]>
Commit | Line | Data |
---|---|---|
a86c6181 AT |
1 | /* |
2 | * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com | |
3 | * Written by Alex Tomas <alex@clusterfs.com> | |
4 | * | |
5 | * Architecture independence: | |
6 | * Copyright (c) 2005, Bull S.A. | |
7 | * Written by Pierre Peiffer <pierre.peiffer@bull.net> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public Licens | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Extents support for EXT4 | |
25 | * | |
26 | * TODO: | |
27 | * - ext4*_error() should be used in some situations | |
28 | * - analyze all BUG()/BUG_ON(), use -EIO where appropriate | |
29 | * - smart tree reduction | |
30 | */ | |
31 | ||
a86c6181 AT |
32 | #include <linux/fs.h> |
33 | #include <linux/time.h> | |
cd02ff0b | 34 | #include <linux/jbd2.h> |
a86c6181 AT |
35 | #include <linux/highuid.h> |
36 | #include <linux/pagemap.h> | |
37 | #include <linux/quotaops.h> | |
38 | #include <linux/string.h> | |
39 | #include <linux/slab.h> | |
a2df2a63 | 40 | #include <linux/falloc.h> |
a86c6181 | 41 | #include <asm/uaccess.h> |
6873fa0d | 42 | #include <linux/fiemap.h> |
3dcf5451 | 43 | #include "ext4_jbd2.h" |
4a092d73 | 44 | #include "ext4_extents.h" |
f19d5870 | 45 | #include "xattr.h" |
a86c6181 | 46 | |
0562e0ba JZ |
47 | #include <trace/events/ext4.h> |
48 | ||
5f95d21f LC |
49 | /* |
50 | * used by extent splitting. | |
51 | */ | |
52 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
53 | due to ENOSPC */ | |
54 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
55 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
56 | ||
dee1f973 DM |
57 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
58 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
59 | ||
7ac5990d DW |
60 | static __le32 ext4_extent_block_csum(struct inode *inode, |
61 | struct ext4_extent_header *eh) | |
62 | { | |
63 | struct ext4_inode_info *ei = EXT4_I(inode); | |
64 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
65 | __u32 csum; | |
66 | ||
67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
68 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
69 | return cpu_to_le32(csum); | |
70 | } | |
71 | ||
72 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
73 | struct ext4_extent_header *eh) | |
74 | { | |
75 | struct ext4_extent_tail *et; | |
76 | ||
77 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
78 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
79 | return 1; | |
80 | ||
81 | et = find_ext4_extent_tail(eh); | |
82 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
83 | return 0; | |
84 | return 1; | |
85 | } | |
86 | ||
87 | static void ext4_extent_block_csum_set(struct inode *inode, | |
88 | struct ext4_extent_header *eh) | |
89 | { | |
90 | struct ext4_extent_tail *et; | |
91 | ||
92 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
93 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
94 | return; | |
95 | ||
96 | et = find_ext4_extent_tail(eh); | |
97 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
98 | } | |
99 | ||
d583fb87 AH |
100 | static int ext4_split_extent(handle_t *handle, |
101 | struct inode *inode, | |
102 | struct ext4_ext_path *path, | |
103 | struct ext4_map_blocks *map, | |
104 | int split_flag, | |
105 | int flags); | |
106 | ||
5f95d21f LC |
107 | static int ext4_split_extent_at(handle_t *handle, |
108 | struct inode *inode, | |
109 | struct ext4_ext_path *path, | |
110 | ext4_lblk_t split, | |
111 | int split_flag, | |
112 | int flags); | |
113 | ||
91dd8c11 | 114 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 115 | struct extent_status *newes); |
91dd8c11 | 116 | |
487caeef JK |
117 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
118 | struct inode *inode, | |
119 | int needed) | |
a86c6181 AT |
120 | { |
121 | int err; | |
122 | ||
0390131b FM |
123 | if (!ext4_handle_valid(handle)) |
124 | return 0; | |
a86c6181 | 125 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
126 | return 0; |
127 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 128 | if (err <= 0) |
9102e4fa | 129 | return err; |
487caeef | 130 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
131 | if (err == 0) |
132 | err = -EAGAIN; | |
487caeef JK |
133 | |
134 | return err; | |
a86c6181 AT |
135 | } |
136 | ||
137 | /* | |
138 | * could return: | |
139 | * - EROFS | |
140 | * - ENOMEM | |
141 | */ | |
142 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
143 | struct ext4_ext_path *path) | |
144 | { | |
145 | if (path->p_bh) { | |
146 | /* path points to block */ | |
147 | return ext4_journal_get_write_access(handle, path->p_bh); | |
148 | } | |
149 | /* path points to leaf/index in inode body */ | |
150 | /* we use in-core data, no need to protect them */ | |
151 | return 0; | |
152 | } | |
153 | ||
154 | /* | |
155 | * could return: | |
156 | * - EROFS | |
157 | * - ENOMEM | |
158 | * - EIO | |
159 | */ | |
9ea7a0df TT |
160 | #define ext4_ext_dirty(handle, inode, path) \ |
161 | __ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path)) | |
162 | static int __ext4_ext_dirty(const char *where, unsigned int line, | |
163 | handle_t *handle, struct inode *inode, | |
164 | struct ext4_ext_path *path) | |
a86c6181 AT |
165 | { |
166 | int err; | |
167 | if (path->p_bh) { | |
7ac5990d | 168 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 169 | /* path points to block */ |
9ea7a0df TT |
170 | err = __ext4_handle_dirty_metadata(where, line, handle, |
171 | inode, path->p_bh); | |
a86c6181 AT |
172 | } else { |
173 | /* path points to leaf/index in inode body */ | |
174 | err = ext4_mark_inode_dirty(handle, inode); | |
175 | } | |
176 | return err; | |
177 | } | |
178 | ||
f65e6fba | 179 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 180 | struct ext4_ext_path *path, |
725d26d3 | 181 | ext4_lblk_t block) |
a86c6181 | 182 | { |
a86c6181 | 183 | if (path) { |
81fdbb4a | 184 | int depth = path->p_depth; |
a86c6181 | 185 | struct ext4_extent *ex; |
a86c6181 | 186 | |
ad4fb9ca KM |
187 | /* |
188 | * Try to predict block placement assuming that we are | |
189 | * filling in a file which will eventually be | |
190 | * non-sparse --- i.e., in the case of libbfd writing | |
191 | * an ELF object sections out-of-order but in a way | |
192 | * the eventually results in a contiguous object or | |
193 | * executable file, or some database extending a table | |
194 | * space file. However, this is actually somewhat | |
195 | * non-ideal if we are writing a sparse file such as | |
196 | * qemu or KVM writing a raw image file that is going | |
197 | * to stay fairly sparse, since it will end up | |
198 | * fragmenting the file system's free space. Maybe we | |
199 | * should have some hueristics or some way to allow | |
200 | * userspace to pass a hint to file system, | |
b8d6568a | 201 | * especially if the latter case turns out to be |
ad4fb9ca KM |
202 | * common. |
203 | */ | |
7e028976 | 204 | ex = path[depth].p_ext; |
ad4fb9ca KM |
205 | if (ex) { |
206 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
207 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
208 | ||
209 | if (block > ext_block) | |
210 | return ext_pblk + (block - ext_block); | |
211 | else | |
212 | return ext_pblk - (ext_block - block); | |
213 | } | |
a86c6181 | 214 | |
d0d856e8 RD |
215 | /* it looks like index is empty; |
216 | * try to find starting block from index itself */ | |
a86c6181 AT |
217 | if (path[depth].p_bh) |
218 | return path[depth].p_bh->b_blocknr; | |
219 | } | |
220 | ||
221 | /* OK. use inode's group */ | |
f86186b4 | 222 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
223 | } |
224 | ||
654b4908 AK |
225 | /* |
226 | * Allocation for a meta data block | |
227 | */ | |
f65e6fba | 228 | static ext4_fsblk_t |
654b4908 | 229 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 230 | struct ext4_ext_path *path, |
55f020db | 231 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 232 | { |
f65e6fba | 233 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
234 | |
235 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
236 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
237 | NULL, err); | |
a86c6181 AT |
238 | return newblock; |
239 | } | |
240 | ||
55ad63bf | 241 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
242 | { |
243 | int size; | |
244 | ||
245 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
246 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 247 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
248 | if (!check && size > 6) |
249 | size = 6; | |
a86c6181 AT |
250 | #endif |
251 | return size; | |
252 | } | |
253 | ||
55ad63bf | 254 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
255 | { |
256 | int size; | |
257 | ||
258 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
259 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 260 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
261 | if (!check && size > 5) |
262 | size = 5; | |
a86c6181 AT |
263 | #endif |
264 | return size; | |
265 | } | |
266 | ||
55ad63bf | 267 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
268 | { |
269 | int size; | |
270 | ||
271 | size = sizeof(EXT4_I(inode)->i_data); | |
272 | size -= sizeof(struct ext4_extent_header); | |
273 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 274 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
275 | if (!check && size > 3) |
276 | size = 3; | |
a86c6181 AT |
277 | #endif |
278 | return size; | |
279 | } | |
280 | ||
55ad63bf | 281 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
282 | { |
283 | int size; | |
284 | ||
285 | size = sizeof(EXT4_I(inode)->i_data); | |
286 | size -= sizeof(struct ext4_extent_header); | |
287 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 288 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
289 | if (!check && size > 4) |
290 | size = 4; | |
a86c6181 AT |
291 | #endif |
292 | return size; | |
293 | } | |
294 | ||
d2a17637 MC |
295 | /* |
296 | * Calculate the number of metadata blocks needed | |
297 | * to allocate @blocks | |
298 | * Worse case is one block per extent | |
299 | */ | |
01f49d0b | 300 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 301 | { |
9d0be502 | 302 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 303 | int idxs; |
d2a17637 | 304 | |
9d0be502 TT |
305 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
306 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
307 | |
308 | /* | |
9d0be502 TT |
309 | * If the new delayed allocation block is contiguous with the |
310 | * previous da block, it can share index blocks with the | |
311 | * previous block, so we only need to allocate a new index | |
312 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
313 | * an additional index block, and at ldxs**3 blocks, yet | |
314 | * another index blocks. | |
d2a17637 | 315 | */ |
9d0be502 TT |
316 | if (ei->i_da_metadata_calc_len && |
317 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
318 | int num = 0; |
319 | ||
9d0be502 TT |
320 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
321 | num++; | |
322 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
323 | num++; | |
324 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
325 | num++; | |
326 | ei->i_da_metadata_calc_len = 0; | |
327 | } else | |
328 | ei->i_da_metadata_calc_len++; | |
329 | ei->i_da_metadata_calc_last_lblock++; | |
330 | return num; | |
331 | } | |
d2a17637 | 332 | |
9d0be502 TT |
333 | /* |
334 | * In the worst case we need a new set of index blocks at | |
335 | * every level of the inode's extent tree. | |
336 | */ | |
337 | ei->i_da_metadata_calc_len = 1; | |
338 | ei->i_da_metadata_calc_last_lblock = lblock; | |
339 | return ext_depth(inode) + 1; | |
d2a17637 MC |
340 | } |
341 | ||
c29c0ae7 AT |
342 | static int |
343 | ext4_ext_max_entries(struct inode *inode, int depth) | |
344 | { | |
345 | int max; | |
346 | ||
347 | if (depth == ext_depth(inode)) { | |
348 | if (depth == 0) | |
55ad63bf | 349 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 350 | else |
55ad63bf | 351 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
352 | } else { |
353 | if (depth == 0) | |
55ad63bf | 354 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 355 | else |
55ad63bf | 356 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
357 | } |
358 | ||
359 | return max; | |
360 | } | |
361 | ||
56b19868 AK |
362 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
363 | { | |
bf89d16f | 364 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 365 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 366 | |
31d4f3a2 TT |
367 | if (len == 0) |
368 | return 0; | |
6fd058f7 | 369 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
370 | } |
371 | ||
372 | static int ext4_valid_extent_idx(struct inode *inode, | |
373 | struct ext4_extent_idx *ext_idx) | |
374 | { | |
bf89d16f | 375 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 376 | |
6fd058f7 | 377 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
378 | } |
379 | ||
380 | static int ext4_valid_extent_entries(struct inode *inode, | |
381 | struct ext4_extent_header *eh, | |
382 | int depth) | |
383 | { | |
56b19868 AK |
384 | unsigned short entries; |
385 | if (eh->eh_entries == 0) | |
386 | return 1; | |
387 | ||
388 | entries = le16_to_cpu(eh->eh_entries); | |
389 | ||
390 | if (depth == 0) { | |
391 | /* leaf entries */ | |
81fdbb4a | 392 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
56b19868 AK |
393 | while (entries) { |
394 | if (!ext4_valid_extent(inode, ext)) | |
395 | return 0; | |
396 | ext++; | |
397 | entries--; | |
398 | } | |
399 | } else { | |
81fdbb4a | 400 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
401 | while (entries) { |
402 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
403 | return 0; | |
404 | ext_idx++; | |
405 | entries--; | |
406 | } | |
407 | } | |
408 | return 1; | |
409 | } | |
410 | ||
c398eda0 TT |
411 | static int __ext4_ext_check(const char *function, unsigned int line, |
412 | struct inode *inode, struct ext4_extent_header *eh, | |
413 | int depth) | |
c29c0ae7 AT |
414 | { |
415 | const char *error_msg; | |
416 | int max = 0; | |
417 | ||
418 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
419 | error_msg = "invalid magic"; | |
420 | goto corrupted; | |
421 | } | |
422 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
423 | error_msg = "unexpected eh_depth"; | |
424 | goto corrupted; | |
425 | } | |
426 | if (unlikely(eh->eh_max == 0)) { | |
427 | error_msg = "invalid eh_max"; | |
428 | goto corrupted; | |
429 | } | |
430 | max = ext4_ext_max_entries(inode, depth); | |
431 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
432 | error_msg = "too large eh_max"; | |
433 | goto corrupted; | |
434 | } | |
435 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
436 | error_msg = "invalid eh_entries"; | |
437 | goto corrupted; | |
438 | } | |
56b19868 AK |
439 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
440 | error_msg = "invalid extent entries"; | |
441 | goto corrupted; | |
442 | } | |
7ac5990d DW |
443 | /* Verify checksum on non-root extent tree nodes */ |
444 | if (ext_depth(inode) != depth && | |
445 | !ext4_extent_block_csum_verify(inode, eh)) { | |
446 | error_msg = "extent tree corrupted"; | |
447 | goto corrupted; | |
448 | } | |
c29c0ae7 AT |
449 | return 0; |
450 | ||
451 | corrupted: | |
c398eda0 | 452 | ext4_error_inode(inode, function, line, 0, |
24676da4 | 453 | "bad header/extent: %s - magic %x, " |
c29c0ae7 | 454 | "entries %u, max %u(%u), depth %u(%u)", |
24676da4 | 455 | error_msg, le16_to_cpu(eh->eh_magic), |
c29c0ae7 AT |
456 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), |
457 | max, le16_to_cpu(eh->eh_depth), depth); | |
458 | ||
459 | return -EIO; | |
460 | } | |
461 | ||
56b19868 | 462 | #define ext4_ext_check(inode, eh, depth) \ |
c398eda0 | 463 | __ext4_ext_check(__func__, __LINE__, inode, eh, depth) |
c29c0ae7 | 464 | |
7a262f7c AK |
465 | int ext4_ext_check_inode(struct inode *inode) |
466 | { | |
467 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode)); | |
468 | } | |
469 | ||
f8489128 DW |
470 | static int __ext4_ext_check_block(const char *function, unsigned int line, |
471 | struct inode *inode, | |
472 | struct ext4_extent_header *eh, | |
473 | int depth, | |
474 | struct buffer_head *bh) | |
475 | { | |
476 | int ret; | |
477 | ||
478 | if (buffer_verified(bh)) | |
479 | return 0; | |
480 | ret = ext4_ext_check(inode, eh, depth); | |
481 | if (ret) | |
482 | return ret; | |
483 | set_buffer_verified(bh); | |
484 | return ret; | |
485 | } | |
486 | ||
487 | #define ext4_ext_check_block(inode, eh, depth, bh) \ | |
488 | __ext4_ext_check_block(__func__, __LINE__, inode, eh, depth, bh) | |
489 | ||
a86c6181 AT |
490 | #ifdef EXT_DEBUG |
491 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
492 | { | |
493 | int k, l = path->p_depth; | |
494 | ||
495 | ext_debug("path:"); | |
496 | for (k = 0; k <= l; k++, path++) { | |
497 | if (path->p_idx) { | |
2ae02107 | 498 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 499 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 500 | } else if (path->p_ext) { |
553f9008 | 501 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 502 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 503 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 504 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 505 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
506 | } else |
507 | ext_debug(" []"); | |
508 | } | |
509 | ext_debug("\n"); | |
510 | } | |
511 | ||
512 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
513 | { | |
514 | int depth = ext_depth(inode); | |
515 | struct ext4_extent_header *eh; | |
516 | struct ext4_extent *ex; | |
517 | int i; | |
518 | ||
519 | if (!path) | |
520 | return; | |
521 | ||
522 | eh = path[depth].p_hdr; | |
523 | ex = EXT_FIRST_EXTENT(eh); | |
524 | ||
553f9008 M |
525 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
526 | ||
a86c6181 | 527 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
528 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
529 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 530 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
531 | } |
532 | ext_debug("\n"); | |
533 | } | |
1b16da77 YY |
534 | |
535 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
536 | ext4_fsblk_t newblock, int level) | |
537 | { | |
538 | int depth = ext_depth(inode); | |
539 | struct ext4_extent *ex; | |
540 | ||
541 | if (depth != level) { | |
542 | struct ext4_extent_idx *idx; | |
543 | idx = path[level].p_idx; | |
544 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
545 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
546 | le32_to_cpu(idx->ei_block), | |
547 | ext4_idx_pblock(idx), | |
548 | newblock); | |
549 | idx++; | |
550 | } | |
551 | ||
552 | return; | |
553 | } | |
554 | ||
555 | ex = path[depth].p_ext; | |
556 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
557 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
558 | le32_to_cpu(ex->ee_block), | |
559 | ext4_ext_pblock(ex), | |
560 | ext4_ext_is_uninitialized(ex), | |
561 | ext4_ext_get_actual_len(ex), | |
562 | newblock); | |
563 | ex++; | |
564 | } | |
565 | } | |
566 | ||
a86c6181 | 567 | #else |
af5bc92d TT |
568 | #define ext4_ext_show_path(inode, path) |
569 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 570 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
571 | #endif |
572 | ||
b35905c1 | 573 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
574 | { |
575 | int depth = path->p_depth; | |
576 | int i; | |
577 | ||
578 | for (i = 0; i <= depth; i++, path++) | |
579 | if (path->p_bh) { | |
580 | brelse(path->p_bh); | |
581 | path->p_bh = NULL; | |
582 | } | |
583 | } | |
584 | ||
585 | /* | |
d0d856e8 RD |
586 | * ext4_ext_binsearch_idx: |
587 | * binary search for the closest index of the given block | |
c29c0ae7 | 588 | * the header must be checked before calling this |
a86c6181 AT |
589 | */ |
590 | static void | |
725d26d3 AK |
591 | ext4_ext_binsearch_idx(struct inode *inode, |
592 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
593 | { |
594 | struct ext4_extent_header *eh = path->p_hdr; | |
595 | struct ext4_extent_idx *r, *l, *m; | |
596 | ||
a86c6181 | 597 | |
bba90743 | 598 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
599 | |
600 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 601 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
602 | while (l <= r) { |
603 | m = l + (r - l) / 2; | |
604 | if (block < le32_to_cpu(m->ei_block)) | |
605 | r = m - 1; | |
606 | else | |
607 | l = m + 1; | |
26d535ed DM |
608 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
609 | m, le32_to_cpu(m->ei_block), | |
610 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
611 | } |
612 | ||
613 | path->p_idx = l - 1; | |
4a3c3a51 | 614 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 615 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
616 | |
617 | #ifdef CHECK_BINSEARCH | |
618 | { | |
619 | struct ext4_extent_idx *chix, *ix; | |
620 | int k; | |
621 | ||
622 | chix = ix = EXT_FIRST_INDEX(eh); | |
623 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
624 | if (k != 0 && | |
625 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
626 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
627 | "first=0x%p\n", k, | |
628 | ix, EXT_FIRST_INDEX(eh)); | |
629 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
630 | le32_to_cpu(ix->ei_block), |
631 | le32_to_cpu(ix[-1].ei_block)); | |
632 | } | |
633 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 634 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
635 | if (block < le32_to_cpu(ix->ei_block)) |
636 | break; | |
637 | chix = ix; | |
638 | } | |
639 | BUG_ON(chix != path->p_idx); | |
640 | } | |
641 | #endif | |
642 | ||
643 | } | |
644 | ||
645 | /* | |
d0d856e8 RD |
646 | * ext4_ext_binsearch: |
647 | * binary search for closest extent of the given block | |
c29c0ae7 | 648 | * the header must be checked before calling this |
a86c6181 AT |
649 | */ |
650 | static void | |
725d26d3 AK |
651 | ext4_ext_binsearch(struct inode *inode, |
652 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
653 | { |
654 | struct ext4_extent_header *eh = path->p_hdr; | |
655 | struct ext4_extent *r, *l, *m; | |
656 | ||
a86c6181 AT |
657 | if (eh->eh_entries == 0) { |
658 | /* | |
d0d856e8 RD |
659 | * this leaf is empty: |
660 | * we get such a leaf in split/add case | |
a86c6181 AT |
661 | */ |
662 | return; | |
663 | } | |
664 | ||
bba90743 | 665 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
666 | |
667 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 668 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
669 | |
670 | while (l <= r) { | |
671 | m = l + (r - l) / 2; | |
672 | if (block < le32_to_cpu(m->ee_block)) | |
673 | r = m - 1; | |
674 | else | |
675 | l = m + 1; | |
26d535ed DM |
676 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
677 | m, le32_to_cpu(m->ee_block), | |
678 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
679 | } |
680 | ||
681 | path->p_ext = l - 1; | |
553f9008 | 682 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 683 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 684 | ext4_ext_pblock(path->p_ext), |
553f9008 | 685 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 686 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
687 | |
688 | #ifdef CHECK_BINSEARCH | |
689 | { | |
690 | struct ext4_extent *chex, *ex; | |
691 | int k; | |
692 | ||
693 | chex = ex = EXT_FIRST_EXTENT(eh); | |
694 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
695 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 696 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
697 | if (block < le32_to_cpu(ex->ee_block)) |
698 | break; | |
699 | chex = ex; | |
700 | } | |
701 | BUG_ON(chex != path->p_ext); | |
702 | } | |
703 | #endif | |
704 | ||
705 | } | |
706 | ||
707 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
708 | { | |
709 | struct ext4_extent_header *eh; | |
710 | ||
711 | eh = ext_inode_hdr(inode); | |
712 | eh->eh_depth = 0; | |
713 | eh->eh_entries = 0; | |
714 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 715 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 716 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
717 | return 0; |
718 | } | |
719 | ||
720 | struct ext4_ext_path * | |
725d26d3 AK |
721 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
722 | struct ext4_ext_path *path) | |
a86c6181 AT |
723 | { |
724 | struct ext4_extent_header *eh; | |
725 | struct buffer_head *bh; | |
726 | short int depth, i, ppos = 0, alloc = 0; | |
860d21e2 | 727 | int ret; |
a86c6181 AT |
728 | |
729 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 730 | depth = ext_depth(inode); |
a86c6181 AT |
731 | |
732 | /* account possible depth increase */ | |
733 | if (!path) { | |
5d4958f9 | 734 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
735 | GFP_NOFS); |
736 | if (!path) | |
737 | return ERR_PTR(-ENOMEM); | |
738 | alloc = 1; | |
739 | } | |
a86c6181 | 740 | path[0].p_hdr = eh; |
1973adcb | 741 | path[0].p_bh = NULL; |
a86c6181 | 742 | |
c29c0ae7 | 743 | i = depth; |
a86c6181 AT |
744 | /* walk through the tree */ |
745 | while (i) { | |
746 | ext_debug("depth %d: num %d, max %d\n", | |
747 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 748 | |
a86c6181 | 749 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 750 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
751 | path[ppos].p_depth = i; |
752 | path[ppos].p_ext = NULL; | |
753 | ||
7a262f7c | 754 | bh = sb_getblk(inode->i_sb, path[ppos].p_block); |
860d21e2 TT |
755 | if (unlikely(!bh)) { |
756 | ret = -ENOMEM; | |
a86c6181 | 757 | goto err; |
860d21e2 | 758 | } |
7a262f7c | 759 | if (!bh_uptodate_or_lock(bh)) { |
0562e0ba JZ |
760 | trace_ext4_ext_load_extent(inode, block, |
761 | path[ppos].p_block); | |
860d21e2 TT |
762 | ret = bh_submit_read(bh); |
763 | if (ret < 0) { | |
7a262f7c AK |
764 | put_bh(bh); |
765 | goto err; | |
766 | } | |
7a262f7c | 767 | } |
a86c6181 AT |
768 | eh = ext_block_hdr(bh); |
769 | ppos++; | |
273df556 FM |
770 | if (unlikely(ppos > depth)) { |
771 | put_bh(bh); | |
772 | EXT4_ERROR_INODE(inode, | |
773 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 774 | ret = -EIO; |
273df556 FM |
775 | goto err; |
776 | } | |
a86c6181 AT |
777 | path[ppos].p_bh = bh; |
778 | path[ppos].p_hdr = eh; | |
779 | i--; | |
780 | ||
860d21e2 TT |
781 | ret = ext4_ext_check_block(inode, eh, i, bh); |
782 | if (ret < 0) | |
a86c6181 AT |
783 | goto err; |
784 | } | |
785 | ||
786 | path[ppos].p_depth = i; | |
a86c6181 AT |
787 | path[ppos].p_ext = NULL; |
788 | path[ppos].p_idx = NULL; | |
789 | ||
a86c6181 AT |
790 | /* find extent */ |
791 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
792 | /* if not an empty leaf */ |
793 | if (path[ppos].p_ext) | |
bf89d16f | 794 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
795 | |
796 | ext4_ext_show_path(inode, path); | |
797 | ||
798 | return path; | |
799 | ||
800 | err: | |
801 | ext4_ext_drop_refs(path); | |
802 | if (alloc) | |
803 | kfree(path); | |
860d21e2 | 804 | return ERR_PTR(ret); |
a86c6181 AT |
805 | } |
806 | ||
807 | /* | |
d0d856e8 RD |
808 | * ext4_ext_insert_index: |
809 | * insert new index [@logical;@ptr] into the block at @curp; | |
810 | * check where to insert: before @curp or after @curp | |
a86c6181 | 811 | */ |
1f109d5a TT |
812 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
813 | struct ext4_ext_path *curp, | |
814 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
815 | { |
816 | struct ext4_extent_idx *ix; | |
817 | int len, err; | |
818 | ||
7e028976 AM |
819 | err = ext4_ext_get_access(handle, inode, curp); |
820 | if (err) | |
a86c6181 AT |
821 | return err; |
822 | ||
273df556 FM |
823 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
824 | EXT4_ERROR_INODE(inode, | |
825 | "logical %d == ei_block %d!", | |
826 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
827 | return -EIO; | |
828 | } | |
d4620315 RD |
829 | |
830 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
831 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
832 | EXT4_ERROR_INODE(inode, | |
833 | "eh_entries %d >= eh_max %d!", | |
834 | le16_to_cpu(curp->p_hdr->eh_entries), | |
835 | le16_to_cpu(curp->p_hdr->eh_max)); | |
836 | return -EIO; | |
837 | } | |
838 | ||
a86c6181 AT |
839 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
840 | /* insert after */ | |
80e675f9 | 841 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
842 | ix = curp->p_idx + 1; |
843 | } else { | |
844 | /* insert before */ | |
80e675f9 | 845 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
846 | ix = curp->p_idx; |
847 | } | |
848 | ||
80e675f9 EG |
849 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
850 | BUG_ON(len < 0); | |
851 | if (len > 0) { | |
852 | ext_debug("insert new index %d: " | |
853 | "move %d indices from 0x%p to 0x%p\n", | |
854 | logical, len, ix, ix + 1); | |
855 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
856 | } | |
857 | ||
f472e026 TM |
858 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
859 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
860 | return -EIO; | |
861 | } | |
862 | ||
a86c6181 | 863 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 864 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 865 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 866 | |
273df556 FM |
867 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
868 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
869 | return -EIO; | |
870 | } | |
a86c6181 AT |
871 | |
872 | err = ext4_ext_dirty(handle, inode, curp); | |
873 | ext4_std_error(inode->i_sb, err); | |
874 | ||
875 | return err; | |
876 | } | |
877 | ||
878 | /* | |
d0d856e8 RD |
879 | * ext4_ext_split: |
880 | * inserts new subtree into the path, using free index entry | |
881 | * at depth @at: | |
882 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
883 | * - makes decision where to split | |
884 | * - moves remaining extents and index entries (right to the split point) | |
885 | * into the newly allocated blocks | |
886 | * - initializes subtree | |
a86c6181 AT |
887 | */ |
888 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
889 | unsigned int flags, |
890 | struct ext4_ext_path *path, | |
891 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
892 | { |
893 | struct buffer_head *bh = NULL; | |
894 | int depth = ext_depth(inode); | |
895 | struct ext4_extent_header *neh; | |
896 | struct ext4_extent_idx *fidx; | |
a86c6181 | 897 | int i = at, k, m, a; |
f65e6fba | 898 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 899 | __le32 border; |
f65e6fba | 900 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
901 | int err = 0; |
902 | ||
903 | /* make decision: where to split? */ | |
d0d856e8 | 904 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 905 | |
d0d856e8 | 906 | /* if current leaf will be split, then we should use |
a86c6181 | 907 | * border from split point */ |
273df556 FM |
908 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
909 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
910 | return -EIO; | |
911 | } | |
a86c6181 AT |
912 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
913 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 914 | ext_debug("leaf will be split." |
a86c6181 | 915 | " next leaf starts at %d\n", |
8c55e204 | 916 | le32_to_cpu(border)); |
a86c6181 AT |
917 | } else { |
918 | border = newext->ee_block; | |
919 | ext_debug("leaf will be added." | |
920 | " next leaf starts at %d\n", | |
8c55e204 | 921 | le32_to_cpu(border)); |
a86c6181 AT |
922 | } |
923 | ||
924 | /* | |
d0d856e8 RD |
925 | * If error occurs, then we break processing |
926 | * and mark filesystem read-only. index won't | |
a86c6181 | 927 | * be inserted and tree will be in consistent |
d0d856e8 | 928 | * state. Next mount will repair buffers too. |
a86c6181 AT |
929 | */ |
930 | ||
931 | /* | |
d0d856e8 RD |
932 | * Get array to track all allocated blocks. |
933 | * We need this to handle errors and free blocks | |
934 | * upon them. | |
a86c6181 | 935 | */ |
5d4958f9 | 936 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
937 | if (!ablocks) |
938 | return -ENOMEM; | |
a86c6181 AT |
939 | |
940 | /* allocate all needed blocks */ | |
941 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
942 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 943 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 944 | newext, &err, flags); |
a86c6181 AT |
945 | if (newblock == 0) |
946 | goto cleanup; | |
947 | ablocks[a] = newblock; | |
948 | } | |
949 | ||
950 | /* initialize new leaf */ | |
951 | newblock = ablocks[--a]; | |
273df556 FM |
952 | if (unlikely(newblock == 0)) { |
953 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
954 | err = -EIO; | |
955 | goto cleanup; | |
956 | } | |
a86c6181 | 957 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 958 | if (unlikely(!bh)) { |
860d21e2 | 959 | err = -ENOMEM; |
a86c6181 AT |
960 | goto cleanup; |
961 | } | |
962 | lock_buffer(bh); | |
963 | ||
7e028976 AM |
964 | err = ext4_journal_get_create_access(handle, bh); |
965 | if (err) | |
a86c6181 AT |
966 | goto cleanup; |
967 | ||
968 | neh = ext_block_hdr(bh); | |
969 | neh->eh_entries = 0; | |
55ad63bf | 970 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
971 | neh->eh_magic = EXT4_EXT_MAGIC; |
972 | neh->eh_depth = 0; | |
a86c6181 | 973 | |
d0d856e8 | 974 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
975 | if (unlikely(path[depth].p_hdr->eh_entries != |
976 | path[depth].p_hdr->eh_max)) { | |
977 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
978 | path[depth].p_hdr->eh_entries, | |
979 | path[depth].p_hdr->eh_max); | |
980 | err = -EIO; | |
981 | goto cleanup; | |
982 | } | |
a86c6181 | 983 | /* start copy from next extent */ |
1b16da77 YY |
984 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
985 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 986 | if (m) { |
1b16da77 YY |
987 | struct ext4_extent *ex; |
988 | ex = EXT_FIRST_EXTENT(neh); | |
989 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 990 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
991 | } |
992 | ||
7ac5990d | 993 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
994 | set_buffer_uptodate(bh); |
995 | unlock_buffer(bh); | |
996 | ||
0390131b | 997 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 998 | if (err) |
a86c6181 AT |
999 | goto cleanup; |
1000 | brelse(bh); | |
1001 | bh = NULL; | |
1002 | ||
1003 | /* correct old leaf */ | |
1004 | if (m) { | |
7e028976 AM |
1005 | err = ext4_ext_get_access(handle, inode, path + depth); |
1006 | if (err) | |
a86c6181 | 1007 | goto cleanup; |
e8546d06 | 1008 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1009 | err = ext4_ext_dirty(handle, inode, path + depth); |
1010 | if (err) | |
a86c6181 AT |
1011 | goto cleanup; |
1012 | ||
1013 | } | |
1014 | ||
1015 | /* create intermediate indexes */ | |
1016 | k = depth - at - 1; | |
273df556 FM |
1017 | if (unlikely(k < 0)) { |
1018 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1019 | err = -EIO; | |
1020 | goto cleanup; | |
1021 | } | |
a86c6181 AT |
1022 | if (k) |
1023 | ext_debug("create %d intermediate indices\n", k); | |
1024 | /* insert new index into current index block */ | |
1025 | /* current depth stored in i var */ | |
1026 | i = depth - 1; | |
1027 | while (k--) { | |
1028 | oldblock = newblock; | |
1029 | newblock = ablocks[--a]; | |
bba90743 | 1030 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1031 | if (unlikely(!bh)) { |
860d21e2 | 1032 | err = -ENOMEM; |
a86c6181 AT |
1033 | goto cleanup; |
1034 | } | |
1035 | lock_buffer(bh); | |
1036 | ||
7e028976 AM |
1037 | err = ext4_journal_get_create_access(handle, bh); |
1038 | if (err) | |
a86c6181 AT |
1039 | goto cleanup; |
1040 | ||
1041 | neh = ext_block_hdr(bh); | |
1042 | neh->eh_entries = cpu_to_le16(1); | |
1043 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1044 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1045 | neh->eh_depth = cpu_to_le16(depth - i); |
1046 | fidx = EXT_FIRST_INDEX(neh); | |
1047 | fidx->ei_block = border; | |
f65e6fba | 1048 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1049 | |
bba90743 ES |
1050 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1051 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1052 | |
1b16da77 | 1053 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1054 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1055 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1056 | EXT4_ERROR_INODE(inode, | |
1057 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1058 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1059 | err = -EIO; | |
1060 | goto cleanup; | |
1061 | } | |
1b16da77 YY |
1062 | /* start copy indexes */ |
1063 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1064 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1065 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1066 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1067 | if (m) { |
1b16da77 | 1068 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1069 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1070 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1071 | } |
7ac5990d | 1072 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1073 | set_buffer_uptodate(bh); |
1074 | unlock_buffer(bh); | |
1075 | ||
0390131b | 1076 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1077 | if (err) |
a86c6181 AT |
1078 | goto cleanup; |
1079 | brelse(bh); | |
1080 | bh = NULL; | |
1081 | ||
1082 | /* correct old index */ | |
1083 | if (m) { | |
1084 | err = ext4_ext_get_access(handle, inode, path + i); | |
1085 | if (err) | |
1086 | goto cleanup; | |
e8546d06 | 1087 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1088 | err = ext4_ext_dirty(handle, inode, path + i); |
1089 | if (err) | |
1090 | goto cleanup; | |
1091 | } | |
1092 | ||
1093 | i--; | |
1094 | } | |
1095 | ||
1096 | /* insert new index */ | |
a86c6181 AT |
1097 | err = ext4_ext_insert_index(handle, inode, path + at, |
1098 | le32_to_cpu(border), newblock); | |
1099 | ||
1100 | cleanup: | |
1101 | if (bh) { | |
1102 | if (buffer_locked(bh)) | |
1103 | unlock_buffer(bh); | |
1104 | brelse(bh); | |
1105 | } | |
1106 | ||
1107 | if (err) { | |
1108 | /* free all allocated blocks in error case */ | |
1109 | for (i = 0; i < depth; i++) { | |
1110 | if (!ablocks[i]) | |
1111 | continue; | |
7dc57615 | 1112 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1113 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1114 | } |
1115 | } | |
1116 | kfree(ablocks); | |
1117 | ||
1118 | return err; | |
1119 | } | |
1120 | ||
1121 | /* | |
d0d856e8 RD |
1122 | * ext4_ext_grow_indepth: |
1123 | * implements tree growing procedure: | |
1124 | * - allocates new block | |
1125 | * - moves top-level data (index block or leaf) into the new block | |
1126 | * - initializes new top-level, creating index that points to the | |
1127 | * just created block | |
a86c6181 AT |
1128 | */ |
1129 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1130 | unsigned int flags, |
55f020db | 1131 | struct ext4_extent *newext) |
a86c6181 | 1132 | { |
a86c6181 | 1133 | struct ext4_extent_header *neh; |
a86c6181 | 1134 | struct buffer_head *bh; |
f65e6fba | 1135 | ext4_fsblk_t newblock; |
a86c6181 AT |
1136 | int err = 0; |
1137 | ||
1939dd84 | 1138 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1139 | newext, &err, flags); |
a86c6181 AT |
1140 | if (newblock == 0) |
1141 | return err; | |
1142 | ||
1143 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1144 | if (unlikely(!bh)) |
860d21e2 | 1145 | return -ENOMEM; |
a86c6181 AT |
1146 | lock_buffer(bh); |
1147 | ||
7e028976 AM |
1148 | err = ext4_journal_get_create_access(handle, bh); |
1149 | if (err) { | |
a86c6181 AT |
1150 | unlock_buffer(bh); |
1151 | goto out; | |
1152 | } | |
1153 | ||
1154 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1155 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1156 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1157 | |
1158 | /* set size of new block */ | |
1159 | neh = ext_block_hdr(bh); | |
1160 | /* old root could have indexes or leaves | |
1161 | * so calculate e_max right way */ | |
1162 | if (ext_depth(inode)) | |
55ad63bf | 1163 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1164 | else |
55ad63bf | 1165 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1166 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1167 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1168 | set_buffer_uptodate(bh); |
1169 | unlock_buffer(bh); | |
1170 | ||
0390131b | 1171 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1172 | if (err) |
a86c6181 AT |
1173 | goto out; |
1174 | ||
1939dd84 | 1175 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1176 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1177 | neh->eh_entries = cpu_to_le16(1); |
1178 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1179 | if (neh->eh_depth == 0) { | |
1180 | /* Root extent block becomes index block */ | |
1181 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1182 | EXT_FIRST_INDEX(neh)->ei_block = | |
1183 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1184 | } | |
2ae02107 | 1185 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1186 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1187 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1188 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1189 | |
ba39ebb6 | 1190 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1191 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1192 | out: |
1193 | brelse(bh); | |
1194 | ||
1195 | return err; | |
1196 | } | |
1197 | ||
1198 | /* | |
d0d856e8 RD |
1199 | * ext4_ext_create_new_leaf: |
1200 | * finds empty index and adds new leaf. | |
1201 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1202 | */ |
1203 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
55f020db AH |
1204 | unsigned int flags, |
1205 | struct ext4_ext_path *path, | |
1206 | struct ext4_extent *newext) | |
a86c6181 AT |
1207 | { |
1208 | struct ext4_ext_path *curp; | |
1209 | int depth, i, err = 0; | |
1210 | ||
1211 | repeat: | |
1212 | i = depth = ext_depth(inode); | |
1213 | ||
1214 | /* walk up to the tree and look for free index entry */ | |
1215 | curp = path + depth; | |
1216 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1217 | i--; | |
1218 | curp--; | |
1219 | } | |
1220 | ||
d0d856e8 RD |
1221 | /* we use already allocated block for index block, |
1222 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1223 | if (EXT_HAS_FREE_INDEX(curp)) { |
1224 | /* if we found index with free entry, then use that | |
1225 | * entry: create all needed subtree and add new leaf */ | |
55f020db | 1226 | err = ext4_ext_split(handle, inode, flags, path, newext, i); |
787e0981 SF |
1227 | if (err) |
1228 | goto out; | |
a86c6181 AT |
1229 | |
1230 | /* refill path */ | |
1231 | ext4_ext_drop_refs(path); | |
1232 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1233 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1234 | path); | |
a86c6181 AT |
1235 | if (IS_ERR(path)) |
1236 | err = PTR_ERR(path); | |
1237 | } else { | |
1238 | /* tree is full, time to grow in depth */ | |
1939dd84 | 1239 | err = ext4_ext_grow_indepth(handle, inode, flags, newext); |
a86c6181 AT |
1240 | if (err) |
1241 | goto out; | |
1242 | ||
1243 | /* refill path */ | |
1244 | ext4_ext_drop_refs(path); | |
1245 | path = ext4_ext_find_extent(inode, | |
725d26d3 AK |
1246 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
1247 | path); | |
a86c6181 AT |
1248 | if (IS_ERR(path)) { |
1249 | err = PTR_ERR(path); | |
1250 | goto out; | |
1251 | } | |
1252 | ||
1253 | /* | |
d0d856e8 RD |
1254 | * only first (depth 0 -> 1) produces free space; |
1255 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1256 | */ |
1257 | depth = ext_depth(inode); | |
1258 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1259 | /* now we need to split */ |
a86c6181 AT |
1260 | goto repeat; |
1261 | } | |
1262 | } | |
1263 | ||
1264 | out: | |
1265 | return err; | |
1266 | } | |
1267 | ||
1988b51e AT |
1268 | /* |
1269 | * search the closest allocated block to the left for *logical | |
1270 | * and returns it at @logical + it's physical address at @phys | |
1271 | * if *logical is the smallest allocated block, the function | |
1272 | * returns 0 at @phys | |
1273 | * return value contains 0 (success) or error code | |
1274 | */ | |
1f109d5a TT |
1275 | static int ext4_ext_search_left(struct inode *inode, |
1276 | struct ext4_ext_path *path, | |
1277 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1278 | { |
1279 | struct ext4_extent_idx *ix; | |
1280 | struct ext4_extent *ex; | |
b939e376 | 1281 | int depth, ee_len; |
1988b51e | 1282 | |
273df556 FM |
1283 | if (unlikely(path == NULL)) { |
1284 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1285 | return -EIO; | |
1286 | } | |
1988b51e AT |
1287 | depth = path->p_depth; |
1288 | *phys = 0; | |
1289 | ||
1290 | if (depth == 0 && path->p_ext == NULL) | |
1291 | return 0; | |
1292 | ||
1293 | /* usually extent in the path covers blocks smaller | |
1294 | * then *logical, but it can be that extent is the | |
1295 | * first one in the file */ | |
1296 | ||
1297 | ex = path[depth].p_ext; | |
b939e376 | 1298 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1299 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1300 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1301 | EXT4_ERROR_INODE(inode, | |
1302 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1303 | *logical, le32_to_cpu(ex->ee_block)); | |
1304 | return -EIO; | |
1305 | } | |
1988b51e AT |
1306 | while (--depth >= 0) { |
1307 | ix = path[depth].p_idx; | |
273df556 FM |
1308 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1309 | EXT4_ERROR_INODE(inode, | |
1310 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1311 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1312 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1313 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1314 | depth); |
1315 | return -EIO; | |
1316 | } | |
1988b51e AT |
1317 | } |
1318 | return 0; | |
1319 | } | |
1320 | ||
273df556 FM |
1321 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1322 | EXT4_ERROR_INODE(inode, | |
1323 | "logical %d < ee_block %d + ee_len %d!", | |
1324 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1325 | return -EIO; | |
1326 | } | |
1988b51e | 1327 | |
b939e376 | 1328 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1329 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1330 | return 0; |
1331 | } | |
1332 | ||
1333 | /* | |
1334 | * search the closest allocated block to the right for *logical | |
1335 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1336 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1337 | * returns 0 at @phys |
1338 | * return value contains 0 (success) or error code | |
1339 | */ | |
1f109d5a TT |
1340 | static int ext4_ext_search_right(struct inode *inode, |
1341 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1342 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1343 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1344 | { |
1345 | struct buffer_head *bh = NULL; | |
1346 | struct ext4_extent_header *eh; | |
1347 | struct ext4_extent_idx *ix; | |
1348 | struct ext4_extent *ex; | |
1349 | ext4_fsblk_t block; | |
395a87bf ES |
1350 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1351 | int ee_len; | |
1988b51e | 1352 | |
273df556 FM |
1353 | if (unlikely(path == NULL)) { |
1354 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1355 | return -EIO; | |
1356 | } | |
1988b51e AT |
1357 | depth = path->p_depth; |
1358 | *phys = 0; | |
1359 | ||
1360 | if (depth == 0 && path->p_ext == NULL) | |
1361 | return 0; | |
1362 | ||
1363 | /* usually extent in the path covers blocks smaller | |
1364 | * then *logical, but it can be that extent is the | |
1365 | * first one in the file */ | |
1366 | ||
1367 | ex = path[depth].p_ext; | |
b939e376 | 1368 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1369 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1370 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1371 | EXT4_ERROR_INODE(inode, | |
1372 | "first_extent(path[%d].p_hdr) != ex", | |
1373 | depth); | |
1374 | return -EIO; | |
1375 | } | |
1988b51e AT |
1376 | while (--depth >= 0) { |
1377 | ix = path[depth].p_idx; | |
273df556 FM |
1378 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1379 | EXT4_ERROR_INODE(inode, | |
1380 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1381 | *logical); | |
1382 | return -EIO; | |
1383 | } | |
1988b51e | 1384 | } |
4d33b1ef | 1385 | goto found_extent; |
1988b51e AT |
1386 | } |
1387 | ||
273df556 FM |
1388 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1389 | EXT4_ERROR_INODE(inode, | |
1390 | "logical %d < ee_block %d + ee_len %d!", | |
1391 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1392 | return -EIO; | |
1393 | } | |
1988b51e AT |
1394 | |
1395 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1396 | /* next allocated block in this leaf */ | |
1397 | ex++; | |
4d33b1ef | 1398 | goto found_extent; |
1988b51e AT |
1399 | } |
1400 | ||
1401 | /* go up and search for index to the right */ | |
1402 | while (--depth >= 0) { | |
1403 | ix = path[depth].p_idx; | |
1404 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1405 | goto got_index; |
1988b51e AT |
1406 | } |
1407 | ||
25f1ee3a WF |
1408 | /* we've gone up to the root and found no index to the right */ |
1409 | return 0; | |
1988b51e | 1410 | |
25f1ee3a | 1411 | got_index: |
1988b51e AT |
1412 | /* we've found index to the right, let's |
1413 | * follow it and find the closest allocated | |
1414 | * block to the right */ | |
1415 | ix++; | |
bf89d16f | 1416 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1417 | while (++depth < path->p_depth) { |
1418 | bh = sb_bread(inode->i_sb, block); | |
1419 | if (bh == NULL) | |
1420 | return -EIO; | |
1421 | eh = ext_block_hdr(bh); | |
395a87bf | 1422 | /* subtract from p_depth to get proper eh_depth */ |
f8489128 DW |
1423 | if (ext4_ext_check_block(inode, eh, |
1424 | path->p_depth - depth, bh)) { | |
1988b51e AT |
1425 | put_bh(bh); |
1426 | return -EIO; | |
1427 | } | |
1428 | ix = EXT_FIRST_INDEX(eh); | |
bf89d16f | 1429 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1430 | put_bh(bh); |
1431 | } | |
1432 | ||
1433 | bh = sb_bread(inode->i_sb, block); | |
1434 | if (bh == NULL) | |
1435 | return -EIO; | |
1436 | eh = ext_block_hdr(bh); | |
f8489128 | 1437 | if (ext4_ext_check_block(inode, eh, path->p_depth - depth, bh)) { |
1988b51e AT |
1438 | put_bh(bh); |
1439 | return -EIO; | |
1440 | } | |
1441 | ex = EXT_FIRST_EXTENT(eh); | |
4d33b1ef | 1442 | found_extent: |
1988b51e | 1443 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1444 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1445 | *ret_ex = ex; |
1446 | if (bh) | |
1447 | put_bh(bh); | |
1988b51e | 1448 | return 0; |
1988b51e AT |
1449 | } |
1450 | ||
a86c6181 | 1451 | /* |
d0d856e8 | 1452 | * ext4_ext_next_allocated_block: |
f17722f9 | 1453 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1454 | * NOTE: it considers block number from index entry as |
1455 | * allocated block. Thus, index entries have to be consistent | |
1456 | * with leaves. | |
a86c6181 | 1457 | */ |
725d26d3 | 1458 | static ext4_lblk_t |
a86c6181 AT |
1459 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1460 | { | |
1461 | int depth; | |
1462 | ||
1463 | BUG_ON(path == NULL); | |
1464 | depth = path->p_depth; | |
1465 | ||
1466 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1467 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1468 | |
1469 | while (depth >= 0) { | |
1470 | if (depth == path->p_depth) { | |
1471 | /* leaf */ | |
6f8ff537 CW |
1472 | if (path[depth].p_ext && |
1473 | path[depth].p_ext != | |
a86c6181 AT |
1474 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1475 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1476 | } else { | |
1477 | /* index */ | |
1478 | if (path[depth].p_idx != | |
1479 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1480 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1481 | } | |
1482 | depth--; | |
1483 | } | |
1484 | ||
f17722f9 | 1485 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1486 | } |
1487 | ||
1488 | /* | |
d0d856e8 | 1489 | * ext4_ext_next_leaf_block: |
f17722f9 | 1490 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1491 | */ |
5718789d | 1492 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1493 | { |
1494 | int depth; | |
1495 | ||
1496 | BUG_ON(path == NULL); | |
1497 | depth = path->p_depth; | |
1498 | ||
1499 | /* zero-tree has no leaf blocks at all */ | |
1500 | if (depth == 0) | |
f17722f9 | 1501 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1502 | |
1503 | /* go to index block */ | |
1504 | depth--; | |
1505 | ||
1506 | while (depth >= 0) { | |
1507 | if (path[depth].p_idx != | |
1508 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1509 | return (ext4_lblk_t) |
1510 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1511 | depth--; |
1512 | } | |
1513 | ||
f17722f9 | 1514 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1515 | } |
1516 | ||
1517 | /* | |
d0d856e8 RD |
1518 | * ext4_ext_correct_indexes: |
1519 | * if leaf gets modified and modified extent is first in the leaf, | |
1520 | * then we have to correct all indexes above. | |
a86c6181 AT |
1521 | * TODO: do we need to correct tree in all cases? |
1522 | */ | |
1d03ec98 | 1523 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1524 | struct ext4_ext_path *path) |
1525 | { | |
1526 | struct ext4_extent_header *eh; | |
1527 | int depth = ext_depth(inode); | |
1528 | struct ext4_extent *ex; | |
1529 | __le32 border; | |
1530 | int k, err = 0; | |
1531 | ||
1532 | eh = path[depth].p_hdr; | |
1533 | ex = path[depth].p_ext; | |
273df556 FM |
1534 | |
1535 | if (unlikely(ex == NULL || eh == NULL)) { | |
1536 | EXT4_ERROR_INODE(inode, | |
1537 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1538 | return -EIO; | |
1539 | } | |
a86c6181 AT |
1540 | |
1541 | if (depth == 0) { | |
1542 | /* there is no tree at all */ | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1546 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1547 | /* we correct tree if first leaf got modified only */ | |
1548 | return 0; | |
1549 | } | |
1550 | ||
1551 | /* | |
d0d856e8 | 1552 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1553 | */ |
1554 | k = depth - 1; | |
1555 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1556 | err = ext4_ext_get_access(handle, inode, path + k); |
1557 | if (err) | |
a86c6181 AT |
1558 | return err; |
1559 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1560 | err = ext4_ext_dirty(handle, inode, path + k); |
1561 | if (err) | |
a86c6181 AT |
1562 | return err; |
1563 | ||
1564 | while (k--) { | |
1565 | /* change all left-side indexes */ | |
1566 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1567 | break; | |
7e028976 AM |
1568 | err = ext4_ext_get_access(handle, inode, path + k); |
1569 | if (err) | |
a86c6181 AT |
1570 | break; |
1571 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1572 | err = ext4_ext_dirty(handle, inode, path + k); |
1573 | if (err) | |
a86c6181 AT |
1574 | break; |
1575 | } | |
1576 | ||
1577 | return err; | |
1578 | } | |
1579 | ||
748de673 | 1580 | int |
a86c6181 AT |
1581 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1582 | struct ext4_extent *ex2) | |
1583 | { | |
749269fa | 1584 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1585 | |
1586 | /* | |
ec22ba8e DM |
1587 | * Make sure that both extents are initialized. We don't merge |
1588 | * uninitialized extents so that we can be sure that end_io code has | |
1589 | * the extent that was written properly split out and conversion to | |
1590 | * initialized is trivial. | |
a2df2a63 | 1591 | */ |
ec22ba8e | 1592 | if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2)) |
a2df2a63 AA |
1593 | return 0; |
1594 | ||
749269fa AA |
1595 | if (ext4_ext_is_uninitialized(ex1)) |
1596 | max_len = EXT_UNINIT_MAX_LEN; | |
1597 | else | |
1598 | max_len = EXT_INIT_MAX_LEN; | |
1599 | ||
a2df2a63 AA |
1600 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1601 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1602 | ||
1603 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1604 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1605 | return 0; |
1606 | ||
471d4011 SB |
1607 | /* |
1608 | * To allow future support for preallocated extents to be added | |
1609 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1610 | * this can result in the top bit of ee_len being set. |
471d4011 | 1611 | */ |
749269fa | 1612 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1613 | return 0; |
bbf2f9fb | 1614 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1615 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1616 | return 0; |
1617 | #endif | |
1618 | ||
bf89d16f | 1619 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1620 | return 1; |
1621 | return 0; | |
1622 | } | |
1623 | ||
56055d3a AA |
1624 | /* |
1625 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1626 | * It always tries to merge towards right. If you want to merge towards | |
1627 | * left, pass "ex - 1" as argument instead of "ex". | |
1628 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1629 | * 1 if they got merged. | |
1630 | */ | |
197217a5 | 1631 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1632 | struct ext4_ext_path *path, |
1633 | struct ext4_extent *ex) | |
56055d3a AA |
1634 | { |
1635 | struct ext4_extent_header *eh; | |
1636 | unsigned int depth, len; | |
1637 | int merge_done = 0; | |
1638 | int uninitialized = 0; | |
1639 | ||
1640 | depth = ext_depth(inode); | |
1641 | BUG_ON(path[depth].p_hdr == NULL); | |
1642 | eh = path[depth].p_hdr; | |
1643 | ||
1644 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1645 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1646 | break; | |
1647 | /* merge with next extent! */ | |
1648 | if (ext4_ext_is_uninitialized(ex)) | |
1649 | uninitialized = 1; | |
1650 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1651 | + ext4_ext_get_actual_len(ex + 1)); | |
1652 | if (uninitialized) | |
1653 | ext4_ext_mark_uninitialized(ex); | |
1654 | ||
1655 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1656 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1657 | * sizeof(struct ext4_extent); | |
1658 | memmove(ex + 1, ex + 2, len); | |
1659 | } | |
e8546d06 | 1660 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1661 | merge_done = 1; |
1662 | WARN_ON(eh->eh_entries == 0); | |
1663 | if (!eh->eh_entries) | |
24676da4 | 1664 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1665 | } |
1666 | ||
1667 | return merge_done; | |
1668 | } | |
1669 | ||
ecb94f5f TT |
1670 | /* |
1671 | * This function does a very simple check to see if we can collapse | |
1672 | * an extent tree with a single extent tree leaf block into the inode. | |
1673 | */ | |
1674 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1675 | struct inode *inode, | |
1676 | struct ext4_ext_path *path) | |
1677 | { | |
1678 | size_t s; | |
1679 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1680 | ext4_fsblk_t blk; | |
1681 | ||
1682 | if ((path[0].p_depth != 1) || | |
1683 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1684 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1685 | return; | |
1686 | ||
1687 | /* | |
1688 | * We need to modify the block allocation bitmap and the block | |
1689 | * group descriptor to release the extent tree block. If we | |
1690 | * can't get the journal credits, give up. | |
1691 | */ | |
1692 | if (ext4_journal_extend(handle, 2)) | |
1693 | return; | |
1694 | ||
1695 | /* | |
1696 | * Copy the extent data up to the inode | |
1697 | */ | |
1698 | blk = ext4_idx_pblock(path[0].p_idx); | |
1699 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1700 | sizeof(struct ext4_extent_idx); | |
1701 | s += sizeof(struct ext4_extent_header); | |
1702 | ||
1703 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1704 | path[0].p_depth = 0; | |
1705 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1706 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1707 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1708 | ||
1709 | brelse(path[1].p_bh); | |
1710 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
1711 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); | |
1712 | } | |
1713 | ||
197217a5 YY |
1714 | /* |
1715 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1716 | * return 1 if merge left else 0. | |
1717 | */ | |
ecb94f5f TT |
1718 | static void ext4_ext_try_to_merge(handle_t *handle, |
1719 | struct inode *inode, | |
197217a5 YY |
1720 | struct ext4_ext_path *path, |
1721 | struct ext4_extent *ex) { | |
1722 | struct ext4_extent_header *eh; | |
1723 | unsigned int depth; | |
1724 | int merge_done = 0; | |
197217a5 YY |
1725 | |
1726 | depth = ext_depth(inode); | |
1727 | BUG_ON(path[depth].p_hdr == NULL); | |
1728 | eh = path[depth].p_hdr; | |
1729 | ||
1730 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1731 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1732 | ||
1733 | if (!merge_done) | |
ecb94f5f | 1734 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1735 | |
ecb94f5f | 1736 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1737 | } |
1738 | ||
25d14f98 AA |
1739 | /* |
1740 | * check if a portion of the "newext" extent overlaps with an | |
1741 | * existing extent. | |
1742 | * | |
1743 | * If there is an overlap discovered, it updates the length of the newext | |
1744 | * such that there will be no overlap, and then returns 1. | |
1745 | * If there is no overlap found, it returns 0. | |
1746 | */ | |
4d33b1ef TT |
1747 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1748 | struct inode *inode, | |
1f109d5a TT |
1749 | struct ext4_extent *newext, |
1750 | struct ext4_ext_path *path) | |
25d14f98 | 1751 | { |
725d26d3 | 1752 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1753 | unsigned int depth, len1; |
1754 | unsigned int ret = 0; | |
1755 | ||
1756 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1757 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1758 | depth = ext_depth(inode); |
1759 | if (!path[depth].p_ext) | |
1760 | goto out; | |
1761 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1762 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1763 | |
1764 | /* | |
1765 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1766 | * is before the requested block(s) |
25d14f98 AA |
1767 | */ |
1768 | if (b2 < b1) { | |
1769 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1770 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1771 | goto out; |
4d33b1ef | 1772 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1773 | } |
1774 | ||
725d26d3 | 1775 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1776 | if (b1 + len1 < b1) { |
f17722f9 | 1777 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1778 | newext->ee_len = cpu_to_le16(len1); |
1779 | ret = 1; | |
1780 | } | |
1781 | ||
1782 | /* check for overlap */ | |
1783 | if (b1 + len1 > b2) { | |
1784 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1785 | ret = 1; | |
1786 | } | |
1787 | out: | |
1788 | return ret; | |
1789 | } | |
1790 | ||
a86c6181 | 1791 | /* |
d0d856e8 RD |
1792 | * ext4_ext_insert_extent: |
1793 | * tries to merge requsted extent into the existing extent or | |
1794 | * inserts requested extent as new one into the tree, | |
1795 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1796 | */ |
1797 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1798 | struct ext4_ext_path *path, | |
0031462b | 1799 | struct ext4_extent *newext, int flag) |
a86c6181 | 1800 | { |
af5bc92d | 1801 | struct ext4_extent_header *eh; |
a86c6181 AT |
1802 | struct ext4_extent *ex, *fex; |
1803 | struct ext4_extent *nearex; /* nearest extent */ | |
1804 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1805 | int depth, len, err; |
1806 | ext4_lblk_t next; | |
a2df2a63 | 1807 | unsigned uninitialized = 0; |
55f020db | 1808 | int flags = 0; |
a86c6181 | 1809 | |
273df556 FM |
1810 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1811 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1812 | return -EIO; | |
1813 | } | |
a86c6181 AT |
1814 | depth = ext_depth(inode); |
1815 | ex = path[depth].p_ext; | |
be8981be | 1816 | eh = path[depth].p_hdr; |
273df556 FM |
1817 | if (unlikely(path[depth].p_hdr == NULL)) { |
1818 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1819 | return -EIO; | |
1820 | } | |
a86c6181 AT |
1821 | |
1822 | /* try to insert block into found extent and return */ | |
be8981be | 1823 | if (ex && !(flag & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1824 | |
1825 | /* | |
be8981be LC |
1826 | * Try to see whether we should rather test the extent on |
1827 | * right from ex, or from the left of ex. This is because | |
1828 | * ext4_ext_find_extent() can return either extent on the | |
1829 | * left, or on the right from the searched position. This | |
1830 | * will make merging more effective. | |
a2df2a63 | 1831 | */ |
be8981be LC |
1832 | if (ex < EXT_LAST_EXTENT(eh) && |
1833 | (le32_to_cpu(ex->ee_block) + | |
1834 | ext4_ext_get_actual_len(ex) < | |
1835 | le32_to_cpu(newext->ee_block))) { | |
1836 | ex += 1; | |
1837 | goto prepend; | |
1838 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1839 | (le32_to_cpu(newext->ee_block) + | |
1840 | ext4_ext_get_actual_len(newext) < | |
1841 | le32_to_cpu(ex->ee_block))) | |
1842 | ex -= 1; | |
1843 | ||
1844 | /* Try to append newex to the ex */ | |
1845 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1846 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
1847 | "(from %llu)\n", | |
1848 | ext4_ext_is_uninitialized(newext), | |
1849 | ext4_ext_get_actual_len(newext), | |
1850 | le32_to_cpu(ex->ee_block), | |
1851 | ext4_ext_is_uninitialized(ex), | |
1852 | ext4_ext_get_actual_len(ex), | |
1853 | ext4_ext_pblock(ex)); | |
1854 | err = ext4_ext_get_access(handle, inode, | |
1855 | path + depth); | |
1856 | if (err) | |
1857 | return err; | |
1858 | ||
1859 | /* | |
1860 | * ext4_can_extents_be_merged should have checked | |
1861 | * that either both extents are uninitialized, or | |
1862 | * both aren't. Thus we need to check only one of | |
1863 | * them here. | |
1864 | */ | |
1865 | if (ext4_ext_is_uninitialized(ex)) | |
1866 | uninitialized = 1; | |
1867 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
a2df2a63 | 1868 | + ext4_ext_get_actual_len(newext)); |
be8981be LC |
1869 | if (uninitialized) |
1870 | ext4_ext_mark_uninitialized(ex); | |
1871 | eh = path[depth].p_hdr; | |
1872 | nearex = ex; | |
1873 | goto merge; | |
1874 | } | |
1875 | ||
1876 | prepend: | |
1877 | /* Try to prepend newex to the ex */ | |
1878 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
1879 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
1880 | "(from %llu)\n", | |
1881 | le32_to_cpu(newext->ee_block), | |
1882 | ext4_ext_is_uninitialized(newext), | |
1883 | ext4_ext_get_actual_len(newext), | |
1884 | le32_to_cpu(ex->ee_block), | |
1885 | ext4_ext_is_uninitialized(ex), | |
1886 | ext4_ext_get_actual_len(ex), | |
1887 | ext4_ext_pblock(ex)); | |
1888 | err = ext4_ext_get_access(handle, inode, | |
1889 | path + depth); | |
1890 | if (err) | |
1891 | return err; | |
1892 | ||
1893 | /* | |
1894 | * ext4_can_extents_be_merged should have checked | |
1895 | * that either both extents are uninitialized, or | |
1896 | * both aren't. Thus we need to check only one of | |
1897 | * them here. | |
1898 | */ | |
1899 | if (ext4_ext_is_uninitialized(ex)) | |
1900 | uninitialized = 1; | |
1901 | ex->ee_block = newext->ee_block; | |
1902 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
1903 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1904 | + ext4_ext_get_actual_len(newext)); | |
1905 | if (uninitialized) | |
1906 | ext4_ext_mark_uninitialized(ex); | |
1907 | eh = path[depth].p_hdr; | |
1908 | nearex = ex; | |
1909 | goto merge; | |
1910 | } | |
a86c6181 AT |
1911 | } |
1912 | ||
a86c6181 AT |
1913 | depth = ext_depth(inode); |
1914 | eh = path[depth].p_hdr; | |
1915 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1916 | goto has_space; | |
1917 | ||
1918 | /* probably next leaf has space for us? */ | |
1919 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1920 | next = EXT_MAX_BLOCKS; |
1921 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1922 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1923 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1924 | ext_debug("next leaf block - %u\n", next); |
a86c6181 AT |
1925 | BUG_ON(npath != NULL); |
1926 | npath = ext4_ext_find_extent(inode, next, NULL); | |
1927 | if (IS_ERR(npath)) | |
1928 | return PTR_ERR(npath); | |
1929 | BUG_ON(npath->p_depth != path->p_depth); | |
1930 | eh = npath[depth].p_hdr; | |
1931 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1932 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1933 | le16_to_cpu(eh->eh_entries)); |
1934 | path = npath; | |
ffb505ff | 1935 | goto has_space; |
a86c6181 AT |
1936 | } |
1937 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1938 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1939 | } | |
1940 | ||
1941 | /* | |
d0d856e8 RD |
1942 | * There is no free space in the found leaf. |
1943 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1944 | */ |
55f020db AH |
1945 | if (flag & EXT4_GET_BLOCKS_PUNCH_OUT_EXT) |
1946 | flags = EXT4_MB_USE_ROOT_BLOCKS; | |
1947 | err = ext4_ext_create_new_leaf(handle, inode, flags, path, newext); | |
a86c6181 AT |
1948 | if (err) |
1949 | goto cleanup; | |
1950 | depth = ext_depth(inode); | |
1951 | eh = path[depth].p_hdr; | |
1952 | ||
1953 | has_space: | |
1954 | nearex = path[depth].p_ext; | |
1955 | ||
7e028976 AM |
1956 | err = ext4_ext_get_access(handle, inode, path + depth); |
1957 | if (err) | |
a86c6181 AT |
1958 | goto cleanup; |
1959 | ||
1960 | if (!nearex) { | |
1961 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1962 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1963 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1964 | ext4_ext_pblock(newext), |
553f9008 | 1965 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1966 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1967 | nearex = EXT_FIRST_EXTENT(eh); |
1968 | } else { | |
1969 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1970 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1971 | /* Insert after */ |
32de6756 YY |
1972 | ext_debug("insert %u:%llu:[%d]%d before: " |
1973 | "nearest %p\n", | |
80e675f9 EG |
1974 | le32_to_cpu(newext->ee_block), |
1975 | ext4_ext_pblock(newext), | |
1976 | ext4_ext_is_uninitialized(newext), | |
1977 | ext4_ext_get_actual_len(newext), | |
1978 | nearex); | |
1979 | nearex++; | |
1980 | } else { | |
1981 | /* Insert before */ | |
1982 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
1983 | ext_debug("insert %u:%llu:[%d]%d after: " |
1984 | "nearest %p\n", | |
8c55e204 | 1985 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1986 | ext4_ext_pblock(newext), |
553f9008 | 1987 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1988 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
1989 | nearex); |
1990 | } | |
1991 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
1992 | if (len > 0) { | |
32de6756 | 1993 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
1994 | "move %d extents from 0x%p to 0x%p\n", |
1995 | le32_to_cpu(newext->ee_block), | |
1996 | ext4_ext_pblock(newext), | |
1997 | ext4_ext_is_uninitialized(newext), | |
1998 | ext4_ext_get_actual_len(newext), | |
1999 | len, nearex, nearex + 1); | |
2000 | memmove(nearex + 1, nearex, | |
2001 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2002 | } |
a86c6181 AT |
2003 | } |
2004 | ||
e8546d06 | 2005 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2006 | path[depth].p_ext = nearex; |
a86c6181 | 2007 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2008 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2009 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2010 | |
2011 | merge: | |
e7bcf823 | 2012 | /* try to merge extents */ |
744692dc | 2013 | if (!(flag & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2014 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2015 | |
a86c6181 AT |
2016 | |
2017 | /* time to correct all indexes above */ | |
2018 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2019 | if (err) | |
2020 | goto cleanup; | |
2021 | ||
ecb94f5f | 2022 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2023 | |
2024 | cleanup: | |
2025 | if (npath) { | |
2026 | ext4_ext_drop_refs(npath); | |
2027 | kfree(npath); | |
2028 | } | |
a86c6181 AT |
2029 | return err; |
2030 | } | |
2031 | ||
91dd8c11 LC |
2032 | static int ext4_fill_fiemap_extents(struct inode *inode, |
2033 | ext4_lblk_t block, ext4_lblk_t num, | |
2034 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
2035 | { |
2036 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 2037 | struct ext4_extent *ex; |
69eb33dc | 2038 | struct extent_status es; |
91dd8c11 | 2039 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 2040 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
2041 | int exists, depth = 0, err = 0; |
2042 | unsigned int flags = 0; | |
2043 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 2044 | |
f17722f9 | 2045 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
2046 | num = last - block; |
2047 | /* find extent for this block */ | |
fab3a549 | 2048 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
2049 | |
2050 | if (path && ext_depth(inode) != depth) { | |
2051 | /* depth was changed. we have to realloc path */ | |
2052 | kfree(path); | |
2053 | path = NULL; | |
2054 | } | |
2055 | ||
6873fa0d ES |
2056 | path = ext4_ext_find_extent(inode, block, path); |
2057 | if (IS_ERR(path)) { | |
91dd8c11 | 2058 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
2059 | err = PTR_ERR(path); |
2060 | path = NULL; | |
2061 | break; | |
2062 | } | |
2063 | ||
2064 | depth = ext_depth(inode); | |
273df556 | 2065 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2066 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2067 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2068 | err = -EIO; | |
2069 | break; | |
2070 | } | |
6873fa0d ES |
2071 | ex = path[depth].p_ext; |
2072 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2073 | ext4_ext_drop_refs(path); |
6873fa0d | 2074 | |
91dd8c11 | 2075 | flags = 0; |
6873fa0d ES |
2076 | exists = 0; |
2077 | if (!ex) { | |
2078 | /* there is no extent yet, so try to allocate | |
2079 | * all requested space */ | |
2080 | start = block; | |
2081 | end = block + num; | |
2082 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2083 | /* need to allocate space before found extent */ | |
2084 | start = block; | |
2085 | end = le32_to_cpu(ex->ee_block); | |
2086 | if (block + num < end) | |
2087 | end = block + num; | |
2088 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2089 | + ext4_ext_get_actual_len(ex)) { | |
2090 | /* need to allocate space after found extent */ | |
2091 | start = block; | |
2092 | end = block + num; | |
2093 | if (end >= next) | |
2094 | end = next; | |
2095 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2096 | /* | |
2097 | * some part of requested space is covered | |
2098 | * by found extent | |
2099 | */ | |
2100 | start = block; | |
2101 | end = le32_to_cpu(ex->ee_block) | |
2102 | + ext4_ext_get_actual_len(ex); | |
2103 | if (block + num < end) | |
2104 | end = block + num; | |
2105 | exists = 1; | |
2106 | } else { | |
2107 | BUG(); | |
2108 | } | |
2109 | BUG_ON(end <= start); | |
2110 | ||
2111 | if (!exists) { | |
69eb33dc ZL |
2112 | es.es_lblk = start; |
2113 | es.es_len = end - start; | |
2114 | es.es_pblk = 0; | |
6873fa0d | 2115 | } else { |
69eb33dc ZL |
2116 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2117 | es.es_len = ext4_ext_get_actual_len(ex); | |
2118 | es.es_pblk = ext4_ext_pblock(ex); | |
91dd8c11 LC |
2119 | if (ext4_ext_is_uninitialized(ex)) |
2120 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
6873fa0d ES |
2121 | } |
2122 | ||
91dd8c11 | 2123 | /* |
69eb33dc ZL |
2124 | * Find delayed extent and update es accordingly. We call |
2125 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2126 | * last existing extent or not. |
2127 | */ | |
69eb33dc | 2128 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2129 | if (!exists && next_del) { |
2130 | exists = 1; | |
2131 | flags |= FIEMAP_EXTENT_DELALLOC; | |
2132 | } | |
2133 | up_read(&EXT4_I(inode)->i_data_sem); | |
2134 | ||
69eb33dc ZL |
2135 | if (unlikely(es.es_len == 0)) { |
2136 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
273df556 FM |
2137 | err = -EIO; |
2138 | break; | |
2139 | } | |
6873fa0d | 2140 | |
f7fec032 ZL |
2141 | /* |
2142 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2143 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2144 | * possible that an extent is with unwritten and delayed | |
2145 | * status due to when an extent is delayed allocated and | |
2146 | * is allocated by fallocate status tree will track both of | |
2147 | * them in a extent. | |
2148 | * | |
2149 | * So we could return a unwritten and delayed extent, and | |
2150 | * its block is equal to 'next'. | |
2151 | */ | |
2152 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2153 | flags |= FIEMAP_EXTENT_LAST; |
2154 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2155 | next != EXT_MAX_BLOCKS)) { | |
2156 | EXT4_ERROR_INODE(inode, | |
2157 | "next extent == %u, next " | |
2158 | "delalloc extent = %u", | |
2159 | next, next_del); | |
2160 | err = -EIO; | |
2161 | break; | |
2162 | } | |
6873fa0d ES |
2163 | } |
2164 | ||
91dd8c11 LC |
2165 | if (exists) { |
2166 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2167 | (__u64)es.es_lblk << blksize_bits, |
2168 | (__u64)es.es_pblk << blksize_bits, | |
2169 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2170 | flags); |
2171 | if (err < 0) | |
2172 | break; | |
2173 | if (err == 1) { | |
2174 | err = 0; | |
2175 | break; | |
2176 | } | |
6873fa0d ES |
2177 | } |
2178 | ||
69eb33dc | 2179 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2180 | } |
2181 | ||
2182 | if (path) { | |
2183 | ext4_ext_drop_refs(path); | |
2184 | kfree(path); | |
2185 | } | |
2186 | ||
2187 | return err; | |
2188 | } | |
2189 | ||
a86c6181 | 2190 | /* |
d0d856e8 RD |
2191 | * ext4_ext_put_gap_in_cache: |
2192 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2193 | * and cache this gap |
2194 | */ | |
09b88252 | 2195 | static void |
a86c6181 | 2196 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2197 | ext4_lblk_t block) |
a86c6181 AT |
2198 | { |
2199 | int depth = ext_depth(inode); | |
725d26d3 AK |
2200 | unsigned long len; |
2201 | ext4_lblk_t lblock; | |
a86c6181 AT |
2202 | struct ext4_extent *ex; |
2203 | ||
2204 | ex = path[depth].p_ext; | |
2205 | if (ex == NULL) { | |
69eb33dc ZL |
2206 | /* |
2207 | * there is no extent yet, so gap is [0;-] and we | |
2208 | * don't cache it | |
2209 | */ | |
a86c6181 AT |
2210 | ext_debug("cache gap(whole file):"); |
2211 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2212 | lblock = block; | |
2213 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2214 | ext_debug("cache gap(before): %u [%u:%u]", |
2215 | block, | |
2216 | le32_to_cpu(ex->ee_block), | |
2217 | ext4_ext_get_actual_len(ex)); | |
d100eef2 ZL |
2218 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2219 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2220 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2221 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2222 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2223 | ext4_lblk_t next; |
8c55e204 | 2224 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2225 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2226 | |
2227 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2228 | ext_debug("cache gap(after): [%u:%u] %u", |
2229 | le32_to_cpu(ex->ee_block), | |
2230 | ext4_ext_get_actual_len(ex), | |
2231 | block); | |
725d26d3 AK |
2232 | BUG_ON(next == lblock); |
2233 | len = next - lblock; | |
d100eef2 ZL |
2234 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2235 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2236 | EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2237 | } else { |
2238 | lblock = len = 0; | |
2239 | BUG(); | |
2240 | } | |
2241 | ||
bba90743 | 2242 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
2243 | } |
2244 | ||
2245 | /* | |
d0d856e8 RD |
2246 | * ext4_ext_rm_idx: |
2247 | * removes index from the index block. | |
a86c6181 | 2248 | */ |
1d03ec98 | 2249 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2250 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2251 | { |
a86c6181 | 2252 | int err; |
f65e6fba | 2253 | ext4_fsblk_t leaf; |
a86c6181 AT |
2254 | |
2255 | /* free index block */ | |
c36575e6 FL |
2256 | depth--; |
2257 | path = path + depth; | |
bf89d16f | 2258 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2259 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2260 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2261 | return -EIO; | |
2262 | } | |
7e028976 AM |
2263 | err = ext4_ext_get_access(handle, inode, path); |
2264 | if (err) | |
a86c6181 | 2265 | return err; |
0e1147b0 RD |
2266 | |
2267 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2268 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2269 | len *= sizeof(struct ext4_extent_idx); | |
2270 | memmove(path->p_idx, path->p_idx + 1, len); | |
2271 | } | |
2272 | ||
e8546d06 | 2273 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2274 | err = ext4_ext_dirty(handle, inode, path); |
2275 | if (err) | |
a86c6181 | 2276 | return err; |
2ae02107 | 2277 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2278 | trace_ext4_ext_rm_idx(inode, leaf); |
2279 | ||
7dc57615 | 2280 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2281 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2282 | |
2283 | while (--depth >= 0) { | |
2284 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2285 | break; | |
2286 | path--; | |
2287 | err = ext4_ext_get_access(handle, inode, path); | |
2288 | if (err) | |
2289 | break; | |
2290 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2291 | err = ext4_ext_dirty(handle, inode, path); | |
2292 | if (err) | |
2293 | break; | |
2294 | } | |
a86c6181 AT |
2295 | return err; |
2296 | } | |
2297 | ||
2298 | /* | |
ee12b630 MC |
2299 | * ext4_ext_calc_credits_for_single_extent: |
2300 | * This routine returns max. credits that needed to insert an extent | |
2301 | * to the extent tree. | |
2302 | * When pass the actual path, the caller should calculate credits | |
2303 | * under i_data_sem. | |
a86c6181 | 2304 | */ |
525f4ed8 | 2305 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2306 | struct ext4_ext_path *path) |
2307 | { | |
a86c6181 | 2308 | if (path) { |
ee12b630 | 2309 | int depth = ext_depth(inode); |
f3bd1f3f | 2310 | int ret = 0; |
ee12b630 | 2311 | |
a86c6181 | 2312 | /* probably there is space in leaf? */ |
a86c6181 | 2313 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2314 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2315 | |
ee12b630 MC |
2316 | /* |
2317 | * There are some space in the leaf tree, no | |
2318 | * need to account for leaf block credit | |
2319 | * | |
2320 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2321 | * and other metadata blocks still need to be |
ee12b630 MC |
2322 | * accounted. |
2323 | */ | |
525f4ed8 | 2324 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2325 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2326 | return ret; |
ee12b630 MC |
2327 | } |
2328 | } | |
a86c6181 | 2329 | |
525f4ed8 | 2330 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2331 | } |
a86c6181 | 2332 | |
ee12b630 MC |
2333 | /* |
2334 | * How many index/leaf blocks need to change/allocate to modify nrblocks? | |
2335 | * | |
2336 | * if nrblocks are fit in a single extent (chunk flag is 1), then | |
2337 | * in the worse case, each tree level index/leaf need to be changed | |
2338 | * if the tree split due to insert a new extent, then the old tree | |
2339 | * index/leaf need to be updated too | |
2340 | * | |
2341 | * If the nrblocks are discontiguous, they could cause | |
2342 | * the whole tree split more than once, but this is really rare. | |
2343 | */ | |
525f4ed8 | 2344 | int ext4_ext_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
ee12b630 MC |
2345 | { |
2346 | int index; | |
f19d5870 TM |
2347 | int depth; |
2348 | ||
2349 | /* If we are converting the inline data, only one is needed here. */ | |
2350 | if (ext4_has_inline_data(inode)) | |
2351 | return 1; | |
2352 | ||
2353 | depth = ext_depth(inode); | |
a86c6181 | 2354 | |
ee12b630 MC |
2355 | if (chunk) |
2356 | index = depth * 2; | |
2357 | else | |
2358 | index = depth * 3; | |
a86c6181 | 2359 | |
ee12b630 | 2360 | return index; |
a86c6181 AT |
2361 | } |
2362 | ||
2363 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2364 | struct ext4_extent *ex, |
2365 | ext4_fsblk_t *partial_cluster, | |
2366 | ext4_lblk_t from, ext4_lblk_t to) | |
a86c6181 | 2367 | { |
0aa06000 | 2368 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2369 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2370 | ext4_fsblk_t pblk; |
18888cf0 | 2371 | int flags = 0; |
a86c6181 | 2372 | |
c9de560d | 2373 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) |
18888cf0 AS |
2374 | flags |= EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2375 | else if (ext4_should_journal_data(inode)) | |
2376 | flags |= EXT4_FREE_BLOCKS_FORGET; | |
2377 | ||
0aa06000 TT |
2378 | /* |
2379 | * For bigalloc file systems, we never free a partial cluster | |
2380 | * at the beginning of the extent. Instead, we make a note | |
2381 | * that we tried freeing the cluster, and check to see if we | |
2382 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2383 | * or at the end of the ext4_truncate() operation. | |
2384 | */ | |
2385 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2386 | ||
d8990240 | 2387 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2388 | /* |
2389 | * If we have a partial cluster, and it's different from the | |
2390 | * cluster of the last block, we need to explicitly free the | |
2391 | * partial cluster here. | |
2392 | */ | |
2393 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
2394 | if (*partial_cluster && (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
2395 | ext4_free_blocks(handle, inode, NULL, | |
2396 | EXT4_C2B(sbi, *partial_cluster), | |
2397 | sbi->s_cluster_ratio, flags); | |
2398 | *partial_cluster = 0; | |
2399 | } | |
2400 | ||
a86c6181 AT |
2401 | #ifdef EXTENTS_STATS |
2402 | { | |
2403 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2404 | spin_lock(&sbi->s_ext_stats_lock); |
2405 | sbi->s_ext_blocks += ee_len; | |
2406 | sbi->s_ext_extents++; | |
2407 | if (ee_len < sbi->s_ext_min) | |
2408 | sbi->s_ext_min = ee_len; | |
2409 | if (ee_len > sbi->s_ext_max) | |
2410 | sbi->s_ext_max = ee_len; | |
2411 | if (ext_depth(inode) > sbi->s_depth_max) | |
2412 | sbi->s_depth_max = ext_depth(inode); | |
2413 | spin_unlock(&sbi->s_ext_stats_lock); | |
2414 | } | |
2415 | #endif | |
2416 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2417 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2418 | /* tail removal */ |
725d26d3 | 2419 | ext4_lblk_t num; |
725d26d3 | 2420 | |
a2df2a63 | 2421 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 TT |
2422 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
2423 | ext_debug("free last %u blocks starting %llu\n", num, pblk); | |
2424 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2425 | /* | |
2426 | * If the block range to be freed didn't start at the | |
2427 | * beginning of a cluster, and we removed the entire | |
2428 | * extent, save the partial cluster here, since we | |
2429 | * might need to delete if we determine that the | |
2430 | * truncate operation has removed all of the blocks in | |
2431 | * the cluster. | |
2432 | */ | |
2433 | if (pblk & (sbi->s_cluster_ratio - 1) && | |
2434 | (ee_len == num)) | |
2435 | *partial_cluster = EXT4_B2C(sbi, pblk); | |
2436 | else | |
2437 | *partial_cluster = 0; | |
a86c6181 | 2438 | } else if (from == le32_to_cpu(ex->ee_block) |
a2df2a63 | 2439 | && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) { |
d583fb87 AH |
2440 | /* head removal */ |
2441 | ext4_lblk_t num; | |
2442 | ext4_fsblk_t start; | |
2443 | ||
2444 | num = to - from; | |
2445 | start = ext4_ext_pblock(ex); | |
2446 | ||
2447 | ext_debug("free first %u blocks starting %llu\n", num, start); | |
ee90d57e | 2448 | ext4_free_blocks(handle, inode, NULL, start, num, flags); |
d583fb87 | 2449 | |
a86c6181 | 2450 | } else { |
725d26d3 AK |
2451 | printk(KERN_INFO "strange request: removal(2) " |
2452 | "%u-%u from %u:%u\n", | |
2453 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2454 | } |
2455 | return 0; | |
2456 | } | |
2457 | ||
d583fb87 AH |
2458 | |
2459 | /* | |
2460 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2461 | * blocks appearing between "start" and "end", and splits the extents | |
2462 | * if "start" and "end" appear in the same extent | |
2463 | * | |
2464 | * @handle: The journal handle | |
2465 | * @inode: The files inode | |
2466 | * @path: The path to the leaf | |
2467 | * @start: The first block to remove | |
2468 | * @end: The last block to remove | |
2469 | */ | |
a86c6181 AT |
2470 | static int |
2471 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
0aa06000 TT |
2472 | struct ext4_ext_path *path, ext4_fsblk_t *partial_cluster, |
2473 | ext4_lblk_t start, ext4_lblk_t end) | |
a86c6181 | 2474 | { |
0aa06000 | 2475 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2476 | int err = 0, correct_index = 0; |
2477 | int depth = ext_depth(inode), credits; | |
2478 | struct ext4_extent_header *eh; | |
750c9c47 | 2479 | ext4_lblk_t a, b; |
725d26d3 AK |
2480 | unsigned num; |
2481 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2482 | unsigned short ex_ee_len; |
a2df2a63 | 2483 | unsigned uninitialized = 0; |
a86c6181 AT |
2484 | struct ext4_extent *ex; |
2485 | ||
c29c0ae7 | 2486 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2487 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2488 | if (!path[depth].p_hdr) |
2489 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2490 | eh = path[depth].p_hdr; | |
273df556 FM |
2491 | if (unlikely(path[depth].p_hdr == NULL)) { |
2492 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2493 | return -EIO; | |
2494 | } | |
a86c6181 AT |
2495 | /* find where to start removing */ |
2496 | ex = EXT_LAST_EXTENT(eh); | |
2497 | ||
2498 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2499 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2500 | |
d8990240 AK |
2501 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2502 | ||
a86c6181 AT |
2503 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2504 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2505 | |
2506 | if (ext4_ext_is_uninitialized(ex)) | |
2507 | uninitialized = 1; | |
2508 | else | |
2509 | uninitialized = 0; | |
2510 | ||
553f9008 M |
2511 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2512 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2513 | path[depth].p_ext = ex; |
2514 | ||
2515 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2516 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2517 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2518 | |
2519 | ext_debug(" border %u:%u\n", a, b); | |
2520 | ||
d583fb87 | 2521 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2522 | if (end < ex_ee_block) { |
d583fb87 AH |
2523 | ex--; |
2524 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2525 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2526 | continue; | |
750c9c47 | 2527 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2528 | EXT4_ERROR_INODE(inode, |
2529 | "can not handle truncate %u:%u " | |
2530 | "on extent %u:%u", | |
2531 | start, end, ex_ee_block, | |
2532 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2533 | err = -EIO; |
2534 | goto out; | |
a86c6181 AT |
2535 | } else if (a != ex_ee_block) { |
2536 | /* remove tail of the extent */ | |
750c9c47 | 2537 | num = a - ex_ee_block; |
a86c6181 AT |
2538 | } else { |
2539 | /* remove whole extent: excellent! */ | |
a86c6181 | 2540 | num = 0; |
a86c6181 | 2541 | } |
34071da7 TT |
2542 | /* |
2543 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2544 | * descriptor) for each block group; assume two block | |
2545 | * groups plus ex_ee_len/blocks_per_block_group for | |
2546 | * the worst case | |
2547 | */ | |
2548 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2549 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2550 | correct_index = 1; | |
2551 | credits += (ext_depth(inode)) + 1; | |
2552 | } | |
5aca07eb | 2553 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2554 | |
487caeef | 2555 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2556 | if (err) |
a86c6181 | 2557 | goto out; |
a86c6181 AT |
2558 | |
2559 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2560 | if (err) | |
2561 | goto out; | |
2562 | ||
0aa06000 TT |
2563 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2564 | a, b); | |
a86c6181 AT |
2565 | if (err) |
2566 | goto out; | |
2567 | ||
750c9c47 | 2568 | if (num == 0) |
d0d856e8 | 2569 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2570 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2571 | |
a86c6181 | 2572 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2573 | /* |
2574 | * Do not mark uninitialized if all the blocks in the | |
2575 | * extent have been removed. | |
2576 | */ | |
2577 | if (uninitialized && num) | |
a2df2a63 | 2578 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2579 | /* |
2580 | * If the extent was completely released, | |
2581 | * we need to remove it from the leaf | |
2582 | */ | |
2583 | if (num == 0) { | |
f17722f9 | 2584 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2585 | /* |
2586 | * For hole punching, we need to scoot all the | |
2587 | * extents up when an extent is removed so that | |
2588 | * we dont have blank extents in the middle | |
2589 | */ | |
2590 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2591 | sizeof(struct ext4_extent)); | |
2592 | ||
2593 | /* Now get rid of the one at the end */ | |
2594 | memset(EXT_LAST_EXTENT(eh), 0, | |
2595 | sizeof(struct ext4_extent)); | |
2596 | } | |
2597 | le16_add_cpu(&eh->eh_entries, -1); | |
0aa06000 TT |
2598 | } else |
2599 | *partial_cluster = 0; | |
d583fb87 | 2600 | |
750c9c47 DM |
2601 | err = ext4_ext_dirty(handle, inode, path + depth); |
2602 | if (err) | |
2603 | goto out; | |
2604 | ||
bf52c6f7 | 2605 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2606 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2607 | ex--; |
2608 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2609 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2610 | } |
2611 | ||
2612 | if (correct_index && eh->eh_entries) | |
2613 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2614 | ||
0aa06000 TT |
2615 | /* |
2616 | * If there is still a entry in the leaf node, check to see if | |
2617 | * it references the partial cluster. This is the only place | |
2618 | * where it could; if it doesn't, we can free the cluster. | |
2619 | */ | |
2620 | if (*partial_cluster && ex >= EXT_FIRST_EXTENT(eh) && | |
2621 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != | |
2622 | *partial_cluster)) { | |
2623 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2624 | ||
2625 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2626 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2627 | ||
2628 | ext4_free_blocks(handle, inode, NULL, | |
2629 | EXT4_C2B(sbi, *partial_cluster), | |
2630 | sbi->s_cluster_ratio, flags); | |
2631 | *partial_cluster = 0; | |
2632 | } | |
2633 | ||
a86c6181 AT |
2634 | /* if this leaf is free, then we should |
2635 | * remove it from index block above */ | |
2636 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2637 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2638 | |
2639 | out: | |
2640 | return err; | |
2641 | } | |
2642 | ||
2643 | /* | |
d0d856e8 RD |
2644 | * ext4_ext_more_to_rm: |
2645 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2646 | */ |
09b88252 | 2647 | static int |
a86c6181 AT |
2648 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2649 | { | |
2650 | BUG_ON(path->p_idx == NULL); | |
2651 | ||
2652 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2653 | return 0; | |
2654 | ||
2655 | /* | |
d0d856e8 | 2656 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2657 | * so we have to consider current index for truncation |
2658 | */ | |
2659 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2660 | return 0; | |
2661 | return 1; | |
2662 | } | |
2663 | ||
26a4c0c6 TT |
2664 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2665 | ext4_lblk_t end) | |
a86c6181 AT |
2666 | { |
2667 | struct super_block *sb = inode->i_sb; | |
2668 | int depth = ext_depth(inode); | |
968dee77 | 2669 | struct ext4_ext_path *path = NULL; |
0aa06000 | 2670 | ext4_fsblk_t partial_cluster = 0; |
a86c6181 | 2671 | handle_t *handle; |
6f2080e6 | 2672 | int i = 0, err = 0; |
a86c6181 | 2673 | |
5f95d21f | 2674 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2675 | |
2676 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2677 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2678 | if (IS_ERR(handle)) |
2679 | return PTR_ERR(handle); | |
2680 | ||
0617b83f | 2681 | again: |
d8990240 AK |
2682 | trace_ext4_ext_remove_space(inode, start, depth); |
2683 | ||
5f95d21f LC |
2684 | /* |
2685 | * Check if we are removing extents inside the extent tree. If that | |
2686 | * is the case, we are going to punch a hole inside the extent tree | |
2687 | * so we have to check whether we need to split the extent covering | |
2688 | * the last block to remove so we can easily remove the part of it | |
2689 | * in ext4_ext_rm_leaf(). | |
2690 | */ | |
2691 | if (end < EXT_MAX_BLOCKS - 1) { | |
2692 | struct ext4_extent *ex; | |
2693 | ext4_lblk_t ee_block; | |
2694 | ||
2695 | /* find extent for this block */ | |
2696 | path = ext4_ext_find_extent(inode, end, NULL); | |
2697 | if (IS_ERR(path)) { | |
2698 | ext4_journal_stop(handle); | |
2699 | return PTR_ERR(path); | |
2700 | } | |
2701 | depth = ext_depth(inode); | |
6f2080e6 | 2702 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2703 | ex = path[depth].p_ext; |
968dee77 | 2704 | if (!ex) { |
6f2080e6 DM |
2705 | if (depth) { |
2706 | EXT4_ERROR_INODE(inode, | |
2707 | "path[%d].p_hdr == NULL", | |
2708 | depth); | |
2709 | err = -EIO; | |
2710 | } | |
2711 | goto out; | |
968dee77 | 2712 | } |
5f95d21f LC |
2713 | |
2714 | ee_block = le32_to_cpu(ex->ee_block); | |
2715 | ||
2716 | /* | |
2717 | * See if the last block is inside the extent, if so split | |
2718 | * the extent at 'end' block so we can easily remove the | |
2719 | * tail of the first part of the split extent in | |
2720 | * ext4_ext_rm_leaf(). | |
2721 | */ | |
2722 | if (end >= ee_block && | |
2723 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2724 | int split_flag = 0; | |
2725 | ||
2726 | if (ext4_ext_is_uninitialized(ex)) | |
2727 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2728 | EXT4_EXT_MARK_UNINIT2; | |
2729 | ||
2730 | /* | |
2731 | * Split the extent in two so that 'end' is the last | |
2732 | * block in the first new extent | |
2733 | */ | |
2734 | err = ext4_split_extent_at(handle, inode, path, | |
2735 | end + 1, split_flag, | |
2736 | EXT4_GET_BLOCKS_PRE_IO | | |
2737 | EXT4_GET_BLOCKS_PUNCH_OUT_EXT); | |
2738 | ||
2739 | if (err < 0) | |
2740 | goto out; | |
2741 | } | |
5f95d21f | 2742 | } |
a86c6181 | 2743 | /* |
d0d856e8 RD |
2744 | * We start scanning from right side, freeing all the blocks |
2745 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2746 | */ |
0617b83f | 2747 | depth = ext_depth(inode); |
968dee77 AS |
2748 | if (path) { |
2749 | int k = i = depth; | |
2750 | while (--k > 0) | |
2751 | path[k].p_block = | |
2752 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2753 | } else { | |
2754 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2755 | GFP_NOFS); | |
2756 | if (path == NULL) { | |
2757 | ext4_journal_stop(handle); | |
2758 | return -ENOMEM; | |
2759 | } | |
2760 | path[0].p_depth = depth; | |
2761 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2762 | i = 0; |
5f95d21f | 2763 | |
968dee77 AS |
2764 | if (ext4_ext_check(inode, path[0].p_hdr, depth)) { |
2765 | err = -EIO; | |
2766 | goto out; | |
2767 | } | |
a86c6181 | 2768 | } |
968dee77 | 2769 | err = 0; |
a86c6181 AT |
2770 | |
2771 | while (i >= 0 && err == 0) { | |
2772 | if (i == depth) { | |
2773 | /* this is leaf block */ | |
d583fb87 | 2774 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2775 | &partial_cluster, start, |
5f95d21f | 2776 | end); |
d0d856e8 | 2777 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2778 | brelse(path[i].p_bh); |
2779 | path[i].p_bh = NULL; | |
2780 | i--; | |
2781 | continue; | |
2782 | } | |
2783 | ||
2784 | /* this is index block */ | |
2785 | if (!path[i].p_hdr) { | |
2786 | ext_debug("initialize header\n"); | |
2787 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2788 | } |
2789 | ||
a86c6181 | 2790 | if (!path[i].p_idx) { |
d0d856e8 | 2791 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2792 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2793 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2794 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2795 | path[i].p_hdr, | |
2796 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2797 | } else { | |
d0d856e8 | 2798 | /* we were already here, see at next index */ |
a86c6181 AT |
2799 | path[i].p_idx--; |
2800 | } | |
2801 | ||
2802 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2803 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2804 | path[i].p_idx); | |
2805 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2806 | struct buffer_head *bh; |
a86c6181 | 2807 | /* go to the next level */ |
2ae02107 | 2808 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2809 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2810 | memset(path + i + 1, 0, sizeof(*path)); |
bf89d16f | 2811 | bh = sb_bread(sb, ext4_idx_pblock(path[i].p_idx)); |
c29c0ae7 | 2812 | if (!bh) { |
a86c6181 AT |
2813 | /* should we reset i_size? */ |
2814 | err = -EIO; | |
2815 | break; | |
2816 | } | |
c29c0ae7 AT |
2817 | if (WARN_ON(i + 1 > depth)) { |
2818 | err = -EIO; | |
2819 | break; | |
2820 | } | |
f8489128 DW |
2821 | if (ext4_ext_check_block(inode, ext_block_hdr(bh), |
2822 | depth - i - 1, bh)) { | |
c29c0ae7 AT |
2823 | err = -EIO; |
2824 | break; | |
2825 | } | |
2826 | path[i + 1].p_bh = bh; | |
a86c6181 | 2827 | |
d0d856e8 RD |
2828 | /* save actual number of indexes since this |
2829 | * number is changed at the next iteration */ | |
a86c6181 AT |
2830 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2831 | i++; | |
2832 | } else { | |
d0d856e8 | 2833 | /* we finished processing this index, go up */ |
a86c6181 | 2834 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2835 | /* index is empty, remove it; |
a86c6181 AT |
2836 | * handle must be already prepared by the |
2837 | * truncatei_leaf() */ | |
c36575e6 | 2838 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2839 | } |
d0d856e8 | 2840 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2841 | brelse(path[i].p_bh); |
2842 | path[i].p_bh = NULL; | |
2843 | i--; | |
2844 | ext_debug("return to level %d\n", i); | |
2845 | } | |
2846 | } | |
2847 | ||
d8990240 AK |
2848 | trace_ext4_ext_remove_space_done(inode, start, depth, partial_cluster, |
2849 | path->p_hdr->eh_entries); | |
2850 | ||
7b415bf6 AK |
2851 | /* If we still have something in the partial cluster and we have removed |
2852 | * even the first extent, then we should free the blocks in the partial | |
2853 | * cluster as well. */ | |
2854 | if (partial_cluster && path->p_hdr->eh_entries == 0) { | |
2855 | int flags = EXT4_FREE_BLOCKS_FORGET; | |
2856 | ||
2857 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2858 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
2859 | ||
2860 | ext4_free_blocks(handle, inode, NULL, | |
2861 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2862 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2863 | partial_cluster = 0; | |
2864 | } | |
2865 | ||
a86c6181 AT |
2866 | /* TODO: flexible tree reduction should be here */ |
2867 | if (path->p_hdr->eh_entries == 0) { | |
2868 | /* | |
d0d856e8 RD |
2869 | * truncate to zero freed all the tree, |
2870 | * so we need to correct eh_depth | |
a86c6181 AT |
2871 | */ |
2872 | err = ext4_ext_get_access(handle, inode, path); | |
2873 | if (err == 0) { | |
2874 | ext_inode_hdr(inode)->eh_depth = 0; | |
2875 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2876 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2877 | err = ext4_ext_dirty(handle, inode, path); |
2878 | } | |
2879 | } | |
2880 | out: | |
a86c6181 AT |
2881 | ext4_ext_drop_refs(path); |
2882 | kfree(path); | |
968dee77 AS |
2883 | if (err == -EAGAIN) { |
2884 | path = NULL; | |
0617b83f | 2885 | goto again; |
968dee77 | 2886 | } |
a86c6181 AT |
2887 | ext4_journal_stop(handle); |
2888 | ||
2889 | return err; | |
2890 | } | |
2891 | ||
2892 | /* | |
2893 | * called at mount time | |
2894 | */ | |
2895 | void ext4_ext_init(struct super_block *sb) | |
2896 | { | |
2897 | /* | |
2898 | * possible initialization would be here | |
2899 | */ | |
2900 | ||
83982b6f | 2901 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2902 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2903 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2904 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2905 | ", aggressive tests" |
a86c6181 AT |
2906 | #endif |
2907 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2908 | ", check binsearch" |
a86c6181 AT |
2909 | #endif |
2910 | #ifdef EXTENTS_STATS | |
92b97816 | 2911 | ", stats" |
a86c6181 | 2912 | #endif |
92b97816 | 2913 | "\n"); |
90576c0b | 2914 | #endif |
a86c6181 AT |
2915 | #ifdef EXTENTS_STATS |
2916 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2917 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2918 | EXT4_SB(sb)->s_ext_max = 0; | |
2919 | #endif | |
2920 | } | |
2921 | } | |
2922 | ||
2923 | /* | |
2924 | * called at umount time | |
2925 | */ | |
2926 | void ext4_ext_release(struct super_block *sb) | |
2927 | { | |
83982b6f | 2928 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2929 | return; |
2930 | ||
2931 | #ifdef EXTENTS_STATS | |
2932 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2933 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2934 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2935 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2936 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2937 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2938 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2939 | } | |
2940 | #endif | |
2941 | } | |
2942 | ||
093a088b AK |
2943 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2944 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2945 | { | |
2407518d LC |
2946 | ext4_fsblk_t ee_pblock; |
2947 | unsigned int ee_len; | |
b720303d | 2948 | int ret; |
093a088b | 2949 | |
093a088b | 2950 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2951 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2952 | |
a107e5a3 | 2953 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2954 | if (ret > 0) |
2955 | ret = 0; | |
093a088b | 2956 | |
2407518d | 2957 | return ret; |
093a088b AK |
2958 | } |
2959 | ||
47ea3bb5 YY |
2960 | /* |
2961 | * ext4_split_extent_at() splits an extent at given block. | |
2962 | * | |
2963 | * @handle: the journal handle | |
2964 | * @inode: the file inode | |
2965 | * @path: the path to the extent | |
2966 | * @split: the logical block where the extent is splitted. | |
2967 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
2968 | * the states(init or uninit) of new extents. | |
2969 | * @flags: flags used to insert new extent to extent tree. | |
2970 | * | |
2971 | * | |
2972 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
2973 | * of which are deterimined by split_flag. | |
2974 | * | |
2975 | * There are two cases: | |
2976 | * a> the extent are splitted into two extent. | |
2977 | * b> split is not needed, and just mark the extent. | |
2978 | * | |
2979 | * return 0 on success. | |
2980 | */ | |
2981 | static int ext4_split_extent_at(handle_t *handle, | |
2982 | struct inode *inode, | |
2983 | struct ext4_ext_path *path, | |
2984 | ext4_lblk_t split, | |
2985 | int split_flag, | |
2986 | int flags) | |
2987 | { | |
2988 | ext4_fsblk_t newblock; | |
2989 | ext4_lblk_t ee_block; | |
adb23551 | 2990 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
2991 | struct ext4_extent *ex2 = NULL; |
2992 | unsigned int ee_len, depth; | |
2993 | int err = 0; | |
2994 | ||
dee1f973 DM |
2995 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
2996 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
2997 | ||
47ea3bb5 YY |
2998 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
2999 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3000 | ||
3001 | ext4_ext_show_leaf(inode, path); | |
3002 | ||
3003 | depth = ext_depth(inode); | |
3004 | ex = path[depth].p_ext; | |
3005 | ee_block = le32_to_cpu(ex->ee_block); | |
3006 | ee_len = ext4_ext_get_actual_len(ex); | |
3007 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3008 | ||
3009 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
357b66fd DM |
3010 | BUG_ON(!ext4_ext_is_uninitialized(ex) && |
3011 | split_flag & (EXT4_EXT_MAY_ZEROOUT | | |
3012 | EXT4_EXT_MARK_UNINIT1 | | |
3013 | EXT4_EXT_MARK_UNINIT2)); | |
47ea3bb5 YY |
3014 | |
3015 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3016 | if (err) | |
3017 | goto out; | |
3018 | ||
3019 | if (split == ee_block) { | |
3020 | /* | |
3021 | * case b: block @split is the block that the extent begins with | |
3022 | * then we just change the state of the extent, and splitting | |
3023 | * is not needed. | |
3024 | */ | |
3025 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3026 | ext4_ext_mark_uninitialized(ex); | |
3027 | else | |
3028 | ext4_ext_mark_initialized(ex); | |
3029 | ||
3030 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3031 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3032 | |
ecb94f5f | 3033 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3034 | goto out; |
3035 | } | |
3036 | ||
3037 | /* case a */ | |
3038 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3039 | ex->ee_len = cpu_to_le16(split - ee_block); | |
3040 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
3041 | ext4_ext_mark_uninitialized(ex); | |
3042 | ||
3043 | /* | |
3044 | * path may lead to new leaf, not to original leaf any more | |
3045 | * after ext4_ext_insert_extent() returns, | |
3046 | */ | |
3047 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3048 | if (err) | |
3049 | goto fix_extent_len; | |
3050 | ||
3051 | ex2 = &newex; | |
3052 | ex2->ee_block = cpu_to_le32(split); | |
3053 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3054 | ext4_ext_store_pblock(ex2, newblock); | |
3055 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3056 | ext4_ext_mark_uninitialized(ex2); | |
3057 | ||
3058 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
3059 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 | 3060 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3061 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3062 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3063 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3064 | zero_ex.ee_len = cpu_to_le16( |
3065 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3066 | ext4_ext_store_pblock(&zero_ex, |
3067 | ext4_ext_pblock(ex2)); | |
3068 | } else { | |
dee1f973 | 3069 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3070 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3071 | zero_ex.ee_len = cpu_to_le16( |
3072 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3073 | ext4_ext_store_pblock(&zero_ex, |
3074 | ext4_ext_pblock(ex)); | |
3075 | } | |
3076 | } else { | |
dee1f973 | 3077 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3078 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3079 | zero_ex.ee_len = cpu_to_le16( |
3080 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3081 | ext4_ext_store_pblock(&zero_ex, |
3082 | ext4_ext_pblock(&orig_ex)); | |
3083 | } | |
dee1f973 | 3084 | |
47ea3bb5 YY |
3085 | if (err) |
3086 | goto fix_extent_len; | |
3087 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3088 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3089 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3090 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3091 | if (err) |
3092 | goto fix_extent_len; | |
3093 | ||
3094 | /* update extent status tree */ | |
3095 | err = ext4_es_zeroout(inode, &zero_ex); | |
3096 | ||
47ea3bb5 YY |
3097 | goto out; |
3098 | } else if (err) | |
3099 | goto fix_extent_len; | |
3100 | ||
3101 | out: | |
3102 | ext4_ext_show_leaf(inode, path); | |
3103 | return err; | |
3104 | ||
3105 | fix_extent_len: | |
3106 | ex->ee_len = orig_ex.ee_len; | |
3107 | ext4_ext_dirty(handle, inode, path + depth); | |
3108 | return err; | |
3109 | } | |
3110 | ||
3111 | /* | |
3112 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3113 | * by @map as split_flags indicates | |
3114 | * | |
3115 | * It may result in splitting the extent into multiple extents (upto three) | |
3116 | * There are three possibilities: | |
3117 | * a> There is no split required | |
3118 | * b> Splits in two extents: Split is happening at either end of the extent | |
3119 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3120 | * | |
3121 | */ | |
3122 | static int ext4_split_extent(handle_t *handle, | |
3123 | struct inode *inode, | |
3124 | struct ext4_ext_path *path, | |
3125 | struct ext4_map_blocks *map, | |
3126 | int split_flag, | |
3127 | int flags) | |
3128 | { | |
3129 | ext4_lblk_t ee_block; | |
3130 | struct ext4_extent *ex; | |
3131 | unsigned int ee_len, depth; | |
3132 | int err = 0; | |
3133 | int uninitialized; | |
3134 | int split_flag1, flags1; | |
3a225670 | 3135 | int allocated = map->m_len; |
47ea3bb5 YY |
3136 | |
3137 | depth = ext_depth(inode); | |
3138 | ex = path[depth].p_ext; | |
3139 | ee_block = le32_to_cpu(ex->ee_block); | |
3140 | ee_len = ext4_ext_get_actual_len(ex); | |
3141 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3142 | ||
3143 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3144 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3145 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3146 | if (uninitialized) | |
3147 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3148 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3149 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3150 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3151 | err = ext4_split_extent_at(handle, inode, path, |
3152 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3153 | if (err) |
3154 | goto out; | |
3a225670 ZL |
3155 | } else { |
3156 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3157 | } |
357b66fd DM |
3158 | /* |
3159 | * Update path is required because previous ext4_split_extent_at() may | |
3160 | * result in split of original leaf or extent zeroout. | |
3161 | */ | |
47ea3bb5 YY |
3162 | ext4_ext_drop_refs(path); |
3163 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3164 | if (IS_ERR(path)) | |
3165 | return PTR_ERR(path); | |
357b66fd DM |
3166 | depth = ext_depth(inode); |
3167 | ex = path[depth].p_ext; | |
3168 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3169 | split_flag1 = 0; | |
47ea3bb5 YY |
3170 | |
3171 | if (map->m_lblk >= ee_block) { | |
357b66fd DM |
3172 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
3173 | if (uninitialized) { | |
47ea3bb5 | 3174 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; |
357b66fd DM |
3175 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3176 | EXT4_EXT_MARK_UNINIT2); | |
3177 | } | |
47ea3bb5 YY |
3178 | err = ext4_split_extent_at(handle, inode, path, |
3179 | map->m_lblk, split_flag1, flags); | |
3180 | if (err) | |
3181 | goto out; | |
3182 | } | |
3183 | ||
3184 | ext4_ext_show_leaf(inode, path); | |
3185 | out: | |
3a225670 | 3186 | return err ? err : allocated; |
47ea3bb5 YY |
3187 | } |
3188 | ||
56055d3a | 3189 | /* |
e35fd660 | 3190 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3191 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3192 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3193 | * uninitialized). |
3194 | * There are three possibilities: | |
3195 | * a> There is no split required: Entire extent should be initialized | |
3196 | * b> Splits in two extents: Write is happening at either end of the extent | |
3197 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3198 | * |
3199 | * Pre-conditions: | |
3200 | * - The extent pointed to by 'path' is uninitialized. | |
3201 | * - The extent pointed to by 'path' contains a superset | |
3202 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3203 | * | |
3204 | * Post-conditions on success: | |
3205 | * - the returned value is the number of blocks beyond map->l_lblk | |
3206 | * that are allocated and initialized. | |
3207 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3208 | */ |
725d26d3 | 3209 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3210 | struct inode *inode, |
3211 | struct ext4_map_blocks *map, | |
3212 | struct ext4_ext_path *path) | |
56055d3a | 3213 | { |
67a5da56 | 3214 | struct ext4_sb_info *sbi; |
6f91bc5f | 3215 | struct ext4_extent_header *eh; |
667eff35 YY |
3216 | struct ext4_map_blocks split_map; |
3217 | struct ext4_extent zero_ex; | |
bc2d9db4 | 3218 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3219 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3220 | unsigned int ee_len, depth, map_len = map->m_len; |
3221 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3222 | int err = 0; |
667eff35 | 3223 | int split_flag = 0; |
21ca087a DM |
3224 | |
3225 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3226 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3227 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3228 | |
67a5da56 | 3229 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3230 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3231 | inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3232 | if (eof_block < map->m_lblk + map_len) |
3233 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3234 | |
3235 | depth = ext_depth(inode); | |
6f91bc5f | 3236 | eh = path[depth].p_hdr; |
56055d3a AA |
3237 | ex = path[depth].p_ext; |
3238 | ee_block = le32_to_cpu(ex->ee_block); | |
3239 | ee_len = ext4_ext_get_actual_len(ex); | |
adb23551 | 3240 | zero_ex.ee_len = 0; |
56055d3a | 3241 | |
6f91bc5f EG |
3242 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3243 | ||
3244 | /* Pre-conditions */ | |
3245 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3246 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3247 | |
3248 | /* | |
3249 | * Attempt to transfer newly initialized blocks from the currently | |
bc2d9db4 | 3250 | * uninitialized extent to its neighbor. This is much cheaper |
6f91bc5f | 3251 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3252 | * memmove() calls. Transferring to the left is the common case in |
3253 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3254 | * followed by append writes. | |
6f91bc5f EG |
3255 | * |
3256 | * Limitations of the current logic: | |
bc2d9db4 | 3257 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3258 | * This would require removing the extent if the transfer |
3259 | * is possible. | |
bc2d9db4 | 3260 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3261 | * same extent tree node. |
3262 | */ | |
bc2d9db4 LC |
3263 | if ((map->m_lblk == ee_block) && |
3264 | /* See if we can merge left */ | |
3265 | (map_len < ee_len) && /*L1*/ | |
3266 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3267 | ext4_lblk_t prev_lblk; |
3268 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3269 | unsigned int prev_len; |
6f91bc5f | 3270 | |
bc2d9db4 LC |
3271 | abut_ex = ex - 1; |
3272 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3273 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3274 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3275 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3276 | |
3277 | /* | |
bc2d9db4 | 3278 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3279 | * upon those conditions: |
bc2d9db4 LC |
3280 | * - C1: abut_ex is initialized, |
3281 | * - C2: abut_ex is logically abutting ex, | |
3282 | * - C3: abut_ex is physically abutting ex, | |
3283 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3284 | * overflowing the (initialized) length limit. |
3285 | */ | |
bc2d9db4 | 3286 | if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3287 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3288 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3289 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3290 | err = ext4_ext_get_access(handle, inode, path + depth); |
3291 | if (err) | |
3292 | goto out; | |
3293 | ||
3294 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3295 | map, ex, abut_ex); |
6f91bc5f | 3296 | |
bc2d9db4 LC |
3297 | /* Shift the start of ex by 'map_len' blocks */ |
3298 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3299 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3300 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
6f91bc5f EG |
3301 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ |
3302 | ||
bc2d9db4 LC |
3303 | /* Extend abut_ex by 'map_len' blocks */ |
3304 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3305 | |
bc2d9db4 LC |
3306 | /* Result: number of initialized blocks past m_lblk */ |
3307 | allocated = map_len; | |
3308 | } | |
3309 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3310 | (map_len < ee_len) && /*L1*/ | |
3311 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3312 | /* See if we can merge right */ | |
3313 | ext4_lblk_t next_lblk; | |
3314 | ext4_fsblk_t next_pblk, ee_pblk; | |
3315 | unsigned int next_len; | |
3316 | ||
3317 | abut_ex = ex + 1; | |
3318 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3319 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3320 | next_pblk = ext4_ext_pblock(abut_ex); | |
3321 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3322 | |
bc2d9db4 LC |
3323 | /* |
3324 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3325 | * upon those conditions: | |
3326 | * - C1: abut_ex is initialized, | |
3327 | * - C2: abut_ex is logically abutting ex, | |
3328 | * - C3: abut_ex is physically abutting ex, | |
3329 | * - C4: abut_ex can receive the additional blocks without | |
3330 | * overflowing the (initialized) length limit. | |
3331 | */ | |
3332 | if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/ | |
3333 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ | |
3334 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3335 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3336 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3337 | if (err) | |
3338 | goto out; | |
3339 | ||
3340 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3341 | map, ex, abut_ex); | |
3342 | ||
3343 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3344 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3345 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3346 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
3347 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3348 | ||
3349 | /* Extend abut_ex by 'map_len' blocks */ | |
3350 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3351 | |
3352 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3353 | allocated = map_len; |
6f91bc5f EG |
3354 | } |
3355 | } | |
bc2d9db4 LC |
3356 | if (allocated) { |
3357 | /* Mark the block containing both extents as dirty */ | |
3358 | ext4_ext_dirty(handle, inode, path + depth); | |
3359 | ||
3360 | /* Update path to point to the right extent */ | |
3361 | path[depth].p_ext = abut_ex; | |
3362 | goto out; | |
3363 | } else | |
3364 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3365 | |
667eff35 | 3366 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3367 | /* |
3368 | * It is safe to convert extent to initialized via explicit | |
3369 | * zeroout only if extent is fully insde i_size or new_size. | |
3370 | */ | |
667eff35 | 3371 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3372 | |
67a5da56 ZL |
3373 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3374 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3375 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 ZL |
3376 | |
3377 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3378 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3379 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3380 | if (err) |
d03856bd | 3381 | goto out; |
adb23551 | 3382 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 | 3383 | zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)); |
adb23551 | 3384 | ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex)); |
d03856bd AK |
3385 | |
3386 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3387 | if (err) | |
3388 | goto out; | |
667eff35 | 3389 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3390 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3391 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3392 | goto out; |
56055d3a | 3393 | } |
667eff35 | 3394 | |
56055d3a | 3395 | /* |
667eff35 YY |
3396 | * four cases: |
3397 | * 1. split the extent into three extents. | |
3398 | * 2. split the extent into two extents, zeroout the first half. | |
3399 | * 3. split the extent into two extents, zeroout the second half. | |
3400 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3401 | */ |
667eff35 YY |
3402 | split_map.m_lblk = map->m_lblk; |
3403 | split_map.m_len = map->m_len; | |
3404 | ||
67a5da56 ZL |
3405 | if (max_zeroout && (allocated > map->m_len)) { |
3406 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3407 | /* case 3 */ |
3408 | zero_ex.ee_block = | |
9b940f8e AH |
3409 | cpu_to_le32(map->m_lblk); |
3410 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3411 | ext4_ext_store_pblock(&zero_ex, |
3412 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3413 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3414 | if (err) |
3415 | goto out; | |
667eff35 YY |
3416 | split_map.m_lblk = map->m_lblk; |
3417 | split_map.m_len = allocated; | |
67a5da56 | 3418 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3419 | /* case 2 */ |
3420 | if (map->m_lblk != ee_block) { | |
3421 | zero_ex.ee_block = ex->ee_block; | |
3422 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3423 | ee_block); | |
3424 | ext4_ext_store_pblock(&zero_ex, | |
3425 | ext4_ext_pblock(ex)); | |
3426 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3427 | if (err) | |
3428 | goto out; | |
3429 | } | |
3430 | ||
667eff35 | 3431 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3432 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3433 | allocated = map->m_len; | |
56055d3a AA |
3434 | } |
3435 | } | |
667eff35 YY |
3436 | |
3437 | allocated = ext4_split_extent(handle, inode, path, | |
67a5da56 | 3438 | &split_map, split_flag, 0); |
667eff35 YY |
3439 | if (allocated < 0) |
3440 | err = allocated; | |
3441 | ||
56055d3a | 3442 | out: |
adb23551 ZL |
3443 | /* If we have gotten a failure, don't zero out status tree */ |
3444 | if (!err) | |
3445 | err = ext4_es_zeroout(inode, &zero_ex); | |
56055d3a AA |
3446 | return err ? err : allocated; |
3447 | } | |
3448 | ||
0031462b | 3449 | /* |
e35fd660 | 3450 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3451 | * ext4_get_blocks_dio_write() when DIO to write |
3452 | * to an uninitialized extent. | |
3453 | * | |
fd018fe8 | 3454 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3455 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3456 | * There are three possibilities: |
3457 | * a> There is no split required: Entire extent should be uninitialized | |
3458 | * b> Splits in two extents: Write is happening at either end of the extent | |
3459 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3460 | * | |
3461 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3462 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3463 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3464 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3465 | * into three uninitialized extent(at most). After IO complete, the part |
3466 | * being filled will be convert to initialized by the end_io callback function | |
3467 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3468 | * |
3469 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3470 | */ |
3471 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3472 | struct inode *inode, | |
e35fd660 | 3473 | struct ext4_map_blocks *map, |
0031462b | 3474 | struct ext4_ext_path *path, |
0031462b MC |
3475 | int flags) |
3476 | { | |
667eff35 YY |
3477 | ext4_lblk_t eof_block; |
3478 | ext4_lblk_t ee_block; | |
3479 | struct ext4_extent *ex; | |
3480 | unsigned int ee_len; | |
3481 | int split_flag = 0, depth; | |
21ca087a DM |
3482 | |
3483 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3484 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3485 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3486 | |
3487 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3488 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3489 | if (eof_block < map->m_lblk + map->m_len) |
3490 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3491 | /* |
3492 | * It is safe to convert extent to initialized via explicit | |
3493 | * zeroout only if extent is fully insde i_size or new_size. | |
3494 | */ | |
667eff35 YY |
3495 | depth = ext_depth(inode); |
3496 | ex = path[depth].p_ext; | |
3497 | ee_block = le32_to_cpu(ex->ee_block); | |
3498 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3499 | |
667eff35 YY |
3500 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3501 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3502 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3503 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3504 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3505 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3506 | } |
197217a5 | 3507 | |
c7064ef1 | 3508 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3509 | struct inode *inode, |
3510 | struct ext4_map_blocks *map, | |
3511 | struct ext4_ext_path *path) | |
0031462b MC |
3512 | { |
3513 | struct ext4_extent *ex; | |
dee1f973 DM |
3514 | ext4_lblk_t ee_block; |
3515 | unsigned int ee_len; | |
0031462b MC |
3516 | int depth; |
3517 | int err = 0; | |
0031462b MC |
3518 | |
3519 | depth = ext_depth(inode); | |
0031462b | 3520 | ex = path[depth].p_ext; |
dee1f973 DM |
3521 | ee_block = le32_to_cpu(ex->ee_block); |
3522 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3523 | |
197217a5 YY |
3524 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3525 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3526 | (unsigned long long)ee_block, ee_len); |
3527 | ||
ff95ec22 DM |
3528 | /* If extent is larger than requested it is a clear sign that we still |
3529 | * have some extent state machine issues left. So extent_split is still | |
3530 | * required. | |
3531 | * TODO: Once all related issues will be fixed this situation should be | |
3532 | * illegal. | |
3533 | */ | |
dee1f973 | 3534 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ff95ec22 DM |
3535 | #ifdef EXT4_DEBUG |
3536 | ext4_warning("Inode (%ld) finished: extent logical block %llu," | |
3537 | " len %u; IO logical block %llu, len %u\n", | |
3538 | inode->i_ino, (unsigned long long)ee_block, ee_len, | |
3539 | (unsigned long long)map->m_lblk, map->m_len); | |
3540 | #endif | |
dee1f973 DM |
3541 | err = ext4_split_unwritten_extents(handle, inode, map, path, |
3542 | EXT4_GET_BLOCKS_CONVERT); | |
3543 | if (err < 0) | |
3544 | goto out; | |
3545 | ext4_ext_drop_refs(path); | |
3546 | path = ext4_ext_find_extent(inode, map->m_lblk, path); | |
3547 | if (IS_ERR(path)) { | |
3548 | err = PTR_ERR(path); | |
3549 | goto out; | |
3550 | } | |
3551 | depth = ext_depth(inode); | |
3552 | ex = path[depth].p_ext; | |
3553 | } | |
197217a5 | 3554 | |
0031462b MC |
3555 | err = ext4_ext_get_access(handle, inode, path + depth); |
3556 | if (err) | |
3557 | goto out; | |
3558 | /* first mark the extent as initialized */ | |
3559 | ext4_ext_mark_initialized(ex); | |
3560 | ||
197217a5 YY |
3561 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3562 | * borders are not changed | |
0031462b | 3563 | */ |
ecb94f5f | 3564 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3565 | |
0031462b | 3566 | /* Mark modified extent as dirty */ |
ecb94f5f | 3567 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3568 | out: |
3569 | ext4_ext_show_leaf(inode, path); | |
3570 | return err; | |
3571 | } | |
3572 | ||
515f41c3 AK |
3573 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3574 | sector_t block, int count) | |
3575 | { | |
3576 | int i; | |
3577 | for (i = 0; i < count; i++) | |
3578 | unmap_underlying_metadata(bdev, block + i); | |
3579 | } | |
3580 | ||
58590b06 TT |
3581 | /* |
3582 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3583 | */ | |
3584 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3585 | ext4_lblk_t lblk, |
58590b06 TT |
3586 | struct ext4_ext_path *path, |
3587 | unsigned int len) | |
3588 | { | |
3589 | int i, depth; | |
3590 | struct ext4_extent_header *eh; | |
65922cb5 | 3591 | struct ext4_extent *last_ex; |
58590b06 TT |
3592 | |
3593 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3594 | return 0; | |
3595 | ||
3596 | depth = ext_depth(inode); | |
3597 | eh = path[depth].p_hdr; | |
58590b06 | 3598 | |
afcff5d8 LC |
3599 | /* |
3600 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3601 | * do not care for this case anymore. Simply remove the flag | |
3602 | * if there are no extents. | |
3603 | */ | |
3604 | if (unlikely(!eh->eh_entries)) | |
3605 | goto out; | |
58590b06 TT |
3606 | last_ex = EXT_LAST_EXTENT(eh); |
3607 | /* | |
3608 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3609 | * last block in the last extent in the file. We test this by | |
3610 | * first checking to see if the caller to | |
3611 | * ext4_ext_get_blocks() was interested in the last block (or | |
3612 | * a block beyond the last block) in the current extent. If | |
3613 | * this turns out to be false, we can bail out from this | |
3614 | * function immediately. | |
3615 | */ | |
d002ebf1 | 3616 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3617 | ext4_ext_get_actual_len(last_ex)) |
3618 | return 0; | |
3619 | /* | |
3620 | * If the caller does appear to be planning to write at or | |
3621 | * beyond the end of the current extent, we then test to see | |
3622 | * if the current extent is the last extent in the file, by | |
3623 | * checking to make sure it was reached via the rightmost node | |
3624 | * at each level of the tree. | |
3625 | */ | |
3626 | for (i = depth-1; i >= 0; i--) | |
3627 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3628 | return 0; | |
afcff5d8 | 3629 | out: |
58590b06 TT |
3630 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3631 | return ext4_mark_inode_dirty(handle, inode); | |
3632 | } | |
3633 | ||
7b415bf6 AK |
3634 | /** |
3635 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3636 | * | |
7d1b1fbc | 3637 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3638 | */ |
f7fec032 ZL |
3639 | int ext4_find_delalloc_range(struct inode *inode, |
3640 | ext4_lblk_t lblk_start, | |
3641 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3642 | { |
7d1b1fbc | 3643 | struct extent_status es; |
7b415bf6 | 3644 | |
be401363 | 3645 | ext4_es_find_delayed_extent(inode, lblk_start, &es); |
06b0c886 | 3646 | if (es.es_len == 0) |
7d1b1fbc | 3647 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3648 | else if (es.es_lblk <= lblk_start && |
3649 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3650 | return 1; |
06b0c886 | 3651 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3652 | return 1; |
7b415bf6 | 3653 | else |
7d1b1fbc | 3654 | return 0; |
7b415bf6 AK |
3655 | } |
3656 | ||
7d1b1fbc | 3657 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3658 | { |
3659 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3660 | ext4_lblk_t lblk_start, lblk_end; | |
3661 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3662 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3663 | ||
7d1b1fbc | 3664 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3665 | } |
3666 | ||
3667 | /** | |
3668 | * Determines how many complete clusters (out of those specified by the 'map') | |
3669 | * are under delalloc and were reserved quota for. | |
3670 | * This function is called when we are writing out the blocks that were | |
3671 | * originally written with their allocation delayed, but then the space was | |
3672 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3673 | * The cases to look for are: | |
3674 | * ('=' indicated delayed allocated blocks | |
3675 | * '-' indicates non-delayed allocated blocks) | |
3676 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3677 | * are not delalloc'ed. | |
3678 | * Ex: | |
3679 | * |----c---=|====c====|====c====|===-c----| | |
3680 | * |++++++ allocated ++++++| | |
3681 | * ==> 4 complete clusters in above example | |
3682 | * | |
3683 | * (b) partial cluster (outside of allocated range) towards either end is | |
3684 | * marked for delayed allocation. In this case, we will exclude that | |
3685 | * cluster. | |
3686 | * Ex: | |
3687 | * |----====c========|========c========| | |
3688 | * |++++++ allocated ++++++| | |
3689 | * ==> 1 complete clusters in above example | |
3690 | * | |
3691 | * Ex: | |
3692 | * |================c================| | |
3693 | * |++++++ allocated ++++++| | |
3694 | * ==> 0 complete clusters in above example | |
3695 | * | |
3696 | * The ext4_da_update_reserve_space will be called only if we | |
3697 | * determine here that there were some "entire" clusters that span | |
3698 | * this 'allocated' range. | |
3699 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3700 | * without ever calling ext4_find_delalloc_range. | |
3701 | */ | |
3702 | static unsigned int | |
3703 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3704 | unsigned int num_blks) | |
3705 | { | |
3706 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3707 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3708 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3709 | unsigned int allocated_clusters = 0; | |
3710 | ||
3711 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3712 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3713 | ||
3714 | /* max possible clusters for this allocation */ | |
3715 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3716 | ||
d8990240 AK |
3717 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3718 | ||
7b415bf6 AK |
3719 | /* Check towards left side */ |
3720 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3721 | if (c_offset) { | |
3722 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3723 | lblk_to = lblk_from + c_offset - 1; | |
3724 | ||
7d1b1fbc | 3725 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3726 | allocated_clusters--; |
3727 | } | |
3728 | ||
3729 | /* Now check towards right. */ | |
3730 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3731 | if (allocated_clusters && c_offset) { | |
3732 | lblk_from = lblk_start + num_blks; | |
3733 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3734 | ||
7d1b1fbc | 3735 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3736 | allocated_clusters--; |
3737 | } | |
3738 | ||
3739 | return allocated_clusters; | |
3740 | } | |
3741 | ||
0031462b MC |
3742 | static int |
3743 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3744 | struct ext4_map_blocks *map, |
0031462b | 3745 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3746 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3747 | { |
3748 | int ret = 0; | |
3749 | int err = 0; | |
f45ee3a1 | 3750 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3751 | |
88635ca2 ZL |
3752 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3753 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3754 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3755 | flags, allocated); |
3756 | ext4_ext_show_leaf(inode, path); | |
3757 | ||
b5645534 ZL |
3758 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3759 | allocated, newblock); | |
d8990240 | 3760 | |
c7064ef1 | 3761 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3762 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3763 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3764 | path, flags); | |
82e54229 DM |
3765 | if (ret <= 0) |
3766 | goto out; | |
5f524950 M |
3767 | /* |
3768 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3769 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3770 | * completed |
3771 | */ | |
0edeb71d TM |
3772 | if (io) |
3773 | ext4_set_io_unwritten_flag(inode, io); | |
3774 | else | |
19f5fb7a | 3775 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
a25a4e1a | 3776 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
744692dc | 3777 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3778 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3779 | goto out; |
3780 | } | |
c7064ef1 | 3781 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3782 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3783 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3784 | path); |
58590b06 | 3785 | if (ret >= 0) { |
b436b9be | 3786 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3787 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3788 | path, map->m_len); | |
58590b06 TT |
3789 | } else |
3790 | err = ret; | |
cdee7843 ZL |
3791 | map->m_flags |= EXT4_MAP_MAPPED; |
3792 | if (allocated > map->m_len) | |
3793 | allocated = map->m_len; | |
3794 | map->m_len = allocated; | |
0031462b MC |
3795 | goto out2; |
3796 | } | |
3797 | /* buffered IO case */ | |
3798 | /* | |
3799 | * repeat fallocate creation request | |
3800 | * we already have an unwritten extent | |
3801 | */ | |
a25a4e1a ZL |
3802 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) { |
3803 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
0031462b | 3804 | goto map_out; |
a25a4e1a | 3805 | } |
0031462b MC |
3806 | |
3807 | /* buffered READ or buffered write_begin() lookup */ | |
3808 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3809 | /* | |
3810 | * We have blocks reserved already. We | |
3811 | * return allocated blocks so that delalloc | |
3812 | * won't do block reservation for us. But | |
3813 | * the buffer head will be unmapped so that | |
3814 | * a read from the block returns 0s. | |
3815 | */ | |
e35fd660 | 3816 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3817 | goto out1; |
3818 | } | |
3819 | ||
3820 | /* buffered write, writepage time, convert*/ | |
e35fd660 | 3821 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path); |
a4e5d88b | 3822 | if (ret >= 0) |
b436b9be | 3823 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3824 | out: |
3825 | if (ret <= 0) { | |
3826 | err = ret; | |
3827 | goto out2; | |
3828 | } else | |
3829 | allocated = ret; | |
e35fd660 | 3830 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3831 | /* |
3832 | * if we allocated more blocks than requested | |
3833 | * we need to make sure we unmap the extra block | |
3834 | * allocated. The actual needed block will get | |
3835 | * unmapped later when we find the buffer_head marked | |
3836 | * new. | |
3837 | */ | |
e35fd660 | 3838 | if (allocated > map->m_len) { |
515f41c3 | 3839 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3840 | newblock + map->m_len, |
3841 | allocated - map->m_len); | |
3842 | allocated = map->m_len; | |
515f41c3 | 3843 | } |
3a225670 | 3844 | map->m_len = allocated; |
5f634d06 AK |
3845 | |
3846 | /* | |
3847 | * If we have done fallocate with the offset that is already | |
3848 | * delayed allocated, we would have block reservation | |
3849 | * and quota reservation done in the delayed write path. | |
3850 | * But fallocate would have already updated quota and block | |
3851 | * count for this offset. So cancel these reservation | |
3852 | */ | |
7b415bf6 AK |
3853 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3854 | unsigned int reserved_clusters; | |
3855 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3856 | map->m_lblk, map->m_len); | |
3857 | if (reserved_clusters) | |
3858 | ext4_da_update_reserve_space(inode, | |
3859 | reserved_clusters, | |
3860 | 0); | |
3861 | } | |
5f634d06 | 3862 | |
0031462b | 3863 | map_out: |
e35fd660 | 3864 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3865 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3866 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3867 | map->m_len); | |
3868 | if (err < 0) | |
3869 | goto out2; | |
3870 | } | |
0031462b | 3871 | out1: |
e35fd660 TT |
3872 | if (allocated > map->m_len) |
3873 | allocated = map->m_len; | |
0031462b | 3874 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3875 | map->m_pblk = newblock; |
3876 | map->m_len = allocated; | |
0031462b MC |
3877 | out2: |
3878 | if (path) { | |
3879 | ext4_ext_drop_refs(path); | |
3880 | kfree(path); | |
3881 | } | |
3882 | return err ? err : allocated; | |
3883 | } | |
58590b06 | 3884 | |
4d33b1ef TT |
3885 | /* |
3886 | * get_implied_cluster_alloc - check to see if the requested | |
3887 | * allocation (in the map structure) overlaps with a cluster already | |
3888 | * allocated in an extent. | |
d8990240 | 3889 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3890 | * @map The requested lblk->pblk mapping |
3891 | * @ex The extent structure which might contain an implied | |
3892 | * cluster allocation | |
3893 | * | |
3894 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3895 | * find blocks that were already in the inode's extent tree. Hence, | |
3896 | * we know that the beginning of the requested region cannot overlap | |
3897 | * the extent from the inode's extent tree. There are three cases we | |
3898 | * want to catch. The first is this case: | |
3899 | * | |
3900 | * |--- cluster # N--| | |
3901 | * |--- extent ---| |---- requested region ---| | |
3902 | * |==========| | |
3903 | * | |
3904 | * The second case that we need to test for is this one: | |
3905 | * | |
3906 | * |--------- cluster # N ----------------| | |
3907 | * |--- requested region --| |------- extent ----| | |
3908 | * |=======================| | |
3909 | * | |
3910 | * The third case is when the requested region lies between two extents | |
3911 | * within the same cluster: | |
3912 | * |------------- cluster # N-------------| | |
3913 | * |----- ex -----| |---- ex_right ----| | |
3914 | * |------ requested region ------| | |
3915 | * |================| | |
3916 | * | |
3917 | * In each of the above cases, we need to set the map->m_pblk and | |
3918 | * map->m_len so it corresponds to the return the extent labelled as | |
3919 | * "|====|" from cluster #N, since it is already in use for data in | |
3920 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3921 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3922 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3923 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3924 | * by calling ext4_mb_new_blocks(). | |
3925 | */ | |
d8990240 | 3926 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3927 | struct ext4_map_blocks *map, |
3928 | struct ext4_extent *ex, | |
3929 | struct ext4_ext_path *path) | |
3930 | { | |
d8990240 | 3931 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3932 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3933 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3934 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3935 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3936 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3937 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3938 | ||
3939 | /* The extent passed in that we are trying to match */ | |
3940 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3941 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3942 | ||
3943 | /* The requested region passed into ext4_map_blocks() */ | |
3944 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3945 | |
3946 | if ((rr_cluster_start == ex_cluster_end) || | |
3947 | (rr_cluster_start == ex_cluster_start)) { | |
3948 | if (rr_cluster_start == ex_cluster_end) | |
3949 | ee_start += ee_len - 1; | |
3950 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
3951 | c_offset; | |
3952 | map->m_len = min(map->m_len, | |
3953 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
3954 | /* | |
3955 | * Check for and handle this case: | |
3956 | * | |
3957 | * |--------- cluster # N-------------| | |
3958 | * |------- extent ----| | |
3959 | * |--- requested region ---| | |
3960 | * |===========| | |
3961 | */ | |
3962 | ||
3963 | if (map->m_lblk < ee_block) | |
3964 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
3965 | ||
3966 | /* | |
3967 | * Check for the case where there is already another allocated | |
3968 | * block to the right of 'ex' but before the end of the cluster. | |
3969 | * | |
3970 | * |------------- cluster # N-------------| | |
3971 | * |----- ex -----| |---- ex_right ----| | |
3972 | * |------ requested region ------| | |
3973 | * |================| | |
3974 | */ | |
3975 | if (map->m_lblk > ee_block) { | |
3976 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
3977 | map->m_len = min(map->m_len, next - map->m_lblk); | |
3978 | } | |
d8990240 AK |
3979 | |
3980 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
3981 | return 1; |
3982 | } | |
d8990240 AK |
3983 | |
3984 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
3985 | return 0; |
3986 | } | |
3987 | ||
3988 | ||
c278bfec | 3989 | /* |
f5ab0d1f MC |
3990 | * Block allocation/map/preallocation routine for extents based files |
3991 | * | |
3992 | * | |
c278bfec | 3993 | * Need to be called with |
0e855ac8 AK |
3994 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
3995 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
3996 | * |
3997 | * return > 0, number of of blocks already mapped/allocated | |
3998 | * if create == 0 and these are pre-allocated blocks | |
3999 | * buffer head is unmapped | |
4000 | * otherwise blocks are mapped | |
4001 | * | |
4002 | * return = 0, if plain look up failed (blocks have not been allocated) | |
4003 | * buffer head is unmapped | |
4004 | * | |
4005 | * return < 0, error case. | |
c278bfec | 4006 | */ |
e35fd660 TT |
4007 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
4008 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
4009 | { |
4010 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
4011 | struct ext4_extent newex, *ex, *ex2; |
4012 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 4013 | ext4_fsblk_t newblock = 0; |
37794732 | 4014 | int free_on_err = 0, err = 0, depth; |
4d33b1ef | 4015 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 4016 | unsigned int allocated_clusters = 0; |
c9de560d | 4017 | struct ext4_allocation_request ar; |
f45ee3a1 | 4018 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 4019 | ext4_lblk_t cluster_offset; |
82e54229 | 4020 | int set_unwritten = 0; |
a86c6181 | 4021 | |
84fe3bef | 4022 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4023 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4024 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4025 | |
a86c6181 | 4026 | /* find extent for this block */ |
e35fd660 | 4027 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL); |
a86c6181 AT |
4028 | if (IS_ERR(path)) { |
4029 | err = PTR_ERR(path); | |
4030 | path = NULL; | |
4031 | goto out2; | |
4032 | } | |
4033 | ||
4034 | depth = ext_depth(inode); | |
4035 | ||
4036 | /* | |
d0d856e8 RD |
4037 | * consistent leaf must not be empty; |
4038 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
4039 | * this is why assert can't be put in ext4_ext_find_extent() |
4040 | */ | |
273df556 FM |
4041 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4042 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4043 | "lblock: %lu, depth: %d pblock %lld", |
4044 | (unsigned long) map->m_lblk, depth, | |
4045 | path[depth].p_block); | |
034fb4c9 SP |
4046 | err = -EIO; |
4047 | goto out2; | |
4048 | } | |
a86c6181 | 4049 | |
7e028976 AM |
4050 | ex = path[depth].p_ext; |
4051 | if (ex) { | |
725d26d3 | 4052 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4053 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4054 | unsigned short ee_len; |
471d4011 SB |
4055 | |
4056 | /* | |
471d4011 | 4057 | * Uninitialized extents are treated as holes, except that |
56055d3a | 4058 | * we split out initialized portions during a write. |
471d4011 | 4059 | */ |
a2df2a63 | 4060 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4061 | |
4062 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4063 | ||
d0d856e8 | 4064 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4065 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4066 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4067 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4068 | allocated = ee_len - (map->m_lblk - ee_block); |
4069 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4070 | ee_block, ee_len, newblock); | |
56055d3a | 4071 | |
69eb33dc | 4072 | if (!ext4_ext_is_uninitialized(ex)) |
7877191c | 4073 | goto out; |
69eb33dc | 4074 | |
37794732 | 4075 | allocated = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
4076 | handle, inode, map, path, flags, |
4077 | allocated, newblock); | |
37794732 | 4078 | goto out3; |
a86c6181 AT |
4079 | } |
4080 | } | |
4081 | ||
7b415bf6 | 4082 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 4083 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
4084 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4085 | ||
a86c6181 | 4086 | /* |
d0d856e8 | 4087 | * requested block isn't allocated yet; |
a86c6181 AT |
4088 | * we couldn't try to create block if create flag is zero |
4089 | */ | |
c2177057 | 4090 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4091 | /* |
4092 | * put just found gap into cache to speed up | |
4093 | * subsequent requests | |
4094 | */ | |
d100eef2 ZL |
4095 | if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0) |
4096 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); | |
a86c6181 AT |
4097 | goto out2; |
4098 | } | |
4d33b1ef | 4099 | |
a86c6181 | 4100 | /* |
c2ea3fde | 4101 | * Okay, we need to do block allocation. |
63f57933 | 4102 | */ |
7b415bf6 | 4103 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4104 | newex.ee_block = cpu_to_le32(map->m_lblk); |
4105 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
4106 | ||
4107 | /* | |
4108 | * If we are doing bigalloc, check to see if the extent returned | |
4109 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4110 | */ | |
4111 | if (cluster_offset && ex && | |
d8990240 | 4112 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4113 | ar.len = allocated = map->m_len; |
4114 | newblock = map->m_pblk; | |
7b415bf6 | 4115 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4116 | goto got_allocated_blocks; |
4117 | } | |
a86c6181 | 4118 | |
c9de560d | 4119 | /* find neighbour allocated blocks */ |
e35fd660 | 4120 | ar.lleft = map->m_lblk; |
c9de560d AT |
4121 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4122 | if (err) | |
4123 | goto out2; | |
e35fd660 | 4124 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4125 | ex2 = NULL; |
4126 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4127 | if (err) |
4128 | goto out2; | |
25d14f98 | 4129 | |
4d33b1ef TT |
4130 | /* Check if the extent after searching to the right implies a |
4131 | * cluster we can use. */ | |
4132 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4133 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4134 | ar.len = allocated = map->m_len; |
4135 | newblock = map->m_pblk; | |
7b415bf6 | 4136 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4137 | goto got_allocated_blocks; |
4138 | } | |
4139 | ||
749269fa AA |
4140 | /* |
4141 | * See if request is beyond maximum number of blocks we can have in | |
4142 | * a single extent. For an initialized extent this limit is | |
4143 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4144 | * EXT_UNINIT_MAX_LEN. | |
4145 | */ | |
e35fd660 | 4146 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4147 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4148 | map->m_len = EXT_INIT_MAX_LEN; |
4149 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4150 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4151 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4152 | |
e35fd660 | 4153 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4154 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4155 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4156 | if (err) |
b939e376 | 4157 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4158 | else |
e35fd660 | 4159 | allocated = map->m_len; |
c9de560d AT |
4160 | |
4161 | /* allocate new block */ | |
4162 | ar.inode = inode; | |
e35fd660 TT |
4163 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4164 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4165 | /* |
4166 | * We calculate the offset from the beginning of the cluster | |
4167 | * for the logical block number, since when we allocate a | |
4168 | * physical cluster, the physical block should start at the | |
4169 | * same offset from the beginning of the cluster. This is | |
4170 | * needed so that future calls to get_implied_cluster_alloc() | |
4171 | * work correctly. | |
4172 | */ | |
4173 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4174 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4175 | ar.goal -= offset; | |
4176 | ar.logical -= offset; | |
c9de560d AT |
4177 | if (S_ISREG(inode->i_mode)) |
4178 | ar.flags = EXT4_MB_HINT_DATA; | |
4179 | else | |
4180 | /* disable in-core preallocation for non-regular files */ | |
4181 | ar.flags = 0; | |
556b27ab VH |
4182 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4183 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4184 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4185 | if (!newblock) |
4186 | goto out2; | |
84fe3bef | 4187 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4188 | ar.goal, newblock, allocated); |
4d33b1ef | 4189 | free_on_err = 1; |
7b415bf6 | 4190 | allocated_clusters = ar.len; |
4d33b1ef TT |
4191 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4192 | if (ar.len > allocated) | |
4193 | ar.len = allocated; | |
a86c6181 | 4194 | |
4d33b1ef | 4195 | got_allocated_blocks: |
a86c6181 | 4196 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4197 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4198 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4199 | /* Mark uninitialized */ |
4200 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4201 | ext4_ext_mark_uninitialized(&newex); |
a25a4e1a | 4202 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4203 | /* |
744692dc | 4204 | * io_end structure was created for every IO write to an |
25985edc | 4205 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4206 | * here we flag the IO that really needs the conversion. |
5f524950 | 4207 | * For non asycn direct IO case, flag the inode state |
25985edc | 4208 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4209 | */ |
82e54229 DM |
4210 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4211 | set_unwritten = 1; | |
744692dc | 4212 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4213 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4214 | } |
c8d46e41 | 4215 | |
a4e5d88b DM |
4216 | err = 0; |
4217 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4218 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4219 | path, ar.len); | |
575a1d4b JZ |
4220 | if (!err) |
4221 | err = ext4_ext_insert_extent(handle, inode, path, | |
4222 | &newex, flags); | |
82e54229 DM |
4223 | |
4224 | if (!err && set_unwritten) { | |
4225 | if (io) | |
4226 | ext4_set_io_unwritten_flag(inode, io); | |
4227 | else | |
4228 | ext4_set_inode_state(inode, | |
4229 | EXT4_STATE_DIO_UNWRITTEN); | |
4230 | } | |
4231 | ||
4d33b1ef | 4232 | if (err && free_on_err) { |
7132de74 MP |
4233 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4234 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4235 | /* free data blocks we just allocated */ |
c9de560d AT |
4236 | /* not a good idea to call discard here directly, |
4237 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4238 | ext4_discard_preallocations(inode); |
7dc57615 | 4239 | ext4_free_blocks(handle, inode, NULL, ext4_ext_pblock(&newex), |
7132de74 | 4240 | ext4_ext_get_actual_len(&newex), fb_flags); |
a86c6181 | 4241 | goto out2; |
315054f0 | 4242 | } |
a86c6181 | 4243 | |
a86c6181 | 4244 | /* previous routine could use block we allocated */ |
bf89d16f | 4245 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4246 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4247 | if (allocated > map->m_len) |
4248 | allocated = map->m_len; | |
4249 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4250 | |
5f634d06 AK |
4251 | /* |
4252 | * Update reserved blocks/metadata blocks after successful | |
4253 | * block allocation which had been deferred till now. | |
4254 | */ | |
7b415bf6 | 4255 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4256 | unsigned int reserved_clusters; |
7b415bf6 | 4257 | /* |
81fdbb4a | 4258 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4259 | */ |
4260 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4261 | map->m_lblk, allocated); | |
4262 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4263 | if (reserved_clusters) { | |
4264 | /* | |
4265 | * We have clusters reserved for this range. | |
4266 | * But since we are not doing actual allocation | |
4267 | * and are simply using blocks from previously | |
4268 | * allocated cluster, we should release the | |
4269 | * reservation and not claim quota. | |
4270 | */ | |
4271 | ext4_da_update_reserve_space(inode, | |
4272 | reserved_clusters, 0); | |
4273 | } | |
4274 | } else { | |
4275 | BUG_ON(allocated_clusters < reserved_clusters); | |
7b415bf6 | 4276 | if (reserved_clusters < allocated_clusters) { |
5356f261 | 4277 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4278 | int reservation = allocated_clusters - |
4279 | reserved_clusters; | |
4280 | /* | |
4281 | * It seems we claimed few clusters outside of | |
4282 | * the range of this allocation. We should give | |
4283 | * it back to the reservation pool. This can | |
4284 | * happen in the following case: | |
4285 | * | |
4286 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4287 | * cluster has 4 blocks. Thus, the clusters | |
4288 | * are [0-3],[4-7],[8-11]... | |
4289 | * * First comes delayed allocation write for | |
4290 | * logical blocks 10 & 11. Since there were no | |
4291 | * previous delayed allocated blocks in the | |
4292 | * range [8-11], we would reserve 1 cluster | |
4293 | * for this write. | |
4294 | * * Next comes write for logical blocks 3 to 8. | |
4295 | * In this case, we will reserve 2 clusters | |
4296 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4297 | * that range has a delayed allocated blocks. | |
4298 | * Thus total reserved clusters now becomes 3. | |
4299 | * * Now, during the delayed allocation writeout | |
4300 | * time, we will first write blocks [3-8] and | |
4301 | * allocate 3 clusters for writing these | |
4302 | * blocks. Also, we would claim all these | |
4303 | * three clusters above. | |
4304 | * * Now when we come here to writeout the | |
4305 | * blocks [10-11], we would expect to claim | |
4306 | * the reservation of 1 cluster we had made | |
4307 | * (and we would claim it since there are no | |
4308 | * more delayed allocated blocks in the range | |
4309 | * [8-11]. But our reserved cluster count had | |
4310 | * already gone to 0. | |
4311 | * | |
4312 | * Thus, at the step 4 above when we determine | |
4313 | * that there are still some unwritten delayed | |
4314 | * allocated blocks outside of our current | |
4315 | * block range, we should increment the | |
4316 | * reserved clusters count so that when the | |
4317 | * remaining blocks finally gets written, we | |
4318 | * could claim them. | |
4319 | */ | |
5356f261 AK |
4320 | dquot_reserve_block(inode, |
4321 | EXT4_C2B(sbi, reservation)); | |
4322 | spin_lock(&ei->i_block_reservation_lock); | |
4323 | ei->i_reserved_data_blocks += reservation; | |
4324 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 | 4325 | } |
232ec872 LC |
4326 | /* |
4327 | * We will claim quota for all newly allocated blocks. | |
4328 | * We're updating the reserved space *after* the | |
4329 | * correction above so we do not accidentally free | |
4330 | * all the metadata reservation because we might | |
4331 | * actually need it later on. | |
4332 | */ | |
4333 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4334 | 1); | |
7b415bf6 AK |
4335 | } |
4336 | } | |
5f634d06 | 4337 | |
b436b9be JK |
4338 | /* |
4339 | * Cache the extent and update transaction to commit on fdatasync only | |
4340 | * when it is _not_ an uninitialized extent. | |
4341 | */ | |
69eb33dc | 4342 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) |
b436b9be | 4343 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4344 | else |
b436b9be | 4345 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4346 | out: |
e35fd660 TT |
4347 | if (allocated > map->m_len) |
4348 | allocated = map->m_len; | |
a86c6181 | 4349 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4350 | map->m_flags |= EXT4_MAP_MAPPED; |
4351 | map->m_pblk = newblock; | |
4352 | map->m_len = allocated; | |
a86c6181 AT |
4353 | out2: |
4354 | if (path) { | |
4355 | ext4_ext_drop_refs(path); | |
4356 | kfree(path); | |
4357 | } | |
e861304b | 4358 | |
37794732 | 4359 | out3: |
19b303d8 | 4360 | trace_ext4_ext_map_blocks_exit(inode, map, err ? err : allocated); |
e7b319e3 | 4361 | |
7877191c | 4362 | return err ? err : allocated; |
a86c6181 AT |
4363 | } |
4364 | ||
819c4920 | 4365 | void ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4366 | { |
a86c6181 | 4367 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4368 | ext4_lblk_t last_block; |
a86c6181 AT |
4369 | int err = 0; |
4370 | ||
a86c6181 | 4371 | /* |
d0d856e8 RD |
4372 | * TODO: optimization is possible here. |
4373 | * Probably we need not scan at all, | |
4374 | * because page truncation is enough. | |
a86c6181 | 4375 | */ |
a86c6181 AT |
4376 | |
4377 | /* we have to know where to truncate from in crash case */ | |
4378 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4379 | ext4_mark_inode_dirty(handle, inode); | |
4380 | ||
4381 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4382 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
51865fda ZL |
4383 | err = ext4_es_remove_extent(inode, last_block, |
4384 | EXT_MAX_BLOCKS - last_block); | |
5f95d21f | 4385 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
a86c6181 AT |
4386 | } |
4387 | ||
fd28784a AK |
4388 | static void ext4_falloc_update_inode(struct inode *inode, |
4389 | int mode, loff_t new_size, int update_ctime) | |
4390 | { | |
4391 | struct timespec now; | |
4392 | ||
4393 | if (update_ctime) { | |
4394 | now = current_fs_time(inode->i_sb); | |
4395 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4396 | inode->i_ctime = now; | |
4397 | } | |
4398 | /* | |
4399 | * Update only when preallocation was requested beyond | |
4400 | * the file size. | |
4401 | */ | |
cf17fea6 AK |
4402 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4403 | if (new_size > i_size_read(inode)) | |
4404 | i_size_write(inode, new_size); | |
4405 | if (new_size > EXT4_I(inode)->i_disksize) | |
4406 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4407 | } else { |
4408 | /* | |
4409 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4410 | * can proceed even if the new size is the same as i_size. | |
4411 | */ | |
4412 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4413 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4414 | } |
4415 | ||
4416 | } | |
4417 | ||
a2df2a63 | 4418 | /* |
2fe17c10 | 4419 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4420 | * operation, which gets called from sys_fallocate system call. |
4421 | * For block-mapped files, posix_fallocate should fall back to the method | |
4422 | * of writing zeroes to the required new blocks (the same behavior which is | |
4423 | * expected for file systems which do not support fallocate() system call). | |
4424 | */ | |
2fe17c10 | 4425 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4426 | { |
496ad9aa | 4427 | struct inode *inode = file_inode(file); |
a2df2a63 | 4428 | handle_t *handle; |
fd28784a | 4429 | loff_t new_size; |
498e5f24 | 4430 | unsigned int max_blocks; |
a2df2a63 AA |
4431 | int ret = 0; |
4432 | int ret2 = 0; | |
4433 | int retries = 0; | |
a4e5d88b | 4434 | int flags; |
2ed88685 | 4435 | struct ext4_map_blocks map; |
a2df2a63 AA |
4436 | unsigned int credits, blkbits = inode->i_blkbits; |
4437 | ||
a4bb6b64 AH |
4438 | /* Return error if mode is not supported */ |
4439 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4440 | return -EOPNOTSUPP; | |
4441 | ||
4442 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
4443 | return ext4_punch_hole(file, offset, len); | |
4444 | ||
0c8d414f TM |
4445 | ret = ext4_convert_inline_data(inode); |
4446 | if (ret) | |
4447 | return ret; | |
4448 | ||
8bad6fc8 ZL |
4449 | /* |
4450 | * currently supporting (pre)allocate mode for extent-based | |
4451 | * files _only_ | |
4452 | */ | |
4453 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4454 | return -EOPNOTSUPP; | |
4455 | ||
0562e0ba | 4456 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4457 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4458 | /* |
4459 | * We can't just convert len to max_blocks because | |
4460 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4461 | */ | |
a2df2a63 | 4462 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4463 | - map.m_lblk; |
a2df2a63 | 4464 | /* |
f3bd1f3f | 4465 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4466 | */ |
f3bd1f3f | 4467 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4468 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4469 | ret = inode_newsize_ok(inode, (len + offset)); |
4470 | if (ret) { | |
4471 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4472 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4473 | return ret; |
4474 | } | |
3c6fe770 | 4475 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4476 | if (mode & FALLOC_FL_KEEP_SIZE) |
4477 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4478 | /* |
4479 | * Don't normalize the request if it can fit in one extent so | |
4480 | * that it doesn't get unnecessarily split into multiple | |
4481 | * extents. | |
4482 | */ | |
4483 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4484 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f | 4485 | |
a2df2a63 AA |
4486 | retry: |
4487 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4488 | map.m_lblk = map.m_lblk + ret; |
4489 | map.m_len = max_blocks = max_blocks - ret; | |
9924a92a TT |
4490 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4491 | credits); | |
a2df2a63 AA |
4492 | if (IS_ERR(handle)) { |
4493 | ret = PTR_ERR(handle); | |
4494 | break; | |
4495 | } | |
a4e5d88b | 4496 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4497 | if (ret <= 0) { |
2c98615d | 4498 | #ifdef EXT4FS_DEBUG |
b06acd38 LC |
4499 | ext4_warning(inode->i_sb, |
4500 | "inode #%lu: block %u: len %u: " | |
4501 | "ext4_ext_map_blocks returned %d", | |
4502 | inode->i_ino, map.m_lblk, | |
4503 | map.m_len, ret); | |
2c98615d | 4504 | #endif |
a2df2a63 AA |
4505 | ext4_mark_inode_dirty(handle, inode); |
4506 | ret2 = ext4_journal_stop(handle); | |
4507 | break; | |
4508 | } | |
2ed88685 | 4509 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4510 | blkbits) >> blkbits)) |
4511 | new_size = offset + len; | |
4512 | else | |
29ae07b7 | 4513 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4514 | |
fd28784a | 4515 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4516 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4517 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4518 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4519 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4520 | ret2 = ext4_journal_stop(handle); |
4521 | if (ret2) | |
4522 | break; | |
4523 | } | |
fd28784a AK |
4524 | if (ret == -ENOSPC && |
4525 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4526 | ret = 0; | |
a2df2a63 | 4527 | goto retry; |
a2df2a63 | 4528 | } |
55bd725a | 4529 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4530 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4531 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4532 | return ret > 0 ? ret2 : ret; |
4533 | } | |
6873fa0d | 4534 | |
0031462b MC |
4535 | /* |
4536 | * This function convert a range of blocks to written extents | |
4537 | * The caller of this function will pass the start offset and the size. | |
4538 | * all unwritten extents within this range will be converted to | |
4539 | * written extents. | |
4540 | * | |
4541 | * This function is called from the direct IO end io call back | |
4542 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4543 | * Returns 0 on success. |
0031462b MC |
4544 | */ |
4545 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | |
a1de02dc | 4546 | ssize_t len) |
0031462b MC |
4547 | { |
4548 | handle_t *handle; | |
0031462b MC |
4549 | unsigned int max_blocks; |
4550 | int ret = 0; | |
4551 | int ret2 = 0; | |
2ed88685 | 4552 | struct ext4_map_blocks map; |
0031462b MC |
4553 | unsigned int credits, blkbits = inode->i_blkbits; |
4554 | ||
2ed88685 | 4555 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4556 | /* |
4557 | * We can't just convert len to max_blocks because | |
4558 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4559 | */ | |
2ed88685 TT |
4560 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4561 | map.m_lblk); | |
0031462b MC |
4562 | /* |
4563 | * credits to insert 1 extent into extent tree | |
4564 | */ | |
4565 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4566 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4567 | map.m_lblk += ret; |
4568 | map.m_len = (max_blocks -= ret); | |
9924a92a | 4569 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, credits); |
0031462b MC |
4570 | if (IS_ERR(handle)) { |
4571 | ret = PTR_ERR(handle); | |
4572 | break; | |
4573 | } | |
2ed88685 | 4574 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4575 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4576 | if (ret <= 0) |
4577 | ext4_warning(inode->i_sb, | |
4578 | "inode #%lu: block %u: len %u: " | |
4579 | "ext4_ext_map_blocks returned %d", | |
4580 | inode->i_ino, map.m_lblk, | |
4581 | map.m_len, ret); | |
0031462b MC |
4582 | ext4_mark_inode_dirty(handle, inode); |
4583 | ret2 = ext4_journal_stop(handle); | |
4584 | if (ret <= 0 || ret2 ) | |
4585 | break; | |
4586 | } | |
4587 | return ret > 0 ? ret2 : ret; | |
4588 | } | |
6d9c85eb | 4589 | |
6873fa0d | 4590 | /* |
69eb33dc ZL |
4591 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
4592 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
4593 | * return start of the next delayed extent. |
4594 | * | |
69eb33dc | 4595 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 4596 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 4597 | * extent found. Leave newes unmodified. |
6873fa0d | 4598 | */ |
91dd8c11 | 4599 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 4600 | struct extent_status *newes) |
6873fa0d | 4601 | { |
b3aff3e3 | 4602 | struct extent_status es; |
be401363 | 4603 | ext4_lblk_t block, next_del; |
6873fa0d | 4604 | |
69eb33dc | 4605 | ext4_es_find_delayed_extent(inode, newes->es_lblk, &es); |
6873fa0d | 4606 | |
69eb33dc | 4607 | if (newes->es_pblk == 0) { |
6d9c85eb | 4608 | /* |
69eb33dc | 4609 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 4610 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 4611 | */ |
06b0c886 | 4612 | if (es.es_len == 0) |
b3aff3e3 | 4613 | /* A hole found. */ |
91dd8c11 | 4614 | return 0; |
b3aff3e3 | 4615 | |
69eb33dc | 4616 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 4617 | /* A hole found. */ |
69eb33dc ZL |
4618 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
4619 | newes->es_len); | |
91dd8c11 | 4620 | return 0; |
6873fa0d | 4621 | } |
6d9c85eb | 4622 | |
69eb33dc | 4623 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
4624 | } |
4625 | ||
69eb33dc | 4626 | block = newes->es_lblk + newes->es_len; |
be401363 ZL |
4627 | ext4_es_find_delayed_extent(inode, block, &es); |
4628 | if (es.es_len == 0) | |
4629 | next_del = EXT_MAX_BLOCKS; | |
4630 | else | |
4631 | next_del = es.es_lblk; | |
4632 | ||
91dd8c11 | 4633 | return next_del; |
6873fa0d | 4634 | } |
6873fa0d ES |
4635 | /* fiemap flags we can handle specified here */ |
4636 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4637 | ||
3a06d778 AK |
4638 | static int ext4_xattr_fiemap(struct inode *inode, |
4639 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4640 | { |
4641 | __u64 physical = 0; | |
4642 | __u64 length; | |
4643 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4644 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4645 | int error = 0; | |
4646 | ||
4647 | /* in-inode? */ | |
19f5fb7a | 4648 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4649 | struct ext4_iloc iloc; |
4650 | int offset; /* offset of xattr in inode */ | |
4651 | ||
4652 | error = ext4_get_inode_loc(inode, &iloc); | |
4653 | if (error) | |
4654 | return error; | |
4655 | physical = iloc.bh->b_blocknr << blockbits; | |
4656 | offset = EXT4_GOOD_OLD_INODE_SIZE + | |
4657 | EXT4_I(inode)->i_extra_isize; | |
4658 | physical += offset; | |
4659 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4660 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4661 | brelse(iloc.bh); |
6873fa0d ES |
4662 | } else { /* external block */ |
4663 | physical = EXT4_I(inode)->i_file_acl << blockbits; | |
4664 | length = inode->i_sb->s_blocksize; | |
4665 | } | |
4666 | ||
4667 | if (physical) | |
4668 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4669 | length, flags); | |
4670 | return (error < 0 ? error : 0); | |
4671 | } | |
4672 | ||
4673 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
4674 | __u64 start, __u64 len) | |
4675 | { | |
4676 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4677 | int error = 0; |
4678 | ||
94191985 TM |
4679 | if (ext4_has_inline_data(inode)) { |
4680 | int has_inline = 1; | |
4681 | ||
4682 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline); | |
4683 | ||
4684 | if (has_inline) | |
4685 | return error; | |
4686 | } | |
4687 | ||
6873fa0d | 4688 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 4689 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4690 | return generic_block_fiemap(inode, fieinfo, start, len, |
4691 | ext4_get_block); | |
4692 | ||
4693 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4694 | return -EBADR; | |
4695 | ||
4696 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4697 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4698 | } else { | |
aca92ff6 LM |
4699 | ext4_lblk_t len_blks; |
4700 | __u64 last_blk; | |
4701 | ||
6873fa0d | 4702 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4703 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4704 | if (last_blk >= EXT_MAX_BLOCKS) |
4705 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4706 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4707 | |
4708 | /* | |
91dd8c11 LC |
4709 | * Walk the extent tree gathering extent information |
4710 | * and pushing extents back to the user. | |
6873fa0d | 4711 | */ |
91dd8c11 LC |
4712 | error = ext4_fill_fiemap_extents(inode, start_blk, |
4713 | len_blks, fieinfo); | |
6873fa0d ES |
4714 | } |
4715 | ||
4716 | return error; | |
4717 | } | |
996bb9fd TT |
4718 | |
4719 | /* | |
4720 | * Migrate a simple extent-based inode to use the i_blocks[] array | |
4721 | */ | |
4722 | int ext4_ind_migrate(struct inode *inode) | |
4723 | { | |
4724 | struct ext4_extent_header *eh; | |
4725 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; | |
4726 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4727 | struct ext4_extent *ex; | |
4728 | unsigned int i, len; | |
4729 | ext4_fsblk_t blk; | |
4730 | handle_t *handle; | |
4731 | int ret; | |
4732 | ||
4733 | if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb, | |
4734 | EXT4_FEATURE_INCOMPAT_EXTENTS) || | |
4735 | (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4736 | return -EINVAL; | |
4737 | ||
e8238f9a DM |
4738 | handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1); |
4739 | if (IS_ERR(handle)) | |
4740 | return PTR_ERR(handle); | |
4741 | ||
996bb9fd TT |
4742 | down_write(&EXT4_I(inode)->i_data_sem); |
4743 | ret = ext4_ext_check_inode(inode); | |
4744 | if (ret) | |
4745 | goto errout; | |
4746 | ||
4747 | eh = ext_inode_hdr(inode); | |
4748 | ex = EXT_FIRST_EXTENT(eh); | |
4749 | if (ext4_blocks_count(es) > EXT4_MAX_BLOCK_FILE_PHYS || | |
4750 | eh->eh_depth != 0 || eh->eh_entries > 1) { | |
4751 | ret = -EOPNOTSUPP; | |
4752 | goto errout; | |
4753 | } | |
4754 | if (eh->eh_entries == 0) | |
4755 | blk = len = 0; | |
4756 | else { | |
4757 | len = le16_to_cpu(ex->ee_len); | |
4758 | blk = ext4_ext_pblock(ex); | |
4759 | if (len > EXT4_NDIR_BLOCKS) { | |
4760 | ret = -EOPNOTSUPP; | |
4761 | goto errout; | |
4762 | } | |
4763 | } | |
4764 | ||
996bb9fd TT |
4765 | ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); |
4766 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
4767 | for (i=0; i < len; i++) | |
4768 | ei->i_data[i] = cpu_to_le32(blk++); | |
4769 | ext4_mark_inode_dirty(handle, inode); | |
996bb9fd | 4770 | errout: |
e8238f9a | 4771 | ext4_journal_stop(handle); |
996bb9fd TT |
4772 | up_write(&EXT4_I(inode)->i_data_sem); |
4773 | return ret; | |
4774 | } |