
386 | .eo_fill_root_el = ocfs2_dx_root_fill_root_el, | |
387 | }; | |
388 | ||
fe924415 TM |
389 | static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et) |
390 | { | |
391 | struct ocfs2_refcount_block *rb = et->et_object; | |
392 | ||
393 | et->et_root_el = &rb->rf_list; | |
394 | } | |
395 | ||
396 | static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
397 | u64 blkno) | |
398 | { | |
399 | struct ocfs2_refcount_block *rb = et->et_object; | |
400 | ||
401 | rb->rf_last_eb_blk = cpu_to_le64(blkno); | |
402 | } | |
403 | ||
404 | static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
405 | { | |
406 | struct ocfs2_refcount_block *rb = et->et_object; | |
407 | ||
408 | return le64_to_cpu(rb->rf_last_eb_blk); | |
409 | } | |
410 | ||
411 | static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et, | |
412 | u32 clusters) | |
413 | { | |
414 | struct ocfs2_refcount_block *rb = et->et_object; | |
415 | ||
416 | le32_add_cpu(&rb->rf_clusters, clusters); | |
417 | } | |
418 | ||
419 | static enum ocfs2_contig_type | |
420 | ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et, | |
421 | struct ocfs2_extent_rec *ext, | |
422 | struct ocfs2_extent_rec *insert_rec) | |
423 | { | |
424 | return CONTIG_NONE; | |
425 | } | |
426 | ||
427 | static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = { | |
428 | .eo_set_last_eb_blk = ocfs2_refcount_tree_set_last_eb_blk, | |
429 | .eo_get_last_eb_blk = ocfs2_refcount_tree_get_last_eb_blk, | |
430 | .eo_update_clusters = ocfs2_refcount_tree_update_clusters, | |
431 | .eo_fill_root_el = ocfs2_refcount_tree_fill_root_el, | |
432 | .eo_extent_contig = ocfs2_refcount_tree_extent_contig, | |
433 | }; | |
434 | ||
8d6220d6 | 435 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 436 | struct ocfs2_caching_info *ci, |
8d6220d6 | 437 | struct buffer_head *bh, |
13723d00 | 438 | ocfs2_journal_access_func access, |
8d6220d6 JB |
439 | void *obj, |
440 | struct ocfs2_extent_tree_operations *ops) | |
e7d4cb6b | 441 | { |
1a09f556 | 442 | et->et_ops = ops; |
ce1d9ea6 | 443 | et->et_root_bh = bh; |
5e404e9e | 444 | et->et_ci = ci; |
13723d00 | 445 | et->et_root_journal_access = access; |
ea5efa15 JB |
446 | if (!obj) |
447 | obj = (void *)bh->b_data; | |
448 | et->et_object = obj; | |
e7d4cb6b | 449 | |
0ce1010f | 450 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
451 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
452 | et->et_max_leaf_clusters = 0; | |
453 | else | |
6136ca5f | 454 | et->et_ops->eo_fill_max_leaf_clusters(et); |
e7d4cb6b TM |
455 | } |
456 | ||
8d6220d6 | 457 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 458 | struct ocfs2_caching_info *ci, |
8d6220d6 | 459 | struct buffer_head *bh) |
1a09f556 | 460 | { |
5e404e9e | 461 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di, |
13723d00 | 462 | NULL, &ocfs2_dinode_et_ops); |
1a09f556 JB |
463 | } |
464 | ||
8d6220d6 | 465 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 466 | struct ocfs2_caching_info *ci, |
8d6220d6 | 467 | struct buffer_head *bh) |
1a09f556 | 468 | { |
5e404e9e | 469 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb, |
13723d00 | 470 | NULL, &ocfs2_xattr_tree_et_ops); |
1a09f556 JB |
471 | } |
472 | ||
8d6220d6 | 473 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 474 | struct ocfs2_caching_info *ci, |
2a50a743 | 475 | struct ocfs2_xattr_value_buf *vb) |
e7d4cb6b | 476 | { |
5e404e9e | 477 | __ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb, |
8d6220d6 | 478 | &ocfs2_xattr_value_et_ops); |
e7d4cb6b TM |
479 | } |
480 | ||
9b7895ef | 481 | void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 482 | struct ocfs2_caching_info *ci, |
9b7895ef MF |
483 | struct buffer_head *bh) |
484 | { | |
5e404e9e | 485 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr, |
9b7895ef MF |
486 | NULL, &ocfs2_dx_root_et_ops); |
487 | } | |
488 | ||
fe924415 TM |
489 | void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et, |
490 | struct ocfs2_caching_info *ci, | |
491 | struct buffer_head *bh) | |
492 | { | |
493 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb, | |
494 | NULL, &ocfs2_refcount_tree_et_ops); | |
495 | } | |
496 | ||
35dc0aa3 JB |
497 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
498 | u64 new_last_eb_blk) | |
e7d4cb6b | 499 | { |
ce1d9ea6 | 500 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
501 | } |
502 | ||
35dc0aa3 | 503 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 504 | { |
ce1d9ea6 | 505 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
506 | } |
507 | ||
6136ca5f | 508 | static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et, |
35dc0aa3 JB |
509 | u32 clusters) |
510 | { | |
6136ca5f | 511 | et->et_ops->eo_update_clusters(et, clusters); |
35dc0aa3 JB |
512 | } |
513 | ||
92ba470c JB |
514 | static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et, |
515 | struct ocfs2_extent_rec *rec) | |
516 | { | |
517 | if (et->et_ops->eo_extent_map_insert) | |
518 | et->et_ops->eo_extent_map_insert(et, rec); | |
519 | } | |
520 | ||
4c911eef JB |
521 | static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et, |
522 | u32 clusters) | |
523 | { | |
524 | if (et->et_ops->eo_extent_map_truncate) | |
525 | et->et_ops->eo_extent_map_truncate(et, clusters); | |
526 | } | |
527 | ||
13723d00 | 528 | static inline int ocfs2_et_root_journal_access(handle_t *handle, |
13723d00 JB |
529 | struct ocfs2_extent_tree *et, |
530 | int type) | |
531 | { | |
d9a0a1f8 | 532 | return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh, |
13723d00 JB |
533 | type); |
534 | } | |
535 | ||
853a3a14 TM |
536 | static inline enum ocfs2_contig_type |
537 | ocfs2_et_extent_contig(struct ocfs2_extent_tree *et, | |
538 | struct ocfs2_extent_rec *rec, | |
539 | struct ocfs2_extent_rec *insert_rec) | |
540 | { | |
541 | if (et->et_ops->eo_extent_contig) | |
542 | return et->et_ops->eo_extent_contig(et, rec, insert_rec); | |
543 | ||
544 | return ocfs2_extent_rec_contig( | |
545 | ocfs2_metadata_cache_get_super(et->et_ci), | |
546 | rec, insert_rec); | |
547 | } | |
548 | ||
6136ca5f | 549 | static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
550 | struct ocfs2_extent_rec *rec) |
551 | { | |
552 | int ret = 0; | |
553 | ||
554 | if (et->et_ops->eo_insert_check) | |
6136ca5f | 555 | ret = et->et_ops->eo_insert_check(et, rec); |
1e61ee79 JB |
556 | return ret; |
557 | } | |
558 | ||
6136ca5f | 559 | static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 560 | { |
1e61ee79 JB |
561 | int ret = 0; |
562 | ||
563 | if (et->et_ops->eo_sanity_check) | |
6136ca5f | 564 | ret = et->et_ops->eo_sanity_check(et); |
1e61ee79 | 565 | return ret; |
e7d4cb6b TM |
566 | } |
567 | ||
dcd0538f | 568 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
569 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
570 | struct ocfs2_extent_block *eb); | |
d401dc12 JB |
571 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
572 | struct ocfs2_extent_tree *et, | |
6b791bcc TM |
573 | struct ocfs2_path *path, |
574 | struct ocfs2_extent_rec *insert_rec); | |
dcd0538f MF |
575 | /* |
576 | * Reset the actual path elements so that we can re-use the structure | |
577 | * to build another path. Generally, this involves freeing the buffer | |
578 | * heads. | |
579 | */ | |
e2e9f608 | 580 | void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) |
dcd0538f MF |
581 | { |
582 | int i, start = 0, depth = 0; | |
583 | struct ocfs2_path_item *node; | |
ccd979bd | 584 | |
dcd0538f MF |
585 | if (keep_root) |
586 | start = 1; | |
ccd979bd | 587 | |
dcd0538f MF |
588 | for(i = start; i < path_num_items(path); i++) { |
589 | node = &path->p_node[i]; | |
590 | ||
591 | brelse(node->bh); | |
592 | node->bh = NULL; | |
593 | node->el = NULL; | |
594 | } | |
595 | ||
596 | /* | |
597 | * Tree depth may change during truncate, or insert. If we're | |
598 | * keeping the root extent list, then make sure that our path | |
599 | * structure reflects the proper depth. | |
600 | */ | |
601 | if (keep_root) | |
602 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
13723d00 JB |
603 | else |
604 | path_root_access(path) = NULL; | |
dcd0538f MF |
605 | |
606 | path->p_tree_depth = depth; | |
607 | } | |
608 | ||
e2e9f608 | 609 | void ocfs2_free_path(struct ocfs2_path *path) |
dcd0538f MF |
610 | { |
611 | if (path) { | |
612 | ocfs2_reinit_path(path, 0); | |
613 | kfree(path); | |
614 | } | |
615 | } | |
616 | ||
328d5752 MF |
617 | /* |
618 | * All the elements of src into dest. After this call, src could be freed | |
619 | * without affecting dest. | |
620 | * | |
621 | * Both paths should have the same root. Any non-root elements of dest | |
622 | * will be freed. | |
623 | */ | |
624 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
625 | { | |
626 | int i; | |
627 | ||
628 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
629 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
13723d00 | 630 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
328d5752 MF |
631 | |
632 | ocfs2_reinit_path(dest, 1); | |
633 | ||
634 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
635 | dest->p_node[i].bh = src->p_node[i].bh; | |
636 | dest->p_node[i].el = src->p_node[i].el; | |
637 | ||
638 | if (dest->p_node[i].bh) | |
639 | get_bh(dest->p_node[i].bh); | |
640 | } | |
641 | } | |
642 | ||
dcd0538f MF |
643 | /* |
644 | * Make the *dest path the same as src and re-initialize src path to | |
645 | * have a root only. | |
646 | */ | |
647 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
648 | { | |
649 | int i; | |
650 | ||
651 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
13723d00 | 652 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
dcd0538f MF |
653 | |
654 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
655 | brelse(dest->p_node[i].bh); | |
656 | ||
657 | dest->p_node[i].bh = src->p_node[i].bh; | |
658 | dest->p_node[i].el = src->p_node[i].el; | |
659 | ||
660 | src->p_node[i].bh = NULL; | |
661 | src->p_node[i].el = NULL; | |
662 | } | |
663 | } | |
664 | ||
665 | /* | |
666 | * Insert an extent block at given index. | |
667 | * | |
668 | * This will not take an additional reference on eb_bh. | |
669 | */ | |
670 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
671 | struct buffer_head *eb_bh) | |
672 | { | |
673 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
674 | ||
675 | /* | |
676 | * Right now, no root bh is an extent block, so this helps | |
677 | * catch code errors with dinode trees. The assertion can be | |
678 | * safely removed if we ever need to insert extent block | |
679 | * structures at the root. | |
680 | */ | |
681 | BUG_ON(index == 0); | |
682 | ||
683 | path->p_node[index].bh = eb_bh; | |
684 | path->p_node[index].el = &eb->h_list; | |
685 | } | |
686 | ||
687 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
13723d00 JB |
688 | struct ocfs2_extent_list *root_el, |
689 | ocfs2_journal_access_func access) | |
dcd0538f MF |
690 | { |
691 | struct ocfs2_path *path; | |
ccd979bd | 692 | |
dcd0538f MF |
693 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
694 | ||
695 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
696 | if (path) { | |
697 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
698 | get_bh(root_bh); | |
699 | path_root_bh(path) = root_bh; | |
700 | path_root_el(path) = root_el; | |
13723d00 | 701 | path_root_access(path) = access; |
dcd0538f MF |
702 | } |
703 | ||
704 | return path; | |
705 | } | |
706 | ||
e2e9f608 | 707 | struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path) |
ffdd7a54 | 708 | { |
13723d00 JB |
709 | return ocfs2_new_path(path_root_bh(path), path_root_el(path), |
710 | path_root_access(path)); | |
ffdd7a54 JB |
711 | } |
712 | ||
e2e9f608 | 713 | struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et) |
ffdd7a54 | 714 | { |
13723d00 JB |
715 | return ocfs2_new_path(et->et_root_bh, et->et_root_el, |
716 | et->et_root_journal_access); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Journal the buffer at depth idx. All idx>0 are extent_blocks, | |
721 | * otherwise it's the root_access function. | |
722 | * | |
723 | * I don't like the way this function's name looks next to | |
724 | * ocfs2_journal_access_path(), but I don't have a better one. | |
725 | */ | |
e2e9f608 TM |
726 | int ocfs2_path_bh_journal_access(handle_t *handle, |
727 | struct ocfs2_caching_info *ci, | |
728 | struct ocfs2_path *path, | |
729 | int idx) | |
13723d00 JB |
730 | { |
731 | ocfs2_journal_access_func access = path_root_access(path); | |
732 | ||
733 | if (!access) | |
734 | access = ocfs2_journal_access; | |
735 | ||
736 | if (idx) | |
737 | access = ocfs2_journal_access_eb; | |
738 | ||
0cf2f763 | 739 | return access(handle, ci, path->p_node[idx].bh, |
13723d00 | 740 | OCFS2_JOURNAL_ACCESS_WRITE); |
ffdd7a54 JB |
741 | } |
742 | ||
dcd0538f MF |
743 | /* |
744 | * Convenience function to journal all components in a path. | |
745 | */ | |
e2e9f608 TM |
746 | int ocfs2_journal_access_path(struct ocfs2_caching_info *ci, |
747 | handle_t *handle, | |
748 | struct ocfs2_path *path) | |
dcd0538f MF |
749 | { |
750 | int i, ret = 0; | |
751 | ||
752 | if (!path) | |
753 | goto out; | |
754 | ||
755 | for(i = 0; i < path_num_items(path); i++) { | |
0cf2f763 | 756 | ret = ocfs2_path_bh_journal_access(handle, ci, path, i); |
dcd0538f MF |
757 | if (ret < 0) { |
758 | mlog_errno(ret); | |
759 | goto out; | |
760 | } | |
761 | } | |
762 | ||
763 | out: | |
764 | return ret; | |
765 | } | |
766 | ||
328d5752 MF |
767 | /* |
768 | * Return the index of the extent record which contains cluster #v_cluster. | |
769 | * -1 is returned if it was not found. | |
770 | * | |
771 | * Should work fine on interior and exterior nodes. | |
772 | */ | |
773 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
774 | { | |
775 | int ret = -1; | |
776 | int i; | |
777 | struct ocfs2_extent_rec *rec; | |
778 | u32 rec_end, rec_start, clusters; | |
779 | ||
780 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
781 | rec = &el->l_recs[i]; | |
782 | ||
783 | rec_start = le32_to_cpu(rec->e_cpos); | |
784 | clusters = ocfs2_rec_clusters(el, rec); | |
785 | ||
786 | rec_end = rec_start + clusters; | |
787 | ||
788 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
789 | ret = i; | |
790 | break; | |
791 | } | |
792 | } | |
793 | ||
794 | return ret; | |
795 | } | |
796 | ||
e48edee2 MF |
797 | /* |
798 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
853a3a14 | 799 | * ocfs2_extent_rec_contig only work properly against leaf nodes! |
e48edee2 | 800 | */ |
dcd0538f MF |
801 | static int ocfs2_block_extent_contig(struct super_block *sb, |
802 | struct ocfs2_extent_rec *ext, | |
803 | u64 blkno) | |
ccd979bd | 804 | { |
e48edee2 MF |
805 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
806 | ||
807 | blk_end += ocfs2_clusters_to_blocks(sb, | |
808 | le16_to_cpu(ext->e_leaf_clusters)); | |
809 | ||
810 | return blkno == blk_end; | |
ccd979bd MF |
811 | } |
812 | ||
dcd0538f MF |
813 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
814 | struct ocfs2_extent_rec *right) | |
815 | { | |
e48edee2 MF |
816 | u32 left_range; |
817 | ||
818 | left_range = le32_to_cpu(left->e_cpos) + | |
819 | le16_to_cpu(left->e_leaf_clusters); | |
820 | ||
821 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
822 | } |
823 | ||
824 | static enum ocfs2_contig_type | |
853a3a14 TM |
825 | ocfs2_extent_rec_contig(struct super_block *sb, |
826 | struct ocfs2_extent_rec *ext, | |
827 | struct ocfs2_extent_rec *insert_rec) | |
dcd0538f MF |
828 | { |
829 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
830 | ||
328d5752 MF |
831 | /* |
832 | * Refuse to coalesce extent records with different flag | |
833 | * fields - we don't want to mix unwritten extents with user | |
834 | * data. | |
835 | */ | |
836 | if (ext->e_flags != insert_rec->e_flags) | |
837 | return CONTIG_NONE; | |
838 | ||
dcd0538f | 839 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
b4a17651 | 840 | ocfs2_block_extent_contig(sb, ext, blkno)) |
dcd0538f MF |
841 | return CONTIG_RIGHT; |
842 | ||
843 | blkno = le64_to_cpu(ext->e_blkno); | |
844 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
b4a17651 | 845 | ocfs2_block_extent_contig(sb, insert_rec, blkno)) |
dcd0538f MF |
846 | return CONTIG_LEFT; |
847 | ||
848 | return CONTIG_NONE; | |
849 | } | |
850 | ||
851 | /* | |
852 | * NOTE: We can have pretty much any combination of contiguousness and | |
853 | * appending. | |
854 | * | |
855 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
856 | * we'll have to update the path to that leaf. | |
857 | */ | |
858 | enum ocfs2_append_type { | |
859 | APPEND_NONE = 0, | |
860 | APPEND_TAIL, | |
861 | }; | |
862 | ||
328d5752 MF |
863 | enum ocfs2_split_type { |
864 | SPLIT_NONE = 0, | |
865 | SPLIT_LEFT, | |
866 | SPLIT_RIGHT, | |
867 | }; | |
868 | ||
dcd0538f | 869 | struct ocfs2_insert_type { |
328d5752 | 870 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
871 | enum ocfs2_append_type ins_appending; |
872 | enum ocfs2_contig_type ins_contig; | |
873 | int ins_contig_index; | |
dcd0538f MF |
874 | int ins_tree_depth; |
875 | }; | |
876 | ||
328d5752 MF |
877 | struct ocfs2_merge_ctxt { |
878 | enum ocfs2_contig_type c_contig_type; | |
879 | int c_has_empty_extent; | |
880 | int c_split_covers_rec; | |
328d5752 MF |
881 | }; |
882 | ||
5e96581a JB |
883 | static int ocfs2_validate_extent_block(struct super_block *sb, |
884 | struct buffer_head *bh) | |
885 | { | |
d6b32bbb | 886 | int rc; |
5e96581a JB |
887 | struct ocfs2_extent_block *eb = |
888 | (struct ocfs2_extent_block *)bh->b_data; | |
889 | ||
970e4936 JB |
890 | mlog(0, "Validating extent block %llu\n", |
891 | (unsigned long long)bh->b_blocknr); | |
892 | ||
d6b32bbb JB |
893 | BUG_ON(!buffer_uptodate(bh)); |
894 | ||
895 | /* | |
896 | * If the ecc fails, we return the error but otherwise | |
897 | * leave the filesystem running. We know any error is | |
898 | * local to this block. | |
899 | */ | |
900 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check); | |
13723d00 JB |
901 | if (rc) { |
902 | mlog(ML_ERROR, "Checksum failed for extent block %llu\n", | |
903 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 904 | return rc; |
13723d00 | 905 | } |
d6b32bbb JB |
906 | |
907 | /* | |
908 | * Errors after here are fatal. | |
909 | */ | |
910 | ||
5e96581a JB |
911 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { |
912 | ocfs2_error(sb, | |
913 | "Extent block #%llu has bad signature %.*s", | |
914 | (unsigned long long)bh->b_blocknr, 7, | |
915 | eb->h_signature); | |
916 | return -EINVAL; | |
917 | } | |
918 | ||
919 | if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) { | |
920 | ocfs2_error(sb, | |
921 | "Extent block #%llu has an invalid h_blkno " | |
922 | "of %llu", | |
923 | (unsigned long long)bh->b_blocknr, | |
924 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
925 | return -EINVAL; | |
926 | } | |
927 | ||
928 | if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) { | |
929 | ocfs2_error(sb, | |
930 | "Extent block #%llu has an invalid " | |
931 | "h_fs_generation of #%u", | |
932 | (unsigned long long)bh->b_blocknr, | |
933 | le32_to_cpu(eb->h_fs_generation)); | |
934 | return -EINVAL; | |
935 | } | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
3d03a305 | 940 | int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno, |
5e96581a JB |
941 | struct buffer_head **bh) |
942 | { | |
943 | int rc; | |
944 | struct buffer_head *tmp = *bh; | |
945 | ||
3d03a305 | 946 | rc = ocfs2_read_block(ci, eb_blkno, &tmp, |
970e4936 | 947 | ocfs2_validate_extent_block); |
5e96581a JB |
948 | |
949 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
970e4936 | 950 | if (!rc && !*bh) |
5e96581a JB |
951 | *bh = tmp; |
952 | ||
5e96581a JB |
953 | return rc; |
954 | } | |
955 | ||
956 | ||
ccd979bd MF |
957 | /* |
958 | * How many free extents have we got before we need more meta data? | |
959 | */ | |
960 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
f99b9b7c | 961 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
962 | { |
963 | int retval; | |
e7d4cb6b | 964 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
965 | struct ocfs2_extent_block *eb; |
966 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 967 | u64 last_eb_blk = 0; |
ccd979bd MF |
968 | |
969 | mlog_entry_void(); | |
970 | ||
f99b9b7c JB |
971 | el = et->et_root_el; |
972 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 973 | |
e7d4cb6b | 974 | if (last_eb_blk) { |
3d03a305 JB |
975 | retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk, |
976 | &eb_bh); | |
ccd979bd MF |
977 | if (retval < 0) { |
978 | mlog_errno(retval); | |
979 | goto bail; | |
980 | } | |
981 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
982 | el = &eb->h_list; | |
e7d4cb6b | 983 | } |
ccd979bd MF |
984 | |
985 | BUG_ON(el->l_tree_depth != 0); | |
986 | ||
987 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
988 | bail: | |
a81cb88b | 989 | brelse(eb_bh); |
ccd979bd MF |
990 | |
991 | mlog_exit(retval); | |
992 | return retval; | |
993 | } | |
994 | ||
995 | /* expects array to already be allocated | |
996 | * | |
997 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
998 | * l_count for you | |
999 | */ | |
42a5a7a9 JB |
1000 | static int ocfs2_create_new_meta_bhs(handle_t *handle, |
1001 | struct ocfs2_extent_tree *et, | |
ccd979bd MF |
1002 | int wanted, |
1003 | struct ocfs2_alloc_context *meta_ac, | |
1004 | struct buffer_head *bhs[]) | |
1005 | { | |
1006 | int count, status, i; | |
1007 | u16 suballoc_bit_start; | |
1008 | u32 num_got; | |
1009 | u64 first_blkno; | |
42a5a7a9 JB |
1010 | struct ocfs2_super *osb = |
1011 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
ccd979bd MF |
1012 | struct ocfs2_extent_block *eb; |
1013 | ||
1014 | mlog_entry_void(); | |
1015 | ||
1016 | count = 0; | |
1017 | while (count < wanted) { | |
1018 | status = ocfs2_claim_metadata(osb, | |
1019 | handle, | |
1020 | meta_ac, | |
1021 | wanted - count, | |
1022 | &suballoc_bit_start, | |
1023 | &num_got, | |
1024 | &first_blkno); | |
1025 | if (status < 0) { | |
1026 | mlog_errno(status); | |
1027 | goto bail; | |
1028 | } | |
1029 | ||
1030 | for(i = count; i < (num_got + count); i++) { | |
1031 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
1032 | if (bhs[i] == NULL) { | |
1033 | status = -EIO; | |
1034 | mlog_errno(status); | |
1035 | goto bail; | |
1036 | } | |
42a5a7a9 | 1037 | ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]); |
ccd979bd | 1038 | |
42a5a7a9 JB |
1039 | status = ocfs2_journal_access_eb(handle, et->et_ci, |
1040 | bhs[i], | |
13723d00 | 1041 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1042 | if (status < 0) { |
1043 | mlog_errno(status); | |
1044 | goto bail; | |
1045 | } | |
1046 | ||
1047 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
1048 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
1049 | /* Ok, setup the minimal stuff here. */ | |
1050 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
1051 | eb->h_blkno = cpu_to_le64(first_blkno); | |
1052 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
ccd979bd | 1053 | eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); |
ccd979bd MF |
1054 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
1055 | eb->h_list.l_count = | |
1056 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
1057 | ||
1058 | suballoc_bit_start++; | |
1059 | first_blkno++; | |
1060 | ||
1061 | /* We'll also be dirtied by the caller, so | |
1062 | * this isn't absolutely necessary. */ | |
1063 | status = ocfs2_journal_dirty(handle, bhs[i]); | |
1064 | if (status < 0) { | |
1065 | mlog_errno(status); | |
1066 | goto bail; | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | count += num_got; | |
1071 | } | |
1072 | ||
1073 | status = 0; | |
1074 | bail: | |
1075 | if (status < 0) { | |
1076 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 1077 | brelse(bhs[i]); |
ccd979bd MF |
1078 | bhs[i] = NULL; |
1079 | } | |
1080 | } | |
1081 | mlog_exit(status); | |
1082 | return status; | |
1083 | } | |
1084 | ||
dcd0538f MF |
1085 | /* |
1086 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
1087 | * | |
1088 | * Returns the sum of the rightmost extent rec logical offset and | |
1089 | * cluster count. | |
1090 | * | |
1091 | * ocfs2_add_branch() uses this to determine what logical cluster | |
1092 | * value should be populated into the leftmost new branch records. | |
1093 | * | |
1094 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
1095 | * value for the new topmost tree record. | |
1096 | */ | |
1097 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
1098 | { | |
1099 | int i; | |
1100 | ||
1101 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1102 | ||
1103 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 1104 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
1105 | } |
1106 | ||
6b791bcc TM |
1107 | /* |
1108 | * Change range of the branches in the right most path according to the leaf | |
1109 | * extent block's rightmost record. | |
1110 | */ | |
1111 | static int ocfs2_adjust_rightmost_branch(handle_t *handle, | |
6b791bcc TM |
1112 | struct ocfs2_extent_tree *et) |
1113 | { | |
1114 | int status; | |
1115 | struct ocfs2_path *path = NULL; | |
1116 | struct ocfs2_extent_list *el; | |
1117 | struct ocfs2_extent_rec *rec; | |
1118 | ||
1119 | path = ocfs2_new_path_from_et(et); | |
1120 | if (!path) { | |
1121 | status = -ENOMEM; | |
1122 | return status; | |
1123 | } | |
1124 | ||
facdb77f | 1125 | status = ocfs2_find_path(et->et_ci, path, UINT_MAX); |
6b791bcc TM |
1126 | if (status < 0) { |
1127 | mlog_errno(status); | |
1128 | goto out; | |
1129 | } | |
1130 | ||
1131 | status = ocfs2_extend_trans(handle, path_num_items(path) + | |
1132 | handle->h_buffer_credits); | |
1133 | if (status < 0) { | |
1134 | mlog_errno(status); | |
1135 | goto out; | |
1136 | } | |
1137 | ||
d401dc12 | 1138 | status = ocfs2_journal_access_path(et->et_ci, handle, path); |
6b791bcc TM |
1139 | if (status < 0) { |
1140 | mlog_errno(status); | |
1141 | goto out; | |
1142 | } | |
1143 | ||
1144 | el = path_leaf_el(path); | |
1145 | rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1]; | |
1146 | ||
d401dc12 | 1147 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
6b791bcc TM |
1148 | |
1149 | out: | |
1150 | ocfs2_free_path(path); | |
1151 | return status; | |
1152 | } | |
1153 | ||
ccd979bd MF |
1154 | /* |
1155 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
d401dc12 | 1156 | * to start at, if we don't want to start the branch at the root |
ccd979bd MF |
1157 | * structure. |
1158 | * | |
1159 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
1160 | * for the new last extent block. | |
1161 | * | |
1162 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 1163 | * contain a single record with cluster count == 0. |
ccd979bd | 1164 | */ |
d401dc12 | 1165 | static int ocfs2_add_branch(handle_t *handle, |
e7d4cb6b | 1166 | struct ocfs2_extent_tree *et, |
ccd979bd | 1167 | struct buffer_head *eb_bh, |
328d5752 | 1168 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
1169 | struct ocfs2_alloc_context *meta_ac) |
1170 | { | |
1171 | int status, new_blocks, i; | |
1172 | u64 next_blkno, new_last_eb_blk; | |
1173 | struct buffer_head *bh; | |
1174 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
1175 | struct ocfs2_extent_block *eb; |
1176 | struct ocfs2_extent_list *eb_el; | |
1177 | struct ocfs2_extent_list *el; | |
6b791bcc | 1178 | u32 new_cpos, root_end; |
ccd979bd MF |
1179 | |
1180 | mlog_entry_void(); | |
1181 | ||
328d5752 | 1182 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 1183 | |
ccd979bd MF |
1184 | if (eb_bh) { |
1185 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
1186 | el = &eb->h_list; | |
1187 | } else | |
ce1d9ea6 | 1188 | el = et->et_root_el; |
ccd979bd MF |
1189 | |
1190 | /* we never add a branch to a leaf. */ | |
1191 | BUG_ON(!el->l_tree_depth); | |
1192 | ||
1193 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
1194 | ||
6b791bcc TM |
1195 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
1196 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); | |
1197 | root_end = ocfs2_sum_rightmost_rec(et->et_root_el); | |
1198 | ||
1199 | /* | |
1200 | * If there is a gap before the root end and the real end | |
1201 | * of the righmost leaf block, we need to remove the gap | |
1202 | * between new_cpos and root_end first so that the tree | |
1203 | * is consistent after we add a new branch(it will start | |
1204 | * from new_cpos). | |
1205 | */ | |
1206 | if (root_end > new_cpos) { | |
1207 | mlog(0, "adjust the cluster end from %u to %u\n", | |
1208 | root_end, new_cpos); | |
d401dc12 | 1209 | status = ocfs2_adjust_rightmost_branch(handle, et); |
6b791bcc TM |
1210 | if (status) { |
1211 | mlog_errno(status); | |
1212 | goto bail; | |
1213 | } | |
1214 | } | |
1215 | ||
ccd979bd MF |
1216 | /* allocate the number of new eb blocks we need */ |
1217 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
1218 | GFP_KERNEL); | |
1219 | if (!new_eb_bhs) { | |
1220 | status = -ENOMEM; | |
1221 | mlog_errno(status); | |
1222 | goto bail; | |
1223 | } | |
1224 | ||
42a5a7a9 | 1225 | status = ocfs2_create_new_meta_bhs(handle, et, new_blocks, |
ccd979bd MF |
1226 | meta_ac, new_eb_bhs); |
1227 | if (status < 0) { | |
1228 | mlog_errno(status); | |
1229 | goto bail; | |
1230 | } | |
1231 | ||
1232 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be | |
1233 | * linked with the rest of the tree. | |
1234 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
1235 | * | |
1236 | * when we leave the loop, new_last_eb_blk will point to the | |
1237 | * newest leaf, and next_blkno will point to the topmost extent | |
1238 | * block. */ | |
1239 | next_blkno = new_last_eb_blk = 0; | |
1240 | for(i = 0; i < new_blocks; i++) { | |
1241 | bh = new_eb_bhs[i]; | |
1242 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
5e96581a JB |
1243 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1244 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1245 | eb_el = &eb->h_list; |
1246 | ||
d401dc12 | 1247 | status = ocfs2_journal_access_eb(handle, et->et_ci, bh, |
13723d00 | 1248 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1249 | if (status < 0) { |
1250 | mlog_errno(status); | |
1251 | goto bail; | |
1252 | } | |
1253 | ||
1254 | eb->h_next_leaf_blk = 0; | |
1255 | eb_el->l_tree_depth = cpu_to_le16(i); | |
1256 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
1257 | /* |
1258 | * This actually counts as an empty extent as | |
1259 | * c_clusters == 0 | |
1260 | */ | |
1261 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 1262 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
1263 | /* |
1264 | * eb_el isn't always an interior node, but even leaf | |
1265 | * nodes want a zero'd flags and reserved field so | |
1266 | * this gets the whole 32 bits regardless of use. | |
1267 | */ | |
1268 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
1269 | if (!eb_el->l_tree_depth) |
1270 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
1271 | ||
1272 | status = ocfs2_journal_dirty(handle, bh); | |
1273 | if (status < 0) { | |
1274 | mlog_errno(status); | |
1275 | goto bail; | |
1276 | } | |
1277 | ||
1278 | next_blkno = le64_to_cpu(eb->h_blkno); | |
1279 | } | |
1280 | ||
1281 | /* This is a bit hairy. We want to update up to three blocks | |
1282 | * here without leaving any of them in an inconsistent state | |
1283 | * in case of error. We don't have to worry about | |
1284 | * journal_dirty erroring as it won't unless we've aborted the | |
1285 | * handle (in which case we would never be here) so reserving | |
1286 | * the write with journal_access is all we need to do. */ | |
d401dc12 | 1287 | status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh, |
13723d00 | 1288 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1289 | if (status < 0) { |
1290 | mlog_errno(status); | |
1291 | goto bail; | |
1292 | } | |
d9a0a1f8 | 1293 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1294 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1295 | if (status < 0) { |
1296 | mlog_errno(status); | |
1297 | goto bail; | |
1298 | } | |
1299 | if (eb_bh) { | |
d401dc12 | 1300 | status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh, |
13723d00 | 1301 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1302 | if (status < 0) { |
1303 | mlog_errno(status); | |
1304 | goto bail; | |
1305 | } | |
1306 | } | |
1307 | ||
1308 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 1309 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
1310 | i = le16_to_cpu(el->l_next_free_rec); |
1311 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 1312 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 1313 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
1314 | le16_add_cpu(&el->l_next_free_rec, 1); |
1315 | ||
1316 | /* fe needs a new last extent block pointer, as does the | |
1317 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 1318 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 1319 | |
328d5752 | 1320 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
1321 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
1322 | ||
328d5752 | 1323 | status = ocfs2_journal_dirty(handle, *last_eb_bh); |
ccd979bd MF |
1324 | if (status < 0) |
1325 | mlog_errno(status); | |
ce1d9ea6 | 1326 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1327 | if (status < 0) |
1328 | mlog_errno(status); | |
1329 | if (eb_bh) { | |
1330 | status = ocfs2_journal_dirty(handle, eb_bh); | |
1331 | if (status < 0) | |
1332 | mlog_errno(status); | |
1333 | } | |
1334 | ||
328d5752 MF |
1335 | /* |
1336 | * Some callers want to track the rightmost leaf so pass it | |
1337 | * back here. | |
1338 | */ | |
1339 | brelse(*last_eb_bh); | |
1340 | get_bh(new_eb_bhs[0]); | |
1341 | *last_eb_bh = new_eb_bhs[0]; | |
1342 | ||
ccd979bd MF |
1343 | status = 0; |
1344 | bail: | |
1345 | if (new_eb_bhs) { | |
1346 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1347 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1348 | kfree(new_eb_bhs); |
1349 | } | |
1350 | ||
1351 | mlog_exit(status); | |
1352 | return status; | |
1353 | } | |
1354 | ||
1355 | /* | |
1356 | * adds another level to the allocation tree. | |
1357 | * returns back the new extent block so you can add a branch to it | |
1358 | * after this call. | |
1359 | */ | |
d401dc12 | 1360 | static int ocfs2_shift_tree_depth(handle_t *handle, |
e7d4cb6b | 1361 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1362 | struct ocfs2_alloc_context *meta_ac, |
1363 | struct buffer_head **ret_new_eb_bh) | |
1364 | { | |
1365 | int status, i; | |
dcd0538f | 1366 | u32 new_clusters; |
ccd979bd | 1367 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1368 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1369 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1370 | struct ocfs2_extent_list *eb_el; |
1371 | ||
1372 | mlog_entry_void(); | |
1373 | ||
42a5a7a9 | 1374 | status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac, |
ccd979bd MF |
1375 | &new_eb_bh); |
1376 | if (status < 0) { | |
1377 | mlog_errno(status); | |
1378 | goto bail; | |
1379 | } | |
1380 | ||
1381 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
5e96581a JB |
1382 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1383 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1384 | |
1385 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1386 | root_el = et->et_root_el; |
ccd979bd | 1387 | |
d401dc12 | 1388 | status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh, |
13723d00 | 1389 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1390 | if (status < 0) { |
1391 | mlog_errno(status); | |
1392 | goto bail; | |
1393 | } | |
1394 | ||
e7d4cb6b TM |
1395 | /* copy the root extent list data into the new extent block */ |
1396 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1397 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1398 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1399 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd MF |
1400 | |
1401 | status = ocfs2_journal_dirty(handle, new_eb_bh); | |
1402 | if (status < 0) { | |
1403 | mlog_errno(status); | |
1404 | goto bail; | |
1405 | } | |
1406 | ||
d9a0a1f8 | 1407 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1408 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1409 | if (status < 0) { |
1410 | mlog_errno(status); | |
1411 | goto bail; | |
1412 | } | |
1413 | ||
dcd0538f MF |
1414 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1415 | ||
e7d4cb6b TM |
1416 | /* update root_bh now */ |
1417 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1418 | root_el->l_recs[0].e_cpos = 0; | |
1419 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1420 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1421 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1422 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1423 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1424 | |
1425 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1426 | * becomes the allocated extent block */ | |
e7d4cb6b | 1427 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1428 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1429 | |
ce1d9ea6 | 1430 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1431 | if (status < 0) { |
1432 | mlog_errno(status); | |
1433 | goto bail; | |
1434 | } | |
1435 | ||
1436 | *ret_new_eb_bh = new_eb_bh; | |
1437 | new_eb_bh = NULL; | |
1438 | status = 0; | |
1439 | bail: | |
a81cb88b | 1440 | brelse(new_eb_bh); |
ccd979bd MF |
1441 | |
1442 | mlog_exit(status); | |
1443 | return status; | |
1444 | } | |
1445 | ||
ccd979bd MF |
1446 | /* |
1447 | * Should only be called when there is no space left in any of the | |
1448 | * leaf nodes. What we want to do is find the lowest tree depth | |
1449 | * non-leaf extent block with room for new records. There are three | |
1450 | * valid results of this search: | |
1451 | * | |
1452 | * 1) a lowest extent block is found, then we pass it back in | |
1453 | * *lowest_eb_bh and return '0' | |
1454 | * | |
e7d4cb6b | 1455 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1456 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1457 | * | |
e7d4cb6b | 1458 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1459 | * which case we return > 0 |
1460 | * | |
1461 | * return status < 0 indicates an error. | |
1462 | */ | |
d401dc12 | 1463 | static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et, |
ccd979bd MF |
1464 | struct buffer_head **target_bh) |
1465 | { | |
1466 | int status = 0, i; | |
1467 | u64 blkno; | |
ccd979bd MF |
1468 | struct ocfs2_extent_block *eb; |
1469 | struct ocfs2_extent_list *el; | |
1470 | struct buffer_head *bh = NULL; | |
1471 | struct buffer_head *lowest_bh = NULL; | |
1472 | ||
1473 | mlog_entry_void(); | |
1474 | ||
1475 | *target_bh = NULL; | |
1476 | ||
ce1d9ea6 | 1477 | el = et->et_root_el; |
ccd979bd MF |
1478 | |
1479 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1480 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3d03a305 JB |
1481 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1482 | "Owner %llu has empty " | |
ccd979bd | 1483 | "extent list (next_free_rec == 0)", |
3d03a305 | 1484 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
ccd979bd MF |
1485 | status = -EIO; |
1486 | goto bail; | |
1487 | } | |
1488 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1489 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1490 | if (!blkno) { | |
3d03a305 JB |
1491 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1492 | "Owner %llu has extent " | |
ccd979bd MF |
1493 | "list where extent # %d has no physical " |
1494 | "block start", | |
3d03a305 | 1495 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i); |
ccd979bd MF |
1496 | status = -EIO; |
1497 | goto bail; | |
1498 | } | |
1499 | ||
a81cb88b MF |
1500 | brelse(bh); |
1501 | bh = NULL; | |
ccd979bd | 1502 | |
3d03a305 | 1503 | status = ocfs2_read_extent_block(et->et_ci, blkno, &bh); |
ccd979bd MF |
1504 | if (status < 0) { |
1505 | mlog_errno(status); | |
1506 | goto bail; | |
1507 | } | |
dcd0538f MF |
1508 | |
1509 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
dcd0538f MF |
1510 | el = &eb->h_list; |
1511 | ||
1512 | if (le16_to_cpu(el->l_next_free_rec) < | |
1513 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1514 | brelse(lowest_bh); |
dcd0538f MF |
1515 | lowest_bh = bh; |
1516 | get_bh(lowest_bh); | |
1517 | } | |
1518 | } | |
1519 | ||
1520 | /* If we didn't find one and the fe doesn't have any room, | |
1521 | * then return '1' */ | |
ce1d9ea6 | 1522 | el = et->et_root_el; |
e7d4cb6b | 1523 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1524 | status = 1; |
1525 | ||
1526 | *target_bh = lowest_bh; | |
1527 | bail: | |
a81cb88b | 1528 | brelse(bh); |
dcd0538f MF |
1529 | |
1530 | mlog_exit(status); | |
1531 | return status; | |
1532 | } | |
1533 | ||
c3afcbb3 MF |
1534 | /* |
1535 | * Grow a b-tree so that it has more records. | |
1536 | * | |
1537 | * We might shift the tree depth in which case existing paths should | |
1538 | * be considered invalid. | |
1539 | * | |
1540 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1541 | * |
1542 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 | 1543 | */ |
d401dc12 JB |
1544 | static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
1545 | int *final_depth, struct buffer_head **last_eb_bh, | |
c3afcbb3 MF |
1546 | struct ocfs2_alloc_context *meta_ac) |
1547 | { | |
1548 | int ret, shift; | |
ce1d9ea6 | 1549 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1550 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1551 | struct buffer_head *bh = NULL; |
1552 | ||
1553 | BUG_ON(meta_ac == NULL); | |
1554 | ||
d401dc12 | 1555 | shift = ocfs2_find_branch_target(et, &bh); |
c3afcbb3 MF |
1556 | if (shift < 0) { |
1557 | ret = shift; | |
1558 | mlog_errno(ret); | |
1559 | goto out; | |
1560 | } | |
1561 | ||
1562 | /* We traveled all the way to the bottom of the allocation tree | |
1563 | * and didn't find room for any more extents - we need to add | |
1564 | * another tree level */ | |
1565 | if (shift) { | |
1566 | BUG_ON(bh); | |
1567 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1568 | ||
1569 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1570 | * the new extent block (so we can pass that to | |
1571 | * ocfs2_add_branch). */ | |
d401dc12 | 1572 | ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh); |
c3afcbb3 MF |
1573 | if (ret < 0) { |
1574 | mlog_errno(ret); | |
1575 | goto out; | |
1576 | } | |
1577 | depth++; | |
328d5752 MF |
1578 | if (depth == 1) { |
1579 | /* | |
1580 | * Special case: we have room now if we shifted from | |
1581 | * tree_depth 0, so no more work needs to be done. | |
1582 | * | |
1583 | * We won't be calling add_branch, so pass | |
1584 | * back *last_eb_bh as the new leaf. At depth | |
1585 | * zero, it should always be null so there's | |
1586 | * no reason to brelse. | |
1587 | */ | |
1588 | BUG_ON(*last_eb_bh); | |
1589 | get_bh(bh); | |
1590 | *last_eb_bh = bh; | |
c3afcbb3 | 1591 | goto out; |
328d5752 | 1592 | } |
c3afcbb3 MF |
1593 | } |
1594 | ||
1595 | /* call ocfs2_add_branch to add the final part of the tree with | |
1596 | * the new data. */ | |
1597 | mlog(0, "add branch. bh = %p\n", bh); | |
d401dc12 | 1598 | ret = ocfs2_add_branch(handle, et, bh, last_eb_bh, |
c3afcbb3 MF |
1599 | meta_ac); |
1600 | if (ret < 0) { | |
1601 | mlog_errno(ret); | |
1602 | goto out; | |
1603 | } | |
1604 | ||
1605 | out: | |
1606 | if (final_depth) | |
1607 | *final_depth = depth; | |
1608 | brelse(bh); | |
1609 | return ret; | |
1610 | } | |
1611 | ||
dcd0538f MF |
1612 | /* |
1613 | * This function will discard the rightmost extent record. | |
1614 | */ | |
1615 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1616 | { | |
1617 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1618 | int count = le16_to_cpu(el->l_count); | |
1619 | unsigned int num_bytes; | |
1620 | ||
1621 | BUG_ON(!next_free); | |
1622 | /* This will cause us to go off the end of our extent list. */ | |
1623 | BUG_ON(next_free >= count); | |
1624 | ||
1625 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1626 | ||
1627 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1628 | } | |
1629 | ||
1630 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1631 | struct ocfs2_extent_rec *insert_rec) | |
1632 | { | |
1633 | int i, insert_index, next_free, has_empty, num_bytes; | |
1634 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1635 | struct ocfs2_extent_rec *rec; | |
1636 | ||
1637 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1638 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1639 | ||
1640 | BUG_ON(!next_free); | |
1641 | ||
1642 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1643 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1644 | |
1645 | /* | |
1646 | * The easiest way to approach this is to just remove the | |
1647 | * empty extent and temporarily decrement next_free. | |
1648 | */ | |
1649 | if (has_empty) { | |
1650 | /* | |
1651 | * If next_free was 1 (only an empty extent), this | |
1652 | * loop won't execute, which is fine. We still want | |
1653 | * the decrement above to happen. | |
1654 | */ | |
1655 | for(i = 0; i < (next_free - 1); i++) | |
1656 | el->l_recs[i] = el->l_recs[i+1]; | |
1657 | ||
1658 | next_free--; | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * Figure out what the new record index should be. | |
1663 | */ | |
1664 | for(i = 0; i < next_free; i++) { | |
1665 | rec = &el->l_recs[i]; | |
1666 | ||
1667 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1668 | break; | |
1669 | } | |
1670 | insert_index = i; | |
1671 | ||
1672 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1673 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1674 | ||
1675 | BUG_ON(insert_index < 0); | |
1676 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1677 | BUG_ON(insert_index > next_free); | |
1678 | ||
1679 | /* | |
1680 | * No need to memmove if we're just adding to the tail. | |
1681 | */ | |
1682 | if (insert_index != next_free) { | |
1683 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1684 | ||
1685 | num_bytes = next_free - insert_index; | |
1686 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1687 | memmove(&el->l_recs[insert_index + 1], | |
1688 | &el->l_recs[insert_index], | |
1689 | num_bytes); | |
1690 | } | |
1691 | ||
1692 | /* | |
1693 | * Either we had an empty extent, and need to re-increment or | |
1694 | * there was no empty extent on a non full rightmost leaf node, | |
1695 | * in which case we still need to increment. | |
1696 | */ | |
1697 | next_free++; | |
1698 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1699 | /* | |
1700 | * Make sure none of the math above just messed up our tree. | |
1701 | */ | |
1702 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1703 | ||
1704 | el->l_recs[insert_index] = *insert_rec; | |
1705 | ||
1706 | } | |
1707 | ||
328d5752 MF |
1708 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1709 | { | |
1710 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1711 | ||
1712 | BUG_ON(num_recs == 0); | |
1713 | ||
1714 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1715 | num_recs--; | |
1716 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1717 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1718 | memset(&el->l_recs[num_recs], 0, | |
1719 | sizeof(struct ocfs2_extent_rec)); | |
1720 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1721 | } | |
1722 | } | |
1723 | ||
dcd0538f MF |
1724 | /* |
1725 | * Create an empty extent record . | |
1726 | * | |
1727 | * l_next_free_rec may be updated. | |
1728 | * | |
1729 | * If an empty extent already exists do nothing. | |
1730 | */ | |
1731 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1732 | { | |
1733 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1734 | ||
e48edee2 MF |
1735 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1736 | ||
dcd0538f MF |
1737 | if (next_free == 0) |
1738 | goto set_and_inc; | |
1739 | ||
1740 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1741 | return; | |
1742 | ||
1743 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1744 | "Asked to create an empty extent in a full list:\n" | |
1745 | "count = %u, tree depth = %u", | |
1746 | le16_to_cpu(el->l_count), | |
1747 | le16_to_cpu(el->l_tree_depth)); | |
1748 | ||
1749 | ocfs2_shift_records_right(el); | |
1750 | ||
1751 | set_and_inc: | |
1752 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1753 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1754 | } | |
1755 | ||
1756 | /* | |
1757 | * For a rotation which involves two leaf nodes, the "root node" is | |
1758 | * the lowest level tree node which contains a path to both leafs. This | |
1759 | * resulting set of information can be used to form a complete "subtree" | |
1760 | * | |
1761 | * This function is passed two full paths from the dinode down to a | |
1762 | * pair of adjacent leaves. It's task is to figure out which path | |
1763 | * index contains the subtree root - this can be the root index itself | |
1764 | * in a worst-case rotation. | |
1765 | * | |
1766 | * The array index of the subtree root is passed back. | |
1767 | */ | |
7dc02805 | 1768 | static int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et, |
dcd0538f MF |
1769 | struct ocfs2_path *left, |
1770 | struct ocfs2_path *right) | |
1771 | { | |
1772 | int i = 0; | |
1773 | ||
1774 | /* | |
1775 | * Check that the caller passed in two paths from the same tree. | |
1776 | */ | |
1777 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1778 | ||
1779 | do { | |
1780 | i++; | |
1781 | ||
1782 | /* | |
1783 | * The caller didn't pass two adjacent paths. | |
1784 | */ | |
1785 | mlog_bug_on_msg(i > left->p_tree_depth, | |
7dc02805 | 1786 | "Owner %llu, left depth %u, right depth %u\n" |
dcd0538f | 1787 | "left leaf blk %llu, right leaf blk %llu\n", |
7dc02805 JB |
1788 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
1789 | left->p_tree_depth, right->p_tree_depth, | |
dcd0538f MF |
1790 | (unsigned long long)path_leaf_bh(left)->b_blocknr, |
1791 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1792 | } while (left->p_node[i].bh->b_blocknr == | |
1793 | right->p_node[i].bh->b_blocknr); | |
1794 | ||
1795 | return i - 1; | |
1796 | } | |
1797 | ||
1798 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1799 | ||
1800 | /* | |
1801 | * Traverse a btree path in search of cpos, starting at root_el. | |
1802 | * | |
1803 | * This code can be called with a cpos larger than the tree, in which | |
1804 | * case it will return the rightmost path. | |
1805 | */ | |
facdb77f | 1806 | static int __ocfs2_find_path(struct ocfs2_caching_info *ci, |
dcd0538f MF |
1807 | struct ocfs2_extent_list *root_el, u32 cpos, |
1808 | path_insert_t *func, void *data) | |
1809 | { | |
1810 | int i, ret = 0; | |
1811 | u32 range; | |
1812 | u64 blkno; | |
1813 | struct buffer_head *bh = NULL; | |
1814 | struct ocfs2_extent_block *eb; | |
1815 | struct ocfs2_extent_list *el; | |
1816 | struct ocfs2_extent_rec *rec; | |
dcd0538f MF |
1817 | |
1818 | el = root_el; | |
1819 | while (el->l_tree_depth) { | |
1820 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
facdb77f JB |
1821 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1822 | "Owner %llu has empty extent list at " | |
dcd0538f | 1823 | "depth %u\n", |
facdb77f | 1824 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1825 | le16_to_cpu(el->l_tree_depth)); |
1826 | ret = -EROFS; | |
1827 | goto out; | |
1828 | ||
1829 | } | |
1830 | ||
1831 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1832 | rec = &el->l_recs[i]; | |
1833 | ||
1834 | /* | |
1835 | * In the case that cpos is off the allocation | |
1836 | * tree, this should just wind up returning the | |
1837 | * rightmost record. | |
1838 | */ | |
1839 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1840 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1841 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1842 | break; | |
1843 | } | |
1844 | ||
1845 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1846 | if (blkno == 0) { | |
facdb77f JB |
1847 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1848 | "Owner %llu has bad blkno in extent list " | |
dcd0538f | 1849 | "at depth %u (index %d)\n", |
facdb77f | 1850 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1851 | le16_to_cpu(el->l_tree_depth), i); |
1852 | ret = -EROFS; | |
1853 | goto out; | |
1854 | } | |
1855 | ||
1856 | brelse(bh); | |
1857 | bh = NULL; | |
facdb77f | 1858 | ret = ocfs2_read_extent_block(ci, blkno, &bh); |
dcd0538f MF |
1859 | if (ret) { |
1860 | mlog_errno(ret); | |
1861 | goto out; | |
1862 | } | |
1863 | ||
1864 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1865 | el = &eb->h_list; | |
dcd0538f MF |
1866 | |
1867 | if (le16_to_cpu(el->l_next_free_rec) > | |
1868 | le16_to_cpu(el->l_count)) { | |
facdb77f JB |
1869 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1870 | "Owner %llu has bad count in extent list " | |
dcd0538f | 1871 | "at block %llu (next free=%u, count=%u)\n", |
facdb77f | 1872 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1873 | (unsigned long long)bh->b_blocknr, |
1874 | le16_to_cpu(el->l_next_free_rec), | |
1875 | le16_to_cpu(el->l_count)); | |
1876 | ret = -EROFS; | |
1877 | goto out; | |
1878 | } | |
1879 | ||
1880 | if (func) | |
1881 | func(data, bh); | |
1882 | } | |
1883 | ||
1884 | out: | |
1885 | /* | |
1886 | * Catch any trailing bh that the loop didn't handle. | |
1887 | */ | |
1888 | brelse(bh); | |
1889 | ||
1890 | return ret; | |
1891 | } | |
1892 | ||
1893 | /* | |
1894 | * Given an initialized path (that is, it has a valid root extent | |
1895 | * list), this function will traverse the btree in search of the path | |
1896 | * which would contain cpos. | |
1897 | * | |
1898 | * The path traveled is recorded in the path structure. | |
1899 | * | |
1900 | * Note that this will not do any comparisons on leaf node extent | |
1901 | * records, so it will work fine in the case that we just added a tree | |
1902 | * branch. | |
1903 | */ | |
1904 | struct find_path_data { | |
1905 | int index; | |
1906 | struct ocfs2_path *path; | |
1907 | }; | |
1908 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1909 | { | |
1910 | struct find_path_data *fp = data; | |
1911 | ||
1912 | get_bh(bh); | |
1913 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1914 | fp->index++; | |
1915 | } | |
e2e9f608 TM |
1916 | int ocfs2_find_path(struct ocfs2_caching_info *ci, |
1917 | struct ocfs2_path *path, u32 cpos) | |
dcd0538f MF |
1918 | { |
1919 | struct find_path_data data; | |
1920 | ||
1921 | data.index = 1; | |
1922 | data.path = path; | |
facdb77f | 1923 | return __ocfs2_find_path(ci, path_root_el(path), cpos, |
dcd0538f MF |
1924 | find_path_ins, &data); |
1925 | } | |
1926 | ||
1927 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1928 | { | |
1929 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1930 | struct ocfs2_extent_list *el = &eb->h_list; | |
1931 | struct buffer_head **ret = data; | |
1932 | ||
1933 | /* We want to retain only the leaf block. */ | |
1934 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1935 | get_bh(bh); | |
1936 | *ret = bh; | |
1937 | } | |
1938 | } | |
1939 | /* | |
1940 | * Find the leaf block in the tree which would contain cpos. No | |
1941 | * checking of the actual leaf is done. | |
1942 | * | |
1943 | * Some paths want to call this instead of allocating a path structure | |
1944 | * and calling ocfs2_find_path(). | |
1945 | * | |
1946 | * This function doesn't handle non btree extent lists. | |
1947 | */ | |
facdb77f JB |
1948 | int ocfs2_find_leaf(struct ocfs2_caching_info *ci, |
1949 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1950 | struct buffer_head **leaf_bh) | |
dcd0538f MF |
1951 | { |
1952 | int ret; | |
1953 | struct buffer_head *bh = NULL; | |
1954 | ||
facdb77f | 1955 | ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh); |
dcd0538f MF |
1956 | if (ret) { |
1957 | mlog_errno(ret); | |
1958 | goto out; | |
1959 | } | |
1960 | ||
1961 | *leaf_bh = bh; | |
1962 | out: | |
1963 | return ret; | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1968 | * | |
1969 | * Basically, we've moved stuff around at the bottom of the tree and | |
1970 | * we need to fix up the extent records above the changes to reflect | |
1971 | * the new changes. | |
1972 | * | |
1973 | * left_rec: the record on the left. | |
1974 | * left_child_el: is the child list pointed to by left_rec | |
1975 | * right_rec: the record to the right of left_rec | |
1976 | * right_child_el: is the child list pointed to by right_rec | |
1977 | * | |
1978 | * By definition, this only works on interior nodes. | |
1979 | */ | |
1980 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1981 | struct ocfs2_extent_list *left_child_el, | |
1982 | struct ocfs2_extent_rec *right_rec, | |
1983 | struct ocfs2_extent_list *right_child_el) | |
1984 | { | |
1985 | u32 left_clusters, right_end; | |
1986 | ||
1987 | /* | |
1988 | * Interior nodes never have holes. Their cpos is the cpos of | |
1989 | * the leftmost record in their child list. Their cluster | |
1990 | * count covers the full theoretical range of their child list | |
1991 | * - the range between their cpos and the cpos of the record | |
1992 | * immediately to their right. | |
1993 | */ | |
1994 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
82e12644 TM |
1995 | if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) { |
1996 | BUG_ON(right_child_el->l_tree_depth); | |
328d5752 MF |
1997 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); |
1998 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1999 | } | |
dcd0538f | 2000 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 2001 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
2002 | |
2003 | /* | |
2004 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 2005 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
2006 | * updating e_cpos to keep the same highest cluster count. |
2007 | */ | |
2008 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 2009 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
2010 | |
2011 | right_rec->e_cpos = left_rec->e_cpos; | |
2012 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
2013 | ||
2014 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 2015 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
2016 | } |
2017 | ||
2018 | /* | |
2019 | * Adjust the adjacent root node records involved in a | |
2020 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
2021 | * find it's index in the root list. | |
2022 | */ | |
2023 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
2024 | struct ocfs2_extent_list *left_el, | |
2025 | struct ocfs2_extent_list *right_el, | |
2026 | u64 left_el_blkno) | |
2027 | { | |
2028 | int i; | |
2029 | ||
2030 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
2031 | le16_to_cpu(left_el->l_tree_depth)); | |
2032 | ||
2033 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
2034 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
2035 | break; | |
2036 | } | |
2037 | ||
2038 | /* | |
2039 | * The path walking code should have never returned a root and | |
2040 | * two paths which are not adjacent. | |
2041 | */ | |
2042 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
2043 | ||
2044 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
2045 | &root_el->l_recs[i + 1], right_el); | |
2046 | } | |
2047 | ||
2048 | /* | |
2049 | * We've changed a leaf block (in right_path) and need to reflect that | |
2050 | * change back up the subtree. | |
2051 | * | |
2052 | * This happens in multiple places: | |
2053 | * - When we've moved an extent record from the left path leaf to the right | |
2054 | * path leaf to make room for an empty extent in the left path leaf. | |
2055 | * - When our insert into the right path leaf is at the leftmost edge | |
2056 | * and requires an update of the path immediately to it's left. This | |
2057 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
2058 | * - When we've adjusted the last extent record in the left path leaf and the |
2059 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f | 2060 | */ |
4619c73e | 2061 | static void ocfs2_complete_edge_insert(handle_t *handle, |
dcd0538f MF |
2062 | struct ocfs2_path *left_path, |
2063 | struct ocfs2_path *right_path, | |
2064 | int subtree_index) | |
2065 | { | |
2066 | int ret, i, idx; | |
2067 | struct ocfs2_extent_list *el, *left_el, *right_el; | |
2068 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
2069 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2070 | ||
2071 | /* | |
2072 | * Update the counts and position values within all the | |
2073 | * interior nodes to reflect the leaf rotation we just did. | |
2074 | * | |
2075 | * The root node is handled below the loop. | |
2076 | * | |
2077 | * We begin the loop with right_el and left_el pointing to the | |
2078 | * leaf lists and work our way up. | |
2079 | * | |
2080 | * NOTE: within this loop, left_el and right_el always refer | |
2081 | * to the *child* lists. | |
2082 | */ | |
2083 | left_el = path_leaf_el(left_path); | |
2084 | right_el = path_leaf_el(right_path); | |
2085 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
2086 | mlog(0, "Adjust records at index %u\n", i); | |
2087 | ||
2088 | /* | |
2089 | * One nice property of knowing that all of these | |
2090 | * nodes are below the root is that we only deal with | |
2091 | * the leftmost right node record and the rightmost | |
2092 | * left node record. | |
2093 | */ | |
2094 | el = left_path->p_node[i].el; | |
2095 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2096 | left_rec = &el->l_recs[idx]; | |
2097 | ||
2098 | el = right_path->p_node[i].el; | |
2099 | right_rec = &el->l_recs[0]; | |
2100 | ||
2101 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
2102 | right_el); | |
2103 | ||
2104 | ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh); | |
2105 | if (ret) | |
2106 | mlog_errno(ret); | |
2107 | ||
2108 | ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
2109 | if (ret) | |
2110 | mlog_errno(ret); | |
2111 | ||
2112 | /* | |
2113 | * Setup our list pointers now so that the current | |
2114 | * parents become children in the next iteration. | |
2115 | */ | |
2116 | left_el = left_path->p_node[i].el; | |
2117 | right_el = right_path->p_node[i].el; | |
2118 | } | |
2119 | ||
2120 | /* | |
2121 | * At the root node, adjust the two adjacent records which | |
2122 | * begin our path to the leaves. | |
2123 | */ | |
2124 | ||
2125 | el = left_path->p_node[subtree_index].el; | |
2126 | left_el = left_path->p_node[subtree_index + 1].el; | |
2127 | right_el = right_path->p_node[subtree_index + 1].el; | |
2128 | ||
2129 | ocfs2_adjust_root_records(el, left_el, right_el, | |
2130 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
2131 | ||
2132 | root_bh = left_path->p_node[subtree_index].bh; | |
2133 | ||
2134 | ret = ocfs2_journal_dirty(handle, root_bh); | |
2135 | if (ret) | |
2136 | mlog_errno(ret); | |
2137 | } | |
2138 | ||
5c601aba JB |
2139 | static int ocfs2_rotate_subtree_right(handle_t *handle, |
2140 | struct ocfs2_extent_tree *et, | |
dcd0538f MF |
2141 | struct ocfs2_path *left_path, |
2142 | struct ocfs2_path *right_path, | |
2143 | int subtree_index) | |
2144 | { | |
2145 | int ret, i; | |
2146 | struct buffer_head *right_leaf_bh; | |
2147 | struct buffer_head *left_leaf_bh = NULL; | |
2148 | struct buffer_head *root_bh; | |
2149 | struct ocfs2_extent_list *right_el, *left_el; | |
2150 | struct ocfs2_extent_rec move_rec; | |
2151 | ||
2152 | left_leaf_bh = path_leaf_bh(left_path); | |
2153 | left_el = path_leaf_el(left_path); | |
2154 | ||
2155 | if (left_el->l_next_free_rec != left_el->l_count) { | |
5c601aba | 2156 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
dcd0538f MF |
2157 | "Inode %llu has non-full interior leaf node %llu" |
2158 | "(next free = %u)", | |
5c601aba | 2159 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2160 | (unsigned long long)left_leaf_bh->b_blocknr, |
2161 | le16_to_cpu(left_el->l_next_free_rec)); | |
2162 | return -EROFS; | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * This extent block may already have an empty record, so we | |
2167 | * return early if so. | |
2168 | */ | |
2169 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
2170 | return 0; | |
2171 | ||
2172 | root_bh = left_path->p_node[subtree_index].bh; | |
2173 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2174 | ||
5c601aba | 2175 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2176 | subtree_index); |
dcd0538f MF |
2177 | if (ret) { |
2178 | mlog_errno(ret); | |
2179 | goto out; | |
2180 | } | |
2181 | ||
2182 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
5c601aba | 2183 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2184 | right_path, i); |
dcd0538f MF |
2185 | if (ret) { |
2186 | mlog_errno(ret); | |
2187 | goto out; | |
2188 | } | |
2189 | ||
5c601aba | 2190 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2191 | left_path, i); |
dcd0538f MF |
2192 | if (ret) { |
2193 | mlog_errno(ret); | |
2194 | goto out; | |
2195 | } | |
2196 | } | |
2197 | ||
2198 | right_leaf_bh = path_leaf_bh(right_path); | |
2199 | right_el = path_leaf_el(right_path); | |
2200 | ||
2201 | /* This is a code error, not a disk corruption. */ | |
2202 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
2203 | "because rightmost leaf block %llu is empty\n", | |
5c601aba | 2204 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2205 | (unsigned long long)right_leaf_bh->b_blocknr); |
2206 | ||
2207 | ocfs2_create_empty_extent(right_el); | |
2208 | ||
2209 | ret = ocfs2_journal_dirty(handle, right_leaf_bh); | |
2210 | if (ret) { | |
2211 | mlog_errno(ret); | |
2212 | goto out; | |
2213 | } | |
2214 | ||
2215 | /* Do the copy now. */ | |
2216 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2217 | move_rec = left_el->l_recs[i]; | |
2218 | right_el->l_recs[0] = move_rec; | |
2219 | ||
2220 | /* | |
2221 | * Clear out the record we just copied and shift everything | |
2222 | * over, leaving an empty extent in the left leaf. | |
2223 | * | |
2224 | * We temporarily subtract from next_free_rec so that the | |
2225 | * shift will lose the tail record (which is now defunct). | |
2226 | */ | |
2227 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
2228 | ocfs2_shift_records_right(left_el); | |
2229 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2230 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
2231 | ||
2232 | ret = ocfs2_journal_dirty(handle, left_leaf_bh); | |
2233 | if (ret) { | |
2234 | mlog_errno(ret); | |
2235 | goto out; | |
2236 | } | |
2237 | ||
4619c73e JB |
2238 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
2239 | subtree_index); | |
dcd0538f MF |
2240 | |
2241 | out: | |
2242 | return ret; | |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Given a full path, determine what cpos value would return us a path | |
2247 | * containing the leaf immediately to the left of the current one. | |
2248 | * | |
2249 | * Will return zero if the path passed in is already the leftmost path. | |
2250 | */ | |
2251 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
2252 | struct ocfs2_path *path, u32 *cpos) | |
2253 | { | |
2254 | int i, j, ret = 0; | |
2255 | u64 blkno; | |
2256 | struct ocfs2_extent_list *el; | |
2257 | ||
e48edee2 MF |
2258 | BUG_ON(path->p_tree_depth == 0); |
2259 | ||
dcd0538f MF |
2260 | *cpos = 0; |
2261 | ||
2262 | blkno = path_leaf_bh(path)->b_blocknr; | |
2263 | ||
2264 | /* Start at the tree node just above the leaf and work our way up. */ | |
2265 | i = path->p_tree_depth - 1; | |
2266 | while (i >= 0) { | |
2267 | el = path->p_node[i].el; | |
2268 | ||
2269 | /* | |
2270 | * Find the extent record just before the one in our | |
2271 | * path. | |
2272 | */ | |
2273 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2274 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2275 | if (j == 0) { | |
2276 | if (i == 0) { | |
2277 | /* | |
2278 | * We've determined that the | |
2279 | * path specified is already | |
2280 | * the leftmost one - return a | |
2281 | * cpos of zero. | |
2282 | */ | |
2283 | goto out; | |
2284 | } | |
2285 | /* | |
2286 | * The leftmost record points to our | |
2287 | * leaf - we need to travel up the | |
2288 | * tree one level. | |
2289 | */ | |
2290 | goto next_node; | |
2291 | } | |
2292 | ||
2293 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
2294 | *cpos = *cpos + ocfs2_rec_clusters(el, |
2295 | &el->l_recs[j - 1]); | |
2296 | *cpos = *cpos - 1; | |
dcd0538f MF |
2297 | goto out; |
2298 | } | |
2299 | } | |
2300 | ||
2301 | /* | |
2302 | * If we got here, we never found a valid node where | |
2303 | * the tree indicated one should be. | |
2304 | */ | |
2305 | ocfs2_error(sb, | |
2306 | "Invalid extent tree at extent block %llu\n", | |
2307 | (unsigned long long)blkno); | |
2308 | ret = -EROFS; | |
2309 | goto out; | |
2310 | ||
2311 | next_node: | |
2312 | blkno = path->p_node[i].bh->b_blocknr; | |
2313 | i--; | |
2314 | } | |
2315 | ||
2316 | out: | |
2317 | return ret; | |
2318 | } | |
2319 | ||
328d5752 MF |
2320 | /* |
2321 | * Extend the transaction by enough credits to complete the rotation, | |
2322 | * and still leave at least the original number of credits allocated | |
2323 | * to this transaction. | |
2324 | */ | |
dcd0538f | 2325 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2326 | int op_credits, |
dcd0538f MF |
2327 | struct ocfs2_path *path) |
2328 | { | |
c18b812d | 2329 | int ret; |
328d5752 | 2330 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f | 2331 | |
c18b812d TM |
2332 | if (handle->h_buffer_credits < credits) { |
2333 | ret = ocfs2_extend_trans(handle, | |
2334 | credits - handle->h_buffer_credits); | |
2335 | if (ret) | |
2336 | return ret; | |
2337 | ||
2338 | if (unlikely(handle->h_buffer_credits < credits)) | |
2339 | return ocfs2_extend_trans(handle, credits); | |
2340 | } | |
dcd0538f MF |
2341 | |
2342 | return 0; | |
2343 | } | |
2344 | ||
2345 | /* | |
2346 | * Trap the case where we're inserting into the theoretical range past | |
2347 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2348 | * whose cpos is less than ours into the right leaf. | |
2349 | * | |
2350 | * It's only necessary to look at the rightmost record of the left | |
2351 | * leaf because the logic that calls us should ensure that the | |
2352 | * theoretical ranges in the path components above the leaves are | |
2353 | * correct. | |
2354 | */ | |
2355 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2356 | u32 insert_cpos) | |
2357 | { | |
2358 | struct ocfs2_extent_list *left_el; | |
2359 | struct ocfs2_extent_rec *rec; | |
2360 | int next_free; | |
2361 | ||
2362 | left_el = path_leaf_el(left_path); | |
2363 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2364 | rec = &left_el->l_recs[next_free - 1]; | |
2365 | ||
2366 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2367 | return 1; | |
2368 | return 0; | |
2369 | } | |
2370 | ||
328d5752 MF |
2371 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2372 | { | |
2373 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2374 | unsigned int range; | |
2375 | struct ocfs2_extent_rec *rec; | |
2376 | ||
2377 | if (next_free == 0) | |
2378 | return 0; | |
2379 | ||
2380 | rec = &el->l_recs[0]; | |
2381 | if (ocfs2_is_empty_extent(rec)) { | |
2382 | /* Empty list. */ | |
2383 | if (next_free == 1) | |
2384 | return 0; | |
2385 | rec = &el->l_recs[1]; | |
2386 | } | |
2387 | ||
2388 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2389 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2390 | return 1; | |
2391 | return 0; | |
2392 | } | |
2393 | ||
dcd0538f MF |
2394 | /* |
2395 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2396 | * | |
2397 | * The path to the rightmost leaf should be passed in. | |
2398 | * | |
2399 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2400 | * | |
af901ca1 | 2401 | * Upon successful return from this function: |
dcd0538f MF |
2402 | * |
2403 | * - The 'right_path' array will contain a path to the leaf block | |
2404 | * whose range contains e_cpos. | |
2405 | * - That leaf block will have a single empty extent in list index 0. | |
2406 | * - In the case that the rotation requires a post-insert update, | |
2407 | * *ret_left_path will contain a valid path which can be passed to | |
2408 | * ocfs2_insert_path(). | |
2409 | */ | |
1bbf0b8d | 2410 | static int ocfs2_rotate_tree_right(handle_t *handle, |
5c601aba | 2411 | struct ocfs2_extent_tree *et, |
328d5752 | 2412 | enum ocfs2_split_type split, |
dcd0538f MF |
2413 | u32 insert_cpos, |
2414 | struct ocfs2_path *right_path, | |
2415 | struct ocfs2_path **ret_left_path) | |
2416 | { | |
328d5752 | 2417 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2418 | u32 cpos; |
2419 | struct ocfs2_path *left_path = NULL; | |
5c601aba | 2420 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
dcd0538f MF |
2421 | |
2422 | *ret_left_path = NULL; | |
2423 | ||
ffdd7a54 | 2424 | left_path = ocfs2_new_path_from_path(right_path); |
dcd0538f MF |
2425 | if (!left_path) { |
2426 | ret = -ENOMEM; | |
2427 | mlog_errno(ret); | |
2428 | goto out; | |
2429 | } | |
2430 | ||
5c601aba | 2431 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2432 | if (ret) { |
2433 | mlog_errno(ret); | |
2434 | goto out; | |
2435 | } | |
2436 | ||
2437 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2438 | ||
2439 | /* | |
2440 | * What we want to do here is: | |
2441 | * | |
2442 | * 1) Start with the rightmost path. | |
2443 | * | |
2444 | * 2) Determine a path to the leaf block directly to the left | |
2445 | * of that leaf. | |
2446 | * | |
2447 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2448 | * which contains a path to both leaves. | |
2449 | * | |
2450 | * 4) Rotate the subtree. | |
2451 | * | |
2452 | * 5) Find the next subtree by considering the left path to be | |
2453 | * the new right path. | |
2454 | * | |
2455 | * The check at the top of this while loop also accepts | |
2456 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2457 | * value to get us the left path - insert_cpos might very well | |
2458 | * be filling that hole. | |
2459 | * | |
2460 | * Stop at a cpos of '0' because we either started at the | |
2461 | * leftmost branch (i.e., a tree with one branch and a | |
2462 | * rotation inside of it), or we've gone as far as we can in | |
2463 | * rotating subtrees. | |
2464 | */ | |
2465 | while (cpos && insert_cpos <= cpos) { | |
2466 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2467 | insert_cpos, cpos); | |
2468 | ||
5c601aba | 2469 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
dcd0538f MF |
2470 | if (ret) { |
2471 | mlog_errno(ret); | |
2472 | goto out; | |
2473 | } | |
2474 | ||
2475 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2476 | path_leaf_bh(right_path), | |
5c601aba | 2477 | "Owner %llu: error during insert of %u " |
dcd0538f MF |
2478 | "(left path cpos %u) results in two identical " |
2479 | "paths ending at %llu\n", | |
5c601aba JB |
2480 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
2481 | insert_cpos, cpos, | |
dcd0538f MF |
2482 | (unsigned long long) |
2483 | path_leaf_bh(left_path)->b_blocknr); | |
2484 | ||
328d5752 MF |
2485 | if (split == SPLIT_NONE && |
2486 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2487 | insert_cpos)) { |
dcd0538f MF |
2488 | |
2489 | /* | |
2490 | * We've rotated the tree as much as we | |
2491 | * should. The rest is up to | |
2492 | * ocfs2_insert_path() to complete, after the | |
2493 | * record insertion. We indicate this | |
2494 | * situation by returning the left path. | |
2495 | * | |
2496 | * The reason we don't adjust the records here | |
2497 | * before the record insert is that an error | |
2498 | * later might break the rule where a parent | |
2499 | * record e_cpos will reflect the actual | |
2500 | * e_cpos of the 1st nonempty record of the | |
2501 | * child list. | |
2502 | */ | |
2503 | *ret_left_path = left_path; | |
2504 | goto out_ret_path; | |
2505 | } | |
2506 | ||
7dc02805 | 2507 | start = ocfs2_find_subtree_root(et, left_path, right_path); |
dcd0538f MF |
2508 | |
2509 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2510 | start, | |
2511 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2512 | right_path->p_tree_depth); | |
2513 | ||
2514 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2515 | orig_credits, right_path); |
dcd0538f MF |
2516 | if (ret) { |
2517 | mlog_errno(ret); | |
2518 | goto out; | |
2519 | } | |
2520 | ||
5c601aba | 2521 | ret = ocfs2_rotate_subtree_right(handle, et, left_path, |
dcd0538f MF |
2522 | right_path, start); |
2523 | if (ret) { | |
2524 | mlog_errno(ret); | |
2525 | goto out; | |
2526 | } | |
2527 | ||
328d5752 MF |
2528 | if (split != SPLIT_NONE && |
2529 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2530 | insert_cpos)) { | |
2531 | /* | |
2532 | * A rotate moves the rightmost left leaf | |
2533 | * record over to the leftmost right leaf | |
2534 | * slot. If we're doing an extent split | |
2535 | * instead of a real insert, then we have to | |
2536 | * check that the extent to be split wasn't | |
2537 | * just moved over. If it was, then we can | |
2538 | * exit here, passing left_path back - | |
2539 | * ocfs2_split_extent() is smart enough to | |
2540 | * search both leaves. | |
2541 | */ | |
2542 | *ret_left_path = left_path; | |
2543 | goto out_ret_path; | |
2544 | } | |
2545 | ||
dcd0538f MF |
2546 | /* |
2547 | * There is no need to re-read the next right path | |
2548 | * as we know that it'll be our current left | |
2549 | * path. Optimize by copying values instead. | |
2550 | */ | |
2551 | ocfs2_mv_path(right_path, left_path); | |
2552 | ||
5c601aba | 2553 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2554 | if (ret) { |
2555 | mlog_errno(ret); | |
2556 | goto out; | |
2557 | } | |
2558 | } | |
2559 | ||
2560 | out: | |
2561 | ocfs2_free_path(left_path); | |
2562 | ||
2563 | out_ret_path: | |
2564 | return ret; | |
2565 | } | |
2566 | ||
09106bae JB |
2567 | static int ocfs2_update_edge_lengths(handle_t *handle, |
2568 | struct ocfs2_extent_tree *et, | |
3c5e1068 | 2569 | int subtree_index, struct ocfs2_path *path) |
dcd0538f | 2570 | { |
3c5e1068 | 2571 | int i, idx, ret; |
dcd0538f | 2572 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2573 | struct ocfs2_extent_list *el; |
2574 | struct ocfs2_extent_block *eb; | |
2575 | u32 range; | |
dcd0538f | 2576 | |
3c5e1068 TM |
2577 | /* |
2578 | * In normal tree rotation process, we will never touch the | |
2579 | * tree branch above subtree_index and ocfs2_extend_rotate_transaction | |
2580 | * doesn't reserve the credits for them either. | |
2581 | * | |
2582 | * But we do have a special case here which will update the rightmost | |
2583 | * records for all the bh in the path. | |
2584 | * So we have to allocate extra credits and access them. | |
2585 | */ | |
2586 | ret = ocfs2_extend_trans(handle, | |
2587 | handle->h_buffer_credits + subtree_index); | |
2588 | if (ret) { | |
2589 | mlog_errno(ret); | |
2590 | goto out; | |
2591 | } | |
2592 | ||
09106bae | 2593 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
3c5e1068 TM |
2594 | if (ret) { |
2595 | mlog_errno(ret); | |
2596 | goto out; | |
2597 | } | |
2598 | ||
328d5752 MF |
2599 | /* Path should always be rightmost. */ |
2600 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2601 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2602 | |
328d5752 MF |
2603 | el = &eb->h_list; |
2604 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2605 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2606 | rec = &el->l_recs[idx]; | |
2607 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2608 | |
328d5752 MF |
2609 | for (i = 0; i < path->p_tree_depth; i++) { |
2610 | el = path->p_node[i].el; | |
2611 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2612 | rec = &el->l_recs[idx]; | |
dcd0538f | 2613 | |
328d5752 MF |
2614 | rec->e_int_clusters = cpu_to_le32(range); |
2615 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2616 | |
328d5752 | 2617 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2618 | } |
3c5e1068 TM |
2619 | out: |
2620 | return ret; | |
dcd0538f MF |
2621 | } |
2622 | ||
6641b0ce JB |
2623 | static void ocfs2_unlink_path(handle_t *handle, |
2624 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2625 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2626 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2627 | { |
328d5752 MF |
2628 | int ret, i; |
2629 | struct ocfs2_extent_block *eb; | |
2630 | struct ocfs2_extent_list *el; | |
2631 | struct buffer_head *bh; | |
2632 | ||
2633 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2634 | bh = path->p_node[i].bh; | |
2635 | ||
2636 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2637 | /* | |
2638 | * Not all nodes might have had their final count | |
2639 | * decremented by the caller - handle this here. | |
2640 | */ | |
2641 | el = &eb->h_list; | |
2642 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2643 | mlog(ML_ERROR, | |
2644 | "Inode %llu, attempted to remove extent block " | |
2645 | "%llu with %u records\n", | |
6641b0ce | 2646 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
2647 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
2648 | le16_to_cpu(el->l_next_free_rec)); | |
2649 | ||
2650 | ocfs2_journal_dirty(handle, bh); | |
6641b0ce | 2651 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 MF |
2652 | continue; |
2653 | } | |
2654 | ||
2655 | el->l_next_free_rec = 0; | |
2656 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2657 | ||
2658 | ocfs2_journal_dirty(handle, bh); | |
2659 | ||
2660 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2661 | if (ret) | |
2662 | mlog_errno(ret); | |
2663 | ||
6641b0ce | 2664 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 | 2665 | } |
dcd0538f MF |
2666 | } |
2667 | ||
6641b0ce JB |
2668 | static void ocfs2_unlink_subtree(handle_t *handle, |
2669 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2670 | struct ocfs2_path *left_path, |
2671 | struct ocfs2_path *right_path, | |
2672 | int subtree_index, | |
2673 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2674 | { |
328d5752 MF |
2675 | int i; |
2676 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2677 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2678 | struct ocfs2_extent_list *el; |
328d5752 | 2679 | struct ocfs2_extent_block *eb; |
dcd0538f | 2680 | |
328d5752 | 2681 | el = path_leaf_el(left_path); |
dcd0538f | 2682 | |
328d5752 | 2683 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2684 | |
328d5752 MF |
2685 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2686 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2687 | break; | |
dcd0538f | 2688 | |
328d5752 | 2689 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2690 | |
328d5752 MF |
2691 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2692 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2693 | ||
2694 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2695 | eb->h_next_leaf_blk = 0; | |
2696 | ||
2697 | ocfs2_journal_dirty(handle, root_bh); | |
2698 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2699 | ||
6641b0ce | 2700 | ocfs2_unlink_path(handle, et, dealloc, right_path, |
328d5752 MF |
2701 | subtree_index + 1); |
2702 | } | |
2703 | ||
1e2dd63f JB |
2704 | static int ocfs2_rotate_subtree_left(handle_t *handle, |
2705 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2706 | struct ocfs2_path *left_path, |
2707 | struct ocfs2_path *right_path, | |
2708 | int subtree_index, | |
2709 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
1e2dd63f | 2710 | int *deleted) |
328d5752 MF |
2711 | { |
2712 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2713 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2714 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2715 | struct ocfs2_extent_block *eb; | |
2716 | ||
2717 | *deleted = 0; | |
2718 | ||
2719 | right_leaf_el = path_leaf_el(right_path); | |
2720 | left_leaf_el = path_leaf_el(left_path); | |
2721 | root_bh = left_path->p_node[subtree_index].bh; | |
2722 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2723 | ||
2724 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2725 | return 0; | |
dcd0538f | 2726 | |
328d5752 MF |
2727 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2728 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2729 | /* |
328d5752 MF |
2730 | * It's legal for us to proceed if the right leaf is |
2731 | * the rightmost one and it has an empty extent. There | |
2732 | * are two cases to handle - whether the leaf will be | |
2733 | * empty after removal or not. If the leaf isn't empty | |
2734 | * then just remove the empty extent up front. The | |
2735 | * next block will handle empty leaves by flagging | |
2736 | * them for unlink. | |
2737 | * | |
2738 | * Non rightmost leaves will throw -EAGAIN and the | |
2739 | * caller can manually move the subtree and retry. | |
dcd0538f | 2740 | */ |
dcd0538f | 2741 | |
328d5752 MF |
2742 | if (eb->h_next_leaf_blk != 0ULL) |
2743 | return -EAGAIN; | |
2744 | ||
2745 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
1e2dd63f | 2746 | ret = ocfs2_journal_access_eb(handle, et->et_ci, |
13723d00 JB |
2747 | path_leaf_bh(right_path), |
2748 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2749 | if (ret) { |
2750 | mlog_errno(ret); | |
2751 | goto out; | |
2752 | } | |
2753 | ||
328d5752 MF |
2754 | ocfs2_remove_empty_extent(right_leaf_el); |
2755 | } else | |
2756 | right_has_empty = 1; | |
dcd0538f MF |
2757 | } |
2758 | ||
328d5752 MF |
2759 | if (eb->h_next_leaf_blk == 0ULL && |
2760 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2761 | /* | |
2762 | * We have to update i_last_eb_blk during the meta | |
2763 | * data delete. | |
2764 | */ | |
d9a0a1f8 | 2765 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 2766 | OCFS2_JOURNAL_ACCESS_WRITE); |
328d5752 MF |
2767 | if (ret) { |
2768 | mlog_errno(ret); | |
2769 | goto out; | |
2770 | } | |
2771 | ||
2772 | del_right_subtree = 1; | |
2773 | } | |
2774 | ||
2775 | /* | |
2776 | * Getting here with an empty extent in the right path implies | |
2777 | * that it's the rightmost path and will be deleted. | |
2778 | */ | |
2779 | BUG_ON(right_has_empty && !del_right_subtree); | |
2780 | ||
1e2dd63f | 2781 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2782 | subtree_index); |
328d5752 MF |
2783 | if (ret) { |
2784 | mlog_errno(ret); | |
2785 | goto out; | |
2786 | } | |
2787 | ||
2788 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1e2dd63f | 2789 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2790 | right_path, i); |
328d5752 MF |
2791 | if (ret) { |
2792 | mlog_errno(ret); | |
2793 | goto out; | |
2794 | } | |
2795 | ||
1e2dd63f | 2796 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2797 | left_path, i); |
328d5752 MF |
2798 | if (ret) { |
2799 | mlog_errno(ret); | |
2800 | goto out; | |
2801 | } | |
2802 | } | |
2803 | ||
2804 | if (!right_has_empty) { | |
2805 | /* | |
2806 | * Only do this if we're moving a real | |
2807 | * record. Otherwise, the action is delayed until | |
2808 | * after removal of the right path in which case we | |
2809 | * can do a simple shift to remove the empty extent. | |
2810 | */ | |
2811 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2812 | memset(&right_leaf_el->l_recs[0], 0, | |
2813 | sizeof(struct ocfs2_extent_rec)); | |
2814 | } | |
2815 | if (eb->h_next_leaf_blk == 0ULL) { | |
2816 | /* | |
2817 | * Move recs over to get rid of empty extent, decrease | |
2818 | * next_free. This is allowed to remove the last | |
2819 | * extent in our leaf (setting l_next_free_rec to | |
2820 | * zero) - the delete code below won't care. | |
2821 | */ | |
2822 | ocfs2_remove_empty_extent(right_leaf_el); | |
2823 | } | |
2824 | ||
2825 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2826 | if (ret) | |
2827 | mlog_errno(ret); | |
2828 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
2829 | if (ret) | |
2830 | mlog_errno(ret); | |
2831 | ||
2832 | if (del_right_subtree) { | |
6641b0ce | 2833 | ocfs2_unlink_subtree(handle, et, left_path, right_path, |
328d5752 | 2834 | subtree_index, dealloc); |
09106bae | 2835 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
2836 | left_path); |
2837 | if (ret) { | |
2838 | mlog_errno(ret); | |
2839 | goto out; | |
2840 | } | |
328d5752 MF |
2841 | |
2842 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2843 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2844 | |
2845 | /* | |
2846 | * Removal of the extent in the left leaf was skipped | |
2847 | * above so we could delete the right path | |
2848 | * 1st. | |
2849 | */ | |
2850 | if (right_has_empty) | |
2851 | ocfs2_remove_empty_extent(left_leaf_el); | |
2852 | ||
e7d4cb6b | 2853 | ret = ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2854 | if (ret) |
2855 | mlog_errno(ret); | |
2856 | ||
2857 | *deleted = 1; | |
2858 | } else | |
4619c73e | 2859 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
328d5752 MF |
2860 | subtree_index); |
2861 | ||
2862 | out: | |
2863 | return ret; | |
2864 | } | |
2865 | ||
2866 | /* | |
2867 | * Given a full path, determine what cpos value would return us a path | |
2868 | * containing the leaf immediately to the right of the current one. | |
2869 | * | |
2870 | * Will return zero if the path passed in is already the rightmost path. | |
2871 | * | |
2872 | * This looks similar, but is subtly different to | |
2873 | * ocfs2_find_cpos_for_left_leaf(). | |
2874 | */ | |
2875 | static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, | |
2876 | struct ocfs2_path *path, u32 *cpos) | |
2877 | { | |
2878 | int i, j, ret = 0; | |
2879 | u64 blkno; | |
2880 | struct ocfs2_extent_list *el; | |
2881 | ||
2882 | *cpos = 0; | |
2883 | ||
2884 | if (path->p_tree_depth == 0) | |
2885 | return 0; | |
2886 | ||
2887 | blkno = path_leaf_bh(path)->b_blocknr; | |
2888 | ||
2889 | /* Start at the tree node just above the leaf and work our way up. */ | |
2890 | i = path->p_tree_depth - 1; | |
2891 | while (i >= 0) { | |
2892 | int next_free; | |
2893 | ||
2894 | el = path->p_node[i].el; | |
2895 | ||
2896 | /* | |
2897 | * Find the extent record just after the one in our | |
2898 | * path. | |
2899 | */ | |
2900 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2901 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2902 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2903 | if (j == (next_free - 1)) { | |
2904 | if (i == 0) { | |
2905 | /* | |
2906 | * We've determined that the | |
2907 | * path specified is already | |
2908 | * the rightmost one - return a | |
2909 | * cpos of zero. | |
2910 | */ | |
2911 | goto out; | |
2912 | } | |
2913 | /* | |
2914 | * The rightmost record points to our | |
2915 | * leaf - we need to travel up the | |
2916 | * tree one level. | |
2917 | */ | |
2918 | goto next_node; | |
2919 | } | |
2920 | ||
2921 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2922 | goto out; | |
2923 | } | |
2924 | } | |
2925 | ||
2926 | /* | |
2927 | * If we got here, we never found a valid node where | |
2928 | * the tree indicated one should be. | |
2929 | */ | |
2930 | ocfs2_error(sb, | |
2931 | "Invalid extent tree at extent block %llu\n", | |
2932 | (unsigned long long)blkno); | |
2933 | ret = -EROFS; | |
2934 | goto out; | |
2935 | ||
2936 | next_node: | |
2937 | blkno = path->p_node[i].bh->b_blocknr; | |
2938 | i--; | |
2939 | } | |
2940 | ||
2941 | out: | |
2942 | return ret; | |
2943 | } | |
2944 | ||
70f18c08 JB |
2945 | static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle, |
2946 | struct ocfs2_extent_tree *et, | |
13723d00 | 2947 | struct ocfs2_path *path) |
328d5752 MF |
2948 | { |
2949 | int ret; | |
13723d00 JB |
2950 | struct buffer_head *bh = path_leaf_bh(path); |
2951 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
328d5752 MF |
2952 | |
2953 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2954 | return 0; | |
2955 | ||
70f18c08 | 2956 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
13723d00 | 2957 | path_num_items(path) - 1); |
328d5752 MF |
2958 | if (ret) { |
2959 | mlog_errno(ret); | |
2960 | goto out; | |
2961 | } | |
2962 | ||
2963 | ocfs2_remove_empty_extent(el); | |
2964 | ||
2965 | ret = ocfs2_journal_dirty(handle, bh); | |
2966 | if (ret) | |
2967 | mlog_errno(ret); | |
2968 | ||
2969 | out: | |
2970 | return ret; | |
2971 | } | |
2972 | ||
e46f74dc JB |
2973 | static int __ocfs2_rotate_tree_left(handle_t *handle, |
2974 | struct ocfs2_extent_tree *et, | |
2975 | int orig_credits, | |
328d5752 MF |
2976 | struct ocfs2_path *path, |
2977 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e46f74dc | 2978 | struct ocfs2_path **empty_extent_path) |
328d5752 MF |
2979 | { |
2980 | int ret, subtree_root, deleted; | |
2981 | u32 right_cpos; | |
2982 | struct ocfs2_path *left_path = NULL; | |
2983 | struct ocfs2_path *right_path = NULL; | |
e46f74dc | 2984 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
328d5752 MF |
2985 | |
2986 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2987 | ||
2988 | *empty_extent_path = NULL; | |
2989 | ||
e46f74dc | 2990 | ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
328d5752 MF |
2991 | if (ret) { |
2992 | mlog_errno(ret); | |
2993 | goto out; | |
2994 | } | |
2995 | ||
ffdd7a54 | 2996 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2997 | if (!left_path) { |
2998 | ret = -ENOMEM; | |
2999 | mlog_errno(ret); | |
3000 | goto out; | |
3001 | } | |
3002 | ||
3003 | ocfs2_cp_path(left_path, path); | |
3004 | ||
ffdd7a54 | 3005 | right_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3006 | if (!right_path) { |
3007 | ret = -ENOMEM; | |
3008 | mlog_errno(ret); | |
3009 | goto out; | |
3010 | } | |
3011 | ||
3012 | while (right_cpos) { | |
facdb77f | 3013 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
328d5752 MF |
3014 | if (ret) { |
3015 | mlog_errno(ret); | |
3016 | goto out; | |
3017 | } | |
3018 | ||
7dc02805 | 3019 | subtree_root = ocfs2_find_subtree_root(et, left_path, |
328d5752 MF |
3020 | right_path); |
3021 | ||
3022 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
3023 | subtree_root, | |
3024 | (unsigned long long) | |
3025 | right_path->p_node[subtree_root].bh->b_blocknr, | |
3026 | right_path->p_tree_depth); | |
3027 | ||
3028 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
3029 | orig_credits, left_path); | |
3030 | if (ret) { | |
3031 | mlog_errno(ret); | |
3032 | goto out; | |
3033 | } | |
3034 | ||
e8aed345 MF |
3035 | /* |
3036 | * Caller might still want to make changes to the | |
3037 | * tree root, so re-add it to the journal here. | |
3038 | */ | |
e46f74dc | 3039 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3040 | left_path, 0); |
e8aed345 MF |
3041 | if (ret) { |
3042 | mlog_errno(ret); | |
3043 | goto out; | |
3044 | } | |
3045 | ||
1e2dd63f | 3046 | ret = ocfs2_rotate_subtree_left(handle, et, left_path, |
328d5752 | 3047 | right_path, subtree_root, |
1e2dd63f | 3048 | dealloc, &deleted); |
328d5752 MF |
3049 | if (ret == -EAGAIN) { |
3050 | /* | |
3051 | * The rotation has to temporarily stop due to | |
3052 | * the right subtree having an empty | |
3053 | * extent. Pass it back to the caller for a | |
3054 | * fixup. | |
3055 | */ | |
3056 | *empty_extent_path = right_path; | |
3057 | right_path = NULL; | |
3058 | goto out; | |
3059 | } | |
3060 | if (ret) { | |
3061 | mlog_errno(ret); | |
3062 | goto out; | |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * The subtree rotate might have removed records on | |
3067 | * the rightmost edge. If so, then rotation is | |
3068 | * complete. | |
3069 | */ | |
3070 | if (deleted) | |
3071 | break; | |
3072 | ||
3073 | ocfs2_mv_path(left_path, right_path); | |
3074 | ||
e46f74dc | 3075 | ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, |
328d5752 MF |
3076 | &right_cpos); |
3077 | if (ret) { | |
3078 | mlog_errno(ret); | |
3079 | goto out; | |
3080 | } | |
3081 | } | |
3082 | ||
3083 | out: | |
3084 | ocfs2_free_path(right_path); | |
3085 | ocfs2_free_path(left_path); | |
3086 | ||
3087 | return ret; | |
3088 | } | |
3089 | ||
70f18c08 JB |
3090 | static int ocfs2_remove_rightmost_path(handle_t *handle, |
3091 | struct ocfs2_extent_tree *et, | |
e7d4cb6b | 3092 | struct ocfs2_path *path, |
70f18c08 | 3093 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3094 | { |
3095 | int ret, subtree_index; | |
3096 | u32 cpos; | |
3097 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
3098 | struct ocfs2_extent_block *eb; |
3099 | struct ocfs2_extent_list *el; | |
3100 | ||
328d5752 | 3101 | |
6136ca5f | 3102 | ret = ocfs2_et_sanity_check(et); |
e7d4cb6b TM |
3103 | if (ret) |
3104 | goto out; | |
328d5752 MF |
3105 | /* |
3106 | * There's two ways we handle this depending on | |
3107 | * whether path is the only existing one. | |
3108 | */ | |
3109 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3110 | handle->h_buffer_credits, | |
3111 | path); | |
3112 | if (ret) { | |
3113 | mlog_errno(ret); | |
3114 | goto out; | |
3115 | } | |
3116 | ||
d9a0a1f8 | 3117 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
328d5752 MF |
3118 | if (ret) { |
3119 | mlog_errno(ret); | |
3120 | goto out; | |
3121 | } | |
3122 | ||
3d03a305 JB |
3123 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3124 | path, &cpos); | |
328d5752 MF |
3125 | if (ret) { |
3126 | mlog_errno(ret); | |
3127 | goto out; | |
3128 | } | |
3129 | ||
3130 | if (cpos) { | |
3131 | /* | |
3132 | * We have a path to the left of this one - it needs | |
3133 | * an update too. | |
3134 | */ | |
ffdd7a54 | 3135 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3136 | if (!left_path) { |
3137 | ret = -ENOMEM; | |
3138 | mlog_errno(ret); | |
3139 | goto out; | |
3140 | } | |
3141 | ||
facdb77f | 3142 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
3143 | if (ret) { |
3144 | mlog_errno(ret); | |
3145 | goto out; | |
3146 | } | |
3147 | ||
d9a0a1f8 | 3148 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
328d5752 MF |
3149 | if (ret) { |
3150 | mlog_errno(ret); | |
3151 | goto out; | |
3152 | } | |
3153 | ||
7dc02805 | 3154 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
328d5752 | 3155 | |
6641b0ce | 3156 | ocfs2_unlink_subtree(handle, et, left_path, path, |
328d5752 | 3157 | subtree_index, dealloc); |
09106bae | 3158 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
3159 | left_path); |
3160 | if (ret) { | |
3161 | mlog_errno(ret); | |
3162 | goto out; | |
3163 | } | |
328d5752 MF |
3164 | |
3165 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 3166 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
3167 | } else { |
3168 | /* | |
3169 | * 'path' is also the leftmost path which | |
3170 | * means it must be the only one. This gets | |
3171 | * handled differently because we want to | |
70f18c08 | 3172 | * revert the root back to having extents |
328d5752 MF |
3173 | * in-line. |
3174 | */ | |
6641b0ce | 3175 | ocfs2_unlink_path(handle, et, dealloc, path, 1); |
328d5752 | 3176 | |
ce1d9ea6 | 3177 | el = et->et_root_el; |
328d5752 MF |
3178 | el->l_tree_depth = 0; |
3179 | el->l_next_free_rec = 0; | |
3180 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3181 | ||
35dc0aa3 | 3182 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
3183 | } |
3184 | ||
3185 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
3186 | ||
3187 | out: | |
3188 | ocfs2_free_path(left_path); | |
3189 | return ret; | |
3190 | } | |
3191 | ||
3192 | /* | |
3193 | * Left rotation of btree records. | |
3194 | * | |
3195 | * In many ways, this is (unsurprisingly) the opposite of right | |
3196 | * rotation. We start at some non-rightmost path containing an empty | |
3197 | * extent in the leaf block. The code works its way to the rightmost | |
3198 | * path by rotating records to the left in every subtree. | |
3199 | * | |
3200 | * This is used by any code which reduces the number of extent records | |
3201 | * in a leaf. After removal, an empty record should be placed in the | |
3202 | * leftmost list position. | |
3203 | * | |
3204 | * This won't handle a length update of the rightmost path records if | |
3205 | * the rightmost tree leaf record is removed so the caller is | |
3206 | * responsible for detecting and correcting that. | |
3207 | */ | |
70f18c08 JB |
3208 | static int ocfs2_rotate_tree_left(handle_t *handle, |
3209 | struct ocfs2_extent_tree *et, | |
328d5752 | 3210 | struct ocfs2_path *path, |
70f18c08 | 3211 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3212 | { |
3213 | int ret, orig_credits = handle->h_buffer_credits; | |
3214 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
3215 | struct ocfs2_extent_block *eb; | |
3216 | struct ocfs2_extent_list *el; | |
3217 | ||
3218 | el = path_leaf_el(path); | |
3219 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
3220 | return 0; | |
3221 | ||
3222 | if (path->p_tree_depth == 0) { | |
3223 | rightmost_no_delete: | |
3224 | /* | |
e7d4cb6b | 3225 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
3226 | * it up front. |
3227 | */ | |
70f18c08 | 3228 | ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path); |
328d5752 MF |
3229 | if (ret) |
3230 | mlog_errno(ret); | |
3231 | goto out; | |
3232 | } | |
3233 | ||
3234 | /* | |
3235 | * Handle rightmost branch now. There's several cases: | |
3236 | * 1) simple rotation leaving records in there. That's trivial. | |
3237 | * 2) rotation requiring a branch delete - there's no more | |
3238 | * records left. Two cases of this: | |
3239 | * a) There are branches to the left. | |
3240 | * b) This is also the leftmost (the only) branch. | |
3241 | * | |
3242 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
3243 | * 2a) we need the left branch so that we can update it with the unlink | |
70f18c08 | 3244 | * 2b) we need to bring the root back to inline extents. |
328d5752 MF |
3245 | */ |
3246 | ||
3247 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
3248 | el = &eb->h_list; | |
3249 | if (eb->h_next_leaf_blk == 0) { | |
3250 | /* | |
3251 | * This gets a bit tricky if we're going to delete the | |
3252 | * rightmost path. Get the other cases out of the way | |
3253 | * 1st. | |
3254 | */ | |
3255 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
3256 | goto rightmost_no_delete; | |
3257 | ||
3258 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3259 | ret = -EIO; | |
70f18c08 JB |
3260 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3261 | "Owner %llu has empty extent block at %llu", | |
3262 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
328d5752 MF |
3263 | (unsigned long long)le64_to_cpu(eb->h_blkno)); |
3264 | goto out; | |
3265 | } | |
3266 | ||
3267 | /* | |
3268 | * XXX: The caller can not trust "path" any more after | |
3269 | * this as it will have been deleted. What do we do? | |
3270 | * | |
3271 | * In theory the rotate-for-merge code will never get | |
3272 | * here because it'll always ask for a rotate in a | |
3273 | * nonempty list. | |
3274 | */ | |
3275 | ||
70f18c08 JB |
3276 | ret = ocfs2_remove_rightmost_path(handle, et, path, |
3277 | dealloc); | |
328d5752 MF |
3278 | if (ret) |
3279 | mlog_errno(ret); | |
3280 | goto out; | |
3281 | } | |
3282 | ||
3283 | /* | |
3284 | * Now we can loop, remembering the path we get from -EAGAIN | |
3285 | * and restarting from there. | |
3286 | */ | |
3287 | try_rotate: | |
e46f74dc JB |
3288 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path, |
3289 | dealloc, &restart_path); | |
328d5752 MF |
3290 | if (ret && ret != -EAGAIN) { |
3291 | mlog_errno(ret); | |
3292 | goto out; | |
3293 | } | |
3294 | ||
3295 | while (ret == -EAGAIN) { | |
3296 | tmp_path = restart_path; | |
3297 | restart_path = NULL; | |
3298 | ||
e46f74dc | 3299 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, |
328d5752 | 3300 | tmp_path, dealloc, |
e46f74dc | 3301 | &restart_path); |
328d5752 MF |
3302 | if (ret && ret != -EAGAIN) { |
3303 | mlog_errno(ret); | |
3304 | goto out; | |
3305 | } | |
3306 | ||
3307 | ocfs2_free_path(tmp_path); | |
3308 | tmp_path = NULL; | |
3309 | ||
3310 | if (ret == 0) | |
3311 | goto try_rotate; | |
3312 | } | |
3313 | ||
3314 | out: | |
3315 | ocfs2_free_path(tmp_path); | |
3316 | ocfs2_free_path(restart_path); | |
3317 | return ret; | |
3318 | } | |
3319 | ||
3320 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
3321 | int index) | |
3322 | { | |
3323 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
3324 | unsigned int size; | |
3325 | ||
3326 | if (rec->e_leaf_clusters == 0) { | |
3327 | /* | |
3328 | * We consumed all of the merged-from record. An empty | |
3329 | * extent cannot exist anywhere but the 1st array | |
3330 | * position, so move things over if the merged-from | |
3331 | * record doesn't occupy that position. | |
3332 | * | |
3333 | * This creates a new empty extent so the caller | |
3334 | * should be smart enough to have removed any existing | |
3335 | * ones. | |
3336 | */ | |
3337 | if (index > 0) { | |
3338 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3339 | size = index * sizeof(struct ocfs2_extent_rec); | |
3340 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
3341 | } | |
3342 | ||
3343 | /* | |
3344 | * Always memset - the caller doesn't check whether it | |
3345 | * created an empty extent, so there could be junk in | |
3346 | * the other fields. | |
3347 | */ | |
3348 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3349 | } | |
3350 | } | |
3351 | ||
4fe82c31 | 3352 | static int ocfs2_get_right_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3353 | struct ocfs2_path *left_path, |
3354 | struct ocfs2_path **ret_right_path) | |
3355 | { | |
3356 | int ret; | |
3357 | u32 right_cpos; | |
3358 | struct ocfs2_path *right_path = NULL; | |
3359 | struct ocfs2_extent_list *left_el; | |
3360 | ||
3361 | *ret_right_path = NULL; | |
3362 | ||
3363 | /* This function shouldn't be called for non-trees. */ | |
3364 | BUG_ON(left_path->p_tree_depth == 0); | |
3365 | ||
3366 | left_el = path_leaf_el(left_path); | |
3367 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3368 | ||
4fe82c31 JB |
3369 | ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3370 | left_path, &right_cpos); | |
677b9752 TM |
3371 | if (ret) { |
3372 | mlog_errno(ret); | |
3373 | goto out; | |
3374 | } | |
3375 | ||
3376 | /* This function shouldn't be called for the rightmost leaf. */ | |
3377 | BUG_ON(right_cpos == 0); | |
3378 | ||
ffdd7a54 | 3379 | right_path = ocfs2_new_path_from_path(left_path); |
677b9752 TM |
3380 | if (!right_path) { |
3381 | ret = -ENOMEM; | |
3382 | mlog_errno(ret); | |
3383 | goto out; | |
3384 | } | |
3385 | ||
4fe82c31 | 3386 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
677b9752 TM |
3387 | if (ret) { |
3388 | mlog_errno(ret); | |
3389 | goto out; | |
3390 | } | |
3391 | ||
3392 | *ret_right_path = right_path; | |
3393 | out: | |
3394 | if (ret) | |
3395 | ocfs2_free_path(right_path); | |
3396 | return ret; | |
3397 | } | |
3398 | ||
328d5752 MF |
3399 | /* |
3400 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3401 | * onto the beginning of the record "next" to it. |
3402 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3403 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3404 | * next extent block. | |
328d5752 | 3405 | */ |
4fe82c31 | 3406 | static int ocfs2_merge_rec_right(struct ocfs2_path *left_path, |
677b9752 | 3407 | handle_t *handle, |
7dc02805 | 3408 | struct ocfs2_extent_tree *et, |
677b9752 TM |
3409 | struct ocfs2_extent_rec *split_rec, |
3410 | int index) | |
328d5752 | 3411 | { |
677b9752 | 3412 | int ret, next_free, i; |
328d5752 MF |
3413 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3414 | struct ocfs2_extent_rec *left_rec; | |
3415 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3416 | struct ocfs2_extent_list *right_el; |
3417 | struct ocfs2_path *right_path = NULL; | |
3418 | int subtree_index = 0; | |
3419 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3420 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3421 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3422 | |
3423 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3424 | left_rec = &el->l_recs[index]; |
677b9752 | 3425 | |
9d8df6aa | 3426 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3427 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3428 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3429 | ret = ocfs2_get_right_path(et, left_path, &right_path); |
677b9752 TM |
3430 | if (ret) { |
3431 | mlog_errno(ret); | |
3432 | goto out; | |
3433 | } | |
3434 | ||
3435 | right_el = path_leaf_el(right_path); | |
3436 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3437 | BUG_ON(next_free <= 0); | |
3438 | right_rec = &right_el->l_recs[0]; | |
3439 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3440 | BUG_ON(next_free <= 1); |
677b9752 TM |
3441 | right_rec = &right_el->l_recs[1]; |
3442 | } | |
3443 | ||
3444 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3445 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3446 | le32_to_cpu(right_rec->e_cpos)); | |
3447 | ||
7dc02805 JB |
3448 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3449 | right_path); | |
677b9752 TM |
3450 | |
3451 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3452 | handle->h_buffer_credits, | |
3453 | right_path); | |
3454 | if (ret) { | |
3455 | mlog_errno(ret); | |
3456 | goto out; | |
3457 | } | |
3458 | ||
3459 | root_bh = left_path->p_node[subtree_index].bh; | |
3460 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3461 | ||
7dc02805 | 3462 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3463 | subtree_index); |
677b9752 TM |
3464 | if (ret) { |
3465 | mlog_errno(ret); | |
3466 | goto out; | |
3467 | } | |
3468 | ||
3469 | for (i = subtree_index + 1; | |
3470 | i < path_num_items(right_path); i++) { | |
7dc02805 | 3471 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3472 | right_path, i); |
677b9752 TM |
3473 | if (ret) { |
3474 | mlog_errno(ret); | |
3475 | goto out; | |
3476 | } | |
3477 | ||
7dc02805 | 3478 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3479 | left_path, i); |
677b9752 TM |
3480 | if (ret) { |
3481 | mlog_errno(ret); | |
3482 | goto out; | |
3483 | } | |
3484 | } | |
3485 | ||
3486 | } else { | |
3487 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3488 | right_rec = &el->l_recs[index + 1]; | |
3489 | } | |
328d5752 | 3490 | |
7dc02805 | 3491 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path, |
13723d00 | 3492 | path_num_items(left_path) - 1); |
328d5752 MF |
3493 | if (ret) { |
3494 | mlog_errno(ret); | |
3495 | goto out; | |
3496 | } | |
3497 | ||
3498 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3499 | ||
3500 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3501 | le64_add_cpu(&right_rec->e_blkno, | |
7dc02805 JB |
3502 | -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3503 | split_clusters)); | |
328d5752 MF |
3504 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); |
3505 | ||
3506 | ocfs2_cleanup_merge(el, index); | |
3507 | ||
3508 | ret = ocfs2_journal_dirty(handle, bh); | |
3509 | if (ret) | |
3510 | mlog_errno(ret); | |
3511 | ||
677b9752 TM |
3512 | if (right_path) { |
3513 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
3514 | if (ret) | |
3515 | mlog_errno(ret); | |
3516 | ||
4619c73e JB |
3517 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
3518 | subtree_index); | |
677b9752 TM |
3519 | } |
3520 | out: | |
3521 | if (right_path) | |
3522 | ocfs2_free_path(right_path); | |
3523 | return ret; | |
3524 | } | |
3525 | ||
4fe82c31 | 3526 | static int ocfs2_get_left_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3527 | struct ocfs2_path *right_path, |
3528 | struct ocfs2_path **ret_left_path) | |
3529 | { | |
3530 | int ret; | |
3531 | u32 left_cpos; | |
3532 | struct ocfs2_path *left_path = NULL; | |
3533 | ||
3534 | *ret_left_path = NULL; | |
3535 | ||
3536 | /* This function shouldn't be called for non-trees. */ | |
3537 | BUG_ON(right_path->p_tree_depth == 0); | |
3538 | ||
4fe82c31 | 3539 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
677b9752 TM |
3540 | right_path, &left_cpos); |
3541 | if (ret) { | |
3542 | mlog_errno(ret); | |
3543 | goto out; | |
3544 | } | |
3545 | ||
3546 | /* This function shouldn't be called for the leftmost leaf. */ | |
3547 | BUG_ON(left_cpos == 0); | |
3548 | ||
ffdd7a54 | 3549 | left_path = ocfs2_new_path_from_path(right_path); |
677b9752 TM |
3550 | if (!left_path) { |
3551 | ret = -ENOMEM; | |
3552 | mlog_errno(ret); | |
3553 | goto out; | |
3554 | } | |
3555 | ||
4fe82c31 | 3556 | ret = ocfs2_find_path(et->et_ci, left_path, left_cpos); |
677b9752 TM |
3557 | if (ret) { |
3558 | mlog_errno(ret); | |
3559 | goto out; | |
3560 | } | |
3561 | ||
3562 | *ret_left_path = left_path; | |
328d5752 | 3563 | out: |
677b9752 TM |
3564 | if (ret) |
3565 | ocfs2_free_path(left_path); | |
328d5752 MF |
3566 | return ret; |
3567 | } | |
3568 | ||
3569 | /* | |
3570 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3571 | * onto the tail of the record "before" it. |
3572 | * For index > 0, the "before" means the extent rec at index - 1. | |
3573 | * | |
3574 | * For index == 0, the "before" means the last record of the previous | |
3575 | * extent block. And there is also a situation that we may need to | |
3576 | * remove the rightmost leaf extent block in the right_path and change | |
3577 | * the right path to indicate the new rightmost path. | |
328d5752 | 3578 | */ |
4fe82c31 | 3579 | static int ocfs2_merge_rec_left(struct ocfs2_path *right_path, |
328d5752 | 3580 | handle_t *handle, |
4fe82c31 | 3581 | struct ocfs2_extent_tree *et, |
328d5752 | 3582 | struct ocfs2_extent_rec *split_rec, |
677b9752 TM |
3583 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
3584 | int index) | |
328d5752 | 3585 | { |
677b9752 | 3586 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3587 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3588 | struct ocfs2_extent_rec *left_rec; | |
3589 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3590 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3591 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3592 | struct buffer_head *root_bh = NULL; | |
3593 | struct ocfs2_path *left_path = NULL; | |
3594 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3595 | |
677b9752 | 3596 | BUG_ON(index < 0); |
328d5752 | 3597 | |
328d5752 | 3598 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3599 | if (index == 0) { |
3600 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3601 | ret = ocfs2_get_left_path(et, right_path, &left_path); |
677b9752 TM |
3602 | if (ret) { |
3603 | mlog_errno(ret); | |
3604 | goto out; | |
3605 | } | |
3606 | ||
3607 | left_el = path_leaf_el(left_path); | |
3608 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3609 | le16_to_cpu(left_el->l_count)); | |
3610 | ||
3611 | left_rec = &left_el->l_recs[ | |
3612 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3613 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3614 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3615 | le32_to_cpu(split_rec->e_cpos)); | |
3616 | ||
7dc02805 JB |
3617 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3618 | right_path); | |
677b9752 TM |
3619 | |
3620 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3621 | handle->h_buffer_credits, | |
3622 | left_path); | |
3623 | if (ret) { | |
3624 | mlog_errno(ret); | |
3625 | goto out; | |
3626 | } | |
3627 | ||
3628 | root_bh = left_path->p_node[subtree_index].bh; | |
3629 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3630 | ||
4fe82c31 | 3631 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3632 | subtree_index); |
677b9752 TM |
3633 | if (ret) { |
3634 | mlog_errno(ret); | |
3635 | goto out; | |
3636 | } | |
3637 | ||
3638 | for (i = subtree_index + 1; | |
3639 | i < path_num_items(right_path); i++) { | |
4fe82c31 | 3640 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3641 | right_path, i); |
677b9752 TM |
3642 | if (ret) { |
3643 | mlog_errno(ret); | |
3644 | goto out; | |
3645 | } | |
3646 | ||
4fe82c31 | 3647 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3648 | left_path, i); |
677b9752 TM |
3649 | if (ret) { |
3650 | mlog_errno(ret); | |
3651 | goto out; | |
3652 | } | |
3653 | } | |
3654 | } else { | |
3655 | left_rec = &el->l_recs[index - 1]; | |
3656 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3657 | has_empty_extent = 1; | |
3658 | } | |
328d5752 | 3659 | |
4fe82c31 | 3660 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
9047beab | 3661 | path_num_items(right_path) - 1); |
328d5752 MF |
3662 | if (ret) { |
3663 | mlog_errno(ret); | |
3664 | goto out; | |
3665 | } | |
3666 | ||
3667 | if (has_empty_extent && index == 1) { | |
3668 | /* | |
3669 | * The easy case - we can just plop the record right in. | |
3670 | */ | |
3671 | *left_rec = *split_rec; | |
3672 | ||
3673 | has_empty_extent = 0; | |
677b9752 | 3674 | } else |
328d5752 | 3675 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3676 | |
3677 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3678 | le64_add_cpu(&right_rec->e_blkno, | |
4fe82c31 JB |
3679 | ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3680 | split_clusters)); | |
328d5752 MF |
3681 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); |
3682 | ||
3683 | ocfs2_cleanup_merge(el, index); | |
3684 | ||
3685 | ret = ocfs2_journal_dirty(handle, bh); | |
3686 | if (ret) | |
3687 | mlog_errno(ret); | |
3688 | ||
677b9752 TM |
3689 | if (left_path) { |
3690 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
3691 | if (ret) | |
3692 | mlog_errno(ret); | |
3693 | ||
3694 | /* | |
3695 | * In the situation that the right_rec is empty and the extent | |
3696 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3697 | * it and we need to delete the right extent block. | |
3698 | */ | |
3699 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3700 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3701 | ||
70f18c08 | 3702 | ret = ocfs2_remove_rightmost_path(handle, et, |
e7d4cb6b | 3703 | right_path, |
70f18c08 | 3704 | dealloc); |
677b9752 TM |
3705 | if (ret) { |
3706 | mlog_errno(ret); | |
3707 | goto out; | |
3708 | } | |
3709 | ||
3710 | /* Now the rightmost extent block has been deleted. | |
3711 | * So we use the new rightmost path. | |
3712 | */ | |
3713 | ocfs2_mv_path(right_path, left_path); | |
3714 | left_path = NULL; | |
3715 | } else | |
4619c73e | 3716 | ocfs2_complete_edge_insert(handle, left_path, |
677b9752 TM |
3717 | right_path, subtree_index); |
3718 | } | |
328d5752 | 3719 | out: |
677b9752 TM |
3720 | if (left_path) |
3721 | ocfs2_free_path(left_path); | |
328d5752 MF |
3722 | return ret; |
3723 | } | |
3724 | ||
c495dd24 JB |
3725 | static int ocfs2_try_to_merge_extent(handle_t *handle, |
3726 | struct ocfs2_extent_tree *et, | |
677b9752 | 3727 | struct ocfs2_path *path, |
328d5752 MF |
3728 | int split_index, |
3729 | struct ocfs2_extent_rec *split_rec, | |
3730 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
c495dd24 | 3731 | struct ocfs2_merge_ctxt *ctxt) |
328d5752 | 3732 | { |
518d7269 | 3733 | int ret = 0; |
677b9752 | 3734 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3735 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3736 | ||
3737 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3738 | ||
518d7269 TM |
3739 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3740 | /* | |
3741 | * The merge code will need to create an empty | |
3742 | * extent to take the place of the newly | |
3743 | * emptied slot. Remove any pre-existing empty | |
3744 | * extents - having more than one in a leaf is | |
3745 | * illegal. | |
3746 | */ | |
70f18c08 | 3747 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
518d7269 TM |
3748 | if (ret) { |
3749 | mlog_errno(ret); | |
3750 | goto out; | |
328d5752 | 3751 | } |
518d7269 TM |
3752 | split_index--; |
3753 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3754 | } |
3755 | ||
3756 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3757 | /* | |
3758 | * Left-right contig implies this. | |
3759 | */ | |
3760 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3761 | |
3762 | /* | |
3763 | * Since the leftright insert always covers the entire | |
3764 | * extent, this call will delete the insert record | |
3765 | * entirely, resulting in an empty extent record added to | |
3766 | * the extent block. | |
3767 | * | |
3768 | * Since the adding of an empty extent shifts | |
3769 | * everything back to the right, there's no need to | |
3770 | * update split_index here. | |
677b9752 TM |
3771 | * |
3772 | * When the split_index is zero, we need to merge it to the | |
3773 | * prevoius extent block. It is more efficient and easier | |
3774 | * if we do merge_right first and merge_left later. | |
328d5752 | 3775 | */ |
4fe82c31 | 3776 | ret = ocfs2_merge_rec_right(path, handle, et, split_rec, |
677b9752 | 3777 | split_index); |
328d5752 MF |
3778 | if (ret) { |
3779 | mlog_errno(ret); | |
3780 | goto out; | |
3781 | } | |
3782 | ||
3783 | /* | |
3784 | * We can only get this from logic error above. | |
3785 | */ | |
3786 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3787 | ||
677b9752 | 3788 | /* The merge left us with an empty extent, remove it. */ |
70f18c08 | 3789 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3790 | if (ret) { |
3791 | mlog_errno(ret); | |
3792 | goto out; | |
3793 | } | |
677b9752 | 3794 | |
328d5752 MF |
3795 | rec = &el->l_recs[split_index]; |
3796 | ||
3797 | /* | |
3798 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3799 | * we've merged it into the rec already. |
328d5752 | 3800 | */ |
4fe82c31 JB |
3801 | ret = ocfs2_merge_rec_left(path, handle, et, rec, |
3802 | dealloc, split_index); | |
677b9752 | 3803 | |
328d5752 MF |
3804 | if (ret) { |
3805 | mlog_errno(ret); | |
3806 | goto out; | |
3807 | } | |
3808 | ||
70f18c08 | 3809 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3810 | /* |
3811 | * Error from this last rotate is not critical, so | |
3812 | * print but don't bubble it up. | |
3813 | */ | |
3814 | if (ret) | |
3815 | mlog_errno(ret); | |
3816 | ret = 0; | |
3817 | } else { | |
3818 | /* | |
3819 | * Merge a record to the left or right. | |
3820 | * | |
3821 | * 'contig_type' is relative to the existing record, | |
3822 | * so for example, if we're "right contig", it's to | |
3823 | * the record on the left (hence the left merge). | |
3824 | */ | |
3825 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
4fe82c31 JB |
3826 | ret = ocfs2_merge_rec_left(path, handle, et, |
3827 | split_rec, dealloc, | |
328d5752 MF |
3828 | split_index); |
3829 | if (ret) { | |
3830 | mlog_errno(ret); | |
3831 | goto out; | |
3832 | } | |
3833 | } else { | |
4fe82c31 | 3834 | ret = ocfs2_merge_rec_right(path, handle, |
7dc02805 | 3835 | et, split_rec, |
328d5752 MF |
3836 | split_index); |
3837 | if (ret) { | |
3838 | mlog_errno(ret); | |
3839 | goto out; | |
3840 | } | |
3841 | } | |
3842 | ||
3843 | if (ctxt->c_split_covers_rec) { | |
3844 | /* | |
3845 | * The merge may have left an empty extent in | |
3846 | * our leaf. Try to rotate it away. | |
3847 | */ | |
70f18c08 JB |
3848 | ret = ocfs2_rotate_tree_left(handle, et, path, |
3849 | dealloc); | |
328d5752 MF |
3850 | if (ret) |
3851 | mlog_errno(ret); | |
3852 | ret = 0; | |
3853 | } | |
3854 | } | |
3855 | ||
3856 | out: | |
3857 | return ret; | |
3858 | } | |
3859 | ||
3860 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3861 | enum ocfs2_split_type split, | |
3862 | struct ocfs2_extent_rec *rec, | |
3863 | struct ocfs2_extent_rec *split_rec) | |
3864 | { | |
3865 | u64 len_blocks; | |
3866 | ||
3867 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3868 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3869 | ||
3870 | if (split == SPLIT_LEFT) { | |
3871 | /* | |
3872 | * Region is on the left edge of the existing | |
3873 | * record. | |
3874 | */ | |
3875 | le32_add_cpu(&rec->e_cpos, | |
3876 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3877 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3878 | le16_add_cpu(&rec->e_leaf_clusters, | |
3879 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3880 | } else { | |
3881 | /* | |
3882 | * Region is on the right edge of the existing | |
3883 | * record. | |
3884 | */ | |
3885 | le16_add_cpu(&rec->e_leaf_clusters, | |
3886 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3887 | } | |
3888 | } | |
3889 | ||
3890 | /* | |
3891 | * Do the final bits of extent record insertion at the target leaf | |
3892 | * list. If this leaf is part of an allocation tree, it is assumed | |
3893 | * that the tree above has been prepared. | |
3894 | */ | |
d5628623 JB |
3895 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et, |
3896 | struct ocfs2_extent_rec *insert_rec, | |
328d5752 | 3897 | struct ocfs2_extent_list *el, |
d5628623 | 3898 | struct ocfs2_insert_type *insert) |
328d5752 MF |
3899 | { |
3900 | int i = insert->ins_contig_index; | |
3901 | unsigned int range; | |
3902 | struct ocfs2_extent_rec *rec; | |
3903 | ||
3904 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3905 | ||
3906 | if (insert->ins_split != SPLIT_NONE) { | |
3907 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3908 | BUG_ON(i == -1); | |
3909 | rec = &el->l_recs[i]; | |
d5628623 JB |
3910 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
3911 | insert->ins_split, rec, | |
328d5752 MF |
3912 | insert_rec); |
3913 | goto rotate; | |
3914 | } | |
3915 | ||
3916 | /* | |
3917 | * Contiguous insert - either left or right. | |
3918 | */ | |
3919 | if (insert->ins_contig != CONTIG_NONE) { | |
3920 | rec = &el->l_recs[i]; | |
3921 | if (insert->ins_contig == CONTIG_LEFT) { | |
3922 | rec->e_blkno = insert_rec->e_blkno; | |
3923 | rec->e_cpos = insert_rec->e_cpos; | |
3924 | } | |
3925 | le16_add_cpu(&rec->e_leaf_clusters, | |
3926 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3927 | return; | |
3928 | } | |
3929 | ||
3930 | /* | |
3931 | * Handle insert into an empty leaf. | |
3932 | */ | |
3933 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3934 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3935 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3936 | el->l_recs[0] = *insert_rec; | |
3937 | el->l_next_free_rec = cpu_to_le16(1); | |
3938 | return; | |
3939 | } | |
3940 | ||
3941 | /* | |
3942 | * Appending insert. | |
3943 | */ | |
3944 | if (insert->ins_appending == APPEND_TAIL) { | |
3945 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3946 | rec = &el->l_recs[i]; | |
3947 | range = le32_to_cpu(rec->e_cpos) | |
3948 | + le16_to_cpu(rec->e_leaf_clusters); | |
3949 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3950 | ||
3951 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3952 | le16_to_cpu(el->l_count), | |
d5628623 | 3953 | "owner %llu, depth %u, count %u, next free %u, " |
328d5752 MF |
3954 | "rec.cpos %u, rec.clusters %u, " |
3955 | "insert.cpos %u, insert.clusters %u\n", | |
d5628623 | 3956 | ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
3957 | le16_to_cpu(el->l_tree_depth), |
3958 | le16_to_cpu(el->l_count), | |
3959 | le16_to_cpu(el->l_next_free_rec), | |
3960 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3961 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3962 | le32_to_cpu(insert_rec->e_cpos), | |
3963 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3964 | i++; | |
3965 | el->l_recs[i] = *insert_rec; | |
3966 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3967 | return; | |
3968 | } | |
3969 | ||
3970 | rotate: | |
3971 | /* | |
3972 | * Ok, we have to rotate. | |
3973 | * | |
3974 | * At this point, it is safe to assume that inserting into an | |
3975 | * empty leaf and appending to a leaf have both been handled | |
3976 | * above. | |
3977 | * | |
3978 | * This leaf needs to have space, either by the empty 1st | |
3979 | * extent record, or by virtue of an l_next_rec < l_count. | |
3980 | */ | |
3981 | ocfs2_rotate_leaf(el, insert_rec); | |
3982 | } | |
3983 | ||
d401dc12 JB |
3984 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
3985 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3986 | struct ocfs2_path *path, |
3987 | struct ocfs2_extent_rec *insert_rec) | |
3988 | { | |
3989 | int ret, i, next_free; | |
3990 | struct buffer_head *bh; | |
3991 | struct ocfs2_extent_list *el; | |
3992 | struct ocfs2_extent_rec *rec; | |
3993 | ||
3994 | /* | |
3995 | * Update everything except the leaf block. | |
3996 | */ | |
3997 | for (i = 0; i < path->p_tree_depth; i++) { | |
3998 | bh = path->p_node[i].bh; | |
3999 | el = path->p_node[i].el; | |
4000 | ||
dcd0538f MF |
4001 | next_free = le16_to_cpu(el->l_next_free_rec); |
4002 | if (next_free == 0) { | |
d401dc12 JB |
4003 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
4004 | "Owner %llu has a bad extent list", | |
4005 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 4006 | ret = -EIO; |
328d5752 MF |
4007 | return; |
4008 | } | |
4009 | ||
4010 | rec = &el->l_recs[next_free - 1]; | |
4011 | ||
4012 | rec->e_int_clusters = insert_rec->e_cpos; | |
4013 | le32_add_cpu(&rec->e_int_clusters, | |
4014 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
4015 | le32_add_cpu(&rec->e_int_clusters, | |
4016 | -le32_to_cpu(rec->e_cpos)); | |
4017 | ||
4018 | ret = ocfs2_journal_dirty(handle, bh); | |
4019 | if (ret) | |
4020 | mlog_errno(ret); | |
4021 | ||
4022 | } | |
4023 | } | |
4024 | ||
d401dc12 JB |
4025 | static int ocfs2_append_rec_to_path(handle_t *handle, |
4026 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
4027 | struct ocfs2_extent_rec *insert_rec, |
4028 | struct ocfs2_path *right_path, | |
4029 | struct ocfs2_path **ret_left_path) | |
4030 | { | |
4031 | int ret, next_free; | |
4032 | struct ocfs2_extent_list *el; | |
4033 | struct ocfs2_path *left_path = NULL; | |
4034 | ||
4035 | *ret_left_path = NULL; | |
4036 | ||
4037 | /* | |
4038 | * This shouldn't happen for non-trees. The extent rec cluster | |
4039 | * count manipulation below only works for interior nodes. | |
4040 | */ | |
4041 | BUG_ON(right_path->p_tree_depth == 0); | |
4042 | ||
4043 | /* | |
4044 | * If our appending insert is at the leftmost edge of a leaf, | |
4045 | * then we might need to update the rightmost records of the | |
4046 | * neighboring path. | |
4047 | */ | |
4048 | el = path_leaf_el(right_path); | |
4049 | next_free = le16_to_cpu(el->l_next_free_rec); | |
4050 | if (next_free == 0 || | |
4051 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
4052 | u32 left_cpos; | |
4053 | ||
d401dc12 JB |
4054 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
4055 | right_path, &left_cpos); | |
328d5752 MF |
4056 | if (ret) { |
4057 | mlog_errno(ret); | |
dcd0538f MF |
4058 | goto out; |
4059 | } | |
4060 | ||
328d5752 MF |
4061 | mlog(0, "Append may need a left path update. cpos: %u, " |
4062 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
4063 | left_cpos); | |
e48edee2 | 4064 | |
328d5752 MF |
4065 | /* |
4066 | * No need to worry if the append is already in the | |
4067 | * leftmost leaf. | |
4068 | */ | |
4069 | if (left_cpos) { | |
ffdd7a54 | 4070 | left_path = ocfs2_new_path_from_path(right_path); |
328d5752 MF |
4071 | if (!left_path) { |
4072 | ret = -ENOMEM; | |
4073 | mlog_errno(ret); | |
4074 | goto out; | |
4075 | } | |
dcd0538f | 4076 | |
d401dc12 | 4077 | ret = ocfs2_find_path(et->et_ci, left_path, |
facdb77f | 4078 | left_cpos); |
328d5752 MF |
4079 | if (ret) { |
4080 | mlog_errno(ret); | |
4081 | goto out; | |
4082 | } | |
dcd0538f | 4083 | |
328d5752 MF |
4084 | /* |
4085 | * ocfs2_insert_path() will pass the left_path to the | |
4086 | * journal for us. | |
4087 | */ | |
4088 | } | |
4089 | } | |
dcd0538f | 4090 | |
d401dc12 | 4091 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
328d5752 MF |
4092 | if (ret) { |
4093 | mlog_errno(ret); | |
4094 | goto out; | |
dcd0538f MF |
4095 | } |
4096 | ||
d401dc12 | 4097 | ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec); |
328d5752 | 4098 | |
dcd0538f MF |
4099 | *ret_left_path = left_path; |
4100 | ret = 0; | |
4101 | out: | |
4102 | if (ret != 0) | |
4103 | ocfs2_free_path(left_path); | |
4104 | ||
4105 | return ret; | |
4106 | } | |
4107 | ||
c38e52bb | 4108 | static void ocfs2_split_record(struct ocfs2_extent_tree *et, |
328d5752 MF |
4109 | struct ocfs2_path *left_path, |
4110 | struct ocfs2_path *right_path, | |
4111 | struct ocfs2_extent_rec *split_rec, | |
4112 | enum ocfs2_split_type split) | |
4113 | { | |
4114 | int index; | |
4115 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
4116 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
4117 | struct ocfs2_extent_rec *rec, *tmprec; | |
4118 | ||
c19a28e1 | 4119 | right_el = path_leaf_el(right_path); |
328d5752 MF |
4120 | if (left_path) |
4121 | left_el = path_leaf_el(left_path); | |
4122 | ||
4123 | el = right_el; | |
4124 | insert_el = right_el; | |
4125 | index = ocfs2_search_extent_list(el, cpos); | |
4126 | if (index != -1) { | |
4127 | if (index == 0 && left_path) { | |
4128 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
4129 | ||
4130 | /* | |
4131 | * This typically means that the record | |
4132 | * started in the left path but moved to the | |
4133 | * right as a result of rotation. We either | |
4134 | * move the existing record to the left, or we | |
4135 | * do the later insert there. | |
4136 | * | |
4137 | * In this case, the left path should always | |
4138 | * exist as the rotate code will have passed | |
4139 | * it back for a post-insert update. | |
4140 | */ | |
4141 | ||
4142 | if (split == SPLIT_LEFT) { | |
4143 | /* | |
4144 | * It's a left split. Since we know | |
4145 | * that the rotate code gave us an | |
4146 | * empty extent in the left path, we | |
4147 | * can just do the insert there. | |
4148 | */ | |
4149 | insert_el = left_el; | |
4150 | } else { | |
4151 | /* | |
4152 | * Right split - we have to move the | |
4153 | * existing record over to the left | |
4154 | * leaf. The insert will be into the | |
4155 | * newly created empty extent in the | |
4156 | * right leaf. | |
4157 | */ | |
4158 | tmprec = &right_el->l_recs[index]; | |
4159 | ocfs2_rotate_leaf(left_el, tmprec); | |
4160 | el = left_el; | |
4161 | ||
4162 | memset(tmprec, 0, sizeof(*tmprec)); | |
4163 | index = ocfs2_search_extent_list(left_el, cpos); | |
4164 | BUG_ON(index == -1); | |
4165 | } | |
4166 | } | |
4167 | } else { | |
4168 | BUG_ON(!left_path); | |
4169 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
4170 | /* | |
4171 | * Left path is easy - we can just allow the insert to | |
4172 | * happen. | |
4173 | */ | |
4174 | el = left_el; | |
4175 | insert_el = left_el; | |
4176 | index = ocfs2_search_extent_list(el, cpos); | |
4177 | BUG_ON(index == -1); | |
4178 | } | |
4179 | ||
4180 | rec = &el->l_recs[index]; | |
c38e52bb JB |
4181 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4182 | split, rec, split_rec); | |
328d5752 MF |
4183 | ocfs2_rotate_leaf(insert_el, split_rec); |
4184 | } | |
4185 | ||
dcd0538f | 4186 | /* |
e7d4cb6b TM |
4187 | * This function only does inserts on an allocation b-tree. For tree |
4188 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
4189 | * |
4190 | * right_path is the path we want to do the actual insert | |
4191 | * in. left_path should only be passed in if we need to update that | |
4192 | * portion of the tree after an edge insert. | |
4193 | */ | |
3505bec0 | 4194 | static int ocfs2_insert_path(handle_t *handle, |
7dc02805 | 4195 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4196 | struct ocfs2_path *left_path, |
4197 | struct ocfs2_path *right_path, | |
4198 | struct ocfs2_extent_rec *insert_rec, | |
4199 | struct ocfs2_insert_type *insert) | |
4200 | { | |
4201 | int ret, subtree_index; | |
4202 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 4203 | |
dcd0538f MF |
4204 | if (left_path) { |
4205 | int credits = handle->h_buffer_credits; | |
4206 | ||
4207 | /* | |
4208 | * There's a chance that left_path got passed back to | |
4209 | * us without being accounted for in the | |
4210 | * journal. Extend our transaction here to be sure we | |
4211 | * can change those blocks. | |
4212 | */ | |
4213 | credits += left_path->p_tree_depth; | |
4214 | ||
4215 | ret = ocfs2_extend_trans(handle, credits); | |
4216 | if (ret < 0) { | |
4217 | mlog_errno(ret); | |
4218 | goto out; | |
4219 | } | |
4220 | ||
7dc02805 | 4221 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
dcd0538f MF |
4222 | if (ret < 0) { |
4223 | mlog_errno(ret); | |
4224 | goto out; | |
4225 | } | |
4226 | } | |
4227 | ||
e8aed345 MF |
4228 | /* |
4229 | * Pass both paths to the journal. The majority of inserts | |
4230 | * will be touching all components anyway. | |
4231 | */ | |
7dc02805 | 4232 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
e8aed345 MF |
4233 | if (ret < 0) { |
4234 | mlog_errno(ret); | |
4235 | goto out; | |
4236 | } | |
4237 | ||
328d5752 MF |
4238 | if (insert->ins_split != SPLIT_NONE) { |
4239 | /* | |
4240 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 4241 | * of splits, but it's easier to just let one separate |
328d5752 MF |
4242 | * function sort it all out. |
4243 | */ | |
c38e52bb | 4244 | ocfs2_split_record(et, left_path, right_path, |
328d5752 | 4245 | insert_rec, insert->ins_split); |
e8aed345 MF |
4246 | |
4247 | /* | |
4248 | * Split might have modified either leaf and we don't | |
4249 | * have a guarantee that the later edge insert will | |
4250 | * dirty this for us. | |
4251 | */ | |
4252 | if (left_path) | |
4253 | ret = ocfs2_journal_dirty(handle, | |
4254 | path_leaf_bh(left_path)); | |
4255 | if (ret) | |
4256 | mlog_errno(ret); | |
328d5752 | 4257 | } else |
d5628623 JB |
4258 | ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path), |
4259 | insert); | |
dcd0538f | 4260 | |
dcd0538f MF |
4261 | ret = ocfs2_journal_dirty(handle, leaf_bh); |
4262 | if (ret) | |
4263 | mlog_errno(ret); | |
4264 | ||
4265 | if (left_path) { | |
4266 | /* | |
4267 | * The rotate code has indicated that we need to fix | |
4268 | * up portions of the tree after the insert. | |
4269 | * | |
4270 | * XXX: Should we extend the transaction here? | |
4271 | */ | |
7dc02805 | 4272 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
dcd0538f | 4273 | right_path); |
4619c73e JB |
4274 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
4275 | subtree_index); | |
dcd0538f MF |
4276 | } |
4277 | ||
4278 | ret = 0; | |
4279 | out: | |
4280 | return ret; | |
4281 | } | |
4282 | ||
3505bec0 | 4283 | static int ocfs2_do_insert_extent(handle_t *handle, |
e7d4cb6b | 4284 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4285 | struct ocfs2_extent_rec *insert_rec, |
4286 | struct ocfs2_insert_type *type) | |
4287 | { | |
4288 | int ret, rotate = 0; | |
4289 | u32 cpos; | |
4290 | struct ocfs2_path *right_path = NULL; | |
4291 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
4292 | struct ocfs2_extent_list *el; |
4293 | ||
ce1d9ea6 | 4294 | el = et->et_root_el; |
dcd0538f | 4295 | |
d9a0a1f8 | 4296 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4297 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
4298 | if (ret) { |
4299 | mlog_errno(ret); | |
4300 | goto out; | |
4301 | } | |
4302 | ||
4303 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
d5628623 | 4304 | ocfs2_insert_at_leaf(et, insert_rec, el, type); |
dcd0538f MF |
4305 | goto out_update_clusters; |
4306 | } | |
4307 | ||
ffdd7a54 | 4308 | right_path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4309 | if (!right_path) { |
4310 | ret = -ENOMEM; | |
4311 | mlog_errno(ret); | |
4312 | goto out; | |
4313 | } | |
4314 | ||
4315 | /* | |
4316 | * Determine the path to start with. Rotations need the | |
4317 | * rightmost path, everything else can go directly to the | |
4318 | * target leaf. | |
4319 | */ | |
4320 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
4321 | if (type->ins_appending == APPEND_NONE && | |
4322 | type->ins_contig == CONTIG_NONE) { | |
4323 | rotate = 1; | |
4324 | cpos = UINT_MAX; | |
4325 | } | |
4326 | ||
facdb77f | 4327 | ret = ocfs2_find_path(et->et_ci, right_path, cpos); |
dcd0538f MF |
4328 | if (ret) { |
4329 | mlog_errno(ret); | |
4330 | goto out; | |
4331 | } | |
4332 | ||
4333 | /* | |
4334 | * Rotations and appends need special treatment - they modify | |
4335 | * parts of the tree's above them. | |
4336 | * | |
4337 | * Both might pass back a path immediate to the left of the | |
4338 | * one being inserted to. This will be cause | |
4339 | * ocfs2_insert_path() to modify the rightmost records of | |
4340 | * left_path to account for an edge insert. | |
4341 | * | |
4342 | * XXX: When modifying this code, keep in mind that an insert | |
4343 | * can wind up skipping both of these two special cases... | |
4344 | */ | |
4345 | if (rotate) { | |
1bbf0b8d | 4346 | ret = ocfs2_rotate_tree_right(handle, et, type->ins_split, |
dcd0538f MF |
4347 | le32_to_cpu(insert_rec->e_cpos), |
4348 | right_path, &left_path); | |
4349 | if (ret) { | |
4350 | mlog_errno(ret); | |
4351 | goto out; | |
4352 | } | |
e8aed345 MF |
4353 | |
4354 | /* | |
4355 | * ocfs2_rotate_tree_right() might have extended the | |
4356 | * transaction without re-journaling our tree root. | |
4357 | */ | |
d9a0a1f8 | 4358 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4359 | OCFS2_JOURNAL_ACCESS_WRITE); |
e8aed345 MF |
4360 | if (ret) { |
4361 | mlog_errno(ret); | |
4362 | goto out; | |
4363 | } | |
dcd0538f MF |
4364 | } else if (type->ins_appending == APPEND_TAIL |
4365 | && type->ins_contig != CONTIG_LEFT) { | |
d401dc12 | 4366 | ret = ocfs2_append_rec_to_path(handle, et, insert_rec, |
dcd0538f MF |
4367 | right_path, &left_path); |
4368 | if (ret) { | |
4369 | mlog_errno(ret); | |
4370 | goto out; | |
4371 | } | |
4372 | } | |
4373 | ||
3505bec0 | 4374 | ret = ocfs2_insert_path(handle, et, left_path, right_path, |
dcd0538f MF |
4375 | insert_rec, type); |
4376 | if (ret) { | |
4377 | mlog_errno(ret); | |
4378 | goto out; | |
4379 | } | |
4380 | ||
4381 | out_update_clusters: | |
328d5752 | 4382 | if (type->ins_split == SPLIT_NONE) |
6136ca5f | 4383 | ocfs2_et_update_clusters(et, |
35dc0aa3 | 4384 | le16_to_cpu(insert_rec->e_leaf_clusters)); |
dcd0538f | 4385 | |
ce1d9ea6 | 4386 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4387 | if (ret) |
4388 | mlog_errno(ret); | |
4389 | ||
4390 | out: | |
4391 | ocfs2_free_path(left_path); | |
4392 | ocfs2_free_path(right_path); | |
4393 | ||
4394 | return ret; | |
4395 | } | |
4396 | ||
328d5752 | 4397 | static enum ocfs2_contig_type |
a2970291 JB |
4398 | ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et, |
4399 | struct ocfs2_path *path, | |
328d5752 MF |
4400 | struct ocfs2_extent_list *el, int index, |
4401 | struct ocfs2_extent_rec *split_rec) | |
4402 | { | |
ad5a4d70 | 4403 | int status; |
328d5752 | 4404 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4405 | u32 left_cpos, right_cpos; |
4406 | struct ocfs2_extent_rec *rec = NULL; | |
4407 | struct ocfs2_extent_list *new_el; | |
4408 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4409 | struct buffer_head *bh; | |
4410 | struct ocfs2_extent_block *eb; | |
a2970291 | 4411 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
ad5a4d70 TM |
4412 | |
4413 | if (index > 0) { | |
4414 | rec = &el->l_recs[index - 1]; | |
4415 | } else if (path->p_tree_depth > 0) { | |
a2970291 | 4416 | status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
ad5a4d70 TM |
4417 | if (status) |
4418 | goto out; | |
4419 | ||
4420 | if (left_cpos != 0) { | |
ffdd7a54 | 4421 | left_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4422 | if (!left_path) |
4423 | goto out; | |
4424 | ||
a2970291 JB |
4425 | status = ocfs2_find_path(et->et_ci, left_path, |
4426 | left_cpos); | |
ad5a4d70 TM |
4427 | if (status) |
4428 | goto out; | |
4429 | ||
4430 | new_el = path_leaf_el(left_path); | |
4431 | ||
4432 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4433 | le16_to_cpu(new_el->l_count)) { | |
4434 | bh = path_leaf_bh(left_path); | |
4435 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4436 | ocfs2_error(sb, |
5e96581a JB |
4437 | "Extent block #%llu has an " |
4438 | "invalid l_next_free_rec of " | |
4439 | "%d. It should have " | |
4440 | "matched the l_count of %d", | |
4441 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4442 | le16_to_cpu(new_el->l_next_free_rec), | |
4443 | le16_to_cpu(new_el->l_count)); | |
4444 | status = -EINVAL; | |
ad5a4d70 TM |
4445 | goto out; |
4446 | } | |
4447 | rec = &new_el->l_recs[ | |
4448 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4449 | } | |
4450 | } | |
328d5752 MF |
4451 | |
4452 | /* | |
4453 | * We're careful to check for an empty extent record here - | |
4454 | * the merge code will know what to do if it sees one. | |
4455 | */ | |
ad5a4d70 | 4456 | if (rec) { |
328d5752 MF |
4457 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4458 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4459 | ret = CONTIG_RIGHT; | |
4460 | } else { | |
853a3a14 | 4461 | ret = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4462 | } |
4463 | } | |
4464 | ||
ad5a4d70 TM |
4465 | rec = NULL; |
4466 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4467 | rec = &el->l_recs[index + 1]; | |
4468 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4469 | path->p_tree_depth > 0) { | |
a2970291 | 4470 | status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
ad5a4d70 TM |
4471 | if (status) |
4472 | goto out; | |
4473 | ||
4474 | if (right_cpos == 0) | |
4475 | goto out; | |
4476 | ||
ffdd7a54 | 4477 | right_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4478 | if (!right_path) |
4479 | goto out; | |
4480 | ||
a2970291 | 4481 | status = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
ad5a4d70 TM |
4482 | if (status) |
4483 | goto out; | |
4484 | ||
4485 | new_el = path_leaf_el(right_path); | |
4486 | rec = &new_el->l_recs[0]; | |
4487 | if (ocfs2_is_empty_extent(rec)) { | |
4488 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4489 | bh = path_leaf_bh(right_path); | |
4490 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4491 | ocfs2_error(sb, |
5e96581a JB |
4492 | "Extent block #%llu has an " |
4493 | "invalid l_next_free_rec of %d", | |
4494 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4495 | le16_to_cpu(new_el->l_next_free_rec)); | |
4496 | status = -EINVAL; | |
ad5a4d70 TM |
4497 | goto out; |
4498 | } | |
4499 | rec = &new_el->l_recs[1]; | |
4500 | } | |
4501 | } | |
4502 | ||
4503 | if (rec) { | |
328d5752 MF |
4504 | enum ocfs2_contig_type contig_type; |
4505 | ||
853a3a14 | 4506 | contig_type = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4507 | |
4508 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4509 | ret = CONTIG_LEFTRIGHT; | |
4510 | else if (ret == CONTIG_NONE) | |
4511 | ret = contig_type; | |
4512 | } | |
4513 | ||
ad5a4d70 TM |
4514 | out: |
4515 | if (left_path) | |
4516 | ocfs2_free_path(left_path); | |
4517 | if (right_path) | |
4518 | ocfs2_free_path(right_path); | |
4519 | ||
328d5752 MF |
4520 | return ret; |
4521 | } | |
4522 | ||
1ef61b33 | 4523 | static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4524 | struct ocfs2_insert_type *insert, |
4525 | struct ocfs2_extent_list *el, | |
1ef61b33 | 4526 | struct ocfs2_extent_rec *insert_rec) |
dcd0538f MF |
4527 | { |
4528 | int i; | |
4529 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4530 | ||
e48edee2 MF |
4531 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4532 | ||
dcd0538f | 4533 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
853a3a14 TM |
4534 | contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i], |
4535 | insert_rec); | |
dcd0538f MF |
4536 | if (contig_type != CONTIG_NONE) { |
4537 | insert->ins_contig_index = i; | |
4538 | break; | |
4539 | } | |
4540 | } | |
4541 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4542 | |
4543 | if (insert->ins_contig != CONTIG_NONE) { | |
4544 | struct ocfs2_extent_rec *rec = | |
4545 | &el->l_recs[insert->ins_contig_index]; | |
4546 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4547 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4548 | ||
4549 | /* | |
4550 | * Caller might want us to limit the size of extents, don't | |
4551 | * calculate contiguousness if we might exceed that limit. | |
4552 | */ | |
ce1d9ea6 JB |
4553 | if (et->et_max_leaf_clusters && |
4554 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4555 | insert->ins_contig = CONTIG_NONE; |
4556 | } | |
dcd0538f MF |
4557 | } |
4558 | ||
4559 | /* | |
4560 | * This should only be called against the righmost leaf extent list. | |
4561 | * | |
4562 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4563 | * insert at the tail of the rightmost leaf. | |
4564 | * | |
e7d4cb6b TM |
4565 | * This should also work against the root extent list for tree's with 0 |
4566 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4567 | * then the logic here makes sense. |
4568 | */ | |
4569 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4570 | struct ocfs2_extent_list *el, | |
4571 | struct ocfs2_extent_rec *insert_rec) | |
4572 | { | |
4573 | int i; | |
4574 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4575 | struct ocfs2_extent_rec *rec; | |
4576 | ||
4577 | insert->ins_appending = APPEND_NONE; | |
4578 | ||
e48edee2 | 4579 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4580 | |
4581 | if (!el->l_next_free_rec) | |
4582 | goto set_tail_append; | |
4583 | ||
4584 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4585 | /* Were all records empty? */ | |
4586 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4587 | goto set_tail_append; | |
4588 | } | |
4589 | ||
4590 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4591 | rec = &el->l_recs[i]; | |
4592 | ||
e48edee2 MF |
4593 | if (cpos >= |
4594 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4595 | goto set_tail_append; |
4596 | ||
4597 | return; | |
4598 | ||
4599 | set_tail_append: | |
4600 | insert->ins_appending = APPEND_TAIL; | |
4601 | } | |
4602 | ||
4603 | /* | |
4604 | * Helper function called at the begining of an insert. | |
4605 | * | |
4606 | * This computes a few things that are commonly used in the process of | |
4607 | * inserting into the btree: | |
4608 | * - Whether the new extent is contiguous with an existing one. | |
4609 | * - The current tree depth. | |
4610 | * - Whether the insert is an appending one. | |
4611 | * - The total # of free records in the tree. | |
4612 | * | |
4613 | * All of the information is stored on the ocfs2_insert_type | |
4614 | * structure. | |
4615 | */ | |
627961b7 | 4616 | static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4617 | struct buffer_head **last_eb_bh, |
4618 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4619 | int *free_records, |
dcd0538f MF |
4620 | struct ocfs2_insert_type *insert) |
4621 | { | |
4622 | int ret; | |
dcd0538f MF |
4623 | struct ocfs2_extent_block *eb; |
4624 | struct ocfs2_extent_list *el; | |
4625 | struct ocfs2_path *path = NULL; | |
4626 | struct buffer_head *bh = NULL; | |
4627 | ||
328d5752 MF |
4628 | insert->ins_split = SPLIT_NONE; |
4629 | ||
ce1d9ea6 | 4630 | el = et->et_root_el; |
dcd0538f MF |
4631 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4632 | ||
4633 | if (el->l_tree_depth) { | |
4634 | /* | |
4635 | * If we have tree depth, we read in the | |
4636 | * rightmost extent block ahead of time as | |
4637 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4638 | * may want it later. | |
4639 | */ | |
3d03a305 | 4640 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
4641 | ocfs2_et_get_last_eb_blk(et), |
4642 | &bh); | |
dcd0538f MF |
4643 | if (ret) { |
4644 | mlog_exit(ret); | |
4645 | goto out; | |
4646 | } | |
ccd979bd | 4647 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4648 | el = &eb->h_list; |
dcd0538f | 4649 | } |
ccd979bd | 4650 | |
dcd0538f MF |
4651 | /* |
4652 | * Unless we have a contiguous insert, we'll need to know if | |
4653 | * there is room left in our allocation tree for another | |
4654 | * extent record. | |
4655 | * | |
4656 | * XXX: This test is simplistic, we can search for empty | |
4657 | * extent records too. | |
4658 | */ | |
c77534f6 | 4659 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4660 | le16_to_cpu(el->l_next_free_rec); |
4661 | ||
4662 | if (!insert->ins_tree_depth) { | |
1ef61b33 | 4663 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4664 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4665 | return 0; | |
ccd979bd MF |
4666 | } |
4667 | ||
ffdd7a54 | 4668 | path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4669 | if (!path) { |
4670 | ret = -ENOMEM; | |
4671 | mlog_errno(ret); | |
4672 | goto out; | |
4673 | } | |
ccd979bd | 4674 | |
dcd0538f MF |
4675 | /* |
4676 | * In the case that we're inserting past what the tree | |
4677 | * currently accounts for, ocfs2_find_path() will return for | |
4678 | * us the rightmost tree path. This is accounted for below in | |
4679 | * the appending code. | |
4680 | */ | |
facdb77f | 4681 | ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos)); |
dcd0538f MF |
4682 | if (ret) { |
4683 | mlog_errno(ret); | |
4684 | goto out; | |
4685 | } | |
ccd979bd | 4686 | |
dcd0538f MF |
4687 | el = path_leaf_el(path); |
4688 | ||
4689 | /* | |
4690 | * Now that we have the path, there's two things we want to determine: | |
4691 | * 1) Contiguousness (also set contig_index if this is so) | |
4692 | * | |
4693 | * 2) Are we doing an append? We can trivially break this up | |
4694 | * into two types of appends: simple record append, or a | |
4695 | * rotate inside the tail leaf. | |
4696 | */ | |
1ef61b33 | 4697 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4698 | |
4699 | /* | |
4700 | * The insert code isn't quite ready to deal with all cases of | |
4701 | * left contiguousness. Specifically, if it's an insert into | |
4702 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4703 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4704 | * left. For now, just catch that case and fool the layers |
4705 | * above us. This works just fine for tree_depth == 0, which | |
4706 | * is why we allow that above. | |
4707 | */ | |
4708 | if (insert->ins_contig == CONTIG_LEFT && | |
4709 | insert->ins_contig_index == 0) | |
4710 | insert->ins_contig = CONTIG_NONE; | |
4711 | ||
4712 | /* | |
4713 | * Ok, so we can simply compare against last_eb to figure out | |
4714 | * whether the path doesn't exist. This will only happen in | |
4715 | * the case that we're doing a tail append, so maybe we can | |
4716 | * take advantage of that information somehow. | |
4717 | */ | |
35dc0aa3 | 4718 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4719 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4720 | /* |
4721 | * Ok, ocfs2_find_path() returned us the rightmost | |
4722 | * tree path. This might be an appending insert. There are | |
4723 | * two cases: | |
4724 | * 1) We're doing a true append at the tail: | |
4725 | * -This might even be off the end of the leaf | |
4726 | * 2) We're "appending" by rotating in the tail | |
4727 | */ | |
4728 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4729 | } | |
4730 | ||
4731 | out: | |
4732 | ocfs2_free_path(path); | |
4733 | ||
4734 | if (ret == 0) | |
4735 | *last_eb_bh = bh; | |
4736 | else | |
4737 | brelse(bh); | |
4738 | return ret; | |
ccd979bd MF |
4739 | } |
4740 | ||
dcd0538f | 4741 | /* |
cc79d8c1 | 4742 | * Insert an extent into a btree. |
dcd0538f | 4743 | * |
cc79d8c1 | 4744 | * The caller needs to update the owning btree's cluster count. |
dcd0538f | 4745 | */ |
cc79d8c1 | 4746 | int ocfs2_insert_extent(handle_t *handle, |
f99b9b7c JB |
4747 | struct ocfs2_extent_tree *et, |
4748 | u32 cpos, | |
4749 | u64 start_blk, | |
4750 | u32 new_clusters, | |
4751 | u8 flags, | |
4752 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4753 | { |
c3afcbb3 | 4754 | int status; |
c77534f6 | 4755 | int uninitialized_var(free_records); |
ccd979bd | 4756 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4757 | struct ocfs2_insert_type insert = {0, }; |
4758 | struct ocfs2_extent_rec rec; | |
4759 | ||
cc79d8c1 JB |
4760 | mlog(0, "add %u clusters at position %u to owner %llu\n", |
4761 | new_clusters, cpos, | |
4762 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 4763 | |
e48edee2 | 4764 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4765 | rec.e_cpos = cpu_to_le32(cpos); |
4766 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4767 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4768 | rec.e_flags = flags; |
6136ca5f | 4769 | status = ocfs2_et_insert_check(et, &rec); |
1e61ee79 JB |
4770 | if (status) { |
4771 | mlog_errno(status); | |
4772 | goto bail; | |
4773 | } | |
dcd0538f | 4774 | |
627961b7 | 4775 | status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec, |
c77534f6 | 4776 | &free_records, &insert); |
dcd0538f MF |
4777 | if (status < 0) { |
4778 | mlog_errno(status); | |
4779 | goto bail; | |
ccd979bd MF |
4780 | } |
4781 | ||
dcd0538f MF |
4782 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4783 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4784 | "Insert.tree_depth: %d\n", | |
4785 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4786 | free_records, insert.ins_tree_depth); |
ccd979bd | 4787 | |
c77534f6 | 4788 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
d401dc12 | 4789 | status = ocfs2_grow_tree(handle, et, |
328d5752 | 4790 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4791 | meta_ac); |
4792 | if (status) { | |
ccd979bd MF |
4793 | mlog_errno(status); |
4794 | goto bail; | |
4795 | } | |
ccd979bd MF |
4796 | } |
4797 | ||
dcd0538f | 4798 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
3505bec0 | 4799 | status = ocfs2_do_insert_extent(handle, et, &rec, &insert); |
ccd979bd MF |
4800 | if (status < 0) |
4801 | mlog_errno(status); | |
92ba470c JB |
4802 | else |
4803 | ocfs2_et_extent_map_insert(et, &rec); | |
ccd979bd MF |
4804 | |
4805 | bail: | |
a81cb88b | 4806 | brelse(last_eb_bh); |
ccd979bd | 4807 | |
f56654c4 TM |
4808 | mlog_exit(status); |
4809 | return status; | |
4810 | } | |
4811 | ||
0eb8d47e TM |
4812 | /* |
4813 | * Allcate and add clusters into the extent b-tree. | |
4814 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4815 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4816 | * it is not limited to the file storage. Any extent tree can use this |
4817 | * function if it implements the proper ocfs2_extent_tree. | |
4818 | */ | |
cbee7e1a JB |
4819 | int ocfs2_add_clusters_in_btree(handle_t *handle, |
4820 | struct ocfs2_extent_tree *et, | |
0eb8d47e TM |
4821 | u32 *logical_offset, |
4822 | u32 clusters_to_add, | |
4823 | int mark_unwritten, | |
0eb8d47e TM |
4824 | struct ocfs2_alloc_context *data_ac, |
4825 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4826 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e TM |
4827 | { |
4828 | int status = 0; | |
4829 | int free_extents; | |
4830 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4831 | u32 bit_off, num_bits; | |
4832 | u64 block; | |
4833 | u8 flags = 0; | |
cbee7e1a JB |
4834 | struct ocfs2_super *osb = |
4835 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
0eb8d47e TM |
4836 | |
4837 | BUG_ON(!clusters_to_add); | |
4838 | ||
4839 | if (mark_unwritten) | |
4840 | flags = OCFS2_EXT_UNWRITTEN; | |
4841 | ||
3d03a305 | 4842 | free_extents = ocfs2_num_free_extents(osb, et); |
0eb8d47e TM |
4843 | if (free_extents < 0) { |
4844 | status = free_extents; | |
4845 | mlog_errno(status); | |
4846 | goto leave; | |
4847 | } | |
4848 | ||
4849 | /* there are two cases which could cause us to EAGAIN in the | |
4850 | * we-need-more-metadata case: | |
4851 | * 1) we haven't reserved *any* | |
4852 | * 2) we are so fragmented, we've needed to add metadata too | |
4853 | * many times. */ | |
4854 | if (!free_extents && !meta_ac) { | |
4855 | mlog(0, "we haven't reserved any metadata!\n"); | |
4856 | status = -EAGAIN; | |
4857 | reason = RESTART_META; | |
4858 | goto leave; | |
4859 | } else if ((!free_extents) | |
4860 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4861 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
0eb8d47e TM |
4862 | mlog(0, "filesystem is really fragmented...\n"); |
4863 | status = -EAGAIN; | |
4864 | reason = RESTART_META; | |
4865 | goto leave; | |
4866 | } | |
4867 | ||
4868 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4869 | clusters_to_add, &bit_off, &num_bits); | |
4870 | if (status < 0) { | |
4871 | if (status != -ENOSPC) | |
4872 | mlog_errno(status); | |
4873 | goto leave; | |
4874 | } | |
4875 | ||
4876 | BUG_ON(num_bits > clusters_to_add); | |
4877 | ||
13723d00 | 4878 | /* reserve our write early -- insert_extent may update the tree root */ |
d9a0a1f8 | 4879 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4880 | OCFS2_JOURNAL_ACCESS_WRITE); |
0eb8d47e TM |
4881 | if (status < 0) { |
4882 | mlog_errno(status); | |
4883 | goto leave; | |
4884 | } | |
4885 | ||
4886 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
cbee7e1a JB |
4887 | mlog(0, "Allocating %u clusters at block %u for owner %llu\n", |
4888 | num_bits, bit_off, | |
4889 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
cc79d8c1 | 4890 | status = ocfs2_insert_extent(handle, et, *logical_offset, block, |
f99b9b7c | 4891 | num_bits, flags, meta_ac); |
0eb8d47e TM |
4892 | if (status < 0) { |
4893 | mlog_errno(status); | |
4894 | goto leave; | |
4895 | } | |
4896 | ||
f99b9b7c | 4897 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4898 | if (status < 0) { |
4899 | mlog_errno(status); | |
4900 | goto leave; | |
4901 | } | |
4902 | ||
4903 | clusters_to_add -= num_bits; | |
4904 | *logical_offset += num_bits; | |
4905 | ||
4906 | if (clusters_to_add) { | |
4907 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4908 | clusters_to_add); | |
4909 | status = -EAGAIN; | |
4910 | reason = RESTART_TRANS; | |
4911 | } | |
4912 | ||
4913 | leave: | |
4914 | mlog_exit(status); | |
4915 | if (reason_ret) | |
4916 | *reason_ret = reason; | |
4917 | return status; | |
4918 | } | |
4919 | ||
328d5752 MF |
4920 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4921 | struct ocfs2_extent_rec *split_rec, | |
4922 | u32 cpos, | |
4923 | struct ocfs2_extent_rec *rec) | |
4924 | { | |
4925 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4926 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4927 | ||
4928 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4929 | ||
4930 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4931 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4932 | ||
4933 | split_rec->e_blkno = rec->e_blkno; | |
4934 | le64_add_cpu(&split_rec->e_blkno, | |
4935 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4936 | ||
4937 | split_rec->e_flags = rec->e_flags; | |
4938 | } | |
4939 | ||
d231129f | 4940 | static int ocfs2_split_and_insert(handle_t *handle, |
e7d4cb6b | 4941 | struct ocfs2_extent_tree *et, |
d231129f | 4942 | struct ocfs2_path *path, |
328d5752 MF |
4943 | struct buffer_head **last_eb_bh, |
4944 | int split_index, | |
4945 | struct ocfs2_extent_rec *orig_split_rec, | |
4946 | struct ocfs2_alloc_context *meta_ac) | |
4947 | { | |
4948 | int ret = 0, depth; | |
4949 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4950 | struct ocfs2_extent_rec tmprec; | |
4951 | struct ocfs2_extent_list *rightmost_el; | |
4952 | struct ocfs2_extent_rec rec; | |
4953 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4954 | struct ocfs2_insert_type insert; | |
4955 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4956 | |
4957 | leftright: | |
4958 | /* | |
4959 | * Store a copy of the record on the stack - it might move | |
4960 | * around as the tree is manipulated below. | |
4961 | */ | |
4962 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4963 | ||
ce1d9ea6 | 4964 | rightmost_el = et->et_root_el; |
328d5752 MF |
4965 | |
4966 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4967 | if (depth) { | |
4968 | BUG_ON(!(*last_eb_bh)); | |
4969 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4970 | rightmost_el = &eb->h_list; | |
4971 | } | |
4972 | ||
4973 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4974 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 4975 | ret = ocfs2_grow_tree(handle, et, |
e7d4cb6b | 4976 | &depth, last_eb_bh, meta_ac); |
328d5752 MF |
4977 | if (ret) { |
4978 | mlog_errno(ret); | |
4979 | goto out; | |
4980 | } | |
328d5752 MF |
4981 | } |
4982 | ||
4983 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4984 | insert.ins_appending = APPEND_NONE; | |
4985 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4986 | insert.ins_tree_depth = depth; |
4987 | ||
4988 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4989 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4990 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4991 | le16_to_cpu(rec.e_leaf_clusters); | |
4992 | ||
4993 | if (split_rec.e_cpos == rec.e_cpos) { | |
4994 | insert.ins_split = SPLIT_LEFT; | |
4995 | } else if (insert_range == rec_range) { | |
4996 | insert.ins_split = SPLIT_RIGHT; | |
4997 | } else { | |
4998 | /* | |
4999 | * Left/right split. We fake this as a right split | |
5000 | * first and then make a second pass as a left split. | |
5001 | */ | |
5002 | insert.ins_split = SPLIT_RIGHT; | |
5003 | ||
d231129f JB |
5004 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
5005 | &tmprec, insert_range, &rec); | |
328d5752 MF |
5006 | |
5007 | split_rec = tmprec; | |
5008 | ||
5009 | BUG_ON(do_leftright); | |
5010 | do_leftright = 1; | |
5011 | } | |
5012 | ||
3505bec0 | 5013 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
328d5752 MF |
5014 | if (ret) { |
5015 | mlog_errno(ret); | |
5016 | goto out; | |
5017 | } | |
5018 | ||
5019 | if (do_leftright == 1) { | |
5020 | u32 cpos; | |
5021 | struct ocfs2_extent_list *el; | |
5022 | ||
5023 | do_leftright++; | |
5024 | split_rec = *orig_split_rec; | |
5025 | ||
5026 | ocfs2_reinit_path(path, 1); | |
5027 | ||
5028 | cpos = le32_to_cpu(split_rec.e_cpos); | |
facdb77f | 5029 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
328d5752 MF |
5030 | if (ret) { |
5031 | mlog_errno(ret); | |
5032 | goto out; | |
5033 | } | |
5034 | ||
5035 | el = path_leaf_el(path); | |
5036 | split_index = ocfs2_search_extent_list(el, cpos); | |
5037 | goto leftright; | |
5038 | } | |
5039 | out: | |
5040 | ||
5041 | return ret; | |
5042 | } | |
5043 | ||
f3868d0f JB |
5044 | static int ocfs2_replace_extent_rec(handle_t *handle, |
5045 | struct ocfs2_extent_tree *et, | |
47be12e4 TM |
5046 | struct ocfs2_path *path, |
5047 | struct ocfs2_extent_list *el, | |
5048 | int split_index, | |
5049 | struct ocfs2_extent_rec *split_rec) | |
5050 | { | |
5051 | int ret; | |
5052 | ||
f3868d0f | 5053 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
47be12e4 TM |
5054 | path_num_items(path) - 1); |
5055 | if (ret) { | |
5056 | mlog_errno(ret); | |
5057 | goto out; | |
5058 | } | |
5059 | ||
5060 | el->l_recs[split_index] = *split_rec; | |
5061 | ||
5062 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5063 | out: | |
5064 | return ret; | |
5065 | } | |
5066 | ||
328d5752 | 5067 | /* |
555936bf TM |
5068 | * Split part or all of the extent record at split_index in the leaf |
5069 | * pointed to by path. Merge with the contiguous extent record if needed. | |
328d5752 MF |
5070 | * |
5071 | * Care is taken to handle contiguousness so as to not grow the tree. | |
5072 | * | |
5073 | * meta_ac is not strictly necessary - we only truly need it if growth | |
5074 | * of the tree is required. All other cases will degrade into a less | |
5075 | * optimal tree layout. | |
5076 | * | |
e7d4cb6b TM |
5077 | * last_eb_bh should be the rightmost leaf block for any extent |
5078 | * btree. Since a split may grow the tree or a merge might shrink it, | |
5079 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
5080 | * |
5081 | * This code is optimized for readability - several passes might be | |
5082 | * made over certain portions of the tree. All of those blocks will | |
5083 | * have been brought into cache (and pinned via the journal), so the | |
5084 | * extra overhead is not expressed in terms of disk reads. | |
5085 | */ | |
e2e9f608 TM |
5086 | int ocfs2_split_extent(handle_t *handle, |
5087 | struct ocfs2_extent_tree *et, | |
5088 | struct ocfs2_path *path, | |
5089 | int split_index, | |
5090 | struct ocfs2_extent_rec *split_rec, | |
5091 | struct ocfs2_alloc_context *meta_ac, | |
5092 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
328d5752 MF |
5093 | { |
5094 | int ret = 0; | |
5095 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 5096 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
5097 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
5098 | struct ocfs2_merge_ctxt ctxt; | |
5099 | struct ocfs2_extent_list *rightmost_el; | |
5100 | ||
328d5752 MF |
5101 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || |
5102 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
5103 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
5104 | ret = -EIO; | |
5105 | mlog_errno(ret); | |
5106 | goto out; | |
5107 | } | |
5108 | ||
a2970291 | 5109 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el, |
328d5752 MF |
5110 | split_index, |
5111 | split_rec); | |
5112 | ||
5113 | /* | |
5114 | * The core merge / split code wants to know how much room is | |
a1cf076b | 5115 | * left in this allocation tree, so we pass the |
328d5752 MF |
5116 | * rightmost extent list. |
5117 | */ | |
5118 | if (path->p_tree_depth) { | |
5119 | struct ocfs2_extent_block *eb; | |
328d5752 | 5120 | |
3d03a305 | 5121 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5122 | ocfs2_et_get_last_eb_blk(et), |
5123 | &last_eb_bh); | |
328d5752 MF |
5124 | if (ret) { |
5125 | mlog_exit(ret); | |
5126 | goto out; | |
5127 | } | |
5128 | ||
5129 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
328d5752 MF |
5130 | rightmost_el = &eb->h_list; |
5131 | } else | |
5132 | rightmost_el = path_root_el(path); | |
5133 | ||
328d5752 MF |
5134 | if (rec->e_cpos == split_rec->e_cpos && |
5135 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
5136 | ctxt.c_split_covers_rec = 1; | |
5137 | else | |
5138 | ctxt.c_split_covers_rec = 0; | |
5139 | ||
5140 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
5141 | ||
015452b1 MF |
5142 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
5143 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
5144 | ctxt.c_split_covers_rec); | |
328d5752 MF |
5145 | |
5146 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
5147 | if (ctxt.c_split_covers_rec) | |
f3868d0f | 5148 | ret = ocfs2_replace_extent_rec(handle, et, path, el, |
47be12e4 | 5149 | split_index, split_rec); |
328d5752 | 5150 | else |
d231129f | 5151 | ret = ocfs2_split_and_insert(handle, et, path, |
328d5752 MF |
5152 | &last_eb_bh, split_index, |
5153 | split_rec, meta_ac); | |
5154 | if (ret) | |
5155 | mlog_errno(ret); | |
5156 | } else { | |
c495dd24 | 5157 | ret = ocfs2_try_to_merge_extent(handle, et, path, |
328d5752 | 5158 | split_index, split_rec, |
c495dd24 | 5159 | dealloc, &ctxt); |
328d5752 MF |
5160 | if (ret) |
5161 | mlog_errno(ret); | |
5162 | } | |
5163 | ||
328d5752 MF |
5164 | out: |
5165 | brelse(last_eb_bh); | |
5166 | return ret; | |
5167 | } | |
5168 | ||
5169 | /* | |
555936bf TM |
5170 | * Change the flags of the already-existing extent at cpos for len clusters. |
5171 | * | |
5172 | * new_flags: the flags we want to set. | |
5173 | * clear_flags: the flags we want to clear. | |
5174 | * phys: the new physical offset we want this new extent starts from. | |
328d5752 MF |
5175 | * |
5176 | * If the existing extent is larger than the request, initiate a | |
5177 | * split. An attempt will be made at merging with adjacent extents. | |
5178 | * | |
5179 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5180 | */ | |
1aa75fea TM |
5181 | int ocfs2_change_extent_flag(handle_t *handle, |
5182 | struct ocfs2_extent_tree *et, | |
5183 | u32 cpos, u32 len, u32 phys, | |
5184 | struct ocfs2_alloc_context *meta_ac, | |
5185 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
5186 | int new_flags, int clear_flags) | |
328d5752 MF |
5187 | { |
5188 | int ret, index; | |
555936bf TM |
5189 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
5190 | u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys); | |
328d5752 MF |
5191 | struct ocfs2_extent_rec split_rec; |
5192 | struct ocfs2_path *left_path = NULL; | |
5193 | struct ocfs2_extent_list *el; | |
555936bf | 5194 | struct ocfs2_extent_rec *rec; |
328d5752 | 5195 | |
ffdd7a54 | 5196 | left_path = ocfs2_new_path_from_et(et); |
328d5752 MF |
5197 | if (!left_path) { |
5198 | ret = -ENOMEM; | |
5199 | mlog_errno(ret); | |
5200 | goto out; | |
5201 | } | |
5202 | ||
facdb77f | 5203 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
5204 | if (ret) { |
5205 | mlog_errno(ret); | |
5206 | goto out; | |
5207 | } | |
5208 | el = path_leaf_el(left_path); | |
5209 | ||
5210 | index = ocfs2_search_extent_list(el, cpos); | |
5211 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
555936bf TM |
5212 | ocfs2_error(sb, |
5213 | "Owner %llu has an extent at cpos %u which can no " | |
328d5752 | 5214 | "longer be found.\n", |
555936bf TM |
5215 | (unsigned long long) |
5216 | ocfs2_metadata_cache_owner(et->et_ci), cpos); | |
328d5752 MF |
5217 | ret = -EROFS; |
5218 | goto out; | |
5219 | } | |
5220 | ||
555936bf TM |
5221 | ret = -EIO; |
5222 | rec = &el->l_recs[index]; | |
5223 | if (new_flags && (rec->e_flags & new_flags)) { | |
5224 | mlog(ML_ERROR, "Owner %llu tried to set %d flags on an " | |
5225 | "extent that already had them", | |
5226 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5227 | new_flags); | |
5228 | goto out; | |
5229 | } | |
5230 | ||
5231 | if (clear_flags && !(rec->e_flags & clear_flags)) { | |
5232 | mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an " | |
5233 | "extent that didn't have them", | |
5234 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5235 | clear_flags); | |
5236 | goto out; | |
5237 | } | |
5238 | ||
328d5752 MF |
5239 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); |
5240 | split_rec.e_cpos = cpu_to_le32(cpos); | |
5241 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
5242 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
555936bf TM |
5243 | split_rec.e_flags = rec->e_flags; |
5244 | if (new_flags) | |
5245 | split_rec.e_flags |= new_flags; | |
5246 | if (clear_flags) | |
5247 | split_rec.e_flags &= ~clear_flags; | |
5248 | ||
e2e9f608 TM |
5249 | ret = ocfs2_split_extent(handle, et, left_path, |
5250 | index, &split_rec, meta_ac, | |
5251 | dealloc); | |
328d5752 MF |
5252 | if (ret) |
5253 | mlog_errno(ret); | |
5254 | ||
5255 | out: | |
5256 | ocfs2_free_path(left_path); | |
5257 | return ret; | |
555936bf TM |
5258 | |
5259 | } | |
5260 | ||
5261 | /* | |
5262 | * Mark the already-existing extent at cpos as written for len clusters. | |
5263 | * This removes the unwritten extent flag. | |
5264 | * | |
5265 | * If the existing extent is larger than the request, initiate a | |
5266 | * split. An attempt will be made at merging with adjacent extents. | |
5267 | * | |
5268 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5269 | */ | |
5270 | int ocfs2_mark_extent_written(struct inode *inode, | |
5271 | struct ocfs2_extent_tree *et, | |
5272 | handle_t *handle, u32 cpos, u32 len, u32 phys, | |
5273 | struct ocfs2_alloc_context *meta_ac, | |
5274 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5275 | { | |
5276 | int ret; | |
5277 | ||
5278 | mlog(0, "Inode %lu cpos %u, len %u, phys clusters %u\n", | |
5279 | inode->i_ino, cpos, len, phys); | |
5280 | ||
5281 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
5282 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
5283 | "that are being written to, but the feature bit " | |
5284 | "is not set in the super block.", | |
5285 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
5286 | ret = -EROFS; | |
5287 | goto out; | |
5288 | } | |
5289 | ||
5290 | /* | |
5291 | * XXX: This should be fixed up so that we just re-insert the | |
5292 | * next extent records. | |
5293 | */ | |
5294 | ocfs2_et_extent_map_truncate(et, 0); | |
5295 | ||
5296 | ret = ocfs2_change_extent_flag(handle, et, cpos, | |
5297 | len, phys, meta_ac, dealloc, | |
5298 | 0, OCFS2_EXT_UNWRITTEN); | |
5299 | if (ret) | |
5300 | mlog_errno(ret); | |
5301 | ||
5302 | out: | |
5303 | return ret; | |
328d5752 MF |
5304 | } |
5305 | ||
dbdcf6a4 JB |
5306 | static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
5307 | struct ocfs2_path *path, | |
d0c7d708 MF |
5308 | int index, u32 new_range, |
5309 | struct ocfs2_alloc_context *meta_ac) | |
5310 | { | |
5311 | int ret, depth, credits = handle->h_buffer_credits; | |
d0c7d708 MF |
5312 | struct buffer_head *last_eb_bh = NULL; |
5313 | struct ocfs2_extent_block *eb; | |
5314 | struct ocfs2_extent_list *rightmost_el, *el; | |
5315 | struct ocfs2_extent_rec split_rec; | |
5316 | struct ocfs2_extent_rec *rec; | |
5317 | struct ocfs2_insert_type insert; | |
5318 | ||
5319 | /* | |
5320 | * Setup the record to split before we grow the tree. | |
5321 | */ | |
5322 | el = path_leaf_el(path); | |
5323 | rec = &el->l_recs[index]; | |
dbdcf6a4 JB |
5324 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
5325 | &split_rec, new_range, rec); | |
d0c7d708 MF |
5326 | |
5327 | depth = path->p_tree_depth; | |
5328 | if (depth > 0) { | |
3d03a305 | 5329 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5330 | ocfs2_et_get_last_eb_blk(et), |
5331 | &last_eb_bh); | |
d0c7d708 MF |
5332 | if (ret < 0) { |
5333 | mlog_errno(ret); | |
5334 | goto out; | |
5335 | } | |
5336 | ||
5337 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
5338 | rightmost_el = &eb->h_list; | |
5339 | } else | |
5340 | rightmost_el = path_leaf_el(path); | |
5341 | ||
811f933d | 5342 | credits += path->p_tree_depth + |
ce1d9ea6 | 5343 | ocfs2_extend_meta_needed(et->et_root_el); |
d0c7d708 MF |
5344 | ret = ocfs2_extend_trans(handle, credits); |
5345 | if (ret) { | |
5346 | mlog_errno(ret); | |
5347 | goto out; | |
5348 | } | |
5349 | ||
5350 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5351 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 5352 | ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5353 | meta_ac); |
5354 | if (ret) { | |
5355 | mlog_errno(ret); | |
5356 | goto out; | |
5357 | } | |
d0c7d708 MF |
5358 | } |
5359 | ||
5360 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5361 | insert.ins_appending = APPEND_NONE; | |
5362 | insert.ins_contig = CONTIG_NONE; | |
5363 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5364 | insert.ins_tree_depth = depth; |
5365 | ||
3505bec0 | 5366 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
d0c7d708 MF |
5367 | if (ret) |
5368 | mlog_errno(ret); | |
5369 | ||
5370 | out: | |
5371 | brelse(last_eb_bh); | |
5372 | return ret; | |
5373 | } | |
5374 | ||
043beebb JB |
5375 | static int ocfs2_truncate_rec(handle_t *handle, |
5376 | struct ocfs2_extent_tree *et, | |
d0c7d708 MF |
5377 | struct ocfs2_path *path, int index, |
5378 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
043beebb | 5379 | u32 cpos, u32 len) |
d0c7d708 MF |
5380 | { |
5381 | int ret; | |
5382 | u32 left_cpos, rec_range, trunc_range; | |
5383 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
043beebb | 5384 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
d0c7d708 MF |
5385 | struct ocfs2_path *left_path = NULL; |
5386 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5387 | struct ocfs2_extent_rec *rec; | |
5388 | struct ocfs2_extent_block *eb; | |
5389 | ||
5390 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
70f18c08 | 5391 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5392 | if (ret) { |
5393 | mlog_errno(ret); | |
5394 | goto out; | |
5395 | } | |
5396 | ||
5397 | index--; | |
5398 | } | |
5399 | ||
5400 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5401 | path->p_tree_depth) { | |
5402 | /* | |
5403 | * Check whether this is the rightmost tree record. If | |
5404 | * we remove all of this record or part of its right | |
5405 | * edge then an update of the record lengths above it | |
5406 | * will be required. | |
5407 | */ | |
5408 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5409 | if (eb->h_next_leaf_blk == 0) | |
5410 | is_rightmost_tree_rec = 1; | |
5411 | } | |
5412 | ||
5413 | rec = &el->l_recs[index]; | |
5414 | if (index == 0 && path->p_tree_depth && | |
5415 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5416 | /* | |
5417 | * Changing the leftmost offset (via partial or whole | |
5418 | * record truncate) of an interior (or rightmost) path | |
5419 | * means we have to update the subtree that is formed | |
5420 | * by this leaf and the one to it's left. | |
5421 | * | |
5422 | * There are two cases we can skip: | |
043beebb | 5423 | * 1) Path is the leftmost one in our btree. |
d0c7d708 MF |
5424 | * 2) The leaf is rightmost and will be empty after |
5425 | * we remove the extent record - the rotate code | |
5426 | * knows how to update the newly formed edge. | |
5427 | */ | |
5428 | ||
043beebb | 5429 | ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
d0c7d708 MF |
5430 | if (ret) { |
5431 | mlog_errno(ret); | |
5432 | goto out; | |
5433 | } | |
5434 | ||
5435 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
ffdd7a54 | 5436 | left_path = ocfs2_new_path_from_path(path); |
d0c7d708 MF |
5437 | if (!left_path) { |
5438 | ret = -ENOMEM; | |
5439 | mlog_errno(ret); | |
5440 | goto out; | |
5441 | } | |
5442 | ||
facdb77f JB |
5443 | ret = ocfs2_find_path(et->et_ci, left_path, |
5444 | left_cpos); | |
d0c7d708 MF |
5445 | if (ret) { |
5446 | mlog_errno(ret); | |
5447 | goto out; | |
5448 | } | |
5449 | } | |
5450 | } | |
5451 | ||
5452 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5453 | handle->h_buffer_credits, | |
5454 | path); | |
5455 | if (ret) { | |
5456 | mlog_errno(ret); | |
5457 | goto out; | |
5458 | } | |
5459 | ||
d9a0a1f8 | 5460 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
d0c7d708 MF |
5461 | if (ret) { |
5462 | mlog_errno(ret); | |
5463 | goto out; | |
5464 | } | |
5465 | ||
d9a0a1f8 | 5466 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
d0c7d708 MF |
5467 | if (ret) { |
5468 | mlog_errno(ret); | |
5469 | goto out; | |
5470 | } | |
5471 | ||
5472 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5473 | trunc_range = cpos + len; | |
5474 | ||
5475 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5476 | int next_free; | |
5477 | ||
5478 | memset(rec, 0, sizeof(*rec)); | |
5479 | ocfs2_cleanup_merge(el, index); | |
5480 | wants_rotate = 1; | |
5481 | ||
5482 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5483 | if (is_rightmost_tree_rec && next_free > 1) { | |
5484 | /* | |
5485 | * We skip the edge update if this path will | |
5486 | * be deleted by the rotate code. | |
5487 | */ | |
5488 | rec = &el->l_recs[next_free - 1]; | |
d401dc12 | 5489 | ocfs2_adjust_rightmost_records(handle, et, path, |
d0c7d708 MF |
5490 | rec); |
5491 | } | |
5492 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5493 | /* Remove leftmost portion of the record. */ | |
5494 | le32_add_cpu(&rec->e_cpos, len); | |
5495 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5496 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5497 | } else if (rec_range == trunc_range) { | |
5498 | /* Remove rightmost portion of the record */ | |
5499 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5500 | if (is_rightmost_tree_rec) | |
d401dc12 | 5501 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
d0c7d708 MF |
5502 | } else { |
5503 | /* Caller should have trapped this. */ | |
043beebb JB |
5504 | mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) " |
5505 | "(%u, %u)\n", | |
5506 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5507 | le32_to_cpu(rec->e_cpos), |
5508 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5509 | BUG(); | |
5510 | } | |
5511 | ||
5512 | if (left_path) { | |
5513 | int subtree_index; | |
5514 | ||
7dc02805 | 5515 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
4619c73e | 5516 | ocfs2_complete_edge_insert(handle, left_path, path, |
d0c7d708 MF |
5517 | subtree_index); |
5518 | } | |
5519 | ||
5520 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5521 | ||
70f18c08 | 5522 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5523 | if (ret) { |
5524 | mlog_errno(ret); | |
5525 | goto out; | |
5526 | } | |
5527 | ||
5528 | out: | |
5529 | ocfs2_free_path(left_path); | |
5530 | return ret; | |
5531 | } | |
5532 | ||
dbdcf6a4 | 5533 | int ocfs2_remove_extent(handle_t *handle, |
f99b9b7c | 5534 | struct ocfs2_extent_tree *et, |
dbdcf6a4 | 5535 | u32 cpos, u32 len, |
063c4561 | 5536 | struct ocfs2_alloc_context *meta_ac, |
f99b9b7c | 5537 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5538 | { |
5539 | int ret, index; | |
5540 | u32 rec_range, trunc_range; | |
5541 | struct ocfs2_extent_rec *rec; | |
5542 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5543 | struct ocfs2_path *path = NULL; |
d0c7d708 | 5544 | |
4c911eef JB |
5545 | /* |
5546 | * XXX: Why are we truncating to 0 instead of wherever this | |
5547 | * affects us? | |
5548 | */ | |
5549 | ocfs2_et_extent_map_truncate(et, 0); | |
d0c7d708 | 5550 | |
ffdd7a54 | 5551 | path = ocfs2_new_path_from_et(et); |
d0c7d708 MF |
5552 | if (!path) { |
5553 | ret = -ENOMEM; | |
5554 | mlog_errno(ret); | |
5555 | goto out; | |
5556 | } | |
5557 | ||
facdb77f | 5558 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5559 | if (ret) { |
5560 | mlog_errno(ret); | |
5561 | goto out; | |
5562 | } | |
5563 | ||
5564 | el = path_leaf_el(path); | |
5565 | index = ocfs2_search_extent_list(el, cpos); | |
5566 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
dbdcf6a4 JB |
5567 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5568 | "Owner %llu has an extent at cpos %u which can no " | |
d0c7d708 | 5569 | "longer be found.\n", |
dbdcf6a4 JB |
5570 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5571 | cpos); | |
d0c7d708 MF |
5572 | ret = -EROFS; |
5573 | goto out; | |
5574 | } | |
5575 | ||
5576 | /* | |
5577 | * We have 3 cases of extent removal: | |
5578 | * 1) Range covers the entire extent rec | |
5579 | * 2) Range begins or ends on one edge of the extent rec | |
5580 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5581 | * | |
5582 | * For case 1 we remove the extent rec and left rotate to | |
5583 | * fill the hole. | |
5584 | * | |
5585 | * For case 2 we just shrink the existing extent rec, with a | |
5586 | * tree update if the shrinking edge is also the edge of an | |
5587 | * extent block. | |
5588 | * | |
5589 | * For case 3 we do a right split to turn the extent rec into | |
5590 | * something case 2 can handle. | |
5591 | */ | |
5592 | rec = &el->l_recs[index]; | |
5593 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5594 | trunc_range = cpos + len; | |
5595 | ||
5596 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5597 | ||
dbdcf6a4 | 5598 | mlog(0, "Owner %llu, remove (cpos %u, len %u). Existing index %d " |
d0c7d708 | 5599 | "(cpos %u, len %u)\n", |
dbdcf6a4 JB |
5600 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5601 | cpos, len, index, | |
d0c7d708 MF |
5602 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); |
5603 | ||
5604 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
043beebb JB |
5605 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5606 | cpos, len); | |
d0c7d708 MF |
5607 | if (ret) { |
5608 | mlog_errno(ret); | |
5609 | goto out; | |
5610 | } | |
5611 | } else { | |
dbdcf6a4 | 5612 | ret = ocfs2_split_tree(handle, et, path, index, |
d0c7d708 MF |
5613 | trunc_range, meta_ac); |
5614 | if (ret) { | |
5615 | mlog_errno(ret); | |
5616 | goto out; | |
5617 | } | |
5618 | ||
5619 | /* | |
5620 | * The split could have manipulated the tree enough to | |
5621 | * move the record location, so we have to look for it again. | |
5622 | */ | |
5623 | ocfs2_reinit_path(path, 1); | |
5624 | ||
facdb77f | 5625 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5626 | if (ret) { |
5627 | mlog_errno(ret); | |
5628 | goto out; | |
5629 | } | |
5630 | ||
5631 | el = path_leaf_el(path); | |
5632 | index = ocfs2_search_extent_list(el, cpos); | |
5633 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
dbdcf6a4 JB |
5634 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5635 | "Owner %llu: split at cpos %u lost record.", | |
5636 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5637 | cpos); |
5638 | ret = -EROFS; | |
5639 | goto out; | |
5640 | } | |
5641 | ||
5642 | /* | |
5643 | * Double check our values here. If anything is fishy, | |
5644 | * it's easier to catch it at the top level. | |
5645 | */ | |
5646 | rec = &el->l_recs[index]; | |
5647 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5648 | ocfs2_rec_clusters(el, rec); | |
5649 | if (rec_range != trunc_range) { | |
dbdcf6a4 JB |
5650 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5651 | "Owner %llu: error after split at cpos %u" | |
d0c7d708 | 5652 | "trunc len %u, existing record is (%u,%u)", |
dbdcf6a4 | 5653 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
d0c7d708 MF |
5654 | cpos, len, le32_to_cpu(rec->e_cpos), |
5655 | ocfs2_rec_clusters(el, rec)); | |
5656 | ret = -EROFS; | |
5657 | goto out; | |
5658 | } | |
5659 | ||
043beebb JB |
5660 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5661 | cpos, len); | |
d0c7d708 MF |
5662 | if (ret) { |
5663 | mlog_errno(ret); | |
5664 | goto out; | |
5665 | } | |
5666 | } | |
5667 | ||
5668 | out: | |
5669 | ocfs2_free_path(path); | |
5670 | return ret; | |
5671 | } | |
5672 | ||
fecc0112 MF |
5673 | int ocfs2_remove_btree_range(struct inode *inode, |
5674 | struct ocfs2_extent_tree *et, | |
5675 | u32 cpos, u32 phys_cpos, u32 len, | |
5676 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5677 | { | |
5678 | int ret; | |
5679 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
5680 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5681 | struct inode *tl_inode = osb->osb_tl_inode; | |
5682 | handle_t *handle; | |
5683 | struct ocfs2_alloc_context *meta_ac = NULL; | |
5684 | ||
5685 | ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac); | |
5686 | if (ret) { | |
5687 | mlog_errno(ret); | |
5688 | return ret; | |
5689 | } | |
5690 | ||
5691 | mutex_lock(&tl_inode->i_mutex); | |
5692 | ||
5693 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5694 | ret = __ocfs2_flush_truncate_log(osb); | |
5695 | if (ret < 0) { | |
5696 | mlog_errno(ret); | |
5697 | goto out; | |
5698 | } | |
5699 | } | |
5700 | ||
a90714c1 | 5701 | handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb)); |
fecc0112 MF |
5702 | if (IS_ERR(handle)) { |
5703 | ret = PTR_ERR(handle); | |
5704 | mlog_errno(ret); | |
5705 | goto out; | |
5706 | } | |
5707 | ||
d9a0a1f8 | 5708 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 5709 | OCFS2_JOURNAL_ACCESS_WRITE); |
fecc0112 MF |
5710 | if (ret) { |
5711 | mlog_errno(ret); | |
5712 | goto out; | |
5713 | } | |
5714 | ||
fd4ef231 MF |
5715 | vfs_dq_free_space_nodirty(inode, |
5716 | ocfs2_clusters_to_bytes(inode->i_sb, len)); | |
5717 | ||
dbdcf6a4 | 5718 | ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc); |
fecc0112 MF |
5719 | if (ret) { |
5720 | mlog_errno(ret); | |
5721 | goto out_commit; | |
5722 | } | |
5723 | ||
6136ca5f | 5724 | ocfs2_et_update_clusters(et, -len); |
fecc0112 MF |
5725 | |
5726 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); | |
5727 | if (ret) { | |
5728 | mlog_errno(ret); | |
5729 | goto out_commit; | |
5730 | } | |
5731 | ||
5732 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
5733 | if (ret) | |
5734 | mlog_errno(ret); | |
5735 | ||
5736 | out_commit: | |
5737 | ocfs2_commit_trans(osb, handle); | |
5738 | out: | |
5739 | mutex_unlock(&tl_inode->i_mutex); | |
5740 | ||
5741 | if (meta_ac) | |
5742 | ocfs2_free_alloc_context(meta_ac); | |
5743 | ||
5744 | return ret; | |
5745 | } | |
5746 | ||
063c4561 | 5747 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5748 | { |
5749 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5750 | struct ocfs2_dinode *di; | |
5751 | struct ocfs2_truncate_log *tl; | |
5752 | ||
5753 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5754 | tl = &di->id2.i_dealloc; | |
5755 | ||
5756 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5757 | "slot %d, invalid truncate log parameters: used = " | |
5758 | "%u, count = %u\n", osb->slot_num, | |
5759 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5760 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5761 | } | |
5762 | ||
5763 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5764 | unsigned int new_start) | |
5765 | { | |
5766 | unsigned int tail_index; | |
5767 | unsigned int current_tail; | |
5768 | ||
5769 | /* No records, nothing to coalesce */ | |
5770 | if (!le16_to_cpu(tl->tl_used)) | |
5771 | return 0; | |
5772 | ||
5773 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5774 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5775 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5776 | ||
5777 | return current_tail == new_start; | |
5778 | } | |
5779 | ||
063c4561 MF |
5780 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5781 | handle_t *handle, | |
5782 | u64 start_blk, | |
5783 | unsigned int num_clusters) | |
ccd979bd MF |
5784 | { |
5785 | int status, index; | |
5786 | unsigned int start_cluster, tl_count; | |
5787 | struct inode *tl_inode = osb->osb_tl_inode; | |
5788 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5789 | struct ocfs2_dinode *di; | |
5790 | struct ocfs2_truncate_log *tl; | |
5791 | ||
b0697053 MF |
5792 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5793 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5794 | |
1b1dcc1b | 5795 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5796 | |
5797 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5798 | ||
5799 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5800 | |
10995aa2 JB |
5801 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5802 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5803 | * corruption is a code bug */ | |
5804 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5805 | ||
5806 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5807 | tl_count = le16_to_cpu(tl->tl_count); |
5808 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5809 | tl_count == 0, | |
b0697053 MF |
5810 | "Truncate record count on #%llu invalid " |
5811 | "wanted %u, actual %u\n", | |
5812 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5813 | ocfs2_truncate_recs_per_inode(osb->sb), |
5814 | le16_to_cpu(tl->tl_count)); | |
5815 | ||
5816 | /* Caller should have known to flush before calling us. */ | |
5817 | index = le16_to_cpu(tl->tl_used); | |
5818 | if (index >= tl_count) { | |
5819 | status = -ENOSPC; | |
5820 | mlog_errno(status); | |
5821 | goto bail; | |
5822 | } | |
5823 | ||
0cf2f763 | 5824 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5825 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5826 | if (status < 0) { |
5827 | mlog_errno(status); | |
5828 | goto bail; | |
5829 | } | |
5830 | ||
5831 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5832 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5833 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5834 | |
5835 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5836 | /* | |
5837 | * Move index back to the record we are coalescing with. | |
5838 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5839 | */ | |
5840 | index--; | |
5841 | ||
5842 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5843 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5844 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5845 | num_clusters); | |
5846 | } else { | |
5847 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5848 | tl->tl_used = cpu_to_le16(index + 1); | |
5849 | } | |
5850 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5851 | ||
5852 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5853 | if (status < 0) { | |
5854 | mlog_errno(status); | |
5855 | goto bail; | |
5856 | } | |
5857 | ||
5858 | bail: | |
5859 | mlog_exit(status); | |
5860 | return status; | |
5861 | } | |
5862 | ||
5863 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5864 | handle_t *handle, |
ccd979bd MF |
5865 | struct inode *data_alloc_inode, |
5866 | struct buffer_head *data_alloc_bh) | |
5867 | { | |
5868 | int status = 0; | |
5869 | int i; | |
5870 | unsigned int num_clusters; | |
5871 | u64 start_blk; | |
5872 | struct ocfs2_truncate_rec rec; | |
5873 | struct ocfs2_dinode *di; | |
5874 | struct ocfs2_truncate_log *tl; | |
5875 | struct inode *tl_inode = osb->osb_tl_inode; | |
5876 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5877 | ||
5878 | mlog_entry_void(); | |
5879 | ||
5880 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5881 | tl = &di->id2.i_dealloc; | |
5882 | i = le16_to_cpu(tl->tl_used) - 1; | |
5883 | while (i >= 0) { | |
5884 | /* Caller has given us at least enough credits to | |
5885 | * update the truncate log dinode */ | |
0cf2f763 | 5886 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5887 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5888 | if (status < 0) { |
5889 | mlog_errno(status); | |
5890 | goto bail; | |
5891 | } | |
5892 | ||
5893 | tl->tl_used = cpu_to_le16(i); | |
5894 | ||
5895 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5896 | if (status < 0) { | |
5897 | mlog_errno(status); | |
5898 | goto bail; | |
5899 | } | |
5900 | ||
5901 | /* TODO: Perhaps we can calculate the bulk of the | |
5902 | * credits up front rather than extending like | |
5903 | * this. */ | |
5904 | status = ocfs2_extend_trans(handle, | |
5905 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5906 | if (status < 0) { | |
5907 | mlog_errno(status); | |
5908 | goto bail; | |
5909 | } | |
5910 | ||
5911 | rec = tl->tl_recs[i]; | |
5912 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5913 | le32_to_cpu(rec.t_start)); | |
5914 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5915 | ||
5916 | /* if start_blk is not set, we ignore the record as | |
5917 | * invalid. */ | |
5918 | if (start_blk) { | |
5919 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5920 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5921 | ||
5922 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5923 | data_alloc_bh, start_blk, | |
5924 | num_clusters); | |
5925 | if (status < 0) { | |
5926 | mlog_errno(status); | |
5927 | goto bail; | |
5928 | } | |
5929 | } | |
5930 | i--; | |
5931 | } | |
5932 | ||
5933 | bail: | |
5934 | mlog_exit(status); | |
5935 | return status; | |
5936 | } | |
5937 | ||
1b1dcc1b | 5938 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5939 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5940 | { |
5941 | int status; | |
5942 | unsigned int num_to_flush; | |
1fabe148 | 5943 | handle_t *handle; |
ccd979bd MF |
5944 | struct inode *tl_inode = osb->osb_tl_inode; |
5945 | struct inode *data_alloc_inode = NULL; | |
5946 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5947 | struct buffer_head *data_alloc_bh = NULL; | |
5948 | struct ocfs2_dinode *di; | |
5949 | struct ocfs2_truncate_log *tl; | |
5950 | ||
5951 | mlog_entry_void(); | |
5952 | ||
1b1dcc1b | 5953 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5954 | |
5955 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5956 | |
10995aa2 JB |
5957 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5958 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5959 | * corruption is a code bug */ | |
5960 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5961 | ||
5962 | tl = &di->id2.i_dealloc; | |
ccd979bd | 5963 | num_to_flush = le16_to_cpu(tl->tl_used); |
b0697053 MF |
5964 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5965 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5966 | if (!num_to_flush) { |
5967 | status = 0; | |
e08dc8b9 | 5968 | goto out; |
ccd979bd MF |
5969 | } |
5970 | ||
5971 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5972 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5973 | OCFS2_INVALID_SLOT); | |
5974 | if (!data_alloc_inode) { | |
5975 | status = -EINVAL; | |
5976 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5977 | goto out; |
ccd979bd MF |
5978 | } |
5979 | ||
e08dc8b9 MF |
5980 | mutex_lock(&data_alloc_inode->i_mutex); |
5981 | ||
e63aecb6 | 5982 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5983 | if (status < 0) { |
5984 | mlog_errno(status); | |
e08dc8b9 | 5985 | goto out_mutex; |
ccd979bd MF |
5986 | } |
5987 | ||
65eff9cc | 5988 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5989 | if (IS_ERR(handle)) { |
5990 | status = PTR_ERR(handle); | |
ccd979bd | 5991 | mlog_errno(status); |
e08dc8b9 | 5992 | goto out_unlock; |
ccd979bd MF |
5993 | } |
5994 | ||
5995 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5996 | data_alloc_bh); | |
e08dc8b9 | 5997 | if (status < 0) |
ccd979bd | 5998 | mlog_errno(status); |
ccd979bd | 5999 | |
02dc1af4 | 6000 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 6001 | |
e08dc8b9 MF |
6002 | out_unlock: |
6003 | brelse(data_alloc_bh); | |
e63aecb6 | 6004 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 6005 | |
e08dc8b9 MF |
6006 | out_mutex: |
6007 | mutex_unlock(&data_alloc_inode->i_mutex); | |
6008 | iput(data_alloc_inode); | |
ccd979bd | 6009 | |
e08dc8b9 | 6010 | out: |
ccd979bd MF |
6011 | mlog_exit(status); |
6012 | return status; | |
6013 | } | |
6014 | ||
6015 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
6016 | { | |
6017 | int status; | |
6018 | struct inode *tl_inode = osb->osb_tl_inode; | |
6019 | ||
1b1dcc1b | 6020 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 6021 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 6022 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6023 | |
6024 | return status; | |
6025 | } | |
6026 | ||
c4028958 | 6027 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
6028 | { |
6029 | int status; | |
c4028958 DH |
6030 | struct ocfs2_super *osb = |
6031 | container_of(work, struct ocfs2_super, | |
6032 | osb_truncate_log_wq.work); | |
ccd979bd MF |
6033 | |
6034 | mlog_entry_void(); | |
6035 | ||
6036 | status = ocfs2_flush_truncate_log(osb); | |
6037 | if (status < 0) | |
6038 | mlog_errno(status); | |
4d0ddb2c TM |
6039 | else |
6040 | ocfs2_init_inode_steal_slot(osb); | |
ccd979bd MF |
6041 | |
6042 | mlog_exit(status); | |
6043 | } | |
6044 | ||
6045 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
6046 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
6047 | int cancel) | |
6048 | { | |
6049 | if (osb->osb_tl_inode) { | |
6050 | /* We want to push off log flushes while truncates are | |
6051 | * still running. */ | |
6052 | if (cancel) | |
6053 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6054 | ||
6055 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
6056 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
6057 | } | |
6058 | } | |
6059 | ||
6060 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
6061 | int slot_num, | |
6062 | struct inode **tl_inode, | |
6063 | struct buffer_head **tl_bh) | |
6064 | { | |
6065 | int status; | |
6066 | struct inode *inode = NULL; | |
6067 | struct buffer_head *bh = NULL; | |
6068 | ||
6069 | inode = ocfs2_get_system_file_inode(osb, | |
6070 | TRUNCATE_LOG_SYSTEM_INODE, | |
6071 | slot_num); | |
6072 | if (!inode) { | |
6073 | status = -EINVAL; | |
6074 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
6075 | goto bail; | |
6076 | } | |
6077 | ||
b657c95c | 6078 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
6079 | if (status < 0) { |
6080 | iput(inode); | |
6081 | mlog_errno(status); | |
6082 | goto bail; | |
6083 | } | |
6084 | ||
6085 | *tl_inode = inode; | |
6086 | *tl_bh = bh; | |
6087 | bail: | |
6088 | mlog_exit(status); | |
6089 | return status; | |
6090 | } | |
6091 | ||
6092 | /* called during the 1st stage of node recovery. we stamp a clean | |
6093 | * truncate log and pass back a copy for processing later. if the | |
6094 | * truncate log does not require processing, a *tl_copy is set to | |
6095 | * NULL. */ | |
6096 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
6097 | int slot_num, | |
6098 | struct ocfs2_dinode **tl_copy) | |
6099 | { | |
6100 | int status; | |
6101 | struct inode *tl_inode = NULL; | |
6102 | struct buffer_head *tl_bh = NULL; | |
6103 | struct ocfs2_dinode *di; | |
6104 | struct ocfs2_truncate_log *tl; | |
6105 | ||
6106 | *tl_copy = NULL; | |
6107 | ||
6108 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
6109 | ||
6110 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
6111 | if (status < 0) { | |
6112 | mlog_errno(status); | |
6113 | goto bail; | |
6114 | } | |
6115 | ||
6116 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 6117 | |
10995aa2 JB |
6118 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
6119 | * validated by the underlying call to ocfs2_read_inode_block(), | |
6120 | * so any corruption is a code bug */ | |
6121 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
6122 | ||
6123 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
6124 | if (le16_to_cpu(tl->tl_used)) { |
6125 | mlog(0, "We'll have %u logs to recover\n", | |
6126 | le16_to_cpu(tl->tl_used)); | |
6127 | ||
6128 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
6129 | if (!(*tl_copy)) { | |
6130 | status = -ENOMEM; | |
6131 | mlog_errno(status); | |
6132 | goto bail; | |
6133 | } | |
6134 | ||
6135 | /* Assuming the write-out below goes well, this copy | |
6136 | * will be passed back to recovery for processing. */ | |
6137 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
6138 | ||
6139 | /* All we need to do to clear the truncate log is set | |
6140 | * tl_used. */ | |
6141 | tl->tl_used = 0; | |
6142 | ||
13723d00 | 6143 | ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check); |
8cb471e8 | 6144 | status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode)); |
ccd979bd MF |
6145 | if (status < 0) { |
6146 | mlog_errno(status); | |
6147 | goto bail; | |
6148 | } | |
6149 | } | |
6150 | ||
6151 | bail: | |
6152 | if (tl_inode) | |
6153 | iput(tl_inode); | |
a81cb88b | 6154 | brelse(tl_bh); |
ccd979bd MF |
6155 | |
6156 | if (status < 0 && (*tl_copy)) { | |
6157 | kfree(*tl_copy); | |
6158 | *tl_copy = NULL; | |
6159 | } | |
6160 | ||
6161 | mlog_exit(status); | |
6162 | return status; | |
6163 | } | |
6164 | ||
6165 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
6166 | struct ocfs2_dinode *tl_copy) | |
6167 | { | |
6168 | int status = 0; | |
6169 | int i; | |
6170 | unsigned int clusters, num_recs, start_cluster; | |
6171 | u64 start_blk; | |
1fabe148 | 6172 | handle_t *handle; |
ccd979bd MF |
6173 | struct inode *tl_inode = osb->osb_tl_inode; |
6174 | struct ocfs2_truncate_log *tl; | |
6175 | ||
6176 | mlog_entry_void(); | |
6177 | ||
6178 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
6179 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
6180 | return -EINVAL; | |
6181 | } | |
6182 | ||
6183 | tl = &tl_copy->id2.i_dealloc; | |
6184 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 6185 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 6186 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 6187 | |
1b1dcc1b | 6188 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6189 | for(i = 0; i < num_recs; i++) { |
6190 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6191 | status = __ocfs2_flush_truncate_log(osb); | |
6192 | if (status < 0) { | |
6193 | mlog_errno(status); | |
6194 | goto bail_up; | |
6195 | } | |
6196 | } | |
6197 | ||
65eff9cc | 6198 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
6199 | if (IS_ERR(handle)) { |
6200 | status = PTR_ERR(handle); | |
6201 | mlog_errno(status); | |
6202 | goto bail_up; | |
6203 | } | |
6204 | ||
6205 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
6206 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
6207 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
6208 | ||
6209 | status = ocfs2_truncate_log_append(osb, handle, | |
6210 | start_blk, clusters); | |
02dc1af4 | 6211 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6212 | if (status < 0) { |
6213 | mlog_errno(status); | |
6214 | goto bail_up; | |
6215 | } | |
6216 | } | |
6217 | ||
6218 | bail_up: | |
1b1dcc1b | 6219 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6220 | |
6221 | mlog_exit(status); | |
6222 | return status; | |
6223 | } | |
6224 | ||
6225 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
6226 | { | |
6227 | int status; | |
6228 | struct inode *tl_inode = osb->osb_tl_inode; | |
6229 | ||
6230 | mlog_entry_void(); | |
6231 | ||
6232 | if (tl_inode) { | |
6233 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6234 | flush_workqueue(ocfs2_wq); | |
6235 | ||
6236 | status = ocfs2_flush_truncate_log(osb); | |
6237 | if (status < 0) | |
6238 | mlog_errno(status); | |
6239 | ||
6240 | brelse(osb->osb_tl_bh); | |
6241 | iput(osb->osb_tl_inode); | |
6242 | } | |
6243 | ||
6244 | mlog_exit_void(); | |
6245 | } | |
6246 | ||
6247 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
6248 | { | |
6249 | int status; | |
6250 | struct inode *tl_inode = NULL; | |
6251 | struct buffer_head *tl_bh = NULL; | |
6252 | ||
6253 | mlog_entry_void(); | |
6254 | ||
6255 | status = ocfs2_get_truncate_log_info(osb, | |
6256 | osb->slot_num, | |
6257 | &tl_inode, | |
6258 | &tl_bh); | |
6259 | if (status < 0) | |
6260 | mlog_errno(status); | |
6261 | ||
6262 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
6263 | * osb->osb_tl_inode so we don't set any of the osb variables | |
6264 | * until we're sure all is well. */ | |
c4028958 DH |
6265 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
6266 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
6267 | osb->osb_tl_bh = tl_bh; |
6268 | osb->osb_tl_inode = tl_inode; | |
6269 | ||
6270 | mlog_exit(status); | |
6271 | return status; | |
6272 | } | |
6273 | ||
2b604351 MF |
6274 | /* |
6275 | * Delayed de-allocation of suballocator blocks. | |
6276 | * | |
6277 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
6278 | * | |
6279 | * The locking for this can get extremely complicated, especially when | |
6280 | * the suballocator inodes to delete from aren't known until deep | |
6281 | * within an unrelated codepath. | |
6282 | * | |
6283 | * ocfs2_extent_block structures are a good example of this - an inode | |
6284 | * btree could have been grown by any number of nodes each allocating | |
6285 | * out of their own suballoc inode. | |
6286 | * | |
6287 | * These structures allow the delay of block de-allocation until a | |
6288 | * later time, when locking of multiple cluster inodes won't cause | |
6289 | * deadlock. | |
6290 | */ | |
6291 | ||
6292 | /* | |
2891d290 TM |
6293 | * Describe a single bit freed from a suballocator. For the block |
6294 | * suballocators, it represents one block. For the global cluster | |
6295 | * allocator, it represents some clusters and free_bit indicates | |
6296 | * clusters number. | |
2b604351 MF |
6297 | */ |
6298 | struct ocfs2_cached_block_free { | |
6299 | struct ocfs2_cached_block_free *free_next; | |
6300 | u64 free_blk; | |
6301 | unsigned int free_bit; | |
6302 | }; | |
6303 | ||
6304 | struct ocfs2_per_slot_free_list { | |
6305 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
6306 | int f_inode_type; | |
6307 | int f_slot; | |
6308 | struct ocfs2_cached_block_free *f_first; | |
6309 | }; | |
6310 | ||
2891d290 TM |
6311 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
6312 | int sysfile_type, | |
6313 | int slot, | |
6314 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
6315 | { |
6316 | int ret; | |
6317 | u64 bg_blkno; | |
6318 | handle_t *handle; | |
6319 | struct inode *inode; | |
6320 | struct buffer_head *di_bh = NULL; | |
6321 | struct ocfs2_cached_block_free *tmp; | |
6322 | ||
6323 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
6324 | if (!inode) { | |
6325 | ret = -EINVAL; | |
6326 | mlog_errno(ret); | |
6327 | goto out; | |
6328 | } | |
6329 | ||
6330 | mutex_lock(&inode->i_mutex); | |
6331 | ||
e63aecb6 | 6332 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
6333 | if (ret) { |
6334 | mlog_errno(ret); | |
6335 | goto out_mutex; | |
6336 | } | |
6337 | ||
6338 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
6339 | if (IS_ERR(handle)) { | |
6340 | ret = PTR_ERR(handle); | |
6341 | mlog_errno(ret); | |
6342 | goto out_unlock; | |
6343 | } | |
6344 | ||
6345 | while (head) { | |
6346 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
6347 | head->free_bit); | |
6348 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
6349 | head->free_bit, (unsigned long long)head->free_blk); | |
6350 | ||
6351 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
6352 | head->free_bit, bg_blkno, 1); | |
6353 | if (ret) { | |
6354 | mlog_errno(ret); | |
6355 | goto out_journal; | |
6356 | } | |
6357 | ||
6358 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
6359 | if (ret) { | |
6360 | mlog_errno(ret); | |
6361 | goto out_journal; | |
6362 | } | |
6363 | ||
6364 | tmp = head; | |
6365 | head = head->free_next; | |
6366 | kfree(tmp); | |
6367 | } | |
6368 | ||
6369 | out_journal: | |
6370 | ocfs2_commit_trans(osb, handle); | |
6371 | ||
6372 | out_unlock: | |
e63aecb6 | 6373 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
6374 | brelse(di_bh); |
6375 | out_mutex: | |
6376 | mutex_unlock(&inode->i_mutex); | |
6377 | iput(inode); | |
6378 | out: | |
6379 | while(head) { | |
6380 | /* Premature exit may have left some dangling items. */ | |
6381 | tmp = head; | |
6382 | head = head->free_next; | |
6383 | kfree(tmp); | |
6384 | } | |
6385 | ||
6386 | return ret; | |
6387 | } | |
6388 | ||
2891d290 TM |
6389 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6390 | u64 blkno, unsigned int bit) | |
6391 | { | |
6392 | int ret = 0; | |
6393 | struct ocfs2_cached_block_free *item; | |
6394 | ||
6395 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6396 | if (item == NULL) { | |
6397 | ret = -ENOMEM; | |
6398 | mlog_errno(ret); | |
6399 | return ret; | |
6400 | } | |
6401 | ||
6402 | mlog(0, "Insert clusters: (bit %u, blk %llu)\n", | |
6403 | bit, (unsigned long long)blkno); | |
6404 | ||
6405 | item->free_blk = blkno; | |
6406 | item->free_bit = bit; | |
6407 | item->free_next = ctxt->c_global_allocator; | |
6408 | ||
6409 | ctxt->c_global_allocator = item; | |
6410 | return ret; | |
6411 | } | |
6412 | ||
6413 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
6414 | struct ocfs2_cached_block_free *head) | |
6415 | { | |
6416 | struct ocfs2_cached_block_free *tmp; | |
6417 | struct inode *tl_inode = osb->osb_tl_inode; | |
6418 | handle_t *handle; | |
6419 | int ret = 0; | |
6420 | ||
6421 | mutex_lock(&tl_inode->i_mutex); | |
6422 | ||
6423 | while (head) { | |
6424 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6425 | ret = __ocfs2_flush_truncate_log(osb); | |
6426 | if (ret < 0) { | |
6427 | mlog_errno(ret); | |
6428 | break; | |
6429 | } | |
6430 | } | |
6431 | ||
6432 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6433 | if (IS_ERR(handle)) { | |
6434 | ret = PTR_ERR(handle); | |
6435 | mlog_errno(ret); | |
6436 | break; | |
6437 | } | |
6438 | ||
6439 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6440 | head->free_bit); | |
6441 | ||
6442 | ocfs2_commit_trans(osb, handle); | |
6443 | tmp = head; | |
6444 | head = head->free_next; | |
6445 | kfree(tmp); | |
6446 | ||
6447 | if (ret < 0) { | |
6448 | mlog_errno(ret); | |
6449 | break; | |
6450 | } | |
6451 | } | |
6452 | ||
6453 | mutex_unlock(&tl_inode->i_mutex); | |
6454 | ||
6455 | while (head) { | |
6456 | /* Premature exit may have left some dangling items. */ | |
6457 | tmp = head; | |
6458 | head = head->free_next; | |
6459 | kfree(tmp); | |
6460 | } | |
6461 | ||
6462 | return ret; | |
6463 | } | |
6464 | ||
2b604351 MF |
6465 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6466 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6467 | { | |
6468 | int ret = 0, ret2; | |
6469 | struct ocfs2_per_slot_free_list *fl; | |
6470 | ||
6471 | if (!ctxt) | |
6472 | return 0; | |
6473 | ||
6474 | while (ctxt->c_first_suballocator) { | |
6475 | fl = ctxt->c_first_suballocator; | |
6476 | ||
6477 | if (fl->f_first) { | |
6478 | mlog(0, "Free items: (type %u, slot %d)\n", | |
6479 | fl->f_inode_type, fl->f_slot); | |
2891d290 TM |
6480 | ret2 = ocfs2_free_cached_blocks(osb, |
6481 | fl->f_inode_type, | |
6482 | fl->f_slot, | |
6483 | fl->f_first); | |
2b604351 MF |
6484 | if (ret2) |
6485 | mlog_errno(ret2); | |
6486 | if (!ret) | |
6487 | ret = ret2; | |
6488 | } | |
6489 | ||
6490 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6491 | kfree(fl); | |
6492 | } | |
6493 | ||
2891d290 TM |
6494 | if (ctxt->c_global_allocator) { |
6495 | ret2 = ocfs2_free_cached_clusters(osb, | |
6496 | ctxt->c_global_allocator); | |
6497 | if (ret2) | |
6498 | mlog_errno(ret2); | |
6499 | if (!ret) | |
6500 | ret = ret2; | |
6501 | ||
6502 | ctxt->c_global_allocator = NULL; | |
6503 | } | |
6504 | ||
2b604351 MF |
6505 | return ret; |
6506 | } | |
6507 | ||
6508 | static struct ocfs2_per_slot_free_list * | |
6509 | ocfs2_find_per_slot_free_list(int type, | |
6510 | int slot, | |
6511 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6512 | { | |
6513 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6514 | ||
6515 | while (fl) { | |
6516 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6517 | return fl; | |
6518 | ||
6519 | fl = fl->f_next_suballocator; | |
6520 | } | |
6521 | ||
6522 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6523 | if (fl) { | |
6524 | fl->f_inode_type = type; | |
6525 | fl->f_slot = slot; | |
6526 | fl->f_first = NULL; | |
6527 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6528 | ||
6529 | ctxt->c_first_suballocator = fl; | |
6530 | } | |
6531 | return fl; | |
6532 | } | |
6533 | ||
1823cb0b TM |
6534 | int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6535 | int type, int slot, u64 blkno, | |
6536 | unsigned int bit) | |
2b604351 MF |
6537 | { |
6538 | int ret; | |
6539 | struct ocfs2_per_slot_free_list *fl; | |
6540 | struct ocfs2_cached_block_free *item; | |
6541 | ||
6542 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6543 | if (fl == NULL) { | |
6544 | ret = -ENOMEM; | |
6545 | mlog_errno(ret); | |
6546 | goto out; | |
6547 | } | |
6548 | ||
6549 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6550 | if (item == NULL) { | |
6551 | ret = -ENOMEM; | |
6552 | mlog_errno(ret); | |
6553 | goto out; | |
6554 | } | |
6555 | ||
6556 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6557 | type, slot, bit, (unsigned long long)blkno); | |
6558 | ||
6559 | item->free_blk = blkno; | |
6560 | item->free_bit = bit; | |
6561 | item->free_next = fl->f_first; | |
6562 | ||
6563 | fl->f_first = item; | |
6564 | ||
6565 | ret = 0; | |
6566 | out: | |
6567 | return ret; | |
6568 | } | |
6569 | ||
59a5e416 MF |
6570 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6571 | struct ocfs2_extent_block *eb) | |
6572 | { | |
6573 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6574 | le16_to_cpu(eb->h_suballoc_slot), | |
6575 | le64_to_cpu(eb->h_blkno), | |
6576 | le16_to_cpu(eb->h_suballoc_bit)); | |
6577 | } | |
6578 | ||
ccd979bd MF |
6579 | /* This function will figure out whether the currently last extent |
6580 | * block will be deleted, and if it will, what the new last extent | |
6581 | * block will be so we can update his h_next_leaf_blk field, as well | |
6582 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6583 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6584 | unsigned int clusters_to_del, |
dcd0538f | 6585 | struct ocfs2_path *path, |
ccd979bd MF |
6586 | struct buffer_head **new_last_eb) |
6587 | { | |
3a0782d0 | 6588 | int next_free, ret = 0; |
dcd0538f | 6589 | u32 cpos; |
3a0782d0 | 6590 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6591 | struct ocfs2_extent_block *eb; |
6592 | struct ocfs2_extent_list *el; | |
6593 | struct buffer_head *bh = NULL; | |
6594 | ||
6595 | *new_last_eb = NULL; | |
6596 | ||
ccd979bd | 6597 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6598 | if (!path->p_tree_depth) |
6599 | goto out; | |
ccd979bd MF |
6600 | |
6601 | /* trunc to zero special case - this makes tree_depth = 0 | |
6602 | * regardless of what it is. */ | |
3a0782d0 | 6603 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6604 | goto out; |
ccd979bd | 6605 | |
dcd0538f | 6606 | el = path_leaf_el(path); |
ccd979bd MF |
6607 | BUG_ON(!el->l_next_free_rec); |
6608 | ||
3a0782d0 MF |
6609 | /* |
6610 | * Make sure that this extent list will actually be empty | |
6611 | * after we clear away the data. We can shortcut out if | |
6612 | * there's more than one non-empty extent in the | |
6613 | * list. Otherwise, a check of the remaining extent is | |
6614 | * necessary. | |
6615 | */ | |
6616 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6617 | rec = NULL; | |
dcd0538f | 6618 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6619 | if (next_free > 2) |
dcd0538f | 6620 | goto out; |
3a0782d0 MF |
6621 | |
6622 | /* We may have a valid extent in index 1, check it. */ | |
6623 | if (next_free == 2) | |
6624 | rec = &el->l_recs[1]; | |
6625 | ||
6626 | /* | |
6627 | * Fall through - no more nonempty extents, so we want | |
6628 | * to delete this leaf. | |
6629 | */ | |
6630 | } else { | |
6631 | if (next_free > 1) | |
6632 | goto out; | |
6633 | ||
6634 | rec = &el->l_recs[0]; | |
6635 | } | |
6636 | ||
6637 | if (rec) { | |
6638 | /* | |
6639 | * Check it we'll only be trimming off the end of this | |
6640 | * cluster. | |
6641 | */ | |
e48edee2 | 6642 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6643 | goto out; |
6644 | } | |
ccd979bd | 6645 | |
dcd0538f MF |
6646 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6647 | if (ret) { | |
6648 | mlog_errno(ret); | |
6649 | goto out; | |
6650 | } | |
ccd979bd | 6651 | |
facdb77f | 6652 | ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh); |
dcd0538f MF |
6653 | if (ret) { |
6654 | mlog_errno(ret); | |
6655 | goto out; | |
6656 | } | |
ccd979bd | 6657 | |
dcd0538f MF |
6658 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6659 | el = &eb->h_list; | |
5e96581a JB |
6660 | |
6661 | /* ocfs2_find_leaf() gets the eb from ocfs2_read_extent_block(). | |
6662 | * Any corruption is a code bug. */ | |
6663 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
6664 | |
6665 | *new_last_eb = bh; | |
6666 | get_bh(*new_last_eb); | |
dcd0538f MF |
6667 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6668 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6669 | out: | |
6670 | brelse(bh); | |
ccd979bd | 6671 | |
dcd0538f | 6672 | return ret; |
ccd979bd MF |
6673 | } |
6674 | ||
3a0782d0 MF |
6675 | /* |
6676 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6677 | * during truncate. | |
6678 | * | |
6679 | * The caller needs to: | |
6680 | * - start journaling of each path component. | |
6681 | * - compute and fully set up any new last ext block | |
6682 | */ | |
6683 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6684 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
bcbbb24a | 6685 | u32 clusters_to_del, u64 *delete_start, u8 *flags) |
3a0782d0 MF |
6686 | { |
6687 | int ret, i, index = path->p_tree_depth; | |
6688 | u32 new_edge = 0; | |
6689 | u64 deleted_eb = 0; | |
6690 | struct buffer_head *bh; | |
6691 | struct ocfs2_extent_list *el; | |
6692 | struct ocfs2_extent_rec *rec; | |
6693 | ||
6694 | *delete_start = 0; | |
bcbbb24a | 6695 | *flags = 0; |
3a0782d0 MF |
6696 | |
6697 | while (index >= 0) { | |
6698 | bh = path->p_node[index].bh; | |
6699 | el = path->p_node[index].el; | |
6700 | ||
6701 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6702 | index, (unsigned long long)bh->b_blocknr); | |
6703 | ||
6704 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6705 | ||
6706 | if (index != | |
6707 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6708 | ocfs2_error(inode->i_sb, | |
6709 | "Inode %lu has invalid ext. block %llu", | |
6710 | inode->i_ino, | |
6711 | (unsigned long long)bh->b_blocknr); | |
6712 | ret = -EROFS; | |
6713 | goto out; | |
6714 | } | |
6715 | ||
6716 | find_tail_record: | |
6717 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6718 | rec = &el->l_recs[i]; | |
6719 | ||
6720 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6721 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6722 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6723 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6724 | le16_to_cpu(el->l_next_free_rec)); | |
6725 | ||
e48edee2 | 6726 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6727 | |
6728 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6729 | /* | |
6730 | * If the leaf block contains a single empty | |
6731 | * extent and no records, we can just remove | |
6732 | * the block. | |
6733 | */ | |
6734 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6735 | memset(rec, 0, | |
6736 | sizeof(struct ocfs2_extent_rec)); | |
6737 | el->l_next_free_rec = cpu_to_le16(0); | |
6738 | ||
6739 | goto delete; | |
6740 | } | |
6741 | ||
6742 | /* | |
6743 | * Remove any empty extents by shifting things | |
6744 | * left. That should make life much easier on | |
6745 | * the code below. This condition is rare | |
6746 | * enough that we shouldn't see a performance | |
6747 | * hit. | |
6748 | */ | |
6749 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6750 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6751 | ||
6752 | for(i = 0; | |
6753 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6754 | el->l_recs[i] = el->l_recs[i + 1]; | |
6755 | ||
6756 | memset(&el->l_recs[i], 0, | |
6757 | sizeof(struct ocfs2_extent_rec)); | |
6758 | ||
6759 | /* | |
6760 | * We've modified our extent list. The | |
6761 | * simplest way to handle this change | |
6762 | * is to being the search from the | |
6763 | * start again. | |
6764 | */ | |
6765 | goto find_tail_record; | |
6766 | } | |
6767 | ||
e48edee2 | 6768 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6769 | |
6770 | /* | |
6771 | * We'll use "new_edge" on our way back up the | |
6772 | * tree to know what our rightmost cpos is. | |
6773 | */ | |
e48edee2 | 6774 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6775 | new_edge += le32_to_cpu(rec->e_cpos); |
6776 | ||
6777 | /* | |
6778 | * The caller will use this to delete data blocks. | |
6779 | */ | |
6780 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6781 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6782 | le16_to_cpu(rec->e_leaf_clusters)); |
bcbbb24a | 6783 | *flags = rec->e_flags; |
3a0782d0 MF |
6784 | |
6785 | /* | |
6786 | * If it's now empty, remove this record. | |
6787 | */ | |
e48edee2 | 6788 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6789 | memset(rec, 0, |
6790 | sizeof(struct ocfs2_extent_rec)); | |
6791 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6792 | } | |
6793 | } else { | |
6794 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6795 | memset(rec, 0, | |
6796 | sizeof(struct ocfs2_extent_rec)); | |
6797 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6798 | ||
6799 | goto delete; | |
6800 | } | |
6801 | ||
6802 | /* Can this actually happen? */ | |
6803 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6804 | goto delete; | |
6805 | ||
6806 | /* | |
6807 | * We never actually deleted any clusters | |
6808 | * because our leaf was empty. There's no | |
6809 | * reason to adjust the rightmost edge then. | |
6810 | */ | |
6811 | if (new_edge == 0) | |
6812 | goto delete; | |
6813 | ||
e48edee2 MF |
6814 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6815 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6816 | -le32_to_cpu(rec->e_cpos)); |
6817 | ||
6818 | /* | |
6819 | * A deleted child record should have been | |
6820 | * caught above. | |
6821 | */ | |
e48edee2 | 6822 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6823 | } |
6824 | ||
6825 | delete: | |
6826 | ret = ocfs2_journal_dirty(handle, bh); | |
6827 | if (ret) { | |
6828 | mlog_errno(ret); | |
6829 | goto out; | |
6830 | } | |
6831 | ||
6832 | mlog(0, "extent list container %llu, after: record %d: " | |
6833 | "(%u, %u, %llu), next = %u.\n", | |
6834 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6835 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6836 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6837 | le16_to_cpu(el->l_next_free_rec)); | |
6838 | ||
6839 | /* | |
6840 | * We must be careful to only attempt delete of an | |
6841 | * extent block (and not the root inode block). | |
6842 | */ | |
6843 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6844 | struct ocfs2_extent_block *eb = | |
6845 | (struct ocfs2_extent_block *)bh->b_data; | |
6846 | ||
6847 | /* | |
6848 | * Save this for use when processing the | |
6849 | * parent block. | |
6850 | */ | |
6851 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6852 | ||
6853 | mlog(0, "deleting this extent block.\n"); | |
6854 | ||
8cb471e8 | 6855 | ocfs2_remove_from_cache(INODE_CACHE(inode), bh); |
3a0782d0 | 6856 | |
e48edee2 | 6857 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6858 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6859 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6860 | ||
59a5e416 MF |
6861 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6862 | /* An error here is not fatal. */ | |
6863 | if (ret < 0) | |
6864 | mlog_errno(ret); | |
3a0782d0 MF |
6865 | } else { |
6866 | deleted_eb = 0; | |
6867 | } | |
6868 | ||
6869 | index--; | |
6870 | } | |
6871 | ||
6872 | ret = 0; | |
6873 | out: | |
6874 | return ret; | |
6875 | } | |
6876 | ||
ccd979bd MF |
6877 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6878 | unsigned int clusters_to_del, | |
6879 | struct inode *inode, | |
6880 | struct buffer_head *fe_bh, | |
1fabe148 | 6881 | handle_t *handle, |
dcd0538f | 6882 | struct ocfs2_truncate_context *tc, |
bcbbb24a TM |
6883 | struct ocfs2_path *path, |
6884 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 6885 | { |
3a0782d0 | 6886 | int status; |
ccd979bd | 6887 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6888 | struct ocfs2_extent_block *last_eb = NULL; |
6889 | struct ocfs2_extent_list *el; | |
ccd979bd | 6890 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd | 6891 | u64 delete_blk = 0; |
bcbbb24a | 6892 | u8 rec_flags; |
ccd979bd MF |
6893 | |
6894 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6895 | ||
3a0782d0 | 6896 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6897 | path, &last_eb_bh); |
ccd979bd MF |
6898 | if (status < 0) { |
6899 | mlog_errno(status); | |
6900 | goto bail; | |
6901 | } | |
dcd0538f MF |
6902 | |
6903 | /* | |
6904 | * Each component will be touched, so we might as well journal | |
6905 | * here to avoid having to handle errors later. | |
6906 | */ | |
0cf2f763 | 6907 | status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path); |
3a0782d0 MF |
6908 | if (status < 0) { |
6909 | mlog_errno(status); | |
6910 | goto bail; | |
dcd0538f MF |
6911 | } |
6912 | ||
6913 | if (last_eb_bh) { | |
0cf2f763 | 6914 | status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh, |
13723d00 | 6915 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
6916 | if (status < 0) { |
6917 | mlog_errno(status); | |
6918 | goto bail; | |
6919 | } | |
6920 | ||
ccd979bd | 6921 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6922 | } |
ccd979bd | 6923 | |
dcd0538f MF |
6924 | el = &(fe->id2.i_list); |
6925 | ||
6926 | /* | |
6927 | * Lower levels depend on this never happening, but it's best | |
6928 | * to check it up here before changing the tree. | |
6929 | */ | |
e48edee2 | 6930 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6931 | ocfs2_error(inode->i_sb, |
6932 | "Inode %lu has an empty extent record, depth %u\n", | |
6933 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6934 | status = -EROFS; |
ccd979bd MF |
6935 | goto bail; |
6936 | } | |
ccd979bd | 6937 | |
a90714c1 JK |
6938 | vfs_dq_free_space_nodirty(inode, |
6939 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_del)); | |
ccd979bd MF |
6940 | spin_lock(&OCFS2_I(inode)->ip_lock); |
6941 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6942 | clusters_to_del; | |
6943 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6944 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6945 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6946 | |
3a0782d0 | 6947 | status = ocfs2_trim_tree(inode, path, handle, tc, |
bcbbb24a | 6948 | clusters_to_del, &delete_blk, &rec_flags); |
3a0782d0 MF |
6949 | if (status) { |
6950 | mlog_errno(status); | |
6951 | goto bail; | |
ccd979bd MF |
6952 | } |
6953 | ||
dcd0538f | 6954 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6955 | /* trunc to zero is a special case. */ |
6956 | el->l_tree_depth = 0; | |
6957 | fe->i_last_eb_blk = 0; | |
6958 | } else if (last_eb) | |
6959 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6960 | ||
6961 | status = ocfs2_journal_dirty(handle, fe_bh); | |
6962 | if (status < 0) { | |
6963 | mlog_errno(status); | |
6964 | goto bail; | |
6965 | } | |
6966 | ||
6967 | if (last_eb) { | |
6968 | /* If there will be a new last extent block, then by | |
6969 | * definition, there cannot be any leaves to the right of | |
6970 | * him. */ | |
ccd979bd MF |
6971 | last_eb->h_next_leaf_blk = 0; |
6972 | status = ocfs2_journal_dirty(handle, last_eb_bh); | |
6973 | if (status < 0) { | |
6974 | mlog_errno(status); | |
6975 | goto bail; | |
6976 | } | |
6977 | } | |
6978 | ||
3a0782d0 | 6979 | if (delete_blk) { |
bcbbb24a TM |
6980 | if (rec_flags & OCFS2_EXT_REFCOUNTED) |
6981 | status = ocfs2_decrease_refcount(inode, handle, | |
6982 | ocfs2_blocks_to_clusters(osb->sb, | |
6983 | delete_blk), | |
6984 | clusters_to_del, meta_ac, | |
6ae23c55 | 6985 | &tc->tc_dealloc, 1); |
bcbbb24a TM |
6986 | else |
6987 | status = ocfs2_truncate_log_append(osb, handle, | |
6988 | delete_blk, | |
6989 | clusters_to_del); | |
ccd979bd MF |
6990 | if (status < 0) { |
6991 | mlog_errno(status); | |
6992 | goto bail; | |
6993 | } | |
ccd979bd MF |
6994 | } |
6995 | status = 0; | |
6996 | bail: | |
60e2ec48 | 6997 | brelse(last_eb_bh); |
ccd979bd MF |
6998 | mlog_exit(status); |
6999 | return status; | |
7000 | } | |
7001 | ||
2b4e30fb | 7002 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
7003 | { |
7004 | set_buffer_uptodate(bh); | |
7005 | mark_buffer_dirty(bh); | |
7006 | return 0; | |
7007 | } | |
7008 | ||
6f70fa51 TM |
7009 | void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
7010 | unsigned int from, unsigned int to, | |
7011 | struct page *page, int zero, u64 *phys) | |
1d410a6e MF |
7012 | { |
7013 | int ret, partial = 0; | |
7014 | ||
7015 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
7016 | if (ret) | |
7017 | mlog_errno(ret); | |
7018 | ||
7019 | if (zero) | |
eebd2aa3 | 7020 | zero_user_segment(page, from, to); |
1d410a6e MF |
7021 | |
7022 | /* | |
7023 | * Need to set the buffers we zero'd into uptodate | |
7024 | * here if they aren't - ocfs2_map_page_blocks() | |
7025 | * might've skipped some | |
7026 | */ | |
2b4e30fb JB |
7027 | ret = walk_page_buffers(handle, page_buffers(page), |
7028 | from, to, &partial, | |
7029 | ocfs2_zero_func); | |
7030 | if (ret < 0) | |
7031 | mlog_errno(ret); | |
7032 | else if (ocfs2_should_order_data(inode)) { | |
7033 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
1d410a6e MF |
7034 | if (ret < 0) |
7035 | mlog_errno(ret); | |
7036 | } | |
7037 | ||
7038 | if (!partial) | |
7039 | SetPageUptodate(page); | |
7040 | ||
7041 | flush_dcache_page(page); | |
7042 | } | |
7043 | ||
35edec1d MF |
7044 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
7045 | loff_t end, struct page **pages, | |
7046 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 7047 | { |
1d410a6e | 7048 | int i; |
60b11392 MF |
7049 | struct page *page; |
7050 | unsigned int from, to = PAGE_CACHE_SIZE; | |
7051 | struct super_block *sb = inode->i_sb; | |
7052 | ||
7053 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
7054 | ||
7055 | if (numpages == 0) | |
7056 | goto out; | |
7057 | ||
35edec1d | 7058 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
7059 | for(i = 0; i < numpages; i++) { |
7060 | page = pages[i]; | |
7061 | ||
35edec1d MF |
7062 | from = start & (PAGE_CACHE_SIZE - 1); |
7063 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
7064 | to = end & (PAGE_CACHE_SIZE - 1); | |
7065 | ||
60b11392 MF |
7066 | BUG_ON(from > PAGE_CACHE_SIZE); |
7067 | BUG_ON(to > PAGE_CACHE_SIZE); | |
7068 | ||
1d410a6e MF |
7069 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
7070 | &phys); | |
60b11392 | 7071 | |
35edec1d | 7072 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
7073 | } |
7074 | out: | |
1d410a6e MF |
7075 | if (pages) |
7076 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
7077 | } |
7078 | ||
6f70fa51 TM |
7079 | int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end, |
7080 | struct page **pages, int *num) | |
60b11392 | 7081 | { |
1d410a6e | 7082 | int numpages, ret = 0; |
60b11392 MF |
7083 | struct address_space *mapping = inode->i_mapping; |
7084 | unsigned long index; | |
35edec1d | 7085 | loff_t last_page_bytes; |
60b11392 | 7086 | |
35edec1d | 7087 | BUG_ON(start > end); |
60b11392 | 7088 | |
1d410a6e | 7089 | numpages = 0; |
35edec1d MF |
7090 | last_page_bytes = PAGE_ALIGN(end); |
7091 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
7092 | do { |
7093 | pages[numpages] = grab_cache_page(mapping, index); | |
7094 | if (!pages[numpages]) { | |
7095 | ret = -ENOMEM; | |
7096 | mlog_errno(ret); | |
7097 | goto out; | |
7098 | } | |
7099 | ||
7100 | numpages++; | |
7101 | index++; | |
35edec1d | 7102 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
7103 | |
7104 | out: | |
7105 | if (ret != 0) { | |
1d410a6e MF |
7106 | if (pages) |
7107 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
7108 | numpages = 0; |
7109 | } | |
7110 | ||
7111 | *num = numpages; | |
7112 | ||
7113 | return ret; | |
7114 | } | |
7115 | ||
6f70fa51 TM |
7116 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
7117 | struct page **pages, int *num) | |
7118 | { | |
7119 | struct super_block *sb = inode->i_sb; | |
7120 | ||
7121 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != | |
7122 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
7123 | ||
7124 | return ocfs2_grab_pages(inode, start, end, pages, num); | |
7125 | } | |
7126 | ||
60b11392 MF |
7127 | /* |
7128 | * Zero the area past i_size but still within an allocated | |
7129 | * cluster. This avoids exposing nonzero data on subsequent file | |
7130 | * extends. | |
7131 | * | |
7132 | * We need to call this before i_size is updated on the inode because | |
7133 | * otherwise block_write_full_page() will skip writeout of pages past | |
7134 | * i_size. The new_i_size parameter is passed for this reason. | |
7135 | */ | |
35edec1d MF |
7136 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
7137 | u64 range_start, u64 range_end) | |
60b11392 | 7138 | { |
1d410a6e | 7139 | int ret = 0, numpages; |
60b11392 MF |
7140 | struct page **pages = NULL; |
7141 | u64 phys; | |
1d410a6e MF |
7142 | unsigned int ext_flags; |
7143 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
7144 | |
7145 | /* | |
7146 | * File systems which don't support sparse files zero on every | |
7147 | * extend. | |
7148 | */ | |
1d410a6e | 7149 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
7150 | return 0; |
7151 | ||
1d410a6e | 7152 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
7153 | sizeof(struct page *), GFP_NOFS); |
7154 | if (pages == NULL) { | |
7155 | ret = -ENOMEM; | |
7156 | mlog_errno(ret); | |
7157 | goto out; | |
7158 | } | |
7159 | ||
1d410a6e MF |
7160 | if (range_start == range_end) |
7161 | goto out; | |
7162 | ||
7163 | ret = ocfs2_extent_map_get_blocks(inode, | |
7164 | range_start >> sb->s_blocksize_bits, | |
7165 | &phys, NULL, &ext_flags); | |
60b11392 MF |
7166 | if (ret) { |
7167 | mlog_errno(ret); | |
7168 | goto out; | |
7169 | } | |
7170 | ||
1d410a6e MF |
7171 | /* |
7172 | * Tail is a hole, or is marked unwritten. In either case, we | |
7173 | * can count on read and write to return/push zero's. | |
7174 | */ | |
7175 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
7176 | goto out; |
7177 | ||
1d410a6e MF |
7178 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
7179 | &numpages); | |
7180 | if (ret) { | |
7181 | mlog_errno(ret); | |
7182 | goto out; | |
7183 | } | |
7184 | ||
35edec1d MF |
7185 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
7186 | numpages, phys, handle); | |
60b11392 MF |
7187 | |
7188 | /* | |
7189 | * Initiate writeout of the pages we zero'd here. We don't | |
7190 | * wait on them - the truncate_inode_pages() call later will | |
7191 | * do that for us. | |
7192 | */ | |
35edec1d MF |
7193 | ret = do_sync_mapping_range(inode->i_mapping, range_start, |
7194 | range_end - 1, SYNC_FILE_RANGE_WRITE); | |
60b11392 MF |
7195 | if (ret) |
7196 | mlog_errno(ret); | |
7197 | ||
7198 | out: | |
7199 | if (pages) | |
7200 | kfree(pages); | |
7201 | ||
7202 | return ret; | |
7203 | } | |
7204 | ||
fdd77704 TY |
7205 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
7206 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
7207 | { |
7208 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 7209 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 7210 | |
fdd77704 TY |
7211 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
7212 | memset(&di->id2, 0, blocksize - | |
7213 | offsetof(struct ocfs2_dinode, id2) - | |
7214 | xattrsize); | |
7215 | else | |
7216 | memset(&di->id2, 0, blocksize - | |
7217 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
7218 | } |
7219 | ||
5b6a3a2b MF |
7220 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
7221 | struct ocfs2_dinode *di) | |
7222 | { | |
fdd77704 | 7223 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
7224 | di->id2.i_list.l_tree_depth = 0; |
7225 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
7226 | di->id2.i_list.l_count = cpu_to_le16( |
7227 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
7228 | } |
7229 | ||
1afc32b9 MF |
7230 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
7231 | { | |
7232 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7233 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7234 | ||
7235 | spin_lock(&oi->ip_lock); | |
7236 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
7237 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7238 | spin_unlock(&oi->ip_lock); | |
7239 | ||
7240 | /* | |
7241 | * We clear the entire i_data structure here so that all | |
7242 | * fields can be properly initialized. | |
7243 | */ | |
fdd77704 | 7244 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 7245 | |
fdd77704 TY |
7246 | idata->id_count = cpu_to_le16( |
7247 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
7248 | } |
7249 | ||
7250 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
7251 | struct buffer_head *di_bh) | |
7252 | { | |
7253 | int ret, i, has_data, num_pages = 0; | |
7254 | handle_t *handle; | |
7255 | u64 uninitialized_var(block); | |
7256 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7257 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7258 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
7259 | struct ocfs2_alloc_context *data_ac = NULL; |
7260 | struct page **pages = NULL; | |
7261 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 7262 | struct ocfs2_extent_tree et; |
a90714c1 | 7263 | int did_quota = 0; |
1afc32b9 MF |
7264 | |
7265 | has_data = i_size_read(inode) ? 1 : 0; | |
7266 | ||
7267 | if (has_data) { | |
7268 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
7269 | sizeof(struct page *), GFP_NOFS); | |
7270 | if (pages == NULL) { | |
7271 | ret = -ENOMEM; | |
7272 | mlog_errno(ret); | |
7273 | goto out; | |
7274 | } | |
7275 | ||
7276 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
7277 | if (ret) { | |
7278 | mlog_errno(ret); | |
7279 | goto out; | |
7280 | } | |
7281 | } | |
7282 | ||
a90714c1 JK |
7283 | handle = ocfs2_start_trans(osb, |
7284 | ocfs2_inline_to_extents_credits(osb->sb)); | |
1afc32b9 MF |
7285 | if (IS_ERR(handle)) { |
7286 | ret = PTR_ERR(handle); | |
7287 | mlog_errno(ret); | |
7288 | goto out_unlock; | |
7289 | } | |
7290 | ||
0cf2f763 | 7291 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7292 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7293 | if (ret) { |
7294 | mlog_errno(ret); | |
7295 | goto out_commit; | |
7296 | } | |
7297 | ||
7298 | if (has_data) { | |
7299 | u32 bit_off, num; | |
7300 | unsigned int page_end; | |
7301 | u64 phys; | |
7302 | ||
a90714c1 JK |
7303 | if (vfs_dq_alloc_space_nodirty(inode, |
7304 | ocfs2_clusters_to_bytes(osb->sb, 1))) { | |
7305 | ret = -EDQUOT; | |
7306 | goto out_commit; | |
7307 | } | |
7308 | did_quota = 1; | |
7309 | ||
1afc32b9 MF |
7310 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, |
7311 | &num); | |
7312 | if (ret) { | |
7313 | mlog_errno(ret); | |
7314 | goto out_commit; | |
7315 | } | |
7316 | ||
7317 | /* | |
7318 | * Save two copies, one for insert, and one that can | |
7319 | * be changed by ocfs2_map_and_dirty_page() below. | |
7320 | */ | |
7321 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
7322 | ||
7323 | /* | |
7324 | * Non sparse file systems zero on extend, so no need | |
7325 | * to do that now. | |
7326 | */ | |
7327 | if (!ocfs2_sparse_alloc(osb) && | |
7328 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
7329 | end = PAGE_CACHE_SIZE; | |
7330 | ||
7331 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
7332 | if (ret) { | |
7333 | mlog_errno(ret); | |
7334 | goto out_commit; | |
7335 | } | |
7336 | ||
7337 | /* | |
7338 | * This should populate the 1st page for us and mark | |
7339 | * it up to date. | |
7340 | */ | |
7341 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
7342 | if (ret) { | |
7343 | mlog_errno(ret); | |
7344 | goto out_commit; | |
7345 | } | |
7346 | ||
7347 | page_end = PAGE_CACHE_SIZE; | |
7348 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
7349 | page_end = osb->s_clustersize; | |
7350 | ||
7351 | for (i = 0; i < num_pages; i++) | |
7352 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
7353 | pages[i], i > 0, &phys); | |
7354 | } | |
7355 | ||
7356 | spin_lock(&oi->ip_lock); | |
7357 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
7358 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7359 | spin_unlock(&oi->ip_lock); | |
7360 | ||
5b6a3a2b | 7361 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
7362 | |
7363 | ocfs2_journal_dirty(handle, di_bh); | |
7364 | ||
7365 | if (has_data) { | |
7366 | /* | |
7367 | * An error at this point should be extremely rare. If | |
7368 | * this proves to be false, we could always re-build | |
7369 | * the in-inode data from our pages. | |
7370 | */ | |
5e404e9e | 7371 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
cc79d8c1 | 7372 | ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL); |
1afc32b9 MF |
7373 | if (ret) { |
7374 | mlog_errno(ret); | |
7375 | goto out_commit; | |
7376 | } | |
7377 | ||
7378 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7379 | } | |
7380 | ||
7381 | out_commit: | |
a90714c1 JK |
7382 | if (ret < 0 && did_quota) |
7383 | vfs_dq_free_space_nodirty(inode, | |
7384 | ocfs2_clusters_to_bytes(osb->sb, 1)); | |
7385 | ||
1afc32b9 MF |
7386 | ocfs2_commit_trans(osb, handle); |
7387 | ||
7388 | out_unlock: | |
7389 | if (data_ac) | |
7390 | ocfs2_free_alloc_context(data_ac); | |
7391 | ||
7392 | out: | |
7393 | if (pages) { | |
7394 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
7395 | kfree(pages); | |
7396 | } | |
7397 | ||
7398 | return ret; | |
7399 | } | |
7400 | ||
ccd979bd MF |
7401 | /* |
7402 | * It is expected, that by the time you call this function, | |
7403 | * inode->i_size and fe->i_size have been adjusted. | |
7404 | * | |
7405 | * WARNING: This will kfree the truncate context | |
7406 | */ | |
7407 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
7408 | struct inode *inode, | |
7409 | struct buffer_head *fe_bh, | |
7410 | struct ocfs2_truncate_context *tc) | |
7411 | { | |
7412 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 7413 | u32 clusters_to_del, new_highest_cpos, range; |
bcbbb24a | 7414 | u64 blkno = 0; |
ccd979bd | 7415 | struct ocfs2_extent_list *el; |
1fabe148 | 7416 | handle_t *handle = NULL; |
ccd979bd | 7417 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 7418 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 7419 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
bcbbb24a TM |
7420 | struct ocfs2_alloc_context *meta_ac = NULL; |
7421 | struct ocfs2_refcount_tree *ref_tree = NULL; | |
ccd979bd MF |
7422 | |
7423 | mlog_entry_void(); | |
7424 | ||
dcd0538f | 7425 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
7426 | i_size_read(inode)); |
7427 | ||
13723d00 JB |
7428 | path = ocfs2_new_path(fe_bh, &di->id2.i_list, |
7429 | ocfs2_journal_access_di); | |
dcd0538f MF |
7430 | if (!path) { |
7431 | status = -ENOMEM; | |
7432 | mlog_errno(status); | |
7433 | goto bail; | |
7434 | } | |
83418978 MF |
7435 | |
7436 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
7437 | ||
ccd979bd | 7438 | start: |
3a0782d0 MF |
7439 | /* |
7440 | * Check that we still have allocation to delete. | |
7441 | */ | |
7442 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
7443 | status = 0; | |
7444 | goto bail; | |
7445 | } | |
7446 | ||
bcbbb24a TM |
7447 | credits = 0; |
7448 | ||
dcd0538f MF |
7449 | /* |
7450 | * Truncate always works against the rightmost tree branch. | |
7451 | */ | |
facdb77f | 7452 | status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX); |
dcd0538f MF |
7453 | if (status) { |
7454 | mlog_errno(status); | |
7455 | goto bail; | |
ccd979bd MF |
7456 | } |
7457 | ||
dcd0538f MF |
7458 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
7459 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
7460 | ||
7461 | /* | |
7462 | * By now, el will point to the extent list on the bottom most | |
7463 | * portion of this tree. Only the tail record is considered in | |
7464 | * each pass. | |
7465 | * | |
7466 | * We handle the following cases, in order: | |
7467 | * - empty extent: delete the remaining branch | |
7468 | * - remove the entire record | |
7469 | * - remove a partial record | |
7470 | * - no record needs to be removed (truncate has completed) | |
7471 | */ | |
7472 | el = path_leaf_el(path); | |
3a0782d0 MF |
7473 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7474 | ocfs2_error(inode->i_sb, | |
7475 | "Inode %llu has empty extent block at %llu\n", | |
7476 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7477 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7478 | status = -EROFS; | |
7479 | goto bail; | |
7480 | } | |
7481 | ||
ccd979bd | 7482 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 7483 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 7484 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
7485 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
7486 | clusters_to_del = 0; | |
7487 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 7488 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
bcbbb24a | 7489 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); |
dcd0538f | 7490 | } else if (range > new_highest_cpos) { |
e48edee2 | 7491 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 7492 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f | 7493 | new_highest_cpos; |
bcbbb24a TM |
7494 | blkno = le64_to_cpu(el->l_recs[i].e_blkno) + |
7495 | ocfs2_clusters_to_blocks(inode->i_sb, | |
7496 | ocfs2_rec_clusters(el, &el->l_recs[i]) - | |
7497 | clusters_to_del); | |
dcd0538f MF |
7498 | } else { |
7499 | status = 0; | |
7500 | goto bail; | |
7501 | } | |
ccd979bd | 7502 | |
dcd0538f MF |
7503 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
7504 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7505 | ||
bcbbb24a TM |
7506 | if (el->l_recs[i].e_flags & OCFS2_EXT_REFCOUNTED && clusters_to_del) { |
7507 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & | |
7508 | OCFS2_HAS_REFCOUNT_FL)); | |
7509 | ||
7510 | status = ocfs2_lock_refcount_tree(osb, | |
7511 | le64_to_cpu(di->i_refcount_loc), | |
7512 | 1, &ref_tree, NULL); | |
7513 | if (status) { | |
7514 | mlog_errno(status); | |
7515 | goto bail; | |
7516 | } | |
7517 | ||
7518 | status = ocfs2_prepare_refcount_change_for_del(inode, fe_bh, | |
7519 | blkno, | |
7520 | clusters_to_del, | |
7521 | &credits, | |
7522 | &meta_ac); | |
7523 | if (status < 0) { | |
7524 | mlog_errno(status); | |
7525 | goto bail; | |
7526 | } | |
7527 | } | |
7528 | ||
1b1dcc1b | 7529 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
7530 | tl_sem = 1; |
7531 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
7532 | * record is free for use. If there isn't any, we flush to get | |
7533 | * an empty truncate log. */ | |
7534 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
7535 | status = __ocfs2_flush_truncate_log(osb); | |
7536 | if (status < 0) { | |
7537 | mlog_errno(status); | |
7538 | goto bail; | |
7539 | } | |
7540 | } | |
7541 | ||
bcbbb24a | 7542 | credits += ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, |
dcd0538f MF |
7543 | (struct ocfs2_dinode *)fe_bh->b_data, |
7544 | el); | |
65eff9cc | 7545 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
7546 | if (IS_ERR(handle)) { |
7547 | status = PTR_ERR(handle); | |
7548 | handle = NULL; | |
7549 | mlog_errno(status); | |
7550 | goto bail; | |
7551 | } | |
7552 | ||
dcd0538f | 7553 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
bcbbb24a | 7554 | tc, path, meta_ac); |
ccd979bd MF |
7555 | if (status < 0) { |
7556 | mlog_errno(status); | |
7557 | goto bail; | |
7558 | } | |
7559 | ||
1b1dcc1b | 7560 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7561 | tl_sem = 0; |
7562 | ||
02dc1af4 | 7563 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
7564 | handle = NULL; |
7565 | ||
dcd0538f MF |
7566 | ocfs2_reinit_path(path, 1); |
7567 | ||
bcbbb24a TM |
7568 | if (meta_ac) { |
7569 | ocfs2_free_alloc_context(meta_ac); | |
7570 | meta_ac = NULL; | |
7571 | } | |
7572 | ||
7573 | if (ref_tree) { | |
7574 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
7575 | ref_tree = NULL; | |
7576 | } | |
7577 | ||
dcd0538f | 7578 | /* |
3a0782d0 MF |
7579 | * The check above will catch the case where we've truncated |
7580 | * away all allocation. | |
dcd0538f | 7581 | */ |
3a0782d0 MF |
7582 | goto start; |
7583 | ||
ccd979bd | 7584 | bail: |
ccd979bd MF |
7585 | |
7586 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7587 | ||
7588 | if (tl_sem) | |
1b1dcc1b | 7589 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7590 | |
7591 | if (handle) | |
02dc1af4 | 7592 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 7593 | |
bcbbb24a TM |
7594 | if (meta_ac) |
7595 | ocfs2_free_alloc_context(meta_ac); | |
7596 | ||
7597 | if (ref_tree) | |
7598 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
7599 | ||
59a5e416 MF |
7600 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
7601 | ||
dcd0538f | 7602 | ocfs2_free_path(path); |
ccd979bd MF |
7603 | |
7604 | /* This will drop the ext_alloc cluster lock for us */ | |
7605 | ocfs2_free_truncate_context(tc); | |
7606 | ||
7607 | mlog_exit(status); | |
7608 | return status; | |
7609 | } | |
7610 | ||
ccd979bd | 7611 | /* |
59a5e416 | 7612 | * Expects the inode to already be locked. |
ccd979bd MF |
7613 | */ |
7614 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7615 | struct inode *inode, | |
7616 | struct buffer_head *fe_bh, | |
7617 | struct ocfs2_truncate_context **tc) | |
7618 | { | |
59a5e416 | 7619 | int status; |
ccd979bd MF |
7620 | unsigned int new_i_clusters; |
7621 | struct ocfs2_dinode *fe; | |
7622 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7623 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7624 | |
7625 | mlog_entry_void(); | |
7626 | ||
7627 | *tc = NULL; | |
7628 | ||
7629 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7630 | i_size_read(inode)); | |
7631 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7632 | ||
7633 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7634 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7635 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7636 | |
cd861280 | 7637 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7638 | if (!(*tc)) { |
7639 | status = -ENOMEM; | |
7640 | mlog_errno(status); | |
7641 | goto bail; | |
7642 | } | |
59a5e416 | 7643 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7644 | |
ccd979bd | 7645 | if (fe->id2.i_list.l_tree_depth) { |
3d03a305 | 7646 | status = ocfs2_read_extent_block(INODE_CACHE(inode), |
5e96581a JB |
7647 | le64_to_cpu(fe->i_last_eb_blk), |
7648 | &last_eb_bh); | |
ccd979bd MF |
7649 | if (status < 0) { |
7650 | mlog_errno(status); | |
7651 | goto bail; | |
7652 | } | |
7653 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
ccd979bd MF |
7654 | } |
7655 | ||
7656 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7657 | ||
ccd979bd MF |
7658 | status = 0; |
7659 | bail: | |
7660 | if (status < 0) { | |
7661 | if (*tc) | |
7662 | ocfs2_free_truncate_context(*tc); | |
7663 | *tc = NULL; | |
7664 | } | |
7665 | mlog_exit_void(); | |
7666 | return status; | |
7667 | } | |
7668 | ||
1afc32b9 MF |
7669 | /* |
7670 | * 'start' is inclusive, 'end' is not. | |
7671 | */ | |
7672 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7673 | unsigned int start, unsigned int end, int trunc) | |
7674 | { | |
7675 | int ret; | |
7676 | unsigned int numbytes; | |
7677 | handle_t *handle; | |
7678 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7679 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7680 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7681 | ||
7682 | if (end > i_size_read(inode)) | |
7683 | end = i_size_read(inode); | |
7684 | ||
7685 | BUG_ON(start >= end); | |
7686 | ||
7687 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7688 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7689 | !ocfs2_supports_inline_data(osb)) { | |
7690 | ocfs2_error(inode->i_sb, | |
7691 | "Inline data flags for inode %llu don't agree! " | |
7692 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7693 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7694 | le16_to_cpu(di->i_dyn_features), | |
7695 | OCFS2_I(inode)->ip_dyn_features, | |
7696 | osb->s_feature_incompat); | |
7697 | ret = -EROFS; | |
7698 | goto out; | |
7699 | } | |
7700 | ||
7701 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7702 | if (IS_ERR(handle)) { | |
7703 | ret = PTR_ERR(handle); | |
7704 | mlog_errno(ret); | |
7705 | goto out; | |
7706 | } | |
7707 | ||
0cf2f763 | 7708 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7709 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7710 | if (ret) { |
7711 | mlog_errno(ret); | |
7712 | goto out_commit; | |
7713 | } | |
7714 | ||
7715 | numbytes = end - start; | |
7716 | memset(idata->id_data + start, 0, numbytes); | |
7717 | ||
7718 | /* | |
7719 | * No need to worry about the data page here - it's been | |
7720 | * truncated already and inline data doesn't need it for | |
7721 | * pushing zero's to disk, so we'll let readpage pick it up | |
7722 | * later. | |
7723 | */ | |
7724 | if (trunc) { | |
7725 | i_size_write(inode, start); | |
7726 | di->i_size = cpu_to_le64(start); | |
7727 | } | |
7728 | ||
7729 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7730 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7731 | ||
7732 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7733 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7734 | ||
7735 | ocfs2_journal_dirty(handle, di_bh); | |
7736 | ||
7737 | out_commit: | |
7738 | ocfs2_commit_trans(osb, handle); | |
7739 | ||
7740 | out: | |
7741 | return ret; | |
7742 | } | |
7743 | ||
ccd979bd MF |
7744 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7745 | { | |
59a5e416 MF |
7746 | /* |
7747 | * The caller is responsible for completing deallocation | |
7748 | * before freeing the context. | |
7749 | */ | |
7750 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7751 | mlog(ML_NOTICE, | |
7752 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd | 7753 | |
a81cb88b | 7754 | brelse(tc->tc_last_eb_bh); |
ccd979bd MF |
7755 | |
7756 | kfree(tc); | |
7757 | } |