Commit | Line | Data |
---|---|---|
a86c6181 AT |
1 | /* |
2 | * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com | |
3 | * Written by Alex Tomas <alex@clusterfs.com> | |
4 | * | |
5 | * Architecture independence: | |
6 | * Copyright (c) 2005, Bull S.A. | |
7 | * Written by Pierre Peiffer <pierre.peiffer@bull.net> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public Licens | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Extents support for EXT4 | |
25 | * | |
26 | * TODO: | |
27 | * - ext4*_error() should be used in some situations | |
28 | * - analyze all BUG()/BUG_ON(), use -EIO where appropriate | |
29 | * - smart tree reduction | |
30 | */ | |
31 | ||
32 | #include <linux/module.h> | |
33 | #include <linux/fs.h> | |
34 | #include <linux/time.h> | |
cd02ff0b | 35 | #include <linux/jbd2.h> |
a86c6181 AT |
36 | #include <linux/highuid.h> |
37 | #include <linux/pagemap.h> | |
38 | #include <linux/quotaops.h> | |
39 | #include <linux/string.h> | |
40 | #include <linux/slab.h> | |
a2df2a63 | 41 | #include <linux/falloc.h> |
a86c6181 | 42 | #include <asm/uaccess.h> |
6873fa0d | 43 | #include <linux/fiemap.h> |
3dcf5451 CH |
44 | #include "ext4_jbd2.h" |
45 | #include "ext4_extents.h" | |
a86c6181 AT |
46 | |
47 | ||
d0d856e8 RD |
48 | /* |
49 | * ext_pblock: | |
50 | * combine low and high parts of physical block number into ext4_fsblk_t | |
51 | */ | |
09b88252 | 52 | static ext4_fsblk_t ext_pblock(struct ext4_extent *ex) |
f65e6fba AT |
53 | { |
54 | ext4_fsblk_t block; | |
55 | ||
b377611d | 56 | block = le32_to_cpu(ex->ee_start_lo); |
9b8f1f01 | 57 | block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1; |
f65e6fba AT |
58 | return block; |
59 | } | |
60 | ||
d0d856e8 RD |
61 | /* |
62 | * idx_pblock: | |
63 | * combine low and high parts of a leaf physical block number into ext4_fsblk_t | |
64 | */ | |
c14c6fd5 | 65 | ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix) |
f65e6fba AT |
66 | { |
67 | ext4_fsblk_t block; | |
68 | ||
d8dd0b45 | 69 | block = le32_to_cpu(ix->ei_leaf_lo); |
9b8f1f01 | 70 | block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1; |
f65e6fba AT |
71 | return block; |
72 | } | |
73 | ||
d0d856e8 RD |
74 | /* |
75 | * ext4_ext_store_pblock: | |
76 | * stores a large physical block number into an extent struct, | |
77 | * breaking it into parts | |
78 | */ | |
c14c6fd5 | 79 | void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb) |
f65e6fba | 80 | { |
b377611d | 81 | ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
9b8f1f01 | 82 | ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
f65e6fba AT |
83 | } |
84 | ||
d0d856e8 RD |
85 | /* |
86 | * ext4_idx_store_pblock: | |
87 | * stores a large physical block number into an index struct, | |
88 | * breaking it into parts | |
89 | */ | |
09b88252 | 90 | static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb) |
f65e6fba | 91 | { |
d8dd0b45 | 92 | ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
9b8f1f01 | 93 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
f65e6fba AT |
94 | } |
95 | ||
9102e4fa | 96 | static int ext4_ext_journal_restart(handle_t *handle, int needed) |
a86c6181 AT |
97 | { |
98 | int err; | |
99 | ||
0390131b FM |
100 | if (!ext4_handle_valid(handle)) |
101 | return 0; | |
a86c6181 | 102 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
103 | return 0; |
104 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 105 | if (err <= 0) |
9102e4fa SF |
106 | return err; |
107 | return ext4_journal_restart(handle, needed); | |
a86c6181 AT |
108 | } |
109 | ||
110 | /* | |
111 | * could return: | |
112 | * - EROFS | |
113 | * - ENOMEM | |
114 | */ | |
115 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
116 | struct ext4_ext_path *path) | |
117 | { | |
118 | if (path->p_bh) { | |
119 | /* path points to block */ | |
120 | return ext4_journal_get_write_access(handle, path->p_bh); | |
121 | } | |
122 | /* path points to leaf/index in inode body */ | |
123 | /* we use in-core data, no need to protect them */ | |
124 | return 0; | |
125 | } | |
126 | ||
127 | /* | |
128 | * could return: | |
129 | * - EROFS | |
130 | * - ENOMEM | |
131 | * - EIO | |
132 | */ | |
133 | static int ext4_ext_dirty(handle_t *handle, struct inode *inode, | |
134 | struct ext4_ext_path *path) | |
135 | { | |
136 | int err; | |
137 | if (path->p_bh) { | |
138 | /* path points to block */ | |
0390131b | 139 | err = ext4_handle_dirty_metadata(handle, inode, path->p_bh); |
a86c6181 AT |
140 | } else { |
141 | /* path points to leaf/index in inode body */ | |
142 | err = ext4_mark_inode_dirty(handle, inode); | |
143 | } | |
144 | return err; | |
145 | } | |
146 | ||
f65e6fba | 147 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 148 | struct ext4_ext_path *path, |
725d26d3 | 149 | ext4_lblk_t block) |
a86c6181 AT |
150 | { |
151 | struct ext4_inode_info *ei = EXT4_I(inode); | |
f65e6fba | 152 | ext4_fsblk_t bg_start; |
74d3487f | 153 | ext4_fsblk_t last_block; |
f65e6fba | 154 | ext4_grpblk_t colour; |
a86c6181 AT |
155 | int depth; |
156 | ||
157 | if (path) { | |
158 | struct ext4_extent *ex; | |
159 | depth = path->p_depth; | |
160 | ||
161 | /* try to predict block placement */ | |
7e028976 AM |
162 | ex = path[depth].p_ext; |
163 | if (ex) | |
f65e6fba | 164 | return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block)); |
a86c6181 | 165 | |
d0d856e8 RD |
166 | /* it looks like index is empty; |
167 | * try to find starting block from index itself */ | |
a86c6181 AT |
168 | if (path[depth].p_bh) |
169 | return path[depth].p_bh->b_blocknr; | |
170 | } | |
171 | ||
172 | /* OK. use inode's group */ | |
173 | bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) + | |
174 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block); | |
74d3487f VC |
175 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
176 | ||
177 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) | |
178 | colour = (current->pid % 16) * | |
a86c6181 | 179 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
180 | else |
181 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
a86c6181 AT |
182 | return bg_start + colour + block; |
183 | } | |
184 | ||
654b4908 AK |
185 | /* |
186 | * Allocation for a meta data block | |
187 | */ | |
f65e6fba | 188 | static ext4_fsblk_t |
654b4908 | 189 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 AT |
190 | struct ext4_ext_path *path, |
191 | struct ext4_extent *ex, int *err) | |
192 | { | |
f65e6fba | 193 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
194 | |
195 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
97df5d15 | 196 | newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err); |
a86c6181 AT |
197 | return newblock; |
198 | } | |
199 | ||
09b88252 | 200 | static int ext4_ext_space_block(struct inode *inode) |
a86c6181 AT |
201 | { |
202 | int size; | |
203 | ||
204 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
205 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 206 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
207 | if (size > 6) |
208 | size = 6; | |
209 | #endif | |
210 | return size; | |
211 | } | |
212 | ||
09b88252 | 213 | static int ext4_ext_space_block_idx(struct inode *inode) |
a86c6181 AT |
214 | { |
215 | int size; | |
216 | ||
217 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
218 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 219 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
220 | if (size > 5) |
221 | size = 5; | |
222 | #endif | |
223 | return size; | |
224 | } | |
225 | ||
09b88252 | 226 | static int ext4_ext_space_root(struct inode *inode) |
a86c6181 AT |
227 | { |
228 | int size; | |
229 | ||
230 | size = sizeof(EXT4_I(inode)->i_data); | |
231 | size -= sizeof(struct ext4_extent_header); | |
232 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 233 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
234 | if (size > 3) |
235 | size = 3; | |
236 | #endif | |
237 | return size; | |
238 | } | |
239 | ||
09b88252 | 240 | static int ext4_ext_space_root_idx(struct inode *inode) |
a86c6181 AT |
241 | { |
242 | int size; | |
243 | ||
244 | size = sizeof(EXT4_I(inode)->i_data); | |
245 | size -= sizeof(struct ext4_extent_header); | |
246 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 247 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
248 | if (size > 4) |
249 | size = 4; | |
250 | #endif | |
251 | return size; | |
252 | } | |
253 | ||
d2a17637 MC |
254 | /* |
255 | * Calculate the number of metadata blocks needed | |
256 | * to allocate @blocks | |
257 | * Worse case is one block per extent | |
258 | */ | |
259 | int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks) | |
260 | { | |
261 | int lcap, icap, rcap, leafs, idxs, num; | |
262 | int newextents = blocks; | |
263 | ||
264 | rcap = ext4_ext_space_root_idx(inode); | |
265 | lcap = ext4_ext_space_block(inode); | |
266 | icap = ext4_ext_space_block_idx(inode); | |
267 | ||
268 | /* number of new leaf blocks needed */ | |
269 | num = leafs = (newextents + lcap - 1) / lcap; | |
270 | ||
271 | /* | |
272 | * Worse case, we need separate index block(s) | |
273 | * to link all new leaf blocks | |
274 | */ | |
275 | idxs = (leafs + icap - 1) / icap; | |
276 | do { | |
277 | num += idxs; | |
278 | idxs = (idxs + icap - 1) / icap; | |
279 | } while (idxs > rcap); | |
280 | ||
281 | return num; | |
282 | } | |
283 | ||
c29c0ae7 AT |
284 | static int |
285 | ext4_ext_max_entries(struct inode *inode, int depth) | |
286 | { | |
287 | int max; | |
288 | ||
289 | if (depth == ext_depth(inode)) { | |
290 | if (depth == 0) | |
291 | max = ext4_ext_space_root(inode); | |
292 | else | |
293 | max = ext4_ext_space_root_idx(inode); | |
294 | } else { | |
295 | if (depth == 0) | |
296 | max = ext4_ext_space_block(inode); | |
297 | else | |
298 | max = ext4_ext_space_block_idx(inode); | |
299 | } | |
300 | ||
301 | return max; | |
302 | } | |
303 | ||
304 | static int __ext4_ext_check_header(const char *function, struct inode *inode, | |
305 | struct ext4_extent_header *eh, | |
306 | int depth) | |
307 | { | |
308 | const char *error_msg; | |
309 | int max = 0; | |
310 | ||
311 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
312 | error_msg = "invalid magic"; | |
313 | goto corrupted; | |
314 | } | |
315 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
316 | error_msg = "unexpected eh_depth"; | |
317 | goto corrupted; | |
318 | } | |
319 | if (unlikely(eh->eh_max == 0)) { | |
320 | error_msg = "invalid eh_max"; | |
321 | goto corrupted; | |
322 | } | |
323 | max = ext4_ext_max_entries(inode, depth); | |
324 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
325 | error_msg = "too large eh_max"; | |
326 | goto corrupted; | |
327 | } | |
328 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
329 | error_msg = "invalid eh_entries"; | |
330 | goto corrupted; | |
331 | } | |
332 | return 0; | |
333 | ||
334 | corrupted: | |
335 | ext4_error(inode->i_sb, function, | |
336 | "bad header in inode #%lu: %s - magic %x, " | |
337 | "entries %u, max %u(%u), depth %u(%u)", | |
338 | inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), | |
339 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
340 | max, le16_to_cpu(eh->eh_depth), depth); | |
341 | ||
342 | return -EIO; | |
343 | } | |
344 | ||
345 | #define ext4_ext_check_header(inode, eh, depth) \ | |
46e665e9 | 346 | __ext4_ext_check_header(__func__, inode, eh, depth) |
c29c0ae7 | 347 | |
a86c6181 AT |
348 | #ifdef EXT_DEBUG |
349 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
350 | { | |
351 | int k, l = path->p_depth; | |
352 | ||
353 | ext_debug("path:"); | |
354 | for (k = 0; k <= l; k++, path++) { | |
355 | if (path->p_idx) { | |
2ae02107 | 356 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
f65e6fba | 357 | idx_pblock(path->p_idx)); |
a86c6181 | 358 | } else if (path->p_ext) { |
2ae02107 | 359 | ext_debug(" %d:%d:%llu ", |
a86c6181 | 360 | le32_to_cpu(path->p_ext->ee_block), |
a2df2a63 | 361 | ext4_ext_get_actual_len(path->p_ext), |
f65e6fba | 362 | ext_pblock(path->p_ext)); |
a86c6181 AT |
363 | } else |
364 | ext_debug(" []"); | |
365 | } | |
366 | ext_debug("\n"); | |
367 | } | |
368 | ||
369 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
370 | { | |
371 | int depth = ext_depth(inode); | |
372 | struct ext4_extent_header *eh; | |
373 | struct ext4_extent *ex; | |
374 | int i; | |
375 | ||
376 | if (!path) | |
377 | return; | |
378 | ||
379 | eh = path[depth].p_hdr; | |
380 | ex = EXT_FIRST_EXTENT(eh); | |
381 | ||
382 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { | |
2ae02107 | 383 | ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block), |
a2df2a63 | 384 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
a86c6181 AT |
385 | } |
386 | ext_debug("\n"); | |
387 | } | |
388 | #else | |
af5bc92d TT |
389 | #define ext4_ext_show_path(inode, path) |
390 | #define ext4_ext_show_leaf(inode, path) | |
a86c6181 AT |
391 | #endif |
392 | ||
b35905c1 | 393 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
394 | { |
395 | int depth = path->p_depth; | |
396 | int i; | |
397 | ||
398 | for (i = 0; i <= depth; i++, path++) | |
399 | if (path->p_bh) { | |
400 | brelse(path->p_bh); | |
401 | path->p_bh = NULL; | |
402 | } | |
403 | } | |
404 | ||
405 | /* | |
d0d856e8 RD |
406 | * ext4_ext_binsearch_idx: |
407 | * binary search for the closest index of the given block | |
c29c0ae7 | 408 | * the header must be checked before calling this |
a86c6181 AT |
409 | */ |
410 | static void | |
725d26d3 AK |
411 | ext4_ext_binsearch_idx(struct inode *inode, |
412 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
413 | { |
414 | struct ext4_extent_header *eh = path->p_hdr; | |
415 | struct ext4_extent_idx *r, *l, *m; | |
416 | ||
a86c6181 | 417 | |
bba90743 | 418 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
419 | |
420 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 421 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
422 | while (l <= r) { |
423 | m = l + (r - l) / 2; | |
424 | if (block < le32_to_cpu(m->ei_block)) | |
425 | r = m - 1; | |
426 | else | |
427 | l = m + 1; | |
26d535ed DM |
428 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
429 | m, le32_to_cpu(m->ei_block), | |
430 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
431 | } |
432 | ||
433 | path->p_idx = l - 1; | |
f65e6fba | 434 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
26d535ed | 435 | idx_pblock(path->p_idx)); |
a86c6181 AT |
436 | |
437 | #ifdef CHECK_BINSEARCH | |
438 | { | |
439 | struct ext4_extent_idx *chix, *ix; | |
440 | int k; | |
441 | ||
442 | chix = ix = EXT_FIRST_INDEX(eh); | |
443 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
444 | if (k != 0 && | |
445 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
446 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
447 | "first=0x%p\n", k, | |
448 | ix, EXT_FIRST_INDEX(eh)); | |
449 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
450 | le32_to_cpu(ix->ei_block), |
451 | le32_to_cpu(ix[-1].ei_block)); | |
452 | } | |
453 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 454 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
455 | if (block < le32_to_cpu(ix->ei_block)) |
456 | break; | |
457 | chix = ix; | |
458 | } | |
459 | BUG_ON(chix != path->p_idx); | |
460 | } | |
461 | #endif | |
462 | ||
463 | } | |
464 | ||
465 | /* | |
d0d856e8 RD |
466 | * ext4_ext_binsearch: |
467 | * binary search for closest extent of the given block | |
c29c0ae7 | 468 | * the header must be checked before calling this |
a86c6181 AT |
469 | */ |
470 | static void | |
725d26d3 AK |
471 | ext4_ext_binsearch(struct inode *inode, |
472 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
473 | { |
474 | struct ext4_extent_header *eh = path->p_hdr; | |
475 | struct ext4_extent *r, *l, *m; | |
476 | ||
a86c6181 AT |
477 | if (eh->eh_entries == 0) { |
478 | /* | |
d0d856e8 RD |
479 | * this leaf is empty: |
480 | * we get such a leaf in split/add case | |
a86c6181 AT |
481 | */ |
482 | return; | |
483 | } | |
484 | ||
bba90743 | 485 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
486 | |
487 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 488 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
489 | |
490 | while (l <= r) { | |
491 | m = l + (r - l) / 2; | |
492 | if (block < le32_to_cpu(m->ee_block)) | |
493 | r = m - 1; | |
494 | else | |
495 | l = m + 1; | |
26d535ed DM |
496 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
497 | m, le32_to_cpu(m->ee_block), | |
498 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
499 | } |
500 | ||
501 | path->p_ext = l - 1; | |
2ae02107 | 502 | ext_debug(" -> %d:%llu:%d ", |
8c55e204 DK |
503 | le32_to_cpu(path->p_ext->ee_block), |
504 | ext_pblock(path->p_ext), | |
a2df2a63 | 505 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
506 | |
507 | #ifdef CHECK_BINSEARCH | |
508 | { | |
509 | struct ext4_extent *chex, *ex; | |
510 | int k; | |
511 | ||
512 | chex = ex = EXT_FIRST_EXTENT(eh); | |
513 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
514 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 515 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
516 | if (block < le32_to_cpu(ex->ee_block)) |
517 | break; | |
518 | chex = ex; | |
519 | } | |
520 | BUG_ON(chex != path->p_ext); | |
521 | } | |
522 | #endif | |
523 | ||
524 | } | |
525 | ||
526 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
527 | { | |
528 | struct ext4_extent_header *eh; | |
529 | ||
530 | eh = ext_inode_hdr(inode); | |
531 | eh->eh_depth = 0; | |
532 | eh->eh_entries = 0; | |
533 | eh->eh_magic = EXT4_EXT_MAGIC; | |
534 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode)); | |
535 | ext4_mark_inode_dirty(handle, inode); | |
536 | ext4_ext_invalidate_cache(inode); | |
537 | return 0; | |
538 | } | |
539 | ||
540 | struct ext4_ext_path * | |
725d26d3 AK |
541 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
542 | struct ext4_ext_path *path) | |
a86c6181 AT |
543 | { |
544 | struct ext4_extent_header *eh; | |
545 | struct buffer_head *bh; | |
546 | short int depth, i, ppos = 0, alloc = 0; | |
547 | ||
548 | eh = ext_inode_hdr(inode); | |
c29c0ae7 AT |
549 | depth = ext_depth(inode); |
550 | if (ext4_ext_check_header(inode, eh, depth)) | |
a86c6181 AT |
551 | return ERR_PTR(-EIO); |
552 | ||
a86c6181 AT |
553 | |
554 | /* account possible depth increase */ | |
555 | if (!path) { | |
5d4958f9 | 556 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
557 | GFP_NOFS); |
558 | if (!path) | |
559 | return ERR_PTR(-ENOMEM); | |
560 | alloc = 1; | |
561 | } | |
a86c6181 | 562 | path[0].p_hdr = eh; |
1973adcb | 563 | path[0].p_bh = NULL; |
a86c6181 | 564 | |
c29c0ae7 | 565 | i = depth; |
a86c6181 AT |
566 | /* walk through the tree */ |
567 | while (i) { | |
568 | ext_debug("depth %d: num %d, max %d\n", | |
569 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 570 | |
a86c6181 | 571 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
f65e6fba | 572 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
573 | path[ppos].p_depth = i; |
574 | path[ppos].p_ext = NULL; | |
575 | ||
576 | bh = sb_bread(inode->i_sb, path[ppos].p_block); | |
577 | if (!bh) | |
578 | goto err; | |
579 | ||
580 | eh = ext_block_hdr(bh); | |
581 | ppos++; | |
582 | BUG_ON(ppos > depth); | |
583 | path[ppos].p_bh = bh; | |
584 | path[ppos].p_hdr = eh; | |
585 | i--; | |
586 | ||
c29c0ae7 | 587 | if (ext4_ext_check_header(inode, eh, i)) |
a86c6181 AT |
588 | goto err; |
589 | } | |
590 | ||
591 | path[ppos].p_depth = i; | |
a86c6181 AT |
592 | path[ppos].p_ext = NULL; |
593 | path[ppos].p_idx = NULL; | |
594 | ||
a86c6181 AT |
595 | /* find extent */ |
596 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
597 | /* if not an empty leaf */ |
598 | if (path[ppos].p_ext) | |
599 | path[ppos].p_block = ext_pblock(path[ppos].p_ext); | |
a86c6181 AT |
600 | |
601 | ext4_ext_show_path(inode, path); | |
602 | ||
603 | return path; | |
604 | ||
605 | err: | |
606 | ext4_ext_drop_refs(path); | |
607 | if (alloc) | |
608 | kfree(path); | |
609 | return ERR_PTR(-EIO); | |
610 | } | |
611 | ||
612 | /* | |
d0d856e8 RD |
613 | * ext4_ext_insert_index: |
614 | * insert new index [@logical;@ptr] into the block at @curp; | |
615 | * check where to insert: before @curp or after @curp | |
a86c6181 AT |
616 | */ |
617 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |
618 | struct ext4_ext_path *curp, | |
f65e6fba | 619 | int logical, ext4_fsblk_t ptr) |
a86c6181 AT |
620 | { |
621 | struct ext4_extent_idx *ix; | |
622 | int len, err; | |
623 | ||
7e028976 AM |
624 | err = ext4_ext_get_access(handle, inode, curp); |
625 | if (err) | |
a86c6181 AT |
626 | return err; |
627 | ||
628 | BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block)); | |
629 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; | |
630 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
631 | /* insert after */ | |
632 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
633 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
634 | len = len < 0 ? 0 : len; | |
26d535ed | 635 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
636 | "move %d from 0x%p to 0x%p\n", |
637 | logical, ptr, len, | |
638 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
639 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
640 | } | |
641 | ix = curp->p_idx + 1; | |
642 | } else { | |
643 | /* insert before */ | |
644 | len = len * sizeof(struct ext4_extent_idx); | |
645 | len = len < 0 ? 0 : len; | |
26d535ed | 646 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
647 | "move %d from 0x%p to 0x%p\n", |
648 | logical, ptr, len, | |
649 | curp->p_idx, (curp->p_idx + 1)); | |
650 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
651 | ix = curp->p_idx; | |
652 | } | |
653 | ||
654 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 655 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 656 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 AT |
657 | |
658 | BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries) | |
8c55e204 | 659 | > le16_to_cpu(curp->p_hdr->eh_max)); |
a86c6181 AT |
660 | BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr)); |
661 | ||
662 | err = ext4_ext_dirty(handle, inode, curp); | |
663 | ext4_std_error(inode->i_sb, err); | |
664 | ||
665 | return err; | |
666 | } | |
667 | ||
668 | /* | |
d0d856e8 RD |
669 | * ext4_ext_split: |
670 | * inserts new subtree into the path, using free index entry | |
671 | * at depth @at: | |
672 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
673 | * - makes decision where to split | |
674 | * - moves remaining extents and index entries (right to the split point) | |
675 | * into the newly allocated blocks | |
676 | * - initializes subtree | |
a86c6181 AT |
677 | */ |
678 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
679 | struct ext4_ext_path *path, | |
680 | struct ext4_extent *newext, int at) | |
681 | { | |
682 | struct buffer_head *bh = NULL; | |
683 | int depth = ext_depth(inode); | |
684 | struct ext4_extent_header *neh; | |
685 | struct ext4_extent_idx *fidx; | |
686 | struct ext4_extent *ex; | |
687 | int i = at, k, m, a; | |
f65e6fba | 688 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 689 | __le32 border; |
f65e6fba | 690 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
691 | int err = 0; |
692 | ||
693 | /* make decision: where to split? */ | |
d0d856e8 | 694 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 695 | |
d0d856e8 | 696 | /* if current leaf will be split, then we should use |
a86c6181 AT |
697 | * border from split point */ |
698 | BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr)); | |
699 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
700 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 701 | ext_debug("leaf will be split." |
a86c6181 | 702 | " next leaf starts at %d\n", |
8c55e204 | 703 | le32_to_cpu(border)); |
a86c6181 AT |
704 | } else { |
705 | border = newext->ee_block; | |
706 | ext_debug("leaf will be added." | |
707 | " next leaf starts at %d\n", | |
8c55e204 | 708 | le32_to_cpu(border)); |
a86c6181 AT |
709 | } |
710 | ||
711 | /* | |
d0d856e8 RD |
712 | * If error occurs, then we break processing |
713 | * and mark filesystem read-only. index won't | |
a86c6181 | 714 | * be inserted and tree will be in consistent |
d0d856e8 | 715 | * state. Next mount will repair buffers too. |
a86c6181 AT |
716 | */ |
717 | ||
718 | /* | |
d0d856e8 RD |
719 | * Get array to track all allocated blocks. |
720 | * We need this to handle errors and free blocks | |
721 | * upon them. | |
a86c6181 | 722 | */ |
5d4958f9 | 723 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
724 | if (!ablocks) |
725 | return -ENOMEM; | |
a86c6181 AT |
726 | |
727 | /* allocate all needed blocks */ | |
728 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
729 | for (a = 0; a < depth - at; a++) { | |
654b4908 AK |
730 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
731 | newext, &err); | |
a86c6181 AT |
732 | if (newblock == 0) |
733 | goto cleanup; | |
734 | ablocks[a] = newblock; | |
735 | } | |
736 | ||
737 | /* initialize new leaf */ | |
738 | newblock = ablocks[--a]; | |
739 | BUG_ON(newblock == 0); | |
740 | bh = sb_getblk(inode->i_sb, newblock); | |
741 | if (!bh) { | |
742 | err = -EIO; | |
743 | goto cleanup; | |
744 | } | |
745 | lock_buffer(bh); | |
746 | ||
7e028976 AM |
747 | err = ext4_journal_get_create_access(handle, bh); |
748 | if (err) | |
a86c6181 AT |
749 | goto cleanup; |
750 | ||
751 | neh = ext_block_hdr(bh); | |
752 | neh->eh_entries = 0; | |
753 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
754 | neh->eh_magic = EXT4_EXT_MAGIC; | |
755 | neh->eh_depth = 0; | |
756 | ex = EXT_FIRST_EXTENT(neh); | |
757 | ||
d0d856e8 | 758 | /* move remainder of path[depth] to the new leaf */ |
a86c6181 AT |
759 | BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max); |
760 | /* start copy from next extent */ | |
761 | /* TODO: we could do it by single memmove */ | |
762 | m = 0; | |
763 | path[depth].p_ext++; | |
764 | while (path[depth].p_ext <= | |
765 | EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
2ae02107 | 766 | ext_debug("move %d:%llu:%d in new leaf %llu\n", |
8c55e204 DK |
767 | le32_to_cpu(path[depth].p_ext->ee_block), |
768 | ext_pblock(path[depth].p_ext), | |
a2df2a63 | 769 | ext4_ext_get_actual_len(path[depth].p_ext), |
a86c6181 AT |
770 | newblock); |
771 | /*memmove(ex++, path[depth].p_ext++, | |
772 | sizeof(struct ext4_extent)); | |
773 | neh->eh_entries++;*/ | |
774 | path[depth].p_ext++; | |
775 | m++; | |
776 | } | |
777 | if (m) { | |
778 | memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); | |
e8546d06 | 779 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
780 | } |
781 | ||
782 | set_buffer_uptodate(bh); | |
783 | unlock_buffer(bh); | |
784 | ||
0390131b | 785 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 786 | if (err) |
a86c6181 AT |
787 | goto cleanup; |
788 | brelse(bh); | |
789 | bh = NULL; | |
790 | ||
791 | /* correct old leaf */ | |
792 | if (m) { | |
7e028976 AM |
793 | err = ext4_ext_get_access(handle, inode, path + depth); |
794 | if (err) | |
a86c6181 | 795 | goto cleanup; |
e8546d06 | 796 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
797 | err = ext4_ext_dirty(handle, inode, path + depth); |
798 | if (err) | |
a86c6181 AT |
799 | goto cleanup; |
800 | ||
801 | } | |
802 | ||
803 | /* create intermediate indexes */ | |
804 | k = depth - at - 1; | |
805 | BUG_ON(k < 0); | |
806 | if (k) | |
807 | ext_debug("create %d intermediate indices\n", k); | |
808 | /* insert new index into current index block */ | |
809 | /* current depth stored in i var */ | |
810 | i = depth - 1; | |
811 | while (k--) { | |
812 | oldblock = newblock; | |
813 | newblock = ablocks[--a]; | |
bba90743 | 814 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
815 | if (!bh) { |
816 | err = -EIO; | |
817 | goto cleanup; | |
818 | } | |
819 | lock_buffer(bh); | |
820 | ||
7e028976 AM |
821 | err = ext4_journal_get_create_access(handle, bh); |
822 | if (err) | |
a86c6181 AT |
823 | goto cleanup; |
824 | ||
825 | neh = ext_block_hdr(bh); | |
826 | neh->eh_entries = cpu_to_le16(1); | |
827 | neh->eh_magic = EXT4_EXT_MAGIC; | |
828 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
829 | neh->eh_depth = cpu_to_le16(depth - i); | |
830 | fidx = EXT_FIRST_INDEX(neh); | |
831 | fidx->ei_block = border; | |
f65e6fba | 832 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 833 | |
bba90743 ES |
834 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
835 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 AT |
836 | /* copy indexes */ |
837 | m = 0; | |
838 | path[i].p_idx++; | |
839 | ||
840 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
841 | EXT_MAX_INDEX(path[i].p_hdr)); | |
842 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != | |
843 | EXT_LAST_INDEX(path[i].p_hdr)); | |
844 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { | |
26d535ed | 845 | ext_debug("%d: move %d:%llu in new index %llu\n", i, |
8c55e204 DK |
846 | le32_to_cpu(path[i].p_idx->ei_block), |
847 | idx_pblock(path[i].p_idx), | |
848 | newblock); | |
a86c6181 AT |
849 | /*memmove(++fidx, path[i].p_idx++, |
850 | sizeof(struct ext4_extent_idx)); | |
851 | neh->eh_entries++; | |
852 | BUG_ON(neh->eh_entries > neh->eh_max);*/ | |
853 | path[i].p_idx++; | |
854 | m++; | |
855 | } | |
856 | if (m) { | |
857 | memmove(++fidx, path[i].p_idx - m, | |
858 | sizeof(struct ext4_extent_idx) * m); | |
e8546d06 | 859 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
860 | } |
861 | set_buffer_uptodate(bh); | |
862 | unlock_buffer(bh); | |
863 | ||
0390131b | 864 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 865 | if (err) |
a86c6181 AT |
866 | goto cleanup; |
867 | brelse(bh); | |
868 | bh = NULL; | |
869 | ||
870 | /* correct old index */ | |
871 | if (m) { | |
872 | err = ext4_ext_get_access(handle, inode, path + i); | |
873 | if (err) | |
874 | goto cleanup; | |
e8546d06 | 875 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
876 | err = ext4_ext_dirty(handle, inode, path + i); |
877 | if (err) | |
878 | goto cleanup; | |
879 | } | |
880 | ||
881 | i--; | |
882 | } | |
883 | ||
884 | /* insert new index */ | |
a86c6181 AT |
885 | err = ext4_ext_insert_index(handle, inode, path + at, |
886 | le32_to_cpu(border), newblock); | |
887 | ||
888 | cleanup: | |
889 | if (bh) { | |
890 | if (buffer_locked(bh)) | |
891 | unlock_buffer(bh); | |
892 | brelse(bh); | |
893 | } | |
894 | ||
895 | if (err) { | |
896 | /* free all allocated blocks in error case */ | |
897 | for (i = 0; i < depth; i++) { | |
898 | if (!ablocks[i]) | |
899 | continue; | |
c9de560d | 900 | ext4_free_blocks(handle, inode, ablocks[i], 1, 1); |
a86c6181 AT |
901 | } |
902 | } | |
903 | kfree(ablocks); | |
904 | ||
905 | return err; | |
906 | } | |
907 | ||
908 | /* | |
d0d856e8 RD |
909 | * ext4_ext_grow_indepth: |
910 | * implements tree growing procedure: | |
911 | * - allocates new block | |
912 | * - moves top-level data (index block or leaf) into the new block | |
913 | * - initializes new top-level, creating index that points to the | |
914 | * just created block | |
a86c6181 AT |
915 | */ |
916 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
917 | struct ext4_ext_path *path, | |
918 | struct ext4_extent *newext) | |
919 | { | |
920 | struct ext4_ext_path *curp = path; | |
921 | struct ext4_extent_header *neh; | |
922 | struct ext4_extent_idx *fidx; | |
923 | struct buffer_head *bh; | |
f65e6fba | 924 | ext4_fsblk_t newblock; |
a86c6181 AT |
925 | int err = 0; |
926 | ||
654b4908 | 927 | newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err); |
a86c6181 AT |
928 | if (newblock == 0) |
929 | return err; | |
930 | ||
931 | bh = sb_getblk(inode->i_sb, newblock); | |
932 | if (!bh) { | |
933 | err = -EIO; | |
934 | ext4_std_error(inode->i_sb, err); | |
935 | return err; | |
936 | } | |
937 | lock_buffer(bh); | |
938 | ||
7e028976 AM |
939 | err = ext4_journal_get_create_access(handle, bh); |
940 | if (err) { | |
a86c6181 AT |
941 | unlock_buffer(bh); |
942 | goto out; | |
943 | } | |
944 | ||
945 | /* move top-level index/leaf into new block */ | |
946 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
947 | ||
948 | /* set size of new block */ | |
949 | neh = ext_block_hdr(bh); | |
950 | /* old root could have indexes or leaves | |
951 | * so calculate e_max right way */ | |
952 | if (ext_depth(inode)) | |
953 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
954 | else | |
955 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
956 | neh->eh_magic = EXT4_EXT_MAGIC; | |
957 | set_buffer_uptodate(bh); | |
958 | unlock_buffer(bh); | |
959 | ||
0390131b | 960 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 961 | if (err) |
a86c6181 AT |
962 | goto out; |
963 | ||
964 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
965 | err = ext4_ext_get_access(handle, inode, curp); |
966 | if (err) | |
a86c6181 AT |
967 | goto out; |
968 | ||
969 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
970 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); | |
971 | curp->p_hdr->eh_entries = cpu_to_le16(1); | |
972 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
973 | |
974 | if (path[0].p_hdr->eh_depth) | |
975 | curp->p_idx->ei_block = | |
976 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
977 | else | |
978 | curp->p_idx->ei_block = | |
979 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 980 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
981 | |
982 | neh = ext_inode_hdr(inode); | |
983 | fidx = EXT_FIRST_INDEX(neh); | |
2ae02107 | 984 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 985 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
f65e6fba | 986 | le32_to_cpu(fidx->ei_block), idx_pblock(fidx)); |
a86c6181 AT |
987 | |
988 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
989 | err = ext4_ext_dirty(handle, inode, curp); | |
990 | out: | |
991 | brelse(bh); | |
992 | ||
993 | return err; | |
994 | } | |
995 | ||
996 | /* | |
d0d856e8 RD |
997 | * ext4_ext_create_new_leaf: |
998 | * finds empty index and adds new leaf. | |
999 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1000 | */ |
1001 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
1002 | struct ext4_ext_path *path, | |
1003 | struct ext4_extent *newext) | |
1004 | { | |
1005 | struct ext4_ext_path *curp; | |
1006 | int depth, i, err = 0; | |
1007 | ||
1008 | repeat: | |
1009 | i = depth = ext_depth(inode); | |
1010 | ||
1011 | /* walk up to the tree and look for free index entry */ | |
1012 | curp = path + depth; | |
1013 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1014 | i--; | |
1015 | curp--; | |
1016 | } | |
1017 | ||
d0d856e8 RD |
1018 | /* we use already allocated block for index block, |
1019 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1020 | if (EXT_HAS_FREE_INDEX(curp)) { |
1021 | /* if we found index with free entry, then use that | |
1022 | * entry: create all needed subtree and add new leaf */ | |
1023 | err = ext4_ext_split(handle, inode, path, newext, i); | |
787e0981 SF |
1024 | if (err) |
1025 | goto out; | |
a86c6181 AT |
1026 | |
1027 | /* refill path */ | |
1028 | ext4_ext_drop_refs(path); | |
1029 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1030 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1031 | path); | |
a86c6181 AT |
1032 | if (IS_ERR(path)) |
1033 | err = PTR_ERR(path); | |
1034 | } else { | |
1035 | /* tree is full, time to grow in depth */ | |
1036 | err = ext4_ext_grow_indepth(handle, inode, path, newext); | |
1037 | if (err) | |
1038 | goto out; | |
1039 | ||
1040 | /* refill path */ | |
1041 | ext4_ext_drop_refs(path); | |
1042 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1043 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1044 | path); | |
a86c6181 AT |
1045 | if (IS_ERR(path)) { |
1046 | err = PTR_ERR(path); | |
1047 | goto out; | |
1048 | } | |
1049 | ||
1050 | /* | |
d0d856e8 RD |
1051 | * only first (depth 0 -> 1) produces free space; |
1052 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1053 | */ |
1054 | depth = ext_depth(inode); | |
1055 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1056 | /* now we need to split */ |
a86c6181 AT |
1057 | goto repeat; |
1058 | } | |
1059 | } | |
1060 | ||
1061 | out: | |
1062 | return err; | |
1063 | } | |
1064 | ||
1988b51e AT |
1065 | /* |
1066 | * search the closest allocated block to the left for *logical | |
1067 | * and returns it at @logical + it's physical address at @phys | |
1068 | * if *logical is the smallest allocated block, the function | |
1069 | * returns 0 at @phys | |
1070 | * return value contains 0 (success) or error code | |
1071 | */ | |
1072 | int | |
1073 | ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path, | |
1074 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1075 | { | |
1076 | struct ext4_extent_idx *ix; | |
1077 | struct ext4_extent *ex; | |
b939e376 | 1078 | int depth, ee_len; |
1988b51e AT |
1079 | |
1080 | BUG_ON(path == NULL); | |
1081 | depth = path->p_depth; | |
1082 | *phys = 0; | |
1083 | ||
1084 | if (depth == 0 && path->p_ext == NULL) | |
1085 | return 0; | |
1086 | ||
1087 | /* usually extent in the path covers blocks smaller | |
1088 | * then *logical, but it can be that extent is the | |
1089 | * first one in the file */ | |
1090 | ||
1091 | ex = path[depth].p_ext; | |
b939e376 | 1092 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1093 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1094 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1095 | while (--depth >= 0) { | |
1096 | ix = path[depth].p_idx; | |
1097 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1098 | } | |
1099 | return 0; | |
1100 | } | |
1101 | ||
b939e376 | 1102 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e | 1103 | |
b939e376 AK |
1104 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
1105 | *phys = ext_pblock(ex) + ee_len - 1; | |
1988b51e AT |
1106 | return 0; |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * search the closest allocated block to the right for *logical | |
1111 | * and returns it at @logical + it's physical address at @phys | |
1112 | * if *logical is the smallest allocated block, the function | |
1113 | * returns 0 at @phys | |
1114 | * return value contains 0 (success) or error code | |
1115 | */ | |
1116 | int | |
1117 | ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path, | |
1118 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1119 | { | |
1120 | struct buffer_head *bh = NULL; | |
1121 | struct ext4_extent_header *eh; | |
1122 | struct ext4_extent_idx *ix; | |
1123 | struct ext4_extent *ex; | |
1124 | ext4_fsblk_t block; | |
395a87bf ES |
1125 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1126 | int ee_len; | |
1988b51e AT |
1127 | |
1128 | BUG_ON(path == NULL); | |
1129 | depth = path->p_depth; | |
1130 | *phys = 0; | |
1131 | ||
1132 | if (depth == 0 && path->p_ext == NULL) | |
1133 | return 0; | |
1134 | ||
1135 | /* usually extent in the path covers blocks smaller | |
1136 | * then *logical, but it can be that extent is the | |
1137 | * first one in the file */ | |
1138 | ||
1139 | ex = path[depth].p_ext; | |
b939e376 | 1140 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1141 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1142 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1143 | while (--depth >= 0) { | |
1144 | ix = path[depth].p_idx; | |
1145 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1146 | } | |
1147 | *logical = le32_to_cpu(ex->ee_block); | |
1148 | *phys = ext_pblock(ex); | |
1149 | return 0; | |
1150 | } | |
1151 | ||
b939e376 | 1152 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e AT |
1153 | |
1154 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1155 | /* next allocated block in this leaf */ | |
1156 | ex++; | |
1157 | *logical = le32_to_cpu(ex->ee_block); | |
1158 | *phys = ext_pblock(ex); | |
1159 | return 0; | |
1160 | } | |
1161 | ||
1162 | /* go up and search for index to the right */ | |
1163 | while (--depth >= 0) { | |
1164 | ix = path[depth].p_idx; | |
1165 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1166 | goto got_index; |
1988b51e AT |
1167 | } |
1168 | ||
25f1ee3a WF |
1169 | /* we've gone up to the root and found no index to the right */ |
1170 | return 0; | |
1988b51e | 1171 | |
25f1ee3a | 1172 | got_index: |
1988b51e AT |
1173 | /* we've found index to the right, let's |
1174 | * follow it and find the closest allocated | |
1175 | * block to the right */ | |
1176 | ix++; | |
1177 | block = idx_pblock(ix); | |
1178 | while (++depth < path->p_depth) { | |
1179 | bh = sb_bread(inode->i_sb, block); | |
1180 | if (bh == NULL) | |
1181 | return -EIO; | |
1182 | eh = ext_block_hdr(bh); | |
395a87bf ES |
1183 | /* subtract from p_depth to get proper eh_depth */ |
1184 | if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) { | |
1988b51e AT |
1185 | put_bh(bh); |
1186 | return -EIO; | |
1187 | } | |
1188 | ix = EXT_FIRST_INDEX(eh); | |
1189 | block = idx_pblock(ix); | |
1190 | put_bh(bh); | |
1191 | } | |
1192 | ||
1193 | bh = sb_bread(inode->i_sb, block); | |
1194 | if (bh == NULL) | |
1195 | return -EIO; | |
1196 | eh = ext_block_hdr(bh); | |
1197 | if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) { | |
1198 | put_bh(bh); | |
1199 | return -EIO; | |
1200 | } | |
1201 | ex = EXT_FIRST_EXTENT(eh); | |
1202 | *logical = le32_to_cpu(ex->ee_block); | |
1203 | *phys = ext_pblock(ex); | |
1204 | put_bh(bh); | |
1205 | return 0; | |
1988b51e AT |
1206 | } |
1207 | ||
a86c6181 | 1208 | /* |
d0d856e8 RD |
1209 | * ext4_ext_next_allocated_block: |
1210 | * returns allocated block in subsequent extent or EXT_MAX_BLOCK. | |
1211 | * NOTE: it considers block number from index entry as | |
1212 | * allocated block. Thus, index entries have to be consistent | |
1213 | * with leaves. | |
a86c6181 | 1214 | */ |
725d26d3 | 1215 | static ext4_lblk_t |
a86c6181 AT |
1216 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1217 | { | |
1218 | int depth; | |
1219 | ||
1220 | BUG_ON(path == NULL); | |
1221 | depth = path->p_depth; | |
1222 | ||
1223 | if (depth == 0 && path->p_ext == NULL) | |
1224 | return EXT_MAX_BLOCK; | |
1225 | ||
1226 | while (depth >= 0) { | |
1227 | if (depth == path->p_depth) { | |
1228 | /* leaf */ | |
1229 | if (path[depth].p_ext != | |
1230 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1231 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1232 | } else { | |
1233 | /* index */ | |
1234 | if (path[depth].p_idx != | |
1235 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1236 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1237 | } | |
1238 | depth--; | |
1239 | } | |
1240 | ||
1241 | return EXT_MAX_BLOCK; | |
1242 | } | |
1243 | ||
1244 | /* | |
d0d856e8 | 1245 | * ext4_ext_next_leaf_block: |
a86c6181 AT |
1246 | * returns first allocated block from next leaf or EXT_MAX_BLOCK |
1247 | */ | |
725d26d3 | 1248 | static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode, |
63f57933 | 1249 | struct ext4_ext_path *path) |
a86c6181 AT |
1250 | { |
1251 | int depth; | |
1252 | ||
1253 | BUG_ON(path == NULL); | |
1254 | depth = path->p_depth; | |
1255 | ||
1256 | /* zero-tree has no leaf blocks at all */ | |
1257 | if (depth == 0) | |
1258 | return EXT_MAX_BLOCK; | |
1259 | ||
1260 | /* go to index block */ | |
1261 | depth--; | |
1262 | ||
1263 | while (depth >= 0) { | |
1264 | if (path[depth].p_idx != | |
1265 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1266 | return (ext4_lblk_t) |
1267 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1268 | depth--; |
1269 | } | |
1270 | ||
1271 | return EXT_MAX_BLOCK; | |
1272 | } | |
1273 | ||
1274 | /* | |
d0d856e8 RD |
1275 | * ext4_ext_correct_indexes: |
1276 | * if leaf gets modified and modified extent is first in the leaf, | |
1277 | * then we have to correct all indexes above. | |
a86c6181 AT |
1278 | * TODO: do we need to correct tree in all cases? |
1279 | */ | |
1d03ec98 | 1280 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1281 | struct ext4_ext_path *path) |
1282 | { | |
1283 | struct ext4_extent_header *eh; | |
1284 | int depth = ext_depth(inode); | |
1285 | struct ext4_extent *ex; | |
1286 | __le32 border; | |
1287 | int k, err = 0; | |
1288 | ||
1289 | eh = path[depth].p_hdr; | |
1290 | ex = path[depth].p_ext; | |
1291 | BUG_ON(ex == NULL); | |
1292 | BUG_ON(eh == NULL); | |
1293 | ||
1294 | if (depth == 0) { | |
1295 | /* there is no tree at all */ | |
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1300 | /* we correct tree if first leaf got modified only */ | |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | /* | |
d0d856e8 | 1305 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1306 | */ |
1307 | k = depth - 1; | |
1308 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1309 | err = ext4_ext_get_access(handle, inode, path + k); |
1310 | if (err) | |
a86c6181 AT |
1311 | return err; |
1312 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1313 | err = ext4_ext_dirty(handle, inode, path + k); |
1314 | if (err) | |
a86c6181 AT |
1315 | return err; |
1316 | ||
1317 | while (k--) { | |
1318 | /* change all left-side indexes */ | |
1319 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1320 | break; | |
7e028976 AM |
1321 | err = ext4_ext_get_access(handle, inode, path + k); |
1322 | if (err) | |
a86c6181 AT |
1323 | break; |
1324 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1325 | err = ext4_ext_dirty(handle, inode, path + k); |
1326 | if (err) | |
a86c6181 AT |
1327 | break; |
1328 | } | |
1329 | ||
1330 | return err; | |
1331 | } | |
1332 | ||
09b88252 | 1333 | static int |
a86c6181 AT |
1334 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1335 | struct ext4_extent *ex2) | |
1336 | { | |
749269fa | 1337 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1338 | |
1339 | /* | |
1340 | * Make sure that either both extents are uninitialized, or | |
1341 | * both are _not_. | |
1342 | */ | |
1343 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1344 | return 0; | |
1345 | ||
749269fa AA |
1346 | if (ext4_ext_is_uninitialized(ex1)) |
1347 | max_len = EXT_UNINIT_MAX_LEN; | |
1348 | else | |
1349 | max_len = EXT_INIT_MAX_LEN; | |
1350 | ||
a2df2a63 AA |
1351 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1352 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1353 | ||
1354 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1355 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1356 | return 0; |
1357 | ||
471d4011 SB |
1358 | /* |
1359 | * To allow future support for preallocated extents to be added | |
1360 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1361 | * this can result in the top bit of ee_len being set. |
471d4011 | 1362 | */ |
749269fa | 1363 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1364 | return 0; |
bbf2f9fb | 1365 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1366 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1367 | return 0; |
1368 | #endif | |
1369 | ||
a2df2a63 | 1370 | if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) |
a86c6181 AT |
1371 | return 1; |
1372 | return 0; | |
1373 | } | |
1374 | ||
56055d3a AA |
1375 | /* |
1376 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1377 | * It always tries to merge towards right. If you want to merge towards | |
1378 | * left, pass "ex - 1" as argument instead of "ex". | |
1379 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1380 | * 1 if they got merged. | |
1381 | */ | |
1382 | int ext4_ext_try_to_merge(struct inode *inode, | |
1383 | struct ext4_ext_path *path, | |
1384 | struct ext4_extent *ex) | |
1385 | { | |
1386 | struct ext4_extent_header *eh; | |
1387 | unsigned int depth, len; | |
1388 | int merge_done = 0; | |
1389 | int uninitialized = 0; | |
1390 | ||
1391 | depth = ext_depth(inode); | |
1392 | BUG_ON(path[depth].p_hdr == NULL); | |
1393 | eh = path[depth].p_hdr; | |
1394 | ||
1395 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1396 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1397 | break; | |
1398 | /* merge with next extent! */ | |
1399 | if (ext4_ext_is_uninitialized(ex)) | |
1400 | uninitialized = 1; | |
1401 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1402 | + ext4_ext_get_actual_len(ex + 1)); | |
1403 | if (uninitialized) | |
1404 | ext4_ext_mark_uninitialized(ex); | |
1405 | ||
1406 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1407 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1408 | * sizeof(struct ext4_extent); | |
1409 | memmove(ex + 1, ex + 2, len); | |
1410 | } | |
e8546d06 | 1411 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1412 | merge_done = 1; |
1413 | WARN_ON(eh->eh_entries == 0); | |
1414 | if (!eh->eh_entries) | |
1415 | ext4_error(inode->i_sb, "ext4_ext_try_to_merge", | |
1416 | "inode#%lu, eh->eh_entries = 0!", inode->i_ino); | |
1417 | } | |
1418 | ||
1419 | return merge_done; | |
1420 | } | |
1421 | ||
25d14f98 AA |
1422 | /* |
1423 | * check if a portion of the "newext" extent overlaps with an | |
1424 | * existing extent. | |
1425 | * | |
1426 | * If there is an overlap discovered, it updates the length of the newext | |
1427 | * such that there will be no overlap, and then returns 1. | |
1428 | * If there is no overlap found, it returns 0. | |
1429 | */ | |
1430 | unsigned int ext4_ext_check_overlap(struct inode *inode, | |
1431 | struct ext4_extent *newext, | |
1432 | struct ext4_ext_path *path) | |
1433 | { | |
725d26d3 | 1434 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1435 | unsigned int depth, len1; |
1436 | unsigned int ret = 0; | |
1437 | ||
1438 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1439 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1440 | depth = ext_depth(inode); |
1441 | if (!path[depth].p_ext) | |
1442 | goto out; | |
1443 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
1444 | ||
1445 | /* | |
1446 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1447 | * is before the requested block(s) |
25d14f98 AA |
1448 | */ |
1449 | if (b2 < b1) { | |
1450 | b2 = ext4_ext_next_allocated_block(path); | |
1451 | if (b2 == EXT_MAX_BLOCK) | |
1452 | goto out; | |
1453 | } | |
1454 | ||
725d26d3 | 1455 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 AA |
1456 | if (b1 + len1 < b1) { |
1457 | len1 = EXT_MAX_BLOCK - b1; | |
1458 | newext->ee_len = cpu_to_le16(len1); | |
1459 | ret = 1; | |
1460 | } | |
1461 | ||
1462 | /* check for overlap */ | |
1463 | if (b1 + len1 > b2) { | |
1464 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1465 | ret = 1; | |
1466 | } | |
1467 | out: | |
1468 | return ret; | |
1469 | } | |
1470 | ||
a86c6181 | 1471 | /* |
d0d856e8 RD |
1472 | * ext4_ext_insert_extent: |
1473 | * tries to merge requsted extent into the existing extent or | |
1474 | * inserts requested extent as new one into the tree, | |
1475 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1476 | */ |
1477 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1478 | struct ext4_ext_path *path, | |
1479 | struct ext4_extent *newext) | |
1480 | { | |
af5bc92d | 1481 | struct ext4_extent_header *eh; |
a86c6181 AT |
1482 | struct ext4_extent *ex, *fex; |
1483 | struct ext4_extent *nearex; /* nearest extent */ | |
1484 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1485 | int depth, len, err; |
1486 | ext4_lblk_t next; | |
a2df2a63 | 1487 | unsigned uninitialized = 0; |
a86c6181 | 1488 | |
a2df2a63 | 1489 | BUG_ON(ext4_ext_get_actual_len(newext) == 0); |
a86c6181 AT |
1490 | depth = ext_depth(inode); |
1491 | ex = path[depth].p_ext; | |
1492 | BUG_ON(path[depth].p_hdr == NULL); | |
1493 | ||
1494 | /* try to insert block into found extent and return */ | |
1495 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | |
2ae02107 | 1496 | ext_debug("append %d block to %d:%d (from %llu)\n", |
a2df2a63 | 1497 | ext4_ext_get_actual_len(newext), |
a86c6181 | 1498 | le32_to_cpu(ex->ee_block), |
a2df2a63 | 1499 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
7e028976 AM |
1500 | err = ext4_ext_get_access(handle, inode, path + depth); |
1501 | if (err) | |
a86c6181 | 1502 | return err; |
a2df2a63 AA |
1503 | |
1504 | /* | |
1505 | * ext4_can_extents_be_merged should have checked that either | |
1506 | * both extents are uninitialized, or both aren't. Thus we | |
1507 | * need to check only one of them here. | |
1508 | */ | |
1509 | if (ext4_ext_is_uninitialized(ex)) | |
1510 | uninitialized = 1; | |
1511 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1512 | + ext4_ext_get_actual_len(newext)); | |
1513 | if (uninitialized) | |
1514 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1515 | eh = path[depth].p_hdr; |
1516 | nearex = ex; | |
1517 | goto merge; | |
1518 | } | |
1519 | ||
1520 | repeat: | |
1521 | depth = ext_depth(inode); | |
1522 | eh = path[depth].p_hdr; | |
1523 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1524 | goto has_space; | |
1525 | ||
1526 | /* probably next leaf has space for us? */ | |
1527 | fex = EXT_LAST_EXTENT(eh); | |
1528 | next = ext4_ext_next_leaf_block(inode, path); | |
1529 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) | |
1530 | && next != EXT_MAX_BLOCK) { | |
1531 | ext_debug("next leaf block - %d\n", next); | |
1532 | BUG_ON(npath != NULL); | |
1533 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1534 | if (IS_ERR(npath)) | |
1535 | return PTR_ERR(npath); | |
1536 | BUG_ON(npath->p_depth != path->p_depth); | |
1537 | eh = npath[depth].p_hdr; | |
1538 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
1539 | ext_debug("next leaf isnt full(%d)\n", | |
1540 | le16_to_cpu(eh->eh_entries)); | |
1541 | path = npath; | |
1542 | goto repeat; | |
1543 | } | |
1544 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1545 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1546 | } | |
1547 | ||
1548 | /* | |
d0d856e8 RD |
1549 | * There is no free space in the found leaf. |
1550 | * We're gonna add a new leaf in the tree. | |
a86c6181 AT |
1551 | */ |
1552 | err = ext4_ext_create_new_leaf(handle, inode, path, newext); | |
1553 | if (err) | |
1554 | goto cleanup; | |
1555 | depth = ext_depth(inode); | |
1556 | eh = path[depth].p_hdr; | |
1557 | ||
1558 | has_space: | |
1559 | nearex = path[depth].p_ext; | |
1560 | ||
7e028976 AM |
1561 | err = ext4_ext_get_access(handle, inode, path + depth); |
1562 | if (err) | |
a86c6181 AT |
1563 | goto cleanup; |
1564 | ||
1565 | if (!nearex) { | |
1566 | /* there is no extent in this leaf, create first one */ | |
2ae02107 | 1567 | ext_debug("first extent in the leaf: %d:%llu:%d\n", |
8c55e204 DK |
1568 | le32_to_cpu(newext->ee_block), |
1569 | ext_pblock(newext), | |
a2df2a63 | 1570 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1571 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1572 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1573 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1574 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1575 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1576 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1577 | len = (len - 1) * sizeof(struct ext4_extent); | |
1578 | len = len < 0 ? 0 : len; | |
2ae02107 | 1579 | ext_debug("insert %d:%llu:%d after: nearest 0x%p, " |
a86c6181 | 1580 | "move %d from 0x%p to 0x%p\n", |
8c55e204 DK |
1581 | le32_to_cpu(newext->ee_block), |
1582 | ext_pblock(newext), | |
a2df2a63 | 1583 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1584 | nearex, len, nearex + 1, nearex + 2); |
1585 | memmove(nearex + 2, nearex + 1, len); | |
1586 | } | |
1587 | path[depth].p_ext = nearex + 1; | |
1588 | } else { | |
1589 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1590 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1591 | len = len < 0 ? 0 : len; | |
2ae02107 | 1592 | ext_debug("insert %d:%llu:%d before: nearest 0x%p, " |
a86c6181 AT |
1593 | "move %d from 0x%p to 0x%p\n", |
1594 | le32_to_cpu(newext->ee_block), | |
f65e6fba | 1595 | ext_pblock(newext), |
a2df2a63 | 1596 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1597 | nearex, len, nearex + 1, nearex + 2); |
1598 | memmove(nearex + 1, nearex, len); | |
1599 | path[depth].p_ext = nearex; | |
1600 | } | |
1601 | ||
e8546d06 | 1602 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1603 | nearex = path[depth].p_ext; |
1604 | nearex->ee_block = newext->ee_block; | |
b377611d | 1605 | ext4_ext_store_pblock(nearex, ext_pblock(newext)); |
a86c6181 | 1606 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1607 | |
1608 | merge: | |
1609 | /* try to merge extents to the right */ | |
56055d3a | 1610 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1611 | |
1612 | /* try to merge extents to the left */ | |
1613 | ||
1614 | /* time to correct all indexes above */ | |
1615 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1616 | if (err) | |
1617 | goto cleanup; | |
1618 | ||
1619 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1620 | ||
1621 | cleanup: | |
1622 | if (npath) { | |
1623 | ext4_ext_drop_refs(npath); | |
1624 | kfree(npath); | |
1625 | } | |
a86c6181 AT |
1626 | ext4_ext_invalidate_cache(inode); |
1627 | return err; | |
1628 | } | |
1629 | ||
6873fa0d ES |
1630 | int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block, |
1631 | ext4_lblk_t num, ext_prepare_callback func, | |
1632 | void *cbdata) | |
1633 | { | |
1634 | struct ext4_ext_path *path = NULL; | |
1635 | struct ext4_ext_cache cbex; | |
1636 | struct ext4_extent *ex; | |
1637 | ext4_lblk_t next, start = 0, end = 0; | |
1638 | ext4_lblk_t last = block + num; | |
1639 | int depth, exists, err = 0; | |
1640 | ||
1641 | BUG_ON(func == NULL); | |
1642 | BUG_ON(inode == NULL); | |
1643 | ||
1644 | while (block < last && block != EXT_MAX_BLOCK) { | |
1645 | num = last - block; | |
1646 | /* find extent for this block */ | |
1647 | path = ext4_ext_find_extent(inode, block, path); | |
1648 | if (IS_ERR(path)) { | |
1649 | err = PTR_ERR(path); | |
1650 | path = NULL; | |
1651 | break; | |
1652 | } | |
1653 | ||
1654 | depth = ext_depth(inode); | |
1655 | BUG_ON(path[depth].p_hdr == NULL); | |
1656 | ex = path[depth].p_ext; | |
1657 | next = ext4_ext_next_allocated_block(path); | |
1658 | ||
1659 | exists = 0; | |
1660 | if (!ex) { | |
1661 | /* there is no extent yet, so try to allocate | |
1662 | * all requested space */ | |
1663 | start = block; | |
1664 | end = block + num; | |
1665 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
1666 | /* need to allocate space before found extent */ | |
1667 | start = block; | |
1668 | end = le32_to_cpu(ex->ee_block); | |
1669 | if (block + num < end) | |
1670 | end = block + num; | |
1671 | } else if (block >= le32_to_cpu(ex->ee_block) | |
1672 | + ext4_ext_get_actual_len(ex)) { | |
1673 | /* need to allocate space after found extent */ | |
1674 | start = block; | |
1675 | end = block + num; | |
1676 | if (end >= next) | |
1677 | end = next; | |
1678 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
1679 | /* | |
1680 | * some part of requested space is covered | |
1681 | * by found extent | |
1682 | */ | |
1683 | start = block; | |
1684 | end = le32_to_cpu(ex->ee_block) | |
1685 | + ext4_ext_get_actual_len(ex); | |
1686 | if (block + num < end) | |
1687 | end = block + num; | |
1688 | exists = 1; | |
1689 | } else { | |
1690 | BUG(); | |
1691 | } | |
1692 | BUG_ON(end <= start); | |
1693 | ||
1694 | if (!exists) { | |
1695 | cbex.ec_block = start; | |
1696 | cbex.ec_len = end - start; | |
1697 | cbex.ec_start = 0; | |
1698 | cbex.ec_type = EXT4_EXT_CACHE_GAP; | |
1699 | } else { | |
1700 | cbex.ec_block = le32_to_cpu(ex->ee_block); | |
1701 | cbex.ec_len = ext4_ext_get_actual_len(ex); | |
1702 | cbex.ec_start = ext_pblock(ex); | |
1703 | cbex.ec_type = EXT4_EXT_CACHE_EXTENT; | |
1704 | } | |
1705 | ||
1706 | BUG_ON(cbex.ec_len == 0); | |
1707 | err = func(inode, path, &cbex, ex, cbdata); | |
1708 | ext4_ext_drop_refs(path); | |
1709 | ||
1710 | if (err < 0) | |
1711 | break; | |
1712 | ||
1713 | if (err == EXT_REPEAT) | |
1714 | continue; | |
1715 | else if (err == EXT_BREAK) { | |
1716 | err = 0; | |
1717 | break; | |
1718 | } | |
1719 | ||
1720 | if (ext_depth(inode) != depth) { | |
1721 | /* depth was changed. we have to realloc path */ | |
1722 | kfree(path); | |
1723 | path = NULL; | |
1724 | } | |
1725 | ||
1726 | block = cbex.ec_block + cbex.ec_len; | |
1727 | } | |
1728 | ||
1729 | if (path) { | |
1730 | ext4_ext_drop_refs(path); | |
1731 | kfree(path); | |
1732 | } | |
1733 | ||
1734 | return err; | |
1735 | } | |
1736 | ||
09b88252 | 1737 | static void |
725d26d3 | 1738 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
dd54567a | 1739 | __u32 len, ext4_fsblk_t start, int type) |
a86c6181 AT |
1740 | { |
1741 | struct ext4_ext_cache *cex; | |
1742 | BUG_ON(len == 0); | |
1743 | cex = &EXT4_I(inode)->i_cached_extent; | |
1744 | cex->ec_type = type; | |
1745 | cex->ec_block = block; | |
1746 | cex->ec_len = len; | |
1747 | cex->ec_start = start; | |
1748 | } | |
1749 | ||
1750 | /* | |
d0d856e8 RD |
1751 | * ext4_ext_put_gap_in_cache: |
1752 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
1753 | * and cache this gap |
1754 | */ | |
09b88252 | 1755 | static void |
a86c6181 | 1756 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 1757 | ext4_lblk_t block) |
a86c6181 AT |
1758 | { |
1759 | int depth = ext_depth(inode); | |
725d26d3 AK |
1760 | unsigned long len; |
1761 | ext4_lblk_t lblock; | |
a86c6181 AT |
1762 | struct ext4_extent *ex; |
1763 | ||
1764 | ex = path[depth].p_ext; | |
1765 | if (ex == NULL) { | |
1766 | /* there is no extent yet, so gap is [0;-] */ | |
1767 | lblock = 0; | |
1768 | len = EXT_MAX_BLOCK; | |
1769 | ext_debug("cache gap(whole file):"); | |
1770 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
1771 | lblock = block; | |
1772 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
1773 | ext_debug("cache gap(before): %u [%u:%u]", |
1774 | block, | |
1775 | le32_to_cpu(ex->ee_block), | |
1776 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 1777 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 1778 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 1779 | ext4_lblk_t next; |
8c55e204 | 1780 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 1781 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
1782 | |
1783 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
1784 | ext_debug("cache gap(after): [%u:%u] %u", |
1785 | le32_to_cpu(ex->ee_block), | |
1786 | ext4_ext_get_actual_len(ex), | |
1787 | block); | |
725d26d3 AK |
1788 | BUG_ON(next == lblock); |
1789 | len = next - lblock; | |
a86c6181 AT |
1790 | } else { |
1791 | lblock = len = 0; | |
1792 | BUG(); | |
1793 | } | |
1794 | ||
bba90743 | 1795 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
1796 | ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP); |
1797 | } | |
1798 | ||
09b88252 | 1799 | static int |
725d26d3 | 1800 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, |
a86c6181 AT |
1801 | struct ext4_extent *ex) |
1802 | { | |
1803 | struct ext4_ext_cache *cex; | |
1804 | ||
1805 | cex = &EXT4_I(inode)->i_cached_extent; | |
1806 | ||
1807 | /* has cache valid data? */ | |
1808 | if (cex->ec_type == EXT4_EXT_CACHE_NO) | |
1809 | return EXT4_EXT_CACHE_NO; | |
1810 | ||
1811 | BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP && | |
1812 | cex->ec_type != EXT4_EXT_CACHE_EXTENT); | |
1813 | if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) { | |
8c55e204 | 1814 | ex->ee_block = cpu_to_le32(cex->ec_block); |
f65e6fba | 1815 | ext4_ext_store_pblock(ex, cex->ec_start); |
8c55e204 | 1816 | ex->ee_len = cpu_to_le16(cex->ec_len); |
bba90743 ES |
1817 | ext_debug("%u cached by %u:%u:%llu\n", |
1818 | block, | |
1819 | cex->ec_block, cex->ec_len, cex->ec_start); | |
a86c6181 AT |
1820 | return cex->ec_type; |
1821 | } | |
1822 | ||
1823 | /* not in cache */ | |
1824 | return EXT4_EXT_CACHE_NO; | |
1825 | } | |
1826 | ||
1827 | /* | |
d0d856e8 RD |
1828 | * ext4_ext_rm_idx: |
1829 | * removes index from the index block. | |
1830 | * It's used in truncate case only, thus all requests are for | |
1831 | * last index in the block only. | |
a86c6181 | 1832 | */ |
1d03ec98 | 1833 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1834 | struct ext4_ext_path *path) |
1835 | { | |
1836 | struct buffer_head *bh; | |
1837 | int err; | |
f65e6fba | 1838 | ext4_fsblk_t leaf; |
a86c6181 AT |
1839 | |
1840 | /* free index block */ | |
1841 | path--; | |
f65e6fba | 1842 | leaf = idx_pblock(path->p_idx); |
a86c6181 | 1843 | BUG_ON(path->p_hdr->eh_entries == 0); |
7e028976 AM |
1844 | err = ext4_ext_get_access(handle, inode, path); |
1845 | if (err) | |
a86c6181 | 1846 | return err; |
e8546d06 | 1847 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
1848 | err = ext4_ext_dirty(handle, inode, path); |
1849 | if (err) | |
a86c6181 | 1850 | return err; |
2ae02107 | 1851 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
a86c6181 AT |
1852 | bh = sb_find_get_block(inode->i_sb, leaf); |
1853 | ext4_forget(handle, 1, inode, bh, leaf); | |
c9de560d | 1854 | ext4_free_blocks(handle, inode, leaf, 1, 1); |
a86c6181 AT |
1855 | return err; |
1856 | } | |
1857 | ||
1858 | /* | |
ee12b630 MC |
1859 | * ext4_ext_calc_credits_for_single_extent: |
1860 | * This routine returns max. credits that needed to insert an extent | |
1861 | * to the extent tree. | |
1862 | * When pass the actual path, the caller should calculate credits | |
1863 | * under i_data_sem. | |
a86c6181 | 1864 | */ |
525f4ed8 | 1865 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
1866 | struct ext4_ext_path *path) |
1867 | { | |
a86c6181 | 1868 | if (path) { |
ee12b630 | 1869 | int depth = ext_depth(inode); |
f3bd1f3f | 1870 | int ret = 0; |
ee12b630 | 1871 | |
a86c6181 | 1872 | /* probably there is space in leaf? */ |
a86c6181 | 1873 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 1874 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 1875 | |
ee12b630 MC |
1876 | /* |
1877 | * There are some space in the leaf tree, no | |
1878 | * need to account for leaf block credit | |
1879 | * | |
1880 | * bitmaps and block group descriptor blocks | |
1881 | * and other metadat blocks still need to be | |
1882 | * accounted. | |
1883 | */ | |
525f4ed8 | 1884 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 MC |
1885 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
1886 | } | |
1887 | } | |
a86c6181 | 1888 | |
525f4ed8 | 1889 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 1890 | } |
a86c6181 | 1891 | |
ee12b630 MC |
1892 | /* |
1893 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
1894 | * | |
1895 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
1896 | * in the worse case, each tree level index/leaf need to be changed | |
1897 | * if the tree split due to insert a new extent, then the old tree | |
1898 | * index/leaf need to be updated too | |
1899 | * | |
1900 | * If the nrblocks are discontiguous, they could cause | |
1901 | * the whole tree split more than once, but this is really rare. | |
1902 | */ | |
525f4ed8 | 1903 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
1904 | { |
1905 | int index; | |
1906 | int depth = ext_depth(inode); | |
a86c6181 | 1907 | |
ee12b630 MC |
1908 | if (chunk) |
1909 | index = depth * 2; | |
1910 | else | |
1911 | index = depth * 3; | |
a86c6181 | 1912 | |
ee12b630 | 1913 | return index; |
a86c6181 AT |
1914 | } |
1915 | ||
1916 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
1917 | struct ext4_extent *ex, | |
725d26d3 | 1918 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 AT |
1919 | { |
1920 | struct buffer_head *bh; | |
a2df2a63 | 1921 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
c9de560d | 1922 | int i, metadata = 0; |
a86c6181 | 1923 | |
c9de560d AT |
1924 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
1925 | metadata = 1; | |
a86c6181 AT |
1926 | #ifdef EXTENTS_STATS |
1927 | { | |
1928 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
1929 | spin_lock(&sbi->s_ext_stats_lock); |
1930 | sbi->s_ext_blocks += ee_len; | |
1931 | sbi->s_ext_extents++; | |
1932 | if (ee_len < sbi->s_ext_min) | |
1933 | sbi->s_ext_min = ee_len; | |
1934 | if (ee_len > sbi->s_ext_max) | |
1935 | sbi->s_ext_max = ee_len; | |
1936 | if (ext_depth(inode) > sbi->s_depth_max) | |
1937 | sbi->s_depth_max = ext_depth(inode); | |
1938 | spin_unlock(&sbi->s_ext_stats_lock); | |
1939 | } | |
1940 | #endif | |
1941 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 1942 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 1943 | /* tail removal */ |
725d26d3 | 1944 | ext4_lblk_t num; |
f65e6fba | 1945 | ext4_fsblk_t start; |
725d26d3 | 1946 | |
a2df2a63 AA |
1947 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
1948 | start = ext_pblock(ex) + ee_len - num; | |
725d26d3 | 1949 | ext_debug("free last %u blocks starting %llu\n", num, start); |
a86c6181 AT |
1950 | for (i = 0; i < num; i++) { |
1951 | bh = sb_find_get_block(inode->i_sb, start + i); | |
1952 | ext4_forget(handle, 0, inode, bh, start + i); | |
1953 | } | |
c9de560d | 1954 | ext4_free_blocks(handle, inode, start, num, metadata); |
a86c6181 | 1955 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 1956 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
725d26d3 | 1957 | printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n", |
a2df2a63 | 1958 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
a86c6181 | 1959 | } else { |
725d26d3 AK |
1960 | printk(KERN_INFO "strange request: removal(2) " |
1961 | "%u-%u from %u:%u\n", | |
1962 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
1963 | } |
1964 | return 0; | |
1965 | } | |
1966 | ||
1967 | static int | |
1968 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
725d26d3 | 1969 | struct ext4_ext_path *path, ext4_lblk_t start) |
a86c6181 AT |
1970 | { |
1971 | int err = 0, correct_index = 0; | |
1972 | int depth = ext_depth(inode), credits; | |
1973 | struct ext4_extent_header *eh; | |
725d26d3 AK |
1974 | ext4_lblk_t a, b, block; |
1975 | unsigned num; | |
1976 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 1977 | unsigned short ex_ee_len; |
a2df2a63 | 1978 | unsigned uninitialized = 0; |
a86c6181 AT |
1979 | struct ext4_extent *ex; |
1980 | ||
c29c0ae7 | 1981 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 1982 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
1983 | if (!path[depth].p_hdr) |
1984 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
1985 | eh = path[depth].p_hdr; | |
1986 | BUG_ON(eh == NULL); | |
a86c6181 AT |
1987 | |
1988 | /* find where to start removing */ | |
1989 | ex = EXT_LAST_EXTENT(eh); | |
1990 | ||
1991 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 AA |
1992 | if (ext4_ext_is_uninitialized(ex)) |
1993 | uninitialized = 1; | |
1994 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
a86c6181 AT |
1995 | |
1996 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
1997 | ex_ee_block + ex_ee_len > start) { | |
1998 | ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len); | |
1999 | path[depth].p_ext = ex; | |
2000 | ||
2001 | a = ex_ee_block > start ? ex_ee_block : start; | |
2002 | b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? | |
2003 | ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; | |
2004 | ||
2005 | ext_debug(" border %u:%u\n", a, b); | |
2006 | ||
2007 | if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { | |
2008 | block = 0; | |
2009 | num = 0; | |
2010 | BUG(); | |
2011 | } else if (a != ex_ee_block) { | |
2012 | /* remove tail of the extent */ | |
2013 | block = ex_ee_block; | |
2014 | num = a - block; | |
2015 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
2016 | /* remove head of the extent */ | |
2017 | block = a; | |
2018 | num = b - a; | |
2019 | /* there is no "make a hole" API yet */ | |
2020 | BUG(); | |
2021 | } else { | |
2022 | /* remove whole extent: excellent! */ | |
2023 | block = ex_ee_block; | |
2024 | num = 0; | |
2025 | BUG_ON(a != ex_ee_block); | |
2026 | BUG_ON(b != ex_ee_block + ex_ee_len - 1); | |
2027 | } | |
2028 | ||
34071da7 TT |
2029 | /* |
2030 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2031 | * descriptor) for each block group; assume two block | |
2032 | * groups plus ex_ee_len/blocks_per_block_group for | |
2033 | * the worst case | |
2034 | */ | |
2035 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2036 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2037 | correct_index = 1; | |
2038 | credits += (ext_depth(inode)) + 1; | |
2039 | } | |
a86c6181 | 2040 | credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2041 | |
9102e4fa SF |
2042 | err = ext4_ext_journal_restart(handle, credits); |
2043 | if (err) | |
a86c6181 | 2044 | goto out; |
a86c6181 AT |
2045 | |
2046 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2047 | if (err) | |
2048 | goto out; | |
2049 | ||
2050 | err = ext4_remove_blocks(handle, inode, ex, a, b); | |
2051 | if (err) | |
2052 | goto out; | |
2053 | ||
2054 | if (num == 0) { | |
d0d856e8 | 2055 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2056 | ext4_ext_store_pblock(ex, 0); |
e8546d06 | 2057 | le16_add_cpu(&eh->eh_entries, -1); |
a86c6181 AT |
2058 | } |
2059 | ||
2060 | ex->ee_block = cpu_to_le32(block); | |
2061 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
2062 | /* |
2063 | * Do not mark uninitialized if all the blocks in the | |
2064 | * extent have been removed. | |
2065 | */ | |
2066 | if (uninitialized && num) | |
a2df2a63 | 2067 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
2068 | |
2069 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2070 | if (err) | |
2071 | goto out; | |
2072 | ||
2ae02107 | 2073 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
f65e6fba | 2074 | ext_pblock(ex)); |
a86c6181 AT |
2075 | ex--; |
2076 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2077 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2078 | } |
2079 | ||
2080 | if (correct_index && eh->eh_entries) | |
2081 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2082 | ||
2083 | /* if this leaf is free, then we should | |
2084 | * remove it from index block above */ | |
2085 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
2086 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
2087 | ||
2088 | out: | |
2089 | return err; | |
2090 | } | |
2091 | ||
2092 | /* | |
d0d856e8 RD |
2093 | * ext4_ext_more_to_rm: |
2094 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2095 | */ |
09b88252 | 2096 | static int |
a86c6181 AT |
2097 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2098 | { | |
2099 | BUG_ON(path->p_idx == NULL); | |
2100 | ||
2101 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2102 | return 0; | |
2103 | ||
2104 | /* | |
d0d856e8 | 2105 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2106 | * so we have to consider current index for truncation |
2107 | */ | |
2108 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2109 | return 0; | |
2110 | return 1; | |
2111 | } | |
2112 | ||
1d03ec98 | 2113 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
2114 | { |
2115 | struct super_block *sb = inode->i_sb; | |
2116 | int depth = ext_depth(inode); | |
2117 | struct ext4_ext_path *path; | |
2118 | handle_t *handle; | |
2119 | int i = 0, err = 0; | |
2120 | ||
725d26d3 | 2121 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
2122 | |
2123 | /* probably first extent we're gonna free will be last in block */ | |
2124 | handle = ext4_journal_start(inode, depth + 1); | |
2125 | if (IS_ERR(handle)) | |
2126 | return PTR_ERR(handle); | |
2127 | ||
2128 | ext4_ext_invalidate_cache(inode); | |
2129 | ||
2130 | /* | |
d0d856e8 RD |
2131 | * We start scanning from right side, freeing all the blocks |
2132 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2133 | */ |
216553c4 | 2134 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
2135 | if (path == NULL) { |
2136 | ext4_journal_stop(handle); | |
2137 | return -ENOMEM; | |
2138 | } | |
a86c6181 | 2139 | path[0].p_hdr = ext_inode_hdr(inode); |
c29c0ae7 | 2140 | if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
2141 | err = -EIO; |
2142 | goto out; | |
2143 | } | |
2144 | path[0].p_depth = depth; | |
2145 | ||
2146 | while (i >= 0 && err == 0) { | |
2147 | if (i == depth) { | |
2148 | /* this is leaf block */ | |
2149 | err = ext4_ext_rm_leaf(handle, inode, path, start); | |
d0d856e8 | 2150 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2151 | brelse(path[i].p_bh); |
2152 | path[i].p_bh = NULL; | |
2153 | i--; | |
2154 | continue; | |
2155 | } | |
2156 | ||
2157 | /* this is index block */ | |
2158 | if (!path[i].p_hdr) { | |
2159 | ext_debug("initialize header\n"); | |
2160 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2161 | } |
2162 | ||
a86c6181 | 2163 | if (!path[i].p_idx) { |
d0d856e8 | 2164 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2165 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2166 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2167 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2168 | path[i].p_hdr, | |
2169 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2170 | } else { | |
d0d856e8 | 2171 | /* we were already here, see at next index */ |
a86c6181 AT |
2172 | path[i].p_idx--; |
2173 | } | |
2174 | ||
2175 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2176 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2177 | path[i].p_idx); | |
2178 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2179 | struct buffer_head *bh; |
a86c6181 | 2180 | /* go to the next level */ |
2ae02107 | 2181 | ext_debug("move to level %d (block %llu)\n", |
f65e6fba | 2182 | i + 1, idx_pblock(path[i].p_idx)); |
a86c6181 | 2183 | memset(path + i + 1, 0, sizeof(*path)); |
c29c0ae7 AT |
2184 | bh = sb_bread(sb, idx_pblock(path[i].p_idx)); |
2185 | if (!bh) { | |
a86c6181 AT |
2186 | /* should we reset i_size? */ |
2187 | err = -EIO; | |
2188 | break; | |
2189 | } | |
c29c0ae7 AT |
2190 | if (WARN_ON(i + 1 > depth)) { |
2191 | err = -EIO; | |
2192 | break; | |
2193 | } | |
2194 | if (ext4_ext_check_header(inode, ext_block_hdr(bh), | |
2195 | depth - i - 1)) { | |
2196 | err = -EIO; | |
2197 | break; | |
2198 | } | |
2199 | path[i + 1].p_bh = bh; | |
a86c6181 | 2200 | |
d0d856e8 RD |
2201 | /* save actual number of indexes since this |
2202 | * number is changed at the next iteration */ | |
a86c6181 AT |
2203 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2204 | i++; | |
2205 | } else { | |
d0d856e8 | 2206 | /* we finished processing this index, go up */ |
a86c6181 | 2207 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2208 | /* index is empty, remove it; |
a86c6181 AT |
2209 | * handle must be already prepared by the |
2210 | * truncatei_leaf() */ | |
2211 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2212 | } | |
d0d856e8 | 2213 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2214 | brelse(path[i].p_bh); |
2215 | path[i].p_bh = NULL; | |
2216 | i--; | |
2217 | ext_debug("return to level %d\n", i); | |
2218 | } | |
2219 | } | |
2220 | ||
2221 | /* TODO: flexible tree reduction should be here */ | |
2222 | if (path->p_hdr->eh_entries == 0) { | |
2223 | /* | |
d0d856e8 RD |
2224 | * truncate to zero freed all the tree, |
2225 | * so we need to correct eh_depth | |
a86c6181 AT |
2226 | */ |
2227 | err = ext4_ext_get_access(handle, inode, path); | |
2228 | if (err == 0) { | |
2229 | ext_inode_hdr(inode)->eh_depth = 0; | |
2230 | ext_inode_hdr(inode)->eh_max = | |
2231 | cpu_to_le16(ext4_ext_space_root(inode)); | |
2232 | err = ext4_ext_dirty(handle, inode, path); | |
2233 | } | |
2234 | } | |
2235 | out: | |
a86c6181 AT |
2236 | ext4_ext_drop_refs(path); |
2237 | kfree(path); | |
2238 | ext4_journal_stop(handle); | |
2239 | ||
2240 | return err; | |
2241 | } | |
2242 | ||
2243 | /* | |
2244 | * called at mount time | |
2245 | */ | |
2246 | void ext4_ext_init(struct super_block *sb) | |
2247 | { | |
2248 | /* | |
2249 | * possible initialization would be here | |
2250 | */ | |
2251 | ||
83982b6f | 2252 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
4776004f | 2253 | printk(KERN_INFO "EXT4-fs: file extents enabled"); |
bbf2f9fb RD |
2254 | #ifdef AGGRESSIVE_TEST |
2255 | printk(", aggressive tests"); | |
a86c6181 AT |
2256 | #endif |
2257 | #ifdef CHECK_BINSEARCH | |
2258 | printk(", check binsearch"); | |
2259 | #endif | |
2260 | #ifdef EXTENTS_STATS | |
2261 | printk(", stats"); | |
2262 | #endif | |
2263 | printk("\n"); | |
2264 | #ifdef EXTENTS_STATS | |
2265 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2266 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2267 | EXT4_SB(sb)->s_ext_max = 0; | |
2268 | #endif | |
2269 | } | |
2270 | } | |
2271 | ||
2272 | /* | |
2273 | * called at umount time | |
2274 | */ | |
2275 | void ext4_ext_release(struct super_block *sb) | |
2276 | { | |
83982b6f | 2277 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2278 | return; |
2279 | ||
2280 | #ifdef EXTENTS_STATS | |
2281 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2282 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2283 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2284 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2285 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2286 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2287 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2288 | } | |
2289 | #endif | |
2290 | } | |
2291 | ||
093a088b AK |
2292 | static void bi_complete(struct bio *bio, int error) |
2293 | { | |
2294 | complete((struct completion *)bio->bi_private); | |
2295 | } | |
2296 | ||
2297 | /* FIXME!! we need to try to merge to left or right after zero-out */ | |
2298 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2299 | { | |
2300 | int ret = -EIO; | |
2301 | struct bio *bio; | |
2302 | int blkbits, blocksize; | |
2303 | sector_t ee_pblock; | |
2304 | struct completion event; | |
2305 | unsigned int ee_len, len, done, offset; | |
2306 | ||
2307 | ||
2308 | blkbits = inode->i_blkbits; | |
2309 | blocksize = inode->i_sb->s_blocksize; | |
2310 | ee_len = ext4_ext_get_actual_len(ex); | |
2311 | ee_pblock = ext_pblock(ex); | |
2312 | ||
2313 | /* convert ee_pblock to 512 byte sectors */ | |
2314 | ee_pblock = ee_pblock << (blkbits - 9); | |
2315 | ||
2316 | while (ee_len > 0) { | |
2317 | ||
2318 | if (ee_len > BIO_MAX_PAGES) | |
2319 | len = BIO_MAX_PAGES; | |
2320 | else | |
2321 | len = ee_len; | |
2322 | ||
2323 | bio = bio_alloc(GFP_NOIO, len); | |
2324 | if (!bio) | |
2325 | return -ENOMEM; | |
2326 | bio->bi_sector = ee_pblock; | |
2327 | bio->bi_bdev = inode->i_sb->s_bdev; | |
2328 | ||
2329 | done = 0; | |
2330 | offset = 0; | |
2331 | while (done < len) { | |
2332 | ret = bio_add_page(bio, ZERO_PAGE(0), | |
2333 | blocksize, offset); | |
2334 | if (ret != blocksize) { | |
2335 | /* | |
2336 | * We can't add any more pages because of | |
2337 | * hardware limitations. Start a new bio. | |
2338 | */ | |
2339 | break; | |
2340 | } | |
2341 | done++; | |
2342 | offset += blocksize; | |
2343 | if (offset >= PAGE_CACHE_SIZE) | |
2344 | offset = 0; | |
2345 | } | |
2346 | ||
2347 | init_completion(&event); | |
2348 | bio->bi_private = &event; | |
2349 | bio->bi_end_io = bi_complete; | |
2350 | submit_bio(WRITE, bio); | |
2351 | wait_for_completion(&event); | |
2352 | ||
2353 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2354 | ret = 0; | |
2355 | else { | |
2356 | ret = -EIO; | |
2357 | break; | |
2358 | } | |
2359 | bio_put(bio); | |
2360 | ee_len -= done; | |
2361 | ee_pblock += done << (blkbits - 9); | |
2362 | } | |
2363 | return ret; | |
2364 | } | |
2365 | ||
3977c965 AK |
2366 | #define EXT4_EXT_ZERO_LEN 7 |
2367 | ||
56055d3a AA |
2368 | /* |
2369 | * This function is called by ext4_ext_get_blocks() if someone tries to write | |
2370 | * to an uninitialized extent. It may result in splitting the uninitialized | |
2371 | * extent into multiple extents (upto three - one initialized and two | |
2372 | * uninitialized). | |
2373 | * There are three possibilities: | |
2374 | * a> There is no split required: Entire extent should be initialized | |
2375 | * b> Splits in two extents: Write is happening at either end of the extent | |
2376 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2377 | */ | |
725d26d3 AK |
2378 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
2379 | struct inode *inode, | |
2380 | struct ext4_ext_path *path, | |
2381 | ext4_lblk_t iblock, | |
498e5f24 | 2382 | unsigned int max_blocks) |
56055d3a | 2383 | { |
95c3889c | 2384 | struct ext4_extent *ex, newex, orig_ex; |
56055d3a AA |
2385 | struct ext4_extent *ex1 = NULL; |
2386 | struct ext4_extent *ex2 = NULL; | |
2387 | struct ext4_extent *ex3 = NULL; | |
2388 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2389 | ext4_lblk_t ee_block; |
2390 | unsigned int allocated, ee_len, depth; | |
56055d3a AA |
2391 | ext4_fsblk_t newblock; |
2392 | int err = 0; | |
2393 | int ret = 0; | |
2394 | ||
2395 | depth = ext_depth(inode); | |
2396 | eh = path[depth].p_hdr; | |
2397 | ex = path[depth].p_ext; | |
2398 | ee_block = le32_to_cpu(ex->ee_block); | |
2399 | ee_len = ext4_ext_get_actual_len(ex); | |
2400 | allocated = ee_len - (iblock - ee_block); | |
2401 | newblock = iblock - ee_block + ext_pblock(ex); | |
2402 | ex2 = ex; | |
95c3889c AK |
2403 | orig_ex.ee_block = ex->ee_block; |
2404 | orig_ex.ee_len = cpu_to_le16(ee_len); | |
2405 | ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); | |
56055d3a | 2406 | |
9df5643a AK |
2407 | err = ext4_ext_get_access(handle, inode, path + depth); |
2408 | if (err) | |
2409 | goto out; | |
3977c965 AK |
2410 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
2411 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN) { | |
2412 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2413 | if (err) | |
2414 | goto fix_extent_len; | |
2415 | /* update the extent length and mark as initialized */ | |
2416 | ex->ee_block = orig_ex.ee_block; | |
2417 | ex->ee_len = orig_ex.ee_len; | |
2418 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2419 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2420 | /* zeroed the full extent */ |
2421 | return allocated; | |
3977c965 | 2422 | } |
9df5643a | 2423 | |
56055d3a AA |
2424 | /* ex1: ee_block to iblock - 1 : uninitialized */ |
2425 | if (iblock > ee_block) { | |
2426 | ex1 = ex; | |
2427 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2428 | ext4_ext_mark_uninitialized(ex1); | |
2429 | ex2 = &newex; | |
2430 | } | |
2431 | /* | |
2432 | * for sanity, update the length of the ex2 extent before | |
2433 | * we insert ex3, if ex1 is NULL. This is to avoid temporary | |
2434 | * overlap of blocks. | |
2435 | */ | |
2436 | if (!ex1 && allocated > max_blocks) | |
2437 | ex2->ee_len = cpu_to_le16(max_blocks); | |
2438 | /* ex3: to ee_block + ee_len : uninitialised */ | |
2439 | if (allocated > max_blocks) { | |
2440 | unsigned int newdepth; | |
3977c965 AK |
2441 | /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */ |
2442 | if (allocated <= EXT4_EXT_ZERO_LEN) { | |
d03856bd AK |
2443 | /* |
2444 | * iblock == ee_block is handled by the zerouout | |
2445 | * at the beginning. | |
2446 | * Mark first half uninitialized. | |
3977c965 AK |
2447 | * Mark second half initialized and zero out the |
2448 | * initialized extent | |
2449 | */ | |
2450 | ex->ee_block = orig_ex.ee_block; | |
2451 | ex->ee_len = cpu_to_le16(ee_len - allocated); | |
2452 | ext4_ext_mark_uninitialized(ex); | |
2453 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2454 | ext4_ext_dirty(handle, inode, path + depth); | |
2455 | ||
2456 | ex3 = &newex; | |
2457 | ex3->ee_block = cpu_to_le32(iblock); | |
2458 | ext4_ext_store_pblock(ex3, newblock); | |
2459 | ex3->ee_len = cpu_to_le16(allocated); | |
2460 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
2461 | if (err == -ENOSPC) { | |
2462 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2463 | if (err) | |
2464 | goto fix_extent_len; | |
2465 | ex->ee_block = orig_ex.ee_block; | |
2466 | ex->ee_len = orig_ex.ee_len; | |
2467 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2468 | ext4_ext_dirty(handle, inode, path + depth); | |
d03856bd | 2469 | /* blocks available from iblock */ |
161e7b7c | 2470 | return allocated; |
3977c965 AK |
2471 | |
2472 | } else if (err) | |
2473 | goto fix_extent_len; | |
2474 | ||
161e7b7c AK |
2475 | /* |
2476 | * We need to zero out the second half because | |
2477 | * an fallocate request can update file size and | |
2478 | * converting the second half to initialized extent | |
2479 | * implies that we can leak some junk data to user | |
2480 | * space. | |
2481 | */ | |
2482 | err = ext4_ext_zeroout(inode, ex3); | |
2483 | if (err) { | |
2484 | /* | |
2485 | * We should actually mark the | |
2486 | * second half as uninit and return error | |
2487 | * Insert would have changed the extent | |
2488 | */ | |
2489 | depth = ext_depth(inode); | |
2490 | ext4_ext_drop_refs(path); | |
2491 | path = ext4_ext_find_extent(inode, | |
2492 | iblock, path); | |
2493 | if (IS_ERR(path)) { | |
2494 | err = PTR_ERR(path); | |
2495 | return err; | |
2496 | } | |
d03856bd | 2497 | /* get the second half extent details */ |
161e7b7c AK |
2498 | ex = path[depth].p_ext; |
2499 | err = ext4_ext_get_access(handle, inode, | |
2500 | path + depth); | |
2501 | if (err) | |
2502 | return err; | |
2503 | ext4_ext_mark_uninitialized(ex); | |
2504 | ext4_ext_dirty(handle, inode, path + depth); | |
2505 | return err; | |
2506 | } | |
2507 | ||
2508 | /* zeroed the second half */ | |
3977c965 AK |
2509 | return allocated; |
2510 | } | |
56055d3a AA |
2511 | ex3 = &newex; |
2512 | ex3->ee_block = cpu_to_le32(iblock + max_blocks); | |
2513 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | |
2514 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | |
2515 | ext4_ext_mark_uninitialized(ex3); | |
2516 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
093a088b AK |
2517 | if (err == -ENOSPC) { |
2518 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2519 | if (err) | |
2520 | goto fix_extent_len; | |
2521 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2522 | ex->ee_block = orig_ex.ee_block; |
2523 | ex->ee_len = orig_ex.ee_len; | |
2524 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2525 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c | 2526 | /* zeroed the full extent */ |
d03856bd | 2527 | /* blocks available from iblock */ |
161e7b7c | 2528 | return allocated; |
093a088b AK |
2529 | |
2530 | } else if (err) | |
2531 | goto fix_extent_len; | |
56055d3a AA |
2532 | /* |
2533 | * The depth, and hence eh & ex might change | |
2534 | * as part of the insert above. | |
2535 | */ | |
2536 | newdepth = ext_depth(inode); | |
95c3889c | 2537 | /* |
73ac36ea | 2538 | * update the extent length after successful insert of the |
95c3889c AK |
2539 | * split extent |
2540 | */ | |
2541 | orig_ex.ee_len = cpu_to_le16(ee_len - | |
2542 | ext4_ext_get_actual_len(ex3)); | |
d03856bd AK |
2543 | depth = newdepth; |
2544 | ext4_ext_drop_refs(path); | |
2545 | path = ext4_ext_find_extent(inode, iblock, path); | |
2546 | if (IS_ERR(path)) { | |
2547 | err = PTR_ERR(path); | |
2548 | goto out; | |
56055d3a | 2549 | } |
d03856bd AK |
2550 | eh = path[depth].p_hdr; |
2551 | ex = path[depth].p_ext; | |
2552 | if (ex2 != &newex) | |
2553 | ex2 = ex; | |
2554 | ||
2555 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2556 | if (err) | |
2557 | goto out; | |
2558 | ||
56055d3a | 2559 | allocated = max_blocks; |
3977c965 AK |
2560 | |
2561 | /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying | |
2562 | * to insert a extent in the middle zerout directly | |
2563 | * otherwise give the extent a chance to merge to left | |
2564 | */ | |
2565 | if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN && | |
2566 | iblock != ee_block) { | |
2567 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2568 | if (err) | |
2569 | goto fix_extent_len; | |
2570 | /* update the extent length and mark as initialized */ | |
2571 | ex->ee_block = orig_ex.ee_block; | |
2572 | ex->ee_len = orig_ex.ee_len; | |
2573 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2574 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c | 2575 | /* zero out the first half */ |
d03856bd | 2576 | /* blocks available from iblock */ |
161e7b7c | 2577 | return allocated; |
3977c965 | 2578 | } |
56055d3a AA |
2579 | } |
2580 | /* | |
2581 | * If there was a change of depth as part of the | |
2582 | * insertion of ex3 above, we need to update the length | |
2583 | * of the ex1 extent again here | |
2584 | */ | |
2585 | if (ex1 && ex1 != ex) { | |
2586 | ex1 = ex; | |
2587 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2588 | ext4_ext_mark_uninitialized(ex1); | |
2589 | ex2 = &newex; | |
2590 | } | |
2591 | /* ex2: iblock to iblock + maxblocks-1 : initialised */ | |
2592 | ex2->ee_block = cpu_to_le32(iblock); | |
56055d3a AA |
2593 | ext4_ext_store_pblock(ex2, newblock); |
2594 | ex2->ee_len = cpu_to_le16(allocated); | |
2595 | if (ex2 != ex) | |
2596 | goto insert; | |
56055d3a AA |
2597 | /* |
2598 | * New (initialized) extent starts from the first block | |
2599 | * in the current extent. i.e., ex2 == ex | |
2600 | * We have to see if it can be merged with the extent | |
2601 | * on the left. | |
2602 | */ | |
2603 | if (ex2 > EXT_FIRST_EXTENT(eh)) { | |
2604 | /* | |
2605 | * To merge left, pass "ex2 - 1" to try_to_merge(), | |
2606 | * since it merges towards right _only_. | |
2607 | */ | |
2608 | ret = ext4_ext_try_to_merge(inode, path, ex2 - 1); | |
2609 | if (ret) { | |
2610 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2611 | if (err) | |
2612 | goto out; | |
2613 | depth = ext_depth(inode); | |
2614 | ex2--; | |
2615 | } | |
2616 | } | |
2617 | /* | |
2618 | * Try to Merge towards right. This might be required | |
2619 | * only when the whole extent is being written to. | |
2620 | * i.e. ex2 == ex and ex3 == NULL. | |
2621 | */ | |
2622 | if (!ex3) { | |
2623 | ret = ext4_ext_try_to_merge(inode, path, ex2); | |
2624 | if (ret) { | |
2625 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2626 | if (err) | |
2627 | goto out; | |
2628 | } | |
2629 | } | |
2630 | /* Mark modified extent as dirty */ | |
2631 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2632 | goto out; | |
2633 | insert: | |
2634 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | |
093a088b AK |
2635 | if (err == -ENOSPC) { |
2636 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2637 | if (err) | |
2638 | goto fix_extent_len; | |
2639 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2640 | ex->ee_block = orig_ex.ee_block; |
2641 | ex->ee_len = orig_ex.ee_len; | |
2642 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2643 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c AK |
2644 | /* zero out the first half */ |
2645 | return allocated; | |
093a088b AK |
2646 | } else if (err) |
2647 | goto fix_extent_len; | |
56055d3a AA |
2648 | out: |
2649 | return err ? err : allocated; | |
093a088b AK |
2650 | |
2651 | fix_extent_len: | |
2652 | ex->ee_block = orig_ex.ee_block; | |
2653 | ex->ee_len = orig_ex.ee_len; | |
2654 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2655 | ext4_ext_mark_uninitialized(ex); | |
2656 | ext4_ext_dirty(handle, inode, path + depth); | |
2657 | return err; | |
56055d3a AA |
2658 | } |
2659 | ||
c278bfec | 2660 | /* |
f5ab0d1f MC |
2661 | * Block allocation/map/preallocation routine for extents based files |
2662 | * | |
2663 | * | |
c278bfec | 2664 | * Need to be called with |
0e855ac8 AK |
2665 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
2666 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
2667 | * |
2668 | * return > 0, number of of blocks already mapped/allocated | |
2669 | * if create == 0 and these are pre-allocated blocks | |
2670 | * buffer head is unmapped | |
2671 | * otherwise blocks are mapped | |
2672 | * | |
2673 | * return = 0, if plain look up failed (blocks have not been allocated) | |
2674 | * buffer head is unmapped | |
2675 | * | |
2676 | * return < 0, error case. | |
c278bfec | 2677 | */ |
f65e6fba | 2678 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, |
725d26d3 | 2679 | ext4_lblk_t iblock, |
498e5f24 | 2680 | unsigned int max_blocks, struct buffer_head *bh_result, |
a86c6181 AT |
2681 | int create, int extend_disksize) |
2682 | { | |
2683 | struct ext4_ext_path *path = NULL; | |
56055d3a | 2684 | struct ext4_extent_header *eh; |
a86c6181 | 2685 | struct ext4_extent newex, *ex; |
498e5f24 TT |
2686 | ext4_fsblk_t newblock; |
2687 | int err = 0, depth, ret, cache_type; | |
2688 | unsigned int allocated = 0; | |
c9de560d | 2689 | struct ext4_allocation_request ar; |
61628a3f | 2690 | loff_t disksize; |
a86c6181 AT |
2691 | |
2692 | __clear_bit(BH_New, &bh_result->b_state); | |
498e5f24 | 2693 | ext_debug("blocks %u/%u requested for inode %u\n", |
bba90743 | 2694 | iblock, max_blocks, inode->i_ino); |
a86c6181 AT |
2695 | |
2696 | /* check in cache */ | |
498e5f24 TT |
2697 | cache_type = ext4_ext_in_cache(inode, iblock, &newex); |
2698 | if (cache_type) { | |
2699 | if (cache_type == EXT4_EXT_CACHE_GAP) { | |
a86c6181 | 2700 | if (!create) { |
56055d3a AA |
2701 | /* |
2702 | * block isn't allocated yet and | |
2703 | * user doesn't want to allocate it | |
2704 | */ | |
a86c6181 AT |
2705 | goto out2; |
2706 | } | |
2707 | /* we should allocate requested block */ | |
498e5f24 | 2708 | } else if (cache_type == EXT4_EXT_CACHE_EXTENT) { |
a86c6181 | 2709 | /* block is already allocated */ |
8c55e204 DK |
2710 | newblock = iblock |
2711 | - le32_to_cpu(newex.ee_block) | |
2712 | + ext_pblock(&newex); | |
d0d856e8 | 2713 | /* number of remaining blocks in the extent */ |
b939e376 | 2714 | allocated = ext4_ext_get_actual_len(&newex) - |
a86c6181 AT |
2715 | (iblock - le32_to_cpu(newex.ee_block)); |
2716 | goto out; | |
2717 | } else { | |
2718 | BUG(); | |
2719 | } | |
2720 | } | |
2721 | ||
2722 | /* find extent for this block */ | |
2723 | path = ext4_ext_find_extent(inode, iblock, NULL); | |
2724 | if (IS_ERR(path)) { | |
2725 | err = PTR_ERR(path); | |
2726 | path = NULL; | |
2727 | goto out2; | |
2728 | } | |
2729 | ||
2730 | depth = ext_depth(inode); | |
2731 | ||
2732 | /* | |
d0d856e8 RD |
2733 | * consistent leaf must not be empty; |
2734 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
2735 | * this is why assert can't be put in ext4_ext_find_extent() |
2736 | */ | |
2737 | BUG_ON(path[depth].p_ext == NULL && depth != 0); | |
56055d3a | 2738 | eh = path[depth].p_hdr; |
a86c6181 | 2739 | |
7e028976 AM |
2740 | ex = path[depth].p_ext; |
2741 | if (ex) { | |
725d26d3 | 2742 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
f65e6fba | 2743 | ext4_fsblk_t ee_start = ext_pblock(ex); |
a2df2a63 | 2744 | unsigned short ee_len; |
471d4011 SB |
2745 | |
2746 | /* | |
471d4011 | 2747 | * Uninitialized extents are treated as holes, except that |
56055d3a | 2748 | * we split out initialized portions during a write. |
471d4011 | 2749 | */ |
a2df2a63 | 2750 | ee_len = ext4_ext_get_actual_len(ex); |
d0d856e8 | 2751 | /* if found extent covers block, simply return it */ |
8c55e204 | 2752 | if (iblock >= ee_block && iblock < ee_block + ee_len) { |
a86c6181 | 2753 | newblock = iblock - ee_block + ee_start; |
d0d856e8 | 2754 | /* number of remaining blocks in the extent */ |
a86c6181 | 2755 | allocated = ee_len - (iblock - ee_block); |
bba90743 | 2756 | ext_debug("%u fit into %lu:%d -> %llu\n", iblock, |
a86c6181 | 2757 | ee_block, ee_len, newblock); |
56055d3a | 2758 | |
a2df2a63 | 2759 | /* Do not put uninitialized extent in the cache */ |
56055d3a | 2760 | if (!ext4_ext_is_uninitialized(ex)) { |
a2df2a63 AA |
2761 | ext4_ext_put_in_cache(inode, ee_block, |
2762 | ee_len, ee_start, | |
2763 | EXT4_EXT_CACHE_EXTENT); | |
56055d3a AA |
2764 | goto out; |
2765 | } | |
2766 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) | |
2767 | goto out; | |
e067ba00 AK |
2768 | if (!create) { |
2769 | /* | |
2770 | * We have blocks reserved already. We | |
2771 | * return allocated blocks so that delalloc | |
2772 | * won't do block reservation for us. But | |
2773 | * the buffer head will be unmapped so that | |
2774 | * a read from the block returns 0s. | |
2775 | */ | |
2776 | if (allocated > max_blocks) | |
2777 | allocated = max_blocks; | |
953e622b | 2778 | set_buffer_unwritten(bh_result); |
56055d3a | 2779 | goto out2; |
e067ba00 | 2780 | } |
56055d3a AA |
2781 | |
2782 | ret = ext4_ext_convert_to_initialized(handle, inode, | |
2783 | path, iblock, | |
2784 | max_blocks); | |
dbf9d7da DM |
2785 | if (ret <= 0) { |
2786 | err = ret; | |
56055d3a | 2787 | goto out2; |
dbf9d7da | 2788 | } else |
56055d3a AA |
2789 | allocated = ret; |
2790 | goto outnew; | |
a86c6181 AT |
2791 | } |
2792 | } | |
2793 | ||
2794 | /* | |
d0d856e8 | 2795 | * requested block isn't allocated yet; |
a86c6181 AT |
2796 | * we couldn't try to create block if create flag is zero |
2797 | */ | |
2798 | if (!create) { | |
56055d3a AA |
2799 | /* |
2800 | * put just found gap into cache to speed up | |
2801 | * subsequent requests | |
2802 | */ | |
a86c6181 AT |
2803 | ext4_ext_put_gap_in_cache(inode, path, iblock); |
2804 | goto out2; | |
2805 | } | |
2806 | /* | |
c2ea3fde | 2807 | * Okay, we need to do block allocation. |
63f57933 | 2808 | */ |
a86c6181 | 2809 | |
c9de560d AT |
2810 | /* find neighbour allocated blocks */ |
2811 | ar.lleft = iblock; | |
2812 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); | |
2813 | if (err) | |
2814 | goto out2; | |
2815 | ar.lright = iblock; | |
2816 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright); | |
2817 | if (err) | |
2818 | goto out2; | |
25d14f98 | 2819 | |
749269fa AA |
2820 | /* |
2821 | * See if request is beyond maximum number of blocks we can have in | |
2822 | * a single extent. For an initialized extent this limit is | |
2823 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
2824 | * EXT_UNINIT_MAX_LEN. | |
2825 | */ | |
2826 | if (max_blocks > EXT_INIT_MAX_LEN && | |
2827 | create != EXT4_CREATE_UNINITIALIZED_EXT) | |
2828 | max_blocks = EXT_INIT_MAX_LEN; | |
2829 | else if (max_blocks > EXT_UNINIT_MAX_LEN && | |
2830 | create == EXT4_CREATE_UNINITIALIZED_EXT) | |
2831 | max_blocks = EXT_UNINIT_MAX_LEN; | |
2832 | ||
25d14f98 AA |
2833 | /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ |
2834 | newex.ee_block = cpu_to_le32(iblock); | |
2835 | newex.ee_len = cpu_to_le16(max_blocks); | |
2836 | err = ext4_ext_check_overlap(inode, &newex, path); | |
2837 | if (err) | |
b939e376 | 2838 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 AA |
2839 | else |
2840 | allocated = max_blocks; | |
c9de560d AT |
2841 | |
2842 | /* allocate new block */ | |
2843 | ar.inode = inode; | |
2844 | ar.goal = ext4_ext_find_goal(inode, path, iblock); | |
2845 | ar.logical = iblock; | |
2846 | ar.len = allocated; | |
2847 | if (S_ISREG(inode->i_mode)) | |
2848 | ar.flags = EXT4_MB_HINT_DATA; | |
2849 | else | |
2850 | /* disable in-core preallocation for non-regular files */ | |
2851 | ar.flags = 0; | |
2852 | newblock = ext4_mb_new_blocks(handle, &ar, &err); | |
a86c6181 AT |
2853 | if (!newblock) |
2854 | goto out2; | |
2ae02107 | 2855 | ext_debug("allocate new block: goal %llu, found %llu/%lu\n", |
498e5f24 | 2856 | ar.goal, newblock, allocated); |
a86c6181 AT |
2857 | |
2858 | /* try to insert new extent into found leaf and return */ | |
f65e6fba | 2859 | ext4_ext_store_pblock(&newex, newblock); |
c9de560d | 2860 | newex.ee_len = cpu_to_le16(ar.len); |
a2df2a63 AA |
2861 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */ |
2862 | ext4_ext_mark_uninitialized(&newex); | |
a86c6181 | 2863 | err = ext4_ext_insert_extent(handle, inode, path, &newex); |
315054f0 AT |
2864 | if (err) { |
2865 | /* free data blocks we just allocated */ | |
c9de560d AT |
2866 | /* not a good idea to call discard here directly, |
2867 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 2868 | ext4_discard_preallocations(inode); |
315054f0 | 2869 | ext4_free_blocks(handle, inode, ext_pblock(&newex), |
b939e376 | 2870 | ext4_ext_get_actual_len(&newex), 0); |
a86c6181 | 2871 | goto out2; |
315054f0 | 2872 | } |
a86c6181 | 2873 | |
a86c6181 | 2874 | /* previous routine could use block we allocated */ |
f65e6fba | 2875 | newblock = ext_pblock(&newex); |
b939e376 | 2876 | allocated = ext4_ext_get_actual_len(&newex); |
56055d3a | 2877 | outnew: |
61628a3f MC |
2878 | if (extend_disksize) { |
2879 | disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits; | |
2880 | if (disksize > i_size_read(inode)) | |
2881 | disksize = i_size_read(inode); | |
2882 | if (disksize > EXT4_I(inode)->i_disksize) | |
2883 | EXT4_I(inode)->i_disksize = disksize; | |
2884 | } | |
a379cd1d | 2885 | |
953e622b | 2886 | set_buffer_new(bh_result); |
a86c6181 | 2887 | |
a2df2a63 AA |
2888 | /* Cache only when it is _not_ an uninitialized extent */ |
2889 | if (create != EXT4_CREATE_UNINITIALIZED_EXT) | |
2890 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, | |
2891 | EXT4_EXT_CACHE_EXTENT); | |
a86c6181 AT |
2892 | out: |
2893 | if (allocated > max_blocks) | |
2894 | allocated = max_blocks; | |
2895 | ext4_ext_show_leaf(inode, path); | |
953e622b | 2896 | set_buffer_mapped(bh_result); |
a86c6181 AT |
2897 | bh_result->b_bdev = inode->i_sb->s_bdev; |
2898 | bh_result->b_blocknr = newblock; | |
2899 | out2: | |
2900 | if (path) { | |
2901 | ext4_ext_drop_refs(path); | |
2902 | kfree(path); | |
2903 | } | |
a86c6181 AT |
2904 | return err ? err : allocated; |
2905 | } | |
2906 | ||
cf108bca | 2907 | void ext4_ext_truncate(struct inode *inode) |
a86c6181 AT |
2908 | { |
2909 | struct address_space *mapping = inode->i_mapping; | |
2910 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 2911 | ext4_lblk_t last_block; |
a86c6181 AT |
2912 | handle_t *handle; |
2913 | int err = 0; | |
2914 | ||
2915 | /* | |
2916 | * probably first extent we're gonna free will be last in block | |
2917 | */ | |
f3bd1f3f | 2918 | err = ext4_writepage_trans_blocks(inode); |
a86c6181 | 2919 | handle = ext4_journal_start(inode, err); |
cf108bca | 2920 | if (IS_ERR(handle)) |
a86c6181 | 2921 | return; |
a86c6181 | 2922 | |
cf108bca JK |
2923 | if (inode->i_size & (sb->s_blocksize - 1)) |
2924 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
a86c6181 | 2925 | |
9ddfc3dc JK |
2926 | if (ext4_orphan_add(handle, inode)) |
2927 | goto out_stop; | |
2928 | ||
0e855ac8 | 2929 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
2930 | ext4_ext_invalidate_cache(inode); |
2931 | ||
c2ea3fde | 2932 | ext4_discard_preallocations(inode); |
c9de560d | 2933 | |
a86c6181 | 2934 | /* |
d0d856e8 RD |
2935 | * TODO: optimization is possible here. |
2936 | * Probably we need not scan at all, | |
2937 | * because page truncation is enough. | |
a86c6181 | 2938 | */ |
a86c6181 AT |
2939 | |
2940 | /* we have to know where to truncate from in crash case */ | |
2941 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2942 | ext4_mark_inode_dirty(handle, inode); | |
2943 | ||
2944 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
2945 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
2946 | err = ext4_ext_remove_space(inode, last_block); | |
2947 | ||
2948 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
2949 | * transaction synchronous. |
2950 | */ | |
a86c6181 | 2951 | if (IS_SYNC(inode)) |
0390131b | 2952 | ext4_handle_sync(handle); |
a86c6181 AT |
2953 | |
2954 | out_stop: | |
9ddfc3dc | 2955 | up_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 | 2956 | /* |
d0d856e8 | 2957 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
2958 | * then we need to clear up the orphan record which we created above. |
2959 | * However, if this was a real unlink then we were called by | |
2960 | * ext4_delete_inode(), and we allow that function to clean up the | |
2961 | * orphan info for us. | |
2962 | */ | |
2963 | if (inode->i_nlink) | |
2964 | ext4_orphan_del(handle, inode); | |
2965 | ||
ef737728 SR |
2966 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
2967 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
2968 | ext4_journal_stop(handle); |
2969 | } | |
2970 | ||
fd28784a AK |
2971 | static void ext4_falloc_update_inode(struct inode *inode, |
2972 | int mode, loff_t new_size, int update_ctime) | |
2973 | { | |
2974 | struct timespec now; | |
2975 | ||
2976 | if (update_ctime) { | |
2977 | now = current_fs_time(inode->i_sb); | |
2978 | if (!timespec_equal(&inode->i_ctime, &now)) | |
2979 | inode->i_ctime = now; | |
2980 | } | |
2981 | /* | |
2982 | * Update only when preallocation was requested beyond | |
2983 | * the file size. | |
2984 | */ | |
cf17fea6 AK |
2985 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
2986 | if (new_size > i_size_read(inode)) | |
2987 | i_size_write(inode, new_size); | |
2988 | if (new_size > EXT4_I(inode)->i_disksize) | |
2989 | ext4_update_i_disksize(inode, new_size); | |
fd28784a AK |
2990 | } |
2991 | ||
2992 | } | |
2993 | ||
a2df2a63 AA |
2994 | /* |
2995 | * preallocate space for a file. This implements ext4's fallocate inode | |
2996 | * operation, which gets called from sys_fallocate system call. | |
2997 | * For block-mapped files, posix_fallocate should fall back to the method | |
2998 | * of writing zeroes to the required new blocks (the same behavior which is | |
2999 | * expected for file systems which do not support fallocate() system call). | |
3000 | */ | |
3001 | long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) | |
3002 | { | |
3003 | handle_t *handle; | |
725d26d3 | 3004 | ext4_lblk_t block; |
fd28784a | 3005 | loff_t new_size; |
498e5f24 | 3006 | unsigned int max_blocks; |
a2df2a63 AA |
3007 | int ret = 0; |
3008 | int ret2 = 0; | |
3009 | int retries = 0; | |
3010 | struct buffer_head map_bh; | |
3011 | unsigned int credits, blkbits = inode->i_blkbits; | |
3012 | ||
3013 | /* | |
3014 | * currently supporting (pre)allocate mode for extent-based | |
3015 | * files _only_ | |
3016 | */ | |
3017 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
3018 | return -EOPNOTSUPP; | |
3019 | ||
3020 | /* preallocation to directories is currently not supported */ | |
3021 | if (S_ISDIR(inode->i_mode)) | |
3022 | return -ENODEV; | |
3023 | ||
3024 | block = offset >> blkbits; | |
fd28784a AK |
3025 | /* |
3026 | * We can't just convert len to max_blocks because | |
3027 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
3028 | */ | |
a2df2a63 | 3029 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
fd28784a | 3030 | - block; |
a2df2a63 | 3031 | /* |
f3bd1f3f | 3032 | * credits to insert 1 extent into extent tree |
a2df2a63 | 3033 | */ |
f3bd1f3f | 3034 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 3035 | mutex_lock(&inode->i_mutex); |
a2df2a63 AA |
3036 | retry: |
3037 | while (ret >= 0 && ret < max_blocks) { | |
3038 | block = block + ret; | |
3039 | max_blocks = max_blocks - ret; | |
3040 | handle = ext4_journal_start(inode, credits); | |
3041 | if (IS_ERR(handle)) { | |
3042 | ret = PTR_ERR(handle); | |
3043 | break; | |
3044 | } | |
55bd725a | 3045 | ret = ext4_get_blocks_wrap(handle, inode, block, |
a2df2a63 | 3046 | max_blocks, &map_bh, |
d2a17637 | 3047 | EXT4_CREATE_UNINITIALIZED_EXT, 0, 0); |
221879c9 | 3048 | if (ret <= 0) { |
2c98615d AK |
3049 | #ifdef EXT4FS_DEBUG |
3050 | WARN_ON(ret <= 0); | |
3051 | printk(KERN_ERR "%s: ext4_ext_get_blocks " | |
3052 | "returned error inode#%lu, block=%u, " | |
9fd9784c | 3053 | "max_blocks=%u", __func__, |
221879c9 | 3054 | inode->i_ino, block, max_blocks); |
2c98615d | 3055 | #endif |
a2df2a63 AA |
3056 | ext4_mark_inode_dirty(handle, inode); |
3057 | ret2 = ext4_journal_stop(handle); | |
3058 | break; | |
3059 | } | |
fd28784a AK |
3060 | if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
3061 | blkbits) >> blkbits)) | |
3062 | new_size = offset + len; | |
3063 | else | |
3064 | new_size = (block + ret) << blkbits; | |
a2df2a63 | 3065 | |
fd28784a AK |
3066 | ext4_falloc_update_inode(inode, mode, new_size, |
3067 | buffer_new(&map_bh)); | |
a2df2a63 AA |
3068 | ext4_mark_inode_dirty(handle, inode); |
3069 | ret2 = ext4_journal_stop(handle); | |
3070 | if (ret2) | |
3071 | break; | |
3072 | } | |
fd28784a AK |
3073 | if (ret == -ENOSPC && |
3074 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
3075 | ret = 0; | |
a2df2a63 | 3076 | goto retry; |
a2df2a63 | 3077 | } |
55bd725a | 3078 | mutex_unlock(&inode->i_mutex); |
a2df2a63 AA |
3079 | return ret > 0 ? ret2 : ret; |
3080 | } | |
6873fa0d ES |
3081 | |
3082 | /* | |
3083 | * Callback function called for each extent to gather FIEMAP information. | |
3084 | */ | |
3a06d778 | 3085 | static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path, |
6873fa0d ES |
3086 | struct ext4_ext_cache *newex, struct ext4_extent *ex, |
3087 | void *data) | |
3088 | { | |
3089 | struct fiemap_extent_info *fieinfo = data; | |
3090 | unsigned long blksize_bits = inode->i_sb->s_blocksize_bits; | |
3091 | __u64 logical; | |
3092 | __u64 physical; | |
3093 | __u64 length; | |
3094 | __u32 flags = 0; | |
3095 | int error; | |
3096 | ||
3097 | logical = (__u64)newex->ec_block << blksize_bits; | |
3098 | ||
3099 | if (newex->ec_type == EXT4_EXT_CACHE_GAP) { | |
3100 | pgoff_t offset; | |
3101 | struct page *page; | |
3102 | struct buffer_head *bh = NULL; | |
3103 | ||
3104 | offset = logical >> PAGE_SHIFT; | |
3105 | page = find_get_page(inode->i_mapping, offset); | |
3106 | if (!page || !page_has_buffers(page)) | |
3107 | return EXT_CONTINUE; | |
3108 | ||
3109 | bh = page_buffers(page); | |
3110 | ||
3111 | if (!bh) | |
3112 | return EXT_CONTINUE; | |
3113 | ||
3114 | if (buffer_delay(bh)) { | |
3115 | flags |= FIEMAP_EXTENT_DELALLOC; | |
3116 | page_cache_release(page); | |
3117 | } else { | |
3118 | page_cache_release(page); | |
3119 | return EXT_CONTINUE; | |
3120 | } | |
3121 | } | |
3122 | ||
3123 | physical = (__u64)newex->ec_start << blksize_bits; | |
3124 | length = (__u64)newex->ec_len << blksize_bits; | |
3125 | ||
3126 | if (ex && ext4_ext_is_uninitialized(ex)) | |
3127 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
3128 | ||
3129 | /* | |
3130 | * If this extent reaches EXT_MAX_BLOCK, it must be last. | |
3131 | * | |
3132 | * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK, | |
3133 | * this also indicates no more allocated blocks. | |
3134 | * | |
3135 | * XXX this might miss a single-block extent at EXT_MAX_BLOCK | |
3136 | */ | |
3137 | if (logical + length - 1 == EXT_MAX_BLOCK || | |
3138 | ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK) | |
3139 | flags |= FIEMAP_EXTENT_LAST; | |
3140 | ||
3141 | error = fiemap_fill_next_extent(fieinfo, logical, physical, | |
3142 | length, flags); | |
3143 | if (error < 0) | |
3144 | return error; | |
3145 | if (error == 1) | |
3146 | return EXT_BREAK; | |
3147 | ||
3148 | return EXT_CONTINUE; | |
3149 | } | |
3150 | ||
3151 | /* fiemap flags we can handle specified here */ | |
3152 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
3153 | ||
3a06d778 AK |
3154 | static int ext4_xattr_fiemap(struct inode *inode, |
3155 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
3156 | { |
3157 | __u64 physical = 0; | |
3158 | __u64 length; | |
3159 | __u32 flags = FIEMAP_EXTENT_LAST; | |
3160 | int blockbits = inode->i_sb->s_blocksize_bits; | |
3161 | int error = 0; | |
3162 | ||
3163 | /* in-inode? */ | |
3164 | if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) { | |
3165 | struct ext4_iloc iloc; | |
3166 | int offset; /* offset of xattr in inode */ | |
3167 | ||
3168 | error = ext4_get_inode_loc(inode, &iloc); | |
3169 | if (error) | |
3170 | return error; | |
3171 | physical = iloc.bh->b_blocknr << blockbits; | |
3172 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
3173 | EXT4_I(inode)->i_extra_isize; | |
3174 | physical += offset; | |
3175 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
3176 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
3177 | } else { /* external block */ | |
3178 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
3179 | length = inode->i_sb->s_blocksize; | |
3180 | } | |
3181 | ||
3182 | if (physical) | |
3183 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
3184 | length, flags); | |
3185 | return (error < 0 ? error : 0); | |
3186 | } | |
3187 | ||
3188 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
3189 | __u64 start, __u64 len) | |
3190 | { | |
3191 | ext4_lblk_t start_blk; | |
3192 | ext4_lblk_t len_blks; | |
3193 | int error = 0; | |
3194 | ||
3195 | /* fallback to generic here if not in extents fmt */ | |
3196 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
3197 | return generic_block_fiemap(inode, fieinfo, start, len, | |
3198 | ext4_get_block); | |
3199 | ||
3200 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
3201 | return -EBADR; | |
3202 | ||
3203 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
3204 | error = ext4_xattr_fiemap(inode, fieinfo); | |
3205 | } else { | |
3206 | start_blk = start >> inode->i_sb->s_blocksize_bits; | |
3207 | len_blks = len >> inode->i_sb->s_blocksize_bits; | |
3208 | ||
3209 | /* | |
3210 | * Walk the extent tree gathering extent information. | |
3211 | * ext4_ext_fiemap_cb will push extents back to user. | |
3212 | */ | |
3213 | down_write(&EXT4_I(inode)->i_data_sem); | |
3214 | error = ext4_ext_walk_space(inode, start_blk, len_blks, | |
3215 | ext4_ext_fiemap_cb, fieinfo); | |
3216 | up_write(&EXT4_I(inode)->i_data_sem); | |
3217 | } | |
3218 | ||
3219 | return error; | |
3220 | } | |
3221 |