]>
Commit | Line | Data |
---|---|---|
a86c6181 AT |
1 | /* |
2 | * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com | |
3 | * Written by Alex Tomas <alex@clusterfs.com> | |
4 | * | |
5 | * Architecture independence: | |
6 | * Copyright (c) 2005, Bull S.A. | |
7 | * Written by Pierre Peiffer <pierre.peiffer@bull.net> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License version 2 as | |
11 | * published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public Licens | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
21 | */ | |
22 | ||
23 | /* | |
24 | * Extents support for EXT4 | |
25 | * | |
26 | * TODO: | |
27 | * - ext4*_error() should be used in some situations | |
28 | * - analyze all BUG()/BUG_ON(), use -EIO where appropriate | |
29 | * - smart tree reduction | |
30 | */ | |
31 | ||
a86c6181 AT |
32 | #include <linux/fs.h> |
33 | #include <linux/time.h> | |
cd02ff0b | 34 | #include <linux/jbd2.h> |
a86c6181 AT |
35 | #include <linux/highuid.h> |
36 | #include <linux/pagemap.h> | |
37 | #include <linux/quotaops.h> | |
38 | #include <linux/string.h> | |
39 | #include <linux/slab.h> | |
a2df2a63 | 40 | #include <linux/falloc.h> |
a86c6181 | 41 | #include <asm/uaccess.h> |
6873fa0d | 42 | #include <linux/fiemap.h> |
3dcf5451 | 43 | #include "ext4_jbd2.h" |
4a092d73 | 44 | #include "ext4_extents.h" |
f19d5870 | 45 | #include "xattr.h" |
a86c6181 | 46 | |
0562e0ba JZ |
47 | #include <trace/events/ext4.h> |
48 | ||
5f95d21f LC |
49 | /* |
50 | * used by extent splitting. | |
51 | */ | |
52 | #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \ | |
53 | due to ENOSPC */ | |
54 | #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */ | |
55 | #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */ | |
56 | ||
dee1f973 DM |
57 | #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */ |
58 | #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */ | |
59 | ||
7ac5990d DW |
60 | static __le32 ext4_extent_block_csum(struct inode *inode, |
61 | struct ext4_extent_header *eh) | |
62 | { | |
63 | struct ext4_inode_info *ei = EXT4_I(inode); | |
64 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
65 | __u32 csum; | |
66 | ||
67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh, | |
68 | EXT4_EXTENT_TAIL_OFFSET(eh)); | |
69 | return cpu_to_le32(csum); | |
70 | } | |
71 | ||
72 | static int ext4_extent_block_csum_verify(struct inode *inode, | |
73 | struct ext4_extent_header *eh) | |
74 | { | |
75 | struct ext4_extent_tail *et; | |
76 | ||
77 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
78 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
79 | return 1; | |
80 | ||
81 | et = find_ext4_extent_tail(eh); | |
82 | if (et->et_checksum != ext4_extent_block_csum(inode, eh)) | |
83 | return 0; | |
84 | return 1; | |
85 | } | |
86 | ||
87 | static void ext4_extent_block_csum_set(struct inode *inode, | |
88 | struct ext4_extent_header *eh) | |
89 | { | |
90 | struct ext4_extent_tail *et; | |
91 | ||
92 | if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
93 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
94 | return; | |
95 | ||
96 | et = find_ext4_extent_tail(eh); | |
97 | et->et_checksum = ext4_extent_block_csum(inode, eh); | |
98 | } | |
99 | ||
d583fb87 AH |
100 | static int ext4_split_extent(handle_t *handle, |
101 | struct inode *inode, | |
102 | struct ext4_ext_path *path, | |
103 | struct ext4_map_blocks *map, | |
104 | int split_flag, | |
105 | int flags); | |
106 | ||
5f95d21f LC |
107 | static int ext4_split_extent_at(handle_t *handle, |
108 | struct inode *inode, | |
109 | struct ext4_ext_path *path, | |
110 | ext4_lblk_t split, | |
111 | int split_flag, | |
112 | int flags); | |
113 | ||
91dd8c11 | 114 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 115 | struct extent_status *newes); |
91dd8c11 | 116 | |
487caeef JK |
117 | static int ext4_ext_truncate_extend_restart(handle_t *handle, |
118 | struct inode *inode, | |
119 | int needed) | |
a86c6181 AT |
120 | { |
121 | int err; | |
122 | ||
0390131b FM |
123 | if (!ext4_handle_valid(handle)) |
124 | return 0; | |
a86c6181 | 125 | if (handle->h_buffer_credits > needed) |
9102e4fa SF |
126 | return 0; |
127 | err = ext4_journal_extend(handle, needed); | |
0123c939 | 128 | if (err <= 0) |
9102e4fa | 129 | return err; |
487caeef | 130 | err = ext4_truncate_restart_trans(handle, inode, needed); |
0617b83f DM |
131 | if (err == 0) |
132 | err = -EAGAIN; | |
487caeef JK |
133 | |
134 | return err; | |
a86c6181 AT |
135 | } |
136 | ||
137 | /* | |
138 | * could return: | |
139 | * - EROFS | |
140 | * - ENOMEM | |
141 | */ | |
142 | static int ext4_ext_get_access(handle_t *handle, struct inode *inode, | |
143 | struct ext4_ext_path *path) | |
144 | { | |
145 | if (path->p_bh) { | |
146 | /* path points to block */ | |
147 | return ext4_journal_get_write_access(handle, path->p_bh); | |
148 | } | |
149 | /* path points to leaf/index in inode body */ | |
150 | /* we use in-core data, no need to protect them */ | |
151 | return 0; | |
152 | } | |
153 | ||
154 | /* | |
155 | * could return: | |
156 | * - EROFS | |
157 | * - ENOMEM | |
158 | * - EIO | |
159 | */ | |
2656497b DW |
160 | int __ext4_ext_dirty(const char *where, unsigned int line, handle_t *handle, |
161 | struct inode *inode, struct ext4_ext_path *path) | |
a86c6181 AT |
162 | { |
163 | int err; | |
164 | if (path->p_bh) { | |
7ac5990d | 165 | ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh)); |
a86c6181 | 166 | /* path points to block */ |
9ea7a0df TT |
167 | err = __ext4_handle_dirty_metadata(where, line, handle, |
168 | inode, path->p_bh); | |
a86c6181 AT |
169 | } else { |
170 | /* path points to leaf/index in inode body */ | |
171 | err = ext4_mark_inode_dirty(handle, inode); | |
172 | } | |
173 | return err; | |
174 | } | |
175 | ||
f65e6fba | 176 | static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode, |
a86c6181 | 177 | struct ext4_ext_path *path, |
725d26d3 | 178 | ext4_lblk_t block) |
a86c6181 | 179 | { |
a86c6181 | 180 | if (path) { |
81fdbb4a | 181 | int depth = path->p_depth; |
a86c6181 | 182 | struct ext4_extent *ex; |
a86c6181 | 183 | |
ad4fb9ca KM |
184 | /* |
185 | * Try to predict block placement assuming that we are | |
186 | * filling in a file which will eventually be | |
187 | * non-sparse --- i.e., in the case of libbfd writing | |
188 | * an ELF object sections out-of-order but in a way | |
189 | * the eventually results in a contiguous object or | |
190 | * executable file, or some database extending a table | |
191 | * space file. However, this is actually somewhat | |
192 | * non-ideal if we are writing a sparse file such as | |
193 | * qemu or KVM writing a raw image file that is going | |
194 | * to stay fairly sparse, since it will end up | |
195 | * fragmenting the file system's free space. Maybe we | |
196 | * should have some hueristics or some way to allow | |
197 | * userspace to pass a hint to file system, | |
b8d6568a | 198 | * especially if the latter case turns out to be |
ad4fb9ca KM |
199 | * common. |
200 | */ | |
7e028976 | 201 | ex = path[depth].p_ext; |
ad4fb9ca KM |
202 | if (ex) { |
203 | ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex); | |
204 | ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block); | |
205 | ||
206 | if (block > ext_block) | |
207 | return ext_pblk + (block - ext_block); | |
208 | else | |
209 | return ext_pblk - (ext_block - block); | |
210 | } | |
a86c6181 | 211 | |
d0d856e8 RD |
212 | /* it looks like index is empty; |
213 | * try to find starting block from index itself */ | |
a86c6181 AT |
214 | if (path[depth].p_bh) |
215 | return path[depth].p_bh->b_blocknr; | |
216 | } | |
217 | ||
218 | /* OK. use inode's group */ | |
f86186b4 | 219 | return ext4_inode_to_goal_block(inode); |
a86c6181 AT |
220 | } |
221 | ||
654b4908 AK |
222 | /* |
223 | * Allocation for a meta data block | |
224 | */ | |
f65e6fba | 225 | static ext4_fsblk_t |
654b4908 | 226 | ext4_ext_new_meta_block(handle_t *handle, struct inode *inode, |
a86c6181 | 227 | struct ext4_ext_path *path, |
55f020db | 228 | struct ext4_extent *ex, int *err, unsigned int flags) |
a86c6181 | 229 | { |
f65e6fba | 230 | ext4_fsblk_t goal, newblock; |
a86c6181 AT |
231 | |
232 | goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); | |
55f020db AH |
233 | newblock = ext4_new_meta_blocks(handle, inode, goal, flags, |
234 | NULL, err); | |
a86c6181 AT |
235 | return newblock; |
236 | } | |
237 | ||
55ad63bf | 238 | static inline int ext4_ext_space_block(struct inode *inode, int check) |
a86c6181 AT |
239 | { |
240 | int size; | |
241 | ||
242 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
243 | / sizeof(struct ext4_extent); | |
bbf2f9fb | 244 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
245 | if (!check && size > 6) |
246 | size = 6; | |
a86c6181 AT |
247 | #endif |
248 | return size; | |
249 | } | |
250 | ||
55ad63bf | 251 | static inline int ext4_ext_space_block_idx(struct inode *inode, int check) |
a86c6181 AT |
252 | { |
253 | int size; | |
254 | ||
255 | size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) | |
256 | / sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 257 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
258 | if (!check && size > 5) |
259 | size = 5; | |
a86c6181 AT |
260 | #endif |
261 | return size; | |
262 | } | |
263 | ||
55ad63bf | 264 | static inline int ext4_ext_space_root(struct inode *inode, int check) |
a86c6181 AT |
265 | { |
266 | int size; | |
267 | ||
268 | size = sizeof(EXT4_I(inode)->i_data); | |
269 | size -= sizeof(struct ext4_extent_header); | |
270 | size /= sizeof(struct ext4_extent); | |
bbf2f9fb | 271 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
272 | if (!check && size > 3) |
273 | size = 3; | |
a86c6181 AT |
274 | #endif |
275 | return size; | |
276 | } | |
277 | ||
55ad63bf | 278 | static inline int ext4_ext_space_root_idx(struct inode *inode, int check) |
a86c6181 AT |
279 | { |
280 | int size; | |
281 | ||
282 | size = sizeof(EXT4_I(inode)->i_data); | |
283 | size -= sizeof(struct ext4_extent_header); | |
284 | size /= sizeof(struct ext4_extent_idx); | |
bbf2f9fb | 285 | #ifdef AGGRESSIVE_TEST |
02dc62fb YY |
286 | if (!check && size > 4) |
287 | size = 4; | |
a86c6181 AT |
288 | #endif |
289 | return size; | |
290 | } | |
291 | ||
d2a17637 MC |
292 | /* |
293 | * Calculate the number of metadata blocks needed | |
294 | * to allocate @blocks | |
295 | * Worse case is one block per extent | |
296 | */ | |
01f49d0b | 297 | int ext4_ext_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 298 | { |
9d0be502 | 299 | struct ext4_inode_info *ei = EXT4_I(inode); |
81fdbb4a | 300 | int idxs; |
d2a17637 | 301 | |
9d0be502 TT |
302 | idxs = ((inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header)) |
303 | / sizeof(struct ext4_extent_idx)); | |
d2a17637 MC |
304 | |
305 | /* | |
9d0be502 TT |
306 | * If the new delayed allocation block is contiguous with the |
307 | * previous da block, it can share index blocks with the | |
308 | * previous block, so we only need to allocate a new index | |
309 | * block every idxs leaf blocks. At ldxs**2 blocks, we need | |
310 | * an additional index block, and at ldxs**3 blocks, yet | |
311 | * another index blocks. | |
d2a17637 | 312 | */ |
9d0be502 TT |
313 | if (ei->i_da_metadata_calc_len && |
314 | ei->i_da_metadata_calc_last_lblock+1 == lblock) { | |
81fdbb4a YY |
315 | int num = 0; |
316 | ||
9d0be502 TT |
317 | if ((ei->i_da_metadata_calc_len % idxs) == 0) |
318 | num++; | |
319 | if ((ei->i_da_metadata_calc_len % (idxs*idxs)) == 0) | |
320 | num++; | |
321 | if ((ei->i_da_metadata_calc_len % (idxs*idxs*idxs)) == 0) { | |
322 | num++; | |
323 | ei->i_da_metadata_calc_len = 0; | |
324 | } else | |
325 | ei->i_da_metadata_calc_len++; | |
326 | ei->i_da_metadata_calc_last_lblock++; | |
327 | return num; | |
328 | } | |
d2a17637 | 329 | |
9d0be502 TT |
330 | /* |
331 | * In the worst case we need a new set of index blocks at | |
332 | * every level of the inode's extent tree. | |
333 | */ | |
334 | ei->i_da_metadata_calc_len = 1; | |
335 | ei->i_da_metadata_calc_last_lblock = lblock; | |
336 | return ext_depth(inode) + 1; | |
d2a17637 MC |
337 | } |
338 | ||
c29c0ae7 AT |
339 | static int |
340 | ext4_ext_max_entries(struct inode *inode, int depth) | |
341 | { | |
342 | int max; | |
343 | ||
344 | if (depth == ext_depth(inode)) { | |
345 | if (depth == 0) | |
55ad63bf | 346 | max = ext4_ext_space_root(inode, 1); |
c29c0ae7 | 347 | else |
55ad63bf | 348 | max = ext4_ext_space_root_idx(inode, 1); |
c29c0ae7 AT |
349 | } else { |
350 | if (depth == 0) | |
55ad63bf | 351 | max = ext4_ext_space_block(inode, 1); |
c29c0ae7 | 352 | else |
55ad63bf | 353 | max = ext4_ext_space_block_idx(inode, 1); |
c29c0ae7 AT |
354 | } |
355 | ||
356 | return max; | |
357 | } | |
358 | ||
56b19868 AK |
359 | static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext) |
360 | { | |
bf89d16f | 361 | ext4_fsblk_t block = ext4_ext_pblock(ext); |
56b19868 | 362 | int len = ext4_ext_get_actual_len(ext); |
e84a26ce | 363 | |
31d4f3a2 TT |
364 | if (len == 0) |
365 | return 0; | |
6fd058f7 | 366 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len); |
56b19868 AK |
367 | } |
368 | ||
369 | static int ext4_valid_extent_idx(struct inode *inode, | |
370 | struct ext4_extent_idx *ext_idx) | |
371 | { | |
bf89d16f | 372 | ext4_fsblk_t block = ext4_idx_pblock(ext_idx); |
e84a26ce | 373 | |
6fd058f7 | 374 | return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1); |
56b19868 AK |
375 | } |
376 | ||
377 | static int ext4_valid_extent_entries(struct inode *inode, | |
378 | struct ext4_extent_header *eh, | |
379 | int depth) | |
380 | { | |
56b19868 AK |
381 | unsigned short entries; |
382 | if (eh->eh_entries == 0) | |
383 | return 1; | |
384 | ||
385 | entries = le16_to_cpu(eh->eh_entries); | |
386 | ||
387 | if (depth == 0) { | |
388 | /* leaf entries */ | |
81fdbb4a | 389 | struct ext4_extent *ext = EXT_FIRST_EXTENT(eh); |
56b19868 AK |
390 | while (entries) { |
391 | if (!ext4_valid_extent(inode, ext)) | |
392 | return 0; | |
393 | ext++; | |
394 | entries--; | |
395 | } | |
396 | } else { | |
81fdbb4a | 397 | struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh); |
56b19868 AK |
398 | while (entries) { |
399 | if (!ext4_valid_extent_idx(inode, ext_idx)) | |
400 | return 0; | |
401 | ext_idx++; | |
402 | entries--; | |
403 | } | |
404 | } | |
405 | return 1; | |
406 | } | |
407 | ||
c398eda0 TT |
408 | static int __ext4_ext_check(const char *function, unsigned int line, |
409 | struct inode *inode, struct ext4_extent_header *eh, | |
c349179b | 410 | int depth, ext4_fsblk_t pblk) |
c29c0ae7 AT |
411 | { |
412 | const char *error_msg; | |
413 | int max = 0; | |
414 | ||
415 | if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) { | |
416 | error_msg = "invalid magic"; | |
417 | goto corrupted; | |
418 | } | |
419 | if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) { | |
420 | error_msg = "unexpected eh_depth"; | |
421 | goto corrupted; | |
422 | } | |
423 | if (unlikely(eh->eh_max == 0)) { | |
424 | error_msg = "invalid eh_max"; | |
425 | goto corrupted; | |
426 | } | |
427 | max = ext4_ext_max_entries(inode, depth); | |
428 | if (unlikely(le16_to_cpu(eh->eh_max) > max)) { | |
429 | error_msg = "too large eh_max"; | |
430 | goto corrupted; | |
431 | } | |
432 | if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) { | |
433 | error_msg = "invalid eh_entries"; | |
434 | goto corrupted; | |
435 | } | |
56b19868 AK |
436 | if (!ext4_valid_extent_entries(inode, eh, depth)) { |
437 | error_msg = "invalid extent entries"; | |
438 | goto corrupted; | |
439 | } | |
7ac5990d DW |
440 | /* Verify checksum on non-root extent tree nodes */ |
441 | if (ext_depth(inode) != depth && | |
442 | !ext4_extent_block_csum_verify(inode, eh)) { | |
443 | error_msg = "extent tree corrupted"; | |
444 | goto corrupted; | |
445 | } | |
c29c0ae7 AT |
446 | return 0; |
447 | ||
448 | corrupted: | |
c398eda0 | 449 | ext4_error_inode(inode, function, line, 0, |
c349179b TT |
450 | "pblk %llu bad header/extent: %s - magic %x, " |
451 | "entries %u, max %u(%u), depth %u(%u)", | |
452 | (unsigned long long) pblk, error_msg, | |
453 | le16_to_cpu(eh->eh_magic), | |
454 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max), | |
455 | max, le16_to_cpu(eh->eh_depth), depth); | |
c29c0ae7 AT |
456 | return -EIO; |
457 | } | |
458 | ||
c349179b TT |
459 | #define ext4_ext_check(inode, eh, depth, pblk) \ |
460 | __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk)) | |
c29c0ae7 | 461 | |
7a262f7c AK |
462 | int ext4_ext_check_inode(struct inode *inode) |
463 | { | |
c349179b | 464 | return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0); |
7a262f7c AK |
465 | } |
466 | ||
7d7ea89e TT |
467 | static struct buffer_head * |
468 | __read_extent_tree_block(const char *function, unsigned int line, | |
107a7bd3 TT |
469 | struct inode *inode, ext4_fsblk_t pblk, int depth, |
470 | int flags) | |
f8489128 | 471 | { |
7d7ea89e TT |
472 | struct buffer_head *bh; |
473 | int err; | |
474 | ||
475 | bh = sb_getblk(inode->i_sb, pblk); | |
476 | if (unlikely(!bh)) | |
477 | return ERR_PTR(-ENOMEM); | |
f8489128 | 478 | |
7d7ea89e TT |
479 | if (!bh_uptodate_or_lock(bh)) { |
480 | trace_ext4_ext_load_extent(inode, pblk, _RET_IP_); | |
481 | err = bh_submit_read(bh); | |
482 | if (err < 0) | |
483 | goto errout; | |
484 | } | |
f8489128 | 485 | if (buffer_verified(bh)) |
7d7ea89e TT |
486 | return bh; |
487 | err = __ext4_ext_check(function, line, inode, | |
c349179b | 488 | ext_block_hdr(bh), depth, pblk); |
7d7ea89e TT |
489 | if (err) |
490 | goto errout; | |
f8489128 | 491 | set_buffer_verified(bh); |
107a7bd3 TT |
492 | /* |
493 | * If this is a leaf block, cache all of its entries | |
494 | */ | |
495 | if (!(flags & EXT4_EX_NOCACHE) && depth == 0) { | |
496 | struct ext4_extent_header *eh = ext_block_hdr(bh); | |
497 | struct ext4_extent *ex = EXT_FIRST_EXTENT(eh); | |
498 | ext4_lblk_t prev = 0; | |
499 | int i; | |
500 | ||
501 | for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) { | |
502 | unsigned int status = EXTENT_STATUS_WRITTEN; | |
503 | ext4_lblk_t lblk = le32_to_cpu(ex->ee_block); | |
504 | int len = ext4_ext_get_actual_len(ex); | |
505 | ||
506 | if (prev && (prev != lblk)) | |
507 | ext4_es_cache_extent(inode, prev, | |
508 | lblk - prev, ~0, | |
509 | EXTENT_STATUS_HOLE); | |
510 | ||
511 | if (ext4_ext_is_uninitialized(ex)) | |
512 | status = EXTENT_STATUS_UNWRITTEN; | |
513 | ext4_es_cache_extent(inode, lblk, len, | |
514 | ext4_ext_pblock(ex), status); | |
515 | prev = lblk + len; | |
516 | } | |
517 | } | |
7d7ea89e TT |
518 | return bh; |
519 | errout: | |
520 | put_bh(bh); | |
521 | return ERR_PTR(err); | |
522 | ||
f8489128 DW |
523 | } |
524 | ||
107a7bd3 TT |
525 | #define read_extent_tree_block(inode, pblk, depth, flags) \ |
526 | __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \ | |
527 | (depth), (flags)) | |
f8489128 | 528 | |
a86c6181 AT |
529 | #ifdef EXT_DEBUG |
530 | static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path) | |
531 | { | |
532 | int k, l = path->p_depth; | |
533 | ||
534 | ext_debug("path:"); | |
535 | for (k = 0; k <= l; k++, path++) { | |
536 | if (path->p_idx) { | |
2ae02107 | 537 | ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 538 | ext4_idx_pblock(path->p_idx)); |
a86c6181 | 539 | } else if (path->p_ext) { |
553f9008 | 540 | ext_debug(" %d:[%d]%d:%llu ", |
a86c6181 | 541 | le32_to_cpu(path->p_ext->ee_block), |
553f9008 | 542 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 543 | ext4_ext_get_actual_len(path->p_ext), |
bf89d16f | 544 | ext4_ext_pblock(path->p_ext)); |
a86c6181 AT |
545 | } else |
546 | ext_debug(" []"); | |
547 | } | |
548 | ext_debug("\n"); | |
549 | } | |
550 | ||
551 | static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path) | |
552 | { | |
553 | int depth = ext_depth(inode); | |
554 | struct ext4_extent_header *eh; | |
555 | struct ext4_extent *ex; | |
556 | int i; | |
557 | ||
558 | if (!path) | |
559 | return; | |
560 | ||
561 | eh = path[depth].p_hdr; | |
562 | ex = EXT_FIRST_EXTENT(eh); | |
563 | ||
553f9008 M |
564 | ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino); |
565 | ||
a86c6181 | 566 | for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) { |
553f9008 M |
567 | ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block), |
568 | ext4_ext_is_uninitialized(ex), | |
bf89d16f | 569 | ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex)); |
a86c6181 AT |
570 | } |
571 | ext_debug("\n"); | |
572 | } | |
1b16da77 YY |
573 | |
574 | static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path, | |
575 | ext4_fsblk_t newblock, int level) | |
576 | { | |
577 | int depth = ext_depth(inode); | |
578 | struct ext4_extent *ex; | |
579 | ||
580 | if (depth != level) { | |
581 | struct ext4_extent_idx *idx; | |
582 | idx = path[level].p_idx; | |
583 | while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) { | |
584 | ext_debug("%d: move %d:%llu in new index %llu\n", level, | |
585 | le32_to_cpu(idx->ei_block), | |
586 | ext4_idx_pblock(idx), | |
587 | newblock); | |
588 | idx++; | |
589 | } | |
590 | ||
591 | return; | |
592 | } | |
593 | ||
594 | ex = path[depth].p_ext; | |
595 | while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) { | |
596 | ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n", | |
597 | le32_to_cpu(ex->ee_block), | |
598 | ext4_ext_pblock(ex), | |
599 | ext4_ext_is_uninitialized(ex), | |
600 | ext4_ext_get_actual_len(ex), | |
601 | newblock); | |
602 | ex++; | |
603 | } | |
604 | } | |
605 | ||
a86c6181 | 606 | #else |
af5bc92d TT |
607 | #define ext4_ext_show_path(inode, path) |
608 | #define ext4_ext_show_leaf(inode, path) | |
1b16da77 | 609 | #define ext4_ext_show_move(inode, path, newblock, level) |
a86c6181 AT |
610 | #endif |
611 | ||
b35905c1 | 612 | void ext4_ext_drop_refs(struct ext4_ext_path *path) |
a86c6181 AT |
613 | { |
614 | int depth = path->p_depth; | |
615 | int i; | |
616 | ||
617 | for (i = 0; i <= depth; i++, path++) | |
618 | if (path->p_bh) { | |
619 | brelse(path->p_bh); | |
620 | path->p_bh = NULL; | |
621 | } | |
622 | } | |
623 | ||
624 | /* | |
d0d856e8 RD |
625 | * ext4_ext_binsearch_idx: |
626 | * binary search for the closest index of the given block | |
c29c0ae7 | 627 | * the header must be checked before calling this |
a86c6181 AT |
628 | */ |
629 | static void | |
725d26d3 AK |
630 | ext4_ext_binsearch_idx(struct inode *inode, |
631 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
632 | { |
633 | struct ext4_extent_header *eh = path->p_hdr; | |
634 | struct ext4_extent_idx *r, *l, *m; | |
635 | ||
a86c6181 | 636 | |
bba90743 | 637 | ext_debug("binsearch for %u(idx): ", block); |
a86c6181 AT |
638 | |
639 | l = EXT_FIRST_INDEX(eh) + 1; | |
e9f410b1 | 640 | r = EXT_LAST_INDEX(eh); |
a86c6181 AT |
641 | while (l <= r) { |
642 | m = l + (r - l) / 2; | |
643 | if (block < le32_to_cpu(m->ei_block)) | |
644 | r = m - 1; | |
645 | else | |
646 | l = m + 1; | |
26d535ed DM |
647 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block), |
648 | m, le32_to_cpu(m->ei_block), | |
649 | r, le32_to_cpu(r->ei_block)); | |
a86c6181 AT |
650 | } |
651 | ||
652 | path->p_idx = l - 1; | |
4a3c3a51 | 653 | ext_debug(" -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block), |
bf89d16f | 654 | ext4_idx_pblock(path->p_idx)); |
a86c6181 AT |
655 | |
656 | #ifdef CHECK_BINSEARCH | |
657 | { | |
658 | struct ext4_extent_idx *chix, *ix; | |
659 | int k; | |
660 | ||
661 | chix = ix = EXT_FIRST_INDEX(eh); | |
662 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) { | |
663 | if (k != 0 && | |
664 | le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) { | |
4776004f TT |
665 | printk(KERN_DEBUG "k=%d, ix=0x%p, " |
666 | "first=0x%p\n", k, | |
667 | ix, EXT_FIRST_INDEX(eh)); | |
668 | printk(KERN_DEBUG "%u <= %u\n", | |
a86c6181 AT |
669 | le32_to_cpu(ix->ei_block), |
670 | le32_to_cpu(ix[-1].ei_block)); | |
671 | } | |
672 | BUG_ON(k && le32_to_cpu(ix->ei_block) | |
8c55e204 | 673 | <= le32_to_cpu(ix[-1].ei_block)); |
a86c6181 AT |
674 | if (block < le32_to_cpu(ix->ei_block)) |
675 | break; | |
676 | chix = ix; | |
677 | } | |
678 | BUG_ON(chix != path->p_idx); | |
679 | } | |
680 | #endif | |
681 | ||
682 | } | |
683 | ||
684 | /* | |
d0d856e8 RD |
685 | * ext4_ext_binsearch: |
686 | * binary search for closest extent of the given block | |
c29c0ae7 | 687 | * the header must be checked before calling this |
a86c6181 AT |
688 | */ |
689 | static void | |
725d26d3 AK |
690 | ext4_ext_binsearch(struct inode *inode, |
691 | struct ext4_ext_path *path, ext4_lblk_t block) | |
a86c6181 AT |
692 | { |
693 | struct ext4_extent_header *eh = path->p_hdr; | |
694 | struct ext4_extent *r, *l, *m; | |
695 | ||
a86c6181 AT |
696 | if (eh->eh_entries == 0) { |
697 | /* | |
d0d856e8 RD |
698 | * this leaf is empty: |
699 | * we get such a leaf in split/add case | |
a86c6181 AT |
700 | */ |
701 | return; | |
702 | } | |
703 | ||
bba90743 | 704 | ext_debug("binsearch for %u: ", block); |
a86c6181 AT |
705 | |
706 | l = EXT_FIRST_EXTENT(eh) + 1; | |
e9f410b1 | 707 | r = EXT_LAST_EXTENT(eh); |
a86c6181 AT |
708 | |
709 | while (l <= r) { | |
710 | m = l + (r - l) / 2; | |
711 | if (block < le32_to_cpu(m->ee_block)) | |
712 | r = m - 1; | |
713 | else | |
714 | l = m + 1; | |
26d535ed DM |
715 | ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block), |
716 | m, le32_to_cpu(m->ee_block), | |
717 | r, le32_to_cpu(r->ee_block)); | |
a86c6181 AT |
718 | } |
719 | ||
720 | path->p_ext = l - 1; | |
553f9008 | 721 | ext_debug(" -> %d:%llu:[%d]%d ", |
8c55e204 | 722 | le32_to_cpu(path->p_ext->ee_block), |
bf89d16f | 723 | ext4_ext_pblock(path->p_ext), |
553f9008 | 724 | ext4_ext_is_uninitialized(path->p_ext), |
a2df2a63 | 725 | ext4_ext_get_actual_len(path->p_ext)); |
a86c6181 AT |
726 | |
727 | #ifdef CHECK_BINSEARCH | |
728 | { | |
729 | struct ext4_extent *chex, *ex; | |
730 | int k; | |
731 | ||
732 | chex = ex = EXT_FIRST_EXTENT(eh); | |
733 | for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) { | |
734 | BUG_ON(k && le32_to_cpu(ex->ee_block) | |
8c55e204 | 735 | <= le32_to_cpu(ex[-1].ee_block)); |
a86c6181 AT |
736 | if (block < le32_to_cpu(ex->ee_block)) |
737 | break; | |
738 | chex = ex; | |
739 | } | |
740 | BUG_ON(chex != path->p_ext); | |
741 | } | |
742 | #endif | |
743 | ||
744 | } | |
745 | ||
746 | int ext4_ext_tree_init(handle_t *handle, struct inode *inode) | |
747 | { | |
748 | struct ext4_extent_header *eh; | |
749 | ||
750 | eh = ext_inode_hdr(inode); | |
751 | eh->eh_depth = 0; | |
752 | eh->eh_entries = 0; | |
753 | eh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 754 | eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 | 755 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
756 | return 0; |
757 | } | |
758 | ||
759 | struct ext4_ext_path * | |
725d26d3 | 760 | ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block, |
107a7bd3 | 761 | struct ext4_ext_path *path, int flags) |
a86c6181 AT |
762 | { |
763 | struct ext4_extent_header *eh; | |
764 | struct buffer_head *bh; | |
765 | short int depth, i, ppos = 0, alloc = 0; | |
860d21e2 | 766 | int ret; |
a86c6181 AT |
767 | |
768 | eh = ext_inode_hdr(inode); | |
c29c0ae7 | 769 | depth = ext_depth(inode); |
a86c6181 AT |
770 | |
771 | /* account possible depth increase */ | |
772 | if (!path) { | |
5d4958f9 | 773 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2), |
a86c6181 AT |
774 | GFP_NOFS); |
775 | if (!path) | |
776 | return ERR_PTR(-ENOMEM); | |
777 | alloc = 1; | |
778 | } | |
a86c6181 | 779 | path[0].p_hdr = eh; |
1973adcb | 780 | path[0].p_bh = NULL; |
a86c6181 | 781 | |
c29c0ae7 | 782 | i = depth; |
a86c6181 AT |
783 | /* walk through the tree */ |
784 | while (i) { | |
785 | ext_debug("depth %d: num %d, max %d\n", | |
786 | ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
c29c0ae7 | 787 | |
a86c6181 | 788 | ext4_ext_binsearch_idx(inode, path + ppos, block); |
bf89d16f | 789 | path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx); |
a86c6181 AT |
790 | path[ppos].p_depth = i; |
791 | path[ppos].p_ext = NULL; | |
792 | ||
107a7bd3 TT |
793 | bh = read_extent_tree_block(inode, path[ppos].p_block, --i, |
794 | flags); | |
7d7ea89e TT |
795 | if (IS_ERR(bh)) { |
796 | ret = PTR_ERR(bh); | |
a86c6181 | 797 | goto err; |
860d21e2 | 798 | } |
7d7ea89e | 799 | |
a86c6181 AT |
800 | eh = ext_block_hdr(bh); |
801 | ppos++; | |
273df556 FM |
802 | if (unlikely(ppos > depth)) { |
803 | put_bh(bh); | |
804 | EXT4_ERROR_INODE(inode, | |
805 | "ppos %d > depth %d", ppos, depth); | |
860d21e2 | 806 | ret = -EIO; |
273df556 FM |
807 | goto err; |
808 | } | |
a86c6181 AT |
809 | path[ppos].p_bh = bh; |
810 | path[ppos].p_hdr = eh; | |
a86c6181 AT |
811 | } |
812 | ||
813 | path[ppos].p_depth = i; | |
a86c6181 AT |
814 | path[ppos].p_ext = NULL; |
815 | path[ppos].p_idx = NULL; | |
816 | ||
a86c6181 AT |
817 | /* find extent */ |
818 | ext4_ext_binsearch(inode, path + ppos, block); | |
1973adcb SF |
819 | /* if not an empty leaf */ |
820 | if (path[ppos].p_ext) | |
bf89d16f | 821 | path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext); |
a86c6181 AT |
822 | |
823 | ext4_ext_show_path(inode, path); | |
824 | ||
825 | return path; | |
826 | ||
827 | err: | |
828 | ext4_ext_drop_refs(path); | |
829 | if (alloc) | |
830 | kfree(path); | |
860d21e2 | 831 | return ERR_PTR(ret); |
a86c6181 AT |
832 | } |
833 | ||
834 | /* | |
d0d856e8 RD |
835 | * ext4_ext_insert_index: |
836 | * insert new index [@logical;@ptr] into the block at @curp; | |
837 | * check where to insert: before @curp or after @curp | |
a86c6181 | 838 | */ |
1f109d5a TT |
839 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
840 | struct ext4_ext_path *curp, | |
841 | int logical, ext4_fsblk_t ptr) | |
a86c6181 AT |
842 | { |
843 | struct ext4_extent_idx *ix; | |
844 | int len, err; | |
845 | ||
7e028976 AM |
846 | err = ext4_ext_get_access(handle, inode, curp); |
847 | if (err) | |
a86c6181 AT |
848 | return err; |
849 | ||
273df556 FM |
850 | if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) { |
851 | EXT4_ERROR_INODE(inode, | |
852 | "logical %d == ei_block %d!", | |
853 | logical, le32_to_cpu(curp->p_idx->ei_block)); | |
854 | return -EIO; | |
855 | } | |
d4620315 RD |
856 | |
857 | if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries) | |
858 | >= le16_to_cpu(curp->p_hdr->eh_max))) { | |
859 | EXT4_ERROR_INODE(inode, | |
860 | "eh_entries %d >= eh_max %d!", | |
861 | le16_to_cpu(curp->p_hdr->eh_entries), | |
862 | le16_to_cpu(curp->p_hdr->eh_max)); | |
863 | return -EIO; | |
864 | } | |
865 | ||
a86c6181 AT |
866 | if (logical > le32_to_cpu(curp->p_idx->ei_block)) { |
867 | /* insert after */ | |
80e675f9 | 868 | ext_debug("insert new index %d after: %llu\n", logical, ptr); |
a86c6181 AT |
869 | ix = curp->p_idx + 1; |
870 | } else { | |
871 | /* insert before */ | |
80e675f9 | 872 | ext_debug("insert new index %d before: %llu\n", logical, ptr); |
a86c6181 AT |
873 | ix = curp->p_idx; |
874 | } | |
875 | ||
80e675f9 EG |
876 | len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1; |
877 | BUG_ON(len < 0); | |
878 | if (len > 0) { | |
879 | ext_debug("insert new index %d: " | |
880 | "move %d indices from 0x%p to 0x%p\n", | |
881 | logical, len, ix, ix + 1); | |
882 | memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx)); | |
883 | } | |
884 | ||
f472e026 TM |
885 | if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) { |
886 | EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!"); | |
887 | return -EIO; | |
888 | } | |
889 | ||
a86c6181 | 890 | ix->ei_block = cpu_to_le32(logical); |
f65e6fba | 891 | ext4_idx_store_pblock(ix, ptr); |
e8546d06 | 892 | le16_add_cpu(&curp->p_hdr->eh_entries, 1); |
a86c6181 | 893 | |
273df556 FM |
894 | if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) { |
895 | EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!"); | |
896 | return -EIO; | |
897 | } | |
a86c6181 AT |
898 | |
899 | err = ext4_ext_dirty(handle, inode, curp); | |
900 | ext4_std_error(inode->i_sb, err); | |
901 | ||
902 | return err; | |
903 | } | |
904 | ||
905 | /* | |
d0d856e8 RD |
906 | * ext4_ext_split: |
907 | * inserts new subtree into the path, using free index entry | |
908 | * at depth @at: | |
909 | * - allocates all needed blocks (new leaf and all intermediate index blocks) | |
910 | * - makes decision where to split | |
911 | * - moves remaining extents and index entries (right to the split point) | |
912 | * into the newly allocated blocks | |
913 | * - initializes subtree | |
a86c6181 AT |
914 | */ |
915 | static int ext4_ext_split(handle_t *handle, struct inode *inode, | |
55f020db AH |
916 | unsigned int flags, |
917 | struct ext4_ext_path *path, | |
918 | struct ext4_extent *newext, int at) | |
a86c6181 AT |
919 | { |
920 | struct buffer_head *bh = NULL; | |
921 | int depth = ext_depth(inode); | |
922 | struct ext4_extent_header *neh; | |
923 | struct ext4_extent_idx *fidx; | |
a86c6181 | 924 | int i = at, k, m, a; |
f65e6fba | 925 | ext4_fsblk_t newblock, oldblock; |
a86c6181 | 926 | __le32 border; |
f65e6fba | 927 | ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */ |
a86c6181 AT |
928 | int err = 0; |
929 | ||
930 | /* make decision: where to split? */ | |
d0d856e8 | 931 | /* FIXME: now decision is simplest: at current extent */ |
a86c6181 | 932 | |
d0d856e8 | 933 | /* if current leaf will be split, then we should use |
a86c6181 | 934 | * border from split point */ |
273df556 FM |
935 | if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) { |
936 | EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!"); | |
937 | return -EIO; | |
938 | } | |
a86c6181 AT |
939 | if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) { |
940 | border = path[depth].p_ext[1].ee_block; | |
d0d856e8 | 941 | ext_debug("leaf will be split." |
a86c6181 | 942 | " next leaf starts at %d\n", |
8c55e204 | 943 | le32_to_cpu(border)); |
a86c6181 AT |
944 | } else { |
945 | border = newext->ee_block; | |
946 | ext_debug("leaf will be added." | |
947 | " next leaf starts at %d\n", | |
8c55e204 | 948 | le32_to_cpu(border)); |
a86c6181 AT |
949 | } |
950 | ||
951 | /* | |
d0d856e8 RD |
952 | * If error occurs, then we break processing |
953 | * and mark filesystem read-only. index won't | |
a86c6181 | 954 | * be inserted and tree will be in consistent |
d0d856e8 | 955 | * state. Next mount will repair buffers too. |
a86c6181 AT |
956 | */ |
957 | ||
958 | /* | |
d0d856e8 RD |
959 | * Get array to track all allocated blocks. |
960 | * We need this to handle errors and free blocks | |
961 | * upon them. | |
a86c6181 | 962 | */ |
5d4958f9 | 963 | ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS); |
a86c6181 AT |
964 | if (!ablocks) |
965 | return -ENOMEM; | |
a86c6181 AT |
966 | |
967 | /* allocate all needed blocks */ | |
968 | ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); | |
969 | for (a = 0; a < depth - at; a++) { | |
654b4908 | 970 | newblock = ext4_ext_new_meta_block(handle, inode, path, |
55f020db | 971 | newext, &err, flags); |
a86c6181 AT |
972 | if (newblock == 0) |
973 | goto cleanup; | |
974 | ablocks[a] = newblock; | |
975 | } | |
976 | ||
977 | /* initialize new leaf */ | |
978 | newblock = ablocks[--a]; | |
273df556 FM |
979 | if (unlikely(newblock == 0)) { |
980 | EXT4_ERROR_INODE(inode, "newblock == 0!"); | |
981 | err = -EIO; | |
982 | goto cleanup; | |
983 | } | |
a86c6181 | 984 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 985 | if (unlikely(!bh)) { |
860d21e2 | 986 | err = -ENOMEM; |
a86c6181 AT |
987 | goto cleanup; |
988 | } | |
989 | lock_buffer(bh); | |
990 | ||
7e028976 AM |
991 | err = ext4_journal_get_create_access(handle, bh); |
992 | if (err) | |
a86c6181 AT |
993 | goto cleanup; |
994 | ||
995 | neh = ext_block_hdr(bh); | |
996 | neh->eh_entries = 0; | |
55ad63bf | 997 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 AT |
998 | neh->eh_magic = EXT4_EXT_MAGIC; |
999 | neh->eh_depth = 0; | |
a86c6181 | 1000 | |
d0d856e8 | 1001 | /* move remainder of path[depth] to the new leaf */ |
273df556 FM |
1002 | if (unlikely(path[depth].p_hdr->eh_entries != |
1003 | path[depth].p_hdr->eh_max)) { | |
1004 | EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!", | |
1005 | path[depth].p_hdr->eh_entries, | |
1006 | path[depth].p_hdr->eh_max); | |
1007 | err = -EIO; | |
1008 | goto cleanup; | |
1009 | } | |
a86c6181 | 1010 | /* start copy from next extent */ |
1b16da77 YY |
1011 | m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++; |
1012 | ext4_ext_show_move(inode, path, newblock, depth); | |
a86c6181 | 1013 | if (m) { |
1b16da77 YY |
1014 | struct ext4_extent *ex; |
1015 | ex = EXT_FIRST_EXTENT(neh); | |
1016 | memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m); | |
e8546d06 | 1017 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 AT |
1018 | } |
1019 | ||
7ac5990d | 1020 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1021 | set_buffer_uptodate(bh); |
1022 | unlock_buffer(bh); | |
1023 | ||
0390131b | 1024 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1025 | if (err) |
a86c6181 AT |
1026 | goto cleanup; |
1027 | brelse(bh); | |
1028 | bh = NULL; | |
1029 | ||
1030 | /* correct old leaf */ | |
1031 | if (m) { | |
7e028976 AM |
1032 | err = ext4_ext_get_access(handle, inode, path + depth); |
1033 | if (err) | |
a86c6181 | 1034 | goto cleanup; |
e8546d06 | 1035 | le16_add_cpu(&path[depth].p_hdr->eh_entries, -m); |
7e028976 AM |
1036 | err = ext4_ext_dirty(handle, inode, path + depth); |
1037 | if (err) | |
a86c6181 AT |
1038 | goto cleanup; |
1039 | ||
1040 | } | |
1041 | ||
1042 | /* create intermediate indexes */ | |
1043 | k = depth - at - 1; | |
273df556 FM |
1044 | if (unlikely(k < 0)) { |
1045 | EXT4_ERROR_INODE(inode, "k %d < 0!", k); | |
1046 | err = -EIO; | |
1047 | goto cleanup; | |
1048 | } | |
a86c6181 AT |
1049 | if (k) |
1050 | ext_debug("create %d intermediate indices\n", k); | |
1051 | /* insert new index into current index block */ | |
1052 | /* current depth stored in i var */ | |
1053 | i = depth - 1; | |
1054 | while (k--) { | |
1055 | oldblock = newblock; | |
1056 | newblock = ablocks[--a]; | |
bba90743 | 1057 | bh = sb_getblk(inode->i_sb, newblock); |
aebf0243 | 1058 | if (unlikely(!bh)) { |
860d21e2 | 1059 | err = -ENOMEM; |
a86c6181 AT |
1060 | goto cleanup; |
1061 | } | |
1062 | lock_buffer(bh); | |
1063 | ||
7e028976 AM |
1064 | err = ext4_journal_get_create_access(handle, bh); |
1065 | if (err) | |
a86c6181 AT |
1066 | goto cleanup; |
1067 | ||
1068 | neh = ext_block_hdr(bh); | |
1069 | neh->eh_entries = cpu_to_le16(1); | |
1070 | neh->eh_magic = EXT4_EXT_MAGIC; | |
55ad63bf | 1071 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 AT |
1072 | neh->eh_depth = cpu_to_le16(depth - i); |
1073 | fidx = EXT_FIRST_INDEX(neh); | |
1074 | fidx->ei_block = border; | |
f65e6fba | 1075 | ext4_idx_store_pblock(fidx, oldblock); |
a86c6181 | 1076 | |
bba90743 ES |
1077 | ext_debug("int.index at %d (block %llu): %u -> %llu\n", |
1078 | i, newblock, le32_to_cpu(border), oldblock); | |
a86c6181 | 1079 | |
1b16da77 | 1080 | /* move remainder of path[i] to the new index block */ |
273df556 FM |
1081 | if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) != |
1082 | EXT_LAST_INDEX(path[i].p_hdr))) { | |
1083 | EXT4_ERROR_INODE(inode, | |
1084 | "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!", | |
1085 | le32_to_cpu(path[i].p_ext->ee_block)); | |
1086 | err = -EIO; | |
1087 | goto cleanup; | |
1088 | } | |
1b16da77 YY |
1089 | /* start copy indexes */ |
1090 | m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++; | |
1091 | ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx, | |
1092 | EXT_MAX_INDEX(path[i].p_hdr)); | |
1093 | ext4_ext_show_move(inode, path, newblock, i); | |
a86c6181 | 1094 | if (m) { |
1b16da77 | 1095 | memmove(++fidx, path[i].p_idx, |
a86c6181 | 1096 | sizeof(struct ext4_extent_idx) * m); |
e8546d06 | 1097 | le16_add_cpu(&neh->eh_entries, m); |
a86c6181 | 1098 | } |
7ac5990d | 1099 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1100 | set_buffer_uptodate(bh); |
1101 | unlock_buffer(bh); | |
1102 | ||
0390131b | 1103 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1104 | if (err) |
a86c6181 AT |
1105 | goto cleanup; |
1106 | brelse(bh); | |
1107 | bh = NULL; | |
1108 | ||
1109 | /* correct old index */ | |
1110 | if (m) { | |
1111 | err = ext4_ext_get_access(handle, inode, path + i); | |
1112 | if (err) | |
1113 | goto cleanup; | |
e8546d06 | 1114 | le16_add_cpu(&path[i].p_hdr->eh_entries, -m); |
a86c6181 AT |
1115 | err = ext4_ext_dirty(handle, inode, path + i); |
1116 | if (err) | |
1117 | goto cleanup; | |
1118 | } | |
1119 | ||
1120 | i--; | |
1121 | } | |
1122 | ||
1123 | /* insert new index */ | |
a86c6181 AT |
1124 | err = ext4_ext_insert_index(handle, inode, path + at, |
1125 | le32_to_cpu(border), newblock); | |
1126 | ||
1127 | cleanup: | |
1128 | if (bh) { | |
1129 | if (buffer_locked(bh)) | |
1130 | unlock_buffer(bh); | |
1131 | brelse(bh); | |
1132 | } | |
1133 | ||
1134 | if (err) { | |
1135 | /* free all allocated blocks in error case */ | |
1136 | for (i = 0; i < depth; i++) { | |
1137 | if (!ablocks[i]) | |
1138 | continue; | |
7dc57615 | 1139 | ext4_free_blocks(handle, inode, NULL, ablocks[i], 1, |
e6362609 | 1140 | EXT4_FREE_BLOCKS_METADATA); |
a86c6181 AT |
1141 | } |
1142 | } | |
1143 | kfree(ablocks); | |
1144 | ||
1145 | return err; | |
1146 | } | |
1147 | ||
1148 | /* | |
d0d856e8 RD |
1149 | * ext4_ext_grow_indepth: |
1150 | * implements tree growing procedure: | |
1151 | * - allocates new block | |
1152 | * - moves top-level data (index block or leaf) into the new block | |
1153 | * - initializes new top-level, creating index that points to the | |
1154 | * just created block | |
a86c6181 AT |
1155 | */ |
1156 | static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode, | |
55f020db | 1157 | unsigned int flags, |
55f020db | 1158 | struct ext4_extent *newext) |
a86c6181 | 1159 | { |
a86c6181 | 1160 | struct ext4_extent_header *neh; |
a86c6181 | 1161 | struct buffer_head *bh; |
f65e6fba | 1162 | ext4_fsblk_t newblock; |
a86c6181 AT |
1163 | int err = 0; |
1164 | ||
1939dd84 | 1165 | newblock = ext4_ext_new_meta_block(handle, inode, NULL, |
55f020db | 1166 | newext, &err, flags); |
a86c6181 AT |
1167 | if (newblock == 0) |
1168 | return err; | |
1169 | ||
1170 | bh = sb_getblk(inode->i_sb, newblock); | |
aebf0243 | 1171 | if (unlikely(!bh)) |
860d21e2 | 1172 | return -ENOMEM; |
a86c6181 AT |
1173 | lock_buffer(bh); |
1174 | ||
7e028976 AM |
1175 | err = ext4_journal_get_create_access(handle, bh); |
1176 | if (err) { | |
a86c6181 AT |
1177 | unlock_buffer(bh); |
1178 | goto out; | |
1179 | } | |
1180 | ||
1181 | /* move top-level index/leaf into new block */ | |
1939dd84 DM |
1182 | memmove(bh->b_data, EXT4_I(inode)->i_data, |
1183 | sizeof(EXT4_I(inode)->i_data)); | |
a86c6181 AT |
1184 | |
1185 | /* set size of new block */ | |
1186 | neh = ext_block_hdr(bh); | |
1187 | /* old root could have indexes or leaves | |
1188 | * so calculate e_max right way */ | |
1189 | if (ext_depth(inode)) | |
55ad63bf | 1190 | neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0)); |
a86c6181 | 1191 | else |
55ad63bf | 1192 | neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0)); |
a86c6181 | 1193 | neh->eh_magic = EXT4_EXT_MAGIC; |
7ac5990d | 1194 | ext4_extent_block_csum_set(inode, neh); |
a86c6181 AT |
1195 | set_buffer_uptodate(bh); |
1196 | unlock_buffer(bh); | |
1197 | ||
0390131b | 1198 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
7e028976 | 1199 | if (err) |
a86c6181 AT |
1200 | goto out; |
1201 | ||
1939dd84 | 1202 | /* Update top-level index: num,max,pointer */ |
a86c6181 | 1203 | neh = ext_inode_hdr(inode); |
1939dd84 DM |
1204 | neh->eh_entries = cpu_to_le16(1); |
1205 | ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock); | |
1206 | if (neh->eh_depth == 0) { | |
1207 | /* Root extent block becomes index block */ | |
1208 | neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0)); | |
1209 | EXT_FIRST_INDEX(neh)->ei_block = | |
1210 | EXT_FIRST_EXTENT(neh)->ee_block; | |
1211 | } | |
2ae02107 | 1212 | ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n", |
a86c6181 | 1213 | le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max), |
5a0790c2 | 1214 | le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block), |
bf89d16f | 1215 | ext4_idx_pblock(EXT_FIRST_INDEX(neh))); |
a86c6181 | 1216 | |
ba39ebb6 | 1217 | le16_add_cpu(&neh->eh_depth, 1); |
1939dd84 | 1218 | ext4_mark_inode_dirty(handle, inode); |
a86c6181 AT |
1219 | out: |
1220 | brelse(bh); | |
1221 | ||
1222 | return err; | |
1223 | } | |
1224 | ||
1225 | /* | |
d0d856e8 RD |
1226 | * ext4_ext_create_new_leaf: |
1227 | * finds empty index and adds new leaf. | |
1228 | * if no free index is found, then it requests in-depth growing. | |
a86c6181 AT |
1229 | */ |
1230 | static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode, | |
107a7bd3 TT |
1231 | unsigned int mb_flags, |
1232 | unsigned int gb_flags, | |
55f020db AH |
1233 | struct ext4_ext_path *path, |
1234 | struct ext4_extent *newext) | |
a86c6181 AT |
1235 | { |
1236 | struct ext4_ext_path *curp; | |
1237 | int depth, i, err = 0; | |
1238 | ||
1239 | repeat: | |
1240 | i = depth = ext_depth(inode); | |
1241 | ||
1242 | /* walk up to the tree and look for free index entry */ | |
1243 | curp = path + depth; | |
1244 | while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) { | |
1245 | i--; | |
1246 | curp--; | |
1247 | } | |
1248 | ||
d0d856e8 RD |
1249 | /* we use already allocated block for index block, |
1250 | * so subsequent data blocks should be contiguous */ | |
a86c6181 AT |
1251 | if (EXT_HAS_FREE_INDEX(curp)) { |
1252 | /* if we found index with free entry, then use that | |
1253 | * entry: create all needed subtree and add new leaf */ | |
107a7bd3 | 1254 | err = ext4_ext_split(handle, inode, mb_flags, path, newext, i); |
787e0981 SF |
1255 | if (err) |
1256 | goto out; | |
a86c6181 AT |
1257 | |
1258 | /* refill path */ | |
1259 | ext4_ext_drop_refs(path); | |
1260 | path = ext4_ext_find_extent(inode, | |
725d26d3 | 1261 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
107a7bd3 | 1262 | path, gb_flags); |
a86c6181 AT |
1263 | if (IS_ERR(path)) |
1264 | err = PTR_ERR(path); | |
1265 | } else { | |
1266 | /* tree is full, time to grow in depth */ | |
107a7bd3 | 1267 | err = ext4_ext_grow_indepth(handle, inode, mb_flags, newext); |
a86c6181 AT |
1268 | if (err) |
1269 | goto out; | |
1270 | ||
1271 | /* refill path */ | |
1272 | ext4_ext_drop_refs(path); | |
1273 | path = ext4_ext_find_extent(inode, | |
725d26d3 | 1274 | (ext4_lblk_t)le32_to_cpu(newext->ee_block), |
107a7bd3 | 1275 | path, gb_flags); |
a86c6181 AT |
1276 | if (IS_ERR(path)) { |
1277 | err = PTR_ERR(path); | |
1278 | goto out; | |
1279 | } | |
1280 | ||
1281 | /* | |
d0d856e8 RD |
1282 | * only first (depth 0 -> 1) produces free space; |
1283 | * in all other cases we have to split the grown tree | |
a86c6181 AT |
1284 | */ |
1285 | depth = ext_depth(inode); | |
1286 | if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) { | |
d0d856e8 | 1287 | /* now we need to split */ |
a86c6181 AT |
1288 | goto repeat; |
1289 | } | |
1290 | } | |
1291 | ||
1292 | out: | |
1293 | return err; | |
1294 | } | |
1295 | ||
1988b51e AT |
1296 | /* |
1297 | * search the closest allocated block to the left for *logical | |
1298 | * and returns it at @logical + it's physical address at @phys | |
1299 | * if *logical is the smallest allocated block, the function | |
1300 | * returns 0 at @phys | |
1301 | * return value contains 0 (success) or error code | |
1302 | */ | |
1f109d5a TT |
1303 | static int ext4_ext_search_left(struct inode *inode, |
1304 | struct ext4_ext_path *path, | |
1305 | ext4_lblk_t *logical, ext4_fsblk_t *phys) | |
1988b51e AT |
1306 | { |
1307 | struct ext4_extent_idx *ix; | |
1308 | struct ext4_extent *ex; | |
b939e376 | 1309 | int depth, ee_len; |
1988b51e | 1310 | |
273df556 FM |
1311 | if (unlikely(path == NULL)) { |
1312 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1313 | return -EIO; | |
1314 | } | |
1988b51e AT |
1315 | depth = path->p_depth; |
1316 | *phys = 0; | |
1317 | ||
1318 | if (depth == 0 && path->p_ext == NULL) | |
1319 | return 0; | |
1320 | ||
1321 | /* usually extent in the path covers blocks smaller | |
1322 | * then *logical, but it can be that extent is the | |
1323 | * first one in the file */ | |
1324 | ||
1325 | ex = path[depth].p_ext; | |
b939e376 | 1326 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1327 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1328 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1329 | EXT4_ERROR_INODE(inode, | |
1330 | "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!", | |
1331 | *logical, le32_to_cpu(ex->ee_block)); | |
1332 | return -EIO; | |
1333 | } | |
1988b51e AT |
1334 | while (--depth >= 0) { |
1335 | ix = path[depth].p_idx; | |
273df556 FM |
1336 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1337 | EXT4_ERROR_INODE(inode, | |
1338 | "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!", | |
6ee3b212 | 1339 | ix != NULL ? le32_to_cpu(ix->ei_block) : 0, |
273df556 | 1340 | EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ? |
6ee3b212 | 1341 | le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0, |
273df556 FM |
1342 | depth); |
1343 | return -EIO; | |
1344 | } | |
1988b51e AT |
1345 | } |
1346 | return 0; | |
1347 | } | |
1348 | ||
273df556 FM |
1349 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1350 | EXT4_ERROR_INODE(inode, | |
1351 | "logical %d < ee_block %d + ee_len %d!", | |
1352 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1353 | return -EIO; | |
1354 | } | |
1988b51e | 1355 | |
b939e376 | 1356 | *logical = le32_to_cpu(ex->ee_block) + ee_len - 1; |
bf89d16f | 1357 | *phys = ext4_ext_pblock(ex) + ee_len - 1; |
1988b51e AT |
1358 | return 0; |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * search the closest allocated block to the right for *logical | |
1363 | * and returns it at @logical + it's physical address at @phys | |
df3ab170 | 1364 | * if *logical is the largest allocated block, the function |
1988b51e AT |
1365 | * returns 0 at @phys |
1366 | * return value contains 0 (success) or error code | |
1367 | */ | |
1f109d5a TT |
1368 | static int ext4_ext_search_right(struct inode *inode, |
1369 | struct ext4_ext_path *path, | |
4d33b1ef TT |
1370 | ext4_lblk_t *logical, ext4_fsblk_t *phys, |
1371 | struct ext4_extent **ret_ex) | |
1988b51e AT |
1372 | { |
1373 | struct buffer_head *bh = NULL; | |
1374 | struct ext4_extent_header *eh; | |
1375 | struct ext4_extent_idx *ix; | |
1376 | struct ext4_extent *ex; | |
1377 | ext4_fsblk_t block; | |
395a87bf ES |
1378 | int depth; /* Note, NOT eh_depth; depth from top of tree */ |
1379 | int ee_len; | |
1988b51e | 1380 | |
273df556 FM |
1381 | if (unlikely(path == NULL)) { |
1382 | EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical); | |
1383 | return -EIO; | |
1384 | } | |
1988b51e AT |
1385 | depth = path->p_depth; |
1386 | *phys = 0; | |
1387 | ||
1388 | if (depth == 0 && path->p_ext == NULL) | |
1389 | return 0; | |
1390 | ||
1391 | /* usually extent in the path covers blocks smaller | |
1392 | * then *logical, but it can be that extent is the | |
1393 | * first one in the file */ | |
1394 | ||
1395 | ex = path[depth].p_ext; | |
b939e376 | 1396 | ee_len = ext4_ext_get_actual_len(ex); |
1988b51e | 1397 | if (*logical < le32_to_cpu(ex->ee_block)) { |
273df556 FM |
1398 | if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) { |
1399 | EXT4_ERROR_INODE(inode, | |
1400 | "first_extent(path[%d].p_hdr) != ex", | |
1401 | depth); | |
1402 | return -EIO; | |
1403 | } | |
1988b51e AT |
1404 | while (--depth >= 0) { |
1405 | ix = path[depth].p_idx; | |
273df556 FM |
1406 | if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) { |
1407 | EXT4_ERROR_INODE(inode, | |
1408 | "ix != EXT_FIRST_INDEX *logical %d!", | |
1409 | *logical); | |
1410 | return -EIO; | |
1411 | } | |
1988b51e | 1412 | } |
4d33b1ef | 1413 | goto found_extent; |
1988b51e AT |
1414 | } |
1415 | ||
273df556 FM |
1416 | if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) { |
1417 | EXT4_ERROR_INODE(inode, | |
1418 | "logical %d < ee_block %d + ee_len %d!", | |
1419 | *logical, le32_to_cpu(ex->ee_block), ee_len); | |
1420 | return -EIO; | |
1421 | } | |
1988b51e AT |
1422 | |
1423 | if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) { | |
1424 | /* next allocated block in this leaf */ | |
1425 | ex++; | |
4d33b1ef | 1426 | goto found_extent; |
1988b51e AT |
1427 | } |
1428 | ||
1429 | /* go up and search for index to the right */ | |
1430 | while (--depth >= 0) { | |
1431 | ix = path[depth].p_idx; | |
1432 | if (ix != EXT_LAST_INDEX(path[depth].p_hdr)) | |
25f1ee3a | 1433 | goto got_index; |
1988b51e AT |
1434 | } |
1435 | ||
25f1ee3a WF |
1436 | /* we've gone up to the root and found no index to the right */ |
1437 | return 0; | |
1988b51e | 1438 | |
25f1ee3a | 1439 | got_index: |
1988b51e AT |
1440 | /* we've found index to the right, let's |
1441 | * follow it and find the closest allocated | |
1442 | * block to the right */ | |
1443 | ix++; | |
bf89d16f | 1444 | block = ext4_idx_pblock(ix); |
1988b51e | 1445 | while (++depth < path->p_depth) { |
395a87bf | 1446 | /* subtract from p_depth to get proper eh_depth */ |
7d7ea89e | 1447 | bh = read_extent_tree_block(inode, block, |
107a7bd3 | 1448 | path->p_depth - depth, 0); |
7d7ea89e TT |
1449 | if (IS_ERR(bh)) |
1450 | return PTR_ERR(bh); | |
1451 | eh = ext_block_hdr(bh); | |
1988b51e | 1452 | ix = EXT_FIRST_INDEX(eh); |
bf89d16f | 1453 | block = ext4_idx_pblock(ix); |
1988b51e AT |
1454 | put_bh(bh); |
1455 | } | |
1456 | ||
107a7bd3 | 1457 | bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0); |
7d7ea89e TT |
1458 | if (IS_ERR(bh)) |
1459 | return PTR_ERR(bh); | |
1988b51e | 1460 | eh = ext_block_hdr(bh); |
1988b51e | 1461 | ex = EXT_FIRST_EXTENT(eh); |
4d33b1ef | 1462 | found_extent: |
1988b51e | 1463 | *logical = le32_to_cpu(ex->ee_block); |
bf89d16f | 1464 | *phys = ext4_ext_pblock(ex); |
4d33b1ef TT |
1465 | *ret_ex = ex; |
1466 | if (bh) | |
1467 | put_bh(bh); | |
1988b51e | 1468 | return 0; |
1988b51e AT |
1469 | } |
1470 | ||
a86c6181 | 1471 | /* |
d0d856e8 | 1472 | * ext4_ext_next_allocated_block: |
f17722f9 | 1473 | * returns allocated block in subsequent extent or EXT_MAX_BLOCKS. |
d0d856e8 RD |
1474 | * NOTE: it considers block number from index entry as |
1475 | * allocated block. Thus, index entries have to be consistent | |
1476 | * with leaves. | |
a86c6181 | 1477 | */ |
725d26d3 | 1478 | static ext4_lblk_t |
a86c6181 AT |
1479 | ext4_ext_next_allocated_block(struct ext4_ext_path *path) |
1480 | { | |
1481 | int depth; | |
1482 | ||
1483 | BUG_ON(path == NULL); | |
1484 | depth = path->p_depth; | |
1485 | ||
1486 | if (depth == 0 && path->p_ext == NULL) | |
f17722f9 | 1487 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1488 | |
1489 | while (depth >= 0) { | |
1490 | if (depth == path->p_depth) { | |
1491 | /* leaf */ | |
6f8ff537 CW |
1492 | if (path[depth].p_ext && |
1493 | path[depth].p_ext != | |
a86c6181 AT |
1494 | EXT_LAST_EXTENT(path[depth].p_hdr)) |
1495 | return le32_to_cpu(path[depth].p_ext[1].ee_block); | |
1496 | } else { | |
1497 | /* index */ | |
1498 | if (path[depth].p_idx != | |
1499 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
1500 | return le32_to_cpu(path[depth].p_idx[1].ei_block); | |
1501 | } | |
1502 | depth--; | |
1503 | } | |
1504 | ||
f17722f9 | 1505 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1506 | } |
1507 | ||
1508 | /* | |
d0d856e8 | 1509 | * ext4_ext_next_leaf_block: |
f17722f9 | 1510 | * returns first allocated block from next leaf or EXT_MAX_BLOCKS |
a86c6181 | 1511 | */ |
5718789d | 1512 | static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path) |
a86c6181 AT |
1513 | { |
1514 | int depth; | |
1515 | ||
1516 | BUG_ON(path == NULL); | |
1517 | depth = path->p_depth; | |
1518 | ||
1519 | /* zero-tree has no leaf blocks at all */ | |
1520 | if (depth == 0) | |
f17722f9 | 1521 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1522 | |
1523 | /* go to index block */ | |
1524 | depth--; | |
1525 | ||
1526 | while (depth >= 0) { | |
1527 | if (path[depth].p_idx != | |
1528 | EXT_LAST_INDEX(path[depth].p_hdr)) | |
725d26d3 AK |
1529 | return (ext4_lblk_t) |
1530 | le32_to_cpu(path[depth].p_idx[1].ei_block); | |
a86c6181 AT |
1531 | depth--; |
1532 | } | |
1533 | ||
f17722f9 | 1534 | return EXT_MAX_BLOCKS; |
a86c6181 AT |
1535 | } |
1536 | ||
1537 | /* | |
d0d856e8 RD |
1538 | * ext4_ext_correct_indexes: |
1539 | * if leaf gets modified and modified extent is first in the leaf, | |
1540 | * then we have to correct all indexes above. | |
a86c6181 AT |
1541 | * TODO: do we need to correct tree in all cases? |
1542 | */ | |
1d03ec98 | 1543 | static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode, |
a86c6181 AT |
1544 | struct ext4_ext_path *path) |
1545 | { | |
1546 | struct ext4_extent_header *eh; | |
1547 | int depth = ext_depth(inode); | |
1548 | struct ext4_extent *ex; | |
1549 | __le32 border; | |
1550 | int k, err = 0; | |
1551 | ||
1552 | eh = path[depth].p_hdr; | |
1553 | ex = path[depth].p_ext; | |
273df556 FM |
1554 | |
1555 | if (unlikely(ex == NULL || eh == NULL)) { | |
1556 | EXT4_ERROR_INODE(inode, | |
1557 | "ex %p == NULL or eh %p == NULL", ex, eh); | |
1558 | return -EIO; | |
1559 | } | |
a86c6181 AT |
1560 | |
1561 | if (depth == 0) { | |
1562 | /* there is no tree at all */ | |
1563 | return 0; | |
1564 | } | |
1565 | ||
1566 | if (ex != EXT_FIRST_EXTENT(eh)) { | |
1567 | /* we correct tree if first leaf got modified only */ | |
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | /* | |
d0d856e8 | 1572 | * TODO: we need correction if border is smaller than current one |
a86c6181 AT |
1573 | */ |
1574 | k = depth - 1; | |
1575 | border = path[depth].p_ext->ee_block; | |
7e028976 AM |
1576 | err = ext4_ext_get_access(handle, inode, path + k); |
1577 | if (err) | |
a86c6181 AT |
1578 | return err; |
1579 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1580 | err = ext4_ext_dirty(handle, inode, path + k); |
1581 | if (err) | |
a86c6181 AT |
1582 | return err; |
1583 | ||
1584 | while (k--) { | |
1585 | /* change all left-side indexes */ | |
1586 | if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr)) | |
1587 | break; | |
7e028976 AM |
1588 | err = ext4_ext_get_access(handle, inode, path + k); |
1589 | if (err) | |
a86c6181 AT |
1590 | break; |
1591 | path[k].p_idx->ei_block = border; | |
7e028976 AM |
1592 | err = ext4_ext_dirty(handle, inode, path + k); |
1593 | if (err) | |
a86c6181 AT |
1594 | break; |
1595 | } | |
1596 | ||
1597 | return err; | |
1598 | } | |
1599 | ||
748de673 | 1600 | int |
a86c6181 AT |
1601 | ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1, |
1602 | struct ext4_extent *ex2) | |
1603 | { | |
749269fa | 1604 | unsigned short ext1_ee_len, ext2_ee_len, max_len; |
a2df2a63 AA |
1605 | |
1606 | /* | |
ec22ba8e DM |
1607 | * Make sure that both extents are initialized. We don't merge |
1608 | * uninitialized extents so that we can be sure that end_io code has | |
1609 | * the extent that was written properly split out and conversion to | |
1610 | * initialized is trivial. | |
a2df2a63 | 1611 | */ |
ec22ba8e | 1612 | if (ext4_ext_is_uninitialized(ex1) || ext4_ext_is_uninitialized(ex2)) |
a2df2a63 AA |
1613 | return 0; |
1614 | ||
749269fa AA |
1615 | if (ext4_ext_is_uninitialized(ex1)) |
1616 | max_len = EXT_UNINIT_MAX_LEN; | |
1617 | else | |
1618 | max_len = EXT_INIT_MAX_LEN; | |
1619 | ||
a2df2a63 AA |
1620 | ext1_ee_len = ext4_ext_get_actual_len(ex1); |
1621 | ext2_ee_len = ext4_ext_get_actual_len(ex2); | |
1622 | ||
1623 | if (le32_to_cpu(ex1->ee_block) + ext1_ee_len != | |
63f57933 | 1624 | le32_to_cpu(ex2->ee_block)) |
a86c6181 AT |
1625 | return 0; |
1626 | ||
471d4011 SB |
1627 | /* |
1628 | * To allow future support for preallocated extents to be added | |
1629 | * as an RO_COMPAT feature, refuse to merge to extents if | |
d0d856e8 | 1630 | * this can result in the top bit of ee_len being set. |
471d4011 | 1631 | */ |
749269fa | 1632 | if (ext1_ee_len + ext2_ee_len > max_len) |
471d4011 | 1633 | return 0; |
bbf2f9fb | 1634 | #ifdef AGGRESSIVE_TEST |
b939e376 | 1635 | if (ext1_ee_len >= 4) |
a86c6181 AT |
1636 | return 0; |
1637 | #endif | |
1638 | ||
bf89d16f | 1639 | if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2)) |
a86c6181 AT |
1640 | return 1; |
1641 | return 0; | |
1642 | } | |
1643 | ||
56055d3a AA |
1644 | /* |
1645 | * This function tries to merge the "ex" extent to the next extent in the tree. | |
1646 | * It always tries to merge towards right. If you want to merge towards | |
1647 | * left, pass "ex - 1" as argument instead of "ex". | |
1648 | * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns | |
1649 | * 1 if they got merged. | |
1650 | */ | |
197217a5 | 1651 | static int ext4_ext_try_to_merge_right(struct inode *inode, |
1f109d5a TT |
1652 | struct ext4_ext_path *path, |
1653 | struct ext4_extent *ex) | |
56055d3a AA |
1654 | { |
1655 | struct ext4_extent_header *eh; | |
1656 | unsigned int depth, len; | |
1657 | int merge_done = 0; | |
1658 | int uninitialized = 0; | |
1659 | ||
1660 | depth = ext_depth(inode); | |
1661 | BUG_ON(path[depth].p_hdr == NULL); | |
1662 | eh = path[depth].p_hdr; | |
1663 | ||
1664 | while (ex < EXT_LAST_EXTENT(eh)) { | |
1665 | if (!ext4_can_extents_be_merged(inode, ex, ex + 1)) | |
1666 | break; | |
1667 | /* merge with next extent! */ | |
1668 | if (ext4_ext_is_uninitialized(ex)) | |
1669 | uninitialized = 1; | |
1670 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1671 | + ext4_ext_get_actual_len(ex + 1)); | |
1672 | if (uninitialized) | |
1673 | ext4_ext_mark_uninitialized(ex); | |
1674 | ||
1675 | if (ex + 1 < EXT_LAST_EXTENT(eh)) { | |
1676 | len = (EXT_LAST_EXTENT(eh) - ex - 1) | |
1677 | * sizeof(struct ext4_extent); | |
1678 | memmove(ex + 1, ex + 2, len); | |
1679 | } | |
e8546d06 | 1680 | le16_add_cpu(&eh->eh_entries, -1); |
56055d3a AA |
1681 | merge_done = 1; |
1682 | WARN_ON(eh->eh_entries == 0); | |
1683 | if (!eh->eh_entries) | |
24676da4 | 1684 | EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!"); |
56055d3a AA |
1685 | } |
1686 | ||
1687 | return merge_done; | |
1688 | } | |
1689 | ||
ecb94f5f TT |
1690 | /* |
1691 | * This function does a very simple check to see if we can collapse | |
1692 | * an extent tree with a single extent tree leaf block into the inode. | |
1693 | */ | |
1694 | static void ext4_ext_try_to_merge_up(handle_t *handle, | |
1695 | struct inode *inode, | |
1696 | struct ext4_ext_path *path) | |
1697 | { | |
1698 | size_t s; | |
1699 | unsigned max_root = ext4_ext_space_root(inode, 0); | |
1700 | ext4_fsblk_t blk; | |
1701 | ||
1702 | if ((path[0].p_depth != 1) || | |
1703 | (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) || | |
1704 | (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root)) | |
1705 | return; | |
1706 | ||
1707 | /* | |
1708 | * We need to modify the block allocation bitmap and the block | |
1709 | * group descriptor to release the extent tree block. If we | |
1710 | * can't get the journal credits, give up. | |
1711 | */ | |
1712 | if (ext4_journal_extend(handle, 2)) | |
1713 | return; | |
1714 | ||
1715 | /* | |
1716 | * Copy the extent data up to the inode | |
1717 | */ | |
1718 | blk = ext4_idx_pblock(path[0].p_idx); | |
1719 | s = le16_to_cpu(path[1].p_hdr->eh_entries) * | |
1720 | sizeof(struct ext4_extent_idx); | |
1721 | s += sizeof(struct ext4_extent_header); | |
1722 | ||
1723 | memcpy(path[0].p_hdr, path[1].p_hdr, s); | |
1724 | path[0].p_depth = 0; | |
1725 | path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) + | |
1726 | (path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr)); | |
1727 | path[0].p_hdr->eh_max = cpu_to_le16(max_root); | |
1728 | ||
1729 | brelse(path[1].p_bh); | |
1730 | ext4_free_blocks(handle, inode, NULL, blk, 1, | |
1731 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); | |
1732 | } | |
1733 | ||
197217a5 YY |
1734 | /* |
1735 | * This function tries to merge the @ex extent to neighbours in the tree. | |
1736 | * return 1 if merge left else 0. | |
1737 | */ | |
ecb94f5f TT |
1738 | static void ext4_ext_try_to_merge(handle_t *handle, |
1739 | struct inode *inode, | |
197217a5 YY |
1740 | struct ext4_ext_path *path, |
1741 | struct ext4_extent *ex) { | |
1742 | struct ext4_extent_header *eh; | |
1743 | unsigned int depth; | |
1744 | int merge_done = 0; | |
197217a5 YY |
1745 | |
1746 | depth = ext_depth(inode); | |
1747 | BUG_ON(path[depth].p_hdr == NULL); | |
1748 | eh = path[depth].p_hdr; | |
1749 | ||
1750 | if (ex > EXT_FIRST_EXTENT(eh)) | |
1751 | merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1); | |
1752 | ||
1753 | if (!merge_done) | |
ecb94f5f | 1754 | (void) ext4_ext_try_to_merge_right(inode, path, ex); |
197217a5 | 1755 | |
ecb94f5f | 1756 | ext4_ext_try_to_merge_up(handle, inode, path); |
197217a5 YY |
1757 | } |
1758 | ||
25d14f98 AA |
1759 | /* |
1760 | * check if a portion of the "newext" extent overlaps with an | |
1761 | * existing extent. | |
1762 | * | |
1763 | * If there is an overlap discovered, it updates the length of the newext | |
1764 | * such that there will be no overlap, and then returns 1. | |
1765 | * If there is no overlap found, it returns 0. | |
1766 | */ | |
4d33b1ef TT |
1767 | static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi, |
1768 | struct inode *inode, | |
1f109d5a TT |
1769 | struct ext4_extent *newext, |
1770 | struct ext4_ext_path *path) | |
25d14f98 | 1771 | { |
725d26d3 | 1772 | ext4_lblk_t b1, b2; |
25d14f98 AA |
1773 | unsigned int depth, len1; |
1774 | unsigned int ret = 0; | |
1775 | ||
1776 | b1 = le32_to_cpu(newext->ee_block); | |
a2df2a63 | 1777 | len1 = ext4_ext_get_actual_len(newext); |
25d14f98 AA |
1778 | depth = ext_depth(inode); |
1779 | if (!path[depth].p_ext) | |
1780 | goto out; | |
1781 | b2 = le32_to_cpu(path[depth].p_ext->ee_block); | |
4d33b1ef | 1782 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1783 | |
1784 | /* | |
1785 | * get the next allocated block if the extent in the path | |
2b2d6d01 | 1786 | * is before the requested block(s) |
25d14f98 AA |
1787 | */ |
1788 | if (b2 < b1) { | |
1789 | b2 = ext4_ext_next_allocated_block(path); | |
f17722f9 | 1790 | if (b2 == EXT_MAX_BLOCKS) |
25d14f98 | 1791 | goto out; |
4d33b1ef | 1792 | b2 &= ~(sbi->s_cluster_ratio - 1); |
25d14f98 AA |
1793 | } |
1794 | ||
725d26d3 | 1795 | /* check for wrap through zero on extent logical start block*/ |
25d14f98 | 1796 | if (b1 + len1 < b1) { |
f17722f9 | 1797 | len1 = EXT_MAX_BLOCKS - b1; |
25d14f98 AA |
1798 | newext->ee_len = cpu_to_le16(len1); |
1799 | ret = 1; | |
1800 | } | |
1801 | ||
1802 | /* check for overlap */ | |
1803 | if (b1 + len1 > b2) { | |
1804 | newext->ee_len = cpu_to_le16(b2 - b1); | |
1805 | ret = 1; | |
1806 | } | |
1807 | out: | |
1808 | return ret; | |
1809 | } | |
1810 | ||
a86c6181 | 1811 | /* |
d0d856e8 RD |
1812 | * ext4_ext_insert_extent: |
1813 | * tries to merge requsted extent into the existing extent or | |
1814 | * inserts requested extent as new one into the tree, | |
1815 | * creating new leaf in the no-space case. | |
a86c6181 AT |
1816 | */ |
1817 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |
1818 | struct ext4_ext_path *path, | |
107a7bd3 | 1819 | struct ext4_extent *newext, int gb_flags) |
a86c6181 | 1820 | { |
af5bc92d | 1821 | struct ext4_extent_header *eh; |
a86c6181 AT |
1822 | struct ext4_extent *ex, *fex; |
1823 | struct ext4_extent *nearex; /* nearest extent */ | |
1824 | struct ext4_ext_path *npath = NULL; | |
725d26d3 AK |
1825 | int depth, len, err; |
1826 | ext4_lblk_t next; | |
a2df2a63 | 1827 | unsigned uninitialized = 0; |
107a7bd3 | 1828 | int mb_flags = 0; |
a86c6181 | 1829 | |
273df556 FM |
1830 | if (unlikely(ext4_ext_get_actual_len(newext) == 0)) { |
1831 | EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0"); | |
1832 | return -EIO; | |
1833 | } | |
a86c6181 AT |
1834 | depth = ext_depth(inode); |
1835 | ex = path[depth].p_ext; | |
be8981be | 1836 | eh = path[depth].p_hdr; |
273df556 FM |
1837 | if (unlikely(path[depth].p_hdr == NULL)) { |
1838 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
1839 | return -EIO; | |
1840 | } | |
a86c6181 AT |
1841 | |
1842 | /* try to insert block into found extent and return */ | |
107a7bd3 | 1843 | if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) { |
a2df2a63 AA |
1844 | |
1845 | /* | |
be8981be LC |
1846 | * Try to see whether we should rather test the extent on |
1847 | * right from ex, or from the left of ex. This is because | |
1848 | * ext4_ext_find_extent() can return either extent on the | |
1849 | * left, or on the right from the searched position. This | |
1850 | * will make merging more effective. | |
a2df2a63 | 1851 | */ |
be8981be LC |
1852 | if (ex < EXT_LAST_EXTENT(eh) && |
1853 | (le32_to_cpu(ex->ee_block) + | |
1854 | ext4_ext_get_actual_len(ex) < | |
1855 | le32_to_cpu(newext->ee_block))) { | |
1856 | ex += 1; | |
1857 | goto prepend; | |
1858 | } else if ((ex > EXT_FIRST_EXTENT(eh)) && | |
1859 | (le32_to_cpu(newext->ee_block) + | |
1860 | ext4_ext_get_actual_len(newext) < | |
1861 | le32_to_cpu(ex->ee_block))) | |
1862 | ex -= 1; | |
1863 | ||
1864 | /* Try to append newex to the ex */ | |
1865 | if (ext4_can_extents_be_merged(inode, ex, newext)) { | |
1866 | ext_debug("append [%d]%d block to %u:[%d]%d" | |
1867 | "(from %llu)\n", | |
1868 | ext4_ext_is_uninitialized(newext), | |
1869 | ext4_ext_get_actual_len(newext), | |
1870 | le32_to_cpu(ex->ee_block), | |
1871 | ext4_ext_is_uninitialized(ex), | |
1872 | ext4_ext_get_actual_len(ex), | |
1873 | ext4_ext_pblock(ex)); | |
1874 | err = ext4_ext_get_access(handle, inode, | |
1875 | path + depth); | |
1876 | if (err) | |
1877 | return err; | |
1878 | ||
1879 | /* | |
1880 | * ext4_can_extents_be_merged should have checked | |
1881 | * that either both extents are uninitialized, or | |
1882 | * both aren't. Thus we need to check only one of | |
1883 | * them here. | |
1884 | */ | |
1885 | if (ext4_ext_is_uninitialized(ex)) | |
1886 | uninitialized = 1; | |
1887 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
a2df2a63 | 1888 | + ext4_ext_get_actual_len(newext)); |
be8981be LC |
1889 | if (uninitialized) |
1890 | ext4_ext_mark_uninitialized(ex); | |
1891 | eh = path[depth].p_hdr; | |
1892 | nearex = ex; | |
1893 | goto merge; | |
1894 | } | |
1895 | ||
1896 | prepend: | |
1897 | /* Try to prepend newex to the ex */ | |
1898 | if (ext4_can_extents_be_merged(inode, newext, ex)) { | |
1899 | ext_debug("prepend %u[%d]%d block to %u:[%d]%d" | |
1900 | "(from %llu)\n", | |
1901 | le32_to_cpu(newext->ee_block), | |
1902 | ext4_ext_is_uninitialized(newext), | |
1903 | ext4_ext_get_actual_len(newext), | |
1904 | le32_to_cpu(ex->ee_block), | |
1905 | ext4_ext_is_uninitialized(ex), | |
1906 | ext4_ext_get_actual_len(ex), | |
1907 | ext4_ext_pblock(ex)); | |
1908 | err = ext4_ext_get_access(handle, inode, | |
1909 | path + depth); | |
1910 | if (err) | |
1911 | return err; | |
1912 | ||
1913 | /* | |
1914 | * ext4_can_extents_be_merged should have checked | |
1915 | * that either both extents are uninitialized, or | |
1916 | * both aren't. Thus we need to check only one of | |
1917 | * them here. | |
1918 | */ | |
1919 | if (ext4_ext_is_uninitialized(ex)) | |
1920 | uninitialized = 1; | |
1921 | ex->ee_block = newext->ee_block; | |
1922 | ext4_ext_store_pblock(ex, ext4_ext_pblock(newext)); | |
1923 | ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex) | |
1924 | + ext4_ext_get_actual_len(newext)); | |
1925 | if (uninitialized) | |
1926 | ext4_ext_mark_uninitialized(ex); | |
1927 | eh = path[depth].p_hdr; | |
1928 | nearex = ex; | |
1929 | goto merge; | |
1930 | } | |
a86c6181 AT |
1931 | } |
1932 | ||
a86c6181 AT |
1933 | depth = ext_depth(inode); |
1934 | eh = path[depth].p_hdr; | |
1935 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) | |
1936 | goto has_space; | |
1937 | ||
1938 | /* probably next leaf has space for us? */ | |
1939 | fex = EXT_LAST_EXTENT(eh); | |
598dbdf2 RD |
1940 | next = EXT_MAX_BLOCKS; |
1941 | if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)) | |
5718789d | 1942 | next = ext4_ext_next_leaf_block(path); |
598dbdf2 | 1943 | if (next != EXT_MAX_BLOCKS) { |
32de6756 | 1944 | ext_debug("next leaf block - %u\n", next); |
a86c6181 | 1945 | BUG_ON(npath != NULL); |
107a7bd3 | 1946 | npath = ext4_ext_find_extent(inode, next, NULL, 0); |
a86c6181 AT |
1947 | if (IS_ERR(npath)) |
1948 | return PTR_ERR(npath); | |
1949 | BUG_ON(npath->p_depth != path->p_depth); | |
1950 | eh = npath[depth].p_hdr; | |
1951 | if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) { | |
25985edc | 1952 | ext_debug("next leaf isn't full(%d)\n", |
a86c6181 AT |
1953 | le16_to_cpu(eh->eh_entries)); |
1954 | path = npath; | |
ffb505ff | 1955 | goto has_space; |
a86c6181 AT |
1956 | } |
1957 | ext_debug("next leaf has no free space(%d,%d)\n", | |
1958 | le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max)); | |
1959 | } | |
1960 | ||
1961 | /* | |
d0d856e8 RD |
1962 | * There is no free space in the found leaf. |
1963 | * We're gonna add a new leaf in the tree. | |
a86c6181 | 1964 | */ |
107a7bd3 TT |
1965 | if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL) |
1966 | mb_flags = EXT4_MB_USE_RESERVED; | |
1967 | err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags, | |
1968 | path, newext); | |
a86c6181 AT |
1969 | if (err) |
1970 | goto cleanup; | |
1971 | depth = ext_depth(inode); | |
1972 | eh = path[depth].p_hdr; | |
1973 | ||
1974 | has_space: | |
1975 | nearex = path[depth].p_ext; | |
1976 | ||
7e028976 AM |
1977 | err = ext4_ext_get_access(handle, inode, path + depth); |
1978 | if (err) | |
a86c6181 AT |
1979 | goto cleanup; |
1980 | ||
1981 | if (!nearex) { | |
1982 | /* there is no extent in this leaf, create first one */ | |
32de6756 | 1983 | ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n", |
8c55e204 | 1984 | le32_to_cpu(newext->ee_block), |
bf89d16f | 1985 | ext4_ext_pblock(newext), |
553f9008 | 1986 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 1987 | ext4_ext_get_actual_len(newext)); |
80e675f9 EG |
1988 | nearex = EXT_FIRST_EXTENT(eh); |
1989 | } else { | |
1990 | if (le32_to_cpu(newext->ee_block) | |
8c55e204 | 1991 | > le32_to_cpu(nearex->ee_block)) { |
80e675f9 | 1992 | /* Insert after */ |
32de6756 YY |
1993 | ext_debug("insert %u:%llu:[%d]%d before: " |
1994 | "nearest %p\n", | |
80e675f9 EG |
1995 | le32_to_cpu(newext->ee_block), |
1996 | ext4_ext_pblock(newext), | |
1997 | ext4_ext_is_uninitialized(newext), | |
1998 | ext4_ext_get_actual_len(newext), | |
1999 | nearex); | |
2000 | nearex++; | |
2001 | } else { | |
2002 | /* Insert before */ | |
2003 | BUG_ON(newext->ee_block == nearex->ee_block); | |
32de6756 YY |
2004 | ext_debug("insert %u:%llu:[%d]%d after: " |
2005 | "nearest %p\n", | |
8c55e204 | 2006 | le32_to_cpu(newext->ee_block), |
bf89d16f | 2007 | ext4_ext_pblock(newext), |
553f9008 | 2008 | ext4_ext_is_uninitialized(newext), |
a2df2a63 | 2009 | ext4_ext_get_actual_len(newext), |
80e675f9 EG |
2010 | nearex); |
2011 | } | |
2012 | len = EXT_LAST_EXTENT(eh) - nearex + 1; | |
2013 | if (len > 0) { | |
32de6756 | 2014 | ext_debug("insert %u:%llu:[%d]%d: " |
80e675f9 EG |
2015 | "move %d extents from 0x%p to 0x%p\n", |
2016 | le32_to_cpu(newext->ee_block), | |
2017 | ext4_ext_pblock(newext), | |
2018 | ext4_ext_is_uninitialized(newext), | |
2019 | ext4_ext_get_actual_len(newext), | |
2020 | len, nearex, nearex + 1); | |
2021 | memmove(nearex + 1, nearex, | |
2022 | len * sizeof(struct ext4_extent)); | |
a86c6181 | 2023 | } |
a86c6181 AT |
2024 | } |
2025 | ||
e8546d06 | 2026 | le16_add_cpu(&eh->eh_entries, 1); |
80e675f9 | 2027 | path[depth].p_ext = nearex; |
a86c6181 | 2028 | nearex->ee_block = newext->ee_block; |
bf89d16f | 2029 | ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext)); |
a86c6181 | 2030 | nearex->ee_len = newext->ee_len; |
a86c6181 AT |
2031 | |
2032 | merge: | |
e7bcf823 | 2033 | /* try to merge extents */ |
107a7bd3 | 2034 | if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) |
ecb94f5f | 2035 | ext4_ext_try_to_merge(handle, inode, path, nearex); |
a86c6181 | 2036 | |
a86c6181 AT |
2037 | |
2038 | /* time to correct all indexes above */ | |
2039 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2040 | if (err) | |
2041 | goto cleanup; | |
2042 | ||
ecb94f5f | 2043 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
a86c6181 AT |
2044 | |
2045 | cleanup: | |
2046 | if (npath) { | |
2047 | ext4_ext_drop_refs(npath); | |
2048 | kfree(npath); | |
2049 | } | |
a86c6181 AT |
2050 | return err; |
2051 | } | |
2052 | ||
91dd8c11 LC |
2053 | static int ext4_fill_fiemap_extents(struct inode *inode, |
2054 | ext4_lblk_t block, ext4_lblk_t num, | |
2055 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
2056 | { |
2057 | struct ext4_ext_path *path = NULL; | |
6873fa0d | 2058 | struct ext4_extent *ex; |
69eb33dc | 2059 | struct extent_status es; |
91dd8c11 | 2060 | ext4_lblk_t next, next_del, start = 0, end = 0; |
6873fa0d | 2061 | ext4_lblk_t last = block + num; |
91dd8c11 LC |
2062 | int exists, depth = 0, err = 0; |
2063 | unsigned int flags = 0; | |
2064 | unsigned char blksize_bits = inode->i_sb->s_blocksize_bits; | |
6873fa0d | 2065 | |
f17722f9 | 2066 | while (block < last && block != EXT_MAX_BLOCKS) { |
6873fa0d ES |
2067 | num = last - block; |
2068 | /* find extent for this block */ | |
fab3a549 | 2069 | down_read(&EXT4_I(inode)->i_data_sem); |
91dd8c11 LC |
2070 | |
2071 | if (path && ext_depth(inode) != depth) { | |
2072 | /* depth was changed. we have to realloc path */ | |
2073 | kfree(path); | |
2074 | path = NULL; | |
2075 | } | |
2076 | ||
107a7bd3 | 2077 | path = ext4_ext_find_extent(inode, block, path, 0); |
6873fa0d | 2078 | if (IS_ERR(path)) { |
91dd8c11 | 2079 | up_read(&EXT4_I(inode)->i_data_sem); |
6873fa0d ES |
2080 | err = PTR_ERR(path); |
2081 | path = NULL; | |
2082 | break; | |
2083 | } | |
2084 | ||
2085 | depth = ext_depth(inode); | |
273df556 | 2086 | if (unlikely(path[depth].p_hdr == NULL)) { |
91dd8c11 | 2087 | up_read(&EXT4_I(inode)->i_data_sem); |
273df556 FM |
2088 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); |
2089 | err = -EIO; | |
2090 | break; | |
2091 | } | |
6873fa0d ES |
2092 | ex = path[depth].p_ext; |
2093 | next = ext4_ext_next_allocated_block(path); | |
91dd8c11 | 2094 | ext4_ext_drop_refs(path); |
6873fa0d | 2095 | |
91dd8c11 | 2096 | flags = 0; |
6873fa0d ES |
2097 | exists = 0; |
2098 | if (!ex) { | |
2099 | /* there is no extent yet, so try to allocate | |
2100 | * all requested space */ | |
2101 | start = block; | |
2102 | end = block + num; | |
2103 | } else if (le32_to_cpu(ex->ee_block) > block) { | |
2104 | /* need to allocate space before found extent */ | |
2105 | start = block; | |
2106 | end = le32_to_cpu(ex->ee_block); | |
2107 | if (block + num < end) | |
2108 | end = block + num; | |
2109 | } else if (block >= le32_to_cpu(ex->ee_block) | |
2110 | + ext4_ext_get_actual_len(ex)) { | |
2111 | /* need to allocate space after found extent */ | |
2112 | start = block; | |
2113 | end = block + num; | |
2114 | if (end >= next) | |
2115 | end = next; | |
2116 | } else if (block >= le32_to_cpu(ex->ee_block)) { | |
2117 | /* | |
2118 | * some part of requested space is covered | |
2119 | * by found extent | |
2120 | */ | |
2121 | start = block; | |
2122 | end = le32_to_cpu(ex->ee_block) | |
2123 | + ext4_ext_get_actual_len(ex); | |
2124 | if (block + num < end) | |
2125 | end = block + num; | |
2126 | exists = 1; | |
2127 | } else { | |
2128 | BUG(); | |
2129 | } | |
2130 | BUG_ON(end <= start); | |
2131 | ||
2132 | if (!exists) { | |
69eb33dc ZL |
2133 | es.es_lblk = start; |
2134 | es.es_len = end - start; | |
2135 | es.es_pblk = 0; | |
6873fa0d | 2136 | } else { |
69eb33dc ZL |
2137 | es.es_lblk = le32_to_cpu(ex->ee_block); |
2138 | es.es_len = ext4_ext_get_actual_len(ex); | |
2139 | es.es_pblk = ext4_ext_pblock(ex); | |
91dd8c11 LC |
2140 | if (ext4_ext_is_uninitialized(ex)) |
2141 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
6873fa0d ES |
2142 | } |
2143 | ||
91dd8c11 | 2144 | /* |
69eb33dc ZL |
2145 | * Find delayed extent and update es accordingly. We call |
2146 | * it even in !exists case to find out whether es is the | |
91dd8c11 LC |
2147 | * last existing extent or not. |
2148 | */ | |
69eb33dc | 2149 | next_del = ext4_find_delayed_extent(inode, &es); |
91dd8c11 LC |
2150 | if (!exists && next_del) { |
2151 | exists = 1; | |
72dac95d JL |
2152 | flags |= (FIEMAP_EXTENT_DELALLOC | |
2153 | FIEMAP_EXTENT_UNKNOWN); | |
91dd8c11 LC |
2154 | } |
2155 | up_read(&EXT4_I(inode)->i_data_sem); | |
2156 | ||
69eb33dc ZL |
2157 | if (unlikely(es.es_len == 0)) { |
2158 | EXT4_ERROR_INODE(inode, "es.es_len == 0"); | |
273df556 FM |
2159 | err = -EIO; |
2160 | break; | |
2161 | } | |
6873fa0d | 2162 | |
f7fec032 ZL |
2163 | /* |
2164 | * This is possible iff next == next_del == EXT_MAX_BLOCKS. | |
2165 | * we need to check next == EXT_MAX_BLOCKS because it is | |
2166 | * possible that an extent is with unwritten and delayed | |
2167 | * status due to when an extent is delayed allocated and | |
2168 | * is allocated by fallocate status tree will track both of | |
2169 | * them in a extent. | |
2170 | * | |
2171 | * So we could return a unwritten and delayed extent, and | |
2172 | * its block is equal to 'next'. | |
2173 | */ | |
2174 | if (next == next_del && next == EXT_MAX_BLOCKS) { | |
91dd8c11 LC |
2175 | flags |= FIEMAP_EXTENT_LAST; |
2176 | if (unlikely(next_del != EXT_MAX_BLOCKS || | |
2177 | next != EXT_MAX_BLOCKS)) { | |
2178 | EXT4_ERROR_INODE(inode, | |
2179 | "next extent == %u, next " | |
2180 | "delalloc extent = %u", | |
2181 | next, next_del); | |
2182 | err = -EIO; | |
2183 | break; | |
2184 | } | |
6873fa0d ES |
2185 | } |
2186 | ||
91dd8c11 LC |
2187 | if (exists) { |
2188 | err = fiemap_fill_next_extent(fieinfo, | |
69eb33dc ZL |
2189 | (__u64)es.es_lblk << blksize_bits, |
2190 | (__u64)es.es_pblk << blksize_bits, | |
2191 | (__u64)es.es_len << blksize_bits, | |
91dd8c11 LC |
2192 | flags); |
2193 | if (err < 0) | |
2194 | break; | |
2195 | if (err == 1) { | |
2196 | err = 0; | |
2197 | break; | |
2198 | } | |
6873fa0d ES |
2199 | } |
2200 | ||
69eb33dc | 2201 | block = es.es_lblk + es.es_len; |
6873fa0d ES |
2202 | } |
2203 | ||
2204 | if (path) { | |
2205 | ext4_ext_drop_refs(path); | |
2206 | kfree(path); | |
2207 | } | |
2208 | ||
2209 | return err; | |
2210 | } | |
2211 | ||
a86c6181 | 2212 | /* |
d0d856e8 RD |
2213 | * ext4_ext_put_gap_in_cache: |
2214 | * calculate boundaries of the gap that the requested block fits into | |
a86c6181 AT |
2215 | * and cache this gap |
2216 | */ | |
09b88252 | 2217 | static void |
a86c6181 | 2218 | ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path, |
725d26d3 | 2219 | ext4_lblk_t block) |
a86c6181 AT |
2220 | { |
2221 | int depth = ext_depth(inode); | |
725d26d3 AK |
2222 | unsigned long len; |
2223 | ext4_lblk_t lblock; | |
a86c6181 AT |
2224 | struct ext4_extent *ex; |
2225 | ||
2226 | ex = path[depth].p_ext; | |
2227 | if (ex == NULL) { | |
69eb33dc ZL |
2228 | /* |
2229 | * there is no extent yet, so gap is [0;-] and we | |
2230 | * don't cache it | |
2231 | */ | |
a86c6181 AT |
2232 | ext_debug("cache gap(whole file):"); |
2233 | } else if (block < le32_to_cpu(ex->ee_block)) { | |
2234 | lblock = block; | |
2235 | len = le32_to_cpu(ex->ee_block) - block; | |
bba90743 ES |
2236 | ext_debug("cache gap(before): %u [%u:%u]", |
2237 | block, | |
2238 | le32_to_cpu(ex->ee_block), | |
2239 | ext4_ext_get_actual_len(ex)); | |
d100eef2 ZL |
2240 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2241 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2242 | EXTENT_STATUS_HOLE); | |
a86c6181 | 2243 | } else if (block >= le32_to_cpu(ex->ee_block) |
a2df2a63 | 2244 | + ext4_ext_get_actual_len(ex)) { |
725d26d3 | 2245 | ext4_lblk_t next; |
8c55e204 | 2246 | lblock = le32_to_cpu(ex->ee_block) |
a2df2a63 | 2247 | + ext4_ext_get_actual_len(ex); |
725d26d3 AK |
2248 | |
2249 | next = ext4_ext_next_allocated_block(path); | |
bba90743 ES |
2250 | ext_debug("cache gap(after): [%u:%u] %u", |
2251 | le32_to_cpu(ex->ee_block), | |
2252 | ext4_ext_get_actual_len(ex), | |
2253 | block); | |
725d26d3 AK |
2254 | BUG_ON(next == lblock); |
2255 | len = next - lblock; | |
d100eef2 ZL |
2256 | if (!ext4_find_delalloc_range(inode, lblock, lblock + len - 1)) |
2257 | ext4_es_insert_extent(inode, lblock, len, ~0, | |
2258 | EXTENT_STATUS_HOLE); | |
a86c6181 AT |
2259 | } else { |
2260 | lblock = len = 0; | |
2261 | BUG(); | |
2262 | } | |
2263 | ||
bba90743 | 2264 | ext_debug(" -> %u:%lu\n", lblock, len); |
a86c6181 AT |
2265 | } |
2266 | ||
2267 | /* | |
d0d856e8 RD |
2268 | * ext4_ext_rm_idx: |
2269 | * removes index from the index block. | |
a86c6181 | 2270 | */ |
1d03ec98 | 2271 | static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode, |
c36575e6 | 2272 | struct ext4_ext_path *path, int depth) |
a86c6181 | 2273 | { |
a86c6181 | 2274 | int err; |
f65e6fba | 2275 | ext4_fsblk_t leaf; |
a86c6181 AT |
2276 | |
2277 | /* free index block */ | |
c36575e6 FL |
2278 | depth--; |
2279 | path = path + depth; | |
bf89d16f | 2280 | leaf = ext4_idx_pblock(path->p_idx); |
273df556 FM |
2281 | if (unlikely(path->p_hdr->eh_entries == 0)) { |
2282 | EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0"); | |
2283 | return -EIO; | |
2284 | } | |
7e028976 AM |
2285 | err = ext4_ext_get_access(handle, inode, path); |
2286 | if (err) | |
a86c6181 | 2287 | return err; |
0e1147b0 RD |
2288 | |
2289 | if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) { | |
2290 | int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx; | |
2291 | len *= sizeof(struct ext4_extent_idx); | |
2292 | memmove(path->p_idx, path->p_idx + 1, len); | |
2293 | } | |
2294 | ||
e8546d06 | 2295 | le16_add_cpu(&path->p_hdr->eh_entries, -1); |
7e028976 AM |
2296 | err = ext4_ext_dirty(handle, inode, path); |
2297 | if (err) | |
a86c6181 | 2298 | return err; |
2ae02107 | 2299 | ext_debug("index is empty, remove it, free block %llu\n", leaf); |
d8990240 AK |
2300 | trace_ext4_ext_rm_idx(inode, leaf); |
2301 | ||
7dc57615 | 2302 | ext4_free_blocks(handle, inode, NULL, leaf, 1, |
e6362609 | 2303 | EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET); |
c36575e6 FL |
2304 | |
2305 | while (--depth >= 0) { | |
2306 | if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr)) | |
2307 | break; | |
2308 | path--; | |
2309 | err = ext4_ext_get_access(handle, inode, path); | |
2310 | if (err) | |
2311 | break; | |
2312 | path->p_idx->ei_block = (path+1)->p_idx->ei_block; | |
2313 | err = ext4_ext_dirty(handle, inode, path); | |
2314 | if (err) | |
2315 | break; | |
2316 | } | |
a86c6181 AT |
2317 | return err; |
2318 | } | |
2319 | ||
2320 | /* | |
ee12b630 MC |
2321 | * ext4_ext_calc_credits_for_single_extent: |
2322 | * This routine returns max. credits that needed to insert an extent | |
2323 | * to the extent tree. | |
2324 | * When pass the actual path, the caller should calculate credits | |
2325 | * under i_data_sem. | |
a86c6181 | 2326 | */ |
525f4ed8 | 2327 | int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks, |
a86c6181 AT |
2328 | struct ext4_ext_path *path) |
2329 | { | |
a86c6181 | 2330 | if (path) { |
ee12b630 | 2331 | int depth = ext_depth(inode); |
f3bd1f3f | 2332 | int ret = 0; |
ee12b630 | 2333 | |
a86c6181 | 2334 | /* probably there is space in leaf? */ |
a86c6181 | 2335 | if (le16_to_cpu(path[depth].p_hdr->eh_entries) |
ee12b630 | 2336 | < le16_to_cpu(path[depth].p_hdr->eh_max)) { |
a86c6181 | 2337 | |
ee12b630 MC |
2338 | /* |
2339 | * There are some space in the leaf tree, no | |
2340 | * need to account for leaf block credit | |
2341 | * | |
2342 | * bitmaps and block group descriptor blocks | |
df3ab170 | 2343 | * and other metadata blocks still need to be |
ee12b630 MC |
2344 | * accounted. |
2345 | */ | |
525f4ed8 | 2346 | /* 1 bitmap, 1 block group descriptor */ |
ee12b630 | 2347 | ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); |
5887e98b | 2348 | return ret; |
ee12b630 MC |
2349 | } |
2350 | } | |
a86c6181 | 2351 | |
525f4ed8 | 2352 | return ext4_chunk_trans_blocks(inode, nrblocks); |
ee12b630 | 2353 | } |
a86c6181 | 2354 | |
ee12b630 | 2355 | /* |
fffb2739 | 2356 | * How many index/leaf blocks need to change/allocate to add @extents extents? |
ee12b630 | 2357 | * |
fffb2739 JK |
2358 | * If we add a single extent, then in the worse case, each tree level |
2359 | * index/leaf need to be changed in case of the tree split. | |
ee12b630 | 2360 | * |
fffb2739 JK |
2361 | * If more extents are inserted, they could cause the whole tree split more |
2362 | * than once, but this is really rare. | |
ee12b630 | 2363 | */ |
fffb2739 | 2364 | int ext4_ext_index_trans_blocks(struct inode *inode, int extents) |
ee12b630 MC |
2365 | { |
2366 | int index; | |
f19d5870 TM |
2367 | int depth; |
2368 | ||
2369 | /* If we are converting the inline data, only one is needed here. */ | |
2370 | if (ext4_has_inline_data(inode)) | |
2371 | return 1; | |
2372 | ||
2373 | depth = ext_depth(inode); | |
a86c6181 | 2374 | |
fffb2739 | 2375 | if (extents <= 1) |
ee12b630 MC |
2376 | index = depth * 2; |
2377 | else | |
2378 | index = depth * 3; | |
a86c6181 | 2379 | |
ee12b630 | 2380 | return index; |
a86c6181 AT |
2381 | } |
2382 | ||
981250ca TT |
2383 | static inline int get_default_free_blocks_flags(struct inode *inode) |
2384 | { | |
2385 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
2386 | return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET; | |
2387 | else if (ext4_should_journal_data(inode)) | |
2388 | return EXT4_FREE_BLOCKS_FORGET; | |
2389 | return 0; | |
2390 | } | |
2391 | ||
a86c6181 | 2392 | static int ext4_remove_blocks(handle_t *handle, struct inode *inode, |
0aa06000 | 2393 | struct ext4_extent *ex, |
d23142c6 | 2394 | long long *partial_cluster, |
0aa06000 | 2395 | ext4_lblk_t from, ext4_lblk_t to) |
a86c6181 | 2396 | { |
0aa06000 | 2397 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a2df2a63 | 2398 | unsigned short ee_len = ext4_ext_get_actual_len(ex); |
0aa06000 | 2399 | ext4_fsblk_t pblk; |
981250ca | 2400 | int flags = get_default_free_blocks_flags(inode); |
18888cf0 | 2401 | |
0aa06000 TT |
2402 | /* |
2403 | * For bigalloc file systems, we never free a partial cluster | |
2404 | * at the beginning of the extent. Instead, we make a note | |
2405 | * that we tried freeing the cluster, and check to see if we | |
2406 | * need to free it on a subsequent call to ext4_remove_blocks, | |
2407 | * or at the end of the ext4_truncate() operation. | |
2408 | */ | |
2409 | flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER; | |
2410 | ||
d8990240 | 2411 | trace_ext4_remove_blocks(inode, ex, from, to, *partial_cluster); |
0aa06000 TT |
2412 | /* |
2413 | * If we have a partial cluster, and it's different from the | |
2414 | * cluster of the last block, we need to explicitly free the | |
2415 | * partial cluster here. | |
2416 | */ | |
2417 | pblk = ext4_ext_pblock(ex) + ee_len - 1; | |
d23142c6 LC |
2418 | if ((*partial_cluster > 0) && |
2419 | (EXT4_B2C(sbi, pblk) != *partial_cluster)) { | |
0aa06000 TT |
2420 | ext4_free_blocks(handle, inode, NULL, |
2421 | EXT4_C2B(sbi, *partial_cluster), | |
2422 | sbi->s_cluster_ratio, flags); | |
2423 | *partial_cluster = 0; | |
2424 | } | |
2425 | ||
a86c6181 AT |
2426 | #ifdef EXTENTS_STATS |
2427 | { | |
2428 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
a86c6181 AT |
2429 | spin_lock(&sbi->s_ext_stats_lock); |
2430 | sbi->s_ext_blocks += ee_len; | |
2431 | sbi->s_ext_extents++; | |
2432 | if (ee_len < sbi->s_ext_min) | |
2433 | sbi->s_ext_min = ee_len; | |
2434 | if (ee_len > sbi->s_ext_max) | |
2435 | sbi->s_ext_max = ee_len; | |
2436 | if (ext_depth(inode) > sbi->s_depth_max) | |
2437 | sbi->s_depth_max = ext_depth(inode); | |
2438 | spin_unlock(&sbi->s_ext_stats_lock); | |
2439 | } | |
2440 | #endif | |
2441 | if (from >= le32_to_cpu(ex->ee_block) | |
a2df2a63 | 2442 | && to == le32_to_cpu(ex->ee_block) + ee_len - 1) { |
a86c6181 | 2443 | /* tail removal */ |
725d26d3 | 2444 | ext4_lblk_t num; |
d23142c6 | 2445 | unsigned int unaligned; |
725d26d3 | 2446 | |
a2df2a63 | 2447 | num = le32_to_cpu(ex->ee_block) + ee_len - from; |
0aa06000 | 2448 | pblk = ext4_ext_pblock(ex) + ee_len - num; |
d23142c6 LC |
2449 | /* |
2450 | * Usually we want to free partial cluster at the end of the | |
2451 | * extent, except for the situation when the cluster is still | |
2452 | * used by any other extent (partial_cluster is negative). | |
2453 | */ | |
2454 | if (*partial_cluster < 0 && | |
2455 | -(*partial_cluster) == EXT4_B2C(sbi, pblk + num - 1)) | |
2456 | flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER; | |
2457 | ||
2458 | ext_debug("free last %u blocks starting %llu partial %lld\n", | |
2459 | num, pblk, *partial_cluster); | |
0aa06000 TT |
2460 | ext4_free_blocks(handle, inode, NULL, pblk, num, flags); |
2461 | /* | |
2462 | * If the block range to be freed didn't start at the | |
2463 | * beginning of a cluster, and we removed the entire | |
d23142c6 LC |
2464 | * extent and the cluster is not used by any other extent, |
2465 | * save the partial cluster here, since we might need to | |
2466 | * delete if we determine that the truncate operation has | |
2467 | * removed all of the blocks in the cluster. | |
2468 | * | |
2469 | * On the other hand, if we did not manage to free the whole | |
2470 | * extent, we have to mark the cluster as used (store negative | |
2471 | * cluster number in partial_cluster). | |
0aa06000 | 2472 | */ |
d23142c6 LC |
2473 | unaligned = pblk & (sbi->s_cluster_ratio - 1); |
2474 | if (unaligned && (ee_len == num) && | |
2475 | (*partial_cluster != -((long long)EXT4_B2C(sbi, pblk)))) | |
0aa06000 | 2476 | *partial_cluster = EXT4_B2C(sbi, pblk); |
d23142c6 LC |
2477 | else if (unaligned) |
2478 | *partial_cluster = -((long long)EXT4_B2C(sbi, pblk)); | |
2479 | else if (*partial_cluster > 0) | |
0aa06000 | 2480 | *partial_cluster = 0; |
78fb9cdf LC |
2481 | } else |
2482 | ext4_error(sbi->s_sb, "strange request: removal(2) " | |
2483 | "%u-%u from %u:%u\n", | |
2484 | from, to, le32_to_cpu(ex->ee_block), ee_len); | |
a86c6181 AT |
2485 | return 0; |
2486 | } | |
2487 | ||
d583fb87 AH |
2488 | |
2489 | /* | |
2490 | * ext4_ext_rm_leaf() Removes the extents associated with the | |
2491 | * blocks appearing between "start" and "end", and splits the extents | |
2492 | * if "start" and "end" appear in the same extent | |
2493 | * | |
2494 | * @handle: The journal handle | |
2495 | * @inode: The files inode | |
2496 | * @path: The path to the leaf | |
d23142c6 LC |
2497 | * @partial_cluster: The cluster which we'll have to free if all extents |
2498 | * has been released from it. It gets negative in case | |
2499 | * that the cluster is still used. | |
d583fb87 AH |
2500 | * @start: The first block to remove |
2501 | * @end: The last block to remove | |
2502 | */ | |
a86c6181 AT |
2503 | static int |
2504 | ext4_ext_rm_leaf(handle_t *handle, struct inode *inode, | |
d23142c6 LC |
2505 | struct ext4_ext_path *path, |
2506 | long long *partial_cluster, | |
0aa06000 | 2507 | ext4_lblk_t start, ext4_lblk_t end) |
a86c6181 | 2508 | { |
0aa06000 | 2509 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
a86c6181 AT |
2510 | int err = 0, correct_index = 0; |
2511 | int depth = ext_depth(inode), credits; | |
2512 | struct ext4_extent_header *eh; | |
750c9c47 | 2513 | ext4_lblk_t a, b; |
725d26d3 AK |
2514 | unsigned num; |
2515 | ext4_lblk_t ex_ee_block; | |
a86c6181 | 2516 | unsigned short ex_ee_len; |
a2df2a63 | 2517 | unsigned uninitialized = 0; |
a86c6181 | 2518 | struct ext4_extent *ex; |
d23142c6 | 2519 | ext4_fsblk_t pblk; |
a86c6181 | 2520 | |
c29c0ae7 | 2521 | /* the header must be checked already in ext4_ext_remove_space() */ |
5f95d21f | 2522 | ext_debug("truncate since %u in leaf to %u\n", start, end); |
a86c6181 AT |
2523 | if (!path[depth].p_hdr) |
2524 | path[depth].p_hdr = ext_block_hdr(path[depth].p_bh); | |
2525 | eh = path[depth].p_hdr; | |
273df556 FM |
2526 | if (unlikely(path[depth].p_hdr == NULL)) { |
2527 | EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth); | |
2528 | return -EIO; | |
2529 | } | |
a86c6181 | 2530 | /* find where to start removing */ |
6ae06ff5 AS |
2531 | ex = path[depth].p_ext; |
2532 | if (!ex) | |
2533 | ex = EXT_LAST_EXTENT(eh); | |
a86c6181 AT |
2534 | |
2535 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2536 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 | 2537 | |
d8990240 AK |
2538 | trace_ext4_ext_rm_leaf(inode, start, ex, *partial_cluster); |
2539 | ||
a86c6181 AT |
2540 | while (ex >= EXT_FIRST_EXTENT(eh) && |
2541 | ex_ee_block + ex_ee_len > start) { | |
a41f2071 AK |
2542 | |
2543 | if (ext4_ext_is_uninitialized(ex)) | |
2544 | uninitialized = 1; | |
2545 | else | |
2546 | uninitialized = 0; | |
2547 | ||
553f9008 M |
2548 | ext_debug("remove ext %u:[%d]%d\n", ex_ee_block, |
2549 | uninitialized, ex_ee_len); | |
a86c6181 AT |
2550 | path[depth].p_ext = ex; |
2551 | ||
2552 | a = ex_ee_block > start ? ex_ee_block : start; | |
d583fb87 AH |
2553 | b = ex_ee_block+ex_ee_len - 1 < end ? |
2554 | ex_ee_block+ex_ee_len - 1 : end; | |
a86c6181 AT |
2555 | |
2556 | ext_debug(" border %u:%u\n", a, b); | |
2557 | ||
d583fb87 | 2558 | /* If this extent is beyond the end of the hole, skip it */ |
5f95d21f | 2559 | if (end < ex_ee_block) { |
d23142c6 LC |
2560 | /* |
2561 | * We're going to skip this extent and move to another, | |
2562 | * so if this extent is not cluster aligned we have | |
2563 | * to mark the current cluster as used to avoid | |
2564 | * accidentally freeing it later on | |
2565 | */ | |
2566 | pblk = ext4_ext_pblock(ex); | |
2567 | if (pblk & (sbi->s_cluster_ratio - 1)) | |
2568 | *partial_cluster = | |
2569 | -((long long)EXT4_B2C(sbi, pblk)); | |
d583fb87 AH |
2570 | ex--; |
2571 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
2572 | ex_ee_len = ext4_ext_get_actual_len(ex); | |
2573 | continue; | |
750c9c47 | 2574 | } else if (b != ex_ee_block + ex_ee_len - 1) { |
dc1841d6 LC |
2575 | EXT4_ERROR_INODE(inode, |
2576 | "can not handle truncate %u:%u " | |
2577 | "on extent %u:%u", | |
2578 | start, end, ex_ee_block, | |
2579 | ex_ee_block + ex_ee_len - 1); | |
750c9c47 DM |
2580 | err = -EIO; |
2581 | goto out; | |
a86c6181 AT |
2582 | } else if (a != ex_ee_block) { |
2583 | /* remove tail of the extent */ | |
750c9c47 | 2584 | num = a - ex_ee_block; |
a86c6181 AT |
2585 | } else { |
2586 | /* remove whole extent: excellent! */ | |
a86c6181 | 2587 | num = 0; |
a86c6181 | 2588 | } |
34071da7 TT |
2589 | /* |
2590 | * 3 for leaf, sb, and inode plus 2 (bmap and group | |
2591 | * descriptor) for each block group; assume two block | |
2592 | * groups plus ex_ee_len/blocks_per_block_group for | |
2593 | * the worst case | |
2594 | */ | |
2595 | credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb)); | |
a86c6181 AT |
2596 | if (ex == EXT_FIRST_EXTENT(eh)) { |
2597 | correct_index = 1; | |
2598 | credits += (ext_depth(inode)) + 1; | |
2599 | } | |
5aca07eb | 2600 | credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
a86c6181 | 2601 | |
487caeef | 2602 | err = ext4_ext_truncate_extend_restart(handle, inode, credits); |
9102e4fa | 2603 | if (err) |
a86c6181 | 2604 | goto out; |
a86c6181 AT |
2605 | |
2606 | err = ext4_ext_get_access(handle, inode, path + depth); | |
2607 | if (err) | |
2608 | goto out; | |
2609 | ||
0aa06000 TT |
2610 | err = ext4_remove_blocks(handle, inode, ex, partial_cluster, |
2611 | a, b); | |
a86c6181 AT |
2612 | if (err) |
2613 | goto out; | |
2614 | ||
750c9c47 | 2615 | if (num == 0) |
d0d856e8 | 2616 | /* this extent is removed; mark slot entirely unused */ |
f65e6fba | 2617 | ext4_ext_store_pblock(ex, 0); |
a86c6181 | 2618 | |
a86c6181 | 2619 | ex->ee_len = cpu_to_le16(num); |
749269fa AA |
2620 | /* |
2621 | * Do not mark uninitialized if all the blocks in the | |
2622 | * extent have been removed. | |
2623 | */ | |
2624 | if (uninitialized && num) | |
a2df2a63 | 2625 | ext4_ext_mark_uninitialized(ex); |
d583fb87 AH |
2626 | /* |
2627 | * If the extent was completely released, | |
2628 | * we need to remove it from the leaf | |
2629 | */ | |
2630 | if (num == 0) { | |
f17722f9 | 2631 | if (end != EXT_MAX_BLOCKS - 1) { |
d583fb87 AH |
2632 | /* |
2633 | * For hole punching, we need to scoot all the | |
2634 | * extents up when an extent is removed so that | |
2635 | * we dont have blank extents in the middle | |
2636 | */ | |
2637 | memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) * | |
2638 | sizeof(struct ext4_extent)); | |
2639 | ||
2640 | /* Now get rid of the one at the end */ | |
2641 | memset(EXT_LAST_EXTENT(eh), 0, | |
2642 | sizeof(struct ext4_extent)); | |
2643 | } | |
2644 | le16_add_cpu(&eh->eh_entries, -1); | |
d23142c6 | 2645 | } else if (*partial_cluster > 0) |
0aa06000 | 2646 | *partial_cluster = 0; |
d583fb87 | 2647 | |
750c9c47 DM |
2648 | err = ext4_ext_dirty(handle, inode, path + depth); |
2649 | if (err) | |
2650 | goto out; | |
2651 | ||
bf52c6f7 | 2652 | ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num, |
bf89d16f | 2653 | ext4_ext_pblock(ex)); |
a86c6181 AT |
2654 | ex--; |
2655 | ex_ee_block = le32_to_cpu(ex->ee_block); | |
a2df2a63 | 2656 | ex_ee_len = ext4_ext_get_actual_len(ex); |
a86c6181 AT |
2657 | } |
2658 | ||
2659 | if (correct_index && eh->eh_entries) | |
2660 | err = ext4_ext_correct_indexes(handle, inode, path); | |
2661 | ||
0aa06000 | 2662 | /* |
d23142c6 LC |
2663 | * Free the partial cluster only if the current extent does not |
2664 | * reference it. Otherwise we might free used cluster. | |
0aa06000 | 2665 | */ |
d23142c6 | 2666 | if (*partial_cluster > 0 && |
0aa06000 TT |
2667 | (EXT4_B2C(sbi, ext4_ext_pblock(ex) + ex_ee_len - 1) != |
2668 | *partial_cluster)) { | |
981250ca | 2669 | int flags = get_default_free_blocks_flags(inode); |
0aa06000 TT |
2670 | |
2671 | ext4_free_blocks(handle, inode, NULL, | |
2672 | EXT4_C2B(sbi, *partial_cluster), | |
2673 | sbi->s_cluster_ratio, flags); | |
2674 | *partial_cluster = 0; | |
2675 | } | |
2676 | ||
a86c6181 AT |
2677 | /* if this leaf is free, then we should |
2678 | * remove it from index block above */ | |
2679 | if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL) | |
c36575e6 | 2680 | err = ext4_ext_rm_idx(handle, inode, path, depth); |
a86c6181 AT |
2681 | |
2682 | out: | |
2683 | return err; | |
2684 | } | |
2685 | ||
2686 | /* | |
d0d856e8 RD |
2687 | * ext4_ext_more_to_rm: |
2688 | * returns 1 if current index has to be freed (even partial) | |
a86c6181 | 2689 | */ |
09b88252 | 2690 | static int |
a86c6181 AT |
2691 | ext4_ext_more_to_rm(struct ext4_ext_path *path) |
2692 | { | |
2693 | BUG_ON(path->p_idx == NULL); | |
2694 | ||
2695 | if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr)) | |
2696 | return 0; | |
2697 | ||
2698 | /* | |
d0d856e8 | 2699 | * if truncate on deeper level happened, it wasn't partial, |
a86c6181 AT |
2700 | * so we have to consider current index for truncation |
2701 | */ | |
2702 | if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block) | |
2703 | return 0; | |
2704 | return 1; | |
2705 | } | |
2706 | ||
26a4c0c6 TT |
2707 | int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start, |
2708 | ext4_lblk_t end) | |
a86c6181 AT |
2709 | { |
2710 | struct super_block *sb = inode->i_sb; | |
2711 | int depth = ext_depth(inode); | |
968dee77 | 2712 | struct ext4_ext_path *path = NULL; |
d23142c6 | 2713 | long long partial_cluster = 0; |
a86c6181 | 2714 | handle_t *handle; |
6f2080e6 | 2715 | int i = 0, err = 0; |
a86c6181 | 2716 | |
5f95d21f | 2717 | ext_debug("truncate since %u to %u\n", start, end); |
a86c6181 AT |
2718 | |
2719 | /* probably first extent we're gonna free will be last in block */ | |
9924a92a | 2720 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, depth + 1); |
a86c6181 AT |
2721 | if (IS_ERR(handle)) |
2722 | return PTR_ERR(handle); | |
2723 | ||
0617b83f | 2724 | again: |
61801325 | 2725 | trace_ext4_ext_remove_space(inode, start, end, depth); |
d8990240 | 2726 | |
5f95d21f LC |
2727 | /* |
2728 | * Check if we are removing extents inside the extent tree. If that | |
2729 | * is the case, we are going to punch a hole inside the extent tree | |
2730 | * so we have to check whether we need to split the extent covering | |
2731 | * the last block to remove so we can easily remove the part of it | |
2732 | * in ext4_ext_rm_leaf(). | |
2733 | */ | |
2734 | if (end < EXT_MAX_BLOCKS - 1) { | |
2735 | struct ext4_extent *ex; | |
2736 | ext4_lblk_t ee_block; | |
2737 | ||
2738 | /* find extent for this block */ | |
107a7bd3 | 2739 | path = ext4_ext_find_extent(inode, end, NULL, EXT4_EX_NOCACHE); |
5f95d21f LC |
2740 | if (IS_ERR(path)) { |
2741 | ext4_journal_stop(handle); | |
2742 | return PTR_ERR(path); | |
2743 | } | |
2744 | depth = ext_depth(inode); | |
6f2080e6 | 2745 | /* Leaf not may not exist only if inode has no blocks at all */ |
5f95d21f | 2746 | ex = path[depth].p_ext; |
968dee77 | 2747 | if (!ex) { |
6f2080e6 DM |
2748 | if (depth) { |
2749 | EXT4_ERROR_INODE(inode, | |
2750 | "path[%d].p_hdr == NULL", | |
2751 | depth); | |
2752 | err = -EIO; | |
2753 | } | |
2754 | goto out; | |
968dee77 | 2755 | } |
5f95d21f LC |
2756 | |
2757 | ee_block = le32_to_cpu(ex->ee_block); | |
2758 | ||
2759 | /* | |
2760 | * See if the last block is inside the extent, if so split | |
2761 | * the extent at 'end' block so we can easily remove the | |
2762 | * tail of the first part of the split extent in | |
2763 | * ext4_ext_rm_leaf(). | |
2764 | */ | |
2765 | if (end >= ee_block && | |
2766 | end < ee_block + ext4_ext_get_actual_len(ex) - 1) { | |
2767 | int split_flag = 0; | |
2768 | ||
2769 | if (ext4_ext_is_uninitialized(ex)) | |
2770 | split_flag = EXT4_EXT_MARK_UNINIT1 | | |
2771 | EXT4_EXT_MARK_UNINIT2; | |
2772 | ||
2773 | /* | |
2774 | * Split the extent in two so that 'end' is the last | |
27dd4385 LC |
2775 | * block in the first new extent. Also we should not |
2776 | * fail removing space due to ENOSPC so try to use | |
2777 | * reserved block if that happens. | |
5f95d21f LC |
2778 | */ |
2779 | err = ext4_split_extent_at(handle, inode, path, | |
27dd4385 | 2780 | end + 1, split_flag, |
107a7bd3 | 2781 | EXT4_EX_NOCACHE | |
27dd4385 LC |
2782 | EXT4_GET_BLOCKS_PRE_IO | |
2783 | EXT4_GET_BLOCKS_METADATA_NOFAIL); | |
5f95d21f LC |
2784 | |
2785 | if (err < 0) | |
2786 | goto out; | |
2787 | } | |
5f95d21f | 2788 | } |
a86c6181 | 2789 | /* |
d0d856e8 RD |
2790 | * We start scanning from right side, freeing all the blocks |
2791 | * after i_size and walking into the tree depth-wise. | |
a86c6181 | 2792 | */ |
0617b83f | 2793 | depth = ext_depth(inode); |
968dee77 AS |
2794 | if (path) { |
2795 | int k = i = depth; | |
2796 | while (--k > 0) | |
2797 | path[k].p_block = | |
2798 | le16_to_cpu(path[k].p_hdr->eh_entries)+1; | |
2799 | } else { | |
2800 | path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), | |
2801 | GFP_NOFS); | |
2802 | if (path == NULL) { | |
2803 | ext4_journal_stop(handle); | |
2804 | return -ENOMEM; | |
2805 | } | |
2806 | path[0].p_depth = depth; | |
2807 | path[0].p_hdr = ext_inode_hdr(inode); | |
89a4e48f | 2808 | i = 0; |
5f95d21f | 2809 | |
c349179b | 2810 | if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) { |
968dee77 AS |
2811 | err = -EIO; |
2812 | goto out; | |
2813 | } | |
a86c6181 | 2814 | } |
968dee77 | 2815 | err = 0; |
a86c6181 AT |
2816 | |
2817 | while (i >= 0 && err == 0) { | |
2818 | if (i == depth) { | |
2819 | /* this is leaf block */ | |
d583fb87 | 2820 | err = ext4_ext_rm_leaf(handle, inode, path, |
0aa06000 | 2821 | &partial_cluster, start, |
5f95d21f | 2822 | end); |
d0d856e8 | 2823 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2824 | brelse(path[i].p_bh); |
2825 | path[i].p_bh = NULL; | |
2826 | i--; | |
2827 | continue; | |
2828 | } | |
2829 | ||
2830 | /* this is index block */ | |
2831 | if (!path[i].p_hdr) { | |
2832 | ext_debug("initialize header\n"); | |
2833 | path[i].p_hdr = ext_block_hdr(path[i].p_bh); | |
a86c6181 AT |
2834 | } |
2835 | ||
a86c6181 | 2836 | if (!path[i].p_idx) { |
d0d856e8 | 2837 | /* this level hasn't been touched yet */ |
a86c6181 AT |
2838 | path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr); |
2839 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1; | |
2840 | ext_debug("init index ptr: hdr 0x%p, num %d\n", | |
2841 | path[i].p_hdr, | |
2842 | le16_to_cpu(path[i].p_hdr->eh_entries)); | |
2843 | } else { | |
d0d856e8 | 2844 | /* we were already here, see at next index */ |
a86c6181 AT |
2845 | path[i].p_idx--; |
2846 | } | |
2847 | ||
2848 | ext_debug("level %d - index, first 0x%p, cur 0x%p\n", | |
2849 | i, EXT_FIRST_INDEX(path[i].p_hdr), | |
2850 | path[i].p_idx); | |
2851 | if (ext4_ext_more_to_rm(path + i)) { | |
c29c0ae7 | 2852 | struct buffer_head *bh; |
a86c6181 | 2853 | /* go to the next level */ |
2ae02107 | 2854 | ext_debug("move to level %d (block %llu)\n", |
bf89d16f | 2855 | i + 1, ext4_idx_pblock(path[i].p_idx)); |
a86c6181 | 2856 | memset(path + i + 1, 0, sizeof(*path)); |
7d7ea89e | 2857 | bh = read_extent_tree_block(inode, |
107a7bd3 TT |
2858 | ext4_idx_pblock(path[i].p_idx), depth - i - 1, |
2859 | EXT4_EX_NOCACHE); | |
7d7ea89e | 2860 | if (IS_ERR(bh)) { |
a86c6181 | 2861 | /* should we reset i_size? */ |
7d7ea89e | 2862 | err = PTR_ERR(bh); |
a86c6181 AT |
2863 | break; |
2864 | } | |
76828c88 TT |
2865 | /* Yield here to deal with large extent trees. |
2866 | * Should be a no-op if we did IO above. */ | |
2867 | cond_resched(); | |
c29c0ae7 AT |
2868 | if (WARN_ON(i + 1 > depth)) { |
2869 | err = -EIO; | |
c29c0ae7 AT |
2870 | break; |
2871 | } | |
2872 | path[i + 1].p_bh = bh; | |
a86c6181 | 2873 | |
d0d856e8 RD |
2874 | /* save actual number of indexes since this |
2875 | * number is changed at the next iteration */ | |
a86c6181 AT |
2876 | path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries); |
2877 | i++; | |
2878 | } else { | |
d0d856e8 | 2879 | /* we finished processing this index, go up */ |
a86c6181 | 2880 | if (path[i].p_hdr->eh_entries == 0 && i > 0) { |
d0d856e8 | 2881 | /* index is empty, remove it; |
a86c6181 AT |
2882 | * handle must be already prepared by the |
2883 | * truncatei_leaf() */ | |
c36575e6 | 2884 | err = ext4_ext_rm_idx(handle, inode, path, i); |
a86c6181 | 2885 | } |
d0d856e8 | 2886 | /* root level has p_bh == NULL, brelse() eats this */ |
a86c6181 AT |
2887 | brelse(path[i].p_bh); |
2888 | path[i].p_bh = NULL; | |
2889 | i--; | |
2890 | ext_debug("return to level %d\n", i); | |
2891 | } | |
2892 | } | |
2893 | ||
61801325 LC |
2894 | trace_ext4_ext_remove_space_done(inode, start, end, depth, |
2895 | partial_cluster, path->p_hdr->eh_entries); | |
d8990240 | 2896 | |
7b415bf6 AK |
2897 | /* If we still have something in the partial cluster and we have removed |
2898 | * even the first extent, then we should free the blocks in the partial | |
2899 | * cluster as well. */ | |
d23142c6 | 2900 | if (partial_cluster > 0 && path->p_hdr->eh_entries == 0) { |
981250ca | 2901 | int flags = get_default_free_blocks_flags(inode); |
7b415bf6 AK |
2902 | |
2903 | ext4_free_blocks(handle, inode, NULL, | |
2904 | EXT4_C2B(EXT4_SB(sb), partial_cluster), | |
2905 | EXT4_SB(sb)->s_cluster_ratio, flags); | |
2906 | partial_cluster = 0; | |
2907 | } | |
2908 | ||
a86c6181 AT |
2909 | /* TODO: flexible tree reduction should be here */ |
2910 | if (path->p_hdr->eh_entries == 0) { | |
2911 | /* | |
d0d856e8 RD |
2912 | * truncate to zero freed all the tree, |
2913 | * so we need to correct eh_depth | |
a86c6181 AT |
2914 | */ |
2915 | err = ext4_ext_get_access(handle, inode, path); | |
2916 | if (err == 0) { | |
2917 | ext_inode_hdr(inode)->eh_depth = 0; | |
2918 | ext_inode_hdr(inode)->eh_max = | |
55ad63bf | 2919 | cpu_to_le16(ext4_ext_space_root(inode, 0)); |
a86c6181 AT |
2920 | err = ext4_ext_dirty(handle, inode, path); |
2921 | } | |
2922 | } | |
2923 | out: | |
a86c6181 AT |
2924 | ext4_ext_drop_refs(path); |
2925 | kfree(path); | |
968dee77 AS |
2926 | if (err == -EAGAIN) { |
2927 | path = NULL; | |
0617b83f | 2928 | goto again; |
968dee77 | 2929 | } |
a86c6181 AT |
2930 | ext4_journal_stop(handle); |
2931 | ||
2932 | return err; | |
2933 | } | |
2934 | ||
2935 | /* | |
2936 | * called at mount time | |
2937 | */ | |
2938 | void ext4_ext_init(struct super_block *sb) | |
2939 | { | |
2940 | /* | |
2941 | * possible initialization would be here | |
2942 | */ | |
2943 | ||
83982b6f | 2944 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
90576c0b | 2945 | #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS) |
92b97816 | 2946 | printk(KERN_INFO "EXT4-fs: file extents enabled" |
bbf2f9fb | 2947 | #ifdef AGGRESSIVE_TEST |
92b97816 | 2948 | ", aggressive tests" |
a86c6181 AT |
2949 | #endif |
2950 | #ifdef CHECK_BINSEARCH | |
92b97816 | 2951 | ", check binsearch" |
a86c6181 AT |
2952 | #endif |
2953 | #ifdef EXTENTS_STATS | |
92b97816 | 2954 | ", stats" |
a86c6181 | 2955 | #endif |
92b97816 | 2956 | "\n"); |
90576c0b | 2957 | #endif |
a86c6181 AT |
2958 | #ifdef EXTENTS_STATS |
2959 | spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock); | |
2960 | EXT4_SB(sb)->s_ext_min = 1 << 30; | |
2961 | EXT4_SB(sb)->s_ext_max = 0; | |
2962 | #endif | |
2963 | } | |
2964 | } | |
2965 | ||
2966 | /* | |
2967 | * called at umount time | |
2968 | */ | |
2969 | void ext4_ext_release(struct super_block *sb) | |
2970 | { | |
83982b6f | 2971 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) |
a86c6181 AT |
2972 | return; |
2973 | ||
2974 | #ifdef EXTENTS_STATS | |
2975 | if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) { | |
2976 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2977 | printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n", | |
2978 | sbi->s_ext_blocks, sbi->s_ext_extents, | |
2979 | sbi->s_ext_blocks / sbi->s_ext_extents); | |
2980 | printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n", | |
2981 | sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max); | |
2982 | } | |
2983 | #endif | |
2984 | } | |
2985 | ||
093a088b AK |
2986 | /* FIXME!! we need to try to merge to left or right after zero-out */ |
2987 | static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |
2988 | { | |
2407518d LC |
2989 | ext4_fsblk_t ee_pblock; |
2990 | unsigned int ee_len; | |
b720303d | 2991 | int ret; |
093a088b | 2992 | |
093a088b | 2993 | ee_len = ext4_ext_get_actual_len(ex); |
bf89d16f | 2994 | ee_pblock = ext4_ext_pblock(ex); |
b720303d | 2995 | |
a107e5a3 | 2996 | ret = sb_issue_zeroout(inode->i_sb, ee_pblock, ee_len, GFP_NOFS); |
2407518d LC |
2997 | if (ret > 0) |
2998 | ret = 0; | |
093a088b | 2999 | |
2407518d | 3000 | return ret; |
093a088b AK |
3001 | } |
3002 | ||
47ea3bb5 YY |
3003 | /* |
3004 | * ext4_split_extent_at() splits an extent at given block. | |
3005 | * | |
3006 | * @handle: the journal handle | |
3007 | * @inode: the file inode | |
3008 | * @path: the path to the extent | |
3009 | * @split: the logical block where the extent is splitted. | |
3010 | * @split_flags: indicates if the extent could be zeroout if split fails, and | |
3011 | * the states(init or uninit) of new extents. | |
3012 | * @flags: flags used to insert new extent to extent tree. | |
3013 | * | |
3014 | * | |
3015 | * Splits extent [a, b] into two extents [a, @split) and [@split, b], states | |
3016 | * of which are deterimined by split_flag. | |
3017 | * | |
3018 | * There are two cases: | |
3019 | * a> the extent are splitted into two extent. | |
3020 | * b> split is not needed, and just mark the extent. | |
3021 | * | |
3022 | * return 0 on success. | |
3023 | */ | |
3024 | static int ext4_split_extent_at(handle_t *handle, | |
3025 | struct inode *inode, | |
3026 | struct ext4_ext_path *path, | |
3027 | ext4_lblk_t split, | |
3028 | int split_flag, | |
3029 | int flags) | |
3030 | { | |
3031 | ext4_fsblk_t newblock; | |
3032 | ext4_lblk_t ee_block; | |
adb23551 | 3033 | struct ext4_extent *ex, newex, orig_ex, zero_ex; |
47ea3bb5 YY |
3034 | struct ext4_extent *ex2 = NULL; |
3035 | unsigned int ee_len, depth; | |
3036 | int err = 0; | |
3037 | ||
dee1f973 DM |
3038 | BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) == |
3039 | (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)); | |
3040 | ||
47ea3bb5 YY |
3041 | ext_debug("ext4_split_extents_at: inode %lu, logical" |
3042 | "block %llu\n", inode->i_ino, (unsigned long long)split); | |
3043 | ||
3044 | ext4_ext_show_leaf(inode, path); | |
3045 | ||
3046 | depth = ext_depth(inode); | |
3047 | ex = path[depth].p_ext; | |
3048 | ee_block = le32_to_cpu(ex->ee_block); | |
3049 | ee_len = ext4_ext_get_actual_len(ex); | |
3050 | newblock = split - ee_block + ext4_ext_pblock(ex); | |
3051 | ||
3052 | BUG_ON(split < ee_block || split >= (ee_block + ee_len)); | |
357b66fd DM |
3053 | BUG_ON(!ext4_ext_is_uninitialized(ex) && |
3054 | split_flag & (EXT4_EXT_MAY_ZEROOUT | | |
3055 | EXT4_EXT_MARK_UNINIT1 | | |
3056 | EXT4_EXT_MARK_UNINIT2)); | |
47ea3bb5 YY |
3057 | |
3058 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3059 | if (err) | |
3060 | goto out; | |
3061 | ||
3062 | if (split == ee_block) { | |
3063 | /* | |
3064 | * case b: block @split is the block that the extent begins with | |
3065 | * then we just change the state of the extent, and splitting | |
3066 | * is not needed. | |
3067 | */ | |
3068 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3069 | ext4_ext_mark_uninitialized(ex); | |
3070 | else | |
3071 | ext4_ext_mark_initialized(ex); | |
3072 | ||
3073 | if (!(flags & EXT4_GET_BLOCKS_PRE_IO)) | |
ecb94f5f | 3074 | ext4_ext_try_to_merge(handle, inode, path, ex); |
47ea3bb5 | 3075 | |
ecb94f5f | 3076 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
47ea3bb5 YY |
3077 | goto out; |
3078 | } | |
3079 | ||
3080 | /* case a */ | |
3081 | memcpy(&orig_ex, ex, sizeof(orig_ex)); | |
3082 | ex->ee_len = cpu_to_le16(split - ee_block); | |
3083 | if (split_flag & EXT4_EXT_MARK_UNINIT1) | |
3084 | ext4_ext_mark_uninitialized(ex); | |
3085 | ||
3086 | /* | |
3087 | * path may lead to new leaf, not to original leaf any more | |
3088 | * after ext4_ext_insert_extent() returns, | |
3089 | */ | |
3090 | err = ext4_ext_dirty(handle, inode, path + depth); | |
3091 | if (err) | |
3092 | goto fix_extent_len; | |
3093 | ||
3094 | ex2 = &newex; | |
3095 | ex2->ee_block = cpu_to_le32(split); | |
3096 | ex2->ee_len = cpu_to_le16(ee_len - (split - ee_block)); | |
3097 | ext4_ext_store_pblock(ex2, newblock); | |
3098 | if (split_flag & EXT4_EXT_MARK_UNINIT2) | |
3099 | ext4_ext_mark_uninitialized(ex2); | |
3100 | ||
3101 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | |
3102 | if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) { | |
dee1f973 | 3103 | if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) { |
adb23551 | 3104 | if (split_flag & EXT4_EXT_DATA_VALID1) { |
dee1f973 | 3105 | err = ext4_ext_zeroout(inode, ex2); |
adb23551 | 3106 | zero_ex.ee_block = ex2->ee_block; |
8cde7ad1 ZL |
3107 | zero_ex.ee_len = cpu_to_le16( |
3108 | ext4_ext_get_actual_len(ex2)); | |
adb23551 ZL |
3109 | ext4_ext_store_pblock(&zero_ex, |
3110 | ext4_ext_pblock(ex2)); | |
3111 | } else { | |
dee1f973 | 3112 | err = ext4_ext_zeroout(inode, ex); |
adb23551 | 3113 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 ZL |
3114 | zero_ex.ee_len = cpu_to_le16( |
3115 | ext4_ext_get_actual_len(ex)); | |
adb23551 ZL |
3116 | ext4_ext_store_pblock(&zero_ex, |
3117 | ext4_ext_pblock(ex)); | |
3118 | } | |
3119 | } else { | |
dee1f973 | 3120 | err = ext4_ext_zeroout(inode, &orig_ex); |
adb23551 | 3121 | zero_ex.ee_block = orig_ex.ee_block; |
8cde7ad1 ZL |
3122 | zero_ex.ee_len = cpu_to_le16( |
3123 | ext4_ext_get_actual_len(&orig_ex)); | |
adb23551 ZL |
3124 | ext4_ext_store_pblock(&zero_ex, |
3125 | ext4_ext_pblock(&orig_ex)); | |
3126 | } | |
dee1f973 | 3127 | |
47ea3bb5 YY |
3128 | if (err) |
3129 | goto fix_extent_len; | |
3130 | /* update the extent length and mark as initialized */ | |
af1584f5 | 3131 | ex->ee_len = cpu_to_le16(ee_len); |
ecb94f5f TT |
3132 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3133 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
adb23551 ZL |
3134 | if (err) |
3135 | goto fix_extent_len; | |
3136 | ||
3137 | /* update extent status tree */ | |
3138 | err = ext4_es_zeroout(inode, &zero_ex); | |
3139 | ||
47ea3bb5 YY |
3140 | goto out; |
3141 | } else if (err) | |
3142 | goto fix_extent_len; | |
3143 | ||
3144 | out: | |
3145 | ext4_ext_show_leaf(inode, path); | |
3146 | return err; | |
3147 | ||
3148 | fix_extent_len: | |
3149 | ex->ee_len = orig_ex.ee_len; | |
3150 | ext4_ext_dirty(handle, inode, path + depth); | |
3151 | return err; | |
3152 | } | |
3153 | ||
3154 | /* | |
3155 | * ext4_split_extents() splits an extent and mark extent which is covered | |
3156 | * by @map as split_flags indicates | |
3157 | * | |
3158 | * It may result in splitting the extent into multiple extents (upto three) | |
3159 | * There are three possibilities: | |
3160 | * a> There is no split required | |
3161 | * b> Splits in two extents: Split is happening at either end of the extent | |
3162 | * c> Splits in three extents: Somone is splitting in middle of the extent | |
3163 | * | |
3164 | */ | |
3165 | static int ext4_split_extent(handle_t *handle, | |
3166 | struct inode *inode, | |
3167 | struct ext4_ext_path *path, | |
3168 | struct ext4_map_blocks *map, | |
3169 | int split_flag, | |
3170 | int flags) | |
3171 | { | |
3172 | ext4_lblk_t ee_block; | |
3173 | struct ext4_extent *ex; | |
3174 | unsigned int ee_len, depth; | |
3175 | int err = 0; | |
3176 | int uninitialized; | |
3177 | int split_flag1, flags1; | |
3a225670 | 3178 | int allocated = map->m_len; |
47ea3bb5 YY |
3179 | |
3180 | depth = ext_depth(inode); | |
3181 | ex = path[depth].p_ext; | |
3182 | ee_block = le32_to_cpu(ex->ee_block); | |
3183 | ee_len = ext4_ext_get_actual_len(ex); | |
3184 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3185 | ||
3186 | if (map->m_lblk + map->m_len < ee_block + ee_len) { | |
dee1f973 | 3187 | split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; |
47ea3bb5 YY |
3188 | flags1 = flags | EXT4_GET_BLOCKS_PRE_IO; |
3189 | if (uninitialized) | |
3190 | split_flag1 |= EXT4_EXT_MARK_UNINIT1 | | |
3191 | EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3192 | if (split_flag & EXT4_EXT_DATA_VALID2) |
3193 | split_flag1 |= EXT4_EXT_DATA_VALID1; | |
47ea3bb5 YY |
3194 | err = ext4_split_extent_at(handle, inode, path, |
3195 | map->m_lblk + map->m_len, split_flag1, flags1); | |
93917411 YY |
3196 | if (err) |
3197 | goto out; | |
3a225670 ZL |
3198 | } else { |
3199 | allocated = ee_len - (map->m_lblk - ee_block); | |
47ea3bb5 | 3200 | } |
357b66fd DM |
3201 | /* |
3202 | * Update path is required because previous ext4_split_extent_at() may | |
3203 | * result in split of original leaf or extent zeroout. | |
3204 | */ | |
47ea3bb5 | 3205 | ext4_ext_drop_refs(path); |
107a7bd3 | 3206 | path = ext4_ext_find_extent(inode, map->m_lblk, path, 0); |
47ea3bb5 YY |
3207 | if (IS_ERR(path)) |
3208 | return PTR_ERR(path); | |
357b66fd DM |
3209 | depth = ext_depth(inode); |
3210 | ex = path[depth].p_ext; | |
3211 | uninitialized = ext4_ext_is_uninitialized(ex); | |
3212 | split_flag1 = 0; | |
47ea3bb5 YY |
3213 | |
3214 | if (map->m_lblk >= ee_block) { | |
357b66fd DM |
3215 | split_flag1 = split_flag & EXT4_EXT_DATA_VALID2; |
3216 | if (uninitialized) { | |
47ea3bb5 | 3217 | split_flag1 |= EXT4_EXT_MARK_UNINIT1; |
357b66fd DM |
3218 | split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT | |
3219 | EXT4_EXT_MARK_UNINIT2); | |
3220 | } | |
47ea3bb5 YY |
3221 | err = ext4_split_extent_at(handle, inode, path, |
3222 | map->m_lblk, split_flag1, flags); | |
3223 | if (err) | |
3224 | goto out; | |
3225 | } | |
3226 | ||
3227 | ext4_ext_show_leaf(inode, path); | |
3228 | out: | |
3a225670 | 3229 | return err ? err : allocated; |
47ea3bb5 YY |
3230 | } |
3231 | ||
56055d3a | 3232 | /* |
e35fd660 | 3233 | * This function is called by ext4_ext_map_blocks() if someone tries to write |
56055d3a | 3234 | * to an uninitialized extent. It may result in splitting the uninitialized |
25985edc | 3235 | * extent into multiple extents (up to three - one initialized and two |
56055d3a AA |
3236 | * uninitialized). |
3237 | * There are three possibilities: | |
3238 | * a> There is no split required: Entire extent should be initialized | |
3239 | * b> Splits in two extents: Write is happening at either end of the extent | |
3240 | * c> Splits in three extents: Somone is writing in middle of the extent | |
6f91bc5f EG |
3241 | * |
3242 | * Pre-conditions: | |
3243 | * - The extent pointed to by 'path' is uninitialized. | |
3244 | * - The extent pointed to by 'path' contains a superset | |
3245 | * of the logical span [map->m_lblk, map->m_lblk + map->m_len). | |
3246 | * | |
3247 | * Post-conditions on success: | |
3248 | * - the returned value is the number of blocks beyond map->l_lblk | |
3249 | * that are allocated and initialized. | |
3250 | * It is guaranteed to be >= map->m_len. | |
56055d3a | 3251 | */ |
725d26d3 | 3252 | static int ext4_ext_convert_to_initialized(handle_t *handle, |
e35fd660 TT |
3253 | struct inode *inode, |
3254 | struct ext4_map_blocks *map, | |
27dd4385 LC |
3255 | struct ext4_ext_path *path, |
3256 | int flags) | |
56055d3a | 3257 | { |
67a5da56 | 3258 | struct ext4_sb_info *sbi; |
6f91bc5f | 3259 | struct ext4_extent_header *eh; |
667eff35 YY |
3260 | struct ext4_map_blocks split_map; |
3261 | struct ext4_extent zero_ex; | |
bc2d9db4 | 3262 | struct ext4_extent *ex, *abut_ex; |
21ca087a | 3263 | ext4_lblk_t ee_block, eof_block; |
bc2d9db4 LC |
3264 | unsigned int ee_len, depth, map_len = map->m_len; |
3265 | int allocated = 0, max_zeroout = 0; | |
56055d3a | 3266 | int err = 0; |
667eff35 | 3267 | int split_flag = 0; |
21ca087a DM |
3268 | |
3269 | ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical" | |
3270 | "block %llu, max_blocks %u\n", inode->i_ino, | |
bc2d9db4 | 3271 | (unsigned long long)map->m_lblk, map_len); |
21ca087a | 3272 | |
67a5da56 | 3273 | sbi = EXT4_SB(inode->i_sb); |
21ca087a DM |
3274 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> |
3275 | inode->i_sb->s_blocksize_bits; | |
bc2d9db4 LC |
3276 | if (eof_block < map->m_lblk + map_len) |
3277 | eof_block = map->m_lblk + map_len; | |
56055d3a AA |
3278 | |
3279 | depth = ext_depth(inode); | |
6f91bc5f | 3280 | eh = path[depth].p_hdr; |
56055d3a AA |
3281 | ex = path[depth].p_ext; |
3282 | ee_block = le32_to_cpu(ex->ee_block); | |
3283 | ee_len = ext4_ext_get_actual_len(ex); | |
adb23551 | 3284 | zero_ex.ee_len = 0; |
56055d3a | 3285 | |
6f91bc5f EG |
3286 | trace_ext4_ext_convert_to_initialized_enter(inode, map, ex); |
3287 | ||
3288 | /* Pre-conditions */ | |
3289 | BUG_ON(!ext4_ext_is_uninitialized(ex)); | |
3290 | BUG_ON(!in_range(map->m_lblk, ee_block, ee_len)); | |
6f91bc5f EG |
3291 | |
3292 | /* | |
3293 | * Attempt to transfer newly initialized blocks from the currently | |
bc2d9db4 | 3294 | * uninitialized extent to its neighbor. This is much cheaper |
6f91bc5f | 3295 | * than an insertion followed by a merge as those involve costly |
bc2d9db4 LC |
3296 | * memmove() calls. Transferring to the left is the common case in |
3297 | * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE) | |
3298 | * followed by append writes. | |
6f91bc5f EG |
3299 | * |
3300 | * Limitations of the current logic: | |
bc2d9db4 | 3301 | * - L1: we do not deal with writes covering the whole extent. |
6f91bc5f EG |
3302 | * This would require removing the extent if the transfer |
3303 | * is possible. | |
bc2d9db4 | 3304 | * - L2: we only attempt to merge with an extent stored in the |
6f91bc5f EG |
3305 | * same extent tree node. |
3306 | */ | |
bc2d9db4 LC |
3307 | if ((map->m_lblk == ee_block) && |
3308 | /* See if we can merge left */ | |
3309 | (map_len < ee_len) && /*L1*/ | |
3310 | (ex > EXT_FIRST_EXTENT(eh))) { /*L2*/ | |
6f91bc5f EG |
3311 | ext4_lblk_t prev_lblk; |
3312 | ext4_fsblk_t prev_pblk, ee_pblk; | |
bc2d9db4 | 3313 | unsigned int prev_len; |
6f91bc5f | 3314 | |
bc2d9db4 LC |
3315 | abut_ex = ex - 1; |
3316 | prev_lblk = le32_to_cpu(abut_ex->ee_block); | |
3317 | prev_len = ext4_ext_get_actual_len(abut_ex); | |
3318 | prev_pblk = ext4_ext_pblock(abut_ex); | |
6f91bc5f | 3319 | ee_pblk = ext4_ext_pblock(ex); |
6f91bc5f EG |
3320 | |
3321 | /* | |
bc2d9db4 | 3322 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed |
6f91bc5f | 3323 | * upon those conditions: |
bc2d9db4 LC |
3324 | * - C1: abut_ex is initialized, |
3325 | * - C2: abut_ex is logically abutting ex, | |
3326 | * - C3: abut_ex is physically abutting ex, | |
3327 | * - C4: abut_ex can receive the additional blocks without | |
6f91bc5f EG |
3328 | * overflowing the (initialized) length limit. |
3329 | */ | |
bc2d9db4 | 3330 | if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/ |
6f91bc5f EG |
3331 | ((prev_lblk + prev_len) == ee_block) && /*C2*/ |
3332 | ((prev_pblk + prev_len) == ee_pblk) && /*C3*/ | |
bc2d9db4 | 3333 | (prev_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ |
6f91bc5f EG |
3334 | err = ext4_ext_get_access(handle, inode, path + depth); |
3335 | if (err) | |
3336 | goto out; | |
3337 | ||
3338 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
bc2d9db4 | 3339 | map, ex, abut_ex); |
6f91bc5f | 3340 | |
bc2d9db4 LC |
3341 | /* Shift the start of ex by 'map_len' blocks */ |
3342 | ex->ee_block = cpu_to_le32(ee_block + map_len); | |
3343 | ext4_ext_store_pblock(ex, ee_pblk + map_len); | |
3344 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
6f91bc5f EG |
3345 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ |
3346 | ||
bc2d9db4 LC |
3347 | /* Extend abut_ex by 'map_len' blocks */ |
3348 | abut_ex->ee_len = cpu_to_le16(prev_len + map_len); | |
6f91bc5f | 3349 | |
bc2d9db4 LC |
3350 | /* Result: number of initialized blocks past m_lblk */ |
3351 | allocated = map_len; | |
3352 | } | |
3353 | } else if (((map->m_lblk + map_len) == (ee_block + ee_len)) && | |
3354 | (map_len < ee_len) && /*L1*/ | |
3355 | ex < EXT_LAST_EXTENT(eh)) { /*L2*/ | |
3356 | /* See if we can merge right */ | |
3357 | ext4_lblk_t next_lblk; | |
3358 | ext4_fsblk_t next_pblk, ee_pblk; | |
3359 | unsigned int next_len; | |
3360 | ||
3361 | abut_ex = ex + 1; | |
3362 | next_lblk = le32_to_cpu(abut_ex->ee_block); | |
3363 | next_len = ext4_ext_get_actual_len(abut_ex); | |
3364 | next_pblk = ext4_ext_pblock(abut_ex); | |
3365 | ee_pblk = ext4_ext_pblock(ex); | |
6f91bc5f | 3366 | |
bc2d9db4 LC |
3367 | /* |
3368 | * A transfer of blocks from 'ex' to 'abut_ex' is allowed | |
3369 | * upon those conditions: | |
3370 | * - C1: abut_ex is initialized, | |
3371 | * - C2: abut_ex is logically abutting ex, | |
3372 | * - C3: abut_ex is physically abutting ex, | |
3373 | * - C4: abut_ex can receive the additional blocks without | |
3374 | * overflowing the (initialized) length limit. | |
3375 | */ | |
3376 | if ((!ext4_ext_is_uninitialized(abut_ex)) && /*C1*/ | |
3377 | ((map->m_lblk + map_len) == next_lblk) && /*C2*/ | |
3378 | ((ee_pblk + ee_len) == next_pblk) && /*C3*/ | |
3379 | (next_len < (EXT_INIT_MAX_LEN - map_len))) { /*C4*/ | |
3380 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3381 | if (err) | |
3382 | goto out; | |
3383 | ||
3384 | trace_ext4_ext_convert_to_initialized_fastpath(inode, | |
3385 | map, ex, abut_ex); | |
3386 | ||
3387 | /* Shift the start of abut_ex by 'map_len' blocks */ | |
3388 | abut_ex->ee_block = cpu_to_le32(next_lblk - map_len); | |
3389 | ext4_ext_store_pblock(abut_ex, next_pblk - map_len); | |
3390 | ex->ee_len = cpu_to_le16(ee_len - map_len); | |
3391 | ext4_ext_mark_uninitialized(ex); /* Restore the flag */ | |
3392 | ||
3393 | /* Extend abut_ex by 'map_len' blocks */ | |
3394 | abut_ex->ee_len = cpu_to_le16(next_len + map_len); | |
6f91bc5f EG |
3395 | |
3396 | /* Result: number of initialized blocks past m_lblk */ | |
bc2d9db4 | 3397 | allocated = map_len; |
6f91bc5f EG |
3398 | } |
3399 | } | |
bc2d9db4 LC |
3400 | if (allocated) { |
3401 | /* Mark the block containing both extents as dirty */ | |
3402 | ext4_ext_dirty(handle, inode, path + depth); | |
3403 | ||
3404 | /* Update path to point to the right extent */ | |
3405 | path[depth].p_ext = abut_ex; | |
3406 | goto out; | |
3407 | } else | |
3408 | allocated = ee_len - (map->m_lblk - ee_block); | |
6f91bc5f | 3409 | |
667eff35 | 3410 | WARN_ON(map->m_lblk < ee_block); |
21ca087a DM |
3411 | /* |
3412 | * It is safe to convert extent to initialized via explicit | |
3413 | * zeroout only if extent is fully insde i_size or new_size. | |
3414 | */ | |
667eff35 | 3415 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
21ca087a | 3416 | |
67a5da56 ZL |
3417 | if (EXT4_EXT_MAY_ZEROOUT & split_flag) |
3418 | max_zeroout = sbi->s_extent_max_zeroout_kb >> | |
4f42f80a | 3419 | (inode->i_sb->s_blocksize_bits - 10); |
67a5da56 ZL |
3420 | |
3421 | /* If extent is less than s_max_zeroout_kb, zeroout directly */ | |
3422 | if (max_zeroout && (ee_len <= max_zeroout)) { | |
667eff35 | 3423 | err = ext4_ext_zeroout(inode, ex); |
3977c965 | 3424 | if (err) |
d03856bd | 3425 | goto out; |
adb23551 | 3426 | zero_ex.ee_block = ex->ee_block; |
8cde7ad1 | 3427 | zero_ex.ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)); |
adb23551 | 3428 | ext4_ext_store_pblock(&zero_ex, ext4_ext_pblock(ex)); |
d03856bd AK |
3429 | |
3430 | err = ext4_ext_get_access(handle, inode, path + depth); | |
3431 | if (err) | |
3432 | goto out; | |
667eff35 | 3433 | ext4_ext_mark_initialized(ex); |
ecb94f5f TT |
3434 | ext4_ext_try_to_merge(handle, inode, path, ex); |
3435 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); | |
667eff35 | 3436 | goto out; |
56055d3a | 3437 | } |
667eff35 | 3438 | |
56055d3a | 3439 | /* |
667eff35 YY |
3440 | * four cases: |
3441 | * 1. split the extent into three extents. | |
3442 | * 2. split the extent into two extents, zeroout the first half. | |
3443 | * 3. split the extent into two extents, zeroout the second half. | |
3444 | * 4. split the extent into two extents with out zeroout. | |
56055d3a | 3445 | */ |
667eff35 YY |
3446 | split_map.m_lblk = map->m_lblk; |
3447 | split_map.m_len = map->m_len; | |
3448 | ||
67a5da56 ZL |
3449 | if (max_zeroout && (allocated > map->m_len)) { |
3450 | if (allocated <= max_zeroout) { | |
667eff35 YY |
3451 | /* case 3 */ |
3452 | zero_ex.ee_block = | |
9b940f8e AH |
3453 | cpu_to_le32(map->m_lblk); |
3454 | zero_ex.ee_len = cpu_to_le16(allocated); | |
667eff35 YY |
3455 | ext4_ext_store_pblock(&zero_ex, |
3456 | ext4_ext_pblock(ex) + map->m_lblk - ee_block); | |
3457 | err = ext4_ext_zeroout(inode, &zero_ex); | |
56055d3a AA |
3458 | if (err) |
3459 | goto out; | |
667eff35 YY |
3460 | split_map.m_lblk = map->m_lblk; |
3461 | split_map.m_len = allocated; | |
67a5da56 | 3462 | } else if (map->m_lblk - ee_block + map->m_len < max_zeroout) { |
667eff35 YY |
3463 | /* case 2 */ |
3464 | if (map->m_lblk != ee_block) { | |
3465 | zero_ex.ee_block = ex->ee_block; | |
3466 | zero_ex.ee_len = cpu_to_le16(map->m_lblk - | |
3467 | ee_block); | |
3468 | ext4_ext_store_pblock(&zero_ex, | |
3469 | ext4_ext_pblock(ex)); | |
3470 | err = ext4_ext_zeroout(inode, &zero_ex); | |
3471 | if (err) | |
3472 | goto out; | |
3473 | } | |
3474 | ||
667eff35 | 3475 | split_map.m_lblk = ee_block; |
9b940f8e AH |
3476 | split_map.m_len = map->m_lblk - ee_block + map->m_len; |
3477 | allocated = map->m_len; | |
56055d3a AA |
3478 | } |
3479 | } | |
667eff35 YY |
3480 | |
3481 | allocated = ext4_split_extent(handle, inode, path, | |
27dd4385 | 3482 | &split_map, split_flag, flags); |
667eff35 YY |
3483 | if (allocated < 0) |
3484 | err = allocated; | |
3485 | ||
56055d3a | 3486 | out: |
adb23551 ZL |
3487 | /* If we have gotten a failure, don't zero out status tree */ |
3488 | if (!err) | |
3489 | err = ext4_es_zeroout(inode, &zero_ex); | |
56055d3a AA |
3490 | return err ? err : allocated; |
3491 | } | |
3492 | ||
0031462b | 3493 | /* |
e35fd660 | 3494 | * This function is called by ext4_ext_map_blocks() from |
0031462b MC |
3495 | * ext4_get_blocks_dio_write() when DIO to write |
3496 | * to an uninitialized extent. | |
3497 | * | |
fd018fe8 | 3498 | * Writing to an uninitialized extent may result in splitting the uninitialized |
30cb27d6 | 3499 | * extent into multiple initialized/uninitialized extents (up to three) |
0031462b MC |
3500 | * There are three possibilities: |
3501 | * a> There is no split required: Entire extent should be uninitialized | |
3502 | * b> Splits in two extents: Write is happening at either end of the extent | |
3503 | * c> Splits in three extents: Somone is writing in middle of the extent | |
3504 | * | |
3505 | * One of more index blocks maybe needed if the extent tree grow after | |
b595076a | 3506 | * the uninitialized extent split. To prevent ENOSPC occur at the IO |
0031462b | 3507 | * complete, we need to split the uninitialized extent before DIO submit |
421f91d2 | 3508 | * the IO. The uninitialized extent called at this time will be split |
0031462b MC |
3509 | * into three uninitialized extent(at most). After IO complete, the part |
3510 | * being filled will be convert to initialized by the end_io callback function | |
3511 | * via ext4_convert_unwritten_extents(). | |
ba230c3f M |
3512 | * |
3513 | * Returns the size of uninitialized extent to be written on success. | |
0031462b MC |
3514 | */ |
3515 | static int ext4_split_unwritten_extents(handle_t *handle, | |
3516 | struct inode *inode, | |
e35fd660 | 3517 | struct ext4_map_blocks *map, |
0031462b | 3518 | struct ext4_ext_path *path, |
0031462b MC |
3519 | int flags) |
3520 | { | |
667eff35 YY |
3521 | ext4_lblk_t eof_block; |
3522 | ext4_lblk_t ee_block; | |
3523 | struct ext4_extent *ex; | |
3524 | unsigned int ee_len; | |
3525 | int split_flag = 0, depth; | |
21ca087a DM |
3526 | |
3527 | ext_debug("ext4_split_unwritten_extents: inode %lu, logical" | |
3528 | "block %llu, max_blocks %u\n", inode->i_ino, | |
e35fd660 | 3529 | (unsigned long long)map->m_lblk, map->m_len); |
21ca087a DM |
3530 | |
3531 | eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >> | |
3532 | inode->i_sb->s_blocksize_bits; | |
e35fd660 TT |
3533 | if (eof_block < map->m_lblk + map->m_len) |
3534 | eof_block = map->m_lblk + map->m_len; | |
21ca087a DM |
3535 | /* |
3536 | * It is safe to convert extent to initialized via explicit | |
3537 | * zeroout only if extent is fully insde i_size or new_size. | |
3538 | */ | |
667eff35 YY |
3539 | depth = ext_depth(inode); |
3540 | ex = path[depth].p_ext; | |
3541 | ee_block = le32_to_cpu(ex->ee_block); | |
3542 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3543 | |
667eff35 YY |
3544 | split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0; |
3545 | split_flag |= EXT4_EXT_MARK_UNINIT2; | |
dee1f973 DM |
3546 | if (flags & EXT4_GET_BLOCKS_CONVERT) |
3547 | split_flag |= EXT4_EXT_DATA_VALID2; | |
667eff35 YY |
3548 | flags |= EXT4_GET_BLOCKS_PRE_IO; |
3549 | return ext4_split_extent(handle, inode, path, map, split_flag, flags); | |
0031462b | 3550 | } |
197217a5 | 3551 | |
c7064ef1 | 3552 | static int ext4_convert_unwritten_extents_endio(handle_t *handle, |
dee1f973 DM |
3553 | struct inode *inode, |
3554 | struct ext4_map_blocks *map, | |
3555 | struct ext4_ext_path *path) | |
0031462b MC |
3556 | { |
3557 | struct ext4_extent *ex; | |
dee1f973 DM |
3558 | ext4_lblk_t ee_block; |
3559 | unsigned int ee_len; | |
0031462b MC |
3560 | int depth; |
3561 | int err = 0; | |
0031462b MC |
3562 | |
3563 | depth = ext_depth(inode); | |
0031462b | 3564 | ex = path[depth].p_ext; |
dee1f973 DM |
3565 | ee_block = le32_to_cpu(ex->ee_block); |
3566 | ee_len = ext4_ext_get_actual_len(ex); | |
0031462b | 3567 | |
197217a5 YY |
3568 | ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical" |
3569 | "block %llu, max_blocks %u\n", inode->i_ino, | |
dee1f973 DM |
3570 | (unsigned long long)ee_block, ee_len); |
3571 | ||
ff95ec22 DM |
3572 | /* If extent is larger than requested it is a clear sign that we still |
3573 | * have some extent state machine issues left. So extent_split is still | |
3574 | * required. | |
3575 | * TODO: Once all related issues will be fixed this situation should be | |
3576 | * illegal. | |
3577 | */ | |
dee1f973 | 3578 | if (ee_block != map->m_lblk || ee_len > map->m_len) { |
ff95ec22 DM |
3579 | #ifdef EXT4_DEBUG |
3580 | ext4_warning("Inode (%ld) finished: extent logical block %llu," | |
3581 | " len %u; IO logical block %llu, len %u\n", | |
3582 | inode->i_ino, (unsigned long long)ee_block, ee_len, | |
3583 | (unsigned long long)map->m_lblk, map->m_len); | |
3584 | #endif | |
dee1f973 DM |
3585 | err = ext4_split_unwritten_extents(handle, inode, map, path, |
3586 | EXT4_GET_BLOCKS_CONVERT); | |
3587 | if (err < 0) | |
3588 | goto out; | |
3589 | ext4_ext_drop_refs(path); | |
107a7bd3 | 3590 | path = ext4_ext_find_extent(inode, map->m_lblk, path, 0); |
dee1f973 DM |
3591 | if (IS_ERR(path)) { |
3592 | err = PTR_ERR(path); | |
3593 | goto out; | |
3594 | } | |
3595 | depth = ext_depth(inode); | |
3596 | ex = path[depth].p_ext; | |
3597 | } | |
197217a5 | 3598 | |
0031462b MC |
3599 | err = ext4_ext_get_access(handle, inode, path + depth); |
3600 | if (err) | |
3601 | goto out; | |
3602 | /* first mark the extent as initialized */ | |
3603 | ext4_ext_mark_initialized(ex); | |
3604 | ||
197217a5 YY |
3605 | /* note: ext4_ext_correct_indexes() isn't needed here because |
3606 | * borders are not changed | |
0031462b | 3607 | */ |
ecb94f5f | 3608 | ext4_ext_try_to_merge(handle, inode, path, ex); |
197217a5 | 3609 | |
0031462b | 3610 | /* Mark modified extent as dirty */ |
ecb94f5f | 3611 | err = ext4_ext_dirty(handle, inode, path + path->p_depth); |
0031462b MC |
3612 | out: |
3613 | ext4_ext_show_leaf(inode, path); | |
3614 | return err; | |
3615 | } | |
3616 | ||
515f41c3 AK |
3617 | static void unmap_underlying_metadata_blocks(struct block_device *bdev, |
3618 | sector_t block, int count) | |
3619 | { | |
3620 | int i; | |
3621 | for (i = 0; i < count; i++) | |
3622 | unmap_underlying_metadata(bdev, block + i); | |
3623 | } | |
3624 | ||
58590b06 TT |
3625 | /* |
3626 | * Handle EOFBLOCKS_FL flag, clearing it if necessary | |
3627 | */ | |
3628 | static int check_eofblocks_fl(handle_t *handle, struct inode *inode, | |
d002ebf1 | 3629 | ext4_lblk_t lblk, |
58590b06 TT |
3630 | struct ext4_ext_path *path, |
3631 | unsigned int len) | |
3632 | { | |
3633 | int i, depth; | |
3634 | struct ext4_extent_header *eh; | |
65922cb5 | 3635 | struct ext4_extent *last_ex; |
58590b06 TT |
3636 | |
3637 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)) | |
3638 | return 0; | |
3639 | ||
3640 | depth = ext_depth(inode); | |
3641 | eh = path[depth].p_hdr; | |
58590b06 | 3642 | |
afcff5d8 LC |
3643 | /* |
3644 | * We're going to remove EOFBLOCKS_FL entirely in future so we | |
3645 | * do not care for this case anymore. Simply remove the flag | |
3646 | * if there are no extents. | |
3647 | */ | |
3648 | if (unlikely(!eh->eh_entries)) | |
3649 | goto out; | |
58590b06 TT |
3650 | last_ex = EXT_LAST_EXTENT(eh); |
3651 | /* | |
3652 | * We should clear the EOFBLOCKS_FL flag if we are writing the | |
3653 | * last block in the last extent in the file. We test this by | |
3654 | * first checking to see if the caller to | |
3655 | * ext4_ext_get_blocks() was interested in the last block (or | |
3656 | * a block beyond the last block) in the current extent. If | |
3657 | * this turns out to be false, we can bail out from this | |
3658 | * function immediately. | |
3659 | */ | |
d002ebf1 | 3660 | if (lblk + len < le32_to_cpu(last_ex->ee_block) + |
58590b06 TT |
3661 | ext4_ext_get_actual_len(last_ex)) |
3662 | return 0; | |
3663 | /* | |
3664 | * If the caller does appear to be planning to write at or | |
3665 | * beyond the end of the current extent, we then test to see | |
3666 | * if the current extent is the last extent in the file, by | |
3667 | * checking to make sure it was reached via the rightmost node | |
3668 | * at each level of the tree. | |
3669 | */ | |
3670 | for (i = depth-1; i >= 0; i--) | |
3671 | if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr)) | |
3672 | return 0; | |
afcff5d8 | 3673 | out: |
58590b06 TT |
3674 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
3675 | return ext4_mark_inode_dirty(handle, inode); | |
3676 | } | |
3677 | ||
7b415bf6 AK |
3678 | /** |
3679 | * ext4_find_delalloc_range: find delayed allocated block in the given range. | |
3680 | * | |
7d1b1fbc | 3681 | * Return 1 if there is a delalloc block in the range, otherwise 0. |
7b415bf6 | 3682 | */ |
f7fec032 ZL |
3683 | int ext4_find_delalloc_range(struct inode *inode, |
3684 | ext4_lblk_t lblk_start, | |
3685 | ext4_lblk_t lblk_end) | |
7b415bf6 | 3686 | { |
7d1b1fbc | 3687 | struct extent_status es; |
7b415bf6 | 3688 | |
e30b5dca | 3689 | ext4_es_find_delayed_extent_range(inode, lblk_start, lblk_end, &es); |
06b0c886 | 3690 | if (es.es_len == 0) |
7d1b1fbc | 3691 | return 0; /* there is no delay extent in this tree */ |
06b0c886 ZL |
3692 | else if (es.es_lblk <= lblk_start && |
3693 | lblk_start < es.es_lblk + es.es_len) | |
7d1b1fbc | 3694 | return 1; |
06b0c886 | 3695 | else if (lblk_start <= es.es_lblk && es.es_lblk <= lblk_end) |
7d1b1fbc | 3696 | return 1; |
7b415bf6 | 3697 | else |
7d1b1fbc | 3698 | return 0; |
7b415bf6 AK |
3699 | } |
3700 | ||
7d1b1fbc | 3701 | int ext4_find_delalloc_cluster(struct inode *inode, ext4_lblk_t lblk) |
7b415bf6 AK |
3702 | { |
3703 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3704 | ext4_lblk_t lblk_start, lblk_end; | |
3705 | lblk_start = lblk & (~(sbi->s_cluster_ratio - 1)); | |
3706 | lblk_end = lblk_start + sbi->s_cluster_ratio - 1; | |
3707 | ||
7d1b1fbc | 3708 | return ext4_find_delalloc_range(inode, lblk_start, lblk_end); |
7b415bf6 AK |
3709 | } |
3710 | ||
3711 | /** | |
3712 | * Determines how many complete clusters (out of those specified by the 'map') | |
3713 | * are under delalloc and were reserved quota for. | |
3714 | * This function is called when we are writing out the blocks that were | |
3715 | * originally written with their allocation delayed, but then the space was | |
3716 | * allocated using fallocate() before the delayed allocation could be resolved. | |
3717 | * The cases to look for are: | |
3718 | * ('=' indicated delayed allocated blocks | |
3719 | * '-' indicates non-delayed allocated blocks) | |
3720 | * (a) partial clusters towards beginning and/or end outside of allocated range | |
3721 | * are not delalloc'ed. | |
3722 | * Ex: | |
3723 | * |----c---=|====c====|====c====|===-c----| | |
3724 | * |++++++ allocated ++++++| | |
3725 | * ==> 4 complete clusters in above example | |
3726 | * | |
3727 | * (b) partial cluster (outside of allocated range) towards either end is | |
3728 | * marked for delayed allocation. In this case, we will exclude that | |
3729 | * cluster. | |
3730 | * Ex: | |
3731 | * |----====c========|========c========| | |
3732 | * |++++++ allocated ++++++| | |
3733 | * ==> 1 complete clusters in above example | |
3734 | * | |
3735 | * Ex: | |
3736 | * |================c================| | |
3737 | * |++++++ allocated ++++++| | |
3738 | * ==> 0 complete clusters in above example | |
3739 | * | |
3740 | * The ext4_da_update_reserve_space will be called only if we | |
3741 | * determine here that there were some "entire" clusters that span | |
3742 | * this 'allocated' range. | |
3743 | * In the non-bigalloc case, this function will just end up returning num_blks | |
3744 | * without ever calling ext4_find_delalloc_range. | |
3745 | */ | |
3746 | static unsigned int | |
3747 | get_reserved_cluster_alloc(struct inode *inode, ext4_lblk_t lblk_start, | |
3748 | unsigned int num_blks) | |
3749 | { | |
3750 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3751 | ext4_lblk_t alloc_cluster_start, alloc_cluster_end; | |
3752 | ext4_lblk_t lblk_from, lblk_to, c_offset; | |
3753 | unsigned int allocated_clusters = 0; | |
3754 | ||
3755 | alloc_cluster_start = EXT4_B2C(sbi, lblk_start); | |
3756 | alloc_cluster_end = EXT4_B2C(sbi, lblk_start + num_blks - 1); | |
3757 | ||
3758 | /* max possible clusters for this allocation */ | |
3759 | allocated_clusters = alloc_cluster_end - alloc_cluster_start + 1; | |
3760 | ||
d8990240 AK |
3761 | trace_ext4_get_reserved_cluster_alloc(inode, lblk_start, num_blks); |
3762 | ||
7b415bf6 AK |
3763 | /* Check towards left side */ |
3764 | c_offset = lblk_start & (sbi->s_cluster_ratio - 1); | |
3765 | if (c_offset) { | |
3766 | lblk_from = lblk_start & (~(sbi->s_cluster_ratio - 1)); | |
3767 | lblk_to = lblk_from + c_offset - 1; | |
3768 | ||
7d1b1fbc | 3769 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3770 | allocated_clusters--; |
3771 | } | |
3772 | ||
3773 | /* Now check towards right. */ | |
3774 | c_offset = (lblk_start + num_blks) & (sbi->s_cluster_ratio - 1); | |
3775 | if (allocated_clusters && c_offset) { | |
3776 | lblk_from = lblk_start + num_blks; | |
3777 | lblk_to = lblk_from + (sbi->s_cluster_ratio - c_offset) - 1; | |
3778 | ||
7d1b1fbc | 3779 | if (ext4_find_delalloc_range(inode, lblk_from, lblk_to)) |
7b415bf6 AK |
3780 | allocated_clusters--; |
3781 | } | |
3782 | ||
3783 | return allocated_clusters; | |
3784 | } | |
3785 | ||
0031462b MC |
3786 | static int |
3787 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | |
e35fd660 | 3788 | struct ext4_map_blocks *map, |
0031462b | 3789 | struct ext4_ext_path *path, int flags, |
e35fd660 | 3790 | unsigned int allocated, ext4_fsblk_t newblock) |
0031462b MC |
3791 | { |
3792 | int ret = 0; | |
3793 | int err = 0; | |
f45ee3a1 | 3794 | ext4_io_end_t *io = ext4_inode_aio(inode); |
0031462b | 3795 | |
88635ca2 ZL |
3796 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical " |
3797 | "block %llu, max_blocks %u, flags %x, allocated %u\n", | |
e35fd660 | 3798 | inode->i_ino, (unsigned long long)map->m_lblk, map->m_len, |
0031462b MC |
3799 | flags, allocated); |
3800 | ext4_ext_show_leaf(inode, path); | |
3801 | ||
27dd4385 LC |
3802 | /* |
3803 | * When writing into uninitialized space, we should not fail to | |
3804 | * allocate metadata blocks for the new extent block if needed. | |
3805 | */ | |
3806 | flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
3807 | ||
b5645534 ZL |
3808 | trace_ext4_ext_handle_uninitialized_extents(inode, map, flags, |
3809 | allocated, newblock); | |
d8990240 | 3810 | |
c7064ef1 | 3811 | /* get_block() before submit the IO, split the extent */ |
744692dc | 3812 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) { |
e35fd660 TT |
3813 | ret = ext4_split_unwritten_extents(handle, inode, map, |
3814 | path, flags); | |
82e54229 DM |
3815 | if (ret <= 0) |
3816 | goto out; | |
5f524950 M |
3817 | /* |
3818 | * Flag the inode(non aio case) or end_io struct (aio case) | |
25985edc | 3819 | * that this IO needs to conversion to written when IO is |
5f524950 M |
3820 | * completed |
3821 | */ | |
0edeb71d TM |
3822 | if (io) |
3823 | ext4_set_io_unwritten_flag(inode, io); | |
3824 | else | |
19f5fb7a | 3825 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
a25a4e1a | 3826 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
744692dc | 3827 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 3828 | map->m_flags |= EXT4_MAP_UNINIT; |
0031462b MC |
3829 | goto out; |
3830 | } | |
c7064ef1 | 3831 | /* IO end_io complete, convert the filled extent to written */ |
744692dc | 3832 | if ((flags & EXT4_GET_BLOCKS_CONVERT)) { |
dee1f973 | 3833 | ret = ext4_convert_unwritten_extents_endio(handle, inode, map, |
0031462b | 3834 | path); |
58590b06 | 3835 | if (ret >= 0) { |
b436b9be | 3836 | ext4_update_inode_fsync_trans(handle, inode, 1); |
d002ebf1 ES |
3837 | err = check_eofblocks_fl(handle, inode, map->m_lblk, |
3838 | path, map->m_len); | |
58590b06 TT |
3839 | } else |
3840 | err = ret; | |
cdee7843 ZL |
3841 | map->m_flags |= EXT4_MAP_MAPPED; |
3842 | if (allocated > map->m_len) | |
3843 | allocated = map->m_len; | |
3844 | map->m_len = allocated; | |
0031462b MC |
3845 | goto out2; |
3846 | } | |
3847 | /* buffered IO case */ | |
3848 | /* | |
3849 | * repeat fallocate creation request | |
3850 | * we already have an unwritten extent | |
3851 | */ | |
a25a4e1a ZL |
3852 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) { |
3853 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
0031462b | 3854 | goto map_out; |
a25a4e1a | 3855 | } |
0031462b MC |
3856 | |
3857 | /* buffered READ or buffered write_begin() lookup */ | |
3858 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | |
3859 | /* | |
3860 | * We have blocks reserved already. We | |
3861 | * return allocated blocks so that delalloc | |
3862 | * won't do block reservation for us. But | |
3863 | * the buffer head will be unmapped so that | |
3864 | * a read from the block returns 0s. | |
3865 | */ | |
e35fd660 | 3866 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
0031462b MC |
3867 | goto out1; |
3868 | } | |
3869 | ||
3870 | /* buffered write, writepage time, convert*/ | |
27dd4385 | 3871 | ret = ext4_ext_convert_to_initialized(handle, inode, map, path, flags); |
a4e5d88b | 3872 | if (ret >= 0) |
b436b9be | 3873 | ext4_update_inode_fsync_trans(handle, inode, 1); |
0031462b MC |
3874 | out: |
3875 | if (ret <= 0) { | |
3876 | err = ret; | |
3877 | goto out2; | |
3878 | } else | |
3879 | allocated = ret; | |
e35fd660 | 3880 | map->m_flags |= EXT4_MAP_NEW; |
515f41c3 AK |
3881 | /* |
3882 | * if we allocated more blocks than requested | |
3883 | * we need to make sure we unmap the extra block | |
3884 | * allocated. The actual needed block will get | |
3885 | * unmapped later when we find the buffer_head marked | |
3886 | * new. | |
3887 | */ | |
e35fd660 | 3888 | if (allocated > map->m_len) { |
515f41c3 | 3889 | unmap_underlying_metadata_blocks(inode->i_sb->s_bdev, |
e35fd660 TT |
3890 | newblock + map->m_len, |
3891 | allocated - map->m_len); | |
3892 | allocated = map->m_len; | |
515f41c3 | 3893 | } |
3a225670 | 3894 | map->m_len = allocated; |
5f634d06 AK |
3895 | |
3896 | /* | |
3897 | * If we have done fallocate with the offset that is already | |
3898 | * delayed allocated, we would have block reservation | |
3899 | * and quota reservation done in the delayed write path. | |
3900 | * But fallocate would have already updated quota and block | |
3901 | * count for this offset. So cancel these reservation | |
3902 | */ | |
7b415bf6 AK |
3903 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
3904 | unsigned int reserved_clusters; | |
3905 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
3906 | map->m_lblk, map->m_len); | |
3907 | if (reserved_clusters) | |
3908 | ext4_da_update_reserve_space(inode, | |
3909 | reserved_clusters, | |
3910 | 0); | |
3911 | } | |
5f634d06 | 3912 | |
0031462b | 3913 | map_out: |
e35fd660 | 3914 | map->m_flags |= EXT4_MAP_MAPPED; |
a4e5d88b DM |
3915 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) { |
3916 | err = check_eofblocks_fl(handle, inode, map->m_lblk, path, | |
3917 | map->m_len); | |
3918 | if (err < 0) | |
3919 | goto out2; | |
3920 | } | |
0031462b | 3921 | out1: |
e35fd660 TT |
3922 | if (allocated > map->m_len) |
3923 | allocated = map->m_len; | |
0031462b | 3924 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
3925 | map->m_pblk = newblock; |
3926 | map->m_len = allocated; | |
0031462b MC |
3927 | out2: |
3928 | if (path) { | |
3929 | ext4_ext_drop_refs(path); | |
3930 | kfree(path); | |
3931 | } | |
3932 | return err ? err : allocated; | |
3933 | } | |
58590b06 | 3934 | |
4d33b1ef TT |
3935 | /* |
3936 | * get_implied_cluster_alloc - check to see if the requested | |
3937 | * allocation (in the map structure) overlaps with a cluster already | |
3938 | * allocated in an extent. | |
d8990240 | 3939 | * @sb The filesystem superblock structure |
4d33b1ef TT |
3940 | * @map The requested lblk->pblk mapping |
3941 | * @ex The extent structure which might contain an implied | |
3942 | * cluster allocation | |
3943 | * | |
3944 | * This function is called by ext4_ext_map_blocks() after we failed to | |
3945 | * find blocks that were already in the inode's extent tree. Hence, | |
3946 | * we know that the beginning of the requested region cannot overlap | |
3947 | * the extent from the inode's extent tree. There are three cases we | |
3948 | * want to catch. The first is this case: | |
3949 | * | |
3950 | * |--- cluster # N--| | |
3951 | * |--- extent ---| |---- requested region ---| | |
3952 | * |==========| | |
3953 | * | |
3954 | * The second case that we need to test for is this one: | |
3955 | * | |
3956 | * |--------- cluster # N ----------------| | |
3957 | * |--- requested region --| |------- extent ----| | |
3958 | * |=======================| | |
3959 | * | |
3960 | * The third case is when the requested region lies between two extents | |
3961 | * within the same cluster: | |
3962 | * |------------- cluster # N-------------| | |
3963 | * |----- ex -----| |---- ex_right ----| | |
3964 | * |------ requested region ------| | |
3965 | * |================| | |
3966 | * | |
3967 | * In each of the above cases, we need to set the map->m_pblk and | |
3968 | * map->m_len so it corresponds to the return the extent labelled as | |
3969 | * "|====|" from cluster #N, since it is already in use for data in | |
3970 | * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to | |
3971 | * signal to ext4_ext_map_blocks() that map->m_pblk should be treated | |
3972 | * as a new "allocated" block region. Otherwise, we will return 0 and | |
3973 | * ext4_ext_map_blocks() will then allocate one or more new clusters | |
3974 | * by calling ext4_mb_new_blocks(). | |
3975 | */ | |
d8990240 | 3976 | static int get_implied_cluster_alloc(struct super_block *sb, |
4d33b1ef TT |
3977 | struct ext4_map_blocks *map, |
3978 | struct ext4_extent *ex, | |
3979 | struct ext4_ext_path *path) | |
3980 | { | |
d8990240 | 3981 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
4d33b1ef TT |
3982 | ext4_lblk_t c_offset = map->m_lblk & (sbi->s_cluster_ratio-1); |
3983 | ext4_lblk_t ex_cluster_start, ex_cluster_end; | |
14d7f3ef | 3984 | ext4_lblk_t rr_cluster_start; |
4d33b1ef TT |
3985 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
3986 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); | |
3987 | unsigned short ee_len = ext4_ext_get_actual_len(ex); | |
3988 | ||
3989 | /* The extent passed in that we are trying to match */ | |
3990 | ex_cluster_start = EXT4_B2C(sbi, ee_block); | |
3991 | ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1); | |
3992 | ||
3993 | /* The requested region passed into ext4_map_blocks() */ | |
3994 | rr_cluster_start = EXT4_B2C(sbi, map->m_lblk); | |
4d33b1ef TT |
3995 | |
3996 | if ((rr_cluster_start == ex_cluster_end) || | |
3997 | (rr_cluster_start == ex_cluster_start)) { | |
3998 | if (rr_cluster_start == ex_cluster_end) | |
3999 | ee_start += ee_len - 1; | |
4000 | map->m_pblk = (ee_start & ~(sbi->s_cluster_ratio - 1)) + | |
4001 | c_offset; | |
4002 | map->m_len = min(map->m_len, | |
4003 | (unsigned) sbi->s_cluster_ratio - c_offset); | |
4004 | /* | |
4005 | * Check for and handle this case: | |
4006 | * | |
4007 | * |--------- cluster # N-------------| | |
4008 | * |------- extent ----| | |
4009 | * |--- requested region ---| | |
4010 | * |===========| | |
4011 | */ | |
4012 | ||
4013 | if (map->m_lblk < ee_block) | |
4014 | map->m_len = min(map->m_len, ee_block - map->m_lblk); | |
4015 | ||
4016 | /* | |
4017 | * Check for the case where there is already another allocated | |
4018 | * block to the right of 'ex' but before the end of the cluster. | |
4019 | * | |
4020 | * |------------- cluster # N-------------| | |
4021 | * |----- ex -----| |---- ex_right ----| | |
4022 | * |------ requested region ------| | |
4023 | * |================| | |
4024 | */ | |
4025 | if (map->m_lblk > ee_block) { | |
4026 | ext4_lblk_t next = ext4_ext_next_allocated_block(path); | |
4027 | map->m_len = min(map->m_len, next - map->m_lblk); | |
4028 | } | |
d8990240 AK |
4029 | |
4030 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1); | |
4d33b1ef TT |
4031 | return 1; |
4032 | } | |
d8990240 AK |
4033 | |
4034 | trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0); | |
4d33b1ef TT |
4035 | return 0; |
4036 | } | |
4037 | ||
4038 | ||
c278bfec | 4039 | /* |
f5ab0d1f MC |
4040 | * Block allocation/map/preallocation routine for extents based files |
4041 | * | |
4042 | * | |
c278bfec | 4043 | * Need to be called with |
0e855ac8 AK |
4044 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
4045 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
f5ab0d1f MC |
4046 | * |
4047 | * return > 0, number of of blocks already mapped/allocated | |
4048 | * if create == 0 and these are pre-allocated blocks | |
4049 | * buffer head is unmapped | |
4050 | * otherwise blocks are mapped | |
4051 | * | |
4052 | * return = 0, if plain look up failed (blocks have not been allocated) | |
4053 | * buffer head is unmapped | |
4054 | * | |
4055 | * return < 0, error case. | |
c278bfec | 4056 | */ |
e35fd660 TT |
4057 | int ext4_ext_map_blocks(handle_t *handle, struct inode *inode, |
4058 | struct ext4_map_blocks *map, int flags) | |
a86c6181 AT |
4059 | { |
4060 | struct ext4_ext_path *path = NULL; | |
4d33b1ef TT |
4061 | struct ext4_extent newex, *ex, *ex2; |
4062 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0562e0ba | 4063 | ext4_fsblk_t newblock = 0; |
37794732 | 4064 | int free_on_err = 0, err = 0, depth; |
4d33b1ef | 4065 | unsigned int allocated = 0, offset = 0; |
81fdbb4a | 4066 | unsigned int allocated_clusters = 0; |
c9de560d | 4067 | struct ext4_allocation_request ar; |
f45ee3a1 | 4068 | ext4_io_end_t *io = ext4_inode_aio(inode); |
4d33b1ef | 4069 | ext4_lblk_t cluster_offset; |
82e54229 | 4070 | int set_unwritten = 0; |
a86c6181 | 4071 | |
84fe3bef | 4072 | ext_debug("blocks %u/%u requested for inode %lu\n", |
e35fd660 | 4073 | map->m_lblk, map->m_len, inode->i_ino); |
0562e0ba | 4074 | trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
a86c6181 | 4075 | |
a86c6181 | 4076 | /* find extent for this block */ |
107a7bd3 | 4077 | path = ext4_ext_find_extent(inode, map->m_lblk, NULL, 0); |
a86c6181 AT |
4078 | if (IS_ERR(path)) { |
4079 | err = PTR_ERR(path); | |
4080 | path = NULL; | |
4081 | goto out2; | |
4082 | } | |
4083 | ||
4084 | depth = ext_depth(inode); | |
4085 | ||
4086 | /* | |
d0d856e8 RD |
4087 | * consistent leaf must not be empty; |
4088 | * this situation is possible, though, _during_ tree modification; | |
a86c6181 AT |
4089 | * this is why assert can't be put in ext4_ext_find_extent() |
4090 | */ | |
273df556 FM |
4091 | if (unlikely(path[depth].p_ext == NULL && depth != 0)) { |
4092 | EXT4_ERROR_INODE(inode, "bad extent address " | |
f70f362b TT |
4093 | "lblock: %lu, depth: %d pblock %lld", |
4094 | (unsigned long) map->m_lblk, depth, | |
4095 | path[depth].p_block); | |
034fb4c9 SP |
4096 | err = -EIO; |
4097 | goto out2; | |
4098 | } | |
a86c6181 | 4099 | |
7e028976 AM |
4100 | ex = path[depth].p_ext; |
4101 | if (ex) { | |
725d26d3 | 4102 | ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block); |
bf89d16f | 4103 | ext4_fsblk_t ee_start = ext4_ext_pblock(ex); |
a2df2a63 | 4104 | unsigned short ee_len; |
471d4011 SB |
4105 | |
4106 | /* | |
471d4011 | 4107 | * Uninitialized extents are treated as holes, except that |
56055d3a | 4108 | * we split out initialized portions during a write. |
471d4011 | 4109 | */ |
a2df2a63 | 4110 | ee_len = ext4_ext_get_actual_len(ex); |
d8990240 AK |
4111 | |
4112 | trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len); | |
4113 | ||
d0d856e8 | 4114 | /* if found extent covers block, simply return it */ |
e35fd660 TT |
4115 | if (in_range(map->m_lblk, ee_block, ee_len)) { |
4116 | newblock = map->m_lblk - ee_block + ee_start; | |
d0d856e8 | 4117 | /* number of remaining blocks in the extent */ |
e35fd660 TT |
4118 | allocated = ee_len - (map->m_lblk - ee_block); |
4119 | ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk, | |
4120 | ee_block, ee_len, newblock); | |
56055d3a | 4121 | |
69eb33dc | 4122 | if (!ext4_ext_is_uninitialized(ex)) |
7877191c | 4123 | goto out; |
69eb33dc | 4124 | |
37794732 | 4125 | allocated = ext4_ext_handle_uninitialized_extents( |
7877191c LC |
4126 | handle, inode, map, path, flags, |
4127 | allocated, newblock); | |
37794732 | 4128 | goto out3; |
a86c6181 AT |
4129 | } |
4130 | } | |
4131 | ||
7b415bf6 | 4132 | if ((sbi->s_cluster_ratio > 1) && |
7d1b1fbc | 4133 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
7b415bf6 AK |
4134 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4135 | ||
a86c6181 | 4136 | /* |
d0d856e8 | 4137 | * requested block isn't allocated yet; |
a86c6181 AT |
4138 | * we couldn't try to create block if create flag is zero |
4139 | */ | |
c2177057 | 4140 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { |
56055d3a AA |
4141 | /* |
4142 | * put just found gap into cache to speed up | |
4143 | * subsequent requests | |
4144 | */ | |
d100eef2 ZL |
4145 | if ((flags & EXT4_GET_BLOCKS_NO_PUT_HOLE) == 0) |
4146 | ext4_ext_put_gap_in_cache(inode, path, map->m_lblk); | |
a86c6181 AT |
4147 | goto out2; |
4148 | } | |
4d33b1ef | 4149 | |
a86c6181 | 4150 | /* |
c2ea3fde | 4151 | * Okay, we need to do block allocation. |
63f57933 | 4152 | */ |
7b415bf6 | 4153 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4154 | newex.ee_block = cpu_to_le32(map->m_lblk); |
4155 | cluster_offset = map->m_lblk & (sbi->s_cluster_ratio-1); | |
4156 | ||
4157 | /* | |
4158 | * If we are doing bigalloc, check to see if the extent returned | |
4159 | * by ext4_ext_find_extent() implies a cluster we can use. | |
4160 | */ | |
4161 | if (cluster_offset && ex && | |
d8990240 | 4162 | get_implied_cluster_alloc(inode->i_sb, map, ex, path)) { |
4d33b1ef TT |
4163 | ar.len = allocated = map->m_len; |
4164 | newblock = map->m_pblk; | |
7b415bf6 | 4165 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4166 | goto got_allocated_blocks; |
4167 | } | |
a86c6181 | 4168 | |
c9de560d | 4169 | /* find neighbour allocated blocks */ |
e35fd660 | 4170 | ar.lleft = map->m_lblk; |
c9de560d AT |
4171 | err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft); |
4172 | if (err) | |
4173 | goto out2; | |
e35fd660 | 4174 | ar.lright = map->m_lblk; |
4d33b1ef TT |
4175 | ex2 = NULL; |
4176 | err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2); | |
c9de560d AT |
4177 | if (err) |
4178 | goto out2; | |
25d14f98 | 4179 | |
4d33b1ef TT |
4180 | /* Check if the extent after searching to the right implies a |
4181 | * cluster we can use. */ | |
4182 | if ((sbi->s_cluster_ratio > 1) && ex2 && | |
d8990240 | 4183 | get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) { |
4d33b1ef TT |
4184 | ar.len = allocated = map->m_len; |
4185 | newblock = map->m_pblk; | |
7b415bf6 | 4186 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
4d33b1ef TT |
4187 | goto got_allocated_blocks; |
4188 | } | |
4189 | ||
749269fa AA |
4190 | /* |
4191 | * See if request is beyond maximum number of blocks we can have in | |
4192 | * a single extent. For an initialized extent this limit is | |
4193 | * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is | |
4194 | * EXT_UNINIT_MAX_LEN. | |
4195 | */ | |
e35fd660 | 4196 | if (map->m_len > EXT_INIT_MAX_LEN && |
c2177057 | 4197 | !(flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 TT |
4198 | map->m_len = EXT_INIT_MAX_LEN; |
4199 | else if (map->m_len > EXT_UNINIT_MAX_LEN && | |
c2177057 | 4200 | (flags & EXT4_GET_BLOCKS_UNINIT_EXT)) |
e35fd660 | 4201 | map->m_len = EXT_UNINIT_MAX_LEN; |
749269fa | 4202 | |
e35fd660 | 4203 | /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */ |
e35fd660 | 4204 | newex.ee_len = cpu_to_le16(map->m_len); |
4d33b1ef | 4205 | err = ext4_ext_check_overlap(sbi, inode, &newex, path); |
25d14f98 | 4206 | if (err) |
b939e376 | 4207 | allocated = ext4_ext_get_actual_len(&newex); |
25d14f98 | 4208 | else |
e35fd660 | 4209 | allocated = map->m_len; |
c9de560d AT |
4210 | |
4211 | /* allocate new block */ | |
4212 | ar.inode = inode; | |
e35fd660 TT |
4213 | ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk); |
4214 | ar.logical = map->m_lblk; | |
4d33b1ef TT |
4215 | /* |
4216 | * We calculate the offset from the beginning of the cluster | |
4217 | * for the logical block number, since when we allocate a | |
4218 | * physical cluster, the physical block should start at the | |
4219 | * same offset from the beginning of the cluster. This is | |
4220 | * needed so that future calls to get_implied_cluster_alloc() | |
4221 | * work correctly. | |
4222 | */ | |
4223 | offset = map->m_lblk & (sbi->s_cluster_ratio - 1); | |
4224 | ar.len = EXT4_NUM_B2C(sbi, offset+allocated); | |
4225 | ar.goal -= offset; | |
4226 | ar.logical -= offset; | |
c9de560d AT |
4227 | if (S_ISREG(inode->i_mode)) |
4228 | ar.flags = EXT4_MB_HINT_DATA; | |
4229 | else | |
4230 | /* disable in-core preallocation for non-regular files */ | |
4231 | ar.flags = 0; | |
556b27ab VH |
4232 | if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE) |
4233 | ar.flags |= EXT4_MB_HINT_NOPREALLOC; | |
c9de560d | 4234 | newblock = ext4_mb_new_blocks(handle, &ar, &err); |
a86c6181 AT |
4235 | if (!newblock) |
4236 | goto out2; | |
84fe3bef | 4237 | ext_debug("allocate new block: goal %llu, found %llu/%u\n", |
498e5f24 | 4238 | ar.goal, newblock, allocated); |
4d33b1ef | 4239 | free_on_err = 1; |
7b415bf6 | 4240 | allocated_clusters = ar.len; |
4d33b1ef TT |
4241 | ar.len = EXT4_C2B(sbi, ar.len) - offset; |
4242 | if (ar.len > allocated) | |
4243 | ar.len = allocated; | |
a86c6181 | 4244 | |
4d33b1ef | 4245 | got_allocated_blocks: |
a86c6181 | 4246 | /* try to insert new extent into found leaf and return */ |
4d33b1ef | 4247 | ext4_ext_store_pblock(&newex, newblock + offset); |
c9de560d | 4248 | newex.ee_len = cpu_to_le16(ar.len); |
8d5d02e6 MC |
4249 | /* Mark uninitialized */ |
4250 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){ | |
a2df2a63 | 4251 | ext4_ext_mark_uninitialized(&newex); |
a25a4e1a | 4252 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
8d5d02e6 | 4253 | /* |
744692dc | 4254 | * io_end structure was created for every IO write to an |
25985edc | 4255 | * uninitialized extent. To avoid unnecessary conversion, |
744692dc | 4256 | * here we flag the IO that really needs the conversion. |
5f524950 | 4257 | * For non asycn direct IO case, flag the inode state |
25985edc | 4258 | * that we need to perform conversion when IO is done. |
8d5d02e6 | 4259 | */ |
82e54229 DM |
4260 | if ((flags & EXT4_GET_BLOCKS_PRE_IO)) |
4261 | set_unwritten = 1; | |
744692dc | 4262 | if (ext4_should_dioread_nolock(inode)) |
e35fd660 | 4263 | map->m_flags |= EXT4_MAP_UNINIT; |
8d5d02e6 | 4264 | } |
c8d46e41 | 4265 | |
a4e5d88b DM |
4266 | err = 0; |
4267 | if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) | |
4268 | err = check_eofblocks_fl(handle, inode, map->m_lblk, | |
4269 | path, ar.len); | |
575a1d4b JZ |
4270 | if (!err) |
4271 | err = ext4_ext_insert_extent(handle, inode, path, | |
4272 | &newex, flags); | |
82e54229 DM |
4273 | |
4274 | if (!err && set_unwritten) { | |
4275 | if (io) | |
4276 | ext4_set_io_unwritten_flag(inode, io); | |
4277 | else | |
4278 | ext4_set_inode_state(inode, | |
4279 | EXT4_STATE_DIO_UNWRITTEN); | |
4280 | } | |
4281 | ||
4d33b1ef | 4282 | if (err && free_on_err) { |
7132de74 MP |
4283 | int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ? |
4284 | EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0; | |
315054f0 | 4285 | /* free data blocks we just allocated */ |
c9de560d AT |
4286 | /* not a good idea to call discard here directly, |
4287 | * but otherwise we'd need to call it every free() */ | |
c2ea3fde | 4288 | ext4_discard_preallocations(inode); |
c8e15130 TT |
4289 | ext4_free_blocks(handle, inode, NULL, newblock, |
4290 | EXT4_C2B(sbi, allocated_clusters), fb_flags); | |
a86c6181 | 4291 | goto out2; |
315054f0 | 4292 | } |
a86c6181 | 4293 | |
a86c6181 | 4294 | /* previous routine could use block we allocated */ |
bf89d16f | 4295 | newblock = ext4_ext_pblock(&newex); |
b939e376 | 4296 | allocated = ext4_ext_get_actual_len(&newex); |
e35fd660 TT |
4297 | if (allocated > map->m_len) |
4298 | allocated = map->m_len; | |
4299 | map->m_flags |= EXT4_MAP_NEW; | |
a86c6181 | 4300 | |
5f634d06 AK |
4301 | /* |
4302 | * Update reserved blocks/metadata blocks after successful | |
4303 | * block allocation which had been deferred till now. | |
4304 | */ | |
7b415bf6 | 4305 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
81fdbb4a | 4306 | unsigned int reserved_clusters; |
7b415bf6 | 4307 | /* |
81fdbb4a | 4308 | * Check how many clusters we had reserved this allocated range |
7b415bf6 AK |
4309 | */ |
4310 | reserved_clusters = get_reserved_cluster_alloc(inode, | |
4311 | map->m_lblk, allocated); | |
4312 | if (map->m_flags & EXT4_MAP_FROM_CLUSTER) { | |
4313 | if (reserved_clusters) { | |
4314 | /* | |
4315 | * We have clusters reserved for this range. | |
4316 | * But since we are not doing actual allocation | |
4317 | * and are simply using blocks from previously | |
4318 | * allocated cluster, we should release the | |
4319 | * reservation and not claim quota. | |
4320 | */ | |
4321 | ext4_da_update_reserve_space(inode, | |
4322 | reserved_clusters, 0); | |
4323 | } | |
4324 | } else { | |
4325 | BUG_ON(allocated_clusters < reserved_clusters); | |
7b415bf6 | 4326 | if (reserved_clusters < allocated_clusters) { |
5356f261 | 4327 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 AK |
4328 | int reservation = allocated_clusters - |
4329 | reserved_clusters; | |
4330 | /* | |
4331 | * It seems we claimed few clusters outside of | |
4332 | * the range of this allocation. We should give | |
4333 | * it back to the reservation pool. This can | |
4334 | * happen in the following case: | |
4335 | * | |
4336 | * * Suppose s_cluster_ratio is 4 (i.e., each | |
4337 | * cluster has 4 blocks. Thus, the clusters | |
4338 | * are [0-3],[4-7],[8-11]... | |
4339 | * * First comes delayed allocation write for | |
4340 | * logical blocks 10 & 11. Since there were no | |
4341 | * previous delayed allocated blocks in the | |
4342 | * range [8-11], we would reserve 1 cluster | |
4343 | * for this write. | |
4344 | * * Next comes write for logical blocks 3 to 8. | |
4345 | * In this case, we will reserve 2 clusters | |
4346 | * (for [0-3] and [4-7]; and not for [8-11] as | |
4347 | * that range has a delayed allocated blocks. | |
4348 | * Thus total reserved clusters now becomes 3. | |
4349 | * * Now, during the delayed allocation writeout | |
4350 | * time, we will first write blocks [3-8] and | |
4351 | * allocate 3 clusters for writing these | |
4352 | * blocks. Also, we would claim all these | |
4353 | * three clusters above. | |
4354 | * * Now when we come here to writeout the | |
4355 | * blocks [10-11], we would expect to claim | |
4356 | * the reservation of 1 cluster we had made | |
4357 | * (and we would claim it since there are no | |
4358 | * more delayed allocated blocks in the range | |
4359 | * [8-11]. But our reserved cluster count had | |
4360 | * already gone to 0. | |
4361 | * | |
4362 | * Thus, at the step 4 above when we determine | |
4363 | * that there are still some unwritten delayed | |
4364 | * allocated blocks outside of our current | |
4365 | * block range, we should increment the | |
4366 | * reserved clusters count so that when the | |
4367 | * remaining blocks finally gets written, we | |
4368 | * could claim them. | |
4369 | */ | |
5356f261 AK |
4370 | dquot_reserve_block(inode, |
4371 | EXT4_C2B(sbi, reservation)); | |
4372 | spin_lock(&ei->i_block_reservation_lock); | |
4373 | ei->i_reserved_data_blocks += reservation; | |
4374 | spin_unlock(&ei->i_block_reservation_lock); | |
7b415bf6 | 4375 | } |
232ec872 LC |
4376 | /* |
4377 | * We will claim quota for all newly allocated blocks. | |
4378 | * We're updating the reserved space *after* the | |
4379 | * correction above so we do not accidentally free | |
4380 | * all the metadata reservation because we might | |
4381 | * actually need it later on. | |
4382 | */ | |
4383 | ext4_da_update_reserve_space(inode, allocated_clusters, | |
4384 | 1); | |
7b415bf6 AK |
4385 | } |
4386 | } | |
5f634d06 | 4387 | |
b436b9be JK |
4388 | /* |
4389 | * Cache the extent and update transaction to commit on fdatasync only | |
4390 | * when it is _not_ an uninitialized extent. | |
4391 | */ | |
69eb33dc | 4392 | if ((flags & EXT4_GET_BLOCKS_UNINIT_EXT) == 0) |
b436b9be | 4393 | ext4_update_inode_fsync_trans(handle, inode, 1); |
69eb33dc | 4394 | else |
b436b9be | 4395 | ext4_update_inode_fsync_trans(handle, inode, 0); |
a86c6181 | 4396 | out: |
e35fd660 TT |
4397 | if (allocated > map->m_len) |
4398 | allocated = map->m_len; | |
a86c6181 | 4399 | ext4_ext_show_leaf(inode, path); |
e35fd660 TT |
4400 | map->m_flags |= EXT4_MAP_MAPPED; |
4401 | map->m_pblk = newblock; | |
4402 | map->m_len = allocated; | |
a86c6181 AT |
4403 | out2: |
4404 | if (path) { | |
4405 | ext4_ext_drop_refs(path); | |
4406 | kfree(path); | |
4407 | } | |
e861304b | 4408 | |
37794732 | 4409 | out3: |
63b99968 TT |
4410 | trace_ext4_ext_map_blocks_exit(inode, flags, map, |
4411 | err ? err : allocated); | |
4412 | ext4_es_lru_add(inode); | |
7877191c | 4413 | return err ? err : allocated; |
a86c6181 AT |
4414 | } |
4415 | ||
819c4920 | 4416 | void ext4_ext_truncate(handle_t *handle, struct inode *inode) |
a86c6181 | 4417 | { |
a86c6181 | 4418 | struct super_block *sb = inode->i_sb; |
725d26d3 | 4419 | ext4_lblk_t last_block; |
a86c6181 AT |
4420 | int err = 0; |
4421 | ||
a86c6181 | 4422 | /* |
d0d856e8 RD |
4423 | * TODO: optimization is possible here. |
4424 | * Probably we need not scan at all, | |
4425 | * because page truncation is enough. | |
a86c6181 | 4426 | */ |
a86c6181 AT |
4427 | |
4428 | /* we have to know where to truncate from in crash case */ | |
4429 | EXT4_I(inode)->i_disksize = inode->i_size; | |
4430 | ext4_mark_inode_dirty(handle, inode); | |
4431 | ||
4432 | last_block = (inode->i_size + sb->s_blocksize - 1) | |
4433 | >> EXT4_BLOCK_SIZE_BITS(sb); | |
8acd5e9b | 4434 | retry: |
51865fda ZL |
4435 | err = ext4_es_remove_extent(inode, last_block, |
4436 | EXT_MAX_BLOCKS - last_block); | |
94eec0fc | 4437 | if (err == -ENOMEM) { |
8acd5e9b TT |
4438 | cond_resched(); |
4439 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
4440 | goto retry; | |
4441 | } | |
4442 | if (err) { | |
4443 | ext4_std_error(inode->i_sb, err); | |
4444 | return; | |
4445 | } | |
5f95d21f | 4446 | err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1); |
8acd5e9b | 4447 | ext4_std_error(inode->i_sb, err); |
a86c6181 AT |
4448 | } |
4449 | ||
fd28784a AK |
4450 | static void ext4_falloc_update_inode(struct inode *inode, |
4451 | int mode, loff_t new_size, int update_ctime) | |
4452 | { | |
4453 | struct timespec now; | |
4454 | ||
4455 | if (update_ctime) { | |
4456 | now = current_fs_time(inode->i_sb); | |
4457 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4458 | inode->i_ctime = now; | |
4459 | } | |
4460 | /* | |
4461 | * Update only when preallocation was requested beyond | |
4462 | * the file size. | |
4463 | */ | |
cf17fea6 AK |
4464 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
4465 | if (new_size > i_size_read(inode)) | |
4466 | i_size_write(inode, new_size); | |
4467 | if (new_size > EXT4_I(inode)->i_disksize) | |
4468 | ext4_update_i_disksize(inode, new_size); | |
c8d46e41 JZ |
4469 | } else { |
4470 | /* | |
4471 | * Mark that we allocate beyond EOF so the subsequent truncate | |
4472 | * can proceed even if the new size is the same as i_size. | |
4473 | */ | |
4474 | if (new_size > i_size_read(inode)) | |
12e9b892 | 4475 | ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
fd28784a AK |
4476 | } |
4477 | ||
4478 | } | |
4479 | ||
a2df2a63 | 4480 | /* |
2fe17c10 | 4481 | * preallocate space for a file. This implements ext4's fallocate file |
a2df2a63 AA |
4482 | * operation, which gets called from sys_fallocate system call. |
4483 | * For block-mapped files, posix_fallocate should fall back to the method | |
4484 | * of writing zeroes to the required new blocks (the same behavior which is | |
4485 | * expected for file systems which do not support fallocate() system call). | |
4486 | */ | |
2fe17c10 | 4487 | long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len) |
a2df2a63 | 4488 | { |
496ad9aa | 4489 | struct inode *inode = file_inode(file); |
a2df2a63 | 4490 | handle_t *handle; |
fd28784a | 4491 | loff_t new_size; |
498e5f24 | 4492 | unsigned int max_blocks; |
a2df2a63 AA |
4493 | int ret = 0; |
4494 | int ret2 = 0; | |
4495 | int retries = 0; | |
a4e5d88b | 4496 | int flags; |
2ed88685 | 4497 | struct ext4_map_blocks map; |
a2df2a63 AA |
4498 | unsigned int credits, blkbits = inode->i_blkbits; |
4499 | ||
a4bb6b64 AH |
4500 | /* Return error if mode is not supported */ |
4501 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
4502 | return -EOPNOTSUPP; | |
4503 | ||
4504 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
aeb2817a | 4505 | return ext4_punch_hole(inode, offset, len); |
a4bb6b64 | 4506 | |
0c8d414f TM |
4507 | ret = ext4_convert_inline_data(inode); |
4508 | if (ret) | |
4509 | return ret; | |
4510 | ||
8bad6fc8 ZL |
4511 | /* |
4512 | * currently supporting (pre)allocate mode for extent-based | |
4513 | * files _only_ | |
4514 | */ | |
4515 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
4516 | return -EOPNOTSUPP; | |
4517 | ||
0562e0ba | 4518 | trace_ext4_fallocate_enter(inode, offset, len, mode); |
2ed88685 | 4519 | map.m_lblk = offset >> blkbits; |
fd28784a AK |
4520 | /* |
4521 | * We can't just convert len to max_blocks because | |
4522 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4523 | */ | |
a2df2a63 | 4524 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) |
2ed88685 | 4525 | - map.m_lblk; |
a2df2a63 | 4526 | /* |
f3bd1f3f | 4527 | * credits to insert 1 extent into extent tree |
a2df2a63 | 4528 | */ |
f3bd1f3f | 4529 | credits = ext4_chunk_trans_blocks(inode, max_blocks); |
55bd725a | 4530 | mutex_lock(&inode->i_mutex); |
6d19c42b NK |
4531 | ret = inode_newsize_ok(inode, (len + offset)); |
4532 | if (ret) { | |
4533 | mutex_unlock(&inode->i_mutex); | |
0562e0ba | 4534 | trace_ext4_fallocate_exit(inode, offset, max_blocks, ret); |
6d19c42b NK |
4535 | return ret; |
4536 | } | |
3c6fe770 | 4537 | flags = EXT4_GET_BLOCKS_CREATE_UNINIT_EXT; |
a4e5d88b DM |
4538 | if (mode & FALLOC_FL_KEEP_SIZE) |
4539 | flags |= EXT4_GET_BLOCKS_KEEP_SIZE; | |
3c6fe770 GH |
4540 | /* |
4541 | * Don't normalize the request if it can fit in one extent so | |
4542 | * that it doesn't get unnecessarily split into multiple | |
4543 | * extents. | |
4544 | */ | |
4545 | if (len <= EXT_UNINIT_MAX_LEN << blkbits) | |
4546 | flags |= EXT4_GET_BLOCKS_NO_NORMALIZE; | |
60d4616f | 4547 | |
a2df2a63 AA |
4548 | retry: |
4549 | while (ret >= 0 && ret < max_blocks) { | |
2ed88685 TT |
4550 | map.m_lblk = map.m_lblk + ret; |
4551 | map.m_len = max_blocks = max_blocks - ret; | |
9924a92a TT |
4552 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
4553 | credits); | |
a2df2a63 AA |
4554 | if (IS_ERR(handle)) { |
4555 | ret = PTR_ERR(handle); | |
4556 | break; | |
4557 | } | |
a4e5d88b | 4558 | ret = ext4_map_blocks(handle, inode, &map, flags); |
221879c9 | 4559 | if (ret <= 0) { |
2c98615d | 4560 | #ifdef EXT4FS_DEBUG |
b06acd38 LC |
4561 | ext4_warning(inode->i_sb, |
4562 | "inode #%lu: block %u: len %u: " | |
4563 | "ext4_ext_map_blocks returned %d", | |
4564 | inode->i_ino, map.m_lblk, | |
4565 | map.m_len, ret); | |
2c98615d | 4566 | #endif |
a2df2a63 AA |
4567 | ext4_mark_inode_dirty(handle, inode); |
4568 | ret2 = ext4_journal_stop(handle); | |
4569 | break; | |
4570 | } | |
2ed88685 | 4571 | if ((map.m_lblk + ret) >= (EXT4_BLOCK_ALIGN(offset + len, |
fd28784a AK |
4572 | blkbits) >> blkbits)) |
4573 | new_size = offset + len; | |
4574 | else | |
29ae07b7 | 4575 | new_size = ((loff_t) map.m_lblk + ret) << blkbits; |
a2df2a63 | 4576 | |
fd28784a | 4577 | ext4_falloc_update_inode(inode, mode, new_size, |
2ed88685 | 4578 | (map.m_flags & EXT4_MAP_NEW)); |
a2df2a63 | 4579 | ext4_mark_inode_dirty(handle, inode); |
f4e95b33 ZL |
4580 | if ((file->f_flags & O_SYNC) && ret >= max_blocks) |
4581 | ext4_handle_sync(handle); | |
a2df2a63 AA |
4582 | ret2 = ext4_journal_stop(handle); |
4583 | if (ret2) | |
4584 | break; | |
4585 | } | |
fd28784a AK |
4586 | if (ret == -ENOSPC && |
4587 | ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
4588 | ret = 0; | |
a2df2a63 | 4589 | goto retry; |
a2df2a63 | 4590 | } |
55bd725a | 4591 | mutex_unlock(&inode->i_mutex); |
0562e0ba JZ |
4592 | trace_ext4_fallocate_exit(inode, offset, max_blocks, |
4593 | ret > 0 ? ret2 : ret); | |
a2df2a63 AA |
4594 | return ret > 0 ? ret2 : ret; |
4595 | } | |
6873fa0d | 4596 | |
0031462b MC |
4597 | /* |
4598 | * This function convert a range of blocks to written extents | |
4599 | * The caller of this function will pass the start offset and the size. | |
4600 | * all unwritten extents within this range will be converted to | |
4601 | * written extents. | |
4602 | * | |
4603 | * This function is called from the direct IO end io call back | |
4604 | * function, to convert the fallocated extents after IO is completed. | |
109f5565 | 4605 | * Returns 0 on success. |
0031462b | 4606 | */ |
6b523df4 JK |
4607 | int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode, |
4608 | loff_t offset, ssize_t len) | |
0031462b | 4609 | { |
0031462b MC |
4610 | unsigned int max_blocks; |
4611 | int ret = 0; | |
4612 | int ret2 = 0; | |
2ed88685 | 4613 | struct ext4_map_blocks map; |
0031462b MC |
4614 | unsigned int credits, blkbits = inode->i_blkbits; |
4615 | ||
2ed88685 | 4616 | map.m_lblk = offset >> blkbits; |
0031462b MC |
4617 | /* |
4618 | * We can't just convert len to max_blocks because | |
4619 | * If blocksize = 4096 offset = 3072 and len = 2048 | |
4620 | */ | |
2ed88685 TT |
4621 | max_blocks = ((EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) - |
4622 | map.m_lblk); | |
0031462b | 4623 | /* |
6b523df4 JK |
4624 | * This is somewhat ugly but the idea is clear: When transaction is |
4625 | * reserved, everything goes into it. Otherwise we rather start several | |
4626 | * smaller transactions for conversion of each extent separately. | |
0031462b | 4627 | */ |
6b523df4 JK |
4628 | if (handle) { |
4629 | handle = ext4_journal_start_reserved(handle, | |
4630 | EXT4_HT_EXT_CONVERT); | |
4631 | if (IS_ERR(handle)) | |
4632 | return PTR_ERR(handle); | |
4633 | credits = 0; | |
4634 | } else { | |
4635 | /* | |
4636 | * credits to insert 1 extent into extent tree | |
4637 | */ | |
4638 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
4639 | } | |
0031462b | 4640 | while (ret >= 0 && ret < max_blocks) { |
2ed88685 TT |
4641 | map.m_lblk += ret; |
4642 | map.m_len = (max_blocks -= ret); | |
6b523df4 JK |
4643 | if (credits) { |
4644 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
4645 | credits); | |
4646 | if (IS_ERR(handle)) { | |
4647 | ret = PTR_ERR(handle); | |
4648 | break; | |
4649 | } | |
0031462b | 4650 | } |
2ed88685 | 4651 | ret = ext4_map_blocks(handle, inode, &map, |
c7064ef1 | 4652 | EXT4_GET_BLOCKS_IO_CONVERT_EXT); |
b06acd38 LC |
4653 | if (ret <= 0) |
4654 | ext4_warning(inode->i_sb, | |
4655 | "inode #%lu: block %u: len %u: " | |
4656 | "ext4_ext_map_blocks returned %d", | |
4657 | inode->i_ino, map.m_lblk, | |
4658 | map.m_len, ret); | |
0031462b | 4659 | ext4_mark_inode_dirty(handle, inode); |
6b523df4 JK |
4660 | if (credits) |
4661 | ret2 = ext4_journal_stop(handle); | |
4662 | if (ret <= 0 || ret2) | |
0031462b MC |
4663 | break; |
4664 | } | |
6b523df4 JK |
4665 | if (!credits) |
4666 | ret2 = ext4_journal_stop(handle); | |
0031462b MC |
4667 | return ret > 0 ? ret2 : ret; |
4668 | } | |
6d9c85eb | 4669 | |
6873fa0d | 4670 | /* |
69eb33dc ZL |
4671 | * If newes is not existing extent (newes->ec_pblk equals zero) find |
4672 | * delayed extent at start of newes and update newes accordingly and | |
91dd8c11 LC |
4673 | * return start of the next delayed extent. |
4674 | * | |
69eb33dc | 4675 | * If newes is existing extent (newes->ec_pblk is not equal zero) |
91dd8c11 | 4676 | * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed |
69eb33dc | 4677 | * extent found. Leave newes unmodified. |
6873fa0d | 4678 | */ |
91dd8c11 | 4679 | static int ext4_find_delayed_extent(struct inode *inode, |
69eb33dc | 4680 | struct extent_status *newes) |
6873fa0d | 4681 | { |
b3aff3e3 | 4682 | struct extent_status es; |
be401363 | 4683 | ext4_lblk_t block, next_del; |
6873fa0d | 4684 | |
69eb33dc | 4685 | if (newes->es_pblk == 0) { |
e30b5dca YZ |
4686 | ext4_es_find_delayed_extent_range(inode, newes->es_lblk, |
4687 | newes->es_lblk + newes->es_len - 1, &es); | |
4688 | ||
6d9c85eb | 4689 | /* |
69eb33dc | 4690 | * No extent in extent-tree contains block @newes->es_pblk, |
6d9c85eb | 4691 | * then the block may stay in 1)a hole or 2)delayed-extent. |
6d9c85eb | 4692 | */ |
06b0c886 | 4693 | if (es.es_len == 0) |
b3aff3e3 | 4694 | /* A hole found. */ |
91dd8c11 | 4695 | return 0; |
b3aff3e3 | 4696 | |
69eb33dc | 4697 | if (es.es_lblk > newes->es_lblk) { |
b3aff3e3 | 4698 | /* A hole found. */ |
69eb33dc ZL |
4699 | newes->es_len = min(es.es_lblk - newes->es_lblk, |
4700 | newes->es_len); | |
91dd8c11 | 4701 | return 0; |
6873fa0d | 4702 | } |
6d9c85eb | 4703 | |
69eb33dc | 4704 | newes->es_len = es.es_lblk + es.es_len - newes->es_lblk; |
6873fa0d ES |
4705 | } |
4706 | ||
69eb33dc | 4707 | block = newes->es_lblk + newes->es_len; |
e30b5dca | 4708 | ext4_es_find_delayed_extent_range(inode, block, EXT_MAX_BLOCKS, &es); |
be401363 ZL |
4709 | if (es.es_len == 0) |
4710 | next_del = EXT_MAX_BLOCKS; | |
4711 | else | |
4712 | next_del = es.es_lblk; | |
4713 | ||
91dd8c11 | 4714 | return next_del; |
6873fa0d | 4715 | } |
6873fa0d ES |
4716 | /* fiemap flags we can handle specified here */ |
4717 | #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) | |
4718 | ||
3a06d778 AK |
4719 | static int ext4_xattr_fiemap(struct inode *inode, |
4720 | struct fiemap_extent_info *fieinfo) | |
6873fa0d ES |
4721 | { |
4722 | __u64 physical = 0; | |
4723 | __u64 length; | |
4724 | __u32 flags = FIEMAP_EXTENT_LAST; | |
4725 | int blockbits = inode->i_sb->s_blocksize_bits; | |
4726 | int error = 0; | |
4727 | ||
4728 | /* in-inode? */ | |
19f5fb7a | 4729 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
6873fa0d ES |
4730 | struct ext4_iloc iloc; |
4731 | int offset; /* offset of xattr in inode */ | |
4732 | ||
4733 | error = ext4_get_inode_loc(inode, &iloc); | |
4734 | if (error) | |
4735 | return error; | |
a60697f4 | 4736 | physical = (__u64)iloc.bh->b_blocknr << blockbits; |
6873fa0d ES |
4737 | offset = EXT4_GOOD_OLD_INODE_SIZE + |
4738 | EXT4_I(inode)->i_extra_isize; | |
4739 | physical += offset; | |
4740 | length = EXT4_SB(inode->i_sb)->s_inode_size - offset; | |
4741 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
fd2dd9fb | 4742 | brelse(iloc.bh); |
6873fa0d | 4743 | } else { /* external block */ |
a60697f4 | 4744 | physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits; |
6873fa0d ES |
4745 | length = inode->i_sb->s_blocksize; |
4746 | } | |
4747 | ||
4748 | if (physical) | |
4749 | error = fiemap_fill_next_extent(fieinfo, 0, physical, | |
4750 | length, flags); | |
4751 | return (error < 0 ? error : 0); | |
4752 | } | |
4753 | ||
4754 | int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, | |
4755 | __u64 start, __u64 len) | |
4756 | { | |
4757 | ext4_lblk_t start_blk; | |
6873fa0d ES |
4758 | int error = 0; |
4759 | ||
94191985 TM |
4760 | if (ext4_has_inline_data(inode)) { |
4761 | int has_inline = 1; | |
4762 | ||
4763 | error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline); | |
4764 | ||
4765 | if (has_inline) | |
4766 | return error; | |
4767 | } | |
4768 | ||
6873fa0d | 4769 | /* fallback to generic here if not in extents fmt */ |
12e9b892 | 4770 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
6873fa0d ES |
4771 | return generic_block_fiemap(inode, fieinfo, start, len, |
4772 | ext4_get_block); | |
4773 | ||
4774 | if (fiemap_check_flags(fieinfo, EXT4_FIEMAP_FLAGS)) | |
4775 | return -EBADR; | |
4776 | ||
4777 | if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { | |
4778 | error = ext4_xattr_fiemap(inode, fieinfo); | |
4779 | } else { | |
aca92ff6 LM |
4780 | ext4_lblk_t len_blks; |
4781 | __u64 last_blk; | |
4782 | ||
6873fa0d | 4783 | start_blk = start >> inode->i_sb->s_blocksize_bits; |
aca92ff6 | 4784 | last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits; |
f17722f9 LC |
4785 | if (last_blk >= EXT_MAX_BLOCKS) |
4786 | last_blk = EXT_MAX_BLOCKS-1; | |
aca92ff6 | 4787 | len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1; |
6873fa0d ES |
4788 | |
4789 | /* | |
91dd8c11 LC |
4790 | * Walk the extent tree gathering extent information |
4791 | * and pushing extents back to the user. | |
6873fa0d | 4792 | */ |
91dd8c11 LC |
4793 | error = ext4_fill_fiemap_extents(inode, start_blk, |
4794 | len_blks, fieinfo); | |
6873fa0d | 4795 | } |
107a7bd3 | 4796 | ext4_es_lru_add(inode); |
6873fa0d ES |
4797 | return error; |
4798 | } |