]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Goal-directed block allocation by Stephen Tweedie | |
16 | * (sct@redhat.com), 1993, 1998 | |
17 | * Big-endian to little-endian byte-swapping/bitmaps by | |
18 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
20 | * (jj@sunsite.ms.mff.cuni.cz) | |
21 | * | |
617ba13b | 22 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
23 | */ |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
dab291af | 28 | #include <linux/jbd2.h> |
ac27a0ec DK |
29 | #include <linux/highuid.h> |
30 | #include <linux/pagemap.h> | |
31 | #include <linux/quotaops.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/buffer_head.h> | |
34 | #include <linux/writeback.h> | |
64769240 | 35 | #include <linux/pagevec.h> |
ac27a0ec | 36 | #include <linux/mpage.h> |
e83c1397 | 37 | #include <linux/namei.h> |
ac27a0ec DK |
38 | #include <linux/uio.h> |
39 | #include <linux/bio.h> | |
4c0425ff | 40 | #include <linux/workqueue.h> |
9bffad1e | 41 | |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
ac27a0ec DK |
43 | #include "xattr.h" |
44 | #include "acl.h" | |
d2a17637 | 45 | #include "ext4_extents.h" |
ac27a0ec | 46 | |
9bffad1e TT |
47 | #include <trace/events/ext4.h> |
48 | ||
a1d6cc56 AK |
49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
50 | ||
678aaf48 JK |
51 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
52 | loff_t new_size) | |
53 | { | |
7f5aa215 JK |
54 | return jbd2_journal_begin_ordered_truncate( |
55 | EXT4_SB(inode->i_sb)->s_journal, | |
56 | &EXT4_I(inode)->jinode, | |
57 | new_size); | |
678aaf48 JK |
58 | } |
59 | ||
64769240 AT |
60 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
61 | ||
ac27a0ec DK |
62 | /* |
63 | * Test whether an inode is a fast symlink. | |
64 | */ | |
617ba13b | 65 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 66 | { |
617ba13b | 67 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
68 | (inode->i_sb->s_blocksize >> 9) : 0; |
69 | ||
70 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
71 | } | |
72 | ||
ac27a0ec DK |
73 | /* |
74 | * Work out how many blocks we need to proceed with the next chunk of a | |
75 | * truncate transaction. | |
76 | */ | |
77 | static unsigned long blocks_for_truncate(struct inode *inode) | |
78 | { | |
725d26d3 | 79 | ext4_lblk_t needed; |
ac27a0ec DK |
80 | |
81 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
82 | ||
83 | /* Give ourselves just enough room to cope with inodes in which | |
84 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
85 | * which resulted in random data in an inode which looked enough | |
617ba13b | 86 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
87 | * will go a bit crazy if that happens, but at least we should |
88 | * try not to panic the whole kernel. */ | |
89 | if (needed < 2) | |
90 | needed = 2; | |
91 | ||
92 | /* But we need to bound the transaction so we don't overflow the | |
93 | * journal. */ | |
617ba13b MC |
94 | if (needed > EXT4_MAX_TRANS_DATA) |
95 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 96 | |
617ba13b | 97 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
98 | } |
99 | ||
100 | /* | |
101 | * Truncate transactions can be complex and absolutely huge. So we need to | |
102 | * be able to restart the transaction at a conventient checkpoint to make | |
103 | * sure we don't overflow the journal. | |
104 | * | |
105 | * start_transaction gets us a new handle for a truncate transaction, | |
106 | * and extend_transaction tries to extend the existing one a bit. If | |
107 | * extend fails, we need to propagate the failure up and restart the | |
108 | * transaction in the top-level truncate loop. --sct | |
109 | */ | |
110 | static handle_t *start_transaction(struct inode *inode) | |
111 | { | |
112 | handle_t *result; | |
113 | ||
617ba13b | 114 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
115 | if (!IS_ERR(result)) |
116 | return result; | |
117 | ||
617ba13b | 118 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
119 | return result; |
120 | } | |
121 | ||
122 | /* | |
123 | * Try to extend this transaction for the purposes of truncation. | |
124 | * | |
125 | * Returns 0 if we managed to create more room. If we can't create more | |
126 | * room, and the transaction must be restarted we return 1. | |
127 | */ | |
128 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
129 | { | |
0390131b FM |
130 | if (!ext4_handle_valid(handle)) |
131 | return 0; | |
132 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
ac27a0ec | 133 | return 0; |
617ba13b | 134 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
135 | return 0; |
136 | return 1; | |
137 | } | |
138 | ||
139 | /* | |
140 | * Restart the transaction associated with *handle. This does a commit, | |
141 | * so before we call here everything must be consistently dirtied against | |
142 | * this transaction. | |
143 | */ | |
fa5d1113 | 144 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 145 | int nblocks) |
ac27a0ec | 146 | { |
487caeef JK |
147 | int ret; |
148 | ||
149 | /* | |
150 | * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this | |
151 | * moment, get_block can be called only for blocks inside i_size since | |
152 | * page cache has been already dropped and writes are blocked by | |
153 | * i_mutex. So we can safely drop the i_data_sem here. | |
154 | */ | |
0390131b | 155 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 156 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef JK |
157 | up_write(&EXT4_I(inode)->i_data_sem); |
158 | ret = ext4_journal_restart(handle, blocks_for_truncate(inode)); | |
159 | down_write(&EXT4_I(inode)->i_data_sem); | |
fa5d1113 | 160 | ext4_discard_preallocations(inode); |
487caeef JK |
161 | |
162 | return ret; | |
ac27a0ec DK |
163 | } |
164 | ||
165 | /* | |
166 | * Called at the last iput() if i_nlink is zero. | |
167 | */ | |
af5bc92d | 168 | void ext4_delete_inode(struct inode *inode) |
ac27a0ec DK |
169 | { |
170 | handle_t *handle; | |
bc965ab3 | 171 | int err; |
ac27a0ec | 172 | |
678aaf48 JK |
173 | if (ext4_should_order_data(inode)) |
174 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
175 | truncate_inode_pages(&inode->i_data, 0); |
176 | ||
177 | if (is_bad_inode(inode)) | |
178 | goto no_delete; | |
179 | ||
bc965ab3 | 180 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 181 | if (IS_ERR(handle)) { |
bc965ab3 | 182 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
183 | /* |
184 | * If we're going to skip the normal cleanup, we still need to | |
185 | * make sure that the in-core orphan linked list is properly | |
186 | * cleaned up. | |
187 | */ | |
617ba13b | 188 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
189 | goto no_delete; |
190 | } | |
191 | ||
192 | if (IS_SYNC(inode)) | |
0390131b | 193 | ext4_handle_sync(handle); |
ac27a0ec | 194 | inode->i_size = 0; |
bc965ab3 TT |
195 | err = ext4_mark_inode_dirty(handle, inode); |
196 | if (err) { | |
197 | ext4_warning(inode->i_sb, __func__, | |
198 | "couldn't mark inode dirty (err %d)", err); | |
199 | goto stop_handle; | |
200 | } | |
ac27a0ec | 201 | if (inode->i_blocks) |
617ba13b | 202 | ext4_truncate(inode); |
bc965ab3 TT |
203 | |
204 | /* | |
205 | * ext4_ext_truncate() doesn't reserve any slop when it | |
206 | * restarts journal transactions; therefore there may not be | |
207 | * enough credits left in the handle to remove the inode from | |
208 | * the orphan list and set the dtime field. | |
209 | */ | |
0390131b | 210 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
211 | err = ext4_journal_extend(handle, 3); |
212 | if (err > 0) | |
213 | err = ext4_journal_restart(handle, 3); | |
214 | if (err != 0) { | |
215 | ext4_warning(inode->i_sb, __func__, | |
216 | "couldn't extend journal (err %d)", err); | |
217 | stop_handle: | |
218 | ext4_journal_stop(handle); | |
219 | goto no_delete; | |
220 | } | |
221 | } | |
222 | ||
ac27a0ec | 223 | /* |
617ba13b | 224 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 225 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 226 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 227 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 228 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
229 | * (Well, we could do this if we need to, but heck - it works) |
230 | */ | |
617ba13b MC |
231 | ext4_orphan_del(handle, inode); |
232 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
233 | |
234 | /* | |
235 | * One subtle ordering requirement: if anything has gone wrong | |
236 | * (transaction abort, IO errors, whatever), then we can still | |
237 | * do these next steps (the fs will already have been marked as | |
238 | * having errors), but we can't free the inode if the mark_dirty | |
239 | * fails. | |
240 | */ | |
617ba13b | 241 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
242 | /* If that failed, just do the required in-core inode clear. */ |
243 | clear_inode(inode); | |
244 | else | |
617ba13b MC |
245 | ext4_free_inode(handle, inode); |
246 | ext4_journal_stop(handle); | |
ac27a0ec DK |
247 | return; |
248 | no_delete: | |
249 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
250 | } | |
251 | ||
252 | typedef struct { | |
253 | __le32 *p; | |
254 | __le32 key; | |
255 | struct buffer_head *bh; | |
256 | } Indirect; | |
257 | ||
258 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
259 | { | |
260 | p->key = *(p->p = v); | |
261 | p->bh = bh; | |
262 | } | |
263 | ||
ac27a0ec | 264 | /** |
617ba13b | 265 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
266 | * @inode: inode in question (we are only interested in its superblock) |
267 | * @i_block: block number to be parsed | |
268 | * @offsets: array to store the offsets in | |
8c55e204 DK |
269 | * @boundary: set this non-zero if the referred-to block is likely to be |
270 | * followed (on disk) by an indirect block. | |
ac27a0ec | 271 | * |
617ba13b | 272 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
273 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
274 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
275 | * This function translates the block number into path in that tree - | |
276 | * return value is the path length and @offsets[n] is the offset of | |
277 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
278 | * (negative or too large) warning is printed and zero returned. | |
279 | * | |
280 | * Note: function doesn't find node addresses, so no IO is needed. All | |
281 | * we need to know is the capacity of indirect blocks (taken from the | |
282 | * inode->i_sb). | |
283 | */ | |
284 | ||
285 | /* | |
286 | * Portability note: the last comparison (check that we fit into triple | |
287 | * indirect block) is spelled differently, because otherwise on an | |
288 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
289 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
290 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
291 | * i_block would have to be negative in the very beginning, so we would not | |
292 | * get there at all. | |
293 | */ | |
294 | ||
617ba13b | 295 | static int ext4_block_to_path(struct inode *inode, |
de9a55b8 TT |
296 | ext4_lblk_t i_block, |
297 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 298 | { |
617ba13b MC |
299 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
300 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
301 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
302 | indirect_blocks = ptrs, |
303 | double_blocks = (1 << (ptrs_bits * 2)); | |
304 | int n = 0; | |
305 | int final = 0; | |
306 | ||
c333e073 | 307 | if (i_block < direct_blocks) { |
ac27a0ec DK |
308 | offsets[n++] = i_block; |
309 | final = direct_blocks; | |
af5bc92d | 310 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
617ba13b | 311 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
312 | offsets[n++] = i_block; |
313 | final = ptrs; | |
314 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 315 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
316 | offsets[n++] = i_block >> ptrs_bits; |
317 | offsets[n++] = i_block & (ptrs - 1); | |
318 | final = ptrs; | |
319 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 320 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
321 | offsets[n++] = i_block >> (ptrs_bits * 2); |
322 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
323 | offsets[n++] = i_block & (ptrs - 1); | |
324 | final = ptrs; | |
325 | } else { | |
e2b46574 | 326 | ext4_warning(inode->i_sb, "ext4_block_to_path", |
de9a55b8 TT |
327 | "block %lu > max in inode %lu", |
328 | i_block + direct_blocks + | |
329 | indirect_blocks + double_blocks, inode->i_ino); | |
ac27a0ec DK |
330 | } |
331 | if (boundary) | |
332 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
333 | return n; | |
334 | } | |
335 | ||
fe2c8191 | 336 | static int __ext4_check_blockref(const char *function, struct inode *inode, |
6fd058f7 TT |
337 | __le32 *p, unsigned int max) |
338 | { | |
f73953c0 | 339 | __le32 *bref = p; |
6fd058f7 TT |
340 | unsigned int blk; |
341 | ||
fe2c8191 | 342 | while (bref < p+max) { |
6fd058f7 | 343 | blk = le32_to_cpu(*bref++); |
de9a55b8 TT |
344 | if (blk && |
345 | unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
6fd058f7 | 346 | blk, 1))) { |
fe2c8191 | 347 | ext4_error(inode->i_sb, function, |
6fd058f7 TT |
348 | "invalid block reference %u " |
349 | "in inode #%lu", blk, inode->i_ino); | |
de9a55b8 TT |
350 | return -EIO; |
351 | } | |
352 | } | |
353 | return 0; | |
fe2c8191 TN |
354 | } |
355 | ||
356 | ||
357 | #define ext4_check_indirect_blockref(inode, bh) \ | |
de9a55b8 | 358 | __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \ |
fe2c8191 TN |
359 | EXT4_ADDR_PER_BLOCK((inode)->i_sb)) |
360 | ||
361 | #define ext4_check_inode_blockref(inode) \ | |
de9a55b8 | 362 | __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \ |
fe2c8191 TN |
363 | EXT4_NDIR_BLOCKS) |
364 | ||
ac27a0ec | 365 | /** |
617ba13b | 366 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
367 | * @inode: inode in question |
368 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
369 | * @offsets: offsets of pointers in inode/indirect blocks | |
370 | * @chain: place to store the result | |
371 | * @err: here we store the error value | |
372 | * | |
373 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
374 | * if everything went OK or the pointer to the last filled triple | |
375 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
376 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
377 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
378 | * number (it points into struct inode for i==0 and into the bh->b_data | |
379 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
380 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
381 | * numbers of the chain, addresses they were taken from (and where we can | |
382 | * verify that chain did not change) and buffer_heads hosting these | |
383 | * numbers. | |
384 | * | |
385 | * Function stops when it stumbles upon zero pointer (absent block) | |
386 | * (pointer to last triple returned, *@err == 0) | |
387 | * or when it gets an IO error reading an indirect block | |
388 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
389 | * or when it reads all @depth-1 indirect blocks successfully and finds |
390 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
391 | * |
392 | * Need to be called with | |
0e855ac8 | 393 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 394 | */ |
725d26d3 AK |
395 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
396 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
397 | Indirect chain[4], int *err) |
398 | { | |
399 | struct super_block *sb = inode->i_sb; | |
400 | Indirect *p = chain; | |
401 | struct buffer_head *bh; | |
402 | ||
403 | *err = 0; | |
404 | /* i_data is not going away, no lock needed */ | |
af5bc92d | 405 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
406 | if (!p->key) |
407 | goto no_block; | |
408 | while (--depth) { | |
fe2c8191 TN |
409 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
410 | if (unlikely(!bh)) | |
ac27a0ec | 411 | goto failure; |
de9a55b8 | 412 | |
fe2c8191 TN |
413 | if (!bh_uptodate_or_lock(bh)) { |
414 | if (bh_submit_read(bh) < 0) { | |
415 | put_bh(bh); | |
416 | goto failure; | |
417 | } | |
418 | /* validate block references */ | |
419 | if (ext4_check_indirect_blockref(inode, bh)) { | |
420 | put_bh(bh); | |
421 | goto failure; | |
422 | } | |
423 | } | |
de9a55b8 | 424 | |
af5bc92d | 425 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
ac27a0ec DK |
426 | /* Reader: end */ |
427 | if (!p->key) | |
428 | goto no_block; | |
429 | } | |
430 | return NULL; | |
431 | ||
ac27a0ec DK |
432 | failure: |
433 | *err = -EIO; | |
434 | no_block: | |
435 | return p; | |
436 | } | |
437 | ||
438 | /** | |
617ba13b | 439 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
440 | * @inode: owner |
441 | * @ind: descriptor of indirect block. | |
442 | * | |
1cc8dcf5 | 443 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
444 | * It is used when heuristic for sequential allocation fails. |
445 | * Rules are: | |
446 | * + if there is a block to the left of our position - allocate near it. | |
447 | * + if pointer will live in indirect block - allocate near that block. | |
448 | * + if pointer will live in inode - allocate in the same | |
449 | * cylinder group. | |
450 | * | |
451 | * In the latter case we colour the starting block by the callers PID to | |
452 | * prevent it from clashing with concurrent allocations for a different inode | |
453 | * in the same block group. The PID is used here so that functionally related | |
454 | * files will be close-by on-disk. | |
455 | * | |
456 | * Caller must make sure that @ind is valid and will stay that way. | |
457 | */ | |
617ba13b | 458 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 459 | { |
617ba13b | 460 | struct ext4_inode_info *ei = EXT4_I(inode); |
af5bc92d | 461 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
ac27a0ec | 462 | __le32 *p; |
617ba13b | 463 | ext4_fsblk_t bg_start; |
74d3487f | 464 | ext4_fsblk_t last_block; |
617ba13b | 465 | ext4_grpblk_t colour; |
a4912123 TT |
466 | ext4_group_t block_group; |
467 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
ac27a0ec DK |
468 | |
469 | /* Try to find previous block */ | |
470 | for (p = ind->p - 1; p >= start; p--) { | |
471 | if (*p) | |
472 | return le32_to_cpu(*p); | |
473 | } | |
474 | ||
475 | /* No such thing, so let's try location of indirect block */ | |
476 | if (ind->bh) | |
477 | return ind->bh->b_blocknr; | |
478 | ||
479 | /* | |
480 | * It is going to be referred to from the inode itself? OK, just put it | |
481 | * into the same cylinder group then. | |
482 | */ | |
a4912123 TT |
483 | block_group = ei->i_block_group; |
484 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
485 | block_group &= ~(flex_size-1); | |
486 | if (S_ISREG(inode->i_mode)) | |
487 | block_group++; | |
488 | } | |
489 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
74d3487f VC |
490 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
491 | ||
a4912123 TT |
492 | /* |
493 | * If we are doing delayed allocation, we don't need take | |
494 | * colour into account. | |
495 | */ | |
496 | if (test_opt(inode->i_sb, DELALLOC)) | |
497 | return bg_start; | |
498 | ||
74d3487f VC |
499 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
500 | colour = (current->pid % 16) * | |
617ba13b | 501 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
502 | else |
503 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
504 | return bg_start + colour; |
505 | } | |
506 | ||
507 | /** | |
1cc8dcf5 | 508 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
509 | * @inode: owner |
510 | * @block: block we want | |
ac27a0ec | 511 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 512 | * |
1cc8dcf5 | 513 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 514 | * returns it. |
fb0a387d ES |
515 | * Because this is only used for non-extent files, we limit the block nr |
516 | * to 32 bits. | |
ac27a0ec | 517 | */ |
725d26d3 | 518 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
de9a55b8 | 519 | Indirect *partial) |
ac27a0ec | 520 | { |
fb0a387d ES |
521 | ext4_fsblk_t goal; |
522 | ||
ac27a0ec | 523 | /* |
c2ea3fde | 524 | * XXX need to get goal block from mballoc's data structures |
ac27a0ec | 525 | */ |
ac27a0ec | 526 | |
fb0a387d ES |
527 | goal = ext4_find_near(inode, partial); |
528 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; | |
529 | return goal; | |
ac27a0ec DK |
530 | } |
531 | ||
532 | /** | |
617ba13b | 533 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
534 | * of direct blocks need to be allocated for the given branch. |
535 | * | |
536 | * @branch: chain of indirect blocks | |
537 | * @k: number of blocks need for indirect blocks | |
538 | * @blks: number of data blocks to be mapped. | |
539 | * @blocks_to_boundary: the offset in the indirect block | |
540 | * | |
541 | * return the total number of blocks to be allocate, including the | |
542 | * direct and indirect blocks. | |
543 | */ | |
498e5f24 | 544 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
de9a55b8 | 545 | int blocks_to_boundary) |
ac27a0ec | 546 | { |
498e5f24 | 547 | unsigned int count = 0; |
ac27a0ec DK |
548 | |
549 | /* | |
550 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
551 | * then it's clear blocks on that path have not allocated | |
552 | */ | |
553 | if (k > 0) { | |
554 | /* right now we don't handle cross boundary allocation */ | |
555 | if (blks < blocks_to_boundary + 1) | |
556 | count += blks; | |
557 | else | |
558 | count += blocks_to_boundary + 1; | |
559 | return count; | |
560 | } | |
561 | ||
562 | count++; | |
563 | while (count < blks && count <= blocks_to_boundary && | |
564 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
565 | count++; | |
566 | } | |
567 | return count; | |
568 | } | |
569 | ||
570 | /** | |
617ba13b | 571 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
572 | * @indirect_blks: the number of blocks need to allocate for indirect |
573 | * blocks | |
574 | * | |
575 | * @new_blocks: on return it will store the new block numbers for | |
576 | * the indirect blocks(if needed) and the first direct block, | |
577 | * @blks: on return it will store the total number of allocated | |
578 | * direct blocks | |
579 | */ | |
617ba13b | 580 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
581 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
582 | int indirect_blks, int blks, | |
583 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec | 584 | { |
815a1130 | 585 | struct ext4_allocation_request ar; |
ac27a0ec | 586 | int target, i; |
7061eba7 | 587 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 588 | int index = 0; |
617ba13b | 589 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
590 | int ret = 0; |
591 | ||
592 | /* | |
593 | * Here we try to allocate the requested multiple blocks at once, | |
594 | * on a best-effort basis. | |
595 | * To build a branch, we should allocate blocks for | |
596 | * the indirect blocks(if not allocated yet), and at least | |
597 | * the first direct block of this branch. That's the | |
598 | * minimum number of blocks need to allocate(required) | |
599 | */ | |
7061eba7 AK |
600 | /* first we try to allocate the indirect blocks */ |
601 | target = indirect_blks; | |
602 | while (target > 0) { | |
ac27a0ec DK |
603 | count = target; |
604 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
605 | current_block = ext4_new_meta_blocks(handle, inode, |
606 | goal, &count, err); | |
ac27a0ec DK |
607 | if (*err) |
608 | goto failed_out; | |
609 | ||
fb0a387d ES |
610 | BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS); |
611 | ||
ac27a0ec DK |
612 | target -= count; |
613 | /* allocate blocks for indirect blocks */ | |
614 | while (index < indirect_blks && count) { | |
615 | new_blocks[index++] = current_block++; | |
616 | count--; | |
617 | } | |
7061eba7 AK |
618 | if (count > 0) { |
619 | /* | |
620 | * save the new block number | |
621 | * for the first direct block | |
622 | */ | |
623 | new_blocks[index] = current_block; | |
624 | printk(KERN_INFO "%s returned more blocks than " | |
625 | "requested\n", __func__); | |
626 | WARN_ON(1); | |
ac27a0ec | 627 | break; |
7061eba7 | 628 | } |
ac27a0ec DK |
629 | } |
630 | ||
7061eba7 AK |
631 | target = blks - count ; |
632 | blk_allocated = count; | |
633 | if (!target) | |
634 | goto allocated; | |
635 | /* Now allocate data blocks */ | |
815a1130 TT |
636 | memset(&ar, 0, sizeof(ar)); |
637 | ar.inode = inode; | |
638 | ar.goal = goal; | |
639 | ar.len = target; | |
640 | ar.logical = iblock; | |
641 | if (S_ISREG(inode->i_mode)) | |
642 | /* enable in-core preallocation only for regular files */ | |
643 | ar.flags = EXT4_MB_HINT_DATA; | |
644 | ||
645 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
fb0a387d | 646 | BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS); |
815a1130 | 647 | |
7061eba7 AK |
648 | if (*err && (target == blks)) { |
649 | /* | |
650 | * if the allocation failed and we didn't allocate | |
651 | * any blocks before | |
652 | */ | |
653 | goto failed_out; | |
654 | } | |
655 | if (!*err) { | |
656 | if (target == blks) { | |
de9a55b8 TT |
657 | /* |
658 | * save the new block number | |
659 | * for the first direct block | |
660 | */ | |
7061eba7 AK |
661 | new_blocks[index] = current_block; |
662 | } | |
815a1130 | 663 | blk_allocated += ar.len; |
7061eba7 AK |
664 | } |
665 | allocated: | |
ac27a0ec | 666 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 667 | ret = blk_allocated; |
ac27a0ec DK |
668 | *err = 0; |
669 | return ret; | |
670 | failed_out: | |
af5bc92d | 671 | for (i = 0; i < index; i++) |
c9de560d | 672 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec DK |
673 | return ret; |
674 | } | |
675 | ||
676 | /** | |
617ba13b | 677 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
678 | * @inode: owner |
679 | * @indirect_blks: number of allocated indirect blocks | |
680 | * @blks: number of allocated direct blocks | |
681 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
682 | * @branch: place to store the chain in. | |
683 | * | |
684 | * This function allocates blocks, zeroes out all but the last one, | |
685 | * links them into chain and (if we are synchronous) writes them to disk. | |
686 | * In other words, it prepares a branch that can be spliced onto the | |
687 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 688 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
689 | * we had read the existing part of chain and partial points to the last |
690 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 691 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
692 | * place chain is disconnected - *branch->p is still zero (we did not |
693 | * set the last link), but branch->key contains the number that should | |
694 | * be placed into *branch->p to fill that gap. | |
695 | * | |
696 | * If allocation fails we free all blocks we've allocated (and forget | |
697 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 698 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
699 | * as described above and return 0. |
700 | */ | |
617ba13b | 701 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
702 | ext4_lblk_t iblock, int indirect_blks, |
703 | int *blks, ext4_fsblk_t goal, | |
704 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
705 | { |
706 | int blocksize = inode->i_sb->s_blocksize; | |
707 | int i, n = 0; | |
708 | int err = 0; | |
709 | struct buffer_head *bh; | |
710 | int num; | |
617ba13b MC |
711 | ext4_fsblk_t new_blocks[4]; |
712 | ext4_fsblk_t current_block; | |
ac27a0ec | 713 | |
7061eba7 | 714 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
715 | *blks, new_blocks, &err); |
716 | if (err) | |
717 | return err; | |
718 | ||
719 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
720 | /* | |
721 | * metadata blocks and data blocks are allocated. | |
722 | */ | |
723 | for (n = 1; n <= indirect_blks; n++) { | |
724 | /* | |
725 | * Get buffer_head for parent block, zero it out | |
726 | * and set the pointer to new one, then send | |
727 | * parent to disk. | |
728 | */ | |
729 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
730 | branch[n].bh = bh; | |
731 | lock_buffer(bh); | |
732 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 733 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 734 | if (err) { |
6487a9d3 CW |
735 | /* Don't brelse(bh) here; it's done in |
736 | * ext4_journal_forget() below */ | |
ac27a0ec | 737 | unlock_buffer(bh); |
ac27a0ec DK |
738 | goto failed; |
739 | } | |
740 | ||
741 | memset(bh->b_data, 0, blocksize); | |
742 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
743 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
744 | *branch[n].p = branch[n].key; | |
af5bc92d | 745 | if (n == indirect_blks) { |
ac27a0ec DK |
746 | current_block = new_blocks[n]; |
747 | /* | |
748 | * End of chain, update the last new metablock of | |
749 | * the chain to point to the new allocated | |
750 | * data blocks numbers | |
751 | */ | |
de9a55b8 | 752 | for (i = 1; i < num; i++) |
ac27a0ec DK |
753 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
754 | } | |
755 | BUFFER_TRACE(bh, "marking uptodate"); | |
756 | set_buffer_uptodate(bh); | |
757 | unlock_buffer(bh); | |
758 | ||
0390131b FM |
759 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
760 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
761 | if (err) |
762 | goto failed; | |
763 | } | |
764 | *blks = num; | |
765 | return err; | |
766 | failed: | |
767 | /* Allocation failed, free what we already allocated */ | |
768 | for (i = 1; i <= n ; i++) { | |
dab291af | 769 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
617ba13b | 770 | ext4_journal_forget(handle, branch[i].bh); |
ac27a0ec | 771 | } |
af5bc92d | 772 | for (i = 0; i < indirect_blks; i++) |
c9de560d | 773 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec | 774 | |
c9de560d | 775 | ext4_free_blocks(handle, inode, new_blocks[i], num, 0); |
ac27a0ec DK |
776 | |
777 | return err; | |
778 | } | |
779 | ||
780 | /** | |
617ba13b | 781 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
782 | * @inode: owner |
783 | * @block: (logical) number of block we are adding | |
784 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 785 | * ext4_alloc_branch) |
ac27a0ec DK |
786 | * @where: location of missing link |
787 | * @num: number of indirect blocks we are adding | |
788 | * @blks: number of direct blocks we are adding | |
789 | * | |
790 | * This function fills the missing link and does all housekeeping needed in | |
791 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
792 | * chain to new block and return 0. | |
793 | */ | |
617ba13b | 794 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
795 | ext4_lblk_t block, Indirect *where, int num, |
796 | int blks) | |
ac27a0ec DK |
797 | { |
798 | int i; | |
799 | int err = 0; | |
617ba13b | 800 | ext4_fsblk_t current_block; |
ac27a0ec | 801 | |
ac27a0ec DK |
802 | /* |
803 | * If we're splicing into a [td]indirect block (as opposed to the | |
804 | * inode) then we need to get write access to the [td]indirect block | |
805 | * before the splice. | |
806 | */ | |
807 | if (where->bh) { | |
808 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 809 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
810 | if (err) |
811 | goto err_out; | |
812 | } | |
813 | /* That's it */ | |
814 | ||
815 | *where->p = where->key; | |
816 | ||
817 | /* | |
818 | * Update the host buffer_head or inode to point to more just allocated | |
819 | * direct blocks blocks | |
820 | */ | |
821 | if (num == 0 && blks > 1) { | |
822 | current_block = le32_to_cpu(where->key) + 1; | |
823 | for (i = 1; i < blks; i++) | |
af5bc92d | 824 | *(where->p + i) = cpu_to_le32(current_block++); |
ac27a0ec DK |
825 | } |
826 | ||
ac27a0ec | 827 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
ac27a0ec DK |
828 | /* had we spliced it onto indirect block? */ |
829 | if (where->bh) { | |
830 | /* | |
831 | * If we spliced it onto an indirect block, we haven't | |
832 | * altered the inode. Note however that if it is being spliced | |
833 | * onto an indirect block at the very end of the file (the | |
834 | * file is growing) then we *will* alter the inode to reflect | |
835 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 836 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
837 | */ |
838 | jbd_debug(5, "splicing indirect only\n"); | |
0390131b FM |
839 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
840 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
ac27a0ec DK |
841 | if (err) |
842 | goto err_out; | |
843 | } else { | |
844 | /* | |
845 | * OK, we spliced it into the inode itself on a direct block. | |
ac27a0ec | 846 | */ |
41591750 | 847 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
848 | jbd_debug(5, "splicing direct\n"); |
849 | } | |
850 | return err; | |
851 | ||
852 | err_out: | |
853 | for (i = 1; i <= num; i++) { | |
dab291af | 854 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
617ba13b | 855 | ext4_journal_forget(handle, where[i].bh); |
c9de560d AT |
856 | ext4_free_blocks(handle, inode, |
857 | le32_to_cpu(where[i-1].key), 1, 0); | |
ac27a0ec | 858 | } |
c9de560d | 859 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0); |
ac27a0ec DK |
860 | |
861 | return err; | |
862 | } | |
863 | ||
864 | /* | |
b920c755 TT |
865 | * The ext4_ind_get_blocks() function handles non-extents inodes |
866 | * (i.e., using the traditional indirect/double-indirect i_blocks | |
867 | * scheme) for ext4_get_blocks(). | |
868 | * | |
ac27a0ec DK |
869 | * Allocation strategy is simple: if we have to allocate something, we will |
870 | * have to go the whole way to leaf. So let's do it before attaching anything | |
871 | * to tree, set linkage between the newborn blocks, write them if sync is | |
872 | * required, recheck the path, free and repeat if check fails, otherwise | |
873 | * set the last missing link (that will protect us from any truncate-generated | |
874 | * removals - all blocks on the path are immune now) and possibly force the | |
875 | * write on the parent block. | |
876 | * That has a nice additional property: no special recovery from the failed | |
877 | * allocations is needed - we simply release blocks and do not touch anything | |
878 | * reachable from inode. | |
879 | * | |
880 | * `handle' can be NULL if create == 0. | |
881 | * | |
ac27a0ec DK |
882 | * return > 0, # of blocks mapped or allocated. |
883 | * return = 0, if plain lookup failed. | |
884 | * return < 0, error case. | |
c278bfec | 885 | * |
b920c755 TT |
886 | * The ext4_ind_get_blocks() function should be called with |
887 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
888 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
889 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
890 | * blocks. | |
ac27a0ec | 891 | */ |
e4d996ca | 892 | static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
893 | ext4_lblk_t iblock, unsigned int maxblocks, |
894 | struct buffer_head *bh_result, | |
895 | int flags) | |
ac27a0ec DK |
896 | { |
897 | int err = -EIO; | |
725d26d3 | 898 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
899 | Indirect chain[4]; |
900 | Indirect *partial; | |
617ba13b | 901 | ext4_fsblk_t goal; |
ac27a0ec DK |
902 | int indirect_blks; |
903 | int blocks_to_boundary = 0; | |
904 | int depth; | |
ac27a0ec | 905 | int count = 0; |
617ba13b | 906 | ext4_fsblk_t first_block = 0; |
ac27a0ec | 907 | |
a86c6181 | 908 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
c2177057 | 909 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
725d26d3 | 910 | depth = ext4_block_to_path(inode, iblock, offsets, |
de9a55b8 | 911 | &blocks_to_boundary); |
ac27a0ec DK |
912 | |
913 | if (depth == 0) | |
914 | goto out; | |
915 | ||
617ba13b | 916 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
917 | |
918 | /* Simplest case - block found, no allocation needed */ | |
919 | if (!partial) { | |
920 | first_block = le32_to_cpu(chain[depth - 1].key); | |
921 | clear_buffer_new(bh_result); | |
922 | count++; | |
923 | /*map more blocks*/ | |
924 | while (count < maxblocks && count <= blocks_to_boundary) { | |
617ba13b | 925 | ext4_fsblk_t blk; |
ac27a0ec | 926 | |
ac27a0ec DK |
927 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
928 | ||
929 | if (blk == first_block + count) | |
930 | count++; | |
931 | else | |
932 | break; | |
933 | } | |
c278bfec | 934 | goto got_it; |
ac27a0ec DK |
935 | } |
936 | ||
937 | /* Next simple case - plain lookup or failed read of indirect block */ | |
c2177057 | 938 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
ac27a0ec DK |
939 | goto cleanup; |
940 | ||
ac27a0ec | 941 | /* |
c2ea3fde | 942 | * Okay, we need to do block allocation. |
ac27a0ec | 943 | */ |
fb01bfda | 944 | goal = ext4_find_goal(inode, iblock, partial); |
ac27a0ec DK |
945 | |
946 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
947 | indirect_blks = (chain + depth) - partial - 1; | |
948 | ||
949 | /* | |
950 | * Next look up the indirect map to count the totoal number of | |
951 | * direct blocks to allocate for this branch. | |
952 | */ | |
617ba13b | 953 | count = ext4_blks_to_allocate(partial, indirect_blks, |
ac27a0ec DK |
954 | maxblocks, blocks_to_boundary); |
955 | /* | |
617ba13b | 956 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 957 | */ |
7061eba7 | 958 | err = ext4_alloc_branch(handle, inode, iblock, indirect_blks, |
de9a55b8 TT |
959 | &count, goal, |
960 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
961 | |
962 | /* | |
617ba13b | 963 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
964 | * on the new chain if there is a failure, but that risks using |
965 | * up transaction credits, especially for bitmaps where the | |
966 | * credits cannot be returned. Can we handle this somehow? We | |
967 | * may need to return -EAGAIN upwards in the worst case. --sct | |
968 | */ | |
969 | if (!err) | |
617ba13b | 970 | err = ext4_splice_branch(handle, inode, iblock, |
de9a55b8 | 971 | partial, indirect_blks, count); |
2bba702d | 972 | if (err) |
ac27a0ec DK |
973 | goto cleanup; |
974 | ||
975 | set_buffer_new(bh_result); | |
976 | got_it: | |
977 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
978 | if (count > blocks_to_boundary) | |
979 | set_buffer_boundary(bh_result); | |
980 | err = count; | |
981 | /* Clean up and exit */ | |
982 | partial = chain + depth - 1; /* the whole chain */ | |
983 | cleanup: | |
984 | while (partial > chain) { | |
985 | BUFFER_TRACE(partial->bh, "call brelse"); | |
986 | brelse(partial->bh); | |
987 | partial--; | |
988 | } | |
989 | BUFFER_TRACE(bh_result, "returned"); | |
990 | out: | |
991 | return err; | |
992 | } | |
993 | ||
60e58e0f MC |
994 | qsize_t ext4_get_reserved_space(struct inode *inode) |
995 | { | |
996 | unsigned long long total; | |
997 | ||
998 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
999 | total = EXT4_I(inode)->i_reserved_data_blocks + | |
1000 | EXT4_I(inode)->i_reserved_meta_blocks; | |
1001 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1002 | ||
1003 | return total; | |
1004 | } | |
12219aea AK |
1005 | /* |
1006 | * Calculate the number of metadata blocks need to reserve | |
1007 | * to allocate @blocks for non extent file based file | |
1008 | */ | |
1009 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks) | |
1010 | { | |
1011 | int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1012 | int ind_blks, dind_blks, tind_blks; | |
1013 | ||
1014 | /* number of new indirect blocks needed */ | |
1015 | ind_blks = (blocks + icap - 1) / icap; | |
1016 | ||
1017 | dind_blks = (ind_blks + icap - 1) / icap; | |
1018 | ||
1019 | tind_blks = 1; | |
1020 | ||
1021 | return ind_blks + dind_blks + tind_blks; | |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * Calculate the number of metadata blocks need to reserve | |
1026 | * to allocate given number of blocks | |
1027 | */ | |
1028 | static int ext4_calc_metadata_amount(struct inode *inode, int blocks) | |
1029 | { | |
cd213226 MC |
1030 | if (!blocks) |
1031 | return 0; | |
1032 | ||
12219aea AK |
1033 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
1034 | return ext4_ext_calc_metadata_amount(inode, blocks); | |
1035 | ||
1036 | return ext4_indirect_calc_metadata_amount(inode, blocks); | |
1037 | } | |
1038 | ||
1039 | static void ext4_da_update_reserve_space(struct inode *inode, int used) | |
1040 | { | |
1041 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1042 | int total, mdb, mdb_free; | |
1043 | ||
1044 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1045 | /* recalculate the number of metablocks still need to be reserved */ | |
1046 | total = EXT4_I(inode)->i_reserved_data_blocks - used; | |
1047 | mdb = ext4_calc_metadata_amount(inode, total); | |
1048 | ||
1049 | /* figure out how many metablocks to release */ | |
1050 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1051 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1052 | ||
6bc6e63f AK |
1053 | if (mdb_free) { |
1054 | /* Account for allocated meta_blocks */ | |
1055 | mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks; | |
1056 | ||
1057 | /* update fs dirty blocks counter */ | |
1058 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free); | |
1059 | EXT4_I(inode)->i_allocated_meta_blocks = 0; | |
1060 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
1061 | } | |
12219aea AK |
1062 | |
1063 | /* update per-inode reservations */ | |
1064 | BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks); | |
1065 | EXT4_I(inode)->i_reserved_data_blocks -= used; | |
12219aea | 1066 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f MC |
1067 | |
1068 | /* | |
1069 | * free those over-booking quota for metadata blocks | |
1070 | */ | |
60e58e0f MC |
1071 | if (mdb_free) |
1072 | vfs_dq_release_reservation_block(inode, mdb_free); | |
d6014301 AK |
1073 | |
1074 | /* | |
1075 | * If we have done all the pending block allocations and if | |
1076 | * there aren't any writers on the inode, we can discard the | |
1077 | * inode's preallocations. | |
1078 | */ | |
1079 | if (!total && (atomic_read(&inode->i_writecount) == 0)) | |
1080 | ext4_discard_preallocations(inode); | |
12219aea AK |
1081 | } |
1082 | ||
80e42468 TT |
1083 | static int check_block_validity(struct inode *inode, const char *msg, |
1084 | sector_t logical, sector_t phys, int len) | |
6fd058f7 TT |
1085 | { |
1086 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) { | |
80e42468 | 1087 | ext4_error(inode->i_sb, msg, |
6fd058f7 TT |
1088 | "inode #%lu logical block %llu mapped to %llu " |
1089 | "(size %d)", inode->i_ino, | |
1090 | (unsigned long long) logical, | |
1091 | (unsigned long long) phys, len); | |
6fd058f7 TT |
1092 | return -EIO; |
1093 | } | |
1094 | return 0; | |
1095 | } | |
1096 | ||
55138e0b | 1097 | /* |
1f94533d TT |
1098 | * Return the number of contiguous dirty pages in a given inode |
1099 | * starting at page frame idx. | |
55138e0b TT |
1100 | */ |
1101 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
1102 | unsigned int max_pages) | |
1103 | { | |
1104 | struct address_space *mapping = inode->i_mapping; | |
1105 | pgoff_t index; | |
1106 | struct pagevec pvec; | |
1107 | pgoff_t num = 0; | |
1108 | int i, nr_pages, done = 0; | |
1109 | ||
1110 | if (max_pages == 0) | |
1111 | return 0; | |
1112 | pagevec_init(&pvec, 0); | |
1113 | while (!done) { | |
1114 | index = idx; | |
1115 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
1116 | PAGECACHE_TAG_DIRTY, | |
1117 | (pgoff_t)PAGEVEC_SIZE); | |
1118 | if (nr_pages == 0) | |
1119 | break; | |
1120 | for (i = 0; i < nr_pages; i++) { | |
1121 | struct page *page = pvec.pages[i]; | |
1122 | struct buffer_head *bh, *head; | |
1123 | ||
1124 | lock_page(page); | |
1125 | if (unlikely(page->mapping != mapping) || | |
1126 | !PageDirty(page) || | |
1127 | PageWriteback(page) || | |
1128 | page->index != idx) { | |
1129 | done = 1; | |
1130 | unlock_page(page); | |
1131 | break; | |
1132 | } | |
1f94533d TT |
1133 | if (page_has_buffers(page)) { |
1134 | bh = head = page_buffers(page); | |
1135 | do { | |
1136 | if (!buffer_delay(bh) && | |
1137 | !buffer_unwritten(bh)) | |
1138 | done = 1; | |
1139 | bh = bh->b_this_page; | |
1140 | } while (!done && (bh != head)); | |
1141 | } | |
55138e0b TT |
1142 | unlock_page(page); |
1143 | if (done) | |
1144 | break; | |
1145 | idx++; | |
1146 | num++; | |
1147 | if (num >= max_pages) | |
1148 | break; | |
1149 | } | |
1150 | pagevec_release(&pvec); | |
1151 | } | |
1152 | return num; | |
1153 | } | |
1154 | ||
f5ab0d1f | 1155 | /* |
12b7ac17 | 1156 | * The ext4_get_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 1157 | * and returns if the blocks are already mapped. |
f5ab0d1f | 1158 | * |
f5ab0d1f MC |
1159 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1160 | * and store the allocated blocks in the result buffer head and mark it | |
1161 | * mapped. | |
1162 | * | |
1163 | * If file type is extents based, it will call ext4_ext_get_blocks(), | |
e4d996ca | 1164 | * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping |
f5ab0d1f MC |
1165 | * based files |
1166 | * | |
1167 | * On success, it returns the number of blocks being mapped or allocate. | |
1168 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1169 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1170 | * the buffer head is mapped. | |
1171 | * | |
1172 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1173 | * that casem, buffer head is unmapped | |
1174 | * | |
1175 | * It returns the error in case of allocation failure. | |
1176 | */ | |
12b7ac17 TT |
1177 | int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block, |
1178 | unsigned int max_blocks, struct buffer_head *bh, | |
c2177057 | 1179 | int flags) |
0e855ac8 AK |
1180 | { |
1181 | int retval; | |
f5ab0d1f MC |
1182 | |
1183 | clear_buffer_mapped(bh); | |
2a8964d6 | 1184 | clear_buffer_unwritten(bh); |
f5ab0d1f | 1185 | |
0031462b MC |
1186 | ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u," |
1187 | "logical block %lu\n", inode->i_ino, flags, max_blocks, | |
1188 | (unsigned long)block); | |
4df3d265 | 1189 | /* |
b920c755 TT |
1190 | * Try to see if we can get the block without requesting a new |
1191 | * file system block. | |
4df3d265 AK |
1192 | */ |
1193 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1194 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { | |
1195 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
c2177057 | 1196 | bh, 0); |
0e855ac8 | 1197 | } else { |
e4d996ca | 1198 | retval = ext4_ind_get_blocks(handle, inode, block, max_blocks, |
c2177057 | 1199 | bh, 0); |
0e855ac8 | 1200 | } |
4df3d265 | 1201 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 1202 | |
6fd058f7 | 1203 | if (retval > 0 && buffer_mapped(bh)) { |
80e42468 TT |
1204 | int ret = check_block_validity(inode, "file system corruption", |
1205 | block, bh->b_blocknr, retval); | |
6fd058f7 TT |
1206 | if (ret != 0) |
1207 | return ret; | |
1208 | } | |
1209 | ||
f5ab0d1f | 1210 | /* If it is only a block(s) look up */ |
c2177057 | 1211 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
1212 | return retval; |
1213 | ||
1214 | /* | |
1215 | * Returns if the blocks have already allocated | |
1216 | * | |
1217 | * Note that if blocks have been preallocated | |
1218 | * ext4_ext_get_block() returns th create = 0 | |
1219 | * with buffer head unmapped. | |
1220 | */ | |
1221 | if (retval > 0 && buffer_mapped(bh)) | |
4df3d265 AK |
1222 | return retval; |
1223 | ||
2a8964d6 AK |
1224 | /* |
1225 | * When we call get_blocks without the create flag, the | |
1226 | * BH_Unwritten flag could have gotten set if the blocks | |
1227 | * requested were part of a uninitialized extent. We need to | |
1228 | * clear this flag now that we are committed to convert all or | |
1229 | * part of the uninitialized extent to be an initialized | |
1230 | * extent. This is because we need to avoid the combination | |
1231 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
1232 | * set on the buffer_head. | |
1233 | */ | |
1234 | clear_buffer_unwritten(bh); | |
1235 | ||
4df3d265 | 1236 | /* |
f5ab0d1f MC |
1237 | * New blocks allocate and/or writing to uninitialized extent |
1238 | * will possibly result in updating i_data, so we take | |
1239 | * the write lock of i_data_sem, and call get_blocks() | |
1240 | * with create == 1 flag. | |
4df3d265 AK |
1241 | */ |
1242 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1243 | |
1244 | /* | |
1245 | * if the caller is from delayed allocation writeout path | |
1246 | * we have already reserved fs blocks for allocation | |
1247 | * let the underlying get_block() function know to | |
1248 | * avoid double accounting | |
1249 | */ | |
c2177057 | 1250 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1251 | EXT4_I(inode)->i_delalloc_reserved_flag = 1; |
4df3d265 AK |
1252 | /* |
1253 | * We need to check for EXT4 here because migrate | |
1254 | * could have changed the inode type in between | |
1255 | */ | |
0e855ac8 AK |
1256 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
1257 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
c2177057 | 1258 | bh, flags); |
0e855ac8 | 1259 | } else { |
e4d996ca | 1260 | retval = ext4_ind_get_blocks(handle, inode, block, |
c2177057 | 1261 | max_blocks, bh, flags); |
267e4db9 AK |
1262 | |
1263 | if (retval > 0 && buffer_new(bh)) { | |
1264 | /* | |
1265 | * We allocated new blocks which will result in | |
1266 | * i_data's format changing. Force the migrate | |
1267 | * to fail by clearing migrate flags | |
1268 | */ | |
1b9c12f4 | 1269 | EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE; |
267e4db9 | 1270 | } |
0e855ac8 | 1271 | } |
d2a17637 | 1272 | |
2ac3b6e0 | 1273 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1274 | EXT4_I(inode)->i_delalloc_reserved_flag = 0; |
2ac3b6e0 TT |
1275 | |
1276 | /* | |
1277 | * Update reserved blocks/metadata blocks after successful | |
1278 | * block allocation which had been deferred till now. | |
1279 | */ | |
1280 | if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE)) | |
1281 | ext4_da_update_reserve_space(inode, retval); | |
d2a17637 | 1282 | |
4df3d265 | 1283 | up_write((&EXT4_I(inode)->i_data_sem)); |
6fd058f7 | 1284 | if (retval > 0 && buffer_mapped(bh)) { |
80e42468 TT |
1285 | int ret = check_block_validity(inode, "file system " |
1286 | "corruption after allocation", | |
1287 | block, bh->b_blocknr, retval); | |
6fd058f7 TT |
1288 | if (ret != 0) |
1289 | return ret; | |
1290 | } | |
0e855ac8 AK |
1291 | return retval; |
1292 | } | |
1293 | ||
f3bd1f3f MC |
1294 | /* Maximum number of blocks we map for direct IO at once. */ |
1295 | #define DIO_MAX_BLOCKS 4096 | |
1296 | ||
6873fa0d ES |
1297 | int ext4_get_block(struct inode *inode, sector_t iblock, |
1298 | struct buffer_head *bh_result, int create) | |
ac27a0ec | 1299 | { |
3e4fdaf8 | 1300 | handle_t *handle = ext4_journal_current_handle(); |
7fb5409d | 1301 | int ret = 0, started = 0; |
ac27a0ec | 1302 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
f3bd1f3f | 1303 | int dio_credits; |
ac27a0ec | 1304 | |
7fb5409d JK |
1305 | if (create && !handle) { |
1306 | /* Direct IO write... */ | |
1307 | if (max_blocks > DIO_MAX_BLOCKS) | |
1308 | max_blocks = DIO_MAX_BLOCKS; | |
f3bd1f3f MC |
1309 | dio_credits = ext4_chunk_trans_blocks(inode, max_blocks); |
1310 | handle = ext4_journal_start(inode, dio_credits); | |
7fb5409d | 1311 | if (IS_ERR(handle)) { |
ac27a0ec | 1312 | ret = PTR_ERR(handle); |
7fb5409d | 1313 | goto out; |
ac27a0ec | 1314 | } |
7fb5409d | 1315 | started = 1; |
ac27a0ec DK |
1316 | } |
1317 | ||
12b7ac17 | 1318 | ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result, |
c2177057 | 1319 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
7fb5409d JK |
1320 | if (ret > 0) { |
1321 | bh_result->b_size = (ret << inode->i_blkbits); | |
1322 | ret = 0; | |
ac27a0ec | 1323 | } |
7fb5409d JK |
1324 | if (started) |
1325 | ext4_journal_stop(handle); | |
1326 | out: | |
ac27a0ec DK |
1327 | return ret; |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * `handle' can be NULL if create is zero | |
1332 | */ | |
617ba13b | 1333 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1334 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec DK |
1335 | { |
1336 | struct buffer_head dummy; | |
1337 | int fatal = 0, err; | |
03f5d8bc | 1338 | int flags = 0; |
ac27a0ec DK |
1339 | |
1340 | J_ASSERT(handle != NULL || create == 0); | |
1341 | ||
1342 | dummy.b_state = 0; | |
1343 | dummy.b_blocknr = -1000; | |
1344 | buffer_trace_init(&dummy.b_history); | |
c2177057 TT |
1345 | if (create) |
1346 | flags |= EXT4_GET_BLOCKS_CREATE; | |
1347 | err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags); | |
ac27a0ec | 1348 | /* |
c2177057 TT |
1349 | * ext4_get_blocks() returns number of blocks mapped. 0 in |
1350 | * case of a HOLE. | |
ac27a0ec DK |
1351 | */ |
1352 | if (err > 0) { | |
1353 | if (err > 1) | |
1354 | WARN_ON(1); | |
1355 | err = 0; | |
1356 | } | |
1357 | *errp = err; | |
1358 | if (!err && buffer_mapped(&dummy)) { | |
1359 | struct buffer_head *bh; | |
1360 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
1361 | if (!bh) { | |
1362 | *errp = -EIO; | |
1363 | goto err; | |
1364 | } | |
1365 | if (buffer_new(&dummy)) { | |
1366 | J_ASSERT(create != 0); | |
ac39849d | 1367 | J_ASSERT(handle != NULL); |
ac27a0ec DK |
1368 | |
1369 | /* | |
1370 | * Now that we do not always journal data, we should | |
1371 | * keep in mind whether this should always journal the | |
1372 | * new buffer as metadata. For now, regular file | |
617ba13b | 1373 | * writes use ext4_get_block instead, so it's not a |
ac27a0ec DK |
1374 | * problem. |
1375 | */ | |
1376 | lock_buffer(bh); | |
1377 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 1378 | fatal = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 1379 | if (!fatal && !buffer_uptodate(bh)) { |
af5bc92d | 1380 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
ac27a0ec DK |
1381 | set_buffer_uptodate(bh); |
1382 | } | |
1383 | unlock_buffer(bh); | |
0390131b FM |
1384 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
1385 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
1386 | if (!fatal) |
1387 | fatal = err; | |
1388 | } else { | |
1389 | BUFFER_TRACE(bh, "not a new buffer"); | |
1390 | } | |
1391 | if (fatal) { | |
1392 | *errp = fatal; | |
1393 | brelse(bh); | |
1394 | bh = NULL; | |
1395 | } | |
1396 | return bh; | |
1397 | } | |
1398 | err: | |
1399 | return NULL; | |
1400 | } | |
1401 | ||
617ba13b | 1402 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1403 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 1404 | { |
af5bc92d | 1405 | struct buffer_head *bh; |
ac27a0ec | 1406 | |
617ba13b | 1407 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1408 | if (!bh) |
1409 | return bh; | |
1410 | if (buffer_uptodate(bh)) | |
1411 | return bh; | |
1412 | ll_rw_block(READ_META, 1, &bh); | |
1413 | wait_on_buffer(bh); | |
1414 | if (buffer_uptodate(bh)) | |
1415 | return bh; | |
1416 | put_bh(bh); | |
1417 | *err = -EIO; | |
1418 | return NULL; | |
1419 | } | |
1420 | ||
af5bc92d TT |
1421 | static int walk_page_buffers(handle_t *handle, |
1422 | struct buffer_head *head, | |
1423 | unsigned from, | |
1424 | unsigned to, | |
1425 | int *partial, | |
1426 | int (*fn)(handle_t *handle, | |
1427 | struct buffer_head *bh)) | |
ac27a0ec DK |
1428 | { |
1429 | struct buffer_head *bh; | |
1430 | unsigned block_start, block_end; | |
1431 | unsigned blocksize = head->b_size; | |
1432 | int err, ret = 0; | |
1433 | struct buffer_head *next; | |
1434 | ||
af5bc92d TT |
1435 | for (bh = head, block_start = 0; |
1436 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1437 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1438 | next = bh->b_this_page; |
1439 | block_end = block_start + blocksize; | |
1440 | if (block_end <= from || block_start >= to) { | |
1441 | if (partial && !buffer_uptodate(bh)) | |
1442 | *partial = 1; | |
1443 | continue; | |
1444 | } | |
1445 | err = (*fn)(handle, bh); | |
1446 | if (!ret) | |
1447 | ret = err; | |
1448 | } | |
1449 | return ret; | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * To preserve ordering, it is essential that the hole instantiation and | |
1454 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1455 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1456 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1457 | * prepare_write() is the right place. |
1458 | * | |
617ba13b MC |
1459 | * Also, this function can nest inside ext4_writepage() -> |
1460 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1461 | * has generated enough buffer credits to do the whole page. So we won't |
1462 | * block on the journal in that case, which is good, because the caller may | |
1463 | * be PF_MEMALLOC. | |
1464 | * | |
617ba13b | 1465 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1466 | * quota file writes. If we were to commit the transaction while thus |
1467 | * reentered, there can be a deadlock - we would be holding a quota | |
1468 | * lock, and the commit would never complete if another thread had a | |
1469 | * transaction open and was blocking on the quota lock - a ranking | |
1470 | * violation. | |
1471 | * | |
dab291af | 1472 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1473 | * will _not_ run commit under these circumstances because handle->h_ref |
1474 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1475 | * write. | |
1476 | */ | |
1477 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 1478 | struct buffer_head *bh) |
ac27a0ec DK |
1479 | { |
1480 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1481 | return 0; | |
617ba13b | 1482 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1483 | } |
1484 | ||
bfc1af65 | 1485 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1486 | loff_t pos, unsigned len, unsigned flags, |
1487 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1488 | { |
af5bc92d | 1489 | struct inode *inode = mapping->host; |
1938a150 | 1490 | int ret, needed_blocks; |
ac27a0ec DK |
1491 | handle_t *handle; |
1492 | int retries = 0; | |
af5bc92d | 1493 | struct page *page; |
de9a55b8 | 1494 | pgoff_t index; |
af5bc92d | 1495 | unsigned from, to; |
bfc1af65 | 1496 | |
9bffad1e | 1497 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1498 | /* |
1499 | * Reserve one block more for addition to orphan list in case | |
1500 | * we allocate blocks but write fails for some reason | |
1501 | */ | |
1502 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1503 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1504 | from = pos & (PAGE_CACHE_SIZE - 1); |
1505 | to = from + len; | |
ac27a0ec DK |
1506 | |
1507 | retry: | |
af5bc92d TT |
1508 | handle = ext4_journal_start(inode, needed_blocks); |
1509 | if (IS_ERR(handle)) { | |
1510 | ret = PTR_ERR(handle); | |
1511 | goto out; | |
7479d2b9 | 1512 | } |
ac27a0ec | 1513 | |
ebd3610b JK |
1514 | /* We cannot recurse into the filesystem as the transaction is already |
1515 | * started */ | |
1516 | flags |= AOP_FLAG_NOFS; | |
1517 | ||
54566b2c | 1518 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
1519 | if (!page) { |
1520 | ext4_journal_stop(handle); | |
1521 | ret = -ENOMEM; | |
1522 | goto out; | |
1523 | } | |
1524 | *pagep = page; | |
1525 | ||
bfc1af65 | 1526 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
ebd3610b | 1527 | ext4_get_block); |
bfc1af65 NP |
1528 | |
1529 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1530 | ret = walk_page_buffers(handle, page_buffers(page), |
1531 | from, to, NULL, do_journal_get_write_access); | |
1532 | } | |
bfc1af65 NP |
1533 | |
1534 | if (ret) { | |
af5bc92d | 1535 | unlock_page(page); |
af5bc92d | 1536 | page_cache_release(page); |
ae4d5372 AK |
1537 | /* |
1538 | * block_write_begin may have instantiated a few blocks | |
1539 | * outside i_size. Trim these off again. Don't need | |
1540 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1541 | * |
1542 | * Add inode to orphan list in case we crash before | |
1543 | * truncate finishes | |
ae4d5372 | 1544 | */ |
ffacfa7a | 1545 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1546 | ext4_orphan_add(handle, inode); |
1547 | ||
1548 | ext4_journal_stop(handle); | |
1549 | if (pos + len > inode->i_size) { | |
ffacfa7a | 1550 | ext4_truncate(inode); |
de9a55b8 | 1551 | /* |
ffacfa7a | 1552 | * If truncate failed early the inode might |
1938a150 AK |
1553 | * still be on the orphan list; we need to |
1554 | * make sure the inode is removed from the | |
1555 | * orphan list in that case. | |
1556 | */ | |
1557 | if (inode->i_nlink) | |
1558 | ext4_orphan_del(NULL, inode); | |
1559 | } | |
bfc1af65 NP |
1560 | } |
1561 | ||
617ba13b | 1562 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1563 | goto retry; |
7479d2b9 | 1564 | out: |
ac27a0ec DK |
1565 | return ret; |
1566 | } | |
1567 | ||
bfc1af65 NP |
1568 | /* For write_end() in data=journal mode */ |
1569 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1570 | { |
1571 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1572 | return 0; | |
1573 | set_buffer_uptodate(bh); | |
0390131b | 1574 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1575 | } |
1576 | ||
f8514083 | 1577 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
1578 | struct address_space *mapping, |
1579 | loff_t pos, unsigned len, unsigned copied, | |
1580 | struct page *page, void *fsdata) | |
f8514083 AK |
1581 | { |
1582 | int i_size_changed = 0; | |
1583 | struct inode *inode = mapping->host; | |
1584 | handle_t *handle = ext4_journal_current_handle(); | |
1585 | ||
1586 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1587 | ||
1588 | /* | |
1589 | * No need to use i_size_read() here, the i_size | |
1590 | * cannot change under us because we hold i_mutex. | |
1591 | * | |
1592 | * But it's important to update i_size while still holding page lock: | |
1593 | * page writeout could otherwise come in and zero beyond i_size. | |
1594 | */ | |
1595 | if (pos + copied > inode->i_size) { | |
1596 | i_size_write(inode, pos + copied); | |
1597 | i_size_changed = 1; | |
1598 | } | |
1599 | ||
1600 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
1601 | /* We need to mark inode dirty even if | |
1602 | * new_i_size is less that inode->i_size | |
1603 | * bu greater than i_disksize.(hint delalloc) | |
1604 | */ | |
1605 | ext4_update_i_disksize(inode, (pos + copied)); | |
1606 | i_size_changed = 1; | |
1607 | } | |
1608 | unlock_page(page); | |
1609 | page_cache_release(page); | |
1610 | ||
1611 | /* | |
1612 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1613 | * makes the holding time of page lock longer. Second, it forces lock | |
1614 | * ordering of page lock and transaction start for journaling | |
1615 | * filesystems. | |
1616 | */ | |
1617 | if (i_size_changed) | |
1618 | ext4_mark_inode_dirty(handle, inode); | |
1619 | ||
1620 | return copied; | |
1621 | } | |
1622 | ||
ac27a0ec DK |
1623 | /* |
1624 | * We need to pick up the new inode size which generic_commit_write gave us | |
1625 | * `file' can be NULL - eg, when called from page_symlink(). | |
1626 | * | |
617ba13b | 1627 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1628 | * buffers are managed internally. |
1629 | */ | |
bfc1af65 | 1630 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1631 | struct address_space *mapping, |
1632 | loff_t pos, unsigned len, unsigned copied, | |
1633 | struct page *page, void *fsdata) | |
ac27a0ec | 1634 | { |
617ba13b | 1635 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1636 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1637 | int ret = 0, ret2; |
1638 | ||
9bffad1e | 1639 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1640 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1641 | |
1642 | if (ret == 0) { | |
f8514083 | 1643 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1644 | page, fsdata); |
f8a87d89 | 1645 | copied = ret2; |
ffacfa7a | 1646 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1647 | /* if we have allocated more blocks and copied |
1648 | * less. We will have blocks allocated outside | |
1649 | * inode->i_size. So truncate them | |
1650 | */ | |
1651 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1652 | if (ret2 < 0) |
1653 | ret = ret2; | |
ac27a0ec | 1654 | } |
617ba13b | 1655 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1656 | if (!ret) |
1657 | ret = ret2; | |
bfc1af65 | 1658 | |
f8514083 | 1659 | if (pos + len > inode->i_size) { |
ffacfa7a | 1660 | ext4_truncate(inode); |
de9a55b8 | 1661 | /* |
ffacfa7a | 1662 | * If truncate failed early the inode might still be |
f8514083 AK |
1663 | * on the orphan list; we need to make sure the inode |
1664 | * is removed from the orphan list in that case. | |
1665 | */ | |
1666 | if (inode->i_nlink) | |
1667 | ext4_orphan_del(NULL, inode); | |
1668 | } | |
1669 | ||
1670 | ||
bfc1af65 | 1671 | return ret ? ret : copied; |
ac27a0ec DK |
1672 | } |
1673 | ||
bfc1af65 | 1674 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1675 | struct address_space *mapping, |
1676 | loff_t pos, unsigned len, unsigned copied, | |
1677 | struct page *page, void *fsdata) | |
ac27a0ec | 1678 | { |
617ba13b | 1679 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1680 | struct inode *inode = mapping->host; |
ac27a0ec | 1681 | int ret = 0, ret2; |
ac27a0ec | 1682 | |
9bffad1e | 1683 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1684 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1685 | page, fsdata); |
f8a87d89 | 1686 | copied = ret2; |
ffacfa7a | 1687 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1688 | /* if we have allocated more blocks and copied |
1689 | * less. We will have blocks allocated outside | |
1690 | * inode->i_size. So truncate them | |
1691 | */ | |
1692 | ext4_orphan_add(handle, inode); | |
1693 | ||
f8a87d89 RK |
1694 | if (ret2 < 0) |
1695 | ret = ret2; | |
ac27a0ec | 1696 | |
617ba13b | 1697 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1698 | if (!ret) |
1699 | ret = ret2; | |
bfc1af65 | 1700 | |
f8514083 | 1701 | if (pos + len > inode->i_size) { |
ffacfa7a | 1702 | ext4_truncate(inode); |
de9a55b8 | 1703 | /* |
ffacfa7a | 1704 | * If truncate failed early the inode might still be |
f8514083 AK |
1705 | * on the orphan list; we need to make sure the inode |
1706 | * is removed from the orphan list in that case. | |
1707 | */ | |
1708 | if (inode->i_nlink) | |
1709 | ext4_orphan_del(NULL, inode); | |
1710 | } | |
1711 | ||
bfc1af65 | 1712 | return ret ? ret : copied; |
ac27a0ec DK |
1713 | } |
1714 | ||
bfc1af65 | 1715 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1716 | struct address_space *mapping, |
1717 | loff_t pos, unsigned len, unsigned copied, | |
1718 | struct page *page, void *fsdata) | |
ac27a0ec | 1719 | { |
617ba13b | 1720 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1721 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1722 | int ret = 0, ret2; |
1723 | int partial = 0; | |
bfc1af65 | 1724 | unsigned from, to; |
cf17fea6 | 1725 | loff_t new_i_size; |
ac27a0ec | 1726 | |
9bffad1e | 1727 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1728 | from = pos & (PAGE_CACHE_SIZE - 1); |
1729 | to = from + len; | |
1730 | ||
1731 | if (copied < len) { | |
1732 | if (!PageUptodate(page)) | |
1733 | copied = 0; | |
1734 | page_zero_new_buffers(page, from+copied, to); | |
1735 | } | |
ac27a0ec DK |
1736 | |
1737 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1738 | to, &partial, write_end_fn); |
ac27a0ec DK |
1739 | if (!partial) |
1740 | SetPageUptodate(page); | |
cf17fea6 AK |
1741 | new_i_size = pos + copied; |
1742 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1743 | i_size_write(inode, pos+copied); |
617ba13b | 1744 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
cf17fea6 AK |
1745 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1746 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1747 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1748 | if (!ret) |
1749 | ret = ret2; | |
1750 | } | |
bfc1af65 | 1751 | |
cf108bca | 1752 | unlock_page(page); |
f8514083 | 1753 | page_cache_release(page); |
ffacfa7a | 1754 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1755 | /* if we have allocated more blocks and copied |
1756 | * less. We will have blocks allocated outside | |
1757 | * inode->i_size. So truncate them | |
1758 | */ | |
1759 | ext4_orphan_add(handle, inode); | |
1760 | ||
617ba13b | 1761 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1762 | if (!ret) |
1763 | ret = ret2; | |
f8514083 | 1764 | if (pos + len > inode->i_size) { |
ffacfa7a | 1765 | ext4_truncate(inode); |
de9a55b8 | 1766 | /* |
ffacfa7a | 1767 | * If truncate failed early the inode might still be |
f8514083 AK |
1768 | * on the orphan list; we need to make sure the inode |
1769 | * is removed from the orphan list in that case. | |
1770 | */ | |
1771 | if (inode->i_nlink) | |
1772 | ext4_orphan_del(NULL, inode); | |
1773 | } | |
bfc1af65 NP |
1774 | |
1775 | return ret ? ret : copied; | |
ac27a0ec | 1776 | } |
d2a17637 MC |
1777 | |
1778 | static int ext4_da_reserve_space(struct inode *inode, int nrblocks) | |
1779 | { | |
030ba6bc | 1780 | int retries = 0; |
60e58e0f MC |
1781 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1782 | unsigned long md_needed, mdblocks, total = 0; | |
d2a17637 MC |
1783 | |
1784 | /* | |
1785 | * recalculate the amount of metadata blocks to reserve | |
1786 | * in order to allocate nrblocks | |
1787 | * worse case is one extent per block | |
1788 | */ | |
030ba6bc | 1789 | repeat: |
d2a17637 MC |
1790 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
1791 | total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks; | |
1792 | mdblocks = ext4_calc_metadata_amount(inode, total); | |
1793 | BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks); | |
1794 | ||
1795 | md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks; | |
1796 | total = md_needed + nrblocks; | |
1797 | ||
60e58e0f MC |
1798 | /* |
1799 | * Make quota reservation here to prevent quota overflow | |
1800 | * later. Real quota accounting is done at pages writeout | |
1801 | * time. | |
1802 | */ | |
1803 | if (vfs_dq_reserve_block(inode, total)) { | |
1804 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1805 | return -EDQUOT; | |
1806 | } | |
1807 | ||
a30d542a | 1808 | if (ext4_claim_free_blocks(sbi, total)) { |
d2a17637 | 1809 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
9f0ccfd8 | 1810 | vfs_dq_release_reservation_block(inode, total); |
030ba6bc AK |
1811 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1812 | yield(); | |
1813 | goto repeat; | |
1814 | } | |
d2a17637 MC |
1815 | return -ENOSPC; |
1816 | } | |
d2a17637 MC |
1817 | EXT4_I(inode)->i_reserved_data_blocks += nrblocks; |
1818 | EXT4_I(inode)->i_reserved_meta_blocks = mdblocks; | |
1819 | ||
1820 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1821 | return 0; /* success */ | |
1822 | } | |
1823 | ||
12219aea | 1824 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1825 | { |
1826 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1827 | int total, mdb, mdb_free, release; | |
1828 | ||
cd213226 MC |
1829 | if (!to_free) |
1830 | return; /* Nothing to release, exit */ | |
1831 | ||
d2a17637 | 1832 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 MC |
1833 | |
1834 | if (!EXT4_I(inode)->i_reserved_data_blocks) { | |
1835 | /* | |
1836 | * if there is no reserved blocks, but we try to free some | |
1837 | * then the counter is messed up somewhere. | |
1838 | * but since this function is called from invalidate | |
1839 | * page, it's harmless to return without any action | |
1840 | */ | |
1841 | printk(KERN_INFO "ext4 delalloc try to release %d reserved " | |
1842 | "blocks for inode %lu, but there is no reserved " | |
1843 | "data blocks\n", to_free, inode->i_ino); | |
1844 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1845 | return; | |
1846 | } | |
1847 | ||
d2a17637 | 1848 | /* recalculate the number of metablocks still need to be reserved */ |
12219aea | 1849 | total = EXT4_I(inode)->i_reserved_data_blocks - to_free; |
d2a17637 MC |
1850 | mdb = ext4_calc_metadata_amount(inode, total); |
1851 | ||
1852 | /* figure out how many metablocks to release */ | |
1853 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1854 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1855 | ||
d2a17637 MC |
1856 | release = to_free + mdb_free; |
1857 | ||
6bc6e63f AK |
1858 | /* update fs dirty blocks counter for truncate case */ |
1859 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, release); | |
d2a17637 MC |
1860 | |
1861 | /* update per-inode reservations */ | |
12219aea AK |
1862 | BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks); |
1863 | EXT4_I(inode)->i_reserved_data_blocks -= to_free; | |
d2a17637 MC |
1864 | |
1865 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1866 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
d2a17637 | 1867 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f MC |
1868 | |
1869 | vfs_dq_release_reservation_block(inode, release); | |
d2a17637 MC |
1870 | } |
1871 | ||
1872 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1873 | unsigned long offset) |
d2a17637 MC |
1874 | { |
1875 | int to_release = 0; | |
1876 | struct buffer_head *head, *bh; | |
1877 | unsigned int curr_off = 0; | |
1878 | ||
1879 | head = page_buffers(page); | |
1880 | bh = head; | |
1881 | do { | |
1882 | unsigned int next_off = curr_off + bh->b_size; | |
1883 | ||
1884 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1885 | to_release++; | |
1886 | clear_buffer_delay(bh); | |
1887 | } | |
1888 | curr_off = next_off; | |
1889 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 1890 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 1891 | } |
ac27a0ec | 1892 | |
64769240 AT |
1893 | /* |
1894 | * Delayed allocation stuff | |
1895 | */ | |
1896 | ||
64769240 AT |
1897 | /* |
1898 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1899 | * them with writepage() call back |
64769240 AT |
1900 | * |
1901 | * @mpd->inode: inode | |
1902 | * @mpd->first_page: first page of the extent | |
1903 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1904 | * |
1905 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1906 | * to be allocated. this may be wrong if allocation failed. | |
1907 | * | |
1908 | * As pages are already locked by write_cache_pages(), we can't use it | |
1909 | */ | |
1910 | static int mpage_da_submit_io(struct mpage_da_data *mpd) | |
1911 | { | |
22208ded | 1912 | long pages_skipped; |
791b7f08 AK |
1913 | struct pagevec pvec; |
1914 | unsigned long index, end; | |
1915 | int ret = 0, err, nr_pages, i; | |
1916 | struct inode *inode = mpd->inode; | |
1917 | struct address_space *mapping = inode->i_mapping; | |
64769240 AT |
1918 | |
1919 | BUG_ON(mpd->next_page <= mpd->first_page); | |
791b7f08 AK |
1920 | /* |
1921 | * We need to start from the first_page to the next_page - 1 | |
1922 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1923 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1924 | * at the currently mapped buffer_heads. |
1925 | */ | |
64769240 AT |
1926 | index = mpd->first_page; |
1927 | end = mpd->next_page - 1; | |
1928 | ||
791b7f08 | 1929 | pagevec_init(&pvec, 0); |
64769240 | 1930 | while (index <= end) { |
791b7f08 | 1931 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1932 | if (nr_pages == 0) |
1933 | break; | |
1934 | for (i = 0; i < nr_pages; i++) { | |
1935 | struct page *page = pvec.pages[i]; | |
1936 | ||
791b7f08 AK |
1937 | index = page->index; |
1938 | if (index > end) | |
1939 | break; | |
1940 | index++; | |
1941 | ||
1942 | BUG_ON(!PageLocked(page)); | |
1943 | BUG_ON(PageWriteback(page)); | |
1944 | ||
22208ded | 1945 | pages_skipped = mpd->wbc->pages_skipped; |
a1d6cc56 | 1946 | err = mapping->a_ops->writepage(page, mpd->wbc); |
22208ded AK |
1947 | if (!err && (pages_skipped == mpd->wbc->pages_skipped)) |
1948 | /* | |
1949 | * have successfully written the page | |
1950 | * without skipping the same | |
1951 | */ | |
a1d6cc56 | 1952 | mpd->pages_written++; |
64769240 AT |
1953 | /* |
1954 | * In error case, we have to continue because | |
1955 | * remaining pages are still locked | |
1956 | * XXX: unlock and re-dirty them? | |
1957 | */ | |
1958 | if (ret == 0) | |
1959 | ret = err; | |
1960 | } | |
1961 | pagevec_release(&pvec); | |
1962 | } | |
64769240 AT |
1963 | return ret; |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers | |
1968 | * | |
1969 | * @mpd->inode - inode to walk through | |
1970 | * @exbh->b_blocknr - first block on a disk | |
1971 | * @exbh->b_size - amount of space in bytes | |
1972 | * @logical - first logical block to start assignment with | |
1973 | * | |
1974 | * the function goes through all passed space and put actual disk | |
29fa89d0 | 1975 | * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten |
64769240 AT |
1976 | */ |
1977 | static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical, | |
1978 | struct buffer_head *exbh) | |
1979 | { | |
1980 | struct inode *inode = mpd->inode; | |
1981 | struct address_space *mapping = inode->i_mapping; | |
1982 | int blocks = exbh->b_size >> inode->i_blkbits; | |
1983 | sector_t pblock = exbh->b_blocknr, cur_logical; | |
1984 | struct buffer_head *head, *bh; | |
a1d6cc56 | 1985 | pgoff_t index, end; |
64769240 AT |
1986 | struct pagevec pvec; |
1987 | int nr_pages, i; | |
1988 | ||
1989 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1990 | end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1991 | cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1992 | ||
1993 | pagevec_init(&pvec, 0); | |
1994 | ||
1995 | while (index <= end) { | |
1996 | /* XXX: optimize tail */ | |
1997 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1998 | if (nr_pages == 0) | |
1999 | break; | |
2000 | for (i = 0; i < nr_pages; i++) { | |
2001 | struct page *page = pvec.pages[i]; | |
2002 | ||
2003 | index = page->index; | |
2004 | if (index > end) | |
2005 | break; | |
2006 | index++; | |
2007 | ||
2008 | BUG_ON(!PageLocked(page)); | |
2009 | BUG_ON(PageWriteback(page)); | |
2010 | BUG_ON(!page_has_buffers(page)); | |
2011 | ||
2012 | bh = page_buffers(page); | |
2013 | head = bh; | |
2014 | ||
2015 | /* skip blocks out of the range */ | |
2016 | do { | |
2017 | if (cur_logical >= logical) | |
2018 | break; | |
2019 | cur_logical++; | |
2020 | } while ((bh = bh->b_this_page) != head); | |
2021 | ||
2022 | do { | |
2023 | if (cur_logical >= logical + blocks) | |
2024 | break; | |
29fa89d0 AK |
2025 | |
2026 | if (buffer_delay(bh) || | |
2027 | buffer_unwritten(bh)) { | |
2028 | ||
2029 | BUG_ON(bh->b_bdev != inode->i_sb->s_bdev); | |
2030 | ||
2031 | if (buffer_delay(bh)) { | |
2032 | clear_buffer_delay(bh); | |
2033 | bh->b_blocknr = pblock; | |
2034 | } else { | |
2035 | /* | |
2036 | * unwritten already should have | |
2037 | * blocknr assigned. Verify that | |
2038 | */ | |
2039 | clear_buffer_unwritten(bh); | |
2040 | BUG_ON(bh->b_blocknr != pblock); | |
2041 | } | |
2042 | ||
61628a3f | 2043 | } else if (buffer_mapped(bh)) |
64769240 | 2044 | BUG_ON(bh->b_blocknr != pblock); |
64769240 AT |
2045 | |
2046 | cur_logical++; | |
2047 | pblock++; | |
2048 | } while ((bh = bh->b_this_page) != head); | |
2049 | } | |
2050 | pagevec_release(&pvec); | |
2051 | } | |
2052 | } | |
2053 | ||
2054 | ||
2055 | /* | |
2056 | * __unmap_underlying_blocks - just a helper function to unmap | |
2057 | * set of blocks described by @bh | |
2058 | */ | |
2059 | static inline void __unmap_underlying_blocks(struct inode *inode, | |
2060 | struct buffer_head *bh) | |
2061 | { | |
2062 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2063 | int blocks, i; | |
2064 | ||
2065 | blocks = bh->b_size >> inode->i_blkbits; | |
2066 | for (i = 0; i < blocks; i++) | |
2067 | unmap_underlying_metadata(bdev, bh->b_blocknr + i); | |
2068 | } | |
2069 | ||
c4a0c46e AK |
2070 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, |
2071 | sector_t logical, long blk_cnt) | |
2072 | { | |
2073 | int nr_pages, i; | |
2074 | pgoff_t index, end; | |
2075 | struct pagevec pvec; | |
2076 | struct inode *inode = mpd->inode; | |
2077 | struct address_space *mapping = inode->i_mapping; | |
2078 | ||
2079 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2080 | end = (logical + blk_cnt - 1) >> | |
2081 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2082 | while (index <= end) { | |
2083 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
2084 | if (nr_pages == 0) | |
2085 | break; | |
2086 | for (i = 0; i < nr_pages; i++) { | |
2087 | struct page *page = pvec.pages[i]; | |
2088 | index = page->index; | |
2089 | if (index > end) | |
2090 | break; | |
2091 | index++; | |
2092 | ||
2093 | BUG_ON(!PageLocked(page)); | |
2094 | BUG_ON(PageWriteback(page)); | |
2095 | block_invalidatepage(page, 0); | |
2096 | ClearPageUptodate(page); | |
2097 | unlock_page(page); | |
2098 | } | |
2099 | } | |
2100 | return; | |
2101 | } | |
2102 | ||
df22291f AK |
2103 | static void ext4_print_free_blocks(struct inode *inode) |
2104 | { | |
2105 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1693918e TT |
2106 | printk(KERN_CRIT "Total free blocks count %lld\n", |
2107 | ext4_count_free_blocks(inode->i_sb)); | |
2108 | printk(KERN_CRIT "Free/Dirty block details\n"); | |
2109 | printk(KERN_CRIT "free_blocks=%lld\n", | |
2110 | (long long) percpu_counter_sum(&sbi->s_freeblocks_counter)); | |
2111 | printk(KERN_CRIT "dirty_blocks=%lld\n", | |
2112 | (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter)); | |
2113 | printk(KERN_CRIT "Block reservation details\n"); | |
2114 | printk(KERN_CRIT "i_reserved_data_blocks=%u\n", | |
2115 | EXT4_I(inode)->i_reserved_data_blocks); | |
2116 | printk(KERN_CRIT "i_reserved_meta_blocks=%u\n", | |
2117 | EXT4_I(inode)->i_reserved_meta_blocks); | |
df22291f AK |
2118 | return; |
2119 | } | |
2120 | ||
64769240 AT |
2121 | /* |
2122 | * mpage_da_map_blocks - go through given space | |
2123 | * | |
8dc207c0 | 2124 | * @mpd - bh describing space |
64769240 AT |
2125 | * |
2126 | * The function skips space we know is already mapped to disk blocks. | |
2127 | * | |
64769240 | 2128 | */ |
ed5bde0b | 2129 | static int mpage_da_map_blocks(struct mpage_da_data *mpd) |
64769240 | 2130 | { |
2ac3b6e0 | 2131 | int err, blks, get_blocks_flags; |
030ba6bc | 2132 | struct buffer_head new; |
2fa3cdfb TT |
2133 | sector_t next = mpd->b_blocknr; |
2134 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
2135 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
2136 | handle_t *handle = NULL; | |
64769240 AT |
2137 | |
2138 | /* | |
2139 | * We consider only non-mapped and non-allocated blocks | |
2140 | */ | |
8dc207c0 | 2141 | if ((mpd->b_state & (1 << BH_Mapped)) && |
29fa89d0 AK |
2142 | !(mpd->b_state & (1 << BH_Delay)) && |
2143 | !(mpd->b_state & (1 << BH_Unwritten))) | |
c4a0c46e | 2144 | return 0; |
2fa3cdfb TT |
2145 | |
2146 | /* | |
2147 | * If we didn't accumulate anything to write simply return | |
2148 | */ | |
2149 | if (!mpd->b_size) | |
2150 | return 0; | |
2151 | ||
2152 | handle = ext4_journal_current_handle(); | |
2153 | BUG_ON(!handle); | |
2154 | ||
79ffab34 | 2155 | /* |
2ac3b6e0 TT |
2156 | * Call ext4_get_blocks() to allocate any delayed allocation |
2157 | * blocks, or to convert an uninitialized extent to be | |
2158 | * initialized (in the case where we have written into | |
2159 | * one or more preallocated blocks). | |
2160 | * | |
2161 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
2162 | * indicate that we are on the delayed allocation path. This | |
2163 | * affects functions in many different parts of the allocation | |
2164 | * call path. This flag exists primarily because we don't | |
2165 | * want to change *many* call functions, so ext4_get_blocks() | |
2166 | * will set the magic i_delalloc_reserved_flag once the | |
2167 | * inode's allocation semaphore is taken. | |
2168 | * | |
2169 | * If the blocks in questions were delalloc blocks, set | |
2170 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
2171 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 2172 | */ |
2ac3b6e0 TT |
2173 | new.b_state = 0; |
2174 | get_blocks_flags = (EXT4_GET_BLOCKS_CREATE | | |
2175 | EXT4_GET_BLOCKS_DELALLOC_RESERVE); | |
2176 | if (mpd->b_state & (1 << BH_Delay)) | |
2177 | get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE; | |
2fa3cdfb | 2178 | blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks, |
2ac3b6e0 | 2179 | &new, get_blocks_flags); |
2fa3cdfb TT |
2180 | if (blks < 0) { |
2181 | err = blks; | |
ed5bde0b TT |
2182 | /* |
2183 | * If get block returns with error we simply | |
2184 | * return. Later writepage will redirty the page and | |
2185 | * writepages will find the dirty page again | |
c4a0c46e AK |
2186 | */ |
2187 | if (err == -EAGAIN) | |
2188 | return 0; | |
df22291f AK |
2189 | |
2190 | if (err == -ENOSPC && | |
ed5bde0b | 2191 | ext4_count_free_blocks(mpd->inode->i_sb)) { |
df22291f AK |
2192 | mpd->retval = err; |
2193 | return 0; | |
2194 | } | |
2195 | ||
c4a0c46e | 2196 | /* |
ed5bde0b TT |
2197 | * get block failure will cause us to loop in |
2198 | * writepages, because a_ops->writepage won't be able | |
2199 | * to make progress. The page will be redirtied by | |
2200 | * writepage and writepages will again try to write | |
2201 | * the same. | |
c4a0c46e | 2202 | */ |
1693918e TT |
2203 | ext4_msg(mpd->inode->i_sb, KERN_CRIT, |
2204 | "delayed block allocation failed for inode %lu at " | |
2205 | "logical offset %llu with max blocks %zd with " | |
2206 | "error %d\n", mpd->inode->i_ino, | |
2207 | (unsigned long long) next, | |
2208 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
2209 | printk(KERN_CRIT "This should not happen!! " | |
2210 | "Data will be lost\n"); | |
030ba6bc | 2211 | if (err == -ENOSPC) { |
df22291f | 2212 | ext4_print_free_blocks(mpd->inode); |
030ba6bc | 2213 | } |
2fa3cdfb | 2214 | /* invalidate all the pages */ |
c4a0c46e | 2215 | ext4_da_block_invalidatepages(mpd, next, |
8dc207c0 | 2216 | mpd->b_size >> mpd->inode->i_blkbits); |
c4a0c46e AK |
2217 | return err; |
2218 | } | |
2fa3cdfb TT |
2219 | BUG_ON(blks == 0); |
2220 | ||
2221 | new.b_size = (blks << mpd->inode->i_blkbits); | |
64769240 | 2222 | |
a1d6cc56 AK |
2223 | if (buffer_new(&new)) |
2224 | __unmap_underlying_blocks(mpd->inode, &new); | |
64769240 | 2225 | |
a1d6cc56 AK |
2226 | /* |
2227 | * If blocks are delayed marked, we need to | |
2228 | * put actual blocknr and drop delayed bit | |
2229 | */ | |
8dc207c0 TT |
2230 | if ((mpd->b_state & (1 << BH_Delay)) || |
2231 | (mpd->b_state & (1 << BH_Unwritten))) | |
a1d6cc56 | 2232 | mpage_put_bnr_to_bhs(mpd, next, &new); |
64769240 | 2233 | |
2fa3cdfb TT |
2234 | if (ext4_should_order_data(mpd->inode)) { |
2235 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
2236 | if (err) | |
2237 | return err; | |
2238 | } | |
2239 | ||
2240 | /* | |
03f5d8bc | 2241 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
2242 | */ |
2243 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
2244 | if (disksize > i_size_read(mpd->inode)) | |
2245 | disksize = i_size_read(mpd->inode); | |
2246 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
2247 | ext4_update_i_disksize(mpd->inode, disksize); | |
2248 | return ext4_mark_inode_dirty(handle, mpd->inode); | |
2249 | } | |
2250 | ||
c4a0c46e | 2251 | return 0; |
64769240 AT |
2252 | } |
2253 | ||
bf068ee2 AK |
2254 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
2255 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
2256 | |
2257 | /* | |
2258 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
2259 | * | |
2260 | * @mpd->lbh - extent of blocks | |
2261 | * @logical - logical number of the block in the file | |
2262 | * @bh - bh of the block (used to access block's state) | |
2263 | * | |
2264 | * the function is used to collect contig. blocks in same state | |
2265 | */ | |
2266 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
2267 | sector_t logical, size_t b_size, |
2268 | unsigned long b_state) | |
64769240 | 2269 | { |
64769240 | 2270 | sector_t next; |
8dc207c0 | 2271 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 2272 | |
525f4ed8 MC |
2273 | /* check if thereserved journal credits might overflow */ |
2274 | if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) { | |
2275 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { | |
2276 | /* | |
2277 | * With non-extent format we are limited by the journal | |
2278 | * credit available. Total credit needed to insert | |
2279 | * nrblocks contiguous blocks is dependent on the | |
2280 | * nrblocks. So limit nrblocks. | |
2281 | */ | |
2282 | goto flush_it; | |
2283 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
2284 | EXT4_MAX_TRANS_DATA) { | |
2285 | /* | |
2286 | * Adding the new buffer_head would make it cross the | |
2287 | * allowed limit for which we have journal credit | |
2288 | * reserved. So limit the new bh->b_size | |
2289 | */ | |
2290 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
2291 | mpd->inode->i_blkbits; | |
2292 | /* we will do mpage_da_submit_io in the next loop */ | |
2293 | } | |
2294 | } | |
64769240 AT |
2295 | /* |
2296 | * First block in the extent | |
2297 | */ | |
8dc207c0 TT |
2298 | if (mpd->b_size == 0) { |
2299 | mpd->b_blocknr = logical; | |
2300 | mpd->b_size = b_size; | |
2301 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
2302 | return; |
2303 | } | |
2304 | ||
8dc207c0 | 2305 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
2306 | /* |
2307 | * Can we merge the block to our big extent? | |
2308 | */ | |
8dc207c0 TT |
2309 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
2310 | mpd->b_size += b_size; | |
64769240 AT |
2311 | return; |
2312 | } | |
2313 | ||
525f4ed8 | 2314 | flush_it: |
64769240 AT |
2315 | /* |
2316 | * We couldn't merge the block to our extent, so we | |
2317 | * need to flush current extent and start new one | |
2318 | */ | |
c4a0c46e AK |
2319 | if (mpage_da_map_blocks(mpd) == 0) |
2320 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2321 | mpd->io_done = 1; |
2322 | return; | |
64769240 AT |
2323 | } |
2324 | ||
c364b22c | 2325 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 2326 | { |
c364b22c | 2327 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
2328 | } |
2329 | ||
64769240 AT |
2330 | /* |
2331 | * __mpage_da_writepage - finds extent of pages and blocks | |
2332 | * | |
2333 | * @page: page to consider | |
2334 | * @wbc: not used, we just follow rules | |
2335 | * @data: context | |
2336 | * | |
2337 | * The function finds extents of pages and scan them for all blocks. | |
2338 | */ | |
2339 | static int __mpage_da_writepage(struct page *page, | |
2340 | struct writeback_control *wbc, void *data) | |
2341 | { | |
2342 | struct mpage_da_data *mpd = data; | |
2343 | struct inode *inode = mpd->inode; | |
8dc207c0 | 2344 | struct buffer_head *bh, *head; |
64769240 AT |
2345 | sector_t logical; |
2346 | ||
a1d6cc56 AK |
2347 | if (mpd->io_done) { |
2348 | /* | |
2349 | * Rest of the page in the page_vec | |
2350 | * redirty then and skip then. We will | |
fd589a8f | 2351 | * try to write them again after |
a1d6cc56 AK |
2352 | * starting a new transaction |
2353 | */ | |
2354 | redirty_page_for_writepage(wbc, page); | |
2355 | unlock_page(page); | |
2356 | return MPAGE_DA_EXTENT_TAIL; | |
2357 | } | |
64769240 AT |
2358 | /* |
2359 | * Can we merge this page to current extent? | |
2360 | */ | |
2361 | if (mpd->next_page != page->index) { | |
2362 | /* | |
2363 | * Nope, we can't. So, we map non-allocated blocks | |
a1d6cc56 | 2364 | * and start IO on them using writepage() |
64769240 AT |
2365 | */ |
2366 | if (mpd->next_page != mpd->first_page) { | |
c4a0c46e AK |
2367 | if (mpage_da_map_blocks(mpd) == 0) |
2368 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2369 | /* |
2370 | * skip rest of the page in the page_vec | |
2371 | */ | |
2372 | mpd->io_done = 1; | |
2373 | redirty_page_for_writepage(wbc, page); | |
2374 | unlock_page(page); | |
2375 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2376 | } |
2377 | ||
2378 | /* | |
2379 | * Start next extent of pages ... | |
2380 | */ | |
2381 | mpd->first_page = page->index; | |
2382 | ||
2383 | /* | |
2384 | * ... and blocks | |
2385 | */ | |
8dc207c0 TT |
2386 | mpd->b_size = 0; |
2387 | mpd->b_state = 0; | |
2388 | mpd->b_blocknr = 0; | |
64769240 AT |
2389 | } |
2390 | ||
2391 | mpd->next_page = page->index + 1; | |
2392 | logical = (sector_t) page->index << | |
2393 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2394 | ||
2395 | if (!page_has_buffers(page)) { | |
8dc207c0 TT |
2396 | mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE, |
2397 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | |
a1d6cc56 AK |
2398 | if (mpd->io_done) |
2399 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2400 | } else { |
2401 | /* | |
2402 | * Page with regular buffer heads, just add all dirty ones | |
2403 | */ | |
2404 | head = page_buffers(page); | |
2405 | bh = head; | |
2406 | do { | |
2407 | BUG_ON(buffer_locked(bh)); | |
791b7f08 AK |
2408 | /* |
2409 | * We need to try to allocate | |
2410 | * unmapped blocks in the same page. | |
2411 | * Otherwise we won't make progress | |
43ce1d23 | 2412 | * with the page in ext4_writepage |
791b7f08 | 2413 | */ |
c364b22c | 2414 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
8dc207c0 TT |
2415 | mpage_add_bh_to_extent(mpd, logical, |
2416 | bh->b_size, | |
2417 | bh->b_state); | |
a1d6cc56 AK |
2418 | if (mpd->io_done) |
2419 | return MPAGE_DA_EXTENT_TAIL; | |
791b7f08 AK |
2420 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2421 | /* | |
2422 | * mapped dirty buffer. We need to update | |
2423 | * the b_state because we look at | |
2424 | * b_state in mpage_da_map_blocks. We don't | |
2425 | * update b_size because if we find an | |
2426 | * unmapped buffer_head later we need to | |
2427 | * use the b_state flag of that buffer_head. | |
2428 | */ | |
8dc207c0 TT |
2429 | if (mpd->b_size == 0) |
2430 | mpd->b_state = bh->b_state & BH_FLAGS; | |
a1d6cc56 | 2431 | } |
64769240 AT |
2432 | logical++; |
2433 | } while ((bh = bh->b_this_page) != head); | |
2434 | } | |
2435 | ||
2436 | return 0; | |
2437 | } | |
2438 | ||
64769240 | 2439 | /* |
b920c755 TT |
2440 | * This is a special get_blocks_t callback which is used by |
2441 | * ext4_da_write_begin(). It will either return mapped block or | |
2442 | * reserve space for a single block. | |
29fa89d0 AK |
2443 | * |
2444 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2445 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2446 | * | |
2447 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2448 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2449 | * initialized properly. | |
64769240 AT |
2450 | */ |
2451 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2452 | struct buffer_head *bh_result, int create) | |
2453 | { | |
2454 | int ret = 0; | |
33b9817e AK |
2455 | sector_t invalid_block = ~((sector_t) 0xffff); |
2456 | ||
2457 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
2458 | invalid_block = ~0; | |
64769240 AT |
2459 | |
2460 | BUG_ON(create == 0); | |
2461 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); | |
2462 | ||
2463 | /* | |
2464 | * first, we need to know whether the block is allocated already | |
2465 | * preallocated blocks are unmapped but should treated | |
2466 | * the same as allocated blocks. | |
2467 | */ | |
c2177057 | 2468 | ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0); |
d2a17637 MC |
2469 | if ((ret == 0) && !buffer_delay(bh_result)) { |
2470 | /* the block isn't (pre)allocated yet, let's reserve space */ | |
64769240 AT |
2471 | /* |
2472 | * XXX: __block_prepare_write() unmaps passed block, | |
2473 | * is it OK? | |
2474 | */ | |
d2a17637 MC |
2475 | ret = ext4_da_reserve_space(inode, 1); |
2476 | if (ret) | |
2477 | /* not enough space to reserve */ | |
2478 | return ret; | |
2479 | ||
33b9817e | 2480 | map_bh(bh_result, inode->i_sb, invalid_block); |
64769240 AT |
2481 | set_buffer_new(bh_result); |
2482 | set_buffer_delay(bh_result); | |
2483 | } else if (ret > 0) { | |
2484 | bh_result->b_size = (ret << inode->i_blkbits); | |
29fa89d0 AK |
2485 | if (buffer_unwritten(bh_result)) { |
2486 | /* A delayed write to unwritten bh should | |
2487 | * be marked new and mapped. Mapped ensures | |
2488 | * that we don't do get_block multiple times | |
2489 | * when we write to the same offset and new | |
2490 | * ensures that we do proper zero out for | |
2491 | * partial write. | |
2492 | */ | |
9c1ee184 | 2493 | set_buffer_new(bh_result); |
29fa89d0 AK |
2494 | set_buffer_mapped(bh_result); |
2495 | } | |
64769240 AT |
2496 | ret = 0; |
2497 | } | |
2498 | ||
2499 | return ret; | |
2500 | } | |
61628a3f | 2501 | |
b920c755 TT |
2502 | /* |
2503 | * This function is used as a standard get_block_t calback function | |
2504 | * when there is no desire to allocate any blocks. It is used as a | |
2505 | * callback function for block_prepare_write(), nobh_writepage(), and | |
2506 | * block_write_full_page(). These functions should only try to map a | |
2507 | * single block at a time. | |
2508 | * | |
2509 | * Since this function doesn't do block allocations even if the caller | |
2510 | * requests it by passing in create=1, it is critically important that | |
2511 | * any caller checks to make sure that any buffer heads are returned | |
2512 | * by this function are either all already mapped or marked for | |
2513 | * delayed allocation before calling nobh_writepage() or | |
2514 | * block_write_full_page(). Otherwise, b_blocknr could be left | |
2515 | * unitialized, and the page write functions will be taken by | |
2516 | * surprise. | |
2517 | */ | |
2518 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
2519 | struct buffer_head *bh_result, int create) |
2520 | { | |
2521 | int ret = 0; | |
2522 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
2523 | ||
a2dc52b5 TT |
2524 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2525 | ||
f0e6c985 AK |
2526 | /* |
2527 | * we don't want to do block allocation in writepage | |
2528 | * so call get_block_wrap with create = 0 | |
2529 | */ | |
c2177057 | 2530 | ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0); |
f0e6c985 AK |
2531 | if (ret > 0) { |
2532 | bh_result->b_size = (ret << inode->i_blkbits); | |
2533 | ret = 0; | |
2534 | } | |
2535 | return ret; | |
61628a3f MC |
2536 | } |
2537 | ||
62e086be AK |
2538 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2539 | { | |
2540 | get_bh(bh); | |
2541 | return 0; | |
2542 | } | |
2543 | ||
2544 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2545 | { | |
2546 | put_bh(bh); | |
2547 | return 0; | |
2548 | } | |
2549 | ||
2550 | static int __ext4_journalled_writepage(struct page *page, | |
2551 | struct writeback_control *wbc, | |
2552 | unsigned int len) | |
2553 | { | |
2554 | struct address_space *mapping = page->mapping; | |
2555 | struct inode *inode = mapping->host; | |
2556 | struct buffer_head *page_bufs; | |
2557 | handle_t *handle = NULL; | |
2558 | int ret = 0; | |
2559 | int err; | |
2560 | ||
2561 | page_bufs = page_buffers(page); | |
2562 | BUG_ON(!page_bufs); | |
2563 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
2564 | /* As soon as we unlock the page, it can go away, but we have | |
2565 | * references to buffers so we are safe */ | |
2566 | unlock_page(page); | |
2567 | ||
2568 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
2569 | if (IS_ERR(handle)) { | |
2570 | ret = PTR_ERR(handle); | |
2571 | goto out; | |
2572 | } | |
2573 | ||
2574 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2575 | do_journal_get_write_access); | |
2576 | ||
2577 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2578 | write_end_fn); | |
2579 | if (ret == 0) | |
2580 | ret = err; | |
2581 | err = ext4_journal_stop(handle); | |
2582 | if (!ret) | |
2583 | ret = err; | |
2584 | ||
2585 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
2586 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; | |
2587 | out: | |
2588 | return ret; | |
2589 | } | |
2590 | ||
61628a3f | 2591 | /* |
43ce1d23 AK |
2592 | * Note that we don't need to start a transaction unless we're journaling data |
2593 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2594 | * need to file the inode to the transaction's list in ordered mode because if | |
2595 | * we are writing back data added by write(), the inode is already there and if | |
2596 | * we are writing back data modified via mmap(), noone guarantees in which | |
2597 | * transaction the data will hit the disk. In case we are journaling data, we | |
2598 | * cannot start transaction directly because transaction start ranks above page | |
2599 | * lock so we have to do some magic. | |
2600 | * | |
b920c755 TT |
2601 | * This function can get called via... |
2602 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2603 | * - journal_submit_inode_data_buffers (no journal handle) | |
2604 | * - shrink_page_list via pdflush (no journal handle) | |
2605 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
2606 | * |
2607 | * We don't do any block allocation in this function. If we have page with | |
2608 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2609 | * is important for mmaped based write. So if we do with blocksize 1K | |
2610 | * truncate(f, 1024); | |
2611 | * a = mmap(f, 0, 4096); | |
2612 | * a[0] = 'a'; | |
2613 | * truncate(f, 4096); | |
2614 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
2615 | * but other bufer_heads would be unmapped but dirty(dirty done via the | |
2616 | * do_wp_page). So writepage should write the first block. If we modify | |
2617 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2618 | * page_mkwrite callback will do the block allocation and mark the | |
2619 | * buffer_heads mapped. | |
2620 | * | |
2621 | * We redirty the page if we have any buffer_heads that is either delay or | |
2622 | * unwritten in the page. | |
2623 | * | |
2624 | * We can get recursively called as show below. | |
2625 | * | |
2626 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2627 | * ext4_writepage() | |
2628 | * | |
2629 | * But since we don't do any block allocation we should not deadlock. | |
2630 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2631 | */ |
43ce1d23 | 2632 | static int ext4_writepage(struct page *page, |
62e086be | 2633 | struct writeback_control *wbc) |
64769240 | 2634 | { |
64769240 | 2635 | int ret = 0; |
61628a3f | 2636 | loff_t size; |
498e5f24 | 2637 | unsigned int len; |
61628a3f MC |
2638 | struct buffer_head *page_bufs; |
2639 | struct inode *inode = page->mapping->host; | |
2640 | ||
43ce1d23 | 2641 | trace_ext4_writepage(inode, page); |
f0e6c985 AK |
2642 | size = i_size_read(inode); |
2643 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2644 | len = size & ~PAGE_CACHE_MASK; | |
2645 | else | |
2646 | len = PAGE_CACHE_SIZE; | |
64769240 | 2647 | |
f0e6c985 | 2648 | if (page_has_buffers(page)) { |
61628a3f | 2649 | page_bufs = page_buffers(page); |
f0e6c985 | 2650 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
c364b22c | 2651 | ext4_bh_delay_or_unwritten)) { |
61628a3f | 2652 | /* |
f0e6c985 AK |
2653 | * We don't want to do block allocation |
2654 | * So redirty the page and return | |
cd1aac32 AK |
2655 | * We may reach here when we do a journal commit |
2656 | * via journal_submit_inode_data_buffers. | |
2657 | * If we don't have mapping block we just ignore | |
f0e6c985 AK |
2658 | * them. We can also reach here via shrink_page_list |
2659 | */ | |
2660 | redirty_page_for_writepage(wbc, page); | |
2661 | unlock_page(page); | |
2662 | return 0; | |
2663 | } | |
2664 | } else { | |
2665 | /* | |
2666 | * The test for page_has_buffers() is subtle: | |
2667 | * We know the page is dirty but it lost buffers. That means | |
2668 | * that at some moment in time after write_begin()/write_end() | |
2669 | * has been called all buffers have been clean and thus they | |
2670 | * must have been written at least once. So they are all | |
2671 | * mapped and we can happily proceed with mapping them | |
2672 | * and writing the page. | |
2673 | * | |
2674 | * Try to initialize the buffer_heads and check whether | |
2675 | * all are mapped and non delay. We don't want to | |
2676 | * do block allocation here. | |
2677 | */ | |
b767e78a | 2678 | ret = block_prepare_write(page, 0, len, |
b920c755 | 2679 | noalloc_get_block_write); |
f0e6c985 AK |
2680 | if (!ret) { |
2681 | page_bufs = page_buffers(page); | |
2682 | /* check whether all are mapped and non delay */ | |
2683 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
c364b22c | 2684 | ext4_bh_delay_or_unwritten)) { |
f0e6c985 AK |
2685 | redirty_page_for_writepage(wbc, page); |
2686 | unlock_page(page); | |
2687 | return 0; | |
2688 | } | |
2689 | } else { | |
2690 | /* | |
2691 | * We can't do block allocation here | |
2692 | * so just redity the page and unlock | |
2693 | * and return | |
61628a3f | 2694 | */ |
61628a3f MC |
2695 | redirty_page_for_writepage(wbc, page); |
2696 | unlock_page(page); | |
2697 | return 0; | |
2698 | } | |
ed9b3e33 | 2699 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 2700 | block_commit_write(page, 0, len); |
64769240 AT |
2701 | } |
2702 | ||
43ce1d23 AK |
2703 | if (PageChecked(page) && ext4_should_journal_data(inode)) { |
2704 | /* | |
2705 | * It's mmapped pagecache. Add buffers and journal it. There | |
2706 | * doesn't seem much point in redirtying the page here. | |
2707 | */ | |
2708 | ClearPageChecked(page); | |
2709 | return __ext4_journalled_writepage(page, wbc, len); | |
2710 | } | |
2711 | ||
64769240 | 2712 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) |
b920c755 | 2713 | ret = nobh_writepage(page, noalloc_get_block_write, wbc); |
64769240 | 2714 | else |
b920c755 TT |
2715 | ret = block_write_full_page(page, noalloc_get_block_write, |
2716 | wbc); | |
64769240 | 2717 | |
64769240 AT |
2718 | return ret; |
2719 | } | |
2720 | ||
61628a3f | 2721 | /* |
525f4ed8 MC |
2722 | * This is called via ext4_da_writepages() to |
2723 | * calulate the total number of credits to reserve to fit | |
2724 | * a single extent allocation into a single transaction, | |
2725 | * ext4_da_writpeages() will loop calling this before | |
2726 | * the block allocation. | |
61628a3f | 2727 | */ |
525f4ed8 MC |
2728 | |
2729 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2730 | { | |
2731 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2732 | ||
2733 | /* | |
2734 | * With non-extent format the journal credit needed to | |
2735 | * insert nrblocks contiguous block is dependent on | |
2736 | * number of contiguous block. So we will limit | |
2737 | * number of contiguous block to a sane value | |
2738 | */ | |
30c6e07a | 2739 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) && |
525f4ed8 MC |
2740 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2741 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2742 | ||
2743 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2744 | } | |
61628a3f | 2745 | |
64769240 | 2746 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2747 | struct writeback_control *wbc) |
64769240 | 2748 | { |
22208ded AK |
2749 | pgoff_t index; |
2750 | int range_whole = 0; | |
61628a3f | 2751 | handle_t *handle = NULL; |
df22291f | 2752 | struct mpage_da_data mpd; |
5e745b04 | 2753 | struct inode *inode = mapping->host; |
22208ded | 2754 | int no_nrwrite_index_update; |
498e5f24 TT |
2755 | int pages_written = 0; |
2756 | long pages_skipped; | |
55138e0b | 2757 | unsigned int max_pages; |
2acf2c26 | 2758 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2759 | int needed_blocks, ret = 0; |
2760 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2761 | loff_t range_start = wbc->range_start; |
5e745b04 | 2762 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
61628a3f | 2763 | |
9bffad1e | 2764 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2765 | |
61628a3f MC |
2766 | /* |
2767 | * No pages to write? This is mainly a kludge to avoid starting | |
2768 | * a transaction for special inodes like journal inode on last iput() | |
2769 | * because that could violate lock ordering on umount | |
2770 | */ | |
a1d6cc56 | 2771 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2772 | return 0; |
2a21e37e TT |
2773 | |
2774 | /* | |
2775 | * If the filesystem has aborted, it is read-only, so return | |
2776 | * right away instead of dumping stack traces later on that | |
2777 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2778 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2779 | * the latter could be true if the filesystem is mounted |
2780 | * read-only, and in that case, ext4_da_writepages should | |
2781 | * *never* be called, so if that ever happens, we would want | |
2782 | * the stack trace. | |
2783 | */ | |
4ab2f15b | 2784 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2785 | return -EROFS; |
2786 | ||
22208ded AK |
2787 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2788 | range_whole = 1; | |
61628a3f | 2789 | |
2acf2c26 AK |
2790 | range_cyclic = wbc->range_cyclic; |
2791 | if (wbc->range_cyclic) { | |
22208ded | 2792 | index = mapping->writeback_index; |
2acf2c26 AK |
2793 | if (index) |
2794 | cycled = 0; | |
2795 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2796 | wbc->range_end = LLONG_MAX; | |
2797 | wbc->range_cyclic = 0; | |
2798 | } else | |
22208ded | 2799 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
a1d6cc56 | 2800 | |
55138e0b TT |
2801 | /* |
2802 | * This works around two forms of stupidity. The first is in | |
2803 | * the writeback code, which caps the maximum number of pages | |
2804 | * written to be 1024 pages. This is wrong on multiple | |
2805 | * levels; different architectues have a different page size, | |
2806 | * which changes the maximum amount of data which gets | |
2807 | * written. Secondly, 4 megabytes is way too small. XFS | |
2808 | * forces this value to be 16 megabytes by multiplying | |
2809 | * nr_to_write parameter by four, and then relies on its | |
2810 | * allocator to allocate larger extents to make them | |
2811 | * contiguous. Unfortunately this brings us to the second | |
2812 | * stupidity, which is that ext4's mballoc code only allocates | |
2813 | * at most 2048 blocks. So we force contiguous writes up to | |
2814 | * the number of dirty blocks in the inode, or | |
2815 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2816 | */ | |
2817 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
2818 | if (!range_cyclic && range_whole) | |
2819 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2820 | else | |
2821 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, | |
2822 | max_pages); | |
2823 | if (desired_nr_to_write > max_pages) | |
2824 | desired_nr_to_write = max_pages; | |
2825 | ||
2826 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2827 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2828 | wbc->nr_to_write = desired_nr_to_write; | |
2829 | } | |
2830 | ||
df22291f AK |
2831 | mpd.wbc = wbc; |
2832 | mpd.inode = mapping->host; | |
2833 | ||
22208ded AK |
2834 | /* |
2835 | * we don't want write_cache_pages to update | |
2836 | * nr_to_write and writeback_index | |
2837 | */ | |
2838 | no_nrwrite_index_update = wbc->no_nrwrite_index_update; | |
2839 | wbc->no_nrwrite_index_update = 1; | |
2840 | pages_skipped = wbc->pages_skipped; | |
2841 | ||
2acf2c26 | 2842 | retry: |
22208ded | 2843 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2844 | |
2845 | /* | |
2846 | * we insert one extent at a time. So we need | |
2847 | * credit needed for single extent allocation. | |
2848 | * journalled mode is currently not supported | |
2849 | * by delalloc | |
2850 | */ | |
2851 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2852 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2853 | |
61628a3f MC |
2854 | /* start a new transaction*/ |
2855 | handle = ext4_journal_start(inode, needed_blocks); | |
2856 | if (IS_ERR(handle)) { | |
2857 | ret = PTR_ERR(handle); | |
1693918e | 2858 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
a1d6cc56 AK |
2859 | "%ld pages, ino %lu; err %d\n", __func__, |
2860 | wbc->nr_to_write, inode->i_ino, ret); | |
61628a3f MC |
2861 | goto out_writepages; |
2862 | } | |
f63e6005 TT |
2863 | |
2864 | /* | |
2865 | * Now call __mpage_da_writepage to find the next | |
2866 | * contiguous region of logical blocks that need | |
2867 | * blocks to be allocated by ext4. We don't actually | |
2868 | * submit the blocks for I/O here, even though | |
2869 | * write_cache_pages thinks it will, and will set the | |
2870 | * pages as clean for write before calling | |
2871 | * __mpage_da_writepage(). | |
2872 | */ | |
2873 | mpd.b_size = 0; | |
2874 | mpd.b_state = 0; | |
2875 | mpd.b_blocknr = 0; | |
2876 | mpd.first_page = 0; | |
2877 | mpd.next_page = 0; | |
2878 | mpd.io_done = 0; | |
2879 | mpd.pages_written = 0; | |
2880 | mpd.retval = 0; | |
2881 | ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, | |
2882 | &mpd); | |
2883 | /* | |
2884 | * If we have a contigous extent of pages and we | |
2885 | * haven't done the I/O yet, map the blocks and submit | |
2886 | * them for I/O. | |
2887 | */ | |
2888 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
2889 | if (mpage_da_map_blocks(&mpd) == 0) | |
2890 | mpage_da_submit_io(&mpd); | |
2891 | mpd.io_done = 1; | |
2892 | ret = MPAGE_DA_EXTENT_TAIL; | |
2893 | } | |
b3a3ca8c | 2894 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2895 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2896 | |
61628a3f | 2897 | ext4_journal_stop(handle); |
df22291f | 2898 | |
8f64b32e | 2899 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2900 | /* commit the transaction which would |
2901 | * free blocks released in the transaction | |
2902 | * and try again | |
2903 | */ | |
df22291f | 2904 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2905 | wbc->pages_skipped = pages_skipped; |
2906 | ret = 0; | |
2907 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
2908 | /* |
2909 | * got one extent now try with | |
2910 | * rest of the pages | |
2911 | */ | |
22208ded AK |
2912 | pages_written += mpd.pages_written; |
2913 | wbc->pages_skipped = pages_skipped; | |
a1d6cc56 | 2914 | ret = 0; |
2acf2c26 | 2915 | io_done = 1; |
22208ded | 2916 | } else if (wbc->nr_to_write) |
61628a3f MC |
2917 | /* |
2918 | * There is no more writeout needed | |
2919 | * or we requested for a noblocking writeout | |
2920 | * and we found the device congested | |
2921 | */ | |
61628a3f | 2922 | break; |
a1d6cc56 | 2923 | } |
2acf2c26 AK |
2924 | if (!io_done && !cycled) { |
2925 | cycled = 1; | |
2926 | index = 0; | |
2927 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2928 | wbc->range_end = mapping->writeback_index - 1; | |
2929 | goto retry; | |
2930 | } | |
22208ded | 2931 | if (pages_skipped != wbc->pages_skipped) |
1693918e TT |
2932 | ext4_msg(inode->i_sb, KERN_CRIT, |
2933 | "This should not happen leaving %s " | |
2934 | "with nr_to_write = %ld ret = %d\n", | |
2935 | __func__, wbc->nr_to_write, ret); | |
22208ded AK |
2936 | |
2937 | /* Update index */ | |
2938 | index += pages_written; | |
2acf2c26 | 2939 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2940 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2941 | /* | |
2942 | * set the writeback_index so that range_cyclic | |
2943 | * mode will write it back later | |
2944 | */ | |
2945 | mapping->writeback_index = index; | |
a1d6cc56 | 2946 | |
61628a3f | 2947 | out_writepages: |
22208ded AK |
2948 | if (!no_nrwrite_index_update) |
2949 | wbc->no_nrwrite_index_update = 0; | |
55138e0b TT |
2950 | if (wbc->nr_to_write > nr_to_writebump) |
2951 | wbc->nr_to_write -= nr_to_writebump; | |
de89de6e | 2952 | wbc->range_start = range_start; |
9bffad1e | 2953 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2954 | return ret; |
64769240 AT |
2955 | } |
2956 | ||
79f0be8d AK |
2957 | #define FALL_BACK_TO_NONDELALLOC 1 |
2958 | static int ext4_nonda_switch(struct super_block *sb) | |
2959 | { | |
2960 | s64 free_blocks, dirty_blocks; | |
2961 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2962 | ||
2963 | /* | |
2964 | * switch to non delalloc mode if we are running low | |
2965 | * on free block. The free block accounting via percpu | |
179f7ebf | 2966 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2967 | * accumulated on each CPU without updating global counters |
2968 | * Delalloc need an accurate free block accounting. So switch | |
2969 | * to non delalloc when we are near to error range. | |
2970 | */ | |
2971 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
2972 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter); | |
2973 | if (2 * free_blocks < 3 * dirty_blocks || | |
2974 | free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) { | |
2975 | /* | |
2976 | * free block count is less that 150% of dirty blocks | |
2977 | * or free blocks is less that watermark | |
2978 | */ | |
2979 | return 1; | |
2980 | } | |
2981 | return 0; | |
2982 | } | |
2983 | ||
64769240 | 2984 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2985 | loff_t pos, unsigned len, unsigned flags, |
2986 | struct page **pagep, void **fsdata) | |
64769240 | 2987 | { |
d2a17637 | 2988 | int ret, retries = 0; |
64769240 AT |
2989 | struct page *page; |
2990 | pgoff_t index; | |
2991 | unsigned from, to; | |
2992 | struct inode *inode = mapping->host; | |
2993 | handle_t *handle; | |
2994 | ||
2995 | index = pos >> PAGE_CACHE_SHIFT; | |
2996 | from = pos & (PAGE_CACHE_SIZE - 1); | |
2997 | to = from + len; | |
79f0be8d AK |
2998 | |
2999 | if (ext4_nonda_switch(inode->i_sb)) { | |
3000 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
3001 | return ext4_write_begin(file, mapping, pos, | |
3002 | len, flags, pagep, fsdata); | |
3003 | } | |
3004 | *fsdata = (void *)0; | |
9bffad1e | 3005 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 3006 | retry: |
64769240 AT |
3007 | /* |
3008 | * With delayed allocation, we don't log the i_disksize update | |
3009 | * if there is delayed block allocation. But we still need | |
3010 | * to journalling the i_disksize update if writes to the end | |
3011 | * of file which has an already mapped buffer. | |
3012 | */ | |
3013 | handle = ext4_journal_start(inode, 1); | |
3014 | if (IS_ERR(handle)) { | |
3015 | ret = PTR_ERR(handle); | |
3016 | goto out; | |
3017 | } | |
ebd3610b JK |
3018 | /* We cannot recurse into the filesystem as the transaction is already |
3019 | * started */ | |
3020 | flags |= AOP_FLAG_NOFS; | |
64769240 | 3021 | |
54566b2c | 3022 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
3023 | if (!page) { |
3024 | ext4_journal_stop(handle); | |
3025 | ret = -ENOMEM; | |
3026 | goto out; | |
3027 | } | |
64769240 AT |
3028 | *pagep = page; |
3029 | ||
3030 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
b920c755 | 3031 | ext4_da_get_block_prep); |
64769240 AT |
3032 | if (ret < 0) { |
3033 | unlock_page(page); | |
3034 | ext4_journal_stop(handle); | |
3035 | page_cache_release(page); | |
ae4d5372 AK |
3036 | /* |
3037 | * block_write_begin may have instantiated a few blocks | |
3038 | * outside i_size. Trim these off again. Don't need | |
3039 | * i_size_read because we hold i_mutex. | |
3040 | */ | |
3041 | if (pos + len > inode->i_size) | |
ffacfa7a | 3042 | ext4_truncate(inode); |
64769240 AT |
3043 | } |
3044 | ||
d2a17637 MC |
3045 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3046 | goto retry; | |
64769240 AT |
3047 | out: |
3048 | return ret; | |
3049 | } | |
3050 | ||
632eaeab MC |
3051 | /* |
3052 | * Check if we should update i_disksize | |
3053 | * when write to the end of file but not require block allocation | |
3054 | */ | |
3055 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3056 | unsigned long offset) |
632eaeab MC |
3057 | { |
3058 | struct buffer_head *bh; | |
3059 | struct inode *inode = page->mapping->host; | |
3060 | unsigned int idx; | |
3061 | int i; | |
3062 | ||
3063 | bh = page_buffers(page); | |
3064 | idx = offset >> inode->i_blkbits; | |
3065 | ||
af5bc92d | 3066 | for (i = 0; i < idx; i++) |
632eaeab MC |
3067 | bh = bh->b_this_page; |
3068 | ||
29fa89d0 | 3069 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3070 | return 0; |
3071 | return 1; | |
3072 | } | |
3073 | ||
64769240 | 3074 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3075 | struct address_space *mapping, |
3076 | loff_t pos, unsigned len, unsigned copied, | |
3077 | struct page *page, void *fsdata) | |
64769240 AT |
3078 | { |
3079 | struct inode *inode = mapping->host; | |
3080 | int ret = 0, ret2; | |
3081 | handle_t *handle = ext4_journal_current_handle(); | |
3082 | loff_t new_i_size; | |
632eaeab | 3083 | unsigned long start, end; |
79f0be8d AK |
3084 | int write_mode = (int)(unsigned long)fsdata; |
3085 | ||
3086 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
3087 | if (ext4_should_order_data(inode)) { | |
3088 | return ext4_ordered_write_end(file, mapping, pos, | |
3089 | len, copied, page, fsdata); | |
3090 | } else if (ext4_should_writeback_data(inode)) { | |
3091 | return ext4_writeback_write_end(file, mapping, pos, | |
3092 | len, copied, page, fsdata); | |
3093 | } else { | |
3094 | BUG(); | |
3095 | } | |
3096 | } | |
632eaeab | 3097 | |
9bffad1e | 3098 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 3099 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 3100 | end = start + copied - 1; |
64769240 AT |
3101 | |
3102 | /* | |
3103 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3104 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3105 | * into that. | |
3106 | */ | |
3107 | ||
3108 | new_i_size = pos + copied; | |
632eaeab MC |
3109 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
3110 | if (ext4_da_should_update_i_disksize(page, end)) { | |
3111 | down_write(&EXT4_I(inode)->i_data_sem); | |
3112 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
3113 | /* | |
3114 | * Updating i_disksize when extending file | |
3115 | * without needing block allocation | |
3116 | */ | |
3117 | if (ext4_should_order_data(inode)) | |
3118 | ret = ext4_jbd2_file_inode(handle, | |
3119 | inode); | |
64769240 | 3120 | |
632eaeab MC |
3121 | EXT4_I(inode)->i_disksize = new_i_size; |
3122 | } | |
3123 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
3124 | /* We need to mark inode dirty even if |
3125 | * new_i_size is less that inode->i_size | |
3126 | * bu greater than i_disksize.(hint delalloc) | |
3127 | */ | |
3128 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3129 | } |
632eaeab | 3130 | } |
64769240 AT |
3131 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
3132 | page, fsdata); | |
3133 | copied = ret2; | |
3134 | if (ret2 < 0) | |
3135 | ret = ret2; | |
3136 | ret2 = ext4_journal_stop(handle); | |
3137 | if (!ret) | |
3138 | ret = ret2; | |
3139 | ||
3140 | return ret ? ret : copied; | |
3141 | } | |
3142 | ||
3143 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
3144 | { | |
64769240 AT |
3145 | /* |
3146 | * Drop reserved blocks | |
3147 | */ | |
3148 | BUG_ON(!PageLocked(page)); | |
3149 | if (!page_has_buffers(page)) | |
3150 | goto out; | |
3151 | ||
d2a17637 | 3152 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
3153 | |
3154 | out: | |
3155 | ext4_invalidatepage(page, offset); | |
3156 | ||
3157 | return; | |
3158 | } | |
3159 | ||
ccd2506b TT |
3160 | /* |
3161 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3162 | */ | |
3163 | int ext4_alloc_da_blocks(struct inode *inode) | |
3164 | { | |
fb40ba0d TT |
3165 | trace_ext4_alloc_da_blocks(inode); |
3166 | ||
ccd2506b TT |
3167 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
3168 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
3169 | return 0; | |
3170 | ||
3171 | /* | |
3172 | * We do something simple for now. The filemap_flush() will | |
3173 | * also start triggering a write of the data blocks, which is | |
3174 | * not strictly speaking necessary (and for users of | |
3175 | * laptop_mode, not even desirable). However, to do otherwise | |
3176 | * would require replicating code paths in: | |
de9a55b8 | 3177 | * |
ccd2506b TT |
3178 | * ext4_da_writepages() -> |
3179 | * write_cache_pages() ---> (via passed in callback function) | |
3180 | * __mpage_da_writepage() --> | |
3181 | * mpage_add_bh_to_extent() | |
3182 | * mpage_da_map_blocks() | |
3183 | * | |
3184 | * The problem is that write_cache_pages(), located in | |
3185 | * mm/page-writeback.c, marks pages clean in preparation for | |
3186 | * doing I/O, which is not desirable if we're not planning on | |
3187 | * doing I/O at all. | |
3188 | * | |
3189 | * We could call write_cache_pages(), and then redirty all of | |
3190 | * the pages by calling redirty_page_for_writeback() but that | |
3191 | * would be ugly in the extreme. So instead we would need to | |
3192 | * replicate parts of the code in the above functions, | |
3193 | * simplifying them becuase we wouldn't actually intend to | |
3194 | * write out the pages, but rather only collect contiguous | |
3195 | * logical block extents, call the multi-block allocator, and | |
3196 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3197 | * |
ccd2506b TT |
3198 | * For now, though, we'll cheat by calling filemap_flush(), |
3199 | * which will map the blocks, and start the I/O, but not | |
3200 | * actually wait for the I/O to complete. | |
3201 | */ | |
3202 | return filemap_flush(inode->i_mapping); | |
3203 | } | |
64769240 | 3204 | |
ac27a0ec DK |
3205 | /* |
3206 | * bmap() is special. It gets used by applications such as lilo and by | |
3207 | * the swapper to find the on-disk block of a specific piece of data. | |
3208 | * | |
3209 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3210 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3211 | * filesystem and enables swap, then they may get a nasty shock when the |
3212 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3213 | * the original zero's written out previously to the journal and | |
3214 | * awaiting writeback in the kernel's buffer cache. | |
3215 | * | |
3216 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3217 | * take extra steps to flush any blocks which might be in the cache. | |
3218 | */ | |
617ba13b | 3219 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3220 | { |
3221 | struct inode *inode = mapping->host; | |
3222 | journal_t *journal; | |
3223 | int err; | |
3224 | ||
64769240 AT |
3225 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3226 | test_opt(inode->i_sb, DELALLOC)) { | |
3227 | /* | |
3228 | * With delalloc we want to sync the file | |
3229 | * so that we can make sure we allocate | |
3230 | * blocks for file | |
3231 | */ | |
3232 | filemap_write_and_wait(mapping); | |
3233 | } | |
3234 | ||
0390131b | 3235 | if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
ac27a0ec DK |
3236 | /* |
3237 | * This is a REALLY heavyweight approach, but the use of | |
3238 | * bmap on dirty files is expected to be extremely rare: | |
3239 | * only if we run lilo or swapon on a freshly made file | |
3240 | * do we expect this to happen. | |
3241 | * | |
3242 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3243 | * represent an unprivileged user DOS attack --- we'd be | |
3244 | * in trouble if mortal users could trigger this path at | |
3245 | * will.) | |
3246 | * | |
617ba13b | 3247 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3248 | * regular files. If somebody wants to bmap a directory |
3249 | * or symlink and gets confused because the buffer | |
3250 | * hasn't yet been flushed to disk, they deserve | |
3251 | * everything they get. | |
3252 | */ | |
3253 | ||
617ba13b MC |
3254 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
3255 | journal = EXT4_JOURNAL(inode); | |
dab291af MC |
3256 | jbd2_journal_lock_updates(journal); |
3257 | err = jbd2_journal_flush(journal); | |
3258 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3259 | |
3260 | if (err) | |
3261 | return 0; | |
3262 | } | |
3263 | ||
af5bc92d | 3264 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3265 | } |
3266 | ||
617ba13b | 3267 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3268 | { |
617ba13b | 3269 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
3270 | } |
3271 | ||
3272 | static int | |
617ba13b | 3273 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3274 | struct list_head *pages, unsigned nr_pages) |
3275 | { | |
617ba13b | 3276 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3277 | } |
3278 | ||
617ba13b | 3279 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3280 | { |
617ba13b | 3281 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3282 | |
3283 | /* | |
3284 | * If it's a full truncate we just forget about the pending dirtying | |
3285 | */ | |
3286 | if (offset == 0) | |
3287 | ClearPageChecked(page); | |
3288 | ||
0390131b FM |
3289 | if (journal) |
3290 | jbd2_journal_invalidatepage(journal, page, offset); | |
3291 | else | |
3292 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
3293 | } |
3294 | ||
617ba13b | 3295 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3296 | { |
617ba13b | 3297 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3298 | |
3299 | WARN_ON(PageChecked(page)); | |
3300 | if (!page_has_buffers(page)) | |
3301 | return 0; | |
0390131b FM |
3302 | if (journal) |
3303 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3304 | else | |
3305 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3306 | } |
3307 | ||
3308 | /* | |
4c0425ff MC |
3309 | * O_DIRECT for ext3 (or indirect map) based files |
3310 | * | |
ac27a0ec DK |
3311 | * If the O_DIRECT write will extend the file then add this inode to the |
3312 | * orphan list. So recovery will truncate it back to the original size | |
3313 | * if the machine crashes during the write. | |
3314 | * | |
3315 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
3316 | * crashes then stale disk data _may_ be exposed inside the file. But current |
3317 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 3318 | */ |
4c0425ff | 3319 | static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, |
de9a55b8 TT |
3320 | const struct iovec *iov, loff_t offset, |
3321 | unsigned long nr_segs) | |
ac27a0ec DK |
3322 | { |
3323 | struct file *file = iocb->ki_filp; | |
3324 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 3325 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 3326 | handle_t *handle; |
ac27a0ec DK |
3327 | ssize_t ret; |
3328 | int orphan = 0; | |
3329 | size_t count = iov_length(iov, nr_segs); | |
fbbf6945 | 3330 | int retries = 0; |
ac27a0ec DK |
3331 | |
3332 | if (rw == WRITE) { | |
3333 | loff_t final_size = offset + count; | |
3334 | ||
ac27a0ec | 3335 | if (final_size > inode->i_size) { |
7fb5409d JK |
3336 | /* Credits for sb + inode write */ |
3337 | handle = ext4_journal_start(inode, 2); | |
3338 | if (IS_ERR(handle)) { | |
3339 | ret = PTR_ERR(handle); | |
3340 | goto out; | |
3341 | } | |
617ba13b | 3342 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
3343 | if (ret) { |
3344 | ext4_journal_stop(handle); | |
3345 | goto out; | |
3346 | } | |
ac27a0ec DK |
3347 | orphan = 1; |
3348 | ei->i_disksize = inode->i_size; | |
7fb5409d | 3349 | ext4_journal_stop(handle); |
ac27a0ec DK |
3350 | } |
3351 | } | |
3352 | ||
fbbf6945 | 3353 | retry: |
ac27a0ec DK |
3354 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, |
3355 | offset, nr_segs, | |
617ba13b | 3356 | ext4_get_block, NULL); |
fbbf6945 ES |
3357 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3358 | goto retry; | |
ac27a0ec | 3359 | |
7fb5409d | 3360 | if (orphan) { |
ac27a0ec DK |
3361 | int err; |
3362 | ||
7fb5409d JK |
3363 | /* Credits for sb + inode write */ |
3364 | handle = ext4_journal_start(inode, 2); | |
3365 | if (IS_ERR(handle)) { | |
3366 | /* This is really bad luck. We've written the data | |
3367 | * but cannot extend i_size. Bail out and pretend | |
3368 | * the write failed... */ | |
3369 | ret = PTR_ERR(handle); | |
3370 | goto out; | |
3371 | } | |
3372 | if (inode->i_nlink) | |
617ba13b | 3373 | ext4_orphan_del(handle, inode); |
7fb5409d | 3374 | if (ret > 0) { |
ac27a0ec DK |
3375 | loff_t end = offset + ret; |
3376 | if (end > inode->i_size) { | |
3377 | ei->i_disksize = end; | |
3378 | i_size_write(inode, end); | |
3379 | /* | |
3380 | * We're going to return a positive `ret' | |
3381 | * here due to non-zero-length I/O, so there's | |
3382 | * no way of reporting error returns from | |
617ba13b | 3383 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
3384 | * ignore it. |
3385 | */ | |
617ba13b | 3386 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3387 | } |
3388 | } | |
617ba13b | 3389 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3390 | if (ret == 0) |
3391 | ret = err; | |
3392 | } | |
3393 | out: | |
3394 | return ret; | |
3395 | } | |
3396 | ||
4c0425ff MC |
3397 | static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock, |
3398 | struct buffer_head *bh_result, int create) | |
3399 | { | |
3400 | handle_t *handle = NULL; | |
3401 | int ret = 0; | |
3402 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
3403 | int dio_credits; | |
3404 | ||
8d5d02e6 MC |
3405 | ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n", |
3406 | inode->i_ino, create); | |
4c0425ff MC |
3407 | /* |
3408 | * DIO VFS code passes create = 0 flag for write to | |
3409 | * the middle of file. It does this to avoid block | |
3410 | * allocation for holes, to prevent expose stale data | |
3411 | * out when there is parallel buffered read (which does | |
3412 | * not hold the i_mutex lock) while direct IO write has | |
3413 | * not completed. DIO request on holes finally falls back | |
3414 | * to buffered IO for this reason. | |
3415 | * | |
3416 | * For ext4 extent based file, since we support fallocate, | |
3417 | * new allocated extent as uninitialized, for holes, we | |
3418 | * could fallocate blocks for holes, thus parallel | |
3419 | * buffered IO read will zero out the page when read on | |
3420 | * a hole while parallel DIO write to the hole has not completed. | |
3421 | * | |
3422 | * when we come here, we know it's a direct IO write to | |
3423 | * to the middle of file (<i_size) | |
3424 | * so it's safe to override the create flag from VFS. | |
3425 | */ | |
3426 | create = EXT4_GET_BLOCKS_DIO_CREATE_EXT; | |
3427 | ||
3428 | if (max_blocks > DIO_MAX_BLOCKS) | |
3429 | max_blocks = DIO_MAX_BLOCKS; | |
3430 | dio_credits = ext4_chunk_trans_blocks(inode, max_blocks); | |
3431 | handle = ext4_journal_start(inode, dio_credits); | |
3432 | if (IS_ERR(handle)) { | |
3433 | ret = PTR_ERR(handle); | |
3434 | goto out; | |
3435 | } | |
3436 | ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result, | |
3437 | create); | |
3438 | if (ret > 0) { | |
3439 | bh_result->b_size = (ret << inode->i_blkbits); | |
3440 | ret = 0; | |
3441 | } | |
3442 | ext4_journal_stop(handle); | |
3443 | out: | |
3444 | return ret; | |
3445 | } | |
3446 | ||
4c0425ff MC |
3447 | static void ext4_free_io_end(ext4_io_end_t *io) |
3448 | { | |
8d5d02e6 MC |
3449 | BUG_ON(!io); |
3450 | iput(io->inode); | |
4c0425ff MC |
3451 | kfree(io); |
3452 | } | |
8d5d02e6 MC |
3453 | static void dump_aio_dio_list(struct inode * inode) |
3454 | { | |
3455 | #ifdef EXT4_DEBUG | |
3456 | struct list_head *cur, *before, *after; | |
3457 | ext4_io_end_t *io, *io0, *io1; | |
3458 | ||
3459 | if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){ | |
3460 | ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino); | |
3461 | return; | |
3462 | } | |
3463 | ||
3464 | ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino); | |
3465 | list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){ | |
3466 | cur = &io->list; | |
3467 | before = cur->prev; | |
3468 | io0 = container_of(before, ext4_io_end_t, list); | |
3469 | after = cur->next; | |
3470 | io1 = container_of(after, ext4_io_end_t, list); | |
3471 | ||
3472 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
3473 | io, inode->i_ino, io0, io1); | |
3474 | } | |
3475 | #endif | |
3476 | } | |
4c0425ff MC |
3477 | |
3478 | /* | |
4c0425ff MC |
3479 | * check a range of space and convert unwritten extents to written. |
3480 | */ | |
8d5d02e6 | 3481 | static int ext4_end_aio_dio_nolock(ext4_io_end_t *io) |
4c0425ff | 3482 | { |
4c0425ff MC |
3483 | struct inode *inode = io->inode; |
3484 | loff_t offset = io->offset; | |
3485 | size_t size = io->size; | |
3486 | int ret = 0; | |
4c0425ff | 3487 | |
8d5d02e6 MC |
3488 | ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p," |
3489 | "list->prev 0x%p\n", | |
3490 | io, inode->i_ino, io->list.next, io->list.prev); | |
3491 | ||
3492 | if (list_empty(&io->list)) | |
3493 | return ret; | |
3494 | ||
3495 | if (io->flag != DIO_AIO_UNWRITTEN) | |
3496 | return ret; | |
3497 | ||
4c0425ff MC |
3498 | if (offset + size <= i_size_read(inode)) |
3499 | ret = ext4_convert_unwritten_extents(inode, offset, size); | |
3500 | ||
8d5d02e6 | 3501 | if (ret < 0) { |
4c0425ff | 3502 | printk(KERN_EMERG "%s: failed to convert unwritten" |
8d5d02e6 MC |
3503 | "extents to written extents, error is %d" |
3504 | " io is still on inode %lu aio dio list\n", | |
3505 | __func__, ret, inode->i_ino); | |
3506 | return ret; | |
3507 | } | |
4c0425ff | 3508 | |
8d5d02e6 MC |
3509 | /* clear the DIO AIO unwritten flag */ |
3510 | io->flag = 0; | |
3511 | return ret; | |
4c0425ff | 3512 | } |
8d5d02e6 MC |
3513 | /* |
3514 | * work on completed aio dio IO, to convert unwritten extents to extents | |
3515 | */ | |
3516 | static void ext4_end_aio_dio_work(struct work_struct *work) | |
3517 | { | |
3518 | ext4_io_end_t *io = container_of(work, ext4_io_end_t, work); | |
3519 | struct inode *inode = io->inode; | |
3520 | int ret = 0; | |
4c0425ff | 3521 | |
8d5d02e6 MC |
3522 | mutex_lock(&inode->i_mutex); |
3523 | ret = ext4_end_aio_dio_nolock(io); | |
3524 | if (ret >= 0) { | |
3525 | if (!list_empty(&io->list)) | |
3526 | list_del_init(&io->list); | |
3527 | ext4_free_io_end(io); | |
3528 | } | |
3529 | mutex_unlock(&inode->i_mutex); | |
3530 | } | |
3531 | /* | |
3532 | * This function is called from ext4_sync_file(). | |
3533 | * | |
3534 | * When AIO DIO IO is completed, the work to convert unwritten | |
3535 | * extents to written is queued on workqueue but may not get immediately | |
3536 | * scheduled. When fsync is called, we need to ensure the | |
3537 | * conversion is complete before fsync returns. | |
3538 | * The inode keeps track of a list of completed AIO from DIO path | |
3539 | * that might needs to do the conversion. This function walks through | |
3540 | * the list and convert the related unwritten extents to written. | |
3541 | */ | |
3542 | int flush_aio_dio_completed_IO(struct inode *inode) | |
3543 | { | |
3544 | ext4_io_end_t *io; | |
3545 | int ret = 0; | |
3546 | int ret2 = 0; | |
3547 | ||
3548 | if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)) | |
3549 | return ret; | |
3550 | ||
3551 | dump_aio_dio_list(inode); | |
3552 | while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){ | |
3553 | io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next, | |
3554 | ext4_io_end_t, list); | |
3555 | /* | |
3556 | * Calling ext4_end_aio_dio_nolock() to convert completed | |
3557 | * IO to written. | |
3558 | * | |
3559 | * When ext4_sync_file() is called, run_queue() may already | |
3560 | * about to flush the work corresponding to this io structure. | |
3561 | * It will be upset if it founds the io structure related | |
3562 | * to the work-to-be schedule is freed. | |
3563 | * | |
3564 | * Thus we need to keep the io structure still valid here after | |
3565 | * convertion finished. The io structure has a flag to | |
3566 | * avoid double converting from both fsync and background work | |
3567 | * queue work. | |
3568 | */ | |
3569 | ret = ext4_end_aio_dio_nolock(io); | |
3570 | if (ret < 0) | |
3571 | ret2 = ret; | |
3572 | else | |
3573 | list_del_init(&io->list); | |
3574 | } | |
3575 | return (ret2 < 0) ? ret2 : 0; | |
3576 | } | |
3577 | ||
3578 | static ext4_io_end_t *ext4_init_io_end (struct inode *inode) | |
4c0425ff MC |
3579 | { |
3580 | ext4_io_end_t *io = NULL; | |
3581 | ||
3582 | io = kmalloc(sizeof(*io), GFP_NOFS); | |
3583 | ||
3584 | if (io) { | |
8d5d02e6 | 3585 | igrab(inode); |
4c0425ff | 3586 | io->inode = inode; |
8d5d02e6 | 3587 | io->flag = 0; |
4c0425ff MC |
3588 | io->offset = 0; |
3589 | io->size = 0; | |
3590 | io->error = 0; | |
8d5d02e6 MC |
3591 | INIT_WORK(&io->work, ext4_end_aio_dio_work); |
3592 | INIT_LIST_HEAD(&io->list); | |
4c0425ff MC |
3593 | } |
3594 | ||
3595 | return io; | |
3596 | } | |
3597 | ||
3598 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, | |
3599 | ssize_t size, void *private) | |
3600 | { | |
3601 | ext4_io_end_t *io_end = iocb->private; | |
3602 | struct workqueue_struct *wq; | |
3603 | ||
4b70df18 M |
3604 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3605 | if (!io_end || !size) | |
3606 | return; | |
3607 | ||
8d5d02e6 MC |
3608 | ext_debug("ext4_end_io_dio(): io_end 0x%p" |
3609 | "for inode %lu, iocb 0x%p, offset %llu, size %llu\n", | |
3610 | iocb->private, io_end->inode->i_ino, iocb, offset, | |
3611 | size); | |
8d5d02e6 MC |
3612 | |
3613 | /* if not aio dio with unwritten extents, just free io and return */ | |
3614 | if (io_end->flag != DIO_AIO_UNWRITTEN){ | |
3615 | ext4_free_io_end(io_end); | |
3616 | iocb->private = NULL; | |
4c0425ff | 3617 | return; |
8d5d02e6 MC |
3618 | } |
3619 | ||
4c0425ff MC |
3620 | io_end->offset = offset; |
3621 | io_end->size = size; | |
3622 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; | |
3623 | ||
8d5d02e6 | 3624 | /* queue the work to convert unwritten extents to written */ |
4c0425ff MC |
3625 | queue_work(wq, &io_end->work); |
3626 | ||
8d5d02e6 MC |
3627 | /* Add the io_end to per-inode completed aio dio list*/ |
3628 | list_add_tail(&io_end->list, | |
3629 | &EXT4_I(io_end->inode)->i_aio_dio_complete_list); | |
4c0425ff MC |
3630 | iocb->private = NULL; |
3631 | } | |
3632 | /* | |
3633 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3634 | * preallocated extents, and those write extend the file, no need to | |
3635 | * fall back to buffered IO. | |
3636 | * | |
3637 | * For holes, we fallocate those blocks, mark them as unintialized | |
3638 | * If those blocks were preallocated, we mark sure they are splited, but | |
3639 | * still keep the range to write as unintialized. | |
3640 | * | |
8d5d02e6 MC |
3641 | * The unwrritten extents will be converted to written when DIO is completed. |
3642 | * For async direct IO, since the IO may still pending when return, we | |
3643 | * set up an end_io call back function, which will do the convertion | |
3644 | * when async direct IO completed. | |
4c0425ff MC |
3645 | * |
3646 | * If the O_DIRECT write will extend the file then add this inode to the | |
3647 | * orphan list. So recovery will truncate it back to the original size | |
3648 | * if the machine crashes during the write. | |
3649 | * | |
3650 | */ | |
3651 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3652 | const struct iovec *iov, loff_t offset, | |
3653 | unsigned long nr_segs) | |
3654 | { | |
3655 | struct file *file = iocb->ki_filp; | |
3656 | struct inode *inode = file->f_mapping->host; | |
3657 | ssize_t ret; | |
3658 | size_t count = iov_length(iov, nr_segs); | |
3659 | ||
3660 | loff_t final_size = offset + count; | |
3661 | if (rw == WRITE && final_size <= inode->i_size) { | |
3662 | /* | |
8d5d02e6 MC |
3663 | * We could direct write to holes and fallocate. |
3664 | * | |
3665 | * Allocated blocks to fill the hole are marked as uninitialized | |
4c0425ff MC |
3666 | * to prevent paralel buffered read to expose the stale data |
3667 | * before DIO complete the data IO. | |
8d5d02e6 MC |
3668 | * |
3669 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
3670 | * will just simply mark the buffer mapped but still |
3671 | * keep the extents uninitialized. | |
3672 | * | |
8d5d02e6 MC |
3673 | * for non AIO case, we will convert those unwritten extents |
3674 | * to written after return back from blockdev_direct_IO. | |
3675 | * | |
3676 | * for async DIO, the conversion needs to be defered when | |
3677 | * the IO is completed. The ext4 end_io callback function | |
3678 | * will be called to take care of the conversion work. | |
3679 | * Here for async case, we allocate an io_end structure to | |
3680 | * hook to the iocb. | |
4c0425ff | 3681 | */ |
8d5d02e6 MC |
3682 | iocb->private = NULL; |
3683 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3684 | if (!is_sync_kiocb(iocb)) { | |
3685 | iocb->private = ext4_init_io_end(inode); | |
3686 | if (!iocb->private) | |
3687 | return -ENOMEM; | |
3688 | /* | |
3689 | * we save the io structure for current async | |
3690 | * direct IO, so that later ext4_get_blocks() | |
3691 | * could flag the io structure whether there | |
3692 | * is a unwritten extents needs to be converted | |
3693 | * when IO is completed. | |
3694 | */ | |
3695 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
3696 | } | |
3697 | ||
4c0425ff MC |
3698 | ret = blockdev_direct_IO(rw, iocb, inode, |
3699 | inode->i_sb->s_bdev, iov, | |
3700 | offset, nr_segs, | |
3701 | ext4_get_block_dio_write, | |
3702 | ext4_end_io_dio); | |
8d5d02e6 MC |
3703 | if (iocb->private) |
3704 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3705 | /* | |
3706 | * The io_end structure takes a reference to the inode, | |
3707 | * that structure needs to be destroyed and the | |
3708 | * reference to the inode need to be dropped, when IO is | |
3709 | * complete, even with 0 byte write, or failed. | |
3710 | * | |
3711 | * In the successful AIO DIO case, the io_end structure will be | |
3712 | * desctroyed and the reference to the inode will be dropped | |
3713 | * after the end_io call back function is called. | |
3714 | * | |
3715 | * In the case there is 0 byte write, or error case, since | |
3716 | * VFS direct IO won't invoke the end_io call back function, | |
3717 | * we need to free the end_io structure here. | |
3718 | */ | |
3719 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3720 | ext4_free_io_end(iocb->private); | |
3721 | iocb->private = NULL; | |
5f524950 M |
3722 | } else if (ret > 0 && (EXT4_I(inode)->i_state & |
3723 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 3724 | int err; |
8d5d02e6 MC |
3725 | /* |
3726 | * for non AIO case, since the IO is already | |
3727 | * completed, we could do the convertion right here | |
3728 | */ | |
109f5565 M |
3729 | err = ext4_convert_unwritten_extents(inode, |
3730 | offset, ret); | |
3731 | if (err < 0) | |
3732 | ret = err; | |
5f524950 | 3733 | EXT4_I(inode)->i_state &= ~EXT4_STATE_DIO_UNWRITTEN; |
109f5565 | 3734 | } |
4c0425ff MC |
3735 | return ret; |
3736 | } | |
8d5d02e6 MC |
3737 | |
3738 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
3739 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
3740 | } | |
3741 | ||
3742 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3743 | const struct iovec *iov, loff_t offset, | |
3744 | unsigned long nr_segs) | |
3745 | { | |
3746 | struct file *file = iocb->ki_filp; | |
3747 | struct inode *inode = file->f_mapping->host; | |
3748 | ||
3749 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) | |
3750 | return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3751 | ||
3752 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3753 | } | |
3754 | ||
ac27a0ec | 3755 | /* |
617ba13b | 3756 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3757 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3758 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3759 | * not necessarily locked. | |
3760 | * | |
3761 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3762 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3763 | * or jbddirty because all the journalling code will explode. | |
3764 | * | |
3765 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3766 | * is called, propagate that into the buffers appropriately. | |
3767 | */ | |
617ba13b | 3768 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3769 | { |
3770 | SetPageChecked(page); | |
3771 | return __set_page_dirty_nobuffers(page); | |
3772 | } | |
3773 | ||
617ba13b | 3774 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3775 | .readpage = ext4_readpage, |
3776 | .readpages = ext4_readpages, | |
43ce1d23 | 3777 | .writepage = ext4_writepage, |
8ab22b9a HH |
3778 | .sync_page = block_sync_page, |
3779 | .write_begin = ext4_write_begin, | |
3780 | .write_end = ext4_ordered_write_end, | |
3781 | .bmap = ext4_bmap, | |
3782 | .invalidatepage = ext4_invalidatepage, | |
3783 | .releasepage = ext4_releasepage, | |
3784 | .direct_IO = ext4_direct_IO, | |
3785 | .migratepage = buffer_migrate_page, | |
3786 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3787 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3788 | }; |
3789 | ||
617ba13b | 3790 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3791 | .readpage = ext4_readpage, |
3792 | .readpages = ext4_readpages, | |
43ce1d23 | 3793 | .writepage = ext4_writepage, |
8ab22b9a HH |
3794 | .sync_page = block_sync_page, |
3795 | .write_begin = ext4_write_begin, | |
3796 | .write_end = ext4_writeback_write_end, | |
3797 | .bmap = ext4_bmap, | |
3798 | .invalidatepage = ext4_invalidatepage, | |
3799 | .releasepage = ext4_releasepage, | |
3800 | .direct_IO = ext4_direct_IO, | |
3801 | .migratepage = buffer_migrate_page, | |
3802 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3803 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3804 | }; |
3805 | ||
617ba13b | 3806 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3807 | .readpage = ext4_readpage, |
3808 | .readpages = ext4_readpages, | |
43ce1d23 | 3809 | .writepage = ext4_writepage, |
8ab22b9a HH |
3810 | .sync_page = block_sync_page, |
3811 | .write_begin = ext4_write_begin, | |
3812 | .write_end = ext4_journalled_write_end, | |
3813 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3814 | .bmap = ext4_bmap, | |
3815 | .invalidatepage = ext4_invalidatepage, | |
3816 | .releasepage = ext4_releasepage, | |
3817 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3818 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3819 | }; |
3820 | ||
64769240 | 3821 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3822 | .readpage = ext4_readpage, |
3823 | .readpages = ext4_readpages, | |
43ce1d23 | 3824 | .writepage = ext4_writepage, |
8ab22b9a HH |
3825 | .writepages = ext4_da_writepages, |
3826 | .sync_page = block_sync_page, | |
3827 | .write_begin = ext4_da_write_begin, | |
3828 | .write_end = ext4_da_write_end, | |
3829 | .bmap = ext4_bmap, | |
3830 | .invalidatepage = ext4_da_invalidatepage, | |
3831 | .releasepage = ext4_releasepage, | |
3832 | .direct_IO = ext4_direct_IO, | |
3833 | .migratepage = buffer_migrate_page, | |
3834 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3835 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3836 | }; |
3837 | ||
617ba13b | 3838 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3839 | { |
cd1aac32 AK |
3840 | if (ext4_should_order_data(inode) && |
3841 | test_opt(inode->i_sb, DELALLOC)) | |
3842 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3843 | else if (ext4_should_order_data(inode)) | |
617ba13b | 3844 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
3845 | else if (ext4_should_writeback_data(inode) && |
3846 | test_opt(inode->i_sb, DELALLOC)) | |
3847 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
3848 | else if (ext4_should_writeback_data(inode)) |
3849 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 3850 | else |
617ba13b | 3851 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
3852 | } |
3853 | ||
3854 | /* | |
617ba13b | 3855 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
3856 | * up to the end of the block which corresponds to `from'. |
3857 | * This required during truncate. We need to physically zero the tail end | |
3858 | * of that block so it doesn't yield old data if the file is later grown. | |
3859 | */ | |
cf108bca | 3860 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
3861 | struct address_space *mapping, loff_t from) |
3862 | { | |
617ba13b | 3863 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 3864 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
3865 | unsigned blocksize, length, pos; |
3866 | ext4_lblk_t iblock; | |
ac27a0ec DK |
3867 | struct inode *inode = mapping->host; |
3868 | struct buffer_head *bh; | |
cf108bca | 3869 | struct page *page; |
ac27a0ec | 3870 | int err = 0; |
ac27a0ec | 3871 | |
f4a01017 TT |
3872 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
3873 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
cf108bca JK |
3874 | if (!page) |
3875 | return -EINVAL; | |
3876 | ||
ac27a0ec DK |
3877 | blocksize = inode->i_sb->s_blocksize; |
3878 | length = blocksize - (offset & (blocksize - 1)); | |
3879 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3880 | ||
3881 | /* | |
3882 | * For "nobh" option, we can only work if we don't need to | |
3883 | * read-in the page - otherwise we create buffers to do the IO. | |
3884 | */ | |
3885 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && | |
617ba13b | 3886 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
eebd2aa3 | 3887 | zero_user(page, offset, length); |
ac27a0ec DK |
3888 | set_page_dirty(page); |
3889 | goto unlock; | |
3890 | } | |
3891 | ||
3892 | if (!page_has_buffers(page)) | |
3893 | create_empty_buffers(page, blocksize, 0); | |
3894 | ||
3895 | /* Find the buffer that contains "offset" */ | |
3896 | bh = page_buffers(page); | |
3897 | pos = blocksize; | |
3898 | while (offset >= pos) { | |
3899 | bh = bh->b_this_page; | |
3900 | iblock++; | |
3901 | pos += blocksize; | |
3902 | } | |
3903 | ||
3904 | err = 0; | |
3905 | if (buffer_freed(bh)) { | |
3906 | BUFFER_TRACE(bh, "freed: skip"); | |
3907 | goto unlock; | |
3908 | } | |
3909 | ||
3910 | if (!buffer_mapped(bh)) { | |
3911 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3912 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3913 | /* unmapped? It's a hole - nothing to do */ |
3914 | if (!buffer_mapped(bh)) { | |
3915 | BUFFER_TRACE(bh, "still unmapped"); | |
3916 | goto unlock; | |
3917 | } | |
3918 | } | |
3919 | ||
3920 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3921 | if (PageUptodate(page)) | |
3922 | set_buffer_uptodate(bh); | |
3923 | ||
3924 | if (!buffer_uptodate(bh)) { | |
3925 | err = -EIO; | |
3926 | ll_rw_block(READ, 1, &bh); | |
3927 | wait_on_buffer(bh); | |
3928 | /* Uhhuh. Read error. Complain and punt. */ | |
3929 | if (!buffer_uptodate(bh)) | |
3930 | goto unlock; | |
3931 | } | |
3932 | ||
617ba13b | 3933 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3934 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3935 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3936 | if (err) |
3937 | goto unlock; | |
3938 | } | |
3939 | ||
eebd2aa3 | 3940 | zero_user(page, offset, length); |
ac27a0ec DK |
3941 | |
3942 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3943 | ||
3944 | err = 0; | |
617ba13b | 3945 | if (ext4_should_journal_data(inode)) { |
0390131b | 3946 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
ac27a0ec | 3947 | } else { |
617ba13b | 3948 | if (ext4_should_order_data(inode)) |
678aaf48 | 3949 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
3950 | mark_buffer_dirty(bh); |
3951 | } | |
3952 | ||
3953 | unlock: | |
3954 | unlock_page(page); | |
3955 | page_cache_release(page); | |
3956 | return err; | |
3957 | } | |
3958 | ||
3959 | /* | |
3960 | * Probably it should be a library function... search for first non-zero word | |
3961 | * or memcmp with zero_page, whatever is better for particular architecture. | |
3962 | * Linus? | |
3963 | */ | |
3964 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
3965 | { | |
3966 | while (p < q) | |
3967 | if (*p++) | |
3968 | return 0; | |
3969 | return 1; | |
3970 | } | |
3971 | ||
3972 | /** | |
617ba13b | 3973 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
3974 | * @inode: inode in question |
3975 | * @depth: depth of the affected branch | |
617ba13b | 3976 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
3977 | * @chain: place to store the pointers to partial indirect blocks |
3978 | * @top: place to the (detached) top of branch | |
3979 | * | |
617ba13b | 3980 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
3981 | * |
3982 | * When we do truncate() we may have to clean the ends of several | |
3983 | * indirect blocks but leave the blocks themselves alive. Block is | |
3984 | * partially truncated if some data below the new i_size is refered | |
3985 | * from it (and it is on the path to the first completely truncated | |
3986 | * data block, indeed). We have to free the top of that path along | |
3987 | * with everything to the right of the path. Since no allocation | |
617ba13b | 3988 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
3989 | * finishes, we may safely do the latter, but top of branch may |
3990 | * require special attention - pageout below the truncation point | |
3991 | * might try to populate it. | |
3992 | * | |
3993 | * We atomically detach the top of branch from the tree, store the | |
3994 | * block number of its root in *@top, pointers to buffer_heads of | |
3995 | * partially truncated blocks - in @chain[].bh and pointers to | |
3996 | * their last elements that should not be removed - in | |
3997 | * @chain[].p. Return value is the pointer to last filled element | |
3998 | * of @chain. | |
3999 | * | |
4000 | * The work left to caller to do the actual freeing of subtrees: | |
4001 | * a) free the subtree starting from *@top | |
4002 | * b) free the subtrees whose roots are stored in | |
4003 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
4004 | * c) free the subtrees growing from the inode past the @chain[0]. | |
4005 | * (no partially truncated stuff there). */ | |
4006 | ||
617ba13b | 4007 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
de9a55b8 TT |
4008 | ext4_lblk_t offsets[4], Indirect chain[4], |
4009 | __le32 *top) | |
ac27a0ec DK |
4010 | { |
4011 | Indirect *partial, *p; | |
4012 | int k, err; | |
4013 | ||
4014 | *top = 0; | |
4015 | /* Make k index the deepest non-null offest + 1 */ | |
4016 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
4017 | ; | |
617ba13b | 4018 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
4019 | /* Writer: pointers */ |
4020 | if (!partial) | |
4021 | partial = chain + k-1; | |
4022 | /* | |
4023 | * If the branch acquired continuation since we've looked at it - | |
4024 | * fine, it should all survive and (new) top doesn't belong to us. | |
4025 | */ | |
4026 | if (!partial->key && *partial->p) | |
4027 | /* Writer: end */ | |
4028 | goto no_top; | |
af5bc92d | 4029 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
ac27a0ec DK |
4030 | ; |
4031 | /* | |
4032 | * OK, we've found the last block that must survive. The rest of our | |
4033 | * branch should be detached before unlocking. However, if that rest | |
4034 | * of branch is all ours and does not grow immediately from the inode | |
4035 | * it's easier to cheat and just decrement partial->p. | |
4036 | */ | |
4037 | if (p == chain + k - 1 && p > chain) { | |
4038 | p->p--; | |
4039 | } else { | |
4040 | *top = *p->p; | |
617ba13b | 4041 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
4042 | #if 0 |
4043 | *p->p = 0; | |
4044 | #endif | |
4045 | } | |
4046 | /* Writer: end */ | |
4047 | ||
af5bc92d | 4048 | while (partial > p) { |
ac27a0ec DK |
4049 | brelse(partial->bh); |
4050 | partial--; | |
4051 | } | |
4052 | no_top: | |
4053 | return partial; | |
4054 | } | |
4055 | ||
4056 | /* | |
4057 | * Zero a number of block pointers in either an inode or an indirect block. | |
4058 | * If we restart the transaction we must again get write access to the | |
4059 | * indirect block for further modification. | |
4060 | * | |
4061 | * We release `count' blocks on disk, but (last - first) may be greater | |
4062 | * than `count' because there can be holes in there. | |
4063 | */ | |
617ba13b | 4064 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
4065 | struct buffer_head *bh, |
4066 | ext4_fsblk_t block_to_free, | |
4067 | unsigned long count, __le32 *first, | |
4068 | __le32 *last) | |
ac27a0ec DK |
4069 | { |
4070 | __le32 *p; | |
50689696 TT |
4071 | int is_metadata = S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode); |
4072 | ||
ac27a0ec DK |
4073 | if (try_to_extend_transaction(handle, inode)) { |
4074 | if (bh) { | |
0390131b FM |
4075 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
4076 | ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec | 4077 | } |
617ba13b | 4078 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4079 | ext4_truncate_restart_trans(handle, inode, |
4080 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4081 | if (bh) { |
4082 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 4083 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
4084 | } |
4085 | } | |
4086 | ||
4087 | /* | |
de9a55b8 TT |
4088 | * Any buffers which are on the journal will be in memory. We |
4089 | * find them on the hash table so jbd2_journal_revoke() will | |
4090 | * run jbd2_journal_forget() on them. We've already detached | |
4091 | * each block from the file, so bforget() in | |
4092 | * jbd2_journal_forget() should be safe. | |
ac27a0ec | 4093 | * |
dab291af | 4094 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
ac27a0ec DK |
4095 | */ |
4096 | for (p = first; p < last; p++) { | |
4097 | u32 nr = le32_to_cpu(*p); | |
4098 | if (nr) { | |
1d03ec98 | 4099 | struct buffer_head *tbh; |
ac27a0ec DK |
4100 | |
4101 | *p = 0; | |
1d03ec98 | 4102 | tbh = sb_find_get_block(inode->i_sb, nr); |
50689696 | 4103 | ext4_forget(handle, is_metadata, inode, tbh, nr); |
ac27a0ec DK |
4104 | } |
4105 | } | |
4106 | ||
50689696 | 4107 | ext4_free_blocks(handle, inode, block_to_free, count, is_metadata); |
ac27a0ec DK |
4108 | } |
4109 | ||
4110 | /** | |
617ba13b | 4111 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
4112 | * @handle: handle for this transaction |
4113 | * @inode: inode we are dealing with | |
4114 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
4115 | * @first: array of block numbers | |
4116 | * @last: points immediately past the end of array | |
4117 | * | |
4118 | * We are freeing all blocks refered from that array (numbers are stored as | |
4119 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
4120 | * | |
4121 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
4122 | * blocks are contiguous then releasing them at one time will only affect one | |
4123 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
4124 | * actually use a lot of journal space. | |
4125 | * | |
4126 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
4127 | * block pointers. | |
4128 | */ | |
617ba13b | 4129 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4130 | struct buffer_head *this_bh, |
4131 | __le32 *first, __le32 *last) | |
4132 | { | |
617ba13b | 4133 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
4134 | unsigned long count = 0; /* Number of blocks in the run */ |
4135 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
4136 | corresponding to | |
4137 | block_to_free */ | |
617ba13b | 4138 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
4139 | __le32 *p; /* Pointer into inode/ind |
4140 | for current block */ | |
4141 | int err; | |
4142 | ||
4143 | if (this_bh) { /* For indirect block */ | |
4144 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 4145 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
4146 | /* Important: if we can't update the indirect pointers |
4147 | * to the blocks, we can't free them. */ | |
4148 | if (err) | |
4149 | return; | |
4150 | } | |
4151 | ||
4152 | for (p = first; p < last; p++) { | |
4153 | nr = le32_to_cpu(*p); | |
4154 | if (nr) { | |
4155 | /* accumulate blocks to free if they're contiguous */ | |
4156 | if (count == 0) { | |
4157 | block_to_free = nr; | |
4158 | block_to_free_p = p; | |
4159 | count = 1; | |
4160 | } else if (nr == block_to_free + count) { | |
4161 | count++; | |
4162 | } else { | |
617ba13b | 4163 | ext4_clear_blocks(handle, inode, this_bh, |
ac27a0ec DK |
4164 | block_to_free, |
4165 | count, block_to_free_p, p); | |
4166 | block_to_free = nr; | |
4167 | block_to_free_p = p; | |
4168 | count = 1; | |
4169 | } | |
4170 | } | |
4171 | } | |
4172 | ||
4173 | if (count > 0) | |
617ba13b | 4174 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
4175 | count, block_to_free_p, p); |
4176 | ||
4177 | if (this_bh) { | |
0390131b | 4178 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
71dc8fbc DG |
4179 | |
4180 | /* | |
4181 | * The buffer head should have an attached journal head at this | |
4182 | * point. However, if the data is corrupted and an indirect | |
4183 | * block pointed to itself, it would have been detached when | |
4184 | * the block was cleared. Check for this instead of OOPSing. | |
4185 | */ | |
e7f07968 | 4186 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
0390131b | 4187 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
71dc8fbc DG |
4188 | else |
4189 | ext4_error(inode->i_sb, __func__, | |
4190 | "circular indirect block detected, " | |
4191 | "inode=%lu, block=%llu", | |
4192 | inode->i_ino, | |
4193 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
4194 | } |
4195 | } | |
4196 | ||
4197 | /** | |
617ba13b | 4198 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
4199 | * @handle: JBD handle for this transaction |
4200 | * @inode: inode we are dealing with | |
4201 | * @parent_bh: the buffer_head which contains *@first and *@last | |
4202 | * @first: array of block numbers | |
4203 | * @last: pointer immediately past the end of array | |
4204 | * @depth: depth of the branches to free | |
4205 | * | |
4206 | * We are freeing all blocks refered from these branches (numbers are | |
4207 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
4208 | * appropriately. | |
4209 | */ | |
617ba13b | 4210 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4211 | struct buffer_head *parent_bh, |
4212 | __le32 *first, __le32 *last, int depth) | |
4213 | { | |
617ba13b | 4214 | ext4_fsblk_t nr; |
ac27a0ec DK |
4215 | __le32 *p; |
4216 | ||
0390131b | 4217 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4218 | return; |
4219 | ||
4220 | if (depth--) { | |
4221 | struct buffer_head *bh; | |
617ba13b | 4222 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
4223 | p = last; |
4224 | while (--p >= first) { | |
4225 | nr = le32_to_cpu(*p); | |
4226 | if (!nr) | |
4227 | continue; /* A hole */ | |
4228 | ||
4229 | /* Go read the buffer for the next level down */ | |
4230 | bh = sb_bread(inode->i_sb, nr); | |
4231 | ||
4232 | /* | |
4233 | * A read failure? Report error and clear slot | |
4234 | * (should be rare). | |
4235 | */ | |
4236 | if (!bh) { | |
617ba13b | 4237 | ext4_error(inode->i_sb, "ext4_free_branches", |
2ae02107 | 4238 | "Read failure, inode=%lu, block=%llu", |
ac27a0ec DK |
4239 | inode->i_ino, nr); |
4240 | continue; | |
4241 | } | |
4242 | ||
4243 | /* This zaps the entire block. Bottom up. */ | |
4244 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 4245 | ext4_free_branches(handle, inode, bh, |
af5bc92d TT |
4246 | (__le32 *) bh->b_data, |
4247 | (__le32 *) bh->b_data + addr_per_block, | |
4248 | depth); | |
ac27a0ec DK |
4249 | |
4250 | /* | |
4251 | * We've probably journalled the indirect block several | |
4252 | * times during the truncate. But it's no longer | |
4253 | * needed and we now drop it from the transaction via | |
dab291af | 4254 | * jbd2_journal_revoke(). |
ac27a0ec DK |
4255 | * |
4256 | * That's easy if it's exclusively part of this | |
4257 | * transaction. But if it's part of the committing | |
dab291af | 4258 | * transaction then jbd2_journal_forget() will simply |
ac27a0ec | 4259 | * brelse() it. That means that if the underlying |
617ba13b | 4260 | * block is reallocated in ext4_get_block(), |
ac27a0ec DK |
4261 | * unmap_underlying_metadata() will find this block |
4262 | * and will try to get rid of it. damn, damn. | |
4263 | * | |
4264 | * If this block has already been committed to the | |
4265 | * journal, a revoke record will be written. And | |
4266 | * revoke records must be emitted *before* clearing | |
4267 | * this block's bit in the bitmaps. | |
4268 | */ | |
617ba13b | 4269 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
ac27a0ec DK |
4270 | |
4271 | /* | |
4272 | * Everything below this this pointer has been | |
4273 | * released. Now let this top-of-subtree go. | |
4274 | * | |
4275 | * We want the freeing of this indirect block to be | |
4276 | * atomic in the journal with the updating of the | |
4277 | * bitmap block which owns it. So make some room in | |
4278 | * the journal. | |
4279 | * | |
4280 | * We zero the parent pointer *after* freeing its | |
4281 | * pointee in the bitmaps, so if extend_transaction() | |
4282 | * for some reason fails to put the bitmap changes and | |
4283 | * the release into the same transaction, recovery | |
4284 | * will merely complain about releasing a free block, | |
4285 | * rather than leaking blocks. | |
4286 | */ | |
0390131b | 4287 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4288 | return; |
4289 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b | 4290 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4291 | ext4_truncate_restart_trans(handle, inode, |
4292 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4293 | } |
4294 | ||
c9de560d | 4295 | ext4_free_blocks(handle, inode, nr, 1, 1); |
ac27a0ec DK |
4296 | |
4297 | if (parent_bh) { | |
4298 | /* | |
4299 | * The block which we have just freed is | |
4300 | * pointed to by an indirect block: journal it | |
4301 | */ | |
4302 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 4303 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
4304 | parent_bh)){ |
4305 | *p = 0; | |
4306 | BUFFER_TRACE(parent_bh, | |
0390131b FM |
4307 | "call ext4_handle_dirty_metadata"); |
4308 | ext4_handle_dirty_metadata(handle, | |
4309 | inode, | |
4310 | parent_bh); | |
ac27a0ec DK |
4311 | } |
4312 | } | |
4313 | } | |
4314 | } else { | |
4315 | /* We have reached the bottom of the tree. */ | |
4316 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 4317 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
4318 | } |
4319 | } | |
4320 | ||
91ef4caf DG |
4321 | int ext4_can_truncate(struct inode *inode) |
4322 | { | |
4323 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4324 | return 0; | |
4325 | if (S_ISREG(inode->i_mode)) | |
4326 | return 1; | |
4327 | if (S_ISDIR(inode->i_mode)) | |
4328 | return 1; | |
4329 | if (S_ISLNK(inode->i_mode)) | |
4330 | return !ext4_inode_is_fast_symlink(inode); | |
4331 | return 0; | |
4332 | } | |
4333 | ||
ac27a0ec | 4334 | /* |
617ba13b | 4335 | * ext4_truncate() |
ac27a0ec | 4336 | * |
617ba13b MC |
4337 | * We block out ext4_get_block() block instantiations across the entire |
4338 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4339 | * simultaneously on behalf of the same inode. |
4340 | * | |
4341 | * As we work through the truncate and commmit bits of it to the journal there | |
4342 | * is one core, guiding principle: the file's tree must always be consistent on | |
4343 | * disk. We must be able to restart the truncate after a crash. | |
4344 | * | |
4345 | * The file's tree may be transiently inconsistent in memory (although it | |
4346 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4347 | * the contents of (the filesystem + the journal) must be consistent and | |
4348 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4349 | * left-to-right works OK too). | |
4350 | * | |
4351 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4352 | * truncate against the orphan inode list. | |
4353 | * | |
4354 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4355 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4356 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4357 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4358 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4359 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4360 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4361 | */ |
617ba13b | 4362 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
4363 | { |
4364 | handle_t *handle; | |
617ba13b | 4365 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 4366 | __le32 *i_data = ei->i_data; |
617ba13b | 4367 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 4368 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 4369 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
4370 | Indirect chain[4]; |
4371 | Indirect *partial; | |
4372 | __le32 nr = 0; | |
4373 | int n; | |
725d26d3 | 4374 | ext4_lblk_t last_block; |
ac27a0ec | 4375 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 4376 | |
91ef4caf | 4377 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4378 | return; |
4379 | ||
5534fb5b | 4380 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
7d8f9f7d TT |
4381 | ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE; |
4382 | ||
1d03ec98 | 4383 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
cf108bca | 4384 | ext4_ext_truncate(inode); |
1d03ec98 AK |
4385 | return; |
4386 | } | |
a86c6181 | 4387 | |
ac27a0ec | 4388 | handle = start_transaction(inode); |
cf108bca | 4389 | if (IS_ERR(handle)) |
ac27a0ec | 4390 | return; /* AKPM: return what? */ |
ac27a0ec DK |
4391 | |
4392 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 4393 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec | 4394 | |
cf108bca JK |
4395 | if (inode->i_size & (blocksize - 1)) |
4396 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
4397 | goto out_stop; | |
ac27a0ec | 4398 | |
617ba13b | 4399 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
4400 | if (n == 0) |
4401 | goto out_stop; /* error */ | |
4402 | ||
4403 | /* | |
4404 | * OK. This truncate is going to happen. We add the inode to the | |
4405 | * orphan list, so that if this truncate spans multiple transactions, | |
4406 | * and we crash, we will resume the truncate when the filesystem | |
4407 | * recovers. It also marks the inode dirty, to catch the new size. | |
4408 | * | |
4409 | * Implication: the file must always be in a sane, consistent | |
4410 | * truncatable state while each transaction commits. | |
4411 | */ | |
617ba13b | 4412 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
4413 | goto out_stop; |
4414 | ||
632eaeab MC |
4415 | /* |
4416 | * From here we block out all ext4_get_block() callers who want to | |
4417 | * modify the block allocation tree. | |
4418 | */ | |
4419 | down_write(&ei->i_data_sem); | |
b4df2030 | 4420 | |
c2ea3fde | 4421 | ext4_discard_preallocations(inode); |
b4df2030 | 4422 | |
ac27a0ec DK |
4423 | /* |
4424 | * The orphan list entry will now protect us from any crash which | |
4425 | * occurs before the truncate completes, so it is now safe to propagate | |
4426 | * the new, shorter inode size (held for now in i_size) into the | |
4427 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 4428 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
4429 | */ |
4430 | ei->i_disksize = inode->i_size; | |
4431 | ||
ac27a0ec | 4432 | if (n == 1) { /* direct blocks */ |
617ba13b MC |
4433 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
4434 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
4435 | goto do_indirects; |
4436 | } | |
4437 | ||
617ba13b | 4438 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
4439 | /* Kill the top of shared branch (not detached) */ |
4440 | if (nr) { | |
4441 | if (partial == chain) { | |
4442 | /* Shared branch grows from the inode */ | |
617ba13b | 4443 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
4444 | &nr, &nr+1, (chain+n-1) - partial); |
4445 | *partial->p = 0; | |
4446 | /* | |
4447 | * We mark the inode dirty prior to restart, | |
4448 | * and prior to stop. No need for it here. | |
4449 | */ | |
4450 | } else { | |
4451 | /* Shared branch grows from an indirect block */ | |
4452 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 4453 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
4454 | partial->p, |
4455 | partial->p+1, (chain+n-1) - partial); | |
4456 | } | |
4457 | } | |
4458 | /* Clear the ends of indirect blocks on the shared branch */ | |
4459 | while (partial > chain) { | |
617ba13b | 4460 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
4461 | (__le32*)partial->bh->b_data+addr_per_block, |
4462 | (chain+n-1) - partial); | |
4463 | BUFFER_TRACE(partial->bh, "call brelse"); | |
de9a55b8 | 4464 | brelse(partial->bh); |
ac27a0ec DK |
4465 | partial--; |
4466 | } | |
4467 | do_indirects: | |
4468 | /* Kill the remaining (whole) subtrees */ | |
4469 | switch (offsets[0]) { | |
4470 | default: | |
617ba13b | 4471 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 4472 | if (nr) { |
617ba13b MC |
4473 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
4474 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 4475 | } |
617ba13b MC |
4476 | case EXT4_IND_BLOCK: |
4477 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 4478 | if (nr) { |
617ba13b MC |
4479 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
4480 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 4481 | } |
617ba13b MC |
4482 | case EXT4_DIND_BLOCK: |
4483 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 4484 | if (nr) { |
617ba13b MC |
4485 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
4486 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 4487 | } |
617ba13b | 4488 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
4489 | ; |
4490 | } | |
4491 | ||
0e855ac8 | 4492 | up_write(&ei->i_data_sem); |
ef7f3835 | 4493 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 4494 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
4495 | |
4496 | /* | |
4497 | * In a multi-transaction truncate, we only make the final transaction | |
4498 | * synchronous | |
4499 | */ | |
4500 | if (IS_SYNC(inode)) | |
0390131b | 4501 | ext4_handle_sync(handle); |
ac27a0ec DK |
4502 | out_stop: |
4503 | /* | |
4504 | * If this was a simple ftruncate(), and the file will remain alive | |
4505 | * then we need to clear up the orphan record which we created above. | |
4506 | * However, if this was a real unlink then we were called by | |
617ba13b | 4507 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
4508 | * orphan info for us. |
4509 | */ | |
4510 | if (inode->i_nlink) | |
617ba13b | 4511 | ext4_orphan_del(handle, inode); |
ac27a0ec | 4512 | |
617ba13b | 4513 | ext4_journal_stop(handle); |
ac27a0ec DK |
4514 | } |
4515 | ||
ac27a0ec | 4516 | /* |
617ba13b | 4517 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4518 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4519 | * data in memory that is needed to recreate the on-disk version of this | |
4520 | * inode. | |
4521 | */ | |
617ba13b MC |
4522 | static int __ext4_get_inode_loc(struct inode *inode, |
4523 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4524 | { |
240799cd TT |
4525 | struct ext4_group_desc *gdp; |
4526 | struct buffer_head *bh; | |
4527 | struct super_block *sb = inode->i_sb; | |
4528 | ext4_fsblk_t block; | |
4529 | int inodes_per_block, inode_offset; | |
4530 | ||
3a06d778 | 4531 | iloc->bh = NULL; |
240799cd TT |
4532 | if (!ext4_valid_inum(sb, inode->i_ino)) |
4533 | return -EIO; | |
ac27a0ec | 4534 | |
240799cd TT |
4535 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4536 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4537 | if (!gdp) | |
ac27a0ec DK |
4538 | return -EIO; |
4539 | ||
240799cd TT |
4540 | /* |
4541 | * Figure out the offset within the block group inode table | |
4542 | */ | |
4543 | inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb)); | |
4544 | inode_offset = ((inode->i_ino - 1) % | |
4545 | EXT4_INODES_PER_GROUP(sb)); | |
4546 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4547 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4548 | ||
4549 | bh = sb_getblk(sb, block); | |
ac27a0ec | 4550 | if (!bh) { |
240799cd TT |
4551 | ext4_error(sb, "ext4_get_inode_loc", "unable to read " |
4552 | "inode block - inode=%lu, block=%llu", | |
4553 | inode->i_ino, block); | |
ac27a0ec DK |
4554 | return -EIO; |
4555 | } | |
4556 | if (!buffer_uptodate(bh)) { | |
4557 | lock_buffer(bh); | |
9c83a923 HK |
4558 | |
4559 | /* | |
4560 | * If the buffer has the write error flag, we have failed | |
4561 | * to write out another inode in the same block. In this | |
4562 | * case, we don't have to read the block because we may | |
4563 | * read the old inode data successfully. | |
4564 | */ | |
4565 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4566 | set_buffer_uptodate(bh); | |
4567 | ||
ac27a0ec DK |
4568 | if (buffer_uptodate(bh)) { |
4569 | /* someone brought it uptodate while we waited */ | |
4570 | unlock_buffer(bh); | |
4571 | goto has_buffer; | |
4572 | } | |
4573 | ||
4574 | /* | |
4575 | * If we have all information of the inode in memory and this | |
4576 | * is the only valid inode in the block, we need not read the | |
4577 | * block. | |
4578 | */ | |
4579 | if (in_mem) { | |
4580 | struct buffer_head *bitmap_bh; | |
240799cd | 4581 | int i, start; |
ac27a0ec | 4582 | |
240799cd | 4583 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4584 | |
240799cd TT |
4585 | /* Is the inode bitmap in cache? */ |
4586 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
4587 | if (!bitmap_bh) |
4588 | goto make_io; | |
4589 | ||
4590 | /* | |
4591 | * If the inode bitmap isn't in cache then the | |
4592 | * optimisation may end up performing two reads instead | |
4593 | * of one, so skip it. | |
4594 | */ | |
4595 | if (!buffer_uptodate(bitmap_bh)) { | |
4596 | brelse(bitmap_bh); | |
4597 | goto make_io; | |
4598 | } | |
240799cd | 4599 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4600 | if (i == inode_offset) |
4601 | continue; | |
617ba13b | 4602 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4603 | break; |
4604 | } | |
4605 | brelse(bitmap_bh); | |
240799cd | 4606 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4607 | /* all other inodes are free, so skip I/O */ |
4608 | memset(bh->b_data, 0, bh->b_size); | |
4609 | set_buffer_uptodate(bh); | |
4610 | unlock_buffer(bh); | |
4611 | goto has_buffer; | |
4612 | } | |
4613 | } | |
4614 | ||
4615 | make_io: | |
240799cd TT |
4616 | /* |
4617 | * If we need to do any I/O, try to pre-readahead extra | |
4618 | * blocks from the inode table. | |
4619 | */ | |
4620 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4621 | ext4_fsblk_t b, end, table; | |
4622 | unsigned num; | |
4623 | ||
4624 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4625 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
4626 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
4627 | if (table > b) | |
4628 | b = table; | |
4629 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
4630 | num = EXT4_INODES_PER_GROUP(sb); | |
4631 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4632 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 4633 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4634 | table += num / inodes_per_block; |
4635 | if (end > table) | |
4636 | end = table; | |
4637 | while (b <= end) | |
4638 | sb_breadahead(sb, b++); | |
4639 | } | |
4640 | ||
ac27a0ec DK |
4641 | /* |
4642 | * There are other valid inodes in the buffer, this inode | |
4643 | * has in-inode xattrs, or we don't have this inode in memory. | |
4644 | * Read the block from disk. | |
4645 | */ | |
4646 | get_bh(bh); | |
4647 | bh->b_end_io = end_buffer_read_sync; | |
4648 | submit_bh(READ_META, bh); | |
4649 | wait_on_buffer(bh); | |
4650 | if (!buffer_uptodate(bh)) { | |
240799cd TT |
4651 | ext4_error(sb, __func__, |
4652 | "unable to read inode block - inode=%lu, " | |
4653 | "block=%llu", inode->i_ino, block); | |
ac27a0ec DK |
4654 | brelse(bh); |
4655 | return -EIO; | |
4656 | } | |
4657 | } | |
4658 | has_buffer: | |
4659 | iloc->bh = bh; | |
4660 | return 0; | |
4661 | } | |
4662 | ||
617ba13b | 4663 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4664 | { |
4665 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b MC |
4666 | return __ext4_get_inode_loc(inode, iloc, |
4667 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); | |
ac27a0ec DK |
4668 | } |
4669 | ||
617ba13b | 4670 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4671 | { |
617ba13b | 4672 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4673 | |
4674 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4675 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4676 | inode->i_flags |= S_SYNC; |
617ba13b | 4677 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4678 | inode->i_flags |= S_APPEND; |
617ba13b | 4679 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4680 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4681 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4682 | inode->i_flags |= S_NOATIME; |
617ba13b | 4683 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4684 | inode->i_flags |= S_DIRSYNC; |
4685 | } | |
4686 | ||
ff9ddf7e JK |
4687 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4688 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4689 | { | |
4690 | unsigned int flags = ei->vfs_inode.i_flags; | |
4691 | ||
4692 | ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4693 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL); | |
4694 | if (flags & S_SYNC) | |
4695 | ei->i_flags |= EXT4_SYNC_FL; | |
4696 | if (flags & S_APPEND) | |
4697 | ei->i_flags |= EXT4_APPEND_FL; | |
4698 | if (flags & S_IMMUTABLE) | |
4699 | ei->i_flags |= EXT4_IMMUTABLE_FL; | |
4700 | if (flags & S_NOATIME) | |
4701 | ei->i_flags |= EXT4_NOATIME_FL; | |
4702 | if (flags & S_DIRSYNC) | |
4703 | ei->i_flags |= EXT4_DIRSYNC_FL; | |
4704 | } | |
de9a55b8 | 4705 | |
0fc1b451 | 4706 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4707 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4708 | { |
4709 | blkcnt_t i_blocks ; | |
8180a562 AK |
4710 | struct inode *inode = &(ei->vfs_inode); |
4711 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4712 | |
4713 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4714 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4715 | /* we are using combined 48 bit field */ | |
4716 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4717 | le32_to_cpu(raw_inode->i_blocks_lo); | |
8180a562 AK |
4718 | if (ei->i_flags & EXT4_HUGE_FILE_FL) { |
4719 | /* i_blocks represent file system block size */ | |
4720 | return i_blocks << (inode->i_blkbits - 9); | |
4721 | } else { | |
4722 | return i_blocks; | |
4723 | } | |
0fc1b451 AK |
4724 | } else { |
4725 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4726 | } | |
4727 | } | |
ff9ddf7e | 4728 | |
1d1fe1ee | 4729 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4730 | { |
617ba13b MC |
4731 | struct ext4_iloc iloc; |
4732 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4733 | struct ext4_inode_info *ei; |
1d1fe1ee DH |
4734 | struct inode *inode; |
4735 | long ret; | |
ac27a0ec DK |
4736 | int block; |
4737 | ||
1d1fe1ee DH |
4738 | inode = iget_locked(sb, ino); |
4739 | if (!inode) | |
4740 | return ERR_PTR(-ENOMEM); | |
4741 | if (!(inode->i_state & I_NEW)) | |
4742 | return inode; | |
4743 | ||
4744 | ei = EXT4_I(inode); | |
567f3e9a | 4745 | iloc.bh = 0; |
ac27a0ec | 4746 | |
1d1fe1ee DH |
4747 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4748 | if (ret < 0) | |
ac27a0ec | 4749 | goto bad_inode; |
617ba13b | 4750 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
4751 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
4752 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
4753 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4754 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4755 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4756 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
4757 | } | |
4758 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec DK |
4759 | |
4760 | ei->i_state = 0; | |
4761 | ei->i_dir_start_lookup = 0; | |
4762 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4763 | /* We now have enough fields to check if the inode was active or not. | |
4764 | * This is needed because nfsd might try to access dead inodes | |
4765 | * the test is that same one that e2fsck uses | |
4766 | * NeilBrown 1999oct15 | |
4767 | */ | |
4768 | if (inode->i_nlink == 0) { | |
4769 | if (inode->i_mode == 0 || | |
617ba13b | 4770 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 4771 | /* this inode is deleted */ |
1d1fe1ee | 4772 | ret = -ESTALE; |
ac27a0ec DK |
4773 | goto bad_inode; |
4774 | } | |
4775 | /* The only unlinked inodes we let through here have | |
4776 | * valid i_mode and are being read by the orphan | |
4777 | * recovery code: that's fine, we're about to complete | |
4778 | * the process of deleting those. */ | |
4779 | } | |
ac27a0ec | 4780 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4781 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4782 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4783 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4784 | ei->i_file_acl |= |
4785 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4786 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec DK |
4787 | ei->i_disksize = inode->i_size; |
4788 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
4789 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4790 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4791 | /* |
4792 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4793 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4794 | */ | |
617ba13b | 4795 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4796 | ei->i_data[block] = raw_inode->i_block[block]; |
4797 | INIT_LIST_HEAD(&ei->i_orphan); | |
4798 | ||
0040d987 | 4799 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 4800 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 4801 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 4802 | EXT4_INODE_SIZE(inode->i_sb)) { |
1d1fe1ee | 4803 | ret = -EIO; |
ac27a0ec | 4804 | goto bad_inode; |
e5d2861f | 4805 | } |
ac27a0ec DK |
4806 | if (ei->i_extra_isize == 0) { |
4807 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4808 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4809 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
4810 | } else { |
4811 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 4812 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 4813 | ei->i_extra_isize; |
617ba13b | 4814 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
de9a55b8 | 4815 | ei->i_state |= EXT4_STATE_XATTR; |
ac27a0ec DK |
4816 | } |
4817 | } else | |
4818 | ei->i_extra_isize = 0; | |
4819 | ||
ef7f3835 KS |
4820 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4821 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4822 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4823 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4824 | ||
25ec56b5 JNC |
4825 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4826 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4827 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4828 | inode->i_version |= | |
4829 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4830 | } | |
4831 | ||
c4b5a614 | 4832 | ret = 0; |
485c26ec | 4833 | if (ei->i_file_acl && |
1032988c | 4834 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
485c26ec TT |
4835 | ext4_error(sb, __func__, |
4836 | "bad extended attribute block %llu in inode #%lu", | |
4837 | ei->i_file_acl, inode->i_ino); | |
4838 | ret = -EIO; | |
4839 | goto bad_inode; | |
4840 | } else if (ei->i_flags & EXT4_EXTENTS_FL) { | |
c4b5a614 TT |
4841 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
4842 | (S_ISLNK(inode->i_mode) && | |
4843 | !ext4_inode_is_fast_symlink(inode))) | |
4844 | /* Validate extent which is part of inode */ | |
4845 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 4846 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
4847 | (S_ISLNK(inode->i_mode) && |
4848 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 4849 | /* Validate block references which are part of inode */ |
fe2c8191 TN |
4850 | ret = ext4_check_inode_blockref(inode); |
4851 | } | |
567f3e9a | 4852 | if (ret) |
de9a55b8 | 4853 | goto bad_inode; |
7a262f7c | 4854 | |
ac27a0ec | 4855 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4856 | inode->i_op = &ext4_file_inode_operations; |
4857 | inode->i_fop = &ext4_file_operations; | |
4858 | ext4_set_aops(inode); | |
ac27a0ec | 4859 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4860 | inode->i_op = &ext4_dir_inode_operations; |
4861 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4862 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4863 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4864 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4865 | nd_terminate_link(ei->i_data, inode->i_size, |
4866 | sizeof(ei->i_data) - 1); | |
4867 | } else { | |
617ba13b MC |
4868 | inode->i_op = &ext4_symlink_inode_operations; |
4869 | ext4_set_aops(inode); | |
ac27a0ec | 4870 | } |
563bdd61 TT |
4871 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4872 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4873 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4874 | if (raw_inode->i_block[0]) |
4875 | init_special_inode(inode, inode->i_mode, | |
4876 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4877 | else | |
4878 | init_special_inode(inode, inode->i_mode, | |
4879 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 4880 | } else { |
563bdd61 | 4881 | ret = -EIO; |
de9a55b8 | 4882 | ext4_error(inode->i_sb, __func__, |
563bdd61 TT |
4883 | "bogus i_mode (%o) for inode=%lu", |
4884 | inode->i_mode, inode->i_ino); | |
4885 | goto bad_inode; | |
ac27a0ec | 4886 | } |
af5bc92d | 4887 | brelse(iloc.bh); |
617ba13b | 4888 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4889 | unlock_new_inode(inode); |
4890 | return inode; | |
ac27a0ec DK |
4891 | |
4892 | bad_inode: | |
567f3e9a | 4893 | brelse(iloc.bh); |
1d1fe1ee DH |
4894 | iget_failed(inode); |
4895 | return ERR_PTR(ret); | |
ac27a0ec DK |
4896 | } |
4897 | ||
0fc1b451 AK |
4898 | static int ext4_inode_blocks_set(handle_t *handle, |
4899 | struct ext4_inode *raw_inode, | |
4900 | struct ext4_inode_info *ei) | |
4901 | { | |
4902 | struct inode *inode = &(ei->vfs_inode); | |
4903 | u64 i_blocks = inode->i_blocks; | |
4904 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4905 | |
4906 | if (i_blocks <= ~0U) { | |
4907 | /* | |
4908 | * i_blocks can be represnted in a 32 bit variable | |
4909 | * as multiple of 512 bytes | |
4910 | */ | |
8180a562 | 4911 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4912 | raw_inode->i_blocks_high = 0; |
8180a562 | 4913 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
f287a1a5 TT |
4914 | return 0; |
4915 | } | |
4916 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4917 | return -EFBIG; | |
4918 | ||
4919 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4920 | /* |
4921 | * i_blocks can be represented in a 48 bit variable | |
4922 | * as multiple of 512 bytes | |
4923 | */ | |
8180a562 | 4924 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4925 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
8180a562 | 4926 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
0fc1b451 | 4927 | } else { |
8180a562 AK |
4928 | ei->i_flags |= EXT4_HUGE_FILE_FL; |
4929 | /* i_block is stored in file system block size */ | |
4930 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4931 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4932 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4933 | } |
f287a1a5 | 4934 | return 0; |
0fc1b451 AK |
4935 | } |
4936 | ||
ac27a0ec DK |
4937 | /* |
4938 | * Post the struct inode info into an on-disk inode location in the | |
4939 | * buffer-cache. This gobbles the caller's reference to the | |
4940 | * buffer_head in the inode location struct. | |
4941 | * | |
4942 | * The caller must have write access to iloc->bh. | |
4943 | */ | |
617ba13b | 4944 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4945 | struct inode *inode, |
830156c7 | 4946 | struct ext4_iloc *iloc) |
ac27a0ec | 4947 | { |
617ba13b MC |
4948 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4949 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4950 | struct buffer_head *bh = iloc->bh; |
4951 | int err = 0, rc, block; | |
4952 | ||
4953 | /* For fields not not tracking in the in-memory inode, | |
4954 | * initialise them to zero for new inodes. */ | |
617ba13b MC |
4955 | if (ei->i_state & EXT4_STATE_NEW) |
4956 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
ac27a0ec | 4957 | |
ff9ddf7e | 4958 | ext4_get_inode_flags(ei); |
ac27a0ec | 4959 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 4960 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4961 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
4962 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
4963 | /* | |
4964 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4965 | * re-used with the upper 16 bits of the uid/gid intact | |
4966 | */ | |
af5bc92d | 4967 | if (!ei->i_dtime) { |
ac27a0ec DK |
4968 | raw_inode->i_uid_high = |
4969 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
4970 | raw_inode->i_gid_high = | |
4971 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
4972 | } else { | |
4973 | raw_inode->i_uid_high = 0; | |
4974 | raw_inode->i_gid_high = 0; | |
4975 | } | |
4976 | } else { | |
4977 | raw_inode->i_uid_low = | |
4978 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
4979 | raw_inode->i_gid_low = | |
4980 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
4981 | raw_inode->i_uid_high = 0; | |
4982 | raw_inode->i_gid_high = 0; | |
4983 | } | |
4984 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4985 | |
4986 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4987 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4988 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4989 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4990 | ||
0fc1b451 AK |
4991 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4992 | goto out_brelse; | |
ac27a0ec | 4993 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
1b9c12f4 | 4994 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); |
9b8f1f01 MC |
4995 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4996 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4997 | raw_inode->i_file_acl_high = |
4998 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4999 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
5000 | ext4_isize_set(raw_inode, ei->i_disksize); |
5001 | if (ei->i_disksize > 0x7fffffffULL) { | |
5002 | struct super_block *sb = inode->i_sb; | |
5003 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
5004 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
5005 | EXT4_SB(sb)->s_es->s_rev_level == | |
5006 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
5007 | /* If this is the first large file | |
5008 | * created, add a flag to the superblock. | |
5009 | */ | |
5010 | err = ext4_journal_get_write_access(handle, | |
5011 | EXT4_SB(sb)->s_sbh); | |
5012 | if (err) | |
5013 | goto out_brelse; | |
5014 | ext4_update_dynamic_rev(sb); | |
5015 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 5016 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 5017 | sb->s_dirt = 1; |
0390131b FM |
5018 | ext4_handle_sync(handle); |
5019 | err = ext4_handle_dirty_metadata(handle, inode, | |
a48380f7 | 5020 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
5021 | } |
5022 | } | |
5023 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5024 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5025 | if (old_valid_dev(inode->i_rdev)) { | |
5026 | raw_inode->i_block[0] = | |
5027 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5028 | raw_inode->i_block[1] = 0; | |
5029 | } else { | |
5030 | raw_inode->i_block[0] = 0; | |
5031 | raw_inode->i_block[1] = | |
5032 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5033 | raw_inode->i_block[2] = 0; | |
5034 | } | |
de9a55b8 TT |
5035 | } else |
5036 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
5037 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 5038 | |
25ec56b5 JNC |
5039 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
5040 | if (ei->i_extra_isize) { | |
5041 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5042 | raw_inode->i_version_hi = | |
5043 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 5044 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
5045 | } |
5046 | ||
830156c7 FM |
5047 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
5048 | rc = ext4_handle_dirty_metadata(handle, inode, bh); | |
5049 | if (!err) | |
5050 | err = rc; | |
617ba13b | 5051 | ei->i_state &= ~EXT4_STATE_NEW; |
ac27a0ec DK |
5052 | |
5053 | out_brelse: | |
af5bc92d | 5054 | brelse(bh); |
617ba13b | 5055 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5056 | return err; |
5057 | } | |
5058 | ||
5059 | /* | |
617ba13b | 5060 | * ext4_write_inode() |
ac27a0ec DK |
5061 | * |
5062 | * We are called from a few places: | |
5063 | * | |
5064 | * - Within generic_file_write() for O_SYNC files. | |
5065 | * Here, there will be no transaction running. We wait for any running | |
5066 | * trasnaction to commit. | |
5067 | * | |
5068 | * - Within sys_sync(), kupdate and such. | |
5069 | * We wait on commit, if tol to. | |
5070 | * | |
5071 | * - Within prune_icache() (PF_MEMALLOC == true) | |
5072 | * Here we simply return. We can't afford to block kswapd on the | |
5073 | * journal commit. | |
5074 | * | |
5075 | * In all cases it is actually safe for us to return without doing anything, | |
5076 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 5077 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
5078 | * knfsd. |
5079 | * | |
5080 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5081 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5082 | * which we are interested. | |
5083 | * | |
5084 | * It would be a bug for them to not do this. The code: | |
5085 | * | |
5086 | * mark_inode_dirty(inode) | |
5087 | * stuff(); | |
5088 | * inode->i_size = expr; | |
5089 | * | |
5090 | * is in error because a kswapd-driven write_inode() could occur while | |
5091 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
5092 | * will no longer be on the superblock's dirty inode list. | |
5093 | */ | |
617ba13b | 5094 | int ext4_write_inode(struct inode *inode, int wait) |
ac27a0ec | 5095 | { |
91ac6f43 FM |
5096 | int err; |
5097 | ||
ac27a0ec DK |
5098 | if (current->flags & PF_MEMALLOC) |
5099 | return 0; | |
5100 | ||
91ac6f43 FM |
5101 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5102 | if (ext4_journal_current_handle()) { | |
5103 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5104 | dump_stack(); | |
5105 | return -EIO; | |
5106 | } | |
ac27a0ec | 5107 | |
91ac6f43 FM |
5108 | if (!wait) |
5109 | return 0; | |
5110 | ||
5111 | err = ext4_force_commit(inode->i_sb); | |
5112 | } else { | |
5113 | struct ext4_iloc iloc; | |
ac27a0ec | 5114 | |
91ac6f43 FM |
5115 | err = ext4_get_inode_loc(inode, &iloc); |
5116 | if (err) | |
5117 | return err; | |
830156c7 FM |
5118 | if (wait) |
5119 | sync_dirty_buffer(iloc.bh); | |
5120 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
5121 | ext4_error(inode->i_sb, __func__, | |
5122 | "IO error syncing inode, " | |
5123 | "inode=%lu, block=%llu", | |
5124 | inode->i_ino, | |
5125 | (unsigned long long)iloc.bh->b_blocknr); | |
5126 | err = -EIO; | |
5127 | } | |
91ac6f43 FM |
5128 | } |
5129 | return err; | |
ac27a0ec DK |
5130 | } |
5131 | ||
5132 | /* | |
617ba13b | 5133 | * ext4_setattr() |
ac27a0ec DK |
5134 | * |
5135 | * Called from notify_change. | |
5136 | * | |
5137 | * We want to trap VFS attempts to truncate the file as soon as | |
5138 | * possible. In particular, we want to make sure that when the VFS | |
5139 | * shrinks i_size, we put the inode on the orphan list and modify | |
5140 | * i_disksize immediately, so that during the subsequent flushing of | |
5141 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5142 | * commit will leave the blocks being flushed in an unused state on | |
5143 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5144 | * be freed, so we have a strong guarantee that no future commit will | |
5145 | * leave these blocks visible to the user.) | |
5146 | * | |
678aaf48 JK |
5147 | * Another thing we have to assure is that if we are in ordered mode |
5148 | * and inode is still attached to the committing transaction, we must | |
5149 | * we start writeout of all the dirty pages which are being truncated. | |
5150 | * This way we are sure that all the data written in the previous | |
5151 | * transaction are already on disk (truncate waits for pages under | |
5152 | * writeback). | |
5153 | * | |
5154 | * Called with inode->i_mutex down. | |
ac27a0ec | 5155 | */ |
617ba13b | 5156 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
5157 | { |
5158 | struct inode *inode = dentry->d_inode; | |
5159 | int error, rc = 0; | |
5160 | const unsigned int ia_valid = attr->ia_valid; | |
5161 | ||
5162 | error = inode_change_ok(inode, attr); | |
5163 | if (error) | |
5164 | return error; | |
5165 | ||
5166 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
5167 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
5168 | handle_t *handle; | |
5169 | ||
5170 | /* (user+group)*(old+new) structure, inode write (sb, | |
5171 | * inode block, ? - but truncate inode update has it) */ | |
617ba13b MC |
5172 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
5173 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
ac27a0ec DK |
5174 | if (IS_ERR(handle)) { |
5175 | error = PTR_ERR(handle); | |
5176 | goto err_out; | |
5177 | } | |
a269eb18 | 5178 | error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; |
ac27a0ec | 5179 | if (error) { |
617ba13b | 5180 | ext4_journal_stop(handle); |
ac27a0ec DK |
5181 | return error; |
5182 | } | |
5183 | /* Update corresponding info in inode so that everything is in | |
5184 | * one transaction */ | |
5185 | if (attr->ia_valid & ATTR_UID) | |
5186 | inode->i_uid = attr->ia_uid; | |
5187 | if (attr->ia_valid & ATTR_GID) | |
5188 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5189 | error = ext4_mark_inode_dirty(handle, inode); |
5190 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5191 | } |
5192 | ||
e2b46574 ES |
5193 | if (attr->ia_valid & ATTR_SIZE) { |
5194 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { | |
5195 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
5196 | ||
5197 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { | |
5198 | error = -EFBIG; | |
5199 | goto err_out; | |
5200 | } | |
5201 | } | |
5202 | } | |
5203 | ||
ac27a0ec DK |
5204 | if (S_ISREG(inode->i_mode) && |
5205 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
5206 | handle_t *handle; | |
5207 | ||
617ba13b | 5208 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
5209 | if (IS_ERR(handle)) { |
5210 | error = PTR_ERR(handle); | |
5211 | goto err_out; | |
5212 | } | |
5213 | ||
617ba13b MC |
5214 | error = ext4_orphan_add(handle, inode); |
5215 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
5216 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
5217 | if (!error) |
5218 | error = rc; | |
617ba13b | 5219 | ext4_journal_stop(handle); |
678aaf48 JK |
5220 | |
5221 | if (ext4_should_order_data(inode)) { | |
5222 | error = ext4_begin_ordered_truncate(inode, | |
5223 | attr->ia_size); | |
5224 | if (error) { | |
5225 | /* Do as much error cleanup as possible */ | |
5226 | handle = ext4_journal_start(inode, 3); | |
5227 | if (IS_ERR(handle)) { | |
5228 | ext4_orphan_del(NULL, inode); | |
5229 | goto err_out; | |
5230 | } | |
5231 | ext4_orphan_del(handle, inode); | |
5232 | ext4_journal_stop(handle); | |
5233 | goto err_out; | |
5234 | } | |
5235 | } | |
ac27a0ec DK |
5236 | } |
5237 | ||
5238 | rc = inode_setattr(inode, attr); | |
5239 | ||
617ba13b | 5240 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
5241 | * transaction handle at all, we need to clean up the in-core |
5242 | * orphan list manually. */ | |
5243 | if (inode->i_nlink) | |
617ba13b | 5244 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5245 | |
5246 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 5247 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
5248 | |
5249 | err_out: | |
617ba13b | 5250 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5251 | if (!error) |
5252 | error = rc; | |
5253 | return error; | |
5254 | } | |
5255 | ||
3e3398a0 MC |
5256 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5257 | struct kstat *stat) | |
5258 | { | |
5259 | struct inode *inode; | |
5260 | unsigned long delalloc_blocks; | |
5261 | ||
5262 | inode = dentry->d_inode; | |
5263 | generic_fillattr(inode, stat); | |
5264 | ||
5265 | /* | |
5266 | * We can't update i_blocks if the block allocation is delayed | |
5267 | * otherwise in the case of system crash before the real block | |
5268 | * allocation is done, we will have i_blocks inconsistent with | |
5269 | * on-disk file blocks. | |
5270 | * We always keep i_blocks updated together with real | |
5271 | * allocation. But to not confuse with user, stat | |
5272 | * will return the blocks that include the delayed allocation | |
5273 | * blocks for this file. | |
5274 | */ | |
5275 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
5276 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
5277 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
5278 | ||
5279 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
5280 | return 0; | |
5281 | } | |
ac27a0ec | 5282 | |
a02908f1 MC |
5283 | static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks, |
5284 | int chunk) | |
5285 | { | |
5286 | int indirects; | |
5287 | ||
5288 | /* if nrblocks are contiguous */ | |
5289 | if (chunk) { | |
5290 | /* | |
5291 | * With N contiguous data blocks, it need at most | |
5292 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks | |
5293 | * 2 dindirect blocks | |
5294 | * 1 tindirect block | |
5295 | */ | |
5296 | indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
5297 | return indirects + 3; | |
5298 | } | |
5299 | /* | |
5300 | * if nrblocks are not contiguous, worse case, each block touch | |
5301 | * a indirect block, and each indirect block touch a double indirect | |
5302 | * block, plus a triple indirect block | |
5303 | */ | |
5304 | indirects = nrblocks * 2 + 1; | |
5305 | return indirects; | |
5306 | } | |
5307 | ||
5308 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5309 | { | |
5310 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
ac51d837 TT |
5311 | return ext4_indirect_trans_blocks(inode, nrblocks, chunk); |
5312 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); | |
a02908f1 | 5313 | } |
ac51d837 | 5314 | |
ac27a0ec | 5315 | /* |
a02908f1 MC |
5316 | * Account for index blocks, block groups bitmaps and block group |
5317 | * descriptor blocks if modify datablocks and index blocks | |
5318 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5319 | * |
a02908f1 MC |
5320 | * If datablocks are discontiguous, they are possible to spread over |
5321 | * different block groups too. If they are contiugous, with flexbg, | |
5322 | * they could still across block group boundary. | |
ac27a0ec | 5323 | * |
a02908f1 MC |
5324 | * Also account for superblock, inode, quota and xattr blocks |
5325 | */ | |
5326 | int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5327 | { | |
8df9675f TT |
5328 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5329 | int gdpblocks; | |
a02908f1 MC |
5330 | int idxblocks; |
5331 | int ret = 0; | |
5332 | ||
5333 | /* | |
5334 | * How many index blocks need to touch to modify nrblocks? | |
5335 | * The "Chunk" flag indicating whether the nrblocks is | |
5336 | * physically contiguous on disk | |
5337 | * | |
5338 | * For Direct IO and fallocate, they calls get_block to allocate | |
5339 | * one single extent at a time, so they could set the "Chunk" flag | |
5340 | */ | |
5341 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
5342 | ||
5343 | ret = idxblocks; | |
5344 | ||
5345 | /* | |
5346 | * Now let's see how many group bitmaps and group descriptors need | |
5347 | * to account | |
5348 | */ | |
5349 | groups = idxblocks; | |
5350 | if (chunk) | |
5351 | groups += 1; | |
5352 | else | |
5353 | groups += nrblocks; | |
5354 | ||
5355 | gdpblocks = groups; | |
8df9675f TT |
5356 | if (groups > ngroups) |
5357 | groups = ngroups; | |
a02908f1 MC |
5358 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5359 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5360 | ||
5361 | /* bitmaps and block group descriptor blocks */ | |
5362 | ret += groups + gdpblocks; | |
5363 | ||
5364 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5365 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5366 | ||
5367 | return ret; | |
5368 | } | |
5369 | ||
5370 | /* | |
5371 | * Calulate the total number of credits to reserve to fit | |
f3bd1f3f MC |
5372 | * the modification of a single pages into a single transaction, |
5373 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5374 | * |
525f4ed8 | 5375 | * This could be called via ext4_write_begin() |
ac27a0ec | 5376 | * |
525f4ed8 | 5377 | * We need to consider the worse case, when |
a02908f1 | 5378 | * one new block per extent. |
ac27a0ec | 5379 | */ |
a86c6181 | 5380 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5381 | { |
617ba13b | 5382 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5383 | int ret; |
5384 | ||
a02908f1 | 5385 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 5386 | |
a02908f1 | 5387 | /* Account for data blocks for journalled mode */ |
617ba13b | 5388 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5389 | ret += bpp; |
ac27a0ec DK |
5390 | return ret; |
5391 | } | |
f3bd1f3f MC |
5392 | |
5393 | /* | |
5394 | * Calculate the journal credits for a chunk of data modification. | |
5395 | * | |
5396 | * This is called from DIO, fallocate or whoever calling | |
12b7ac17 | 5397 | * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks. |
f3bd1f3f MC |
5398 | * |
5399 | * journal buffers for data blocks are not included here, as DIO | |
5400 | * and fallocate do no need to journal data buffers. | |
5401 | */ | |
5402 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5403 | { | |
5404 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5405 | } | |
5406 | ||
ac27a0ec | 5407 | /* |
617ba13b | 5408 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5409 | * Give this, we know that the caller already has write access to iloc->bh. |
5410 | */ | |
617ba13b | 5411 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5412 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5413 | { |
5414 | int err = 0; | |
5415 | ||
25ec56b5 JNC |
5416 | if (test_opt(inode->i_sb, I_VERSION)) |
5417 | inode_inc_iversion(inode); | |
5418 | ||
ac27a0ec DK |
5419 | /* the do_update_inode consumes one bh->b_count */ |
5420 | get_bh(iloc->bh); | |
5421 | ||
dab291af | 5422 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5423 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5424 | put_bh(iloc->bh); |
5425 | return err; | |
5426 | } | |
5427 | ||
5428 | /* | |
5429 | * On success, We end up with an outstanding reference count against | |
5430 | * iloc->bh. This _must_ be cleaned up later. | |
5431 | */ | |
5432 | ||
5433 | int | |
617ba13b MC |
5434 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5435 | struct ext4_iloc *iloc) | |
ac27a0ec | 5436 | { |
0390131b FM |
5437 | int err; |
5438 | ||
5439 | err = ext4_get_inode_loc(inode, iloc); | |
5440 | if (!err) { | |
5441 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5442 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5443 | if (err) { | |
5444 | brelse(iloc->bh); | |
5445 | iloc->bh = NULL; | |
ac27a0ec DK |
5446 | } |
5447 | } | |
617ba13b | 5448 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5449 | return err; |
5450 | } | |
5451 | ||
6dd4ee7c KS |
5452 | /* |
5453 | * Expand an inode by new_extra_isize bytes. | |
5454 | * Returns 0 on success or negative error number on failure. | |
5455 | */ | |
1d03ec98 AK |
5456 | static int ext4_expand_extra_isize(struct inode *inode, |
5457 | unsigned int new_extra_isize, | |
5458 | struct ext4_iloc iloc, | |
5459 | handle_t *handle) | |
6dd4ee7c KS |
5460 | { |
5461 | struct ext4_inode *raw_inode; | |
5462 | struct ext4_xattr_ibody_header *header; | |
5463 | struct ext4_xattr_entry *entry; | |
5464 | ||
5465 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5466 | return 0; | |
5467 | ||
5468 | raw_inode = ext4_raw_inode(&iloc); | |
5469 | ||
5470 | header = IHDR(inode, raw_inode); | |
5471 | entry = IFIRST(header); | |
5472 | ||
5473 | /* No extended attributes present */ | |
5474 | if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) || | |
5475 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5476 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, | |
5477 | new_extra_isize); | |
5478 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5479 | return 0; | |
5480 | } | |
5481 | ||
5482 | /* try to expand with EAs present */ | |
5483 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5484 | raw_inode, handle); | |
5485 | } | |
5486 | ||
ac27a0ec DK |
5487 | /* |
5488 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5489 | * dirtiness (it may still be data-dirty). | |
5490 | * This means that the in-core inode may be reaped by prune_icache | |
5491 | * without having to perform any I/O. This is a very good thing, | |
5492 | * because *any* task may call prune_icache - even ones which | |
5493 | * have a transaction open against a different journal. | |
5494 | * | |
5495 | * Is this cheating? Not really. Sure, we haven't written the | |
5496 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5497 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5498 | * we start and wait on commits. | |
5499 | * | |
5500 | * Is this efficient/effective? Well, we're being nice to the system | |
5501 | * by cleaning up our inodes proactively so they can be reaped | |
5502 | * without I/O. But we are potentially leaving up to five seconds' | |
5503 | * worth of inodes floating about which prune_icache wants us to | |
5504 | * write out. One way to fix that would be to get prune_icache() | |
5505 | * to do a write_super() to free up some memory. It has the desired | |
5506 | * effect. | |
5507 | */ | |
617ba13b | 5508 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5509 | { |
617ba13b | 5510 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5511 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5512 | static unsigned int mnt_count; | |
5513 | int err, ret; | |
ac27a0ec DK |
5514 | |
5515 | might_sleep(); | |
617ba13b | 5516 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5517 | if (ext4_handle_valid(handle) && |
5518 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
6dd4ee7c KS |
5519 | !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { |
5520 | /* | |
5521 | * We need extra buffer credits since we may write into EA block | |
5522 | * with this same handle. If journal_extend fails, then it will | |
5523 | * only result in a minor loss of functionality for that inode. | |
5524 | * If this is felt to be critical, then e2fsck should be run to | |
5525 | * force a large enough s_min_extra_isize. | |
5526 | */ | |
5527 | if ((jbd2_journal_extend(handle, | |
5528 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5529 | ret = ext4_expand_extra_isize(inode, | |
5530 | sbi->s_want_extra_isize, | |
5531 | iloc, handle); | |
5532 | if (ret) { | |
5533 | EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND; | |
c1bddad9 AK |
5534 | if (mnt_count != |
5535 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
46e665e9 | 5536 | ext4_warning(inode->i_sb, __func__, |
6dd4ee7c KS |
5537 | "Unable to expand inode %lu. Delete" |
5538 | " some EAs or run e2fsck.", | |
5539 | inode->i_ino); | |
c1bddad9 AK |
5540 | mnt_count = |
5541 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5542 | } |
5543 | } | |
5544 | } | |
5545 | } | |
ac27a0ec | 5546 | if (!err) |
617ba13b | 5547 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5548 | return err; |
5549 | } | |
5550 | ||
5551 | /* | |
617ba13b | 5552 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5553 | * |
5554 | * We're really interested in the case where a file is being extended. | |
5555 | * i_size has been changed by generic_commit_write() and we thus need | |
5556 | * to include the updated inode in the current transaction. | |
5557 | * | |
a269eb18 | 5558 | * Also, vfs_dq_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5559 | * are allocated to the file. |
5560 | * | |
5561 | * If the inode is marked synchronous, we don't honour that here - doing | |
5562 | * so would cause a commit on atime updates, which we don't bother doing. | |
5563 | * We handle synchronous inodes at the highest possible level. | |
5564 | */ | |
617ba13b | 5565 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 5566 | { |
ac27a0ec DK |
5567 | handle_t *handle; |
5568 | ||
617ba13b | 5569 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
5570 | if (IS_ERR(handle)) |
5571 | goto out; | |
f3dc272f | 5572 | |
f3dc272f CW |
5573 | ext4_mark_inode_dirty(handle, inode); |
5574 | ||
617ba13b | 5575 | ext4_journal_stop(handle); |
ac27a0ec DK |
5576 | out: |
5577 | return; | |
5578 | } | |
5579 | ||
5580 | #if 0 | |
5581 | /* | |
5582 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5583 | * it from being flushed to disk early. Unlike | |
617ba13b | 5584 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5585 | * returns no iloc structure, so the caller needs to repeat the iloc |
5586 | * lookup to mark the inode dirty later. | |
5587 | */ | |
617ba13b | 5588 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5589 | { |
617ba13b | 5590 | struct ext4_iloc iloc; |
ac27a0ec DK |
5591 | |
5592 | int err = 0; | |
5593 | if (handle) { | |
617ba13b | 5594 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5595 | if (!err) { |
5596 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5597 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5598 | if (!err) |
0390131b FM |
5599 | err = ext4_handle_dirty_metadata(handle, |
5600 | inode, | |
5601 | iloc.bh); | |
ac27a0ec DK |
5602 | brelse(iloc.bh); |
5603 | } | |
5604 | } | |
617ba13b | 5605 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5606 | return err; |
5607 | } | |
5608 | #endif | |
5609 | ||
617ba13b | 5610 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5611 | { |
5612 | journal_t *journal; | |
5613 | handle_t *handle; | |
5614 | int err; | |
5615 | ||
5616 | /* | |
5617 | * We have to be very careful here: changing a data block's | |
5618 | * journaling status dynamically is dangerous. If we write a | |
5619 | * data block to the journal, change the status and then delete | |
5620 | * that block, we risk forgetting to revoke the old log record | |
5621 | * from the journal and so a subsequent replay can corrupt data. | |
5622 | * So, first we make sure that the journal is empty and that | |
5623 | * nobody is changing anything. | |
5624 | */ | |
5625 | ||
617ba13b | 5626 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5627 | if (!journal) |
5628 | return 0; | |
d699594d | 5629 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5630 | return -EROFS; |
5631 | ||
dab291af MC |
5632 | jbd2_journal_lock_updates(journal); |
5633 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
5634 | |
5635 | /* | |
5636 | * OK, there are no updates running now, and all cached data is | |
5637 | * synced to disk. We are now in a completely consistent state | |
5638 | * which doesn't have anything in the journal, and we know that | |
5639 | * no filesystem updates are running, so it is safe to modify | |
5640 | * the inode's in-core data-journaling state flag now. | |
5641 | */ | |
5642 | ||
5643 | if (val) | |
617ba13b | 5644 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
ac27a0ec | 5645 | else |
617ba13b MC |
5646 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
5647 | ext4_set_aops(inode); | |
ac27a0ec | 5648 | |
dab291af | 5649 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
5650 | |
5651 | /* Finally we can mark the inode as dirty. */ | |
5652 | ||
617ba13b | 5653 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
5654 | if (IS_ERR(handle)) |
5655 | return PTR_ERR(handle); | |
5656 | ||
617ba13b | 5657 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5658 | ext4_handle_sync(handle); |
617ba13b MC |
5659 | ext4_journal_stop(handle); |
5660 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5661 | |
5662 | return err; | |
5663 | } | |
2e9ee850 AK |
5664 | |
5665 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5666 | { | |
5667 | return !buffer_mapped(bh); | |
5668 | } | |
5669 | ||
c2ec175c | 5670 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5671 | { |
c2ec175c | 5672 | struct page *page = vmf->page; |
2e9ee850 AK |
5673 | loff_t size; |
5674 | unsigned long len; | |
5675 | int ret = -EINVAL; | |
79f0be8d | 5676 | void *fsdata; |
2e9ee850 AK |
5677 | struct file *file = vma->vm_file; |
5678 | struct inode *inode = file->f_path.dentry->d_inode; | |
5679 | struct address_space *mapping = inode->i_mapping; | |
5680 | ||
5681 | /* | |
5682 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
5683 | * get i_mutex because we are already holding mmap_sem. | |
5684 | */ | |
5685 | down_read(&inode->i_alloc_sem); | |
5686 | size = i_size_read(inode); | |
5687 | if (page->mapping != mapping || size <= page_offset(page) | |
5688 | || !PageUptodate(page)) { | |
5689 | /* page got truncated from under us? */ | |
5690 | goto out_unlock; | |
5691 | } | |
5692 | ret = 0; | |
5693 | if (PageMappedToDisk(page)) | |
5694 | goto out_unlock; | |
5695 | ||
5696 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5697 | len = size & ~PAGE_CACHE_MASK; | |
5698 | else | |
5699 | len = PAGE_CACHE_SIZE; | |
5700 | ||
a827eaff AK |
5701 | lock_page(page); |
5702 | /* | |
5703 | * return if we have all the buffers mapped. This avoid | |
5704 | * the need to call write_begin/write_end which does a | |
5705 | * journal_start/journal_stop which can block and take | |
5706 | * long time | |
5707 | */ | |
2e9ee850 | 5708 | if (page_has_buffers(page)) { |
2e9ee850 | 5709 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff AK |
5710 | ext4_bh_unmapped)) { |
5711 | unlock_page(page); | |
2e9ee850 | 5712 | goto out_unlock; |
a827eaff | 5713 | } |
2e9ee850 | 5714 | } |
a827eaff | 5715 | unlock_page(page); |
2e9ee850 AK |
5716 | /* |
5717 | * OK, we need to fill the hole... Do write_begin write_end | |
5718 | * to do block allocation/reservation.We are not holding | |
5719 | * inode.i__mutex here. That allow * parallel write_begin, | |
5720 | * write_end call. lock_page prevent this from happening | |
5721 | * on the same page though | |
5722 | */ | |
5723 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
79f0be8d | 5724 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); |
2e9ee850 AK |
5725 | if (ret < 0) |
5726 | goto out_unlock; | |
5727 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
79f0be8d | 5728 | len, len, page, fsdata); |
2e9ee850 AK |
5729 | if (ret < 0) |
5730 | goto out_unlock; | |
5731 | ret = 0; | |
5732 | out_unlock: | |
c2ec175c NP |
5733 | if (ret) |
5734 | ret = VM_FAULT_SIGBUS; | |
2e9ee850 AK |
5735 | up_read(&inode->i_alloc_sem); |
5736 | return ret; | |
5737 | } |