]>
Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a86c6181 AT |
2 | /* |
3 | * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com | |
4 | * Written by Alex Tomas <alex@clusterfs.com> | |
5 | * | |
6 | * Architecture independence: | |
7 | * Copyright (c) 2005, Bull S.A. | |
8 | * Written by Pierre Peiffer <pierre.peiffer@bull.net> | |
a86c6181 AT |
9 | */ |
10 | ||
11 | /* | |
12 | * Extents support for EXT4 | |
13 | * | |
14 | * TODO: | |
15 | * - ext4*_error() should be used in some situations | |
16 | * - analyze all BUG()/BUG_ON(), use -EIO where appropriate | |
17 | * - smart tree reduction | |
18 | */ | |
19 | ||
a86c6181 AT |
20 | #include <linux/fs.h> |
21 | #include <linux/time.h> | |
cd02ff0b | 22 | #include <linux/jbd2.h> |
a86c6181 AT |
23 | #include <linux/highuid.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/quotaops.h> | |
26 | #include <linux/string.h> | |
27 | #include <linux/slab.h> | |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
6873fa0d | 29 | #include <linux/fiemap.h> |
66114cad | 30 | #include <linux/backing-dev.h> |
3dcf5451 | 31 | #include "ext4_jbd2.h" |
4a092d73 | 32 | #include "ext4_extents.h" |
f19d5870 | 33 | #include "xattr.h" |
a86c6181 | 34 | |
0562e0ba JZ |
35 | #include <trace/events/ext4.h> |
36 | ||
5f95d21f LC |
37 | /* |
38 | * used by extent splitting. | |
39 | */ | |
40 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
41 | due to ENOSPC */ | |
556615dc LC |
42 | #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */ |
43 | #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */ | |
5f95d21f | 44 | |
dee1f973 DM |
45 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
46 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
47 | ||
7ac5990d DW |
48 | static __le32 ext4_extent_block_csum(struct inode *inode, |
49 | struct ext4_extent_header *eh) | |
50 | { | |
51 | struct ext4_inode_info *ei = EXT4_I(inode); | |
52 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
53 | __u32 csum; | |
54 | ||
55 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
56 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
57 | return cpu_to_le32(csum); | |
58 | } | |
59 | ||
60 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
61 | struct ext4_extent_header *eh) | |
62 | { | |
63 | struct ext4_extent_tail *et; | |
64 | ||
9aa5d32b | 65 | if (!ext4_has_metadata_csum(inode->i_sb)) |
7ac5990d DW |
66 | return 1; |
67 | ||
68 | et = find_ext4_extent_tail(eh); | |
69 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
70 | return 0; | |
71 | return 1; | |
72 | } | |
73 | ||
74 | static void ext4_extent_block_csum_set(struct inode *inode, | |
75 | struct ext4_extent_header *eh) | |
76 | { | |
77 | struct ext4_extent_tail *et; | |
78 | ||
9aa5d32b | 79 | if (!ext4_has_metadata_csum(inode->i_sb)) |
7ac5990d DW |
80 | return; |
81 | ||
82 | et = find_ext4_extent_tail(eh); | |
83 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
84 | } | |
85 | ||
d583fb87 AH |
86 | static int ext4_split_extent(handle_t *handle, |
87 | struct inode *inode, | |
dfe50809 | 88 | struct ext4_ext_path **ppath, |
d583fb87 AH |
89 | struct ext4_map_blocks *map, |
90 | int split_flag, | |
91 | int flags); | |
92 | ||
5f95d21f LC |
93 | static int ext4_split_extent_at(handle_t *handle, |
94 | struct inode *inode, | |
dfe50809 | 95 | struct ext4_ext_path **ppath, |
5f95d21f LC |
96 | ext4_lblk_t split, |
97 | int split_flag, | |
98 | int flags); | |
99 | ||
91dd8c11 | 100 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 101 | struct extent_status *newes); |
91dd8c11 | 102 | |
487caeef JK |
103 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
104 | struct inode *inode, | |
105 | int needed) | |
a86c6181 AT |
106 | { |
107 | int err; | |
108 | ||
0390131b FM |
109 | if (!ext4_handle_valid(handle)) |
110 | return 0; | |
7b808191 | 111 | if (handle->h_buffer_credits >= needed) |
9102e4fa | 112 | return 0; |
7b808191 TT |
113 | /* |
114 | * If we need to extend the journal get a few extra blocks | |
115 | * while we're at it for efficiency's sake. | |
116 | */ | |
117 | needed += 3; | |
118 | err = ext4_journal_extend(handle, needed - handle->h_buffer_credits); | |
0123c939 | 119 | if (err <= 0) |
9102e4fa | 120 | return err; |
487caeef | 121 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
122 | if (err == 0) |
123 | err = -EAGAIN; | |
487caeef JK |
124 | |
125 | return err; | |
a86c6181 AT |
126 | } |
127 | ||
128 | /* | |
129 | * could return: | |
130 | * - EROFS | |
131 | * - ENOMEM | |
132 | */ | |
133 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
134 | struct ext4_ext_path *path) | |
135 | { | |
136 | if (path->p_bh) { | |
137 | /* path points to block */ | |
5d601255 | 138 | BUFFER_TRACE(path->p_bh, "get_write_access"); |
a86c6181 AT |
139 | return ext4_journal_get_write_access(handle, path->p_bh); |
140 | } | |
141 | /* path points to leaf/index in inode body */ | |
142 | /* we use in-core data, no need to protect them */ | |
143 | return 0; | |
144 | } | |
145 | ||
146 | /* | |
147 | * could return: | |
148 | * - EROFS | |
149 | * - ENOMEM | |
150 | * - EIO | |
151 | */ | |
2656497b DW |
152 | int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle, |
153 | struct inode *inode, struct ext4_ext_path *path) | |
a86c6181 AT |
154 | { |
155 | int err; | |
4b1f1660 DM |
156 | |
157 | WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem)); | |
a86c6181 | 158 | if (path->p_bh) { |
7ac5990d | 159 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 160 | /* path points to block */ |
9ea7a0df TT |
161 | err = __ext4_handle_dirty_metadata(where, line, handle, |
162 | inode, path->p_bh); | |
a86c6181 AT |
163 | } else { |
164 | /* path points to leaf/index in inode body */ | |
165 | err = ext4_mark_inode_dirty(handle, inode); | |
166 | } | |
167 | return err; | |
168 | } | |
169 | ||
f65e6fba | 170 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 171 | struct ext4_ext_path *path, |
725d26d3 | 172 | ext4_lblk_t block) |
a86c6181 | 173 | { |
a86c6181 | 174 | if (path) { |
81fdbb4a | 175 | int depth = path->p_depth; |
a86c6181 | 176 | struct ext4_extent *ex; |
a86c6181 | 177 | |
ad4fb9ca KM |
178 | /* |
179 | * Try to predict block placement assuming that we are | |
180 | * filling in a file which will eventually be | |
181 | * non-sparse --- i.e., in the case of libbfd writing | |
182 | * an ELF object sections out-of-order but in a way | |
183 | * the eventually results in a contiguous object or | |
184 | * executable file, or some database extending a table | |
185 | * space file. However, this is actually somewhat | |
186 | * non-ideal if we are writing a sparse file such as | |
187 | * qemu or KVM writing a raw image file that is going | |
188 | * to stay fairly sparse, since it will end up | |
189 | * fragmenting the file system's free space. Maybe we | |
190 | * should have some hueristics or some way to allow | |
191 | * userspace to pass a hint to file system, | |
b8d6568a | 192 | * especially if the latter case turns out to be |
ad4fb9ca KM |
193 | * common. |
194 | */ | |
7e028976 | 195 | ex = path[depth].p_ext; |
ad4fb9ca KM |
196 | if (ex) { |
197 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
198 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
199 | ||
200 | if (block > ext_block) | |
201 | return ext_pblk + (block - ext_block); | |
202 | else | |
203 | return ext_pblk - (ext_block - block); | |
204 | } | |
a86c6181 | 205 | |
d0d856e8 RD |
206 | /* it looks like index is empty; |
207 | * try to find starting block from index itself */ | |
a86c6181 AT |
208 | if (path[depth].p_bh) |
209 | return path[depth].p_bh->b_blocknr; | |
210 | } | |
211 | ||
212 | /* OK. use inode's group */ | |
f86186b4 | 213 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
214 | } |
215 | ||
654b4908 AK |
216 | /* |
217 | * Allocation for a meta data block | |
218 | */ | |
f65e6fba | 219 | static ext4_fsblk_t |
654b4908 | 220 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 221 | struct ext4_ext_path *path, |
55f020db | 222 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 223 | { |
f65e6fba | 224 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
225 | |
226 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
227 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
228 | NULL, err); | |
a86c6181 AT |
229 | return newblock; |
230 | } | |
231 | ||
55ad63bf | 232 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
233 | { |
234 | int size; | |
235 | ||
236 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
237 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 238 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
239 | if (!check && size > 6) |
240 | size = 6; | |
a86c6181 AT |
241 | #endif |
242 | return size; | |
243 | } | |
244 | ||
55ad63bf | 245 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
246 | { |
247 | int size; | |
248 | ||
249 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
250 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 251 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
252 | if (!check && size > 5) |
253 | size = 5; | |
a86c6181 AT |
254 | #endif |
255 | return size; | |
256 | } | |
257 | ||
55ad63bf | 258 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
259 | { |
260 | int size; | |
261 | ||
262 | size = sizeof(EXT4_I(inode)->i_data); | |
263 | size -= sizeof(struct ext4_extent_header); | |
264 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 265 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
266 | if (!check && size > 3) |
267 | size = 3; | |
a86c6181 AT |
268 | #endif |
269 | return size; | |
270 | } | |
271 | ||
55ad63bf | 272 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
273 | { |
274 | int size; | |
275 | ||
276 | size = sizeof(EXT4_I(inode)->i_data); | |
277 | size -= sizeof(struct ext4_extent_header); | |
278 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 279 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
280 | if (!check && size > 4) |
281 | size = 4; | |
a86c6181 AT |
282 | #endif |
283 | return size; | |
284 | } | |
285 | ||
fcf6b1b7 DM |
286 | static inline int |
287 | ext4_force_split_extent_at(handle_t *handle, struct inode *inode, | |
dfe50809 | 288 | struct ext4_ext_path **ppath, ext4_lblk_t lblk, |
fcf6b1b7 DM |
289 | int nofail) |
290 | { | |
dfe50809 | 291 | struct ext4_ext_path *path = *ppath; |
fcf6b1b7 DM |
292 | int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext); |
293 | ||
dfe50809 | 294 | return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ? |
fcf6b1b7 DM |
295 | EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0, |
296 | EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO | | |
297 | (nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0)); | |
298 | } | |
299 | ||
d2a17637 MC |
300 | /* |
301 | * Calculate the number of metadata blocks needed | |
302 | * to allocate @blocks | |
303 | * Worse case is one block per extent | |
304 | */ | |
01f49d0b | 305 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 306 | { |
9d0be502 | 307 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 308 | int idxs; |
d2a17637 | 309 | |
9d0be502 TT |
310 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
311 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
312 | |
313 | /* | |
9d0be502 TT |
314 | * If the new delayed allocation block is contiguous with the |
315 | * previous da block, it can share index blocks with the | |
316 | * previous block, so we only need to allocate a new index | |
317 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
318 | * an additional index block, and at ldxs**3 blocks, yet | |
319 | * another index blocks. | |
d2a17637 | 320 | */ |
9d0be502 TT |
321 | if (ei->i_da_metadata_calc_len && |
322 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
323 | int num = 0; |
324 | ||
9d0be502 TT |
325 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
326 | num++; | |
327 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
328 | num++; | |
329 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
330 | num++; | |
331 | ei->i_da_metadata_calc_len = 0; | |
332 | } else | |
333 | ei->i_da_metadata_calc_len++; | |
334 | ei->i_da_metadata_calc_last_lblock++; | |
335 | return num; | |
336 | } | |
d2a17637 | 337 | |
9d0be502 TT |
338 | /* |
339 | * In the worst case we need a new set of index blocks at | |
340 | * every level of the inode's extent tree. | |
341 | */ | |
342 | ei->i_da_metadata_calc_len = 1; | |
343 | ei->i_da_metadata_calc_last_lblock = lblock; | |
344 | return ext_depth(inode) + 1; | |
d2a17637 MC |
345 | } |
346 | ||
c29c0ae7 AT |
347 | static int |
348 | ext4_ext_max_entries(struct inode *inode, int depth) | |
349 | { | |
350 | int max; | |
351 | ||
352 | if (depth == ext_depth(inode)) { | |
353 | if (depth == 0) | |
55ad63bf | 354 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 355 | else |
55ad63bf | 356 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
357 | } else { |
358 | if (depth == 0) | |
55ad63bf | 359 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 360 | else |
55ad63bf | 361 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
362 | } |
363 | ||
364 | return max; | |
365 | } | |
366 | ||
56b19868 AK |
367 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
368 | { | |
bf89d16f | 369 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 370 | int len = ext4_ext_get_actual_len(ext); |
5946d089 | 371 | ext4_lblk_t lblock = le32_to_cpu(ext->ee_block); |
e84a26ce | 372 | |
f70749ca VN |
373 | /* |
374 | * We allow neither: | |
375 | * - zero length | |
376 | * - overflow/wrap-around | |
377 | */ | |
378 | if (lblock + len <= lblock) | |
31d4f3a2 | 379 | return 0; |
6fd058f7 | 380 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
381 | } |
382 | ||
383 | static int ext4_valid_extent_idx(struct inode *inode, | |
384 | struct ext4_extent_idx *ext_idx) | |
385 | { | |
bf89d16f | 386 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 387 | |
6fd058f7 | 388 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
389 | } |
390 | ||
391 | static int ext4_valid_extent_entries(struct inode *inode, | |
392 | struct ext4_extent_header *eh, | |
393 | int depth) | |
394 | { | |
56b19868 AK |
395 | unsigned short entries; |
396 | if (eh->eh_entries == 0) | |
397 | return 1; | |
398 | ||
399 | entries = le16_to_cpu(eh->eh_entries); | |
400 | ||
401 | if (depth == 0) { | |
402 | /* leaf entries */ | |
81fdbb4a | 403 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
5946d089 EG |
404 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
405 | ext4_fsblk_t pblock = 0; | |
406 | ext4_lblk_t lblock = 0; | |
407 | ext4_lblk_t prev = 0; | |
408 | int len = 0; | |
56b19868 AK |
409 | while (entries) { |
410 | if (!ext4_valid_extent(inode, ext)) | |
411 | return 0; | |
5946d089 EG |
412 | |
413 | /* Check for overlapping extents */ | |
414 | lblock = le32_to_cpu(ext->ee_block); | |
415 | len = ext4_ext_get_actual_len(ext); | |
416 | if ((lblock <= prev) && prev) { | |
417 | pblock = ext4_ext_pblock(ext); | |
418 | es->s_last_error_block = cpu_to_le64(pblock); | |
419 | return 0; | |
420 | } | |
56b19868 AK |
421 | ext++; |
422 | entries--; | |
5946d089 | 423 | prev = lblock + len - 1; |
56b19868 AK |
424 | } |
425 | } else { | |
81fdbb4a | 426 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
427 | while (entries) { |
428 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
429 | return 0; | |
430 | ext_idx++; | |
431 | entries--; | |
432 | } | |
433 | } | |
434 | return 1; | |
435 | } | |
436 | ||
c398eda0 TT |
437 | static int __ext4_ext_check(const char *function, unsigned int line, |
438 | struct inode *inode, struct ext4_extent_header *eh, | |
c349179b | 439 | int depth, ext4_fsblk_t pblk) |
c29c0ae7 AT |
440 | { |
441 | const char *error_msg; | |
6a797d27 | 442 | int max = 0, err = -EFSCORRUPTED; |
c29c0ae7 AT |
443 | |
444 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
445 | error_msg = "invalid magic"; | |
446 | goto corrupted; | |
447 | } | |
448 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
449 | error_msg = "unexpected eh_depth"; | |
450 | goto corrupted; | |
451 | } | |
452 | if (unlikely(eh->eh_max == 0)) { | |
453 | error_msg = "invalid eh_max"; | |
454 | goto corrupted; | |
455 | } | |
456 | max = ext4_ext_max_entries(inode, depth); | |
457 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
458 | error_msg = "too large eh_max"; | |
459 | goto corrupted; | |
460 | } | |
461 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
462 | error_msg = "invalid eh_entries"; | |
463 | goto corrupted; | |
464 | } | |
56b19868 AK |
465 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
466 | error_msg = "invalid extent entries"; | |
467 | goto corrupted; | |
468 | } | |
7bc94916 VN |
469 | if (unlikely(depth > 32)) { |
470 | error_msg = "too large eh_depth"; | |
471 | goto corrupted; | |
472 | } | |
7ac5990d DW |
473 | /* Verify checksum on non-root extent tree nodes */ |
474 | if (ext_depth(inode) != depth && | |
475 | !ext4_extent_block_csum_verify(inode, eh)) { | |
476 | error_msg = "extent tree corrupted"; | |
6a797d27 | 477 | err = -EFSBADCRC; |
7ac5990d DW |
478 | goto corrupted; |
479 | } | |
c29c0ae7 AT |
480 | return 0; |
481 | ||
482 | corrupted: | |
c398eda0 | 483 | ext4_error_inode(inode, function, line, 0, |
c349179b TT |
484 | "pblk %llu bad header/extent: %s - magic %x, " |
485 | "entries %u, max %u(%u), depth %u(%u)", | |
486 | (unsigned long long) pblk, error_msg, | |
487 | le16_to_cpu(eh->eh_magic), | |
488 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
489 | max, le16_to_cpu(eh->eh_depth), depth); | |
6a797d27 | 490 | return err; |
c29c0ae7 AT |
491 | } |
492 | ||
c349179b TT |
493 | #define ext4_ext_check(inode, eh, depth, pblk) \ |
494 | __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk)) | |
c29c0ae7 | 495 | |
7a262f7c AK |
496 | int ext4_ext_check_inode(struct inode *inode) |
497 | { | |
c349179b | 498 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
7a262f7c AK |
499 | } |
500 | ||
7d7ea89e TT |
501 | static struct buffer_head * |
502 | __read_extent_tree_block(const char *function, unsigned int line, | |
107a7bd3 TT |
503 | struct inode *inode, ext4_fsblk_t pblk, int depth, |
504 | int flags) | |
f8489128 | 505 | { |
7d7ea89e TT |
506 | struct buffer_head *bh; |
507 | int err; | |
508 | ||
c45653c3 | 509 | bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS); |
7d7ea89e TT |
510 | if (unlikely(!bh)) |
511 | return ERR_PTR(-ENOMEM); | |
f8489128 | 512 | |
7d7ea89e TT |
513 | if (!bh_uptodate_or_lock(bh)) { |
514 | trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); | |
515 | err = bh_submit_read(bh); | |
516 | if (err < 0) | |
517 | goto errout; | |
518 | } | |
7869a4a6 | 519 | if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE)) |
7d7ea89e | 520 | return bh; |
0a944e8a TT |
521 | if (!ext4_has_feature_journal(inode->i_sb) || |
522 | (inode->i_ino != | |
523 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) { | |
524 | err = __ext4_ext_check(function, line, inode, | |
525 | ext_block_hdr(bh), depth, pblk); | |
526 | if (err) | |
527 | goto errout; | |
528 | } | |
f8489128 | 529 | set_buffer_verified(bh); |
107a7bd3 TT |
530 | /* |
531 | * If this is a leaf block, cache all of its entries | |
532 | */ | |
533 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { | |
534 | struct ext4_extent_header *eh = ext_block_hdr(bh); | |
535 | struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); | |
536 | ext4_lblk_t prev = 0; | |
537 | int i; | |
538 | ||
539 | for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { | |
540 | unsigned int status = EXTENT_STATUS_WRITTEN; | |
541 | ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); | |
542 | int len = ext4_ext_get_actual_len(ex); | |
543 | ||
544 | if (prev && (prev != lblk)) | |
545 | ext4_es_cache_extent(inode, prev, | |
546 | lblk - prev, ~0, | |
547 | EXTENT_STATUS_HOLE); | |
548 | ||
556615dc | 549 | if (ext4_ext_is_unwritten(ex)) |
107a7bd3 TT |
550 | status = EXTENT_STATUS_UNWRITTEN; |
551 | ext4_es_cache_extent(inode, lblk, len, | |
552 | ext4_ext_pblock(ex), status); | |
553 | prev = lblk + len; | |
554 | } | |
555 | } | |
7d7ea89e TT |
556 | return bh; |
557 | errout: | |
558 | put_bh(bh); | |
559 | return ERR_PTR(err); | |
560 | ||
f8489128 DW |
561 | } |
562 | ||
107a7bd3 TT |
563 | #define read_extent_tree_block(inode, pblk, depth, flags) \ |
564 | __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \ | |
565 | (depth), (flags)) | |
f8489128 | 566 | |
7869a4a6 TT |
567 | /* |
568 | * This function is called to cache a file's extent information in the | |
569 | * extent status tree | |
570 | */ | |
571 | int ext4_ext_precache(struct inode *inode) | |
572 | { | |
573 | struct ext4_inode_info *ei = EXT4_I(inode); | |
574 | struct ext4_ext_path *path = NULL; | |
575 | struct buffer_head *bh; | |
576 | int i = 0, depth, ret = 0; | |
577 | ||
578 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
579 | return 0; /* not an extent-mapped inode */ | |
580 | ||
581 | down_read(&ei->i_data_sem); | |
582 | depth = ext_depth(inode); | |
583 | ||
6396bb22 | 584 | path = kcalloc(depth + 1, sizeof(struct ext4_ext_path), |
7869a4a6 TT |
585 | GFP_NOFS); |
586 | if (path == NULL) { | |
587 | up_read(&ei->i_data_sem); | |
588 | return -ENOMEM; | |
589 | } | |
590 | ||
591 | /* Don't cache anything if there are no external extent blocks */ | |
592 | if (depth == 0) | |
593 | goto out; | |
594 | path[0].p_hdr = ext_inode_hdr(inode); | |
595 | ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0); | |
596 | if (ret) | |
597 | goto out; | |
598 | path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr); | |
599 | while (i >= 0) { | |
600 | /* | |
601 | * If this is a leaf block or we've reached the end of | |
602 | * the index block, go up | |
603 | */ | |
604 | if ((i == depth) || | |
605 | path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) { | |
606 | brelse(path[i].p_bh); | |
607 | path[i].p_bh = NULL; | |
608 | i--; | |
609 | continue; | |
610 | } | |
611 | bh = read_extent_tree_block(inode, | |
612 | ext4_idx_pblock(path[i].p_idx++), | |
613 | depth - i - 1, | |
614 | EXT4_EX_FORCE_CACHE); | |
615 | if (IS_ERR(bh)) { | |
616 | ret = PTR_ERR(bh); | |
617 | break; | |
618 | } | |
619 | i++; | |
620 | path[i].p_bh = bh; | |
621 | path[i].p_hdr = ext_block_hdr(bh); | |
622 | path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr); | |
623 | } | |
624 | ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED); | |
625 | out: | |
626 | up_read(&ei->i_data_sem); | |
627 | ext4_ext_drop_refs(path); | |
628 | kfree(path); | |
629 | return ret; | |
630 | } | |
631 | ||
a86c6181 AT |
632 | #ifdef EXT_DEBUG |
633 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
634 | { | |
635 | int k, l = path->p_depth; | |
636 | ||
637 | ext_debug("path:"); | |
638 | for (k = 0; k <= l; k++, path++) { | |
639 | if (path->p_idx) { | |
2ae02107 | 640 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 641 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 642 | } else if (path->p_ext) { |
553f9008 | 643 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 644 | le32_to_cpu(path->p_ext->ee_block), |
556615dc | 645 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 646 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 647 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
648 | } else |
649 | ext_debug(" []"); | |
650 | } | |
651 | ext_debug("\n"); | |
652 | } | |
653 | ||
654 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
655 | { | |
656 | int depth = ext_depth(inode); | |
657 | struct ext4_extent_header *eh; | |
658 | struct ext4_extent *ex; | |
659 | int i; | |
660 | ||
661 | if (!path) | |
662 | return; | |
663 | ||
664 | eh = path[depth].p_hdr; | |
665 | ex = EXT_FIRST_EXTENT(eh); | |
666 | ||
553f9008 M |
667 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
668 | ||
a86c6181 | 669 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 | 670 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
556615dc | 671 | ext4_ext_is_unwritten(ex), |
bf89d16f | 672 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
673 | } |
674 | ext_debug("\n"); | |
675 | } | |
1b16da77 YY |
676 | |
677 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
678 | ext4_fsblk_t newblock, int level) | |
679 | { | |
680 | int depth = ext_depth(inode); | |
681 | struct ext4_extent *ex; | |
682 | ||
683 | if (depth != level) { | |
684 | struct ext4_extent_idx *idx; | |
685 | idx = path[level].p_idx; | |
686 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
687 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
688 | le32_to_cpu(idx->ei_block), | |
689 | ext4_idx_pblock(idx), | |
690 | newblock); | |
691 | idx++; | |
692 | } | |
693 | ||
694 | return; | |
695 | } | |
696 | ||
697 | ex = path[depth].p_ext; | |
698 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
699 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
700 | le32_to_cpu(ex->ee_block), | |
701 | ext4_ext_pblock(ex), | |
556615dc | 702 | ext4_ext_is_unwritten(ex), |
1b16da77 YY |
703 | ext4_ext_get_actual_len(ex), |
704 | newblock); | |
705 | ex++; | |
706 | } | |
707 | } | |
708 | ||
a86c6181 | 709 | #else |
af5bc92d TT |
710 | #define ext4_ext_show_path(inode, path) |
711 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 712 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
713 | #endif |
714 | ||
b35905c1 | 715 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 | 716 | { |
b7ea89ad | 717 | int depth, i; |
a86c6181 | 718 | |
b7ea89ad TT |
719 | if (!path) |
720 | return; | |
721 | depth = path->p_depth; | |
a86c6181 AT |
722 | for (i = 0; i <= depth; i++, path++) |
723 | if (path->p_bh) { | |
724 | brelse(path->p_bh); | |
725 | path->p_bh = NULL; | |
726 | } | |
727 | } | |
728 | ||
729 | /* | |
d0d856e8 RD |
730 | * ext4_ext_binsearch_idx: |
731 | * binary search for the closest index of the given block | |
c29c0ae7 | 732 | * the header must be checked before calling this |
a86c6181 AT |
733 | */ |
734 | static void | |
725d26d3 AK |
735 | ext4_ext_binsearch_idx(struct inode *inode, |
736 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
737 | { |
738 | struct ext4_extent_header *eh = path->p_hdr; | |
739 | struct ext4_extent_idx *r, *l, *m; | |
740 | ||
a86c6181 | 741 | |
bba90743 | 742 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
743 | |
744 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 745 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
746 | while (l <= r) { |
747 | m = l + (r - l) / 2; | |
748 | if (block < le32_to_cpu(m->ei_block)) | |
749 | r = m - 1; | |
750 | else | |
751 | l = m + 1; | |
26d535ed DM |
752 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
753 | m, le32_to_cpu(m->ei_block), | |
754 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
755 | } |
756 | ||
757 | path->p_idx = l - 1; | |
4a3c3a51 | 758 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 759 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
760 | |
761 | #ifdef CHECK_BINSEARCH | |
762 | { | |
763 | struct ext4_extent_idx *chix, *ix; | |
764 | int k; | |
765 | ||
766 | chix = ix = EXT_FIRST_INDEX(eh); | |
767 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
768 | if (k != 0 && | |
769 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
770 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
771 | "first=0x%p\n", k, | |
772 | ix, EXT_FIRST_INDEX(eh)); | |
773 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
774 | le32_to_cpu(ix->ei_block), |
775 | le32_to_cpu(ix[-1].ei_block)); | |
776 | } | |
777 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 778 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
779 | if (block < le32_to_cpu(ix->ei_block)) |
780 | break; | |
781 | chix = ix; | |
782 | } | |
783 | BUG_ON(chix != path->p_idx); | |
784 | } | |
785 | #endif | |
786 | ||
787 | } | |
788 | ||
789 | /* | |
d0d856e8 RD |
790 | * ext4_ext_binsearch: |
791 | * binary search for closest extent of the given block | |
c29c0ae7 | 792 | * the header must be checked before calling this |
a86c6181 AT |
793 | */ |
794 | static void | |
725d26d3 AK |
795 | ext4_ext_binsearch(struct inode *inode, |
796 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
797 | { |
798 | struct ext4_extent_header *eh = path->p_hdr; | |
799 | struct ext4_extent *r, *l, *m; | |
800 | ||
a86c6181 AT |
801 | if (eh->eh_entries == 0) { |
802 | /* | |
d0d856e8 RD |
803 | * this leaf is empty: |
804 | * we get such a leaf in split/add case | |
a86c6181 AT |
805 | */ |
806 | return; | |
807 | } | |
808 | ||
bba90743 | 809 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
810 | |
811 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 812 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
813 | |
814 | while (l <= r) { | |
815 | m = l + (r - l) / 2; | |
816 | if (block < le32_to_cpu(m->ee_block)) | |
817 | r = m - 1; | |
818 | else | |
819 | l = m + 1; | |
26d535ed DM |
820 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
821 | m, le32_to_cpu(m->ee_block), | |
822 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
823 | } |
824 | ||
825 | path->p_ext = l - 1; | |
553f9008 | 826 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 827 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 828 | ext4_ext_pblock(path->p_ext), |
556615dc | 829 | ext4_ext_is_unwritten(path->p_ext), |
a2df2a63 | 830 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
831 | |
832 | #ifdef CHECK_BINSEARCH | |
833 | { | |
834 | struct ext4_extent *chex, *ex; | |
835 | int k; | |
836 | ||
837 | chex = ex = EXT_FIRST_EXTENT(eh); | |
838 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
839 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 840 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
841 | if (block < le32_to_cpu(ex->ee_block)) |
842 | break; | |
843 | chex = ex; | |
844 | } | |
845 | BUG_ON(chex != path->p_ext); | |
846 | } | |
847 | #endif | |
848 | ||
849 | } | |
850 | ||
851 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
852 | { | |
853 | struct ext4_extent_header *eh; | |
854 | ||
855 | eh = ext_inode_hdr(inode); | |
856 | eh->eh_depth = 0; | |
857 | eh->eh_entries = 0; | |
858 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 859 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 860 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
861 | return 0; |
862 | } | |
863 | ||
864 | struct ext4_ext_path * | |
ed8a1a76 TT |
865 | ext4_find_extent(struct inode *inode, ext4_lblk_t block, |
866 | struct ext4_ext_path **orig_path, int flags) | |
a86c6181 AT |
867 | { |
868 | struct ext4_extent_header *eh; | |
869 | struct buffer_head *bh; | |
705912ca TT |
870 | struct ext4_ext_path *path = orig_path ? *orig_path : NULL; |
871 | short int depth, i, ppos = 0; | |
860d21e2 | 872 | int ret; |
a86c6181 AT |
873 | |
874 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 875 | depth = ext_depth(inode); |
bc890a60 TT |
876 | if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) { |
877 | EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d", | |
878 | depth); | |
879 | ret = -EFSCORRUPTED; | |
880 | goto err; | |
881 | } | |
a86c6181 | 882 | |
10809df8 | 883 | if (path) { |
523f431c | 884 | ext4_ext_drop_refs(path); |
10809df8 TT |
885 | if (depth > path[0].p_maxdepth) { |
886 | kfree(path); | |
887 | *orig_path = path = NULL; | |
888 | } | |
889 | } | |
890 | if (!path) { | |
523f431c | 891 | /* account possible depth increase */ |
6396bb22 | 892 | path = kcalloc(depth + 2, sizeof(struct ext4_ext_path), |
a86c6181 | 893 | GFP_NOFS); |
19008f6d | 894 | if (unlikely(!path)) |
a86c6181 | 895 | return ERR_PTR(-ENOMEM); |
10809df8 | 896 | path[0].p_maxdepth = depth + 1; |
a86c6181 | 897 | } |
a86c6181 | 898 | path[0].p_hdr = eh; |
1973adcb | 899 | path[0].p_bh = NULL; |
a86c6181 | 900 | |
c29c0ae7 | 901 | i = depth; |
a86c6181 AT |
902 | /* walk through the tree */ |
903 | while (i) { | |
904 | ext_debug("depth %d: num %d, max %d\n", | |
905 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 906 | |
a86c6181 | 907 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 908 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
909 | path[ppos].p_depth = i; |
910 | path[ppos].p_ext = NULL; | |
911 | ||
107a7bd3 TT |
912 | bh = read_extent_tree_block(inode, path[ppos].p_block, --i, |
913 | flags); | |
a1c83681 | 914 | if (IS_ERR(bh)) { |
7d7ea89e | 915 | ret = PTR_ERR(bh); |
a86c6181 | 916 | goto err; |
860d21e2 | 917 | } |
7d7ea89e | 918 | |
a86c6181 AT |
919 | eh = ext_block_hdr(bh); |
920 | ppos++; | |
a86c6181 AT |
921 | path[ppos].p_bh = bh; |
922 | path[ppos].p_hdr = eh; | |
a86c6181 AT |
923 | } |
924 | ||
925 | path[ppos].p_depth = i; | |
a86c6181 AT |
926 | path[ppos].p_ext = NULL; |
927 | path[ppos].p_idx = NULL; | |
928 | ||
a86c6181 AT |
929 | /* find extent */ |
930 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
931 | /* if not an empty leaf */ |
932 | if (path[ppos].p_ext) | |
bf89d16f | 933 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
934 | |
935 | ext4_ext_show_path(inode, path); | |
936 | ||
937 | return path; | |
938 | ||
939 | err: | |
940 | ext4_ext_drop_refs(path); | |
dfe50809 TT |
941 | kfree(path); |
942 | if (orig_path) | |
943 | *orig_path = NULL; | |
860d21e2 | 944 | return ERR_PTR(ret); |
a86c6181 AT |
945 | } |
946 | ||
947 | /* | |
d0d856e8 RD |
948 | * ext4_ext_insert_index: |
949 | * insert new index [@logical;@ptr] into the block at @curp; | |
950 | * check where to insert: before @curp or after @curp | |
a86c6181 | 951 | */ |
1f109d5a TT |
952 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
953 | struct ext4_ext_path *curp, | |
954 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
955 | { |
956 | struct ext4_extent_idx *ix; | |
957 | int len, err; | |
958 | ||
7e028976 AM |
959 | err = ext4_ext_get_access(handle, inode, curp); |
960 | if (err) | |
a86c6181 AT |
961 | return err; |
962 | ||
273df556 FM |
963 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
964 | EXT4_ERROR_INODE(inode, | |
965 | "logical %d == ei_block %d!", | |
966 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
6a797d27 | 967 | return -EFSCORRUPTED; |
273df556 | 968 | } |
d4620315 RD |
969 | |
970 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
971 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
972 | EXT4_ERROR_INODE(inode, | |
973 | "eh_entries %d >= eh_max %d!", | |
974 | le16_to_cpu(curp->p_hdr->eh_entries), | |
975 | le16_to_cpu(curp->p_hdr->eh_max)); | |
6a797d27 | 976 | return -EFSCORRUPTED; |
d4620315 RD |
977 | } |
978 | ||
a86c6181 AT |
979 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
980 | /* insert after */ | |
80e675f9 | 981 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
982 | ix = curp->p_idx + 1; |
983 | } else { | |
984 | /* insert before */ | |
80e675f9 | 985 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
986 | ix = curp->p_idx; |
987 | } | |
988 | ||
80e675f9 EG |
989 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
990 | BUG_ON(len < 0); | |
991 | if (len > 0) { | |
992 | ext_debug("insert new index %d: " | |
993 | "move %d indices from 0x%p to 0x%p\n", | |
994 | logical, len, ix, ix + 1); | |
995 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
996 | } | |
997 | ||
f472e026 TM |
998 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
999 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
6a797d27 | 1000 | return -EFSCORRUPTED; |
f472e026 TM |
1001 | } |
1002 | ||
a86c6181 | 1003 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 1004 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 1005 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 1006 | |
273df556 FM |
1007 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
1008 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
6a797d27 | 1009 | return -EFSCORRUPTED; |
273df556 | 1010 | } |
a86c6181 AT |
1011 | |
1012 | err = ext4_ext_dirty(handle, inode, curp); | |
1013 | ext4_std_error(inode->i_sb, err); | |
1014 | ||
1015 | return err; | |
1016 | } | |
1017 | ||
1018 | /* | |
d0d856e8 RD |
1019 | * ext4_ext_split: |
1020 | * inserts new subtree into the path, using free index entry | |
1021 | * at depth @at: | |
1022 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
1023 | * - makes decision where to split | |
1024 | * - moves remaining extents and index entries (right to the split point) | |
1025 | * into the newly allocated blocks | |
1026 | * - initializes subtree | |
a86c6181 AT |
1027 | */ |
1028 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
1029 | unsigned int flags, |
1030 | struct ext4_ext_path *path, | |
1031 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
1032 | { |
1033 | struct buffer_head *bh = NULL; | |
1034 | int depth = ext_depth(inode); | |
1035 | struct ext4_extent_header *neh; | |
1036 | struct ext4_extent_idx *fidx; | |
a86c6181 | 1037 | int i = at, k, m, a; |
f65e6fba | 1038 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 1039 | __le32 border; |
f65e6fba | 1040 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 | 1041 | int err = 0; |
592acbf1 | 1042 | size_t ext_size = 0; |
a86c6181 AT |
1043 | |
1044 | /* make decision: where to split? */ | |
d0d856e8 | 1045 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 1046 | |
d0d856e8 | 1047 | /* if current leaf will be split, then we should use |
a86c6181 | 1048 | * border from split point */ |
273df556 FM |
1049 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
1050 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
6a797d27 | 1051 | return -EFSCORRUPTED; |
273df556 | 1052 | } |
a86c6181 AT |
1053 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
1054 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 1055 | ext_debug("leaf will be split." |
a86c6181 | 1056 | " next leaf starts at %d\n", |
8c55e204 | 1057 | le32_to_cpu(border)); |
a86c6181 AT |
1058 | } else { |
1059 | border = newext->ee_block; | |
1060 | ext_debug("leaf will be added." | |
1061 | " next leaf starts at %d\n", | |
8c55e204 | 1062 | le32_to_cpu(border)); |
a86c6181 AT |
1063 | } |
1064 | ||
1065 | /* | |
d0d856e8 RD |
1066 | * If error occurs, then we break processing |
1067 | * and mark filesystem read-only. index won't | |
a86c6181 | 1068 | * be inserted and tree will be in consistent |
d0d856e8 | 1069 | * state. Next mount will repair buffers too. |
a86c6181 AT |
1070 | */ |
1071 | ||
1072 | /* | |
d0d856e8 RD |
1073 | * Get array to track all allocated blocks. |
1074 | * We need this to handle errors and free blocks | |
1075 | * upon them. | |
a86c6181 | 1076 | */ |
6396bb22 | 1077 | ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS); |
a86c6181 AT |
1078 | if (!ablocks) |
1079 | return -ENOMEM; | |
a86c6181 AT |
1080 | |
1081 | /* allocate all needed blocks */ | |
1082 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
1083 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 1084 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 1085 | newext, &err, flags); |
a86c6181 AT |
1086 | if (newblock == 0) |
1087 | goto cleanup; | |
1088 | ablocks[a] = newblock; | |
1089 | } | |
1090 | ||
1091 | /* initialize new leaf */ | |
1092 | newblock = ablocks[--a]; | |
273df556 FM |
1093 | if (unlikely(newblock == 0)) { |
1094 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
6a797d27 | 1095 | err = -EFSCORRUPTED; |
273df556 FM |
1096 | goto cleanup; |
1097 | } | |
c45653c3 | 1098 | bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS); |
aebf0243 | 1099 | if (unlikely(!bh)) { |
860d21e2 | 1100 | err = -ENOMEM; |
a86c6181 AT |
1101 | goto cleanup; |
1102 | } | |
1103 | lock_buffer(bh); | |
1104 | ||
7e028976 AM |
1105 | err = ext4_journal_get_create_access(handle, bh); |
1106 | if (err) | |
a86c6181 AT |
1107 | goto cleanup; |
1108 | ||
1109 | neh = ext_block_hdr(bh); | |
1110 | neh->eh_entries = 0; | |
55ad63bf | 1111 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
1112 | neh->eh_magic = EXT4_EXT_MAGIC; |
1113 | neh->eh_depth = 0; | |
a86c6181 | 1114 | |
d0d856e8 | 1115 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
1116 | if (unlikely(path[depth].p_hdr->eh_entries != |
1117 | path[depth].p_hdr->eh_max)) { | |
1118 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
1119 | path[depth].p_hdr->eh_entries, | |
1120 | path[depth].p_hdr->eh_max); | |
6a797d27 | 1121 | err = -EFSCORRUPTED; |
273df556 FM |
1122 | goto cleanup; |
1123 | } | |
a86c6181 | 1124 | /* start copy from next extent */ |
1b16da77 YY |
1125 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
1126 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 1127 | if (m) { |
1b16da77 YY |
1128 | struct ext4_extent *ex; |
1129 | ex = EXT_FIRST_EXTENT(neh); | |
1130 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 1131 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1132 | } |
1133 | ||
592acbf1 SR |
1134 | /* zero out unused area in the extent block */ |
1135 | ext_size = sizeof(struct ext4_extent_header) + | |
1136 | sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries); | |
1137 | memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size); | |
7ac5990d | 1138 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1139 | set_buffer_uptodate(bh); |
1140 | unlock_buffer(bh); | |
1141 | ||
0390131b | 1142 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1143 | if (err) |
a86c6181 AT |
1144 | goto cleanup; |
1145 | brelse(bh); | |
1146 | bh = NULL; | |
1147 | ||
1148 | /* correct old leaf */ | |
1149 | if (m) { | |
7e028976 AM |
1150 | err = ext4_ext_get_access(handle, inode, path + depth); |
1151 | if (err) | |
a86c6181 | 1152 | goto cleanup; |
e8546d06 | 1153 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1154 | err = ext4_ext_dirty(handle, inode, path + depth); |
1155 | if (err) | |
a86c6181 AT |
1156 | goto cleanup; |
1157 | ||
1158 | } | |
1159 | ||
1160 | /* create intermediate indexes */ | |
1161 | k = depth - at - 1; | |
273df556 FM |
1162 | if (unlikely(k < 0)) { |
1163 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
6a797d27 | 1164 | err = -EFSCORRUPTED; |
273df556 FM |
1165 | goto cleanup; |
1166 | } | |
a86c6181 AT |
1167 | if (k) |
1168 | ext_debug("create %d intermediate indices\n", k); | |
1169 | /* insert new index into current index block */ | |
1170 | /* current depth stored in i var */ | |
1171 | i = depth - 1; | |
1172 | while (k--) { | |
1173 | oldblock = newblock; | |
1174 | newblock = ablocks[--a]; | |
bba90743 | 1175 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1176 | if (unlikely(!bh)) { |
860d21e2 | 1177 | err = -ENOMEM; |
a86c6181 AT |
1178 | goto cleanup; |
1179 | } | |
1180 | lock_buffer(bh); | |
1181 | ||
7e028976 AM |
1182 | err = ext4_journal_get_create_access(handle, bh); |
1183 | if (err) | |
a86c6181 AT |
1184 | goto cleanup; |
1185 | ||
1186 | neh = ext_block_hdr(bh); | |
1187 | neh->eh_entries = cpu_to_le16(1); | |
1188 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1189 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1190 | neh->eh_depth = cpu_to_le16(depth - i); |
1191 | fidx = EXT_FIRST_INDEX(neh); | |
1192 | fidx->ei_block = border; | |
f65e6fba | 1193 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1194 | |
bba90743 ES |
1195 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1196 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1197 | |
1b16da77 | 1198 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1199 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1200 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1201 | EXT4_ERROR_INODE(inode, | |
1202 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1203 | le32_to_cpu(path[i].p_ext->ee_block)); | |
6a797d27 | 1204 | err = -EFSCORRUPTED; |
273df556 FM |
1205 | goto cleanup; |
1206 | } | |
1b16da77 YY |
1207 | /* start copy indexes */ |
1208 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1209 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1210 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1211 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1212 | if (m) { |
1b16da77 | 1213 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1214 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1215 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1216 | } |
592acbf1 SR |
1217 | /* zero out unused area in the extent block */ |
1218 | ext_size = sizeof(struct ext4_extent_header) + | |
1219 | (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries)); | |
1220 | memset(bh->b_data + ext_size, 0, | |
1221 | inode->i_sb->s_blocksize - ext_size); | |
7ac5990d | 1222 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1223 | set_buffer_uptodate(bh); |
1224 | unlock_buffer(bh); | |
1225 | ||
0390131b | 1226 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1227 | if (err) |
a86c6181 AT |
1228 | goto cleanup; |
1229 | brelse(bh); | |
1230 | bh = NULL; | |
1231 | ||
1232 | /* correct old index */ | |
1233 | if (m) { | |
1234 | err = ext4_ext_get_access(handle, inode, path + i); | |
1235 | if (err) | |
1236 | goto cleanup; | |
e8546d06 | 1237 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1238 | err = ext4_ext_dirty(handle, inode, path + i); |
1239 | if (err) | |
1240 | goto cleanup; | |
1241 | } | |
1242 | ||
1243 | i--; | |
1244 | } | |
1245 | ||
1246 | /* insert new index */ | |
a86c6181 AT |
1247 | err = ext4_ext_insert_index(handle, inode, path + at, |
1248 | le32_to_cpu(border), newblock); | |
1249 | ||
1250 | cleanup: | |
1251 | if (bh) { | |
1252 | if (buffer_locked(bh)) | |
1253 | unlock_buffer(bh); | |
1254 | brelse(bh); | |
1255 | } | |
1256 | ||
1257 | if (err) { | |
1258 | /* free all allocated blocks in error case */ | |
1259 | for (i = 0; i < depth; i++) { | |
1260 | if (!ablocks[i]) | |
1261 | continue; | |
7dc57615 | 1262 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1263 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1264 | } |
1265 | } | |
1266 | kfree(ablocks); | |
1267 | ||
1268 | return err; | |
1269 | } | |
1270 | ||
1271 | /* | |
d0d856e8 RD |
1272 | * ext4_ext_grow_indepth: |
1273 | * implements tree growing procedure: | |
1274 | * - allocates new block | |
1275 | * - moves top-level data (index block or leaf) into the new block | |
1276 | * - initializes new top-level, creating index that points to the | |
1277 | * just created block | |
a86c6181 AT |
1278 | */ |
1279 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
be5cd90d | 1280 | unsigned int flags) |
a86c6181 | 1281 | { |
a86c6181 | 1282 | struct ext4_extent_header *neh; |
a86c6181 | 1283 | struct buffer_head *bh; |
be5cd90d DM |
1284 | ext4_fsblk_t newblock, goal = 0; |
1285 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; | |
a86c6181 | 1286 | int err = 0; |
592acbf1 | 1287 | size_t ext_size = 0; |
a86c6181 | 1288 | |
be5cd90d DM |
1289 | /* Try to prepend new index to old one */ |
1290 | if (ext_depth(inode)) | |
1291 | goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode))); | |
1292 | if (goal > le32_to_cpu(es->s_first_data_block)) { | |
1293 | flags |= EXT4_MB_HINT_TRY_GOAL; | |
1294 | goal--; | |
1295 | } else | |
1296 | goal = ext4_inode_to_goal_block(inode); | |
1297 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, | |
1298 | NULL, &err); | |
a86c6181 AT |
1299 | if (newblock == 0) |
1300 | return err; | |
1301 | ||
c45653c3 | 1302 | bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS); |
aebf0243 | 1303 | if (unlikely(!bh)) |
860d21e2 | 1304 | return -ENOMEM; |
a86c6181 AT |
1305 | lock_buffer(bh); |
1306 | ||
7e028976 AM |
1307 | err = ext4_journal_get_create_access(handle, bh); |
1308 | if (err) { | |
a86c6181 AT |
1309 | unlock_buffer(bh); |
1310 | goto out; | |
1311 | } | |
1312 | ||
592acbf1 | 1313 | ext_size = sizeof(EXT4_I(inode)->i_data); |
a86c6181 | 1314 | /* move top-level index/leaf into new block */ |
592acbf1 SR |
1315 | memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size); |
1316 | /* zero out unused area in the extent block */ | |
1317 | memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size); | |
a86c6181 AT |
1318 | |
1319 | /* set size of new block */ | |
1320 | neh = ext_block_hdr(bh); | |
1321 | /* old root could have indexes or leaves | |
1322 | * so calculate e_max right way */ | |
1323 | if (ext_depth(inode)) | |
55ad63bf | 1324 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1325 | else |
55ad63bf | 1326 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1327 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1328 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1329 | set_buffer_uptodate(bh); |
1330 | unlock_buffer(bh); | |
1331 | ||
0390131b | 1332 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1333 | if (err) |
a86c6181 AT |
1334 | goto out; |
1335 | ||
1939dd84 | 1336 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1337 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1338 | neh->eh_entries = cpu_to_le16(1); |
1339 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1340 | if (neh->eh_depth == 0) { | |
1341 | /* Root extent block becomes index block */ | |
1342 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1343 | EXT_FIRST_INDEX(neh)->ei_block = | |
1344 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1345 | } | |
2ae02107 | 1346 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1347 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1348 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1349 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1350 | |
ba39ebb6 | 1351 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1352 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1353 | out: |
1354 | brelse(bh); | |
1355 | ||
1356 | return err; | |
1357 | } | |
1358 | ||
1359 | /* | |
d0d856e8 RD |
1360 | * ext4_ext_create_new_leaf: |
1361 | * finds empty index and adds new leaf. | |
1362 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1363 | */ |
1364 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
107a7bd3 TT |
1365 | unsigned int mb_flags, |
1366 | unsigned int gb_flags, | |
dfe50809 | 1367 | struct ext4_ext_path **ppath, |
55f020db | 1368 | struct ext4_extent *newext) |
a86c6181 | 1369 | { |
dfe50809 | 1370 | struct ext4_ext_path *path = *ppath; |
a86c6181 AT |
1371 | struct ext4_ext_path *curp; |
1372 | int depth, i, err = 0; | |
1373 | ||
1374 | repeat: | |
1375 | i = depth = ext_depth(inode); | |
1376 | ||
1377 | /* walk up to the tree and look for free index entry */ | |
1378 | curp = path + depth; | |
1379 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1380 | i--; | |
1381 | curp--; | |
1382 | } | |
1383 | ||
d0d856e8 RD |
1384 | /* we use already allocated block for index block, |
1385 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1386 | if (EXT_HAS_FREE_INDEX(curp)) { |
1387 | /* if we found index with free entry, then use that | |
1388 | * entry: create all needed subtree and add new leaf */ | |
107a7bd3 | 1389 | err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
787e0981 SF |
1390 | if (err) |
1391 | goto out; | |
a86c6181 AT |
1392 | |
1393 | /* refill path */ | |
ed8a1a76 | 1394 | path = ext4_find_extent(inode, |
725d26d3 | 1395 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
dfe50809 | 1396 | ppath, gb_flags); |
a86c6181 AT |
1397 | if (IS_ERR(path)) |
1398 | err = PTR_ERR(path); | |
1399 | } else { | |
1400 | /* tree is full, time to grow in depth */ | |
be5cd90d | 1401 | err = ext4_ext_grow_indepth(handle, inode, mb_flags); |
a86c6181 AT |
1402 | if (err) |
1403 | goto out; | |
1404 | ||
1405 | /* refill path */ | |
ed8a1a76 | 1406 | path = ext4_find_extent(inode, |
725d26d3 | 1407 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
dfe50809 | 1408 | ppath, gb_flags); |
a86c6181 AT |
1409 | if (IS_ERR(path)) { |
1410 | err = PTR_ERR(path); | |
1411 | goto out; | |
1412 | } | |
1413 | ||
1414 | /* | |
d0d856e8 RD |
1415 | * only first (depth 0 -> 1) produces free space; |
1416 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1417 | */ |
1418 | depth = ext_depth(inode); | |
1419 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1420 | /* now we need to split */ |
a86c6181 AT |
1421 | goto repeat; |
1422 | } | |
1423 | } | |
1424 | ||
1425 | out: | |
1426 | return err; | |
1427 | } | |
1428 | ||
1988b51e AT |
1429 | /* |
1430 | * search the closest allocated block to the left for *logical | |
1431 | * and returns it at @logical + it's physical address at @phys | |
1432 | * if *logical is the smallest allocated block, the function | |
1433 | * returns 0 at @phys | |
1434 | * return value contains 0 (success) or error code | |
1435 | */ | |
1f109d5a TT |
1436 | static int ext4_ext_search_left(struct inode *inode, |
1437 | struct ext4_ext_path *path, | |
1438 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1439 | { |
1440 | struct ext4_extent_idx *ix; | |
1441 | struct ext4_extent *ex; | |
b939e376 | 1442 | int depth, ee_len; |
1988b51e | 1443 | |
273df556 FM |
1444 | if (unlikely(path == NULL)) { |
1445 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
6a797d27 | 1446 | return -EFSCORRUPTED; |
273df556 | 1447 | } |
1988b51e AT |
1448 | depth = path->p_depth; |
1449 | *phys = 0; | |
1450 | ||
1451 | if (depth == 0 && path->p_ext == NULL) | |
1452 | return 0; | |
1453 | ||
1454 | /* usually extent in the path covers blocks smaller | |
1455 | * then *logical, but it can be that extent is the | |
1456 | * first one in the file */ | |
1457 | ||
1458 | ex = path[depth].p_ext; | |
b939e376 | 1459 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1460 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1461 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1462 | EXT4_ERROR_INODE(inode, | |
1463 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1464 | *logical, le32_to_cpu(ex->ee_block)); | |
6a797d27 | 1465 | return -EFSCORRUPTED; |
273df556 | 1466 | } |
1988b51e AT |
1467 | while (--depth >= 0) { |
1468 | ix = path[depth].p_idx; | |
273df556 FM |
1469 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1470 | EXT4_ERROR_INODE(inode, | |
1471 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1472 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1473 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1474 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 | 1475 | depth); |
6a797d27 | 1476 | return -EFSCORRUPTED; |
273df556 | 1477 | } |
1988b51e AT |
1478 | } |
1479 | return 0; | |
1480 | } | |
1481 | ||
273df556 FM |
1482 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1483 | EXT4_ERROR_INODE(inode, | |
1484 | "logical %d < ee_block %d + ee_len %d!", | |
1485 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
6a797d27 | 1486 | return -EFSCORRUPTED; |
273df556 | 1487 | } |
1988b51e | 1488 | |
b939e376 | 1489 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1490 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1491 | return 0; |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * search the closest allocated block to the right for *logical | |
1496 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1497 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1498 | * returns 0 at @phys |
1499 | * return value contains 0 (success) or error code | |
1500 | */ | |
1f109d5a TT |
1501 | static int ext4_ext_search_right(struct inode *inode, |
1502 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1503 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1504 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1505 | { |
1506 | struct buffer_head *bh = NULL; | |
1507 | struct ext4_extent_header *eh; | |
1508 | struct ext4_extent_idx *ix; | |
1509 | struct ext4_extent *ex; | |
1510 | ext4_fsblk_t block; | |
395a87bf ES |
1511 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1512 | int ee_len; | |
1988b51e | 1513 | |
273df556 FM |
1514 | if (unlikely(path == NULL)) { |
1515 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
6a797d27 | 1516 | return -EFSCORRUPTED; |
273df556 | 1517 | } |
1988b51e AT |
1518 | depth = path->p_depth; |
1519 | *phys = 0; | |
1520 | ||
1521 | if (depth == 0 && path->p_ext == NULL) | |
1522 | return 0; | |
1523 | ||
1524 | /* usually extent in the path covers blocks smaller | |
1525 | * then *logical, but it can be that extent is the | |
1526 | * first one in the file */ | |
1527 | ||
1528 | ex = path[depth].p_ext; | |
b939e376 | 1529 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1530 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1531 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1532 | EXT4_ERROR_INODE(inode, | |
1533 | "first_extent(path[%d].p_hdr) != ex", | |
1534 | depth); | |
6a797d27 | 1535 | return -EFSCORRUPTED; |
273df556 | 1536 | } |
1988b51e AT |
1537 | while (--depth >= 0) { |
1538 | ix = path[depth].p_idx; | |
273df556 FM |
1539 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1540 | EXT4_ERROR_INODE(inode, | |
1541 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1542 | *logical); | |
6a797d27 | 1543 | return -EFSCORRUPTED; |
273df556 | 1544 | } |
1988b51e | 1545 | } |
4d33b1ef | 1546 | goto found_extent; |
1988b51e AT |
1547 | } |
1548 | ||
273df556 FM |
1549 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1550 | EXT4_ERROR_INODE(inode, | |
1551 | "logical %d < ee_block %d + ee_len %d!", | |
1552 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
6a797d27 | 1553 | return -EFSCORRUPTED; |
273df556 | 1554 | } |
1988b51e AT |
1555 | |
1556 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1557 | /* next allocated block in this leaf */ | |
1558 | ex++; | |
4d33b1ef | 1559 | goto found_extent; |
1988b51e AT |
1560 | } |
1561 | ||
1562 | /* go up and search for index to the right */ | |
1563 | while (--depth >= 0) { | |
1564 | ix = path[depth].p_idx; | |
1565 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1566 | goto got_index; |
1988b51e AT |
1567 | } |
1568 | ||
25f1ee3a WF |
1569 | /* we've gone up to the root and found no index to the right */ |
1570 | return 0; | |
1988b51e | 1571 | |
25f1ee3a | 1572 | got_index: |
1988b51e AT |
1573 | /* we've found index to the right, let's |
1574 | * follow it and find the closest allocated | |
1575 | * block to the right */ | |
1576 | ix++; | |
bf89d16f | 1577 | block = ext4_idx_pblock(ix); |
1988b51e | 1578 | while (++depth < path->p_depth) { |
395a87bf | 1579 | /* subtract from p_depth to get proper eh_depth */ |
7d7ea89e | 1580 | bh = read_extent_tree_block(inode, block, |
107a7bd3 | 1581 | path->p_depth - depth, 0); |
7d7ea89e TT |
1582 | if (IS_ERR(bh)) |
1583 | return PTR_ERR(bh); | |
1584 | eh = ext_block_hdr(bh); | |
1988b51e | 1585 | ix = EXT_FIRST_INDEX(eh); |
bf89d16f | 1586 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1587 | put_bh(bh); |
1588 | } | |
1589 | ||
107a7bd3 | 1590 | bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0); |
7d7ea89e TT |
1591 | if (IS_ERR(bh)) |
1592 | return PTR_ERR(bh); | |
1988b51e | 1593 | eh = ext_block_hdr(bh); |
1988b51e | 1594 | ex = EXT_FIRST_EXTENT(eh); |
4d33b1ef | 1595 | found_extent: |
1988b51e | 1596 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1597 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1598 | *ret_ex = ex; |
1599 | if (bh) | |
1600 | put_bh(bh); | |
1988b51e | 1601 | return 0; |
1988b51e AT |
1602 | } |
1603 | ||
a86c6181 | 1604 | /* |
d0d856e8 | 1605 | * ext4_ext_next_allocated_block: |
f17722f9 | 1606 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1607 | * NOTE: it considers block number from index entry as |
1608 | * allocated block. Thus, index entries have to be consistent | |
1609 | * with leaves. | |
a86c6181 | 1610 | */ |
fcf6b1b7 | 1611 | ext4_lblk_t |
a86c6181 AT |
1612 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1613 | { | |
1614 | int depth; | |
1615 | ||
1616 | BUG_ON(path == NULL); | |
1617 | depth = path->p_depth; | |
1618 | ||
1619 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1620 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1621 | |
1622 | while (depth >= 0) { | |
1623 | if (depth == path->p_depth) { | |
1624 | /* leaf */ | |
6f8ff537 CW |
1625 | if (path[depth].p_ext && |
1626 | path[depth].p_ext != | |
a86c6181 AT |
1627 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1628 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1629 | } else { | |
1630 | /* index */ | |
1631 | if (path[depth].p_idx != | |
1632 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1633 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1634 | } | |
1635 | depth--; | |
1636 | } | |
1637 | ||
f17722f9 | 1638 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1639 | } |
1640 | ||
1641 | /* | |
d0d856e8 | 1642 | * ext4_ext_next_leaf_block: |
f17722f9 | 1643 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1644 | */ |
5718789d | 1645 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1646 | { |
1647 | int depth; | |
1648 | ||
1649 | BUG_ON(path == NULL); | |
1650 | depth = path->p_depth; | |
1651 | ||
1652 | /* zero-tree has no leaf blocks at all */ | |
1653 | if (depth == 0) | |
f17722f9 | 1654 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1655 | |
1656 | /* go to index block */ | |
1657 | depth--; | |
1658 | ||
1659 | while (depth >= 0) { | |
1660 | if (path[depth].p_idx != | |
1661 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1662 | return (ext4_lblk_t) |
1663 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1664 | depth--; |
1665 | } | |
1666 | ||
f17722f9 | 1667 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1668 | } |
1669 | ||
1670 | /* | |
d0d856e8 RD |
1671 | * ext4_ext_correct_indexes: |
1672 | * if leaf gets modified and modified extent is first in the leaf, | |
1673 | * then we have to correct all indexes above. | |
a86c6181 AT |
1674 | * TODO: do we need to correct tree in all cases? |
1675 | */ | |
1d03ec98 | 1676 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1677 | struct ext4_ext_path *path) |
1678 | { | |
1679 | struct ext4_extent_header *eh; | |
1680 | int depth = ext_depth(inode); | |
1681 | struct ext4_extent *ex; | |
1682 | __le32 border; | |
1683 | int k, err = 0; | |
1684 | ||
1685 | eh = path[depth].p_hdr; | |
1686 | ex = path[depth].p_ext; | |
273df556 FM |
1687 | |
1688 | if (unlikely(ex == NULL || eh == NULL)) { | |
1689 | EXT4_ERROR_INODE(inode, | |
1690 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
6a797d27 | 1691 | return -EFSCORRUPTED; |
273df556 | 1692 | } |
a86c6181 AT |
1693 | |
1694 | if (depth == 0) { | |
1695 | /* there is no tree at all */ | |
1696 | return 0; | |
1697 | } | |
1698 | ||
1699 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1700 | /* we correct tree if first leaf got modified only */ | |
1701 | return 0; | |
1702 | } | |
1703 | ||
1704 | /* | |
d0d856e8 | 1705 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1706 | */ |
1707 | k = depth - 1; | |
1708 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1709 | err = ext4_ext_get_access(handle, inode, path + k); |
1710 | if (err) | |
a86c6181 AT |
1711 | return err; |
1712 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1713 | err = ext4_ext_dirty(handle, inode, path + k); |
1714 | if (err) | |
a86c6181 AT |
1715 | return err; |
1716 | ||
1717 | while (k--) { | |
1718 | /* change all left-side indexes */ | |
1719 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1720 | break; | |
7e028976 AM |
1721 | err = ext4_ext_get_access(handle, inode, path + k); |
1722 | if (err) | |
a86c6181 AT |
1723 | break; |
1724 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1725 | err = ext4_ext_dirty(handle, inode, path + k); |
1726 | if (err) | |
a86c6181 AT |
1727 | break; |
1728 | } | |
1729 | ||
1730 | return err; | |
1731 | } | |
1732 | ||
748de673 | 1733 | int |
a86c6181 AT |
1734 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1735 | struct ext4_extent *ex2) | |
1736 | { | |
da0169b3 | 1737 | unsigned short ext1_ee_len, ext2_ee_len; |
a2df2a63 | 1738 | |
556615dc | 1739 | if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2)) |
a2df2a63 AA |
1740 | return 0; |
1741 | ||
1742 | ext1_ee_len = ext4_ext_get_actual_len(ex1); | |
1743 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1744 | ||
1745 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1746 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1747 | return 0; |
1748 | ||
471d4011 SB |
1749 | /* |
1750 | * To allow future support for preallocated extents to be added | |
1751 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1752 | * this can result in the top bit of ee_len being set. |
471d4011 | 1753 | */ |
da0169b3 | 1754 | if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN) |
471d4011 | 1755 | return 0; |
109811c2 JK |
1756 | /* |
1757 | * The check for IO to unwritten extent is somewhat racy as we | |
1758 | * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after | |
1759 | * dropping i_data_sem. But reserved blocks should save us in that | |
1760 | * case. | |
1761 | */ | |
556615dc | 1762 | if (ext4_ext_is_unwritten(ex1) && |
a9b82415 DW |
1763 | (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) || |
1764 | atomic_read(&EXT4_I(inode)->i_unwritten) || | |
556615dc | 1765 | (ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN))) |
a9b82415 | 1766 | return 0; |
bbf2f9fb | 1767 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1768 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1769 | return 0; |
1770 | #endif | |
1771 | ||
bf89d16f | 1772 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1773 | return 1; |
1774 | return 0; | |
1775 | } | |
1776 | ||
56055d3a AA |
1777 | /* |
1778 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1779 | * It always tries to merge towards right. If you want to merge towards | |
1780 | * left, pass "ex - 1" as argument instead of "ex". | |
1781 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1782 | * 1 if they got merged. | |
1783 | */ | |
197217a5 | 1784 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1785 | struct ext4_ext_path *path, |
1786 | struct ext4_extent *ex) | |
56055d3a AA |
1787 | { |
1788 | struct ext4_extent_header *eh; | |
1789 | unsigned int depth, len; | |
556615dc | 1790 | int merge_done = 0, unwritten; |
56055d3a AA |
1791 | |
1792 | depth = ext_depth(inode); | |
1793 | BUG_ON(path[depth].p_hdr == NULL); | |
1794 | eh = path[depth].p_hdr; | |
1795 | ||
1796 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1797 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1798 | break; | |
1799 | /* merge with next extent! */ | |
556615dc | 1800 | unwritten = ext4_ext_is_unwritten(ex); |
56055d3a AA |
1801 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
1802 | + ext4_ext_get_actual_len(ex + 1)); | |
556615dc LC |
1803 | if (unwritten) |
1804 | ext4_ext_mark_unwritten(ex); | |
56055d3a AA |
1805 | |
1806 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1807 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1808 | * sizeof(struct ext4_extent); | |
1809 | memmove(ex + 1, ex + 2, len); | |
1810 | } | |
e8546d06 | 1811 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1812 | merge_done = 1; |
1813 | WARN_ON(eh->eh_entries == 0); | |
1814 | if (!eh->eh_entries) | |
24676da4 | 1815 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1816 | } |
1817 | ||
1818 | return merge_done; | |
1819 | } | |
1820 | ||
ecb94f5f TT |
1821 | /* |
1822 | * This function does a very simple check to see if we can collapse | |
1823 | * an extent tree with a single extent tree leaf block into the inode. | |
1824 | */ | |
1825 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1826 | struct inode *inode, | |
1827 | struct ext4_ext_path *path) | |
1828 | { | |
1829 | size_t s; | |
1830 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1831 | ext4_fsblk_t blk; | |
1832 | ||
1833 | if ((path[0].p_depth != 1) || | |
1834 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1835 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1836 | return; | |
1837 | ||
1838 | /* | |
1839 | * We need to modify the block allocation bitmap and the block | |
1840 | * group descriptor to release the extent tree block. If we | |
1841 | * can't get the journal credits, give up. | |
1842 | */ | |
1843 | if (ext4_journal_extend(handle, 2)) | |
1844 | return; | |
1845 | ||
1846 | /* | |
1847 | * Copy the extent data up to the inode | |
1848 | */ | |
1849 | blk = ext4_idx_pblock(path[0].p_idx); | |
1850 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1851 | sizeof(struct ext4_extent_idx); | |
1852 | s += sizeof(struct ext4_extent_header); | |
1853 | ||
10809df8 | 1854 | path[1].p_maxdepth = path[0].p_maxdepth; |
ecb94f5f TT |
1855 | memcpy(path[0].p_hdr, path[1].p_hdr, s); |
1856 | path[0].p_depth = 0; | |
1857 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1858 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1859 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1860 | ||
1861 | brelse(path[1].p_bh); | |
1862 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
71d4f7d0 | 1863 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
ecb94f5f TT |
1864 | } |
1865 | ||
197217a5 YY |
1866 | /* |
1867 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1868 | * return 1 if merge left else 0. | |
1869 | */ | |
ecb94f5f TT |
1870 | static void ext4_ext_try_to_merge(handle_t *handle, |
1871 | struct inode *inode, | |
197217a5 YY |
1872 | struct ext4_ext_path *path, |
1873 | struct ext4_extent *ex) { | |
1874 | struct ext4_extent_header *eh; | |
1875 | unsigned int depth; | |
1876 | int merge_done = 0; | |
197217a5 YY |
1877 | |
1878 | depth = ext_depth(inode); | |
1879 | BUG_ON(path[depth].p_hdr == NULL); | |
1880 | eh = path[depth].p_hdr; | |
1881 | ||
1882 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1883 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1884 | ||
1885 | if (!merge_done) | |
ecb94f5f | 1886 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1887 | |
ecb94f5f | 1888 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1889 | } |
1890 | ||
25d14f98 AA |
1891 | /* |
1892 | * check if a portion of the "newext" extent overlaps with an | |
1893 | * existing extent. | |
1894 | * | |
1895 | * If there is an overlap discovered, it updates the length of the newext | |
1896 | * such that there will be no overlap, and then returns 1. | |
1897 | * If there is no overlap found, it returns 0. | |
1898 | */ | |
4d33b1ef TT |
1899 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1900 | struct inode *inode, | |
1f109d5a TT |
1901 | struct ext4_extent *newext, |
1902 | struct ext4_ext_path *path) | |
25d14f98 | 1903 | { |
725d26d3 | 1904 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1905 | unsigned int depth, len1; |
1906 | unsigned int ret = 0; | |
1907 | ||
1908 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1909 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1910 | depth = ext_depth(inode); |
1911 | if (!path[depth].p_ext) | |
1912 | goto out; | |
f5a44db5 | 1913 | b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block)); |
25d14f98 AA |
1914 | |
1915 | /* | |
1916 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1917 | * is before the requested block(s) |
25d14f98 AA |
1918 | */ |
1919 | if (b2 < b1) { | |
1920 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1921 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1922 | goto out; |
f5a44db5 | 1923 | b2 = EXT4_LBLK_CMASK(sbi, b2); |
25d14f98 AA |
1924 | } |
1925 | ||
725d26d3 | 1926 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1927 | if (b1 + len1 < b1) { |
f17722f9 | 1928 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1929 | newext->ee_len = cpu_to_le16(len1); |
1930 | ret = 1; | |
1931 | } | |
1932 | ||
1933 | /* check for overlap */ | |
1934 | if (b1 + len1 > b2) { | |
1935 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1936 | ret = 1; | |
1937 | } | |
1938 | out: | |
1939 | return ret; | |
1940 | } | |
1941 | ||
a86c6181 | 1942 | /* |
d0d856e8 RD |
1943 | * ext4_ext_insert_extent: |
1944 | * tries to merge requsted extent into the existing extent or | |
1945 | * inserts requested extent as new one into the tree, | |
1946 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1947 | */ |
1948 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
dfe50809 | 1949 | struct ext4_ext_path **ppath, |
107a7bd3 | 1950 | struct ext4_extent *newext, int gb_flags) |
a86c6181 | 1951 | { |
dfe50809 | 1952 | struct ext4_ext_path *path = *ppath; |
af5bc92d | 1953 | struct ext4_extent_header *eh; |
a86c6181 AT |
1954 | struct ext4_extent *ex, *fex; |
1955 | struct ext4_extent *nearex; /* nearest extent */ | |
1956 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1957 | int depth, len, err; |
1958 | ext4_lblk_t next; | |
556615dc | 1959 | int mb_flags = 0, unwritten; |
a86c6181 | 1960 | |
e3cf5d5d TT |
1961 | if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
1962 | mb_flags |= EXT4_MB_DELALLOC_RESERVED; | |
273df556 FM |
1963 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1964 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
6a797d27 | 1965 | return -EFSCORRUPTED; |
273df556 | 1966 | } |
a86c6181 AT |
1967 | depth = ext_depth(inode); |
1968 | ex = path[depth].p_ext; | |
be8981be | 1969 | eh = path[depth].p_hdr; |
273df556 FM |
1970 | if (unlikely(path[depth].p_hdr == NULL)) { |
1971 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
6a797d27 | 1972 | return -EFSCORRUPTED; |
273df556 | 1973 | } |
a86c6181 AT |
1974 | |
1975 | /* try to insert block into found extent and return */ | |
107a7bd3 | 1976 | if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1977 | |
1978 | /* | |
be8981be LC |
1979 | * Try to see whether we should rather test the extent on |
1980 | * right from ex, or from the left of ex. This is because | |
ed8a1a76 | 1981 | * ext4_find_extent() can return either extent on the |
be8981be LC |
1982 | * left, or on the right from the searched position. This |
1983 | * will make merging more effective. | |
a2df2a63 | 1984 | */ |
be8981be LC |
1985 | if (ex < EXT_LAST_EXTENT(eh) && |
1986 | (le32_to_cpu(ex->ee_block) + | |
1987 | ext4_ext_get_actual_len(ex) < | |
1988 | le32_to_cpu(newext->ee_block))) { | |
1989 | ex += 1; | |
1990 | goto prepend; | |
1991 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1992 | (le32_to_cpu(newext->ee_block) + | |
1993 | ext4_ext_get_actual_len(newext) < | |
1994 | le32_to_cpu(ex->ee_block))) | |
1995 | ex -= 1; | |
1996 | ||
1997 | /* Try to append newex to the ex */ | |
1998 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1999 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
2000 | "(from %llu)\n", | |
556615dc | 2001 | ext4_ext_is_unwritten(newext), |
be8981be LC |
2002 | ext4_ext_get_actual_len(newext), |
2003 | le32_to_cpu(ex->ee_block), | |
556615dc | 2004 | ext4_ext_is_unwritten(ex), |
be8981be LC |
2005 | ext4_ext_get_actual_len(ex), |
2006 | ext4_ext_pblock(ex)); | |
2007 | err = ext4_ext_get_access(handle, inode, | |
2008 | path + depth); | |
2009 | if (err) | |
2010 | return err; | |
556615dc | 2011 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be | 2012 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) |
a2df2a63 | 2013 | + ext4_ext_get_actual_len(newext)); |
556615dc LC |
2014 | if (unwritten) |
2015 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
2016 | eh = path[depth].p_hdr; |
2017 | nearex = ex; | |
2018 | goto merge; | |
2019 | } | |
2020 | ||
2021 | prepend: | |
2022 | /* Try to prepend newex to the ex */ | |
2023 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
2024 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
2025 | "(from %llu)\n", | |
2026 | le32_to_cpu(newext->ee_block), | |
556615dc | 2027 | ext4_ext_is_unwritten(newext), |
be8981be LC |
2028 | ext4_ext_get_actual_len(newext), |
2029 | le32_to_cpu(ex->ee_block), | |
556615dc | 2030 | ext4_ext_is_unwritten(ex), |
be8981be LC |
2031 | ext4_ext_get_actual_len(ex), |
2032 | ext4_ext_pblock(ex)); | |
2033 | err = ext4_ext_get_access(handle, inode, | |
2034 | path + depth); | |
2035 | if (err) | |
2036 | return err; | |
2037 | ||
556615dc | 2038 | unwritten = ext4_ext_is_unwritten(ex); |
be8981be LC |
2039 | ex->ee_block = newext->ee_block; |
2040 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
2041 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
2042 | + ext4_ext_get_actual_len(newext)); | |
556615dc LC |
2043 | if (unwritten) |
2044 | ext4_ext_mark_unwritten(ex); | |
be8981be LC |
2045 | eh = path[depth].p_hdr; |
2046 | nearex = ex; | |
2047 | goto merge; | |
2048 | } | |
a86c6181 AT |
2049 | } |
2050 | ||
a86c6181 AT |
2051 | depth = ext_depth(inode); |
2052 | eh = path[depth].p_hdr; | |
2053 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
2054 | goto has_space; | |
2055 | ||
2056 | /* probably next leaf has space for us? */ | |
2057 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
2058 | next = EXT_MAX_BLOCKS; |
2059 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 2060 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 2061 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 2062 | ext_debug("next leaf block - %u\n", next); |
a86c6181 | 2063 | BUG_ON(npath != NULL); |
ed8a1a76 | 2064 | npath = ext4_find_extent(inode, next, NULL, 0); |
a86c6181 AT |
2065 | if (IS_ERR(npath)) |
2066 | return PTR_ERR(npath); | |
2067 | BUG_ON(npath->p_depth != path->p_depth); | |
2068 | eh = npath[depth].p_hdr; | |
2069 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 2070 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
2071 | le16_to_cpu(eh->eh_entries)); |
2072 | path = npath; | |
ffb505ff | 2073 | goto has_space; |
a86c6181 AT |
2074 | } |
2075 | ext_debug("next leaf has no free space(%d,%d)\n", | |
2076 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
2077 | } | |
2078 | ||
2079 | /* | |
d0d856e8 RD |
2080 | * There is no free space in the found leaf. |
2081 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 2082 | */ |
107a7bd3 | 2083 | if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
e3cf5d5d | 2084 | mb_flags |= EXT4_MB_USE_RESERVED; |
107a7bd3 | 2085 | err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, |
dfe50809 | 2086 | ppath, newext); |
a86c6181 AT |
2087 | if (err) |
2088 | goto cleanup; | |
2089 | depth = ext_depth(inode); | |
2090 | eh = path[depth].p_hdr; | |
2091 | ||
2092 | has_space: | |
2093 | nearex = path[depth].p_ext; | |
2094 | ||
7e028976 AM |
2095 | err = ext4_ext_get_access(handle, inode, path + depth); |
2096 | if (err) | |
a86c6181 AT |
2097 | goto cleanup; |
2098 | ||
2099 | if (!nearex) { | |
2100 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 2101 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 2102 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2103 | ext4_ext_pblock(newext), |
556615dc | 2104 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2105 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
2106 | nearex = EXT_FIRST_EXTENT(eh); |
2107 | } else { | |
2108 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 2109 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 2110 | /* Insert after */ |
32de6756 YY |
2111 | ext_debug("insert %u:%llu:[%d]%d before: " |
2112 | "nearest %p\n", | |
80e675f9 EG |
2113 | le32_to_cpu(newext->ee_block), |
2114 | ext4_ext_pblock(newext), | |
556615dc | 2115 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2116 | ext4_ext_get_actual_len(newext), |
2117 | nearex); | |
2118 | nearex++; | |
2119 | } else { | |
2120 | /* Insert before */ | |
2121 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
2122 | ext_debug("insert %u:%llu:[%d]%d after: " |
2123 | "nearest %p\n", | |
8c55e204 | 2124 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2125 | ext4_ext_pblock(newext), |
556615dc | 2126 | ext4_ext_is_unwritten(newext), |
a2df2a63 | 2127 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
2128 | nearex); |
2129 | } | |
2130 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
2131 | if (len > 0) { | |
32de6756 | 2132 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
2133 | "move %d extents from 0x%p to 0x%p\n", |
2134 | le32_to_cpu(newext->ee_block), | |
2135 | ext4_ext_pblock(newext), | |
556615dc | 2136 | ext4_ext_is_unwritten(newext), |
80e675f9 EG |
2137 | ext4_ext_get_actual_len(newext), |
2138 | len, nearex, nearex + 1); | |
2139 | memmove(nearex + 1, nearex, | |
2140 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2141 | } |
a86c6181 AT |
2142 | } |
2143 | ||
e8546d06 | 2144 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2145 | path[depth].p_ext = nearex; |
a86c6181 | 2146 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2147 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2148 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2149 | |
2150 | merge: | |
e7bcf823 | 2151 | /* try to merge extents */ |
107a7bd3 | 2152 | if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2153 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2154 | |
a86c6181 AT |
2155 | |
2156 | /* time to correct all indexes above */ | |
2157 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2158 | if (err) | |
2159 | goto cleanup; | |
2160 | ||
ecb94f5f | 2161 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2162 | |
2163 | cleanup: | |
b7ea89ad TT |
2164 | ext4_ext_drop_refs(npath); |
2165 | kfree(npath); | |
a86c6181 AT |
2166 | return err; |
2167 | } | |
2168 | ||
91dd8c11 LC |
2169 | static int ext4_fill_fiemap_extents(struct inode *inode, |
2170 | ext4_lblk_t block, ext4_lblk_t num, | |
2171 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
2172 | { |
2173 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 2174 | struct ext4_extent *ex; |
69eb33dc | 2175 | struct extent_status es; |
91dd8c11 | 2176 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 2177 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
2178 | int exists, depth = 0, err = 0; |
2179 | unsigned int flags = 0; | |
2180 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 2181 | |
f17722f9 | 2182 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
2183 | num = last - block; |
2184 | /* find extent for this block */ | |
fab3a549 | 2185 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 | 2186 | |
ed8a1a76 | 2187 | path = ext4_find_extent(inode, block, &path, 0); |
6873fa0d | 2188 | if (IS_ERR(path)) { |
91dd8c11 | 2189 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
2190 | err = PTR_ERR(path); |
2191 | path = NULL; | |
2192 | break; | |
2193 | } | |
2194 | ||
2195 | depth = ext_depth(inode); | |
273df556 | 2196 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2197 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 | 2198 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
6a797d27 | 2199 | err = -EFSCORRUPTED; |
273df556 FM |
2200 | break; |
2201 | } | |
6873fa0d ES |
2202 | ex = path[depth].p_ext; |
2203 | next = ext4_ext_next_allocated_block(path); | |
2204 | ||
91dd8c11 | 2205 | flags = 0; |
6873fa0d ES |
2206 | exists = 0; |
2207 | if (!ex) { | |
2208 | /* there is no extent yet, so try to allocate | |
2209 | * all requested space */ | |
2210 | start = block; | |
2211 | end = block + num; | |
2212 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2213 | /* need to allocate space before found extent */ | |
2214 | start = block; | |
2215 | end = le32_to_cpu(ex->ee_block); | |
2216 | if (block + num < end) | |
2217 | end = block + num; | |
2218 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2219 | + ext4_ext_get_actual_len(ex)) { | |
2220 | /* need to allocate space after found extent */ | |
2221 | start = block; | |
2222 | end = block + num; | |
2223 | if (end >= next) | |
2224 | end = next; | |
2225 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2226 | /* | |
2227 | * some part of requested space is covered | |
2228 | * by found extent | |
2229 | */ | |
2230 | start = block; | |
2231 | end = le32_to_cpu(ex->ee_block) | |
2232 | + ext4_ext_get_actual_len(ex); | |
2233 | if (block + num < end) | |
2234 | end = block + num; | |
2235 | exists = 1; | |
2236 | } else { | |
2237 | BUG(); | |
2238 | } | |
2239 | BUG_ON(end <= start); | |
2240 | ||
2241 | if (!exists) { | |
69eb33dc ZL |
2242 | es.es_lblk = start; |
2243 | es.es_len = end - start; | |
2244 | es.es_pblk = 0; | |
6873fa0d | 2245 | } else { |
69eb33dc ZL |
2246 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2247 | es.es_len = ext4_ext_get_actual_len(ex); | |
2248 | es.es_pblk = ext4_ext_pblock(ex); | |
556615dc | 2249 | if (ext4_ext_is_unwritten(ex)) |
91dd8c11 | 2250 | flags |= FIEMAP_EXTENT_UNWRITTEN; |
6873fa0d ES |
2251 | } |
2252 | ||
91dd8c11 | 2253 | /* |
69eb33dc ZL |
2254 | * Find delayed extent and update es accordingly. We call |
2255 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2256 | * last existing extent or not. |
2257 | */ | |
69eb33dc | 2258 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2259 | if (!exists && next_del) { |
2260 | exists = 1; | |
72dac95d JL |
2261 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2262 | FIEMAP_EXTENT_UNKNOWN); | |
91dd8c11 LC |
2263 | } |
2264 | up_read(&EXT4_I(inode)->i_data_sem); | |
2265 | ||
69eb33dc ZL |
2266 | if (unlikely(es.es_len == 0)) { |
2267 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
6a797d27 | 2268 | err = -EFSCORRUPTED; |
273df556 FM |
2269 | break; |
2270 | } | |
6873fa0d | 2271 | |
f7fec032 ZL |
2272 | /* |
2273 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2274 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2275 | * possible that an extent is with unwritten and delayed | |
2276 | * status due to when an extent is delayed allocated and | |
2277 | * is allocated by fallocate status tree will track both of | |
2278 | * them in a extent. | |
2279 | * | |
2280 | * So we could return a unwritten and delayed extent, and | |
2281 | * its block is equal to 'next'. | |
2282 | */ | |
2283 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2284 | flags |= FIEMAP_EXTENT_LAST; |
2285 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2286 | next != EXT_MAX_BLOCKS)) { | |
2287 | EXT4_ERROR_INODE(inode, | |
2288 | "next extent == %u, next " | |
2289 | "delalloc extent = %u", | |
2290 | next, next_del); | |
6a797d27 | 2291 | err = -EFSCORRUPTED; |
91dd8c11 LC |
2292 | break; |
2293 | } | |
6873fa0d ES |
2294 | } |
2295 | ||
91dd8c11 LC |
2296 | if (exists) { |
2297 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2298 | (__u64)es.es_lblk << blksize_bits, |
2299 | (__u64)es.es_pblk << blksize_bits, | |
2300 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2301 | flags); |
2302 | if (err < 0) | |
2303 | break; | |
2304 | if (err == 1) { | |
2305 | err = 0; | |
2306 | break; | |
2307 | } | |
6873fa0d ES |
2308 | } |
2309 | ||
69eb33dc | 2310 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2311 | } |
2312 | ||
b7ea89ad TT |
2313 | ext4_ext_drop_refs(path); |
2314 | kfree(path); | |
6873fa0d ES |
2315 | return err; |
2316 | } | |
2317 | ||
a86c6181 | 2318 | /* |
140a5250 JK |
2319 | * ext4_ext_determine_hole - determine hole around given block |
2320 | * @inode: inode we lookup in | |
2321 | * @path: path in extent tree to @lblk | |
2322 | * @lblk: pointer to logical block around which we want to determine hole | |
2323 | * | |
2324 | * Determine hole length (and start if easily possible) around given logical | |
2325 | * block. We don't try too hard to find the beginning of the hole but @path | |
2326 | * actually points to extent before @lblk, we provide it. | |
2327 | * | |
2328 | * The function returns the length of a hole starting at @lblk. We update @lblk | |
2329 | * to the beginning of the hole if we managed to find it. | |
a86c6181 | 2330 | */ |
140a5250 JK |
2331 | static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode, |
2332 | struct ext4_ext_path *path, | |
2333 | ext4_lblk_t *lblk) | |
a86c6181 AT |
2334 | { |
2335 | int depth = ext_depth(inode); | |
a86c6181 | 2336 | struct ext4_extent *ex; |
140a5250 | 2337 | ext4_lblk_t len; |
a86c6181 AT |
2338 | |
2339 | ex = path[depth].p_ext; | |
2340 | if (ex == NULL) { | |
2f8e0a7c | 2341 | /* there is no extent yet, so gap is [0;-] */ |
140a5250 | 2342 | *lblk = 0; |
2f8e0a7c | 2343 | len = EXT_MAX_BLOCKS; |
140a5250 JK |
2344 | } else if (*lblk < le32_to_cpu(ex->ee_block)) { |
2345 | len = le32_to_cpu(ex->ee_block) - *lblk; | |
2346 | } else if (*lblk >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2347 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2348 | ext4_lblk_t next; |
725d26d3 | 2349 | |
140a5250 | 2350 | *lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex); |
725d26d3 | 2351 | next = ext4_ext_next_allocated_block(path); |
140a5250 JK |
2352 | BUG_ON(next == *lblk); |
2353 | len = next - *lblk; | |
a86c6181 | 2354 | } else { |
a86c6181 AT |
2355 | BUG(); |
2356 | } | |
140a5250 JK |
2357 | return len; |
2358 | } | |
a86c6181 | 2359 | |
140a5250 JK |
2360 | /* |
2361 | * ext4_ext_put_gap_in_cache: | |
2362 | * calculate boundaries of the gap that the requested block fits into | |
2363 | * and cache this gap | |
2364 | */ | |
2365 | static void | |
2366 | ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start, | |
2367 | ext4_lblk_t hole_len) | |
2368 | { | |
2369 | struct extent_status es; | |
2370 | ||
ad431025 EW |
2371 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start, |
2372 | hole_start + hole_len - 1, &es); | |
2f8e0a7c ZL |
2373 | if (es.es_len) { |
2374 | /* There's delayed extent containing lblock? */ | |
140a5250 | 2375 | if (es.es_lblk <= hole_start) |
2f8e0a7c | 2376 | return; |
140a5250 | 2377 | hole_len = min(es.es_lblk - hole_start, hole_len); |
2f8e0a7c | 2378 | } |
140a5250 JK |
2379 | ext_debug(" -> %u:%u\n", hole_start, hole_len); |
2380 | ext4_es_insert_extent(inode, hole_start, hole_len, ~0, | |
2381 | EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2382 | } |
2383 | ||
2384 | /* | |
d0d856e8 RD |
2385 | * ext4_ext_rm_idx: |
2386 | * removes index from the index block. | |
a86c6181 | 2387 | */ |
1d03ec98 | 2388 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2389 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2390 | { |
a86c6181 | 2391 | int err; |
f65e6fba | 2392 | ext4_fsblk_t leaf; |
a86c6181 AT |
2393 | |
2394 | /* free index block */ | |
c36575e6 FL |
2395 | depth--; |
2396 | path = path + depth; | |
bf89d16f | 2397 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2398 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2399 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
6a797d27 | 2400 | return -EFSCORRUPTED; |
273df556 | 2401 | } |
7e028976 AM |
2402 | err = ext4_ext_get_access(handle, inode, path); |
2403 | if (err) | |
a86c6181 | 2404 | return err; |
0e1147b0 RD |
2405 | |
2406 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2407 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2408 | len *= sizeof(struct ext4_extent_idx); | |
2409 | memmove(path->p_idx, path->p_idx + 1, len); | |
2410 | } | |
2411 | ||
e8546d06 | 2412 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2413 | err = ext4_ext_dirty(handle, inode, path); |
2414 | if (err) | |
a86c6181 | 2415 | return err; |
2ae02107 | 2416 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2417 | trace_ext4_ext_rm_idx(inode, leaf); |
2418 | ||
7dc57615 | 2419 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2420 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2421 | |
2422 | while (--depth >= 0) { | |
2423 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2424 | break; | |
2425 | path--; | |
2426 | err = ext4_ext_get_access(handle, inode, path); | |
2427 | if (err) | |
2428 | break; | |
2429 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2430 | err = ext4_ext_dirty(handle, inode, path); | |
2431 | if (err) | |
2432 | break; | |
2433 | } | |
a86c6181 AT |
2434 | return err; |
2435 | } | |
2436 | ||
2437 | /* | |
ee12b630 MC |
2438 | * ext4_ext_calc_credits_for_single_extent: |
2439 | * This routine returns max. credits that needed to insert an extent | |
2440 | * to the extent tree. | |
2441 | * When pass the actual path, the caller should calculate credits | |
2442 | * under i_data_sem. | |
a86c6181 | 2443 | */ |
525f4ed8 | 2444 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2445 | struct ext4_ext_path *path) |
2446 | { | |
a86c6181 | 2447 | if (path) { |
ee12b630 | 2448 | int depth = ext_depth(inode); |
f3bd1f3f | 2449 | int ret = 0; |
ee12b630 | 2450 | |
a86c6181 | 2451 | /* probably there is space in leaf? */ |
a86c6181 | 2452 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2453 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2454 | |
ee12b630 MC |
2455 | /* |
2456 | * There are some space in the leaf tree, no | |
2457 | * need to account for leaf block credit | |
2458 | * | |
2459 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2460 | * and other metadata blocks still need to be |
ee12b630 MC |
2461 | * accounted. |
2462 | */ | |
525f4ed8 | 2463 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2464 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2465 | return ret; |
ee12b630 MC |
2466 | } |
2467 | } | |
a86c6181 | 2468 | |
525f4ed8 | 2469 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2470 | } |
a86c6181 | 2471 | |
ee12b630 | 2472 | /* |
fffb2739 | 2473 | * How many index/leaf blocks need to change/allocate to add @extents extents? |
ee12b630 | 2474 | * |
fffb2739 JK |
2475 | * If we add a single extent, then in the worse case, each tree level |
2476 | * index/leaf need to be changed in case of the tree split. | |
ee12b630 | 2477 | * |
fffb2739 JK |
2478 | * If more extents are inserted, they could cause the whole tree split more |
2479 | * than once, but this is really rare. | |
ee12b630 | 2480 | */ |
fffb2739 | 2481 | int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
ee12b630 MC |
2482 | { |
2483 | int index; | |
f19d5870 TM |
2484 | int depth; |
2485 | ||
2486 | /* If we are converting the inline data, only one is needed here. */ | |
2487 | if (ext4_has_inline_data(inode)) | |
2488 | return 1; | |
2489 | ||
2490 | depth = ext_depth(inode); | |
a86c6181 | 2491 | |
fffb2739 | 2492 | if (extents <= 1) |
ee12b630 MC |
2493 | index = depth * 2; |
2494 | else | |
2495 | index = depth * 3; | |
a86c6181 | 2496 | |
ee12b630 | 2497 | return index; |
a86c6181 AT |
2498 | } |
2499 | ||
981250ca TT |
2500 | static inline int get_default_free_blocks_flags(struct inode *inode) |
2501 | { | |
ddfa17e4 TE |
2502 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) || |
2503 | ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE)) | |
981250ca TT |
2504 | return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; |
2505 | else if (ext4_should_journal_data(inode)) | |
2506 | return EXT4_FREE_BLOCKS_FORGET; | |
2507 | return 0; | |
2508 | } | |
2509 | ||
9fe67149 EW |
2510 | /* |
2511 | * ext4_rereserve_cluster - increment the reserved cluster count when | |
2512 | * freeing a cluster with a pending reservation | |
2513 | * | |
2514 | * @inode - file containing the cluster | |
2515 | * @lblk - logical block in cluster to be reserved | |
2516 | * | |
2517 | * Increments the reserved cluster count and adjusts quota in a bigalloc | |
2518 | * file system when freeing a partial cluster containing at least one | |
2519 | * delayed and unwritten block. A partial cluster meeting that | |
2520 | * requirement will have a pending reservation. If so, the | |
2521 | * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to | |
2522 | * defer reserved and allocated space accounting to a subsequent call | |
2523 | * to this function. | |
2524 | */ | |
2525 | static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk) | |
2526 | { | |
2527 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
2528 | struct ext4_inode_info *ei = EXT4_I(inode); | |
2529 | ||
2530 | dquot_reclaim_block(inode, EXT4_C2B(sbi, 1)); | |
2531 | ||
2532 | spin_lock(&ei->i_block_reservation_lock); | |
2533 | ei->i_reserved_data_blocks++; | |
2534 | percpu_counter_add(&sbi->s_dirtyclusters_counter, 1); | |
2535 | spin_unlock(&ei->i_block_reservation_lock); | |
2536 | ||
2537 | percpu_counter_add(&sbi->s_freeclusters_counter, 1); | |
2538 | ext4_remove_pending(inode, lblk); | |
2539 | } | |
2540 | ||
a86c6181 | 2541 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
0aa06000 | 2542 | struct ext4_extent *ex, |
9fe67149 | 2543 | struct partial_cluster *partial, |
0aa06000 | 2544 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2545 | { |
0aa06000 | 2546 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
345ee947 | 2547 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
9fe67149 EW |
2548 | ext4_fsblk_t last_pblk, pblk; |
2549 | ext4_lblk_t num; | |
2550 | int flags; | |
2551 | ||
2552 | /* only extent tail removal is allowed */ | |
2553 | if (from < le32_to_cpu(ex->ee_block) || | |
2554 | to != le32_to_cpu(ex->ee_block) + ee_len - 1) { | |
2555 | ext4_error(sbi->s_sb, | |
2556 | "strange request: removal(2) %u-%u from %u:%u", | |
2557 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
2558 | return 0; | |
2559 | } | |
2560 | ||
2561 | #ifdef EXTENTS_STATS | |
2562 | spin_lock(&sbi->s_ext_stats_lock); | |
2563 | sbi->s_ext_blocks += ee_len; | |
2564 | sbi->s_ext_extents++; | |
2565 | if (ee_len < sbi->s_ext_min) | |
2566 | sbi->s_ext_min = ee_len; | |
2567 | if (ee_len > sbi->s_ext_max) | |
2568 | sbi->s_ext_max = ee_len; | |
2569 | if (ext_depth(inode) > sbi->s_depth_max) | |
2570 | sbi->s_depth_max = ext_depth(inode); | |
2571 | spin_unlock(&sbi->s_ext_stats_lock); | |
2572 | #endif | |
2573 | ||
2574 | trace_ext4_remove_blocks(inode, ex, from, to, partial); | |
18888cf0 | 2575 | |
0aa06000 | 2576 | /* |
9fe67149 EW |
2577 | * if we have a partial cluster, and it's different from the |
2578 | * cluster of the last block in the extent, we free it | |
0aa06000 | 2579 | */ |
9fe67149 EW |
2580 | last_pblk = ext4_ext_pblock(ex) + ee_len - 1; |
2581 | ||
2582 | if (partial->state != initial && | |
2583 | partial->pclu != EXT4_B2C(sbi, last_pblk)) { | |
2584 | if (partial->state == tofree) { | |
2585 | flags = get_default_free_blocks_flags(inode); | |
2586 | if (ext4_is_pending(inode, partial->lblk)) | |
2587 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
2588 | ext4_free_blocks(handle, inode, NULL, | |
2589 | EXT4_C2B(sbi, partial->pclu), | |
2590 | sbi->s_cluster_ratio, flags); | |
2591 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) | |
2592 | ext4_rereserve_cluster(inode, partial->lblk); | |
2593 | } | |
2594 | partial->state = initial; | |
2595 | } | |
2596 | ||
2597 | num = le32_to_cpu(ex->ee_block) + ee_len - from; | |
2598 | pblk = ext4_ext_pblock(ex) + ee_len - num; | |
0aa06000 TT |
2599 | |
2600 | /* | |
9fe67149 EW |
2601 | * We free the partial cluster at the end of the extent (if any), |
2602 | * unless the cluster is used by another extent (partial_cluster | |
2603 | * state is nofree). If a partial cluster exists here, it must be | |
2604 | * shared with the last block in the extent. | |
0aa06000 | 2605 | */ |
9fe67149 EW |
2606 | flags = get_default_free_blocks_flags(inode); |
2607 | ||
2608 | /* partial, left end cluster aligned, right end unaligned */ | |
2609 | if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) && | |
2610 | (EXT4_LBLK_CMASK(sbi, to) >= from) && | |
2611 | (partial->state != nofree)) { | |
2612 | if (ext4_is_pending(inode, to)) | |
2613 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
0aa06000 | 2614 | ext4_free_blocks(handle, inode, NULL, |
9fe67149 | 2615 | EXT4_PBLK_CMASK(sbi, last_pblk), |
0aa06000 | 2616 | sbi->s_cluster_ratio, flags); |
9fe67149 EW |
2617 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) |
2618 | ext4_rereserve_cluster(inode, to); | |
2619 | partial->state = initial; | |
2620 | flags = get_default_free_blocks_flags(inode); | |
0aa06000 TT |
2621 | } |
2622 | ||
9fe67149 | 2623 | flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; |
d23142c6 | 2624 | |
9fe67149 EW |
2625 | /* |
2626 | * For bigalloc file systems, we never free a partial cluster | |
2627 | * at the beginning of the extent. Instead, we check to see if we | |
2628 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2629 | * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space. | |
2630 | */ | |
2631 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2632 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); | |
2633 | ||
2634 | /* reset the partial cluster if we've freed past it */ | |
2635 | if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk)) | |
2636 | partial->state = initial; | |
2637 | ||
2638 | /* | |
2639 | * If we've freed the entire extent but the beginning is not left | |
2640 | * cluster aligned and is not marked as ineligible for freeing we | |
2641 | * record the partial cluster at the beginning of the extent. It | |
2642 | * wasn't freed by the preceding ext4_free_blocks() call, and we | |
2643 | * need to look farther to the left to determine if it's to be freed | |
2644 | * (not shared with another extent). Else, reset the partial | |
2645 | * cluster - we're either done freeing or the beginning of the | |
2646 | * extent is left cluster aligned. | |
2647 | */ | |
2648 | if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) { | |
2649 | if (partial->state == initial) { | |
2650 | partial->pclu = EXT4_B2C(sbi, pblk); | |
2651 | partial->lblk = from; | |
2652 | partial->state = tofree; | |
345ee947 | 2653 | } |
9fe67149 EW |
2654 | } else { |
2655 | partial->state = initial; | |
2656 | } | |
2657 | ||
a86c6181 AT |
2658 | return 0; |
2659 | } | |
2660 | ||
d583fb87 AH |
2661 | /* |
2662 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
5bf43760 EW |
2663 | * blocks appearing between "start" and "end". Both "start" |
2664 | * and "end" must appear in the same extent or EIO is returned. | |
d583fb87 AH |
2665 | * |
2666 | * @handle: The journal handle | |
2667 | * @inode: The files inode | |
2668 | * @path: The path to the leaf | |
d23142c6 | 2669 | * @partial_cluster: The cluster which we'll have to free if all extents |
5bf43760 EW |
2670 | * has been released from it. However, if this value is |
2671 | * negative, it's a cluster just to the right of the | |
2672 | * punched region and it must not be freed. | |
d583fb87 AH |
2673 | * @start: The first block to remove |
2674 | * @end: The last block to remove | |
2675 | */ | |
a86c6181 AT |
2676 | static int |
2677 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
d23142c6 | 2678 | struct ext4_ext_path *path, |
9fe67149 | 2679 | struct partial_cluster *partial, |
0aa06000 | 2680 | ext4_lblk_t start, ext4_lblk_t end) |
a86c6181 | 2681 | { |
0aa06000 | 2682 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2683 | int err = 0, correct_index = 0; |
2684 | int depth = ext_depth(inode), credits; | |
2685 | struct ext4_extent_header *eh; | |
750c9c47 | 2686 | ext4_lblk_t a, b; |
725d26d3 AK |
2687 | unsigned num; |
2688 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2689 | unsigned short ex_ee_len; |
556615dc | 2690 | unsigned unwritten = 0; |
a86c6181 | 2691 | struct ext4_extent *ex; |
d23142c6 | 2692 | ext4_fsblk_t pblk; |
a86c6181 | 2693 | |
c29c0ae7 | 2694 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2695 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2696 | if (!path[depth].p_hdr) |
2697 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2698 | eh = path[depth].p_hdr; | |
273df556 FM |
2699 | if (unlikely(path[depth].p_hdr == NULL)) { |
2700 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
6a797d27 | 2701 | return -EFSCORRUPTED; |
273df556 | 2702 | } |
a86c6181 | 2703 | /* find where to start removing */ |
6ae06ff5 AS |
2704 | ex = path[depth].p_ext; |
2705 | if (!ex) | |
2706 | ex = EXT_LAST_EXTENT(eh); | |
a86c6181 AT |
2707 | |
2708 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2709 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2710 | |
9fe67149 | 2711 | trace_ext4_ext_rm_leaf(inode, start, ex, partial); |
d8990240 | 2712 | |
a86c6181 AT |
2713 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2714 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 | 2715 | |
556615dc LC |
2716 | if (ext4_ext_is_unwritten(ex)) |
2717 | unwritten = 1; | |
a41f2071 | 2718 | else |
556615dc | 2719 | unwritten = 0; |
a41f2071 | 2720 | |
553f9008 | 2721 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
556615dc | 2722 | unwritten, ex_ee_len); |
a86c6181 AT |
2723 | path[depth].p_ext = ex; |
2724 | ||
2725 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2726 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2727 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2728 | |
2729 | ext_debug(" border %u:%u\n", a, b); | |
2730 | ||
d583fb87 | 2731 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2732 | if (end < ex_ee_block) { |
d23142c6 LC |
2733 | /* |
2734 | * We're going to skip this extent and move to another, | |
f4226d9e EW |
2735 | * so note that its first cluster is in use to avoid |
2736 | * freeing it when removing blocks. Eventually, the | |
2737 | * right edge of the truncated/punched region will | |
2738 | * be just to the left. | |
d23142c6 | 2739 | */ |
f4226d9e EW |
2740 | if (sbi->s_cluster_ratio > 1) { |
2741 | pblk = ext4_ext_pblock(ex); | |
9fe67149 EW |
2742 | partial->pclu = EXT4_B2C(sbi, pblk); |
2743 | partial->state = nofree; | |
f4226d9e | 2744 | } |
d583fb87 AH |
2745 | ex--; |
2746 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2747 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2748 | continue; | |
750c9c47 | 2749 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2750 | EXT4_ERROR_INODE(inode, |
2751 | "can not handle truncate %u:%u " | |
2752 | "on extent %u:%u", | |
2753 | start, end, ex_ee_block, | |
2754 | ex_ee_block + ex_ee_len - 1); | |
6a797d27 | 2755 | err = -EFSCORRUPTED; |
750c9c47 | 2756 | goto out; |
a86c6181 AT |
2757 | } else if (a != ex_ee_block) { |
2758 | /* remove tail of the extent */ | |
750c9c47 | 2759 | num = a - ex_ee_block; |
a86c6181 AT |
2760 | } else { |
2761 | /* remove whole extent: excellent! */ | |
a86c6181 | 2762 | num = 0; |
a86c6181 | 2763 | } |
34071da7 TT |
2764 | /* |
2765 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2766 | * descriptor) for each block group; assume two block | |
2767 | * groups plus ex_ee_len/blocks_per_block_group for | |
2768 | * the worst case | |
2769 | */ | |
2770 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2771 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2772 | correct_index = 1; | |
2773 | credits += (ext_depth(inode)) + 1; | |
2774 | } | |
5aca07eb | 2775 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2776 | |
487caeef | 2777 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2778 | if (err) |
a86c6181 | 2779 | goto out; |
a86c6181 AT |
2780 | |
2781 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2782 | if (err) | |
2783 | goto out; | |
2784 | ||
9fe67149 | 2785 | err = ext4_remove_blocks(handle, inode, ex, partial, a, b); |
a86c6181 AT |
2786 | if (err) |
2787 | goto out; | |
2788 | ||
750c9c47 | 2789 | if (num == 0) |
d0d856e8 | 2790 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2791 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2792 | |
a86c6181 | 2793 | ex->ee_len = cpu_to_le16(num); |
749269fa | 2794 | /* |
556615dc | 2795 | * Do not mark unwritten if all the blocks in the |
749269fa AA |
2796 | * extent have been removed. |
2797 | */ | |
556615dc LC |
2798 | if (unwritten && num) |
2799 | ext4_ext_mark_unwritten(ex); | |
d583fb87 AH |
2800 | /* |
2801 | * If the extent was completely released, | |
2802 | * we need to remove it from the leaf | |
2803 | */ | |
2804 | if (num == 0) { | |
f17722f9 | 2805 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2806 | /* |
2807 | * For hole punching, we need to scoot all the | |
2808 | * extents up when an extent is removed so that | |
2809 | * we dont have blank extents in the middle | |
2810 | */ | |
2811 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2812 | sizeof(struct ext4_extent)); | |
2813 | ||
2814 | /* Now get rid of the one at the end */ | |
2815 | memset(EXT_LAST_EXTENT(eh), 0, | |
2816 | sizeof(struct ext4_extent)); | |
2817 | } | |
2818 | le16_add_cpu(&eh->eh_entries, -1); | |
5bf43760 | 2819 | } |
d583fb87 | 2820 | |
750c9c47 DM |
2821 | err = ext4_ext_dirty(handle, inode, path + depth); |
2822 | if (err) | |
2823 | goto out; | |
2824 | ||
bf52c6f7 | 2825 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2826 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2827 | ex--; |
2828 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2829 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2830 | } |
2831 | ||
2832 | if (correct_index && eh->eh_entries) | |
2833 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2834 | ||
0aa06000 | 2835 | /* |
ad6599ab EW |
2836 | * If there's a partial cluster and at least one extent remains in |
2837 | * the leaf, free the partial cluster if it isn't shared with the | |
5bf43760 | 2838 | * current extent. If it is shared with the current extent |
9fe67149 | 2839 | * we reset the partial cluster because we've reached the start of the |
5bf43760 | 2840 | * truncated/punched region and we're done removing blocks. |
0aa06000 | 2841 | */ |
9fe67149 | 2842 | if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) { |
5bf43760 | 2843 | pblk = ext4_ext_pblock(ex) + ex_ee_len - 1; |
9fe67149 EW |
2844 | if (partial->pclu != EXT4_B2C(sbi, pblk)) { |
2845 | int flags = get_default_free_blocks_flags(inode); | |
2846 | ||
2847 | if (ext4_is_pending(inode, partial->lblk)) | |
2848 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
5bf43760 | 2849 | ext4_free_blocks(handle, inode, NULL, |
9fe67149 EW |
2850 | EXT4_C2B(sbi, partial->pclu), |
2851 | sbi->s_cluster_ratio, flags); | |
2852 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) | |
2853 | ext4_rereserve_cluster(inode, partial->lblk); | |
5bf43760 | 2854 | } |
9fe67149 | 2855 | partial->state = initial; |
0aa06000 TT |
2856 | } |
2857 | ||
a86c6181 AT |
2858 | /* if this leaf is free, then we should |
2859 | * remove it from index block above */ | |
2860 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2861 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2862 | |
2863 | out: | |
2864 | return err; | |
2865 | } | |
2866 | ||
2867 | /* | |
d0d856e8 RD |
2868 | * ext4_ext_more_to_rm: |
2869 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2870 | */ |
09b88252 | 2871 | static int |
a86c6181 AT |
2872 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2873 | { | |
2874 | BUG_ON(path->p_idx == NULL); | |
2875 | ||
2876 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2877 | return 0; | |
2878 | ||
2879 | /* | |
d0d856e8 | 2880 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2881 | * so we have to consider current index for truncation |
2882 | */ | |
2883 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2884 | return 0; | |
2885 | return 1; | |
2886 | } | |
2887 | ||
26a4c0c6 TT |
2888 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2889 | ext4_lblk_t end) | |
a86c6181 | 2890 | { |
f4226d9e | 2891 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 | 2892 | int depth = ext_depth(inode); |
968dee77 | 2893 | struct ext4_ext_path *path = NULL; |
9fe67149 | 2894 | struct partial_cluster partial; |
a86c6181 | 2895 | handle_t *handle; |
6f2080e6 | 2896 | int i = 0, err = 0; |
a86c6181 | 2897 | |
9fe67149 EW |
2898 | partial.pclu = 0; |
2899 | partial.lblk = 0; | |
2900 | partial.state = initial; | |
2901 | ||
5f95d21f | 2902 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2903 | |
2904 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2905 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2906 | if (IS_ERR(handle)) |
2907 | return PTR_ERR(handle); | |
2908 | ||
0617b83f | 2909 | again: |
61801325 | 2910 | trace_ext4_ext_remove_space(inode, start, end, depth); |
d8990240 | 2911 | |
5f95d21f LC |
2912 | /* |
2913 | * Check if we are removing extents inside the extent tree. If that | |
2914 | * is the case, we are going to punch a hole inside the extent tree | |
2915 | * so we have to check whether we need to split the extent covering | |
2916 | * the last block to remove so we can easily remove the part of it | |
2917 | * in ext4_ext_rm_leaf(). | |
2918 | */ | |
2919 | if (end < EXT_MAX_BLOCKS - 1) { | |
2920 | struct ext4_extent *ex; | |
f4226d9e EW |
2921 | ext4_lblk_t ee_block, ex_end, lblk; |
2922 | ext4_fsblk_t pblk; | |
5f95d21f | 2923 | |
f4226d9e | 2924 | /* find extent for or closest extent to this block */ |
ed8a1a76 | 2925 | path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE); |
5f95d21f LC |
2926 | if (IS_ERR(path)) { |
2927 | ext4_journal_stop(handle); | |
2928 | return PTR_ERR(path); | |
2929 | } | |
2930 | depth = ext_depth(inode); | |
6f2080e6 | 2931 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2932 | ex = path[depth].p_ext; |
968dee77 | 2933 | if (!ex) { |
6f2080e6 DM |
2934 | if (depth) { |
2935 | EXT4_ERROR_INODE(inode, | |
2936 | "path[%d].p_hdr == NULL", | |
2937 | depth); | |
6a797d27 | 2938 | err = -EFSCORRUPTED; |
6f2080e6 DM |
2939 | } |
2940 | goto out; | |
968dee77 | 2941 | } |
5f95d21f LC |
2942 | |
2943 | ee_block = le32_to_cpu(ex->ee_block); | |
f4226d9e | 2944 | ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1; |
5f95d21f LC |
2945 | |
2946 | /* | |
2947 | * See if the last block is inside the extent, if so split | |
2948 | * the extent at 'end' block so we can easily remove the | |
2949 | * tail of the first part of the split extent in | |
2950 | * ext4_ext_rm_leaf(). | |
2951 | */ | |
f4226d9e EW |
2952 | if (end >= ee_block && end < ex_end) { |
2953 | ||
2954 | /* | |
2955 | * If we're going to split the extent, note that | |
2956 | * the cluster containing the block after 'end' is | |
2957 | * in use to avoid freeing it when removing blocks. | |
2958 | */ | |
2959 | if (sbi->s_cluster_ratio > 1) { | |
2960 | pblk = ext4_ext_pblock(ex) + end - ee_block + 2; | |
9fe67149 EW |
2961 | partial.pclu = EXT4_B2C(sbi, pblk); |
2962 | partial.state = nofree; | |
f4226d9e EW |
2963 | } |
2964 | ||
5f95d21f LC |
2965 | /* |
2966 | * Split the extent in two so that 'end' is the last | |
27dd4385 LC |
2967 | * block in the first new extent. Also we should not |
2968 | * fail removing space due to ENOSPC so try to use | |
2969 | * reserved block if that happens. | |
5f95d21f | 2970 | */ |
dfe50809 | 2971 | err = ext4_force_split_extent_at(handle, inode, &path, |
fcf6b1b7 | 2972 | end + 1, 1); |
5f95d21f LC |
2973 | if (err < 0) |
2974 | goto out; | |
f4226d9e | 2975 | |
7bd75230 EW |
2976 | } else if (sbi->s_cluster_ratio > 1 && end >= ex_end && |
2977 | partial.state == initial) { | |
f4226d9e | 2978 | /* |
7bd75230 EW |
2979 | * If we're punching, there's an extent to the right. |
2980 | * If the partial cluster hasn't been set, set it to | |
2981 | * that extent's first cluster and its state to nofree | |
2982 | * so it won't be freed should it contain blocks to be | |
2983 | * removed. If it's already set (tofree/nofree), we're | |
2984 | * retrying and keep the original partial cluster info | |
2985 | * so a cluster marked tofree as a result of earlier | |
2986 | * extent removal is not lost. | |
f4226d9e EW |
2987 | */ |
2988 | lblk = ex_end + 1; | |
2989 | err = ext4_ext_search_right(inode, path, &lblk, &pblk, | |
2990 | &ex); | |
2991 | if (err) | |
2992 | goto out; | |
9fe67149 EW |
2993 | if (pblk) { |
2994 | partial.pclu = EXT4_B2C(sbi, pblk); | |
2995 | partial.state = nofree; | |
2996 | } | |
5f95d21f | 2997 | } |
5f95d21f | 2998 | } |
a86c6181 | 2999 | /* |
d0d856e8 RD |
3000 | * We start scanning from right side, freeing all the blocks |
3001 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 3002 | */ |
0617b83f | 3003 | depth = ext_depth(inode); |
968dee77 AS |
3004 | if (path) { |
3005 | int k = i = depth; | |
3006 | while (--k > 0) | |
3007 | path[k].p_block = | |
3008 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
3009 | } else { | |
6396bb22 | 3010 | path = kcalloc(depth + 1, sizeof(struct ext4_ext_path), |
968dee77 AS |
3011 | GFP_NOFS); |
3012 | if (path == NULL) { | |
3013 | ext4_journal_stop(handle); | |
3014 | return -ENOMEM; | |
3015 | } | |
10809df8 | 3016 | path[0].p_maxdepth = path[0].p_depth = depth; |
968dee77 | 3017 | path[0].p_hdr = ext_inode_hdr(inode); |
89a4e48f | 3018 | i = 0; |
5f95d21f | 3019 | |
c349179b | 3020 | if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
6a797d27 | 3021 | err = -EFSCORRUPTED; |
968dee77 AS |
3022 | goto out; |
3023 | } | |
a86c6181 | 3024 | } |
968dee77 | 3025 | err = 0; |
a86c6181 AT |
3026 | |
3027 | while (i >= 0 && err == 0) { | |
3028 | if (i == depth) { | |
3029 | /* this is leaf block */ | |
d583fb87 | 3030 | err = ext4_ext_rm_leaf(handle, inode, path, |
9fe67149 | 3031 | &partial, start, end); |
d0d856e8 | 3032 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
3033 | brelse(path[i].p_bh); |
3034 | path[i].p_bh = NULL; | |
3035 | i--; | |
3036 | continue; | |
3037 | } | |
3038 | ||
3039 | /* this is index block */ | |
3040 | if (!path[i].p_hdr) { | |
3041 | ext_debug("initialize header\n"); | |
3042 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
3043 | } |
3044 | ||
a86c6181 | 3045 | if (!path[i].p_idx) { |
d0d856e8 | 3046 | /* this level hasn't been touched yet */ |
a86c6181 AT |
3047 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
3048 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
3049 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
3050 | path[i].p_hdr, | |
3051 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
3052 | } else { | |
d0d856e8 | 3053 | /* we were already here, see at next index */ |
a86c6181 AT |
3054 | path[i].p_idx--; |
3055 | } | |
3056 | ||
3057 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
3058 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
3059 | path[i].p_idx); | |
3060 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 3061 | struct buffer_head *bh; |
a86c6181 | 3062 | /* go to the next level */ |
2ae02107 | 3063 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 3064 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 3065 | memset(path + i + 1, 0, sizeof(*path)); |
7d7ea89e | 3066 | bh = read_extent_tree_block(inode, |
107a7bd3 TT |
3067 | ext4_idx_pblock(path[i].p_idx), depth - i - 1, |
3068 | EXT4_EX_NOCACHE); | |
7d7ea89e | 3069 | if (IS_ERR(bh)) { |
a86c6181 | 3070 | /* should we reset i_size? */ |
7d7ea89e | 3071 | err = PTR_ERR(bh); |
a86c6181 AT |
3072 | break; |
3073 | } | |
76828c88 TT |
3074 | /* Yield here to deal with large extent trees. |
3075 | * Should be a no-op if we did IO above. */ | |
3076 | cond_resched(); | |
c29c0ae7 | 3077 | if (WARN_ON(i + 1 > depth)) { |
6a797d27 | 3078 | err = -EFSCORRUPTED; |
c29c0ae7 AT |
3079 | break; |
3080 | } | |
3081 | path[i + 1].p_bh = bh; | |
a86c6181 | 3082 | |
d0d856e8 RD |
3083 | /* save actual number of indexes since this |
3084 | * number is changed at the next iteration */ | |
a86c6181 AT |
3085 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
3086 | i++; | |
3087 | } else { | |
d0d856e8 | 3088 | /* we finished processing this index, go up */ |
a86c6181 | 3089 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 3090 | /* index is empty, remove it; |
a86c6181 AT |
3091 | * handle must be already prepared by the |
3092 | * truncatei_leaf() */ | |
c36575e6 | 3093 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 3094 | } |
d0d856e8 | 3095 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
3096 | brelse(path[i].p_bh); |
3097 | path[i].p_bh = NULL; | |
3098 | i--; | |
3099 | ext_debug("return to level %d\n", i); | |
3100 | } | |
3101 | } | |
3102 | ||
9fe67149 EW |
3103 | trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial, |
3104 | path->p_hdr->eh_entries); | |
d8990240 | 3105 | |
0756b908 | 3106 | /* |
9fe67149 EW |
3107 | * if there's a partial cluster and we have removed the first extent |
3108 | * in the file, then we also free the partial cluster, if any | |
0756b908 | 3109 | */ |
9fe67149 EW |
3110 | if (partial.state == tofree && err == 0) { |
3111 | int flags = get_default_free_blocks_flags(inode); | |
3112 | ||
3113 | if (ext4_is_pending(inode, partial.lblk)) | |
3114 | flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER; | |
7b415bf6 | 3115 | ext4_free_blocks(handle, inode, NULL, |
9fe67149 EW |
3116 | EXT4_C2B(sbi, partial.pclu), |
3117 | sbi->s_cluster_ratio, flags); | |
3118 | if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER) | |
3119 | ext4_rereserve_cluster(inode, partial.lblk); | |
3120 | partial.state = initial; | |
7b415bf6 AK |
3121 | } |
3122 | ||
a86c6181 AT |
3123 | /* TODO: flexible tree reduction should be here */ |
3124 | if (path->p_hdr->eh_entries == 0) { | |
3125 | /* | |
d0d856e8 RD |
3126 | * truncate to zero freed all the tree, |
3127 | * so we need to correct eh_depth | |
a86c6181 AT |
3128 | */ |
3129 | err = ext4_ext_get_access(handle, inode, path); | |
3130 | if (err == 0) { | |
3131 | ext_inode_hdr(inode)->eh_depth = 0; | |
3132 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 3133 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
3134 | err = ext4_ext_dirty(handle, inode, path); |
3135 | } | |
3136 | } | |
3137 | out: | |
b7ea89ad TT |
3138 | ext4_ext_drop_refs(path); |
3139 | kfree(path); | |
3140 | path = NULL; | |
dfe50809 TT |
3141 | if (err == -EAGAIN) |
3142 | goto again; | |
a86c6181 AT |
3143 | ext4_journal_stop(handle); |
3144 | ||
3145 | return err; | |
3146 | } | |
3147 | ||
3148 | /* | |
3149 | * called at mount time | |
3150 | */ | |
3151 | void ext4_ext_init(struct super_block *sb) | |
3152 | { | |
3153 | /* | |
3154 | * possible initialization would be here | |
3155 | */ | |
3156 | ||
e2b911c5 | 3157 | if (ext4_has_feature_extents(sb)) { |
90576c0b | 3158 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 3159 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 3160 | #ifdef AGGRESSIVE_TEST |
92b97816 | 3161 | ", aggressive tests" |
a86c6181 AT |
3162 | #endif |
3163 | #ifdef CHECK_BINSEARCH | |
92b97816 | 3164 | ", check binsearch" |
a86c6181 AT |
3165 | #endif |
3166 | #ifdef EXTENTS_STATS | |
92b97816 | 3167 | ", stats" |
a86c6181 | 3168 | #endif |
92b97816 | 3169 | "\n"); |
90576c0b | 3170 | #endif |
a86c6181 AT |
3171 | #ifdef EXTENTS_STATS |
3172 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
3173 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
3174 | EXT4_SB(sb)->s_ext_max = 0; | |
3175 | #endif | |
3176 | } | |
3177 | } | |
3178 | ||
3179 | /* | |
3180 | * called at umount time | |
3181 | */ | |
3182 | void ext4_ext_release(struct super_block *sb) | |
3183 | { | |
e2b911c5 | 3184 | if (!ext4_has_feature_extents(sb)) |
a86c6181 AT |
3185 | return; |
3186 | ||
3187 | #ifdef EXTENTS_STATS | |
3188 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
3189 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3190 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
3191 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
3192 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
3193 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
3194 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
3195 | } | |
3196 | #endif | |
3197 | } | |
3198 | ||
d7b2a00c ZL |
3199 | static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex) |
3200 | { | |
3201 | ext4_lblk_t ee_block; | |
3202 | ext4_fsblk_t ee_pblock; | |
3203 | unsigned int ee_len; | |
3204 | ||
3205 | ee_block = le32_to_cpu(ex->ee_block); | |
3206 | ee_len = ext4_ext_get_actual_len(ex); | |
3207 | ee_pblock = ext4_ext_pblock(ex); | |
3208 | ||
3209 | if (ee_len == 0) | |
3210 | return 0; | |
3211 | ||
3212 | return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock, | |
3213 | EXTENT_STATUS_WRITTEN); | |
3214 | } | |
3215 | ||
093a088b AK |
3216 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
3217 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
3218 | { | |
2407518d LC |
3219 | ext4_fsblk_t ee_pblock; |
3220 | unsigned int ee_len; | |
093a088b | 3221 | |
093a088b | 3222 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 3223 | ee_pblock = ext4_ext_pblock(ex); |
53085fac JK |
3224 | return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock, |
3225 | ee_len); | |
093a088b AK |
3226 | } |
3227 | ||
47ea3bb5 YY |
3228 | /* |
3229 | * ext4_split_extent_at() splits an extent at given block. | |
3230 | * | |
3231 | * @handle: the journal handle | |
3232 | * @inode: the file inode | |
3233 | * @path: the path to the extent | |
3234 | * @split: the logical block where the extent is splitted. | |
3235 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
556615dc | 3236 | * the states(init or unwritten) of new extents. |
47ea3bb5 YY |
3237 | * @flags: flags used to insert new extent to extent tree. |
3238 | * | |
3239 | * | |
3240 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
3241 | * of which are deterimined by split_flag. | |
3242 | * | |
3243 | * There are two cases: | |
3244 | * a> the extent are splitted into two extent. | |
3245 | * b> split is not needed, and just mark the extent. | |
3246 | * | |
3247 | * return 0 on success. | |
3248 | */ | |
3249 | static int ext4_split_extent_at(handle_t *handle, | |
3250 | struct inode *inode, | |
dfe50809 | 3251 | struct ext4_ext_path **ppath, |
47ea3bb5 YY |
3252 | ext4_lblk_t split, |
3253 | int split_flag, | |
3254 | int flags) | |
3255 | { | |
dfe50809 | 3256 | struct ext4_ext_path *path = *ppath; |
47ea3bb5 YY |
3257 | ext4_fsblk_t newblock; |
3258 | ext4_lblk_t ee_block; | |
adb23551 | 3259 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
3260 | struct ext4_extent *ex2 = NULL; |
3261 | unsigned int ee_len, depth; | |
3262 | int err = 0; | |
3263 | ||
dee1f973 DM |
3264 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
3265 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
3266 | ||
47ea3bb5 YY |
3267 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
3268 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3269 | ||
3270 | ext4_ext_show_leaf(inode, path); | |
3271 | ||
3272 | depth = ext_depth(inode); | |
3273 | ex = path[depth].p_ext; | |
3274 | ee_block = le32_to_cpu(ex->ee_block); | |
3275 | ee_len = ext4_ext_get_actual_len(ex); | |
3276 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3277 | ||
3278 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
556615dc | 3279 | BUG_ON(!ext4_ext_is_unwritten(ex) && |
357b66fd | 3280 | split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc LC |
3281 | EXT4_EXT_MARK_UNWRIT1 | |
3282 | EXT4_EXT_MARK_UNWRIT2)); | |
47ea3bb5 YY |
3283 | |
3284 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3285 | if (err) | |
3286 | goto out; | |
3287 | ||
3288 | if (split == ee_block) { | |
3289 | /* | |
3290 | * case b: block @split is the block that the extent begins with | |
3291 | * then we just change the state of the extent, and splitting | |
3292 | * is not needed. | |
3293 | */ | |
556615dc LC |
3294 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3295 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3296 | else |
3297 | ext4_ext_mark_initialized(ex); | |
3298 | ||
3299 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3300 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3301 | |
ecb94f5f | 3302 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3303 | goto out; |
3304 | } | |
3305 | ||
3306 | /* case a */ | |
3307 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3308 | ex->ee_len = cpu_to_le16(split - ee_block); | |
556615dc LC |
3309 | if (split_flag & EXT4_EXT_MARK_UNWRIT1) |
3310 | ext4_ext_mark_unwritten(ex); | |
47ea3bb5 YY |
3311 | |
3312 | /* | |
3313 | * path may lead to new leaf, not to original leaf any more | |
3314 | * after ext4_ext_insert_extent() returns, | |
3315 | */ | |
3316 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3317 | if (err) | |
3318 | goto fix_extent_len; | |
3319 | ||
3320 | ex2 = &newex; | |
3321 | ex2->ee_block = cpu_to_le32(split); | |
3322 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3323 | ext4_ext_store_pblock(ex2, newblock); | |
556615dc LC |
3324 | if (split_flag & EXT4_EXT_MARK_UNWRIT2) |
3325 | ext4_ext_mark_unwritten(ex2); | |
47ea3bb5 | 3326 | |
dfe50809 | 3327 | err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags); |
47ea3bb5 | 3328 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { |
dee1f973 | 3329 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3330 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3331 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3332 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3333 | zero_ex.ee_len = cpu_to_le16( |
3334 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3335 | ext4_ext_store_pblock(&zero_ex, |
3336 | ext4_ext_pblock(ex2)); | |
3337 | } else { | |
dee1f973 | 3338 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3339 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3340 | zero_ex.ee_len = cpu_to_le16( |
3341 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3342 | ext4_ext_store_pblock(&zero_ex, |
3343 | ext4_ext_pblock(ex)); | |
3344 | } | |
3345 | } else { | |
dee1f973 | 3346 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3347 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3348 | zero_ex.ee_len = cpu_to_le16( |
3349 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3350 | ext4_ext_store_pblock(&zero_ex, |
3351 | ext4_ext_pblock(&orig_ex)); | |
3352 | } | |
dee1f973 | 3353 | |
47ea3bb5 YY |
3354 | if (err) |
3355 | goto fix_extent_len; | |
3356 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3357 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3358 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3359 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3360 | if (err) |
3361 | goto fix_extent_len; | |
3362 | ||
3363 | /* update extent status tree */ | |
d7b2a00c | 3364 | err = ext4_zeroout_es(inode, &zero_ex); |
adb23551 | 3365 | |
47ea3bb5 YY |
3366 | goto out; |
3367 | } else if (err) | |
3368 | goto fix_extent_len; | |
3369 | ||
3370 | out: | |
3371 | ext4_ext_show_leaf(inode, path); | |
3372 | return err; | |
3373 | ||
3374 | fix_extent_len: | |
3375 | ex->ee_len = orig_ex.ee_len; | |
29faed16 | 3376 | ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3377 | return err; |
3378 | } | |
3379 | ||
3380 | /* | |
3381 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3382 | * by @map as split_flags indicates | |
3383 | * | |
70261f56 | 3384 | * It may result in splitting the extent into multiple extents (up to three) |
47ea3bb5 YY |
3385 | * There are three possibilities: |
3386 | * a> There is no split required | |
3387 | * b> Splits in two extents: Split is happening at either end of the extent | |
3388 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3389 | * | |
3390 | */ | |
3391 | static int ext4_split_extent(handle_t *handle, | |
3392 | struct inode *inode, | |
dfe50809 | 3393 | struct ext4_ext_path **ppath, |
47ea3bb5 YY |
3394 | struct ext4_map_blocks *map, |
3395 | int split_flag, | |
3396 | int flags) | |
3397 | { | |
dfe50809 | 3398 | struct ext4_ext_path *path = *ppath; |
47ea3bb5 YY |
3399 | ext4_lblk_t ee_block; |
3400 | struct ext4_extent *ex; | |
3401 | unsigned int ee_len, depth; | |
3402 | int err = 0; | |
556615dc | 3403 | int unwritten; |
47ea3bb5 | 3404 | int split_flag1, flags1; |
3a225670 | 3405 | int allocated = map->m_len; |
47ea3bb5 YY |
3406 | |
3407 | depth = ext_depth(inode); | |
3408 | ex = path[depth].p_ext; | |
3409 | ee_block = le32_to_cpu(ex->ee_block); | |
3410 | ee_len = ext4_ext_get_actual_len(ex); | |
556615dc | 3411 | unwritten = ext4_ext_is_unwritten(ex); |
47ea3bb5 YY |
3412 | |
3413 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3414 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 | 3415 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
556615dc LC |
3416 | if (unwritten) |
3417 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1 | | |
3418 | EXT4_EXT_MARK_UNWRIT2; | |
dee1f973 DM |
3419 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3420 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
dfe50809 | 3421 | err = ext4_split_extent_at(handle, inode, ppath, |
47ea3bb5 | 3422 | map->m_lblk + map->m_len, split_flag1, flags1); |
93917411 YY |
3423 | if (err) |
3424 | goto out; | |
3a225670 ZL |
3425 | } else { |
3426 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3427 | } |
357b66fd DM |
3428 | /* |
3429 | * Update path is required because previous ext4_split_extent_at() may | |
3430 | * result in split of original leaf or extent zeroout. | |
3431 | */ | |
ed8a1a76 | 3432 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
47ea3bb5 YY |
3433 | if (IS_ERR(path)) |
3434 | return PTR_ERR(path); | |
357b66fd DM |
3435 | depth = ext_depth(inode); |
3436 | ex = path[depth].p_ext; | |
a18ed359 DM |
3437 | if (!ex) { |
3438 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3439 | (unsigned long) map->m_lblk); | |
6a797d27 | 3440 | return -EFSCORRUPTED; |
a18ed359 | 3441 | } |
556615dc | 3442 | unwritten = ext4_ext_is_unwritten(ex); |
357b66fd | 3443 | split_flag1 = 0; |
47ea3bb5 YY |
3444 | |
3445 | if (map->m_lblk >= ee_block) { | |
357b66fd | 3446 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
556615dc LC |
3447 | if (unwritten) { |
3448 | split_flag1 |= EXT4_EXT_MARK_UNWRIT1; | |
357b66fd | 3449 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
556615dc | 3450 | EXT4_EXT_MARK_UNWRIT2); |
357b66fd | 3451 | } |
dfe50809 | 3452 | err = ext4_split_extent_at(handle, inode, ppath, |
47ea3bb5 YY |
3453 | map->m_lblk, split_flag1, flags); |
3454 | if (err) | |
3455 | goto out; | |
3456 | } | |
3457 | ||
3458 | ext4_ext_show_leaf(inode, path); | |
3459 | out: | |
3a225670 | 3460 | return err ? err : allocated; |
47ea3bb5 YY |
3461 | } |
3462 | ||
56055d3a | 3463 | /* |
e35fd660 | 3464 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
556615dc | 3465 | * to an unwritten extent. It may result in splitting the unwritten |
25985edc | 3466 | * extent into multiple extents (up to three - one initialized and two |
556615dc | 3467 | * unwritten). |
56055d3a AA |
3468 | * There are three possibilities: |
3469 | * a> There is no split required: Entire extent should be initialized | |
3470 | * b> Splits in two extents: Write is happening at either end of the extent | |
3471 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3472 | * |
3473 | * Pre-conditions: | |
556615dc | 3474 | * - The extent pointed to by 'path' is unwritten. |
6f91bc5f EG |
3475 | * - The extent pointed to by 'path' contains a superset |
3476 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3477 | * | |
3478 | * Post-conditions on success: | |
3479 | * - the returned value is the number of blocks beyond map->l_lblk | |
3480 | * that are allocated and initialized. | |
3481 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3482 | */ |
725d26d3 | 3483 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3484 | struct inode *inode, |
3485 | struct ext4_map_blocks *map, | |
dfe50809 | 3486 | struct ext4_ext_path **ppath, |
27dd4385 | 3487 | int flags) |
56055d3a | 3488 | { |
dfe50809 | 3489 | struct ext4_ext_path *path = *ppath; |
67a5da56 | 3490 | struct ext4_sb_info *sbi; |
6f91bc5f | 3491 | struct ext4_extent_header *eh; |
667eff35 | 3492 | struct ext4_map_blocks split_map; |
4f8caa60 | 3493 | struct ext4_extent zero_ex1, zero_ex2; |
bc2d9db4 | 3494 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3495 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3496 | unsigned int ee_len, depth, map_len = map->m_len; |
3497 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3498 | int err = 0; |
4f8caa60 | 3499 | int split_flag = EXT4_EXT_DATA_VALID2; |
21ca087a DM |
3500 | |
3501 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3502 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3503 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3504 | |
67a5da56 | 3505 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3506 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3507 | inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3508 | if (eof_block < map->m_lblk + map_len) |
3509 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3510 | |
3511 | depth = ext_depth(inode); | |
6f91bc5f | 3512 | eh = path[depth].p_hdr; |
56055d3a AA |
3513 | ex = path[depth].p_ext; |
3514 | ee_block = le32_to_cpu(ex->ee_block); | |
3515 | ee_len = ext4_ext_get_actual_len(ex); | |
4f8caa60 JK |
3516 | zero_ex1.ee_len = 0; |
3517 | zero_ex2.ee_len = 0; | |
56055d3a | 3518 | |
6f91bc5f EG |
3519 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3520 | ||
3521 | /* Pre-conditions */ | |
556615dc | 3522 | BUG_ON(!ext4_ext_is_unwritten(ex)); |
6f91bc5f | 3523 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); |
6f91bc5f EG |
3524 | |
3525 | /* | |
3526 | * Attempt to transfer newly initialized blocks from the currently | |
556615dc | 3527 | * unwritten extent to its neighbor. This is much cheaper |
6f91bc5f | 3528 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3529 | * memmove() calls. Transferring to the left is the common case in |
3530 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3531 | * followed by append writes. | |
6f91bc5f EG |
3532 | * |
3533 | * Limitations of the current logic: | |
bc2d9db4 | 3534 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3535 | * This would require removing the extent if the transfer |
3536 | * is possible. | |
bc2d9db4 | 3537 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3538 | * same extent tree node. |
3539 | */ | |
bc2d9db4 LC |
3540 | if ((map->m_lblk == ee_block) && |
3541 | /* See if we can merge left */ | |
3542 | (map_len < ee_len) && /*L1*/ | |
3543 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3544 | ext4_lblk_t prev_lblk; |
3545 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3546 | unsigned int prev_len; |
6f91bc5f | 3547 | |
bc2d9db4 LC |
3548 | abut_ex = ex - 1; |
3549 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3550 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3551 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3552 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3553 | |
3554 | /* | |
bc2d9db4 | 3555 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3556 | * upon those conditions: |
bc2d9db4 LC |
3557 | * - C1: abut_ex is initialized, |
3558 | * - C2: abut_ex is logically abutting ex, | |
3559 | * - C3: abut_ex is physically abutting ex, | |
3560 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3561 | * overflowing the (initialized) length limit. |
3562 | */ | |
556615dc | 3563 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3564 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3565 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3566 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3567 | err = ext4_ext_get_access(handle, inode, path + depth); |
3568 | if (err) | |
3569 | goto out; | |
3570 | ||
3571 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3572 | map, ex, abut_ex); |
6f91bc5f | 3573 | |
bc2d9db4 LC |
3574 | /* Shift the start of ex by 'map_len' blocks */ |
3575 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3576 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3577 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3578 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
6f91bc5f | 3579 | |
bc2d9db4 LC |
3580 | /* Extend abut_ex by 'map_len' blocks */ |
3581 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3582 | |
bc2d9db4 LC |
3583 | /* Result: number of initialized blocks past m_lblk */ |
3584 | allocated = map_len; | |
3585 | } | |
3586 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3587 | (map_len < ee_len) && /*L1*/ | |
3588 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3589 | /* See if we can merge right */ | |
3590 | ext4_lblk_t next_lblk; | |
3591 | ext4_fsblk_t next_pblk, ee_pblk; | |
3592 | unsigned int next_len; | |
3593 | ||
3594 | abut_ex = ex + 1; | |
3595 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3596 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3597 | next_pblk = ext4_ext_pblock(abut_ex); | |
3598 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3599 | |
bc2d9db4 LC |
3600 | /* |
3601 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3602 | * upon those conditions: | |
3603 | * - C1: abut_ex is initialized, | |
3604 | * - C2: abut_ex is logically abutting ex, | |
3605 | * - C3: abut_ex is physically abutting ex, | |
3606 | * - C4: abut_ex can receive the additional blocks without | |
3607 | * overflowing the (initialized) length limit. | |
3608 | */ | |
556615dc | 3609 | if ((!ext4_ext_is_unwritten(abut_ex)) && /*C1*/ |
bc2d9db4 LC |
3610 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ |
3611 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3612 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3613 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3614 | if (err) | |
3615 | goto out; | |
3616 | ||
3617 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3618 | map, ex, abut_ex); | |
3619 | ||
3620 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3621 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3622 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3623 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
556615dc | 3624 | ext4_ext_mark_unwritten(ex); /* Restore the flag */ |
bc2d9db4 LC |
3625 | |
3626 | /* Extend abut_ex by 'map_len' blocks */ | |
3627 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3628 | |
3629 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3630 | allocated = map_len; |
6f91bc5f EG |
3631 | } |
3632 | } | |
bc2d9db4 LC |
3633 | if (allocated) { |
3634 | /* Mark the block containing both extents as dirty */ | |
3635 | ext4_ext_dirty(handle, inode, path + depth); | |
3636 | ||
3637 | /* Update path to point to the right extent */ | |
3638 | path[depth].p_ext = abut_ex; | |
3639 | goto out; | |
3640 | } else | |
3641 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3642 | |
667eff35 | 3643 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3644 | /* |
3645 | * It is safe to convert extent to initialized via explicit | |
9e740568 | 3646 | * zeroout only if extent is fully inside i_size or new_size. |
21ca087a | 3647 | */ |
667eff35 | 3648 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3649 | |
67a5da56 ZL |
3650 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3651 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3652 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 | 3653 | |
592ddec7 | 3654 | if (IS_ENCRYPTED(inode)) |
36086d43 TT |
3655 | max_zeroout = 0; |
3656 | ||
56055d3a | 3657 | /* |
4f8caa60 | 3658 | * five cases: |
667eff35 | 3659 | * 1. split the extent into three extents. |
4f8caa60 JK |
3660 | * 2. split the extent into two extents, zeroout the head of the first |
3661 | * extent. | |
3662 | * 3. split the extent into two extents, zeroout the tail of the second | |
3663 | * extent. | |
667eff35 | 3664 | * 4. split the extent into two extents with out zeroout. |
4f8caa60 JK |
3665 | * 5. no splitting needed, just possibly zeroout the head and / or the |
3666 | * tail of the extent. | |
56055d3a | 3667 | */ |
667eff35 YY |
3668 | split_map.m_lblk = map->m_lblk; |
3669 | split_map.m_len = map->m_len; | |
3670 | ||
4f8caa60 | 3671 | if (max_zeroout && (allocated > split_map.m_len)) { |
67a5da56 | 3672 | if (allocated <= max_zeroout) { |
4f8caa60 JK |
3673 | /* case 3 or 5 */ |
3674 | zero_ex1.ee_block = | |
3675 | cpu_to_le32(split_map.m_lblk + | |
3676 | split_map.m_len); | |
3677 | zero_ex1.ee_len = | |
3678 | cpu_to_le16(allocated - split_map.m_len); | |
3679 | ext4_ext_store_pblock(&zero_ex1, | |
3680 | ext4_ext_pblock(ex) + split_map.m_lblk + | |
3681 | split_map.m_len - ee_block); | |
3682 | err = ext4_ext_zeroout(inode, &zero_ex1); | |
56055d3a AA |
3683 | if (err) |
3684 | goto out; | |
667eff35 | 3685 | split_map.m_len = allocated; |
4f8caa60 JK |
3686 | } |
3687 | if (split_map.m_lblk - ee_block + split_map.m_len < | |
3688 | max_zeroout) { | |
3689 | /* case 2 or 5 */ | |
3690 | if (split_map.m_lblk != ee_block) { | |
3691 | zero_ex2.ee_block = ex->ee_block; | |
3692 | zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk - | |
667eff35 | 3693 | ee_block); |
4f8caa60 | 3694 | ext4_ext_store_pblock(&zero_ex2, |
667eff35 | 3695 | ext4_ext_pblock(ex)); |
4f8caa60 | 3696 | err = ext4_ext_zeroout(inode, &zero_ex2); |
667eff35 YY |
3697 | if (err) |
3698 | goto out; | |
3699 | } | |
3700 | ||
4f8caa60 | 3701 | split_map.m_len += split_map.m_lblk - ee_block; |
667eff35 | 3702 | split_map.m_lblk = ee_block; |
9b940f8e | 3703 | allocated = map->m_len; |
56055d3a AA |
3704 | } |
3705 | } | |
667eff35 | 3706 | |
ae9e9c6a JK |
3707 | err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag, |
3708 | flags); | |
3709 | if (err > 0) | |
3710 | err = 0; | |
56055d3a | 3711 | out: |
adb23551 | 3712 | /* If we have gotten a failure, don't zero out status tree */ |
4f8caa60 JK |
3713 | if (!err) { |
3714 | err = ext4_zeroout_es(inode, &zero_ex1); | |
3715 | if (!err) | |
3716 | err = ext4_zeroout_es(inode, &zero_ex2); | |
3717 | } | |
56055d3a AA |
3718 | return err ? err : allocated; |
3719 | } | |
3720 | ||
0031462b | 3721 | /* |
e35fd660 | 3722 | * This function is called by ext4_ext_map_blocks() from |
0031462b | 3723 | * ext4_get_blocks_dio_write() when DIO to write |
556615dc | 3724 | * to an unwritten extent. |
0031462b | 3725 | * |
556615dc LC |
3726 | * Writing to an unwritten extent may result in splitting the unwritten |
3727 | * extent into multiple initialized/unwritten extents (up to three) | |
0031462b | 3728 | * There are three possibilities: |
556615dc | 3729 | * a> There is no split required: Entire extent should be unwritten |
0031462b MC |
3730 | * b> Splits in two extents: Write is happening at either end of the extent |
3731 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3732 | * | |
b8a86845 LC |
3733 | * This works the same way in the case of initialized -> unwritten conversion. |
3734 | * | |
0031462b | 3735 | * One of more index blocks maybe needed if the extent tree grow after |
556615dc LC |
3736 | * the unwritten extent split. To prevent ENOSPC occur at the IO |
3737 | * complete, we need to split the unwritten extent before DIO submit | |
3738 | * the IO. The unwritten extent called at this time will be split | |
3739 | * into three unwritten extent(at most). After IO complete, the part | |
0031462b MC |
3740 | * being filled will be convert to initialized by the end_io callback function |
3741 | * via ext4_convert_unwritten_extents(). | |
ba230c3f | 3742 | * |
556615dc | 3743 | * Returns the size of unwritten extent to be written on success. |
0031462b | 3744 | */ |
b8a86845 | 3745 | static int ext4_split_convert_extents(handle_t *handle, |
0031462b | 3746 | struct inode *inode, |
e35fd660 | 3747 | struct ext4_map_blocks *map, |
dfe50809 | 3748 | struct ext4_ext_path **ppath, |
0031462b MC |
3749 | int flags) |
3750 | { | |
dfe50809 | 3751 | struct ext4_ext_path *path = *ppath; |
667eff35 YY |
3752 | ext4_lblk_t eof_block; |
3753 | ext4_lblk_t ee_block; | |
3754 | struct ext4_extent *ex; | |
3755 | unsigned int ee_len; | |
3756 | int split_flag = 0, depth; | |
21ca087a | 3757 | |
b8a86845 LC |
3758 | ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n", |
3759 | __func__, inode->i_ino, | |
3760 | (unsigned long long)map->m_lblk, map->m_len); | |
21ca087a DM |
3761 | |
3762 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3763 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3764 | if (eof_block < map->m_lblk + map->m_len) |
3765 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3766 | /* |
3767 | * It is safe to convert extent to initialized via explicit | |
3768 | * zeroout only if extent is fully insde i_size or new_size. | |
3769 | */ | |
667eff35 YY |
3770 | depth = ext_depth(inode); |
3771 | ex = path[depth].p_ext; | |
3772 | ee_block = le32_to_cpu(ex->ee_block); | |
3773 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3774 | |
b8a86845 LC |
3775 | /* Convert to unwritten */ |
3776 | if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) { | |
3777 | split_flag |= EXT4_EXT_DATA_VALID1; | |
3778 | /* Convert to initialized */ | |
3779 | } else if (flags & EXT4_GET_BLOCKS_CONVERT) { | |
3780 | split_flag |= ee_block + ee_len <= eof_block ? | |
3781 | EXT4_EXT_MAY_ZEROOUT : 0; | |
556615dc | 3782 | split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2); |
b8a86845 | 3783 | } |
667eff35 | 3784 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
dfe50809 | 3785 | return ext4_split_extent(handle, inode, ppath, map, split_flag, flags); |
0031462b | 3786 | } |
197217a5 | 3787 | |
c7064ef1 | 3788 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3789 | struct inode *inode, |
3790 | struct ext4_map_blocks *map, | |
dfe50809 | 3791 | struct ext4_ext_path **ppath) |
0031462b | 3792 | { |
dfe50809 | 3793 | struct ext4_ext_path *path = *ppath; |
0031462b | 3794 | struct ext4_extent *ex; |
dee1f973 DM |
3795 | ext4_lblk_t ee_block; |
3796 | unsigned int ee_len; | |
0031462b MC |
3797 | int depth; |
3798 | int err = 0; | |
0031462b MC |
3799 | |
3800 | depth = ext_depth(inode); | |
0031462b | 3801 | ex = path[depth].p_ext; |
dee1f973 DM |
3802 | ee_block = le32_to_cpu(ex->ee_block); |
3803 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3804 | |
197217a5 YY |
3805 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3806 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3807 | (unsigned long long)ee_block, ee_len); |
3808 | ||
ff95ec22 DM |
3809 | /* If extent is larger than requested it is a clear sign that we still |
3810 | * have some extent state machine issues left. So extent_split is still | |
3811 | * required. | |
3812 | * TODO: Once all related issues will be fixed this situation should be | |
3813 | * illegal. | |
3814 | */ | |
dee1f973 | 3815 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ac62b54c RP |
3816 | #ifdef CONFIG_EXT4_DEBUG |
3817 | ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu," | |
8d2ae1cb | 3818 | " len %u; IO logical block %llu, len %u", |
ff95ec22 DM |
3819 | inode->i_ino, (unsigned long long)ee_block, ee_len, |
3820 | (unsigned long long)map->m_lblk, map->m_len); | |
3821 | #endif | |
dfe50809 | 3822 | err = ext4_split_convert_extents(handle, inode, map, ppath, |
b8a86845 | 3823 | EXT4_GET_BLOCKS_CONVERT); |
dee1f973 | 3824 | if (err < 0) |
dfe50809 | 3825 | return err; |
ed8a1a76 | 3826 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
dfe50809 TT |
3827 | if (IS_ERR(path)) |
3828 | return PTR_ERR(path); | |
dee1f973 DM |
3829 | depth = ext_depth(inode); |
3830 | ex = path[depth].p_ext; | |
3831 | } | |
197217a5 | 3832 | |
0031462b MC |
3833 | err = ext4_ext_get_access(handle, inode, path + depth); |
3834 | if (err) | |
3835 | goto out; | |
3836 | /* first mark the extent as initialized */ | |
3837 | ext4_ext_mark_initialized(ex); | |
3838 | ||
197217a5 YY |
3839 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3840 | * borders are not changed | |
0031462b | 3841 | */ |
ecb94f5f | 3842 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3843 | |
0031462b | 3844 | /* Mark modified extent as dirty */ |
ecb94f5f | 3845 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3846 | out: |
3847 | ext4_ext_show_leaf(inode, path); | |
3848 | return err; | |
3849 | } | |
3850 | ||
58590b06 TT |
3851 | /* |
3852 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3853 | */ | |
3854 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3855 | ext4_lblk_t lblk, |
58590b06 TT |
3856 | struct ext4_ext_path *path, |
3857 | unsigned int len) | |
3858 | { | |
3859 | int i, depth; | |
3860 | struct ext4_extent_header *eh; | |
65922cb5 | 3861 | struct ext4_extent *last_ex; |
58590b06 TT |
3862 | |
3863 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3864 | return 0; | |
3865 | ||
3866 | depth = ext_depth(inode); | |
3867 | eh = path[depth].p_hdr; | |
58590b06 | 3868 | |
afcff5d8 LC |
3869 | /* |
3870 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3871 | * do not care for this case anymore. Simply remove the flag | |
3872 | * if there are no extents. | |
3873 | */ | |
3874 | if (unlikely(!eh->eh_entries)) | |
3875 | goto out; | |
58590b06 TT |
3876 | last_ex = EXT_LAST_EXTENT(eh); |
3877 | /* | |
3878 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3879 | * last block in the last extent in the file. We test this by | |
3880 | * first checking to see if the caller to | |
3881 | * ext4_ext_get_blocks() was interested in the last block (or | |
3882 | * a block beyond the last block) in the current extent. If | |
3883 | * this turns out to be false, we can bail out from this | |
3884 | * function immediately. | |
3885 | */ | |
d002ebf1 | 3886 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3887 | ext4_ext_get_actual_len(last_ex)) |
3888 | return 0; | |
3889 | /* | |
3890 | * If the caller does appear to be planning to write at or | |
3891 | * beyond the end of the current extent, we then test to see | |
3892 | * if the current extent is the last extent in the file, by | |
3893 | * checking to make sure it was reached via the rightmost node | |
3894 | * at each level of the tree. | |
3895 | */ | |
3896 | for (i = depth-1; i >= 0; i--) | |
3897 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3898 | return 0; | |
afcff5d8 | 3899 | out: |
58590b06 TT |
3900 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3901 | return ext4_mark_inode_dirty(handle, inode); | |
3902 | } | |
3903 | ||
b8a86845 | 3904 | static int |
e8b83d93 TT |
3905 | convert_initialized_extent(handle_t *handle, struct inode *inode, |
3906 | struct ext4_map_blocks *map, | |
29c6eaff | 3907 | struct ext4_ext_path **ppath, |
56263b4c | 3908 | unsigned int allocated) |
b8a86845 | 3909 | { |
4f224b8b | 3910 | struct ext4_ext_path *path = *ppath; |
e8b83d93 TT |
3911 | struct ext4_extent *ex; |
3912 | ext4_lblk_t ee_block; | |
3913 | unsigned int ee_len; | |
3914 | int depth; | |
b8a86845 LC |
3915 | int err = 0; |
3916 | ||
3917 | /* | |
3918 | * Make sure that the extent is no bigger than we support with | |
556615dc | 3919 | * unwritten extent |
b8a86845 | 3920 | */ |
556615dc LC |
3921 | if (map->m_len > EXT_UNWRITTEN_MAX_LEN) |
3922 | map->m_len = EXT_UNWRITTEN_MAX_LEN / 2; | |
b8a86845 | 3923 | |
e8b83d93 TT |
3924 | depth = ext_depth(inode); |
3925 | ex = path[depth].p_ext; | |
3926 | ee_block = le32_to_cpu(ex->ee_block); | |
3927 | ee_len = ext4_ext_get_actual_len(ex); | |
3928 | ||
3929 | ext_debug("%s: inode %lu, logical" | |
3930 | "block %llu, max_blocks %u\n", __func__, inode->i_ino, | |
3931 | (unsigned long long)ee_block, ee_len); | |
3932 | ||
3933 | if (ee_block != map->m_lblk || ee_len > map->m_len) { | |
dfe50809 | 3934 | err = ext4_split_convert_extents(handle, inode, map, ppath, |
e8b83d93 TT |
3935 | EXT4_GET_BLOCKS_CONVERT_UNWRITTEN); |
3936 | if (err < 0) | |
3937 | return err; | |
ed8a1a76 | 3938 | path = ext4_find_extent(inode, map->m_lblk, ppath, 0); |
e8b83d93 TT |
3939 | if (IS_ERR(path)) |
3940 | return PTR_ERR(path); | |
3941 | depth = ext_depth(inode); | |
3942 | ex = path[depth].p_ext; | |
3943 | if (!ex) { | |
3944 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
3945 | (unsigned long) map->m_lblk); | |
6a797d27 | 3946 | return -EFSCORRUPTED; |
e8b83d93 TT |
3947 | } |
3948 | } | |
3949 | ||
3950 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3951 | if (err) | |
3952 | return err; | |
3953 | /* first mark the extent as unwritten */ | |
3954 | ext4_ext_mark_unwritten(ex); | |
3955 | ||
3956 | /* note: ext4_ext_correct_indexes() isn't needed here because | |
3957 | * borders are not changed | |
3958 | */ | |
3959 | ext4_ext_try_to_merge(handle, inode, path, ex); | |
3960 | ||
3961 | /* Mark modified extent as dirty */ | |
3962 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
3963 | if (err) | |
3964 | return err; | |
3965 | ext4_ext_show_leaf(inode, path); | |
3966 | ||
3967 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
3968 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len); | |
3969 | if (err) | |
3970 | return err; | |
b8a86845 LC |
3971 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
3972 | if (allocated > map->m_len) | |
3973 | allocated = map->m_len; | |
3974 | map->m_len = allocated; | |
e8b83d93 | 3975 | return allocated; |
b8a86845 LC |
3976 | } |
3977 | ||
0031462b | 3978 | static int |
556615dc | 3979 | ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode, |
e35fd660 | 3980 | struct ext4_map_blocks *map, |
dfe50809 | 3981 | struct ext4_ext_path **ppath, int flags, |
e35fd660 | 3982 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b | 3983 | { |
dfe50809 | 3984 | struct ext4_ext_path *path = *ppath; |
0031462b MC |
3985 | int ret = 0; |
3986 | int err = 0; | |
3987 | ||
556615dc | 3988 | ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical " |
88635ca2 | 3989 | "block %llu, max_blocks %u, flags %x, allocated %u\n", |
e35fd660 | 3990 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3991 | flags, allocated); |
3992 | ext4_ext_show_leaf(inode, path); | |
3993 | ||
27dd4385 | 3994 | /* |
556615dc | 3995 | * When writing into unwritten space, we should not fail to |
27dd4385 LC |
3996 | * allocate metadata blocks for the new extent block if needed. |
3997 | */ | |
3998 | flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
3999 | ||
556615dc | 4000 | trace_ext4_ext_handle_unwritten_extents(inode, map, flags, |
b5645534 | 4001 | allocated, newblock); |
d8990240 | 4002 | |
c7064ef1 | 4003 | /* get_block() before submit the IO, split the extent */ |
c8b459f4 | 4004 | if (flags & EXT4_GET_BLOCKS_PRE_IO) { |
dfe50809 TT |
4005 | ret = ext4_split_convert_extents(handle, inode, map, ppath, |
4006 | flags | EXT4_GET_BLOCKS_CONVERT); | |
82e54229 DM |
4007 | if (ret <= 0) |
4008 | goto out; | |
a25a4e1a | 4009 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
4010 | goto out; |
4011 | } | |
c7064ef1 | 4012 | /* IO end_io complete, convert the filled extent to written */ |
c8b459f4 | 4013 | if (flags & EXT4_GET_BLOCKS_CONVERT) { |
c86d8db3 JK |
4014 | if (flags & EXT4_GET_BLOCKS_ZERO) { |
4015 | if (allocated > map->m_len) | |
4016 | allocated = map->m_len; | |
4017 | err = ext4_issue_zeroout(inode, map->m_lblk, newblock, | |
4018 | allocated); | |
4019 | if (err < 0) | |
4020 | goto out2; | |
4021 | } | |
dee1f973 | 4022 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
dfe50809 | 4023 | ppath); |
58590b06 | 4024 | if (ret >= 0) { |
b436b9be | 4025 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
4026 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
4027 | path, map->m_len); | |
58590b06 TT |
4028 | } else |
4029 | err = ret; | |
cdee7843 | 4030 | map->m_flags |= EXT4_MAP_MAPPED; |
15cc1767 | 4031 | map->m_pblk = newblock; |
cdee7843 ZL |
4032 | if (allocated > map->m_len) |
4033 | allocated = map->m_len; | |
4034 | map->m_len = allocated; | |
0031462b MC |
4035 | goto out2; |
4036 | } | |
4037 | /* buffered IO case */ | |
4038 | /* | |
4039 | * repeat fallocate creation request | |
4040 | * we already have an unwritten extent | |
4041 | */ | |
556615dc | 4042 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) { |
a25a4e1a | 4043 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b | 4044 | goto map_out; |
a25a4e1a | 4045 | } |
0031462b MC |
4046 | |
4047 | /* buffered READ or buffered write_begin() lookup */ | |
4048 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
4049 | /* | |
4050 | * We have blocks reserved already. We | |
4051 | * return allocated blocks so that delalloc | |
4052 | * won't do block reservation for us. But | |
4053 | * the buffer head will be unmapped so that | |
4054 | * a read from the block returns 0s. | |
4055 | */ | |
e35fd660 | 4056 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
4057 | goto out1; |
4058 | } | |
4059 | ||
4060 | /* buffered write, writepage time, convert*/ | |
dfe50809 | 4061 | ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags); |
a4e5d88b | 4062 | if (ret >= 0) |
b436b9be | 4063 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
4064 | out: |
4065 | if (ret <= 0) { | |
4066 | err = ret; | |
4067 | goto out2; | |
4068 | } else | |
4069 | allocated = ret; | |
e35fd660 | 4070 | map->m_flags |= EXT4_MAP_NEW; |
16e08b14 | 4071 | if (allocated > map->m_len) |
e35fd660 | 4072 | allocated = map->m_len; |
3a225670 | 4073 | map->m_len = allocated; |
5f634d06 | 4074 | |
0031462b | 4075 | map_out: |
e35fd660 | 4076 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
4077 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
4078 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
4079 | map->m_len); | |
4080 | if (err < 0) | |
4081 | goto out2; | |
4082 | } | |
0031462b | 4083 | out1: |
e35fd660 TT |
4084 | if (allocated > map->m_len) |
4085 | allocated = map->m_len; | |
0031462b | 4086 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4087 | map->m_pblk = newblock; |
4088 | map->m_len = allocated; | |
0031462b | 4089 | out2: |
0031462b MC |
4090 | return err ? err : allocated; |
4091 | } | |
58590b06 | 4092 | |
4d33b1ef TT |
4093 | /* |
4094 | * get_implied_cluster_alloc - check to see if the requested | |
4095 | * allocation (in the map structure) overlaps with a cluster already | |
4096 | * allocated in an extent. | |
d8990240 | 4097 | * @sb The filesystem superblock structure |
4d33b1ef TT |
4098 | * @map The requested lblk->pblk mapping |
4099 | * @ex The extent structure which might contain an implied | |
4100 | * cluster allocation | |
4101 | * | |
4102 | * This function is called by ext4_ext_map_blocks() after we failed to | |
4103 | * find blocks that were already in the inode's extent tree. Hence, | |
4104 | * we know that the beginning of the requested region cannot overlap | |
4105 | * the extent from the inode's extent tree. There are three cases we | |
4106 | * want to catch. The first is this case: | |
4107 | * | |
4108 | * |--- cluster # N--| | |
4109 | * |--- extent ---| |---- requested region ---| | |
4110 | * |==========| | |
4111 | * | |
4112 | * The second case that we need to test for is this one: | |
4113 | * | |
4114 | * |--------- cluster # N ----------------| | |
4115 | * |--- requested region --| |------- extent ----| | |
4116 | * |=======================| | |
4117 | * | |
4118 | * The third case is when the requested region lies between two extents | |
4119 | * within the same cluster: | |
4120 | * |------------- cluster # N-------------| | |
4121 | * |----- ex -----| |---- ex_right ----| | |
4122 | * |------ requested region ------| | |
4123 | * |================| | |
4124 | * | |
4125 | * In each of the above cases, we need to set the map->m_pblk and | |
4126 | * map->m_len so it corresponds to the return the extent labelled as | |
4127 | * "|====|" from cluster #N, since it is already in use for data in | |
4128 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
4129 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
4130 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
4131 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
4132 | * by calling ext4_mb_new_blocks(). | |
4133 | */ | |
d8990240 | 4134 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
4135 | struct ext4_map_blocks *map, |
4136 | struct ext4_extent *ex, | |
4137 | struct ext4_ext_path *path) | |
4138 | { | |
d8990240 | 4139 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
f5a44db5 | 4140 | ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef | 4141 | ext4_lblk_t ex_cluster_start, ex_cluster_end; |
14d7f3ef | 4142 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
4143 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
4144 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
4145 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
4146 | ||
4147 | /* The extent passed in that we are trying to match */ | |
4148 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
4149 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
4150 | ||
4151 | /* The requested region passed into ext4_map_blocks() */ | |
4152 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
4153 | |
4154 | if ((rr_cluster_start == ex_cluster_end) || | |
4155 | (rr_cluster_start == ex_cluster_start)) { | |
4156 | if (rr_cluster_start == ex_cluster_end) | |
4157 | ee_start += ee_len - 1; | |
f5a44db5 | 4158 | map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset; |
4d33b1ef TT |
4159 | map->m_len = min(map->m_len, |
4160 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
4161 | /* | |
4162 | * Check for and handle this case: | |
4163 | * | |
4164 | * |--------- cluster # N-------------| | |
4165 | * |------- extent ----| | |
4166 | * |--- requested region ---| | |
4167 | * |===========| | |
4168 | */ | |
4169 | ||
4170 | if (map->m_lblk < ee_block) | |
4171 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
4172 | ||
4173 | /* | |
4174 | * Check for the case where there is already another allocated | |
4175 | * block to the right of 'ex' but before the end of the cluster. | |
4176 | * | |
4177 | * |------------- cluster # N-------------| | |
4178 | * |----- ex -----| |---- ex_right ----| | |
4179 | * |------ requested region ------| | |
4180 | * |================| | |
4181 | */ | |
4182 | if (map->m_lblk > ee_block) { | |
4183 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
4184 | map->m_len = min(map->m_len, next - map->m_lblk); | |
4185 | } | |
d8990240 AK |
4186 | |
4187 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
4188 | return 1; |
4189 | } | |
d8990240 AK |
4190 | |
4191 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
4192 | return 0; |
4193 | } | |
4194 | ||
4195 | ||
c278bfec | 4196 | /* |
f5ab0d1f MC |
4197 | * Block allocation/map/preallocation routine for extents based files |
4198 | * | |
4199 | * | |
c278bfec | 4200 | * Need to be called with |
0e855ac8 AK |
4201 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
4202 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
4203 | * |
4204 | * return > 0, number of of blocks already mapped/allocated | |
4205 | * if create == 0 and these are pre-allocated blocks | |
4206 | * buffer head is unmapped | |
4207 | * otherwise blocks are mapped | |
4208 | * | |
4209 | * return = 0, if plain look up failed (blocks have not been allocated) | |
4210 | * buffer head is unmapped | |
4211 | * | |
4212 | * return < 0, error case. | |
c278bfec | 4213 | */ |
e35fd660 TT |
4214 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
4215 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
4216 | { |
4217 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
4218 | struct ext4_extent newex, *ex, *ex2; |
4219 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 4220 | ext4_fsblk_t newblock = 0; |
ce37c429 | 4221 | int free_on_err = 0, err = 0, depth, ret; |
4d33b1ef | 4222 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 4223 | unsigned int allocated_clusters = 0; |
c9de560d | 4224 | struct ext4_allocation_request ar; |
4d33b1ef | 4225 | ext4_lblk_t cluster_offset; |
cbd7584e | 4226 | bool map_from_cluster = false; |
a86c6181 | 4227 | |
84fe3bef | 4228 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4229 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4230 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4231 | |
a86c6181 | 4232 | /* find extent for this block */ |
ed8a1a76 | 4233 | path = ext4_find_extent(inode, map->m_lblk, NULL, 0); |
a86c6181 AT |
4234 | if (IS_ERR(path)) { |
4235 | err = PTR_ERR(path); | |
4236 | path = NULL; | |
4237 | goto out2; | |
4238 | } | |
4239 | ||
4240 | depth = ext_depth(inode); | |
4241 | ||
4242 | /* | |
d0d856e8 RD |
4243 | * consistent leaf must not be empty; |
4244 | * this situation is possible, though, _during_ tree modification; | |
ed8a1a76 | 4245 | * this is why assert can't be put in ext4_find_extent() |
a86c6181 | 4246 | */ |
273df556 FM |
4247 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4248 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4249 | "lblock: %lu, depth: %d pblock %lld", |
4250 | (unsigned long) map->m_lblk, depth, | |
4251 | path[depth].p_block); | |
6a797d27 | 4252 | err = -EFSCORRUPTED; |
034fb4c9 SP |
4253 | goto out2; |
4254 | } | |
a86c6181 | 4255 | |
7e028976 AM |
4256 | ex = path[depth].p_ext; |
4257 | if (ex) { | |
725d26d3 | 4258 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4259 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4260 | unsigned short ee_len; |
471d4011 | 4261 | |
b8a86845 | 4262 | |
471d4011 | 4263 | /* |
556615dc | 4264 | * unwritten extents are treated as holes, except that |
56055d3a | 4265 | * we split out initialized portions during a write. |
471d4011 | 4266 | */ |
a2df2a63 | 4267 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4268 | |
4269 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4270 | ||
d0d856e8 | 4271 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4272 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4273 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4274 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4275 | allocated = ee_len - (map->m_lblk - ee_block); |
4276 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4277 | ee_block, ee_len, newblock); | |
56055d3a | 4278 | |
b8a86845 LC |
4279 | /* |
4280 | * If the extent is initialized check whether the | |
4281 | * caller wants to convert it to unwritten. | |
4282 | */ | |
556615dc | 4283 | if ((!ext4_ext_is_unwritten(ex)) && |
b8a86845 | 4284 | (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) { |
e8b83d93 | 4285 | allocated = convert_initialized_extent( |
4f224b8b | 4286 | handle, inode, map, &path, |
29c6eaff | 4287 | allocated); |
b8a86845 | 4288 | goto out2; |
556615dc | 4289 | } else if (!ext4_ext_is_unwritten(ex)) |
7877191c | 4290 | goto out; |
69eb33dc | 4291 | |
556615dc | 4292 | ret = ext4_ext_handle_unwritten_extents( |
dfe50809 | 4293 | handle, inode, map, &path, flags, |
7877191c | 4294 | allocated, newblock); |
ce37c429 EW |
4295 | if (ret < 0) |
4296 | err = ret; | |
4297 | else | |
4298 | allocated = ret; | |
31cf0f2c | 4299 | goto out2; |
a86c6181 AT |
4300 | } |
4301 | } | |
4302 | ||
4303 | /* | |
d0d856e8 | 4304 | * requested block isn't allocated yet; |
a86c6181 AT |
4305 | * we couldn't try to create block if create flag is zero |
4306 | */ | |
c2177057 | 4307 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
140a5250 JK |
4308 | ext4_lblk_t hole_start, hole_len; |
4309 | ||
facab4d9 JK |
4310 | hole_start = map->m_lblk; |
4311 | hole_len = ext4_ext_determine_hole(inode, path, &hole_start); | |
56055d3a AA |
4312 | /* |
4313 | * put just found gap into cache to speed up | |
4314 | * subsequent requests | |
4315 | */ | |
140a5250 | 4316 | ext4_ext_put_gap_in_cache(inode, hole_start, hole_len); |
facab4d9 JK |
4317 | |
4318 | /* Update hole_len to reflect hole size after map->m_lblk */ | |
4319 | if (hole_start != map->m_lblk) | |
4320 | hole_len -= map->m_lblk - hole_start; | |
4321 | map->m_pblk = 0; | |
4322 | map->m_len = min_t(unsigned int, map->m_len, hole_len); | |
4323 | ||
a86c6181 AT |
4324 | goto out2; |
4325 | } | |
4d33b1ef | 4326 | |
a86c6181 | 4327 | /* |
c2ea3fde | 4328 | * Okay, we need to do block allocation. |
63f57933 | 4329 | */ |
4d33b1ef | 4330 | newex.ee_block = cpu_to_le32(map->m_lblk); |
d0abafac | 4331 | cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4332 | |
4333 | /* | |
4334 | * If we are doing bigalloc, check to see if the extent returned | |
ed8a1a76 | 4335 | * by ext4_find_extent() implies a cluster we can use. |
4d33b1ef TT |
4336 | */ |
4337 | if (cluster_offset && ex && | |
d8990240 | 4338 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4339 | ar.len = allocated = map->m_len; |
4340 | newblock = map->m_pblk; | |
cbd7584e | 4341 | map_from_cluster = true; |
4d33b1ef TT |
4342 | goto got_allocated_blocks; |
4343 | } | |
a86c6181 | 4344 | |
c9de560d | 4345 | /* find neighbour allocated blocks */ |
e35fd660 | 4346 | ar.lleft = map->m_lblk; |
c9de560d AT |
4347 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4348 | if (err) | |
4349 | goto out2; | |
e35fd660 | 4350 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4351 | ex2 = NULL; |
4352 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4353 | if (err) |
4354 | goto out2; | |
25d14f98 | 4355 | |
4d33b1ef TT |
4356 | /* Check if the extent after searching to the right implies a |
4357 | * cluster we can use. */ | |
4358 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4359 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4360 | ar.len = allocated = map->m_len; |
4361 | newblock = map->m_pblk; | |
cbd7584e | 4362 | map_from_cluster = true; |
4d33b1ef TT |
4363 | goto got_allocated_blocks; |
4364 | } | |
4365 | ||
749269fa AA |
4366 | /* |
4367 | * See if request is beyond maximum number of blocks we can have in | |
4368 | * a single extent. For an initialized extent this limit is | |
556615dc LC |
4369 | * EXT_INIT_MAX_LEN and for an unwritten extent this limit is |
4370 | * EXT_UNWRITTEN_MAX_LEN. | |
749269fa | 4371 | */ |
e35fd660 | 4372 | if (map->m_len > EXT_INIT_MAX_LEN && |
556615dc | 4373 | !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) |
e35fd660 | 4374 | map->m_len = EXT_INIT_MAX_LEN; |
556615dc LC |
4375 | else if (map->m_len > EXT_UNWRITTEN_MAX_LEN && |
4376 | (flags & EXT4_GET_BLOCKS_UNWRIT_EXT)) | |
4377 | map->m_len = EXT_UNWRITTEN_MAX_LEN; | |
749269fa | 4378 | |
e35fd660 | 4379 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4380 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4381 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4382 | if (err) |
b939e376 | 4383 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4384 | else |
e35fd660 | 4385 | allocated = map->m_len; |
c9de560d AT |
4386 | |
4387 | /* allocate new block */ | |
4388 | ar.inode = inode; | |
e35fd660 TT |
4389 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4390 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4391 | /* |
4392 | * We calculate the offset from the beginning of the cluster | |
4393 | * for the logical block number, since when we allocate a | |
4394 | * physical cluster, the physical block should start at the | |
4395 | * same offset from the beginning of the cluster. This is | |
4396 | * needed so that future calls to get_implied_cluster_alloc() | |
4397 | * work correctly. | |
4398 | */ | |
f5a44db5 | 4399 | offset = EXT4_LBLK_COFF(sbi, map->m_lblk); |
4d33b1ef TT |
4400 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); |
4401 | ar.goal -= offset; | |
4402 | ar.logical -= offset; | |
c9de560d AT |
4403 | if (S_ISREG(inode->i_mode)) |
4404 | ar.flags = EXT4_MB_HINT_DATA; | |
4405 | else | |
4406 | /* disable in-core preallocation for non-regular files */ | |
4407 | ar.flags = 0; | |
556b27ab VH |
4408 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4409 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
e3cf5d5d TT |
4410 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
4411 | ar.flags |= EXT4_MB_DELALLOC_RESERVED; | |
c5e298ae TT |
4412 | if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
4413 | ar.flags |= EXT4_MB_USE_RESERVED; | |
c9de560d | 4414 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4415 | if (!newblock) |
4416 | goto out2; | |
84fe3bef | 4417 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4418 | ar.goal, newblock, allocated); |
4d33b1ef | 4419 | free_on_err = 1; |
7b415bf6 | 4420 | allocated_clusters = ar.len; |
4d33b1ef TT |
4421 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4422 | if (ar.len > allocated) | |
4423 | ar.len = allocated; | |
a86c6181 | 4424 | |
4d33b1ef | 4425 | got_allocated_blocks: |
a86c6181 | 4426 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4427 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4428 | newex.ee_len = cpu_to_le16(ar.len); |
556615dc LC |
4429 | /* Mark unwritten */ |
4430 | if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){ | |
4431 | ext4_ext_mark_unwritten(&newex); | |
a25a4e1a | 4432 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4433 | } |
c8d46e41 | 4434 | |
a4e5d88b DM |
4435 | err = 0; |
4436 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4437 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4438 | path, ar.len); | |
575a1d4b | 4439 | if (!err) |
dfe50809 | 4440 | err = ext4_ext_insert_extent(handle, inode, &path, |
575a1d4b | 4441 | &newex, flags); |
82e54229 | 4442 | |
4d33b1ef | 4443 | if (err && free_on_err) { |
7132de74 MP |
4444 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4445 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4446 | /* free data blocks we just allocated */ |
c9de560d AT |
4447 | /* not a good idea to call discard here directly, |
4448 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4449 | ext4_discard_preallocations(inode); |
c8e15130 TT |
4450 | ext4_free_blocks(handle, inode, NULL, newblock, |
4451 | EXT4_C2B(sbi, allocated_clusters), fb_flags); | |
a86c6181 | 4452 | goto out2; |
315054f0 | 4453 | } |
a86c6181 | 4454 | |
a86c6181 | 4455 | /* previous routine could use block we allocated */ |
bf89d16f | 4456 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4457 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4458 | if (allocated > map->m_len) |
4459 | allocated = map->m_len; | |
4460 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4461 | |
5f634d06 | 4462 | /* |
b6bf9171 EW |
4463 | * Reduce the reserved cluster count to reflect successful deferred |
4464 | * allocation of delayed allocated clusters or direct allocation of | |
4465 | * clusters discovered to be delayed allocated. Once allocated, a | |
4466 | * cluster is not included in the reserved count. | |
5f634d06 | 4467 | */ |
b6bf9171 EW |
4468 | if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) { |
4469 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { | |
232ec872 | 4470 | /* |
b6bf9171 EW |
4471 | * When allocating delayed allocated clusters, simply |
4472 | * reduce the reserved cluster count and claim quota | |
232ec872 LC |
4473 | */ |
4474 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4475 | 1); | |
b6bf9171 EW |
4476 | } else { |
4477 | ext4_lblk_t lblk, len; | |
4478 | unsigned int n; | |
4479 | ||
4480 | /* | |
4481 | * When allocating non-delayed allocated clusters | |
4482 | * (from fallocate, filemap, DIO, or clusters | |
4483 | * allocated when delalloc has been disabled by | |
4484 | * ext4_nonda_switch), reduce the reserved cluster | |
4485 | * count by the number of allocated clusters that | |
4486 | * have previously been delayed allocated. Quota | |
4487 | * has been claimed by ext4_mb_new_blocks() above, | |
4488 | * so release the quota reservations made for any | |
4489 | * previously delayed allocated clusters. | |
4490 | */ | |
4491 | lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk); | |
4492 | len = allocated_clusters << sbi->s_cluster_bits; | |
4493 | n = ext4_es_delayed_clu(inode, lblk, len); | |
4494 | if (n > 0) | |
4495 | ext4_da_update_reserve_space(inode, (int) n, 0); | |
7b415bf6 AK |
4496 | } |
4497 | } | |
5f634d06 | 4498 | |
b436b9be JK |
4499 | /* |
4500 | * Cache the extent and update transaction to commit on fdatasync only | |
556615dc | 4501 | * when it is _not_ an unwritten extent. |
b436b9be | 4502 | */ |
556615dc | 4503 | if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0) |
b436b9be | 4504 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4505 | else |
b436b9be | 4506 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4507 | out: |
e35fd660 TT |
4508 | if (allocated > map->m_len) |
4509 | allocated = map->m_len; | |
a86c6181 | 4510 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4511 | map->m_flags |= EXT4_MAP_MAPPED; |
4512 | map->m_pblk = newblock; | |
4513 | map->m_len = allocated; | |
a86c6181 | 4514 | out2: |
b7ea89ad TT |
4515 | ext4_ext_drop_refs(path); |
4516 | kfree(path); | |
e861304b | 4517 | |
63b99968 TT |
4518 | trace_ext4_ext_map_blocks_exit(inode, flags, map, |
4519 | err ? err : allocated); | |
7877191c | 4520 | return err ? err : allocated; |
a86c6181 AT |
4521 | } |
4522 | ||
d0abb36d | 4523 | int ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4524 | { |
a86c6181 | 4525 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4526 | ext4_lblk_t last_block; |
a86c6181 AT |
4527 | int err = 0; |
4528 | ||
a86c6181 | 4529 | /* |
d0d856e8 RD |
4530 | * TODO: optimization is possible here. |
4531 | * Probably we need not scan at all, | |
4532 | * because page truncation is enough. | |
a86c6181 | 4533 | */ |
a86c6181 AT |
4534 | |
4535 | /* we have to know where to truncate from in crash case */ | |
4536 | EXT4_I(inode)->i_disksize = inode->i_size; | |
d0abb36d TT |
4537 | err = ext4_mark_inode_dirty(handle, inode); |
4538 | if (err) | |
4539 | return err; | |
a86c6181 AT |
4540 | |
4541 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4542 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
8acd5e9b | 4543 | retry: |
51865fda ZL |
4544 | err = ext4_es_remove_extent(inode, last_block, |
4545 | EXT_MAX_BLOCKS - last_block); | |
94eec0fc | 4546 | if (err == -ENOMEM) { |
8acd5e9b TT |
4547 | cond_resched(); |
4548 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
4549 | goto retry; | |
4550 | } | |
d0abb36d TT |
4551 | if (err) |
4552 | return err; | |
4553 | return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); | |
a86c6181 AT |
4554 | } |
4555 | ||
0e8b6879 | 4556 | static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset, |
c174e6d6 | 4557 | ext4_lblk_t len, loff_t new_size, |
77a2e84d | 4558 | int flags) |
0e8b6879 LC |
4559 | { |
4560 | struct inode *inode = file_inode(file); | |
4561 | handle_t *handle; | |
4562 | int ret = 0; | |
4563 | int ret2 = 0; | |
4564 | int retries = 0; | |
4134f5c8 | 4565 | int depth = 0; |
0e8b6879 LC |
4566 | struct ext4_map_blocks map; |
4567 | unsigned int credits; | |
c174e6d6 | 4568 | loff_t epos; |
0e8b6879 | 4569 | |
c3fe493c | 4570 | BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)); |
0e8b6879 | 4571 | map.m_lblk = offset; |
c174e6d6 | 4572 | map.m_len = len; |
0e8b6879 LC |
4573 | /* |
4574 | * Don't normalize the request if it can fit in one extent so | |
4575 | * that it doesn't get unnecessarily split into multiple | |
4576 | * extents. | |
4577 | */ | |
556615dc | 4578 | if (len <= EXT_UNWRITTEN_MAX_LEN) |
0e8b6879 LC |
4579 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; |
4580 | ||
4581 | /* | |
4582 | * credits to insert 1 extent into extent tree | |
4583 | */ | |
4584 | credits = ext4_chunk_trans_blocks(inode, len); | |
c3fe493c | 4585 | depth = ext_depth(inode); |
0e8b6879 LC |
4586 | |
4587 | retry: | |
c174e6d6 | 4588 | while (ret >= 0 && len) { |
4134f5c8 LC |
4589 | /* |
4590 | * Recalculate credits when extent tree depth changes. | |
4591 | */ | |
011c88e3 | 4592 | if (depth != ext_depth(inode)) { |
4134f5c8 LC |
4593 | credits = ext4_chunk_trans_blocks(inode, len); |
4594 | depth = ext_depth(inode); | |
4595 | } | |
4596 | ||
0e8b6879 LC |
4597 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4598 | credits); | |
4599 | if (IS_ERR(handle)) { | |
4600 | ret = PTR_ERR(handle); | |
4601 | break; | |
4602 | } | |
4603 | ret = ext4_map_blocks(handle, inode, &map, flags); | |
4604 | if (ret <= 0) { | |
4605 | ext4_debug("inode #%lu: block %u: len %u: " | |
4606 | "ext4_ext_map_blocks returned %d", | |
4607 | inode->i_ino, map.m_lblk, | |
4608 | map.m_len, ret); | |
4609 | ext4_mark_inode_dirty(handle, inode); | |
4610 | ret2 = ext4_journal_stop(handle); | |
4611 | break; | |
4612 | } | |
c174e6d6 DM |
4613 | map.m_lblk += ret; |
4614 | map.m_len = len = len - ret; | |
4615 | epos = (loff_t)map.m_lblk << inode->i_blkbits; | |
eeca7ea1 | 4616 | inode->i_ctime = current_time(inode); |
c174e6d6 DM |
4617 | if (new_size) { |
4618 | if (epos > new_size) | |
4619 | epos = new_size; | |
4620 | if (ext4_update_inode_size(inode, epos) & 0x1) | |
4621 | inode->i_mtime = inode->i_ctime; | |
4622 | } else { | |
4623 | if (epos > inode->i_size) | |
4624 | ext4_set_inode_flag(inode, | |
4625 | EXT4_INODE_EOFBLOCKS); | |
4626 | } | |
4627 | ext4_mark_inode_dirty(handle, inode); | |
c894aa97 | 4628 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0e8b6879 LC |
4629 | ret2 = ext4_journal_stop(handle); |
4630 | if (ret2) | |
4631 | break; | |
4632 | } | |
4633 | if (ret == -ENOSPC && | |
4634 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4635 | ret = 0; | |
4636 | goto retry; | |
4637 | } | |
4638 | ||
4639 | return ret > 0 ? ret2 : ret; | |
4640 | } | |
4641 | ||
b8a86845 LC |
4642 | static long ext4_zero_range(struct file *file, loff_t offset, |
4643 | loff_t len, int mode) | |
4644 | { | |
4645 | struct inode *inode = file_inode(file); | |
4646 | handle_t *handle = NULL; | |
4647 | unsigned int max_blocks; | |
4648 | loff_t new_size = 0; | |
4649 | int ret = 0; | |
4650 | int flags; | |
69dc9536 | 4651 | int credits; |
c174e6d6 | 4652 | int partial_begin, partial_end; |
b8a86845 LC |
4653 | loff_t start, end; |
4654 | ext4_lblk_t lblk; | |
b8a86845 LC |
4655 | unsigned int blkbits = inode->i_blkbits; |
4656 | ||
4657 | trace_ext4_zero_range(inode, offset, len, mode); | |
4658 | ||
6c5e73d3 | 4659 | if (!S_ISREG(inode->i_mode)) |
4660 | return -EINVAL; | |
4661 | ||
e1ee60fd NJ |
4662 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
4663 | if (ext4_should_journal_data(inode)) { | |
4664 | ret = ext4_force_commit(inode->i_sb); | |
4665 | if (ret) | |
4666 | return ret; | |
4667 | } | |
4668 | ||
b8a86845 LC |
4669 | /* |
4670 | * Round up offset. This is not fallocate, we neet to zero out | |
4671 | * blocks, so convert interior block aligned part of the range to | |
4672 | * unwritten and possibly manually zero out unaligned parts of the | |
4673 | * range. | |
4674 | */ | |
4675 | start = round_up(offset, 1 << blkbits); | |
4676 | end = round_down((offset + len), 1 << blkbits); | |
4677 | ||
4678 | if (start < offset || end > offset + len) | |
4679 | return -EINVAL; | |
c174e6d6 DM |
4680 | partial_begin = offset & ((1 << blkbits) - 1); |
4681 | partial_end = (offset + len) & ((1 << blkbits) - 1); | |
b8a86845 LC |
4682 | |
4683 | lblk = start >> blkbits; | |
4684 | max_blocks = (end >> blkbits); | |
4685 | if (max_blocks < lblk) | |
4686 | max_blocks = 0; | |
4687 | else | |
4688 | max_blocks -= lblk; | |
4689 | ||
5955102c | 4690 | inode_lock(inode); |
b8a86845 LC |
4691 | |
4692 | /* | |
4693 | * Indirect files do not support unwritten extnets | |
4694 | */ | |
4695 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
4696 | ret = -EOPNOTSUPP; | |
4697 | goto out_mutex; | |
4698 | } | |
4699 | ||
4700 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
51e3ae81 TT |
4701 | (offset + len > i_size_read(inode) || |
4702 | offset + len > EXT4_I(inode)->i_disksize)) { | |
b8a86845 LC |
4703 | new_size = offset + len; |
4704 | ret = inode_newsize_ok(inode, new_size); | |
4705 | if (ret) | |
4706 | goto out_mutex; | |
b8a86845 LC |
4707 | } |
4708 | ||
0f2af21a LC |
4709 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
4710 | if (mode & FALLOC_FL_KEEP_SIZE) | |
4711 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4712 | ||
17048e8a | 4713 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
17048e8a JK |
4714 | inode_dio_wait(inode); |
4715 | ||
0f2af21a LC |
4716 | /* Preallocate the range including the unaligned edges */ |
4717 | if (partial_begin || partial_end) { | |
4718 | ret = ext4_alloc_file_blocks(file, | |
4719 | round_down(offset, 1 << blkbits) >> blkbits, | |
4720 | (round_up((offset + len), 1 << blkbits) - | |
4721 | round_down(offset, 1 << blkbits)) >> blkbits, | |
77a2e84d | 4722 | new_size, flags); |
0f2af21a | 4723 | if (ret) |
1d39834f | 4724 | goto out_mutex; |
0f2af21a LC |
4725 | |
4726 | } | |
4727 | ||
4728 | /* Zero range excluding the unaligned edges */ | |
b8a86845 | 4729 | if (max_blocks > 0) { |
0f2af21a LC |
4730 | flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN | |
4731 | EXT4_EX_NOCACHE); | |
b8a86845 | 4732 | |
ea3d7209 JK |
4733 | /* |
4734 | * Prevent page faults from reinstantiating pages we have | |
4735 | * released from page cache. | |
4736 | */ | |
4737 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
4738 | |
4739 | ret = ext4_break_layouts(inode); | |
4740 | if (ret) { | |
4741 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4742 | goto out_mutex; | |
4743 | } | |
4744 | ||
01127848 JK |
4745 | ret = ext4_update_disksize_before_punch(inode, offset, len); |
4746 | if (ret) { | |
4747 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
1d39834f | 4748 | goto out_mutex; |
01127848 | 4749 | } |
ea3d7209 JK |
4750 | /* Now release the pages and zero block aligned part of pages */ |
4751 | truncate_pagecache_range(inode, start, end - 1); | |
eeca7ea1 | 4752 | inode->i_mtime = inode->i_ctime = current_time(inode); |
ea3d7209 | 4753 | |
713e8dde | 4754 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, |
77a2e84d | 4755 | flags); |
ea3d7209 | 4756 | up_write(&EXT4_I(inode)->i_mmap_sem); |
713e8dde | 4757 | if (ret) |
1d39834f | 4758 | goto out_mutex; |
b8a86845 | 4759 | } |
c174e6d6 | 4760 | if (!partial_begin && !partial_end) |
1d39834f | 4761 | goto out_mutex; |
c174e6d6 | 4762 | |
69dc9536 DM |
4763 | /* |
4764 | * In worst case we have to writeout two nonadjacent unwritten | |
4765 | * blocks and update the inode | |
4766 | */ | |
4767 | credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1; | |
4768 | if (ext4_should_journal_data(inode)) | |
4769 | credits += 2; | |
4770 | handle = ext4_journal_start(inode, EXT4_HT_MISC, credits); | |
b8a86845 LC |
4771 | if (IS_ERR(handle)) { |
4772 | ret = PTR_ERR(handle); | |
4773 | ext4_std_error(inode->i_sb, ret); | |
1d39834f | 4774 | goto out_mutex; |
b8a86845 LC |
4775 | } |
4776 | ||
eeca7ea1 | 4777 | inode->i_mtime = inode->i_ctime = current_time(inode); |
e5b30416 | 4778 | if (new_size) { |
4631dbf6 | 4779 | ext4_update_inode_size(inode, new_size); |
e5b30416 | 4780 | } else { |
b8a86845 LC |
4781 | /* |
4782 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4783 | * can proceed even if the new size is the same as i_size. | |
4784 | */ | |
4785 | if ((offset + len) > i_size_read(inode)) | |
4786 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); | |
4787 | } | |
b8a86845 LC |
4788 | ext4_mark_inode_dirty(handle, inode); |
4789 | ||
4790 | /* Zero out partial block at the edges of the range */ | |
4791 | ret = ext4_zero_partial_blocks(handle, inode, offset, len); | |
67a7d5f5 JK |
4792 | if (ret >= 0) |
4793 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
b8a86845 LC |
4794 | |
4795 | if (file->f_flags & O_SYNC) | |
4796 | ext4_handle_sync(handle); | |
4797 | ||
4798 | ext4_journal_stop(handle); | |
b8a86845 | 4799 | out_mutex: |
5955102c | 4800 | inode_unlock(inode); |
b8a86845 LC |
4801 | return ret; |
4802 | } | |
4803 | ||
a2df2a63 | 4804 | /* |
2fe17c10 | 4805 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4806 | * operation, which gets called from sys_fallocate system call. |
4807 | * For block-mapped files, posix_fallocate should fall back to the method | |
4808 | * of writing zeroes to the required new blocks (the same behavior which is | |
4809 | * expected for file systems which do not support fallocate() system call). | |
4810 | */ | |
2fe17c10 | 4811 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4812 | { |
496ad9aa | 4813 | struct inode *inode = file_inode(file); |
f282ac19 | 4814 | loff_t new_size = 0; |
498e5f24 | 4815 | unsigned int max_blocks; |
a2df2a63 | 4816 | int ret = 0; |
a4e5d88b | 4817 | int flags; |
0e8b6879 | 4818 | ext4_lblk_t lblk; |
0e8b6879 | 4819 | unsigned int blkbits = inode->i_blkbits; |
a2df2a63 | 4820 | |
2058f83a MH |
4821 | /* |
4822 | * Encrypted inodes can't handle collapse range or insert | |
4823 | * range since we would need to re-encrypt blocks with a | |
4824 | * different IV or XTS tweak (which are based on the logical | |
4825 | * block number). | |
4826 | * | |
4827 | * XXX It's not clear why zero range isn't working, but we'll | |
4828 | * leave it disabled for encrypted inodes for now. This is a | |
4829 | * bug we should fix.... | |
4830 | */ | |
592ddec7 | 4831 | if (IS_ENCRYPTED(inode) && |
331573fe NJ |
4832 | (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE | |
4833 | FALLOC_FL_ZERO_RANGE))) | |
2058f83a MH |
4834 | return -EOPNOTSUPP; |
4835 | ||
a4bb6b64 | 4836 | /* Return error if mode is not supported */ |
9eb79482 | 4837 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
331573fe NJ |
4838 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | |
4839 | FALLOC_FL_INSERT_RANGE)) | |
a4bb6b64 AH |
4840 | return -EOPNOTSUPP; |
4841 | ||
4842 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
aeb2817a | 4843 | return ext4_punch_hole(inode, offset, len); |
a4bb6b64 | 4844 | |
0c8d414f TM |
4845 | ret = ext4_convert_inline_data(inode); |
4846 | if (ret) | |
4847 | return ret; | |
4848 | ||
40c406c7 TT |
4849 | if (mode & FALLOC_FL_COLLAPSE_RANGE) |
4850 | return ext4_collapse_range(inode, offset, len); | |
4851 | ||
331573fe NJ |
4852 | if (mode & FALLOC_FL_INSERT_RANGE) |
4853 | return ext4_insert_range(inode, offset, len); | |
4854 | ||
b8a86845 LC |
4855 | if (mode & FALLOC_FL_ZERO_RANGE) |
4856 | return ext4_zero_range(file, offset, len, mode); | |
4857 | ||
0562e0ba | 4858 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
0e8b6879 | 4859 | lblk = offset >> blkbits; |
0e8b6879 | 4860 | |
518eaa63 | 4861 | max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits); |
556615dc | 4862 | flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT; |
0e8b6879 LC |
4863 | if (mode & FALLOC_FL_KEEP_SIZE) |
4864 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
4865 | ||
5955102c | 4866 | inode_lock(inode); |
f282ac19 | 4867 | |
280227a7 DI |
4868 | /* |
4869 | * We only support preallocation for extent-based files only | |
4870 | */ | |
4871 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
4872 | ret = -EOPNOTSUPP; | |
4873 | goto out; | |
4874 | } | |
4875 | ||
f282ac19 | 4876 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
51e3ae81 TT |
4877 | (offset + len > i_size_read(inode) || |
4878 | offset + len > EXT4_I(inode)->i_disksize)) { | |
f282ac19 LC |
4879 | new_size = offset + len; |
4880 | ret = inode_newsize_ok(inode, new_size); | |
4881 | if (ret) | |
4882 | goto out; | |
6d19c42b | 4883 | } |
f282ac19 | 4884 | |
17048e8a | 4885 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
17048e8a JK |
4886 | inode_dio_wait(inode); |
4887 | ||
77a2e84d | 4888 | ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags); |
0e8b6879 LC |
4889 | if (ret) |
4890 | goto out; | |
f282ac19 | 4891 | |
c174e6d6 DM |
4892 | if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) { |
4893 | ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal, | |
4894 | EXT4_I(inode)->i_sync_tid); | |
f282ac19 | 4895 | } |
f282ac19 | 4896 | out: |
5955102c | 4897 | inode_unlock(inode); |
0e8b6879 LC |
4898 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
4899 | return ret; | |
a2df2a63 | 4900 | } |
6873fa0d | 4901 | |
0031462b MC |
4902 | /* |
4903 | * This function convert a range of blocks to written extents | |
4904 | * The caller of this function will pass the start offset and the size. | |
4905 | * all unwritten extents within this range will be converted to | |
4906 | * written extents. | |
4907 | * | |
4908 | * This function is called from the direct IO end io call back | |
4909 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4910 | * Returns 0 on success. |
0031462b | 4911 | */ |
6b523df4 JK |
4912 | int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
4913 | loff_t offset, ssize_t len) | |
0031462b | 4914 | { |
0031462b MC |
4915 | unsigned int max_blocks; |
4916 | int ret = 0; | |
4917 | int ret2 = 0; | |
2ed88685 | 4918 | struct ext4_map_blocks map; |
0031462b MC |
4919 | unsigned int credits, blkbits = inode->i_blkbits; |
4920 | ||
2ed88685 | 4921 | map.m_lblk = offset >> blkbits; |
518eaa63 FF |
4922 | max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits); |
4923 | ||
0031462b | 4924 | /* |
6b523df4 JK |
4925 | * This is somewhat ugly but the idea is clear: When transaction is |
4926 | * reserved, everything goes into it. Otherwise we rather start several | |
4927 | * smaller transactions for conversion of each extent separately. | |
0031462b | 4928 | */ |
6b523df4 JK |
4929 | if (handle) { |
4930 | handle = ext4_journal_start_reserved(handle, | |
4931 | EXT4_HT_EXT_CONVERT); | |
4932 | if (IS_ERR(handle)) | |
4933 | return PTR_ERR(handle); | |
4934 | credits = 0; | |
4935 | } else { | |
4936 | /* | |
4937 | * credits to insert 1 extent into extent tree | |
4938 | */ | |
4939 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4940 | } | |
0031462b | 4941 | while (ret >= 0 && ret < max_blocks) { |
2ed88685 TT |
4942 | map.m_lblk += ret; |
4943 | map.m_len = (max_blocks -= ret); | |
6b523df4 JK |
4944 | if (credits) { |
4945 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
4946 | credits); | |
4947 | if (IS_ERR(handle)) { | |
4948 | ret = PTR_ERR(handle); | |
4949 | break; | |
4950 | } | |
0031462b | 4951 | } |
2ed88685 | 4952 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4953 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4954 | if (ret <= 0) |
4955 | ext4_warning(inode->i_sb, | |
4956 | "inode #%lu: block %u: len %u: " | |
4957 | "ext4_ext_map_blocks returned %d", | |
4958 | inode->i_ino, map.m_lblk, | |
4959 | map.m_len, ret); | |
0031462b | 4960 | ext4_mark_inode_dirty(handle, inode); |
6b523df4 JK |
4961 | if (credits) |
4962 | ret2 = ext4_journal_stop(handle); | |
4963 | if (ret <= 0 || ret2) | |
0031462b MC |
4964 | break; |
4965 | } | |
6b523df4 JK |
4966 | if (!credits) |
4967 | ret2 = ext4_journal_stop(handle); | |
0031462b MC |
4968 | return ret > 0 ? ret2 : ret; |
4969 | } | |
6d9c85eb | 4970 | |
6873fa0d | 4971 | /* |
69eb33dc ZL |
4972 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
4973 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
4974 | * return start of the next delayed extent. |
4975 | * | |
69eb33dc | 4976 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 4977 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 4978 | * extent found. Leave newes unmodified. |
6873fa0d | 4979 | */ |
91dd8c11 | 4980 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 4981 | struct extent_status *newes) |
6873fa0d | 4982 | { |
b3aff3e3 | 4983 | struct extent_status es; |
be401363 | 4984 | ext4_lblk_t block, next_del; |
6873fa0d | 4985 | |
69eb33dc | 4986 | if (newes->es_pblk == 0) { |
ad431025 EW |
4987 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
4988 | newes->es_lblk, | |
4989 | newes->es_lblk + newes->es_len - 1, | |
4990 | &es); | |
e30b5dca | 4991 | |
6d9c85eb | 4992 | /* |
69eb33dc | 4993 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 4994 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 4995 | */ |
06b0c886 | 4996 | if (es.es_len == 0) |
b3aff3e3 | 4997 | /* A hole found. */ |
91dd8c11 | 4998 | return 0; |
b3aff3e3 | 4999 | |
69eb33dc | 5000 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 5001 | /* A hole found. */ |
69eb33dc ZL |
5002 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
5003 | newes->es_len); | |
91dd8c11 | 5004 | return 0; |
6873fa0d | 5005 | } |
6d9c85eb | 5006 | |
69eb33dc | 5007 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
5008 | } |
5009 | ||
69eb33dc | 5010 | block = newes->es_lblk + newes->es_len; |
ad431025 EW |
5011 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block, |
5012 | EXT_MAX_BLOCKS, &es); | |
be401363 ZL |
5013 | if (es.es_len == 0) |
5014 | next_del = EXT_MAX_BLOCKS; | |
5015 | else | |
5016 | next_del = es.es_lblk; | |
5017 | ||
91dd8c11 | 5018 | return next_del; |
6873fa0d | 5019 | } |
6873fa0d ES |
5020 | /* fiemap flags we can handle specified here */ |
5021 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
5022 | ||
3a06d778 AK |
5023 | static int ext4_xattr_fiemap(struct inode *inode, |
5024 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
5025 | { |
5026 | __u64 physical = 0; | |
5027 | __u64 length; | |
5028 | __u32 flags = FIEMAP_EXTENT_LAST; | |
5029 | int blockbits = inode->i_sb->s_blocksize_bits; | |
5030 | int error = 0; | |
5031 | ||
5032 | /* in-inode? */ | |
19f5fb7a | 5033 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
5034 | struct ext4_iloc iloc; |
5035 | int offset; /* offset of xattr in inode */ | |
5036 | ||
5037 | error = ext4_get_inode_loc(inode, &iloc); | |
5038 | if (error) | |
5039 | return error; | |
a60697f4 | 5040 | physical = (__u64)iloc.bh->b_blocknr << blockbits; |
6873fa0d ES |
5041 | offset = EXT4_GOOD_OLD_INODE_SIZE + |
5042 | EXT4_I(inode)->i_extra_isize; | |
5043 | physical += offset; | |
5044 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
5045 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 5046 | brelse(iloc.bh); |
6873fa0d | 5047 | } else { /* external block */ |
a60697f4 | 5048 | physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
6873fa0d ES |
5049 | length = inode->i_sb->s_blocksize; |
5050 | } | |
5051 | ||
5052 | if (physical) | |
5053 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
5054 | length, flags); | |
5055 | return (error < 0 ? error : 0); | |
5056 | } | |
5057 | ||
5058 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
5059 | __u64 start, __u64 len) | |
5060 | { | |
5061 | ext4_lblk_t start_blk; | |
6873fa0d ES |
5062 | int error = 0; |
5063 | ||
94191985 TM |
5064 | if (ext4_has_inline_data(inode)) { |
5065 | int has_inline = 1; | |
5066 | ||
d952d69e DM |
5067 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline, |
5068 | start, len); | |
94191985 TM |
5069 | |
5070 | if (has_inline) | |
5071 | return error; | |
5072 | } | |
5073 | ||
7869a4a6 TT |
5074 | if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) { |
5075 | error = ext4_ext_precache(inode); | |
5076 | if (error) | |
5077 | return error; | |
5078 | } | |
5079 | ||
6873fa0d | 5080 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 5081 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
ad7fefb1 TT |
5082 | return generic_block_fiemap(inode, fieinfo, start, len, |
5083 | ext4_get_block); | |
6873fa0d ES |
5084 | |
5085 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
5086 | return -EBADR; | |
5087 | ||
5088 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
5089 | error = ext4_xattr_fiemap(inode, fieinfo); | |
5090 | } else { | |
aca92ff6 LM |
5091 | ext4_lblk_t len_blks; |
5092 | __u64 last_blk; | |
5093 | ||
6873fa0d | 5094 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 5095 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
5096 | if (last_blk >= EXT_MAX_BLOCKS) |
5097 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 5098 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
5099 | |
5100 | /* | |
91dd8c11 LC |
5101 | * Walk the extent tree gathering extent information |
5102 | * and pushing extents back to the user. | |
6873fa0d | 5103 | */ |
91dd8c11 LC |
5104 | error = ext4_fill_fiemap_extents(inode, start_blk, |
5105 | len_blks, fieinfo); | |
6873fa0d | 5106 | } |
6873fa0d ES |
5107 | return error; |
5108 | } | |
9eb79482 NJ |
5109 | |
5110 | /* | |
5111 | * ext4_access_path: | |
5112 | * Function to access the path buffer for marking it dirty. | |
5113 | * It also checks if there are sufficient credits left in the journal handle | |
5114 | * to update path. | |
5115 | */ | |
5116 | static int | |
5117 | ext4_access_path(handle_t *handle, struct inode *inode, | |
5118 | struct ext4_ext_path *path) | |
5119 | { | |
5120 | int credits, err; | |
5121 | ||
5122 | if (!ext4_handle_valid(handle)) | |
5123 | return 0; | |
5124 | ||
5125 | /* | |
5126 | * Check if need to extend journal credits | |
5127 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
5128 | * descriptor) for each block group; assume two block | |
5129 | * groups | |
5130 | */ | |
5131 | if (handle->h_buffer_credits < 7) { | |
5132 | credits = ext4_writepage_trans_blocks(inode); | |
5133 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); | |
5134 | /* EAGAIN is success */ | |
5135 | if (err && err != -EAGAIN) | |
5136 | return err; | |
5137 | } | |
5138 | ||
5139 | err = ext4_ext_get_access(handle, inode, path); | |
5140 | return err; | |
5141 | } | |
5142 | ||
5143 | /* | |
5144 | * ext4_ext_shift_path_extents: | |
5145 | * Shift the extents of a path structure lying between path[depth].p_ext | |
331573fe NJ |
5146 | * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells |
5147 | * if it is right shift or left shift operation. | |
9eb79482 NJ |
5148 | */ |
5149 | static int | |
5150 | ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift, | |
5151 | struct inode *inode, handle_t *handle, | |
331573fe | 5152 | enum SHIFT_DIRECTION SHIFT) |
9eb79482 NJ |
5153 | { |
5154 | int depth, err = 0; | |
5155 | struct ext4_extent *ex_start, *ex_last; | |
5156 | bool update = 0; | |
5157 | depth = path->p_depth; | |
5158 | ||
5159 | while (depth >= 0) { | |
5160 | if (depth == path->p_depth) { | |
5161 | ex_start = path[depth].p_ext; | |
5162 | if (!ex_start) | |
6a797d27 | 5163 | return -EFSCORRUPTED; |
9eb79482 NJ |
5164 | |
5165 | ex_last = EXT_LAST_EXTENT(path[depth].p_hdr); | |
9eb79482 NJ |
5166 | |
5167 | err = ext4_access_path(handle, inode, path + depth); | |
5168 | if (err) | |
5169 | goto out; | |
5170 | ||
5171 | if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
5172 | update = 1; | |
5173 | ||
9eb79482 | 5174 | while (ex_start <= ex_last) { |
331573fe NJ |
5175 | if (SHIFT == SHIFT_LEFT) { |
5176 | le32_add_cpu(&ex_start->ee_block, | |
5177 | -shift); | |
5178 | /* Try to merge to the left. */ | |
5179 | if ((ex_start > | |
5180 | EXT_FIRST_EXTENT(path[depth].p_hdr)) | |
5181 | && | |
5182 | ext4_ext_try_to_merge_right(inode, | |
5183 | path, ex_start - 1)) | |
5184 | ex_last--; | |
5185 | else | |
5186 | ex_start++; | |
5187 | } else { | |
5188 | le32_add_cpu(&ex_last->ee_block, shift); | |
5189 | ext4_ext_try_to_merge_right(inode, path, | |
5190 | ex_last); | |
6dd834ef | 5191 | ex_last--; |
331573fe | 5192 | } |
9eb79482 NJ |
5193 | } |
5194 | err = ext4_ext_dirty(handle, inode, path + depth); | |
5195 | if (err) | |
5196 | goto out; | |
5197 | ||
5198 | if (--depth < 0 || !update) | |
5199 | break; | |
5200 | } | |
5201 | ||
5202 | /* Update index too */ | |
5203 | err = ext4_access_path(handle, inode, path + depth); | |
5204 | if (err) | |
5205 | goto out; | |
5206 | ||
331573fe NJ |
5207 | if (SHIFT == SHIFT_LEFT) |
5208 | le32_add_cpu(&path[depth].p_idx->ei_block, -shift); | |
5209 | else | |
5210 | le32_add_cpu(&path[depth].p_idx->ei_block, shift); | |
9eb79482 NJ |
5211 | err = ext4_ext_dirty(handle, inode, path + depth); |
5212 | if (err) | |
5213 | goto out; | |
5214 | ||
5215 | /* we are done if current index is not a starting index */ | |
5216 | if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr)) | |
5217 | break; | |
5218 | ||
5219 | depth--; | |
5220 | } | |
5221 | ||
5222 | out: | |
5223 | return err; | |
5224 | } | |
5225 | ||
5226 | /* | |
5227 | * ext4_ext_shift_extents: | |
331573fe NJ |
5228 | * All the extents which lies in the range from @start to the last allocated |
5229 | * block for the @inode are shifted either towards left or right (depending | |
5230 | * upon @SHIFT) by @shift blocks. | |
9eb79482 NJ |
5231 | * On success, 0 is returned, error otherwise. |
5232 | */ | |
5233 | static int | |
5234 | ext4_ext_shift_extents(struct inode *inode, handle_t *handle, | |
331573fe NJ |
5235 | ext4_lblk_t start, ext4_lblk_t shift, |
5236 | enum SHIFT_DIRECTION SHIFT) | |
9eb79482 NJ |
5237 | { |
5238 | struct ext4_ext_path *path; | |
5239 | int ret = 0, depth; | |
5240 | struct ext4_extent *extent; | |
331573fe | 5241 | ext4_lblk_t stop, *iterator, ex_start, ex_end; |
9eb79482 NJ |
5242 | |
5243 | /* Let path point to the last extent */ | |
03e916fa RP |
5244 | path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, |
5245 | EXT4_EX_NOCACHE); | |
9eb79482 NJ |
5246 | if (IS_ERR(path)) |
5247 | return PTR_ERR(path); | |
5248 | ||
5249 | depth = path->p_depth; | |
5250 | extent = path[depth].p_ext; | |
ee4bd0d9 TT |
5251 | if (!extent) |
5252 | goto out; | |
9eb79482 | 5253 | |
2a9b8cba | 5254 | stop = le32_to_cpu(extent->ee_block); |
9eb79482 | 5255 | |
331573fe | 5256 | /* |
349fa7d6 EB |
5257 | * For left shifts, make sure the hole on the left is big enough to |
5258 | * accommodate the shift. For right shifts, make sure the last extent | |
5259 | * won't be shifted beyond EXT_MAX_BLOCKS. | |
331573fe NJ |
5260 | */ |
5261 | if (SHIFT == SHIFT_LEFT) { | |
03e916fa RP |
5262 | path = ext4_find_extent(inode, start - 1, &path, |
5263 | EXT4_EX_NOCACHE); | |
331573fe NJ |
5264 | if (IS_ERR(path)) |
5265 | return PTR_ERR(path); | |
5266 | depth = path->p_depth; | |
5267 | extent = path[depth].p_ext; | |
5268 | if (extent) { | |
5269 | ex_start = le32_to_cpu(extent->ee_block); | |
5270 | ex_end = le32_to_cpu(extent->ee_block) + | |
5271 | ext4_ext_get_actual_len(extent); | |
5272 | } else { | |
5273 | ex_start = 0; | |
5274 | ex_end = 0; | |
5275 | } | |
9eb79482 | 5276 | |
331573fe NJ |
5277 | if ((start == ex_start && shift > ex_start) || |
5278 | (shift > start - ex_end)) { | |
349fa7d6 EB |
5279 | ret = -EINVAL; |
5280 | goto out; | |
5281 | } | |
5282 | } else { | |
5283 | if (shift > EXT_MAX_BLOCKS - | |
5284 | (stop + ext4_ext_get_actual_len(extent))) { | |
5285 | ret = -EINVAL; | |
5286 | goto out; | |
331573fe | 5287 | } |
8dc79ec4 | 5288 | } |
9eb79482 | 5289 | |
331573fe NJ |
5290 | /* |
5291 | * In case of left shift, iterator points to start and it is increased | |
5292 | * till we reach stop. In case of right shift, iterator points to stop | |
5293 | * and it is decreased till we reach start. | |
5294 | */ | |
5295 | if (SHIFT == SHIFT_LEFT) | |
5296 | iterator = &start; | |
5297 | else | |
5298 | iterator = &stop; | |
9eb79482 | 5299 | |
2a9b8cba RP |
5300 | /* |
5301 | * Its safe to start updating extents. Start and stop are unsigned, so | |
5302 | * in case of right shift if extent with 0 block is reached, iterator | |
5303 | * becomes NULL to indicate the end of the loop. | |
5304 | */ | |
5305 | while (iterator && start <= stop) { | |
03e916fa RP |
5306 | path = ext4_find_extent(inode, *iterator, &path, |
5307 | EXT4_EX_NOCACHE); | |
9eb79482 NJ |
5308 | if (IS_ERR(path)) |
5309 | return PTR_ERR(path); | |
5310 | depth = path->p_depth; | |
5311 | extent = path[depth].p_ext; | |
a18ed359 DM |
5312 | if (!extent) { |
5313 | EXT4_ERROR_INODE(inode, "unexpected hole at %lu", | |
331573fe | 5314 | (unsigned long) *iterator); |
6a797d27 | 5315 | return -EFSCORRUPTED; |
a18ed359 | 5316 | } |
331573fe NJ |
5317 | if (SHIFT == SHIFT_LEFT && *iterator > |
5318 | le32_to_cpu(extent->ee_block)) { | |
9eb79482 | 5319 | /* Hole, move to the next extent */ |
f8fb4f41 DM |
5320 | if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) { |
5321 | path[depth].p_ext++; | |
5322 | } else { | |
331573fe | 5323 | *iterator = ext4_ext_next_allocated_block(path); |
f8fb4f41 | 5324 | continue; |
9eb79482 NJ |
5325 | } |
5326 | } | |
331573fe NJ |
5327 | |
5328 | if (SHIFT == SHIFT_LEFT) { | |
5329 | extent = EXT_LAST_EXTENT(path[depth].p_hdr); | |
5330 | *iterator = le32_to_cpu(extent->ee_block) + | |
5331 | ext4_ext_get_actual_len(extent); | |
5332 | } else { | |
5333 | extent = EXT_FIRST_EXTENT(path[depth].p_hdr); | |
2a9b8cba RP |
5334 | if (le32_to_cpu(extent->ee_block) > 0) |
5335 | *iterator = le32_to_cpu(extent->ee_block) - 1; | |
5336 | else | |
5337 | /* Beginning is reached, end of the loop */ | |
5338 | iterator = NULL; | |
331573fe NJ |
5339 | /* Update path extent in case we need to stop */ |
5340 | while (le32_to_cpu(extent->ee_block) < start) | |
5341 | extent++; | |
5342 | path[depth].p_ext = extent; | |
5343 | } | |
9eb79482 | 5344 | ret = ext4_ext_shift_path_extents(path, shift, inode, |
331573fe | 5345 | handle, SHIFT); |
9eb79482 NJ |
5346 | if (ret) |
5347 | break; | |
5348 | } | |
ee4bd0d9 TT |
5349 | out: |
5350 | ext4_ext_drop_refs(path); | |
5351 | kfree(path); | |
9eb79482 NJ |
5352 | return ret; |
5353 | } | |
5354 | ||
5355 | /* | |
5356 | * ext4_collapse_range: | |
5357 | * This implements the fallocate's collapse range functionality for ext4 | |
5358 | * Returns: 0 and non-zero on error. | |
5359 | */ | |
5360 | int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len) | |
5361 | { | |
5362 | struct super_block *sb = inode->i_sb; | |
5363 | ext4_lblk_t punch_start, punch_stop; | |
5364 | handle_t *handle; | |
5365 | unsigned int credits; | |
a8680e0d | 5366 | loff_t new_size, ioffset; |
9eb79482 NJ |
5367 | int ret; |
5368 | ||
b9576fc3 TT |
5369 | /* |
5370 | * We need to test this early because xfstests assumes that a | |
5371 | * collapse range of (0, 1) will return EOPNOTSUPP if the file | |
5372 | * system does not support collapse range. | |
5373 | */ | |
5374 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
5375 | return -EOPNOTSUPP; | |
5376 | ||
9eb79482 | 5377 | /* Collapse range works only on fs block size aligned offsets. */ |
ee98fa3a NJ |
5378 | if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) || |
5379 | len & (EXT4_CLUSTER_SIZE(sb) - 1)) | |
9eb79482 NJ |
5380 | return -EINVAL; |
5381 | ||
5382 | if (!S_ISREG(inode->i_mode)) | |
86f1ca38 | 5383 | return -EINVAL; |
9eb79482 NJ |
5384 | |
5385 | trace_ext4_collapse_range(inode, offset, len); | |
5386 | ||
5387 | punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5388 | punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb); | |
5389 | ||
1ce01c4a NJ |
5390 | /* Call ext4_force_commit to flush all data in case of data=journal. */ |
5391 | if (ext4_should_journal_data(inode)) { | |
5392 | ret = ext4_force_commit(inode->i_sb); | |
5393 | if (ret) | |
5394 | return ret; | |
5395 | } | |
5396 | ||
5955102c | 5397 | inode_lock(inode); |
23fffa92 LC |
5398 | /* |
5399 | * There is no need to overlap collapse range with EOF, in which case | |
5400 | * it is effectively a truncate operation | |
5401 | */ | |
5402 | if (offset + len >= i_size_read(inode)) { | |
5403 | ret = -EINVAL; | |
5404 | goto out_mutex; | |
5405 | } | |
5406 | ||
9eb79482 NJ |
5407 | /* Currently just for extent based files */ |
5408 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5409 | ret = -EOPNOTSUPP; | |
5410 | goto out_mutex; | |
5411 | } | |
5412 | ||
9eb79482 | 5413 | /* Wait for existing dio to complete */ |
9eb79482 NJ |
5414 | inode_dio_wait(inode); |
5415 | ||
ea3d7209 JK |
5416 | /* |
5417 | * Prevent page faults from reinstantiating pages we have released from | |
5418 | * page cache. | |
5419 | */ | |
5420 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
5421 | |
5422 | ret = ext4_break_layouts(inode); | |
5423 | if (ret) | |
5424 | goto out_mmap; | |
5425 | ||
32ebffd3 JK |
5426 | /* |
5427 | * Need to round down offset to be aligned with page size boundary | |
5428 | * for page size > block size. | |
5429 | */ | |
5430 | ioffset = round_down(offset, PAGE_SIZE); | |
5431 | /* | |
5432 | * Write tail of the last page before removed range since it will get | |
5433 | * removed from the page cache below. | |
5434 | */ | |
5435 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset); | |
5436 | if (ret) | |
5437 | goto out_mmap; | |
5438 | /* | |
5439 | * Write data that will be shifted to preserve them when discarding | |
5440 | * page cache below. We are also protected from pages becoming dirty | |
5441 | * by i_mmap_sem. | |
5442 | */ | |
5443 | ret = filemap_write_and_wait_range(inode->i_mapping, offset + len, | |
5444 | LLONG_MAX); | |
5445 | if (ret) | |
5446 | goto out_mmap; | |
ea3d7209 JK |
5447 | truncate_pagecache(inode, ioffset); |
5448 | ||
9eb79482 NJ |
5449 | credits = ext4_writepage_trans_blocks(inode); |
5450 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5451 | if (IS_ERR(handle)) { | |
5452 | ret = PTR_ERR(handle); | |
ea3d7209 | 5453 | goto out_mmap; |
9eb79482 NJ |
5454 | } |
5455 | ||
5456 | down_write(&EXT4_I(inode)->i_data_sem); | |
5457 | ext4_discard_preallocations(inode); | |
5458 | ||
5459 | ret = ext4_es_remove_extent(inode, punch_start, | |
2c1d2328 | 5460 | EXT_MAX_BLOCKS - punch_start); |
9eb79482 NJ |
5461 | if (ret) { |
5462 | up_write(&EXT4_I(inode)->i_data_sem); | |
5463 | goto out_stop; | |
5464 | } | |
5465 | ||
5466 | ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1); | |
5467 | if (ret) { | |
5468 | up_write(&EXT4_I(inode)->i_data_sem); | |
5469 | goto out_stop; | |
5470 | } | |
ef24f6c2 | 5471 | ext4_discard_preallocations(inode); |
9eb79482 NJ |
5472 | |
5473 | ret = ext4_ext_shift_extents(inode, handle, punch_stop, | |
331573fe | 5474 | punch_stop - punch_start, SHIFT_LEFT); |
9eb79482 NJ |
5475 | if (ret) { |
5476 | up_write(&EXT4_I(inode)->i_data_sem); | |
5477 | goto out_stop; | |
5478 | } | |
5479 | ||
5480 | new_size = i_size_read(inode) - len; | |
9337d5d3 | 5481 | i_size_write(inode, new_size); |
9eb79482 NJ |
5482 | EXT4_I(inode)->i_disksize = new_size; |
5483 | ||
9eb79482 NJ |
5484 | up_write(&EXT4_I(inode)->i_data_sem); |
5485 | if (IS_SYNC(inode)) | |
5486 | ext4_handle_sync(handle); | |
eeca7ea1 | 5487 | inode->i_mtime = inode->i_ctime = current_time(inode); |
9eb79482 | 5488 | ext4_mark_inode_dirty(handle, inode); |
67a7d5f5 | 5489 | ext4_update_inode_fsync_trans(handle, inode, 1); |
9eb79482 NJ |
5490 | |
5491 | out_stop: | |
5492 | ext4_journal_stop(handle); | |
ea3d7209 JK |
5493 | out_mmap: |
5494 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
9eb79482 | 5495 | out_mutex: |
5955102c | 5496 | inode_unlock(inode); |
9eb79482 NJ |
5497 | return ret; |
5498 | } | |
fcf6b1b7 | 5499 | |
331573fe NJ |
5500 | /* |
5501 | * ext4_insert_range: | |
5502 | * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate. | |
5503 | * The data blocks starting from @offset to the EOF are shifted by @len | |
5504 | * towards right to create a hole in the @inode. Inode size is increased | |
5505 | * by len bytes. | |
5506 | * Returns 0 on success, error otherwise. | |
5507 | */ | |
5508 | int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len) | |
5509 | { | |
5510 | struct super_block *sb = inode->i_sb; | |
5511 | handle_t *handle; | |
5512 | struct ext4_ext_path *path; | |
5513 | struct ext4_extent *extent; | |
5514 | ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0; | |
5515 | unsigned int credits, ee_len; | |
5516 | int ret = 0, depth, split_flag = 0; | |
5517 | loff_t ioffset; | |
5518 | ||
5519 | /* | |
5520 | * We need to test this early because xfstests assumes that an | |
5521 | * insert range of (0, 1) will return EOPNOTSUPP if the file | |
5522 | * system does not support insert range. | |
5523 | */ | |
5524 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
5525 | return -EOPNOTSUPP; | |
5526 | ||
5527 | /* Insert range works only on fs block size aligned offsets. */ | |
5528 | if (offset & (EXT4_CLUSTER_SIZE(sb) - 1) || | |
5529 | len & (EXT4_CLUSTER_SIZE(sb) - 1)) | |
5530 | return -EINVAL; | |
5531 | ||
5532 | if (!S_ISREG(inode->i_mode)) | |
5533 | return -EOPNOTSUPP; | |
5534 | ||
5535 | trace_ext4_insert_range(inode, offset, len); | |
5536 | ||
5537 | offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb); | |
5538 | len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb); | |
5539 | ||
5540 | /* Call ext4_force_commit to flush all data in case of data=journal */ | |
5541 | if (ext4_should_journal_data(inode)) { | |
5542 | ret = ext4_force_commit(inode->i_sb); | |
5543 | if (ret) | |
5544 | return ret; | |
5545 | } | |
5546 | ||
5955102c | 5547 | inode_lock(inode); |
331573fe NJ |
5548 | /* Currently just for extent based files */ |
5549 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
5550 | ret = -EOPNOTSUPP; | |
5551 | goto out_mutex; | |
5552 | } | |
5553 | ||
5554 | /* Check for wrap through zero */ | |
5555 | if (inode->i_size + len > inode->i_sb->s_maxbytes) { | |
5556 | ret = -EFBIG; | |
5557 | goto out_mutex; | |
5558 | } | |
5559 | ||
5560 | /* Offset should be less than i_size */ | |
5561 | if (offset >= i_size_read(inode)) { | |
5562 | ret = -EINVAL; | |
5563 | goto out_mutex; | |
5564 | } | |
5565 | ||
331573fe | 5566 | /* Wait for existing dio to complete */ |
331573fe NJ |
5567 | inode_dio_wait(inode); |
5568 | ||
ea3d7209 JK |
5569 | /* |
5570 | * Prevent page faults from reinstantiating pages we have released from | |
5571 | * page cache. | |
5572 | */ | |
5573 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
5574 | |
5575 | ret = ext4_break_layouts(inode); | |
5576 | if (ret) | |
5577 | goto out_mmap; | |
5578 | ||
32ebffd3 JK |
5579 | /* |
5580 | * Need to round down to align start offset to page size boundary | |
5581 | * for page size > block size. | |
5582 | */ | |
5583 | ioffset = round_down(offset, PAGE_SIZE); | |
5584 | /* Write out all dirty pages */ | |
5585 | ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, | |
5586 | LLONG_MAX); | |
5587 | if (ret) | |
5588 | goto out_mmap; | |
ea3d7209 JK |
5589 | truncate_pagecache(inode, ioffset); |
5590 | ||
331573fe NJ |
5591 | credits = ext4_writepage_trans_blocks(inode); |
5592 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
5593 | if (IS_ERR(handle)) { | |
5594 | ret = PTR_ERR(handle); | |
ea3d7209 | 5595 | goto out_mmap; |
331573fe NJ |
5596 | } |
5597 | ||
5598 | /* Expand file to avoid data loss if there is error while shifting */ | |
5599 | inode->i_size += len; | |
5600 | EXT4_I(inode)->i_disksize += len; | |
eeca7ea1 | 5601 | inode->i_mtime = inode->i_ctime = current_time(inode); |
331573fe NJ |
5602 | ret = ext4_mark_inode_dirty(handle, inode); |
5603 | if (ret) | |
5604 | goto out_stop; | |
5605 | ||
5606 | down_write(&EXT4_I(inode)->i_data_sem); | |
5607 | ext4_discard_preallocations(inode); | |
5608 | ||
5609 | path = ext4_find_extent(inode, offset_lblk, NULL, 0); | |
5610 | if (IS_ERR(path)) { | |
5611 | up_write(&EXT4_I(inode)->i_data_sem); | |
5612 | goto out_stop; | |
5613 | } | |
5614 | ||
5615 | depth = ext_depth(inode); | |
5616 | extent = path[depth].p_ext; | |
5617 | if (extent) { | |
5618 | ee_start_lblk = le32_to_cpu(extent->ee_block); | |
5619 | ee_len = ext4_ext_get_actual_len(extent); | |
5620 | ||
5621 | /* | |
5622 | * If offset_lblk is not the starting block of extent, split | |
5623 | * the extent @offset_lblk | |
5624 | */ | |
5625 | if ((offset_lblk > ee_start_lblk) && | |
5626 | (offset_lblk < (ee_start_lblk + ee_len))) { | |
5627 | if (ext4_ext_is_unwritten(extent)) | |
5628 | split_flag = EXT4_EXT_MARK_UNWRIT1 | | |
5629 | EXT4_EXT_MARK_UNWRIT2; | |
5630 | ret = ext4_split_extent_at(handle, inode, &path, | |
5631 | offset_lblk, split_flag, | |
5632 | EXT4_EX_NOCACHE | | |
5633 | EXT4_GET_BLOCKS_PRE_IO | | |
5634 | EXT4_GET_BLOCKS_METADATA_NOFAIL); | |
5635 | } | |
5636 | ||
5637 | ext4_ext_drop_refs(path); | |
5638 | kfree(path); | |
5639 | if (ret < 0) { | |
5640 | up_write(&EXT4_I(inode)->i_data_sem); | |
5641 | goto out_stop; | |
5642 | } | |
edf15aa1 FF |
5643 | } else { |
5644 | ext4_ext_drop_refs(path); | |
5645 | kfree(path); | |
331573fe NJ |
5646 | } |
5647 | ||
5648 | ret = ext4_es_remove_extent(inode, offset_lblk, | |
5649 | EXT_MAX_BLOCKS - offset_lblk); | |
5650 | if (ret) { | |
5651 | up_write(&EXT4_I(inode)->i_data_sem); | |
5652 | goto out_stop; | |
5653 | } | |
5654 | ||
5655 | /* | |
5656 | * if offset_lblk lies in a hole which is at start of file, use | |
5657 | * ee_start_lblk to shift extents | |
5658 | */ | |
5659 | ret = ext4_ext_shift_extents(inode, handle, | |
5660 | ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk, | |
5661 | len_lblk, SHIFT_RIGHT); | |
5662 | ||
5663 | up_write(&EXT4_I(inode)->i_data_sem); | |
5664 | if (IS_SYNC(inode)) | |
5665 | ext4_handle_sync(handle); | |
67a7d5f5 JK |
5666 | if (ret >= 0) |
5667 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
331573fe NJ |
5668 | |
5669 | out_stop: | |
5670 | ext4_journal_stop(handle); | |
ea3d7209 JK |
5671 | out_mmap: |
5672 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
331573fe | 5673 | out_mutex: |
5955102c | 5674 | inode_unlock(inode); |
331573fe NJ |
5675 | return ret; |
5676 | } | |
5677 | ||
fcf6b1b7 | 5678 | /** |
c60990b3 TT |
5679 | * ext4_swap_extents() - Swap extents between two inodes |
5680 | * @handle: handle for this transaction | |
fcf6b1b7 DM |
5681 | * @inode1: First inode |
5682 | * @inode2: Second inode | |
5683 | * @lblk1: Start block for first inode | |
5684 | * @lblk2: Start block for second inode | |
5685 | * @count: Number of blocks to swap | |
dcae058a | 5686 | * @unwritten: Mark second inode's extents as unwritten after swap |
fcf6b1b7 DM |
5687 | * @erp: Pointer to save error value |
5688 | * | |
5689 | * This helper routine does exactly what is promise "swap extents". All other | |
5690 | * stuff such as page-cache locking consistency, bh mapping consistency or | |
5691 | * extent's data copying must be performed by caller. | |
5692 | * Locking: | |
5693 | * i_mutex is held for both inodes | |
5694 | * i_data_sem is locked for write for both inodes | |
5695 | * Assumptions: | |
5696 | * All pages from requested range are locked for both inodes | |
5697 | */ | |
5698 | int | |
5699 | ext4_swap_extents(handle_t *handle, struct inode *inode1, | |
dcae058a | 5700 | struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2, |
fcf6b1b7 DM |
5701 | ext4_lblk_t count, int unwritten, int *erp) |
5702 | { | |
5703 | struct ext4_ext_path *path1 = NULL; | |
5704 | struct ext4_ext_path *path2 = NULL; | |
5705 | int replaced_count = 0; | |
5706 | ||
5707 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem)); | |
5708 | BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem)); | |
5955102c AV |
5709 | BUG_ON(!inode_is_locked(inode1)); |
5710 | BUG_ON(!inode_is_locked(inode2)); | |
fcf6b1b7 DM |
5711 | |
5712 | *erp = ext4_es_remove_extent(inode1, lblk1, count); | |
19008f6d | 5713 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5714 | return 0; |
5715 | *erp = ext4_es_remove_extent(inode2, lblk2, count); | |
19008f6d | 5716 | if (unlikely(*erp)) |
fcf6b1b7 DM |
5717 | return 0; |
5718 | ||
5719 | while (count) { | |
5720 | struct ext4_extent *ex1, *ex2, tmp_ex; | |
5721 | ext4_lblk_t e1_blk, e2_blk; | |
5722 | int e1_len, e2_len, len; | |
5723 | int split = 0; | |
5724 | ||
ed8a1a76 | 5725 | path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE); |
a1c83681 | 5726 | if (IS_ERR(path1)) { |
fcf6b1b7 | 5727 | *erp = PTR_ERR(path1); |
19008f6d TT |
5728 | path1 = NULL; |
5729 | finish: | |
5730 | count = 0; | |
5731 | goto repeat; | |
fcf6b1b7 | 5732 | } |
ed8a1a76 | 5733 | path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE); |
a1c83681 | 5734 | if (IS_ERR(path2)) { |
fcf6b1b7 | 5735 | *erp = PTR_ERR(path2); |
19008f6d TT |
5736 | path2 = NULL; |
5737 | goto finish; | |
fcf6b1b7 DM |
5738 | } |
5739 | ex1 = path1[path1->p_depth].p_ext; | |
5740 | ex2 = path2[path2->p_depth].p_ext; | |
5741 | /* Do we have somthing to swap ? */ | |
5742 | if (unlikely(!ex2 || !ex1)) | |
19008f6d | 5743 | goto finish; |
fcf6b1b7 DM |
5744 | |
5745 | e1_blk = le32_to_cpu(ex1->ee_block); | |
5746 | e2_blk = le32_to_cpu(ex2->ee_block); | |
5747 | e1_len = ext4_ext_get_actual_len(ex1); | |
5748 | e2_len = ext4_ext_get_actual_len(ex2); | |
5749 | ||
5750 | /* Hole handling */ | |
5751 | if (!in_range(lblk1, e1_blk, e1_len) || | |
5752 | !in_range(lblk2, e2_blk, e2_len)) { | |
5753 | ext4_lblk_t next1, next2; | |
5754 | ||
5755 | /* if hole after extent, then go to next extent */ | |
5756 | next1 = ext4_ext_next_allocated_block(path1); | |
5757 | next2 = ext4_ext_next_allocated_block(path2); | |
5758 | /* If hole before extent, then shift to that extent */ | |
5759 | if (e1_blk > lblk1) | |
5760 | next1 = e1_blk; | |
5761 | if (e2_blk > lblk2) | |
4e562013 | 5762 | next2 = e2_blk; |
fcf6b1b7 DM |
5763 | /* Do we have something to swap */ |
5764 | if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS) | |
19008f6d | 5765 | goto finish; |
fcf6b1b7 DM |
5766 | /* Move to the rightest boundary */ |
5767 | len = next1 - lblk1; | |
5768 | if (len < next2 - lblk2) | |
5769 | len = next2 - lblk2; | |
5770 | if (len > count) | |
5771 | len = count; | |
5772 | lblk1 += len; | |
5773 | lblk2 += len; | |
5774 | count -= len; | |
5775 | goto repeat; | |
5776 | } | |
5777 | ||
5778 | /* Prepare left boundary */ | |
5779 | if (e1_blk < lblk1) { | |
5780 | split = 1; | |
5781 | *erp = ext4_force_split_extent_at(handle, inode1, | |
dfe50809 | 5782 | &path1, lblk1, 0); |
19008f6d TT |
5783 | if (unlikely(*erp)) |
5784 | goto finish; | |
fcf6b1b7 DM |
5785 | } |
5786 | if (e2_blk < lblk2) { | |
5787 | split = 1; | |
5788 | *erp = ext4_force_split_extent_at(handle, inode2, | |
dfe50809 | 5789 | &path2, lblk2, 0); |
19008f6d TT |
5790 | if (unlikely(*erp)) |
5791 | goto finish; | |
fcf6b1b7 | 5792 | } |
dfe50809 | 5793 | /* ext4_split_extent_at() may result in leaf extent split, |
fcf6b1b7 DM |
5794 | * path must to be revalidated. */ |
5795 | if (split) | |
5796 | goto repeat; | |
5797 | ||
5798 | /* Prepare right boundary */ | |
5799 | len = count; | |
5800 | if (len > e1_blk + e1_len - lblk1) | |
5801 | len = e1_blk + e1_len - lblk1; | |
5802 | if (len > e2_blk + e2_len - lblk2) | |
5803 | len = e2_blk + e2_len - lblk2; | |
5804 | ||
5805 | if (len != e1_len) { | |
5806 | split = 1; | |
5807 | *erp = ext4_force_split_extent_at(handle, inode1, | |
dfe50809 | 5808 | &path1, lblk1 + len, 0); |
19008f6d TT |
5809 | if (unlikely(*erp)) |
5810 | goto finish; | |
fcf6b1b7 DM |
5811 | } |
5812 | if (len != e2_len) { | |
5813 | split = 1; | |
5814 | *erp = ext4_force_split_extent_at(handle, inode2, | |
dfe50809 | 5815 | &path2, lblk2 + len, 0); |
fcf6b1b7 | 5816 | if (*erp) |
19008f6d | 5817 | goto finish; |
fcf6b1b7 | 5818 | } |
dfe50809 | 5819 | /* ext4_split_extent_at() may result in leaf extent split, |
fcf6b1b7 DM |
5820 | * path must to be revalidated. */ |
5821 | if (split) | |
5822 | goto repeat; | |
5823 | ||
5824 | BUG_ON(e2_len != e1_len); | |
5825 | *erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth); | |
19008f6d TT |
5826 | if (unlikely(*erp)) |
5827 | goto finish; | |
fcf6b1b7 | 5828 | *erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth); |
19008f6d TT |
5829 | if (unlikely(*erp)) |
5830 | goto finish; | |
fcf6b1b7 DM |
5831 | |
5832 | /* Both extents are fully inside boundaries. Swap it now */ | |
5833 | tmp_ex = *ex1; | |
5834 | ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2)); | |
5835 | ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex)); | |
5836 | ex1->ee_len = cpu_to_le16(e2_len); | |
5837 | ex2->ee_len = cpu_to_le16(e1_len); | |
5838 | if (unwritten) | |
5839 | ext4_ext_mark_unwritten(ex2); | |
5840 | if (ext4_ext_is_unwritten(&tmp_ex)) | |
5841 | ext4_ext_mark_unwritten(ex1); | |
5842 | ||
5843 | ext4_ext_try_to_merge(handle, inode2, path2, ex2); | |
5844 | ext4_ext_try_to_merge(handle, inode1, path1, ex1); | |
5845 | *erp = ext4_ext_dirty(handle, inode2, path2 + | |
5846 | path2->p_depth); | |
19008f6d TT |
5847 | if (unlikely(*erp)) |
5848 | goto finish; | |
fcf6b1b7 DM |
5849 | *erp = ext4_ext_dirty(handle, inode1, path1 + |
5850 | path1->p_depth); | |
5851 | /* | |
5852 | * Looks scarry ah..? second inode already points to new blocks, | |
5853 | * and it was successfully dirtied. But luckily error may happen | |
5854 | * only due to journal error, so full transaction will be | |
5855 | * aborted anyway. | |
5856 | */ | |
19008f6d TT |
5857 | if (unlikely(*erp)) |
5858 | goto finish; | |
fcf6b1b7 DM |
5859 | lblk1 += len; |
5860 | lblk2 += len; | |
5861 | replaced_count += len; | |
5862 | count -= len; | |
5863 | ||
5864 | repeat: | |
b7ea89ad TT |
5865 | ext4_ext_drop_refs(path1); |
5866 | kfree(path1); | |
5867 | ext4_ext_drop_refs(path2); | |
5868 | kfree(path2); | |
5869 | path1 = path2 = NULL; | |
fcf6b1b7 | 5870 | } |
fcf6b1b7 DM |
5871 | return replaced_count; |
5872 | } | |
0b02f4c0 EW |
5873 | |
5874 | /* | |
5875 | * ext4_clu_mapped - determine whether any block in a logical cluster has | |
5876 | * been mapped to a physical cluster | |
5877 | * | |
5878 | * @inode - file containing the logical cluster | |
5879 | * @lclu - logical cluster of interest | |
5880 | * | |
5881 | * Returns 1 if any block in the logical cluster is mapped, signifying | |
5882 | * that a physical cluster has been allocated for it. Otherwise, | |
5883 | * returns 0. Can also return negative error codes. Derived from | |
5884 | * ext4_ext_map_blocks(). | |
5885 | */ | |
5886 | int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu) | |
5887 | { | |
5888 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
5889 | struct ext4_ext_path *path; | |
5890 | int depth, mapped = 0, err = 0; | |
5891 | struct ext4_extent *extent; | |
5892 | ext4_lblk_t first_lblk, first_lclu, last_lclu; | |
5893 | ||
5894 | /* search for the extent closest to the first block in the cluster */ | |
5895 | path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0); | |
5896 | if (IS_ERR(path)) { | |
5897 | err = PTR_ERR(path); | |
5898 | path = NULL; | |
5899 | goto out; | |
5900 | } | |
5901 | ||
5902 | depth = ext_depth(inode); | |
5903 | ||
5904 | /* | |
5905 | * A consistent leaf must not be empty. This situation is possible, | |
5906 | * though, _during_ tree modification, and it's why an assert can't | |
5907 | * be put in ext4_find_extent(). | |
5908 | */ | |
5909 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { | |
5910 | EXT4_ERROR_INODE(inode, | |
5911 | "bad extent address - lblock: %lu, depth: %d, pblock: %lld", | |
5912 | (unsigned long) EXT4_C2B(sbi, lclu), | |
5913 | depth, path[depth].p_block); | |
5914 | err = -EFSCORRUPTED; | |
5915 | goto out; | |
5916 | } | |
5917 | ||
5918 | extent = path[depth].p_ext; | |
5919 | ||
5920 | /* can't be mapped if the extent tree is empty */ | |
5921 | if (extent == NULL) | |
5922 | goto out; | |
5923 | ||
5924 | first_lblk = le32_to_cpu(extent->ee_block); | |
5925 | first_lclu = EXT4_B2C(sbi, first_lblk); | |
5926 | ||
5927 | /* | |
5928 | * Three possible outcomes at this point - found extent spanning | |
5929 | * the target cluster, to the left of the target cluster, or to the | |
5930 | * right of the target cluster. The first two cases are handled here. | |
5931 | * The last case indicates the target cluster is not mapped. | |
5932 | */ | |
5933 | if (lclu >= first_lclu) { | |
5934 | last_lclu = EXT4_B2C(sbi, first_lblk + | |
5935 | ext4_ext_get_actual_len(extent) - 1); | |
5936 | if (lclu <= last_lclu) { | |
5937 | mapped = 1; | |
5938 | } else { | |
5939 | first_lblk = ext4_ext_next_allocated_block(path); | |
5940 | first_lclu = EXT4_B2C(sbi, first_lblk); | |
5941 | if (lclu == first_lclu) | |
5942 | mapped = 1; | |
5943 | } | |
5944 | } | |
5945 | ||
5946 | out: | |
5947 | ext4_ext_drop_refs(path); | |
5948 | kfree(path); | |
5949 | ||
5950 | return err ? err : mapped; | |
5951 | } |