diff options
Diffstat (limited to 'fs/ext4/extents.c')
-rw-r--r-- | fs/ext4/extents.c | 682 |
1 files changed, 548 insertions, 134 deletions
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c index b9ce24129070..750c46f7d893 100644 --- a/fs/ext4/extents.c +++ b/fs/ext4/extents.c | |||
@@ -39,6 +39,7 @@ | |||
39 | #include <linux/quotaops.h> | 39 | #include <linux/quotaops.h> |
40 | #include <linux/string.h> | 40 | #include <linux/string.h> |
41 | #include <linux/slab.h> | 41 | #include <linux/slab.h> |
42 | #include <linux/falloc.h> | ||
42 | #include <linux/ext4_fs_extents.h> | 43 | #include <linux/ext4_fs_extents.h> |
43 | #include <asm/uaccess.h> | 44 | #include <asm/uaccess.h> |
44 | 45 | ||
@@ -91,36 +92,6 @@ static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb) | |||
91 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); | 92 | ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); |
92 | } | 93 | } |
93 | 94 | ||
94 | static int ext4_ext_check_header(const char *function, struct inode *inode, | ||
95 | struct ext4_extent_header *eh) | ||
96 | { | ||
97 | const char *error_msg = NULL; | ||
98 | |||
99 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | ||
100 | error_msg = "invalid magic"; | ||
101 | goto corrupted; | ||
102 | } | ||
103 | if (unlikely(eh->eh_max == 0)) { | ||
104 | error_msg = "invalid eh_max"; | ||
105 | goto corrupted; | ||
106 | } | ||
107 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | ||
108 | error_msg = "invalid eh_entries"; | ||
109 | goto corrupted; | ||
110 | } | ||
111 | return 0; | ||
112 | |||
113 | corrupted: | ||
114 | ext4_error(inode->i_sb, function, | ||
115 | "bad header in inode #%lu: %s - magic %x, " | ||
116 | "entries %u, max %u, depth %u", | ||
117 | inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), | ||
118 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | ||
119 | le16_to_cpu(eh->eh_depth)); | ||
120 | |||
121 | return -EIO; | ||
122 | } | ||
123 | |||
124 | static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed) | 95 | static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed) |
125 | { | 96 | { |
126 | int err; | 97 | int err; |
@@ -269,6 +240,70 @@ static int ext4_ext_space_root_idx(struct inode *inode) | |||
269 | return size; | 240 | return size; |
270 | } | 241 | } |
271 | 242 | ||
243 | static int | ||
244 | ext4_ext_max_entries(struct inode *inode, int depth) | ||
245 | { | ||
246 | int max; | ||
247 | |||
248 | if (depth == ext_depth(inode)) { | ||
249 | if (depth == 0) | ||
250 | max = ext4_ext_space_root(inode); | ||
251 | else | ||
252 | max = ext4_ext_space_root_idx(inode); | ||
253 | } else { | ||
254 | if (depth == 0) | ||
255 | max = ext4_ext_space_block(inode); | ||
256 | else | ||
257 | max = ext4_ext_space_block_idx(inode); | ||
258 | } | ||
259 | |||
260 | return max; | ||
261 | } | ||
262 | |||
263 | static int __ext4_ext_check_header(const char *function, struct inode *inode, | ||
264 | struct ext4_extent_header *eh, | ||
265 | int depth) | ||
266 | { | ||
267 | const char *error_msg; | ||
268 | int max = 0; | ||
269 | |||
270 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | ||
271 | error_msg = "invalid magic"; | ||
272 | goto corrupted; | ||
273 | } | ||
274 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | ||
275 | error_msg = "unexpected eh_depth"; | ||
276 | goto corrupted; | ||
277 | } | ||
278 | if (unlikely(eh->eh_max == 0)) { | ||
279 | error_msg = "invalid eh_max"; | ||
280 | goto corrupted; | ||
281 | } | ||
282 | max = ext4_ext_max_entries(inode, depth); | ||
283 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | ||
284 | error_msg = "too large eh_max"; | ||
285 | goto corrupted; | ||
286 | } | ||
287 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | ||
288 | error_msg = "invalid eh_entries"; | ||
289 | goto corrupted; | ||
290 | } | ||
291 | return 0; | ||
292 | |||
293 | corrupted: | ||
294 | ext4_error(inode->i_sb, function, | ||
295 | "bad header in inode #%lu: %s - magic %x, " | ||
296 | "entries %u, max %u(%u), depth %u(%u)", | ||
297 | inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic), | ||
298 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | ||
299 | max, le16_to_cpu(eh->eh_depth), depth); | ||
300 | |||
301 | return -EIO; | ||
302 | } | ||
303 | |||
304 | #define ext4_ext_check_header(inode, eh, depth) \ | ||
305 | __ext4_ext_check_header(__FUNCTION__, inode, eh, depth) | ||
306 | |||
272 | #ifdef EXT_DEBUG | 307 | #ifdef EXT_DEBUG |
273 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | 308 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) |
274 | { | 309 | { |
@@ -282,7 +317,7 @@ static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |||
282 | } else if (path->p_ext) { | 317 | } else if (path->p_ext) { |
283 | ext_debug(" %d:%d:%llu ", | 318 | ext_debug(" %d:%d:%llu ", |
284 | le32_to_cpu(path->p_ext->ee_block), | 319 | le32_to_cpu(path->p_ext->ee_block), |
285 | le16_to_cpu(path->p_ext->ee_len), | 320 | ext4_ext_get_actual_len(path->p_ext), |
286 | ext_pblock(path->p_ext)); | 321 | ext_pblock(path->p_ext)); |
287 | } else | 322 | } else |
288 | ext_debug(" []"); | 323 | ext_debug(" []"); |
@@ -305,7 +340,7 @@ static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |||
305 | 340 | ||
306 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { | 341 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
307 | ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block), | 342 | ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block), |
308 | le16_to_cpu(ex->ee_len), ext_pblock(ex)); | 343 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
309 | } | 344 | } |
310 | ext_debug("\n"); | 345 | ext_debug("\n"); |
311 | } | 346 | } |
@@ -329,6 +364,7 @@ static void ext4_ext_drop_refs(struct ext4_ext_path *path) | |||
329 | /* | 364 | /* |
330 | * ext4_ext_binsearch_idx: | 365 | * ext4_ext_binsearch_idx: |
331 | * binary search for the closest index of the given block | 366 | * binary search for the closest index of the given block |
367 | * the header must be checked before calling this | ||
332 | */ | 368 | */ |
333 | static void | 369 | static void |
334 | ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block) | 370 | ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block) |
@@ -336,27 +372,25 @@ ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int bloc | |||
336 | struct ext4_extent_header *eh = path->p_hdr; | 372 | struct ext4_extent_header *eh = path->p_hdr; |
337 | struct ext4_extent_idx *r, *l, *m; | 373 | struct ext4_extent_idx *r, *l, *m; |
338 | 374 | ||
339 | BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC); | ||
340 | BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max)); | ||
341 | BUG_ON(le16_to_cpu(eh->eh_entries) <= 0); | ||
342 | 375 | ||
343 | ext_debug("binsearch for %d(idx): ", block); | 376 | ext_debug("binsearch for %d(idx): ", block); |
344 | 377 | ||
345 | l = EXT_FIRST_INDEX(eh) + 1; | 378 | l = EXT_FIRST_INDEX(eh) + 1; |
346 | r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1; | 379 | r = EXT_LAST_INDEX(eh); |
347 | while (l <= r) { | 380 | while (l <= r) { |
348 | m = l + (r - l) / 2; | 381 | m = l + (r - l) / 2; |
349 | if (block < le32_to_cpu(m->ei_block)) | 382 | if (block < le32_to_cpu(m->ei_block)) |
350 | r = m - 1; | 383 | r = m - 1; |
351 | else | 384 | else |
352 | l = m + 1; | 385 | l = m + 1; |
353 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block, | 386 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
354 | m, m->ei_block, r, r->ei_block); | 387 | m, le32_to_cpu(m->ei_block), |
388 | r, le32_to_cpu(r->ei_block)); | ||
355 | } | 389 | } |
356 | 390 | ||
357 | path->p_idx = l - 1; | 391 | path->p_idx = l - 1; |
358 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), | 392 | ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block), |
359 | idx_block(path->p_idx)); | 393 | idx_pblock(path->p_idx)); |
360 | 394 | ||
361 | #ifdef CHECK_BINSEARCH | 395 | #ifdef CHECK_BINSEARCH |
362 | { | 396 | { |
@@ -388,6 +422,7 @@ ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int bloc | |||
388 | /* | 422 | /* |
389 | * ext4_ext_binsearch: | 423 | * ext4_ext_binsearch: |
390 | * binary search for closest extent of the given block | 424 | * binary search for closest extent of the given block |
425 | * the header must be checked before calling this | ||
391 | */ | 426 | */ |
392 | static void | 427 | static void |
393 | ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) | 428 | ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) |
@@ -395,9 +430,6 @@ ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) | |||
395 | struct ext4_extent_header *eh = path->p_hdr; | 430 | struct ext4_extent_header *eh = path->p_hdr; |
396 | struct ext4_extent *r, *l, *m; | 431 | struct ext4_extent *r, *l, *m; |
397 | 432 | ||
398 | BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC); | ||
399 | BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max)); | ||
400 | |||
401 | if (eh->eh_entries == 0) { | 433 | if (eh->eh_entries == 0) { |
402 | /* | 434 | /* |
403 | * this leaf is empty: | 435 | * this leaf is empty: |
@@ -409,7 +441,7 @@ ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) | |||
409 | ext_debug("binsearch for %d: ", block); | 441 | ext_debug("binsearch for %d: ", block); |
410 | 442 | ||
411 | l = EXT_FIRST_EXTENT(eh) + 1; | 443 | l = EXT_FIRST_EXTENT(eh) + 1; |
412 | r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1; | 444 | r = EXT_LAST_EXTENT(eh); |
413 | 445 | ||
414 | while (l <= r) { | 446 | while (l <= r) { |
415 | m = l + (r - l) / 2; | 447 | m = l + (r - l) / 2; |
@@ -417,15 +449,16 @@ ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block) | |||
417 | r = m - 1; | 449 | r = m - 1; |
418 | else | 450 | else |
419 | l = m + 1; | 451 | l = m + 1; |
420 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block, | 452 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
421 | m, m->ee_block, r, r->ee_block); | 453 | m, le32_to_cpu(m->ee_block), |
454 | r, le32_to_cpu(r->ee_block)); | ||
422 | } | 455 | } |
423 | 456 | ||
424 | path->p_ext = l - 1; | 457 | path->p_ext = l - 1; |
425 | ext_debug(" -> %d:%llu:%d ", | 458 | ext_debug(" -> %d:%llu:%d ", |
426 | le32_to_cpu(path->p_ext->ee_block), | 459 | le32_to_cpu(path->p_ext->ee_block), |
427 | ext_pblock(path->p_ext), | 460 | ext_pblock(path->p_ext), |
428 | le16_to_cpu(path->p_ext->ee_len)); | 461 | ext4_ext_get_actual_len(path->p_ext)); |
429 | 462 | ||
430 | #ifdef CHECK_BINSEARCH | 463 | #ifdef CHECK_BINSEARCH |
431 | { | 464 | { |
@@ -468,11 +501,10 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path) | |||
468 | short int depth, i, ppos = 0, alloc = 0; | 501 | short int depth, i, ppos = 0, alloc = 0; |
469 | 502 | ||
470 | eh = ext_inode_hdr(inode); | 503 | eh = ext_inode_hdr(inode); |
471 | BUG_ON(eh == NULL); | 504 | depth = ext_depth(inode); |
472 | if (ext4_ext_check_header(__FUNCTION__, inode, eh)) | 505 | if (ext4_ext_check_header(inode, eh, depth)) |
473 | return ERR_PTR(-EIO); | 506 | return ERR_PTR(-EIO); |
474 | 507 | ||
475 | i = depth = ext_depth(inode); | ||
476 | 508 | ||
477 | /* account possible depth increase */ | 509 | /* account possible depth increase */ |
478 | if (!path) { | 510 | if (!path) { |
@@ -484,10 +516,12 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path) | |||
484 | } | 516 | } |
485 | path[0].p_hdr = eh; | 517 | path[0].p_hdr = eh; |
486 | 518 | ||
519 | i = depth; | ||
487 | /* walk through the tree */ | 520 | /* walk through the tree */ |
488 | while (i) { | 521 | while (i) { |
489 | ext_debug("depth %d: num %d, max %d\n", | 522 | ext_debug("depth %d: num %d, max %d\n", |
490 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | 523 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); |
524 | |||
491 | ext4_ext_binsearch_idx(inode, path + ppos, block); | 525 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
492 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); | 526 | path[ppos].p_block = idx_pblock(path[ppos].p_idx); |
493 | path[ppos].p_depth = i; | 527 | path[ppos].p_depth = i; |
@@ -504,7 +538,7 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path) | |||
504 | path[ppos].p_hdr = eh; | 538 | path[ppos].p_hdr = eh; |
505 | i--; | 539 | i--; |
506 | 540 | ||
507 | if (ext4_ext_check_header(__FUNCTION__, inode, eh)) | 541 | if (ext4_ext_check_header(inode, eh, i)) |
508 | goto err; | 542 | goto err; |
509 | } | 543 | } |
510 | 544 | ||
@@ -513,9 +547,6 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path) | |||
513 | path[ppos].p_ext = NULL; | 547 | path[ppos].p_ext = NULL; |
514 | path[ppos].p_idx = NULL; | 548 | path[ppos].p_idx = NULL; |
515 | 549 | ||
516 | if (ext4_ext_check_header(__FUNCTION__, inode, eh)) | ||
517 | goto err; | ||
518 | |||
519 | /* find extent */ | 550 | /* find extent */ |
520 | ext4_ext_binsearch(inode, path + ppos, block); | 551 | ext4_ext_binsearch(inode, path + ppos, block); |
521 | 552 | ||
@@ -553,7 +584,7 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |||
553 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { | 584 | if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) { |
554 | len = (len - 1) * sizeof(struct ext4_extent_idx); | 585 | len = (len - 1) * sizeof(struct ext4_extent_idx); |
555 | len = len < 0 ? 0 : len; | 586 | len = len < 0 ? 0 : len; |
556 | ext_debug("insert new index %d after: %d. " | 587 | ext_debug("insert new index %d after: %llu. " |
557 | "move %d from 0x%p to 0x%p\n", | 588 | "move %d from 0x%p to 0x%p\n", |
558 | logical, ptr, len, | 589 | logical, ptr, len, |
559 | (curp->p_idx + 1), (curp->p_idx + 2)); | 590 | (curp->p_idx + 1), (curp->p_idx + 2)); |
@@ -564,7 +595,7 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | |||
564 | /* insert before */ | 595 | /* insert before */ |
565 | len = len * sizeof(struct ext4_extent_idx); | 596 | len = len * sizeof(struct ext4_extent_idx); |
566 | len = len < 0 ? 0 : len; | 597 | len = len < 0 ? 0 : len; |
567 | ext_debug("insert new index %d before: %d. " | 598 | ext_debug("insert new index %d before: %llu. " |
568 | "move %d from 0x%p to 0x%p\n", | 599 | "move %d from 0x%p to 0x%p\n", |
569 | logical, ptr, len, | 600 | logical, ptr, len, |
570 | curp->p_idx, (curp->p_idx + 1)); | 601 | curp->p_idx, (curp->p_idx + 1)); |
@@ -686,7 +717,7 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode, | |||
686 | ext_debug("move %d:%llu:%d in new leaf %llu\n", | 717 | ext_debug("move %d:%llu:%d in new leaf %llu\n", |
687 | le32_to_cpu(path[depth].p_ext->ee_block), | 718 | le32_to_cpu(path[depth].p_ext->ee_block), |
688 | ext_pblock(path[depth].p_ext), | 719 | ext_pblock(path[depth].p_ext), |
689 | le16_to_cpu(path[depth].p_ext->ee_len), | 720 | ext4_ext_get_actual_len(path[depth].p_ext), |
690 | newblock); | 721 | newblock); |
691 | /*memmove(ex++, path[depth].p_ext++, | 722 | /*memmove(ex++, path[depth].p_ext++, |
692 | sizeof(struct ext4_extent)); | 723 | sizeof(struct ext4_extent)); |
@@ -764,7 +795,7 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode, | |||
764 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != | 795 | BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) != |
765 | EXT_LAST_INDEX(path[i].p_hdr)); | 796 | EXT_LAST_INDEX(path[i].p_hdr)); |
766 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { | 797 | while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) { |
767 | ext_debug("%d: move %d:%d in new index %llu\n", i, | 798 | ext_debug("%d: move %d:%llu in new index %llu\n", i, |
768 | le32_to_cpu(path[i].p_idx->ei_block), | 799 | le32_to_cpu(path[i].p_idx->ei_block), |
769 | idx_pblock(path[i].p_idx), | 800 | idx_pblock(path[i].p_idx), |
770 | newblock); | 801 | newblock); |
@@ -893,8 +924,13 @@ static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |||
893 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); | 924 | curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode)); |
894 | curp->p_hdr->eh_entries = cpu_to_le16(1); | 925 | curp->p_hdr->eh_entries = cpu_to_le16(1); |
895 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); | 926 | curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr); |
896 | /* FIXME: it works, but actually path[0] can be index */ | 927 | |
897 | curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | 928 | if (path[0].p_hdr->eh_depth) |
929 | curp->p_idx->ei_block = | ||
930 | EXT_FIRST_INDEX(path[0].p_hdr)->ei_block; | ||
931 | else | ||
932 | curp->p_idx->ei_block = | ||
933 | EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block; | ||
898 | ext4_idx_store_pblock(curp->p_idx, newblock); | 934 | ext4_idx_store_pblock(curp->p_idx, newblock); |
899 | 935 | ||
900 | neh = ext_inode_hdr(inode); | 936 | neh = ext_inode_hdr(inode); |
@@ -1106,7 +1142,24 @@ static int | |||
1106 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, | 1142 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1107 | struct ext4_extent *ex2) | 1143 | struct ext4_extent *ex2) |
1108 | { | 1144 | { |
1109 | if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len) != | 1145 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
1146 | |||
1147 | /* | ||
1148 | * Make sure that either both extents are uninitialized, or | ||
1149 | * both are _not_. | ||
1150 | */ | ||
1151 | if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2)) | ||
1152 | return 0; | ||
1153 | |||
1154 | if (ext4_ext_is_uninitialized(ex1)) | ||
1155 | max_len = EXT_UNINIT_MAX_LEN; | ||
1156 | else | ||
1157 | max_len = EXT_INIT_MAX_LEN; | ||
1158 | |||
1159 | ext1_ee_len = ext4_ext_get_actual_len(ex1); | ||
1160 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | ||
1161 | |||
1162 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | ||
1110 | le32_to_cpu(ex2->ee_block)) | 1163 | le32_to_cpu(ex2->ee_block)) |
1111 | return 0; | 1164 | return 0; |
1112 | 1165 | ||
@@ -1115,19 +1168,66 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, | |||
1115 | * as an RO_COMPAT feature, refuse to merge to extents if | 1168 | * as an RO_COMPAT feature, refuse to merge to extents if |
1116 | * this can result in the top bit of ee_len being set. | 1169 | * this can result in the top bit of ee_len being set. |
1117 | */ | 1170 | */ |
1118 | if (le16_to_cpu(ex1->ee_len) + le16_to_cpu(ex2->ee_len) > EXT_MAX_LEN) | 1171 | if (ext1_ee_len + ext2_ee_len > max_len) |
1119 | return 0; | 1172 | return 0; |
1120 | #ifdef AGGRESSIVE_TEST | 1173 | #ifdef AGGRESSIVE_TEST |
1121 | if (le16_to_cpu(ex1->ee_len) >= 4) | 1174 | if (le16_to_cpu(ex1->ee_len) >= 4) |
1122 | return 0; | 1175 | return 0; |
1123 | #endif | 1176 | #endif |
1124 | 1177 | ||
1125 | if (ext_pblock(ex1) + le16_to_cpu(ex1->ee_len) == ext_pblock(ex2)) | 1178 | if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2)) |
1126 | return 1; | 1179 | return 1; |
1127 | return 0; | 1180 | return 0; |
1128 | } | 1181 | } |
1129 | 1182 | ||
1130 | /* | 1183 | /* |
1184 | * This function tries to merge the "ex" extent to the next extent in the tree. | ||
1185 | * It always tries to merge towards right. If you want to merge towards | ||
1186 | * left, pass "ex - 1" as argument instead of "ex". | ||
1187 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | ||
1188 | * 1 if they got merged. | ||
1189 | */ | ||
1190 | int ext4_ext_try_to_merge(struct inode *inode, | ||
1191 | struct ext4_ext_path *path, | ||
1192 | struct ext4_extent *ex) | ||
1193 | { | ||
1194 | struct ext4_extent_header *eh; | ||
1195 | unsigned int depth, len; | ||
1196 | int merge_done = 0; | ||
1197 | int uninitialized = 0; | ||
1198 | |||
1199 | depth = ext_depth(inode); | ||
1200 | BUG_ON(path[depth].p_hdr == NULL); | ||
1201 | eh = path[depth].p_hdr; | ||
1202 | |||
1203 | while (ex < EXT_LAST_EXTENT(eh)) { | ||
1204 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | ||
1205 | break; | ||
1206 | /* merge with next extent! */ | ||
1207 | if (ext4_ext_is_uninitialized(ex)) | ||
1208 | uninitialized = 1; | ||
1209 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | ||
1210 | + ext4_ext_get_actual_len(ex + 1)); | ||
1211 | if (uninitialized) | ||
1212 | ext4_ext_mark_uninitialized(ex); | ||
1213 | |||
1214 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | ||
1215 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | ||
1216 | * sizeof(struct ext4_extent); | ||
1217 | memmove(ex + 1, ex + 2, len); | ||
1218 | } | ||
1219 | eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1); | ||
1220 | merge_done = 1; | ||
1221 | WARN_ON(eh->eh_entries == 0); | ||
1222 | if (!eh->eh_entries) | ||
1223 | ext4_error(inode->i_sb, "ext4_ext_try_to_merge", | ||
1224 | "inode#%lu, eh->eh_entries = 0!", inode->i_ino); | ||
1225 | } | ||
1226 | |||
1227 | return merge_done; | ||
1228 | } | ||
1229 | |||
1230 | /* | ||
1131 | * check if a portion of the "newext" extent overlaps with an | 1231 | * check if a portion of the "newext" extent overlaps with an |
1132 | * existing extent. | 1232 | * existing extent. |
1133 | * | 1233 | * |
@@ -1144,7 +1244,7 @@ unsigned int ext4_ext_check_overlap(struct inode *inode, | |||
1144 | unsigned int ret = 0; | 1244 | unsigned int ret = 0; |
1145 | 1245 | ||
1146 | b1 = le32_to_cpu(newext->ee_block); | 1246 | b1 = le32_to_cpu(newext->ee_block); |
1147 | len1 = le16_to_cpu(newext->ee_len); | 1247 | len1 = ext4_ext_get_actual_len(newext); |
1148 | depth = ext_depth(inode); | 1248 | depth = ext_depth(inode); |
1149 | if (!path[depth].p_ext) | 1249 | if (!path[depth].p_ext) |
1150 | goto out; | 1250 | goto out; |
@@ -1191,8 +1291,9 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |||
1191 | struct ext4_extent *nearex; /* nearest extent */ | 1291 | struct ext4_extent *nearex; /* nearest extent */ |
1192 | struct ext4_ext_path *npath = NULL; | 1292 | struct ext4_ext_path *npath = NULL; |
1193 | int depth, len, err, next; | 1293 | int depth, len, err, next; |
1294 | unsigned uninitialized = 0; | ||
1194 | 1295 | ||
1195 | BUG_ON(newext->ee_len == 0); | 1296 | BUG_ON(ext4_ext_get_actual_len(newext) == 0); |
1196 | depth = ext_depth(inode); | 1297 | depth = ext_depth(inode); |
1197 | ex = path[depth].p_ext; | 1298 | ex = path[depth].p_ext; |
1198 | BUG_ON(path[depth].p_hdr == NULL); | 1299 | BUG_ON(path[depth].p_hdr == NULL); |
@@ -1200,14 +1301,24 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |||
1200 | /* try to insert block into found extent and return */ | 1301 | /* try to insert block into found extent and return */ |
1201 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | 1302 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { |
1202 | ext_debug("append %d block to %d:%d (from %llu)\n", | 1303 | ext_debug("append %d block to %d:%d (from %llu)\n", |
1203 | le16_to_cpu(newext->ee_len), | 1304 | ext4_ext_get_actual_len(newext), |
1204 | le32_to_cpu(ex->ee_block), | 1305 | le32_to_cpu(ex->ee_block), |
1205 | le16_to_cpu(ex->ee_len), ext_pblock(ex)); | 1306 | ext4_ext_get_actual_len(ex), ext_pblock(ex)); |
1206 | err = ext4_ext_get_access(handle, inode, path + depth); | 1307 | err = ext4_ext_get_access(handle, inode, path + depth); |
1207 | if (err) | 1308 | if (err) |
1208 | return err; | 1309 | return err; |
1209 | ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len) | 1310 | |
1210 | + le16_to_cpu(newext->ee_len)); | 1311 | /* |
1312 | * ext4_can_extents_be_merged should have checked that either | ||
1313 | * both extents are uninitialized, or both aren't. Thus we | ||
1314 | * need to check only one of them here. | ||
1315 | */ | ||
1316 | if (ext4_ext_is_uninitialized(ex)) | ||
1317 | uninitialized = 1; | ||
1318 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | ||
1319 | + ext4_ext_get_actual_len(newext)); | ||
1320 | if (uninitialized) | ||
1321 | ext4_ext_mark_uninitialized(ex); | ||
1211 | eh = path[depth].p_hdr; | 1322 | eh = path[depth].p_hdr; |
1212 | nearex = ex; | 1323 | nearex = ex; |
1213 | goto merge; | 1324 | goto merge; |
@@ -1263,7 +1374,7 @@ has_space: | |||
1263 | ext_debug("first extent in the leaf: %d:%llu:%d\n", | 1374 | ext_debug("first extent in the leaf: %d:%llu:%d\n", |
1264 | le32_to_cpu(newext->ee_block), | 1375 | le32_to_cpu(newext->ee_block), |
1265 | ext_pblock(newext), | 1376 | ext_pblock(newext), |
1266 | le16_to_cpu(newext->ee_len)); | 1377 | ext4_ext_get_actual_len(newext)); |
1267 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); | 1378 | path[depth].p_ext = EXT_FIRST_EXTENT(eh); |
1268 | } else if (le32_to_cpu(newext->ee_block) | 1379 | } else if (le32_to_cpu(newext->ee_block) |
1269 | > le32_to_cpu(nearex->ee_block)) { | 1380 | > le32_to_cpu(nearex->ee_block)) { |
@@ -1276,7 +1387,7 @@ has_space: | |||
1276 | "move %d from 0x%p to 0x%p\n", | 1387 | "move %d from 0x%p to 0x%p\n", |
1277 | le32_to_cpu(newext->ee_block), | 1388 | le32_to_cpu(newext->ee_block), |
1278 | ext_pblock(newext), | 1389 | ext_pblock(newext), |
1279 | le16_to_cpu(newext->ee_len), | 1390 | ext4_ext_get_actual_len(newext), |
1280 | nearex, len, nearex + 1, nearex + 2); | 1391 | nearex, len, nearex + 1, nearex + 2); |
1281 | memmove(nearex + 2, nearex + 1, len); | 1392 | memmove(nearex + 2, nearex + 1, len); |
1282 | } | 1393 | } |
@@ -1289,7 +1400,7 @@ has_space: | |||
1289 | "move %d from 0x%p to 0x%p\n", | 1400 | "move %d from 0x%p to 0x%p\n", |
1290 | le32_to_cpu(newext->ee_block), | 1401 | le32_to_cpu(newext->ee_block), |
1291 | ext_pblock(newext), | 1402 | ext_pblock(newext), |
1292 | le16_to_cpu(newext->ee_len), | 1403 | ext4_ext_get_actual_len(newext), |
1293 | nearex, len, nearex + 1, nearex + 2); | 1404 | nearex, len, nearex + 1, nearex + 2); |
1294 | memmove(nearex + 1, nearex, len); | 1405 | memmove(nearex + 1, nearex, len); |
1295 | path[depth].p_ext = nearex; | 1406 | path[depth].p_ext = nearex; |
@@ -1304,20 +1415,7 @@ has_space: | |||
1304 | 1415 | ||
1305 | merge: | 1416 | merge: |
1306 | /* try to merge extents to the right */ | 1417 | /* try to merge extents to the right */ |
1307 | while (nearex < EXT_LAST_EXTENT(eh)) { | 1418 | ext4_ext_try_to_merge(inode, path, nearex); |
1308 | if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1)) | ||
1309 | break; | ||
1310 | /* merge with next extent! */ | ||
1311 | nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len) | ||
1312 | + le16_to_cpu(nearex[1].ee_len)); | ||
1313 | if (nearex + 1 < EXT_LAST_EXTENT(eh)) { | ||
1314 | len = (EXT_LAST_EXTENT(eh) - nearex - 1) | ||
1315 | * sizeof(struct ext4_extent); | ||
1316 | memmove(nearex + 1, nearex + 2, len); | ||
1317 | } | ||
1318 | eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1); | ||
1319 | BUG_ON(eh->eh_entries == 0); | ||
1320 | } | ||
1321 | 1419 | ||
1322 | /* try to merge extents to the left */ | 1420 | /* try to merge extents to the left */ |
1323 | 1421 | ||
@@ -1379,8 +1477,8 @@ int ext4_ext_walk_space(struct inode *inode, unsigned long block, | |||
1379 | end = le32_to_cpu(ex->ee_block); | 1477 | end = le32_to_cpu(ex->ee_block); |
1380 | if (block + num < end) | 1478 | if (block + num < end) |
1381 | end = block + num; | 1479 | end = block + num; |
1382 | } else if (block >= | 1480 | } else if (block >= le32_to_cpu(ex->ee_block) |
1383 | le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) { | 1481 | + ext4_ext_get_actual_len(ex)) { |
1384 | /* need to allocate space after found extent */ | 1482 | /* need to allocate space after found extent */ |
1385 | start = block; | 1483 | start = block; |
1386 | end = block + num; | 1484 | end = block + num; |
@@ -1392,7 +1490,8 @@ int ext4_ext_walk_space(struct inode *inode, unsigned long block, | |||
1392 | * by found extent | 1490 | * by found extent |
1393 | */ | 1491 | */ |
1394 | start = block; | 1492 | start = block; |
1395 | end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len); | 1493 | end = le32_to_cpu(ex->ee_block) |
1494 | + ext4_ext_get_actual_len(ex); | ||
1396 | if (block + num < end) | 1495 | if (block + num < end) |
1397 | end = block + num; | 1496 | end = block + num; |
1398 | exists = 1; | 1497 | exists = 1; |
@@ -1408,7 +1507,7 @@ int ext4_ext_walk_space(struct inode *inode, unsigned long block, | |||
1408 | cbex.ec_type = EXT4_EXT_CACHE_GAP; | 1507 | cbex.ec_type = EXT4_EXT_CACHE_GAP; |
1409 | } else { | 1508 | } else { |
1410 | cbex.ec_block = le32_to_cpu(ex->ee_block); | 1509 | cbex.ec_block = le32_to_cpu(ex->ee_block); |
1411 | cbex.ec_len = le16_to_cpu(ex->ee_len); | 1510 | cbex.ec_len = ext4_ext_get_actual_len(ex); |
1412 | cbex.ec_start = ext_pblock(ex); | 1511 | cbex.ec_start = ext_pblock(ex); |
1413 | cbex.ec_type = EXT4_EXT_CACHE_EXTENT; | 1512 | cbex.ec_type = EXT4_EXT_CACHE_EXTENT; |
1414 | } | 1513 | } |
@@ -1481,15 +1580,15 @@ ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, | |||
1481 | ext_debug("cache gap(before): %lu [%lu:%lu]", | 1580 | ext_debug("cache gap(before): %lu [%lu:%lu]", |
1482 | (unsigned long) block, | 1581 | (unsigned long) block, |
1483 | (unsigned long) le32_to_cpu(ex->ee_block), | 1582 | (unsigned long) le32_to_cpu(ex->ee_block), |
1484 | (unsigned long) le16_to_cpu(ex->ee_len)); | 1583 | (unsigned long) ext4_ext_get_actual_len(ex)); |
1485 | } else if (block >= le32_to_cpu(ex->ee_block) | 1584 | } else if (block >= le32_to_cpu(ex->ee_block) |
1486 | + le16_to_cpu(ex->ee_len)) { | 1585 | + ext4_ext_get_actual_len(ex)) { |
1487 | lblock = le32_to_cpu(ex->ee_block) | 1586 | lblock = le32_to_cpu(ex->ee_block) |
1488 | + le16_to_cpu(ex->ee_len); | 1587 | + ext4_ext_get_actual_len(ex); |
1489 | len = ext4_ext_next_allocated_block(path); | 1588 | len = ext4_ext_next_allocated_block(path); |
1490 | ext_debug("cache gap(after): [%lu:%lu] %lu", | 1589 | ext_debug("cache gap(after): [%lu:%lu] %lu", |
1491 | (unsigned long) le32_to_cpu(ex->ee_block), | 1590 | (unsigned long) le32_to_cpu(ex->ee_block), |
1492 | (unsigned long) le16_to_cpu(ex->ee_len), | 1591 | (unsigned long) ext4_ext_get_actual_len(ex), |
1493 | (unsigned long) block); | 1592 | (unsigned long) block); |
1494 | BUG_ON(len == lblock); | 1593 | BUG_ON(len == lblock); |
1495 | len = len - lblock; | 1594 | len = len - lblock; |
@@ -1619,12 +1718,12 @@ static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |||
1619 | unsigned long from, unsigned long to) | 1718 | unsigned long from, unsigned long to) |
1620 | { | 1719 | { |
1621 | struct buffer_head *bh; | 1720 | struct buffer_head *bh; |
1721 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | ||
1622 | int i; | 1722 | int i; |
1623 | 1723 | ||
1624 | #ifdef EXTENTS_STATS | 1724 | #ifdef EXTENTS_STATS |
1625 | { | 1725 | { |
1626 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | 1726 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1627 | unsigned short ee_len = le16_to_cpu(ex->ee_len); | ||
1628 | spin_lock(&sbi->s_ext_stats_lock); | 1727 | spin_lock(&sbi->s_ext_stats_lock); |
1629 | sbi->s_ext_blocks += ee_len; | 1728 | sbi->s_ext_blocks += ee_len; |
1630 | sbi->s_ext_extents++; | 1729 | sbi->s_ext_extents++; |
@@ -1638,12 +1737,12 @@ static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |||
1638 | } | 1737 | } |
1639 | #endif | 1738 | #endif |
1640 | if (from >= le32_to_cpu(ex->ee_block) | 1739 | if (from >= le32_to_cpu(ex->ee_block) |
1641 | && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) { | 1740 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
1642 | /* tail removal */ | 1741 | /* tail removal */ |
1643 | unsigned long num; | 1742 | unsigned long num; |
1644 | ext4_fsblk_t start; | 1743 | ext4_fsblk_t start; |
1645 | num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from; | 1744 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
1646 | start = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - num; | 1745 | start = ext_pblock(ex) + ee_len - num; |
1647 | ext_debug("free last %lu blocks starting %llu\n", num, start); | 1746 | ext_debug("free last %lu blocks starting %llu\n", num, start); |
1648 | for (i = 0; i < num; i++) { | 1747 | for (i = 0; i < num; i++) { |
1649 | bh = sb_find_get_block(inode->i_sb, start + i); | 1748 | bh = sb_find_get_block(inode->i_sb, start + i); |
@@ -1651,12 +1750,12 @@ static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |||
1651 | } | 1750 | } |
1652 | ext4_free_blocks(handle, inode, start, num); | 1751 | ext4_free_blocks(handle, inode, start, num); |
1653 | } else if (from == le32_to_cpu(ex->ee_block) | 1752 | } else if (from == le32_to_cpu(ex->ee_block) |
1654 | && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) { | 1753 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
1655 | printk("strange request: removal %lu-%lu from %u:%u\n", | 1754 | printk("strange request: removal %lu-%lu from %u:%u\n", |
1656 | from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len)); | 1755 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
1657 | } else { | 1756 | } else { |
1658 | printk("strange request: removal(2) %lu-%lu from %u:%u\n", | 1757 | printk("strange request: removal(2) %lu-%lu from %u:%u\n", |
1659 | from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len)); | 1758 | from, to, le32_to_cpu(ex->ee_block), ee_len); |
1660 | } | 1759 | } |
1661 | return 0; | 1760 | return 0; |
1662 | } | 1761 | } |
@@ -1671,21 +1770,23 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |||
1671 | unsigned a, b, block, num; | 1770 | unsigned a, b, block, num; |
1672 | unsigned long ex_ee_block; | 1771 | unsigned long ex_ee_block; |
1673 | unsigned short ex_ee_len; | 1772 | unsigned short ex_ee_len; |
1773 | unsigned uninitialized = 0; | ||
1674 | struct ext4_extent *ex; | 1774 | struct ext4_extent *ex; |
1675 | 1775 | ||
1776 | /* the header must be checked already in ext4_ext_remove_space() */ | ||
1676 | ext_debug("truncate since %lu in leaf\n", start); | 1777 | ext_debug("truncate since %lu in leaf\n", start); |
1677 | if (!path[depth].p_hdr) | 1778 | if (!path[depth].p_hdr) |
1678 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | 1779 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); |
1679 | eh = path[depth].p_hdr; | 1780 | eh = path[depth].p_hdr; |
1680 | BUG_ON(eh == NULL); | 1781 | BUG_ON(eh == NULL); |
1681 | BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max)); | ||
1682 | BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC); | ||
1683 | 1782 | ||
1684 | /* find where to start removing */ | 1783 | /* find where to start removing */ |
1685 | ex = EXT_LAST_EXTENT(eh); | 1784 | ex = EXT_LAST_EXTENT(eh); |
1686 | 1785 | ||
1687 | ex_ee_block = le32_to_cpu(ex->ee_block); | 1786 | ex_ee_block = le32_to_cpu(ex->ee_block); |
1688 | ex_ee_len = le16_to_cpu(ex->ee_len); | 1787 | if (ext4_ext_is_uninitialized(ex)) |
1788 | uninitialized = 1; | ||
1789 | ex_ee_len = ext4_ext_get_actual_len(ex); | ||
1689 | 1790 | ||
1690 | while (ex >= EXT_FIRST_EXTENT(eh) && | 1791 | while (ex >= EXT_FIRST_EXTENT(eh) && |
1691 | ex_ee_block + ex_ee_len > start) { | 1792 | ex_ee_block + ex_ee_len > start) { |
@@ -1753,6 +1854,12 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |||
1753 | 1854 | ||
1754 | ex->ee_block = cpu_to_le32(block); | 1855 | ex->ee_block = cpu_to_le32(block); |
1755 | ex->ee_len = cpu_to_le16(num); | 1856 | ex->ee_len = cpu_to_le16(num); |
1857 | /* | ||
1858 | * Do not mark uninitialized if all the blocks in the | ||
1859 | * extent have been removed. | ||
1860 | */ | ||
1861 | if (uninitialized && num) | ||
1862 | ext4_ext_mark_uninitialized(ex); | ||
1756 | 1863 | ||
1757 | err = ext4_ext_dirty(handle, inode, path + depth); | 1864 | err = ext4_ext_dirty(handle, inode, path + depth); |
1758 | if (err) | 1865 | if (err) |
@@ -1762,7 +1869,7 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |||
1762 | ext_pblock(ex)); | 1869 | ext_pblock(ex)); |
1763 | ex--; | 1870 | ex--; |
1764 | ex_ee_block = le32_to_cpu(ex->ee_block); | 1871 | ex_ee_block = le32_to_cpu(ex->ee_block); |
1765 | ex_ee_len = le16_to_cpu(ex->ee_len); | 1872 | ex_ee_len = ext4_ext_get_actual_len(ex); |
1766 | } | 1873 | } |
1767 | 1874 | ||
1768 | if (correct_index && eh->eh_entries) | 1875 | if (correct_index && eh->eh_entries) |
@@ -1825,7 +1932,7 @@ int ext4_ext_remove_space(struct inode *inode, unsigned long start) | |||
1825 | return -ENOMEM; | 1932 | return -ENOMEM; |
1826 | } | 1933 | } |
1827 | path[0].p_hdr = ext_inode_hdr(inode); | 1934 | path[0].p_hdr = ext_inode_hdr(inode); |
1828 | if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) { | 1935 | if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) { |
1829 | err = -EIO; | 1936 | err = -EIO; |
1830 | goto out; | 1937 | goto out; |
1831 | } | 1938 | } |
@@ -1846,17 +1953,8 @@ int ext4_ext_remove_space(struct inode *inode, unsigned long start) | |||
1846 | if (!path[i].p_hdr) { | 1953 | if (!path[i].p_hdr) { |
1847 | ext_debug("initialize header\n"); | 1954 | ext_debug("initialize header\n"); |
1848 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | 1955 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); |
1849 | if (ext4_ext_check_header(__FUNCTION__, inode, | ||
1850 | path[i].p_hdr)) { | ||
1851 | err = -EIO; | ||
1852 | goto out; | ||
1853 | } | ||
1854 | } | 1956 | } |
1855 | 1957 | ||
1856 | BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries) | ||
1857 | > le16_to_cpu(path[i].p_hdr->eh_max)); | ||
1858 | BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC); | ||
1859 | |||
1860 | if (!path[i].p_idx) { | 1958 | if (!path[i].p_idx) { |
1861 | /* this level hasn't been touched yet */ | 1959 | /* this level hasn't been touched yet */ |
1862 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); | 1960 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
@@ -1873,17 +1971,27 @@ int ext4_ext_remove_space(struct inode *inode, unsigned long start) | |||
1873 | i, EXT_FIRST_INDEX(path[i].p_hdr), | 1971 | i, EXT_FIRST_INDEX(path[i].p_hdr), |
1874 | path[i].p_idx); | 1972 | path[i].p_idx); |
1875 | if (ext4_ext_more_to_rm(path + i)) { | 1973 | if (ext4_ext_more_to_rm(path + i)) { |
1974 | struct buffer_head *bh; | ||
1876 | /* go to the next level */ | 1975 | /* go to the next level */ |
1877 | ext_debug("move to level %d (block %llu)\n", | 1976 | ext_debug("move to level %d (block %llu)\n", |
1878 | i + 1, idx_pblock(path[i].p_idx)); | 1977 | i + 1, idx_pblock(path[i].p_idx)); |
1879 | memset(path + i + 1, 0, sizeof(*path)); | 1978 | memset(path + i + 1, 0, sizeof(*path)); |
1880 | path[i+1].p_bh = | 1979 | bh = sb_bread(sb, idx_pblock(path[i].p_idx)); |
1881 | sb_bread(sb, idx_pblock(path[i].p_idx)); | 1980 | if (!bh) { |
1882 | if (!path[i+1].p_bh) { | ||
1883 | /* should we reset i_size? */ | 1981 | /* should we reset i_size? */ |
1884 | err = -EIO; | 1982 | err = -EIO; |
1885 | break; | 1983 | break; |
1886 | } | 1984 | } |
1985 | if (WARN_ON(i + 1 > depth)) { | ||
1986 | err = -EIO; | ||
1987 | break; | ||
1988 | } | ||
1989 | if (ext4_ext_check_header(inode, ext_block_hdr(bh), | ||
1990 | depth - i - 1)) { | ||
1991 | err = -EIO; | ||
1992 | break; | ||
1993 | } | ||
1994 | path[i + 1].p_bh = bh; | ||
1887 | 1995 | ||
1888 | /* save actual number of indexes since this | 1996 | /* save actual number of indexes since this |
1889 | * number is changed at the next iteration */ | 1997 | * number is changed at the next iteration */ |
@@ -1977,15 +2085,158 @@ void ext4_ext_release(struct super_block *sb) | |||
1977 | #endif | 2085 | #endif |
1978 | } | 2086 | } |
1979 | 2087 | ||
2088 | /* | ||
2089 | * This function is called by ext4_ext_get_blocks() if someone tries to write | ||
2090 | * to an uninitialized extent. It may result in splitting the uninitialized | ||
2091 | * extent into multiple extents (upto three - one initialized and two | ||
2092 | * uninitialized). | ||
2093 | * There are three possibilities: | ||
2094 | * a> There is no split required: Entire extent should be initialized | ||
2095 | * b> Splits in two extents: Write is happening at either end of the extent | ||
2096 | * c> Splits in three extents: Somone is writing in middle of the extent | ||
2097 | */ | ||
2098 | int ext4_ext_convert_to_initialized(handle_t *handle, struct inode *inode, | ||
2099 | struct ext4_ext_path *path, | ||
2100 | ext4_fsblk_t iblock, | ||
2101 | unsigned long max_blocks) | ||
2102 | { | ||
2103 | struct ext4_extent *ex, newex; | ||
2104 | struct ext4_extent *ex1 = NULL; | ||
2105 | struct ext4_extent *ex2 = NULL; | ||
2106 | struct ext4_extent *ex3 = NULL; | ||
2107 | struct ext4_extent_header *eh; | ||
2108 | unsigned int allocated, ee_block, ee_len, depth; | ||
2109 | ext4_fsblk_t newblock; | ||
2110 | int err = 0; | ||
2111 | int ret = 0; | ||
2112 | |||
2113 | depth = ext_depth(inode); | ||
2114 | eh = path[depth].p_hdr; | ||
2115 | ex = path[depth].p_ext; | ||
2116 | ee_block = le32_to_cpu(ex->ee_block); | ||
2117 | ee_len = ext4_ext_get_actual_len(ex); | ||
2118 | allocated = ee_len - (iblock - ee_block); | ||
2119 | newblock = iblock - ee_block + ext_pblock(ex); | ||
2120 | ex2 = ex; | ||
2121 | |||
2122 | /* ex1: ee_block to iblock - 1 : uninitialized */ | ||
2123 | if (iblock > ee_block) { | ||
2124 | ex1 = ex; | ||
2125 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | ||
2126 | ext4_ext_mark_uninitialized(ex1); | ||
2127 | ex2 = &newex; | ||
2128 | } | ||
2129 | /* | ||
2130 | * for sanity, update the length of the ex2 extent before | ||
2131 | * we insert ex3, if ex1 is NULL. This is to avoid temporary | ||
2132 | * overlap of blocks. | ||
2133 | */ | ||
2134 | if (!ex1 && allocated > max_blocks) | ||
2135 | ex2->ee_len = cpu_to_le16(max_blocks); | ||
2136 | /* ex3: to ee_block + ee_len : uninitialised */ | ||
2137 | if (allocated > max_blocks) { | ||
2138 | unsigned int newdepth; | ||
2139 | ex3 = &newex; | ||
2140 | ex3->ee_block = cpu_to_le32(iblock + max_blocks); | ||
2141 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | ||
2142 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | ||
2143 | ext4_ext_mark_uninitialized(ex3); | ||
2144 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | ||
2145 | if (err) | ||
2146 | goto out; | ||
2147 | /* | ||
2148 | * The depth, and hence eh & ex might change | ||
2149 | * as part of the insert above. | ||
2150 | */ | ||
2151 | newdepth = ext_depth(inode); | ||
2152 | if (newdepth != depth) { | ||
2153 | depth = newdepth; | ||
2154 | path = ext4_ext_find_extent(inode, iblock, NULL); | ||
2155 | if (IS_ERR(path)) { | ||
2156 | err = PTR_ERR(path); | ||
2157 | path = NULL; | ||
2158 | goto out; | ||
2159 | } | ||
2160 | eh = path[depth].p_hdr; | ||
2161 | ex = path[depth].p_ext; | ||
2162 | if (ex2 != &newex) | ||
2163 | ex2 = ex; | ||
2164 | } | ||
2165 | allocated = max_blocks; | ||
2166 | } | ||
2167 | /* | ||
2168 | * If there was a change of depth as part of the | ||
2169 | * insertion of ex3 above, we need to update the length | ||
2170 | * of the ex1 extent again here | ||
2171 | */ | ||
2172 | if (ex1 && ex1 != ex) { | ||
2173 | ex1 = ex; | ||
2174 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | ||
2175 | ext4_ext_mark_uninitialized(ex1); | ||
2176 | ex2 = &newex; | ||
2177 | } | ||
2178 | /* ex2: iblock to iblock + maxblocks-1 : initialised */ | ||
2179 | ex2->ee_block = cpu_to_le32(iblock); | ||
2180 | ex2->ee_start = cpu_to_le32(newblock); | ||
2181 | ext4_ext_store_pblock(ex2, newblock); | ||
2182 | ex2->ee_len = cpu_to_le16(allocated); | ||
2183 | if (ex2 != ex) | ||
2184 | goto insert; | ||
2185 | err = ext4_ext_get_access(handle, inode, path + depth); | ||
2186 | if (err) | ||
2187 | goto out; | ||
2188 | /* | ||
2189 | * New (initialized) extent starts from the first block | ||
2190 | * in the current extent. i.e., ex2 == ex | ||
2191 | * We have to see if it can be merged with the extent | ||
2192 | * on the left. | ||
2193 | */ | ||
2194 | if (ex2 > EXT_FIRST_EXTENT(eh)) { | ||
2195 | /* | ||
2196 | * To merge left, pass "ex2 - 1" to try_to_merge(), | ||
2197 | * since it merges towards right _only_. | ||
2198 | */ | ||
2199 | ret = ext4_ext_try_to_merge(inode, path, ex2 - 1); | ||
2200 | if (ret) { | ||
2201 | err = ext4_ext_correct_indexes(handle, inode, path); | ||
2202 | if (err) | ||
2203 | goto out; | ||
2204 | depth = ext_depth(inode); | ||
2205 | ex2--; | ||
2206 | } | ||
2207 | } | ||
2208 | /* | ||
2209 | * Try to Merge towards right. This might be required | ||
2210 | * only when the whole extent is being written to. | ||
2211 | * i.e. ex2 == ex and ex3 == NULL. | ||
2212 | */ | ||
2213 | if (!ex3) { | ||
2214 | ret = ext4_ext_try_to_merge(inode, path, ex2); | ||
2215 | if (ret) { | ||
2216 | err = ext4_ext_correct_indexes(handle, inode, path); | ||
2217 | if (err) | ||
2218 | goto out; | ||
2219 | } | ||
2220 | } | ||
2221 | /* Mark modified extent as dirty */ | ||
2222 | err = ext4_ext_dirty(handle, inode, path + depth); | ||
2223 | goto out; | ||
2224 | insert: | ||
2225 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | ||
2226 | out: | ||
2227 | return err ? err : allocated; | ||
2228 | } | ||
2229 | |||
1980 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | 2230 | int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, |
1981 | ext4_fsblk_t iblock, | 2231 | ext4_fsblk_t iblock, |
1982 | unsigned long max_blocks, struct buffer_head *bh_result, | 2232 | unsigned long max_blocks, struct buffer_head *bh_result, |
1983 | int create, int extend_disksize) | 2233 | int create, int extend_disksize) |
1984 | { | 2234 | { |
1985 | struct ext4_ext_path *path = NULL; | 2235 | struct ext4_ext_path *path = NULL; |
2236 | struct ext4_extent_header *eh; | ||
1986 | struct ext4_extent newex, *ex; | 2237 | struct ext4_extent newex, *ex; |
1987 | ext4_fsblk_t goal, newblock; | 2238 | ext4_fsblk_t goal, newblock; |
1988 | int err = 0, depth; | 2239 | int err = 0, depth, ret; |
1989 | unsigned long allocated = 0; | 2240 | unsigned long allocated = 0; |
1990 | 2241 | ||
1991 | __clear_bit(BH_New, &bh_result->b_state); | 2242 | __clear_bit(BH_New, &bh_result->b_state); |
@@ -1998,8 +2249,10 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
1998 | if (goal) { | 2249 | if (goal) { |
1999 | if (goal == EXT4_EXT_CACHE_GAP) { | 2250 | if (goal == EXT4_EXT_CACHE_GAP) { |
2000 | if (!create) { | 2251 | if (!create) { |
2001 | /* block isn't allocated yet and | 2252 | /* |
2002 | * user doesn't want to allocate it */ | 2253 | * block isn't allocated yet and |
2254 | * user doesn't want to allocate it | ||
2255 | */ | ||
2003 | goto out2; | 2256 | goto out2; |
2004 | } | 2257 | } |
2005 | /* we should allocate requested block */ | 2258 | /* we should allocate requested block */ |
@@ -2033,21 +2286,19 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
2033 | * this is why assert can't be put in ext4_ext_find_extent() | 2286 | * this is why assert can't be put in ext4_ext_find_extent() |
2034 | */ | 2287 | */ |
2035 | BUG_ON(path[depth].p_ext == NULL && depth != 0); | 2288 | BUG_ON(path[depth].p_ext == NULL && depth != 0); |
2289 | eh = path[depth].p_hdr; | ||
2036 | 2290 | ||
2037 | ex = path[depth].p_ext; | 2291 | ex = path[depth].p_ext; |
2038 | if (ex) { | 2292 | if (ex) { |
2039 | unsigned long ee_block = le32_to_cpu(ex->ee_block); | 2293 | unsigned long ee_block = le32_to_cpu(ex->ee_block); |
2040 | ext4_fsblk_t ee_start = ext_pblock(ex); | 2294 | ext4_fsblk_t ee_start = ext_pblock(ex); |
2041 | unsigned short ee_len = le16_to_cpu(ex->ee_len); | 2295 | unsigned short ee_len; |
2042 | 2296 | ||
2043 | /* | 2297 | /* |
2044 | * Allow future support for preallocated extents to be added | ||
2045 | * as an RO_COMPAT feature: | ||
2046 | * Uninitialized extents are treated as holes, except that | 2298 | * Uninitialized extents are treated as holes, except that |
2047 | * we avoid (fail) allocating new blocks during a write. | 2299 | * we split out initialized portions during a write. |
2048 | */ | 2300 | */ |
2049 | if (ee_len > EXT_MAX_LEN) | 2301 | ee_len = ext4_ext_get_actual_len(ex); |
2050 | goto out2; | ||
2051 | /* if found extent covers block, simply return it */ | 2302 | /* if found extent covers block, simply return it */ |
2052 | if (iblock >= ee_block && iblock < ee_block + ee_len) { | 2303 | if (iblock >= ee_block && iblock < ee_block + ee_len) { |
2053 | newblock = iblock - ee_block + ee_start; | 2304 | newblock = iblock - ee_block + ee_start; |
@@ -2055,9 +2306,27 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
2055 | allocated = ee_len - (iblock - ee_block); | 2306 | allocated = ee_len - (iblock - ee_block); |
2056 | ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock, | 2307 | ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock, |
2057 | ee_block, ee_len, newblock); | 2308 | ee_block, ee_len, newblock); |
2058 | ext4_ext_put_in_cache(inode, ee_block, ee_len, | 2309 | |
2059 | ee_start, EXT4_EXT_CACHE_EXTENT); | 2310 | /* Do not put uninitialized extent in the cache */ |
2060 | goto out; | 2311 | if (!ext4_ext_is_uninitialized(ex)) { |
2312 | ext4_ext_put_in_cache(inode, ee_block, | ||
2313 | ee_len, ee_start, | ||
2314 | EXT4_EXT_CACHE_EXTENT); | ||
2315 | goto out; | ||
2316 | } | ||
2317 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) | ||
2318 | goto out; | ||
2319 | if (!create) | ||
2320 | goto out2; | ||
2321 | |||
2322 | ret = ext4_ext_convert_to_initialized(handle, inode, | ||
2323 | path, iblock, | ||
2324 | max_blocks); | ||
2325 | if (ret <= 0) | ||
2326 | goto out2; | ||
2327 | else | ||
2328 | allocated = ret; | ||
2329 | goto outnew; | ||
2061 | } | 2330 | } |
2062 | } | 2331 | } |
2063 | 2332 | ||
@@ -2066,8 +2335,10 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
2066 | * we couldn't try to create block if create flag is zero | 2335 | * we couldn't try to create block if create flag is zero |
2067 | */ | 2336 | */ |
2068 | if (!create) { | 2337 | if (!create) { |
2069 | /* put just found gap into cache to speed up | 2338 | /* |
2070 | * subsequent requests */ | 2339 | * put just found gap into cache to speed up |
2340 | * subsequent requests | ||
2341 | */ | ||
2071 | ext4_ext_put_gap_in_cache(inode, path, iblock); | 2342 | ext4_ext_put_gap_in_cache(inode, path, iblock); |
2072 | goto out2; | 2343 | goto out2; |
2073 | } | 2344 | } |
@@ -2081,6 +2352,19 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
2081 | /* allocate new block */ | 2352 | /* allocate new block */ |
2082 | goal = ext4_ext_find_goal(inode, path, iblock); | 2353 | goal = ext4_ext_find_goal(inode, path, iblock); |
2083 | 2354 | ||
2355 | /* | ||
2356 | * See if request is beyond maximum number of blocks we can have in | ||
2357 | * a single extent. For an initialized extent this limit is | ||
2358 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | ||
2359 | * EXT_UNINIT_MAX_LEN. | ||
2360 | */ | ||
2361 | if (max_blocks > EXT_INIT_MAX_LEN && | ||
2362 | create != EXT4_CREATE_UNINITIALIZED_EXT) | ||
2363 | max_blocks = EXT_INIT_MAX_LEN; | ||
2364 | else if (max_blocks > EXT_UNINIT_MAX_LEN && | ||
2365 | create == EXT4_CREATE_UNINITIALIZED_EXT) | ||
2366 | max_blocks = EXT_UNINIT_MAX_LEN; | ||
2367 | |||
2084 | /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ | 2368 | /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */ |
2085 | newex.ee_block = cpu_to_le32(iblock); | 2369 | newex.ee_block = cpu_to_le32(iblock); |
2086 | newex.ee_len = cpu_to_le16(max_blocks); | 2370 | newex.ee_len = cpu_to_le16(max_blocks); |
@@ -2098,6 +2382,8 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
2098 | /* try to insert new extent into found leaf and return */ | 2382 | /* try to insert new extent into found leaf and return */ |
2099 | ext4_ext_store_pblock(&newex, newblock); | 2383 | ext4_ext_store_pblock(&newex, newblock); |
2100 | newex.ee_len = cpu_to_le16(allocated); | 2384 | newex.ee_len = cpu_to_le16(allocated); |
2385 | if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */ | ||
2386 | ext4_ext_mark_uninitialized(&newex); | ||
2101 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | 2387 | err = ext4_ext_insert_extent(handle, inode, path, &newex); |
2102 | if (err) { | 2388 | if (err) { |
2103 | /* free data blocks we just allocated */ | 2389 | /* free data blocks we just allocated */ |
@@ -2111,10 +2397,13 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
2111 | 2397 | ||
2112 | /* previous routine could use block we allocated */ | 2398 | /* previous routine could use block we allocated */ |
2113 | newblock = ext_pblock(&newex); | 2399 | newblock = ext_pblock(&newex); |
2400 | outnew: | ||
2114 | __set_bit(BH_New, &bh_result->b_state); | 2401 | __set_bit(BH_New, &bh_result->b_state); |
2115 | 2402 | ||
2116 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, | 2403 | /* Cache only when it is _not_ an uninitialized extent */ |
2117 | EXT4_EXT_CACHE_EXTENT); | 2404 | if (create != EXT4_CREATE_UNINITIALIZED_EXT) |
2405 | ext4_ext_put_in_cache(inode, iblock, allocated, newblock, | ||
2406 | EXT4_EXT_CACHE_EXTENT); | ||
2118 | out: | 2407 | out: |
2119 | if (allocated > max_blocks) | 2408 | if (allocated > max_blocks) |
2120 | allocated = max_blocks; | 2409 | allocated = max_blocks; |
@@ -2178,7 +2467,8 @@ void ext4_ext_truncate(struct inode * inode, struct page *page) | |||
2178 | err = ext4_ext_remove_space(inode, last_block); | 2467 | err = ext4_ext_remove_space(inode, last_block); |
2179 | 2468 | ||
2180 | /* In a multi-transaction truncate, we only make the final | 2469 | /* In a multi-transaction truncate, we only make the final |
2181 | * transaction synchronous. */ | 2470 | * transaction synchronous. |
2471 | */ | ||
2182 | if (IS_SYNC(inode)) | 2472 | if (IS_SYNC(inode)) |
2183 | handle->h_sync = 1; | 2473 | handle->h_sync = 1; |
2184 | 2474 | ||
@@ -2217,3 +2507,127 @@ int ext4_ext_writepage_trans_blocks(struct inode *inode, int num) | |||
2217 | 2507 | ||
2218 | return needed; | 2508 | return needed; |
2219 | } | 2509 | } |
2510 | |||
2511 | /* | ||
2512 | * preallocate space for a file. This implements ext4's fallocate inode | ||
2513 | * operation, which gets called from sys_fallocate system call. | ||
2514 | * For block-mapped files, posix_fallocate should fall back to the method | ||
2515 | * of writing zeroes to the required new blocks (the same behavior which is | ||
2516 | * expected for file systems which do not support fallocate() system call). | ||
2517 | */ | ||
2518 | long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len) | ||
2519 | { | ||
2520 | handle_t *handle; | ||
2521 | ext4_fsblk_t block, max_blocks; | ||
2522 | ext4_fsblk_t nblocks = 0; | ||
2523 | int ret = 0; | ||
2524 | int ret2 = 0; | ||
2525 | int retries = 0; | ||
2526 | struct buffer_head map_bh; | ||
2527 | unsigned int credits, blkbits = inode->i_blkbits; | ||
2528 | |||
2529 | /* | ||
2530 | * currently supporting (pre)allocate mode for extent-based | ||
2531 | * files _only_ | ||
2532 | */ | ||
2533 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | ||
2534 | return -EOPNOTSUPP; | ||
2535 | |||
2536 | /* preallocation to directories is currently not supported */ | ||
2537 | if (S_ISDIR(inode->i_mode)) | ||
2538 | return -ENODEV; | ||
2539 | |||
2540 | block = offset >> blkbits; | ||
2541 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) | ||
2542 | - block; | ||
2543 | |||
2544 | /* | ||
2545 | * credits to insert 1 extent into extent tree + buffers to be able to | ||
2546 | * modify 1 super block, 1 block bitmap and 1 group descriptor. | ||
2547 | */ | ||
2548 | credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3; | ||
2549 | retry: | ||
2550 | while (ret >= 0 && ret < max_blocks) { | ||
2551 | block = block + ret; | ||
2552 | max_blocks = max_blocks - ret; | ||
2553 | handle = ext4_journal_start(inode, credits); | ||
2554 | if (IS_ERR(handle)) { | ||
2555 | ret = PTR_ERR(handle); | ||
2556 | break; | ||
2557 | } | ||
2558 | |||
2559 | ret = ext4_ext_get_blocks(handle, inode, block, | ||
2560 | max_blocks, &map_bh, | ||
2561 | EXT4_CREATE_UNINITIALIZED_EXT, 0); | ||
2562 | WARN_ON(!ret); | ||
2563 | if (!ret) { | ||
2564 | ext4_error(inode->i_sb, "ext4_fallocate", | ||
2565 | "ext4_ext_get_blocks returned 0! inode#%lu" | ||
2566 | ", block=%llu, max_blocks=%llu", | ||
2567 | inode->i_ino, block, max_blocks); | ||
2568 | ret = -EIO; | ||
2569 | ext4_mark_inode_dirty(handle, inode); | ||
2570 | ret2 = ext4_journal_stop(handle); | ||
2571 | break; | ||
2572 | } | ||
2573 | if (ret > 0) { | ||
2574 | /* check wrap through sign-bit/zero here */ | ||
2575 | if ((block + ret) < 0 || (block + ret) < block) { | ||
2576 | ret = -EIO; | ||
2577 | ext4_mark_inode_dirty(handle, inode); | ||
2578 | ret2 = ext4_journal_stop(handle); | ||
2579 | break; | ||
2580 | } | ||
2581 | if (buffer_new(&map_bh) && ((block + ret) > | ||
2582 | (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits) | ||
2583 | >> blkbits))) | ||
2584 | nblocks = nblocks + ret; | ||
2585 | } | ||
2586 | |||
2587 | /* Update ctime if new blocks get allocated */ | ||
2588 | if (nblocks) { | ||
2589 | struct timespec now; | ||
2590 | |||
2591 | now = current_fs_time(inode->i_sb); | ||
2592 | if (!timespec_equal(&inode->i_ctime, &now)) | ||
2593 | inode->i_ctime = now; | ||
2594 | } | ||
2595 | |||
2596 | ext4_mark_inode_dirty(handle, inode); | ||
2597 | ret2 = ext4_journal_stop(handle); | ||
2598 | if (ret2) | ||
2599 | break; | ||
2600 | } | ||
2601 | |||
2602 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | ||
2603 | goto retry; | ||
2604 | |||
2605 | /* | ||
2606 | * Time to update the file size. | ||
2607 | * Update only when preallocation was requested beyond the file size. | ||
2608 | */ | ||
2609 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | ||
2610 | (offset + len) > i_size_read(inode)) { | ||
2611 | if (ret > 0) { | ||
2612 | /* | ||
2613 | * if no error, we assume preallocation succeeded | ||
2614 | * completely | ||
2615 | */ | ||
2616 | mutex_lock(&inode->i_mutex); | ||
2617 | i_size_write(inode, offset + len); | ||
2618 | EXT4_I(inode)->i_disksize = i_size_read(inode); | ||
2619 | mutex_unlock(&inode->i_mutex); | ||
2620 | } else if (ret < 0 && nblocks) { | ||
2621 | /* Handle partial allocation scenario */ | ||
2622 | loff_t newsize; | ||
2623 | |||
2624 | mutex_lock(&inode->i_mutex); | ||
2625 | newsize = (nblocks << blkbits) + i_size_read(inode); | ||
2626 | i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits)); | ||
2627 | EXT4_I(inode)->i_disksize = i_size_read(inode); | ||
2628 | mutex_unlock(&inode->i_mutex); | ||
2629 | } | ||
2630 | } | ||
2631 | |||
2632 | return ret > 0 ? ret2 : ret; | ||
2633 | } | ||