blob: 50322a09bd01af54d66cf2d344765b8b92075332 [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>
Mingming Caocd02ff02007-10-16 18:38:25 -040035#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070036#include <linux/highuid.h>
37#include <linux/pagemap.h>
38#include <linux/quotaops.h>
39#include <linux/string.h>
40#include <linux/slab.h>
Amit Aroraa2df2a62007-07-17 21:42:41 -040041#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070042#include <asm/uaccess.h>
Eric Sandeen6873fa02008-10-07 00:46:36 -040043#include <linux/fiemap.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040044#include "ext4_jbd2.h"
45#include "ext4_extents.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070046
47
Randy Dunlapd0d856e2006-10-11 01:21:07 -070048/*
49 * ext_pblock:
50 * combine low and high parts of physical block number into ext4_fsblk_t
51 */
Akira Fujita748de672009-06-17 19:24:03 -040052ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070053{
54 ext4_fsblk_t block;
55
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040056 block = le32_to_cpu(ex->ee_start_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070057 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070058 return block;
59}
60
Randy Dunlapd0d856e2006-10-11 01:21:07 -070061/*
62 * idx_pblock:
63 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
64 */
Aneesh Kumar K.Vc14c6fd2008-01-28 23:58:26 -050065ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070066{
67 ext4_fsblk_t block;
68
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040069 block = le32_to_cpu(ix->ei_leaf_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070070 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070071 return block;
72}
73
Randy Dunlapd0d856e2006-10-11 01:21:07 -070074/*
75 * ext4_ext_store_pblock:
76 * stores a large physical block number into an extent struct,
77 * breaking it into parts
78 */
Aneesh Kumar K.Vc14c6fd2008-01-28 23:58:26 -050079void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070080{
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040081 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070082 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070083}
84
Randy Dunlapd0d856e2006-10-11 01:21:07 -070085/*
86 * ext4_idx_store_pblock:
87 * stores a large physical block number into an index struct,
88 * breaking it into parts
89 */
Avantika Mathur09b88252006-12-06 20:41:36 -080090static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070091{
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040092 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070093 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070094}
95
Shen Feng9102e4f2008-07-11 19:27:31 -040096static int ext4_ext_journal_restart(handle_t *handle, int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -070097{
98 int err;
99
Frank Mayhar03901312009-01-07 00:06:22 -0500100 if (!ext4_handle_valid(handle))
101 return 0;
Alex Tomasa86c6182006-10-11 01:21:03 -0700102 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -0400103 return 0;
104 err = ext4_journal_extend(handle, needed);
Theodore Ts'o0123c932008-08-01 20:57:54 -0400105 if (err <= 0)
Shen Feng9102e4f2008-07-11 19:27:31 -0400106 return err;
107 return ext4_journal_restart(handle, needed);
Alex Tomasa86c6182006-10-11 01:21:03 -0700108}
109
110/*
111 * could return:
112 * - EROFS
113 * - ENOMEM
114 */
115static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
116 struct ext4_ext_path *path)
117{
118 if (path->p_bh) {
119 /* path points to block */
120 return ext4_journal_get_write_access(handle, path->p_bh);
121 }
122 /* path points to leaf/index in inode body */
123 /* we use in-core data, no need to protect them */
124 return 0;
125}
126
127/*
128 * could return:
129 * - EROFS
130 * - ENOMEM
131 * - EIO
132 */
133static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
134 struct ext4_ext_path *path)
135{
136 int err;
137 if (path->p_bh) {
138 /* path points to block */
Frank Mayhar03901312009-01-07 00:06:22 -0500139 err = ext4_handle_dirty_metadata(handle, inode, path->p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700140 } else {
141 /* path points to leaf/index in inode body */
142 err = ext4_mark_inode_dirty(handle, inode);
143 }
144 return err;
145}
146
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700147static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700148 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500149 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700150{
151 struct ext4_inode_info *ei = EXT4_I(inode);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700152 ext4_fsblk_t bg_start;
Valerie Clement74d34872008-02-15 13:43:07 -0500153 ext4_fsblk_t last_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700154 ext4_grpblk_t colour;
Theodore Ts'oa4912122009-03-12 12:18:34 -0400155 ext4_group_t block_group;
156 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -0700157 int depth;
158
159 if (path) {
160 struct ext4_extent *ex;
161 depth = path->p_depth;
162
163 /* try to predict block placement */
Avantika Mathur7e028972006-12-06 20:41:33 -0800164 ex = path[depth].p_ext;
165 if (ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700166 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700167
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700168 /* it looks like index is empty;
169 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700170 if (path[depth].p_bh)
171 return path[depth].p_bh->b_blocknr;
172 }
173
174 /* OK. use inode's group */
Theodore Ts'oa4912122009-03-12 12:18:34 -0400175 block_group = ei->i_block_group;
176 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
177 /*
178 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
179 * block groups per flexgroup, reserve the first block
180 * group for directories and special files. Regular
181 * files will start at the second block group. This
182 * tends to speed up directory access and improves
183 * fsck times.
184 */
185 block_group &= ~(flex_size-1);
186 if (S_ISREG(inode->i_mode))
187 block_group++;
188 }
189 bg_start = (block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
Alex Tomasa86c6182006-10-11 01:21:03 -0700190 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
Valerie Clement74d34872008-02-15 13:43:07 -0500191 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
192
Theodore Ts'oa4912122009-03-12 12:18:34 -0400193 /*
194 * If we are doing delayed allocation, we don't need take
195 * colour into account.
196 */
197 if (test_opt(inode->i_sb, DELALLOC))
198 return bg_start;
199
Valerie Clement74d34872008-02-15 13:43:07 -0500200 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
201 colour = (current->pid % 16) *
Alex Tomasa86c6182006-10-11 01:21:03 -0700202 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Valerie Clement74d34872008-02-15 13:43:07 -0500203 else
204 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
Alex Tomasa86c6182006-10-11 01:21:03 -0700205 return bg_start + colour + block;
206}
207
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400208/*
209 * Allocation for a meta data block
210 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700211static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400212ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700213 struct ext4_ext_path *path,
214 struct ext4_extent *ex, int *err)
215{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700216 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700217
218 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Theodore Ts'o97df5d12008-12-12 12:41:28 -0500219 newblock = ext4_new_meta_blocks(handle, inode, goal, NULL, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700220 return newblock;
221}
222
Avantika Mathur09b88252006-12-06 20:41:36 -0800223static int ext4_ext_space_block(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700224{
225 int size;
226
227 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
228 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100229#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700230 if (size > 6)
231 size = 6;
232#endif
233 return size;
234}
235
Avantika Mathur09b88252006-12-06 20:41:36 -0800236static int ext4_ext_space_block_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700237{
238 int size;
239
240 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
241 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100242#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700243 if (size > 5)
244 size = 5;
245#endif
246 return size;
247}
248
Avantika Mathur09b88252006-12-06 20:41:36 -0800249static int ext4_ext_space_root(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700250{
251 int size;
252
253 size = sizeof(EXT4_I(inode)->i_data);
254 size -= sizeof(struct ext4_extent_header);
255 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100256#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700257 if (size > 3)
258 size = 3;
259#endif
260 return size;
261}
262
Avantika Mathur09b88252006-12-06 20:41:36 -0800263static int ext4_ext_space_root_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700264{
265 int size;
266
267 size = sizeof(EXT4_I(inode)->i_data);
268 size -= sizeof(struct ext4_extent_header);
269 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100270#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700271 if (size > 4)
272 size = 4;
273#endif
274 return size;
275}
276
Mingming Caod2a17632008-07-14 17:52:37 -0400277/*
278 * Calculate the number of metadata blocks needed
279 * to allocate @blocks
280 * Worse case is one block per extent
281 */
282int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
283{
284 int lcap, icap, rcap, leafs, idxs, num;
285 int newextents = blocks;
286
287 rcap = ext4_ext_space_root_idx(inode);
288 lcap = ext4_ext_space_block(inode);
289 icap = ext4_ext_space_block_idx(inode);
290
291 /* number of new leaf blocks needed */
292 num = leafs = (newextents + lcap - 1) / lcap;
293
294 /*
295 * Worse case, we need separate index block(s)
296 * to link all new leaf blocks
297 */
298 idxs = (leafs + icap - 1) / icap;
299 do {
300 num += idxs;
301 idxs = (idxs + icap - 1) / icap;
302 } while (idxs > rcap);
303
304 return num;
305}
306
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400307static int
308ext4_ext_max_entries(struct inode *inode, int depth)
309{
310 int max;
311
312 if (depth == ext_depth(inode)) {
313 if (depth == 0)
314 max = ext4_ext_space_root(inode);
315 else
316 max = ext4_ext_space_root_idx(inode);
317 } else {
318 if (depth == 0)
319 max = ext4_ext_space_block(inode);
320 else
321 max = ext4_ext_space_block_idx(inode);
322 }
323
324 return max;
325}
326
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400327static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
328{
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400329 ext4_fsblk_t block = ext_pblock(ext);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400330 int len = ext4_ext_get_actual_len(ext);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400331
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400332 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400333}
334
335static int ext4_valid_extent_idx(struct inode *inode,
336 struct ext4_extent_idx *ext_idx)
337{
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400338 ext4_fsblk_t block = idx_pblock(ext_idx);
Theodore Ts'oe84a26c2009-04-22 20:52:25 -0400339
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400340 return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400341}
342
343static int ext4_valid_extent_entries(struct inode *inode,
344 struct ext4_extent_header *eh,
345 int depth)
346{
347 struct ext4_extent *ext;
348 struct ext4_extent_idx *ext_idx;
349 unsigned short entries;
350 if (eh->eh_entries == 0)
351 return 1;
352
353 entries = le16_to_cpu(eh->eh_entries);
354
355 if (depth == 0) {
356 /* leaf entries */
357 ext = EXT_FIRST_EXTENT(eh);
358 while (entries) {
359 if (!ext4_valid_extent(inode, ext))
360 return 0;
361 ext++;
362 entries--;
363 }
364 } else {
365 ext_idx = EXT_FIRST_INDEX(eh);
366 while (entries) {
367 if (!ext4_valid_extent_idx(inode, ext_idx))
368 return 0;
369 ext_idx++;
370 entries--;
371 }
372 }
373 return 1;
374}
375
376static int __ext4_ext_check(const char *function, struct inode *inode,
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400377 struct ext4_extent_header *eh,
378 int depth)
379{
380 const char *error_msg;
381 int max = 0;
382
383 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
384 error_msg = "invalid magic";
385 goto corrupted;
386 }
387 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
388 error_msg = "unexpected eh_depth";
389 goto corrupted;
390 }
391 if (unlikely(eh->eh_max == 0)) {
392 error_msg = "invalid eh_max";
393 goto corrupted;
394 }
395 max = ext4_ext_max_entries(inode, depth);
396 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
397 error_msg = "too large eh_max";
398 goto corrupted;
399 }
400 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
401 error_msg = "invalid eh_entries";
402 goto corrupted;
403 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400404 if (!ext4_valid_extent_entries(inode, eh, depth)) {
405 error_msg = "invalid extent entries";
406 goto corrupted;
407 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400408 return 0;
409
410corrupted:
411 ext4_error(inode->i_sb, function,
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400412 "bad header/extent in inode #%lu: %s - magic %x, "
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400413 "entries %u, max %u(%u), depth %u(%u)",
414 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
415 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
416 max, le16_to_cpu(eh->eh_depth), depth);
417
418 return -EIO;
419}
420
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -0400421#define ext4_ext_check(inode, eh, depth) \
422 __ext4_ext_check(__func__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400423
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400424int ext4_ext_check_inode(struct inode *inode)
425{
426 return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode));
427}
428
Alex Tomasa86c6182006-10-11 01:21:03 -0700429#ifdef EXT_DEBUG
430static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
431{
432 int k, l = path->p_depth;
433
434 ext_debug("path:");
435 for (k = 0; k <= l; k++, path++) {
436 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700437 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700438 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700439 } else if (path->p_ext) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700440 ext_debug(" %d:%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700441 le32_to_cpu(path->p_ext->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400442 ext4_ext_get_actual_len(path->p_ext),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700443 ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700444 } else
445 ext_debug(" []");
446 }
447 ext_debug("\n");
448}
449
450static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
451{
452 int depth = ext_depth(inode);
453 struct ext4_extent_header *eh;
454 struct ext4_extent *ex;
455 int i;
456
457 if (!path)
458 return;
459
460 eh = path[depth].p_hdr;
461 ex = EXT_FIRST_EXTENT(eh);
462
463 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700464 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400465 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700466 }
467 ext_debug("\n");
468}
469#else
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400470#define ext4_ext_show_path(inode, path)
471#define ext4_ext_show_leaf(inode, path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700472#endif
473
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500474void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700475{
476 int depth = path->p_depth;
477 int i;
478
479 for (i = 0; i <= depth; i++, path++)
480 if (path->p_bh) {
481 brelse(path->p_bh);
482 path->p_bh = NULL;
483 }
484}
485
486/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700487 * ext4_ext_binsearch_idx:
488 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400489 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700490 */
491static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500492ext4_ext_binsearch_idx(struct inode *inode,
493 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700494{
495 struct ext4_extent_header *eh = path->p_hdr;
496 struct ext4_extent_idx *r, *l, *m;
497
Alex Tomasa86c6182006-10-11 01:21:03 -0700498
Eric Sandeenbba90742008-01-28 23:58:27 -0500499 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700500
501 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400502 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700503 while (l <= r) {
504 m = l + (r - l) / 2;
505 if (block < le32_to_cpu(m->ei_block))
506 r = m - 1;
507 else
508 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400509 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
510 m, le32_to_cpu(m->ei_block),
511 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700512 }
513
514 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700515 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400516 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700517
518#ifdef CHECK_BINSEARCH
519 {
520 struct ext4_extent_idx *chix, *ix;
521 int k;
522
523 chix = ix = EXT_FIRST_INDEX(eh);
524 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
525 if (k != 0 &&
526 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400527 printk(KERN_DEBUG "k=%d, ix=0x%p, "
528 "first=0x%p\n", k,
529 ix, EXT_FIRST_INDEX(eh));
530 printk(KERN_DEBUG "%u <= %u\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700531 le32_to_cpu(ix->ei_block),
532 le32_to_cpu(ix[-1].ei_block));
533 }
534 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400535 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700536 if (block < le32_to_cpu(ix->ei_block))
537 break;
538 chix = ix;
539 }
540 BUG_ON(chix != path->p_idx);
541 }
542#endif
543
544}
545
546/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700547 * ext4_ext_binsearch:
548 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400549 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700550 */
551static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500552ext4_ext_binsearch(struct inode *inode,
553 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700554{
555 struct ext4_extent_header *eh = path->p_hdr;
556 struct ext4_extent *r, *l, *m;
557
Alex Tomasa86c6182006-10-11 01:21:03 -0700558 if (eh->eh_entries == 0) {
559 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700560 * this leaf is empty:
561 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700562 */
563 return;
564 }
565
Eric Sandeenbba90742008-01-28 23:58:27 -0500566 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700567
568 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400569 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700570
571 while (l <= r) {
572 m = l + (r - l) / 2;
573 if (block < le32_to_cpu(m->ee_block))
574 r = m - 1;
575 else
576 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400577 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
578 m, le32_to_cpu(m->ee_block),
579 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700580 }
581
582 path->p_ext = l - 1;
Mingming Cao2ae02102006-10-11 01:21:11 -0700583 ext_debug(" -> %d:%llu:%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400584 le32_to_cpu(path->p_ext->ee_block),
585 ext_pblock(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400586 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700587
588#ifdef CHECK_BINSEARCH
589 {
590 struct ext4_extent *chex, *ex;
591 int k;
592
593 chex = ex = EXT_FIRST_EXTENT(eh);
594 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
595 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400596 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700597 if (block < le32_to_cpu(ex->ee_block))
598 break;
599 chex = ex;
600 }
601 BUG_ON(chex != path->p_ext);
602 }
603#endif
604
605}
606
607int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
608{
609 struct ext4_extent_header *eh;
610
611 eh = ext_inode_hdr(inode);
612 eh->eh_depth = 0;
613 eh->eh_entries = 0;
614 eh->eh_magic = EXT4_EXT_MAGIC;
615 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
616 ext4_mark_inode_dirty(handle, inode);
617 ext4_ext_invalidate_cache(inode);
618 return 0;
619}
620
621struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500622ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
623 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700624{
625 struct ext4_extent_header *eh;
626 struct buffer_head *bh;
627 short int depth, i, ppos = 0, alloc = 0;
628
629 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400630 depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -0700631
632 /* account possible depth increase */
633 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800634 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700635 GFP_NOFS);
636 if (!path)
637 return ERR_PTR(-ENOMEM);
638 alloc = 1;
639 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700640 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400641 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700642
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400643 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700644 /* walk through the tree */
645 while (i) {
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400646 int need_to_validate = 0;
647
Alex Tomasa86c6182006-10-11 01:21:03 -0700648 ext_debug("depth %d: num %d, max %d\n",
649 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400650
Alex Tomasa86c6182006-10-11 01:21:03 -0700651 ext4_ext_binsearch_idx(inode, path + ppos, block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700652 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700653 path[ppos].p_depth = i;
654 path[ppos].p_ext = NULL;
655
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400656 bh = sb_getblk(inode->i_sb, path[ppos].p_block);
657 if (unlikely(!bh))
Alex Tomasa86c6182006-10-11 01:21:03 -0700658 goto err;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400659 if (!bh_uptodate_or_lock(bh)) {
660 if (bh_submit_read(bh) < 0) {
661 put_bh(bh);
662 goto err;
663 }
664 /* validate the extent entries */
665 need_to_validate = 1;
666 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700667 eh = ext_block_hdr(bh);
668 ppos++;
669 BUG_ON(ppos > depth);
670 path[ppos].p_bh = bh;
671 path[ppos].p_hdr = eh;
672 i--;
673
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -0400674 if (need_to_validate && ext4_ext_check(inode, eh, i))
Alex Tomasa86c6182006-10-11 01:21:03 -0700675 goto err;
676 }
677
678 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700679 path[ppos].p_ext = NULL;
680 path[ppos].p_idx = NULL;
681
Alex Tomasa86c6182006-10-11 01:21:03 -0700682 /* find extent */
683 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400684 /* if not an empty leaf */
685 if (path[ppos].p_ext)
686 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700687
688 ext4_ext_show_path(inode, path);
689
690 return path;
691
692err:
693 ext4_ext_drop_refs(path);
694 if (alloc)
695 kfree(path);
696 return ERR_PTR(-EIO);
697}
698
699/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700700 * ext4_ext_insert_index:
701 * insert new index [@logical;@ptr] into the block at @curp;
702 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700703 */
704static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
705 struct ext4_ext_path *curp,
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700706 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700707{
708 struct ext4_extent_idx *ix;
709 int len, err;
710
Avantika Mathur7e028972006-12-06 20:41:33 -0800711 err = ext4_ext_get_access(handle, inode, curp);
712 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700713 return err;
714
715 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
716 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
717 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
718 /* insert after */
719 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
720 len = (len - 1) * sizeof(struct ext4_extent_idx);
721 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400722 ext_debug("insert new index %d after: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700723 "move %d from 0x%p to 0x%p\n",
724 logical, ptr, len,
725 (curp->p_idx + 1), (curp->p_idx + 2));
726 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
727 }
728 ix = curp->p_idx + 1;
729 } else {
730 /* insert before */
731 len = len * sizeof(struct ext4_extent_idx);
732 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400733 ext_debug("insert new index %d before: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700734 "move %d from 0x%p to 0x%p\n",
735 logical, ptr, len,
736 curp->p_idx, (curp->p_idx + 1));
737 memmove(curp->p_idx + 1, curp->p_idx, len);
738 ix = curp->p_idx;
739 }
740
741 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700742 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400743 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700744
745 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400746 > le16_to_cpu(curp->p_hdr->eh_max));
Alex Tomasa86c6182006-10-11 01:21:03 -0700747 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
748
749 err = ext4_ext_dirty(handle, inode, curp);
750 ext4_std_error(inode->i_sb, err);
751
752 return err;
753}
754
755/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700756 * ext4_ext_split:
757 * inserts new subtree into the path, using free index entry
758 * at depth @at:
759 * - allocates all needed blocks (new leaf and all intermediate index blocks)
760 * - makes decision where to split
761 * - moves remaining extents and index entries (right to the split point)
762 * into the newly allocated blocks
763 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700764 */
765static int ext4_ext_split(handle_t *handle, struct inode *inode,
766 struct ext4_ext_path *path,
767 struct ext4_extent *newext, int at)
768{
769 struct buffer_head *bh = NULL;
770 int depth = ext_depth(inode);
771 struct ext4_extent_header *neh;
772 struct ext4_extent_idx *fidx;
773 struct ext4_extent *ex;
774 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700775 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700776 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700777 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700778 int err = 0;
779
780 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700781 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700782
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700783 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700784 * border from split point */
785 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
786 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
787 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700788 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700789 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400790 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700791 } else {
792 border = newext->ee_block;
793 ext_debug("leaf will be added."
794 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400795 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700796 }
797
798 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700799 * If error occurs, then we break processing
800 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700801 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700802 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700803 */
804
805 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700806 * Get array to track all allocated blocks.
807 * We need this to handle errors and free blocks
808 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700809 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800810 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700811 if (!ablocks)
812 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700813
814 /* allocate all needed blocks */
815 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
816 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400817 newblock = ext4_ext_new_meta_block(handle, inode, path,
818 newext, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700819 if (newblock == 0)
820 goto cleanup;
821 ablocks[a] = newblock;
822 }
823
824 /* initialize new leaf */
825 newblock = ablocks[--a];
826 BUG_ON(newblock == 0);
827 bh = sb_getblk(inode->i_sb, newblock);
828 if (!bh) {
829 err = -EIO;
830 goto cleanup;
831 }
832 lock_buffer(bh);
833
Avantika Mathur7e028972006-12-06 20:41:33 -0800834 err = ext4_journal_get_create_access(handle, bh);
835 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700836 goto cleanup;
837
838 neh = ext_block_hdr(bh);
839 neh->eh_entries = 0;
840 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
841 neh->eh_magic = EXT4_EXT_MAGIC;
842 neh->eh_depth = 0;
843 ex = EXT_FIRST_EXTENT(neh);
844
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700845 /* move remainder of path[depth] to the new leaf */
Alex Tomasa86c6182006-10-11 01:21:03 -0700846 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
847 /* start copy from next extent */
848 /* TODO: we could do it by single memmove */
849 m = 0;
850 path[depth].p_ext++;
851 while (path[depth].p_ext <=
852 EXT_MAX_EXTENT(path[depth].p_hdr)) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700853 ext_debug("move %d:%llu:%d in new leaf %llu\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400854 le32_to_cpu(path[depth].p_ext->ee_block),
855 ext_pblock(path[depth].p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400856 ext4_ext_get_actual_len(path[depth].p_ext),
Alex Tomasa86c6182006-10-11 01:21:03 -0700857 newblock);
858 /*memmove(ex++, path[depth].p_ext++,
859 sizeof(struct ext4_extent));
860 neh->eh_entries++;*/
861 path[depth].p_ext++;
862 m++;
863 }
864 if (m) {
865 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400866 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700867 }
868
869 set_buffer_uptodate(bh);
870 unlock_buffer(bh);
871
Frank Mayhar03901312009-01-07 00:06:22 -0500872 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800873 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700874 goto cleanup;
875 brelse(bh);
876 bh = NULL;
877
878 /* correct old leaf */
879 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -0800880 err = ext4_ext_get_access(handle, inode, path + depth);
881 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700882 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400883 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -0800884 err = ext4_ext_dirty(handle, inode, path + depth);
885 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700886 goto cleanup;
887
888 }
889
890 /* create intermediate indexes */
891 k = depth - at - 1;
892 BUG_ON(k < 0);
893 if (k)
894 ext_debug("create %d intermediate indices\n", k);
895 /* insert new index into current index block */
896 /* current depth stored in i var */
897 i = depth - 1;
898 while (k--) {
899 oldblock = newblock;
900 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -0500901 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700902 if (!bh) {
903 err = -EIO;
904 goto cleanup;
905 }
906 lock_buffer(bh);
907
Avantika Mathur7e028972006-12-06 20:41:33 -0800908 err = ext4_journal_get_create_access(handle, bh);
909 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700910 goto cleanup;
911
912 neh = ext_block_hdr(bh);
913 neh->eh_entries = cpu_to_le16(1);
914 neh->eh_magic = EXT4_EXT_MAGIC;
915 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
916 neh->eh_depth = cpu_to_le16(depth - i);
917 fidx = EXT_FIRST_INDEX(neh);
918 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700919 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700920
Eric Sandeenbba90742008-01-28 23:58:27 -0500921 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
922 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700923 /* copy indexes */
924 m = 0;
925 path[i].p_idx++;
926
927 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
928 EXT_MAX_INDEX(path[i].p_hdr));
929 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
930 EXT_LAST_INDEX(path[i].p_hdr));
931 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400932 ext_debug("%d: move %d:%llu in new index %llu\n", i,
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400933 le32_to_cpu(path[i].p_idx->ei_block),
934 idx_pblock(path[i].p_idx),
935 newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700936 /*memmove(++fidx, path[i].p_idx++,
937 sizeof(struct ext4_extent_idx));
938 neh->eh_entries++;
939 BUG_ON(neh->eh_entries > neh->eh_max);*/
940 path[i].p_idx++;
941 m++;
942 }
943 if (m) {
944 memmove(++fidx, path[i].p_idx - m,
945 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400946 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700947 }
948 set_buffer_uptodate(bh);
949 unlock_buffer(bh);
950
Frank Mayhar03901312009-01-07 00:06:22 -0500951 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -0800952 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700953 goto cleanup;
954 brelse(bh);
955 bh = NULL;
956
957 /* correct old index */
958 if (m) {
959 err = ext4_ext_get_access(handle, inode, path + i);
960 if (err)
961 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400962 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700963 err = ext4_ext_dirty(handle, inode, path + i);
964 if (err)
965 goto cleanup;
966 }
967
968 i--;
969 }
970
971 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -0700972 err = ext4_ext_insert_index(handle, inode, path + at,
973 le32_to_cpu(border), newblock);
974
975cleanup:
976 if (bh) {
977 if (buffer_locked(bh))
978 unlock_buffer(bh);
979 brelse(bh);
980 }
981
982 if (err) {
983 /* free all allocated blocks in error case */
984 for (i = 0; i < depth; i++) {
985 if (!ablocks[i])
986 continue;
Alex Tomasc9de5602008-01-29 00:19:52 -0500987 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700988 }
989 }
990 kfree(ablocks);
991
992 return err;
993}
994
995/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700996 * ext4_ext_grow_indepth:
997 * implements tree growing procedure:
998 * - allocates new block
999 * - moves top-level data (index block or leaf) into the new block
1000 * - initializes new top-level, creating index that points to the
1001 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -07001002 */
1003static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1004 struct ext4_ext_path *path,
1005 struct ext4_extent *newext)
1006{
1007 struct ext4_ext_path *curp = path;
1008 struct ext4_extent_header *neh;
1009 struct ext4_extent_idx *fidx;
1010 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001011 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001012 int err = 0;
1013
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -04001014 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07001015 if (newblock == 0)
1016 return err;
1017
1018 bh = sb_getblk(inode->i_sb, newblock);
1019 if (!bh) {
1020 err = -EIO;
1021 ext4_std_error(inode->i_sb, err);
1022 return err;
1023 }
1024 lock_buffer(bh);
1025
Avantika Mathur7e028972006-12-06 20:41:33 -08001026 err = ext4_journal_get_create_access(handle, bh);
1027 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001028 unlock_buffer(bh);
1029 goto out;
1030 }
1031
1032 /* move top-level index/leaf into new block */
1033 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
1034
1035 /* set size of new block */
1036 neh = ext_block_hdr(bh);
1037 /* old root could have indexes or leaves
1038 * so calculate e_max right way */
1039 if (ext_depth(inode))
1040 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
1041 else
1042 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
1043 neh->eh_magic = EXT4_EXT_MAGIC;
1044 set_buffer_uptodate(bh);
1045 unlock_buffer(bh);
1046
Frank Mayhar03901312009-01-07 00:06:22 -05001047 err = ext4_handle_dirty_metadata(handle, inode, bh);
Avantika Mathur7e028972006-12-06 20:41:33 -08001048 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001049 goto out;
1050
1051 /* create index in new top-level index: num,max,pointer */
Avantika Mathur7e028972006-12-06 20:41:33 -08001052 err = ext4_ext_get_access(handle, inode, curp);
1053 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001054 goto out;
1055
1056 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
1057 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
1058 curp->p_hdr->eh_entries = cpu_to_le16(1);
1059 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
Dmitry Monakhove9f410b2007-07-18 09:09:15 -04001060
1061 if (path[0].p_hdr->eh_depth)
1062 curp->p_idx->ei_block =
1063 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
1064 else
1065 curp->p_idx->ei_block =
1066 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001067 ext4_idx_store_pblock(curp->p_idx, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001068
1069 neh = ext_inode_hdr(inode);
1070 fidx = EXT_FIRST_INDEX(neh);
Mingming Cao2ae02102006-10-11 01:21:11 -07001071 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07001072 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001073 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
Alex Tomasa86c6182006-10-11 01:21:03 -07001074
1075 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
1076 err = ext4_ext_dirty(handle, inode, curp);
1077out:
1078 brelse(bh);
1079
1080 return err;
1081}
1082
1083/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001084 * ext4_ext_create_new_leaf:
1085 * finds empty index and adds new leaf.
1086 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -07001087 */
1088static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1089 struct ext4_ext_path *path,
1090 struct ext4_extent *newext)
1091{
1092 struct ext4_ext_path *curp;
1093 int depth, i, err = 0;
1094
1095repeat:
1096 i = depth = ext_depth(inode);
1097
1098 /* walk up to the tree and look for free index entry */
1099 curp = path + depth;
1100 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1101 i--;
1102 curp--;
1103 }
1104
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001105 /* we use already allocated block for index block,
1106 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001107 if (EXT_HAS_FREE_INDEX(curp)) {
1108 /* if we found index with free entry, then use that
1109 * entry: create all needed subtree and add new leaf */
1110 err = ext4_ext_split(handle, inode, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001111 if (err)
1112 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001113
1114 /* refill path */
1115 ext4_ext_drop_refs(path);
1116 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001117 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1118 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001119 if (IS_ERR(path))
1120 err = PTR_ERR(path);
1121 } else {
1122 /* tree is full, time to grow in depth */
1123 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1124 if (err)
1125 goto out;
1126
1127 /* refill path */
1128 ext4_ext_drop_refs(path);
1129 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001130 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1131 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001132 if (IS_ERR(path)) {
1133 err = PTR_ERR(path);
1134 goto out;
1135 }
1136
1137 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001138 * only first (depth 0 -> 1) produces free space;
1139 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001140 */
1141 depth = ext_depth(inode);
1142 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001143 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001144 goto repeat;
1145 }
1146 }
1147
1148out:
1149 return err;
1150}
1151
1152/*
Alex Tomas1988b512008-01-28 23:58:27 -05001153 * search the closest allocated block to the left for *logical
1154 * and returns it at @logical + it's physical address at @phys
1155 * if *logical is the smallest allocated block, the function
1156 * returns 0 at @phys
1157 * return value contains 0 (success) or error code
1158 */
1159int
1160ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1161 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1162{
1163 struct ext4_extent_idx *ix;
1164 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001165 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001166
1167 BUG_ON(path == NULL);
1168 depth = path->p_depth;
1169 *phys = 0;
1170
1171 if (depth == 0 && path->p_ext == NULL)
1172 return 0;
1173
1174 /* usually extent in the path covers blocks smaller
1175 * then *logical, but it can be that extent is the
1176 * first one in the file */
1177
1178 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001179 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001180 if (*logical < le32_to_cpu(ex->ee_block)) {
1181 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1182 while (--depth >= 0) {
1183 ix = path[depth].p_idx;
1184 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1185 }
1186 return 0;
1187 }
1188
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001189 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
Alex Tomas1988b512008-01-28 23:58:27 -05001190
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001191 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1192 *phys = ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001193 return 0;
1194}
1195
1196/*
1197 * search the closest allocated block to the right for *logical
1198 * and returns it at @logical + it's physical address at @phys
1199 * if *logical is the smallest allocated block, the function
1200 * returns 0 at @phys
1201 * return value contains 0 (success) or error code
1202 */
1203int
1204ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1205 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1206{
1207 struct buffer_head *bh = NULL;
1208 struct ext4_extent_header *eh;
1209 struct ext4_extent_idx *ix;
1210 struct ext4_extent *ex;
1211 ext4_fsblk_t block;
Eric Sandeen395a87b2009-03-10 18:18:47 -04001212 int depth; /* Note, NOT eh_depth; depth from top of tree */
1213 int ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001214
1215 BUG_ON(path == NULL);
1216 depth = path->p_depth;
1217 *phys = 0;
1218
1219 if (depth == 0 && path->p_ext == NULL)
1220 return 0;
1221
1222 /* usually extent in the path covers blocks smaller
1223 * then *logical, but it can be that extent is the
1224 * first one in the file */
1225
1226 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001227 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001228 if (*logical < le32_to_cpu(ex->ee_block)) {
1229 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1230 while (--depth >= 0) {
1231 ix = path[depth].p_idx;
1232 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1233 }
1234 *logical = le32_to_cpu(ex->ee_block);
1235 *phys = ext_pblock(ex);
1236 return 0;
1237 }
1238
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001239 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
Alex Tomas1988b512008-01-28 23:58:27 -05001240
1241 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1242 /* next allocated block in this leaf */
1243 ex++;
1244 *logical = le32_to_cpu(ex->ee_block);
1245 *phys = ext_pblock(ex);
1246 return 0;
1247 }
1248
1249 /* go up and search for index to the right */
1250 while (--depth >= 0) {
1251 ix = path[depth].p_idx;
1252 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001253 goto got_index;
Alex Tomas1988b512008-01-28 23:58:27 -05001254 }
1255
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001256 /* we've gone up to the root and found no index to the right */
1257 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001258
Wu Fengguang25f1ee32008-11-25 17:24:23 -05001259got_index:
Alex Tomas1988b512008-01-28 23:58:27 -05001260 /* we've found index to the right, let's
1261 * follow it and find the closest allocated
1262 * block to the right */
1263 ix++;
1264 block = idx_pblock(ix);
1265 while (++depth < path->p_depth) {
1266 bh = sb_bread(inode->i_sb, block);
1267 if (bh == NULL)
1268 return -EIO;
1269 eh = ext_block_hdr(bh);
Eric Sandeen395a87b2009-03-10 18:18:47 -04001270 /* subtract from p_depth to get proper eh_depth */
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04001271 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001272 put_bh(bh);
1273 return -EIO;
1274 }
1275 ix = EXT_FIRST_INDEX(eh);
1276 block = idx_pblock(ix);
1277 put_bh(bh);
1278 }
1279
1280 bh = sb_bread(inode->i_sb, block);
1281 if (bh == NULL)
1282 return -EIO;
1283 eh = ext_block_hdr(bh);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04001284 if (ext4_ext_check(inode, eh, path->p_depth - depth)) {
Alex Tomas1988b512008-01-28 23:58:27 -05001285 put_bh(bh);
1286 return -EIO;
1287 }
1288 ex = EXT_FIRST_EXTENT(eh);
1289 *logical = le32_to_cpu(ex->ee_block);
1290 *phys = ext_pblock(ex);
1291 put_bh(bh);
1292 return 0;
Alex Tomas1988b512008-01-28 23:58:27 -05001293}
1294
1295/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001296 * ext4_ext_next_allocated_block:
1297 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1298 * NOTE: it considers block number from index entry as
1299 * allocated block. Thus, index entries have to be consistent
1300 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001301 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001302static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001303ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1304{
1305 int depth;
1306
1307 BUG_ON(path == NULL);
1308 depth = path->p_depth;
1309
1310 if (depth == 0 && path->p_ext == NULL)
1311 return EXT_MAX_BLOCK;
1312
1313 while (depth >= 0) {
1314 if (depth == path->p_depth) {
1315 /* leaf */
1316 if (path[depth].p_ext !=
1317 EXT_LAST_EXTENT(path[depth].p_hdr))
1318 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1319 } else {
1320 /* index */
1321 if (path[depth].p_idx !=
1322 EXT_LAST_INDEX(path[depth].p_hdr))
1323 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1324 }
1325 depth--;
1326 }
1327
1328 return EXT_MAX_BLOCK;
1329}
1330
1331/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001332 * ext4_ext_next_leaf_block:
Alex Tomasa86c6182006-10-11 01:21:03 -07001333 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1334 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001335static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
Andrew Morton63f57932006-10-11 01:21:24 -07001336 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001337{
1338 int depth;
1339
1340 BUG_ON(path == NULL);
1341 depth = path->p_depth;
1342
1343 /* zero-tree has no leaf blocks at all */
1344 if (depth == 0)
1345 return EXT_MAX_BLOCK;
1346
1347 /* go to index block */
1348 depth--;
1349
1350 while (depth >= 0) {
1351 if (path[depth].p_idx !=
1352 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001353 return (ext4_lblk_t)
1354 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001355 depth--;
1356 }
1357
1358 return EXT_MAX_BLOCK;
1359}
1360
1361/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001362 * ext4_ext_correct_indexes:
1363 * if leaf gets modified and modified extent is first in the leaf,
1364 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001365 * TODO: do we need to correct tree in all cases?
1366 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001367static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001368 struct ext4_ext_path *path)
1369{
1370 struct ext4_extent_header *eh;
1371 int depth = ext_depth(inode);
1372 struct ext4_extent *ex;
1373 __le32 border;
1374 int k, err = 0;
1375
1376 eh = path[depth].p_hdr;
1377 ex = path[depth].p_ext;
1378 BUG_ON(ex == NULL);
1379 BUG_ON(eh == NULL);
1380
1381 if (depth == 0) {
1382 /* there is no tree at all */
1383 return 0;
1384 }
1385
1386 if (ex != EXT_FIRST_EXTENT(eh)) {
1387 /* we correct tree if first leaf got modified only */
1388 return 0;
1389 }
1390
1391 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001392 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001393 */
1394 k = depth - 1;
1395 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001396 err = ext4_ext_get_access(handle, inode, path + k);
1397 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001398 return err;
1399 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001400 err = ext4_ext_dirty(handle, inode, path + k);
1401 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001402 return err;
1403
1404 while (k--) {
1405 /* change all left-side indexes */
1406 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1407 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001408 err = ext4_ext_get_access(handle, inode, path + k);
1409 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001410 break;
1411 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001412 err = ext4_ext_dirty(handle, inode, path + k);
1413 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001414 break;
1415 }
1416
1417 return err;
1418}
1419
Akira Fujita748de672009-06-17 19:24:03 -04001420int
Alex Tomasa86c6182006-10-11 01:21:03 -07001421ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1422 struct ext4_extent *ex2)
1423{
Amit Arora749269f2007-07-18 09:02:56 -04001424 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001425
1426 /*
1427 * Make sure that either both extents are uninitialized, or
1428 * both are _not_.
1429 */
1430 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1431 return 0;
1432
Amit Arora749269f2007-07-18 09:02:56 -04001433 if (ext4_ext_is_uninitialized(ex1))
1434 max_len = EXT_UNINIT_MAX_LEN;
1435 else
1436 max_len = EXT_INIT_MAX_LEN;
1437
Amit Aroraa2df2a62007-07-17 21:42:41 -04001438 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1439 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1440
1441 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001442 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001443 return 0;
1444
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001445 /*
1446 * To allow future support for preallocated extents to be added
1447 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001448 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001449 */
Amit Arora749269f2007-07-18 09:02:56 -04001450 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001451 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001452#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001453 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001454 return 0;
1455#endif
1456
Amit Aroraa2df2a62007-07-17 21:42:41 -04001457 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001458 return 1;
1459 return 0;
1460}
1461
1462/*
Amit Arora56055d32007-07-17 21:42:38 -04001463 * This function tries to merge the "ex" extent to the next extent in the tree.
1464 * It always tries to merge towards right. If you want to merge towards
1465 * left, pass "ex - 1" as argument instead of "ex".
1466 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1467 * 1 if they got merged.
1468 */
1469int ext4_ext_try_to_merge(struct inode *inode,
1470 struct ext4_ext_path *path,
1471 struct ext4_extent *ex)
1472{
1473 struct ext4_extent_header *eh;
1474 unsigned int depth, len;
1475 int merge_done = 0;
1476 int uninitialized = 0;
1477
1478 depth = ext_depth(inode);
1479 BUG_ON(path[depth].p_hdr == NULL);
1480 eh = path[depth].p_hdr;
1481
1482 while (ex < EXT_LAST_EXTENT(eh)) {
1483 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1484 break;
1485 /* merge with next extent! */
1486 if (ext4_ext_is_uninitialized(ex))
1487 uninitialized = 1;
1488 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1489 + ext4_ext_get_actual_len(ex + 1));
1490 if (uninitialized)
1491 ext4_ext_mark_uninitialized(ex);
1492
1493 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1494 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1495 * sizeof(struct ext4_extent);
1496 memmove(ex + 1, ex + 2, len);
1497 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001498 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001499 merge_done = 1;
1500 WARN_ON(eh->eh_entries == 0);
1501 if (!eh->eh_entries)
1502 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1503 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1504 }
1505
1506 return merge_done;
1507}
1508
1509/*
Amit Arora25d14f92007-05-24 13:04:13 -04001510 * check if a portion of the "newext" extent overlaps with an
1511 * existing extent.
1512 *
1513 * If there is an overlap discovered, it updates the length of the newext
1514 * such that there will be no overlap, and then returns 1.
1515 * If there is no overlap found, it returns 0.
1516 */
1517unsigned int ext4_ext_check_overlap(struct inode *inode,
1518 struct ext4_extent *newext,
1519 struct ext4_ext_path *path)
1520{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001521 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001522 unsigned int depth, len1;
1523 unsigned int ret = 0;
1524
1525 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001526 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001527 depth = ext_depth(inode);
1528 if (!path[depth].p_ext)
1529 goto out;
1530 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1531
1532 /*
1533 * get the next allocated block if the extent in the path
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001534 * is before the requested block(s)
Amit Arora25d14f92007-05-24 13:04:13 -04001535 */
1536 if (b2 < b1) {
1537 b2 = ext4_ext_next_allocated_block(path);
1538 if (b2 == EXT_MAX_BLOCK)
1539 goto out;
1540 }
1541
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001542 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001543 if (b1 + len1 < b1) {
1544 len1 = EXT_MAX_BLOCK - b1;
1545 newext->ee_len = cpu_to_le16(len1);
1546 ret = 1;
1547 }
1548
1549 /* check for overlap */
1550 if (b1 + len1 > b2) {
1551 newext->ee_len = cpu_to_le16(b2 - b1);
1552 ret = 1;
1553 }
1554out:
1555 return ret;
1556}
1557
1558/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001559 * ext4_ext_insert_extent:
1560 * tries to merge requsted extent into the existing extent or
1561 * inserts requested extent as new one into the tree,
1562 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001563 */
1564int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1565 struct ext4_ext_path *path,
1566 struct ext4_extent *newext)
1567{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001568 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07001569 struct ext4_extent *ex, *fex;
1570 struct ext4_extent *nearex; /* nearest extent */
1571 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001572 int depth, len, err;
1573 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001574 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001575
Amit Aroraa2df2a62007-07-17 21:42:41 -04001576 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001577 depth = ext_depth(inode);
1578 ex = path[depth].p_ext;
1579 BUG_ON(path[depth].p_hdr == NULL);
1580
1581 /* try to insert block into found extent and return */
1582 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming Cao2ae02102006-10-11 01:21:11 -07001583 ext_debug("append %d block to %d:%d (from %llu)\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001584 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001585 le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001586 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001587 err = ext4_ext_get_access(handle, inode, path + depth);
1588 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001589 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001590
1591 /*
1592 * ext4_can_extents_be_merged should have checked that either
1593 * both extents are uninitialized, or both aren't. Thus we
1594 * need to check only one of them here.
1595 */
1596 if (ext4_ext_is_uninitialized(ex))
1597 uninitialized = 1;
1598 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1599 + ext4_ext_get_actual_len(newext));
1600 if (uninitialized)
1601 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001602 eh = path[depth].p_hdr;
1603 nearex = ex;
1604 goto merge;
1605 }
1606
1607repeat:
1608 depth = ext_depth(inode);
1609 eh = path[depth].p_hdr;
1610 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1611 goto has_space;
1612
1613 /* probably next leaf has space for us? */
1614 fex = EXT_LAST_EXTENT(eh);
1615 next = ext4_ext_next_leaf_block(inode, path);
1616 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1617 && next != EXT_MAX_BLOCK) {
1618 ext_debug("next leaf block - %d\n", next);
1619 BUG_ON(npath != NULL);
1620 npath = ext4_ext_find_extent(inode, next, NULL);
1621 if (IS_ERR(npath))
1622 return PTR_ERR(npath);
1623 BUG_ON(npath->p_depth != path->p_depth);
1624 eh = npath[depth].p_hdr;
1625 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1626 ext_debug("next leaf isnt full(%d)\n",
1627 le16_to_cpu(eh->eh_entries));
1628 path = npath;
1629 goto repeat;
1630 }
1631 ext_debug("next leaf has no free space(%d,%d)\n",
1632 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1633 }
1634
1635 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001636 * There is no free space in the found leaf.
1637 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001638 */
1639 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1640 if (err)
1641 goto cleanup;
1642 depth = ext_depth(inode);
1643 eh = path[depth].p_hdr;
1644
1645has_space:
1646 nearex = path[depth].p_ext;
1647
Avantika Mathur7e028972006-12-06 20:41:33 -08001648 err = ext4_ext_get_access(handle, inode, path + depth);
1649 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001650 goto cleanup;
1651
1652 if (!nearex) {
1653 /* there is no extent in this leaf, create first one */
Mingming Cao2ae02102006-10-11 01:21:11 -07001654 ext_debug("first extent in the leaf: %d:%llu:%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001655 le32_to_cpu(newext->ee_block),
1656 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001657 ext4_ext_get_actual_len(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001658 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1659 } else if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001660 > le32_to_cpu(nearex->ee_block)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001661/* BUG_ON(newext->ee_block == nearex->ee_block); */
1662 if (nearex != EXT_LAST_EXTENT(eh)) {
1663 len = EXT_MAX_EXTENT(eh) - nearex;
1664 len = (len - 1) * sizeof(struct ext4_extent);
1665 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001666 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001667 "move %d from 0x%p to 0x%p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001668 le32_to_cpu(newext->ee_block),
1669 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001670 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001671 nearex, len, nearex + 1, nearex + 2);
1672 memmove(nearex + 2, nearex + 1, len);
1673 }
1674 path[depth].p_ext = nearex + 1;
1675 } else {
1676 BUG_ON(newext->ee_block == nearex->ee_block);
1677 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1678 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001679 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001680 "move %d from 0x%p to 0x%p\n",
1681 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001682 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001683 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001684 nearex, len, nearex + 1, nearex + 2);
1685 memmove(nearex + 1, nearex, len);
1686 path[depth].p_ext = nearex;
1687 }
1688
Marcin Slusarze8546d02008-04-17 10:38:59 -04001689 le16_add_cpu(&eh->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001690 nearex = path[depth].p_ext;
1691 nearex->ee_block = newext->ee_block;
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -04001692 ext4_ext_store_pblock(nearex, ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001693 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001694
1695merge:
1696 /* try to merge extents to the right */
Amit Arora56055d32007-07-17 21:42:38 -04001697 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001698
1699 /* try to merge extents to the left */
1700
1701 /* time to correct all indexes above */
1702 err = ext4_ext_correct_indexes(handle, inode, path);
1703 if (err)
1704 goto cleanup;
1705
1706 err = ext4_ext_dirty(handle, inode, path + depth);
1707
1708cleanup:
1709 if (npath) {
1710 ext4_ext_drop_refs(npath);
1711 kfree(npath);
1712 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001713 ext4_ext_invalidate_cache(inode);
1714 return err;
1715}
1716
Eric Sandeen6873fa02008-10-07 00:46:36 -04001717int ext4_ext_walk_space(struct inode *inode, ext4_lblk_t block,
1718 ext4_lblk_t num, ext_prepare_callback func,
1719 void *cbdata)
1720{
1721 struct ext4_ext_path *path = NULL;
1722 struct ext4_ext_cache cbex;
1723 struct ext4_extent *ex;
1724 ext4_lblk_t next, start = 0, end = 0;
1725 ext4_lblk_t last = block + num;
1726 int depth, exists, err = 0;
1727
1728 BUG_ON(func == NULL);
1729 BUG_ON(inode == NULL);
1730
1731 while (block < last && block != EXT_MAX_BLOCK) {
1732 num = last - block;
1733 /* find extent for this block */
1734 path = ext4_ext_find_extent(inode, block, path);
1735 if (IS_ERR(path)) {
1736 err = PTR_ERR(path);
1737 path = NULL;
1738 break;
1739 }
1740
1741 depth = ext_depth(inode);
1742 BUG_ON(path[depth].p_hdr == NULL);
1743 ex = path[depth].p_ext;
1744 next = ext4_ext_next_allocated_block(path);
1745
1746 exists = 0;
1747 if (!ex) {
1748 /* there is no extent yet, so try to allocate
1749 * all requested space */
1750 start = block;
1751 end = block + num;
1752 } else if (le32_to_cpu(ex->ee_block) > block) {
1753 /* need to allocate space before found extent */
1754 start = block;
1755 end = le32_to_cpu(ex->ee_block);
1756 if (block + num < end)
1757 end = block + num;
1758 } else if (block >= le32_to_cpu(ex->ee_block)
1759 + ext4_ext_get_actual_len(ex)) {
1760 /* need to allocate space after found extent */
1761 start = block;
1762 end = block + num;
1763 if (end >= next)
1764 end = next;
1765 } else if (block >= le32_to_cpu(ex->ee_block)) {
1766 /*
1767 * some part of requested space is covered
1768 * by found extent
1769 */
1770 start = block;
1771 end = le32_to_cpu(ex->ee_block)
1772 + ext4_ext_get_actual_len(ex);
1773 if (block + num < end)
1774 end = block + num;
1775 exists = 1;
1776 } else {
1777 BUG();
1778 }
1779 BUG_ON(end <= start);
1780
1781 if (!exists) {
1782 cbex.ec_block = start;
1783 cbex.ec_len = end - start;
1784 cbex.ec_start = 0;
1785 cbex.ec_type = EXT4_EXT_CACHE_GAP;
1786 } else {
1787 cbex.ec_block = le32_to_cpu(ex->ee_block);
1788 cbex.ec_len = ext4_ext_get_actual_len(ex);
1789 cbex.ec_start = ext_pblock(ex);
1790 cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1791 }
1792
1793 BUG_ON(cbex.ec_len == 0);
1794 err = func(inode, path, &cbex, ex, cbdata);
1795 ext4_ext_drop_refs(path);
1796
1797 if (err < 0)
1798 break;
1799
1800 if (err == EXT_REPEAT)
1801 continue;
1802 else if (err == EXT_BREAK) {
1803 err = 0;
1804 break;
1805 }
1806
1807 if (ext_depth(inode) != depth) {
1808 /* depth was changed. we have to realloc path */
1809 kfree(path);
1810 path = NULL;
1811 }
1812
1813 block = cbex.ec_block + cbex.ec_len;
1814 }
1815
1816 if (path) {
1817 ext4_ext_drop_refs(path);
1818 kfree(path);
1819 }
1820
1821 return err;
1822}
1823
Avantika Mathur09b88252006-12-06 20:41:36 -08001824static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001825ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Mingming Caodd545672007-07-31 00:37:46 -07001826 __u32 len, ext4_fsblk_t start, int type)
Alex Tomasa86c6182006-10-11 01:21:03 -07001827{
1828 struct ext4_ext_cache *cex;
1829 BUG_ON(len == 0);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001830 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001831 cex = &EXT4_I(inode)->i_cached_extent;
1832 cex->ec_type = type;
1833 cex->ec_block = block;
1834 cex->ec_len = len;
1835 cex->ec_start = start;
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001836 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001837}
1838
1839/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001840 * ext4_ext_put_gap_in_cache:
1841 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001842 * and cache this gap
1843 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001844static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001845ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001846 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001847{
1848 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001849 unsigned long len;
1850 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001851 struct ext4_extent *ex;
1852
1853 ex = path[depth].p_ext;
1854 if (ex == NULL) {
1855 /* there is no extent yet, so gap is [0;-] */
1856 lblock = 0;
1857 len = EXT_MAX_BLOCK;
1858 ext_debug("cache gap(whole file):");
1859 } else if (block < le32_to_cpu(ex->ee_block)) {
1860 lblock = block;
1861 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05001862 ext_debug("cache gap(before): %u [%u:%u]",
1863 block,
1864 le32_to_cpu(ex->ee_block),
1865 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001866 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001867 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001868 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001869 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001870 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001871
1872 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05001873 ext_debug("cache gap(after): [%u:%u] %u",
1874 le32_to_cpu(ex->ee_block),
1875 ext4_ext_get_actual_len(ex),
1876 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001877 BUG_ON(next == lblock);
1878 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001879 } else {
1880 lblock = len = 0;
1881 BUG();
1882 }
1883
Eric Sandeenbba90742008-01-28 23:58:27 -05001884 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001885 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1886}
1887
Avantika Mathur09b88252006-12-06 20:41:36 -08001888static int
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001889ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
Alex Tomasa86c6182006-10-11 01:21:03 -07001890 struct ext4_extent *ex)
1891{
1892 struct ext4_ext_cache *cex;
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001893 int ret = EXT4_EXT_CACHE_NO;
Alex Tomasa86c6182006-10-11 01:21:03 -07001894
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001895 /*
1896 * We borrow i_block_reservation_lock to protect i_cached_extent
1897 */
1898 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Alex Tomasa86c6182006-10-11 01:21:03 -07001899 cex = &EXT4_I(inode)->i_cached_extent;
1900
1901 /* has cache valid data? */
1902 if (cex->ec_type == EXT4_EXT_CACHE_NO)
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001903 goto errout;
Alex Tomasa86c6182006-10-11 01:21:03 -07001904
1905 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1906 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1907 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001908 ex->ee_block = cpu_to_le32(cex->ec_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001909 ext4_ext_store_pblock(ex, cex->ec_start);
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001910 ex->ee_len = cpu_to_le16(cex->ec_len);
Eric Sandeenbba90742008-01-28 23:58:27 -05001911 ext_debug("%u cached by %u:%u:%llu\n",
1912 block,
1913 cex->ec_block, cex->ec_len, cex->ec_start);
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001914 ret = cex->ec_type;
Alex Tomasa86c6182006-10-11 01:21:03 -07001915 }
Theodore Ts'o2ec0ae32009-05-15 09:07:28 -04001916errout:
1917 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1918 return ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07001919}
1920
1921/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001922 * ext4_ext_rm_idx:
1923 * removes index from the index block.
1924 * It's used in truncate case only, thus all requests are for
1925 * last index in the block only.
Alex Tomasa86c6182006-10-11 01:21:03 -07001926 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001927static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001928 struct ext4_ext_path *path)
1929{
1930 struct buffer_head *bh;
1931 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001932 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07001933
1934 /* free index block */
1935 path--;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001936 leaf = idx_pblock(path->p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -07001937 BUG_ON(path->p_hdr->eh_entries == 0);
Avantika Mathur7e028972006-12-06 20:41:33 -08001938 err = ext4_ext_get_access(handle, inode, path);
1939 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001940 return err;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001941 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08001942 err = ext4_ext_dirty(handle, inode, path);
1943 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001944 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07001945 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Alex Tomasa86c6182006-10-11 01:21:03 -07001946 bh = sb_find_get_block(inode->i_sb, leaf);
1947 ext4_forget(handle, 1, inode, bh, leaf);
Alex Tomasc9de5602008-01-29 00:19:52 -05001948 ext4_free_blocks(handle, inode, leaf, 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001949 return err;
1950}
1951
1952/*
Mingming Caoee12b632008-08-19 22:16:05 -04001953 * ext4_ext_calc_credits_for_single_extent:
1954 * This routine returns max. credits that needed to insert an extent
1955 * to the extent tree.
1956 * When pass the actual path, the caller should calculate credits
1957 * under i_data_sem.
Alex Tomasa86c6182006-10-11 01:21:03 -07001958 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04001959int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
Alex Tomasa86c6182006-10-11 01:21:03 -07001960 struct ext4_ext_path *path)
1961{
Alex Tomasa86c6182006-10-11 01:21:03 -07001962 if (path) {
Mingming Caoee12b632008-08-19 22:16:05 -04001963 int depth = ext_depth(inode);
Mingming Caof3bd1f32008-08-19 22:16:03 -04001964 int ret = 0;
Mingming Caoee12b632008-08-19 22:16:05 -04001965
Alex Tomasa86c6182006-10-11 01:21:03 -07001966 /* probably there is space in leaf? */
Alex Tomasa86c6182006-10-11 01:21:03 -07001967 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
Mingming Caoee12b632008-08-19 22:16:05 -04001968 < le16_to_cpu(path[depth].p_hdr->eh_max)) {
1969
1970 /*
1971 * There are some space in the leaf tree, no
1972 * need to account for leaf block credit
1973 *
1974 * bitmaps and block group descriptor blocks
1975 * and other metadat blocks still need to be
1976 * accounted.
1977 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04001978 /* 1 bitmap, 1 block group descriptor */
Mingming Caoee12b632008-08-19 22:16:05 -04001979 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1980 }
Alex Tomasa86c6182006-10-11 01:21:03 -07001981 }
1982
Mingming Cao525f4ed2008-08-19 22:15:58 -04001983 return ext4_chunk_trans_blocks(inode, nrblocks);
Mingming Caoee12b632008-08-19 22:16:05 -04001984}
Alex Tomasa86c6182006-10-11 01:21:03 -07001985
Mingming Caoee12b632008-08-19 22:16:05 -04001986/*
1987 * How many index/leaf blocks need to change/allocate to modify nrblocks?
1988 *
1989 * if nrblocks are fit in a single extent (chunk flag is 1), then
1990 * in the worse case, each tree level index/leaf need to be changed
1991 * if the tree split due to insert a new extent, then the old tree
1992 * index/leaf need to be updated too
1993 *
1994 * If the nrblocks are discontiguous, they could cause
1995 * the whole tree split more than once, but this is really rare.
1996 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04001997int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoee12b632008-08-19 22:16:05 -04001998{
1999 int index;
2000 int depth = ext_depth(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002001
Mingming Caoee12b632008-08-19 22:16:05 -04002002 if (chunk)
2003 index = depth * 2;
2004 else
2005 index = depth * 3;
Alex Tomasa86c6182006-10-11 01:21:03 -07002006
Mingming Caoee12b632008-08-19 22:16:05 -04002007 return index;
Alex Tomasa86c6182006-10-11 01:21:03 -07002008}
2009
2010static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2011 struct ext4_extent *ex,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002012 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07002013{
2014 struct buffer_head *bh;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002015 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasc9de5602008-01-29 00:19:52 -05002016 int i, metadata = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002017
Alex Tomasc9de5602008-01-29 00:19:52 -05002018 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
2019 metadata = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07002020#ifdef EXTENTS_STATS
2021 {
2022 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002023 spin_lock(&sbi->s_ext_stats_lock);
2024 sbi->s_ext_blocks += ee_len;
2025 sbi->s_ext_extents++;
2026 if (ee_len < sbi->s_ext_min)
2027 sbi->s_ext_min = ee_len;
2028 if (ee_len > sbi->s_ext_max)
2029 sbi->s_ext_max = ee_len;
2030 if (ext_depth(inode) > sbi->s_depth_max)
2031 sbi->s_depth_max = ext_depth(inode);
2032 spin_unlock(&sbi->s_ext_stats_lock);
2033 }
2034#endif
2035 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002036 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002037 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002038 ext4_lblk_t num;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002039 ext4_fsblk_t start;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002040
Amit Aroraa2df2a62007-07-17 21:42:41 -04002041 num = le32_to_cpu(ex->ee_block) + ee_len - from;
2042 start = ext_pblock(ex) + ee_len - num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002043 ext_debug("free last %u blocks starting %llu\n", num, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002044 for (i = 0; i < num; i++) {
2045 bh = sb_find_get_block(inode->i_sb, start + i);
2046 ext4_forget(handle, 0, inode, bh, start + i);
2047 }
Alex Tomasc9de5602008-01-29 00:19:52 -05002048 ext4_free_blocks(handle, inode, start, num, metadata);
Alex Tomasa86c6182006-10-11 01:21:03 -07002049 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002050 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002051 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04002052 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002053 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002054 printk(KERN_INFO "strange request: removal(2) "
2055 "%u-%u from %u:%u\n",
2056 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07002057 }
2058 return 0;
2059}
2060
2061static int
2062ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002063 struct ext4_ext_path *path, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07002064{
2065 int err = 0, correct_index = 0;
2066 int depth = ext_depth(inode), credits;
2067 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002068 ext4_lblk_t a, b, block;
2069 unsigned num;
2070 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002071 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002072 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07002073 struct ext4_extent *ex;
2074
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002075 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002076 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002077 if (!path[depth].p_hdr)
2078 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2079 eh = path[depth].p_hdr;
2080 BUG_ON(eh == NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07002081
2082 /* find where to start removing */
2083 ex = EXT_LAST_EXTENT(eh);
2084
2085 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002086 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002087
2088 while (ex >= EXT_FIRST_EXTENT(eh) &&
2089 ex_ee_block + ex_ee_len > start) {
Aneesh Kumar K.Va41f2072009-06-10 14:22:55 -04002090
2091 if (ext4_ext_is_uninitialized(ex))
2092 uninitialized = 1;
2093 else
2094 uninitialized = 0;
2095
Alex Tomasa86c6182006-10-11 01:21:03 -07002096 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
2097 path[depth].p_ext = ex;
2098
2099 a = ex_ee_block > start ? ex_ee_block : start;
2100 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
2101 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
2102
2103 ext_debug(" border %u:%u\n", a, b);
2104
2105 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
2106 block = 0;
2107 num = 0;
2108 BUG();
2109 } else if (a != ex_ee_block) {
2110 /* remove tail of the extent */
2111 block = ex_ee_block;
2112 num = a - block;
2113 } else if (b != ex_ee_block + ex_ee_len - 1) {
2114 /* remove head of the extent */
2115 block = a;
2116 num = b - a;
2117 /* there is no "make a hole" API yet */
2118 BUG();
2119 } else {
2120 /* remove whole extent: excellent! */
2121 block = ex_ee_block;
2122 num = 0;
2123 BUG_ON(a != ex_ee_block);
2124 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
2125 }
2126
Theodore Ts'o34071da2008-08-01 21:59:19 -04002127 /*
2128 * 3 for leaf, sb, and inode plus 2 (bmap and group
2129 * descriptor) for each block group; assume two block
2130 * groups plus ex_ee_len/blocks_per_block_group for
2131 * the worst case
2132 */
2133 credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
Alex Tomasa86c6182006-10-11 01:21:03 -07002134 if (ex == EXT_FIRST_EXTENT(eh)) {
2135 correct_index = 1;
2136 credits += (ext_depth(inode)) + 1;
2137 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002138 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07002139
Shen Feng9102e4f2008-07-11 19:27:31 -04002140 err = ext4_ext_journal_restart(handle, credits);
2141 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07002142 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07002143
2144 err = ext4_ext_get_access(handle, inode, path + depth);
2145 if (err)
2146 goto out;
2147
2148 err = ext4_remove_blocks(handle, inode, ex, a, b);
2149 if (err)
2150 goto out;
2151
2152 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002153 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002154 ext4_ext_store_pblock(ex, 0);
Marcin Slusarze8546d02008-04-17 10:38:59 -04002155 le16_add_cpu(&eh->eh_entries, -1);
Alex Tomasa86c6182006-10-11 01:21:03 -07002156 }
2157
2158 ex->ee_block = cpu_to_le32(block);
2159 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04002160 /*
2161 * Do not mark uninitialized if all the blocks in the
2162 * extent have been removed.
2163 */
2164 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002165 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002166
2167 err = ext4_ext_dirty(handle, inode, path + depth);
2168 if (err)
2169 goto out;
2170
Mingming Cao2ae02102006-10-11 01:21:11 -07002171 ext_debug("new extent: %u:%u:%llu\n", block, num,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002172 ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07002173 ex--;
2174 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002175 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002176 }
2177
2178 if (correct_index && eh->eh_entries)
2179 err = ext4_ext_correct_indexes(handle, inode, path);
2180
2181 /* if this leaf is free, then we should
2182 * remove it from index block above */
2183 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2184 err = ext4_ext_rm_idx(handle, inode, path + depth);
2185
2186out:
2187 return err;
2188}
2189
2190/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002191 * ext4_ext_more_to_rm:
2192 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07002193 */
Avantika Mathur09b88252006-12-06 20:41:36 -08002194static int
Alex Tomasa86c6182006-10-11 01:21:03 -07002195ext4_ext_more_to_rm(struct ext4_ext_path *path)
2196{
2197 BUG_ON(path->p_idx == NULL);
2198
2199 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2200 return 0;
2201
2202 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002203 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07002204 * so we have to consider current index for truncation
2205 */
2206 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2207 return 0;
2208 return 1;
2209}
2210
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05002211static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07002212{
2213 struct super_block *sb = inode->i_sb;
2214 int depth = ext_depth(inode);
2215 struct ext4_ext_path *path;
2216 handle_t *handle;
2217 int i = 0, err = 0;
2218
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002219 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002220
2221 /* probably first extent we're gonna free will be last in block */
2222 handle = ext4_journal_start(inode, depth + 1);
2223 if (IS_ERR(handle))
2224 return PTR_ERR(handle);
2225
2226 ext4_ext_invalidate_cache(inode);
2227
2228 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002229 * We start scanning from right side, freeing all the blocks
2230 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002231 */
Josef Bacik216553c2008-04-29 22:02:02 -04002232 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07002233 if (path == NULL) {
2234 ext4_journal_stop(handle);
2235 return -ENOMEM;
2236 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002237 path[0].p_hdr = ext_inode_hdr(inode);
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04002238 if (ext4_ext_check(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002239 err = -EIO;
2240 goto out;
2241 }
2242 path[0].p_depth = depth;
2243
2244 while (i >= 0 && err == 0) {
2245 if (i == depth) {
2246 /* this is leaf block */
2247 err = ext4_ext_rm_leaf(handle, inode, path, start);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002248 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002249 brelse(path[i].p_bh);
2250 path[i].p_bh = NULL;
2251 i--;
2252 continue;
2253 }
2254
2255 /* this is index block */
2256 if (!path[i].p_hdr) {
2257 ext_debug("initialize header\n");
2258 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002259 }
2260
Alex Tomasa86c6182006-10-11 01:21:03 -07002261 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002262 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002263 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2264 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2265 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2266 path[i].p_hdr,
2267 le16_to_cpu(path[i].p_hdr->eh_entries));
2268 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002269 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002270 path[i].p_idx--;
2271 }
2272
2273 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2274 i, EXT_FIRST_INDEX(path[i].p_hdr),
2275 path[i].p_idx);
2276 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002277 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002278 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002279 ext_debug("move to level %d (block %llu)\n",
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002280 i + 1, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002281 memset(path + i + 1, 0, sizeof(*path));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002282 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2283 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002284 /* should we reset i_size? */
2285 err = -EIO;
2286 break;
2287 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002288 if (WARN_ON(i + 1 > depth)) {
2289 err = -EIO;
2290 break;
2291 }
Aneesh Kumar K.V56b19862009-03-12 09:51:20 -04002292 if (ext4_ext_check(inode, ext_block_hdr(bh),
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002293 depth - i - 1)) {
2294 err = -EIO;
2295 break;
2296 }
2297 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002298
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002299 /* save actual number of indexes since this
2300 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002301 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2302 i++;
2303 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002304 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002305 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002306 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002307 * handle must be already prepared by the
2308 * truncatei_leaf() */
2309 err = ext4_ext_rm_idx(handle, inode, path + i);
2310 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002311 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002312 brelse(path[i].p_bh);
2313 path[i].p_bh = NULL;
2314 i--;
2315 ext_debug("return to level %d\n", i);
2316 }
2317 }
2318
2319 /* TODO: flexible tree reduction should be here */
2320 if (path->p_hdr->eh_entries == 0) {
2321 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002322 * truncate to zero freed all the tree,
2323 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002324 */
2325 err = ext4_ext_get_access(handle, inode, path);
2326 if (err == 0) {
2327 ext_inode_hdr(inode)->eh_depth = 0;
2328 ext_inode_hdr(inode)->eh_max =
2329 cpu_to_le16(ext4_ext_space_root(inode));
2330 err = ext4_ext_dirty(handle, inode, path);
2331 }
2332 }
2333out:
Alex Tomasa86c6182006-10-11 01:21:03 -07002334 ext4_ext_drop_refs(path);
2335 kfree(path);
2336 ext4_journal_stop(handle);
2337
2338 return err;
2339}
2340
2341/*
2342 * called at mount time
2343 */
2344void ext4_ext_init(struct super_block *sb)
2345{
2346 /*
2347 * possible initialization would be here
2348 */
2349
Theodore Ts'o83982b62009-01-06 14:53:16 -05002350 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -04002351 printk(KERN_INFO "EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002352#ifdef AGGRESSIVE_TEST
2353 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07002354#endif
2355#ifdef CHECK_BINSEARCH
2356 printk(", check binsearch");
2357#endif
2358#ifdef EXTENTS_STATS
2359 printk(", stats");
2360#endif
2361 printk("\n");
2362#ifdef EXTENTS_STATS
2363 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2364 EXT4_SB(sb)->s_ext_min = 1 << 30;
2365 EXT4_SB(sb)->s_ext_max = 0;
2366#endif
2367 }
2368}
2369
2370/*
2371 * called at umount time
2372 */
2373void ext4_ext_release(struct super_block *sb)
2374{
Theodore Ts'o83982b62009-01-06 14:53:16 -05002375 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
Alex Tomasa86c6182006-10-11 01:21:03 -07002376 return;
2377
2378#ifdef EXTENTS_STATS
2379 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2380 struct ext4_sb_info *sbi = EXT4_SB(sb);
2381 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2382 sbi->s_ext_blocks, sbi->s_ext_extents,
2383 sbi->s_ext_blocks / sbi->s_ext_extents);
2384 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2385 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2386 }
2387#endif
2388}
2389
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002390static void bi_complete(struct bio *bio, int error)
2391{
2392 complete((struct completion *)bio->bi_private);
2393}
2394
2395/* FIXME!! we need to try to merge to left or right after zero-out */
2396static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2397{
2398 int ret = -EIO;
2399 struct bio *bio;
2400 int blkbits, blocksize;
2401 sector_t ee_pblock;
2402 struct completion event;
2403 unsigned int ee_len, len, done, offset;
2404
2405
2406 blkbits = inode->i_blkbits;
2407 blocksize = inode->i_sb->s_blocksize;
2408 ee_len = ext4_ext_get_actual_len(ex);
2409 ee_pblock = ext_pblock(ex);
2410
2411 /* convert ee_pblock to 512 byte sectors */
2412 ee_pblock = ee_pblock << (blkbits - 9);
2413
2414 while (ee_len > 0) {
2415
2416 if (ee_len > BIO_MAX_PAGES)
2417 len = BIO_MAX_PAGES;
2418 else
2419 len = ee_len;
2420
2421 bio = bio_alloc(GFP_NOIO, len);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002422 bio->bi_sector = ee_pblock;
2423 bio->bi_bdev = inode->i_sb->s_bdev;
2424
2425 done = 0;
2426 offset = 0;
2427 while (done < len) {
2428 ret = bio_add_page(bio, ZERO_PAGE(0),
2429 blocksize, offset);
2430 if (ret != blocksize) {
2431 /*
2432 * We can't add any more pages because of
2433 * hardware limitations. Start a new bio.
2434 */
2435 break;
2436 }
2437 done++;
2438 offset += blocksize;
2439 if (offset >= PAGE_CACHE_SIZE)
2440 offset = 0;
2441 }
2442
2443 init_completion(&event);
2444 bio->bi_private = &event;
2445 bio->bi_end_io = bi_complete;
2446 submit_bio(WRITE, bio);
2447 wait_for_completion(&event);
2448
2449 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2450 ret = 0;
2451 else {
2452 ret = -EIO;
2453 break;
2454 }
2455 bio_put(bio);
2456 ee_len -= done;
2457 ee_pblock += done << (blkbits - 9);
2458 }
2459 return ret;
2460}
2461
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002462#define EXT4_EXT_ZERO_LEN 7
2463
Amit Arora56055d32007-07-17 21:42:38 -04002464/*
2465 * This function is called by ext4_ext_get_blocks() if someone tries to write
2466 * to an uninitialized extent. It may result in splitting the uninitialized
2467 * extent into multiple extents (upto three - one initialized and two
2468 * uninitialized).
2469 * There are three possibilities:
2470 * a> There is no split required: Entire extent should be initialized
2471 * b> Splits in two extents: Write is happening at either end of the extent
2472 * c> Splits in three extents: Somone is writing in middle of the extent
2473 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002474static int ext4_ext_convert_to_initialized(handle_t *handle,
2475 struct inode *inode,
2476 struct ext4_ext_path *path,
2477 ext4_lblk_t iblock,
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002478 unsigned int max_blocks)
Amit Arora56055d32007-07-17 21:42:38 -04002479{
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002480 struct ext4_extent *ex, newex, orig_ex;
Amit Arora56055d32007-07-17 21:42:38 -04002481 struct ext4_extent *ex1 = NULL;
2482 struct ext4_extent *ex2 = NULL;
2483 struct ext4_extent *ex3 = NULL;
2484 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002485 ext4_lblk_t ee_block;
2486 unsigned int allocated, ee_len, depth;
Amit Arora56055d32007-07-17 21:42:38 -04002487 ext4_fsblk_t newblock;
2488 int err = 0;
2489 int ret = 0;
2490
2491 depth = ext_depth(inode);
2492 eh = path[depth].p_hdr;
2493 ex = path[depth].p_ext;
2494 ee_block = le32_to_cpu(ex->ee_block);
2495 ee_len = ext4_ext_get_actual_len(ex);
2496 allocated = ee_len - (iblock - ee_block);
2497 newblock = iblock - ee_block + ext_pblock(ex);
2498 ex2 = ex;
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002499 orig_ex.ee_block = ex->ee_block;
2500 orig_ex.ee_len = cpu_to_le16(ee_len);
2501 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
Amit Arora56055d32007-07-17 21:42:38 -04002502
Aneesh Kumar K.V9df56432008-02-22 06:17:31 -05002503 err = ext4_ext_get_access(handle, inode, path + depth);
2504 if (err)
2505 goto out;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002506 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2507 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2508 err = ext4_ext_zeroout(inode, &orig_ex);
2509 if (err)
2510 goto fix_extent_len;
2511 /* update the extent length and mark as initialized */
2512 ex->ee_block = orig_ex.ee_block;
2513 ex->ee_len = orig_ex.ee_len;
2514 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2515 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002516 /* zeroed the full extent */
2517 return allocated;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002518 }
Aneesh Kumar K.V9df56432008-02-22 06:17:31 -05002519
Amit Arora56055d32007-07-17 21:42:38 -04002520 /* ex1: ee_block to iblock - 1 : uninitialized */
2521 if (iblock > ee_block) {
2522 ex1 = ex;
2523 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2524 ext4_ext_mark_uninitialized(ex1);
2525 ex2 = &newex;
2526 }
2527 /*
2528 * for sanity, update the length of the ex2 extent before
2529 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2530 * overlap of blocks.
2531 */
2532 if (!ex1 && allocated > max_blocks)
2533 ex2->ee_len = cpu_to_le16(max_blocks);
2534 /* ex3: to ee_block + ee_len : uninitialised */
2535 if (allocated > max_blocks) {
2536 unsigned int newdepth;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002537 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2538 if (allocated <= EXT4_EXT_ZERO_LEN) {
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002539 /*
2540 * iblock == ee_block is handled by the zerouout
2541 * at the beginning.
2542 * Mark first half uninitialized.
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002543 * Mark second half initialized and zero out the
2544 * initialized extent
2545 */
2546 ex->ee_block = orig_ex.ee_block;
2547 ex->ee_len = cpu_to_le16(ee_len - allocated);
2548 ext4_ext_mark_uninitialized(ex);
2549 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2550 ext4_ext_dirty(handle, inode, path + depth);
2551
2552 ex3 = &newex;
2553 ex3->ee_block = cpu_to_le32(iblock);
2554 ext4_ext_store_pblock(ex3, newblock);
2555 ex3->ee_len = cpu_to_le16(allocated);
2556 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2557 if (err == -ENOSPC) {
2558 err = ext4_ext_zeroout(inode, &orig_ex);
2559 if (err)
2560 goto fix_extent_len;
2561 ex->ee_block = orig_ex.ee_block;
2562 ex->ee_len = orig_ex.ee_len;
2563 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2564 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002565 /* blocks available from iblock */
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002566 return allocated;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002567
2568 } else if (err)
2569 goto fix_extent_len;
2570
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002571 /*
2572 * We need to zero out the second half because
2573 * an fallocate request can update file size and
2574 * converting the second half to initialized extent
2575 * implies that we can leak some junk data to user
2576 * space.
2577 */
2578 err = ext4_ext_zeroout(inode, ex3);
2579 if (err) {
2580 /*
2581 * We should actually mark the
2582 * second half as uninit and return error
2583 * Insert would have changed the extent
2584 */
2585 depth = ext_depth(inode);
2586 ext4_ext_drop_refs(path);
2587 path = ext4_ext_find_extent(inode,
2588 iblock, path);
2589 if (IS_ERR(path)) {
2590 err = PTR_ERR(path);
2591 return err;
2592 }
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002593 /* get the second half extent details */
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002594 ex = path[depth].p_ext;
2595 err = ext4_ext_get_access(handle, inode,
2596 path + depth);
2597 if (err)
2598 return err;
2599 ext4_ext_mark_uninitialized(ex);
2600 ext4_ext_dirty(handle, inode, path + depth);
2601 return err;
2602 }
2603
2604 /* zeroed the second half */
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002605 return allocated;
2606 }
Amit Arora56055d32007-07-17 21:42:38 -04002607 ex3 = &newex;
2608 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2609 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2610 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2611 ext4_ext_mark_uninitialized(ex3);
2612 err = ext4_ext_insert_extent(handle, inode, path, ex3);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002613 if (err == -ENOSPC) {
2614 err = ext4_ext_zeroout(inode, &orig_ex);
2615 if (err)
2616 goto fix_extent_len;
2617 /* update the extent length and mark as initialized */
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002618 ex->ee_block = orig_ex.ee_block;
2619 ex->ee_len = orig_ex.ee_len;
2620 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002621 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002622 /* zeroed the full extent */
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002623 /* blocks available from iblock */
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002624 return allocated;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002625
2626 } else if (err)
2627 goto fix_extent_len;
Amit Arora56055d32007-07-17 21:42:38 -04002628 /*
2629 * The depth, and hence eh & ex might change
2630 * as part of the insert above.
2631 */
2632 newdepth = ext_depth(inode);
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002633 /*
Coly Li73ac36e2009-01-07 18:09:16 -08002634 * update the extent length after successful insert of the
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002635 * split extent
2636 */
2637 orig_ex.ee_len = cpu_to_le16(ee_len -
2638 ext4_ext_get_actual_len(ex3));
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002639 depth = newdepth;
2640 ext4_ext_drop_refs(path);
2641 path = ext4_ext_find_extent(inode, iblock, path);
2642 if (IS_ERR(path)) {
2643 err = PTR_ERR(path);
2644 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04002645 }
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002646 eh = path[depth].p_hdr;
2647 ex = path[depth].p_ext;
2648 if (ex2 != &newex)
2649 ex2 = ex;
2650
2651 err = ext4_ext_get_access(handle, inode, path + depth);
2652 if (err)
2653 goto out;
2654
Amit Arora56055d32007-07-17 21:42:38 -04002655 allocated = max_blocks;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002656
2657 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2658 * to insert a extent in the middle zerout directly
2659 * otherwise give the extent a chance to merge to left
2660 */
2661 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2662 iblock != ee_block) {
2663 err = ext4_ext_zeroout(inode, &orig_ex);
2664 if (err)
2665 goto fix_extent_len;
2666 /* update the extent length and mark as initialized */
2667 ex->ee_block = orig_ex.ee_block;
2668 ex->ee_len = orig_ex.ee_len;
2669 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2670 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002671 /* zero out the first half */
Aneesh Kumar K.Vd03856b2008-08-02 18:51:32 -04002672 /* blocks available from iblock */
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002673 return allocated;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002674 }
Amit Arora56055d32007-07-17 21:42:38 -04002675 }
2676 /*
2677 * If there was a change of depth as part of the
2678 * insertion of ex3 above, we need to update the length
2679 * of the ex1 extent again here
2680 */
2681 if (ex1 && ex1 != ex) {
2682 ex1 = ex;
2683 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2684 ext4_ext_mark_uninitialized(ex1);
2685 ex2 = &newex;
2686 }
2687 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2688 ex2->ee_block = cpu_to_le32(iblock);
Amit Arora56055d32007-07-17 21:42:38 -04002689 ext4_ext_store_pblock(ex2, newblock);
2690 ex2->ee_len = cpu_to_le16(allocated);
2691 if (ex2 != ex)
2692 goto insert;
Amit Arora56055d32007-07-17 21:42:38 -04002693 /*
2694 * New (initialized) extent starts from the first block
2695 * in the current extent. i.e., ex2 == ex
2696 * We have to see if it can be merged with the extent
2697 * on the left.
2698 */
2699 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2700 /*
2701 * To merge left, pass "ex2 - 1" to try_to_merge(),
2702 * since it merges towards right _only_.
2703 */
2704 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2705 if (ret) {
2706 err = ext4_ext_correct_indexes(handle, inode, path);
2707 if (err)
2708 goto out;
2709 depth = ext_depth(inode);
2710 ex2--;
2711 }
2712 }
2713 /*
2714 * Try to Merge towards right. This might be required
2715 * only when the whole extent is being written to.
2716 * i.e. ex2 == ex and ex3 == NULL.
2717 */
2718 if (!ex3) {
2719 ret = ext4_ext_try_to_merge(inode, path, ex2);
2720 if (ret) {
2721 err = ext4_ext_correct_indexes(handle, inode, path);
2722 if (err)
2723 goto out;
2724 }
2725 }
2726 /* Mark modified extent as dirty */
2727 err = ext4_ext_dirty(handle, inode, path + depth);
2728 goto out;
2729insert:
2730 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002731 if (err == -ENOSPC) {
2732 err = ext4_ext_zeroout(inode, &orig_ex);
2733 if (err)
2734 goto fix_extent_len;
2735 /* update the extent length and mark as initialized */
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002736 ex->ee_block = orig_ex.ee_block;
2737 ex->ee_len = orig_ex.ee_len;
2738 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002739 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002740 /* zero out the first half */
2741 return allocated;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002742 } else if (err)
2743 goto fix_extent_len;
Amit Arora56055d32007-07-17 21:42:38 -04002744out:
2745 return err ? err : allocated;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002746
2747fix_extent_len:
2748 ex->ee_block = orig_ex.ee_block;
2749 ex->ee_len = orig_ex.ee_len;
2750 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2751 ext4_ext_mark_uninitialized(ex);
2752 ext4_ext_dirty(handle, inode, path + depth);
2753 return err;
Amit Arora56055d32007-07-17 21:42:38 -04002754}
2755
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002756/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05002757 * Block allocation/map/preallocation routine for extents based files
2758 *
2759 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002760 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002761 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2762 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05002763 *
2764 * return > 0, number of of blocks already mapped/allocated
2765 * if create == 0 and these are pre-allocated blocks
2766 * buffer head is unmapped
2767 * otherwise blocks are mapped
2768 *
2769 * return = 0, if plain look up failed (blocks have not been allocated)
2770 * buffer head is unmapped
2771 *
2772 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002773 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002774int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002775 ext4_lblk_t iblock,
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002776 unsigned int max_blocks, struct buffer_head *bh_result,
Theodore Ts'oc2177052009-05-14 00:58:52 -04002777 int flags)
Alex Tomasa86c6182006-10-11 01:21:03 -07002778{
2779 struct ext4_ext_path *path = NULL;
Amit Arora56055d32007-07-17 21:42:38 -04002780 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002781 struct ext4_extent newex, *ex;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002782 ext4_fsblk_t newblock;
2783 int err = 0, depth, ret, cache_type;
2784 unsigned int allocated = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05002785 struct ext4_allocation_request ar;
Alex Tomasa86c6182006-10-11 01:21:03 -07002786
2787 __clear_bit(BH_New, &bh_result->b_state);
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002788 ext_debug("blocks %u/%u requested for inode %u\n",
Eric Sandeenbba90742008-01-28 23:58:27 -05002789 iblock, max_blocks, inode->i_ino);
Alex Tomasa86c6182006-10-11 01:21:03 -07002790
2791 /* check in cache */
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002792 cache_type = ext4_ext_in_cache(inode, iblock, &newex);
2793 if (cache_type) {
2794 if (cache_type == EXT4_EXT_CACHE_GAP) {
Theodore Ts'oc2177052009-05-14 00:58:52 -04002795 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04002796 /*
2797 * block isn't allocated yet and
2798 * user doesn't want to allocate it
2799 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002800 goto out2;
2801 }
2802 /* we should allocate requested block */
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002803 } else if (cache_type == EXT4_EXT_CACHE_EXTENT) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002804 /* block is already allocated */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002805 newblock = iblock
2806 - le32_to_cpu(newex.ee_block)
2807 + ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002808 /* number of remaining blocks in the extent */
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002809 allocated = ext4_ext_get_actual_len(&newex) -
Alex Tomasa86c6182006-10-11 01:21:03 -07002810 (iblock - le32_to_cpu(newex.ee_block));
2811 goto out;
2812 } else {
2813 BUG();
2814 }
2815 }
2816
2817 /* find extent for this block */
2818 path = ext4_ext_find_extent(inode, iblock, NULL);
2819 if (IS_ERR(path)) {
2820 err = PTR_ERR(path);
2821 path = NULL;
2822 goto out2;
2823 }
2824
2825 depth = ext_depth(inode);
2826
2827 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002828 * consistent leaf must not be empty;
2829 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07002830 * this is why assert can't be put in ext4_ext_find_extent()
2831 */
2832 BUG_ON(path[depth].p_ext == NULL && depth != 0);
Amit Arora56055d32007-07-17 21:42:38 -04002833 eh = path[depth].p_hdr;
Alex Tomasa86c6182006-10-11 01:21:03 -07002834
Avantika Mathur7e028972006-12-06 20:41:33 -08002835 ex = path[depth].p_ext;
2836 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002837 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002838 ext4_fsblk_t ee_start = ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002839 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002840
2841 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002842 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04002843 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002844 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002845 ee_len = ext4_ext_get_actual_len(ex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002846 /* if found extent covers block, simply return it */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002847 if (iblock >= ee_block && iblock < ee_block + ee_len) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002848 newblock = iblock - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002849 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002850 allocated = ee_len - (iblock - ee_block);
Eric Sandeenbba90742008-01-28 23:58:27 -05002851 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002852 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04002853
Amit Aroraa2df2a62007-07-17 21:42:41 -04002854 /* Do not put uninitialized extent in the cache */
Amit Arora56055d32007-07-17 21:42:38 -04002855 if (!ext4_ext_is_uninitialized(ex)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002856 ext4_ext_put_in_cache(inode, ee_block,
2857 ee_len, ee_start,
2858 EXT4_EXT_CACHE_EXTENT);
Amit Arora56055d32007-07-17 21:42:38 -04002859 goto out;
2860 }
Theodore Ts'oc2177052009-05-14 00:58:52 -04002861 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)
Amit Arora56055d32007-07-17 21:42:38 -04002862 goto out;
Theodore Ts'oc2177052009-05-14 00:58:52 -04002863 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002864 if (allocated > max_blocks)
2865 allocated = max_blocks;
Aneesh Kumar K.Ve067ba02008-04-29 08:11:12 -04002866 /*
2867 * We have blocks reserved already. We
2868 * return allocated blocks so that delalloc
2869 * won't do block reservation for us. But
2870 * the buffer head will be unmapped so that
2871 * a read from the block returns 0s.
2872 */
Eric Sandeen953e6222008-07-11 19:27:31 -04002873 set_buffer_unwritten(bh_result);
Aneesh Kumar K.V9c1ee182009-05-13 18:36:58 -04002874 bh_result->b_bdev = inode->i_sb->s_bdev;
2875 bh_result->b_blocknr = newblock;
Amit Arora56055d32007-07-17 21:42:38 -04002876 goto out2;
Aneesh Kumar K.Ve067ba02008-04-29 08:11:12 -04002877 }
Amit Arora56055d32007-07-17 21:42:38 -04002878
2879 ret = ext4_ext_convert_to_initialized(handle, inode,
2880 path, iblock,
2881 max_blocks);
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002882 if (ret <= 0) {
2883 err = ret;
Amit Arora56055d32007-07-17 21:42:38 -04002884 goto out2;
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002885 } else
Amit Arora56055d32007-07-17 21:42:38 -04002886 allocated = ret;
2887 goto outnew;
Alex Tomasa86c6182006-10-11 01:21:03 -07002888 }
2889 }
2890
2891 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002892 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07002893 * we couldn't try to create block if create flag is zero
2894 */
Theodore Ts'oc2177052009-05-14 00:58:52 -04002895 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
Amit Arora56055d32007-07-17 21:42:38 -04002896 /*
2897 * put just found gap into cache to speed up
2898 * subsequent requests
2899 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002900 ext4_ext_put_gap_in_cache(inode, path, iblock);
2901 goto out2;
2902 }
2903 /*
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04002904 * Okay, we need to do block allocation.
Andrew Morton63f57932006-10-11 01:21:24 -07002905 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002906
Alex Tomasc9de5602008-01-29 00:19:52 -05002907 /* find neighbour allocated blocks */
2908 ar.lleft = iblock;
2909 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2910 if (err)
2911 goto out2;
2912 ar.lright = iblock;
2913 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2914 if (err)
2915 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04002916
Amit Arora749269f2007-07-18 09:02:56 -04002917 /*
2918 * See if request is beyond maximum number of blocks we can have in
2919 * a single extent. For an initialized extent this limit is
2920 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2921 * EXT_UNINIT_MAX_LEN.
2922 */
2923 if (max_blocks > EXT_INIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04002924 !(flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Amit Arora749269f2007-07-18 09:02:56 -04002925 max_blocks = EXT_INIT_MAX_LEN;
2926 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
Theodore Ts'oc2177052009-05-14 00:58:52 -04002927 (flags & EXT4_GET_BLOCKS_UNINIT_EXT))
Amit Arora749269f2007-07-18 09:02:56 -04002928 max_blocks = EXT_UNINIT_MAX_LEN;
2929
Amit Arora25d14f92007-05-24 13:04:13 -04002930 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2931 newex.ee_block = cpu_to_le32(iblock);
2932 newex.ee_len = cpu_to_le16(max_blocks);
2933 err = ext4_ext_check_overlap(inode, &newex, path);
2934 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002935 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04002936 else
2937 allocated = max_blocks;
Alex Tomasc9de5602008-01-29 00:19:52 -05002938
2939 /* allocate new block */
2940 ar.inode = inode;
2941 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2942 ar.logical = iblock;
2943 ar.len = allocated;
2944 if (S_ISREG(inode->i_mode))
2945 ar.flags = EXT4_MB_HINT_DATA;
2946 else
2947 /* disable in-core preallocation for non-regular files */
2948 ar.flags = 0;
2949 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07002950 if (!newblock)
2951 goto out2;
Mingming Cao2ae02102006-10-11 01:21:11 -07002952 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002953 ar.goal, newblock, allocated);
Alex Tomasa86c6182006-10-11 01:21:03 -07002954
2955 /* try to insert new extent into found leaf and return */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002956 ext4_ext_store_pblock(&newex, newblock);
Alex Tomasc9de5602008-01-29 00:19:52 -05002957 newex.ee_len = cpu_to_le16(ar.len);
Theodore Ts'oc2177052009-05-14 00:58:52 -04002958 if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) /* Mark uninitialized */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002959 ext4_ext_mark_uninitialized(&newex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002960 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Alex Tomas315054f2007-05-24 13:04:25 -04002961 if (err) {
2962 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05002963 /* not a good idea to call discard here directly,
2964 * but otherwise we'd need to call it every free() */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04002965 ext4_discard_preallocations(inode);
Alex Tomas315054f2007-05-24 13:04:25 -04002966 ext4_free_blocks(handle, inode, ext_pblock(&newex),
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002967 ext4_ext_get_actual_len(&newex), 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002968 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04002969 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002970
Alex Tomasa86c6182006-10-11 01:21:03 -07002971 /* previous routine could use block we allocated */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002972 newblock = ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002973 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora56055d32007-07-17 21:42:38 -04002974outnew:
Eric Sandeen953e6222008-07-11 19:27:31 -04002975 set_buffer_new(bh_result);
Alex Tomasa86c6182006-10-11 01:21:03 -07002976
Amit Aroraa2df2a62007-07-17 21:42:41 -04002977 /* Cache only when it is _not_ an uninitialized extent */
Theodore Ts'oc2177052009-05-14 00:58:52 -04002978 if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0)
Amit Aroraa2df2a62007-07-17 21:42:41 -04002979 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2980 EXT4_EXT_CACHE_EXTENT);
Alex Tomasa86c6182006-10-11 01:21:03 -07002981out:
2982 if (allocated > max_blocks)
2983 allocated = max_blocks;
2984 ext4_ext_show_leaf(inode, path);
Eric Sandeen953e6222008-07-11 19:27:31 -04002985 set_buffer_mapped(bh_result);
Alex Tomasa86c6182006-10-11 01:21:03 -07002986 bh_result->b_bdev = inode->i_sb->s_bdev;
2987 bh_result->b_blocknr = newblock;
2988out2:
2989 if (path) {
2990 ext4_ext_drop_refs(path);
2991 kfree(path);
2992 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002993 return err ? err : allocated;
2994}
2995
Jan Karacf108bc2008-07-11 19:27:31 -04002996void ext4_ext_truncate(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07002997{
2998 struct address_space *mapping = inode->i_mapping;
2999 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003000 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07003001 handle_t *handle;
3002 int err = 0;
3003
3004 /*
3005 * probably first extent we're gonna free will be last in block
3006 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04003007 err = ext4_writepage_trans_blocks(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07003008 handle = ext4_journal_start(inode, err);
Jan Karacf108bc2008-07-11 19:27:31 -04003009 if (IS_ERR(handle))
Alex Tomasa86c6182006-10-11 01:21:03 -07003010 return;
Alex Tomasa86c6182006-10-11 01:21:03 -07003011
Jan Karacf108bc2008-07-11 19:27:31 -04003012 if (inode->i_size & (sb->s_blocksize - 1))
3013 ext4_block_truncate_page(handle, mapping, inode->i_size);
Alex Tomasa86c6182006-10-11 01:21:03 -07003014
Jan Kara9ddfc3d2008-07-11 19:27:31 -04003015 if (ext4_orphan_add(handle, inode))
3016 goto out_stop;
3017
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05003018 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07003019 ext4_ext_invalidate_cache(inode);
3020
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04003021 ext4_discard_preallocations(inode);
Alex Tomasc9de5602008-01-29 00:19:52 -05003022
Alex Tomasa86c6182006-10-11 01:21:03 -07003023 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003024 * TODO: optimization is possible here.
3025 * Probably we need not scan at all,
3026 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07003027 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003028
3029 /* we have to know where to truncate from in crash case */
3030 EXT4_I(inode)->i_disksize = inode->i_size;
3031 ext4_mark_inode_dirty(handle, inode);
3032
3033 last_block = (inode->i_size + sb->s_blocksize - 1)
3034 >> EXT4_BLOCK_SIZE_BITS(sb);
3035 err = ext4_ext_remove_space(inode, last_block);
3036
3037 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04003038 * transaction synchronous.
3039 */
Alex Tomasa86c6182006-10-11 01:21:03 -07003040 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -05003041 ext4_handle_sync(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07003042
3043out_stop:
Jan Kara9ddfc3d2008-07-11 19:27:31 -04003044 up_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07003045 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07003046 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07003047 * then we need to clear up the orphan record which we created above.
3048 * However, if this was a real unlink then we were called by
3049 * ext4_delete_inode(), and we allow that function to clean up the
3050 * orphan info for us.
3051 */
3052 if (inode->i_nlink)
3053 ext4_orphan_del(handle, inode);
3054
Solofo Ramangalahyef737722008-04-29 22:00:41 -04003055 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3056 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07003057 ext4_journal_stop(handle);
3058}
3059
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003060static void ext4_falloc_update_inode(struct inode *inode,
3061 int mode, loff_t new_size, int update_ctime)
3062{
3063 struct timespec now;
3064
3065 if (update_ctime) {
3066 now = current_fs_time(inode->i_sb);
3067 if (!timespec_equal(&inode->i_ctime, &now))
3068 inode->i_ctime = now;
3069 }
3070 /*
3071 * Update only when preallocation was requested beyond
3072 * the file size.
3073 */
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04003074 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3075 if (new_size > i_size_read(inode))
3076 i_size_write(inode, new_size);
3077 if (new_size > EXT4_I(inode)->i_disksize)
3078 ext4_update_i_disksize(inode, new_size);
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003079 }
3080
3081}
3082
Amit Aroraa2df2a62007-07-17 21:42:41 -04003083/*
3084 * preallocate space for a file. This implements ext4's fallocate inode
3085 * operation, which gets called from sys_fallocate system call.
3086 * For block-mapped files, posix_fallocate should fall back to the method
3087 * of writing zeroes to the required new blocks (the same behavior which is
3088 * expected for file systems which do not support fallocate() system call).
3089 */
3090long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
3091{
3092 handle_t *handle;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05003093 ext4_lblk_t block;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003094 loff_t new_size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05003095 unsigned int max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003096 int ret = 0;
3097 int ret2 = 0;
3098 int retries = 0;
3099 struct buffer_head map_bh;
3100 unsigned int credits, blkbits = inode->i_blkbits;
3101
3102 /*
3103 * currently supporting (pre)allocate mode for extent-based
3104 * files _only_
3105 */
3106 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3107 return -EOPNOTSUPP;
3108
3109 /* preallocation to directories is currently not supported */
3110 if (S_ISDIR(inode->i_mode))
3111 return -ENODEV;
3112
3113 block = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003114 /*
3115 * We can't just convert len to max_blocks because
3116 * If blocksize = 4096 offset = 3072 and len = 2048
3117 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04003118 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003119 - block;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003120 /*
Mingming Caof3bd1f32008-08-19 22:16:03 -04003121 * credits to insert 1 extent into extent tree
Amit Aroraa2df2a62007-07-17 21:42:41 -04003122 */
Mingming Caof3bd1f32008-08-19 22:16:03 -04003123 credits = ext4_chunk_trans_blocks(inode, max_blocks);
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05003124 mutex_lock(&inode->i_mutex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04003125retry:
3126 while (ret >= 0 && ret < max_blocks) {
3127 block = block + ret;
3128 max_blocks = max_blocks - ret;
3129 handle = ext4_journal_start(inode, credits);
3130 if (IS_ERR(handle)) {
3131 ret = PTR_ERR(handle);
3132 break;
3133 }
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04003134 map_bh.b_state = 0;
Theodore Ts'o12b7ac12009-05-14 00:57:44 -04003135 ret = ext4_get_blocks(handle, inode, block,
3136 max_blocks, &map_bh,
Theodore Ts'oc2177052009-05-14 00:58:52 -04003137 EXT4_GET_BLOCKS_CREATE_UNINIT_EXT);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05003138 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05003139#ifdef EXT4FS_DEBUG
3140 WARN_ON(ret <= 0);
3141 printk(KERN_ERR "%s: ext4_ext_get_blocks "
3142 "returned error inode#%lu, block=%u, "
Thadeu Lima de Souza Cascardo9fd97842009-01-26 19:26:26 -05003143 "max_blocks=%u", __func__,
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05003144 inode->i_ino, block, max_blocks);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05003145#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04003146 ext4_mark_inode_dirty(handle, inode);
3147 ret2 = ext4_journal_stop(handle);
3148 break;
3149 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003150 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
3151 blkbits) >> blkbits))
3152 new_size = offset + len;
3153 else
3154 new_size = (block + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003155
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003156 ext4_falloc_update_inode(inode, mode, new_size,
3157 buffer_new(&map_bh));
Amit Aroraa2df2a62007-07-17 21:42:41 -04003158 ext4_mark_inode_dirty(handle, inode);
3159 ret2 = ext4_journal_stop(handle);
3160 if (ret2)
3161 break;
3162 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04003163 if (ret == -ENOSPC &&
3164 ext4_should_retry_alloc(inode->i_sb, &retries)) {
3165 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003166 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04003167 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05003168 mutex_unlock(&inode->i_mutex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04003169 return ret > 0 ? ret2 : ret;
3170}
Eric Sandeen6873fa02008-10-07 00:46:36 -04003171
3172/*
3173 * Callback function called for each extent to gather FIEMAP information.
3174 */
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003175static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path,
Eric Sandeen6873fa02008-10-07 00:46:36 -04003176 struct ext4_ext_cache *newex, struct ext4_extent *ex,
3177 void *data)
3178{
3179 struct fiemap_extent_info *fieinfo = data;
Eric Sandeenc9877b22009-05-01 23:32:06 -04003180 unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
Eric Sandeen6873fa02008-10-07 00:46:36 -04003181 __u64 logical;
3182 __u64 physical;
3183 __u64 length;
3184 __u32 flags = 0;
3185 int error;
3186
3187 logical = (__u64)newex->ec_block << blksize_bits;
3188
3189 if (newex->ec_type == EXT4_EXT_CACHE_GAP) {
3190 pgoff_t offset;
3191 struct page *page;
3192 struct buffer_head *bh = NULL;
3193
3194 offset = logical >> PAGE_SHIFT;
3195 page = find_get_page(inode->i_mapping, offset);
3196 if (!page || !page_has_buffers(page))
3197 return EXT_CONTINUE;
3198
3199 bh = page_buffers(page);
3200
3201 if (!bh)
3202 return EXT_CONTINUE;
3203
3204 if (buffer_delay(bh)) {
3205 flags |= FIEMAP_EXTENT_DELALLOC;
3206 page_cache_release(page);
3207 } else {
3208 page_cache_release(page);
3209 return EXT_CONTINUE;
3210 }
3211 }
3212
3213 physical = (__u64)newex->ec_start << blksize_bits;
3214 length = (__u64)newex->ec_len << blksize_bits;
3215
3216 if (ex && ext4_ext_is_uninitialized(ex))
3217 flags |= FIEMAP_EXTENT_UNWRITTEN;
3218
3219 /*
3220 * If this extent reaches EXT_MAX_BLOCK, it must be last.
3221 *
3222 * Or if ext4_ext_next_allocated_block is EXT_MAX_BLOCK,
3223 * this also indicates no more allocated blocks.
3224 *
3225 * XXX this might miss a single-block extent at EXT_MAX_BLOCK
3226 */
Eric Sandeenc9877b22009-05-01 23:32:06 -04003227 if (ext4_ext_next_allocated_block(path) == EXT_MAX_BLOCK ||
Theodore Ts'oeefd7f02009-05-02 19:05:37 -04003228 newex->ec_block + newex->ec_len - 1 == EXT_MAX_BLOCK) {
3229 loff_t size = i_size_read(inode);
3230 loff_t bs = EXT4_BLOCK_SIZE(inode->i_sb);
3231
Eric Sandeen6873fa02008-10-07 00:46:36 -04003232 flags |= FIEMAP_EXTENT_LAST;
Theodore Ts'oeefd7f02009-05-02 19:05:37 -04003233 if ((flags & FIEMAP_EXTENT_DELALLOC) &&
3234 logical+length > size)
3235 length = (size - logical + bs - 1) & ~(bs-1);
3236 }
Eric Sandeen6873fa02008-10-07 00:46:36 -04003237
3238 error = fiemap_fill_next_extent(fieinfo, logical, physical,
3239 length, flags);
3240 if (error < 0)
3241 return error;
3242 if (error == 1)
3243 return EXT_BREAK;
3244
3245 return EXT_CONTINUE;
3246}
3247
3248/* fiemap flags we can handle specified here */
3249#define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
3250
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003251static int ext4_xattr_fiemap(struct inode *inode,
3252 struct fiemap_extent_info *fieinfo)
Eric Sandeen6873fa02008-10-07 00:46:36 -04003253{
3254 __u64 physical = 0;
3255 __u64 length;
3256 __u32 flags = FIEMAP_EXTENT_LAST;
3257 int blockbits = inode->i_sb->s_blocksize_bits;
3258 int error = 0;
3259
3260 /* in-inode? */
3261 if (EXT4_I(inode)->i_state & EXT4_STATE_XATTR) {
3262 struct ext4_iloc iloc;
3263 int offset; /* offset of xattr in inode */
3264
3265 error = ext4_get_inode_loc(inode, &iloc);
3266 if (error)
3267 return error;
3268 physical = iloc.bh->b_blocknr << blockbits;
3269 offset = EXT4_GOOD_OLD_INODE_SIZE +
3270 EXT4_I(inode)->i_extra_isize;
3271 physical += offset;
3272 length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
3273 flags |= FIEMAP_EXTENT_DATA_INLINE;
3274 } else { /* external block */
3275 physical = EXT4_I(inode)->i_file_acl << blockbits;
3276 length = inode->i_sb->s_blocksize;
3277 }
3278
3279 if (physical)
3280 error = fiemap_fill_next_extent(fieinfo, 0, physical,
3281 length, flags);
3282 return (error < 0 ? error : 0);
3283}
3284
3285int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
3286 __u64 start, __u64 len)
3287{
3288 ext4_lblk_t start_blk;
3289 ext4_lblk_t len_blks;
3290 int error = 0;
3291
3292 /* fallback to generic here if not in extents fmt */
3293 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
3294 return generic_block_fiemap(inode, fieinfo, start, len,
3295 ext4_get_block);
3296
3297 if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS))
3298 return -EBADR;
3299
3300 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
3301 error = ext4_xattr_fiemap(inode, fieinfo);
3302 } else {
3303 start_blk = start >> inode->i_sb->s_blocksize_bits;
3304 len_blks = len >> inode->i_sb->s_blocksize_bits;
3305
3306 /*
3307 * Walk the extent tree gathering extent information.
3308 * ext4_ext_fiemap_cb will push extents back to user.
3309 */
Theodore Ts'o0568c512009-05-17 23:31:23 -04003310 down_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04003311 error = ext4_ext_walk_space(inode, start_blk, len_blks,
3312 ext4_ext_fiemap_cb, fieinfo);
Theodore Ts'o0568c512009-05-17 23:31:23 -04003313 up_read(&EXT4_I(inode)->i_data_sem);
Eric Sandeen6873fa02008-10-07 00:46:36 -04003314 }
3315
3316 return error;
3317}
3318