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> |
3dcf5451 CH |
43 | #include "ext4_jbd2.h" |
44 | #include "ext4_extents.h" | |
a86c6181 AT |
45 | |
46 | ||
d0d856e8 RD |
47 | /* |
48 | * ext_pblock: | |
49 | * combine low and high parts of physical block number into ext4_fsblk_t | |
50 | */ | |
09b88252 | 51 | static ext4_fsblk_t ext_pblock(struct ext4_extent *ex) |
f65e6fba AT |
52 | { |
53 | ext4_fsblk_t block; | |
54 | ||
b377611d | 55 | block = le32_to_cpu(ex->ee_start_lo); |
9b8f1f01 | 56 | block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1; |
f65e6fba AT |
57 | return block; |
58 | } | |
59 | ||
d0d856e8 RD |
60 | /* |
61 | * idx_pblock: | |
62 | * combine low and high parts of a leaf physical block number into ext4_fsblk_t | |
63 | */ | |
c14c6fd5 | 64 | ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix) |
f65e6fba AT |
65 | { |
66 | ext4_fsblk_t block; | |
67 | ||
d8dd0b45 | 68 | block = le32_to_cpu(ix->ei_leaf_lo); |
9b8f1f01 | 69 | block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1; |
f65e6fba AT |
70 | return block; |
71 | } | |
72 | ||
d0d856e8 RD |
73 | /* |
74 | * ext4_ext_store_pblock: | |
75 | * stores a large physical block number into an extent struct, | |
76 | * breaking it into parts | |
77 | */ | |
c14c6fd5 | 78 | void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb) |
f65e6fba | 79 | { |
b377611d | 80 | ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
9b8f1f01 | 81 | ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
f65e6fba AT |
82 | } |
83 | ||
d0d856e8 RD |
84 | /* |
85 | * ext4_idx_store_pblock: | |
86 | * stores a large physical block number into an index struct, | |
87 | * breaking it into parts | |
88 | */ | |
09b88252 | 89 | static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb) |
f65e6fba | 90 | { |
d8dd0b45 | 91 | ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff)); |
9b8f1f01 | 92 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
f65e6fba AT |
93 | } |
94 | ||
a86c6181 AT |
95 | static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed) |
96 | { | |
97 | int err; | |
98 | ||
99 | if (handle->h_buffer_credits > needed) | |
100 | return handle; | |
101 | if (!ext4_journal_extend(handle, needed)) | |
102 | return handle; | |
103 | err = ext4_journal_restart(handle, needed); | |
104 | ||
105 | return handle; | |
106 | } | |
107 | ||
108 | /* | |
109 | * could return: | |
110 | * - EROFS | |
111 | * - ENOMEM | |
112 | */ | |
113 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
114 | struct ext4_ext_path *path) | |
115 | { | |
116 | if (path->p_bh) { | |
117 | /* path points to block */ | |
118 | return ext4_journal_get_write_access(handle, path->p_bh); | |
119 | } | |
120 | /* path points to leaf/index in inode body */ | |
121 | /* we use in-core data, no need to protect them */ | |
122 | return 0; | |
123 | } | |
124 | ||
125 | /* | |
126 | * could return: | |
127 | * - EROFS | |
128 | * - ENOMEM | |
129 | * - EIO | |
130 | */ | |
131 | static int ext4_ext_dirty(handle_t *handle, struct inode *inode, | |
132 | struct ext4_ext_path *path) | |
133 | { | |
134 | int err; | |
135 | if (path->p_bh) { | |
136 | /* path points to block */ | |
137 | err = ext4_journal_dirty_metadata(handle, path->p_bh); | |
138 | } else { | |
139 | /* path points to leaf/index in inode body */ | |
140 | err = ext4_mark_inode_dirty(handle, inode); | |
141 | } | |
142 | return err; | |
143 | } | |
144 | ||
f65e6fba | 145 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 146 | struct ext4_ext_path *path, |
725d26d3 | 147 | ext4_lblk_t block) |
a86c6181 AT |
148 | { |
149 | struct ext4_inode_info *ei = EXT4_I(inode); | |
f65e6fba | 150 | ext4_fsblk_t bg_start; |
74d3487f | 151 | ext4_fsblk_t last_block; |
f65e6fba | 152 | ext4_grpblk_t colour; |
a86c6181 AT |
153 | int depth; |
154 | ||
155 | if (path) { | |
156 | struct ext4_extent *ex; | |
157 | depth = path->p_depth; | |
158 | ||
159 | /* try to predict block placement */ | |
7e028976 AM |
160 | ex = path[depth].p_ext; |
161 | if (ex) | |
f65e6fba | 162 | return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block)); |
a86c6181 | 163 | |
d0d856e8 RD |
164 | /* it looks like index is empty; |
165 | * try to find starting block from index itself */ | |
a86c6181 AT |
166 | if (path[depth].p_bh) |
167 | return path[depth].p_bh->b_blocknr; | |
168 | } | |
169 | ||
170 | /* OK. use inode's group */ | |
171 | bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) + | |
172 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block); | |
74d3487f VC |
173 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
174 | ||
175 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) | |
176 | colour = (current->pid % 16) * | |
a86c6181 | 177 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
178 | else |
179 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
a86c6181 AT |
180 | return bg_start + colour + block; |
181 | } | |
182 | ||
f65e6fba | 183 | static ext4_fsblk_t |
a86c6181 AT |
184 | ext4_ext_new_block(handle_t *handle, struct inode *inode, |
185 | struct ext4_ext_path *path, | |
186 | struct ext4_extent *ex, int *err) | |
187 | { | |
f65e6fba | 188 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
189 | |
190 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
191 | newblock = ext4_new_block(handle, inode, goal, err); | |
192 | return newblock; | |
193 | } | |
194 | ||
09b88252 | 195 | static int ext4_ext_space_block(struct inode *inode) |
a86c6181 AT |
196 | { |
197 | int size; | |
198 | ||
199 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
200 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 201 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
202 | if (size > 6) |
203 | size = 6; | |
204 | #endif | |
205 | return size; | |
206 | } | |
207 | ||
09b88252 | 208 | static int ext4_ext_space_block_idx(struct inode *inode) |
a86c6181 AT |
209 | { |
210 | int size; | |
211 | ||
212 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
213 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 214 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
215 | if (size > 5) |
216 | size = 5; | |
217 | #endif | |
218 | return size; | |
219 | } | |
220 | ||
09b88252 | 221 | static int ext4_ext_space_root(struct inode *inode) |
a86c6181 AT |
222 | { |
223 | int size; | |
224 | ||
225 | size = sizeof(EXT4_I(inode)->i_data); | |
226 | size -= sizeof(struct ext4_extent_header); | |
227 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 228 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
229 | if (size > 3) |
230 | size = 3; | |
231 | #endif | |
232 | return size; | |
233 | } | |
234 | ||
09b88252 | 235 | static int ext4_ext_space_root_idx(struct inode *inode) |
a86c6181 AT |
236 | { |
237 | int size; | |
238 | ||
239 | size = sizeof(EXT4_I(inode)->i_data); | |
240 | size -= sizeof(struct ext4_extent_header); | |
241 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 242 | #ifdef AGGRESSIVE_TEST |
a86c6181 AT |
243 | if (size > 4) |
244 | size = 4; | |
245 | #endif | |
246 | return size; | |
247 | } | |
248 | ||
c29c0ae7 AT |
249 | static int |
250 | ext4_ext_max_entries(struct inode *inode, int depth) | |
251 | { | |
252 | int max; | |
253 | ||
254 | if (depth == ext_depth(inode)) { | |
255 | if (depth == 0) | |
256 | max = ext4_ext_space_root(inode); | |
257 | else | |
258 | max = ext4_ext_space_root_idx(inode); | |
259 | } else { | |
260 | if (depth == 0) | |
261 | max = ext4_ext_space_block(inode); | |
262 | else | |
263 | max = ext4_ext_space_block_idx(inode); | |
264 | } | |
265 | ||
266 | return max; | |
267 | } | |
268 | ||
269 | static int __ext4_ext_check_header(const char *function, struct inode *inode, | |
270 | struct ext4_extent_header *eh, | |
271 | int depth) | |
272 | { | |
273 | const char *error_msg; | |
274 | int max = 0; | |
275 | ||
276 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
277 | error_msg = "invalid magic"; | |
278 | goto corrupted; | |
279 | } | |
280 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
281 | error_msg = "unexpected eh_depth"; | |
282 | goto corrupted; | |
283 | } | |
284 | if (unlikely(eh->eh_max == 0)) { | |
285 | error_msg = "invalid eh_max"; | |
286 | goto corrupted; | |
287 | } | |
288 | max = ext4_ext_max_entries(inode, depth); | |
289 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
290 | error_msg = "too large eh_max"; | |
291 | goto corrupted; | |
292 | } | |
293 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
294 | error_msg = "invalid eh_entries"; | |
295 | goto corrupted; | |
296 | } | |
297 | return 0; | |
298 | ||
299 | corrupted: | |
300 | ext4_error(inode->i_sb, function, | |
301 | "bad header in inode #%lu: %s - magic %x, " | |
302 | "entries %u, max %u(%u), depth %u(%u)", | |
303 | inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), | |
304 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
305 | max, le16_to_cpu(eh->eh_depth), depth); | |
306 | ||
307 | return -EIO; | |
308 | } | |
309 | ||
310 | #define ext4_ext_check_header(inode, eh, depth) \ | |
46e665e9 | 311 | __ext4_ext_check_header(__func__, inode, eh, depth) |
c29c0ae7 | 312 | |
a86c6181 AT |
313 | #ifdef EXT_DEBUG |
314 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
315 | { | |
316 | int k, l = path->p_depth; | |
317 | ||
318 | ext_debug("path:"); | |
319 | for (k = 0; k <= l; k++, path++) { | |
320 | if (path->p_idx) { | |
2ae02107 | 321 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
f65e6fba | 322 | idx_pblock(path->p_idx)); |
a86c6181 | 323 | } else if (path->p_ext) { |
2ae02107 | 324 | ext_debug(" %d:%d:%llu ", |
a86c6181 | 325 | le32_to_cpu(path->p_ext->ee_block), |
a2df2a63 | 326 | ext4_ext_get_actual_len(path->p_ext), |
f65e6fba | 327 | ext_pblock(path->p_ext)); |
a86c6181 AT |
328 | } else |
329 | ext_debug(" []"); | |
330 | } | |
331 | ext_debug("\n"); | |
332 | } | |
333 | ||
334 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
335 | { | |
336 | int depth = ext_depth(inode); | |
337 | struct ext4_extent_header *eh; | |
338 | struct ext4_extent *ex; | |
339 | int i; | |
340 | ||
341 | if (!path) | |
342 | return; | |
343 | ||
344 | eh = path[depth].p_hdr; | |
345 | ex = EXT_FIRST_EXTENT(eh); | |
346 | ||
347 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { | |
2ae02107 | 348 | ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block), |
a2df2a63 | 349 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
a86c6181 AT |
350 | } |
351 | ext_debug("\n"); | |
352 | } | |
353 | #else | |
354 | #define ext4_ext_show_path(inode,path) | |
355 | #define ext4_ext_show_leaf(inode,path) | |
356 | #endif | |
357 | ||
b35905c1 | 358 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
359 | { |
360 | int depth = path->p_depth; | |
361 | int i; | |
362 | ||
363 | for (i = 0; i <= depth; i++, path++) | |
364 | if (path->p_bh) { | |
365 | brelse(path->p_bh); | |
366 | path->p_bh = NULL; | |
367 | } | |
368 | } | |
369 | ||
370 | /* | |
d0d856e8 RD |
371 | * ext4_ext_binsearch_idx: |
372 | * binary search for the closest index of the given block | |
c29c0ae7 | 373 | * the header must be checked before calling this |
a86c6181 AT |
374 | */ |
375 | static void | |
725d26d3 AK |
376 | ext4_ext_binsearch_idx(struct inode *inode, |
377 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
378 | { |
379 | struct ext4_extent_header *eh = path->p_hdr; | |
380 | struct ext4_extent_idx *r, *l, *m; | |
381 | ||
a86c6181 | 382 | |
bba90743 | 383 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
384 | |
385 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 386 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
387 | while (l <= r) { |
388 | m = l + (r - l) / 2; | |
389 | if (block < le32_to_cpu(m->ei_block)) | |
390 | r = m - 1; | |
391 | else | |
392 | l = m + 1; | |
26d535ed DM |
393 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
394 | m, le32_to_cpu(m->ei_block), | |
395 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
396 | } |
397 | ||
398 | path->p_idx = l - 1; | |
f65e6fba | 399 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
26d535ed | 400 | idx_pblock(path->p_idx)); |
a86c6181 AT |
401 | |
402 | #ifdef CHECK_BINSEARCH | |
403 | { | |
404 | struct ext4_extent_idx *chix, *ix; | |
405 | int k; | |
406 | ||
407 | chix = ix = EXT_FIRST_INDEX(eh); | |
408 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
409 | if (k != 0 && | |
410 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
411 | printk("k=%d, ix=0x%p, first=0x%p\n", k, | |
412 | ix, EXT_FIRST_INDEX(eh)); | |
413 | printk("%u <= %u\n", | |
414 | le32_to_cpu(ix->ei_block), | |
415 | le32_to_cpu(ix[-1].ei_block)); | |
416 | } | |
417 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 418 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
419 | if (block < le32_to_cpu(ix->ei_block)) |
420 | break; | |
421 | chix = ix; | |
422 | } | |
423 | BUG_ON(chix != path->p_idx); | |
424 | } | |
425 | #endif | |
426 | ||
427 | } | |
428 | ||
429 | /* | |
d0d856e8 RD |
430 | * ext4_ext_binsearch: |
431 | * binary search for closest extent of the given block | |
c29c0ae7 | 432 | * the header must be checked before calling this |
a86c6181 AT |
433 | */ |
434 | static void | |
725d26d3 AK |
435 | ext4_ext_binsearch(struct inode *inode, |
436 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
437 | { |
438 | struct ext4_extent_header *eh = path->p_hdr; | |
439 | struct ext4_extent *r, *l, *m; | |
440 | ||
a86c6181 AT |
441 | if (eh->eh_entries == 0) { |
442 | /* | |
d0d856e8 RD |
443 | * this leaf is empty: |
444 | * we get such a leaf in split/add case | |
a86c6181 AT |
445 | */ |
446 | return; | |
447 | } | |
448 | ||
bba90743 | 449 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
450 | |
451 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 452 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
453 | |
454 | while (l <= r) { | |
455 | m = l + (r - l) / 2; | |
456 | if (block < le32_to_cpu(m->ee_block)) | |
457 | r = m - 1; | |
458 | else | |
459 | l = m + 1; | |
26d535ed DM |
460 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
461 | m, le32_to_cpu(m->ee_block), | |
462 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
463 | } |
464 | ||
465 | path->p_ext = l - 1; | |
2ae02107 | 466 | ext_debug(" -> %d:%llu:%d ", |
8c55e204 DK |
467 | le32_to_cpu(path->p_ext->ee_block), |
468 | ext_pblock(path->p_ext), | |
a2df2a63 | 469 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
470 | |
471 | #ifdef CHECK_BINSEARCH | |
472 | { | |
473 | struct ext4_extent *chex, *ex; | |
474 | int k; | |
475 | ||
476 | chex = ex = EXT_FIRST_EXTENT(eh); | |
477 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
478 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 479 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
480 | if (block < le32_to_cpu(ex->ee_block)) |
481 | break; | |
482 | chex = ex; | |
483 | } | |
484 | BUG_ON(chex != path->p_ext); | |
485 | } | |
486 | #endif | |
487 | ||
488 | } | |
489 | ||
490 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
491 | { | |
492 | struct ext4_extent_header *eh; | |
493 | ||
494 | eh = ext_inode_hdr(inode); | |
495 | eh->eh_depth = 0; | |
496 | eh->eh_entries = 0; | |
497 | eh->eh_magic = EXT4_EXT_MAGIC; | |
498 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode)); | |
499 | ext4_mark_inode_dirty(handle, inode); | |
500 | ext4_ext_invalidate_cache(inode); | |
501 | return 0; | |
502 | } | |
503 | ||
504 | struct ext4_ext_path * | |
725d26d3 AK |
505 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
506 | struct ext4_ext_path *path) | |
a86c6181 AT |
507 | { |
508 | struct ext4_extent_header *eh; | |
509 | struct buffer_head *bh; | |
510 | short int depth, i, ppos = 0, alloc = 0; | |
511 | ||
512 | eh = ext_inode_hdr(inode); | |
c29c0ae7 AT |
513 | depth = ext_depth(inode); |
514 | if (ext4_ext_check_header(inode, eh, depth)) | |
a86c6181 AT |
515 | return ERR_PTR(-EIO); |
516 | ||
a86c6181 AT |
517 | |
518 | /* account possible depth increase */ | |
519 | if (!path) { | |
5d4958f9 | 520 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
521 | GFP_NOFS); |
522 | if (!path) | |
523 | return ERR_PTR(-ENOMEM); | |
524 | alloc = 1; | |
525 | } | |
a86c6181 AT |
526 | path[0].p_hdr = eh; |
527 | ||
c29c0ae7 | 528 | i = depth; |
a86c6181 AT |
529 | /* walk through the tree */ |
530 | while (i) { | |
531 | ext_debug("depth %d: num %d, max %d\n", | |
532 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 533 | |
a86c6181 | 534 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
f65e6fba | 535 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
536 | path[ppos].p_depth = i; |
537 | path[ppos].p_ext = NULL; | |
538 | ||
539 | bh = sb_bread(inode->i_sb, path[ppos].p_block); | |
540 | if (!bh) | |
541 | goto err; | |
542 | ||
543 | eh = ext_block_hdr(bh); | |
544 | ppos++; | |
545 | BUG_ON(ppos > depth); | |
546 | path[ppos].p_bh = bh; | |
547 | path[ppos].p_hdr = eh; | |
548 | i--; | |
549 | ||
c29c0ae7 | 550 | if (ext4_ext_check_header(inode, eh, i)) |
a86c6181 AT |
551 | goto err; |
552 | } | |
553 | ||
554 | path[ppos].p_depth = i; | |
555 | path[ppos].p_hdr = eh; | |
556 | path[ppos].p_ext = NULL; | |
557 | path[ppos].p_idx = NULL; | |
558 | ||
a86c6181 AT |
559 | /* find extent */ |
560 | ext4_ext_binsearch(inode, path + ppos, block); | |
561 | ||
562 | ext4_ext_show_path(inode, path); | |
563 | ||
564 | return path; | |
565 | ||
566 | err: | |
567 | ext4_ext_drop_refs(path); | |
568 | if (alloc) | |
569 | kfree(path); | |
570 | return ERR_PTR(-EIO); | |
571 | } | |
572 | ||
573 | /* | |
d0d856e8 RD |
574 | * ext4_ext_insert_index: |
575 | * insert new index [@logical;@ptr] into the block at @curp; | |
576 | * check where to insert: before @curp or after @curp | |
a86c6181 AT |
577 | */ |
578 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |
579 | struct ext4_ext_path *curp, | |
f65e6fba | 580 | int logical, ext4_fsblk_t ptr) |
a86c6181 AT |
581 | { |
582 | struct ext4_extent_idx *ix; | |
583 | int len, err; | |
584 | ||
7e028976 AM |
585 | err = ext4_ext_get_access(handle, inode, curp); |
586 | if (err) | |
a86c6181 AT |
587 | return err; |
588 | ||
589 | BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block)); | |
590 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; | |
591 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
592 | /* insert after */ | |
593 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
594 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
595 | len = len < 0 ? 0 : len; | |
26d535ed | 596 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
597 | "move %d from 0x%p to 0x%p\n", |
598 | logical, ptr, len, | |
599 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
600 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
601 | } | |
602 | ix = curp->p_idx + 1; | |
603 | } else { | |
604 | /* insert before */ | |
605 | len = len * sizeof(struct ext4_extent_idx); | |
606 | len = len < 0 ? 0 : len; | |
26d535ed | 607 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
608 | "move %d from 0x%p to 0x%p\n", |
609 | logical, ptr, len, | |
610 | curp->p_idx, (curp->p_idx + 1)); | |
611 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
612 | ix = curp->p_idx; | |
613 | } | |
614 | ||
615 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 616 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 617 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 AT |
618 | |
619 | BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries) | |
8c55e204 | 620 | > le16_to_cpu(curp->p_hdr->eh_max)); |
a86c6181 AT |
621 | BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr)); |
622 | ||
623 | err = ext4_ext_dirty(handle, inode, curp); | |
624 | ext4_std_error(inode->i_sb, err); | |
625 | ||
626 | return err; | |
627 | } | |
628 | ||
629 | /* | |
d0d856e8 RD |
630 | * ext4_ext_split: |
631 | * inserts new subtree into the path, using free index entry | |
632 | * at depth @at: | |
633 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
634 | * - makes decision where to split | |
635 | * - moves remaining extents and index entries (right to the split point) | |
636 | * into the newly allocated blocks | |
637 | * - initializes subtree | |
a86c6181 AT |
638 | */ |
639 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
640 | struct ext4_ext_path *path, | |
641 | struct ext4_extent *newext, int at) | |
642 | { | |
643 | struct buffer_head *bh = NULL; | |
644 | int depth = ext_depth(inode); | |
645 | struct ext4_extent_header *neh; | |
646 | struct ext4_extent_idx *fidx; | |
647 | struct ext4_extent *ex; | |
648 | int i = at, k, m, a; | |
f65e6fba | 649 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 650 | __le32 border; |
f65e6fba | 651 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
652 | int err = 0; |
653 | ||
654 | /* make decision: where to split? */ | |
d0d856e8 | 655 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 656 | |
d0d856e8 | 657 | /* if current leaf will be split, then we should use |
a86c6181 AT |
658 | * border from split point */ |
659 | BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr)); | |
660 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
661 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 662 | ext_debug("leaf will be split." |
a86c6181 | 663 | " next leaf starts at %d\n", |
8c55e204 | 664 | le32_to_cpu(border)); |
a86c6181 AT |
665 | } else { |
666 | border = newext->ee_block; | |
667 | ext_debug("leaf will be added." | |
668 | " next leaf starts at %d\n", | |
8c55e204 | 669 | le32_to_cpu(border)); |
a86c6181 AT |
670 | } |
671 | ||
672 | /* | |
d0d856e8 RD |
673 | * If error occurs, then we break processing |
674 | * and mark filesystem read-only. index won't | |
a86c6181 | 675 | * be inserted and tree will be in consistent |
d0d856e8 | 676 | * state. Next mount will repair buffers too. |
a86c6181 AT |
677 | */ |
678 | ||
679 | /* | |
d0d856e8 RD |
680 | * Get array to track all allocated blocks. |
681 | * We need this to handle errors and free blocks | |
682 | * upon them. | |
a86c6181 | 683 | */ |
5d4958f9 | 684 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
685 | if (!ablocks) |
686 | return -ENOMEM; | |
a86c6181 AT |
687 | |
688 | /* allocate all needed blocks */ | |
689 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
690 | for (a = 0; a < depth - at; a++) { | |
691 | newblock = ext4_ext_new_block(handle, inode, path, newext, &err); | |
692 | if (newblock == 0) | |
693 | goto cleanup; | |
694 | ablocks[a] = newblock; | |
695 | } | |
696 | ||
697 | /* initialize new leaf */ | |
698 | newblock = ablocks[--a]; | |
699 | BUG_ON(newblock == 0); | |
700 | bh = sb_getblk(inode->i_sb, newblock); | |
701 | if (!bh) { | |
702 | err = -EIO; | |
703 | goto cleanup; | |
704 | } | |
705 | lock_buffer(bh); | |
706 | ||
7e028976 AM |
707 | err = ext4_journal_get_create_access(handle, bh); |
708 | if (err) | |
a86c6181 AT |
709 | goto cleanup; |
710 | ||
711 | neh = ext_block_hdr(bh); | |
712 | neh->eh_entries = 0; | |
713 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
714 | neh->eh_magic = EXT4_EXT_MAGIC; | |
715 | neh->eh_depth = 0; | |
716 | ex = EXT_FIRST_EXTENT(neh); | |
717 | ||
d0d856e8 | 718 | /* move remainder of path[depth] to the new leaf */ |
a86c6181 AT |
719 | BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max); |
720 | /* start copy from next extent */ | |
721 | /* TODO: we could do it by single memmove */ | |
722 | m = 0; | |
723 | path[depth].p_ext++; | |
724 | while (path[depth].p_ext <= | |
725 | EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
2ae02107 | 726 | ext_debug("move %d:%llu:%d in new leaf %llu\n", |
8c55e204 DK |
727 | le32_to_cpu(path[depth].p_ext->ee_block), |
728 | ext_pblock(path[depth].p_ext), | |
a2df2a63 | 729 | ext4_ext_get_actual_len(path[depth].p_ext), |
a86c6181 AT |
730 | newblock); |
731 | /*memmove(ex++, path[depth].p_ext++, | |
732 | sizeof(struct ext4_extent)); | |
733 | neh->eh_entries++;*/ | |
734 | path[depth].p_ext++; | |
735 | m++; | |
736 | } | |
737 | if (m) { | |
738 | memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); | |
e8546d06 | 739 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
740 | } |
741 | ||
742 | set_buffer_uptodate(bh); | |
743 | unlock_buffer(bh); | |
744 | ||
7e028976 AM |
745 | err = ext4_journal_dirty_metadata(handle, bh); |
746 | if (err) | |
a86c6181 AT |
747 | goto cleanup; |
748 | brelse(bh); | |
749 | bh = NULL; | |
750 | ||
751 | /* correct old leaf */ | |
752 | if (m) { | |
7e028976 AM |
753 | err = ext4_ext_get_access(handle, inode, path + depth); |
754 | if (err) | |
a86c6181 | 755 | goto cleanup; |
e8546d06 | 756 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
757 | err = ext4_ext_dirty(handle, inode, path + depth); |
758 | if (err) | |
a86c6181 AT |
759 | goto cleanup; |
760 | ||
761 | } | |
762 | ||
763 | /* create intermediate indexes */ | |
764 | k = depth - at - 1; | |
765 | BUG_ON(k < 0); | |
766 | if (k) | |
767 | ext_debug("create %d intermediate indices\n", k); | |
768 | /* insert new index into current index block */ | |
769 | /* current depth stored in i var */ | |
770 | i = depth - 1; | |
771 | while (k--) { | |
772 | oldblock = newblock; | |
773 | newblock = ablocks[--a]; | |
bba90743 | 774 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
775 | if (!bh) { |
776 | err = -EIO; | |
777 | goto cleanup; | |
778 | } | |
779 | lock_buffer(bh); | |
780 | ||
7e028976 AM |
781 | err = ext4_journal_get_create_access(handle, bh); |
782 | if (err) | |
a86c6181 AT |
783 | goto cleanup; |
784 | ||
785 | neh = ext_block_hdr(bh); | |
786 | neh->eh_entries = cpu_to_le16(1); | |
787 | neh->eh_magic = EXT4_EXT_MAGIC; | |
788 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
789 | neh->eh_depth = cpu_to_le16(depth - i); | |
790 | fidx = EXT_FIRST_INDEX(neh); | |
791 | fidx->ei_block = border; | |
f65e6fba | 792 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 793 | |
bba90743 ES |
794 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
795 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 AT |
796 | /* copy indexes */ |
797 | m = 0; | |
798 | path[i].p_idx++; | |
799 | ||
800 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
801 | EXT_MAX_INDEX(path[i].p_hdr)); | |
802 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != | |
803 | EXT_LAST_INDEX(path[i].p_hdr)); | |
804 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { | |
26d535ed | 805 | ext_debug("%d: move %d:%llu in new index %llu\n", i, |
8c55e204 DK |
806 | le32_to_cpu(path[i].p_idx->ei_block), |
807 | idx_pblock(path[i].p_idx), | |
808 | newblock); | |
a86c6181 AT |
809 | /*memmove(++fidx, path[i].p_idx++, |
810 | sizeof(struct ext4_extent_idx)); | |
811 | neh->eh_entries++; | |
812 | BUG_ON(neh->eh_entries > neh->eh_max);*/ | |
813 | path[i].p_idx++; | |
814 | m++; | |
815 | } | |
816 | if (m) { | |
817 | memmove(++fidx, path[i].p_idx - m, | |
818 | sizeof(struct ext4_extent_idx) * m); | |
e8546d06 | 819 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
820 | } |
821 | set_buffer_uptodate(bh); | |
822 | unlock_buffer(bh); | |
823 | ||
7e028976 AM |
824 | err = ext4_journal_dirty_metadata(handle, bh); |
825 | if (err) | |
a86c6181 AT |
826 | goto cleanup; |
827 | brelse(bh); | |
828 | bh = NULL; | |
829 | ||
830 | /* correct old index */ | |
831 | if (m) { | |
832 | err = ext4_ext_get_access(handle, inode, path + i); | |
833 | if (err) | |
834 | goto cleanup; | |
e8546d06 | 835 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
836 | err = ext4_ext_dirty(handle, inode, path + i); |
837 | if (err) | |
838 | goto cleanup; | |
839 | } | |
840 | ||
841 | i--; | |
842 | } | |
843 | ||
844 | /* insert new index */ | |
a86c6181 AT |
845 | err = ext4_ext_insert_index(handle, inode, path + at, |
846 | le32_to_cpu(border), newblock); | |
847 | ||
848 | cleanup: | |
849 | if (bh) { | |
850 | if (buffer_locked(bh)) | |
851 | unlock_buffer(bh); | |
852 | brelse(bh); | |
853 | } | |
854 | ||
855 | if (err) { | |
856 | /* free all allocated blocks in error case */ | |
857 | for (i = 0; i < depth; i++) { | |
858 | if (!ablocks[i]) | |
859 | continue; | |
c9de560d | 860 | ext4_free_blocks(handle, inode, ablocks[i], 1, 1); |
a86c6181 AT |
861 | } |
862 | } | |
863 | kfree(ablocks); | |
864 | ||
865 | return err; | |
866 | } | |
867 | ||
868 | /* | |
d0d856e8 RD |
869 | * ext4_ext_grow_indepth: |
870 | * implements tree growing procedure: | |
871 | * - allocates new block | |
872 | * - moves top-level data (index block or leaf) into the new block | |
873 | * - initializes new top-level, creating index that points to the | |
874 | * just created block | |
a86c6181 AT |
875 | */ |
876 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
877 | struct ext4_ext_path *path, | |
878 | struct ext4_extent *newext) | |
879 | { | |
880 | struct ext4_ext_path *curp = path; | |
881 | struct ext4_extent_header *neh; | |
882 | struct ext4_extent_idx *fidx; | |
883 | struct buffer_head *bh; | |
f65e6fba | 884 | ext4_fsblk_t newblock; |
a86c6181 AT |
885 | int err = 0; |
886 | ||
887 | newblock = ext4_ext_new_block(handle, inode, path, newext, &err); | |
888 | if (newblock == 0) | |
889 | return err; | |
890 | ||
891 | bh = sb_getblk(inode->i_sb, newblock); | |
892 | if (!bh) { | |
893 | err = -EIO; | |
894 | ext4_std_error(inode->i_sb, err); | |
895 | return err; | |
896 | } | |
897 | lock_buffer(bh); | |
898 | ||
7e028976 AM |
899 | err = ext4_journal_get_create_access(handle, bh); |
900 | if (err) { | |
a86c6181 AT |
901 | unlock_buffer(bh); |
902 | goto out; | |
903 | } | |
904 | ||
905 | /* move top-level index/leaf into new block */ | |
906 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
907 | ||
908 | /* set size of new block */ | |
909 | neh = ext_block_hdr(bh); | |
910 | /* old root could have indexes or leaves | |
911 | * so calculate e_max right way */ | |
912 | if (ext_depth(inode)) | |
913 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
914 | else | |
915 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
916 | neh->eh_magic = EXT4_EXT_MAGIC; | |
917 | set_buffer_uptodate(bh); | |
918 | unlock_buffer(bh); | |
919 | ||
7e028976 AM |
920 | err = ext4_journal_dirty_metadata(handle, bh); |
921 | if (err) | |
a86c6181 AT |
922 | goto out; |
923 | ||
924 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
925 | err = ext4_ext_get_access(handle, inode, curp); |
926 | if (err) | |
a86c6181 AT |
927 | goto out; |
928 | ||
929 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
930 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); | |
931 | curp->p_hdr->eh_entries = cpu_to_le16(1); | |
932 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
933 | |
934 | if (path[0].p_hdr->eh_depth) | |
935 | curp->p_idx->ei_block = | |
936 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
937 | else | |
938 | curp->p_idx->ei_block = | |
939 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 940 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
941 | |
942 | neh = ext_inode_hdr(inode); | |
943 | fidx = EXT_FIRST_INDEX(neh); | |
2ae02107 | 944 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 945 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
f65e6fba | 946 | le32_to_cpu(fidx->ei_block), idx_pblock(fidx)); |
a86c6181 AT |
947 | |
948 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
949 | err = ext4_ext_dirty(handle, inode, curp); | |
950 | out: | |
951 | brelse(bh); | |
952 | ||
953 | return err; | |
954 | } | |
955 | ||
956 | /* | |
d0d856e8 RD |
957 | * ext4_ext_create_new_leaf: |
958 | * finds empty index and adds new leaf. | |
959 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
960 | */ |
961 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
962 | struct ext4_ext_path *path, | |
963 | struct ext4_extent *newext) | |
964 | { | |
965 | struct ext4_ext_path *curp; | |
966 | int depth, i, err = 0; | |
967 | ||
968 | repeat: | |
969 | i = depth = ext_depth(inode); | |
970 | ||
971 | /* walk up to the tree and look for free index entry */ | |
972 | curp = path + depth; | |
973 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
974 | i--; | |
975 | curp--; | |
976 | } | |
977 | ||
d0d856e8 RD |
978 | /* we use already allocated block for index block, |
979 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
980 | if (EXT_HAS_FREE_INDEX(curp)) { |
981 | /* if we found index with free entry, then use that | |
982 | * entry: create all needed subtree and add new leaf */ | |
983 | err = ext4_ext_split(handle, inode, path, newext, i); | |
984 | ||
985 | /* refill path */ | |
986 | ext4_ext_drop_refs(path); | |
987 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
988 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
989 | path); | |
a86c6181 AT |
990 | if (IS_ERR(path)) |
991 | err = PTR_ERR(path); | |
992 | } else { | |
993 | /* tree is full, time to grow in depth */ | |
994 | err = ext4_ext_grow_indepth(handle, inode, path, newext); | |
995 | if (err) | |
996 | goto out; | |
997 | ||
998 | /* refill path */ | |
999 | ext4_ext_drop_refs(path); | |
1000 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1001 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1002 | path); | |
a86c6181 AT |
1003 | if (IS_ERR(path)) { |
1004 | err = PTR_ERR(path); | |
1005 | goto out; | |
1006 | } | |
1007 | ||
1008 | /* | |
d0d856e8 RD |
1009 | * only first (depth 0 -> 1) produces free space; |
1010 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1011 | */ |
1012 | depth = ext_depth(inode); | |
1013 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1014 | /* now we need to split */ |
a86c6181 AT |
1015 | goto repeat; |
1016 | } | |
1017 | } | |
1018 | ||
1019 | out: | |
1020 | return err; | |
1021 | } | |
1022 | ||
1988b51e AT |
1023 | /* |
1024 | * search the closest allocated block to the left for *logical | |
1025 | * and returns it at @logical + it's physical address at @phys | |
1026 | * if *logical is the smallest allocated block, the function | |
1027 | * returns 0 at @phys | |
1028 | * return value contains 0 (success) or error code | |
1029 | */ | |
1030 | int | |
1031 | ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path, | |
1032 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1033 | { | |
1034 | struct ext4_extent_idx *ix; | |
1035 | struct ext4_extent *ex; | |
b939e376 | 1036 | int depth, ee_len; |
1988b51e AT |
1037 | |
1038 | BUG_ON(path == NULL); | |
1039 | depth = path->p_depth; | |
1040 | *phys = 0; | |
1041 | ||
1042 | if (depth == 0 && path->p_ext == NULL) | |
1043 | return 0; | |
1044 | ||
1045 | /* usually extent in the path covers blocks smaller | |
1046 | * then *logical, but it can be that extent is the | |
1047 | * first one in the file */ | |
1048 | ||
1049 | ex = path[depth].p_ext; | |
b939e376 | 1050 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1051 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1052 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1053 | while (--depth >= 0) { | |
1054 | ix = path[depth].p_idx; | |
1055 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1056 | } | |
1057 | return 0; | |
1058 | } | |
1059 | ||
b939e376 | 1060 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e | 1061 | |
b939e376 AK |
1062 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
1063 | *phys = ext_pblock(ex) + ee_len - 1; | |
1988b51e AT |
1064 | return 0; |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * search the closest allocated block to the right for *logical | |
1069 | * and returns it at @logical + it's physical address at @phys | |
1070 | * if *logical is the smallest allocated block, the function | |
1071 | * returns 0 at @phys | |
1072 | * return value contains 0 (success) or error code | |
1073 | */ | |
1074 | int | |
1075 | ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path, | |
1076 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1077 | { | |
1078 | struct buffer_head *bh = NULL; | |
1079 | struct ext4_extent_header *eh; | |
1080 | struct ext4_extent_idx *ix; | |
1081 | struct ext4_extent *ex; | |
1082 | ext4_fsblk_t block; | |
b939e376 | 1083 | int depth, ee_len; |
1988b51e AT |
1084 | |
1085 | BUG_ON(path == NULL); | |
1086 | depth = path->p_depth; | |
1087 | *phys = 0; | |
1088 | ||
1089 | if (depth == 0 && path->p_ext == NULL) | |
1090 | return 0; | |
1091 | ||
1092 | /* usually extent in the path covers blocks smaller | |
1093 | * then *logical, but it can be that extent is the | |
1094 | * first one in the file */ | |
1095 | ||
1096 | ex = path[depth].p_ext; | |
b939e376 | 1097 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1098 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1099 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1100 | while (--depth >= 0) { | |
1101 | ix = path[depth].p_idx; | |
1102 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1103 | } | |
1104 | *logical = le32_to_cpu(ex->ee_block); | |
1105 | *phys = ext_pblock(ex); | |
1106 | return 0; | |
1107 | } | |
1108 | ||
b939e376 | 1109 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e AT |
1110 | |
1111 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1112 | /* next allocated block in this leaf */ | |
1113 | ex++; | |
1114 | *logical = le32_to_cpu(ex->ee_block); | |
1115 | *phys = ext_pblock(ex); | |
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | /* go up and search for index to the right */ | |
1120 | while (--depth >= 0) { | |
1121 | ix = path[depth].p_idx; | |
1122 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
1123 | break; | |
1124 | } | |
1125 | ||
1126 | if (depth < 0) { | |
1127 | /* we've gone up to the root and | |
1128 | * found no index to the right */ | |
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | /* we've found index to the right, let's | |
1133 | * follow it and find the closest allocated | |
1134 | * block to the right */ | |
1135 | ix++; | |
1136 | block = idx_pblock(ix); | |
1137 | while (++depth < path->p_depth) { | |
1138 | bh = sb_bread(inode->i_sb, block); | |
1139 | if (bh == NULL) | |
1140 | return -EIO; | |
1141 | eh = ext_block_hdr(bh); | |
1142 | if (ext4_ext_check_header(inode, eh, depth)) { | |
1143 | put_bh(bh); | |
1144 | return -EIO; | |
1145 | } | |
1146 | ix = EXT_FIRST_INDEX(eh); | |
1147 | block = idx_pblock(ix); | |
1148 | put_bh(bh); | |
1149 | } | |
1150 | ||
1151 | bh = sb_bread(inode->i_sb, block); | |
1152 | if (bh == NULL) | |
1153 | return -EIO; | |
1154 | eh = ext_block_hdr(bh); | |
1155 | if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) { | |
1156 | put_bh(bh); | |
1157 | return -EIO; | |
1158 | } | |
1159 | ex = EXT_FIRST_EXTENT(eh); | |
1160 | *logical = le32_to_cpu(ex->ee_block); | |
1161 | *phys = ext_pblock(ex); | |
1162 | put_bh(bh); | |
1163 | return 0; | |
1164 | ||
1165 | } | |
1166 | ||
a86c6181 | 1167 | /* |
d0d856e8 RD |
1168 | * ext4_ext_next_allocated_block: |
1169 | * returns allocated block in subsequent extent or EXT_MAX_BLOCK. | |
1170 | * NOTE: it considers block number from index entry as | |
1171 | * allocated block. Thus, index entries have to be consistent | |
1172 | * with leaves. | |
a86c6181 | 1173 | */ |
725d26d3 | 1174 | static ext4_lblk_t |
a86c6181 AT |
1175 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1176 | { | |
1177 | int depth; | |
1178 | ||
1179 | BUG_ON(path == NULL); | |
1180 | depth = path->p_depth; | |
1181 | ||
1182 | if (depth == 0 && path->p_ext == NULL) | |
1183 | return EXT_MAX_BLOCK; | |
1184 | ||
1185 | while (depth >= 0) { | |
1186 | if (depth == path->p_depth) { | |
1187 | /* leaf */ | |
1188 | if (path[depth].p_ext != | |
1189 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1190 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1191 | } else { | |
1192 | /* index */ | |
1193 | if (path[depth].p_idx != | |
1194 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1195 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1196 | } | |
1197 | depth--; | |
1198 | } | |
1199 | ||
1200 | return EXT_MAX_BLOCK; | |
1201 | } | |
1202 | ||
1203 | /* | |
d0d856e8 | 1204 | * ext4_ext_next_leaf_block: |
a86c6181 AT |
1205 | * returns first allocated block from next leaf or EXT_MAX_BLOCK |
1206 | */ | |
725d26d3 | 1207 | static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode, |
63f57933 | 1208 | struct ext4_ext_path *path) |
a86c6181 AT |
1209 | { |
1210 | int depth; | |
1211 | ||
1212 | BUG_ON(path == NULL); | |
1213 | depth = path->p_depth; | |
1214 | ||
1215 | /* zero-tree has no leaf blocks at all */ | |
1216 | if (depth == 0) | |
1217 | return EXT_MAX_BLOCK; | |
1218 | ||
1219 | /* go to index block */ | |
1220 | depth--; | |
1221 | ||
1222 | while (depth >= 0) { | |
1223 | if (path[depth].p_idx != | |
1224 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1225 | return (ext4_lblk_t) |
1226 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1227 | depth--; |
1228 | } | |
1229 | ||
1230 | return EXT_MAX_BLOCK; | |
1231 | } | |
1232 | ||
1233 | /* | |
d0d856e8 RD |
1234 | * ext4_ext_correct_indexes: |
1235 | * if leaf gets modified and modified extent is first in the leaf, | |
1236 | * then we have to correct all indexes above. | |
a86c6181 AT |
1237 | * TODO: do we need to correct tree in all cases? |
1238 | */ | |
1d03ec98 | 1239 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1240 | struct ext4_ext_path *path) |
1241 | { | |
1242 | struct ext4_extent_header *eh; | |
1243 | int depth = ext_depth(inode); | |
1244 | struct ext4_extent *ex; | |
1245 | __le32 border; | |
1246 | int k, err = 0; | |
1247 | ||
1248 | eh = path[depth].p_hdr; | |
1249 | ex = path[depth].p_ext; | |
1250 | BUG_ON(ex == NULL); | |
1251 | BUG_ON(eh == NULL); | |
1252 | ||
1253 | if (depth == 0) { | |
1254 | /* there is no tree at all */ | |
1255 | return 0; | |
1256 | } | |
1257 | ||
1258 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1259 | /* we correct tree if first leaf got modified only */ | |
1260 | return 0; | |
1261 | } | |
1262 | ||
1263 | /* | |
d0d856e8 | 1264 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1265 | */ |
1266 | k = depth - 1; | |
1267 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1268 | err = ext4_ext_get_access(handle, inode, path + k); |
1269 | if (err) | |
a86c6181 AT |
1270 | return err; |
1271 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1272 | err = ext4_ext_dirty(handle, inode, path + k); |
1273 | if (err) | |
a86c6181 AT |
1274 | return err; |
1275 | ||
1276 | while (k--) { | |
1277 | /* change all left-side indexes */ | |
1278 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1279 | break; | |
7e028976 AM |
1280 | err = ext4_ext_get_access(handle, inode, path + k); |
1281 | if (err) | |
a86c6181 AT |
1282 | break; |
1283 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1284 | err = ext4_ext_dirty(handle, inode, path + k); |
1285 | if (err) | |
a86c6181 AT |
1286 | break; |
1287 | } | |
1288 | ||
1289 | return err; | |
1290 | } | |
1291 | ||
09b88252 | 1292 | static int |
a86c6181 AT |
1293 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1294 | struct ext4_extent *ex2) | |
1295 | { | |
749269fa | 1296 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1297 | |
1298 | /* | |
1299 | * Make sure that either both extents are uninitialized, or | |
1300 | * both are _not_. | |
1301 | */ | |
1302 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1303 | return 0; | |
1304 | ||
749269fa AA |
1305 | if (ext4_ext_is_uninitialized(ex1)) |
1306 | max_len = EXT_UNINIT_MAX_LEN; | |
1307 | else | |
1308 | max_len = EXT_INIT_MAX_LEN; | |
1309 | ||
a2df2a63 AA |
1310 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1311 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1312 | ||
1313 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1314 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1315 | return 0; |
1316 | ||
471d4011 SB |
1317 | /* |
1318 | * To allow future support for preallocated extents to be added | |
1319 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1320 | * this can result in the top bit of ee_len being set. |
471d4011 | 1321 | */ |
749269fa | 1322 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1323 | return 0; |
bbf2f9fb | 1324 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1325 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1326 | return 0; |
1327 | #endif | |
1328 | ||
a2df2a63 | 1329 | if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) |
a86c6181 AT |
1330 | return 1; |
1331 | return 0; | |
1332 | } | |
1333 | ||
56055d3a AA |
1334 | /* |
1335 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1336 | * It always tries to merge towards right. If you want to merge towards | |
1337 | * left, pass "ex - 1" as argument instead of "ex". | |
1338 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1339 | * 1 if they got merged. | |
1340 | */ | |
1341 | int ext4_ext_try_to_merge(struct inode *inode, | |
1342 | struct ext4_ext_path *path, | |
1343 | struct ext4_extent *ex) | |
1344 | { | |
1345 | struct ext4_extent_header *eh; | |
1346 | unsigned int depth, len; | |
1347 | int merge_done = 0; | |
1348 | int uninitialized = 0; | |
1349 | ||
1350 | depth = ext_depth(inode); | |
1351 | BUG_ON(path[depth].p_hdr == NULL); | |
1352 | eh = path[depth].p_hdr; | |
1353 | ||
1354 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1355 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1356 | break; | |
1357 | /* merge with next extent! */ | |
1358 | if (ext4_ext_is_uninitialized(ex)) | |
1359 | uninitialized = 1; | |
1360 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1361 | + ext4_ext_get_actual_len(ex + 1)); | |
1362 | if (uninitialized) | |
1363 | ext4_ext_mark_uninitialized(ex); | |
1364 | ||
1365 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1366 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1367 | * sizeof(struct ext4_extent); | |
1368 | memmove(ex + 1, ex + 2, len); | |
1369 | } | |
e8546d06 | 1370 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1371 | merge_done = 1; |
1372 | WARN_ON(eh->eh_entries == 0); | |
1373 | if (!eh->eh_entries) | |
1374 | ext4_error(inode->i_sb, "ext4_ext_try_to_merge", | |
1375 | "inode#%lu, eh->eh_entries = 0!", inode->i_ino); | |
1376 | } | |
1377 | ||
1378 | return merge_done; | |
1379 | } | |
1380 | ||
25d14f98 AA |
1381 | /* |
1382 | * check if a portion of the "newext" extent overlaps with an | |
1383 | * existing extent. | |
1384 | * | |
1385 | * If there is an overlap discovered, it updates the length of the newext | |
1386 | * such that there will be no overlap, and then returns 1. | |
1387 | * If there is no overlap found, it returns 0. | |
1388 | */ | |
1389 | unsigned int ext4_ext_check_overlap(struct inode *inode, | |
1390 | struct ext4_extent *newext, | |
1391 | struct ext4_ext_path *path) | |
1392 | { | |
725d26d3 | 1393 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1394 | unsigned int depth, len1; |
1395 | unsigned int ret = 0; | |
1396 | ||
1397 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1398 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1399 | depth = ext_depth(inode); |
1400 | if (!path[depth].p_ext) | |
1401 | goto out; | |
1402 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
1403 | ||
1404 | /* | |
1405 | * get the next allocated block if the extent in the path | |
1406 | * is before the requested block(s) | |
1407 | */ | |
1408 | if (b2 < b1) { | |
1409 | b2 = ext4_ext_next_allocated_block(path); | |
1410 | if (b2 == EXT_MAX_BLOCK) | |
1411 | goto out; | |
1412 | } | |
1413 | ||
725d26d3 | 1414 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 AA |
1415 | if (b1 + len1 < b1) { |
1416 | len1 = EXT_MAX_BLOCK - b1; | |
1417 | newext->ee_len = cpu_to_le16(len1); | |
1418 | ret = 1; | |
1419 | } | |
1420 | ||
1421 | /* check for overlap */ | |
1422 | if (b1 + len1 > b2) { | |
1423 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1424 | ret = 1; | |
1425 | } | |
1426 | out: | |
1427 | return ret; | |
1428 | } | |
1429 | ||
a86c6181 | 1430 | /* |
d0d856e8 RD |
1431 | * ext4_ext_insert_extent: |
1432 | * tries to merge requsted extent into the existing extent or | |
1433 | * inserts requested extent as new one into the tree, | |
1434 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1435 | */ |
1436 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1437 | struct ext4_ext_path *path, | |
1438 | struct ext4_extent *newext) | |
1439 | { | |
1440 | struct ext4_extent_header * eh; | |
1441 | struct ext4_extent *ex, *fex; | |
1442 | struct ext4_extent *nearex; /* nearest extent */ | |
1443 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1444 | int depth, len, err; |
1445 | ext4_lblk_t next; | |
a2df2a63 | 1446 | unsigned uninitialized = 0; |
a86c6181 | 1447 | |
a2df2a63 | 1448 | BUG_ON(ext4_ext_get_actual_len(newext) == 0); |
a86c6181 AT |
1449 | depth = ext_depth(inode); |
1450 | ex = path[depth].p_ext; | |
1451 | BUG_ON(path[depth].p_hdr == NULL); | |
1452 | ||
1453 | /* try to insert block into found extent and return */ | |
1454 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | |
2ae02107 | 1455 | ext_debug("append %d block to %d:%d (from %llu)\n", |
a2df2a63 | 1456 | ext4_ext_get_actual_len(newext), |
a86c6181 | 1457 | le32_to_cpu(ex->ee_block), |
a2df2a63 | 1458 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
7e028976 AM |
1459 | err = ext4_ext_get_access(handle, inode, path + depth); |
1460 | if (err) | |
a86c6181 | 1461 | return err; |
a2df2a63 AA |
1462 | |
1463 | /* | |
1464 | * ext4_can_extents_be_merged should have checked that either | |
1465 | * both extents are uninitialized, or both aren't. Thus we | |
1466 | * need to check only one of them here. | |
1467 | */ | |
1468 | if (ext4_ext_is_uninitialized(ex)) | |
1469 | uninitialized = 1; | |
1470 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1471 | + ext4_ext_get_actual_len(newext)); | |
1472 | if (uninitialized) | |
1473 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1474 | eh = path[depth].p_hdr; |
1475 | nearex = ex; | |
1476 | goto merge; | |
1477 | } | |
1478 | ||
1479 | repeat: | |
1480 | depth = ext_depth(inode); | |
1481 | eh = path[depth].p_hdr; | |
1482 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1483 | goto has_space; | |
1484 | ||
1485 | /* probably next leaf has space for us? */ | |
1486 | fex = EXT_LAST_EXTENT(eh); | |
1487 | next = ext4_ext_next_leaf_block(inode, path); | |
1488 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) | |
1489 | && next != EXT_MAX_BLOCK) { | |
1490 | ext_debug("next leaf block - %d\n", next); | |
1491 | BUG_ON(npath != NULL); | |
1492 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1493 | if (IS_ERR(npath)) | |
1494 | return PTR_ERR(npath); | |
1495 | BUG_ON(npath->p_depth != path->p_depth); | |
1496 | eh = npath[depth].p_hdr; | |
1497 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
1498 | ext_debug("next leaf isnt full(%d)\n", | |
1499 | le16_to_cpu(eh->eh_entries)); | |
1500 | path = npath; | |
1501 | goto repeat; | |
1502 | } | |
1503 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1504 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1505 | } | |
1506 | ||
1507 | /* | |
d0d856e8 RD |
1508 | * There is no free space in the found leaf. |
1509 | * We're gonna add a new leaf in the tree. | |
a86c6181 AT |
1510 | */ |
1511 | err = ext4_ext_create_new_leaf(handle, inode, path, newext); | |
1512 | if (err) | |
1513 | goto cleanup; | |
1514 | depth = ext_depth(inode); | |
1515 | eh = path[depth].p_hdr; | |
1516 | ||
1517 | has_space: | |
1518 | nearex = path[depth].p_ext; | |
1519 | ||
7e028976 AM |
1520 | err = ext4_ext_get_access(handle, inode, path + depth); |
1521 | if (err) | |
a86c6181 AT |
1522 | goto cleanup; |
1523 | ||
1524 | if (!nearex) { | |
1525 | /* there is no extent in this leaf, create first one */ | |
2ae02107 | 1526 | ext_debug("first extent in the leaf: %d:%llu:%d\n", |
8c55e204 DK |
1527 | le32_to_cpu(newext->ee_block), |
1528 | ext_pblock(newext), | |
a2df2a63 | 1529 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1530 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1531 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1532 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1533 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1534 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1535 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1536 | len = (len - 1) * sizeof(struct ext4_extent); | |
1537 | len = len < 0 ? 0 : len; | |
2ae02107 | 1538 | ext_debug("insert %d:%llu:%d after: nearest 0x%p, " |
a86c6181 | 1539 | "move %d from 0x%p to 0x%p\n", |
8c55e204 DK |
1540 | le32_to_cpu(newext->ee_block), |
1541 | ext_pblock(newext), | |
a2df2a63 | 1542 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1543 | nearex, len, nearex + 1, nearex + 2); |
1544 | memmove(nearex + 2, nearex + 1, len); | |
1545 | } | |
1546 | path[depth].p_ext = nearex + 1; | |
1547 | } else { | |
1548 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1549 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1550 | len = len < 0 ? 0 : len; | |
2ae02107 | 1551 | ext_debug("insert %d:%llu:%d before: nearest 0x%p, " |
a86c6181 AT |
1552 | "move %d from 0x%p to 0x%p\n", |
1553 | le32_to_cpu(newext->ee_block), | |
f65e6fba | 1554 | ext_pblock(newext), |
a2df2a63 | 1555 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1556 | nearex, len, nearex + 1, nearex + 2); |
1557 | memmove(nearex + 1, nearex, len); | |
1558 | path[depth].p_ext = nearex; | |
1559 | } | |
1560 | ||
e8546d06 | 1561 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1562 | nearex = path[depth].p_ext; |
1563 | nearex->ee_block = newext->ee_block; | |
b377611d | 1564 | ext4_ext_store_pblock(nearex, ext_pblock(newext)); |
a86c6181 | 1565 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1566 | |
1567 | merge: | |
1568 | /* try to merge extents to the right */ | |
56055d3a | 1569 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1570 | |
1571 | /* try to merge extents to the left */ | |
1572 | ||
1573 | /* time to correct all indexes above */ | |
1574 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1575 | if (err) | |
1576 | goto cleanup; | |
1577 | ||
1578 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1579 | ||
1580 | cleanup: | |
1581 | if (npath) { | |
1582 | ext4_ext_drop_refs(npath); | |
1583 | kfree(npath); | |
1584 | } | |
1585 | ext4_ext_tree_changed(inode); | |
1586 | ext4_ext_invalidate_cache(inode); | |
1587 | return err; | |
1588 | } | |
1589 | ||
09b88252 | 1590 | static void |
725d26d3 | 1591 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
dd54567a | 1592 | __u32 len, ext4_fsblk_t start, int type) |
a86c6181 AT |
1593 | { |
1594 | struct ext4_ext_cache *cex; | |
1595 | BUG_ON(len == 0); | |
1596 | cex = &EXT4_I(inode)->i_cached_extent; | |
1597 | cex->ec_type = type; | |
1598 | cex->ec_block = block; | |
1599 | cex->ec_len = len; | |
1600 | cex->ec_start = start; | |
1601 | } | |
1602 | ||
1603 | /* | |
d0d856e8 RD |
1604 | * ext4_ext_put_gap_in_cache: |
1605 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
1606 | * and cache this gap |
1607 | */ | |
09b88252 | 1608 | static void |
a86c6181 | 1609 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 1610 | ext4_lblk_t block) |
a86c6181 AT |
1611 | { |
1612 | int depth = ext_depth(inode); | |
725d26d3 AK |
1613 | unsigned long len; |
1614 | ext4_lblk_t lblock; | |
a86c6181 AT |
1615 | struct ext4_extent *ex; |
1616 | ||
1617 | ex = path[depth].p_ext; | |
1618 | if (ex == NULL) { | |
1619 | /* there is no extent yet, so gap is [0;-] */ | |
1620 | lblock = 0; | |
1621 | len = EXT_MAX_BLOCK; | |
1622 | ext_debug("cache gap(whole file):"); | |
1623 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
1624 | lblock = block; | |
1625 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
1626 | ext_debug("cache gap(before): %u [%u:%u]", |
1627 | block, | |
1628 | le32_to_cpu(ex->ee_block), | |
1629 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 1630 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 1631 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 1632 | ext4_lblk_t next; |
8c55e204 | 1633 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 1634 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
1635 | |
1636 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
1637 | ext_debug("cache gap(after): [%u:%u] %u", |
1638 | le32_to_cpu(ex->ee_block), | |
1639 | ext4_ext_get_actual_len(ex), | |
1640 | block); | |
725d26d3 AK |
1641 | BUG_ON(next == lblock); |
1642 | len = next - lblock; | |
a86c6181 AT |
1643 | } else { |
1644 | lblock = len = 0; | |
1645 | BUG(); | |
1646 | } | |
1647 | ||
bba90743 | 1648 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
1649 | ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP); |
1650 | } | |
1651 | ||
09b88252 | 1652 | static int |
725d26d3 | 1653 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, |
a86c6181 AT |
1654 | struct ext4_extent *ex) |
1655 | { | |
1656 | struct ext4_ext_cache *cex; | |
1657 | ||
1658 | cex = &EXT4_I(inode)->i_cached_extent; | |
1659 | ||
1660 | /* has cache valid data? */ | |
1661 | if (cex->ec_type == EXT4_EXT_CACHE_NO) | |
1662 | return EXT4_EXT_CACHE_NO; | |
1663 | ||
1664 | BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP && | |
1665 | cex->ec_type != EXT4_EXT_CACHE_EXTENT); | |
1666 | if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) { | |
8c55e204 | 1667 | ex->ee_block = cpu_to_le32(cex->ec_block); |
f65e6fba | 1668 | ext4_ext_store_pblock(ex, cex->ec_start); |
8c55e204 | 1669 | ex->ee_len = cpu_to_le16(cex->ec_len); |
bba90743 ES |
1670 | ext_debug("%u cached by %u:%u:%llu\n", |
1671 | block, | |
1672 | cex->ec_block, cex->ec_len, cex->ec_start); | |
a86c6181 AT |
1673 | return cex->ec_type; |
1674 | } | |
1675 | ||
1676 | /* not in cache */ | |
1677 | return EXT4_EXT_CACHE_NO; | |
1678 | } | |
1679 | ||
1680 | /* | |
d0d856e8 RD |
1681 | * ext4_ext_rm_idx: |
1682 | * removes index from the index block. | |
1683 | * It's used in truncate case only, thus all requests are for | |
1684 | * last index in the block only. | |
a86c6181 | 1685 | */ |
1d03ec98 | 1686 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1687 | struct ext4_ext_path *path) |
1688 | { | |
1689 | struct buffer_head *bh; | |
1690 | int err; | |
f65e6fba | 1691 | ext4_fsblk_t leaf; |
a86c6181 AT |
1692 | |
1693 | /* free index block */ | |
1694 | path--; | |
f65e6fba | 1695 | leaf = idx_pblock(path->p_idx); |
a86c6181 | 1696 | BUG_ON(path->p_hdr->eh_entries == 0); |
7e028976 AM |
1697 | err = ext4_ext_get_access(handle, inode, path); |
1698 | if (err) | |
a86c6181 | 1699 | return err; |
e8546d06 | 1700 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
1701 | err = ext4_ext_dirty(handle, inode, path); |
1702 | if (err) | |
a86c6181 | 1703 | return err; |
2ae02107 | 1704 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
a86c6181 AT |
1705 | bh = sb_find_get_block(inode->i_sb, leaf); |
1706 | ext4_forget(handle, 1, inode, bh, leaf); | |
c9de560d | 1707 | ext4_free_blocks(handle, inode, leaf, 1, 1); |
a86c6181 AT |
1708 | return err; |
1709 | } | |
1710 | ||
1711 | /* | |
d0d856e8 RD |
1712 | * ext4_ext_calc_credits_for_insert: |
1713 | * This routine returns max. credits that the extent tree can consume. | |
a86c6181 | 1714 | * It should be OK for low-performance paths like ->writepage() |
d0d856e8 | 1715 | * To allow many writing processes to fit into a single transaction, |
0e855ac8 | 1716 | * the caller should calculate credits under i_data_sem and |
d0d856e8 | 1717 | * pass the actual path. |
a86c6181 | 1718 | */ |
09b88252 | 1719 | int ext4_ext_calc_credits_for_insert(struct inode *inode, |
a86c6181 AT |
1720 | struct ext4_ext_path *path) |
1721 | { | |
1722 | int depth, needed; | |
1723 | ||
1724 | if (path) { | |
1725 | /* probably there is space in leaf? */ | |
1726 | depth = ext_depth(inode); | |
1727 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) | |
1728 | < le16_to_cpu(path[depth].p_hdr->eh_max)) | |
1729 | return 1; | |
1730 | } | |
1731 | ||
1732 | /* | |
d0d856e8 | 1733 | * given 32-bit logical block (4294967296 blocks), max. tree |
a86c6181 | 1734 | * can be 4 levels in depth -- 4 * 340^4 == 53453440000. |
d0d856e8 | 1735 | * Let's also add one more level for imbalance. |
a86c6181 AT |
1736 | */ |
1737 | depth = 5; | |
1738 | ||
1739 | /* allocation of new data block(s) */ | |
1740 | needed = 2; | |
1741 | ||
1742 | /* | |
d0d856e8 | 1743 | * tree can be full, so it would need to grow in depth: |
feb18927 JL |
1744 | * we need one credit to modify old root, credits for |
1745 | * new root will be added in split accounting | |
a86c6181 | 1746 | */ |
feb18927 | 1747 | needed += 1; |
a86c6181 AT |
1748 | |
1749 | /* | |
d0d856e8 | 1750 | * Index split can happen, we would need: |
a86c6181 AT |
1751 | * allocate intermediate indexes (bitmap + group) |
1752 | * + change two blocks at each level, but root (already included) | |
1753 | */ | |
feb18927 | 1754 | needed += (depth * 2) + (depth * 2); |
a86c6181 AT |
1755 | |
1756 | /* any allocation modifies superblock */ | |
1757 | needed += 1; | |
1758 | ||
1759 | return needed; | |
1760 | } | |
1761 | ||
1762 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
1763 | struct ext4_extent *ex, | |
725d26d3 | 1764 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 AT |
1765 | { |
1766 | struct buffer_head *bh; | |
a2df2a63 | 1767 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
c9de560d | 1768 | int i, metadata = 0; |
a86c6181 | 1769 | |
c9de560d AT |
1770 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
1771 | metadata = 1; | |
a86c6181 AT |
1772 | #ifdef EXTENTS_STATS |
1773 | { | |
1774 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
1775 | spin_lock(&sbi->s_ext_stats_lock); |
1776 | sbi->s_ext_blocks += ee_len; | |
1777 | sbi->s_ext_extents++; | |
1778 | if (ee_len < sbi->s_ext_min) | |
1779 | sbi->s_ext_min = ee_len; | |
1780 | if (ee_len > sbi->s_ext_max) | |
1781 | sbi->s_ext_max = ee_len; | |
1782 | if (ext_depth(inode) > sbi->s_depth_max) | |
1783 | sbi->s_depth_max = ext_depth(inode); | |
1784 | spin_unlock(&sbi->s_ext_stats_lock); | |
1785 | } | |
1786 | #endif | |
1787 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 1788 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 1789 | /* tail removal */ |
725d26d3 | 1790 | ext4_lblk_t num; |
f65e6fba | 1791 | ext4_fsblk_t start; |
725d26d3 | 1792 | |
a2df2a63 AA |
1793 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
1794 | start = ext_pblock(ex) + ee_len - num; | |
725d26d3 | 1795 | ext_debug("free last %u blocks starting %llu\n", num, start); |
a86c6181 AT |
1796 | for (i = 0; i < num; i++) { |
1797 | bh = sb_find_get_block(inode->i_sb, start + i); | |
1798 | ext4_forget(handle, 0, inode, bh, start + i); | |
1799 | } | |
c9de560d | 1800 | ext4_free_blocks(handle, inode, start, num, metadata); |
a86c6181 | 1801 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 1802 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
725d26d3 | 1803 | printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n", |
a2df2a63 | 1804 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
a86c6181 | 1805 | } else { |
725d26d3 AK |
1806 | printk(KERN_INFO "strange request: removal(2) " |
1807 | "%u-%u from %u:%u\n", | |
1808 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
1809 | } |
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | static int | |
1814 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
725d26d3 | 1815 | struct ext4_ext_path *path, ext4_lblk_t start) |
a86c6181 AT |
1816 | { |
1817 | int err = 0, correct_index = 0; | |
1818 | int depth = ext_depth(inode), credits; | |
1819 | struct ext4_extent_header *eh; | |
725d26d3 AK |
1820 | ext4_lblk_t a, b, block; |
1821 | unsigned num; | |
1822 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 1823 | unsigned short ex_ee_len; |
a2df2a63 | 1824 | unsigned uninitialized = 0; |
a86c6181 AT |
1825 | struct ext4_extent *ex; |
1826 | ||
c29c0ae7 | 1827 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 1828 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
1829 | if (!path[depth].p_hdr) |
1830 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
1831 | eh = path[depth].p_hdr; | |
1832 | BUG_ON(eh == NULL); | |
a86c6181 AT |
1833 | |
1834 | /* find where to start removing */ | |
1835 | ex = EXT_LAST_EXTENT(eh); | |
1836 | ||
1837 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 AA |
1838 | if (ext4_ext_is_uninitialized(ex)) |
1839 | uninitialized = 1; | |
1840 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
a86c6181 AT |
1841 | |
1842 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
1843 | ex_ee_block + ex_ee_len > start) { | |
1844 | ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len); | |
1845 | path[depth].p_ext = ex; | |
1846 | ||
1847 | a = ex_ee_block > start ? ex_ee_block : start; | |
1848 | b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? | |
1849 | ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; | |
1850 | ||
1851 | ext_debug(" border %u:%u\n", a, b); | |
1852 | ||
1853 | if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { | |
1854 | block = 0; | |
1855 | num = 0; | |
1856 | BUG(); | |
1857 | } else if (a != ex_ee_block) { | |
1858 | /* remove tail of the extent */ | |
1859 | block = ex_ee_block; | |
1860 | num = a - block; | |
1861 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
1862 | /* remove head of the extent */ | |
1863 | block = a; | |
1864 | num = b - a; | |
1865 | /* there is no "make a hole" API yet */ | |
1866 | BUG(); | |
1867 | } else { | |
1868 | /* remove whole extent: excellent! */ | |
1869 | block = ex_ee_block; | |
1870 | num = 0; | |
1871 | BUG_ON(a != ex_ee_block); | |
1872 | BUG_ON(b != ex_ee_block + ex_ee_len - 1); | |
1873 | } | |
1874 | ||
d0d856e8 | 1875 | /* at present, extent can't cross block group: */ |
a86c6181 AT |
1876 | /* leaf + bitmap + group desc + sb + inode */ |
1877 | credits = 5; | |
1878 | if (ex == EXT_FIRST_EXTENT(eh)) { | |
1879 | correct_index = 1; | |
1880 | credits += (ext_depth(inode)) + 1; | |
1881 | } | |
1882 | #ifdef CONFIG_QUOTA | |
1883 | credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); | |
1884 | #endif | |
1885 | ||
1886 | handle = ext4_ext_journal_restart(handle, credits); | |
1887 | if (IS_ERR(handle)) { | |
1888 | err = PTR_ERR(handle); | |
1889 | goto out; | |
1890 | } | |
1891 | ||
1892 | err = ext4_ext_get_access(handle, inode, path + depth); | |
1893 | if (err) | |
1894 | goto out; | |
1895 | ||
1896 | err = ext4_remove_blocks(handle, inode, ex, a, b); | |
1897 | if (err) | |
1898 | goto out; | |
1899 | ||
1900 | if (num == 0) { | |
d0d856e8 | 1901 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 1902 | ext4_ext_store_pblock(ex, 0); |
e8546d06 | 1903 | le16_add_cpu(&eh->eh_entries, -1); |
a86c6181 AT |
1904 | } |
1905 | ||
1906 | ex->ee_block = cpu_to_le32(block); | |
1907 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
1908 | /* |
1909 | * Do not mark uninitialized if all the blocks in the | |
1910 | * extent have been removed. | |
1911 | */ | |
1912 | if (uninitialized && num) | |
a2df2a63 | 1913 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
1914 | |
1915 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1916 | if (err) | |
1917 | goto out; | |
1918 | ||
2ae02107 | 1919 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
f65e6fba | 1920 | ext_pblock(ex)); |
a86c6181 AT |
1921 | ex--; |
1922 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 1923 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
1924 | } |
1925 | ||
1926 | if (correct_index && eh->eh_entries) | |
1927 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1928 | ||
1929 | /* if this leaf is free, then we should | |
1930 | * remove it from index block above */ | |
1931 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
1932 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
1933 | ||
1934 | out: | |
1935 | return err; | |
1936 | } | |
1937 | ||
1938 | /* | |
d0d856e8 RD |
1939 | * ext4_ext_more_to_rm: |
1940 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 1941 | */ |
09b88252 | 1942 | static int |
a86c6181 AT |
1943 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
1944 | { | |
1945 | BUG_ON(path->p_idx == NULL); | |
1946 | ||
1947 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
1948 | return 0; | |
1949 | ||
1950 | /* | |
d0d856e8 | 1951 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
1952 | * so we have to consider current index for truncation |
1953 | */ | |
1954 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
1955 | return 0; | |
1956 | return 1; | |
1957 | } | |
1958 | ||
1d03ec98 | 1959 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
1960 | { |
1961 | struct super_block *sb = inode->i_sb; | |
1962 | int depth = ext_depth(inode); | |
1963 | struct ext4_ext_path *path; | |
1964 | handle_t *handle; | |
1965 | int i = 0, err = 0; | |
1966 | ||
725d26d3 | 1967 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
1968 | |
1969 | /* probably first extent we're gonna free will be last in block */ | |
1970 | handle = ext4_journal_start(inode, depth + 1); | |
1971 | if (IS_ERR(handle)) | |
1972 | return PTR_ERR(handle); | |
1973 | ||
1974 | ext4_ext_invalidate_cache(inode); | |
1975 | ||
1976 | /* | |
d0d856e8 RD |
1977 | * We start scanning from right side, freeing all the blocks |
1978 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 1979 | */ |
216553c4 | 1980 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
1981 | if (path == NULL) { |
1982 | ext4_journal_stop(handle); | |
1983 | return -ENOMEM; | |
1984 | } | |
a86c6181 | 1985 | path[0].p_hdr = ext_inode_hdr(inode); |
c29c0ae7 | 1986 | if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
1987 | err = -EIO; |
1988 | goto out; | |
1989 | } | |
1990 | path[0].p_depth = depth; | |
1991 | ||
1992 | while (i >= 0 && err == 0) { | |
1993 | if (i == depth) { | |
1994 | /* this is leaf block */ | |
1995 | err = ext4_ext_rm_leaf(handle, inode, path, start); | |
d0d856e8 | 1996 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
1997 | brelse(path[i].p_bh); |
1998 | path[i].p_bh = NULL; | |
1999 | i--; | |
2000 | continue; | |
2001 | } | |
2002 | ||
2003 | /* this is index block */ | |
2004 | if (!path[i].p_hdr) { | |
2005 | ext_debug("initialize header\n"); | |
2006 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2007 | } |
2008 | ||
a86c6181 | 2009 | if (!path[i].p_idx) { |
d0d856e8 | 2010 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2011 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2012 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2013 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2014 | path[i].p_hdr, | |
2015 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2016 | } else { | |
d0d856e8 | 2017 | /* we were already here, see at next index */ |
a86c6181 AT |
2018 | path[i].p_idx--; |
2019 | } | |
2020 | ||
2021 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2022 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2023 | path[i].p_idx); | |
2024 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2025 | struct buffer_head *bh; |
a86c6181 | 2026 | /* go to the next level */ |
2ae02107 | 2027 | ext_debug("move to level %d (block %llu)\n", |
f65e6fba | 2028 | i + 1, idx_pblock(path[i].p_idx)); |
a86c6181 | 2029 | memset(path + i + 1, 0, sizeof(*path)); |
c29c0ae7 AT |
2030 | bh = sb_bread(sb, idx_pblock(path[i].p_idx)); |
2031 | if (!bh) { | |
a86c6181 AT |
2032 | /* should we reset i_size? */ |
2033 | err = -EIO; | |
2034 | break; | |
2035 | } | |
c29c0ae7 AT |
2036 | if (WARN_ON(i + 1 > depth)) { |
2037 | err = -EIO; | |
2038 | break; | |
2039 | } | |
2040 | if (ext4_ext_check_header(inode, ext_block_hdr(bh), | |
2041 | depth - i - 1)) { | |
2042 | err = -EIO; | |
2043 | break; | |
2044 | } | |
2045 | path[i + 1].p_bh = bh; | |
a86c6181 | 2046 | |
d0d856e8 RD |
2047 | /* save actual number of indexes since this |
2048 | * number is changed at the next iteration */ | |
a86c6181 AT |
2049 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2050 | i++; | |
2051 | } else { | |
d0d856e8 | 2052 | /* we finished processing this index, go up */ |
a86c6181 | 2053 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2054 | /* index is empty, remove it; |
a86c6181 AT |
2055 | * handle must be already prepared by the |
2056 | * truncatei_leaf() */ | |
2057 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2058 | } | |
d0d856e8 | 2059 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2060 | brelse(path[i].p_bh); |
2061 | path[i].p_bh = NULL; | |
2062 | i--; | |
2063 | ext_debug("return to level %d\n", i); | |
2064 | } | |
2065 | } | |
2066 | ||
2067 | /* TODO: flexible tree reduction should be here */ | |
2068 | if (path->p_hdr->eh_entries == 0) { | |
2069 | /* | |
d0d856e8 RD |
2070 | * truncate to zero freed all the tree, |
2071 | * so we need to correct eh_depth | |
a86c6181 AT |
2072 | */ |
2073 | err = ext4_ext_get_access(handle, inode, path); | |
2074 | if (err == 0) { | |
2075 | ext_inode_hdr(inode)->eh_depth = 0; | |
2076 | ext_inode_hdr(inode)->eh_max = | |
2077 | cpu_to_le16(ext4_ext_space_root(inode)); | |
2078 | err = ext4_ext_dirty(handle, inode, path); | |
2079 | } | |
2080 | } | |
2081 | out: | |
2082 | ext4_ext_tree_changed(inode); | |
2083 | ext4_ext_drop_refs(path); | |
2084 | kfree(path); | |
2085 | ext4_journal_stop(handle); | |
2086 | ||
2087 | return err; | |
2088 | } | |
2089 | ||
2090 | /* | |
2091 | * called at mount time | |
2092 | */ | |
2093 | void ext4_ext_init(struct super_block *sb) | |
2094 | { | |
2095 | /* | |
2096 | * possible initialization would be here | |
2097 | */ | |
2098 | ||
2099 | if (test_opt(sb, EXTENTS)) { | |
2100 | printk("EXT4-fs: file extents enabled"); | |
bbf2f9fb RD |
2101 | #ifdef AGGRESSIVE_TEST |
2102 | printk(", aggressive tests"); | |
a86c6181 AT |
2103 | #endif |
2104 | #ifdef CHECK_BINSEARCH | |
2105 | printk(", check binsearch"); | |
2106 | #endif | |
2107 | #ifdef EXTENTS_STATS | |
2108 | printk(", stats"); | |
2109 | #endif | |
2110 | printk("\n"); | |
2111 | #ifdef EXTENTS_STATS | |
2112 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2113 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2114 | EXT4_SB(sb)->s_ext_max = 0; | |
2115 | #endif | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * called at umount time | |
2121 | */ | |
2122 | void ext4_ext_release(struct super_block *sb) | |
2123 | { | |
2124 | if (!test_opt(sb, EXTENTS)) | |
2125 | return; | |
2126 | ||
2127 | #ifdef EXTENTS_STATS | |
2128 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2129 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2130 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2131 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2132 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2133 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2134 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2135 | } | |
2136 | #endif | |
2137 | } | |
2138 | ||
093a088b AK |
2139 | static void bi_complete(struct bio *bio, int error) |
2140 | { | |
2141 | complete((struct completion *)bio->bi_private); | |
2142 | } | |
2143 | ||
2144 | /* FIXME!! we need to try to merge to left or right after zero-out */ | |
2145 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2146 | { | |
2147 | int ret = -EIO; | |
2148 | struct bio *bio; | |
2149 | int blkbits, blocksize; | |
2150 | sector_t ee_pblock; | |
2151 | struct completion event; | |
2152 | unsigned int ee_len, len, done, offset; | |
2153 | ||
2154 | ||
2155 | blkbits = inode->i_blkbits; | |
2156 | blocksize = inode->i_sb->s_blocksize; | |
2157 | ee_len = ext4_ext_get_actual_len(ex); | |
2158 | ee_pblock = ext_pblock(ex); | |
2159 | ||
2160 | /* convert ee_pblock to 512 byte sectors */ | |
2161 | ee_pblock = ee_pblock << (blkbits - 9); | |
2162 | ||
2163 | while (ee_len > 0) { | |
2164 | ||
2165 | if (ee_len > BIO_MAX_PAGES) | |
2166 | len = BIO_MAX_PAGES; | |
2167 | else | |
2168 | len = ee_len; | |
2169 | ||
2170 | bio = bio_alloc(GFP_NOIO, len); | |
2171 | if (!bio) | |
2172 | return -ENOMEM; | |
2173 | bio->bi_sector = ee_pblock; | |
2174 | bio->bi_bdev = inode->i_sb->s_bdev; | |
2175 | ||
2176 | done = 0; | |
2177 | offset = 0; | |
2178 | while (done < len) { | |
2179 | ret = bio_add_page(bio, ZERO_PAGE(0), | |
2180 | blocksize, offset); | |
2181 | if (ret != blocksize) { | |
2182 | /* | |
2183 | * We can't add any more pages because of | |
2184 | * hardware limitations. Start a new bio. | |
2185 | */ | |
2186 | break; | |
2187 | } | |
2188 | done++; | |
2189 | offset += blocksize; | |
2190 | if (offset >= PAGE_CACHE_SIZE) | |
2191 | offset = 0; | |
2192 | } | |
2193 | ||
2194 | init_completion(&event); | |
2195 | bio->bi_private = &event; | |
2196 | bio->bi_end_io = bi_complete; | |
2197 | submit_bio(WRITE, bio); | |
2198 | wait_for_completion(&event); | |
2199 | ||
2200 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2201 | ret = 0; | |
2202 | else { | |
2203 | ret = -EIO; | |
2204 | break; | |
2205 | } | |
2206 | bio_put(bio); | |
2207 | ee_len -= done; | |
2208 | ee_pblock += done << (blkbits - 9); | |
2209 | } | |
2210 | return ret; | |
2211 | } | |
2212 | ||
3977c965 AK |
2213 | #define EXT4_EXT_ZERO_LEN 7 |
2214 | ||
56055d3a AA |
2215 | /* |
2216 | * This function is called by ext4_ext_get_blocks() if someone tries to write | |
2217 | * to an uninitialized extent. It may result in splitting the uninitialized | |
2218 | * extent into multiple extents (upto three - one initialized and two | |
2219 | * uninitialized). | |
2220 | * There are three possibilities: | |
2221 | * a> There is no split required: Entire extent should be initialized | |
2222 | * b> Splits in two extents: Write is happening at either end of the extent | |
2223 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2224 | */ | |
725d26d3 AK |
2225 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
2226 | struct inode *inode, | |
2227 | struct ext4_ext_path *path, | |
2228 | ext4_lblk_t iblock, | |
2229 | unsigned long max_blocks) | |
56055d3a | 2230 | { |
95c3889c | 2231 | struct ext4_extent *ex, newex, orig_ex; |
56055d3a AA |
2232 | struct ext4_extent *ex1 = NULL; |
2233 | struct ext4_extent *ex2 = NULL; | |
2234 | struct ext4_extent *ex3 = NULL; | |
2235 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2236 | ext4_lblk_t ee_block; |
2237 | unsigned int allocated, ee_len, depth; | |
56055d3a AA |
2238 | ext4_fsblk_t newblock; |
2239 | int err = 0; | |
2240 | int ret = 0; | |
2241 | ||
2242 | depth = ext_depth(inode); | |
2243 | eh = path[depth].p_hdr; | |
2244 | ex = path[depth].p_ext; | |
2245 | ee_block = le32_to_cpu(ex->ee_block); | |
2246 | ee_len = ext4_ext_get_actual_len(ex); | |
2247 | allocated = ee_len - (iblock - ee_block); | |
2248 | newblock = iblock - ee_block + ext_pblock(ex); | |
2249 | ex2 = ex; | |
95c3889c AK |
2250 | orig_ex.ee_block = ex->ee_block; |
2251 | orig_ex.ee_len = cpu_to_le16(ee_len); | |
2252 | ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); | |
56055d3a | 2253 | |
9df5643a AK |
2254 | err = ext4_ext_get_access(handle, inode, path + depth); |
2255 | if (err) | |
2256 | goto out; | |
3977c965 AK |
2257 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
2258 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN) { | |
2259 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2260 | if (err) | |
2261 | goto fix_extent_len; | |
2262 | /* update the extent length and mark as initialized */ | |
2263 | ex->ee_block = orig_ex.ee_block; | |
2264 | ex->ee_len = orig_ex.ee_len; | |
2265 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2266 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2267 | /* zeroed the full extent */ |
2268 | return allocated; | |
3977c965 | 2269 | } |
9df5643a | 2270 | |
56055d3a AA |
2271 | /* ex1: ee_block to iblock - 1 : uninitialized */ |
2272 | if (iblock > ee_block) { | |
2273 | ex1 = ex; | |
2274 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2275 | ext4_ext_mark_uninitialized(ex1); | |
2276 | ex2 = &newex; | |
2277 | } | |
2278 | /* | |
2279 | * for sanity, update the length of the ex2 extent before | |
2280 | * we insert ex3, if ex1 is NULL. This is to avoid temporary | |
2281 | * overlap of blocks. | |
2282 | */ | |
2283 | if (!ex1 && allocated > max_blocks) | |
2284 | ex2->ee_len = cpu_to_le16(max_blocks); | |
2285 | /* ex3: to ee_block + ee_len : uninitialised */ | |
2286 | if (allocated > max_blocks) { | |
2287 | unsigned int newdepth; | |
3977c965 AK |
2288 | /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */ |
2289 | if (allocated <= EXT4_EXT_ZERO_LEN) { | |
2290 | /* Mark first half uninitialized. | |
2291 | * Mark second half initialized and zero out the | |
2292 | * initialized extent | |
2293 | */ | |
2294 | ex->ee_block = orig_ex.ee_block; | |
2295 | ex->ee_len = cpu_to_le16(ee_len - allocated); | |
2296 | ext4_ext_mark_uninitialized(ex); | |
2297 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2298 | ext4_ext_dirty(handle, inode, path + depth); | |
2299 | ||
2300 | ex3 = &newex; | |
2301 | ex3->ee_block = cpu_to_le32(iblock); | |
2302 | ext4_ext_store_pblock(ex3, newblock); | |
2303 | ex3->ee_len = cpu_to_le16(allocated); | |
2304 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
2305 | if (err == -ENOSPC) { | |
2306 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2307 | if (err) | |
2308 | goto fix_extent_len; | |
2309 | ex->ee_block = orig_ex.ee_block; | |
2310 | ex->ee_len = orig_ex.ee_len; | |
2311 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2312 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2313 | /* zeroed the full extent */ |
2314 | return allocated; | |
3977c965 AK |
2315 | |
2316 | } else if (err) | |
2317 | goto fix_extent_len; | |
2318 | ||
161e7b7c AK |
2319 | /* |
2320 | * We need to zero out the second half because | |
2321 | * an fallocate request can update file size and | |
2322 | * converting the second half to initialized extent | |
2323 | * implies that we can leak some junk data to user | |
2324 | * space. | |
2325 | */ | |
2326 | err = ext4_ext_zeroout(inode, ex3); | |
2327 | if (err) { | |
2328 | /* | |
2329 | * We should actually mark the | |
2330 | * second half as uninit and return error | |
2331 | * Insert would have changed the extent | |
2332 | */ | |
2333 | depth = ext_depth(inode); | |
2334 | ext4_ext_drop_refs(path); | |
2335 | path = ext4_ext_find_extent(inode, | |
2336 | iblock, path); | |
2337 | if (IS_ERR(path)) { | |
2338 | err = PTR_ERR(path); | |
2339 | return err; | |
2340 | } | |
2341 | ex = path[depth].p_ext; | |
2342 | err = ext4_ext_get_access(handle, inode, | |
2343 | path + depth); | |
2344 | if (err) | |
2345 | return err; | |
2346 | ext4_ext_mark_uninitialized(ex); | |
2347 | ext4_ext_dirty(handle, inode, path + depth); | |
2348 | return err; | |
2349 | } | |
2350 | ||
2351 | /* zeroed the second half */ | |
3977c965 AK |
2352 | return allocated; |
2353 | } | |
56055d3a AA |
2354 | ex3 = &newex; |
2355 | ex3->ee_block = cpu_to_le32(iblock + max_blocks); | |
2356 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | |
2357 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | |
2358 | ext4_ext_mark_uninitialized(ex3); | |
2359 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
093a088b AK |
2360 | if (err == -ENOSPC) { |
2361 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2362 | if (err) | |
2363 | goto fix_extent_len; | |
2364 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2365 | ex->ee_block = orig_ex.ee_block; |
2366 | ex->ee_len = orig_ex.ee_len; | |
2367 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2368 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c AK |
2369 | /* zeroed the full extent */ |
2370 | return allocated; | |
093a088b AK |
2371 | |
2372 | } else if (err) | |
2373 | goto fix_extent_len; | |
56055d3a AA |
2374 | /* |
2375 | * The depth, and hence eh & ex might change | |
2376 | * as part of the insert above. | |
2377 | */ | |
2378 | newdepth = ext_depth(inode); | |
95c3889c AK |
2379 | /* |
2380 | * update the extent length after successfull insert of the | |
2381 | * split extent | |
2382 | */ | |
2383 | orig_ex.ee_len = cpu_to_le16(ee_len - | |
2384 | ext4_ext_get_actual_len(ex3)); | |
56055d3a AA |
2385 | if (newdepth != depth) { |
2386 | depth = newdepth; | |
b35905c1 AK |
2387 | ext4_ext_drop_refs(path); |
2388 | path = ext4_ext_find_extent(inode, iblock, path); | |
56055d3a AA |
2389 | if (IS_ERR(path)) { |
2390 | err = PTR_ERR(path); | |
56055d3a AA |
2391 | goto out; |
2392 | } | |
2393 | eh = path[depth].p_hdr; | |
2394 | ex = path[depth].p_ext; | |
2395 | if (ex2 != &newex) | |
2396 | ex2 = ex; | |
9df5643a AK |
2397 | |
2398 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2399 | if (err) | |
2400 | goto out; | |
56055d3a AA |
2401 | } |
2402 | allocated = max_blocks; | |
3977c965 AK |
2403 | |
2404 | /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying | |
2405 | * to insert a extent in the middle zerout directly | |
2406 | * otherwise give the extent a chance to merge to left | |
2407 | */ | |
2408 | if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN && | |
2409 | iblock != ee_block) { | |
2410 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2411 | if (err) | |
2412 | goto fix_extent_len; | |
2413 | /* update the extent length and mark as initialized */ | |
2414 | ex->ee_block = orig_ex.ee_block; | |
2415 | ex->ee_len = orig_ex.ee_len; | |
2416 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2417 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2418 | /* zero out the first half */ |
2419 | return allocated; | |
3977c965 | 2420 | } |
56055d3a AA |
2421 | } |
2422 | /* | |
2423 | * If there was a change of depth as part of the | |
2424 | * insertion of ex3 above, we need to update the length | |
2425 | * of the ex1 extent again here | |
2426 | */ | |
2427 | if (ex1 && ex1 != ex) { | |
2428 | ex1 = ex; | |
2429 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2430 | ext4_ext_mark_uninitialized(ex1); | |
2431 | ex2 = &newex; | |
2432 | } | |
2433 | /* ex2: iblock to iblock + maxblocks-1 : initialised */ | |
2434 | ex2->ee_block = cpu_to_le32(iblock); | |
56055d3a AA |
2435 | ext4_ext_store_pblock(ex2, newblock); |
2436 | ex2->ee_len = cpu_to_le16(allocated); | |
2437 | if (ex2 != ex) | |
2438 | goto insert; | |
56055d3a AA |
2439 | /* |
2440 | * New (initialized) extent starts from the first block | |
2441 | * in the current extent. i.e., ex2 == ex | |
2442 | * We have to see if it can be merged with the extent | |
2443 | * on the left. | |
2444 | */ | |
2445 | if (ex2 > EXT_FIRST_EXTENT(eh)) { | |
2446 | /* | |
2447 | * To merge left, pass "ex2 - 1" to try_to_merge(), | |
2448 | * since it merges towards right _only_. | |
2449 | */ | |
2450 | ret = ext4_ext_try_to_merge(inode, path, ex2 - 1); | |
2451 | if (ret) { | |
2452 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2453 | if (err) | |
2454 | goto out; | |
2455 | depth = ext_depth(inode); | |
2456 | ex2--; | |
2457 | } | |
2458 | } | |
2459 | /* | |
2460 | * Try to Merge towards right. This might be required | |
2461 | * only when the whole extent is being written to. | |
2462 | * i.e. ex2 == ex and ex3 == NULL. | |
2463 | */ | |
2464 | if (!ex3) { | |
2465 | ret = ext4_ext_try_to_merge(inode, path, ex2); | |
2466 | if (ret) { | |
2467 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2468 | if (err) | |
2469 | goto out; | |
2470 | } | |
2471 | } | |
2472 | /* Mark modified extent as dirty */ | |
2473 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2474 | goto out; | |
2475 | insert: | |
2476 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | |
093a088b AK |
2477 | if (err == -ENOSPC) { |
2478 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2479 | if (err) | |
2480 | goto fix_extent_len; | |
2481 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2482 | ex->ee_block = orig_ex.ee_block; |
2483 | ex->ee_len = orig_ex.ee_len; | |
2484 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2485 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c AK |
2486 | /* zero out the first half */ |
2487 | return allocated; | |
093a088b AK |
2488 | } else if (err) |
2489 | goto fix_extent_len; | |
56055d3a AA |
2490 | out: |
2491 | return err ? err : allocated; | |
093a088b AK |
2492 | |
2493 | fix_extent_len: | |
2494 | ex->ee_block = orig_ex.ee_block; | |
2495 | ex->ee_len = orig_ex.ee_len; | |
2496 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2497 | ext4_ext_mark_uninitialized(ex); | |
2498 | ext4_ext_dirty(handle, inode, path + depth); | |
2499 | return err; | |
56055d3a AA |
2500 | } |
2501 | ||
c278bfec | 2502 | /* |
f5ab0d1f MC |
2503 | * Block allocation/map/preallocation routine for extents based files |
2504 | * | |
2505 | * | |
c278bfec | 2506 | * Need to be called with |
0e855ac8 AK |
2507 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
2508 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
2509 | * |
2510 | * return > 0, number of of blocks already mapped/allocated | |
2511 | * if create == 0 and these are pre-allocated blocks | |
2512 | * buffer head is unmapped | |
2513 | * otherwise blocks are mapped | |
2514 | * | |
2515 | * return = 0, if plain look up failed (blocks have not been allocated) | |
2516 | * buffer head is unmapped | |
2517 | * | |
2518 | * return < 0, error case. | |
c278bfec | 2519 | */ |
f65e6fba | 2520 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, |
725d26d3 | 2521 | ext4_lblk_t iblock, |
a86c6181 AT |
2522 | unsigned long max_blocks, struct buffer_head *bh_result, |
2523 | int create, int extend_disksize) | |
2524 | { | |
2525 | struct ext4_ext_path *path = NULL; | |
56055d3a | 2526 | struct ext4_extent_header *eh; |
a86c6181 | 2527 | struct ext4_extent newex, *ex; |
f65e6fba | 2528 | ext4_fsblk_t goal, newblock; |
56055d3a | 2529 | int err = 0, depth, ret; |
a86c6181 | 2530 | unsigned long allocated = 0; |
c9de560d | 2531 | struct ext4_allocation_request ar; |
a86c6181 AT |
2532 | |
2533 | __clear_bit(BH_New, &bh_result->b_state); | |
bba90743 ES |
2534 | ext_debug("blocks %u/%lu requested for inode %u\n", |
2535 | iblock, max_blocks, inode->i_ino); | |
a86c6181 AT |
2536 | |
2537 | /* check in cache */ | |
7e028976 AM |
2538 | goal = ext4_ext_in_cache(inode, iblock, &newex); |
2539 | if (goal) { | |
a86c6181 AT |
2540 | if (goal == EXT4_EXT_CACHE_GAP) { |
2541 | if (!create) { | |
56055d3a AA |
2542 | /* |
2543 | * block isn't allocated yet and | |
2544 | * user doesn't want to allocate it | |
2545 | */ | |
a86c6181 AT |
2546 | goto out2; |
2547 | } | |
2548 | /* we should allocate requested block */ | |
2549 | } else if (goal == EXT4_EXT_CACHE_EXTENT) { | |
2550 | /* block is already allocated */ | |
8c55e204 DK |
2551 | newblock = iblock |
2552 | - le32_to_cpu(newex.ee_block) | |
2553 | + ext_pblock(&newex); | |
d0d856e8 | 2554 | /* number of remaining blocks in the extent */ |
b939e376 | 2555 | allocated = ext4_ext_get_actual_len(&newex) - |
a86c6181 AT |
2556 | (iblock - le32_to_cpu(newex.ee_block)); |
2557 | goto out; | |
2558 | } else { | |
2559 | BUG(); | |
2560 | } | |
2561 | } | |
2562 | ||
2563 | /* find extent for this block */ | |
2564 | path = ext4_ext_find_extent(inode, iblock, NULL); | |
2565 | if (IS_ERR(path)) { | |
2566 | err = PTR_ERR(path); | |
2567 | path = NULL; | |
2568 | goto out2; | |
2569 | } | |
2570 | ||
2571 | depth = ext_depth(inode); | |
2572 | ||
2573 | /* | |
d0d856e8 RD |
2574 | * consistent leaf must not be empty; |
2575 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
2576 | * this is why assert can't be put in ext4_ext_find_extent() |
2577 | */ | |
2578 | BUG_ON(path[depth].p_ext == NULL && depth != 0); | |
56055d3a | 2579 | eh = path[depth].p_hdr; |
a86c6181 | 2580 | |
7e028976 AM |
2581 | ex = path[depth].p_ext; |
2582 | if (ex) { | |
725d26d3 | 2583 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
f65e6fba | 2584 | ext4_fsblk_t ee_start = ext_pblock(ex); |
a2df2a63 | 2585 | unsigned short ee_len; |
471d4011 SB |
2586 | |
2587 | /* | |
471d4011 | 2588 | * Uninitialized extents are treated as holes, except that |
56055d3a | 2589 | * we split out initialized portions during a write. |
471d4011 | 2590 | */ |
a2df2a63 | 2591 | ee_len = ext4_ext_get_actual_len(ex); |
d0d856e8 | 2592 | /* if found extent covers block, simply return it */ |
8c55e204 | 2593 | if (iblock >= ee_block && iblock < ee_block + ee_len) { |
a86c6181 | 2594 | newblock = iblock - ee_block + ee_start; |
d0d856e8 | 2595 | /* number of remaining blocks in the extent */ |
a86c6181 | 2596 | allocated = ee_len - (iblock - ee_block); |
bba90743 | 2597 | ext_debug("%u fit into %lu:%d -> %llu\n", iblock, |
a86c6181 | 2598 | ee_block, ee_len, newblock); |
56055d3a | 2599 | |
a2df2a63 | 2600 | /* Do not put uninitialized extent in the cache */ |
56055d3a | 2601 | if (!ext4_ext_is_uninitialized(ex)) { |
a2df2a63 AA |
2602 | ext4_ext_put_in_cache(inode, ee_block, |
2603 | ee_len, ee_start, | |
2604 | EXT4_EXT_CACHE_EXTENT); | |
56055d3a AA |
2605 | goto out; |
2606 | } | |
2607 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) | |
2608 | goto out; | |
e067ba00 AK |
2609 | if (!create) { |
2610 | /* | |
2611 | * We have blocks reserved already. We | |
2612 | * return allocated blocks so that delalloc | |
2613 | * won't do block reservation for us. But | |
2614 | * the buffer head will be unmapped so that | |
2615 | * a read from the block returns 0s. | |
2616 | */ | |
2617 | if (allocated > max_blocks) | |
2618 | allocated = max_blocks; | |
1a89734d AK |
2619 | /* mark the buffer unwritten */ |
2620 | __set_bit(BH_Unwritten, &bh_result->b_state); | |
56055d3a | 2621 | goto out2; |
e067ba00 | 2622 | } |
56055d3a AA |
2623 | |
2624 | ret = ext4_ext_convert_to_initialized(handle, inode, | |
2625 | path, iblock, | |
2626 | max_blocks); | |
dbf9d7da DM |
2627 | if (ret <= 0) { |
2628 | err = ret; | |
56055d3a | 2629 | goto out2; |
dbf9d7da | 2630 | } else |
56055d3a AA |
2631 | allocated = ret; |
2632 | goto outnew; | |
a86c6181 AT |
2633 | } |
2634 | } | |
2635 | ||
2636 | /* | |
d0d856e8 | 2637 | * requested block isn't allocated yet; |
a86c6181 AT |
2638 | * we couldn't try to create block if create flag is zero |
2639 | */ | |
2640 | if (!create) { | |
56055d3a AA |
2641 | /* |
2642 | * put just found gap into cache to speed up | |
2643 | * subsequent requests | |
2644 | */ | |
a86c6181 AT |
2645 | ext4_ext_put_gap_in_cache(inode, path, iblock); |
2646 | goto out2; | |
2647 | } | |
2648 | /* | |
63f57933 AM |
2649 | * Okay, we need to do block allocation. Lazily initialize the block |
2650 | * allocation info here if necessary. | |
2651 | */ | |
a86c6181 AT |
2652 | if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info)) |
2653 | ext4_init_block_alloc_info(inode); | |
2654 | ||
c9de560d AT |
2655 | /* find neighbour allocated blocks */ |
2656 | ar.lleft = iblock; | |
2657 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); | |
2658 | if (err) | |
2659 | goto out2; | |
2660 | ar.lright = iblock; | |
2661 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright); | |
2662 | if (err) | |
2663 | goto out2; | |
25d14f98 | 2664 | |
749269fa AA |
2665 | /* |
2666 | * See if request is beyond maximum number of blocks we can have in | |
2667 | * a single extent. For an initialized extent this limit is | |
2668 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
2669 | * EXT_UNINIT_MAX_LEN. | |
2670 | */ | |
2671 | if (max_blocks > EXT_INIT_MAX_LEN && | |
2672 | create != EXT4_CREATE_UNINITIALIZED_EXT) | |
2673 | max_blocks = EXT_INIT_MAX_LEN; | |
2674 | else if (max_blocks > EXT_UNINIT_MAX_LEN && | |
2675 | create == EXT4_CREATE_UNINITIALIZED_EXT) | |
2676 | max_blocks = EXT_UNINIT_MAX_LEN; | |
2677 | ||
25d14f98 AA |
2678 | /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ |
2679 | newex.ee_block = cpu_to_le32(iblock); | |
2680 | newex.ee_len = cpu_to_le16(max_blocks); | |
2681 | err = ext4_ext_check_overlap(inode, &newex, path); | |
2682 | if (err) | |
b939e376 | 2683 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 AA |
2684 | else |
2685 | allocated = max_blocks; | |
c9de560d AT |
2686 | |
2687 | /* allocate new block */ | |
2688 | ar.inode = inode; | |
2689 | ar.goal = ext4_ext_find_goal(inode, path, iblock); | |
2690 | ar.logical = iblock; | |
2691 | ar.len = allocated; | |
2692 | if (S_ISREG(inode->i_mode)) | |
2693 | ar.flags = EXT4_MB_HINT_DATA; | |
2694 | else | |
2695 | /* disable in-core preallocation for non-regular files */ | |
2696 | ar.flags = 0; | |
2697 | newblock = ext4_mb_new_blocks(handle, &ar, &err); | |
a86c6181 AT |
2698 | if (!newblock) |
2699 | goto out2; | |
2ae02107 | 2700 | ext_debug("allocate new block: goal %llu, found %llu/%lu\n", |
a86c6181 AT |
2701 | goal, newblock, allocated); |
2702 | ||
2703 | /* try to insert new extent into found leaf and return */ | |
f65e6fba | 2704 | ext4_ext_store_pblock(&newex, newblock); |
c9de560d | 2705 | newex.ee_len = cpu_to_le16(ar.len); |
a2df2a63 AA |
2706 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */ |
2707 | ext4_ext_mark_uninitialized(&newex); | |
a86c6181 | 2708 | err = ext4_ext_insert_extent(handle, inode, path, &newex); |
315054f0 AT |
2709 | if (err) { |
2710 | /* free data blocks we just allocated */ | |
c9de560d AT |
2711 | /* not a good idea to call discard here directly, |
2712 | * but otherwise we'd need to call it every free() */ | |
2713 | ext4_mb_discard_inode_preallocations(inode); | |
315054f0 | 2714 | ext4_free_blocks(handle, inode, ext_pblock(&newex), |
b939e376 | 2715 | ext4_ext_get_actual_len(&newex), 0); |
a86c6181 | 2716 | goto out2; |
315054f0 | 2717 | } |
a86c6181 AT |
2718 | |
2719 | if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize) | |
2720 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2721 | ||
2722 | /* previous routine could use block we allocated */ | |
f65e6fba | 2723 | newblock = ext_pblock(&newex); |
b939e376 | 2724 | allocated = ext4_ext_get_actual_len(&newex); |
56055d3a | 2725 | outnew: |
a86c6181 AT |
2726 | __set_bit(BH_New, &bh_result->b_state); |
2727 | ||
a2df2a63 AA |
2728 | /* Cache only when it is _not_ an uninitialized extent */ |
2729 | if (create != EXT4_CREATE_UNINITIALIZED_EXT) | |
2730 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, | |
2731 | EXT4_EXT_CACHE_EXTENT); | |
a86c6181 AT |
2732 | out: |
2733 | if (allocated > max_blocks) | |
2734 | allocated = max_blocks; | |
2735 | ext4_ext_show_leaf(inode, path); | |
2736 | __set_bit(BH_Mapped, &bh_result->b_state); | |
2737 | bh_result->b_bdev = inode->i_sb->s_bdev; | |
2738 | bh_result->b_blocknr = newblock; | |
2739 | out2: | |
2740 | if (path) { | |
2741 | ext4_ext_drop_refs(path); | |
2742 | kfree(path); | |
2743 | } | |
a86c6181 AT |
2744 | return err ? err : allocated; |
2745 | } | |
2746 | ||
2747 | void ext4_ext_truncate(struct inode * inode, struct page *page) | |
2748 | { | |
2749 | struct address_space *mapping = inode->i_mapping; | |
2750 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 2751 | ext4_lblk_t last_block; |
a86c6181 AT |
2752 | handle_t *handle; |
2753 | int err = 0; | |
2754 | ||
2755 | /* | |
2756 | * probably first extent we're gonna free will be last in block | |
2757 | */ | |
2758 | err = ext4_writepage_trans_blocks(inode) + 3; | |
2759 | handle = ext4_journal_start(inode, err); | |
2760 | if (IS_ERR(handle)) { | |
2761 | if (page) { | |
2762 | clear_highpage(page); | |
2763 | flush_dcache_page(page); | |
2764 | unlock_page(page); | |
2765 | page_cache_release(page); | |
2766 | } | |
2767 | return; | |
2768 | } | |
2769 | ||
2770 | if (page) | |
2771 | ext4_block_truncate_page(handle, page, mapping, inode->i_size); | |
2772 | ||
0e855ac8 | 2773 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
2774 | ext4_ext_invalidate_cache(inode); |
2775 | ||
c9de560d AT |
2776 | ext4_mb_discard_inode_preallocations(inode); |
2777 | ||
a86c6181 | 2778 | /* |
d0d856e8 RD |
2779 | * TODO: optimization is possible here. |
2780 | * Probably we need not scan at all, | |
2781 | * because page truncation is enough. | |
a86c6181 AT |
2782 | */ |
2783 | if (ext4_orphan_add(handle, inode)) | |
2784 | goto out_stop; | |
2785 | ||
2786 | /* we have to know where to truncate from in crash case */ | |
2787 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2788 | ext4_mark_inode_dirty(handle, inode); | |
2789 | ||
2790 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
2791 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
2792 | err = ext4_ext_remove_space(inode, last_block); | |
2793 | ||
2794 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
2795 | * transaction synchronous. |
2796 | */ | |
a86c6181 AT |
2797 | if (IS_SYNC(inode)) |
2798 | handle->h_sync = 1; | |
2799 | ||
2800 | out_stop: | |
2801 | /* | |
d0d856e8 | 2802 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
2803 | * then we need to clear up the orphan record which we created above. |
2804 | * However, if this was a real unlink then we were called by | |
2805 | * ext4_delete_inode(), and we allow that function to clean up the | |
2806 | * orphan info for us. | |
2807 | */ | |
2808 | if (inode->i_nlink) | |
2809 | ext4_orphan_del(handle, inode); | |
2810 | ||
0e855ac8 | 2811 | up_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
2812 | ext4_journal_stop(handle); |
2813 | } | |
2814 | ||
2815 | /* | |
d0d856e8 RD |
2816 | * ext4_ext_writepage_trans_blocks: |
2817 | * calculate max number of blocks we could modify | |
a86c6181 AT |
2818 | * in order to allocate new block for an inode |
2819 | */ | |
2820 | int ext4_ext_writepage_trans_blocks(struct inode *inode, int num) | |
2821 | { | |
2822 | int needed; | |
2823 | ||
2824 | needed = ext4_ext_calc_credits_for_insert(inode, NULL); | |
2825 | ||
d0d856e8 | 2826 | /* caller wants to allocate num blocks, but note it includes sb */ |
a86c6181 AT |
2827 | needed = needed * num - (num - 1); |
2828 | ||
2829 | #ifdef CONFIG_QUOTA | |
2830 | needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); | |
2831 | #endif | |
2832 | ||
2833 | return needed; | |
2834 | } | |
a2df2a63 | 2835 | |
fd28784a AK |
2836 | static void ext4_falloc_update_inode(struct inode *inode, |
2837 | int mode, loff_t new_size, int update_ctime) | |
2838 | { | |
2839 | struct timespec now; | |
2840 | ||
2841 | if (update_ctime) { | |
2842 | now = current_fs_time(inode->i_sb); | |
2843 | if (!timespec_equal(&inode->i_ctime, &now)) | |
2844 | inode->i_ctime = now; | |
2845 | } | |
2846 | /* | |
2847 | * Update only when preallocation was requested beyond | |
2848 | * the file size. | |
2849 | */ | |
2850 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
2851 | new_size > i_size_read(inode)) { | |
2852 | i_size_write(inode, new_size); | |
2853 | EXT4_I(inode)->i_disksize = new_size; | |
2854 | } | |
2855 | ||
2856 | } | |
2857 | ||
a2df2a63 AA |
2858 | /* |
2859 | * preallocate space for a file. This implements ext4's fallocate inode | |
2860 | * operation, which gets called from sys_fallocate system call. | |
2861 | * For block-mapped files, posix_fallocate should fall back to the method | |
2862 | * of writing zeroes to the required new blocks (the same behavior which is | |
2863 | * expected for file systems which do not support fallocate() system call). | |
2864 | */ | |
2865 | long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) | |
2866 | { | |
2867 | handle_t *handle; | |
725d26d3 | 2868 | ext4_lblk_t block; |
fd28784a | 2869 | loff_t new_size; |
725d26d3 | 2870 | unsigned long max_blocks; |
a2df2a63 AA |
2871 | int ret = 0; |
2872 | int ret2 = 0; | |
2873 | int retries = 0; | |
2874 | struct buffer_head map_bh; | |
2875 | unsigned int credits, blkbits = inode->i_blkbits; | |
2876 | ||
2877 | /* | |
2878 | * currently supporting (pre)allocate mode for extent-based | |
2879 | * files _only_ | |
2880 | */ | |
2881 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
2882 | return -EOPNOTSUPP; | |
2883 | ||
2884 | /* preallocation to directories is currently not supported */ | |
2885 | if (S_ISDIR(inode->i_mode)) | |
2886 | return -ENODEV; | |
2887 | ||
2888 | block = offset >> blkbits; | |
fd28784a AK |
2889 | /* |
2890 | * We can't just convert len to max_blocks because | |
2891 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
2892 | */ | |
a2df2a63 | 2893 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
fd28784a | 2894 | - block; |
a2df2a63 AA |
2895 | /* |
2896 | * credits to insert 1 extent into extent tree + buffers to be able to | |
2897 | * modify 1 super block, 1 block bitmap and 1 group descriptor. | |
2898 | */ | |
2899 | credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3; | |
55bd725a | 2900 | mutex_lock(&inode->i_mutex); |
a2df2a63 AA |
2901 | retry: |
2902 | while (ret >= 0 && ret < max_blocks) { | |
2903 | block = block + ret; | |
2904 | max_blocks = max_blocks - ret; | |
2905 | handle = ext4_journal_start(inode, credits); | |
2906 | if (IS_ERR(handle)) { | |
2907 | ret = PTR_ERR(handle); | |
2908 | break; | |
2909 | } | |
55bd725a | 2910 | ret = ext4_get_blocks_wrap(handle, inode, block, |
a2df2a63 AA |
2911 | max_blocks, &map_bh, |
2912 | EXT4_CREATE_UNINITIALIZED_EXT, 0); | |
221879c9 | 2913 | if (ret <= 0) { |
2c98615d AK |
2914 | #ifdef EXT4FS_DEBUG |
2915 | WARN_ON(ret <= 0); | |
2916 | printk(KERN_ERR "%s: ext4_ext_get_blocks " | |
2917 | "returned error inode#%lu, block=%u, " | |
2918 | "max_blocks=%lu", __func__, | |
221879c9 | 2919 | inode->i_ino, block, max_blocks); |
2c98615d | 2920 | #endif |
a2df2a63 AA |
2921 | ext4_mark_inode_dirty(handle, inode); |
2922 | ret2 = ext4_journal_stop(handle); | |
2923 | break; | |
2924 | } | |
fd28784a AK |
2925 | if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
2926 | blkbits) >> blkbits)) | |
2927 | new_size = offset + len; | |
2928 | else | |
2929 | new_size = (block + ret) << blkbits; | |
a2df2a63 | 2930 | |
fd28784a AK |
2931 | ext4_falloc_update_inode(inode, mode, new_size, |
2932 | buffer_new(&map_bh)); | |
a2df2a63 AA |
2933 | ext4_mark_inode_dirty(handle, inode); |
2934 | ret2 = ext4_journal_stop(handle); | |
2935 | if (ret2) | |
2936 | break; | |
2937 | } | |
fd28784a AK |
2938 | if (ret == -ENOSPC && |
2939 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
2940 | ret = 0; | |
a2df2a63 | 2941 | goto retry; |
a2df2a63 | 2942 | } |
55bd725a | 2943 | mutex_unlock(&inode->i_mutex); |
a2df2a63 AA |
2944 | return ret > 0 ? ret2 : ret; |
2945 | } |