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 | 526 | path[0].p_hdr = eh; |
1973adcb | 527 | path[0].p_bh = NULL; |
a86c6181 | 528 | |
c29c0ae7 | 529 | i = depth; |
a86c6181 AT |
530 | /* walk through the tree */ |
531 | while (i) { | |
532 | ext_debug("depth %d: num %d, max %d\n", | |
533 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 534 | |
a86c6181 | 535 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
f65e6fba | 536 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
537 | path[ppos].p_depth = i; |
538 | path[ppos].p_ext = NULL; | |
539 | ||
540 | bh = sb_bread(inode->i_sb, path[ppos].p_block); | |
541 | if (!bh) | |
542 | goto err; | |
543 | ||
544 | eh = ext_block_hdr(bh); | |
545 | ppos++; | |
546 | BUG_ON(ppos > depth); | |
547 | path[ppos].p_bh = bh; | |
548 | path[ppos].p_hdr = eh; | |
549 | i--; | |
550 | ||
c29c0ae7 | 551 | if (ext4_ext_check_header(inode, eh, i)) |
a86c6181 AT |
552 | goto err; |
553 | } | |
554 | ||
555 | path[ppos].p_depth = i; | |
a86c6181 AT |
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); | |
1973adcb SF |
561 | /* if not an empty leaf */ |
562 | if (path[ppos].p_ext) | |
563 | path[ppos].p_block = ext_pblock(path[ppos].p_ext); | |
a86c6181 AT |
564 | |
565 | ext4_ext_show_path(inode, path); | |
566 | ||
567 | return path; | |
568 | ||
569 | err: | |
570 | ext4_ext_drop_refs(path); | |
571 | if (alloc) | |
572 | kfree(path); | |
573 | return ERR_PTR(-EIO); | |
574 | } | |
575 | ||
576 | /* | |
d0d856e8 RD |
577 | * ext4_ext_insert_index: |
578 | * insert new index [@logical;@ptr] into the block at @curp; | |
579 | * check where to insert: before @curp or after @curp | |
a86c6181 AT |
580 | */ |
581 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |
582 | struct ext4_ext_path *curp, | |
f65e6fba | 583 | int logical, ext4_fsblk_t ptr) |
a86c6181 AT |
584 | { |
585 | struct ext4_extent_idx *ix; | |
586 | int len, err; | |
587 | ||
7e028976 AM |
588 | err = ext4_ext_get_access(handle, inode, curp); |
589 | if (err) | |
a86c6181 AT |
590 | return err; |
591 | ||
592 | BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block)); | |
593 | len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx; | |
594 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { | |
595 | /* insert after */ | |
596 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | |
597 | len = (len - 1) * sizeof(struct ext4_extent_idx); | |
598 | len = len < 0 ? 0 : len; | |
26d535ed | 599 | ext_debug("insert new index %d after: %llu. " |
a86c6181 AT |
600 | "move %d from 0x%p to 0x%p\n", |
601 | logical, ptr, len, | |
602 | (curp->p_idx + 1), (curp->p_idx + 2)); | |
603 | memmove(curp->p_idx + 2, curp->p_idx + 1, len); | |
604 | } | |
605 | ix = curp->p_idx + 1; | |
606 | } else { | |
607 | /* insert before */ | |
608 | len = len * sizeof(struct ext4_extent_idx); | |
609 | len = len < 0 ? 0 : len; | |
26d535ed | 610 | ext_debug("insert new index %d before: %llu. " |
a86c6181 AT |
611 | "move %d from 0x%p to 0x%p\n", |
612 | logical, ptr, len, | |
613 | curp->p_idx, (curp->p_idx + 1)); | |
614 | memmove(curp->p_idx + 1, curp->p_idx, len); | |
615 | ix = curp->p_idx; | |
616 | } | |
617 | ||
618 | ix->ei_block = cpu_to_le32(logical); | |
f65e6fba | 619 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 620 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 AT |
621 | |
622 | BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries) | |
8c55e204 | 623 | > le16_to_cpu(curp->p_hdr->eh_max)); |
a86c6181 AT |
624 | BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr)); |
625 | ||
626 | err = ext4_ext_dirty(handle, inode, curp); | |
627 | ext4_std_error(inode->i_sb, err); | |
628 | ||
629 | return err; | |
630 | } | |
631 | ||
632 | /* | |
d0d856e8 RD |
633 | * ext4_ext_split: |
634 | * inserts new subtree into the path, using free index entry | |
635 | * at depth @at: | |
636 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
637 | * - makes decision where to split | |
638 | * - moves remaining extents and index entries (right to the split point) | |
639 | * into the newly allocated blocks | |
640 | * - initializes subtree | |
a86c6181 AT |
641 | */ |
642 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
643 | struct ext4_ext_path *path, | |
644 | struct ext4_extent *newext, int at) | |
645 | { | |
646 | struct buffer_head *bh = NULL; | |
647 | int depth = ext_depth(inode); | |
648 | struct ext4_extent_header *neh; | |
649 | struct ext4_extent_idx *fidx; | |
650 | struct ext4_extent *ex; | |
651 | int i = at, k, m, a; | |
f65e6fba | 652 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 653 | __le32 border; |
f65e6fba | 654 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
655 | int err = 0; |
656 | ||
657 | /* make decision: where to split? */ | |
d0d856e8 | 658 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 659 | |
d0d856e8 | 660 | /* if current leaf will be split, then we should use |
a86c6181 AT |
661 | * border from split point */ |
662 | BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr)); | |
663 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
664 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 665 | ext_debug("leaf will be split." |
a86c6181 | 666 | " next leaf starts at %d\n", |
8c55e204 | 667 | le32_to_cpu(border)); |
a86c6181 AT |
668 | } else { |
669 | border = newext->ee_block; | |
670 | ext_debug("leaf will be added." | |
671 | " next leaf starts at %d\n", | |
8c55e204 | 672 | le32_to_cpu(border)); |
a86c6181 AT |
673 | } |
674 | ||
675 | /* | |
d0d856e8 RD |
676 | * If error occurs, then we break processing |
677 | * and mark filesystem read-only. index won't | |
a86c6181 | 678 | * be inserted and tree will be in consistent |
d0d856e8 | 679 | * state. Next mount will repair buffers too. |
a86c6181 AT |
680 | */ |
681 | ||
682 | /* | |
d0d856e8 RD |
683 | * Get array to track all allocated blocks. |
684 | * We need this to handle errors and free blocks | |
685 | * upon them. | |
a86c6181 | 686 | */ |
5d4958f9 | 687 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
688 | if (!ablocks) |
689 | return -ENOMEM; | |
a86c6181 AT |
690 | |
691 | /* allocate all needed blocks */ | |
692 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
693 | for (a = 0; a < depth - at; a++) { | |
694 | newblock = ext4_ext_new_block(handle, inode, path, newext, &err); | |
695 | if (newblock == 0) | |
696 | goto cleanup; | |
697 | ablocks[a] = newblock; | |
698 | } | |
699 | ||
700 | /* initialize new leaf */ | |
701 | newblock = ablocks[--a]; | |
702 | BUG_ON(newblock == 0); | |
703 | bh = sb_getblk(inode->i_sb, newblock); | |
704 | if (!bh) { | |
705 | err = -EIO; | |
706 | goto cleanup; | |
707 | } | |
708 | lock_buffer(bh); | |
709 | ||
7e028976 AM |
710 | err = ext4_journal_get_create_access(handle, bh); |
711 | if (err) | |
a86c6181 AT |
712 | goto cleanup; |
713 | ||
714 | neh = ext_block_hdr(bh); | |
715 | neh->eh_entries = 0; | |
716 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
717 | neh->eh_magic = EXT4_EXT_MAGIC; | |
718 | neh->eh_depth = 0; | |
719 | ex = EXT_FIRST_EXTENT(neh); | |
720 | ||
d0d856e8 | 721 | /* move remainder of path[depth] to the new leaf */ |
a86c6181 AT |
722 | BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max); |
723 | /* start copy from next extent */ | |
724 | /* TODO: we could do it by single memmove */ | |
725 | m = 0; | |
726 | path[depth].p_ext++; | |
727 | while (path[depth].p_ext <= | |
728 | EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
2ae02107 | 729 | ext_debug("move %d:%llu:%d in new leaf %llu\n", |
8c55e204 DK |
730 | le32_to_cpu(path[depth].p_ext->ee_block), |
731 | ext_pblock(path[depth].p_ext), | |
a2df2a63 | 732 | ext4_ext_get_actual_len(path[depth].p_ext), |
a86c6181 AT |
733 | newblock); |
734 | /*memmove(ex++, path[depth].p_ext++, | |
735 | sizeof(struct ext4_extent)); | |
736 | neh->eh_entries++;*/ | |
737 | path[depth].p_ext++; | |
738 | m++; | |
739 | } | |
740 | if (m) { | |
741 | memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m); | |
e8546d06 | 742 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
743 | } |
744 | ||
745 | set_buffer_uptodate(bh); | |
746 | unlock_buffer(bh); | |
747 | ||
7e028976 AM |
748 | err = ext4_journal_dirty_metadata(handle, bh); |
749 | if (err) | |
a86c6181 AT |
750 | goto cleanup; |
751 | brelse(bh); | |
752 | bh = NULL; | |
753 | ||
754 | /* correct old leaf */ | |
755 | if (m) { | |
7e028976 AM |
756 | err = ext4_ext_get_access(handle, inode, path + depth); |
757 | if (err) | |
a86c6181 | 758 | goto cleanup; |
e8546d06 | 759 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
760 | err = ext4_ext_dirty(handle, inode, path + depth); |
761 | if (err) | |
a86c6181 AT |
762 | goto cleanup; |
763 | ||
764 | } | |
765 | ||
766 | /* create intermediate indexes */ | |
767 | k = depth - at - 1; | |
768 | BUG_ON(k < 0); | |
769 | if (k) | |
770 | ext_debug("create %d intermediate indices\n", k); | |
771 | /* insert new index into current index block */ | |
772 | /* current depth stored in i var */ | |
773 | i = depth - 1; | |
774 | while (k--) { | |
775 | oldblock = newblock; | |
776 | newblock = ablocks[--a]; | |
bba90743 | 777 | bh = sb_getblk(inode->i_sb, newblock); |
a86c6181 AT |
778 | if (!bh) { |
779 | err = -EIO; | |
780 | goto cleanup; | |
781 | } | |
782 | lock_buffer(bh); | |
783 | ||
7e028976 AM |
784 | err = ext4_journal_get_create_access(handle, bh); |
785 | if (err) | |
a86c6181 AT |
786 | goto cleanup; |
787 | ||
788 | neh = ext_block_hdr(bh); | |
789 | neh->eh_entries = cpu_to_le16(1); | |
790 | neh->eh_magic = EXT4_EXT_MAGIC; | |
791 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
792 | neh->eh_depth = cpu_to_le16(depth - i); | |
793 | fidx = EXT_FIRST_INDEX(neh); | |
794 | fidx->ei_block = border; | |
f65e6fba | 795 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 796 | |
bba90743 ES |
797 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
798 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 AT |
799 | /* copy indexes */ |
800 | m = 0; | |
801 | path[i].p_idx++; | |
802 | ||
803 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
804 | EXT_MAX_INDEX(path[i].p_hdr)); | |
805 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != | |
806 | EXT_LAST_INDEX(path[i].p_hdr)); | |
807 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { | |
26d535ed | 808 | ext_debug("%d: move %d:%llu in new index %llu\n", i, |
8c55e204 DK |
809 | le32_to_cpu(path[i].p_idx->ei_block), |
810 | idx_pblock(path[i].p_idx), | |
811 | newblock); | |
a86c6181 AT |
812 | /*memmove(++fidx, path[i].p_idx++, |
813 | sizeof(struct ext4_extent_idx)); | |
814 | neh->eh_entries++; | |
815 | BUG_ON(neh->eh_entries > neh->eh_max);*/ | |
816 | path[i].p_idx++; | |
817 | m++; | |
818 | } | |
819 | if (m) { | |
820 | memmove(++fidx, path[i].p_idx - m, | |
821 | sizeof(struct ext4_extent_idx) * m); | |
e8546d06 | 822 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
823 | } |
824 | set_buffer_uptodate(bh); | |
825 | unlock_buffer(bh); | |
826 | ||
7e028976 AM |
827 | err = ext4_journal_dirty_metadata(handle, bh); |
828 | if (err) | |
a86c6181 AT |
829 | goto cleanup; |
830 | brelse(bh); | |
831 | bh = NULL; | |
832 | ||
833 | /* correct old index */ | |
834 | if (m) { | |
835 | err = ext4_ext_get_access(handle, inode, path + i); | |
836 | if (err) | |
837 | goto cleanup; | |
e8546d06 | 838 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
839 | err = ext4_ext_dirty(handle, inode, path + i); |
840 | if (err) | |
841 | goto cleanup; | |
842 | } | |
843 | ||
844 | i--; | |
845 | } | |
846 | ||
847 | /* insert new index */ | |
a86c6181 AT |
848 | err = ext4_ext_insert_index(handle, inode, path + at, |
849 | le32_to_cpu(border), newblock); | |
850 | ||
851 | cleanup: | |
852 | if (bh) { | |
853 | if (buffer_locked(bh)) | |
854 | unlock_buffer(bh); | |
855 | brelse(bh); | |
856 | } | |
857 | ||
858 | if (err) { | |
859 | /* free all allocated blocks in error case */ | |
860 | for (i = 0; i < depth; i++) { | |
861 | if (!ablocks[i]) | |
862 | continue; | |
c9de560d | 863 | ext4_free_blocks(handle, inode, ablocks[i], 1, 1); |
a86c6181 AT |
864 | } |
865 | } | |
866 | kfree(ablocks); | |
867 | ||
868 | return err; | |
869 | } | |
870 | ||
871 | /* | |
d0d856e8 RD |
872 | * ext4_ext_grow_indepth: |
873 | * implements tree growing procedure: | |
874 | * - allocates new block | |
875 | * - moves top-level data (index block or leaf) into the new block | |
876 | * - initializes new top-level, creating index that points to the | |
877 | * just created block | |
a86c6181 AT |
878 | */ |
879 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
880 | struct ext4_ext_path *path, | |
881 | struct ext4_extent *newext) | |
882 | { | |
883 | struct ext4_ext_path *curp = path; | |
884 | struct ext4_extent_header *neh; | |
885 | struct ext4_extent_idx *fidx; | |
886 | struct buffer_head *bh; | |
f65e6fba | 887 | ext4_fsblk_t newblock; |
a86c6181 AT |
888 | int err = 0; |
889 | ||
890 | newblock = ext4_ext_new_block(handle, inode, path, newext, &err); | |
891 | if (newblock == 0) | |
892 | return err; | |
893 | ||
894 | bh = sb_getblk(inode->i_sb, newblock); | |
895 | if (!bh) { | |
896 | err = -EIO; | |
897 | ext4_std_error(inode->i_sb, err); | |
898 | return err; | |
899 | } | |
900 | lock_buffer(bh); | |
901 | ||
7e028976 AM |
902 | err = ext4_journal_get_create_access(handle, bh); |
903 | if (err) { | |
a86c6181 AT |
904 | unlock_buffer(bh); |
905 | goto out; | |
906 | } | |
907 | ||
908 | /* move top-level index/leaf into new block */ | |
909 | memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data)); | |
910 | ||
911 | /* set size of new block */ | |
912 | neh = ext_block_hdr(bh); | |
913 | /* old root could have indexes or leaves | |
914 | * so calculate e_max right way */ | |
915 | if (ext_depth(inode)) | |
916 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode)); | |
917 | else | |
918 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode)); | |
919 | neh->eh_magic = EXT4_EXT_MAGIC; | |
920 | set_buffer_uptodate(bh); | |
921 | unlock_buffer(bh); | |
922 | ||
7e028976 AM |
923 | err = ext4_journal_dirty_metadata(handle, bh); |
924 | if (err) | |
a86c6181 AT |
925 | goto out; |
926 | ||
927 | /* create index in new top-level index: num,max,pointer */ | |
7e028976 AM |
928 | err = ext4_ext_get_access(handle, inode, curp); |
929 | if (err) | |
a86c6181 AT |
930 | goto out; |
931 | ||
932 | curp->p_hdr->eh_magic = EXT4_EXT_MAGIC; | |
933 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); | |
934 | curp->p_hdr->eh_entries = cpu_to_le16(1); | |
935 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | |
e9f410b1 DM |
936 | |
937 | if (path[0].p_hdr->eh_depth) | |
938 | curp->p_idx->ei_block = | |
939 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | |
940 | else | |
941 | curp->p_idx->ei_block = | |
942 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | |
f65e6fba | 943 | ext4_idx_store_pblock(curp->p_idx, newblock); |
a86c6181 AT |
944 | |
945 | neh = ext_inode_hdr(inode); | |
946 | fidx = EXT_FIRST_INDEX(neh); | |
2ae02107 | 947 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 948 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
f65e6fba | 949 | le32_to_cpu(fidx->ei_block), idx_pblock(fidx)); |
a86c6181 AT |
950 | |
951 | neh->eh_depth = cpu_to_le16(path->p_depth + 1); | |
952 | err = ext4_ext_dirty(handle, inode, curp); | |
953 | out: | |
954 | brelse(bh); | |
955 | ||
956 | return err; | |
957 | } | |
958 | ||
959 | /* | |
d0d856e8 RD |
960 | * ext4_ext_create_new_leaf: |
961 | * finds empty index and adds new leaf. | |
962 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
963 | */ |
964 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
965 | struct ext4_ext_path *path, | |
966 | struct ext4_extent *newext) | |
967 | { | |
968 | struct ext4_ext_path *curp; | |
969 | int depth, i, err = 0; | |
970 | ||
971 | repeat: | |
972 | i = depth = ext_depth(inode); | |
973 | ||
974 | /* walk up to the tree and look for free index entry */ | |
975 | curp = path + depth; | |
976 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
977 | i--; | |
978 | curp--; | |
979 | } | |
980 | ||
d0d856e8 RD |
981 | /* we use already allocated block for index block, |
982 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
983 | if (EXT_HAS_FREE_INDEX(curp)) { |
984 | /* if we found index with free entry, then use that | |
985 | * entry: create all needed subtree and add new leaf */ | |
986 | err = ext4_ext_split(handle, inode, path, newext, i); | |
787e0981 SF |
987 | if (err) |
988 | goto out; | |
a86c6181 AT |
989 | |
990 | /* refill path */ | |
991 | ext4_ext_drop_refs(path); | |
992 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
993 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
994 | path); | |
a86c6181 AT |
995 | if (IS_ERR(path)) |
996 | err = PTR_ERR(path); | |
997 | } else { | |
998 | /* tree is full, time to grow in depth */ | |
999 | err = ext4_ext_grow_indepth(handle, inode, path, newext); | |
1000 | if (err) | |
1001 | goto out; | |
1002 | ||
1003 | /* refill path */ | |
1004 | ext4_ext_drop_refs(path); | |
1005 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1006 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1007 | path); | |
a86c6181 AT |
1008 | if (IS_ERR(path)) { |
1009 | err = PTR_ERR(path); | |
1010 | goto out; | |
1011 | } | |
1012 | ||
1013 | /* | |
d0d856e8 RD |
1014 | * only first (depth 0 -> 1) produces free space; |
1015 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1016 | */ |
1017 | depth = ext_depth(inode); | |
1018 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1019 | /* now we need to split */ |
a86c6181 AT |
1020 | goto repeat; |
1021 | } | |
1022 | } | |
1023 | ||
1024 | out: | |
1025 | return err; | |
1026 | } | |
1027 | ||
1988b51e AT |
1028 | /* |
1029 | * search the closest allocated block to the left for *logical | |
1030 | * and returns it at @logical + it's physical address at @phys | |
1031 | * if *logical is the smallest allocated block, the function | |
1032 | * returns 0 at @phys | |
1033 | * return value contains 0 (success) or error code | |
1034 | */ | |
1035 | int | |
1036 | ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path, | |
1037 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1038 | { | |
1039 | struct ext4_extent_idx *ix; | |
1040 | struct ext4_extent *ex; | |
b939e376 | 1041 | int depth, ee_len; |
1988b51e AT |
1042 | |
1043 | BUG_ON(path == NULL); | |
1044 | depth = path->p_depth; | |
1045 | *phys = 0; | |
1046 | ||
1047 | if (depth == 0 && path->p_ext == NULL) | |
1048 | return 0; | |
1049 | ||
1050 | /* usually extent in the path covers blocks smaller | |
1051 | * then *logical, but it can be that extent is the | |
1052 | * first one in the file */ | |
1053 | ||
1054 | ex = path[depth].p_ext; | |
b939e376 | 1055 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1056 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1057 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1058 | while (--depth >= 0) { | |
1059 | ix = path[depth].p_idx; | |
1060 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1061 | } | |
1062 | return 0; | |
1063 | } | |
1064 | ||
b939e376 | 1065 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e | 1066 | |
b939e376 AK |
1067 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
1068 | *phys = ext_pblock(ex) + ee_len - 1; | |
1988b51e AT |
1069 | return 0; |
1070 | } | |
1071 | ||
1072 | /* | |
1073 | * search the closest allocated block to the right for *logical | |
1074 | * and returns it at @logical + it's physical address at @phys | |
1075 | * if *logical is the smallest allocated block, the function | |
1076 | * returns 0 at @phys | |
1077 | * return value contains 0 (success) or error code | |
1078 | */ | |
1079 | int | |
1080 | ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path, | |
1081 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1082 | { | |
1083 | struct buffer_head *bh = NULL; | |
1084 | struct ext4_extent_header *eh; | |
1085 | struct ext4_extent_idx *ix; | |
1086 | struct ext4_extent *ex; | |
1087 | ext4_fsblk_t block; | |
b939e376 | 1088 | int depth, ee_len; |
1988b51e AT |
1089 | |
1090 | BUG_ON(path == NULL); | |
1091 | depth = path->p_depth; | |
1092 | *phys = 0; | |
1093 | ||
1094 | if (depth == 0 && path->p_ext == NULL) | |
1095 | return 0; | |
1096 | ||
1097 | /* usually extent in the path covers blocks smaller | |
1098 | * then *logical, but it can be that extent is the | |
1099 | * first one in the file */ | |
1100 | ||
1101 | ex = path[depth].p_ext; | |
b939e376 | 1102 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e AT |
1103 | if (*logical < le32_to_cpu(ex->ee_block)) { |
1104 | BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex); | |
1105 | while (--depth >= 0) { | |
1106 | ix = path[depth].p_idx; | |
1107 | BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr)); | |
1108 | } | |
1109 | *logical = le32_to_cpu(ex->ee_block); | |
1110 | *phys = ext_pblock(ex); | |
1111 | return 0; | |
1112 | } | |
1113 | ||
b939e376 | 1114 | BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len)); |
1988b51e AT |
1115 | |
1116 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1117 | /* next allocated block in this leaf */ | |
1118 | ex++; | |
1119 | *logical = le32_to_cpu(ex->ee_block); | |
1120 | *phys = ext_pblock(ex); | |
1121 | return 0; | |
1122 | } | |
1123 | ||
1124 | /* go up and search for index to the right */ | |
1125 | while (--depth >= 0) { | |
1126 | ix = path[depth].p_idx; | |
1127 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
1128 | break; | |
1129 | } | |
1130 | ||
1131 | if (depth < 0) { | |
1132 | /* we've gone up to the root and | |
1133 | * found no index to the right */ | |
1134 | return 0; | |
1135 | } | |
1136 | ||
1137 | /* we've found index to the right, let's | |
1138 | * follow it and find the closest allocated | |
1139 | * block to the right */ | |
1140 | ix++; | |
1141 | block = idx_pblock(ix); | |
1142 | while (++depth < path->p_depth) { | |
1143 | bh = sb_bread(inode->i_sb, block); | |
1144 | if (bh == NULL) | |
1145 | return -EIO; | |
1146 | eh = ext_block_hdr(bh); | |
1147 | if (ext4_ext_check_header(inode, eh, depth)) { | |
1148 | put_bh(bh); | |
1149 | return -EIO; | |
1150 | } | |
1151 | ix = EXT_FIRST_INDEX(eh); | |
1152 | block = idx_pblock(ix); | |
1153 | put_bh(bh); | |
1154 | } | |
1155 | ||
1156 | bh = sb_bread(inode->i_sb, block); | |
1157 | if (bh == NULL) | |
1158 | return -EIO; | |
1159 | eh = ext_block_hdr(bh); | |
1160 | if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) { | |
1161 | put_bh(bh); | |
1162 | return -EIO; | |
1163 | } | |
1164 | ex = EXT_FIRST_EXTENT(eh); | |
1165 | *logical = le32_to_cpu(ex->ee_block); | |
1166 | *phys = ext_pblock(ex); | |
1167 | put_bh(bh); | |
1168 | return 0; | |
1169 | ||
1170 | } | |
1171 | ||
a86c6181 | 1172 | /* |
d0d856e8 RD |
1173 | * ext4_ext_next_allocated_block: |
1174 | * returns allocated block in subsequent extent or EXT_MAX_BLOCK. | |
1175 | * NOTE: it considers block number from index entry as | |
1176 | * allocated block. Thus, index entries have to be consistent | |
1177 | * with leaves. | |
a86c6181 | 1178 | */ |
725d26d3 | 1179 | static ext4_lblk_t |
a86c6181 AT |
1180 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1181 | { | |
1182 | int depth; | |
1183 | ||
1184 | BUG_ON(path == NULL); | |
1185 | depth = path->p_depth; | |
1186 | ||
1187 | if (depth == 0 && path->p_ext == NULL) | |
1188 | return EXT_MAX_BLOCK; | |
1189 | ||
1190 | while (depth >= 0) { | |
1191 | if (depth == path->p_depth) { | |
1192 | /* leaf */ | |
1193 | if (path[depth].p_ext != | |
1194 | EXT_LAST_EXTENT(path[depth].p_hdr)) | |
1195 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1196 | } else { | |
1197 | /* index */ | |
1198 | if (path[depth].p_idx != | |
1199 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1200 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1201 | } | |
1202 | depth--; | |
1203 | } | |
1204 | ||
1205 | return EXT_MAX_BLOCK; | |
1206 | } | |
1207 | ||
1208 | /* | |
d0d856e8 | 1209 | * ext4_ext_next_leaf_block: |
a86c6181 AT |
1210 | * returns first allocated block from next leaf or EXT_MAX_BLOCK |
1211 | */ | |
725d26d3 | 1212 | static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode, |
63f57933 | 1213 | struct ext4_ext_path *path) |
a86c6181 AT |
1214 | { |
1215 | int depth; | |
1216 | ||
1217 | BUG_ON(path == NULL); | |
1218 | depth = path->p_depth; | |
1219 | ||
1220 | /* zero-tree has no leaf blocks at all */ | |
1221 | if (depth == 0) | |
1222 | return EXT_MAX_BLOCK; | |
1223 | ||
1224 | /* go to index block */ | |
1225 | depth--; | |
1226 | ||
1227 | while (depth >= 0) { | |
1228 | if (path[depth].p_idx != | |
1229 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1230 | return (ext4_lblk_t) |
1231 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1232 | depth--; |
1233 | } | |
1234 | ||
1235 | return EXT_MAX_BLOCK; | |
1236 | } | |
1237 | ||
1238 | /* | |
d0d856e8 RD |
1239 | * ext4_ext_correct_indexes: |
1240 | * if leaf gets modified and modified extent is first in the leaf, | |
1241 | * then we have to correct all indexes above. | |
a86c6181 AT |
1242 | * TODO: do we need to correct tree in all cases? |
1243 | */ | |
1d03ec98 | 1244 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1245 | struct ext4_ext_path *path) |
1246 | { | |
1247 | struct ext4_extent_header *eh; | |
1248 | int depth = ext_depth(inode); | |
1249 | struct ext4_extent *ex; | |
1250 | __le32 border; | |
1251 | int k, err = 0; | |
1252 | ||
1253 | eh = path[depth].p_hdr; | |
1254 | ex = path[depth].p_ext; | |
1255 | BUG_ON(ex == NULL); | |
1256 | BUG_ON(eh == NULL); | |
1257 | ||
1258 | if (depth == 0) { | |
1259 | /* there is no tree at all */ | |
1260 | return 0; | |
1261 | } | |
1262 | ||
1263 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1264 | /* we correct tree if first leaf got modified only */ | |
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | /* | |
d0d856e8 | 1269 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1270 | */ |
1271 | k = depth - 1; | |
1272 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1273 | err = ext4_ext_get_access(handle, inode, path + k); |
1274 | if (err) | |
a86c6181 AT |
1275 | return err; |
1276 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1277 | err = ext4_ext_dirty(handle, inode, path + k); |
1278 | if (err) | |
a86c6181 AT |
1279 | return err; |
1280 | ||
1281 | while (k--) { | |
1282 | /* change all left-side indexes */ | |
1283 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1284 | break; | |
7e028976 AM |
1285 | err = ext4_ext_get_access(handle, inode, path + k); |
1286 | if (err) | |
a86c6181 AT |
1287 | break; |
1288 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1289 | err = ext4_ext_dirty(handle, inode, path + k); |
1290 | if (err) | |
a86c6181 AT |
1291 | break; |
1292 | } | |
1293 | ||
1294 | return err; | |
1295 | } | |
1296 | ||
09b88252 | 1297 | static int |
a86c6181 AT |
1298 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1299 | struct ext4_extent *ex2) | |
1300 | { | |
749269fa | 1301 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1302 | |
1303 | /* | |
1304 | * Make sure that either both extents are uninitialized, or | |
1305 | * both are _not_. | |
1306 | */ | |
1307 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | |
1308 | return 0; | |
1309 | ||
749269fa AA |
1310 | if (ext4_ext_is_uninitialized(ex1)) |
1311 | max_len = EXT_UNINIT_MAX_LEN; | |
1312 | else | |
1313 | max_len = EXT_INIT_MAX_LEN; | |
1314 | ||
a2df2a63 AA |
1315 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1316 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1317 | ||
1318 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1319 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1320 | return 0; |
1321 | ||
471d4011 SB |
1322 | /* |
1323 | * To allow future support for preallocated extents to be added | |
1324 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1325 | * this can result in the top bit of ee_len being set. |
471d4011 | 1326 | */ |
749269fa | 1327 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1328 | return 0; |
bbf2f9fb | 1329 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1330 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1331 | return 0; |
1332 | #endif | |
1333 | ||
a2df2a63 | 1334 | if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) |
a86c6181 AT |
1335 | return 1; |
1336 | return 0; | |
1337 | } | |
1338 | ||
56055d3a AA |
1339 | /* |
1340 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1341 | * It always tries to merge towards right. If you want to merge towards | |
1342 | * left, pass "ex - 1" as argument instead of "ex". | |
1343 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1344 | * 1 if they got merged. | |
1345 | */ | |
1346 | int ext4_ext_try_to_merge(struct inode *inode, | |
1347 | struct ext4_ext_path *path, | |
1348 | struct ext4_extent *ex) | |
1349 | { | |
1350 | struct ext4_extent_header *eh; | |
1351 | unsigned int depth, len; | |
1352 | int merge_done = 0; | |
1353 | int uninitialized = 0; | |
1354 | ||
1355 | depth = ext_depth(inode); | |
1356 | BUG_ON(path[depth].p_hdr == NULL); | |
1357 | eh = path[depth].p_hdr; | |
1358 | ||
1359 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1360 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1361 | break; | |
1362 | /* merge with next extent! */ | |
1363 | if (ext4_ext_is_uninitialized(ex)) | |
1364 | uninitialized = 1; | |
1365 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1366 | + ext4_ext_get_actual_len(ex + 1)); | |
1367 | if (uninitialized) | |
1368 | ext4_ext_mark_uninitialized(ex); | |
1369 | ||
1370 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1371 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1372 | * sizeof(struct ext4_extent); | |
1373 | memmove(ex + 1, ex + 2, len); | |
1374 | } | |
e8546d06 | 1375 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1376 | merge_done = 1; |
1377 | WARN_ON(eh->eh_entries == 0); | |
1378 | if (!eh->eh_entries) | |
1379 | ext4_error(inode->i_sb, "ext4_ext_try_to_merge", | |
1380 | "inode#%lu, eh->eh_entries = 0!", inode->i_ino); | |
1381 | } | |
1382 | ||
1383 | return merge_done; | |
1384 | } | |
1385 | ||
25d14f98 AA |
1386 | /* |
1387 | * check if a portion of the "newext" extent overlaps with an | |
1388 | * existing extent. | |
1389 | * | |
1390 | * If there is an overlap discovered, it updates the length of the newext | |
1391 | * such that there will be no overlap, and then returns 1. | |
1392 | * If there is no overlap found, it returns 0. | |
1393 | */ | |
1394 | unsigned int ext4_ext_check_overlap(struct inode *inode, | |
1395 | struct ext4_extent *newext, | |
1396 | struct ext4_ext_path *path) | |
1397 | { | |
725d26d3 | 1398 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1399 | unsigned int depth, len1; |
1400 | unsigned int ret = 0; | |
1401 | ||
1402 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1403 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1404 | depth = ext_depth(inode); |
1405 | if (!path[depth].p_ext) | |
1406 | goto out; | |
1407 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
1408 | ||
1409 | /* | |
1410 | * get the next allocated block if the extent in the path | |
1411 | * is before the requested block(s) | |
1412 | */ | |
1413 | if (b2 < b1) { | |
1414 | b2 = ext4_ext_next_allocated_block(path); | |
1415 | if (b2 == EXT_MAX_BLOCK) | |
1416 | goto out; | |
1417 | } | |
1418 | ||
725d26d3 | 1419 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 AA |
1420 | if (b1 + len1 < b1) { |
1421 | len1 = EXT_MAX_BLOCK - b1; | |
1422 | newext->ee_len = cpu_to_le16(len1); | |
1423 | ret = 1; | |
1424 | } | |
1425 | ||
1426 | /* check for overlap */ | |
1427 | if (b1 + len1 > b2) { | |
1428 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1429 | ret = 1; | |
1430 | } | |
1431 | out: | |
1432 | return ret; | |
1433 | } | |
1434 | ||
a86c6181 | 1435 | /* |
d0d856e8 RD |
1436 | * ext4_ext_insert_extent: |
1437 | * tries to merge requsted extent into the existing extent or | |
1438 | * inserts requested extent as new one into the tree, | |
1439 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1440 | */ |
1441 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1442 | struct ext4_ext_path *path, | |
1443 | struct ext4_extent *newext) | |
1444 | { | |
1445 | struct ext4_extent_header * eh; | |
1446 | struct ext4_extent *ex, *fex; | |
1447 | struct ext4_extent *nearex; /* nearest extent */ | |
1448 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1449 | int depth, len, err; |
1450 | ext4_lblk_t next; | |
a2df2a63 | 1451 | unsigned uninitialized = 0; |
a86c6181 | 1452 | |
a2df2a63 | 1453 | BUG_ON(ext4_ext_get_actual_len(newext) == 0); |
a86c6181 AT |
1454 | depth = ext_depth(inode); |
1455 | ex = path[depth].p_ext; | |
1456 | BUG_ON(path[depth].p_hdr == NULL); | |
1457 | ||
1458 | /* try to insert block into found extent and return */ | |
1459 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | |
2ae02107 | 1460 | ext_debug("append %d block to %d:%d (from %llu)\n", |
a2df2a63 | 1461 | ext4_ext_get_actual_len(newext), |
a86c6181 | 1462 | le32_to_cpu(ex->ee_block), |
a2df2a63 | 1463 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
7e028976 AM |
1464 | err = ext4_ext_get_access(handle, inode, path + depth); |
1465 | if (err) | |
a86c6181 | 1466 | return err; |
a2df2a63 AA |
1467 | |
1468 | /* | |
1469 | * ext4_can_extents_be_merged should have checked that either | |
1470 | * both extents are uninitialized, or both aren't. Thus we | |
1471 | * need to check only one of them here. | |
1472 | */ | |
1473 | if (ext4_ext_is_uninitialized(ex)) | |
1474 | uninitialized = 1; | |
1475 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1476 | + ext4_ext_get_actual_len(newext)); | |
1477 | if (uninitialized) | |
1478 | ext4_ext_mark_uninitialized(ex); | |
a86c6181 AT |
1479 | eh = path[depth].p_hdr; |
1480 | nearex = ex; | |
1481 | goto merge; | |
1482 | } | |
1483 | ||
1484 | repeat: | |
1485 | depth = ext_depth(inode); | |
1486 | eh = path[depth].p_hdr; | |
1487 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1488 | goto has_space; | |
1489 | ||
1490 | /* probably next leaf has space for us? */ | |
1491 | fex = EXT_LAST_EXTENT(eh); | |
1492 | next = ext4_ext_next_leaf_block(inode, path); | |
1493 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block) | |
1494 | && next != EXT_MAX_BLOCK) { | |
1495 | ext_debug("next leaf block - %d\n", next); | |
1496 | BUG_ON(npath != NULL); | |
1497 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1498 | if (IS_ERR(npath)) | |
1499 | return PTR_ERR(npath); | |
1500 | BUG_ON(npath->p_depth != path->p_depth); | |
1501 | eh = npath[depth].p_hdr; | |
1502 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
1503 | ext_debug("next leaf isnt full(%d)\n", | |
1504 | le16_to_cpu(eh->eh_entries)); | |
1505 | path = npath; | |
1506 | goto repeat; | |
1507 | } | |
1508 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1509 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1510 | } | |
1511 | ||
1512 | /* | |
d0d856e8 RD |
1513 | * There is no free space in the found leaf. |
1514 | * We're gonna add a new leaf in the tree. | |
a86c6181 AT |
1515 | */ |
1516 | err = ext4_ext_create_new_leaf(handle, inode, path, newext); | |
1517 | if (err) | |
1518 | goto cleanup; | |
1519 | depth = ext_depth(inode); | |
1520 | eh = path[depth].p_hdr; | |
1521 | ||
1522 | has_space: | |
1523 | nearex = path[depth].p_ext; | |
1524 | ||
7e028976 AM |
1525 | err = ext4_ext_get_access(handle, inode, path + depth); |
1526 | if (err) | |
a86c6181 AT |
1527 | goto cleanup; |
1528 | ||
1529 | if (!nearex) { | |
1530 | /* there is no extent in this leaf, create first one */ | |
2ae02107 | 1531 | ext_debug("first extent in the leaf: %d:%llu:%d\n", |
8c55e204 DK |
1532 | le32_to_cpu(newext->ee_block), |
1533 | ext_pblock(newext), | |
a2df2a63 | 1534 | ext4_ext_get_actual_len(newext)); |
a86c6181 AT |
1535 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1536 | } else if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1537 | > le32_to_cpu(nearex->ee_block)) { |
a86c6181 AT |
1538 | /* BUG_ON(newext->ee_block == nearex->ee_block); */ |
1539 | if (nearex != EXT_LAST_EXTENT(eh)) { | |
1540 | len = EXT_MAX_EXTENT(eh) - nearex; | |
1541 | len = (len - 1) * sizeof(struct ext4_extent); | |
1542 | len = len < 0 ? 0 : len; | |
2ae02107 | 1543 | ext_debug("insert %d:%llu:%d after: nearest 0x%p, " |
a86c6181 | 1544 | "move %d from 0x%p to 0x%p\n", |
8c55e204 DK |
1545 | le32_to_cpu(newext->ee_block), |
1546 | ext_pblock(newext), | |
a2df2a63 | 1547 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1548 | nearex, len, nearex + 1, nearex + 2); |
1549 | memmove(nearex + 2, nearex + 1, len); | |
1550 | } | |
1551 | path[depth].p_ext = nearex + 1; | |
1552 | } else { | |
1553 | BUG_ON(newext->ee_block == nearex->ee_block); | |
1554 | len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent); | |
1555 | len = len < 0 ? 0 : len; | |
2ae02107 | 1556 | ext_debug("insert %d:%llu:%d before: nearest 0x%p, " |
a86c6181 AT |
1557 | "move %d from 0x%p to 0x%p\n", |
1558 | le32_to_cpu(newext->ee_block), | |
f65e6fba | 1559 | ext_pblock(newext), |
a2df2a63 | 1560 | ext4_ext_get_actual_len(newext), |
a86c6181 AT |
1561 | nearex, len, nearex + 1, nearex + 2); |
1562 | memmove(nearex + 1, nearex, len); | |
1563 | path[depth].p_ext = nearex; | |
1564 | } | |
1565 | ||
e8546d06 | 1566 | le16_add_cpu(&eh->eh_entries, 1); |
a86c6181 AT |
1567 | nearex = path[depth].p_ext; |
1568 | nearex->ee_block = newext->ee_block; | |
b377611d | 1569 | ext4_ext_store_pblock(nearex, ext_pblock(newext)); |
a86c6181 | 1570 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
1571 | |
1572 | merge: | |
1573 | /* try to merge extents to the right */ | |
56055d3a | 1574 | ext4_ext_try_to_merge(inode, path, nearex); |
a86c6181 AT |
1575 | |
1576 | /* try to merge extents to the left */ | |
1577 | ||
1578 | /* time to correct all indexes above */ | |
1579 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1580 | if (err) | |
1581 | goto cleanup; | |
1582 | ||
1583 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1584 | ||
1585 | cleanup: | |
1586 | if (npath) { | |
1587 | ext4_ext_drop_refs(npath); | |
1588 | kfree(npath); | |
1589 | } | |
1590 | ext4_ext_tree_changed(inode); | |
1591 | ext4_ext_invalidate_cache(inode); | |
1592 | return err; | |
1593 | } | |
1594 | ||
09b88252 | 1595 | static void |
725d26d3 | 1596 | ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block, |
dd54567a | 1597 | __u32 len, ext4_fsblk_t start, int type) |
a86c6181 AT |
1598 | { |
1599 | struct ext4_ext_cache *cex; | |
1600 | BUG_ON(len == 0); | |
1601 | cex = &EXT4_I(inode)->i_cached_extent; | |
1602 | cex->ec_type = type; | |
1603 | cex->ec_block = block; | |
1604 | cex->ec_len = len; | |
1605 | cex->ec_start = start; | |
1606 | } | |
1607 | ||
1608 | /* | |
d0d856e8 RD |
1609 | * ext4_ext_put_gap_in_cache: |
1610 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
1611 | * and cache this gap |
1612 | */ | |
09b88252 | 1613 | static void |
a86c6181 | 1614 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 1615 | ext4_lblk_t block) |
a86c6181 AT |
1616 | { |
1617 | int depth = ext_depth(inode); | |
725d26d3 AK |
1618 | unsigned long len; |
1619 | ext4_lblk_t lblock; | |
a86c6181 AT |
1620 | struct ext4_extent *ex; |
1621 | ||
1622 | ex = path[depth].p_ext; | |
1623 | if (ex == NULL) { | |
1624 | /* there is no extent yet, so gap is [0;-] */ | |
1625 | lblock = 0; | |
1626 | len = EXT_MAX_BLOCK; | |
1627 | ext_debug("cache gap(whole file):"); | |
1628 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
1629 | lblock = block; | |
1630 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
1631 | ext_debug("cache gap(before): %u [%u:%u]", |
1632 | block, | |
1633 | le32_to_cpu(ex->ee_block), | |
1634 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 1635 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 1636 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 1637 | ext4_lblk_t next; |
8c55e204 | 1638 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 1639 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
1640 | |
1641 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
1642 | ext_debug("cache gap(after): [%u:%u] %u", |
1643 | le32_to_cpu(ex->ee_block), | |
1644 | ext4_ext_get_actual_len(ex), | |
1645 | block); | |
725d26d3 AK |
1646 | BUG_ON(next == lblock); |
1647 | len = next - lblock; | |
a86c6181 AT |
1648 | } else { |
1649 | lblock = len = 0; | |
1650 | BUG(); | |
1651 | } | |
1652 | ||
bba90743 | 1653 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
1654 | ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP); |
1655 | } | |
1656 | ||
09b88252 | 1657 | static int |
725d26d3 | 1658 | ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block, |
a86c6181 AT |
1659 | struct ext4_extent *ex) |
1660 | { | |
1661 | struct ext4_ext_cache *cex; | |
1662 | ||
1663 | cex = &EXT4_I(inode)->i_cached_extent; | |
1664 | ||
1665 | /* has cache valid data? */ | |
1666 | if (cex->ec_type == EXT4_EXT_CACHE_NO) | |
1667 | return EXT4_EXT_CACHE_NO; | |
1668 | ||
1669 | BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP && | |
1670 | cex->ec_type != EXT4_EXT_CACHE_EXTENT); | |
1671 | if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) { | |
8c55e204 | 1672 | ex->ee_block = cpu_to_le32(cex->ec_block); |
f65e6fba | 1673 | ext4_ext_store_pblock(ex, cex->ec_start); |
8c55e204 | 1674 | ex->ee_len = cpu_to_le16(cex->ec_len); |
bba90743 ES |
1675 | ext_debug("%u cached by %u:%u:%llu\n", |
1676 | block, | |
1677 | cex->ec_block, cex->ec_len, cex->ec_start); | |
a86c6181 AT |
1678 | return cex->ec_type; |
1679 | } | |
1680 | ||
1681 | /* not in cache */ | |
1682 | return EXT4_EXT_CACHE_NO; | |
1683 | } | |
1684 | ||
1685 | /* | |
d0d856e8 RD |
1686 | * ext4_ext_rm_idx: |
1687 | * removes index from the index block. | |
1688 | * It's used in truncate case only, thus all requests are for | |
1689 | * last index in the block only. | |
a86c6181 | 1690 | */ |
1d03ec98 | 1691 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1692 | struct ext4_ext_path *path) |
1693 | { | |
1694 | struct buffer_head *bh; | |
1695 | int err; | |
f65e6fba | 1696 | ext4_fsblk_t leaf; |
a86c6181 AT |
1697 | |
1698 | /* free index block */ | |
1699 | path--; | |
f65e6fba | 1700 | leaf = idx_pblock(path->p_idx); |
a86c6181 | 1701 | BUG_ON(path->p_hdr->eh_entries == 0); |
7e028976 AM |
1702 | err = ext4_ext_get_access(handle, inode, path); |
1703 | if (err) | |
a86c6181 | 1704 | return err; |
e8546d06 | 1705 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
1706 | err = ext4_ext_dirty(handle, inode, path); |
1707 | if (err) | |
a86c6181 | 1708 | return err; |
2ae02107 | 1709 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
a86c6181 AT |
1710 | bh = sb_find_get_block(inode->i_sb, leaf); |
1711 | ext4_forget(handle, 1, inode, bh, leaf); | |
c9de560d | 1712 | ext4_free_blocks(handle, inode, leaf, 1, 1); |
a86c6181 AT |
1713 | return err; |
1714 | } | |
1715 | ||
1716 | /* | |
d0d856e8 RD |
1717 | * ext4_ext_calc_credits_for_insert: |
1718 | * This routine returns max. credits that the extent tree can consume. | |
a86c6181 | 1719 | * It should be OK for low-performance paths like ->writepage() |
d0d856e8 | 1720 | * To allow many writing processes to fit into a single transaction, |
0e855ac8 | 1721 | * the caller should calculate credits under i_data_sem and |
d0d856e8 | 1722 | * pass the actual path. |
a86c6181 | 1723 | */ |
09b88252 | 1724 | int ext4_ext_calc_credits_for_insert(struct inode *inode, |
a86c6181 AT |
1725 | struct ext4_ext_path *path) |
1726 | { | |
1727 | int depth, needed; | |
1728 | ||
1729 | if (path) { | |
1730 | /* probably there is space in leaf? */ | |
1731 | depth = ext_depth(inode); | |
1732 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) | |
1733 | < le16_to_cpu(path[depth].p_hdr->eh_max)) | |
1734 | return 1; | |
1735 | } | |
1736 | ||
1737 | /* | |
d0d856e8 | 1738 | * given 32-bit logical block (4294967296 blocks), max. tree |
a86c6181 | 1739 | * can be 4 levels in depth -- 4 * 340^4 == 53453440000. |
d0d856e8 | 1740 | * Let's also add one more level for imbalance. |
a86c6181 AT |
1741 | */ |
1742 | depth = 5; | |
1743 | ||
1744 | /* allocation of new data block(s) */ | |
1745 | needed = 2; | |
1746 | ||
1747 | /* | |
d0d856e8 | 1748 | * tree can be full, so it would need to grow in depth: |
feb18927 JL |
1749 | * we need one credit to modify old root, credits for |
1750 | * new root will be added in split accounting | |
a86c6181 | 1751 | */ |
feb18927 | 1752 | needed += 1; |
a86c6181 AT |
1753 | |
1754 | /* | |
d0d856e8 | 1755 | * Index split can happen, we would need: |
a86c6181 AT |
1756 | * allocate intermediate indexes (bitmap + group) |
1757 | * + change two blocks at each level, but root (already included) | |
1758 | */ | |
feb18927 | 1759 | needed += (depth * 2) + (depth * 2); |
a86c6181 AT |
1760 | |
1761 | /* any allocation modifies superblock */ | |
1762 | needed += 1; | |
1763 | ||
1764 | return needed; | |
1765 | } | |
1766 | ||
1767 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
1768 | struct ext4_extent *ex, | |
725d26d3 | 1769 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 AT |
1770 | { |
1771 | struct buffer_head *bh; | |
a2df2a63 | 1772 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
c9de560d | 1773 | int i, metadata = 0; |
a86c6181 | 1774 | |
c9de560d AT |
1775 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
1776 | metadata = 1; | |
a86c6181 AT |
1777 | #ifdef EXTENTS_STATS |
1778 | { | |
1779 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
1780 | spin_lock(&sbi->s_ext_stats_lock); |
1781 | sbi->s_ext_blocks += ee_len; | |
1782 | sbi->s_ext_extents++; | |
1783 | if (ee_len < sbi->s_ext_min) | |
1784 | sbi->s_ext_min = ee_len; | |
1785 | if (ee_len > sbi->s_ext_max) | |
1786 | sbi->s_ext_max = ee_len; | |
1787 | if (ext_depth(inode) > sbi->s_depth_max) | |
1788 | sbi->s_depth_max = ext_depth(inode); | |
1789 | spin_unlock(&sbi->s_ext_stats_lock); | |
1790 | } | |
1791 | #endif | |
1792 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 1793 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 1794 | /* tail removal */ |
725d26d3 | 1795 | ext4_lblk_t num; |
f65e6fba | 1796 | ext4_fsblk_t start; |
725d26d3 | 1797 | |
a2df2a63 AA |
1798 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
1799 | start = ext_pblock(ex) + ee_len - num; | |
725d26d3 | 1800 | ext_debug("free last %u blocks starting %llu\n", num, start); |
a86c6181 AT |
1801 | for (i = 0; i < num; i++) { |
1802 | bh = sb_find_get_block(inode->i_sb, start + i); | |
1803 | ext4_forget(handle, 0, inode, bh, start + i); | |
1804 | } | |
c9de560d | 1805 | ext4_free_blocks(handle, inode, start, num, metadata); |
a86c6181 | 1806 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 1807 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
725d26d3 | 1808 | printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n", |
a2df2a63 | 1809 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
a86c6181 | 1810 | } else { |
725d26d3 AK |
1811 | printk(KERN_INFO "strange request: removal(2) " |
1812 | "%u-%u from %u:%u\n", | |
1813 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
1814 | } |
1815 | return 0; | |
1816 | } | |
1817 | ||
1818 | static int | |
1819 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
725d26d3 | 1820 | struct ext4_ext_path *path, ext4_lblk_t start) |
a86c6181 AT |
1821 | { |
1822 | int err = 0, correct_index = 0; | |
1823 | int depth = ext_depth(inode), credits; | |
1824 | struct ext4_extent_header *eh; | |
725d26d3 AK |
1825 | ext4_lblk_t a, b, block; |
1826 | unsigned num; | |
1827 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 1828 | unsigned short ex_ee_len; |
a2df2a63 | 1829 | unsigned uninitialized = 0; |
a86c6181 AT |
1830 | struct ext4_extent *ex; |
1831 | ||
c29c0ae7 | 1832 | /* the header must be checked already in ext4_ext_remove_space() */ |
725d26d3 | 1833 | ext_debug("truncate since %u in leaf\n", start); |
a86c6181 AT |
1834 | if (!path[depth].p_hdr) |
1835 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
1836 | eh = path[depth].p_hdr; | |
1837 | BUG_ON(eh == NULL); | |
a86c6181 AT |
1838 | |
1839 | /* find where to start removing */ | |
1840 | ex = EXT_LAST_EXTENT(eh); | |
1841 | ||
1842 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 AA |
1843 | if (ext4_ext_is_uninitialized(ex)) |
1844 | uninitialized = 1; | |
1845 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
a86c6181 AT |
1846 | |
1847 | while (ex >= EXT_FIRST_EXTENT(eh) && | |
1848 | ex_ee_block + ex_ee_len > start) { | |
1849 | ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len); | |
1850 | path[depth].p_ext = ex; | |
1851 | ||
1852 | a = ex_ee_block > start ? ex_ee_block : start; | |
1853 | b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ? | |
1854 | ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK; | |
1855 | ||
1856 | ext_debug(" border %u:%u\n", a, b); | |
1857 | ||
1858 | if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) { | |
1859 | block = 0; | |
1860 | num = 0; | |
1861 | BUG(); | |
1862 | } else if (a != ex_ee_block) { | |
1863 | /* remove tail of the extent */ | |
1864 | block = ex_ee_block; | |
1865 | num = a - block; | |
1866 | } else if (b != ex_ee_block + ex_ee_len - 1) { | |
1867 | /* remove head of the extent */ | |
1868 | block = a; | |
1869 | num = b - a; | |
1870 | /* there is no "make a hole" API yet */ | |
1871 | BUG(); | |
1872 | } else { | |
1873 | /* remove whole extent: excellent! */ | |
1874 | block = ex_ee_block; | |
1875 | num = 0; | |
1876 | BUG_ON(a != ex_ee_block); | |
1877 | BUG_ON(b != ex_ee_block + ex_ee_len - 1); | |
1878 | } | |
1879 | ||
d0d856e8 | 1880 | /* at present, extent can't cross block group: */ |
a86c6181 AT |
1881 | /* leaf + bitmap + group desc + sb + inode */ |
1882 | credits = 5; | |
1883 | if (ex == EXT_FIRST_EXTENT(eh)) { | |
1884 | correct_index = 1; | |
1885 | credits += (ext_depth(inode)) + 1; | |
1886 | } | |
1887 | #ifdef CONFIG_QUOTA | |
1888 | credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); | |
1889 | #endif | |
1890 | ||
1891 | handle = ext4_ext_journal_restart(handle, credits); | |
1892 | if (IS_ERR(handle)) { | |
1893 | err = PTR_ERR(handle); | |
1894 | goto out; | |
1895 | } | |
1896 | ||
1897 | err = ext4_ext_get_access(handle, inode, path + depth); | |
1898 | if (err) | |
1899 | goto out; | |
1900 | ||
1901 | err = ext4_remove_blocks(handle, inode, ex, a, b); | |
1902 | if (err) | |
1903 | goto out; | |
1904 | ||
1905 | if (num == 0) { | |
d0d856e8 | 1906 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 1907 | ext4_ext_store_pblock(ex, 0); |
e8546d06 | 1908 | le16_add_cpu(&eh->eh_entries, -1); |
a86c6181 AT |
1909 | } |
1910 | ||
1911 | ex->ee_block = cpu_to_le32(block); | |
1912 | ex->ee_len = cpu_to_le16(num); | |
749269fa AA |
1913 | /* |
1914 | * Do not mark uninitialized if all the blocks in the | |
1915 | * extent have been removed. | |
1916 | */ | |
1917 | if (uninitialized && num) | |
a2df2a63 | 1918 | ext4_ext_mark_uninitialized(ex); |
a86c6181 AT |
1919 | |
1920 | err = ext4_ext_dirty(handle, inode, path + depth); | |
1921 | if (err) | |
1922 | goto out; | |
1923 | ||
2ae02107 | 1924 | ext_debug("new extent: %u:%u:%llu\n", block, num, |
f65e6fba | 1925 | ext_pblock(ex)); |
a86c6181 AT |
1926 | ex--; |
1927 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 1928 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
1929 | } |
1930 | ||
1931 | if (correct_index && eh->eh_entries) | |
1932 | err = ext4_ext_correct_indexes(handle, inode, path); | |
1933 | ||
1934 | /* if this leaf is free, then we should | |
1935 | * remove it from index block above */ | |
1936 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
1937 | err = ext4_ext_rm_idx(handle, inode, path + depth); | |
1938 | ||
1939 | out: | |
1940 | return err; | |
1941 | } | |
1942 | ||
1943 | /* | |
d0d856e8 RD |
1944 | * ext4_ext_more_to_rm: |
1945 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 1946 | */ |
09b88252 | 1947 | static int |
a86c6181 AT |
1948 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
1949 | { | |
1950 | BUG_ON(path->p_idx == NULL); | |
1951 | ||
1952 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
1953 | return 0; | |
1954 | ||
1955 | /* | |
d0d856e8 | 1956 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
1957 | * so we have to consider current index for truncation |
1958 | */ | |
1959 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
1960 | return 0; | |
1961 | return 1; | |
1962 | } | |
1963 | ||
1d03ec98 | 1964 | static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start) |
a86c6181 AT |
1965 | { |
1966 | struct super_block *sb = inode->i_sb; | |
1967 | int depth = ext_depth(inode); | |
1968 | struct ext4_ext_path *path; | |
1969 | handle_t *handle; | |
1970 | int i = 0, err = 0; | |
1971 | ||
725d26d3 | 1972 | ext_debug("truncate since %u\n", start); |
a86c6181 AT |
1973 | |
1974 | /* probably first extent we're gonna free will be last in block */ | |
1975 | handle = ext4_journal_start(inode, depth + 1); | |
1976 | if (IS_ERR(handle)) | |
1977 | return PTR_ERR(handle); | |
1978 | ||
1979 | ext4_ext_invalidate_cache(inode); | |
1980 | ||
1981 | /* | |
d0d856e8 RD |
1982 | * We start scanning from right side, freeing all the blocks |
1983 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 1984 | */ |
216553c4 | 1985 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS); |
a86c6181 AT |
1986 | if (path == NULL) { |
1987 | ext4_journal_stop(handle); | |
1988 | return -ENOMEM; | |
1989 | } | |
a86c6181 | 1990 | path[0].p_hdr = ext_inode_hdr(inode); |
c29c0ae7 | 1991 | if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) { |
a86c6181 AT |
1992 | err = -EIO; |
1993 | goto out; | |
1994 | } | |
1995 | path[0].p_depth = depth; | |
1996 | ||
1997 | while (i >= 0 && err == 0) { | |
1998 | if (i == depth) { | |
1999 | /* this is leaf block */ | |
2000 | err = ext4_ext_rm_leaf(handle, inode, path, start); | |
d0d856e8 | 2001 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2002 | brelse(path[i].p_bh); |
2003 | path[i].p_bh = NULL; | |
2004 | i--; | |
2005 | continue; | |
2006 | } | |
2007 | ||
2008 | /* this is index block */ | |
2009 | if (!path[i].p_hdr) { | |
2010 | ext_debug("initialize header\n"); | |
2011 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2012 | } |
2013 | ||
a86c6181 | 2014 | if (!path[i].p_idx) { |
d0d856e8 | 2015 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2016 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2017 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2018 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2019 | path[i].p_hdr, | |
2020 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2021 | } else { | |
d0d856e8 | 2022 | /* we were already here, see at next index */ |
a86c6181 AT |
2023 | path[i].p_idx--; |
2024 | } | |
2025 | ||
2026 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2027 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2028 | path[i].p_idx); | |
2029 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2030 | struct buffer_head *bh; |
a86c6181 | 2031 | /* go to the next level */ |
2ae02107 | 2032 | ext_debug("move to level %d (block %llu)\n", |
f65e6fba | 2033 | i + 1, idx_pblock(path[i].p_idx)); |
a86c6181 | 2034 | memset(path + i + 1, 0, sizeof(*path)); |
c29c0ae7 AT |
2035 | bh = sb_bread(sb, idx_pblock(path[i].p_idx)); |
2036 | if (!bh) { | |
a86c6181 AT |
2037 | /* should we reset i_size? */ |
2038 | err = -EIO; | |
2039 | break; | |
2040 | } | |
c29c0ae7 AT |
2041 | if (WARN_ON(i + 1 > depth)) { |
2042 | err = -EIO; | |
2043 | break; | |
2044 | } | |
2045 | if (ext4_ext_check_header(inode, ext_block_hdr(bh), | |
2046 | depth - i - 1)) { | |
2047 | err = -EIO; | |
2048 | break; | |
2049 | } | |
2050 | path[i + 1].p_bh = bh; | |
a86c6181 | 2051 | |
d0d856e8 RD |
2052 | /* save actual number of indexes since this |
2053 | * number is changed at the next iteration */ | |
a86c6181 AT |
2054 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2055 | i++; | |
2056 | } else { | |
d0d856e8 | 2057 | /* we finished processing this index, go up */ |
a86c6181 | 2058 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2059 | /* index is empty, remove it; |
a86c6181 AT |
2060 | * handle must be already prepared by the |
2061 | * truncatei_leaf() */ | |
2062 | err = ext4_ext_rm_idx(handle, inode, path + i); | |
2063 | } | |
d0d856e8 | 2064 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2065 | brelse(path[i].p_bh); |
2066 | path[i].p_bh = NULL; | |
2067 | i--; | |
2068 | ext_debug("return to level %d\n", i); | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | /* TODO: flexible tree reduction should be here */ | |
2073 | if (path->p_hdr->eh_entries == 0) { | |
2074 | /* | |
d0d856e8 RD |
2075 | * truncate to zero freed all the tree, |
2076 | * so we need to correct eh_depth | |
a86c6181 AT |
2077 | */ |
2078 | err = ext4_ext_get_access(handle, inode, path); | |
2079 | if (err == 0) { | |
2080 | ext_inode_hdr(inode)->eh_depth = 0; | |
2081 | ext_inode_hdr(inode)->eh_max = | |
2082 | cpu_to_le16(ext4_ext_space_root(inode)); | |
2083 | err = ext4_ext_dirty(handle, inode, path); | |
2084 | } | |
2085 | } | |
2086 | out: | |
2087 | ext4_ext_tree_changed(inode); | |
2088 | ext4_ext_drop_refs(path); | |
2089 | kfree(path); | |
2090 | ext4_journal_stop(handle); | |
2091 | ||
2092 | return err; | |
2093 | } | |
2094 | ||
2095 | /* | |
2096 | * called at mount time | |
2097 | */ | |
2098 | void ext4_ext_init(struct super_block *sb) | |
2099 | { | |
2100 | /* | |
2101 | * possible initialization would be here | |
2102 | */ | |
2103 | ||
2104 | if (test_opt(sb, EXTENTS)) { | |
2105 | printk("EXT4-fs: file extents enabled"); | |
bbf2f9fb RD |
2106 | #ifdef AGGRESSIVE_TEST |
2107 | printk(", aggressive tests"); | |
a86c6181 AT |
2108 | #endif |
2109 | #ifdef CHECK_BINSEARCH | |
2110 | printk(", check binsearch"); | |
2111 | #endif | |
2112 | #ifdef EXTENTS_STATS | |
2113 | printk(", stats"); | |
2114 | #endif | |
2115 | printk("\n"); | |
2116 | #ifdef EXTENTS_STATS | |
2117 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2118 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2119 | EXT4_SB(sb)->s_ext_max = 0; | |
2120 | #endif | |
2121 | } | |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * called at umount time | |
2126 | */ | |
2127 | void ext4_ext_release(struct super_block *sb) | |
2128 | { | |
2129 | if (!test_opt(sb, EXTENTS)) | |
2130 | return; | |
2131 | ||
2132 | #ifdef EXTENTS_STATS | |
2133 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2134 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2135 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2136 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2137 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2138 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2139 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2140 | } | |
2141 | #endif | |
2142 | } | |
2143 | ||
093a088b AK |
2144 | static void bi_complete(struct bio *bio, int error) |
2145 | { | |
2146 | complete((struct completion *)bio->bi_private); | |
2147 | } | |
2148 | ||
2149 | /* FIXME!! we need to try to merge to left or right after zero-out */ | |
2150 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2151 | { | |
2152 | int ret = -EIO; | |
2153 | struct bio *bio; | |
2154 | int blkbits, blocksize; | |
2155 | sector_t ee_pblock; | |
2156 | struct completion event; | |
2157 | unsigned int ee_len, len, done, offset; | |
2158 | ||
2159 | ||
2160 | blkbits = inode->i_blkbits; | |
2161 | blocksize = inode->i_sb->s_blocksize; | |
2162 | ee_len = ext4_ext_get_actual_len(ex); | |
2163 | ee_pblock = ext_pblock(ex); | |
2164 | ||
2165 | /* convert ee_pblock to 512 byte sectors */ | |
2166 | ee_pblock = ee_pblock << (blkbits - 9); | |
2167 | ||
2168 | while (ee_len > 0) { | |
2169 | ||
2170 | if (ee_len > BIO_MAX_PAGES) | |
2171 | len = BIO_MAX_PAGES; | |
2172 | else | |
2173 | len = ee_len; | |
2174 | ||
2175 | bio = bio_alloc(GFP_NOIO, len); | |
2176 | if (!bio) | |
2177 | return -ENOMEM; | |
2178 | bio->bi_sector = ee_pblock; | |
2179 | bio->bi_bdev = inode->i_sb->s_bdev; | |
2180 | ||
2181 | done = 0; | |
2182 | offset = 0; | |
2183 | while (done < len) { | |
2184 | ret = bio_add_page(bio, ZERO_PAGE(0), | |
2185 | blocksize, offset); | |
2186 | if (ret != blocksize) { | |
2187 | /* | |
2188 | * We can't add any more pages because of | |
2189 | * hardware limitations. Start a new bio. | |
2190 | */ | |
2191 | break; | |
2192 | } | |
2193 | done++; | |
2194 | offset += blocksize; | |
2195 | if (offset >= PAGE_CACHE_SIZE) | |
2196 | offset = 0; | |
2197 | } | |
2198 | ||
2199 | init_completion(&event); | |
2200 | bio->bi_private = &event; | |
2201 | bio->bi_end_io = bi_complete; | |
2202 | submit_bio(WRITE, bio); | |
2203 | wait_for_completion(&event); | |
2204 | ||
2205 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
2206 | ret = 0; | |
2207 | else { | |
2208 | ret = -EIO; | |
2209 | break; | |
2210 | } | |
2211 | bio_put(bio); | |
2212 | ee_len -= done; | |
2213 | ee_pblock += done << (blkbits - 9); | |
2214 | } | |
2215 | return ret; | |
2216 | } | |
2217 | ||
3977c965 AK |
2218 | #define EXT4_EXT_ZERO_LEN 7 |
2219 | ||
56055d3a AA |
2220 | /* |
2221 | * This function is called by ext4_ext_get_blocks() if someone tries to write | |
2222 | * to an uninitialized extent. It may result in splitting the uninitialized | |
2223 | * extent into multiple extents (upto three - one initialized and two | |
2224 | * uninitialized). | |
2225 | * There are three possibilities: | |
2226 | * a> There is no split required: Entire extent should be initialized | |
2227 | * b> Splits in two extents: Write is happening at either end of the extent | |
2228 | * c> Splits in three extents: Somone is writing in middle of the extent | |
2229 | */ | |
725d26d3 AK |
2230 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
2231 | struct inode *inode, | |
2232 | struct ext4_ext_path *path, | |
2233 | ext4_lblk_t iblock, | |
2234 | unsigned long max_blocks) | |
56055d3a | 2235 | { |
95c3889c | 2236 | struct ext4_extent *ex, newex, orig_ex; |
56055d3a AA |
2237 | struct ext4_extent *ex1 = NULL; |
2238 | struct ext4_extent *ex2 = NULL; | |
2239 | struct ext4_extent *ex3 = NULL; | |
2240 | struct ext4_extent_header *eh; | |
725d26d3 AK |
2241 | ext4_lblk_t ee_block; |
2242 | unsigned int allocated, ee_len, depth; | |
56055d3a AA |
2243 | ext4_fsblk_t newblock; |
2244 | int err = 0; | |
2245 | int ret = 0; | |
2246 | ||
2247 | depth = ext_depth(inode); | |
2248 | eh = path[depth].p_hdr; | |
2249 | ex = path[depth].p_ext; | |
2250 | ee_block = le32_to_cpu(ex->ee_block); | |
2251 | ee_len = ext4_ext_get_actual_len(ex); | |
2252 | allocated = ee_len - (iblock - ee_block); | |
2253 | newblock = iblock - ee_block + ext_pblock(ex); | |
2254 | ex2 = ex; | |
95c3889c AK |
2255 | orig_ex.ee_block = ex->ee_block; |
2256 | orig_ex.ee_len = cpu_to_le16(ee_len); | |
2257 | ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); | |
56055d3a | 2258 | |
9df5643a AK |
2259 | err = ext4_ext_get_access(handle, inode, path + depth); |
2260 | if (err) | |
2261 | goto out; | |
3977c965 AK |
2262 | /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */ |
2263 | if (ee_len <= 2*EXT4_EXT_ZERO_LEN) { | |
2264 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2265 | if (err) | |
2266 | goto fix_extent_len; | |
2267 | /* update the extent length and mark as initialized */ | |
2268 | ex->ee_block = orig_ex.ee_block; | |
2269 | ex->ee_len = orig_ex.ee_len; | |
2270 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2271 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2272 | /* zeroed the full extent */ |
2273 | return allocated; | |
3977c965 | 2274 | } |
9df5643a | 2275 | |
56055d3a AA |
2276 | /* ex1: ee_block to iblock - 1 : uninitialized */ |
2277 | if (iblock > ee_block) { | |
2278 | ex1 = ex; | |
2279 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2280 | ext4_ext_mark_uninitialized(ex1); | |
2281 | ex2 = &newex; | |
2282 | } | |
2283 | /* | |
2284 | * for sanity, update the length of the ex2 extent before | |
2285 | * we insert ex3, if ex1 is NULL. This is to avoid temporary | |
2286 | * overlap of blocks. | |
2287 | */ | |
2288 | if (!ex1 && allocated > max_blocks) | |
2289 | ex2->ee_len = cpu_to_le16(max_blocks); | |
2290 | /* ex3: to ee_block + ee_len : uninitialised */ | |
2291 | if (allocated > max_blocks) { | |
2292 | unsigned int newdepth; | |
3977c965 AK |
2293 | /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */ |
2294 | if (allocated <= EXT4_EXT_ZERO_LEN) { | |
2295 | /* Mark first half uninitialized. | |
2296 | * Mark second half initialized and zero out the | |
2297 | * initialized extent | |
2298 | */ | |
2299 | ex->ee_block = orig_ex.ee_block; | |
2300 | ex->ee_len = cpu_to_le16(ee_len - allocated); | |
2301 | ext4_ext_mark_uninitialized(ex); | |
2302 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2303 | ext4_ext_dirty(handle, inode, path + depth); | |
2304 | ||
2305 | ex3 = &newex; | |
2306 | ex3->ee_block = cpu_to_le32(iblock); | |
2307 | ext4_ext_store_pblock(ex3, newblock); | |
2308 | ex3->ee_len = cpu_to_le16(allocated); | |
2309 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
2310 | if (err == -ENOSPC) { | |
2311 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2312 | if (err) | |
2313 | goto fix_extent_len; | |
2314 | ex->ee_block = orig_ex.ee_block; | |
2315 | ex->ee_len = orig_ex.ee_len; | |
2316 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2317 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2318 | /* zeroed the full extent */ |
2319 | return allocated; | |
3977c965 AK |
2320 | |
2321 | } else if (err) | |
2322 | goto fix_extent_len; | |
2323 | ||
161e7b7c AK |
2324 | /* |
2325 | * We need to zero out the second half because | |
2326 | * an fallocate request can update file size and | |
2327 | * converting the second half to initialized extent | |
2328 | * implies that we can leak some junk data to user | |
2329 | * space. | |
2330 | */ | |
2331 | err = ext4_ext_zeroout(inode, ex3); | |
2332 | if (err) { | |
2333 | /* | |
2334 | * We should actually mark the | |
2335 | * second half as uninit and return error | |
2336 | * Insert would have changed the extent | |
2337 | */ | |
2338 | depth = ext_depth(inode); | |
2339 | ext4_ext_drop_refs(path); | |
2340 | path = ext4_ext_find_extent(inode, | |
2341 | iblock, path); | |
2342 | if (IS_ERR(path)) { | |
2343 | err = PTR_ERR(path); | |
2344 | return err; | |
2345 | } | |
2346 | ex = path[depth].p_ext; | |
2347 | err = ext4_ext_get_access(handle, inode, | |
2348 | path + depth); | |
2349 | if (err) | |
2350 | return err; | |
2351 | ext4_ext_mark_uninitialized(ex); | |
2352 | ext4_ext_dirty(handle, inode, path + depth); | |
2353 | return err; | |
2354 | } | |
2355 | ||
2356 | /* zeroed the second half */ | |
3977c965 AK |
2357 | return allocated; |
2358 | } | |
56055d3a AA |
2359 | ex3 = &newex; |
2360 | ex3->ee_block = cpu_to_le32(iblock + max_blocks); | |
2361 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | |
2362 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | |
2363 | ext4_ext_mark_uninitialized(ex3); | |
2364 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | |
093a088b AK |
2365 | if (err == -ENOSPC) { |
2366 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2367 | if (err) | |
2368 | goto fix_extent_len; | |
2369 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2370 | ex->ee_block = orig_ex.ee_block; |
2371 | ex->ee_len = orig_ex.ee_len; | |
2372 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2373 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c AK |
2374 | /* zeroed the full extent */ |
2375 | return allocated; | |
093a088b AK |
2376 | |
2377 | } else if (err) | |
2378 | goto fix_extent_len; | |
56055d3a AA |
2379 | /* |
2380 | * The depth, and hence eh & ex might change | |
2381 | * as part of the insert above. | |
2382 | */ | |
2383 | newdepth = ext_depth(inode); | |
95c3889c AK |
2384 | /* |
2385 | * update the extent length after successfull insert of the | |
2386 | * split extent | |
2387 | */ | |
2388 | orig_ex.ee_len = cpu_to_le16(ee_len - | |
2389 | ext4_ext_get_actual_len(ex3)); | |
56055d3a AA |
2390 | if (newdepth != depth) { |
2391 | depth = newdepth; | |
b35905c1 AK |
2392 | ext4_ext_drop_refs(path); |
2393 | path = ext4_ext_find_extent(inode, iblock, path); | |
56055d3a AA |
2394 | if (IS_ERR(path)) { |
2395 | err = PTR_ERR(path); | |
56055d3a AA |
2396 | goto out; |
2397 | } | |
2398 | eh = path[depth].p_hdr; | |
2399 | ex = path[depth].p_ext; | |
2400 | if (ex2 != &newex) | |
2401 | ex2 = ex; | |
9df5643a AK |
2402 | |
2403 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2404 | if (err) | |
2405 | goto out; | |
56055d3a AA |
2406 | } |
2407 | allocated = max_blocks; | |
3977c965 AK |
2408 | |
2409 | /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying | |
2410 | * to insert a extent in the middle zerout directly | |
2411 | * otherwise give the extent a chance to merge to left | |
2412 | */ | |
2413 | if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN && | |
2414 | iblock != ee_block) { | |
2415 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2416 | if (err) | |
2417 | goto fix_extent_len; | |
2418 | /* update the extent length and mark as initialized */ | |
2419 | ex->ee_block = orig_ex.ee_block; | |
2420 | ex->ee_len = orig_ex.ee_len; | |
2421 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2422 | ext4_ext_dirty(handle, inode, path + depth); | |
161e7b7c AK |
2423 | /* zero out the first half */ |
2424 | return allocated; | |
3977c965 | 2425 | } |
56055d3a AA |
2426 | } |
2427 | /* | |
2428 | * If there was a change of depth as part of the | |
2429 | * insertion of ex3 above, we need to update the length | |
2430 | * of the ex1 extent again here | |
2431 | */ | |
2432 | if (ex1 && ex1 != ex) { | |
2433 | ex1 = ex; | |
2434 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | |
2435 | ext4_ext_mark_uninitialized(ex1); | |
2436 | ex2 = &newex; | |
2437 | } | |
2438 | /* ex2: iblock to iblock + maxblocks-1 : initialised */ | |
2439 | ex2->ee_block = cpu_to_le32(iblock); | |
56055d3a AA |
2440 | ext4_ext_store_pblock(ex2, newblock); |
2441 | ex2->ee_len = cpu_to_le16(allocated); | |
2442 | if (ex2 != ex) | |
2443 | goto insert; | |
56055d3a AA |
2444 | /* |
2445 | * New (initialized) extent starts from the first block | |
2446 | * in the current extent. i.e., ex2 == ex | |
2447 | * We have to see if it can be merged with the extent | |
2448 | * on the left. | |
2449 | */ | |
2450 | if (ex2 > EXT_FIRST_EXTENT(eh)) { | |
2451 | /* | |
2452 | * To merge left, pass "ex2 - 1" to try_to_merge(), | |
2453 | * since it merges towards right _only_. | |
2454 | */ | |
2455 | ret = ext4_ext_try_to_merge(inode, path, ex2 - 1); | |
2456 | if (ret) { | |
2457 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2458 | if (err) | |
2459 | goto out; | |
2460 | depth = ext_depth(inode); | |
2461 | ex2--; | |
2462 | } | |
2463 | } | |
2464 | /* | |
2465 | * Try to Merge towards right. This might be required | |
2466 | * only when the whole extent is being written to. | |
2467 | * i.e. ex2 == ex and ex3 == NULL. | |
2468 | */ | |
2469 | if (!ex3) { | |
2470 | ret = ext4_ext_try_to_merge(inode, path, ex2); | |
2471 | if (ret) { | |
2472 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2473 | if (err) | |
2474 | goto out; | |
2475 | } | |
2476 | } | |
2477 | /* Mark modified extent as dirty */ | |
2478 | err = ext4_ext_dirty(handle, inode, path + depth); | |
2479 | goto out; | |
2480 | insert: | |
2481 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | |
093a088b AK |
2482 | if (err == -ENOSPC) { |
2483 | err = ext4_ext_zeroout(inode, &orig_ex); | |
2484 | if (err) | |
2485 | goto fix_extent_len; | |
2486 | /* update the extent length and mark as initialized */ | |
95c3889c AK |
2487 | ex->ee_block = orig_ex.ee_block; |
2488 | ex->ee_len = orig_ex.ee_len; | |
2489 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
95c3889c | 2490 | ext4_ext_dirty(handle, inode, path + depth); |
161e7b7c AK |
2491 | /* zero out the first half */ |
2492 | return allocated; | |
093a088b AK |
2493 | } else if (err) |
2494 | goto fix_extent_len; | |
56055d3a AA |
2495 | out: |
2496 | return err ? err : allocated; | |
093a088b AK |
2497 | |
2498 | fix_extent_len: | |
2499 | ex->ee_block = orig_ex.ee_block; | |
2500 | ex->ee_len = orig_ex.ee_len; | |
2501 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | |
2502 | ext4_ext_mark_uninitialized(ex); | |
2503 | ext4_ext_dirty(handle, inode, path + depth); | |
2504 | return err; | |
56055d3a AA |
2505 | } |
2506 | ||
c278bfec | 2507 | /* |
f5ab0d1f MC |
2508 | * Block allocation/map/preallocation routine for extents based files |
2509 | * | |
2510 | * | |
c278bfec | 2511 | * Need to be called with |
0e855ac8 AK |
2512 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
2513 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
2514 | * |
2515 | * return > 0, number of of blocks already mapped/allocated | |
2516 | * if create == 0 and these are pre-allocated blocks | |
2517 | * buffer head is unmapped | |
2518 | * otherwise blocks are mapped | |
2519 | * | |
2520 | * return = 0, if plain look up failed (blocks have not been allocated) | |
2521 | * buffer head is unmapped | |
2522 | * | |
2523 | * return < 0, error case. | |
c278bfec | 2524 | */ |
f65e6fba | 2525 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, |
725d26d3 | 2526 | ext4_lblk_t iblock, |
a86c6181 AT |
2527 | unsigned long max_blocks, struct buffer_head *bh_result, |
2528 | int create, int extend_disksize) | |
2529 | { | |
2530 | struct ext4_ext_path *path = NULL; | |
56055d3a | 2531 | struct ext4_extent_header *eh; |
a86c6181 | 2532 | struct ext4_extent newex, *ex; |
f65e6fba | 2533 | ext4_fsblk_t goal, newblock; |
56055d3a | 2534 | int err = 0, depth, ret; |
a86c6181 | 2535 | unsigned long allocated = 0; |
c9de560d | 2536 | struct ext4_allocation_request ar; |
a86c6181 AT |
2537 | |
2538 | __clear_bit(BH_New, &bh_result->b_state); | |
bba90743 ES |
2539 | ext_debug("blocks %u/%lu requested for inode %u\n", |
2540 | iblock, max_blocks, inode->i_ino); | |
a86c6181 AT |
2541 | |
2542 | /* check in cache */ | |
7e028976 AM |
2543 | goal = ext4_ext_in_cache(inode, iblock, &newex); |
2544 | if (goal) { | |
a86c6181 AT |
2545 | if (goal == EXT4_EXT_CACHE_GAP) { |
2546 | if (!create) { | |
56055d3a AA |
2547 | /* |
2548 | * block isn't allocated yet and | |
2549 | * user doesn't want to allocate it | |
2550 | */ | |
a86c6181 AT |
2551 | goto out2; |
2552 | } | |
2553 | /* we should allocate requested block */ | |
2554 | } else if (goal == EXT4_EXT_CACHE_EXTENT) { | |
2555 | /* block is already allocated */ | |
8c55e204 DK |
2556 | newblock = iblock |
2557 | - le32_to_cpu(newex.ee_block) | |
2558 | + ext_pblock(&newex); | |
d0d856e8 | 2559 | /* number of remaining blocks in the extent */ |
b939e376 | 2560 | allocated = ext4_ext_get_actual_len(&newex) - |
a86c6181 AT |
2561 | (iblock - le32_to_cpu(newex.ee_block)); |
2562 | goto out; | |
2563 | } else { | |
2564 | BUG(); | |
2565 | } | |
2566 | } | |
2567 | ||
2568 | /* find extent for this block */ | |
2569 | path = ext4_ext_find_extent(inode, iblock, NULL); | |
2570 | if (IS_ERR(path)) { | |
2571 | err = PTR_ERR(path); | |
2572 | path = NULL; | |
2573 | goto out2; | |
2574 | } | |
2575 | ||
2576 | depth = ext_depth(inode); | |
2577 | ||
2578 | /* | |
d0d856e8 RD |
2579 | * consistent leaf must not be empty; |
2580 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
2581 | * this is why assert can't be put in ext4_ext_find_extent() |
2582 | */ | |
2583 | BUG_ON(path[depth].p_ext == NULL && depth != 0); | |
56055d3a | 2584 | eh = path[depth].p_hdr; |
a86c6181 | 2585 | |
7e028976 AM |
2586 | ex = path[depth].p_ext; |
2587 | if (ex) { | |
725d26d3 | 2588 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
f65e6fba | 2589 | ext4_fsblk_t ee_start = ext_pblock(ex); |
a2df2a63 | 2590 | unsigned short ee_len; |
471d4011 SB |
2591 | |
2592 | /* | |
471d4011 | 2593 | * Uninitialized extents are treated as holes, except that |
56055d3a | 2594 | * we split out initialized portions during a write. |
471d4011 | 2595 | */ |
a2df2a63 | 2596 | ee_len = ext4_ext_get_actual_len(ex); |
d0d856e8 | 2597 | /* if found extent covers block, simply return it */ |
8c55e204 | 2598 | if (iblock >= ee_block && iblock < ee_block + ee_len) { |
a86c6181 | 2599 | newblock = iblock - ee_block + ee_start; |
d0d856e8 | 2600 | /* number of remaining blocks in the extent */ |
a86c6181 | 2601 | allocated = ee_len - (iblock - ee_block); |
bba90743 | 2602 | ext_debug("%u fit into %lu:%d -> %llu\n", iblock, |
a86c6181 | 2603 | ee_block, ee_len, newblock); |
56055d3a | 2604 | |
a2df2a63 | 2605 | /* Do not put uninitialized extent in the cache */ |
56055d3a | 2606 | if (!ext4_ext_is_uninitialized(ex)) { |
a2df2a63 AA |
2607 | ext4_ext_put_in_cache(inode, ee_block, |
2608 | ee_len, ee_start, | |
2609 | EXT4_EXT_CACHE_EXTENT); | |
56055d3a AA |
2610 | goto out; |
2611 | } | |
2612 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) | |
2613 | goto out; | |
e067ba00 AK |
2614 | if (!create) { |
2615 | /* | |
2616 | * We have blocks reserved already. We | |
2617 | * return allocated blocks so that delalloc | |
2618 | * won't do block reservation for us. But | |
2619 | * the buffer head will be unmapped so that | |
2620 | * a read from the block returns 0s. | |
2621 | */ | |
2622 | if (allocated > max_blocks) | |
2623 | allocated = max_blocks; | |
1a89734d AK |
2624 | /* mark the buffer unwritten */ |
2625 | __set_bit(BH_Unwritten, &bh_result->b_state); | |
56055d3a | 2626 | goto out2; |
e067ba00 | 2627 | } |
56055d3a AA |
2628 | |
2629 | ret = ext4_ext_convert_to_initialized(handle, inode, | |
2630 | path, iblock, | |
2631 | max_blocks); | |
dbf9d7da DM |
2632 | if (ret <= 0) { |
2633 | err = ret; | |
56055d3a | 2634 | goto out2; |
dbf9d7da | 2635 | } else |
56055d3a AA |
2636 | allocated = ret; |
2637 | goto outnew; | |
a86c6181 AT |
2638 | } |
2639 | } | |
2640 | ||
2641 | /* | |
d0d856e8 | 2642 | * requested block isn't allocated yet; |
a86c6181 AT |
2643 | * we couldn't try to create block if create flag is zero |
2644 | */ | |
2645 | if (!create) { | |
56055d3a AA |
2646 | /* |
2647 | * put just found gap into cache to speed up | |
2648 | * subsequent requests | |
2649 | */ | |
a86c6181 AT |
2650 | ext4_ext_put_gap_in_cache(inode, path, iblock); |
2651 | goto out2; | |
2652 | } | |
2653 | /* | |
63f57933 AM |
2654 | * Okay, we need to do block allocation. Lazily initialize the block |
2655 | * allocation info here if necessary. | |
2656 | */ | |
a86c6181 AT |
2657 | if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info)) |
2658 | ext4_init_block_alloc_info(inode); | |
2659 | ||
c9de560d AT |
2660 | /* find neighbour allocated blocks */ |
2661 | ar.lleft = iblock; | |
2662 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); | |
2663 | if (err) | |
2664 | goto out2; | |
2665 | ar.lright = iblock; | |
2666 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright); | |
2667 | if (err) | |
2668 | goto out2; | |
25d14f98 | 2669 | |
749269fa AA |
2670 | /* |
2671 | * See if request is beyond maximum number of blocks we can have in | |
2672 | * a single extent. For an initialized extent this limit is | |
2673 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
2674 | * EXT_UNINIT_MAX_LEN. | |
2675 | */ | |
2676 | if (max_blocks > EXT_INIT_MAX_LEN && | |
2677 | create != EXT4_CREATE_UNINITIALIZED_EXT) | |
2678 | max_blocks = EXT_INIT_MAX_LEN; | |
2679 | else if (max_blocks > EXT_UNINIT_MAX_LEN && | |
2680 | create == EXT4_CREATE_UNINITIALIZED_EXT) | |
2681 | max_blocks = EXT_UNINIT_MAX_LEN; | |
2682 | ||
25d14f98 AA |
2683 | /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ |
2684 | newex.ee_block = cpu_to_le32(iblock); | |
2685 | newex.ee_len = cpu_to_le16(max_blocks); | |
2686 | err = ext4_ext_check_overlap(inode, &newex, path); | |
2687 | if (err) | |
b939e376 | 2688 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 AA |
2689 | else |
2690 | allocated = max_blocks; | |
c9de560d AT |
2691 | |
2692 | /* allocate new block */ | |
2693 | ar.inode = inode; | |
2694 | ar.goal = ext4_ext_find_goal(inode, path, iblock); | |
2695 | ar.logical = iblock; | |
2696 | ar.len = allocated; | |
2697 | if (S_ISREG(inode->i_mode)) | |
2698 | ar.flags = EXT4_MB_HINT_DATA; | |
2699 | else | |
2700 | /* disable in-core preallocation for non-regular files */ | |
2701 | ar.flags = 0; | |
2702 | newblock = ext4_mb_new_blocks(handle, &ar, &err); | |
a86c6181 AT |
2703 | if (!newblock) |
2704 | goto out2; | |
2ae02107 | 2705 | ext_debug("allocate new block: goal %llu, found %llu/%lu\n", |
a86c6181 AT |
2706 | goal, newblock, allocated); |
2707 | ||
2708 | /* try to insert new extent into found leaf and return */ | |
f65e6fba | 2709 | ext4_ext_store_pblock(&newex, newblock); |
c9de560d | 2710 | newex.ee_len = cpu_to_le16(ar.len); |
a2df2a63 AA |
2711 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */ |
2712 | ext4_ext_mark_uninitialized(&newex); | |
a86c6181 | 2713 | err = ext4_ext_insert_extent(handle, inode, path, &newex); |
315054f0 AT |
2714 | if (err) { |
2715 | /* free data blocks we just allocated */ | |
c9de560d AT |
2716 | /* not a good idea to call discard here directly, |
2717 | * but otherwise we'd need to call it every free() */ | |
2718 | ext4_mb_discard_inode_preallocations(inode); | |
315054f0 | 2719 | ext4_free_blocks(handle, inode, ext_pblock(&newex), |
b939e376 | 2720 | ext4_ext_get_actual_len(&newex), 0); |
a86c6181 | 2721 | goto out2; |
315054f0 | 2722 | } |
a86c6181 | 2723 | |
a86c6181 | 2724 | /* previous routine could use block we allocated */ |
f65e6fba | 2725 | newblock = ext_pblock(&newex); |
b939e376 | 2726 | allocated = ext4_ext_get_actual_len(&newex); |
56055d3a | 2727 | outnew: |
a379cd1d AK |
2728 | if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize) |
2729 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2730 | ||
a86c6181 AT |
2731 | __set_bit(BH_New, &bh_result->b_state); |
2732 | ||
a2df2a63 AA |
2733 | /* Cache only when it is _not_ an uninitialized extent */ |
2734 | if (create != EXT4_CREATE_UNINITIALIZED_EXT) | |
2735 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, | |
2736 | EXT4_EXT_CACHE_EXTENT); | |
a86c6181 AT |
2737 | out: |
2738 | if (allocated > max_blocks) | |
2739 | allocated = max_blocks; | |
2740 | ext4_ext_show_leaf(inode, path); | |
2741 | __set_bit(BH_Mapped, &bh_result->b_state); | |
2742 | bh_result->b_bdev = inode->i_sb->s_bdev; | |
2743 | bh_result->b_blocknr = newblock; | |
2744 | out2: | |
2745 | if (path) { | |
2746 | ext4_ext_drop_refs(path); | |
2747 | kfree(path); | |
2748 | } | |
a86c6181 AT |
2749 | return err ? err : allocated; |
2750 | } | |
2751 | ||
2752 | void ext4_ext_truncate(struct inode * inode, struct page *page) | |
2753 | { | |
2754 | struct address_space *mapping = inode->i_mapping; | |
2755 | struct super_block *sb = inode->i_sb; | |
725d26d3 | 2756 | ext4_lblk_t last_block; |
a86c6181 AT |
2757 | handle_t *handle; |
2758 | int err = 0; | |
2759 | ||
2760 | /* | |
2761 | * probably first extent we're gonna free will be last in block | |
2762 | */ | |
2763 | err = ext4_writepage_trans_blocks(inode) + 3; | |
2764 | handle = ext4_journal_start(inode, err); | |
2765 | if (IS_ERR(handle)) { | |
2766 | if (page) { | |
2767 | clear_highpage(page); | |
2768 | flush_dcache_page(page); | |
2769 | unlock_page(page); | |
2770 | page_cache_release(page); | |
2771 | } | |
2772 | return; | |
2773 | } | |
2774 | ||
2775 | if (page) | |
2776 | ext4_block_truncate_page(handle, page, mapping, inode->i_size); | |
2777 | ||
0e855ac8 | 2778 | down_write(&EXT4_I(inode)->i_data_sem); |
a86c6181 AT |
2779 | ext4_ext_invalidate_cache(inode); |
2780 | ||
c9de560d AT |
2781 | ext4_mb_discard_inode_preallocations(inode); |
2782 | ||
a86c6181 | 2783 | /* |
d0d856e8 RD |
2784 | * TODO: optimization is possible here. |
2785 | * Probably we need not scan at all, | |
2786 | * because page truncation is enough. | |
a86c6181 AT |
2787 | */ |
2788 | if (ext4_orphan_add(handle, inode)) | |
2789 | goto out_stop; | |
2790 | ||
2791 | /* we have to know where to truncate from in crash case */ | |
2792 | EXT4_I(inode)->i_disksize = inode->i_size; | |
2793 | ext4_mark_inode_dirty(handle, inode); | |
2794 | ||
2795 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
2796 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
2797 | err = ext4_ext_remove_space(inode, last_block); | |
2798 | ||
2799 | /* In a multi-transaction truncate, we only make the final | |
56055d3a AA |
2800 | * transaction synchronous. |
2801 | */ | |
a86c6181 AT |
2802 | if (IS_SYNC(inode)) |
2803 | handle->h_sync = 1; | |
2804 | ||
2805 | out_stop: | |
2806 | /* | |
d0d856e8 | 2807 | * If this was a simple ftruncate() and the file will remain alive, |
a86c6181 AT |
2808 | * then we need to clear up the orphan record which we created above. |
2809 | * However, if this was a real unlink then we were called by | |
2810 | * ext4_delete_inode(), and we allow that function to clean up the | |
2811 | * orphan info for us. | |
2812 | */ | |
2813 | if (inode->i_nlink) | |
2814 | ext4_orphan_del(handle, inode); | |
2815 | ||
0e855ac8 | 2816 | up_write(&EXT4_I(inode)->i_data_sem); |
ef737728 SR |
2817 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
2818 | ext4_mark_inode_dirty(handle, inode); | |
a86c6181 AT |
2819 | ext4_journal_stop(handle); |
2820 | } | |
2821 | ||
2822 | /* | |
d0d856e8 RD |
2823 | * ext4_ext_writepage_trans_blocks: |
2824 | * calculate max number of blocks we could modify | |
a86c6181 AT |
2825 | * in order to allocate new block for an inode |
2826 | */ | |
2827 | int ext4_ext_writepage_trans_blocks(struct inode *inode, int num) | |
2828 | { | |
2829 | int needed; | |
2830 | ||
2831 | needed = ext4_ext_calc_credits_for_insert(inode, NULL); | |
2832 | ||
d0d856e8 | 2833 | /* caller wants to allocate num blocks, but note it includes sb */ |
a86c6181 AT |
2834 | needed = needed * num - (num - 1); |
2835 | ||
2836 | #ifdef CONFIG_QUOTA | |
2837 | needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); | |
2838 | #endif | |
2839 | ||
2840 | return needed; | |
2841 | } | |
a2df2a63 | 2842 | |
fd28784a AK |
2843 | static void ext4_falloc_update_inode(struct inode *inode, |
2844 | int mode, loff_t new_size, int update_ctime) | |
2845 | { | |
2846 | struct timespec now; | |
2847 | ||
2848 | if (update_ctime) { | |
2849 | now = current_fs_time(inode->i_sb); | |
2850 | if (!timespec_equal(&inode->i_ctime, &now)) | |
2851 | inode->i_ctime = now; | |
2852 | } | |
2853 | /* | |
2854 | * Update only when preallocation was requested beyond | |
2855 | * the file size. | |
2856 | */ | |
2857 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
2858 | new_size > i_size_read(inode)) { | |
2859 | i_size_write(inode, new_size); | |
2860 | EXT4_I(inode)->i_disksize = new_size; | |
2861 | } | |
2862 | ||
2863 | } | |
2864 | ||
a2df2a63 AA |
2865 | /* |
2866 | * preallocate space for a file. This implements ext4's fallocate inode | |
2867 | * operation, which gets called from sys_fallocate system call. | |
2868 | * For block-mapped files, posix_fallocate should fall back to the method | |
2869 | * of writing zeroes to the required new blocks (the same behavior which is | |
2870 | * expected for file systems which do not support fallocate() system call). | |
2871 | */ | |
2872 | long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) | |
2873 | { | |
2874 | handle_t *handle; | |
725d26d3 | 2875 | ext4_lblk_t block; |
fd28784a | 2876 | loff_t new_size; |
725d26d3 | 2877 | unsigned long max_blocks; |
a2df2a63 AA |
2878 | int ret = 0; |
2879 | int ret2 = 0; | |
2880 | int retries = 0; | |
2881 | struct buffer_head map_bh; | |
2882 | unsigned int credits, blkbits = inode->i_blkbits; | |
2883 | ||
2884 | /* | |
2885 | * currently supporting (pre)allocate mode for extent-based | |
2886 | * files _only_ | |
2887 | */ | |
2888 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
2889 | return -EOPNOTSUPP; | |
2890 | ||
2891 | /* preallocation to directories is currently not supported */ | |
2892 | if (S_ISDIR(inode->i_mode)) | |
2893 | return -ENODEV; | |
2894 | ||
2895 | block = offset >> blkbits; | |
fd28784a AK |
2896 | /* |
2897 | * We can't just convert len to max_blocks because | |
2898 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
2899 | */ | |
a2df2a63 | 2900 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
fd28784a | 2901 | - block; |
a2df2a63 AA |
2902 | /* |
2903 | * credits to insert 1 extent into extent tree + buffers to be able to | |
2904 | * modify 1 super block, 1 block bitmap and 1 group descriptor. | |
2905 | */ | |
2906 | credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3; | |
55bd725a | 2907 | mutex_lock(&inode->i_mutex); |
a2df2a63 AA |
2908 | retry: |
2909 | while (ret >= 0 && ret < max_blocks) { | |
2910 | block = block + ret; | |
2911 | max_blocks = max_blocks - ret; | |
2912 | handle = ext4_journal_start(inode, credits); | |
2913 | if (IS_ERR(handle)) { | |
2914 | ret = PTR_ERR(handle); | |
2915 | break; | |
2916 | } | |
55bd725a | 2917 | ret = ext4_get_blocks_wrap(handle, inode, block, |
a2df2a63 AA |
2918 | max_blocks, &map_bh, |
2919 | EXT4_CREATE_UNINITIALIZED_EXT, 0); | |
221879c9 | 2920 | if (ret <= 0) { |
2c98615d AK |
2921 | #ifdef EXT4FS_DEBUG |
2922 | WARN_ON(ret <= 0); | |
2923 | printk(KERN_ERR "%s: ext4_ext_get_blocks " | |
2924 | "returned error inode#%lu, block=%u, " | |
2925 | "max_blocks=%lu", __func__, | |
221879c9 | 2926 | inode->i_ino, block, max_blocks); |
2c98615d | 2927 | #endif |
a2df2a63 AA |
2928 | ext4_mark_inode_dirty(handle, inode); |
2929 | ret2 = ext4_journal_stop(handle); | |
2930 | break; | |
2931 | } | |
fd28784a AK |
2932 | if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
2933 | blkbits) >> blkbits)) | |
2934 | new_size = offset + len; | |
2935 | else | |
2936 | new_size = (block + ret) << blkbits; | |
a2df2a63 | 2937 | |
fd28784a AK |
2938 | ext4_falloc_update_inode(inode, mode, new_size, |
2939 | buffer_new(&map_bh)); | |
a2df2a63 AA |
2940 | ext4_mark_inode_dirty(handle, inode); |
2941 | ret2 = ext4_journal_stop(handle); | |
2942 | if (ret2) | |
2943 | break; | |
2944 | } | |
fd28784a AK |
2945 | if (ret == -ENOSPC && |
2946 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
2947 | ret = 0; | |
a2df2a63 | 2948 | goto retry; |
a2df2a63 | 2949 | } |
55bd725a | 2950 | mutex_unlock(&inode->i_mutex); |
a2df2a63 AA |
2951 | return ret > 0 ? ret2 : ret; |
2952 | } |