blob: 926186a787a8bebb74b0199ddd24d57eeef24c9d [file] [log] [blame]
Alex Tomasa86c6182006-10-11 01:21:03 -07001/*
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>
35#include <linux/ext4_jbd2.h>
36#include <linux/jbd.h>
37#include <linux/smp_lock.h>
38#include <linux/highuid.h>
39#include <linux/pagemap.h>
40#include <linux/quotaops.h>
41#include <linux/string.h>
42#include <linux/slab.h>
43#include <linux/ext4_fs_extents.h>
44#include <asm/uaccess.h>
45
46
Randy Dunlapd0d856e2006-10-11 01:21:07 -070047/*
48 * ext_pblock:
49 * combine low and high parts of physical block number into ext4_fsblk_t
50 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -070051static inline ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
52{
53 ext4_fsblk_t block;
54
55 block = le32_to_cpu(ex->ee_start);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070056 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070057 return block;
58}
59
Randy Dunlapd0d856e2006-10-11 01:21:07 -070060/*
61 * idx_pblock:
62 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
63 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -070064static inline ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
65{
66 ext4_fsblk_t block;
67
68 block = le32_to_cpu(ix->ei_leaf);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070069 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070070 return block;
71}
72
Randy Dunlapd0d856e2006-10-11 01:21:07 -070073/*
74 * ext4_ext_store_pblock:
75 * stores a large physical block number into an extent struct,
76 * breaking it into parts
77 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -070078static inline void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
79{
80 ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070081 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070082}
83
Randy Dunlapd0d856e2006-10-11 01:21:07 -070084/*
85 * ext4_idx_store_pblock:
86 * stores a large physical block number into an index struct,
87 * breaking it into parts
88 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -070089static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
90{
91 ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070092 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070093}
94
Alex Tomasa86c6182006-10-11 01:21:03 -070095static int ext4_ext_check_header(const char *function, struct inode *inode,
96 struct ext4_extent_header *eh)
97{
98 const char *error_msg = NULL;
99
100 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
101 error_msg = "invalid magic";
102 goto corrupted;
103 }
104 if (unlikely(eh->eh_max == 0)) {
105 error_msg = "invalid eh_max";
106 goto corrupted;
107 }
108 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
109 error_msg = "invalid eh_entries";
110 goto corrupted;
111 }
112 return 0;
113
114corrupted:
115 ext4_error(inode->i_sb, function,
116 "bad header in inode #%lu: %s - magic %x, "
117 "entries %u, max %u, depth %u",
118 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
119 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
120 le16_to_cpu(eh->eh_depth));
121
122 return -EIO;
123}
124
125static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
126{
127 int err;
128
129 if (handle->h_buffer_credits > needed)
130 return handle;
131 if (!ext4_journal_extend(handle, needed))
132 return handle;
133 err = ext4_journal_restart(handle, needed);
134
135 return handle;
136}
137
138/*
139 * could return:
140 * - EROFS
141 * - ENOMEM
142 */
143static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
144 struct ext4_ext_path *path)
145{
146 if (path->p_bh) {
147 /* path points to block */
148 return ext4_journal_get_write_access(handle, path->p_bh);
149 }
150 /* path points to leaf/index in inode body */
151 /* we use in-core data, no need to protect them */
152 return 0;
153}
154
155/*
156 * could return:
157 * - EROFS
158 * - ENOMEM
159 * - EIO
160 */
161static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
162 struct ext4_ext_path *path)
163{
164 int err;
165 if (path->p_bh) {
166 /* path points to block */
167 err = ext4_journal_dirty_metadata(handle, path->p_bh);
168 } else {
169 /* path points to leaf/index in inode body */
170 err = ext4_mark_inode_dirty(handle, inode);
171 }
172 return err;
173}
174
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700175static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700176 struct ext4_ext_path *path,
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700177 ext4_fsblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700178{
179 struct ext4_inode_info *ei = EXT4_I(inode);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700180 ext4_fsblk_t bg_start;
181 ext4_grpblk_t colour;
Alex Tomasa86c6182006-10-11 01:21:03 -0700182 int depth;
183
184 if (path) {
185 struct ext4_extent *ex;
186 depth = path->p_depth;
187
188 /* try to predict block placement */
189 if ((ex = path[depth].p_ext))
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700190 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700191
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700192 /* it looks like index is empty;
193 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700194 if (path[depth].p_bh)
195 return path[depth].p_bh->b_blocknr;
196 }
197
198 /* OK. use inode's group */
199 bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
200 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
201 colour = (current->pid % 16) *
202 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
203 return bg_start + colour + block;
204}
205
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700206static ext4_fsblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -0700207ext4_ext_new_block(handle_t *handle, struct inode *inode,
208 struct ext4_ext_path *path,
209 struct ext4_extent *ex, int *err)
210{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700211 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700212
213 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
214 newblock = ext4_new_block(handle, inode, goal, err);
215 return newblock;
216}
217
218static inline int ext4_ext_space_block(struct inode *inode)
219{
220 int size;
221
222 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
223 / sizeof(struct ext4_extent);
224#ifdef AGRESSIVE_TEST
225 if (size > 6)
226 size = 6;
227#endif
228 return size;
229}
230
231static inline int ext4_ext_space_block_idx(struct inode *inode)
232{
233 int size;
234
235 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
236 / sizeof(struct ext4_extent_idx);
237#ifdef AGRESSIVE_TEST
238 if (size > 5)
239 size = 5;
240#endif
241 return size;
242}
243
244static inline int ext4_ext_space_root(struct inode *inode)
245{
246 int size;
247
248 size = sizeof(EXT4_I(inode)->i_data);
249 size -= sizeof(struct ext4_extent_header);
250 size /= sizeof(struct ext4_extent);
251#ifdef AGRESSIVE_TEST
252 if (size > 3)
253 size = 3;
254#endif
255 return size;
256}
257
258static inline int ext4_ext_space_root_idx(struct inode *inode)
259{
260 int size;
261
262 size = sizeof(EXT4_I(inode)->i_data);
263 size -= sizeof(struct ext4_extent_header);
264 size /= sizeof(struct ext4_extent_idx);
265#ifdef AGRESSIVE_TEST
266 if (size > 4)
267 size = 4;
268#endif
269 return size;
270}
271
272#ifdef EXT_DEBUG
273static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
274{
275 int k, l = path->p_depth;
276
277 ext_debug("path:");
278 for (k = 0; k <= l; k++, path++) {
279 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700280 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700281 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700282 } else if (path->p_ext) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700283 ext_debug(" %d:%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700284 le32_to_cpu(path->p_ext->ee_block),
285 le16_to_cpu(path->p_ext->ee_len),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700286 ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700287 } else
288 ext_debug(" []");
289 }
290 ext_debug("\n");
291}
292
293static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
294{
295 int depth = ext_depth(inode);
296 struct ext4_extent_header *eh;
297 struct ext4_extent *ex;
298 int i;
299
300 if (!path)
301 return;
302
303 eh = path[depth].p_hdr;
304 ex = EXT_FIRST_EXTENT(eh);
305
306 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700307 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700308 le16_to_cpu(ex->ee_len), ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700309 }
310 ext_debug("\n");
311}
312#else
313#define ext4_ext_show_path(inode,path)
314#define ext4_ext_show_leaf(inode,path)
315#endif
316
317static void ext4_ext_drop_refs(struct ext4_ext_path *path)
318{
319 int depth = path->p_depth;
320 int i;
321
322 for (i = 0; i <= depth; i++, path++)
323 if (path->p_bh) {
324 brelse(path->p_bh);
325 path->p_bh = NULL;
326 }
327}
328
329/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700330 * ext4_ext_binsearch_idx:
331 * binary search for the closest index of the given block
Alex Tomasa86c6182006-10-11 01:21:03 -0700332 */
333static void
334ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block)
335{
336 struct ext4_extent_header *eh = path->p_hdr;
337 struct ext4_extent_idx *r, *l, *m;
338
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
343 ext_debug("binsearch for %d(idx): ", block);
344
345 l = EXT_FIRST_INDEX(eh) + 1;
346 r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
347 while (l <= r) {
348 m = l + (r - l) / 2;
349 if (block < le32_to_cpu(m->ei_block))
350 r = m - 1;
351 else
352 l = m + 1;
353 ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block,
354 m, m->ei_block, r, r->ei_block);
355 }
356
357 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700358 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
359 idx_block(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700360
361#ifdef CHECK_BINSEARCH
362 {
363 struct ext4_extent_idx *chix, *ix;
364 int k;
365
366 chix = ix = EXT_FIRST_INDEX(eh);
367 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
368 if (k != 0 &&
369 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
370 printk("k=%d, ix=0x%p, first=0x%p\n", k,
371 ix, EXT_FIRST_INDEX(eh));
372 printk("%u <= %u\n",
373 le32_to_cpu(ix->ei_block),
374 le32_to_cpu(ix[-1].ei_block));
375 }
376 BUG_ON(k && le32_to_cpu(ix->ei_block)
377 <= le32_to_cpu(ix[-1].ei_block));
378 if (block < le32_to_cpu(ix->ei_block))
379 break;
380 chix = ix;
381 }
382 BUG_ON(chix != path->p_idx);
383 }
384#endif
385
386}
387
388/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700389 * ext4_ext_binsearch:
390 * binary search for closest extent of the given block
Alex Tomasa86c6182006-10-11 01:21:03 -0700391 */
392static void
393ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
394{
395 struct ext4_extent_header *eh = path->p_hdr;
396 struct ext4_extent *r, *l, *m;
397
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) {
402 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700403 * this leaf is empty:
404 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700405 */
406 return;
407 }
408
409 ext_debug("binsearch for %d: ", block);
410
411 l = EXT_FIRST_EXTENT(eh) + 1;
412 r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
413
414 while (l <= r) {
415 m = l + (r - l) / 2;
416 if (block < le32_to_cpu(m->ee_block))
417 r = m - 1;
418 else
419 l = m + 1;
420 ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block,
421 m, m->ee_block, r, r->ee_block);
422 }
423
424 path->p_ext = l - 1;
Mingming Cao2ae02102006-10-11 01:21:11 -0700425 ext_debug(" -> %d:%llu:%d ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700426 le32_to_cpu(path->p_ext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700427 ext_pblock(path->p_ext),
428 le16_to_cpu(path->p_ext->ee_len));
Alex Tomasa86c6182006-10-11 01:21:03 -0700429
430#ifdef CHECK_BINSEARCH
431 {
432 struct ext4_extent *chex, *ex;
433 int k;
434
435 chex = ex = EXT_FIRST_EXTENT(eh);
436 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
437 BUG_ON(k && le32_to_cpu(ex->ee_block)
438 <= le32_to_cpu(ex[-1].ee_block));
439 if (block < le32_to_cpu(ex->ee_block))
440 break;
441 chex = ex;
442 }
443 BUG_ON(chex != path->p_ext);
444 }
445#endif
446
447}
448
449int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
450{
451 struct ext4_extent_header *eh;
452
453 eh = ext_inode_hdr(inode);
454 eh->eh_depth = 0;
455 eh->eh_entries = 0;
456 eh->eh_magic = EXT4_EXT_MAGIC;
457 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
458 ext4_mark_inode_dirty(handle, inode);
459 ext4_ext_invalidate_cache(inode);
460 return 0;
461}
462
463struct ext4_ext_path *
464ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
465{
466 struct ext4_extent_header *eh;
467 struct buffer_head *bh;
468 short int depth, i, ppos = 0, alloc = 0;
469
470 eh = ext_inode_hdr(inode);
471 BUG_ON(eh == NULL);
472 if (ext4_ext_check_header(__FUNCTION__, inode, eh))
473 return ERR_PTR(-EIO);
474
475 i = depth = ext_depth(inode);
476
477 /* account possible depth increase */
478 if (!path) {
479 path = kmalloc(sizeof(struct ext4_ext_path) * (depth + 2),
480 GFP_NOFS);
481 if (!path)
482 return ERR_PTR(-ENOMEM);
483 alloc = 1;
484 }
485 memset(path, 0, sizeof(struct ext4_ext_path) * (depth + 1));
486 path[0].p_hdr = eh;
487
488 /* walk through the tree */
489 while (i) {
490 ext_debug("depth %d: num %d, max %d\n",
491 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
492 ext4_ext_binsearch_idx(inode, path + ppos, block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700493 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700494 path[ppos].p_depth = i;
495 path[ppos].p_ext = NULL;
496
497 bh = sb_bread(inode->i_sb, path[ppos].p_block);
498 if (!bh)
499 goto err;
500
501 eh = ext_block_hdr(bh);
502 ppos++;
503 BUG_ON(ppos > depth);
504 path[ppos].p_bh = bh;
505 path[ppos].p_hdr = eh;
506 i--;
507
508 if (ext4_ext_check_header(__FUNCTION__, inode, eh))
509 goto err;
510 }
511
512 path[ppos].p_depth = i;
513 path[ppos].p_hdr = eh;
514 path[ppos].p_ext = NULL;
515 path[ppos].p_idx = NULL;
516
517 if (ext4_ext_check_header(__FUNCTION__, inode, eh))
518 goto err;
519
520 /* find extent */
521 ext4_ext_binsearch(inode, path + ppos, block);
522
523 ext4_ext_show_path(inode, path);
524
525 return path;
526
527err:
528 ext4_ext_drop_refs(path);
529 if (alloc)
530 kfree(path);
531 return ERR_PTR(-EIO);
532}
533
534/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700535 * ext4_ext_insert_index:
536 * insert new index [@logical;@ptr] into the block at @curp;
537 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700538 */
539static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
540 struct ext4_ext_path *curp,
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700541 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700542{
543 struct ext4_extent_idx *ix;
544 int len, err;
545
546 if ((err = ext4_ext_get_access(handle, inode, curp)))
547 return err;
548
549 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
550 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
551 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
552 /* insert after */
553 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
554 len = (len - 1) * sizeof(struct ext4_extent_idx);
555 len = len < 0 ? 0 : len;
556 ext_debug("insert new index %d after: %d. "
557 "move %d from 0x%p to 0x%p\n",
558 logical, ptr, len,
559 (curp->p_idx + 1), (curp->p_idx + 2));
560 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
561 }
562 ix = curp->p_idx + 1;
563 } else {
564 /* insert before */
565 len = len * sizeof(struct ext4_extent_idx);
566 len = len < 0 ? 0 : len;
567 ext_debug("insert new index %d before: %d. "
568 "move %d from 0x%p to 0x%p\n",
569 logical, ptr, len,
570 curp->p_idx, (curp->p_idx + 1));
571 memmove(curp->p_idx + 1, curp->p_idx, len);
572 ix = curp->p_idx;
573 }
574
575 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700576 ext4_idx_store_pblock(ix, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700577 curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
578
579 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
580 > le16_to_cpu(curp->p_hdr->eh_max));
581 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
582
583 err = ext4_ext_dirty(handle, inode, curp);
584 ext4_std_error(inode->i_sb, err);
585
586 return err;
587}
588
589/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700590 * ext4_ext_split:
591 * inserts new subtree into the path, using free index entry
592 * at depth @at:
593 * - allocates all needed blocks (new leaf and all intermediate index blocks)
594 * - makes decision where to split
595 * - moves remaining extents and index entries (right to the split point)
596 * into the newly allocated blocks
597 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700598 */
599static int ext4_ext_split(handle_t *handle, struct inode *inode,
600 struct ext4_ext_path *path,
601 struct ext4_extent *newext, int at)
602{
603 struct buffer_head *bh = NULL;
604 int depth = ext_depth(inode);
605 struct ext4_extent_header *neh;
606 struct ext4_extent_idx *fidx;
607 struct ext4_extent *ex;
608 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700609 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700610 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700611 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700612 int err = 0;
613
614 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700615 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700616
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700617 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700618 * border from split point */
619 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
620 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
621 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700622 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700623 " next leaf starts at %d\n",
624 le32_to_cpu(border));
625 } else {
626 border = newext->ee_block;
627 ext_debug("leaf will be added."
628 " next leaf starts at %d\n",
629 le32_to_cpu(border));
630 }
631
632 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700633 * If error occurs, then we break processing
634 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700635 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700636 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700637 */
638
639 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700640 * Get array to track all allocated blocks.
641 * We need this to handle errors and free blocks
642 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700643 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700644 ablocks = kmalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700645 if (!ablocks)
646 return -ENOMEM;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700647 memset(ablocks, 0, sizeof(ext4_fsblk_t) * depth);
Alex Tomasa86c6182006-10-11 01:21:03 -0700648
649 /* allocate all needed blocks */
650 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
651 for (a = 0; a < depth - at; a++) {
652 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
653 if (newblock == 0)
654 goto cleanup;
655 ablocks[a] = newblock;
656 }
657
658 /* initialize new leaf */
659 newblock = ablocks[--a];
660 BUG_ON(newblock == 0);
661 bh = sb_getblk(inode->i_sb, newblock);
662 if (!bh) {
663 err = -EIO;
664 goto cleanup;
665 }
666 lock_buffer(bh);
667
668 if ((err = ext4_journal_get_create_access(handle, bh)))
669 goto cleanup;
670
671 neh = ext_block_hdr(bh);
672 neh->eh_entries = 0;
673 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
674 neh->eh_magic = EXT4_EXT_MAGIC;
675 neh->eh_depth = 0;
676 ex = EXT_FIRST_EXTENT(neh);
677
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700678 /* move remainder of path[depth] to the new leaf */
Alex Tomasa86c6182006-10-11 01:21:03 -0700679 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
680 /* start copy from next extent */
681 /* TODO: we could do it by single memmove */
682 m = 0;
683 path[depth].p_ext++;
684 while (path[depth].p_ext <=
685 EXT_MAX_EXTENT(path[depth].p_hdr)) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700686 ext_debug("move %d:%llu:%d in new leaf %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700687 le32_to_cpu(path[depth].p_ext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700688 ext_pblock(path[depth].p_ext),
Alex Tomasa86c6182006-10-11 01:21:03 -0700689 le16_to_cpu(path[depth].p_ext->ee_len),
690 newblock);
691 /*memmove(ex++, path[depth].p_ext++,
692 sizeof(struct ext4_extent));
693 neh->eh_entries++;*/
694 path[depth].p_ext++;
695 m++;
696 }
697 if (m) {
698 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
699 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
700 }
701
702 set_buffer_uptodate(bh);
703 unlock_buffer(bh);
704
705 if ((err = ext4_journal_dirty_metadata(handle, bh)))
706 goto cleanup;
707 brelse(bh);
708 bh = NULL;
709
710 /* correct old leaf */
711 if (m) {
712 if ((err = ext4_ext_get_access(handle, inode, path + depth)))
713 goto cleanup;
714 path[depth].p_hdr->eh_entries =
715 cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
716 if ((err = ext4_ext_dirty(handle, inode, path + depth)))
717 goto cleanup;
718
719 }
720
721 /* create intermediate indexes */
722 k = depth - at - 1;
723 BUG_ON(k < 0);
724 if (k)
725 ext_debug("create %d intermediate indices\n", k);
726 /* insert new index into current index block */
727 /* current depth stored in i var */
728 i = depth - 1;
729 while (k--) {
730 oldblock = newblock;
731 newblock = ablocks[--a];
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700732 bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700733 if (!bh) {
734 err = -EIO;
735 goto cleanup;
736 }
737 lock_buffer(bh);
738
739 if ((err = ext4_journal_get_create_access(handle, bh)))
740 goto cleanup;
741
742 neh = ext_block_hdr(bh);
743 neh->eh_entries = cpu_to_le16(1);
744 neh->eh_magic = EXT4_EXT_MAGIC;
745 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
746 neh->eh_depth = cpu_to_le16(depth - i);
747 fidx = EXT_FIRST_INDEX(neh);
748 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700749 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700750
Mingming Cao2ae02102006-10-11 01:21:11 -0700751 ext_debug("int.index at %d (block %llu): %lu -> %llu\n", i,
Alex Tomasa86c6182006-10-11 01:21:03 -0700752 newblock, (unsigned long) le32_to_cpu(border),
753 oldblock);
754 /* copy indexes */
755 m = 0;
756 path[i].p_idx++;
757
758 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
759 EXT_MAX_INDEX(path[i].p_hdr));
760 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
761 EXT_LAST_INDEX(path[i].p_hdr));
762 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700763 ext_debug("%d: move %d:%d in new index %llu\n", i,
Alex Tomasa86c6182006-10-11 01:21:03 -0700764 le32_to_cpu(path[i].p_idx->ei_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700765 idx_pblock(path[i].p_idx),
Alex Tomasa86c6182006-10-11 01:21:03 -0700766 newblock);
767 /*memmove(++fidx, path[i].p_idx++,
768 sizeof(struct ext4_extent_idx));
769 neh->eh_entries++;
770 BUG_ON(neh->eh_entries > neh->eh_max);*/
771 path[i].p_idx++;
772 m++;
773 }
774 if (m) {
775 memmove(++fidx, path[i].p_idx - m,
776 sizeof(struct ext4_extent_idx) * m);
777 neh->eh_entries =
778 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
779 }
780 set_buffer_uptodate(bh);
781 unlock_buffer(bh);
782
783 if ((err = ext4_journal_dirty_metadata(handle, bh)))
784 goto cleanup;
785 brelse(bh);
786 bh = NULL;
787
788 /* correct old index */
789 if (m) {
790 err = ext4_ext_get_access(handle, inode, path + i);
791 if (err)
792 goto cleanup;
793 path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
794 err = ext4_ext_dirty(handle, inode, path + i);
795 if (err)
796 goto cleanup;
797 }
798
799 i--;
800 }
801
802 /* insert new index */
803 if (err)
804 goto cleanup;
805
806 err = ext4_ext_insert_index(handle, inode, path + at,
807 le32_to_cpu(border), newblock);
808
809cleanup:
810 if (bh) {
811 if (buffer_locked(bh))
812 unlock_buffer(bh);
813 brelse(bh);
814 }
815
816 if (err) {
817 /* free all allocated blocks in error case */
818 for (i = 0; i < depth; i++) {
819 if (!ablocks[i])
820 continue;
821 ext4_free_blocks(handle, inode, ablocks[i], 1);
822 }
823 }
824 kfree(ablocks);
825
826 return err;
827}
828
829/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700830 * ext4_ext_grow_indepth:
831 * implements tree growing procedure:
832 * - allocates new block
833 * - moves top-level data (index block or leaf) into the new block
834 * - initializes new top-level, creating index that points to the
835 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -0700836 */
837static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
838 struct ext4_ext_path *path,
839 struct ext4_extent *newext)
840{
841 struct ext4_ext_path *curp = path;
842 struct ext4_extent_header *neh;
843 struct ext4_extent_idx *fidx;
844 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700845 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700846 int err = 0;
847
848 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
849 if (newblock == 0)
850 return err;
851
852 bh = sb_getblk(inode->i_sb, newblock);
853 if (!bh) {
854 err = -EIO;
855 ext4_std_error(inode->i_sb, err);
856 return err;
857 }
858 lock_buffer(bh);
859
860 if ((err = ext4_journal_get_create_access(handle, bh))) {
861 unlock_buffer(bh);
862 goto out;
863 }
864
865 /* move top-level index/leaf into new block */
866 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
867
868 /* set size of new block */
869 neh = ext_block_hdr(bh);
870 /* old root could have indexes or leaves
871 * so calculate e_max right way */
872 if (ext_depth(inode))
873 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
874 else
875 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
876 neh->eh_magic = EXT4_EXT_MAGIC;
877 set_buffer_uptodate(bh);
878 unlock_buffer(bh);
879
880 if ((err = ext4_journal_dirty_metadata(handle, bh)))
881 goto out;
882
883 /* create index in new top-level index: num,max,pointer */
884 if ((err = ext4_ext_get_access(handle, inode, curp)))
885 goto out;
886
887 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
888 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
889 curp->p_hdr->eh_entries = cpu_to_le16(1);
890 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
891 /* FIXME: it works, but actually path[0] can be index */
892 curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700893 ext4_idx_store_pblock(curp->p_idx, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700894
895 neh = ext_inode_hdr(inode);
896 fidx = EXT_FIRST_INDEX(neh);
Mingming Cao2ae02102006-10-11 01:21:11 -0700897 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700898 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700899 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700900
901 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
902 err = ext4_ext_dirty(handle, inode, curp);
903out:
904 brelse(bh);
905
906 return err;
907}
908
909/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700910 * ext4_ext_create_new_leaf:
911 * finds empty index and adds new leaf.
912 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -0700913 */
914static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
915 struct ext4_ext_path *path,
916 struct ext4_extent *newext)
917{
918 struct ext4_ext_path *curp;
919 int depth, i, err = 0;
920
921repeat:
922 i = depth = ext_depth(inode);
923
924 /* walk up to the tree and look for free index entry */
925 curp = path + depth;
926 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
927 i--;
928 curp--;
929 }
930
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700931 /* we use already allocated block for index block,
932 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -0700933 if (EXT_HAS_FREE_INDEX(curp)) {
934 /* if we found index with free entry, then use that
935 * entry: create all needed subtree and add new leaf */
936 err = ext4_ext_split(handle, inode, path, newext, i);
937
938 /* refill path */
939 ext4_ext_drop_refs(path);
940 path = ext4_ext_find_extent(inode,
941 le32_to_cpu(newext->ee_block),
942 path);
943 if (IS_ERR(path))
944 err = PTR_ERR(path);
945 } else {
946 /* tree is full, time to grow in depth */
947 err = ext4_ext_grow_indepth(handle, inode, path, newext);
948 if (err)
949 goto out;
950
951 /* refill path */
952 ext4_ext_drop_refs(path);
953 path = ext4_ext_find_extent(inode,
954 le32_to_cpu(newext->ee_block),
955 path);
956 if (IS_ERR(path)) {
957 err = PTR_ERR(path);
958 goto out;
959 }
960
961 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700962 * only first (depth 0 -> 1) produces free space;
963 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -0700964 */
965 depth = ext_depth(inode);
966 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700967 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -0700968 goto repeat;
969 }
970 }
971
972out:
973 return err;
974}
975
976/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700977 * ext4_ext_next_allocated_block:
978 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
979 * NOTE: it considers block number from index entry as
980 * allocated block. Thus, index entries have to be consistent
981 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -0700982 */
983static unsigned long
984ext4_ext_next_allocated_block(struct ext4_ext_path *path)
985{
986 int depth;
987
988 BUG_ON(path == NULL);
989 depth = path->p_depth;
990
991 if (depth == 0 && path->p_ext == NULL)
992 return EXT_MAX_BLOCK;
993
994 while (depth >= 0) {
995 if (depth == path->p_depth) {
996 /* leaf */
997 if (path[depth].p_ext !=
998 EXT_LAST_EXTENT(path[depth].p_hdr))
999 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1000 } else {
1001 /* index */
1002 if (path[depth].p_idx !=
1003 EXT_LAST_INDEX(path[depth].p_hdr))
1004 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1005 }
1006 depth--;
1007 }
1008
1009 return EXT_MAX_BLOCK;
1010}
1011
1012/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001013 * ext4_ext_next_leaf_block:
Alex Tomasa86c6182006-10-11 01:21:03 -07001014 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1015 */
1016static unsigned ext4_ext_next_leaf_block(struct inode *inode,
1017 struct ext4_ext_path *path)
1018{
1019 int depth;
1020
1021 BUG_ON(path == NULL);
1022 depth = path->p_depth;
1023
1024 /* zero-tree has no leaf blocks at all */
1025 if (depth == 0)
1026 return EXT_MAX_BLOCK;
1027
1028 /* go to index block */
1029 depth--;
1030
1031 while (depth >= 0) {
1032 if (path[depth].p_idx !=
1033 EXT_LAST_INDEX(path[depth].p_hdr))
1034 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1035 depth--;
1036 }
1037
1038 return EXT_MAX_BLOCK;
1039}
1040
1041/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001042 * ext4_ext_correct_indexes:
1043 * if leaf gets modified and modified extent is first in the leaf,
1044 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001045 * TODO: do we need to correct tree in all cases?
1046 */
1047int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1048 struct ext4_ext_path *path)
1049{
1050 struct ext4_extent_header *eh;
1051 int depth = ext_depth(inode);
1052 struct ext4_extent *ex;
1053 __le32 border;
1054 int k, err = 0;
1055
1056 eh = path[depth].p_hdr;
1057 ex = path[depth].p_ext;
1058 BUG_ON(ex == NULL);
1059 BUG_ON(eh == NULL);
1060
1061 if (depth == 0) {
1062 /* there is no tree at all */
1063 return 0;
1064 }
1065
1066 if (ex != EXT_FIRST_EXTENT(eh)) {
1067 /* we correct tree if first leaf got modified only */
1068 return 0;
1069 }
1070
1071 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001072 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001073 */
1074 k = depth - 1;
1075 border = path[depth].p_ext->ee_block;
1076 if ((err = ext4_ext_get_access(handle, inode, path + k)))
1077 return err;
1078 path[k].p_idx->ei_block = border;
1079 if ((err = ext4_ext_dirty(handle, inode, path + k)))
1080 return err;
1081
1082 while (k--) {
1083 /* change all left-side indexes */
1084 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1085 break;
1086 if ((err = ext4_ext_get_access(handle, inode, path + k)))
1087 break;
1088 path[k].p_idx->ei_block = border;
1089 if ((err = ext4_ext_dirty(handle, inode, path + k)))
1090 break;
1091 }
1092
1093 return err;
1094}
1095
1096static int inline
1097ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1098 struct ext4_extent *ex2)
1099{
Alex Tomasa86c6182006-10-11 01:21:03 -07001100 if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len)
1101 != le32_to_cpu(ex2->ee_block))
1102 return 0;
1103
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001104 /*
1105 * To allow future support for preallocated extents to be added
1106 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001107 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001108 */
1109 if (le16_to_cpu(ex1->ee_len) + le16_to_cpu(ex2->ee_len) > EXT_MAX_LEN)
1110 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001111#ifdef AGRESSIVE_TEST
1112 if (le16_to_cpu(ex1->ee_len) >= 4)
1113 return 0;
1114#endif
1115
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001116 if (ext_pblock(ex1) + le16_to_cpu(ex1->ee_len) == ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001117 return 1;
1118 return 0;
1119}
1120
1121/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001122 * ext4_ext_insert_extent:
1123 * tries to merge requsted extent into the existing extent or
1124 * inserts requested extent as new one into the tree,
1125 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001126 */
1127int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1128 struct ext4_ext_path *path,
1129 struct ext4_extent *newext)
1130{
1131 struct ext4_extent_header * eh;
1132 struct ext4_extent *ex, *fex;
1133 struct ext4_extent *nearex; /* nearest extent */
1134 struct ext4_ext_path *npath = NULL;
1135 int depth, len, err, next;
1136
1137 BUG_ON(newext->ee_len == 0);
1138 depth = ext_depth(inode);
1139 ex = path[depth].p_ext;
1140 BUG_ON(path[depth].p_hdr == NULL);
1141
1142 /* try to insert block into found extent and return */
1143 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming Cao2ae02102006-10-11 01:21:11 -07001144 ext_debug("append %d block to %d:%d (from %llu)\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001145 le16_to_cpu(newext->ee_len),
1146 le32_to_cpu(ex->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001147 le16_to_cpu(ex->ee_len), ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001148 if ((err = ext4_ext_get_access(handle, inode, path + depth)))
1149 return err;
1150 ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len)
1151 + le16_to_cpu(newext->ee_len));
1152 eh = path[depth].p_hdr;
1153 nearex = ex;
1154 goto merge;
1155 }
1156
1157repeat:
1158 depth = ext_depth(inode);
1159 eh = path[depth].p_hdr;
1160 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1161 goto has_space;
1162
1163 /* probably next leaf has space for us? */
1164 fex = EXT_LAST_EXTENT(eh);
1165 next = ext4_ext_next_leaf_block(inode, path);
1166 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1167 && next != EXT_MAX_BLOCK) {
1168 ext_debug("next leaf block - %d\n", next);
1169 BUG_ON(npath != NULL);
1170 npath = ext4_ext_find_extent(inode, next, NULL);
1171 if (IS_ERR(npath))
1172 return PTR_ERR(npath);
1173 BUG_ON(npath->p_depth != path->p_depth);
1174 eh = npath[depth].p_hdr;
1175 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1176 ext_debug("next leaf isnt full(%d)\n",
1177 le16_to_cpu(eh->eh_entries));
1178 path = npath;
1179 goto repeat;
1180 }
1181 ext_debug("next leaf has no free space(%d,%d)\n",
1182 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1183 }
1184
1185 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001186 * There is no free space in the found leaf.
1187 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001188 */
1189 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1190 if (err)
1191 goto cleanup;
1192 depth = ext_depth(inode);
1193 eh = path[depth].p_hdr;
1194
1195has_space:
1196 nearex = path[depth].p_ext;
1197
1198 if ((err = ext4_ext_get_access(handle, inode, path + depth)))
1199 goto cleanup;
1200
1201 if (!nearex) {
1202 /* there is no extent in this leaf, create first one */
Mingming Cao2ae02102006-10-11 01:21:11 -07001203 ext_debug("first extent in the leaf: %d:%llu:%d\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001204 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001205 ext_pblock(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001206 le16_to_cpu(newext->ee_len));
1207 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1208 } else if (le32_to_cpu(newext->ee_block)
1209 > le32_to_cpu(nearex->ee_block)) {
1210/* BUG_ON(newext->ee_block == nearex->ee_block); */
1211 if (nearex != EXT_LAST_EXTENT(eh)) {
1212 len = EXT_MAX_EXTENT(eh) - nearex;
1213 len = (len - 1) * sizeof(struct ext4_extent);
1214 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001215 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001216 "move %d from 0x%p to 0x%p\n",
1217 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001218 ext_pblock(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001219 le16_to_cpu(newext->ee_len),
1220 nearex, len, nearex + 1, nearex + 2);
1221 memmove(nearex + 2, nearex + 1, len);
1222 }
1223 path[depth].p_ext = nearex + 1;
1224 } else {
1225 BUG_ON(newext->ee_block == nearex->ee_block);
1226 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1227 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001228 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001229 "move %d from 0x%p to 0x%p\n",
1230 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001231 ext_pblock(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001232 le16_to_cpu(newext->ee_len),
1233 nearex, len, nearex + 1, nearex + 2);
1234 memmove(nearex + 1, nearex, len);
1235 path[depth].p_ext = nearex;
1236 }
1237
1238 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1239 nearex = path[depth].p_ext;
1240 nearex->ee_block = newext->ee_block;
1241 nearex->ee_start = newext->ee_start;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001242 nearex->ee_start_hi = newext->ee_start_hi;
Alex Tomasa86c6182006-10-11 01:21:03 -07001243 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001244
1245merge:
1246 /* try to merge extents to the right */
1247 while (nearex < EXT_LAST_EXTENT(eh)) {
1248 if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1))
1249 break;
1250 /* merge with next extent! */
1251 nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len)
1252 + le16_to_cpu(nearex[1].ee_len));
1253 if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1254 len = (EXT_LAST_EXTENT(eh) - nearex - 1)
1255 * sizeof(struct ext4_extent);
1256 memmove(nearex + 1, nearex + 2, len);
1257 }
1258 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1259 BUG_ON(eh->eh_entries == 0);
1260 }
1261
1262 /* try to merge extents to the left */
1263
1264 /* time to correct all indexes above */
1265 err = ext4_ext_correct_indexes(handle, inode, path);
1266 if (err)
1267 goto cleanup;
1268
1269 err = ext4_ext_dirty(handle, inode, path + depth);
1270
1271cleanup:
1272 if (npath) {
1273 ext4_ext_drop_refs(npath);
1274 kfree(npath);
1275 }
1276 ext4_ext_tree_changed(inode);
1277 ext4_ext_invalidate_cache(inode);
1278 return err;
1279}
1280
1281int ext4_ext_walk_space(struct inode *inode, unsigned long block,
1282 unsigned long num, ext_prepare_callback func,
1283 void *cbdata)
1284{
1285 struct ext4_ext_path *path = NULL;
1286 struct ext4_ext_cache cbex;
1287 struct ext4_extent *ex;
1288 unsigned long next, start = 0, end = 0;
1289 unsigned long last = block + num;
1290 int depth, exists, err = 0;
1291
1292 BUG_ON(func == NULL);
1293 BUG_ON(inode == NULL);
1294
1295 while (block < last && block != EXT_MAX_BLOCK) {
1296 num = last - block;
1297 /* find extent for this block */
1298 path = ext4_ext_find_extent(inode, block, path);
1299 if (IS_ERR(path)) {
1300 err = PTR_ERR(path);
1301 path = NULL;
1302 break;
1303 }
1304
1305 depth = ext_depth(inode);
1306 BUG_ON(path[depth].p_hdr == NULL);
1307 ex = path[depth].p_ext;
1308 next = ext4_ext_next_allocated_block(path);
1309
1310 exists = 0;
1311 if (!ex) {
1312 /* there is no extent yet, so try to allocate
1313 * all requested space */
1314 start = block;
1315 end = block + num;
1316 } else if (le32_to_cpu(ex->ee_block) > block) {
1317 /* need to allocate space before found extent */
1318 start = block;
1319 end = le32_to_cpu(ex->ee_block);
1320 if (block + num < end)
1321 end = block + num;
1322 } else if (block >=
1323 le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) {
1324 /* need to allocate space after found extent */
1325 start = block;
1326 end = block + num;
1327 if (end >= next)
1328 end = next;
1329 } else if (block >= le32_to_cpu(ex->ee_block)) {
1330 /*
1331 * some part of requested space is covered
1332 * by found extent
1333 */
1334 start = block;
1335 end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len);
1336 if (block + num < end)
1337 end = block + num;
1338 exists = 1;
1339 } else {
1340 BUG();
1341 }
1342 BUG_ON(end <= start);
1343
1344 if (!exists) {
1345 cbex.ec_block = start;
1346 cbex.ec_len = end - start;
1347 cbex.ec_start = 0;
1348 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1349 } else {
1350 cbex.ec_block = le32_to_cpu(ex->ee_block);
1351 cbex.ec_len = le16_to_cpu(ex->ee_len);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001352 cbex.ec_start = ext_pblock(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001353 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1354 }
1355
1356 BUG_ON(cbex.ec_len == 0);
1357 err = func(inode, path, &cbex, cbdata);
1358 ext4_ext_drop_refs(path);
1359
1360 if (err < 0)
1361 break;
1362 if (err == EXT_REPEAT)
1363 continue;
1364 else if (err == EXT_BREAK) {
1365 err = 0;
1366 break;
1367 }
1368
1369 if (ext_depth(inode) != depth) {
1370 /* depth was changed. we have to realloc path */
1371 kfree(path);
1372 path = NULL;
1373 }
1374
1375 block = cbex.ec_block + cbex.ec_len;
1376 }
1377
1378 if (path) {
1379 ext4_ext_drop_refs(path);
1380 kfree(path);
1381 }
1382
1383 return err;
1384}
1385
1386static inline void
1387ext4_ext_put_in_cache(struct inode *inode, __u32 block,
1388 __u32 len, __u32 start, int type)
1389{
1390 struct ext4_ext_cache *cex;
1391 BUG_ON(len == 0);
1392 cex = &EXT4_I(inode)->i_cached_extent;
1393 cex->ec_type = type;
1394 cex->ec_block = block;
1395 cex->ec_len = len;
1396 cex->ec_start = start;
1397}
1398
1399/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001400 * ext4_ext_put_gap_in_cache:
1401 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001402 * and cache this gap
1403 */
1404static inline void
1405ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1406 unsigned long block)
1407{
1408 int depth = ext_depth(inode);
1409 unsigned long lblock, len;
1410 struct ext4_extent *ex;
1411
1412 ex = path[depth].p_ext;
1413 if (ex == NULL) {
1414 /* there is no extent yet, so gap is [0;-] */
1415 lblock = 0;
1416 len = EXT_MAX_BLOCK;
1417 ext_debug("cache gap(whole file):");
1418 } else if (block < le32_to_cpu(ex->ee_block)) {
1419 lblock = block;
1420 len = le32_to_cpu(ex->ee_block) - block;
1421 ext_debug("cache gap(before): %lu [%lu:%lu]",
1422 (unsigned long) block,
1423 (unsigned long) le32_to_cpu(ex->ee_block),
1424 (unsigned long) le16_to_cpu(ex->ee_len));
1425 } else if (block >= le32_to_cpu(ex->ee_block)
1426 + le16_to_cpu(ex->ee_len)) {
1427 lblock = le32_to_cpu(ex->ee_block)
1428 + le16_to_cpu(ex->ee_len);
1429 len = ext4_ext_next_allocated_block(path);
1430 ext_debug("cache gap(after): [%lu:%lu] %lu",
1431 (unsigned long) le32_to_cpu(ex->ee_block),
1432 (unsigned long) le16_to_cpu(ex->ee_len),
1433 (unsigned long) block);
1434 BUG_ON(len == lblock);
1435 len = len - lblock;
1436 } else {
1437 lblock = len = 0;
1438 BUG();
1439 }
1440
1441 ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len);
1442 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1443}
1444
1445static inline int
1446ext4_ext_in_cache(struct inode *inode, unsigned long block,
1447 struct ext4_extent *ex)
1448{
1449 struct ext4_ext_cache *cex;
1450
1451 cex = &EXT4_I(inode)->i_cached_extent;
1452
1453 /* has cache valid data? */
1454 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1455 return EXT4_EXT_CACHE_NO;
1456
1457 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1458 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1459 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1460 ex->ee_block = cpu_to_le32(cex->ec_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001461 ext4_ext_store_pblock(ex, cex->ec_start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001462 ex->ee_len = cpu_to_le16(cex->ec_len);
Mingming Cao2ae02102006-10-11 01:21:11 -07001463 ext_debug("%lu cached by %lu:%lu:%llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001464 (unsigned long) block,
1465 (unsigned long) cex->ec_block,
1466 (unsigned long) cex->ec_len,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001467 cex->ec_start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001468 return cex->ec_type;
1469 }
1470
1471 /* not in cache */
1472 return EXT4_EXT_CACHE_NO;
1473}
1474
1475/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001476 * ext4_ext_rm_idx:
1477 * removes index from the index block.
1478 * It's used in truncate case only, thus all requests are for
1479 * last index in the block only.
Alex Tomasa86c6182006-10-11 01:21:03 -07001480 */
1481int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1482 struct ext4_ext_path *path)
1483{
1484 struct buffer_head *bh;
1485 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001486 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07001487
1488 /* free index block */
1489 path--;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001490 leaf = idx_pblock(path->p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -07001491 BUG_ON(path->p_hdr->eh_entries == 0);
1492 if ((err = ext4_ext_get_access(handle, inode, path)))
1493 return err;
1494 path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1495 if ((err = ext4_ext_dirty(handle, inode, path)))
1496 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07001497 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Alex Tomasa86c6182006-10-11 01:21:03 -07001498 bh = sb_find_get_block(inode->i_sb, leaf);
1499 ext4_forget(handle, 1, inode, bh, leaf);
1500 ext4_free_blocks(handle, inode, leaf, 1);
1501 return err;
1502}
1503
1504/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001505 * ext4_ext_calc_credits_for_insert:
1506 * This routine returns max. credits that the extent tree can consume.
Alex Tomasa86c6182006-10-11 01:21:03 -07001507 * It should be OK for low-performance paths like ->writepage()
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001508 * To allow many writing processes to fit into a single transaction,
1509 * the caller should calculate credits under truncate_mutex and
1510 * pass the actual path.
Alex Tomasa86c6182006-10-11 01:21:03 -07001511 */
1512int inline ext4_ext_calc_credits_for_insert(struct inode *inode,
1513 struct ext4_ext_path *path)
1514{
1515 int depth, needed;
1516
1517 if (path) {
1518 /* probably there is space in leaf? */
1519 depth = ext_depth(inode);
1520 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1521 < le16_to_cpu(path[depth].p_hdr->eh_max))
1522 return 1;
1523 }
1524
1525 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001526 * given 32-bit logical block (4294967296 blocks), max. tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001527 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001528 * Let's also add one more level for imbalance.
Alex Tomasa86c6182006-10-11 01:21:03 -07001529 */
1530 depth = 5;
1531
1532 /* allocation of new data block(s) */
1533 needed = 2;
1534
1535 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001536 * tree can be full, so it would need to grow in depth:
Alex Tomasa86c6182006-10-11 01:21:03 -07001537 * allocation + old root + new root
1538 */
1539 needed += 2 + 1 + 1;
1540
1541 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001542 * Index split can happen, we would need:
Alex Tomasa86c6182006-10-11 01:21:03 -07001543 * allocate intermediate indexes (bitmap + group)
1544 * + change two blocks at each level, but root (already included)
1545 */
1546 needed = (depth * 2) + (depth * 2);
1547
1548 /* any allocation modifies superblock */
1549 needed += 1;
1550
1551 return needed;
1552}
1553
1554static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1555 struct ext4_extent *ex,
1556 unsigned long from, unsigned long to)
1557{
1558 struct buffer_head *bh;
1559 int i;
1560
1561#ifdef EXTENTS_STATS
1562 {
1563 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1564 unsigned short ee_len = le16_to_cpu(ex->ee_len);
1565 spin_lock(&sbi->s_ext_stats_lock);
1566 sbi->s_ext_blocks += ee_len;
1567 sbi->s_ext_extents++;
1568 if (ee_len < sbi->s_ext_min)
1569 sbi->s_ext_min = ee_len;
1570 if (ee_len > sbi->s_ext_max)
1571 sbi->s_ext_max = ee_len;
1572 if (ext_depth(inode) > sbi->s_depth_max)
1573 sbi->s_depth_max = ext_depth(inode);
1574 spin_unlock(&sbi->s_ext_stats_lock);
1575 }
1576#endif
1577 if (from >= le32_to_cpu(ex->ee_block)
1578 && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1579 /* tail removal */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001580 unsigned long num;
1581 ext4_fsblk_t start;
Alex Tomasa86c6182006-10-11 01:21:03 -07001582 num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001583 start = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - num;
Mingming Cao2ae02102006-10-11 01:21:11 -07001584 ext_debug("free last %lu blocks starting %llu\n", num, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001585 for (i = 0; i < num; i++) {
1586 bh = sb_find_get_block(inode->i_sb, start + i);
1587 ext4_forget(handle, 0, inode, bh, start + i);
1588 }
1589 ext4_free_blocks(handle, inode, start, num);
1590 } else if (from == le32_to_cpu(ex->ee_block)
1591 && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1592 printk("strange request: removal %lu-%lu from %u:%u\n",
1593 from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1594 } else {
1595 printk("strange request: removal(2) %lu-%lu from %u:%u\n",
1596 from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1597 }
1598 return 0;
1599}
1600
1601static int
1602ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1603 struct ext4_ext_path *path, unsigned long start)
1604{
1605 int err = 0, correct_index = 0;
1606 int depth = ext_depth(inode), credits;
1607 struct ext4_extent_header *eh;
1608 unsigned a, b, block, num;
1609 unsigned long ex_ee_block;
1610 unsigned short ex_ee_len;
1611 struct ext4_extent *ex;
1612
1613 ext_debug("truncate since %lu in leaf\n", start);
1614 if (!path[depth].p_hdr)
1615 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1616 eh = path[depth].p_hdr;
1617 BUG_ON(eh == NULL);
1618 BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
1619 BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
1620
1621 /* find where to start removing */
1622 ex = EXT_LAST_EXTENT(eh);
1623
1624 ex_ee_block = le32_to_cpu(ex->ee_block);
1625 ex_ee_len = le16_to_cpu(ex->ee_len);
1626
1627 while (ex >= EXT_FIRST_EXTENT(eh) &&
1628 ex_ee_block + ex_ee_len > start) {
1629 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1630 path[depth].p_ext = ex;
1631
1632 a = ex_ee_block > start ? ex_ee_block : start;
1633 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1634 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1635
1636 ext_debug(" border %u:%u\n", a, b);
1637
1638 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1639 block = 0;
1640 num = 0;
1641 BUG();
1642 } else if (a != ex_ee_block) {
1643 /* remove tail of the extent */
1644 block = ex_ee_block;
1645 num = a - block;
1646 } else if (b != ex_ee_block + ex_ee_len - 1) {
1647 /* remove head of the extent */
1648 block = a;
1649 num = b - a;
1650 /* there is no "make a hole" API yet */
1651 BUG();
1652 } else {
1653 /* remove whole extent: excellent! */
1654 block = ex_ee_block;
1655 num = 0;
1656 BUG_ON(a != ex_ee_block);
1657 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1658 }
1659
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001660 /* at present, extent can't cross block group: */
Alex Tomasa86c6182006-10-11 01:21:03 -07001661 /* leaf + bitmap + group desc + sb + inode */
1662 credits = 5;
1663 if (ex == EXT_FIRST_EXTENT(eh)) {
1664 correct_index = 1;
1665 credits += (ext_depth(inode)) + 1;
1666 }
1667#ifdef CONFIG_QUOTA
1668 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1669#endif
1670
1671 handle = ext4_ext_journal_restart(handle, credits);
1672 if (IS_ERR(handle)) {
1673 err = PTR_ERR(handle);
1674 goto out;
1675 }
1676
1677 err = ext4_ext_get_access(handle, inode, path + depth);
1678 if (err)
1679 goto out;
1680
1681 err = ext4_remove_blocks(handle, inode, ex, a, b);
1682 if (err)
1683 goto out;
1684
1685 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001686 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001687 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001688 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1689 }
1690
1691 ex->ee_block = cpu_to_le32(block);
1692 ex->ee_len = cpu_to_le16(num);
1693
1694 err = ext4_ext_dirty(handle, inode, path + depth);
1695 if (err)
1696 goto out;
1697
Mingming Cao2ae02102006-10-11 01:21:11 -07001698 ext_debug("new extent: %u:%u:%llu\n", block, num,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001699 ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001700 ex--;
1701 ex_ee_block = le32_to_cpu(ex->ee_block);
1702 ex_ee_len = le16_to_cpu(ex->ee_len);
1703 }
1704
1705 if (correct_index && eh->eh_entries)
1706 err = ext4_ext_correct_indexes(handle, inode, path);
1707
1708 /* if this leaf is free, then we should
1709 * remove it from index block above */
1710 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1711 err = ext4_ext_rm_idx(handle, inode, path + depth);
1712
1713out:
1714 return err;
1715}
1716
1717/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001718 * ext4_ext_more_to_rm:
1719 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07001720 */
1721static int inline
1722ext4_ext_more_to_rm(struct ext4_ext_path *path)
1723{
1724 BUG_ON(path->p_idx == NULL);
1725
1726 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1727 return 0;
1728
1729 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001730 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07001731 * so we have to consider current index for truncation
1732 */
1733 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1734 return 0;
1735 return 1;
1736}
1737
1738int ext4_ext_remove_space(struct inode *inode, unsigned long start)
1739{
1740 struct super_block *sb = inode->i_sb;
1741 int depth = ext_depth(inode);
1742 struct ext4_ext_path *path;
1743 handle_t *handle;
1744 int i = 0, err = 0;
1745
1746 ext_debug("truncate since %lu\n", start);
1747
1748 /* probably first extent we're gonna free will be last in block */
1749 handle = ext4_journal_start(inode, depth + 1);
1750 if (IS_ERR(handle))
1751 return PTR_ERR(handle);
1752
1753 ext4_ext_invalidate_cache(inode);
1754
1755 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001756 * We start scanning from right side, freeing all the blocks
1757 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07001758 */
1759 path = kmalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1760 if (path == NULL) {
1761 ext4_journal_stop(handle);
1762 return -ENOMEM;
1763 }
1764 memset(path, 0, sizeof(struct ext4_ext_path) * (depth + 1));
1765 path[0].p_hdr = ext_inode_hdr(inode);
1766 if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) {
1767 err = -EIO;
1768 goto out;
1769 }
1770 path[0].p_depth = depth;
1771
1772 while (i >= 0 && err == 0) {
1773 if (i == depth) {
1774 /* this is leaf block */
1775 err = ext4_ext_rm_leaf(handle, inode, path, start);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001776 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07001777 brelse(path[i].p_bh);
1778 path[i].p_bh = NULL;
1779 i--;
1780 continue;
1781 }
1782
1783 /* this is index block */
1784 if (!path[i].p_hdr) {
1785 ext_debug("initialize header\n");
1786 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1787 if (ext4_ext_check_header(__FUNCTION__, inode,
1788 path[i].p_hdr)) {
1789 err = -EIO;
1790 goto out;
1791 }
1792 }
1793
1794 BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries)
1795 > le16_to_cpu(path[i].p_hdr->eh_max));
1796 BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC);
1797
1798 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001799 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07001800 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1801 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1802 ext_debug("init index ptr: hdr 0x%p, num %d\n",
1803 path[i].p_hdr,
1804 le16_to_cpu(path[i].p_hdr->eh_entries));
1805 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001806 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001807 path[i].p_idx--;
1808 }
1809
1810 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
1811 i, EXT_FIRST_INDEX(path[i].p_hdr),
1812 path[i].p_idx);
1813 if (ext4_ext_more_to_rm(path + i)) {
1814 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07001815 ext_debug("move to level %d (block %llu)\n",
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001816 i + 1, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07001817 memset(path + i + 1, 0, sizeof(*path));
1818 path[i+1].p_bh =
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001819 sb_bread(sb, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07001820 if (!path[i+1].p_bh) {
1821 /* should we reset i_size? */
1822 err = -EIO;
1823 break;
1824 }
1825
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001826 /* save actual number of indexes since this
1827 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07001828 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1829 i++;
1830 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001831 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07001832 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001833 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07001834 * handle must be already prepared by the
1835 * truncatei_leaf() */
1836 err = ext4_ext_rm_idx(handle, inode, path + i);
1837 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001838 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07001839 brelse(path[i].p_bh);
1840 path[i].p_bh = NULL;
1841 i--;
1842 ext_debug("return to level %d\n", i);
1843 }
1844 }
1845
1846 /* TODO: flexible tree reduction should be here */
1847 if (path->p_hdr->eh_entries == 0) {
1848 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001849 * truncate to zero freed all the tree,
1850 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07001851 */
1852 err = ext4_ext_get_access(handle, inode, path);
1853 if (err == 0) {
1854 ext_inode_hdr(inode)->eh_depth = 0;
1855 ext_inode_hdr(inode)->eh_max =
1856 cpu_to_le16(ext4_ext_space_root(inode));
1857 err = ext4_ext_dirty(handle, inode, path);
1858 }
1859 }
1860out:
1861 ext4_ext_tree_changed(inode);
1862 ext4_ext_drop_refs(path);
1863 kfree(path);
1864 ext4_journal_stop(handle);
1865
1866 return err;
1867}
1868
1869/*
1870 * called at mount time
1871 */
1872void ext4_ext_init(struct super_block *sb)
1873{
1874 /*
1875 * possible initialization would be here
1876 */
1877
1878 if (test_opt(sb, EXTENTS)) {
1879 printk("EXT4-fs: file extents enabled");
1880#ifdef AGRESSIVE_TEST
1881 printk(", agressive tests");
1882#endif
1883#ifdef CHECK_BINSEARCH
1884 printk(", check binsearch");
1885#endif
1886#ifdef EXTENTS_STATS
1887 printk(", stats");
1888#endif
1889 printk("\n");
1890#ifdef EXTENTS_STATS
1891 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
1892 EXT4_SB(sb)->s_ext_min = 1 << 30;
1893 EXT4_SB(sb)->s_ext_max = 0;
1894#endif
1895 }
1896}
1897
1898/*
1899 * called at umount time
1900 */
1901void ext4_ext_release(struct super_block *sb)
1902{
1903 if (!test_opt(sb, EXTENTS))
1904 return;
1905
1906#ifdef EXTENTS_STATS
1907 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
1908 struct ext4_sb_info *sbi = EXT4_SB(sb);
1909 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
1910 sbi->s_ext_blocks, sbi->s_ext_extents,
1911 sbi->s_ext_blocks / sbi->s_ext_extents);
1912 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
1913 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
1914 }
1915#endif
1916}
1917
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001918int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
1919 ext4_fsblk_t iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07001920 unsigned long max_blocks, struct buffer_head *bh_result,
1921 int create, int extend_disksize)
1922{
1923 struct ext4_ext_path *path = NULL;
1924 struct ext4_extent newex, *ex;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001925 ext4_fsblk_t goal, newblock;
1926 int err = 0, depth;
Alex Tomasa86c6182006-10-11 01:21:03 -07001927 unsigned long allocated = 0;
1928
1929 __clear_bit(BH_New, &bh_result->b_state);
1930 ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock,
1931 max_blocks, (unsigned) inode->i_ino);
1932 mutex_lock(&EXT4_I(inode)->truncate_mutex);
1933
1934 /* check in cache */
1935 if ((goal = ext4_ext_in_cache(inode, iblock, &newex))) {
1936 if (goal == EXT4_EXT_CACHE_GAP) {
1937 if (!create) {
1938 /* block isn't allocated yet and
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001939 * user doesn't want to allocate it */
Alex Tomasa86c6182006-10-11 01:21:03 -07001940 goto out2;
1941 }
1942 /* we should allocate requested block */
1943 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
1944 /* block is already allocated */
1945 newblock = iblock
1946 - le32_to_cpu(newex.ee_block)
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001947 + ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001948 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07001949 allocated = le16_to_cpu(newex.ee_len) -
1950 (iblock - le32_to_cpu(newex.ee_block));
1951 goto out;
1952 } else {
1953 BUG();
1954 }
1955 }
1956
1957 /* find extent for this block */
1958 path = ext4_ext_find_extent(inode, iblock, NULL);
1959 if (IS_ERR(path)) {
1960 err = PTR_ERR(path);
1961 path = NULL;
1962 goto out2;
1963 }
1964
1965 depth = ext_depth(inode);
1966
1967 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001968 * consistent leaf must not be empty;
1969 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07001970 * this is why assert can't be put in ext4_ext_find_extent()
1971 */
1972 BUG_ON(path[depth].p_ext == NULL && depth != 0);
1973
1974 if ((ex = path[depth].p_ext)) {
1975 unsigned long ee_block = le32_to_cpu(ex->ee_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001976 ext4_fsblk_t ee_start = ext_pblock(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001977 unsigned short ee_len = le16_to_cpu(ex->ee_len);
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001978
1979 /*
1980 * Allow future support for preallocated extents to be added
1981 * as an RO_COMPAT feature:
1982 * Uninitialized extents are treated as holes, except that
1983 * we avoid (fail) allocating new blocks during a write.
1984 */
1985 if (ee_len > EXT_MAX_LEN)
1986 goto out2;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001987 /* if found extent covers block, simply return it */
Alex Tomasa86c6182006-10-11 01:21:03 -07001988 if (iblock >= ee_block && iblock < ee_block + ee_len) {
1989 newblock = iblock - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001990 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07001991 allocated = ee_len - (iblock - ee_block);
Mingming Cao2ae02102006-10-11 01:21:11 -07001992 ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07001993 ee_block, ee_len, newblock);
1994 ext4_ext_put_in_cache(inode, ee_block, ee_len,
1995 ee_start, EXT4_EXT_CACHE_EXTENT);
1996 goto out;
1997 }
1998 }
1999
2000 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002001 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07002002 * we couldn't try to create block if create flag is zero
2003 */
2004 if (!create) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002005 /* put just found gap into cache to speed up
2006 * subsequent requests */
Alex Tomasa86c6182006-10-11 01:21:03 -07002007 ext4_ext_put_gap_in_cache(inode, path, iblock);
2008 goto out2;
2009 }
2010 /*
2011 * Okay, we need to do block allocation. Lazily initialize the block
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002012 * allocation info here if necessary.
Alex Tomasa86c6182006-10-11 01:21:03 -07002013 */
2014 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2015 ext4_init_block_alloc_info(inode);
2016
2017 /* allocate new block */
2018 goal = ext4_ext_find_goal(inode, path, iblock);
2019 allocated = max_blocks;
2020 newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2021 if (!newblock)
2022 goto out2;
Mingming Cao2ae02102006-10-11 01:21:11 -07002023 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002024 goal, newblock, allocated);
2025
2026 /* try to insert new extent into found leaf and return */
2027 newex.ee_block = cpu_to_le32(iblock);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002028 ext4_ext_store_pblock(&newex, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07002029 newex.ee_len = cpu_to_le16(allocated);
2030 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2031 if (err)
2032 goto out2;
2033
2034 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2035 EXT4_I(inode)->i_disksize = inode->i_size;
2036
2037 /* previous routine could use block we allocated */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002038 newblock = ext_pblock(&newex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002039 __set_bit(BH_New, &bh_result->b_state);
2040
2041 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2042 EXT4_EXT_CACHE_EXTENT);
2043out:
2044 if (allocated > max_blocks)
2045 allocated = max_blocks;
2046 ext4_ext_show_leaf(inode, path);
2047 __set_bit(BH_Mapped, &bh_result->b_state);
2048 bh_result->b_bdev = inode->i_sb->s_bdev;
2049 bh_result->b_blocknr = newblock;
2050out2:
2051 if (path) {
2052 ext4_ext_drop_refs(path);
2053 kfree(path);
2054 }
2055 mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2056
2057 return err ? err : allocated;
2058}
2059
2060void ext4_ext_truncate(struct inode * inode, struct page *page)
2061{
2062 struct address_space *mapping = inode->i_mapping;
2063 struct super_block *sb = inode->i_sb;
2064 unsigned long last_block;
2065 handle_t *handle;
2066 int err = 0;
2067
2068 /*
2069 * probably first extent we're gonna free will be last in block
2070 */
2071 err = ext4_writepage_trans_blocks(inode) + 3;
2072 handle = ext4_journal_start(inode, err);
2073 if (IS_ERR(handle)) {
2074 if (page) {
2075 clear_highpage(page);
2076 flush_dcache_page(page);
2077 unlock_page(page);
2078 page_cache_release(page);
2079 }
2080 return;
2081 }
2082
2083 if (page)
2084 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2085
2086 mutex_lock(&EXT4_I(inode)->truncate_mutex);
2087 ext4_ext_invalidate_cache(inode);
2088
2089 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002090 * TODO: optimization is possible here.
2091 * Probably we need not scan at all,
2092 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07002093 */
2094 if (ext4_orphan_add(handle, inode))
2095 goto out_stop;
2096
2097 /* we have to know where to truncate from in crash case */
2098 EXT4_I(inode)->i_disksize = inode->i_size;
2099 ext4_mark_inode_dirty(handle, inode);
2100
2101 last_block = (inode->i_size + sb->s_blocksize - 1)
2102 >> EXT4_BLOCK_SIZE_BITS(sb);
2103 err = ext4_ext_remove_space(inode, last_block);
2104
2105 /* In a multi-transaction truncate, we only make the final
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002106 * transaction synchronous. */
Alex Tomasa86c6182006-10-11 01:21:03 -07002107 if (IS_SYNC(inode))
2108 handle->h_sync = 1;
2109
2110out_stop:
2111 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002112 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07002113 * then we need to clear up the orphan record which we created above.
2114 * However, if this was a real unlink then we were called by
2115 * ext4_delete_inode(), and we allow that function to clean up the
2116 * orphan info for us.
2117 */
2118 if (inode->i_nlink)
2119 ext4_orphan_del(handle, inode);
2120
2121 mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2122 ext4_journal_stop(handle);
2123}
2124
2125/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002126 * ext4_ext_writepage_trans_blocks:
2127 * calculate max number of blocks we could modify
Alex Tomasa86c6182006-10-11 01:21:03 -07002128 * in order to allocate new block for an inode
2129 */
2130int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2131{
2132 int needed;
2133
2134 needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2135
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002136 /* caller wants to allocate num blocks, but note it includes sb */
Alex Tomasa86c6182006-10-11 01:21:03 -07002137 needed = needed * num - (num - 1);
2138
2139#ifdef CONFIG_QUOTA
2140 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2141#endif
2142
2143 return needed;
2144}
2145
2146EXPORT_SYMBOL(ext4_mark_inode_dirty);
2147EXPORT_SYMBOL(ext4_ext_invalidate_cache);
2148EXPORT_SYMBOL(ext4_ext_insert_extent);
2149EXPORT_SYMBOL(ext4_ext_walk_space);
2150EXPORT_SYMBOL(ext4_ext_find_goal);
2151EXPORT_SYMBOL(ext4_ext_calc_credits_for_insert);
2152