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 TM |
52 | /* |
53 | * ocfs2_extent_tree and ocfs2_extent_tree_operations are used to abstract | |
54 | * the b-tree operations in ocfs2. Now all the b-tree operations are not | |
55 | * limited to ocfs2_dinode only. Any data which need to allocate clusters | |
56 | * to store can use b-tree. And it only needs to implement its ocfs2_extent_tree | |
57 | * and operation. | |
58 | * | |
59 | * ocfs2_extent_tree contains info for the root of the b-tree, it must have a | |
60 | * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree | |
61 | * functions. | |
62 | * ocfs2_extent_tree_operations abstract the normal operations we do for | |
63 | * the root of extent b-tree. | |
64 | */ | |
65 | struct ocfs2_extent_tree; | |
66 | ||
67 | struct ocfs2_extent_tree_operations { | |
35dc0aa3 JB |
68 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
69 | u64 blkno); | |
70 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
71 | void (*eo_update_clusters)(struct inode *inode, | |
72 | struct ocfs2_extent_tree *et, | |
73 | u32 new_clusters); | |
74 | int (*eo_sanity_check)(struct inode *inode, struct ocfs2_extent_tree *et); | |
e7d4cb6b TM |
75 | }; |
76 | ||
77 | struct ocfs2_extent_tree { | |
78 | enum ocfs2_extent_tree_type type; | |
79 | struct ocfs2_extent_tree_operations *eops; | |
80 | struct buffer_head *root_bh; | |
81 | struct ocfs2_extent_list *root_el; | |
f56654c4 | 82 | void *private; |
ca12b7c4 | 83 | unsigned int max_leaf_clusters; |
e7d4cb6b TM |
84 | }; |
85 | ||
86 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
87 | u64 blkno) | |
88 | { | |
89 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)et->root_bh->b_data; | |
90 | ||
91 | BUG_ON(et->type != OCFS2_DINODE_EXTENT); | |
92 | di->i_last_eb_blk = cpu_to_le64(blkno); | |
93 | } | |
94 | ||
95 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
96 | { | |
97 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)et->root_bh->b_data; | |
98 | ||
99 | BUG_ON(et->type != OCFS2_DINODE_EXTENT); | |
100 | return le64_to_cpu(di->i_last_eb_blk); | |
101 | } | |
102 | ||
103 | static void ocfs2_dinode_update_clusters(struct inode *inode, | |
104 | struct ocfs2_extent_tree *et, | |
105 | u32 clusters) | |
106 | { | |
107 | struct ocfs2_dinode *di = | |
108 | (struct ocfs2_dinode *)et->root_bh->b_data; | |
109 | ||
110 | le32_add_cpu(&di->i_clusters, clusters); | |
111 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
112 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters); | |
113 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
114 | } | |
115 | ||
116 | static int ocfs2_dinode_sanity_check(struct inode *inode, | |
117 | struct ocfs2_extent_tree *et) | |
118 | { | |
119 | int ret = 0; | |
120 | struct ocfs2_dinode *di; | |
121 | ||
122 | BUG_ON(et->type != OCFS2_DINODE_EXTENT); | |
123 | ||
124 | di = (struct ocfs2_dinode *)et->root_bh->b_data; | |
125 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
126 | ret = -EIO; | |
127 | ocfs2_error(inode->i_sb, | |
128 | "Inode %llu has invalid path root", | |
129 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
130 | } | |
131 | ||
132 | return ret; | |
133 | } | |
134 | ||
135 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
35dc0aa3 JB |
136 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, |
137 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
138 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
139 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
e7d4cb6b TM |
140 | }; |
141 | ||
f56654c4 TM |
142 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
143 | u64 blkno) | |
144 | { | |
145 | struct ocfs2_xattr_value_root *xv = | |
146 | (struct ocfs2_xattr_value_root *)et->private; | |
147 | ||
148 | xv->xr_last_eb_blk = cpu_to_le64(blkno); | |
149 | } | |
150 | ||
151 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
152 | { | |
153 | struct ocfs2_xattr_value_root *xv = | |
154 | (struct ocfs2_xattr_value_root *) et->private; | |
155 | ||
156 | return le64_to_cpu(xv->xr_last_eb_blk); | |
157 | } | |
158 | ||
159 | static void ocfs2_xattr_value_update_clusters(struct inode *inode, | |
160 | struct ocfs2_extent_tree *et, | |
161 | u32 clusters) | |
162 | { | |
163 | struct ocfs2_xattr_value_root *xv = | |
164 | (struct ocfs2_xattr_value_root *)et->private; | |
165 | ||
166 | le32_add_cpu(&xv->xr_clusters, clusters); | |
167 | } | |
168 | ||
169 | static int ocfs2_xattr_value_sanity_check(struct inode *inode, | |
170 | struct ocfs2_extent_tree *et) | |
171 | { | |
172 | return 0; | |
173 | } | |
174 | ||
175 | static struct ocfs2_extent_tree_operations ocfs2_xattr_et_ops = { | |
35dc0aa3 JB |
176 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
177 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
178 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
179 | .eo_sanity_check = ocfs2_xattr_value_sanity_check, | |
f56654c4 TM |
180 | }; |
181 | ||
ba492615 TM |
182 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
183 | u64 blkno) | |
184 | { | |
185 | struct ocfs2_xattr_block *xb = | |
186 | (struct ocfs2_xattr_block *) et->root_bh->b_data; | |
187 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; | |
188 | ||
189 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
190 | } | |
191 | ||
192 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
193 | { | |
194 | struct ocfs2_xattr_block *xb = | |
195 | (struct ocfs2_xattr_block *) et->root_bh->b_data; | |
196 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; | |
197 | ||
198 | return le64_to_cpu(xt->xt_last_eb_blk); | |
199 | } | |
200 | ||
201 | static void ocfs2_xattr_tree_update_clusters(struct inode *inode, | |
202 | struct ocfs2_extent_tree *et, | |
203 | u32 clusters) | |
204 | { | |
205 | struct ocfs2_xattr_block *xb = | |
206 | (struct ocfs2_xattr_block *)et->root_bh->b_data; | |
207 | ||
208 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
209 | } | |
210 | ||
211 | static int ocfs2_xattr_tree_sanity_check(struct inode *inode, | |
212 | struct ocfs2_extent_tree *et) | |
213 | { | |
214 | return 0; | |
215 | } | |
216 | ||
217 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { | |
35dc0aa3 JB |
218 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
219 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
220 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
221 | .eo_sanity_check = ocfs2_xattr_tree_sanity_check, | |
ba492615 TM |
222 | }; |
223 | ||
e7d4cb6b | 224 | static struct ocfs2_extent_tree* |
ca12b7c4 TM |
225 | ocfs2_new_extent_tree(struct inode *inode, |
226 | struct buffer_head *bh, | |
f56654c4 TM |
227 | enum ocfs2_extent_tree_type et_type, |
228 | void *private) | |
e7d4cb6b TM |
229 | { |
230 | struct ocfs2_extent_tree *et; | |
231 | ||
232 | et = kzalloc(sizeof(*et), GFP_NOFS); | |
233 | if (!et) | |
234 | return NULL; | |
235 | ||
236 | et->type = et_type; | |
237 | get_bh(bh); | |
238 | et->root_bh = bh; | |
f56654c4 | 239 | et->private = private; |
e7d4cb6b | 240 | |
e7d4cb6b TM |
241 | if (et_type == OCFS2_DINODE_EXTENT) { |
242 | et->root_el = &((struct ocfs2_dinode *)bh->b_data)->id2.i_list; | |
243 | et->eops = &ocfs2_dinode_et_ops; | |
f56654c4 TM |
244 | } else if (et_type == OCFS2_XATTR_VALUE_EXTENT) { |
245 | struct ocfs2_xattr_value_root *xv = | |
246 | (struct ocfs2_xattr_value_root *) private; | |
247 | et->root_el = &xv->xr_list; | |
248 | et->eops = &ocfs2_xattr_et_ops; | |
ba492615 TM |
249 | } else if (et_type == OCFS2_XATTR_TREE_EXTENT) { |
250 | struct ocfs2_xattr_block *xb = | |
251 | (struct ocfs2_xattr_block *)bh->b_data; | |
252 | et->root_el = &xb->xb_attrs.xb_root.xt_list; | |
253 | et->eops = &ocfs2_xattr_tree_et_ops; | |
ca12b7c4 TM |
254 | et->max_leaf_clusters = ocfs2_clusters_for_bytes(inode->i_sb, |
255 | OCFS2_MAX_XATTR_TREE_LEAF_SIZE); | |
e7d4cb6b TM |
256 | } |
257 | ||
258 | return et; | |
259 | } | |
260 | ||
261 | static void ocfs2_free_extent_tree(struct ocfs2_extent_tree *et) | |
262 | { | |
263 | if (et) { | |
264 | brelse(et->root_bh); | |
265 | kfree(et); | |
266 | } | |
267 | } | |
268 | ||
35dc0aa3 JB |
269 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
270 | u64 new_last_eb_blk) | |
e7d4cb6b | 271 | { |
35dc0aa3 | 272 | et->eops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
273 | } |
274 | ||
35dc0aa3 | 275 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 276 | { |
35dc0aa3 | 277 | return et->eops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
278 | } |
279 | ||
35dc0aa3 JB |
280 | static inline void ocfs2_et_update_clusters(struct inode *inode, |
281 | struct ocfs2_extent_tree *et, | |
282 | u32 clusters) | |
283 | { | |
284 | et->eops->eo_update_clusters(inode, et, clusters); | |
285 | } | |
286 | ||
287 | static inline int ocfs2_et_sanity_check(struct inode *inode, | |
288 | struct ocfs2_extent_tree *et) | |
e7d4cb6b | 289 | { |
35dc0aa3 | 290 | return et->eops->eo_sanity_check(inode, et); |
e7d4cb6b TM |
291 | } |
292 | ||
dcd0538f | 293 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
294 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
295 | struct ocfs2_extent_block *eb); | |
ccd979bd | 296 | |
dcd0538f MF |
297 | /* |
298 | * Structures which describe a path through a btree, and functions to | |
299 | * manipulate them. | |
300 | * | |
301 | * The idea here is to be as generic as possible with the tree | |
302 | * manipulation code. | |
303 | */ | |
304 | struct ocfs2_path_item { | |
305 | struct buffer_head *bh; | |
306 | struct ocfs2_extent_list *el; | |
307 | }; | |
ccd979bd | 308 | |
dcd0538f | 309 | #define OCFS2_MAX_PATH_DEPTH 5 |
ccd979bd | 310 | |
dcd0538f MF |
311 | struct ocfs2_path { |
312 | int p_tree_depth; | |
313 | struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH]; | |
314 | }; | |
ccd979bd | 315 | |
dcd0538f MF |
316 | #define path_root_bh(_path) ((_path)->p_node[0].bh) |
317 | #define path_root_el(_path) ((_path)->p_node[0].el) | |
318 | #define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh) | |
319 | #define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el) | |
320 | #define path_num_items(_path) ((_path)->p_tree_depth + 1) | |
ccd979bd | 321 | |
dcd0538f MF |
322 | /* |
323 | * Reset the actual path elements so that we can re-use the structure | |
324 | * to build another path. Generally, this involves freeing the buffer | |
325 | * heads. | |
326 | */ | |
327 | static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) | |
328 | { | |
329 | int i, start = 0, depth = 0; | |
330 | struct ocfs2_path_item *node; | |
ccd979bd | 331 | |
dcd0538f MF |
332 | if (keep_root) |
333 | start = 1; | |
ccd979bd | 334 | |
dcd0538f MF |
335 | for(i = start; i < path_num_items(path); i++) { |
336 | node = &path->p_node[i]; | |
337 | ||
338 | brelse(node->bh); | |
339 | node->bh = NULL; | |
340 | node->el = NULL; | |
341 | } | |
342 | ||
343 | /* | |
344 | * Tree depth may change during truncate, or insert. If we're | |
345 | * keeping the root extent list, then make sure that our path | |
346 | * structure reflects the proper depth. | |
347 | */ | |
348 | if (keep_root) | |
349 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
350 | ||
351 | path->p_tree_depth = depth; | |
352 | } | |
353 | ||
354 | static void ocfs2_free_path(struct ocfs2_path *path) | |
355 | { | |
356 | if (path) { | |
357 | ocfs2_reinit_path(path, 0); | |
358 | kfree(path); | |
359 | } | |
360 | } | |
361 | ||
328d5752 MF |
362 | /* |
363 | * All the elements of src into dest. After this call, src could be freed | |
364 | * without affecting dest. | |
365 | * | |
366 | * Both paths should have the same root. Any non-root elements of dest | |
367 | * will be freed. | |
368 | */ | |
369 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
370 | { | |
371 | int i; | |
372 | ||
373 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
374 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
375 | ||
376 | ocfs2_reinit_path(dest, 1); | |
377 | ||
378 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
379 | dest->p_node[i].bh = src->p_node[i].bh; | |
380 | dest->p_node[i].el = src->p_node[i].el; | |
381 | ||
382 | if (dest->p_node[i].bh) | |
383 | get_bh(dest->p_node[i].bh); | |
384 | } | |
385 | } | |
386 | ||
dcd0538f MF |
387 | /* |
388 | * Make the *dest path the same as src and re-initialize src path to | |
389 | * have a root only. | |
390 | */ | |
391 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
392 | { | |
393 | int i; | |
394 | ||
395 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
396 | ||
397 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
398 | brelse(dest->p_node[i].bh); | |
399 | ||
400 | dest->p_node[i].bh = src->p_node[i].bh; | |
401 | dest->p_node[i].el = src->p_node[i].el; | |
402 | ||
403 | src->p_node[i].bh = NULL; | |
404 | src->p_node[i].el = NULL; | |
405 | } | |
406 | } | |
407 | ||
408 | /* | |
409 | * Insert an extent block at given index. | |
410 | * | |
411 | * This will not take an additional reference on eb_bh. | |
412 | */ | |
413 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
414 | struct buffer_head *eb_bh) | |
415 | { | |
416 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
417 | ||
418 | /* | |
419 | * Right now, no root bh is an extent block, so this helps | |
420 | * catch code errors with dinode trees. The assertion can be | |
421 | * safely removed if we ever need to insert extent block | |
422 | * structures at the root. | |
423 | */ | |
424 | BUG_ON(index == 0); | |
425 | ||
426 | path->p_node[index].bh = eb_bh; | |
427 | path->p_node[index].el = &eb->h_list; | |
428 | } | |
429 | ||
430 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
431 | struct ocfs2_extent_list *root_el) | |
432 | { | |
433 | struct ocfs2_path *path; | |
ccd979bd | 434 | |
dcd0538f MF |
435 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
436 | ||
437 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
438 | if (path) { | |
439 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
440 | get_bh(root_bh); | |
441 | path_root_bh(path) = root_bh; | |
442 | path_root_el(path) = root_el; | |
443 | } | |
444 | ||
445 | return path; | |
446 | } | |
447 | ||
dcd0538f MF |
448 | /* |
449 | * Convenience function to journal all components in a path. | |
450 | */ | |
451 | static int ocfs2_journal_access_path(struct inode *inode, handle_t *handle, | |
452 | struct ocfs2_path *path) | |
453 | { | |
454 | int i, ret = 0; | |
455 | ||
456 | if (!path) | |
457 | goto out; | |
458 | ||
459 | for(i = 0; i < path_num_items(path); i++) { | |
460 | ret = ocfs2_journal_access(handle, inode, path->p_node[i].bh, | |
461 | OCFS2_JOURNAL_ACCESS_WRITE); | |
462 | if (ret < 0) { | |
463 | mlog_errno(ret); | |
464 | goto out; | |
465 | } | |
466 | } | |
467 | ||
468 | out: | |
469 | return ret; | |
470 | } | |
471 | ||
328d5752 MF |
472 | /* |
473 | * Return the index of the extent record which contains cluster #v_cluster. | |
474 | * -1 is returned if it was not found. | |
475 | * | |
476 | * Should work fine on interior and exterior nodes. | |
477 | */ | |
478 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
479 | { | |
480 | int ret = -1; | |
481 | int i; | |
482 | struct ocfs2_extent_rec *rec; | |
483 | u32 rec_end, rec_start, clusters; | |
484 | ||
485 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
486 | rec = &el->l_recs[i]; | |
487 | ||
488 | rec_start = le32_to_cpu(rec->e_cpos); | |
489 | clusters = ocfs2_rec_clusters(el, rec); | |
490 | ||
491 | rec_end = rec_start + clusters; | |
492 | ||
493 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
494 | ret = i; | |
495 | break; | |
496 | } | |
497 | } | |
498 | ||
499 | return ret; | |
500 | } | |
501 | ||
dcd0538f MF |
502 | enum ocfs2_contig_type { |
503 | CONTIG_NONE = 0, | |
504 | CONTIG_LEFT, | |
328d5752 MF |
505 | CONTIG_RIGHT, |
506 | CONTIG_LEFTRIGHT, | |
dcd0538f MF |
507 | }; |
508 | ||
e48edee2 MF |
509 | |
510 | /* | |
511 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
512 | * ocfs2_extent_contig only work properly against leaf nodes! | |
513 | */ | |
dcd0538f MF |
514 | static int ocfs2_block_extent_contig(struct super_block *sb, |
515 | struct ocfs2_extent_rec *ext, | |
516 | u64 blkno) | |
ccd979bd | 517 | { |
e48edee2 MF |
518 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
519 | ||
520 | blk_end += ocfs2_clusters_to_blocks(sb, | |
521 | le16_to_cpu(ext->e_leaf_clusters)); | |
522 | ||
523 | return blkno == blk_end; | |
ccd979bd MF |
524 | } |
525 | ||
dcd0538f MF |
526 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
527 | struct ocfs2_extent_rec *right) | |
528 | { | |
e48edee2 MF |
529 | u32 left_range; |
530 | ||
531 | left_range = le32_to_cpu(left->e_cpos) + | |
532 | le16_to_cpu(left->e_leaf_clusters); | |
533 | ||
534 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
535 | } |
536 | ||
537 | static enum ocfs2_contig_type | |
538 | ocfs2_extent_contig(struct inode *inode, | |
539 | struct ocfs2_extent_rec *ext, | |
540 | struct ocfs2_extent_rec *insert_rec) | |
541 | { | |
542 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
543 | ||
328d5752 MF |
544 | /* |
545 | * Refuse to coalesce extent records with different flag | |
546 | * fields - we don't want to mix unwritten extents with user | |
547 | * data. | |
548 | */ | |
549 | if (ext->e_flags != insert_rec->e_flags) | |
550 | return CONTIG_NONE; | |
551 | ||
dcd0538f MF |
552 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
553 | ocfs2_block_extent_contig(inode->i_sb, ext, blkno)) | |
554 | return CONTIG_RIGHT; | |
555 | ||
556 | blkno = le64_to_cpu(ext->e_blkno); | |
557 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
558 | ocfs2_block_extent_contig(inode->i_sb, insert_rec, blkno)) | |
559 | return CONTIG_LEFT; | |
560 | ||
561 | return CONTIG_NONE; | |
562 | } | |
563 | ||
564 | /* | |
565 | * NOTE: We can have pretty much any combination of contiguousness and | |
566 | * appending. | |
567 | * | |
568 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
569 | * we'll have to update the path to that leaf. | |
570 | */ | |
571 | enum ocfs2_append_type { | |
572 | APPEND_NONE = 0, | |
573 | APPEND_TAIL, | |
574 | }; | |
575 | ||
328d5752 MF |
576 | enum ocfs2_split_type { |
577 | SPLIT_NONE = 0, | |
578 | SPLIT_LEFT, | |
579 | SPLIT_RIGHT, | |
580 | }; | |
581 | ||
dcd0538f | 582 | struct ocfs2_insert_type { |
328d5752 | 583 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
584 | enum ocfs2_append_type ins_appending; |
585 | enum ocfs2_contig_type ins_contig; | |
586 | int ins_contig_index; | |
dcd0538f MF |
587 | int ins_tree_depth; |
588 | }; | |
589 | ||
328d5752 MF |
590 | struct ocfs2_merge_ctxt { |
591 | enum ocfs2_contig_type c_contig_type; | |
592 | int c_has_empty_extent; | |
593 | int c_split_covers_rec; | |
328d5752 MF |
594 | }; |
595 | ||
ccd979bd MF |
596 | /* |
597 | * How many free extents have we got before we need more meta data? | |
598 | */ | |
599 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
600 | struct inode *inode, | |
e7d4cb6b | 601 | struct buffer_head *root_bh, |
f56654c4 TM |
602 | enum ocfs2_extent_tree_type type, |
603 | void *private) | |
ccd979bd MF |
604 | { |
605 | int retval; | |
e7d4cb6b | 606 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
607 | struct ocfs2_extent_block *eb; |
608 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 609 | u64 last_eb_blk = 0; |
ccd979bd MF |
610 | |
611 | mlog_entry_void(); | |
612 | ||
e7d4cb6b TM |
613 | if (type == OCFS2_DINODE_EXTENT) { |
614 | struct ocfs2_dinode *fe = | |
615 | (struct ocfs2_dinode *)root_bh->b_data; | |
616 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
617 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
618 | retval = -EIO; | |
619 | goto bail; | |
620 | } | |
621 | ||
622 | if (fe->i_last_eb_blk) | |
623 | last_eb_blk = le64_to_cpu(fe->i_last_eb_blk); | |
624 | el = &fe->id2.i_list; | |
f56654c4 TM |
625 | } else if (type == OCFS2_XATTR_VALUE_EXTENT) { |
626 | struct ocfs2_xattr_value_root *xv = | |
627 | (struct ocfs2_xattr_value_root *) private; | |
628 | ||
629 | last_eb_blk = le64_to_cpu(xv->xr_last_eb_blk); | |
630 | el = &xv->xr_list; | |
ba492615 TM |
631 | } else if (type == OCFS2_XATTR_TREE_EXTENT) { |
632 | struct ocfs2_xattr_block *xb = | |
633 | (struct ocfs2_xattr_block *)root_bh->b_data; | |
634 | ||
635 | last_eb_blk = le64_to_cpu(xb->xb_attrs.xb_root.xt_last_eb_blk); | |
636 | el = &xb->xb_attrs.xb_root.xt_list; | |
ccd979bd MF |
637 | } |
638 | ||
e7d4cb6b TM |
639 | if (last_eb_blk) { |
640 | retval = ocfs2_read_block(osb, last_eb_blk, | |
ccd979bd MF |
641 | &eb_bh, OCFS2_BH_CACHED, inode); |
642 | if (retval < 0) { | |
643 | mlog_errno(retval); | |
644 | goto bail; | |
645 | } | |
646 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
647 | el = &eb->h_list; | |
e7d4cb6b | 648 | } |
ccd979bd MF |
649 | |
650 | BUG_ON(el->l_tree_depth != 0); | |
651 | ||
652 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
653 | bail: | |
654 | if (eb_bh) | |
655 | brelse(eb_bh); | |
656 | ||
657 | mlog_exit(retval); | |
658 | return retval; | |
659 | } | |
660 | ||
661 | /* expects array to already be allocated | |
662 | * | |
663 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
664 | * l_count for you | |
665 | */ | |
666 | static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb, | |
1fabe148 | 667 | handle_t *handle, |
ccd979bd MF |
668 | struct inode *inode, |
669 | int wanted, | |
670 | struct ocfs2_alloc_context *meta_ac, | |
671 | struct buffer_head *bhs[]) | |
672 | { | |
673 | int count, status, i; | |
674 | u16 suballoc_bit_start; | |
675 | u32 num_got; | |
676 | u64 first_blkno; | |
677 | struct ocfs2_extent_block *eb; | |
678 | ||
679 | mlog_entry_void(); | |
680 | ||
681 | count = 0; | |
682 | while (count < wanted) { | |
683 | status = ocfs2_claim_metadata(osb, | |
684 | handle, | |
685 | meta_ac, | |
686 | wanted - count, | |
687 | &suballoc_bit_start, | |
688 | &num_got, | |
689 | &first_blkno); | |
690 | if (status < 0) { | |
691 | mlog_errno(status); | |
692 | goto bail; | |
693 | } | |
694 | ||
695 | for(i = count; i < (num_got + count); i++) { | |
696 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
697 | if (bhs[i] == NULL) { | |
698 | status = -EIO; | |
699 | mlog_errno(status); | |
700 | goto bail; | |
701 | } | |
702 | ocfs2_set_new_buffer_uptodate(inode, bhs[i]); | |
703 | ||
704 | status = ocfs2_journal_access(handle, inode, bhs[i], | |
705 | OCFS2_JOURNAL_ACCESS_CREATE); | |
706 | if (status < 0) { | |
707 | mlog_errno(status); | |
708 | goto bail; | |
709 | } | |
710 | ||
711 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
712 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
713 | /* Ok, setup the minimal stuff here. */ | |
714 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
715 | eb->h_blkno = cpu_to_le64(first_blkno); | |
716 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
ccd979bd | 717 | eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); |
ccd979bd MF |
718 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
719 | eb->h_list.l_count = | |
720 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
721 | ||
722 | suballoc_bit_start++; | |
723 | first_blkno++; | |
724 | ||
725 | /* We'll also be dirtied by the caller, so | |
726 | * this isn't absolutely necessary. */ | |
727 | status = ocfs2_journal_dirty(handle, bhs[i]); | |
728 | if (status < 0) { | |
729 | mlog_errno(status); | |
730 | goto bail; | |
731 | } | |
732 | } | |
733 | ||
734 | count += num_got; | |
735 | } | |
736 | ||
737 | status = 0; | |
738 | bail: | |
739 | if (status < 0) { | |
740 | for(i = 0; i < wanted; i++) { | |
741 | if (bhs[i]) | |
742 | brelse(bhs[i]); | |
743 | bhs[i] = NULL; | |
744 | } | |
745 | } | |
746 | mlog_exit(status); | |
747 | return status; | |
748 | } | |
749 | ||
dcd0538f MF |
750 | /* |
751 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
752 | * | |
753 | * Returns the sum of the rightmost extent rec logical offset and | |
754 | * cluster count. | |
755 | * | |
756 | * ocfs2_add_branch() uses this to determine what logical cluster | |
757 | * value should be populated into the leftmost new branch records. | |
758 | * | |
759 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
760 | * value for the new topmost tree record. | |
761 | */ | |
762 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
763 | { | |
764 | int i; | |
765 | ||
766 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
767 | ||
768 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 769 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
770 | } |
771 | ||
ccd979bd MF |
772 | /* |
773 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
774 | * to start at, if we don't want to start the branch at the dinode | |
775 | * structure. | |
776 | * | |
777 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
778 | * for the new last extent block. | |
779 | * | |
780 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 781 | * contain a single record with cluster count == 0. |
ccd979bd MF |
782 | */ |
783 | static int ocfs2_add_branch(struct ocfs2_super *osb, | |
1fabe148 | 784 | handle_t *handle, |
ccd979bd | 785 | struct inode *inode, |
e7d4cb6b | 786 | struct ocfs2_extent_tree *et, |
ccd979bd | 787 | struct buffer_head *eb_bh, |
328d5752 | 788 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
789 | struct ocfs2_alloc_context *meta_ac) |
790 | { | |
791 | int status, new_blocks, i; | |
792 | u64 next_blkno, new_last_eb_blk; | |
793 | struct buffer_head *bh; | |
794 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
795 | struct ocfs2_extent_block *eb; |
796 | struct ocfs2_extent_list *eb_el; | |
797 | struct ocfs2_extent_list *el; | |
dcd0538f | 798 | u32 new_cpos; |
ccd979bd MF |
799 | |
800 | mlog_entry_void(); | |
801 | ||
328d5752 | 802 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 803 | |
ccd979bd MF |
804 | if (eb_bh) { |
805 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
806 | el = &eb->h_list; | |
807 | } else | |
e7d4cb6b | 808 | el = et->root_el; |
ccd979bd MF |
809 | |
810 | /* we never add a branch to a leaf. */ | |
811 | BUG_ON(!el->l_tree_depth); | |
812 | ||
813 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
814 | ||
815 | /* allocate the number of new eb blocks we need */ | |
816 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
817 | GFP_KERNEL); | |
818 | if (!new_eb_bhs) { | |
819 | status = -ENOMEM; | |
820 | mlog_errno(status); | |
821 | goto bail; | |
822 | } | |
823 | ||
824 | status = ocfs2_create_new_meta_bhs(osb, handle, inode, new_blocks, | |
825 | meta_ac, new_eb_bhs); | |
826 | if (status < 0) { | |
827 | mlog_errno(status); | |
828 | goto bail; | |
829 | } | |
830 | ||
328d5752 | 831 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
dcd0538f MF |
832 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); |
833 | ||
ccd979bd MF |
834 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be |
835 | * linked with the rest of the tree. | |
836 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
837 | * | |
838 | * when we leave the loop, new_last_eb_blk will point to the | |
839 | * newest leaf, and next_blkno will point to the topmost extent | |
840 | * block. */ | |
841 | next_blkno = new_last_eb_blk = 0; | |
842 | for(i = 0; i < new_blocks; i++) { | |
843 | bh = new_eb_bhs[i]; | |
844 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
845 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
846 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
847 | status = -EIO; | |
848 | goto bail; | |
849 | } | |
850 | eb_el = &eb->h_list; | |
851 | ||
852 | status = ocfs2_journal_access(handle, inode, bh, | |
853 | OCFS2_JOURNAL_ACCESS_CREATE); | |
854 | if (status < 0) { | |
855 | mlog_errno(status); | |
856 | goto bail; | |
857 | } | |
858 | ||
859 | eb->h_next_leaf_blk = 0; | |
860 | eb_el->l_tree_depth = cpu_to_le16(i); | |
861 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
862 | /* |
863 | * This actually counts as an empty extent as | |
864 | * c_clusters == 0 | |
865 | */ | |
866 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 867 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
868 | /* |
869 | * eb_el isn't always an interior node, but even leaf | |
870 | * nodes want a zero'd flags and reserved field so | |
871 | * this gets the whole 32 bits regardless of use. | |
872 | */ | |
873 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
874 | if (!eb_el->l_tree_depth) |
875 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
876 | ||
877 | status = ocfs2_journal_dirty(handle, bh); | |
878 | if (status < 0) { | |
879 | mlog_errno(status); | |
880 | goto bail; | |
881 | } | |
882 | ||
883 | next_blkno = le64_to_cpu(eb->h_blkno); | |
884 | } | |
885 | ||
886 | /* This is a bit hairy. We want to update up to three blocks | |
887 | * here without leaving any of them in an inconsistent state | |
888 | * in case of error. We don't have to worry about | |
889 | * journal_dirty erroring as it won't unless we've aborted the | |
890 | * handle (in which case we would never be here) so reserving | |
891 | * the write with journal_access is all we need to do. */ | |
328d5752 | 892 | status = ocfs2_journal_access(handle, inode, *last_eb_bh, |
ccd979bd MF |
893 | OCFS2_JOURNAL_ACCESS_WRITE); |
894 | if (status < 0) { | |
895 | mlog_errno(status); | |
896 | goto bail; | |
897 | } | |
e7d4cb6b | 898 | status = ocfs2_journal_access(handle, inode, et->root_bh, |
ccd979bd MF |
899 | OCFS2_JOURNAL_ACCESS_WRITE); |
900 | if (status < 0) { | |
901 | mlog_errno(status); | |
902 | goto bail; | |
903 | } | |
904 | if (eb_bh) { | |
905 | status = ocfs2_journal_access(handle, inode, eb_bh, | |
906 | OCFS2_JOURNAL_ACCESS_WRITE); | |
907 | if (status < 0) { | |
908 | mlog_errno(status); | |
909 | goto bail; | |
910 | } | |
911 | } | |
912 | ||
913 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 914 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
915 | i = le16_to_cpu(el->l_next_free_rec); |
916 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 917 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 918 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
919 | le16_add_cpu(&el->l_next_free_rec, 1); |
920 | ||
921 | /* fe needs a new last extent block pointer, as does the | |
922 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 923 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 924 | |
328d5752 | 925 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
926 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
927 | ||
328d5752 | 928 | status = ocfs2_journal_dirty(handle, *last_eb_bh); |
ccd979bd MF |
929 | if (status < 0) |
930 | mlog_errno(status); | |
e7d4cb6b | 931 | status = ocfs2_journal_dirty(handle, et->root_bh); |
ccd979bd MF |
932 | if (status < 0) |
933 | mlog_errno(status); | |
934 | if (eb_bh) { | |
935 | status = ocfs2_journal_dirty(handle, eb_bh); | |
936 | if (status < 0) | |
937 | mlog_errno(status); | |
938 | } | |
939 | ||
328d5752 MF |
940 | /* |
941 | * Some callers want to track the rightmost leaf so pass it | |
942 | * back here. | |
943 | */ | |
944 | brelse(*last_eb_bh); | |
945 | get_bh(new_eb_bhs[0]); | |
946 | *last_eb_bh = new_eb_bhs[0]; | |
947 | ||
ccd979bd MF |
948 | status = 0; |
949 | bail: | |
950 | if (new_eb_bhs) { | |
951 | for (i = 0; i < new_blocks; i++) | |
952 | if (new_eb_bhs[i]) | |
953 | brelse(new_eb_bhs[i]); | |
954 | kfree(new_eb_bhs); | |
955 | } | |
956 | ||
957 | mlog_exit(status); | |
958 | return status; | |
959 | } | |
960 | ||
961 | /* | |
962 | * adds another level to the allocation tree. | |
963 | * returns back the new extent block so you can add a branch to it | |
964 | * after this call. | |
965 | */ | |
966 | static int ocfs2_shift_tree_depth(struct ocfs2_super *osb, | |
1fabe148 | 967 | handle_t *handle, |
ccd979bd | 968 | struct inode *inode, |
e7d4cb6b | 969 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
970 | struct ocfs2_alloc_context *meta_ac, |
971 | struct buffer_head **ret_new_eb_bh) | |
972 | { | |
973 | int status, i; | |
dcd0538f | 974 | u32 new_clusters; |
ccd979bd | 975 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 976 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 977 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
978 | struct ocfs2_extent_list *eb_el; |
979 | ||
980 | mlog_entry_void(); | |
981 | ||
982 | status = ocfs2_create_new_meta_bhs(osb, handle, inode, 1, meta_ac, | |
983 | &new_eb_bh); | |
984 | if (status < 0) { | |
985 | mlog_errno(status); | |
986 | goto bail; | |
987 | } | |
988 | ||
989 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
990 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
991 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
992 | status = -EIO; | |
993 | goto bail; | |
994 | } | |
995 | ||
996 | eb_el = &eb->h_list; | |
e7d4cb6b | 997 | root_el = et->root_el; |
ccd979bd MF |
998 | |
999 | status = ocfs2_journal_access(handle, inode, new_eb_bh, | |
1000 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1001 | if (status < 0) { | |
1002 | mlog_errno(status); | |
1003 | goto bail; | |
1004 | } | |
1005 | ||
e7d4cb6b TM |
1006 | /* copy the root extent list data into the new extent block */ |
1007 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1008 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1009 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1010 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd MF |
1011 | |
1012 | status = ocfs2_journal_dirty(handle, new_eb_bh); | |
1013 | if (status < 0) { | |
1014 | mlog_errno(status); | |
1015 | goto bail; | |
1016 | } | |
1017 | ||
e7d4cb6b | 1018 | status = ocfs2_journal_access(handle, inode, et->root_bh, |
ccd979bd MF |
1019 | OCFS2_JOURNAL_ACCESS_WRITE); |
1020 | if (status < 0) { | |
1021 | mlog_errno(status); | |
1022 | goto bail; | |
1023 | } | |
1024 | ||
dcd0538f MF |
1025 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1026 | ||
e7d4cb6b TM |
1027 | /* update root_bh now */ |
1028 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1029 | root_el->l_recs[0].e_cpos = 0; | |
1030 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1031 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1032 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1033 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1034 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1035 | |
1036 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1037 | * becomes the allocated extent block */ | |
e7d4cb6b | 1038 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1039 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1040 | |
e7d4cb6b | 1041 | status = ocfs2_journal_dirty(handle, et->root_bh); |
ccd979bd MF |
1042 | if (status < 0) { |
1043 | mlog_errno(status); | |
1044 | goto bail; | |
1045 | } | |
1046 | ||
1047 | *ret_new_eb_bh = new_eb_bh; | |
1048 | new_eb_bh = NULL; | |
1049 | status = 0; | |
1050 | bail: | |
1051 | if (new_eb_bh) | |
1052 | brelse(new_eb_bh); | |
1053 | ||
1054 | mlog_exit(status); | |
1055 | return status; | |
1056 | } | |
1057 | ||
ccd979bd MF |
1058 | /* |
1059 | * Should only be called when there is no space left in any of the | |
1060 | * leaf nodes. What we want to do is find the lowest tree depth | |
1061 | * non-leaf extent block with room for new records. There are three | |
1062 | * valid results of this search: | |
1063 | * | |
1064 | * 1) a lowest extent block is found, then we pass it back in | |
1065 | * *lowest_eb_bh and return '0' | |
1066 | * | |
e7d4cb6b | 1067 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1068 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1069 | * | |
e7d4cb6b | 1070 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1071 | * which case we return > 0 |
1072 | * | |
1073 | * return status < 0 indicates an error. | |
1074 | */ | |
1075 | static int ocfs2_find_branch_target(struct ocfs2_super *osb, | |
1076 | struct inode *inode, | |
e7d4cb6b | 1077 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1078 | struct buffer_head **target_bh) |
1079 | { | |
1080 | int status = 0, i; | |
1081 | u64 blkno; | |
ccd979bd MF |
1082 | struct ocfs2_extent_block *eb; |
1083 | struct ocfs2_extent_list *el; | |
1084 | struct buffer_head *bh = NULL; | |
1085 | struct buffer_head *lowest_bh = NULL; | |
1086 | ||
1087 | mlog_entry_void(); | |
1088 | ||
1089 | *target_bh = NULL; | |
1090 | ||
e7d4cb6b | 1091 | el = et->root_el; |
ccd979bd MF |
1092 | |
1093 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1094 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
b0697053 | 1095 | ocfs2_error(inode->i_sb, "Dinode %llu has empty " |
ccd979bd | 1096 | "extent list (next_free_rec == 0)", |
b0697053 | 1097 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1098 | status = -EIO; |
1099 | goto bail; | |
1100 | } | |
1101 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1102 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1103 | if (!blkno) { | |
b0697053 | 1104 | ocfs2_error(inode->i_sb, "Dinode %llu has extent " |
ccd979bd MF |
1105 | "list where extent # %d has no physical " |
1106 | "block start", | |
b0697053 | 1107 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i); |
ccd979bd MF |
1108 | status = -EIO; |
1109 | goto bail; | |
1110 | } | |
1111 | ||
1112 | if (bh) { | |
1113 | brelse(bh); | |
1114 | bh = NULL; | |
1115 | } | |
1116 | ||
1117 | status = ocfs2_read_block(osb, blkno, &bh, OCFS2_BH_CACHED, | |
1118 | inode); | |
1119 | if (status < 0) { | |
1120 | mlog_errno(status); | |
1121 | goto bail; | |
1122 | } | |
dcd0538f MF |
1123 | |
1124 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1125 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1126 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1127 | status = -EIO; | |
1128 | goto bail; | |
1129 | } | |
1130 | el = &eb->h_list; | |
1131 | ||
1132 | if (le16_to_cpu(el->l_next_free_rec) < | |
1133 | le16_to_cpu(el->l_count)) { | |
1134 | if (lowest_bh) | |
1135 | brelse(lowest_bh); | |
1136 | lowest_bh = bh; | |
1137 | get_bh(lowest_bh); | |
1138 | } | |
1139 | } | |
1140 | ||
1141 | /* If we didn't find one and the fe doesn't have any room, | |
1142 | * then return '1' */ | |
e7d4cb6b TM |
1143 | el = et->root_el; |
1144 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) | |
dcd0538f MF |
1145 | status = 1; |
1146 | ||
1147 | *target_bh = lowest_bh; | |
1148 | bail: | |
1149 | if (bh) | |
1150 | brelse(bh); | |
1151 | ||
1152 | mlog_exit(status); | |
1153 | return status; | |
1154 | } | |
1155 | ||
c3afcbb3 MF |
1156 | /* |
1157 | * Grow a b-tree so that it has more records. | |
1158 | * | |
1159 | * We might shift the tree depth in which case existing paths should | |
1160 | * be considered invalid. | |
1161 | * | |
1162 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1163 | * |
1164 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 MF |
1165 | */ |
1166 | static int ocfs2_grow_tree(struct inode *inode, handle_t *handle, | |
e7d4cb6b | 1167 | struct ocfs2_extent_tree *et, int *final_depth, |
328d5752 | 1168 | struct buffer_head **last_eb_bh, |
c3afcbb3 MF |
1169 | struct ocfs2_alloc_context *meta_ac) |
1170 | { | |
1171 | int ret, shift; | |
e7d4cb6b TM |
1172 | struct ocfs2_extent_list *el = et->root_el; |
1173 | int depth = le16_to_cpu(el->l_tree_depth); | |
c3afcbb3 MF |
1174 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1175 | struct buffer_head *bh = NULL; | |
1176 | ||
1177 | BUG_ON(meta_ac == NULL); | |
1178 | ||
e7d4cb6b | 1179 | shift = ocfs2_find_branch_target(osb, inode, et, &bh); |
c3afcbb3 MF |
1180 | if (shift < 0) { |
1181 | ret = shift; | |
1182 | mlog_errno(ret); | |
1183 | goto out; | |
1184 | } | |
1185 | ||
1186 | /* We traveled all the way to the bottom of the allocation tree | |
1187 | * and didn't find room for any more extents - we need to add | |
1188 | * another tree level */ | |
1189 | if (shift) { | |
1190 | BUG_ON(bh); | |
1191 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1192 | ||
1193 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1194 | * the new extent block (so we can pass that to | |
1195 | * ocfs2_add_branch). */ | |
e7d4cb6b | 1196 | ret = ocfs2_shift_tree_depth(osb, handle, inode, et, |
c3afcbb3 MF |
1197 | meta_ac, &bh); |
1198 | if (ret < 0) { | |
1199 | mlog_errno(ret); | |
1200 | goto out; | |
1201 | } | |
1202 | depth++; | |
328d5752 MF |
1203 | if (depth == 1) { |
1204 | /* | |
1205 | * Special case: we have room now if we shifted from | |
1206 | * tree_depth 0, so no more work needs to be done. | |
1207 | * | |
1208 | * We won't be calling add_branch, so pass | |
1209 | * back *last_eb_bh as the new leaf. At depth | |
1210 | * zero, it should always be null so there's | |
1211 | * no reason to brelse. | |
1212 | */ | |
1213 | BUG_ON(*last_eb_bh); | |
1214 | get_bh(bh); | |
1215 | *last_eb_bh = bh; | |
c3afcbb3 | 1216 | goto out; |
328d5752 | 1217 | } |
c3afcbb3 MF |
1218 | } |
1219 | ||
1220 | /* call ocfs2_add_branch to add the final part of the tree with | |
1221 | * the new data. */ | |
1222 | mlog(0, "add branch. bh = %p\n", bh); | |
e7d4cb6b | 1223 | ret = ocfs2_add_branch(osb, handle, inode, et, bh, last_eb_bh, |
c3afcbb3 MF |
1224 | meta_ac); |
1225 | if (ret < 0) { | |
1226 | mlog_errno(ret); | |
1227 | goto out; | |
1228 | } | |
1229 | ||
1230 | out: | |
1231 | if (final_depth) | |
1232 | *final_depth = depth; | |
1233 | brelse(bh); | |
1234 | return ret; | |
1235 | } | |
1236 | ||
dcd0538f MF |
1237 | /* |
1238 | * This function will discard the rightmost extent record. | |
1239 | */ | |
1240 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1241 | { | |
1242 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1243 | int count = le16_to_cpu(el->l_count); | |
1244 | unsigned int num_bytes; | |
1245 | ||
1246 | BUG_ON(!next_free); | |
1247 | /* This will cause us to go off the end of our extent list. */ | |
1248 | BUG_ON(next_free >= count); | |
1249 | ||
1250 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1251 | ||
1252 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1253 | } | |
1254 | ||
1255 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1256 | struct ocfs2_extent_rec *insert_rec) | |
1257 | { | |
1258 | int i, insert_index, next_free, has_empty, num_bytes; | |
1259 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1260 | struct ocfs2_extent_rec *rec; | |
1261 | ||
1262 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1263 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1264 | ||
1265 | BUG_ON(!next_free); | |
1266 | ||
1267 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1268 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1269 | |
1270 | /* | |
1271 | * The easiest way to approach this is to just remove the | |
1272 | * empty extent and temporarily decrement next_free. | |
1273 | */ | |
1274 | if (has_empty) { | |
1275 | /* | |
1276 | * If next_free was 1 (only an empty extent), this | |
1277 | * loop won't execute, which is fine. We still want | |
1278 | * the decrement above to happen. | |
1279 | */ | |
1280 | for(i = 0; i < (next_free - 1); i++) | |
1281 | el->l_recs[i] = el->l_recs[i+1]; | |
1282 | ||
1283 | next_free--; | |
1284 | } | |
1285 | ||
1286 | /* | |
1287 | * Figure out what the new record index should be. | |
1288 | */ | |
1289 | for(i = 0; i < next_free; i++) { | |
1290 | rec = &el->l_recs[i]; | |
1291 | ||
1292 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1293 | break; | |
1294 | } | |
1295 | insert_index = i; | |
1296 | ||
1297 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1298 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1299 | ||
1300 | BUG_ON(insert_index < 0); | |
1301 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1302 | BUG_ON(insert_index > next_free); | |
1303 | ||
1304 | /* | |
1305 | * No need to memmove if we're just adding to the tail. | |
1306 | */ | |
1307 | if (insert_index != next_free) { | |
1308 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1309 | ||
1310 | num_bytes = next_free - insert_index; | |
1311 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1312 | memmove(&el->l_recs[insert_index + 1], | |
1313 | &el->l_recs[insert_index], | |
1314 | num_bytes); | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Either we had an empty extent, and need to re-increment or | |
1319 | * there was no empty extent on a non full rightmost leaf node, | |
1320 | * in which case we still need to increment. | |
1321 | */ | |
1322 | next_free++; | |
1323 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1324 | /* | |
1325 | * Make sure none of the math above just messed up our tree. | |
1326 | */ | |
1327 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1328 | ||
1329 | el->l_recs[insert_index] = *insert_rec; | |
1330 | ||
1331 | } | |
1332 | ||
328d5752 MF |
1333 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1334 | { | |
1335 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1336 | ||
1337 | BUG_ON(num_recs == 0); | |
1338 | ||
1339 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1340 | num_recs--; | |
1341 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1342 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1343 | memset(&el->l_recs[num_recs], 0, | |
1344 | sizeof(struct ocfs2_extent_rec)); | |
1345 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1346 | } | |
1347 | } | |
1348 | ||
dcd0538f MF |
1349 | /* |
1350 | * Create an empty extent record . | |
1351 | * | |
1352 | * l_next_free_rec may be updated. | |
1353 | * | |
1354 | * If an empty extent already exists do nothing. | |
1355 | */ | |
1356 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1357 | { | |
1358 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1359 | ||
e48edee2 MF |
1360 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1361 | ||
dcd0538f MF |
1362 | if (next_free == 0) |
1363 | goto set_and_inc; | |
1364 | ||
1365 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1366 | return; | |
1367 | ||
1368 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1369 | "Asked to create an empty extent in a full list:\n" | |
1370 | "count = %u, tree depth = %u", | |
1371 | le16_to_cpu(el->l_count), | |
1372 | le16_to_cpu(el->l_tree_depth)); | |
1373 | ||
1374 | ocfs2_shift_records_right(el); | |
1375 | ||
1376 | set_and_inc: | |
1377 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1378 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1379 | } | |
1380 | ||
1381 | /* | |
1382 | * For a rotation which involves two leaf nodes, the "root node" is | |
1383 | * the lowest level tree node which contains a path to both leafs. This | |
1384 | * resulting set of information can be used to form a complete "subtree" | |
1385 | * | |
1386 | * This function is passed two full paths from the dinode down to a | |
1387 | * pair of adjacent leaves. It's task is to figure out which path | |
1388 | * index contains the subtree root - this can be the root index itself | |
1389 | * in a worst-case rotation. | |
1390 | * | |
1391 | * The array index of the subtree root is passed back. | |
1392 | */ | |
1393 | static int ocfs2_find_subtree_root(struct inode *inode, | |
1394 | struct ocfs2_path *left, | |
1395 | struct ocfs2_path *right) | |
1396 | { | |
1397 | int i = 0; | |
1398 | ||
1399 | /* | |
1400 | * Check that the caller passed in two paths from the same tree. | |
1401 | */ | |
1402 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1403 | ||
1404 | do { | |
1405 | i++; | |
1406 | ||
1407 | /* | |
1408 | * The caller didn't pass two adjacent paths. | |
1409 | */ | |
1410 | mlog_bug_on_msg(i > left->p_tree_depth, | |
1411 | "Inode %lu, left depth %u, right depth %u\n" | |
1412 | "left leaf blk %llu, right leaf blk %llu\n", | |
1413 | inode->i_ino, left->p_tree_depth, | |
1414 | right->p_tree_depth, | |
1415 | (unsigned long long)path_leaf_bh(left)->b_blocknr, | |
1416 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1417 | } while (left->p_node[i].bh->b_blocknr == | |
1418 | right->p_node[i].bh->b_blocknr); | |
1419 | ||
1420 | return i - 1; | |
1421 | } | |
1422 | ||
1423 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1424 | ||
1425 | /* | |
1426 | * Traverse a btree path in search of cpos, starting at root_el. | |
1427 | * | |
1428 | * This code can be called with a cpos larger than the tree, in which | |
1429 | * case it will return the rightmost path. | |
1430 | */ | |
1431 | static int __ocfs2_find_path(struct inode *inode, | |
1432 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1433 | path_insert_t *func, void *data) | |
1434 | { | |
1435 | int i, ret = 0; | |
1436 | u32 range; | |
1437 | u64 blkno; | |
1438 | struct buffer_head *bh = NULL; | |
1439 | struct ocfs2_extent_block *eb; | |
1440 | struct ocfs2_extent_list *el; | |
1441 | struct ocfs2_extent_rec *rec; | |
1442 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1443 | ||
1444 | el = root_el; | |
1445 | while (el->l_tree_depth) { | |
1446 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
1447 | ocfs2_error(inode->i_sb, | |
1448 | "Inode %llu has empty extent list at " | |
1449 | "depth %u\n", | |
1450 | (unsigned long long)oi->ip_blkno, | |
1451 | le16_to_cpu(el->l_tree_depth)); | |
1452 | ret = -EROFS; | |
1453 | goto out; | |
1454 | ||
1455 | } | |
1456 | ||
1457 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1458 | rec = &el->l_recs[i]; | |
1459 | ||
1460 | /* | |
1461 | * In the case that cpos is off the allocation | |
1462 | * tree, this should just wind up returning the | |
1463 | * rightmost record. | |
1464 | */ | |
1465 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1466 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1467 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1468 | break; | |
1469 | } | |
1470 | ||
1471 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1472 | if (blkno == 0) { | |
1473 | ocfs2_error(inode->i_sb, | |
1474 | "Inode %llu has bad blkno in extent list " | |
1475 | "at depth %u (index %d)\n", | |
1476 | (unsigned long long)oi->ip_blkno, | |
1477 | le16_to_cpu(el->l_tree_depth), i); | |
1478 | ret = -EROFS; | |
1479 | goto out; | |
1480 | } | |
1481 | ||
1482 | brelse(bh); | |
1483 | bh = NULL; | |
1484 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, | |
1485 | &bh, OCFS2_BH_CACHED, inode); | |
1486 | if (ret) { | |
1487 | mlog_errno(ret); | |
1488 | goto out; | |
1489 | } | |
1490 | ||
1491 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1492 | el = &eb->h_list; | |
1493 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1494 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1495 | ret = -EIO; | |
1496 | goto out; | |
1497 | } | |
1498 | ||
1499 | if (le16_to_cpu(el->l_next_free_rec) > | |
1500 | le16_to_cpu(el->l_count)) { | |
1501 | ocfs2_error(inode->i_sb, | |
1502 | "Inode %llu has bad count in extent list " | |
1503 | "at block %llu (next free=%u, count=%u)\n", | |
1504 | (unsigned long long)oi->ip_blkno, | |
1505 | (unsigned long long)bh->b_blocknr, | |
1506 | le16_to_cpu(el->l_next_free_rec), | |
1507 | le16_to_cpu(el->l_count)); | |
1508 | ret = -EROFS; | |
1509 | goto out; | |
1510 | } | |
1511 | ||
1512 | if (func) | |
1513 | func(data, bh); | |
1514 | } | |
1515 | ||
1516 | out: | |
1517 | /* | |
1518 | * Catch any trailing bh that the loop didn't handle. | |
1519 | */ | |
1520 | brelse(bh); | |
1521 | ||
1522 | return ret; | |
1523 | } | |
1524 | ||
1525 | /* | |
1526 | * Given an initialized path (that is, it has a valid root extent | |
1527 | * list), this function will traverse the btree in search of the path | |
1528 | * which would contain cpos. | |
1529 | * | |
1530 | * The path traveled is recorded in the path structure. | |
1531 | * | |
1532 | * Note that this will not do any comparisons on leaf node extent | |
1533 | * records, so it will work fine in the case that we just added a tree | |
1534 | * branch. | |
1535 | */ | |
1536 | struct find_path_data { | |
1537 | int index; | |
1538 | struct ocfs2_path *path; | |
1539 | }; | |
1540 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1541 | { | |
1542 | struct find_path_data *fp = data; | |
1543 | ||
1544 | get_bh(bh); | |
1545 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1546 | fp->index++; | |
1547 | } | |
1548 | static int ocfs2_find_path(struct inode *inode, struct ocfs2_path *path, | |
1549 | u32 cpos) | |
1550 | { | |
1551 | struct find_path_data data; | |
1552 | ||
1553 | data.index = 1; | |
1554 | data.path = path; | |
1555 | return __ocfs2_find_path(inode, path_root_el(path), cpos, | |
1556 | find_path_ins, &data); | |
1557 | } | |
1558 | ||
1559 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1560 | { | |
1561 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1562 | struct ocfs2_extent_list *el = &eb->h_list; | |
1563 | struct buffer_head **ret = data; | |
1564 | ||
1565 | /* We want to retain only the leaf block. */ | |
1566 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1567 | get_bh(bh); | |
1568 | *ret = bh; | |
1569 | } | |
1570 | } | |
1571 | /* | |
1572 | * Find the leaf block in the tree which would contain cpos. No | |
1573 | * checking of the actual leaf is done. | |
1574 | * | |
1575 | * Some paths want to call this instead of allocating a path structure | |
1576 | * and calling ocfs2_find_path(). | |
1577 | * | |
1578 | * This function doesn't handle non btree extent lists. | |
1579 | */ | |
363041a5 MF |
1580 | int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el, |
1581 | u32 cpos, struct buffer_head **leaf_bh) | |
dcd0538f MF |
1582 | { |
1583 | int ret; | |
1584 | struct buffer_head *bh = NULL; | |
1585 | ||
1586 | ret = __ocfs2_find_path(inode, root_el, cpos, find_leaf_ins, &bh); | |
1587 | if (ret) { | |
1588 | mlog_errno(ret); | |
1589 | goto out; | |
1590 | } | |
1591 | ||
1592 | *leaf_bh = bh; | |
1593 | out: | |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1599 | * | |
1600 | * Basically, we've moved stuff around at the bottom of the tree and | |
1601 | * we need to fix up the extent records above the changes to reflect | |
1602 | * the new changes. | |
1603 | * | |
1604 | * left_rec: the record on the left. | |
1605 | * left_child_el: is the child list pointed to by left_rec | |
1606 | * right_rec: the record to the right of left_rec | |
1607 | * right_child_el: is the child list pointed to by right_rec | |
1608 | * | |
1609 | * By definition, this only works on interior nodes. | |
1610 | */ | |
1611 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1612 | struct ocfs2_extent_list *left_child_el, | |
1613 | struct ocfs2_extent_rec *right_rec, | |
1614 | struct ocfs2_extent_list *right_child_el) | |
1615 | { | |
1616 | u32 left_clusters, right_end; | |
1617 | ||
1618 | /* | |
1619 | * Interior nodes never have holes. Their cpos is the cpos of | |
1620 | * the leftmost record in their child list. Their cluster | |
1621 | * count covers the full theoretical range of their child list | |
1622 | * - the range between their cpos and the cpos of the record | |
1623 | * immediately to their right. | |
1624 | */ | |
1625 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
328d5752 MF |
1626 | if (ocfs2_is_empty_extent(&right_child_el->l_recs[0])) { |
1627 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); | |
1628 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1629 | } | |
dcd0538f | 1630 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1631 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1632 | |
1633 | /* | |
1634 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1635 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1636 | * updating e_cpos to keep the same highest cluster count. |
1637 | */ | |
1638 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1639 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1640 | |
1641 | right_rec->e_cpos = left_rec->e_cpos; | |
1642 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1643 | ||
1644 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1645 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1646 | } |
1647 | ||
1648 | /* | |
1649 | * Adjust the adjacent root node records involved in a | |
1650 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1651 | * find it's index in the root list. | |
1652 | */ | |
1653 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
1654 | struct ocfs2_extent_list *left_el, | |
1655 | struct ocfs2_extent_list *right_el, | |
1656 | u64 left_el_blkno) | |
1657 | { | |
1658 | int i; | |
1659 | ||
1660 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
1661 | le16_to_cpu(left_el->l_tree_depth)); | |
1662 | ||
1663 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
1664 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
1665 | break; | |
1666 | } | |
1667 | ||
1668 | /* | |
1669 | * The path walking code should have never returned a root and | |
1670 | * two paths which are not adjacent. | |
1671 | */ | |
1672 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
1673 | ||
1674 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
1675 | &root_el->l_recs[i + 1], right_el); | |
1676 | } | |
1677 | ||
1678 | /* | |
1679 | * We've changed a leaf block (in right_path) and need to reflect that | |
1680 | * change back up the subtree. | |
1681 | * | |
1682 | * This happens in multiple places: | |
1683 | * - When we've moved an extent record from the left path leaf to the right | |
1684 | * path leaf to make room for an empty extent in the left path leaf. | |
1685 | * - When our insert into the right path leaf is at the leftmost edge | |
1686 | * and requires an update of the path immediately to it's left. This | |
1687 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
1688 | * - When we've adjusted the last extent record in the left path leaf and the |
1689 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f MF |
1690 | */ |
1691 | static void ocfs2_complete_edge_insert(struct inode *inode, handle_t *handle, | |
1692 | struct ocfs2_path *left_path, | |
1693 | struct ocfs2_path *right_path, | |
1694 | int subtree_index) | |
1695 | { | |
1696 | int ret, i, idx; | |
1697 | struct ocfs2_extent_list *el, *left_el, *right_el; | |
1698 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
1699 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
1700 | ||
1701 | /* | |
1702 | * Update the counts and position values within all the | |
1703 | * interior nodes to reflect the leaf rotation we just did. | |
1704 | * | |
1705 | * The root node is handled below the loop. | |
1706 | * | |
1707 | * We begin the loop with right_el and left_el pointing to the | |
1708 | * leaf lists and work our way up. | |
1709 | * | |
1710 | * NOTE: within this loop, left_el and right_el always refer | |
1711 | * to the *child* lists. | |
1712 | */ | |
1713 | left_el = path_leaf_el(left_path); | |
1714 | right_el = path_leaf_el(right_path); | |
1715 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
1716 | mlog(0, "Adjust records at index %u\n", i); | |
1717 | ||
1718 | /* | |
1719 | * One nice property of knowing that all of these | |
1720 | * nodes are below the root is that we only deal with | |
1721 | * the leftmost right node record and the rightmost | |
1722 | * left node record. | |
1723 | */ | |
1724 | el = left_path->p_node[i].el; | |
1725 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1726 | left_rec = &el->l_recs[idx]; | |
1727 | ||
1728 | el = right_path->p_node[i].el; | |
1729 | right_rec = &el->l_recs[0]; | |
1730 | ||
1731 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
1732 | right_el); | |
1733 | ||
1734 | ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh); | |
1735 | if (ret) | |
1736 | mlog_errno(ret); | |
1737 | ||
1738 | ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
1739 | if (ret) | |
1740 | mlog_errno(ret); | |
1741 | ||
1742 | /* | |
1743 | * Setup our list pointers now so that the current | |
1744 | * parents become children in the next iteration. | |
1745 | */ | |
1746 | left_el = left_path->p_node[i].el; | |
1747 | right_el = right_path->p_node[i].el; | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * At the root node, adjust the two adjacent records which | |
1752 | * begin our path to the leaves. | |
1753 | */ | |
1754 | ||
1755 | el = left_path->p_node[subtree_index].el; | |
1756 | left_el = left_path->p_node[subtree_index + 1].el; | |
1757 | right_el = right_path->p_node[subtree_index + 1].el; | |
1758 | ||
1759 | ocfs2_adjust_root_records(el, left_el, right_el, | |
1760 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
1761 | ||
1762 | root_bh = left_path->p_node[subtree_index].bh; | |
1763 | ||
1764 | ret = ocfs2_journal_dirty(handle, root_bh); | |
1765 | if (ret) | |
1766 | mlog_errno(ret); | |
1767 | } | |
1768 | ||
1769 | static int ocfs2_rotate_subtree_right(struct inode *inode, | |
1770 | handle_t *handle, | |
1771 | struct ocfs2_path *left_path, | |
1772 | struct ocfs2_path *right_path, | |
1773 | int subtree_index) | |
1774 | { | |
1775 | int ret, i; | |
1776 | struct buffer_head *right_leaf_bh; | |
1777 | struct buffer_head *left_leaf_bh = NULL; | |
1778 | struct buffer_head *root_bh; | |
1779 | struct ocfs2_extent_list *right_el, *left_el; | |
1780 | struct ocfs2_extent_rec move_rec; | |
1781 | ||
1782 | left_leaf_bh = path_leaf_bh(left_path); | |
1783 | left_el = path_leaf_el(left_path); | |
1784 | ||
1785 | if (left_el->l_next_free_rec != left_el->l_count) { | |
1786 | ocfs2_error(inode->i_sb, | |
1787 | "Inode %llu has non-full interior leaf node %llu" | |
1788 | "(next free = %u)", | |
1789 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1790 | (unsigned long long)left_leaf_bh->b_blocknr, | |
1791 | le16_to_cpu(left_el->l_next_free_rec)); | |
1792 | return -EROFS; | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * This extent block may already have an empty record, so we | |
1797 | * return early if so. | |
1798 | */ | |
1799 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
1800 | return 0; | |
1801 | ||
1802 | root_bh = left_path->p_node[subtree_index].bh; | |
1803 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
1804 | ||
1805 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
1806 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1807 | if (ret) { | |
1808 | mlog_errno(ret); | |
1809 | goto out; | |
1810 | } | |
1811 | ||
1812 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1813 | ret = ocfs2_journal_access(handle, inode, | |
1814 | right_path->p_node[i].bh, | |
1815 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1816 | if (ret) { | |
1817 | mlog_errno(ret); | |
1818 | goto out; | |
1819 | } | |
1820 | ||
1821 | ret = ocfs2_journal_access(handle, inode, | |
1822 | left_path->p_node[i].bh, | |
1823 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1824 | if (ret) { | |
1825 | mlog_errno(ret); | |
1826 | goto out; | |
1827 | } | |
1828 | } | |
1829 | ||
1830 | right_leaf_bh = path_leaf_bh(right_path); | |
1831 | right_el = path_leaf_el(right_path); | |
1832 | ||
1833 | /* This is a code error, not a disk corruption. */ | |
1834 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
1835 | "because rightmost leaf block %llu is empty\n", | |
1836 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1837 | (unsigned long long)right_leaf_bh->b_blocknr); | |
1838 | ||
1839 | ocfs2_create_empty_extent(right_el); | |
1840 | ||
1841 | ret = ocfs2_journal_dirty(handle, right_leaf_bh); | |
1842 | if (ret) { | |
1843 | mlog_errno(ret); | |
1844 | goto out; | |
1845 | } | |
1846 | ||
1847 | /* Do the copy now. */ | |
1848 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1849 | move_rec = left_el->l_recs[i]; | |
1850 | right_el->l_recs[0] = move_rec; | |
1851 | ||
1852 | /* | |
1853 | * Clear out the record we just copied and shift everything | |
1854 | * over, leaving an empty extent in the left leaf. | |
1855 | * | |
1856 | * We temporarily subtract from next_free_rec so that the | |
1857 | * shift will lose the tail record (which is now defunct). | |
1858 | */ | |
1859 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
1860 | ocfs2_shift_records_right(left_el); | |
1861 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1862 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
1863 | ||
1864 | ret = ocfs2_journal_dirty(handle, left_leaf_bh); | |
1865 | if (ret) { | |
1866 | mlog_errno(ret); | |
1867 | goto out; | |
1868 | } | |
1869 | ||
1870 | ocfs2_complete_edge_insert(inode, handle, left_path, right_path, | |
1871 | subtree_index); | |
1872 | ||
1873 | out: | |
1874 | return ret; | |
1875 | } | |
1876 | ||
1877 | /* | |
1878 | * Given a full path, determine what cpos value would return us a path | |
1879 | * containing the leaf immediately to the left of the current one. | |
1880 | * | |
1881 | * Will return zero if the path passed in is already the leftmost path. | |
1882 | */ | |
1883 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
1884 | struct ocfs2_path *path, u32 *cpos) | |
1885 | { | |
1886 | int i, j, ret = 0; | |
1887 | u64 blkno; | |
1888 | struct ocfs2_extent_list *el; | |
1889 | ||
e48edee2 MF |
1890 | BUG_ON(path->p_tree_depth == 0); |
1891 | ||
dcd0538f MF |
1892 | *cpos = 0; |
1893 | ||
1894 | blkno = path_leaf_bh(path)->b_blocknr; | |
1895 | ||
1896 | /* Start at the tree node just above the leaf and work our way up. */ | |
1897 | i = path->p_tree_depth - 1; | |
1898 | while (i >= 0) { | |
1899 | el = path->p_node[i].el; | |
1900 | ||
1901 | /* | |
1902 | * Find the extent record just before the one in our | |
1903 | * path. | |
1904 | */ | |
1905 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
1906 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
1907 | if (j == 0) { | |
1908 | if (i == 0) { | |
1909 | /* | |
1910 | * We've determined that the | |
1911 | * path specified is already | |
1912 | * the leftmost one - return a | |
1913 | * cpos of zero. | |
1914 | */ | |
1915 | goto out; | |
1916 | } | |
1917 | /* | |
1918 | * The leftmost record points to our | |
1919 | * leaf - we need to travel up the | |
1920 | * tree one level. | |
1921 | */ | |
1922 | goto next_node; | |
1923 | } | |
1924 | ||
1925 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
1926 | *cpos = *cpos + ocfs2_rec_clusters(el, |
1927 | &el->l_recs[j - 1]); | |
1928 | *cpos = *cpos - 1; | |
dcd0538f MF |
1929 | goto out; |
1930 | } | |
1931 | } | |
1932 | ||
1933 | /* | |
1934 | * If we got here, we never found a valid node where | |
1935 | * the tree indicated one should be. | |
1936 | */ | |
1937 | ocfs2_error(sb, | |
1938 | "Invalid extent tree at extent block %llu\n", | |
1939 | (unsigned long long)blkno); | |
1940 | ret = -EROFS; | |
1941 | goto out; | |
1942 | ||
1943 | next_node: | |
1944 | blkno = path->p_node[i].bh->b_blocknr; | |
1945 | i--; | |
1946 | } | |
1947 | ||
1948 | out: | |
1949 | return ret; | |
1950 | } | |
1951 | ||
328d5752 MF |
1952 | /* |
1953 | * Extend the transaction by enough credits to complete the rotation, | |
1954 | * and still leave at least the original number of credits allocated | |
1955 | * to this transaction. | |
1956 | */ | |
dcd0538f | 1957 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 1958 | int op_credits, |
dcd0538f MF |
1959 | struct ocfs2_path *path) |
1960 | { | |
328d5752 | 1961 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f MF |
1962 | |
1963 | if (handle->h_buffer_credits < credits) | |
1964 | return ocfs2_extend_trans(handle, credits); | |
1965 | ||
1966 | return 0; | |
1967 | } | |
1968 | ||
1969 | /* | |
1970 | * Trap the case where we're inserting into the theoretical range past | |
1971 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
1972 | * whose cpos is less than ours into the right leaf. | |
1973 | * | |
1974 | * It's only necessary to look at the rightmost record of the left | |
1975 | * leaf because the logic that calls us should ensure that the | |
1976 | * theoretical ranges in the path components above the leaves are | |
1977 | * correct. | |
1978 | */ | |
1979 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
1980 | u32 insert_cpos) | |
1981 | { | |
1982 | struct ocfs2_extent_list *left_el; | |
1983 | struct ocfs2_extent_rec *rec; | |
1984 | int next_free; | |
1985 | ||
1986 | left_el = path_leaf_el(left_path); | |
1987 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
1988 | rec = &left_el->l_recs[next_free - 1]; | |
1989 | ||
1990 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
1991 | return 1; | |
1992 | return 0; | |
1993 | } | |
1994 | ||
328d5752 MF |
1995 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
1996 | { | |
1997 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1998 | unsigned int range; | |
1999 | struct ocfs2_extent_rec *rec; | |
2000 | ||
2001 | if (next_free == 0) | |
2002 | return 0; | |
2003 | ||
2004 | rec = &el->l_recs[0]; | |
2005 | if (ocfs2_is_empty_extent(rec)) { | |
2006 | /* Empty list. */ | |
2007 | if (next_free == 1) | |
2008 | return 0; | |
2009 | rec = &el->l_recs[1]; | |
2010 | } | |
2011 | ||
2012 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2013 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2014 | return 1; | |
2015 | return 0; | |
2016 | } | |
2017 | ||
dcd0538f MF |
2018 | /* |
2019 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2020 | * | |
2021 | * The path to the rightmost leaf should be passed in. | |
2022 | * | |
2023 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2024 | * | |
2025 | * Upon succesful return from this function: | |
2026 | * | |
2027 | * - The 'right_path' array will contain a path to the leaf block | |
2028 | * whose range contains e_cpos. | |
2029 | * - That leaf block will have a single empty extent in list index 0. | |
2030 | * - In the case that the rotation requires a post-insert update, | |
2031 | * *ret_left_path will contain a valid path which can be passed to | |
2032 | * ocfs2_insert_path(). | |
2033 | */ | |
2034 | static int ocfs2_rotate_tree_right(struct inode *inode, | |
2035 | handle_t *handle, | |
328d5752 | 2036 | enum ocfs2_split_type split, |
dcd0538f MF |
2037 | u32 insert_cpos, |
2038 | struct ocfs2_path *right_path, | |
2039 | struct ocfs2_path **ret_left_path) | |
2040 | { | |
328d5752 | 2041 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2042 | u32 cpos; |
2043 | struct ocfs2_path *left_path = NULL; | |
2044 | ||
2045 | *ret_left_path = NULL; | |
2046 | ||
2047 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
2048 | path_root_el(right_path)); | |
2049 | if (!left_path) { | |
2050 | ret = -ENOMEM; | |
2051 | mlog_errno(ret); | |
2052 | goto out; | |
2053 | } | |
2054 | ||
2055 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, &cpos); | |
2056 | if (ret) { | |
2057 | mlog_errno(ret); | |
2058 | goto out; | |
2059 | } | |
2060 | ||
2061 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2062 | ||
2063 | /* | |
2064 | * What we want to do here is: | |
2065 | * | |
2066 | * 1) Start with the rightmost path. | |
2067 | * | |
2068 | * 2) Determine a path to the leaf block directly to the left | |
2069 | * of that leaf. | |
2070 | * | |
2071 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2072 | * which contains a path to both leaves. | |
2073 | * | |
2074 | * 4) Rotate the subtree. | |
2075 | * | |
2076 | * 5) Find the next subtree by considering the left path to be | |
2077 | * the new right path. | |
2078 | * | |
2079 | * The check at the top of this while loop also accepts | |
2080 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2081 | * value to get us the left path - insert_cpos might very well | |
2082 | * be filling that hole. | |
2083 | * | |
2084 | * Stop at a cpos of '0' because we either started at the | |
2085 | * leftmost branch (i.e., a tree with one branch and a | |
2086 | * rotation inside of it), or we've gone as far as we can in | |
2087 | * rotating subtrees. | |
2088 | */ | |
2089 | while (cpos && insert_cpos <= cpos) { | |
2090 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2091 | insert_cpos, cpos); | |
2092 | ||
2093 | ret = ocfs2_find_path(inode, left_path, cpos); | |
2094 | if (ret) { | |
2095 | mlog_errno(ret); | |
2096 | goto out; | |
2097 | } | |
2098 | ||
2099 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2100 | path_leaf_bh(right_path), | |
2101 | "Inode %lu: error during insert of %u " | |
2102 | "(left path cpos %u) results in two identical " | |
2103 | "paths ending at %llu\n", | |
2104 | inode->i_ino, insert_cpos, cpos, | |
2105 | (unsigned long long) | |
2106 | path_leaf_bh(left_path)->b_blocknr); | |
2107 | ||
328d5752 MF |
2108 | if (split == SPLIT_NONE && |
2109 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2110 | insert_cpos)) { |
dcd0538f MF |
2111 | |
2112 | /* | |
2113 | * We've rotated the tree as much as we | |
2114 | * should. The rest is up to | |
2115 | * ocfs2_insert_path() to complete, after the | |
2116 | * record insertion. We indicate this | |
2117 | * situation by returning the left path. | |
2118 | * | |
2119 | * The reason we don't adjust the records here | |
2120 | * before the record insert is that an error | |
2121 | * later might break the rule where a parent | |
2122 | * record e_cpos will reflect the actual | |
2123 | * e_cpos of the 1st nonempty record of the | |
2124 | * child list. | |
2125 | */ | |
2126 | *ret_left_path = left_path; | |
2127 | goto out_ret_path; | |
2128 | } | |
2129 | ||
2130 | start = ocfs2_find_subtree_root(inode, left_path, right_path); | |
2131 | ||
2132 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2133 | start, | |
2134 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2135 | right_path->p_tree_depth); | |
2136 | ||
2137 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2138 | orig_credits, right_path); |
dcd0538f MF |
2139 | if (ret) { |
2140 | mlog_errno(ret); | |
2141 | goto out; | |
2142 | } | |
2143 | ||
2144 | ret = ocfs2_rotate_subtree_right(inode, handle, left_path, | |
2145 | right_path, start); | |
2146 | if (ret) { | |
2147 | mlog_errno(ret); | |
2148 | goto out; | |
2149 | } | |
2150 | ||
328d5752 MF |
2151 | if (split != SPLIT_NONE && |
2152 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2153 | insert_cpos)) { | |
2154 | /* | |
2155 | * A rotate moves the rightmost left leaf | |
2156 | * record over to the leftmost right leaf | |
2157 | * slot. If we're doing an extent split | |
2158 | * instead of a real insert, then we have to | |
2159 | * check that the extent to be split wasn't | |
2160 | * just moved over. If it was, then we can | |
2161 | * exit here, passing left_path back - | |
2162 | * ocfs2_split_extent() is smart enough to | |
2163 | * search both leaves. | |
2164 | */ | |
2165 | *ret_left_path = left_path; | |
2166 | goto out_ret_path; | |
2167 | } | |
2168 | ||
dcd0538f MF |
2169 | /* |
2170 | * There is no need to re-read the next right path | |
2171 | * as we know that it'll be our current left | |
2172 | * path. Optimize by copying values instead. | |
2173 | */ | |
2174 | ocfs2_mv_path(right_path, left_path); | |
2175 | ||
2176 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, | |
2177 | &cpos); | |
2178 | if (ret) { | |
2179 | mlog_errno(ret); | |
2180 | goto out; | |
2181 | } | |
2182 | } | |
2183 | ||
2184 | out: | |
2185 | ocfs2_free_path(left_path); | |
2186 | ||
2187 | out_ret_path: | |
2188 | return ret; | |
2189 | } | |
2190 | ||
328d5752 MF |
2191 | static void ocfs2_update_edge_lengths(struct inode *inode, handle_t *handle, |
2192 | struct ocfs2_path *path) | |
dcd0538f | 2193 | { |
328d5752 | 2194 | int i, idx; |
dcd0538f | 2195 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2196 | struct ocfs2_extent_list *el; |
2197 | struct ocfs2_extent_block *eb; | |
2198 | u32 range; | |
dcd0538f | 2199 | |
328d5752 MF |
2200 | /* Path should always be rightmost. */ |
2201 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2202 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2203 | |
328d5752 MF |
2204 | el = &eb->h_list; |
2205 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2206 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2207 | rec = &el->l_recs[idx]; | |
2208 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2209 | |
328d5752 MF |
2210 | for (i = 0; i < path->p_tree_depth; i++) { |
2211 | el = path->p_node[i].el; | |
2212 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2213 | rec = &el->l_recs[idx]; | |
dcd0538f | 2214 | |
328d5752 MF |
2215 | rec->e_int_clusters = cpu_to_le32(range); |
2216 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2217 | |
328d5752 | 2218 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2219 | } |
dcd0538f MF |
2220 | } |
2221 | ||
328d5752 MF |
2222 | static void ocfs2_unlink_path(struct inode *inode, handle_t *handle, |
2223 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
2224 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2225 | { |
328d5752 MF |
2226 | int ret, i; |
2227 | struct ocfs2_extent_block *eb; | |
2228 | struct ocfs2_extent_list *el; | |
2229 | struct buffer_head *bh; | |
2230 | ||
2231 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2232 | bh = path->p_node[i].bh; | |
2233 | ||
2234 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2235 | /* | |
2236 | * Not all nodes might have had their final count | |
2237 | * decremented by the caller - handle this here. | |
2238 | */ | |
2239 | el = &eb->h_list; | |
2240 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2241 | mlog(ML_ERROR, | |
2242 | "Inode %llu, attempted to remove extent block " | |
2243 | "%llu with %u records\n", | |
2244 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2245 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
2246 | le16_to_cpu(el->l_next_free_rec)); | |
2247 | ||
2248 | ocfs2_journal_dirty(handle, bh); | |
2249 | ocfs2_remove_from_cache(inode, bh); | |
2250 | continue; | |
2251 | } | |
2252 | ||
2253 | el->l_next_free_rec = 0; | |
2254 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2255 | ||
2256 | ocfs2_journal_dirty(handle, bh); | |
2257 | ||
2258 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2259 | if (ret) | |
2260 | mlog_errno(ret); | |
2261 | ||
2262 | ocfs2_remove_from_cache(inode, bh); | |
2263 | } | |
dcd0538f MF |
2264 | } |
2265 | ||
328d5752 MF |
2266 | static void ocfs2_unlink_subtree(struct inode *inode, handle_t *handle, |
2267 | struct ocfs2_path *left_path, | |
2268 | struct ocfs2_path *right_path, | |
2269 | int subtree_index, | |
2270 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2271 | { |
328d5752 MF |
2272 | int i; |
2273 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2274 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2275 | struct ocfs2_extent_list *el; |
328d5752 | 2276 | struct ocfs2_extent_block *eb; |
dcd0538f | 2277 | |
328d5752 | 2278 | el = path_leaf_el(left_path); |
dcd0538f | 2279 | |
328d5752 | 2280 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2281 | |
328d5752 MF |
2282 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2283 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2284 | break; | |
dcd0538f | 2285 | |
328d5752 | 2286 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2287 | |
328d5752 MF |
2288 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2289 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2290 | ||
2291 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2292 | eb->h_next_leaf_blk = 0; | |
2293 | ||
2294 | ocfs2_journal_dirty(handle, root_bh); | |
2295 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2296 | ||
2297 | ocfs2_unlink_path(inode, handle, dealloc, right_path, | |
2298 | subtree_index + 1); | |
2299 | } | |
2300 | ||
2301 | static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle, | |
2302 | struct ocfs2_path *left_path, | |
2303 | struct ocfs2_path *right_path, | |
2304 | int subtree_index, | |
2305 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
2306 | int *deleted, |
2307 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2308 | { |
2309 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2310 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2311 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2312 | struct ocfs2_extent_block *eb; | |
2313 | ||
2314 | *deleted = 0; | |
2315 | ||
2316 | right_leaf_el = path_leaf_el(right_path); | |
2317 | left_leaf_el = path_leaf_el(left_path); | |
2318 | root_bh = left_path->p_node[subtree_index].bh; | |
2319 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2320 | ||
2321 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2322 | return 0; | |
dcd0538f | 2323 | |
328d5752 MF |
2324 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2325 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2326 | /* |
328d5752 MF |
2327 | * It's legal for us to proceed if the right leaf is |
2328 | * the rightmost one and it has an empty extent. There | |
2329 | * are two cases to handle - whether the leaf will be | |
2330 | * empty after removal or not. If the leaf isn't empty | |
2331 | * then just remove the empty extent up front. The | |
2332 | * next block will handle empty leaves by flagging | |
2333 | * them for unlink. | |
2334 | * | |
2335 | * Non rightmost leaves will throw -EAGAIN and the | |
2336 | * caller can manually move the subtree and retry. | |
dcd0538f | 2337 | */ |
dcd0538f | 2338 | |
328d5752 MF |
2339 | if (eb->h_next_leaf_blk != 0ULL) |
2340 | return -EAGAIN; | |
2341 | ||
2342 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
2343 | ret = ocfs2_journal_access(handle, inode, | |
2344 | path_leaf_bh(right_path), | |
2345 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2346 | if (ret) { |
2347 | mlog_errno(ret); | |
2348 | goto out; | |
2349 | } | |
2350 | ||
328d5752 MF |
2351 | ocfs2_remove_empty_extent(right_leaf_el); |
2352 | } else | |
2353 | right_has_empty = 1; | |
dcd0538f MF |
2354 | } |
2355 | ||
328d5752 MF |
2356 | if (eb->h_next_leaf_blk == 0ULL && |
2357 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2358 | /* | |
2359 | * We have to update i_last_eb_blk during the meta | |
2360 | * data delete. | |
2361 | */ | |
e7d4cb6b | 2362 | ret = ocfs2_journal_access(handle, inode, et_root_bh, |
328d5752 MF |
2363 | OCFS2_JOURNAL_ACCESS_WRITE); |
2364 | if (ret) { | |
2365 | mlog_errno(ret); | |
2366 | goto out; | |
2367 | } | |
2368 | ||
2369 | del_right_subtree = 1; | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * Getting here with an empty extent in the right path implies | |
2374 | * that it's the rightmost path and will be deleted. | |
2375 | */ | |
2376 | BUG_ON(right_has_empty && !del_right_subtree); | |
2377 | ||
2378 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
2379 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2380 | if (ret) { | |
2381 | mlog_errno(ret); | |
2382 | goto out; | |
2383 | } | |
2384 | ||
2385 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
2386 | ret = ocfs2_journal_access(handle, inode, | |
2387 | right_path->p_node[i].bh, | |
2388 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2389 | if (ret) { | |
2390 | mlog_errno(ret); | |
2391 | goto out; | |
2392 | } | |
2393 | ||
2394 | ret = ocfs2_journal_access(handle, inode, | |
2395 | left_path->p_node[i].bh, | |
2396 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2397 | if (ret) { | |
2398 | mlog_errno(ret); | |
2399 | goto out; | |
2400 | } | |
2401 | } | |
2402 | ||
2403 | if (!right_has_empty) { | |
2404 | /* | |
2405 | * Only do this if we're moving a real | |
2406 | * record. Otherwise, the action is delayed until | |
2407 | * after removal of the right path in which case we | |
2408 | * can do a simple shift to remove the empty extent. | |
2409 | */ | |
2410 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2411 | memset(&right_leaf_el->l_recs[0], 0, | |
2412 | sizeof(struct ocfs2_extent_rec)); | |
2413 | } | |
2414 | if (eb->h_next_leaf_blk == 0ULL) { | |
2415 | /* | |
2416 | * Move recs over to get rid of empty extent, decrease | |
2417 | * next_free. This is allowed to remove the last | |
2418 | * extent in our leaf (setting l_next_free_rec to | |
2419 | * zero) - the delete code below won't care. | |
2420 | */ | |
2421 | ocfs2_remove_empty_extent(right_leaf_el); | |
2422 | } | |
2423 | ||
2424 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2425 | if (ret) | |
2426 | mlog_errno(ret); | |
2427 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
2428 | if (ret) | |
2429 | mlog_errno(ret); | |
2430 | ||
2431 | if (del_right_subtree) { | |
2432 | ocfs2_unlink_subtree(inode, handle, left_path, right_path, | |
2433 | subtree_index, dealloc); | |
2434 | ocfs2_update_edge_lengths(inode, handle, left_path); | |
2435 | ||
2436 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2437 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2438 | |
2439 | /* | |
2440 | * Removal of the extent in the left leaf was skipped | |
2441 | * above so we could delete the right path | |
2442 | * 1st. | |
2443 | */ | |
2444 | if (right_has_empty) | |
2445 | ocfs2_remove_empty_extent(left_leaf_el); | |
2446 | ||
e7d4cb6b | 2447 | ret = ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2448 | if (ret) |
2449 | mlog_errno(ret); | |
2450 | ||
2451 | *deleted = 1; | |
2452 | } else | |
2453 | ocfs2_complete_edge_insert(inode, handle, left_path, right_path, | |
2454 | subtree_index); | |
2455 | ||
2456 | out: | |
2457 | return ret; | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * Given a full path, determine what cpos value would return us a path | |
2462 | * containing the leaf immediately to the right of the current one. | |
2463 | * | |
2464 | * Will return zero if the path passed in is already the rightmost path. | |
2465 | * | |
2466 | * This looks similar, but is subtly different to | |
2467 | * ocfs2_find_cpos_for_left_leaf(). | |
2468 | */ | |
2469 | static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, | |
2470 | struct ocfs2_path *path, u32 *cpos) | |
2471 | { | |
2472 | int i, j, ret = 0; | |
2473 | u64 blkno; | |
2474 | struct ocfs2_extent_list *el; | |
2475 | ||
2476 | *cpos = 0; | |
2477 | ||
2478 | if (path->p_tree_depth == 0) | |
2479 | return 0; | |
2480 | ||
2481 | blkno = path_leaf_bh(path)->b_blocknr; | |
2482 | ||
2483 | /* Start at the tree node just above the leaf and work our way up. */ | |
2484 | i = path->p_tree_depth - 1; | |
2485 | while (i >= 0) { | |
2486 | int next_free; | |
2487 | ||
2488 | el = path->p_node[i].el; | |
2489 | ||
2490 | /* | |
2491 | * Find the extent record just after the one in our | |
2492 | * path. | |
2493 | */ | |
2494 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2495 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2496 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2497 | if (j == (next_free - 1)) { | |
2498 | if (i == 0) { | |
2499 | /* | |
2500 | * We've determined that the | |
2501 | * path specified is already | |
2502 | * the rightmost one - return a | |
2503 | * cpos of zero. | |
2504 | */ | |
2505 | goto out; | |
2506 | } | |
2507 | /* | |
2508 | * The rightmost record points to our | |
2509 | * leaf - we need to travel up the | |
2510 | * tree one level. | |
2511 | */ | |
2512 | goto next_node; | |
2513 | } | |
2514 | ||
2515 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2516 | goto out; | |
2517 | } | |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * If we got here, we never found a valid node where | |
2522 | * the tree indicated one should be. | |
2523 | */ | |
2524 | ocfs2_error(sb, | |
2525 | "Invalid extent tree at extent block %llu\n", | |
2526 | (unsigned long long)blkno); | |
2527 | ret = -EROFS; | |
2528 | goto out; | |
2529 | ||
2530 | next_node: | |
2531 | blkno = path->p_node[i].bh->b_blocknr; | |
2532 | i--; | |
2533 | } | |
2534 | ||
2535 | out: | |
2536 | return ret; | |
2537 | } | |
2538 | ||
2539 | static int ocfs2_rotate_rightmost_leaf_left(struct inode *inode, | |
2540 | handle_t *handle, | |
2541 | struct buffer_head *bh, | |
2542 | struct ocfs2_extent_list *el) | |
2543 | { | |
2544 | int ret; | |
2545 | ||
2546 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2547 | return 0; | |
2548 | ||
2549 | ret = ocfs2_journal_access(handle, inode, bh, | |
2550 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2551 | if (ret) { | |
2552 | mlog_errno(ret); | |
2553 | goto out; | |
2554 | } | |
2555 | ||
2556 | ocfs2_remove_empty_extent(el); | |
2557 | ||
2558 | ret = ocfs2_journal_dirty(handle, bh); | |
2559 | if (ret) | |
2560 | mlog_errno(ret); | |
2561 | ||
2562 | out: | |
2563 | return ret; | |
2564 | } | |
2565 | ||
2566 | static int __ocfs2_rotate_tree_left(struct inode *inode, | |
2567 | handle_t *handle, int orig_credits, | |
2568 | struct ocfs2_path *path, | |
2569 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
2570 | struct ocfs2_path **empty_extent_path, |
2571 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2572 | { |
2573 | int ret, subtree_root, deleted; | |
2574 | u32 right_cpos; | |
2575 | struct ocfs2_path *left_path = NULL; | |
2576 | struct ocfs2_path *right_path = NULL; | |
2577 | ||
2578 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2579 | ||
2580 | *empty_extent_path = NULL; | |
2581 | ||
2582 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, path, | |
2583 | &right_cpos); | |
2584 | if (ret) { | |
2585 | mlog_errno(ret); | |
2586 | goto out; | |
2587 | } | |
2588 | ||
2589 | left_path = ocfs2_new_path(path_root_bh(path), | |
2590 | path_root_el(path)); | |
2591 | if (!left_path) { | |
2592 | ret = -ENOMEM; | |
2593 | mlog_errno(ret); | |
2594 | goto out; | |
2595 | } | |
2596 | ||
2597 | ocfs2_cp_path(left_path, path); | |
2598 | ||
2599 | right_path = ocfs2_new_path(path_root_bh(path), | |
2600 | path_root_el(path)); | |
2601 | if (!right_path) { | |
2602 | ret = -ENOMEM; | |
2603 | mlog_errno(ret); | |
2604 | goto out; | |
2605 | } | |
2606 | ||
2607 | while (right_cpos) { | |
2608 | ret = ocfs2_find_path(inode, right_path, right_cpos); | |
2609 | if (ret) { | |
2610 | mlog_errno(ret); | |
2611 | goto out; | |
2612 | } | |
2613 | ||
2614 | subtree_root = ocfs2_find_subtree_root(inode, left_path, | |
2615 | right_path); | |
2616 | ||
2617 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2618 | subtree_root, | |
2619 | (unsigned long long) | |
2620 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2621 | right_path->p_tree_depth); | |
2622 | ||
2623 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
2624 | orig_credits, left_path); | |
2625 | if (ret) { | |
2626 | mlog_errno(ret); | |
2627 | goto out; | |
2628 | } | |
2629 | ||
e8aed345 MF |
2630 | /* |
2631 | * Caller might still want to make changes to the | |
2632 | * tree root, so re-add it to the journal here. | |
2633 | */ | |
2634 | ret = ocfs2_journal_access(handle, inode, | |
2635 | path_root_bh(left_path), | |
2636 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2637 | if (ret) { | |
2638 | mlog_errno(ret); | |
2639 | goto out; | |
2640 | } | |
2641 | ||
328d5752 MF |
2642 | ret = ocfs2_rotate_subtree_left(inode, handle, left_path, |
2643 | right_path, subtree_root, | |
e7d4cb6b | 2644 | dealloc, &deleted, et); |
328d5752 MF |
2645 | if (ret == -EAGAIN) { |
2646 | /* | |
2647 | * The rotation has to temporarily stop due to | |
2648 | * the right subtree having an empty | |
2649 | * extent. Pass it back to the caller for a | |
2650 | * fixup. | |
2651 | */ | |
2652 | *empty_extent_path = right_path; | |
2653 | right_path = NULL; | |
2654 | goto out; | |
2655 | } | |
2656 | if (ret) { | |
2657 | mlog_errno(ret); | |
2658 | goto out; | |
2659 | } | |
2660 | ||
2661 | /* | |
2662 | * The subtree rotate might have removed records on | |
2663 | * the rightmost edge. If so, then rotation is | |
2664 | * complete. | |
2665 | */ | |
2666 | if (deleted) | |
2667 | break; | |
2668 | ||
2669 | ocfs2_mv_path(left_path, right_path); | |
2670 | ||
2671 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path, | |
2672 | &right_cpos); | |
2673 | if (ret) { | |
2674 | mlog_errno(ret); | |
2675 | goto out; | |
2676 | } | |
2677 | } | |
2678 | ||
2679 | out: | |
2680 | ocfs2_free_path(right_path); | |
2681 | ocfs2_free_path(left_path); | |
2682 | ||
2683 | return ret; | |
2684 | } | |
2685 | ||
2686 | static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle, | |
e7d4cb6b TM |
2687 | struct ocfs2_path *path, |
2688 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
2689 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2690 | { |
2691 | int ret, subtree_index; | |
2692 | u32 cpos; | |
2693 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
2694 | struct ocfs2_extent_block *eb; |
2695 | struct ocfs2_extent_list *el; | |
2696 | ||
328d5752 | 2697 | |
35dc0aa3 | 2698 | ret = ocfs2_et_sanity_check(inode, et); |
e7d4cb6b TM |
2699 | if (ret) |
2700 | goto out; | |
328d5752 MF |
2701 | /* |
2702 | * There's two ways we handle this depending on | |
2703 | * whether path is the only existing one. | |
2704 | */ | |
2705 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
2706 | handle->h_buffer_credits, | |
2707 | path); | |
2708 | if (ret) { | |
2709 | mlog_errno(ret); | |
2710 | goto out; | |
2711 | } | |
2712 | ||
2713 | ret = ocfs2_journal_access_path(inode, handle, path); | |
2714 | if (ret) { | |
2715 | mlog_errno(ret); | |
2716 | goto out; | |
2717 | } | |
2718 | ||
2719 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); | |
2720 | if (ret) { | |
2721 | mlog_errno(ret); | |
2722 | goto out; | |
2723 | } | |
2724 | ||
2725 | if (cpos) { | |
2726 | /* | |
2727 | * We have a path to the left of this one - it needs | |
2728 | * an update too. | |
2729 | */ | |
2730 | left_path = ocfs2_new_path(path_root_bh(path), | |
2731 | path_root_el(path)); | |
2732 | if (!left_path) { | |
2733 | ret = -ENOMEM; | |
2734 | mlog_errno(ret); | |
2735 | goto out; | |
2736 | } | |
2737 | ||
2738 | ret = ocfs2_find_path(inode, left_path, cpos); | |
2739 | if (ret) { | |
2740 | mlog_errno(ret); | |
2741 | goto out; | |
2742 | } | |
2743 | ||
2744 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
2745 | if (ret) { | |
2746 | mlog_errno(ret); | |
2747 | goto out; | |
2748 | } | |
2749 | ||
2750 | subtree_index = ocfs2_find_subtree_root(inode, left_path, path); | |
2751 | ||
2752 | ocfs2_unlink_subtree(inode, handle, left_path, path, | |
2753 | subtree_index, dealloc); | |
2754 | ocfs2_update_edge_lengths(inode, handle, left_path); | |
2755 | ||
2756 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2757 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2758 | } else { |
2759 | /* | |
2760 | * 'path' is also the leftmost path which | |
2761 | * means it must be the only one. This gets | |
2762 | * handled differently because we want to | |
2763 | * revert the inode back to having extents | |
2764 | * in-line. | |
2765 | */ | |
2766 | ocfs2_unlink_path(inode, handle, dealloc, path, 1); | |
2767 | ||
e7d4cb6b | 2768 | el = et->root_el; |
328d5752 MF |
2769 | el->l_tree_depth = 0; |
2770 | el->l_next_free_rec = 0; | |
2771 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2772 | ||
35dc0aa3 | 2773 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
2774 | } |
2775 | ||
2776 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
2777 | ||
2778 | out: | |
2779 | ocfs2_free_path(left_path); | |
2780 | return ret; | |
2781 | } | |
2782 | ||
2783 | /* | |
2784 | * Left rotation of btree records. | |
2785 | * | |
2786 | * In many ways, this is (unsurprisingly) the opposite of right | |
2787 | * rotation. We start at some non-rightmost path containing an empty | |
2788 | * extent in the leaf block. The code works its way to the rightmost | |
2789 | * path by rotating records to the left in every subtree. | |
2790 | * | |
2791 | * This is used by any code which reduces the number of extent records | |
2792 | * in a leaf. After removal, an empty record should be placed in the | |
2793 | * leftmost list position. | |
2794 | * | |
2795 | * This won't handle a length update of the rightmost path records if | |
2796 | * the rightmost tree leaf record is removed so the caller is | |
2797 | * responsible for detecting and correcting that. | |
2798 | */ | |
2799 | static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle, | |
2800 | struct ocfs2_path *path, | |
e7d4cb6b TM |
2801 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2802 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2803 | { |
2804 | int ret, orig_credits = handle->h_buffer_credits; | |
2805 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
2806 | struct ocfs2_extent_block *eb; | |
2807 | struct ocfs2_extent_list *el; | |
2808 | ||
2809 | el = path_leaf_el(path); | |
2810 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2811 | return 0; | |
2812 | ||
2813 | if (path->p_tree_depth == 0) { | |
2814 | rightmost_no_delete: | |
2815 | /* | |
e7d4cb6b | 2816 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
2817 | * it up front. |
2818 | */ | |
2819 | ret = ocfs2_rotate_rightmost_leaf_left(inode, handle, | |
2820 | path_leaf_bh(path), | |
2821 | path_leaf_el(path)); | |
2822 | if (ret) | |
2823 | mlog_errno(ret); | |
2824 | goto out; | |
2825 | } | |
2826 | ||
2827 | /* | |
2828 | * Handle rightmost branch now. There's several cases: | |
2829 | * 1) simple rotation leaving records in there. That's trivial. | |
2830 | * 2) rotation requiring a branch delete - there's no more | |
2831 | * records left. Two cases of this: | |
2832 | * a) There are branches to the left. | |
2833 | * b) This is also the leftmost (the only) branch. | |
2834 | * | |
2835 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
2836 | * 2a) we need the left branch so that we can update it with the unlink | |
2837 | * 2b) we need to bring the inode back to inline extents. | |
2838 | */ | |
2839 | ||
2840 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2841 | el = &eb->h_list; | |
2842 | if (eb->h_next_leaf_blk == 0) { | |
2843 | /* | |
2844 | * This gets a bit tricky if we're going to delete the | |
2845 | * rightmost path. Get the other cases out of the way | |
2846 | * 1st. | |
2847 | */ | |
2848 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
2849 | goto rightmost_no_delete; | |
2850 | ||
2851 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
2852 | ret = -EIO; | |
2853 | ocfs2_error(inode->i_sb, | |
2854 | "Inode %llu has empty extent block at %llu", | |
2855 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2856 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
2857 | goto out; | |
2858 | } | |
2859 | ||
2860 | /* | |
2861 | * XXX: The caller can not trust "path" any more after | |
2862 | * this as it will have been deleted. What do we do? | |
2863 | * | |
2864 | * In theory the rotate-for-merge code will never get | |
2865 | * here because it'll always ask for a rotate in a | |
2866 | * nonempty list. | |
2867 | */ | |
2868 | ||
2869 | ret = ocfs2_remove_rightmost_path(inode, handle, path, | |
e7d4cb6b | 2870 | dealloc, et); |
328d5752 MF |
2871 | if (ret) |
2872 | mlog_errno(ret); | |
2873 | goto out; | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * Now we can loop, remembering the path we get from -EAGAIN | |
2878 | * and restarting from there. | |
2879 | */ | |
2880 | try_rotate: | |
2881 | ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, path, | |
e7d4cb6b | 2882 | dealloc, &restart_path, et); |
328d5752 MF |
2883 | if (ret && ret != -EAGAIN) { |
2884 | mlog_errno(ret); | |
2885 | goto out; | |
2886 | } | |
2887 | ||
2888 | while (ret == -EAGAIN) { | |
2889 | tmp_path = restart_path; | |
2890 | restart_path = NULL; | |
2891 | ||
2892 | ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, | |
2893 | tmp_path, dealloc, | |
e7d4cb6b | 2894 | &restart_path, et); |
328d5752 MF |
2895 | if (ret && ret != -EAGAIN) { |
2896 | mlog_errno(ret); | |
2897 | goto out; | |
2898 | } | |
2899 | ||
2900 | ocfs2_free_path(tmp_path); | |
2901 | tmp_path = NULL; | |
2902 | ||
2903 | if (ret == 0) | |
2904 | goto try_rotate; | |
2905 | } | |
2906 | ||
2907 | out: | |
2908 | ocfs2_free_path(tmp_path); | |
2909 | ocfs2_free_path(restart_path); | |
2910 | return ret; | |
2911 | } | |
2912 | ||
2913 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
2914 | int index) | |
2915 | { | |
2916 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
2917 | unsigned int size; | |
2918 | ||
2919 | if (rec->e_leaf_clusters == 0) { | |
2920 | /* | |
2921 | * We consumed all of the merged-from record. An empty | |
2922 | * extent cannot exist anywhere but the 1st array | |
2923 | * position, so move things over if the merged-from | |
2924 | * record doesn't occupy that position. | |
2925 | * | |
2926 | * This creates a new empty extent so the caller | |
2927 | * should be smart enough to have removed any existing | |
2928 | * ones. | |
2929 | */ | |
2930 | if (index > 0) { | |
2931 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
2932 | size = index * sizeof(struct ocfs2_extent_rec); | |
2933 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
2934 | } | |
2935 | ||
2936 | /* | |
2937 | * Always memset - the caller doesn't check whether it | |
2938 | * created an empty extent, so there could be junk in | |
2939 | * the other fields. | |
2940 | */ | |
2941 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2942 | } | |
2943 | } | |
2944 | ||
677b9752 TM |
2945 | static int ocfs2_get_right_path(struct inode *inode, |
2946 | struct ocfs2_path *left_path, | |
2947 | struct ocfs2_path **ret_right_path) | |
2948 | { | |
2949 | int ret; | |
2950 | u32 right_cpos; | |
2951 | struct ocfs2_path *right_path = NULL; | |
2952 | struct ocfs2_extent_list *left_el; | |
2953 | ||
2954 | *ret_right_path = NULL; | |
2955 | ||
2956 | /* This function shouldn't be called for non-trees. */ | |
2957 | BUG_ON(left_path->p_tree_depth == 0); | |
2958 | ||
2959 | left_el = path_leaf_el(left_path); | |
2960 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
2961 | ||
2962 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path, | |
2963 | &right_cpos); | |
2964 | if (ret) { | |
2965 | mlog_errno(ret); | |
2966 | goto out; | |
2967 | } | |
2968 | ||
2969 | /* This function shouldn't be called for the rightmost leaf. */ | |
2970 | BUG_ON(right_cpos == 0); | |
2971 | ||
2972 | right_path = ocfs2_new_path(path_root_bh(left_path), | |
2973 | path_root_el(left_path)); | |
2974 | if (!right_path) { | |
2975 | ret = -ENOMEM; | |
2976 | mlog_errno(ret); | |
2977 | goto out; | |
2978 | } | |
2979 | ||
2980 | ret = ocfs2_find_path(inode, right_path, right_cpos); | |
2981 | if (ret) { | |
2982 | mlog_errno(ret); | |
2983 | goto out; | |
2984 | } | |
2985 | ||
2986 | *ret_right_path = right_path; | |
2987 | out: | |
2988 | if (ret) | |
2989 | ocfs2_free_path(right_path); | |
2990 | return ret; | |
2991 | } | |
2992 | ||
328d5752 MF |
2993 | /* |
2994 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
2995 | * onto the beginning of the record "next" to it. |
2996 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
2997 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
2998 | * next extent block. | |
328d5752 | 2999 | */ |
677b9752 TM |
3000 | static int ocfs2_merge_rec_right(struct inode *inode, |
3001 | struct ocfs2_path *left_path, | |
3002 | handle_t *handle, | |
3003 | struct ocfs2_extent_rec *split_rec, | |
3004 | int index) | |
328d5752 | 3005 | { |
677b9752 | 3006 | int ret, next_free, i; |
328d5752 MF |
3007 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3008 | struct ocfs2_extent_rec *left_rec; | |
3009 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3010 | struct ocfs2_extent_list *right_el; |
3011 | struct ocfs2_path *right_path = NULL; | |
3012 | int subtree_index = 0; | |
3013 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3014 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3015 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3016 | |
3017 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3018 | left_rec = &el->l_recs[index]; |
677b9752 | 3019 | |
9d8df6aa | 3020 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3021 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3022 | /* we meet with a cross extent block merge. */ | |
3023 | ret = ocfs2_get_right_path(inode, left_path, &right_path); | |
3024 | if (ret) { | |
3025 | mlog_errno(ret); | |
3026 | goto out; | |
3027 | } | |
3028 | ||
3029 | right_el = path_leaf_el(right_path); | |
3030 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3031 | BUG_ON(next_free <= 0); | |
3032 | right_rec = &right_el->l_recs[0]; | |
3033 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3034 | BUG_ON(next_free <= 1); |
677b9752 TM |
3035 | right_rec = &right_el->l_recs[1]; |
3036 | } | |
3037 | ||
3038 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3039 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3040 | le32_to_cpu(right_rec->e_cpos)); | |
3041 | ||
3042 | subtree_index = ocfs2_find_subtree_root(inode, | |
3043 | left_path, right_path); | |
3044 | ||
3045 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3046 | handle->h_buffer_credits, | |
3047 | right_path); | |
3048 | if (ret) { | |
3049 | mlog_errno(ret); | |
3050 | goto out; | |
3051 | } | |
3052 | ||
3053 | root_bh = left_path->p_node[subtree_index].bh; | |
3054 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3055 | ||
3056 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
3057 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3058 | if (ret) { | |
3059 | mlog_errno(ret); | |
3060 | goto out; | |
3061 | } | |
3062 | ||
3063 | for (i = subtree_index + 1; | |
3064 | i < path_num_items(right_path); i++) { | |
3065 | ret = ocfs2_journal_access(handle, inode, | |
3066 | right_path->p_node[i].bh, | |
3067 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3068 | if (ret) { | |
3069 | mlog_errno(ret); | |
3070 | goto out; | |
3071 | } | |
3072 | ||
3073 | ret = ocfs2_journal_access(handle, inode, | |
3074 | left_path->p_node[i].bh, | |
3075 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3076 | if (ret) { | |
3077 | mlog_errno(ret); | |
3078 | goto out; | |
3079 | } | |
3080 | } | |
3081 | ||
3082 | } else { | |
3083 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3084 | right_rec = &el->l_recs[index + 1]; | |
3085 | } | |
328d5752 MF |
3086 | |
3087 | ret = ocfs2_journal_access(handle, inode, bh, | |
3088 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3089 | if (ret) { | |
3090 | mlog_errno(ret); | |
3091 | goto out; | |
3092 | } | |
3093 | ||
3094 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3095 | ||
3096 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3097 | le64_add_cpu(&right_rec->e_blkno, | |
3098 | -ocfs2_clusters_to_blocks(inode->i_sb, split_clusters)); | |
3099 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); | |
3100 | ||
3101 | ocfs2_cleanup_merge(el, index); | |
3102 | ||
3103 | ret = ocfs2_journal_dirty(handle, bh); | |
3104 | if (ret) | |
3105 | mlog_errno(ret); | |
3106 | ||
677b9752 TM |
3107 | if (right_path) { |
3108 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
3109 | if (ret) | |
3110 | mlog_errno(ret); | |
3111 | ||
3112 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3113 | right_path, subtree_index); | |
3114 | } | |
3115 | out: | |
3116 | if (right_path) | |
3117 | ocfs2_free_path(right_path); | |
3118 | return ret; | |
3119 | } | |
3120 | ||
3121 | static int ocfs2_get_left_path(struct inode *inode, | |
3122 | struct ocfs2_path *right_path, | |
3123 | struct ocfs2_path **ret_left_path) | |
3124 | { | |
3125 | int ret; | |
3126 | u32 left_cpos; | |
3127 | struct ocfs2_path *left_path = NULL; | |
3128 | ||
3129 | *ret_left_path = NULL; | |
3130 | ||
3131 | /* This function shouldn't be called for non-trees. */ | |
3132 | BUG_ON(right_path->p_tree_depth == 0); | |
3133 | ||
3134 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
3135 | right_path, &left_cpos); | |
3136 | if (ret) { | |
3137 | mlog_errno(ret); | |
3138 | goto out; | |
3139 | } | |
3140 | ||
3141 | /* This function shouldn't be called for the leftmost leaf. */ | |
3142 | BUG_ON(left_cpos == 0); | |
3143 | ||
3144 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
3145 | path_root_el(right_path)); | |
3146 | if (!left_path) { | |
3147 | ret = -ENOMEM; | |
3148 | mlog_errno(ret); | |
3149 | goto out; | |
3150 | } | |
3151 | ||
3152 | ret = ocfs2_find_path(inode, left_path, left_cpos); | |
3153 | if (ret) { | |
3154 | mlog_errno(ret); | |
3155 | goto out; | |
3156 | } | |
3157 | ||
3158 | *ret_left_path = left_path; | |
328d5752 | 3159 | out: |
677b9752 TM |
3160 | if (ret) |
3161 | ocfs2_free_path(left_path); | |
328d5752 MF |
3162 | return ret; |
3163 | } | |
3164 | ||
3165 | /* | |
3166 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3167 | * onto the tail of the record "before" it. |
3168 | * For index > 0, the "before" means the extent rec at index - 1. | |
3169 | * | |
3170 | * For index == 0, the "before" means the last record of the previous | |
3171 | * extent block. And there is also a situation that we may need to | |
3172 | * remove the rightmost leaf extent block in the right_path and change | |
3173 | * the right path to indicate the new rightmost path. | |
328d5752 | 3174 | */ |
677b9752 TM |
3175 | static int ocfs2_merge_rec_left(struct inode *inode, |
3176 | struct ocfs2_path *right_path, | |
328d5752 MF |
3177 | handle_t *handle, |
3178 | struct ocfs2_extent_rec *split_rec, | |
677b9752 | 3179 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
e7d4cb6b | 3180 | struct ocfs2_extent_tree *et, |
677b9752 | 3181 | int index) |
328d5752 | 3182 | { |
677b9752 | 3183 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3184 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3185 | struct ocfs2_extent_rec *left_rec; | |
3186 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3187 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3188 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3189 | struct buffer_head *root_bh = NULL; | |
3190 | struct ocfs2_path *left_path = NULL; | |
3191 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3192 | |
677b9752 | 3193 | BUG_ON(index < 0); |
328d5752 | 3194 | |
328d5752 | 3195 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3196 | if (index == 0) { |
3197 | /* we meet with a cross extent block merge. */ | |
3198 | ret = ocfs2_get_left_path(inode, right_path, &left_path); | |
3199 | if (ret) { | |
3200 | mlog_errno(ret); | |
3201 | goto out; | |
3202 | } | |
3203 | ||
3204 | left_el = path_leaf_el(left_path); | |
3205 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3206 | le16_to_cpu(left_el->l_count)); | |
3207 | ||
3208 | left_rec = &left_el->l_recs[ | |
3209 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3210 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3211 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3212 | le32_to_cpu(split_rec->e_cpos)); | |
3213 | ||
3214 | subtree_index = ocfs2_find_subtree_root(inode, | |
3215 | left_path, right_path); | |
3216 | ||
3217 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3218 | handle->h_buffer_credits, | |
3219 | left_path); | |
3220 | if (ret) { | |
3221 | mlog_errno(ret); | |
3222 | goto out; | |
3223 | } | |
3224 | ||
3225 | root_bh = left_path->p_node[subtree_index].bh; | |
3226 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3227 | ||
3228 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
3229 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3230 | if (ret) { | |
3231 | mlog_errno(ret); | |
3232 | goto out; | |
3233 | } | |
3234 | ||
3235 | for (i = subtree_index + 1; | |
3236 | i < path_num_items(right_path); i++) { | |
3237 | ret = ocfs2_journal_access(handle, inode, | |
3238 | right_path->p_node[i].bh, | |
3239 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3240 | if (ret) { | |
3241 | mlog_errno(ret); | |
3242 | goto out; | |
3243 | } | |
3244 | ||
3245 | ret = ocfs2_journal_access(handle, inode, | |
3246 | left_path->p_node[i].bh, | |
3247 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3248 | if (ret) { | |
3249 | mlog_errno(ret); | |
3250 | goto out; | |
3251 | } | |
3252 | } | |
3253 | } else { | |
3254 | left_rec = &el->l_recs[index - 1]; | |
3255 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3256 | has_empty_extent = 1; | |
3257 | } | |
328d5752 MF |
3258 | |
3259 | ret = ocfs2_journal_access(handle, inode, bh, | |
3260 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3261 | if (ret) { | |
3262 | mlog_errno(ret); | |
3263 | goto out; | |
3264 | } | |
3265 | ||
3266 | if (has_empty_extent && index == 1) { | |
3267 | /* | |
3268 | * The easy case - we can just plop the record right in. | |
3269 | */ | |
3270 | *left_rec = *split_rec; | |
3271 | ||
3272 | has_empty_extent = 0; | |
677b9752 | 3273 | } else |
328d5752 | 3274 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3275 | |
3276 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3277 | le64_add_cpu(&right_rec->e_blkno, | |
3278 | ocfs2_clusters_to_blocks(inode->i_sb, split_clusters)); | |
3279 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); | |
3280 | ||
3281 | ocfs2_cleanup_merge(el, index); | |
3282 | ||
3283 | ret = ocfs2_journal_dirty(handle, bh); | |
3284 | if (ret) | |
3285 | mlog_errno(ret); | |
3286 | ||
677b9752 TM |
3287 | if (left_path) { |
3288 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
3289 | if (ret) | |
3290 | mlog_errno(ret); | |
3291 | ||
3292 | /* | |
3293 | * In the situation that the right_rec is empty and the extent | |
3294 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3295 | * it and we need to delete the right extent block. | |
3296 | */ | |
3297 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3298 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3299 | ||
3300 | ret = ocfs2_remove_rightmost_path(inode, handle, | |
e7d4cb6b TM |
3301 | right_path, |
3302 | dealloc, et); | |
677b9752 TM |
3303 | if (ret) { |
3304 | mlog_errno(ret); | |
3305 | goto out; | |
3306 | } | |
3307 | ||
3308 | /* Now the rightmost extent block has been deleted. | |
3309 | * So we use the new rightmost path. | |
3310 | */ | |
3311 | ocfs2_mv_path(right_path, left_path); | |
3312 | left_path = NULL; | |
3313 | } else | |
3314 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3315 | right_path, subtree_index); | |
3316 | } | |
328d5752 | 3317 | out: |
677b9752 TM |
3318 | if (left_path) |
3319 | ocfs2_free_path(left_path); | |
328d5752 MF |
3320 | return ret; |
3321 | } | |
3322 | ||
3323 | static int ocfs2_try_to_merge_extent(struct inode *inode, | |
3324 | handle_t *handle, | |
677b9752 | 3325 | struct ocfs2_path *path, |
328d5752 MF |
3326 | int split_index, |
3327 | struct ocfs2_extent_rec *split_rec, | |
3328 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
3329 | struct ocfs2_merge_ctxt *ctxt, |
3330 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
3331 | |
3332 | { | |
518d7269 | 3333 | int ret = 0; |
677b9752 | 3334 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3335 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3336 | ||
3337 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3338 | ||
518d7269 TM |
3339 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3340 | /* | |
3341 | * The merge code will need to create an empty | |
3342 | * extent to take the place of the newly | |
3343 | * emptied slot. Remove any pre-existing empty | |
3344 | * extents - having more than one in a leaf is | |
3345 | * illegal. | |
3346 | */ | |
677b9752 | 3347 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3348 | dealloc, et); |
518d7269 TM |
3349 | if (ret) { |
3350 | mlog_errno(ret); | |
3351 | goto out; | |
328d5752 | 3352 | } |
518d7269 TM |
3353 | split_index--; |
3354 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3355 | } |
3356 | ||
3357 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3358 | /* | |
3359 | * Left-right contig implies this. | |
3360 | */ | |
3361 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3362 | |
3363 | /* | |
3364 | * Since the leftright insert always covers the entire | |
3365 | * extent, this call will delete the insert record | |
3366 | * entirely, resulting in an empty extent record added to | |
3367 | * the extent block. | |
3368 | * | |
3369 | * Since the adding of an empty extent shifts | |
3370 | * everything back to the right, there's no need to | |
3371 | * update split_index here. | |
677b9752 TM |
3372 | * |
3373 | * When the split_index is zero, we need to merge it to the | |
3374 | * prevoius extent block. It is more efficient and easier | |
3375 | * if we do merge_right first and merge_left later. | |
328d5752 | 3376 | */ |
677b9752 TM |
3377 | ret = ocfs2_merge_rec_right(inode, path, |
3378 | handle, split_rec, | |
3379 | split_index); | |
328d5752 MF |
3380 | if (ret) { |
3381 | mlog_errno(ret); | |
3382 | goto out; | |
3383 | } | |
3384 | ||
3385 | /* | |
3386 | * We can only get this from logic error above. | |
3387 | */ | |
3388 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3389 | ||
677b9752 | 3390 | /* The merge left us with an empty extent, remove it. */ |
e7d4cb6b TM |
3391 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
3392 | dealloc, et); | |
328d5752 MF |
3393 | if (ret) { |
3394 | mlog_errno(ret); | |
3395 | goto out; | |
3396 | } | |
677b9752 | 3397 | |
328d5752 MF |
3398 | rec = &el->l_recs[split_index]; |
3399 | ||
3400 | /* | |
3401 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3402 | * we've merged it into the rec already. |
328d5752 | 3403 | */ |
677b9752 TM |
3404 | ret = ocfs2_merge_rec_left(inode, path, |
3405 | handle, rec, | |
e7d4cb6b | 3406 | dealloc, et, |
677b9752 TM |
3407 | split_index); |
3408 | ||
328d5752 MF |
3409 | if (ret) { |
3410 | mlog_errno(ret); | |
3411 | goto out; | |
3412 | } | |
3413 | ||
677b9752 | 3414 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3415 | dealloc, et); |
328d5752 MF |
3416 | /* |
3417 | * Error from this last rotate is not critical, so | |
3418 | * print but don't bubble it up. | |
3419 | */ | |
3420 | if (ret) | |
3421 | mlog_errno(ret); | |
3422 | ret = 0; | |
3423 | } else { | |
3424 | /* | |
3425 | * Merge a record to the left or right. | |
3426 | * | |
3427 | * 'contig_type' is relative to the existing record, | |
3428 | * so for example, if we're "right contig", it's to | |
3429 | * the record on the left (hence the left merge). | |
3430 | */ | |
3431 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
3432 | ret = ocfs2_merge_rec_left(inode, | |
677b9752 TM |
3433 | path, |
3434 | handle, split_rec, | |
e7d4cb6b | 3435 | dealloc, et, |
328d5752 MF |
3436 | split_index); |
3437 | if (ret) { | |
3438 | mlog_errno(ret); | |
3439 | goto out; | |
3440 | } | |
3441 | } else { | |
3442 | ret = ocfs2_merge_rec_right(inode, | |
677b9752 TM |
3443 | path, |
3444 | handle, split_rec, | |
328d5752 MF |
3445 | split_index); |
3446 | if (ret) { | |
3447 | mlog_errno(ret); | |
3448 | goto out; | |
3449 | } | |
3450 | } | |
3451 | ||
3452 | if (ctxt->c_split_covers_rec) { | |
3453 | /* | |
3454 | * The merge may have left an empty extent in | |
3455 | * our leaf. Try to rotate it away. | |
3456 | */ | |
677b9752 | 3457 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3458 | dealloc, et); |
328d5752 MF |
3459 | if (ret) |
3460 | mlog_errno(ret); | |
3461 | ret = 0; | |
3462 | } | |
3463 | } | |
3464 | ||
3465 | out: | |
3466 | return ret; | |
3467 | } | |
3468 | ||
3469 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3470 | enum ocfs2_split_type split, | |
3471 | struct ocfs2_extent_rec *rec, | |
3472 | struct ocfs2_extent_rec *split_rec) | |
3473 | { | |
3474 | u64 len_blocks; | |
3475 | ||
3476 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3477 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3478 | ||
3479 | if (split == SPLIT_LEFT) { | |
3480 | /* | |
3481 | * Region is on the left edge of the existing | |
3482 | * record. | |
3483 | */ | |
3484 | le32_add_cpu(&rec->e_cpos, | |
3485 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3486 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3487 | le16_add_cpu(&rec->e_leaf_clusters, | |
3488 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3489 | } else { | |
3490 | /* | |
3491 | * Region is on the right edge of the existing | |
3492 | * record. | |
3493 | */ | |
3494 | le16_add_cpu(&rec->e_leaf_clusters, | |
3495 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3496 | } | |
3497 | } | |
3498 | ||
3499 | /* | |
3500 | * Do the final bits of extent record insertion at the target leaf | |
3501 | * list. If this leaf is part of an allocation tree, it is assumed | |
3502 | * that the tree above has been prepared. | |
3503 | */ | |
3504 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec, | |
3505 | struct ocfs2_extent_list *el, | |
3506 | struct ocfs2_insert_type *insert, | |
3507 | struct inode *inode) | |
3508 | { | |
3509 | int i = insert->ins_contig_index; | |
3510 | unsigned int range; | |
3511 | struct ocfs2_extent_rec *rec; | |
3512 | ||
3513 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3514 | ||
3515 | if (insert->ins_split != SPLIT_NONE) { | |
3516 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3517 | BUG_ON(i == -1); | |
3518 | rec = &el->l_recs[i]; | |
3519 | ocfs2_subtract_from_rec(inode->i_sb, insert->ins_split, rec, | |
3520 | insert_rec); | |
3521 | goto rotate; | |
3522 | } | |
3523 | ||
3524 | /* | |
3525 | * Contiguous insert - either left or right. | |
3526 | */ | |
3527 | if (insert->ins_contig != CONTIG_NONE) { | |
3528 | rec = &el->l_recs[i]; | |
3529 | if (insert->ins_contig == CONTIG_LEFT) { | |
3530 | rec->e_blkno = insert_rec->e_blkno; | |
3531 | rec->e_cpos = insert_rec->e_cpos; | |
3532 | } | |
3533 | le16_add_cpu(&rec->e_leaf_clusters, | |
3534 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3535 | return; | |
3536 | } | |
3537 | ||
3538 | /* | |
3539 | * Handle insert into an empty leaf. | |
3540 | */ | |
3541 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3542 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3543 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3544 | el->l_recs[0] = *insert_rec; | |
3545 | el->l_next_free_rec = cpu_to_le16(1); | |
3546 | return; | |
3547 | } | |
3548 | ||
3549 | /* | |
3550 | * Appending insert. | |
3551 | */ | |
3552 | if (insert->ins_appending == APPEND_TAIL) { | |
3553 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3554 | rec = &el->l_recs[i]; | |
3555 | range = le32_to_cpu(rec->e_cpos) | |
3556 | + le16_to_cpu(rec->e_leaf_clusters); | |
3557 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3558 | ||
3559 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3560 | le16_to_cpu(el->l_count), | |
3561 | "inode %lu, depth %u, count %u, next free %u, " | |
3562 | "rec.cpos %u, rec.clusters %u, " | |
3563 | "insert.cpos %u, insert.clusters %u\n", | |
3564 | inode->i_ino, | |
3565 | le16_to_cpu(el->l_tree_depth), | |
3566 | le16_to_cpu(el->l_count), | |
3567 | le16_to_cpu(el->l_next_free_rec), | |
3568 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3569 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3570 | le32_to_cpu(insert_rec->e_cpos), | |
3571 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3572 | i++; | |
3573 | el->l_recs[i] = *insert_rec; | |
3574 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3575 | return; | |
3576 | } | |
3577 | ||
3578 | rotate: | |
3579 | /* | |
3580 | * Ok, we have to rotate. | |
3581 | * | |
3582 | * At this point, it is safe to assume that inserting into an | |
3583 | * empty leaf and appending to a leaf have both been handled | |
3584 | * above. | |
3585 | * | |
3586 | * This leaf needs to have space, either by the empty 1st | |
3587 | * extent record, or by virtue of an l_next_rec < l_count. | |
3588 | */ | |
3589 | ocfs2_rotate_leaf(el, insert_rec); | |
3590 | } | |
3591 | ||
328d5752 MF |
3592 | static void ocfs2_adjust_rightmost_records(struct inode *inode, |
3593 | handle_t *handle, | |
3594 | struct ocfs2_path *path, | |
3595 | struct ocfs2_extent_rec *insert_rec) | |
3596 | { | |
3597 | int ret, i, next_free; | |
3598 | struct buffer_head *bh; | |
3599 | struct ocfs2_extent_list *el; | |
3600 | struct ocfs2_extent_rec *rec; | |
3601 | ||
3602 | /* | |
3603 | * Update everything except the leaf block. | |
3604 | */ | |
3605 | for (i = 0; i < path->p_tree_depth; i++) { | |
3606 | bh = path->p_node[i].bh; | |
3607 | el = path->p_node[i].el; | |
3608 | ||
dcd0538f MF |
3609 | next_free = le16_to_cpu(el->l_next_free_rec); |
3610 | if (next_free == 0) { | |
3611 | ocfs2_error(inode->i_sb, | |
3612 | "Dinode %llu has a bad extent list", | |
3613 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
3614 | ret = -EIO; | |
328d5752 MF |
3615 | return; |
3616 | } | |
3617 | ||
3618 | rec = &el->l_recs[next_free - 1]; | |
3619 | ||
3620 | rec->e_int_clusters = insert_rec->e_cpos; | |
3621 | le32_add_cpu(&rec->e_int_clusters, | |
3622 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3623 | le32_add_cpu(&rec->e_int_clusters, | |
3624 | -le32_to_cpu(rec->e_cpos)); | |
3625 | ||
3626 | ret = ocfs2_journal_dirty(handle, bh); | |
3627 | if (ret) | |
3628 | mlog_errno(ret); | |
3629 | ||
3630 | } | |
3631 | } | |
3632 | ||
3633 | static int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle, | |
3634 | struct ocfs2_extent_rec *insert_rec, | |
3635 | struct ocfs2_path *right_path, | |
3636 | struct ocfs2_path **ret_left_path) | |
3637 | { | |
3638 | int ret, next_free; | |
3639 | struct ocfs2_extent_list *el; | |
3640 | struct ocfs2_path *left_path = NULL; | |
3641 | ||
3642 | *ret_left_path = NULL; | |
3643 | ||
3644 | /* | |
3645 | * This shouldn't happen for non-trees. The extent rec cluster | |
3646 | * count manipulation below only works for interior nodes. | |
3647 | */ | |
3648 | BUG_ON(right_path->p_tree_depth == 0); | |
3649 | ||
3650 | /* | |
3651 | * If our appending insert is at the leftmost edge of a leaf, | |
3652 | * then we might need to update the rightmost records of the | |
3653 | * neighboring path. | |
3654 | */ | |
3655 | el = path_leaf_el(right_path); | |
3656 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3657 | if (next_free == 0 || | |
3658 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3659 | u32 left_cpos; | |
3660 | ||
3661 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, | |
3662 | &left_cpos); | |
3663 | if (ret) { | |
3664 | mlog_errno(ret); | |
dcd0538f MF |
3665 | goto out; |
3666 | } | |
3667 | ||
328d5752 MF |
3668 | mlog(0, "Append may need a left path update. cpos: %u, " |
3669 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3670 | left_cpos); | |
e48edee2 | 3671 | |
328d5752 MF |
3672 | /* |
3673 | * No need to worry if the append is already in the | |
3674 | * leftmost leaf. | |
3675 | */ | |
3676 | if (left_cpos) { | |
3677 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
3678 | path_root_el(right_path)); | |
3679 | if (!left_path) { | |
3680 | ret = -ENOMEM; | |
3681 | mlog_errno(ret); | |
3682 | goto out; | |
3683 | } | |
dcd0538f | 3684 | |
328d5752 MF |
3685 | ret = ocfs2_find_path(inode, left_path, left_cpos); |
3686 | if (ret) { | |
3687 | mlog_errno(ret); | |
3688 | goto out; | |
3689 | } | |
dcd0538f | 3690 | |
328d5752 MF |
3691 | /* |
3692 | * ocfs2_insert_path() will pass the left_path to the | |
3693 | * journal for us. | |
3694 | */ | |
3695 | } | |
3696 | } | |
dcd0538f | 3697 | |
328d5752 MF |
3698 | ret = ocfs2_journal_access_path(inode, handle, right_path); |
3699 | if (ret) { | |
3700 | mlog_errno(ret); | |
3701 | goto out; | |
dcd0538f MF |
3702 | } |
3703 | ||
328d5752 MF |
3704 | ocfs2_adjust_rightmost_records(inode, handle, right_path, insert_rec); |
3705 | ||
dcd0538f MF |
3706 | *ret_left_path = left_path; |
3707 | ret = 0; | |
3708 | out: | |
3709 | if (ret != 0) | |
3710 | ocfs2_free_path(left_path); | |
3711 | ||
3712 | return ret; | |
3713 | } | |
3714 | ||
328d5752 MF |
3715 | static void ocfs2_split_record(struct inode *inode, |
3716 | struct ocfs2_path *left_path, | |
3717 | struct ocfs2_path *right_path, | |
3718 | struct ocfs2_extent_rec *split_rec, | |
3719 | enum ocfs2_split_type split) | |
3720 | { | |
3721 | int index; | |
3722 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
3723 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
3724 | struct ocfs2_extent_rec *rec, *tmprec; | |
3725 | ||
3726 | right_el = path_leaf_el(right_path);; | |
3727 | if (left_path) | |
3728 | left_el = path_leaf_el(left_path); | |
3729 | ||
3730 | el = right_el; | |
3731 | insert_el = right_el; | |
3732 | index = ocfs2_search_extent_list(el, cpos); | |
3733 | if (index != -1) { | |
3734 | if (index == 0 && left_path) { | |
3735 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3736 | ||
3737 | /* | |
3738 | * This typically means that the record | |
3739 | * started in the left path but moved to the | |
3740 | * right as a result of rotation. We either | |
3741 | * move the existing record to the left, or we | |
3742 | * do the later insert there. | |
3743 | * | |
3744 | * In this case, the left path should always | |
3745 | * exist as the rotate code will have passed | |
3746 | * it back for a post-insert update. | |
3747 | */ | |
3748 | ||
3749 | if (split == SPLIT_LEFT) { | |
3750 | /* | |
3751 | * It's a left split. Since we know | |
3752 | * that the rotate code gave us an | |
3753 | * empty extent in the left path, we | |
3754 | * can just do the insert there. | |
3755 | */ | |
3756 | insert_el = left_el; | |
3757 | } else { | |
3758 | /* | |
3759 | * Right split - we have to move the | |
3760 | * existing record over to the left | |
3761 | * leaf. The insert will be into the | |
3762 | * newly created empty extent in the | |
3763 | * right leaf. | |
3764 | */ | |
3765 | tmprec = &right_el->l_recs[index]; | |
3766 | ocfs2_rotate_leaf(left_el, tmprec); | |
3767 | el = left_el; | |
3768 | ||
3769 | memset(tmprec, 0, sizeof(*tmprec)); | |
3770 | index = ocfs2_search_extent_list(left_el, cpos); | |
3771 | BUG_ON(index == -1); | |
3772 | } | |
3773 | } | |
3774 | } else { | |
3775 | BUG_ON(!left_path); | |
3776 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
3777 | /* | |
3778 | * Left path is easy - we can just allow the insert to | |
3779 | * happen. | |
3780 | */ | |
3781 | el = left_el; | |
3782 | insert_el = left_el; | |
3783 | index = ocfs2_search_extent_list(el, cpos); | |
3784 | BUG_ON(index == -1); | |
3785 | } | |
3786 | ||
3787 | rec = &el->l_recs[index]; | |
3788 | ocfs2_subtract_from_rec(inode->i_sb, split, rec, split_rec); | |
3789 | ocfs2_rotate_leaf(insert_el, split_rec); | |
3790 | } | |
3791 | ||
dcd0538f | 3792 | /* |
e7d4cb6b TM |
3793 | * This function only does inserts on an allocation b-tree. For tree |
3794 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
3795 | * |
3796 | * right_path is the path we want to do the actual insert | |
3797 | * in. left_path should only be passed in if we need to update that | |
3798 | * portion of the tree after an edge insert. | |
3799 | */ | |
3800 | static int ocfs2_insert_path(struct inode *inode, | |
3801 | handle_t *handle, | |
3802 | struct ocfs2_path *left_path, | |
3803 | struct ocfs2_path *right_path, | |
3804 | struct ocfs2_extent_rec *insert_rec, | |
3805 | struct ocfs2_insert_type *insert) | |
3806 | { | |
3807 | int ret, subtree_index; | |
3808 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 3809 | |
dcd0538f MF |
3810 | if (left_path) { |
3811 | int credits = handle->h_buffer_credits; | |
3812 | ||
3813 | /* | |
3814 | * There's a chance that left_path got passed back to | |
3815 | * us without being accounted for in the | |
3816 | * journal. Extend our transaction here to be sure we | |
3817 | * can change those blocks. | |
3818 | */ | |
3819 | credits += left_path->p_tree_depth; | |
3820 | ||
3821 | ret = ocfs2_extend_trans(handle, credits); | |
3822 | if (ret < 0) { | |
3823 | mlog_errno(ret); | |
3824 | goto out; | |
3825 | } | |
3826 | ||
3827 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
3828 | if (ret < 0) { | |
3829 | mlog_errno(ret); | |
3830 | goto out; | |
3831 | } | |
3832 | } | |
3833 | ||
e8aed345 MF |
3834 | /* |
3835 | * Pass both paths to the journal. The majority of inserts | |
3836 | * will be touching all components anyway. | |
3837 | */ | |
3838 | ret = ocfs2_journal_access_path(inode, handle, right_path); | |
3839 | if (ret < 0) { | |
3840 | mlog_errno(ret); | |
3841 | goto out; | |
3842 | } | |
3843 | ||
328d5752 MF |
3844 | if (insert->ins_split != SPLIT_NONE) { |
3845 | /* | |
3846 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 3847 | * of splits, but it's easier to just let one separate |
328d5752 MF |
3848 | * function sort it all out. |
3849 | */ | |
3850 | ocfs2_split_record(inode, left_path, right_path, | |
3851 | insert_rec, insert->ins_split); | |
e8aed345 MF |
3852 | |
3853 | /* | |
3854 | * Split might have modified either leaf and we don't | |
3855 | * have a guarantee that the later edge insert will | |
3856 | * dirty this for us. | |
3857 | */ | |
3858 | if (left_path) | |
3859 | ret = ocfs2_journal_dirty(handle, | |
3860 | path_leaf_bh(left_path)); | |
3861 | if (ret) | |
3862 | mlog_errno(ret); | |
328d5752 MF |
3863 | } else |
3864 | ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path), | |
3865 | insert, inode); | |
dcd0538f | 3866 | |
dcd0538f MF |
3867 | ret = ocfs2_journal_dirty(handle, leaf_bh); |
3868 | if (ret) | |
3869 | mlog_errno(ret); | |
3870 | ||
3871 | if (left_path) { | |
3872 | /* | |
3873 | * The rotate code has indicated that we need to fix | |
3874 | * up portions of the tree after the insert. | |
3875 | * | |
3876 | * XXX: Should we extend the transaction here? | |
3877 | */ | |
3878 | subtree_index = ocfs2_find_subtree_root(inode, left_path, | |
3879 | right_path); | |
3880 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3881 | right_path, subtree_index); | |
3882 | } | |
3883 | ||
3884 | ret = 0; | |
3885 | out: | |
3886 | return ret; | |
3887 | } | |
3888 | ||
3889 | static int ocfs2_do_insert_extent(struct inode *inode, | |
3890 | handle_t *handle, | |
e7d4cb6b | 3891 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
3892 | struct ocfs2_extent_rec *insert_rec, |
3893 | struct ocfs2_insert_type *type) | |
3894 | { | |
3895 | int ret, rotate = 0; | |
3896 | u32 cpos; | |
3897 | struct ocfs2_path *right_path = NULL; | |
3898 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
3899 | struct ocfs2_extent_list *el; |
3900 | ||
e7d4cb6b | 3901 | el = et->root_el; |
dcd0538f | 3902 | |
e7d4cb6b | 3903 | ret = ocfs2_journal_access(handle, inode, et->root_bh, |
dcd0538f MF |
3904 | OCFS2_JOURNAL_ACCESS_WRITE); |
3905 | if (ret) { | |
3906 | mlog_errno(ret); | |
3907 | goto out; | |
3908 | } | |
3909 | ||
3910 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
3911 | ocfs2_insert_at_leaf(insert_rec, el, type, inode); | |
3912 | goto out_update_clusters; | |
3913 | } | |
3914 | ||
e7d4cb6b | 3915 | right_path = ocfs2_new_path(et->root_bh, et->root_el); |
dcd0538f MF |
3916 | if (!right_path) { |
3917 | ret = -ENOMEM; | |
3918 | mlog_errno(ret); | |
3919 | goto out; | |
3920 | } | |
3921 | ||
3922 | /* | |
3923 | * Determine the path to start with. Rotations need the | |
3924 | * rightmost path, everything else can go directly to the | |
3925 | * target leaf. | |
3926 | */ | |
3927 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
3928 | if (type->ins_appending == APPEND_NONE && | |
3929 | type->ins_contig == CONTIG_NONE) { | |
3930 | rotate = 1; | |
3931 | cpos = UINT_MAX; | |
3932 | } | |
3933 | ||
3934 | ret = ocfs2_find_path(inode, right_path, cpos); | |
3935 | if (ret) { | |
3936 | mlog_errno(ret); | |
3937 | goto out; | |
3938 | } | |
3939 | ||
3940 | /* | |
3941 | * Rotations and appends need special treatment - they modify | |
3942 | * parts of the tree's above them. | |
3943 | * | |
3944 | * Both might pass back a path immediate to the left of the | |
3945 | * one being inserted to. This will be cause | |
3946 | * ocfs2_insert_path() to modify the rightmost records of | |
3947 | * left_path to account for an edge insert. | |
3948 | * | |
3949 | * XXX: When modifying this code, keep in mind that an insert | |
3950 | * can wind up skipping both of these two special cases... | |
3951 | */ | |
3952 | if (rotate) { | |
328d5752 | 3953 | ret = ocfs2_rotate_tree_right(inode, handle, type->ins_split, |
dcd0538f MF |
3954 | le32_to_cpu(insert_rec->e_cpos), |
3955 | right_path, &left_path); | |
3956 | if (ret) { | |
3957 | mlog_errno(ret); | |
3958 | goto out; | |
3959 | } | |
e8aed345 MF |
3960 | |
3961 | /* | |
3962 | * ocfs2_rotate_tree_right() might have extended the | |
3963 | * transaction without re-journaling our tree root. | |
3964 | */ | |
e7d4cb6b | 3965 | ret = ocfs2_journal_access(handle, inode, et->root_bh, |
e8aed345 MF |
3966 | OCFS2_JOURNAL_ACCESS_WRITE); |
3967 | if (ret) { | |
3968 | mlog_errno(ret); | |
3969 | goto out; | |
3970 | } | |
dcd0538f MF |
3971 | } else if (type->ins_appending == APPEND_TAIL |
3972 | && type->ins_contig != CONTIG_LEFT) { | |
3973 | ret = ocfs2_append_rec_to_path(inode, handle, insert_rec, | |
3974 | right_path, &left_path); | |
3975 | if (ret) { | |
3976 | mlog_errno(ret); | |
3977 | goto out; | |
3978 | } | |
3979 | } | |
3980 | ||
3981 | ret = ocfs2_insert_path(inode, handle, left_path, right_path, | |
3982 | insert_rec, type); | |
3983 | if (ret) { | |
3984 | mlog_errno(ret); | |
3985 | goto out; | |
3986 | } | |
3987 | ||
3988 | out_update_clusters: | |
328d5752 | 3989 | if (type->ins_split == SPLIT_NONE) |
35dc0aa3 JB |
3990 | ocfs2_et_update_clusters(inode, et, |
3991 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
dcd0538f | 3992 | |
e7d4cb6b | 3993 | ret = ocfs2_journal_dirty(handle, et->root_bh); |
dcd0538f MF |
3994 | if (ret) |
3995 | mlog_errno(ret); | |
3996 | ||
3997 | out: | |
3998 | ocfs2_free_path(left_path); | |
3999 | ocfs2_free_path(right_path); | |
4000 | ||
4001 | return ret; | |
4002 | } | |
4003 | ||
328d5752 | 4004 | static enum ocfs2_contig_type |
ad5a4d70 | 4005 | ocfs2_figure_merge_contig_type(struct inode *inode, struct ocfs2_path *path, |
328d5752 MF |
4006 | struct ocfs2_extent_list *el, int index, |
4007 | struct ocfs2_extent_rec *split_rec) | |
4008 | { | |
ad5a4d70 | 4009 | int status; |
328d5752 | 4010 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4011 | u32 left_cpos, right_cpos; |
4012 | struct ocfs2_extent_rec *rec = NULL; | |
4013 | struct ocfs2_extent_list *new_el; | |
4014 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4015 | struct buffer_head *bh; | |
4016 | struct ocfs2_extent_block *eb; | |
4017 | ||
4018 | if (index > 0) { | |
4019 | rec = &el->l_recs[index - 1]; | |
4020 | } else if (path->p_tree_depth > 0) { | |
4021 | status = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
4022 | path, &left_cpos); | |
4023 | if (status) | |
4024 | goto out; | |
4025 | ||
4026 | if (left_cpos != 0) { | |
4027 | left_path = ocfs2_new_path(path_root_bh(path), | |
4028 | path_root_el(path)); | |
4029 | if (!left_path) | |
4030 | goto out; | |
4031 | ||
4032 | status = ocfs2_find_path(inode, left_path, left_cpos); | |
4033 | if (status) | |
4034 | goto out; | |
4035 | ||
4036 | new_el = path_leaf_el(left_path); | |
4037 | ||
4038 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4039 | le16_to_cpu(new_el->l_count)) { | |
4040 | bh = path_leaf_bh(left_path); | |
4041 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
4042 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, | |
4043 | eb); | |
4044 | goto out; | |
4045 | } | |
4046 | rec = &new_el->l_recs[ | |
4047 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4048 | } | |
4049 | } | |
328d5752 MF |
4050 | |
4051 | /* | |
4052 | * We're careful to check for an empty extent record here - | |
4053 | * the merge code will know what to do if it sees one. | |
4054 | */ | |
ad5a4d70 | 4055 | if (rec) { |
328d5752 MF |
4056 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4057 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4058 | ret = CONTIG_RIGHT; | |
4059 | } else { | |
4060 | ret = ocfs2_extent_contig(inode, rec, split_rec); | |
4061 | } | |
4062 | } | |
4063 | ||
ad5a4d70 TM |
4064 | rec = NULL; |
4065 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4066 | rec = &el->l_recs[index + 1]; | |
4067 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4068 | path->p_tree_depth > 0) { | |
4069 | status = ocfs2_find_cpos_for_right_leaf(inode->i_sb, | |
4070 | path, &right_cpos); | |
4071 | if (status) | |
4072 | goto out; | |
4073 | ||
4074 | if (right_cpos == 0) | |
4075 | goto out; | |
4076 | ||
4077 | right_path = ocfs2_new_path(path_root_bh(path), | |
4078 | path_root_el(path)); | |
4079 | if (!right_path) | |
4080 | goto out; | |
4081 | ||
4082 | status = ocfs2_find_path(inode, right_path, right_cpos); | |
4083 | if (status) | |
4084 | goto out; | |
4085 | ||
4086 | new_el = path_leaf_el(right_path); | |
4087 | rec = &new_el->l_recs[0]; | |
4088 | if (ocfs2_is_empty_extent(rec)) { | |
4089 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4090 | bh = path_leaf_bh(right_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[1]; | |
4097 | } | |
4098 | } | |
4099 | ||
4100 | if (rec) { | |
328d5752 MF |
4101 | enum ocfs2_contig_type contig_type; |
4102 | ||
328d5752 MF |
4103 | contig_type = ocfs2_extent_contig(inode, rec, split_rec); |
4104 | ||
4105 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4106 | ret = CONTIG_LEFTRIGHT; | |
4107 | else if (ret == CONTIG_NONE) | |
4108 | ret = contig_type; | |
4109 | } | |
4110 | ||
ad5a4d70 TM |
4111 | out: |
4112 | if (left_path) | |
4113 | ocfs2_free_path(left_path); | |
4114 | if (right_path) | |
4115 | ocfs2_free_path(right_path); | |
4116 | ||
328d5752 MF |
4117 | return ret; |
4118 | } | |
4119 | ||
dcd0538f MF |
4120 | static void ocfs2_figure_contig_type(struct inode *inode, |
4121 | struct ocfs2_insert_type *insert, | |
4122 | struct ocfs2_extent_list *el, | |
ca12b7c4 TM |
4123 | struct ocfs2_extent_rec *insert_rec, |
4124 | struct ocfs2_extent_tree *et) | |
dcd0538f MF |
4125 | { |
4126 | int i; | |
4127 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4128 | ||
e48edee2 MF |
4129 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4130 | ||
dcd0538f MF |
4131 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
4132 | contig_type = ocfs2_extent_contig(inode, &el->l_recs[i], | |
4133 | insert_rec); | |
4134 | if (contig_type != CONTIG_NONE) { | |
4135 | insert->ins_contig_index = i; | |
4136 | break; | |
4137 | } | |
4138 | } | |
4139 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4140 | |
4141 | if (insert->ins_contig != CONTIG_NONE) { | |
4142 | struct ocfs2_extent_rec *rec = | |
4143 | &el->l_recs[insert->ins_contig_index]; | |
4144 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4145 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4146 | ||
4147 | /* | |
4148 | * Caller might want us to limit the size of extents, don't | |
4149 | * calculate contiguousness if we might exceed that limit. | |
4150 | */ | |
4151 | if (et->max_leaf_clusters && len > et->max_leaf_clusters) | |
4152 | insert->ins_contig = CONTIG_NONE; | |
4153 | } | |
dcd0538f MF |
4154 | } |
4155 | ||
4156 | /* | |
4157 | * This should only be called against the righmost leaf extent list. | |
4158 | * | |
4159 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4160 | * insert at the tail of the rightmost leaf. | |
4161 | * | |
e7d4cb6b TM |
4162 | * This should also work against the root extent list for tree's with 0 |
4163 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4164 | * then the logic here makes sense. |
4165 | */ | |
4166 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4167 | struct ocfs2_extent_list *el, | |
4168 | struct ocfs2_extent_rec *insert_rec) | |
4169 | { | |
4170 | int i; | |
4171 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4172 | struct ocfs2_extent_rec *rec; | |
4173 | ||
4174 | insert->ins_appending = APPEND_NONE; | |
4175 | ||
e48edee2 | 4176 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4177 | |
4178 | if (!el->l_next_free_rec) | |
4179 | goto set_tail_append; | |
4180 | ||
4181 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4182 | /* Were all records empty? */ | |
4183 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4184 | goto set_tail_append; | |
4185 | } | |
4186 | ||
4187 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4188 | rec = &el->l_recs[i]; | |
4189 | ||
e48edee2 MF |
4190 | if (cpos >= |
4191 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4192 | goto set_tail_append; |
4193 | ||
4194 | return; | |
4195 | ||
4196 | set_tail_append: | |
4197 | insert->ins_appending = APPEND_TAIL; | |
4198 | } | |
4199 | ||
4200 | /* | |
4201 | * Helper function called at the begining of an insert. | |
4202 | * | |
4203 | * This computes a few things that are commonly used in the process of | |
4204 | * inserting into the btree: | |
4205 | * - Whether the new extent is contiguous with an existing one. | |
4206 | * - The current tree depth. | |
4207 | * - Whether the insert is an appending one. | |
4208 | * - The total # of free records in the tree. | |
4209 | * | |
4210 | * All of the information is stored on the ocfs2_insert_type | |
4211 | * structure. | |
4212 | */ | |
4213 | static int ocfs2_figure_insert_type(struct inode *inode, | |
e7d4cb6b | 4214 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4215 | struct buffer_head **last_eb_bh, |
4216 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4217 | int *free_records, |
dcd0538f MF |
4218 | struct ocfs2_insert_type *insert) |
4219 | { | |
4220 | int ret; | |
dcd0538f MF |
4221 | struct ocfs2_extent_block *eb; |
4222 | struct ocfs2_extent_list *el; | |
4223 | struct ocfs2_path *path = NULL; | |
4224 | struct buffer_head *bh = NULL; | |
4225 | ||
328d5752 MF |
4226 | insert->ins_split = SPLIT_NONE; |
4227 | ||
e7d4cb6b | 4228 | el = et->root_el; |
dcd0538f MF |
4229 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4230 | ||
4231 | if (el->l_tree_depth) { | |
4232 | /* | |
4233 | * If we have tree depth, we read in the | |
4234 | * rightmost extent block ahead of time as | |
4235 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4236 | * may want it later. | |
4237 | */ | |
4238 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
35dc0aa3 | 4239 | ocfs2_et_get_last_eb_blk(et), &bh, |
dcd0538f MF |
4240 | OCFS2_BH_CACHED, inode); |
4241 | if (ret) { | |
4242 | mlog_exit(ret); | |
4243 | goto out; | |
4244 | } | |
ccd979bd | 4245 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4246 | el = &eb->h_list; |
dcd0538f | 4247 | } |
ccd979bd | 4248 | |
dcd0538f MF |
4249 | /* |
4250 | * Unless we have a contiguous insert, we'll need to know if | |
4251 | * there is room left in our allocation tree for another | |
4252 | * extent record. | |
4253 | * | |
4254 | * XXX: This test is simplistic, we can search for empty | |
4255 | * extent records too. | |
4256 | */ | |
c77534f6 | 4257 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4258 | le16_to_cpu(el->l_next_free_rec); |
4259 | ||
4260 | if (!insert->ins_tree_depth) { | |
ca12b7c4 | 4261 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4262 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4263 | return 0; | |
ccd979bd MF |
4264 | } |
4265 | ||
e7d4cb6b | 4266 | path = ocfs2_new_path(et->root_bh, et->root_el); |
dcd0538f MF |
4267 | if (!path) { |
4268 | ret = -ENOMEM; | |
4269 | mlog_errno(ret); | |
4270 | goto out; | |
4271 | } | |
ccd979bd | 4272 | |
dcd0538f MF |
4273 | /* |
4274 | * In the case that we're inserting past what the tree | |
4275 | * currently accounts for, ocfs2_find_path() will return for | |
4276 | * us the rightmost tree path. This is accounted for below in | |
4277 | * the appending code. | |
4278 | */ | |
4279 | ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos)); | |
4280 | if (ret) { | |
4281 | mlog_errno(ret); | |
4282 | goto out; | |
4283 | } | |
ccd979bd | 4284 | |
dcd0538f MF |
4285 | el = path_leaf_el(path); |
4286 | ||
4287 | /* | |
4288 | * Now that we have the path, there's two things we want to determine: | |
4289 | * 1) Contiguousness (also set contig_index if this is so) | |
4290 | * | |
4291 | * 2) Are we doing an append? We can trivially break this up | |
4292 | * into two types of appends: simple record append, or a | |
4293 | * rotate inside the tail leaf. | |
4294 | */ | |
ca12b7c4 | 4295 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4296 | |
4297 | /* | |
4298 | * The insert code isn't quite ready to deal with all cases of | |
4299 | * left contiguousness. Specifically, if it's an insert into | |
4300 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4301 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4302 | * left. For now, just catch that case and fool the layers |
4303 | * above us. This works just fine for tree_depth == 0, which | |
4304 | * is why we allow that above. | |
4305 | */ | |
4306 | if (insert->ins_contig == CONTIG_LEFT && | |
4307 | insert->ins_contig_index == 0) | |
4308 | insert->ins_contig = CONTIG_NONE; | |
4309 | ||
4310 | /* | |
4311 | * Ok, so we can simply compare against last_eb to figure out | |
4312 | * whether the path doesn't exist. This will only happen in | |
4313 | * the case that we're doing a tail append, so maybe we can | |
4314 | * take advantage of that information somehow. | |
4315 | */ | |
35dc0aa3 | 4316 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4317 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4318 | /* |
4319 | * Ok, ocfs2_find_path() returned us the rightmost | |
4320 | * tree path. This might be an appending insert. There are | |
4321 | * two cases: | |
4322 | * 1) We're doing a true append at the tail: | |
4323 | * -This might even be off the end of the leaf | |
4324 | * 2) We're "appending" by rotating in the tail | |
4325 | */ | |
4326 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4327 | } | |
4328 | ||
4329 | out: | |
4330 | ocfs2_free_path(path); | |
4331 | ||
4332 | if (ret == 0) | |
4333 | *last_eb_bh = bh; | |
4334 | else | |
4335 | brelse(bh); | |
4336 | return ret; | |
ccd979bd MF |
4337 | } |
4338 | ||
dcd0538f MF |
4339 | /* |
4340 | * Insert an extent into an inode btree. | |
4341 | * | |
4342 | * The caller needs to update fe->i_clusters | |
4343 | */ | |
f56654c4 TM |
4344 | static int ocfs2_insert_extent(struct ocfs2_super *osb, |
4345 | handle_t *handle, | |
4346 | struct inode *inode, | |
4347 | struct buffer_head *root_bh, | |
4348 | u32 cpos, | |
4349 | u64 start_blk, | |
4350 | u32 new_clusters, | |
4351 | u8 flags, | |
4352 | struct ocfs2_alloc_context *meta_ac, | |
4353 | struct ocfs2_extent_tree *et) | |
ccd979bd | 4354 | { |
c3afcbb3 | 4355 | int status; |
c77534f6 | 4356 | int uninitialized_var(free_records); |
ccd979bd | 4357 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4358 | struct ocfs2_insert_type insert = {0, }; |
4359 | struct ocfs2_extent_rec rec; | |
4360 | ||
1afc32b9 MF |
4361 | BUG_ON(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
4362 | ||
dcd0538f MF |
4363 | mlog(0, "add %u clusters at position %u to inode %llu\n", |
4364 | new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4365 | ||
4366 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && | |
4367 | (OCFS2_I(inode)->ip_clusters != cpos), | |
4368 | "Device %s, asking for sparse allocation: inode %llu, " | |
4369 | "cpos %u, clusters %u\n", | |
4370 | osb->dev_str, | |
4371 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, | |
4372 | OCFS2_I(inode)->ip_clusters); | |
4373 | ||
e48edee2 | 4374 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4375 | rec.e_cpos = cpu_to_le32(cpos); |
4376 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4377 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4378 | rec.e_flags = flags; |
dcd0538f | 4379 | |
e7d4cb6b | 4380 | status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec, |
c77534f6 | 4381 | &free_records, &insert); |
dcd0538f MF |
4382 | if (status < 0) { |
4383 | mlog_errno(status); | |
4384 | goto bail; | |
ccd979bd MF |
4385 | } |
4386 | ||
dcd0538f MF |
4387 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4388 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4389 | "Insert.tree_depth: %d\n", | |
4390 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4391 | free_records, insert.ins_tree_depth); |
ccd979bd | 4392 | |
c77534f6 | 4393 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
e7d4cb6b | 4394 | status = ocfs2_grow_tree(inode, handle, et, |
328d5752 | 4395 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4396 | meta_ac); |
4397 | if (status) { | |
ccd979bd MF |
4398 | mlog_errno(status); |
4399 | goto bail; | |
4400 | } | |
ccd979bd MF |
4401 | } |
4402 | ||
dcd0538f | 4403 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
e7d4cb6b | 4404 | status = ocfs2_do_insert_extent(inode, handle, et, &rec, &insert); |
ccd979bd MF |
4405 | if (status < 0) |
4406 | mlog_errno(status); | |
e7d4cb6b | 4407 | else if (et->type == OCFS2_DINODE_EXTENT) |
83418978 | 4408 | ocfs2_extent_map_insert_rec(inode, &rec); |
ccd979bd MF |
4409 | |
4410 | bail: | |
ccd979bd MF |
4411 | if (last_eb_bh) |
4412 | brelse(last_eb_bh); | |
4413 | ||
f56654c4 TM |
4414 | mlog_exit(status); |
4415 | return status; | |
4416 | } | |
4417 | ||
4418 | int ocfs2_dinode_insert_extent(struct ocfs2_super *osb, | |
4419 | handle_t *handle, | |
4420 | struct inode *inode, | |
4421 | struct buffer_head *root_bh, | |
4422 | u32 cpos, | |
4423 | u64 start_blk, | |
4424 | u32 new_clusters, | |
4425 | u8 flags, | |
4426 | struct ocfs2_alloc_context *meta_ac) | |
4427 | { | |
4428 | int status; | |
4429 | struct ocfs2_extent_tree *et = NULL; | |
4430 | ||
ca12b7c4 | 4431 | et = ocfs2_new_extent_tree(inode, root_bh, OCFS2_DINODE_EXTENT, NULL); |
f56654c4 TM |
4432 | if (!et) { |
4433 | status = -ENOMEM; | |
4434 | mlog_errno(status); | |
4435 | goto bail; | |
4436 | } | |
4437 | ||
4438 | status = ocfs2_insert_extent(osb, handle, inode, root_bh, | |
4439 | cpos, start_blk, new_clusters, | |
4440 | flags, meta_ac, et); | |
4441 | ||
e7d4cb6b TM |
4442 | if (et) |
4443 | ocfs2_free_extent_tree(et); | |
f56654c4 TM |
4444 | bail: |
4445 | return status; | |
4446 | } | |
4447 | ||
4448 | int ocfs2_xattr_value_insert_extent(struct ocfs2_super *osb, | |
4449 | handle_t *handle, | |
4450 | struct inode *inode, | |
4451 | struct buffer_head *root_bh, | |
4452 | u32 cpos, | |
4453 | u64 start_blk, | |
4454 | u32 new_clusters, | |
4455 | u8 flags, | |
4456 | struct ocfs2_alloc_context *meta_ac, | |
4457 | void *private) | |
4458 | { | |
4459 | int status; | |
4460 | struct ocfs2_extent_tree *et = NULL; | |
4461 | ||
ca12b7c4 TM |
4462 | et = ocfs2_new_extent_tree(inode, root_bh, |
4463 | OCFS2_XATTR_VALUE_EXTENT, private); | |
f56654c4 TM |
4464 | if (!et) { |
4465 | status = -ENOMEM; | |
4466 | mlog_errno(status); | |
4467 | goto bail; | |
4468 | } | |
4469 | ||
4470 | status = ocfs2_insert_extent(osb, handle, inode, root_bh, | |
4471 | cpos, start_blk, new_clusters, | |
4472 | flags, meta_ac, et); | |
4473 | ||
4474 | if (et) | |
4475 | ocfs2_free_extent_tree(et); | |
4476 | bail: | |
ccd979bd MF |
4477 | return status; |
4478 | } | |
4479 | ||
ba492615 TM |
4480 | int ocfs2_xattr_tree_insert_extent(struct ocfs2_super *osb, |
4481 | handle_t *handle, | |
4482 | struct inode *inode, | |
4483 | struct buffer_head *root_bh, | |
4484 | u32 cpos, | |
4485 | u64 start_blk, | |
4486 | u32 new_clusters, | |
4487 | u8 flags, | |
4488 | struct ocfs2_alloc_context *meta_ac) | |
4489 | { | |
4490 | int status; | |
4491 | struct ocfs2_extent_tree *et = NULL; | |
4492 | ||
ca12b7c4 TM |
4493 | et = ocfs2_new_extent_tree(inode, root_bh, OCFS2_XATTR_TREE_EXTENT, |
4494 | NULL); | |
ba492615 TM |
4495 | if (!et) { |
4496 | status = -ENOMEM; | |
4497 | mlog_errno(status); | |
4498 | goto bail; | |
4499 | } | |
4500 | ||
4501 | status = ocfs2_insert_extent(osb, handle, inode, root_bh, | |
4502 | cpos, start_blk, new_clusters, | |
4503 | flags, meta_ac, et); | |
4504 | ||
4505 | if (et) | |
4506 | ocfs2_free_extent_tree(et); | |
4507 | bail: | |
4508 | return status; | |
4509 | } | |
4510 | ||
0eb8d47e TM |
4511 | /* |
4512 | * Allcate and add clusters into the extent b-tree. | |
4513 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
4514 | * The extent b-tree's root is root_el and it should be in root_bh, and | |
4515 | * it is not limited to the file storage. Any extent tree can use this | |
4516 | * function if it implements the proper ocfs2_extent_tree. | |
4517 | */ | |
4518 | int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb, | |
4519 | struct inode *inode, | |
4520 | u32 *logical_offset, | |
4521 | u32 clusters_to_add, | |
4522 | int mark_unwritten, | |
4523 | struct buffer_head *root_bh, | |
4524 | struct ocfs2_extent_list *root_el, | |
4525 | handle_t *handle, | |
4526 | struct ocfs2_alloc_context *data_ac, | |
4527 | struct ocfs2_alloc_context *meta_ac, | |
4528 | enum ocfs2_alloc_restarted *reason_ret, | |
f56654c4 TM |
4529 | enum ocfs2_extent_tree_type type, |
4530 | void *private) | |
0eb8d47e TM |
4531 | { |
4532 | int status = 0; | |
4533 | int free_extents; | |
4534 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4535 | u32 bit_off, num_bits; | |
4536 | u64 block; | |
4537 | u8 flags = 0; | |
4538 | ||
4539 | BUG_ON(!clusters_to_add); | |
4540 | ||
4541 | if (mark_unwritten) | |
4542 | flags = OCFS2_EXT_UNWRITTEN; | |
4543 | ||
f56654c4 TM |
4544 | free_extents = ocfs2_num_free_extents(osb, inode, root_bh, type, |
4545 | private); | |
0eb8d47e TM |
4546 | if (free_extents < 0) { |
4547 | status = free_extents; | |
4548 | mlog_errno(status); | |
4549 | goto leave; | |
4550 | } | |
4551 | ||
4552 | /* there are two cases which could cause us to EAGAIN in the | |
4553 | * we-need-more-metadata case: | |
4554 | * 1) we haven't reserved *any* | |
4555 | * 2) we are so fragmented, we've needed to add metadata too | |
4556 | * many times. */ | |
4557 | if (!free_extents && !meta_ac) { | |
4558 | mlog(0, "we haven't reserved any metadata!\n"); | |
4559 | status = -EAGAIN; | |
4560 | reason = RESTART_META; | |
4561 | goto leave; | |
4562 | } else if ((!free_extents) | |
4563 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
4564 | < ocfs2_extend_meta_needed(root_el))) { | |
4565 | mlog(0, "filesystem is really fragmented...\n"); | |
4566 | status = -EAGAIN; | |
4567 | reason = RESTART_META; | |
4568 | goto leave; | |
4569 | } | |
4570 | ||
4571 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4572 | clusters_to_add, &bit_off, &num_bits); | |
4573 | if (status < 0) { | |
4574 | if (status != -ENOSPC) | |
4575 | mlog_errno(status); | |
4576 | goto leave; | |
4577 | } | |
4578 | ||
4579 | BUG_ON(num_bits > clusters_to_add); | |
4580 | ||
4581 | /* reserve our write early -- insert_extent may update the inode */ | |
4582 | status = ocfs2_journal_access(handle, inode, root_bh, | |
4583 | OCFS2_JOURNAL_ACCESS_WRITE); | |
4584 | if (status < 0) { | |
4585 | mlog_errno(status); | |
4586 | goto leave; | |
4587 | } | |
4588 | ||
4589 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
4590 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", | |
4591 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
f56654c4 TM |
4592 | if (type == OCFS2_DINODE_EXTENT) |
4593 | status = ocfs2_dinode_insert_extent(osb, handle, inode, root_bh, | |
4594 | *logical_offset, block, | |
4595 | num_bits, flags, meta_ac); | |
ba492615 TM |
4596 | else if (type == OCFS2_XATTR_TREE_EXTENT) |
4597 | status = ocfs2_xattr_tree_insert_extent(osb, handle, | |
4598 | inode, root_bh, | |
4599 | *logical_offset, | |
4600 | block, num_bits, flags, | |
4601 | meta_ac); | |
f56654c4 TM |
4602 | else |
4603 | status = ocfs2_xattr_value_insert_extent(osb, handle, | |
4604 | inode, root_bh, | |
4605 | *logical_offset, | |
4606 | block, num_bits, flags, | |
4607 | meta_ac, private); | |
0eb8d47e TM |
4608 | if (status < 0) { |
4609 | mlog_errno(status); | |
4610 | goto leave; | |
4611 | } | |
4612 | ||
4613 | status = ocfs2_journal_dirty(handle, root_bh); | |
4614 | if (status < 0) { | |
4615 | mlog_errno(status); | |
4616 | goto leave; | |
4617 | } | |
4618 | ||
4619 | clusters_to_add -= num_bits; | |
4620 | *logical_offset += num_bits; | |
4621 | ||
4622 | if (clusters_to_add) { | |
4623 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4624 | clusters_to_add); | |
4625 | status = -EAGAIN; | |
4626 | reason = RESTART_TRANS; | |
4627 | } | |
4628 | ||
4629 | leave: | |
4630 | mlog_exit(status); | |
4631 | if (reason_ret) | |
4632 | *reason_ret = reason; | |
4633 | return status; | |
4634 | } | |
4635 | ||
328d5752 MF |
4636 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4637 | struct ocfs2_extent_rec *split_rec, | |
4638 | u32 cpos, | |
4639 | struct ocfs2_extent_rec *rec) | |
4640 | { | |
4641 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4642 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4643 | ||
4644 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4645 | ||
4646 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4647 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4648 | ||
4649 | split_rec->e_blkno = rec->e_blkno; | |
4650 | le64_add_cpu(&split_rec->e_blkno, | |
4651 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4652 | ||
4653 | split_rec->e_flags = rec->e_flags; | |
4654 | } | |
4655 | ||
4656 | static int ocfs2_split_and_insert(struct inode *inode, | |
4657 | handle_t *handle, | |
4658 | struct ocfs2_path *path, | |
e7d4cb6b | 4659 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4660 | struct buffer_head **last_eb_bh, |
4661 | int split_index, | |
4662 | struct ocfs2_extent_rec *orig_split_rec, | |
4663 | struct ocfs2_alloc_context *meta_ac) | |
4664 | { | |
4665 | int ret = 0, depth; | |
4666 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4667 | struct ocfs2_extent_rec tmprec; | |
4668 | struct ocfs2_extent_list *rightmost_el; | |
4669 | struct ocfs2_extent_rec rec; | |
4670 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4671 | struct ocfs2_insert_type insert; | |
4672 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4673 | |
4674 | leftright: | |
4675 | /* | |
4676 | * Store a copy of the record on the stack - it might move | |
4677 | * around as the tree is manipulated below. | |
4678 | */ | |
4679 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4680 | ||
e7d4cb6b | 4681 | rightmost_el = et->root_el; |
328d5752 MF |
4682 | |
4683 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4684 | if (depth) { | |
4685 | BUG_ON(!(*last_eb_bh)); | |
4686 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4687 | rightmost_el = &eb->h_list; | |
4688 | } | |
4689 | ||
4690 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4691 | le16_to_cpu(rightmost_el->l_count)) { | |
e7d4cb6b TM |
4692 | ret = ocfs2_grow_tree(inode, handle, et, |
4693 | &depth, last_eb_bh, meta_ac); | |
328d5752 MF |
4694 | if (ret) { |
4695 | mlog_errno(ret); | |
4696 | goto out; | |
4697 | } | |
328d5752 MF |
4698 | } |
4699 | ||
4700 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4701 | insert.ins_appending = APPEND_NONE; | |
4702 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4703 | insert.ins_tree_depth = depth; |
4704 | ||
4705 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4706 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4707 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4708 | le16_to_cpu(rec.e_leaf_clusters); | |
4709 | ||
4710 | if (split_rec.e_cpos == rec.e_cpos) { | |
4711 | insert.ins_split = SPLIT_LEFT; | |
4712 | } else if (insert_range == rec_range) { | |
4713 | insert.ins_split = SPLIT_RIGHT; | |
4714 | } else { | |
4715 | /* | |
4716 | * Left/right split. We fake this as a right split | |
4717 | * first and then make a second pass as a left split. | |
4718 | */ | |
4719 | insert.ins_split = SPLIT_RIGHT; | |
4720 | ||
4721 | ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range, | |
4722 | &rec); | |
4723 | ||
4724 | split_rec = tmprec; | |
4725 | ||
4726 | BUG_ON(do_leftright); | |
4727 | do_leftright = 1; | |
4728 | } | |
4729 | ||
e7d4cb6b | 4730 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
328d5752 MF |
4731 | if (ret) { |
4732 | mlog_errno(ret); | |
4733 | goto out; | |
4734 | } | |
4735 | ||
4736 | if (do_leftright == 1) { | |
4737 | u32 cpos; | |
4738 | struct ocfs2_extent_list *el; | |
4739 | ||
4740 | do_leftright++; | |
4741 | split_rec = *orig_split_rec; | |
4742 | ||
4743 | ocfs2_reinit_path(path, 1); | |
4744 | ||
4745 | cpos = le32_to_cpu(split_rec.e_cpos); | |
4746 | ret = ocfs2_find_path(inode, path, cpos); | |
4747 | if (ret) { | |
4748 | mlog_errno(ret); | |
4749 | goto out; | |
4750 | } | |
4751 | ||
4752 | el = path_leaf_el(path); | |
4753 | split_index = ocfs2_search_extent_list(el, cpos); | |
4754 | goto leftright; | |
4755 | } | |
4756 | out: | |
4757 | ||
4758 | return ret; | |
4759 | } | |
4760 | ||
4761 | /* | |
4762 | * Mark part or all of the extent record at split_index in the leaf | |
4763 | * pointed to by path as written. This removes the unwritten | |
4764 | * extent flag. | |
4765 | * | |
4766 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4767 | * | |
4768 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4769 | * of the tree is required. All other cases will degrade into a less | |
4770 | * optimal tree layout. | |
4771 | * | |
e7d4cb6b TM |
4772 | * last_eb_bh should be the rightmost leaf block for any extent |
4773 | * btree. Since a split may grow the tree or a merge might shrink it, | |
4774 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
4775 | * |
4776 | * This code is optimized for readability - several passes might be | |
4777 | * made over certain portions of the tree. All of those blocks will | |
4778 | * have been brought into cache (and pinned via the journal), so the | |
4779 | * extra overhead is not expressed in terms of disk reads. | |
4780 | */ | |
4781 | static int __ocfs2_mark_extent_written(struct inode *inode, | |
e7d4cb6b | 4782 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4783 | handle_t *handle, |
4784 | struct ocfs2_path *path, | |
4785 | int split_index, | |
4786 | struct ocfs2_extent_rec *split_rec, | |
4787 | struct ocfs2_alloc_context *meta_ac, | |
4788 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
4789 | { | |
4790 | int ret = 0; | |
4791 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 4792 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
4793 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
4794 | struct ocfs2_merge_ctxt ctxt; | |
4795 | struct ocfs2_extent_list *rightmost_el; | |
4796 | ||
3cf0c507 | 4797 | if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) { |
328d5752 MF |
4798 | ret = -EIO; |
4799 | mlog_errno(ret); | |
4800 | goto out; | |
4801 | } | |
4802 | ||
4803 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || | |
4804 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
4805 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
4806 | ret = -EIO; | |
4807 | mlog_errno(ret); | |
4808 | goto out; | |
4809 | } | |
4810 | ||
ad5a4d70 | 4811 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, path, el, |
328d5752 MF |
4812 | split_index, |
4813 | split_rec); | |
4814 | ||
4815 | /* | |
4816 | * The core merge / split code wants to know how much room is | |
4817 | * left in this inodes allocation tree, so we pass the | |
4818 | * rightmost extent list. | |
4819 | */ | |
4820 | if (path->p_tree_depth) { | |
4821 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4822 | |
4823 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
35dc0aa3 | 4824 | ocfs2_et_get_last_eb_blk(et), |
328d5752 MF |
4825 | &last_eb_bh, OCFS2_BH_CACHED, inode); |
4826 | if (ret) { | |
4827 | mlog_exit(ret); | |
4828 | goto out; | |
4829 | } | |
4830 | ||
4831 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
4832 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
4833 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
4834 | ret = -EROFS; | |
4835 | goto out; | |
4836 | } | |
4837 | ||
4838 | rightmost_el = &eb->h_list; | |
4839 | } else | |
4840 | rightmost_el = path_root_el(path); | |
4841 | ||
328d5752 MF |
4842 | if (rec->e_cpos == split_rec->e_cpos && |
4843 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
4844 | ctxt.c_split_covers_rec = 1; | |
4845 | else | |
4846 | ctxt.c_split_covers_rec = 0; | |
4847 | ||
4848 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
4849 | ||
015452b1 MF |
4850 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
4851 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
4852 | ctxt.c_split_covers_rec); | |
328d5752 MF |
4853 | |
4854 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
4855 | if (ctxt.c_split_covers_rec) | |
4856 | el->l_recs[split_index] = *split_rec; | |
4857 | else | |
e7d4cb6b | 4858 | ret = ocfs2_split_and_insert(inode, handle, path, et, |
328d5752 MF |
4859 | &last_eb_bh, split_index, |
4860 | split_rec, meta_ac); | |
4861 | if (ret) | |
4862 | mlog_errno(ret); | |
4863 | } else { | |
4864 | ret = ocfs2_try_to_merge_extent(inode, handle, path, | |
4865 | split_index, split_rec, | |
e7d4cb6b | 4866 | dealloc, &ctxt, et); |
328d5752 MF |
4867 | if (ret) |
4868 | mlog_errno(ret); | |
4869 | } | |
4870 | ||
328d5752 MF |
4871 | out: |
4872 | brelse(last_eb_bh); | |
4873 | return ret; | |
4874 | } | |
4875 | ||
4876 | /* | |
4877 | * Mark the already-existing extent at cpos as written for len clusters. | |
4878 | * | |
4879 | * If the existing extent is larger than the request, initiate a | |
4880 | * split. An attempt will be made at merging with adjacent extents. | |
4881 | * | |
4882 | * The caller is responsible for passing down meta_ac if we'll need it. | |
4883 | */ | |
e7d4cb6b | 4884 | int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *root_bh, |
328d5752 MF |
4885 | handle_t *handle, u32 cpos, u32 len, u32 phys, |
4886 | struct ocfs2_alloc_context *meta_ac, | |
e7d4cb6b | 4887 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
f56654c4 TM |
4888 | enum ocfs2_extent_tree_type et_type, |
4889 | void *private) | |
328d5752 MF |
4890 | { |
4891 | int ret, index; | |
4892 | u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys); | |
4893 | struct ocfs2_extent_rec split_rec; | |
4894 | struct ocfs2_path *left_path = NULL; | |
4895 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 4896 | struct ocfs2_extent_tree *et = NULL; |
328d5752 MF |
4897 | |
4898 | mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n", | |
4899 | inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno); | |
4900 | ||
4901 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
4902 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
4903 | "that are being written to, but the feature bit " | |
4904 | "is not set in the super block.", | |
4905 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4906 | ret = -EROFS; | |
4907 | goto out; | |
4908 | } | |
4909 | ||
ca12b7c4 | 4910 | et = ocfs2_new_extent_tree(inode, root_bh, et_type, private); |
e7d4cb6b TM |
4911 | if (!et) { |
4912 | ret = -ENOMEM; | |
4913 | mlog_errno(ret); | |
4914 | goto out; | |
4915 | } | |
4916 | ||
328d5752 MF |
4917 | /* |
4918 | * XXX: This should be fixed up so that we just re-insert the | |
4919 | * next extent records. | |
4920 | */ | |
e7d4cb6b TM |
4921 | if (et_type == OCFS2_DINODE_EXTENT) |
4922 | ocfs2_extent_map_trunc(inode, 0); | |
328d5752 | 4923 | |
e7d4cb6b | 4924 | left_path = ocfs2_new_path(et->root_bh, et->root_el); |
328d5752 MF |
4925 | if (!left_path) { |
4926 | ret = -ENOMEM; | |
4927 | mlog_errno(ret); | |
4928 | goto out; | |
4929 | } | |
4930 | ||
4931 | ret = ocfs2_find_path(inode, left_path, cpos); | |
4932 | if (ret) { | |
4933 | mlog_errno(ret); | |
4934 | goto out; | |
4935 | } | |
4936 | el = path_leaf_el(left_path); | |
4937 | ||
4938 | index = ocfs2_search_extent_list(el, cpos); | |
4939 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
4940 | ocfs2_error(inode->i_sb, | |
4941 | "Inode %llu has an extent at cpos %u which can no " | |
4942 | "longer be found.\n", | |
4943 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
4944 | ret = -EROFS; | |
4945 | goto out; | |
4946 | } | |
4947 | ||
4948 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4949 | split_rec.e_cpos = cpu_to_le32(cpos); | |
4950 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
4951 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
4952 | split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags; | |
4953 | split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN; | |
4954 | ||
e7d4cb6b TM |
4955 | ret = __ocfs2_mark_extent_written(inode, et, handle, left_path, |
4956 | index, &split_rec, meta_ac, | |
4957 | dealloc); | |
328d5752 MF |
4958 | if (ret) |
4959 | mlog_errno(ret); | |
4960 | ||
4961 | out: | |
4962 | ocfs2_free_path(left_path); | |
e7d4cb6b TM |
4963 | if (et) |
4964 | ocfs2_free_extent_tree(et); | |
328d5752 MF |
4965 | return ret; |
4966 | } | |
4967 | ||
e7d4cb6b | 4968 | static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et, |
d0c7d708 MF |
4969 | handle_t *handle, struct ocfs2_path *path, |
4970 | int index, u32 new_range, | |
4971 | struct ocfs2_alloc_context *meta_ac) | |
4972 | { | |
4973 | int ret, depth, credits = handle->h_buffer_credits; | |
d0c7d708 MF |
4974 | struct buffer_head *last_eb_bh = NULL; |
4975 | struct ocfs2_extent_block *eb; | |
4976 | struct ocfs2_extent_list *rightmost_el, *el; | |
4977 | struct ocfs2_extent_rec split_rec; | |
4978 | struct ocfs2_extent_rec *rec; | |
4979 | struct ocfs2_insert_type insert; | |
4980 | ||
4981 | /* | |
4982 | * Setup the record to split before we grow the tree. | |
4983 | */ | |
4984 | el = path_leaf_el(path); | |
4985 | rec = &el->l_recs[index]; | |
4986 | ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec); | |
4987 | ||
4988 | depth = path->p_tree_depth; | |
4989 | if (depth > 0) { | |
4990 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
35dc0aa3 | 4991 | ocfs2_et_get_last_eb_blk(et), |
d0c7d708 MF |
4992 | &last_eb_bh, OCFS2_BH_CACHED, inode); |
4993 | if (ret < 0) { | |
4994 | mlog_errno(ret); | |
4995 | goto out; | |
4996 | } | |
4997 | ||
4998 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
4999 | rightmost_el = &eb->h_list; | |
5000 | } else | |
5001 | rightmost_el = path_leaf_el(path); | |
5002 | ||
811f933d | 5003 | credits += path->p_tree_depth + |
e7d4cb6b | 5004 | ocfs2_extend_meta_needed(et->root_el); |
d0c7d708 MF |
5005 | ret = ocfs2_extend_trans(handle, credits); |
5006 | if (ret) { | |
5007 | mlog_errno(ret); | |
5008 | goto out; | |
5009 | } | |
5010 | ||
5011 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5012 | le16_to_cpu(rightmost_el->l_count)) { | |
e7d4cb6b | 5013 | ret = ocfs2_grow_tree(inode, handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5014 | meta_ac); |
5015 | if (ret) { | |
5016 | mlog_errno(ret); | |
5017 | goto out; | |
5018 | } | |
d0c7d708 MF |
5019 | } |
5020 | ||
5021 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5022 | insert.ins_appending = APPEND_NONE; | |
5023 | insert.ins_contig = CONTIG_NONE; | |
5024 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5025 | insert.ins_tree_depth = depth; |
5026 | ||
e7d4cb6b | 5027 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
d0c7d708 MF |
5028 | if (ret) |
5029 | mlog_errno(ret); | |
5030 | ||
5031 | out: | |
5032 | brelse(last_eb_bh); | |
5033 | return ret; | |
5034 | } | |
5035 | ||
5036 | static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle, | |
5037 | struct ocfs2_path *path, int index, | |
5038 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
5039 | u32 cpos, u32 len, |
5040 | struct ocfs2_extent_tree *et) | |
d0c7d708 MF |
5041 | { |
5042 | int ret; | |
5043 | u32 left_cpos, rec_range, trunc_range; | |
5044 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
5045 | struct super_block *sb = inode->i_sb; | |
5046 | struct ocfs2_path *left_path = NULL; | |
5047 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5048 | struct ocfs2_extent_rec *rec; | |
5049 | struct ocfs2_extent_block *eb; | |
5050 | ||
5051 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
e7d4cb6b | 5052 | ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et); |
d0c7d708 MF |
5053 | if (ret) { |
5054 | mlog_errno(ret); | |
5055 | goto out; | |
5056 | } | |
5057 | ||
5058 | index--; | |
5059 | } | |
5060 | ||
5061 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5062 | path->p_tree_depth) { | |
5063 | /* | |
5064 | * Check whether this is the rightmost tree record. If | |
5065 | * we remove all of this record or part of its right | |
5066 | * edge then an update of the record lengths above it | |
5067 | * will be required. | |
5068 | */ | |
5069 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5070 | if (eb->h_next_leaf_blk == 0) | |
5071 | is_rightmost_tree_rec = 1; | |
5072 | } | |
5073 | ||
5074 | rec = &el->l_recs[index]; | |
5075 | if (index == 0 && path->p_tree_depth && | |
5076 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5077 | /* | |
5078 | * Changing the leftmost offset (via partial or whole | |
5079 | * record truncate) of an interior (or rightmost) path | |
5080 | * means we have to update the subtree that is formed | |
5081 | * by this leaf and the one to it's left. | |
5082 | * | |
5083 | * There are two cases we can skip: | |
5084 | * 1) Path is the leftmost one in our inode tree. | |
5085 | * 2) The leaf is rightmost and will be empty after | |
5086 | * we remove the extent record - the rotate code | |
5087 | * knows how to update the newly formed edge. | |
5088 | */ | |
5089 | ||
5090 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, | |
5091 | &left_cpos); | |
5092 | if (ret) { | |
5093 | mlog_errno(ret); | |
5094 | goto out; | |
5095 | } | |
5096 | ||
5097 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
5098 | left_path = ocfs2_new_path(path_root_bh(path), | |
5099 | path_root_el(path)); | |
5100 | if (!left_path) { | |
5101 | ret = -ENOMEM; | |
5102 | mlog_errno(ret); | |
5103 | goto out; | |
5104 | } | |
5105 | ||
5106 | ret = ocfs2_find_path(inode, left_path, left_cpos); | |
5107 | if (ret) { | |
5108 | mlog_errno(ret); | |
5109 | goto out; | |
5110 | } | |
5111 | } | |
5112 | } | |
5113 | ||
5114 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5115 | handle->h_buffer_credits, | |
5116 | path); | |
5117 | if (ret) { | |
5118 | mlog_errno(ret); | |
5119 | goto out; | |
5120 | } | |
5121 | ||
5122 | ret = ocfs2_journal_access_path(inode, handle, path); | |
5123 | if (ret) { | |
5124 | mlog_errno(ret); | |
5125 | goto out; | |
5126 | } | |
5127 | ||
5128 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
5129 | if (ret) { | |
5130 | mlog_errno(ret); | |
5131 | goto out; | |
5132 | } | |
5133 | ||
5134 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5135 | trunc_range = cpos + len; | |
5136 | ||
5137 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5138 | int next_free; | |
5139 | ||
5140 | memset(rec, 0, sizeof(*rec)); | |
5141 | ocfs2_cleanup_merge(el, index); | |
5142 | wants_rotate = 1; | |
5143 | ||
5144 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5145 | if (is_rightmost_tree_rec && next_free > 1) { | |
5146 | /* | |
5147 | * We skip the edge update if this path will | |
5148 | * be deleted by the rotate code. | |
5149 | */ | |
5150 | rec = &el->l_recs[next_free - 1]; | |
5151 | ocfs2_adjust_rightmost_records(inode, handle, path, | |
5152 | rec); | |
5153 | } | |
5154 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5155 | /* Remove leftmost portion of the record. */ | |
5156 | le32_add_cpu(&rec->e_cpos, len); | |
5157 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5158 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5159 | } else if (rec_range == trunc_range) { | |
5160 | /* Remove rightmost portion of the record */ | |
5161 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5162 | if (is_rightmost_tree_rec) | |
5163 | ocfs2_adjust_rightmost_records(inode, handle, path, rec); | |
5164 | } else { | |
5165 | /* Caller should have trapped this. */ | |
5166 | mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) " | |
5167 | "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5168 | le32_to_cpu(rec->e_cpos), | |
5169 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5170 | BUG(); | |
5171 | } | |
5172 | ||
5173 | if (left_path) { | |
5174 | int subtree_index; | |
5175 | ||
5176 | subtree_index = ocfs2_find_subtree_root(inode, left_path, path); | |
5177 | ocfs2_complete_edge_insert(inode, handle, left_path, path, | |
5178 | subtree_index); | |
5179 | } | |
5180 | ||
5181 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5182 | ||
e7d4cb6b | 5183 | ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et); |
d0c7d708 MF |
5184 | if (ret) { |
5185 | mlog_errno(ret); | |
5186 | goto out; | |
5187 | } | |
5188 | ||
5189 | out: | |
5190 | ocfs2_free_path(left_path); | |
5191 | return ret; | |
5192 | } | |
5193 | ||
e7d4cb6b | 5194 | int ocfs2_remove_extent(struct inode *inode, struct buffer_head *root_bh, |
063c4561 MF |
5195 | u32 cpos, u32 len, handle_t *handle, |
5196 | struct ocfs2_alloc_context *meta_ac, | |
e7d4cb6b | 5197 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
f56654c4 TM |
5198 | enum ocfs2_extent_tree_type et_type, |
5199 | void *private) | |
d0c7d708 MF |
5200 | { |
5201 | int ret, index; | |
5202 | u32 rec_range, trunc_range; | |
5203 | struct ocfs2_extent_rec *rec; | |
5204 | struct ocfs2_extent_list *el; | |
e7d4cb6b TM |
5205 | struct ocfs2_path *path = NULL; |
5206 | struct ocfs2_extent_tree *et = NULL; | |
5207 | ||
ca12b7c4 | 5208 | et = ocfs2_new_extent_tree(inode, root_bh, et_type, private); |
e7d4cb6b TM |
5209 | if (!et) { |
5210 | ret = -ENOMEM; | |
5211 | mlog_errno(ret); | |
5212 | goto out; | |
5213 | } | |
d0c7d708 MF |
5214 | |
5215 | ocfs2_extent_map_trunc(inode, 0); | |
5216 | ||
e7d4cb6b | 5217 | path = ocfs2_new_path(et->root_bh, et->root_el); |
d0c7d708 MF |
5218 | if (!path) { |
5219 | ret = -ENOMEM; | |
5220 | mlog_errno(ret); | |
5221 | goto out; | |
5222 | } | |
5223 | ||
5224 | ret = ocfs2_find_path(inode, path, cpos); | |
5225 | if (ret) { | |
5226 | mlog_errno(ret); | |
5227 | goto out; | |
5228 | } | |
5229 | ||
5230 | el = path_leaf_el(path); | |
5231 | index = ocfs2_search_extent_list(el, cpos); | |
5232 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5233 | ocfs2_error(inode->i_sb, | |
5234 | "Inode %llu has an extent at cpos %u which can no " | |
5235 | "longer be found.\n", | |
5236 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5237 | ret = -EROFS; | |
5238 | goto out; | |
5239 | } | |
5240 | ||
5241 | /* | |
5242 | * We have 3 cases of extent removal: | |
5243 | * 1) Range covers the entire extent rec | |
5244 | * 2) Range begins or ends on one edge of the extent rec | |
5245 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5246 | * | |
5247 | * For case 1 we remove the extent rec and left rotate to | |
5248 | * fill the hole. | |
5249 | * | |
5250 | * For case 2 we just shrink the existing extent rec, with a | |
5251 | * tree update if the shrinking edge is also the edge of an | |
5252 | * extent block. | |
5253 | * | |
5254 | * For case 3 we do a right split to turn the extent rec into | |
5255 | * something case 2 can handle. | |
5256 | */ | |
5257 | rec = &el->l_recs[index]; | |
5258 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5259 | trunc_range = cpos + len; | |
5260 | ||
5261 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5262 | ||
5263 | mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d " | |
5264 | "(cpos %u, len %u)\n", | |
5265 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index, | |
5266 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); | |
5267 | ||
5268 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
5269 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
e7d4cb6b | 5270 | cpos, len, et); |
d0c7d708 MF |
5271 | if (ret) { |
5272 | mlog_errno(ret); | |
5273 | goto out; | |
5274 | } | |
5275 | } else { | |
e7d4cb6b | 5276 | ret = ocfs2_split_tree(inode, et, handle, path, index, |
d0c7d708 MF |
5277 | trunc_range, meta_ac); |
5278 | if (ret) { | |
5279 | mlog_errno(ret); | |
5280 | goto out; | |
5281 | } | |
5282 | ||
5283 | /* | |
5284 | * The split could have manipulated the tree enough to | |
5285 | * move the record location, so we have to look for it again. | |
5286 | */ | |
5287 | ocfs2_reinit_path(path, 1); | |
5288 | ||
5289 | ret = ocfs2_find_path(inode, path, cpos); | |
5290 | if (ret) { | |
5291 | mlog_errno(ret); | |
5292 | goto out; | |
5293 | } | |
5294 | ||
5295 | el = path_leaf_el(path); | |
5296 | index = ocfs2_search_extent_list(el, cpos); | |
5297 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5298 | ocfs2_error(inode->i_sb, | |
5299 | "Inode %llu: split at cpos %u lost record.", | |
5300 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5301 | cpos); | |
5302 | ret = -EROFS; | |
5303 | goto out; | |
5304 | } | |
5305 | ||
5306 | /* | |
5307 | * Double check our values here. If anything is fishy, | |
5308 | * it's easier to catch it at the top level. | |
5309 | */ | |
5310 | rec = &el->l_recs[index]; | |
5311 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5312 | ocfs2_rec_clusters(el, rec); | |
5313 | if (rec_range != trunc_range) { | |
5314 | ocfs2_error(inode->i_sb, | |
5315 | "Inode %llu: error after split at cpos %u" | |
5316 | "trunc len %u, existing record is (%u,%u)", | |
5317 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5318 | cpos, len, le32_to_cpu(rec->e_cpos), | |
5319 | ocfs2_rec_clusters(el, rec)); | |
5320 | ret = -EROFS; | |
5321 | goto out; | |
5322 | } | |
5323 | ||
5324 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
e7d4cb6b | 5325 | cpos, len, et); |
d0c7d708 MF |
5326 | if (ret) { |
5327 | mlog_errno(ret); | |
5328 | goto out; | |
5329 | } | |
5330 | } | |
5331 | ||
5332 | out: | |
5333 | ocfs2_free_path(path); | |
e7d4cb6b TM |
5334 | if (et) |
5335 | ocfs2_free_extent_tree(et); | |
d0c7d708 MF |
5336 | return ret; |
5337 | } | |
5338 | ||
063c4561 | 5339 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5340 | { |
5341 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5342 | struct ocfs2_dinode *di; | |
5343 | struct ocfs2_truncate_log *tl; | |
5344 | ||
5345 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5346 | tl = &di->id2.i_dealloc; | |
5347 | ||
5348 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5349 | "slot %d, invalid truncate log parameters: used = " | |
5350 | "%u, count = %u\n", osb->slot_num, | |
5351 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5352 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5353 | } | |
5354 | ||
5355 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5356 | unsigned int new_start) | |
5357 | { | |
5358 | unsigned int tail_index; | |
5359 | unsigned int current_tail; | |
5360 | ||
5361 | /* No records, nothing to coalesce */ | |
5362 | if (!le16_to_cpu(tl->tl_used)) | |
5363 | return 0; | |
5364 | ||
5365 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5366 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5367 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5368 | ||
5369 | return current_tail == new_start; | |
5370 | } | |
5371 | ||
063c4561 MF |
5372 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5373 | handle_t *handle, | |
5374 | u64 start_blk, | |
5375 | unsigned int num_clusters) | |
ccd979bd MF |
5376 | { |
5377 | int status, index; | |
5378 | unsigned int start_cluster, tl_count; | |
5379 | struct inode *tl_inode = osb->osb_tl_inode; | |
5380 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5381 | struct ocfs2_dinode *di; | |
5382 | struct ocfs2_truncate_log *tl; | |
5383 | ||
b0697053 MF |
5384 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5385 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5386 | |
1b1dcc1b | 5387 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5388 | |
5389 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5390 | ||
5391 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5392 | tl = &di->id2.i_dealloc; | |
5393 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
5394 | OCFS2_RO_ON_INVALID_DINODE(osb->sb, di); | |
5395 | status = -EIO; | |
5396 | goto bail; | |
5397 | } | |
5398 | ||
5399 | tl_count = le16_to_cpu(tl->tl_count); | |
5400 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5401 | tl_count == 0, | |
b0697053 MF |
5402 | "Truncate record count on #%llu invalid " |
5403 | "wanted %u, actual %u\n", | |
5404 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5405 | ocfs2_truncate_recs_per_inode(osb->sb), |
5406 | le16_to_cpu(tl->tl_count)); | |
5407 | ||
5408 | /* Caller should have known to flush before calling us. */ | |
5409 | index = le16_to_cpu(tl->tl_used); | |
5410 | if (index >= tl_count) { | |
5411 | status = -ENOSPC; | |
5412 | mlog_errno(status); | |
5413 | goto bail; | |
5414 | } | |
5415 | ||
5416 | status = ocfs2_journal_access(handle, tl_inode, tl_bh, | |
5417 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5418 | if (status < 0) { | |
5419 | mlog_errno(status); | |
5420 | goto bail; | |
5421 | } | |
5422 | ||
5423 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5424 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5425 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5426 | |
5427 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5428 | /* | |
5429 | * Move index back to the record we are coalescing with. | |
5430 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5431 | */ | |
5432 | index--; | |
5433 | ||
5434 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5435 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5436 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5437 | num_clusters); | |
5438 | } else { | |
5439 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5440 | tl->tl_used = cpu_to_le16(index + 1); | |
5441 | } | |
5442 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5443 | ||
5444 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5445 | if (status < 0) { | |
5446 | mlog_errno(status); | |
5447 | goto bail; | |
5448 | } | |
5449 | ||
5450 | bail: | |
5451 | mlog_exit(status); | |
5452 | return status; | |
5453 | } | |
5454 | ||
5455 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5456 | handle_t *handle, |
ccd979bd MF |
5457 | struct inode *data_alloc_inode, |
5458 | struct buffer_head *data_alloc_bh) | |
5459 | { | |
5460 | int status = 0; | |
5461 | int i; | |
5462 | unsigned int num_clusters; | |
5463 | u64 start_blk; | |
5464 | struct ocfs2_truncate_rec rec; | |
5465 | struct ocfs2_dinode *di; | |
5466 | struct ocfs2_truncate_log *tl; | |
5467 | struct inode *tl_inode = osb->osb_tl_inode; | |
5468 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5469 | ||
5470 | mlog_entry_void(); | |
5471 | ||
5472 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5473 | tl = &di->id2.i_dealloc; | |
5474 | i = le16_to_cpu(tl->tl_used) - 1; | |
5475 | while (i >= 0) { | |
5476 | /* Caller has given us at least enough credits to | |
5477 | * update the truncate log dinode */ | |
5478 | status = ocfs2_journal_access(handle, tl_inode, tl_bh, | |
5479 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5480 | if (status < 0) { | |
5481 | mlog_errno(status); | |
5482 | goto bail; | |
5483 | } | |
5484 | ||
5485 | tl->tl_used = cpu_to_le16(i); | |
5486 | ||
5487 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5488 | if (status < 0) { | |
5489 | mlog_errno(status); | |
5490 | goto bail; | |
5491 | } | |
5492 | ||
5493 | /* TODO: Perhaps we can calculate the bulk of the | |
5494 | * credits up front rather than extending like | |
5495 | * this. */ | |
5496 | status = ocfs2_extend_trans(handle, | |
5497 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5498 | if (status < 0) { | |
5499 | mlog_errno(status); | |
5500 | goto bail; | |
5501 | } | |
5502 | ||
5503 | rec = tl->tl_recs[i]; | |
5504 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5505 | le32_to_cpu(rec.t_start)); | |
5506 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5507 | ||
5508 | /* if start_blk is not set, we ignore the record as | |
5509 | * invalid. */ | |
5510 | if (start_blk) { | |
5511 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5512 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5513 | ||
5514 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5515 | data_alloc_bh, start_blk, | |
5516 | num_clusters); | |
5517 | if (status < 0) { | |
5518 | mlog_errno(status); | |
5519 | goto bail; | |
5520 | } | |
5521 | } | |
5522 | i--; | |
5523 | } | |
5524 | ||
5525 | bail: | |
5526 | mlog_exit(status); | |
5527 | return status; | |
5528 | } | |
5529 | ||
1b1dcc1b | 5530 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5531 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5532 | { |
5533 | int status; | |
5534 | unsigned int num_to_flush; | |
1fabe148 | 5535 | handle_t *handle; |
ccd979bd MF |
5536 | struct inode *tl_inode = osb->osb_tl_inode; |
5537 | struct inode *data_alloc_inode = NULL; | |
5538 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5539 | struct buffer_head *data_alloc_bh = NULL; | |
5540 | struct ocfs2_dinode *di; | |
5541 | struct ocfs2_truncate_log *tl; | |
5542 | ||
5543 | mlog_entry_void(); | |
5544 | ||
1b1dcc1b | 5545 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5546 | |
5547 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5548 | tl = &di->id2.i_dealloc; | |
5549 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
5550 | OCFS2_RO_ON_INVALID_DINODE(osb->sb, di); | |
5551 | status = -EIO; | |
e08dc8b9 | 5552 | goto out; |
ccd979bd MF |
5553 | } |
5554 | ||
5555 | num_to_flush = le16_to_cpu(tl->tl_used); | |
b0697053 MF |
5556 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5557 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5558 | if (!num_to_flush) { |
5559 | status = 0; | |
e08dc8b9 | 5560 | goto out; |
ccd979bd MF |
5561 | } |
5562 | ||
5563 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5564 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5565 | OCFS2_INVALID_SLOT); | |
5566 | if (!data_alloc_inode) { | |
5567 | status = -EINVAL; | |
5568 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5569 | goto out; |
ccd979bd MF |
5570 | } |
5571 | ||
e08dc8b9 MF |
5572 | mutex_lock(&data_alloc_inode->i_mutex); |
5573 | ||
e63aecb6 | 5574 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5575 | if (status < 0) { |
5576 | mlog_errno(status); | |
e08dc8b9 | 5577 | goto out_mutex; |
ccd979bd MF |
5578 | } |
5579 | ||
65eff9cc | 5580 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5581 | if (IS_ERR(handle)) { |
5582 | status = PTR_ERR(handle); | |
ccd979bd | 5583 | mlog_errno(status); |
e08dc8b9 | 5584 | goto out_unlock; |
ccd979bd MF |
5585 | } |
5586 | ||
5587 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5588 | data_alloc_bh); | |
e08dc8b9 | 5589 | if (status < 0) |
ccd979bd | 5590 | mlog_errno(status); |
ccd979bd | 5591 | |
02dc1af4 | 5592 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 5593 | |
e08dc8b9 MF |
5594 | out_unlock: |
5595 | brelse(data_alloc_bh); | |
e63aecb6 | 5596 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 5597 | |
e08dc8b9 MF |
5598 | out_mutex: |
5599 | mutex_unlock(&data_alloc_inode->i_mutex); | |
5600 | iput(data_alloc_inode); | |
ccd979bd | 5601 | |
e08dc8b9 | 5602 | out: |
ccd979bd MF |
5603 | mlog_exit(status); |
5604 | return status; | |
5605 | } | |
5606 | ||
5607 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
5608 | { | |
5609 | int status; | |
5610 | struct inode *tl_inode = osb->osb_tl_inode; | |
5611 | ||
1b1dcc1b | 5612 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 5613 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 5614 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5615 | |
5616 | return status; | |
5617 | } | |
5618 | ||
c4028958 | 5619 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
5620 | { |
5621 | int status; | |
c4028958 DH |
5622 | struct ocfs2_super *osb = |
5623 | container_of(work, struct ocfs2_super, | |
5624 | osb_truncate_log_wq.work); | |
ccd979bd MF |
5625 | |
5626 | mlog_entry_void(); | |
5627 | ||
5628 | status = ocfs2_flush_truncate_log(osb); | |
5629 | if (status < 0) | |
5630 | mlog_errno(status); | |
4d0ddb2c TM |
5631 | else |
5632 | ocfs2_init_inode_steal_slot(osb); | |
ccd979bd MF |
5633 | |
5634 | mlog_exit(status); | |
5635 | } | |
5636 | ||
5637 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
5638 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
5639 | int cancel) | |
5640 | { | |
5641 | if (osb->osb_tl_inode) { | |
5642 | /* We want to push off log flushes while truncates are | |
5643 | * still running. */ | |
5644 | if (cancel) | |
5645 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5646 | ||
5647 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
5648 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
5649 | } | |
5650 | } | |
5651 | ||
5652 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
5653 | int slot_num, | |
5654 | struct inode **tl_inode, | |
5655 | struct buffer_head **tl_bh) | |
5656 | { | |
5657 | int status; | |
5658 | struct inode *inode = NULL; | |
5659 | struct buffer_head *bh = NULL; | |
5660 | ||
5661 | inode = ocfs2_get_system_file_inode(osb, | |
5662 | TRUNCATE_LOG_SYSTEM_INODE, | |
5663 | slot_num); | |
5664 | if (!inode) { | |
5665 | status = -EINVAL; | |
5666 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
5667 | goto bail; | |
5668 | } | |
5669 | ||
5670 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, | |
5671 | OCFS2_BH_CACHED, inode); | |
5672 | if (status < 0) { | |
5673 | iput(inode); | |
5674 | mlog_errno(status); | |
5675 | goto bail; | |
5676 | } | |
5677 | ||
5678 | *tl_inode = inode; | |
5679 | *tl_bh = bh; | |
5680 | bail: | |
5681 | mlog_exit(status); | |
5682 | return status; | |
5683 | } | |
5684 | ||
5685 | /* called during the 1st stage of node recovery. we stamp a clean | |
5686 | * truncate log and pass back a copy for processing later. if the | |
5687 | * truncate log does not require processing, a *tl_copy is set to | |
5688 | * NULL. */ | |
5689 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
5690 | int slot_num, | |
5691 | struct ocfs2_dinode **tl_copy) | |
5692 | { | |
5693 | int status; | |
5694 | struct inode *tl_inode = NULL; | |
5695 | struct buffer_head *tl_bh = NULL; | |
5696 | struct ocfs2_dinode *di; | |
5697 | struct ocfs2_truncate_log *tl; | |
5698 | ||
5699 | *tl_copy = NULL; | |
5700 | ||
5701 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
5702 | ||
5703 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
5704 | if (status < 0) { | |
5705 | mlog_errno(status); | |
5706 | goto bail; | |
5707 | } | |
5708 | ||
5709 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5710 | tl = &di->id2.i_dealloc; | |
5711 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
5712 | OCFS2_RO_ON_INVALID_DINODE(tl_inode->i_sb, di); | |
5713 | status = -EIO; | |
5714 | goto bail; | |
5715 | } | |
5716 | ||
5717 | if (le16_to_cpu(tl->tl_used)) { | |
5718 | mlog(0, "We'll have %u logs to recover\n", | |
5719 | le16_to_cpu(tl->tl_used)); | |
5720 | ||
5721 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
5722 | if (!(*tl_copy)) { | |
5723 | status = -ENOMEM; | |
5724 | mlog_errno(status); | |
5725 | goto bail; | |
5726 | } | |
5727 | ||
5728 | /* Assuming the write-out below goes well, this copy | |
5729 | * will be passed back to recovery for processing. */ | |
5730 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
5731 | ||
5732 | /* All we need to do to clear the truncate log is set | |
5733 | * tl_used. */ | |
5734 | tl->tl_used = 0; | |
5735 | ||
5736 | status = ocfs2_write_block(osb, tl_bh, tl_inode); | |
5737 | if (status < 0) { | |
5738 | mlog_errno(status); | |
5739 | goto bail; | |
5740 | } | |
5741 | } | |
5742 | ||
5743 | bail: | |
5744 | if (tl_inode) | |
5745 | iput(tl_inode); | |
5746 | if (tl_bh) | |
5747 | brelse(tl_bh); | |
5748 | ||
5749 | if (status < 0 && (*tl_copy)) { | |
5750 | kfree(*tl_copy); | |
5751 | *tl_copy = NULL; | |
5752 | } | |
5753 | ||
5754 | mlog_exit(status); | |
5755 | return status; | |
5756 | } | |
5757 | ||
5758 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
5759 | struct ocfs2_dinode *tl_copy) | |
5760 | { | |
5761 | int status = 0; | |
5762 | int i; | |
5763 | unsigned int clusters, num_recs, start_cluster; | |
5764 | u64 start_blk; | |
1fabe148 | 5765 | handle_t *handle; |
ccd979bd MF |
5766 | struct inode *tl_inode = osb->osb_tl_inode; |
5767 | struct ocfs2_truncate_log *tl; | |
5768 | ||
5769 | mlog_entry_void(); | |
5770 | ||
5771 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
5772 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
5773 | return -EINVAL; | |
5774 | } | |
5775 | ||
5776 | tl = &tl_copy->id2.i_dealloc; | |
5777 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 5778 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 5779 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 5780 | |
1b1dcc1b | 5781 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
5782 | for(i = 0; i < num_recs; i++) { |
5783 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5784 | status = __ocfs2_flush_truncate_log(osb); | |
5785 | if (status < 0) { | |
5786 | mlog_errno(status); | |
5787 | goto bail_up; | |
5788 | } | |
5789 | } | |
5790 | ||
65eff9cc | 5791 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5792 | if (IS_ERR(handle)) { |
5793 | status = PTR_ERR(handle); | |
5794 | mlog_errno(status); | |
5795 | goto bail_up; | |
5796 | } | |
5797 | ||
5798 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
5799 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
5800 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
5801 | ||
5802 | status = ocfs2_truncate_log_append(osb, handle, | |
5803 | start_blk, clusters); | |
02dc1af4 | 5804 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
5805 | if (status < 0) { |
5806 | mlog_errno(status); | |
5807 | goto bail_up; | |
5808 | } | |
5809 | } | |
5810 | ||
5811 | bail_up: | |
1b1dcc1b | 5812 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5813 | |
5814 | mlog_exit(status); | |
5815 | return status; | |
5816 | } | |
5817 | ||
5818 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
5819 | { | |
5820 | int status; | |
5821 | struct inode *tl_inode = osb->osb_tl_inode; | |
5822 | ||
5823 | mlog_entry_void(); | |
5824 | ||
5825 | if (tl_inode) { | |
5826 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5827 | flush_workqueue(ocfs2_wq); | |
5828 | ||
5829 | status = ocfs2_flush_truncate_log(osb); | |
5830 | if (status < 0) | |
5831 | mlog_errno(status); | |
5832 | ||
5833 | brelse(osb->osb_tl_bh); | |
5834 | iput(osb->osb_tl_inode); | |
5835 | } | |
5836 | ||
5837 | mlog_exit_void(); | |
5838 | } | |
5839 | ||
5840 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
5841 | { | |
5842 | int status; | |
5843 | struct inode *tl_inode = NULL; | |
5844 | struct buffer_head *tl_bh = NULL; | |
5845 | ||
5846 | mlog_entry_void(); | |
5847 | ||
5848 | status = ocfs2_get_truncate_log_info(osb, | |
5849 | osb->slot_num, | |
5850 | &tl_inode, | |
5851 | &tl_bh); | |
5852 | if (status < 0) | |
5853 | mlog_errno(status); | |
5854 | ||
5855 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
5856 | * osb->osb_tl_inode so we don't set any of the osb variables | |
5857 | * until we're sure all is well. */ | |
c4028958 DH |
5858 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
5859 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
5860 | osb->osb_tl_bh = tl_bh; |
5861 | osb->osb_tl_inode = tl_inode; | |
5862 | ||
5863 | mlog_exit(status); | |
5864 | return status; | |
5865 | } | |
5866 | ||
2b604351 MF |
5867 | /* |
5868 | * Delayed de-allocation of suballocator blocks. | |
5869 | * | |
5870 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
5871 | * | |
5872 | * The locking for this can get extremely complicated, especially when | |
5873 | * the suballocator inodes to delete from aren't known until deep | |
5874 | * within an unrelated codepath. | |
5875 | * | |
5876 | * ocfs2_extent_block structures are a good example of this - an inode | |
5877 | * btree could have been grown by any number of nodes each allocating | |
5878 | * out of their own suballoc inode. | |
5879 | * | |
5880 | * These structures allow the delay of block de-allocation until a | |
5881 | * later time, when locking of multiple cluster inodes won't cause | |
5882 | * deadlock. | |
5883 | */ | |
5884 | ||
5885 | /* | |
5886 | * Describes a single block free from a suballocator | |
5887 | */ | |
5888 | struct ocfs2_cached_block_free { | |
5889 | struct ocfs2_cached_block_free *free_next; | |
5890 | u64 free_blk; | |
5891 | unsigned int free_bit; | |
5892 | }; | |
5893 | ||
5894 | struct ocfs2_per_slot_free_list { | |
5895 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
5896 | int f_inode_type; | |
5897 | int f_slot; | |
5898 | struct ocfs2_cached_block_free *f_first; | |
5899 | }; | |
5900 | ||
5901 | static int ocfs2_free_cached_items(struct ocfs2_super *osb, | |
5902 | int sysfile_type, | |
5903 | int slot, | |
5904 | struct ocfs2_cached_block_free *head) | |
5905 | { | |
5906 | int ret; | |
5907 | u64 bg_blkno; | |
5908 | handle_t *handle; | |
5909 | struct inode *inode; | |
5910 | struct buffer_head *di_bh = NULL; | |
5911 | struct ocfs2_cached_block_free *tmp; | |
5912 | ||
5913 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
5914 | if (!inode) { | |
5915 | ret = -EINVAL; | |
5916 | mlog_errno(ret); | |
5917 | goto out; | |
5918 | } | |
5919 | ||
5920 | mutex_lock(&inode->i_mutex); | |
5921 | ||
e63aecb6 | 5922 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
5923 | if (ret) { |
5924 | mlog_errno(ret); | |
5925 | goto out_mutex; | |
5926 | } | |
5927 | ||
5928 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
5929 | if (IS_ERR(handle)) { | |
5930 | ret = PTR_ERR(handle); | |
5931 | mlog_errno(ret); | |
5932 | goto out_unlock; | |
5933 | } | |
5934 | ||
5935 | while (head) { | |
5936 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
5937 | head->free_bit); | |
5938 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
5939 | head->free_bit, (unsigned long long)head->free_blk); | |
5940 | ||
5941 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
5942 | head->free_bit, bg_blkno, 1); | |
5943 | if (ret) { | |
5944 | mlog_errno(ret); | |
5945 | goto out_journal; | |
5946 | } | |
5947 | ||
5948 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
5949 | if (ret) { | |
5950 | mlog_errno(ret); | |
5951 | goto out_journal; | |
5952 | } | |
5953 | ||
5954 | tmp = head; | |
5955 | head = head->free_next; | |
5956 | kfree(tmp); | |
5957 | } | |
5958 | ||
5959 | out_journal: | |
5960 | ocfs2_commit_trans(osb, handle); | |
5961 | ||
5962 | out_unlock: | |
e63aecb6 | 5963 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
5964 | brelse(di_bh); |
5965 | out_mutex: | |
5966 | mutex_unlock(&inode->i_mutex); | |
5967 | iput(inode); | |
5968 | out: | |
5969 | while(head) { | |
5970 | /* Premature exit may have left some dangling items. */ | |
5971 | tmp = head; | |
5972 | head = head->free_next; | |
5973 | kfree(tmp); | |
5974 | } | |
5975 | ||
5976 | return ret; | |
5977 | } | |
5978 | ||
5979 | int ocfs2_run_deallocs(struct ocfs2_super *osb, | |
5980 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
5981 | { | |
5982 | int ret = 0, ret2; | |
5983 | struct ocfs2_per_slot_free_list *fl; | |
5984 | ||
5985 | if (!ctxt) | |
5986 | return 0; | |
5987 | ||
5988 | while (ctxt->c_first_suballocator) { | |
5989 | fl = ctxt->c_first_suballocator; | |
5990 | ||
5991 | if (fl->f_first) { | |
5992 | mlog(0, "Free items: (type %u, slot %d)\n", | |
5993 | fl->f_inode_type, fl->f_slot); | |
5994 | ret2 = ocfs2_free_cached_items(osb, fl->f_inode_type, | |
5995 | fl->f_slot, fl->f_first); | |
5996 | if (ret2) | |
5997 | mlog_errno(ret2); | |
5998 | if (!ret) | |
5999 | ret = ret2; | |
6000 | } | |
6001 | ||
6002 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6003 | kfree(fl); | |
6004 | } | |
6005 | ||
6006 | return ret; | |
6007 | } | |
6008 | ||
6009 | static struct ocfs2_per_slot_free_list * | |
6010 | ocfs2_find_per_slot_free_list(int type, | |
6011 | int slot, | |
6012 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6013 | { | |
6014 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6015 | ||
6016 | while (fl) { | |
6017 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6018 | return fl; | |
6019 | ||
6020 | fl = fl->f_next_suballocator; | |
6021 | } | |
6022 | ||
6023 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6024 | if (fl) { | |
6025 | fl->f_inode_type = type; | |
6026 | fl->f_slot = slot; | |
6027 | fl->f_first = NULL; | |
6028 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6029 | ||
6030 | ctxt->c_first_suballocator = fl; | |
6031 | } | |
6032 | return fl; | |
6033 | } | |
6034 | ||
6035 | static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, | |
6036 | int type, int slot, u64 blkno, | |
6037 | unsigned int bit) | |
6038 | { | |
6039 | int ret; | |
6040 | struct ocfs2_per_slot_free_list *fl; | |
6041 | struct ocfs2_cached_block_free *item; | |
6042 | ||
6043 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6044 | if (fl == NULL) { | |
6045 | ret = -ENOMEM; | |
6046 | mlog_errno(ret); | |
6047 | goto out; | |
6048 | } | |
6049 | ||
6050 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6051 | if (item == NULL) { | |
6052 | ret = -ENOMEM; | |
6053 | mlog_errno(ret); | |
6054 | goto out; | |
6055 | } | |
6056 | ||
6057 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6058 | type, slot, bit, (unsigned long long)blkno); | |
6059 | ||
6060 | item->free_blk = blkno; | |
6061 | item->free_bit = bit; | |
6062 | item->free_next = fl->f_first; | |
6063 | ||
6064 | fl->f_first = item; | |
6065 | ||
6066 | ret = 0; | |
6067 | out: | |
6068 | return ret; | |
6069 | } | |
6070 | ||
59a5e416 MF |
6071 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6072 | struct ocfs2_extent_block *eb) | |
6073 | { | |
6074 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6075 | le16_to_cpu(eb->h_suballoc_slot), | |
6076 | le64_to_cpu(eb->h_blkno), | |
6077 | le16_to_cpu(eb->h_suballoc_bit)); | |
6078 | } | |
6079 | ||
ccd979bd MF |
6080 | /* This function will figure out whether the currently last extent |
6081 | * block will be deleted, and if it will, what the new last extent | |
6082 | * block will be so we can update his h_next_leaf_blk field, as well | |
6083 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6084 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6085 | unsigned int clusters_to_del, |
dcd0538f | 6086 | struct ocfs2_path *path, |
ccd979bd MF |
6087 | struct buffer_head **new_last_eb) |
6088 | { | |
3a0782d0 | 6089 | int next_free, ret = 0; |
dcd0538f | 6090 | u32 cpos; |
3a0782d0 | 6091 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6092 | struct ocfs2_extent_block *eb; |
6093 | struct ocfs2_extent_list *el; | |
6094 | struct buffer_head *bh = NULL; | |
6095 | ||
6096 | *new_last_eb = NULL; | |
6097 | ||
ccd979bd | 6098 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6099 | if (!path->p_tree_depth) |
6100 | goto out; | |
ccd979bd MF |
6101 | |
6102 | /* trunc to zero special case - this makes tree_depth = 0 | |
6103 | * regardless of what it is. */ | |
3a0782d0 | 6104 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6105 | goto out; |
ccd979bd | 6106 | |
dcd0538f | 6107 | el = path_leaf_el(path); |
ccd979bd MF |
6108 | BUG_ON(!el->l_next_free_rec); |
6109 | ||
3a0782d0 MF |
6110 | /* |
6111 | * Make sure that this extent list will actually be empty | |
6112 | * after we clear away the data. We can shortcut out if | |
6113 | * there's more than one non-empty extent in the | |
6114 | * list. Otherwise, a check of the remaining extent is | |
6115 | * necessary. | |
6116 | */ | |
6117 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6118 | rec = NULL; | |
dcd0538f | 6119 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6120 | if (next_free > 2) |
dcd0538f | 6121 | goto out; |
3a0782d0 MF |
6122 | |
6123 | /* We may have a valid extent in index 1, check it. */ | |
6124 | if (next_free == 2) | |
6125 | rec = &el->l_recs[1]; | |
6126 | ||
6127 | /* | |
6128 | * Fall through - no more nonempty extents, so we want | |
6129 | * to delete this leaf. | |
6130 | */ | |
6131 | } else { | |
6132 | if (next_free > 1) | |
6133 | goto out; | |
6134 | ||
6135 | rec = &el->l_recs[0]; | |
6136 | } | |
6137 | ||
6138 | if (rec) { | |
6139 | /* | |
6140 | * Check it we'll only be trimming off the end of this | |
6141 | * cluster. | |
6142 | */ | |
e48edee2 | 6143 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6144 | goto out; |
6145 | } | |
ccd979bd | 6146 | |
dcd0538f MF |
6147 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6148 | if (ret) { | |
6149 | mlog_errno(ret); | |
6150 | goto out; | |
6151 | } | |
ccd979bd | 6152 | |
dcd0538f MF |
6153 | ret = ocfs2_find_leaf(inode, path_root_el(path), cpos, &bh); |
6154 | if (ret) { | |
6155 | mlog_errno(ret); | |
6156 | goto out; | |
6157 | } | |
ccd979bd | 6158 | |
dcd0538f MF |
6159 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6160 | el = &eb->h_list; | |
6161 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
6162 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
6163 | ret = -EROFS; | |
6164 | goto out; | |
6165 | } | |
ccd979bd MF |
6166 | |
6167 | *new_last_eb = bh; | |
6168 | get_bh(*new_last_eb); | |
dcd0538f MF |
6169 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6170 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6171 | out: | |
6172 | brelse(bh); | |
ccd979bd | 6173 | |
dcd0538f | 6174 | return ret; |
ccd979bd MF |
6175 | } |
6176 | ||
3a0782d0 MF |
6177 | /* |
6178 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6179 | * during truncate. | |
6180 | * | |
6181 | * The caller needs to: | |
6182 | * - start journaling of each path component. | |
6183 | * - compute and fully set up any new last ext block | |
6184 | */ | |
6185 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6186 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
6187 | u32 clusters_to_del, u64 *delete_start) | |
6188 | { | |
6189 | int ret, i, index = path->p_tree_depth; | |
6190 | u32 new_edge = 0; | |
6191 | u64 deleted_eb = 0; | |
6192 | struct buffer_head *bh; | |
6193 | struct ocfs2_extent_list *el; | |
6194 | struct ocfs2_extent_rec *rec; | |
6195 | ||
6196 | *delete_start = 0; | |
6197 | ||
6198 | while (index >= 0) { | |
6199 | bh = path->p_node[index].bh; | |
6200 | el = path->p_node[index].el; | |
6201 | ||
6202 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6203 | index, (unsigned long long)bh->b_blocknr); | |
6204 | ||
6205 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6206 | ||
6207 | if (index != | |
6208 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6209 | ocfs2_error(inode->i_sb, | |
6210 | "Inode %lu has invalid ext. block %llu", | |
6211 | inode->i_ino, | |
6212 | (unsigned long long)bh->b_blocknr); | |
6213 | ret = -EROFS; | |
6214 | goto out; | |
6215 | } | |
6216 | ||
6217 | find_tail_record: | |
6218 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6219 | rec = &el->l_recs[i]; | |
6220 | ||
6221 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6222 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6223 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6224 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6225 | le16_to_cpu(el->l_next_free_rec)); | |
6226 | ||
e48edee2 | 6227 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6228 | |
6229 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6230 | /* | |
6231 | * If the leaf block contains a single empty | |
6232 | * extent and no records, we can just remove | |
6233 | * the block. | |
6234 | */ | |
6235 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6236 | memset(rec, 0, | |
6237 | sizeof(struct ocfs2_extent_rec)); | |
6238 | el->l_next_free_rec = cpu_to_le16(0); | |
6239 | ||
6240 | goto delete; | |
6241 | } | |
6242 | ||
6243 | /* | |
6244 | * Remove any empty extents by shifting things | |
6245 | * left. That should make life much easier on | |
6246 | * the code below. This condition is rare | |
6247 | * enough that we shouldn't see a performance | |
6248 | * hit. | |
6249 | */ | |
6250 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6251 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6252 | ||
6253 | for(i = 0; | |
6254 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6255 | el->l_recs[i] = el->l_recs[i + 1]; | |
6256 | ||
6257 | memset(&el->l_recs[i], 0, | |
6258 | sizeof(struct ocfs2_extent_rec)); | |
6259 | ||
6260 | /* | |
6261 | * We've modified our extent list. The | |
6262 | * simplest way to handle this change | |
6263 | * is to being the search from the | |
6264 | * start again. | |
6265 | */ | |
6266 | goto find_tail_record; | |
6267 | } | |
6268 | ||
e48edee2 | 6269 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6270 | |
6271 | /* | |
6272 | * We'll use "new_edge" on our way back up the | |
6273 | * tree to know what our rightmost cpos is. | |
6274 | */ | |
e48edee2 | 6275 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6276 | new_edge += le32_to_cpu(rec->e_cpos); |
6277 | ||
6278 | /* | |
6279 | * The caller will use this to delete data blocks. | |
6280 | */ | |
6281 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6282 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6283 | le16_to_cpu(rec->e_leaf_clusters)); |
3a0782d0 MF |
6284 | |
6285 | /* | |
6286 | * If it's now empty, remove this record. | |
6287 | */ | |
e48edee2 | 6288 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6289 | memset(rec, 0, |
6290 | sizeof(struct ocfs2_extent_rec)); | |
6291 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6292 | } | |
6293 | } else { | |
6294 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6295 | memset(rec, 0, | |
6296 | sizeof(struct ocfs2_extent_rec)); | |
6297 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6298 | ||
6299 | goto delete; | |
6300 | } | |
6301 | ||
6302 | /* Can this actually happen? */ | |
6303 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6304 | goto delete; | |
6305 | ||
6306 | /* | |
6307 | * We never actually deleted any clusters | |
6308 | * because our leaf was empty. There's no | |
6309 | * reason to adjust the rightmost edge then. | |
6310 | */ | |
6311 | if (new_edge == 0) | |
6312 | goto delete; | |
6313 | ||
e48edee2 MF |
6314 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6315 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6316 | -le32_to_cpu(rec->e_cpos)); |
6317 | ||
6318 | /* | |
6319 | * A deleted child record should have been | |
6320 | * caught above. | |
6321 | */ | |
e48edee2 | 6322 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6323 | } |
6324 | ||
6325 | delete: | |
6326 | ret = ocfs2_journal_dirty(handle, bh); | |
6327 | if (ret) { | |
6328 | mlog_errno(ret); | |
6329 | goto out; | |
6330 | } | |
6331 | ||
6332 | mlog(0, "extent list container %llu, after: record %d: " | |
6333 | "(%u, %u, %llu), next = %u.\n", | |
6334 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6335 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6336 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6337 | le16_to_cpu(el->l_next_free_rec)); | |
6338 | ||
6339 | /* | |
6340 | * We must be careful to only attempt delete of an | |
6341 | * extent block (and not the root inode block). | |
6342 | */ | |
6343 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6344 | struct ocfs2_extent_block *eb = | |
6345 | (struct ocfs2_extent_block *)bh->b_data; | |
6346 | ||
6347 | /* | |
6348 | * Save this for use when processing the | |
6349 | * parent block. | |
6350 | */ | |
6351 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6352 | ||
6353 | mlog(0, "deleting this extent block.\n"); | |
6354 | ||
6355 | ocfs2_remove_from_cache(inode, bh); | |
6356 | ||
e48edee2 | 6357 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6358 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6359 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6360 | ||
59a5e416 MF |
6361 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6362 | /* An error here is not fatal. */ | |
6363 | if (ret < 0) | |
6364 | mlog_errno(ret); | |
3a0782d0 MF |
6365 | } else { |
6366 | deleted_eb = 0; | |
6367 | } | |
6368 | ||
6369 | index--; | |
6370 | } | |
6371 | ||
6372 | ret = 0; | |
6373 | out: | |
6374 | return ret; | |
6375 | } | |
6376 | ||
ccd979bd MF |
6377 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6378 | unsigned int clusters_to_del, | |
6379 | struct inode *inode, | |
6380 | struct buffer_head *fe_bh, | |
1fabe148 | 6381 | handle_t *handle, |
dcd0538f MF |
6382 | struct ocfs2_truncate_context *tc, |
6383 | struct ocfs2_path *path) | |
ccd979bd | 6384 | { |
3a0782d0 | 6385 | int status; |
ccd979bd | 6386 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6387 | struct ocfs2_extent_block *last_eb = NULL; |
6388 | struct ocfs2_extent_list *el; | |
ccd979bd | 6389 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
6390 | u64 delete_blk = 0; |
6391 | ||
6392 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6393 | ||
3a0782d0 | 6394 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6395 | path, &last_eb_bh); |
ccd979bd MF |
6396 | if (status < 0) { |
6397 | mlog_errno(status); | |
6398 | goto bail; | |
6399 | } | |
dcd0538f MF |
6400 | |
6401 | /* | |
6402 | * Each component will be touched, so we might as well journal | |
6403 | * here to avoid having to handle errors later. | |
6404 | */ | |
3a0782d0 MF |
6405 | status = ocfs2_journal_access_path(inode, handle, path); |
6406 | if (status < 0) { | |
6407 | mlog_errno(status); | |
6408 | goto bail; | |
dcd0538f MF |
6409 | } |
6410 | ||
6411 | if (last_eb_bh) { | |
6412 | status = ocfs2_journal_access(handle, inode, last_eb_bh, | |
6413 | OCFS2_JOURNAL_ACCESS_WRITE); | |
6414 | if (status < 0) { | |
6415 | mlog_errno(status); | |
6416 | goto bail; | |
6417 | } | |
6418 | ||
ccd979bd | 6419 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6420 | } |
ccd979bd | 6421 | |
dcd0538f MF |
6422 | el = &(fe->id2.i_list); |
6423 | ||
6424 | /* | |
6425 | * Lower levels depend on this never happening, but it's best | |
6426 | * to check it up here before changing the tree. | |
6427 | */ | |
e48edee2 | 6428 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6429 | ocfs2_error(inode->i_sb, |
6430 | "Inode %lu has an empty extent record, depth %u\n", | |
6431 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6432 | status = -EROFS; |
ccd979bd MF |
6433 | goto bail; |
6434 | } | |
ccd979bd MF |
6435 | |
6436 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
6437 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6438 | clusters_to_del; | |
6439 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6440 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6441 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6442 | |
3a0782d0 MF |
6443 | status = ocfs2_trim_tree(inode, path, handle, tc, |
6444 | clusters_to_del, &delete_blk); | |
6445 | if (status) { | |
6446 | mlog_errno(status); | |
6447 | goto bail; | |
ccd979bd MF |
6448 | } |
6449 | ||
dcd0538f | 6450 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6451 | /* trunc to zero is a special case. */ |
6452 | el->l_tree_depth = 0; | |
6453 | fe->i_last_eb_blk = 0; | |
6454 | } else if (last_eb) | |
6455 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6456 | ||
6457 | status = ocfs2_journal_dirty(handle, fe_bh); | |
6458 | if (status < 0) { | |
6459 | mlog_errno(status); | |
6460 | goto bail; | |
6461 | } | |
6462 | ||
6463 | if (last_eb) { | |
6464 | /* If there will be a new last extent block, then by | |
6465 | * definition, there cannot be any leaves to the right of | |
6466 | * him. */ | |
ccd979bd MF |
6467 | last_eb->h_next_leaf_blk = 0; |
6468 | status = ocfs2_journal_dirty(handle, last_eb_bh); | |
6469 | if (status < 0) { | |
6470 | mlog_errno(status); | |
6471 | goto bail; | |
6472 | } | |
6473 | } | |
6474 | ||
3a0782d0 MF |
6475 | if (delete_blk) { |
6476 | status = ocfs2_truncate_log_append(osb, handle, delete_blk, | |
6477 | clusters_to_del); | |
ccd979bd MF |
6478 | if (status < 0) { |
6479 | mlog_errno(status); | |
6480 | goto bail; | |
6481 | } | |
ccd979bd MF |
6482 | } |
6483 | status = 0; | |
6484 | bail: | |
dcd0538f | 6485 | |
ccd979bd MF |
6486 | mlog_exit(status); |
6487 | return status; | |
6488 | } | |
6489 | ||
60b11392 MF |
6490 | static int ocfs2_writeback_zero_func(handle_t *handle, struct buffer_head *bh) |
6491 | { | |
6492 | set_buffer_uptodate(bh); | |
6493 | mark_buffer_dirty(bh); | |
6494 | return 0; | |
6495 | } | |
6496 | ||
6497 | static int ocfs2_ordered_zero_func(handle_t *handle, struct buffer_head *bh) | |
6498 | { | |
6499 | set_buffer_uptodate(bh); | |
6500 | mark_buffer_dirty(bh); | |
6501 | return ocfs2_journal_dirty_data(handle, bh); | |
6502 | } | |
6503 | ||
1d410a6e MF |
6504 | static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6505 | unsigned int from, unsigned int to, | |
6506 | struct page *page, int zero, u64 *phys) | |
6507 | { | |
6508 | int ret, partial = 0; | |
6509 | ||
6510 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6511 | if (ret) | |
6512 | mlog_errno(ret); | |
6513 | ||
6514 | if (zero) | |
eebd2aa3 | 6515 | zero_user_segment(page, from, to); |
1d410a6e MF |
6516 | |
6517 | /* | |
6518 | * Need to set the buffers we zero'd into uptodate | |
6519 | * here if they aren't - ocfs2_map_page_blocks() | |
6520 | * might've skipped some | |
6521 | */ | |
6522 | if (ocfs2_should_order_data(inode)) { | |
6523 | ret = walk_page_buffers(handle, | |
6524 | page_buffers(page), | |
6525 | from, to, &partial, | |
6526 | ocfs2_ordered_zero_func); | |
6527 | if (ret < 0) | |
6528 | mlog_errno(ret); | |
6529 | } else { | |
6530 | ret = walk_page_buffers(handle, page_buffers(page), | |
6531 | from, to, &partial, | |
6532 | ocfs2_writeback_zero_func); | |
6533 | if (ret < 0) | |
6534 | mlog_errno(ret); | |
6535 | } | |
6536 | ||
6537 | if (!partial) | |
6538 | SetPageUptodate(page); | |
6539 | ||
6540 | flush_dcache_page(page); | |
6541 | } | |
6542 | ||
35edec1d MF |
6543 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6544 | loff_t end, struct page **pages, | |
6545 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6546 | { |
1d410a6e | 6547 | int i; |
60b11392 MF |
6548 | struct page *page; |
6549 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6550 | struct super_block *sb = inode->i_sb; | |
6551 | ||
6552 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6553 | ||
6554 | if (numpages == 0) | |
6555 | goto out; | |
6556 | ||
35edec1d | 6557 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6558 | for(i = 0; i < numpages; i++) { |
6559 | page = pages[i]; | |
6560 | ||
35edec1d MF |
6561 | from = start & (PAGE_CACHE_SIZE - 1); |
6562 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6563 | to = end & (PAGE_CACHE_SIZE - 1); | |
6564 | ||
60b11392 MF |
6565 | BUG_ON(from > PAGE_CACHE_SIZE); |
6566 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6567 | ||
1d410a6e MF |
6568 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6569 | &phys); | |
60b11392 | 6570 | |
35edec1d | 6571 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6572 | } |
6573 | out: | |
1d410a6e MF |
6574 | if (pages) |
6575 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6576 | } |
6577 | ||
35edec1d | 6578 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
1d410a6e | 6579 | struct page **pages, int *num) |
60b11392 | 6580 | { |
1d410a6e | 6581 | int numpages, ret = 0; |
60b11392 MF |
6582 | struct super_block *sb = inode->i_sb; |
6583 | struct address_space *mapping = inode->i_mapping; | |
6584 | unsigned long index; | |
35edec1d | 6585 | loff_t last_page_bytes; |
60b11392 | 6586 | |
35edec1d | 6587 | BUG_ON(start > end); |
60b11392 | 6588 | |
35edec1d MF |
6589 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != |
6590 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6591 | ||
1d410a6e | 6592 | numpages = 0; |
35edec1d MF |
6593 | last_page_bytes = PAGE_ALIGN(end); |
6594 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
6595 | do { |
6596 | pages[numpages] = grab_cache_page(mapping, index); | |
6597 | if (!pages[numpages]) { | |
6598 | ret = -ENOMEM; | |
6599 | mlog_errno(ret); | |
6600 | goto out; | |
6601 | } | |
6602 | ||
6603 | numpages++; | |
6604 | index++; | |
35edec1d | 6605 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6606 | |
6607 | out: | |
6608 | if (ret != 0) { | |
1d410a6e MF |
6609 | if (pages) |
6610 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6611 | numpages = 0; |
6612 | } | |
6613 | ||
6614 | *num = numpages; | |
6615 | ||
6616 | return ret; | |
6617 | } | |
6618 | ||
6619 | /* | |
6620 | * Zero the area past i_size but still within an allocated | |
6621 | * cluster. This avoids exposing nonzero data on subsequent file | |
6622 | * extends. | |
6623 | * | |
6624 | * We need to call this before i_size is updated on the inode because | |
6625 | * otherwise block_write_full_page() will skip writeout of pages past | |
6626 | * i_size. The new_i_size parameter is passed for this reason. | |
6627 | */ | |
35edec1d MF |
6628 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6629 | u64 range_start, u64 range_end) | |
60b11392 | 6630 | { |
1d410a6e | 6631 | int ret = 0, numpages; |
60b11392 MF |
6632 | struct page **pages = NULL; |
6633 | u64 phys; | |
1d410a6e MF |
6634 | unsigned int ext_flags; |
6635 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6636 | |
6637 | /* | |
6638 | * File systems which don't support sparse files zero on every | |
6639 | * extend. | |
6640 | */ | |
1d410a6e | 6641 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
6642 | return 0; |
6643 | ||
1d410a6e | 6644 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
6645 | sizeof(struct page *), GFP_NOFS); |
6646 | if (pages == NULL) { | |
6647 | ret = -ENOMEM; | |
6648 | mlog_errno(ret); | |
6649 | goto out; | |
6650 | } | |
6651 | ||
1d410a6e MF |
6652 | if (range_start == range_end) |
6653 | goto out; | |
6654 | ||
6655 | ret = ocfs2_extent_map_get_blocks(inode, | |
6656 | range_start >> sb->s_blocksize_bits, | |
6657 | &phys, NULL, &ext_flags); | |
60b11392 MF |
6658 | if (ret) { |
6659 | mlog_errno(ret); | |
6660 | goto out; | |
6661 | } | |
6662 | ||
1d410a6e MF |
6663 | /* |
6664 | * Tail is a hole, or is marked unwritten. In either case, we | |
6665 | * can count on read and write to return/push zero's. | |
6666 | */ | |
6667 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
6668 | goto out; |
6669 | ||
1d410a6e MF |
6670 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
6671 | &numpages); | |
6672 | if (ret) { | |
6673 | mlog_errno(ret); | |
6674 | goto out; | |
6675 | } | |
6676 | ||
35edec1d MF |
6677 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
6678 | numpages, phys, handle); | |
60b11392 MF |
6679 | |
6680 | /* | |
6681 | * Initiate writeout of the pages we zero'd here. We don't | |
6682 | * wait on them - the truncate_inode_pages() call later will | |
6683 | * do that for us. | |
6684 | */ | |
35edec1d MF |
6685 | ret = do_sync_mapping_range(inode->i_mapping, range_start, |
6686 | range_end - 1, SYNC_FILE_RANGE_WRITE); | |
60b11392 MF |
6687 | if (ret) |
6688 | mlog_errno(ret); | |
6689 | ||
6690 | out: | |
6691 | if (pages) | |
6692 | kfree(pages); | |
6693 | ||
6694 | return ret; | |
6695 | } | |
6696 | ||
fdd77704 TY |
6697 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
6698 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
6699 | { |
6700 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 6701 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 6702 | |
fdd77704 TY |
6703 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
6704 | memset(&di->id2, 0, blocksize - | |
6705 | offsetof(struct ocfs2_dinode, id2) - | |
6706 | xattrsize); | |
6707 | else | |
6708 | memset(&di->id2, 0, blocksize - | |
6709 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
6710 | } |
6711 | ||
5b6a3a2b MF |
6712 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
6713 | struct ocfs2_dinode *di) | |
6714 | { | |
fdd77704 | 6715 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
6716 | di->id2.i_list.l_tree_depth = 0; |
6717 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
6718 | di->id2.i_list.l_count = cpu_to_le16( |
6719 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
6720 | } |
6721 | ||
1afc32b9 MF |
6722 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
6723 | { | |
6724 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6725 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
6726 | ||
6727 | spin_lock(&oi->ip_lock); | |
6728 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
6729 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6730 | spin_unlock(&oi->ip_lock); | |
6731 | ||
6732 | /* | |
6733 | * We clear the entire i_data structure here so that all | |
6734 | * fields can be properly initialized. | |
6735 | */ | |
fdd77704 | 6736 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 6737 | |
fdd77704 TY |
6738 | idata->id_count = cpu_to_le16( |
6739 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
6740 | } |
6741 | ||
6742 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
6743 | struct buffer_head *di_bh) | |
6744 | { | |
6745 | int ret, i, has_data, num_pages = 0; | |
6746 | handle_t *handle; | |
6747 | u64 uninitialized_var(block); | |
6748 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6749 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
6750 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
6751 | struct ocfs2_alloc_context *data_ac = NULL; |
6752 | struct page **pages = NULL; | |
6753 | loff_t end = osb->s_clustersize; | |
6754 | ||
6755 | has_data = i_size_read(inode) ? 1 : 0; | |
6756 | ||
6757 | if (has_data) { | |
6758 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
6759 | sizeof(struct page *), GFP_NOFS); | |
6760 | if (pages == NULL) { | |
6761 | ret = -ENOMEM; | |
6762 | mlog_errno(ret); | |
6763 | goto out; | |
6764 | } | |
6765 | ||
6766 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
6767 | if (ret) { | |
6768 | mlog_errno(ret); | |
6769 | goto out; | |
6770 | } | |
6771 | } | |
6772 | ||
6773 | handle = ocfs2_start_trans(osb, OCFS2_INLINE_TO_EXTENTS_CREDITS); | |
6774 | if (IS_ERR(handle)) { | |
6775 | ret = PTR_ERR(handle); | |
6776 | mlog_errno(ret); | |
6777 | goto out_unlock; | |
6778 | } | |
6779 | ||
6780 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
6781 | OCFS2_JOURNAL_ACCESS_WRITE); | |
6782 | if (ret) { | |
6783 | mlog_errno(ret); | |
6784 | goto out_commit; | |
6785 | } | |
6786 | ||
6787 | if (has_data) { | |
6788 | u32 bit_off, num; | |
6789 | unsigned int page_end; | |
6790 | u64 phys; | |
6791 | ||
6792 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, | |
6793 | &num); | |
6794 | if (ret) { | |
6795 | mlog_errno(ret); | |
6796 | goto out_commit; | |
6797 | } | |
6798 | ||
6799 | /* | |
6800 | * Save two copies, one for insert, and one that can | |
6801 | * be changed by ocfs2_map_and_dirty_page() below. | |
6802 | */ | |
6803 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
6804 | ||
6805 | /* | |
6806 | * Non sparse file systems zero on extend, so no need | |
6807 | * to do that now. | |
6808 | */ | |
6809 | if (!ocfs2_sparse_alloc(osb) && | |
6810 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
6811 | end = PAGE_CACHE_SIZE; | |
6812 | ||
6813 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
6814 | if (ret) { | |
6815 | mlog_errno(ret); | |
6816 | goto out_commit; | |
6817 | } | |
6818 | ||
6819 | /* | |
6820 | * This should populate the 1st page for us and mark | |
6821 | * it up to date. | |
6822 | */ | |
6823 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
6824 | if (ret) { | |
6825 | mlog_errno(ret); | |
6826 | goto out_commit; | |
6827 | } | |
6828 | ||
6829 | page_end = PAGE_CACHE_SIZE; | |
6830 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
6831 | page_end = osb->s_clustersize; | |
6832 | ||
6833 | for (i = 0; i < num_pages; i++) | |
6834 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
6835 | pages[i], i > 0, &phys); | |
6836 | } | |
6837 | ||
6838 | spin_lock(&oi->ip_lock); | |
6839 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
6840 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6841 | spin_unlock(&oi->ip_lock); | |
6842 | ||
5b6a3a2b | 6843 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
6844 | |
6845 | ocfs2_journal_dirty(handle, di_bh); | |
6846 | ||
6847 | if (has_data) { | |
6848 | /* | |
6849 | * An error at this point should be extremely rare. If | |
6850 | * this proves to be false, we could always re-build | |
6851 | * the in-inode data from our pages. | |
6852 | */ | |
f56654c4 TM |
6853 | ret = ocfs2_dinode_insert_extent(osb, handle, inode, di_bh, |
6854 | 0, block, 1, 0, NULL); | |
1afc32b9 MF |
6855 | if (ret) { |
6856 | mlog_errno(ret); | |
6857 | goto out_commit; | |
6858 | } | |
6859 | ||
6860 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
6861 | } | |
6862 | ||
6863 | out_commit: | |
6864 | ocfs2_commit_trans(osb, handle); | |
6865 | ||
6866 | out_unlock: | |
6867 | if (data_ac) | |
6868 | ocfs2_free_alloc_context(data_ac); | |
6869 | ||
6870 | out: | |
6871 | if (pages) { | |
6872 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
6873 | kfree(pages); | |
6874 | } | |
6875 | ||
6876 | return ret; | |
6877 | } | |
6878 | ||
ccd979bd MF |
6879 | /* |
6880 | * It is expected, that by the time you call this function, | |
6881 | * inode->i_size and fe->i_size have been adjusted. | |
6882 | * | |
6883 | * WARNING: This will kfree the truncate context | |
6884 | */ | |
6885 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
6886 | struct inode *inode, | |
6887 | struct buffer_head *fe_bh, | |
6888 | struct ocfs2_truncate_context *tc) | |
6889 | { | |
6890 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 6891 | u32 clusters_to_del, new_highest_cpos, range; |
ccd979bd | 6892 | struct ocfs2_extent_list *el; |
1fabe148 | 6893 | handle_t *handle = NULL; |
ccd979bd | 6894 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 6895 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 6896 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
ccd979bd MF |
6897 | |
6898 | mlog_entry_void(); | |
6899 | ||
dcd0538f | 6900 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
6901 | i_size_read(inode)); |
6902 | ||
e7d4cb6b | 6903 | path = ocfs2_new_path(fe_bh, &di->id2.i_list); |
dcd0538f MF |
6904 | if (!path) { |
6905 | status = -ENOMEM; | |
6906 | mlog_errno(status); | |
6907 | goto bail; | |
6908 | } | |
83418978 MF |
6909 | |
6910 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
6911 | ||
ccd979bd | 6912 | start: |
3a0782d0 MF |
6913 | /* |
6914 | * Check that we still have allocation to delete. | |
6915 | */ | |
6916 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
6917 | status = 0; | |
6918 | goto bail; | |
6919 | } | |
6920 | ||
dcd0538f MF |
6921 | /* |
6922 | * Truncate always works against the rightmost tree branch. | |
6923 | */ | |
6924 | status = ocfs2_find_path(inode, path, UINT_MAX); | |
6925 | if (status) { | |
6926 | mlog_errno(status); | |
6927 | goto bail; | |
ccd979bd MF |
6928 | } |
6929 | ||
dcd0538f MF |
6930 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
6931 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
6932 | ||
6933 | /* | |
6934 | * By now, el will point to the extent list on the bottom most | |
6935 | * portion of this tree. Only the tail record is considered in | |
6936 | * each pass. | |
6937 | * | |
6938 | * We handle the following cases, in order: | |
6939 | * - empty extent: delete the remaining branch | |
6940 | * - remove the entire record | |
6941 | * - remove a partial record | |
6942 | * - no record needs to be removed (truncate has completed) | |
6943 | */ | |
6944 | el = path_leaf_el(path); | |
3a0782d0 MF |
6945 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
6946 | ocfs2_error(inode->i_sb, | |
6947 | "Inode %llu has empty extent block at %llu\n", | |
6948 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
6949 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
6950 | status = -EROFS; | |
6951 | goto bail; | |
6952 | } | |
6953 | ||
ccd979bd | 6954 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 6955 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 6956 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
6957 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
6958 | clusters_to_del = 0; | |
6959 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 6960 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f | 6961 | } else if (range > new_highest_cpos) { |
e48edee2 | 6962 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 6963 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f MF |
6964 | new_highest_cpos; |
6965 | } else { | |
6966 | status = 0; | |
6967 | goto bail; | |
6968 | } | |
ccd979bd | 6969 | |
dcd0538f MF |
6970 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
6971 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
6972 | ||
1b1dcc1b | 6973 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6974 | tl_sem = 1; |
6975 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
6976 | * record is free for use. If there isn't any, we flush to get | |
6977 | * an empty truncate log. */ | |
6978 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6979 | status = __ocfs2_flush_truncate_log(osb); | |
6980 | if (status < 0) { | |
6981 | mlog_errno(status); | |
6982 | goto bail; | |
6983 | } | |
6984 | } | |
6985 | ||
6986 | credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, | |
dcd0538f MF |
6987 | (struct ocfs2_dinode *)fe_bh->b_data, |
6988 | el); | |
65eff9cc | 6989 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
6990 | if (IS_ERR(handle)) { |
6991 | status = PTR_ERR(handle); | |
6992 | handle = NULL; | |
6993 | mlog_errno(status); | |
6994 | goto bail; | |
6995 | } | |
6996 | ||
dcd0538f MF |
6997 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
6998 | tc, path); | |
ccd979bd MF |
6999 | if (status < 0) { |
7000 | mlog_errno(status); | |
7001 | goto bail; | |
7002 | } | |
7003 | ||
1b1dcc1b | 7004 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7005 | tl_sem = 0; |
7006 | ||
02dc1af4 | 7007 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
7008 | handle = NULL; |
7009 | ||
dcd0538f MF |
7010 | ocfs2_reinit_path(path, 1); |
7011 | ||
7012 | /* | |
3a0782d0 MF |
7013 | * The check above will catch the case where we've truncated |
7014 | * away all allocation. | |
dcd0538f | 7015 | */ |
3a0782d0 MF |
7016 | goto start; |
7017 | ||
ccd979bd | 7018 | bail: |
ccd979bd MF |
7019 | |
7020 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7021 | ||
7022 | if (tl_sem) | |
1b1dcc1b | 7023 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7024 | |
7025 | if (handle) | |
02dc1af4 | 7026 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 7027 | |
59a5e416 MF |
7028 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
7029 | ||
dcd0538f | 7030 | ocfs2_free_path(path); |
ccd979bd MF |
7031 | |
7032 | /* This will drop the ext_alloc cluster lock for us */ | |
7033 | ocfs2_free_truncate_context(tc); | |
7034 | ||
7035 | mlog_exit(status); | |
7036 | return status; | |
7037 | } | |
7038 | ||
ccd979bd | 7039 | /* |
59a5e416 | 7040 | * Expects the inode to already be locked. |
ccd979bd MF |
7041 | */ |
7042 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7043 | struct inode *inode, | |
7044 | struct buffer_head *fe_bh, | |
7045 | struct ocfs2_truncate_context **tc) | |
7046 | { | |
59a5e416 | 7047 | int status; |
ccd979bd MF |
7048 | unsigned int new_i_clusters; |
7049 | struct ocfs2_dinode *fe; | |
7050 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7051 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7052 | |
7053 | mlog_entry_void(); | |
7054 | ||
7055 | *tc = NULL; | |
7056 | ||
7057 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7058 | i_size_read(inode)); | |
7059 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7060 | ||
7061 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7062 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7063 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7064 | |
cd861280 | 7065 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7066 | if (!(*tc)) { |
7067 | status = -ENOMEM; | |
7068 | mlog_errno(status); | |
7069 | goto bail; | |
7070 | } | |
59a5e416 | 7071 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7072 | |
ccd979bd | 7073 | if (fe->id2.i_list.l_tree_depth) { |
ccd979bd MF |
7074 | status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk), |
7075 | &last_eb_bh, OCFS2_BH_CACHED, inode); | |
7076 | if (status < 0) { | |
7077 | mlog_errno(status); | |
7078 | goto bail; | |
7079 | } | |
7080 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
7081 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
7082 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
7083 | ||
7084 | brelse(last_eb_bh); | |
7085 | status = -EIO; | |
7086 | goto bail; | |
7087 | } | |
ccd979bd MF |
7088 | } |
7089 | ||
7090 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7091 | ||
ccd979bd MF |
7092 | status = 0; |
7093 | bail: | |
7094 | if (status < 0) { | |
7095 | if (*tc) | |
7096 | ocfs2_free_truncate_context(*tc); | |
7097 | *tc = NULL; | |
7098 | } | |
7099 | mlog_exit_void(); | |
7100 | return status; | |
7101 | } | |
7102 | ||
1afc32b9 MF |
7103 | /* |
7104 | * 'start' is inclusive, 'end' is not. | |
7105 | */ | |
7106 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7107 | unsigned int start, unsigned int end, int trunc) | |
7108 | { | |
7109 | int ret; | |
7110 | unsigned int numbytes; | |
7111 | handle_t *handle; | |
7112 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7113 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7114 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7115 | ||
7116 | if (end > i_size_read(inode)) | |
7117 | end = i_size_read(inode); | |
7118 | ||
7119 | BUG_ON(start >= end); | |
7120 | ||
7121 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7122 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7123 | !ocfs2_supports_inline_data(osb)) { | |
7124 | ocfs2_error(inode->i_sb, | |
7125 | "Inline data flags for inode %llu don't agree! " | |
7126 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7127 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7128 | le16_to_cpu(di->i_dyn_features), | |
7129 | OCFS2_I(inode)->ip_dyn_features, | |
7130 | osb->s_feature_incompat); | |
7131 | ret = -EROFS; | |
7132 | goto out; | |
7133 | } | |
7134 | ||
7135 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7136 | if (IS_ERR(handle)) { | |
7137 | ret = PTR_ERR(handle); | |
7138 | mlog_errno(ret); | |
7139 | goto out; | |
7140 | } | |
7141 | ||
7142 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
7143 | OCFS2_JOURNAL_ACCESS_WRITE); | |
7144 | if (ret) { | |
7145 | mlog_errno(ret); | |
7146 | goto out_commit; | |
7147 | } | |
7148 | ||
7149 | numbytes = end - start; | |
7150 | memset(idata->id_data + start, 0, numbytes); | |
7151 | ||
7152 | /* | |
7153 | * No need to worry about the data page here - it's been | |
7154 | * truncated already and inline data doesn't need it for | |
7155 | * pushing zero's to disk, so we'll let readpage pick it up | |
7156 | * later. | |
7157 | */ | |
7158 | if (trunc) { | |
7159 | i_size_write(inode, start); | |
7160 | di->i_size = cpu_to_le64(start); | |
7161 | } | |
7162 | ||
7163 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7164 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7165 | ||
7166 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7167 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7168 | ||
7169 | ocfs2_journal_dirty(handle, di_bh); | |
7170 | ||
7171 | out_commit: | |
7172 | ocfs2_commit_trans(osb, handle); | |
7173 | ||
7174 | out: | |
7175 | return ret; | |
7176 | } | |
7177 | ||
ccd979bd MF |
7178 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7179 | { | |
59a5e416 MF |
7180 | /* |
7181 | * The caller is responsible for completing deallocation | |
7182 | * before freeing the context. | |
7183 | */ | |
7184 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7185 | mlog(ML_NOTICE, | |
7186 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd MF |
7187 | |
7188 | if (tc->tc_last_eb_bh) | |
7189 | brelse(tc->tc_last_eb_bh); | |
7190 | ||
7191 | kfree(tc); | |
7192 | } |