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> |
ccd979bd MF |
31 | |
32 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
60b11392 | 38 | #include "aops.h" |
ccd979bd MF |
39 | #include "dlmglue.h" |
40 | #include "extent_map.h" | |
41 | #include "inode.h" | |
42 | #include "journal.h" | |
43 | #include "localalloc.h" | |
44 | #include "suballoc.h" | |
45 | #include "sysfile.h" | |
46 | #include "file.h" | |
47 | #include "super.h" | |
48 | #include "uptodate.h" | |
49 | ||
50 | #include "buffer_head_io.h" | |
51 | ||
e7d4cb6b | 52 | |
1625f8ac JB |
53 | /* |
54 | * Operations for a specific extent tree type. | |
55 | * | |
56 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
57 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 58 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
59 | * for the allocation portion of the extent tree. |
60 | */ | |
e7d4cb6b | 61 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
62 | /* |
63 | * last_eb_blk is the block number of the right most leaf extent | |
64 | * block. Most on-disk structures containing an extent tree store | |
65 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
66 | * ->eo_get_last_eb_blk() operations access this value. They are | |
67 | * both required. | |
68 | */ | |
35dc0aa3 JB |
69 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
70 | u64 blkno); | |
71 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
72 | |
73 | /* | |
74 | * The on-disk structure usually keeps track of how many total | |
75 | * clusters are stored in this extent tree. This function updates | |
76 | * that value. new_clusters is the delta, and must be | |
77 | * added to the total. Required. | |
78 | */ | |
35dc0aa3 JB |
79 | void (*eo_update_clusters)(struct inode *inode, |
80 | struct ocfs2_extent_tree *et, | |
81 | u32 new_clusters); | |
1625f8ac JB |
82 | |
83 | /* | |
84 | * If ->eo_insert_check() exists, it is called before rec is | |
85 | * inserted into the extent tree. It is optional. | |
86 | */ | |
1e61ee79 JB |
87 | int (*eo_insert_check)(struct inode *inode, |
88 | struct ocfs2_extent_tree *et, | |
89 | struct ocfs2_extent_rec *rec); | |
35dc0aa3 | 90 | int (*eo_sanity_check)(struct inode *inode, struct ocfs2_extent_tree *et); |
0ce1010f | 91 | |
1625f8ac JB |
92 | /* |
93 | * -------------------------------------------------------------- | |
94 | * The remaining are internal to ocfs2_extent_tree and don't have | |
95 | * accessor functions | |
96 | */ | |
97 | ||
98 | /* | |
99 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
100 | * It is required. | |
101 | */ | |
0ce1010f | 102 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
103 | |
104 | /* | |
105 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
106 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
107 | * to 0 (unlimited). Optional. | |
108 | */ | |
943cced3 JB |
109 | void (*eo_fill_max_leaf_clusters)(struct inode *inode, |
110 | struct ocfs2_extent_tree *et); | |
e7d4cb6b TM |
111 | }; |
112 | ||
e7d4cb6b | 113 | |
f99b9b7c JB |
114 | /* |
115 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
116 | * in the methods. | |
117 | */ | |
118 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
119 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
120 | u64 blkno); | |
121 | static void ocfs2_dinode_update_clusters(struct inode *inode, | |
122 | struct ocfs2_extent_tree *et, | |
123 | u32 clusters); | |
124 | static int ocfs2_dinode_insert_check(struct inode *inode, | |
125 | struct ocfs2_extent_tree *et, | |
126 | struct ocfs2_extent_rec *rec); | |
127 | static int ocfs2_dinode_sanity_check(struct inode *inode, | |
128 | struct ocfs2_extent_tree *et); | |
129 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); | |
130 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
131 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, | |
132 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
133 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
134 | .eo_insert_check = ocfs2_dinode_insert_check, | |
135 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
136 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
137 | }; | |
0ce1010f | 138 | |
e7d4cb6b TM |
139 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
140 | u64 blkno) | |
141 | { | |
ea5efa15 | 142 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 143 | |
f99b9b7c | 144 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
145 | di->i_last_eb_blk = cpu_to_le64(blkno); |
146 | } | |
147 | ||
148 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
149 | { | |
ea5efa15 | 150 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 151 | |
f99b9b7c | 152 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
153 | return le64_to_cpu(di->i_last_eb_blk); |
154 | } | |
155 | ||
156 | static void ocfs2_dinode_update_clusters(struct inode *inode, | |
157 | struct ocfs2_extent_tree *et, | |
158 | u32 clusters) | |
159 | { | |
ea5efa15 | 160 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
161 | |
162 | le32_add_cpu(&di->i_clusters, clusters); | |
163 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
164 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters); | |
165 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
166 | } | |
167 | ||
1e61ee79 JB |
168 | static int ocfs2_dinode_insert_check(struct inode *inode, |
169 | struct ocfs2_extent_tree *et, | |
170 | struct ocfs2_extent_rec *rec) | |
171 | { | |
172 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
173 | ||
174 | BUG_ON(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL); | |
175 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && | |
176 | (OCFS2_I(inode)->ip_clusters != rec->e_cpos), | |
177 | "Device %s, asking for sparse allocation: inode %llu, " | |
178 | "cpos %u, clusters %u\n", | |
179 | osb->dev_str, | |
180 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
181 | rec->e_cpos, | |
182 | OCFS2_I(inode)->ip_clusters); | |
183 | ||
184 | return 0; | |
185 | } | |
186 | ||
e7d4cb6b TM |
187 | static int ocfs2_dinode_sanity_check(struct inode *inode, |
188 | struct ocfs2_extent_tree *et) | |
189 | { | |
10995aa2 | 190 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 191 | |
f99b9b7c | 192 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 193 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 194 | |
10995aa2 | 195 | return 0; |
e7d4cb6b TM |
196 | } |
197 | ||
f99b9b7c JB |
198 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
199 | { | |
200 | struct ocfs2_dinode *di = et->et_object; | |
201 | ||
202 | et->et_root_el = &di->id2.i_list; | |
203 | } | |
204 | ||
e7d4cb6b | 205 | |
0ce1010f JB |
206 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
207 | { | |
208 | struct ocfs2_xattr_value_root *xv = et->et_object; | |
209 | ||
210 | et->et_root_el = &xv->xr_list; | |
211 | } | |
212 | ||
f56654c4 TM |
213 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
214 | u64 blkno) | |
215 | { | |
216 | struct ocfs2_xattr_value_root *xv = | |
ea5efa15 | 217 | (struct ocfs2_xattr_value_root *)et->et_object; |
f56654c4 TM |
218 | |
219 | xv->xr_last_eb_blk = cpu_to_le64(blkno); | |
220 | } | |
221 | ||
222 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
223 | { | |
224 | struct ocfs2_xattr_value_root *xv = | |
ea5efa15 | 225 | (struct ocfs2_xattr_value_root *) et->et_object; |
f56654c4 TM |
226 | |
227 | return le64_to_cpu(xv->xr_last_eb_blk); | |
228 | } | |
229 | ||
230 | static void ocfs2_xattr_value_update_clusters(struct inode *inode, | |
231 | struct ocfs2_extent_tree *et, | |
232 | u32 clusters) | |
233 | { | |
234 | struct ocfs2_xattr_value_root *xv = | |
ea5efa15 | 235 | (struct ocfs2_xattr_value_root *)et->et_object; |
f56654c4 TM |
236 | |
237 | le32_add_cpu(&xv->xr_clusters, clusters); | |
238 | } | |
239 | ||
1a09f556 | 240 | static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
241 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
242 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
243 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 244 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
245 | }; |
246 | ||
0ce1010f JB |
247 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
248 | { | |
249 | struct ocfs2_xattr_block *xb = et->et_object; | |
250 | ||
251 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
252 | } | |
253 | ||
943cced3 JB |
254 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct inode *inode, |
255 | struct ocfs2_extent_tree *et) | |
256 | { | |
257 | et->et_max_leaf_clusters = | |
258 | ocfs2_clusters_for_bytes(inode->i_sb, | |
259 | OCFS2_MAX_XATTR_TREE_LEAF_SIZE); | |
260 | } | |
261 | ||
ba492615 TM |
262 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
263 | u64 blkno) | |
264 | { | |
ea5efa15 | 265 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
266 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
267 | ||
268 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
269 | } | |
270 | ||
271 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
272 | { | |
ea5efa15 | 273 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
274 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
275 | ||
276 | return le64_to_cpu(xt->xt_last_eb_blk); | |
277 | } | |
278 | ||
279 | static void ocfs2_xattr_tree_update_clusters(struct inode *inode, | |
280 | struct ocfs2_extent_tree *et, | |
281 | u32 clusters) | |
282 | { | |
ea5efa15 | 283 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
284 | |
285 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
286 | } | |
287 | ||
ba492615 | 288 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
289 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
290 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
291 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 292 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 293 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
294 | }; |
295 | ||
8d6220d6 JB |
296 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
297 | struct inode *inode, | |
298 | struct buffer_head *bh, | |
299 | void *obj, | |
300 | struct ocfs2_extent_tree_operations *ops) | |
e7d4cb6b | 301 | { |
1a09f556 | 302 | et->et_ops = ops; |
ce1d9ea6 | 303 | et->et_root_bh = bh; |
ea5efa15 JB |
304 | if (!obj) |
305 | obj = (void *)bh->b_data; | |
306 | et->et_object = obj; | |
e7d4cb6b | 307 | |
0ce1010f | 308 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
309 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
310 | et->et_max_leaf_clusters = 0; | |
311 | else | |
312 | et->et_ops->eo_fill_max_leaf_clusters(inode, et); | |
e7d4cb6b TM |
313 | } |
314 | ||
8d6220d6 JB |
315 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
316 | struct inode *inode, | |
317 | struct buffer_head *bh) | |
1a09f556 | 318 | { |
8d6220d6 | 319 | __ocfs2_init_extent_tree(et, inode, bh, NULL, &ocfs2_dinode_et_ops); |
1a09f556 JB |
320 | } |
321 | ||
8d6220d6 | 322 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
f99b9b7c | 323 | struct inode *inode, |
8d6220d6 | 324 | struct buffer_head *bh) |
1a09f556 | 325 | { |
8d6220d6 JB |
326 | __ocfs2_init_extent_tree(et, inode, bh, NULL, |
327 | &ocfs2_xattr_tree_et_ops); | |
1a09f556 JB |
328 | } |
329 | ||
8d6220d6 JB |
330 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
331 | struct inode *inode, | |
332 | struct buffer_head *bh, | |
333 | struct ocfs2_xattr_value_root *xv) | |
e7d4cb6b | 334 | { |
8d6220d6 JB |
335 | __ocfs2_init_extent_tree(et, inode, bh, xv, |
336 | &ocfs2_xattr_value_et_ops); | |
e7d4cb6b TM |
337 | } |
338 | ||
35dc0aa3 JB |
339 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
340 | u64 new_last_eb_blk) | |
e7d4cb6b | 341 | { |
ce1d9ea6 | 342 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
343 | } |
344 | ||
35dc0aa3 | 345 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 346 | { |
ce1d9ea6 | 347 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
348 | } |
349 | ||
35dc0aa3 JB |
350 | static inline void ocfs2_et_update_clusters(struct inode *inode, |
351 | struct ocfs2_extent_tree *et, | |
352 | u32 clusters) | |
353 | { | |
ce1d9ea6 | 354 | et->et_ops->eo_update_clusters(inode, et, clusters); |
35dc0aa3 JB |
355 | } |
356 | ||
1e61ee79 JB |
357 | static inline int ocfs2_et_insert_check(struct inode *inode, |
358 | struct ocfs2_extent_tree *et, | |
359 | struct ocfs2_extent_rec *rec) | |
360 | { | |
361 | int ret = 0; | |
362 | ||
363 | if (et->et_ops->eo_insert_check) | |
364 | ret = et->et_ops->eo_insert_check(inode, et, rec); | |
365 | return ret; | |
366 | } | |
367 | ||
35dc0aa3 JB |
368 | static inline int ocfs2_et_sanity_check(struct inode *inode, |
369 | struct ocfs2_extent_tree *et) | |
e7d4cb6b | 370 | { |
1e61ee79 JB |
371 | int ret = 0; |
372 | ||
373 | if (et->et_ops->eo_sanity_check) | |
374 | ret = et->et_ops->eo_sanity_check(inode, et); | |
375 | return ret; | |
e7d4cb6b TM |
376 | } |
377 | ||
dcd0538f | 378 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
379 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
380 | struct ocfs2_extent_block *eb); | |
ccd979bd | 381 | |
dcd0538f MF |
382 | /* |
383 | * Structures which describe a path through a btree, and functions to | |
384 | * manipulate them. | |
385 | * | |
386 | * The idea here is to be as generic as possible with the tree | |
387 | * manipulation code. | |
388 | */ | |
389 | struct ocfs2_path_item { | |
390 | struct buffer_head *bh; | |
391 | struct ocfs2_extent_list *el; | |
392 | }; | |
ccd979bd | 393 | |
dcd0538f | 394 | #define OCFS2_MAX_PATH_DEPTH 5 |
ccd979bd | 395 | |
dcd0538f MF |
396 | struct ocfs2_path { |
397 | int p_tree_depth; | |
398 | struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH]; | |
399 | }; | |
ccd979bd | 400 | |
dcd0538f MF |
401 | #define path_root_bh(_path) ((_path)->p_node[0].bh) |
402 | #define path_root_el(_path) ((_path)->p_node[0].el) | |
403 | #define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh) | |
404 | #define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el) | |
405 | #define path_num_items(_path) ((_path)->p_tree_depth + 1) | |
ccd979bd | 406 | |
dcd0538f MF |
407 | /* |
408 | * Reset the actual path elements so that we can re-use the structure | |
409 | * to build another path. Generally, this involves freeing the buffer | |
410 | * heads. | |
411 | */ | |
412 | static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) | |
413 | { | |
414 | int i, start = 0, depth = 0; | |
415 | struct ocfs2_path_item *node; | |
ccd979bd | 416 | |
dcd0538f MF |
417 | if (keep_root) |
418 | start = 1; | |
ccd979bd | 419 | |
dcd0538f MF |
420 | for(i = start; i < path_num_items(path); i++) { |
421 | node = &path->p_node[i]; | |
422 | ||
423 | brelse(node->bh); | |
424 | node->bh = NULL; | |
425 | node->el = NULL; | |
426 | } | |
427 | ||
428 | /* | |
429 | * Tree depth may change during truncate, or insert. If we're | |
430 | * keeping the root extent list, then make sure that our path | |
431 | * structure reflects the proper depth. | |
432 | */ | |
433 | if (keep_root) | |
434 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
435 | ||
436 | path->p_tree_depth = depth; | |
437 | } | |
438 | ||
439 | static void ocfs2_free_path(struct ocfs2_path *path) | |
440 | { | |
441 | if (path) { | |
442 | ocfs2_reinit_path(path, 0); | |
443 | kfree(path); | |
444 | } | |
445 | } | |
446 | ||
328d5752 MF |
447 | /* |
448 | * All the elements of src into dest. After this call, src could be freed | |
449 | * without affecting dest. | |
450 | * | |
451 | * Both paths should have the same root. Any non-root elements of dest | |
452 | * will be freed. | |
453 | */ | |
454 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
455 | { | |
456 | int i; | |
457 | ||
458 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
459 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
460 | ||
461 | ocfs2_reinit_path(dest, 1); | |
462 | ||
463 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
464 | dest->p_node[i].bh = src->p_node[i].bh; | |
465 | dest->p_node[i].el = src->p_node[i].el; | |
466 | ||
467 | if (dest->p_node[i].bh) | |
468 | get_bh(dest->p_node[i].bh); | |
469 | } | |
470 | } | |
471 | ||
dcd0538f MF |
472 | /* |
473 | * Make the *dest path the same as src and re-initialize src path to | |
474 | * have a root only. | |
475 | */ | |
476 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
477 | { | |
478 | int i; | |
479 | ||
480 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
481 | ||
482 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
483 | brelse(dest->p_node[i].bh); | |
484 | ||
485 | dest->p_node[i].bh = src->p_node[i].bh; | |
486 | dest->p_node[i].el = src->p_node[i].el; | |
487 | ||
488 | src->p_node[i].bh = NULL; | |
489 | src->p_node[i].el = NULL; | |
490 | } | |
491 | } | |
492 | ||
493 | /* | |
494 | * Insert an extent block at given index. | |
495 | * | |
496 | * This will not take an additional reference on eb_bh. | |
497 | */ | |
498 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
499 | struct buffer_head *eb_bh) | |
500 | { | |
501 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
502 | ||
503 | /* | |
504 | * Right now, no root bh is an extent block, so this helps | |
505 | * catch code errors with dinode trees. The assertion can be | |
506 | * safely removed if we ever need to insert extent block | |
507 | * structures at the root. | |
508 | */ | |
509 | BUG_ON(index == 0); | |
510 | ||
511 | path->p_node[index].bh = eb_bh; | |
512 | path->p_node[index].el = &eb->h_list; | |
513 | } | |
514 | ||
515 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
516 | struct ocfs2_extent_list *root_el) | |
517 | { | |
518 | struct ocfs2_path *path; | |
ccd979bd | 519 | |
dcd0538f MF |
520 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
521 | ||
522 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
523 | if (path) { | |
524 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
525 | get_bh(root_bh); | |
526 | path_root_bh(path) = root_bh; | |
527 | path_root_el(path) = root_el; | |
528 | } | |
529 | ||
530 | return path; | |
531 | } | |
532 | ||
dcd0538f MF |
533 | /* |
534 | * Convenience function to journal all components in a path. | |
535 | */ | |
536 | static int ocfs2_journal_access_path(struct inode *inode, handle_t *handle, | |
537 | struct ocfs2_path *path) | |
538 | { | |
539 | int i, ret = 0; | |
540 | ||
541 | if (!path) | |
542 | goto out; | |
543 | ||
544 | for(i = 0; i < path_num_items(path); i++) { | |
545 | ret = ocfs2_journal_access(handle, inode, path->p_node[i].bh, | |
546 | OCFS2_JOURNAL_ACCESS_WRITE); | |
547 | if (ret < 0) { | |
548 | mlog_errno(ret); | |
549 | goto out; | |
550 | } | |
551 | } | |
552 | ||
553 | out: | |
554 | return ret; | |
555 | } | |
556 | ||
328d5752 MF |
557 | /* |
558 | * Return the index of the extent record which contains cluster #v_cluster. | |
559 | * -1 is returned if it was not found. | |
560 | * | |
561 | * Should work fine on interior and exterior nodes. | |
562 | */ | |
563 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
564 | { | |
565 | int ret = -1; | |
566 | int i; | |
567 | struct ocfs2_extent_rec *rec; | |
568 | u32 rec_end, rec_start, clusters; | |
569 | ||
570 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
571 | rec = &el->l_recs[i]; | |
572 | ||
573 | rec_start = le32_to_cpu(rec->e_cpos); | |
574 | clusters = ocfs2_rec_clusters(el, rec); | |
575 | ||
576 | rec_end = rec_start + clusters; | |
577 | ||
578 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
579 | ret = i; | |
580 | break; | |
581 | } | |
582 | } | |
583 | ||
584 | return ret; | |
585 | } | |
586 | ||
dcd0538f MF |
587 | enum ocfs2_contig_type { |
588 | CONTIG_NONE = 0, | |
589 | CONTIG_LEFT, | |
328d5752 MF |
590 | CONTIG_RIGHT, |
591 | CONTIG_LEFTRIGHT, | |
dcd0538f MF |
592 | }; |
593 | ||
e48edee2 MF |
594 | |
595 | /* | |
596 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
597 | * ocfs2_extent_contig only work properly against leaf nodes! | |
598 | */ | |
dcd0538f MF |
599 | static int ocfs2_block_extent_contig(struct super_block *sb, |
600 | struct ocfs2_extent_rec *ext, | |
601 | u64 blkno) | |
ccd979bd | 602 | { |
e48edee2 MF |
603 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
604 | ||
605 | blk_end += ocfs2_clusters_to_blocks(sb, | |
606 | le16_to_cpu(ext->e_leaf_clusters)); | |
607 | ||
608 | return blkno == blk_end; | |
ccd979bd MF |
609 | } |
610 | ||
dcd0538f MF |
611 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
612 | struct ocfs2_extent_rec *right) | |
613 | { | |
e48edee2 MF |
614 | u32 left_range; |
615 | ||
616 | left_range = le32_to_cpu(left->e_cpos) + | |
617 | le16_to_cpu(left->e_leaf_clusters); | |
618 | ||
619 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
620 | } |
621 | ||
622 | static enum ocfs2_contig_type | |
623 | ocfs2_extent_contig(struct inode *inode, | |
624 | struct ocfs2_extent_rec *ext, | |
625 | struct ocfs2_extent_rec *insert_rec) | |
626 | { | |
627 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
628 | ||
328d5752 MF |
629 | /* |
630 | * Refuse to coalesce extent records with different flag | |
631 | * fields - we don't want to mix unwritten extents with user | |
632 | * data. | |
633 | */ | |
634 | if (ext->e_flags != insert_rec->e_flags) | |
635 | return CONTIG_NONE; | |
636 | ||
dcd0538f MF |
637 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
638 | ocfs2_block_extent_contig(inode->i_sb, ext, blkno)) | |
639 | return CONTIG_RIGHT; | |
640 | ||
641 | blkno = le64_to_cpu(ext->e_blkno); | |
642 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
643 | ocfs2_block_extent_contig(inode->i_sb, insert_rec, blkno)) | |
644 | return CONTIG_LEFT; | |
645 | ||
646 | return CONTIG_NONE; | |
647 | } | |
648 | ||
649 | /* | |
650 | * NOTE: We can have pretty much any combination of contiguousness and | |
651 | * appending. | |
652 | * | |
653 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
654 | * we'll have to update the path to that leaf. | |
655 | */ | |
656 | enum ocfs2_append_type { | |
657 | APPEND_NONE = 0, | |
658 | APPEND_TAIL, | |
659 | }; | |
660 | ||
328d5752 MF |
661 | enum ocfs2_split_type { |
662 | SPLIT_NONE = 0, | |
663 | SPLIT_LEFT, | |
664 | SPLIT_RIGHT, | |
665 | }; | |
666 | ||
dcd0538f | 667 | struct ocfs2_insert_type { |
328d5752 | 668 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
669 | enum ocfs2_append_type ins_appending; |
670 | enum ocfs2_contig_type ins_contig; | |
671 | int ins_contig_index; | |
dcd0538f MF |
672 | int ins_tree_depth; |
673 | }; | |
674 | ||
328d5752 MF |
675 | struct ocfs2_merge_ctxt { |
676 | enum ocfs2_contig_type c_contig_type; | |
677 | int c_has_empty_extent; | |
678 | int c_split_covers_rec; | |
328d5752 MF |
679 | }; |
680 | ||
ccd979bd MF |
681 | /* |
682 | * How many free extents have we got before we need more meta data? | |
683 | */ | |
684 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
685 | struct inode *inode, | |
f99b9b7c | 686 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
687 | { |
688 | int retval; | |
e7d4cb6b | 689 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
690 | struct ocfs2_extent_block *eb; |
691 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 692 | u64 last_eb_blk = 0; |
ccd979bd MF |
693 | |
694 | mlog_entry_void(); | |
695 | ||
f99b9b7c JB |
696 | el = et->et_root_el; |
697 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 698 | |
e7d4cb6b | 699 | if (last_eb_blk) { |
31d33073 | 700 | retval = ocfs2_read_block(inode, last_eb_blk, |
0fcaa56a | 701 | &eb_bh); |
ccd979bd MF |
702 | if (retval < 0) { |
703 | mlog_errno(retval); | |
704 | goto bail; | |
705 | } | |
706 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
707 | el = &eb->h_list; | |
e7d4cb6b | 708 | } |
ccd979bd MF |
709 | |
710 | BUG_ON(el->l_tree_depth != 0); | |
711 | ||
712 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
713 | bail: | |
a81cb88b | 714 | brelse(eb_bh); |
ccd979bd MF |
715 | |
716 | mlog_exit(retval); | |
717 | return retval; | |
718 | } | |
719 | ||
720 | /* expects array to already be allocated | |
721 | * | |
722 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
723 | * l_count for you | |
724 | */ | |
725 | static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb, | |
1fabe148 | 726 | handle_t *handle, |
ccd979bd MF |
727 | struct inode *inode, |
728 | int wanted, | |
729 | struct ocfs2_alloc_context *meta_ac, | |
730 | struct buffer_head *bhs[]) | |
731 | { | |
732 | int count, status, i; | |
733 | u16 suballoc_bit_start; | |
734 | u32 num_got; | |
735 | u64 first_blkno; | |
736 | struct ocfs2_extent_block *eb; | |
737 | ||
738 | mlog_entry_void(); | |
739 | ||
740 | count = 0; | |
741 | while (count < wanted) { | |
742 | status = ocfs2_claim_metadata(osb, | |
743 | handle, | |
744 | meta_ac, | |
745 | wanted - count, | |
746 | &suballoc_bit_start, | |
747 | &num_got, | |
748 | &first_blkno); | |
749 | if (status < 0) { | |
750 | mlog_errno(status); | |
751 | goto bail; | |
752 | } | |
753 | ||
754 | for(i = count; i < (num_got + count); i++) { | |
755 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
756 | if (bhs[i] == NULL) { | |
757 | status = -EIO; | |
758 | mlog_errno(status); | |
759 | goto bail; | |
760 | } | |
761 | ocfs2_set_new_buffer_uptodate(inode, bhs[i]); | |
762 | ||
763 | status = ocfs2_journal_access(handle, inode, bhs[i], | |
764 | OCFS2_JOURNAL_ACCESS_CREATE); | |
765 | if (status < 0) { | |
766 | mlog_errno(status); | |
767 | goto bail; | |
768 | } | |
769 | ||
770 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
771 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
772 | /* Ok, setup the minimal stuff here. */ | |
773 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
774 | eb->h_blkno = cpu_to_le64(first_blkno); | |
775 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
ccd979bd | 776 | eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); |
ccd979bd MF |
777 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
778 | eb->h_list.l_count = | |
779 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
780 | ||
781 | suballoc_bit_start++; | |
782 | first_blkno++; | |
783 | ||
784 | /* We'll also be dirtied by the caller, so | |
785 | * this isn't absolutely necessary. */ | |
786 | status = ocfs2_journal_dirty(handle, bhs[i]); | |
787 | if (status < 0) { | |
788 | mlog_errno(status); | |
789 | goto bail; | |
790 | } | |
791 | } | |
792 | ||
793 | count += num_got; | |
794 | } | |
795 | ||
796 | status = 0; | |
797 | bail: | |
798 | if (status < 0) { | |
799 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 800 | brelse(bhs[i]); |
ccd979bd MF |
801 | bhs[i] = NULL; |
802 | } | |
803 | } | |
804 | mlog_exit(status); | |
805 | return status; | |
806 | } | |
807 | ||
dcd0538f MF |
808 | /* |
809 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
810 | * | |
811 | * Returns the sum of the rightmost extent rec logical offset and | |
812 | * cluster count. | |
813 | * | |
814 | * ocfs2_add_branch() uses this to determine what logical cluster | |
815 | * value should be populated into the leftmost new branch records. | |
816 | * | |
817 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
818 | * value for the new topmost tree record. | |
819 | */ | |
820 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
821 | { | |
822 | int i; | |
823 | ||
824 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
825 | ||
826 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 827 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
828 | } |
829 | ||
ccd979bd MF |
830 | /* |
831 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
832 | * to start at, if we don't want to start the branch at the dinode | |
833 | * structure. | |
834 | * | |
835 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
836 | * for the new last extent block. | |
837 | * | |
838 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 839 | * contain a single record with cluster count == 0. |
ccd979bd MF |
840 | */ |
841 | static int ocfs2_add_branch(struct ocfs2_super *osb, | |
1fabe148 | 842 | handle_t *handle, |
ccd979bd | 843 | struct inode *inode, |
e7d4cb6b | 844 | struct ocfs2_extent_tree *et, |
ccd979bd | 845 | struct buffer_head *eb_bh, |
328d5752 | 846 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
847 | struct ocfs2_alloc_context *meta_ac) |
848 | { | |
849 | int status, new_blocks, i; | |
850 | u64 next_blkno, new_last_eb_blk; | |
851 | struct buffer_head *bh; | |
852 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
853 | struct ocfs2_extent_block *eb; |
854 | struct ocfs2_extent_list *eb_el; | |
855 | struct ocfs2_extent_list *el; | |
dcd0538f | 856 | u32 new_cpos; |
ccd979bd MF |
857 | |
858 | mlog_entry_void(); | |
859 | ||
328d5752 | 860 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 861 | |
ccd979bd MF |
862 | if (eb_bh) { |
863 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
864 | el = &eb->h_list; | |
865 | } else | |
ce1d9ea6 | 866 | el = et->et_root_el; |
ccd979bd MF |
867 | |
868 | /* we never add a branch to a leaf. */ | |
869 | BUG_ON(!el->l_tree_depth); | |
870 | ||
871 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
872 | ||
873 | /* allocate the number of new eb blocks we need */ | |
874 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
875 | GFP_KERNEL); | |
876 | if (!new_eb_bhs) { | |
877 | status = -ENOMEM; | |
878 | mlog_errno(status); | |
879 | goto bail; | |
880 | } | |
881 | ||
882 | status = ocfs2_create_new_meta_bhs(osb, handle, inode, new_blocks, | |
883 | meta_ac, new_eb_bhs); | |
884 | if (status < 0) { | |
885 | mlog_errno(status); | |
886 | goto bail; | |
887 | } | |
888 | ||
328d5752 | 889 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
dcd0538f MF |
890 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); |
891 | ||
ccd979bd MF |
892 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be |
893 | * linked with the rest of the tree. | |
894 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
895 | * | |
896 | * when we leave the loop, new_last_eb_blk will point to the | |
897 | * newest leaf, and next_blkno will point to the topmost extent | |
898 | * block. */ | |
899 | next_blkno = new_last_eb_blk = 0; | |
900 | for(i = 0; i < new_blocks; i++) { | |
901 | bh = new_eb_bhs[i]; | |
902 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
903 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
904 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
905 | status = -EIO; | |
906 | goto bail; | |
907 | } | |
908 | eb_el = &eb->h_list; | |
909 | ||
910 | status = ocfs2_journal_access(handle, inode, bh, | |
911 | OCFS2_JOURNAL_ACCESS_CREATE); | |
912 | if (status < 0) { | |
913 | mlog_errno(status); | |
914 | goto bail; | |
915 | } | |
916 | ||
917 | eb->h_next_leaf_blk = 0; | |
918 | eb_el->l_tree_depth = cpu_to_le16(i); | |
919 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
920 | /* |
921 | * This actually counts as an empty extent as | |
922 | * c_clusters == 0 | |
923 | */ | |
924 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 925 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
926 | /* |
927 | * eb_el isn't always an interior node, but even leaf | |
928 | * nodes want a zero'd flags and reserved field so | |
929 | * this gets the whole 32 bits regardless of use. | |
930 | */ | |
931 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
932 | if (!eb_el->l_tree_depth) |
933 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
934 | ||
935 | status = ocfs2_journal_dirty(handle, bh); | |
936 | if (status < 0) { | |
937 | mlog_errno(status); | |
938 | goto bail; | |
939 | } | |
940 | ||
941 | next_blkno = le64_to_cpu(eb->h_blkno); | |
942 | } | |
943 | ||
944 | /* This is a bit hairy. We want to update up to three blocks | |
945 | * here without leaving any of them in an inconsistent state | |
946 | * in case of error. We don't have to worry about | |
947 | * journal_dirty erroring as it won't unless we've aborted the | |
948 | * handle (in which case we would never be here) so reserving | |
949 | * the write with journal_access is all we need to do. */ | |
328d5752 | 950 | status = ocfs2_journal_access(handle, inode, *last_eb_bh, |
ccd979bd MF |
951 | OCFS2_JOURNAL_ACCESS_WRITE); |
952 | if (status < 0) { | |
953 | mlog_errno(status); | |
954 | goto bail; | |
955 | } | |
ce1d9ea6 | 956 | status = ocfs2_journal_access(handle, inode, et->et_root_bh, |
ccd979bd MF |
957 | OCFS2_JOURNAL_ACCESS_WRITE); |
958 | if (status < 0) { | |
959 | mlog_errno(status); | |
960 | goto bail; | |
961 | } | |
962 | if (eb_bh) { | |
963 | status = ocfs2_journal_access(handle, inode, eb_bh, | |
964 | OCFS2_JOURNAL_ACCESS_WRITE); | |
965 | if (status < 0) { | |
966 | mlog_errno(status); | |
967 | goto bail; | |
968 | } | |
969 | } | |
970 | ||
971 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 972 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
973 | i = le16_to_cpu(el->l_next_free_rec); |
974 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 975 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 976 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
977 | le16_add_cpu(&el->l_next_free_rec, 1); |
978 | ||
979 | /* fe needs a new last extent block pointer, as does the | |
980 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 981 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 982 | |
328d5752 | 983 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
984 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
985 | ||
328d5752 | 986 | status = ocfs2_journal_dirty(handle, *last_eb_bh); |
ccd979bd MF |
987 | if (status < 0) |
988 | mlog_errno(status); | |
ce1d9ea6 | 989 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
990 | if (status < 0) |
991 | mlog_errno(status); | |
992 | if (eb_bh) { | |
993 | status = ocfs2_journal_dirty(handle, eb_bh); | |
994 | if (status < 0) | |
995 | mlog_errno(status); | |
996 | } | |
997 | ||
328d5752 MF |
998 | /* |
999 | * Some callers want to track the rightmost leaf so pass it | |
1000 | * back here. | |
1001 | */ | |
1002 | brelse(*last_eb_bh); | |
1003 | get_bh(new_eb_bhs[0]); | |
1004 | *last_eb_bh = new_eb_bhs[0]; | |
1005 | ||
ccd979bd MF |
1006 | status = 0; |
1007 | bail: | |
1008 | if (new_eb_bhs) { | |
1009 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1010 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1011 | kfree(new_eb_bhs); |
1012 | } | |
1013 | ||
1014 | mlog_exit(status); | |
1015 | return status; | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * adds another level to the allocation tree. | |
1020 | * returns back the new extent block so you can add a branch to it | |
1021 | * after this call. | |
1022 | */ | |
1023 | static int ocfs2_shift_tree_depth(struct ocfs2_super *osb, | |
1fabe148 | 1024 | handle_t *handle, |
ccd979bd | 1025 | struct inode *inode, |
e7d4cb6b | 1026 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1027 | struct ocfs2_alloc_context *meta_ac, |
1028 | struct buffer_head **ret_new_eb_bh) | |
1029 | { | |
1030 | int status, i; | |
dcd0538f | 1031 | u32 new_clusters; |
ccd979bd | 1032 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1033 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1034 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1035 | struct ocfs2_extent_list *eb_el; |
1036 | ||
1037 | mlog_entry_void(); | |
1038 | ||
1039 | status = ocfs2_create_new_meta_bhs(osb, handle, inode, 1, meta_ac, | |
1040 | &new_eb_bh); | |
1041 | if (status < 0) { | |
1042 | mlog_errno(status); | |
1043 | goto bail; | |
1044 | } | |
1045 | ||
1046 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
1047 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1048 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1049 | status = -EIO; | |
1050 | goto bail; | |
1051 | } | |
1052 | ||
1053 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1054 | root_el = et->et_root_el; |
ccd979bd MF |
1055 | |
1056 | status = ocfs2_journal_access(handle, inode, new_eb_bh, | |
1057 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1058 | if (status < 0) { | |
1059 | mlog_errno(status); | |
1060 | goto bail; | |
1061 | } | |
1062 | ||
e7d4cb6b TM |
1063 | /* copy the root extent list data into the new extent block */ |
1064 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1065 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1066 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1067 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd MF |
1068 | |
1069 | status = ocfs2_journal_dirty(handle, new_eb_bh); | |
1070 | if (status < 0) { | |
1071 | mlog_errno(status); | |
1072 | goto bail; | |
1073 | } | |
1074 | ||
ce1d9ea6 | 1075 | status = ocfs2_journal_access(handle, inode, et->et_root_bh, |
ccd979bd MF |
1076 | OCFS2_JOURNAL_ACCESS_WRITE); |
1077 | if (status < 0) { | |
1078 | mlog_errno(status); | |
1079 | goto bail; | |
1080 | } | |
1081 | ||
dcd0538f MF |
1082 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1083 | ||
e7d4cb6b TM |
1084 | /* update root_bh now */ |
1085 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1086 | root_el->l_recs[0].e_cpos = 0; | |
1087 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1088 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1089 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1090 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1091 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1092 | |
1093 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1094 | * becomes the allocated extent block */ | |
e7d4cb6b | 1095 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1096 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1097 | |
ce1d9ea6 | 1098 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1099 | if (status < 0) { |
1100 | mlog_errno(status); | |
1101 | goto bail; | |
1102 | } | |
1103 | ||
1104 | *ret_new_eb_bh = new_eb_bh; | |
1105 | new_eb_bh = NULL; | |
1106 | status = 0; | |
1107 | bail: | |
a81cb88b | 1108 | brelse(new_eb_bh); |
ccd979bd MF |
1109 | |
1110 | mlog_exit(status); | |
1111 | return status; | |
1112 | } | |
1113 | ||
ccd979bd MF |
1114 | /* |
1115 | * Should only be called when there is no space left in any of the | |
1116 | * leaf nodes. What we want to do is find the lowest tree depth | |
1117 | * non-leaf extent block with room for new records. There are three | |
1118 | * valid results of this search: | |
1119 | * | |
1120 | * 1) a lowest extent block is found, then we pass it back in | |
1121 | * *lowest_eb_bh and return '0' | |
1122 | * | |
e7d4cb6b | 1123 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1124 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1125 | * | |
e7d4cb6b | 1126 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1127 | * which case we return > 0 |
1128 | * | |
1129 | * return status < 0 indicates an error. | |
1130 | */ | |
1131 | static int ocfs2_find_branch_target(struct ocfs2_super *osb, | |
1132 | struct inode *inode, | |
e7d4cb6b | 1133 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1134 | struct buffer_head **target_bh) |
1135 | { | |
1136 | int status = 0, i; | |
1137 | u64 blkno; | |
ccd979bd MF |
1138 | struct ocfs2_extent_block *eb; |
1139 | struct ocfs2_extent_list *el; | |
1140 | struct buffer_head *bh = NULL; | |
1141 | struct buffer_head *lowest_bh = NULL; | |
1142 | ||
1143 | mlog_entry_void(); | |
1144 | ||
1145 | *target_bh = NULL; | |
1146 | ||
ce1d9ea6 | 1147 | el = et->et_root_el; |
ccd979bd MF |
1148 | |
1149 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1150 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
b0697053 | 1151 | ocfs2_error(inode->i_sb, "Dinode %llu has empty " |
ccd979bd | 1152 | "extent list (next_free_rec == 0)", |
b0697053 | 1153 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1154 | status = -EIO; |
1155 | goto bail; | |
1156 | } | |
1157 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1158 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1159 | if (!blkno) { | |
b0697053 | 1160 | ocfs2_error(inode->i_sb, "Dinode %llu has extent " |
ccd979bd MF |
1161 | "list where extent # %d has no physical " |
1162 | "block start", | |
b0697053 | 1163 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i); |
ccd979bd MF |
1164 | status = -EIO; |
1165 | goto bail; | |
1166 | } | |
1167 | ||
a81cb88b MF |
1168 | brelse(bh); |
1169 | bh = NULL; | |
ccd979bd | 1170 | |
0fcaa56a | 1171 | status = ocfs2_read_block(inode, blkno, &bh); |
ccd979bd MF |
1172 | if (status < 0) { |
1173 | mlog_errno(status); | |
1174 | goto bail; | |
1175 | } | |
dcd0538f MF |
1176 | |
1177 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1178 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1179 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1180 | status = -EIO; | |
1181 | goto bail; | |
1182 | } | |
1183 | el = &eb->h_list; | |
1184 | ||
1185 | if (le16_to_cpu(el->l_next_free_rec) < | |
1186 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1187 | brelse(lowest_bh); |
dcd0538f MF |
1188 | lowest_bh = bh; |
1189 | get_bh(lowest_bh); | |
1190 | } | |
1191 | } | |
1192 | ||
1193 | /* If we didn't find one and the fe doesn't have any room, | |
1194 | * then return '1' */ | |
ce1d9ea6 | 1195 | el = et->et_root_el; |
e7d4cb6b | 1196 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1197 | status = 1; |
1198 | ||
1199 | *target_bh = lowest_bh; | |
1200 | bail: | |
a81cb88b | 1201 | brelse(bh); |
dcd0538f MF |
1202 | |
1203 | mlog_exit(status); | |
1204 | return status; | |
1205 | } | |
1206 | ||
c3afcbb3 MF |
1207 | /* |
1208 | * Grow a b-tree so that it has more records. | |
1209 | * | |
1210 | * We might shift the tree depth in which case existing paths should | |
1211 | * be considered invalid. | |
1212 | * | |
1213 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1214 | * |
1215 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 MF |
1216 | */ |
1217 | static int ocfs2_grow_tree(struct inode *inode, handle_t *handle, | |
e7d4cb6b | 1218 | struct ocfs2_extent_tree *et, int *final_depth, |
328d5752 | 1219 | struct buffer_head **last_eb_bh, |
c3afcbb3 MF |
1220 | struct ocfs2_alloc_context *meta_ac) |
1221 | { | |
1222 | int ret, shift; | |
ce1d9ea6 | 1223 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1224 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1225 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1226 | struct buffer_head *bh = NULL; | |
1227 | ||
1228 | BUG_ON(meta_ac == NULL); | |
1229 | ||
e7d4cb6b | 1230 | shift = ocfs2_find_branch_target(osb, inode, et, &bh); |
c3afcbb3 MF |
1231 | if (shift < 0) { |
1232 | ret = shift; | |
1233 | mlog_errno(ret); | |
1234 | goto out; | |
1235 | } | |
1236 | ||
1237 | /* We traveled all the way to the bottom of the allocation tree | |
1238 | * and didn't find room for any more extents - we need to add | |
1239 | * another tree level */ | |
1240 | if (shift) { | |
1241 | BUG_ON(bh); | |
1242 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1243 | ||
1244 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1245 | * the new extent block (so we can pass that to | |
1246 | * ocfs2_add_branch). */ | |
e7d4cb6b | 1247 | ret = ocfs2_shift_tree_depth(osb, handle, inode, et, |
c3afcbb3 MF |
1248 | meta_ac, &bh); |
1249 | if (ret < 0) { | |
1250 | mlog_errno(ret); | |
1251 | goto out; | |
1252 | } | |
1253 | depth++; | |
328d5752 MF |
1254 | if (depth == 1) { |
1255 | /* | |
1256 | * Special case: we have room now if we shifted from | |
1257 | * tree_depth 0, so no more work needs to be done. | |
1258 | * | |
1259 | * We won't be calling add_branch, so pass | |
1260 | * back *last_eb_bh as the new leaf. At depth | |
1261 | * zero, it should always be null so there's | |
1262 | * no reason to brelse. | |
1263 | */ | |
1264 | BUG_ON(*last_eb_bh); | |
1265 | get_bh(bh); | |
1266 | *last_eb_bh = bh; | |
c3afcbb3 | 1267 | goto out; |
328d5752 | 1268 | } |
c3afcbb3 MF |
1269 | } |
1270 | ||
1271 | /* call ocfs2_add_branch to add the final part of the tree with | |
1272 | * the new data. */ | |
1273 | mlog(0, "add branch. bh = %p\n", bh); | |
e7d4cb6b | 1274 | ret = ocfs2_add_branch(osb, handle, inode, et, bh, last_eb_bh, |
c3afcbb3 MF |
1275 | meta_ac); |
1276 | if (ret < 0) { | |
1277 | mlog_errno(ret); | |
1278 | goto out; | |
1279 | } | |
1280 | ||
1281 | out: | |
1282 | if (final_depth) | |
1283 | *final_depth = depth; | |
1284 | brelse(bh); | |
1285 | return ret; | |
1286 | } | |
1287 | ||
dcd0538f MF |
1288 | /* |
1289 | * This function will discard the rightmost extent record. | |
1290 | */ | |
1291 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1292 | { | |
1293 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1294 | int count = le16_to_cpu(el->l_count); | |
1295 | unsigned int num_bytes; | |
1296 | ||
1297 | BUG_ON(!next_free); | |
1298 | /* This will cause us to go off the end of our extent list. */ | |
1299 | BUG_ON(next_free >= count); | |
1300 | ||
1301 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1302 | ||
1303 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1304 | } | |
1305 | ||
1306 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1307 | struct ocfs2_extent_rec *insert_rec) | |
1308 | { | |
1309 | int i, insert_index, next_free, has_empty, num_bytes; | |
1310 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1311 | struct ocfs2_extent_rec *rec; | |
1312 | ||
1313 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1314 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1315 | ||
1316 | BUG_ON(!next_free); | |
1317 | ||
1318 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1319 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1320 | |
1321 | /* | |
1322 | * The easiest way to approach this is to just remove the | |
1323 | * empty extent and temporarily decrement next_free. | |
1324 | */ | |
1325 | if (has_empty) { | |
1326 | /* | |
1327 | * If next_free was 1 (only an empty extent), this | |
1328 | * loop won't execute, which is fine. We still want | |
1329 | * the decrement above to happen. | |
1330 | */ | |
1331 | for(i = 0; i < (next_free - 1); i++) | |
1332 | el->l_recs[i] = el->l_recs[i+1]; | |
1333 | ||
1334 | next_free--; | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * Figure out what the new record index should be. | |
1339 | */ | |
1340 | for(i = 0; i < next_free; i++) { | |
1341 | rec = &el->l_recs[i]; | |
1342 | ||
1343 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1344 | break; | |
1345 | } | |
1346 | insert_index = i; | |
1347 | ||
1348 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1349 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1350 | ||
1351 | BUG_ON(insert_index < 0); | |
1352 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1353 | BUG_ON(insert_index > next_free); | |
1354 | ||
1355 | /* | |
1356 | * No need to memmove if we're just adding to the tail. | |
1357 | */ | |
1358 | if (insert_index != next_free) { | |
1359 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1360 | ||
1361 | num_bytes = next_free - insert_index; | |
1362 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1363 | memmove(&el->l_recs[insert_index + 1], | |
1364 | &el->l_recs[insert_index], | |
1365 | num_bytes); | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Either we had an empty extent, and need to re-increment or | |
1370 | * there was no empty extent on a non full rightmost leaf node, | |
1371 | * in which case we still need to increment. | |
1372 | */ | |
1373 | next_free++; | |
1374 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1375 | /* | |
1376 | * Make sure none of the math above just messed up our tree. | |
1377 | */ | |
1378 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1379 | ||
1380 | el->l_recs[insert_index] = *insert_rec; | |
1381 | ||
1382 | } | |
1383 | ||
328d5752 MF |
1384 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1385 | { | |
1386 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1387 | ||
1388 | BUG_ON(num_recs == 0); | |
1389 | ||
1390 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1391 | num_recs--; | |
1392 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1393 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1394 | memset(&el->l_recs[num_recs], 0, | |
1395 | sizeof(struct ocfs2_extent_rec)); | |
1396 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1397 | } | |
1398 | } | |
1399 | ||
dcd0538f MF |
1400 | /* |
1401 | * Create an empty extent record . | |
1402 | * | |
1403 | * l_next_free_rec may be updated. | |
1404 | * | |
1405 | * If an empty extent already exists do nothing. | |
1406 | */ | |
1407 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1408 | { | |
1409 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1410 | ||
e48edee2 MF |
1411 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1412 | ||
dcd0538f MF |
1413 | if (next_free == 0) |
1414 | goto set_and_inc; | |
1415 | ||
1416 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1417 | return; | |
1418 | ||
1419 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1420 | "Asked to create an empty extent in a full list:\n" | |
1421 | "count = %u, tree depth = %u", | |
1422 | le16_to_cpu(el->l_count), | |
1423 | le16_to_cpu(el->l_tree_depth)); | |
1424 | ||
1425 | ocfs2_shift_records_right(el); | |
1426 | ||
1427 | set_and_inc: | |
1428 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1429 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * For a rotation which involves two leaf nodes, the "root node" is | |
1434 | * the lowest level tree node which contains a path to both leafs. This | |
1435 | * resulting set of information can be used to form a complete "subtree" | |
1436 | * | |
1437 | * This function is passed two full paths from the dinode down to a | |
1438 | * pair of adjacent leaves. It's task is to figure out which path | |
1439 | * index contains the subtree root - this can be the root index itself | |
1440 | * in a worst-case rotation. | |
1441 | * | |
1442 | * The array index of the subtree root is passed back. | |
1443 | */ | |
1444 | static int ocfs2_find_subtree_root(struct inode *inode, | |
1445 | struct ocfs2_path *left, | |
1446 | struct ocfs2_path *right) | |
1447 | { | |
1448 | int i = 0; | |
1449 | ||
1450 | /* | |
1451 | * Check that the caller passed in two paths from the same tree. | |
1452 | */ | |
1453 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1454 | ||
1455 | do { | |
1456 | i++; | |
1457 | ||
1458 | /* | |
1459 | * The caller didn't pass two adjacent paths. | |
1460 | */ | |
1461 | mlog_bug_on_msg(i > left->p_tree_depth, | |
1462 | "Inode %lu, left depth %u, right depth %u\n" | |
1463 | "left leaf blk %llu, right leaf blk %llu\n", | |
1464 | inode->i_ino, left->p_tree_depth, | |
1465 | right->p_tree_depth, | |
1466 | (unsigned long long)path_leaf_bh(left)->b_blocknr, | |
1467 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1468 | } while (left->p_node[i].bh->b_blocknr == | |
1469 | right->p_node[i].bh->b_blocknr); | |
1470 | ||
1471 | return i - 1; | |
1472 | } | |
1473 | ||
1474 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1475 | ||
1476 | /* | |
1477 | * Traverse a btree path in search of cpos, starting at root_el. | |
1478 | * | |
1479 | * This code can be called with a cpos larger than the tree, in which | |
1480 | * case it will return the rightmost path. | |
1481 | */ | |
1482 | static int __ocfs2_find_path(struct inode *inode, | |
1483 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1484 | path_insert_t *func, void *data) | |
1485 | { | |
1486 | int i, ret = 0; | |
1487 | u32 range; | |
1488 | u64 blkno; | |
1489 | struct buffer_head *bh = NULL; | |
1490 | struct ocfs2_extent_block *eb; | |
1491 | struct ocfs2_extent_list *el; | |
1492 | struct ocfs2_extent_rec *rec; | |
1493 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1494 | ||
1495 | el = root_el; | |
1496 | while (el->l_tree_depth) { | |
1497 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
1498 | ocfs2_error(inode->i_sb, | |
1499 | "Inode %llu has empty extent list at " | |
1500 | "depth %u\n", | |
1501 | (unsigned long long)oi->ip_blkno, | |
1502 | le16_to_cpu(el->l_tree_depth)); | |
1503 | ret = -EROFS; | |
1504 | goto out; | |
1505 | ||
1506 | } | |
1507 | ||
1508 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1509 | rec = &el->l_recs[i]; | |
1510 | ||
1511 | /* | |
1512 | * In the case that cpos is off the allocation | |
1513 | * tree, this should just wind up returning the | |
1514 | * rightmost record. | |
1515 | */ | |
1516 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1517 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1518 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1519 | break; | |
1520 | } | |
1521 | ||
1522 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1523 | if (blkno == 0) { | |
1524 | ocfs2_error(inode->i_sb, | |
1525 | "Inode %llu has bad blkno in extent list " | |
1526 | "at depth %u (index %d)\n", | |
1527 | (unsigned long long)oi->ip_blkno, | |
1528 | le16_to_cpu(el->l_tree_depth), i); | |
1529 | ret = -EROFS; | |
1530 | goto out; | |
1531 | } | |
1532 | ||
1533 | brelse(bh); | |
1534 | bh = NULL; | |
0fcaa56a | 1535 | ret = ocfs2_read_block(inode, blkno, &bh); |
dcd0538f MF |
1536 | if (ret) { |
1537 | mlog_errno(ret); | |
1538 | goto out; | |
1539 | } | |
1540 | ||
1541 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1542 | el = &eb->h_list; | |
1543 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1544 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1545 | ret = -EIO; | |
1546 | goto out; | |
1547 | } | |
1548 | ||
1549 | if (le16_to_cpu(el->l_next_free_rec) > | |
1550 | le16_to_cpu(el->l_count)) { | |
1551 | ocfs2_error(inode->i_sb, | |
1552 | "Inode %llu has bad count in extent list " | |
1553 | "at block %llu (next free=%u, count=%u)\n", | |
1554 | (unsigned long long)oi->ip_blkno, | |
1555 | (unsigned long long)bh->b_blocknr, | |
1556 | le16_to_cpu(el->l_next_free_rec), | |
1557 | le16_to_cpu(el->l_count)); | |
1558 | ret = -EROFS; | |
1559 | goto out; | |
1560 | } | |
1561 | ||
1562 | if (func) | |
1563 | func(data, bh); | |
1564 | } | |
1565 | ||
1566 | out: | |
1567 | /* | |
1568 | * Catch any trailing bh that the loop didn't handle. | |
1569 | */ | |
1570 | brelse(bh); | |
1571 | ||
1572 | return ret; | |
1573 | } | |
1574 | ||
1575 | /* | |
1576 | * Given an initialized path (that is, it has a valid root extent | |
1577 | * list), this function will traverse the btree in search of the path | |
1578 | * which would contain cpos. | |
1579 | * | |
1580 | * The path traveled is recorded in the path structure. | |
1581 | * | |
1582 | * Note that this will not do any comparisons on leaf node extent | |
1583 | * records, so it will work fine in the case that we just added a tree | |
1584 | * branch. | |
1585 | */ | |
1586 | struct find_path_data { | |
1587 | int index; | |
1588 | struct ocfs2_path *path; | |
1589 | }; | |
1590 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1591 | { | |
1592 | struct find_path_data *fp = data; | |
1593 | ||
1594 | get_bh(bh); | |
1595 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1596 | fp->index++; | |
1597 | } | |
1598 | static int ocfs2_find_path(struct inode *inode, struct ocfs2_path *path, | |
1599 | u32 cpos) | |
1600 | { | |
1601 | struct find_path_data data; | |
1602 | ||
1603 | data.index = 1; | |
1604 | data.path = path; | |
1605 | return __ocfs2_find_path(inode, path_root_el(path), cpos, | |
1606 | find_path_ins, &data); | |
1607 | } | |
1608 | ||
1609 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1610 | { | |
1611 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1612 | struct ocfs2_extent_list *el = &eb->h_list; | |
1613 | struct buffer_head **ret = data; | |
1614 | ||
1615 | /* We want to retain only the leaf block. */ | |
1616 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1617 | get_bh(bh); | |
1618 | *ret = bh; | |
1619 | } | |
1620 | } | |
1621 | /* | |
1622 | * Find the leaf block in the tree which would contain cpos. No | |
1623 | * checking of the actual leaf is done. | |
1624 | * | |
1625 | * Some paths want to call this instead of allocating a path structure | |
1626 | * and calling ocfs2_find_path(). | |
1627 | * | |
1628 | * This function doesn't handle non btree extent lists. | |
1629 | */ | |
363041a5 MF |
1630 | int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el, |
1631 | u32 cpos, struct buffer_head **leaf_bh) | |
dcd0538f MF |
1632 | { |
1633 | int ret; | |
1634 | struct buffer_head *bh = NULL; | |
1635 | ||
1636 | ret = __ocfs2_find_path(inode, root_el, cpos, find_leaf_ins, &bh); | |
1637 | if (ret) { | |
1638 | mlog_errno(ret); | |
1639 | goto out; | |
1640 | } | |
1641 | ||
1642 | *leaf_bh = bh; | |
1643 | out: | |
1644 | return ret; | |
1645 | } | |
1646 | ||
1647 | /* | |
1648 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1649 | * | |
1650 | * Basically, we've moved stuff around at the bottom of the tree and | |
1651 | * we need to fix up the extent records above the changes to reflect | |
1652 | * the new changes. | |
1653 | * | |
1654 | * left_rec: the record on the left. | |
1655 | * left_child_el: is the child list pointed to by left_rec | |
1656 | * right_rec: the record to the right of left_rec | |
1657 | * right_child_el: is the child list pointed to by right_rec | |
1658 | * | |
1659 | * By definition, this only works on interior nodes. | |
1660 | */ | |
1661 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1662 | struct ocfs2_extent_list *left_child_el, | |
1663 | struct ocfs2_extent_rec *right_rec, | |
1664 | struct ocfs2_extent_list *right_child_el) | |
1665 | { | |
1666 | u32 left_clusters, right_end; | |
1667 | ||
1668 | /* | |
1669 | * Interior nodes never have holes. Their cpos is the cpos of | |
1670 | * the leftmost record in their child list. Their cluster | |
1671 | * count covers the full theoretical range of their child list | |
1672 | * - the range between their cpos and the cpos of the record | |
1673 | * immediately to their right. | |
1674 | */ | |
1675 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
328d5752 MF |
1676 | if (ocfs2_is_empty_extent(&right_child_el->l_recs[0])) { |
1677 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); | |
1678 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1679 | } | |
dcd0538f | 1680 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1681 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1682 | |
1683 | /* | |
1684 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1685 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1686 | * updating e_cpos to keep the same highest cluster count. |
1687 | */ | |
1688 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1689 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1690 | |
1691 | right_rec->e_cpos = left_rec->e_cpos; | |
1692 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1693 | ||
1694 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1695 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1696 | } |
1697 | ||
1698 | /* | |
1699 | * Adjust the adjacent root node records involved in a | |
1700 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1701 | * find it's index in the root list. | |
1702 | */ | |
1703 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
1704 | struct ocfs2_extent_list *left_el, | |
1705 | struct ocfs2_extent_list *right_el, | |
1706 | u64 left_el_blkno) | |
1707 | { | |
1708 | int i; | |
1709 | ||
1710 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
1711 | le16_to_cpu(left_el->l_tree_depth)); | |
1712 | ||
1713 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
1714 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
1715 | break; | |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * The path walking code should have never returned a root and | |
1720 | * two paths which are not adjacent. | |
1721 | */ | |
1722 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
1723 | ||
1724 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
1725 | &root_el->l_recs[i + 1], right_el); | |
1726 | } | |
1727 | ||
1728 | /* | |
1729 | * We've changed a leaf block (in right_path) and need to reflect that | |
1730 | * change back up the subtree. | |
1731 | * | |
1732 | * This happens in multiple places: | |
1733 | * - When we've moved an extent record from the left path leaf to the right | |
1734 | * path leaf to make room for an empty extent in the left path leaf. | |
1735 | * - When our insert into the right path leaf is at the leftmost edge | |
1736 | * and requires an update of the path immediately to it's left. This | |
1737 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
1738 | * - When we've adjusted the last extent record in the left path leaf and the |
1739 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f MF |
1740 | */ |
1741 | static void ocfs2_complete_edge_insert(struct inode *inode, handle_t *handle, | |
1742 | struct ocfs2_path *left_path, | |
1743 | struct ocfs2_path *right_path, | |
1744 | int subtree_index) | |
1745 | { | |
1746 | int ret, i, idx; | |
1747 | struct ocfs2_extent_list *el, *left_el, *right_el; | |
1748 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
1749 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
1750 | ||
1751 | /* | |
1752 | * Update the counts and position values within all the | |
1753 | * interior nodes to reflect the leaf rotation we just did. | |
1754 | * | |
1755 | * The root node is handled below the loop. | |
1756 | * | |
1757 | * We begin the loop with right_el and left_el pointing to the | |
1758 | * leaf lists and work our way up. | |
1759 | * | |
1760 | * NOTE: within this loop, left_el and right_el always refer | |
1761 | * to the *child* lists. | |
1762 | */ | |
1763 | left_el = path_leaf_el(left_path); | |
1764 | right_el = path_leaf_el(right_path); | |
1765 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
1766 | mlog(0, "Adjust records at index %u\n", i); | |
1767 | ||
1768 | /* | |
1769 | * One nice property of knowing that all of these | |
1770 | * nodes are below the root is that we only deal with | |
1771 | * the leftmost right node record and the rightmost | |
1772 | * left node record. | |
1773 | */ | |
1774 | el = left_path->p_node[i].el; | |
1775 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1776 | left_rec = &el->l_recs[idx]; | |
1777 | ||
1778 | el = right_path->p_node[i].el; | |
1779 | right_rec = &el->l_recs[0]; | |
1780 | ||
1781 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
1782 | right_el); | |
1783 | ||
1784 | ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh); | |
1785 | if (ret) | |
1786 | mlog_errno(ret); | |
1787 | ||
1788 | ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
1789 | if (ret) | |
1790 | mlog_errno(ret); | |
1791 | ||
1792 | /* | |
1793 | * Setup our list pointers now so that the current | |
1794 | * parents become children in the next iteration. | |
1795 | */ | |
1796 | left_el = left_path->p_node[i].el; | |
1797 | right_el = right_path->p_node[i].el; | |
1798 | } | |
1799 | ||
1800 | /* | |
1801 | * At the root node, adjust the two adjacent records which | |
1802 | * begin our path to the leaves. | |
1803 | */ | |
1804 | ||
1805 | el = left_path->p_node[subtree_index].el; | |
1806 | left_el = left_path->p_node[subtree_index + 1].el; | |
1807 | right_el = right_path->p_node[subtree_index + 1].el; | |
1808 | ||
1809 | ocfs2_adjust_root_records(el, left_el, right_el, | |
1810 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
1811 | ||
1812 | root_bh = left_path->p_node[subtree_index].bh; | |
1813 | ||
1814 | ret = ocfs2_journal_dirty(handle, root_bh); | |
1815 | if (ret) | |
1816 | mlog_errno(ret); | |
1817 | } | |
1818 | ||
1819 | static int ocfs2_rotate_subtree_right(struct inode *inode, | |
1820 | handle_t *handle, | |
1821 | struct ocfs2_path *left_path, | |
1822 | struct ocfs2_path *right_path, | |
1823 | int subtree_index) | |
1824 | { | |
1825 | int ret, i; | |
1826 | struct buffer_head *right_leaf_bh; | |
1827 | struct buffer_head *left_leaf_bh = NULL; | |
1828 | struct buffer_head *root_bh; | |
1829 | struct ocfs2_extent_list *right_el, *left_el; | |
1830 | struct ocfs2_extent_rec move_rec; | |
1831 | ||
1832 | left_leaf_bh = path_leaf_bh(left_path); | |
1833 | left_el = path_leaf_el(left_path); | |
1834 | ||
1835 | if (left_el->l_next_free_rec != left_el->l_count) { | |
1836 | ocfs2_error(inode->i_sb, | |
1837 | "Inode %llu has non-full interior leaf node %llu" | |
1838 | "(next free = %u)", | |
1839 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1840 | (unsigned long long)left_leaf_bh->b_blocknr, | |
1841 | le16_to_cpu(left_el->l_next_free_rec)); | |
1842 | return -EROFS; | |
1843 | } | |
1844 | ||
1845 | /* | |
1846 | * This extent block may already have an empty record, so we | |
1847 | * return early if so. | |
1848 | */ | |
1849 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
1850 | return 0; | |
1851 | ||
1852 | root_bh = left_path->p_node[subtree_index].bh; | |
1853 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
1854 | ||
1855 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
1856 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1857 | if (ret) { | |
1858 | mlog_errno(ret); | |
1859 | goto out; | |
1860 | } | |
1861 | ||
1862 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1863 | ret = ocfs2_journal_access(handle, inode, | |
1864 | right_path->p_node[i].bh, | |
1865 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1866 | if (ret) { | |
1867 | mlog_errno(ret); | |
1868 | goto out; | |
1869 | } | |
1870 | ||
1871 | ret = ocfs2_journal_access(handle, inode, | |
1872 | left_path->p_node[i].bh, | |
1873 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1874 | if (ret) { | |
1875 | mlog_errno(ret); | |
1876 | goto out; | |
1877 | } | |
1878 | } | |
1879 | ||
1880 | right_leaf_bh = path_leaf_bh(right_path); | |
1881 | right_el = path_leaf_el(right_path); | |
1882 | ||
1883 | /* This is a code error, not a disk corruption. */ | |
1884 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
1885 | "because rightmost leaf block %llu is empty\n", | |
1886 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1887 | (unsigned long long)right_leaf_bh->b_blocknr); | |
1888 | ||
1889 | ocfs2_create_empty_extent(right_el); | |
1890 | ||
1891 | ret = ocfs2_journal_dirty(handle, right_leaf_bh); | |
1892 | if (ret) { | |
1893 | mlog_errno(ret); | |
1894 | goto out; | |
1895 | } | |
1896 | ||
1897 | /* Do the copy now. */ | |
1898 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1899 | move_rec = left_el->l_recs[i]; | |
1900 | right_el->l_recs[0] = move_rec; | |
1901 | ||
1902 | /* | |
1903 | * Clear out the record we just copied and shift everything | |
1904 | * over, leaving an empty extent in the left leaf. | |
1905 | * | |
1906 | * We temporarily subtract from next_free_rec so that the | |
1907 | * shift will lose the tail record (which is now defunct). | |
1908 | */ | |
1909 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
1910 | ocfs2_shift_records_right(left_el); | |
1911 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1912 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
1913 | ||
1914 | ret = ocfs2_journal_dirty(handle, left_leaf_bh); | |
1915 | if (ret) { | |
1916 | mlog_errno(ret); | |
1917 | goto out; | |
1918 | } | |
1919 | ||
1920 | ocfs2_complete_edge_insert(inode, handle, left_path, right_path, | |
1921 | subtree_index); | |
1922 | ||
1923 | out: | |
1924 | return ret; | |
1925 | } | |
1926 | ||
1927 | /* | |
1928 | * Given a full path, determine what cpos value would return us a path | |
1929 | * containing the leaf immediately to the left of the current one. | |
1930 | * | |
1931 | * Will return zero if the path passed in is already the leftmost path. | |
1932 | */ | |
1933 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
1934 | struct ocfs2_path *path, u32 *cpos) | |
1935 | { | |
1936 | int i, j, ret = 0; | |
1937 | u64 blkno; | |
1938 | struct ocfs2_extent_list *el; | |
1939 | ||
e48edee2 MF |
1940 | BUG_ON(path->p_tree_depth == 0); |
1941 | ||
dcd0538f MF |
1942 | *cpos = 0; |
1943 | ||
1944 | blkno = path_leaf_bh(path)->b_blocknr; | |
1945 | ||
1946 | /* Start at the tree node just above the leaf and work our way up. */ | |
1947 | i = path->p_tree_depth - 1; | |
1948 | while (i >= 0) { | |
1949 | el = path->p_node[i].el; | |
1950 | ||
1951 | /* | |
1952 | * Find the extent record just before the one in our | |
1953 | * path. | |
1954 | */ | |
1955 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
1956 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
1957 | if (j == 0) { | |
1958 | if (i == 0) { | |
1959 | /* | |
1960 | * We've determined that the | |
1961 | * path specified is already | |
1962 | * the leftmost one - return a | |
1963 | * cpos of zero. | |
1964 | */ | |
1965 | goto out; | |
1966 | } | |
1967 | /* | |
1968 | * The leftmost record points to our | |
1969 | * leaf - we need to travel up the | |
1970 | * tree one level. | |
1971 | */ | |
1972 | goto next_node; | |
1973 | } | |
1974 | ||
1975 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
1976 | *cpos = *cpos + ocfs2_rec_clusters(el, |
1977 | &el->l_recs[j - 1]); | |
1978 | *cpos = *cpos - 1; | |
dcd0538f MF |
1979 | goto out; |
1980 | } | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * If we got here, we never found a valid node where | |
1985 | * the tree indicated one should be. | |
1986 | */ | |
1987 | ocfs2_error(sb, | |
1988 | "Invalid extent tree at extent block %llu\n", | |
1989 | (unsigned long long)blkno); | |
1990 | ret = -EROFS; | |
1991 | goto out; | |
1992 | ||
1993 | next_node: | |
1994 | blkno = path->p_node[i].bh->b_blocknr; | |
1995 | i--; | |
1996 | } | |
1997 | ||
1998 | out: | |
1999 | return ret; | |
2000 | } | |
2001 | ||
328d5752 MF |
2002 | /* |
2003 | * Extend the transaction by enough credits to complete the rotation, | |
2004 | * and still leave at least the original number of credits allocated | |
2005 | * to this transaction. | |
2006 | */ | |
dcd0538f | 2007 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2008 | int op_credits, |
dcd0538f MF |
2009 | struct ocfs2_path *path) |
2010 | { | |
328d5752 | 2011 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f MF |
2012 | |
2013 | if (handle->h_buffer_credits < credits) | |
2014 | return ocfs2_extend_trans(handle, credits); | |
2015 | ||
2016 | return 0; | |
2017 | } | |
2018 | ||
2019 | /* | |
2020 | * Trap the case where we're inserting into the theoretical range past | |
2021 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2022 | * whose cpos is less than ours into the right leaf. | |
2023 | * | |
2024 | * It's only necessary to look at the rightmost record of the left | |
2025 | * leaf because the logic that calls us should ensure that the | |
2026 | * theoretical ranges in the path components above the leaves are | |
2027 | * correct. | |
2028 | */ | |
2029 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2030 | u32 insert_cpos) | |
2031 | { | |
2032 | struct ocfs2_extent_list *left_el; | |
2033 | struct ocfs2_extent_rec *rec; | |
2034 | int next_free; | |
2035 | ||
2036 | left_el = path_leaf_el(left_path); | |
2037 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2038 | rec = &left_el->l_recs[next_free - 1]; | |
2039 | ||
2040 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2041 | return 1; | |
2042 | return 0; | |
2043 | } | |
2044 | ||
328d5752 MF |
2045 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2046 | { | |
2047 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2048 | unsigned int range; | |
2049 | struct ocfs2_extent_rec *rec; | |
2050 | ||
2051 | if (next_free == 0) | |
2052 | return 0; | |
2053 | ||
2054 | rec = &el->l_recs[0]; | |
2055 | if (ocfs2_is_empty_extent(rec)) { | |
2056 | /* Empty list. */ | |
2057 | if (next_free == 1) | |
2058 | return 0; | |
2059 | rec = &el->l_recs[1]; | |
2060 | } | |
2061 | ||
2062 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2063 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2064 | return 1; | |
2065 | return 0; | |
2066 | } | |
2067 | ||
dcd0538f MF |
2068 | /* |
2069 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2070 | * | |
2071 | * The path to the rightmost leaf should be passed in. | |
2072 | * | |
2073 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2074 | * | |
2075 | * Upon succesful return from this function: | |
2076 | * | |
2077 | * - The 'right_path' array will contain a path to the leaf block | |
2078 | * whose range contains e_cpos. | |
2079 | * - That leaf block will have a single empty extent in list index 0. | |
2080 | * - In the case that the rotation requires a post-insert update, | |
2081 | * *ret_left_path will contain a valid path which can be passed to | |
2082 | * ocfs2_insert_path(). | |
2083 | */ | |
2084 | static int ocfs2_rotate_tree_right(struct inode *inode, | |
2085 | handle_t *handle, | |
328d5752 | 2086 | enum ocfs2_split_type split, |
dcd0538f MF |
2087 | u32 insert_cpos, |
2088 | struct ocfs2_path *right_path, | |
2089 | struct ocfs2_path **ret_left_path) | |
2090 | { | |
328d5752 | 2091 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2092 | u32 cpos; |
2093 | struct ocfs2_path *left_path = NULL; | |
2094 | ||
2095 | *ret_left_path = NULL; | |
2096 | ||
2097 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
2098 | path_root_el(right_path)); | |
2099 | if (!left_path) { | |
2100 | ret = -ENOMEM; | |
2101 | mlog_errno(ret); | |
2102 | goto out; | |
2103 | } | |
2104 | ||
2105 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, &cpos); | |
2106 | if (ret) { | |
2107 | mlog_errno(ret); | |
2108 | goto out; | |
2109 | } | |
2110 | ||
2111 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2112 | ||
2113 | /* | |
2114 | * What we want to do here is: | |
2115 | * | |
2116 | * 1) Start with the rightmost path. | |
2117 | * | |
2118 | * 2) Determine a path to the leaf block directly to the left | |
2119 | * of that leaf. | |
2120 | * | |
2121 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2122 | * which contains a path to both leaves. | |
2123 | * | |
2124 | * 4) Rotate the subtree. | |
2125 | * | |
2126 | * 5) Find the next subtree by considering the left path to be | |
2127 | * the new right path. | |
2128 | * | |
2129 | * The check at the top of this while loop also accepts | |
2130 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2131 | * value to get us the left path - insert_cpos might very well | |
2132 | * be filling that hole. | |
2133 | * | |
2134 | * Stop at a cpos of '0' because we either started at the | |
2135 | * leftmost branch (i.e., a tree with one branch and a | |
2136 | * rotation inside of it), or we've gone as far as we can in | |
2137 | * rotating subtrees. | |
2138 | */ | |
2139 | while (cpos && insert_cpos <= cpos) { | |
2140 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2141 | insert_cpos, cpos); | |
2142 | ||
2143 | ret = ocfs2_find_path(inode, left_path, cpos); | |
2144 | if (ret) { | |
2145 | mlog_errno(ret); | |
2146 | goto out; | |
2147 | } | |
2148 | ||
2149 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2150 | path_leaf_bh(right_path), | |
2151 | "Inode %lu: error during insert of %u " | |
2152 | "(left path cpos %u) results in two identical " | |
2153 | "paths ending at %llu\n", | |
2154 | inode->i_ino, insert_cpos, cpos, | |
2155 | (unsigned long long) | |
2156 | path_leaf_bh(left_path)->b_blocknr); | |
2157 | ||
328d5752 MF |
2158 | if (split == SPLIT_NONE && |
2159 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2160 | insert_cpos)) { |
dcd0538f MF |
2161 | |
2162 | /* | |
2163 | * We've rotated the tree as much as we | |
2164 | * should. The rest is up to | |
2165 | * ocfs2_insert_path() to complete, after the | |
2166 | * record insertion. We indicate this | |
2167 | * situation by returning the left path. | |
2168 | * | |
2169 | * The reason we don't adjust the records here | |
2170 | * before the record insert is that an error | |
2171 | * later might break the rule where a parent | |
2172 | * record e_cpos will reflect the actual | |
2173 | * e_cpos of the 1st nonempty record of the | |
2174 | * child list. | |
2175 | */ | |
2176 | *ret_left_path = left_path; | |
2177 | goto out_ret_path; | |
2178 | } | |
2179 | ||
2180 | start = ocfs2_find_subtree_root(inode, left_path, right_path); | |
2181 | ||
2182 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2183 | start, | |
2184 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2185 | right_path->p_tree_depth); | |
2186 | ||
2187 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2188 | orig_credits, right_path); |
dcd0538f MF |
2189 | if (ret) { |
2190 | mlog_errno(ret); | |
2191 | goto out; | |
2192 | } | |
2193 | ||
2194 | ret = ocfs2_rotate_subtree_right(inode, handle, left_path, | |
2195 | right_path, start); | |
2196 | if (ret) { | |
2197 | mlog_errno(ret); | |
2198 | goto out; | |
2199 | } | |
2200 | ||
328d5752 MF |
2201 | if (split != SPLIT_NONE && |
2202 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2203 | insert_cpos)) { | |
2204 | /* | |
2205 | * A rotate moves the rightmost left leaf | |
2206 | * record over to the leftmost right leaf | |
2207 | * slot. If we're doing an extent split | |
2208 | * instead of a real insert, then we have to | |
2209 | * check that the extent to be split wasn't | |
2210 | * just moved over. If it was, then we can | |
2211 | * exit here, passing left_path back - | |
2212 | * ocfs2_split_extent() is smart enough to | |
2213 | * search both leaves. | |
2214 | */ | |
2215 | *ret_left_path = left_path; | |
2216 | goto out_ret_path; | |
2217 | } | |
2218 | ||
dcd0538f MF |
2219 | /* |
2220 | * There is no need to re-read the next right path | |
2221 | * as we know that it'll be our current left | |
2222 | * path. Optimize by copying values instead. | |
2223 | */ | |
2224 | ocfs2_mv_path(right_path, left_path); | |
2225 | ||
2226 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, | |
2227 | &cpos); | |
2228 | if (ret) { | |
2229 | mlog_errno(ret); | |
2230 | goto out; | |
2231 | } | |
2232 | } | |
2233 | ||
2234 | out: | |
2235 | ocfs2_free_path(left_path); | |
2236 | ||
2237 | out_ret_path: | |
2238 | return ret; | |
2239 | } | |
2240 | ||
328d5752 MF |
2241 | static void ocfs2_update_edge_lengths(struct inode *inode, handle_t *handle, |
2242 | struct ocfs2_path *path) | |
dcd0538f | 2243 | { |
328d5752 | 2244 | int i, idx; |
dcd0538f | 2245 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2246 | struct ocfs2_extent_list *el; |
2247 | struct ocfs2_extent_block *eb; | |
2248 | u32 range; | |
dcd0538f | 2249 | |
328d5752 MF |
2250 | /* Path should always be rightmost. */ |
2251 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2252 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2253 | |
328d5752 MF |
2254 | el = &eb->h_list; |
2255 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2256 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2257 | rec = &el->l_recs[idx]; | |
2258 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2259 | |
328d5752 MF |
2260 | for (i = 0; i < path->p_tree_depth; i++) { |
2261 | el = path->p_node[i].el; | |
2262 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2263 | rec = &el->l_recs[idx]; | |
dcd0538f | 2264 | |
328d5752 MF |
2265 | rec->e_int_clusters = cpu_to_le32(range); |
2266 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2267 | |
328d5752 | 2268 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2269 | } |
dcd0538f MF |
2270 | } |
2271 | ||
328d5752 MF |
2272 | static void ocfs2_unlink_path(struct inode *inode, handle_t *handle, |
2273 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
2274 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2275 | { |
328d5752 MF |
2276 | int ret, i; |
2277 | struct ocfs2_extent_block *eb; | |
2278 | struct ocfs2_extent_list *el; | |
2279 | struct buffer_head *bh; | |
2280 | ||
2281 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2282 | bh = path->p_node[i].bh; | |
2283 | ||
2284 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2285 | /* | |
2286 | * Not all nodes might have had their final count | |
2287 | * decremented by the caller - handle this here. | |
2288 | */ | |
2289 | el = &eb->h_list; | |
2290 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2291 | mlog(ML_ERROR, | |
2292 | "Inode %llu, attempted to remove extent block " | |
2293 | "%llu with %u records\n", | |
2294 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2295 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
2296 | le16_to_cpu(el->l_next_free_rec)); | |
2297 | ||
2298 | ocfs2_journal_dirty(handle, bh); | |
2299 | ocfs2_remove_from_cache(inode, bh); | |
2300 | continue; | |
2301 | } | |
2302 | ||
2303 | el->l_next_free_rec = 0; | |
2304 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2305 | ||
2306 | ocfs2_journal_dirty(handle, bh); | |
2307 | ||
2308 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2309 | if (ret) | |
2310 | mlog_errno(ret); | |
2311 | ||
2312 | ocfs2_remove_from_cache(inode, bh); | |
2313 | } | |
dcd0538f MF |
2314 | } |
2315 | ||
328d5752 MF |
2316 | static void ocfs2_unlink_subtree(struct inode *inode, handle_t *handle, |
2317 | struct ocfs2_path *left_path, | |
2318 | struct ocfs2_path *right_path, | |
2319 | int subtree_index, | |
2320 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2321 | { |
328d5752 MF |
2322 | int i; |
2323 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2324 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2325 | struct ocfs2_extent_list *el; |
328d5752 | 2326 | struct ocfs2_extent_block *eb; |
dcd0538f | 2327 | |
328d5752 | 2328 | el = path_leaf_el(left_path); |
dcd0538f | 2329 | |
328d5752 | 2330 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2331 | |
328d5752 MF |
2332 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2333 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2334 | break; | |
dcd0538f | 2335 | |
328d5752 | 2336 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2337 | |
328d5752 MF |
2338 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2339 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2340 | ||
2341 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2342 | eb->h_next_leaf_blk = 0; | |
2343 | ||
2344 | ocfs2_journal_dirty(handle, root_bh); | |
2345 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2346 | ||
2347 | ocfs2_unlink_path(inode, handle, dealloc, right_path, | |
2348 | subtree_index + 1); | |
2349 | } | |
2350 | ||
2351 | static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle, | |
2352 | struct ocfs2_path *left_path, | |
2353 | struct ocfs2_path *right_path, | |
2354 | int subtree_index, | |
2355 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
2356 | int *deleted, |
2357 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2358 | { |
2359 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2360 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2361 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2362 | struct ocfs2_extent_block *eb; | |
2363 | ||
2364 | *deleted = 0; | |
2365 | ||
2366 | right_leaf_el = path_leaf_el(right_path); | |
2367 | left_leaf_el = path_leaf_el(left_path); | |
2368 | root_bh = left_path->p_node[subtree_index].bh; | |
2369 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2370 | ||
2371 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2372 | return 0; | |
dcd0538f | 2373 | |
328d5752 MF |
2374 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2375 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2376 | /* |
328d5752 MF |
2377 | * It's legal for us to proceed if the right leaf is |
2378 | * the rightmost one and it has an empty extent. There | |
2379 | * are two cases to handle - whether the leaf will be | |
2380 | * empty after removal or not. If the leaf isn't empty | |
2381 | * then just remove the empty extent up front. The | |
2382 | * next block will handle empty leaves by flagging | |
2383 | * them for unlink. | |
2384 | * | |
2385 | * Non rightmost leaves will throw -EAGAIN and the | |
2386 | * caller can manually move the subtree and retry. | |
dcd0538f | 2387 | */ |
dcd0538f | 2388 | |
328d5752 MF |
2389 | if (eb->h_next_leaf_blk != 0ULL) |
2390 | return -EAGAIN; | |
2391 | ||
2392 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
2393 | ret = ocfs2_journal_access(handle, inode, | |
2394 | path_leaf_bh(right_path), | |
2395 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2396 | if (ret) { |
2397 | mlog_errno(ret); | |
2398 | goto out; | |
2399 | } | |
2400 | ||
328d5752 MF |
2401 | ocfs2_remove_empty_extent(right_leaf_el); |
2402 | } else | |
2403 | right_has_empty = 1; | |
dcd0538f MF |
2404 | } |
2405 | ||
328d5752 MF |
2406 | if (eb->h_next_leaf_blk == 0ULL && |
2407 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2408 | /* | |
2409 | * We have to update i_last_eb_blk during the meta | |
2410 | * data delete. | |
2411 | */ | |
e7d4cb6b | 2412 | ret = ocfs2_journal_access(handle, inode, et_root_bh, |
328d5752 MF |
2413 | OCFS2_JOURNAL_ACCESS_WRITE); |
2414 | if (ret) { | |
2415 | mlog_errno(ret); | |
2416 | goto out; | |
2417 | } | |
2418 | ||
2419 | del_right_subtree = 1; | |
2420 | } | |
2421 | ||
2422 | /* | |
2423 | * Getting here with an empty extent in the right path implies | |
2424 | * that it's the rightmost path and will be deleted. | |
2425 | */ | |
2426 | BUG_ON(right_has_empty && !del_right_subtree); | |
2427 | ||
2428 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
2429 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2430 | if (ret) { | |
2431 | mlog_errno(ret); | |
2432 | goto out; | |
2433 | } | |
2434 | ||
2435 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
2436 | ret = ocfs2_journal_access(handle, inode, | |
2437 | right_path->p_node[i].bh, | |
2438 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2439 | if (ret) { | |
2440 | mlog_errno(ret); | |
2441 | goto out; | |
2442 | } | |
2443 | ||
2444 | ret = ocfs2_journal_access(handle, inode, | |
2445 | left_path->p_node[i].bh, | |
2446 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2447 | if (ret) { | |
2448 | mlog_errno(ret); | |
2449 | goto out; | |
2450 | } | |
2451 | } | |
2452 | ||
2453 | if (!right_has_empty) { | |
2454 | /* | |
2455 | * Only do this if we're moving a real | |
2456 | * record. Otherwise, the action is delayed until | |
2457 | * after removal of the right path in which case we | |
2458 | * can do a simple shift to remove the empty extent. | |
2459 | */ | |
2460 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2461 | memset(&right_leaf_el->l_recs[0], 0, | |
2462 | sizeof(struct ocfs2_extent_rec)); | |
2463 | } | |
2464 | if (eb->h_next_leaf_blk == 0ULL) { | |
2465 | /* | |
2466 | * Move recs over to get rid of empty extent, decrease | |
2467 | * next_free. This is allowed to remove the last | |
2468 | * extent in our leaf (setting l_next_free_rec to | |
2469 | * zero) - the delete code below won't care. | |
2470 | */ | |
2471 | ocfs2_remove_empty_extent(right_leaf_el); | |
2472 | } | |
2473 | ||
2474 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2475 | if (ret) | |
2476 | mlog_errno(ret); | |
2477 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
2478 | if (ret) | |
2479 | mlog_errno(ret); | |
2480 | ||
2481 | if (del_right_subtree) { | |
2482 | ocfs2_unlink_subtree(inode, handle, left_path, right_path, | |
2483 | subtree_index, dealloc); | |
2484 | ocfs2_update_edge_lengths(inode, handle, left_path); | |
2485 | ||
2486 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2487 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2488 | |
2489 | /* | |
2490 | * Removal of the extent in the left leaf was skipped | |
2491 | * above so we could delete the right path | |
2492 | * 1st. | |
2493 | */ | |
2494 | if (right_has_empty) | |
2495 | ocfs2_remove_empty_extent(left_leaf_el); | |
2496 | ||
e7d4cb6b | 2497 | ret = ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2498 | if (ret) |
2499 | mlog_errno(ret); | |
2500 | ||
2501 | *deleted = 1; | |
2502 | } else | |
2503 | ocfs2_complete_edge_insert(inode, handle, left_path, right_path, | |
2504 | subtree_index); | |
2505 | ||
2506 | out: | |
2507 | return ret; | |
2508 | } | |
2509 | ||
2510 | /* | |
2511 | * Given a full path, determine what cpos value would return us a path | |
2512 | * containing the leaf immediately to the right of the current one. | |
2513 | * | |
2514 | * Will return zero if the path passed in is already the rightmost path. | |
2515 | * | |
2516 | * This looks similar, but is subtly different to | |
2517 | * ocfs2_find_cpos_for_left_leaf(). | |
2518 | */ | |
2519 | static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, | |
2520 | struct ocfs2_path *path, u32 *cpos) | |
2521 | { | |
2522 | int i, j, ret = 0; | |
2523 | u64 blkno; | |
2524 | struct ocfs2_extent_list *el; | |
2525 | ||
2526 | *cpos = 0; | |
2527 | ||
2528 | if (path->p_tree_depth == 0) | |
2529 | return 0; | |
2530 | ||
2531 | blkno = path_leaf_bh(path)->b_blocknr; | |
2532 | ||
2533 | /* Start at the tree node just above the leaf and work our way up. */ | |
2534 | i = path->p_tree_depth - 1; | |
2535 | while (i >= 0) { | |
2536 | int next_free; | |
2537 | ||
2538 | el = path->p_node[i].el; | |
2539 | ||
2540 | /* | |
2541 | * Find the extent record just after the one in our | |
2542 | * path. | |
2543 | */ | |
2544 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2545 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2546 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2547 | if (j == (next_free - 1)) { | |
2548 | if (i == 0) { | |
2549 | /* | |
2550 | * We've determined that the | |
2551 | * path specified is already | |
2552 | * the rightmost one - return a | |
2553 | * cpos of zero. | |
2554 | */ | |
2555 | goto out; | |
2556 | } | |
2557 | /* | |
2558 | * The rightmost record points to our | |
2559 | * leaf - we need to travel up the | |
2560 | * tree one level. | |
2561 | */ | |
2562 | goto next_node; | |
2563 | } | |
2564 | ||
2565 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2566 | goto out; | |
2567 | } | |
2568 | } | |
2569 | ||
2570 | /* | |
2571 | * If we got here, we never found a valid node where | |
2572 | * the tree indicated one should be. | |
2573 | */ | |
2574 | ocfs2_error(sb, | |
2575 | "Invalid extent tree at extent block %llu\n", | |
2576 | (unsigned long long)blkno); | |
2577 | ret = -EROFS; | |
2578 | goto out; | |
2579 | ||
2580 | next_node: | |
2581 | blkno = path->p_node[i].bh->b_blocknr; | |
2582 | i--; | |
2583 | } | |
2584 | ||
2585 | out: | |
2586 | return ret; | |
2587 | } | |
2588 | ||
2589 | static int ocfs2_rotate_rightmost_leaf_left(struct inode *inode, | |
2590 | handle_t *handle, | |
2591 | struct buffer_head *bh, | |
2592 | struct ocfs2_extent_list *el) | |
2593 | { | |
2594 | int ret; | |
2595 | ||
2596 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2597 | return 0; | |
2598 | ||
2599 | ret = ocfs2_journal_access(handle, inode, bh, | |
2600 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2601 | if (ret) { | |
2602 | mlog_errno(ret); | |
2603 | goto out; | |
2604 | } | |
2605 | ||
2606 | ocfs2_remove_empty_extent(el); | |
2607 | ||
2608 | ret = ocfs2_journal_dirty(handle, bh); | |
2609 | if (ret) | |
2610 | mlog_errno(ret); | |
2611 | ||
2612 | out: | |
2613 | return ret; | |
2614 | } | |
2615 | ||
2616 | static int __ocfs2_rotate_tree_left(struct inode *inode, | |
2617 | handle_t *handle, int orig_credits, | |
2618 | struct ocfs2_path *path, | |
2619 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
2620 | struct ocfs2_path **empty_extent_path, |
2621 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2622 | { |
2623 | int ret, subtree_root, deleted; | |
2624 | u32 right_cpos; | |
2625 | struct ocfs2_path *left_path = NULL; | |
2626 | struct ocfs2_path *right_path = NULL; | |
2627 | ||
2628 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2629 | ||
2630 | *empty_extent_path = NULL; | |
2631 | ||
2632 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, path, | |
2633 | &right_cpos); | |
2634 | if (ret) { | |
2635 | mlog_errno(ret); | |
2636 | goto out; | |
2637 | } | |
2638 | ||
2639 | left_path = ocfs2_new_path(path_root_bh(path), | |
2640 | path_root_el(path)); | |
2641 | if (!left_path) { | |
2642 | ret = -ENOMEM; | |
2643 | mlog_errno(ret); | |
2644 | goto out; | |
2645 | } | |
2646 | ||
2647 | ocfs2_cp_path(left_path, path); | |
2648 | ||
2649 | right_path = ocfs2_new_path(path_root_bh(path), | |
2650 | path_root_el(path)); | |
2651 | if (!right_path) { | |
2652 | ret = -ENOMEM; | |
2653 | mlog_errno(ret); | |
2654 | goto out; | |
2655 | } | |
2656 | ||
2657 | while (right_cpos) { | |
2658 | ret = ocfs2_find_path(inode, right_path, right_cpos); | |
2659 | if (ret) { | |
2660 | mlog_errno(ret); | |
2661 | goto out; | |
2662 | } | |
2663 | ||
2664 | subtree_root = ocfs2_find_subtree_root(inode, left_path, | |
2665 | right_path); | |
2666 | ||
2667 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2668 | subtree_root, | |
2669 | (unsigned long long) | |
2670 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2671 | right_path->p_tree_depth); | |
2672 | ||
2673 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
2674 | orig_credits, left_path); | |
2675 | if (ret) { | |
2676 | mlog_errno(ret); | |
2677 | goto out; | |
2678 | } | |
2679 | ||
e8aed345 MF |
2680 | /* |
2681 | * Caller might still want to make changes to the | |
2682 | * tree root, so re-add it to the journal here. | |
2683 | */ | |
2684 | ret = ocfs2_journal_access(handle, inode, | |
2685 | path_root_bh(left_path), | |
2686 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2687 | if (ret) { | |
2688 | mlog_errno(ret); | |
2689 | goto out; | |
2690 | } | |
2691 | ||
328d5752 MF |
2692 | ret = ocfs2_rotate_subtree_left(inode, handle, left_path, |
2693 | right_path, subtree_root, | |
e7d4cb6b | 2694 | dealloc, &deleted, et); |
328d5752 MF |
2695 | if (ret == -EAGAIN) { |
2696 | /* | |
2697 | * The rotation has to temporarily stop due to | |
2698 | * the right subtree having an empty | |
2699 | * extent. Pass it back to the caller for a | |
2700 | * fixup. | |
2701 | */ | |
2702 | *empty_extent_path = right_path; | |
2703 | right_path = NULL; | |
2704 | goto out; | |
2705 | } | |
2706 | if (ret) { | |
2707 | mlog_errno(ret); | |
2708 | goto out; | |
2709 | } | |
2710 | ||
2711 | /* | |
2712 | * The subtree rotate might have removed records on | |
2713 | * the rightmost edge. If so, then rotation is | |
2714 | * complete. | |
2715 | */ | |
2716 | if (deleted) | |
2717 | break; | |
2718 | ||
2719 | ocfs2_mv_path(left_path, right_path); | |
2720 | ||
2721 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path, | |
2722 | &right_cpos); | |
2723 | if (ret) { | |
2724 | mlog_errno(ret); | |
2725 | goto out; | |
2726 | } | |
2727 | } | |
2728 | ||
2729 | out: | |
2730 | ocfs2_free_path(right_path); | |
2731 | ocfs2_free_path(left_path); | |
2732 | ||
2733 | return ret; | |
2734 | } | |
2735 | ||
2736 | static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle, | |
e7d4cb6b TM |
2737 | struct ocfs2_path *path, |
2738 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
2739 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2740 | { |
2741 | int ret, subtree_index; | |
2742 | u32 cpos; | |
2743 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
2744 | struct ocfs2_extent_block *eb; |
2745 | struct ocfs2_extent_list *el; | |
2746 | ||
328d5752 | 2747 | |
35dc0aa3 | 2748 | ret = ocfs2_et_sanity_check(inode, et); |
e7d4cb6b TM |
2749 | if (ret) |
2750 | goto out; | |
328d5752 MF |
2751 | /* |
2752 | * There's two ways we handle this depending on | |
2753 | * whether path is the only existing one. | |
2754 | */ | |
2755 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
2756 | handle->h_buffer_credits, | |
2757 | path); | |
2758 | if (ret) { | |
2759 | mlog_errno(ret); | |
2760 | goto out; | |
2761 | } | |
2762 | ||
2763 | ret = ocfs2_journal_access_path(inode, handle, path); | |
2764 | if (ret) { | |
2765 | mlog_errno(ret); | |
2766 | goto out; | |
2767 | } | |
2768 | ||
2769 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); | |
2770 | if (ret) { | |
2771 | mlog_errno(ret); | |
2772 | goto out; | |
2773 | } | |
2774 | ||
2775 | if (cpos) { | |
2776 | /* | |
2777 | * We have a path to the left of this one - it needs | |
2778 | * an update too. | |
2779 | */ | |
2780 | left_path = ocfs2_new_path(path_root_bh(path), | |
2781 | path_root_el(path)); | |
2782 | if (!left_path) { | |
2783 | ret = -ENOMEM; | |
2784 | mlog_errno(ret); | |
2785 | goto out; | |
2786 | } | |
2787 | ||
2788 | ret = ocfs2_find_path(inode, left_path, cpos); | |
2789 | if (ret) { | |
2790 | mlog_errno(ret); | |
2791 | goto out; | |
2792 | } | |
2793 | ||
2794 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
2795 | if (ret) { | |
2796 | mlog_errno(ret); | |
2797 | goto out; | |
2798 | } | |
2799 | ||
2800 | subtree_index = ocfs2_find_subtree_root(inode, left_path, path); | |
2801 | ||
2802 | ocfs2_unlink_subtree(inode, handle, left_path, path, | |
2803 | subtree_index, dealloc); | |
2804 | ocfs2_update_edge_lengths(inode, handle, left_path); | |
2805 | ||
2806 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2807 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2808 | } else { |
2809 | /* | |
2810 | * 'path' is also the leftmost path which | |
2811 | * means it must be the only one. This gets | |
2812 | * handled differently because we want to | |
2813 | * revert the inode back to having extents | |
2814 | * in-line. | |
2815 | */ | |
2816 | ocfs2_unlink_path(inode, handle, dealloc, path, 1); | |
2817 | ||
ce1d9ea6 | 2818 | el = et->et_root_el; |
328d5752 MF |
2819 | el->l_tree_depth = 0; |
2820 | el->l_next_free_rec = 0; | |
2821 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2822 | ||
35dc0aa3 | 2823 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
2824 | } |
2825 | ||
2826 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
2827 | ||
2828 | out: | |
2829 | ocfs2_free_path(left_path); | |
2830 | return ret; | |
2831 | } | |
2832 | ||
2833 | /* | |
2834 | * Left rotation of btree records. | |
2835 | * | |
2836 | * In many ways, this is (unsurprisingly) the opposite of right | |
2837 | * rotation. We start at some non-rightmost path containing an empty | |
2838 | * extent in the leaf block. The code works its way to the rightmost | |
2839 | * path by rotating records to the left in every subtree. | |
2840 | * | |
2841 | * This is used by any code which reduces the number of extent records | |
2842 | * in a leaf. After removal, an empty record should be placed in the | |
2843 | * leftmost list position. | |
2844 | * | |
2845 | * This won't handle a length update of the rightmost path records if | |
2846 | * the rightmost tree leaf record is removed so the caller is | |
2847 | * responsible for detecting and correcting that. | |
2848 | */ | |
2849 | static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle, | |
2850 | struct ocfs2_path *path, | |
e7d4cb6b TM |
2851 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2852 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2853 | { |
2854 | int ret, orig_credits = handle->h_buffer_credits; | |
2855 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
2856 | struct ocfs2_extent_block *eb; | |
2857 | struct ocfs2_extent_list *el; | |
2858 | ||
2859 | el = path_leaf_el(path); | |
2860 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2861 | return 0; | |
2862 | ||
2863 | if (path->p_tree_depth == 0) { | |
2864 | rightmost_no_delete: | |
2865 | /* | |
e7d4cb6b | 2866 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
2867 | * it up front. |
2868 | */ | |
2869 | ret = ocfs2_rotate_rightmost_leaf_left(inode, handle, | |
2870 | path_leaf_bh(path), | |
2871 | path_leaf_el(path)); | |
2872 | if (ret) | |
2873 | mlog_errno(ret); | |
2874 | goto out; | |
2875 | } | |
2876 | ||
2877 | /* | |
2878 | * Handle rightmost branch now. There's several cases: | |
2879 | * 1) simple rotation leaving records in there. That's trivial. | |
2880 | * 2) rotation requiring a branch delete - there's no more | |
2881 | * records left. Two cases of this: | |
2882 | * a) There are branches to the left. | |
2883 | * b) This is also the leftmost (the only) branch. | |
2884 | * | |
2885 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
2886 | * 2a) we need the left branch so that we can update it with the unlink | |
2887 | * 2b) we need to bring the inode back to inline extents. | |
2888 | */ | |
2889 | ||
2890 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2891 | el = &eb->h_list; | |
2892 | if (eb->h_next_leaf_blk == 0) { | |
2893 | /* | |
2894 | * This gets a bit tricky if we're going to delete the | |
2895 | * rightmost path. Get the other cases out of the way | |
2896 | * 1st. | |
2897 | */ | |
2898 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
2899 | goto rightmost_no_delete; | |
2900 | ||
2901 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
2902 | ret = -EIO; | |
2903 | ocfs2_error(inode->i_sb, | |
2904 | "Inode %llu has empty extent block at %llu", | |
2905 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2906 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
2907 | goto out; | |
2908 | } | |
2909 | ||
2910 | /* | |
2911 | * XXX: The caller can not trust "path" any more after | |
2912 | * this as it will have been deleted. What do we do? | |
2913 | * | |
2914 | * In theory the rotate-for-merge code will never get | |
2915 | * here because it'll always ask for a rotate in a | |
2916 | * nonempty list. | |
2917 | */ | |
2918 | ||
2919 | ret = ocfs2_remove_rightmost_path(inode, handle, path, | |
e7d4cb6b | 2920 | dealloc, et); |
328d5752 MF |
2921 | if (ret) |
2922 | mlog_errno(ret); | |
2923 | goto out; | |
2924 | } | |
2925 | ||
2926 | /* | |
2927 | * Now we can loop, remembering the path we get from -EAGAIN | |
2928 | * and restarting from there. | |
2929 | */ | |
2930 | try_rotate: | |
2931 | ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, path, | |
e7d4cb6b | 2932 | dealloc, &restart_path, et); |
328d5752 MF |
2933 | if (ret && ret != -EAGAIN) { |
2934 | mlog_errno(ret); | |
2935 | goto out; | |
2936 | } | |
2937 | ||
2938 | while (ret == -EAGAIN) { | |
2939 | tmp_path = restart_path; | |
2940 | restart_path = NULL; | |
2941 | ||
2942 | ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, | |
2943 | tmp_path, dealloc, | |
e7d4cb6b | 2944 | &restart_path, et); |
328d5752 MF |
2945 | if (ret && ret != -EAGAIN) { |
2946 | mlog_errno(ret); | |
2947 | goto out; | |
2948 | } | |
2949 | ||
2950 | ocfs2_free_path(tmp_path); | |
2951 | tmp_path = NULL; | |
2952 | ||
2953 | if (ret == 0) | |
2954 | goto try_rotate; | |
2955 | } | |
2956 | ||
2957 | out: | |
2958 | ocfs2_free_path(tmp_path); | |
2959 | ocfs2_free_path(restart_path); | |
2960 | return ret; | |
2961 | } | |
2962 | ||
2963 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
2964 | int index) | |
2965 | { | |
2966 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
2967 | unsigned int size; | |
2968 | ||
2969 | if (rec->e_leaf_clusters == 0) { | |
2970 | /* | |
2971 | * We consumed all of the merged-from record. An empty | |
2972 | * extent cannot exist anywhere but the 1st array | |
2973 | * position, so move things over if the merged-from | |
2974 | * record doesn't occupy that position. | |
2975 | * | |
2976 | * This creates a new empty extent so the caller | |
2977 | * should be smart enough to have removed any existing | |
2978 | * ones. | |
2979 | */ | |
2980 | if (index > 0) { | |
2981 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
2982 | size = index * sizeof(struct ocfs2_extent_rec); | |
2983 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
2984 | } | |
2985 | ||
2986 | /* | |
2987 | * Always memset - the caller doesn't check whether it | |
2988 | * created an empty extent, so there could be junk in | |
2989 | * the other fields. | |
2990 | */ | |
2991 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2992 | } | |
2993 | } | |
2994 | ||
677b9752 TM |
2995 | static int ocfs2_get_right_path(struct inode *inode, |
2996 | struct ocfs2_path *left_path, | |
2997 | struct ocfs2_path **ret_right_path) | |
2998 | { | |
2999 | int ret; | |
3000 | u32 right_cpos; | |
3001 | struct ocfs2_path *right_path = NULL; | |
3002 | struct ocfs2_extent_list *left_el; | |
3003 | ||
3004 | *ret_right_path = NULL; | |
3005 | ||
3006 | /* This function shouldn't be called for non-trees. */ | |
3007 | BUG_ON(left_path->p_tree_depth == 0); | |
3008 | ||
3009 | left_el = path_leaf_el(left_path); | |
3010 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3011 | ||
3012 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path, | |
3013 | &right_cpos); | |
3014 | if (ret) { | |
3015 | mlog_errno(ret); | |
3016 | goto out; | |
3017 | } | |
3018 | ||
3019 | /* This function shouldn't be called for the rightmost leaf. */ | |
3020 | BUG_ON(right_cpos == 0); | |
3021 | ||
3022 | right_path = ocfs2_new_path(path_root_bh(left_path), | |
3023 | path_root_el(left_path)); | |
3024 | if (!right_path) { | |
3025 | ret = -ENOMEM; | |
3026 | mlog_errno(ret); | |
3027 | goto out; | |
3028 | } | |
3029 | ||
3030 | ret = ocfs2_find_path(inode, right_path, right_cpos); | |
3031 | if (ret) { | |
3032 | mlog_errno(ret); | |
3033 | goto out; | |
3034 | } | |
3035 | ||
3036 | *ret_right_path = right_path; | |
3037 | out: | |
3038 | if (ret) | |
3039 | ocfs2_free_path(right_path); | |
3040 | return ret; | |
3041 | } | |
3042 | ||
328d5752 MF |
3043 | /* |
3044 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3045 | * onto the beginning of the record "next" to it. |
3046 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3047 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3048 | * next extent block. | |
328d5752 | 3049 | */ |
677b9752 TM |
3050 | static int ocfs2_merge_rec_right(struct inode *inode, |
3051 | struct ocfs2_path *left_path, | |
3052 | handle_t *handle, | |
3053 | struct ocfs2_extent_rec *split_rec, | |
3054 | int index) | |
328d5752 | 3055 | { |
677b9752 | 3056 | int ret, next_free, i; |
328d5752 MF |
3057 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3058 | struct ocfs2_extent_rec *left_rec; | |
3059 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3060 | struct ocfs2_extent_list *right_el; |
3061 | struct ocfs2_path *right_path = NULL; | |
3062 | int subtree_index = 0; | |
3063 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3064 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3065 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3066 | |
3067 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3068 | left_rec = &el->l_recs[index]; |
677b9752 | 3069 | |
9d8df6aa | 3070 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3071 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3072 | /* we meet with a cross extent block merge. */ | |
3073 | ret = ocfs2_get_right_path(inode, left_path, &right_path); | |
3074 | if (ret) { | |
3075 | mlog_errno(ret); | |
3076 | goto out; | |
3077 | } | |
3078 | ||
3079 | right_el = path_leaf_el(right_path); | |
3080 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3081 | BUG_ON(next_free <= 0); | |
3082 | right_rec = &right_el->l_recs[0]; | |
3083 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3084 | BUG_ON(next_free <= 1); |
677b9752 TM |
3085 | right_rec = &right_el->l_recs[1]; |
3086 | } | |
3087 | ||
3088 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3089 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3090 | le32_to_cpu(right_rec->e_cpos)); | |
3091 | ||
3092 | subtree_index = ocfs2_find_subtree_root(inode, | |
3093 | left_path, right_path); | |
3094 | ||
3095 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3096 | handle->h_buffer_credits, | |
3097 | right_path); | |
3098 | if (ret) { | |
3099 | mlog_errno(ret); | |
3100 | goto out; | |
3101 | } | |
3102 | ||
3103 | root_bh = left_path->p_node[subtree_index].bh; | |
3104 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3105 | ||
3106 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
3107 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3108 | if (ret) { | |
3109 | mlog_errno(ret); | |
3110 | goto out; | |
3111 | } | |
3112 | ||
3113 | for (i = subtree_index + 1; | |
3114 | i < path_num_items(right_path); i++) { | |
3115 | ret = ocfs2_journal_access(handle, inode, | |
3116 | right_path->p_node[i].bh, | |
3117 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3118 | if (ret) { | |
3119 | mlog_errno(ret); | |
3120 | goto out; | |
3121 | } | |
3122 | ||
3123 | ret = ocfs2_journal_access(handle, inode, | |
3124 | left_path->p_node[i].bh, | |
3125 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3126 | if (ret) { | |
3127 | mlog_errno(ret); | |
3128 | goto out; | |
3129 | } | |
3130 | } | |
3131 | ||
3132 | } else { | |
3133 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3134 | right_rec = &el->l_recs[index + 1]; | |
3135 | } | |
328d5752 MF |
3136 | |
3137 | ret = ocfs2_journal_access(handle, inode, bh, | |
3138 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3139 | if (ret) { | |
3140 | mlog_errno(ret); | |
3141 | goto out; | |
3142 | } | |
3143 | ||
3144 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3145 | ||
3146 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3147 | le64_add_cpu(&right_rec->e_blkno, | |
3148 | -ocfs2_clusters_to_blocks(inode->i_sb, split_clusters)); | |
3149 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); | |
3150 | ||
3151 | ocfs2_cleanup_merge(el, index); | |
3152 | ||
3153 | ret = ocfs2_journal_dirty(handle, bh); | |
3154 | if (ret) | |
3155 | mlog_errno(ret); | |
3156 | ||
677b9752 TM |
3157 | if (right_path) { |
3158 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
3159 | if (ret) | |
3160 | mlog_errno(ret); | |
3161 | ||
3162 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3163 | right_path, subtree_index); | |
3164 | } | |
3165 | out: | |
3166 | if (right_path) | |
3167 | ocfs2_free_path(right_path); | |
3168 | return ret; | |
3169 | } | |
3170 | ||
3171 | static int ocfs2_get_left_path(struct inode *inode, | |
3172 | struct ocfs2_path *right_path, | |
3173 | struct ocfs2_path **ret_left_path) | |
3174 | { | |
3175 | int ret; | |
3176 | u32 left_cpos; | |
3177 | struct ocfs2_path *left_path = NULL; | |
3178 | ||
3179 | *ret_left_path = NULL; | |
3180 | ||
3181 | /* This function shouldn't be called for non-trees. */ | |
3182 | BUG_ON(right_path->p_tree_depth == 0); | |
3183 | ||
3184 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
3185 | right_path, &left_cpos); | |
3186 | if (ret) { | |
3187 | mlog_errno(ret); | |
3188 | goto out; | |
3189 | } | |
3190 | ||
3191 | /* This function shouldn't be called for the leftmost leaf. */ | |
3192 | BUG_ON(left_cpos == 0); | |
3193 | ||
3194 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
3195 | path_root_el(right_path)); | |
3196 | if (!left_path) { | |
3197 | ret = -ENOMEM; | |
3198 | mlog_errno(ret); | |
3199 | goto out; | |
3200 | } | |
3201 | ||
3202 | ret = ocfs2_find_path(inode, left_path, left_cpos); | |
3203 | if (ret) { | |
3204 | mlog_errno(ret); | |
3205 | goto out; | |
3206 | } | |
3207 | ||
3208 | *ret_left_path = left_path; | |
328d5752 | 3209 | out: |
677b9752 TM |
3210 | if (ret) |
3211 | ocfs2_free_path(left_path); | |
328d5752 MF |
3212 | return ret; |
3213 | } | |
3214 | ||
3215 | /* | |
3216 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3217 | * onto the tail of the record "before" it. |
3218 | * For index > 0, the "before" means the extent rec at index - 1. | |
3219 | * | |
3220 | * For index == 0, the "before" means the last record of the previous | |
3221 | * extent block. And there is also a situation that we may need to | |
3222 | * remove the rightmost leaf extent block in the right_path and change | |
3223 | * the right path to indicate the new rightmost path. | |
328d5752 | 3224 | */ |
677b9752 TM |
3225 | static int ocfs2_merge_rec_left(struct inode *inode, |
3226 | struct ocfs2_path *right_path, | |
328d5752 MF |
3227 | handle_t *handle, |
3228 | struct ocfs2_extent_rec *split_rec, | |
677b9752 | 3229 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
e7d4cb6b | 3230 | struct ocfs2_extent_tree *et, |
677b9752 | 3231 | int index) |
328d5752 | 3232 | { |
677b9752 | 3233 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3234 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3235 | struct ocfs2_extent_rec *left_rec; | |
3236 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3237 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3238 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3239 | struct buffer_head *root_bh = NULL; | |
3240 | struct ocfs2_path *left_path = NULL; | |
3241 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3242 | |
677b9752 | 3243 | BUG_ON(index < 0); |
328d5752 | 3244 | |
328d5752 | 3245 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3246 | if (index == 0) { |
3247 | /* we meet with a cross extent block merge. */ | |
3248 | ret = ocfs2_get_left_path(inode, right_path, &left_path); | |
3249 | if (ret) { | |
3250 | mlog_errno(ret); | |
3251 | goto out; | |
3252 | } | |
3253 | ||
3254 | left_el = path_leaf_el(left_path); | |
3255 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3256 | le16_to_cpu(left_el->l_count)); | |
3257 | ||
3258 | left_rec = &left_el->l_recs[ | |
3259 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3260 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3261 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3262 | le32_to_cpu(split_rec->e_cpos)); | |
3263 | ||
3264 | subtree_index = ocfs2_find_subtree_root(inode, | |
3265 | left_path, right_path); | |
3266 | ||
3267 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3268 | handle->h_buffer_credits, | |
3269 | left_path); | |
3270 | if (ret) { | |
3271 | mlog_errno(ret); | |
3272 | goto out; | |
3273 | } | |
3274 | ||
3275 | root_bh = left_path->p_node[subtree_index].bh; | |
3276 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3277 | ||
3278 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
3279 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3280 | if (ret) { | |
3281 | mlog_errno(ret); | |
3282 | goto out; | |
3283 | } | |
3284 | ||
3285 | for (i = subtree_index + 1; | |
3286 | i < path_num_items(right_path); i++) { | |
3287 | ret = ocfs2_journal_access(handle, inode, | |
3288 | right_path->p_node[i].bh, | |
3289 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3290 | if (ret) { | |
3291 | mlog_errno(ret); | |
3292 | goto out; | |
3293 | } | |
3294 | ||
3295 | ret = ocfs2_journal_access(handle, inode, | |
3296 | left_path->p_node[i].bh, | |
3297 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3298 | if (ret) { | |
3299 | mlog_errno(ret); | |
3300 | goto out; | |
3301 | } | |
3302 | } | |
3303 | } else { | |
3304 | left_rec = &el->l_recs[index - 1]; | |
3305 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3306 | has_empty_extent = 1; | |
3307 | } | |
328d5752 MF |
3308 | |
3309 | ret = ocfs2_journal_access(handle, inode, bh, | |
3310 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3311 | if (ret) { | |
3312 | mlog_errno(ret); | |
3313 | goto out; | |
3314 | } | |
3315 | ||
3316 | if (has_empty_extent && index == 1) { | |
3317 | /* | |
3318 | * The easy case - we can just plop the record right in. | |
3319 | */ | |
3320 | *left_rec = *split_rec; | |
3321 | ||
3322 | has_empty_extent = 0; | |
677b9752 | 3323 | } else |
328d5752 | 3324 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3325 | |
3326 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3327 | le64_add_cpu(&right_rec->e_blkno, | |
3328 | ocfs2_clusters_to_blocks(inode->i_sb, split_clusters)); | |
3329 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); | |
3330 | ||
3331 | ocfs2_cleanup_merge(el, index); | |
3332 | ||
3333 | ret = ocfs2_journal_dirty(handle, bh); | |
3334 | if (ret) | |
3335 | mlog_errno(ret); | |
3336 | ||
677b9752 TM |
3337 | if (left_path) { |
3338 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
3339 | if (ret) | |
3340 | mlog_errno(ret); | |
3341 | ||
3342 | /* | |
3343 | * In the situation that the right_rec is empty and the extent | |
3344 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3345 | * it and we need to delete the right extent block. | |
3346 | */ | |
3347 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3348 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3349 | ||
3350 | ret = ocfs2_remove_rightmost_path(inode, handle, | |
e7d4cb6b TM |
3351 | right_path, |
3352 | dealloc, et); | |
677b9752 TM |
3353 | if (ret) { |
3354 | mlog_errno(ret); | |
3355 | goto out; | |
3356 | } | |
3357 | ||
3358 | /* Now the rightmost extent block has been deleted. | |
3359 | * So we use the new rightmost path. | |
3360 | */ | |
3361 | ocfs2_mv_path(right_path, left_path); | |
3362 | left_path = NULL; | |
3363 | } else | |
3364 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3365 | right_path, subtree_index); | |
3366 | } | |
328d5752 | 3367 | out: |
677b9752 TM |
3368 | if (left_path) |
3369 | ocfs2_free_path(left_path); | |
328d5752 MF |
3370 | return ret; |
3371 | } | |
3372 | ||
3373 | static int ocfs2_try_to_merge_extent(struct inode *inode, | |
3374 | handle_t *handle, | |
677b9752 | 3375 | struct ocfs2_path *path, |
328d5752 MF |
3376 | int split_index, |
3377 | struct ocfs2_extent_rec *split_rec, | |
3378 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
3379 | struct ocfs2_merge_ctxt *ctxt, |
3380 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
3381 | |
3382 | { | |
518d7269 | 3383 | int ret = 0; |
677b9752 | 3384 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3385 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3386 | ||
3387 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3388 | ||
518d7269 TM |
3389 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3390 | /* | |
3391 | * The merge code will need to create an empty | |
3392 | * extent to take the place of the newly | |
3393 | * emptied slot. Remove any pre-existing empty | |
3394 | * extents - having more than one in a leaf is | |
3395 | * illegal. | |
3396 | */ | |
677b9752 | 3397 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3398 | dealloc, et); |
518d7269 TM |
3399 | if (ret) { |
3400 | mlog_errno(ret); | |
3401 | goto out; | |
328d5752 | 3402 | } |
518d7269 TM |
3403 | split_index--; |
3404 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3405 | } |
3406 | ||
3407 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3408 | /* | |
3409 | * Left-right contig implies this. | |
3410 | */ | |
3411 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3412 | |
3413 | /* | |
3414 | * Since the leftright insert always covers the entire | |
3415 | * extent, this call will delete the insert record | |
3416 | * entirely, resulting in an empty extent record added to | |
3417 | * the extent block. | |
3418 | * | |
3419 | * Since the adding of an empty extent shifts | |
3420 | * everything back to the right, there's no need to | |
3421 | * update split_index here. | |
677b9752 TM |
3422 | * |
3423 | * When the split_index is zero, we need to merge it to the | |
3424 | * prevoius extent block. It is more efficient and easier | |
3425 | * if we do merge_right first and merge_left later. | |
328d5752 | 3426 | */ |
677b9752 TM |
3427 | ret = ocfs2_merge_rec_right(inode, path, |
3428 | handle, split_rec, | |
3429 | split_index); | |
328d5752 MF |
3430 | if (ret) { |
3431 | mlog_errno(ret); | |
3432 | goto out; | |
3433 | } | |
3434 | ||
3435 | /* | |
3436 | * We can only get this from logic error above. | |
3437 | */ | |
3438 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3439 | ||
677b9752 | 3440 | /* The merge left us with an empty extent, remove it. */ |
e7d4cb6b TM |
3441 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
3442 | dealloc, et); | |
328d5752 MF |
3443 | if (ret) { |
3444 | mlog_errno(ret); | |
3445 | goto out; | |
3446 | } | |
677b9752 | 3447 | |
328d5752 MF |
3448 | rec = &el->l_recs[split_index]; |
3449 | ||
3450 | /* | |
3451 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3452 | * we've merged it into the rec already. |
328d5752 | 3453 | */ |
677b9752 TM |
3454 | ret = ocfs2_merge_rec_left(inode, path, |
3455 | handle, rec, | |
e7d4cb6b | 3456 | dealloc, et, |
677b9752 TM |
3457 | split_index); |
3458 | ||
328d5752 MF |
3459 | if (ret) { |
3460 | mlog_errno(ret); | |
3461 | goto out; | |
3462 | } | |
3463 | ||
677b9752 | 3464 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3465 | dealloc, et); |
328d5752 MF |
3466 | /* |
3467 | * Error from this last rotate is not critical, so | |
3468 | * print but don't bubble it up. | |
3469 | */ | |
3470 | if (ret) | |
3471 | mlog_errno(ret); | |
3472 | ret = 0; | |
3473 | } else { | |
3474 | /* | |
3475 | * Merge a record to the left or right. | |
3476 | * | |
3477 | * 'contig_type' is relative to the existing record, | |
3478 | * so for example, if we're "right contig", it's to | |
3479 | * the record on the left (hence the left merge). | |
3480 | */ | |
3481 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
3482 | ret = ocfs2_merge_rec_left(inode, | |
677b9752 TM |
3483 | path, |
3484 | handle, split_rec, | |
e7d4cb6b | 3485 | dealloc, et, |
328d5752 MF |
3486 | split_index); |
3487 | if (ret) { | |
3488 | mlog_errno(ret); | |
3489 | goto out; | |
3490 | } | |
3491 | } else { | |
3492 | ret = ocfs2_merge_rec_right(inode, | |
677b9752 TM |
3493 | path, |
3494 | handle, split_rec, | |
328d5752 MF |
3495 | split_index); |
3496 | if (ret) { | |
3497 | mlog_errno(ret); | |
3498 | goto out; | |
3499 | } | |
3500 | } | |
3501 | ||
3502 | if (ctxt->c_split_covers_rec) { | |
3503 | /* | |
3504 | * The merge may have left an empty extent in | |
3505 | * our leaf. Try to rotate it away. | |
3506 | */ | |
677b9752 | 3507 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3508 | dealloc, et); |
328d5752 MF |
3509 | if (ret) |
3510 | mlog_errno(ret); | |
3511 | ret = 0; | |
3512 | } | |
3513 | } | |
3514 | ||
3515 | out: | |
3516 | return ret; | |
3517 | } | |
3518 | ||
3519 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3520 | enum ocfs2_split_type split, | |
3521 | struct ocfs2_extent_rec *rec, | |
3522 | struct ocfs2_extent_rec *split_rec) | |
3523 | { | |
3524 | u64 len_blocks; | |
3525 | ||
3526 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3527 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3528 | ||
3529 | if (split == SPLIT_LEFT) { | |
3530 | /* | |
3531 | * Region is on the left edge of the existing | |
3532 | * record. | |
3533 | */ | |
3534 | le32_add_cpu(&rec->e_cpos, | |
3535 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3536 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3537 | le16_add_cpu(&rec->e_leaf_clusters, | |
3538 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3539 | } else { | |
3540 | /* | |
3541 | * Region is on the right edge of the existing | |
3542 | * record. | |
3543 | */ | |
3544 | le16_add_cpu(&rec->e_leaf_clusters, | |
3545 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3546 | } | |
3547 | } | |
3548 | ||
3549 | /* | |
3550 | * Do the final bits of extent record insertion at the target leaf | |
3551 | * list. If this leaf is part of an allocation tree, it is assumed | |
3552 | * that the tree above has been prepared. | |
3553 | */ | |
3554 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec, | |
3555 | struct ocfs2_extent_list *el, | |
3556 | struct ocfs2_insert_type *insert, | |
3557 | struct inode *inode) | |
3558 | { | |
3559 | int i = insert->ins_contig_index; | |
3560 | unsigned int range; | |
3561 | struct ocfs2_extent_rec *rec; | |
3562 | ||
3563 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3564 | ||
3565 | if (insert->ins_split != SPLIT_NONE) { | |
3566 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3567 | BUG_ON(i == -1); | |
3568 | rec = &el->l_recs[i]; | |
3569 | ocfs2_subtract_from_rec(inode->i_sb, insert->ins_split, rec, | |
3570 | insert_rec); | |
3571 | goto rotate; | |
3572 | } | |
3573 | ||
3574 | /* | |
3575 | * Contiguous insert - either left or right. | |
3576 | */ | |
3577 | if (insert->ins_contig != CONTIG_NONE) { | |
3578 | rec = &el->l_recs[i]; | |
3579 | if (insert->ins_contig == CONTIG_LEFT) { | |
3580 | rec->e_blkno = insert_rec->e_blkno; | |
3581 | rec->e_cpos = insert_rec->e_cpos; | |
3582 | } | |
3583 | le16_add_cpu(&rec->e_leaf_clusters, | |
3584 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3585 | return; | |
3586 | } | |
3587 | ||
3588 | /* | |
3589 | * Handle insert into an empty leaf. | |
3590 | */ | |
3591 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3592 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3593 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3594 | el->l_recs[0] = *insert_rec; | |
3595 | el->l_next_free_rec = cpu_to_le16(1); | |
3596 | return; | |
3597 | } | |
3598 | ||
3599 | /* | |
3600 | * Appending insert. | |
3601 | */ | |
3602 | if (insert->ins_appending == APPEND_TAIL) { | |
3603 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3604 | rec = &el->l_recs[i]; | |
3605 | range = le32_to_cpu(rec->e_cpos) | |
3606 | + le16_to_cpu(rec->e_leaf_clusters); | |
3607 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3608 | ||
3609 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3610 | le16_to_cpu(el->l_count), | |
3611 | "inode %lu, depth %u, count %u, next free %u, " | |
3612 | "rec.cpos %u, rec.clusters %u, " | |
3613 | "insert.cpos %u, insert.clusters %u\n", | |
3614 | inode->i_ino, | |
3615 | le16_to_cpu(el->l_tree_depth), | |
3616 | le16_to_cpu(el->l_count), | |
3617 | le16_to_cpu(el->l_next_free_rec), | |
3618 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3619 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3620 | le32_to_cpu(insert_rec->e_cpos), | |
3621 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3622 | i++; | |
3623 | el->l_recs[i] = *insert_rec; | |
3624 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3625 | return; | |
3626 | } | |
3627 | ||
3628 | rotate: | |
3629 | /* | |
3630 | * Ok, we have to rotate. | |
3631 | * | |
3632 | * At this point, it is safe to assume that inserting into an | |
3633 | * empty leaf and appending to a leaf have both been handled | |
3634 | * above. | |
3635 | * | |
3636 | * This leaf needs to have space, either by the empty 1st | |
3637 | * extent record, or by virtue of an l_next_rec < l_count. | |
3638 | */ | |
3639 | ocfs2_rotate_leaf(el, insert_rec); | |
3640 | } | |
3641 | ||
328d5752 MF |
3642 | static void ocfs2_adjust_rightmost_records(struct inode *inode, |
3643 | handle_t *handle, | |
3644 | struct ocfs2_path *path, | |
3645 | struct ocfs2_extent_rec *insert_rec) | |
3646 | { | |
3647 | int ret, i, next_free; | |
3648 | struct buffer_head *bh; | |
3649 | struct ocfs2_extent_list *el; | |
3650 | struct ocfs2_extent_rec *rec; | |
3651 | ||
3652 | /* | |
3653 | * Update everything except the leaf block. | |
3654 | */ | |
3655 | for (i = 0; i < path->p_tree_depth; i++) { | |
3656 | bh = path->p_node[i].bh; | |
3657 | el = path->p_node[i].el; | |
3658 | ||
dcd0538f MF |
3659 | next_free = le16_to_cpu(el->l_next_free_rec); |
3660 | if (next_free == 0) { | |
3661 | ocfs2_error(inode->i_sb, | |
3662 | "Dinode %llu has a bad extent list", | |
3663 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
3664 | ret = -EIO; | |
328d5752 MF |
3665 | return; |
3666 | } | |
3667 | ||
3668 | rec = &el->l_recs[next_free - 1]; | |
3669 | ||
3670 | rec->e_int_clusters = insert_rec->e_cpos; | |
3671 | le32_add_cpu(&rec->e_int_clusters, | |
3672 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3673 | le32_add_cpu(&rec->e_int_clusters, | |
3674 | -le32_to_cpu(rec->e_cpos)); | |
3675 | ||
3676 | ret = ocfs2_journal_dirty(handle, bh); | |
3677 | if (ret) | |
3678 | mlog_errno(ret); | |
3679 | ||
3680 | } | |
3681 | } | |
3682 | ||
3683 | static int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle, | |
3684 | struct ocfs2_extent_rec *insert_rec, | |
3685 | struct ocfs2_path *right_path, | |
3686 | struct ocfs2_path **ret_left_path) | |
3687 | { | |
3688 | int ret, next_free; | |
3689 | struct ocfs2_extent_list *el; | |
3690 | struct ocfs2_path *left_path = NULL; | |
3691 | ||
3692 | *ret_left_path = NULL; | |
3693 | ||
3694 | /* | |
3695 | * This shouldn't happen for non-trees. The extent rec cluster | |
3696 | * count manipulation below only works for interior nodes. | |
3697 | */ | |
3698 | BUG_ON(right_path->p_tree_depth == 0); | |
3699 | ||
3700 | /* | |
3701 | * If our appending insert is at the leftmost edge of a leaf, | |
3702 | * then we might need to update the rightmost records of the | |
3703 | * neighboring path. | |
3704 | */ | |
3705 | el = path_leaf_el(right_path); | |
3706 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3707 | if (next_free == 0 || | |
3708 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3709 | u32 left_cpos; | |
3710 | ||
3711 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, | |
3712 | &left_cpos); | |
3713 | if (ret) { | |
3714 | mlog_errno(ret); | |
dcd0538f MF |
3715 | goto out; |
3716 | } | |
3717 | ||
328d5752 MF |
3718 | mlog(0, "Append may need a left path update. cpos: %u, " |
3719 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3720 | left_cpos); | |
e48edee2 | 3721 | |
328d5752 MF |
3722 | /* |
3723 | * No need to worry if the append is already in the | |
3724 | * leftmost leaf. | |
3725 | */ | |
3726 | if (left_cpos) { | |
3727 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
3728 | path_root_el(right_path)); | |
3729 | if (!left_path) { | |
3730 | ret = -ENOMEM; | |
3731 | mlog_errno(ret); | |
3732 | goto out; | |
3733 | } | |
dcd0538f | 3734 | |
328d5752 MF |
3735 | ret = ocfs2_find_path(inode, left_path, left_cpos); |
3736 | if (ret) { | |
3737 | mlog_errno(ret); | |
3738 | goto out; | |
3739 | } | |
dcd0538f | 3740 | |
328d5752 MF |
3741 | /* |
3742 | * ocfs2_insert_path() will pass the left_path to the | |
3743 | * journal for us. | |
3744 | */ | |
3745 | } | |
3746 | } | |
dcd0538f | 3747 | |
328d5752 MF |
3748 | ret = ocfs2_journal_access_path(inode, handle, right_path); |
3749 | if (ret) { | |
3750 | mlog_errno(ret); | |
3751 | goto out; | |
dcd0538f MF |
3752 | } |
3753 | ||
328d5752 MF |
3754 | ocfs2_adjust_rightmost_records(inode, handle, right_path, insert_rec); |
3755 | ||
dcd0538f MF |
3756 | *ret_left_path = left_path; |
3757 | ret = 0; | |
3758 | out: | |
3759 | if (ret != 0) | |
3760 | ocfs2_free_path(left_path); | |
3761 | ||
3762 | return ret; | |
3763 | } | |
3764 | ||
328d5752 MF |
3765 | static void ocfs2_split_record(struct inode *inode, |
3766 | struct ocfs2_path *left_path, | |
3767 | struct ocfs2_path *right_path, | |
3768 | struct ocfs2_extent_rec *split_rec, | |
3769 | enum ocfs2_split_type split) | |
3770 | { | |
3771 | int index; | |
3772 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
3773 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
3774 | struct ocfs2_extent_rec *rec, *tmprec; | |
3775 | ||
3776 | right_el = path_leaf_el(right_path);; | |
3777 | if (left_path) | |
3778 | left_el = path_leaf_el(left_path); | |
3779 | ||
3780 | el = right_el; | |
3781 | insert_el = right_el; | |
3782 | index = ocfs2_search_extent_list(el, cpos); | |
3783 | if (index != -1) { | |
3784 | if (index == 0 && left_path) { | |
3785 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3786 | ||
3787 | /* | |
3788 | * This typically means that the record | |
3789 | * started in the left path but moved to the | |
3790 | * right as a result of rotation. We either | |
3791 | * move the existing record to the left, or we | |
3792 | * do the later insert there. | |
3793 | * | |
3794 | * In this case, the left path should always | |
3795 | * exist as the rotate code will have passed | |
3796 | * it back for a post-insert update. | |
3797 | */ | |
3798 | ||
3799 | if (split == SPLIT_LEFT) { | |
3800 | /* | |
3801 | * It's a left split. Since we know | |
3802 | * that the rotate code gave us an | |
3803 | * empty extent in the left path, we | |
3804 | * can just do the insert there. | |
3805 | */ | |
3806 | insert_el = left_el; | |
3807 | } else { | |
3808 | /* | |
3809 | * Right split - we have to move the | |
3810 | * existing record over to the left | |
3811 | * leaf. The insert will be into the | |
3812 | * newly created empty extent in the | |
3813 | * right leaf. | |
3814 | */ | |
3815 | tmprec = &right_el->l_recs[index]; | |
3816 | ocfs2_rotate_leaf(left_el, tmprec); | |
3817 | el = left_el; | |
3818 | ||
3819 | memset(tmprec, 0, sizeof(*tmprec)); | |
3820 | index = ocfs2_search_extent_list(left_el, cpos); | |
3821 | BUG_ON(index == -1); | |
3822 | } | |
3823 | } | |
3824 | } else { | |
3825 | BUG_ON(!left_path); | |
3826 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
3827 | /* | |
3828 | * Left path is easy - we can just allow the insert to | |
3829 | * happen. | |
3830 | */ | |
3831 | el = left_el; | |
3832 | insert_el = left_el; | |
3833 | index = ocfs2_search_extent_list(el, cpos); | |
3834 | BUG_ON(index == -1); | |
3835 | } | |
3836 | ||
3837 | rec = &el->l_recs[index]; | |
3838 | ocfs2_subtract_from_rec(inode->i_sb, split, rec, split_rec); | |
3839 | ocfs2_rotate_leaf(insert_el, split_rec); | |
3840 | } | |
3841 | ||
dcd0538f | 3842 | /* |
e7d4cb6b TM |
3843 | * This function only does inserts on an allocation b-tree. For tree |
3844 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
3845 | * |
3846 | * right_path is the path we want to do the actual insert | |
3847 | * in. left_path should only be passed in if we need to update that | |
3848 | * portion of the tree after an edge insert. | |
3849 | */ | |
3850 | static int ocfs2_insert_path(struct inode *inode, | |
3851 | handle_t *handle, | |
3852 | struct ocfs2_path *left_path, | |
3853 | struct ocfs2_path *right_path, | |
3854 | struct ocfs2_extent_rec *insert_rec, | |
3855 | struct ocfs2_insert_type *insert) | |
3856 | { | |
3857 | int ret, subtree_index; | |
3858 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 3859 | |
dcd0538f MF |
3860 | if (left_path) { |
3861 | int credits = handle->h_buffer_credits; | |
3862 | ||
3863 | /* | |
3864 | * There's a chance that left_path got passed back to | |
3865 | * us without being accounted for in the | |
3866 | * journal. Extend our transaction here to be sure we | |
3867 | * can change those blocks. | |
3868 | */ | |
3869 | credits += left_path->p_tree_depth; | |
3870 | ||
3871 | ret = ocfs2_extend_trans(handle, credits); | |
3872 | if (ret < 0) { | |
3873 | mlog_errno(ret); | |
3874 | goto out; | |
3875 | } | |
3876 | ||
3877 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
3878 | if (ret < 0) { | |
3879 | mlog_errno(ret); | |
3880 | goto out; | |
3881 | } | |
3882 | } | |
3883 | ||
e8aed345 MF |
3884 | /* |
3885 | * Pass both paths to the journal. The majority of inserts | |
3886 | * will be touching all components anyway. | |
3887 | */ | |
3888 | ret = ocfs2_journal_access_path(inode, handle, right_path); | |
3889 | if (ret < 0) { | |
3890 | mlog_errno(ret); | |
3891 | goto out; | |
3892 | } | |
3893 | ||
328d5752 MF |
3894 | if (insert->ins_split != SPLIT_NONE) { |
3895 | /* | |
3896 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 3897 | * of splits, but it's easier to just let one separate |
328d5752 MF |
3898 | * function sort it all out. |
3899 | */ | |
3900 | ocfs2_split_record(inode, left_path, right_path, | |
3901 | insert_rec, insert->ins_split); | |
e8aed345 MF |
3902 | |
3903 | /* | |
3904 | * Split might have modified either leaf and we don't | |
3905 | * have a guarantee that the later edge insert will | |
3906 | * dirty this for us. | |
3907 | */ | |
3908 | if (left_path) | |
3909 | ret = ocfs2_journal_dirty(handle, | |
3910 | path_leaf_bh(left_path)); | |
3911 | if (ret) | |
3912 | mlog_errno(ret); | |
328d5752 MF |
3913 | } else |
3914 | ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path), | |
3915 | insert, inode); | |
dcd0538f | 3916 | |
dcd0538f MF |
3917 | ret = ocfs2_journal_dirty(handle, leaf_bh); |
3918 | if (ret) | |
3919 | mlog_errno(ret); | |
3920 | ||
3921 | if (left_path) { | |
3922 | /* | |
3923 | * The rotate code has indicated that we need to fix | |
3924 | * up portions of the tree after the insert. | |
3925 | * | |
3926 | * XXX: Should we extend the transaction here? | |
3927 | */ | |
3928 | subtree_index = ocfs2_find_subtree_root(inode, left_path, | |
3929 | right_path); | |
3930 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3931 | right_path, subtree_index); | |
3932 | } | |
3933 | ||
3934 | ret = 0; | |
3935 | out: | |
3936 | return ret; | |
3937 | } | |
3938 | ||
3939 | static int ocfs2_do_insert_extent(struct inode *inode, | |
3940 | handle_t *handle, | |
e7d4cb6b | 3941 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
3942 | struct ocfs2_extent_rec *insert_rec, |
3943 | struct ocfs2_insert_type *type) | |
3944 | { | |
3945 | int ret, rotate = 0; | |
3946 | u32 cpos; | |
3947 | struct ocfs2_path *right_path = NULL; | |
3948 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
3949 | struct ocfs2_extent_list *el; |
3950 | ||
ce1d9ea6 | 3951 | el = et->et_root_el; |
dcd0538f | 3952 | |
ce1d9ea6 | 3953 | ret = ocfs2_journal_access(handle, inode, et->et_root_bh, |
dcd0538f MF |
3954 | OCFS2_JOURNAL_ACCESS_WRITE); |
3955 | if (ret) { | |
3956 | mlog_errno(ret); | |
3957 | goto out; | |
3958 | } | |
3959 | ||
3960 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
3961 | ocfs2_insert_at_leaf(insert_rec, el, type, inode); | |
3962 | goto out_update_clusters; | |
3963 | } | |
3964 | ||
ce1d9ea6 | 3965 | right_path = ocfs2_new_path(et->et_root_bh, et->et_root_el); |
dcd0538f MF |
3966 | if (!right_path) { |
3967 | ret = -ENOMEM; | |
3968 | mlog_errno(ret); | |
3969 | goto out; | |
3970 | } | |
3971 | ||
3972 | /* | |
3973 | * Determine the path to start with. Rotations need the | |
3974 | * rightmost path, everything else can go directly to the | |
3975 | * target leaf. | |
3976 | */ | |
3977 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
3978 | if (type->ins_appending == APPEND_NONE && | |
3979 | type->ins_contig == CONTIG_NONE) { | |
3980 | rotate = 1; | |
3981 | cpos = UINT_MAX; | |
3982 | } | |
3983 | ||
3984 | ret = ocfs2_find_path(inode, right_path, cpos); | |
3985 | if (ret) { | |
3986 | mlog_errno(ret); | |
3987 | goto out; | |
3988 | } | |
3989 | ||
3990 | /* | |
3991 | * Rotations and appends need special treatment - they modify | |
3992 | * parts of the tree's above them. | |
3993 | * | |
3994 | * Both might pass back a path immediate to the left of the | |
3995 | * one being inserted to. This will be cause | |
3996 | * ocfs2_insert_path() to modify the rightmost records of | |
3997 | * left_path to account for an edge insert. | |
3998 | * | |
3999 | * XXX: When modifying this code, keep in mind that an insert | |
4000 | * can wind up skipping both of these two special cases... | |
4001 | */ | |
4002 | if (rotate) { | |
328d5752 | 4003 | ret = ocfs2_rotate_tree_right(inode, handle, type->ins_split, |
dcd0538f MF |
4004 | le32_to_cpu(insert_rec->e_cpos), |
4005 | right_path, &left_path); | |
4006 | if (ret) { | |
4007 | mlog_errno(ret); | |
4008 | goto out; | |
4009 | } | |
e8aed345 MF |
4010 | |
4011 | /* | |
4012 | * ocfs2_rotate_tree_right() might have extended the | |
4013 | * transaction without re-journaling our tree root. | |
4014 | */ | |
ce1d9ea6 | 4015 | ret = ocfs2_journal_access(handle, inode, et->et_root_bh, |
e8aed345 MF |
4016 | OCFS2_JOURNAL_ACCESS_WRITE); |
4017 | if (ret) { | |
4018 | mlog_errno(ret); | |
4019 | goto out; | |
4020 | } | |
dcd0538f MF |
4021 | } else if (type->ins_appending == APPEND_TAIL |
4022 | && type->ins_contig != CONTIG_LEFT) { | |
4023 | ret = ocfs2_append_rec_to_path(inode, handle, insert_rec, | |
4024 | right_path, &left_path); | |
4025 | if (ret) { | |
4026 | mlog_errno(ret); | |
4027 | goto out; | |
4028 | } | |
4029 | } | |
4030 | ||
4031 | ret = ocfs2_insert_path(inode, handle, left_path, right_path, | |
4032 | insert_rec, type); | |
4033 | if (ret) { | |
4034 | mlog_errno(ret); | |
4035 | goto out; | |
4036 | } | |
4037 | ||
4038 | out_update_clusters: | |
328d5752 | 4039 | if (type->ins_split == SPLIT_NONE) |
35dc0aa3 JB |
4040 | ocfs2_et_update_clusters(inode, et, |
4041 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
dcd0538f | 4042 | |
ce1d9ea6 | 4043 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4044 | if (ret) |
4045 | mlog_errno(ret); | |
4046 | ||
4047 | out: | |
4048 | ocfs2_free_path(left_path); | |
4049 | ocfs2_free_path(right_path); | |
4050 | ||
4051 | return ret; | |
4052 | } | |
4053 | ||
328d5752 | 4054 | static enum ocfs2_contig_type |
ad5a4d70 | 4055 | ocfs2_figure_merge_contig_type(struct inode *inode, struct ocfs2_path *path, |
328d5752 MF |
4056 | struct ocfs2_extent_list *el, int index, |
4057 | struct ocfs2_extent_rec *split_rec) | |
4058 | { | |
ad5a4d70 | 4059 | int status; |
328d5752 | 4060 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4061 | u32 left_cpos, right_cpos; |
4062 | struct ocfs2_extent_rec *rec = NULL; | |
4063 | struct ocfs2_extent_list *new_el; | |
4064 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4065 | struct buffer_head *bh; | |
4066 | struct ocfs2_extent_block *eb; | |
4067 | ||
4068 | if (index > 0) { | |
4069 | rec = &el->l_recs[index - 1]; | |
4070 | } else if (path->p_tree_depth > 0) { | |
4071 | status = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
4072 | path, &left_cpos); | |
4073 | if (status) | |
4074 | goto out; | |
4075 | ||
4076 | if (left_cpos != 0) { | |
4077 | left_path = ocfs2_new_path(path_root_bh(path), | |
4078 | path_root_el(path)); | |
4079 | if (!left_path) | |
4080 | goto out; | |
4081 | ||
4082 | status = ocfs2_find_path(inode, left_path, left_cpos); | |
4083 | if (status) | |
4084 | goto out; | |
4085 | ||
4086 | new_el = path_leaf_el(left_path); | |
4087 | ||
4088 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4089 | le16_to_cpu(new_el->l_count)) { | |
4090 | bh = path_leaf_bh(left_path); | |
4091 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
4092 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, | |
4093 | eb); | |
4094 | goto out; | |
4095 | } | |
4096 | rec = &new_el->l_recs[ | |
4097 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4098 | } | |
4099 | } | |
328d5752 MF |
4100 | |
4101 | /* | |
4102 | * We're careful to check for an empty extent record here - | |
4103 | * the merge code will know what to do if it sees one. | |
4104 | */ | |
ad5a4d70 | 4105 | if (rec) { |
328d5752 MF |
4106 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4107 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4108 | ret = CONTIG_RIGHT; | |
4109 | } else { | |
4110 | ret = ocfs2_extent_contig(inode, rec, split_rec); | |
4111 | } | |
4112 | } | |
4113 | ||
ad5a4d70 TM |
4114 | rec = NULL; |
4115 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4116 | rec = &el->l_recs[index + 1]; | |
4117 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4118 | path->p_tree_depth > 0) { | |
4119 | status = ocfs2_find_cpos_for_right_leaf(inode->i_sb, | |
4120 | path, &right_cpos); | |
4121 | if (status) | |
4122 | goto out; | |
4123 | ||
4124 | if (right_cpos == 0) | |
4125 | goto out; | |
4126 | ||
4127 | right_path = ocfs2_new_path(path_root_bh(path), | |
4128 | path_root_el(path)); | |
4129 | if (!right_path) | |
4130 | goto out; | |
4131 | ||
4132 | status = ocfs2_find_path(inode, right_path, right_cpos); | |
4133 | if (status) | |
4134 | goto out; | |
4135 | ||
4136 | new_el = path_leaf_el(right_path); | |
4137 | rec = &new_el->l_recs[0]; | |
4138 | if (ocfs2_is_empty_extent(rec)) { | |
4139 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4140 | bh = path_leaf_bh(right_path); | |
4141 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
4142 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, | |
4143 | eb); | |
4144 | goto out; | |
4145 | } | |
4146 | rec = &new_el->l_recs[1]; | |
4147 | } | |
4148 | } | |
4149 | ||
4150 | if (rec) { | |
328d5752 MF |
4151 | enum ocfs2_contig_type contig_type; |
4152 | ||
328d5752 MF |
4153 | contig_type = ocfs2_extent_contig(inode, rec, split_rec); |
4154 | ||
4155 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4156 | ret = CONTIG_LEFTRIGHT; | |
4157 | else if (ret == CONTIG_NONE) | |
4158 | ret = contig_type; | |
4159 | } | |
4160 | ||
ad5a4d70 TM |
4161 | out: |
4162 | if (left_path) | |
4163 | ocfs2_free_path(left_path); | |
4164 | if (right_path) | |
4165 | ocfs2_free_path(right_path); | |
4166 | ||
328d5752 MF |
4167 | return ret; |
4168 | } | |
4169 | ||
dcd0538f MF |
4170 | static void ocfs2_figure_contig_type(struct inode *inode, |
4171 | struct ocfs2_insert_type *insert, | |
4172 | struct ocfs2_extent_list *el, | |
ca12b7c4 TM |
4173 | struct ocfs2_extent_rec *insert_rec, |
4174 | struct ocfs2_extent_tree *et) | |
dcd0538f MF |
4175 | { |
4176 | int i; | |
4177 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4178 | ||
e48edee2 MF |
4179 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4180 | ||
dcd0538f MF |
4181 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
4182 | contig_type = ocfs2_extent_contig(inode, &el->l_recs[i], | |
4183 | insert_rec); | |
4184 | if (contig_type != CONTIG_NONE) { | |
4185 | insert->ins_contig_index = i; | |
4186 | break; | |
4187 | } | |
4188 | } | |
4189 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4190 | |
4191 | if (insert->ins_contig != CONTIG_NONE) { | |
4192 | struct ocfs2_extent_rec *rec = | |
4193 | &el->l_recs[insert->ins_contig_index]; | |
4194 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4195 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4196 | ||
4197 | /* | |
4198 | * Caller might want us to limit the size of extents, don't | |
4199 | * calculate contiguousness if we might exceed that limit. | |
4200 | */ | |
ce1d9ea6 JB |
4201 | if (et->et_max_leaf_clusters && |
4202 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4203 | insert->ins_contig = CONTIG_NONE; |
4204 | } | |
dcd0538f MF |
4205 | } |
4206 | ||
4207 | /* | |
4208 | * This should only be called against the righmost leaf extent list. | |
4209 | * | |
4210 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4211 | * insert at the tail of the rightmost leaf. | |
4212 | * | |
e7d4cb6b TM |
4213 | * This should also work against the root extent list for tree's with 0 |
4214 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4215 | * then the logic here makes sense. |
4216 | */ | |
4217 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4218 | struct ocfs2_extent_list *el, | |
4219 | struct ocfs2_extent_rec *insert_rec) | |
4220 | { | |
4221 | int i; | |
4222 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4223 | struct ocfs2_extent_rec *rec; | |
4224 | ||
4225 | insert->ins_appending = APPEND_NONE; | |
4226 | ||
e48edee2 | 4227 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4228 | |
4229 | if (!el->l_next_free_rec) | |
4230 | goto set_tail_append; | |
4231 | ||
4232 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4233 | /* Were all records empty? */ | |
4234 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4235 | goto set_tail_append; | |
4236 | } | |
4237 | ||
4238 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4239 | rec = &el->l_recs[i]; | |
4240 | ||
e48edee2 MF |
4241 | if (cpos >= |
4242 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4243 | goto set_tail_append; |
4244 | ||
4245 | return; | |
4246 | ||
4247 | set_tail_append: | |
4248 | insert->ins_appending = APPEND_TAIL; | |
4249 | } | |
4250 | ||
4251 | /* | |
4252 | * Helper function called at the begining of an insert. | |
4253 | * | |
4254 | * This computes a few things that are commonly used in the process of | |
4255 | * inserting into the btree: | |
4256 | * - Whether the new extent is contiguous with an existing one. | |
4257 | * - The current tree depth. | |
4258 | * - Whether the insert is an appending one. | |
4259 | * - The total # of free records in the tree. | |
4260 | * | |
4261 | * All of the information is stored on the ocfs2_insert_type | |
4262 | * structure. | |
4263 | */ | |
4264 | static int ocfs2_figure_insert_type(struct inode *inode, | |
e7d4cb6b | 4265 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4266 | struct buffer_head **last_eb_bh, |
4267 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4268 | int *free_records, |
dcd0538f MF |
4269 | struct ocfs2_insert_type *insert) |
4270 | { | |
4271 | int ret; | |
dcd0538f MF |
4272 | struct ocfs2_extent_block *eb; |
4273 | struct ocfs2_extent_list *el; | |
4274 | struct ocfs2_path *path = NULL; | |
4275 | struct buffer_head *bh = NULL; | |
4276 | ||
328d5752 MF |
4277 | insert->ins_split = SPLIT_NONE; |
4278 | ||
ce1d9ea6 | 4279 | el = et->et_root_el; |
dcd0538f MF |
4280 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4281 | ||
4282 | if (el->l_tree_depth) { | |
4283 | /* | |
4284 | * If we have tree depth, we read in the | |
4285 | * rightmost extent block ahead of time as | |
4286 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4287 | * may want it later. | |
4288 | */ | |
0fcaa56a | 4289 | ret = ocfs2_read_block(inode, ocfs2_et_get_last_eb_blk(et), &bh); |
dcd0538f MF |
4290 | if (ret) { |
4291 | mlog_exit(ret); | |
4292 | goto out; | |
4293 | } | |
ccd979bd | 4294 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4295 | el = &eb->h_list; |
dcd0538f | 4296 | } |
ccd979bd | 4297 | |
dcd0538f MF |
4298 | /* |
4299 | * Unless we have a contiguous insert, we'll need to know if | |
4300 | * there is room left in our allocation tree for another | |
4301 | * extent record. | |
4302 | * | |
4303 | * XXX: This test is simplistic, we can search for empty | |
4304 | * extent records too. | |
4305 | */ | |
c77534f6 | 4306 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4307 | le16_to_cpu(el->l_next_free_rec); |
4308 | ||
4309 | if (!insert->ins_tree_depth) { | |
ca12b7c4 | 4310 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4311 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4312 | return 0; | |
ccd979bd MF |
4313 | } |
4314 | ||
ce1d9ea6 | 4315 | path = ocfs2_new_path(et->et_root_bh, et->et_root_el); |
dcd0538f MF |
4316 | if (!path) { |
4317 | ret = -ENOMEM; | |
4318 | mlog_errno(ret); | |
4319 | goto out; | |
4320 | } | |
ccd979bd | 4321 | |
dcd0538f MF |
4322 | /* |
4323 | * In the case that we're inserting past what the tree | |
4324 | * currently accounts for, ocfs2_find_path() will return for | |
4325 | * us the rightmost tree path. This is accounted for below in | |
4326 | * the appending code. | |
4327 | */ | |
4328 | ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos)); | |
4329 | if (ret) { | |
4330 | mlog_errno(ret); | |
4331 | goto out; | |
4332 | } | |
ccd979bd | 4333 | |
dcd0538f MF |
4334 | el = path_leaf_el(path); |
4335 | ||
4336 | /* | |
4337 | * Now that we have the path, there's two things we want to determine: | |
4338 | * 1) Contiguousness (also set contig_index if this is so) | |
4339 | * | |
4340 | * 2) Are we doing an append? We can trivially break this up | |
4341 | * into two types of appends: simple record append, or a | |
4342 | * rotate inside the tail leaf. | |
4343 | */ | |
ca12b7c4 | 4344 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4345 | |
4346 | /* | |
4347 | * The insert code isn't quite ready to deal with all cases of | |
4348 | * left contiguousness. Specifically, if it's an insert into | |
4349 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4350 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4351 | * left. For now, just catch that case and fool the layers |
4352 | * above us. This works just fine for tree_depth == 0, which | |
4353 | * is why we allow that above. | |
4354 | */ | |
4355 | if (insert->ins_contig == CONTIG_LEFT && | |
4356 | insert->ins_contig_index == 0) | |
4357 | insert->ins_contig = CONTIG_NONE; | |
4358 | ||
4359 | /* | |
4360 | * Ok, so we can simply compare against last_eb to figure out | |
4361 | * whether the path doesn't exist. This will only happen in | |
4362 | * the case that we're doing a tail append, so maybe we can | |
4363 | * take advantage of that information somehow. | |
4364 | */ | |
35dc0aa3 | 4365 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4366 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4367 | /* |
4368 | * Ok, ocfs2_find_path() returned us the rightmost | |
4369 | * tree path. This might be an appending insert. There are | |
4370 | * two cases: | |
4371 | * 1) We're doing a true append at the tail: | |
4372 | * -This might even be off the end of the leaf | |
4373 | * 2) We're "appending" by rotating in the tail | |
4374 | */ | |
4375 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4376 | } | |
4377 | ||
4378 | out: | |
4379 | ocfs2_free_path(path); | |
4380 | ||
4381 | if (ret == 0) | |
4382 | *last_eb_bh = bh; | |
4383 | else | |
4384 | brelse(bh); | |
4385 | return ret; | |
ccd979bd MF |
4386 | } |
4387 | ||
dcd0538f MF |
4388 | /* |
4389 | * Insert an extent into an inode btree. | |
4390 | * | |
4391 | * The caller needs to update fe->i_clusters | |
4392 | */ | |
f99b9b7c JB |
4393 | int ocfs2_insert_extent(struct ocfs2_super *osb, |
4394 | handle_t *handle, | |
4395 | struct inode *inode, | |
4396 | struct ocfs2_extent_tree *et, | |
4397 | u32 cpos, | |
4398 | u64 start_blk, | |
4399 | u32 new_clusters, | |
4400 | u8 flags, | |
4401 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4402 | { |
c3afcbb3 | 4403 | int status; |
c77534f6 | 4404 | int uninitialized_var(free_records); |
ccd979bd | 4405 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4406 | struct ocfs2_insert_type insert = {0, }; |
4407 | struct ocfs2_extent_rec rec; | |
4408 | ||
4409 | mlog(0, "add %u clusters at position %u to inode %llu\n", | |
4410 | new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4411 | ||
e48edee2 | 4412 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4413 | rec.e_cpos = cpu_to_le32(cpos); |
4414 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4415 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4416 | rec.e_flags = flags; |
1e61ee79 JB |
4417 | status = ocfs2_et_insert_check(inode, et, &rec); |
4418 | if (status) { | |
4419 | mlog_errno(status); | |
4420 | goto bail; | |
4421 | } | |
dcd0538f | 4422 | |
e7d4cb6b | 4423 | status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec, |
c77534f6 | 4424 | &free_records, &insert); |
dcd0538f MF |
4425 | if (status < 0) { |
4426 | mlog_errno(status); | |
4427 | goto bail; | |
ccd979bd MF |
4428 | } |
4429 | ||
dcd0538f MF |
4430 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4431 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4432 | "Insert.tree_depth: %d\n", | |
4433 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4434 | free_records, insert.ins_tree_depth); |
ccd979bd | 4435 | |
c77534f6 | 4436 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
e7d4cb6b | 4437 | status = ocfs2_grow_tree(inode, handle, et, |
328d5752 | 4438 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4439 | meta_ac); |
4440 | if (status) { | |
ccd979bd MF |
4441 | mlog_errno(status); |
4442 | goto bail; | |
4443 | } | |
ccd979bd MF |
4444 | } |
4445 | ||
dcd0538f | 4446 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
e7d4cb6b | 4447 | status = ocfs2_do_insert_extent(inode, handle, et, &rec, &insert); |
ccd979bd MF |
4448 | if (status < 0) |
4449 | mlog_errno(status); | |
f99b9b7c | 4450 | else if (et->et_ops == &ocfs2_dinode_et_ops) |
83418978 | 4451 | ocfs2_extent_map_insert_rec(inode, &rec); |
ccd979bd MF |
4452 | |
4453 | bail: | |
a81cb88b | 4454 | brelse(last_eb_bh); |
ccd979bd | 4455 | |
f56654c4 TM |
4456 | mlog_exit(status); |
4457 | return status; | |
4458 | } | |
4459 | ||
0eb8d47e TM |
4460 | /* |
4461 | * Allcate and add clusters into the extent b-tree. | |
4462 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4463 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4464 | * it is not limited to the file storage. Any extent tree can use this |
4465 | * function if it implements the proper ocfs2_extent_tree. | |
4466 | */ | |
4467 | int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb, | |
4468 | struct inode *inode, | |
4469 | u32 *logical_offset, | |
4470 | u32 clusters_to_add, | |
4471 | int mark_unwritten, | |
f99b9b7c | 4472 | struct ocfs2_extent_tree *et, |
0eb8d47e TM |
4473 | handle_t *handle, |
4474 | struct ocfs2_alloc_context *data_ac, | |
4475 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4476 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e TM |
4477 | { |
4478 | int status = 0; | |
4479 | int free_extents; | |
4480 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4481 | u32 bit_off, num_bits; | |
4482 | u64 block; | |
4483 | u8 flags = 0; | |
4484 | ||
4485 | BUG_ON(!clusters_to_add); | |
4486 | ||
4487 | if (mark_unwritten) | |
4488 | flags = OCFS2_EXT_UNWRITTEN; | |
4489 | ||
f99b9b7c | 4490 | free_extents = ocfs2_num_free_extents(osb, inode, et); |
0eb8d47e TM |
4491 | if (free_extents < 0) { |
4492 | status = free_extents; | |
4493 | mlog_errno(status); | |
4494 | goto leave; | |
4495 | } | |
4496 | ||
4497 | /* there are two cases which could cause us to EAGAIN in the | |
4498 | * we-need-more-metadata case: | |
4499 | * 1) we haven't reserved *any* | |
4500 | * 2) we are so fragmented, we've needed to add metadata too | |
4501 | * many times. */ | |
4502 | if (!free_extents && !meta_ac) { | |
4503 | mlog(0, "we haven't reserved any metadata!\n"); | |
4504 | status = -EAGAIN; | |
4505 | reason = RESTART_META; | |
4506 | goto leave; | |
4507 | } else if ((!free_extents) | |
4508 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4509 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
0eb8d47e TM |
4510 | mlog(0, "filesystem is really fragmented...\n"); |
4511 | status = -EAGAIN; | |
4512 | reason = RESTART_META; | |
4513 | goto leave; | |
4514 | } | |
4515 | ||
4516 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4517 | clusters_to_add, &bit_off, &num_bits); | |
4518 | if (status < 0) { | |
4519 | if (status != -ENOSPC) | |
4520 | mlog_errno(status); | |
4521 | goto leave; | |
4522 | } | |
4523 | ||
4524 | BUG_ON(num_bits > clusters_to_add); | |
4525 | ||
4526 | /* reserve our write early -- insert_extent may update the inode */ | |
f99b9b7c | 4527 | status = ocfs2_journal_access(handle, inode, et->et_root_bh, |
0eb8d47e TM |
4528 | OCFS2_JOURNAL_ACCESS_WRITE); |
4529 | if (status < 0) { | |
4530 | mlog_errno(status); | |
4531 | goto leave; | |
4532 | } | |
4533 | ||
4534 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
4535 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", | |
4536 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
f99b9b7c JB |
4537 | status = ocfs2_insert_extent(osb, handle, inode, et, |
4538 | *logical_offset, block, | |
4539 | num_bits, flags, meta_ac); | |
0eb8d47e TM |
4540 | if (status < 0) { |
4541 | mlog_errno(status); | |
4542 | goto leave; | |
4543 | } | |
4544 | ||
f99b9b7c | 4545 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4546 | if (status < 0) { |
4547 | mlog_errno(status); | |
4548 | goto leave; | |
4549 | } | |
4550 | ||
4551 | clusters_to_add -= num_bits; | |
4552 | *logical_offset += num_bits; | |
4553 | ||
4554 | if (clusters_to_add) { | |
4555 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4556 | clusters_to_add); | |
4557 | status = -EAGAIN; | |
4558 | reason = RESTART_TRANS; | |
4559 | } | |
4560 | ||
4561 | leave: | |
4562 | mlog_exit(status); | |
4563 | if (reason_ret) | |
4564 | *reason_ret = reason; | |
4565 | return status; | |
4566 | } | |
4567 | ||
328d5752 MF |
4568 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4569 | struct ocfs2_extent_rec *split_rec, | |
4570 | u32 cpos, | |
4571 | struct ocfs2_extent_rec *rec) | |
4572 | { | |
4573 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4574 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4575 | ||
4576 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4577 | ||
4578 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4579 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4580 | ||
4581 | split_rec->e_blkno = rec->e_blkno; | |
4582 | le64_add_cpu(&split_rec->e_blkno, | |
4583 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4584 | ||
4585 | split_rec->e_flags = rec->e_flags; | |
4586 | } | |
4587 | ||
4588 | static int ocfs2_split_and_insert(struct inode *inode, | |
4589 | handle_t *handle, | |
4590 | struct ocfs2_path *path, | |
e7d4cb6b | 4591 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4592 | struct buffer_head **last_eb_bh, |
4593 | int split_index, | |
4594 | struct ocfs2_extent_rec *orig_split_rec, | |
4595 | struct ocfs2_alloc_context *meta_ac) | |
4596 | { | |
4597 | int ret = 0, depth; | |
4598 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4599 | struct ocfs2_extent_rec tmprec; | |
4600 | struct ocfs2_extent_list *rightmost_el; | |
4601 | struct ocfs2_extent_rec rec; | |
4602 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4603 | struct ocfs2_insert_type insert; | |
4604 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4605 | |
4606 | leftright: | |
4607 | /* | |
4608 | * Store a copy of the record on the stack - it might move | |
4609 | * around as the tree is manipulated below. | |
4610 | */ | |
4611 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4612 | ||
ce1d9ea6 | 4613 | rightmost_el = et->et_root_el; |
328d5752 MF |
4614 | |
4615 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4616 | if (depth) { | |
4617 | BUG_ON(!(*last_eb_bh)); | |
4618 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4619 | rightmost_el = &eb->h_list; | |
4620 | } | |
4621 | ||
4622 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4623 | le16_to_cpu(rightmost_el->l_count)) { | |
e7d4cb6b TM |
4624 | ret = ocfs2_grow_tree(inode, handle, et, |
4625 | &depth, last_eb_bh, meta_ac); | |
328d5752 MF |
4626 | if (ret) { |
4627 | mlog_errno(ret); | |
4628 | goto out; | |
4629 | } | |
328d5752 MF |
4630 | } |
4631 | ||
4632 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4633 | insert.ins_appending = APPEND_NONE; | |
4634 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4635 | insert.ins_tree_depth = depth; |
4636 | ||
4637 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4638 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4639 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4640 | le16_to_cpu(rec.e_leaf_clusters); | |
4641 | ||
4642 | if (split_rec.e_cpos == rec.e_cpos) { | |
4643 | insert.ins_split = SPLIT_LEFT; | |
4644 | } else if (insert_range == rec_range) { | |
4645 | insert.ins_split = SPLIT_RIGHT; | |
4646 | } else { | |
4647 | /* | |
4648 | * Left/right split. We fake this as a right split | |
4649 | * first and then make a second pass as a left split. | |
4650 | */ | |
4651 | insert.ins_split = SPLIT_RIGHT; | |
4652 | ||
4653 | ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range, | |
4654 | &rec); | |
4655 | ||
4656 | split_rec = tmprec; | |
4657 | ||
4658 | BUG_ON(do_leftright); | |
4659 | do_leftright = 1; | |
4660 | } | |
4661 | ||
e7d4cb6b | 4662 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
328d5752 MF |
4663 | if (ret) { |
4664 | mlog_errno(ret); | |
4665 | goto out; | |
4666 | } | |
4667 | ||
4668 | if (do_leftright == 1) { | |
4669 | u32 cpos; | |
4670 | struct ocfs2_extent_list *el; | |
4671 | ||
4672 | do_leftright++; | |
4673 | split_rec = *orig_split_rec; | |
4674 | ||
4675 | ocfs2_reinit_path(path, 1); | |
4676 | ||
4677 | cpos = le32_to_cpu(split_rec.e_cpos); | |
4678 | ret = ocfs2_find_path(inode, path, cpos); | |
4679 | if (ret) { | |
4680 | mlog_errno(ret); | |
4681 | goto out; | |
4682 | } | |
4683 | ||
4684 | el = path_leaf_el(path); | |
4685 | split_index = ocfs2_search_extent_list(el, cpos); | |
4686 | goto leftright; | |
4687 | } | |
4688 | out: | |
4689 | ||
4690 | return ret; | |
4691 | } | |
4692 | ||
4693 | /* | |
4694 | * Mark part or all of the extent record at split_index in the leaf | |
4695 | * pointed to by path as written. This removes the unwritten | |
4696 | * extent flag. | |
4697 | * | |
4698 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4699 | * | |
4700 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4701 | * of the tree is required. All other cases will degrade into a less | |
4702 | * optimal tree layout. | |
4703 | * | |
e7d4cb6b TM |
4704 | * last_eb_bh should be the rightmost leaf block for any extent |
4705 | * btree. Since a split may grow the tree or a merge might shrink it, | |
4706 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
4707 | * |
4708 | * This code is optimized for readability - several passes might be | |
4709 | * made over certain portions of the tree. All of those blocks will | |
4710 | * have been brought into cache (and pinned via the journal), so the | |
4711 | * extra overhead is not expressed in terms of disk reads. | |
4712 | */ | |
4713 | static int __ocfs2_mark_extent_written(struct inode *inode, | |
e7d4cb6b | 4714 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4715 | handle_t *handle, |
4716 | struct ocfs2_path *path, | |
4717 | int split_index, | |
4718 | struct ocfs2_extent_rec *split_rec, | |
4719 | struct ocfs2_alloc_context *meta_ac, | |
4720 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
4721 | { | |
4722 | int ret = 0; | |
4723 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 4724 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
4725 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
4726 | struct ocfs2_merge_ctxt ctxt; | |
4727 | struct ocfs2_extent_list *rightmost_el; | |
4728 | ||
3cf0c507 | 4729 | if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) { |
328d5752 MF |
4730 | ret = -EIO; |
4731 | mlog_errno(ret); | |
4732 | goto out; | |
4733 | } | |
4734 | ||
4735 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || | |
4736 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
4737 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
4738 | ret = -EIO; | |
4739 | mlog_errno(ret); | |
4740 | goto out; | |
4741 | } | |
4742 | ||
ad5a4d70 | 4743 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, path, el, |
328d5752 MF |
4744 | split_index, |
4745 | split_rec); | |
4746 | ||
4747 | /* | |
4748 | * The core merge / split code wants to know how much room is | |
4749 | * left in this inodes allocation tree, so we pass the | |
4750 | * rightmost extent list. | |
4751 | */ | |
4752 | if (path->p_tree_depth) { | |
4753 | struct ocfs2_extent_block *eb; | |
328d5752 | 4754 | |
0fcaa56a JB |
4755 | ret = ocfs2_read_block(inode, ocfs2_et_get_last_eb_blk(et), |
4756 | &last_eb_bh); | |
328d5752 MF |
4757 | if (ret) { |
4758 | mlog_exit(ret); | |
4759 | goto out; | |
4760 | } | |
4761 | ||
4762 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
4763 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
4764 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
4765 | ret = -EROFS; | |
4766 | goto out; | |
4767 | } | |
4768 | ||
4769 | rightmost_el = &eb->h_list; | |
4770 | } else | |
4771 | rightmost_el = path_root_el(path); | |
4772 | ||
328d5752 MF |
4773 | if (rec->e_cpos == split_rec->e_cpos && |
4774 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
4775 | ctxt.c_split_covers_rec = 1; | |
4776 | else | |
4777 | ctxt.c_split_covers_rec = 0; | |
4778 | ||
4779 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
4780 | ||
015452b1 MF |
4781 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
4782 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
4783 | ctxt.c_split_covers_rec); | |
328d5752 MF |
4784 | |
4785 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
4786 | if (ctxt.c_split_covers_rec) | |
4787 | el->l_recs[split_index] = *split_rec; | |
4788 | else | |
e7d4cb6b | 4789 | ret = ocfs2_split_and_insert(inode, handle, path, et, |
328d5752 MF |
4790 | &last_eb_bh, split_index, |
4791 | split_rec, meta_ac); | |
4792 | if (ret) | |
4793 | mlog_errno(ret); | |
4794 | } else { | |
4795 | ret = ocfs2_try_to_merge_extent(inode, handle, path, | |
4796 | split_index, split_rec, | |
e7d4cb6b | 4797 | dealloc, &ctxt, et); |
328d5752 MF |
4798 | if (ret) |
4799 | mlog_errno(ret); | |
4800 | } | |
4801 | ||
328d5752 MF |
4802 | out: |
4803 | brelse(last_eb_bh); | |
4804 | return ret; | |
4805 | } | |
4806 | ||
4807 | /* | |
4808 | * Mark the already-existing extent at cpos as written for len clusters. | |
4809 | * | |
4810 | * If the existing extent is larger than the request, initiate a | |
4811 | * split. An attempt will be made at merging with adjacent extents. | |
4812 | * | |
4813 | * The caller is responsible for passing down meta_ac if we'll need it. | |
4814 | */ | |
f99b9b7c JB |
4815 | int ocfs2_mark_extent_written(struct inode *inode, |
4816 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
4817 | handle_t *handle, u32 cpos, u32 len, u32 phys, |
4818 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4819 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
4820 | { |
4821 | int ret, index; | |
4822 | u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys); | |
4823 | struct ocfs2_extent_rec split_rec; | |
4824 | struct ocfs2_path *left_path = NULL; | |
4825 | struct ocfs2_extent_list *el; | |
4826 | ||
4827 | mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n", | |
4828 | inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno); | |
4829 | ||
4830 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
4831 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
4832 | "that are being written to, but the feature bit " | |
4833 | "is not set in the super block.", | |
4834 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4835 | ret = -EROFS; | |
4836 | goto out; | |
4837 | } | |
4838 | ||
4839 | /* | |
4840 | * XXX: This should be fixed up so that we just re-insert the | |
4841 | * next extent records. | |
f99b9b7c JB |
4842 | * |
4843 | * XXX: This is a hack on the extent tree, maybe it should be | |
4844 | * an op? | |
328d5752 | 4845 | */ |
f99b9b7c | 4846 | if (et->et_ops == &ocfs2_dinode_et_ops) |
e7d4cb6b | 4847 | ocfs2_extent_map_trunc(inode, 0); |
328d5752 | 4848 | |
f99b9b7c | 4849 | left_path = ocfs2_new_path(et->et_root_bh, et->et_root_el); |
328d5752 MF |
4850 | if (!left_path) { |
4851 | ret = -ENOMEM; | |
4852 | mlog_errno(ret); | |
4853 | goto out; | |
4854 | } | |
4855 | ||
4856 | ret = ocfs2_find_path(inode, left_path, cpos); | |
4857 | if (ret) { | |
4858 | mlog_errno(ret); | |
4859 | goto out; | |
4860 | } | |
4861 | el = path_leaf_el(left_path); | |
4862 | ||
4863 | index = ocfs2_search_extent_list(el, cpos); | |
4864 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
4865 | ocfs2_error(inode->i_sb, | |
4866 | "Inode %llu has an extent at cpos %u which can no " | |
4867 | "longer be found.\n", | |
4868 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
4869 | ret = -EROFS; | |
4870 | goto out; | |
4871 | } | |
4872 | ||
4873 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4874 | split_rec.e_cpos = cpu_to_le32(cpos); | |
4875 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
4876 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
4877 | split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags; | |
4878 | split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN; | |
4879 | ||
f99b9b7c | 4880 | ret = __ocfs2_mark_extent_written(inode, et, handle, left_path, |
e7d4cb6b TM |
4881 | index, &split_rec, meta_ac, |
4882 | dealloc); | |
328d5752 MF |
4883 | if (ret) |
4884 | mlog_errno(ret); | |
4885 | ||
4886 | out: | |
4887 | ocfs2_free_path(left_path); | |
4888 | return ret; | |
4889 | } | |
4890 | ||
e7d4cb6b | 4891 | static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et, |
d0c7d708 MF |
4892 | handle_t *handle, struct ocfs2_path *path, |
4893 | int index, u32 new_range, | |
4894 | struct ocfs2_alloc_context *meta_ac) | |
4895 | { | |
4896 | int ret, depth, credits = handle->h_buffer_credits; | |
d0c7d708 MF |
4897 | struct buffer_head *last_eb_bh = NULL; |
4898 | struct ocfs2_extent_block *eb; | |
4899 | struct ocfs2_extent_list *rightmost_el, *el; | |
4900 | struct ocfs2_extent_rec split_rec; | |
4901 | struct ocfs2_extent_rec *rec; | |
4902 | struct ocfs2_insert_type insert; | |
4903 | ||
4904 | /* | |
4905 | * Setup the record to split before we grow the tree. | |
4906 | */ | |
4907 | el = path_leaf_el(path); | |
4908 | rec = &el->l_recs[index]; | |
4909 | ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec); | |
4910 | ||
4911 | depth = path->p_tree_depth; | |
4912 | if (depth > 0) { | |
0fcaa56a JB |
4913 | ret = ocfs2_read_block(inode, ocfs2_et_get_last_eb_blk(et), |
4914 | &last_eb_bh); | |
d0c7d708 MF |
4915 | if (ret < 0) { |
4916 | mlog_errno(ret); | |
4917 | goto out; | |
4918 | } | |
4919 | ||
4920 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
4921 | rightmost_el = &eb->h_list; | |
4922 | } else | |
4923 | rightmost_el = path_leaf_el(path); | |
4924 | ||
811f933d | 4925 | credits += path->p_tree_depth + |
ce1d9ea6 | 4926 | ocfs2_extend_meta_needed(et->et_root_el); |
d0c7d708 MF |
4927 | ret = ocfs2_extend_trans(handle, credits); |
4928 | if (ret) { | |
4929 | mlog_errno(ret); | |
4930 | goto out; | |
4931 | } | |
4932 | ||
4933 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4934 | le16_to_cpu(rightmost_el->l_count)) { | |
e7d4cb6b | 4935 | ret = ocfs2_grow_tree(inode, handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
4936 | meta_ac); |
4937 | if (ret) { | |
4938 | mlog_errno(ret); | |
4939 | goto out; | |
4940 | } | |
d0c7d708 MF |
4941 | } |
4942 | ||
4943 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4944 | insert.ins_appending = APPEND_NONE; | |
4945 | insert.ins_contig = CONTIG_NONE; | |
4946 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
4947 | insert.ins_tree_depth = depth; |
4948 | ||
e7d4cb6b | 4949 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
d0c7d708 MF |
4950 | if (ret) |
4951 | mlog_errno(ret); | |
4952 | ||
4953 | out: | |
4954 | brelse(last_eb_bh); | |
4955 | return ret; | |
4956 | } | |
4957 | ||
4958 | static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle, | |
4959 | struct ocfs2_path *path, int index, | |
4960 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
4961 | u32 cpos, u32 len, |
4962 | struct ocfs2_extent_tree *et) | |
d0c7d708 MF |
4963 | { |
4964 | int ret; | |
4965 | u32 left_cpos, rec_range, trunc_range; | |
4966 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
4967 | struct super_block *sb = inode->i_sb; | |
4968 | struct ocfs2_path *left_path = NULL; | |
4969 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
4970 | struct ocfs2_extent_rec *rec; | |
4971 | struct ocfs2_extent_block *eb; | |
4972 | ||
4973 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
e7d4cb6b | 4974 | ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et); |
d0c7d708 MF |
4975 | if (ret) { |
4976 | mlog_errno(ret); | |
4977 | goto out; | |
4978 | } | |
4979 | ||
4980 | index--; | |
4981 | } | |
4982 | ||
4983 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
4984 | path->p_tree_depth) { | |
4985 | /* | |
4986 | * Check whether this is the rightmost tree record. If | |
4987 | * we remove all of this record or part of its right | |
4988 | * edge then an update of the record lengths above it | |
4989 | * will be required. | |
4990 | */ | |
4991 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
4992 | if (eb->h_next_leaf_blk == 0) | |
4993 | is_rightmost_tree_rec = 1; | |
4994 | } | |
4995 | ||
4996 | rec = &el->l_recs[index]; | |
4997 | if (index == 0 && path->p_tree_depth && | |
4998 | le32_to_cpu(rec->e_cpos) == cpos) { | |
4999 | /* | |
5000 | * Changing the leftmost offset (via partial or whole | |
5001 | * record truncate) of an interior (or rightmost) path | |
5002 | * means we have to update the subtree that is formed | |
5003 | * by this leaf and the one to it's left. | |
5004 | * | |
5005 | * There are two cases we can skip: | |
5006 | * 1) Path is the leftmost one in our inode tree. | |
5007 | * 2) The leaf is rightmost and will be empty after | |
5008 | * we remove the extent record - the rotate code | |
5009 | * knows how to update the newly formed edge. | |
5010 | */ | |
5011 | ||
5012 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, | |
5013 | &left_cpos); | |
5014 | if (ret) { | |
5015 | mlog_errno(ret); | |
5016 | goto out; | |
5017 | } | |
5018 | ||
5019 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
5020 | left_path = ocfs2_new_path(path_root_bh(path), | |
5021 | path_root_el(path)); | |
5022 | if (!left_path) { | |
5023 | ret = -ENOMEM; | |
5024 | mlog_errno(ret); | |
5025 | goto out; | |
5026 | } | |
5027 | ||
5028 | ret = ocfs2_find_path(inode, left_path, left_cpos); | |
5029 | if (ret) { | |
5030 | mlog_errno(ret); | |
5031 | goto out; | |
5032 | } | |
5033 | } | |
5034 | } | |
5035 | ||
5036 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5037 | handle->h_buffer_credits, | |
5038 | path); | |
5039 | if (ret) { | |
5040 | mlog_errno(ret); | |
5041 | goto out; | |
5042 | } | |
5043 | ||
5044 | ret = ocfs2_journal_access_path(inode, handle, path); | |
5045 | if (ret) { | |
5046 | mlog_errno(ret); | |
5047 | goto out; | |
5048 | } | |
5049 | ||
5050 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
5051 | if (ret) { | |
5052 | mlog_errno(ret); | |
5053 | goto out; | |
5054 | } | |
5055 | ||
5056 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5057 | trunc_range = cpos + len; | |
5058 | ||
5059 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5060 | int next_free; | |
5061 | ||
5062 | memset(rec, 0, sizeof(*rec)); | |
5063 | ocfs2_cleanup_merge(el, index); | |
5064 | wants_rotate = 1; | |
5065 | ||
5066 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5067 | if (is_rightmost_tree_rec && next_free > 1) { | |
5068 | /* | |
5069 | * We skip the edge update if this path will | |
5070 | * be deleted by the rotate code. | |
5071 | */ | |
5072 | rec = &el->l_recs[next_free - 1]; | |
5073 | ocfs2_adjust_rightmost_records(inode, handle, path, | |
5074 | rec); | |
5075 | } | |
5076 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5077 | /* Remove leftmost portion of the record. */ | |
5078 | le32_add_cpu(&rec->e_cpos, len); | |
5079 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5080 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5081 | } else if (rec_range == trunc_range) { | |
5082 | /* Remove rightmost portion of the record */ | |
5083 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5084 | if (is_rightmost_tree_rec) | |
5085 | ocfs2_adjust_rightmost_records(inode, handle, path, rec); | |
5086 | } else { | |
5087 | /* Caller should have trapped this. */ | |
5088 | mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) " | |
5089 | "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5090 | le32_to_cpu(rec->e_cpos), | |
5091 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5092 | BUG(); | |
5093 | } | |
5094 | ||
5095 | if (left_path) { | |
5096 | int subtree_index; | |
5097 | ||
5098 | subtree_index = ocfs2_find_subtree_root(inode, left_path, path); | |
5099 | ocfs2_complete_edge_insert(inode, handle, left_path, path, | |
5100 | subtree_index); | |
5101 | } | |
5102 | ||
5103 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5104 | ||
e7d4cb6b | 5105 | ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et); |
d0c7d708 MF |
5106 | if (ret) { |
5107 | mlog_errno(ret); | |
5108 | goto out; | |
5109 | } | |
5110 | ||
5111 | out: | |
5112 | ocfs2_free_path(left_path); | |
5113 | return ret; | |
5114 | } | |
5115 | ||
f99b9b7c JB |
5116 | int ocfs2_remove_extent(struct inode *inode, |
5117 | struct ocfs2_extent_tree *et, | |
063c4561 MF |
5118 | u32 cpos, u32 len, handle_t *handle, |
5119 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 5120 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5121 | { |
5122 | int ret, index; | |
5123 | u32 rec_range, trunc_range; | |
5124 | struct ocfs2_extent_rec *rec; | |
5125 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5126 | struct ocfs2_path *path = NULL; |
d0c7d708 MF |
5127 | |
5128 | ocfs2_extent_map_trunc(inode, 0); | |
5129 | ||
f99b9b7c | 5130 | path = ocfs2_new_path(et->et_root_bh, et->et_root_el); |
d0c7d708 MF |
5131 | if (!path) { |
5132 | ret = -ENOMEM; | |
5133 | mlog_errno(ret); | |
5134 | goto out; | |
5135 | } | |
5136 | ||
5137 | ret = ocfs2_find_path(inode, path, cpos); | |
5138 | if (ret) { | |
5139 | mlog_errno(ret); | |
5140 | goto out; | |
5141 | } | |
5142 | ||
5143 | el = path_leaf_el(path); | |
5144 | index = ocfs2_search_extent_list(el, cpos); | |
5145 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5146 | ocfs2_error(inode->i_sb, | |
5147 | "Inode %llu has an extent at cpos %u which can no " | |
5148 | "longer be found.\n", | |
5149 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5150 | ret = -EROFS; | |
5151 | goto out; | |
5152 | } | |
5153 | ||
5154 | /* | |
5155 | * We have 3 cases of extent removal: | |
5156 | * 1) Range covers the entire extent rec | |
5157 | * 2) Range begins or ends on one edge of the extent rec | |
5158 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5159 | * | |
5160 | * For case 1 we remove the extent rec and left rotate to | |
5161 | * fill the hole. | |
5162 | * | |
5163 | * For case 2 we just shrink the existing extent rec, with a | |
5164 | * tree update if the shrinking edge is also the edge of an | |
5165 | * extent block. | |
5166 | * | |
5167 | * For case 3 we do a right split to turn the extent rec into | |
5168 | * something case 2 can handle. | |
5169 | */ | |
5170 | rec = &el->l_recs[index]; | |
5171 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5172 | trunc_range = cpos + len; | |
5173 | ||
5174 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5175 | ||
5176 | mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d " | |
5177 | "(cpos %u, len %u)\n", | |
5178 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index, | |
5179 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); | |
5180 | ||
5181 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
5182 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
f99b9b7c | 5183 | cpos, len, et); |
d0c7d708 MF |
5184 | if (ret) { |
5185 | mlog_errno(ret); | |
5186 | goto out; | |
5187 | } | |
5188 | } else { | |
f99b9b7c | 5189 | ret = ocfs2_split_tree(inode, et, handle, path, index, |
d0c7d708 MF |
5190 | trunc_range, meta_ac); |
5191 | if (ret) { | |
5192 | mlog_errno(ret); | |
5193 | goto out; | |
5194 | } | |
5195 | ||
5196 | /* | |
5197 | * The split could have manipulated the tree enough to | |
5198 | * move the record location, so we have to look for it again. | |
5199 | */ | |
5200 | ocfs2_reinit_path(path, 1); | |
5201 | ||
5202 | ret = ocfs2_find_path(inode, path, cpos); | |
5203 | if (ret) { | |
5204 | mlog_errno(ret); | |
5205 | goto out; | |
5206 | } | |
5207 | ||
5208 | el = path_leaf_el(path); | |
5209 | index = ocfs2_search_extent_list(el, cpos); | |
5210 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5211 | ocfs2_error(inode->i_sb, | |
5212 | "Inode %llu: split at cpos %u lost record.", | |
5213 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5214 | cpos); | |
5215 | ret = -EROFS; | |
5216 | goto out; | |
5217 | } | |
5218 | ||
5219 | /* | |
5220 | * Double check our values here. If anything is fishy, | |
5221 | * it's easier to catch it at the top level. | |
5222 | */ | |
5223 | rec = &el->l_recs[index]; | |
5224 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5225 | ocfs2_rec_clusters(el, rec); | |
5226 | if (rec_range != trunc_range) { | |
5227 | ocfs2_error(inode->i_sb, | |
5228 | "Inode %llu: error after split at cpos %u" | |
5229 | "trunc len %u, existing record is (%u,%u)", | |
5230 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5231 | cpos, len, le32_to_cpu(rec->e_cpos), | |
5232 | ocfs2_rec_clusters(el, rec)); | |
5233 | ret = -EROFS; | |
5234 | goto out; | |
5235 | } | |
5236 | ||
5237 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
f99b9b7c | 5238 | cpos, len, et); |
d0c7d708 MF |
5239 | if (ret) { |
5240 | mlog_errno(ret); | |
5241 | goto out; | |
5242 | } | |
5243 | } | |
5244 | ||
5245 | out: | |
5246 | ocfs2_free_path(path); | |
5247 | return ret; | |
5248 | } | |
5249 | ||
fecc0112 MF |
5250 | int ocfs2_remove_btree_range(struct inode *inode, |
5251 | struct ocfs2_extent_tree *et, | |
5252 | u32 cpos, u32 phys_cpos, u32 len, | |
5253 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5254 | { | |
5255 | int ret; | |
5256 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
5257 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5258 | struct inode *tl_inode = osb->osb_tl_inode; | |
5259 | handle_t *handle; | |
5260 | struct ocfs2_alloc_context *meta_ac = NULL; | |
5261 | ||
5262 | ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac); | |
5263 | if (ret) { | |
5264 | mlog_errno(ret); | |
5265 | return ret; | |
5266 | } | |
5267 | ||
5268 | mutex_lock(&tl_inode->i_mutex); | |
5269 | ||
5270 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5271 | ret = __ocfs2_flush_truncate_log(osb); | |
5272 | if (ret < 0) { | |
5273 | mlog_errno(ret); | |
5274 | goto out; | |
5275 | } | |
5276 | } | |
5277 | ||
5278 | handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); | |
5279 | if (IS_ERR(handle)) { | |
5280 | ret = PTR_ERR(handle); | |
5281 | mlog_errno(ret); | |
5282 | goto out; | |
5283 | } | |
5284 | ||
5285 | ret = ocfs2_journal_access(handle, inode, et->et_root_bh, | |
5286 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5287 | if (ret) { | |
5288 | mlog_errno(ret); | |
5289 | goto out; | |
5290 | } | |
5291 | ||
5292 | ret = ocfs2_remove_extent(inode, et, cpos, len, handle, meta_ac, | |
5293 | dealloc); | |
5294 | if (ret) { | |
5295 | mlog_errno(ret); | |
5296 | goto out_commit; | |
5297 | } | |
5298 | ||
5299 | ocfs2_et_update_clusters(inode, et, -len); | |
5300 | ||
5301 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); | |
5302 | if (ret) { | |
5303 | mlog_errno(ret); | |
5304 | goto out_commit; | |
5305 | } | |
5306 | ||
5307 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
5308 | if (ret) | |
5309 | mlog_errno(ret); | |
5310 | ||
5311 | out_commit: | |
5312 | ocfs2_commit_trans(osb, handle); | |
5313 | out: | |
5314 | mutex_unlock(&tl_inode->i_mutex); | |
5315 | ||
5316 | if (meta_ac) | |
5317 | ocfs2_free_alloc_context(meta_ac); | |
5318 | ||
5319 | return ret; | |
5320 | } | |
5321 | ||
063c4561 | 5322 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5323 | { |
5324 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5325 | struct ocfs2_dinode *di; | |
5326 | struct ocfs2_truncate_log *tl; | |
5327 | ||
5328 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5329 | tl = &di->id2.i_dealloc; | |
5330 | ||
5331 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5332 | "slot %d, invalid truncate log parameters: used = " | |
5333 | "%u, count = %u\n", osb->slot_num, | |
5334 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5335 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5336 | } | |
5337 | ||
5338 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5339 | unsigned int new_start) | |
5340 | { | |
5341 | unsigned int tail_index; | |
5342 | unsigned int current_tail; | |
5343 | ||
5344 | /* No records, nothing to coalesce */ | |
5345 | if (!le16_to_cpu(tl->tl_used)) | |
5346 | return 0; | |
5347 | ||
5348 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5349 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5350 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5351 | ||
5352 | return current_tail == new_start; | |
5353 | } | |
5354 | ||
063c4561 MF |
5355 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5356 | handle_t *handle, | |
5357 | u64 start_blk, | |
5358 | unsigned int num_clusters) | |
ccd979bd MF |
5359 | { |
5360 | int status, index; | |
5361 | unsigned int start_cluster, tl_count; | |
5362 | struct inode *tl_inode = osb->osb_tl_inode; | |
5363 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5364 | struct ocfs2_dinode *di; | |
5365 | struct ocfs2_truncate_log *tl; | |
5366 | ||
b0697053 MF |
5367 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5368 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5369 | |
1b1dcc1b | 5370 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5371 | |
5372 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5373 | ||
5374 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5375 | |
10995aa2 JB |
5376 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5377 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5378 | * corruption is a code bug */ | |
5379 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5380 | ||
5381 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5382 | tl_count = le16_to_cpu(tl->tl_count); |
5383 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5384 | tl_count == 0, | |
b0697053 MF |
5385 | "Truncate record count on #%llu invalid " |
5386 | "wanted %u, actual %u\n", | |
5387 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5388 | ocfs2_truncate_recs_per_inode(osb->sb), |
5389 | le16_to_cpu(tl->tl_count)); | |
5390 | ||
5391 | /* Caller should have known to flush before calling us. */ | |
5392 | index = le16_to_cpu(tl->tl_used); | |
5393 | if (index >= tl_count) { | |
5394 | status = -ENOSPC; | |
5395 | mlog_errno(status); | |
5396 | goto bail; | |
5397 | } | |
5398 | ||
5399 | status = ocfs2_journal_access(handle, tl_inode, tl_bh, | |
5400 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5401 | if (status < 0) { | |
5402 | mlog_errno(status); | |
5403 | goto bail; | |
5404 | } | |
5405 | ||
5406 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5407 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5408 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5409 | |
5410 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5411 | /* | |
5412 | * Move index back to the record we are coalescing with. | |
5413 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5414 | */ | |
5415 | index--; | |
5416 | ||
5417 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5418 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5419 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5420 | num_clusters); | |
5421 | } else { | |
5422 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5423 | tl->tl_used = cpu_to_le16(index + 1); | |
5424 | } | |
5425 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5426 | ||
5427 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5428 | if (status < 0) { | |
5429 | mlog_errno(status); | |
5430 | goto bail; | |
5431 | } | |
5432 | ||
5433 | bail: | |
5434 | mlog_exit(status); | |
5435 | return status; | |
5436 | } | |
5437 | ||
5438 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5439 | handle_t *handle, |
ccd979bd MF |
5440 | struct inode *data_alloc_inode, |
5441 | struct buffer_head *data_alloc_bh) | |
5442 | { | |
5443 | int status = 0; | |
5444 | int i; | |
5445 | unsigned int num_clusters; | |
5446 | u64 start_blk; | |
5447 | struct ocfs2_truncate_rec rec; | |
5448 | struct ocfs2_dinode *di; | |
5449 | struct ocfs2_truncate_log *tl; | |
5450 | struct inode *tl_inode = osb->osb_tl_inode; | |
5451 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5452 | ||
5453 | mlog_entry_void(); | |
5454 | ||
5455 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5456 | tl = &di->id2.i_dealloc; | |
5457 | i = le16_to_cpu(tl->tl_used) - 1; | |
5458 | while (i >= 0) { | |
5459 | /* Caller has given us at least enough credits to | |
5460 | * update the truncate log dinode */ | |
5461 | status = ocfs2_journal_access(handle, tl_inode, tl_bh, | |
5462 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5463 | if (status < 0) { | |
5464 | mlog_errno(status); | |
5465 | goto bail; | |
5466 | } | |
5467 | ||
5468 | tl->tl_used = cpu_to_le16(i); | |
5469 | ||
5470 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5471 | if (status < 0) { | |
5472 | mlog_errno(status); | |
5473 | goto bail; | |
5474 | } | |
5475 | ||
5476 | /* TODO: Perhaps we can calculate the bulk of the | |
5477 | * credits up front rather than extending like | |
5478 | * this. */ | |
5479 | status = ocfs2_extend_trans(handle, | |
5480 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5481 | if (status < 0) { | |
5482 | mlog_errno(status); | |
5483 | goto bail; | |
5484 | } | |
5485 | ||
5486 | rec = tl->tl_recs[i]; | |
5487 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5488 | le32_to_cpu(rec.t_start)); | |
5489 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5490 | ||
5491 | /* if start_blk is not set, we ignore the record as | |
5492 | * invalid. */ | |
5493 | if (start_blk) { | |
5494 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5495 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5496 | ||
5497 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5498 | data_alloc_bh, start_blk, | |
5499 | num_clusters); | |
5500 | if (status < 0) { | |
5501 | mlog_errno(status); | |
5502 | goto bail; | |
5503 | } | |
5504 | } | |
5505 | i--; | |
5506 | } | |
5507 | ||
5508 | bail: | |
5509 | mlog_exit(status); | |
5510 | return status; | |
5511 | } | |
5512 | ||
1b1dcc1b | 5513 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5514 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5515 | { |
5516 | int status; | |
5517 | unsigned int num_to_flush; | |
1fabe148 | 5518 | handle_t *handle; |
ccd979bd MF |
5519 | struct inode *tl_inode = osb->osb_tl_inode; |
5520 | struct inode *data_alloc_inode = NULL; | |
5521 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5522 | struct buffer_head *data_alloc_bh = NULL; | |
5523 | struct ocfs2_dinode *di; | |
5524 | struct ocfs2_truncate_log *tl; | |
5525 | ||
5526 | mlog_entry_void(); | |
5527 | ||
1b1dcc1b | 5528 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5529 | |
5530 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5531 | |
10995aa2 JB |
5532 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5533 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5534 | * corruption is a code bug */ | |
5535 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5536 | ||
5537 | tl = &di->id2.i_dealloc; | |
ccd979bd | 5538 | num_to_flush = le16_to_cpu(tl->tl_used); |
b0697053 MF |
5539 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5540 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5541 | if (!num_to_flush) { |
5542 | status = 0; | |
e08dc8b9 | 5543 | goto out; |
ccd979bd MF |
5544 | } |
5545 | ||
5546 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5547 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5548 | OCFS2_INVALID_SLOT); | |
5549 | if (!data_alloc_inode) { | |
5550 | status = -EINVAL; | |
5551 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5552 | goto out; |
ccd979bd MF |
5553 | } |
5554 | ||
e08dc8b9 MF |
5555 | mutex_lock(&data_alloc_inode->i_mutex); |
5556 | ||
e63aecb6 | 5557 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5558 | if (status < 0) { |
5559 | mlog_errno(status); | |
e08dc8b9 | 5560 | goto out_mutex; |
ccd979bd MF |
5561 | } |
5562 | ||
65eff9cc | 5563 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5564 | if (IS_ERR(handle)) { |
5565 | status = PTR_ERR(handle); | |
ccd979bd | 5566 | mlog_errno(status); |
e08dc8b9 | 5567 | goto out_unlock; |
ccd979bd MF |
5568 | } |
5569 | ||
5570 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5571 | data_alloc_bh); | |
e08dc8b9 | 5572 | if (status < 0) |
ccd979bd | 5573 | mlog_errno(status); |
ccd979bd | 5574 | |
02dc1af4 | 5575 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 5576 | |
e08dc8b9 MF |
5577 | out_unlock: |
5578 | brelse(data_alloc_bh); | |
e63aecb6 | 5579 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 5580 | |
e08dc8b9 MF |
5581 | out_mutex: |
5582 | mutex_unlock(&data_alloc_inode->i_mutex); | |
5583 | iput(data_alloc_inode); | |
ccd979bd | 5584 | |
e08dc8b9 | 5585 | out: |
ccd979bd MF |
5586 | mlog_exit(status); |
5587 | return status; | |
5588 | } | |
5589 | ||
5590 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
5591 | { | |
5592 | int status; | |
5593 | struct inode *tl_inode = osb->osb_tl_inode; | |
5594 | ||
1b1dcc1b | 5595 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 5596 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 5597 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5598 | |
5599 | return status; | |
5600 | } | |
5601 | ||
c4028958 | 5602 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
5603 | { |
5604 | int status; | |
c4028958 DH |
5605 | struct ocfs2_super *osb = |
5606 | container_of(work, struct ocfs2_super, | |
5607 | osb_truncate_log_wq.work); | |
ccd979bd MF |
5608 | |
5609 | mlog_entry_void(); | |
5610 | ||
5611 | status = ocfs2_flush_truncate_log(osb); | |
5612 | if (status < 0) | |
5613 | mlog_errno(status); | |
4d0ddb2c TM |
5614 | else |
5615 | ocfs2_init_inode_steal_slot(osb); | |
ccd979bd MF |
5616 | |
5617 | mlog_exit(status); | |
5618 | } | |
5619 | ||
5620 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
5621 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
5622 | int cancel) | |
5623 | { | |
5624 | if (osb->osb_tl_inode) { | |
5625 | /* We want to push off log flushes while truncates are | |
5626 | * still running. */ | |
5627 | if (cancel) | |
5628 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5629 | ||
5630 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
5631 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
5632 | } | |
5633 | } | |
5634 | ||
5635 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
5636 | int slot_num, | |
5637 | struct inode **tl_inode, | |
5638 | struct buffer_head **tl_bh) | |
5639 | { | |
5640 | int status; | |
5641 | struct inode *inode = NULL; | |
5642 | struct buffer_head *bh = NULL; | |
5643 | ||
5644 | inode = ocfs2_get_system_file_inode(osb, | |
5645 | TRUNCATE_LOG_SYSTEM_INODE, | |
5646 | slot_num); | |
5647 | if (!inode) { | |
5648 | status = -EINVAL; | |
5649 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
5650 | goto bail; | |
5651 | } | |
5652 | ||
b657c95c | 5653 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
5654 | if (status < 0) { |
5655 | iput(inode); | |
5656 | mlog_errno(status); | |
5657 | goto bail; | |
5658 | } | |
5659 | ||
5660 | *tl_inode = inode; | |
5661 | *tl_bh = bh; | |
5662 | bail: | |
5663 | mlog_exit(status); | |
5664 | return status; | |
5665 | } | |
5666 | ||
5667 | /* called during the 1st stage of node recovery. we stamp a clean | |
5668 | * truncate log and pass back a copy for processing later. if the | |
5669 | * truncate log does not require processing, a *tl_copy is set to | |
5670 | * NULL. */ | |
5671 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
5672 | int slot_num, | |
5673 | struct ocfs2_dinode **tl_copy) | |
5674 | { | |
5675 | int status; | |
5676 | struct inode *tl_inode = NULL; | |
5677 | struct buffer_head *tl_bh = NULL; | |
5678 | struct ocfs2_dinode *di; | |
5679 | struct ocfs2_truncate_log *tl; | |
5680 | ||
5681 | *tl_copy = NULL; | |
5682 | ||
5683 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
5684 | ||
5685 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
5686 | if (status < 0) { | |
5687 | mlog_errno(status); | |
5688 | goto bail; | |
5689 | } | |
5690 | ||
5691 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5692 | |
10995aa2 JB |
5693 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
5694 | * validated by the underlying call to ocfs2_read_inode_block(), | |
5695 | * so any corruption is a code bug */ | |
5696 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5697 | ||
5698 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5699 | if (le16_to_cpu(tl->tl_used)) { |
5700 | mlog(0, "We'll have %u logs to recover\n", | |
5701 | le16_to_cpu(tl->tl_used)); | |
5702 | ||
5703 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
5704 | if (!(*tl_copy)) { | |
5705 | status = -ENOMEM; | |
5706 | mlog_errno(status); | |
5707 | goto bail; | |
5708 | } | |
5709 | ||
5710 | /* Assuming the write-out below goes well, this copy | |
5711 | * will be passed back to recovery for processing. */ | |
5712 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
5713 | ||
5714 | /* All we need to do to clear the truncate log is set | |
5715 | * tl_used. */ | |
5716 | tl->tl_used = 0; | |
5717 | ||
5718 | status = ocfs2_write_block(osb, tl_bh, tl_inode); | |
5719 | if (status < 0) { | |
5720 | mlog_errno(status); | |
5721 | goto bail; | |
5722 | } | |
5723 | } | |
5724 | ||
5725 | bail: | |
5726 | if (tl_inode) | |
5727 | iput(tl_inode); | |
a81cb88b | 5728 | brelse(tl_bh); |
ccd979bd MF |
5729 | |
5730 | if (status < 0 && (*tl_copy)) { | |
5731 | kfree(*tl_copy); | |
5732 | *tl_copy = NULL; | |
5733 | } | |
5734 | ||
5735 | mlog_exit(status); | |
5736 | return status; | |
5737 | } | |
5738 | ||
5739 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
5740 | struct ocfs2_dinode *tl_copy) | |
5741 | { | |
5742 | int status = 0; | |
5743 | int i; | |
5744 | unsigned int clusters, num_recs, start_cluster; | |
5745 | u64 start_blk; | |
1fabe148 | 5746 | handle_t *handle; |
ccd979bd MF |
5747 | struct inode *tl_inode = osb->osb_tl_inode; |
5748 | struct ocfs2_truncate_log *tl; | |
5749 | ||
5750 | mlog_entry_void(); | |
5751 | ||
5752 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
5753 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
5754 | return -EINVAL; | |
5755 | } | |
5756 | ||
5757 | tl = &tl_copy->id2.i_dealloc; | |
5758 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 5759 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 5760 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 5761 | |
1b1dcc1b | 5762 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
5763 | for(i = 0; i < num_recs; i++) { |
5764 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5765 | status = __ocfs2_flush_truncate_log(osb); | |
5766 | if (status < 0) { | |
5767 | mlog_errno(status); | |
5768 | goto bail_up; | |
5769 | } | |
5770 | } | |
5771 | ||
65eff9cc | 5772 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5773 | if (IS_ERR(handle)) { |
5774 | status = PTR_ERR(handle); | |
5775 | mlog_errno(status); | |
5776 | goto bail_up; | |
5777 | } | |
5778 | ||
5779 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
5780 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
5781 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
5782 | ||
5783 | status = ocfs2_truncate_log_append(osb, handle, | |
5784 | start_blk, clusters); | |
02dc1af4 | 5785 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
5786 | if (status < 0) { |
5787 | mlog_errno(status); | |
5788 | goto bail_up; | |
5789 | } | |
5790 | } | |
5791 | ||
5792 | bail_up: | |
1b1dcc1b | 5793 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5794 | |
5795 | mlog_exit(status); | |
5796 | return status; | |
5797 | } | |
5798 | ||
5799 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
5800 | { | |
5801 | int status; | |
5802 | struct inode *tl_inode = osb->osb_tl_inode; | |
5803 | ||
5804 | mlog_entry_void(); | |
5805 | ||
5806 | if (tl_inode) { | |
5807 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5808 | flush_workqueue(ocfs2_wq); | |
5809 | ||
5810 | status = ocfs2_flush_truncate_log(osb); | |
5811 | if (status < 0) | |
5812 | mlog_errno(status); | |
5813 | ||
5814 | brelse(osb->osb_tl_bh); | |
5815 | iput(osb->osb_tl_inode); | |
5816 | } | |
5817 | ||
5818 | mlog_exit_void(); | |
5819 | } | |
5820 | ||
5821 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
5822 | { | |
5823 | int status; | |
5824 | struct inode *tl_inode = NULL; | |
5825 | struct buffer_head *tl_bh = NULL; | |
5826 | ||
5827 | mlog_entry_void(); | |
5828 | ||
5829 | status = ocfs2_get_truncate_log_info(osb, | |
5830 | osb->slot_num, | |
5831 | &tl_inode, | |
5832 | &tl_bh); | |
5833 | if (status < 0) | |
5834 | mlog_errno(status); | |
5835 | ||
5836 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
5837 | * osb->osb_tl_inode so we don't set any of the osb variables | |
5838 | * until we're sure all is well. */ | |
c4028958 DH |
5839 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
5840 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
5841 | osb->osb_tl_bh = tl_bh; |
5842 | osb->osb_tl_inode = tl_inode; | |
5843 | ||
5844 | mlog_exit(status); | |
5845 | return status; | |
5846 | } | |
5847 | ||
2b604351 MF |
5848 | /* |
5849 | * Delayed de-allocation of suballocator blocks. | |
5850 | * | |
5851 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
5852 | * | |
5853 | * The locking for this can get extremely complicated, especially when | |
5854 | * the suballocator inodes to delete from aren't known until deep | |
5855 | * within an unrelated codepath. | |
5856 | * | |
5857 | * ocfs2_extent_block structures are a good example of this - an inode | |
5858 | * btree could have been grown by any number of nodes each allocating | |
5859 | * out of their own suballoc inode. | |
5860 | * | |
5861 | * These structures allow the delay of block de-allocation until a | |
5862 | * later time, when locking of multiple cluster inodes won't cause | |
5863 | * deadlock. | |
5864 | */ | |
5865 | ||
5866 | /* | |
2891d290 TM |
5867 | * Describe a single bit freed from a suballocator. For the block |
5868 | * suballocators, it represents one block. For the global cluster | |
5869 | * allocator, it represents some clusters and free_bit indicates | |
5870 | * clusters number. | |
2b604351 MF |
5871 | */ |
5872 | struct ocfs2_cached_block_free { | |
5873 | struct ocfs2_cached_block_free *free_next; | |
5874 | u64 free_blk; | |
5875 | unsigned int free_bit; | |
5876 | }; | |
5877 | ||
5878 | struct ocfs2_per_slot_free_list { | |
5879 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
5880 | int f_inode_type; | |
5881 | int f_slot; | |
5882 | struct ocfs2_cached_block_free *f_first; | |
5883 | }; | |
5884 | ||
2891d290 TM |
5885 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
5886 | int sysfile_type, | |
5887 | int slot, | |
5888 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
5889 | { |
5890 | int ret; | |
5891 | u64 bg_blkno; | |
5892 | handle_t *handle; | |
5893 | struct inode *inode; | |
5894 | struct buffer_head *di_bh = NULL; | |
5895 | struct ocfs2_cached_block_free *tmp; | |
5896 | ||
5897 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
5898 | if (!inode) { | |
5899 | ret = -EINVAL; | |
5900 | mlog_errno(ret); | |
5901 | goto out; | |
5902 | } | |
5903 | ||
5904 | mutex_lock(&inode->i_mutex); | |
5905 | ||
e63aecb6 | 5906 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
5907 | if (ret) { |
5908 | mlog_errno(ret); | |
5909 | goto out_mutex; | |
5910 | } | |
5911 | ||
5912 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
5913 | if (IS_ERR(handle)) { | |
5914 | ret = PTR_ERR(handle); | |
5915 | mlog_errno(ret); | |
5916 | goto out_unlock; | |
5917 | } | |
5918 | ||
5919 | while (head) { | |
5920 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
5921 | head->free_bit); | |
5922 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
5923 | head->free_bit, (unsigned long long)head->free_blk); | |
5924 | ||
5925 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
5926 | head->free_bit, bg_blkno, 1); | |
5927 | if (ret) { | |
5928 | mlog_errno(ret); | |
5929 | goto out_journal; | |
5930 | } | |
5931 | ||
5932 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
5933 | if (ret) { | |
5934 | mlog_errno(ret); | |
5935 | goto out_journal; | |
5936 | } | |
5937 | ||
5938 | tmp = head; | |
5939 | head = head->free_next; | |
5940 | kfree(tmp); | |
5941 | } | |
5942 | ||
5943 | out_journal: | |
5944 | ocfs2_commit_trans(osb, handle); | |
5945 | ||
5946 | out_unlock: | |
e63aecb6 | 5947 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
5948 | brelse(di_bh); |
5949 | out_mutex: | |
5950 | mutex_unlock(&inode->i_mutex); | |
5951 | iput(inode); | |
5952 | out: | |
5953 | while(head) { | |
5954 | /* Premature exit may have left some dangling items. */ | |
5955 | tmp = head; | |
5956 | head = head->free_next; | |
5957 | kfree(tmp); | |
5958 | } | |
5959 | ||
5960 | return ret; | |
5961 | } | |
5962 | ||
2891d290 TM |
5963 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
5964 | u64 blkno, unsigned int bit) | |
5965 | { | |
5966 | int ret = 0; | |
5967 | struct ocfs2_cached_block_free *item; | |
5968 | ||
5969 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
5970 | if (item == NULL) { | |
5971 | ret = -ENOMEM; | |
5972 | mlog_errno(ret); | |
5973 | return ret; | |
5974 | } | |
5975 | ||
5976 | mlog(0, "Insert clusters: (bit %u, blk %llu)\n", | |
5977 | bit, (unsigned long long)blkno); | |
5978 | ||
5979 | item->free_blk = blkno; | |
5980 | item->free_bit = bit; | |
5981 | item->free_next = ctxt->c_global_allocator; | |
5982 | ||
5983 | ctxt->c_global_allocator = item; | |
5984 | return ret; | |
5985 | } | |
5986 | ||
5987 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
5988 | struct ocfs2_cached_block_free *head) | |
5989 | { | |
5990 | struct ocfs2_cached_block_free *tmp; | |
5991 | struct inode *tl_inode = osb->osb_tl_inode; | |
5992 | handle_t *handle; | |
5993 | int ret = 0; | |
5994 | ||
5995 | mutex_lock(&tl_inode->i_mutex); | |
5996 | ||
5997 | while (head) { | |
5998 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5999 | ret = __ocfs2_flush_truncate_log(osb); | |
6000 | if (ret < 0) { | |
6001 | mlog_errno(ret); | |
6002 | break; | |
6003 | } | |
6004 | } | |
6005 | ||
6006 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6007 | if (IS_ERR(handle)) { | |
6008 | ret = PTR_ERR(handle); | |
6009 | mlog_errno(ret); | |
6010 | break; | |
6011 | } | |
6012 | ||
6013 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6014 | head->free_bit); | |
6015 | ||
6016 | ocfs2_commit_trans(osb, handle); | |
6017 | tmp = head; | |
6018 | head = head->free_next; | |
6019 | kfree(tmp); | |
6020 | ||
6021 | if (ret < 0) { | |
6022 | mlog_errno(ret); | |
6023 | break; | |
6024 | } | |
6025 | } | |
6026 | ||
6027 | mutex_unlock(&tl_inode->i_mutex); | |
6028 | ||
6029 | while (head) { | |
6030 | /* Premature exit may have left some dangling items. */ | |
6031 | tmp = head; | |
6032 | head = head->free_next; | |
6033 | kfree(tmp); | |
6034 | } | |
6035 | ||
6036 | return ret; | |
6037 | } | |
6038 | ||
2b604351 MF |
6039 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6040 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6041 | { | |
6042 | int ret = 0, ret2; | |
6043 | struct ocfs2_per_slot_free_list *fl; | |
6044 | ||
6045 | if (!ctxt) | |
6046 | return 0; | |
6047 | ||
6048 | while (ctxt->c_first_suballocator) { | |
6049 | fl = ctxt->c_first_suballocator; | |
6050 | ||
6051 | if (fl->f_first) { | |
6052 | mlog(0, "Free items: (type %u, slot %d)\n", | |
6053 | fl->f_inode_type, fl->f_slot); | |
2891d290 TM |
6054 | ret2 = ocfs2_free_cached_blocks(osb, |
6055 | fl->f_inode_type, | |
6056 | fl->f_slot, | |
6057 | fl->f_first); | |
2b604351 MF |
6058 | if (ret2) |
6059 | mlog_errno(ret2); | |
6060 | if (!ret) | |
6061 | ret = ret2; | |
6062 | } | |
6063 | ||
6064 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6065 | kfree(fl); | |
6066 | } | |
6067 | ||
2891d290 TM |
6068 | if (ctxt->c_global_allocator) { |
6069 | ret2 = ocfs2_free_cached_clusters(osb, | |
6070 | ctxt->c_global_allocator); | |
6071 | if (ret2) | |
6072 | mlog_errno(ret2); | |
6073 | if (!ret) | |
6074 | ret = ret2; | |
6075 | ||
6076 | ctxt->c_global_allocator = NULL; | |
6077 | } | |
6078 | ||
2b604351 MF |
6079 | return ret; |
6080 | } | |
6081 | ||
6082 | static struct ocfs2_per_slot_free_list * | |
6083 | ocfs2_find_per_slot_free_list(int type, | |
6084 | int slot, | |
6085 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6086 | { | |
6087 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6088 | ||
6089 | while (fl) { | |
6090 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6091 | return fl; | |
6092 | ||
6093 | fl = fl->f_next_suballocator; | |
6094 | } | |
6095 | ||
6096 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6097 | if (fl) { | |
6098 | fl->f_inode_type = type; | |
6099 | fl->f_slot = slot; | |
6100 | fl->f_first = NULL; | |
6101 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6102 | ||
6103 | ctxt->c_first_suballocator = fl; | |
6104 | } | |
6105 | return fl; | |
6106 | } | |
6107 | ||
6108 | static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, | |
6109 | int type, int slot, u64 blkno, | |
6110 | unsigned int bit) | |
6111 | { | |
6112 | int ret; | |
6113 | struct ocfs2_per_slot_free_list *fl; | |
6114 | struct ocfs2_cached_block_free *item; | |
6115 | ||
6116 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6117 | if (fl == NULL) { | |
6118 | ret = -ENOMEM; | |
6119 | mlog_errno(ret); | |
6120 | goto out; | |
6121 | } | |
6122 | ||
6123 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6124 | if (item == NULL) { | |
6125 | ret = -ENOMEM; | |
6126 | mlog_errno(ret); | |
6127 | goto out; | |
6128 | } | |
6129 | ||
6130 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6131 | type, slot, bit, (unsigned long long)blkno); | |
6132 | ||
6133 | item->free_blk = blkno; | |
6134 | item->free_bit = bit; | |
6135 | item->free_next = fl->f_first; | |
6136 | ||
6137 | fl->f_first = item; | |
6138 | ||
6139 | ret = 0; | |
6140 | out: | |
6141 | return ret; | |
6142 | } | |
6143 | ||
59a5e416 MF |
6144 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6145 | struct ocfs2_extent_block *eb) | |
6146 | { | |
6147 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6148 | le16_to_cpu(eb->h_suballoc_slot), | |
6149 | le64_to_cpu(eb->h_blkno), | |
6150 | le16_to_cpu(eb->h_suballoc_bit)); | |
6151 | } | |
6152 | ||
ccd979bd MF |
6153 | /* This function will figure out whether the currently last extent |
6154 | * block will be deleted, and if it will, what the new last extent | |
6155 | * block will be so we can update his h_next_leaf_blk field, as well | |
6156 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6157 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6158 | unsigned int clusters_to_del, |
dcd0538f | 6159 | struct ocfs2_path *path, |
ccd979bd MF |
6160 | struct buffer_head **new_last_eb) |
6161 | { | |
3a0782d0 | 6162 | int next_free, ret = 0; |
dcd0538f | 6163 | u32 cpos; |
3a0782d0 | 6164 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6165 | struct ocfs2_extent_block *eb; |
6166 | struct ocfs2_extent_list *el; | |
6167 | struct buffer_head *bh = NULL; | |
6168 | ||
6169 | *new_last_eb = NULL; | |
6170 | ||
ccd979bd | 6171 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6172 | if (!path->p_tree_depth) |
6173 | goto out; | |
ccd979bd MF |
6174 | |
6175 | /* trunc to zero special case - this makes tree_depth = 0 | |
6176 | * regardless of what it is. */ | |
3a0782d0 | 6177 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6178 | goto out; |
ccd979bd | 6179 | |
dcd0538f | 6180 | el = path_leaf_el(path); |
ccd979bd MF |
6181 | BUG_ON(!el->l_next_free_rec); |
6182 | ||
3a0782d0 MF |
6183 | /* |
6184 | * Make sure that this extent list will actually be empty | |
6185 | * after we clear away the data. We can shortcut out if | |
6186 | * there's more than one non-empty extent in the | |
6187 | * list. Otherwise, a check of the remaining extent is | |
6188 | * necessary. | |
6189 | */ | |
6190 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6191 | rec = NULL; | |
dcd0538f | 6192 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6193 | if (next_free > 2) |
dcd0538f | 6194 | goto out; |
3a0782d0 MF |
6195 | |
6196 | /* We may have a valid extent in index 1, check it. */ | |
6197 | if (next_free == 2) | |
6198 | rec = &el->l_recs[1]; | |
6199 | ||
6200 | /* | |
6201 | * Fall through - no more nonempty extents, so we want | |
6202 | * to delete this leaf. | |
6203 | */ | |
6204 | } else { | |
6205 | if (next_free > 1) | |
6206 | goto out; | |
6207 | ||
6208 | rec = &el->l_recs[0]; | |
6209 | } | |
6210 | ||
6211 | if (rec) { | |
6212 | /* | |
6213 | * Check it we'll only be trimming off the end of this | |
6214 | * cluster. | |
6215 | */ | |
e48edee2 | 6216 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6217 | goto out; |
6218 | } | |
ccd979bd | 6219 | |
dcd0538f MF |
6220 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6221 | if (ret) { | |
6222 | mlog_errno(ret); | |
6223 | goto out; | |
6224 | } | |
ccd979bd | 6225 | |
dcd0538f MF |
6226 | ret = ocfs2_find_leaf(inode, path_root_el(path), cpos, &bh); |
6227 | if (ret) { | |
6228 | mlog_errno(ret); | |
6229 | goto out; | |
6230 | } | |
ccd979bd | 6231 | |
dcd0538f MF |
6232 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6233 | el = &eb->h_list; | |
6234 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
6235 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
6236 | ret = -EROFS; | |
6237 | goto out; | |
6238 | } | |
ccd979bd MF |
6239 | |
6240 | *new_last_eb = bh; | |
6241 | get_bh(*new_last_eb); | |
dcd0538f MF |
6242 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6243 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6244 | out: | |
6245 | brelse(bh); | |
ccd979bd | 6246 | |
dcd0538f | 6247 | return ret; |
ccd979bd MF |
6248 | } |
6249 | ||
3a0782d0 MF |
6250 | /* |
6251 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6252 | * during truncate. | |
6253 | * | |
6254 | * The caller needs to: | |
6255 | * - start journaling of each path component. | |
6256 | * - compute and fully set up any new last ext block | |
6257 | */ | |
6258 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6259 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
6260 | u32 clusters_to_del, u64 *delete_start) | |
6261 | { | |
6262 | int ret, i, index = path->p_tree_depth; | |
6263 | u32 new_edge = 0; | |
6264 | u64 deleted_eb = 0; | |
6265 | struct buffer_head *bh; | |
6266 | struct ocfs2_extent_list *el; | |
6267 | struct ocfs2_extent_rec *rec; | |
6268 | ||
6269 | *delete_start = 0; | |
6270 | ||
6271 | while (index >= 0) { | |
6272 | bh = path->p_node[index].bh; | |
6273 | el = path->p_node[index].el; | |
6274 | ||
6275 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6276 | index, (unsigned long long)bh->b_blocknr); | |
6277 | ||
6278 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6279 | ||
6280 | if (index != | |
6281 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6282 | ocfs2_error(inode->i_sb, | |
6283 | "Inode %lu has invalid ext. block %llu", | |
6284 | inode->i_ino, | |
6285 | (unsigned long long)bh->b_blocknr); | |
6286 | ret = -EROFS; | |
6287 | goto out; | |
6288 | } | |
6289 | ||
6290 | find_tail_record: | |
6291 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6292 | rec = &el->l_recs[i]; | |
6293 | ||
6294 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6295 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6296 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6297 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6298 | le16_to_cpu(el->l_next_free_rec)); | |
6299 | ||
e48edee2 | 6300 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6301 | |
6302 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6303 | /* | |
6304 | * If the leaf block contains a single empty | |
6305 | * extent and no records, we can just remove | |
6306 | * the block. | |
6307 | */ | |
6308 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6309 | memset(rec, 0, | |
6310 | sizeof(struct ocfs2_extent_rec)); | |
6311 | el->l_next_free_rec = cpu_to_le16(0); | |
6312 | ||
6313 | goto delete; | |
6314 | } | |
6315 | ||
6316 | /* | |
6317 | * Remove any empty extents by shifting things | |
6318 | * left. That should make life much easier on | |
6319 | * the code below. This condition is rare | |
6320 | * enough that we shouldn't see a performance | |
6321 | * hit. | |
6322 | */ | |
6323 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6324 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6325 | ||
6326 | for(i = 0; | |
6327 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6328 | el->l_recs[i] = el->l_recs[i + 1]; | |
6329 | ||
6330 | memset(&el->l_recs[i], 0, | |
6331 | sizeof(struct ocfs2_extent_rec)); | |
6332 | ||
6333 | /* | |
6334 | * We've modified our extent list. The | |
6335 | * simplest way to handle this change | |
6336 | * is to being the search from the | |
6337 | * start again. | |
6338 | */ | |
6339 | goto find_tail_record; | |
6340 | } | |
6341 | ||
e48edee2 | 6342 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6343 | |
6344 | /* | |
6345 | * We'll use "new_edge" on our way back up the | |
6346 | * tree to know what our rightmost cpos is. | |
6347 | */ | |
e48edee2 | 6348 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6349 | new_edge += le32_to_cpu(rec->e_cpos); |
6350 | ||
6351 | /* | |
6352 | * The caller will use this to delete data blocks. | |
6353 | */ | |
6354 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6355 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6356 | le16_to_cpu(rec->e_leaf_clusters)); |
3a0782d0 MF |
6357 | |
6358 | /* | |
6359 | * If it's now empty, remove this record. | |
6360 | */ | |
e48edee2 | 6361 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6362 | memset(rec, 0, |
6363 | sizeof(struct ocfs2_extent_rec)); | |
6364 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6365 | } | |
6366 | } else { | |
6367 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6368 | memset(rec, 0, | |
6369 | sizeof(struct ocfs2_extent_rec)); | |
6370 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6371 | ||
6372 | goto delete; | |
6373 | } | |
6374 | ||
6375 | /* Can this actually happen? */ | |
6376 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6377 | goto delete; | |
6378 | ||
6379 | /* | |
6380 | * We never actually deleted any clusters | |
6381 | * because our leaf was empty. There's no | |
6382 | * reason to adjust the rightmost edge then. | |
6383 | */ | |
6384 | if (new_edge == 0) | |
6385 | goto delete; | |
6386 | ||
e48edee2 MF |
6387 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6388 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6389 | -le32_to_cpu(rec->e_cpos)); |
6390 | ||
6391 | /* | |
6392 | * A deleted child record should have been | |
6393 | * caught above. | |
6394 | */ | |
e48edee2 | 6395 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6396 | } |
6397 | ||
6398 | delete: | |
6399 | ret = ocfs2_journal_dirty(handle, bh); | |
6400 | if (ret) { | |
6401 | mlog_errno(ret); | |
6402 | goto out; | |
6403 | } | |
6404 | ||
6405 | mlog(0, "extent list container %llu, after: record %d: " | |
6406 | "(%u, %u, %llu), next = %u.\n", | |
6407 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6408 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6409 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6410 | le16_to_cpu(el->l_next_free_rec)); | |
6411 | ||
6412 | /* | |
6413 | * We must be careful to only attempt delete of an | |
6414 | * extent block (and not the root inode block). | |
6415 | */ | |
6416 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6417 | struct ocfs2_extent_block *eb = | |
6418 | (struct ocfs2_extent_block *)bh->b_data; | |
6419 | ||
6420 | /* | |
6421 | * Save this for use when processing the | |
6422 | * parent block. | |
6423 | */ | |
6424 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6425 | ||
6426 | mlog(0, "deleting this extent block.\n"); | |
6427 | ||
6428 | ocfs2_remove_from_cache(inode, bh); | |
6429 | ||
e48edee2 | 6430 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6431 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6432 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6433 | ||
59a5e416 MF |
6434 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6435 | /* An error here is not fatal. */ | |
6436 | if (ret < 0) | |
6437 | mlog_errno(ret); | |
3a0782d0 MF |
6438 | } else { |
6439 | deleted_eb = 0; | |
6440 | } | |
6441 | ||
6442 | index--; | |
6443 | } | |
6444 | ||
6445 | ret = 0; | |
6446 | out: | |
6447 | return ret; | |
6448 | } | |
6449 | ||
ccd979bd MF |
6450 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6451 | unsigned int clusters_to_del, | |
6452 | struct inode *inode, | |
6453 | struct buffer_head *fe_bh, | |
1fabe148 | 6454 | handle_t *handle, |
dcd0538f MF |
6455 | struct ocfs2_truncate_context *tc, |
6456 | struct ocfs2_path *path) | |
ccd979bd | 6457 | { |
3a0782d0 | 6458 | int status; |
ccd979bd | 6459 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6460 | struct ocfs2_extent_block *last_eb = NULL; |
6461 | struct ocfs2_extent_list *el; | |
ccd979bd | 6462 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
6463 | u64 delete_blk = 0; |
6464 | ||
6465 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6466 | ||
3a0782d0 | 6467 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6468 | path, &last_eb_bh); |
ccd979bd MF |
6469 | if (status < 0) { |
6470 | mlog_errno(status); | |
6471 | goto bail; | |
6472 | } | |
dcd0538f MF |
6473 | |
6474 | /* | |
6475 | * Each component will be touched, so we might as well journal | |
6476 | * here to avoid having to handle errors later. | |
6477 | */ | |
3a0782d0 MF |
6478 | status = ocfs2_journal_access_path(inode, handle, path); |
6479 | if (status < 0) { | |
6480 | mlog_errno(status); | |
6481 | goto bail; | |
dcd0538f MF |
6482 | } |
6483 | ||
6484 | if (last_eb_bh) { | |
6485 | status = ocfs2_journal_access(handle, inode, last_eb_bh, | |
6486 | OCFS2_JOURNAL_ACCESS_WRITE); | |
6487 | if (status < 0) { | |
6488 | mlog_errno(status); | |
6489 | goto bail; | |
6490 | } | |
6491 | ||
ccd979bd | 6492 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6493 | } |
ccd979bd | 6494 | |
dcd0538f MF |
6495 | el = &(fe->id2.i_list); |
6496 | ||
6497 | /* | |
6498 | * Lower levels depend on this never happening, but it's best | |
6499 | * to check it up here before changing the tree. | |
6500 | */ | |
e48edee2 | 6501 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6502 | ocfs2_error(inode->i_sb, |
6503 | "Inode %lu has an empty extent record, depth %u\n", | |
6504 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6505 | status = -EROFS; |
ccd979bd MF |
6506 | goto bail; |
6507 | } | |
ccd979bd MF |
6508 | |
6509 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
6510 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6511 | clusters_to_del; | |
6512 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6513 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6514 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6515 | |
3a0782d0 MF |
6516 | status = ocfs2_trim_tree(inode, path, handle, tc, |
6517 | clusters_to_del, &delete_blk); | |
6518 | if (status) { | |
6519 | mlog_errno(status); | |
6520 | goto bail; | |
ccd979bd MF |
6521 | } |
6522 | ||
dcd0538f | 6523 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6524 | /* trunc to zero is a special case. */ |
6525 | el->l_tree_depth = 0; | |
6526 | fe->i_last_eb_blk = 0; | |
6527 | } else if (last_eb) | |
6528 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6529 | ||
6530 | status = ocfs2_journal_dirty(handle, fe_bh); | |
6531 | if (status < 0) { | |
6532 | mlog_errno(status); | |
6533 | goto bail; | |
6534 | } | |
6535 | ||
6536 | if (last_eb) { | |
6537 | /* If there will be a new last extent block, then by | |
6538 | * definition, there cannot be any leaves to the right of | |
6539 | * him. */ | |
ccd979bd MF |
6540 | last_eb->h_next_leaf_blk = 0; |
6541 | status = ocfs2_journal_dirty(handle, last_eb_bh); | |
6542 | if (status < 0) { | |
6543 | mlog_errno(status); | |
6544 | goto bail; | |
6545 | } | |
6546 | } | |
6547 | ||
3a0782d0 MF |
6548 | if (delete_blk) { |
6549 | status = ocfs2_truncate_log_append(osb, handle, delete_blk, | |
6550 | clusters_to_del); | |
ccd979bd MF |
6551 | if (status < 0) { |
6552 | mlog_errno(status); | |
6553 | goto bail; | |
6554 | } | |
ccd979bd MF |
6555 | } |
6556 | status = 0; | |
6557 | bail: | |
dcd0538f | 6558 | |
ccd979bd MF |
6559 | mlog_exit(status); |
6560 | return status; | |
6561 | } | |
6562 | ||
2b4e30fb | 6563 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
6564 | { |
6565 | set_buffer_uptodate(bh); | |
6566 | mark_buffer_dirty(bh); | |
6567 | return 0; | |
6568 | } | |
6569 | ||
1d410a6e MF |
6570 | static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6571 | unsigned int from, unsigned int to, | |
6572 | struct page *page, int zero, u64 *phys) | |
6573 | { | |
6574 | int ret, partial = 0; | |
6575 | ||
6576 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6577 | if (ret) | |
6578 | mlog_errno(ret); | |
6579 | ||
6580 | if (zero) | |
eebd2aa3 | 6581 | zero_user_segment(page, from, to); |
1d410a6e MF |
6582 | |
6583 | /* | |
6584 | * Need to set the buffers we zero'd into uptodate | |
6585 | * here if they aren't - ocfs2_map_page_blocks() | |
6586 | * might've skipped some | |
6587 | */ | |
2b4e30fb JB |
6588 | ret = walk_page_buffers(handle, page_buffers(page), |
6589 | from, to, &partial, | |
6590 | ocfs2_zero_func); | |
6591 | if (ret < 0) | |
6592 | mlog_errno(ret); | |
6593 | else if (ocfs2_should_order_data(inode)) { | |
6594 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
6595 | #ifdef CONFIG_OCFS2_COMPAT_JBD | |
1d410a6e MF |
6596 | ret = walk_page_buffers(handle, page_buffers(page), |
6597 | from, to, &partial, | |
2b4e30fb JB |
6598 | ocfs2_journal_dirty_data); |
6599 | #endif | |
1d410a6e MF |
6600 | if (ret < 0) |
6601 | mlog_errno(ret); | |
6602 | } | |
6603 | ||
6604 | if (!partial) | |
6605 | SetPageUptodate(page); | |
6606 | ||
6607 | flush_dcache_page(page); | |
6608 | } | |
6609 | ||
35edec1d MF |
6610 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6611 | loff_t end, struct page **pages, | |
6612 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6613 | { |
1d410a6e | 6614 | int i; |
60b11392 MF |
6615 | struct page *page; |
6616 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6617 | struct super_block *sb = inode->i_sb; | |
6618 | ||
6619 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6620 | ||
6621 | if (numpages == 0) | |
6622 | goto out; | |
6623 | ||
35edec1d | 6624 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6625 | for(i = 0; i < numpages; i++) { |
6626 | page = pages[i]; | |
6627 | ||
35edec1d MF |
6628 | from = start & (PAGE_CACHE_SIZE - 1); |
6629 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6630 | to = end & (PAGE_CACHE_SIZE - 1); | |
6631 | ||
60b11392 MF |
6632 | BUG_ON(from > PAGE_CACHE_SIZE); |
6633 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6634 | ||
1d410a6e MF |
6635 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6636 | &phys); | |
60b11392 | 6637 | |
35edec1d | 6638 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6639 | } |
6640 | out: | |
1d410a6e MF |
6641 | if (pages) |
6642 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6643 | } |
6644 | ||
35edec1d | 6645 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
1d410a6e | 6646 | struct page **pages, int *num) |
60b11392 | 6647 | { |
1d410a6e | 6648 | int numpages, ret = 0; |
60b11392 MF |
6649 | struct super_block *sb = inode->i_sb; |
6650 | struct address_space *mapping = inode->i_mapping; | |
6651 | unsigned long index; | |
35edec1d | 6652 | loff_t last_page_bytes; |
60b11392 | 6653 | |
35edec1d | 6654 | BUG_ON(start > end); |
60b11392 | 6655 | |
35edec1d MF |
6656 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != |
6657 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6658 | ||
1d410a6e | 6659 | numpages = 0; |
35edec1d MF |
6660 | last_page_bytes = PAGE_ALIGN(end); |
6661 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
6662 | do { |
6663 | pages[numpages] = grab_cache_page(mapping, index); | |
6664 | if (!pages[numpages]) { | |
6665 | ret = -ENOMEM; | |
6666 | mlog_errno(ret); | |
6667 | goto out; | |
6668 | } | |
6669 | ||
6670 | numpages++; | |
6671 | index++; | |
35edec1d | 6672 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6673 | |
6674 | out: | |
6675 | if (ret != 0) { | |
1d410a6e MF |
6676 | if (pages) |
6677 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6678 | numpages = 0; |
6679 | } | |
6680 | ||
6681 | *num = numpages; | |
6682 | ||
6683 | return ret; | |
6684 | } | |
6685 | ||
6686 | /* | |
6687 | * Zero the area past i_size but still within an allocated | |
6688 | * cluster. This avoids exposing nonzero data on subsequent file | |
6689 | * extends. | |
6690 | * | |
6691 | * We need to call this before i_size is updated on the inode because | |
6692 | * otherwise block_write_full_page() will skip writeout of pages past | |
6693 | * i_size. The new_i_size parameter is passed for this reason. | |
6694 | */ | |
35edec1d MF |
6695 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6696 | u64 range_start, u64 range_end) | |
60b11392 | 6697 | { |
1d410a6e | 6698 | int ret = 0, numpages; |
60b11392 MF |
6699 | struct page **pages = NULL; |
6700 | u64 phys; | |
1d410a6e MF |
6701 | unsigned int ext_flags; |
6702 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6703 | |
6704 | /* | |
6705 | * File systems which don't support sparse files zero on every | |
6706 | * extend. | |
6707 | */ | |
1d410a6e | 6708 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
6709 | return 0; |
6710 | ||
1d410a6e | 6711 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
6712 | sizeof(struct page *), GFP_NOFS); |
6713 | if (pages == NULL) { | |
6714 | ret = -ENOMEM; | |
6715 | mlog_errno(ret); | |
6716 | goto out; | |
6717 | } | |
6718 | ||
1d410a6e MF |
6719 | if (range_start == range_end) |
6720 | goto out; | |
6721 | ||
6722 | ret = ocfs2_extent_map_get_blocks(inode, | |
6723 | range_start >> sb->s_blocksize_bits, | |
6724 | &phys, NULL, &ext_flags); | |
60b11392 MF |
6725 | if (ret) { |
6726 | mlog_errno(ret); | |
6727 | goto out; | |
6728 | } | |
6729 | ||
1d410a6e MF |
6730 | /* |
6731 | * Tail is a hole, or is marked unwritten. In either case, we | |
6732 | * can count on read and write to return/push zero's. | |
6733 | */ | |
6734 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
6735 | goto out; |
6736 | ||
1d410a6e MF |
6737 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
6738 | &numpages); | |
6739 | if (ret) { | |
6740 | mlog_errno(ret); | |
6741 | goto out; | |
6742 | } | |
6743 | ||
35edec1d MF |
6744 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
6745 | numpages, phys, handle); | |
60b11392 MF |
6746 | |
6747 | /* | |
6748 | * Initiate writeout of the pages we zero'd here. We don't | |
6749 | * wait on them - the truncate_inode_pages() call later will | |
6750 | * do that for us. | |
6751 | */ | |
35edec1d MF |
6752 | ret = do_sync_mapping_range(inode->i_mapping, range_start, |
6753 | range_end - 1, SYNC_FILE_RANGE_WRITE); | |
60b11392 MF |
6754 | if (ret) |
6755 | mlog_errno(ret); | |
6756 | ||
6757 | out: | |
6758 | if (pages) | |
6759 | kfree(pages); | |
6760 | ||
6761 | return ret; | |
6762 | } | |
6763 | ||
fdd77704 TY |
6764 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
6765 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
6766 | { |
6767 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 6768 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 6769 | |
fdd77704 TY |
6770 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
6771 | memset(&di->id2, 0, blocksize - | |
6772 | offsetof(struct ocfs2_dinode, id2) - | |
6773 | xattrsize); | |
6774 | else | |
6775 | memset(&di->id2, 0, blocksize - | |
6776 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
6777 | } |
6778 | ||
5b6a3a2b MF |
6779 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
6780 | struct ocfs2_dinode *di) | |
6781 | { | |
fdd77704 | 6782 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
6783 | di->id2.i_list.l_tree_depth = 0; |
6784 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
6785 | di->id2.i_list.l_count = cpu_to_le16( |
6786 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
6787 | } |
6788 | ||
1afc32b9 MF |
6789 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
6790 | { | |
6791 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6792 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
6793 | ||
6794 | spin_lock(&oi->ip_lock); | |
6795 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
6796 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6797 | spin_unlock(&oi->ip_lock); | |
6798 | ||
6799 | /* | |
6800 | * We clear the entire i_data structure here so that all | |
6801 | * fields can be properly initialized. | |
6802 | */ | |
fdd77704 | 6803 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 6804 | |
fdd77704 TY |
6805 | idata->id_count = cpu_to_le16( |
6806 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
6807 | } |
6808 | ||
6809 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
6810 | struct buffer_head *di_bh) | |
6811 | { | |
6812 | int ret, i, has_data, num_pages = 0; | |
6813 | handle_t *handle; | |
6814 | u64 uninitialized_var(block); | |
6815 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6816 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
6817 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
6818 | struct ocfs2_alloc_context *data_ac = NULL; |
6819 | struct page **pages = NULL; | |
6820 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 6821 | struct ocfs2_extent_tree et; |
1afc32b9 MF |
6822 | |
6823 | has_data = i_size_read(inode) ? 1 : 0; | |
6824 | ||
6825 | if (has_data) { | |
6826 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
6827 | sizeof(struct page *), GFP_NOFS); | |
6828 | if (pages == NULL) { | |
6829 | ret = -ENOMEM; | |
6830 | mlog_errno(ret); | |
6831 | goto out; | |
6832 | } | |
6833 | ||
6834 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
6835 | if (ret) { | |
6836 | mlog_errno(ret); | |
6837 | goto out; | |
6838 | } | |
6839 | } | |
6840 | ||
6841 | handle = ocfs2_start_trans(osb, OCFS2_INLINE_TO_EXTENTS_CREDITS); | |
6842 | if (IS_ERR(handle)) { | |
6843 | ret = PTR_ERR(handle); | |
6844 | mlog_errno(ret); | |
6845 | goto out_unlock; | |
6846 | } | |
6847 | ||
6848 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
6849 | OCFS2_JOURNAL_ACCESS_WRITE); | |
6850 | if (ret) { | |
6851 | mlog_errno(ret); | |
6852 | goto out_commit; | |
6853 | } | |
6854 | ||
6855 | if (has_data) { | |
6856 | u32 bit_off, num; | |
6857 | unsigned int page_end; | |
6858 | u64 phys; | |
6859 | ||
6860 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, | |
6861 | &num); | |
6862 | if (ret) { | |
6863 | mlog_errno(ret); | |
6864 | goto out_commit; | |
6865 | } | |
6866 | ||
6867 | /* | |
6868 | * Save two copies, one for insert, and one that can | |
6869 | * be changed by ocfs2_map_and_dirty_page() below. | |
6870 | */ | |
6871 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
6872 | ||
6873 | /* | |
6874 | * Non sparse file systems zero on extend, so no need | |
6875 | * to do that now. | |
6876 | */ | |
6877 | if (!ocfs2_sparse_alloc(osb) && | |
6878 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
6879 | end = PAGE_CACHE_SIZE; | |
6880 | ||
6881 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
6882 | if (ret) { | |
6883 | mlog_errno(ret); | |
6884 | goto out_commit; | |
6885 | } | |
6886 | ||
6887 | /* | |
6888 | * This should populate the 1st page for us and mark | |
6889 | * it up to date. | |
6890 | */ | |
6891 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
6892 | if (ret) { | |
6893 | mlog_errno(ret); | |
6894 | goto out_commit; | |
6895 | } | |
6896 | ||
6897 | page_end = PAGE_CACHE_SIZE; | |
6898 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
6899 | page_end = osb->s_clustersize; | |
6900 | ||
6901 | for (i = 0; i < num_pages; i++) | |
6902 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
6903 | pages[i], i > 0, &phys); | |
6904 | } | |
6905 | ||
6906 | spin_lock(&oi->ip_lock); | |
6907 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
6908 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6909 | spin_unlock(&oi->ip_lock); | |
6910 | ||
5b6a3a2b | 6911 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
6912 | |
6913 | ocfs2_journal_dirty(handle, di_bh); | |
6914 | ||
6915 | if (has_data) { | |
6916 | /* | |
6917 | * An error at this point should be extremely rare. If | |
6918 | * this proves to be false, we could always re-build | |
6919 | * the in-inode data from our pages. | |
6920 | */ | |
8d6220d6 | 6921 | ocfs2_init_dinode_extent_tree(&et, inode, di_bh); |
f99b9b7c JB |
6922 | ret = ocfs2_insert_extent(osb, handle, inode, &et, |
6923 | 0, block, 1, 0, NULL); | |
1afc32b9 MF |
6924 | if (ret) { |
6925 | mlog_errno(ret); | |
6926 | goto out_commit; | |
6927 | } | |
6928 | ||
6929 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
6930 | } | |
6931 | ||
6932 | out_commit: | |
6933 | ocfs2_commit_trans(osb, handle); | |
6934 | ||
6935 | out_unlock: | |
6936 | if (data_ac) | |
6937 | ocfs2_free_alloc_context(data_ac); | |
6938 | ||
6939 | out: | |
6940 | if (pages) { | |
6941 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
6942 | kfree(pages); | |
6943 | } | |
6944 | ||
6945 | return ret; | |
6946 | } | |
6947 | ||
ccd979bd MF |
6948 | /* |
6949 | * It is expected, that by the time you call this function, | |
6950 | * inode->i_size and fe->i_size have been adjusted. | |
6951 | * | |
6952 | * WARNING: This will kfree the truncate context | |
6953 | */ | |
6954 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
6955 | struct inode *inode, | |
6956 | struct buffer_head *fe_bh, | |
6957 | struct ocfs2_truncate_context *tc) | |
6958 | { | |
6959 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 6960 | u32 clusters_to_del, new_highest_cpos, range; |
ccd979bd | 6961 | struct ocfs2_extent_list *el; |
1fabe148 | 6962 | handle_t *handle = NULL; |
ccd979bd | 6963 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 6964 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 6965 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
ccd979bd MF |
6966 | |
6967 | mlog_entry_void(); | |
6968 | ||
dcd0538f | 6969 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
6970 | i_size_read(inode)); |
6971 | ||
e7d4cb6b | 6972 | path = ocfs2_new_path(fe_bh, &di->id2.i_list); |
dcd0538f MF |
6973 | if (!path) { |
6974 | status = -ENOMEM; | |
6975 | mlog_errno(status); | |
6976 | goto bail; | |
6977 | } | |
83418978 MF |
6978 | |
6979 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
6980 | ||
ccd979bd | 6981 | start: |
3a0782d0 MF |
6982 | /* |
6983 | * Check that we still have allocation to delete. | |
6984 | */ | |
6985 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
6986 | status = 0; | |
6987 | goto bail; | |
6988 | } | |
6989 | ||
dcd0538f MF |
6990 | /* |
6991 | * Truncate always works against the rightmost tree branch. | |
6992 | */ | |
6993 | status = ocfs2_find_path(inode, path, UINT_MAX); | |
6994 | if (status) { | |
6995 | mlog_errno(status); | |
6996 | goto bail; | |
ccd979bd MF |
6997 | } |
6998 | ||
dcd0538f MF |
6999 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
7000 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
7001 | ||
7002 | /* | |
7003 | * By now, el will point to the extent list on the bottom most | |
7004 | * portion of this tree. Only the tail record is considered in | |
7005 | * each pass. | |
7006 | * | |
7007 | * We handle the following cases, in order: | |
7008 | * - empty extent: delete the remaining branch | |
7009 | * - remove the entire record | |
7010 | * - remove a partial record | |
7011 | * - no record needs to be removed (truncate has completed) | |
7012 | */ | |
7013 | el = path_leaf_el(path); | |
3a0782d0 MF |
7014 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7015 | ocfs2_error(inode->i_sb, | |
7016 | "Inode %llu has empty extent block at %llu\n", | |
7017 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7018 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7019 | status = -EROFS; | |
7020 | goto bail; | |
7021 | } | |
7022 | ||
ccd979bd | 7023 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 7024 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 7025 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
7026 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
7027 | clusters_to_del = 0; | |
7028 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 7029 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f | 7030 | } else if (range > new_highest_cpos) { |
e48edee2 | 7031 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 7032 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f MF |
7033 | new_highest_cpos; |
7034 | } else { | |
7035 | status = 0; | |
7036 | goto bail; | |
7037 | } | |
ccd979bd | 7038 | |
dcd0538f MF |
7039 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
7040 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7041 | ||
1b1dcc1b | 7042 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
7043 | tl_sem = 1; |
7044 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
7045 | * record is free for use. If there isn't any, we flush to get | |
7046 | * an empty truncate log. */ | |
7047 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
7048 | status = __ocfs2_flush_truncate_log(osb); | |
7049 | if (status < 0) { | |
7050 | mlog_errno(status); | |
7051 | goto bail; | |
7052 | } | |
7053 | } | |
7054 | ||
7055 | credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, | |
dcd0538f MF |
7056 | (struct ocfs2_dinode *)fe_bh->b_data, |
7057 | el); | |
65eff9cc | 7058 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
7059 | if (IS_ERR(handle)) { |
7060 | status = PTR_ERR(handle); | |
7061 | handle = NULL; | |
7062 | mlog_errno(status); | |
7063 | goto bail; | |
7064 | } | |
7065 | ||
dcd0538f MF |
7066 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
7067 | tc, path); | |
ccd979bd MF |
7068 | if (status < 0) { |
7069 | mlog_errno(status); | |
7070 | goto bail; | |
7071 | } | |
7072 | ||
1b1dcc1b | 7073 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7074 | tl_sem = 0; |
7075 | ||
02dc1af4 | 7076 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
7077 | handle = NULL; |
7078 | ||
dcd0538f MF |
7079 | ocfs2_reinit_path(path, 1); |
7080 | ||
7081 | /* | |
3a0782d0 MF |
7082 | * The check above will catch the case where we've truncated |
7083 | * away all allocation. | |
dcd0538f | 7084 | */ |
3a0782d0 MF |
7085 | goto start; |
7086 | ||
ccd979bd | 7087 | bail: |
ccd979bd MF |
7088 | |
7089 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7090 | ||
7091 | if (tl_sem) | |
1b1dcc1b | 7092 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7093 | |
7094 | if (handle) | |
02dc1af4 | 7095 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 7096 | |
59a5e416 MF |
7097 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
7098 | ||
dcd0538f | 7099 | ocfs2_free_path(path); |
ccd979bd MF |
7100 | |
7101 | /* This will drop the ext_alloc cluster lock for us */ | |
7102 | ocfs2_free_truncate_context(tc); | |
7103 | ||
7104 | mlog_exit(status); | |
7105 | return status; | |
7106 | } | |
7107 | ||
ccd979bd | 7108 | /* |
59a5e416 | 7109 | * Expects the inode to already be locked. |
ccd979bd MF |
7110 | */ |
7111 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7112 | struct inode *inode, | |
7113 | struct buffer_head *fe_bh, | |
7114 | struct ocfs2_truncate_context **tc) | |
7115 | { | |
59a5e416 | 7116 | int status; |
ccd979bd MF |
7117 | unsigned int new_i_clusters; |
7118 | struct ocfs2_dinode *fe; | |
7119 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7120 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7121 | |
7122 | mlog_entry_void(); | |
7123 | ||
7124 | *tc = NULL; | |
7125 | ||
7126 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7127 | i_size_read(inode)); | |
7128 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7129 | ||
7130 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7131 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7132 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7133 | |
cd861280 | 7134 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7135 | if (!(*tc)) { |
7136 | status = -ENOMEM; | |
7137 | mlog_errno(status); | |
7138 | goto bail; | |
7139 | } | |
59a5e416 | 7140 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7141 | |
ccd979bd | 7142 | if (fe->id2.i_list.l_tree_depth) { |
31d33073 | 7143 | status = ocfs2_read_block(inode, le64_to_cpu(fe->i_last_eb_blk), |
0fcaa56a | 7144 | &last_eb_bh); |
ccd979bd MF |
7145 | if (status < 0) { |
7146 | mlog_errno(status); | |
7147 | goto bail; | |
7148 | } | |
7149 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
7150 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
7151 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
7152 | ||
7153 | brelse(last_eb_bh); | |
7154 | status = -EIO; | |
7155 | goto bail; | |
7156 | } | |
ccd979bd MF |
7157 | } |
7158 | ||
7159 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7160 | ||
ccd979bd MF |
7161 | status = 0; |
7162 | bail: | |
7163 | if (status < 0) { | |
7164 | if (*tc) | |
7165 | ocfs2_free_truncate_context(*tc); | |
7166 | *tc = NULL; | |
7167 | } | |
7168 | mlog_exit_void(); | |
7169 | return status; | |
7170 | } | |
7171 | ||
1afc32b9 MF |
7172 | /* |
7173 | * 'start' is inclusive, 'end' is not. | |
7174 | */ | |
7175 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7176 | unsigned int start, unsigned int end, int trunc) | |
7177 | { | |
7178 | int ret; | |
7179 | unsigned int numbytes; | |
7180 | handle_t *handle; | |
7181 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7182 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7183 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7184 | ||
7185 | if (end > i_size_read(inode)) | |
7186 | end = i_size_read(inode); | |
7187 | ||
7188 | BUG_ON(start >= end); | |
7189 | ||
7190 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7191 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7192 | !ocfs2_supports_inline_data(osb)) { | |
7193 | ocfs2_error(inode->i_sb, | |
7194 | "Inline data flags for inode %llu don't agree! " | |
7195 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7196 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7197 | le16_to_cpu(di->i_dyn_features), | |
7198 | OCFS2_I(inode)->ip_dyn_features, | |
7199 | osb->s_feature_incompat); | |
7200 | ret = -EROFS; | |
7201 | goto out; | |
7202 | } | |
7203 | ||
7204 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7205 | if (IS_ERR(handle)) { | |
7206 | ret = PTR_ERR(handle); | |
7207 | mlog_errno(ret); | |
7208 | goto out; | |
7209 | } | |
7210 | ||
7211 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
7212 | OCFS2_JOURNAL_ACCESS_WRITE); | |
7213 | if (ret) { | |
7214 | mlog_errno(ret); | |
7215 | goto out_commit; | |
7216 | } | |
7217 | ||
7218 | numbytes = end - start; | |
7219 | memset(idata->id_data + start, 0, numbytes); | |
7220 | ||
7221 | /* | |
7222 | * No need to worry about the data page here - it's been | |
7223 | * truncated already and inline data doesn't need it for | |
7224 | * pushing zero's to disk, so we'll let readpage pick it up | |
7225 | * later. | |
7226 | */ | |
7227 | if (trunc) { | |
7228 | i_size_write(inode, start); | |
7229 | di->i_size = cpu_to_le64(start); | |
7230 | } | |
7231 | ||
7232 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7233 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7234 | ||
7235 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7236 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7237 | ||
7238 | ocfs2_journal_dirty(handle, di_bh); | |
7239 | ||
7240 | out_commit: | |
7241 | ocfs2_commit_trans(osb, handle); | |
7242 | ||
7243 | out: | |
7244 | return ret; | |
7245 | } | |
7246 | ||
ccd979bd MF |
7247 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7248 | { | |
59a5e416 MF |
7249 | /* |
7250 | * The caller is responsible for completing deallocation | |
7251 | * before freeing the context. | |
7252 | */ | |
7253 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7254 | mlog(ML_NOTICE, | |
7255 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd | 7256 | |
a81cb88b | 7257 | brelse(tc->tc_last_eb_bh); |
ccd979bd MF |
7258 | |
7259 | kfree(tc); | |
7260 | } |