]>
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> | |
a86c6181 | 40 | #include <asm/uaccess.h> |
6873fa0d | 41 | #include <linux/fiemap.h> |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
4a092d73 | 43 | #include "ext4_extents.h" |
f19d5870 | 44 | #include "xattr.h" |
a86c6181 | 45 | |
0562e0ba JZ |
46 | #include <trace/events/ext4.h> |
47 | ||
5f95d21f LC |
48 | /* |
49 | * used by extent splitting. | |
50 | */ | |
51 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
52 | due to ENOSPC */ | |
556615dc LC |
53 | #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */ |
54 | #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */ | |
5f95d21f | 55 | |
dee1f973 DM |
56 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
57 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
58 | ||
7ac5990d DW |
59 | static __le32 ext4_extent_block_csum(struct inode *inode, |
60 | struct ext4_extent_header *eh) | |
61 | { | |
62 | struct ext4_inode_info *ei = EXT4_I(inode); | |
63 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
64 | __u32 csum; | |
65 | ||
66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
67 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
68 | return cpu_to_le32(csum); | |
69 | } | |
70 | ||
71 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
72 | struct ext4_extent_header *eh) | |
73 | { | |
74 | struct ext4_extent_tail *et; | |
75 | ||
76 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
77 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
78 | return 1; | |
79 | ||
80 | et = find_ext4_extent_tail(eh); | |
81 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
82 | return 0; | |
83 | return 1; | |
84 | } | |
85 | ||
86 | static void ext4_extent_block_csum_set(struct inode *inode, | |
87 | struct ext4_extent_header *eh) | |
88 | { | |
89 | struct ext4_extent_tail *et; | |
90 | ||
91 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
92 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
93 | return; | |
94 | ||
95 | et = find_ext4_extent_tail(eh); | |
96 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
97 | } | |
98 | ||
d583fb87 AH |
99 | static int ext4_split_extent(handle_t *handle, |
100 | struct inode *inode, | |
101 | struct ext4_ext_path *path, | |
102 | struct ext4_map_blocks *map, | |
103 | int split_flag, | |
104 | int flags); | |
105 | ||
5f95d21f LC |
106 | static int ext4_split_extent_at(handle_t *handle, |
107 | struct inode *inode, | |
108 | struct ext4_ext_path *path, | |
109 | ext4_lblk_t split, | |
110 | int split_flag, | |
111 | int flags); | |
112 | ||
91dd8c11 | 113 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 114 | struct extent_status *newes); |
91dd8c11 | 115 | |
487caeef JK |
116 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
117 | struct inode *inode, | |
118 | int needed) | |
a86c6181 AT |
119 | { |
120 | int err; | |
121 | ||
0390131b FM |
122 | if (!ext4_handle_valid(handle)) |
123 | return 0; | |
a86c6181 | 124 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
125 | return 0; |
126 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 127 | if (err <= 0) |
9102e4fa | 128 | return err; |
487caeef | 129 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
130 | if (err == 0) |
131 | err = -EAGAIN; | |
487caeef JK |
132 | |
133 | return err; | |
a86c6181 AT |
134 | } |
135 | ||
136 | /* | |
137 | * could return: | |
138 | * - EROFS | |
139 | * - ENOMEM | |
140 | */ | |
141 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
142 | struct ext4_ext_path *path) | |
143 | { | |
144 | if (path->p_bh) { | |
145 | /* path points to block */ | |
5d601255 | 146 | BUFFER_TRACE(path->p_bh, "get_write_access"); |
a86c6181 AT |
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 | */ | |
2656497b DW |
160 | int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle, |
161 | struct inode *inode, struct ext4_ext_path *path) | |
a86c6181 AT |
162 | { |
163 | int err; | |
4b1f1660 DM |
164 | |
165 | WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
a86c6181 | 166 | if (path->p_bh) { |
7ac5990d | 167 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 168 | /* path points to block */ |
9ea7a0df TT |
169 | err = __ext4_handle_dirty_metadata(where, line, handle, |
170 | inode, path->p_bh); | |
a86c6181 AT |
171 | } else { |
172 | /* path points to leaf/index in inode body */ | |
173 | err = ext4_mark_inode_dirty(handle, inode); | |
174 | } | |
175 | return err; | |
176 | } | |
177 | ||
f65e6fba | 178 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 179 | struct ext4_ext_path *path, |
725d26d3 | 180 | ext4_lblk_t block) |
a86c6181 | 181 | { |
a86c6181 | 182 | if (path) { |
81fdbb4a | 183 | int depth = path->p_depth; |
a86c6181 | 184 | struct ext4_extent *ex; |
a86c6181 | 185 | |
ad4fb9ca KM |
186 | /* |
187 | * Try to predict block placement assuming that we are | |
188 | * filling in a file which will eventually be | |
189 | * non-sparse --- i.e., in the case of libbfd writing | |
190 | * an ELF object sections out-of-order but in a way | |
191 | * the eventually results in a contiguous object or | |
192 | * executable file, or some database extending a table | |
193 | * space file. However, this is actually somewhat | |
194 | * non-ideal if we are writing a sparse file such as | |
195 | * qemu or KVM writing a raw image file that is going | |
196 | * to stay fairly sparse, since it will end up | |
197 | * fragmenting the file system's free space. Maybe we | |
198 | * should have some hueristics or some way to allow | |
199 | * userspace to pass a hint to file system, | |
b8d6568a | 200 | * especially if the latter case turns out to be |
ad4fb9ca KM |
201 | * common. |
202 | */ | |
7e028976 | 203 | ex = path[depth].p_ext; |
ad4fb9ca KM |
204 | if (ex) { |
205 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
206 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
207 | ||
208 | if (block > ext_block) | |
209 | return ext_pblk + (block - ext_block); | |
210 | else | |
211 | return ext_pblk - (ext_block - block); | |
212 | } | |
a86c6181 | 213 | |
d0d856e8 RD |
214 | /* it looks like index is empty; |
215 | * try to find starting block from index itself */ | |
a86c6181 AT |
216 | if (path[depth].p_bh) |
217 | return path[depth].p_bh->b_blocknr; | |
218 | } | |
219 | ||
220 | /* OK. use inode's group */ | |
f86186b4 | 221 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
222 | } |
223 | ||
654b4908 AK |
224 | /* |
225 | * Allocation for a meta data block | |
226 | */ | |
f65e6fba | 227 | static ext4_fsblk_t |
654b4908 | 228 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 229 | struct ext4_ext_path *path, |
55f020db | 230 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 231 | { |
f65e6fba | 232 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
233 | |
234 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
235 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
236 | NULL, err); | |
a86c6181 AT |
237 | return newblock; |
238 | } | |
239 | ||
55ad63bf | 240 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
241 | { |
242 | int size; | |
243 | ||
244 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
245 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 246 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
247 | if (!check && size > 6) |
248 | size = 6; | |
a86c6181 AT |
249 | #endif |
250 | return size; | |
251 | } | |
252 | ||
55ad63bf | 253 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
254 | { |
255 | int size; | |
256 | ||
257 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
258 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 259 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
260 | if (!check && size > 5) |
261 | size = 5; | |
a86c6181 AT |
262 | #endif |
263 | return size; | |
264 | } | |
265 | ||
55ad63bf | 266 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
267 | { |
268 | int size; | |
269 | ||
270 | size = sizeof(EXT4_I(inode)->i_data); | |
271 | size -= sizeof(struct ext4_extent_header); | |
272 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 273 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
274 | if (!check && size > 3) |
275 | size = 3; | |
a86c6181 AT |
276 | #endif |
277 | return size; | |
278 | } | |
279 | ||
55ad63bf | 280 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
281 | { |
282 | int size; | |
283 | ||
284 | size = sizeof(EXT4_I(inode)->i_data); | |
285 | size -= sizeof(struct ext4_extent_header); | |
286 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 287 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
288 | if (!check && size > 4) |
289 | size = 4; | |
a86c6181 AT |
290 | #endif |
291 | return size; | |
292 | } | |
293 | ||
fcf6b1b7 DM |
294 | static inline int |
295 | ext4_force_split_extent_at(handle_t *handle, struct inode *inode, | |
296 | struct ext4_ext_path *path, ext4_lblk_t lblk, | |
297 | int nofail) | |
298 | { | |
299 | int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext); | |
300 | ||
301 | return ext4_split_extent_at(handle, inode, path, lblk, unwritten ? | |
302 | EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0, | |
303 | EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO | | |
304 | (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0)); | |
305 | } | |
306 | ||
d2a17637 MC |
307 | /* |
308 | * Calculate the number of metadata blocks needed | |
309 | * to allocate @blocks | |
310 | * Worse case is one block per extent | |
311 | */ | |
01f49d0b | 312 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 313 | { |
9d0be502 | 314 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 315 | int idxs; |
d2a17637 | 316 | |
9d0be502 TT |
317 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
318 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
319 | |
320 | /* | |
9d0be502 TT |
321 | * If the new delayed allocation block is contiguous with the |
322 | * previous da block, it can share index blocks with the | |
323 | * previous block, so we only need to allocate a new index | |
324 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
325 | * an additional index block, and at ldxs**3 blocks, yet | |
326 | * another index blocks. | |
d2a17637 | 327 | */ |
9d0be502 TT |
328 | if (ei->i_da_metadata_calc_len && |
329 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
330 | int num = 0; |
331 | ||
9d0be502 TT |
332 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
333 | num++; | |
334 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
335 | num++; | |
336 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
337 | num++; | |
338 | ei->i_da_metadata_calc_len = 0; | |
339 | } else | |
340 | ei->i_da_metadata_calc_len++; | |
341 | ei->i_da_metadata_calc_last_lblock++; | |
342 | return num; | |
343 | } | |
d2a17637 | 344 | |
9d0be502 TT |
345 | /* |
346 | * In the worst case we need a new set of index blocks at | |
347 | * every level of the inode's extent tree. | |
348 | */ | |
349 | ei->i_da_metadata_calc_len = 1; | |
350 | ei->i_da_metadata_calc_last_lblock = lblock; | |
351 | return ext_depth(inode) + 1; | |
d2a17637 MC |
352 | } |
353 | ||
c29c0ae7 AT |
354 | static int |
355 | ext4_ext_max_entries(struct inode *inode, int depth) | |
356 | { | |
357 | int max; | |
358 | ||
359 | if (depth == ext_depth(inode)) { | |
360 | if (depth == 0) | |
55ad63bf | 361 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 362 | else |
55ad63bf | 363 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
364 | } else { |
365 | if (depth == 0) | |
55ad63bf | 366 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 367 | else |
55ad63bf | 368 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
369 | } |
370 | ||
371 | return max; | |
372 | } | |
373 | ||
56b19868 AK |
374 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
375 | { | |
bf89d16f | 376 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 377 | int len = ext4_ext_get_actual_len(ext); |
5946d089 EG |
378 | ext4_lblk_t lblock = le32_to_cpu(ext->ee_block); |
379 | ext4_lblk_t last = lblock + len - 1; | |
e84a26ce | 380 | |
5946d089 | 381 | if (lblock > last) |
31d4f3a2 | 382 | return 0; |
6fd058f7 | 383 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
384 | } |
385 | ||
386 | static int ext4_valid_extent_idx(struct inode *inode, | |
387 | struct ext4_extent_idx *ext_idx) | |
388 | { | |
bf89d16f | 389 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 390 | |
6fd058f7 | 391 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
392 | } |
393 | ||
394 | static int ext4_valid_extent_entries(struct inode *inode, | |
395 | struct ext4_extent_header *eh, | |
396 | int depth) | |
397 | { | |
56b19868 AK |
398 | unsigned short entries; |
399 | if (eh->eh_entries == 0) | |
400 | return 1; | |
401 | ||
402 | entries = le16_to_cpu(eh->eh_entries); | |
403 | ||
404 | if (depth == 0) { | |
405 | /* leaf entries */ | |
81fdbb4a | 406 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
5946d089 EG |
407 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
408 | ext4_fsblk_t pblock = 0; | |
409 | ext4_lblk_t lblock = 0; | |
410 | ext4_lblk_t prev = 0; | |
411 | int len = 0; | |
56b19868 AK |
412 | while (entries) { |
413 | if (!ext4_valid_extent(inode, ext)) | |
414 | return 0; | |
5946d089 EG |
415 | |
416 | /* Check for overlapping extents */ | |
417 | lblock = le32_to_cpu(ext->ee_block); | |
418 | len = ext4_ext_get_actual_len(ext); | |
419 | if ((lblock <= prev) && prev) { | |
420 | pblock = ext4_ext_pblock(ext); | |
421 | es->s_last_error_block = cpu_to_le64(pblock); | |
422 | return 0; | |
423 | } | |
56b19868 AK |
424 | ext++; |
425 | entries--; | |
5946d089 | 426 | prev = lblock + len - 1; |
56b19868 AK |
427 | } |
428 | } else { | |
81fdbb4a | 429 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
430 | while (entries) { |
431 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
432 | return 0; | |
433 | ext_idx++; | |
434 | entries--; | |
435 | } | |
436 | } | |
437 | return 1; | |
438 | } | |
439 | ||
c398eda0 TT |
440 | static int __ext4_ext_check(const char *function, unsigned int line, |
441 | struct inode *inode, struct ext4_extent_header *eh, | |
c349179b | 442 | int depth, ext4_fsblk_t pblk) |
c29c0ae7 AT |
443 | { |
444 | const char *error_msg; | |
445 | int max = 0; | |
446 | ||
447 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
448 | error_msg = "invalid magic"; | |
449 | goto corrupted; | |
450 | } | |
451 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
452 | error_msg = "unexpected eh_depth"; | |
453 | goto corrupted; | |
454 | } | |
455 | if (unlikely(eh->eh_max == 0)) { | |
456 | error_msg = "invalid eh_max"; | |
457 | goto corrupted; | |
458 | } | |
459 | max = ext4_ext_max_entries(inode, depth); | |
460 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
461 | error_msg = "too large eh_max"; | |
462 | goto corrupted; | |
463 | } | |
464 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
465 | error_msg = "invalid eh_entries"; | |
466 | goto corrupted; | |
467 | } | |
56b19868 AK |
468 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
469 | error_msg = "invalid extent entries"; | |
470 | goto corrupted; | |
471 | } | |
7ac5990d DW |
472 | /* Verify checksum on non-root extent tree nodes */ |
473 | if (ext_depth(inode) != depth && | |
474 | !ext4_extent_block_csum_verify(inode, eh)) { | |
475 | error_msg = "extent tree corrupted"; | |
476 | goto corrupted; | |
477 | } | |
c29c0ae7 AT |
478 | return 0; |
479 | ||
480 | corrupted: | |
c398eda0 | 481 | ext4_error_inode(inode, function, line, 0, |
c349179b TT |
482 | "pblk %llu bad header/extent: %s - magic %x, " |
483 | "entries %u, max %u(%u), depth %u(%u)", | |
484 | (unsigned long long) pblk, error_msg, | |
485 | le16_to_cpu(eh->eh_magic), | |
486 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
487 | max, le16_to_cpu(eh->eh_depth), depth); | |
c29c0ae7 AT |
488 | return -EIO; |
489 | } | |
490 | ||
c349179b TT |
491 | #define ext4_ext_check(inode, eh, depth, pblk) \ |
492 | __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk)) | |
c29c0ae7 | 493 | |
7a262f7c AK |
494 | int ext4_ext_check_inode(struct inode *inode) |
495 | { | |
c349179b | 496 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
7a262f7c AK |
497 | } |
498 | ||
7d7ea89e TT |
499 | static struct buffer_head * |
500 | __read_extent_tree_block(const char *function, unsigned int line, | |
107a7bd3 TT |
501 | struct inode *inode, ext4_fsblk_t pblk, int depth, |
502 | int flags) | |
f8489128 | 503 | { |
7d7ea89e TT |
504 | struct buffer_head *bh; |
505 | int err; | |
506 | ||
507 | bh = sb_getblk(inode->i_sb, pblk); | |
508 | if (unlikely(!bh)) | |
509 | return ERR_PTR(-ENOMEM); | |
f8489128 | 510 | |
7d7ea89e TT |
511 | if (!bh_uptodate_or_lock(bh)) { |
512 | trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); | |
513 | err = bh_submit_read(bh); | |
514 | if (err < 0) | |
515 | goto errout; | |
516 | } | |
7869a4a6 | 517 | if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE)) |
7d7ea89e TT |
518 | return bh; |
519 | err = __ext4_ext_check(function, line, inode, | |
c349179b | 520 | ext_block_hdr(bh), depth, pblk); |
7d7ea89e TT |
521 | if (err) |
522 | goto errout; | |
f8489128 | 523 | set_buffer_verified(bh); |
107a7bd3 TT |
524 | /* |
525 | * If this is a leaf block, cache all of its entries | |
526 | */ | |
527 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { | |
528 | struct ext4_extent_header *eh = ext_block_hdr(bh); | |
529 | struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); | |
530 | ext4_lblk_t prev = 0; | |
531 | int i; | |
532 | ||
533 | for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { | |
534 | unsigned int status = EXTENT_STATUS_WRITTEN; | |
535 | ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); | |
536 | int len = ext4_ext_get_actual_len(ex); | |
537 | ||
538 | if (prev && (prev != lblk)) | |
539 | ext4_es_cache_extent(inode, prev, | |
540 | lblk - prev, ~0, | |
541 | EXTENT_STATUS_HOLE); | |
542 | ||
556615dc | 543 | if (ext4_ext_is_unwritten(ex)) |
107a7bd3 TT |
544 | status = EXTENT_STATUS_UNWRITTEN; |
545 | ext4_es_cache_extent(inode, lblk, len, | |
546 | ext4_ext_pblock(ex), status); | |
547 | prev = lblk + len; | |
548 | } | |
549 | } | |
7d7ea89e TT |
550 | return bh; |
551 | errout: | |
552 | put_bh(bh); | |
553 | return ERR_PTR(err); | |
554 | ||
f8489128 DW |
555 | } |
556 | ||
107a7bd3 TT |
557 | #define read_extent_tree_block(inode, pblk, depth, flags) \ |
558 | __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \ | |
559 | (depth), (flags)) | |
f8489128 | 560 | |
7869a4a6 TT |
561 | /* |
562 | * This function is called to cache a file's extent information in the | |
563 | * extent status tree | |
564 | */ | |
565 | int ext4_ext_precache(struct inode *inode) | |
566 | { | |
567 | struct ext4_inode_info *ei = EXT4_I(inode); | |
568 | struct ext4_ext_path *path = NULL; | |
569 | struct buffer_head *bh; | |
570 | int i = 0, depth, ret = 0; | |
571 | ||
572 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
573 | return 0; /* not an extent-mapped inode */ | |
574 | ||
575 | down_read(&ei->i_data_sem); | |
576 | depth = ext_depth(inode); | |
577 | ||
578 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
579 | GFP_NOFS); | |
580 | if (path == NULL) { | |
581 | up_read(&ei->i_data_sem); | |
582 | return -ENOMEM; | |
583 | } | |
584 | ||
585 | /* Don't cache anything if there are no external extent blocks */ | |
586 | if (depth == 0) | |
587 | goto out; | |
588 | path[0].p_hdr = ext_inode_hdr(inode); | |
589 | ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0); | |
590 | if (ret) | |
591 | goto out; | |
592 | path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr); | |
593 | while (i >= 0) { | |
594 | /* | |
595 | * If this is a leaf block or we've reached the end of | |
596 | * the index block, go up | |
597 | */ | |
598 | if ((i == depth) || | |
599 | path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) { | |
600 | brelse(path[i].p_bh); | |
601 | path[i].p_bh = NULL; | |
602 | i--; | |
603 | continue; | |
604 | } | |
605 | bh = read_extent_tree_block(inode, | |
606 | ext4_idx_pblock(path[i].p_idx++), | |
607 | depth - i - 1, | |
608 | EXT4_EX_FORCE_CACHE); | |
609 | if (IS_ERR(bh)) { | |
610 | ret = PTR_ERR(bh); | |
611 | break; | |
612 | } | |
613 | i++; | |
614 | path[i].p_bh = bh; | |
615 | path[i].p_hdr = ext_block_hdr(bh); | |
616 | path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr); | |
617 | } | |
618 | ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED); | |
619 | out: | |
620 | up_read(&ei->i_data_sem); | |
621 | ext4_ext_drop_refs(path); | |
622 | kfree(path); | |
623 | return ret; | |
624 | } | |
625 | ||
a86c6181 AT |
626 | #ifdef EXT_DEBUG |
627 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
628 | { | |
629 | int k, l = path->p_depth; | |
630 | ||
631 | ext_debug("path:"); | |
632 | for (k = 0; k <= l; k++, path++) { | |
633 | if (path->p_idx) { | |
2ae02107 | 634 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 635 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 636 | } else if (path->p_ext) { |
553f9008 | 637 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 638 | le32_to_cpu(path->p_ext->ee_block), |
556615dc | 639 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 640 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 641 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
642 | } else |
643 | ext_debug(" []"); | |
644 | } | |
645 | ext_debug("\n"); | |
646 | } | |
647 | ||
648 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
649 | { | |
650 | int depth = ext_depth(inode); | |
651 | struct ext4_extent_header *eh; | |
652 | struct ext4_extent *ex; | |
653 | int i; | |
654 | ||
655 | if (!path) | |
656 | return; | |
657 | ||
658 | eh = path[depth].p_hdr; | |
659 | ex = EXT_FIRST_EXTENT(eh); | |
660 | ||
553f9008 M |
661 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
662 | ||
a86c6181 | 663 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 | 664 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
556615dc | 665 | ext4_ext_is_unwritten(ex), |
bf89d16f | 666 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
667 | } |
668 | ext_debug("\n"); | |
669 | } | |
1b16da77 YY |
670 | |
671 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
672 | ext4_fsblk_t newblock, int level) | |
673 | { | |
674 | int depth = ext_depth(inode); | |
675 | struct ext4_extent *ex; | |
676 | ||
677 | if (depth != level) { | |
678 | struct ext4_extent_idx *idx; | |
679 | idx = path[level].p_idx; | |
680 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
681 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
682 | le32_to_cpu(idx->ei_block), | |
683 | ext4_idx_pblock(idx), | |
684 | newblock); | |
685 | idx++; | |
686 | } | |
687 | ||
688 | return; | |
689 | } | |
690 | ||
691 | ex = path[depth].p_ext; | |
692 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
693 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
694 | le32_to_cpu(ex->ee_block), | |
695 | ext4_ext_pblock(ex), | |
556615dc | 696 | ext4_ext_is_unwritten(ex), |
1b16da77 YY |
697 | ext4_ext_get_actual_len(ex), |
698 | newblock); | |
699 | ex++; | |
700 | } | |
701 | } | |
702 | ||
a86c6181 | 703 | #else |
af5bc92d TT |
704 | #define ext4_ext_show_path(inode, path) |
705 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 706 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
707 | #endif |
708 | ||
b35905c1 | 709 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
710 | { |
711 | int depth = path->p_depth; | |
712 | int i; | |
713 | ||
714 | for (i = 0; i <= depth; i++, path++) | |
715 | if (path->p_bh) { | |
716 | brelse(path->p_bh); | |
717 | path->p_bh = NULL; | |
718 | } | |
719 | } | |
720 | ||
721 | /* | |
d0d856e8 RD |
722 | * ext4_ext_binsearch_idx: |
723 | * binary search for the closest index of the given block | |
c29c0ae7 | 724 | * the header must be checked before calling this |
a86c6181 AT |
725 | */ |
726 | static void | |
725d26d3 AK |
727 | ext4_ext_binsearch_idx(struct inode *inode, |
728 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
729 | { |
730 | struct ext4_extent_header *eh = path->p_hdr; | |
731 | struct ext4_extent_idx *r, *l, *m; | |
732 | ||
a86c6181 | 733 | |
bba90743 | 734 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
735 | |
736 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 737 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
738 | while (l <= r) { |
739 | m = l + (r - l) / 2; | |
740 | if (block < le32_to_cpu(m->ei_block)) | |
741 | r = m - 1; | |
742 | else | |
743 | l = m + 1; | |
26d535ed DM |
744 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
745 | m, le32_to_cpu(m->ei_block), | |
746 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
747 | } |
748 | ||
749 | path->p_idx = l - 1; | |
4a3c3a51 | 750 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 751 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
752 | |
753 | #ifdef CHECK_BINSEARCH | |
754 | { | |
755 | struct ext4_extent_idx *chix, *ix; | |
756 | int k; | |
757 | ||
758 | chix = ix = EXT_FIRST_INDEX(eh); | |
759 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
760 | if (k != 0 && | |
761 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
762 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
763 | "first=0x%p\n", k, | |
764 | ix, EXT_FIRST_INDEX(eh)); | |
765 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
766 | le32_to_cpu(ix->ei_block), |
767 | le32_to_cpu(ix[-1].ei_block)); | |
768 | } | |
769 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 770 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
771 | if (block < le32_to_cpu(ix->ei_block)) |
772 | break; | |
773 | chix = ix; | |
774 | } | |
775 | BUG_ON(chix != path->p_idx); | |
776 | } | |
777 | #endif | |
778 | ||
779 | } | |
780 | ||
781 | /* | |
d0d856e8 RD |
782 | * ext4_ext_binsearch: |
783 | * binary search for closest extent of the given block | |
c29c0ae7 | 784 | * the header must be checked before calling this |
a86c6181 AT |
785 | */ |
786 | static void | |
725d26d3 AK |
787 | ext4_ext_binsearch(struct inode *inode, |
788 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
789 | { |
790 | struct ext4_extent_header *eh = path->p_hdr; | |
791 | struct ext4_extent *r, *l, *m; | |
792 | ||
a86c6181 AT |
793 | if (eh->eh_entries == 0) { |
794 | /* | |
d0d856e8 RD |
795 | * this leaf is empty: |
796 | * we get such a leaf in split/add case | |
a86c6181 AT |
797 | */ |
798 | return; | |
799 | } | |
800 | ||
bba90743 | 801 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
802 | |
803 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 804 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
805 | |
806 | while (l <= r) { | |
807 | m = l + (r - l) / 2; | |
808 | if (block < le32_to_cpu(m->ee_block)) | |
809 | r = m - 1; | |
810 | else | |
811 | l = m + 1; | |
26d535ed DM |
812 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
813 | m, le32_to_cpu(m->ee_block), | |
814 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
815 | } |
816 | ||
817 | path->p_ext = l - 1; | |
553f9008 | 818 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 819 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 820 | ext4_ext_pblock(path->p_ext), |
556615dc | 821 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 822 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
823 | |
824 | #ifdef CHECK_BINSEARCH | |
825 | { | |
826 | struct ext4_extent *chex, *ex; | |
827 | int k; | |
828 | ||
829 | chex = ex = EXT_FIRST_EXTENT(eh); | |
830 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
831 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 832 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
833 | if (block < le32_to_cpu(ex->ee_block)) |
834 | break; | |
835 | chex = ex; | |
836 | } | |
837 | BUG_ON(chex != path->p_ext); | |
838 | } | |
839 | #endif | |
840 | ||
841 | } | |
842 | ||
843 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
844 | { | |
845 | struct ext4_extent_header *eh; | |
846 | ||
847 | eh = ext_inode_hdr(inode); | |
848 | eh->eh_depth = 0; | |
849 | eh->eh_entries = 0; | |
850 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 851 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 852 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
853 | return 0; |
854 | } | |
855 | ||
856 | struct ext4_ext_path * | |
725d26d3 | 857 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
705912ca | 858 | struct ext4_ext_path **orig_path, int flags) |
a86c6181 AT |
859 | { |
860 | struct ext4_extent_header *eh; | |
861 | struct buffer_head *bh; | |
705912ca TT |
862 | struct ext4_ext_path *path = orig_path ? *orig_path : NULL; |
863 | short int depth, i, ppos = 0; | |
864 | short free_on_err = (flags & EXT4_EX_NOFREE_ON_ERR) == 0; | |
860d21e2 | 865 | int ret; |
a86c6181 AT |
866 | |
867 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 868 | depth = ext_depth(inode); |
a86c6181 AT |
869 | |
870 | /* account possible depth increase */ | |
871 | if (!path) { | |
5d4958f9 | 872 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 | 873 | GFP_NOFS); |
19008f6d | 874 | if (unlikely(!path)) |
a86c6181 | 875 | return ERR_PTR(-ENOMEM); |
705912ca | 876 | free_on_err = 1; |
a86c6181 | 877 | } |
a86c6181 | 878 | path[0].p_hdr = eh; |
1973adcb | 879 | path[0].p_bh = NULL; |
a86c6181 | 880 | |
c29c0ae7 | 881 | i = depth; |
a86c6181 AT |
882 | /* walk through the tree */ |
883 | while (i) { | |
884 | ext_debug("depth %d: num %d, max %d\n", | |
885 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 886 | |
a86c6181 | 887 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 888 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
889 | path[ppos].p_depth = i; |
890 | path[ppos].p_ext = NULL; | |
891 | ||
107a7bd3 TT |
892 | bh = read_extent_tree_block(inode, path[ppos].p_block, --i, |
893 | flags); | |
19008f6d | 894 | if (unlikely(IS_ERR(bh))) { |
7d7ea89e | 895 | ret = PTR_ERR(bh); |
a86c6181 | 896 | goto err; |
860d21e2 | 897 | } |
7d7ea89e | 898 | |
a86c6181 AT |
899 | eh = ext_block_hdr(bh); |
900 | ppos++; | |
273df556 FM |
901 | if (unlikely(ppos > depth)) { |
902 | put_bh(bh); | |
903 | EXT4_ERROR_INODE(inode, | |
904 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 905 | ret = -EIO; |
273df556 FM |
906 | goto err; |
907 | } | |
a86c6181 AT |
908 | path[ppos].p_bh = bh; |
909 | path[ppos].p_hdr = eh; | |
a86c6181 AT |
910 | } |
911 | ||
912 | path[ppos].p_depth = i; | |
a86c6181 AT |
913 | path[ppos].p_ext = NULL; |
914 | path[ppos].p_idx = NULL; | |
915 | ||
a86c6181 AT |
916 | /* find extent */ |
917 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
918 | /* if not an empty leaf */ |
919 | if (path[ppos].p_ext) | |
bf89d16f | 920 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
921 | |
922 | ext4_ext_show_path(inode, path); | |
923 | ||
924 | return path; | |
925 | ||
926 | err: | |
927 | ext4_ext_drop_refs(path); | |
705912ca | 928 | if (free_on_err) { |
a86c6181 | 929 | kfree(path); |
705912ca TT |
930 | if (orig_path) |
931 | *orig_path = NULL; | |
932 | } | |
860d21e2 | 933 | return ERR_PTR(ret); |
a86c6181 AT |
934 | } |
935 | ||
936 | /* | |
d0d856e8 RD |
937 | * ext4_ext_insert_index: |
938 | * insert new index [@logical;@ptr] into the block at @curp; | |
939 | * check where to insert: before @curp or after @curp | |
a86c6181 | 940 | */ |
1f109d5a TT |
941 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
942 | struct ext4_ext_path *curp, | |
943 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
944 | { |
945 | struct ext4_extent_idx *ix; | |
946 | int len, err; | |
947 | ||
7e028976 AM |
948 | err = ext4_ext_get_access(handle, inode, curp); |
949 | if (err) | |
a86c6181 AT |
950 | return err; |
951 | ||
273df556 FM |
952 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
953 | EXT4_ERROR_INODE(inode, | |
954 | "logical %d == ei_block %d!", | |
955 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
956 | return -EIO; | |
957 | } | |
d4620315 RD |
958 | |
959 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
960 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
961 | EXT4_ERROR_INODE(inode, | |
962 | "eh_entries %d >= eh_max %d!", | |
963 | le16_to_cpu(curp->p_hdr->eh_entries), | |
964 | le16_to_cpu(curp->p_hdr->eh_max)); | |
965 | return -EIO; | |
966 | } | |
967 | ||
a86c6181 AT |
968 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
969 | /* insert after */ | |
80e675f9 | 970 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
971 | ix = curp->p_idx + 1; |
972 | } else { | |
973 | /* insert before */ | |
80e675f9 | 974 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
975 | ix = curp->p_idx; |
976 | } | |
977 | ||
80e675f9 EG |
978 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
979 | BUG_ON(len < 0); | |
980 | if (len > 0) { | |
981 | ext_debug("insert new index %d: " | |
982 | "move %d indices from 0x%p to 0x%p\n", | |
983 | logical, len, ix, ix + 1); | |
984 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
985 | } | |
986 | ||
f472e026 TM |
987 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
988 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
989 | return -EIO; | |
990 | } | |
991 | ||
a86c6181 | 992 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 993 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 994 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 995 | |
273df556 FM |
996 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
997 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
998 | return -EIO; | |
999 | } | |
a86c6181 AT |
1000 | |
1001 | err = ext4_ext_dirty(handle, inode, curp); | |
1002 | ext4_std_error(inode->i_sb, err); | |
1003 | ||
1004 | return err; | |
1005 | } | |
1006 | ||
1007 | /* | |
d0d856e8 RD |
1008 | * ext4_ext_split: |
1009 | * inserts new subtree into the path, using free index entry | |
1010 | * at depth @at: | |
1011 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
1012 | * - makes decision where to split | |
1013 | * - moves remaining extents and index entries (right to the split point) | |
1014 | * into the newly allocated blocks | |
1015 | * - initializes subtree | |
a86c6181 AT |
1016 | */ |
1017 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
1018 | unsigned int flags, |
1019 | struct ext4_ext_path *path, | |
1020 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
1021 | { |
1022 | struct buffer_head *bh = NULL; | |
1023 | int depth = ext_depth(inode); | |
1024 | struct ext4_extent_header *neh; | |
1025 | struct ext4_extent_idx *fidx; | |
a86c6181 | 1026 | int i = at, k, m, a; |
f65e6fba | 1027 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 1028 | __le32 border; |
f65e6fba | 1029 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
1030 | int err = 0; |
1031 | ||
1032 | /* make decision: where to split? */ | |
d0d856e8 | 1033 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 1034 | |
d0d856e8 | 1035 | /* if current leaf will be split, then we should use |
a86c6181 | 1036 | * border from split point */ |
273df556 FM |
1037 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
1038 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
1039 | return -EIO; | |
1040 | } | |
a86c6181 AT |
1041 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
1042 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 1043 | ext_debug("leaf will be split." |
a86c6181 | 1044 | " next leaf starts at %d\n", |
8c55e204 | 1045 | le32_to_cpu(border)); |
a86c6181 AT |
1046 | } else { |
1047 | border = newext->ee_block; | |
1048 | ext_debug("leaf will be added." | |
1049 | " next leaf starts at %d\n", | |
8c55e204 | 1050 | le32_to_cpu(border)); |
a86c6181 AT |
1051 | } |
1052 | ||
1053 | /* | |
d0d856e8 RD |
1054 | * If error occurs, then we break processing |
1055 | * and mark filesystem read-only. index won't | |
a86c6181 | 1056 | * be inserted and tree will be in consistent |
d0d856e8 | 1057 | * state. Next mount will repair buffers too. |
a86c6181 AT |
1058 | */ |
1059 | ||
1060 | /* | |
d0d856e8 RD |
1061 | * Get array to track all allocated blocks. |
1062 | * We need this to handle errors and free blocks | |
1063 | * upon them. | |
a86c6181 | 1064 | */ |
5d4958f9 | 1065 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
1066 | if (!ablocks) |
1067 | return -ENOMEM; | |
a86c6181 AT |
1068 | |
1069 | /* allocate all needed blocks */ | |
1070 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
1071 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 1072 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 1073 | newext, &err, flags); |
a86c6181 AT |
1074 | if (newblock == 0) |
1075 | goto cleanup; | |
1076 | ablocks[a] = newblock; | |
1077 | } | |
1078 | ||
1079 | /* initialize new leaf */ | |
1080 | newblock = ablocks[--a]; | |
273df556 FM |
1081 | if (unlikely(newblock == 0)) { |
1082 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
1083 | err = -EIO; | |
1084 | goto cleanup; | |
1085 | } | |
a86c6181 | 1086 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1087 | if (unlikely(!bh)) { |
860d21e2 | 1088 | err = -ENOMEM; |
a86c6181 AT |
1089 | goto cleanup; |
1090 | } | |
1091 | lock_buffer(bh); | |
1092 | ||
7e028976 AM |
1093 | err = ext4_journal_get_create_access(handle, bh); |
1094 | if (err) | |
a86c6181 AT |
1095 | goto cleanup; |
1096 | ||
1097 | neh = ext_block_hdr(bh); | |
1098 | neh->eh_entries = 0; | |
55ad63bf | 1099 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1100 | neh->eh_magic = EXT4_EXT_MAGIC; |
1101 | neh->eh_depth = 0; | |
a86c6181 | 1102 | |
d0d856e8 | 1103 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
1104 | if (unlikely(path[depth].p_hdr->eh_entries != |
1105 | path[depth].p_hdr->eh_max)) { | |
1106 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
1107 | path[depth].p_hdr->eh_entries, | |
1108 | path[depth].p_hdr->eh_max); | |
1109 | err = -EIO; | |
1110 | goto cleanup; | |
1111 | } | |
a86c6181 | 1112 | /* start copy from next extent */ |
1b16da77 YY |
1113 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
1114 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 1115 | if (m) { |
1b16da77 YY |
1116 | struct ext4_extent *ex; |
1117 | ex = EXT_FIRST_EXTENT(neh); | |
1118 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 1119 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1120 | } |
1121 | ||
7ac5990d | 1122 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1123 | set_buffer_uptodate(bh); |
1124 | unlock_buffer(bh); | |
1125 | ||
0390131b | 1126 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1127 | if (err) |
a86c6181 AT |
1128 | goto cleanup; |
1129 | brelse(bh); | |
1130 | bh = NULL; | |
1131 | ||
1132 | /* correct old leaf */ | |
1133 | if (m) { | |
7e028976 AM |
1134 | err = ext4_ext_get_access(handle, inode, path + depth); |
1135 | if (err) | |
a86c6181 | 1136 | goto cleanup; |
e8546d06 | 1137 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1138 | err = ext4_ext_dirty(handle, inode, path + depth); |
1139 | if (err) | |
a86c6181 AT |
1140 | goto cleanup; |
1141 | ||
1142 | } | |
1143 | ||
1144 | /* create intermediate indexes */ | |
1145 | k = depth - at - 1; | |
273df556 FM |
1146 | if (unlikely(k < 0)) { |
1147 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1148 | err = -EIO; | |
1149 | goto cleanup; | |
1150 | } | |
a86c6181 AT |
1151 | if (k) |
1152 | ext_debug("create %d intermediate indices\n", k); | |
1153 | /* insert new index into current index block */ | |
1154 | /* current depth stored in i var */ | |
1155 | i = depth - 1; | |
1156 | while (k--) { | |
1157 | oldblock = newblock; | |
1158 | newblock = ablocks[--a]; | |
bba90743 | 1159 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1160 | if (unlikely(!bh)) { |
860d21e2 | 1161 | err = -ENOMEM; |
a86c6181 AT |
1162 | goto cleanup; |
1163 | } | |
1164 | lock_buffer(bh); | |
1165 | ||
7e028976 AM |
1166 | err = ext4_journal_get_create_access(handle, bh); |
1167 | if (err) | |
a86c6181 AT |
1168 | goto cleanup; |
1169 | ||
1170 | neh = ext_block_hdr(bh); | |
1171 | neh->eh_entries = cpu_to_le16(1); | |
1172 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1173 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1174 | neh->eh_depth = cpu_to_le16(depth - i); |
1175 | fidx = EXT_FIRST_INDEX(neh); | |
1176 | fidx->ei_block = border; | |
f65e6fba | 1177 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1178 | |
bba90743 ES |
1179 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1180 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1181 | |
1b16da77 | 1182 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1183 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1184 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1185 | EXT4_ERROR_INODE(inode, | |
1186 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1187 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1188 | err = -EIO; | |
1189 | goto cleanup; | |
1190 | } | |
1b16da77 YY |
1191 | /* start copy indexes */ |
1192 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1193 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1194 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1195 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1196 | if (m) { |
1b16da77 | 1197 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1198 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1199 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1200 | } |
7ac5990d | 1201 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1202 | set_buffer_uptodate(bh); |
1203 | unlock_buffer(bh); | |
1204 | ||
0390131b | 1205 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1206 | if (err) |
a86c6181 AT |
1207 | goto cleanup; |
1208 | brelse(bh); | |
1209 | bh = NULL; | |
1210 | ||
1211 | /* correct old index */ | |
1212 | if (m) { | |
1213 | err = ext4_ext_get_access(handle, inode, path + i); | |
1214 | if (err) | |
1215 | goto cleanup; | |
e8546d06 | 1216 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1217 | err = ext4_ext_dirty(handle, inode, path + i); |
1218 | if (err) | |
1219 | goto cleanup; | |
1220 | } | |
1221 | ||
1222 | i--; | |
1223 | } | |
1224 | ||
1225 | /* insert new index */ | |
a86c6181 AT |
1226 | err = ext4_ext_insert_index(handle, inode, path + at, |
1227 | le32_to_cpu(border), newblock); | |
1228 | ||
1229 | cleanup: | |
1230 | if (bh) { | |
1231 | if (buffer_locked(bh)) | |
1232 | unlock_buffer(bh); | |
1233 | brelse(bh); | |
1234 | } | |
1235 | ||
1236 | if (err) { | |
1237 | /* free all allocated blocks in error case */ | |
1238 | for (i = 0; i < depth; i++) { | |
1239 | if (!ablocks[i]) | |
1240 | continue; | |
7dc57615 | 1241 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1242 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1243 | } |
1244 | } | |
1245 | kfree(ablocks); | |
1246 | ||
1247 | return err; | |
1248 | } | |
1249 | ||
1250 | /* | |
d0d856e8 RD |
1251 | * ext4_ext_grow_indepth: |
1252 | * implements tree growing procedure: | |
1253 | * - allocates new block | |
1254 | * - moves top-level data (index block or leaf) into the new block | |
1255 | * - initializes new top-level, creating index that points to the | |
1256 | * just created block | |
a86c6181 AT |
1257 | */ |
1258 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1259 | unsigned int flags, |
55f020db | 1260 | struct ext4_extent *newext) |
a86c6181 | 1261 | { |
a86c6181 | 1262 | struct ext4_extent_header *neh; |
a86c6181 | 1263 | struct buffer_head *bh; |
f65e6fba | 1264 | ext4_fsblk_t newblock; |
a86c6181 AT |
1265 | int err = 0; |
1266 | ||
1939dd84 | 1267 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1268 | newext, &err, flags); |
a86c6181 AT |
1269 | if (newblock == 0) |
1270 | return err; | |
1271 | ||
1272 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1273 | if (unlikely(!bh)) |
860d21e2 | 1274 | return -ENOMEM; |
a86c6181 AT |
1275 | lock_buffer(bh); |
1276 | ||
7e028976 AM |
1277 | err = ext4_journal_get_create_access(handle, bh); |
1278 | if (err) { | |
a86c6181 AT |
1279 | unlock_buffer(bh); |
1280 | goto out; | |
1281 | } | |
1282 | ||
1283 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1284 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1285 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1286 | |
1287 | /* set size of new block */ | |
1288 | neh = ext_block_hdr(bh); | |
1289 | /* old root could have indexes or leaves | |
1290 | * so calculate e_max right way */ | |
1291 | if (ext_depth(inode)) | |
55ad63bf | 1292 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1293 | else |
55ad63bf | 1294 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1295 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1296 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1297 | set_buffer_uptodate(bh); |
1298 | unlock_buffer(bh); | |
1299 | ||
0390131b | 1300 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1301 | if (err) |
a86c6181 AT |
1302 | goto out; |
1303 | ||
1939dd84 | 1304 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1305 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1306 | neh->eh_entries = cpu_to_le16(1); |
1307 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1308 | if (neh->eh_depth == 0) { | |
1309 | /* Root extent block becomes index block */ | |
1310 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1311 | EXT_FIRST_INDEX(neh)->ei_block = | |
1312 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1313 | } | |
2ae02107 | 1314 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1315 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1316 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1317 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1318 | |
ba39ebb6 | 1319 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1320 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1321 | out: |
1322 | brelse(bh); | |
1323 | ||
1324 | return err; | |
1325 | } | |
1326 | ||
1327 | /* | |
d0d856e8 RD |
1328 | * ext4_ext_create_new_leaf: |
1329 | * finds empty index and adds new leaf. | |
1330 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1331 | */ |
1332 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
107a7bd3 TT |
1333 | unsigned int mb_flags, |
1334 | unsigned int gb_flags, | |
55f020db AH |
1335 | struct ext4_ext_path *path, |
1336 | struct ext4_extent *newext) | |
a86c6181 AT |
1337 | { |
1338 | struct ext4_ext_path *curp; | |
1339 | int depth, i, err = 0; | |
1340 | ||
1341 | repeat: | |
1342 | i = depth = ext_depth(inode); | |
1343 | ||
1344 | /* walk up to the tree and look for free index entry */ | |
1345 | curp = path + depth; | |
1346 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1347 | i--; | |
1348 | curp--; | |
1349 | } | |
1350 | ||
d0d856e8 RD |
1351 | /* we use already allocated block for index block, |
1352 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1353 | if (EXT_HAS_FREE_INDEX(curp)) { |
1354 | /* if we found index with free entry, then use that | |
1355 | * entry: create all needed subtree and add new leaf */ | |
107a7bd3 | 1356 | err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
787e0981 SF |
1357 | if (err) |
1358 | goto out; | |
a86c6181 AT |
1359 | |
1360 | /* refill path */ | |
1361 | ext4_ext_drop_refs(path); | |
1362 | path = ext4_ext_find_extent(inode, | |
725d26d3 | 1363 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
705912ca | 1364 | &path, gb_flags | EXT4_EX_NOFREE_ON_ERR); |
a86c6181 AT |
1365 | if (IS_ERR(path)) |
1366 | err = PTR_ERR(path); | |
1367 | } else { | |
1368 | /* tree is full, time to grow in depth */ | |
107a7bd3 | 1369 | err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext); |
a86c6181 AT |
1370 | if (err) |
1371 | goto out; | |
1372 | ||
1373 | /* refill path */ | |
1374 | ext4_ext_drop_refs(path); | |
1375 | path = ext4_ext_find_extent(inode, | |
725d26d3 | 1376 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
705912ca | 1377 | &path, gb_flags | EXT4_EX_NOFREE_ON_ERR); |
a86c6181 AT |
1378 | if (IS_ERR(path)) { |
1379 | err = PTR_ERR(path); | |
1380 | goto out; | |
1381 | } | |
1382 | ||
1383 | /* | |
d0d856e8 RD |
1384 | * only first (depth 0 -> 1) produces free space; |
1385 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1386 | */ |
1387 | depth = ext_depth(inode); | |
1388 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1389 | /* now we need to split */ |
a86c6181 AT |
1390 | goto repeat; |
1391 | } | |
1392 | } | |
1393 | ||
1394 | out: | |
1395 | return err; | |
1396 | } | |
1397 | ||
1988b51e AT |
1398 | /* |
1399 | * search the closest allocated block to the left for *logical | |
1400 | * and returns it at @logical + it's physical address at @phys | |
1401 | * if *logical is the smallest allocated block, the function | |
1402 | * returns 0 at @phys | |
1403 | * return value contains 0 (success) or error code | |
1404 | */ | |
1f109d5a TT |
1405 | static int ext4_ext_search_left(struct inode *inode, |
1406 | struct ext4_ext_path *path, | |
1407 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1408 | { |
1409 | struct ext4_extent_idx *ix; | |
1410 | struct ext4_extent *ex; | |
b939e376 | 1411 | int depth, ee_len; |
1988b51e | 1412 | |
273df556 FM |
1413 | if (unlikely(path == NULL)) { |
1414 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1415 | return -EIO; | |
1416 | } | |
1988b51e AT |
1417 | depth = path->p_depth; |
1418 | *phys = 0; | |
1419 | ||
1420 | if (depth == 0 && path->p_ext == NULL) | |
1421 | return 0; | |
1422 | ||
1423 | /* usually extent in the path covers blocks smaller | |
1424 | * then *logical, but it can be that extent is the | |
1425 | * first one in the file */ | |
1426 | ||
1427 | ex = path[depth].p_ext; | |
b939e376 | 1428 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1429 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1430 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1431 | EXT4_ERROR_INODE(inode, | |
1432 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1433 | *logical, le32_to_cpu(ex->ee_block)); | |
1434 | return -EIO; | |
1435 | } | |
1988b51e AT |
1436 | while (--depth >= 0) { |
1437 | ix = path[depth].p_idx; | |
273df556 FM |
1438 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1439 | EXT4_ERROR_INODE(inode, | |
1440 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1441 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1442 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1443 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1444 | depth); |
1445 | return -EIO; | |
1446 | } | |
1988b51e AT |
1447 | } |
1448 | return 0; | |
1449 | } | |
1450 | ||
273df556 FM |
1451 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1452 | EXT4_ERROR_INODE(inode, | |
1453 | "logical %d < ee_block %d + ee_len %d!", | |
1454 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1455 | return -EIO; | |
1456 | } | |
1988b51e | 1457 | |
b939e376 | 1458 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1459 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1460 | return 0; |
1461 | } | |
1462 | ||
1463 | /* | |
1464 | * search the closest allocated block to the right for *logical | |
1465 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1466 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1467 | * returns 0 at @phys |
1468 | * return value contains 0 (success) or error code | |
1469 | */ | |
1f109d5a TT |
1470 | static int ext4_ext_search_right(struct inode *inode, |
1471 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1472 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1473 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1474 | { |
1475 | struct buffer_head *bh = NULL; | |
1476 | struct ext4_extent_header *eh; | |
1477 | struct ext4_extent_idx *ix; | |
1478 | struct ext4_extent *ex; | |
1479 | ext4_fsblk_t block; | |
395a87bf ES |
1480 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1481 | int ee_len; | |
1988b51e | 1482 | |
273df556 FM |
1483 | if (unlikely(path == NULL)) { |
1484 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1485 | return -EIO; | |
1486 | } | |
1988b51e AT |
1487 | depth = path->p_depth; |
1488 | *phys = 0; | |
1489 | ||
1490 | if (depth == 0 && path->p_ext == NULL) | |
1491 | return 0; | |
1492 | ||
1493 | /* usually extent in the path covers blocks smaller | |
1494 | * then *logical, but it can be that extent is the | |
1495 | * first one in the file */ | |
1496 | ||
1497 | ex = path[depth].p_ext; | |
b939e376 | 1498 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1499 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1500 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1501 | EXT4_ERROR_INODE(inode, | |
1502 | "first_extent(path[%d].p_hdr) != ex", | |
1503 | depth); | |
1504 | return -EIO; | |
1505 | } | |
1988b51e AT |
1506 | while (--depth >= 0) { |
1507 | ix = path[depth].p_idx; | |
273df556 FM |
1508 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1509 | EXT4_ERROR_INODE(inode, | |
1510 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1511 | *logical); | |
1512 | return -EIO; | |
1513 | } | |
1988b51e | 1514 | } |
4d33b1ef | 1515 | goto found_extent; |
1988b51e AT |
1516 | } |
1517 | ||
273df556 FM |
1518 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1519 | EXT4_ERROR_INODE(inode, | |
1520 | "logical %d < ee_block %d + ee_len %d!", | |
1521 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1522 | return -EIO; | |
1523 | } | |
1988b51e AT |
1524 | |
1525 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1526 | /* next allocated block in this leaf */ | |
1527 | ex++; | |
4d33b1ef | 1528 | goto found_extent; |
1988b51e AT |
1529 | } |
1530 | ||
1531 | /* go up and search for index to the right */ | |
1532 | while (--depth >= 0) { | |
1533 | ix = path[depth].p_idx; | |
1534 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1535 | goto got_index; |
1988b51e AT |
1536 | } |
1537 | ||
25f1ee3a WF |
1538 | /* we've gone up to the root and found no index to the right */ |
1539 | return 0; | |
1988b51e | 1540 | |
25f1ee3a | 1541 | got_index: |
1988b51e AT |
1542 | /* we've found index to the right, let's |
1543 | * follow it and find the closest allocated | |
1544 | * block to the right */ | |
1545 | ix++; | |
bf89d16f | 1546 | block = ext4_idx_pblock(ix); |
1988b51e | 1547 | while (++depth < path->p_depth) { |
395a87bf | 1548 | /* subtract from p_depth to get proper eh_depth */ |
7d7ea89e | 1549 | bh = read_extent_tree_block(inode, block, |
107a7bd3 | 1550 | path->p_depth - depth, 0); |
7d7ea89e TT |
1551 | if (IS_ERR(bh)) |
1552 | return PTR_ERR(bh); | |
1553 | eh = ext_block_hdr(bh); | |
1988b51e | 1554 | ix = EXT_FIRST_INDEX(eh); |
bf89d16f | 1555 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1556 | put_bh(bh); |
1557 | } | |
1558 | ||
107a7bd3 | 1559 | bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0); |
7d7ea89e TT |
1560 | if (IS_ERR(bh)) |
1561 | return PTR_ERR(bh); | |
1988b51e | 1562 | eh = ext_block_hdr(bh); |
1988b51e | 1563 | ex = EXT_FIRST_EXTENT(eh); |
4d33b1ef | 1564 | found_extent: |
1988b51e | 1565 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1566 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1567 | *ret_ex = ex; |
1568 | if (bh) | |
1569 | put_bh(bh); | |
1988b51e | 1570 | return 0; |
1988b51e AT |
1571 | } |
1572 | ||
a86c6181 | 1573 | /* |
d0d856e8 | 1574 | * ext4_ext_next_allocated_block: |
f17722f9 | 1575 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1576 | * NOTE: it considers block number from index entry as |
1577 | * allocated block. Thus, index entries have to be consistent | |
1578 | * with leaves. | |
a86c6181 | 1579 | */ |
fcf6b1b7 | 1580 | ext4_lblk_t |
a86c6181 AT |
1581 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1582 | { | |
1583 | int depth; | |
1584 | ||
1585 | BUG_ON(path == NULL); | |
1586 | depth = path->p_depth; | |
1587 | ||
1588 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1589 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1590 | |
1591 | while (depth >= 0) { | |
1592 | if (depth == path->p_depth) { | |
1593 | /* leaf */ | |
6f8ff537 CW |
1594 | if (path[depth].p_ext && |
1595 | path[depth].p_ext != | |
a86c6181 AT |
1596 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1597 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1598 | } else { | |
1599 | /* index */ | |
1600 | if (path[depth].p_idx != | |
1601 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1602 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1603 | } | |
1604 | depth--; | |
1605 | } | |
1606 | ||
f17722f9 | 1607 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1608 | } |
1609 | ||
1610 | /* | |
d0d856e8 | 1611 | * ext4_ext_next_leaf_block: |
f17722f9 | 1612 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1613 | */ |
5718789d | 1614 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1615 | { |
1616 | int depth; | |
1617 | ||
1618 | BUG_ON(path == NULL); | |
1619 | depth = path->p_depth; | |
1620 | ||
1621 | /* zero-tree has no leaf blocks at all */ | |
1622 | if (depth == 0) | |
f17722f9 | 1623 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1624 | |
1625 | /* go to index block */ | |
1626 | depth--; | |
1627 | ||
1628 | while (depth >= 0) { | |
1629 | if (path[depth].p_idx != | |
1630 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1631 | return (ext4_lblk_t) |
1632 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1633 | depth--; |
1634 | } | |
1635 | ||
f17722f9 | 1636 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1637 | } |
1638 | ||
1639 | /* | |
d0d856e8 RD |
1640 | * ext4_ext_correct_indexes: |
1641 | * if leaf gets modified and modified extent is first in the leaf, | |
1642 | * then we have to correct all indexes above. | |
a86c6181 AT |
1643 | * TODO: do we need to correct tree in all cases? |
1644 | */ | |
1d03ec98 | 1645 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1646 | struct ext4_ext_path *path) |
1647 | { | |
1648 | struct ext4_extent_header *eh; | |
1649 | int depth = ext_depth(inode); | |
1650 | struct ext4_extent *ex; | |
1651 | __le32 border; | |
1652 | int k, err = 0; | |
1653 | ||
1654 | eh = path[depth].p_hdr; | |
1655 | ex = path[depth].p_ext; | |
273df556 FM |
1656 | |
1657 | if (unlikely(ex == NULL || eh == NULL)) { | |
1658 | EXT4_ERROR_INODE(inode, | |
1659 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1660 | return -EIO; | |
1661 | } | |
a86c6181 AT |
1662 | |
1663 | if (depth == 0) { | |
1664 | /* there is no tree at all */ | |
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1669 | /* we correct tree if first leaf got modified only */ | |
1670 | return 0; | |
1671 | } | |
1672 | ||
1673 | /* | |
d0d856e8 | 1674 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1675 | */ |
1676 | k = depth - 1; | |
1677 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1678 | err = ext4_ext_get_access(handle, inode, path + k); |
1679 | if (err) | |
a86c6181 AT |
1680 | return err; |
1681 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1682 | err = ext4_ext_dirty(handle, inode, path + k); |
1683 | if (err) | |
a86c6181 AT |
1684 | return err; |
1685 | ||
1686 | while (k--) { | |
1687 | /* change all left-side indexes */ | |
1688 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1689 | break; | |
7e028976 AM |
1690 | err = ext4_ext_get_access(handle, inode, path + k); |
1691 | if (err) | |
a86c6181 AT |
1692 | break; |
1693 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1694 | err = ext4_ext_dirty(handle, inode, path + k); |
1695 | if (err) | |
a86c6181 AT |
1696 | break; |
1697 | } | |
1698 | ||
1699 | return err; | |
1700 | } | |
1701 | ||
748de673 | 1702 | int |
a86c6181 AT |
1703 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1704 | struct ext4_extent *ex2) | |
1705 | { | |
da0169b3 | 1706 | unsigned short ext1_ee_len, ext2_ee_len; |
a2df2a63 AA |
1707 | |
1708 | /* | |
ec22ba8e | 1709 | * Make sure that both extents are initialized. We don't merge |
556615dc | 1710 | * unwritten extents so that we can be sure that end_io code has |
ec22ba8e DM |
1711 | * the extent that was written properly split out and conversion to |
1712 | * initialized is trivial. | |
a2df2a63 | 1713 | */ |
556615dc | 1714 | if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2)) |
a2df2a63 AA |
1715 | return 0; |
1716 | ||
1717 | ext1_ee_len = ext4_ext_get_actual_len(ex1); | |
1718 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1719 | ||
1720 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1721 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1722 | return 0; |
1723 | ||
471d4011 SB |
1724 | /* |
1725 | * To allow future support for preallocated extents to be added | |
1726 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1727 | * this can result in the top bit of ee_len being set. |
471d4011 | 1728 | */ |
da0169b3 | 1729 | if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN) |
471d4011 | 1730 | return 0; |
556615dc | 1731 | if (ext4_ext_is_unwritten(ex1) && |
a9b82415 DW |
1732 | (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) || |
1733 | atomic_read(&EXT4_I(inode)->i_unwritten) || | |
556615dc | 1734 | (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN))) |
a9b82415 | 1735 | return 0; |
bbf2f9fb | 1736 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1737 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1738 | return 0; |
1739 | #endif | |
1740 | ||
bf89d16f | 1741 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1742 | return 1; |
1743 | return 0; | |
1744 | } | |
1745 | ||
56055d3a AA |
1746 | /* |
1747 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1748 | * It always tries to merge towards right. If you want to merge towards | |
1749 | * left, pass "ex - 1" as argument instead of "ex". | |
1750 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1751 | * 1 if they got merged. | |
1752 | */ | |
197217a5 | 1753 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1754 | struct ext4_ext_path *path, |
1755 | struct ext4_extent *ex) | |
56055d3a AA |
1756 | { |
1757 | struct ext4_extent_header *eh; | |
1758 | unsigned int depth, len; | |
556615dc | 1759 | int merge_done = 0, unwritten; |
56055d3a AA |
1760 | |
1761 | depth = ext_depth(inode); | |
1762 | BUG_ON(path[depth].p_hdr == NULL); | |
1763 | eh = path[depth].p_hdr; | |
1764 | ||
1765 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1766 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1767 | break; | |
1768 | /* merge with next extent! */ | |
556615dc | 1769 | unwritten = ext4_ext_is_unwritten(ex); |
56055d3a AA |
1770 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
1771 | + ext4_ext_get_actual_len(ex + 1)); | |
556615dc LC |
1772 | if (unwritten) |
1773 | ext4_ext_mark_unwritten(ex); | |
56055d3a AA |
1774 | |
1775 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1776 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1777 | * sizeof(struct ext4_extent); | |
1778 | memmove(ex + 1, ex + 2, len); | |
1779 | } | |
e8546d06 | 1780 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1781 | merge_done = 1; |
1782 | WARN_ON(eh->eh_entries == 0); | |
1783 | if (!eh->eh_entries) | |
24676da4 | 1784 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1785 | } |
1786 | ||
1787 | return merge_done; | |
1788 | } | |
1789 | ||
ecb94f5f TT |
1790 | /* |
1791 | * This function does a very simple check to see if we can collapse | |
1792 | * an extent tree with a single extent tree leaf block into the inode. | |
1793 | */ | |
1794 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1795 | struct inode *inode, | |
1796 | struct ext4_ext_path *path) | |
1797 | { | |
1798 | size_t s; | |
1799 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1800 | ext4_fsblk_t blk; | |
1801 | ||
1802 | if ((path[0].p_depth != 1) || | |
1803 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1804 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1805 | return; | |
1806 | ||
1807 | /* | |
1808 | * We need to modify the block allocation bitmap and the block | |
1809 | * group descriptor to release the extent tree block. If we | |
1810 | * can't get the journal credits, give up. | |
1811 | */ | |
1812 | if (ext4_journal_extend(handle, 2)) | |
1813 | return; | |
1814 | ||
1815 | /* | |
1816 | * Copy the extent data up to the inode | |
1817 | */ | |
1818 | blk = ext4_idx_pblock(path[0].p_idx); | |
1819 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1820 | sizeof(struct ext4_extent_idx); | |
1821 | s += sizeof(struct ext4_extent_header); | |
1822 | ||
1823 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1824 | path[0].p_depth = 0; | |
1825 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1826 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1827 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1828 | ||
1829 | brelse(path[1].p_bh); | |
1830 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
71d4f7d0 | 1831 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
ecb94f5f TT |
1832 | } |
1833 | ||
197217a5 YY |
1834 | /* |
1835 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1836 | * return 1 if merge left else 0. | |
1837 | */ | |
ecb94f5f TT |
1838 | static void ext4_ext_try_to_merge(handle_t *handle, |
1839 | struct inode *inode, | |
197217a5 YY |
1840 | struct ext4_ext_path *path, |
1841 | struct ext4_extent *ex) { | |
1842 | struct ext4_extent_header *eh; | |
1843 | unsigned int depth; | |
1844 | int merge_done = 0; | |
197217a5 YY |
1845 | |
1846 | depth = ext_depth(inode); | |
1847 | BUG_ON(path[depth].p_hdr == NULL); | |
1848 | eh = path[depth].p_hdr; | |
1849 | ||
1850 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1851 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1852 | ||
1853 | if (!merge_done) | |
ecb94f5f | 1854 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1855 | |
ecb94f5f | 1856 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1857 | } |
1858 | ||
25d14f98 AA |
1859 | /* |
1860 | * check if a portion of the "newext" extent overlaps with an | |
1861 | * existing extent. | |
1862 | * | |
1863 | * If there is an overlap discovered, it updates the length of the newext | |
1864 | * such that there will be no overlap, and then returns 1. | |
1865 | * If there is no overlap found, it returns 0. | |
1866 | */ | |
4d33b1ef TT |
1867 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1868 | struct inode *inode, | |
1f109d5a TT |
1869 | struct ext4_extent *newext, |
1870 | struct ext4_ext_path *path) | |
25d14f98 | 1871 | { |
725d26d3 | 1872 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1873 | unsigned int depth, len1; |
1874 | unsigned int ret = 0; | |
1875 | ||
1876 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1877 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1878 | depth = ext_depth(inode); |
1879 | if (!path[depth].p_ext) | |
1880 | goto out; | |
f5a44db5 | 1881 | b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block)); |
25d14f98 AA |
1882 | |
1883 | /* | |
1884 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1885 | * is before the requested block(s) |
25d14f98 AA |
1886 | */ |
1887 | if (b2 < b1) { | |
1888 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1889 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1890 | goto out; |
f5a44db5 | 1891 | b2 = EXT4_LBLK_CMASK(sbi, b2); |
25d14f98 AA |
1892 | } |
1893 | ||
725d26d3 | 1894 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1895 | if (b1 + len1 < b1) { |
f17722f9 | 1896 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1897 | newext->ee_len = cpu_to_le16(len1); |
1898 | ret = 1; | |
1899 | } | |
1900 | ||
1901 | /* check for overlap */ | |
1902 | if (b1 + len1 > b2) { | |
1903 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1904 | ret = 1; | |
1905 | } | |
1906 | out: | |
1907 | return ret; | |
1908 | } | |
1909 | ||
a86c6181 | 1910 | /* |
d0d856e8 RD |
1911 | * ext4_ext_insert_extent: |
1912 | * tries to merge requsted extent into the existing extent or | |
1913 | * inserts requested extent as new one into the tree, | |
1914 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1915 | */ |
1916 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1917 | struct ext4_ext_path *path, | |
107a7bd3 | 1918 | struct ext4_extent *newext, int gb_flags) |
a86c6181 | 1919 | { |
af5bc92d | 1920 | struct ext4_extent_header *eh; |
a86c6181 AT |
1921 | struct ext4_extent *ex, *fex; |
1922 | struct ext4_extent *nearex; /* nearest extent */ | |
1923 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1924 | int depth, len, err; |
1925 | ext4_lblk_t next; | |
556615dc | 1926 | int mb_flags = 0, unwritten; |
a86c6181 | 1927 | |
273df556 FM |
1928 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1929 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1930 | return -EIO; | |
1931 | } | |
a86c6181 AT |
1932 | depth = ext_depth(inode); |
1933 | ex = path[depth].p_ext; | |
be8981be | 1934 | eh = path[depth].p_hdr; |
273df556 FM |
1935 | if (unlikely(path[depth].p_hdr == NULL)) { |
1936 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1937 | return -EIO; | |
1938 | } | |
a86c6181 AT |
1939 | |
1940 | /* try to insert block into found extent and return */ | |
107a7bd3 | 1941 | if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1942 | |
1943 | /* | |
be8981be LC |
1944 | * Try to see whether we should rather test the extent on |
1945 | * right from ex, or from the left of ex. This is because | |
1946 | * ext4_ext_find_extent() can return either extent on the | |
1947 | * left, or on the right from the searched position. This | |
1948 | * will make merging more effective. | |
a2df2a63 | 1949 | */ |
be8981be LC |
1950 | if (ex < EXT_LAST_EXTENT(eh) && |
1951 | (le32_to_cpu(ex->ee_block) + | |
1952 | ext4_ext_get_actual_len(ex) < | |
1953 | le32_to_cpu(newext->ee_block))) { | |
1954 | ex += 1; | |
1955 | goto prepend; | |
1956 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1957 | (le32_to_cpu(newext->ee_block) + | |
1958 | ext4_ext_get_actual_len(newext) < | |
1959 | le32_to_cpu(ex->ee_block))) | |
1960 | ex -= 1; | |
1961 | ||
1962 | /* Try to append newex to the ex */ | |
1963 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1964 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
1965 | "(from %llu)\n", | |
556615dc | 1966 | ext4_ext_is_unwritten(newext), |
be8981be LC |
1967 | ext4_ext_get_actual_len(newext), |
1968 | le32_to_cpu(ex->ee_block), | |
556615dc | 1969 | ext4_ext_is_unwritten(ex), |
be8981be LC |
1970 | ext4_ext_get_actual_len(ex), |
1971 | ext4_ext_pblock(ex)); | |
1972 | err = ext4_ext_get_access(handle, inode, | |
1973 | path + depth); | |
1974 | if (err) | |
1975 | return err; | |
556615dc | 1976 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be | 1977 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
a2df2a63 | 1978 | + ext4_ext_get_actual_len(newext)); |
556615dc LC |
1979 | if (unwritten) |
1980 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
1981 | eh = path[depth].p_hdr; |
1982 | nearex = ex; | |
1983 | goto merge; | |
1984 | } | |
1985 | ||
1986 | prepend: | |
1987 | /* Try to prepend newex to the ex */ | |
1988 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
1989 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
1990 | "(from %llu)\n", | |
1991 | le32_to_cpu(newext->ee_block), | |
556615dc | 1992 | ext4_ext_is_unwritten(newext), |
be8981be LC |
1993 | ext4_ext_get_actual_len(newext), |
1994 | le32_to_cpu(ex->ee_block), | |
556615dc | 1995 | ext4_ext_is_unwritten(ex), |
be8981be LC |
1996 | ext4_ext_get_actual_len(ex), |
1997 | ext4_ext_pblock(ex)); | |
1998 | err = ext4_ext_get_access(handle, inode, | |
1999 | path + depth); | |
2000 | if (err) | |
2001 | return err; | |
2002 | ||
556615dc | 2003 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be LC |
2004 | ex->ee_block = newext->ee_block; |
2005 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
2006 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
2007 | + ext4_ext_get_actual_len(newext)); | |
556615dc LC |
2008 | if (unwritten) |
2009 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
2010 | eh = path[depth].p_hdr; |
2011 | nearex = ex; | |
2012 | goto merge; | |
2013 | } | |
a86c6181 AT |
2014 | } |
2015 | ||
a86c6181 AT |
2016 | depth = ext_depth(inode); |
2017 | eh = path[depth].p_hdr; | |
2018 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
2019 | goto has_space; | |
2020 | ||
2021 | /* probably next leaf has space for us? */ | |
2022 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
2023 | next = EXT_MAX_BLOCKS; |
2024 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 2025 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 2026 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 2027 | ext_debug("next leaf block - %u\n", next); |
a86c6181 | 2028 | BUG_ON(npath != NULL); |
107a7bd3 | 2029 | npath = ext4_ext_find_extent(inode, next, NULL, 0); |
a86c6181 AT |
2030 | if (IS_ERR(npath)) |
2031 | return PTR_ERR(npath); | |
2032 | BUG_ON(npath->p_depth != path->p_depth); | |
2033 | eh = npath[depth].p_hdr; | |
2034 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 2035 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
2036 | le16_to_cpu(eh->eh_entries)); |
2037 | path = npath; | |
ffb505ff | 2038 | goto has_space; |
a86c6181 AT |
2039 | } |
2040 | ext_debug("next leaf has no free space(%d,%d)\n", | |
2041 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
2042 | } | |
2043 | ||
2044 | /* | |
d0d856e8 RD |
2045 | * There is no free space in the found leaf. |
2046 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 2047 | */ |
107a7bd3 TT |
2048 | if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
2049 | mb_flags = EXT4_MB_USE_RESERVED; | |
2050 | err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, | |
2051 | path, newext); | |
a86c6181 AT |
2052 | if (err) |
2053 | goto cleanup; | |
2054 | depth = ext_depth(inode); | |
2055 | eh = path[depth].p_hdr; | |
2056 | ||
2057 | has_space: | |
2058 | nearex = path[depth].p_ext; | |
2059 | ||
7e028976 AM |
2060 | err = ext4_ext_get_access(handle, inode, path + depth); |
2061 | if (err) | |
a86c6181 AT |
2062 | goto cleanup; |
2063 | ||
2064 | if (!nearex) { | |
2065 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 2066 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 2067 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2068 | ext4_ext_pblock(newext), |
556615dc | 2069 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2070 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
2071 | nearex = EXT_FIRST_EXTENT(eh); |
2072 | } else { | |
2073 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 2074 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 2075 | /* Insert after */ |
32de6756 YY |
2076 | ext_debug("insert %u:%llu:[%d]%d before: " |
2077 | "nearest %p\n", | |
80e675f9 EG |
2078 | le32_to_cpu(newext->ee_block), |
2079 | ext4_ext_pblock(newext), | |
556615dc | 2080 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2081 | ext4_ext_get_actual_len(newext), |
2082 | nearex); | |
2083 | nearex++; | |
2084 | } else { | |
2085 | /* Insert before */ | |
2086 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
2087 | ext_debug("insert %u:%llu:[%d]%d after: " |
2088 | "nearest %p\n", | |
8c55e204 | 2089 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2090 | ext4_ext_pblock(newext), |
556615dc | 2091 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2092 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
2093 | nearex); |
2094 | } | |
2095 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
2096 | if (len > 0) { | |
32de6756 | 2097 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
2098 | "move %d extents from 0x%p to 0x%p\n", |
2099 | le32_to_cpu(newext->ee_block), | |
2100 | ext4_ext_pblock(newext), | |
556615dc | 2101 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2102 | ext4_ext_get_actual_len(newext), |
2103 | len, nearex, nearex + 1); | |
2104 | memmove(nearex + 1, nearex, | |
2105 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2106 | } |
a86c6181 AT |
2107 | } |
2108 | ||
e8546d06 | 2109 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2110 | path[depth].p_ext = nearex; |
a86c6181 | 2111 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2112 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2113 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2114 | |
2115 | merge: | |
e7bcf823 | 2116 | /* try to merge extents */ |
107a7bd3 | 2117 | if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2118 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2119 | |
a86c6181 AT |
2120 | |
2121 | /* time to correct all indexes above */ | |
2122 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2123 | if (err) | |
2124 | goto cleanup; | |
2125 | ||
ecb94f5f | 2126 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2127 | |
2128 | cleanup: | |
2129 | if (npath) { | |
2130 | ext4_ext_drop_refs(npath); | |
2131 | kfree(npath); | |
2132 | } | |
a86c6181 AT |
2133 | return err; |
2134 | } | |
2135 | ||
91dd8c11 LC |
2136 | static int ext4_fill_fiemap_extents(struct inode *inode, |
2137 | ext4_lblk_t block, ext4_lblk_t num, | |
2138 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
2139 | { |
2140 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 2141 | struct ext4_extent *ex; |
69eb33dc | 2142 | struct extent_status es; |
91dd8c11 | 2143 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 2144 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
2145 | int exists, depth = 0, err = 0; |
2146 | unsigned int flags = 0; | |
2147 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 2148 | |
f17722f9 | 2149 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
2150 | num = last - block; |
2151 | /* find extent for this block */ | |
fab3a549 | 2152 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
2153 | |
2154 | if (path && ext_depth(inode) != depth) { | |
2155 | /* depth was changed. we have to realloc path */ | |
2156 | kfree(path); | |
2157 | path = NULL; | |
2158 | } | |
2159 | ||
705912ca | 2160 | path = ext4_ext_find_extent(inode, block, &path, 0); |
6873fa0d | 2161 | if (IS_ERR(path)) { |
91dd8c11 | 2162 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
2163 | err = PTR_ERR(path); |
2164 | path = NULL; | |
2165 | break; | |
2166 | } | |
2167 | ||
2168 | depth = ext_depth(inode); | |
273df556 | 2169 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2170 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2171 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2172 | err = -EIO; | |
2173 | break; | |
2174 | } | |
6873fa0d ES |
2175 | ex = path[depth].p_ext; |
2176 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2177 | ext4_ext_drop_refs(path); |
6873fa0d | 2178 | |
91dd8c11 | 2179 | flags = 0; |
6873fa0d ES |
2180 | exists = 0; |
2181 | if (!ex) { | |
2182 | /* there is no extent yet, so try to allocate | |
2183 | * all requested space */ | |
2184 | start = block; | |
2185 | end = block + num; | |
2186 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2187 | /* need to allocate space before found extent */ | |
2188 | start = block; | |
2189 | end = le32_to_cpu(ex->ee_block); | |
2190 | if (block + num < end) | |
2191 | end = block + num; | |
2192 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2193 | + ext4_ext_get_actual_len(ex)) { | |
2194 | /* need to allocate space after found extent */ | |
2195 | start = block; | |
2196 | end = block + num; | |
2197 | if (end >= next) | |
2198 | end = next; | |
2199 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2200 | /* | |
2201 | * some part of requested space is covered | |
2202 | * by found extent | |
2203 | */ | |
2204 | start = block; | |
2205 | end = le32_to_cpu(ex->ee_block) | |
2206 | + ext4_ext_get_actual_len(ex); | |
2207 | if (block + num < end) | |
2208 | end = block + num; | |
2209 | exists = 1; | |
2210 | } else { | |
2211 | BUG(); | |
2212 | } | |
2213 | BUG_ON(end <= start); | |
2214 | ||
2215 | if (!exists) { | |
69eb33dc ZL |
2216 | es.es_lblk = start; |
2217 | es.es_len = end - start; | |
2218 | es.es_pblk = 0; | |
6873fa0d | 2219 | } else { |
69eb33dc ZL |
2220 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2221 | es.es_len = ext4_ext_get_actual_len(ex); | |
2222 | es.es_pblk = ext4_ext_pblock(ex); | |
556615dc | 2223 | if (ext4_ext_is_unwritten(ex)) |
91dd8c11 | 2224 | flags |= FIEMAP_EXTENT_UNWRITTEN; |
6873fa0d ES |
2225 | } |
2226 | ||
91dd8c11 | 2227 | /* |
69eb33dc ZL |
2228 | * Find delayed extent and update es accordingly. We call |
2229 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2230 | * last existing extent or not. |
2231 | */ | |
69eb33dc | 2232 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2233 | if (!exists && next_del) { |
2234 | exists = 1; | |
72dac95d JL |
2235 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2236 | FIEMAP_EXTENT_UNKNOWN); | |
91dd8c11 LC |
2237 | } |
2238 | up_read(&EXT4_I(inode)->i_data_sem); | |
2239 | ||
69eb33dc ZL |
2240 | if (unlikely(es.es_len == 0)) { |
2241 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
273df556 FM |
2242 | err = -EIO; |
2243 | break; | |
2244 | } | |
6873fa0d | 2245 | |
f7fec032 ZL |
2246 | /* |
2247 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2248 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2249 | * possible that an extent is with unwritten and delayed | |
2250 | * status due to when an extent is delayed allocated and | |
2251 | * is allocated by fallocate status tree will track both of | |
2252 | * them in a extent. | |
2253 | * | |
2254 | * So we could return a unwritten and delayed extent, and | |
2255 | * its block is equal to 'next'. | |
2256 | */ | |
2257 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2258 | flags |= FIEMAP_EXTENT_LAST; |
2259 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2260 | next != EXT_MAX_BLOCKS)) { | |
2261 | EXT4_ERROR_INODE(inode, | |
2262 | "next extent == %u, next " | |
2263 | "delalloc extent = %u", | |
2264 | next, next_del); | |
2265 | err = -EIO; | |
2266 | break; | |
2267 | } | |
6873fa0d ES |
2268 | } |
2269 | ||
91dd8c11 LC |
2270 | if (exists) { |
2271 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2272 | (__u64)es.es_lblk << blksize_bits, |
2273 | (__u64)es.es_pblk << blksize_bits, | |
2274 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2275 | flags); |
2276 | if (err < 0) | |
2277 | break; | |
2278 | if (err == 1) { | |
2279 | err = 0; | |
2280 | break; | |
2281 | } | |
6873fa0d ES |
2282 | } |
2283 | ||
69eb33dc | 2284 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2285 | } |
2286 | ||
2287 | if (path) { | |
2288 | ext4_ext_drop_refs(path); | |
2289 | kfree(path); | |
2290 | } | |
2291 | ||
2292 | return err; | |
2293 | } | |
2294 | ||
a86c6181 | 2295 | /* |
d0d856e8 RD |
2296 | * ext4_ext_put_gap_in_cache: |
2297 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2298 | * and cache this gap |
2299 | */ | |
09b88252 | 2300 | static void |
a86c6181 | 2301 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2302 | ext4_lblk_t block) |
a86c6181 AT |
2303 | { |
2304 | int depth = ext_depth(inode); | |
27b1b228 AS |
2305 | unsigned long len = 0; |
2306 | ext4_lblk_t lblock = 0; | |
a86c6181 AT |
2307 | struct ext4_extent *ex; |
2308 | ||
2309 | ex = path[depth].p_ext; | |
2310 | if (ex == NULL) { | |
69eb33dc ZL |
2311 | /* |
2312 | * there is no extent yet, so gap is [0;-] and we | |
2313 | * don't cache it | |
2314 | */ | |
a86c6181 AT |
2315 | ext_debug("cache gap(whole file):"); |
2316 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2317 | lblock = block; | |
2318 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2319 | ext_debug("cache gap(before): %u [%u:%u]", |
2320 | block, | |
2321 | le32_to_cpu(ex->ee_block), | |
2322 | ext4_ext_get_actual_len(ex)); | |
d100eef2 ZL |
2323 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2324 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2325 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2326 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2327 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2328 | ext4_lblk_t next; |
8c55e204 | 2329 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2330 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2331 | |
2332 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2333 | ext_debug("cache gap(after): [%u:%u] %u", |
2334 | le32_to_cpu(ex->ee_block), | |
2335 | ext4_ext_get_actual_len(ex), | |
2336 | block); | |
725d26d3 AK |
2337 | BUG_ON(next == lblock); |
2338 | len = next - lblock; | |
d100eef2 ZL |
2339 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2340 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2341 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2342 | } else { |
a86c6181 AT |
2343 | BUG(); |
2344 | } | |
2345 | ||
bba90743 | 2346 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
2347 | } |
2348 | ||
2349 | /* | |
d0d856e8 RD |
2350 | * ext4_ext_rm_idx: |
2351 | * removes index from the index block. | |
a86c6181 | 2352 | */ |
1d03ec98 | 2353 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2354 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2355 | { |
a86c6181 | 2356 | int err; |
f65e6fba | 2357 | ext4_fsblk_t leaf; |
a86c6181 AT |
2358 | |
2359 | /* free index block */ | |
c36575e6 FL |
2360 | depth--; |
2361 | path = path + depth; | |
bf89d16f | 2362 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2363 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2364 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2365 | return -EIO; | |
2366 | } | |
7e028976 AM |
2367 | err = ext4_ext_get_access(handle, inode, path); |
2368 | if (err) | |
a86c6181 | 2369 | return err; |
0e1147b0 RD |
2370 | |
2371 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2372 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2373 | len *= sizeof(struct ext4_extent_idx); | |
2374 | memmove(path->p_idx, path->p_idx + 1, len); | |
2375 | } | |
2376 | ||
e8546d06 | 2377 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2378 | err = ext4_ext_dirty(handle, inode, path); |
2379 | if (err) | |
a86c6181 | 2380 | return err; |
2ae02107 | 2381 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2382 | trace_ext4_ext_rm_idx(inode, leaf); |
2383 | ||
7dc57615 | 2384 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2385 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2386 | |
2387 | while (--depth >= 0) { | |
2388 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2389 | break; | |
2390 | path--; | |
2391 | err = ext4_ext_get_access(handle, inode, path); | |
2392 | if (err) | |
2393 | break; | |
2394 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2395 | err = ext4_ext_dirty(handle, inode, path); | |
2396 | if (err) | |
2397 | break; | |
2398 | } | |
a86c6181 AT |
2399 | return err; |
2400 | } | |
2401 | ||
2402 | /* | |
ee12b630 MC |
2403 | * ext4_ext_calc_credits_for_single_extent: |
2404 | * This routine returns max. credits that needed to insert an extent | |
2405 | * to the extent tree. | |
2406 | * When pass the actual path, the caller should calculate credits | |
2407 | * under i_data_sem. | |
a86c6181 | 2408 | */ |
525f4ed8 | 2409 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2410 | struct ext4_ext_path *path) |
2411 | { | |
a86c6181 | 2412 | if (path) { |
ee12b630 | 2413 | int depth = ext_depth(inode); |
f3bd1f3f | 2414 | int ret = 0; |
ee12b630 | 2415 | |
a86c6181 | 2416 | /* probably there is space in leaf? */ |
a86c6181 | 2417 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2418 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2419 | |
ee12b630 MC |
2420 | /* |
2421 | * There are some space in the leaf tree, no | |
2422 | * need to account for leaf block credit | |
2423 | * | |
2424 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2425 | * and other metadata blocks still need to be |
ee12b630 MC |
2426 | * accounted. |
2427 | */ | |
525f4ed8 | 2428 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2429 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2430 | return ret; |
ee12b630 MC |
2431 | } |
2432 | } | |
a86c6181 | 2433 | |
525f4ed8 | 2434 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2435 | } |
a86c6181 | 2436 | |
ee12b630 | 2437 | /* |
fffb2739 | 2438 | * How many index/leaf blocks need to change/allocate to add @extents extents? |
ee12b630 | 2439 | * |
fffb2739 JK |
2440 | * If we add a single extent, then in the worse case, each tree level |
2441 | * index/leaf need to be changed in case of the tree split. | |
ee12b630 | 2442 | * |
fffb2739 JK |
2443 | * If more extents are inserted, they could cause the whole tree split more |
2444 | * than once, but this is really rare. | |
ee12b630 | 2445 | */ |
fffb2739 | 2446 | int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
ee12b630 MC |
2447 | { |
2448 | int index; | |
f19d5870 TM |
2449 | int depth; |
2450 | ||
2451 | /* If we are converting the inline data, only one is needed here. */ | |
2452 | if (ext4_has_inline_data(inode)) | |
2453 | return 1; | |
2454 | ||
2455 | depth = ext_depth(inode); | |
a86c6181 | 2456 | |
fffb2739 | 2457 | if (extents <= 1) |
ee12b630 MC |
2458 | index = depth * 2; |
2459 | else | |
2460 | index = depth * 3; | |
a86c6181 | 2461 | |
ee12b630 | 2462 | return index; |
a86c6181 AT |
2463 | } |
2464 | ||
981250ca TT |
2465 | static inline int get_default_free_blocks_flags(struct inode *inode) |
2466 | { | |
2467 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2468 | return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; | |
2469 | else if (ext4_should_journal_data(inode)) | |
2470 | return EXT4_FREE_BLOCKS_FORGET; | |
2471 | return 0; | |
2472 | } | |
2473 | ||
a86c6181 | 2474 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
0aa06000 | 2475 | struct ext4_extent *ex, |
d23142c6 | 2476 | long long *partial_cluster, |
0aa06000 | 2477 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2478 | { |
0aa06000 | 2479 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2480 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2481 | ext4_fsblk_t pblk; |
981250ca | 2482 | int flags = get_default_free_blocks_flags(inode); |
18888cf0 | 2483 | |
0aa06000 TT |
2484 | /* |
2485 | * For bigalloc file systems, we never free a partial cluster | |
2486 | * at the beginning of the extent. Instead, we make a note | |
2487 | * that we tried freeing the cluster, and check to see if we | |
2488 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2489 | * or at the end of the ext4_truncate() operation. | |
2490 | */ | |
2491 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2492 | ||
d8990240 | 2493 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2494 | /* |
2495 | * If we have a partial cluster, and it's different from the | |
2496 | * cluster of the last block, we need to explicitly free the | |
2497 | * partial cluster here. | |
2498 | */ | |
2499 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
d23142c6 LC |
2500 | if ((*partial_cluster > 0) && |
2501 | (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
0aa06000 TT |
2502 | ext4_free_blocks(handle, inode, NULL, |
2503 | EXT4_C2B(sbi, *partial_cluster), | |
2504 | sbi->s_cluster_ratio, flags); | |
2505 | *partial_cluster = 0; | |
2506 | } | |
2507 | ||
a86c6181 AT |
2508 | #ifdef EXTENTS_STATS |
2509 | { | |
2510 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2511 | spin_lock(&sbi->s_ext_stats_lock); |
2512 | sbi->s_ext_blocks += ee_len; | |
2513 | sbi->s_ext_extents++; | |
2514 | if (ee_len < sbi->s_ext_min) | |
2515 | sbi->s_ext_min = ee_len; | |
2516 | if (ee_len > sbi->s_ext_max) | |
2517 | sbi->s_ext_max = ee_len; | |
2518 | if (ext_depth(inode) > sbi->s_depth_max) | |
2519 | sbi->s_depth_max = ext_depth(inode); | |
2520 | spin_unlock(&sbi->s_ext_stats_lock); | |
2521 | } | |
2522 | #endif | |
2523 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2524 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2525 | /* tail removal */ |
725d26d3 | 2526 | ext4_lblk_t num; |
d23142c6 | 2527 | unsigned int unaligned; |
725d26d3 | 2528 | |
a2df2a63 | 2529 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 | 2530 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
d23142c6 LC |
2531 | /* |
2532 | * Usually we want to free partial cluster at the end of the | |
2533 | * extent, except for the situation when the cluster is still | |
2534 | * used by any other extent (partial_cluster is negative). | |
2535 | */ | |
2536 | if (*partial_cluster < 0 && | |
2537 | -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1)) | |
2538 | flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; | |
2539 | ||
2540 | ext_debug("free last %u blocks starting %llu partial %lld\n", | |
2541 | num, pblk, *partial_cluster); | |
0aa06000 TT |
2542 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); |
2543 | /* | |
2544 | * If the block range to be freed didn't start at the | |
2545 | * beginning of a cluster, and we removed the entire | |
d23142c6 LC |
2546 | * extent and the cluster is not used by any other extent, |
2547 | * save the partial cluster here, since we might need to | |
2548 | * delete if we determine that the truncate operation has | |
2549 | * removed all of the blocks in the cluster. | |
2550 | * | |
2551 | * On the other hand, if we did not manage to free the whole | |
2552 | * extent, we have to mark the cluster as used (store negative | |
2553 | * cluster number in partial_cluster). | |
0aa06000 | 2554 | */ |
f5a44db5 | 2555 | unaligned = EXT4_PBLK_COFF(sbi, pblk); |
d23142c6 LC |
2556 | if (unaligned && (ee_len == num) && |
2557 | (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk)))) | |
0aa06000 | 2558 | *partial_cluster = EXT4_B2C(sbi, pblk); |
d23142c6 LC |
2559 | else if (unaligned) |
2560 | *partial_cluster = -((long long)EXT4_B2C(sbi, pblk)); | |
2561 | else if (*partial_cluster > 0) | |
0aa06000 | 2562 | *partial_cluster = 0; |
78fb9cdf LC |
2563 | } else |
2564 | ext4_error(sbi->s_sb, "strange request: removal(2) " | |
2565 | "%u-%u from %u:%u\n", | |
2566 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2567 | return 0; |
2568 | } | |
2569 | ||
d583fb87 AH |
2570 | |
2571 | /* | |
2572 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2573 | * blocks appearing between "start" and "end", and splits the extents | |
2574 | * if "start" and "end" appear in the same extent | |
2575 | * | |
2576 | * @handle: The journal handle | |
2577 | * @inode: The files inode | |
2578 | * @path: The path to the leaf | |
d23142c6 LC |
2579 | * @partial_cluster: The cluster which we'll have to free if all extents |
2580 | * has been released from it. It gets negative in case | |
2581 | * that the cluster is still used. | |
d583fb87 AH |
2582 | * @start: The first block to remove |
2583 | * @end: The last block to remove | |
2584 | */ | |
a86c6181 AT |
2585 | static int |
2586 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
d23142c6 LC |
2587 | struct ext4_ext_path *path, |
2588 | long long *partial_cluster, | |
0aa06000 | 2589 | ext4_lblk_t start, ext4_lblk_t end) |
a86c6181 | 2590 | { |
0aa06000 | 2591 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2592 | int err = 0, correct_index = 0; |
2593 | int depth = ext_depth(inode), credits; | |
2594 | struct ext4_extent_header *eh; | |
750c9c47 | 2595 | ext4_lblk_t a, b; |
725d26d3 AK |
2596 | unsigned num; |
2597 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2598 | unsigned short ex_ee_len; |
556615dc | 2599 | unsigned unwritten = 0; |
a86c6181 | 2600 | struct ext4_extent *ex; |
d23142c6 | 2601 | ext4_fsblk_t pblk; |
a86c6181 | 2602 | |
c29c0ae7 | 2603 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2604 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2605 | if (!path[depth].p_hdr) |
2606 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2607 | eh = path[depth].p_hdr; | |
273df556 FM |
2608 | if (unlikely(path[depth].p_hdr == NULL)) { |
2609 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2610 | return -EIO; | |
2611 | } | |
a86c6181 | 2612 | /* find where to start removing */ |
6ae06ff5 AS |
2613 | ex = path[depth].p_ext; |
2614 | if (!ex) | |
2615 | ex = EXT_LAST_EXTENT(eh); | |
a86c6181 AT |
2616 | |
2617 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2618 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2619 | |
c0634493 EW |
2620 | /* |
2621 | * If we're starting with an extent other than the last one in the | |
2622 | * node, we need to see if it shares a cluster with the extent to | |
2623 | * the right (towards the end of the file). If its leftmost cluster | |
2624 | * is this extent's rightmost cluster and it is not cluster aligned, | |
2625 | * we'll mark it as a partial that is not to be deallocated. | |
2626 | */ | |
2627 | ||
2628 | if (ex != EXT_LAST_EXTENT(eh)) { | |
2629 | ext4_fsblk_t current_pblk, right_pblk; | |
2630 | long long current_cluster, right_cluster; | |
2631 | ||
2632 | current_pblk = ext4_ext_pblock(ex) + ex_ee_len - 1; | |
2633 | current_cluster = (long long)EXT4_B2C(sbi, current_pblk); | |
2634 | right_pblk = ext4_ext_pblock(ex + 1); | |
2635 | right_cluster = (long long)EXT4_B2C(sbi, right_pblk); | |
2636 | if (current_cluster == right_cluster && | |
2637 | EXT4_PBLK_COFF(sbi, right_pblk)) | |
2638 | *partial_cluster = -right_cluster; | |
2639 | } | |
2640 | ||
d8990240 AK |
2641 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2642 | ||
a86c6181 AT |
2643 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2644 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 | 2645 | |
556615dc LC |
2646 | if (ext4_ext_is_unwritten(ex)) |
2647 | unwritten = 1; | |
a41f2071 | 2648 | else |
556615dc | 2649 | unwritten = 0; |
a41f2071 | 2650 | |
553f9008 | 2651 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
556615dc | 2652 | unwritten, ex_ee_len); |
a86c6181 AT |
2653 | path[depth].p_ext = ex; |
2654 | ||
2655 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2656 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2657 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2658 | |
2659 | ext_debug(" border %u:%u\n", a, b); | |
2660 | ||
d583fb87 | 2661 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2662 | if (end < ex_ee_block) { |
d23142c6 LC |
2663 | /* |
2664 | * We're going to skip this extent and move to another, | |
2665 | * so if this extent is not cluster aligned we have | |
2666 | * to mark the current cluster as used to avoid | |
2667 | * accidentally freeing it later on | |
2668 | */ | |
2669 | pblk = ext4_ext_pblock(ex); | |
f5a44db5 | 2670 | if (EXT4_PBLK_COFF(sbi, pblk)) |
d23142c6 LC |
2671 | *partial_cluster = |
2672 | -((long long)EXT4_B2C(sbi, pblk)); | |
d583fb87 AH |
2673 | ex--; |
2674 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2675 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2676 | continue; | |
750c9c47 | 2677 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2678 | EXT4_ERROR_INODE(inode, |
2679 | "can not handle truncate %u:%u " | |
2680 | "on extent %u:%u", | |
2681 | start, end, ex_ee_block, | |
2682 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2683 | err = -EIO; |
2684 | goto out; | |
a86c6181 AT |
2685 | } else if (a != ex_ee_block) { |
2686 | /* remove tail of the extent */ | |
750c9c47 | 2687 | num = a - ex_ee_block; |
a86c6181 AT |
2688 | } else { |
2689 | /* remove whole extent: excellent! */ | |
a86c6181 | 2690 | num = 0; |
a86c6181 | 2691 | } |
34071da7 TT |
2692 | /* |
2693 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2694 | * descriptor) for each block group; assume two block | |
2695 | * groups plus ex_ee_len/blocks_per_block_group for | |
2696 | * the worst case | |
2697 | */ | |
2698 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2699 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2700 | correct_index = 1; | |
2701 | credits += (ext_depth(inode)) + 1; | |
2702 | } | |
5aca07eb | 2703 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2704 | |
487caeef | 2705 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2706 | if (err) |
a86c6181 | 2707 | goto out; |
a86c6181 AT |
2708 | |
2709 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2710 | if (err) | |
2711 | goto out; | |
2712 | ||
0aa06000 TT |
2713 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2714 | a, b); | |
a86c6181 AT |
2715 | if (err) |
2716 | goto out; | |
2717 | ||
750c9c47 | 2718 | if (num == 0) |
d0d856e8 | 2719 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2720 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2721 | |
a86c6181 | 2722 | ex->ee_len = cpu_to_le16(num); |
749269fa | 2723 | /* |
556615dc | 2724 | * Do not mark unwritten if all the blocks in the |
749269fa AA |
2725 | * extent have been removed. |
2726 | */ | |
556615dc LC |
2727 | if (unwritten && num) |
2728 | ext4_ext_mark_unwritten(ex); | |
d583fb87 AH |
2729 | /* |
2730 | * If the extent was completely released, | |
2731 | * we need to remove it from the leaf | |
2732 | */ | |
2733 | if (num == 0) { | |
f17722f9 | 2734 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2735 | /* |
2736 | * For hole punching, we need to scoot all the | |
2737 | * extents up when an extent is removed so that | |
2738 | * we dont have blank extents in the middle | |
2739 | */ | |
2740 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2741 | sizeof(struct ext4_extent)); | |
2742 | ||
2743 | /* Now get rid of the one at the end */ | |
2744 | memset(EXT_LAST_EXTENT(eh), 0, | |
2745 | sizeof(struct ext4_extent)); | |
2746 | } | |
2747 | le16_add_cpu(&eh->eh_entries, -1); | |
d23142c6 | 2748 | } else if (*partial_cluster > 0) |
0aa06000 | 2749 | *partial_cluster = 0; |
d583fb87 | 2750 | |
750c9c47 DM |
2751 | err = ext4_ext_dirty(handle, inode, path + depth); |
2752 | if (err) | |
2753 | goto out; | |
2754 | ||
bf52c6f7 | 2755 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2756 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2757 | ex--; |
2758 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2759 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2760 | } |
2761 | ||
2762 | if (correct_index && eh->eh_entries) | |
2763 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2764 | ||
0aa06000 | 2765 | /* |
ad6599ab EW |
2766 | * If there's a partial cluster and at least one extent remains in |
2767 | * the leaf, free the partial cluster if it isn't shared with the | |
2768 | * current extent. If there's a partial cluster and no extents | |
2769 | * remain in the leaf, it can't be freed here. It can only be | |
2770 | * freed when it's possible to determine if it's not shared with | |
2771 | * any other extent - when the next leaf is processed or when space | |
2772 | * removal is complete. | |
0aa06000 | 2773 | */ |
ad6599ab | 2774 | if (*partial_cluster > 0 && eh->eh_entries && |
0aa06000 TT |
2775 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != |
2776 | *partial_cluster)) { | |
981250ca | 2777 | int flags = get_default_free_blocks_flags(inode); |
0aa06000 TT |
2778 | |
2779 | ext4_free_blocks(handle, inode, NULL, | |
2780 | EXT4_C2B(sbi, *partial_cluster), | |
2781 | sbi->s_cluster_ratio, flags); | |
2782 | *partial_cluster = 0; | |
2783 | } | |
2784 | ||
a86c6181 AT |
2785 | /* if this leaf is free, then we should |
2786 | * remove it from index block above */ | |
2787 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2788 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2789 | |
2790 | out: | |
2791 | return err; | |
2792 | } | |
2793 | ||
2794 | /* | |
d0d856e8 RD |
2795 | * ext4_ext_more_to_rm: |
2796 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2797 | */ |
09b88252 | 2798 | static int |
a86c6181 AT |
2799 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2800 | { | |
2801 | BUG_ON(path->p_idx == NULL); | |
2802 | ||
2803 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2804 | return 0; | |
2805 | ||
2806 | /* | |
d0d856e8 | 2807 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2808 | * so we have to consider current index for truncation |
2809 | */ | |
2810 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2811 | return 0; | |
2812 | return 1; | |
2813 | } | |
2814 | ||
26a4c0c6 TT |
2815 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2816 | ext4_lblk_t end) | |
a86c6181 AT |
2817 | { |
2818 | struct super_block *sb = inode->i_sb; | |
2819 | int depth = ext_depth(inode); | |
968dee77 | 2820 | struct ext4_ext_path *path = NULL; |
d23142c6 | 2821 | long long partial_cluster = 0; |
a86c6181 | 2822 | handle_t *handle; |
6f2080e6 | 2823 | int i = 0, err = 0; |
a86c6181 | 2824 | |
5f95d21f | 2825 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2826 | |
2827 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2828 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2829 | if (IS_ERR(handle)) |
2830 | return PTR_ERR(handle); | |
2831 | ||
0617b83f | 2832 | again: |
61801325 | 2833 | trace_ext4_ext_remove_space(inode, start, end, depth); |
d8990240 | 2834 | |
5f95d21f LC |
2835 | /* |
2836 | * Check if we are removing extents inside the extent tree. If that | |
2837 | * is the case, we are going to punch a hole inside the extent tree | |
2838 | * so we have to check whether we need to split the extent covering | |
2839 | * the last block to remove so we can easily remove the part of it | |
2840 | * in ext4_ext_rm_leaf(). | |
2841 | */ | |
2842 | if (end < EXT_MAX_BLOCKS - 1) { | |
2843 | struct ext4_extent *ex; | |
2844 | ext4_lblk_t ee_block; | |
2845 | ||
2846 | /* find extent for this block */ | |
107a7bd3 | 2847 | path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE); |
5f95d21f LC |
2848 | if (IS_ERR(path)) { |
2849 | ext4_journal_stop(handle); | |
2850 | return PTR_ERR(path); | |
2851 | } | |
2852 | depth = ext_depth(inode); | |
6f2080e6 | 2853 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2854 | ex = path[depth].p_ext; |
968dee77 | 2855 | if (!ex) { |
6f2080e6 DM |
2856 | if (depth) { |
2857 | EXT4_ERROR_INODE(inode, | |
2858 | "path[%d].p_hdr == NULL", | |
2859 | depth); | |
2860 | err = -EIO; | |
2861 | } | |
2862 | goto out; | |
968dee77 | 2863 | } |
5f95d21f LC |
2864 | |
2865 | ee_block = le32_to_cpu(ex->ee_block); | |
2866 | ||
2867 | /* | |
2868 | * See if the last block is inside the extent, if so split | |
2869 | * the extent at 'end' block so we can easily remove the | |
2870 | * tail of the first part of the split extent in | |
2871 | * ext4_ext_rm_leaf(). | |
2872 | */ | |
2873 | if (end >= ee_block && | |
2874 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
5f95d21f LC |
2875 | /* |
2876 | * Split the extent in two so that 'end' is the last | |
27dd4385 LC |
2877 | * block in the first new extent. Also we should not |
2878 | * fail removing space due to ENOSPC so try to use | |
2879 | * reserved block if that happens. | |
5f95d21f | 2880 | */ |
fcf6b1b7 DM |
2881 | err = ext4_force_split_extent_at(handle, inode, path, |
2882 | end + 1, 1); | |
5f95d21f LC |
2883 | if (err < 0) |
2884 | goto out; | |
2885 | } | |
5f95d21f | 2886 | } |
a86c6181 | 2887 | /* |
d0d856e8 RD |
2888 | * We start scanning from right side, freeing all the blocks |
2889 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2890 | */ |
0617b83f | 2891 | depth = ext_depth(inode); |
968dee77 AS |
2892 | if (path) { |
2893 | int k = i = depth; | |
2894 | while (--k > 0) | |
2895 | path[k].p_block = | |
2896 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2897 | } else { | |
2898 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2899 | GFP_NOFS); | |
2900 | if (path == NULL) { | |
2901 | ext4_journal_stop(handle); | |
2902 | return -ENOMEM; | |
2903 | } | |
2904 | path[0].p_depth = depth; | |
2905 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2906 | i = 0; |
5f95d21f | 2907 | |
c349179b | 2908 | if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
968dee77 AS |
2909 | err = -EIO; |
2910 | goto out; | |
2911 | } | |
a86c6181 | 2912 | } |
968dee77 | 2913 | err = 0; |
a86c6181 AT |
2914 | |
2915 | while (i >= 0 && err == 0) { | |
2916 | if (i == depth) { | |
2917 | /* this is leaf block */ | |
d583fb87 | 2918 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2919 | &partial_cluster, start, |
5f95d21f | 2920 | end); |
d0d856e8 | 2921 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2922 | brelse(path[i].p_bh); |
2923 | path[i].p_bh = NULL; | |
2924 | i--; | |
2925 | continue; | |
2926 | } | |
2927 | ||
2928 | /* this is index block */ | |
2929 | if (!path[i].p_hdr) { | |
2930 | ext_debug("initialize header\n"); | |
2931 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2932 | } |
2933 | ||
a86c6181 | 2934 | if (!path[i].p_idx) { |
d0d856e8 | 2935 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2936 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2937 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2938 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2939 | path[i].p_hdr, | |
2940 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2941 | } else { | |
d0d856e8 | 2942 | /* we were already here, see at next index */ |
a86c6181 AT |
2943 | path[i].p_idx--; |
2944 | } | |
2945 | ||
2946 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2947 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2948 | path[i].p_idx); | |
2949 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2950 | struct buffer_head *bh; |
a86c6181 | 2951 | /* go to the next level */ |
2ae02107 | 2952 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2953 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2954 | memset(path + i + 1, 0, sizeof(*path)); |
7d7ea89e | 2955 | bh = read_extent_tree_block(inode, |
107a7bd3 TT |
2956 | ext4_idx_pblock(path[i].p_idx), depth - i - 1, |
2957 | EXT4_EX_NOCACHE); | |
7d7ea89e | 2958 | if (IS_ERR(bh)) { |
a86c6181 | 2959 | /* should we reset i_size? */ |
7d7ea89e | 2960 | err = PTR_ERR(bh); |
a86c6181 AT |
2961 | break; |
2962 | } | |
76828c88 TT |
2963 | /* Yield here to deal with large extent trees. |
2964 | * Should be a no-op if we did IO above. */ | |
2965 | cond_resched(); | |
c29c0ae7 AT |
2966 | if (WARN_ON(i + 1 > depth)) { |
2967 | err = -EIO; | |
c29c0ae7 AT |
2968 | break; |
2969 | } | |
2970 | path[i + 1].p_bh = bh; | |
a86c6181 | 2971 | |
d0d856e8 RD |
2972 | /* save actual number of indexes since this |
2973 | * number is changed at the next iteration */ | |
a86c6181 AT |
2974 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2975 | i++; | |
2976 | } else { | |
d0d856e8 | 2977 | /* we finished processing this index, go up */ |
a86c6181 | 2978 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2979 | /* index is empty, remove it; |
a86c6181 AT |
2980 | * handle must be already prepared by the |
2981 | * truncatei_leaf() */ | |
c36575e6 | 2982 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2983 | } |
d0d856e8 | 2984 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2985 | brelse(path[i].p_bh); |
2986 | path[i].p_bh = NULL; | |
2987 | i--; | |
2988 | ext_debug("return to level %d\n", i); | |
2989 | } | |
2990 | } | |
2991 | ||
61801325 LC |
2992 | trace_ext4_ext_remove_space_done(inode, start, end, depth, |
2993 | partial_cluster, path->p_hdr->eh_entries); | |
d8990240 | 2994 | |
7b415bf6 AK |
2995 | /* If we still have something in the partial cluster and we have removed |
2996 | * even the first extent, then we should free the blocks in the partial | |
2997 | * cluster as well. */ | |
d23142c6 | 2998 | if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) { |
981250ca | 2999 | int flags = get_default_free_blocks_flags(inode); |
7b415bf6 AK |
3000 | |
3001 | ext4_free_blocks(handle, inode, NULL, | |
3002 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
3003 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
3004 | partial_cluster = 0; | |
3005 | } | |
3006 | ||
a86c6181 AT |
3007 | /* TODO: flexible tree reduction should be here */ |
3008 | if (path->p_hdr->eh_entries == 0) { | |
3009 | /* | |
d0d856e8 RD |
3010 | * truncate to zero freed all the tree, |
3011 | * so we need to correct eh_depth | |
a86c6181 AT |
3012 | */ |
3013 | err = ext4_ext_get_access(handle, inode, path); | |
3014 | if (err == 0) { | |
3015 | ext_inode_hdr(inode)->eh_depth = 0; | |
3016 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 3017 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
3018 | err = ext4_ext_dirty(handle, inode, path); |
3019 | } | |
3020 | } | |
3021 | out: | |
a86c6181 AT |
3022 | ext4_ext_drop_refs(path); |
3023 | kfree(path); | |
968dee77 AS |
3024 | if (err == -EAGAIN) { |
3025 | path = NULL; | |
0617b83f | 3026 | goto again; |
968dee77 | 3027 | } |
a86c6181 AT |
3028 | ext4_journal_stop(handle); |
3029 | ||
3030 | return err; | |
3031 | } | |
3032 | ||
3033 | /* | |
3034 | * called at mount time | |
3035 | */ | |
3036 | void ext4_ext_init(struct super_block *sb) | |
3037 | { | |
3038 | /* | |
3039 | * possible initialization would be here | |
3040 | */ | |
3041 | ||
83982b6f | 3042 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 3043 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 3044 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 3045 | #ifdef AGGRESSIVE_TEST |
92b97816 | 3046 | ", aggressive tests" |
a86c6181 AT |
3047 | #endif |
3048 | #ifdef CHECK_BINSEARCH | |
92b97816 | 3049 | ", check binsearch" |
a86c6181 AT |
3050 | #endif |
3051 | #ifdef EXTENTS_STATS | |
92b97816 | 3052 | ", stats" |
a86c6181 | 3053 | #endif |
92b97816 | 3054 | "\n"); |
90576c0b | 3055 | #endif |
a86c6181 AT |
3056 | #ifdef EXTENTS_STATS |
3057 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
3058 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
3059 | EXT4_SB(sb)->s_ext_max = 0; | |
3060 | #endif | |
3061 | } | |
3062 | } | |
3063 | ||
3064 | /* | |
3065 | * called at umount time | |
3066 | */ | |
3067 | void ext4_ext_release(struct super_block *sb) | |
3068 | { | |
83982b6f | 3069 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
3070 | return; |
3071 | ||
3072 | #ifdef EXTENTS_STATS | |
3073 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
3074 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3075 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
3076 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
3077 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
3078 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
3079 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
3080 | } | |
3081 | #endif | |
3082 | } | |
3083 | ||
d7b2a00c ZL |
3084 | static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex) |
3085 | { | |
3086 | ext4_lblk_t ee_block; | |
3087 | ext4_fsblk_t ee_pblock; | |
3088 | unsigned int ee_len; | |
3089 | ||
3090 | ee_block = le32_to_cpu(ex->ee_block); | |
3091 | ee_len = ext4_ext_get_actual_len(ex); | |
3092 | ee_pblock = ext4_ext_pblock(ex); | |
3093 | ||
3094 | if (ee_len == 0) | |
3095 | return 0; | |
3096 | ||
3097 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
3098 | EXTENT_STATUS_WRITTEN); | |
3099 | } | |
3100 | ||
093a088b AK |
3101 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
3102 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
3103 | { | |
2407518d LC |
3104 | ext4_fsblk_t ee_pblock; |
3105 | unsigned int ee_len; | |
b720303d | 3106 | int ret; |
093a088b | 3107 | |
093a088b | 3108 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 3109 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 3110 | |
a107e5a3 | 3111 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
3112 | if (ret > 0) |
3113 | ret = 0; | |
093a088b | 3114 | |
2407518d | 3115 | return ret; |
093a088b AK |
3116 | } |
3117 | ||
47ea3bb5 YY |
3118 | /* |
3119 | * ext4_split_extent_at() splits an extent at given block. | |
3120 | * | |
3121 | * @handle: the journal handle | |
3122 | * @inode: the file inode | |
3123 | * @path: the path to the extent | |
3124 | * @split: the logical block where the extent is splitted. | |
3125 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
556615dc | 3126 | * the states(init or unwritten) of new extents. |
47ea3bb5 YY |
3127 | * @flags: flags used to insert new extent to extent tree. |
3128 | * | |
3129 | * | |
3130 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
3131 | * of which are deterimined by split_flag. | |
3132 | * | |
3133 | * There are two cases: | |
3134 | * a> the extent are splitted into two extent. | |
3135 | * b> split is not needed, and just mark the extent. | |
3136 | * | |
3137 | * return 0 on success. | |
3138 | */ | |
3139 | static int ext4_split_extent_at(handle_t *handle, | |
3140 | struct inode *inode, | |
3141 | struct ext4_ext_path *path, | |
3142 | ext4_lblk_t split, | |
3143 | int split_flag, | |
3144 | int flags) | |
3145 | { | |
3146 | ext4_fsblk_t newblock; | |
3147 | ext4_lblk_t ee_block; | |
adb23551 | 3148 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
3149 | struct ext4_extent *ex2 = NULL; |
3150 | unsigned int ee_len, depth; | |
3151 | int err = 0; | |
3152 | ||
dee1f973 DM |
3153 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
3154 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
3155 | ||
47ea3bb5 YY |
3156 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
3157 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3158 | ||
3159 | ext4_ext_show_leaf(inode, path); | |
3160 | ||
3161 | depth = ext_depth(inode); | |
3162 | ex = path[depth].p_ext; | |
3163 | ee_block = le32_to_cpu(ex->ee_block); | |
3164 | ee_len = ext4_ext_get_actual_len(ex); | |
3165 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3166 | ||
3167 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
556615dc | 3168 | BUG_ON(!ext4_ext_is_unwritten(ex) && |
357b66fd | 3169 | split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc LC |
3170 | EXT4_EXT_MARK_UNWRIT1 | |
3171 | EXT4_EXT_MARK_UNWRIT2)); | |
47ea3bb5 YY |
3172 | |
3173 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3174 | if (err) | |
3175 | goto out; | |
3176 | ||
3177 | if (split == ee_block) { | |
3178 | /* | |
3179 | * case b: block @split is the block that the extent begins with | |
3180 | * then we just change the state of the extent, and splitting | |
3181 | * is not needed. | |
3182 | */ | |
556615dc LC |
3183 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3184 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3185 | else |
3186 | ext4_ext_mark_initialized(ex); | |
3187 | ||
3188 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3189 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3190 | |
ecb94f5f | 3191 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3192 | goto out; |
3193 | } | |
3194 | ||
3195 | /* case a */ | |
3196 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3197 | ex->ee_len = cpu_to_le16(split - ee_block); | |
556615dc LC |
3198 | if (split_flag & EXT4_EXT_MARK_UNWRIT1) |
3199 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3200 | |
3201 | /* | |
3202 | * path may lead to new leaf, not to original leaf any more | |
3203 | * after ext4_ext_insert_extent() returns, | |
3204 | */ | |
3205 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3206 | if (err) | |
3207 | goto fix_extent_len; | |
3208 | ||
3209 | ex2 = &newex; | |
3210 | ex2->ee_block = cpu_to_le32(split); | |
3211 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3212 | ext4_ext_store_pblock(ex2, newblock); | |
556615dc LC |
3213 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3214 | ext4_ext_mark_unwritten(ex2); | |
47ea3bb5 YY |
3215 | |
3216 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
3217 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 | 3218 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3219 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3220 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3221 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3222 | zero_ex.ee_len = cpu_to_le16( |
3223 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3224 | ext4_ext_store_pblock(&zero_ex, |
3225 | ext4_ext_pblock(ex2)); | |
3226 | } else { | |
dee1f973 | 3227 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3228 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3229 | zero_ex.ee_len = cpu_to_le16( |
3230 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3231 | ext4_ext_store_pblock(&zero_ex, |
3232 | ext4_ext_pblock(ex)); | |
3233 | } | |
3234 | } else { | |
dee1f973 | 3235 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3236 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3237 | zero_ex.ee_len = cpu_to_le16( |
3238 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3239 | ext4_ext_store_pblock(&zero_ex, |
3240 | ext4_ext_pblock(&orig_ex)); | |
3241 | } | |
dee1f973 | 3242 | |
47ea3bb5 YY |
3243 | if (err) |
3244 | goto fix_extent_len; | |
3245 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3246 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3247 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3248 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3249 | if (err) |
3250 | goto fix_extent_len; | |
3251 | ||
3252 | /* update extent status tree */ | |
d7b2a00c | 3253 | err = ext4_zeroout_es(inode, &zero_ex); |
adb23551 | 3254 | |
47ea3bb5 YY |
3255 | goto out; |
3256 | } else if (err) | |
3257 | goto fix_extent_len; | |
3258 | ||
3259 | out: | |
3260 | ext4_ext_show_leaf(inode, path); | |
3261 | return err; | |
3262 | ||
3263 | fix_extent_len: | |
3264 | ex->ee_len = orig_ex.ee_len; | |
29faed16 | 3265 | ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3266 | return err; |
3267 | } | |
3268 | ||
3269 | /* | |
3270 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3271 | * by @map as split_flags indicates | |
3272 | * | |
70261f56 | 3273 | * It may result in splitting the extent into multiple extents (up to three) |
47ea3bb5 YY |
3274 | * There are three possibilities: |
3275 | * a> There is no split required | |
3276 | * b> Splits in two extents: Split is happening at either end of the extent | |
3277 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3278 | * | |
3279 | */ | |
3280 | static int ext4_split_extent(handle_t *handle, | |
3281 | struct inode *inode, | |
3282 | struct ext4_ext_path *path, | |
3283 | struct ext4_map_blocks *map, | |
3284 | int split_flag, | |
3285 | int flags) | |
3286 | { | |
3287 | ext4_lblk_t ee_block; | |
3288 | struct ext4_extent *ex; | |
3289 | unsigned int ee_len, depth; | |
3290 | int err = 0; | |
556615dc | 3291 | int unwritten; |
47ea3bb5 | 3292 | int split_flag1, flags1; |
3a225670 | 3293 | int allocated = map->m_len; |
47ea3bb5 YY |
3294 | |
3295 | depth = ext_depth(inode); | |
3296 | ex = path[depth].p_ext; | |
3297 | ee_block = le32_to_cpu(ex->ee_block); | |
3298 | ee_len = ext4_ext_get_actual_len(ex); | |
556615dc | 3299 | unwritten = ext4_ext_is_unwritten(ex); |
47ea3bb5 YY |
3300 | |
3301 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3302 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 | 3303 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
556615dc LC |
3304 | if (unwritten) |
3305 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1 | | |
3306 | EXT4_EXT_MARK_UNWRIT2; | |
dee1f973 DM |
3307 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3308 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3309 | err = ext4_split_extent_at(handle, inode, path, |
3310 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3311 | if (err) |
3312 | goto out; | |
3a225670 ZL |
3313 | } else { |
3314 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3315 | } |
357b66fd DM |
3316 | /* |
3317 | * Update path is required because previous ext4_split_extent_at() may | |
3318 | * result in split of original leaf or extent zeroout. | |
3319 | */ | |
47ea3bb5 | 3320 | ext4_ext_drop_refs(path); |
705912ca TT |
3321 | path = ext4_ext_find_extent(inode, map->m_lblk, &path, |
3322 | EXT4_EX_NOFREE_ON_ERR); | |
47ea3bb5 YY |
3323 | if (IS_ERR(path)) |
3324 | return PTR_ERR(path); | |
357b66fd DM |
3325 | depth = ext_depth(inode); |
3326 | ex = path[depth].p_ext; | |
a18ed359 DM |
3327 | if (!ex) { |
3328 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3329 | (unsigned long) map->m_lblk); | |
3330 | return -EIO; | |
3331 | } | |
556615dc | 3332 | unwritten = ext4_ext_is_unwritten(ex); |
357b66fd | 3333 | split_flag1 = 0; |
47ea3bb5 YY |
3334 | |
3335 | if (map->m_lblk >= ee_block) { | |
357b66fd | 3336 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
556615dc LC |
3337 | if (unwritten) { |
3338 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1; | |
357b66fd | 3339 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc | 3340 | EXT4_EXT_MARK_UNWRIT2); |
357b66fd | 3341 | } |
47ea3bb5 YY |
3342 | err = ext4_split_extent_at(handle, inode, path, |
3343 | map->m_lblk, split_flag1, flags); | |
3344 | if (err) | |
3345 | goto out; | |
3346 | } | |
3347 | ||
3348 | ext4_ext_show_leaf(inode, path); | |
3349 | out: | |
3a225670 | 3350 | return err ? err : allocated; |
47ea3bb5 YY |
3351 | } |
3352 | ||
56055d3a | 3353 | /* |
e35fd660 | 3354 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
556615dc | 3355 | * to an unwritten extent. It may result in splitting the unwritten |
25985edc | 3356 | * extent into multiple extents (up to three - one initialized and two |
556615dc | 3357 | * unwritten). |
56055d3a AA |
3358 | * There are three possibilities: |
3359 | * a> There is no split required: Entire extent should be initialized | |
3360 | * b> Splits in two extents: Write is happening at either end of the extent | |
3361 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3362 | * |
3363 | * Pre-conditions: | |
556615dc | 3364 | * - The extent pointed to by 'path' is unwritten. |
6f91bc5f EG |
3365 | * - The extent pointed to by 'path' contains a superset |
3366 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3367 | * | |
3368 | * Post-conditions on success: | |
3369 | * - the returned value is the number of blocks beyond map->l_lblk | |
3370 | * that are allocated and initialized. | |
3371 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3372 | */ |
725d26d3 | 3373 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3374 | struct inode *inode, |
3375 | struct ext4_map_blocks *map, | |
27dd4385 LC |
3376 | struct ext4_ext_path *path, |
3377 | int flags) | |
56055d3a | 3378 | { |
67a5da56 | 3379 | struct ext4_sb_info *sbi; |
6f91bc5f | 3380 | struct ext4_extent_header *eh; |
667eff35 YY |
3381 | struct ext4_map_blocks split_map; |
3382 | struct ext4_extent zero_ex; | |
bc2d9db4 | 3383 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3384 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3385 | unsigned int ee_len, depth, map_len = map->m_len; |
3386 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3387 | int err = 0; |
667eff35 | 3388 | int split_flag = 0; |
21ca087a DM |
3389 | |
3390 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3391 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3392 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3393 | |
67a5da56 | 3394 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3395 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3396 | inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3397 | if (eof_block < map->m_lblk + map_len) |
3398 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3399 | |
3400 | depth = ext_depth(inode); | |
6f91bc5f | 3401 | eh = path[depth].p_hdr; |
56055d3a AA |
3402 | ex = path[depth].p_ext; |
3403 | ee_block = le32_to_cpu(ex->ee_block); | |
3404 | ee_len = ext4_ext_get_actual_len(ex); | |
adb23551 | 3405 | zero_ex.ee_len = 0; |
56055d3a | 3406 | |
6f91bc5f EG |
3407 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3408 | ||
3409 | /* Pre-conditions */ | |
556615dc | 3410 | BUG_ON(!ext4_ext_is_unwritten(ex)); |
6f91bc5f | 3411 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); |
6f91bc5f EG |
3412 | |
3413 | /* | |
3414 | * Attempt to transfer newly initialized blocks from the currently | |
556615dc | 3415 | * unwritten extent to its neighbor. This is much cheaper |
6f91bc5f | 3416 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3417 | * memmove() calls. Transferring to the left is the common case in |
3418 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3419 | * followed by append writes. | |
6f91bc5f EG |
3420 | * |
3421 | * Limitations of the current logic: | |
bc2d9db4 | 3422 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3423 | * This would require removing the extent if the transfer |
3424 | * is possible. | |
bc2d9db4 | 3425 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3426 | * same extent tree node. |
3427 | */ | |
bc2d9db4 LC |
3428 | if ((map->m_lblk == ee_block) && |
3429 | /* See if we can merge left */ | |
3430 | (map_len < ee_len) && /*L1*/ | |
3431 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3432 | ext4_lblk_t prev_lblk; |
3433 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3434 | unsigned int prev_len; |
6f91bc5f | 3435 | |
bc2d9db4 LC |
3436 | abut_ex = ex - 1; |
3437 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3438 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3439 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3440 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3441 | |
3442 | /* | |
bc2d9db4 | 3443 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3444 | * upon those conditions: |
bc2d9db4 LC |
3445 | * - C1: abut_ex is initialized, |
3446 | * - C2: abut_ex is logically abutting ex, | |
3447 | * - C3: abut_ex is physically abutting ex, | |
3448 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3449 | * overflowing the (initialized) length limit. |
3450 | */ | |
556615dc | 3451 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3452 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3453 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3454 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3455 | err = ext4_ext_get_access(handle, inode, path + depth); |
3456 | if (err) | |
3457 | goto out; | |
3458 | ||
3459 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3460 | map, ex, abut_ex); |
6f91bc5f | 3461 | |
bc2d9db4 LC |
3462 | /* Shift the start of ex by 'map_len' blocks */ |
3463 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3464 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3465 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3466 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
6f91bc5f | 3467 | |
bc2d9db4 LC |
3468 | /* Extend abut_ex by 'map_len' blocks */ |
3469 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3470 | |
bc2d9db4 LC |
3471 | /* Result: number of initialized blocks past m_lblk */ |
3472 | allocated = map_len; | |
3473 | } | |
3474 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3475 | (map_len < ee_len) && /*L1*/ | |
3476 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3477 | /* See if we can merge right */ | |
3478 | ext4_lblk_t next_lblk; | |
3479 | ext4_fsblk_t next_pblk, ee_pblk; | |
3480 | unsigned int next_len; | |
3481 | ||
3482 | abut_ex = ex + 1; | |
3483 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3484 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3485 | next_pblk = ext4_ext_pblock(abut_ex); | |
3486 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3487 | |
bc2d9db4 LC |
3488 | /* |
3489 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3490 | * upon those conditions: | |
3491 | * - C1: abut_ex is initialized, | |
3492 | * - C2: abut_ex is logically abutting ex, | |
3493 | * - C3: abut_ex is physically abutting ex, | |
3494 | * - C4: abut_ex can receive the additional blocks without | |
3495 | * overflowing the (initialized) length limit. | |
3496 | */ | |
556615dc | 3497 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
bc2d9db4 LC |
3498 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ |
3499 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3500 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3501 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3502 | if (err) | |
3503 | goto out; | |
3504 | ||
3505 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3506 | map, ex, abut_ex); | |
3507 | ||
3508 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3509 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3510 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3511 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3512 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
bc2d9db4 LC |
3513 | |
3514 | /* Extend abut_ex by 'map_len' blocks */ | |
3515 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3516 | |
3517 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3518 | allocated = map_len; |
6f91bc5f EG |
3519 | } |
3520 | } | |
bc2d9db4 LC |
3521 | if (allocated) { |
3522 | /* Mark the block containing both extents as dirty */ | |
3523 | ext4_ext_dirty(handle, inode, path + depth); | |
3524 | ||
3525 | /* Update path to point to the right extent */ | |
3526 | path[depth].p_ext = abut_ex; | |
3527 | goto out; | |
3528 | } else | |
3529 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3530 | |
667eff35 | 3531 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3532 | /* |
3533 | * It is safe to convert extent to initialized via explicit | |
9e740568 | 3534 | * zeroout only if extent is fully inside i_size or new_size. |
21ca087a | 3535 | */ |
667eff35 | 3536 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3537 | |
67a5da56 ZL |
3538 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3539 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3540 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 ZL |
3541 | |
3542 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3543 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3544 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3545 | if (err) |
d03856bd | 3546 | goto out; |
adb23551 | 3547 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 | 3548 | zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)); |
adb23551 | 3549 | ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex)); |
d03856bd AK |
3550 | |
3551 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3552 | if (err) | |
3553 | goto out; | |
667eff35 | 3554 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3555 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3556 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3557 | goto out; |
56055d3a | 3558 | } |
667eff35 | 3559 | |
56055d3a | 3560 | /* |
667eff35 YY |
3561 | * four cases: |
3562 | * 1. split the extent into three extents. | |
3563 | * 2. split the extent into two extents, zeroout the first half. | |
3564 | * 3. split the extent into two extents, zeroout the second half. | |
3565 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3566 | */ |
667eff35 YY |
3567 | split_map.m_lblk = map->m_lblk; |
3568 | split_map.m_len = map->m_len; | |
3569 | ||
67a5da56 ZL |
3570 | if (max_zeroout && (allocated > map->m_len)) { |
3571 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3572 | /* case 3 */ |
3573 | zero_ex.ee_block = | |
9b940f8e AH |
3574 | cpu_to_le32(map->m_lblk); |
3575 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3576 | ext4_ext_store_pblock(&zero_ex, |
3577 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3578 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3579 | if (err) |
3580 | goto out; | |
667eff35 YY |
3581 | split_map.m_lblk = map->m_lblk; |
3582 | split_map.m_len = allocated; | |
67a5da56 | 3583 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3584 | /* case 2 */ |
3585 | if (map->m_lblk != ee_block) { | |
3586 | zero_ex.ee_block = ex->ee_block; | |
3587 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3588 | ee_block); | |
3589 | ext4_ext_store_pblock(&zero_ex, | |
3590 | ext4_ext_pblock(ex)); | |
3591 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3592 | if (err) | |
3593 | goto out; | |
3594 | } | |
3595 | ||
667eff35 | 3596 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3597 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3598 | allocated = map->m_len; | |
56055d3a AA |
3599 | } |
3600 | } | |
667eff35 YY |
3601 | |
3602 | allocated = ext4_split_extent(handle, inode, path, | |
27dd4385 | 3603 | &split_map, split_flag, flags); |
667eff35 YY |
3604 | if (allocated < 0) |
3605 | err = allocated; | |
3606 | ||
56055d3a | 3607 | out: |
adb23551 ZL |
3608 | /* If we have gotten a failure, don't zero out status tree */ |
3609 | if (!err) | |
d7b2a00c | 3610 | err = ext4_zeroout_es(inode, &zero_ex); |
56055d3a AA |
3611 | return err ? err : allocated; |
3612 | } | |
3613 | ||
0031462b | 3614 | /* |
e35fd660 | 3615 | * This function is called by ext4_ext_map_blocks() from |
0031462b | 3616 | * ext4_get_blocks_dio_write() when DIO to write |
556615dc | 3617 | * to an unwritten extent. |
0031462b | 3618 | * |
556615dc LC |
3619 | * Writing to an unwritten extent may result in splitting the unwritten |
3620 | * extent into multiple initialized/unwritten extents (up to three) | |
0031462b | 3621 | * There are three possibilities: |
556615dc | 3622 | * a> There is no split required: Entire extent should be unwritten |
0031462b MC |
3623 | * b> Splits in two extents: Write is happening at either end of the extent |
3624 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3625 | * | |
b8a86845 LC |
3626 | * This works the same way in the case of initialized -> unwritten conversion. |
3627 | * | |
0031462b | 3628 | * One of more index blocks maybe needed if the extent tree grow after |
556615dc LC |
3629 | * the unwritten extent split. To prevent ENOSPC occur at the IO |
3630 | * complete, we need to split the unwritten extent before DIO submit | |
3631 | * the IO. The unwritten extent called at this time will be split | |
3632 | * into three unwritten extent(at most). After IO complete, the part | |
0031462b MC |
3633 | * being filled will be convert to initialized by the end_io callback function |
3634 | * via ext4_convert_unwritten_extents(). | |
ba230c3f | 3635 | * |
556615dc | 3636 | * Returns the size of unwritten extent to be written on success. |
0031462b | 3637 | */ |
b8a86845 | 3638 | static int ext4_split_convert_extents(handle_t *handle, |
0031462b | 3639 | struct inode *inode, |
e35fd660 | 3640 | struct ext4_map_blocks *map, |
0031462b | 3641 | struct ext4_ext_path *path, |
0031462b MC |
3642 | int flags) |
3643 | { | |
667eff35 YY |
3644 | ext4_lblk_t eof_block; |
3645 | ext4_lblk_t ee_block; | |
3646 | struct ext4_extent *ex; | |
3647 | unsigned int ee_len; | |
3648 | int split_flag = 0, depth; | |
21ca087a | 3649 | |
b8a86845 LC |
3650 | ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n", |
3651 | __func__, inode->i_ino, | |
3652 | (unsigned long long)map->m_lblk, map->m_len); | |
21ca087a DM |
3653 | |
3654 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3655 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3656 | if (eof_block < map->m_lblk + map->m_len) |
3657 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3658 | /* |
3659 | * It is safe to convert extent to initialized via explicit | |
3660 | * zeroout only if extent is fully insde i_size or new_size. | |
3661 | */ | |
667eff35 YY |
3662 | depth = ext_depth(inode); |
3663 | ex = path[depth].p_ext; | |
3664 | ee_block = le32_to_cpu(ex->ee_block); | |
3665 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3666 | |
b8a86845 LC |
3667 | /* Convert to unwritten */ |
3668 | if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) { | |
3669 | split_flag |= EXT4_EXT_DATA_VALID1; | |
3670 | /* Convert to initialized */ | |
3671 | } else if (flags & EXT4_GET_BLOCKS_CONVERT) { | |
3672 | split_flag |= ee_block + ee_len <= eof_block ? | |
3673 | EXT4_EXT_MAY_ZEROOUT : 0; | |
556615dc | 3674 | split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2); |
b8a86845 | 3675 | } |
667eff35 YY |
3676 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3677 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3678 | } |
197217a5 | 3679 | |
b8a86845 LC |
3680 | static int ext4_convert_initialized_extents(handle_t *handle, |
3681 | struct inode *inode, | |
3682 | struct ext4_map_blocks *map, | |
3683 | struct ext4_ext_path *path) | |
3684 | { | |
3685 | struct ext4_extent *ex; | |
3686 | ext4_lblk_t ee_block; | |
3687 | unsigned int ee_len; | |
3688 | int depth; | |
3689 | int err = 0; | |
3690 | ||
3691 | depth = ext_depth(inode); | |
3692 | ex = path[depth].p_ext; | |
3693 | ee_block = le32_to_cpu(ex->ee_block); | |
3694 | ee_len = ext4_ext_get_actual_len(ex); | |
3695 | ||
3696 | ext_debug("%s: inode %lu, logical" | |
3697 | "block %llu, max_blocks %u\n", __func__, inode->i_ino, | |
3698 | (unsigned long long)ee_block, ee_len); | |
3699 | ||
3700 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
3701 | err = ext4_split_convert_extents(handle, inode, map, path, | |
3702 | EXT4_GET_BLOCKS_CONVERT_UNWRITTEN); | |
3703 | if (err < 0) | |
3704 | goto out; | |
3705 | ext4_ext_drop_refs(path); | |
705912ca TT |
3706 | path = ext4_ext_find_extent(inode, map->m_lblk, &path, |
3707 | EXT4_EX_NOFREE_ON_ERR); | |
b8a86845 LC |
3708 | if (IS_ERR(path)) { |
3709 | err = PTR_ERR(path); | |
3710 | goto out; | |
3711 | } | |
3712 | depth = ext_depth(inode); | |
3713 | ex = path[depth].p_ext; | |
a18ed359 DM |
3714 | if (!ex) { |
3715 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3716 | (unsigned long) map->m_lblk); | |
3717 | err = -EIO; | |
3718 | goto out; | |
3719 | } | |
b8a86845 LC |
3720 | } |
3721 | ||
3722 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3723 | if (err) | |
3724 | goto out; | |
556615dc LC |
3725 | /* first mark the extent as unwritten */ |
3726 | ext4_ext_mark_unwritten(ex); | |
b8a86845 LC |
3727 | |
3728 | /* note: ext4_ext_correct_indexes() isn't needed here because | |
3729 | * borders are not changed | |
3730 | */ | |
3731 | ext4_ext_try_to_merge(handle, inode, path, ex); | |
3732 | ||
3733 | /* Mark modified extent as dirty */ | |
3734 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
3735 | out: | |
3736 | ext4_ext_show_leaf(inode, path); | |
3737 | return err; | |
3738 | } | |
3739 | ||
3740 | ||
c7064ef1 | 3741 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3742 | struct inode *inode, |
3743 | struct ext4_map_blocks *map, | |
3744 | struct ext4_ext_path *path) | |
0031462b MC |
3745 | { |
3746 | struct ext4_extent *ex; | |
dee1f973 DM |
3747 | ext4_lblk_t ee_block; |
3748 | unsigned int ee_len; | |
0031462b MC |
3749 | int depth; |
3750 | int err = 0; | |
0031462b MC |
3751 | |
3752 | depth = ext_depth(inode); | |
0031462b | 3753 | ex = path[depth].p_ext; |
dee1f973 DM |
3754 | ee_block = le32_to_cpu(ex->ee_block); |
3755 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3756 | |
197217a5 YY |
3757 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3758 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3759 | (unsigned long long)ee_block, ee_len); |
3760 | ||
ff95ec22 DM |
3761 | /* If extent is larger than requested it is a clear sign that we still |
3762 | * have some extent state machine issues left. So extent_split is still | |
3763 | * required. | |
3764 | * TODO: Once all related issues will be fixed this situation should be | |
3765 | * illegal. | |
3766 | */ | |
dee1f973 | 3767 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ff95ec22 DM |
3768 | #ifdef EXT4_DEBUG |
3769 | ext4_warning("Inode (%ld) finished: extent logical block %llu," | |
3770 | " len %u; IO logical block %llu, len %u\n", | |
3771 | inode->i_ino, (unsigned long long)ee_block, ee_len, | |
3772 | (unsigned long long)map->m_lblk, map->m_len); | |
3773 | #endif | |
b8a86845 LC |
3774 | err = ext4_split_convert_extents(handle, inode, map, path, |
3775 | EXT4_GET_BLOCKS_CONVERT); | |
dee1f973 DM |
3776 | if (err < 0) |
3777 | goto out; | |
3778 | ext4_ext_drop_refs(path); | |
705912ca TT |
3779 | path = ext4_ext_find_extent(inode, map->m_lblk, &path, |
3780 | EXT4_EX_NOFREE_ON_ERR); | |
dee1f973 DM |
3781 | if (IS_ERR(path)) { |
3782 | err = PTR_ERR(path); | |
3783 | goto out; | |
3784 | } | |
3785 | depth = ext_depth(inode); | |
3786 | ex = path[depth].p_ext; | |
3787 | } | |
197217a5 | 3788 | |
0031462b MC |
3789 | err = ext4_ext_get_access(handle, inode, path + depth); |
3790 | if (err) | |
3791 | goto out; | |
3792 | /* first mark the extent as initialized */ | |
3793 | ext4_ext_mark_initialized(ex); | |
3794 | ||
197217a5 YY |
3795 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3796 | * borders are not changed | |
0031462b | 3797 | */ |
ecb94f5f | 3798 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3799 | |
0031462b | 3800 | /* Mark modified extent as dirty */ |
ecb94f5f | 3801 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3802 | out: |
3803 | ext4_ext_show_leaf(inode, path); | |
3804 | return err; | |
3805 | } | |
3806 | ||
515f41c3 AK |
3807 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3808 | sector_t block, int count) | |
3809 | { | |
3810 | int i; | |
3811 | for (i = 0; i < count; i++) | |
3812 | unmap_underlying_metadata(bdev, block + i); | |
3813 | } | |
3814 | ||
58590b06 TT |
3815 | /* |
3816 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3817 | */ | |
3818 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3819 | ext4_lblk_t lblk, |
58590b06 TT |
3820 | struct ext4_ext_path *path, |
3821 | unsigned int len) | |
3822 | { | |
3823 | int i, depth; | |
3824 | struct ext4_extent_header *eh; | |
65922cb5 | 3825 | struct ext4_extent *last_ex; |
58590b06 TT |
3826 | |
3827 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3828 | return 0; | |
3829 | ||
3830 | depth = ext_depth(inode); | |
3831 | eh = path[depth].p_hdr; | |
58590b06 | 3832 | |
afcff5d8 LC |
3833 | /* |
3834 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3835 | * do not care for this case anymore. Simply remove the flag | |
3836 | * if there are no extents. | |
3837 | */ | |
3838 | if (unlikely(!eh->eh_entries)) | |
3839 | goto out; | |
58590b06 TT |
3840 | last_ex = EXT_LAST_EXTENT(eh); |
3841 | /* | |
3842 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3843 | * last block in the last extent in the file. We test this by | |
3844 | * first checking to see if the caller to | |
3845 | * ext4_ext_get_blocks() was interested in the last block (or | |
3846 | * a block beyond the last block) in the current extent. If | |
3847 | * this turns out to be false, we can bail out from this | |
3848 | * function immediately. | |
3849 | */ | |
d002ebf1 | 3850 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3851 | ext4_ext_get_actual_len(last_ex)) |
3852 | return 0; | |
3853 | /* | |
3854 | * If the caller does appear to be planning to write at or | |
3855 | * beyond the end of the current extent, we then test to see | |
3856 | * if the current extent is the last extent in the file, by | |
3857 | * checking to make sure it was reached via the rightmost node | |
3858 | * at each level of the tree. | |
3859 | */ | |
3860 | for (i = depth-1; i >= 0; i--) | |
3861 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3862 | return 0; | |
afcff5d8 | 3863 | out: |
58590b06 TT |
3864 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3865 | return ext4_mark_inode_dirty(handle, inode); | |
3866 | } | |
3867 | ||
7b415bf6 AK |
3868 | /** |
3869 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3870 | * | |
7d1b1fbc | 3871 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3872 | */ |
f7fec032 ZL |
3873 | int ext4_find_delalloc_range(struct inode *inode, |
3874 | ext4_lblk_t lblk_start, | |
3875 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3876 | { |
7d1b1fbc | 3877 | struct extent_status es; |
7b415bf6 | 3878 | |
e30b5dca | 3879 | ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es); |
06b0c886 | 3880 | if (es.es_len == 0) |
7d1b1fbc | 3881 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3882 | else if (es.es_lblk <= lblk_start && |
3883 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3884 | return 1; |
06b0c886 | 3885 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3886 | return 1; |
7b415bf6 | 3887 | else |
7d1b1fbc | 3888 | return 0; |
7b415bf6 AK |
3889 | } |
3890 | ||
7d1b1fbc | 3891 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3892 | { |
3893 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3894 | ext4_lblk_t lblk_start, lblk_end; | |
f5a44db5 | 3895 | lblk_start = EXT4_LBLK_CMASK(sbi, lblk); |
7b415bf6 AK |
3896 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; |
3897 | ||
7d1b1fbc | 3898 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3899 | } |
3900 | ||
3901 | /** | |
3902 | * Determines how many complete clusters (out of those specified by the 'map') | |
3903 | * are under delalloc and were reserved quota for. | |
3904 | * This function is called when we are writing out the blocks that were | |
3905 | * originally written with their allocation delayed, but then the space was | |
3906 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3907 | * The cases to look for are: | |
3908 | * ('=' indicated delayed allocated blocks | |
3909 | * '-' indicates non-delayed allocated blocks) | |
3910 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3911 | * are not delalloc'ed. | |
3912 | * Ex: | |
3913 | * |----c---=|====c====|====c====|===-c----| | |
3914 | * |++++++ allocated ++++++| | |
3915 | * ==> 4 complete clusters in above example | |
3916 | * | |
3917 | * (b) partial cluster (outside of allocated range) towards either end is | |
3918 | * marked for delayed allocation. In this case, we will exclude that | |
3919 | * cluster. | |
3920 | * Ex: | |
3921 | * |----====c========|========c========| | |
3922 | * |++++++ allocated ++++++| | |
3923 | * ==> 1 complete clusters in above example | |
3924 | * | |
3925 | * Ex: | |
3926 | * |================c================| | |
3927 | * |++++++ allocated ++++++| | |
3928 | * ==> 0 complete clusters in above example | |
3929 | * | |
3930 | * The ext4_da_update_reserve_space will be called only if we | |
3931 | * determine here that there were some "entire" clusters that span | |
3932 | * this 'allocated' range. | |
3933 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3934 | * without ever calling ext4_find_delalloc_range. | |
3935 | */ | |
3936 | static unsigned int | |
3937 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3938 | unsigned int num_blks) | |
3939 | { | |
3940 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3941 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3942 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3943 | unsigned int allocated_clusters = 0; | |
3944 | ||
3945 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3946 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3947 | ||
3948 | /* max possible clusters for this allocation */ | |
3949 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3950 | ||
d8990240 AK |
3951 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3952 | ||
7b415bf6 | 3953 | /* Check towards left side */ |
f5a44db5 | 3954 | c_offset = EXT4_LBLK_COFF(sbi, lblk_start); |
7b415bf6 | 3955 | if (c_offset) { |
f5a44db5 | 3956 | lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start); |
7b415bf6 AK |
3957 | lblk_to = lblk_from + c_offset - 1; |
3958 | ||
7d1b1fbc | 3959 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3960 | allocated_clusters--; |
3961 | } | |
3962 | ||
3963 | /* Now check towards right. */ | |
f5a44db5 | 3964 | c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks); |
7b415bf6 AK |
3965 | if (allocated_clusters && c_offset) { |
3966 | lblk_from = lblk_start + num_blks; | |
3967 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3968 | ||
7d1b1fbc | 3969 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3970 | allocated_clusters--; |
3971 | } | |
3972 | ||
3973 | return allocated_clusters; | |
3974 | } | |
3975 | ||
b8a86845 LC |
3976 | static int |
3977 | ext4_ext_convert_initialized_extent(handle_t *handle, struct inode *inode, | |
3978 | struct ext4_map_blocks *map, | |
3979 | struct ext4_ext_path *path, int flags, | |
3980 | unsigned int allocated, ext4_fsblk_t newblock) | |
3981 | { | |
3982 | int ret = 0; | |
3983 | int err = 0; | |
3984 | ||
3985 | /* | |
3986 | * Make sure that the extent is no bigger than we support with | |
556615dc | 3987 | * unwritten extent |
b8a86845 | 3988 | */ |
556615dc LC |
3989 | if (map->m_len > EXT_UNWRITTEN_MAX_LEN) |
3990 | map->m_len = EXT_UNWRITTEN_MAX_LEN / 2; | |
b8a86845 LC |
3991 | |
3992 | ret = ext4_convert_initialized_extents(handle, inode, map, | |
3993 | path); | |
3994 | if (ret >= 0) { | |
3995 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
3996 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
3997 | path, map->m_len); | |
3998 | } else | |
3999 | err = ret; | |
4000 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
4001 | if (allocated > map->m_len) | |
4002 | allocated = map->m_len; | |
4003 | map->m_len = allocated; | |
4004 | ||
4005 | return err ? err : allocated; | |
4006 | } | |
4007 | ||
0031462b | 4008 | static int |
556615dc | 4009 | ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode, |
e35fd660 | 4010 | struct ext4_map_blocks *map, |
0031462b | 4011 | struct ext4_ext_path *path, int flags, |
e35fd660 | 4012 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
4013 | { |
4014 | int ret = 0; | |
4015 | int err = 0; | |
f45ee3a1 | 4016 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 4017 | |
556615dc | 4018 | ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical " |
88635ca2 | 4019 | "block %llu, max_blocks %u, flags %x, allocated %u\n", |
e35fd660 | 4020 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
4021 | flags, allocated); |
4022 | ext4_ext_show_leaf(inode, path); | |
4023 | ||
27dd4385 | 4024 | /* |
556615dc | 4025 | * When writing into unwritten space, we should not fail to |
27dd4385 LC |
4026 | * allocate metadata blocks for the new extent block if needed. |
4027 | */ | |
4028 | flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
4029 | ||
556615dc | 4030 | trace_ext4_ext_handle_unwritten_extents(inode, map, flags, |
b5645534 | 4031 | allocated, newblock); |
d8990240 | 4032 | |
c7064ef1 | 4033 | /* get_block() before submit the IO, split the extent */ |
c8b459f4 | 4034 | if (flags & EXT4_GET_BLOCKS_PRE_IO) { |
b8a86845 LC |
4035 | ret = ext4_split_convert_extents(handle, inode, map, |
4036 | path, flags | EXT4_GET_BLOCKS_CONVERT); | |
82e54229 DM |
4037 | if (ret <= 0) |
4038 | goto out; | |
5f524950 M |
4039 | /* |
4040 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 4041 | * that this IO needs to conversion to written when IO is |
5f524950 M |
4042 | * completed |
4043 | */ | |
0edeb71d TM |
4044 | if (io) |
4045 | ext4_set_io_unwritten_flag(inode, io); | |
4046 | else | |
19f5fb7a | 4047 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
a25a4e1a | 4048 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
4049 | goto out; |
4050 | } | |
c7064ef1 | 4051 | /* IO end_io complete, convert the filled extent to written */ |
c8b459f4 | 4052 | if (flags & EXT4_GET_BLOCKS_CONVERT) { |
dee1f973 | 4053 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 4054 | path); |
58590b06 | 4055 | if (ret >= 0) { |
b436b9be | 4056 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
4057 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
4058 | path, map->m_len); | |
58590b06 TT |
4059 | } else |
4060 | err = ret; | |
cdee7843 | 4061 | map->m_flags |= EXT4_MAP_MAPPED; |
15cc1767 | 4062 | map->m_pblk = newblock; |
cdee7843 ZL |
4063 | if (allocated > map->m_len) |
4064 | allocated = map->m_len; | |
4065 | map->m_len = allocated; | |
0031462b MC |
4066 | goto out2; |
4067 | } | |
4068 | /* buffered IO case */ | |
4069 | /* | |
4070 | * repeat fallocate creation request | |
4071 | * we already have an unwritten extent | |
4072 | */ | |
556615dc | 4073 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) { |
a25a4e1a | 4074 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b | 4075 | goto map_out; |
a25a4e1a | 4076 | } |
0031462b MC |
4077 | |
4078 | /* buffered READ or buffered write_begin() lookup */ | |
4079 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
4080 | /* | |
4081 | * We have blocks reserved already. We | |
4082 | * return allocated blocks so that delalloc | |
4083 | * won't do block reservation for us. But | |
4084 | * the buffer head will be unmapped so that | |
4085 | * a read from the block returns 0s. | |
4086 | */ | |
e35fd660 | 4087 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
4088 | goto out1; |
4089 | } | |
4090 | ||
4091 | /* buffered write, writepage time, convert*/ | |
27dd4385 | 4092 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags); |
a4e5d88b | 4093 | if (ret >= 0) |
b436b9be | 4094 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
4095 | out: |
4096 | if (ret <= 0) { | |
4097 | err = ret; | |
4098 | goto out2; | |
4099 | } else | |
4100 | allocated = ret; | |
e35fd660 | 4101 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
4102 | /* |
4103 | * if we allocated more blocks than requested | |
4104 | * we need to make sure we unmap the extra block | |
4105 | * allocated. The actual needed block will get | |
4106 | * unmapped later when we find the buffer_head marked | |
4107 | * new. | |
4108 | */ | |
e35fd660 | 4109 | if (allocated > map->m_len) { |
515f41c3 | 4110 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
4111 | newblock + map->m_len, |
4112 | allocated - map->m_len); | |
4113 | allocated = map->m_len; | |
515f41c3 | 4114 | } |
3a225670 | 4115 | map->m_len = allocated; |
5f634d06 AK |
4116 | |
4117 | /* | |
4118 | * If we have done fallocate with the offset that is already | |
4119 | * delayed allocated, we would have block reservation | |
4120 | * and quota reservation done in the delayed write path. | |
4121 | * But fallocate would have already updated quota and block | |
4122 | * count for this offset. So cancel these reservation | |
4123 | */ | |
7b415bf6 AK |
4124 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
4125 | unsigned int reserved_clusters; | |
4126 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4127 | map->m_lblk, map->m_len); | |
4128 | if (reserved_clusters) | |
4129 | ext4_da_update_reserve_space(inode, | |
4130 | reserved_clusters, | |
4131 | 0); | |
4132 | } | |
5f634d06 | 4133 | |
0031462b | 4134 | map_out: |
e35fd660 | 4135 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
4136 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
4137 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
4138 | map->m_len); | |
4139 | if (err < 0) | |
4140 | goto out2; | |
4141 | } | |
0031462b | 4142 | out1: |
e35fd660 TT |
4143 | if (allocated > map->m_len) |
4144 | allocated = map->m_len; | |
0031462b | 4145 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4146 | map->m_pblk = newblock; |
4147 | map->m_len = allocated; | |
0031462b | 4148 | out2: |
0031462b MC |
4149 | return err ? err : allocated; |
4150 | } | |
58590b06 | 4151 | |
4d33b1ef TT |
4152 | /* |
4153 | * get_implied_cluster_alloc - check to see if the requested | |
4154 | * allocation (in the map structure) overlaps with a cluster already | |
4155 | * allocated in an extent. | |
d8990240 | 4156 | * @sb The filesystem superblock structure |
4d33b1ef TT |
4157 | * @map The requested lblk->pblk mapping |
4158 | * @ex The extent structure which might contain an implied | |
4159 | * cluster allocation | |
4160 | * | |
4161 | * This function is called by ext4_ext_map_blocks() after we failed to | |
4162 | * find blocks that were already in the inode's extent tree. Hence, | |
4163 | * we know that the beginning of the requested region cannot overlap | |
4164 | * the extent from the inode's extent tree. There are three cases we | |
4165 | * want to catch. The first is this case: | |
4166 | * | |
4167 | * |--- cluster # N--| | |
4168 | * |--- extent ---| |---- requested region ---| | |
4169 | * |==========| | |
4170 | * | |
4171 | * The second case that we need to test for is this one: | |
4172 | * | |
4173 | * |--------- cluster # N ----------------| | |
4174 | * |--- requested region --| |------- extent ----| | |
4175 | * |=======================| | |
4176 | * | |
4177 | * The third case is when the requested region lies between two extents | |
4178 | * within the same cluster: | |
4179 | * |------------- cluster # N-------------| | |
4180 | * |----- ex -----| |---- ex_right ----| | |
4181 | * |------ requested region ------| | |
4182 | * |================| | |
4183 | * | |
4184 | * In each of the above cases, we need to set the map->m_pblk and | |
4185 | * map->m_len so it corresponds to the return the extent labelled as | |
4186 | * "|====|" from cluster #N, since it is already in use for data in | |
4187 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
4188 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
4189 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
4190 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
4191 | * by calling ext4_mb_new_blocks(). | |
4192 | */ | |
d8990240 | 4193 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
4194 | struct ext4_map_blocks *map, |
4195 | struct ext4_extent *ex, | |
4196 | struct ext4_ext_path *path) | |
4197 | { | |
d8990240 | 4198 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
f5a44db5 | 4199 | ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef | 4200 | ext4_lblk_t ex_cluster_start, ex_cluster_end; |
14d7f3ef | 4201 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
4202 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
4203 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
4204 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
4205 | ||
4206 | /* The extent passed in that we are trying to match */ | |
4207 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
4208 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
4209 | ||
4210 | /* The requested region passed into ext4_map_blocks() */ | |
4211 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
4212 | |
4213 | if ((rr_cluster_start == ex_cluster_end) || | |
4214 | (rr_cluster_start == ex_cluster_start)) { | |
4215 | if (rr_cluster_start == ex_cluster_end) | |
4216 | ee_start += ee_len - 1; | |
f5a44db5 | 4217 | map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset; |
4d33b1ef TT |
4218 | map->m_len = min(map->m_len, |
4219 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
4220 | /* | |
4221 | * Check for and handle this case: | |
4222 | * | |
4223 | * |--------- cluster # N-------------| | |
4224 | * |------- extent ----| | |
4225 | * |--- requested region ---| | |
4226 | * |===========| | |
4227 | */ | |
4228 | ||
4229 | if (map->m_lblk < ee_block) | |
4230 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
4231 | ||
4232 | /* | |
4233 | * Check for the case where there is already another allocated | |
4234 | * block to the right of 'ex' but before the end of the cluster. | |
4235 | * | |
4236 | * |------------- cluster # N-------------| | |
4237 | * |----- ex -----| |---- ex_right ----| | |
4238 | * |------ requested region ------| | |
4239 | * |================| | |
4240 | */ | |
4241 | if (map->m_lblk > ee_block) { | |
4242 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
4243 | map->m_len = min(map->m_len, next - map->m_lblk); | |
4244 | } | |
d8990240 AK |
4245 | |
4246 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
4247 | return 1; |
4248 | } | |
d8990240 AK |
4249 | |
4250 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
4251 | return 0; |
4252 | } | |
4253 | ||
4254 | ||
c278bfec | 4255 | /* |
f5ab0d1f MC |
4256 | * Block allocation/map/preallocation routine for extents based files |
4257 | * | |
4258 | * | |
c278bfec | 4259 | * Need to be called with |
0e855ac8 AK |
4260 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
4261 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
4262 | * |
4263 | * return > 0, number of of blocks already mapped/allocated | |
4264 | * if create == 0 and these are pre-allocated blocks | |
4265 | * buffer head is unmapped | |
4266 | * otherwise blocks are mapped | |
4267 | * | |
4268 | * return = 0, if plain look up failed (blocks have not been allocated) | |
4269 | * buffer head is unmapped | |
4270 | * | |
4271 | * return < 0, error case. | |
c278bfec | 4272 | */ |
e35fd660 TT |
4273 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
4274 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
4275 | { |
4276 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
4277 | struct ext4_extent newex, *ex, *ex2; |
4278 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 4279 | ext4_fsblk_t newblock = 0; |
ce37c429 | 4280 | int free_on_err = 0, err = 0, depth, ret; |
4d33b1ef | 4281 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 4282 | unsigned int allocated_clusters = 0; |
c9de560d | 4283 | struct ext4_allocation_request ar; |
f45ee3a1 | 4284 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 4285 | ext4_lblk_t cluster_offset; |
82e54229 | 4286 | int set_unwritten = 0; |
a86c6181 | 4287 | |
84fe3bef | 4288 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4289 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4290 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4291 | |
a86c6181 | 4292 | /* find extent for this block */ |
107a7bd3 | 4293 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0); |
a86c6181 AT |
4294 | if (IS_ERR(path)) { |
4295 | err = PTR_ERR(path); | |
4296 | path = NULL; | |
4297 | goto out2; | |
4298 | } | |
4299 | ||
4300 | depth = ext_depth(inode); | |
4301 | ||
4302 | /* | |
d0d856e8 RD |
4303 | * consistent leaf must not be empty; |
4304 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
4305 | * this is why assert can't be put in ext4_ext_find_extent() |
4306 | */ | |
273df556 FM |
4307 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4308 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4309 | "lblock: %lu, depth: %d pblock %lld", |
4310 | (unsigned long) map->m_lblk, depth, | |
4311 | path[depth].p_block); | |
034fb4c9 SP |
4312 | err = -EIO; |
4313 | goto out2; | |
4314 | } | |
a86c6181 | 4315 | |
7e028976 AM |
4316 | ex = path[depth].p_ext; |
4317 | if (ex) { | |
725d26d3 | 4318 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4319 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4320 | unsigned short ee_len; |
471d4011 | 4321 | |
b8a86845 | 4322 | |
471d4011 | 4323 | /* |
556615dc | 4324 | * unwritten extents are treated as holes, except that |
56055d3a | 4325 | * we split out initialized portions during a write. |
471d4011 | 4326 | */ |
a2df2a63 | 4327 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4328 | |
4329 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4330 | ||
d0d856e8 | 4331 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4332 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4333 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4334 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4335 | allocated = ee_len - (map->m_lblk - ee_block); |
4336 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4337 | ee_block, ee_len, newblock); | |
56055d3a | 4338 | |
b8a86845 LC |
4339 | /* |
4340 | * If the extent is initialized check whether the | |
4341 | * caller wants to convert it to unwritten. | |
4342 | */ | |
556615dc | 4343 | if ((!ext4_ext_is_unwritten(ex)) && |
b8a86845 LC |
4344 | (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) { |
4345 | allocated = ext4_ext_convert_initialized_extent( | |
4346 | handle, inode, map, path, flags, | |
4347 | allocated, newblock); | |
4348 | goto out2; | |
556615dc | 4349 | } else if (!ext4_ext_is_unwritten(ex)) |
7877191c | 4350 | goto out; |
69eb33dc | 4351 | |
556615dc | 4352 | ret = ext4_ext_handle_unwritten_extents( |
7877191c LC |
4353 | handle, inode, map, path, flags, |
4354 | allocated, newblock); | |
ce37c429 EW |
4355 | if (ret < 0) |
4356 | err = ret; | |
4357 | else | |
4358 | allocated = ret; | |
31cf0f2c | 4359 | goto out2; |
a86c6181 AT |
4360 | } |
4361 | } | |
4362 | ||
7b415bf6 | 4363 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 4364 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
4365 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4366 | ||
a86c6181 | 4367 | /* |
d0d856e8 | 4368 | * requested block isn't allocated yet; |
a86c6181 AT |
4369 | * we couldn't try to create block if create flag is zero |
4370 | */ | |
c2177057 | 4371 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4372 | /* |
4373 | * put just found gap into cache to speed up | |
4374 | * subsequent requests | |
4375 | */ | |
d100eef2 ZL |
4376 | if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0) |
4377 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); | |
a86c6181 AT |
4378 | goto out2; |
4379 | } | |
4d33b1ef | 4380 | |
a86c6181 | 4381 | /* |
c2ea3fde | 4382 | * Okay, we need to do block allocation. |
63f57933 | 4383 | */ |
7b415bf6 | 4384 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef | 4385 | newex.ee_block = cpu_to_le32(map->m_lblk); |
d0abafac | 4386 | cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4387 | |
4388 | /* | |
4389 | * If we are doing bigalloc, check to see if the extent returned | |
4390 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4391 | */ | |
4392 | if (cluster_offset && ex && | |
d8990240 | 4393 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4394 | ar.len = allocated = map->m_len; |
4395 | newblock = map->m_pblk; | |
7b415bf6 | 4396 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4397 | goto got_allocated_blocks; |
4398 | } | |
a86c6181 | 4399 | |
c9de560d | 4400 | /* find neighbour allocated blocks */ |
e35fd660 | 4401 | ar.lleft = map->m_lblk; |
c9de560d AT |
4402 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4403 | if (err) | |
4404 | goto out2; | |
e35fd660 | 4405 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4406 | ex2 = NULL; |
4407 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4408 | if (err) |
4409 | goto out2; | |
25d14f98 | 4410 | |
4d33b1ef TT |
4411 | /* Check if the extent after searching to the right implies a |
4412 | * cluster we can use. */ | |
4413 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4414 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4415 | ar.len = allocated = map->m_len; |
4416 | newblock = map->m_pblk; | |
7b415bf6 | 4417 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4418 | goto got_allocated_blocks; |
4419 | } | |
4420 | ||
749269fa AA |
4421 | /* |
4422 | * See if request is beyond maximum number of blocks we can have in | |
4423 | * a single extent. For an initialized extent this limit is | |
556615dc LC |
4424 | * EXT_INIT_MAX_LEN and for an unwritten extent this limit is |
4425 | * EXT_UNWRITTEN_MAX_LEN. | |
749269fa | 4426 | */ |
e35fd660 | 4427 | if (map->m_len > EXT_INIT_MAX_LEN && |
556615dc | 4428 | !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) |
e35fd660 | 4429 | map->m_len = EXT_INIT_MAX_LEN; |
556615dc LC |
4430 | else if (map->m_len > EXT_UNWRITTEN_MAX_LEN && |
4431 | (flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) | |
4432 | map->m_len = EXT_UNWRITTEN_MAX_LEN; | |
749269fa | 4433 | |
e35fd660 | 4434 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4435 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4436 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4437 | if (err) |
b939e376 | 4438 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4439 | else |
e35fd660 | 4440 | allocated = map->m_len; |
c9de560d AT |
4441 | |
4442 | /* allocate new block */ | |
4443 | ar.inode = inode; | |
e35fd660 TT |
4444 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4445 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4446 | /* |
4447 | * We calculate the offset from the beginning of the cluster | |
4448 | * for the logical block number, since when we allocate a | |
4449 | * physical cluster, the physical block should start at the | |
4450 | * same offset from the beginning of the cluster. This is | |
4451 | * needed so that future calls to get_implied_cluster_alloc() | |
4452 | * work correctly. | |
4453 | */ | |
f5a44db5 | 4454 | offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4455 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); |
4456 | ar.goal -= offset; | |
4457 | ar.logical -= offset; | |
c9de560d AT |
4458 | if (S_ISREG(inode->i_mode)) |
4459 | ar.flags = EXT4_MB_HINT_DATA; | |
4460 | else | |
4461 | /* disable in-core preallocation for non-regular files */ | |
4462 | ar.flags = 0; | |
556b27ab VH |
4463 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4464 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4465 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4466 | if (!newblock) |
4467 | goto out2; | |
84fe3bef | 4468 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4469 | ar.goal, newblock, allocated); |
4d33b1ef | 4470 | free_on_err = 1; |
7b415bf6 | 4471 | allocated_clusters = ar.len; |
4d33b1ef TT |
4472 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4473 | if (ar.len > allocated) | |
4474 | ar.len = allocated; | |
a86c6181 | 4475 | |
4d33b1ef | 4476 | got_allocated_blocks: |
a86c6181 | 4477 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4478 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4479 | newex.ee_len = cpu_to_le16(ar.len); |
556615dc LC |
4480 | /* Mark unwritten */ |
4481 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){ | |
4482 | ext4_ext_mark_unwritten(&newex); | |
a25a4e1a | 4483 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4484 | /* |
744692dc | 4485 | * io_end structure was created for every IO write to an |
556615dc | 4486 | * unwritten extent. To avoid unnecessary conversion, |
744692dc | 4487 | * here we flag the IO that really needs the conversion. |
5f524950 | 4488 | * For non asycn direct IO case, flag the inode state |
25985edc | 4489 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4490 | */ |
c8b459f4 | 4491 | if (flags & EXT4_GET_BLOCKS_PRE_IO) |
82e54229 | 4492 | set_unwritten = 1; |
8d5d02e6 | 4493 | } |
c8d46e41 | 4494 | |
a4e5d88b DM |
4495 | err = 0; |
4496 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4497 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4498 | path, ar.len); | |
575a1d4b JZ |
4499 | if (!err) |
4500 | err = ext4_ext_insert_extent(handle, inode, path, | |
4501 | &newex, flags); | |
82e54229 DM |
4502 | |
4503 | if (!err && set_unwritten) { | |
4504 | if (io) | |
4505 | ext4_set_io_unwritten_flag(inode, io); | |
4506 | else | |
4507 | ext4_set_inode_state(inode, | |
4508 | EXT4_STATE_DIO_UNWRITTEN); | |
4509 | } | |
4510 | ||
4d33b1ef | 4511 | if (err && free_on_err) { |
7132de74 MP |
4512 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4513 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4514 | /* free data blocks we just allocated */ |
c9de560d AT |
4515 | /* not a good idea to call discard here directly, |
4516 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4517 | ext4_discard_preallocations(inode); |
c8e15130 TT |
4518 | ext4_free_blocks(handle, inode, NULL, newblock, |
4519 | EXT4_C2B(sbi, allocated_clusters), fb_flags); | |
a86c6181 | 4520 | goto out2; |
315054f0 | 4521 | } |
a86c6181 | 4522 | |
a86c6181 | 4523 | /* previous routine could use block we allocated */ |
bf89d16f | 4524 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4525 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4526 | if (allocated > map->m_len) |
4527 | allocated = map->m_len; | |
4528 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4529 | |
5f634d06 AK |
4530 | /* |
4531 | * Update reserved blocks/metadata blocks after successful | |
4532 | * block allocation which had been deferred till now. | |
4533 | */ | |
7b415bf6 | 4534 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4535 | unsigned int reserved_clusters; |
7b415bf6 | 4536 | /* |
81fdbb4a | 4537 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4538 | */ |
4539 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4540 | map->m_lblk, allocated); | |
4541 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4542 | if (reserved_clusters) { | |
4543 | /* | |
4544 | * We have clusters reserved for this range. | |
4545 | * But since we are not doing actual allocation | |
4546 | * and are simply using blocks from previously | |
4547 | * allocated cluster, we should release the | |
4548 | * reservation and not claim quota. | |
4549 | */ | |
4550 | ext4_da_update_reserve_space(inode, | |
4551 | reserved_clusters, 0); | |
4552 | } | |
4553 | } else { | |
4554 | BUG_ON(allocated_clusters < reserved_clusters); | |
7b415bf6 | 4555 | if (reserved_clusters < allocated_clusters) { |
5356f261 | 4556 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4557 | int reservation = allocated_clusters - |
4558 | reserved_clusters; | |
4559 | /* | |
4560 | * It seems we claimed few clusters outside of | |
4561 | * the range of this allocation. We should give | |
4562 | * it back to the reservation pool. This can | |
4563 | * happen in the following case: | |
4564 | * | |
4565 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4566 | * cluster has 4 blocks. Thus, the clusters | |
4567 | * are [0-3],[4-7],[8-11]... | |
4568 | * * First comes delayed allocation write for | |
4569 | * logical blocks 10 & 11. Since there were no | |
4570 | * previous delayed allocated blocks in the | |
4571 | * range [8-11], we would reserve 1 cluster | |
4572 | * for this write. | |
4573 | * * Next comes write for logical blocks 3 to 8. | |
4574 | * In this case, we will reserve 2 clusters | |
4575 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4576 | * that range has a delayed allocated blocks. | |
4577 | * Thus total reserved clusters now becomes 3. | |
4578 | * * Now, during the delayed allocation writeout | |
4579 | * time, we will first write blocks [3-8] and | |
4580 | * allocate 3 clusters for writing these | |
4581 | * blocks. Also, we would claim all these | |
4582 | * three clusters above. | |
4583 | * * Now when we come here to writeout the | |
4584 | * blocks [10-11], we would expect to claim | |
4585 | * the reservation of 1 cluster we had made | |
4586 | * (and we would claim it since there are no | |
4587 | * more delayed allocated blocks in the range | |
4588 | * [8-11]. But our reserved cluster count had | |
4589 | * already gone to 0. | |
4590 | * | |
4591 | * Thus, at the step 4 above when we determine | |
4592 | * that there are still some unwritten delayed | |
4593 | * allocated blocks outside of our current | |
4594 | * block range, we should increment the | |
4595 | * reserved clusters count so that when the | |
4596 | * remaining blocks finally gets written, we | |
4597 | * could claim them. | |
4598 | */ | |
5356f261 AK |
4599 | dquot_reserve_block(inode, |
4600 | EXT4_C2B(sbi, reservation)); | |
4601 | spin_lock(&ei->i_block_reservation_lock); | |
4602 | ei->i_reserved_data_blocks += reservation; | |
4603 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 | 4604 | } |
232ec872 LC |
4605 | /* |
4606 | * We will claim quota for all newly allocated blocks. | |
4607 | * We're updating the reserved space *after* the | |
4608 | * correction above so we do not accidentally free | |
4609 | * all the metadata reservation because we might | |
4610 | * actually need it later on. | |
4611 | */ | |
4612 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4613 | 1); | |
7b415bf6 AK |
4614 | } |
4615 | } | |
5f634d06 | 4616 | |
b436b9be JK |
4617 | /* |
4618 | * Cache the extent and update transaction to commit on fdatasync only | |
556615dc | 4619 | * when it is _not_ an unwritten extent. |
b436b9be | 4620 | */ |
556615dc | 4621 | if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0) |
b436b9be | 4622 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4623 | else |
b436b9be | 4624 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4625 | out: |
e35fd660 TT |
4626 | if (allocated > map->m_len) |
4627 | allocated = map->m_len; | |
a86c6181 | 4628 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4629 | map->m_flags |= EXT4_MAP_MAPPED; |
4630 | map->m_pblk = newblock; | |
4631 | map->m_len = allocated; | |
a86c6181 AT |
4632 | out2: |
4633 | if (path) { | |
4634 | ext4_ext_drop_refs(path); | |
4635 | kfree(path); | |
4636 | } | |
e861304b | 4637 | |
63b99968 TT |
4638 | trace_ext4_ext_map_blocks_exit(inode, flags, map, |
4639 | err ? err : allocated); | |
4640 | ext4_es_lru_add(inode); | |
7877191c | 4641 | return err ? err : allocated; |
a86c6181 AT |
4642 | } |
4643 | ||
819c4920 | 4644 | void ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4645 | { |
a86c6181 | 4646 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4647 | ext4_lblk_t last_block; |
a86c6181 AT |
4648 | int err = 0; |
4649 | ||
a86c6181 | 4650 | /* |
d0d856e8 RD |
4651 | * TODO: optimization is possible here. |
4652 | * Probably we need not scan at all, | |
4653 | * because page truncation is enough. | |
a86c6181 | 4654 | */ |
a86c6181 AT |
4655 | |
4656 | /* we have to know where to truncate from in crash case */ | |
4657 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4658 | ext4_mark_inode_dirty(handle, inode); | |
4659 | ||
4660 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4661 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
8acd5e9b | 4662 | retry: |
51865fda ZL |
4663 | err = ext4_es_remove_extent(inode, last_block, |
4664 | EXT_MAX_BLOCKS - last_block); | |
94eec0fc | 4665 | if (err == -ENOMEM) { |
8acd5e9b TT |
4666 | cond_resched(); |
4667 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
4668 | goto retry; | |
4669 | } | |
4670 | if (err) { | |
4671 | ext4_std_error(inode->i_sb, err); | |
4672 | return; | |
4673 | } | |
5f95d21f | 4674 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
8acd5e9b | 4675 | ext4_std_error(inode->i_sb, err); |
a86c6181 AT |
4676 | } |
4677 | ||
0e8b6879 | 4678 | static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset, |
c174e6d6 DM |
4679 | ext4_lblk_t len, loff_t new_size, |
4680 | int flags, int mode) | |
0e8b6879 LC |
4681 | { |
4682 | struct inode *inode = file_inode(file); | |
4683 | handle_t *handle; | |
4684 | int ret = 0; | |
4685 | int ret2 = 0; | |
4686 | int retries = 0; | |
4687 | struct ext4_map_blocks map; | |
4688 | unsigned int credits; | |
c174e6d6 | 4689 | loff_t epos; |
0e8b6879 LC |
4690 | |
4691 | map.m_lblk = offset; | |
c174e6d6 | 4692 | map.m_len = len; |
0e8b6879 LC |
4693 | /* |
4694 | * Don't normalize the request if it can fit in one extent so | |
4695 | * that it doesn't get unnecessarily split into multiple | |
4696 | * extents. | |
4697 | */ | |
556615dc | 4698 | if (len <= EXT_UNWRITTEN_MAX_LEN) |
0e8b6879 LC |
4699 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; |
4700 | ||
4701 | /* | |
4702 | * credits to insert 1 extent into extent tree | |
4703 | */ | |
4704 | credits = ext4_chunk_trans_blocks(inode, len); | |
4705 | ||
4706 | retry: | |
c174e6d6 | 4707 | while (ret >= 0 && len) { |
0e8b6879 LC |
4708 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4709 | credits); | |
4710 | if (IS_ERR(handle)) { | |
4711 | ret = PTR_ERR(handle); | |
4712 | break; | |
4713 | } | |
4714 | ret = ext4_map_blocks(handle, inode, &map, flags); | |
4715 | if (ret <= 0) { | |
4716 | ext4_debug("inode #%lu: block %u: len %u: " | |
4717 | "ext4_ext_map_blocks returned %d", | |
4718 | inode->i_ino, map.m_lblk, | |
4719 | map.m_len, ret); | |
4720 | ext4_mark_inode_dirty(handle, inode); | |
4721 | ret2 = ext4_journal_stop(handle); | |
4722 | break; | |
4723 | } | |
c174e6d6 DM |
4724 | map.m_lblk += ret; |
4725 | map.m_len = len = len - ret; | |
4726 | epos = (loff_t)map.m_lblk << inode->i_blkbits; | |
4727 | inode->i_ctime = ext4_current_time(inode); | |
4728 | if (new_size) { | |
4729 | if (epos > new_size) | |
4730 | epos = new_size; | |
4731 | if (ext4_update_inode_size(inode, epos) & 0x1) | |
4732 | inode->i_mtime = inode->i_ctime; | |
4733 | } else { | |
4734 | if (epos > inode->i_size) | |
4735 | ext4_set_inode_flag(inode, | |
4736 | EXT4_INODE_EOFBLOCKS); | |
4737 | } | |
4738 | ext4_mark_inode_dirty(handle, inode); | |
0e8b6879 LC |
4739 | ret2 = ext4_journal_stop(handle); |
4740 | if (ret2) | |
4741 | break; | |
4742 | } | |
4743 | if (ret == -ENOSPC && | |
4744 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4745 | ret = 0; | |
4746 | goto retry; | |
4747 | } | |
4748 | ||
4749 | return ret > 0 ? ret2 : ret; | |
4750 | } | |
4751 | ||
b8a86845 LC |
4752 | static long ext4_zero_range(struct file *file, loff_t offset, |
4753 | loff_t len, int mode) | |
4754 | { | |
4755 | struct inode *inode = file_inode(file); | |
4756 | handle_t *handle = NULL; | |
4757 | unsigned int max_blocks; | |
4758 | loff_t new_size = 0; | |
4759 | int ret = 0; | |
4760 | int flags; | |
69dc9536 | 4761 | int credits; |
c174e6d6 | 4762 | int partial_begin, partial_end; |
b8a86845 LC |
4763 | loff_t start, end; |
4764 | ext4_lblk_t lblk; | |
4765 | struct address_space *mapping = inode->i_mapping; | |
4766 | unsigned int blkbits = inode->i_blkbits; | |
4767 | ||
4768 | trace_ext4_zero_range(inode, offset, len, mode); | |
4769 | ||
6c5e73d3 | 4770 | if (!S_ISREG(inode->i_mode)) |
4771 | return -EINVAL; | |
4772 | ||
e1ee60fd NJ |
4773 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
4774 | if (ext4_should_journal_data(inode)) { | |
4775 | ret = ext4_force_commit(inode->i_sb); | |
4776 | if (ret) | |
4777 | return ret; | |
4778 | } | |
4779 | ||
b8a86845 LC |
4780 | /* |
4781 | * Write out all dirty pages to avoid race conditions | |
4782 | * Then release them. | |
4783 | */ | |
4784 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4785 | ret = filemap_write_and_wait_range(mapping, offset, | |
4786 | offset + len - 1); | |
4787 | if (ret) | |
4788 | return ret; | |
4789 | } | |
4790 | ||
4791 | /* | |
4792 | * Round up offset. This is not fallocate, we neet to zero out | |
4793 | * blocks, so convert interior block aligned part of the range to | |
4794 | * unwritten and possibly manually zero out unaligned parts of the | |
4795 | * range. | |
4796 | */ | |
4797 | start = round_up(offset, 1 << blkbits); | |
4798 | end = round_down((offset + len), 1 << blkbits); | |
4799 | ||
4800 | if (start < offset || end > offset + len) | |
4801 | return -EINVAL; | |
c174e6d6 DM |
4802 | partial_begin = offset & ((1 << blkbits) - 1); |
4803 | partial_end = (offset + len) & ((1 << blkbits) - 1); | |
b8a86845 LC |
4804 | |
4805 | lblk = start >> blkbits; | |
4806 | max_blocks = (end >> blkbits); | |
4807 | if (max_blocks < lblk) | |
4808 | max_blocks = 0; | |
4809 | else | |
4810 | max_blocks -= lblk; | |
4811 | ||
556615dc | 4812 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT | |
713e8dde TT |
4813 | EXT4_GET_BLOCKS_CONVERT_UNWRITTEN | |
4814 | EXT4_EX_NOCACHE; | |
b8a86845 LC |
4815 | if (mode & FALLOC_FL_KEEP_SIZE) |
4816 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4817 | ||
4818 | mutex_lock(&inode->i_mutex); | |
4819 | ||
4820 | /* | |
4821 | * Indirect files do not support unwritten extnets | |
4822 | */ | |
4823 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
4824 | ret = -EOPNOTSUPP; | |
4825 | goto out_mutex; | |
4826 | } | |
4827 | ||
4828 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4829 | offset + len > i_size_read(inode)) { | |
4830 | new_size = offset + len; | |
4831 | ret = inode_newsize_ok(inode, new_size); | |
4832 | if (ret) | |
4833 | goto out_mutex; | |
4834 | /* | |
4835 | * If we have a partial block after EOF we have to allocate | |
4836 | * the entire block. | |
4837 | */ | |
c174e6d6 | 4838 | if (partial_end) |
b8a86845 LC |
4839 | max_blocks += 1; |
4840 | } | |
4841 | ||
4842 | if (max_blocks > 0) { | |
4843 | ||
4844 | /* Now release the pages and zero block aligned part of pages*/ | |
4845 | truncate_pagecache_range(inode, start, end - 1); | |
c174e6d6 | 4846 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
b8a86845 LC |
4847 | |
4848 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
4849 | ext4_inode_block_unlocked_dio(inode); | |
4850 | inode_dio_wait(inode); | |
4851 | ||
713e8dde TT |
4852 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, |
4853 | flags, mode); | |
4854 | if (ret) | |
4855 | goto out_dio; | |
b8a86845 LC |
4856 | /* |
4857 | * Remove entire range from the extent status tree. | |
713e8dde TT |
4858 | * |
4859 | * ext4_es_remove_extent(inode, lblk, max_blocks) is | |
4860 | * NOT sufficient. I'm not sure why this is the case, | |
4861 | * but let's be conservative and remove the extent | |
4862 | * status tree for the entire inode. There should be | |
4863 | * no outstanding delalloc extents thanks to the | |
4864 | * filemap_write_and_wait_range() call above. | |
b8a86845 | 4865 | */ |
713e8dde | 4866 | ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); |
b8a86845 LC |
4867 | if (ret) |
4868 | goto out_dio; | |
4869 | } | |
c174e6d6 DM |
4870 | if (!partial_begin && !partial_end) |
4871 | goto out_dio; | |
4872 | ||
69dc9536 DM |
4873 | /* |
4874 | * In worst case we have to writeout two nonadjacent unwritten | |
4875 | * blocks and update the inode | |
4876 | */ | |
4877 | credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1; | |
4878 | if (ext4_should_journal_data(inode)) | |
4879 | credits += 2; | |
4880 | handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); | |
b8a86845 LC |
4881 | if (IS_ERR(handle)) { |
4882 | ret = PTR_ERR(handle); | |
4883 | ext4_std_error(inode->i_sb, ret); | |
4884 | goto out_dio; | |
4885 | } | |
4886 | ||
4887 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
e5b30416 | 4888 | if (new_size) { |
4631dbf6 | 4889 | ext4_update_inode_size(inode, new_size); |
e5b30416 | 4890 | } else { |
b8a86845 LC |
4891 | /* |
4892 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4893 | * can proceed even if the new size is the same as i_size. | |
4894 | */ | |
4895 | if ((offset + len) > i_size_read(inode)) | |
4896 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); | |
4897 | } | |
b8a86845 LC |
4898 | ext4_mark_inode_dirty(handle, inode); |
4899 | ||
4900 | /* Zero out partial block at the edges of the range */ | |
4901 | ret = ext4_zero_partial_blocks(handle, inode, offset, len); | |
4902 | ||
4903 | if (file->f_flags & O_SYNC) | |
4904 | ext4_handle_sync(handle); | |
4905 | ||
4906 | ext4_journal_stop(handle); | |
4907 | out_dio: | |
4908 | ext4_inode_resume_unlocked_dio(inode); | |
4909 | out_mutex: | |
4910 | mutex_unlock(&inode->i_mutex); | |
4911 | return ret; | |
4912 | } | |
4913 | ||
a2df2a63 | 4914 | /* |
2fe17c10 | 4915 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4916 | * operation, which gets called from sys_fallocate system call. |
4917 | * For block-mapped files, posix_fallocate should fall back to the method | |
4918 | * of writing zeroes to the required new blocks (the same behavior which is | |
4919 | * expected for file systems which do not support fallocate() system call). | |
4920 | */ | |
2fe17c10 | 4921 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4922 | { |
496ad9aa | 4923 | struct inode *inode = file_inode(file); |
f282ac19 | 4924 | loff_t new_size = 0; |
498e5f24 | 4925 | unsigned int max_blocks; |
a2df2a63 | 4926 | int ret = 0; |
a4e5d88b | 4927 | int flags; |
0e8b6879 | 4928 | ext4_lblk_t lblk; |
0e8b6879 | 4929 | unsigned int blkbits = inode->i_blkbits; |
a2df2a63 | 4930 | |
a4bb6b64 | 4931 | /* Return error if mode is not supported */ |
9eb79482 | 4932 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
b8a86845 | 4933 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
a4bb6b64 AH |
4934 | return -EOPNOTSUPP; |
4935 | ||
4936 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
aeb2817a | 4937 | return ext4_punch_hole(inode, offset, len); |
a4bb6b64 | 4938 | |
0c8d414f TM |
4939 | ret = ext4_convert_inline_data(inode); |
4940 | if (ret) | |
4941 | return ret; | |
4942 | ||
8bad6fc8 ZL |
4943 | /* |
4944 | * currently supporting (pre)allocate mode for extent-based | |
4945 | * files _only_ | |
4946 | */ | |
4947 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4948 | return -EOPNOTSUPP; | |
4949 | ||
40c406c7 TT |
4950 | if (mode & FALLOC_FL_COLLAPSE_RANGE) |
4951 | return ext4_collapse_range(inode, offset, len); | |
4952 | ||
b8a86845 LC |
4953 | if (mode & FALLOC_FL_ZERO_RANGE) |
4954 | return ext4_zero_range(file, offset, len, mode); | |
4955 | ||
0562e0ba | 4956 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
0e8b6879 | 4957 | lblk = offset >> blkbits; |
fd28784a AK |
4958 | /* |
4959 | * We can't just convert len to max_blocks because | |
4960 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4961 | */ | |
a2df2a63 | 4962 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
0e8b6879 LC |
4963 | - lblk; |
4964 | ||
556615dc | 4965 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
0e8b6879 LC |
4966 | if (mode & FALLOC_FL_KEEP_SIZE) |
4967 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4968 | ||
55bd725a | 4969 | mutex_lock(&inode->i_mutex); |
f282ac19 LC |
4970 | |
4971 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4972 | offset + len > i_size_read(inode)) { | |
4973 | new_size = offset + len; | |
4974 | ret = inode_newsize_ok(inode, new_size); | |
4975 | if (ret) | |
4976 | goto out; | |
6d19c42b | 4977 | } |
f282ac19 | 4978 | |
c174e6d6 DM |
4979 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, |
4980 | flags, mode); | |
0e8b6879 LC |
4981 | if (ret) |
4982 | goto out; | |
f282ac19 | 4983 | |
c174e6d6 DM |
4984 | if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) { |
4985 | ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal, | |
4986 | EXT4_I(inode)->i_sync_tid); | |
f282ac19 | 4987 | } |
f282ac19 | 4988 | out: |
55bd725a | 4989 | mutex_unlock(&inode->i_mutex); |
0e8b6879 LC |
4990 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
4991 | return ret; | |
a2df2a63 | 4992 | } |
6873fa0d | 4993 | |
0031462b MC |
4994 | /* |
4995 | * This function convert a range of blocks to written extents | |
4996 | * The caller of this function will pass the start offset and the size. | |
4997 | * all unwritten extents within this range will be converted to | |
4998 | * written extents. | |
4999 | * | |
5000 | * This function is called from the direct IO end io call back | |
5001 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 5002 | * Returns 0 on success. |
0031462b | 5003 | */ |
6b523df4 JK |
5004 | int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
5005 | loff_t offset, ssize_t len) | |
0031462b | 5006 | { |
0031462b MC |
5007 | unsigned int max_blocks; |
5008 | int ret = 0; | |
5009 | int ret2 = 0; | |
2ed88685 | 5010 | struct ext4_map_blocks map; |
0031462b MC |
5011 | unsigned int credits, blkbits = inode->i_blkbits; |
5012 | ||
2ed88685 | 5013 | map.m_lblk = offset >> blkbits; |
0031462b MC |
5014 | /* |
5015 | * We can't just convert len to max_blocks because | |
5016 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
5017 | */ | |
2ed88685 TT |
5018 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
5019 | map.m_lblk); | |
0031462b | 5020 | /* |
6b523df4 JK |
5021 | * This is somewhat ugly but the idea is clear: When transaction is |
5022 | * reserved, everything goes into it. Otherwise we rather start several | |
5023 | * smaller transactions for conversion of each extent separately. | |
0031462b | 5024 | */ |
6b523df4 JK |
5025 | if (handle) { |
5026 | handle = ext4_journal_start_reserved(handle, | |
5027 | EXT4_HT_EXT_CONVERT); | |
5028 | if (IS_ERR(handle)) | |
5029 | return PTR_ERR(handle); | |
5030 | credits = 0; | |
5031 | } else { | |
5032 | /* | |
5033 | * credits to insert 1 extent into extent tree | |
5034 | */ | |
5035 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
5036 | } | |
0031462b | 5037 | while (ret >= 0 && ret < max_blocks) { |
2ed88685 TT |
5038 | map.m_lblk += ret; |
5039 | map.m_len = (max_blocks -= ret); | |
6b523df4 JK |
5040 | if (credits) { |
5041 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
5042 | credits); | |
5043 | if (IS_ERR(handle)) { | |
5044 | ret = PTR_ERR(handle); | |
5045 | break; | |
5046 | } | |
0031462b | 5047 | } |
2ed88685 | 5048 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 5049 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
5050 | if (ret <= 0) |
5051 | ext4_warning(inode->i_sb, | |
5052 | "inode #%lu: block %u: len %u: " | |
5053 | "ext4_ext_map_blocks returned %d", | |
5054 | inode->i_ino, map.m_lblk, | |
5055 | map.m_len, ret); | |
0031462b | 5056 | ext4_mark_inode_dirty(handle, inode); |
6b523df4 JK |
5057 | if (credits) |
5058 | ret2 = ext4_journal_stop(handle); | |
5059 | if (ret <= 0 || ret2) | |
0031462b MC |
5060 | break; |
5061 | } | |
6b523df4 JK |
5062 | if (!credits) |
5063 | ret2 = ext4_journal_stop(handle); | |
0031462b MC |
5064 | return ret > 0 ? ret2 : ret; |
5065 | } | |
6d9c85eb | 5066 | |
6873fa0d | 5067 | /* |
69eb33dc ZL |
5068 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
5069 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
5070 | * return start of the next delayed extent. |
5071 | * | |
69eb33dc | 5072 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 5073 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 5074 | * extent found. Leave newes unmodified. |
6873fa0d | 5075 | */ |
91dd8c11 | 5076 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 5077 | struct extent_status *newes) |
6873fa0d | 5078 | { |
b3aff3e3 | 5079 | struct extent_status es; |
be401363 | 5080 | ext4_lblk_t block, next_del; |
6873fa0d | 5081 | |
69eb33dc | 5082 | if (newes->es_pblk == 0) { |
e30b5dca YZ |
5083 | ext4_es_find_delayed_extent_range(inode, newes->es_lblk, |
5084 | newes->es_lblk + newes->es_len - 1, &es); | |
5085 | ||
6d9c85eb | 5086 | /* |
69eb33dc | 5087 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 5088 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 5089 | */ |
06b0c886 | 5090 | if (es.es_len == 0) |
b3aff3e3 | 5091 | /* A hole found. */ |
91dd8c11 | 5092 | return 0; |
b3aff3e3 | 5093 | |
69eb33dc | 5094 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 5095 | /* A hole found. */ |
69eb33dc ZL |
5096 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
5097 | newes->es_len); | |
91dd8c11 | 5098 | return 0; |
6873fa0d | 5099 | } |
6d9c85eb | 5100 | |
69eb33dc | 5101 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
5102 | } |
5103 | ||
69eb33dc | 5104 | block = newes->es_lblk + newes->es_len; |
e30b5dca | 5105 | ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es); |
be401363 ZL |
5106 | if (es.es_len == 0) |
5107 | next_del = EXT_MAX_BLOCKS; | |
5108 | else | |
5109 | next_del = es.es_lblk; | |
5110 | ||
91dd8c11 | 5111 | return next_del; |
6873fa0d | 5112 | } |
6873fa0d ES |
5113 | /* fiemap flags we can handle specified here */ |
5114 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
5115 | ||
3a06d778 AK |
5116 | static int ext4_xattr_fiemap(struct inode *inode, |
5117 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
5118 | { |
5119 | __u64 physical = 0; | |
5120 | __u64 length; | |
5121 | __u32 flags = FIEMAP_EXTENT_LAST; | |
5122 | int blockbits = inode->i_sb->s_blocksize_bits; | |
5123 | int error = 0; | |
5124 | ||
5125 | /* in-inode? */ | |
19f5fb7a | 5126 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
5127 | struct ext4_iloc iloc; |
5128 | int offset; /* offset of xattr in inode */ | |
5129 | ||
5130 | error = ext4_get_inode_loc(inode, &iloc); | |
5131 | if (error) | |
5132 | return error; | |
a60697f4 | 5133 | physical = (__u64)iloc.bh->b_blocknr << blockbits; |
6873fa0d ES |
5134 | offset = EXT4_GOOD_OLD_INODE_SIZE + |
5135 | EXT4_I(inode)->i_extra_isize; | |
5136 | physical += offset; | |
5137 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
5138 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 5139 | brelse(iloc.bh); |
6873fa0d | 5140 | } else { /* external block */ |
a60697f4 | 5141 | physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
6873fa0d ES |
5142 | length = inode->i_sb->s_blocksize; |
5143 | } | |
5144 | ||
5145 | if (physical) | |
5146 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
5147 | length, flags); | |
5148 | return (error < 0 ? error : 0); | |
5149 | } | |
5150 | ||
5151 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
5152 | __u64 start, __u64 len) | |
5153 | { | |
5154 | ext4_lblk_t start_blk; | |
6873fa0d ES |
5155 | int error = 0; |
5156 | ||
94191985 TM |
5157 | if (ext4_has_inline_data(inode)) { |
5158 | int has_inline = 1; | |
5159 | ||
5160 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline); | |
5161 | ||
5162 | if (has_inline) | |
5163 | return error; | |
5164 | } | |
5165 | ||
7869a4a6 TT |
5166 | if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
5167 | error = ext4_ext_precache(inode); | |
5168 | if (error) | |
5169 | return error; | |
5170 | } | |
5171 | ||
6873fa0d | 5172 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 5173 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
5174 | return generic_block_fiemap(inode, fieinfo, start, len, |
5175 | ext4_get_block); | |
5176 | ||
5177 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
5178 | return -EBADR; | |
5179 | ||
5180 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
5181 | error = ext4_xattr_fiemap(inode, fieinfo); | |
5182 | } else { | |
aca92ff6 LM |
5183 | ext4_lblk_t len_blks; |
5184 | __u64 last_blk; | |
5185 | ||
6873fa0d | 5186 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 5187 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
5188 | if (last_blk >= EXT_MAX_BLOCKS) |
5189 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 5190 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
5191 | |
5192 | /* | |
91dd8c11 LC |
5193 | * Walk the extent tree gathering extent information |
5194 | * and pushing extents back to the user. | |
6873fa0d | 5195 | */ |
91dd8c11 LC |
5196 | error = ext4_fill_fiemap_extents(inode, start_blk, |
5197 | len_blks, fieinfo); | |
6873fa0d | 5198 | } |
107a7bd3 | 5199 | ext4_es_lru_add(inode); |
6873fa0d ES |
5200 | return error; |
5201 | } | |
9eb79482 NJ |
5202 | |
5203 | /* | |
5204 | * ext4_access_path: | |
5205 | * Function to access the path buffer for marking it dirty. | |
5206 | * It also checks if there are sufficient credits left in the journal handle | |
5207 | * to update path. | |
5208 | */ | |
5209 | static int | |
5210 | ext4_access_path(handle_t *handle, struct inode *inode, | |
5211 | struct ext4_ext_path *path) | |
5212 | { | |
5213 | int credits, err; | |
5214 | ||
5215 | if (!ext4_handle_valid(handle)) | |
5216 | return 0; | |
5217 | ||
5218 | /* | |
5219 | * Check if need to extend journal credits | |
5220 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
5221 | * descriptor) for each block group; assume two block | |
5222 | * groups | |
5223 | */ | |
5224 | if (handle->h_buffer_credits < 7) { | |
5225 | credits = ext4_writepage_trans_blocks(inode); | |
5226 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); | |
5227 | /* EAGAIN is success */ | |
5228 | if (err && err != -EAGAIN) | |
5229 | return err; | |
5230 | } | |
5231 | ||
5232 | err = ext4_ext_get_access(handle, inode, path); | |
5233 | return err; | |
5234 | } | |
5235 | ||
5236 | /* | |
5237 | * ext4_ext_shift_path_extents: | |
5238 | * Shift the extents of a path structure lying between path[depth].p_ext | |
5239 | * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift | |
5240 | * from starting block for each extent. | |
5241 | */ | |
5242 | static int | |
5243 | ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift, | |
5244 | struct inode *inode, handle_t *handle, | |
5245 | ext4_lblk_t *start) | |
5246 | { | |
5247 | int depth, err = 0; | |
5248 | struct ext4_extent *ex_start, *ex_last; | |
5249 | bool update = 0; | |
5250 | depth = path->p_depth; | |
5251 | ||
5252 | while (depth >= 0) { | |
5253 | if (depth == path->p_depth) { | |
5254 | ex_start = path[depth].p_ext; | |
5255 | if (!ex_start) | |
5256 | return -EIO; | |
5257 | ||
5258 | ex_last = EXT_LAST_EXTENT(path[depth].p_hdr); | |
5259 | if (!ex_last) | |
5260 | return -EIO; | |
5261 | ||
5262 | err = ext4_access_path(handle, inode, path + depth); | |
5263 | if (err) | |
5264 | goto out; | |
5265 | ||
5266 | if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
5267 | update = 1; | |
5268 | ||
847c6c42 | 5269 | *start = le32_to_cpu(ex_last->ee_block) + |
9eb79482 NJ |
5270 | ext4_ext_get_actual_len(ex_last); |
5271 | ||
5272 | while (ex_start <= ex_last) { | |
847c6c42 | 5273 | le32_add_cpu(&ex_start->ee_block, -shift); |
6dd834ef LC |
5274 | /* Try to merge to the left. */ |
5275 | if ((ex_start > | |
5276 | EXT_FIRST_EXTENT(path[depth].p_hdr)) && | |
5277 | ext4_ext_try_to_merge_right(inode, | |
5278 | path, ex_start - 1)) | |
5279 | ex_last--; | |
5280 | else | |
5281 | ex_start++; | |
9eb79482 NJ |
5282 | } |
5283 | err = ext4_ext_dirty(handle, inode, path + depth); | |
5284 | if (err) | |
5285 | goto out; | |
5286 | ||
5287 | if (--depth < 0 || !update) | |
5288 | break; | |
5289 | } | |
5290 | ||
5291 | /* Update index too */ | |
5292 | err = ext4_access_path(handle, inode, path + depth); | |
5293 | if (err) | |
5294 | goto out; | |
5295 | ||
847c6c42 | 5296 | le32_add_cpu(&path[depth].p_idx->ei_block, -shift); |
9eb79482 NJ |
5297 | err = ext4_ext_dirty(handle, inode, path + depth); |
5298 | if (err) | |
5299 | goto out; | |
5300 | ||
5301 | /* we are done if current index is not a starting index */ | |
5302 | if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr)) | |
5303 | break; | |
5304 | ||
5305 | depth--; | |
5306 | } | |
5307 | ||
5308 | out: | |
5309 | return err; | |
5310 | } | |
5311 | ||
5312 | /* | |
5313 | * ext4_ext_shift_extents: | |
5314 | * All the extents which lies in the range from start to the last allocated | |
5315 | * block for the file are shifted downwards by shift blocks. | |
5316 | * On success, 0 is returned, error otherwise. | |
5317 | */ | |
5318 | static int | |
5319 | ext4_ext_shift_extents(struct inode *inode, handle_t *handle, | |
5320 | ext4_lblk_t start, ext4_lblk_t shift) | |
5321 | { | |
5322 | struct ext4_ext_path *path; | |
5323 | int ret = 0, depth; | |
5324 | struct ext4_extent *extent; | |
f8fb4f41 | 5325 | ext4_lblk_t stop_block; |
9eb79482 NJ |
5326 | ext4_lblk_t ex_start, ex_end; |
5327 | ||
5328 | /* Let path point to the last extent */ | |
5329 | path = ext4_ext_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0); | |
5330 | if (IS_ERR(path)) | |
5331 | return PTR_ERR(path); | |
5332 | ||
5333 | depth = path->p_depth; | |
5334 | extent = path[depth].p_ext; | |
5335 | if (!extent) { | |
5336 | ext4_ext_drop_refs(path); | |
5337 | kfree(path); | |
5338 | return ret; | |
5339 | } | |
5340 | ||
847c6c42 ZL |
5341 | stop_block = le32_to_cpu(extent->ee_block) + |
5342 | ext4_ext_get_actual_len(extent); | |
9eb79482 NJ |
5343 | ext4_ext_drop_refs(path); |
5344 | kfree(path); | |
5345 | ||
5346 | /* Nothing to shift, if hole is at the end of file */ | |
5347 | if (start >= stop_block) | |
5348 | return ret; | |
5349 | ||
5350 | /* | |
5351 | * Don't start shifting extents until we make sure the hole is big | |
5352 | * enough to accomodate the shift. | |
5353 | */ | |
5354 | path = ext4_ext_find_extent(inode, start - 1, NULL, 0); | |
8dc79ec4 DM |
5355 | if (IS_ERR(path)) |
5356 | return PTR_ERR(path); | |
9eb79482 NJ |
5357 | depth = path->p_depth; |
5358 | extent = path[depth].p_ext; | |
8dc79ec4 DM |
5359 | if (extent) { |
5360 | ex_start = le32_to_cpu(extent->ee_block); | |
5361 | ex_end = le32_to_cpu(extent->ee_block) + | |
847c6c42 | 5362 | ext4_ext_get_actual_len(extent); |
8dc79ec4 DM |
5363 | } else { |
5364 | ex_start = 0; | |
5365 | ex_end = 0; | |
5366 | } | |
9eb79482 NJ |
5367 | ext4_ext_drop_refs(path); |
5368 | kfree(path); | |
5369 | ||
5370 | if ((start == ex_start && shift > ex_start) || | |
5371 | (shift > start - ex_end)) | |
5372 | return -EINVAL; | |
5373 | ||
5374 | /* Its safe to start updating extents */ | |
5375 | while (start < stop_block) { | |
5376 | path = ext4_ext_find_extent(inode, start, NULL, 0); | |
5377 | if (IS_ERR(path)) | |
5378 | return PTR_ERR(path); | |
5379 | depth = path->p_depth; | |
5380 | extent = path[depth].p_ext; | |
a18ed359 DM |
5381 | if (!extent) { |
5382 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
5383 | (unsigned long) start); | |
5384 | return -EIO; | |
5385 | } | |
f8fb4f41 | 5386 | if (start > le32_to_cpu(extent->ee_block)) { |
9eb79482 | 5387 | /* Hole, move to the next extent */ |
f8fb4f41 DM |
5388 | if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) { |
5389 | path[depth].p_ext++; | |
5390 | } else { | |
5391 | start = ext4_ext_next_allocated_block(path); | |
9eb79482 NJ |
5392 | ext4_ext_drop_refs(path); |
5393 | kfree(path); | |
f8fb4f41 | 5394 | continue; |
9eb79482 NJ |
5395 | } |
5396 | } | |
5397 | ret = ext4_ext_shift_path_extents(path, shift, inode, | |
5398 | handle, &start); | |
5399 | ext4_ext_drop_refs(path); | |
5400 | kfree(path); | |
5401 | if (ret) | |
5402 | break; | |
5403 | } | |
5404 | ||
5405 | return ret; | |
5406 | } | |
5407 | ||
5408 | /* | |
5409 | * ext4_collapse_range: | |
5410 | * This implements the fallocate's collapse range functionality for ext4 | |
5411 | * Returns: 0 and non-zero on error. | |
5412 | */ | |
5413 | int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len) | |
5414 | { | |
5415 | struct super_block *sb = inode->i_sb; | |
5416 | ext4_lblk_t punch_start, punch_stop; | |
5417 | handle_t *handle; | |
5418 | unsigned int credits; | |
a8680e0d | 5419 | loff_t new_size, ioffset; |
9eb79482 NJ |
5420 | int ret; |
5421 | ||
9eb79482 | 5422 | /* Collapse range works only on fs block size aligned offsets. */ |
ee98fa3a NJ |
5423 | if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) || |
5424 | len & (EXT4_CLUSTER_SIZE(sb) - 1)) | |
9eb79482 NJ |
5425 | return -EINVAL; |
5426 | ||
5427 | if (!S_ISREG(inode->i_mode)) | |
86f1ca38 | 5428 | return -EINVAL; |
9eb79482 NJ |
5429 | |
5430 | trace_ext4_collapse_range(inode, offset, len); | |
5431 | ||
5432 | punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5433 | punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb); | |
5434 | ||
1ce01c4a NJ |
5435 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
5436 | if (ext4_should_journal_data(inode)) { | |
5437 | ret = ext4_force_commit(inode->i_sb); | |
5438 | if (ret) | |
5439 | return ret; | |
5440 | } | |
5441 | ||
a8680e0d NJ |
5442 | /* |
5443 | * Need to round down offset to be aligned with page size boundary | |
5444 | * for page size > block size. | |
5445 | */ | |
5446 | ioffset = round_down(offset, PAGE_SIZE); | |
5447 | ||
9eb79482 | 5448 | /* Write out all dirty pages */ |
a8680e0d NJ |
5449 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, |
5450 | LLONG_MAX); | |
9eb79482 NJ |
5451 | if (ret) |
5452 | return ret; | |
5453 | ||
5454 | /* Take mutex lock */ | |
5455 | mutex_lock(&inode->i_mutex); | |
5456 | ||
23fffa92 LC |
5457 | /* |
5458 | * There is no need to overlap collapse range with EOF, in which case | |
5459 | * it is effectively a truncate operation | |
5460 | */ | |
5461 | if (offset + len >= i_size_read(inode)) { | |
5462 | ret = -EINVAL; | |
5463 | goto out_mutex; | |
5464 | } | |
5465 | ||
9eb79482 NJ |
5466 | /* Currently just for extent based files */ |
5467 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5468 | ret = -EOPNOTSUPP; | |
5469 | goto out_mutex; | |
5470 | } | |
5471 | ||
a8680e0d | 5472 | truncate_pagecache(inode, ioffset); |
9eb79482 NJ |
5473 | |
5474 | /* Wait for existing dio to complete */ | |
5475 | ext4_inode_block_unlocked_dio(inode); | |
5476 | inode_dio_wait(inode); | |
5477 | ||
5478 | credits = ext4_writepage_trans_blocks(inode); | |
5479 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5480 | if (IS_ERR(handle)) { | |
5481 | ret = PTR_ERR(handle); | |
5482 | goto out_dio; | |
5483 | } | |
5484 | ||
5485 | down_write(&EXT4_I(inode)->i_data_sem); | |
5486 | ext4_discard_preallocations(inode); | |
5487 | ||
5488 | ret = ext4_es_remove_extent(inode, punch_start, | |
2c1d2328 | 5489 | EXT_MAX_BLOCKS - punch_start); |
9eb79482 NJ |
5490 | if (ret) { |
5491 | up_write(&EXT4_I(inode)->i_data_sem); | |
5492 | goto out_stop; | |
5493 | } | |
5494 | ||
5495 | ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1); | |
5496 | if (ret) { | |
5497 | up_write(&EXT4_I(inode)->i_data_sem); | |
5498 | goto out_stop; | |
5499 | } | |
ef24f6c2 | 5500 | ext4_discard_preallocations(inode); |
9eb79482 NJ |
5501 | |
5502 | ret = ext4_ext_shift_extents(inode, handle, punch_stop, | |
5503 | punch_stop - punch_start); | |
5504 | if (ret) { | |
5505 | up_write(&EXT4_I(inode)->i_data_sem); | |
5506 | goto out_stop; | |
5507 | } | |
5508 | ||
5509 | new_size = i_size_read(inode) - len; | |
9337d5d3 | 5510 | i_size_write(inode, new_size); |
9eb79482 NJ |
5511 | EXT4_I(inode)->i_disksize = new_size; |
5512 | ||
9eb79482 NJ |
5513 | up_write(&EXT4_I(inode)->i_data_sem); |
5514 | if (IS_SYNC(inode)) | |
5515 | ext4_handle_sync(handle); | |
5516 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
5517 | ext4_mark_inode_dirty(handle, inode); | |
5518 | ||
5519 | out_stop: | |
5520 | ext4_journal_stop(handle); | |
5521 | out_dio: | |
5522 | ext4_inode_resume_unlocked_dio(inode); | |
5523 | out_mutex: | |
5524 | mutex_unlock(&inode->i_mutex); | |
5525 | return ret; | |
5526 | } | |
fcf6b1b7 DM |
5527 | |
5528 | /** | |
5529 | * ext4_swap_extents - Swap extents between two inodes | |
5530 | * | |
5531 | * @inode1: First inode | |
5532 | * @inode2: Second inode | |
5533 | * @lblk1: Start block for first inode | |
5534 | * @lblk2: Start block for second inode | |
5535 | * @count: Number of blocks to swap | |
5536 | * @mark_unwritten: Mark second inode's extents as unwritten after swap | |
5537 | * @erp: Pointer to save error value | |
5538 | * | |
5539 | * This helper routine does exactly what is promise "swap extents". All other | |
5540 | * stuff such as page-cache locking consistency, bh mapping consistency or | |
5541 | * extent's data copying must be performed by caller. | |
5542 | * Locking: | |
5543 | * i_mutex is held for both inodes | |
5544 | * i_data_sem is locked for write for both inodes | |
5545 | * Assumptions: | |
5546 | * All pages from requested range are locked for both inodes | |
5547 | */ | |
5548 | int | |
5549 | ext4_swap_extents(handle_t *handle, struct inode *inode1, | |
5550 | struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2, | |
5551 | ext4_lblk_t count, int unwritten, int *erp) | |
5552 | { | |
5553 | struct ext4_ext_path *path1 = NULL; | |
5554 | struct ext4_ext_path *path2 = NULL; | |
5555 | int replaced_count = 0; | |
5556 | ||
5557 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem)); | |
5558 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem)); | |
5559 | BUG_ON(!mutex_is_locked(&inode1->i_mutex)); | |
5560 | BUG_ON(!mutex_is_locked(&inode1->i_mutex)); | |
5561 | ||
5562 | *erp = ext4_es_remove_extent(inode1, lblk1, count); | |
19008f6d | 5563 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5564 | return 0; |
5565 | *erp = ext4_es_remove_extent(inode2, lblk2, count); | |
19008f6d | 5566 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5567 | return 0; |
5568 | ||
5569 | while (count) { | |
5570 | struct ext4_extent *ex1, *ex2, tmp_ex; | |
5571 | ext4_lblk_t e1_blk, e2_blk; | |
5572 | int e1_len, e2_len, len; | |
5573 | int split = 0; | |
5574 | ||
5575 | path1 = ext4_ext_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE); | |
19008f6d | 5576 | if (unlikely(IS_ERR(path1))) { |
fcf6b1b7 | 5577 | *erp = PTR_ERR(path1); |
19008f6d TT |
5578 | path1 = NULL; |
5579 | finish: | |
5580 | count = 0; | |
5581 | goto repeat; | |
fcf6b1b7 DM |
5582 | } |
5583 | path2 = ext4_ext_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE); | |
19008f6d | 5584 | if (unlikely(IS_ERR(path2))) { |
fcf6b1b7 | 5585 | *erp = PTR_ERR(path2); |
19008f6d TT |
5586 | path2 = NULL; |
5587 | goto finish; | |
fcf6b1b7 DM |
5588 | } |
5589 | ex1 = path1[path1->p_depth].p_ext; | |
5590 | ex2 = path2[path2->p_depth].p_ext; | |
5591 | /* Do we have somthing to swap ? */ | |
5592 | if (unlikely(!ex2 || !ex1)) | |
19008f6d | 5593 | goto finish; |
fcf6b1b7 DM |
5594 | |
5595 | e1_blk = le32_to_cpu(ex1->ee_block); | |
5596 | e2_blk = le32_to_cpu(ex2->ee_block); | |
5597 | e1_len = ext4_ext_get_actual_len(ex1); | |
5598 | e2_len = ext4_ext_get_actual_len(ex2); | |
5599 | ||
5600 | /* Hole handling */ | |
5601 | if (!in_range(lblk1, e1_blk, e1_len) || | |
5602 | !in_range(lblk2, e2_blk, e2_len)) { | |
5603 | ext4_lblk_t next1, next2; | |
5604 | ||
5605 | /* if hole after extent, then go to next extent */ | |
5606 | next1 = ext4_ext_next_allocated_block(path1); | |
5607 | next2 = ext4_ext_next_allocated_block(path2); | |
5608 | /* If hole before extent, then shift to that extent */ | |
5609 | if (e1_blk > lblk1) | |
5610 | next1 = e1_blk; | |
5611 | if (e2_blk > lblk2) | |
5612 | next2 = e1_blk; | |
5613 | /* Do we have something to swap */ | |
5614 | if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS) | |
19008f6d | 5615 | goto finish; |
fcf6b1b7 DM |
5616 | /* Move to the rightest boundary */ |
5617 | len = next1 - lblk1; | |
5618 | if (len < next2 - lblk2) | |
5619 | len = next2 - lblk2; | |
5620 | if (len > count) | |
5621 | len = count; | |
5622 | lblk1 += len; | |
5623 | lblk2 += len; | |
5624 | count -= len; | |
5625 | goto repeat; | |
5626 | } | |
5627 | ||
5628 | /* Prepare left boundary */ | |
5629 | if (e1_blk < lblk1) { | |
5630 | split = 1; | |
5631 | *erp = ext4_force_split_extent_at(handle, inode1, | |
5632 | path1, lblk1, 0); | |
19008f6d TT |
5633 | if (unlikely(*erp)) |
5634 | goto finish; | |
fcf6b1b7 DM |
5635 | } |
5636 | if (e2_blk < lblk2) { | |
5637 | split = 1; | |
5638 | *erp = ext4_force_split_extent_at(handle, inode2, | |
5639 | path2, lblk2, 0); | |
19008f6d TT |
5640 | if (unlikely(*erp)) |
5641 | goto finish; | |
fcf6b1b7 DM |
5642 | } |
5643 | /* ext4_split_extent_at() may retult in leaf extent split, | |
5644 | * path must to be revalidated. */ | |
5645 | if (split) | |
5646 | goto repeat; | |
5647 | ||
5648 | /* Prepare right boundary */ | |
5649 | len = count; | |
5650 | if (len > e1_blk + e1_len - lblk1) | |
5651 | len = e1_blk + e1_len - lblk1; | |
5652 | if (len > e2_blk + e2_len - lblk2) | |
5653 | len = e2_blk + e2_len - lblk2; | |
5654 | ||
5655 | if (len != e1_len) { | |
5656 | split = 1; | |
5657 | *erp = ext4_force_split_extent_at(handle, inode1, | |
5658 | path1, lblk1 + len, 0); | |
19008f6d TT |
5659 | if (unlikely(*erp)) |
5660 | goto finish; | |
fcf6b1b7 DM |
5661 | } |
5662 | if (len != e2_len) { | |
5663 | split = 1; | |
5664 | *erp = ext4_force_split_extent_at(handle, inode2, | |
5665 | path2, lblk2 + len, 0); | |
5666 | if (*erp) | |
19008f6d | 5667 | goto finish; |
fcf6b1b7 DM |
5668 | } |
5669 | /* ext4_split_extent_at() may retult in leaf extent split, | |
5670 | * path must to be revalidated. */ | |
5671 | if (split) | |
5672 | goto repeat; | |
5673 | ||
5674 | BUG_ON(e2_len != e1_len); | |
5675 | *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth); | |
19008f6d TT |
5676 | if (unlikely(*erp)) |
5677 | goto finish; | |
fcf6b1b7 | 5678 | *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth); |
19008f6d TT |
5679 | if (unlikely(*erp)) |
5680 | goto finish; | |
fcf6b1b7 DM |
5681 | |
5682 | /* Both extents are fully inside boundaries. Swap it now */ | |
5683 | tmp_ex = *ex1; | |
5684 | ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2)); | |
5685 | ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex)); | |
5686 | ex1->ee_len = cpu_to_le16(e2_len); | |
5687 | ex2->ee_len = cpu_to_le16(e1_len); | |
5688 | if (unwritten) | |
5689 | ext4_ext_mark_unwritten(ex2); | |
5690 | if (ext4_ext_is_unwritten(&tmp_ex)) | |
5691 | ext4_ext_mark_unwritten(ex1); | |
5692 | ||
5693 | ext4_ext_try_to_merge(handle, inode2, path2, ex2); | |
5694 | ext4_ext_try_to_merge(handle, inode1, path1, ex1); | |
5695 | *erp = ext4_ext_dirty(handle, inode2, path2 + | |
5696 | path2->p_depth); | |
19008f6d TT |
5697 | if (unlikely(*erp)) |
5698 | goto finish; | |
fcf6b1b7 DM |
5699 | *erp = ext4_ext_dirty(handle, inode1, path1 + |
5700 | path1->p_depth); | |
5701 | /* | |
5702 | * Looks scarry ah..? second inode already points to new blocks, | |
5703 | * and it was successfully dirtied. But luckily error may happen | |
5704 | * only due to journal error, so full transaction will be | |
5705 | * aborted anyway. | |
5706 | */ | |
19008f6d TT |
5707 | if (unlikely(*erp)) |
5708 | goto finish; | |
fcf6b1b7 DM |
5709 | lblk1 += len; |
5710 | lblk2 += len; | |
5711 | replaced_count += len; | |
5712 | count -= len; | |
5713 | ||
5714 | repeat: | |
5715 | if (path1) { | |
5716 | ext4_ext_drop_refs(path1); | |
5717 | kfree(path1); | |
5718 | path1 = NULL; | |
5719 | } | |
5720 | if (path2) { | |
5721 | ext4_ext_drop_refs(path2); | |
5722 | kfree(path2); | |
5723 | path2 = NULL; | |
5724 | } | |
5725 | } | |
fcf6b1b7 DM |
5726 | return replaced_count; |
5727 | } |