]>
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> | |
3dcf5451 | 40 | #include "ext4_jbd2.h" |
ac27a0ec DK |
41 | #include "xattr.h" |
42 | #include "acl.h" | |
d2a17637 | 43 | #include "ext4_extents.h" |
ac27a0ec | 44 | |
a1d6cc56 AK |
45 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
46 | ||
678aaf48 JK |
47 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
48 | loff_t new_size) | |
49 | { | |
7f5aa215 JK |
50 | return jbd2_journal_begin_ordered_truncate( |
51 | EXT4_SB(inode->i_sb)->s_journal, | |
52 | &EXT4_I(inode)->jinode, | |
53 | new_size); | |
678aaf48 JK |
54 | } |
55 | ||
64769240 AT |
56 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
57 | ||
ac27a0ec DK |
58 | /* |
59 | * Test whether an inode is a fast symlink. | |
60 | */ | |
617ba13b | 61 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 62 | { |
617ba13b | 63 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
64 | (inode->i_sb->s_blocksize >> 9) : 0; |
65 | ||
66 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
67 | } | |
68 | ||
69 | /* | |
617ba13b | 70 | * The ext4 forget function must perform a revoke if we are freeing data |
ac27a0ec DK |
71 | * which has been journaled. Metadata (eg. indirect blocks) must be |
72 | * revoked in all cases. | |
73 | * | |
74 | * "bh" may be NULL: a metadata block may have been freed from memory | |
75 | * but there may still be a record of it in the journal, and that record | |
76 | * still needs to be revoked. | |
0390131b FM |
77 | * |
78 | * If the handle isn't valid we're not journaling so there's nothing to do. | |
ac27a0ec | 79 | */ |
617ba13b MC |
80 | int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, |
81 | struct buffer_head *bh, ext4_fsblk_t blocknr) | |
ac27a0ec DK |
82 | { |
83 | int err; | |
84 | ||
0390131b FM |
85 | if (!ext4_handle_valid(handle)) |
86 | return 0; | |
87 | ||
ac27a0ec DK |
88 | might_sleep(); |
89 | ||
90 | BUFFER_TRACE(bh, "enter"); | |
91 | ||
92 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
93 | "data mode %lx\n", | |
94 | bh, is_metadata, inode->i_mode, | |
95 | test_opt(inode->i_sb, DATA_FLAGS)); | |
96 | ||
97 | /* Never use the revoke function if we are doing full data | |
98 | * journaling: there is no need to, and a V1 superblock won't | |
99 | * support it. Otherwise, only skip the revoke on un-journaled | |
100 | * data blocks. */ | |
101 | ||
617ba13b MC |
102 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
103 | (!is_metadata && !ext4_should_journal_data(inode))) { | |
ac27a0ec | 104 | if (bh) { |
dab291af | 105 | BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
617ba13b | 106 | return ext4_journal_forget(handle, bh); |
ac27a0ec DK |
107 | } |
108 | return 0; | |
109 | } | |
110 | ||
111 | /* | |
112 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
113 | */ | |
617ba13b MC |
114 | BUFFER_TRACE(bh, "call ext4_journal_revoke"); |
115 | err = ext4_journal_revoke(handle, blocknr, bh); | |
ac27a0ec | 116 | if (err) |
46e665e9 | 117 | ext4_abort(inode->i_sb, __func__, |
ac27a0ec DK |
118 | "error %d when attempting revoke", err); |
119 | BUFFER_TRACE(bh, "exit"); | |
120 | return err; | |
121 | } | |
122 | ||
123 | /* | |
124 | * Work out how many blocks we need to proceed with the next chunk of a | |
125 | * truncate transaction. | |
126 | */ | |
127 | static unsigned long blocks_for_truncate(struct inode *inode) | |
128 | { | |
725d26d3 | 129 | ext4_lblk_t needed; |
ac27a0ec DK |
130 | |
131 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
132 | ||
133 | /* Give ourselves just enough room to cope with inodes in which | |
134 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
135 | * which resulted in random data in an inode which looked enough | |
617ba13b | 136 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
137 | * will go a bit crazy if that happens, but at least we should |
138 | * try not to panic the whole kernel. */ | |
139 | if (needed < 2) | |
140 | needed = 2; | |
141 | ||
142 | /* But we need to bound the transaction so we don't overflow the | |
143 | * journal. */ | |
617ba13b MC |
144 | if (needed > EXT4_MAX_TRANS_DATA) |
145 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 146 | |
617ba13b | 147 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
148 | } |
149 | ||
150 | /* | |
151 | * Truncate transactions can be complex and absolutely huge. So we need to | |
152 | * be able to restart the transaction at a conventient checkpoint to make | |
153 | * sure we don't overflow the journal. | |
154 | * | |
155 | * start_transaction gets us a new handle for a truncate transaction, | |
156 | * and extend_transaction tries to extend the existing one a bit. If | |
157 | * extend fails, we need to propagate the failure up and restart the | |
158 | * transaction in the top-level truncate loop. --sct | |
159 | */ | |
160 | static handle_t *start_transaction(struct inode *inode) | |
161 | { | |
162 | handle_t *result; | |
163 | ||
617ba13b | 164 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
165 | if (!IS_ERR(result)) |
166 | return result; | |
167 | ||
617ba13b | 168 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
169 | return result; |
170 | } | |
171 | ||
172 | /* | |
173 | * Try to extend this transaction for the purposes of truncation. | |
174 | * | |
175 | * Returns 0 if we managed to create more room. If we can't create more | |
176 | * room, and the transaction must be restarted we return 1. | |
177 | */ | |
178 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
179 | { | |
0390131b FM |
180 | if (!ext4_handle_valid(handle)) |
181 | return 0; | |
182 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
ac27a0ec | 183 | return 0; |
617ba13b | 184 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
185 | return 0; |
186 | return 1; | |
187 | } | |
188 | ||
189 | /* | |
190 | * Restart the transaction associated with *handle. This does a commit, | |
191 | * so before we call here everything must be consistently dirtied against | |
192 | * this transaction. | |
193 | */ | |
617ba13b | 194 | static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) |
ac27a0ec | 195 | { |
0390131b | 196 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 197 | jbd_debug(2, "restarting handle %p\n", handle); |
617ba13b | 198 | return ext4_journal_restart(handle, blocks_for_truncate(inode)); |
ac27a0ec DK |
199 | } |
200 | ||
201 | /* | |
202 | * Called at the last iput() if i_nlink is zero. | |
203 | */ | |
af5bc92d | 204 | void ext4_delete_inode(struct inode *inode) |
ac27a0ec DK |
205 | { |
206 | handle_t *handle; | |
bc965ab3 | 207 | int err; |
ac27a0ec | 208 | |
678aaf48 JK |
209 | if (ext4_should_order_data(inode)) |
210 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
211 | truncate_inode_pages(&inode->i_data, 0); |
212 | ||
213 | if (is_bad_inode(inode)) | |
214 | goto no_delete; | |
215 | ||
bc965ab3 | 216 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 217 | if (IS_ERR(handle)) { |
bc965ab3 | 218 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
219 | /* |
220 | * If we're going to skip the normal cleanup, we still need to | |
221 | * make sure that the in-core orphan linked list is properly | |
222 | * cleaned up. | |
223 | */ | |
617ba13b | 224 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
225 | goto no_delete; |
226 | } | |
227 | ||
228 | if (IS_SYNC(inode)) | |
0390131b | 229 | ext4_handle_sync(handle); |
ac27a0ec | 230 | inode->i_size = 0; |
bc965ab3 TT |
231 | err = ext4_mark_inode_dirty(handle, inode); |
232 | if (err) { | |
233 | ext4_warning(inode->i_sb, __func__, | |
234 | "couldn't mark inode dirty (err %d)", err); | |
235 | goto stop_handle; | |
236 | } | |
ac27a0ec | 237 | if (inode->i_blocks) |
617ba13b | 238 | ext4_truncate(inode); |
bc965ab3 TT |
239 | |
240 | /* | |
241 | * ext4_ext_truncate() doesn't reserve any slop when it | |
242 | * restarts journal transactions; therefore there may not be | |
243 | * enough credits left in the handle to remove the inode from | |
244 | * the orphan list and set the dtime field. | |
245 | */ | |
0390131b | 246 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
247 | err = ext4_journal_extend(handle, 3); |
248 | if (err > 0) | |
249 | err = ext4_journal_restart(handle, 3); | |
250 | if (err != 0) { | |
251 | ext4_warning(inode->i_sb, __func__, | |
252 | "couldn't extend journal (err %d)", err); | |
253 | stop_handle: | |
254 | ext4_journal_stop(handle); | |
255 | goto no_delete; | |
256 | } | |
257 | } | |
258 | ||
ac27a0ec | 259 | /* |
617ba13b | 260 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 261 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 262 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 263 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 264 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
265 | * (Well, we could do this if we need to, but heck - it works) |
266 | */ | |
617ba13b MC |
267 | ext4_orphan_del(handle, inode); |
268 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
269 | |
270 | /* | |
271 | * One subtle ordering requirement: if anything has gone wrong | |
272 | * (transaction abort, IO errors, whatever), then we can still | |
273 | * do these next steps (the fs will already have been marked as | |
274 | * having errors), but we can't free the inode if the mark_dirty | |
275 | * fails. | |
276 | */ | |
617ba13b | 277 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
278 | /* If that failed, just do the required in-core inode clear. */ |
279 | clear_inode(inode); | |
280 | else | |
617ba13b MC |
281 | ext4_free_inode(handle, inode); |
282 | ext4_journal_stop(handle); | |
ac27a0ec DK |
283 | return; |
284 | no_delete: | |
285 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
286 | } | |
287 | ||
288 | typedef struct { | |
289 | __le32 *p; | |
290 | __le32 key; | |
291 | struct buffer_head *bh; | |
292 | } Indirect; | |
293 | ||
294 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
295 | { | |
296 | p->key = *(p->p = v); | |
297 | p->bh = bh; | |
298 | } | |
299 | ||
ac27a0ec | 300 | /** |
617ba13b | 301 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
302 | * @inode: inode in question (we are only interested in its superblock) |
303 | * @i_block: block number to be parsed | |
304 | * @offsets: array to store the offsets in | |
8c55e204 DK |
305 | * @boundary: set this non-zero if the referred-to block is likely to be |
306 | * followed (on disk) by an indirect block. | |
ac27a0ec | 307 | * |
617ba13b | 308 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
309 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
310 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
311 | * This function translates the block number into path in that tree - | |
312 | * return value is the path length and @offsets[n] is the offset of | |
313 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
314 | * (negative or too large) warning is printed and zero returned. | |
315 | * | |
316 | * Note: function doesn't find node addresses, so no IO is needed. All | |
317 | * we need to know is the capacity of indirect blocks (taken from the | |
318 | * inode->i_sb). | |
319 | */ | |
320 | ||
321 | /* | |
322 | * Portability note: the last comparison (check that we fit into triple | |
323 | * indirect block) is spelled differently, because otherwise on an | |
324 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
325 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
326 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
327 | * i_block would have to be negative in the very beginning, so we would not | |
328 | * get there at all. | |
329 | */ | |
330 | ||
617ba13b | 331 | static int ext4_block_to_path(struct inode *inode, |
725d26d3 AK |
332 | ext4_lblk_t i_block, |
333 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 334 | { |
617ba13b MC |
335 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
336 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
337 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
338 | indirect_blocks = ptrs, |
339 | double_blocks = (1 << (ptrs_bits * 2)); | |
340 | int n = 0; | |
341 | int final = 0; | |
342 | ||
343 | if (i_block < 0) { | |
af5bc92d | 344 | ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0"); |
ac27a0ec DK |
345 | } else if (i_block < direct_blocks) { |
346 | offsets[n++] = i_block; | |
347 | final = direct_blocks; | |
af5bc92d | 348 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
617ba13b | 349 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
350 | offsets[n++] = i_block; |
351 | final = ptrs; | |
352 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 353 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
354 | offsets[n++] = i_block >> ptrs_bits; |
355 | offsets[n++] = i_block & (ptrs - 1); | |
356 | final = ptrs; | |
357 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 358 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
359 | offsets[n++] = i_block >> (ptrs_bits * 2); |
360 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
361 | offsets[n++] = i_block & (ptrs - 1); | |
362 | final = ptrs; | |
363 | } else { | |
e2b46574 | 364 | ext4_warning(inode->i_sb, "ext4_block_to_path", |
06a279d6 | 365 | "block %lu > max in inode %lu", |
e2b46574 | 366 | i_block + direct_blocks + |
06a279d6 | 367 | indirect_blocks + double_blocks, inode->i_ino); |
ac27a0ec DK |
368 | } |
369 | if (boundary) | |
370 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
371 | return n; | |
372 | } | |
373 | ||
fe2c8191 | 374 | static int __ext4_check_blockref(const char *function, struct inode *inode, |
f73953c0 | 375 | __le32 *p, unsigned int max) { |
fe2c8191 TN |
376 | |
377 | unsigned int maxblocks = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es); | |
f73953c0 | 378 | __le32 *bref = p; |
fe2c8191 | 379 | while (bref < p+max) { |
f73953c0 | 380 | if (unlikely(le32_to_cpu(*bref) >= maxblocks)) { |
fe2c8191 TN |
381 | ext4_error(inode->i_sb, function, |
382 | "block reference %u >= max (%u) " | |
383 | "in inode #%lu, offset=%d", | |
f73953c0 | 384 | le32_to_cpu(*bref), maxblocks, |
fe2c8191 TN |
385 | inode->i_ino, (int)(bref-p)); |
386 | return -EIO; | |
387 | } | |
388 | bref++; | |
389 | } | |
390 | return 0; | |
391 | } | |
392 | ||
393 | ||
394 | #define ext4_check_indirect_blockref(inode, bh) \ | |
395 | __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \ | |
396 | EXT4_ADDR_PER_BLOCK((inode)->i_sb)) | |
397 | ||
398 | #define ext4_check_inode_blockref(inode) \ | |
399 | __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \ | |
400 | EXT4_NDIR_BLOCKS) | |
401 | ||
ac27a0ec | 402 | /** |
617ba13b | 403 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
404 | * @inode: inode in question |
405 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
406 | * @offsets: offsets of pointers in inode/indirect blocks | |
407 | * @chain: place to store the result | |
408 | * @err: here we store the error value | |
409 | * | |
410 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
411 | * if everything went OK or the pointer to the last filled triple | |
412 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
413 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
414 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
415 | * number (it points into struct inode for i==0 and into the bh->b_data | |
416 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
417 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
418 | * numbers of the chain, addresses they were taken from (and where we can | |
419 | * verify that chain did not change) and buffer_heads hosting these | |
420 | * numbers. | |
421 | * | |
422 | * Function stops when it stumbles upon zero pointer (absent block) | |
423 | * (pointer to last triple returned, *@err == 0) | |
424 | * or when it gets an IO error reading an indirect block | |
425 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
426 | * or when it reads all @depth-1 indirect blocks successfully and finds |
427 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
428 | * |
429 | * Need to be called with | |
0e855ac8 | 430 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 431 | */ |
725d26d3 AK |
432 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
433 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
434 | Indirect chain[4], int *err) |
435 | { | |
436 | struct super_block *sb = inode->i_sb; | |
437 | Indirect *p = chain; | |
438 | struct buffer_head *bh; | |
439 | ||
440 | *err = 0; | |
441 | /* i_data is not going away, no lock needed */ | |
af5bc92d | 442 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
443 | if (!p->key) |
444 | goto no_block; | |
445 | while (--depth) { | |
fe2c8191 TN |
446 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
447 | if (unlikely(!bh)) | |
ac27a0ec | 448 | goto failure; |
fe2c8191 TN |
449 | |
450 | if (!bh_uptodate_or_lock(bh)) { | |
451 | if (bh_submit_read(bh) < 0) { | |
452 | put_bh(bh); | |
453 | goto failure; | |
454 | } | |
455 | /* validate block references */ | |
456 | if (ext4_check_indirect_blockref(inode, bh)) { | |
457 | put_bh(bh); | |
458 | goto failure; | |
459 | } | |
460 | } | |
461 | ||
af5bc92d | 462 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
ac27a0ec DK |
463 | /* Reader: end */ |
464 | if (!p->key) | |
465 | goto no_block; | |
466 | } | |
467 | return NULL; | |
468 | ||
ac27a0ec DK |
469 | failure: |
470 | *err = -EIO; | |
471 | no_block: | |
472 | return p; | |
473 | } | |
474 | ||
475 | /** | |
617ba13b | 476 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
477 | * @inode: owner |
478 | * @ind: descriptor of indirect block. | |
479 | * | |
1cc8dcf5 | 480 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
481 | * It is used when heuristic for sequential allocation fails. |
482 | * Rules are: | |
483 | * + if there is a block to the left of our position - allocate near it. | |
484 | * + if pointer will live in indirect block - allocate near that block. | |
485 | * + if pointer will live in inode - allocate in the same | |
486 | * cylinder group. | |
487 | * | |
488 | * In the latter case we colour the starting block by the callers PID to | |
489 | * prevent it from clashing with concurrent allocations for a different inode | |
490 | * in the same block group. The PID is used here so that functionally related | |
491 | * files will be close-by on-disk. | |
492 | * | |
493 | * Caller must make sure that @ind is valid and will stay that way. | |
494 | */ | |
617ba13b | 495 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 496 | { |
617ba13b | 497 | struct ext4_inode_info *ei = EXT4_I(inode); |
af5bc92d | 498 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
ac27a0ec | 499 | __le32 *p; |
617ba13b | 500 | ext4_fsblk_t bg_start; |
74d3487f | 501 | ext4_fsblk_t last_block; |
617ba13b | 502 | ext4_grpblk_t colour; |
a4912123 TT |
503 | ext4_group_t block_group; |
504 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
ac27a0ec DK |
505 | |
506 | /* Try to find previous block */ | |
507 | for (p = ind->p - 1; p >= start; p--) { | |
508 | if (*p) | |
509 | return le32_to_cpu(*p); | |
510 | } | |
511 | ||
512 | /* No such thing, so let's try location of indirect block */ | |
513 | if (ind->bh) | |
514 | return ind->bh->b_blocknr; | |
515 | ||
516 | /* | |
517 | * It is going to be referred to from the inode itself? OK, just put it | |
518 | * into the same cylinder group then. | |
519 | */ | |
a4912123 TT |
520 | block_group = ei->i_block_group; |
521 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
522 | block_group &= ~(flex_size-1); | |
523 | if (S_ISREG(inode->i_mode)) | |
524 | block_group++; | |
525 | } | |
526 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
74d3487f VC |
527 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
528 | ||
a4912123 TT |
529 | /* |
530 | * If we are doing delayed allocation, we don't need take | |
531 | * colour into account. | |
532 | */ | |
533 | if (test_opt(inode->i_sb, DELALLOC)) | |
534 | return bg_start; | |
535 | ||
74d3487f VC |
536 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
537 | colour = (current->pid % 16) * | |
617ba13b | 538 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
539 | else |
540 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
541 | return bg_start + colour; |
542 | } | |
543 | ||
544 | /** | |
1cc8dcf5 | 545 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
546 | * @inode: owner |
547 | * @block: block we want | |
ac27a0ec | 548 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 549 | * |
1cc8dcf5 | 550 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 551 | * returns it. |
ac27a0ec | 552 | */ |
725d26d3 | 553 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
fb01bfda | 554 | Indirect *partial) |
ac27a0ec | 555 | { |
ac27a0ec | 556 | /* |
c2ea3fde | 557 | * XXX need to get goal block from mballoc's data structures |
ac27a0ec | 558 | */ |
ac27a0ec | 559 | |
617ba13b | 560 | return ext4_find_near(inode, partial); |
ac27a0ec DK |
561 | } |
562 | ||
563 | /** | |
617ba13b | 564 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
565 | * of direct blocks need to be allocated for the given branch. |
566 | * | |
567 | * @branch: chain of indirect blocks | |
568 | * @k: number of blocks need for indirect blocks | |
569 | * @blks: number of data blocks to be mapped. | |
570 | * @blocks_to_boundary: the offset in the indirect block | |
571 | * | |
572 | * return the total number of blocks to be allocate, including the | |
573 | * direct and indirect blocks. | |
574 | */ | |
498e5f24 | 575 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
ac27a0ec DK |
576 | int blocks_to_boundary) |
577 | { | |
498e5f24 | 578 | unsigned int count = 0; |
ac27a0ec DK |
579 | |
580 | /* | |
581 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
582 | * then it's clear blocks on that path have not allocated | |
583 | */ | |
584 | if (k > 0) { | |
585 | /* right now we don't handle cross boundary allocation */ | |
586 | if (blks < blocks_to_boundary + 1) | |
587 | count += blks; | |
588 | else | |
589 | count += blocks_to_boundary + 1; | |
590 | return count; | |
591 | } | |
592 | ||
593 | count++; | |
594 | while (count < blks && count <= blocks_to_boundary && | |
595 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
596 | count++; | |
597 | } | |
598 | return count; | |
599 | } | |
600 | ||
601 | /** | |
617ba13b | 602 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
603 | * @indirect_blks: the number of blocks need to allocate for indirect |
604 | * blocks | |
605 | * | |
606 | * @new_blocks: on return it will store the new block numbers for | |
607 | * the indirect blocks(if needed) and the first direct block, | |
608 | * @blks: on return it will store the total number of allocated | |
609 | * direct blocks | |
610 | */ | |
617ba13b | 611 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
7061eba7 AK |
612 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
613 | int indirect_blks, int blks, | |
614 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec | 615 | { |
815a1130 | 616 | struct ext4_allocation_request ar; |
ac27a0ec | 617 | int target, i; |
7061eba7 | 618 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 619 | int index = 0; |
617ba13b | 620 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
621 | int ret = 0; |
622 | ||
623 | /* | |
624 | * Here we try to allocate the requested multiple blocks at once, | |
625 | * on a best-effort basis. | |
626 | * To build a branch, we should allocate blocks for | |
627 | * the indirect blocks(if not allocated yet), and at least | |
628 | * the first direct block of this branch. That's the | |
629 | * minimum number of blocks need to allocate(required) | |
630 | */ | |
7061eba7 AK |
631 | /* first we try to allocate the indirect blocks */ |
632 | target = indirect_blks; | |
633 | while (target > 0) { | |
ac27a0ec DK |
634 | count = target; |
635 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
636 | current_block = ext4_new_meta_blocks(handle, inode, |
637 | goal, &count, err); | |
ac27a0ec DK |
638 | if (*err) |
639 | goto failed_out; | |
640 | ||
641 | target -= count; | |
642 | /* allocate blocks for indirect blocks */ | |
643 | while (index < indirect_blks && count) { | |
644 | new_blocks[index++] = current_block++; | |
645 | count--; | |
646 | } | |
7061eba7 AK |
647 | if (count > 0) { |
648 | /* | |
649 | * save the new block number | |
650 | * for the first direct block | |
651 | */ | |
652 | new_blocks[index] = current_block; | |
653 | printk(KERN_INFO "%s returned more blocks than " | |
654 | "requested\n", __func__); | |
655 | WARN_ON(1); | |
ac27a0ec | 656 | break; |
7061eba7 | 657 | } |
ac27a0ec DK |
658 | } |
659 | ||
7061eba7 AK |
660 | target = blks - count ; |
661 | blk_allocated = count; | |
662 | if (!target) | |
663 | goto allocated; | |
664 | /* Now allocate data blocks */ | |
815a1130 TT |
665 | memset(&ar, 0, sizeof(ar)); |
666 | ar.inode = inode; | |
667 | ar.goal = goal; | |
668 | ar.len = target; | |
669 | ar.logical = iblock; | |
670 | if (S_ISREG(inode->i_mode)) | |
671 | /* enable in-core preallocation only for regular files */ | |
672 | ar.flags = EXT4_MB_HINT_DATA; | |
673 | ||
674 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
675 | ||
7061eba7 AK |
676 | if (*err && (target == blks)) { |
677 | /* | |
678 | * if the allocation failed and we didn't allocate | |
679 | * any blocks before | |
680 | */ | |
681 | goto failed_out; | |
682 | } | |
683 | if (!*err) { | |
684 | if (target == blks) { | |
685 | /* | |
686 | * save the new block number | |
687 | * for the first direct block | |
688 | */ | |
689 | new_blocks[index] = current_block; | |
690 | } | |
815a1130 | 691 | blk_allocated += ar.len; |
7061eba7 AK |
692 | } |
693 | allocated: | |
ac27a0ec | 694 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 695 | ret = blk_allocated; |
ac27a0ec DK |
696 | *err = 0; |
697 | return ret; | |
698 | failed_out: | |
af5bc92d | 699 | for (i = 0; i < index; i++) |
c9de560d | 700 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec DK |
701 | return ret; |
702 | } | |
703 | ||
704 | /** | |
617ba13b | 705 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
706 | * @inode: owner |
707 | * @indirect_blks: number of allocated indirect blocks | |
708 | * @blks: number of allocated direct blocks | |
709 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
710 | * @branch: place to store the chain in. | |
711 | * | |
712 | * This function allocates blocks, zeroes out all but the last one, | |
713 | * links them into chain and (if we are synchronous) writes them to disk. | |
714 | * In other words, it prepares a branch that can be spliced onto the | |
715 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 716 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
717 | * we had read the existing part of chain and partial points to the last |
718 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 719 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
720 | * place chain is disconnected - *branch->p is still zero (we did not |
721 | * set the last link), but branch->key contains the number that should | |
722 | * be placed into *branch->p to fill that gap. | |
723 | * | |
724 | * If allocation fails we free all blocks we've allocated (and forget | |
725 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 726 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
727 | * as described above and return 0. |
728 | */ | |
617ba13b | 729 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
7061eba7 AK |
730 | ext4_lblk_t iblock, int indirect_blks, |
731 | int *blks, ext4_fsblk_t goal, | |
732 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
733 | { |
734 | int blocksize = inode->i_sb->s_blocksize; | |
735 | int i, n = 0; | |
736 | int err = 0; | |
737 | struct buffer_head *bh; | |
738 | int num; | |
617ba13b MC |
739 | ext4_fsblk_t new_blocks[4]; |
740 | ext4_fsblk_t current_block; | |
ac27a0ec | 741 | |
7061eba7 | 742 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
743 | *blks, new_blocks, &err); |
744 | if (err) | |
745 | return err; | |
746 | ||
747 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
748 | /* | |
749 | * metadata blocks and data blocks are allocated. | |
750 | */ | |
751 | for (n = 1; n <= indirect_blks; n++) { | |
752 | /* | |
753 | * Get buffer_head for parent block, zero it out | |
754 | * and set the pointer to new one, then send | |
755 | * parent to disk. | |
756 | */ | |
757 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
758 | branch[n].bh = bh; | |
759 | lock_buffer(bh); | |
760 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 761 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
762 | if (err) { |
763 | unlock_buffer(bh); | |
764 | brelse(bh); | |
765 | goto failed; | |
766 | } | |
767 | ||
768 | memset(bh->b_data, 0, blocksize); | |
769 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
770 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
771 | *branch[n].p = branch[n].key; | |
af5bc92d | 772 | if (n == indirect_blks) { |
ac27a0ec DK |
773 | current_block = new_blocks[n]; |
774 | /* | |
775 | * End of chain, update the last new metablock of | |
776 | * the chain to point to the new allocated | |
777 | * data blocks numbers | |
778 | */ | |
779 | for (i=1; i < num; i++) | |
780 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
781 | } | |
782 | BUFFER_TRACE(bh, "marking uptodate"); | |
783 | set_buffer_uptodate(bh); | |
784 | unlock_buffer(bh); | |
785 | ||
0390131b FM |
786 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
787 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
788 | if (err) |
789 | goto failed; | |
790 | } | |
791 | *blks = num; | |
792 | return err; | |
793 | failed: | |
794 | /* Allocation failed, free what we already allocated */ | |
795 | for (i = 1; i <= n ; i++) { | |
dab291af | 796 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
617ba13b | 797 | ext4_journal_forget(handle, branch[i].bh); |
ac27a0ec | 798 | } |
af5bc92d | 799 | for (i = 0; i < indirect_blks; i++) |
c9de560d | 800 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec | 801 | |
c9de560d | 802 | ext4_free_blocks(handle, inode, new_blocks[i], num, 0); |
ac27a0ec DK |
803 | |
804 | return err; | |
805 | } | |
806 | ||
807 | /** | |
617ba13b | 808 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
809 | * @inode: owner |
810 | * @block: (logical) number of block we are adding | |
811 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 812 | * ext4_alloc_branch) |
ac27a0ec DK |
813 | * @where: location of missing link |
814 | * @num: number of indirect blocks we are adding | |
815 | * @blks: number of direct blocks we are adding | |
816 | * | |
817 | * This function fills the missing link and does all housekeeping needed in | |
818 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
819 | * chain to new block and return 0. | |
820 | */ | |
617ba13b | 821 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
725d26d3 | 822 | ext4_lblk_t block, Indirect *where, int num, int blks) |
ac27a0ec DK |
823 | { |
824 | int i; | |
825 | int err = 0; | |
617ba13b | 826 | ext4_fsblk_t current_block; |
ac27a0ec | 827 | |
ac27a0ec DK |
828 | /* |
829 | * If we're splicing into a [td]indirect block (as opposed to the | |
830 | * inode) then we need to get write access to the [td]indirect block | |
831 | * before the splice. | |
832 | */ | |
833 | if (where->bh) { | |
834 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 835 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
836 | if (err) |
837 | goto err_out; | |
838 | } | |
839 | /* That's it */ | |
840 | ||
841 | *where->p = where->key; | |
842 | ||
843 | /* | |
844 | * Update the host buffer_head or inode to point to more just allocated | |
845 | * direct blocks blocks | |
846 | */ | |
847 | if (num == 0 && blks > 1) { | |
848 | current_block = le32_to_cpu(where->key) + 1; | |
849 | for (i = 1; i < blks; i++) | |
af5bc92d | 850 | *(where->p + i) = cpu_to_le32(current_block++); |
ac27a0ec DK |
851 | } |
852 | ||
ac27a0ec DK |
853 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
854 | ||
ef7f3835 | 855 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 856 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
857 | |
858 | /* had we spliced it onto indirect block? */ | |
859 | if (where->bh) { | |
860 | /* | |
861 | * If we spliced it onto an indirect block, we haven't | |
862 | * altered the inode. Note however that if it is being spliced | |
863 | * onto an indirect block at the very end of the file (the | |
864 | * file is growing) then we *will* alter the inode to reflect | |
865 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 866 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
867 | */ |
868 | jbd_debug(5, "splicing indirect only\n"); | |
0390131b FM |
869 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
870 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
ac27a0ec DK |
871 | if (err) |
872 | goto err_out; | |
873 | } else { | |
874 | /* | |
875 | * OK, we spliced it into the inode itself on a direct block. | |
876 | * Inode was dirtied above. | |
877 | */ | |
878 | jbd_debug(5, "splicing direct\n"); | |
879 | } | |
880 | return err; | |
881 | ||
882 | err_out: | |
883 | for (i = 1; i <= num; i++) { | |
dab291af | 884 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
617ba13b | 885 | ext4_journal_forget(handle, where[i].bh); |
c9de560d AT |
886 | ext4_free_blocks(handle, inode, |
887 | le32_to_cpu(where[i-1].key), 1, 0); | |
ac27a0ec | 888 | } |
c9de560d | 889 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0); |
ac27a0ec DK |
890 | |
891 | return err; | |
892 | } | |
893 | ||
894 | /* | |
b920c755 TT |
895 | * The ext4_ind_get_blocks() function handles non-extents inodes |
896 | * (i.e., using the traditional indirect/double-indirect i_blocks | |
897 | * scheme) for ext4_get_blocks(). | |
898 | * | |
ac27a0ec DK |
899 | * Allocation strategy is simple: if we have to allocate something, we will |
900 | * have to go the whole way to leaf. So let's do it before attaching anything | |
901 | * to tree, set linkage between the newborn blocks, write them if sync is | |
902 | * required, recheck the path, free and repeat if check fails, otherwise | |
903 | * set the last missing link (that will protect us from any truncate-generated | |
904 | * removals - all blocks on the path are immune now) and possibly force the | |
905 | * write on the parent block. | |
906 | * That has a nice additional property: no special recovery from the failed | |
907 | * allocations is needed - we simply release blocks and do not touch anything | |
908 | * reachable from inode. | |
909 | * | |
910 | * `handle' can be NULL if create == 0. | |
911 | * | |
ac27a0ec DK |
912 | * return > 0, # of blocks mapped or allocated. |
913 | * return = 0, if plain lookup failed. | |
914 | * return < 0, error case. | |
c278bfec | 915 | * |
b920c755 TT |
916 | * The ext4_ind_get_blocks() function should be called with |
917 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
918 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
919 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
920 | * blocks. | |
ac27a0ec | 921 | */ |
e4d996ca | 922 | static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode, |
498e5f24 TT |
923 | ext4_lblk_t iblock, unsigned int maxblocks, |
924 | struct buffer_head *bh_result, | |
c2177057 | 925 | int flags) |
ac27a0ec DK |
926 | { |
927 | int err = -EIO; | |
725d26d3 | 928 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
929 | Indirect chain[4]; |
930 | Indirect *partial; | |
617ba13b | 931 | ext4_fsblk_t goal; |
ac27a0ec DK |
932 | int indirect_blks; |
933 | int blocks_to_boundary = 0; | |
934 | int depth; | |
617ba13b | 935 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 936 | int count = 0; |
617ba13b | 937 | ext4_fsblk_t first_block = 0; |
61628a3f | 938 | loff_t disksize; |
ac27a0ec DK |
939 | |
940 | ||
a86c6181 | 941 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
c2177057 | 942 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
725d26d3 AK |
943 | depth = ext4_block_to_path(inode, iblock, offsets, |
944 | &blocks_to_boundary); | |
ac27a0ec DK |
945 | |
946 | if (depth == 0) | |
947 | goto out; | |
948 | ||
617ba13b | 949 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
950 | |
951 | /* Simplest case - block found, no allocation needed */ | |
952 | if (!partial) { | |
953 | first_block = le32_to_cpu(chain[depth - 1].key); | |
954 | clear_buffer_new(bh_result); | |
955 | count++; | |
956 | /*map more blocks*/ | |
957 | while (count < maxblocks && count <= blocks_to_boundary) { | |
617ba13b | 958 | ext4_fsblk_t blk; |
ac27a0ec | 959 | |
ac27a0ec DK |
960 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
961 | ||
962 | if (blk == first_block + count) | |
963 | count++; | |
964 | else | |
965 | break; | |
966 | } | |
c278bfec | 967 | goto got_it; |
ac27a0ec DK |
968 | } |
969 | ||
970 | /* Next simple case - plain lookup or failed read of indirect block */ | |
c2177057 | 971 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
ac27a0ec DK |
972 | goto cleanup; |
973 | ||
ac27a0ec | 974 | /* |
c2ea3fde | 975 | * Okay, we need to do block allocation. |
ac27a0ec | 976 | */ |
fb01bfda | 977 | goal = ext4_find_goal(inode, iblock, partial); |
ac27a0ec DK |
978 | |
979 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
980 | indirect_blks = (chain + depth) - partial - 1; | |
981 | ||
982 | /* | |
983 | * Next look up the indirect map to count the totoal number of | |
984 | * direct blocks to allocate for this branch. | |
985 | */ | |
617ba13b | 986 | count = ext4_blks_to_allocate(partial, indirect_blks, |
ac27a0ec DK |
987 | maxblocks, blocks_to_boundary); |
988 | /* | |
617ba13b | 989 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 990 | */ |
7061eba7 AK |
991 | err = ext4_alloc_branch(handle, inode, iblock, indirect_blks, |
992 | &count, goal, | |
993 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
994 | |
995 | /* | |
617ba13b | 996 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
997 | * on the new chain if there is a failure, but that risks using |
998 | * up transaction credits, especially for bitmaps where the | |
999 | * credits cannot be returned. Can we handle this somehow? We | |
1000 | * may need to return -EAGAIN upwards in the worst case. --sct | |
1001 | */ | |
1002 | if (!err) | |
617ba13b | 1003 | err = ext4_splice_branch(handle, inode, iblock, |
ac27a0ec DK |
1004 | partial, indirect_blks, count); |
1005 | /* | |
0e855ac8 | 1006 | * i_disksize growing is protected by i_data_sem. Don't forget to |
ac27a0ec | 1007 | * protect it if you're about to implement concurrent |
617ba13b | 1008 | * ext4_get_block() -bzzz |
ac27a0ec | 1009 | */ |
c2177057 | 1010 | if (!err && (flags & EXT4_GET_BLOCKS_EXTEND_DISKSIZE)) { |
61628a3f MC |
1011 | disksize = ((loff_t) iblock + count) << inode->i_blkbits; |
1012 | if (disksize > i_size_read(inode)) | |
1013 | disksize = i_size_read(inode); | |
1014 | if (disksize > ei->i_disksize) | |
1015 | ei->i_disksize = disksize; | |
1016 | } | |
ac27a0ec DK |
1017 | if (err) |
1018 | goto cleanup; | |
1019 | ||
1020 | set_buffer_new(bh_result); | |
1021 | got_it: | |
1022 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
1023 | if (count > blocks_to_boundary) | |
1024 | set_buffer_boundary(bh_result); | |
1025 | err = count; | |
1026 | /* Clean up and exit */ | |
1027 | partial = chain + depth - 1; /* the whole chain */ | |
1028 | cleanup: | |
1029 | while (partial > chain) { | |
1030 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1031 | brelse(partial->bh); | |
1032 | partial--; | |
1033 | } | |
1034 | BUFFER_TRACE(bh_result, "returned"); | |
1035 | out: | |
1036 | return err; | |
1037 | } | |
1038 | ||
60e58e0f MC |
1039 | qsize_t ext4_get_reserved_space(struct inode *inode) |
1040 | { | |
1041 | unsigned long long total; | |
1042 | ||
1043 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1044 | total = EXT4_I(inode)->i_reserved_data_blocks + | |
1045 | EXT4_I(inode)->i_reserved_meta_blocks; | |
1046 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1047 | ||
1048 | return total; | |
1049 | } | |
12219aea AK |
1050 | /* |
1051 | * Calculate the number of metadata blocks need to reserve | |
1052 | * to allocate @blocks for non extent file based file | |
1053 | */ | |
1054 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks) | |
1055 | { | |
1056 | int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1057 | int ind_blks, dind_blks, tind_blks; | |
1058 | ||
1059 | /* number of new indirect blocks needed */ | |
1060 | ind_blks = (blocks + icap - 1) / icap; | |
1061 | ||
1062 | dind_blks = (ind_blks + icap - 1) / icap; | |
1063 | ||
1064 | tind_blks = 1; | |
1065 | ||
1066 | return ind_blks + dind_blks + tind_blks; | |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * Calculate the number of metadata blocks need to reserve | |
1071 | * to allocate given number of blocks | |
1072 | */ | |
1073 | static int ext4_calc_metadata_amount(struct inode *inode, int blocks) | |
1074 | { | |
cd213226 MC |
1075 | if (!blocks) |
1076 | return 0; | |
1077 | ||
12219aea AK |
1078 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
1079 | return ext4_ext_calc_metadata_amount(inode, blocks); | |
1080 | ||
1081 | return ext4_indirect_calc_metadata_amount(inode, blocks); | |
1082 | } | |
1083 | ||
1084 | static void ext4_da_update_reserve_space(struct inode *inode, int used) | |
1085 | { | |
1086 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1087 | int total, mdb, mdb_free; | |
1088 | ||
1089 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1090 | /* recalculate the number of metablocks still need to be reserved */ | |
1091 | total = EXT4_I(inode)->i_reserved_data_blocks - used; | |
1092 | mdb = ext4_calc_metadata_amount(inode, total); | |
1093 | ||
1094 | /* figure out how many metablocks to release */ | |
1095 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1096 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1097 | ||
6bc6e63f AK |
1098 | if (mdb_free) { |
1099 | /* Account for allocated meta_blocks */ | |
1100 | mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks; | |
1101 | ||
1102 | /* update fs dirty blocks counter */ | |
1103 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free); | |
1104 | EXT4_I(inode)->i_allocated_meta_blocks = 0; | |
1105 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
1106 | } | |
12219aea AK |
1107 | |
1108 | /* update per-inode reservations */ | |
1109 | BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks); | |
1110 | EXT4_I(inode)->i_reserved_data_blocks -= used; | |
12219aea | 1111 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f MC |
1112 | |
1113 | /* | |
1114 | * free those over-booking quota for metadata blocks | |
1115 | */ | |
60e58e0f MC |
1116 | if (mdb_free) |
1117 | vfs_dq_release_reservation_block(inode, mdb_free); | |
d6014301 AK |
1118 | |
1119 | /* | |
1120 | * If we have done all the pending block allocations and if | |
1121 | * there aren't any writers on the inode, we can discard the | |
1122 | * inode's preallocations. | |
1123 | */ | |
1124 | if (!total && (atomic_read(&inode->i_writecount) == 0)) | |
1125 | ext4_discard_preallocations(inode); | |
12219aea AK |
1126 | } |
1127 | ||
f5ab0d1f | 1128 | /* |
12b7ac17 | 1129 | * The ext4_get_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 1130 | * and returns if the blocks are already mapped. |
f5ab0d1f | 1131 | * |
f5ab0d1f MC |
1132 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1133 | * and store the allocated blocks in the result buffer head and mark it | |
1134 | * mapped. | |
1135 | * | |
1136 | * If file type is extents based, it will call ext4_ext_get_blocks(), | |
e4d996ca | 1137 | * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping |
f5ab0d1f MC |
1138 | * based files |
1139 | * | |
1140 | * On success, it returns the number of blocks being mapped or allocate. | |
1141 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1142 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1143 | * the buffer head is mapped. | |
1144 | * | |
1145 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1146 | * that casem, buffer head is unmapped | |
1147 | * | |
1148 | * It returns the error in case of allocation failure. | |
1149 | */ | |
12b7ac17 TT |
1150 | int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block, |
1151 | unsigned int max_blocks, struct buffer_head *bh, | |
c2177057 | 1152 | int flags) |
0e855ac8 AK |
1153 | { |
1154 | int retval; | |
f5ab0d1f MC |
1155 | |
1156 | clear_buffer_mapped(bh); | |
2a8964d6 | 1157 | clear_buffer_unwritten(bh); |
f5ab0d1f | 1158 | |
4df3d265 | 1159 | /* |
b920c755 TT |
1160 | * Try to see if we can get the block without requesting a new |
1161 | * file system block. | |
4df3d265 AK |
1162 | */ |
1163 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1164 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { | |
1165 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
c2177057 | 1166 | bh, 0); |
0e855ac8 | 1167 | } else { |
e4d996ca | 1168 | retval = ext4_ind_get_blocks(handle, inode, block, max_blocks, |
c2177057 | 1169 | bh, 0); |
0e855ac8 | 1170 | } |
4df3d265 | 1171 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f MC |
1172 | |
1173 | /* If it is only a block(s) look up */ | |
c2177057 | 1174 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
1175 | return retval; |
1176 | ||
1177 | /* | |
1178 | * Returns if the blocks have already allocated | |
1179 | * | |
1180 | * Note that if blocks have been preallocated | |
1181 | * ext4_ext_get_block() returns th create = 0 | |
1182 | * with buffer head unmapped. | |
1183 | */ | |
1184 | if (retval > 0 && buffer_mapped(bh)) | |
4df3d265 AK |
1185 | return retval; |
1186 | ||
2a8964d6 AK |
1187 | /* |
1188 | * When we call get_blocks without the create flag, the | |
1189 | * BH_Unwritten flag could have gotten set if the blocks | |
1190 | * requested were part of a uninitialized extent. We need to | |
1191 | * clear this flag now that we are committed to convert all or | |
1192 | * part of the uninitialized extent to be an initialized | |
1193 | * extent. This is because we need to avoid the combination | |
1194 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
1195 | * set on the buffer_head. | |
1196 | */ | |
1197 | clear_buffer_unwritten(bh); | |
1198 | ||
4df3d265 | 1199 | /* |
f5ab0d1f MC |
1200 | * New blocks allocate and/or writing to uninitialized extent |
1201 | * will possibly result in updating i_data, so we take | |
1202 | * the write lock of i_data_sem, and call get_blocks() | |
1203 | * with create == 1 flag. | |
4df3d265 AK |
1204 | */ |
1205 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1206 | |
1207 | /* | |
1208 | * if the caller is from delayed allocation writeout path | |
1209 | * we have already reserved fs blocks for allocation | |
1210 | * let the underlying get_block() function know to | |
1211 | * avoid double accounting | |
1212 | */ | |
c2177057 | 1213 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1214 | EXT4_I(inode)->i_delalloc_reserved_flag = 1; |
4df3d265 AK |
1215 | /* |
1216 | * We need to check for EXT4 here because migrate | |
1217 | * could have changed the inode type in between | |
1218 | */ | |
0e855ac8 AK |
1219 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
1220 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
c2177057 | 1221 | bh, flags); |
0e855ac8 | 1222 | } else { |
e4d996ca | 1223 | retval = ext4_ind_get_blocks(handle, inode, block, |
c2177057 | 1224 | max_blocks, bh, flags); |
267e4db9 AK |
1225 | |
1226 | if (retval > 0 && buffer_new(bh)) { | |
1227 | /* | |
1228 | * We allocated new blocks which will result in | |
1229 | * i_data's format changing. Force the migrate | |
1230 | * to fail by clearing migrate flags | |
1231 | */ | |
1232 | EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags & | |
1233 | ~EXT4_EXT_MIGRATE; | |
1234 | } | |
0e855ac8 | 1235 | } |
d2a17637 | 1236 | |
c2177057 | 1237 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
d2a17637 MC |
1238 | EXT4_I(inode)->i_delalloc_reserved_flag = 0; |
1239 | /* | |
1240 | * Update reserved blocks/metadata blocks | |
1241 | * after successful block allocation | |
1242 | * which were deferred till now | |
1243 | */ | |
1244 | if ((retval > 0) && buffer_delay(bh)) | |
12219aea | 1245 | ext4_da_update_reserve_space(inode, retval); |
d2a17637 MC |
1246 | } |
1247 | ||
4df3d265 | 1248 | up_write((&EXT4_I(inode)->i_data_sem)); |
0e855ac8 AK |
1249 | return retval; |
1250 | } | |
1251 | ||
f3bd1f3f MC |
1252 | /* Maximum number of blocks we map for direct IO at once. */ |
1253 | #define DIO_MAX_BLOCKS 4096 | |
1254 | ||
6873fa0d ES |
1255 | int ext4_get_block(struct inode *inode, sector_t iblock, |
1256 | struct buffer_head *bh_result, int create) | |
ac27a0ec | 1257 | { |
3e4fdaf8 | 1258 | handle_t *handle = ext4_journal_current_handle(); |
7fb5409d | 1259 | int ret = 0, started = 0; |
ac27a0ec | 1260 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
f3bd1f3f | 1261 | int dio_credits; |
ac27a0ec | 1262 | |
7fb5409d JK |
1263 | if (create && !handle) { |
1264 | /* Direct IO write... */ | |
1265 | if (max_blocks > DIO_MAX_BLOCKS) | |
1266 | max_blocks = DIO_MAX_BLOCKS; | |
f3bd1f3f MC |
1267 | dio_credits = ext4_chunk_trans_blocks(inode, max_blocks); |
1268 | handle = ext4_journal_start(inode, dio_credits); | |
7fb5409d | 1269 | if (IS_ERR(handle)) { |
ac27a0ec | 1270 | ret = PTR_ERR(handle); |
7fb5409d | 1271 | goto out; |
ac27a0ec | 1272 | } |
7fb5409d | 1273 | started = 1; |
ac27a0ec DK |
1274 | } |
1275 | ||
12b7ac17 | 1276 | ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result, |
c2177057 | 1277 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
7fb5409d JK |
1278 | if (ret > 0) { |
1279 | bh_result->b_size = (ret << inode->i_blkbits); | |
1280 | ret = 0; | |
ac27a0ec | 1281 | } |
7fb5409d JK |
1282 | if (started) |
1283 | ext4_journal_stop(handle); | |
1284 | out: | |
ac27a0ec DK |
1285 | return ret; |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * `handle' can be NULL if create is zero | |
1290 | */ | |
617ba13b | 1291 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1292 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec DK |
1293 | { |
1294 | struct buffer_head dummy; | |
1295 | int fatal = 0, err; | |
c2177057 | 1296 | int flags = EXT4_GET_BLOCKS_EXTEND_DISKSIZE; |
ac27a0ec DK |
1297 | |
1298 | J_ASSERT(handle != NULL || create == 0); | |
1299 | ||
1300 | dummy.b_state = 0; | |
1301 | dummy.b_blocknr = -1000; | |
1302 | buffer_trace_init(&dummy.b_history); | |
c2177057 TT |
1303 | if (create) |
1304 | flags |= EXT4_GET_BLOCKS_CREATE; | |
1305 | err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags); | |
ac27a0ec | 1306 | /* |
c2177057 TT |
1307 | * ext4_get_blocks() returns number of blocks mapped. 0 in |
1308 | * case of a HOLE. | |
ac27a0ec DK |
1309 | */ |
1310 | if (err > 0) { | |
1311 | if (err > 1) | |
1312 | WARN_ON(1); | |
1313 | err = 0; | |
1314 | } | |
1315 | *errp = err; | |
1316 | if (!err && buffer_mapped(&dummy)) { | |
1317 | struct buffer_head *bh; | |
1318 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
1319 | if (!bh) { | |
1320 | *errp = -EIO; | |
1321 | goto err; | |
1322 | } | |
1323 | if (buffer_new(&dummy)) { | |
1324 | J_ASSERT(create != 0); | |
ac39849d | 1325 | J_ASSERT(handle != NULL); |
ac27a0ec DK |
1326 | |
1327 | /* | |
1328 | * Now that we do not always journal data, we should | |
1329 | * keep in mind whether this should always journal the | |
1330 | * new buffer as metadata. For now, regular file | |
617ba13b | 1331 | * writes use ext4_get_block instead, so it's not a |
ac27a0ec DK |
1332 | * problem. |
1333 | */ | |
1334 | lock_buffer(bh); | |
1335 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 1336 | fatal = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 1337 | if (!fatal && !buffer_uptodate(bh)) { |
af5bc92d | 1338 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
ac27a0ec DK |
1339 | set_buffer_uptodate(bh); |
1340 | } | |
1341 | unlock_buffer(bh); | |
0390131b FM |
1342 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
1343 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
1344 | if (!fatal) |
1345 | fatal = err; | |
1346 | } else { | |
1347 | BUFFER_TRACE(bh, "not a new buffer"); | |
1348 | } | |
1349 | if (fatal) { | |
1350 | *errp = fatal; | |
1351 | brelse(bh); | |
1352 | bh = NULL; | |
1353 | } | |
1354 | return bh; | |
1355 | } | |
1356 | err: | |
1357 | return NULL; | |
1358 | } | |
1359 | ||
617ba13b | 1360 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1361 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 1362 | { |
af5bc92d | 1363 | struct buffer_head *bh; |
ac27a0ec | 1364 | |
617ba13b | 1365 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1366 | if (!bh) |
1367 | return bh; | |
1368 | if (buffer_uptodate(bh)) | |
1369 | return bh; | |
1370 | ll_rw_block(READ_META, 1, &bh); | |
1371 | wait_on_buffer(bh); | |
1372 | if (buffer_uptodate(bh)) | |
1373 | return bh; | |
1374 | put_bh(bh); | |
1375 | *err = -EIO; | |
1376 | return NULL; | |
1377 | } | |
1378 | ||
af5bc92d TT |
1379 | static int walk_page_buffers(handle_t *handle, |
1380 | struct buffer_head *head, | |
1381 | unsigned from, | |
1382 | unsigned to, | |
1383 | int *partial, | |
1384 | int (*fn)(handle_t *handle, | |
1385 | struct buffer_head *bh)) | |
ac27a0ec DK |
1386 | { |
1387 | struct buffer_head *bh; | |
1388 | unsigned block_start, block_end; | |
1389 | unsigned blocksize = head->b_size; | |
1390 | int err, ret = 0; | |
1391 | struct buffer_head *next; | |
1392 | ||
af5bc92d TT |
1393 | for (bh = head, block_start = 0; |
1394 | ret == 0 && (bh != head || !block_start); | |
1395 | block_start = block_end, bh = next) | |
ac27a0ec DK |
1396 | { |
1397 | next = bh->b_this_page; | |
1398 | block_end = block_start + blocksize; | |
1399 | if (block_end <= from || block_start >= to) { | |
1400 | if (partial && !buffer_uptodate(bh)) | |
1401 | *partial = 1; | |
1402 | continue; | |
1403 | } | |
1404 | err = (*fn)(handle, bh); | |
1405 | if (!ret) | |
1406 | ret = err; | |
1407 | } | |
1408 | return ret; | |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * To preserve ordering, it is essential that the hole instantiation and | |
1413 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1414 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1415 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1416 | * prepare_write() is the right place. |
1417 | * | |
617ba13b MC |
1418 | * Also, this function can nest inside ext4_writepage() -> |
1419 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1420 | * has generated enough buffer credits to do the whole page. So we won't |
1421 | * block on the journal in that case, which is good, because the caller may | |
1422 | * be PF_MEMALLOC. | |
1423 | * | |
617ba13b | 1424 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1425 | * quota file writes. If we were to commit the transaction while thus |
1426 | * reentered, there can be a deadlock - we would be holding a quota | |
1427 | * lock, and the commit would never complete if another thread had a | |
1428 | * transaction open and was blocking on the quota lock - a ranking | |
1429 | * violation. | |
1430 | * | |
dab291af | 1431 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1432 | * will _not_ run commit under these circumstances because handle->h_ref |
1433 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1434 | * write. | |
1435 | */ | |
1436 | static int do_journal_get_write_access(handle_t *handle, | |
1437 | struct buffer_head *bh) | |
1438 | { | |
1439 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1440 | return 0; | |
617ba13b | 1441 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1442 | } |
1443 | ||
bfc1af65 NP |
1444 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
1445 | loff_t pos, unsigned len, unsigned flags, | |
1446 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1447 | { |
af5bc92d | 1448 | struct inode *inode = mapping->host; |
7479d2b9 | 1449 | int ret, needed_blocks = ext4_writepage_trans_blocks(inode); |
ac27a0ec DK |
1450 | handle_t *handle; |
1451 | int retries = 0; | |
af5bc92d | 1452 | struct page *page; |
bfc1af65 | 1453 | pgoff_t index; |
af5bc92d | 1454 | unsigned from, to; |
bfc1af65 | 1455 | |
ba80b101 TT |
1456 | trace_mark(ext4_write_begin, |
1457 | "dev %s ino %lu pos %llu len %u flags %u", | |
1458 | inode->i_sb->s_id, inode->i_ino, | |
1459 | (unsigned long long) pos, len, flags); | |
bfc1af65 | 1460 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1461 | from = pos & (PAGE_CACHE_SIZE - 1); |
1462 | to = from + len; | |
ac27a0ec DK |
1463 | |
1464 | retry: | |
af5bc92d TT |
1465 | handle = ext4_journal_start(inode, needed_blocks); |
1466 | if (IS_ERR(handle)) { | |
1467 | ret = PTR_ERR(handle); | |
1468 | goto out; | |
7479d2b9 | 1469 | } |
ac27a0ec | 1470 | |
ebd3610b JK |
1471 | /* We cannot recurse into the filesystem as the transaction is already |
1472 | * started */ | |
1473 | flags |= AOP_FLAG_NOFS; | |
1474 | ||
54566b2c | 1475 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
1476 | if (!page) { |
1477 | ext4_journal_stop(handle); | |
1478 | ret = -ENOMEM; | |
1479 | goto out; | |
1480 | } | |
1481 | *pagep = page; | |
1482 | ||
bfc1af65 | 1483 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
ebd3610b | 1484 | ext4_get_block); |
bfc1af65 NP |
1485 | |
1486 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1487 | ret = walk_page_buffers(handle, page_buffers(page), |
1488 | from, to, NULL, do_journal_get_write_access); | |
1489 | } | |
bfc1af65 NP |
1490 | |
1491 | if (ret) { | |
af5bc92d | 1492 | unlock_page(page); |
cf108bca | 1493 | ext4_journal_stop(handle); |
af5bc92d | 1494 | page_cache_release(page); |
ae4d5372 AK |
1495 | /* |
1496 | * block_write_begin may have instantiated a few blocks | |
1497 | * outside i_size. Trim these off again. Don't need | |
1498 | * i_size_read because we hold i_mutex. | |
1499 | */ | |
1500 | if (pos + len > inode->i_size) | |
1501 | vmtruncate(inode, inode->i_size); | |
bfc1af65 NP |
1502 | } |
1503 | ||
617ba13b | 1504 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1505 | goto retry; |
7479d2b9 | 1506 | out: |
ac27a0ec DK |
1507 | return ret; |
1508 | } | |
1509 | ||
bfc1af65 NP |
1510 | /* For write_end() in data=journal mode */ |
1511 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1512 | { |
1513 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1514 | return 0; | |
1515 | set_buffer_uptodate(bh); | |
0390131b | 1516 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1517 | } |
1518 | ||
1519 | /* | |
1520 | * We need to pick up the new inode size which generic_commit_write gave us | |
1521 | * `file' can be NULL - eg, when called from page_symlink(). | |
1522 | * | |
617ba13b | 1523 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1524 | * buffers are managed internally. |
1525 | */ | |
bfc1af65 NP |
1526 | static int ext4_ordered_write_end(struct file *file, |
1527 | struct address_space *mapping, | |
1528 | loff_t pos, unsigned len, unsigned copied, | |
1529 | struct page *page, void *fsdata) | |
ac27a0ec | 1530 | { |
617ba13b | 1531 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1532 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1533 | int ret = 0, ret2; |
1534 | ||
ba80b101 TT |
1535 | trace_mark(ext4_ordered_write_end, |
1536 | "dev %s ino %lu pos %llu len %u copied %u", | |
1537 | inode->i_sb->s_id, inode->i_ino, | |
1538 | (unsigned long long) pos, len, copied); | |
678aaf48 | 1539 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1540 | |
1541 | if (ret == 0) { | |
ac27a0ec DK |
1542 | loff_t new_i_size; |
1543 | ||
bfc1af65 | 1544 | new_i_size = pos + copied; |
cf17fea6 AK |
1545 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1546 | ext4_update_i_disksize(inode, new_i_size); | |
1547 | /* We need to mark inode dirty even if | |
1548 | * new_i_size is less that inode->i_size | |
1549 | * bu greater than i_disksize.(hint delalloc) | |
1550 | */ | |
1551 | ext4_mark_inode_dirty(handle, inode); | |
1552 | } | |
1553 | ||
cf108bca | 1554 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1555 | page, fsdata); |
f8a87d89 RK |
1556 | copied = ret2; |
1557 | if (ret2 < 0) | |
1558 | ret = ret2; | |
ac27a0ec | 1559 | } |
617ba13b | 1560 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1561 | if (!ret) |
1562 | ret = ret2; | |
bfc1af65 NP |
1563 | |
1564 | return ret ? ret : copied; | |
ac27a0ec DK |
1565 | } |
1566 | ||
bfc1af65 NP |
1567 | static int ext4_writeback_write_end(struct file *file, |
1568 | struct address_space *mapping, | |
1569 | loff_t pos, unsigned len, unsigned copied, | |
1570 | struct page *page, void *fsdata) | |
ac27a0ec | 1571 | { |
617ba13b | 1572 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1573 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1574 | int ret = 0, ret2; |
1575 | loff_t new_i_size; | |
1576 | ||
ba80b101 TT |
1577 | trace_mark(ext4_writeback_write_end, |
1578 | "dev %s ino %lu pos %llu len %u copied %u", | |
1579 | inode->i_sb->s_id, inode->i_ino, | |
1580 | (unsigned long long) pos, len, copied); | |
bfc1af65 | 1581 | new_i_size = pos + copied; |
cf17fea6 AK |
1582 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1583 | ext4_update_i_disksize(inode, new_i_size); | |
1584 | /* We need to mark inode dirty even if | |
1585 | * new_i_size is less that inode->i_size | |
1586 | * bu greater than i_disksize.(hint delalloc) | |
1587 | */ | |
1588 | ext4_mark_inode_dirty(handle, inode); | |
1589 | } | |
ac27a0ec | 1590 | |
cf108bca | 1591 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1592 | page, fsdata); |
f8a87d89 RK |
1593 | copied = ret2; |
1594 | if (ret2 < 0) | |
1595 | ret = ret2; | |
ac27a0ec | 1596 | |
617ba13b | 1597 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1598 | if (!ret) |
1599 | ret = ret2; | |
bfc1af65 NP |
1600 | |
1601 | return ret ? ret : copied; | |
ac27a0ec DK |
1602 | } |
1603 | ||
bfc1af65 NP |
1604 | static int ext4_journalled_write_end(struct file *file, |
1605 | struct address_space *mapping, | |
1606 | loff_t pos, unsigned len, unsigned copied, | |
1607 | struct page *page, void *fsdata) | |
ac27a0ec | 1608 | { |
617ba13b | 1609 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1610 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1611 | int ret = 0, ret2; |
1612 | int partial = 0; | |
bfc1af65 | 1613 | unsigned from, to; |
cf17fea6 | 1614 | loff_t new_i_size; |
ac27a0ec | 1615 | |
ba80b101 TT |
1616 | trace_mark(ext4_journalled_write_end, |
1617 | "dev %s ino %lu pos %llu len %u copied %u", | |
1618 | inode->i_sb->s_id, inode->i_ino, | |
1619 | (unsigned long long) pos, len, copied); | |
bfc1af65 NP |
1620 | from = pos & (PAGE_CACHE_SIZE - 1); |
1621 | to = from + len; | |
1622 | ||
1623 | if (copied < len) { | |
1624 | if (!PageUptodate(page)) | |
1625 | copied = 0; | |
1626 | page_zero_new_buffers(page, from+copied, to); | |
1627 | } | |
ac27a0ec DK |
1628 | |
1629 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1630 | to, &partial, write_end_fn); |
ac27a0ec DK |
1631 | if (!partial) |
1632 | SetPageUptodate(page); | |
cf17fea6 AK |
1633 | new_i_size = pos + copied; |
1634 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1635 | i_size_write(inode, pos+copied); |
617ba13b | 1636 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
cf17fea6 AK |
1637 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1638 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1639 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1640 | if (!ret) |
1641 | ret = ret2; | |
1642 | } | |
bfc1af65 | 1643 | |
cf108bca | 1644 | unlock_page(page); |
617ba13b | 1645 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1646 | if (!ret) |
1647 | ret = ret2; | |
bfc1af65 NP |
1648 | page_cache_release(page); |
1649 | ||
1650 | return ret ? ret : copied; | |
ac27a0ec | 1651 | } |
d2a17637 MC |
1652 | |
1653 | static int ext4_da_reserve_space(struct inode *inode, int nrblocks) | |
1654 | { | |
030ba6bc | 1655 | int retries = 0; |
60e58e0f MC |
1656 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1657 | unsigned long md_needed, mdblocks, total = 0; | |
d2a17637 MC |
1658 | |
1659 | /* | |
1660 | * recalculate the amount of metadata blocks to reserve | |
1661 | * in order to allocate nrblocks | |
1662 | * worse case is one extent per block | |
1663 | */ | |
030ba6bc | 1664 | repeat: |
d2a17637 MC |
1665 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
1666 | total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks; | |
1667 | mdblocks = ext4_calc_metadata_amount(inode, total); | |
1668 | BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks); | |
1669 | ||
1670 | md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks; | |
1671 | total = md_needed + nrblocks; | |
1672 | ||
60e58e0f MC |
1673 | /* |
1674 | * Make quota reservation here to prevent quota overflow | |
1675 | * later. Real quota accounting is done at pages writeout | |
1676 | * time. | |
1677 | */ | |
1678 | if (vfs_dq_reserve_block(inode, total)) { | |
1679 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1680 | return -EDQUOT; | |
1681 | } | |
1682 | ||
a30d542a | 1683 | if (ext4_claim_free_blocks(sbi, total)) { |
d2a17637 | 1684 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
030ba6bc AK |
1685 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1686 | yield(); | |
1687 | goto repeat; | |
1688 | } | |
60e58e0f | 1689 | vfs_dq_release_reservation_block(inode, total); |
d2a17637 MC |
1690 | return -ENOSPC; |
1691 | } | |
d2a17637 MC |
1692 | EXT4_I(inode)->i_reserved_data_blocks += nrblocks; |
1693 | EXT4_I(inode)->i_reserved_meta_blocks = mdblocks; | |
1694 | ||
1695 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1696 | return 0; /* success */ | |
1697 | } | |
1698 | ||
12219aea | 1699 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1700 | { |
1701 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1702 | int total, mdb, mdb_free, release; | |
1703 | ||
cd213226 MC |
1704 | if (!to_free) |
1705 | return; /* Nothing to release, exit */ | |
1706 | ||
d2a17637 | 1707 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 MC |
1708 | |
1709 | if (!EXT4_I(inode)->i_reserved_data_blocks) { | |
1710 | /* | |
1711 | * if there is no reserved blocks, but we try to free some | |
1712 | * then the counter is messed up somewhere. | |
1713 | * but since this function is called from invalidate | |
1714 | * page, it's harmless to return without any action | |
1715 | */ | |
1716 | printk(KERN_INFO "ext4 delalloc try to release %d reserved " | |
1717 | "blocks for inode %lu, but there is no reserved " | |
1718 | "data blocks\n", to_free, inode->i_ino); | |
1719 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1720 | return; | |
1721 | } | |
1722 | ||
d2a17637 | 1723 | /* recalculate the number of metablocks still need to be reserved */ |
12219aea | 1724 | total = EXT4_I(inode)->i_reserved_data_blocks - to_free; |
d2a17637 MC |
1725 | mdb = ext4_calc_metadata_amount(inode, total); |
1726 | ||
1727 | /* figure out how many metablocks to release */ | |
1728 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1729 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1730 | ||
d2a17637 MC |
1731 | release = to_free + mdb_free; |
1732 | ||
6bc6e63f AK |
1733 | /* update fs dirty blocks counter for truncate case */ |
1734 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, release); | |
d2a17637 MC |
1735 | |
1736 | /* update per-inode reservations */ | |
12219aea AK |
1737 | BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks); |
1738 | EXT4_I(inode)->i_reserved_data_blocks -= to_free; | |
d2a17637 MC |
1739 | |
1740 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1741 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
d2a17637 | 1742 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f MC |
1743 | |
1744 | vfs_dq_release_reservation_block(inode, release); | |
d2a17637 MC |
1745 | } |
1746 | ||
1747 | static void ext4_da_page_release_reservation(struct page *page, | |
1748 | unsigned long offset) | |
1749 | { | |
1750 | int to_release = 0; | |
1751 | struct buffer_head *head, *bh; | |
1752 | unsigned int curr_off = 0; | |
1753 | ||
1754 | head = page_buffers(page); | |
1755 | bh = head; | |
1756 | do { | |
1757 | unsigned int next_off = curr_off + bh->b_size; | |
1758 | ||
1759 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1760 | to_release++; | |
1761 | clear_buffer_delay(bh); | |
1762 | } | |
1763 | curr_off = next_off; | |
1764 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 1765 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 1766 | } |
ac27a0ec | 1767 | |
64769240 AT |
1768 | /* |
1769 | * Delayed allocation stuff | |
1770 | */ | |
1771 | ||
1772 | struct mpage_da_data { | |
1773 | struct inode *inode; | |
8dc207c0 TT |
1774 | sector_t b_blocknr; /* start block number of extent */ |
1775 | size_t b_size; /* size of extent */ | |
1776 | unsigned long b_state; /* state of the extent */ | |
64769240 | 1777 | unsigned long first_page, next_page; /* extent of pages */ |
64769240 | 1778 | struct writeback_control *wbc; |
a1d6cc56 | 1779 | int io_done; |
498e5f24 | 1780 | int pages_written; |
df22291f | 1781 | int retval; |
64769240 AT |
1782 | }; |
1783 | ||
1784 | /* | |
1785 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1786 | * them with writepage() call back |
64769240 AT |
1787 | * |
1788 | * @mpd->inode: inode | |
1789 | * @mpd->first_page: first page of the extent | |
1790 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1791 | * |
1792 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1793 | * to be allocated. this may be wrong if allocation failed. | |
1794 | * | |
1795 | * As pages are already locked by write_cache_pages(), we can't use it | |
1796 | */ | |
1797 | static int mpage_da_submit_io(struct mpage_da_data *mpd) | |
1798 | { | |
22208ded | 1799 | long pages_skipped; |
791b7f08 AK |
1800 | struct pagevec pvec; |
1801 | unsigned long index, end; | |
1802 | int ret = 0, err, nr_pages, i; | |
1803 | struct inode *inode = mpd->inode; | |
1804 | struct address_space *mapping = inode->i_mapping; | |
64769240 AT |
1805 | |
1806 | BUG_ON(mpd->next_page <= mpd->first_page); | |
791b7f08 AK |
1807 | /* |
1808 | * We need to start from the first_page to the next_page - 1 | |
1809 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1810 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1811 | * at the currently mapped buffer_heads. |
1812 | */ | |
64769240 AT |
1813 | index = mpd->first_page; |
1814 | end = mpd->next_page - 1; | |
1815 | ||
791b7f08 | 1816 | pagevec_init(&pvec, 0); |
64769240 | 1817 | while (index <= end) { |
791b7f08 | 1818 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1819 | if (nr_pages == 0) |
1820 | break; | |
1821 | for (i = 0; i < nr_pages; i++) { | |
1822 | struct page *page = pvec.pages[i]; | |
1823 | ||
791b7f08 AK |
1824 | index = page->index; |
1825 | if (index > end) | |
1826 | break; | |
1827 | index++; | |
1828 | ||
1829 | BUG_ON(!PageLocked(page)); | |
1830 | BUG_ON(PageWriteback(page)); | |
1831 | ||
22208ded | 1832 | pages_skipped = mpd->wbc->pages_skipped; |
a1d6cc56 | 1833 | err = mapping->a_ops->writepage(page, mpd->wbc); |
22208ded AK |
1834 | if (!err && (pages_skipped == mpd->wbc->pages_skipped)) |
1835 | /* | |
1836 | * have successfully written the page | |
1837 | * without skipping the same | |
1838 | */ | |
a1d6cc56 | 1839 | mpd->pages_written++; |
64769240 AT |
1840 | /* |
1841 | * In error case, we have to continue because | |
1842 | * remaining pages are still locked | |
1843 | * XXX: unlock and re-dirty them? | |
1844 | */ | |
1845 | if (ret == 0) | |
1846 | ret = err; | |
1847 | } | |
1848 | pagevec_release(&pvec); | |
1849 | } | |
64769240 AT |
1850 | return ret; |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers | |
1855 | * | |
1856 | * @mpd->inode - inode to walk through | |
1857 | * @exbh->b_blocknr - first block on a disk | |
1858 | * @exbh->b_size - amount of space in bytes | |
1859 | * @logical - first logical block to start assignment with | |
1860 | * | |
1861 | * the function goes through all passed space and put actual disk | |
29fa89d0 | 1862 | * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten |
64769240 AT |
1863 | */ |
1864 | static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical, | |
1865 | struct buffer_head *exbh) | |
1866 | { | |
1867 | struct inode *inode = mpd->inode; | |
1868 | struct address_space *mapping = inode->i_mapping; | |
1869 | int blocks = exbh->b_size >> inode->i_blkbits; | |
1870 | sector_t pblock = exbh->b_blocknr, cur_logical; | |
1871 | struct buffer_head *head, *bh; | |
a1d6cc56 | 1872 | pgoff_t index, end; |
64769240 AT |
1873 | struct pagevec pvec; |
1874 | int nr_pages, i; | |
1875 | ||
1876 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1877 | end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1878 | cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1879 | ||
1880 | pagevec_init(&pvec, 0); | |
1881 | ||
1882 | while (index <= end) { | |
1883 | /* XXX: optimize tail */ | |
1884 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1885 | if (nr_pages == 0) | |
1886 | break; | |
1887 | for (i = 0; i < nr_pages; i++) { | |
1888 | struct page *page = pvec.pages[i]; | |
1889 | ||
1890 | index = page->index; | |
1891 | if (index > end) | |
1892 | break; | |
1893 | index++; | |
1894 | ||
1895 | BUG_ON(!PageLocked(page)); | |
1896 | BUG_ON(PageWriteback(page)); | |
1897 | BUG_ON(!page_has_buffers(page)); | |
1898 | ||
1899 | bh = page_buffers(page); | |
1900 | head = bh; | |
1901 | ||
1902 | /* skip blocks out of the range */ | |
1903 | do { | |
1904 | if (cur_logical >= logical) | |
1905 | break; | |
1906 | cur_logical++; | |
1907 | } while ((bh = bh->b_this_page) != head); | |
1908 | ||
1909 | do { | |
1910 | if (cur_logical >= logical + blocks) | |
1911 | break; | |
29fa89d0 AK |
1912 | |
1913 | if (buffer_delay(bh) || | |
1914 | buffer_unwritten(bh)) { | |
1915 | ||
1916 | BUG_ON(bh->b_bdev != inode->i_sb->s_bdev); | |
1917 | ||
1918 | if (buffer_delay(bh)) { | |
1919 | clear_buffer_delay(bh); | |
1920 | bh->b_blocknr = pblock; | |
1921 | } else { | |
1922 | /* | |
1923 | * unwritten already should have | |
1924 | * blocknr assigned. Verify that | |
1925 | */ | |
1926 | clear_buffer_unwritten(bh); | |
1927 | BUG_ON(bh->b_blocknr != pblock); | |
1928 | } | |
1929 | ||
61628a3f | 1930 | } else if (buffer_mapped(bh)) |
64769240 | 1931 | BUG_ON(bh->b_blocknr != pblock); |
64769240 AT |
1932 | |
1933 | cur_logical++; | |
1934 | pblock++; | |
1935 | } while ((bh = bh->b_this_page) != head); | |
1936 | } | |
1937 | pagevec_release(&pvec); | |
1938 | } | |
1939 | } | |
1940 | ||
1941 | ||
1942 | /* | |
1943 | * __unmap_underlying_blocks - just a helper function to unmap | |
1944 | * set of blocks described by @bh | |
1945 | */ | |
1946 | static inline void __unmap_underlying_blocks(struct inode *inode, | |
1947 | struct buffer_head *bh) | |
1948 | { | |
1949 | struct block_device *bdev = inode->i_sb->s_bdev; | |
1950 | int blocks, i; | |
1951 | ||
1952 | blocks = bh->b_size >> inode->i_blkbits; | |
1953 | for (i = 0; i < blocks; i++) | |
1954 | unmap_underlying_metadata(bdev, bh->b_blocknr + i); | |
1955 | } | |
1956 | ||
c4a0c46e AK |
1957 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, |
1958 | sector_t logical, long blk_cnt) | |
1959 | { | |
1960 | int nr_pages, i; | |
1961 | pgoff_t index, end; | |
1962 | struct pagevec pvec; | |
1963 | struct inode *inode = mpd->inode; | |
1964 | struct address_space *mapping = inode->i_mapping; | |
1965 | ||
1966 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1967 | end = (logical + blk_cnt - 1) >> | |
1968 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1969 | while (index <= end) { | |
1970 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1971 | if (nr_pages == 0) | |
1972 | break; | |
1973 | for (i = 0; i < nr_pages; i++) { | |
1974 | struct page *page = pvec.pages[i]; | |
1975 | index = page->index; | |
1976 | if (index > end) | |
1977 | break; | |
1978 | index++; | |
1979 | ||
1980 | BUG_ON(!PageLocked(page)); | |
1981 | BUG_ON(PageWriteback(page)); | |
1982 | block_invalidatepage(page, 0); | |
1983 | ClearPageUptodate(page); | |
1984 | unlock_page(page); | |
1985 | } | |
1986 | } | |
1987 | return; | |
1988 | } | |
1989 | ||
df22291f AK |
1990 | static void ext4_print_free_blocks(struct inode *inode) |
1991 | { | |
1992 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1993 | printk(KERN_EMERG "Total free blocks count %lld\n", | |
1994 | ext4_count_free_blocks(inode->i_sb)); | |
1995 | printk(KERN_EMERG "Free/Dirty block details\n"); | |
1996 | printk(KERN_EMERG "free_blocks=%lld\n", | |
8f72fbdf | 1997 | (long long)percpu_counter_sum(&sbi->s_freeblocks_counter)); |
df22291f | 1998 | printk(KERN_EMERG "dirty_blocks=%lld\n", |
8f72fbdf | 1999 | (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter)); |
df22291f | 2000 | printk(KERN_EMERG "Block reservation details\n"); |
498e5f24 | 2001 | printk(KERN_EMERG "i_reserved_data_blocks=%u\n", |
df22291f | 2002 | EXT4_I(inode)->i_reserved_data_blocks); |
498e5f24 | 2003 | printk(KERN_EMERG "i_reserved_meta_blocks=%u\n", |
df22291f AK |
2004 | EXT4_I(inode)->i_reserved_meta_blocks); |
2005 | return; | |
2006 | } | |
2007 | ||
64769240 AT |
2008 | /* |
2009 | * mpage_da_map_blocks - go through given space | |
2010 | * | |
8dc207c0 | 2011 | * @mpd - bh describing space |
64769240 AT |
2012 | * |
2013 | * The function skips space we know is already mapped to disk blocks. | |
2014 | * | |
64769240 | 2015 | */ |
ed5bde0b | 2016 | static int mpage_da_map_blocks(struct mpage_da_data *mpd) |
64769240 | 2017 | { |
2fa3cdfb | 2018 | int err, blks; |
030ba6bc | 2019 | struct buffer_head new; |
2fa3cdfb TT |
2020 | sector_t next = mpd->b_blocknr; |
2021 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
2022 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
2023 | handle_t *handle = NULL; | |
64769240 AT |
2024 | |
2025 | /* | |
2026 | * We consider only non-mapped and non-allocated blocks | |
2027 | */ | |
8dc207c0 | 2028 | if ((mpd->b_state & (1 << BH_Mapped)) && |
29fa89d0 AK |
2029 | !(mpd->b_state & (1 << BH_Delay)) && |
2030 | !(mpd->b_state & (1 << BH_Unwritten))) | |
c4a0c46e | 2031 | return 0; |
2fa3cdfb TT |
2032 | |
2033 | /* | |
2034 | * If we didn't accumulate anything to write simply return | |
2035 | */ | |
2036 | if (!mpd->b_size) | |
2037 | return 0; | |
2038 | ||
2039 | handle = ext4_journal_current_handle(); | |
2040 | BUG_ON(!handle); | |
2041 | ||
79ffab34 AK |
2042 | /* |
2043 | * We need to make sure the BH_Delay flag is passed down to | |
c2177057 TT |
2044 | * ext4_da_get_block_write(), since it calls ext4_get_blocks() |
2045 | * with the EXT4_GET_BLOCKS_DELALLOC_RESERVE flag. This flag | |
2046 | * causes ext4_get_blocks() to call | |
79ffab34 AK |
2047 | * ext4_da_update_reserve_space() if the passed buffer head |
2048 | * has the BH_Delay flag set. In the future, once we clean up | |
c2177057 TT |
2049 | * the interfaces to ext4_get_blocks(), we should pass in a |
2050 | * separate flag which requests that the delayed allocation | |
79ffab34 AK |
2051 | * statistics should be updated, instead of depending on the |
2052 | * state information getting passed down via the map_bh's | |
c2177057 TT |
2053 | * state bitmasks plus the magic |
2054 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE flag. | |
79ffab34 AK |
2055 | */ |
2056 | new.b_state = mpd->b_state & (1 << BH_Delay); | |
2fa3cdfb TT |
2057 | blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks, |
2058 | &new, EXT4_GET_BLOCKS_CREATE| | |
2059 | EXT4_GET_BLOCKS_DELALLOC_RESERVE); | |
2060 | if (blks < 0) { | |
2061 | err = blks; | |
ed5bde0b TT |
2062 | /* |
2063 | * If get block returns with error we simply | |
2064 | * return. Later writepage will redirty the page and | |
2065 | * writepages will find the dirty page again | |
c4a0c46e AK |
2066 | */ |
2067 | if (err == -EAGAIN) | |
2068 | return 0; | |
df22291f AK |
2069 | |
2070 | if (err == -ENOSPC && | |
ed5bde0b | 2071 | ext4_count_free_blocks(mpd->inode->i_sb)) { |
df22291f AK |
2072 | mpd->retval = err; |
2073 | return 0; | |
2074 | } | |
2075 | ||
c4a0c46e | 2076 | /* |
ed5bde0b TT |
2077 | * get block failure will cause us to loop in |
2078 | * writepages, because a_ops->writepage won't be able | |
2079 | * to make progress. The page will be redirtied by | |
2080 | * writepage and writepages will again try to write | |
2081 | * the same. | |
c4a0c46e AK |
2082 | */ |
2083 | printk(KERN_EMERG "%s block allocation failed for inode %lu " | |
2084 | "at logical offset %llu with max blocks " | |
2085 | "%zd with error %d\n", | |
2086 | __func__, mpd->inode->i_ino, | |
2087 | (unsigned long long)next, | |
8dc207c0 | 2088 | mpd->b_size >> mpd->inode->i_blkbits, err); |
c4a0c46e AK |
2089 | printk(KERN_EMERG "This should not happen.!! " |
2090 | "Data will be lost\n"); | |
030ba6bc | 2091 | if (err == -ENOSPC) { |
df22291f | 2092 | ext4_print_free_blocks(mpd->inode); |
030ba6bc | 2093 | } |
2fa3cdfb | 2094 | /* invalidate all the pages */ |
c4a0c46e | 2095 | ext4_da_block_invalidatepages(mpd, next, |
8dc207c0 | 2096 | mpd->b_size >> mpd->inode->i_blkbits); |
c4a0c46e AK |
2097 | return err; |
2098 | } | |
2fa3cdfb TT |
2099 | BUG_ON(blks == 0); |
2100 | ||
2101 | new.b_size = (blks << mpd->inode->i_blkbits); | |
64769240 | 2102 | |
a1d6cc56 AK |
2103 | if (buffer_new(&new)) |
2104 | __unmap_underlying_blocks(mpd->inode, &new); | |
64769240 | 2105 | |
a1d6cc56 AK |
2106 | /* |
2107 | * If blocks are delayed marked, we need to | |
2108 | * put actual blocknr and drop delayed bit | |
2109 | */ | |
8dc207c0 TT |
2110 | if ((mpd->b_state & (1 << BH_Delay)) || |
2111 | (mpd->b_state & (1 << BH_Unwritten))) | |
a1d6cc56 | 2112 | mpage_put_bnr_to_bhs(mpd, next, &new); |
64769240 | 2113 | |
2fa3cdfb TT |
2114 | if (ext4_should_order_data(mpd->inode)) { |
2115 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
2116 | if (err) | |
2117 | return err; | |
2118 | } | |
2119 | ||
2120 | /* | |
2121 | * Update on-disk size along with block allocation we don't | |
2122 | * use EXT4_GET_BLOCKS_EXTEND_DISKSIZE as size may change | |
2123 | * within already allocated block -bzzz | |
2124 | */ | |
2125 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
2126 | if (disksize > i_size_read(mpd->inode)) | |
2127 | disksize = i_size_read(mpd->inode); | |
2128 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
2129 | ext4_update_i_disksize(mpd->inode, disksize); | |
2130 | return ext4_mark_inode_dirty(handle, mpd->inode); | |
2131 | } | |
2132 | ||
c4a0c46e | 2133 | return 0; |
64769240 AT |
2134 | } |
2135 | ||
bf068ee2 AK |
2136 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
2137 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
2138 | |
2139 | /* | |
2140 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
2141 | * | |
2142 | * @mpd->lbh - extent of blocks | |
2143 | * @logical - logical number of the block in the file | |
2144 | * @bh - bh of the block (used to access block's state) | |
2145 | * | |
2146 | * the function is used to collect contig. blocks in same state | |
2147 | */ | |
2148 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
2149 | sector_t logical, size_t b_size, |
2150 | unsigned long b_state) | |
64769240 | 2151 | { |
64769240 | 2152 | sector_t next; |
8dc207c0 | 2153 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 2154 | |
525f4ed8 MC |
2155 | /* check if thereserved journal credits might overflow */ |
2156 | if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) { | |
2157 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { | |
2158 | /* | |
2159 | * With non-extent format we are limited by the journal | |
2160 | * credit available. Total credit needed to insert | |
2161 | * nrblocks contiguous blocks is dependent on the | |
2162 | * nrblocks. So limit nrblocks. | |
2163 | */ | |
2164 | goto flush_it; | |
2165 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
2166 | EXT4_MAX_TRANS_DATA) { | |
2167 | /* | |
2168 | * Adding the new buffer_head would make it cross the | |
2169 | * allowed limit for which we have journal credit | |
2170 | * reserved. So limit the new bh->b_size | |
2171 | */ | |
2172 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
2173 | mpd->inode->i_blkbits; | |
2174 | /* we will do mpage_da_submit_io in the next loop */ | |
2175 | } | |
2176 | } | |
64769240 AT |
2177 | /* |
2178 | * First block in the extent | |
2179 | */ | |
8dc207c0 TT |
2180 | if (mpd->b_size == 0) { |
2181 | mpd->b_blocknr = logical; | |
2182 | mpd->b_size = b_size; | |
2183 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
2184 | return; |
2185 | } | |
2186 | ||
8dc207c0 | 2187 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
2188 | /* |
2189 | * Can we merge the block to our big extent? | |
2190 | */ | |
8dc207c0 TT |
2191 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
2192 | mpd->b_size += b_size; | |
64769240 AT |
2193 | return; |
2194 | } | |
2195 | ||
525f4ed8 | 2196 | flush_it: |
64769240 AT |
2197 | /* |
2198 | * We couldn't merge the block to our extent, so we | |
2199 | * need to flush current extent and start new one | |
2200 | */ | |
c4a0c46e AK |
2201 | if (mpage_da_map_blocks(mpd) == 0) |
2202 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2203 | mpd->io_done = 1; |
2204 | return; | |
64769240 AT |
2205 | } |
2206 | ||
29fa89d0 AK |
2207 | static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh) |
2208 | { | |
2209 | /* | |
2210 | * unmapped buffer is possible for holes. | |
2211 | * delay buffer is possible with delayed allocation. | |
2212 | * We also need to consider unwritten buffer as unmapped. | |
2213 | */ | |
2214 | return (!buffer_mapped(bh) || buffer_delay(bh) || | |
2215 | buffer_unwritten(bh)) && buffer_dirty(bh); | |
2216 | } | |
2217 | ||
64769240 AT |
2218 | /* |
2219 | * __mpage_da_writepage - finds extent of pages and blocks | |
2220 | * | |
2221 | * @page: page to consider | |
2222 | * @wbc: not used, we just follow rules | |
2223 | * @data: context | |
2224 | * | |
2225 | * The function finds extents of pages and scan them for all blocks. | |
2226 | */ | |
2227 | static int __mpage_da_writepage(struct page *page, | |
2228 | struct writeback_control *wbc, void *data) | |
2229 | { | |
2230 | struct mpage_da_data *mpd = data; | |
2231 | struct inode *inode = mpd->inode; | |
8dc207c0 | 2232 | struct buffer_head *bh, *head; |
64769240 AT |
2233 | sector_t logical; |
2234 | ||
a1d6cc56 AK |
2235 | if (mpd->io_done) { |
2236 | /* | |
2237 | * Rest of the page in the page_vec | |
2238 | * redirty then and skip then. We will | |
2239 | * try to to write them again after | |
2240 | * starting a new transaction | |
2241 | */ | |
2242 | redirty_page_for_writepage(wbc, page); | |
2243 | unlock_page(page); | |
2244 | return MPAGE_DA_EXTENT_TAIL; | |
2245 | } | |
64769240 AT |
2246 | /* |
2247 | * Can we merge this page to current extent? | |
2248 | */ | |
2249 | if (mpd->next_page != page->index) { | |
2250 | /* | |
2251 | * Nope, we can't. So, we map non-allocated blocks | |
a1d6cc56 | 2252 | * and start IO on them using writepage() |
64769240 AT |
2253 | */ |
2254 | if (mpd->next_page != mpd->first_page) { | |
c4a0c46e AK |
2255 | if (mpage_da_map_blocks(mpd) == 0) |
2256 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2257 | /* |
2258 | * skip rest of the page in the page_vec | |
2259 | */ | |
2260 | mpd->io_done = 1; | |
2261 | redirty_page_for_writepage(wbc, page); | |
2262 | unlock_page(page); | |
2263 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2264 | } |
2265 | ||
2266 | /* | |
2267 | * Start next extent of pages ... | |
2268 | */ | |
2269 | mpd->first_page = page->index; | |
2270 | ||
2271 | /* | |
2272 | * ... and blocks | |
2273 | */ | |
8dc207c0 TT |
2274 | mpd->b_size = 0; |
2275 | mpd->b_state = 0; | |
2276 | mpd->b_blocknr = 0; | |
64769240 AT |
2277 | } |
2278 | ||
2279 | mpd->next_page = page->index + 1; | |
2280 | logical = (sector_t) page->index << | |
2281 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2282 | ||
2283 | if (!page_has_buffers(page)) { | |
8dc207c0 TT |
2284 | mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE, |
2285 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | |
a1d6cc56 AK |
2286 | if (mpd->io_done) |
2287 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2288 | } else { |
2289 | /* | |
2290 | * Page with regular buffer heads, just add all dirty ones | |
2291 | */ | |
2292 | head = page_buffers(page); | |
2293 | bh = head; | |
2294 | do { | |
2295 | BUG_ON(buffer_locked(bh)); | |
791b7f08 AK |
2296 | /* |
2297 | * We need to try to allocate | |
2298 | * unmapped blocks in the same page. | |
2299 | * Otherwise we won't make progress | |
2300 | * with the page in ext4_da_writepage | |
2301 | */ | |
29fa89d0 | 2302 | if (ext4_bh_unmapped_or_delay(NULL, bh)) { |
8dc207c0 TT |
2303 | mpage_add_bh_to_extent(mpd, logical, |
2304 | bh->b_size, | |
2305 | bh->b_state); | |
a1d6cc56 AK |
2306 | if (mpd->io_done) |
2307 | return MPAGE_DA_EXTENT_TAIL; | |
791b7f08 AK |
2308 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2309 | /* | |
2310 | * mapped dirty buffer. We need to update | |
2311 | * the b_state because we look at | |
2312 | * b_state in mpage_da_map_blocks. We don't | |
2313 | * update b_size because if we find an | |
2314 | * unmapped buffer_head later we need to | |
2315 | * use the b_state flag of that buffer_head. | |
2316 | */ | |
8dc207c0 TT |
2317 | if (mpd->b_size == 0) |
2318 | mpd->b_state = bh->b_state & BH_FLAGS; | |
a1d6cc56 | 2319 | } |
64769240 AT |
2320 | logical++; |
2321 | } while ((bh = bh->b_this_page) != head); | |
2322 | } | |
2323 | ||
2324 | return 0; | |
2325 | } | |
2326 | ||
64769240 | 2327 | /* |
b920c755 TT |
2328 | * This is a special get_blocks_t callback which is used by |
2329 | * ext4_da_write_begin(). It will either return mapped block or | |
2330 | * reserve space for a single block. | |
29fa89d0 AK |
2331 | * |
2332 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2333 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2334 | * | |
2335 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2336 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2337 | * initialized properly. | |
64769240 AT |
2338 | */ |
2339 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2340 | struct buffer_head *bh_result, int create) | |
2341 | { | |
2342 | int ret = 0; | |
33b9817e AK |
2343 | sector_t invalid_block = ~((sector_t) 0xffff); |
2344 | ||
2345 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
2346 | invalid_block = ~0; | |
64769240 AT |
2347 | |
2348 | BUG_ON(create == 0); | |
2349 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); | |
2350 | ||
2351 | /* | |
2352 | * first, we need to know whether the block is allocated already | |
2353 | * preallocated blocks are unmapped but should treated | |
2354 | * the same as allocated blocks. | |
2355 | */ | |
c2177057 | 2356 | ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0); |
d2a17637 MC |
2357 | if ((ret == 0) && !buffer_delay(bh_result)) { |
2358 | /* the block isn't (pre)allocated yet, let's reserve space */ | |
64769240 AT |
2359 | /* |
2360 | * XXX: __block_prepare_write() unmaps passed block, | |
2361 | * is it OK? | |
2362 | */ | |
d2a17637 MC |
2363 | ret = ext4_da_reserve_space(inode, 1); |
2364 | if (ret) | |
2365 | /* not enough space to reserve */ | |
2366 | return ret; | |
2367 | ||
33b9817e | 2368 | map_bh(bh_result, inode->i_sb, invalid_block); |
64769240 AT |
2369 | set_buffer_new(bh_result); |
2370 | set_buffer_delay(bh_result); | |
2371 | } else if (ret > 0) { | |
2372 | bh_result->b_size = (ret << inode->i_blkbits); | |
29fa89d0 AK |
2373 | if (buffer_unwritten(bh_result)) { |
2374 | /* A delayed write to unwritten bh should | |
2375 | * be marked new and mapped. Mapped ensures | |
2376 | * that we don't do get_block multiple times | |
2377 | * when we write to the same offset and new | |
2378 | * ensures that we do proper zero out for | |
2379 | * partial write. | |
2380 | */ | |
9c1ee184 | 2381 | set_buffer_new(bh_result); |
29fa89d0 AK |
2382 | set_buffer_mapped(bh_result); |
2383 | } | |
64769240 AT |
2384 | ret = 0; |
2385 | } | |
2386 | ||
2387 | return ret; | |
2388 | } | |
61628a3f | 2389 | |
b920c755 TT |
2390 | /* |
2391 | * This function is used as a standard get_block_t calback function | |
2392 | * when there is no desire to allocate any blocks. It is used as a | |
2393 | * callback function for block_prepare_write(), nobh_writepage(), and | |
2394 | * block_write_full_page(). These functions should only try to map a | |
2395 | * single block at a time. | |
2396 | * | |
2397 | * Since this function doesn't do block allocations even if the caller | |
2398 | * requests it by passing in create=1, it is critically important that | |
2399 | * any caller checks to make sure that any buffer heads are returned | |
2400 | * by this function are either all already mapped or marked for | |
2401 | * delayed allocation before calling nobh_writepage() or | |
2402 | * block_write_full_page(). Otherwise, b_blocknr could be left | |
2403 | * unitialized, and the page write functions will be taken by | |
2404 | * surprise. | |
2405 | */ | |
2406 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
2407 | struct buffer_head *bh_result, int create) |
2408 | { | |
2409 | int ret = 0; | |
2410 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
2411 | ||
a2dc52b5 TT |
2412 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2413 | ||
f0e6c985 AK |
2414 | /* |
2415 | * we don't want to do block allocation in writepage | |
2416 | * so call get_block_wrap with create = 0 | |
2417 | */ | |
c2177057 | 2418 | ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0); |
a2dc52b5 | 2419 | BUG_ON(create && ret == 0); |
f0e6c985 AK |
2420 | if (ret > 0) { |
2421 | bh_result->b_size = (ret << inode->i_blkbits); | |
2422 | ret = 0; | |
2423 | } | |
2424 | return ret; | |
61628a3f MC |
2425 | } |
2426 | ||
61628a3f | 2427 | /* |
b920c755 TT |
2428 | * This function can get called via... |
2429 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2430 | * - journal_submit_inode_data_buffers (no journal handle) | |
2431 | * - shrink_page_list via pdflush (no journal handle) | |
2432 | * - grab_page_cache when doing write_begin (have journal handle) | |
61628a3f | 2433 | */ |
64769240 AT |
2434 | static int ext4_da_writepage(struct page *page, |
2435 | struct writeback_control *wbc) | |
2436 | { | |
64769240 | 2437 | int ret = 0; |
61628a3f | 2438 | loff_t size; |
498e5f24 | 2439 | unsigned int len; |
61628a3f MC |
2440 | struct buffer_head *page_bufs; |
2441 | struct inode *inode = page->mapping->host; | |
2442 | ||
ba80b101 TT |
2443 | trace_mark(ext4_da_writepage, |
2444 | "dev %s ino %lu page_index %lu", | |
2445 | inode->i_sb->s_id, inode->i_ino, page->index); | |
f0e6c985 AK |
2446 | size = i_size_read(inode); |
2447 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2448 | len = size & ~PAGE_CACHE_MASK; | |
2449 | else | |
2450 | len = PAGE_CACHE_SIZE; | |
64769240 | 2451 | |
f0e6c985 | 2452 | if (page_has_buffers(page)) { |
61628a3f | 2453 | page_bufs = page_buffers(page); |
f0e6c985 AK |
2454 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2455 | ext4_bh_unmapped_or_delay)) { | |
61628a3f | 2456 | /* |
f0e6c985 AK |
2457 | * We don't want to do block allocation |
2458 | * So redirty the page and return | |
cd1aac32 AK |
2459 | * We may reach here when we do a journal commit |
2460 | * via journal_submit_inode_data_buffers. | |
2461 | * If we don't have mapping block we just ignore | |
f0e6c985 AK |
2462 | * them. We can also reach here via shrink_page_list |
2463 | */ | |
2464 | redirty_page_for_writepage(wbc, page); | |
2465 | unlock_page(page); | |
2466 | return 0; | |
2467 | } | |
2468 | } else { | |
2469 | /* | |
2470 | * The test for page_has_buffers() is subtle: | |
2471 | * We know the page is dirty but it lost buffers. That means | |
2472 | * that at some moment in time after write_begin()/write_end() | |
2473 | * has been called all buffers have been clean and thus they | |
2474 | * must have been written at least once. So they are all | |
2475 | * mapped and we can happily proceed with mapping them | |
2476 | * and writing the page. | |
2477 | * | |
2478 | * Try to initialize the buffer_heads and check whether | |
2479 | * all are mapped and non delay. We don't want to | |
2480 | * do block allocation here. | |
2481 | */ | |
2482 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
b920c755 | 2483 | noalloc_get_block_write); |
f0e6c985 AK |
2484 | if (!ret) { |
2485 | page_bufs = page_buffers(page); | |
2486 | /* check whether all are mapped and non delay */ | |
2487 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2488 | ext4_bh_unmapped_or_delay)) { | |
2489 | redirty_page_for_writepage(wbc, page); | |
2490 | unlock_page(page); | |
2491 | return 0; | |
2492 | } | |
2493 | } else { | |
2494 | /* | |
2495 | * We can't do block allocation here | |
2496 | * so just redity the page and unlock | |
2497 | * and return | |
61628a3f | 2498 | */ |
61628a3f MC |
2499 | redirty_page_for_writepage(wbc, page); |
2500 | unlock_page(page); | |
2501 | return 0; | |
2502 | } | |
ed9b3e33 AK |
2503 | /* now mark the buffer_heads as dirty and uptodate */ |
2504 | block_commit_write(page, 0, PAGE_CACHE_SIZE); | |
64769240 AT |
2505 | } |
2506 | ||
2507 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) | |
b920c755 | 2508 | ret = nobh_writepage(page, noalloc_get_block_write, wbc); |
64769240 | 2509 | else |
b920c755 TT |
2510 | ret = block_write_full_page(page, noalloc_get_block_write, |
2511 | wbc); | |
64769240 | 2512 | |
64769240 AT |
2513 | return ret; |
2514 | } | |
2515 | ||
61628a3f | 2516 | /* |
525f4ed8 MC |
2517 | * This is called via ext4_da_writepages() to |
2518 | * calulate the total number of credits to reserve to fit | |
2519 | * a single extent allocation into a single transaction, | |
2520 | * ext4_da_writpeages() will loop calling this before | |
2521 | * the block allocation. | |
61628a3f | 2522 | */ |
525f4ed8 MC |
2523 | |
2524 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2525 | { | |
2526 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2527 | ||
2528 | /* | |
2529 | * With non-extent format the journal credit needed to | |
2530 | * insert nrblocks contiguous block is dependent on | |
2531 | * number of contiguous block. So we will limit | |
2532 | * number of contiguous block to a sane value | |
2533 | */ | |
2534 | if (!(inode->i_flags & EXT4_EXTENTS_FL) && | |
2535 | (max_blocks > EXT4_MAX_TRANS_DATA)) | |
2536 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2537 | ||
2538 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2539 | } | |
61628a3f | 2540 | |
64769240 | 2541 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2542 | struct writeback_control *wbc) |
64769240 | 2543 | { |
22208ded AK |
2544 | pgoff_t index; |
2545 | int range_whole = 0; | |
61628a3f | 2546 | handle_t *handle = NULL; |
df22291f | 2547 | struct mpage_da_data mpd; |
5e745b04 | 2548 | struct inode *inode = mapping->host; |
22208ded | 2549 | int no_nrwrite_index_update; |
498e5f24 TT |
2550 | int pages_written = 0; |
2551 | long pages_skipped; | |
2acf2c26 | 2552 | int range_cyclic, cycled = 1, io_done = 0; |
5e745b04 | 2553 | int needed_blocks, ret = 0, nr_to_writebump = 0; |
5e745b04 | 2554 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
61628a3f | 2555 | |
ba80b101 TT |
2556 | trace_mark(ext4_da_writepages, |
2557 | "dev %s ino %lu nr_t_write %ld " | |
2558 | "pages_skipped %ld range_start %llu " | |
2559 | "range_end %llu nonblocking %d " | |
2560 | "for_kupdate %d for_reclaim %d " | |
2561 | "for_writepages %d range_cyclic %d", | |
2562 | inode->i_sb->s_id, inode->i_ino, | |
2563 | wbc->nr_to_write, wbc->pages_skipped, | |
2564 | (unsigned long long) wbc->range_start, | |
2565 | (unsigned long long) wbc->range_end, | |
2566 | wbc->nonblocking, wbc->for_kupdate, | |
2567 | wbc->for_reclaim, wbc->for_writepages, | |
2568 | wbc->range_cyclic); | |
2569 | ||
61628a3f MC |
2570 | /* |
2571 | * No pages to write? This is mainly a kludge to avoid starting | |
2572 | * a transaction for special inodes like journal inode on last iput() | |
2573 | * because that could violate lock ordering on umount | |
2574 | */ | |
a1d6cc56 | 2575 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2576 | return 0; |
2a21e37e TT |
2577 | |
2578 | /* | |
2579 | * If the filesystem has aborted, it is read-only, so return | |
2580 | * right away instead of dumping stack traces later on that | |
2581 | * will obscure the real source of the problem. We test | |
2582 | * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because | |
2583 | * the latter could be true if the filesystem is mounted | |
2584 | * read-only, and in that case, ext4_da_writepages should | |
2585 | * *never* be called, so if that ever happens, we would want | |
2586 | * the stack trace. | |
2587 | */ | |
2588 | if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT)) | |
2589 | return -EROFS; | |
2590 | ||
5e745b04 AK |
2591 | /* |
2592 | * Make sure nr_to_write is >= sbi->s_mb_stream_request | |
2593 | * This make sure small files blocks are allocated in | |
2594 | * single attempt. This ensure that small files | |
2595 | * get less fragmented. | |
2596 | */ | |
2597 | if (wbc->nr_to_write < sbi->s_mb_stream_request) { | |
2598 | nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write; | |
2599 | wbc->nr_to_write = sbi->s_mb_stream_request; | |
2600 | } | |
22208ded AK |
2601 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2602 | range_whole = 1; | |
61628a3f | 2603 | |
2acf2c26 AK |
2604 | range_cyclic = wbc->range_cyclic; |
2605 | if (wbc->range_cyclic) { | |
22208ded | 2606 | index = mapping->writeback_index; |
2acf2c26 AK |
2607 | if (index) |
2608 | cycled = 0; | |
2609 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2610 | wbc->range_end = LLONG_MAX; | |
2611 | wbc->range_cyclic = 0; | |
2612 | } else | |
22208ded | 2613 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
a1d6cc56 | 2614 | |
df22291f AK |
2615 | mpd.wbc = wbc; |
2616 | mpd.inode = mapping->host; | |
2617 | ||
22208ded AK |
2618 | /* |
2619 | * we don't want write_cache_pages to update | |
2620 | * nr_to_write and writeback_index | |
2621 | */ | |
2622 | no_nrwrite_index_update = wbc->no_nrwrite_index_update; | |
2623 | wbc->no_nrwrite_index_update = 1; | |
2624 | pages_skipped = wbc->pages_skipped; | |
2625 | ||
2acf2c26 | 2626 | retry: |
22208ded | 2627 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2628 | |
2629 | /* | |
2630 | * we insert one extent at a time. So we need | |
2631 | * credit needed for single extent allocation. | |
2632 | * journalled mode is currently not supported | |
2633 | * by delalloc | |
2634 | */ | |
2635 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2636 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2637 | |
61628a3f MC |
2638 | /* start a new transaction*/ |
2639 | handle = ext4_journal_start(inode, needed_blocks); | |
2640 | if (IS_ERR(handle)) { | |
2641 | ret = PTR_ERR(handle); | |
2a21e37e | 2642 | printk(KERN_CRIT "%s: jbd2_start: " |
a1d6cc56 AK |
2643 | "%ld pages, ino %lu; err %d\n", __func__, |
2644 | wbc->nr_to_write, inode->i_ino, ret); | |
2645 | dump_stack(); | |
61628a3f MC |
2646 | goto out_writepages; |
2647 | } | |
f63e6005 TT |
2648 | |
2649 | /* | |
2650 | * Now call __mpage_da_writepage to find the next | |
2651 | * contiguous region of logical blocks that need | |
2652 | * blocks to be allocated by ext4. We don't actually | |
2653 | * submit the blocks for I/O here, even though | |
2654 | * write_cache_pages thinks it will, and will set the | |
2655 | * pages as clean for write before calling | |
2656 | * __mpage_da_writepage(). | |
2657 | */ | |
2658 | mpd.b_size = 0; | |
2659 | mpd.b_state = 0; | |
2660 | mpd.b_blocknr = 0; | |
2661 | mpd.first_page = 0; | |
2662 | mpd.next_page = 0; | |
2663 | mpd.io_done = 0; | |
2664 | mpd.pages_written = 0; | |
2665 | mpd.retval = 0; | |
2666 | ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, | |
2667 | &mpd); | |
2668 | /* | |
2669 | * If we have a contigous extent of pages and we | |
2670 | * haven't done the I/O yet, map the blocks and submit | |
2671 | * them for I/O. | |
2672 | */ | |
2673 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
2674 | if (mpage_da_map_blocks(&mpd) == 0) | |
2675 | mpage_da_submit_io(&mpd); | |
2676 | mpd.io_done = 1; | |
2677 | ret = MPAGE_DA_EXTENT_TAIL; | |
2678 | } | |
2679 | wbc->nr_to_write -= mpd.pages_written; | |
df22291f | 2680 | |
61628a3f | 2681 | ext4_journal_stop(handle); |
df22291f | 2682 | |
8f64b32e | 2683 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2684 | /* commit the transaction which would |
2685 | * free blocks released in the transaction | |
2686 | * and try again | |
2687 | */ | |
df22291f | 2688 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2689 | wbc->pages_skipped = pages_skipped; |
2690 | ret = 0; | |
2691 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
2692 | /* |
2693 | * got one extent now try with | |
2694 | * rest of the pages | |
2695 | */ | |
22208ded AK |
2696 | pages_written += mpd.pages_written; |
2697 | wbc->pages_skipped = pages_skipped; | |
a1d6cc56 | 2698 | ret = 0; |
2acf2c26 | 2699 | io_done = 1; |
22208ded | 2700 | } else if (wbc->nr_to_write) |
61628a3f MC |
2701 | /* |
2702 | * There is no more writeout needed | |
2703 | * or we requested for a noblocking writeout | |
2704 | * and we found the device congested | |
2705 | */ | |
61628a3f | 2706 | break; |
a1d6cc56 | 2707 | } |
2acf2c26 AK |
2708 | if (!io_done && !cycled) { |
2709 | cycled = 1; | |
2710 | index = 0; | |
2711 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2712 | wbc->range_end = mapping->writeback_index - 1; | |
2713 | goto retry; | |
2714 | } | |
22208ded AK |
2715 | if (pages_skipped != wbc->pages_skipped) |
2716 | printk(KERN_EMERG "This should not happen leaving %s " | |
2717 | "with nr_to_write = %ld ret = %d\n", | |
2718 | __func__, wbc->nr_to_write, ret); | |
2719 | ||
2720 | /* Update index */ | |
2721 | index += pages_written; | |
2acf2c26 | 2722 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2723 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2724 | /* | |
2725 | * set the writeback_index so that range_cyclic | |
2726 | * mode will write it back later | |
2727 | */ | |
2728 | mapping->writeback_index = index; | |
a1d6cc56 | 2729 | |
61628a3f | 2730 | out_writepages: |
22208ded AK |
2731 | if (!no_nrwrite_index_update) |
2732 | wbc->no_nrwrite_index_update = 0; | |
2733 | wbc->nr_to_write -= nr_to_writebump; | |
ba80b101 TT |
2734 | trace_mark(ext4_da_writepage_result, |
2735 | "dev %s ino %lu ret %d pages_written %d " | |
2736 | "pages_skipped %ld congestion %d " | |
2737 | "more_io %d no_nrwrite_index_update %d", | |
2738 | inode->i_sb->s_id, inode->i_ino, ret, | |
2739 | pages_written, wbc->pages_skipped, | |
2740 | wbc->encountered_congestion, wbc->more_io, | |
2741 | wbc->no_nrwrite_index_update); | |
61628a3f | 2742 | return ret; |
64769240 AT |
2743 | } |
2744 | ||
79f0be8d AK |
2745 | #define FALL_BACK_TO_NONDELALLOC 1 |
2746 | static int ext4_nonda_switch(struct super_block *sb) | |
2747 | { | |
2748 | s64 free_blocks, dirty_blocks; | |
2749 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2750 | ||
2751 | /* | |
2752 | * switch to non delalloc mode if we are running low | |
2753 | * on free block. The free block accounting via percpu | |
179f7ebf | 2754 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2755 | * accumulated on each CPU without updating global counters |
2756 | * Delalloc need an accurate free block accounting. So switch | |
2757 | * to non delalloc when we are near to error range. | |
2758 | */ | |
2759 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
2760 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter); | |
2761 | if (2 * free_blocks < 3 * dirty_blocks || | |
2762 | free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) { | |
2763 | /* | |
2764 | * free block count is less that 150% of dirty blocks | |
2765 | * or free blocks is less that watermark | |
2766 | */ | |
2767 | return 1; | |
2768 | } | |
2769 | return 0; | |
2770 | } | |
2771 | ||
64769240 AT |
2772 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
2773 | loff_t pos, unsigned len, unsigned flags, | |
2774 | struct page **pagep, void **fsdata) | |
2775 | { | |
d2a17637 | 2776 | int ret, retries = 0; |
64769240 AT |
2777 | struct page *page; |
2778 | pgoff_t index; | |
2779 | unsigned from, to; | |
2780 | struct inode *inode = mapping->host; | |
2781 | handle_t *handle; | |
2782 | ||
2783 | index = pos >> PAGE_CACHE_SHIFT; | |
2784 | from = pos & (PAGE_CACHE_SIZE - 1); | |
2785 | to = from + len; | |
79f0be8d AK |
2786 | |
2787 | if (ext4_nonda_switch(inode->i_sb)) { | |
2788 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2789 | return ext4_write_begin(file, mapping, pos, | |
2790 | len, flags, pagep, fsdata); | |
2791 | } | |
2792 | *fsdata = (void *)0; | |
ba80b101 TT |
2793 | |
2794 | trace_mark(ext4_da_write_begin, | |
2795 | "dev %s ino %lu pos %llu len %u flags %u", | |
2796 | inode->i_sb->s_id, inode->i_ino, | |
2797 | (unsigned long long) pos, len, flags); | |
d2a17637 | 2798 | retry: |
64769240 AT |
2799 | /* |
2800 | * With delayed allocation, we don't log the i_disksize update | |
2801 | * if there is delayed block allocation. But we still need | |
2802 | * to journalling the i_disksize update if writes to the end | |
2803 | * of file which has an already mapped buffer. | |
2804 | */ | |
2805 | handle = ext4_journal_start(inode, 1); | |
2806 | if (IS_ERR(handle)) { | |
2807 | ret = PTR_ERR(handle); | |
2808 | goto out; | |
2809 | } | |
ebd3610b JK |
2810 | /* We cannot recurse into the filesystem as the transaction is already |
2811 | * started */ | |
2812 | flags |= AOP_FLAG_NOFS; | |
64769240 | 2813 | |
54566b2c | 2814 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
2815 | if (!page) { |
2816 | ext4_journal_stop(handle); | |
2817 | ret = -ENOMEM; | |
2818 | goto out; | |
2819 | } | |
64769240 AT |
2820 | *pagep = page; |
2821 | ||
2822 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
b920c755 | 2823 | ext4_da_get_block_prep); |
64769240 AT |
2824 | if (ret < 0) { |
2825 | unlock_page(page); | |
2826 | ext4_journal_stop(handle); | |
2827 | page_cache_release(page); | |
ae4d5372 AK |
2828 | /* |
2829 | * block_write_begin may have instantiated a few blocks | |
2830 | * outside i_size. Trim these off again. Don't need | |
2831 | * i_size_read because we hold i_mutex. | |
2832 | */ | |
2833 | if (pos + len > inode->i_size) | |
2834 | vmtruncate(inode, inode->i_size); | |
64769240 AT |
2835 | } |
2836 | ||
d2a17637 MC |
2837 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2838 | goto retry; | |
64769240 AT |
2839 | out: |
2840 | return ret; | |
2841 | } | |
2842 | ||
632eaeab MC |
2843 | /* |
2844 | * Check if we should update i_disksize | |
2845 | * when write to the end of file but not require block allocation | |
2846 | */ | |
2847 | static int ext4_da_should_update_i_disksize(struct page *page, | |
2848 | unsigned long offset) | |
2849 | { | |
2850 | struct buffer_head *bh; | |
2851 | struct inode *inode = page->mapping->host; | |
2852 | unsigned int idx; | |
2853 | int i; | |
2854 | ||
2855 | bh = page_buffers(page); | |
2856 | idx = offset >> inode->i_blkbits; | |
2857 | ||
af5bc92d | 2858 | for (i = 0; i < idx; i++) |
632eaeab MC |
2859 | bh = bh->b_this_page; |
2860 | ||
29fa89d0 | 2861 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2862 | return 0; |
2863 | return 1; | |
2864 | } | |
2865 | ||
64769240 AT |
2866 | static int ext4_da_write_end(struct file *file, |
2867 | struct address_space *mapping, | |
2868 | loff_t pos, unsigned len, unsigned copied, | |
2869 | struct page *page, void *fsdata) | |
2870 | { | |
2871 | struct inode *inode = mapping->host; | |
2872 | int ret = 0, ret2; | |
2873 | handle_t *handle = ext4_journal_current_handle(); | |
2874 | loff_t new_i_size; | |
632eaeab | 2875 | unsigned long start, end; |
79f0be8d AK |
2876 | int write_mode = (int)(unsigned long)fsdata; |
2877 | ||
2878 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
2879 | if (ext4_should_order_data(inode)) { | |
2880 | return ext4_ordered_write_end(file, mapping, pos, | |
2881 | len, copied, page, fsdata); | |
2882 | } else if (ext4_should_writeback_data(inode)) { | |
2883 | return ext4_writeback_write_end(file, mapping, pos, | |
2884 | len, copied, page, fsdata); | |
2885 | } else { | |
2886 | BUG(); | |
2887 | } | |
2888 | } | |
632eaeab | 2889 | |
ba80b101 TT |
2890 | trace_mark(ext4_da_write_end, |
2891 | "dev %s ino %lu pos %llu len %u copied %u", | |
2892 | inode->i_sb->s_id, inode->i_ino, | |
2893 | (unsigned long long) pos, len, copied); | |
632eaeab | 2894 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2895 | end = start + copied - 1; |
64769240 AT |
2896 | |
2897 | /* | |
2898 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2899 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2900 | * into that. | |
2901 | */ | |
2902 | ||
2903 | new_i_size = pos + copied; | |
632eaeab MC |
2904 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
2905 | if (ext4_da_should_update_i_disksize(page, end)) { | |
2906 | down_write(&EXT4_I(inode)->i_data_sem); | |
2907 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2908 | /* | |
2909 | * Updating i_disksize when extending file | |
2910 | * without needing block allocation | |
2911 | */ | |
2912 | if (ext4_should_order_data(inode)) | |
2913 | ret = ext4_jbd2_file_inode(handle, | |
2914 | inode); | |
64769240 | 2915 | |
632eaeab MC |
2916 | EXT4_I(inode)->i_disksize = new_i_size; |
2917 | } | |
2918 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
2919 | /* We need to mark inode dirty even if |
2920 | * new_i_size is less that inode->i_size | |
2921 | * bu greater than i_disksize.(hint delalloc) | |
2922 | */ | |
2923 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2924 | } |
632eaeab | 2925 | } |
64769240 AT |
2926 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
2927 | page, fsdata); | |
2928 | copied = ret2; | |
2929 | if (ret2 < 0) | |
2930 | ret = ret2; | |
2931 | ret2 = ext4_journal_stop(handle); | |
2932 | if (!ret) | |
2933 | ret = ret2; | |
2934 | ||
2935 | return ret ? ret : copied; | |
2936 | } | |
2937 | ||
2938 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2939 | { | |
64769240 AT |
2940 | /* |
2941 | * Drop reserved blocks | |
2942 | */ | |
2943 | BUG_ON(!PageLocked(page)); | |
2944 | if (!page_has_buffers(page)) | |
2945 | goto out; | |
2946 | ||
d2a17637 | 2947 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2948 | |
2949 | out: | |
2950 | ext4_invalidatepage(page, offset); | |
2951 | ||
2952 | return; | |
2953 | } | |
2954 | ||
ccd2506b TT |
2955 | /* |
2956 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2957 | */ | |
2958 | int ext4_alloc_da_blocks(struct inode *inode) | |
2959 | { | |
2960 | if (!EXT4_I(inode)->i_reserved_data_blocks && | |
2961 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2962 | return 0; | |
2963 | ||
2964 | /* | |
2965 | * We do something simple for now. The filemap_flush() will | |
2966 | * also start triggering a write of the data blocks, which is | |
2967 | * not strictly speaking necessary (and for users of | |
2968 | * laptop_mode, not even desirable). However, to do otherwise | |
2969 | * would require replicating code paths in: | |
2970 | * | |
2971 | * ext4_da_writepages() -> | |
2972 | * write_cache_pages() ---> (via passed in callback function) | |
2973 | * __mpage_da_writepage() --> | |
2974 | * mpage_add_bh_to_extent() | |
2975 | * mpage_da_map_blocks() | |
2976 | * | |
2977 | * The problem is that write_cache_pages(), located in | |
2978 | * mm/page-writeback.c, marks pages clean in preparation for | |
2979 | * doing I/O, which is not desirable if we're not planning on | |
2980 | * doing I/O at all. | |
2981 | * | |
2982 | * We could call write_cache_pages(), and then redirty all of | |
2983 | * the pages by calling redirty_page_for_writeback() but that | |
2984 | * would be ugly in the extreme. So instead we would need to | |
2985 | * replicate parts of the code in the above functions, | |
2986 | * simplifying them becuase we wouldn't actually intend to | |
2987 | * write out the pages, but rather only collect contiguous | |
2988 | * logical block extents, call the multi-block allocator, and | |
2989 | * then update the buffer heads with the block allocations. | |
2990 | * | |
2991 | * For now, though, we'll cheat by calling filemap_flush(), | |
2992 | * which will map the blocks, and start the I/O, but not | |
2993 | * actually wait for the I/O to complete. | |
2994 | */ | |
2995 | return filemap_flush(inode->i_mapping); | |
2996 | } | |
64769240 | 2997 | |
ac27a0ec DK |
2998 | /* |
2999 | * bmap() is special. It gets used by applications such as lilo and by | |
3000 | * the swapper to find the on-disk block of a specific piece of data. | |
3001 | * | |
3002 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3003 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3004 | * filesystem and enables swap, then they may get a nasty shock when the |
3005 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3006 | * the original zero's written out previously to the journal and | |
3007 | * awaiting writeback in the kernel's buffer cache. | |
3008 | * | |
3009 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3010 | * take extra steps to flush any blocks which might be in the cache. | |
3011 | */ | |
617ba13b | 3012 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3013 | { |
3014 | struct inode *inode = mapping->host; | |
3015 | journal_t *journal; | |
3016 | int err; | |
3017 | ||
64769240 AT |
3018 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3019 | test_opt(inode->i_sb, DELALLOC)) { | |
3020 | /* | |
3021 | * With delalloc we want to sync the file | |
3022 | * so that we can make sure we allocate | |
3023 | * blocks for file | |
3024 | */ | |
3025 | filemap_write_and_wait(mapping); | |
3026 | } | |
3027 | ||
0390131b | 3028 | if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
ac27a0ec DK |
3029 | /* |
3030 | * This is a REALLY heavyweight approach, but the use of | |
3031 | * bmap on dirty files is expected to be extremely rare: | |
3032 | * only if we run lilo or swapon on a freshly made file | |
3033 | * do we expect this to happen. | |
3034 | * | |
3035 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3036 | * represent an unprivileged user DOS attack --- we'd be | |
3037 | * in trouble if mortal users could trigger this path at | |
3038 | * will.) | |
3039 | * | |
617ba13b | 3040 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3041 | * regular files. If somebody wants to bmap a directory |
3042 | * or symlink and gets confused because the buffer | |
3043 | * hasn't yet been flushed to disk, they deserve | |
3044 | * everything they get. | |
3045 | */ | |
3046 | ||
617ba13b MC |
3047 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
3048 | journal = EXT4_JOURNAL(inode); | |
dab291af MC |
3049 | jbd2_journal_lock_updates(journal); |
3050 | err = jbd2_journal_flush(journal); | |
3051 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3052 | |
3053 | if (err) | |
3054 | return 0; | |
3055 | } | |
3056 | ||
af5bc92d | 3057 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3058 | } |
3059 | ||
3060 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
3061 | { | |
3062 | get_bh(bh); | |
3063 | return 0; | |
3064 | } | |
3065 | ||
3066 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
3067 | { | |
3068 | put_bh(bh); | |
3069 | return 0; | |
3070 | } | |
3071 | ||
ac27a0ec | 3072 | /* |
678aaf48 JK |
3073 | * Note that we don't need to start a transaction unless we're journaling data |
3074 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
3075 | * need to file the inode to the transaction's list in ordered mode because if | |
3076 | * we are writing back data added by write(), the inode is already there and if | |
3077 | * we are writing back data modified via mmap(), noone guarantees in which | |
3078 | * transaction the data will hit the disk. In case we are journaling data, we | |
3079 | * cannot start transaction directly because transaction start ranks above page | |
3080 | * lock so we have to do some magic. | |
ac27a0ec | 3081 | * |
678aaf48 | 3082 | * In all journaling modes block_write_full_page() will start the I/O. |
ac27a0ec DK |
3083 | * |
3084 | * Problem: | |
3085 | * | |
617ba13b MC |
3086 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
3087 | * ext4_writepage() | |
ac27a0ec DK |
3088 | * |
3089 | * Similar for: | |
3090 | * | |
617ba13b | 3091 | * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
ac27a0ec | 3092 | * |
617ba13b | 3093 | * Same applies to ext4_get_block(). We will deadlock on various things like |
0e855ac8 | 3094 | * lock_journal and i_data_sem |
ac27a0ec DK |
3095 | * |
3096 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
3097 | * allocations fail. | |
3098 | * | |
3099 | * 16May01: If we're reentered then journal_current_handle() will be | |
3100 | * non-zero. We simply *return*. | |
3101 | * | |
3102 | * 1 July 2001: @@@ FIXME: | |
3103 | * In journalled data mode, a data buffer may be metadata against the | |
3104 | * current transaction. But the same file is part of a shared mapping | |
3105 | * and someone does a writepage() on it. | |
3106 | * | |
3107 | * We will move the buffer onto the async_data list, but *after* it has | |
3108 | * been dirtied. So there's a small window where we have dirty data on | |
3109 | * BJ_Metadata. | |
3110 | * | |
3111 | * Note that this only applies to the last partial page in the file. The | |
3112 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
3113 | * broken code anyway: it's wrong for msync()). | |
3114 | * | |
3115 | * It's a rare case: affects the final partial page, for journalled data | |
3116 | * where the file is subject to bith write() and writepage() in the same | |
3117 | * transction. To fix it we'll need a custom block_write_full_page(). | |
3118 | * We'll probably need that anyway for journalling writepage() output. | |
3119 | * | |
3120 | * We don't honour synchronous mounts for writepage(). That would be | |
3121 | * disastrous. Any write() or metadata operation will sync the fs for | |
3122 | * us. | |
3123 | * | |
ac27a0ec | 3124 | */ |
678aaf48 | 3125 | static int __ext4_normal_writepage(struct page *page, |
cf108bca JK |
3126 | struct writeback_control *wbc) |
3127 | { | |
3128 | struct inode *inode = page->mapping->host; | |
3129 | ||
3130 | if (test_opt(inode->i_sb, NOBH)) | |
b920c755 | 3131 | return nobh_writepage(page, noalloc_get_block_write, wbc); |
cf108bca | 3132 | else |
b920c755 TT |
3133 | return block_write_full_page(page, noalloc_get_block_write, |
3134 | wbc); | |
cf108bca JK |
3135 | } |
3136 | ||
678aaf48 | 3137 | static int ext4_normal_writepage(struct page *page, |
ac27a0ec DK |
3138 | struct writeback_control *wbc) |
3139 | { | |
3140 | struct inode *inode = page->mapping->host; | |
cf108bca JK |
3141 | loff_t size = i_size_read(inode); |
3142 | loff_t len; | |
3143 | ||
ba80b101 TT |
3144 | trace_mark(ext4_normal_writepage, |
3145 | "dev %s ino %lu page_index %lu", | |
3146 | inode->i_sb->s_id, inode->i_ino, page->index); | |
cf108bca | 3147 | J_ASSERT(PageLocked(page)); |
cf108bca JK |
3148 | if (page->index == size >> PAGE_CACHE_SHIFT) |
3149 | len = size & ~PAGE_CACHE_MASK; | |
3150 | else | |
3151 | len = PAGE_CACHE_SIZE; | |
f0e6c985 AK |
3152 | |
3153 | if (page_has_buffers(page)) { | |
3154 | /* if page has buffers it should all be mapped | |
3155 | * and allocated. If there are not buffers attached | |
3156 | * to the page we know the page is dirty but it lost | |
3157 | * buffers. That means that at some moment in time | |
3158 | * after write_begin() / write_end() has been called | |
3159 | * all buffers have been clean and thus they must have been | |
3160 | * written at least once. So they are all mapped and we can | |
3161 | * happily proceed with mapping them and writing the page. | |
3162 | */ | |
3163 | BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
3164 | ext4_bh_unmapped_or_delay)); | |
3165 | } | |
cf108bca JK |
3166 | |
3167 | if (!ext4_journal_current_handle()) | |
678aaf48 | 3168 | return __ext4_normal_writepage(page, wbc); |
cf108bca JK |
3169 | |
3170 | redirty_page_for_writepage(wbc, page); | |
3171 | unlock_page(page); | |
3172 | return 0; | |
3173 | } | |
3174 | ||
3175 | static int __ext4_journalled_writepage(struct page *page, | |
3176 | struct writeback_control *wbc) | |
3177 | { | |
3178 | struct address_space *mapping = page->mapping; | |
3179 | struct inode *inode = mapping->host; | |
3180 | struct buffer_head *page_bufs; | |
ac27a0ec DK |
3181 | handle_t *handle = NULL; |
3182 | int ret = 0; | |
3183 | int err; | |
3184 | ||
f0e6c985 | 3185 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, |
b920c755 | 3186 | noalloc_get_block_write); |
cf108bca JK |
3187 | if (ret != 0) |
3188 | goto out_unlock; | |
3189 | ||
3190 | page_bufs = page_buffers(page); | |
3191 | walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL, | |
3192 | bget_one); | |
3193 | /* As soon as we unlock the page, it can go away, but we have | |
3194 | * references to buffers so we are safe */ | |
3195 | unlock_page(page); | |
ac27a0ec | 3196 | |
617ba13b | 3197 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
3198 | if (IS_ERR(handle)) { |
3199 | ret = PTR_ERR(handle); | |
cf108bca | 3200 | goto out; |
ac27a0ec DK |
3201 | } |
3202 | ||
cf108bca JK |
3203 | ret = walk_page_buffers(handle, page_bufs, 0, |
3204 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
ac27a0ec | 3205 | |
cf108bca JK |
3206 | err = walk_page_buffers(handle, page_bufs, 0, |
3207 | PAGE_CACHE_SIZE, NULL, write_end_fn); | |
3208 | if (ret == 0) | |
3209 | ret = err; | |
617ba13b | 3210 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3211 | if (!ret) |
3212 | ret = err; | |
ac27a0ec | 3213 | |
cf108bca JK |
3214 | walk_page_buffers(handle, page_bufs, 0, |
3215 | PAGE_CACHE_SIZE, NULL, bput_one); | |
3216 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; | |
3217 | goto out; | |
3218 | ||
3219 | out_unlock: | |
ac27a0ec | 3220 | unlock_page(page); |
cf108bca | 3221 | out: |
ac27a0ec DK |
3222 | return ret; |
3223 | } | |
3224 | ||
617ba13b | 3225 | static int ext4_journalled_writepage(struct page *page, |
ac27a0ec DK |
3226 | struct writeback_control *wbc) |
3227 | { | |
3228 | struct inode *inode = page->mapping->host; | |
cf108bca JK |
3229 | loff_t size = i_size_read(inode); |
3230 | loff_t len; | |
ac27a0ec | 3231 | |
ba80b101 TT |
3232 | trace_mark(ext4_journalled_writepage, |
3233 | "dev %s ino %lu page_index %lu", | |
3234 | inode->i_sb->s_id, inode->i_ino, page->index); | |
cf108bca | 3235 | J_ASSERT(PageLocked(page)); |
cf108bca JK |
3236 | if (page->index == size >> PAGE_CACHE_SHIFT) |
3237 | len = size & ~PAGE_CACHE_MASK; | |
3238 | else | |
3239 | len = PAGE_CACHE_SIZE; | |
f0e6c985 AK |
3240 | |
3241 | if (page_has_buffers(page)) { | |
3242 | /* if page has buffers it should all be mapped | |
3243 | * and allocated. If there are not buffers attached | |
3244 | * to the page we know the page is dirty but it lost | |
3245 | * buffers. That means that at some moment in time | |
3246 | * after write_begin() / write_end() has been called | |
3247 | * all buffers have been clean and thus they must have been | |
3248 | * written at least once. So they are all mapped and we can | |
3249 | * happily proceed with mapping them and writing the page. | |
3250 | */ | |
3251 | BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
3252 | ext4_bh_unmapped_or_delay)); | |
3253 | } | |
ac27a0ec | 3254 | |
cf108bca | 3255 | if (ext4_journal_current_handle()) |
ac27a0ec | 3256 | goto no_write; |
ac27a0ec | 3257 | |
cf108bca | 3258 | if (PageChecked(page)) { |
ac27a0ec DK |
3259 | /* |
3260 | * It's mmapped pagecache. Add buffers and journal it. There | |
3261 | * doesn't seem much point in redirtying the page here. | |
3262 | */ | |
3263 | ClearPageChecked(page); | |
cf108bca | 3264 | return __ext4_journalled_writepage(page, wbc); |
ac27a0ec DK |
3265 | } else { |
3266 | /* | |
3267 | * It may be a page full of checkpoint-mode buffers. We don't | |
3268 | * really know unless we go poke around in the buffer_heads. | |
3269 | * But block_write_full_page will do the right thing. | |
3270 | */ | |
b920c755 TT |
3271 | return block_write_full_page(page, noalloc_get_block_write, |
3272 | wbc); | |
ac27a0ec | 3273 | } |
ac27a0ec DK |
3274 | no_write: |
3275 | redirty_page_for_writepage(wbc, page); | |
ac27a0ec | 3276 | unlock_page(page); |
cf108bca | 3277 | return 0; |
ac27a0ec DK |
3278 | } |
3279 | ||
617ba13b | 3280 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3281 | { |
617ba13b | 3282 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
3283 | } |
3284 | ||
3285 | static int | |
617ba13b | 3286 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3287 | struct list_head *pages, unsigned nr_pages) |
3288 | { | |
617ba13b | 3289 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3290 | } |
3291 | ||
617ba13b | 3292 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3293 | { |
617ba13b | 3294 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3295 | |
3296 | /* | |
3297 | * If it's a full truncate we just forget about the pending dirtying | |
3298 | */ | |
3299 | if (offset == 0) | |
3300 | ClearPageChecked(page); | |
3301 | ||
0390131b FM |
3302 | if (journal) |
3303 | jbd2_journal_invalidatepage(journal, page, offset); | |
3304 | else | |
3305 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
3306 | } |
3307 | ||
617ba13b | 3308 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3309 | { |
617ba13b | 3310 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3311 | |
3312 | WARN_ON(PageChecked(page)); | |
3313 | if (!page_has_buffers(page)) | |
3314 | return 0; | |
0390131b FM |
3315 | if (journal) |
3316 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3317 | else | |
3318 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3319 | } |
3320 | ||
3321 | /* | |
3322 | * If the O_DIRECT write will extend the file then add this inode to the | |
3323 | * orphan list. So recovery will truncate it back to the original size | |
3324 | * if the machine crashes during the write. | |
3325 | * | |
3326 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
3327 | * crashes then stale disk data _may_ be exposed inside the file. But current |
3328 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 3329 | */ |
617ba13b | 3330 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
ac27a0ec DK |
3331 | const struct iovec *iov, loff_t offset, |
3332 | unsigned long nr_segs) | |
3333 | { | |
3334 | struct file *file = iocb->ki_filp; | |
3335 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 3336 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 3337 | handle_t *handle; |
ac27a0ec DK |
3338 | ssize_t ret; |
3339 | int orphan = 0; | |
3340 | size_t count = iov_length(iov, nr_segs); | |
3341 | ||
3342 | if (rw == WRITE) { | |
3343 | loff_t final_size = offset + count; | |
3344 | ||
ac27a0ec | 3345 | if (final_size > inode->i_size) { |
7fb5409d JK |
3346 | /* Credits for sb + inode write */ |
3347 | handle = ext4_journal_start(inode, 2); | |
3348 | if (IS_ERR(handle)) { | |
3349 | ret = PTR_ERR(handle); | |
3350 | goto out; | |
3351 | } | |
617ba13b | 3352 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
3353 | if (ret) { |
3354 | ext4_journal_stop(handle); | |
3355 | goto out; | |
3356 | } | |
ac27a0ec DK |
3357 | orphan = 1; |
3358 | ei->i_disksize = inode->i_size; | |
7fb5409d | 3359 | ext4_journal_stop(handle); |
ac27a0ec DK |
3360 | } |
3361 | } | |
3362 | ||
3363 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, | |
3364 | offset, nr_segs, | |
617ba13b | 3365 | ext4_get_block, NULL); |
ac27a0ec | 3366 | |
7fb5409d | 3367 | if (orphan) { |
ac27a0ec DK |
3368 | int err; |
3369 | ||
7fb5409d JK |
3370 | /* Credits for sb + inode write */ |
3371 | handle = ext4_journal_start(inode, 2); | |
3372 | if (IS_ERR(handle)) { | |
3373 | /* This is really bad luck. We've written the data | |
3374 | * but cannot extend i_size. Bail out and pretend | |
3375 | * the write failed... */ | |
3376 | ret = PTR_ERR(handle); | |
3377 | goto out; | |
3378 | } | |
3379 | if (inode->i_nlink) | |
617ba13b | 3380 | ext4_orphan_del(handle, inode); |
7fb5409d | 3381 | if (ret > 0) { |
ac27a0ec DK |
3382 | loff_t end = offset + ret; |
3383 | if (end > inode->i_size) { | |
3384 | ei->i_disksize = end; | |
3385 | i_size_write(inode, end); | |
3386 | /* | |
3387 | * We're going to return a positive `ret' | |
3388 | * here due to non-zero-length I/O, so there's | |
3389 | * no way of reporting error returns from | |
617ba13b | 3390 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
3391 | * ignore it. |
3392 | */ | |
617ba13b | 3393 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3394 | } |
3395 | } | |
617ba13b | 3396 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3397 | if (ret == 0) |
3398 | ret = err; | |
3399 | } | |
3400 | out: | |
3401 | return ret; | |
3402 | } | |
3403 | ||
3404 | /* | |
617ba13b | 3405 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3406 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3407 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3408 | * not necessarily locked. | |
3409 | * | |
3410 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3411 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3412 | * or jbddirty because all the journalling code will explode. | |
3413 | * | |
3414 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3415 | * is called, propagate that into the buffers appropriately. | |
3416 | */ | |
617ba13b | 3417 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3418 | { |
3419 | SetPageChecked(page); | |
3420 | return __set_page_dirty_nobuffers(page); | |
3421 | } | |
3422 | ||
617ba13b | 3423 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3424 | .readpage = ext4_readpage, |
3425 | .readpages = ext4_readpages, | |
3426 | .writepage = ext4_normal_writepage, | |
3427 | .sync_page = block_sync_page, | |
3428 | .write_begin = ext4_write_begin, | |
3429 | .write_end = ext4_ordered_write_end, | |
3430 | .bmap = ext4_bmap, | |
3431 | .invalidatepage = ext4_invalidatepage, | |
3432 | .releasepage = ext4_releasepage, | |
3433 | .direct_IO = ext4_direct_IO, | |
3434 | .migratepage = buffer_migrate_page, | |
3435 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
3436 | }; |
3437 | ||
617ba13b | 3438 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3439 | .readpage = ext4_readpage, |
3440 | .readpages = ext4_readpages, | |
3441 | .writepage = ext4_normal_writepage, | |
3442 | .sync_page = block_sync_page, | |
3443 | .write_begin = ext4_write_begin, | |
3444 | .write_end = ext4_writeback_write_end, | |
3445 | .bmap = ext4_bmap, | |
3446 | .invalidatepage = ext4_invalidatepage, | |
3447 | .releasepage = ext4_releasepage, | |
3448 | .direct_IO = ext4_direct_IO, | |
3449 | .migratepage = buffer_migrate_page, | |
3450 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
3451 | }; |
3452 | ||
617ba13b | 3453 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3454 | .readpage = ext4_readpage, |
3455 | .readpages = ext4_readpages, | |
3456 | .writepage = ext4_journalled_writepage, | |
3457 | .sync_page = block_sync_page, | |
3458 | .write_begin = ext4_write_begin, | |
3459 | .write_end = ext4_journalled_write_end, | |
3460 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3461 | .bmap = ext4_bmap, | |
3462 | .invalidatepage = ext4_invalidatepage, | |
3463 | .releasepage = ext4_releasepage, | |
3464 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
3465 | }; |
3466 | ||
64769240 | 3467 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3468 | .readpage = ext4_readpage, |
3469 | .readpages = ext4_readpages, | |
3470 | .writepage = ext4_da_writepage, | |
3471 | .writepages = ext4_da_writepages, | |
3472 | .sync_page = block_sync_page, | |
3473 | .write_begin = ext4_da_write_begin, | |
3474 | .write_end = ext4_da_write_end, | |
3475 | .bmap = ext4_bmap, | |
3476 | .invalidatepage = ext4_da_invalidatepage, | |
3477 | .releasepage = ext4_releasepage, | |
3478 | .direct_IO = ext4_direct_IO, | |
3479 | .migratepage = buffer_migrate_page, | |
3480 | .is_partially_uptodate = block_is_partially_uptodate, | |
64769240 AT |
3481 | }; |
3482 | ||
617ba13b | 3483 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3484 | { |
cd1aac32 AK |
3485 | if (ext4_should_order_data(inode) && |
3486 | test_opt(inode->i_sb, DELALLOC)) | |
3487 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3488 | else if (ext4_should_order_data(inode)) | |
617ba13b | 3489 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
3490 | else if (ext4_should_writeback_data(inode) && |
3491 | test_opt(inode->i_sb, DELALLOC)) | |
3492 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
3493 | else if (ext4_should_writeback_data(inode)) |
3494 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 3495 | else |
617ba13b | 3496 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
3497 | } |
3498 | ||
3499 | /* | |
617ba13b | 3500 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
3501 | * up to the end of the block which corresponds to `from'. |
3502 | * This required during truncate. We need to physically zero the tail end | |
3503 | * of that block so it doesn't yield old data if the file is later grown. | |
3504 | */ | |
cf108bca | 3505 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
3506 | struct address_space *mapping, loff_t from) |
3507 | { | |
617ba13b | 3508 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 3509 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
3510 | unsigned blocksize, length, pos; |
3511 | ext4_lblk_t iblock; | |
ac27a0ec DK |
3512 | struct inode *inode = mapping->host; |
3513 | struct buffer_head *bh; | |
cf108bca | 3514 | struct page *page; |
ac27a0ec | 3515 | int err = 0; |
ac27a0ec | 3516 | |
cf108bca JK |
3517 | page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT); |
3518 | if (!page) | |
3519 | return -EINVAL; | |
3520 | ||
ac27a0ec DK |
3521 | blocksize = inode->i_sb->s_blocksize; |
3522 | length = blocksize - (offset & (blocksize - 1)); | |
3523 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3524 | ||
3525 | /* | |
3526 | * For "nobh" option, we can only work if we don't need to | |
3527 | * read-in the page - otherwise we create buffers to do the IO. | |
3528 | */ | |
3529 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && | |
617ba13b | 3530 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
eebd2aa3 | 3531 | zero_user(page, offset, length); |
ac27a0ec DK |
3532 | set_page_dirty(page); |
3533 | goto unlock; | |
3534 | } | |
3535 | ||
3536 | if (!page_has_buffers(page)) | |
3537 | create_empty_buffers(page, blocksize, 0); | |
3538 | ||
3539 | /* Find the buffer that contains "offset" */ | |
3540 | bh = page_buffers(page); | |
3541 | pos = blocksize; | |
3542 | while (offset >= pos) { | |
3543 | bh = bh->b_this_page; | |
3544 | iblock++; | |
3545 | pos += blocksize; | |
3546 | } | |
3547 | ||
3548 | err = 0; | |
3549 | if (buffer_freed(bh)) { | |
3550 | BUFFER_TRACE(bh, "freed: skip"); | |
3551 | goto unlock; | |
3552 | } | |
3553 | ||
3554 | if (!buffer_mapped(bh)) { | |
3555 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3556 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3557 | /* unmapped? It's a hole - nothing to do */ |
3558 | if (!buffer_mapped(bh)) { | |
3559 | BUFFER_TRACE(bh, "still unmapped"); | |
3560 | goto unlock; | |
3561 | } | |
3562 | } | |
3563 | ||
3564 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3565 | if (PageUptodate(page)) | |
3566 | set_buffer_uptodate(bh); | |
3567 | ||
3568 | if (!buffer_uptodate(bh)) { | |
3569 | err = -EIO; | |
3570 | ll_rw_block(READ, 1, &bh); | |
3571 | wait_on_buffer(bh); | |
3572 | /* Uhhuh. Read error. Complain and punt. */ | |
3573 | if (!buffer_uptodate(bh)) | |
3574 | goto unlock; | |
3575 | } | |
3576 | ||
617ba13b | 3577 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3578 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3579 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3580 | if (err) |
3581 | goto unlock; | |
3582 | } | |
3583 | ||
eebd2aa3 | 3584 | zero_user(page, offset, length); |
ac27a0ec DK |
3585 | |
3586 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3587 | ||
3588 | err = 0; | |
617ba13b | 3589 | if (ext4_should_journal_data(inode)) { |
0390131b | 3590 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
ac27a0ec | 3591 | } else { |
617ba13b | 3592 | if (ext4_should_order_data(inode)) |
678aaf48 | 3593 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
3594 | mark_buffer_dirty(bh); |
3595 | } | |
3596 | ||
3597 | unlock: | |
3598 | unlock_page(page); | |
3599 | page_cache_release(page); | |
3600 | return err; | |
3601 | } | |
3602 | ||
3603 | /* | |
3604 | * Probably it should be a library function... search for first non-zero word | |
3605 | * or memcmp with zero_page, whatever is better for particular architecture. | |
3606 | * Linus? | |
3607 | */ | |
3608 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
3609 | { | |
3610 | while (p < q) | |
3611 | if (*p++) | |
3612 | return 0; | |
3613 | return 1; | |
3614 | } | |
3615 | ||
3616 | /** | |
617ba13b | 3617 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
3618 | * @inode: inode in question |
3619 | * @depth: depth of the affected branch | |
617ba13b | 3620 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
3621 | * @chain: place to store the pointers to partial indirect blocks |
3622 | * @top: place to the (detached) top of branch | |
3623 | * | |
617ba13b | 3624 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
3625 | * |
3626 | * When we do truncate() we may have to clean the ends of several | |
3627 | * indirect blocks but leave the blocks themselves alive. Block is | |
3628 | * partially truncated if some data below the new i_size is refered | |
3629 | * from it (and it is on the path to the first completely truncated | |
3630 | * data block, indeed). We have to free the top of that path along | |
3631 | * with everything to the right of the path. Since no allocation | |
617ba13b | 3632 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
3633 | * finishes, we may safely do the latter, but top of branch may |
3634 | * require special attention - pageout below the truncation point | |
3635 | * might try to populate it. | |
3636 | * | |
3637 | * We atomically detach the top of branch from the tree, store the | |
3638 | * block number of its root in *@top, pointers to buffer_heads of | |
3639 | * partially truncated blocks - in @chain[].bh and pointers to | |
3640 | * their last elements that should not be removed - in | |
3641 | * @chain[].p. Return value is the pointer to last filled element | |
3642 | * of @chain. | |
3643 | * | |
3644 | * The work left to caller to do the actual freeing of subtrees: | |
3645 | * a) free the subtree starting from *@top | |
3646 | * b) free the subtrees whose roots are stored in | |
3647 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
3648 | * c) free the subtrees growing from the inode past the @chain[0]. | |
3649 | * (no partially truncated stuff there). */ | |
3650 | ||
617ba13b | 3651 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
725d26d3 | 3652 | ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top) |
ac27a0ec DK |
3653 | { |
3654 | Indirect *partial, *p; | |
3655 | int k, err; | |
3656 | ||
3657 | *top = 0; | |
3658 | /* Make k index the deepest non-null offest + 1 */ | |
3659 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
3660 | ; | |
617ba13b | 3661 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
3662 | /* Writer: pointers */ |
3663 | if (!partial) | |
3664 | partial = chain + k-1; | |
3665 | /* | |
3666 | * If the branch acquired continuation since we've looked at it - | |
3667 | * fine, it should all survive and (new) top doesn't belong to us. | |
3668 | */ | |
3669 | if (!partial->key && *partial->p) | |
3670 | /* Writer: end */ | |
3671 | goto no_top; | |
af5bc92d | 3672 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
ac27a0ec DK |
3673 | ; |
3674 | /* | |
3675 | * OK, we've found the last block that must survive. The rest of our | |
3676 | * branch should be detached before unlocking. However, if that rest | |
3677 | * of branch is all ours and does not grow immediately from the inode | |
3678 | * it's easier to cheat and just decrement partial->p. | |
3679 | */ | |
3680 | if (p == chain + k - 1 && p > chain) { | |
3681 | p->p--; | |
3682 | } else { | |
3683 | *top = *p->p; | |
617ba13b | 3684 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
3685 | #if 0 |
3686 | *p->p = 0; | |
3687 | #endif | |
3688 | } | |
3689 | /* Writer: end */ | |
3690 | ||
af5bc92d | 3691 | while (partial > p) { |
ac27a0ec DK |
3692 | brelse(partial->bh); |
3693 | partial--; | |
3694 | } | |
3695 | no_top: | |
3696 | return partial; | |
3697 | } | |
3698 | ||
3699 | /* | |
3700 | * Zero a number of block pointers in either an inode or an indirect block. | |
3701 | * If we restart the transaction we must again get write access to the | |
3702 | * indirect block for further modification. | |
3703 | * | |
3704 | * We release `count' blocks on disk, but (last - first) may be greater | |
3705 | * than `count' because there can be holes in there. | |
3706 | */ | |
617ba13b MC |
3707 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
3708 | struct buffer_head *bh, ext4_fsblk_t block_to_free, | |
ac27a0ec DK |
3709 | unsigned long count, __le32 *first, __le32 *last) |
3710 | { | |
3711 | __le32 *p; | |
3712 | if (try_to_extend_transaction(handle, inode)) { | |
3713 | if (bh) { | |
0390131b FM |
3714 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
3715 | ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec | 3716 | } |
617ba13b MC |
3717 | ext4_mark_inode_dirty(handle, inode); |
3718 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
3719 | if (bh) { |
3720 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 3721 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3722 | } |
3723 | } | |
3724 | ||
3725 | /* | |
3726 | * Any buffers which are on the journal will be in memory. We find | |
dab291af | 3727 | * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget() |
ac27a0ec | 3728 | * on them. We've already detached each block from the file, so |
dab291af | 3729 | * bforget() in jbd2_journal_forget() should be safe. |
ac27a0ec | 3730 | * |
dab291af | 3731 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
ac27a0ec DK |
3732 | */ |
3733 | for (p = first; p < last; p++) { | |
3734 | u32 nr = le32_to_cpu(*p); | |
3735 | if (nr) { | |
1d03ec98 | 3736 | struct buffer_head *tbh; |
ac27a0ec DK |
3737 | |
3738 | *p = 0; | |
1d03ec98 AK |
3739 | tbh = sb_find_get_block(inode->i_sb, nr); |
3740 | ext4_forget(handle, 0, inode, tbh, nr); | |
ac27a0ec DK |
3741 | } |
3742 | } | |
3743 | ||
c9de560d | 3744 | ext4_free_blocks(handle, inode, block_to_free, count, 0); |
ac27a0ec DK |
3745 | } |
3746 | ||
3747 | /** | |
617ba13b | 3748 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
3749 | * @handle: handle for this transaction |
3750 | * @inode: inode we are dealing with | |
3751 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
3752 | * @first: array of block numbers | |
3753 | * @last: points immediately past the end of array | |
3754 | * | |
3755 | * We are freeing all blocks refered from that array (numbers are stored as | |
3756 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
3757 | * | |
3758 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
3759 | * blocks are contiguous then releasing them at one time will only affect one | |
3760 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
3761 | * actually use a lot of journal space. | |
3762 | * | |
3763 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
3764 | * block pointers. | |
3765 | */ | |
617ba13b | 3766 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
3767 | struct buffer_head *this_bh, |
3768 | __le32 *first, __le32 *last) | |
3769 | { | |
617ba13b | 3770 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
3771 | unsigned long count = 0; /* Number of blocks in the run */ |
3772 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
3773 | corresponding to | |
3774 | block_to_free */ | |
617ba13b | 3775 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
3776 | __le32 *p; /* Pointer into inode/ind |
3777 | for current block */ | |
3778 | int err; | |
3779 | ||
3780 | if (this_bh) { /* For indirect block */ | |
3781 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 3782 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
3783 | /* Important: if we can't update the indirect pointers |
3784 | * to the blocks, we can't free them. */ | |
3785 | if (err) | |
3786 | return; | |
3787 | } | |
3788 | ||
3789 | for (p = first; p < last; p++) { | |
3790 | nr = le32_to_cpu(*p); | |
3791 | if (nr) { | |
3792 | /* accumulate blocks to free if they're contiguous */ | |
3793 | if (count == 0) { | |
3794 | block_to_free = nr; | |
3795 | block_to_free_p = p; | |
3796 | count = 1; | |
3797 | } else if (nr == block_to_free + count) { | |
3798 | count++; | |
3799 | } else { | |
617ba13b | 3800 | ext4_clear_blocks(handle, inode, this_bh, |
ac27a0ec DK |
3801 | block_to_free, |
3802 | count, block_to_free_p, p); | |
3803 | block_to_free = nr; | |
3804 | block_to_free_p = p; | |
3805 | count = 1; | |
3806 | } | |
3807 | } | |
3808 | } | |
3809 | ||
3810 | if (count > 0) | |
617ba13b | 3811 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
3812 | count, block_to_free_p, p); |
3813 | ||
3814 | if (this_bh) { | |
0390131b | 3815 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
71dc8fbc DG |
3816 | |
3817 | /* | |
3818 | * The buffer head should have an attached journal head at this | |
3819 | * point. However, if the data is corrupted and an indirect | |
3820 | * block pointed to itself, it would have been detached when | |
3821 | * the block was cleared. Check for this instead of OOPSing. | |
3822 | */ | |
e7f07968 | 3823 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
0390131b | 3824 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
71dc8fbc DG |
3825 | else |
3826 | ext4_error(inode->i_sb, __func__, | |
3827 | "circular indirect block detected, " | |
3828 | "inode=%lu, block=%llu", | |
3829 | inode->i_ino, | |
3830 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
3831 | } |
3832 | } | |
3833 | ||
3834 | /** | |
617ba13b | 3835 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
3836 | * @handle: JBD handle for this transaction |
3837 | * @inode: inode we are dealing with | |
3838 | * @parent_bh: the buffer_head which contains *@first and *@last | |
3839 | * @first: array of block numbers | |
3840 | * @last: pointer immediately past the end of array | |
3841 | * @depth: depth of the branches to free | |
3842 | * | |
3843 | * We are freeing all blocks refered from these branches (numbers are | |
3844 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
3845 | * appropriately. | |
3846 | */ | |
617ba13b | 3847 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
3848 | struct buffer_head *parent_bh, |
3849 | __le32 *first, __le32 *last, int depth) | |
3850 | { | |
617ba13b | 3851 | ext4_fsblk_t nr; |
ac27a0ec DK |
3852 | __le32 *p; |
3853 | ||
0390131b | 3854 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
3855 | return; |
3856 | ||
3857 | if (depth--) { | |
3858 | struct buffer_head *bh; | |
617ba13b | 3859 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
3860 | p = last; |
3861 | while (--p >= first) { | |
3862 | nr = le32_to_cpu(*p); | |
3863 | if (!nr) | |
3864 | continue; /* A hole */ | |
3865 | ||
3866 | /* Go read the buffer for the next level down */ | |
3867 | bh = sb_bread(inode->i_sb, nr); | |
3868 | ||
3869 | /* | |
3870 | * A read failure? Report error and clear slot | |
3871 | * (should be rare). | |
3872 | */ | |
3873 | if (!bh) { | |
617ba13b | 3874 | ext4_error(inode->i_sb, "ext4_free_branches", |
2ae02107 | 3875 | "Read failure, inode=%lu, block=%llu", |
ac27a0ec DK |
3876 | inode->i_ino, nr); |
3877 | continue; | |
3878 | } | |
3879 | ||
3880 | /* This zaps the entire block. Bottom up. */ | |
3881 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 3882 | ext4_free_branches(handle, inode, bh, |
af5bc92d TT |
3883 | (__le32 *) bh->b_data, |
3884 | (__le32 *) bh->b_data + addr_per_block, | |
3885 | depth); | |
ac27a0ec DK |
3886 | |
3887 | /* | |
3888 | * We've probably journalled the indirect block several | |
3889 | * times during the truncate. But it's no longer | |
3890 | * needed and we now drop it from the transaction via | |
dab291af | 3891 | * jbd2_journal_revoke(). |
ac27a0ec DK |
3892 | * |
3893 | * That's easy if it's exclusively part of this | |
3894 | * transaction. But if it's part of the committing | |
dab291af | 3895 | * transaction then jbd2_journal_forget() will simply |
ac27a0ec | 3896 | * brelse() it. That means that if the underlying |
617ba13b | 3897 | * block is reallocated in ext4_get_block(), |
ac27a0ec DK |
3898 | * unmap_underlying_metadata() will find this block |
3899 | * and will try to get rid of it. damn, damn. | |
3900 | * | |
3901 | * If this block has already been committed to the | |
3902 | * journal, a revoke record will be written. And | |
3903 | * revoke records must be emitted *before* clearing | |
3904 | * this block's bit in the bitmaps. | |
3905 | */ | |
617ba13b | 3906 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
ac27a0ec DK |
3907 | |
3908 | /* | |
3909 | * Everything below this this pointer has been | |
3910 | * released. Now let this top-of-subtree go. | |
3911 | * | |
3912 | * We want the freeing of this indirect block to be | |
3913 | * atomic in the journal with the updating of the | |
3914 | * bitmap block which owns it. So make some room in | |
3915 | * the journal. | |
3916 | * | |
3917 | * We zero the parent pointer *after* freeing its | |
3918 | * pointee in the bitmaps, so if extend_transaction() | |
3919 | * for some reason fails to put the bitmap changes and | |
3920 | * the release into the same transaction, recovery | |
3921 | * will merely complain about releasing a free block, | |
3922 | * rather than leaking blocks. | |
3923 | */ | |
0390131b | 3924 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
3925 | return; |
3926 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b MC |
3927 | ext4_mark_inode_dirty(handle, inode); |
3928 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
3929 | } |
3930 | ||
c9de560d | 3931 | ext4_free_blocks(handle, inode, nr, 1, 1); |
ac27a0ec DK |
3932 | |
3933 | if (parent_bh) { | |
3934 | /* | |
3935 | * The block which we have just freed is | |
3936 | * pointed to by an indirect block: journal it | |
3937 | */ | |
3938 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 3939 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
3940 | parent_bh)){ |
3941 | *p = 0; | |
3942 | BUFFER_TRACE(parent_bh, | |
0390131b FM |
3943 | "call ext4_handle_dirty_metadata"); |
3944 | ext4_handle_dirty_metadata(handle, | |
3945 | inode, | |
3946 | parent_bh); | |
ac27a0ec DK |
3947 | } |
3948 | } | |
3949 | } | |
3950 | } else { | |
3951 | /* We have reached the bottom of the tree. */ | |
3952 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 3953 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
3954 | } |
3955 | } | |
3956 | ||
91ef4caf DG |
3957 | int ext4_can_truncate(struct inode *inode) |
3958 | { | |
3959 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3960 | return 0; | |
3961 | if (S_ISREG(inode->i_mode)) | |
3962 | return 1; | |
3963 | if (S_ISDIR(inode->i_mode)) | |
3964 | return 1; | |
3965 | if (S_ISLNK(inode->i_mode)) | |
3966 | return !ext4_inode_is_fast_symlink(inode); | |
3967 | return 0; | |
3968 | } | |
3969 | ||
ac27a0ec | 3970 | /* |
617ba13b | 3971 | * ext4_truncate() |
ac27a0ec | 3972 | * |
617ba13b MC |
3973 | * We block out ext4_get_block() block instantiations across the entire |
3974 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3975 | * simultaneously on behalf of the same inode. |
3976 | * | |
3977 | * As we work through the truncate and commmit bits of it to the journal there | |
3978 | * is one core, guiding principle: the file's tree must always be consistent on | |
3979 | * disk. We must be able to restart the truncate after a crash. | |
3980 | * | |
3981 | * The file's tree may be transiently inconsistent in memory (although it | |
3982 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3983 | * the contents of (the filesystem + the journal) must be consistent and | |
3984 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3985 | * left-to-right works OK too). | |
3986 | * | |
3987 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3988 | * truncate against the orphan inode list. | |
3989 | * | |
3990 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3991 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3992 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3993 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3994 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3995 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3996 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3997 | */ |
617ba13b | 3998 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
3999 | { |
4000 | handle_t *handle; | |
617ba13b | 4001 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 4002 | __le32 *i_data = ei->i_data; |
617ba13b | 4003 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 4004 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 4005 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
4006 | Indirect chain[4]; |
4007 | Indirect *partial; | |
4008 | __le32 nr = 0; | |
4009 | int n; | |
725d26d3 | 4010 | ext4_lblk_t last_block; |
ac27a0ec | 4011 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 4012 | |
91ef4caf | 4013 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4014 | return; |
4015 | ||
afd4672d | 4016 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
7d8f9f7d TT |
4017 | ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE; |
4018 | ||
1d03ec98 | 4019 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
cf108bca | 4020 | ext4_ext_truncate(inode); |
1d03ec98 AK |
4021 | return; |
4022 | } | |
a86c6181 | 4023 | |
ac27a0ec | 4024 | handle = start_transaction(inode); |
cf108bca | 4025 | if (IS_ERR(handle)) |
ac27a0ec | 4026 | return; /* AKPM: return what? */ |
ac27a0ec DK |
4027 | |
4028 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 4029 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec | 4030 | |
cf108bca JK |
4031 | if (inode->i_size & (blocksize - 1)) |
4032 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
4033 | goto out_stop; | |
ac27a0ec | 4034 | |
617ba13b | 4035 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
4036 | if (n == 0) |
4037 | goto out_stop; /* error */ | |
4038 | ||
4039 | /* | |
4040 | * OK. This truncate is going to happen. We add the inode to the | |
4041 | * orphan list, so that if this truncate spans multiple transactions, | |
4042 | * and we crash, we will resume the truncate when the filesystem | |
4043 | * recovers. It also marks the inode dirty, to catch the new size. | |
4044 | * | |
4045 | * Implication: the file must always be in a sane, consistent | |
4046 | * truncatable state while each transaction commits. | |
4047 | */ | |
617ba13b | 4048 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
4049 | goto out_stop; |
4050 | ||
632eaeab MC |
4051 | /* |
4052 | * From here we block out all ext4_get_block() callers who want to | |
4053 | * modify the block allocation tree. | |
4054 | */ | |
4055 | down_write(&ei->i_data_sem); | |
b4df2030 | 4056 | |
c2ea3fde | 4057 | ext4_discard_preallocations(inode); |
b4df2030 | 4058 | |
ac27a0ec DK |
4059 | /* |
4060 | * The orphan list entry will now protect us from any crash which | |
4061 | * occurs before the truncate completes, so it is now safe to propagate | |
4062 | * the new, shorter inode size (held for now in i_size) into the | |
4063 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 4064 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
4065 | */ |
4066 | ei->i_disksize = inode->i_size; | |
4067 | ||
ac27a0ec | 4068 | if (n == 1) { /* direct blocks */ |
617ba13b MC |
4069 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
4070 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
4071 | goto do_indirects; |
4072 | } | |
4073 | ||
617ba13b | 4074 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
4075 | /* Kill the top of shared branch (not detached) */ |
4076 | if (nr) { | |
4077 | if (partial == chain) { | |
4078 | /* Shared branch grows from the inode */ | |
617ba13b | 4079 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
4080 | &nr, &nr+1, (chain+n-1) - partial); |
4081 | *partial->p = 0; | |
4082 | /* | |
4083 | * We mark the inode dirty prior to restart, | |
4084 | * and prior to stop. No need for it here. | |
4085 | */ | |
4086 | } else { | |
4087 | /* Shared branch grows from an indirect block */ | |
4088 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 4089 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
4090 | partial->p, |
4091 | partial->p+1, (chain+n-1) - partial); | |
4092 | } | |
4093 | } | |
4094 | /* Clear the ends of indirect blocks on the shared branch */ | |
4095 | while (partial > chain) { | |
617ba13b | 4096 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
4097 | (__le32*)partial->bh->b_data+addr_per_block, |
4098 | (chain+n-1) - partial); | |
4099 | BUFFER_TRACE(partial->bh, "call brelse"); | |
4100 | brelse (partial->bh); | |
4101 | partial--; | |
4102 | } | |
4103 | do_indirects: | |
4104 | /* Kill the remaining (whole) subtrees */ | |
4105 | switch (offsets[0]) { | |
4106 | default: | |
617ba13b | 4107 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 4108 | if (nr) { |
617ba13b MC |
4109 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
4110 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 4111 | } |
617ba13b MC |
4112 | case EXT4_IND_BLOCK: |
4113 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 4114 | if (nr) { |
617ba13b MC |
4115 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
4116 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 4117 | } |
617ba13b MC |
4118 | case EXT4_DIND_BLOCK: |
4119 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 4120 | if (nr) { |
617ba13b MC |
4121 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
4122 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 4123 | } |
617ba13b | 4124 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
4125 | ; |
4126 | } | |
4127 | ||
0e855ac8 | 4128 | up_write(&ei->i_data_sem); |
ef7f3835 | 4129 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 4130 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
4131 | |
4132 | /* | |
4133 | * In a multi-transaction truncate, we only make the final transaction | |
4134 | * synchronous | |
4135 | */ | |
4136 | if (IS_SYNC(inode)) | |
0390131b | 4137 | ext4_handle_sync(handle); |
ac27a0ec DK |
4138 | out_stop: |
4139 | /* | |
4140 | * If this was a simple ftruncate(), and the file will remain alive | |
4141 | * then we need to clear up the orphan record which we created above. | |
4142 | * However, if this was a real unlink then we were called by | |
617ba13b | 4143 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
4144 | * orphan info for us. |
4145 | */ | |
4146 | if (inode->i_nlink) | |
617ba13b | 4147 | ext4_orphan_del(handle, inode); |
ac27a0ec | 4148 | |
617ba13b | 4149 | ext4_journal_stop(handle); |
ac27a0ec DK |
4150 | } |
4151 | ||
ac27a0ec | 4152 | /* |
617ba13b | 4153 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4154 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4155 | * data in memory that is needed to recreate the on-disk version of this | |
4156 | * inode. | |
4157 | */ | |
617ba13b MC |
4158 | static int __ext4_get_inode_loc(struct inode *inode, |
4159 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4160 | { |
240799cd TT |
4161 | struct ext4_group_desc *gdp; |
4162 | struct buffer_head *bh; | |
4163 | struct super_block *sb = inode->i_sb; | |
4164 | ext4_fsblk_t block; | |
4165 | int inodes_per_block, inode_offset; | |
4166 | ||
3a06d778 | 4167 | iloc->bh = NULL; |
240799cd TT |
4168 | if (!ext4_valid_inum(sb, inode->i_ino)) |
4169 | return -EIO; | |
ac27a0ec | 4170 | |
240799cd TT |
4171 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4172 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4173 | if (!gdp) | |
ac27a0ec DK |
4174 | return -EIO; |
4175 | ||
240799cd TT |
4176 | /* |
4177 | * Figure out the offset within the block group inode table | |
4178 | */ | |
4179 | inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb)); | |
4180 | inode_offset = ((inode->i_ino - 1) % | |
4181 | EXT4_INODES_PER_GROUP(sb)); | |
4182 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4183 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4184 | ||
4185 | bh = sb_getblk(sb, block); | |
ac27a0ec | 4186 | if (!bh) { |
240799cd TT |
4187 | ext4_error(sb, "ext4_get_inode_loc", "unable to read " |
4188 | "inode block - inode=%lu, block=%llu", | |
4189 | inode->i_ino, block); | |
ac27a0ec DK |
4190 | return -EIO; |
4191 | } | |
4192 | if (!buffer_uptodate(bh)) { | |
4193 | lock_buffer(bh); | |
9c83a923 HK |
4194 | |
4195 | /* | |
4196 | * If the buffer has the write error flag, we have failed | |
4197 | * to write out another inode in the same block. In this | |
4198 | * case, we don't have to read the block because we may | |
4199 | * read the old inode data successfully. | |
4200 | */ | |
4201 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4202 | set_buffer_uptodate(bh); | |
4203 | ||
ac27a0ec DK |
4204 | if (buffer_uptodate(bh)) { |
4205 | /* someone brought it uptodate while we waited */ | |
4206 | unlock_buffer(bh); | |
4207 | goto has_buffer; | |
4208 | } | |
4209 | ||
4210 | /* | |
4211 | * If we have all information of the inode in memory and this | |
4212 | * is the only valid inode in the block, we need not read the | |
4213 | * block. | |
4214 | */ | |
4215 | if (in_mem) { | |
4216 | struct buffer_head *bitmap_bh; | |
240799cd | 4217 | int i, start; |
ac27a0ec | 4218 | |
240799cd | 4219 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4220 | |
240799cd TT |
4221 | /* Is the inode bitmap in cache? */ |
4222 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
4223 | if (!bitmap_bh) |
4224 | goto make_io; | |
4225 | ||
4226 | /* | |
4227 | * If the inode bitmap isn't in cache then the | |
4228 | * optimisation may end up performing two reads instead | |
4229 | * of one, so skip it. | |
4230 | */ | |
4231 | if (!buffer_uptodate(bitmap_bh)) { | |
4232 | brelse(bitmap_bh); | |
4233 | goto make_io; | |
4234 | } | |
240799cd | 4235 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4236 | if (i == inode_offset) |
4237 | continue; | |
617ba13b | 4238 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4239 | break; |
4240 | } | |
4241 | brelse(bitmap_bh); | |
240799cd | 4242 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4243 | /* all other inodes are free, so skip I/O */ |
4244 | memset(bh->b_data, 0, bh->b_size); | |
4245 | set_buffer_uptodate(bh); | |
4246 | unlock_buffer(bh); | |
4247 | goto has_buffer; | |
4248 | } | |
4249 | } | |
4250 | ||
4251 | make_io: | |
240799cd TT |
4252 | /* |
4253 | * If we need to do any I/O, try to pre-readahead extra | |
4254 | * blocks from the inode table. | |
4255 | */ | |
4256 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4257 | ext4_fsblk_t b, end, table; | |
4258 | unsigned num; | |
4259 | ||
4260 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4261 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
4262 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
4263 | if (table > b) | |
4264 | b = table; | |
4265 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
4266 | num = EXT4_INODES_PER_GROUP(sb); | |
4267 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4268 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 4269 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4270 | table += num / inodes_per_block; |
4271 | if (end > table) | |
4272 | end = table; | |
4273 | while (b <= end) | |
4274 | sb_breadahead(sb, b++); | |
4275 | } | |
4276 | ||
ac27a0ec DK |
4277 | /* |
4278 | * There are other valid inodes in the buffer, this inode | |
4279 | * has in-inode xattrs, or we don't have this inode in memory. | |
4280 | * Read the block from disk. | |
4281 | */ | |
4282 | get_bh(bh); | |
4283 | bh->b_end_io = end_buffer_read_sync; | |
4284 | submit_bh(READ_META, bh); | |
4285 | wait_on_buffer(bh); | |
4286 | if (!buffer_uptodate(bh)) { | |
240799cd TT |
4287 | ext4_error(sb, __func__, |
4288 | "unable to read inode block - inode=%lu, " | |
4289 | "block=%llu", inode->i_ino, block); | |
ac27a0ec DK |
4290 | brelse(bh); |
4291 | return -EIO; | |
4292 | } | |
4293 | } | |
4294 | has_buffer: | |
4295 | iloc->bh = bh; | |
4296 | return 0; | |
4297 | } | |
4298 | ||
617ba13b | 4299 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4300 | { |
4301 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b MC |
4302 | return __ext4_get_inode_loc(inode, iloc, |
4303 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); | |
ac27a0ec DK |
4304 | } |
4305 | ||
617ba13b | 4306 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4307 | { |
617ba13b | 4308 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4309 | |
4310 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4311 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4312 | inode->i_flags |= S_SYNC; |
617ba13b | 4313 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4314 | inode->i_flags |= S_APPEND; |
617ba13b | 4315 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4316 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4317 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4318 | inode->i_flags |= S_NOATIME; |
617ba13b | 4319 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4320 | inode->i_flags |= S_DIRSYNC; |
4321 | } | |
4322 | ||
ff9ddf7e JK |
4323 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4324 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4325 | { | |
4326 | unsigned int flags = ei->vfs_inode.i_flags; | |
4327 | ||
4328 | ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4329 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL); | |
4330 | if (flags & S_SYNC) | |
4331 | ei->i_flags |= EXT4_SYNC_FL; | |
4332 | if (flags & S_APPEND) | |
4333 | ei->i_flags |= EXT4_APPEND_FL; | |
4334 | if (flags & S_IMMUTABLE) | |
4335 | ei->i_flags |= EXT4_IMMUTABLE_FL; | |
4336 | if (flags & S_NOATIME) | |
4337 | ei->i_flags |= EXT4_NOATIME_FL; | |
4338 | if (flags & S_DIRSYNC) | |
4339 | ei->i_flags |= EXT4_DIRSYNC_FL; | |
4340 | } | |
0fc1b451 AK |
4341 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
4342 | struct ext4_inode_info *ei) | |
4343 | { | |
4344 | blkcnt_t i_blocks ; | |
8180a562 AK |
4345 | struct inode *inode = &(ei->vfs_inode); |
4346 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4347 | |
4348 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4349 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4350 | /* we are using combined 48 bit field */ | |
4351 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4352 | le32_to_cpu(raw_inode->i_blocks_lo); | |
8180a562 AK |
4353 | if (ei->i_flags & EXT4_HUGE_FILE_FL) { |
4354 | /* i_blocks represent file system block size */ | |
4355 | return i_blocks << (inode->i_blkbits - 9); | |
4356 | } else { | |
4357 | return i_blocks; | |
4358 | } | |
0fc1b451 AK |
4359 | } else { |
4360 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4361 | } | |
4362 | } | |
ff9ddf7e | 4363 | |
1d1fe1ee | 4364 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4365 | { |
617ba13b MC |
4366 | struct ext4_iloc iloc; |
4367 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4368 | struct ext4_inode_info *ei; |
ac27a0ec | 4369 | struct buffer_head *bh; |
1d1fe1ee DH |
4370 | struct inode *inode; |
4371 | long ret; | |
ac27a0ec DK |
4372 | int block; |
4373 | ||
1d1fe1ee DH |
4374 | inode = iget_locked(sb, ino); |
4375 | if (!inode) | |
4376 | return ERR_PTR(-ENOMEM); | |
4377 | if (!(inode->i_state & I_NEW)) | |
4378 | return inode; | |
4379 | ||
4380 | ei = EXT4_I(inode); | |
03010a33 | 4381 | #ifdef CONFIG_EXT4_FS_POSIX_ACL |
617ba13b MC |
4382 | ei->i_acl = EXT4_ACL_NOT_CACHED; |
4383 | ei->i_default_acl = EXT4_ACL_NOT_CACHED; | |
ac27a0ec | 4384 | #endif |
ac27a0ec | 4385 | |
1d1fe1ee DH |
4386 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4387 | if (ret < 0) | |
ac27a0ec DK |
4388 | goto bad_inode; |
4389 | bh = iloc.bh; | |
617ba13b | 4390 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
4391 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
4392 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
4393 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4394 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4395 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4396 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
4397 | } | |
4398 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec DK |
4399 | |
4400 | ei->i_state = 0; | |
4401 | ei->i_dir_start_lookup = 0; | |
4402 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4403 | /* We now have enough fields to check if the inode was active or not. | |
4404 | * This is needed because nfsd might try to access dead inodes | |
4405 | * the test is that same one that e2fsck uses | |
4406 | * NeilBrown 1999oct15 | |
4407 | */ | |
4408 | if (inode->i_nlink == 0) { | |
4409 | if (inode->i_mode == 0 || | |
617ba13b | 4410 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 4411 | /* this inode is deleted */ |
af5bc92d | 4412 | brelse(bh); |
1d1fe1ee | 4413 | ret = -ESTALE; |
ac27a0ec DK |
4414 | goto bad_inode; |
4415 | } | |
4416 | /* The only unlinked inodes we let through here have | |
4417 | * valid i_mode and are being read by the orphan | |
4418 | * recovery code: that's fine, we're about to complete | |
4419 | * the process of deleting those. */ | |
4420 | } | |
ac27a0ec | 4421 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4422 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4423 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4424 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4425 | ei->i_file_acl |= |
4426 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4427 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec DK |
4428 | ei->i_disksize = inode->i_size; |
4429 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
4430 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4431 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4432 | /* |
4433 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4434 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4435 | */ | |
617ba13b | 4436 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4437 | ei->i_data[block] = raw_inode->i_block[block]; |
4438 | INIT_LIST_HEAD(&ei->i_orphan); | |
4439 | ||
0040d987 | 4440 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 4441 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 4442 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 4443 | EXT4_INODE_SIZE(inode->i_sb)) { |
af5bc92d | 4444 | brelse(bh); |
1d1fe1ee | 4445 | ret = -EIO; |
ac27a0ec | 4446 | goto bad_inode; |
e5d2861f | 4447 | } |
ac27a0ec DK |
4448 | if (ei->i_extra_isize == 0) { |
4449 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4450 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4451 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
4452 | } else { |
4453 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 4454 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 4455 | ei->i_extra_isize; |
617ba13b MC |
4456 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
4457 | ei->i_state |= EXT4_STATE_XATTR; | |
ac27a0ec DK |
4458 | } |
4459 | } else | |
4460 | ei->i_extra_isize = 0; | |
4461 | ||
ef7f3835 KS |
4462 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4463 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4464 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4465 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4466 | ||
25ec56b5 JNC |
4467 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4468 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4469 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4470 | inode->i_version |= | |
4471 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4472 | } | |
4473 | ||
c4b5a614 | 4474 | ret = 0; |
485c26ec TT |
4475 | if (ei->i_file_acl && |
4476 | ((ei->i_file_acl < | |
4477 | (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) + | |
4478 | EXT4_SB(sb)->s_gdb_count)) || | |
4479 | (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) { | |
4480 | ext4_error(sb, __func__, | |
4481 | "bad extended attribute block %llu in inode #%lu", | |
4482 | ei->i_file_acl, inode->i_ino); | |
4483 | ret = -EIO; | |
4484 | goto bad_inode; | |
4485 | } else if (ei->i_flags & EXT4_EXTENTS_FL) { | |
c4b5a614 TT |
4486 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
4487 | (S_ISLNK(inode->i_mode) && | |
4488 | !ext4_inode_is_fast_symlink(inode))) | |
4489 | /* Validate extent which is part of inode */ | |
4490 | ret = ext4_ext_check_inode(inode); | |
fe2c8191 TN |
4491 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
4492 | (S_ISLNK(inode->i_mode) && | |
4493 | !ext4_inode_is_fast_symlink(inode))) { | |
4494 | /* Validate block references which are part of inode */ | |
4495 | ret = ext4_check_inode_blockref(inode); | |
4496 | } | |
4497 | if (ret) { | |
4498 | brelse(bh); | |
4499 | goto bad_inode; | |
7a262f7c AK |
4500 | } |
4501 | ||
ac27a0ec | 4502 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4503 | inode->i_op = &ext4_file_inode_operations; |
4504 | inode->i_fop = &ext4_file_operations; | |
4505 | ext4_set_aops(inode); | |
ac27a0ec | 4506 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4507 | inode->i_op = &ext4_dir_inode_operations; |
4508 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4509 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4510 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4511 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4512 | nd_terminate_link(ei->i_data, inode->i_size, |
4513 | sizeof(ei->i_data) - 1); | |
4514 | } else { | |
617ba13b MC |
4515 | inode->i_op = &ext4_symlink_inode_operations; |
4516 | ext4_set_aops(inode); | |
ac27a0ec | 4517 | } |
563bdd61 TT |
4518 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4519 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4520 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4521 | if (raw_inode->i_block[0]) |
4522 | init_special_inode(inode, inode->i_mode, | |
4523 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4524 | else | |
4525 | init_special_inode(inode, inode->i_mode, | |
4526 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 TT |
4527 | } else { |
4528 | brelse(bh); | |
4529 | ret = -EIO; | |
4530 | ext4_error(inode->i_sb, __func__, | |
4531 | "bogus i_mode (%o) for inode=%lu", | |
4532 | inode->i_mode, inode->i_ino); | |
4533 | goto bad_inode; | |
ac27a0ec | 4534 | } |
af5bc92d | 4535 | brelse(iloc.bh); |
617ba13b | 4536 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4537 | unlock_new_inode(inode); |
4538 | return inode; | |
ac27a0ec DK |
4539 | |
4540 | bad_inode: | |
1d1fe1ee DH |
4541 | iget_failed(inode); |
4542 | return ERR_PTR(ret); | |
ac27a0ec DK |
4543 | } |
4544 | ||
0fc1b451 AK |
4545 | static int ext4_inode_blocks_set(handle_t *handle, |
4546 | struct ext4_inode *raw_inode, | |
4547 | struct ext4_inode_info *ei) | |
4548 | { | |
4549 | struct inode *inode = &(ei->vfs_inode); | |
4550 | u64 i_blocks = inode->i_blocks; | |
4551 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4552 | |
4553 | if (i_blocks <= ~0U) { | |
4554 | /* | |
4555 | * i_blocks can be represnted in a 32 bit variable | |
4556 | * as multiple of 512 bytes | |
4557 | */ | |
8180a562 | 4558 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4559 | raw_inode->i_blocks_high = 0; |
8180a562 | 4560 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
f287a1a5 TT |
4561 | return 0; |
4562 | } | |
4563 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4564 | return -EFBIG; | |
4565 | ||
4566 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4567 | /* |
4568 | * i_blocks can be represented in a 48 bit variable | |
4569 | * as multiple of 512 bytes | |
4570 | */ | |
8180a562 | 4571 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4572 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
8180a562 | 4573 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
0fc1b451 | 4574 | } else { |
8180a562 AK |
4575 | ei->i_flags |= EXT4_HUGE_FILE_FL; |
4576 | /* i_block is stored in file system block size */ | |
4577 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4578 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4579 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4580 | } |
f287a1a5 | 4581 | return 0; |
0fc1b451 AK |
4582 | } |
4583 | ||
ac27a0ec DK |
4584 | /* |
4585 | * Post the struct inode info into an on-disk inode location in the | |
4586 | * buffer-cache. This gobbles the caller's reference to the | |
4587 | * buffer_head in the inode location struct. | |
4588 | * | |
4589 | * The caller must have write access to iloc->bh. | |
4590 | */ | |
617ba13b | 4591 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4592 | struct inode *inode, |
617ba13b | 4593 | struct ext4_iloc *iloc) |
ac27a0ec | 4594 | { |
617ba13b MC |
4595 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4596 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4597 | struct buffer_head *bh = iloc->bh; |
4598 | int err = 0, rc, block; | |
4599 | ||
4600 | /* For fields not not tracking in the in-memory inode, | |
4601 | * initialise them to zero for new inodes. */ | |
617ba13b MC |
4602 | if (ei->i_state & EXT4_STATE_NEW) |
4603 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
ac27a0ec | 4604 | |
ff9ddf7e | 4605 | ext4_get_inode_flags(ei); |
ac27a0ec | 4606 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 4607 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4608 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
4609 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
4610 | /* | |
4611 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4612 | * re-used with the upper 16 bits of the uid/gid intact | |
4613 | */ | |
af5bc92d | 4614 | if (!ei->i_dtime) { |
ac27a0ec DK |
4615 | raw_inode->i_uid_high = |
4616 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
4617 | raw_inode->i_gid_high = | |
4618 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
4619 | } else { | |
4620 | raw_inode->i_uid_high = 0; | |
4621 | raw_inode->i_gid_high = 0; | |
4622 | } | |
4623 | } else { | |
4624 | raw_inode->i_uid_low = | |
4625 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
4626 | raw_inode->i_gid_low = | |
4627 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
4628 | raw_inode->i_uid_high = 0; | |
4629 | raw_inode->i_gid_high = 0; | |
4630 | } | |
4631 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4632 | |
4633 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4634 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4635 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4636 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4637 | ||
0fc1b451 AK |
4638 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4639 | goto out_brelse; | |
ac27a0ec | 4640 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
267e4db9 AK |
4641 | /* clear the migrate flag in the raw_inode */ |
4642 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE); | |
9b8f1f01 MC |
4643 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4644 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4645 | raw_inode->i_file_acl_high = |
4646 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4647 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
4648 | ext4_isize_set(raw_inode, ei->i_disksize); |
4649 | if (ei->i_disksize > 0x7fffffffULL) { | |
4650 | struct super_block *sb = inode->i_sb; | |
4651 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4652 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4653 | EXT4_SB(sb)->s_es->s_rev_level == | |
4654 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4655 | /* If this is the first large file | |
4656 | * created, add a flag to the superblock. | |
4657 | */ | |
4658 | err = ext4_journal_get_write_access(handle, | |
4659 | EXT4_SB(sb)->s_sbh); | |
4660 | if (err) | |
4661 | goto out_brelse; | |
4662 | ext4_update_dynamic_rev(sb); | |
4663 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4664 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 4665 | sb->s_dirt = 1; |
0390131b FM |
4666 | ext4_handle_sync(handle); |
4667 | err = ext4_handle_dirty_metadata(handle, inode, | |
a48380f7 | 4668 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
4669 | } |
4670 | } | |
4671 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4672 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4673 | if (old_valid_dev(inode->i_rdev)) { | |
4674 | raw_inode->i_block[0] = | |
4675 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4676 | raw_inode->i_block[1] = 0; | |
4677 | } else { | |
4678 | raw_inode->i_block[0] = 0; | |
4679 | raw_inode->i_block[1] = | |
4680 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4681 | raw_inode->i_block[2] = 0; | |
4682 | } | |
617ba13b | 4683 | } else for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4684 | raw_inode->i_block[block] = ei->i_data[block]; |
4685 | ||
25ec56b5 JNC |
4686 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4687 | if (ei->i_extra_isize) { | |
4688 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4689 | raw_inode->i_version_hi = | |
4690 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4691 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4692 | } |
4693 | ||
0390131b FM |
4694 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
4695 | rc = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
4696 | if (!err) |
4697 | err = rc; | |
617ba13b | 4698 | ei->i_state &= ~EXT4_STATE_NEW; |
ac27a0ec DK |
4699 | |
4700 | out_brelse: | |
af5bc92d | 4701 | brelse(bh); |
617ba13b | 4702 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4703 | return err; |
4704 | } | |
4705 | ||
4706 | /* | |
617ba13b | 4707 | * ext4_write_inode() |
ac27a0ec DK |
4708 | * |
4709 | * We are called from a few places: | |
4710 | * | |
4711 | * - Within generic_file_write() for O_SYNC files. | |
4712 | * Here, there will be no transaction running. We wait for any running | |
4713 | * trasnaction to commit. | |
4714 | * | |
4715 | * - Within sys_sync(), kupdate and such. | |
4716 | * We wait on commit, if tol to. | |
4717 | * | |
4718 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4719 | * Here we simply return. We can't afford to block kswapd on the | |
4720 | * journal commit. | |
4721 | * | |
4722 | * In all cases it is actually safe for us to return without doing anything, | |
4723 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4724 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4725 | * knfsd. |
4726 | * | |
4727 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4728 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4729 | * which we are interested. | |
4730 | * | |
4731 | * It would be a bug for them to not do this. The code: | |
4732 | * | |
4733 | * mark_inode_dirty(inode) | |
4734 | * stuff(); | |
4735 | * inode->i_size = expr; | |
4736 | * | |
4737 | * is in error because a kswapd-driven write_inode() could occur while | |
4738 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4739 | * will no longer be on the superblock's dirty inode list. | |
4740 | */ | |
617ba13b | 4741 | int ext4_write_inode(struct inode *inode, int wait) |
ac27a0ec DK |
4742 | { |
4743 | if (current->flags & PF_MEMALLOC) | |
4744 | return 0; | |
4745 | ||
617ba13b | 4746 | if (ext4_journal_current_handle()) { |
b38bd33a | 4747 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
ac27a0ec DK |
4748 | dump_stack(); |
4749 | return -EIO; | |
4750 | } | |
4751 | ||
4752 | if (!wait) | |
4753 | return 0; | |
4754 | ||
617ba13b | 4755 | return ext4_force_commit(inode->i_sb); |
ac27a0ec DK |
4756 | } |
4757 | ||
0390131b FM |
4758 | int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh) |
4759 | { | |
4760 | int err = 0; | |
4761 | ||
4762 | mark_buffer_dirty(bh); | |
4763 | if (inode && inode_needs_sync(inode)) { | |
4764 | sync_dirty_buffer(bh); | |
4765 | if (buffer_req(bh) && !buffer_uptodate(bh)) { | |
4766 | ext4_error(inode->i_sb, __func__, | |
4767 | "IO error syncing inode, " | |
4768 | "inode=%lu, block=%llu", | |
4769 | inode->i_ino, | |
4770 | (unsigned long long)bh->b_blocknr); | |
4771 | err = -EIO; | |
4772 | } | |
4773 | } | |
4774 | return err; | |
4775 | } | |
4776 | ||
ac27a0ec | 4777 | /* |
617ba13b | 4778 | * ext4_setattr() |
ac27a0ec DK |
4779 | * |
4780 | * Called from notify_change. | |
4781 | * | |
4782 | * We want to trap VFS attempts to truncate the file as soon as | |
4783 | * possible. In particular, we want to make sure that when the VFS | |
4784 | * shrinks i_size, we put the inode on the orphan list and modify | |
4785 | * i_disksize immediately, so that during the subsequent flushing of | |
4786 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4787 | * commit will leave the blocks being flushed in an unused state on | |
4788 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4789 | * be freed, so we have a strong guarantee that no future commit will | |
4790 | * leave these blocks visible to the user.) | |
4791 | * | |
678aaf48 JK |
4792 | * Another thing we have to assure is that if we are in ordered mode |
4793 | * and inode is still attached to the committing transaction, we must | |
4794 | * we start writeout of all the dirty pages which are being truncated. | |
4795 | * This way we are sure that all the data written in the previous | |
4796 | * transaction are already on disk (truncate waits for pages under | |
4797 | * writeback). | |
4798 | * | |
4799 | * Called with inode->i_mutex down. | |
ac27a0ec | 4800 | */ |
617ba13b | 4801 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4802 | { |
4803 | struct inode *inode = dentry->d_inode; | |
4804 | int error, rc = 0; | |
4805 | const unsigned int ia_valid = attr->ia_valid; | |
4806 | ||
4807 | error = inode_change_ok(inode, attr); | |
4808 | if (error) | |
4809 | return error; | |
4810 | ||
4811 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
4812 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
4813 | handle_t *handle; | |
4814 | ||
4815 | /* (user+group)*(old+new) structure, inode write (sb, | |
4816 | * inode block, ? - but truncate inode update has it) */ | |
617ba13b MC |
4817 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
4818 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
ac27a0ec DK |
4819 | if (IS_ERR(handle)) { |
4820 | error = PTR_ERR(handle); | |
4821 | goto err_out; | |
4822 | } | |
a269eb18 | 4823 | error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; |
ac27a0ec | 4824 | if (error) { |
617ba13b | 4825 | ext4_journal_stop(handle); |
ac27a0ec DK |
4826 | return error; |
4827 | } | |
4828 | /* Update corresponding info in inode so that everything is in | |
4829 | * one transaction */ | |
4830 | if (attr->ia_valid & ATTR_UID) | |
4831 | inode->i_uid = attr->ia_uid; | |
4832 | if (attr->ia_valid & ATTR_GID) | |
4833 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4834 | error = ext4_mark_inode_dirty(handle, inode); |
4835 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4836 | } |
4837 | ||
e2b46574 ES |
4838 | if (attr->ia_valid & ATTR_SIZE) { |
4839 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { | |
4840 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4841 | ||
4842 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { | |
4843 | error = -EFBIG; | |
4844 | goto err_out; | |
4845 | } | |
4846 | } | |
4847 | } | |
4848 | ||
ac27a0ec DK |
4849 | if (S_ISREG(inode->i_mode) && |
4850 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
4851 | handle_t *handle; | |
4852 | ||
617ba13b | 4853 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4854 | if (IS_ERR(handle)) { |
4855 | error = PTR_ERR(handle); | |
4856 | goto err_out; | |
4857 | } | |
4858 | ||
617ba13b MC |
4859 | error = ext4_orphan_add(handle, inode); |
4860 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
4861 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4862 | if (!error) |
4863 | error = rc; | |
617ba13b | 4864 | ext4_journal_stop(handle); |
678aaf48 JK |
4865 | |
4866 | if (ext4_should_order_data(inode)) { | |
4867 | error = ext4_begin_ordered_truncate(inode, | |
4868 | attr->ia_size); | |
4869 | if (error) { | |
4870 | /* Do as much error cleanup as possible */ | |
4871 | handle = ext4_journal_start(inode, 3); | |
4872 | if (IS_ERR(handle)) { | |
4873 | ext4_orphan_del(NULL, inode); | |
4874 | goto err_out; | |
4875 | } | |
4876 | ext4_orphan_del(handle, inode); | |
4877 | ext4_journal_stop(handle); | |
4878 | goto err_out; | |
4879 | } | |
4880 | } | |
ac27a0ec DK |
4881 | } |
4882 | ||
4883 | rc = inode_setattr(inode, attr); | |
4884 | ||
617ba13b | 4885 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
4886 | * transaction handle at all, we need to clean up the in-core |
4887 | * orphan list manually. */ | |
4888 | if (inode->i_nlink) | |
617ba13b | 4889 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4890 | |
4891 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4892 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4893 | |
4894 | err_out: | |
617ba13b | 4895 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4896 | if (!error) |
4897 | error = rc; | |
4898 | return error; | |
4899 | } | |
4900 | ||
3e3398a0 MC |
4901 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4902 | struct kstat *stat) | |
4903 | { | |
4904 | struct inode *inode; | |
4905 | unsigned long delalloc_blocks; | |
4906 | ||
4907 | inode = dentry->d_inode; | |
4908 | generic_fillattr(inode, stat); | |
4909 | ||
4910 | /* | |
4911 | * We can't update i_blocks if the block allocation is delayed | |
4912 | * otherwise in the case of system crash before the real block | |
4913 | * allocation is done, we will have i_blocks inconsistent with | |
4914 | * on-disk file blocks. | |
4915 | * We always keep i_blocks updated together with real | |
4916 | * allocation. But to not confuse with user, stat | |
4917 | * will return the blocks that include the delayed allocation | |
4918 | * blocks for this file. | |
4919 | */ | |
4920 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
4921 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
4922 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
4923 | ||
4924 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4925 | return 0; | |
4926 | } | |
ac27a0ec | 4927 | |
a02908f1 MC |
4928 | static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks, |
4929 | int chunk) | |
4930 | { | |
4931 | int indirects; | |
4932 | ||
4933 | /* if nrblocks are contiguous */ | |
4934 | if (chunk) { | |
4935 | /* | |
4936 | * With N contiguous data blocks, it need at most | |
4937 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks | |
4938 | * 2 dindirect blocks | |
4939 | * 1 tindirect block | |
4940 | */ | |
4941 | indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
4942 | return indirects + 3; | |
4943 | } | |
4944 | /* | |
4945 | * if nrblocks are not contiguous, worse case, each block touch | |
4946 | * a indirect block, and each indirect block touch a double indirect | |
4947 | * block, plus a triple indirect block | |
4948 | */ | |
4949 | indirects = nrblocks * 2 + 1; | |
4950 | return indirects; | |
4951 | } | |
4952 | ||
4953 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
4954 | { | |
4955 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
ac51d837 TT |
4956 | return ext4_indirect_trans_blocks(inode, nrblocks, chunk); |
4957 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); | |
a02908f1 | 4958 | } |
ac51d837 | 4959 | |
ac27a0ec | 4960 | /* |
a02908f1 MC |
4961 | * Account for index blocks, block groups bitmaps and block group |
4962 | * descriptor blocks if modify datablocks and index blocks | |
4963 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4964 | * |
a02908f1 MC |
4965 | * If datablocks are discontiguous, they are possible to spread over |
4966 | * different block groups too. If they are contiugous, with flexbg, | |
4967 | * they could still across block group boundary. | |
ac27a0ec | 4968 | * |
a02908f1 MC |
4969 | * Also account for superblock, inode, quota and xattr blocks |
4970 | */ | |
4971 | int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
4972 | { | |
8df9675f TT |
4973 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4974 | int gdpblocks; | |
a02908f1 MC |
4975 | int idxblocks; |
4976 | int ret = 0; | |
4977 | ||
4978 | /* | |
4979 | * How many index blocks need to touch to modify nrblocks? | |
4980 | * The "Chunk" flag indicating whether the nrblocks is | |
4981 | * physically contiguous on disk | |
4982 | * | |
4983 | * For Direct IO and fallocate, they calls get_block to allocate | |
4984 | * one single extent at a time, so they could set the "Chunk" flag | |
4985 | */ | |
4986 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4987 | ||
4988 | ret = idxblocks; | |
4989 | ||
4990 | /* | |
4991 | * Now let's see how many group bitmaps and group descriptors need | |
4992 | * to account | |
4993 | */ | |
4994 | groups = idxblocks; | |
4995 | if (chunk) | |
4996 | groups += 1; | |
4997 | else | |
4998 | groups += nrblocks; | |
4999 | ||
5000 | gdpblocks = groups; | |
8df9675f TT |
5001 | if (groups > ngroups) |
5002 | groups = ngroups; | |
a02908f1 MC |
5003 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5004 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5005 | ||
5006 | /* bitmaps and block group descriptor blocks */ | |
5007 | ret += groups + gdpblocks; | |
5008 | ||
5009 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5010 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5011 | ||
5012 | return ret; | |
5013 | } | |
5014 | ||
5015 | /* | |
5016 | * Calulate the total number of credits to reserve to fit | |
f3bd1f3f MC |
5017 | * the modification of a single pages into a single transaction, |
5018 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5019 | * |
525f4ed8 | 5020 | * This could be called via ext4_write_begin() |
ac27a0ec | 5021 | * |
525f4ed8 | 5022 | * We need to consider the worse case, when |
a02908f1 | 5023 | * one new block per extent. |
ac27a0ec | 5024 | */ |
a86c6181 | 5025 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5026 | { |
617ba13b | 5027 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5028 | int ret; |
5029 | ||
a02908f1 | 5030 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 5031 | |
a02908f1 | 5032 | /* Account for data blocks for journalled mode */ |
617ba13b | 5033 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5034 | ret += bpp; |
ac27a0ec DK |
5035 | return ret; |
5036 | } | |
f3bd1f3f MC |
5037 | |
5038 | /* | |
5039 | * Calculate the journal credits for a chunk of data modification. | |
5040 | * | |
5041 | * This is called from DIO, fallocate or whoever calling | |
12b7ac17 | 5042 | * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks. |
f3bd1f3f MC |
5043 | * |
5044 | * journal buffers for data blocks are not included here, as DIO | |
5045 | * and fallocate do no need to journal data buffers. | |
5046 | */ | |
5047 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5048 | { | |
5049 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5050 | } | |
5051 | ||
ac27a0ec | 5052 | /* |
617ba13b | 5053 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5054 | * Give this, we know that the caller already has write access to iloc->bh. |
5055 | */ | |
617ba13b MC |
5056 | int ext4_mark_iloc_dirty(handle_t *handle, |
5057 | struct inode *inode, struct ext4_iloc *iloc) | |
ac27a0ec DK |
5058 | { |
5059 | int err = 0; | |
5060 | ||
25ec56b5 JNC |
5061 | if (test_opt(inode->i_sb, I_VERSION)) |
5062 | inode_inc_iversion(inode); | |
5063 | ||
ac27a0ec DK |
5064 | /* the do_update_inode consumes one bh->b_count */ |
5065 | get_bh(iloc->bh); | |
5066 | ||
dab291af | 5067 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
617ba13b | 5068 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5069 | put_bh(iloc->bh); |
5070 | return err; | |
5071 | } | |
5072 | ||
5073 | /* | |
5074 | * On success, We end up with an outstanding reference count against | |
5075 | * iloc->bh. This _must_ be cleaned up later. | |
5076 | */ | |
5077 | ||
5078 | int | |
617ba13b MC |
5079 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5080 | struct ext4_iloc *iloc) | |
ac27a0ec | 5081 | { |
0390131b FM |
5082 | int err; |
5083 | ||
5084 | err = ext4_get_inode_loc(inode, iloc); | |
5085 | if (!err) { | |
5086 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5087 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5088 | if (err) { | |
5089 | brelse(iloc->bh); | |
5090 | iloc->bh = NULL; | |
ac27a0ec DK |
5091 | } |
5092 | } | |
617ba13b | 5093 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5094 | return err; |
5095 | } | |
5096 | ||
6dd4ee7c KS |
5097 | /* |
5098 | * Expand an inode by new_extra_isize bytes. | |
5099 | * Returns 0 on success or negative error number on failure. | |
5100 | */ | |
1d03ec98 AK |
5101 | static int ext4_expand_extra_isize(struct inode *inode, |
5102 | unsigned int new_extra_isize, | |
5103 | struct ext4_iloc iloc, | |
5104 | handle_t *handle) | |
6dd4ee7c KS |
5105 | { |
5106 | struct ext4_inode *raw_inode; | |
5107 | struct ext4_xattr_ibody_header *header; | |
5108 | struct ext4_xattr_entry *entry; | |
5109 | ||
5110 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5111 | return 0; | |
5112 | ||
5113 | raw_inode = ext4_raw_inode(&iloc); | |
5114 | ||
5115 | header = IHDR(inode, raw_inode); | |
5116 | entry = IFIRST(header); | |
5117 | ||
5118 | /* No extended attributes present */ | |
5119 | if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) || | |
5120 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5121 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, | |
5122 | new_extra_isize); | |
5123 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5124 | return 0; | |
5125 | } | |
5126 | ||
5127 | /* try to expand with EAs present */ | |
5128 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5129 | raw_inode, handle); | |
5130 | } | |
5131 | ||
ac27a0ec DK |
5132 | /* |
5133 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5134 | * dirtiness (it may still be data-dirty). | |
5135 | * This means that the in-core inode may be reaped by prune_icache | |
5136 | * without having to perform any I/O. This is a very good thing, | |
5137 | * because *any* task may call prune_icache - even ones which | |
5138 | * have a transaction open against a different journal. | |
5139 | * | |
5140 | * Is this cheating? Not really. Sure, we haven't written the | |
5141 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5142 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5143 | * we start and wait on commits. | |
5144 | * | |
5145 | * Is this efficient/effective? Well, we're being nice to the system | |
5146 | * by cleaning up our inodes proactively so they can be reaped | |
5147 | * without I/O. But we are potentially leaving up to five seconds' | |
5148 | * worth of inodes floating about which prune_icache wants us to | |
5149 | * write out. One way to fix that would be to get prune_icache() | |
5150 | * to do a write_super() to free up some memory. It has the desired | |
5151 | * effect. | |
5152 | */ | |
617ba13b | 5153 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5154 | { |
617ba13b | 5155 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5156 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5157 | static unsigned int mnt_count; | |
5158 | int err, ret; | |
ac27a0ec DK |
5159 | |
5160 | might_sleep(); | |
617ba13b | 5161 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5162 | if (ext4_handle_valid(handle) && |
5163 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
6dd4ee7c KS |
5164 | !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { |
5165 | /* | |
5166 | * We need extra buffer credits since we may write into EA block | |
5167 | * with this same handle. If journal_extend fails, then it will | |
5168 | * only result in a minor loss of functionality for that inode. | |
5169 | * If this is felt to be critical, then e2fsck should be run to | |
5170 | * force a large enough s_min_extra_isize. | |
5171 | */ | |
5172 | if ((jbd2_journal_extend(handle, | |
5173 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5174 | ret = ext4_expand_extra_isize(inode, | |
5175 | sbi->s_want_extra_isize, | |
5176 | iloc, handle); | |
5177 | if (ret) { | |
5178 | EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND; | |
c1bddad9 AK |
5179 | if (mnt_count != |
5180 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
46e665e9 | 5181 | ext4_warning(inode->i_sb, __func__, |
6dd4ee7c KS |
5182 | "Unable to expand inode %lu. Delete" |
5183 | " some EAs or run e2fsck.", | |
5184 | inode->i_ino); | |
c1bddad9 AK |
5185 | mnt_count = |
5186 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5187 | } |
5188 | } | |
5189 | } | |
5190 | } | |
ac27a0ec | 5191 | if (!err) |
617ba13b | 5192 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5193 | return err; |
5194 | } | |
5195 | ||
5196 | /* | |
617ba13b | 5197 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5198 | * |
5199 | * We're really interested in the case where a file is being extended. | |
5200 | * i_size has been changed by generic_commit_write() and we thus need | |
5201 | * to include the updated inode in the current transaction. | |
5202 | * | |
a269eb18 | 5203 | * Also, vfs_dq_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5204 | * are allocated to the file. |
5205 | * | |
5206 | * If the inode is marked synchronous, we don't honour that here - doing | |
5207 | * so would cause a commit on atime updates, which we don't bother doing. | |
5208 | * We handle synchronous inodes at the highest possible level. | |
5209 | */ | |
617ba13b | 5210 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 5211 | { |
617ba13b | 5212 | handle_t *current_handle = ext4_journal_current_handle(); |
ac27a0ec DK |
5213 | handle_t *handle; |
5214 | ||
0390131b FM |
5215 | if (!ext4_handle_valid(current_handle)) { |
5216 | ext4_mark_inode_dirty(current_handle, inode); | |
5217 | return; | |
5218 | } | |
5219 | ||
617ba13b | 5220 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
5221 | if (IS_ERR(handle)) |
5222 | goto out; | |
5223 | if (current_handle && | |
5224 | current_handle->h_transaction != handle->h_transaction) { | |
5225 | /* This task has a transaction open against a different fs */ | |
5226 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
46e665e9 | 5227 | __func__); |
ac27a0ec DK |
5228 | } else { |
5229 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
5230 | current_handle); | |
617ba13b | 5231 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 5232 | } |
617ba13b | 5233 | ext4_journal_stop(handle); |
ac27a0ec DK |
5234 | out: |
5235 | return; | |
5236 | } | |
5237 | ||
5238 | #if 0 | |
5239 | /* | |
5240 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5241 | * it from being flushed to disk early. Unlike | |
617ba13b | 5242 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5243 | * returns no iloc structure, so the caller needs to repeat the iloc |
5244 | * lookup to mark the inode dirty later. | |
5245 | */ | |
617ba13b | 5246 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5247 | { |
617ba13b | 5248 | struct ext4_iloc iloc; |
ac27a0ec DK |
5249 | |
5250 | int err = 0; | |
5251 | if (handle) { | |
617ba13b | 5252 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5253 | if (!err) { |
5254 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5255 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5256 | if (!err) |
0390131b FM |
5257 | err = ext4_handle_dirty_metadata(handle, |
5258 | inode, | |
5259 | iloc.bh); | |
ac27a0ec DK |
5260 | brelse(iloc.bh); |
5261 | } | |
5262 | } | |
617ba13b | 5263 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5264 | return err; |
5265 | } | |
5266 | #endif | |
5267 | ||
617ba13b | 5268 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5269 | { |
5270 | journal_t *journal; | |
5271 | handle_t *handle; | |
5272 | int err; | |
5273 | ||
5274 | /* | |
5275 | * We have to be very careful here: changing a data block's | |
5276 | * journaling status dynamically is dangerous. If we write a | |
5277 | * data block to the journal, change the status and then delete | |
5278 | * that block, we risk forgetting to revoke the old log record | |
5279 | * from the journal and so a subsequent replay can corrupt data. | |
5280 | * So, first we make sure that the journal is empty and that | |
5281 | * nobody is changing anything. | |
5282 | */ | |
5283 | ||
617ba13b | 5284 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5285 | if (!journal) |
5286 | return 0; | |
d699594d | 5287 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5288 | return -EROFS; |
5289 | ||
dab291af MC |
5290 | jbd2_journal_lock_updates(journal); |
5291 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
5292 | |
5293 | /* | |
5294 | * OK, there are no updates running now, and all cached data is | |
5295 | * synced to disk. We are now in a completely consistent state | |
5296 | * which doesn't have anything in the journal, and we know that | |
5297 | * no filesystem updates are running, so it is safe to modify | |
5298 | * the inode's in-core data-journaling state flag now. | |
5299 | */ | |
5300 | ||
5301 | if (val) | |
617ba13b | 5302 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
ac27a0ec | 5303 | else |
617ba13b MC |
5304 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
5305 | ext4_set_aops(inode); | |
ac27a0ec | 5306 | |
dab291af | 5307 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
5308 | |
5309 | /* Finally we can mark the inode as dirty. */ | |
5310 | ||
617ba13b | 5311 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
5312 | if (IS_ERR(handle)) |
5313 | return PTR_ERR(handle); | |
5314 | ||
617ba13b | 5315 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5316 | ext4_handle_sync(handle); |
617ba13b MC |
5317 | ext4_journal_stop(handle); |
5318 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5319 | |
5320 | return err; | |
5321 | } | |
2e9ee850 AK |
5322 | |
5323 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5324 | { | |
5325 | return !buffer_mapped(bh); | |
5326 | } | |
5327 | ||
c2ec175c | 5328 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5329 | { |
c2ec175c | 5330 | struct page *page = vmf->page; |
2e9ee850 AK |
5331 | loff_t size; |
5332 | unsigned long len; | |
5333 | int ret = -EINVAL; | |
79f0be8d | 5334 | void *fsdata; |
2e9ee850 AK |
5335 | struct file *file = vma->vm_file; |
5336 | struct inode *inode = file->f_path.dentry->d_inode; | |
5337 | struct address_space *mapping = inode->i_mapping; | |
5338 | ||
5339 | /* | |
5340 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
5341 | * get i_mutex because we are already holding mmap_sem. | |
5342 | */ | |
5343 | down_read(&inode->i_alloc_sem); | |
5344 | size = i_size_read(inode); | |
5345 | if (page->mapping != mapping || size <= page_offset(page) | |
5346 | || !PageUptodate(page)) { | |
5347 | /* page got truncated from under us? */ | |
5348 | goto out_unlock; | |
5349 | } | |
5350 | ret = 0; | |
5351 | if (PageMappedToDisk(page)) | |
5352 | goto out_unlock; | |
5353 | ||
5354 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5355 | len = size & ~PAGE_CACHE_MASK; | |
5356 | else | |
5357 | len = PAGE_CACHE_SIZE; | |
5358 | ||
5359 | if (page_has_buffers(page)) { | |
5360 | /* return if we have all the buffers mapped */ | |
5361 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
5362 | ext4_bh_unmapped)) | |
5363 | goto out_unlock; | |
5364 | } | |
5365 | /* | |
5366 | * OK, we need to fill the hole... Do write_begin write_end | |
5367 | * to do block allocation/reservation.We are not holding | |
5368 | * inode.i__mutex here. That allow * parallel write_begin, | |
5369 | * write_end call. lock_page prevent this from happening | |
5370 | * on the same page though | |
5371 | */ | |
5372 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
79f0be8d | 5373 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); |
2e9ee850 AK |
5374 | if (ret < 0) |
5375 | goto out_unlock; | |
5376 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
79f0be8d | 5377 | len, len, page, fsdata); |
2e9ee850 AK |
5378 | if (ret < 0) |
5379 | goto out_unlock; | |
5380 | ret = 0; | |
5381 | out_unlock: | |
c2ec175c NP |
5382 | if (ret) |
5383 | ret = VM_FAULT_SIGBUS; | |
2e9ee850 AK |
5384 | up_read(&inode->i_alloc_sem); |
5385 | return ret; | |
5386 | } |