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 | ||
9aa5d32b | 76 | if (!ext4_has_metadata_csum(inode->i_sb)) |
7ac5990d DW |
77 | return 1; |
78 | ||
79 | et = find_ext4_extent_tail(eh); | |
80 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
81 | return 0; | |
82 | return 1; | |
83 | } | |
84 | ||
85 | static void ext4_extent_block_csum_set(struct inode *inode, | |
86 | struct ext4_extent_header *eh) | |
87 | { | |
88 | struct ext4_extent_tail *et; | |
89 | ||
9aa5d32b | 90 | if (!ext4_has_metadata_csum(inode->i_sb)) |
7ac5990d DW |
91 | return; |
92 | ||
93 | et = find_ext4_extent_tail(eh); | |
94 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
95 | } | |
96 | ||
d583fb87 AH |
97 | static int ext4_split_extent(handle_t *handle, |
98 | struct inode *inode, | |
dfe50809 | 99 | struct ext4_ext_path **ppath, |
d583fb87 AH |
100 | struct ext4_map_blocks *map, |
101 | int split_flag, | |
102 | int flags); | |
103 | ||
5f95d21f LC |
104 | static int ext4_split_extent_at(handle_t *handle, |
105 | struct inode *inode, | |
dfe50809 | 106 | struct ext4_ext_path **ppath, |
5f95d21f LC |
107 | ext4_lblk_t split, |
108 | int split_flag, | |
109 | int flags); | |
110 | ||
91dd8c11 | 111 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 112 | struct extent_status *newes); |
91dd8c11 | 113 | |
487caeef JK |
114 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
115 | struct inode *inode, | |
116 | int needed) | |
a86c6181 AT |
117 | { |
118 | int err; | |
119 | ||
0390131b FM |
120 | if (!ext4_handle_valid(handle)) |
121 | return 0; | |
a86c6181 | 122 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
123 | return 0; |
124 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 125 | if (err <= 0) |
9102e4fa | 126 | return err; |
487caeef | 127 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
128 | if (err == 0) |
129 | err = -EAGAIN; | |
487caeef JK |
130 | |
131 | return err; | |
a86c6181 AT |
132 | } |
133 | ||
134 | /* | |
135 | * could return: | |
136 | * - EROFS | |
137 | * - ENOMEM | |
138 | */ | |
139 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
140 | struct ext4_ext_path *path) | |
141 | { | |
142 | if (path->p_bh) { | |
143 | /* path points to block */ | |
5d601255 | 144 | BUFFER_TRACE(path->p_bh, "get_write_access"); |
a86c6181 AT |
145 | return ext4_journal_get_write_access(handle, path->p_bh); |
146 | } | |
147 | /* path points to leaf/index in inode body */ | |
148 | /* we use in-core data, no need to protect them */ | |
149 | return 0; | |
150 | } | |
151 | ||
152 | /* | |
153 | * could return: | |
154 | * - EROFS | |
155 | * - ENOMEM | |
156 | * - EIO | |
157 | */ | |
2656497b DW |
158 | int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle, |
159 | struct inode *inode, struct ext4_ext_path *path) | |
a86c6181 AT |
160 | { |
161 | int err; | |
4b1f1660 DM |
162 | |
163 | WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
a86c6181 | 164 | if (path->p_bh) { |
7ac5990d | 165 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 166 | /* path points to block */ |
9ea7a0df TT |
167 | err = __ext4_handle_dirty_metadata(where, line, handle, |
168 | inode, path->p_bh); | |
a86c6181 AT |
169 | } else { |
170 | /* path points to leaf/index in inode body */ | |
171 | err = ext4_mark_inode_dirty(handle, inode); | |
172 | } | |
173 | return err; | |
174 | } | |
175 | ||
f65e6fba | 176 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 177 | struct ext4_ext_path *path, |
725d26d3 | 178 | ext4_lblk_t block) |
a86c6181 | 179 | { |
a86c6181 | 180 | if (path) { |
81fdbb4a | 181 | int depth = path->p_depth; |
a86c6181 | 182 | struct ext4_extent *ex; |
a86c6181 | 183 | |
ad4fb9ca KM |
184 | /* |
185 | * Try to predict block placement assuming that we are | |
186 | * filling in a file which will eventually be | |
187 | * non-sparse --- i.e., in the case of libbfd writing | |
188 | * an ELF object sections out-of-order but in a way | |
189 | * the eventually results in a contiguous object or | |
190 | * executable file, or some database extending a table | |
191 | * space file. However, this is actually somewhat | |
192 | * non-ideal if we are writing a sparse file such as | |
193 | * qemu or KVM writing a raw image file that is going | |
194 | * to stay fairly sparse, since it will end up | |
195 | * fragmenting the file system's free space. Maybe we | |
196 | * should have some hueristics or some way to allow | |
197 | * userspace to pass a hint to file system, | |
b8d6568a | 198 | * especially if the latter case turns out to be |
ad4fb9ca KM |
199 | * common. |
200 | */ | |
7e028976 | 201 | ex = path[depth].p_ext; |
ad4fb9ca KM |
202 | if (ex) { |
203 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
204 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
205 | ||
206 | if (block > ext_block) | |
207 | return ext_pblk + (block - ext_block); | |
208 | else | |
209 | return ext_pblk - (ext_block - block); | |
210 | } | |
a86c6181 | 211 | |
d0d856e8 RD |
212 | /* it looks like index is empty; |
213 | * try to find starting block from index itself */ | |
a86c6181 AT |
214 | if (path[depth].p_bh) |
215 | return path[depth].p_bh->b_blocknr; | |
216 | } | |
217 | ||
218 | /* OK. use inode's group */ | |
f86186b4 | 219 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
220 | } |
221 | ||
654b4908 AK |
222 | /* |
223 | * Allocation for a meta data block | |
224 | */ | |
f65e6fba | 225 | static ext4_fsblk_t |
654b4908 | 226 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 227 | struct ext4_ext_path *path, |
55f020db | 228 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 229 | { |
f65e6fba | 230 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
231 | |
232 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
233 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
234 | NULL, err); | |
a86c6181 AT |
235 | return newblock; |
236 | } | |
237 | ||
55ad63bf | 238 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
239 | { |
240 | int size; | |
241 | ||
242 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
243 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 244 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
245 | if (!check && size > 6) |
246 | size = 6; | |
a86c6181 AT |
247 | #endif |
248 | return size; | |
249 | } | |
250 | ||
55ad63bf | 251 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
252 | { |
253 | int size; | |
254 | ||
255 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
256 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 257 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
258 | if (!check && size > 5) |
259 | size = 5; | |
a86c6181 AT |
260 | #endif |
261 | return size; | |
262 | } | |
263 | ||
55ad63bf | 264 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
265 | { |
266 | int size; | |
267 | ||
268 | size = sizeof(EXT4_I(inode)->i_data); | |
269 | size -= sizeof(struct ext4_extent_header); | |
270 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 271 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
272 | if (!check && size > 3) |
273 | size = 3; | |
a86c6181 AT |
274 | #endif |
275 | return size; | |
276 | } | |
277 | ||
55ad63bf | 278 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
279 | { |
280 | int size; | |
281 | ||
282 | size = sizeof(EXT4_I(inode)->i_data); | |
283 | size -= sizeof(struct ext4_extent_header); | |
284 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 285 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
286 | if (!check && size > 4) |
287 | size = 4; | |
a86c6181 AT |
288 | #endif |
289 | return size; | |
290 | } | |
291 | ||
fcf6b1b7 DM |
292 | static inline int |
293 | ext4_force_split_extent_at(handle_t *handle, struct inode *inode, | |
dfe50809 | 294 | struct ext4_ext_path **ppath, ext4_lblk_t lblk, |
fcf6b1b7 DM |
295 | int nofail) |
296 | { | |
dfe50809 | 297 | struct ext4_ext_path *path = *ppath; |
fcf6b1b7 DM |
298 | int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext); |
299 | ||
dfe50809 | 300 | return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ? |
fcf6b1b7 DM |
301 | EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0, |
302 | EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO | | |
303 | (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0)); | |
304 | } | |
305 | ||
d2a17637 MC |
306 | /* |
307 | * Calculate the number of metadata blocks needed | |
308 | * to allocate @blocks | |
309 | * Worse case is one block per extent | |
310 | */ | |
01f49d0b | 311 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 312 | { |
9d0be502 | 313 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 314 | int idxs; |
d2a17637 | 315 | |
9d0be502 TT |
316 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
317 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
318 | |
319 | /* | |
9d0be502 TT |
320 | * If the new delayed allocation block is contiguous with the |
321 | * previous da block, it can share index blocks with the | |
322 | * previous block, so we only need to allocate a new index | |
323 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
324 | * an additional index block, and at ldxs**3 blocks, yet | |
325 | * another index blocks. | |
d2a17637 | 326 | */ |
9d0be502 TT |
327 | if (ei->i_da_metadata_calc_len && |
328 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
329 | int num = 0; |
330 | ||
9d0be502 TT |
331 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
332 | num++; | |
333 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
334 | num++; | |
335 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
336 | num++; | |
337 | ei->i_da_metadata_calc_len = 0; | |
338 | } else | |
339 | ei->i_da_metadata_calc_len++; | |
340 | ei->i_da_metadata_calc_last_lblock++; | |
341 | return num; | |
342 | } | |
d2a17637 | 343 | |
9d0be502 TT |
344 | /* |
345 | * In the worst case we need a new set of index blocks at | |
346 | * every level of the inode's extent tree. | |
347 | */ | |
348 | ei->i_da_metadata_calc_len = 1; | |
349 | ei->i_da_metadata_calc_last_lblock = lblock; | |
350 | return ext_depth(inode) + 1; | |
d2a17637 MC |
351 | } |
352 | ||
c29c0ae7 AT |
353 | static int |
354 | ext4_ext_max_entries(struct inode *inode, int depth) | |
355 | { | |
356 | int max; | |
357 | ||
358 | if (depth == ext_depth(inode)) { | |
359 | if (depth == 0) | |
55ad63bf | 360 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 361 | else |
55ad63bf | 362 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
363 | } else { |
364 | if (depth == 0) | |
55ad63bf | 365 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 366 | else |
55ad63bf | 367 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
368 | } |
369 | ||
370 | return max; | |
371 | } | |
372 | ||
56b19868 AK |
373 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
374 | { | |
bf89d16f | 375 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 376 | int len = ext4_ext_get_actual_len(ext); |
5946d089 EG |
377 | ext4_lblk_t lblock = le32_to_cpu(ext->ee_block); |
378 | ext4_lblk_t last = lblock + len - 1; | |
e84a26ce | 379 | |
5946d089 | 380 | if (lblock > last) |
31d4f3a2 | 381 | return 0; |
6fd058f7 | 382 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
383 | } |
384 | ||
385 | static int ext4_valid_extent_idx(struct inode *inode, | |
386 | struct ext4_extent_idx *ext_idx) | |
387 | { | |
bf89d16f | 388 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 389 | |
6fd058f7 | 390 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
391 | } |
392 | ||
393 | static int ext4_valid_extent_entries(struct inode *inode, | |
394 | struct ext4_extent_header *eh, | |
395 | int depth) | |
396 | { | |
56b19868 AK |
397 | unsigned short entries; |
398 | if (eh->eh_entries == 0) | |
399 | return 1; | |
400 | ||
401 | entries = le16_to_cpu(eh->eh_entries); | |
402 | ||
403 | if (depth == 0) { | |
404 | /* leaf entries */ | |
81fdbb4a | 405 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
5946d089 EG |
406 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
407 | ext4_fsblk_t pblock = 0; | |
408 | ext4_lblk_t lblock = 0; | |
409 | ext4_lblk_t prev = 0; | |
410 | int len = 0; | |
56b19868 AK |
411 | while (entries) { |
412 | if (!ext4_valid_extent(inode, ext)) | |
413 | return 0; | |
5946d089 EG |
414 | |
415 | /* Check for overlapping extents */ | |
416 | lblock = le32_to_cpu(ext->ee_block); | |
417 | len = ext4_ext_get_actual_len(ext); | |
418 | if ((lblock <= prev) && prev) { | |
419 | pblock = ext4_ext_pblock(ext); | |
420 | es->s_last_error_block = cpu_to_le64(pblock); | |
421 | return 0; | |
422 | } | |
56b19868 AK |
423 | ext++; |
424 | entries--; | |
5946d089 | 425 | prev = lblock + len - 1; |
56b19868 AK |
426 | } |
427 | } else { | |
81fdbb4a | 428 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
429 | while (entries) { |
430 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
431 | return 0; | |
432 | ext_idx++; | |
433 | entries--; | |
434 | } | |
435 | } | |
436 | return 1; | |
437 | } | |
438 | ||
c398eda0 TT |
439 | static int __ext4_ext_check(const char *function, unsigned int line, |
440 | struct inode *inode, struct ext4_extent_header *eh, | |
c349179b | 441 | int depth, ext4_fsblk_t pblk) |
c29c0ae7 AT |
442 | { |
443 | const char *error_msg; | |
444 | int max = 0; | |
445 | ||
446 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
447 | error_msg = "invalid magic"; | |
448 | goto corrupted; | |
449 | } | |
450 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
451 | error_msg = "unexpected eh_depth"; | |
452 | goto corrupted; | |
453 | } | |
454 | if (unlikely(eh->eh_max == 0)) { | |
455 | error_msg = "invalid eh_max"; | |
456 | goto corrupted; | |
457 | } | |
458 | max = ext4_ext_max_entries(inode, depth); | |
459 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
460 | error_msg = "too large eh_max"; | |
461 | goto corrupted; | |
462 | } | |
463 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
464 | error_msg = "invalid eh_entries"; | |
465 | goto corrupted; | |
466 | } | |
56b19868 AK |
467 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
468 | error_msg = "invalid extent entries"; | |
469 | goto corrupted; | |
470 | } | |
7ac5990d DW |
471 | /* Verify checksum on non-root extent tree nodes */ |
472 | if (ext_depth(inode) != depth && | |
473 | !ext4_extent_block_csum_verify(inode, eh)) { | |
474 | error_msg = "extent tree corrupted"; | |
475 | goto corrupted; | |
476 | } | |
c29c0ae7 AT |
477 | return 0; |
478 | ||
479 | corrupted: | |
c398eda0 | 480 | ext4_error_inode(inode, function, line, 0, |
c349179b TT |
481 | "pblk %llu bad header/extent: %s - magic %x, " |
482 | "entries %u, max %u(%u), depth %u(%u)", | |
483 | (unsigned long long) pblk, error_msg, | |
484 | le16_to_cpu(eh->eh_magic), | |
485 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
486 | max, le16_to_cpu(eh->eh_depth), depth); | |
c29c0ae7 AT |
487 | return -EIO; |
488 | } | |
489 | ||
c349179b TT |
490 | #define ext4_ext_check(inode, eh, depth, pblk) \ |
491 | __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk)) | |
c29c0ae7 | 492 | |
7a262f7c AK |
493 | int ext4_ext_check_inode(struct inode *inode) |
494 | { | |
c349179b | 495 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
7a262f7c AK |
496 | } |
497 | ||
7d7ea89e TT |
498 | static struct buffer_head * |
499 | __read_extent_tree_block(const char *function, unsigned int line, | |
107a7bd3 TT |
500 | struct inode *inode, ext4_fsblk_t pblk, int depth, |
501 | int flags) | |
f8489128 | 502 | { |
7d7ea89e TT |
503 | struct buffer_head *bh; |
504 | int err; | |
505 | ||
506 | bh = sb_getblk(inode->i_sb, pblk); | |
507 | if (unlikely(!bh)) | |
508 | return ERR_PTR(-ENOMEM); | |
f8489128 | 509 | |
7d7ea89e TT |
510 | if (!bh_uptodate_or_lock(bh)) { |
511 | trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); | |
512 | err = bh_submit_read(bh); | |
513 | if (err < 0) | |
514 | goto errout; | |
515 | } | |
7869a4a6 | 516 | if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE)) |
7d7ea89e TT |
517 | return bh; |
518 | err = __ext4_ext_check(function, line, inode, | |
c349179b | 519 | ext_block_hdr(bh), depth, pblk); |
7d7ea89e TT |
520 | if (err) |
521 | goto errout; | |
f8489128 | 522 | set_buffer_verified(bh); |
107a7bd3 TT |
523 | /* |
524 | * If this is a leaf block, cache all of its entries | |
525 | */ | |
526 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { | |
527 | struct ext4_extent_header *eh = ext_block_hdr(bh); | |
528 | struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); | |
529 | ext4_lblk_t prev = 0; | |
530 | int i; | |
531 | ||
532 | for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { | |
533 | unsigned int status = EXTENT_STATUS_WRITTEN; | |
534 | ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); | |
535 | int len = ext4_ext_get_actual_len(ex); | |
536 | ||
537 | if (prev && (prev != lblk)) | |
538 | ext4_es_cache_extent(inode, prev, | |
539 | lblk - prev, ~0, | |
540 | EXTENT_STATUS_HOLE); | |
541 | ||
556615dc | 542 | if (ext4_ext_is_unwritten(ex)) |
107a7bd3 TT |
543 | status = EXTENT_STATUS_UNWRITTEN; |
544 | ext4_es_cache_extent(inode, lblk, len, | |
545 | ext4_ext_pblock(ex), status); | |
546 | prev = lblk + len; | |
547 | } | |
548 | } | |
7d7ea89e TT |
549 | return bh; |
550 | errout: | |
551 | put_bh(bh); | |
552 | return ERR_PTR(err); | |
553 | ||
f8489128 DW |
554 | } |
555 | ||
107a7bd3 TT |
556 | #define read_extent_tree_block(inode, pblk, depth, flags) \ |
557 | __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \ | |
558 | (depth), (flags)) | |
f8489128 | 559 | |
7869a4a6 TT |
560 | /* |
561 | * This function is called to cache a file's extent information in the | |
562 | * extent status tree | |
563 | */ | |
564 | int ext4_ext_precache(struct inode *inode) | |
565 | { | |
566 | struct ext4_inode_info *ei = EXT4_I(inode); | |
567 | struct ext4_ext_path *path = NULL; | |
568 | struct buffer_head *bh; | |
569 | int i = 0, depth, ret = 0; | |
570 | ||
571 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
572 | return 0; /* not an extent-mapped inode */ | |
573 | ||
574 | down_read(&ei->i_data_sem); | |
575 | depth = ext_depth(inode); | |
576 | ||
577 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
578 | GFP_NOFS); | |
579 | if (path == NULL) { | |
580 | up_read(&ei->i_data_sem); | |
581 | return -ENOMEM; | |
582 | } | |
583 | ||
584 | /* Don't cache anything if there are no external extent blocks */ | |
585 | if (depth == 0) | |
586 | goto out; | |
587 | path[0].p_hdr = ext_inode_hdr(inode); | |
588 | ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0); | |
589 | if (ret) | |
590 | goto out; | |
591 | path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr); | |
592 | while (i >= 0) { | |
593 | /* | |
594 | * If this is a leaf block or we've reached the end of | |
595 | * the index block, go up | |
596 | */ | |
597 | if ((i == depth) || | |
598 | path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) { | |
599 | brelse(path[i].p_bh); | |
600 | path[i].p_bh = NULL; | |
601 | i--; | |
602 | continue; | |
603 | } | |
604 | bh = read_extent_tree_block(inode, | |
605 | ext4_idx_pblock(path[i].p_idx++), | |
606 | depth - i - 1, | |
607 | EXT4_EX_FORCE_CACHE); | |
608 | if (IS_ERR(bh)) { | |
609 | ret = PTR_ERR(bh); | |
610 | break; | |
611 | } | |
612 | i++; | |
613 | path[i].p_bh = bh; | |
614 | path[i].p_hdr = ext_block_hdr(bh); | |
615 | path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr); | |
616 | } | |
617 | ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED); | |
618 | out: | |
619 | up_read(&ei->i_data_sem); | |
620 | ext4_ext_drop_refs(path); | |
621 | kfree(path); | |
622 | return ret; | |
623 | } | |
624 | ||
a86c6181 AT |
625 | #ifdef EXT_DEBUG |
626 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
627 | { | |
628 | int k, l = path->p_depth; | |
629 | ||
630 | ext_debug("path:"); | |
631 | for (k = 0; k <= l; k++, path++) { | |
632 | if (path->p_idx) { | |
2ae02107 | 633 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 634 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 635 | } else if (path->p_ext) { |
553f9008 | 636 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 637 | le32_to_cpu(path->p_ext->ee_block), |
556615dc | 638 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 639 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 640 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
641 | } else |
642 | ext_debug(" []"); | |
643 | } | |
644 | ext_debug("\n"); | |
645 | } | |
646 | ||
647 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
648 | { | |
649 | int depth = ext_depth(inode); | |
650 | struct ext4_extent_header *eh; | |
651 | struct ext4_extent *ex; | |
652 | int i; | |
653 | ||
654 | if (!path) | |
655 | return; | |
656 | ||
657 | eh = path[depth].p_hdr; | |
658 | ex = EXT_FIRST_EXTENT(eh); | |
659 | ||
553f9008 M |
660 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
661 | ||
a86c6181 | 662 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 | 663 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
556615dc | 664 | ext4_ext_is_unwritten(ex), |
bf89d16f | 665 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
666 | } |
667 | ext_debug("\n"); | |
668 | } | |
1b16da77 YY |
669 | |
670 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
671 | ext4_fsblk_t newblock, int level) | |
672 | { | |
673 | int depth = ext_depth(inode); | |
674 | struct ext4_extent *ex; | |
675 | ||
676 | if (depth != level) { | |
677 | struct ext4_extent_idx *idx; | |
678 | idx = path[level].p_idx; | |
679 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
680 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
681 | le32_to_cpu(idx->ei_block), | |
682 | ext4_idx_pblock(idx), | |
683 | newblock); | |
684 | idx++; | |
685 | } | |
686 | ||
687 | return; | |
688 | } | |
689 | ||
690 | ex = path[depth].p_ext; | |
691 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
692 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
693 | le32_to_cpu(ex->ee_block), | |
694 | ext4_ext_pblock(ex), | |
556615dc | 695 | ext4_ext_is_unwritten(ex), |
1b16da77 YY |
696 | ext4_ext_get_actual_len(ex), |
697 | newblock); | |
698 | ex++; | |
699 | } | |
700 | } | |
701 | ||
a86c6181 | 702 | #else |
af5bc92d TT |
703 | #define ext4_ext_show_path(inode, path) |
704 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 705 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
706 | #endif |
707 | ||
b35905c1 | 708 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 | 709 | { |
b7ea89ad | 710 | int depth, i; |
a86c6181 | 711 | |
b7ea89ad TT |
712 | if (!path) |
713 | return; | |
714 | depth = path->p_depth; | |
a86c6181 AT |
715 | for (i = 0; i <= depth; i++, path++) |
716 | if (path->p_bh) { | |
717 | brelse(path->p_bh); | |
718 | path->p_bh = NULL; | |
719 | } | |
720 | } | |
721 | ||
722 | /* | |
d0d856e8 RD |
723 | * ext4_ext_binsearch_idx: |
724 | * binary search for the closest index of the given block | |
c29c0ae7 | 725 | * the header must be checked before calling this |
a86c6181 AT |
726 | */ |
727 | static void | |
725d26d3 AK |
728 | ext4_ext_binsearch_idx(struct inode *inode, |
729 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
730 | { |
731 | struct ext4_extent_header *eh = path->p_hdr; | |
732 | struct ext4_extent_idx *r, *l, *m; | |
733 | ||
a86c6181 | 734 | |
bba90743 | 735 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
736 | |
737 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 738 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
739 | while (l <= r) { |
740 | m = l + (r - l) / 2; | |
741 | if (block < le32_to_cpu(m->ei_block)) | |
742 | r = m - 1; | |
743 | else | |
744 | l = m + 1; | |
26d535ed DM |
745 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
746 | m, le32_to_cpu(m->ei_block), | |
747 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
748 | } |
749 | ||
750 | path->p_idx = l - 1; | |
4a3c3a51 | 751 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 752 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
753 | |
754 | #ifdef CHECK_BINSEARCH | |
755 | { | |
756 | struct ext4_extent_idx *chix, *ix; | |
757 | int k; | |
758 | ||
759 | chix = ix = EXT_FIRST_INDEX(eh); | |
760 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
761 | if (k != 0 && | |
762 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
763 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
764 | "first=0x%p\n", k, | |
765 | ix, EXT_FIRST_INDEX(eh)); | |
766 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
767 | le32_to_cpu(ix->ei_block), |
768 | le32_to_cpu(ix[-1].ei_block)); | |
769 | } | |
770 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 771 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
772 | if (block < le32_to_cpu(ix->ei_block)) |
773 | break; | |
774 | chix = ix; | |
775 | } | |
776 | BUG_ON(chix != path->p_idx); | |
777 | } | |
778 | #endif | |
779 | ||
780 | } | |
781 | ||
782 | /* | |
d0d856e8 RD |
783 | * ext4_ext_binsearch: |
784 | * binary search for closest extent of the given block | |
c29c0ae7 | 785 | * the header must be checked before calling this |
a86c6181 AT |
786 | */ |
787 | static void | |
725d26d3 AK |
788 | ext4_ext_binsearch(struct inode *inode, |
789 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
790 | { |
791 | struct ext4_extent_header *eh = path->p_hdr; | |
792 | struct ext4_extent *r, *l, *m; | |
793 | ||
a86c6181 AT |
794 | if (eh->eh_entries == 0) { |
795 | /* | |
d0d856e8 RD |
796 | * this leaf is empty: |
797 | * we get such a leaf in split/add case | |
a86c6181 AT |
798 | */ |
799 | return; | |
800 | } | |
801 | ||
bba90743 | 802 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
803 | |
804 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 805 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
806 | |
807 | while (l <= r) { | |
808 | m = l + (r - l) / 2; | |
809 | if (block < le32_to_cpu(m->ee_block)) | |
810 | r = m - 1; | |
811 | else | |
812 | l = m + 1; | |
26d535ed DM |
813 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
814 | m, le32_to_cpu(m->ee_block), | |
815 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
816 | } |
817 | ||
818 | path->p_ext = l - 1; | |
553f9008 | 819 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 820 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 821 | ext4_ext_pblock(path->p_ext), |
556615dc | 822 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 823 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
824 | |
825 | #ifdef CHECK_BINSEARCH | |
826 | { | |
827 | struct ext4_extent *chex, *ex; | |
828 | int k; | |
829 | ||
830 | chex = ex = EXT_FIRST_EXTENT(eh); | |
831 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
832 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 833 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
834 | if (block < le32_to_cpu(ex->ee_block)) |
835 | break; | |
836 | chex = ex; | |
837 | } | |
838 | BUG_ON(chex != path->p_ext); | |
839 | } | |
840 | #endif | |
841 | ||
842 | } | |
843 | ||
844 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
845 | { | |
846 | struct ext4_extent_header *eh; | |
847 | ||
848 | eh = ext_inode_hdr(inode); | |
849 | eh->eh_depth = 0; | |
850 | eh->eh_entries = 0; | |
851 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 852 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 853 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
854 | return 0; |
855 | } | |
856 | ||
857 | struct ext4_ext_path * | |
ed8a1a76 TT |
858 | ext4_find_extent(struct inode *inode, ext4_lblk_t block, |
859 | struct ext4_ext_path **orig_path, int flags) | |
a86c6181 AT |
860 | { |
861 | struct ext4_extent_header *eh; | |
862 | struct buffer_head *bh; | |
705912ca TT |
863 | struct ext4_ext_path *path = orig_path ? *orig_path : NULL; |
864 | short int depth, i, ppos = 0; | |
860d21e2 | 865 | int ret; |
a86c6181 AT |
866 | |
867 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 868 | depth = ext_depth(inode); |
a86c6181 | 869 | |
10809df8 | 870 | if (path) { |
523f431c | 871 | ext4_ext_drop_refs(path); |
10809df8 TT |
872 | if (depth > path[0].p_maxdepth) { |
873 | kfree(path); | |
874 | *orig_path = path = NULL; | |
875 | } | |
876 | } | |
877 | if (!path) { | |
523f431c | 878 | /* account possible depth increase */ |
5d4958f9 | 879 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 | 880 | GFP_NOFS); |
19008f6d | 881 | if (unlikely(!path)) |
a86c6181 | 882 | return ERR_PTR(-ENOMEM); |
10809df8 | 883 | path[0].p_maxdepth = depth + 1; |
a86c6181 | 884 | } |
a86c6181 | 885 | path[0].p_hdr = eh; |
1973adcb | 886 | path[0].p_bh = NULL; |
a86c6181 | 887 | |
c29c0ae7 | 888 | i = depth; |
a86c6181 AT |
889 | /* walk through the tree */ |
890 | while (i) { | |
891 | ext_debug("depth %d: num %d, max %d\n", | |
892 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 893 | |
a86c6181 | 894 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 895 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
896 | path[ppos].p_depth = i; |
897 | path[ppos].p_ext = NULL; | |
898 | ||
107a7bd3 TT |
899 | bh = read_extent_tree_block(inode, path[ppos].p_block, --i, |
900 | flags); | |
19008f6d | 901 | if (unlikely(IS_ERR(bh))) { |
7d7ea89e | 902 | ret = PTR_ERR(bh); |
a86c6181 | 903 | goto err; |
860d21e2 | 904 | } |
7d7ea89e | 905 | |
a86c6181 AT |
906 | eh = ext_block_hdr(bh); |
907 | ppos++; | |
273df556 FM |
908 | if (unlikely(ppos > depth)) { |
909 | put_bh(bh); | |
910 | EXT4_ERROR_INODE(inode, | |
911 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 912 | ret = -EIO; |
273df556 FM |
913 | goto err; |
914 | } | |
a86c6181 AT |
915 | path[ppos].p_bh = bh; |
916 | path[ppos].p_hdr = eh; | |
a86c6181 AT |
917 | } |
918 | ||
919 | path[ppos].p_depth = i; | |
a86c6181 AT |
920 | path[ppos].p_ext = NULL; |
921 | path[ppos].p_idx = NULL; | |
922 | ||
a86c6181 AT |
923 | /* find extent */ |
924 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
925 | /* if not an empty leaf */ |
926 | if (path[ppos].p_ext) | |
bf89d16f | 927 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
928 | |
929 | ext4_ext_show_path(inode, path); | |
930 | ||
931 | return path; | |
932 | ||
933 | err: | |
934 | ext4_ext_drop_refs(path); | |
dfe50809 TT |
935 | kfree(path); |
936 | if (orig_path) | |
937 | *orig_path = NULL; | |
860d21e2 | 938 | return ERR_PTR(ret); |
a86c6181 AT |
939 | } |
940 | ||
941 | /* | |
d0d856e8 RD |
942 | * ext4_ext_insert_index: |
943 | * insert new index [@logical;@ptr] into the block at @curp; | |
944 | * check where to insert: before @curp or after @curp | |
a86c6181 | 945 | */ |
1f109d5a TT |
946 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
947 | struct ext4_ext_path *curp, | |
948 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
949 | { |
950 | struct ext4_extent_idx *ix; | |
951 | int len, err; | |
952 | ||
7e028976 AM |
953 | err = ext4_ext_get_access(handle, inode, curp); |
954 | if (err) | |
a86c6181 AT |
955 | return err; |
956 | ||
273df556 FM |
957 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
958 | EXT4_ERROR_INODE(inode, | |
959 | "logical %d == ei_block %d!", | |
960 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
961 | return -EIO; | |
962 | } | |
d4620315 RD |
963 | |
964 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
965 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
966 | EXT4_ERROR_INODE(inode, | |
967 | "eh_entries %d >= eh_max %d!", | |
968 | le16_to_cpu(curp->p_hdr->eh_entries), | |
969 | le16_to_cpu(curp->p_hdr->eh_max)); | |
970 | return -EIO; | |
971 | } | |
972 | ||
a86c6181 AT |
973 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
974 | /* insert after */ | |
80e675f9 | 975 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
976 | ix = curp->p_idx + 1; |
977 | } else { | |
978 | /* insert before */ | |
80e675f9 | 979 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
980 | ix = curp->p_idx; |
981 | } | |
982 | ||
80e675f9 EG |
983 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
984 | BUG_ON(len < 0); | |
985 | if (len > 0) { | |
986 | ext_debug("insert new index %d: " | |
987 | "move %d indices from 0x%p to 0x%p\n", | |
988 | logical, len, ix, ix + 1); | |
989 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
990 | } | |
991 | ||
f472e026 TM |
992 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
993 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
994 | return -EIO; | |
995 | } | |
996 | ||
a86c6181 | 997 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 998 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 999 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 1000 | |
273df556 FM |
1001 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
1002 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
1003 | return -EIO; | |
1004 | } | |
a86c6181 AT |
1005 | |
1006 | err = ext4_ext_dirty(handle, inode, curp); | |
1007 | ext4_std_error(inode->i_sb, err); | |
1008 | ||
1009 | return err; | |
1010 | } | |
1011 | ||
1012 | /* | |
d0d856e8 RD |
1013 | * ext4_ext_split: |
1014 | * inserts new subtree into the path, using free index entry | |
1015 | * at depth @at: | |
1016 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
1017 | * - makes decision where to split | |
1018 | * - moves remaining extents and index entries (right to the split point) | |
1019 | * into the newly allocated blocks | |
1020 | * - initializes subtree | |
a86c6181 AT |
1021 | */ |
1022 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
1023 | unsigned int flags, |
1024 | struct ext4_ext_path *path, | |
1025 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
1026 | { |
1027 | struct buffer_head *bh = NULL; | |
1028 | int depth = ext_depth(inode); | |
1029 | struct ext4_extent_header *neh; | |
1030 | struct ext4_extent_idx *fidx; | |
a86c6181 | 1031 | int i = at, k, m, a; |
f65e6fba | 1032 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 1033 | __le32 border; |
f65e6fba | 1034 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
1035 | int err = 0; |
1036 | ||
1037 | /* make decision: where to split? */ | |
d0d856e8 | 1038 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 1039 | |
d0d856e8 | 1040 | /* if current leaf will be split, then we should use |
a86c6181 | 1041 | * border from split point */ |
273df556 FM |
1042 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
1043 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
1044 | return -EIO; | |
1045 | } | |
a86c6181 AT |
1046 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
1047 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 1048 | ext_debug("leaf will be split." |
a86c6181 | 1049 | " next leaf starts at %d\n", |
8c55e204 | 1050 | le32_to_cpu(border)); |
a86c6181 AT |
1051 | } else { |
1052 | border = newext->ee_block; | |
1053 | ext_debug("leaf will be added." | |
1054 | " next leaf starts at %d\n", | |
8c55e204 | 1055 | le32_to_cpu(border)); |
a86c6181 AT |
1056 | } |
1057 | ||
1058 | /* | |
d0d856e8 RD |
1059 | * If error occurs, then we break processing |
1060 | * and mark filesystem read-only. index won't | |
a86c6181 | 1061 | * be inserted and tree will be in consistent |
d0d856e8 | 1062 | * state. Next mount will repair buffers too. |
a86c6181 AT |
1063 | */ |
1064 | ||
1065 | /* | |
d0d856e8 RD |
1066 | * Get array to track all allocated blocks. |
1067 | * We need this to handle errors and free blocks | |
1068 | * upon them. | |
a86c6181 | 1069 | */ |
5d4958f9 | 1070 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
1071 | if (!ablocks) |
1072 | return -ENOMEM; | |
a86c6181 AT |
1073 | |
1074 | /* allocate all needed blocks */ | |
1075 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
1076 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 1077 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 1078 | newext, &err, flags); |
a86c6181 AT |
1079 | if (newblock == 0) |
1080 | goto cleanup; | |
1081 | ablocks[a] = newblock; | |
1082 | } | |
1083 | ||
1084 | /* initialize new leaf */ | |
1085 | newblock = ablocks[--a]; | |
273df556 FM |
1086 | if (unlikely(newblock == 0)) { |
1087 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
1088 | err = -EIO; | |
1089 | goto cleanup; | |
1090 | } | |
a86c6181 | 1091 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1092 | if (unlikely(!bh)) { |
860d21e2 | 1093 | err = -ENOMEM; |
a86c6181 AT |
1094 | goto cleanup; |
1095 | } | |
1096 | lock_buffer(bh); | |
1097 | ||
7e028976 AM |
1098 | err = ext4_journal_get_create_access(handle, bh); |
1099 | if (err) | |
a86c6181 AT |
1100 | goto cleanup; |
1101 | ||
1102 | neh = ext_block_hdr(bh); | |
1103 | neh->eh_entries = 0; | |
55ad63bf | 1104 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1105 | neh->eh_magic = EXT4_EXT_MAGIC; |
1106 | neh->eh_depth = 0; | |
a86c6181 | 1107 | |
d0d856e8 | 1108 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
1109 | if (unlikely(path[depth].p_hdr->eh_entries != |
1110 | path[depth].p_hdr->eh_max)) { | |
1111 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
1112 | path[depth].p_hdr->eh_entries, | |
1113 | path[depth].p_hdr->eh_max); | |
1114 | err = -EIO; | |
1115 | goto cleanup; | |
1116 | } | |
a86c6181 | 1117 | /* start copy from next extent */ |
1b16da77 YY |
1118 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
1119 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 1120 | if (m) { |
1b16da77 YY |
1121 | struct ext4_extent *ex; |
1122 | ex = EXT_FIRST_EXTENT(neh); | |
1123 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 1124 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1125 | } |
1126 | ||
7ac5990d | 1127 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1128 | set_buffer_uptodate(bh); |
1129 | unlock_buffer(bh); | |
1130 | ||
0390131b | 1131 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1132 | if (err) |
a86c6181 AT |
1133 | goto cleanup; |
1134 | brelse(bh); | |
1135 | bh = NULL; | |
1136 | ||
1137 | /* correct old leaf */ | |
1138 | if (m) { | |
7e028976 AM |
1139 | err = ext4_ext_get_access(handle, inode, path + depth); |
1140 | if (err) | |
a86c6181 | 1141 | goto cleanup; |
e8546d06 | 1142 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1143 | err = ext4_ext_dirty(handle, inode, path + depth); |
1144 | if (err) | |
a86c6181 AT |
1145 | goto cleanup; |
1146 | ||
1147 | } | |
1148 | ||
1149 | /* create intermediate indexes */ | |
1150 | k = depth - at - 1; | |
273df556 FM |
1151 | if (unlikely(k < 0)) { |
1152 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1153 | err = -EIO; | |
1154 | goto cleanup; | |
1155 | } | |
a86c6181 AT |
1156 | if (k) |
1157 | ext_debug("create %d intermediate indices\n", k); | |
1158 | /* insert new index into current index block */ | |
1159 | /* current depth stored in i var */ | |
1160 | i = depth - 1; | |
1161 | while (k--) { | |
1162 | oldblock = newblock; | |
1163 | newblock = ablocks[--a]; | |
bba90743 | 1164 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1165 | if (unlikely(!bh)) { |
860d21e2 | 1166 | err = -ENOMEM; |
a86c6181 AT |
1167 | goto cleanup; |
1168 | } | |
1169 | lock_buffer(bh); | |
1170 | ||
7e028976 AM |
1171 | err = ext4_journal_get_create_access(handle, bh); |
1172 | if (err) | |
a86c6181 AT |
1173 | goto cleanup; |
1174 | ||
1175 | neh = ext_block_hdr(bh); | |
1176 | neh->eh_entries = cpu_to_le16(1); | |
1177 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1178 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1179 | neh->eh_depth = cpu_to_le16(depth - i); |
1180 | fidx = EXT_FIRST_INDEX(neh); | |
1181 | fidx->ei_block = border; | |
f65e6fba | 1182 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1183 | |
bba90743 ES |
1184 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1185 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1186 | |
1b16da77 | 1187 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1188 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1189 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1190 | EXT4_ERROR_INODE(inode, | |
1191 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1192 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1193 | err = -EIO; | |
1194 | goto cleanup; | |
1195 | } | |
1b16da77 YY |
1196 | /* start copy indexes */ |
1197 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1198 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1199 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1200 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1201 | if (m) { |
1b16da77 | 1202 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1203 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1204 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1205 | } |
7ac5990d | 1206 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1207 | set_buffer_uptodate(bh); |
1208 | unlock_buffer(bh); | |
1209 | ||
0390131b | 1210 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1211 | if (err) |
a86c6181 AT |
1212 | goto cleanup; |
1213 | brelse(bh); | |
1214 | bh = NULL; | |
1215 | ||
1216 | /* correct old index */ | |
1217 | if (m) { | |
1218 | err = ext4_ext_get_access(handle, inode, path + i); | |
1219 | if (err) | |
1220 | goto cleanup; | |
e8546d06 | 1221 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1222 | err = ext4_ext_dirty(handle, inode, path + i); |
1223 | if (err) | |
1224 | goto cleanup; | |
1225 | } | |
1226 | ||
1227 | i--; | |
1228 | } | |
1229 | ||
1230 | /* insert new index */ | |
a86c6181 AT |
1231 | err = ext4_ext_insert_index(handle, inode, path + at, |
1232 | le32_to_cpu(border), newblock); | |
1233 | ||
1234 | cleanup: | |
1235 | if (bh) { | |
1236 | if (buffer_locked(bh)) | |
1237 | unlock_buffer(bh); | |
1238 | brelse(bh); | |
1239 | } | |
1240 | ||
1241 | if (err) { | |
1242 | /* free all allocated blocks in error case */ | |
1243 | for (i = 0; i < depth; i++) { | |
1244 | if (!ablocks[i]) | |
1245 | continue; | |
7dc57615 | 1246 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1247 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1248 | } |
1249 | } | |
1250 | kfree(ablocks); | |
1251 | ||
1252 | return err; | |
1253 | } | |
1254 | ||
1255 | /* | |
d0d856e8 RD |
1256 | * ext4_ext_grow_indepth: |
1257 | * implements tree growing procedure: | |
1258 | * - allocates new block | |
1259 | * - moves top-level data (index block or leaf) into the new block | |
1260 | * - initializes new top-level, creating index that points to the | |
1261 | * just created block | |
a86c6181 AT |
1262 | */ |
1263 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
be5cd90d | 1264 | unsigned int flags) |
a86c6181 | 1265 | { |
a86c6181 | 1266 | struct ext4_extent_header *neh; |
a86c6181 | 1267 | struct buffer_head *bh; |
be5cd90d DM |
1268 | ext4_fsblk_t newblock, goal = 0; |
1269 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; | |
a86c6181 AT |
1270 | int err = 0; |
1271 | ||
be5cd90d DM |
1272 | /* Try to prepend new index to old one */ |
1273 | if (ext_depth(inode)) | |
1274 | goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode))); | |
1275 | if (goal > le32_to_cpu(es->s_first_data_block)) { | |
1276 | flags |= EXT4_MB_HINT_TRY_GOAL; | |
1277 | goal--; | |
1278 | } else | |
1279 | goal = ext4_inode_to_goal_block(inode); | |
1280 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, | |
1281 | NULL, &err); | |
a86c6181 AT |
1282 | if (newblock == 0) |
1283 | return err; | |
1284 | ||
1285 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1286 | if (unlikely(!bh)) |
860d21e2 | 1287 | return -ENOMEM; |
a86c6181 AT |
1288 | lock_buffer(bh); |
1289 | ||
7e028976 AM |
1290 | err = ext4_journal_get_create_access(handle, bh); |
1291 | if (err) { | |
a86c6181 AT |
1292 | unlock_buffer(bh); |
1293 | goto out; | |
1294 | } | |
1295 | ||
1296 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1297 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1298 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1299 | |
1300 | /* set size of new block */ | |
1301 | neh = ext_block_hdr(bh); | |
1302 | /* old root could have indexes or leaves | |
1303 | * so calculate e_max right way */ | |
1304 | if (ext_depth(inode)) | |
55ad63bf | 1305 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1306 | else |
55ad63bf | 1307 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1308 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1309 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1310 | set_buffer_uptodate(bh); |
1311 | unlock_buffer(bh); | |
1312 | ||
0390131b | 1313 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1314 | if (err) |
a86c6181 AT |
1315 | goto out; |
1316 | ||
1939dd84 | 1317 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1318 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1319 | neh->eh_entries = cpu_to_le16(1); |
1320 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1321 | if (neh->eh_depth == 0) { | |
1322 | /* Root extent block becomes index block */ | |
1323 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1324 | EXT_FIRST_INDEX(neh)->ei_block = | |
1325 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1326 | } | |
2ae02107 | 1327 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1328 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1329 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1330 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1331 | |
ba39ebb6 | 1332 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1333 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1334 | out: |
1335 | brelse(bh); | |
1336 | ||
1337 | return err; | |
1338 | } | |
1339 | ||
1340 | /* | |
d0d856e8 RD |
1341 | * ext4_ext_create_new_leaf: |
1342 | * finds empty index and adds new leaf. | |
1343 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1344 | */ |
1345 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
107a7bd3 TT |
1346 | unsigned int mb_flags, |
1347 | unsigned int gb_flags, | |
dfe50809 | 1348 | struct ext4_ext_path **ppath, |
55f020db | 1349 | struct ext4_extent *newext) |
a86c6181 | 1350 | { |
dfe50809 | 1351 | struct ext4_ext_path *path = *ppath; |
a86c6181 AT |
1352 | struct ext4_ext_path *curp; |
1353 | int depth, i, err = 0; | |
1354 | ||
1355 | repeat: | |
1356 | i = depth = ext_depth(inode); | |
1357 | ||
1358 | /* walk up to the tree and look for free index entry */ | |
1359 | curp = path + depth; | |
1360 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1361 | i--; | |
1362 | curp--; | |
1363 | } | |
1364 | ||
d0d856e8 RD |
1365 | /* we use already allocated block for index block, |
1366 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1367 | if (EXT_HAS_FREE_INDEX(curp)) { |
1368 | /* if we found index with free entry, then use that | |
1369 | * entry: create all needed subtree and add new leaf */ | |
107a7bd3 | 1370 | err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
787e0981 SF |
1371 | if (err) |
1372 | goto out; | |
a86c6181 AT |
1373 | |
1374 | /* refill path */ | |
ed8a1a76 | 1375 | path = ext4_find_extent(inode, |
725d26d3 | 1376 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
dfe50809 | 1377 | ppath, gb_flags); |
a86c6181 AT |
1378 | if (IS_ERR(path)) |
1379 | err = PTR_ERR(path); | |
1380 | } else { | |
1381 | /* tree is full, time to grow in depth */ | |
be5cd90d | 1382 | err = ext4_ext_grow_indepth(handle, inode, mb_flags); |
a86c6181 AT |
1383 | if (err) |
1384 | goto out; | |
1385 | ||
1386 | /* refill path */ | |
ed8a1a76 | 1387 | path = ext4_find_extent(inode, |
725d26d3 | 1388 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
dfe50809 | 1389 | ppath, gb_flags); |
a86c6181 AT |
1390 | if (IS_ERR(path)) { |
1391 | err = PTR_ERR(path); | |
1392 | goto out; | |
1393 | } | |
1394 | ||
1395 | /* | |
d0d856e8 RD |
1396 | * only first (depth 0 -> 1) produces free space; |
1397 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1398 | */ |
1399 | depth = ext_depth(inode); | |
1400 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1401 | /* now we need to split */ |
a86c6181 AT |
1402 | goto repeat; |
1403 | } | |
1404 | } | |
1405 | ||
1406 | out: | |
1407 | return err; | |
1408 | } | |
1409 | ||
1988b51e AT |
1410 | /* |
1411 | * search the closest allocated block to the left for *logical | |
1412 | * and returns it at @logical + it's physical address at @phys | |
1413 | * if *logical is the smallest allocated block, the function | |
1414 | * returns 0 at @phys | |
1415 | * return value contains 0 (success) or error code | |
1416 | */ | |
1f109d5a TT |
1417 | static int ext4_ext_search_left(struct inode *inode, |
1418 | struct ext4_ext_path *path, | |
1419 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1420 | { |
1421 | struct ext4_extent_idx *ix; | |
1422 | struct ext4_extent *ex; | |
b939e376 | 1423 | int depth, ee_len; |
1988b51e | 1424 | |
273df556 FM |
1425 | if (unlikely(path == NULL)) { |
1426 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1427 | return -EIO; | |
1428 | } | |
1988b51e AT |
1429 | depth = path->p_depth; |
1430 | *phys = 0; | |
1431 | ||
1432 | if (depth == 0 && path->p_ext == NULL) | |
1433 | return 0; | |
1434 | ||
1435 | /* usually extent in the path covers blocks smaller | |
1436 | * then *logical, but it can be that extent is the | |
1437 | * first one in the file */ | |
1438 | ||
1439 | ex = path[depth].p_ext; | |
b939e376 | 1440 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1441 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1442 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1443 | EXT4_ERROR_INODE(inode, | |
1444 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1445 | *logical, le32_to_cpu(ex->ee_block)); | |
1446 | return -EIO; | |
1447 | } | |
1988b51e AT |
1448 | while (--depth >= 0) { |
1449 | ix = path[depth].p_idx; | |
273df556 FM |
1450 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1451 | EXT4_ERROR_INODE(inode, | |
1452 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1453 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1454 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1455 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1456 | depth); |
1457 | return -EIO; | |
1458 | } | |
1988b51e AT |
1459 | } |
1460 | return 0; | |
1461 | } | |
1462 | ||
273df556 FM |
1463 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1464 | EXT4_ERROR_INODE(inode, | |
1465 | "logical %d < ee_block %d + ee_len %d!", | |
1466 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1467 | return -EIO; | |
1468 | } | |
1988b51e | 1469 | |
b939e376 | 1470 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1471 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1472 | return 0; |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * search the closest allocated block to the right for *logical | |
1477 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1478 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1479 | * returns 0 at @phys |
1480 | * return value contains 0 (success) or error code | |
1481 | */ | |
1f109d5a TT |
1482 | static int ext4_ext_search_right(struct inode *inode, |
1483 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1484 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1485 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1486 | { |
1487 | struct buffer_head *bh = NULL; | |
1488 | struct ext4_extent_header *eh; | |
1489 | struct ext4_extent_idx *ix; | |
1490 | struct ext4_extent *ex; | |
1491 | ext4_fsblk_t block; | |
395a87bf ES |
1492 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1493 | int ee_len; | |
1988b51e | 1494 | |
273df556 FM |
1495 | if (unlikely(path == NULL)) { |
1496 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1497 | return -EIO; | |
1498 | } | |
1988b51e AT |
1499 | depth = path->p_depth; |
1500 | *phys = 0; | |
1501 | ||
1502 | if (depth == 0 && path->p_ext == NULL) | |
1503 | return 0; | |
1504 | ||
1505 | /* usually extent in the path covers blocks smaller | |
1506 | * then *logical, but it can be that extent is the | |
1507 | * first one in the file */ | |
1508 | ||
1509 | ex = path[depth].p_ext; | |
b939e376 | 1510 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1511 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1512 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1513 | EXT4_ERROR_INODE(inode, | |
1514 | "first_extent(path[%d].p_hdr) != ex", | |
1515 | depth); | |
1516 | return -EIO; | |
1517 | } | |
1988b51e AT |
1518 | while (--depth >= 0) { |
1519 | ix = path[depth].p_idx; | |
273df556 FM |
1520 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1521 | EXT4_ERROR_INODE(inode, | |
1522 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1523 | *logical); | |
1524 | return -EIO; | |
1525 | } | |
1988b51e | 1526 | } |
4d33b1ef | 1527 | goto found_extent; |
1988b51e AT |
1528 | } |
1529 | ||
273df556 FM |
1530 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1531 | EXT4_ERROR_INODE(inode, | |
1532 | "logical %d < ee_block %d + ee_len %d!", | |
1533 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1534 | return -EIO; | |
1535 | } | |
1988b51e AT |
1536 | |
1537 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1538 | /* next allocated block in this leaf */ | |
1539 | ex++; | |
4d33b1ef | 1540 | goto found_extent; |
1988b51e AT |
1541 | } |
1542 | ||
1543 | /* go up and search for index to the right */ | |
1544 | while (--depth >= 0) { | |
1545 | ix = path[depth].p_idx; | |
1546 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1547 | goto got_index; |
1988b51e AT |
1548 | } |
1549 | ||
25f1ee3a WF |
1550 | /* we've gone up to the root and found no index to the right */ |
1551 | return 0; | |
1988b51e | 1552 | |
25f1ee3a | 1553 | got_index: |
1988b51e AT |
1554 | /* we've found index to the right, let's |
1555 | * follow it and find the closest allocated | |
1556 | * block to the right */ | |
1557 | ix++; | |
bf89d16f | 1558 | block = ext4_idx_pblock(ix); |
1988b51e | 1559 | while (++depth < path->p_depth) { |
395a87bf | 1560 | /* subtract from p_depth to get proper eh_depth */ |
7d7ea89e | 1561 | bh = read_extent_tree_block(inode, block, |
107a7bd3 | 1562 | path->p_depth - depth, 0); |
7d7ea89e TT |
1563 | if (IS_ERR(bh)) |
1564 | return PTR_ERR(bh); | |
1565 | eh = ext_block_hdr(bh); | |
1988b51e | 1566 | ix = EXT_FIRST_INDEX(eh); |
bf89d16f | 1567 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1568 | put_bh(bh); |
1569 | } | |
1570 | ||
107a7bd3 | 1571 | bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0); |
7d7ea89e TT |
1572 | if (IS_ERR(bh)) |
1573 | return PTR_ERR(bh); | |
1988b51e | 1574 | eh = ext_block_hdr(bh); |
1988b51e | 1575 | ex = EXT_FIRST_EXTENT(eh); |
4d33b1ef | 1576 | found_extent: |
1988b51e | 1577 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1578 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1579 | *ret_ex = ex; |
1580 | if (bh) | |
1581 | put_bh(bh); | |
1988b51e | 1582 | return 0; |
1988b51e AT |
1583 | } |
1584 | ||
a86c6181 | 1585 | /* |
d0d856e8 | 1586 | * ext4_ext_next_allocated_block: |
f17722f9 | 1587 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1588 | * NOTE: it considers block number from index entry as |
1589 | * allocated block. Thus, index entries have to be consistent | |
1590 | * with leaves. | |
a86c6181 | 1591 | */ |
fcf6b1b7 | 1592 | ext4_lblk_t |
a86c6181 AT |
1593 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1594 | { | |
1595 | int depth; | |
1596 | ||
1597 | BUG_ON(path == NULL); | |
1598 | depth = path->p_depth; | |
1599 | ||
1600 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1601 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1602 | |
1603 | while (depth >= 0) { | |
1604 | if (depth == path->p_depth) { | |
1605 | /* leaf */ | |
6f8ff537 CW |
1606 | if (path[depth].p_ext && |
1607 | path[depth].p_ext != | |
a86c6181 AT |
1608 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1609 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1610 | } else { | |
1611 | /* index */ | |
1612 | if (path[depth].p_idx != | |
1613 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1614 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1615 | } | |
1616 | depth--; | |
1617 | } | |
1618 | ||
f17722f9 | 1619 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1620 | } |
1621 | ||
1622 | /* | |
d0d856e8 | 1623 | * ext4_ext_next_leaf_block: |
f17722f9 | 1624 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1625 | */ |
5718789d | 1626 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1627 | { |
1628 | int depth; | |
1629 | ||
1630 | BUG_ON(path == NULL); | |
1631 | depth = path->p_depth; | |
1632 | ||
1633 | /* zero-tree has no leaf blocks at all */ | |
1634 | if (depth == 0) | |
f17722f9 | 1635 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1636 | |
1637 | /* go to index block */ | |
1638 | depth--; | |
1639 | ||
1640 | while (depth >= 0) { | |
1641 | if (path[depth].p_idx != | |
1642 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1643 | return (ext4_lblk_t) |
1644 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1645 | depth--; |
1646 | } | |
1647 | ||
f17722f9 | 1648 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1649 | } |
1650 | ||
1651 | /* | |
d0d856e8 RD |
1652 | * ext4_ext_correct_indexes: |
1653 | * if leaf gets modified and modified extent is first in the leaf, | |
1654 | * then we have to correct all indexes above. | |
a86c6181 AT |
1655 | * TODO: do we need to correct tree in all cases? |
1656 | */ | |
1d03ec98 | 1657 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1658 | struct ext4_ext_path *path) |
1659 | { | |
1660 | struct ext4_extent_header *eh; | |
1661 | int depth = ext_depth(inode); | |
1662 | struct ext4_extent *ex; | |
1663 | __le32 border; | |
1664 | int k, err = 0; | |
1665 | ||
1666 | eh = path[depth].p_hdr; | |
1667 | ex = path[depth].p_ext; | |
273df556 FM |
1668 | |
1669 | if (unlikely(ex == NULL || eh == NULL)) { | |
1670 | EXT4_ERROR_INODE(inode, | |
1671 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1672 | return -EIO; | |
1673 | } | |
a86c6181 AT |
1674 | |
1675 | if (depth == 0) { | |
1676 | /* there is no tree at all */ | |
1677 | return 0; | |
1678 | } | |
1679 | ||
1680 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1681 | /* we correct tree if first leaf got modified only */ | |
1682 | return 0; | |
1683 | } | |
1684 | ||
1685 | /* | |
d0d856e8 | 1686 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1687 | */ |
1688 | k = depth - 1; | |
1689 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1690 | err = ext4_ext_get_access(handle, inode, path + k); |
1691 | if (err) | |
a86c6181 AT |
1692 | return err; |
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 | return err; |
1697 | ||
1698 | while (k--) { | |
1699 | /* change all left-side indexes */ | |
1700 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1701 | break; | |
7e028976 AM |
1702 | err = ext4_ext_get_access(handle, inode, path + k); |
1703 | if (err) | |
a86c6181 AT |
1704 | break; |
1705 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1706 | err = ext4_ext_dirty(handle, inode, path + k); |
1707 | if (err) | |
a86c6181 AT |
1708 | break; |
1709 | } | |
1710 | ||
1711 | return err; | |
1712 | } | |
1713 | ||
748de673 | 1714 | int |
a86c6181 AT |
1715 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1716 | struct ext4_extent *ex2) | |
1717 | { | |
da0169b3 | 1718 | unsigned short ext1_ee_len, ext2_ee_len; |
a2df2a63 AA |
1719 | |
1720 | /* | |
ec22ba8e | 1721 | * Make sure that both extents are initialized. We don't merge |
556615dc | 1722 | * unwritten extents so that we can be sure that end_io code has |
ec22ba8e DM |
1723 | * the extent that was written properly split out and conversion to |
1724 | * initialized is trivial. | |
a2df2a63 | 1725 | */ |
556615dc | 1726 | if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2)) |
a2df2a63 AA |
1727 | return 0; |
1728 | ||
1729 | ext1_ee_len = ext4_ext_get_actual_len(ex1); | |
1730 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1731 | ||
1732 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1733 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1734 | return 0; |
1735 | ||
471d4011 SB |
1736 | /* |
1737 | * To allow future support for preallocated extents to be added | |
1738 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1739 | * this can result in the top bit of ee_len being set. |
471d4011 | 1740 | */ |
da0169b3 | 1741 | if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN) |
471d4011 | 1742 | return 0; |
556615dc | 1743 | if (ext4_ext_is_unwritten(ex1) && |
a9b82415 DW |
1744 | (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) || |
1745 | atomic_read(&EXT4_I(inode)->i_unwritten) || | |
556615dc | 1746 | (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN))) |
a9b82415 | 1747 | return 0; |
bbf2f9fb | 1748 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1749 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1750 | return 0; |
1751 | #endif | |
1752 | ||
bf89d16f | 1753 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1754 | return 1; |
1755 | return 0; | |
1756 | } | |
1757 | ||
56055d3a AA |
1758 | /* |
1759 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1760 | * It always tries to merge towards right. If you want to merge towards | |
1761 | * left, pass "ex - 1" as argument instead of "ex". | |
1762 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1763 | * 1 if they got merged. | |
1764 | */ | |
197217a5 | 1765 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1766 | struct ext4_ext_path *path, |
1767 | struct ext4_extent *ex) | |
56055d3a AA |
1768 | { |
1769 | struct ext4_extent_header *eh; | |
1770 | unsigned int depth, len; | |
556615dc | 1771 | int merge_done = 0, unwritten; |
56055d3a AA |
1772 | |
1773 | depth = ext_depth(inode); | |
1774 | BUG_ON(path[depth].p_hdr == NULL); | |
1775 | eh = path[depth].p_hdr; | |
1776 | ||
1777 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1778 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1779 | break; | |
1780 | /* merge with next extent! */ | |
556615dc | 1781 | unwritten = ext4_ext_is_unwritten(ex); |
56055d3a AA |
1782 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
1783 | + ext4_ext_get_actual_len(ex + 1)); | |
556615dc LC |
1784 | if (unwritten) |
1785 | ext4_ext_mark_unwritten(ex); | |
56055d3a AA |
1786 | |
1787 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1788 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1789 | * sizeof(struct ext4_extent); | |
1790 | memmove(ex + 1, ex + 2, len); | |
1791 | } | |
e8546d06 | 1792 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1793 | merge_done = 1; |
1794 | WARN_ON(eh->eh_entries == 0); | |
1795 | if (!eh->eh_entries) | |
24676da4 | 1796 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1797 | } |
1798 | ||
1799 | return merge_done; | |
1800 | } | |
1801 | ||
ecb94f5f TT |
1802 | /* |
1803 | * This function does a very simple check to see if we can collapse | |
1804 | * an extent tree with a single extent tree leaf block into the inode. | |
1805 | */ | |
1806 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1807 | struct inode *inode, | |
1808 | struct ext4_ext_path *path) | |
1809 | { | |
1810 | size_t s; | |
1811 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1812 | ext4_fsblk_t blk; | |
1813 | ||
1814 | if ((path[0].p_depth != 1) || | |
1815 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1816 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1817 | return; | |
1818 | ||
1819 | /* | |
1820 | * We need to modify the block allocation bitmap and the block | |
1821 | * group descriptor to release the extent tree block. If we | |
1822 | * can't get the journal credits, give up. | |
1823 | */ | |
1824 | if (ext4_journal_extend(handle, 2)) | |
1825 | return; | |
1826 | ||
1827 | /* | |
1828 | * Copy the extent data up to the inode | |
1829 | */ | |
1830 | blk = ext4_idx_pblock(path[0].p_idx); | |
1831 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1832 | sizeof(struct ext4_extent_idx); | |
1833 | s += sizeof(struct ext4_extent_header); | |
1834 | ||
10809df8 | 1835 | path[1].p_maxdepth = path[0].p_maxdepth; |
ecb94f5f TT |
1836 | memcpy(path[0].p_hdr, path[1].p_hdr, s); |
1837 | path[0].p_depth = 0; | |
1838 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1839 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1840 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1841 | ||
1842 | brelse(path[1].p_bh); | |
1843 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
71d4f7d0 | 1844 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
ecb94f5f TT |
1845 | } |
1846 | ||
197217a5 YY |
1847 | /* |
1848 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1849 | * return 1 if merge left else 0. | |
1850 | */ | |
ecb94f5f TT |
1851 | static void ext4_ext_try_to_merge(handle_t *handle, |
1852 | struct inode *inode, | |
197217a5 YY |
1853 | struct ext4_ext_path *path, |
1854 | struct ext4_extent *ex) { | |
1855 | struct ext4_extent_header *eh; | |
1856 | unsigned int depth; | |
1857 | int merge_done = 0; | |
197217a5 YY |
1858 | |
1859 | depth = ext_depth(inode); | |
1860 | BUG_ON(path[depth].p_hdr == NULL); | |
1861 | eh = path[depth].p_hdr; | |
1862 | ||
1863 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1864 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1865 | ||
1866 | if (!merge_done) | |
ecb94f5f | 1867 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1868 | |
ecb94f5f | 1869 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1870 | } |
1871 | ||
25d14f98 AA |
1872 | /* |
1873 | * check if a portion of the "newext" extent overlaps with an | |
1874 | * existing extent. | |
1875 | * | |
1876 | * If there is an overlap discovered, it updates the length of the newext | |
1877 | * such that there will be no overlap, and then returns 1. | |
1878 | * If there is no overlap found, it returns 0. | |
1879 | */ | |
4d33b1ef TT |
1880 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1881 | struct inode *inode, | |
1f109d5a TT |
1882 | struct ext4_extent *newext, |
1883 | struct ext4_ext_path *path) | |
25d14f98 | 1884 | { |
725d26d3 | 1885 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1886 | unsigned int depth, len1; |
1887 | unsigned int ret = 0; | |
1888 | ||
1889 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1890 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1891 | depth = ext_depth(inode); |
1892 | if (!path[depth].p_ext) | |
1893 | goto out; | |
f5a44db5 | 1894 | b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block)); |
25d14f98 AA |
1895 | |
1896 | /* | |
1897 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1898 | * is before the requested block(s) |
25d14f98 AA |
1899 | */ |
1900 | if (b2 < b1) { | |
1901 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1902 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1903 | goto out; |
f5a44db5 | 1904 | b2 = EXT4_LBLK_CMASK(sbi, b2); |
25d14f98 AA |
1905 | } |
1906 | ||
725d26d3 | 1907 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1908 | if (b1 + len1 < b1) { |
f17722f9 | 1909 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1910 | newext->ee_len = cpu_to_le16(len1); |
1911 | ret = 1; | |
1912 | } | |
1913 | ||
1914 | /* check for overlap */ | |
1915 | if (b1 + len1 > b2) { | |
1916 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1917 | ret = 1; | |
1918 | } | |
1919 | out: | |
1920 | return ret; | |
1921 | } | |
1922 | ||
a86c6181 | 1923 | /* |
d0d856e8 RD |
1924 | * ext4_ext_insert_extent: |
1925 | * tries to merge requsted extent into the existing extent or | |
1926 | * inserts requested extent as new one into the tree, | |
1927 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1928 | */ |
1929 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
dfe50809 | 1930 | struct ext4_ext_path **ppath, |
107a7bd3 | 1931 | struct ext4_extent *newext, int gb_flags) |
a86c6181 | 1932 | { |
dfe50809 | 1933 | struct ext4_ext_path *path = *ppath; |
af5bc92d | 1934 | struct ext4_extent_header *eh; |
a86c6181 AT |
1935 | struct ext4_extent *ex, *fex; |
1936 | struct ext4_extent *nearex; /* nearest extent */ | |
1937 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1938 | int depth, len, err; |
1939 | ext4_lblk_t next; | |
556615dc | 1940 | int mb_flags = 0, unwritten; |
a86c6181 | 1941 | |
e3cf5d5d TT |
1942 | if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
1943 | mb_flags |= EXT4_MB_DELALLOC_RESERVED; | |
273df556 FM |
1944 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1945 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1946 | return -EIO; | |
1947 | } | |
a86c6181 AT |
1948 | depth = ext_depth(inode); |
1949 | ex = path[depth].p_ext; | |
be8981be | 1950 | eh = path[depth].p_hdr; |
273df556 FM |
1951 | if (unlikely(path[depth].p_hdr == NULL)) { |
1952 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1953 | return -EIO; | |
1954 | } | |
a86c6181 AT |
1955 | |
1956 | /* try to insert block into found extent and return */ | |
107a7bd3 | 1957 | if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1958 | |
1959 | /* | |
be8981be LC |
1960 | * Try to see whether we should rather test the extent on |
1961 | * right from ex, or from the left of ex. This is because | |
ed8a1a76 | 1962 | * ext4_find_extent() can return either extent on the |
be8981be LC |
1963 | * left, or on the right from the searched position. This |
1964 | * will make merging more effective. | |
a2df2a63 | 1965 | */ |
be8981be LC |
1966 | if (ex < EXT_LAST_EXTENT(eh) && |
1967 | (le32_to_cpu(ex->ee_block) + | |
1968 | ext4_ext_get_actual_len(ex) < | |
1969 | le32_to_cpu(newext->ee_block))) { | |
1970 | ex += 1; | |
1971 | goto prepend; | |
1972 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1973 | (le32_to_cpu(newext->ee_block) + | |
1974 | ext4_ext_get_actual_len(newext) < | |
1975 | le32_to_cpu(ex->ee_block))) | |
1976 | ex -= 1; | |
1977 | ||
1978 | /* Try to append newex to the ex */ | |
1979 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1980 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
1981 | "(from %llu)\n", | |
556615dc | 1982 | ext4_ext_is_unwritten(newext), |
be8981be LC |
1983 | ext4_ext_get_actual_len(newext), |
1984 | le32_to_cpu(ex->ee_block), | |
556615dc | 1985 | ext4_ext_is_unwritten(ex), |
be8981be LC |
1986 | ext4_ext_get_actual_len(ex), |
1987 | ext4_ext_pblock(ex)); | |
1988 | err = ext4_ext_get_access(handle, inode, | |
1989 | path + depth); | |
1990 | if (err) | |
1991 | return err; | |
556615dc | 1992 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be | 1993 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
a2df2a63 | 1994 | + ext4_ext_get_actual_len(newext)); |
556615dc LC |
1995 | if (unwritten) |
1996 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
1997 | eh = path[depth].p_hdr; |
1998 | nearex = ex; | |
1999 | goto merge; | |
2000 | } | |
2001 | ||
2002 | prepend: | |
2003 | /* Try to prepend newex to the ex */ | |
2004 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
2005 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
2006 | "(from %llu)\n", | |
2007 | le32_to_cpu(newext->ee_block), | |
556615dc | 2008 | ext4_ext_is_unwritten(newext), |
be8981be LC |
2009 | ext4_ext_get_actual_len(newext), |
2010 | le32_to_cpu(ex->ee_block), | |
556615dc | 2011 | ext4_ext_is_unwritten(ex), |
be8981be LC |
2012 | ext4_ext_get_actual_len(ex), |
2013 | ext4_ext_pblock(ex)); | |
2014 | err = ext4_ext_get_access(handle, inode, | |
2015 | path + depth); | |
2016 | if (err) | |
2017 | return err; | |
2018 | ||
556615dc | 2019 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be LC |
2020 | ex->ee_block = newext->ee_block; |
2021 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
2022 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
2023 | + ext4_ext_get_actual_len(newext)); | |
556615dc LC |
2024 | if (unwritten) |
2025 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
2026 | eh = path[depth].p_hdr; |
2027 | nearex = ex; | |
2028 | goto merge; | |
2029 | } | |
a86c6181 AT |
2030 | } |
2031 | ||
a86c6181 AT |
2032 | depth = ext_depth(inode); |
2033 | eh = path[depth].p_hdr; | |
2034 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
2035 | goto has_space; | |
2036 | ||
2037 | /* probably next leaf has space for us? */ | |
2038 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
2039 | next = EXT_MAX_BLOCKS; |
2040 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 2041 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 2042 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 2043 | ext_debug("next leaf block - %u\n", next); |
a86c6181 | 2044 | BUG_ON(npath != NULL); |
ed8a1a76 | 2045 | npath = ext4_find_extent(inode, next, NULL, 0); |
a86c6181 AT |
2046 | if (IS_ERR(npath)) |
2047 | return PTR_ERR(npath); | |
2048 | BUG_ON(npath->p_depth != path->p_depth); | |
2049 | eh = npath[depth].p_hdr; | |
2050 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 2051 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
2052 | le16_to_cpu(eh->eh_entries)); |
2053 | path = npath; | |
ffb505ff | 2054 | goto has_space; |
a86c6181 AT |
2055 | } |
2056 | ext_debug("next leaf has no free space(%d,%d)\n", | |
2057 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
2058 | } | |
2059 | ||
2060 | /* | |
d0d856e8 RD |
2061 | * There is no free space in the found leaf. |
2062 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 2063 | */ |
107a7bd3 | 2064 | if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
e3cf5d5d | 2065 | mb_flags |= EXT4_MB_USE_RESERVED; |
107a7bd3 | 2066 | err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, |
dfe50809 | 2067 | ppath, newext); |
a86c6181 AT |
2068 | if (err) |
2069 | goto cleanup; | |
2070 | depth = ext_depth(inode); | |
2071 | eh = path[depth].p_hdr; | |
2072 | ||
2073 | has_space: | |
2074 | nearex = path[depth].p_ext; | |
2075 | ||
7e028976 AM |
2076 | err = ext4_ext_get_access(handle, inode, path + depth); |
2077 | if (err) | |
a86c6181 AT |
2078 | goto cleanup; |
2079 | ||
2080 | if (!nearex) { | |
2081 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 2082 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 2083 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2084 | ext4_ext_pblock(newext), |
556615dc | 2085 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2086 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
2087 | nearex = EXT_FIRST_EXTENT(eh); |
2088 | } else { | |
2089 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 2090 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 2091 | /* Insert after */ |
32de6756 YY |
2092 | ext_debug("insert %u:%llu:[%d]%d before: " |
2093 | "nearest %p\n", | |
80e675f9 EG |
2094 | le32_to_cpu(newext->ee_block), |
2095 | ext4_ext_pblock(newext), | |
556615dc | 2096 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2097 | ext4_ext_get_actual_len(newext), |
2098 | nearex); | |
2099 | nearex++; | |
2100 | } else { | |
2101 | /* Insert before */ | |
2102 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
2103 | ext_debug("insert %u:%llu:[%d]%d after: " |
2104 | "nearest %p\n", | |
8c55e204 | 2105 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2106 | ext4_ext_pblock(newext), |
556615dc | 2107 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2108 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
2109 | nearex); |
2110 | } | |
2111 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
2112 | if (len > 0) { | |
32de6756 | 2113 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
2114 | "move %d extents from 0x%p to 0x%p\n", |
2115 | le32_to_cpu(newext->ee_block), | |
2116 | ext4_ext_pblock(newext), | |
556615dc | 2117 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2118 | ext4_ext_get_actual_len(newext), |
2119 | len, nearex, nearex + 1); | |
2120 | memmove(nearex + 1, nearex, | |
2121 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2122 | } |
a86c6181 AT |
2123 | } |
2124 | ||
e8546d06 | 2125 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2126 | path[depth].p_ext = nearex; |
a86c6181 | 2127 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2128 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2129 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2130 | |
2131 | merge: | |
e7bcf823 | 2132 | /* try to merge extents */ |
107a7bd3 | 2133 | if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2134 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2135 | |
a86c6181 AT |
2136 | |
2137 | /* time to correct all indexes above */ | |
2138 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2139 | if (err) | |
2140 | goto cleanup; | |
2141 | ||
ecb94f5f | 2142 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2143 | |
2144 | cleanup: | |
b7ea89ad TT |
2145 | ext4_ext_drop_refs(npath); |
2146 | kfree(npath); | |
a86c6181 AT |
2147 | return err; |
2148 | } | |
2149 | ||
91dd8c11 LC |
2150 | static int ext4_fill_fiemap_extents(struct inode *inode, |
2151 | ext4_lblk_t block, ext4_lblk_t num, | |
2152 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
2153 | { |
2154 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 2155 | struct ext4_extent *ex; |
69eb33dc | 2156 | struct extent_status es; |
91dd8c11 | 2157 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 2158 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
2159 | int exists, depth = 0, err = 0; |
2160 | unsigned int flags = 0; | |
2161 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 2162 | |
f17722f9 | 2163 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
2164 | num = last - block; |
2165 | /* find extent for this block */ | |
fab3a549 | 2166 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 | 2167 | |
ed8a1a76 | 2168 | path = ext4_find_extent(inode, block, &path, 0); |
6873fa0d | 2169 | if (IS_ERR(path)) { |
91dd8c11 | 2170 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
2171 | err = PTR_ERR(path); |
2172 | path = NULL; | |
2173 | break; | |
2174 | } | |
2175 | ||
2176 | depth = ext_depth(inode); | |
273df556 | 2177 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2178 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2179 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2180 | err = -EIO; | |
2181 | break; | |
2182 | } | |
6873fa0d ES |
2183 | ex = path[depth].p_ext; |
2184 | next = ext4_ext_next_allocated_block(path); | |
2185 | ||
91dd8c11 | 2186 | flags = 0; |
6873fa0d ES |
2187 | exists = 0; |
2188 | if (!ex) { | |
2189 | /* there is no extent yet, so try to allocate | |
2190 | * all requested space */ | |
2191 | start = block; | |
2192 | end = block + num; | |
2193 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2194 | /* need to allocate space before found extent */ | |
2195 | start = block; | |
2196 | end = le32_to_cpu(ex->ee_block); | |
2197 | if (block + num < end) | |
2198 | end = block + num; | |
2199 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2200 | + ext4_ext_get_actual_len(ex)) { | |
2201 | /* need to allocate space after found extent */ | |
2202 | start = block; | |
2203 | end = block + num; | |
2204 | if (end >= next) | |
2205 | end = next; | |
2206 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2207 | /* | |
2208 | * some part of requested space is covered | |
2209 | * by found extent | |
2210 | */ | |
2211 | start = block; | |
2212 | end = le32_to_cpu(ex->ee_block) | |
2213 | + ext4_ext_get_actual_len(ex); | |
2214 | if (block + num < end) | |
2215 | end = block + num; | |
2216 | exists = 1; | |
2217 | } else { | |
2218 | BUG(); | |
2219 | } | |
2220 | BUG_ON(end <= start); | |
2221 | ||
2222 | if (!exists) { | |
69eb33dc ZL |
2223 | es.es_lblk = start; |
2224 | es.es_len = end - start; | |
2225 | es.es_pblk = 0; | |
6873fa0d | 2226 | } else { |
69eb33dc ZL |
2227 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2228 | es.es_len = ext4_ext_get_actual_len(ex); | |
2229 | es.es_pblk = ext4_ext_pblock(ex); | |
556615dc | 2230 | if (ext4_ext_is_unwritten(ex)) |
91dd8c11 | 2231 | flags |= FIEMAP_EXTENT_UNWRITTEN; |
6873fa0d ES |
2232 | } |
2233 | ||
91dd8c11 | 2234 | /* |
69eb33dc ZL |
2235 | * Find delayed extent and update es accordingly. We call |
2236 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2237 | * last existing extent or not. |
2238 | */ | |
69eb33dc | 2239 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2240 | if (!exists && next_del) { |
2241 | exists = 1; | |
72dac95d JL |
2242 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2243 | FIEMAP_EXTENT_UNKNOWN); | |
91dd8c11 LC |
2244 | } |
2245 | up_read(&EXT4_I(inode)->i_data_sem); | |
2246 | ||
69eb33dc ZL |
2247 | if (unlikely(es.es_len == 0)) { |
2248 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
273df556 FM |
2249 | err = -EIO; |
2250 | break; | |
2251 | } | |
6873fa0d | 2252 | |
f7fec032 ZL |
2253 | /* |
2254 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2255 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2256 | * possible that an extent is with unwritten and delayed | |
2257 | * status due to when an extent is delayed allocated and | |
2258 | * is allocated by fallocate status tree will track both of | |
2259 | * them in a extent. | |
2260 | * | |
2261 | * So we could return a unwritten and delayed extent, and | |
2262 | * its block is equal to 'next'. | |
2263 | */ | |
2264 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2265 | flags |= FIEMAP_EXTENT_LAST; |
2266 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2267 | next != EXT_MAX_BLOCKS)) { | |
2268 | EXT4_ERROR_INODE(inode, | |
2269 | "next extent == %u, next " | |
2270 | "delalloc extent = %u", | |
2271 | next, next_del); | |
2272 | err = -EIO; | |
2273 | break; | |
2274 | } | |
6873fa0d ES |
2275 | } |
2276 | ||
91dd8c11 LC |
2277 | if (exists) { |
2278 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2279 | (__u64)es.es_lblk << blksize_bits, |
2280 | (__u64)es.es_pblk << blksize_bits, | |
2281 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2282 | flags); |
2283 | if (err < 0) | |
2284 | break; | |
2285 | if (err == 1) { | |
2286 | err = 0; | |
2287 | break; | |
2288 | } | |
6873fa0d ES |
2289 | } |
2290 | ||
69eb33dc | 2291 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2292 | } |
2293 | ||
b7ea89ad TT |
2294 | ext4_ext_drop_refs(path); |
2295 | kfree(path); | |
6873fa0d ES |
2296 | return err; |
2297 | } | |
2298 | ||
a86c6181 | 2299 | /* |
d0d856e8 RD |
2300 | * ext4_ext_put_gap_in_cache: |
2301 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2302 | * and cache this gap |
2303 | */ | |
09b88252 | 2304 | static void |
a86c6181 | 2305 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2306 | ext4_lblk_t block) |
a86c6181 AT |
2307 | { |
2308 | int depth = ext_depth(inode); | |
2f8e0a7c ZL |
2309 | ext4_lblk_t len; |
2310 | ext4_lblk_t lblock; | |
a86c6181 | 2311 | struct ext4_extent *ex; |
2f8e0a7c | 2312 | struct extent_status es; |
a86c6181 AT |
2313 | |
2314 | ex = path[depth].p_ext; | |
2315 | if (ex == NULL) { | |
2f8e0a7c ZL |
2316 | /* there is no extent yet, so gap is [0;-] */ |
2317 | lblock = 0; | |
2318 | len = EXT_MAX_BLOCKS; | |
a86c6181 AT |
2319 | ext_debug("cache gap(whole file):"); |
2320 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2321 | lblock = block; | |
2322 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2323 | ext_debug("cache gap(before): %u [%u:%u]", |
2324 | block, | |
2325 | le32_to_cpu(ex->ee_block), | |
2326 | ext4_ext_get_actual_len(ex)); | |
a86c6181 | 2327 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2328 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2329 | ext4_lblk_t next; |
8c55e204 | 2330 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2331 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2332 | |
2333 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2334 | ext_debug("cache gap(after): [%u:%u] %u", |
2335 | le32_to_cpu(ex->ee_block), | |
2336 | ext4_ext_get_actual_len(ex), | |
2337 | block); | |
725d26d3 AK |
2338 | BUG_ON(next == lblock); |
2339 | len = next - lblock; | |
a86c6181 | 2340 | } else { |
a86c6181 AT |
2341 | BUG(); |
2342 | } | |
2343 | ||
2f8e0a7c ZL |
2344 | ext4_es_find_delayed_extent_range(inode, lblock, lblock + len - 1, &es); |
2345 | if (es.es_len) { | |
2346 | /* There's delayed extent containing lblock? */ | |
2347 | if (es.es_lblk <= lblock) | |
2348 | return; | |
2349 | len = min(es.es_lblk - lblock, len); | |
2350 | } | |
2351 | ext_debug(" -> %u:%u\n", lblock, len); | |
2352 | ext4_es_insert_extent(inode, lblock, len, ~0, EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2353 | } |
2354 | ||
2355 | /* | |
d0d856e8 RD |
2356 | * ext4_ext_rm_idx: |
2357 | * removes index from the index block. | |
a86c6181 | 2358 | */ |
1d03ec98 | 2359 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2360 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2361 | { |
a86c6181 | 2362 | int err; |
f65e6fba | 2363 | ext4_fsblk_t leaf; |
a86c6181 AT |
2364 | |
2365 | /* free index block */ | |
c36575e6 FL |
2366 | depth--; |
2367 | path = path + depth; | |
bf89d16f | 2368 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2369 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2370 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2371 | return -EIO; | |
2372 | } | |
7e028976 AM |
2373 | err = ext4_ext_get_access(handle, inode, path); |
2374 | if (err) | |
a86c6181 | 2375 | return err; |
0e1147b0 RD |
2376 | |
2377 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2378 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2379 | len *= sizeof(struct ext4_extent_idx); | |
2380 | memmove(path->p_idx, path->p_idx + 1, len); | |
2381 | } | |
2382 | ||
e8546d06 | 2383 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2384 | err = ext4_ext_dirty(handle, inode, path); |
2385 | if (err) | |
a86c6181 | 2386 | return err; |
2ae02107 | 2387 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2388 | trace_ext4_ext_rm_idx(inode, leaf); |
2389 | ||
7dc57615 | 2390 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2391 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2392 | |
2393 | while (--depth >= 0) { | |
2394 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2395 | break; | |
2396 | path--; | |
2397 | err = ext4_ext_get_access(handle, inode, path); | |
2398 | if (err) | |
2399 | break; | |
2400 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2401 | err = ext4_ext_dirty(handle, inode, path); | |
2402 | if (err) | |
2403 | break; | |
2404 | } | |
a86c6181 AT |
2405 | return err; |
2406 | } | |
2407 | ||
2408 | /* | |
ee12b630 MC |
2409 | * ext4_ext_calc_credits_for_single_extent: |
2410 | * This routine returns max. credits that needed to insert an extent | |
2411 | * to the extent tree. | |
2412 | * When pass the actual path, the caller should calculate credits | |
2413 | * under i_data_sem. | |
a86c6181 | 2414 | */ |
525f4ed8 | 2415 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2416 | struct ext4_ext_path *path) |
2417 | { | |
a86c6181 | 2418 | if (path) { |
ee12b630 | 2419 | int depth = ext_depth(inode); |
f3bd1f3f | 2420 | int ret = 0; |
ee12b630 | 2421 | |
a86c6181 | 2422 | /* probably there is space in leaf? */ |
a86c6181 | 2423 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2424 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2425 | |
ee12b630 MC |
2426 | /* |
2427 | * There are some space in the leaf tree, no | |
2428 | * need to account for leaf block credit | |
2429 | * | |
2430 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2431 | * and other metadata blocks still need to be |
ee12b630 MC |
2432 | * accounted. |
2433 | */ | |
525f4ed8 | 2434 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2435 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2436 | return ret; |
ee12b630 MC |
2437 | } |
2438 | } | |
a86c6181 | 2439 | |
525f4ed8 | 2440 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2441 | } |
a86c6181 | 2442 | |
ee12b630 | 2443 | /* |
fffb2739 | 2444 | * How many index/leaf blocks need to change/allocate to add @extents extents? |
ee12b630 | 2445 | * |
fffb2739 JK |
2446 | * If we add a single extent, then in the worse case, each tree level |
2447 | * index/leaf need to be changed in case of the tree split. | |
ee12b630 | 2448 | * |
fffb2739 JK |
2449 | * If more extents are inserted, they could cause the whole tree split more |
2450 | * than once, but this is really rare. | |
ee12b630 | 2451 | */ |
fffb2739 | 2452 | int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
ee12b630 MC |
2453 | { |
2454 | int index; | |
f19d5870 TM |
2455 | int depth; |
2456 | ||
2457 | /* If we are converting the inline data, only one is needed here. */ | |
2458 | if (ext4_has_inline_data(inode)) | |
2459 | return 1; | |
2460 | ||
2461 | depth = ext_depth(inode); | |
a86c6181 | 2462 | |
fffb2739 | 2463 | if (extents <= 1) |
ee12b630 MC |
2464 | index = depth * 2; |
2465 | else | |
2466 | index = depth * 3; | |
a86c6181 | 2467 | |
ee12b630 | 2468 | return index; |
a86c6181 AT |
2469 | } |
2470 | ||
981250ca TT |
2471 | static inline int get_default_free_blocks_flags(struct inode *inode) |
2472 | { | |
2473 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2474 | return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; | |
2475 | else if (ext4_should_journal_data(inode)) | |
2476 | return EXT4_FREE_BLOCKS_FORGET; | |
2477 | return 0; | |
2478 | } | |
2479 | ||
a86c6181 | 2480 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
0aa06000 | 2481 | struct ext4_extent *ex, |
d23142c6 | 2482 | long long *partial_cluster, |
0aa06000 | 2483 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2484 | { |
0aa06000 | 2485 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
345ee947 | 2486 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2487 | ext4_fsblk_t pblk; |
981250ca | 2488 | int flags = get_default_free_blocks_flags(inode); |
18888cf0 | 2489 | |
0aa06000 TT |
2490 | /* |
2491 | * For bigalloc file systems, we never free a partial cluster | |
2492 | * at the beginning of the extent. Instead, we make a note | |
2493 | * that we tried freeing the cluster, and check to see if we | |
2494 | * need to free it on a subsequent call to ext4_remove_blocks, | |
345ee947 | 2495 | * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space. |
0aa06000 TT |
2496 | */ |
2497 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2498 | ||
d8990240 | 2499 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2500 | /* |
2501 | * If we have a partial cluster, and it's different from the | |
2502 | * cluster of the last block, we need to explicitly free the | |
2503 | * partial cluster here. | |
2504 | */ | |
2505 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
345ee947 EW |
2506 | if (*partial_cluster > 0 && |
2507 | *partial_cluster != (long long) EXT4_B2C(sbi, pblk)) { | |
0aa06000 TT |
2508 | ext4_free_blocks(handle, inode, NULL, |
2509 | EXT4_C2B(sbi, *partial_cluster), | |
2510 | sbi->s_cluster_ratio, flags); | |
2511 | *partial_cluster = 0; | |
2512 | } | |
2513 | ||
a86c6181 AT |
2514 | #ifdef EXTENTS_STATS |
2515 | { | |
2516 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2517 | spin_lock(&sbi->s_ext_stats_lock); |
2518 | sbi->s_ext_blocks += ee_len; | |
2519 | sbi->s_ext_extents++; | |
2520 | if (ee_len < sbi->s_ext_min) | |
2521 | sbi->s_ext_min = ee_len; | |
2522 | if (ee_len > sbi->s_ext_max) | |
2523 | sbi->s_ext_max = ee_len; | |
2524 | if (ext_depth(inode) > sbi->s_depth_max) | |
2525 | sbi->s_depth_max = ext_depth(inode); | |
2526 | spin_unlock(&sbi->s_ext_stats_lock); | |
2527 | } | |
2528 | #endif | |
2529 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2530 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2531 | /* tail removal */ |
725d26d3 | 2532 | ext4_lblk_t num; |
345ee947 | 2533 | long long first_cluster; |
725d26d3 | 2534 | |
a2df2a63 | 2535 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 | 2536 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
d23142c6 LC |
2537 | /* |
2538 | * Usually we want to free partial cluster at the end of the | |
2539 | * extent, except for the situation when the cluster is still | |
2540 | * used by any other extent (partial_cluster is negative). | |
2541 | */ | |
2542 | if (*partial_cluster < 0 && | |
345ee947 | 2543 | *partial_cluster == -(long long) EXT4_B2C(sbi, pblk+num-1)) |
d23142c6 LC |
2544 | flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; |
2545 | ||
2546 | ext_debug("free last %u blocks starting %llu partial %lld\n", | |
2547 | num, pblk, *partial_cluster); | |
0aa06000 TT |
2548 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); |
2549 | /* | |
2550 | * If the block range to be freed didn't start at the | |
2551 | * beginning of a cluster, and we removed the entire | |
d23142c6 LC |
2552 | * extent and the cluster is not used by any other extent, |
2553 | * save the partial cluster here, since we might need to | |
345ee947 EW |
2554 | * delete if we determine that the truncate or punch hole |
2555 | * operation has removed all of the blocks in the cluster. | |
2556 | * If that cluster is used by another extent, preserve its | |
2557 | * negative value so it isn't freed later on. | |
d23142c6 | 2558 | * |
345ee947 EW |
2559 | * If the whole extent wasn't freed, we've reached the |
2560 | * start of the truncated/punched region and have finished | |
2561 | * removing blocks. If there's a partial cluster here it's | |
2562 | * shared with the remainder of the extent and is no longer | |
2563 | * a candidate for removal. | |
0aa06000 | 2564 | */ |
345ee947 EW |
2565 | if (EXT4_PBLK_COFF(sbi, pblk) && ee_len == num) { |
2566 | first_cluster = (long long) EXT4_B2C(sbi, pblk); | |
2567 | if (first_cluster != -*partial_cluster) | |
2568 | *partial_cluster = first_cluster; | |
2569 | } else { | |
0aa06000 | 2570 | *partial_cluster = 0; |
345ee947 | 2571 | } |
78fb9cdf LC |
2572 | } else |
2573 | ext4_error(sbi->s_sb, "strange request: removal(2) " | |
2574 | "%u-%u from %u:%u\n", | |
2575 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2576 | return 0; |
2577 | } | |
2578 | ||
d583fb87 AH |
2579 | |
2580 | /* | |
2581 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
5bf43760 EW |
2582 | * blocks appearing between "start" and "end". Both "start" |
2583 | * and "end" must appear in the same extent or EIO is returned. | |
d583fb87 AH |
2584 | * |
2585 | * @handle: The journal handle | |
2586 | * @inode: The files inode | |
2587 | * @path: The path to the leaf | |
d23142c6 | 2588 | * @partial_cluster: The cluster which we'll have to free if all extents |
5bf43760 EW |
2589 | * has been released from it. However, if this value is |
2590 | * negative, it's a cluster just to the right of the | |
2591 | * punched region and it must not be freed. | |
d583fb87 AH |
2592 | * @start: The first block to remove |
2593 | * @end: The last block to remove | |
2594 | */ | |
a86c6181 AT |
2595 | static int |
2596 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
d23142c6 LC |
2597 | struct ext4_ext_path *path, |
2598 | long long *partial_cluster, | |
0aa06000 | 2599 | ext4_lblk_t start, ext4_lblk_t end) |
a86c6181 | 2600 | { |
0aa06000 | 2601 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2602 | int err = 0, correct_index = 0; |
2603 | int depth = ext_depth(inode), credits; | |
2604 | struct ext4_extent_header *eh; | |
750c9c47 | 2605 | ext4_lblk_t a, b; |
725d26d3 AK |
2606 | unsigned num; |
2607 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2608 | unsigned short ex_ee_len; |
556615dc | 2609 | unsigned unwritten = 0; |
a86c6181 | 2610 | struct ext4_extent *ex; |
d23142c6 | 2611 | ext4_fsblk_t pblk; |
a86c6181 | 2612 | |
c29c0ae7 | 2613 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2614 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2615 | if (!path[depth].p_hdr) |
2616 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2617 | eh = path[depth].p_hdr; | |
273df556 FM |
2618 | if (unlikely(path[depth].p_hdr == NULL)) { |
2619 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2620 | return -EIO; | |
2621 | } | |
a86c6181 | 2622 | /* find where to start removing */ |
6ae06ff5 AS |
2623 | ex = path[depth].p_ext; |
2624 | if (!ex) | |
2625 | ex = EXT_LAST_EXTENT(eh); | |
a86c6181 AT |
2626 | |
2627 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2628 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2629 | |
d8990240 AK |
2630 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2631 | ||
a86c6181 AT |
2632 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2633 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 | 2634 | |
556615dc LC |
2635 | if (ext4_ext_is_unwritten(ex)) |
2636 | unwritten = 1; | |
a41f2071 | 2637 | else |
556615dc | 2638 | unwritten = 0; |
a41f2071 | 2639 | |
553f9008 | 2640 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
556615dc | 2641 | unwritten, ex_ee_len); |
a86c6181 AT |
2642 | path[depth].p_ext = ex; |
2643 | ||
2644 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2645 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2646 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2647 | |
2648 | ext_debug(" border %u:%u\n", a, b); | |
2649 | ||
d583fb87 | 2650 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2651 | if (end < ex_ee_block) { |
d23142c6 LC |
2652 | /* |
2653 | * We're going to skip this extent and move to another, | |
f4226d9e EW |
2654 | * so note that its first cluster is in use to avoid |
2655 | * freeing it when removing blocks. Eventually, the | |
2656 | * right edge of the truncated/punched region will | |
2657 | * be just to the left. | |
d23142c6 | 2658 | */ |
f4226d9e EW |
2659 | if (sbi->s_cluster_ratio > 1) { |
2660 | pblk = ext4_ext_pblock(ex); | |
d23142c6 | 2661 | *partial_cluster = |
f4226d9e EW |
2662 | -(long long) EXT4_B2C(sbi, pblk); |
2663 | } | |
d583fb87 AH |
2664 | ex--; |
2665 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2666 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2667 | continue; | |
750c9c47 | 2668 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2669 | EXT4_ERROR_INODE(inode, |
2670 | "can not handle truncate %u:%u " | |
2671 | "on extent %u:%u", | |
2672 | start, end, ex_ee_block, | |
2673 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2674 | err = -EIO; |
2675 | goto out; | |
a86c6181 AT |
2676 | } else if (a != ex_ee_block) { |
2677 | /* remove tail of the extent */ | |
750c9c47 | 2678 | num = a - ex_ee_block; |
a86c6181 AT |
2679 | } else { |
2680 | /* remove whole extent: excellent! */ | |
a86c6181 | 2681 | num = 0; |
a86c6181 | 2682 | } |
34071da7 TT |
2683 | /* |
2684 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2685 | * descriptor) for each block group; assume two block | |
2686 | * groups plus ex_ee_len/blocks_per_block_group for | |
2687 | * the worst case | |
2688 | */ | |
2689 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2690 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2691 | correct_index = 1; | |
2692 | credits += (ext_depth(inode)) + 1; | |
2693 | } | |
5aca07eb | 2694 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2695 | |
487caeef | 2696 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2697 | if (err) |
a86c6181 | 2698 | goto out; |
a86c6181 AT |
2699 | |
2700 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2701 | if (err) | |
2702 | goto out; | |
2703 | ||
0aa06000 TT |
2704 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2705 | a, b); | |
a86c6181 AT |
2706 | if (err) |
2707 | goto out; | |
2708 | ||
750c9c47 | 2709 | if (num == 0) |
d0d856e8 | 2710 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2711 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2712 | |
a86c6181 | 2713 | ex->ee_len = cpu_to_le16(num); |
749269fa | 2714 | /* |
556615dc | 2715 | * Do not mark unwritten if all the blocks in the |
749269fa AA |
2716 | * extent have been removed. |
2717 | */ | |
556615dc LC |
2718 | if (unwritten && num) |
2719 | ext4_ext_mark_unwritten(ex); | |
d583fb87 AH |
2720 | /* |
2721 | * If the extent was completely released, | |
2722 | * we need to remove it from the leaf | |
2723 | */ | |
2724 | if (num == 0) { | |
f17722f9 | 2725 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2726 | /* |
2727 | * For hole punching, we need to scoot all the | |
2728 | * extents up when an extent is removed so that | |
2729 | * we dont have blank extents in the middle | |
2730 | */ | |
2731 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2732 | sizeof(struct ext4_extent)); | |
2733 | ||
2734 | /* Now get rid of the one at the end */ | |
2735 | memset(EXT_LAST_EXTENT(eh), 0, | |
2736 | sizeof(struct ext4_extent)); | |
2737 | } | |
2738 | le16_add_cpu(&eh->eh_entries, -1); | |
5bf43760 | 2739 | } |
d583fb87 | 2740 | |
750c9c47 DM |
2741 | err = ext4_ext_dirty(handle, inode, path + depth); |
2742 | if (err) | |
2743 | goto out; | |
2744 | ||
bf52c6f7 | 2745 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2746 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2747 | ex--; |
2748 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2749 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2750 | } |
2751 | ||
2752 | if (correct_index && eh->eh_entries) | |
2753 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2754 | ||
0aa06000 | 2755 | /* |
ad6599ab EW |
2756 | * If there's a partial cluster and at least one extent remains in |
2757 | * the leaf, free the partial cluster if it isn't shared with the | |
5bf43760 EW |
2758 | * current extent. If it is shared with the current extent |
2759 | * we zero partial_cluster because we've reached the start of the | |
2760 | * truncated/punched region and we're done removing blocks. | |
0aa06000 | 2761 | */ |
5bf43760 EW |
2762 | if (*partial_cluster > 0 && ex >= EXT_FIRST_EXTENT(eh)) { |
2763 | pblk = ext4_ext_pblock(ex) + ex_ee_len - 1; | |
2764 | if (*partial_cluster != (long long) EXT4_B2C(sbi, pblk)) { | |
2765 | ext4_free_blocks(handle, inode, NULL, | |
2766 | EXT4_C2B(sbi, *partial_cluster), | |
2767 | sbi->s_cluster_ratio, | |
2768 | get_default_free_blocks_flags(inode)); | |
2769 | } | |
0aa06000 TT |
2770 | *partial_cluster = 0; |
2771 | } | |
2772 | ||
a86c6181 AT |
2773 | /* if this leaf is free, then we should |
2774 | * remove it from index block above */ | |
2775 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2776 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2777 | |
2778 | out: | |
2779 | return err; | |
2780 | } | |
2781 | ||
2782 | /* | |
d0d856e8 RD |
2783 | * ext4_ext_more_to_rm: |
2784 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2785 | */ |
09b88252 | 2786 | static int |
a86c6181 AT |
2787 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2788 | { | |
2789 | BUG_ON(path->p_idx == NULL); | |
2790 | ||
2791 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2792 | return 0; | |
2793 | ||
2794 | /* | |
d0d856e8 | 2795 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2796 | * so we have to consider current index for truncation |
2797 | */ | |
2798 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2799 | return 0; | |
2800 | return 1; | |
2801 | } | |
2802 | ||
26a4c0c6 TT |
2803 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2804 | ext4_lblk_t end) | |
a86c6181 | 2805 | { |
f4226d9e | 2806 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 | 2807 | int depth = ext_depth(inode); |
968dee77 | 2808 | struct ext4_ext_path *path = NULL; |
d23142c6 | 2809 | long long partial_cluster = 0; |
a86c6181 | 2810 | handle_t *handle; |
6f2080e6 | 2811 | int i = 0, err = 0; |
a86c6181 | 2812 | |
5f95d21f | 2813 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2814 | |
2815 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2816 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2817 | if (IS_ERR(handle)) |
2818 | return PTR_ERR(handle); | |
2819 | ||
0617b83f | 2820 | again: |
61801325 | 2821 | trace_ext4_ext_remove_space(inode, start, end, depth); |
d8990240 | 2822 | |
5f95d21f LC |
2823 | /* |
2824 | * Check if we are removing extents inside the extent tree. If that | |
2825 | * is the case, we are going to punch a hole inside the extent tree | |
2826 | * so we have to check whether we need to split the extent covering | |
2827 | * the last block to remove so we can easily remove the part of it | |
2828 | * in ext4_ext_rm_leaf(). | |
2829 | */ | |
2830 | if (end < EXT_MAX_BLOCKS - 1) { | |
2831 | struct ext4_extent *ex; | |
f4226d9e EW |
2832 | ext4_lblk_t ee_block, ex_end, lblk; |
2833 | ext4_fsblk_t pblk; | |
5f95d21f | 2834 | |
f4226d9e | 2835 | /* find extent for or closest extent to this block */ |
ed8a1a76 | 2836 | path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE); |
5f95d21f LC |
2837 | if (IS_ERR(path)) { |
2838 | ext4_journal_stop(handle); | |
2839 | return PTR_ERR(path); | |
2840 | } | |
2841 | depth = ext_depth(inode); | |
6f2080e6 | 2842 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2843 | ex = path[depth].p_ext; |
968dee77 | 2844 | if (!ex) { |
6f2080e6 DM |
2845 | if (depth) { |
2846 | EXT4_ERROR_INODE(inode, | |
2847 | "path[%d].p_hdr == NULL", | |
2848 | depth); | |
2849 | err = -EIO; | |
2850 | } | |
2851 | goto out; | |
968dee77 | 2852 | } |
5f95d21f LC |
2853 | |
2854 | ee_block = le32_to_cpu(ex->ee_block); | |
f4226d9e | 2855 | ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1; |
5f95d21f LC |
2856 | |
2857 | /* | |
2858 | * See if the last block is inside the extent, if so split | |
2859 | * the extent at 'end' block so we can easily remove the | |
2860 | * tail of the first part of the split extent in | |
2861 | * ext4_ext_rm_leaf(). | |
2862 | */ | |
f4226d9e EW |
2863 | if (end >= ee_block && end < ex_end) { |
2864 | ||
2865 | /* | |
2866 | * If we're going to split the extent, note that | |
2867 | * the cluster containing the block after 'end' is | |
2868 | * in use to avoid freeing it when removing blocks. | |
2869 | */ | |
2870 | if (sbi->s_cluster_ratio > 1) { | |
2871 | pblk = ext4_ext_pblock(ex) + end - ee_block + 2; | |
2872 | partial_cluster = | |
2873 | -(long long) EXT4_B2C(sbi, pblk); | |
2874 | } | |
2875 | ||
5f95d21f LC |
2876 | /* |
2877 | * Split the extent in two so that 'end' is the last | |
27dd4385 LC |
2878 | * block in the first new extent. Also we should not |
2879 | * fail removing space due to ENOSPC so try to use | |
2880 | * reserved block if that happens. | |
5f95d21f | 2881 | */ |
dfe50809 | 2882 | err = ext4_force_split_extent_at(handle, inode, &path, |
fcf6b1b7 | 2883 | end + 1, 1); |
5f95d21f LC |
2884 | if (err < 0) |
2885 | goto out; | |
f4226d9e EW |
2886 | |
2887 | } else if (sbi->s_cluster_ratio > 1 && end >= ex_end) { | |
2888 | /* | |
2889 | * If there's an extent to the right its first cluster | |
2890 | * contains the immediate right boundary of the | |
2891 | * truncated/punched region. Set partial_cluster to | |
2892 | * its negative value so it won't be freed if shared | |
2893 | * with the current extent. The end < ee_block case | |
2894 | * is handled in ext4_ext_rm_leaf(). | |
2895 | */ | |
2896 | lblk = ex_end + 1; | |
2897 | err = ext4_ext_search_right(inode, path, &lblk, &pblk, | |
2898 | &ex); | |
2899 | if (err) | |
2900 | goto out; | |
2901 | if (pblk) | |
2902 | partial_cluster = | |
2903 | -(long long) EXT4_B2C(sbi, pblk); | |
5f95d21f | 2904 | } |
5f95d21f | 2905 | } |
a86c6181 | 2906 | /* |
d0d856e8 RD |
2907 | * We start scanning from right side, freeing all the blocks |
2908 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2909 | */ |
0617b83f | 2910 | depth = ext_depth(inode); |
968dee77 AS |
2911 | if (path) { |
2912 | int k = i = depth; | |
2913 | while (--k > 0) | |
2914 | path[k].p_block = | |
2915 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2916 | } else { | |
2917 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2918 | GFP_NOFS); | |
2919 | if (path == NULL) { | |
2920 | ext4_journal_stop(handle); | |
2921 | return -ENOMEM; | |
2922 | } | |
10809df8 | 2923 | path[0].p_maxdepth = path[0].p_depth = depth; |
968dee77 | 2924 | path[0].p_hdr = ext_inode_hdr(inode); |
89a4e48f | 2925 | i = 0; |
5f95d21f | 2926 | |
c349179b | 2927 | if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
968dee77 AS |
2928 | err = -EIO; |
2929 | goto out; | |
2930 | } | |
a86c6181 | 2931 | } |
968dee77 | 2932 | err = 0; |
a86c6181 AT |
2933 | |
2934 | while (i >= 0 && err == 0) { | |
2935 | if (i == depth) { | |
2936 | /* this is leaf block */ | |
d583fb87 | 2937 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2938 | &partial_cluster, start, |
5f95d21f | 2939 | end); |
d0d856e8 | 2940 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2941 | brelse(path[i].p_bh); |
2942 | path[i].p_bh = NULL; | |
2943 | i--; | |
2944 | continue; | |
2945 | } | |
2946 | ||
2947 | /* this is index block */ | |
2948 | if (!path[i].p_hdr) { | |
2949 | ext_debug("initialize header\n"); | |
2950 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2951 | } |
2952 | ||
a86c6181 | 2953 | if (!path[i].p_idx) { |
d0d856e8 | 2954 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2955 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2956 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2957 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2958 | path[i].p_hdr, | |
2959 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2960 | } else { | |
d0d856e8 | 2961 | /* we were already here, see at next index */ |
a86c6181 AT |
2962 | path[i].p_idx--; |
2963 | } | |
2964 | ||
2965 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2966 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2967 | path[i].p_idx); | |
2968 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2969 | struct buffer_head *bh; |
a86c6181 | 2970 | /* go to the next level */ |
2ae02107 | 2971 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2972 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2973 | memset(path + i + 1, 0, sizeof(*path)); |
7d7ea89e | 2974 | bh = read_extent_tree_block(inode, |
107a7bd3 TT |
2975 | ext4_idx_pblock(path[i].p_idx), depth - i - 1, |
2976 | EXT4_EX_NOCACHE); | |
7d7ea89e | 2977 | if (IS_ERR(bh)) { |
a86c6181 | 2978 | /* should we reset i_size? */ |
7d7ea89e | 2979 | err = PTR_ERR(bh); |
a86c6181 AT |
2980 | break; |
2981 | } | |
76828c88 TT |
2982 | /* Yield here to deal with large extent trees. |
2983 | * Should be a no-op if we did IO above. */ | |
2984 | cond_resched(); | |
c29c0ae7 AT |
2985 | if (WARN_ON(i + 1 > depth)) { |
2986 | err = -EIO; | |
c29c0ae7 AT |
2987 | break; |
2988 | } | |
2989 | path[i + 1].p_bh = bh; | |
a86c6181 | 2990 | |
d0d856e8 RD |
2991 | /* save actual number of indexes since this |
2992 | * number is changed at the next iteration */ | |
a86c6181 AT |
2993 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2994 | i++; | |
2995 | } else { | |
d0d856e8 | 2996 | /* we finished processing this index, go up */ |
a86c6181 | 2997 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2998 | /* index is empty, remove it; |
a86c6181 AT |
2999 | * handle must be already prepared by the |
3000 | * truncatei_leaf() */ | |
c36575e6 | 3001 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 3002 | } |
d0d856e8 | 3003 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
3004 | brelse(path[i].p_bh); |
3005 | path[i].p_bh = NULL; | |
3006 | i--; | |
3007 | ext_debug("return to level %d\n", i); | |
3008 | } | |
3009 | } | |
3010 | ||
61801325 LC |
3011 | trace_ext4_ext_remove_space_done(inode, start, end, depth, |
3012 | partial_cluster, path->p_hdr->eh_entries); | |
d8990240 | 3013 | |
0756b908 EW |
3014 | /* |
3015 | * If we still have something in the partial cluster and we have removed | |
7b415bf6 | 3016 | * even the first extent, then we should free the blocks in the partial |
0756b908 EW |
3017 | * cluster as well. (This code will only run when there are no leaves |
3018 | * to the immediate left of the truncated/punched region.) | |
3019 | */ | |
3020 | if (partial_cluster > 0 && err == 0) { | |
3021 | /* don't zero partial_cluster since it's not used afterwards */ | |
7b415bf6 | 3022 | ext4_free_blocks(handle, inode, NULL, |
f4226d9e | 3023 | EXT4_C2B(sbi, partial_cluster), |
0756b908 EW |
3024 | sbi->s_cluster_ratio, |
3025 | get_default_free_blocks_flags(inode)); | |
7b415bf6 AK |
3026 | } |
3027 | ||
a86c6181 AT |
3028 | /* TODO: flexible tree reduction should be here */ |
3029 | if (path->p_hdr->eh_entries == 0) { | |
3030 | /* | |
d0d856e8 RD |
3031 | * truncate to zero freed all the tree, |
3032 | * so we need to correct eh_depth | |
a86c6181 AT |
3033 | */ |
3034 | err = ext4_ext_get_access(handle, inode, path); | |
3035 | if (err == 0) { | |
3036 | ext_inode_hdr(inode)->eh_depth = 0; | |
3037 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 3038 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
3039 | err = ext4_ext_dirty(handle, inode, path); |
3040 | } | |
3041 | } | |
3042 | out: | |
b7ea89ad TT |
3043 | ext4_ext_drop_refs(path); |
3044 | kfree(path); | |
3045 | path = NULL; | |
dfe50809 TT |
3046 | if (err == -EAGAIN) |
3047 | goto again; | |
a86c6181 AT |
3048 | ext4_journal_stop(handle); |
3049 | ||
3050 | return err; | |
3051 | } | |
3052 | ||
3053 | /* | |
3054 | * called at mount time | |
3055 | */ | |
3056 | void ext4_ext_init(struct super_block *sb) | |
3057 | { | |
3058 | /* | |
3059 | * possible initialization would be here | |
3060 | */ | |
3061 | ||
83982b6f | 3062 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 3063 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 3064 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 3065 | #ifdef AGGRESSIVE_TEST |
92b97816 | 3066 | ", aggressive tests" |
a86c6181 AT |
3067 | #endif |
3068 | #ifdef CHECK_BINSEARCH | |
92b97816 | 3069 | ", check binsearch" |
a86c6181 AT |
3070 | #endif |
3071 | #ifdef EXTENTS_STATS | |
92b97816 | 3072 | ", stats" |
a86c6181 | 3073 | #endif |
92b97816 | 3074 | "\n"); |
90576c0b | 3075 | #endif |
a86c6181 AT |
3076 | #ifdef EXTENTS_STATS |
3077 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
3078 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
3079 | EXT4_SB(sb)->s_ext_max = 0; | |
3080 | #endif | |
3081 | } | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * called at umount time | |
3086 | */ | |
3087 | void ext4_ext_release(struct super_block *sb) | |
3088 | { | |
83982b6f | 3089 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
3090 | return; |
3091 | ||
3092 | #ifdef EXTENTS_STATS | |
3093 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
3094 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3095 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
3096 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
3097 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
3098 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
3099 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
3100 | } | |
3101 | #endif | |
3102 | } | |
3103 | ||
d7b2a00c ZL |
3104 | static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex) |
3105 | { | |
3106 | ext4_lblk_t ee_block; | |
3107 | ext4_fsblk_t ee_pblock; | |
3108 | unsigned int ee_len; | |
3109 | ||
3110 | ee_block = le32_to_cpu(ex->ee_block); | |
3111 | ee_len = ext4_ext_get_actual_len(ex); | |
3112 | ee_pblock = ext4_ext_pblock(ex); | |
3113 | ||
3114 | if (ee_len == 0) | |
3115 | return 0; | |
3116 | ||
3117 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
3118 | EXTENT_STATUS_WRITTEN); | |
3119 | } | |
3120 | ||
093a088b AK |
3121 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
3122 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
3123 | { | |
2407518d LC |
3124 | ext4_fsblk_t ee_pblock; |
3125 | unsigned int ee_len; | |
b720303d | 3126 | int ret; |
093a088b | 3127 | |
093a088b | 3128 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 3129 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 3130 | |
a107e5a3 | 3131 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
3132 | if (ret > 0) |
3133 | ret = 0; | |
093a088b | 3134 | |
2407518d | 3135 | return ret; |
093a088b AK |
3136 | } |
3137 | ||
47ea3bb5 YY |
3138 | /* |
3139 | * ext4_split_extent_at() splits an extent at given block. | |
3140 | * | |
3141 | * @handle: the journal handle | |
3142 | * @inode: the file inode | |
3143 | * @path: the path to the extent | |
3144 | * @split: the logical block where the extent is splitted. | |
3145 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
556615dc | 3146 | * the states(init or unwritten) of new extents. |
47ea3bb5 YY |
3147 | * @flags: flags used to insert new extent to extent tree. |
3148 | * | |
3149 | * | |
3150 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
3151 | * of which are deterimined by split_flag. | |
3152 | * | |
3153 | * There are two cases: | |
3154 | * a> the extent are splitted into two extent. | |
3155 | * b> split is not needed, and just mark the extent. | |
3156 | * | |
3157 | * return 0 on success. | |
3158 | */ | |
3159 | static int ext4_split_extent_at(handle_t *handle, | |
3160 | struct inode *inode, | |
dfe50809 | 3161 | struct ext4_ext_path **ppath, |
47ea3bb5 YY |
3162 | ext4_lblk_t split, |
3163 | int split_flag, | |
3164 | int flags) | |
3165 | { | |
dfe50809 | 3166 | struct ext4_ext_path *path = *ppath; |
47ea3bb5 YY |
3167 | ext4_fsblk_t newblock; |
3168 | ext4_lblk_t ee_block; | |
adb23551 | 3169 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
3170 | struct ext4_extent *ex2 = NULL; |
3171 | unsigned int ee_len, depth; | |
3172 | int err = 0; | |
3173 | ||
dee1f973 DM |
3174 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
3175 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
3176 | ||
47ea3bb5 YY |
3177 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
3178 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3179 | ||
3180 | ext4_ext_show_leaf(inode, path); | |
3181 | ||
3182 | depth = ext_depth(inode); | |
3183 | ex = path[depth].p_ext; | |
3184 | ee_block = le32_to_cpu(ex->ee_block); | |
3185 | ee_len = ext4_ext_get_actual_len(ex); | |
3186 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3187 | ||
3188 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
556615dc | 3189 | BUG_ON(!ext4_ext_is_unwritten(ex) && |
357b66fd | 3190 | split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc LC |
3191 | EXT4_EXT_MARK_UNWRIT1 | |
3192 | EXT4_EXT_MARK_UNWRIT2)); | |
47ea3bb5 YY |
3193 | |
3194 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3195 | if (err) | |
3196 | goto out; | |
3197 | ||
3198 | if (split == ee_block) { | |
3199 | /* | |
3200 | * case b: block @split is the block that the extent begins with | |
3201 | * then we just change the state of the extent, and splitting | |
3202 | * is not needed. | |
3203 | */ | |
556615dc LC |
3204 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3205 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3206 | else |
3207 | ext4_ext_mark_initialized(ex); | |
3208 | ||
3209 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3210 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3211 | |
ecb94f5f | 3212 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3213 | goto out; |
3214 | } | |
3215 | ||
3216 | /* case a */ | |
3217 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3218 | ex->ee_len = cpu_to_le16(split - ee_block); | |
556615dc LC |
3219 | if (split_flag & EXT4_EXT_MARK_UNWRIT1) |
3220 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3221 | |
3222 | /* | |
3223 | * path may lead to new leaf, not to original leaf any more | |
3224 | * after ext4_ext_insert_extent() returns, | |
3225 | */ | |
3226 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3227 | if (err) | |
3228 | goto fix_extent_len; | |
3229 | ||
3230 | ex2 = &newex; | |
3231 | ex2->ee_block = cpu_to_le32(split); | |
3232 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3233 | ext4_ext_store_pblock(ex2, newblock); | |
556615dc LC |
3234 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3235 | ext4_ext_mark_unwritten(ex2); | |
47ea3bb5 | 3236 | |
dfe50809 | 3237 | err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags); |
47ea3bb5 | 3238 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { |
dee1f973 | 3239 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3240 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3241 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3242 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3243 | zero_ex.ee_len = cpu_to_le16( |
3244 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3245 | ext4_ext_store_pblock(&zero_ex, |
3246 | ext4_ext_pblock(ex2)); | |
3247 | } else { | |
dee1f973 | 3248 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3249 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3250 | zero_ex.ee_len = cpu_to_le16( |
3251 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3252 | ext4_ext_store_pblock(&zero_ex, |
3253 | ext4_ext_pblock(ex)); | |
3254 | } | |
3255 | } else { | |
dee1f973 | 3256 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3257 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3258 | zero_ex.ee_len = cpu_to_le16( |
3259 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3260 | ext4_ext_store_pblock(&zero_ex, |
3261 | ext4_ext_pblock(&orig_ex)); | |
3262 | } | |
dee1f973 | 3263 | |
47ea3bb5 YY |
3264 | if (err) |
3265 | goto fix_extent_len; | |
3266 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3267 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3268 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3269 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3270 | if (err) |
3271 | goto fix_extent_len; | |
3272 | ||
3273 | /* update extent status tree */ | |
d7b2a00c | 3274 | err = ext4_zeroout_es(inode, &zero_ex); |
adb23551 | 3275 | |
47ea3bb5 YY |
3276 | goto out; |
3277 | } else if (err) | |
3278 | goto fix_extent_len; | |
3279 | ||
3280 | out: | |
3281 | ext4_ext_show_leaf(inode, path); | |
3282 | return err; | |
3283 | ||
3284 | fix_extent_len: | |
3285 | ex->ee_len = orig_ex.ee_len; | |
29faed16 | 3286 | ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3287 | return err; |
3288 | } | |
3289 | ||
3290 | /* | |
3291 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3292 | * by @map as split_flags indicates | |
3293 | * | |
70261f56 | 3294 | * It may result in splitting the extent into multiple extents (up to three) |
47ea3bb5 YY |
3295 | * There are three possibilities: |
3296 | * a> There is no split required | |
3297 | * b> Splits in two extents: Split is happening at either end of the extent | |
3298 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3299 | * | |
3300 | */ | |
3301 | static int ext4_split_extent(handle_t *handle, | |
3302 | struct inode *inode, | |
dfe50809 | 3303 | struct ext4_ext_path **ppath, |
47ea3bb5 YY |
3304 | struct ext4_map_blocks *map, |
3305 | int split_flag, | |
3306 | int flags) | |
3307 | { | |
dfe50809 | 3308 | struct ext4_ext_path *path = *ppath; |
47ea3bb5 YY |
3309 | ext4_lblk_t ee_block; |
3310 | struct ext4_extent *ex; | |
3311 | unsigned int ee_len, depth; | |
3312 | int err = 0; | |
556615dc | 3313 | int unwritten; |
47ea3bb5 | 3314 | int split_flag1, flags1; |
3a225670 | 3315 | int allocated = map->m_len; |
47ea3bb5 YY |
3316 | |
3317 | depth = ext_depth(inode); | |
3318 | ex = path[depth].p_ext; | |
3319 | ee_block = le32_to_cpu(ex->ee_block); | |
3320 | ee_len = ext4_ext_get_actual_len(ex); | |
556615dc | 3321 | unwritten = ext4_ext_is_unwritten(ex); |
47ea3bb5 YY |
3322 | |
3323 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3324 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 | 3325 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
556615dc LC |
3326 | if (unwritten) |
3327 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1 | | |
3328 | EXT4_EXT_MARK_UNWRIT2; | |
dee1f973 DM |
3329 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3330 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
dfe50809 | 3331 | err = ext4_split_extent_at(handle, inode, ppath, |
47ea3bb5 | 3332 | map->m_lblk + map->m_len, split_flag1, flags1); |
93917411 YY |
3333 | if (err) |
3334 | goto out; | |
3a225670 ZL |
3335 | } else { |
3336 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3337 | } |
357b66fd DM |
3338 | /* |
3339 | * Update path is required because previous ext4_split_extent_at() may | |
3340 | * result in split of original leaf or extent zeroout. | |
3341 | */ | |
ed8a1a76 | 3342 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
47ea3bb5 YY |
3343 | if (IS_ERR(path)) |
3344 | return PTR_ERR(path); | |
357b66fd DM |
3345 | depth = ext_depth(inode); |
3346 | ex = path[depth].p_ext; | |
a18ed359 DM |
3347 | if (!ex) { |
3348 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3349 | (unsigned long) map->m_lblk); | |
3350 | return -EIO; | |
3351 | } | |
556615dc | 3352 | unwritten = ext4_ext_is_unwritten(ex); |
357b66fd | 3353 | split_flag1 = 0; |
47ea3bb5 YY |
3354 | |
3355 | if (map->m_lblk >= ee_block) { | |
357b66fd | 3356 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
556615dc LC |
3357 | if (unwritten) { |
3358 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1; | |
357b66fd | 3359 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc | 3360 | EXT4_EXT_MARK_UNWRIT2); |
357b66fd | 3361 | } |
dfe50809 | 3362 | err = ext4_split_extent_at(handle, inode, ppath, |
47ea3bb5 YY |
3363 | map->m_lblk, split_flag1, flags); |
3364 | if (err) | |
3365 | goto out; | |
3366 | } | |
3367 | ||
3368 | ext4_ext_show_leaf(inode, path); | |
3369 | out: | |
3a225670 | 3370 | return err ? err : allocated; |
47ea3bb5 YY |
3371 | } |
3372 | ||
56055d3a | 3373 | /* |
e35fd660 | 3374 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
556615dc | 3375 | * to an unwritten extent. It may result in splitting the unwritten |
25985edc | 3376 | * extent into multiple extents (up to three - one initialized and two |
556615dc | 3377 | * unwritten). |
56055d3a AA |
3378 | * There are three possibilities: |
3379 | * a> There is no split required: Entire extent should be initialized | |
3380 | * b> Splits in two extents: Write is happening at either end of the extent | |
3381 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3382 | * |
3383 | * Pre-conditions: | |
556615dc | 3384 | * - The extent pointed to by 'path' is unwritten. |
6f91bc5f EG |
3385 | * - The extent pointed to by 'path' contains a superset |
3386 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3387 | * | |
3388 | * Post-conditions on success: | |
3389 | * - the returned value is the number of blocks beyond map->l_lblk | |
3390 | * that are allocated and initialized. | |
3391 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3392 | */ |
725d26d3 | 3393 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3394 | struct inode *inode, |
3395 | struct ext4_map_blocks *map, | |
dfe50809 | 3396 | struct ext4_ext_path **ppath, |
27dd4385 | 3397 | int flags) |
56055d3a | 3398 | { |
dfe50809 | 3399 | struct ext4_ext_path *path = *ppath; |
67a5da56 | 3400 | struct ext4_sb_info *sbi; |
6f91bc5f | 3401 | struct ext4_extent_header *eh; |
667eff35 YY |
3402 | struct ext4_map_blocks split_map; |
3403 | struct ext4_extent zero_ex; | |
bc2d9db4 | 3404 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3405 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3406 | unsigned int ee_len, depth, map_len = map->m_len; |
3407 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3408 | int err = 0; |
667eff35 | 3409 | int split_flag = 0; |
21ca087a DM |
3410 | |
3411 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3412 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3413 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3414 | |
67a5da56 | 3415 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3416 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3417 | inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3418 | if (eof_block < map->m_lblk + map_len) |
3419 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3420 | |
3421 | depth = ext_depth(inode); | |
6f91bc5f | 3422 | eh = path[depth].p_hdr; |
56055d3a AA |
3423 | ex = path[depth].p_ext; |
3424 | ee_block = le32_to_cpu(ex->ee_block); | |
3425 | ee_len = ext4_ext_get_actual_len(ex); | |
adb23551 | 3426 | zero_ex.ee_len = 0; |
56055d3a | 3427 | |
6f91bc5f EG |
3428 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3429 | ||
3430 | /* Pre-conditions */ | |
556615dc | 3431 | BUG_ON(!ext4_ext_is_unwritten(ex)); |
6f91bc5f | 3432 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); |
6f91bc5f EG |
3433 | |
3434 | /* | |
3435 | * Attempt to transfer newly initialized blocks from the currently | |
556615dc | 3436 | * unwritten extent to its neighbor. This is much cheaper |
6f91bc5f | 3437 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3438 | * memmove() calls. Transferring to the left is the common case in |
3439 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3440 | * followed by append writes. | |
6f91bc5f EG |
3441 | * |
3442 | * Limitations of the current logic: | |
bc2d9db4 | 3443 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3444 | * This would require removing the extent if the transfer |
3445 | * is possible. | |
bc2d9db4 | 3446 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3447 | * same extent tree node. |
3448 | */ | |
bc2d9db4 LC |
3449 | if ((map->m_lblk == ee_block) && |
3450 | /* See if we can merge left */ | |
3451 | (map_len < ee_len) && /*L1*/ | |
3452 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3453 | ext4_lblk_t prev_lblk; |
3454 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3455 | unsigned int prev_len; |
6f91bc5f | 3456 | |
bc2d9db4 LC |
3457 | abut_ex = ex - 1; |
3458 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3459 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3460 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3461 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3462 | |
3463 | /* | |
bc2d9db4 | 3464 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3465 | * upon those conditions: |
bc2d9db4 LC |
3466 | * - C1: abut_ex is initialized, |
3467 | * - C2: abut_ex is logically abutting ex, | |
3468 | * - C3: abut_ex is physically abutting ex, | |
3469 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3470 | * overflowing the (initialized) length limit. |
3471 | */ | |
556615dc | 3472 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3473 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3474 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3475 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3476 | err = ext4_ext_get_access(handle, inode, path + depth); |
3477 | if (err) | |
3478 | goto out; | |
3479 | ||
3480 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3481 | map, ex, abut_ex); |
6f91bc5f | 3482 | |
bc2d9db4 LC |
3483 | /* Shift the start of ex by 'map_len' blocks */ |
3484 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3485 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3486 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3487 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
6f91bc5f | 3488 | |
bc2d9db4 LC |
3489 | /* Extend abut_ex by 'map_len' blocks */ |
3490 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3491 | |
bc2d9db4 LC |
3492 | /* Result: number of initialized blocks past m_lblk */ |
3493 | allocated = map_len; | |
3494 | } | |
3495 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3496 | (map_len < ee_len) && /*L1*/ | |
3497 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3498 | /* See if we can merge right */ | |
3499 | ext4_lblk_t next_lblk; | |
3500 | ext4_fsblk_t next_pblk, ee_pblk; | |
3501 | unsigned int next_len; | |
3502 | ||
3503 | abut_ex = ex + 1; | |
3504 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3505 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3506 | next_pblk = ext4_ext_pblock(abut_ex); | |
3507 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3508 | |
bc2d9db4 LC |
3509 | /* |
3510 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3511 | * upon those conditions: | |
3512 | * - C1: abut_ex is initialized, | |
3513 | * - C2: abut_ex is logically abutting ex, | |
3514 | * - C3: abut_ex is physically abutting ex, | |
3515 | * - C4: abut_ex can receive the additional blocks without | |
3516 | * overflowing the (initialized) length limit. | |
3517 | */ | |
556615dc | 3518 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
bc2d9db4 LC |
3519 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ |
3520 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3521 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3522 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3523 | if (err) | |
3524 | goto out; | |
3525 | ||
3526 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3527 | map, ex, abut_ex); | |
3528 | ||
3529 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3530 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3531 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3532 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3533 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
bc2d9db4 LC |
3534 | |
3535 | /* Extend abut_ex by 'map_len' blocks */ | |
3536 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3537 | |
3538 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3539 | allocated = map_len; |
6f91bc5f EG |
3540 | } |
3541 | } | |
bc2d9db4 LC |
3542 | if (allocated) { |
3543 | /* Mark the block containing both extents as dirty */ | |
3544 | ext4_ext_dirty(handle, inode, path + depth); | |
3545 | ||
3546 | /* Update path to point to the right extent */ | |
3547 | path[depth].p_ext = abut_ex; | |
3548 | goto out; | |
3549 | } else | |
3550 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3551 | |
667eff35 | 3552 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3553 | /* |
3554 | * It is safe to convert extent to initialized via explicit | |
9e740568 | 3555 | * zeroout only if extent is fully inside i_size or new_size. |
21ca087a | 3556 | */ |
667eff35 | 3557 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3558 | |
67a5da56 ZL |
3559 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3560 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3561 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 ZL |
3562 | |
3563 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3564 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3565 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3566 | if (err) |
d03856bd | 3567 | goto out; |
adb23551 | 3568 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 | 3569 | zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)); |
adb23551 | 3570 | ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex)); |
d03856bd AK |
3571 | |
3572 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3573 | if (err) | |
3574 | goto out; | |
667eff35 | 3575 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3576 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3577 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3578 | goto out; |
56055d3a | 3579 | } |
667eff35 | 3580 | |
56055d3a | 3581 | /* |
667eff35 YY |
3582 | * four cases: |
3583 | * 1. split the extent into three extents. | |
3584 | * 2. split the extent into two extents, zeroout the first half. | |
3585 | * 3. split the extent into two extents, zeroout the second half. | |
3586 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3587 | */ |
667eff35 YY |
3588 | split_map.m_lblk = map->m_lblk; |
3589 | split_map.m_len = map->m_len; | |
3590 | ||
67a5da56 ZL |
3591 | if (max_zeroout && (allocated > map->m_len)) { |
3592 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3593 | /* case 3 */ |
3594 | zero_ex.ee_block = | |
9b940f8e AH |
3595 | cpu_to_le32(map->m_lblk); |
3596 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3597 | ext4_ext_store_pblock(&zero_ex, |
3598 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3599 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3600 | if (err) |
3601 | goto out; | |
667eff35 YY |
3602 | split_map.m_lblk = map->m_lblk; |
3603 | split_map.m_len = allocated; | |
67a5da56 | 3604 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3605 | /* case 2 */ |
3606 | if (map->m_lblk != ee_block) { | |
3607 | zero_ex.ee_block = ex->ee_block; | |
3608 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3609 | ee_block); | |
3610 | ext4_ext_store_pblock(&zero_ex, | |
3611 | ext4_ext_pblock(ex)); | |
3612 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3613 | if (err) | |
3614 | goto out; | |
3615 | } | |
3616 | ||
667eff35 | 3617 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3618 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3619 | allocated = map->m_len; | |
56055d3a AA |
3620 | } |
3621 | } | |
667eff35 | 3622 | |
ae9e9c6a JK |
3623 | err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag, |
3624 | flags); | |
3625 | if (err > 0) | |
3626 | err = 0; | |
56055d3a | 3627 | out: |
adb23551 ZL |
3628 | /* If we have gotten a failure, don't zero out status tree */ |
3629 | if (!err) | |
d7b2a00c | 3630 | err = ext4_zeroout_es(inode, &zero_ex); |
56055d3a AA |
3631 | return err ? err : allocated; |
3632 | } | |
3633 | ||
0031462b | 3634 | /* |
e35fd660 | 3635 | * This function is called by ext4_ext_map_blocks() from |
0031462b | 3636 | * ext4_get_blocks_dio_write() when DIO to write |
556615dc | 3637 | * to an unwritten extent. |
0031462b | 3638 | * |
556615dc LC |
3639 | * Writing to an unwritten extent may result in splitting the unwritten |
3640 | * extent into multiple initialized/unwritten extents (up to three) | |
0031462b | 3641 | * There are three possibilities: |
556615dc | 3642 | * a> There is no split required: Entire extent should be unwritten |
0031462b MC |
3643 | * b> Splits in two extents: Write is happening at either end of the extent |
3644 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3645 | * | |
b8a86845 LC |
3646 | * This works the same way in the case of initialized -> unwritten conversion. |
3647 | * | |
0031462b | 3648 | * One of more index blocks maybe needed if the extent tree grow after |
556615dc LC |
3649 | * the unwritten extent split. To prevent ENOSPC occur at the IO |
3650 | * complete, we need to split the unwritten extent before DIO submit | |
3651 | * the IO. The unwritten extent called at this time will be split | |
3652 | * into three unwritten extent(at most). After IO complete, the part | |
0031462b MC |
3653 | * being filled will be convert to initialized by the end_io callback function |
3654 | * via ext4_convert_unwritten_extents(). | |
ba230c3f | 3655 | * |
556615dc | 3656 | * Returns the size of unwritten extent to be written on success. |
0031462b | 3657 | */ |
b8a86845 | 3658 | static int ext4_split_convert_extents(handle_t *handle, |
0031462b | 3659 | struct inode *inode, |
e35fd660 | 3660 | struct ext4_map_blocks *map, |
dfe50809 | 3661 | struct ext4_ext_path **ppath, |
0031462b MC |
3662 | int flags) |
3663 | { | |
dfe50809 | 3664 | struct ext4_ext_path *path = *ppath; |
667eff35 YY |
3665 | ext4_lblk_t eof_block; |
3666 | ext4_lblk_t ee_block; | |
3667 | struct ext4_extent *ex; | |
3668 | unsigned int ee_len; | |
3669 | int split_flag = 0, depth; | |
21ca087a | 3670 | |
b8a86845 LC |
3671 | ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n", |
3672 | __func__, inode->i_ino, | |
3673 | (unsigned long long)map->m_lblk, map->m_len); | |
21ca087a DM |
3674 | |
3675 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3676 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3677 | if (eof_block < map->m_lblk + map->m_len) |
3678 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3679 | /* |
3680 | * It is safe to convert extent to initialized via explicit | |
3681 | * zeroout only if extent is fully insde i_size or new_size. | |
3682 | */ | |
667eff35 YY |
3683 | depth = ext_depth(inode); |
3684 | ex = path[depth].p_ext; | |
3685 | ee_block = le32_to_cpu(ex->ee_block); | |
3686 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3687 | |
b8a86845 LC |
3688 | /* Convert to unwritten */ |
3689 | if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) { | |
3690 | split_flag |= EXT4_EXT_DATA_VALID1; | |
3691 | /* Convert to initialized */ | |
3692 | } else if (flags & EXT4_GET_BLOCKS_CONVERT) { | |
3693 | split_flag |= ee_block + ee_len <= eof_block ? | |
3694 | EXT4_EXT_MAY_ZEROOUT : 0; | |
556615dc | 3695 | split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2); |
b8a86845 | 3696 | } |
667eff35 | 3697 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
dfe50809 | 3698 | return ext4_split_extent(handle, inode, ppath, map, split_flag, flags); |
0031462b | 3699 | } |
197217a5 | 3700 | |
c7064ef1 | 3701 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3702 | struct inode *inode, |
3703 | struct ext4_map_blocks *map, | |
dfe50809 | 3704 | struct ext4_ext_path **ppath) |
0031462b | 3705 | { |
dfe50809 | 3706 | struct ext4_ext_path *path = *ppath; |
0031462b | 3707 | struct ext4_extent *ex; |
dee1f973 DM |
3708 | ext4_lblk_t ee_block; |
3709 | unsigned int ee_len; | |
0031462b MC |
3710 | int depth; |
3711 | int err = 0; | |
0031462b MC |
3712 | |
3713 | depth = ext_depth(inode); | |
0031462b | 3714 | ex = path[depth].p_ext; |
dee1f973 DM |
3715 | ee_block = le32_to_cpu(ex->ee_block); |
3716 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3717 | |
197217a5 YY |
3718 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3719 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3720 | (unsigned long long)ee_block, ee_len); |
3721 | ||
ff95ec22 DM |
3722 | /* If extent is larger than requested it is a clear sign that we still |
3723 | * have some extent state machine issues left. So extent_split is still | |
3724 | * required. | |
3725 | * TODO: Once all related issues will be fixed this situation should be | |
3726 | * illegal. | |
3727 | */ | |
dee1f973 | 3728 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ff95ec22 DM |
3729 | #ifdef EXT4_DEBUG |
3730 | ext4_warning("Inode (%ld) finished: extent logical block %llu," | |
3731 | " len %u; IO logical block %llu, len %u\n", | |
3732 | inode->i_ino, (unsigned long long)ee_block, ee_len, | |
3733 | (unsigned long long)map->m_lblk, map->m_len); | |
3734 | #endif | |
dfe50809 | 3735 | err = ext4_split_convert_extents(handle, inode, map, ppath, |
b8a86845 | 3736 | EXT4_GET_BLOCKS_CONVERT); |
dee1f973 | 3737 | if (err < 0) |
dfe50809 | 3738 | return err; |
ed8a1a76 | 3739 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
dfe50809 TT |
3740 | if (IS_ERR(path)) |
3741 | return PTR_ERR(path); | |
dee1f973 DM |
3742 | depth = ext_depth(inode); |
3743 | ex = path[depth].p_ext; | |
3744 | } | |
197217a5 | 3745 | |
0031462b MC |
3746 | err = ext4_ext_get_access(handle, inode, path + depth); |
3747 | if (err) | |
3748 | goto out; | |
3749 | /* first mark the extent as initialized */ | |
3750 | ext4_ext_mark_initialized(ex); | |
3751 | ||
197217a5 YY |
3752 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3753 | * borders are not changed | |
0031462b | 3754 | */ |
ecb94f5f | 3755 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3756 | |
0031462b | 3757 | /* Mark modified extent as dirty */ |
ecb94f5f | 3758 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3759 | out: |
3760 | ext4_ext_show_leaf(inode, path); | |
3761 | return err; | |
3762 | } | |
3763 | ||
515f41c3 AK |
3764 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3765 | sector_t block, int count) | |
3766 | { | |
3767 | int i; | |
3768 | for (i = 0; i < count; i++) | |
3769 | unmap_underlying_metadata(bdev, block + i); | |
3770 | } | |
3771 | ||
58590b06 TT |
3772 | /* |
3773 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3774 | */ | |
3775 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3776 | ext4_lblk_t lblk, |
58590b06 TT |
3777 | struct ext4_ext_path *path, |
3778 | unsigned int len) | |
3779 | { | |
3780 | int i, depth; | |
3781 | struct ext4_extent_header *eh; | |
65922cb5 | 3782 | struct ext4_extent *last_ex; |
58590b06 TT |
3783 | |
3784 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3785 | return 0; | |
3786 | ||
3787 | depth = ext_depth(inode); | |
3788 | eh = path[depth].p_hdr; | |
58590b06 | 3789 | |
afcff5d8 LC |
3790 | /* |
3791 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3792 | * do not care for this case anymore. Simply remove the flag | |
3793 | * if there are no extents. | |
3794 | */ | |
3795 | if (unlikely(!eh->eh_entries)) | |
3796 | goto out; | |
58590b06 TT |
3797 | last_ex = EXT_LAST_EXTENT(eh); |
3798 | /* | |
3799 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3800 | * last block in the last extent in the file. We test this by | |
3801 | * first checking to see if the caller to | |
3802 | * ext4_ext_get_blocks() was interested in the last block (or | |
3803 | * a block beyond the last block) in the current extent. If | |
3804 | * this turns out to be false, we can bail out from this | |
3805 | * function immediately. | |
3806 | */ | |
d002ebf1 | 3807 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3808 | ext4_ext_get_actual_len(last_ex)) |
3809 | return 0; | |
3810 | /* | |
3811 | * If the caller does appear to be planning to write at or | |
3812 | * beyond the end of the current extent, we then test to see | |
3813 | * if the current extent is the last extent in the file, by | |
3814 | * checking to make sure it was reached via the rightmost node | |
3815 | * at each level of the tree. | |
3816 | */ | |
3817 | for (i = depth-1; i >= 0; i--) | |
3818 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3819 | return 0; | |
afcff5d8 | 3820 | out: |
58590b06 TT |
3821 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3822 | return ext4_mark_inode_dirty(handle, inode); | |
3823 | } | |
3824 | ||
7b415bf6 AK |
3825 | /** |
3826 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3827 | * | |
7d1b1fbc | 3828 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3829 | */ |
f7fec032 ZL |
3830 | int ext4_find_delalloc_range(struct inode *inode, |
3831 | ext4_lblk_t lblk_start, | |
3832 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3833 | { |
7d1b1fbc | 3834 | struct extent_status es; |
7b415bf6 | 3835 | |
e30b5dca | 3836 | ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es); |
06b0c886 | 3837 | if (es.es_len == 0) |
7d1b1fbc | 3838 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3839 | else if (es.es_lblk <= lblk_start && |
3840 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3841 | return 1; |
06b0c886 | 3842 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3843 | return 1; |
7b415bf6 | 3844 | else |
7d1b1fbc | 3845 | return 0; |
7b415bf6 AK |
3846 | } |
3847 | ||
7d1b1fbc | 3848 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3849 | { |
3850 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3851 | ext4_lblk_t lblk_start, lblk_end; | |
f5a44db5 | 3852 | lblk_start = EXT4_LBLK_CMASK(sbi, lblk); |
7b415bf6 AK |
3853 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; |
3854 | ||
7d1b1fbc | 3855 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3856 | } |
3857 | ||
3858 | /** | |
3859 | * Determines how many complete clusters (out of those specified by the 'map') | |
3860 | * are under delalloc and were reserved quota for. | |
3861 | * This function is called when we are writing out the blocks that were | |
3862 | * originally written with their allocation delayed, but then the space was | |
3863 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3864 | * The cases to look for are: | |
3865 | * ('=' indicated delayed allocated blocks | |
3866 | * '-' indicates non-delayed allocated blocks) | |
3867 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3868 | * are not delalloc'ed. | |
3869 | * Ex: | |
3870 | * |----c---=|====c====|====c====|===-c----| | |
3871 | * |++++++ allocated ++++++| | |
3872 | * ==> 4 complete clusters in above example | |
3873 | * | |
3874 | * (b) partial cluster (outside of allocated range) towards either end is | |
3875 | * marked for delayed allocation. In this case, we will exclude that | |
3876 | * cluster. | |
3877 | * Ex: | |
3878 | * |----====c========|========c========| | |
3879 | * |++++++ allocated ++++++| | |
3880 | * ==> 1 complete clusters in above example | |
3881 | * | |
3882 | * Ex: | |
3883 | * |================c================| | |
3884 | * |++++++ allocated ++++++| | |
3885 | * ==> 0 complete clusters in above example | |
3886 | * | |
3887 | * The ext4_da_update_reserve_space will be called only if we | |
3888 | * determine here that there were some "entire" clusters that span | |
3889 | * this 'allocated' range. | |
3890 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3891 | * without ever calling ext4_find_delalloc_range. | |
3892 | */ | |
3893 | static unsigned int | |
3894 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3895 | unsigned int num_blks) | |
3896 | { | |
3897 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3898 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3899 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3900 | unsigned int allocated_clusters = 0; | |
3901 | ||
3902 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3903 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3904 | ||
3905 | /* max possible clusters for this allocation */ | |
3906 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3907 | ||
d8990240 AK |
3908 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3909 | ||
7b415bf6 | 3910 | /* Check towards left side */ |
f5a44db5 | 3911 | c_offset = EXT4_LBLK_COFF(sbi, lblk_start); |
7b415bf6 | 3912 | if (c_offset) { |
f5a44db5 | 3913 | lblk_from = EXT4_LBLK_CMASK(sbi, lblk_start); |
7b415bf6 AK |
3914 | lblk_to = lblk_from + c_offset - 1; |
3915 | ||
7d1b1fbc | 3916 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3917 | allocated_clusters--; |
3918 | } | |
3919 | ||
3920 | /* Now check towards right. */ | |
f5a44db5 | 3921 | c_offset = EXT4_LBLK_COFF(sbi, lblk_start + num_blks); |
7b415bf6 AK |
3922 | if (allocated_clusters && c_offset) { |
3923 | lblk_from = lblk_start + num_blks; | |
3924 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3925 | ||
7d1b1fbc | 3926 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3927 | allocated_clusters--; |
3928 | } | |
3929 | ||
3930 | return allocated_clusters; | |
3931 | } | |
3932 | ||
b8a86845 | 3933 | static int |
e8b83d93 TT |
3934 | convert_initialized_extent(handle_t *handle, struct inode *inode, |
3935 | struct ext4_map_blocks *map, | |
4f224b8b | 3936 | struct ext4_ext_path **ppath, int flags, |
e8b83d93 | 3937 | unsigned int allocated, ext4_fsblk_t newblock) |
b8a86845 | 3938 | { |
4f224b8b | 3939 | struct ext4_ext_path *path = *ppath; |
e8b83d93 TT |
3940 | struct ext4_extent *ex; |
3941 | ext4_lblk_t ee_block; | |
3942 | unsigned int ee_len; | |
3943 | int depth; | |
b8a86845 LC |
3944 | int err = 0; |
3945 | ||
3946 | /* | |
3947 | * Make sure that the extent is no bigger than we support with | |
556615dc | 3948 | * unwritten extent |
b8a86845 | 3949 | */ |
556615dc LC |
3950 | if (map->m_len > EXT_UNWRITTEN_MAX_LEN) |
3951 | map->m_len = EXT_UNWRITTEN_MAX_LEN / 2; | |
b8a86845 | 3952 | |
e8b83d93 TT |
3953 | depth = ext_depth(inode); |
3954 | ex = path[depth].p_ext; | |
3955 | ee_block = le32_to_cpu(ex->ee_block); | |
3956 | ee_len = ext4_ext_get_actual_len(ex); | |
3957 | ||
3958 | ext_debug("%s: inode %lu, logical" | |
3959 | "block %llu, max_blocks %u\n", __func__, inode->i_ino, | |
3960 | (unsigned long long)ee_block, ee_len); | |
3961 | ||
3962 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
dfe50809 | 3963 | err = ext4_split_convert_extents(handle, inode, map, ppath, |
e8b83d93 TT |
3964 | EXT4_GET_BLOCKS_CONVERT_UNWRITTEN); |
3965 | if (err < 0) | |
3966 | return err; | |
ed8a1a76 | 3967 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
e8b83d93 TT |
3968 | if (IS_ERR(path)) |
3969 | return PTR_ERR(path); | |
3970 | depth = ext_depth(inode); | |
3971 | ex = path[depth].p_ext; | |
3972 | if (!ex) { | |
3973 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3974 | (unsigned long) map->m_lblk); | |
3975 | return -EIO; | |
3976 | } | |
3977 | } | |
3978 | ||
3979 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3980 | if (err) | |
3981 | return err; | |
3982 | /* first mark the extent as unwritten */ | |
3983 | ext4_ext_mark_unwritten(ex); | |
3984 | ||
3985 | /* note: ext4_ext_correct_indexes() isn't needed here because | |
3986 | * borders are not changed | |
3987 | */ | |
3988 | ext4_ext_try_to_merge(handle, inode, path, ex); | |
3989 | ||
3990 | /* Mark modified extent as dirty */ | |
3991 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
3992 | if (err) | |
3993 | return err; | |
3994 | ext4_ext_show_leaf(inode, path); | |
3995 | ||
3996 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
3997 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len); | |
3998 | if (err) | |
3999 | return err; | |
b8a86845 LC |
4000 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
4001 | if (allocated > map->m_len) | |
4002 | allocated = map->m_len; | |
4003 | map->m_len = allocated; | |
e8b83d93 | 4004 | return allocated; |
b8a86845 LC |
4005 | } |
4006 | ||
0031462b | 4007 | static int |
556615dc | 4008 | ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode, |
e35fd660 | 4009 | struct ext4_map_blocks *map, |
dfe50809 | 4010 | struct ext4_ext_path **ppath, int flags, |
e35fd660 | 4011 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b | 4012 | { |
dfe50809 | 4013 | struct ext4_ext_path *path = *ppath; |
0031462b MC |
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) { |
dfe50809 TT |
4035 | ret = ext4_split_convert_extents(handle, inode, map, ppath, |
4036 | 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, |
dfe50809 | 4054 | ppath); |
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*/ | |
dfe50809 | 4092 | ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, 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; |
cbd7584e | 4287 | bool map_from_cluster = false; |
a86c6181 | 4288 | |
84fe3bef | 4289 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4290 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4291 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4292 | |
a86c6181 | 4293 | /* find extent for this block */ |
ed8a1a76 | 4294 | path = ext4_find_extent(inode, map->m_lblk, NULL, 0); |
a86c6181 AT |
4295 | if (IS_ERR(path)) { |
4296 | err = PTR_ERR(path); | |
4297 | path = NULL; | |
4298 | goto out2; | |
4299 | } | |
4300 | ||
4301 | depth = ext_depth(inode); | |
4302 | ||
4303 | /* | |
d0d856e8 RD |
4304 | * consistent leaf must not be empty; |
4305 | * this situation is possible, though, _during_ tree modification; | |
ed8a1a76 | 4306 | * this is why assert can't be put in ext4_find_extent() |
a86c6181 | 4307 | */ |
273df556 FM |
4308 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4309 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4310 | "lblock: %lu, depth: %d pblock %lld", |
4311 | (unsigned long) map->m_lblk, depth, | |
4312 | path[depth].p_block); | |
034fb4c9 SP |
4313 | err = -EIO; |
4314 | goto out2; | |
4315 | } | |
a86c6181 | 4316 | |
7e028976 AM |
4317 | ex = path[depth].p_ext; |
4318 | if (ex) { | |
725d26d3 | 4319 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4320 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4321 | unsigned short ee_len; |
471d4011 | 4322 | |
b8a86845 | 4323 | |
471d4011 | 4324 | /* |
556615dc | 4325 | * unwritten extents are treated as holes, except that |
56055d3a | 4326 | * we split out initialized portions during a write. |
471d4011 | 4327 | */ |
a2df2a63 | 4328 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4329 | |
4330 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4331 | ||
d0d856e8 | 4332 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4333 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4334 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4335 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4336 | allocated = ee_len - (map->m_lblk - ee_block); |
4337 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4338 | ee_block, ee_len, newblock); | |
56055d3a | 4339 | |
b8a86845 LC |
4340 | /* |
4341 | * If the extent is initialized check whether the | |
4342 | * caller wants to convert it to unwritten. | |
4343 | */ | |
556615dc | 4344 | if ((!ext4_ext_is_unwritten(ex)) && |
b8a86845 | 4345 | (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) { |
e8b83d93 | 4346 | allocated = convert_initialized_extent( |
4f224b8b TT |
4347 | handle, inode, map, &path, |
4348 | flags, allocated, newblock); | |
b8a86845 | 4349 | goto out2; |
556615dc | 4350 | } else if (!ext4_ext_is_unwritten(ex)) |
7877191c | 4351 | goto out; |
69eb33dc | 4352 | |
556615dc | 4353 | ret = ext4_ext_handle_unwritten_extents( |
dfe50809 | 4354 | handle, inode, map, &path, flags, |
7877191c | 4355 | allocated, newblock); |
ce37c429 EW |
4356 | if (ret < 0) |
4357 | err = ret; | |
4358 | else | |
4359 | allocated = ret; | |
31cf0f2c | 4360 | goto out2; |
a86c6181 AT |
4361 | } |
4362 | } | |
4363 | ||
4364 | /* | |
d0d856e8 | 4365 | * requested block isn't allocated yet; |
a86c6181 AT |
4366 | * we couldn't try to create block if create flag is zero |
4367 | */ | |
c2177057 | 4368 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4369 | /* |
4370 | * put just found gap into cache to speed up | |
4371 | * subsequent requests | |
4372 | */ | |
2f8e0a7c | 4373 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); |
a86c6181 AT |
4374 | goto out2; |
4375 | } | |
4d33b1ef | 4376 | |
a86c6181 | 4377 | /* |
c2ea3fde | 4378 | * Okay, we need to do block allocation. |
63f57933 | 4379 | */ |
4d33b1ef | 4380 | newex.ee_block = cpu_to_le32(map->m_lblk); |
d0abafac | 4381 | cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4382 | |
4383 | /* | |
4384 | * If we are doing bigalloc, check to see if the extent returned | |
ed8a1a76 | 4385 | * by ext4_find_extent() implies a cluster we can use. |
4d33b1ef TT |
4386 | */ |
4387 | if (cluster_offset && ex && | |
d8990240 | 4388 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4389 | ar.len = allocated = map->m_len; |
4390 | newblock = map->m_pblk; | |
cbd7584e | 4391 | map_from_cluster = true; |
4d33b1ef TT |
4392 | goto got_allocated_blocks; |
4393 | } | |
a86c6181 | 4394 | |
c9de560d | 4395 | /* find neighbour allocated blocks */ |
e35fd660 | 4396 | ar.lleft = map->m_lblk; |
c9de560d AT |
4397 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4398 | if (err) | |
4399 | goto out2; | |
e35fd660 | 4400 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4401 | ex2 = NULL; |
4402 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4403 | if (err) |
4404 | goto out2; | |
25d14f98 | 4405 | |
4d33b1ef TT |
4406 | /* Check if the extent after searching to the right implies a |
4407 | * cluster we can use. */ | |
4408 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4409 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4410 | ar.len = allocated = map->m_len; |
4411 | newblock = map->m_pblk; | |
cbd7584e | 4412 | map_from_cluster = true; |
4d33b1ef TT |
4413 | goto got_allocated_blocks; |
4414 | } | |
4415 | ||
749269fa AA |
4416 | /* |
4417 | * See if request is beyond maximum number of blocks we can have in | |
4418 | * a single extent. For an initialized extent this limit is | |
556615dc LC |
4419 | * EXT_INIT_MAX_LEN and for an unwritten extent this limit is |
4420 | * EXT_UNWRITTEN_MAX_LEN. | |
749269fa | 4421 | */ |
e35fd660 | 4422 | if (map->m_len > EXT_INIT_MAX_LEN && |
556615dc | 4423 | !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) |
e35fd660 | 4424 | map->m_len = EXT_INIT_MAX_LEN; |
556615dc LC |
4425 | else if (map->m_len > EXT_UNWRITTEN_MAX_LEN && |
4426 | (flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) | |
4427 | map->m_len = EXT_UNWRITTEN_MAX_LEN; | |
749269fa | 4428 | |
e35fd660 | 4429 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4430 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4431 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4432 | if (err) |
b939e376 | 4433 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4434 | else |
e35fd660 | 4435 | allocated = map->m_len; |
c9de560d AT |
4436 | |
4437 | /* allocate new block */ | |
4438 | ar.inode = inode; | |
e35fd660 TT |
4439 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4440 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4441 | /* |
4442 | * We calculate the offset from the beginning of the cluster | |
4443 | * for the logical block number, since when we allocate a | |
4444 | * physical cluster, the physical block should start at the | |
4445 | * same offset from the beginning of the cluster. This is | |
4446 | * needed so that future calls to get_implied_cluster_alloc() | |
4447 | * work correctly. | |
4448 | */ | |
f5a44db5 | 4449 | offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4450 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); |
4451 | ar.goal -= offset; | |
4452 | ar.logical -= offset; | |
c9de560d AT |
4453 | if (S_ISREG(inode->i_mode)) |
4454 | ar.flags = EXT4_MB_HINT_DATA; | |
4455 | else | |
4456 | /* disable in-core preallocation for non-regular files */ | |
4457 | ar.flags = 0; | |
556b27ab VH |
4458 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4459 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
e3cf5d5d TT |
4460 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
4461 | ar.flags |= EXT4_MB_DELALLOC_RESERVED; | |
c9de560d | 4462 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4463 | if (!newblock) |
4464 | goto out2; | |
84fe3bef | 4465 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4466 | ar.goal, newblock, allocated); |
4d33b1ef | 4467 | free_on_err = 1; |
7b415bf6 | 4468 | allocated_clusters = ar.len; |
4d33b1ef TT |
4469 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4470 | if (ar.len > allocated) | |
4471 | ar.len = allocated; | |
a86c6181 | 4472 | |
4d33b1ef | 4473 | got_allocated_blocks: |
a86c6181 | 4474 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4475 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4476 | newex.ee_len = cpu_to_le16(ar.len); |
556615dc LC |
4477 | /* Mark unwritten */ |
4478 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){ | |
4479 | ext4_ext_mark_unwritten(&newex); | |
a25a4e1a | 4480 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4481 | /* |
744692dc | 4482 | * io_end structure was created for every IO write to an |
556615dc | 4483 | * unwritten extent. To avoid unnecessary conversion, |
744692dc | 4484 | * here we flag the IO that really needs the conversion. |
5f524950 | 4485 | * For non asycn direct IO case, flag the inode state |
25985edc | 4486 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4487 | */ |
c8b459f4 | 4488 | if (flags & EXT4_GET_BLOCKS_PRE_IO) |
82e54229 | 4489 | set_unwritten = 1; |
8d5d02e6 | 4490 | } |
c8d46e41 | 4491 | |
a4e5d88b DM |
4492 | err = 0; |
4493 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4494 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4495 | path, ar.len); | |
575a1d4b | 4496 | if (!err) |
dfe50809 | 4497 | err = ext4_ext_insert_extent(handle, inode, &path, |
575a1d4b | 4498 | &newex, flags); |
82e54229 DM |
4499 | |
4500 | if (!err && set_unwritten) { | |
4501 | if (io) | |
4502 | ext4_set_io_unwritten_flag(inode, io); | |
4503 | else | |
4504 | ext4_set_inode_state(inode, | |
4505 | EXT4_STATE_DIO_UNWRITTEN); | |
4506 | } | |
4507 | ||
4d33b1ef | 4508 | if (err && free_on_err) { |
7132de74 MP |
4509 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4510 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4511 | /* free data blocks we just allocated */ |
c9de560d AT |
4512 | /* not a good idea to call discard here directly, |
4513 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4514 | ext4_discard_preallocations(inode); |
c8e15130 TT |
4515 | ext4_free_blocks(handle, inode, NULL, newblock, |
4516 | EXT4_C2B(sbi, allocated_clusters), fb_flags); | |
a86c6181 | 4517 | goto out2; |
315054f0 | 4518 | } |
a86c6181 | 4519 | |
a86c6181 | 4520 | /* previous routine could use block we allocated */ |
bf89d16f | 4521 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4522 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4523 | if (allocated > map->m_len) |
4524 | allocated = map->m_len; | |
4525 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4526 | |
5f634d06 AK |
4527 | /* |
4528 | * Update reserved blocks/metadata blocks after successful | |
4529 | * block allocation which had been deferred till now. | |
4530 | */ | |
7b415bf6 | 4531 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4532 | unsigned int reserved_clusters; |
7b415bf6 | 4533 | /* |
81fdbb4a | 4534 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4535 | */ |
4536 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4537 | map->m_lblk, allocated); | |
cbd7584e | 4538 | if (map_from_cluster) { |
7b415bf6 AK |
4539 | if (reserved_clusters) { |
4540 | /* | |
4541 | * We have clusters reserved for this range. | |
4542 | * But since we are not doing actual allocation | |
4543 | * and are simply using blocks from previously | |
4544 | * allocated cluster, we should release the | |
4545 | * reservation and not claim quota. | |
4546 | */ | |
4547 | ext4_da_update_reserve_space(inode, | |
4548 | reserved_clusters, 0); | |
4549 | } | |
4550 | } else { | |
4551 | BUG_ON(allocated_clusters < reserved_clusters); | |
7b415bf6 | 4552 | if (reserved_clusters < allocated_clusters) { |
5356f261 | 4553 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4554 | int reservation = allocated_clusters - |
4555 | reserved_clusters; | |
4556 | /* | |
4557 | * It seems we claimed few clusters outside of | |
4558 | * the range of this allocation. We should give | |
4559 | * it back to the reservation pool. This can | |
4560 | * happen in the following case: | |
4561 | * | |
4562 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4563 | * cluster has 4 blocks. Thus, the clusters | |
4564 | * are [0-3],[4-7],[8-11]... | |
4565 | * * First comes delayed allocation write for | |
4566 | * logical blocks 10 & 11. Since there were no | |
4567 | * previous delayed allocated blocks in the | |
4568 | * range [8-11], we would reserve 1 cluster | |
4569 | * for this write. | |
4570 | * * Next comes write for logical blocks 3 to 8. | |
4571 | * In this case, we will reserve 2 clusters | |
4572 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4573 | * that range has a delayed allocated blocks. | |
4574 | * Thus total reserved clusters now becomes 3. | |
4575 | * * Now, during the delayed allocation writeout | |
4576 | * time, we will first write blocks [3-8] and | |
4577 | * allocate 3 clusters for writing these | |
4578 | * blocks. Also, we would claim all these | |
4579 | * three clusters above. | |
4580 | * * Now when we come here to writeout the | |
4581 | * blocks [10-11], we would expect to claim | |
4582 | * the reservation of 1 cluster we had made | |
4583 | * (and we would claim it since there are no | |
4584 | * more delayed allocated blocks in the range | |
4585 | * [8-11]. But our reserved cluster count had | |
4586 | * already gone to 0. | |
4587 | * | |
4588 | * Thus, at the step 4 above when we determine | |
4589 | * that there are still some unwritten delayed | |
4590 | * allocated blocks outside of our current | |
4591 | * block range, we should increment the | |
4592 | * reserved clusters count so that when the | |
4593 | * remaining blocks finally gets written, we | |
4594 | * could claim them. | |
4595 | */ | |
5356f261 AK |
4596 | dquot_reserve_block(inode, |
4597 | EXT4_C2B(sbi, reservation)); | |
4598 | spin_lock(&ei->i_block_reservation_lock); | |
4599 | ei->i_reserved_data_blocks += reservation; | |
4600 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 | 4601 | } |
232ec872 LC |
4602 | /* |
4603 | * We will claim quota for all newly allocated blocks. | |
4604 | * We're updating the reserved space *after* the | |
4605 | * correction above so we do not accidentally free | |
4606 | * all the metadata reservation because we might | |
4607 | * actually need it later on. | |
4608 | */ | |
4609 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4610 | 1); | |
7b415bf6 AK |
4611 | } |
4612 | } | |
5f634d06 | 4613 | |
b436b9be JK |
4614 | /* |
4615 | * Cache the extent and update transaction to commit on fdatasync only | |
556615dc | 4616 | * when it is _not_ an unwritten extent. |
b436b9be | 4617 | */ |
556615dc | 4618 | if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0) |
b436b9be | 4619 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4620 | else |
b436b9be | 4621 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4622 | out: |
e35fd660 TT |
4623 | if (allocated > map->m_len) |
4624 | allocated = map->m_len; | |
a86c6181 | 4625 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4626 | map->m_flags |= EXT4_MAP_MAPPED; |
4627 | map->m_pblk = newblock; | |
4628 | map->m_len = allocated; | |
a86c6181 | 4629 | out2: |
b7ea89ad TT |
4630 | ext4_ext_drop_refs(path); |
4631 | kfree(path); | |
e861304b | 4632 | |
63b99968 TT |
4633 | trace_ext4_ext_map_blocks_exit(inode, flags, map, |
4634 | err ? err : allocated); | |
7877191c | 4635 | return err ? err : allocated; |
a86c6181 AT |
4636 | } |
4637 | ||
819c4920 | 4638 | void ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4639 | { |
a86c6181 | 4640 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4641 | ext4_lblk_t last_block; |
a86c6181 AT |
4642 | int err = 0; |
4643 | ||
a86c6181 | 4644 | /* |
d0d856e8 RD |
4645 | * TODO: optimization is possible here. |
4646 | * Probably we need not scan at all, | |
4647 | * because page truncation is enough. | |
a86c6181 | 4648 | */ |
a86c6181 AT |
4649 | |
4650 | /* we have to know where to truncate from in crash case */ | |
4651 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4652 | ext4_mark_inode_dirty(handle, inode); | |
4653 | ||
4654 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4655 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
8acd5e9b | 4656 | retry: |
51865fda ZL |
4657 | err = ext4_es_remove_extent(inode, last_block, |
4658 | EXT_MAX_BLOCKS - last_block); | |
94eec0fc | 4659 | if (err == -ENOMEM) { |
8acd5e9b TT |
4660 | cond_resched(); |
4661 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
4662 | goto retry; | |
4663 | } | |
4664 | if (err) { | |
4665 | ext4_std_error(inode->i_sb, err); | |
4666 | return; | |
4667 | } | |
5f95d21f | 4668 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
8acd5e9b | 4669 | ext4_std_error(inode->i_sb, err); |
a86c6181 AT |
4670 | } |
4671 | ||
0e8b6879 | 4672 | static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset, |
c174e6d6 DM |
4673 | ext4_lblk_t len, loff_t new_size, |
4674 | int flags, int mode) | |
0e8b6879 LC |
4675 | { |
4676 | struct inode *inode = file_inode(file); | |
4677 | handle_t *handle; | |
4678 | int ret = 0; | |
4679 | int ret2 = 0; | |
4680 | int retries = 0; | |
4681 | struct ext4_map_blocks map; | |
4682 | unsigned int credits; | |
c174e6d6 | 4683 | loff_t epos; |
0e8b6879 LC |
4684 | |
4685 | map.m_lblk = offset; | |
c174e6d6 | 4686 | map.m_len = len; |
0e8b6879 LC |
4687 | /* |
4688 | * Don't normalize the request if it can fit in one extent so | |
4689 | * that it doesn't get unnecessarily split into multiple | |
4690 | * extents. | |
4691 | */ | |
556615dc | 4692 | if (len <= EXT_UNWRITTEN_MAX_LEN) |
0e8b6879 LC |
4693 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; |
4694 | ||
4695 | /* | |
4696 | * credits to insert 1 extent into extent tree | |
4697 | */ | |
4698 | credits = ext4_chunk_trans_blocks(inode, len); | |
4699 | ||
4700 | retry: | |
c174e6d6 | 4701 | while (ret >= 0 && len) { |
0e8b6879 LC |
4702 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4703 | credits); | |
4704 | if (IS_ERR(handle)) { | |
4705 | ret = PTR_ERR(handle); | |
4706 | break; | |
4707 | } | |
4708 | ret = ext4_map_blocks(handle, inode, &map, flags); | |
4709 | if (ret <= 0) { | |
4710 | ext4_debug("inode #%lu: block %u: len %u: " | |
4711 | "ext4_ext_map_blocks returned %d", | |
4712 | inode->i_ino, map.m_lblk, | |
4713 | map.m_len, ret); | |
4714 | ext4_mark_inode_dirty(handle, inode); | |
4715 | ret2 = ext4_journal_stop(handle); | |
4716 | break; | |
4717 | } | |
c174e6d6 DM |
4718 | map.m_lblk += ret; |
4719 | map.m_len = len = len - ret; | |
4720 | epos = (loff_t)map.m_lblk << inode->i_blkbits; | |
4721 | inode->i_ctime = ext4_current_time(inode); | |
4722 | if (new_size) { | |
4723 | if (epos > new_size) | |
4724 | epos = new_size; | |
4725 | if (ext4_update_inode_size(inode, epos) & 0x1) | |
4726 | inode->i_mtime = inode->i_ctime; | |
4727 | } else { | |
4728 | if (epos > inode->i_size) | |
4729 | ext4_set_inode_flag(inode, | |
4730 | EXT4_INODE_EOFBLOCKS); | |
4731 | } | |
4732 | ext4_mark_inode_dirty(handle, inode); | |
0e8b6879 LC |
4733 | ret2 = ext4_journal_stop(handle); |
4734 | if (ret2) | |
4735 | break; | |
4736 | } | |
4737 | if (ret == -ENOSPC && | |
4738 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4739 | ret = 0; | |
4740 | goto retry; | |
4741 | } | |
4742 | ||
4743 | return ret > 0 ? ret2 : ret; | |
4744 | } | |
4745 | ||
b8a86845 LC |
4746 | static long ext4_zero_range(struct file *file, loff_t offset, |
4747 | loff_t len, int mode) | |
4748 | { | |
4749 | struct inode *inode = file_inode(file); | |
4750 | handle_t *handle = NULL; | |
4751 | unsigned int max_blocks; | |
4752 | loff_t new_size = 0; | |
4753 | int ret = 0; | |
4754 | int flags; | |
69dc9536 | 4755 | int credits; |
c174e6d6 | 4756 | int partial_begin, partial_end; |
b8a86845 LC |
4757 | loff_t start, end; |
4758 | ext4_lblk_t lblk; | |
4759 | struct address_space *mapping = inode->i_mapping; | |
4760 | unsigned int blkbits = inode->i_blkbits; | |
4761 | ||
4762 | trace_ext4_zero_range(inode, offset, len, mode); | |
4763 | ||
6c5e73d3 | 4764 | if (!S_ISREG(inode->i_mode)) |
4765 | return -EINVAL; | |
4766 | ||
e1ee60fd NJ |
4767 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
4768 | if (ext4_should_journal_data(inode)) { | |
4769 | ret = ext4_force_commit(inode->i_sb); | |
4770 | if (ret) | |
4771 | return ret; | |
4772 | } | |
4773 | ||
b8a86845 LC |
4774 | /* |
4775 | * Write out all dirty pages to avoid race conditions | |
4776 | * Then release them. | |
4777 | */ | |
4778 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
4779 | ret = filemap_write_and_wait_range(mapping, offset, | |
4780 | offset + len - 1); | |
4781 | if (ret) | |
4782 | return ret; | |
4783 | } | |
4784 | ||
4785 | /* | |
4786 | * Round up offset. This is not fallocate, we neet to zero out | |
4787 | * blocks, so convert interior block aligned part of the range to | |
4788 | * unwritten and possibly manually zero out unaligned parts of the | |
4789 | * range. | |
4790 | */ | |
4791 | start = round_up(offset, 1 << blkbits); | |
4792 | end = round_down((offset + len), 1 << blkbits); | |
4793 | ||
4794 | if (start < offset || end > offset + len) | |
4795 | return -EINVAL; | |
c174e6d6 DM |
4796 | partial_begin = offset & ((1 << blkbits) - 1); |
4797 | partial_end = (offset + len) & ((1 << blkbits) - 1); | |
b8a86845 LC |
4798 | |
4799 | lblk = start >> blkbits; | |
4800 | max_blocks = (end >> blkbits); | |
4801 | if (max_blocks < lblk) | |
4802 | max_blocks = 0; | |
4803 | else | |
4804 | max_blocks -= lblk; | |
4805 | ||
556615dc | 4806 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT | |
713e8dde TT |
4807 | EXT4_GET_BLOCKS_CONVERT_UNWRITTEN | |
4808 | EXT4_EX_NOCACHE; | |
b8a86845 LC |
4809 | if (mode & FALLOC_FL_KEEP_SIZE) |
4810 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4811 | ||
4812 | mutex_lock(&inode->i_mutex); | |
4813 | ||
4814 | /* | |
4815 | * Indirect files do not support unwritten extnets | |
4816 | */ | |
4817 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
4818 | ret = -EOPNOTSUPP; | |
4819 | goto out_mutex; | |
4820 | } | |
4821 | ||
4822 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4823 | offset + len > i_size_read(inode)) { | |
4824 | new_size = offset + len; | |
4825 | ret = inode_newsize_ok(inode, new_size); | |
4826 | if (ret) | |
4827 | goto out_mutex; | |
4828 | /* | |
4829 | * If we have a partial block after EOF we have to allocate | |
4830 | * the entire block. | |
4831 | */ | |
c174e6d6 | 4832 | if (partial_end) |
b8a86845 LC |
4833 | max_blocks += 1; |
4834 | } | |
4835 | ||
4836 | if (max_blocks > 0) { | |
4837 | ||
4838 | /* Now release the pages and zero block aligned part of pages*/ | |
4839 | truncate_pagecache_range(inode, start, end - 1); | |
c174e6d6 | 4840 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
b8a86845 LC |
4841 | |
4842 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
4843 | ext4_inode_block_unlocked_dio(inode); | |
4844 | inode_dio_wait(inode); | |
4845 | ||
713e8dde TT |
4846 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, |
4847 | flags, mode); | |
4848 | if (ret) | |
4849 | goto out_dio; | |
b8a86845 LC |
4850 | /* |
4851 | * Remove entire range from the extent status tree. | |
713e8dde TT |
4852 | * |
4853 | * ext4_es_remove_extent(inode, lblk, max_blocks) is | |
4854 | * NOT sufficient. I'm not sure why this is the case, | |
4855 | * but let's be conservative and remove the extent | |
4856 | * status tree for the entire inode. There should be | |
4857 | * no outstanding delalloc extents thanks to the | |
4858 | * filemap_write_and_wait_range() call above. | |
b8a86845 | 4859 | */ |
713e8dde | 4860 | ret = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS); |
b8a86845 LC |
4861 | if (ret) |
4862 | goto out_dio; | |
4863 | } | |
c174e6d6 DM |
4864 | if (!partial_begin && !partial_end) |
4865 | goto out_dio; | |
4866 | ||
69dc9536 DM |
4867 | /* |
4868 | * In worst case we have to writeout two nonadjacent unwritten | |
4869 | * blocks and update the inode | |
4870 | */ | |
4871 | credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1; | |
4872 | if (ext4_should_journal_data(inode)) | |
4873 | credits += 2; | |
4874 | handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); | |
b8a86845 LC |
4875 | if (IS_ERR(handle)) { |
4876 | ret = PTR_ERR(handle); | |
4877 | ext4_std_error(inode->i_sb, ret); | |
4878 | goto out_dio; | |
4879 | } | |
4880 | ||
4881 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
e5b30416 | 4882 | if (new_size) { |
4631dbf6 | 4883 | ext4_update_inode_size(inode, new_size); |
e5b30416 | 4884 | } else { |
b8a86845 LC |
4885 | /* |
4886 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4887 | * can proceed even if the new size is the same as i_size. | |
4888 | */ | |
4889 | if ((offset + len) > i_size_read(inode)) | |
4890 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); | |
4891 | } | |
b8a86845 LC |
4892 | ext4_mark_inode_dirty(handle, inode); |
4893 | ||
4894 | /* Zero out partial block at the edges of the range */ | |
4895 | ret = ext4_zero_partial_blocks(handle, inode, offset, len); | |
4896 | ||
4897 | if (file->f_flags & O_SYNC) | |
4898 | ext4_handle_sync(handle); | |
4899 | ||
4900 | ext4_journal_stop(handle); | |
4901 | out_dio: | |
4902 | ext4_inode_resume_unlocked_dio(inode); | |
4903 | out_mutex: | |
4904 | mutex_unlock(&inode->i_mutex); | |
4905 | return ret; | |
4906 | } | |
4907 | ||
a2df2a63 | 4908 | /* |
2fe17c10 | 4909 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4910 | * operation, which gets called from sys_fallocate system call. |
4911 | * For block-mapped files, posix_fallocate should fall back to the method | |
4912 | * of writing zeroes to the required new blocks (the same behavior which is | |
4913 | * expected for file systems which do not support fallocate() system call). | |
4914 | */ | |
2fe17c10 | 4915 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4916 | { |
496ad9aa | 4917 | struct inode *inode = file_inode(file); |
f282ac19 | 4918 | loff_t new_size = 0; |
498e5f24 | 4919 | unsigned int max_blocks; |
a2df2a63 | 4920 | int ret = 0; |
a4e5d88b | 4921 | int flags; |
0e8b6879 | 4922 | ext4_lblk_t lblk; |
0e8b6879 | 4923 | unsigned int blkbits = inode->i_blkbits; |
a2df2a63 | 4924 | |
a4bb6b64 | 4925 | /* Return error if mode is not supported */ |
9eb79482 | 4926 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
b8a86845 | 4927 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
a4bb6b64 AH |
4928 | return -EOPNOTSUPP; |
4929 | ||
4930 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
aeb2817a | 4931 | return ext4_punch_hole(inode, offset, len); |
a4bb6b64 | 4932 | |
0c8d414f TM |
4933 | ret = ext4_convert_inline_data(inode); |
4934 | if (ret) | |
4935 | return ret; | |
4936 | ||
8bad6fc8 ZL |
4937 | /* |
4938 | * currently supporting (pre)allocate mode for extent-based | |
4939 | * files _only_ | |
4940 | */ | |
4941 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4942 | return -EOPNOTSUPP; | |
4943 | ||
40c406c7 TT |
4944 | if (mode & FALLOC_FL_COLLAPSE_RANGE) |
4945 | return ext4_collapse_range(inode, offset, len); | |
4946 | ||
b8a86845 LC |
4947 | if (mode & FALLOC_FL_ZERO_RANGE) |
4948 | return ext4_zero_range(file, offset, len, mode); | |
4949 | ||
0562e0ba | 4950 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
0e8b6879 | 4951 | lblk = offset >> blkbits; |
fd28784a AK |
4952 | /* |
4953 | * We can't just convert len to max_blocks because | |
4954 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4955 | */ | |
a2df2a63 | 4956 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
0e8b6879 LC |
4957 | - lblk; |
4958 | ||
556615dc | 4959 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
0e8b6879 LC |
4960 | if (mode & FALLOC_FL_KEEP_SIZE) |
4961 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4962 | ||
55bd725a | 4963 | mutex_lock(&inode->i_mutex); |
f282ac19 LC |
4964 | |
4965 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
4966 | offset + len > i_size_read(inode)) { | |
4967 | new_size = offset + len; | |
4968 | ret = inode_newsize_ok(inode, new_size); | |
4969 | if (ret) | |
4970 | goto out; | |
6d19c42b | 4971 | } |
f282ac19 | 4972 | |
c174e6d6 DM |
4973 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, |
4974 | flags, mode); | |
0e8b6879 LC |
4975 | if (ret) |
4976 | goto out; | |
f282ac19 | 4977 | |
c174e6d6 DM |
4978 | if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) { |
4979 | ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal, | |
4980 | EXT4_I(inode)->i_sync_tid); | |
f282ac19 | 4981 | } |
f282ac19 | 4982 | out: |
55bd725a | 4983 | mutex_unlock(&inode->i_mutex); |
0e8b6879 LC |
4984 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
4985 | return ret; | |
a2df2a63 | 4986 | } |
6873fa0d | 4987 | |
0031462b MC |
4988 | /* |
4989 | * This function convert a range of blocks to written extents | |
4990 | * The caller of this function will pass the start offset and the size. | |
4991 | * all unwritten extents within this range will be converted to | |
4992 | * written extents. | |
4993 | * | |
4994 | * This function is called from the direct IO end io call back | |
4995 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4996 | * Returns 0 on success. |
0031462b | 4997 | */ |
6b523df4 JK |
4998 | int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
4999 | loff_t offset, ssize_t len) | |
0031462b | 5000 | { |
0031462b MC |
5001 | unsigned int max_blocks; |
5002 | int ret = 0; | |
5003 | int ret2 = 0; | |
2ed88685 | 5004 | struct ext4_map_blocks map; |
0031462b MC |
5005 | unsigned int credits, blkbits = inode->i_blkbits; |
5006 | ||
2ed88685 | 5007 | map.m_lblk = offset >> blkbits; |
0031462b MC |
5008 | /* |
5009 | * We can't just convert len to max_blocks because | |
5010 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
5011 | */ | |
2ed88685 TT |
5012 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
5013 | map.m_lblk); | |
0031462b | 5014 | /* |
6b523df4 JK |
5015 | * This is somewhat ugly but the idea is clear: When transaction is |
5016 | * reserved, everything goes into it. Otherwise we rather start several | |
5017 | * smaller transactions for conversion of each extent separately. | |
0031462b | 5018 | */ |
6b523df4 JK |
5019 | if (handle) { |
5020 | handle = ext4_journal_start_reserved(handle, | |
5021 | EXT4_HT_EXT_CONVERT); | |
5022 | if (IS_ERR(handle)) | |
5023 | return PTR_ERR(handle); | |
5024 | credits = 0; | |
5025 | } else { | |
5026 | /* | |
5027 | * credits to insert 1 extent into extent tree | |
5028 | */ | |
5029 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
5030 | } | |
0031462b | 5031 | while (ret >= 0 && ret < max_blocks) { |
2ed88685 TT |
5032 | map.m_lblk += ret; |
5033 | map.m_len = (max_blocks -= ret); | |
6b523df4 JK |
5034 | if (credits) { |
5035 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
5036 | credits); | |
5037 | if (IS_ERR(handle)) { | |
5038 | ret = PTR_ERR(handle); | |
5039 | break; | |
5040 | } | |
0031462b | 5041 | } |
2ed88685 | 5042 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 5043 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
5044 | if (ret <= 0) |
5045 | ext4_warning(inode->i_sb, | |
5046 | "inode #%lu: block %u: len %u: " | |
5047 | "ext4_ext_map_blocks returned %d", | |
5048 | inode->i_ino, map.m_lblk, | |
5049 | map.m_len, ret); | |
0031462b | 5050 | ext4_mark_inode_dirty(handle, inode); |
6b523df4 JK |
5051 | if (credits) |
5052 | ret2 = ext4_journal_stop(handle); | |
5053 | if (ret <= 0 || ret2) | |
0031462b MC |
5054 | break; |
5055 | } | |
6b523df4 JK |
5056 | if (!credits) |
5057 | ret2 = ext4_journal_stop(handle); | |
0031462b MC |
5058 | return ret > 0 ? ret2 : ret; |
5059 | } | |
6d9c85eb | 5060 | |
6873fa0d | 5061 | /* |
69eb33dc ZL |
5062 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
5063 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
5064 | * return start of the next delayed extent. |
5065 | * | |
69eb33dc | 5066 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 5067 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 5068 | * extent found. Leave newes unmodified. |
6873fa0d | 5069 | */ |
91dd8c11 | 5070 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 5071 | struct extent_status *newes) |
6873fa0d | 5072 | { |
b3aff3e3 | 5073 | struct extent_status es; |
be401363 | 5074 | ext4_lblk_t block, next_del; |
6873fa0d | 5075 | |
69eb33dc | 5076 | if (newes->es_pblk == 0) { |
e30b5dca YZ |
5077 | ext4_es_find_delayed_extent_range(inode, newes->es_lblk, |
5078 | newes->es_lblk + newes->es_len - 1, &es); | |
5079 | ||
6d9c85eb | 5080 | /* |
69eb33dc | 5081 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 5082 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 5083 | */ |
06b0c886 | 5084 | if (es.es_len == 0) |
b3aff3e3 | 5085 | /* A hole found. */ |
91dd8c11 | 5086 | return 0; |
b3aff3e3 | 5087 | |
69eb33dc | 5088 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 5089 | /* A hole found. */ |
69eb33dc ZL |
5090 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
5091 | newes->es_len); | |
91dd8c11 | 5092 | return 0; |
6873fa0d | 5093 | } |
6d9c85eb | 5094 | |
69eb33dc | 5095 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
5096 | } |
5097 | ||
69eb33dc | 5098 | block = newes->es_lblk + newes->es_len; |
e30b5dca | 5099 | ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es); |
be401363 ZL |
5100 | if (es.es_len == 0) |
5101 | next_del = EXT_MAX_BLOCKS; | |
5102 | else | |
5103 | next_del = es.es_lblk; | |
5104 | ||
91dd8c11 | 5105 | return next_del; |
6873fa0d | 5106 | } |
6873fa0d ES |
5107 | /* fiemap flags we can handle specified here */ |
5108 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
5109 | ||
3a06d778 AK |
5110 | static int ext4_xattr_fiemap(struct inode *inode, |
5111 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
5112 | { |
5113 | __u64 physical = 0; | |
5114 | __u64 length; | |
5115 | __u32 flags = FIEMAP_EXTENT_LAST; | |
5116 | int blockbits = inode->i_sb->s_blocksize_bits; | |
5117 | int error = 0; | |
5118 | ||
5119 | /* in-inode? */ | |
19f5fb7a | 5120 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
5121 | struct ext4_iloc iloc; |
5122 | int offset; /* offset of xattr in inode */ | |
5123 | ||
5124 | error = ext4_get_inode_loc(inode, &iloc); | |
5125 | if (error) | |
5126 | return error; | |
a60697f4 | 5127 | physical = (__u64)iloc.bh->b_blocknr << blockbits; |
6873fa0d ES |
5128 | offset = EXT4_GOOD_OLD_INODE_SIZE + |
5129 | EXT4_I(inode)->i_extra_isize; | |
5130 | physical += offset; | |
5131 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
5132 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 5133 | brelse(iloc.bh); |
6873fa0d | 5134 | } else { /* external block */ |
a60697f4 | 5135 | physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
6873fa0d ES |
5136 | length = inode->i_sb->s_blocksize; |
5137 | } | |
5138 | ||
5139 | if (physical) | |
5140 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
5141 | length, flags); | |
5142 | return (error < 0 ? error : 0); | |
5143 | } | |
5144 | ||
5145 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
5146 | __u64 start, __u64 len) | |
5147 | { | |
5148 | ext4_lblk_t start_blk; | |
6873fa0d ES |
5149 | int error = 0; |
5150 | ||
94191985 TM |
5151 | if (ext4_has_inline_data(inode)) { |
5152 | int has_inline = 1; | |
5153 | ||
d952d69e DM |
5154 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline, |
5155 | start, len); | |
94191985 TM |
5156 | |
5157 | if (has_inline) | |
5158 | return error; | |
5159 | } | |
5160 | ||
7869a4a6 TT |
5161 | if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
5162 | error = ext4_ext_precache(inode); | |
5163 | if (error) | |
5164 | return error; | |
5165 | } | |
5166 | ||
6873fa0d | 5167 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 5168 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
14516bb7 DM |
5169 | return __generic_block_fiemap(inode, fieinfo, start, len, |
5170 | ext4_get_block); | |
6873fa0d ES |
5171 | |
5172 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
5173 | return -EBADR; | |
5174 | ||
5175 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
5176 | error = ext4_xattr_fiemap(inode, fieinfo); | |
5177 | } else { | |
aca92ff6 LM |
5178 | ext4_lblk_t len_blks; |
5179 | __u64 last_blk; | |
5180 | ||
6873fa0d | 5181 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 5182 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
5183 | if (last_blk >= EXT_MAX_BLOCKS) |
5184 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 5185 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
5186 | |
5187 | /* | |
91dd8c11 LC |
5188 | * Walk the extent tree gathering extent information |
5189 | * and pushing extents back to the user. | |
6873fa0d | 5190 | */ |
91dd8c11 LC |
5191 | error = ext4_fill_fiemap_extents(inode, start_blk, |
5192 | len_blks, fieinfo); | |
6873fa0d | 5193 | } |
6873fa0d ES |
5194 | return error; |
5195 | } | |
9eb79482 NJ |
5196 | |
5197 | /* | |
5198 | * ext4_access_path: | |
5199 | * Function to access the path buffer for marking it dirty. | |
5200 | * It also checks if there are sufficient credits left in the journal handle | |
5201 | * to update path. | |
5202 | */ | |
5203 | static int | |
5204 | ext4_access_path(handle_t *handle, struct inode *inode, | |
5205 | struct ext4_ext_path *path) | |
5206 | { | |
5207 | int credits, err; | |
5208 | ||
5209 | if (!ext4_handle_valid(handle)) | |
5210 | return 0; | |
5211 | ||
5212 | /* | |
5213 | * Check if need to extend journal credits | |
5214 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
5215 | * descriptor) for each block group; assume two block | |
5216 | * groups | |
5217 | */ | |
5218 | if (handle->h_buffer_credits < 7) { | |
5219 | credits = ext4_writepage_trans_blocks(inode); | |
5220 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); | |
5221 | /* EAGAIN is success */ | |
5222 | if (err && err != -EAGAIN) | |
5223 | return err; | |
5224 | } | |
5225 | ||
5226 | err = ext4_ext_get_access(handle, inode, path); | |
5227 | return err; | |
5228 | } | |
5229 | ||
5230 | /* | |
5231 | * ext4_ext_shift_path_extents: | |
5232 | * Shift the extents of a path structure lying between path[depth].p_ext | |
5233 | * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift | |
5234 | * from starting block for each extent. | |
5235 | */ | |
5236 | static int | |
5237 | ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift, | |
5238 | struct inode *inode, handle_t *handle, | |
5239 | ext4_lblk_t *start) | |
5240 | { | |
5241 | int depth, err = 0; | |
5242 | struct ext4_extent *ex_start, *ex_last; | |
5243 | bool update = 0; | |
5244 | depth = path->p_depth; | |
5245 | ||
5246 | while (depth >= 0) { | |
5247 | if (depth == path->p_depth) { | |
5248 | ex_start = path[depth].p_ext; | |
5249 | if (!ex_start) | |
5250 | return -EIO; | |
5251 | ||
5252 | ex_last = EXT_LAST_EXTENT(path[depth].p_hdr); | |
9eb79482 NJ |
5253 | |
5254 | err = ext4_access_path(handle, inode, path + depth); | |
5255 | if (err) | |
5256 | goto out; | |
5257 | ||
5258 | if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
5259 | update = 1; | |
5260 | ||
847c6c42 | 5261 | *start = le32_to_cpu(ex_last->ee_block) + |
9eb79482 NJ |
5262 | ext4_ext_get_actual_len(ex_last); |
5263 | ||
5264 | while (ex_start <= ex_last) { | |
847c6c42 | 5265 | le32_add_cpu(&ex_start->ee_block, -shift); |
6dd834ef LC |
5266 | /* Try to merge to the left. */ |
5267 | if ((ex_start > | |
5268 | EXT_FIRST_EXTENT(path[depth].p_hdr)) && | |
5269 | ext4_ext_try_to_merge_right(inode, | |
5270 | path, ex_start - 1)) | |
5271 | ex_last--; | |
5272 | else | |
5273 | ex_start++; | |
9eb79482 NJ |
5274 | } |
5275 | err = ext4_ext_dirty(handle, inode, path + depth); | |
5276 | if (err) | |
5277 | goto out; | |
5278 | ||
5279 | if (--depth < 0 || !update) | |
5280 | break; | |
5281 | } | |
5282 | ||
5283 | /* Update index too */ | |
5284 | err = ext4_access_path(handle, inode, path + depth); | |
5285 | if (err) | |
5286 | goto out; | |
5287 | ||
847c6c42 | 5288 | le32_add_cpu(&path[depth].p_idx->ei_block, -shift); |
9eb79482 NJ |
5289 | err = ext4_ext_dirty(handle, inode, path + depth); |
5290 | if (err) | |
5291 | goto out; | |
5292 | ||
5293 | /* we are done if current index is not a starting index */ | |
5294 | if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr)) | |
5295 | break; | |
5296 | ||
5297 | depth--; | |
5298 | } | |
5299 | ||
5300 | out: | |
5301 | return err; | |
5302 | } | |
5303 | ||
5304 | /* | |
5305 | * ext4_ext_shift_extents: | |
5306 | * All the extents which lies in the range from start to the last allocated | |
5307 | * block for the file are shifted downwards by shift blocks. | |
5308 | * On success, 0 is returned, error otherwise. | |
5309 | */ | |
5310 | static int | |
5311 | ext4_ext_shift_extents(struct inode *inode, handle_t *handle, | |
5312 | ext4_lblk_t start, ext4_lblk_t shift) | |
5313 | { | |
5314 | struct ext4_ext_path *path; | |
5315 | int ret = 0, depth; | |
5316 | struct ext4_extent *extent; | |
f8fb4f41 | 5317 | ext4_lblk_t stop_block; |
9eb79482 NJ |
5318 | ext4_lblk_t ex_start, ex_end; |
5319 | ||
5320 | /* Let path point to the last extent */ | |
ed8a1a76 | 5321 | path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0); |
9eb79482 NJ |
5322 | if (IS_ERR(path)) |
5323 | return PTR_ERR(path); | |
5324 | ||
5325 | depth = path->p_depth; | |
5326 | extent = path[depth].p_ext; | |
ee4bd0d9 TT |
5327 | if (!extent) |
5328 | goto out; | |
9eb79482 | 5329 | |
847c6c42 ZL |
5330 | stop_block = le32_to_cpu(extent->ee_block) + |
5331 | ext4_ext_get_actual_len(extent); | |
9eb79482 NJ |
5332 | |
5333 | /* Nothing to shift, if hole is at the end of file */ | |
5334 | if (start >= stop_block) | |
ee4bd0d9 | 5335 | goto out; |
9eb79482 NJ |
5336 | |
5337 | /* | |
5338 | * Don't start shifting extents until we make sure the hole is big | |
5339 | * enough to accomodate the shift. | |
5340 | */ | |
ed8a1a76 | 5341 | path = ext4_find_extent(inode, start - 1, &path, 0); |
8dc79ec4 DM |
5342 | if (IS_ERR(path)) |
5343 | return PTR_ERR(path); | |
9eb79482 NJ |
5344 | depth = path->p_depth; |
5345 | extent = path[depth].p_ext; | |
8dc79ec4 DM |
5346 | if (extent) { |
5347 | ex_start = le32_to_cpu(extent->ee_block); | |
5348 | ex_end = le32_to_cpu(extent->ee_block) + | |
847c6c42 | 5349 | ext4_ext_get_actual_len(extent); |
8dc79ec4 DM |
5350 | } else { |
5351 | ex_start = 0; | |
5352 | ex_end = 0; | |
5353 | } | |
9eb79482 NJ |
5354 | |
5355 | if ((start == ex_start && shift > ex_start) || | |
5356 | (shift > start - ex_end)) | |
5357 | return -EINVAL; | |
5358 | ||
5359 | /* Its safe to start updating extents */ | |
5360 | while (start < stop_block) { | |
ed8a1a76 | 5361 | path = ext4_find_extent(inode, start, &path, 0); |
9eb79482 NJ |
5362 | if (IS_ERR(path)) |
5363 | return PTR_ERR(path); | |
5364 | depth = path->p_depth; | |
5365 | extent = path[depth].p_ext; | |
a18ed359 DM |
5366 | if (!extent) { |
5367 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
5368 | (unsigned long) start); | |
5369 | return -EIO; | |
5370 | } | |
f8fb4f41 | 5371 | if (start > le32_to_cpu(extent->ee_block)) { |
9eb79482 | 5372 | /* Hole, move to the next extent */ |
f8fb4f41 DM |
5373 | if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) { |
5374 | path[depth].p_ext++; | |
5375 | } else { | |
5376 | start = ext4_ext_next_allocated_block(path); | |
f8fb4f41 | 5377 | continue; |
9eb79482 NJ |
5378 | } |
5379 | } | |
5380 | ret = ext4_ext_shift_path_extents(path, shift, inode, | |
5381 | handle, &start); | |
9eb79482 NJ |
5382 | if (ret) |
5383 | break; | |
5384 | } | |
ee4bd0d9 TT |
5385 | out: |
5386 | ext4_ext_drop_refs(path); | |
5387 | kfree(path); | |
9eb79482 NJ |
5388 | return ret; |
5389 | } | |
5390 | ||
5391 | /* | |
5392 | * ext4_collapse_range: | |
5393 | * This implements the fallocate's collapse range functionality for ext4 | |
5394 | * Returns: 0 and non-zero on error. | |
5395 | */ | |
5396 | int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len) | |
5397 | { | |
5398 | struct super_block *sb = inode->i_sb; | |
5399 | ext4_lblk_t punch_start, punch_stop; | |
5400 | handle_t *handle; | |
5401 | unsigned int credits; | |
a8680e0d | 5402 | loff_t new_size, ioffset; |
9eb79482 NJ |
5403 | int ret; |
5404 | ||
9eb79482 | 5405 | /* Collapse range works only on fs block size aligned offsets. */ |
ee98fa3a NJ |
5406 | if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) || |
5407 | len & (EXT4_CLUSTER_SIZE(sb) - 1)) | |
9eb79482 NJ |
5408 | return -EINVAL; |
5409 | ||
5410 | if (!S_ISREG(inode->i_mode)) | |
86f1ca38 | 5411 | return -EINVAL; |
9eb79482 NJ |
5412 | |
5413 | trace_ext4_collapse_range(inode, offset, len); | |
5414 | ||
5415 | punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5416 | punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb); | |
5417 | ||
1ce01c4a NJ |
5418 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
5419 | if (ext4_should_journal_data(inode)) { | |
5420 | ret = ext4_force_commit(inode->i_sb); | |
5421 | if (ret) | |
5422 | return ret; | |
5423 | } | |
5424 | ||
a8680e0d NJ |
5425 | /* |
5426 | * Need to round down offset to be aligned with page size boundary | |
5427 | * for page size > block size. | |
5428 | */ | |
5429 | ioffset = round_down(offset, PAGE_SIZE); | |
5430 | ||
9eb79482 | 5431 | /* Write out all dirty pages */ |
a8680e0d NJ |
5432 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, |
5433 | LLONG_MAX); | |
9eb79482 NJ |
5434 | if (ret) |
5435 | return ret; | |
5436 | ||
5437 | /* Take mutex lock */ | |
5438 | mutex_lock(&inode->i_mutex); | |
5439 | ||
23fffa92 LC |
5440 | /* |
5441 | * There is no need to overlap collapse range with EOF, in which case | |
5442 | * it is effectively a truncate operation | |
5443 | */ | |
5444 | if (offset + len >= i_size_read(inode)) { | |
5445 | ret = -EINVAL; | |
5446 | goto out_mutex; | |
5447 | } | |
5448 | ||
9eb79482 NJ |
5449 | /* Currently just for extent based files */ |
5450 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5451 | ret = -EOPNOTSUPP; | |
5452 | goto out_mutex; | |
5453 | } | |
5454 | ||
a8680e0d | 5455 | truncate_pagecache(inode, ioffset); |
9eb79482 NJ |
5456 | |
5457 | /* Wait for existing dio to complete */ | |
5458 | ext4_inode_block_unlocked_dio(inode); | |
5459 | inode_dio_wait(inode); | |
5460 | ||
5461 | credits = ext4_writepage_trans_blocks(inode); | |
5462 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5463 | if (IS_ERR(handle)) { | |
5464 | ret = PTR_ERR(handle); | |
5465 | goto out_dio; | |
5466 | } | |
5467 | ||
5468 | down_write(&EXT4_I(inode)->i_data_sem); | |
5469 | ext4_discard_preallocations(inode); | |
5470 | ||
5471 | ret = ext4_es_remove_extent(inode, punch_start, | |
2c1d2328 | 5472 | EXT_MAX_BLOCKS - punch_start); |
9eb79482 NJ |
5473 | if (ret) { |
5474 | up_write(&EXT4_I(inode)->i_data_sem); | |
5475 | goto out_stop; | |
5476 | } | |
5477 | ||
5478 | ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1); | |
5479 | if (ret) { | |
5480 | up_write(&EXT4_I(inode)->i_data_sem); | |
5481 | goto out_stop; | |
5482 | } | |
ef24f6c2 | 5483 | ext4_discard_preallocations(inode); |
9eb79482 NJ |
5484 | |
5485 | ret = ext4_ext_shift_extents(inode, handle, punch_stop, | |
5486 | punch_stop - punch_start); | |
5487 | if (ret) { | |
5488 | up_write(&EXT4_I(inode)->i_data_sem); | |
5489 | goto out_stop; | |
5490 | } | |
5491 | ||
5492 | new_size = i_size_read(inode) - len; | |
9337d5d3 | 5493 | i_size_write(inode, new_size); |
9eb79482 NJ |
5494 | EXT4_I(inode)->i_disksize = new_size; |
5495 | ||
9eb79482 NJ |
5496 | up_write(&EXT4_I(inode)->i_data_sem); |
5497 | if (IS_SYNC(inode)) | |
5498 | ext4_handle_sync(handle); | |
5499 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
5500 | ext4_mark_inode_dirty(handle, inode); | |
5501 | ||
5502 | out_stop: | |
5503 | ext4_journal_stop(handle); | |
5504 | out_dio: | |
5505 | ext4_inode_resume_unlocked_dio(inode); | |
5506 | out_mutex: | |
5507 | mutex_unlock(&inode->i_mutex); | |
5508 | return ret; | |
5509 | } | |
fcf6b1b7 DM |
5510 | |
5511 | /** | |
5512 | * ext4_swap_extents - Swap extents between two inodes | |
5513 | * | |
5514 | * @inode1: First inode | |
5515 | * @inode2: Second inode | |
5516 | * @lblk1: Start block for first inode | |
5517 | * @lblk2: Start block for second inode | |
5518 | * @count: Number of blocks to swap | |
5519 | * @mark_unwritten: Mark second inode's extents as unwritten after swap | |
5520 | * @erp: Pointer to save error value | |
5521 | * | |
5522 | * This helper routine does exactly what is promise "swap extents". All other | |
5523 | * stuff such as page-cache locking consistency, bh mapping consistency or | |
5524 | * extent's data copying must be performed by caller. | |
5525 | * Locking: | |
5526 | * i_mutex is held for both inodes | |
5527 | * i_data_sem is locked for write for both inodes | |
5528 | * Assumptions: | |
5529 | * All pages from requested range are locked for both inodes | |
5530 | */ | |
5531 | int | |
5532 | ext4_swap_extents(handle_t *handle, struct inode *inode1, | |
5533 | struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2, | |
5534 | ext4_lblk_t count, int unwritten, int *erp) | |
5535 | { | |
5536 | struct ext4_ext_path *path1 = NULL; | |
5537 | struct ext4_ext_path *path2 = NULL; | |
5538 | int replaced_count = 0; | |
5539 | ||
5540 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem)); | |
5541 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem)); | |
5542 | BUG_ON(!mutex_is_locked(&inode1->i_mutex)); | |
5543 | BUG_ON(!mutex_is_locked(&inode1->i_mutex)); | |
5544 | ||
5545 | *erp = ext4_es_remove_extent(inode1, lblk1, count); | |
19008f6d | 5546 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5547 | return 0; |
5548 | *erp = ext4_es_remove_extent(inode2, lblk2, count); | |
19008f6d | 5549 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5550 | return 0; |
5551 | ||
5552 | while (count) { | |
5553 | struct ext4_extent *ex1, *ex2, tmp_ex; | |
5554 | ext4_lblk_t e1_blk, e2_blk; | |
5555 | int e1_len, e2_len, len; | |
5556 | int split = 0; | |
5557 | ||
ed8a1a76 | 5558 | path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE); |
19008f6d | 5559 | if (unlikely(IS_ERR(path1))) { |
fcf6b1b7 | 5560 | *erp = PTR_ERR(path1); |
19008f6d TT |
5561 | path1 = NULL; |
5562 | finish: | |
5563 | count = 0; | |
5564 | goto repeat; | |
fcf6b1b7 | 5565 | } |
ed8a1a76 | 5566 | path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE); |
19008f6d | 5567 | if (unlikely(IS_ERR(path2))) { |
fcf6b1b7 | 5568 | *erp = PTR_ERR(path2); |
19008f6d TT |
5569 | path2 = NULL; |
5570 | goto finish; | |
fcf6b1b7 DM |
5571 | } |
5572 | ex1 = path1[path1->p_depth].p_ext; | |
5573 | ex2 = path2[path2->p_depth].p_ext; | |
5574 | /* Do we have somthing to swap ? */ | |
5575 | if (unlikely(!ex2 || !ex1)) | |
19008f6d | 5576 | goto finish; |
fcf6b1b7 DM |
5577 | |
5578 | e1_blk = le32_to_cpu(ex1->ee_block); | |
5579 | e2_blk = le32_to_cpu(ex2->ee_block); | |
5580 | e1_len = ext4_ext_get_actual_len(ex1); | |
5581 | e2_len = ext4_ext_get_actual_len(ex2); | |
5582 | ||
5583 | /* Hole handling */ | |
5584 | if (!in_range(lblk1, e1_blk, e1_len) || | |
5585 | !in_range(lblk2, e2_blk, e2_len)) { | |
5586 | ext4_lblk_t next1, next2; | |
5587 | ||
5588 | /* if hole after extent, then go to next extent */ | |
5589 | next1 = ext4_ext_next_allocated_block(path1); | |
5590 | next2 = ext4_ext_next_allocated_block(path2); | |
5591 | /* If hole before extent, then shift to that extent */ | |
5592 | if (e1_blk > lblk1) | |
5593 | next1 = e1_blk; | |
5594 | if (e2_blk > lblk2) | |
5595 | next2 = e1_blk; | |
5596 | /* Do we have something to swap */ | |
5597 | if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS) | |
19008f6d | 5598 | goto finish; |
fcf6b1b7 DM |
5599 | /* Move to the rightest boundary */ |
5600 | len = next1 - lblk1; | |
5601 | if (len < next2 - lblk2) | |
5602 | len = next2 - lblk2; | |
5603 | if (len > count) | |
5604 | len = count; | |
5605 | lblk1 += len; | |
5606 | lblk2 += len; | |
5607 | count -= len; | |
5608 | goto repeat; | |
5609 | } | |
5610 | ||
5611 | /* Prepare left boundary */ | |
5612 | if (e1_blk < lblk1) { | |
5613 | split = 1; | |
5614 | *erp = ext4_force_split_extent_at(handle, inode1, | |
dfe50809 | 5615 | &path1, lblk1, 0); |
19008f6d TT |
5616 | if (unlikely(*erp)) |
5617 | goto finish; | |
fcf6b1b7 DM |
5618 | } |
5619 | if (e2_blk < lblk2) { | |
5620 | split = 1; | |
5621 | *erp = ext4_force_split_extent_at(handle, inode2, | |
dfe50809 | 5622 | &path2, lblk2, 0); |
19008f6d TT |
5623 | if (unlikely(*erp)) |
5624 | goto finish; | |
fcf6b1b7 | 5625 | } |
dfe50809 | 5626 | /* ext4_split_extent_at() may result in leaf extent split, |
fcf6b1b7 DM |
5627 | * path must to be revalidated. */ |
5628 | if (split) | |
5629 | goto repeat; | |
5630 | ||
5631 | /* Prepare right boundary */ | |
5632 | len = count; | |
5633 | if (len > e1_blk + e1_len - lblk1) | |
5634 | len = e1_blk + e1_len - lblk1; | |
5635 | if (len > e2_blk + e2_len - lblk2) | |
5636 | len = e2_blk + e2_len - lblk2; | |
5637 | ||
5638 | if (len != e1_len) { | |
5639 | split = 1; | |
5640 | *erp = ext4_force_split_extent_at(handle, inode1, | |
dfe50809 | 5641 | &path1, lblk1 + len, 0); |
19008f6d TT |
5642 | if (unlikely(*erp)) |
5643 | goto finish; | |
fcf6b1b7 DM |
5644 | } |
5645 | if (len != e2_len) { | |
5646 | split = 1; | |
5647 | *erp = ext4_force_split_extent_at(handle, inode2, | |
dfe50809 | 5648 | &path2, lblk2 + len, 0); |
fcf6b1b7 | 5649 | if (*erp) |
19008f6d | 5650 | goto finish; |
fcf6b1b7 | 5651 | } |
dfe50809 | 5652 | /* ext4_split_extent_at() may result in leaf extent split, |
fcf6b1b7 DM |
5653 | * path must to be revalidated. */ |
5654 | if (split) | |
5655 | goto repeat; | |
5656 | ||
5657 | BUG_ON(e2_len != e1_len); | |
5658 | *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth); | |
19008f6d TT |
5659 | if (unlikely(*erp)) |
5660 | goto finish; | |
fcf6b1b7 | 5661 | *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth); |
19008f6d TT |
5662 | if (unlikely(*erp)) |
5663 | goto finish; | |
fcf6b1b7 DM |
5664 | |
5665 | /* Both extents are fully inside boundaries. Swap it now */ | |
5666 | tmp_ex = *ex1; | |
5667 | ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2)); | |
5668 | ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex)); | |
5669 | ex1->ee_len = cpu_to_le16(e2_len); | |
5670 | ex2->ee_len = cpu_to_le16(e1_len); | |
5671 | if (unwritten) | |
5672 | ext4_ext_mark_unwritten(ex2); | |
5673 | if (ext4_ext_is_unwritten(&tmp_ex)) | |
5674 | ext4_ext_mark_unwritten(ex1); | |
5675 | ||
5676 | ext4_ext_try_to_merge(handle, inode2, path2, ex2); | |
5677 | ext4_ext_try_to_merge(handle, inode1, path1, ex1); | |
5678 | *erp = ext4_ext_dirty(handle, inode2, path2 + | |
5679 | path2->p_depth); | |
19008f6d TT |
5680 | if (unlikely(*erp)) |
5681 | goto finish; | |
fcf6b1b7 DM |
5682 | *erp = ext4_ext_dirty(handle, inode1, path1 + |
5683 | path1->p_depth); | |
5684 | /* | |
5685 | * Looks scarry ah..? second inode already points to new blocks, | |
5686 | * and it was successfully dirtied. But luckily error may happen | |
5687 | * only due to journal error, so full transaction will be | |
5688 | * aborted anyway. | |
5689 | */ | |
19008f6d TT |
5690 | if (unlikely(*erp)) |
5691 | goto finish; | |
fcf6b1b7 DM |
5692 | lblk1 += len; |
5693 | lblk2 += len; | |
5694 | replaced_count += len; | |
5695 | count -= len; | |
5696 | ||
5697 | repeat: | |
b7ea89ad TT |
5698 | ext4_ext_drop_refs(path1); |
5699 | kfree(path1); | |
5700 | ext4_ext_drop_refs(path2); | |
5701 | kfree(path2); | |
5702 | path1 = path2 = NULL; | |
fcf6b1b7 | 5703 | } |
fcf6b1b7 DM |
5704 | return replaced_count; |
5705 | } |