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 MC |
28 | #include <linux/ext4_jbd2.h> |
29 | #include <linux/jbd2.h> | |
ac27a0ec DK |
30 | #include <linux/smp_lock.h> |
31 | #include <linux/highuid.h> | |
32 | #include <linux/pagemap.h> | |
33 | #include <linux/quotaops.h> | |
34 | #include <linux/string.h> | |
35 | #include <linux/buffer_head.h> | |
36 | #include <linux/writeback.h> | |
37 | #include <linux/mpage.h> | |
38 | #include <linux/uio.h> | |
39 | #include <linux/bio.h> | |
40 | #include "xattr.h" | |
41 | #include "acl.h" | |
42 | ||
ac27a0ec DK |
43 | /* |
44 | * Test whether an inode is a fast symlink. | |
45 | */ | |
617ba13b | 46 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 47 | { |
617ba13b | 48 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
49 | (inode->i_sb->s_blocksize >> 9) : 0; |
50 | ||
51 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
52 | } | |
53 | ||
54 | /* | |
617ba13b | 55 | * The ext4 forget function must perform a revoke if we are freeing data |
ac27a0ec DK |
56 | * which has been journaled. Metadata (eg. indirect blocks) must be |
57 | * revoked in all cases. | |
58 | * | |
59 | * "bh" may be NULL: a metadata block may have been freed from memory | |
60 | * but there may still be a record of it in the journal, and that record | |
61 | * still needs to be revoked. | |
62 | */ | |
617ba13b MC |
63 | int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, |
64 | struct buffer_head *bh, ext4_fsblk_t blocknr) | |
ac27a0ec DK |
65 | { |
66 | int err; | |
67 | ||
68 | might_sleep(); | |
69 | ||
70 | BUFFER_TRACE(bh, "enter"); | |
71 | ||
72 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
73 | "data mode %lx\n", | |
74 | bh, is_metadata, inode->i_mode, | |
75 | test_opt(inode->i_sb, DATA_FLAGS)); | |
76 | ||
77 | /* Never use the revoke function if we are doing full data | |
78 | * journaling: there is no need to, and a V1 superblock won't | |
79 | * support it. Otherwise, only skip the revoke on un-journaled | |
80 | * data blocks. */ | |
81 | ||
617ba13b MC |
82 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
83 | (!is_metadata && !ext4_should_journal_data(inode))) { | |
ac27a0ec | 84 | if (bh) { |
dab291af | 85 | BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
617ba13b | 86 | return ext4_journal_forget(handle, bh); |
ac27a0ec DK |
87 | } |
88 | return 0; | |
89 | } | |
90 | ||
91 | /* | |
92 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
93 | */ | |
617ba13b MC |
94 | BUFFER_TRACE(bh, "call ext4_journal_revoke"); |
95 | err = ext4_journal_revoke(handle, blocknr, bh); | |
ac27a0ec | 96 | if (err) |
617ba13b | 97 | ext4_abort(inode->i_sb, __FUNCTION__, |
ac27a0ec DK |
98 | "error %d when attempting revoke", err); |
99 | BUFFER_TRACE(bh, "exit"); | |
100 | return err; | |
101 | } | |
102 | ||
103 | /* | |
104 | * Work out how many blocks we need to proceed with the next chunk of a | |
105 | * truncate transaction. | |
106 | */ | |
107 | static unsigned long blocks_for_truncate(struct inode *inode) | |
108 | { | |
109 | unsigned long needed; | |
110 | ||
111 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
112 | ||
113 | /* Give ourselves just enough room to cope with inodes in which | |
114 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
115 | * which resulted in random data in an inode which looked enough | |
617ba13b | 116 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
117 | * will go a bit crazy if that happens, but at least we should |
118 | * try not to panic the whole kernel. */ | |
119 | if (needed < 2) | |
120 | needed = 2; | |
121 | ||
122 | /* But we need to bound the transaction so we don't overflow the | |
123 | * journal. */ | |
617ba13b MC |
124 | if (needed > EXT4_MAX_TRANS_DATA) |
125 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 126 | |
617ba13b | 127 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
128 | } |
129 | ||
130 | /* | |
131 | * Truncate transactions can be complex and absolutely huge. So we need to | |
132 | * be able to restart the transaction at a conventient checkpoint to make | |
133 | * sure we don't overflow the journal. | |
134 | * | |
135 | * start_transaction gets us a new handle for a truncate transaction, | |
136 | * and extend_transaction tries to extend the existing one a bit. If | |
137 | * extend fails, we need to propagate the failure up and restart the | |
138 | * transaction in the top-level truncate loop. --sct | |
139 | */ | |
140 | static handle_t *start_transaction(struct inode *inode) | |
141 | { | |
142 | handle_t *result; | |
143 | ||
617ba13b | 144 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
145 | if (!IS_ERR(result)) |
146 | return result; | |
147 | ||
617ba13b | 148 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
149 | return result; |
150 | } | |
151 | ||
152 | /* | |
153 | * Try to extend this transaction for the purposes of truncation. | |
154 | * | |
155 | * Returns 0 if we managed to create more room. If we can't create more | |
156 | * room, and the transaction must be restarted we return 1. | |
157 | */ | |
158 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
159 | { | |
617ba13b | 160 | if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS) |
ac27a0ec | 161 | return 0; |
617ba13b | 162 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
163 | return 0; |
164 | return 1; | |
165 | } | |
166 | ||
167 | /* | |
168 | * Restart the transaction associated with *handle. This does a commit, | |
169 | * so before we call here everything must be consistently dirtied against | |
170 | * this transaction. | |
171 | */ | |
617ba13b | 172 | static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
173 | { |
174 | jbd_debug(2, "restarting handle %p\n", handle); | |
617ba13b | 175 | return ext4_journal_restart(handle, blocks_for_truncate(inode)); |
ac27a0ec DK |
176 | } |
177 | ||
178 | /* | |
179 | * Called at the last iput() if i_nlink is zero. | |
180 | */ | |
617ba13b | 181 | void ext4_delete_inode (struct inode * inode) |
ac27a0ec DK |
182 | { |
183 | handle_t *handle; | |
184 | ||
185 | truncate_inode_pages(&inode->i_data, 0); | |
186 | ||
187 | if (is_bad_inode(inode)) | |
188 | goto no_delete; | |
189 | ||
190 | handle = start_transaction(inode); | |
191 | if (IS_ERR(handle)) { | |
192 | /* | |
193 | * If we're going to skip the normal cleanup, we still need to | |
194 | * make sure that the in-core orphan linked list is properly | |
195 | * cleaned up. | |
196 | */ | |
617ba13b | 197 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
198 | goto no_delete; |
199 | } | |
200 | ||
201 | if (IS_SYNC(inode)) | |
202 | handle->h_sync = 1; | |
203 | inode->i_size = 0; | |
204 | if (inode->i_blocks) | |
617ba13b | 205 | ext4_truncate(inode); |
ac27a0ec | 206 | /* |
617ba13b | 207 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 208 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 209 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 210 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 211 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
212 | * (Well, we could do this if we need to, but heck - it works) |
213 | */ | |
617ba13b MC |
214 | ext4_orphan_del(handle, inode); |
215 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
216 | |
217 | /* | |
218 | * One subtle ordering requirement: if anything has gone wrong | |
219 | * (transaction abort, IO errors, whatever), then we can still | |
220 | * do these next steps (the fs will already have been marked as | |
221 | * having errors), but we can't free the inode if the mark_dirty | |
222 | * fails. | |
223 | */ | |
617ba13b | 224 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
225 | /* If that failed, just do the required in-core inode clear. */ |
226 | clear_inode(inode); | |
227 | else | |
617ba13b MC |
228 | ext4_free_inode(handle, inode); |
229 | ext4_journal_stop(handle); | |
ac27a0ec DK |
230 | return; |
231 | no_delete: | |
232 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
233 | } | |
234 | ||
235 | typedef struct { | |
236 | __le32 *p; | |
237 | __le32 key; | |
238 | struct buffer_head *bh; | |
239 | } Indirect; | |
240 | ||
241 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
242 | { | |
243 | p->key = *(p->p = v); | |
244 | p->bh = bh; | |
245 | } | |
246 | ||
247 | static int verify_chain(Indirect *from, Indirect *to) | |
248 | { | |
249 | while (from <= to && from->key == *from->p) | |
250 | from++; | |
251 | return (from > to); | |
252 | } | |
253 | ||
254 | /** | |
617ba13b | 255 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
256 | * @inode: inode in question (we are only interested in its superblock) |
257 | * @i_block: block number to be parsed | |
258 | * @offsets: array to store the offsets in | |
259 | * @boundary: set this non-zero if the referred-to block is likely to be | |
260 | * followed (on disk) by an indirect block. | |
261 | * | |
617ba13b | 262 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
263 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
264 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
265 | * This function translates the block number into path in that tree - | |
266 | * return value is the path length and @offsets[n] is the offset of | |
267 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
268 | * (negative or too large) warning is printed and zero returned. | |
269 | * | |
270 | * Note: function doesn't find node addresses, so no IO is needed. All | |
271 | * we need to know is the capacity of indirect blocks (taken from the | |
272 | * inode->i_sb). | |
273 | */ | |
274 | ||
275 | /* | |
276 | * Portability note: the last comparison (check that we fit into triple | |
277 | * indirect block) is spelled differently, because otherwise on an | |
278 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
279 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
280 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
281 | * i_block would have to be negative in the very beginning, so we would not | |
282 | * get there at all. | |
283 | */ | |
284 | ||
617ba13b | 285 | static int ext4_block_to_path(struct inode *inode, |
ac27a0ec DK |
286 | long i_block, int offsets[4], int *boundary) |
287 | { | |
617ba13b MC |
288 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
289 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
290 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
291 | indirect_blocks = ptrs, |
292 | double_blocks = (1 << (ptrs_bits * 2)); | |
293 | int n = 0; | |
294 | int final = 0; | |
295 | ||
296 | if (i_block < 0) { | |
617ba13b | 297 | ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0"); |
ac27a0ec DK |
298 | } else if (i_block < direct_blocks) { |
299 | offsets[n++] = i_block; | |
300 | final = direct_blocks; | |
301 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { | |
617ba13b | 302 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
303 | offsets[n++] = i_block; |
304 | final = ptrs; | |
305 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 306 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
307 | offsets[n++] = i_block >> ptrs_bits; |
308 | offsets[n++] = i_block & (ptrs - 1); | |
309 | final = ptrs; | |
310 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 311 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
312 | offsets[n++] = i_block >> (ptrs_bits * 2); |
313 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
314 | offsets[n++] = i_block & (ptrs - 1); | |
315 | final = ptrs; | |
316 | } else { | |
617ba13b | 317 | ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big"); |
ac27a0ec DK |
318 | } |
319 | if (boundary) | |
320 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
321 | return n; | |
322 | } | |
323 | ||
324 | /** | |
617ba13b | 325 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
326 | * @inode: inode in question |
327 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
328 | * @offsets: offsets of pointers in inode/indirect blocks | |
329 | * @chain: place to store the result | |
330 | * @err: here we store the error value | |
331 | * | |
332 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
333 | * if everything went OK or the pointer to the last filled triple | |
334 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
335 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
336 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
337 | * number (it points into struct inode for i==0 and into the bh->b_data | |
338 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
339 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
340 | * numbers of the chain, addresses they were taken from (and where we can | |
341 | * verify that chain did not change) and buffer_heads hosting these | |
342 | * numbers. | |
343 | * | |
344 | * Function stops when it stumbles upon zero pointer (absent block) | |
345 | * (pointer to last triple returned, *@err == 0) | |
346 | * or when it gets an IO error reading an indirect block | |
347 | * (ditto, *@err == -EIO) | |
348 | * or when it notices that chain had been changed while it was reading | |
349 | * (ditto, *@err == -EAGAIN) | |
350 | * or when it reads all @depth-1 indirect blocks successfully and finds | |
351 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
352 | */ | |
617ba13b | 353 | static Indirect *ext4_get_branch(struct inode *inode, int depth, int *offsets, |
ac27a0ec DK |
354 | Indirect chain[4], int *err) |
355 | { | |
356 | struct super_block *sb = inode->i_sb; | |
357 | Indirect *p = chain; | |
358 | struct buffer_head *bh; | |
359 | ||
360 | *err = 0; | |
361 | /* i_data is not going away, no lock needed */ | |
617ba13b | 362 | add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
363 | if (!p->key) |
364 | goto no_block; | |
365 | while (--depth) { | |
366 | bh = sb_bread(sb, le32_to_cpu(p->key)); | |
367 | if (!bh) | |
368 | goto failure; | |
369 | /* Reader: pointers */ | |
370 | if (!verify_chain(chain, p)) | |
371 | goto changed; | |
372 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); | |
373 | /* Reader: end */ | |
374 | if (!p->key) | |
375 | goto no_block; | |
376 | } | |
377 | return NULL; | |
378 | ||
379 | changed: | |
380 | brelse(bh); | |
381 | *err = -EAGAIN; | |
382 | goto no_block; | |
383 | failure: | |
384 | *err = -EIO; | |
385 | no_block: | |
386 | return p; | |
387 | } | |
388 | ||
389 | /** | |
617ba13b | 390 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
391 | * @inode: owner |
392 | * @ind: descriptor of indirect block. | |
393 | * | |
394 | * This function returns the prefered place for block allocation. | |
395 | * It is used when heuristic for sequential allocation fails. | |
396 | * Rules are: | |
397 | * + if there is a block to the left of our position - allocate near it. | |
398 | * + if pointer will live in indirect block - allocate near that block. | |
399 | * + if pointer will live in inode - allocate in the same | |
400 | * cylinder group. | |
401 | * | |
402 | * In the latter case we colour the starting block by the callers PID to | |
403 | * prevent it from clashing with concurrent allocations for a different inode | |
404 | * in the same block group. The PID is used here so that functionally related | |
405 | * files will be close-by on-disk. | |
406 | * | |
407 | * Caller must make sure that @ind is valid and will stay that way. | |
408 | */ | |
617ba13b | 409 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 410 | { |
617ba13b | 411 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec DK |
412 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; |
413 | __le32 *p; | |
617ba13b MC |
414 | ext4_fsblk_t bg_start; |
415 | ext4_grpblk_t colour; | |
ac27a0ec DK |
416 | |
417 | /* Try to find previous block */ | |
418 | for (p = ind->p - 1; p >= start; p--) { | |
419 | if (*p) | |
420 | return le32_to_cpu(*p); | |
421 | } | |
422 | ||
423 | /* No such thing, so let's try location of indirect block */ | |
424 | if (ind->bh) | |
425 | return ind->bh->b_blocknr; | |
426 | ||
427 | /* | |
428 | * It is going to be referred to from the inode itself? OK, just put it | |
429 | * into the same cylinder group then. | |
430 | */ | |
617ba13b | 431 | bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group); |
ac27a0ec | 432 | colour = (current->pid % 16) * |
617ba13b | 433 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
ac27a0ec DK |
434 | return bg_start + colour; |
435 | } | |
436 | ||
437 | /** | |
617ba13b | 438 | * ext4_find_goal - find a prefered place for allocation. |
ac27a0ec DK |
439 | * @inode: owner |
440 | * @block: block we want | |
441 | * @chain: chain of indirect blocks | |
442 | * @partial: pointer to the last triple within a chain | |
443 | * @goal: place to store the result. | |
444 | * | |
445 | * Normally this function find the prefered place for block allocation, | |
446 | * stores it in *@goal and returns zero. | |
447 | */ | |
448 | ||
617ba13b | 449 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, long block, |
ac27a0ec DK |
450 | Indirect chain[4], Indirect *partial) |
451 | { | |
617ba13b | 452 | struct ext4_block_alloc_info *block_i; |
ac27a0ec | 453 | |
617ba13b | 454 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
455 | |
456 | /* | |
457 | * try the heuristic for sequential allocation, | |
458 | * failing that at least try to get decent locality. | |
459 | */ | |
460 | if (block_i && (block == block_i->last_alloc_logical_block + 1) | |
461 | && (block_i->last_alloc_physical_block != 0)) { | |
462 | return block_i->last_alloc_physical_block + 1; | |
463 | } | |
464 | ||
617ba13b | 465 | return ext4_find_near(inode, partial); |
ac27a0ec DK |
466 | } |
467 | ||
468 | /** | |
617ba13b | 469 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
470 | * of direct blocks need to be allocated for the given branch. |
471 | * | |
472 | * @branch: chain of indirect blocks | |
473 | * @k: number of blocks need for indirect blocks | |
474 | * @blks: number of data blocks to be mapped. | |
475 | * @blocks_to_boundary: the offset in the indirect block | |
476 | * | |
477 | * return the total number of blocks to be allocate, including the | |
478 | * direct and indirect blocks. | |
479 | */ | |
617ba13b | 480 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
ac27a0ec DK |
481 | int blocks_to_boundary) |
482 | { | |
483 | unsigned long count = 0; | |
484 | ||
485 | /* | |
486 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
487 | * then it's clear blocks on that path have not allocated | |
488 | */ | |
489 | if (k > 0) { | |
490 | /* right now we don't handle cross boundary allocation */ | |
491 | if (blks < blocks_to_boundary + 1) | |
492 | count += blks; | |
493 | else | |
494 | count += blocks_to_boundary + 1; | |
495 | return count; | |
496 | } | |
497 | ||
498 | count++; | |
499 | while (count < blks && count <= blocks_to_boundary && | |
500 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
501 | count++; | |
502 | } | |
503 | return count; | |
504 | } | |
505 | ||
506 | /** | |
617ba13b | 507 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
508 | * @indirect_blks: the number of blocks need to allocate for indirect |
509 | * blocks | |
510 | * | |
511 | * @new_blocks: on return it will store the new block numbers for | |
512 | * the indirect blocks(if needed) and the first direct block, | |
513 | * @blks: on return it will store the total number of allocated | |
514 | * direct blocks | |
515 | */ | |
617ba13b MC |
516 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
517 | ext4_fsblk_t goal, int indirect_blks, int blks, | |
518 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec DK |
519 | { |
520 | int target, i; | |
521 | unsigned long count = 0; | |
522 | int index = 0; | |
617ba13b | 523 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
524 | int ret = 0; |
525 | ||
526 | /* | |
527 | * Here we try to allocate the requested multiple blocks at once, | |
528 | * on a best-effort basis. | |
529 | * To build a branch, we should allocate blocks for | |
530 | * the indirect blocks(if not allocated yet), and at least | |
531 | * the first direct block of this branch. That's the | |
532 | * minimum number of blocks need to allocate(required) | |
533 | */ | |
534 | target = blks + indirect_blks; | |
535 | ||
536 | while (1) { | |
537 | count = target; | |
538 | /* allocating blocks for indirect blocks and direct blocks */ | |
617ba13b | 539 | current_block = ext4_new_blocks(handle,inode,goal,&count,err); |
ac27a0ec DK |
540 | if (*err) |
541 | goto failed_out; | |
542 | ||
543 | target -= count; | |
544 | /* allocate blocks for indirect blocks */ | |
545 | while (index < indirect_blks && count) { | |
546 | new_blocks[index++] = current_block++; | |
547 | count--; | |
548 | } | |
549 | ||
550 | if (count > 0) | |
551 | break; | |
552 | } | |
553 | ||
554 | /* save the new block number for the first direct block */ | |
555 | new_blocks[index] = current_block; | |
556 | ||
557 | /* total number of blocks allocated for direct blocks */ | |
558 | ret = count; | |
559 | *err = 0; | |
560 | return ret; | |
561 | failed_out: | |
562 | for (i = 0; i <index; i++) | |
617ba13b | 563 | ext4_free_blocks(handle, inode, new_blocks[i], 1); |
ac27a0ec DK |
564 | return ret; |
565 | } | |
566 | ||
567 | /** | |
617ba13b | 568 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
569 | * @inode: owner |
570 | * @indirect_blks: number of allocated indirect blocks | |
571 | * @blks: number of allocated direct blocks | |
572 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
573 | * @branch: place to store the chain in. | |
574 | * | |
575 | * This function allocates blocks, zeroes out all but the last one, | |
576 | * links them into chain and (if we are synchronous) writes them to disk. | |
577 | * In other words, it prepares a branch that can be spliced onto the | |
578 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 579 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
580 | * we had read the existing part of chain and partial points to the last |
581 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 582 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
583 | * place chain is disconnected - *branch->p is still zero (we did not |
584 | * set the last link), but branch->key contains the number that should | |
585 | * be placed into *branch->p to fill that gap. | |
586 | * | |
587 | * If allocation fails we free all blocks we've allocated (and forget | |
588 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 589 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
590 | * as described above and return 0. |
591 | */ | |
617ba13b MC |
592 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
593 | int indirect_blks, int *blks, ext4_fsblk_t goal, | |
ac27a0ec DK |
594 | int *offsets, Indirect *branch) |
595 | { | |
596 | int blocksize = inode->i_sb->s_blocksize; | |
597 | int i, n = 0; | |
598 | int err = 0; | |
599 | struct buffer_head *bh; | |
600 | int num; | |
617ba13b MC |
601 | ext4_fsblk_t new_blocks[4]; |
602 | ext4_fsblk_t current_block; | |
ac27a0ec | 603 | |
617ba13b | 604 | num = ext4_alloc_blocks(handle, inode, goal, indirect_blks, |
ac27a0ec DK |
605 | *blks, new_blocks, &err); |
606 | if (err) | |
607 | return err; | |
608 | ||
609 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
610 | /* | |
611 | * metadata blocks and data blocks are allocated. | |
612 | */ | |
613 | for (n = 1; n <= indirect_blks; n++) { | |
614 | /* | |
615 | * Get buffer_head for parent block, zero it out | |
616 | * and set the pointer to new one, then send | |
617 | * parent to disk. | |
618 | */ | |
619 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
620 | branch[n].bh = bh; | |
621 | lock_buffer(bh); | |
622 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 623 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
624 | if (err) { |
625 | unlock_buffer(bh); | |
626 | brelse(bh); | |
627 | goto failed; | |
628 | } | |
629 | ||
630 | memset(bh->b_data, 0, blocksize); | |
631 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
632 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
633 | *branch[n].p = branch[n].key; | |
634 | if ( n == indirect_blks) { | |
635 | current_block = new_blocks[n]; | |
636 | /* | |
637 | * End of chain, update the last new metablock of | |
638 | * the chain to point to the new allocated | |
639 | * data blocks numbers | |
640 | */ | |
641 | for (i=1; i < num; i++) | |
642 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
643 | } | |
644 | BUFFER_TRACE(bh, "marking uptodate"); | |
645 | set_buffer_uptodate(bh); | |
646 | unlock_buffer(bh); | |
647 | ||
617ba13b MC |
648 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
649 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
650 | if (err) |
651 | goto failed; | |
652 | } | |
653 | *blks = num; | |
654 | return err; | |
655 | failed: | |
656 | /* Allocation failed, free what we already allocated */ | |
657 | for (i = 1; i <= n ; i++) { | |
dab291af | 658 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
617ba13b | 659 | ext4_journal_forget(handle, branch[i].bh); |
ac27a0ec DK |
660 | } |
661 | for (i = 0; i <indirect_blks; i++) | |
617ba13b | 662 | ext4_free_blocks(handle, inode, new_blocks[i], 1); |
ac27a0ec | 663 | |
617ba13b | 664 | ext4_free_blocks(handle, inode, new_blocks[i], num); |
ac27a0ec DK |
665 | |
666 | return err; | |
667 | } | |
668 | ||
669 | /** | |
617ba13b | 670 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
671 | * @inode: owner |
672 | * @block: (logical) number of block we are adding | |
673 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 674 | * ext4_alloc_branch) |
ac27a0ec DK |
675 | * @where: location of missing link |
676 | * @num: number of indirect blocks we are adding | |
677 | * @blks: number of direct blocks we are adding | |
678 | * | |
679 | * This function fills the missing link and does all housekeeping needed in | |
680 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
681 | * chain to new block and return 0. | |
682 | */ | |
617ba13b | 683 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
684 | long block, Indirect *where, int num, int blks) |
685 | { | |
686 | int i; | |
687 | int err = 0; | |
617ba13b MC |
688 | struct ext4_block_alloc_info *block_i; |
689 | ext4_fsblk_t current_block; | |
ac27a0ec | 690 | |
617ba13b | 691 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
692 | /* |
693 | * If we're splicing into a [td]indirect block (as opposed to the | |
694 | * inode) then we need to get write access to the [td]indirect block | |
695 | * before the splice. | |
696 | */ | |
697 | if (where->bh) { | |
698 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 699 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
700 | if (err) |
701 | goto err_out; | |
702 | } | |
703 | /* That's it */ | |
704 | ||
705 | *where->p = where->key; | |
706 | ||
707 | /* | |
708 | * Update the host buffer_head or inode to point to more just allocated | |
709 | * direct blocks blocks | |
710 | */ | |
711 | if (num == 0 && blks > 1) { | |
712 | current_block = le32_to_cpu(where->key) + 1; | |
713 | for (i = 1; i < blks; i++) | |
714 | *(where->p + i ) = cpu_to_le32(current_block++); | |
715 | } | |
716 | ||
717 | /* | |
718 | * update the most recently allocated logical & physical block | |
719 | * in i_block_alloc_info, to assist find the proper goal block for next | |
720 | * allocation | |
721 | */ | |
722 | if (block_i) { | |
723 | block_i->last_alloc_logical_block = block + blks - 1; | |
724 | block_i->last_alloc_physical_block = | |
725 | le32_to_cpu(where[num].key) + blks - 1; | |
726 | } | |
727 | ||
728 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
729 | ||
730 | inode->i_ctime = CURRENT_TIME_SEC; | |
617ba13b | 731 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
732 | |
733 | /* had we spliced it onto indirect block? */ | |
734 | if (where->bh) { | |
735 | /* | |
736 | * If we spliced it onto an indirect block, we haven't | |
737 | * altered the inode. Note however that if it is being spliced | |
738 | * onto an indirect block at the very end of the file (the | |
739 | * file is growing) then we *will* alter the inode to reflect | |
740 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 741 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
742 | */ |
743 | jbd_debug(5, "splicing indirect only\n"); | |
617ba13b MC |
744 | BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata"); |
745 | err = ext4_journal_dirty_metadata(handle, where->bh); | |
ac27a0ec DK |
746 | if (err) |
747 | goto err_out; | |
748 | } else { | |
749 | /* | |
750 | * OK, we spliced it into the inode itself on a direct block. | |
751 | * Inode was dirtied above. | |
752 | */ | |
753 | jbd_debug(5, "splicing direct\n"); | |
754 | } | |
755 | return err; | |
756 | ||
757 | err_out: | |
758 | for (i = 1; i <= num; i++) { | |
dab291af | 759 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
617ba13b MC |
760 | ext4_journal_forget(handle, where[i].bh); |
761 | ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); | |
ac27a0ec | 762 | } |
617ba13b | 763 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
ac27a0ec DK |
764 | |
765 | return err; | |
766 | } | |
767 | ||
768 | /* | |
769 | * Allocation strategy is simple: if we have to allocate something, we will | |
770 | * have to go the whole way to leaf. So let's do it before attaching anything | |
771 | * to tree, set linkage between the newborn blocks, write them if sync is | |
772 | * required, recheck the path, free and repeat if check fails, otherwise | |
773 | * set the last missing link (that will protect us from any truncate-generated | |
774 | * removals - all blocks on the path are immune now) and possibly force the | |
775 | * write on the parent block. | |
776 | * That has a nice additional property: no special recovery from the failed | |
777 | * allocations is needed - we simply release blocks and do not touch anything | |
778 | * reachable from inode. | |
779 | * | |
780 | * `handle' can be NULL if create == 0. | |
781 | * | |
782 | * The BKL may not be held on entry here. Be sure to take it early. | |
783 | * return > 0, # of blocks mapped or allocated. | |
784 | * return = 0, if plain lookup failed. | |
785 | * return < 0, error case. | |
786 | */ | |
617ba13b | 787 | int ext4_get_blocks_handle(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
788 | sector_t iblock, unsigned long maxblocks, |
789 | struct buffer_head *bh_result, | |
790 | int create, int extend_disksize) | |
791 | { | |
792 | int err = -EIO; | |
793 | int offsets[4]; | |
794 | Indirect chain[4]; | |
795 | Indirect *partial; | |
617ba13b | 796 | ext4_fsblk_t goal; |
ac27a0ec DK |
797 | int indirect_blks; |
798 | int blocks_to_boundary = 0; | |
799 | int depth; | |
617ba13b | 800 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 801 | int count = 0; |
617ba13b | 802 | ext4_fsblk_t first_block = 0; |
ac27a0ec DK |
803 | |
804 | ||
a86c6181 | 805 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
ac27a0ec | 806 | J_ASSERT(handle != NULL || create == 0); |
617ba13b | 807 | depth = ext4_block_to_path(inode,iblock,offsets,&blocks_to_boundary); |
ac27a0ec DK |
808 | |
809 | if (depth == 0) | |
810 | goto out; | |
811 | ||
617ba13b | 812 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
813 | |
814 | /* Simplest case - block found, no allocation needed */ | |
815 | if (!partial) { | |
816 | first_block = le32_to_cpu(chain[depth - 1].key); | |
817 | clear_buffer_new(bh_result); | |
818 | count++; | |
819 | /*map more blocks*/ | |
820 | while (count < maxblocks && count <= blocks_to_boundary) { | |
617ba13b | 821 | ext4_fsblk_t blk; |
ac27a0ec DK |
822 | |
823 | if (!verify_chain(chain, partial)) { | |
824 | /* | |
825 | * Indirect block might be removed by | |
826 | * truncate while we were reading it. | |
827 | * Handling of that case: forget what we've | |
828 | * got now. Flag the err as EAGAIN, so it | |
829 | * will reread. | |
830 | */ | |
831 | err = -EAGAIN; | |
832 | count = 0; | |
833 | break; | |
834 | } | |
835 | blk = le32_to_cpu(*(chain[depth-1].p + count)); | |
836 | ||
837 | if (blk == first_block + count) | |
838 | count++; | |
839 | else | |
840 | break; | |
841 | } | |
842 | if (err != -EAGAIN) | |
843 | goto got_it; | |
844 | } | |
845 | ||
846 | /* Next simple case - plain lookup or failed read of indirect block */ | |
847 | if (!create || err == -EIO) | |
848 | goto cleanup; | |
849 | ||
850 | mutex_lock(&ei->truncate_mutex); | |
851 | ||
852 | /* | |
853 | * If the indirect block is missing while we are reading | |
617ba13b | 854 | * the chain(ext4_get_branch() returns -EAGAIN err), or |
ac27a0ec DK |
855 | * if the chain has been changed after we grab the semaphore, |
856 | * (either because another process truncated this branch, or | |
857 | * another get_block allocated this branch) re-grab the chain to see if | |
858 | * the request block has been allocated or not. | |
859 | * | |
860 | * Since we already block the truncate/other get_block | |
861 | * at this point, we will have the current copy of the chain when we | |
862 | * splice the branch into the tree. | |
863 | */ | |
864 | if (err == -EAGAIN || !verify_chain(chain, partial)) { | |
865 | while (partial > chain) { | |
866 | brelse(partial->bh); | |
867 | partial--; | |
868 | } | |
617ba13b | 869 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
870 | if (!partial) { |
871 | count++; | |
872 | mutex_unlock(&ei->truncate_mutex); | |
873 | if (err) | |
874 | goto cleanup; | |
875 | clear_buffer_new(bh_result); | |
876 | goto got_it; | |
877 | } | |
878 | } | |
879 | ||
880 | /* | |
881 | * Okay, we need to do block allocation. Lazily initialize the block | |
882 | * allocation info here if necessary | |
883 | */ | |
884 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) | |
617ba13b | 885 | ext4_init_block_alloc_info(inode); |
ac27a0ec | 886 | |
617ba13b | 887 | goal = ext4_find_goal(inode, iblock, chain, partial); |
ac27a0ec DK |
888 | |
889 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
890 | indirect_blks = (chain + depth) - partial - 1; | |
891 | ||
892 | /* | |
893 | * Next look up the indirect map to count the totoal number of | |
894 | * direct blocks to allocate for this branch. | |
895 | */ | |
617ba13b | 896 | count = ext4_blks_to_allocate(partial, indirect_blks, |
ac27a0ec DK |
897 | maxblocks, blocks_to_boundary); |
898 | /* | |
617ba13b | 899 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 900 | */ |
617ba13b | 901 | err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal, |
ac27a0ec DK |
902 | offsets + (partial - chain), partial); |
903 | ||
904 | /* | |
617ba13b | 905 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
906 | * on the new chain if there is a failure, but that risks using |
907 | * up transaction credits, especially for bitmaps where the | |
908 | * credits cannot be returned. Can we handle this somehow? We | |
909 | * may need to return -EAGAIN upwards in the worst case. --sct | |
910 | */ | |
911 | if (!err) | |
617ba13b | 912 | err = ext4_splice_branch(handle, inode, iblock, |
ac27a0ec DK |
913 | partial, indirect_blks, count); |
914 | /* | |
915 | * i_disksize growing is protected by truncate_mutex. Don't forget to | |
916 | * protect it if you're about to implement concurrent | |
617ba13b | 917 | * ext4_get_block() -bzzz |
ac27a0ec DK |
918 | */ |
919 | if (!err && extend_disksize && inode->i_size > ei->i_disksize) | |
920 | ei->i_disksize = inode->i_size; | |
921 | mutex_unlock(&ei->truncate_mutex); | |
922 | if (err) | |
923 | goto cleanup; | |
924 | ||
925 | set_buffer_new(bh_result); | |
926 | got_it: | |
927 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
928 | if (count > blocks_to_boundary) | |
929 | set_buffer_boundary(bh_result); | |
930 | err = count; | |
931 | /* Clean up and exit */ | |
932 | partial = chain + depth - 1; /* the whole chain */ | |
933 | cleanup: | |
934 | while (partial > chain) { | |
935 | BUFFER_TRACE(partial->bh, "call brelse"); | |
936 | brelse(partial->bh); | |
937 | partial--; | |
938 | } | |
939 | BUFFER_TRACE(bh_result, "returned"); | |
940 | out: | |
941 | return err; | |
942 | } | |
943 | ||
617ba13b | 944 | #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32) |
ac27a0ec | 945 | |
617ba13b | 946 | static int ext4_get_block(struct inode *inode, sector_t iblock, |
ac27a0ec DK |
947 | struct buffer_head *bh_result, int create) |
948 | { | |
949 | handle_t *handle = journal_current_handle(); | |
950 | int ret = 0; | |
951 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
952 | ||
953 | if (!create) | |
954 | goto get_block; /* A read */ | |
955 | ||
956 | if (max_blocks == 1) | |
957 | goto get_block; /* A single block get */ | |
958 | ||
959 | if (handle->h_transaction->t_state == T_LOCKED) { | |
960 | /* | |
961 | * Huge direct-io writes can hold off commits for long | |
962 | * periods of time. Let this commit run. | |
963 | */ | |
617ba13b MC |
964 | ext4_journal_stop(handle); |
965 | handle = ext4_journal_start(inode, DIO_CREDITS); | |
ac27a0ec DK |
966 | if (IS_ERR(handle)) |
967 | ret = PTR_ERR(handle); | |
968 | goto get_block; | |
969 | } | |
970 | ||
617ba13b | 971 | if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) { |
ac27a0ec DK |
972 | /* |
973 | * Getting low on buffer credits... | |
974 | */ | |
617ba13b | 975 | ret = ext4_journal_extend(handle, DIO_CREDITS); |
ac27a0ec DK |
976 | if (ret > 0) { |
977 | /* | |
978 | * Couldn't extend the transaction. Start a new one. | |
979 | */ | |
617ba13b | 980 | ret = ext4_journal_restart(handle, DIO_CREDITS); |
ac27a0ec DK |
981 | } |
982 | } | |
983 | ||
984 | get_block: | |
985 | if (ret == 0) { | |
a86c6181 | 986 | ret = ext4_get_blocks_wrap(handle, inode, iblock, |
ac27a0ec DK |
987 | max_blocks, bh_result, create, 0); |
988 | if (ret > 0) { | |
989 | bh_result->b_size = (ret << inode->i_blkbits); | |
990 | ret = 0; | |
991 | } | |
992 | } | |
993 | return ret; | |
994 | } | |
995 | ||
996 | /* | |
997 | * `handle' can be NULL if create is zero | |
998 | */ | |
617ba13b | 999 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
1000 | long block, int create, int *errp) |
1001 | { | |
1002 | struct buffer_head dummy; | |
1003 | int fatal = 0, err; | |
1004 | ||
1005 | J_ASSERT(handle != NULL || create == 0); | |
1006 | ||
1007 | dummy.b_state = 0; | |
1008 | dummy.b_blocknr = -1000; | |
1009 | buffer_trace_init(&dummy.b_history); | |
a86c6181 | 1010 | err = ext4_get_blocks_wrap(handle, inode, block, 1, |
ac27a0ec DK |
1011 | &dummy, create, 1); |
1012 | /* | |
617ba13b | 1013 | * ext4_get_blocks_handle() returns number of blocks |
ac27a0ec DK |
1014 | * mapped. 0 in case of a HOLE. |
1015 | */ | |
1016 | if (err > 0) { | |
1017 | if (err > 1) | |
1018 | WARN_ON(1); | |
1019 | err = 0; | |
1020 | } | |
1021 | *errp = err; | |
1022 | if (!err && buffer_mapped(&dummy)) { | |
1023 | struct buffer_head *bh; | |
1024 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
1025 | if (!bh) { | |
1026 | *errp = -EIO; | |
1027 | goto err; | |
1028 | } | |
1029 | if (buffer_new(&dummy)) { | |
1030 | J_ASSERT(create != 0); | |
1031 | J_ASSERT(handle != 0); | |
1032 | ||
1033 | /* | |
1034 | * Now that we do not always journal data, we should | |
1035 | * keep in mind whether this should always journal the | |
1036 | * new buffer as metadata. For now, regular file | |
617ba13b | 1037 | * writes use ext4_get_block instead, so it's not a |
ac27a0ec DK |
1038 | * problem. |
1039 | */ | |
1040 | lock_buffer(bh); | |
1041 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 1042 | fatal = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
1043 | if (!fatal && !buffer_uptodate(bh)) { |
1044 | memset(bh->b_data,0,inode->i_sb->s_blocksize); | |
1045 | set_buffer_uptodate(bh); | |
1046 | } | |
1047 | unlock_buffer(bh); | |
617ba13b MC |
1048 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
1049 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
1050 | if (!fatal) |
1051 | fatal = err; | |
1052 | } else { | |
1053 | BUFFER_TRACE(bh, "not a new buffer"); | |
1054 | } | |
1055 | if (fatal) { | |
1056 | *errp = fatal; | |
1057 | brelse(bh); | |
1058 | bh = NULL; | |
1059 | } | |
1060 | return bh; | |
1061 | } | |
1062 | err: | |
1063 | return NULL; | |
1064 | } | |
1065 | ||
617ba13b | 1066 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
1067 | int block, int create, int *err) |
1068 | { | |
1069 | struct buffer_head * bh; | |
1070 | ||
617ba13b | 1071 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1072 | if (!bh) |
1073 | return bh; | |
1074 | if (buffer_uptodate(bh)) | |
1075 | return bh; | |
1076 | ll_rw_block(READ_META, 1, &bh); | |
1077 | wait_on_buffer(bh); | |
1078 | if (buffer_uptodate(bh)) | |
1079 | return bh; | |
1080 | put_bh(bh); | |
1081 | *err = -EIO; | |
1082 | return NULL; | |
1083 | } | |
1084 | ||
1085 | static int walk_page_buffers( handle_t *handle, | |
1086 | struct buffer_head *head, | |
1087 | unsigned from, | |
1088 | unsigned to, | |
1089 | int *partial, | |
1090 | int (*fn)( handle_t *handle, | |
1091 | struct buffer_head *bh)) | |
1092 | { | |
1093 | struct buffer_head *bh; | |
1094 | unsigned block_start, block_end; | |
1095 | unsigned blocksize = head->b_size; | |
1096 | int err, ret = 0; | |
1097 | struct buffer_head *next; | |
1098 | ||
1099 | for ( bh = head, block_start = 0; | |
1100 | ret == 0 && (bh != head || !block_start); | |
1101 | block_start = block_end, bh = next) | |
1102 | { | |
1103 | next = bh->b_this_page; | |
1104 | block_end = block_start + blocksize; | |
1105 | if (block_end <= from || block_start >= to) { | |
1106 | if (partial && !buffer_uptodate(bh)) | |
1107 | *partial = 1; | |
1108 | continue; | |
1109 | } | |
1110 | err = (*fn)(handle, bh); | |
1111 | if (!ret) | |
1112 | ret = err; | |
1113 | } | |
1114 | return ret; | |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * To preserve ordering, it is essential that the hole instantiation and | |
1119 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1120 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1121 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1122 | * prepare_write() is the right place. |
1123 | * | |
617ba13b MC |
1124 | * Also, this function can nest inside ext4_writepage() -> |
1125 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1126 | * has generated enough buffer credits to do the whole page. So we won't |
1127 | * block on the journal in that case, which is good, because the caller may | |
1128 | * be PF_MEMALLOC. | |
1129 | * | |
617ba13b | 1130 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1131 | * quota file writes. If we were to commit the transaction while thus |
1132 | * reentered, there can be a deadlock - we would be holding a quota | |
1133 | * lock, and the commit would never complete if another thread had a | |
1134 | * transaction open and was blocking on the quota lock - a ranking | |
1135 | * violation. | |
1136 | * | |
dab291af | 1137 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1138 | * will _not_ run commit under these circumstances because handle->h_ref |
1139 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1140 | * write. | |
1141 | */ | |
1142 | static int do_journal_get_write_access(handle_t *handle, | |
1143 | struct buffer_head *bh) | |
1144 | { | |
1145 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1146 | return 0; | |
617ba13b | 1147 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1148 | } |
1149 | ||
b46be050 AS |
1150 | /* |
1151 | * The idea of this helper function is following: | |
1152 | * if prepare_write has allocated some blocks, but not all of them, the | |
1153 | * transaction must include the content of the newly allocated blocks. | |
1154 | * This content is expected to be set to zeroes by block_prepare_write(). | |
1155 | * 2006/10/14 SAW | |
1156 | */ | |
1157 | static int ext4_prepare_failure(struct file *file, struct page *page, | |
1158 | unsigned from, unsigned to) | |
1159 | { | |
1160 | struct address_space *mapping; | |
1161 | struct buffer_head *bh, *head, *next; | |
1162 | unsigned block_start, block_end; | |
1163 | unsigned blocksize; | |
1164 | int ret; | |
1165 | handle_t *handle = ext4_journal_current_handle(); | |
1166 | ||
1167 | mapping = page->mapping; | |
1168 | if (ext4_should_writeback_data(mapping->host)) { | |
1169 | /* optimization: no constraints about data */ | |
1170 | skip: | |
1171 | return ext4_journal_stop(handle); | |
1172 | } | |
1173 | ||
1174 | head = page_buffers(page); | |
1175 | blocksize = head->b_size; | |
1176 | for ( bh = head, block_start = 0; | |
1177 | bh != head || !block_start; | |
1178 | block_start = block_end, bh = next) | |
1179 | { | |
1180 | next = bh->b_this_page; | |
1181 | block_end = block_start + blocksize; | |
1182 | if (block_end <= from) | |
1183 | continue; | |
1184 | if (block_start >= to) { | |
1185 | block_start = to; | |
1186 | break; | |
1187 | } | |
1188 | if (!buffer_mapped(bh)) | |
1189 | /* prepare_write failed on this bh */ | |
1190 | break; | |
1191 | if (ext4_should_journal_data(mapping->host)) { | |
1192 | ret = do_journal_get_write_access(handle, bh); | |
1193 | if (ret) { | |
1194 | ext4_journal_stop(handle); | |
1195 | return ret; | |
1196 | } | |
1197 | } | |
1198 | /* | |
1199 | * block_start here becomes the first block where the current iteration | |
1200 | * of prepare_write failed. | |
1201 | */ | |
1202 | } | |
1203 | if (block_start <= from) | |
1204 | goto skip; | |
1205 | ||
1206 | /* commit allocated and zeroed buffers */ | |
1207 | return mapping->a_ops->commit_write(file, page, from, block_start); | |
1208 | } | |
1209 | ||
617ba13b | 1210 | static int ext4_prepare_write(struct file *file, struct page *page, |
ac27a0ec DK |
1211 | unsigned from, unsigned to) |
1212 | { | |
1213 | struct inode *inode = page->mapping->host; | |
b46be050 AS |
1214 | int ret, ret2; |
1215 | int needed_blocks = ext4_writepage_trans_blocks(inode); | |
ac27a0ec DK |
1216 | handle_t *handle; |
1217 | int retries = 0; | |
1218 | ||
1219 | retry: | |
617ba13b | 1220 | handle = ext4_journal_start(inode, needed_blocks); |
b46be050 AS |
1221 | if (IS_ERR(handle)) |
1222 | return PTR_ERR(handle); | |
617ba13b MC |
1223 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) |
1224 | ret = nobh_prepare_write(page, from, to, ext4_get_block); | |
ac27a0ec | 1225 | else |
617ba13b | 1226 | ret = block_prepare_write(page, from, to, ext4_get_block); |
ac27a0ec | 1227 | if (ret) |
b46be050 | 1228 | goto failure; |
ac27a0ec | 1229 | |
617ba13b | 1230 | if (ext4_should_journal_data(inode)) { |
ac27a0ec DK |
1231 | ret = walk_page_buffers(handle, page_buffers(page), |
1232 | from, to, NULL, do_journal_get_write_access); | |
b46be050 AS |
1233 | if (ret) |
1234 | /* fatal error, just put the handle and return */ | |
1235 | journal_stop(handle); | |
ac27a0ec | 1236 | } |
b46be050 AS |
1237 | return ret; |
1238 | ||
1239 | failure: | |
1240 | ret2 = ext4_prepare_failure(file, page, from, to); | |
1241 | if (ret2 < 0) | |
1242 | return ret2; | |
617ba13b | 1243 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1244 | goto retry; |
b46be050 | 1245 | /* retry number exceeded, or other error like -EDQUOT */ |
ac27a0ec DK |
1246 | return ret; |
1247 | } | |
1248 | ||
617ba13b | 1249 | int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
ac27a0ec | 1250 | { |
dab291af | 1251 | int err = jbd2_journal_dirty_data(handle, bh); |
ac27a0ec | 1252 | if (err) |
617ba13b | 1253 | ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__, |
ac27a0ec DK |
1254 | bh, handle,err); |
1255 | return err; | |
1256 | } | |
1257 | ||
1258 | /* For commit_write() in data=journal mode */ | |
1259 | static int commit_write_fn(handle_t *handle, struct buffer_head *bh) | |
1260 | { | |
1261 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1262 | return 0; | |
1263 | set_buffer_uptodate(bh); | |
617ba13b | 1264 | return ext4_journal_dirty_metadata(handle, bh); |
ac27a0ec DK |
1265 | } |
1266 | ||
1267 | /* | |
1268 | * We need to pick up the new inode size which generic_commit_write gave us | |
1269 | * `file' can be NULL - eg, when called from page_symlink(). | |
1270 | * | |
617ba13b | 1271 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1272 | * buffers are managed internally. |
1273 | */ | |
617ba13b | 1274 | static int ext4_ordered_commit_write(struct file *file, struct page *page, |
ac27a0ec DK |
1275 | unsigned from, unsigned to) |
1276 | { | |
617ba13b | 1277 | handle_t *handle = ext4_journal_current_handle(); |
ac27a0ec DK |
1278 | struct inode *inode = page->mapping->host; |
1279 | int ret = 0, ret2; | |
1280 | ||
1281 | ret = walk_page_buffers(handle, page_buffers(page), | |
617ba13b | 1282 | from, to, NULL, ext4_journal_dirty_data); |
ac27a0ec DK |
1283 | |
1284 | if (ret == 0) { | |
1285 | /* | |
1286 | * generic_commit_write() will run mark_inode_dirty() if i_size | |
1287 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
1288 | * into that. | |
1289 | */ | |
1290 | loff_t new_i_size; | |
1291 | ||
1292 | new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
617ba13b MC |
1293 | if (new_i_size > EXT4_I(inode)->i_disksize) |
1294 | EXT4_I(inode)->i_disksize = new_i_size; | |
ac27a0ec DK |
1295 | ret = generic_commit_write(file, page, from, to); |
1296 | } | |
617ba13b | 1297 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1298 | if (!ret) |
1299 | ret = ret2; | |
1300 | return ret; | |
1301 | } | |
1302 | ||
617ba13b | 1303 | static int ext4_writeback_commit_write(struct file *file, struct page *page, |
ac27a0ec DK |
1304 | unsigned from, unsigned to) |
1305 | { | |
617ba13b | 1306 | handle_t *handle = ext4_journal_current_handle(); |
ac27a0ec DK |
1307 | struct inode *inode = page->mapping->host; |
1308 | int ret = 0, ret2; | |
1309 | loff_t new_i_size; | |
1310 | ||
1311 | new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
617ba13b MC |
1312 | if (new_i_size > EXT4_I(inode)->i_disksize) |
1313 | EXT4_I(inode)->i_disksize = new_i_size; | |
ac27a0ec | 1314 | |
617ba13b | 1315 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) |
ac27a0ec DK |
1316 | ret = nobh_commit_write(file, page, from, to); |
1317 | else | |
1318 | ret = generic_commit_write(file, page, from, to); | |
1319 | ||
617ba13b | 1320 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1321 | if (!ret) |
1322 | ret = ret2; | |
1323 | return ret; | |
1324 | } | |
1325 | ||
617ba13b | 1326 | static int ext4_journalled_commit_write(struct file *file, |
ac27a0ec DK |
1327 | struct page *page, unsigned from, unsigned to) |
1328 | { | |
617ba13b | 1329 | handle_t *handle = ext4_journal_current_handle(); |
ac27a0ec DK |
1330 | struct inode *inode = page->mapping->host; |
1331 | int ret = 0, ret2; | |
1332 | int partial = 0; | |
1333 | loff_t pos; | |
1334 | ||
1335 | /* | |
1336 | * Here we duplicate the generic_commit_write() functionality | |
1337 | */ | |
1338 | pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; | |
1339 | ||
1340 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
1341 | to, &partial, commit_write_fn); | |
1342 | if (!partial) | |
1343 | SetPageUptodate(page); | |
1344 | if (pos > inode->i_size) | |
1345 | i_size_write(inode, pos); | |
617ba13b MC |
1346 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
1347 | if (inode->i_size > EXT4_I(inode)->i_disksize) { | |
1348 | EXT4_I(inode)->i_disksize = inode->i_size; | |
1349 | ret2 = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
1350 | if (!ret) |
1351 | ret = ret2; | |
1352 | } | |
617ba13b | 1353 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1354 | if (!ret) |
1355 | ret = ret2; | |
1356 | return ret; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * bmap() is special. It gets used by applications such as lilo and by | |
1361 | * the swapper to find the on-disk block of a specific piece of data. | |
1362 | * | |
1363 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 1364 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
1365 | * filesystem and enables swap, then they may get a nasty shock when the |
1366 | * data getting swapped to that swapfile suddenly gets overwritten by | |
1367 | * the original zero's written out previously to the journal and | |
1368 | * awaiting writeback in the kernel's buffer cache. | |
1369 | * | |
1370 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
1371 | * take extra steps to flush any blocks which might be in the cache. | |
1372 | */ | |
617ba13b | 1373 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
1374 | { |
1375 | struct inode *inode = mapping->host; | |
1376 | journal_t *journal; | |
1377 | int err; | |
1378 | ||
617ba13b | 1379 | if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
ac27a0ec DK |
1380 | /* |
1381 | * This is a REALLY heavyweight approach, but the use of | |
1382 | * bmap on dirty files is expected to be extremely rare: | |
1383 | * only if we run lilo or swapon on a freshly made file | |
1384 | * do we expect this to happen. | |
1385 | * | |
1386 | * (bmap requires CAP_SYS_RAWIO so this does not | |
1387 | * represent an unprivileged user DOS attack --- we'd be | |
1388 | * in trouble if mortal users could trigger this path at | |
1389 | * will.) | |
1390 | * | |
617ba13b | 1391 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
1392 | * regular files. If somebody wants to bmap a directory |
1393 | * or symlink and gets confused because the buffer | |
1394 | * hasn't yet been flushed to disk, they deserve | |
1395 | * everything they get. | |
1396 | */ | |
1397 | ||
617ba13b MC |
1398 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
1399 | journal = EXT4_JOURNAL(inode); | |
dab291af MC |
1400 | jbd2_journal_lock_updates(journal); |
1401 | err = jbd2_journal_flush(journal); | |
1402 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
1403 | |
1404 | if (err) | |
1405 | return 0; | |
1406 | } | |
1407 | ||
617ba13b | 1408 | return generic_block_bmap(mapping,block,ext4_get_block); |
ac27a0ec DK |
1409 | } |
1410 | ||
1411 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
1412 | { | |
1413 | get_bh(bh); | |
1414 | return 0; | |
1415 | } | |
1416 | ||
1417 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1418 | { | |
1419 | put_bh(bh); | |
1420 | return 0; | |
1421 | } | |
1422 | ||
dab291af | 1423 | static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) |
ac27a0ec DK |
1424 | { |
1425 | if (buffer_mapped(bh)) | |
617ba13b | 1426 | return ext4_journal_dirty_data(handle, bh); |
ac27a0ec DK |
1427 | return 0; |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * Note that we always start a transaction even if we're not journalling | |
1432 | * data. This is to preserve ordering: any hole instantiation within | |
617ba13b | 1433 | * __block_write_full_page -> ext4_get_block() should be journalled |
ac27a0ec DK |
1434 | * along with the data so we don't crash and then get metadata which |
1435 | * refers to old data. | |
1436 | * | |
1437 | * In all journalling modes block_write_full_page() will start the I/O. | |
1438 | * | |
1439 | * Problem: | |
1440 | * | |
617ba13b MC |
1441 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
1442 | * ext4_writepage() | |
ac27a0ec DK |
1443 | * |
1444 | * Similar for: | |
1445 | * | |
617ba13b | 1446 | * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
ac27a0ec | 1447 | * |
617ba13b | 1448 | * Same applies to ext4_get_block(). We will deadlock on various things like |
ac27a0ec DK |
1449 | * lock_journal and i_truncate_mutex. |
1450 | * | |
1451 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
1452 | * allocations fail. | |
1453 | * | |
1454 | * 16May01: If we're reentered then journal_current_handle() will be | |
1455 | * non-zero. We simply *return*. | |
1456 | * | |
1457 | * 1 July 2001: @@@ FIXME: | |
1458 | * In journalled data mode, a data buffer may be metadata against the | |
1459 | * current transaction. But the same file is part of a shared mapping | |
1460 | * and someone does a writepage() on it. | |
1461 | * | |
1462 | * We will move the buffer onto the async_data list, but *after* it has | |
1463 | * been dirtied. So there's a small window where we have dirty data on | |
1464 | * BJ_Metadata. | |
1465 | * | |
1466 | * Note that this only applies to the last partial page in the file. The | |
1467 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
1468 | * broken code anyway: it's wrong for msync()). | |
1469 | * | |
1470 | * It's a rare case: affects the final partial page, for journalled data | |
1471 | * where the file is subject to bith write() and writepage() in the same | |
1472 | * transction. To fix it we'll need a custom block_write_full_page(). | |
1473 | * We'll probably need that anyway for journalling writepage() output. | |
1474 | * | |
1475 | * We don't honour synchronous mounts for writepage(). That would be | |
1476 | * disastrous. Any write() or metadata operation will sync the fs for | |
1477 | * us. | |
1478 | * | |
1479 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, | |
1480 | * we don't need to open a transaction here. | |
1481 | */ | |
617ba13b | 1482 | static int ext4_ordered_writepage(struct page *page, |
ac27a0ec DK |
1483 | struct writeback_control *wbc) |
1484 | { | |
1485 | struct inode *inode = page->mapping->host; | |
1486 | struct buffer_head *page_bufs; | |
1487 | handle_t *handle = NULL; | |
1488 | int ret = 0; | |
1489 | int err; | |
1490 | ||
1491 | J_ASSERT(PageLocked(page)); | |
1492 | ||
1493 | /* | |
1494 | * We give up here if we're reentered, because it might be for a | |
1495 | * different filesystem. | |
1496 | */ | |
617ba13b | 1497 | if (ext4_journal_current_handle()) |
ac27a0ec DK |
1498 | goto out_fail; |
1499 | ||
617ba13b | 1500 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
1501 | |
1502 | if (IS_ERR(handle)) { | |
1503 | ret = PTR_ERR(handle); | |
1504 | goto out_fail; | |
1505 | } | |
1506 | ||
1507 | if (!page_has_buffers(page)) { | |
1508 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
1509 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
1510 | } | |
1511 | page_bufs = page_buffers(page); | |
1512 | walk_page_buffers(handle, page_bufs, 0, | |
1513 | PAGE_CACHE_SIZE, NULL, bget_one); | |
1514 | ||
617ba13b | 1515 | ret = block_write_full_page(page, ext4_get_block, wbc); |
ac27a0ec DK |
1516 | |
1517 | /* | |
1518 | * The page can become unlocked at any point now, and | |
1519 | * truncate can then come in and change things. So we | |
1520 | * can't touch *page from now on. But *page_bufs is | |
1521 | * safe due to elevated refcount. | |
1522 | */ | |
1523 | ||
1524 | /* | |
1525 | * And attach them to the current transaction. But only if | |
1526 | * block_write_full_page() succeeded. Otherwise they are unmapped, | |
1527 | * and generally junk. | |
1528 | */ | |
1529 | if (ret == 0) { | |
1530 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, | |
dab291af | 1531 | NULL, jbd2_journal_dirty_data_fn); |
ac27a0ec DK |
1532 | if (!ret) |
1533 | ret = err; | |
1534 | } | |
1535 | walk_page_buffers(handle, page_bufs, 0, | |
1536 | PAGE_CACHE_SIZE, NULL, bput_one); | |
617ba13b | 1537 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1538 | if (!ret) |
1539 | ret = err; | |
1540 | return ret; | |
1541 | ||
1542 | out_fail: | |
1543 | redirty_page_for_writepage(wbc, page); | |
1544 | unlock_page(page); | |
1545 | return ret; | |
1546 | } | |
1547 | ||
617ba13b | 1548 | static int ext4_writeback_writepage(struct page *page, |
ac27a0ec DK |
1549 | struct writeback_control *wbc) |
1550 | { | |
1551 | struct inode *inode = page->mapping->host; | |
1552 | handle_t *handle = NULL; | |
1553 | int ret = 0; | |
1554 | int err; | |
1555 | ||
617ba13b | 1556 | if (ext4_journal_current_handle()) |
ac27a0ec DK |
1557 | goto out_fail; |
1558 | ||
617ba13b | 1559 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
1560 | if (IS_ERR(handle)) { |
1561 | ret = PTR_ERR(handle); | |
1562 | goto out_fail; | |
1563 | } | |
1564 | ||
617ba13b MC |
1565 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) |
1566 | ret = nobh_writepage(page, ext4_get_block, wbc); | |
ac27a0ec | 1567 | else |
617ba13b | 1568 | ret = block_write_full_page(page, ext4_get_block, wbc); |
ac27a0ec | 1569 | |
617ba13b | 1570 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1571 | if (!ret) |
1572 | ret = err; | |
1573 | return ret; | |
1574 | ||
1575 | out_fail: | |
1576 | redirty_page_for_writepage(wbc, page); | |
1577 | unlock_page(page); | |
1578 | return ret; | |
1579 | } | |
1580 | ||
617ba13b | 1581 | static int ext4_journalled_writepage(struct page *page, |
ac27a0ec DK |
1582 | struct writeback_control *wbc) |
1583 | { | |
1584 | struct inode *inode = page->mapping->host; | |
1585 | handle_t *handle = NULL; | |
1586 | int ret = 0; | |
1587 | int err; | |
1588 | ||
617ba13b | 1589 | if (ext4_journal_current_handle()) |
ac27a0ec DK |
1590 | goto no_write; |
1591 | ||
617ba13b | 1592 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
1593 | if (IS_ERR(handle)) { |
1594 | ret = PTR_ERR(handle); | |
1595 | goto no_write; | |
1596 | } | |
1597 | ||
1598 | if (!page_has_buffers(page) || PageChecked(page)) { | |
1599 | /* | |
1600 | * It's mmapped pagecache. Add buffers and journal it. There | |
1601 | * doesn't seem much point in redirtying the page here. | |
1602 | */ | |
1603 | ClearPageChecked(page); | |
1604 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
617ba13b | 1605 | ext4_get_block); |
ac27a0ec | 1606 | if (ret != 0) { |
617ba13b | 1607 | ext4_journal_stop(handle); |
ac27a0ec DK |
1608 | goto out_unlock; |
1609 | } | |
1610 | ret = walk_page_buffers(handle, page_buffers(page), 0, | |
1611 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
1612 | ||
1613 | err = walk_page_buffers(handle, page_buffers(page), 0, | |
1614 | PAGE_CACHE_SIZE, NULL, commit_write_fn); | |
1615 | if (ret == 0) | |
1616 | ret = err; | |
617ba13b | 1617 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
ac27a0ec DK |
1618 | unlock_page(page); |
1619 | } else { | |
1620 | /* | |
1621 | * It may be a page full of checkpoint-mode buffers. We don't | |
1622 | * really know unless we go poke around in the buffer_heads. | |
1623 | * But block_write_full_page will do the right thing. | |
1624 | */ | |
617ba13b | 1625 | ret = block_write_full_page(page, ext4_get_block, wbc); |
ac27a0ec | 1626 | } |
617ba13b | 1627 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1628 | if (!ret) |
1629 | ret = err; | |
1630 | out: | |
1631 | return ret; | |
1632 | ||
1633 | no_write: | |
1634 | redirty_page_for_writepage(wbc, page); | |
1635 | out_unlock: | |
1636 | unlock_page(page); | |
1637 | goto out; | |
1638 | } | |
1639 | ||
617ba13b | 1640 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 1641 | { |
617ba13b | 1642 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
1643 | } |
1644 | ||
1645 | static int | |
617ba13b | 1646 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
1647 | struct list_head *pages, unsigned nr_pages) |
1648 | { | |
617ba13b | 1649 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
1650 | } |
1651 | ||
617ba13b | 1652 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 1653 | { |
617ba13b | 1654 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
1655 | |
1656 | /* | |
1657 | * If it's a full truncate we just forget about the pending dirtying | |
1658 | */ | |
1659 | if (offset == 0) | |
1660 | ClearPageChecked(page); | |
1661 | ||
dab291af | 1662 | jbd2_journal_invalidatepage(journal, page, offset); |
ac27a0ec DK |
1663 | } |
1664 | ||
617ba13b | 1665 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 1666 | { |
617ba13b | 1667 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
1668 | |
1669 | WARN_ON(PageChecked(page)); | |
1670 | if (!page_has_buffers(page)) | |
1671 | return 0; | |
dab291af | 1672 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
ac27a0ec DK |
1673 | } |
1674 | ||
1675 | /* | |
1676 | * If the O_DIRECT write will extend the file then add this inode to the | |
1677 | * orphan list. So recovery will truncate it back to the original size | |
1678 | * if the machine crashes during the write. | |
1679 | * | |
1680 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
1681 | * crashes then stale disk data _may_ be exposed inside the file. | |
1682 | */ | |
617ba13b | 1683 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
ac27a0ec DK |
1684 | const struct iovec *iov, loff_t offset, |
1685 | unsigned long nr_segs) | |
1686 | { | |
1687 | struct file *file = iocb->ki_filp; | |
1688 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 1689 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec DK |
1690 | handle_t *handle = NULL; |
1691 | ssize_t ret; | |
1692 | int orphan = 0; | |
1693 | size_t count = iov_length(iov, nr_segs); | |
1694 | ||
1695 | if (rw == WRITE) { | |
1696 | loff_t final_size = offset + count; | |
1697 | ||
617ba13b | 1698 | handle = ext4_journal_start(inode, DIO_CREDITS); |
ac27a0ec DK |
1699 | if (IS_ERR(handle)) { |
1700 | ret = PTR_ERR(handle); | |
1701 | goto out; | |
1702 | } | |
1703 | if (final_size > inode->i_size) { | |
617ba13b | 1704 | ret = ext4_orphan_add(handle, inode); |
ac27a0ec DK |
1705 | if (ret) |
1706 | goto out_stop; | |
1707 | orphan = 1; | |
1708 | ei->i_disksize = inode->i_size; | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, | |
1713 | offset, nr_segs, | |
617ba13b | 1714 | ext4_get_block, NULL); |
ac27a0ec DK |
1715 | |
1716 | /* | |
617ba13b | 1717 | * Reacquire the handle: ext4_get_block() can restart the transaction |
ac27a0ec DK |
1718 | */ |
1719 | handle = journal_current_handle(); | |
1720 | ||
1721 | out_stop: | |
1722 | if (handle) { | |
1723 | int err; | |
1724 | ||
1725 | if (orphan && inode->i_nlink) | |
617ba13b | 1726 | ext4_orphan_del(handle, inode); |
ac27a0ec DK |
1727 | if (orphan && ret > 0) { |
1728 | loff_t end = offset + ret; | |
1729 | if (end > inode->i_size) { | |
1730 | ei->i_disksize = end; | |
1731 | i_size_write(inode, end); | |
1732 | /* | |
1733 | * We're going to return a positive `ret' | |
1734 | * here due to non-zero-length I/O, so there's | |
1735 | * no way of reporting error returns from | |
617ba13b | 1736 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
1737 | * ignore it. |
1738 | */ | |
617ba13b | 1739 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1740 | } |
1741 | } | |
617ba13b | 1742 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
1743 | if (ret == 0) |
1744 | ret = err; | |
1745 | } | |
1746 | out: | |
1747 | return ret; | |
1748 | } | |
1749 | ||
1750 | /* | |
617ba13b | 1751 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
1752 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
1753 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
1754 | * not necessarily locked. | |
1755 | * | |
1756 | * We cannot just dirty the page and leave attached buffers clean, because the | |
1757 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
1758 | * or jbddirty because all the journalling code will explode. | |
1759 | * | |
1760 | * So what we do is to mark the page "pending dirty" and next time writepage | |
1761 | * is called, propagate that into the buffers appropriately. | |
1762 | */ | |
617ba13b | 1763 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
1764 | { |
1765 | SetPageChecked(page); | |
1766 | return __set_page_dirty_nobuffers(page); | |
1767 | } | |
1768 | ||
617ba13b MC |
1769 | static const struct address_space_operations ext4_ordered_aops = { |
1770 | .readpage = ext4_readpage, | |
1771 | .readpages = ext4_readpages, | |
1772 | .writepage = ext4_ordered_writepage, | |
ac27a0ec | 1773 | .sync_page = block_sync_page, |
617ba13b MC |
1774 | .prepare_write = ext4_prepare_write, |
1775 | .commit_write = ext4_ordered_commit_write, | |
1776 | .bmap = ext4_bmap, | |
1777 | .invalidatepage = ext4_invalidatepage, | |
1778 | .releasepage = ext4_releasepage, | |
1779 | .direct_IO = ext4_direct_IO, | |
ac27a0ec DK |
1780 | .migratepage = buffer_migrate_page, |
1781 | }; | |
1782 | ||
617ba13b MC |
1783 | static const struct address_space_operations ext4_writeback_aops = { |
1784 | .readpage = ext4_readpage, | |
1785 | .readpages = ext4_readpages, | |
1786 | .writepage = ext4_writeback_writepage, | |
ac27a0ec | 1787 | .sync_page = block_sync_page, |
617ba13b MC |
1788 | .prepare_write = ext4_prepare_write, |
1789 | .commit_write = ext4_writeback_commit_write, | |
1790 | .bmap = ext4_bmap, | |
1791 | .invalidatepage = ext4_invalidatepage, | |
1792 | .releasepage = ext4_releasepage, | |
1793 | .direct_IO = ext4_direct_IO, | |
ac27a0ec DK |
1794 | .migratepage = buffer_migrate_page, |
1795 | }; | |
1796 | ||
617ba13b MC |
1797 | static const struct address_space_operations ext4_journalled_aops = { |
1798 | .readpage = ext4_readpage, | |
1799 | .readpages = ext4_readpages, | |
1800 | .writepage = ext4_journalled_writepage, | |
ac27a0ec | 1801 | .sync_page = block_sync_page, |
617ba13b MC |
1802 | .prepare_write = ext4_prepare_write, |
1803 | .commit_write = ext4_journalled_commit_write, | |
1804 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
1805 | .bmap = ext4_bmap, | |
1806 | .invalidatepage = ext4_invalidatepage, | |
1807 | .releasepage = ext4_releasepage, | |
ac27a0ec DK |
1808 | }; |
1809 | ||
617ba13b | 1810 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 1811 | { |
617ba13b MC |
1812 | if (ext4_should_order_data(inode)) |
1813 | inode->i_mapping->a_ops = &ext4_ordered_aops; | |
1814 | else if (ext4_should_writeback_data(inode)) | |
1815 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 1816 | else |
617ba13b | 1817 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
1818 | } |
1819 | ||
1820 | /* | |
617ba13b | 1821 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
1822 | * up to the end of the block which corresponds to `from'. |
1823 | * This required during truncate. We need to physically zero the tail end | |
1824 | * of that block so it doesn't yield old data if the file is later grown. | |
1825 | */ | |
a86c6181 | 1826 | int ext4_block_truncate_page(handle_t *handle, struct page *page, |
ac27a0ec DK |
1827 | struct address_space *mapping, loff_t from) |
1828 | { | |
617ba13b | 1829 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec DK |
1830 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
1831 | unsigned blocksize, iblock, length, pos; | |
1832 | struct inode *inode = mapping->host; | |
1833 | struct buffer_head *bh; | |
1834 | int err = 0; | |
1835 | void *kaddr; | |
1836 | ||
1837 | blocksize = inode->i_sb->s_blocksize; | |
1838 | length = blocksize - (offset & (blocksize - 1)); | |
1839 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1840 | ||
1841 | /* | |
1842 | * For "nobh" option, we can only work if we don't need to | |
1843 | * read-in the page - otherwise we create buffers to do the IO. | |
1844 | */ | |
1845 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && | |
617ba13b | 1846 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
ac27a0ec DK |
1847 | kaddr = kmap_atomic(page, KM_USER0); |
1848 | memset(kaddr + offset, 0, length); | |
1849 | flush_dcache_page(page); | |
1850 | kunmap_atomic(kaddr, KM_USER0); | |
1851 | set_page_dirty(page); | |
1852 | goto unlock; | |
1853 | } | |
1854 | ||
1855 | if (!page_has_buffers(page)) | |
1856 | create_empty_buffers(page, blocksize, 0); | |
1857 | ||
1858 | /* Find the buffer that contains "offset" */ | |
1859 | bh = page_buffers(page); | |
1860 | pos = blocksize; | |
1861 | while (offset >= pos) { | |
1862 | bh = bh->b_this_page; | |
1863 | iblock++; | |
1864 | pos += blocksize; | |
1865 | } | |
1866 | ||
1867 | err = 0; | |
1868 | if (buffer_freed(bh)) { | |
1869 | BUFFER_TRACE(bh, "freed: skip"); | |
1870 | goto unlock; | |
1871 | } | |
1872 | ||
1873 | if (!buffer_mapped(bh)) { | |
1874 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 1875 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
1876 | /* unmapped? It's a hole - nothing to do */ |
1877 | if (!buffer_mapped(bh)) { | |
1878 | BUFFER_TRACE(bh, "still unmapped"); | |
1879 | goto unlock; | |
1880 | } | |
1881 | } | |
1882 | ||
1883 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
1884 | if (PageUptodate(page)) | |
1885 | set_buffer_uptodate(bh); | |
1886 | ||
1887 | if (!buffer_uptodate(bh)) { | |
1888 | err = -EIO; | |
1889 | ll_rw_block(READ, 1, &bh); | |
1890 | wait_on_buffer(bh); | |
1891 | /* Uhhuh. Read error. Complain and punt. */ | |
1892 | if (!buffer_uptodate(bh)) | |
1893 | goto unlock; | |
1894 | } | |
1895 | ||
617ba13b | 1896 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 1897 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 1898 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1899 | if (err) |
1900 | goto unlock; | |
1901 | } | |
1902 | ||
1903 | kaddr = kmap_atomic(page, KM_USER0); | |
1904 | memset(kaddr + offset, 0, length); | |
1905 | flush_dcache_page(page); | |
1906 | kunmap_atomic(kaddr, KM_USER0); | |
1907 | ||
1908 | BUFFER_TRACE(bh, "zeroed end of block"); | |
1909 | ||
1910 | err = 0; | |
617ba13b MC |
1911 | if (ext4_should_journal_data(inode)) { |
1912 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 1913 | } else { |
617ba13b MC |
1914 | if (ext4_should_order_data(inode)) |
1915 | err = ext4_journal_dirty_data(handle, bh); | |
ac27a0ec DK |
1916 | mark_buffer_dirty(bh); |
1917 | } | |
1918 | ||
1919 | unlock: | |
1920 | unlock_page(page); | |
1921 | page_cache_release(page); | |
1922 | return err; | |
1923 | } | |
1924 | ||
1925 | /* | |
1926 | * Probably it should be a library function... search for first non-zero word | |
1927 | * or memcmp with zero_page, whatever is better for particular architecture. | |
1928 | * Linus? | |
1929 | */ | |
1930 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
1931 | { | |
1932 | while (p < q) | |
1933 | if (*p++) | |
1934 | return 0; | |
1935 | return 1; | |
1936 | } | |
1937 | ||
1938 | /** | |
617ba13b | 1939 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
1940 | * @inode: inode in question |
1941 | * @depth: depth of the affected branch | |
617ba13b | 1942 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
1943 | * @chain: place to store the pointers to partial indirect blocks |
1944 | * @top: place to the (detached) top of branch | |
1945 | * | |
617ba13b | 1946 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
1947 | * |
1948 | * When we do truncate() we may have to clean the ends of several | |
1949 | * indirect blocks but leave the blocks themselves alive. Block is | |
1950 | * partially truncated if some data below the new i_size is refered | |
1951 | * from it (and it is on the path to the first completely truncated | |
1952 | * data block, indeed). We have to free the top of that path along | |
1953 | * with everything to the right of the path. Since no allocation | |
617ba13b | 1954 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
1955 | * finishes, we may safely do the latter, but top of branch may |
1956 | * require special attention - pageout below the truncation point | |
1957 | * might try to populate it. | |
1958 | * | |
1959 | * We atomically detach the top of branch from the tree, store the | |
1960 | * block number of its root in *@top, pointers to buffer_heads of | |
1961 | * partially truncated blocks - in @chain[].bh and pointers to | |
1962 | * their last elements that should not be removed - in | |
1963 | * @chain[].p. Return value is the pointer to last filled element | |
1964 | * of @chain. | |
1965 | * | |
1966 | * The work left to caller to do the actual freeing of subtrees: | |
1967 | * a) free the subtree starting from *@top | |
1968 | * b) free the subtrees whose roots are stored in | |
1969 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
1970 | * c) free the subtrees growing from the inode past the @chain[0]. | |
1971 | * (no partially truncated stuff there). */ | |
1972 | ||
617ba13b | 1973 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
ac27a0ec DK |
1974 | int offsets[4], Indirect chain[4], __le32 *top) |
1975 | { | |
1976 | Indirect *partial, *p; | |
1977 | int k, err; | |
1978 | ||
1979 | *top = 0; | |
1980 | /* Make k index the deepest non-null offest + 1 */ | |
1981 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
1982 | ; | |
617ba13b | 1983 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
1984 | /* Writer: pointers */ |
1985 | if (!partial) | |
1986 | partial = chain + k-1; | |
1987 | /* | |
1988 | * If the branch acquired continuation since we've looked at it - | |
1989 | * fine, it should all survive and (new) top doesn't belong to us. | |
1990 | */ | |
1991 | if (!partial->key && *partial->p) | |
1992 | /* Writer: end */ | |
1993 | goto no_top; | |
1994 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) | |
1995 | ; | |
1996 | /* | |
1997 | * OK, we've found the last block that must survive. The rest of our | |
1998 | * branch should be detached before unlocking. However, if that rest | |
1999 | * of branch is all ours and does not grow immediately from the inode | |
2000 | * it's easier to cheat and just decrement partial->p. | |
2001 | */ | |
2002 | if (p == chain + k - 1 && p > chain) { | |
2003 | p->p--; | |
2004 | } else { | |
2005 | *top = *p->p; | |
617ba13b | 2006 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
2007 | #if 0 |
2008 | *p->p = 0; | |
2009 | #endif | |
2010 | } | |
2011 | /* Writer: end */ | |
2012 | ||
2013 | while(partial > p) { | |
2014 | brelse(partial->bh); | |
2015 | partial--; | |
2016 | } | |
2017 | no_top: | |
2018 | return partial; | |
2019 | } | |
2020 | ||
2021 | /* | |
2022 | * Zero a number of block pointers in either an inode or an indirect block. | |
2023 | * If we restart the transaction we must again get write access to the | |
2024 | * indirect block for further modification. | |
2025 | * | |
2026 | * We release `count' blocks on disk, but (last - first) may be greater | |
2027 | * than `count' because there can be holes in there. | |
2028 | */ | |
617ba13b MC |
2029 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
2030 | struct buffer_head *bh, ext4_fsblk_t block_to_free, | |
ac27a0ec DK |
2031 | unsigned long count, __le32 *first, __le32 *last) |
2032 | { | |
2033 | __le32 *p; | |
2034 | if (try_to_extend_transaction(handle, inode)) { | |
2035 | if (bh) { | |
617ba13b MC |
2036 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
2037 | ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 2038 | } |
617ba13b MC |
2039 | ext4_mark_inode_dirty(handle, inode); |
2040 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
2041 | if (bh) { |
2042 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 2043 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
2044 | } |
2045 | } | |
2046 | ||
2047 | /* | |
2048 | * Any buffers which are on the journal will be in memory. We find | |
dab291af | 2049 | * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget() |
ac27a0ec | 2050 | * on them. We've already detached each block from the file, so |
dab291af | 2051 | * bforget() in jbd2_journal_forget() should be safe. |
ac27a0ec | 2052 | * |
dab291af | 2053 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
ac27a0ec DK |
2054 | */ |
2055 | for (p = first; p < last; p++) { | |
2056 | u32 nr = le32_to_cpu(*p); | |
2057 | if (nr) { | |
2058 | struct buffer_head *bh; | |
2059 | ||
2060 | *p = 0; | |
2061 | bh = sb_find_get_block(inode->i_sb, nr); | |
617ba13b | 2062 | ext4_forget(handle, 0, inode, bh, nr); |
ac27a0ec DK |
2063 | } |
2064 | } | |
2065 | ||
617ba13b | 2066 | ext4_free_blocks(handle, inode, block_to_free, count); |
ac27a0ec DK |
2067 | } |
2068 | ||
2069 | /** | |
617ba13b | 2070 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
2071 | * @handle: handle for this transaction |
2072 | * @inode: inode we are dealing with | |
2073 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
2074 | * @first: array of block numbers | |
2075 | * @last: points immediately past the end of array | |
2076 | * | |
2077 | * We are freeing all blocks refered from that array (numbers are stored as | |
2078 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
2079 | * | |
2080 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
2081 | * blocks are contiguous then releasing them at one time will only affect one | |
2082 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
2083 | * actually use a lot of journal space. | |
2084 | * | |
2085 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
2086 | * block pointers. | |
2087 | */ | |
617ba13b | 2088 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
2089 | struct buffer_head *this_bh, |
2090 | __le32 *first, __le32 *last) | |
2091 | { | |
617ba13b | 2092 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
2093 | unsigned long count = 0; /* Number of blocks in the run */ |
2094 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
2095 | corresponding to | |
2096 | block_to_free */ | |
617ba13b | 2097 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
2098 | __le32 *p; /* Pointer into inode/ind |
2099 | for current block */ | |
2100 | int err; | |
2101 | ||
2102 | if (this_bh) { /* For indirect block */ | |
2103 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 2104 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
2105 | /* Important: if we can't update the indirect pointers |
2106 | * to the blocks, we can't free them. */ | |
2107 | if (err) | |
2108 | return; | |
2109 | } | |
2110 | ||
2111 | for (p = first; p < last; p++) { | |
2112 | nr = le32_to_cpu(*p); | |
2113 | if (nr) { | |
2114 | /* accumulate blocks to free if they're contiguous */ | |
2115 | if (count == 0) { | |
2116 | block_to_free = nr; | |
2117 | block_to_free_p = p; | |
2118 | count = 1; | |
2119 | } else if (nr == block_to_free + count) { | |
2120 | count++; | |
2121 | } else { | |
617ba13b | 2122 | ext4_clear_blocks(handle, inode, this_bh, |
ac27a0ec DK |
2123 | block_to_free, |
2124 | count, block_to_free_p, p); | |
2125 | block_to_free = nr; | |
2126 | block_to_free_p = p; | |
2127 | count = 1; | |
2128 | } | |
2129 | } | |
2130 | } | |
2131 | ||
2132 | if (count > 0) | |
617ba13b | 2133 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
2134 | count, block_to_free_p, p); |
2135 | ||
2136 | if (this_bh) { | |
617ba13b MC |
2137 | BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata"); |
2138 | ext4_journal_dirty_metadata(handle, this_bh); | |
ac27a0ec DK |
2139 | } |
2140 | } | |
2141 | ||
2142 | /** | |
617ba13b | 2143 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
2144 | * @handle: JBD handle for this transaction |
2145 | * @inode: inode we are dealing with | |
2146 | * @parent_bh: the buffer_head which contains *@first and *@last | |
2147 | * @first: array of block numbers | |
2148 | * @last: pointer immediately past the end of array | |
2149 | * @depth: depth of the branches to free | |
2150 | * | |
2151 | * We are freeing all blocks refered from these branches (numbers are | |
2152 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
2153 | * appropriately. | |
2154 | */ | |
617ba13b | 2155 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
2156 | struct buffer_head *parent_bh, |
2157 | __le32 *first, __le32 *last, int depth) | |
2158 | { | |
617ba13b | 2159 | ext4_fsblk_t nr; |
ac27a0ec DK |
2160 | __le32 *p; |
2161 | ||
2162 | if (is_handle_aborted(handle)) | |
2163 | return; | |
2164 | ||
2165 | if (depth--) { | |
2166 | struct buffer_head *bh; | |
617ba13b | 2167 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
2168 | p = last; |
2169 | while (--p >= first) { | |
2170 | nr = le32_to_cpu(*p); | |
2171 | if (!nr) | |
2172 | continue; /* A hole */ | |
2173 | ||
2174 | /* Go read the buffer for the next level down */ | |
2175 | bh = sb_bread(inode->i_sb, nr); | |
2176 | ||
2177 | /* | |
2178 | * A read failure? Report error and clear slot | |
2179 | * (should be rare). | |
2180 | */ | |
2181 | if (!bh) { | |
617ba13b | 2182 | ext4_error(inode->i_sb, "ext4_free_branches", |
2ae02107 | 2183 | "Read failure, inode=%lu, block=%llu", |
ac27a0ec DK |
2184 | inode->i_ino, nr); |
2185 | continue; | |
2186 | } | |
2187 | ||
2188 | /* This zaps the entire block. Bottom up. */ | |
2189 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 2190 | ext4_free_branches(handle, inode, bh, |
ac27a0ec DK |
2191 | (__le32*)bh->b_data, |
2192 | (__le32*)bh->b_data + addr_per_block, | |
2193 | depth); | |
2194 | ||
2195 | /* | |
2196 | * We've probably journalled the indirect block several | |
2197 | * times during the truncate. But it's no longer | |
2198 | * needed and we now drop it from the transaction via | |
dab291af | 2199 | * jbd2_journal_revoke(). |
ac27a0ec DK |
2200 | * |
2201 | * That's easy if it's exclusively part of this | |
2202 | * transaction. But if it's part of the committing | |
dab291af | 2203 | * transaction then jbd2_journal_forget() will simply |
ac27a0ec | 2204 | * brelse() it. That means that if the underlying |
617ba13b | 2205 | * block is reallocated in ext4_get_block(), |
ac27a0ec DK |
2206 | * unmap_underlying_metadata() will find this block |
2207 | * and will try to get rid of it. damn, damn. | |
2208 | * | |
2209 | * If this block has already been committed to the | |
2210 | * journal, a revoke record will be written. And | |
2211 | * revoke records must be emitted *before* clearing | |
2212 | * this block's bit in the bitmaps. | |
2213 | */ | |
617ba13b | 2214 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
ac27a0ec DK |
2215 | |
2216 | /* | |
2217 | * Everything below this this pointer has been | |
2218 | * released. Now let this top-of-subtree go. | |
2219 | * | |
2220 | * We want the freeing of this indirect block to be | |
2221 | * atomic in the journal with the updating of the | |
2222 | * bitmap block which owns it. So make some room in | |
2223 | * the journal. | |
2224 | * | |
2225 | * We zero the parent pointer *after* freeing its | |
2226 | * pointee in the bitmaps, so if extend_transaction() | |
2227 | * for some reason fails to put the bitmap changes and | |
2228 | * the release into the same transaction, recovery | |
2229 | * will merely complain about releasing a free block, | |
2230 | * rather than leaking blocks. | |
2231 | */ | |
2232 | if (is_handle_aborted(handle)) | |
2233 | return; | |
2234 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b MC |
2235 | ext4_mark_inode_dirty(handle, inode); |
2236 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
2237 | } |
2238 | ||
617ba13b | 2239 | ext4_free_blocks(handle, inode, nr, 1); |
ac27a0ec DK |
2240 | |
2241 | if (parent_bh) { | |
2242 | /* | |
2243 | * The block which we have just freed is | |
2244 | * pointed to by an indirect block: journal it | |
2245 | */ | |
2246 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 2247 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
2248 | parent_bh)){ |
2249 | *p = 0; | |
2250 | BUFFER_TRACE(parent_bh, | |
617ba13b MC |
2251 | "call ext4_journal_dirty_metadata"); |
2252 | ext4_journal_dirty_metadata(handle, | |
ac27a0ec DK |
2253 | parent_bh); |
2254 | } | |
2255 | } | |
2256 | } | |
2257 | } else { | |
2258 | /* We have reached the bottom of the tree. */ | |
2259 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 2260 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
2261 | } |
2262 | } | |
2263 | ||
2264 | /* | |
617ba13b | 2265 | * ext4_truncate() |
ac27a0ec | 2266 | * |
617ba13b MC |
2267 | * We block out ext4_get_block() block instantiations across the entire |
2268 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
2269 | * simultaneously on behalf of the same inode. |
2270 | * | |
2271 | * As we work through the truncate and commmit bits of it to the journal there | |
2272 | * is one core, guiding principle: the file's tree must always be consistent on | |
2273 | * disk. We must be able to restart the truncate after a crash. | |
2274 | * | |
2275 | * The file's tree may be transiently inconsistent in memory (although it | |
2276 | * probably isn't), but whenever we close off and commit a journal transaction, | |
2277 | * the contents of (the filesystem + the journal) must be consistent and | |
2278 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
2279 | * left-to-right works OK too). | |
2280 | * | |
2281 | * Note that at recovery time, journal replay occurs *before* the restart of | |
2282 | * truncate against the orphan inode list. | |
2283 | * | |
2284 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 2285 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 2286 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
2287 | * ext4_truncate() to have another go. So there will be instantiated blocks |
2288 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 2289 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 2290 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 2291 | */ |
617ba13b | 2292 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
2293 | { |
2294 | handle_t *handle; | |
617ba13b | 2295 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 2296 | __le32 *i_data = ei->i_data; |
617ba13b | 2297 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
2298 | struct address_space *mapping = inode->i_mapping; |
2299 | int offsets[4]; | |
2300 | Indirect chain[4]; | |
2301 | Indirect *partial; | |
2302 | __le32 nr = 0; | |
2303 | int n; | |
2304 | long last_block; | |
2305 | unsigned blocksize = inode->i_sb->s_blocksize; | |
2306 | struct page *page; | |
2307 | ||
2308 | if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
2309 | S_ISLNK(inode->i_mode))) | |
2310 | return; | |
617ba13b | 2311 | if (ext4_inode_is_fast_symlink(inode)) |
ac27a0ec DK |
2312 | return; |
2313 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
2314 | return; | |
2315 | ||
2316 | /* | |
2317 | * We have to lock the EOF page here, because lock_page() nests | |
dab291af | 2318 | * outside jbd2_journal_start(). |
ac27a0ec DK |
2319 | */ |
2320 | if ((inode->i_size & (blocksize - 1)) == 0) { | |
2321 | /* Block boundary? Nothing to do */ | |
2322 | page = NULL; | |
2323 | } else { | |
2324 | page = grab_cache_page(mapping, | |
2325 | inode->i_size >> PAGE_CACHE_SHIFT); | |
2326 | if (!page) | |
2327 | return; | |
2328 | } | |
2329 | ||
a86c6181 AT |
2330 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
2331 | return ext4_ext_truncate(inode, page); | |
2332 | ||
ac27a0ec DK |
2333 | handle = start_transaction(inode); |
2334 | if (IS_ERR(handle)) { | |
2335 | if (page) { | |
2336 | clear_highpage(page); | |
2337 | flush_dcache_page(page); | |
2338 | unlock_page(page); | |
2339 | page_cache_release(page); | |
2340 | } | |
2341 | return; /* AKPM: return what? */ | |
2342 | } | |
2343 | ||
2344 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 2345 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec DK |
2346 | |
2347 | if (page) | |
617ba13b | 2348 | ext4_block_truncate_page(handle, page, mapping, inode->i_size); |
ac27a0ec | 2349 | |
617ba13b | 2350 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
2351 | if (n == 0) |
2352 | goto out_stop; /* error */ | |
2353 | ||
2354 | /* | |
2355 | * OK. This truncate is going to happen. We add the inode to the | |
2356 | * orphan list, so that if this truncate spans multiple transactions, | |
2357 | * and we crash, we will resume the truncate when the filesystem | |
2358 | * recovers. It also marks the inode dirty, to catch the new size. | |
2359 | * | |
2360 | * Implication: the file must always be in a sane, consistent | |
2361 | * truncatable state while each transaction commits. | |
2362 | */ | |
617ba13b | 2363 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
2364 | goto out_stop; |
2365 | ||
2366 | /* | |
2367 | * The orphan list entry will now protect us from any crash which | |
2368 | * occurs before the truncate completes, so it is now safe to propagate | |
2369 | * the new, shorter inode size (held for now in i_size) into the | |
2370 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 2371 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
2372 | */ |
2373 | ei->i_disksize = inode->i_size; | |
2374 | ||
2375 | /* | |
617ba13b | 2376 | * From here we block out all ext4_get_block() callers who want to |
ac27a0ec DK |
2377 | * modify the block allocation tree. |
2378 | */ | |
2379 | mutex_lock(&ei->truncate_mutex); | |
2380 | ||
2381 | if (n == 1) { /* direct blocks */ | |
617ba13b MC |
2382 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
2383 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
2384 | goto do_indirects; |
2385 | } | |
2386 | ||
617ba13b | 2387 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
2388 | /* Kill the top of shared branch (not detached) */ |
2389 | if (nr) { | |
2390 | if (partial == chain) { | |
2391 | /* Shared branch grows from the inode */ | |
617ba13b | 2392 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
2393 | &nr, &nr+1, (chain+n-1) - partial); |
2394 | *partial->p = 0; | |
2395 | /* | |
2396 | * We mark the inode dirty prior to restart, | |
2397 | * and prior to stop. No need for it here. | |
2398 | */ | |
2399 | } else { | |
2400 | /* Shared branch grows from an indirect block */ | |
2401 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 2402 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
2403 | partial->p, |
2404 | partial->p+1, (chain+n-1) - partial); | |
2405 | } | |
2406 | } | |
2407 | /* Clear the ends of indirect blocks on the shared branch */ | |
2408 | while (partial > chain) { | |
617ba13b | 2409 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
2410 | (__le32*)partial->bh->b_data+addr_per_block, |
2411 | (chain+n-1) - partial); | |
2412 | BUFFER_TRACE(partial->bh, "call brelse"); | |
2413 | brelse (partial->bh); | |
2414 | partial--; | |
2415 | } | |
2416 | do_indirects: | |
2417 | /* Kill the remaining (whole) subtrees */ | |
2418 | switch (offsets[0]) { | |
2419 | default: | |
617ba13b | 2420 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 2421 | if (nr) { |
617ba13b MC |
2422 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
2423 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 2424 | } |
617ba13b MC |
2425 | case EXT4_IND_BLOCK: |
2426 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 2427 | if (nr) { |
617ba13b MC |
2428 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
2429 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 2430 | } |
617ba13b MC |
2431 | case EXT4_DIND_BLOCK: |
2432 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 2433 | if (nr) { |
617ba13b MC |
2434 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
2435 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 2436 | } |
617ba13b | 2437 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
2438 | ; |
2439 | } | |
2440 | ||
617ba13b | 2441 | ext4_discard_reservation(inode); |
ac27a0ec DK |
2442 | |
2443 | mutex_unlock(&ei->truncate_mutex); | |
2444 | inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC; | |
617ba13b | 2445 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2446 | |
2447 | /* | |
2448 | * In a multi-transaction truncate, we only make the final transaction | |
2449 | * synchronous | |
2450 | */ | |
2451 | if (IS_SYNC(inode)) | |
2452 | handle->h_sync = 1; | |
2453 | out_stop: | |
2454 | /* | |
2455 | * If this was a simple ftruncate(), and the file will remain alive | |
2456 | * then we need to clear up the orphan record which we created above. | |
2457 | * However, if this was a real unlink then we were called by | |
617ba13b | 2458 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
2459 | * orphan info for us. |
2460 | */ | |
2461 | if (inode->i_nlink) | |
617ba13b | 2462 | ext4_orphan_del(handle, inode); |
ac27a0ec | 2463 | |
617ba13b | 2464 | ext4_journal_stop(handle); |
ac27a0ec DK |
2465 | } |
2466 | ||
617ba13b MC |
2467 | static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb, |
2468 | unsigned long ino, struct ext4_iloc *iloc) | |
ac27a0ec DK |
2469 | { |
2470 | unsigned long desc, group_desc, block_group; | |
2471 | unsigned long offset; | |
617ba13b | 2472 | ext4_fsblk_t block; |
ac27a0ec | 2473 | struct buffer_head *bh; |
617ba13b | 2474 | struct ext4_group_desc * gdp; |
ac27a0ec | 2475 | |
617ba13b | 2476 | if (!ext4_valid_inum(sb, ino)) { |
ac27a0ec DK |
2477 | /* |
2478 | * This error is already checked for in namei.c unless we are | |
2479 | * looking at an NFS filehandle, in which case no error | |
2480 | * report is needed | |
2481 | */ | |
2482 | return 0; | |
2483 | } | |
2484 | ||
617ba13b MC |
2485 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
2486 | if (block_group >= EXT4_SB(sb)->s_groups_count) { | |
2487 | ext4_error(sb,"ext4_get_inode_block","group >= groups count"); | |
ac27a0ec DK |
2488 | return 0; |
2489 | } | |
2490 | smp_rmb(); | |
617ba13b MC |
2491 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
2492 | desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
2493 | bh = EXT4_SB(sb)->s_group_desc[group_desc]; | |
ac27a0ec | 2494 | if (!bh) { |
617ba13b | 2495 | ext4_error (sb, "ext4_get_inode_block", |
ac27a0ec DK |
2496 | "Descriptor not loaded"); |
2497 | return 0; | |
2498 | } | |
2499 | ||
0d1ee42f AR |
2500 | gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data + |
2501 | desc * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
2502 | /* |
2503 | * Figure out the offset within the block group inode table | |
2504 | */ | |
617ba13b MC |
2505 | offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) * |
2506 | EXT4_INODE_SIZE(sb); | |
8fadc143 AR |
2507 | block = ext4_inode_table(sb, gdp) + |
2508 | (offset >> EXT4_BLOCK_SIZE_BITS(sb)); | |
ac27a0ec DK |
2509 | |
2510 | iloc->block_group = block_group; | |
617ba13b | 2511 | iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1); |
ac27a0ec DK |
2512 | return block; |
2513 | } | |
2514 | ||
2515 | /* | |
617ba13b | 2516 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
2517 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
2518 | * data in memory that is needed to recreate the on-disk version of this | |
2519 | * inode. | |
2520 | */ | |
617ba13b MC |
2521 | static int __ext4_get_inode_loc(struct inode *inode, |
2522 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 2523 | { |
617ba13b | 2524 | ext4_fsblk_t block; |
ac27a0ec DK |
2525 | struct buffer_head *bh; |
2526 | ||
617ba13b | 2527 | block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc); |
ac27a0ec DK |
2528 | if (!block) |
2529 | return -EIO; | |
2530 | ||
2531 | bh = sb_getblk(inode->i_sb, block); | |
2532 | if (!bh) { | |
617ba13b | 2533 | ext4_error (inode->i_sb, "ext4_get_inode_loc", |
ac27a0ec | 2534 | "unable to read inode block - " |
2ae02107 | 2535 | "inode=%lu, block=%llu", |
ac27a0ec DK |
2536 | inode->i_ino, block); |
2537 | return -EIO; | |
2538 | } | |
2539 | if (!buffer_uptodate(bh)) { | |
2540 | lock_buffer(bh); | |
2541 | if (buffer_uptodate(bh)) { | |
2542 | /* someone brought it uptodate while we waited */ | |
2543 | unlock_buffer(bh); | |
2544 | goto has_buffer; | |
2545 | } | |
2546 | ||
2547 | /* | |
2548 | * If we have all information of the inode in memory and this | |
2549 | * is the only valid inode in the block, we need not read the | |
2550 | * block. | |
2551 | */ | |
2552 | if (in_mem) { | |
2553 | struct buffer_head *bitmap_bh; | |
617ba13b | 2554 | struct ext4_group_desc *desc; |
ac27a0ec DK |
2555 | int inodes_per_buffer; |
2556 | int inode_offset, i; | |
2557 | int block_group; | |
2558 | int start; | |
2559 | ||
2560 | block_group = (inode->i_ino - 1) / | |
617ba13b | 2561 | EXT4_INODES_PER_GROUP(inode->i_sb); |
ac27a0ec | 2562 | inodes_per_buffer = bh->b_size / |
617ba13b | 2563 | EXT4_INODE_SIZE(inode->i_sb); |
ac27a0ec | 2564 | inode_offset = ((inode->i_ino - 1) % |
617ba13b | 2565 | EXT4_INODES_PER_GROUP(inode->i_sb)); |
ac27a0ec DK |
2566 | start = inode_offset & ~(inodes_per_buffer - 1); |
2567 | ||
2568 | /* Is the inode bitmap in cache? */ | |
617ba13b | 2569 | desc = ext4_get_group_desc(inode->i_sb, |
ac27a0ec DK |
2570 | block_group, NULL); |
2571 | if (!desc) | |
2572 | goto make_io; | |
2573 | ||
2574 | bitmap_bh = sb_getblk(inode->i_sb, | |
8fadc143 | 2575 | ext4_inode_bitmap(inode->i_sb, desc)); |
ac27a0ec DK |
2576 | if (!bitmap_bh) |
2577 | goto make_io; | |
2578 | ||
2579 | /* | |
2580 | * If the inode bitmap isn't in cache then the | |
2581 | * optimisation may end up performing two reads instead | |
2582 | * of one, so skip it. | |
2583 | */ | |
2584 | if (!buffer_uptodate(bitmap_bh)) { | |
2585 | brelse(bitmap_bh); | |
2586 | goto make_io; | |
2587 | } | |
2588 | for (i = start; i < start + inodes_per_buffer; i++) { | |
2589 | if (i == inode_offset) | |
2590 | continue; | |
617ba13b | 2591 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
2592 | break; |
2593 | } | |
2594 | brelse(bitmap_bh); | |
2595 | if (i == start + inodes_per_buffer) { | |
2596 | /* all other inodes are free, so skip I/O */ | |
2597 | memset(bh->b_data, 0, bh->b_size); | |
2598 | set_buffer_uptodate(bh); | |
2599 | unlock_buffer(bh); | |
2600 | goto has_buffer; | |
2601 | } | |
2602 | } | |
2603 | ||
2604 | make_io: | |
2605 | /* | |
2606 | * There are other valid inodes in the buffer, this inode | |
2607 | * has in-inode xattrs, or we don't have this inode in memory. | |
2608 | * Read the block from disk. | |
2609 | */ | |
2610 | get_bh(bh); | |
2611 | bh->b_end_io = end_buffer_read_sync; | |
2612 | submit_bh(READ_META, bh); | |
2613 | wait_on_buffer(bh); | |
2614 | if (!buffer_uptodate(bh)) { | |
617ba13b | 2615 | ext4_error(inode->i_sb, "ext4_get_inode_loc", |
ac27a0ec | 2616 | "unable to read inode block - " |
2ae02107 | 2617 | "inode=%lu, block=%llu", |
ac27a0ec DK |
2618 | inode->i_ino, block); |
2619 | brelse(bh); | |
2620 | return -EIO; | |
2621 | } | |
2622 | } | |
2623 | has_buffer: | |
2624 | iloc->bh = bh; | |
2625 | return 0; | |
2626 | } | |
2627 | ||
617ba13b | 2628 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
2629 | { |
2630 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b MC |
2631 | return __ext4_get_inode_loc(inode, iloc, |
2632 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); | |
ac27a0ec DK |
2633 | } |
2634 | ||
617ba13b | 2635 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 2636 | { |
617ba13b | 2637 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
2638 | |
2639 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 2640 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 2641 | inode->i_flags |= S_SYNC; |
617ba13b | 2642 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 2643 | inode->i_flags |= S_APPEND; |
617ba13b | 2644 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 2645 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 2646 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 2647 | inode->i_flags |= S_NOATIME; |
617ba13b | 2648 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
2649 | inode->i_flags |= S_DIRSYNC; |
2650 | } | |
2651 | ||
617ba13b | 2652 | void ext4_read_inode(struct inode * inode) |
ac27a0ec | 2653 | { |
617ba13b MC |
2654 | struct ext4_iloc iloc; |
2655 | struct ext4_inode *raw_inode; | |
2656 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
2657 | struct buffer_head *bh; |
2658 | int block; | |
2659 | ||
617ba13b MC |
2660 | #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL |
2661 | ei->i_acl = EXT4_ACL_NOT_CACHED; | |
2662 | ei->i_default_acl = EXT4_ACL_NOT_CACHED; | |
ac27a0ec DK |
2663 | #endif |
2664 | ei->i_block_alloc_info = NULL; | |
2665 | ||
617ba13b | 2666 | if (__ext4_get_inode_loc(inode, &iloc, 0)) |
ac27a0ec DK |
2667 | goto bad_inode; |
2668 | bh = iloc.bh; | |
617ba13b | 2669 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
2670 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
2671 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
2672 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
2673 | if(!(test_opt (inode->i_sb, NO_UID32))) { | |
2674 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; | |
2675 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
2676 | } | |
2677 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
2678 | inode->i_size = le32_to_cpu(raw_inode->i_size); | |
2679 | inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime); | |
2680 | inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime); | |
2681 | inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime); | |
2682 | inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0; | |
2683 | ||
2684 | ei->i_state = 0; | |
2685 | ei->i_dir_start_lookup = 0; | |
2686 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
2687 | /* We now have enough fields to check if the inode was active or not. | |
2688 | * This is needed because nfsd might try to access dead inodes | |
2689 | * the test is that same one that e2fsck uses | |
2690 | * NeilBrown 1999oct15 | |
2691 | */ | |
2692 | if (inode->i_nlink == 0) { | |
2693 | if (inode->i_mode == 0 || | |
617ba13b | 2694 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec DK |
2695 | /* this inode is deleted */ |
2696 | brelse (bh); | |
2697 | goto bad_inode; | |
2698 | } | |
2699 | /* The only unlinked inodes we let through here have | |
2700 | * valid i_mode and are being read by the orphan | |
2701 | * recovery code: that's fine, we're about to complete | |
2702 | * the process of deleting those. */ | |
2703 | } | |
2704 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); | |
2705 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); | |
617ba13b | 2706 | #ifdef EXT4_FRAGMENTS |
ac27a0ec DK |
2707 | ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); |
2708 | ei->i_frag_no = raw_inode->i_frag; | |
2709 | ei->i_frag_size = raw_inode->i_fsize; | |
2710 | #endif | |
2711 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); | |
9b8f1f01 MC |
2712 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
2713 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
2714 | ei->i_file_acl |= |
2715 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
ac27a0ec DK |
2716 | if (!S_ISREG(inode->i_mode)) { |
2717 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); | |
2718 | } else { | |
2719 | inode->i_size |= | |
2720 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; | |
2721 | } | |
2722 | ei->i_disksize = inode->i_size; | |
2723 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
2724 | ei->i_block_group = iloc.block_group; | |
2725 | /* | |
2726 | * NOTE! The in-memory inode i_data array is in little-endian order | |
2727 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
2728 | */ | |
617ba13b | 2729 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
2730 | ei->i_data[block] = raw_inode->i_block[block]; |
2731 | INIT_LIST_HEAD(&ei->i_orphan); | |
2732 | ||
617ba13b MC |
2733 | if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 && |
2734 | EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
ac27a0ec DK |
2735 | /* |
2736 | * When mke2fs creates big inodes it does not zero out | |
617ba13b | 2737 | * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE, |
ac27a0ec DK |
2738 | * so ignore those first few inodes. |
2739 | */ | |
2740 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
617ba13b MC |
2741 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
2742 | EXT4_INODE_SIZE(inode->i_sb)) | |
ac27a0ec DK |
2743 | goto bad_inode; |
2744 | if (ei->i_extra_isize == 0) { | |
2745 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
2746 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
2747 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
2748 | } else { |
2749 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 2750 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 2751 | ei->i_extra_isize; |
617ba13b MC |
2752 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
2753 | ei->i_state |= EXT4_STATE_XATTR; | |
ac27a0ec DK |
2754 | } |
2755 | } else | |
2756 | ei->i_extra_isize = 0; | |
2757 | ||
2758 | if (S_ISREG(inode->i_mode)) { | |
617ba13b MC |
2759 | inode->i_op = &ext4_file_inode_operations; |
2760 | inode->i_fop = &ext4_file_operations; | |
2761 | ext4_set_aops(inode); | |
ac27a0ec | 2762 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
2763 | inode->i_op = &ext4_dir_inode_operations; |
2764 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 2765 | } else if (S_ISLNK(inode->i_mode)) { |
617ba13b MC |
2766 | if (ext4_inode_is_fast_symlink(inode)) |
2767 | inode->i_op = &ext4_fast_symlink_inode_operations; | |
ac27a0ec | 2768 | else { |
617ba13b MC |
2769 | inode->i_op = &ext4_symlink_inode_operations; |
2770 | ext4_set_aops(inode); | |
ac27a0ec DK |
2771 | } |
2772 | } else { | |
617ba13b | 2773 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
2774 | if (raw_inode->i_block[0]) |
2775 | init_special_inode(inode, inode->i_mode, | |
2776 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
2777 | else | |
2778 | init_special_inode(inode, inode->i_mode, | |
2779 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
2780 | } | |
2781 | brelse (iloc.bh); | |
617ba13b | 2782 | ext4_set_inode_flags(inode); |
ac27a0ec DK |
2783 | return; |
2784 | ||
2785 | bad_inode: | |
2786 | make_bad_inode(inode); | |
2787 | return; | |
2788 | } | |
2789 | ||
2790 | /* | |
2791 | * Post the struct inode info into an on-disk inode location in the | |
2792 | * buffer-cache. This gobbles the caller's reference to the | |
2793 | * buffer_head in the inode location struct. | |
2794 | * | |
2795 | * The caller must have write access to iloc->bh. | |
2796 | */ | |
617ba13b | 2797 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 2798 | struct inode *inode, |
617ba13b | 2799 | struct ext4_iloc *iloc) |
ac27a0ec | 2800 | { |
617ba13b MC |
2801 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
2802 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
2803 | struct buffer_head *bh = iloc->bh; |
2804 | int err = 0, rc, block; | |
2805 | ||
2806 | /* For fields not not tracking in the in-memory inode, | |
2807 | * initialise them to zero for new inodes. */ | |
617ba13b MC |
2808 | if (ei->i_state & EXT4_STATE_NEW) |
2809 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
ac27a0ec DK |
2810 | |
2811 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
2812 | if(!(test_opt(inode->i_sb, NO_UID32))) { | |
2813 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); | |
2814 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
2815 | /* | |
2816 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
2817 | * re-used with the upper 16 bits of the uid/gid intact | |
2818 | */ | |
2819 | if(!ei->i_dtime) { | |
2820 | raw_inode->i_uid_high = | |
2821 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
2822 | raw_inode->i_gid_high = | |
2823 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
2824 | } else { | |
2825 | raw_inode->i_uid_high = 0; | |
2826 | raw_inode->i_gid_high = 0; | |
2827 | } | |
2828 | } else { | |
2829 | raw_inode->i_uid_low = | |
2830 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
2831 | raw_inode->i_gid_low = | |
2832 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
2833 | raw_inode->i_uid_high = 0; | |
2834 | raw_inode->i_gid_high = 0; | |
2835 | } | |
2836 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
2837 | raw_inode->i_size = cpu_to_le32(ei->i_disksize); | |
2838 | raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
2839 | raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
2840 | raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
2841 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); | |
2842 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
2843 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); | |
617ba13b | 2844 | #ifdef EXT4_FRAGMENTS |
ac27a0ec DK |
2845 | raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); |
2846 | raw_inode->i_frag = ei->i_frag_no; | |
2847 | raw_inode->i_fsize = ei->i_frag_size; | |
2848 | #endif | |
9b8f1f01 MC |
2849 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
2850 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
2851 | raw_inode->i_file_acl_high = |
2852 | cpu_to_le16(ei->i_file_acl >> 32); | |
ac27a0ec DK |
2853 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); |
2854 | if (!S_ISREG(inode->i_mode)) { | |
2855 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); | |
2856 | } else { | |
2857 | raw_inode->i_size_high = | |
2858 | cpu_to_le32(ei->i_disksize >> 32); | |
2859 | if (ei->i_disksize > 0x7fffffffULL) { | |
2860 | struct super_block *sb = inode->i_sb; | |
617ba13b MC |
2861 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
2862 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
2863 | EXT4_SB(sb)->s_es->s_rev_level == | |
2864 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
ac27a0ec DK |
2865 | /* If this is the first large file |
2866 | * created, add a flag to the superblock. | |
2867 | */ | |
617ba13b MC |
2868 | err = ext4_journal_get_write_access(handle, |
2869 | EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
2870 | if (err) |
2871 | goto out_brelse; | |
617ba13b MC |
2872 | ext4_update_dynamic_rev(sb); |
2873 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
2874 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); | |
ac27a0ec DK |
2875 | sb->s_dirt = 1; |
2876 | handle->h_sync = 1; | |
617ba13b MC |
2877 | err = ext4_journal_dirty_metadata(handle, |
2878 | EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
2879 | } |
2880 | } | |
2881 | } | |
2882 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
2883 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
2884 | if (old_valid_dev(inode->i_rdev)) { | |
2885 | raw_inode->i_block[0] = | |
2886 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
2887 | raw_inode->i_block[1] = 0; | |
2888 | } else { | |
2889 | raw_inode->i_block[0] = 0; | |
2890 | raw_inode->i_block[1] = | |
2891 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
2892 | raw_inode->i_block[2] = 0; | |
2893 | } | |
617ba13b | 2894 | } else for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
2895 | raw_inode->i_block[block] = ei->i_data[block]; |
2896 | ||
2897 | if (ei->i_extra_isize) | |
2898 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); | |
2899 | ||
617ba13b MC |
2900 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
2901 | rc = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
2902 | if (!err) |
2903 | err = rc; | |
617ba13b | 2904 | ei->i_state &= ~EXT4_STATE_NEW; |
ac27a0ec DK |
2905 | |
2906 | out_brelse: | |
2907 | brelse (bh); | |
617ba13b | 2908 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
2909 | return err; |
2910 | } | |
2911 | ||
2912 | /* | |
617ba13b | 2913 | * ext4_write_inode() |
ac27a0ec DK |
2914 | * |
2915 | * We are called from a few places: | |
2916 | * | |
2917 | * - Within generic_file_write() for O_SYNC files. | |
2918 | * Here, there will be no transaction running. We wait for any running | |
2919 | * trasnaction to commit. | |
2920 | * | |
2921 | * - Within sys_sync(), kupdate and such. | |
2922 | * We wait on commit, if tol to. | |
2923 | * | |
2924 | * - Within prune_icache() (PF_MEMALLOC == true) | |
2925 | * Here we simply return. We can't afford to block kswapd on the | |
2926 | * journal commit. | |
2927 | * | |
2928 | * In all cases it is actually safe for us to return without doing anything, | |
2929 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 2930 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
2931 | * knfsd. |
2932 | * | |
2933 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
2934 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
2935 | * which we are interested. | |
2936 | * | |
2937 | * It would be a bug for them to not do this. The code: | |
2938 | * | |
2939 | * mark_inode_dirty(inode) | |
2940 | * stuff(); | |
2941 | * inode->i_size = expr; | |
2942 | * | |
2943 | * is in error because a kswapd-driven write_inode() could occur while | |
2944 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
2945 | * will no longer be on the superblock's dirty inode list. | |
2946 | */ | |
617ba13b | 2947 | int ext4_write_inode(struct inode *inode, int wait) |
ac27a0ec DK |
2948 | { |
2949 | if (current->flags & PF_MEMALLOC) | |
2950 | return 0; | |
2951 | ||
617ba13b | 2952 | if (ext4_journal_current_handle()) { |
ac27a0ec DK |
2953 | jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n"); |
2954 | dump_stack(); | |
2955 | return -EIO; | |
2956 | } | |
2957 | ||
2958 | if (!wait) | |
2959 | return 0; | |
2960 | ||
617ba13b | 2961 | return ext4_force_commit(inode->i_sb); |
ac27a0ec DK |
2962 | } |
2963 | ||
2964 | /* | |
617ba13b | 2965 | * ext4_setattr() |
ac27a0ec DK |
2966 | * |
2967 | * Called from notify_change. | |
2968 | * | |
2969 | * We want to trap VFS attempts to truncate the file as soon as | |
2970 | * possible. In particular, we want to make sure that when the VFS | |
2971 | * shrinks i_size, we put the inode on the orphan list and modify | |
2972 | * i_disksize immediately, so that during the subsequent flushing of | |
2973 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
2974 | * commit will leave the blocks being flushed in an unused state on | |
2975 | * disk. (On recovery, the inode will get truncated and the blocks will | |
2976 | * be freed, so we have a strong guarantee that no future commit will | |
2977 | * leave these blocks visible to the user.) | |
2978 | * | |
2979 | * Called with inode->sem down. | |
2980 | */ | |
617ba13b | 2981 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
2982 | { |
2983 | struct inode *inode = dentry->d_inode; | |
2984 | int error, rc = 0; | |
2985 | const unsigned int ia_valid = attr->ia_valid; | |
2986 | ||
2987 | error = inode_change_ok(inode, attr); | |
2988 | if (error) | |
2989 | return error; | |
2990 | ||
2991 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
2992 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
2993 | handle_t *handle; | |
2994 | ||
2995 | /* (user+group)*(old+new) structure, inode write (sb, | |
2996 | * inode block, ? - but truncate inode update has it) */ | |
617ba13b MC |
2997 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
2998 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
ac27a0ec DK |
2999 | if (IS_ERR(handle)) { |
3000 | error = PTR_ERR(handle); | |
3001 | goto err_out; | |
3002 | } | |
3003 | error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0; | |
3004 | if (error) { | |
617ba13b | 3005 | ext4_journal_stop(handle); |
ac27a0ec DK |
3006 | return error; |
3007 | } | |
3008 | /* Update corresponding info in inode so that everything is in | |
3009 | * one transaction */ | |
3010 | if (attr->ia_valid & ATTR_UID) | |
3011 | inode->i_uid = attr->ia_uid; | |
3012 | if (attr->ia_valid & ATTR_GID) | |
3013 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
3014 | error = ext4_mark_inode_dirty(handle, inode); |
3015 | ext4_journal_stop(handle); | |
ac27a0ec DK |
3016 | } |
3017 | ||
3018 | if (S_ISREG(inode->i_mode) && | |
3019 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
3020 | handle_t *handle; | |
3021 | ||
617ba13b | 3022 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
3023 | if (IS_ERR(handle)) { |
3024 | error = PTR_ERR(handle); | |
3025 | goto err_out; | |
3026 | } | |
3027 | ||
617ba13b MC |
3028 | error = ext4_orphan_add(handle, inode); |
3029 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
3030 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
3031 | if (!error) |
3032 | error = rc; | |
617ba13b | 3033 | ext4_journal_stop(handle); |
ac27a0ec DK |
3034 | } |
3035 | ||
3036 | rc = inode_setattr(inode, attr); | |
3037 | ||
617ba13b | 3038 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
3039 | * transaction handle at all, we need to clean up the in-core |
3040 | * orphan list manually. */ | |
3041 | if (inode->i_nlink) | |
617ba13b | 3042 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
3043 | |
3044 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 3045 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
3046 | |
3047 | err_out: | |
617ba13b | 3048 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
3049 | if (!error) |
3050 | error = rc; | |
3051 | return error; | |
3052 | } | |
3053 | ||
3054 | ||
3055 | /* | |
3056 | * How many blocks doth make a writepage()? | |
3057 | * | |
3058 | * With N blocks per page, it may be: | |
3059 | * N data blocks | |
3060 | * 2 indirect block | |
3061 | * 2 dindirect | |
3062 | * 1 tindirect | |
3063 | * N+5 bitmap blocks (from the above) | |
3064 | * N+5 group descriptor summary blocks | |
3065 | * 1 inode block | |
3066 | * 1 superblock. | |
617ba13b | 3067 | * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files |
ac27a0ec | 3068 | * |
617ba13b | 3069 | * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS |
ac27a0ec DK |
3070 | * |
3071 | * With ordered or writeback data it's the same, less the N data blocks. | |
3072 | * | |
3073 | * If the inode's direct blocks can hold an integral number of pages then a | |
3074 | * page cannot straddle two indirect blocks, and we can only touch one indirect | |
3075 | * and dindirect block, and the "5" above becomes "3". | |
3076 | * | |
3077 | * This still overestimates under most circumstances. If we were to pass the | |
3078 | * start and end offsets in here as well we could do block_to_path() on each | |
3079 | * block and work out the exact number of indirects which are touched. Pah. | |
3080 | */ | |
3081 | ||
a86c6181 | 3082 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 3083 | { |
617ba13b MC |
3084 | int bpp = ext4_journal_blocks_per_page(inode); |
3085 | int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3; | |
ac27a0ec DK |
3086 | int ret; |
3087 | ||
a86c6181 AT |
3088 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
3089 | return ext4_ext_writepage_trans_blocks(inode, bpp); | |
3090 | ||
617ba13b | 3091 | if (ext4_should_journal_data(inode)) |
ac27a0ec DK |
3092 | ret = 3 * (bpp + indirects) + 2; |
3093 | else | |
3094 | ret = 2 * (bpp + indirects) + 2; | |
3095 | ||
3096 | #ifdef CONFIG_QUOTA | |
3097 | /* We know that structure was already allocated during DQUOT_INIT so | |
3098 | * we will be updating only the data blocks + inodes */ | |
617ba13b | 3099 | ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); |
ac27a0ec DK |
3100 | #endif |
3101 | ||
3102 | return ret; | |
3103 | } | |
3104 | ||
3105 | /* | |
617ba13b | 3106 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
3107 | * Give this, we know that the caller already has write access to iloc->bh. |
3108 | */ | |
617ba13b MC |
3109 | int ext4_mark_iloc_dirty(handle_t *handle, |
3110 | struct inode *inode, struct ext4_iloc *iloc) | |
ac27a0ec DK |
3111 | { |
3112 | int err = 0; | |
3113 | ||
3114 | /* the do_update_inode consumes one bh->b_count */ | |
3115 | get_bh(iloc->bh); | |
3116 | ||
dab291af | 3117 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
617ba13b | 3118 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
3119 | put_bh(iloc->bh); |
3120 | return err; | |
3121 | } | |
3122 | ||
3123 | /* | |
3124 | * On success, We end up with an outstanding reference count against | |
3125 | * iloc->bh. This _must_ be cleaned up later. | |
3126 | */ | |
3127 | ||
3128 | int | |
617ba13b MC |
3129 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
3130 | struct ext4_iloc *iloc) | |
ac27a0ec DK |
3131 | { |
3132 | int err = 0; | |
3133 | if (handle) { | |
617ba13b | 3134 | err = ext4_get_inode_loc(inode, iloc); |
ac27a0ec DK |
3135 | if (!err) { |
3136 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
617ba13b | 3137 | err = ext4_journal_get_write_access(handle, iloc->bh); |
ac27a0ec DK |
3138 | if (err) { |
3139 | brelse(iloc->bh); | |
3140 | iloc->bh = NULL; | |
3141 | } | |
3142 | } | |
3143 | } | |
617ba13b | 3144 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
3145 | return err; |
3146 | } | |
3147 | ||
3148 | /* | |
3149 | * What we do here is to mark the in-core inode as clean with respect to inode | |
3150 | * dirtiness (it may still be data-dirty). | |
3151 | * This means that the in-core inode may be reaped by prune_icache | |
3152 | * without having to perform any I/O. This is a very good thing, | |
3153 | * because *any* task may call prune_icache - even ones which | |
3154 | * have a transaction open against a different journal. | |
3155 | * | |
3156 | * Is this cheating? Not really. Sure, we haven't written the | |
3157 | * inode out, but prune_icache isn't a user-visible syncing function. | |
3158 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
3159 | * we start and wait on commits. | |
3160 | * | |
3161 | * Is this efficient/effective? Well, we're being nice to the system | |
3162 | * by cleaning up our inodes proactively so they can be reaped | |
3163 | * without I/O. But we are potentially leaving up to five seconds' | |
3164 | * worth of inodes floating about which prune_icache wants us to | |
3165 | * write out. One way to fix that would be to get prune_icache() | |
3166 | * to do a write_super() to free up some memory. It has the desired | |
3167 | * effect. | |
3168 | */ | |
617ba13b | 3169 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 3170 | { |
617ba13b | 3171 | struct ext4_iloc iloc; |
ac27a0ec DK |
3172 | int err; |
3173 | ||
3174 | might_sleep(); | |
617ba13b | 3175 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
ac27a0ec | 3176 | if (!err) |
617ba13b | 3177 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
3178 | return err; |
3179 | } | |
3180 | ||
3181 | /* | |
617ba13b | 3182 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
3183 | * |
3184 | * We're really interested in the case where a file is being extended. | |
3185 | * i_size has been changed by generic_commit_write() and we thus need | |
3186 | * to include the updated inode in the current transaction. | |
3187 | * | |
3188 | * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks | |
3189 | * are allocated to the file. | |
3190 | * | |
3191 | * If the inode is marked synchronous, we don't honour that here - doing | |
3192 | * so would cause a commit on atime updates, which we don't bother doing. | |
3193 | * We handle synchronous inodes at the highest possible level. | |
3194 | */ | |
617ba13b | 3195 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 3196 | { |
617ba13b | 3197 | handle_t *current_handle = ext4_journal_current_handle(); |
ac27a0ec DK |
3198 | handle_t *handle; |
3199 | ||
617ba13b | 3200 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
3201 | if (IS_ERR(handle)) |
3202 | goto out; | |
3203 | if (current_handle && | |
3204 | current_handle->h_transaction != handle->h_transaction) { | |
3205 | /* This task has a transaction open against a different fs */ | |
3206 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
3207 | __FUNCTION__); | |
3208 | } else { | |
3209 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
3210 | current_handle); | |
617ba13b | 3211 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 3212 | } |
617ba13b | 3213 | ext4_journal_stop(handle); |
ac27a0ec DK |
3214 | out: |
3215 | return; | |
3216 | } | |
3217 | ||
3218 | #if 0 | |
3219 | /* | |
3220 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
3221 | * it from being flushed to disk early. Unlike | |
617ba13b | 3222 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
3223 | * returns no iloc structure, so the caller needs to repeat the iloc |
3224 | * lookup to mark the inode dirty later. | |
3225 | */ | |
617ba13b | 3226 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 3227 | { |
617ba13b | 3228 | struct ext4_iloc iloc; |
ac27a0ec DK |
3229 | |
3230 | int err = 0; | |
3231 | if (handle) { | |
617ba13b | 3232 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
3233 | if (!err) { |
3234 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 3235 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 3236 | if (!err) |
617ba13b | 3237 | err = ext4_journal_dirty_metadata(handle, |
ac27a0ec DK |
3238 | iloc.bh); |
3239 | brelse(iloc.bh); | |
3240 | } | |
3241 | } | |
617ba13b | 3242 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
3243 | return err; |
3244 | } | |
3245 | #endif | |
3246 | ||
617ba13b | 3247 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
3248 | { |
3249 | journal_t *journal; | |
3250 | handle_t *handle; | |
3251 | int err; | |
3252 | ||
3253 | /* | |
3254 | * We have to be very careful here: changing a data block's | |
3255 | * journaling status dynamically is dangerous. If we write a | |
3256 | * data block to the journal, change the status and then delete | |
3257 | * that block, we risk forgetting to revoke the old log record | |
3258 | * from the journal and so a subsequent replay can corrupt data. | |
3259 | * So, first we make sure that the journal is empty and that | |
3260 | * nobody is changing anything. | |
3261 | */ | |
3262 | ||
617ba13b | 3263 | journal = EXT4_JOURNAL(inode); |
ac27a0ec DK |
3264 | if (is_journal_aborted(journal) || IS_RDONLY(inode)) |
3265 | return -EROFS; | |
3266 | ||
dab291af MC |
3267 | jbd2_journal_lock_updates(journal); |
3268 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
3269 | |
3270 | /* | |
3271 | * OK, there are no updates running now, and all cached data is | |
3272 | * synced to disk. We are now in a completely consistent state | |
3273 | * which doesn't have anything in the journal, and we know that | |
3274 | * no filesystem updates are running, so it is safe to modify | |
3275 | * the inode's in-core data-journaling state flag now. | |
3276 | */ | |
3277 | ||
3278 | if (val) | |
617ba13b | 3279 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
ac27a0ec | 3280 | else |
617ba13b MC |
3281 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
3282 | ext4_set_aops(inode); | |
ac27a0ec | 3283 | |
dab291af | 3284 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
3285 | |
3286 | /* Finally we can mark the inode as dirty. */ | |
3287 | ||
617ba13b | 3288 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
3289 | if (IS_ERR(handle)) |
3290 | return PTR_ERR(handle); | |
3291 | ||
617ba13b | 3292 | err = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 3293 | handle->h_sync = 1; |
617ba13b MC |
3294 | ext4_journal_stop(handle); |
3295 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
3296 | |
3297 | return err; | |
3298 | } |