blob: 68537229ee1ca466a28690f4f10253fdd27babbe [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>
Mingming Caocd02ff02007-10-16 18:38:25 -040036#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070037#include <linux/highuid.h>
38#include <linux/pagemap.h>
39#include <linux/quotaops.h>
40#include <linux/string.h>
41#include <linux/slab.h>
Amit Aroraa2df2a62007-07-17 21:42:41 -040042#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070043#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 */
Avantika Mathur09b88252006-12-06 20:41:36 -080051static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070052{
53 ext4_fsblk_t block;
54
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040055 block = le32_to_cpu(ex->ee_start_lo);
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 */
Avantika Mathur09b88252006-12-06 20:41:36 -080064static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070065{
66 ext4_fsblk_t block;
67
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040068 block = le32_to_cpu(ix->ei_leaf_lo);
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 */
Avantika Mathur09b88252006-12-06 20:41:36 -080078static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070079{
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040080 ex->ee_start_lo = 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 */
Avantika Mathur09b88252006-12-06 20:41:36 -080089static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070090{
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040091 ix->ei_leaf_lo = 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 handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
96{
97 int err;
98
99 if (handle->h_buffer_credits > needed)
100 return handle;
101 if (!ext4_journal_extend(handle, needed))
102 return handle;
103 err = ext4_journal_restart(handle, needed);
104
105 return handle;
106}
107
108/*
109 * could return:
110 * - EROFS
111 * - ENOMEM
112 */
113static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
114 struct ext4_ext_path *path)
115{
116 if (path->p_bh) {
117 /* path points to block */
118 return ext4_journal_get_write_access(handle, path->p_bh);
119 }
120 /* path points to leaf/index in inode body */
121 /* we use in-core data, no need to protect them */
122 return 0;
123}
124
125/*
126 * could return:
127 * - EROFS
128 * - ENOMEM
129 * - EIO
130 */
131static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
132 struct ext4_ext_path *path)
133{
134 int err;
135 if (path->p_bh) {
136 /* path points to block */
137 err = ext4_journal_dirty_metadata(handle, path->p_bh);
138 } else {
139 /* path points to leaf/index in inode body */
140 err = ext4_mark_inode_dirty(handle, inode);
141 }
142 return err;
143}
144
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700145static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700146 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500147 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700148{
149 struct ext4_inode_info *ei = EXT4_I(inode);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700150 ext4_fsblk_t bg_start;
151 ext4_grpblk_t colour;
Alex Tomasa86c6182006-10-11 01:21:03 -0700152 int depth;
153
154 if (path) {
155 struct ext4_extent *ex;
156 depth = path->p_depth;
157
158 /* try to predict block placement */
Avantika Mathur7e028972006-12-06 20:41:33 -0800159 ex = path[depth].p_ext;
160 if (ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700161 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700162
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700163 /* it looks like index is empty;
164 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700165 if (path[depth].p_bh)
166 return path[depth].p_bh->b_blocknr;
167 }
168
169 /* OK. use inode's group */
170 bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
171 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
172 colour = (current->pid % 16) *
173 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
174 return bg_start + colour + block;
175}
176
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700177static ext4_fsblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -0700178ext4_ext_new_block(handle_t *handle, struct inode *inode,
179 struct ext4_ext_path *path,
180 struct ext4_extent *ex, int *err)
181{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700182 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700183
184 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
185 newblock = ext4_new_block(handle, inode, goal, err);
186 return newblock;
187}
188
Avantika Mathur09b88252006-12-06 20:41:36 -0800189static int ext4_ext_space_block(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700190{
191 int size;
192
193 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
194 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100195#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700196 if (size > 6)
197 size = 6;
198#endif
199 return size;
200}
201
Avantika Mathur09b88252006-12-06 20:41:36 -0800202static int ext4_ext_space_block_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700203{
204 int size;
205
206 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
207 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100208#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700209 if (size > 5)
210 size = 5;
211#endif
212 return size;
213}
214
Avantika Mathur09b88252006-12-06 20:41:36 -0800215static int ext4_ext_space_root(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700216{
217 int size;
218
219 size = sizeof(EXT4_I(inode)->i_data);
220 size -= sizeof(struct ext4_extent_header);
221 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100222#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700223 if (size > 3)
224 size = 3;
225#endif
226 return size;
227}
228
Avantika Mathur09b88252006-12-06 20:41:36 -0800229static int ext4_ext_space_root_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700230{
231 int size;
232
233 size = sizeof(EXT4_I(inode)->i_data);
234 size -= sizeof(struct ext4_extent_header);
235 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100236#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700237 if (size > 4)
238 size = 4;
239#endif
240 return size;
241}
242
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400243static int
244ext4_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
263static 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
293corrupted:
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
Alex Tomasa86c6182006-10-11 01:21:03 -0700307#ifdef EXT_DEBUG
308static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
309{
310 int k, l = path->p_depth;
311
312 ext_debug("path:");
313 for (k = 0; k <= l; k++, path++) {
314 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700315 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700316 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700317 } else if (path->p_ext) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700318 ext_debug(" %d:%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700319 le32_to_cpu(path->p_ext->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400320 ext4_ext_get_actual_len(path->p_ext),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700321 ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700322 } else
323 ext_debug(" []");
324 }
325 ext_debug("\n");
326}
327
328static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
329{
330 int depth = ext_depth(inode);
331 struct ext4_extent_header *eh;
332 struct ext4_extent *ex;
333 int i;
334
335 if (!path)
336 return;
337
338 eh = path[depth].p_hdr;
339 ex = EXT_FIRST_EXTENT(eh);
340
341 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700342 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400343 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700344 }
345 ext_debug("\n");
346}
347#else
348#define ext4_ext_show_path(inode,path)
349#define ext4_ext_show_leaf(inode,path)
350#endif
351
352static void ext4_ext_drop_refs(struct ext4_ext_path *path)
353{
354 int depth = path->p_depth;
355 int i;
356
357 for (i = 0; i <= depth; i++, path++)
358 if (path->p_bh) {
359 brelse(path->p_bh);
360 path->p_bh = NULL;
361 }
362}
363
364/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700365 * ext4_ext_binsearch_idx:
366 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400367 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700368 */
369static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500370ext4_ext_binsearch_idx(struct inode *inode,
371 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700372{
373 struct ext4_extent_header *eh = path->p_hdr;
374 struct ext4_extent_idx *r, *l, *m;
375
Alex Tomasa86c6182006-10-11 01:21:03 -0700376
Eric Sandeenbba90742008-01-28 23:58:27 -0500377 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700378
379 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400380 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700381 while (l <= r) {
382 m = l + (r - l) / 2;
383 if (block < le32_to_cpu(m->ei_block))
384 r = m - 1;
385 else
386 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400387 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
388 m, le32_to_cpu(m->ei_block),
389 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700390 }
391
392 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700393 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400394 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700395
396#ifdef CHECK_BINSEARCH
397 {
398 struct ext4_extent_idx *chix, *ix;
399 int k;
400
401 chix = ix = EXT_FIRST_INDEX(eh);
402 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
403 if (k != 0 &&
404 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
405 printk("k=%d, ix=0x%p, first=0x%p\n", k,
406 ix, EXT_FIRST_INDEX(eh));
407 printk("%u <= %u\n",
408 le32_to_cpu(ix->ei_block),
409 le32_to_cpu(ix[-1].ei_block));
410 }
411 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400412 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700413 if (block < le32_to_cpu(ix->ei_block))
414 break;
415 chix = ix;
416 }
417 BUG_ON(chix != path->p_idx);
418 }
419#endif
420
421}
422
423/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700424 * ext4_ext_binsearch:
425 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400426 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700427 */
428static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500429ext4_ext_binsearch(struct inode *inode,
430 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700431{
432 struct ext4_extent_header *eh = path->p_hdr;
433 struct ext4_extent *r, *l, *m;
434
Alex Tomasa86c6182006-10-11 01:21:03 -0700435 if (eh->eh_entries == 0) {
436 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700437 * this leaf is empty:
438 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700439 */
440 return;
441 }
442
Eric Sandeenbba90742008-01-28 23:58:27 -0500443 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700444
445 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400446 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700447
448 while (l <= r) {
449 m = l + (r - l) / 2;
450 if (block < le32_to_cpu(m->ee_block))
451 r = m - 1;
452 else
453 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400454 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
455 m, le32_to_cpu(m->ee_block),
456 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700457 }
458
459 path->p_ext = l - 1;
Mingming Cao2ae02102006-10-11 01:21:11 -0700460 ext_debug(" -> %d:%llu:%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400461 le32_to_cpu(path->p_ext->ee_block),
462 ext_pblock(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400463 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700464
465#ifdef CHECK_BINSEARCH
466 {
467 struct ext4_extent *chex, *ex;
468 int k;
469
470 chex = ex = EXT_FIRST_EXTENT(eh);
471 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
472 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400473 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700474 if (block < le32_to_cpu(ex->ee_block))
475 break;
476 chex = ex;
477 }
478 BUG_ON(chex != path->p_ext);
479 }
480#endif
481
482}
483
484int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
485{
486 struct ext4_extent_header *eh;
487
488 eh = ext_inode_hdr(inode);
489 eh->eh_depth = 0;
490 eh->eh_entries = 0;
491 eh->eh_magic = EXT4_EXT_MAGIC;
492 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
493 ext4_mark_inode_dirty(handle, inode);
494 ext4_ext_invalidate_cache(inode);
495 return 0;
496}
497
498struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500499ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
500 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700501{
502 struct ext4_extent_header *eh;
503 struct buffer_head *bh;
504 short int depth, i, ppos = 0, alloc = 0;
505
506 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400507 depth = ext_depth(inode);
508 if (ext4_ext_check_header(inode, eh, depth))
Alex Tomasa86c6182006-10-11 01:21:03 -0700509 return ERR_PTR(-EIO);
510
Alex Tomasa86c6182006-10-11 01:21:03 -0700511
512 /* account possible depth increase */
513 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800514 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700515 GFP_NOFS);
516 if (!path)
517 return ERR_PTR(-ENOMEM);
518 alloc = 1;
519 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700520 path[0].p_hdr = eh;
521
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400522 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700523 /* walk through the tree */
524 while (i) {
525 ext_debug("depth %d: num %d, max %d\n",
526 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400527
Alex Tomasa86c6182006-10-11 01:21:03 -0700528 ext4_ext_binsearch_idx(inode, path + ppos, block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700529 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700530 path[ppos].p_depth = i;
531 path[ppos].p_ext = NULL;
532
533 bh = sb_bread(inode->i_sb, path[ppos].p_block);
534 if (!bh)
535 goto err;
536
537 eh = ext_block_hdr(bh);
538 ppos++;
539 BUG_ON(ppos > depth);
540 path[ppos].p_bh = bh;
541 path[ppos].p_hdr = eh;
542 i--;
543
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400544 if (ext4_ext_check_header(inode, eh, i))
Alex Tomasa86c6182006-10-11 01:21:03 -0700545 goto err;
546 }
547
548 path[ppos].p_depth = i;
549 path[ppos].p_hdr = eh;
550 path[ppos].p_ext = NULL;
551 path[ppos].p_idx = NULL;
552
Alex Tomasa86c6182006-10-11 01:21:03 -0700553 /* find extent */
554 ext4_ext_binsearch(inode, path + ppos, block);
555
556 ext4_ext_show_path(inode, path);
557
558 return path;
559
560err:
561 ext4_ext_drop_refs(path);
562 if (alloc)
563 kfree(path);
564 return ERR_PTR(-EIO);
565}
566
567/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700568 * ext4_ext_insert_index:
569 * insert new index [@logical;@ptr] into the block at @curp;
570 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700571 */
572static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
573 struct ext4_ext_path *curp,
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700574 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700575{
576 struct ext4_extent_idx *ix;
577 int len, err;
578
Avantika Mathur7e028972006-12-06 20:41:33 -0800579 err = ext4_ext_get_access(handle, inode, curp);
580 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700581 return err;
582
583 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
584 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
585 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
586 /* insert after */
587 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
588 len = (len - 1) * sizeof(struct ext4_extent_idx);
589 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400590 ext_debug("insert new index %d after: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700591 "move %d from 0x%p to 0x%p\n",
592 logical, ptr, len,
593 (curp->p_idx + 1), (curp->p_idx + 2));
594 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
595 }
596 ix = curp->p_idx + 1;
597 } else {
598 /* insert before */
599 len = len * sizeof(struct ext4_extent_idx);
600 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400601 ext_debug("insert new index %d before: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700602 "move %d from 0x%p to 0x%p\n",
603 logical, ptr, len,
604 curp->p_idx, (curp->p_idx + 1));
605 memmove(curp->p_idx + 1, curp->p_idx, len);
606 ix = curp->p_idx;
607 }
608
609 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700610 ext4_idx_store_pblock(ix, ptr);
Alex Tomasa86c6182006-10-11 01:21:03 -0700611 curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
612
613 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400614 > le16_to_cpu(curp->p_hdr->eh_max));
Alex Tomasa86c6182006-10-11 01:21:03 -0700615 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
616
617 err = ext4_ext_dirty(handle, inode, curp);
618 ext4_std_error(inode->i_sb, err);
619
620 return err;
621}
622
623/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700624 * ext4_ext_split:
625 * inserts new subtree into the path, using free index entry
626 * at depth @at:
627 * - allocates all needed blocks (new leaf and all intermediate index blocks)
628 * - makes decision where to split
629 * - moves remaining extents and index entries (right to the split point)
630 * into the newly allocated blocks
631 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700632 */
633static int ext4_ext_split(handle_t *handle, struct inode *inode,
634 struct ext4_ext_path *path,
635 struct ext4_extent *newext, int at)
636{
637 struct buffer_head *bh = NULL;
638 int depth = ext_depth(inode);
639 struct ext4_extent_header *neh;
640 struct ext4_extent_idx *fidx;
641 struct ext4_extent *ex;
642 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700643 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700644 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700645 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700646 int err = 0;
647
648 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700649 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700650
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700651 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700652 * border from split point */
653 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
654 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
655 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700656 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700657 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400658 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700659 } else {
660 border = newext->ee_block;
661 ext_debug("leaf will be added."
662 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400663 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700664 }
665
666 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700667 * If error occurs, then we break processing
668 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700669 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700670 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700671 */
672
673 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700674 * Get array to track all allocated blocks.
675 * We need this to handle errors and free blocks
676 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700677 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800678 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700679 if (!ablocks)
680 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700681
682 /* allocate all needed blocks */
683 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
684 for (a = 0; a < depth - at; a++) {
685 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
686 if (newblock == 0)
687 goto cleanup;
688 ablocks[a] = newblock;
689 }
690
691 /* initialize new leaf */
692 newblock = ablocks[--a];
693 BUG_ON(newblock == 0);
694 bh = sb_getblk(inode->i_sb, newblock);
695 if (!bh) {
696 err = -EIO;
697 goto cleanup;
698 }
699 lock_buffer(bh);
700
Avantika Mathur7e028972006-12-06 20:41:33 -0800701 err = ext4_journal_get_create_access(handle, bh);
702 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700703 goto cleanup;
704
705 neh = ext_block_hdr(bh);
706 neh->eh_entries = 0;
707 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
708 neh->eh_magic = EXT4_EXT_MAGIC;
709 neh->eh_depth = 0;
710 ex = EXT_FIRST_EXTENT(neh);
711
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700712 /* move remainder of path[depth] to the new leaf */
Alex Tomasa86c6182006-10-11 01:21:03 -0700713 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
714 /* start copy from next extent */
715 /* TODO: we could do it by single memmove */
716 m = 0;
717 path[depth].p_ext++;
718 while (path[depth].p_ext <=
719 EXT_MAX_EXTENT(path[depth].p_hdr)) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700720 ext_debug("move %d:%llu:%d in new leaf %llu\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400721 le32_to_cpu(path[depth].p_ext->ee_block),
722 ext_pblock(path[depth].p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400723 ext4_ext_get_actual_len(path[depth].p_ext),
Alex Tomasa86c6182006-10-11 01:21:03 -0700724 newblock);
725 /*memmove(ex++, path[depth].p_ext++,
726 sizeof(struct ext4_extent));
727 neh->eh_entries++;*/
728 path[depth].p_ext++;
729 m++;
730 }
731 if (m) {
732 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
733 neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
734 }
735
736 set_buffer_uptodate(bh);
737 unlock_buffer(bh);
738
Avantika Mathur7e028972006-12-06 20:41:33 -0800739 err = ext4_journal_dirty_metadata(handle, bh);
740 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700741 goto cleanup;
742 brelse(bh);
743 bh = NULL;
744
745 /* correct old leaf */
746 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -0800747 err = ext4_ext_get_access(handle, inode, path + depth);
748 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700749 goto cleanup;
750 path[depth].p_hdr->eh_entries =
751 cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
Avantika Mathur7e028972006-12-06 20:41:33 -0800752 err = ext4_ext_dirty(handle, inode, path + depth);
753 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700754 goto cleanup;
755
756 }
757
758 /* create intermediate indexes */
759 k = depth - at - 1;
760 BUG_ON(k < 0);
761 if (k)
762 ext_debug("create %d intermediate indices\n", k);
763 /* insert new index into current index block */
764 /* current depth stored in i var */
765 i = depth - 1;
766 while (k--) {
767 oldblock = newblock;
768 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -0500769 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700770 if (!bh) {
771 err = -EIO;
772 goto cleanup;
773 }
774 lock_buffer(bh);
775
Avantika Mathur7e028972006-12-06 20:41:33 -0800776 err = ext4_journal_get_create_access(handle, bh);
777 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700778 goto cleanup;
779
780 neh = ext_block_hdr(bh);
781 neh->eh_entries = cpu_to_le16(1);
782 neh->eh_magic = EXT4_EXT_MAGIC;
783 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
784 neh->eh_depth = cpu_to_le16(depth - i);
785 fidx = EXT_FIRST_INDEX(neh);
786 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700787 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700788
Eric Sandeenbba90742008-01-28 23:58:27 -0500789 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
790 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700791 /* copy indexes */
792 m = 0;
793 path[i].p_idx++;
794
795 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
796 EXT_MAX_INDEX(path[i].p_hdr));
797 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
798 EXT_LAST_INDEX(path[i].p_hdr));
799 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400800 ext_debug("%d: move %d:%llu in new index %llu\n", i,
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400801 le32_to_cpu(path[i].p_idx->ei_block),
802 idx_pblock(path[i].p_idx),
803 newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700804 /*memmove(++fidx, path[i].p_idx++,
805 sizeof(struct ext4_extent_idx));
806 neh->eh_entries++;
807 BUG_ON(neh->eh_entries > neh->eh_max);*/
808 path[i].p_idx++;
809 m++;
810 }
811 if (m) {
812 memmove(++fidx, path[i].p_idx - m,
813 sizeof(struct ext4_extent_idx) * m);
814 neh->eh_entries =
815 cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
816 }
817 set_buffer_uptodate(bh);
818 unlock_buffer(bh);
819
Avantika Mathur7e028972006-12-06 20:41:33 -0800820 err = ext4_journal_dirty_metadata(handle, bh);
821 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700822 goto cleanup;
823 brelse(bh);
824 bh = NULL;
825
826 /* correct old index */
827 if (m) {
828 err = ext4_ext_get_access(handle, inode, path + i);
829 if (err)
830 goto cleanup;
831 path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
832 err = ext4_ext_dirty(handle, inode, path + i);
833 if (err)
834 goto cleanup;
835 }
836
837 i--;
838 }
839
840 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -0700841 err = ext4_ext_insert_index(handle, inode, path + at,
842 le32_to_cpu(border), newblock);
843
844cleanup:
845 if (bh) {
846 if (buffer_locked(bh))
847 unlock_buffer(bh);
848 brelse(bh);
849 }
850
851 if (err) {
852 /* free all allocated blocks in error case */
853 for (i = 0; i < depth; i++) {
854 if (!ablocks[i])
855 continue;
856 ext4_free_blocks(handle, inode, ablocks[i], 1);
857 }
858 }
859 kfree(ablocks);
860
861 return err;
862}
863
864/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700865 * ext4_ext_grow_indepth:
866 * implements tree growing procedure:
867 * - allocates new block
868 * - moves top-level data (index block or leaf) into the new block
869 * - initializes new top-level, creating index that points to the
870 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -0700871 */
872static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
873 struct ext4_ext_path *path,
874 struct ext4_extent *newext)
875{
876 struct ext4_ext_path *curp = path;
877 struct ext4_extent_header *neh;
878 struct ext4_extent_idx *fidx;
879 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700880 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700881 int err = 0;
882
883 newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
884 if (newblock == 0)
885 return err;
886
887 bh = sb_getblk(inode->i_sb, newblock);
888 if (!bh) {
889 err = -EIO;
890 ext4_std_error(inode->i_sb, err);
891 return err;
892 }
893 lock_buffer(bh);
894
Avantika Mathur7e028972006-12-06 20:41:33 -0800895 err = ext4_journal_get_create_access(handle, bh);
896 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -0700897 unlock_buffer(bh);
898 goto out;
899 }
900
901 /* move top-level index/leaf into new block */
902 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
903
904 /* set size of new block */
905 neh = ext_block_hdr(bh);
906 /* old root could have indexes or leaves
907 * so calculate e_max right way */
908 if (ext_depth(inode))
909 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
910 else
911 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
912 neh->eh_magic = EXT4_EXT_MAGIC;
913 set_buffer_uptodate(bh);
914 unlock_buffer(bh);
915
Avantika Mathur7e028972006-12-06 20:41:33 -0800916 err = ext4_journal_dirty_metadata(handle, bh);
917 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700918 goto out;
919
920 /* create index in new top-level index: num,max,pointer */
Avantika Mathur7e028972006-12-06 20:41:33 -0800921 err = ext4_ext_get_access(handle, inode, curp);
922 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700923 goto out;
924
925 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
926 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
927 curp->p_hdr->eh_entries = cpu_to_le16(1);
928 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400929
930 if (path[0].p_hdr->eh_depth)
931 curp->p_idx->ei_block =
932 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
933 else
934 curp->p_idx->ei_block =
935 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700936 ext4_idx_store_pblock(curp->p_idx, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700937
938 neh = ext_inode_hdr(inode);
939 fidx = EXT_FIRST_INDEX(neh);
Mingming Cao2ae02102006-10-11 01:21:11 -0700940 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700941 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700942 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700943
944 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
945 err = ext4_ext_dirty(handle, inode, curp);
946out:
947 brelse(bh);
948
949 return err;
950}
951
952/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700953 * ext4_ext_create_new_leaf:
954 * finds empty index and adds new leaf.
955 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -0700956 */
957static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
958 struct ext4_ext_path *path,
959 struct ext4_extent *newext)
960{
961 struct ext4_ext_path *curp;
962 int depth, i, err = 0;
963
964repeat:
965 i = depth = ext_depth(inode);
966
967 /* walk up to the tree and look for free index entry */
968 curp = path + depth;
969 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
970 i--;
971 curp--;
972 }
973
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700974 /* we use already allocated block for index block,
975 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -0700976 if (EXT_HAS_FREE_INDEX(curp)) {
977 /* if we found index with free entry, then use that
978 * entry: create all needed subtree and add new leaf */
979 err = ext4_ext_split(handle, inode, path, newext, i);
980
981 /* refill path */
982 ext4_ext_drop_refs(path);
983 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500984 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
985 path);
Alex Tomasa86c6182006-10-11 01:21:03 -0700986 if (IS_ERR(path))
987 err = PTR_ERR(path);
988 } else {
989 /* tree is full, time to grow in depth */
990 err = ext4_ext_grow_indepth(handle, inode, path, newext);
991 if (err)
992 goto out;
993
994 /* refill path */
995 ext4_ext_drop_refs(path);
996 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500997 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
998 path);
Alex Tomasa86c6182006-10-11 01:21:03 -0700999 if (IS_ERR(path)) {
1000 err = PTR_ERR(path);
1001 goto out;
1002 }
1003
1004 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001005 * only first (depth 0 -> 1) produces free space;
1006 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001007 */
1008 depth = ext_depth(inode);
1009 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001010 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001011 goto repeat;
1012 }
1013 }
1014
1015out:
1016 return err;
1017}
1018
1019/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001020 * ext4_ext_next_allocated_block:
1021 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1022 * NOTE: it considers block number from index entry as
1023 * allocated block. Thus, index entries have to be consistent
1024 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001025 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001026static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001027ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1028{
1029 int depth;
1030
1031 BUG_ON(path == NULL);
1032 depth = path->p_depth;
1033
1034 if (depth == 0 && path->p_ext == NULL)
1035 return EXT_MAX_BLOCK;
1036
1037 while (depth >= 0) {
1038 if (depth == path->p_depth) {
1039 /* leaf */
1040 if (path[depth].p_ext !=
1041 EXT_LAST_EXTENT(path[depth].p_hdr))
1042 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1043 } else {
1044 /* index */
1045 if (path[depth].p_idx !=
1046 EXT_LAST_INDEX(path[depth].p_hdr))
1047 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1048 }
1049 depth--;
1050 }
1051
1052 return EXT_MAX_BLOCK;
1053}
1054
1055/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001056 * ext4_ext_next_leaf_block:
Alex Tomasa86c6182006-10-11 01:21:03 -07001057 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1058 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001059static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
Andrew Morton63f57932006-10-11 01:21:24 -07001060 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001061{
1062 int depth;
1063
1064 BUG_ON(path == NULL);
1065 depth = path->p_depth;
1066
1067 /* zero-tree has no leaf blocks at all */
1068 if (depth == 0)
1069 return EXT_MAX_BLOCK;
1070
1071 /* go to index block */
1072 depth--;
1073
1074 while (depth >= 0) {
1075 if (path[depth].p_idx !=
1076 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001077 return (ext4_lblk_t)
1078 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001079 depth--;
1080 }
1081
1082 return EXT_MAX_BLOCK;
1083}
1084
1085/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001086 * ext4_ext_correct_indexes:
1087 * if leaf gets modified and modified extent is first in the leaf,
1088 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001089 * TODO: do we need to correct tree in all cases?
1090 */
1091int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1092 struct ext4_ext_path *path)
1093{
1094 struct ext4_extent_header *eh;
1095 int depth = ext_depth(inode);
1096 struct ext4_extent *ex;
1097 __le32 border;
1098 int k, err = 0;
1099
1100 eh = path[depth].p_hdr;
1101 ex = path[depth].p_ext;
1102 BUG_ON(ex == NULL);
1103 BUG_ON(eh == NULL);
1104
1105 if (depth == 0) {
1106 /* there is no tree at all */
1107 return 0;
1108 }
1109
1110 if (ex != EXT_FIRST_EXTENT(eh)) {
1111 /* we correct tree if first leaf got modified only */
1112 return 0;
1113 }
1114
1115 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001116 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001117 */
1118 k = depth - 1;
1119 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001120 err = ext4_ext_get_access(handle, inode, path + k);
1121 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001122 return err;
1123 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001124 err = ext4_ext_dirty(handle, inode, path + k);
1125 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001126 return err;
1127
1128 while (k--) {
1129 /* change all left-side indexes */
1130 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1131 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001132 err = ext4_ext_get_access(handle, inode, path + k);
1133 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001134 break;
1135 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001136 err = ext4_ext_dirty(handle, inode, path + k);
1137 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001138 break;
1139 }
1140
1141 return err;
1142}
1143
Avantika Mathur09b88252006-12-06 20:41:36 -08001144static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001145ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1146 struct ext4_extent *ex2)
1147{
Amit Arora749269f2007-07-18 09:02:56 -04001148 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001149
1150 /*
1151 * Make sure that either both extents are uninitialized, or
1152 * both are _not_.
1153 */
1154 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1155 return 0;
1156
Amit Arora749269f2007-07-18 09:02:56 -04001157 if (ext4_ext_is_uninitialized(ex1))
1158 max_len = EXT_UNINIT_MAX_LEN;
1159 else
1160 max_len = EXT_INIT_MAX_LEN;
1161
Amit Aroraa2df2a62007-07-17 21:42:41 -04001162 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1163 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1164
1165 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001166 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001167 return 0;
1168
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001169 /*
1170 * To allow future support for preallocated extents to be added
1171 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001172 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001173 */
Amit Arora749269f2007-07-18 09:02:56 -04001174 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001175 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001176#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -07001177 if (le16_to_cpu(ex1->ee_len) >= 4)
1178 return 0;
1179#endif
1180
Amit Aroraa2df2a62007-07-17 21:42:41 -04001181 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001182 return 1;
1183 return 0;
1184}
1185
1186/*
Amit Arora56055d32007-07-17 21:42:38 -04001187 * This function tries to merge the "ex" extent to the next extent in the tree.
1188 * It always tries to merge towards right. If you want to merge towards
1189 * left, pass "ex - 1" as argument instead of "ex".
1190 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1191 * 1 if they got merged.
1192 */
1193int ext4_ext_try_to_merge(struct inode *inode,
1194 struct ext4_ext_path *path,
1195 struct ext4_extent *ex)
1196{
1197 struct ext4_extent_header *eh;
1198 unsigned int depth, len;
1199 int merge_done = 0;
1200 int uninitialized = 0;
1201
1202 depth = ext_depth(inode);
1203 BUG_ON(path[depth].p_hdr == NULL);
1204 eh = path[depth].p_hdr;
1205
1206 while (ex < EXT_LAST_EXTENT(eh)) {
1207 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1208 break;
1209 /* merge with next extent! */
1210 if (ext4_ext_is_uninitialized(ex))
1211 uninitialized = 1;
1212 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1213 + ext4_ext_get_actual_len(ex + 1));
1214 if (uninitialized)
1215 ext4_ext_mark_uninitialized(ex);
1216
1217 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1218 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1219 * sizeof(struct ext4_extent);
1220 memmove(ex + 1, ex + 2, len);
1221 }
1222 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
1223 merge_done = 1;
1224 WARN_ON(eh->eh_entries == 0);
1225 if (!eh->eh_entries)
1226 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1227 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1228 }
1229
1230 return merge_done;
1231}
1232
1233/*
Amit Arora25d14f92007-05-24 13:04:13 -04001234 * check if a portion of the "newext" extent overlaps with an
1235 * existing extent.
1236 *
1237 * If there is an overlap discovered, it updates the length of the newext
1238 * such that there will be no overlap, and then returns 1.
1239 * If there is no overlap found, it returns 0.
1240 */
1241unsigned int ext4_ext_check_overlap(struct inode *inode,
1242 struct ext4_extent *newext,
1243 struct ext4_ext_path *path)
1244{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001245 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001246 unsigned int depth, len1;
1247 unsigned int ret = 0;
1248
1249 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001250 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001251 depth = ext_depth(inode);
1252 if (!path[depth].p_ext)
1253 goto out;
1254 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1255
1256 /*
1257 * get the next allocated block if the extent in the path
1258 * is before the requested block(s)
1259 */
1260 if (b2 < b1) {
1261 b2 = ext4_ext_next_allocated_block(path);
1262 if (b2 == EXT_MAX_BLOCK)
1263 goto out;
1264 }
1265
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001266 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001267 if (b1 + len1 < b1) {
1268 len1 = EXT_MAX_BLOCK - b1;
1269 newext->ee_len = cpu_to_le16(len1);
1270 ret = 1;
1271 }
1272
1273 /* check for overlap */
1274 if (b1 + len1 > b2) {
1275 newext->ee_len = cpu_to_le16(b2 - b1);
1276 ret = 1;
1277 }
1278out:
1279 return ret;
1280}
1281
1282/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001283 * ext4_ext_insert_extent:
1284 * tries to merge requsted extent into the existing extent or
1285 * inserts requested extent as new one into the tree,
1286 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001287 */
1288int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1289 struct ext4_ext_path *path,
1290 struct ext4_extent *newext)
1291{
1292 struct ext4_extent_header * eh;
1293 struct ext4_extent *ex, *fex;
1294 struct ext4_extent *nearex; /* nearest extent */
1295 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001296 int depth, len, err;
1297 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001298 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001299
Amit Aroraa2df2a62007-07-17 21:42:41 -04001300 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001301 depth = ext_depth(inode);
1302 ex = path[depth].p_ext;
1303 BUG_ON(path[depth].p_hdr == NULL);
1304
1305 /* try to insert block into found extent and return */
1306 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming Cao2ae02102006-10-11 01:21:11 -07001307 ext_debug("append %d block to %d:%d (from %llu)\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001308 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001309 le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001310 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001311 err = ext4_ext_get_access(handle, inode, path + depth);
1312 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001313 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001314
1315 /*
1316 * ext4_can_extents_be_merged should have checked that either
1317 * both extents are uninitialized, or both aren't. Thus we
1318 * need to check only one of them here.
1319 */
1320 if (ext4_ext_is_uninitialized(ex))
1321 uninitialized = 1;
1322 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1323 + ext4_ext_get_actual_len(newext));
1324 if (uninitialized)
1325 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001326 eh = path[depth].p_hdr;
1327 nearex = ex;
1328 goto merge;
1329 }
1330
1331repeat:
1332 depth = ext_depth(inode);
1333 eh = path[depth].p_hdr;
1334 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1335 goto has_space;
1336
1337 /* probably next leaf has space for us? */
1338 fex = EXT_LAST_EXTENT(eh);
1339 next = ext4_ext_next_leaf_block(inode, path);
1340 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1341 && next != EXT_MAX_BLOCK) {
1342 ext_debug("next leaf block - %d\n", next);
1343 BUG_ON(npath != NULL);
1344 npath = ext4_ext_find_extent(inode, next, NULL);
1345 if (IS_ERR(npath))
1346 return PTR_ERR(npath);
1347 BUG_ON(npath->p_depth != path->p_depth);
1348 eh = npath[depth].p_hdr;
1349 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1350 ext_debug("next leaf isnt full(%d)\n",
1351 le16_to_cpu(eh->eh_entries));
1352 path = npath;
1353 goto repeat;
1354 }
1355 ext_debug("next leaf has no free space(%d,%d)\n",
1356 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1357 }
1358
1359 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001360 * There is no free space in the found leaf.
1361 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001362 */
1363 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1364 if (err)
1365 goto cleanup;
1366 depth = ext_depth(inode);
1367 eh = path[depth].p_hdr;
1368
1369has_space:
1370 nearex = path[depth].p_ext;
1371
Avantika Mathur7e028972006-12-06 20:41:33 -08001372 err = ext4_ext_get_access(handle, inode, path + depth);
1373 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001374 goto cleanup;
1375
1376 if (!nearex) {
1377 /* there is no extent in this leaf, create first one */
Mingming Cao2ae02102006-10-11 01:21:11 -07001378 ext_debug("first extent in the leaf: %d:%llu:%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001379 le32_to_cpu(newext->ee_block),
1380 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001381 ext4_ext_get_actual_len(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001382 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1383 } else if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001384 > le32_to_cpu(nearex->ee_block)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001385/* BUG_ON(newext->ee_block == nearex->ee_block); */
1386 if (nearex != EXT_LAST_EXTENT(eh)) {
1387 len = EXT_MAX_EXTENT(eh) - nearex;
1388 len = (len - 1) * sizeof(struct ext4_extent);
1389 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001390 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001391 "move %d from 0x%p to 0x%p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001392 le32_to_cpu(newext->ee_block),
1393 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001394 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001395 nearex, len, nearex + 1, nearex + 2);
1396 memmove(nearex + 2, nearex + 1, len);
1397 }
1398 path[depth].p_ext = nearex + 1;
1399 } else {
1400 BUG_ON(newext->ee_block == nearex->ee_block);
1401 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1402 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001403 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001404 "move %d from 0x%p to 0x%p\n",
1405 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001406 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001407 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001408 nearex, len, nearex + 1, nearex + 2);
1409 memmove(nearex + 1, nearex, len);
1410 path[depth].p_ext = nearex;
1411 }
1412
1413 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1414 nearex = path[depth].p_ext;
1415 nearex->ee_block = newext->ee_block;
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -04001416 ext4_ext_store_pblock(nearex, ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001417 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001418
1419merge:
1420 /* try to merge extents to the right */
Amit Arora56055d32007-07-17 21:42:38 -04001421 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001422
1423 /* try to merge extents to the left */
1424
1425 /* time to correct all indexes above */
1426 err = ext4_ext_correct_indexes(handle, inode, path);
1427 if (err)
1428 goto cleanup;
1429
1430 err = ext4_ext_dirty(handle, inode, path + depth);
1431
1432cleanup:
1433 if (npath) {
1434 ext4_ext_drop_refs(npath);
1435 kfree(npath);
1436 }
1437 ext4_ext_tree_changed(inode);
1438 ext4_ext_invalidate_cache(inode);
1439 return err;
1440}
1441
Avantika Mathur09b88252006-12-06 20:41:36 -08001442static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001443ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Mingming Caodd545672007-07-31 00:37:46 -07001444 __u32 len, ext4_fsblk_t start, int type)
Alex Tomasa86c6182006-10-11 01:21:03 -07001445{
1446 struct ext4_ext_cache *cex;
1447 BUG_ON(len == 0);
1448 cex = &EXT4_I(inode)->i_cached_extent;
1449 cex->ec_type = type;
1450 cex->ec_block = block;
1451 cex->ec_len = len;
1452 cex->ec_start = start;
1453}
1454
1455/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001456 * ext4_ext_put_gap_in_cache:
1457 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001458 * and cache this gap
1459 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001460static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001461ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001462 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001463{
1464 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001465 unsigned long len;
1466 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001467 struct ext4_extent *ex;
1468
1469 ex = path[depth].p_ext;
1470 if (ex == NULL) {
1471 /* there is no extent yet, so gap is [0;-] */
1472 lblock = 0;
1473 len = EXT_MAX_BLOCK;
1474 ext_debug("cache gap(whole file):");
1475 } else if (block < le32_to_cpu(ex->ee_block)) {
1476 lblock = block;
1477 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05001478 ext_debug("cache gap(before): %u [%u:%u]",
1479 block,
1480 le32_to_cpu(ex->ee_block),
1481 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001482 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001483 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001484 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001485 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001486 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001487
1488 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05001489 ext_debug("cache gap(after): [%u:%u] %u",
1490 le32_to_cpu(ex->ee_block),
1491 ext4_ext_get_actual_len(ex),
1492 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001493 BUG_ON(next == lblock);
1494 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001495 } else {
1496 lblock = len = 0;
1497 BUG();
1498 }
1499
Eric Sandeenbba90742008-01-28 23:58:27 -05001500 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001501 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1502}
1503
Avantika Mathur09b88252006-12-06 20:41:36 -08001504static int
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001505ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
Alex Tomasa86c6182006-10-11 01:21:03 -07001506 struct ext4_extent *ex)
1507{
1508 struct ext4_ext_cache *cex;
1509
1510 cex = &EXT4_I(inode)->i_cached_extent;
1511
1512 /* has cache valid data? */
1513 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1514 return EXT4_EXT_CACHE_NO;
1515
1516 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1517 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1518 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001519 ex->ee_block = cpu_to_le32(cex->ec_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001520 ext4_ext_store_pblock(ex, cex->ec_start);
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001521 ex->ee_len = cpu_to_le16(cex->ec_len);
Eric Sandeenbba90742008-01-28 23:58:27 -05001522 ext_debug("%u cached by %u:%u:%llu\n",
1523 block,
1524 cex->ec_block, cex->ec_len, cex->ec_start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001525 return cex->ec_type;
1526 }
1527
1528 /* not in cache */
1529 return EXT4_EXT_CACHE_NO;
1530}
1531
1532/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001533 * ext4_ext_rm_idx:
1534 * removes index from the index block.
1535 * It's used in truncate case only, thus all requests are for
1536 * last index in the block only.
Alex Tomasa86c6182006-10-11 01:21:03 -07001537 */
1538int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1539 struct ext4_ext_path *path)
1540{
1541 struct buffer_head *bh;
1542 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001543 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07001544
1545 /* free index block */
1546 path--;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001547 leaf = idx_pblock(path->p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -07001548 BUG_ON(path->p_hdr->eh_entries == 0);
Avantika Mathur7e028972006-12-06 20:41:33 -08001549 err = ext4_ext_get_access(handle, inode, path);
1550 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001551 return err;
1552 path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
Avantika Mathur7e028972006-12-06 20:41:33 -08001553 err = ext4_ext_dirty(handle, inode, path);
1554 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001555 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07001556 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Alex Tomasa86c6182006-10-11 01:21:03 -07001557 bh = sb_find_get_block(inode->i_sb, leaf);
1558 ext4_forget(handle, 1, inode, bh, leaf);
1559 ext4_free_blocks(handle, inode, leaf, 1);
1560 return err;
1561}
1562
1563/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001564 * ext4_ext_calc_credits_for_insert:
1565 * This routine returns max. credits that the extent tree can consume.
Alex Tomasa86c6182006-10-11 01:21:03 -07001566 * It should be OK for low-performance paths like ->writepage()
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001567 * To allow many writing processes to fit into a single transaction,
1568 * the caller should calculate credits under truncate_mutex and
1569 * pass the actual path.
Alex Tomasa86c6182006-10-11 01:21:03 -07001570 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001571int ext4_ext_calc_credits_for_insert(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001572 struct ext4_ext_path *path)
1573{
1574 int depth, needed;
1575
1576 if (path) {
1577 /* probably there is space in leaf? */
1578 depth = ext_depth(inode);
1579 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1580 < le16_to_cpu(path[depth].p_hdr->eh_max))
1581 return 1;
1582 }
1583
1584 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001585 * given 32-bit logical block (4294967296 blocks), max. tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001586 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001587 * Let's also add one more level for imbalance.
Alex Tomasa86c6182006-10-11 01:21:03 -07001588 */
1589 depth = 5;
1590
1591 /* allocation of new data block(s) */
1592 needed = 2;
1593
1594 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001595 * tree can be full, so it would need to grow in depth:
Johann Lombardifeb18922006-12-06 20:40:46 -08001596 * we need one credit to modify old root, credits for
1597 * new root will be added in split accounting
Alex Tomasa86c6182006-10-11 01:21:03 -07001598 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001599 needed += 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001600
1601 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001602 * Index split can happen, we would need:
Alex Tomasa86c6182006-10-11 01:21:03 -07001603 * allocate intermediate indexes (bitmap + group)
1604 * + change two blocks at each level, but root (already included)
1605 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001606 needed += (depth * 2) + (depth * 2);
Alex Tomasa86c6182006-10-11 01:21:03 -07001607
1608 /* any allocation modifies superblock */
1609 needed += 1;
1610
1611 return needed;
1612}
1613
1614static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1615 struct ext4_extent *ex,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001616 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07001617{
1618 struct buffer_head *bh;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001619 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001620 int i;
1621
1622#ifdef EXTENTS_STATS
1623 {
1624 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07001625 spin_lock(&sbi->s_ext_stats_lock);
1626 sbi->s_ext_blocks += ee_len;
1627 sbi->s_ext_extents++;
1628 if (ee_len < sbi->s_ext_min)
1629 sbi->s_ext_min = ee_len;
1630 if (ee_len > sbi->s_ext_max)
1631 sbi->s_ext_max = ee_len;
1632 if (ext_depth(inode) > sbi->s_depth_max)
1633 sbi->s_depth_max = ext_depth(inode);
1634 spin_unlock(&sbi->s_ext_stats_lock);
1635 }
1636#endif
1637 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001638 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001639 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001640 ext4_lblk_t num;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001641 ext4_fsblk_t start;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001642
Amit Aroraa2df2a62007-07-17 21:42:41 -04001643 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1644 start = ext_pblock(ex) + ee_len - num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001645 ext_debug("free last %u blocks starting %llu\n", num, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001646 for (i = 0; i < num; i++) {
1647 bh = sb_find_get_block(inode->i_sb, start + i);
1648 ext4_forget(handle, 0, inode, bh, start + i);
1649 }
1650 ext4_free_blocks(handle, inode, start, num);
1651 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001652 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001653 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001654 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001655 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001656 printk(KERN_INFO "strange request: removal(2) "
1657 "%u-%u from %u:%u\n",
1658 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001659 }
1660 return 0;
1661}
1662
1663static int
1664ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001665 struct ext4_ext_path *path, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001666{
1667 int err = 0, correct_index = 0;
1668 int depth = ext_depth(inode), credits;
1669 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001670 ext4_lblk_t a, b, block;
1671 unsigned num;
1672 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07001673 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001674 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001675 struct ext4_extent *ex;
1676
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001677 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001678 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001679 if (!path[depth].p_hdr)
1680 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1681 eh = path[depth].p_hdr;
1682 BUG_ON(eh == NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001683
1684 /* find where to start removing */
1685 ex = EXT_LAST_EXTENT(eh);
1686
1687 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001688 if (ext4_ext_is_uninitialized(ex))
1689 uninitialized = 1;
1690 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001691
1692 while (ex >= EXT_FIRST_EXTENT(eh) &&
1693 ex_ee_block + ex_ee_len > start) {
1694 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1695 path[depth].p_ext = ex;
1696
1697 a = ex_ee_block > start ? ex_ee_block : start;
1698 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1699 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1700
1701 ext_debug(" border %u:%u\n", a, b);
1702
1703 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1704 block = 0;
1705 num = 0;
1706 BUG();
1707 } else if (a != ex_ee_block) {
1708 /* remove tail of the extent */
1709 block = ex_ee_block;
1710 num = a - block;
1711 } else if (b != ex_ee_block + ex_ee_len - 1) {
1712 /* remove head of the extent */
1713 block = a;
1714 num = b - a;
1715 /* there is no "make a hole" API yet */
1716 BUG();
1717 } else {
1718 /* remove whole extent: excellent! */
1719 block = ex_ee_block;
1720 num = 0;
1721 BUG_ON(a != ex_ee_block);
1722 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1723 }
1724
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001725 /* at present, extent can't cross block group: */
Alex Tomasa86c6182006-10-11 01:21:03 -07001726 /* leaf + bitmap + group desc + sb + inode */
1727 credits = 5;
1728 if (ex == EXT_FIRST_EXTENT(eh)) {
1729 correct_index = 1;
1730 credits += (ext_depth(inode)) + 1;
1731 }
1732#ifdef CONFIG_QUOTA
1733 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1734#endif
1735
1736 handle = ext4_ext_journal_restart(handle, credits);
1737 if (IS_ERR(handle)) {
1738 err = PTR_ERR(handle);
1739 goto out;
1740 }
1741
1742 err = ext4_ext_get_access(handle, inode, path + depth);
1743 if (err)
1744 goto out;
1745
1746 err = ext4_remove_blocks(handle, inode, ex, a, b);
1747 if (err)
1748 goto out;
1749
1750 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001751 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001752 ext4_ext_store_pblock(ex, 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001753 eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1754 }
1755
1756 ex->ee_block = cpu_to_le32(block);
1757 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04001758 /*
1759 * Do not mark uninitialized if all the blocks in the
1760 * extent have been removed.
1761 */
1762 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001763 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001764
1765 err = ext4_ext_dirty(handle, inode, path + depth);
1766 if (err)
1767 goto out;
1768
Mingming Cao2ae02102006-10-11 01:21:11 -07001769 ext_debug("new extent: %u:%u:%llu\n", block, num,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001770 ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001771 ex--;
1772 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001773 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001774 }
1775
1776 if (correct_index && eh->eh_entries)
1777 err = ext4_ext_correct_indexes(handle, inode, path);
1778
1779 /* if this leaf is free, then we should
1780 * remove it from index block above */
1781 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1782 err = ext4_ext_rm_idx(handle, inode, path + depth);
1783
1784out:
1785 return err;
1786}
1787
1788/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001789 * ext4_ext_more_to_rm:
1790 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07001791 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001792static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001793ext4_ext_more_to_rm(struct ext4_ext_path *path)
1794{
1795 BUG_ON(path->p_idx == NULL);
1796
1797 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1798 return 0;
1799
1800 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001801 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07001802 * so we have to consider current index for truncation
1803 */
1804 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1805 return 0;
1806 return 1;
1807}
1808
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001809int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001810{
1811 struct super_block *sb = inode->i_sb;
1812 int depth = ext_depth(inode);
1813 struct ext4_ext_path *path;
1814 handle_t *handle;
1815 int i = 0, err = 0;
1816
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001817 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001818
1819 /* probably first extent we're gonna free will be last in block */
1820 handle = ext4_journal_start(inode, depth + 1);
1821 if (IS_ERR(handle))
1822 return PTR_ERR(handle);
1823
1824 ext4_ext_invalidate_cache(inode);
1825
1826 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001827 * We start scanning from right side, freeing all the blocks
1828 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07001829 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -08001830 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001831 if (path == NULL) {
1832 ext4_journal_stop(handle);
1833 return -ENOMEM;
1834 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001835 path[0].p_hdr = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001836 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001837 err = -EIO;
1838 goto out;
1839 }
1840 path[0].p_depth = depth;
1841
1842 while (i >= 0 && err == 0) {
1843 if (i == depth) {
1844 /* this is leaf block */
1845 err = ext4_ext_rm_leaf(handle, inode, path, start);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001846 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07001847 brelse(path[i].p_bh);
1848 path[i].p_bh = NULL;
1849 i--;
1850 continue;
1851 }
1852
1853 /* this is index block */
1854 if (!path[i].p_hdr) {
1855 ext_debug("initialize header\n");
1856 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07001857 }
1858
Alex Tomasa86c6182006-10-11 01:21:03 -07001859 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001860 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07001861 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1862 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1863 ext_debug("init index ptr: hdr 0x%p, num %d\n",
1864 path[i].p_hdr,
1865 le16_to_cpu(path[i].p_hdr->eh_entries));
1866 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001867 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07001868 path[i].p_idx--;
1869 }
1870
1871 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
1872 i, EXT_FIRST_INDEX(path[i].p_hdr),
1873 path[i].p_idx);
1874 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001875 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001876 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07001877 ext_debug("move to level %d (block %llu)\n",
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001878 i + 1, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07001879 memset(path + i + 1, 0, sizeof(*path));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001880 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
1881 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001882 /* should we reset i_size? */
1883 err = -EIO;
1884 break;
1885 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001886 if (WARN_ON(i + 1 > depth)) {
1887 err = -EIO;
1888 break;
1889 }
1890 if (ext4_ext_check_header(inode, ext_block_hdr(bh),
1891 depth - i - 1)) {
1892 err = -EIO;
1893 break;
1894 }
1895 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001896
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001897 /* save actual number of indexes since this
1898 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07001899 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1900 i++;
1901 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001902 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07001903 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001904 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07001905 * handle must be already prepared by the
1906 * truncatei_leaf() */
1907 err = ext4_ext_rm_idx(handle, inode, path + i);
1908 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001909 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07001910 brelse(path[i].p_bh);
1911 path[i].p_bh = NULL;
1912 i--;
1913 ext_debug("return to level %d\n", i);
1914 }
1915 }
1916
1917 /* TODO: flexible tree reduction should be here */
1918 if (path->p_hdr->eh_entries == 0) {
1919 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001920 * truncate to zero freed all the tree,
1921 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07001922 */
1923 err = ext4_ext_get_access(handle, inode, path);
1924 if (err == 0) {
1925 ext_inode_hdr(inode)->eh_depth = 0;
1926 ext_inode_hdr(inode)->eh_max =
1927 cpu_to_le16(ext4_ext_space_root(inode));
1928 err = ext4_ext_dirty(handle, inode, path);
1929 }
1930 }
1931out:
1932 ext4_ext_tree_changed(inode);
1933 ext4_ext_drop_refs(path);
1934 kfree(path);
1935 ext4_journal_stop(handle);
1936
1937 return err;
1938}
1939
1940/*
1941 * called at mount time
1942 */
1943void ext4_ext_init(struct super_block *sb)
1944{
1945 /*
1946 * possible initialization would be here
1947 */
1948
1949 if (test_opt(sb, EXTENTS)) {
1950 printk("EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001951#ifdef AGGRESSIVE_TEST
1952 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07001953#endif
1954#ifdef CHECK_BINSEARCH
1955 printk(", check binsearch");
1956#endif
1957#ifdef EXTENTS_STATS
1958 printk(", stats");
1959#endif
1960 printk("\n");
1961#ifdef EXTENTS_STATS
1962 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
1963 EXT4_SB(sb)->s_ext_min = 1 << 30;
1964 EXT4_SB(sb)->s_ext_max = 0;
1965#endif
1966 }
1967}
1968
1969/*
1970 * called at umount time
1971 */
1972void ext4_ext_release(struct super_block *sb)
1973{
1974 if (!test_opt(sb, EXTENTS))
1975 return;
1976
1977#ifdef EXTENTS_STATS
1978 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
1979 struct ext4_sb_info *sbi = EXT4_SB(sb);
1980 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
1981 sbi->s_ext_blocks, sbi->s_ext_extents,
1982 sbi->s_ext_blocks / sbi->s_ext_extents);
1983 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
1984 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
1985 }
1986#endif
1987}
1988
Amit Arora56055d32007-07-17 21:42:38 -04001989/*
1990 * This function is called by ext4_ext_get_blocks() if someone tries to write
1991 * to an uninitialized extent. It may result in splitting the uninitialized
1992 * extent into multiple extents (upto three - one initialized and two
1993 * uninitialized).
1994 * There are three possibilities:
1995 * a> There is no split required: Entire extent should be initialized
1996 * b> Splits in two extents: Write is happening at either end of the extent
1997 * c> Splits in three extents: Somone is writing in middle of the extent
1998 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001999static int ext4_ext_convert_to_initialized(handle_t *handle,
2000 struct inode *inode,
2001 struct ext4_ext_path *path,
2002 ext4_lblk_t iblock,
2003 unsigned long max_blocks)
Amit Arora56055d32007-07-17 21:42:38 -04002004{
2005 struct ext4_extent *ex, newex;
2006 struct ext4_extent *ex1 = NULL;
2007 struct ext4_extent *ex2 = NULL;
2008 struct ext4_extent *ex3 = NULL;
2009 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002010 ext4_lblk_t ee_block;
2011 unsigned int allocated, ee_len, depth;
Amit Arora56055d32007-07-17 21:42:38 -04002012 ext4_fsblk_t newblock;
2013 int err = 0;
2014 int ret = 0;
2015
2016 depth = ext_depth(inode);
2017 eh = path[depth].p_hdr;
2018 ex = path[depth].p_ext;
2019 ee_block = le32_to_cpu(ex->ee_block);
2020 ee_len = ext4_ext_get_actual_len(ex);
2021 allocated = ee_len - (iblock - ee_block);
2022 newblock = iblock - ee_block + ext_pblock(ex);
2023 ex2 = ex;
2024
2025 /* ex1: ee_block to iblock - 1 : uninitialized */
2026 if (iblock > ee_block) {
2027 ex1 = ex;
2028 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2029 ext4_ext_mark_uninitialized(ex1);
2030 ex2 = &newex;
2031 }
2032 /*
2033 * for sanity, update the length of the ex2 extent before
2034 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2035 * overlap of blocks.
2036 */
2037 if (!ex1 && allocated > max_blocks)
2038 ex2->ee_len = cpu_to_le16(max_blocks);
2039 /* ex3: to ee_block + ee_len : uninitialised */
2040 if (allocated > max_blocks) {
2041 unsigned int newdepth;
2042 ex3 = &newex;
2043 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2044 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2045 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2046 ext4_ext_mark_uninitialized(ex3);
2047 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2048 if (err)
2049 goto out;
2050 /*
2051 * The depth, and hence eh & ex might change
2052 * as part of the insert above.
2053 */
2054 newdepth = ext_depth(inode);
2055 if (newdepth != depth) {
2056 depth = newdepth;
2057 path = ext4_ext_find_extent(inode, iblock, NULL);
2058 if (IS_ERR(path)) {
2059 err = PTR_ERR(path);
2060 path = NULL;
2061 goto out;
2062 }
2063 eh = path[depth].p_hdr;
2064 ex = path[depth].p_ext;
2065 if (ex2 != &newex)
2066 ex2 = ex;
2067 }
2068 allocated = max_blocks;
2069 }
2070 /*
2071 * If there was a change of depth as part of the
2072 * insertion of ex3 above, we need to update the length
2073 * of the ex1 extent again here
2074 */
2075 if (ex1 && ex1 != ex) {
2076 ex1 = ex;
2077 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2078 ext4_ext_mark_uninitialized(ex1);
2079 ex2 = &newex;
2080 }
2081 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2082 ex2->ee_block = cpu_to_le32(iblock);
Amit Arora56055d32007-07-17 21:42:38 -04002083 ext4_ext_store_pblock(ex2, newblock);
2084 ex2->ee_len = cpu_to_le16(allocated);
2085 if (ex2 != ex)
2086 goto insert;
2087 err = ext4_ext_get_access(handle, inode, path + depth);
2088 if (err)
2089 goto out;
2090 /*
2091 * New (initialized) extent starts from the first block
2092 * in the current extent. i.e., ex2 == ex
2093 * We have to see if it can be merged with the extent
2094 * on the left.
2095 */
2096 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2097 /*
2098 * To merge left, pass "ex2 - 1" to try_to_merge(),
2099 * since it merges towards right _only_.
2100 */
2101 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2102 if (ret) {
2103 err = ext4_ext_correct_indexes(handle, inode, path);
2104 if (err)
2105 goto out;
2106 depth = ext_depth(inode);
2107 ex2--;
2108 }
2109 }
2110 /*
2111 * Try to Merge towards right. This might be required
2112 * only when the whole extent is being written to.
2113 * i.e. ex2 == ex and ex3 == NULL.
2114 */
2115 if (!ex3) {
2116 ret = ext4_ext_try_to_merge(inode, path, ex2);
2117 if (ret) {
2118 err = ext4_ext_correct_indexes(handle, inode, path);
2119 if (err)
2120 goto out;
2121 }
2122 }
2123 /* Mark modified extent as dirty */
2124 err = ext4_ext_dirty(handle, inode, path + depth);
2125 goto out;
2126insert:
2127 err = ext4_ext_insert_extent(handle, inode, path, &newex);
2128out:
2129 return err ? err : allocated;
2130}
2131
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002132int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002133 ext4_lblk_t iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002134 unsigned long max_blocks, struct buffer_head *bh_result,
2135 int create, int extend_disksize)
2136{
2137 struct ext4_ext_path *path = NULL;
Amit Arora56055d32007-07-17 21:42:38 -04002138 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002139 struct ext4_extent newex, *ex;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002140 ext4_fsblk_t goal, newblock;
Amit Arora56055d32007-07-17 21:42:38 -04002141 int err = 0, depth, ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002142 unsigned long allocated = 0;
2143
2144 __clear_bit(BH_New, &bh_result->b_state);
Eric Sandeenbba90742008-01-28 23:58:27 -05002145 ext_debug("blocks %u/%lu requested for inode %u\n",
2146 iblock, max_blocks, inode->i_ino);
Alex Tomasa86c6182006-10-11 01:21:03 -07002147 mutex_lock(&EXT4_I(inode)->truncate_mutex);
2148
2149 /* check in cache */
Avantika Mathur7e028972006-12-06 20:41:33 -08002150 goal = ext4_ext_in_cache(inode, iblock, &newex);
2151 if (goal) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002152 if (goal == EXT4_EXT_CACHE_GAP) {
2153 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002154 /*
2155 * block isn't allocated yet and
2156 * user doesn't want to allocate it
2157 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002158 goto out2;
2159 }
2160 /* we should allocate requested block */
2161 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2162 /* block is already allocated */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002163 newblock = iblock
2164 - le32_to_cpu(newex.ee_block)
2165 + ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002166 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002167 allocated = le16_to_cpu(newex.ee_len) -
2168 (iblock - le32_to_cpu(newex.ee_block));
2169 goto out;
2170 } else {
2171 BUG();
2172 }
2173 }
2174
2175 /* find extent for this block */
2176 path = ext4_ext_find_extent(inode, iblock, NULL);
2177 if (IS_ERR(path)) {
2178 err = PTR_ERR(path);
2179 path = NULL;
2180 goto out2;
2181 }
2182
2183 depth = ext_depth(inode);
2184
2185 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002186 * consistent leaf must not be empty;
2187 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07002188 * this is why assert can't be put in ext4_ext_find_extent()
2189 */
2190 BUG_ON(path[depth].p_ext == NULL && depth != 0);
Amit Arora56055d32007-07-17 21:42:38 -04002191 eh = path[depth].p_hdr;
Alex Tomasa86c6182006-10-11 01:21:03 -07002192
Avantika Mathur7e028972006-12-06 20:41:33 -08002193 ex = path[depth].p_ext;
2194 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002195 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002196 ext4_fsblk_t ee_start = ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002197 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002198
2199 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002200 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04002201 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002202 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002203 ee_len = ext4_ext_get_actual_len(ex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002204 /* if found extent covers block, simply return it */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002205 if (iblock >= ee_block && iblock < ee_block + ee_len) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002206 newblock = iblock - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002207 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002208 allocated = ee_len - (iblock - ee_block);
Eric Sandeenbba90742008-01-28 23:58:27 -05002209 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002210 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04002211
Amit Aroraa2df2a62007-07-17 21:42:41 -04002212 /* Do not put uninitialized extent in the cache */
Amit Arora56055d32007-07-17 21:42:38 -04002213 if (!ext4_ext_is_uninitialized(ex)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002214 ext4_ext_put_in_cache(inode, ee_block,
2215 ee_len, ee_start,
2216 EXT4_EXT_CACHE_EXTENT);
Amit Arora56055d32007-07-17 21:42:38 -04002217 goto out;
2218 }
2219 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2220 goto out;
2221 if (!create)
2222 goto out2;
2223
2224 ret = ext4_ext_convert_to_initialized(handle, inode,
2225 path, iblock,
2226 max_blocks);
2227 if (ret <= 0)
2228 goto out2;
2229 else
2230 allocated = ret;
2231 goto outnew;
Alex Tomasa86c6182006-10-11 01:21:03 -07002232 }
2233 }
2234
2235 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002236 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07002237 * we couldn't try to create block if create flag is zero
2238 */
2239 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002240 /*
2241 * put just found gap into cache to speed up
2242 * subsequent requests
2243 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002244 ext4_ext_put_gap_in_cache(inode, path, iblock);
2245 goto out2;
2246 }
2247 /*
Andrew Morton63f57932006-10-11 01:21:24 -07002248 * Okay, we need to do block allocation. Lazily initialize the block
2249 * allocation info here if necessary.
2250 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002251 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2252 ext4_init_block_alloc_info(inode);
2253
2254 /* allocate new block */
2255 goal = ext4_ext_find_goal(inode, path, iblock);
Amit Arora25d14f92007-05-24 13:04:13 -04002256
Amit Arora749269f2007-07-18 09:02:56 -04002257 /*
2258 * See if request is beyond maximum number of blocks we can have in
2259 * a single extent. For an initialized extent this limit is
2260 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2261 * EXT_UNINIT_MAX_LEN.
2262 */
2263 if (max_blocks > EXT_INIT_MAX_LEN &&
2264 create != EXT4_CREATE_UNINITIALIZED_EXT)
2265 max_blocks = EXT_INIT_MAX_LEN;
2266 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2267 create == EXT4_CREATE_UNINITIALIZED_EXT)
2268 max_blocks = EXT_UNINIT_MAX_LEN;
2269
Amit Arora25d14f92007-05-24 13:04:13 -04002270 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2271 newex.ee_block = cpu_to_le32(iblock);
2272 newex.ee_len = cpu_to_le16(max_blocks);
2273 err = ext4_ext_check_overlap(inode, &newex, path);
2274 if (err)
2275 allocated = le16_to_cpu(newex.ee_len);
2276 else
2277 allocated = max_blocks;
Alex Tomasa86c6182006-10-11 01:21:03 -07002278 newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2279 if (!newblock)
2280 goto out2;
Mingming Cao2ae02102006-10-11 01:21:11 -07002281 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002282 goal, newblock, allocated);
2283
2284 /* try to insert new extent into found leaf and return */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002285 ext4_ext_store_pblock(&newex, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07002286 newex.ee_len = cpu_to_le16(allocated);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002287 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2288 ext4_ext_mark_uninitialized(&newex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002289 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Alex Tomas315054f2007-05-24 13:04:25 -04002290 if (err) {
2291 /* free data blocks we just allocated */
2292 ext4_free_blocks(handle, inode, ext_pblock(&newex),
2293 le16_to_cpu(newex.ee_len));
Alex Tomasa86c6182006-10-11 01:21:03 -07002294 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04002295 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002296
2297 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2298 EXT4_I(inode)->i_disksize = inode->i_size;
2299
2300 /* previous routine could use block we allocated */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002301 newblock = ext_pblock(&newex);
Amit Arora56055d32007-07-17 21:42:38 -04002302outnew:
Alex Tomasa86c6182006-10-11 01:21:03 -07002303 __set_bit(BH_New, &bh_result->b_state);
2304
Amit Aroraa2df2a62007-07-17 21:42:41 -04002305 /* Cache only when it is _not_ an uninitialized extent */
2306 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2307 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2308 EXT4_EXT_CACHE_EXTENT);
Alex Tomasa86c6182006-10-11 01:21:03 -07002309out:
2310 if (allocated > max_blocks)
2311 allocated = max_blocks;
2312 ext4_ext_show_leaf(inode, path);
2313 __set_bit(BH_Mapped, &bh_result->b_state);
2314 bh_result->b_bdev = inode->i_sb->s_bdev;
2315 bh_result->b_blocknr = newblock;
2316out2:
2317 if (path) {
2318 ext4_ext_drop_refs(path);
2319 kfree(path);
2320 }
2321 mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2322
2323 return err ? err : allocated;
2324}
2325
2326void ext4_ext_truncate(struct inode * inode, struct page *page)
2327{
2328 struct address_space *mapping = inode->i_mapping;
2329 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002330 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002331 handle_t *handle;
2332 int err = 0;
2333
2334 /*
2335 * probably first extent we're gonna free will be last in block
2336 */
2337 err = ext4_writepage_trans_blocks(inode) + 3;
2338 handle = ext4_journal_start(inode, err);
2339 if (IS_ERR(handle)) {
2340 if (page) {
2341 clear_highpage(page);
2342 flush_dcache_page(page);
2343 unlock_page(page);
2344 page_cache_release(page);
2345 }
2346 return;
2347 }
2348
2349 if (page)
2350 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2351
2352 mutex_lock(&EXT4_I(inode)->truncate_mutex);
2353 ext4_ext_invalidate_cache(inode);
2354
2355 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002356 * TODO: optimization is possible here.
2357 * Probably we need not scan at all,
2358 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07002359 */
2360 if (ext4_orphan_add(handle, inode))
2361 goto out_stop;
2362
2363 /* we have to know where to truncate from in crash case */
2364 EXT4_I(inode)->i_disksize = inode->i_size;
2365 ext4_mark_inode_dirty(handle, inode);
2366
2367 last_block = (inode->i_size + sb->s_blocksize - 1)
2368 >> EXT4_BLOCK_SIZE_BITS(sb);
2369 err = ext4_ext_remove_space(inode, last_block);
2370
2371 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04002372 * transaction synchronous.
2373 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002374 if (IS_SYNC(inode))
2375 handle->h_sync = 1;
2376
2377out_stop:
2378 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002379 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07002380 * then we need to clear up the orphan record which we created above.
2381 * However, if this was a real unlink then we were called by
2382 * ext4_delete_inode(), and we allow that function to clean up the
2383 * orphan info for us.
2384 */
2385 if (inode->i_nlink)
2386 ext4_orphan_del(handle, inode);
2387
2388 mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2389 ext4_journal_stop(handle);
2390}
2391
2392/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002393 * ext4_ext_writepage_trans_blocks:
2394 * calculate max number of blocks we could modify
Alex Tomasa86c6182006-10-11 01:21:03 -07002395 * in order to allocate new block for an inode
2396 */
2397int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2398{
2399 int needed;
2400
2401 needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2402
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002403 /* caller wants to allocate num blocks, but note it includes sb */
Alex Tomasa86c6182006-10-11 01:21:03 -07002404 needed = needed * num - (num - 1);
2405
2406#ifdef CONFIG_QUOTA
2407 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2408#endif
2409
2410 return needed;
2411}
Amit Aroraa2df2a62007-07-17 21:42:41 -04002412
2413/*
2414 * preallocate space for a file. This implements ext4's fallocate inode
2415 * operation, which gets called from sys_fallocate system call.
2416 * For block-mapped files, posix_fallocate should fall back to the method
2417 * of writing zeroes to the required new blocks (the same behavior which is
2418 * expected for file systems which do not support fallocate() system call).
2419 */
2420long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2421{
2422 handle_t *handle;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002423 ext4_lblk_t block;
2424 unsigned long max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002425 ext4_fsblk_t nblocks = 0;
2426 int ret = 0;
2427 int ret2 = 0;
2428 int retries = 0;
2429 struct buffer_head map_bh;
2430 unsigned int credits, blkbits = inode->i_blkbits;
2431
2432 /*
2433 * currently supporting (pre)allocate mode for extent-based
2434 * files _only_
2435 */
2436 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2437 return -EOPNOTSUPP;
2438
2439 /* preallocation to directories is currently not supported */
2440 if (S_ISDIR(inode->i_mode))
2441 return -ENODEV;
2442
2443 block = offset >> blkbits;
2444 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
2445 - block;
2446
2447 /*
2448 * credits to insert 1 extent into extent tree + buffers to be able to
2449 * modify 1 super block, 1 block bitmap and 1 group descriptor.
2450 */
2451 credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
2452retry:
2453 while (ret >= 0 && ret < max_blocks) {
2454 block = block + ret;
2455 max_blocks = max_blocks - ret;
2456 handle = ext4_journal_start(inode, credits);
2457 if (IS_ERR(handle)) {
2458 ret = PTR_ERR(handle);
2459 break;
2460 }
2461
2462 ret = ext4_ext_get_blocks(handle, inode, block,
2463 max_blocks, &map_bh,
2464 EXT4_CREATE_UNINITIALIZED_EXT, 0);
2465 WARN_ON(!ret);
2466 if (!ret) {
2467 ext4_error(inode->i_sb, "ext4_fallocate",
2468 "ext4_ext_get_blocks returned 0! inode#%lu"
Eric Sandeenbba90742008-01-28 23:58:27 -05002469 ", block=%u, max_blocks=%lu",
2470 inode->i_ino, block, max_blocks);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002471 ret = -EIO;
2472 ext4_mark_inode_dirty(handle, inode);
2473 ret2 = ext4_journal_stop(handle);
2474 break;
2475 }
2476 if (ret > 0) {
2477 /* check wrap through sign-bit/zero here */
2478 if ((block + ret) < 0 || (block + ret) < block) {
2479 ret = -EIO;
2480 ext4_mark_inode_dirty(handle, inode);
2481 ret2 = ext4_journal_stop(handle);
2482 break;
2483 }
2484 if (buffer_new(&map_bh) && ((block + ret) >
2485 (EXT4_BLOCK_ALIGN(i_size_read(inode), blkbits)
2486 >> blkbits)))
2487 nblocks = nblocks + ret;
2488 }
2489
2490 /* Update ctime if new blocks get allocated */
2491 if (nblocks) {
2492 struct timespec now;
2493
2494 now = current_fs_time(inode->i_sb);
2495 if (!timespec_equal(&inode->i_ctime, &now))
2496 inode->i_ctime = now;
2497 }
2498
2499 ext4_mark_inode_dirty(handle, inode);
2500 ret2 = ext4_journal_stop(handle);
2501 if (ret2)
2502 break;
2503 }
2504
2505 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2506 goto retry;
2507
2508 /*
2509 * Time to update the file size.
2510 * Update only when preallocation was requested beyond the file size.
2511 */
2512 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2513 (offset + len) > i_size_read(inode)) {
2514 if (ret > 0) {
2515 /*
2516 * if no error, we assume preallocation succeeded
2517 * completely
2518 */
2519 mutex_lock(&inode->i_mutex);
2520 i_size_write(inode, offset + len);
2521 EXT4_I(inode)->i_disksize = i_size_read(inode);
2522 mutex_unlock(&inode->i_mutex);
2523 } else if (ret < 0 && nblocks) {
2524 /* Handle partial allocation scenario */
2525 loff_t newsize;
2526
2527 mutex_lock(&inode->i_mutex);
2528 newsize = (nblocks << blkbits) + i_size_read(inode);
2529 i_size_write(inode, EXT4_BLOCK_ALIGN(newsize, blkbits));
2530 EXT4_I(inode)->i_disksize = i_size_read(inode);
2531 mutex_unlock(&inode->i_mutex);
2532 }
2533 }
2534
2535 return ret > 0 ? ret2 : ret;
2536}