]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Goal-directed block allocation by Stephen Tweedie | |
16 | * (sct@redhat.com), 1993, 1998 | |
17 | * Big-endian to little-endian byte-swapping/bitmaps by | |
18 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
20 | * (jj@sunsite.ms.mff.cuni.cz) | |
21 | * | |
617ba13b | 22 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
23 | */ |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
dab291af | 28 | #include <linux/jbd2.h> |
ac27a0ec DK |
29 | #include <linux/highuid.h> |
30 | #include <linux/pagemap.h> | |
31 | #include <linux/quotaops.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/buffer_head.h> | |
34 | #include <linux/writeback.h> | |
64769240 | 35 | #include <linux/pagevec.h> |
ac27a0ec DK |
36 | #include <linux/mpage.h> |
37 | #include <linux/uio.h> | |
38 | #include <linux/bio.h> | |
3dcf5451 | 39 | #include "ext4_jbd2.h" |
ac27a0ec DK |
40 | #include "xattr.h" |
41 | #include "acl.h" | |
d2a17637 | 42 | #include "ext4_extents.h" |
ac27a0ec | 43 | |
678aaf48 JK |
44 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
45 | loff_t new_size) | |
46 | { | |
47 | return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode, | |
48 | new_size); | |
49 | } | |
50 | ||
64769240 AT |
51 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
52 | ||
ac27a0ec DK |
53 | /* |
54 | * Test whether an inode is a fast symlink. | |
55 | */ | |
617ba13b | 56 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 57 | { |
617ba13b | 58 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
59 | (inode->i_sb->s_blocksize >> 9) : 0; |
60 | ||
61 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
62 | } | |
63 | ||
64 | /* | |
617ba13b | 65 | * The ext4 forget function must perform a revoke if we are freeing data |
ac27a0ec DK |
66 | * which has been journaled. Metadata (eg. indirect blocks) must be |
67 | * revoked in all cases. | |
68 | * | |
69 | * "bh" may be NULL: a metadata block may have been freed from memory | |
70 | * but there may still be a record of it in the journal, and that record | |
71 | * still needs to be revoked. | |
72 | */ | |
617ba13b MC |
73 | int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, |
74 | struct buffer_head *bh, ext4_fsblk_t blocknr) | |
ac27a0ec DK |
75 | { |
76 | int err; | |
77 | ||
78 | might_sleep(); | |
79 | ||
80 | BUFFER_TRACE(bh, "enter"); | |
81 | ||
82 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
83 | "data mode %lx\n", | |
84 | bh, is_metadata, inode->i_mode, | |
85 | test_opt(inode->i_sb, DATA_FLAGS)); | |
86 | ||
87 | /* Never use the revoke function if we are doing full data | |
88 | * journaling: there is no need to, and a V1 superblock won't | |
89 | * support it. Otherwise, only skip the revoke on un-journaled | |
90 | * data blocks. */ | |
91 | ||
617ba13b MC |
92 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
93 | (!is_metadata && !ext4_should_journal_data(inode))) { | |
ac27a0ec | 94 | if (bh) { |
dab291af | 95 | BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
617ba13b | 96 | return ext4_journal_forget(handle, bh); |
ac27a0ec DK |
97 | } |
98 | return 0; | |
99 | } | |
100 | ||
101 | /* | |
102 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
103 | */ | |
617ba13b MC |
104 | BUFFER_TRACE(bh, "call ext4_journal_revoke"); |
105 | err = ext4_journal_revoke(handle, blocknr, bh); | |
ac27a0ec | 106 | if (err) |
46e665e9 | 107 | ext4_abort(inode->i_sb, __func__, |
ac27a0ec DK |
108 | "error %d when attempting revoke", err); |
109 | BUFFER_TRACE(bh, "exit"); | |
110 | return err; | |
111 | } | |
112 | ||
113 | /* | |
114 | * Work out how many blocks we need to proceed with the next chunk of a | |
115 | * truncate transaction. | |
116 | */ | |
117 | static unsigned long blocks_for_truncate(struct inode *inode) | |
118 | { | |
725d26d3 | 119 | ext4_lblk_t needed; |
ac27a0ec DK |
120 | |
121 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
122 | ||
123 | /* Give ourselves just enough room to cope with inodes in which | |
124 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
125 | * which resulted in random data in an inode which looked enough | |
617ba13b | 126 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
127 | * will go a bit crazy if that happens, but at least we should |
128 | * try not to panic the whole kernel. */ | |
129 | if (needed < 2) | |
130 | needed = 2; | |
131 | ||
132 | /* But we need to bound the transaction so we don't overflow the | |
133 | * journal. */ | |
617ba13b MC |
134 | if (needed > EXT4_MAX_TRANS_DATA) |
135 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 136 | |
617ba13b | 137 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
138 | } |
139 | ||
140 | /* | |
141 | * Truncate transactions can be complex and absolutely huge. So we need to | |
142 | * be able to restart the transaction at a conventient checkpoint to make | |
143 | * sure we don't overflow the journal. | |
144 | * | |
145 | * start_transaction gets us a new handle for a truncate transaction, | |
146 | * and extend_transaction tries to extend the existing one a bit. If | |
147 | * extend fails, we need to propagate the failure up and restart the | |
148 | * transaction in the top-level truncate loop. --sct | |
149 | */ | |
150 | static handle_t *start_transaction(struct inode *inode) | |
151 | { | |
152 | handle_t *result; | |
153 | ||
617ba13b | 154 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
155 | if (!IS_ERR(result)) |
156 | return result; | |
157 | ||
617ba13b | 158 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
159 | return result; |
160 | } | |
161 | ||
162 | /* | |
163 | * Try to extend this transaction for the purposes of truncation. | |
164 | * | |
165 | * Returns 0 if we managed to create more room. If we can't create more | |
166 | * room, and the transaction must be restarted we return 1. | |
167 | */ | |
168 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
169 | { | |
617ba13b | 170 | if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS) |
ac27a0ec | 171 | return 0; |
617ba13b | 172 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
173 | return 0; |
174 | return 1; | |
175 | } | |
176 | ||
177 | /* | |
178 | * Restart the transaction associated with *handle. This does a commit, | |
179 | * so before we call here everything must be consistently dirtied against | |
180 | * this transaction. | |
181 | */ | |
617ba13b | 182 | static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) |
ac27a0ec DK |
183 | { |
184 | jbd_debug(2, "restarting handle %p\n", handle); | |
617ba13b | 185 | return ext4_journal_restart(handle, blocks_for_truncate(inode)); |
ac27a0ec DK |
186 | } |
187 | ||
188 | /* | |
189 | * Called at the last iput() if i_nlink is zero. | |
190 | */ | |
617ba13b | 191 | void ext4_delete_inode (struct inode * inode) |
ac27a0ec DK |
192 | { |
193 | handle_t *handle; | |
bc965ab3 | 194 | int err; |
ac27a0ec | 195 | |
678aaf48 JK |
196 | if (ext4_should_order_data(inode)) |
197 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
198 | truncate_inode_pages(&inode->i_data, 0); |
199 | ||
200 | if (is_bad_inode(inode)) | |
201 | goto no_delete; | |
202 | ||
bc965ab3 | 203 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 204 | if (IS_ERR(handle)) { |
bc965ab3 | 205 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
206 | /* |
207 | * If we're going to skip the normal cleanup, we still need to | |
208 | * make sure that the in-core orphan linked list is properly | |
209 | * cleaned up. | |
210 | */ | |
617ba13b | 211 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
212 | goto no_delete; |
213 | } | |
214 | ||
215 | if (IS_SYNC(inode)) | |
216 | handle->h_sync = 1; | |
217 | inode->i_size = 0; | |
bc965ab3 TT |
218 | err = ext4_mark_inode_dirty(handle, inode); |
219 | if (err) { | |
220 | ext4_warning(inode->i_sb, __func__, | |
221 | "couldn't mark inode dirty (err %d)", err); | |
222 | goto stop_handle; | |
223 | } | |
ac27a0ec | 224 | if (inode->i_blocks) |
617ba13b | 225 | ext4_truncate(inode); |
bc965ab3 TT |
226 | |
227 | /* | |
228 | * ext4_ext_truncate() doesn't reserve any slop when it | |
229 | * restarts journal transactions; therefore there may not be | |
230 | * enough credits left in the handle to remove the inode from | |
231 | * the orphan list and set the dtime field. | |
232 | */ | |
233 | if (handle->h_buffer_credits < 3) { | |
234 | err = ext4_journal_extend(handle, 3); | |
235 | if (err > 0) | |
236 | err = ext4_journal_restart(handle, 3); | |
237 | if (err != 0) { | |
238 | ext4_warning(inode->i_sb, __func__, | |
239 | "couldn't extend journal (err %d)", err); | |
240 | stop_handle: | |
241 | ext4_journal_stop(handle); | |
242 | goto no_delete; | |
243 | } | |
244 | } | |
245 | ||
ac27a0ec | 246 | /* |
617ba13b | 247 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 248 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 249 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 250 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 251 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
252 | * (Well, we could do this if we need to, but heck - it works) |
253 | */ | |
617ba13b MC |
254 | ext4_orphan_del(handle, inode); |
255 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
256 | |
257 | /* | |
258 | * One subtle ordering requirement: if anything has gone wrong | |
259 | * (transaction abort, IO errors, whatever), then we can still | |
260 | * do these next steps (the fs will already have been marked as | |
261 | * having errors), but we can't free the inode if the mark_dirty | |
262 | * fails. | |
263 | */ | |
617ba13b | 264 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
265 | /* If that failed, just do the required in-core inode clear. */ |
266 | clear_inode(inode); | |
267 | else | |
617ba13b MC |
268 | ext4_free_inode(handle, inode); |
269 | ext4_journal_stop(handle); | |
ac27a0ec DK |
270 | return; |
271 | no_delete: | |
272 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
273 | } | |
274 | ||
275 | typedef struct { | |
276 | __le32 *p; | |
277 | __le32 key; | |
278 | struct buffer_head *bh; | |
279 | } Indirect; | |
280 | ||
281 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
282 | { | |
283 | p->key = *(p->p = v); | |
284 | p->bh = bh; | |
285 | } | |
286 | ||
ac27a0ec | 287 | /** |
617ba13b | 288 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
289 | * @inode: inode in question (we are only interested in its superblock) |
290 | * @i_block: block number to be parsed | |
291 | * @offsets: array to store the offsets in | |
8c55e204 DK |
292 | * @boundary: set this non-zero if the referred-to block is likely to be |
293 | * followed (on disk) by an indirect block. | |
ac27a0ec | 294 | * |
617ba13b | 295 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
296 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
297 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
298 | * This function translates the block number into path in that tree - | |
299 | * return value is the path length and @offsets[n] is the offset of | |
300 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
301 | * (negative or too large) warning is printed and zero returned. | |
302 | * | |
303 | * Note: function doesn't find node addresses, so no IO is needed. All | |
304 | * we need to know is the capacity of indirect blocks (taken from the | |
305 | * inode->i_sb). | |
306 | */ | |
307 | ||
308 | /* | |
309 | * Portability note: the last comparison (check that we fit into triple | |
310 | * indirect block) is spelled differently, because otherwise on an | |
311 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
312 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
313 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
314 | * i_block would have to be negative in the very beginning, so we would not | |
315 | * get there at all. | |
316 | */ | |
317 | ||
617ba13b | 318 | static int ext4_block_to_path(struct inode *inode, |
725d26d3 AK |
319 | ext4_lblk_t i_block, |
320 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 321 | { |
617ba13b MC |
322 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
323 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
324 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
325 | indirect_blocks = ptrs, |
326 | double_blocks = (1 << (ptrs_bits * 2)); | |
327 | int n = 0; | |
328 | int final = 0; | |
329 | ||
330 | if (i_block < 0) { | |
617ba13b | 331 | ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0"); |
ac27a0ec DK |
332 | } else if (i_block < direct_blocks) { |
333 | offsets[n++] = i_block; | |
334 | final = direct_blocks; | |
335 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { | |
617ba13b | 336 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
337 | offsets[n++] = i_block; |
338 | final = ptrs; | |
339 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 340 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
341 | offsets[n++] = i_block >> ptrs_bits; |
342 | offsets[n++] = i_block & (ptrs - 1); | |
343 | final = ptrs; | |
344 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 345 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
346 | offsets[n++] = i_block >> (ptrs_bits * 2); |
347 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
348 | offsets[n++] = i_block & (ptrs - 1); | |
349 | final = ptrs; | |
350 | } else { | |
e2b46574 | 351 | ext4_warning(inode->i_sb, "ext4_block_to_path", |
0e855ac8 | 352 | "block %lu > max", |
e2b46574 ES |
353 | i_block + direct_blocks + |
354 | indirect_blocks + double_blocks); | |
ac27a0ec DK |
355 | } |
356 | if (boundary) | |
357 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
358 | return n; | |
359 | } | |
360 | ||
361 | /** | |
617ba13b | 362 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
363 | * @inode: inode in question |
364 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
365 | * @offsets: offsets of pointers in inode/indirect blocks | |
366 | * @chain: place to store the result | |
367 | * @err: here we store the error value | |
368 | * | |
369 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
370 | * if everything went OK or the pointer to the last filled triple | |
371 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
372 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
373 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
374 | * number (it points into struct inode for i==0 and into the bh->b_data | |
375 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
376 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
377 | * numbers of the chain, addresses they were taken from (and where we can | |
378 | * verify that chain did not change) and buffer_heads hosting these | |
379 | * numbers. | |
380 | * | |
381 | * Function stops when it stumbles upon zero pointer (absent block) | |
382 | * (pointer to last triple returned, *@err == 0) | |
383 | * or when it gets an IO error reading an indirect block | |
384 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
385 | * or when it reads all @depth-1 indirect blocks successfully and finds |
386 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
387 | * |
388 | * Need to be called with | |
0e855ac8 | 389 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 390 | */ |
725d26d3 AK |
391 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
392 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
393 | Indirect chain[4], int *err) |
394 | { | |
395 | struct super_block *sb = inode->i_sb; | |
396 | Indirect *p = chain; | |
397 | struct buffer_head *bh; | |
398 | ||
399 | *err = 0; | |
400 | /* i_data is not going away, no lock needed */ | |
617ba13b | 401 | add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
402 | if (!p->key) |
403 | goto no_block; | |
404 | while (--depth) { | |
405 | bh = sb_bread(sb, le32_to_cpu(p->key)); | |
406 | if (!bh) | |
407 | goto failure; | |
ac27a0ec DK |
408 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); |
409 | /* Reader: end */ | |
410 | if (!p->key) | |
411 | goto no_block; | |
412 | } | |
413 | return NULL; | |
414 | ||
ac27a0ec DK |
415 | failure: |
416 | *err = -EIO; | |
417 | no_block: | |
418 | return p; | |
419 | } | |
420 | ||
421 | /** | |
617ba13b | 422 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
423 | * @inode: owner |
424 | * @ind: descriptor of indirect block. | |
425 | * | |
1cc8dcf5 | 426 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
427 | * It is used when heuristic for sequential allocation fails. |
428 | * Rules are: | |
429 | * + if there is a block to the left of our position - allocate near it. | |
430 | * + if pointer will live in indirect block - allocate near that block. | |
431 | * + if pointer will live in inode - allocate in the same | |
432 | * cylinder group. | |
433 | * | |
434 | * In the latter case we colour the starting block by the callers PID to | |
435 | * prevent it from clashing with concurrent allocations for a different inode | |
436 | * in the same block group. The PID is used here so that functionally related | |
437 | * files will be close-by on-disk. | |
438 | * | |
439 | * Caller must make sure that @ind is valid and will stay that way. | |
440 | */ | |
617ba13b | 441 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 442 | { |
617ba13b | 443 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec DK |
444 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; |
445 | __le32 *p; | |
617ba13b | 446 | ext4_fsblk_t bg_start; |
74d3487f | 447 | ext4_fsblk_t last_block; |
617ba13b | 448 | ext4_grpblk_t colour; |
ac27a0ec DK |
449 | |
450 | /* Try to find previous block */ | |
451 | for (p = ind->p - 1; p >= start; p--) { | |
452 | if (*p) | |
453 | return le32_to_cpu(*p); | |
454 | } | |
455 | ||
456 | /* No such thing, so let's try location of indirect block */ | |
457 | if (ind->bh) | |
458 | return ind->bh->b_blocknr; | |
459 | ||
460 | /* | |
461 | * It is going to be referred to from the inode itself? OK, just put it | |
462 | * into the same cylinder group then. | |
463 | */ | |
617ba13b | 464 | bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group); |
74d3487f VC |
465 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
466 | ||
467 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) | |
468 | colour = (current->pid % 16) * | |
617ba13b | 469 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
470 | else |
471 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
472 | return bg_start + colour; |
473 | } | |
474 | ||
475 | /** | |
1cc8dcf5 | 476 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
477 | * @inode: owner |
478 | * @block: block we want | |
ac27a0ec | 479 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 480 | * |
1cc8dcf5 | 481 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 482 | * returns it. |
ac27a0ec | 483 | */ |
725d26d3 | 484 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
fb01bfda | 485 | Indirect *partial) |
ac27a0ec | 486 | { |
617ba13b | 487 | struct ext4_block_alloc_info *block_i; |
ac27a0ec | 488 | |
617ba13b | 489 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
490 | |
491 | /* | |
492 | * try the heuristic for sequential allocation, | |
493 | * failing that at least try to get decent locality. | |
494 | */ | |
495 | if (block_i && (block == block_i->last_alloc_logical_block + 1) | |
496 | && (block_i->last_alloc_physical_block != 0)) { | |
497 | return block_i->last_alloc_physical_block + 1; | |
498 | } | |
499 | ||
617ba13b | 500 | return ext4_find_near(inode, partial); |
ac27a0ec DK |
501 | } |
502 | ||
503 | /** | |
617ba13b | 504 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
505 | * of direct blocks need to be allocated for the given branch. |
506 | * | |
507 | * @branch: chain of indirect blocks | |
508 | * @k: number of blocks need for indirect blocks | |
509 | * @blks: number of data blocks to be mapped. | |
510 | * @blocks_to_boundary: the offset in the indirect block | |
511 | * | |
512 | * return the total number of blocks to be allocate, including the | |
513 | * direct and indirect blocks. | |
514 | */ | |
617ba13b | 515 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
ac27a0ec DK |
516 | int blocks_to_boundary) |
517 | { | |
518 | unsigned long count = 0; | |
519 | ||
520 | /* | |
521 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
522 | * then it's clear blocks on that path have not allocated | |
523 | */ | |
524 | if (k > 0) { | |
525 | /* right now we don't handle cross boundary allocation */ | |
526 | if (blks < blocks_to_boundary + 1) | |
527 | count += blks; | |
528 | else | |
529 | count += blocks_to_boundary + 1; | |
530 | return count; | |
531 | } | |
532 | ||
533 | count++; | |
534 | while (count < blks && count <= blocks_to_boundary && | |
535 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
536 | count++; | |
537 | } | |
538 | return count; | |
539 | } | |
540 | ||
541 | /** | |
617ba13b | 542 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
543 | * @indirect_blks: the number of blocks need to allocate for indirect |
544 | * blocks | |
545 | * | |
546 | * @new_blocks: on return it will store the new block numbers for | |
547 | * the indirect blocks(if needed) and the first direct block, | |
548 | * @blks: on return it will store the total number of allocated | |
549 | * direct blocks | |
550 | */ | |
617ba13b | 551 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
7061eba7 AK |
552 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
553 | int indirect_blks, int blks, | |
554 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec DK |
555 | { |
556 | int target, i; | |
7061eba7 | 557 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 558 | int index = 0; |
617ba13b | 559 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
560 | int ret = 0; |
561 | ||
562 | /* | |
563 | * Here we try to allocate the requested multiple blocks at once, | |
564 | * on a best-effort basis. | |
565 | * To build a branch, we should allocate blocks for | |
566 | * the indirect blocks(if not allocated yet), and at least | |
567 | * the first direct block of this branch. That's the | |
568 | * minimum number of blocks need to allocate(required) | |
569 | */ | |
7061eba7 AK |
570 | /* first we try to allocate the indirect blocks */ |
571 | target = indirect_blks; | |
572 | while (target > 0) { | |
ac27a0ec DK |
573 | count = target; |
574 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
575 | current_block = ext4_new_meta_blocks(handle, inode, |
576 | goal, &count, err); | |
ac27a0ec DK |
577 | if (*err) |
578 | goto failed_out; | |
579 | ||
580 | target -= count; | |
581 | /* allocate blocks for indirect blocks */ | |
582 | while (index < indirect_blks && count) { | |
583 | new_blocks[index++] = current_block++; | |
584 | count--; | |
585 | } | |
7061eba7 AK |
586 | if (count > 0) { |
587 | /* | |
588 | * save the new block number | |
589 | * for the first direct block | |
590 | */ | |
591 | new_blocks[index] = current_block; | |
592 | printk(KERN_INFO "%s returned more blocks than " | |
593 | "requested\n", __func__); | |
594 | WARN_ON(1); | |
ac27a0ec | 595 | break; |
7061eba7 | 596 | } |
ac27a0ec DK |
597 | } |
598 | ||
7061eba7 AK |
599 | target = blks - count ; |
600 | blk_allocated = count; | |
601 | if (!target) | |
602 | goto allocated; | |
603 | /* Now allocate data blocks */ | |
604 | count = target; | |
654b4908 | 605 | /* allocating blocks for data blocks */ |
7061eba7 AK |
606 | current_block = ext4_new_blocks(handle, inode, iblock, |
607 | goal, &count, err); | |
608 | if (*err && (target == blks)) { | |
609 | /* | |
610 | * if the allocation failed and we didn't allocate | |
611 | * any blocks before | |
612 | */ | |
613 | goto failed_out; | |
614 | } | |
615 | if (!*err) { | |
616 | if (target == blks) { | |
617 | /* | |
618 | * save the new block number | |
619 | * for the first direct block | |
620 | */ | |
621 | new_blocks[index] = current_block; | |
622 | } | |
623 | blk_allocated += count; | |
624 | } | |
625 | allocated: | |
ac27a0ec | 626 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 627 | ret = blk_allocated; |
ac27a0ec DK |
628 | *err = 0; |
629 | return ret; | |
630 | failed_out: | |
631 | for (i = 0; i <index; i++) | |
c9de560d | 632 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec DK |
633 | return ret; |
634 | } | |
635 | ||
636 | /** | |
617ba13b | 637 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
638 | * @inode: owner |
639 | * @indirect_blks: number of allocated indirect blocks | |
640 | * @blks: number of allocated direct blocks | |
641 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
642 | * @branch: place to store the chain in. | |
643 | * | |
644 | * This function allocates blocks, zeroes out all but the last one, | |
645 | * links them into chain and (if we are synchronous) writes them to disk. | |
646 | * In other words, it prepares a branch that can be spliced onto the | |
647 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 648 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
649 | * we had read the existing part of chain and partial points to the last |
650 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 651 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
652 | * place chain is disconnected - *branch->p is still zero (we did not |
653 | * set the last link), but branch->key contains the number that should | |
654 | * be placed into *branch->p to fill that gap. | |
655 | * | |
656 | * If allocation fails we free all blocks we've allocated (and forget | |
657 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 658 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
659 | * as described above and return 0. |
660 | */ | |
617ba13b | 661 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
7061eba7 AK |
662 | ext4_lblk_t iblock, int indirect_blks, |
663 | int *blks, ext4_fsblk_t goal, | |
664 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
665 | { |
666 | int blocksize = inode->i_sb->s_blocksize; | |
667 | int i, n = 0; | |
668 | int err = 0; | |
669 | struct buffer_head *bh; | |
670 | int num; | |
617ba13b MC |
671 | ext4_fsblk_t new_blocks[4]; |
672 | ext4_fsblk_t current_block; | |
ac27a0ec | 673 | |
7061eba7 | 674 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
675 | *blks, new_blocks, &err); |
676 | if (err) | |
677 | return err; | |
678 | ||
679 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
680 | /* | |
681 | * metadata blocks and data blocks are allocated. | |
682 | */ | |
683 | for (n = 1; n <= indirect_blks; n++) { | |
684 | /* | |
685 | * Get buffer_head for parent block, zero it out | |
686 | * and set the pointer to new one, then send | |
687 | * parent to disk. | |
688 | */ | |
689 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
690 | branch[n].bh = bh; | |
691 | lock_buffer(bh); | |
692 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 693 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
694 | if (err) { |
695 | unlock_buffer(bh); | |
696 | brelse(bh); | |
697 | goto failed; | |
698 | } | |
699 | ||
700 | memset(bh->b_data, 0, blocksize); | |
701 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
702 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
703 | *branch[n].p = branch[n].key; | |
704 | if ( n == indirect_blks) { | |
705 | current_block = new_blocks[n]; | |
706 | /* | |
707 | * End of chain, update the last new metablock of | |
708 | * the chain to point to the new allocated | |
709 | * data blocks numbers | |
710 | */ | |
711 | for (i=1; i < num; i++) | |
712 | *(branch[n].p + i) = cpu_to_le32(++current_block); | |
713 | } | |
714 | BUFFER_TRACE(bh, "marking uptodate"); | |
715 | set_buffer_uptodate(bh); | |
716 | unlock_buffer(bh); | |
717 | ||
617ba13b MC |
718 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
719 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
720 | if (err) |
721 | goto failed; | |
722 | } | |
723 | *blks = num; | |
724 | return err; | |
725 | failed: | |
726 | /* Allocation failed, free what we already allocated */ | |
727 | for (i = 1; i <= n ; i++) { | |
dab291af | 728 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
617ba13b | 729 | ext4_journal_forget(handle, branch[i].bh); |
ac27a0ec DK |
730 | } |
731 | for (i = 0; i <indirect_blks; i++) | |
c9de560d | 732 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec | 733 | |
c9de560d | 734 | ext4_free_blocks(handle, inode, new_blocks[i], num, 0); |
ac27a0ec DK |
735 | |
736 | return err; | |
737 | } | |
738 | ||
739 | /** | |
617ba13b | 740 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
741 | * @inode: owner |
742 | * @block: (logical) number of block we are adding | |
743 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 744 | * ext4_alloc_branch) |
ac27a0ec DK |
745 | * @where: location of missing link |
746 | * @num: number of indirect blocks we are adding | |
747 | * @blks: number of direct blocks we are adding | |
748 | * | |
749 | * This function fills the missing link and does all housekeeping needed in | |
750 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
751 | * chain to new block and return 0. | |
752 | */ | |
617ba13b | 753 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
725d26d3 | 754 | ext4_lblk_t block, Indirect *where, int num, int blks) |
ac27a0ec DK |
755 | { |
756 | int i; | |
757 | int err = 0; | |
617ba13b MC |
758 | struct ext4_block_alloc_info *block_i; |
759 | ext4_fsblk_t current_block; | |
ac27a0ec | 760 | |
617ba13b | 761 | block_i = EXT4_I(inode)->i_block_alloc_info; |
ac27a0ec DK |
762 | /* |
763 | * If we're splicing into a [td]indirect block (as opposed to the | |
764 | * inode) then we need to get write access to the [td]indirect block | |
765 | * before the splice. | |
766 | */ | |
767 | if (where->bh) { | |
768 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 769 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
770 | if (err) |
771 | goto err_out; | |
772 | } | |
773 | /* That's it */ | |
774 | ||
775 | *where->p = where->key; | |
776 | ||
777 | /* | |
778 | * Update the host buffer_head or inode to point to more just allocated | |
779 | * direct blocks blocks | |
780 | */ | |
781 | if (num == 0 && blks > 1) { | |
782 | current_block = le32_to_cpu(where->key) + 1; | |
783 | for (i = 1; i < blks; i++) | |
784 | *(where->p + i ) = cpu_to_le32(current_block++); | |
785 | } | |
786 | ||
787 | /* | |
788 | * update the most recently allocated logical & physical block | |
789 | * in i_block_alloc_info, to assist find the proper goal block for next | |
790 | * allocation | |
791 | */ | |
792 | if (block_i) { | |
793 | block_i->last_alloc_logical_block = block + blks - 1; | |
794 | block_i->last_alloc_physical_block = | |
795 | le32_to_cpu(where[num].key) + blks - 1; | |
796 | } | |
797 | ||
798 | /* We are done with atomic stuff, now do the rest of housekeeping */ | |
799 | ||
ef7f3835 | 800 | inode->i_ctime = ext4_current_time(inode); |
617ba13b | 801 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
802 | |
803 | /* had we spliced it onto indirect block? */ | |
804 | if (where->bh) { | |
805 | /* | |
806 | * If we spliced it onto an indirect block, we haven't | |
807 | * altered the inode. Note however that if it is being spliced | |
808 | * onto an indirect block at the very end of the file (the | |
809 | * file is growing) then we *will* alter the inode to reflect | |
810 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 811 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
812 | */ |
813 | jbd_debug(5, "splicing indirect only\n"); | |
617ba13b MC |
814 | BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata"); |
815 | err = ext4_journal_dirty_metadata(handle, where->bh); | |
ac27a0ec DK |
816 | if (err) |
817 | goto err_out; | |
818 | } else { | |
819 | /* | |
820 | * OK, we spliced it into the inode itself on a direct block. | |
821 | * Inode was dirtied above. | |
822 | */ | |
823 | jbd_debug(5, "splicing direct\n"); | |
824 | } | |
825 | return err; | |
826 | ||
827 | err_out: | |
828 | for (i = 1; i <= num; i++) { | |
dab291af | 829 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
617ba13b | 830 | ext4_journal_forget(handle, where[i].bh); |
c9de560d AT |
831 | ext4_free_blocks(handle, inode, |
832 | le32_to_cpu(where[i-1].key), 1, 0); | |
ac27a0ec | 833 | } |
c9de560d | 834 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0); |
ac27a0ec DK |
835 | |
836 | return err; | |
837 | } | |
838 | ||
839 | /* | |
840 | * Allocation strategy is simple: if we have to allocate something, we will | |
841 | * have to go the whole way to leaf. So let's do it before attaching anything | |
842 | * to tree, set linkage between the newborn blocks, write them if sync is | |
843 | * required, recheck the path, free and repeat if check fails, otherwise | |
844 | * set the last missing link (that will protect us from any truncate-generated | |
845 | * removals - all blocks on the path are immune now) and possibly force the | |
846 | * write on the parent block. | |
847 | * That has a nice additional property: no special recovery from the failed | |
848 | * allocations is needed - we simply release blocks and do not touch anything | |
849 | * reachable from inode. | |
850 | * | |
851 | * `handle' can be NULL if create == 0. | |
852 | * | |
ac27a0ec DK |
853 | * return > 0, # of blocks mapped or allocated. |
854 | * return = 0, if plain lookup failed. | |
855 | * return < 0, error case. | |
c278bfec AK |
856 | * |
857 | * | |
858 | * Need to be called with | |
0e855ac8 AK |
859 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block |
860 | * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) | |
ac27a0ec | 861 | */ |
617ba13b | 862 | int ext4_get_blocks_handle(handle_t *handle, struct inode *inode, |
725d26d3 | 863 | ext4_lblk_t iblock, unsigned long maxblocks, |
ac27a0ec DK |
864 | struct buffer_head *bh_result, |
865 | int create, int extend_disksize) | |
866 | { | |
867 | int err = -EIO; | |
725d26d3 | 868 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
869 | Indirect chain[4]; |
870 | Indirect *partial; | |
617ba13b | 871 | ext4_fsblk_t goal; |
ac27a0ec DK |
872 | int indirect_blks; |
873 | int blocks_to_boundary = 0; | |
874 | int depth; | |
617ba13b | 875 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 876 | int count = 0; |
617ba13b | 877 | ext4_fsblk_t first_block = 0; |
61628a3f | 878 | loff_t disksize; |
ac27a0ec DK |
879 | |
880 | ||
a86c6181 | 881 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
ac27a0ec | 882 | J_ASSERT(handle != NULL || create == 0); |
725d26d3 AK |
883 | depth = ext4_block_to_path(inode, iblock, offsets, |
884 | &blocks_to_boundary); | |
ac27a0ec DK |
885 | |
886 | if (depth == 0) | |
887 | goto out; | |
888 | ||
617ba13b | 889 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
890 | |
891 | /* Simplest case - block found, no allocation needed */ | |
892 | if (!partial) { | |
893 | first_block = le32_to_cpu(chain[depth - 1].key); | |
894 | clear_buffer_new(bh_result); | |
895 | count++; | |
896 | /*map more blocks*/ | |
897 | while (count < maxblocks && count <= blocks_to_boundary) { | |
617ba13b | 898 | ext4_fsblk_t blk; |
ac27a0ec | 899 | |
ac27a0ec DK |
900 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
901 | ||
902 | if (blk == first_block + count) | |
903 | count++; | |
904 | else | |
905 | break; | |
906 | } | |
c278bfec | 907 | goto got_it; |
ac27a0ec DK |
908 | } |
909 | ||
910 | /* Next simple case - plain lookup or failed read of indirect block */ | |
911 | if (!create || err == -EIO) | |
912 | goto cleanup; | |
913 | ||
ac27a0ec DK |
914 | /* |
915 | * Okay, we need to do block allocation. Lazily initialize the block | |
916 | * allocation info here if necessary | |
917 | */ | |
918 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) | |
617ba13b | 919 | ext4_init_block_alloc_info(inode); |
ac27a0ec | 920 | |
fb01bfda | 921 | goal = ext4_find_goal(inode, iblock, partial); |
ac27a0ec DK |
922 | |
923 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
924 | indirect_blks = (chain + depth) - partial - 1; | |
925 | ||
926 | /* | |
927 | * Next look up the indirect map to count the totoal number of | |
928 | * direct blocks to allocate for this branch. | |
929 | */ | |
617ba13b | 930 | count = ext4_blks_to_allocate(partial, indirect_blks, |
ac27a0ec DK |
931 | maxblocks, blocks_to_boundary); |
932 | /* | |
617ba13b | 933 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 934 | */ |
7061eba7 AK |
935 | err = ext4_alloc_branch(handle, inode, iblock, indirect_blks, |
936 | &count, goal, | |
937 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
938 | |
939 | /* | |
617ba13b | 940 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
941 | * on the new chain if there is a failure, but that risks using |
942 | * up transaction credits, especially for bitmaps where the | |
943 | * credits cannot be returned. Can we handle this somehow? We | |
944 | * may need to return -EAGAIN upwards in the worst case. --sct | |
945 | */ | |
946 | if (!err) | |
617ba13b | 947 | err = ext4_splice_branch(handle, inode, iblock, |
ac27a0ec DK |
948 | partial, indirect_blks, count); |
949 | /* | |
0e855ac8 | 950 | * i_disksize growing is protected by i_data_sem. Don't forget to |
ac27a0ec | 951 | * protect it if you're about to implement concurrent |
617ba13b | 952 | * ext4_get_block() -bzzz |
ac27a0ec | 953 | */ |
61628a3f MC |
954 | if (!err && extend_disksize) { |
955 | disksize = ((loff_t) iblock + count) << inode->i_blkbits; | |
956 | if (disksize > i_size_read(inode)) | |
957 | disksize = i_size_read(inode); | |
958 | if (disksize > ei->i_disksize) | |
959 | ei->i_disksize = disksize; | |
960 | } | |
ac27a0ec DK |
961 | if (err) |
962 | goto cleanup; | |
963 | ||
964 | set_buffer_new(bh_result); | |
965 | got_it: | |
966 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
967 | if (count > blocks_to_boundary) | |
968 | set_buffer_boundary(bh_result); | |
969 | err = count; | |
970 | /* Clean up and exit */ | |
971 | partial = chain + depth - 1; /* the whole chain */ | |
972 | cleanup: | |
973 | while (partial > chain) { | |
974 | BUFFER_TRACE(partial->bh, "call brelse"); | |
975 | brelse(partial->bh); | |
976 | partial--; | |
977 | } | |
978 | BUFFER_TRACE(bh_result, "returned"); | |
979 | out: | |
980 | return err; | |
981 | } | |
982 | ||
12219aea AK |
983 | /* |
984 | * Calculate the number of metadata blocks need to reserve | |
985 | * to allocate @blocks for non extent file based file | |
986 | */ | |
987 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks) | |
988 | { | |
989 | int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
990 | int ind_blks, dind_blks, tind_blks; | |
991 | ||
992 | /* number of new indirect blocks needed */ | |
993 | ind_blks = (blocks + icap - 1) / icap; | |
994 | ||
995 | dind_blks = (ind_blks + icap - 1) / icap; | |
996 | ||
997 | tind_blks = 1; | |
998 | ||
999 | return ind_blks + dind_blks + tind_blks; | |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * Calculate the number of metadata blocks need to reserve | |
1004 | * to allocate given number of blocks | |
1005 | */ | |
1006 | static int ext4_calc_metadata_amount(struct inode *inode, int blocks) | |
1007 | { | |
1008 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) | |
1009 | return ext4_ext_calc_metadata_amount(inode, blocks); | |
1010 | ||
1011 | return ext4_indirect_calc_metadata_amount(inode, blocks); | |
1012 | } | |
1013 | ||
1014 | static void ext4_da_update_reserve_space(struct inode *inode, int used) | |
1015 | { | |
1016 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1017 | int total, mdb, mdb_free; | |
1018 | ||
1019 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1020 | /* recalculate the number of metablocks still need to be reserved */ | |
1021 | total = EXT4_I(inode)->i_reserved_data_blocks - used; | |
1022 | mdb = ext4_calc_metadata_amount(inode, total); | |
1023 | ||
1024 | /* figure out how many metablocks to release */ | |
1025 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1026 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1027 | ||
1028 | /* Account for allocated meta_blocks */ | |
1029 | mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks; | |
1030 | ||
1031 | /* update fs free blocks counter for truncate case */ | |
1032 | percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free); | |
1033 | ||
1034 | /* update per-inode reservations */ | |
1035 | BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks); | |
1036 | EXT4_I(inode)->i_reserved_data_blocks -= used; | |
1037 | ||
1038 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1039 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
1040 | EXT4_I(inode)->i_allocated_meta_blocks = 0; | |
1041 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1042 | } | |
1043 | ||
7fb5409d JK |
1044 | /* Maximum number of blocks we map for direct IO at once. */ |
1045 | #define DIO_MAX_BLOCKS 4096 | |
1046 | /* | |
1047 | * Number of credits we need for writing DIO_MAX_BLOCKS: | |
1048 | * We need sb + group descriptor + bitmap + inode -> 4 | |
1049 | * For B blocks with A block pointers per block we need: | |
1050 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). | |
1051 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. | |
1052 | */ | |
1053 | #define DIO_CREDITS 25 | |
ac27a0ec | 1054 | |
f5ab0d1f MC |
1055 | |
1056 | /* | |
2b2d6d01 TT |
1057 | * The ext4_get_blocks_wrap() function try to look up the requested blocks, |
1058 | * and returns if the blocks are already mapped. | |
f5ab0d1f | 1059 | * |
f5ab0d1f MC |
1060 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1061 | * and store the allocated blocks in the result buffer head and mark it | |
1062 | * mapped. | |
1063 | * | |
1064 | * If file type is extents based, it will call ext4_ext_get_blocks(), | |
1065 | * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping | |
1066 | * based files | |
1067 | * | |
1068 | * On success, it returns the number of blocks being mapped or allocate. | |
1069 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1070 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1071 | * the buffer head is mapped. | |
1072 | * | |
1073 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1074 | * that casem, buffer head is unmapped | |
1075 | * | |
1076 | * It returns the error in case of allocation failure. | |
1077 | */ | |
0e855ac8 AK |
1078 | int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block, |
1079 | unsigned long max_blocks, struct buffer_head *bh, | |
d2a17637 | 1080 | int create, int extend_disksize, int flag) |
0e855ac8 AK |
1081 | { |
1082 | int retval; | |
f5ab0d1f MC |
1083 | |
1084 | clear_buffer_mapped(bh); | |
1085 | ||
4df3d265 AK |
1086 | /* |
1087 | * Try to see if we can get the block without requesting | |
1088 | * for new file system block. | |
1089 | */ | |
1090 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1091 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { | |
1092 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
1093 | bh, 0, 0); | |
0e855ac8 | 1094 | } else { |
4df3d265 AK |
1095 | retval = ext4_get_blocks_handle(handle, |
1096 | inode, block, max_blocks, bh, 0, 0); | |
0e855ac8 | 1097 | } |
4df3d265 | 1098 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f MC |
1099 | |
1100 | /* If it is only a block(s) look up */ | |
1101 | if (!create) | |
1102 | return retval; | |
1103 | ||
1104 | /* | |
1105 | * Returns if the blocks have already allocated | |
1106 | * | |
1107 | * Note that if blocks have been preallocated | |
1108 | * ext4_ext_get_block() returns th create = 0 | |
1109 | * with buffer head unmapped. | |
1110 | */ | |
1111 | if (retval > 0 && buffer_mapped(bh)) | |
4df3d265 AK |
1112 | return retval; |
1113 | ||
1114 | /* | |
f5ab0d1f MC |
1115 | * New blocks allocate and/or writing to uninitialized extent |
1116 | * will possibly result in updating i_data, so we take | |
1117 | * the write lock of i_data_sem, and call get_blocks() | |
1118 | * with create == 1 flag. | |
4df3d265 AK |
1119 | */ |
1120 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1121 | |
1122 | /* | |
1123 | * if the caller is from delayed allocation writeout path | |
1124 | * we have already reserved fs blocks for allocation | |
1125 | * let the underlying get_block() function know to | |
1126 | * avoid double accounting | |
1127 | */ | |
1128 | if (flag) | |
1129 | EXT4_I(inode)->i_delalloc_reserved_flag = 1; | |
4df3d265 AK |
1130 | /* |
1131 | * We need to check for EXT4 here because migrate | |
1132 | * could have changed the inode type in between | |
1133 | */ | |
0e855ac8 AK |
1134 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
1135 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
1136 | bh, create, extend_disksize); | |
1137 | } else { | |
1138 | retval = ext4_get_blocks_handle(handle, inode, block, | |
1139 | max_blocks, bh, create, extend_disksize); | |
267e4db9 AK |
1140 | |
1141 | if (retval > 0 && buffer_new(bh)) { | |
1142 | /* | |
1143 | * We allocated new blocks which will result in | |
1144 | * i_data's format changing. Force the migrate | |
1145 | * to fail by clearing migrate flags | |
1146 | */ | |
1147 | EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags & | |
1148 | ~EXT4_EXT_MIGRATE; | |
1149 | } | |
0e855ac8 | 1150 | } |
d2a17637 MC |
1151 | |
1152 | if (flag) { | |
1153 | EXT4_I(inode)->i_delalloc_reserved_flag = 0; | |
1154 | /* | |
1155 | * Update reserved blocks/metadata blocks | |
1156 | * after successful block allocation | |
1157 | * which were deferred till now | |
1158 | */ | |
1159 | if ((retval > 0) && buffer_delay(bh)) | |
12219aea | 1160 | ext4_da_update_reserve_space(inode, retval); |
d2a17637 MC |
1161 | } |
1162 | ||
4df3d265 | 1163 | up_write((&EXT4_I(inode)->i_data_sem)); |
0e855ac8 AK |
1164 | return retval; |
1165 | } | |
1166 | ||
617ba13b | 1167 | static int ext4_get_block(struct inode *inode, sector_t iblock, |
ac27a0ec DK |
1168 | struct buffer_head *bh_result, int create) |
1169 | { | |
3e4fdaf8 | 1170 | handle_t *handle = ext4_journal_current_handle(); |
7fb5409d | 1171 | int ret = 0, started = 0; |
ac27a0ec DK |
1172 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
1173 | ||
7fb5409d JK |
1174 | if (create && !handle) { |
1175 | /* Direct IO write... */ | |
1176 | if (max_blocks > DIO_MAX_BLOCKS) | |
1177 | max_blocks = DIO_MAX_BLOCKS; | |
1178 | handle = ext4_journal_start(inode, DIO_CREDITS + | |
1179 | 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb)); | |
1180 | if (IS_ERR(handle)) { | |
ac27a0ec | 1181 | ret = PTR_ERR(handle); |
7fb5409d | 1182 | goto out; |
ac27a0ec | 1183 | } |
7fb5409d | 1184 | started = 1; |
ac27a0ec DK |
1185 | } |
1186 | ||
7fb5409d | 1187 | ret = ext4_get_blocks_wrap(handle, inode, iblock, |
d2a17637 | 1188 | max_blocks, bh_result, create, 0, 0); |
7fb5409d JK |
1189 | if (ret > 0) { |
1190 | bh_result->b_size = (ret << inode->i_blkbits); | |
1191 | ret = 0; | |
ac27a0ec | 1192 | } |
7fb5409d JK |
1193 | if (started) |
1194 | ext4_journal_stop(handle); | |
1195 | out: | |
ac27a0ec DK |
1196 | return ret; |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * `handle' can be NULL if create is zero | |
1201 | */ | |
617ba13b | 1202 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1203 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec DK |
1204 | { |
1205 | struct buffer_head dummy; | |
1206 | int fatal = 0, err; | |
1207 | ||
1208 | J_ASSERT(handle != NULL || create == 0); | |
1209 | ||
1210 | dummy.b_state = 0; | |
1211 | dummy.b_blocknr = -1000; | |
1212 | buffer_trace_init(&dummy.b_history); | |
a86c6181 | 1213 | err = ext4_get_blocks_wrap(handle, inode, block, 1, |
d2a17637 | 1214 | &dummy, create, 1, 0); |
ac27a0ec | 1215 | /* |
617ba13b | 1216 | * ext4_get_blocks_handle() returns number of blocks |
ac27a0ec DK |
1217 | * mapped. 0 in case of a HOLE. |
1218 | */ | |
1219 | if (err > 0) { | |
1220 | if (err > 1) | |
1221 | WARN_ON(1); | |
1222 | err = 0; | |
1223 | } | |
1224 | *errp = err; | |
1225 | if (!err && buffer_mapped(&dummy)) { | |
1226 | struct buffer_head *bh; | |
1227 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
1228 | if (!bh) { | |
1229 | *errp = -EIO; | |
1230 | goto err; | |
1231 | } | |
1232 | if (buffer_new(&dummy)) { | |
1233 | J_ASSERT(create != 0); | |
ac39849d | 1234 | J_ASSERT(handle != NULL); |
ac27a0ec DK |
1235 | |
1236 | /* | |
1237 | * Now that we do not always journal data, we should | |
1238 | * keep in mind whether this should always journal the | |
1239 | * new buffer as metadata. For now, regular file | |
617ba13b | 1240 | * writes use ext4_get_block instead, so it's not a |
ac27a0ec DK |
1241 | * problem. |
1242 | */ | |
1243 | lock_buffer(bh); | |
1244 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 1245 | fatal = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
1246 | if (!fatal && !buffer_uptodate(bh)) { |
1247 | memset(bh->b_data,0,inode->i_sb->s_blocksize); | |
1248 | set_buffer_uptodate(bh); | |
1249 | } | |
1250 | unlock_buffer(bh); | |
617ba13b MC |
1251 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
1252 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
1253 | if (!fatal) |
1254 | fatal = err; | |
1255 | } else { | |
1256 | BUFFER_TRACE(bh, "not a new buffer"); | |
1257 | } | |
1258 | if (fatal) { | |
1259 | *errp = fatal; | |
1260 | brelse(bh); | |
1261 | bh = NULL; | |
1262 | } | |
1263 | return bh; | |
1264 | } | |
1265 | err: | |
1266 | return NULL; | |
1267 | } | |
1268 | ||
617ba13b | 1269 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1270 | ext4_lblk_t block, int create, int *err) |
ac27a0ec DK |
1271 | { |
1272 | struct buffer_head * bh; | |
1273 | ||
617ba13b | 1274 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1275 | if (!bh) |
1276 | return bh; | |
1277 | if (buffer_uptodate(bh)) | |
1278 | return bh; | |
1279 | ll_rw_block(READ_META, 1, &bh); | |
1280 | wait_on_buffer(bh); | |
1281 | if (buffer_uptodate(bh)) | |
1282 | return bh; | |
1283 | put_bh(bh); | |
1284 | *err = -EIO; | |
1285 | return NULL; | |
1286 | } | |
1287 | ||
1288 | static int walk_page_buffers( handle_t *handle, | |
1289 | struct buffer_head *head, | |
1290 | unsigned from, | |
1291 | unsigned to, | |
1292 | int *partial, | |
1293 | int (*fn)( handle_t *handle, | |
1294 | struct buffer_head *bh)) | |
1295 | { | |
1296 | struct buffer_head *bh; | |
1297 | unsigned block_start, block_end; | |
1298 | unsigned blocksize = head->b_size; | |
1299 | int err, ret = 0; | |
1300 | struct buffer_head *next; | |
1301 | ||
1302 | for ( bh = head, block_start = 0; | |
1303 | ret == 0 && (bh != head || !block_start); | |
1304 | block_start = block_end, bh = next) | |
1305 | { | |
1306 | next = bh->b_this_page; | |
1307 | block_end = block_start + blocksize; | |
1308 | if (block_end <= from || block_start >= to) { | |
1309 | if (partial && !buffer_uptodate(bh)) | |
1310 | *partial = 1; | |
1311 | continue; | |
1312 | } | |
1313 | err = (*fn)(handle, bh); | |
1314 | if (!ret) | |
1315 | ret = err; | |
1316 | } | |
1317 | return ret; | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * To preserve ordering, it is essential that the hole instantiation and | |
1322 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1323 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1324 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1325 | * prepare_write() is the right place. |
1326 | * | |
617ba13b MC |
1327 | * Also, this function can nest inside ext4_writepage() -> |
1328 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1329 | * has generated enough buffer credits to do the whole page. So we won't |
1330 | * block on the journal in that case, which is good, because the caller may | |
1331 | * be PF_MEMALLOC. | |
1332 | * | |
617ba13b | 1333 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1334 | * quota file writes. If we were to commit the transaction while thus |
1335 | * reentered, there can be a deadlock - we would be holding a quota | |
1336 | * lock, and the commit would never complete if another thread had a | |
1337 | * transaction open and was blocking on the quota lock - a ranking | |
1338 | * violation. | |
1339 | * | |
dab291af | 1340 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1341 | * will _not_ run commit under these circumstances because handle->h_ref |
1342 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1343 | * write. | |
1344 | */ | |
1345 | static int do_journal_get_write_access(handle_t *handle, | |
1346 | struct buffer_head *bh) | |
1347 | { | |
1348 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1349 | return 0; | |
617ba13b | 1350 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1351 | } |
1352 | ||
bfc1af65 NP |
1353 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
1354 | loff_t pos, unsigned len, unsigned flags, | |
1355 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1356 | { |
bfc1af65 | 1357 | struct inode *inode = mapping->host; |
7479d2b9 | 1358 | int ret, needed_blocks = ext4_writepage_trans_blocks(inode); |
ac27a0ec DK |
1359 | handle_t *handle; |
1360 | int retries = 0; | |
bfc1af65 NP |
1361 | struct page *page; |
1362 | pgoff_t index; | |
1363 | unsigned from, to; | |
1364 | ||
1365 | index = pos >> PAGE_CACHE_SHIFT; | |
1366 | from = pos & (PAGE_CACHE_SIZE - 1); | |
1367 | to = from + len; | |
ac27a0ec DK |
1368 | |
1369 | retry: | |
bfc1af65 NP |
1370 | handle = ext4_journal_start(inode, needed_blocks); |
1371 | if (IS_ERR(handle)) { | |
bfc1af65 NP |
1372 | ret = PTR_ERR(handle); |
1373 | goto out; | |
7479d2b9 | 1374 | } |
ac27a0ec | 1375 | |
cf108bca JK |
1376 | page = __grab_cache_page(mapping, index); |
1377 | if (!page) { | |
1378 | ext4_journal_stop(handle); | |
1379 | ret = -ENOMEM; | |
1380 | goto out; | |
1381 | } | |
1382 | *pagep = page; | |
1383 | ||
bfc1af65 NP |
1384 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
1385 | ext4_get_block); | |
1386 | ||
1387 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1388 | ret = walk_page_buffers(handle, page_buffers(page), |
1389 | from, to, NULL, do_journal_get_write_access); | |
1390 | } | |
bfc1af65 NP |
1391 | |
1392 | if (ret) { | |
bfc1af65 | 1393 | unlock_page(page); |
cf108bca | 1394 | ext4_journal_stop(handle); |
bfc1af65 NP |
1395 | page_cache_release(page); |
1396 | } | |
1397 | ||
617ba13b | 1398 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1399 | goto retry; |
7479d2b9 | 1400 | out: |
ac27a0ec DK |
1401 | return ret; |
1402 | } | |
1403 | ||
bfc1af65 NP |
1404 | /* For write_end() in data=journal mode */ |
1405 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1406 | { |
1407 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1408 | return 0; | |
1409 | set_buffer_uptodate(bh); | |
617ba13b | 1410 | return ext4_journal_dirty_metadata(handle, bh); |
ac27a0ec DK |
1411 | } |
1412 | ||
1413 | /* | |
1414 | * We need to pick up the new inode size which generic_commit_write gave us | |
1415 | * `file' can be NULL - eg, when called from page_symlink(). | |
1416 | * | |
617ba13b | 1417 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1418 | * buffers are managed internally. |
1419 | */ | |
bfc1af65 NP |
1420 | static int ext4_ordered_write_end(struct file *file, |
1421 | struct address_space *mapping, | |
1422 | loff_t pos, unsigned len, unsigned copied, | |
1423 | struct page *page, void *fsdata) | |
ac27a0ec | 1424 | { |
617ba13b | 1425 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1426 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1427 | int ret = 0, ret2; |
1428 | ||
678aaf48 | 1429 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1430 | |
1431 | if (ret == 0) { | |
1432 | /* | |
bfc1af65 | 1433 | * generic_write_end() will run mark_inode_dirty() if i_size |
ac27a0ec DK |
1434 | * changes. So let's piggyback the i_disksize mark_inode_dirty |
1435 | * into that. | |
1436 | */ | |
1437 | loff_t new_i_size; | |
1438 | ||
bfc1af65 | 1439 | new_i_size = pos + copied; |
617ba13b MC |
1440 | if (new_i_size > EXT4_I(inode)->i_disksize) |
1441 | EXT4_I(inode)->i_disksize = new_i_size; | |
cf108bca | 1442 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1443 | page, fsdata); |
f8a87d89 RK |
1444 | copied = ret2; |
1445 | if (ret2 < 0) | |
1446 | ret = ret2; | |
ac27a0ec | 1447 | } |
617ba13b | 1448 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1449 | if (!ret) |
1450 | ret = ret2; | |
bfc1af65 NP |
1451 | |
1452 | return ret ? ret : copied; | |
ac27a0ec DK |
1453 | } |
1454 | ||
bfc1af65 NP |
1455 | static int ext4_writeback_write_end(struct file *file, |
1456 | struct address_space *mapping, | |
1457 | loff_t pos, unsigned len, unsigned copied, | |
1458 | struct page *page, void *fsdata) | |
ac27a0ec | 1459 | { |
617ba13b | 1460 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1461 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1462 | int ret = 0, ret2; |
1463 | loff_t new_i_size; | |
1464 | ||
bfc1af65 | 1465 | new_i_size = pos + copied; |
617ba13b MC |
1466 | if (new_i_size > EXT4_I(inode)->i_disksize) |
1467 | EXT4_I(inode)->i_disksize = new_i_size; | |
ac27a0ec | 1468 | |
cf108bca | 1469 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1470 | page, fsdata); |
f8a87d89 RK |
1471 | copied = ret2; |
1472 | if (ret2 < 0) | |
1473 | ret = ret2; | |
ac27a0ec | 1474 | |
617ba13b | 1475 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1476 | if (!ret) |
1477 | ret = ret2; | |
bfc1af65 NP |
1478 | |
1479 | return ret ? ret : copied; | |
ac27a0ec DK |
1480 | } |
1481 | ||
bfc1af65 NP |
1482 | static int ext4_journalled_write_end(struct file *file, |
1483 | struct address_space *mapping, | |
1484 | loff_t pos, unsigned len, unsigned copied, | |
1485 | struct page *page, void *fsdata) | |
ac27a0ec | 1486 | { |
617ba13b | 1487 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1488 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1489 | int ret = 0, ret2; |
1490 | int partial = 0; | |
bfc1af65 | 1491 | unsigned from, to; |
ac27a0ec | 1492 | |
bfc1af65 NP |
1493 | from = pos & (PAGE_CACHE_SIZE - 1); |
1494 | to = from + len; | |
1495 | ||
1496 | if (copied < len) { | |
1497 | if (!PageUptodate(page)) | |
1498 | copied = 0; | |
1499 | page_zero_new_buffers(page, from+copied, to); | |
1500 | } | |
ac27a0ec DK |
1501 | |
1502 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1503 | to, &partial, write_end_fn); |
ac27a0ec DK |
1504 | if (!partial) |
1505 | SetPageUptodate(page); | |
bfc1af65 NP |
1506 | if (pos+copied > inode->i_size) |
1507 | i_size_write(inode, pos+copied); | |
617ba13b MC |
1508 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
1509 | if (inode->i_size > EXT4_I(inode)->i_disksize) { | |
1510 | EXT4_I(inode)->i_disksize = inode->i_size; | |
1511 | ret2 = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
1512 | if (!ret) |
1513 | ret = ret2; | |
1514 | } | |
bfc1af65 | 1515 | |
cf108bca | 1516 | unlock_page(page); |
617ba13b | 1517 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1518 | if (!ret) |
1519 | ret = ret2; | |
bfc1af65 NP |
1520 | page_cache_release(page); |
1521 | ||
1522 | return ret ? ret : copied; | |
ac27a0ec | 1523 | } |
d2a17637 MC |
1524 | |
1525 | static int ext4_da_reserve_space(struct inode *inode, int nrblocks) | |
1526 | { | |
1527 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1528 | unsigned long md_needed, mdblocks, total = 0; | |
1529 | ||
1530 | /* | |
1531 | * recalculate the amount of metadata blocks to reserve | |
1532 | * in order to allocate nrblocks | |
1533 | * worse case is one extent per block | |
1534 | */ | |
1535 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1536 | total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks; | |
1537 | mdblocks = ext4_calc_metadata_amount(inode, total); | |
1538 | BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks); | |
1539 | ||
1540 | md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks; | |
1541 | total = md_needed + nrblocks; | |
1542 | ||
1543 | if (ext4_has_free_blocks(sbi, total) < total) { | |
1544 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1545 | return -ENOSPC; | |
1546 | } | |
d2a17637 MC |
1547 | /* reduce fs free blocks counter */ |
1548 | percpu_counter_sub(&sbi->s_freeblocks_counter, total); | |
1549 | ||
1550 | EXT4_I(inode)->i_reserved_data_blocks += nrblocks; | |
1551 | EXT4_I(inode)->i_reserved_meta_blocks = mdblocks; | |
1552 | ||
1553 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1554 | return 0; /* success */ | |
1555 | } | |
1556 | ||
12219aea | 1557 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1558 | { |
1559 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1560 | int total, mdb, mdb_free, release; | |
1561 | ||
1562 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1563 | /* recalculate the number of metablocks still need to be reserved */ | |
12219aea | 1564 | total = EXT4_I(inode)->i_reserved_data_blocks - to_free; |
d2a17637 MC |
1565 | mdb = ext4_calc_metadata_amount(inode, total); |
1566 | ||
1567 | /* figure out how many metablocks to release */ | |
1568 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1569 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1570 | ||
d2a17637 MC |
1571 | release = to_free + mdb_free; |
1572 | ||
1573 | /* update fs free blocks counter for truncate case */ | |
1574 | percpu_counter_add(&sbi->s_freeblocks_counter, release); | |
1575 | ||
1576 | /* update per-inode reservations */ | |
12219aea AK |
1577 | BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks); |
1578 | EXT4_I(inode)->i_reserved_data_blocks -= to_free; | |
d2a17637 MC |
1579 | |
1580 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1581 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
d2a17637 MC |
1582 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
1583 | } | |
1584 | ||
1585 | static void ext4_da_page_release_reservation(struct page *page, | |
1586 | unsigned long offset) | |
1587 | { | |
1588 | int to_release = 0; | |
1589 | struct buffer_head *head, *bh; | |
1590 | unsigned int curr_off = 0; | |
1591 | ||
1592 | head = page_buffers(page); | |
1593 | bh = head; | |
1594 | do { | |
1595 | unsigned int next_off = curr_off + bh->b_size; | |
1596 | ||
1597 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1598 | to_release++; | |
1599 | clear_buffer_delay(bh); | |
1600 | } | |
1601 | curr_off = next_off; | |
1602 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 1603 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 1604 | } |
ac27a0ec | 1605 | |
64769240 AT |
1606 | /* |
1607 | * Delayed allocation stuff | |
1608 | */ | |
1609 | ||
1610 | struct mpage_da_data { | |
1611 | struct inode *inode; | |
1612 | struct buffer_head lbh; /* extent of blocks */ | |
1613 | unsigned long first_page, next_page; /* extent of pages */ | |
1614 | get_block_t *get_block; | |
1615 | struct writeback_control *wbc; | |
1616 | }; | |
1617 | ||
1618 | /* | |
1619 | * mpage_da_submit_io - walks through extent of pages and try to write | |
1620 | * them with __mpage_writepage() | |
1621 | * | |
1622 | * @mpd->inode: inode | |
1623 | * @mpd->first_page: first page of the extent | |
1624 | * @mpd->next_page: page after the last page of the extent | |
1625 | * @mpd->get_block: the filesystem's block mapper function | |
1626 | * | |
1627 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1628 | * to be allocated. this may be wrong if allocation failed. | |
1629 | * | |
1630 | * As pages are already locked by write_cache_pages(), we can't use it | |
1631 | */ | |
1632 | static int mpage_da_submit_io(struct mpage_da_data *mpd) | |
1633 | { | |
1634 | struct address_space *mapping = mpd->inode->i_mapping; | |
1635 | struct mpage_data mpd_pp = { | |
1636 | .bio = NULL, | |
1637 | .last_block_in_bio = 0, | |
1638 | .get_block = mpd->get_block, | |
1639 | .use_writepage = 1, | |
1640 | }; | |
1641 | int ret = 0, err, nr_pages, i; | |
1642 | unsigned long index, end; | |
1643 | struct pagevec pvec; | |
1644 | ||
1645 | BUG_ON(mpd->next_page <= mpd->first_page); | |
1646 | ||
1647 | pagevec_init(&pvec, 0); | |
1648 | index = mpd->first_page; | |
1649 | end = mpd->next_page - 1; | |
1650 | ||
1651 | while (index <= end) { | |
1652 | /* XXX: optimize tail */ | |
1653 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1654 | if (nr_pages == 0) | |
1655 | break; | |
1656 | for (i = 0; i < nr_pages; i++) { | |
1657 | struct page *page = pvec.pages[i]; | |
1658 | ||
1659 | index = page->index; | |
1660 | if (index > end) | |
1661 | break; | |
1662 | index++; | |
1663 | ||
1664 | err = __mpage_writepage(page, mpd->wbc, &mpd_pp); | |
1665 | ||
1666 | /* | |
1667 | * In error case, we have to continue because | |
1668 | * remaining pages are still locked | |
1669 | * XXX: unlock and re-dirty them? | |
1670 | */ | |
1671 | if (ret == 0) | |
1672 | ret = err; | |
1673 | } | |
1674 | pagevec_release(&pvec); | |
1675 | } | |
1676 | if (mpd_pp.bio) | |
1677 | mpage_bio_submit(WRITE, mpd_pp.bio); | |
1678 | ||
1679 | return ret; | |
1680 | } | |
1681 | ||
1682 | /* | |
1683 | * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers | |
1684 | * | |
1685 | * @mpd->inode - inode to walk through | |
1686 | * @exbh->b_blocknr - first block on a disk | |
1687 | * @exbh->b_size - amount of space in bytes | |
1688 | * @logical - first logical block to start assignment with | |
1689 | * | |
1690 | * the function goes through all passed space and put actual disk | |
1691 | * block numbers into buffer heads, dropping BH_Delay | |
1692 | */ | |
1693 | static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical, | |
1694 | struct buffer_head *exbh) | |
1695 | { | |
1696 | struct inode *inode = mpd->inode; | |
1697 | struct address_space *mapping = inode->i_mapping; | |
1698 | int blocks = exbh->b_size >> inode->i_blkbits; | |
1699 | sector_t pblock = exbh->b_blocknr, cur_logical; | |
1700 | struct buffer_head *head, *bh; | |
1701 | unsigned long index, end; | |
1702 | struct pagevec pvec; | |
1703 | int nr_pages, i; | |
1704 | ||
1705 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1706 | end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1707 | cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1708 | ||
1709 | pagevec_init(&pvec, 0); | |
1710 | ||
1711 | while (index <= end) { | |
1712 | /* XXX: optimize tail */ | |
1713 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1714 | if (nr_pages == 0) | |
1715 | break; | |
1716 | for (i = 0; i < nr_pages; i++) { | |
1717 | struct page *page = pvec.pages[i]; | |
1718 | ||
1719 | index = page->index; | |
1720 | if (index > end) | |
1721 | break; | |
1722 | index++; | |
1723 | ||
1724 | BUG_ON(!PageLocked(page)); | |
1725 | BUG_ON(PageWriteback(page)); | |
1726 | BUG_ON(!page_has_buffers(page)); | |
1727 | ||
1728 | bh = page_buffers(page); | |
1729 | head = bh; | |
1730 | ||
1731 | /* skip blocks out of the range */ | |
1732 | do { | |
1733 | if (cur_logical >= logical) | |
1734 | break; | |
1735 | cur_logical++; | |
1736 | } while ((bh = bh->b_this_page) != head); | |
1737 | ||
1738 | do { | |
1739 | if (cur_logical >= logical + blocks) | |
1740 | break; | |
64769240 AT |
1741 | if (buffer_delay(bh)) { |
1742 | bh->b_blocknr = pblock; | |
1743 | clear_buffer_delay(bh); | |
bf068ee2 AK |
1744 | bh->b_bdev = inode->i_sb->s_bdev; |
1745 | } else if (buffer_unwritten(bh)) { | |
1746 | bh->b_blocknr = pblock; | |
1747 | clear_buffer_unwritten(bh); | |
1748 | set_buffer_mapped(bh); | |
1749 | set_buffer_new(bh); | |
1750 | bh->b_bdev = inode->i_sb->s_bdev; | |
61628a3f | 1751 | } else if (buffer_mapped(bh)) |
64769240 | 1752 | BUG_ON(bh->b_blocknr != pblock); |
64769240 AT |
1753 | |
1754 | cur_logical++; | |
1755 | pblock++; | |
1756 | } while ((bh = bh->b_this_page) != head); | |
1757 | } | |
1758 | pagevec_release(&pvec); | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | ||
1763 | /* | |
1764 | * __unmap_underlying_blocks - just a helper function to unmap | |
1765 | * set of blocks described by @bh | |
1766 | */ | |
1767 | static inline void __unmap_underlying_blocks(struct inode *inode, | |
1768 | struct buffer_head *bh) | |
1769 | { | |
1770 | struct block_device *bdev = inode->i_sb->s_bdev; | |
1771 | int blocks, i; | |
1772 | ||
1773 | blocks = bh->b_size >> inode->i_blkbits; | |
1774 | for (i = 0; i < blocks; i++) | |
1775 | unmap_underlying_metadata(bdev, bh->b_blocknr + i); | |
1776 | } | |
1777 | ||
1778 | /* | |
1779 | * mpage_da_map_blocks - go through given space | |
1780 | * | |
1781 | * @mpd->lbh - bh describing space | |
1782 | * @mpd->get_block - the filesystem's block mapper function | |
1783 | * | |
1784 | * The function skips space we know is already mapped to disk blocks. | |
1785 | * | |
1786 | * The function ignores errors ->get_block() returns, thus real | |
1787 | * error handling is postponed to __mpage_writepage() | |
1788 | */ | |
1789 | static void mpage_da_map_blocks(struct mpage_da_data *mpd) | |
1790 | { | |
1791 | struct buffer_head *lbh = &mpd->lbh; | |
1792 | int err = 0, remain = lbh->b_size; | |
1793 | sector_t next = lbh->b_blocknr; | |
1794 | struct buffer_head new; | |
1795 | ||
1796 | /* | |
1797 | * We consider only non-mapped and non-allocated blocks | |
1798 | */ | |
1799 | if (buffer_mapped(lbh) && !buffer_delay(lbh)) | |
1800 | return; | |
1801 | ||
1802 | while (remain) { | |
1803 | new.b_state = lbh->b_state; | |
1804 | new.b_blocknr = 0; | |
1805 | new.b_size = remain; | |
1806 | err = mpd->get_block(mpd->inode, next, &new, 1); | |
1807 | if (err) { | |
1808 | /* | |
1809 | * Rather than implement own error handling | |
1810 | * here, we just leave remaining blocks | |
1811 | * unallocated and try again with ->writepage() | |
1812 | */ | |
1813 | break; | |
1814 | } | |
1815 | BUG_ON(new.b_size == 0); | |
1816 | ||
1817 | if (buffer_new(&new)) | |
1818 | __unmap_underlying_blocks(mpd->inode, &new); | |
1819 | ||
1820 | /* | |
1821 | * If blocks are delayed marked, we need to | |
1822 | * put actual blocknr and drop delayed bit | |
1823 | */ | |
bf068ee2 | 1824 | if (buffer_delay(lbh) || buffer_unwritten(lbh)) |
64769240 AT |
1825 | mpage_put_bnr_to_bhs(mpd, next, &new); |
1826 | ||
61628a3f MC |
1827 | /* go for the remaining blocks */ |
1828 | next += new.b_size >> mpd->inode->i_blkbits; | |
1829 | remain -= new.b_size; | |
1830 | } | |
64769240 AT |
1831 | } |
1832 | ||
bf068ee2 AK |
1833 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1834 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1835 | |
1836 | /* | |
1837 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1838 | * | |
1839 | * @mpd->lbh - extent of blocks | |
1840 | * @logical - logical number of the block in the file | |
1841 | * @bh - bh of the block (used to access block's state) | |
1842 | * | |
1843 | * the function is used to collect contig. blocks in same state | |
1844 | */ | |
1845 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
1846 | sector_t logical, struct buffer_head *bh) | |
1847 | { | |
1848 | struct buffer_head *lbh = &mpd->lbh; | |
1849 | sector_t next; | |
1850 | ||
1851 | next = lbh->b_blocknr + (lbh->b_size >> mpd->inode->i_blkbits); | |
1852 | ||
1853 | /* | |
1854 | * First block in the extent | |
1855 | */ | |
1856 | if (lbh->b_size == 0) { | |
1857 | lbh->b_blocknr = logical; | |
1858 | lbh->b_size = bh->b_size; | |
1859 | lbh->b_state = bh->b_state & BH_FLAGS; | |
1860 | return; | |
1861 | } | |
1862 | ||
1863 | /* | |
1864 | * Can we merge the block to our big extent? | |
1865 | */ | |
1866 | if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) { | |
1867 | lbh->b_size += bh->b_size; | |
1868 | return; | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * We couldn't merge the block to our extent, so we | |
1873 | * need to flush current extent and start new one | |
1874 | */ | |
1875 | mpage_da_map_blocks(mpd); | |
1876 | ||
1877 | /* | |
1878 | * Now start a new extent | |
1879 | */ | |
1880 | lbh->b_size = bh->b_size; | |
1881 | lbh->b_state = bh->b_state & BH_FLAGS; | |
1882 | lbh->b_blocknr = logical; | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * __mpage_da_writepage - finds extent of pages and blocks | |
1887 | * | |
1888 | * @page: page to consider | |
1889 | * @wbc: not used, we just follow rules | |
1890 | * @data: context | |
1891 | * | |
1892 | * The function finds extents of pages and scan them for all blocks. | |
1893 | */ | |
1894 | static int __mpage_da_writepage(struct page *page, | |
1895 | struct writeback_control *wbc, void *data) | |
1896 | { | |
1897 | struct mpage_da_data *mpd = data; | |
1898 | struct inode *inode = mpd->inode; | |
1899 | struct buffer_head *bh, *head, fake; | |
1900 | sector_t logical; | |
1901 | ||
1902 | /* | |
1903 | * Can we merge this page to current extent? | |
1904 | */ | |
1905 | if (mpd->next_page != page->index) { | |
1906 | /* | |
1907 | * Nope, we can't. So, we map non-allocated blocks | |
1908 | * and start IO on them using __mpage_writepage() | |
1909 | */ | |
1910 | if (mpd->next_page != mpd->first_page) { | |
1911 | mpage_da_map_blocks(mpd); | |
1912 | mpage_da_submit_io(mpd); | |
1913 | } | |
1914 | ||
1915 | /* | |
1916 | * Start next extent of pages ... | |
1917 | */ | |
1918 | mpd->first_page = page->index; | |
1919 | ||
1920 | /* | |
1921 | * ... and blocks | |
1922 | */ | |
1923 | mpd->lbh.b_size = 0; | |
1924 | mpd->lbh.b_state = 0; | |
1925 | mpd->lbh.b_blocknr = 0; | |
1926 | } | |
1927 | ||
1928 | mpd->next_page = page->index + 1; | |
1929 | logical = (sector_t) page->index << | |
1930 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1931 | ||
1932 | if (!page_has_buffers(page)) { | |
1933 | /* | |
1934 | * There is no attached buffer heads yet (mmap?) | |
1935 | * we treat the page asfull of dirty blocks | |
1936 | */ | |
1937 | bh = &fake; | |
1938 | bh->b_size = PAGE_CACHE_SIZE; | |
1939 | bh->b_state = 0; | |
1940 | set_buffer_dirty(bh); | |
1941 | set_buffer_uptodate(bh); | |
1942 | mpage_add_bh_to_extent(mpd, logical, bh); | |
1943 | } else { | |
1944 | /* | |
1945 | * Page with regular buffer heads, just add all dirty ones | |
1946 | */ | |
1947 | head = page_buffers(page); | |
1948 | bh = head; | |
1949 | do { | |
1950 | BUG_ON(buffer_locked(bh)); | |
1951 | if (buffer_dirty(bh)) | |
1952 | mpage_add_bh_to_extent(mpd, logical, bh); | |
1953 | logical++; | |
1954 | } while ((bh = bh->b_this_page) != head); | |
1955 | } | |
1956 | ||
1957 | return 0; | |
1958 | } | |
1959 | ||
1960 | /* | |
1961 | * mpage_da_writepages - walk the list of dirty pages of the given | |
1962 | * address space, allocates non-allocated blocks, maps newly-allocated | |
1963 | * blocks to existing bhs and issue IO them | |
1964 | * | |
1965 | * @mapping: address space structure to write | |
1966 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
1967 | * @get_block: the filesystem's block mapper function. | |
1968 | * | |
1969 | * This is a library function, which implements the writepages() | |
1970 | * address_space_operation. | |
1971 | * | |
1972 | * In order to avoid duplication of logic that deals with partial pages, | |
1973 | * multiple bio per page, etc, we find non-allocated blocks, allocate | |
1974 | * them with minimal calls to ->get_block() and re-use __mpage_writepage() | |
1975 | * | |
1976 | * It's important that we call __mpage_writepage() only once for each | |
1977 | * involved page, otherwise we'd have to implement more complicated logic | |
1978 | * to deal with pages w/o PG_lock or w/ PG_writeback and so on. | |
1979 | * | |
1980 | * See comments to mpage_writepages() | |
1981 | */ | |
1982 | static int mpage_da_writepages(struct address_space *mapping, | |
1983 | struct writeback_control *wbc, | |
1984 | get_block_t get_block) | |
1985 | { | |
1986 | struct mpage_da_data mpd; | |
1987 | int ret; | |
1988 | ||
1989 | if (!get_block) | |
1990 | return generic_writepages(mapping, wbc); | |
1991 | ||
1992 | mpd.wbc = wbc; | |
1993 | mpd.inode = mapping->host; | |
1994 | mpd.lbh.b_size = 0; | |
1995 | mpd.lbh.b_state = 0; | |
1996 | mpd.lbh.b_blocknr = 0; | |
1997 | mpd.first_page = 0; | |
1998 | mpd.next_page = 0; | |
1999 | mpd.get_block = get_block; | |
2000 | ||
2001 | ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd); | |
2002 | ||
2003 | /* | |
2004 | * Handle last extent of pages | |
2005 | */ | |
2006 | if (mpd.next_page != mpd.first_page) { | |
2007 | mpage_da_map_blocks(&mpd); | |
2008 | mpage_da_submit_io(&mpd); | |
2009 | } | |
2010 | ||
2011 | return ret; | |
2012 | } | |
2013 | ||
2014 | /* | |
2015 | * this is a special callback for ->write_begin() only | |
2016 | * it's intention is to return mapped block or reserve space | |
2017 | */ | |
2018 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2019 | struct buffer_head *bh_result, int create) | |
2020 | { | |
2021 | int ret = 0; | |
2022 | ||
2023 | BUG_ON(create == 0); | |
2024 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); | |
2025 | ||
2026 | /* | |
2027 | * first, we need to know whether the block is allocated already | |
2028 | * preallocated blocks are unmapped but should treated | |
2029 | * the same as allocated blocks. | |
2030 | */ | |
d2a17637 MC |
2031 | ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0); |
2032 | if ((ret == 0) && !buffer_delay(bh_result)) { | |
2033 | /* the block isn't (pre)allocated yet, let's reserve space */ | |
64769240 AT |
2034 | /* |
2035 | * XXX: __block_prepare_write() unmaps passed block, | |
2036 | * is it OK? | |
2037 | */ | |
d2a17637 MC |
2038 | ret = ext4_da_reserve_space(inode, 1); |
2039 | if (ret) | |
2040 | /* not enough space to reserve */ | |
2041 | return ret; | |
2042 | ||
64769240 AT |
2043 | map_bh(bh_result, inode->i_sb, 0); |
2044 | set_buffer_new(bh_result); | |
2045 | set_buffer_delay(bh_result); | |
2046 | } else if (ret > 0) { | |
2047 | bh_result->b_size = (ret << inode->i_blkbits); | |
2048 | ret = 0; | |
2049 | } | |
2050 | ||
2051 | return ret; | |
2052 | } | |
d2a17637 | 2053 | #define EXT4_DELALLOC_RSVED 1 |
64769240 AT |
2054 | static int ext4_da_get_block_write(struct inode *inode, sector_t iblock, |
2055 | struct buffer_head *bh_result, int create) | |
2056 | { | |
61628a3f | 2057 | int ret; |
64769240 AT |
2058 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
2059 | loff_t disksize = EXT4_I(inode)->i_disksize; | |
2060 | handle_t *handle = NULL; | |
2061 | ||
61628a3f | 2062 | handle = ext4_journal_current_handle(); |
f0e6c985 AK |
2063 | if (!handle) { |
2064 | ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks, | |
2065 | bh_result, 0, 0, 0); | |
2066 | BUG_ON(!ret); | |
2067 | } else { | |
2068 | ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks, | |
d2a17637 | 2069 | bh_result, create, 0, EXT4_DELALLOC_RSVED); |
f0e6c985 AK |
2070 | } |
2071 | ||
64769240 AT |
2072 | if (ret > 0) { |
2073 | bh_result->b_size = (ret << inode->i_blkbits); | |
2074 | ||
2075 | /* | |
2076 | * Update on-disk size along with block allocation | |
2077 | * we don't use 'extend_disksize' as size may change | |
2078 | * within already allocated block -bzzz | |
2079 | */ | |
2080 | disksize = ((loff_t) iblock + ret) << inode->i_blkbits; | |
2081 | if (disksize > i_size_read(inode)) | |
2082 | disksize = i_size_read(inode); | |
2083 | if (disksize > EXT4_I(inode)->i_disksize) { | |
2084 | /* | |
2085 | * XXX: replace with spinlock if seen contended -bzzz | |
2086 | */ | |
2087 | down_write(&EXT4_I(inode)->i_data_sem); | |
2088 | if (disksize > EXT4_I(inode)->i_disksize) | |
2089 | EXT4_I(inode)->i_disksize = disksize; | |
2090 | up_write(&EXT4_I(inode)->i_data_sem); | |
2091 | ||
2092 | if (EXT4_I(inode)->i_disksize == disksize) { | |
61628a3f MC |
2093 | ret = ext4_mark_inode_dirty(handle, inode); |
2094 | return ret; | |
64769240 AT |
2095 | } |
2096 | } | |
64769240 AT |
2097 | ret = 0; |
2098 | } | |
64769240 AT |
2099 | return ret; |
2100 | } | |
61628a3f MC |
2101 | |
2102 | static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh) | |
2103 | { | |
f0e6c985 AK |
2104 | /* |
2105 | * unmapped buffer is possible for holes. | |
2106 | * delay buffer is possible with delayed allocation | |
2107 | */ | |
2108 | return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh)); | |
2109 | } | |
2110 | ||
2111 | static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock, | |
2112 | struct buffer_head *bh_result, int create) | |
2113 | { | |
2114 | int ret = 0; | |
2115 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
2116 | ||
2117 | /* | |
2118 | * we don't want to do block allocation in writepage | |
2119 | * so call get_block_wrap with create = 0 | |
2120 | */ | |
2121 | ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks, | |
2122 | bh_result, 0, 0, 0); | |
2123 | if (ret > 0) { | |
2124 | bh_result->b_size = (ret << inode->i_blkbits); | |
2125 | ret = 0; | |
2126 | } | |
2127 | return ret; | |
61628a3f MC |
2128 | } |
2129 | ||
61628a3f | 2130 | /* |
f0e6c985 AK |
2131 | * get called vi ext4_da_writepages after taking page lock (have journal handle) |
2132 | * get called via journal_submit_inode_data_buffers (no journal handle) | |
2133 | * get called via shrink_page_list via pdflush (no journal handle) | |
2134 | * or grab_page_cache when doing write_begin (have journal handle) | |
61628a3f | 2135 | */ |
64769240 AT |
2136 | static int ext4_da_writepage(struct page *page, |
2137 | struct writeback_control *wbc) | |
2138 | { | |
64769240 | 2139 | int ret = 0; |
61628a3f MC |
2140 | loff_t size; |
2141 | unsigned long len; | |
61628a3f MC |
2142 | struct buffer_head *page_bufs; |
2143 | struct inode *inode = page->mapping->host; | |
2144 | ||
f0e6c985 AK |
2145 | size = i_size_read(inode); |
2146 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2147 | len = size & ~PAGE_CACHE_MASK; | |
2148 | else | |
2149 | len = PAGE_CACHE_SIZE; | |
64769240 | 2150 | |
f0e6c985 | 2151 | if (page_has_buffers(page)) { |
61628a3f | 2152 | page_bufs = page_buffers(page); |
f0e6c985 AK |
2153 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2154 | ext4_bh_unmapped_or_delay)) { | |
61628a3f | 2155 | /* |
f0e6c985 AK |
2156 | * We don't want to do block allocation |
2157 | * So redirty the page and return | |
cd1aac32 AK |
2158 | * We may reach here when we do a journal commit |
2159 | * via journal_submit_inode_data_buffers. | |
2160 | * If we don't have mapping block we just ignore | |
f0e6c985 AK |
2161 | * them. We can also reach here via shrink_page_list |
2162 | */ | |
2163 | redirty_page_for_writepage(wbc, page); | |
2164 | unlock_page(page); | |
2165 | return 0; | |
2166 | } | |
2167 | } else { | |
2168 | /* | |
2169 | * The test for page_has_buffers() is subtle: | |
2170 | * We know the page is dirty but it lost buffers. That means | |
2171 | * that at some moment in time after write_begin()/write_end() | |
2172 | * has been called all buffers have been clean and thus they | |
2173 | * must have been written at least once. So they are all | |
2174 | * mapped and we can happily proceed with mapping them | |
2175 | * and writing the page. | |
2176 | * | |
2177 | * Try to initialize the buffer_heads and check whether | |
2178 | * all are mapped and non delay. We don't want to | |
2179 | * do block allocation here. | |
2180 | */ | |
2181 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
2182 | ext4_normal_get_block_write); | |
2183 | if (!ret) { | |
2184 | page_bufs = page_buffers(page); | |
2185 | /* check whether all are mapped and non delay */ | |
2186 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2187 | ext4_bh_unmapped_or_delay)) { | |
2188 | redirty_page_for_writepage(wbc, page); | |
2189 | unlock_page(page); | |
2190 | return 0; | |
2191 | } | |
2192 | } else { | |
2193 | /* | |
2194 | * We can't do block allocation here | |
2195 | * so just redity the page and unlock | |
2196 | * and return | |
61628a3f | 2197 | */ |
61628a3f MC |
2198 | redirty_page_for_writepage(wbc, page); |
2199 | unlock_page(page); | |
2200 | return 0; | |
2201 | } | |
64769240 AT |
2202 | } |
2203 | ||
2204 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) | |
f0e6c985 | 2205 | ret = nobh_writepage(page, ext4_normal_get_block_write, wbc); |
64769240 | 2206 | else |
f0e6c985 AK |
2207 | ret = block_write_full_page(page, |
2208 | ext4_normal_get_block_write, | |
2209 | wbc); | |
64769240 | 2210 | |
64769240 AT |
2211 | return ret; |
2212 | } | |
2213 | ||
61628a3f MC |
2214 | /* |
2215 | * For now just follow the DIO way to estimate the max credits | |
2216 | * needed to write out EXT4_MAX_WRITEBACK_PAGES. | |
2217 | * todo: need to calculate the max credits need for | |
2218 | * extent based files, currently the DIO credits is based on | |
2219 | * indirect-blocks mapping way. | |
2220 | * | |
2221 | * Probably should have a generic way to calculate credits | |
2222 | * for DIO, writepages, and truncate | |
2223 | */ | |
2224 | #define EXT4_MAX_WRITEBACK_PAGES DIO_MAX_BLOCKS | |
2225 | #define EXT4_MAX_WRITEBACK_CREDITS DIO_CREDITS | |
2226 | ||
64769240 AT |
2227 | static int ext4_da_writepages(struct address_space *mapping, |
2228 | struct writeback_control *wbc) | |
2229 | { | |
61628a3f MC |
2230 | struct inode *inode = mapping->host; |
2231 | handle_t *handle = NULL; | |
2232 | int needed_blocks; | |
2233 | int ret = 0; | |
2234 | long to_write; | |
2235 | loff_t range_start = 0; | |
2236 | ||
2237 | /* | |
2238 | * No pages to write? This is mainly a kludge to avoid starting | |
2239 | * a transaction for special inodes like journal inode on last iput() | |
2240 | * because that could violate lock ordering on umount | |
2241 | */ | |
2242 | if (!mapping->nrpages) | |
2243 | return 0; | |
2244 | ||
2245 | /* | |
cd1aac32 | 2246 | * Estimate the worse case needed credits to write out |
61628a3f MC |
2247 | * EXT4_MAX_BUF_BLOCKS pages |
2248 | */ | |
2249 | needed_blocks = EXT4_MAX_WRITEBACK_CREDITS; | |
2250 | ||
2251 | to_write = wbc->nr_to_write; | |
2252 | if (!wbc->range_cyclic) { | |
2253 | /* | |
2254 | * If range_cyclic is not set force range_cont | |
2255 | * and save the old writeback_index | |
2256 | */ | |
2257 | wbc->range_cont = 1; | |
2258 | range_start = wbc->range_start; | |
2259 | } | |
2260 | ||
2261 | while (!ret && to_write) { | |
2262 | /* start a new transaction*/ | |
2263 | handle = ext4_journal_start(inode, needed_blocks); | |
2264 | if (IS_ERR(handle)) { | |
2265 | ret = PTR_ERR(handle); | |
2266 | goto out_writepages; | |
2267 | } | |
cd1aac32 AK |
2268 | if (ext4_should_order_data(inode)) { |
2269 | /* | |
2270 | * With ordered mode we need to add | |
2271 | * the inode to the journal handle | |
2272 | * when we do block allocation. | |
2273 | */ | |
2274 | ret = ext4_jbd2_file_inode(handle, inode); | |
2275 | if (ret) { | |
2276 | ext4_journal_stop(handle); | |
2277 | goto out_writepages; | |
2278 | } | |
2279 | ||
2280 | } | |
61628a3f MC |
2281 | /* |
2282 | * set the max dirty pages could be write at a time | |
2283 | * to fit into the reserved transaction credits | |
2284 | */ | |
2285 | if (wbc->nr_to_write > EXT4_MAX_WRITEBACK_PAGES) | |
2286 | wbc->nr_to_write = EXT4_MAX_WRITEBACK_PAGES; | |
2287 | ||
2288 | to_write -= wbc->nr_to_write; | |
2289 | ret = mpage_da_writepages(mapping, wbc, | |
2290 | ext4_da_get_block_write); | |
2291 | ext4_journal_stop(handle); | |
2292 | if (wbc->nr_to_write) { | |
2293 | /* | |
2294 | * There is no more writeout needed | |
2295 | * or we requested for a noblocking writeout | |
2296 | * and we found the device congested | |
2297 | */ | |
2298 | to_write += wbc->nr_to_write; | |
2299 | break; | |
2300 | } | |
2301 | wbc->nr_to_write = to_write; | |
2302 | } | |
2303 | ||
2304 | out_writepages: | |
2305 | wbc->nr_to_write = to_write; | |
2306 | if (range_start) | |
2307 | wbc->range_start = range_start; | |
2308 | return ret; | |
64769240 AT |
2309 | } |
2310 | ||
2311 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, | |
2312 | loff_t pos, unsigned len, unsigned flags, | |
2313 | struct page **pagep, void **fsdata) | |
2314 | { | |
d2a17637 | 2315 | int ret, retries = 0; |
64769240 AT |
2316 | struct page *page; |
2317 | pgoff_t index; | |
2318 | unsigned from, to; | |
2319 | struct inode *inode = mapping->host; | |
2320 | handle_t *handle; | |
2321 | ||
2322 | index = pos >> PAGE_CACHE_SHIFT; | |
2323 | from = pos & (PAGE_CACHE_SIZE - 1); | |
2324 | to = from + len; | |
2325 | ||
d2a17637 | 2326 | retry: |
64769240 AT |
2327 | /* |
2328 | * With delayed allocation, we don't log the i_disksize update | |
2329 | * if there is delayed block allocation. But we still need | |
2330 | * to journalling the i_disksize update if writes to the end | |
2331 | * of file which has an already mapped buffer. | |
2332 | */ | |
2333 | handle = ext4_journal_start(inode, 1); | |
2334 | if (IS_ERR(handle)) { | |
2335 | ret = PTR_ERR(handle); | |
2336 | goto out; | |
2337 | } | |
2338 | ||
2339 | page = __grab_cache_page(mapping, index); | |
d5a0d4f7 ES |
2340 | if (!page) { |
2341 | ext4_journal_stop(handle); | |
2342 | ret = -ENOMEM; | |
2343 | goto out; | |
2344 | } | |
64769240 AT |
2345 | *pagep = page; |
2346 | ||
2347 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
2348 | ext4_da_get_block_prep); | |
2349 | if (ret < 0) { | |
2350 | unlock_page(page); | |
2351 | ext4_journal_stop(handle); | |
2352 | page_cache_release(page); | |
2353 | } | |
2354 | ||
d2a17637 MC |
2355 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2356 | goto retry; | |
64769240 AT |
2357 | out: |
2358 | return ret; | |
2359 | } | |
2360 | ||
632eaeab MC |
2361 | /* |
2362 | * Check if we should update i_disksize | |
2363 | * when write to the end of file but not require block allocation | |
2364 | */ | |
2365 | static int ext4_da_should_update_i_disksize(struct page *page, | |
2366 | unsigned long offset) | |
2367 | { | |
2368 | struct buffer_head *bh; | |
2369 | struct inode *inode = page->mapping->host; | |
2370 | unsigned int idx; | |
2371 | int i; | |
2372 | ||
2373 | bh = page_buffers(page); | |
2374 | idx = offset >> inode->i_blkbits; | |
2375 | ||
2376 | for (i=0; i < idx; i++) | |
2377 | bh = bh->b_this_page; | |
2378 | ||
2379 | if (!buffer_mapped(bh) || (buffer_delay(bh))) | |
2380 | return 0; | |
2381 | return 1; | |
2382 | } | |
2383 | ||
64769240 AT |
2384 | static int ext4_da_write_end(struct file *file, |
2385 | struct address_space *mapping, | |
2386 | loff_t pos, unsigned len, unsigned copied, | |
2387 | struct page *page, void *fsdata) | |
2388 | { | |
2389 | struct inode *inode = mapping->host; | |
2390 | int ret = 0, ret2; | |
2391 | handle_t *handle = ext4_journal_current_handle(); | |
2392 | loff_t new_i_size; | |
632eaeab MC |
2393 | unsigned long start, end; |
2394 | ||
2395 | start = pos & (PAGE_CACHE_SIZE - 1); | |
2396 | end = start + copied -1; | |
64769240 AT |
2397 | |
2398 | /* | |
2399 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2400 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2401 | * into that. | |
2402 | */ | |
2403 | ||
2404 | new_i_size = pos + copied; | |
632eaeab MC |
2405 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
2406 | if (ext4_da_should_update_i_disksize(page, end)) { | |
2407 | down_write(&EXT4_I(inode)->i_data_sem); | |
2408 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2409 | /* | |
2410 | * Updating i_disksize when extending file | |
2411 | * without needing block allocation | |
2412 | */ | |
2413 | if (ext4_should_order_data(inode)) | |
2414 | ret = ext4_jbd2_file_inode(handle, | |
2415 | inode); | |
64769240 | 2416 | |
632eaeab MC |
2417 | EXT4_I(inode)->i_disksize = new_i_size; |
2418 | } | |
2419 | up_write(&EXT4_I(inode)->i_data_sem); | |
64769240 | 2420 | } |
632eaeab | 2421 | } |
64769240 AT |
2422 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
2423 | page, fsdata); | |
2424 | copied = ret2; | |
2425 | if (ret2 < 0) | |
2426 | ret = ret2; | |
2427 | ret2 = ext4_journal_stop(handle); | |
2428 | if (!ret) | |
2429 | ret = ret2; | |
2430 | ||
2431 | return ret ? ret : copied; | |
2432 | } | |
2433 | ||
2434 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2435 | { | |
64769240 AT |
2436 | /* |
2437 | * Drop reserved blocks | |
2438 | */ | |
2439 | BUG_ON(!PageLocked(page)); | |
2440 | if (!page_has_buffers(page)) | |
2441 | goto out; | |
2442 | ||
d2a17637 | 2443 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2444 | |
2445 | out: | |
2446 | ext4_invalidatepage(page, offset); | |
2447 | ||
2448 | return; | |
2449 | } | |
2450 | ||
2451 | ||
ac27a0ec DK |
2452 | /* |
2453 | * bmap() is special. It gets used by applications such as lilo and by | |
2454 | * the swapper to find the on-disk block of a specific piece of data. | |
2455 | * | |
2456 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2457 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2458 | * filesystem and enables swap, then they may get a nasty shock when the |
2459 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2460 | * the original zero's written out previously to the journal and | |
2461 | * awaiting writeback in the kernel's buffer cache. | |
2462 | * | |
2463 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2464 | * take extra steps to flush any blocks which might be in the cache. | |
2465 | */ | |
617ba13b | 2466 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2467 | { |
2468 | struct inode *inode = mapping->host; | |
2469 | journal_t *journal; | |
2470 | int err; | |
2471 | ||
64769240 AT |
2472 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2473 | test_opt(inode->i_sb, DELALLOC)) { | |
2474 | /* | |
2475 | * With delalloc we want to sync the file | |
2476 | * so that we can make sure we allocate | |
2477 | * blocks for file | |
2478 | */ | |
2479 | filemap_write_and_wait(mapping); | |
2480 | } | |
2481 | ||
617ba13b | 2482 | if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
ac27a0ec DK |
2483 | /* |
2484 | * This is a REALLY heavyweight approach, but the use of | |
2485 | * bmap on dirty files is expected to be extremely rare: | |
2486 | * only if we run lilo or swapon on a freshly made file | |
2487 | * do we expect this to happen. | |
2488 | * | |
2489 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2490 | * represent an unprivileged user DOS attack --- we'd be | |
2491 | * in trouble if mortal users could trigger this path at | |
2492 | * will.) | |
2493 | * | |
617ba13b | 2494 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2495 | * regular files. If somebody wants to bmap a directory |
2496 | * or symlink and gets confused because the buffer | |
2497 | * hasn't yet been flushed to disk, they deserve | |
2498 | * everything they get. | |
2499 | */ | |
2500 | ||
617ba13b MC |
2501 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
2502 | journal = EXT4_JOURNAL(inode); | |
dab291af MC |
2503 | jbd2_journal_lock_updates(journal); |
2504 | err = jbd2_journal_flush(journal); | |
2505 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2506 | |
2507 | if (err) | |
2508 | return 0; | |
2509 | } | |
2510 | ||
617ba13b | 2511 | return generic_block_bmap(mapping,block,ext4_get_block); |
ac27a0ec DK |
2512 | } |
2513 | ||
2514 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
2515 | { | |
2516 | get_bh(bh); | |
2517 | return 0; | |
2518 | } | |
2519 | ||
2520 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2521 | { | |
2522 | put_bh(bh); | |
2523 | return 0; | |
2524 | } | |
2525 | ||
ac27a0ec | 2526 | /* |
678aaf48 JK |
2527 | * Note that we don't need to start a transaction unless we're journaling data |
2528 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2529 | * need to file the inode to the transaction's list in ordered mode because if | |
2530 | * we are writing back data added by write(), the inode is already there and if | |
2531 | * we are writing back data modified via mmap(), noone guarantees in which | |
2532 | * transaction the data will hit the disk. In case we are journaling data, we | |
2533 | * cannot start transaction directly because transaction start ranks above page | |
2534 | * lock so we have to do some magic. | |
ac27a0ec | 2535 | * |
678aaf48 | 2536 | * In all journaling modes block_write_full_page() will start the I/O. |
ac27a0ec DK |
2537 | * |
2538 | * Problem: | |
2539 | * | |
617ba13b MC |
2540 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
2541 | * ext4_writepage() | |
ac27a0ec DK |
2542 | * |
2543 | * Similar for: | |
2544 | * | |
617ba13b | 2545 | * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
ac27a0ec | 2546 | * |
617ba13b | 2547 | * Same applies to ext4_get_block(). We will deadlock on various things like |
0e855ac8 | 2548 | * lock_journal and i_data_sem |
ac27a0ec DK |
2549 | * |
2550 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
2551 | * allocations fail. | |
2552 | * | |
2553 | * 16May01: If we're reentered then journal_current_handle() will be | |
2554 | * non-zero. We simply *return*. | |
2555 | * | |
2556 | * 1 July 2001: @@@ FIXME: | |
2557 | * In journalled data mode, a data buffer may be metadata against the | |
2558 | * current transaction. But the same file is part of a shared mapping | |
2559 | * and someone does a writepage() on it. | |
2560 | * | |
2561 | * We will move the buffer onto the async_data list, but *after* it has | |
2562 | * been dirtied. So there's a small window where we have dirty data on | |
2563 | * BJ_Metadata. | |
2564 | * | |
2565 | * Note that this only applies to the last partial page in the file. The | |
2566 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
2567 | * broken code anyway: it's wrong for msync()). | |
2568 | * | |
2569 | * It's a rare case: affects the final partial page, for journalled data | |
2570 | * where the file is subject to bith write() and writepage() in the same | |
2571 | * transction. To fix it we'll need a custom block_write_full_page(). | |
2572 | * We'll probably need that anyway for journalling writepage() output. | |
2573 | * | |
2574 | * We don't honour synchronous mounts for writepage(). That would be | |
2575 | * disastrous. Any write() or metadata operation will sync the fs for | |
2576 | * us. | |
2577 | * | |
ac27a0ec | 2578 | */ |
678aaf48 | 2579 | static int __ext4_normal_writepage(struct page *page, |
cf108bca JK |
2580 | struct writeback_control *wbc) |
2581 | { | |
2582 | struct inode *inode = page->mapping->host; | |
2583 | ||
2584 | if (test_opt(inode->i_sb, NOBH)) | |
f0e6c985 AK |
2585 | return nobh_writepage(page, |
2586 | ext4_normal_get_block_write, wbc); | |
cf108bca | 2587 | else |
f0e6c985 AK |
2588 | return block_write_full_page(page, |
2589 | ext4_normal_get_block_write, | |
2590 | wbc); | |
cf108bca JK |
2591 | } |
2592 | ||
678aaf48 | 2593 | static int ext4_normal_writepage(struct page *page, |
ac27a0ec DK |
2594 | struct writeback_control *wbc) |
2595 | { | |
2596 | struct inode *inode = page->mapping->host; | |
cf108bca JK |
2597 | loff_t size = i_size_read(inode); |
2598 | loff_t len; | |
2599 | ||
2600 | J_ASSERT(PageLocked(page)); | |
cf108bca JK |
2601 | if (page->index == size >> PAGE_CACHE_SHIFT) |
2602 | len = size & ~PAGE_CACHE_MASK; | |
2603 | else | |
2604 | len = PAGE_CACHE_SIZE; | |
f0e6c985 AK |
2605 | |
2606 | if (page_has_buffers(page)) { | |
2607 | /* if page has buffers it should all be mapped | |
2608 | * and allocated. If there are not buffers attached | |
2609 | * to the page we know the page is dirty but it lost | |
2610 | * buffers. That means that at some moment in time | |
2611 | * after write_begin() / write_end() has been called | |
2612 | * all buffers have been clean and thus they must have been | |
2613 | * written at least once. So they are all mapped and we can | |
2614 | * happily proceed with mapping them and writing the page. | |
2615 | */ | |
2616 | BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
2617 | ext4_bh_unmapped_or_delay)); | |
2618 | } | |
cf108bca JK |
2619 | |
2620 | if (!ext4_journal_current_handle()) | |
678aaf48 | 2621 | return __ext4_normal_writepage(page, wbc); |
cf108bca JK |
2622 | |
2623 | redirty_page_for_writepage(wbc, page); | |
2624 | unlock_page(page); | |
2625 | return 0; | |
2626 | } | |
2627 | ||
2628 | static int __ext4_journalled_writepage(struct page *page, | |
2629 | struct writeback_control *wbc) | |
2630 | { | |
2631 | struct address_space *mapping = page->mapping; | |
2632 | struct inode *inode = mapping->host; | |
2633 | struct buffer_head *page_bufs; | |
ac27a0ec DK |
2634 | handle_t *handle = NULL; |
2635 | int ret = 0; | |
2636 | int err; | |
2637 | ||
f0e6c985 AK |
2638 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, |
2639 | ext4_normal_get_block_write); | |
cf108bca JK |
2640 | if (ret != 0) |
2641 | goto out_unlock; | |
2642 | ||
2643 | page_bufs = page_buffers(page); | |
2644 | walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL, | |
2645 | bget_one); | |
2646 | /* As soon as we unlock the page, it can go away, but we have | |
2647 | * references to buffers so we are safe */ | |
2648 | unlock_page(page); | |
ac27a0ec | 2649 | |
617ba13b | 2650 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
2651 | if (IS_ERR(handle)) { |
2652 | ret = PTR_ERR(handle); | |
cf108bca | 2653 | goto out; |
ac27a0ec DK |
2654 | } |
2655 | ||
cf108bca JK |
2656 | ret = walk_page_buffers(handle, page_bufs, 0, |
2657 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
ac27a0ec | 2658 | |
cf108bca JK |
2659 | err = walk_page_buffers(handle, page_bufs, 0, |
2660 | PAGE_CACHE_SIZE, NULL, write_end_fn); | |
2661 | if (ret == 0) | |
2662 | ret = err; | |
617ba13b | 2663 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
2664 | if (!ret) |
2665 | ret = err; | |
ac27a0ec | 2666 | |
cf108bca JK |
2667 | walk_page_buffers(handle, page_bufs, 0, |
2668 | PAGE_CACHE_SIZE, NULL, bput_one); | |
2669 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; | |
2670 | goto out; | |
2671 | ||
2672 | out_unlock: | |
ac27a0ec | 2673 | unlock_page(page); |
cf108bca | 2674 | out: |
ac27a0ec DK |
2675 | return ret; |
2676 | } | |
2677 | ||
617ba13b | 2678 | static int ext4_journalled_writepage(struct page *page, |
ac27a0ec DK |
2679 | struct writeback_control *wbc) |
2680 | { | |
2681 | struct inode *inode = page->mapping->host; | |
cf108bca JK |
2682 | loff_t size = i_size_read(inode); |
2683 | loff_t len; | |
ac27a0ec | 2684 | |
cf108bca | 2685 | J_ASSERT(PageLocked(page)); |
cf108bca JK |
2686 | if (page->index == size >> PAGE_CACHE_SHIFT) |
2687 | len = size & ~PAGE_CACHE_MASK; | |
2688 | else | |
2689 | len = PAGE_CACHE_SIZE; | |
f0e6c985 AK |
2690 | |
2691 | if (page_has_buffers(page)) { | |
2692 | /* if page has buffers it should all be mapped | |
2693 | * and allocated. If there are not buffers attached | |
2694 | * to the page we know the page is dirty but it lost | |
2695 | * buffers. That means that at some moment in time | |
2696 | * after write_begin() / write_end() has been called | |
2697 | * all buffers have been clean and thus they must have been | |
2698 | * written at least once. So they are all mapped and we can | |
2699 | * happily proceed with mapping them and writing the page. | |
2700 | */ | |
2701 | BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
2702 | ext4_bh_unmapped_or_delay)); | |
2703 | } | |
ac27a0ec | 2704 | |
cf108bca | 2705 | if (ext4_journal_current_handle()) |
ac27a0ec | 2706 | goto no_write; |
ac27a0ec | 2707 | |
cf108bca | 2708 | if (PageChecked(page)) { |
ac27a0ec DK |
2709 | /* |
2710 | * It's mmapped pagecache. Add buffers and journal it. There | |
2711 | * doesn't seem much point in redirtying the page here. | |
2712 | */ | |
2713 | ClearPageChecked(page); | |
cf108bca | 2714 | return __ext4_journalled_writepage(page, wbc); |
ac27a0ec DK |
2715 | } else { |
2716 | /* | |
2717 | * It may be a page full of checkpoint-mode buffers. We don't | |
2718 | * really know unless we go poke around in the buffer_heads. | |
2719 | * But block_write_full_page will do the right thing. | |
2720 | */ | |
f0e6c985 AK |
2721 | return block_write_full_page(page, |
2722 | ext4_normal_get_block_write, | |
2723 | wbc); | |
ac27a0ec | 2724 | } |
ac27a0ec DK |
2725 | no_write: |
2726 | redirty_page_for_writepage(wbc, page); | |
ac27a0ec | 2727 | unlock_page(page); |
cf108bca | 2728 | return 0; |
ac27a0ec DK |
2729 | } |
2730 | ||
617ba13b | 2731 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2732 | { |
617ba13b | 2733 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
2734 | } |
2735 | ||
2736 | static int | |
617ba13b | 2737 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2738 | struct list_head *pages, unsigned nr_pages) |
2739 | { | |
617ba13b | 2740 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2741 | } |
2742 | ||
617ba13b | 2743 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2744 | { |
617ba13b | 2745 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
2746 | |
2747 | /* | |
2748 | * If it's a full truncate we just forget about the pending dirtying | |
2749 | */ | |
2750 | if (offset == 0) | |
2751 | ClearPageChecked(page); | |
2752 | ||
dab291af | 2753 | jbd2_journal_invalidatepage(journal, page, offset); |
ac27a0ec DK |
2754 | } |
2755 | ||
617ba13b | 2756 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2757 | { |
617ba13b | 2758 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
2759 | |
2760 | WARN_ON(PageChecked(page)); | |
2761 | if (!page_has_buffers(page)) | |
2762 | return 0; | |
dab291af | 2763 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
ac27a0ec DK |
2764 | } |
2765 | ||
2766 | /* | |
2767 | * If the O_DIRECT write will extend the file then add this inode to the | |
2768 | * orphan list. So recovery will truncate it back to the original size | |
2769 | * if the machine crashes during the write. | |
2770 | * | |
2771 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
2772 | * crashes then stale disk data _may_ be exposed inside the file. But current |
2773 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 2774 | */ |
617ba13b | 2775 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
ac27a0ec DK |
2776 | const struct iovec *iov, loff_t offset, |
2777 | unsigned long nr_segs) | |
2778 | { | |
2779 | struct file *file = iocb->ki_filp; | |
2780 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 2781 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 2782 | handle_t *handle; |
ac27a0ec DK |
2783 | ssize_t ret; |
2784 | int orphan = 0; | |
2785 | size_t count = iov_length(iov, nr_segs); | |
2786 | ||
2787 | if (rw == WRITE) { | |
2788 | loff_t final_size = offset + count; | |
2789 | ||
ac27a0ec | 2790 | if (final_size > inode->i_size) { |
7fb5409d JK |
2791 | /* Credits for sb + inode write */ |
2792 | handle = ext4_journal_start(inode, 2); | |
2793 | if (IS_ERR(handle)) { | |
2794 | ret = PTR_ERR(handle); | |
2795 | goto out; | |
2796 | } | |
617ba13b | 2797 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
2798 | if (ret) { |
2799 | ext4_journal_stop(handle); | |
2800 | goto out; | |
2801 | } | |
ac27a0ec DK |
2802 | orphan = 1; |
2803 | ei->i_disksize = inode->i_size; | |
7fb5409d | 2804 | ext4_journal_stop(handle); |
ac27a0ec DK |
2805 | } |
2806 | } | |
2807 | ||
2808 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, | |
2809 | offset, nr_segs, | |
617ba13b | 2810 | ext4_get_block, NULL); |
ac27a0ec | 2811 | |
7fb5409d | 2812 | if (orphan) { |
ac27a0ec DK |
2813 | int err; |
2814 | ||
7fb5409d JK |
2815 | /* Credits for sb + inode write */ |
2816 | handle = ext4_journal_start(inode, 2); | |
2817 | if (IS_ERR(handle)) { | |
2818 | /* This is really bad luck. We've written the data | |
2819 | * but cannot extend i_size. Bail out and pretend | |
2820 | * the write failed... */ | |
2821 | ret = PTR_ERR(handle); | |
2822 | goto out; | |
2823 | } | |
2824 | if (inode->i_nlink) | |
617ba13b | 2825 | ext4_orphan_del(handle, inode); |
7fb5409d | 2826 | if (ret > 0) { |
ac27a0ec DK |
2827 | loff_t end = offset + ret; |
2828 | if (end > inode->i_size) { | |
2829 | ei->i_disksize = end; | |
2830 | i_size_write(inode, end); | |
2831 | /* | |
2832 | * We're going to return a positive `ret' | |
2833 | * here due to non-zero-length I/O, so there's | |
2834 | * no way of reporting error returns from | |
617ba13b | 2835 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
2836 | * ignore it. |
2837 | */ | |
617ba13b | 2838 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
2839 | } |
2840 | } | |
617ba13b | 2841 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
2842 | if (ret == 0) |
2843 | ret = err; | |
2844 | } | |
2845 | out: | |
2846 | return ret; | |
2847 | } | |
2848 | ||
2849 | /* | |
617ba13b | 2850 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
2851 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
2852 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
2853 | * not necessarily locked. | |
2854 | * | |
2855 | * We cannot just dirty the page and leave attached buffers clean, because the | |
2856 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
2857 | * or jbddirty because all the journalling code will explode. | |
2858 | * | |
2859 | * So what we do is to mark the page "pending dirty" and next time writepage | |
2860 | * is called, propagate that into the buffers appropriately. | |
2861 | */ | |
617ba13b | 2862 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
2863 | { |
2864 | SetPageChecked(page); | |
2865 | return __set_page_dirty_nobuffers(page); | |
2866 | } | |
2867 | ||
617ba13b | 2868 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
2869 | .readpage = ext4_readpage, |
2870 | .readpages = ext4_readpages, | |
2871 | .writepage = ext4_normal_writepage, | |
2872 | .sync_page = block_sync_page, | |
2873 | .write_begin = ext4_write_begin, | |
2874 | .write_end = ext4_ordered_write_end, | |
2875 | .bmap = ext4_bmap, | |
2876 | .invalidatepage = ext4_invalidatepage, | |
2877 | .releasepage = ext4_releasepage, | |
2878 | .direct_IO = ext4_direct_IO, | |
2879 | .migratepage = buffer_migrate_page, | |
2880 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
2881 | }; |
2882 | ||
617ba13b | 2883 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
2884 | .readpage = ext4_readpage, |
2885 | .readpages = ext4_readpages, | |
2886 | .writepage = ext4_normal_writepage, | |
2887 | .sync_page = block_sync_page, | |
2888 | .write_begin = ext4_write_begin, | |
2889 | .write_end = ext4_writeback_write_end, | |
2890 | .bmap = ext4_bmap, | |
2891 | .invalidatepage = ext4_invalidatepage, | |
2892 | .releasepage = ext4_releasepage, | |
2893 | .direct_IO = ext4_direct_IO, | |
2894 | .migratepage = buffer_migrate_page, | |
2895 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
2896 | }; |
2897 | ||
617ba13b | 2898 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
2899 | .readpage = ext4_readpage, |
2900 | .readpages = ext4_readpages, | |
2901 | .writepage = ext4_journalled_writepage, | |
2902 | .sync_page = block_sync_page, | |
2903 | .write_begin = ext4_write_begin, | |
2904 | .write_end = ext4_journalled_write_end, | |
2905 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
2906 | .bmap = ext4_bmap, | |
2907 | .invalidatepage = ext4_invalidatepage, | |
2908 | .releasepage = ext4_releasepage, | |
2909 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
2910 | }; |
2911 | ||
64769240 | 2912 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
2913 | .readpage = ext4_readpage, |
2914 | .readpages = ext4_readpages, | |
2915 | .writepage = ext4_da_writepage, | |
2916 | .writepages = ext4_da_writepages, | |
2917 | .sync_page = block_sync_page, | |
2918 | .write_begin = ext4_da_write_begin, | |
2919 | .write_end = ext4_da_write_end, | |
2920 | .bmap = ext4_bmap, | |
2921 | .invalidatepage = ext4_da_invalidatepage, | |
2922 | .releasepage = ext4_releasepage, | |
2923 | .direct_IO = ext4_direct_IO, | |
2924 | .migratepage = buffer_migrate_page, | |
2925 | .is_partially_uptodate = block_is_partially_uptodate, | |
64769240 AT |
2926 | }; |
2927 | ||
617ba13b | 2928 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 2929 | { |
cd1aac32 AK |
2930 | if (ext4_should_order_data(inode) && |
2931 | test_opt(inode->i_sb, DELALLOC)) | |
2932 | inode->i_mapping->a_ops = &ext4_da_aops; | |
2933 | else if (ext4_should_order_data(inode)) | |
617ba13b | 2934 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
2935 | else if (ext4_should_writeback_data(inode) && |
2936 | test_opt(inode->i_sb, DELALLOC)) | |
2937 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
2938 | else if (ext4_should_writeback_data(inode)) |
2939 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 2940 | else |
617ba13b | 2941 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
2942 | } |
2943 | ||
2944 | /* | |
617ba13b | 2945 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
2946 | * up to the end of the block which corresponds to `from'. |
2947 | * This required during truncate. We need to physically zero the tail end | |
2948 | * of that block so it doesn't yield old data if the file is later grown. | |
2949 | */ | |
cf108bca | 2950 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
2951 | struct address_space *mapping, loff_t from) |
2952 | { | |
617ba13b | 2953 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 2954 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
2955 | unsigned blocksize, length, pos; |
2956 | ext4_lblk_t iblock; | |
ac27a0ec DK |
2957 | struct inode *inode = mapping->host; |
2958 | struct buffer_head *bh; | |
cf108bca | 2959 | struct page *page; |
ac27a0ec | 2960 | int err = 0; |
ac27a0ec | 2961 | |
cf108bca JK |
2962 | page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT); |
2963 | if (!page) | |
2964 | return -EINVAL; | |
2965 | ||
ac27a0ec DK |
2966 | blocksize = inode->i_sb->s_blocksize; |
2967 | length = blocksize - (offset & (blocksize - 1)); | |
2968 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
2969 | ||
2970 | /* | |
2971 | * For "nobh" option, we can only work if we don't need to | |
2972 | * read-in the page - otherwise we create buffers to do the IO. | |
2973 | */ | |
2974 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && | |
617ba13b | 2975 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
eebd2aa3 | 2976 | zero_user(page, offset, length); |
ac27a0ec DK |
2977 | set_page_dirty(page); |
2978 | goto unlock; | |
2979 | } | |
2980 | ||
2981 | if (!page_has_buffers(page)) | |
2982 | create_empty_buffers(page, blocksize, 0); | |
2983 | ||
2984 | /* Find the buffer that contains "offset" */ | |
2985 | bh = page_buffers(page); | |
2986 | pos = blocksize; | |
2987 | while (offset >= pos) { | |
2988 | bh = bh->b_this_page; | |
2989 | iblock++; | |
2990 | pos += blocksize; | |
2991 | } | |
2992 | ||
2993 | err = 0; | |
2994 | if (buffer_freed(bh)) { | |
2995 | BUFFER_TRACE(bh, "freed: skip"); | |
2996 | goto unlock; | |
2997 | } | |
2998 | ||
2999 | if (!buffer_mapped(bh)) { | |
3000 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3001 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3002 | /* unmapped? It's a hole - nothing to do */ |
3003 | if (!buffer_mapped(bh)) { | |
3004 | BUFFER_TRACE(bh, "still unmapped"); | |
3005 | goto unlock; | |
3006 | } | |
3007 | } | |
3008 | ||
3009 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3010 | if (PageUptodate(page)) | |
3011 | set_buffer_uptodate(bh); | |
3012 | ||
3013 | if (!buffer_uptodate(bh)) { | |
3014 | err = -EIO; | |
3015 | ll_rw_block(READ, 1, &bh); | |
3016 | wait_on_buffer(bh); | |
3017 | /* Uhhuh. Read error. Complain and punt. */ | |
3018 | if (!buffer_uptodate(bh)) | |
3019 | goto unlock; | |
3020 | } | |
3021 | ||
617ba13b | 3022 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3023 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3024 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3025 | if (err) |
3026 | goto unlock; | |
3027 | } | |
3028 | ||
eebd2aa3 | 3029 | zero_user(page, offset, length); |
ac27a0ec DK |
3030 | |
3031 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3032 | ||
3033 | err = 0; | |
617ba13b MC |
3034 | if (ext4_should_journal_data(inode)) { |
3035 | err = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 3036 | } else { |
617ba13b | 3037 | if (ext4_should_order_data(inode)) |
678aaf48 | 3038 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
3039 | mark_buffer_dirty(bh); |
3040 | } | |
3041 | ||
3042 | unlock: | |
3043 | unlock_page(page); | |
3044 | page_cache_release(page); | |
3045 | return err; | |
3046 | } | |
3047 | ||
3048 | /* | |
3049 | * Probably it should be a library function... search for first non-zero word | |
3050 | * or memcmp with zero_page, whatever is better for particular architecture. | |
3051 | * Linus? | |
3052 | */ | |
3053 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
3054 | { | |
3055 | while (p < q) | |
3056 | if (*p++) | |
3057 | return 0; | |
3058 | return 1; | |
3059 | } | |
3060 | ||
3061 | /** | |
617ba13b | 3062 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
3063 | * @inode: inode in question |
3064 | * @depth: depth of the affected branch | |
617ba13b | 3065 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
3066 | * @chain: place to store the pointers to partial indirect blocks |
3067 | * @top: place to the (detached) top of branch | |
3068 | * | |
617ba13b | 3069 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
3070 | * |
3071 | * When we do truncate() we may have to clean the ends of several | |
3072 | * indirect blocks but leave the blocks themselves alive. Block is | |
3073 | * partially truncated if some data below the new i_size is refered | |
3074 | * from it (and it is on the path to the first completely truncated | |
3075 | * data block, indeed). We have to free the top of that path along | |
3076 | * with everything to the right of the path. Since no allocation | |
617ba13b | 3077 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
3078 | * finishes, we may safely do the latter, but top of branch may |
3079 | * require special attention - pageout below the truncation point | |
3080 | * might try to populate it. | |
3081 | * | |
3082 | * We atomically detach the top of branch from the tree, store the | |
3083 | * block number of its root in *@top, pointers to buffer_heads of | |
3084 | * partially truncated blocks - in @chain[].bh and pointers to | |
3085 | * their last elements that should not be removed - in | |
3086 | * @chain[].p. Return value is the pointer to last filled element | |
3087 | * of @chain. | |
3088 | * | |
3089 | * The work left to caller to do the actual freeing of subtrees: | |
3090 | * a) free the subtree starting from *@top | |
3091 | * b) free the subtrees whose roots are stored in | |
3092 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
3093 | * c) free the subtrees growing from the inode past the @chain[0]. | |
3094 | * (no partially truncated stuff there). */ | |
3095 | ||
617ba13b | 3096 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
725d26d3 | 3097 | ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top) |
ac27a0ec DK |
3098 | { |
3099 | Indirect *partial, *p; | |
3100 | int k, err; | |
3101 | ||
3102 | *top = 0; | |
3103 | /* Make k index the deepest non-null offest + 1 */ | |
3104 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
3105 | ; | |
617ba13b | 3106 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
3107 | /* Writer: pointers */ |
3108 | if (!partial) | |
3109 | partial = chain + k-1; | |
3110 | /* | |
3111 | * If the branch acquired continuation since we've looked at it - | |
3112 | * fine, it should all survive and (new) top doesn't belong to us. | |
3113 | */ | |
3114 | if (!partial->key && *partial->p) | |
3115 | /* Writer: end */ | |
3116 | goto no_top; | |
3117 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) | |
3118 | ; | |
3119 | /* | |
3120 | * OK, we've found the last block that must survive. The rest of our | |
3121 | * branch should be detached before unlocking. However, if that rest | |
3122 | * of branch is all ours and does not grow immediately from the inode | |
3123 | * it's easier to cheat and just decrement partial->p. | |
3124 | */ | |
3125 | if (p == chain + k - 1 && p > chain) { | |
3126 | p->p--; | |
3127 | } else { | |
3128 | *top = *p->p; | |
617ba13b | 3129 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
3130 | #if 0 |
3131 | *p->p = 0; | |
3132 | #endif | |
3133 | } | |
3134 | /* Writer: end */ | |
3135 | ||
3136 | while(partial > p) { | |
3137 | brelse(partial->bh); | |
3138 | partial--; | |
3139 | } | |
3140 | no_top: | |
3141 | return partial; | |
3142 | } | |
3143 | ||
3144 | /* | |
3145 | * Zero a number of block pointers in either an inode or an indirect block. | |
3146 | * If we restart the transaction we must again get write access to the | |
3147 | * indirect block for further modification. | |
3148 | * | |
3149 | * We release `count' blocks on disk, but (last - first) may be greater | |
3150 | * than `count' because there can be holes in there. | |
3151 | */ | |
617ba13b MC |
3152 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
3153 | struct buffer_head *bh, ext4_fsblk_t block_to_free, | |
ac27a0ec DK |
3154 | unsigned long count, __le32 *first, __le32 *last) |
3155 | { | |
3156 | __le32 *p; | |
3157 | if (try_to_extend_transaction(handle, inode)) { | |
3158 | if (bh) { | |
617ba13b MC |
3159 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
3160 | ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec | 3161 | } |
617ba13b MC |
3162 | ext4_mark_inode_dirty(handle, inode); |
3163 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
3164 | if (bh) { |
3165 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 3166 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3167 | } |
3168 | } | |
3169 | ||
3170 | /* | |
3171 | * Any buffers which are on the journal will be in memory. We find | |
dab291af | 3172 | * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget() |
ac27a0ec | 3173 | * on them. We've already detached each block from the file, so |
dab291af | 3174 | * bforget() in jbd2_journal_forget() should be safe. |
ac27a0ec | 3175 | * |
dab291af | 3176 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
ac27a0ec DK |
3177 | */ |
3178 | for (p = first; p < last; p++) { | |
3179 | u32 nr = le32_to_cpu(*p); | |
3180 | if (nr) { | |
1d03ec98 | 3181 | struct buffer_head *tbh; |
ac27a0ec DK |
3182 | |
3183 | *p = 0; | |
1d03ec98 AK |
3184 | tbh = sb_find_get_block(inode->i_sb, nr); |
3185 | ext4_forget(handle, 0, inode, tbh, nr); | |
ac27a0ec DK |
3186 | } |
3187 | } | |
3188 | ||
c9de560d | 3189 | ext4_free_blocks(handle, inode, block_to_free, count, 0); |
ac27a0ec DK |
3190 | } |
3191 | ||
3192 | /** | |
617ba13b | 3193 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
3194 | * @handle: handle for this transaction |
3195 | * @inode: inode we are dealing with | |
3196 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
3197 | * @first: array of block numbers | |
3198 | * @last: points immediately past the end of array | |
3199 | * | |
3200 | * We are freeing all blocks refered from that array (numbers are stored as | |
3201 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
3202 | * | |
3203 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
3204 | * blocks are contiguous then releasing them at one time will only affect one | |
3205 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
3206 | * actually use a lot of journal space. | |
3207 | * | |
3208 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
3209 | * block pointers. | |
3210 | */ | |
617ba13b | 3211 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
3212 | struct buffer_head *this_bh, |
3213 | __le32 *first, __le32 *last) | |
3214 | { | |
617ba13b | 3215 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
3216 | unsigned long count = 0; /* Number of blocks in the run */ |
3217 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
3218 | corresponding to | |
3219 | block_to_free */ | |
617ba13b | 3220 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
3221 | __le32 *p; /* Pointer into inode/ind |
3222 | for current block */ | |
3223 | int err; | |
3224 | ||
3225 | if (this_bh) { /* For indirect block */ | |
3226 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 3227 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
3228 | /* Important: if we can't update the indirect pointers |
3229 | * to the blocks, we can't free them. */ | |
3230 | if (err) | |
3231 | return; | |
3232 | } | |
3233 | ||
3234 | for (p = first; p < last; p++) { | |
3235 | nr = le32_to_cpu(*p); | |
3236 | if (nr) { | |
3237 | /* accumulate blocks to free if they're contiguous */ | |
3238 | if (count == 0) { | |
3239 | block_to_free = nr; | |
3240 | block_to_free_p = p; | |
3241 | count = 1; | |
3242 | } else if (nr == block_to_free + count) { | |
3243 | count++; | |
3244 | } else { | |
617ba13b | 3245 | ext4_clear_blocks(handle, inode, this_bh, |
ac27a0ec DK |
3246 | block_to_free, |
3247 | count, block_to_free_p, p); | |
3248 | block_to_free = nr; | |
3249 | block_to_free_p = p; | |
3250 | count = 1; | |
3251 | } | |
3252 | } | |
3253 | } | |
3254 | ||
3255 | if (count > 0) | |
617ba13b | 3256 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
3257 | count, block_to_free_p, p); |
3258 | ||
3259 | if (this_bh) { | |
617ba13b | 3260 | BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata"); |
71dc8fbc DG |
3261 | |
3262 | /* | |
3263 | * The buffer head should have an attached journal head at this | |
3264 | * point. However, if the data is corrupted and an indirect | |
3265 | * block pointed to itself, it would have been detached when | |
3266 | * the block was cleared. Check for this instead of OOPSing. | |
3267 | */ | |
3268 | if (bh2jh(this_bh)) | |
3269 | ext4_journal_dirty_metadata(handle, this_bh); | |
3270 | else | |
3271 | ext4_error(inode->i_sb, __func__, | |
3272 | "circular indirect block detected, " | |
3273 | "inode=%lu, block=%llu", | |
3274 | inode->i_ino, | |
3275 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
3276 | } |
3277 | } | |
3278 | ||
3279 | /** | |
617ba13b | 3280 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
3281 | * @handle: JBD handle for this transaction |
3282 | * @inode: inode we are dealing with | |
3283 | * @parent_bh: the buffer_head which contains *@first and *@last | |
3284 | * @first: array of block numbers | |
3285 | * @last: pointer immediately past the end of array | |
3286 | * @depth: depth of the branches to free | |
3287 | * | |
3288 | * We are freeing all blocks refered from these branches (numbers are | |
3289 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
3290 | * appropriately. | |
3291 | */ | |
617ba13b | 3292 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
3293 | struct buffer_head *parent_bh, |
3294 | __le32 *first, __le32 *last, int depth) | |
3295 | { | |
617ba13b | 3296 | ext4_fsblk_t nr; |
ac27a0ec DK |
3297 | __le32 *p; |
3298 | ||
3299 | if (is_handle_aborted(handle)) | |
3300 | return; | |
3301 | ||
3302 | if (depth--) { | |
3303 | struct buffer_head *bh; | |
617ba13b | 3304 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
3305 | p = last; |
3306 | while (--p >= first) { | |
3307 | nr = le32_to_cpu(*p); | |
3308 | if (!nr) | |
3309 | continue; /* A hole */ | |
3310 | ||
3311 | /* Go read the buffer for the next level down */ | |
3312 | bh = sb_bread(inode->i_sb, nr); | |
3313 | ||
3314 | /* | |
3315 | * A read failure? Report error and clear slot | |
3316 | * (should be rare). | |
3317 | */ | |
3318 | if (!bh) { | |
617ba13b | 3319 | ext4_error(inode->i_sb, "ext4_free_branches", |
2ae02107 | 3320 | "Read failure, inode=%lu, block=%llu", |
ac27a0ec DK |
3321 | inode->i_ino, nr); |
3322 | continue; | |
3323 | } | |
3324 | ||
3325 | /* This zaps the entire block. Bottom up. */ | |
3326 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 3327 | ext4_free_branches(handle, inode, bh, |
ac27a0ec DK |
3328 | (__le32*)bh->b_data, |
3329 | (__le32*)bh->b_data + addr_per_block, | |
3330 | depth); | |
3331 | ||
3332 | /* | |
3333 | * We've probably journalled the indirect block several | |
3334 | * times during the truncate. But it's no longer | |
3335 | * needed and we now drop it from the transaction via | |
dab291af | 3336 | * jbd2_journal_revoke(). |
ac27a0ec DK |
3337 | * |
3338 | * That's easy if it's exclusively part of this | |
3339 | * transaction. But if it's part of the committing | |
dab291af | 3340 | * transaction then jbd2_journal_forget() will simply |
ac27a0ec | 3341 | * brelse() it. That means that if the underlying |
617ba13b | 3342 | * block is reallocated in ext4_get_block(), |
ac27a0ec DK |
3343 | * unmap_underlying_metadata() will find this block |
3344 | * and will try to get rid of it. damn, damn. | |
3345 | * | |
3346 | * If this block has already been committed to the | |
3347 | * journal, a revoke record will be written. And | |
3348 | * revoke records must be emitted *before* clearing | |
3349 | * this block's bit in the bitmaps. | |
3350 | */ | |
617ba13b | 3351 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
ac27a0ec DK |
3352 | |
3353 | /* | |
3354 | * Everything below this this pointer has been | |
3355 | * released. Now let this top-of-subtree go. | |
3356 | * | |
3357 | * We want the freeing of this indirect block to be | |
3358 | * atomic in the journal with the updating of the | |
3359 | * bitmap block which owns it. So make some room in | |
3360 | * the journal. | |
3361 | * | |
3362 | * We zero the parent pointer *after* freeing its | |
3363 | * pointee in the bitmaps, so if extend_transaction() | |
3364 | * for some reason fails to put the bitmap changes and | |
3365 | * the release into the same transaction, recovery | |
3366 | * will merely complain about releasing a free block, | |
3367 | * rather than leaking blocks. | |
3368 | */ | |
3369 | if (is_handle_aborted(handle)) | |
3370 | return; | |
3371 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b MC |
3372 | ext4_mark_inode_dirty(handle, inode); |
3373 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
3374 | } |
3375 | ||
c9de560d | 3376 | ext4_free_blocks(handle, inode, nr, 1, 1); |
ac27a0ec DK |
3377 | |
3378 | if (parent_bh) { | |
3379 | /* | |
3380 | * The block which we have just freed is | |
3381 | * pointed to by an indirect block: journal it | |
3382 | */ | |
3383 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 3384 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
3385 | parent_bh)){ |
3386 | *p = 0; | |
3387 | BUFFER_TRACE(parent_bh, | |
617ba13b MC |
3388 | "call ext4_journal_dirty_metadata"); |
3389 | ext4_journal_dirty_metadata(handle, | |
ac27a0ec DK |
3390 | parent_bh); |
3391 | } | |
3392 | } | |
3393 | } | |
3394 | } else { | |
3395 | /* We have reached the bottom of the tree. */ | |
3396 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 3397 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
3398 | } |
3399 | } | |
3400 | ||
91ef4caf DG |
3401 | int ext4_can_truncate(struct inode *inode) |
3402 | { | |
3403 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
3404 | return 0; | |
3405 | if (S_ISREG(inode->i_mode)) | |
3406 | return 1; | |
3407 | if (S_ISDIR(inode->i_mode)) | |
3408 | return 1; | |
3409 | if (S_ISLNK(inode->i_mode)) | |
3410 | return !ext4_inode_is_fast_symlink(inode); | |
3411 | return 0; | |
3412 | } | |
3413 | ||
ac27a0ec | 3414 | /* |
617ba13b | 3415 | * ext4_truncate() |
ac27a0ec | 3416 | * |
617ba13b MC |
3417 | * We block out ext4_get_block() block instantiations across the entire |
3418 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3419 | * simultaneously on behalf of the same inode. |
3420 | * | |
3421 | * As we work through the truncate and commmit bits of it to the journal there | |
3422 | * is one core, guiding principle: the file's tree must always be consistent on | |
3423 | * disk. We must be able to restart the truncate after a crash. | |
3424 | * | |
3425 | * The file's tree may be transiently inconsistent in memory (although it | |
3426 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3427 | * the contents of (the filesystem + the journal) must be consistent and | |
3428 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3429 | * left-to-right works OK too). | |
3430 | * | |
3431 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3432 | * truncate against the orphan inode list. | |
3433 | * | |
3434 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3435 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3436 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3437 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3438 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3439 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3440 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3441 | */ |
617ba13b | 3442 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
3443 | { |
3444 | handle_t *handle; | |
617ba13b | 3445 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 3446 | __le32 *i_data = ei->i_data; |
617ba13b | 3447 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 3448 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 3449 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
3450 | Indirect chain[4]; |
3451 | Indirect *partial; | |
3452 | __le32 nr = 0; | |
3453 | int n; | |
725d26d3 | 3454 | ext4_lblk_t last_block; |
ac27a0ec | 3455 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 3456 | |
91ef4caf | 3457 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3458 | return; |
3459 | ||
1d03ec98 | 3460 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
cf108bca | 3461 | ext4_ext_truncate(inode); |
1d03ec98 AK |
3462 | return; |
3463 | } | |
a86c6181 | 3464 | |
ac27a0ec | 3465 | handle = start_transaction(inode); |
cf108bca | 3466 | if (IS_ERR(handle)) |
ac27a0ec | 3467 | return; /* AKPM: return what? */ |
ac27a0ec DK |
3468 | |
3469 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 3470 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec | 3471 | |
cf108bca JK |
3472 | if (inode->i_size & (blocksize - 1)) |
3473 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
3474 | goto out_stop; | |
ac27a0ec | 3475 | |
617ba13b | 3476 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
3477 | if (n == 0) |
3478 | goto out_stop; /* error */ | |
3479 | ||
3480 | /* | |
3481 | * OK. This truncate is going to happen. We add the inode to the | |
3482 | * orphan list, so that if this truncate spans multiple transactions, | |
3483 | * and we crash, we will resume the truncate when the filesystem | |
3484 | * recovers. It also marks the inode dirty, to catch the new size. | |
3485 | * | |
3486 | * Implication: the file must always be in a sane, consistent | |
3487 | * truncatable state while each transaction commits. | |
3488 | */ | |
617ba13b | 3489 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
3490 | goto out_stop; |
3491 | ||
632eaeab MC |
3492 | /* |
3493 | * From here we block out all ext4_get_block() callers who want to | |
3494 | * modify the block allocation tree. | |
3495 | */ | |
3496 | down_write(&ei->i_data_sem); | |
b4df2030 TT |
3497 | |
3498 | ext4_discard_reservation(inode); | |
3499 | ||
ac27a0ec DK |
3500 | /* |
3501 | * The orphan list entry will now protect us from any crash which | |
3502 | * occurs before the truncate completes, so it is now safe to propagate | |
3503 | * the new, shorter inode size (held for now in i_size) into the | |
3504 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 3505 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
3506 | */ |
3507 | ei->i_disksize = inode->i_size; | |
3508 | ||
ac27a0ec | 3509 | if (n == 1) { /* direct blocks */ |
617ba13b MC |
3510 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
3511 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
3512 | goto do_indirects; |
3513 | } | |
3514 | ||
617ba13b | 3515 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
3516 | /* Kill the top of shared branch (not detached) */ |
3517 | if (nr) { | |
3518 | if (partial == chain) { | |
3519 | /* Shared branch grows from the inode */ | |
617ba13b | 3520 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
3521 | &nr, &nr+1, (chain+n-1) - partial); |
3522 | *partial->p = 0; | |
3523 | /* | |
3524 | * We mark the inode dirty prior to restart, | |
3525 | * and prior to stop. No need for it here. | |
3526 | */ | |
3527 | } else { | |
3528 | /* Shared branch grows from an indirect block */ | |
3529 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 3530 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
3531 | partial->p, |
3532 | partial->p+1, (chain+n-1) - partial); | |
3533 | } | |
3534 | } | |
3535 | /* Clear the ends of indirect blocks on the shared branch */ | |
3536 | while (partial > chain) { | |
617ba13b | 3537 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
3538 | (__le32*)partial->bh->b_data+addr_per_block, |
3539 | (chain+n-1) - partial); | |
3540 | BUFFER_TRACE(partial->bh, "call brelse"); | |
3541 | brelse (partial->bh); | |
3542 | partial--; | |
3543 | } | |
3544 | do_indirects: | |
3545 | /* Kill the remaining (whole) subtrees */ | |
3546 | switch (offsets[0]) { | |
3547 | default: | |
617ba13b | 3548 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 3549 | if (nr) { |
617ba13b MC |
3550 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
3551 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 3552 | } |
617ba13b MC |
3553 | case EXT4_IND_BLOCK: |
3554 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 3555 | if (nr) { |
617ba13b MC |
3556 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
3557 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 3558 | } |
617ba13b MC |
3559 | case EXT4_DIND_BLOCK: |
3560 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 3561 | if (nr) { |
617ba13b MC |
3562 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
3563 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 3564 | } |
617ba13b | 3565 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
3566 | ; |
3567 | } | |
3568 | ||
0e855ac8 | 3569 | up_write(&ei->i_data_sem); |
ef7f3835 | 3570 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 3571 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3572 | |
3573 | /* | |
3574 | * In a multi-transaction truncate, we only make the final transaction | |
3575 | * synchronous | |
3576 | */ | |
3577 | if (IS_SYNC(inode)) | |
3578 | handle->h_sync = 1; | |
3579 | out_stop: | |
3580 | /* | |
3581 | * If this was a simple ftruncate(), and the file will remain alive | |
3582 | * then we need to clear up the orphan record which we created above. | |
3583 | * However, if this was a real unlink then we were called by | |
617ba13b | 3584 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
3585 | * orphan info for us. |
3586 | */ | |
3587 | if (inode->i_nlink) | |
617ba13b | 3588 | ext4_orphan_del(handle, inode); |
ac27a0ec | 3589 | |
617ba13b | 3590 | ext4_journal_stop(handle); |
ac27a0ec DK |
3591 | } |
3592 | ||
617ba13b MC |
3593 | static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb, |
3594 | unsigned long ino, struct ext4_iloc *iloc) | |
ac27a0ec | 3595 | { |
fd2d4291 | 3596 | ext4_group_t block_group; |
ac27a0ec | 3597 | unsigned long offset; |
617ba13b | 3598 | ext4_fsblk_t block; |
c0a4ef38 | 3599 | struct ext4_group_desc *gdp; |
ac27a0ec | 3600 | |
617ba13b | 3601 | if (!ext4_valid_inum(sb, ino)) { |
ac27a0ec DK |
3602 | /* |
3603 | * This error is already checked for in namei.c unless we are | |
3604 | * looking at an NFS filehandle, in which case no error | |
3605 | * report is needed | |
3606 | */ | |
3607 | return 0; | |
3608 | } | |
3609 | ||
617ba13b | 3610 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
c0a4ef38 AM |
3611 | gdp = ext4_get_group_desc(sb, block_group, NULL); |
3612 | if (!gdp) | |
ac27a0ec | 3613 | return 0; |
ac27a0ec | 3614 | |
ac27a0ec DK |
3615 | /* |
3616 | * Figure out the offset within the block group inode table | |
3617 | */ | |
617ba13b MC |
3618 | offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) * |
3619 | EXT4_INODE_SIZE(sb); | |
8fadc143 AR |
3620 | block = ext4_inode_table(sb, gdp) + |
3621 | (offset >> EXT4_BLOCK_SIZE_BITS(sb)); | |
ac27a0ec DK |
3622 | |
3623 | iloc->block_group = block_group; | |
617ba13b | 3624 | iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1); |
ac27a0ec DK |
3625 | return block; |
3626 | } | |
3627 | ||
3628 | /* | |
617ba13b | 3629 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3630 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3631 | * data in memory that is needed to recreate the on-disk version of this | |
3632 | * inode. | |
3633 | */ | |
617ba13b MC |
3634 | static int __ext4_get_inode_loc(struct inode *inode, |
3635 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3636 | { |
617ba13b | 3637 | ext4_fsblk_t block; |
ac27a0ec DK |
3638 | struct buffer_head *bh; |
3639 | ||
617ba13b | 3640 | block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc); |
ac27a0ec DK |
3641 | if (!block) |
3642 | return -EIO; | |
3643 | ||
3644 | bh = sb_getblk(inode->i_sb, block); | |
3645 | if (!bh) { | |
617ba13b | 3646 | ext4_error (inode->i_sb, "ext4_get_inode_loc", |
ac27a0ec | 3647 | "unable to read inode block - " |
2ae02107 | 3648 | "inode=%lu, block=%llu", |
ac27a0ec DK |
3649 | inode->i_ino, block); |
3650 | return -EIO; | |
3651 | } | |
3652 | if (!buffer_uptodate(bh)) { | |
3653 | lock_buffer(bh); | |
9c83a923 HK |
3654 | |
3655 | /* | |
3656 | * If the buffer has the write error flag, we have failed | |
3657 | * to write out another inode in the same block. In this | |
3658 | * case, we don't have to read the block because we may | |
3659 | * read the old inode data successfully. | |
3660 | */ | |
3661 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3662 | set_buffer_uptodate(bh); | |
3663 | ||
ac27a0ec DK |
3664 | if (buffer_uptodate(bh)) { |
3665 | /* someone brought it uptodate while we waited */ | |
3666 | unlock_buffer(bh); | |
3667 | goto has_buffer; | |
3668 | } | |
3669 | ||
3670 | /* | |
3671 | * If we have all information of the inode in memory and this | |
3672 | * is the only valid inode in the block, we need not read the | |
3673 | * block. | |
3674 | */ | |
3675 | if (in_mem) { | |
3676 | struct buffer_head *bitmap_bh; | |
617ba13b | 3677 | struct ext4_group_desc *desc; |
ac27a0ec DK |
3678 | int inodes_per_buffer; |
3679 | int inode_offset, i; | |
fd2d4291 | 3680 | ext4_group_t block_group; |
ac27a0ec DK |
3681 | int start; |
3682 | ||
3683 | block_group = (inode->i_ino - 1) / | |
617ba13b | 3684 | EXT4_INODES_PER_GROUP(inode->i_sb); |
ac27a0ec | 3685 | inodes_per_buffer = bh->b_size / |
617ba13b | 3686 | EXT4_INODE_SIZE(inode->i_sb); |
ac27a0ec | 3687 | inode_offset = ((inode->i_ino - 1) % |
617ba13b | 3688 | EXT4_INODES_PER_GROUP(inode->i_sb)); |
ac27a0ec DK |
3689 | start = inode_offset & ~(inodes_per_buffer - 1); |
3690 | ||
3691 | /* Is the inode bitmap in cache? */ | |
617ba13b | 3692 | desc = ext4_get_group_desc(inode->i_sb, |
ac27a0ec DK |
3693 | block_group, NULL); |
3694 | if (!desc) | |
3695 | goto make_io; | |
3696 | ||
3697 | bitmap_bh = sb_getblk(inode->i_sb, | |
8fadc143 | 3698 | ext4_inode_bitmap(inode->i_sb, desc)); |
ac27a0ec DK |
3699 | if (!bitmap_bh) |
3700 | goto make_io; | |
3701 | ||
3702 | /* | |
3703 | * If the inode bitmap isn't in cache then the | |
3704 | * optimisation may end up performing two reads instead | |
3705 | * of one, so skip it. | |
3706 | */ | |
3707 | if (!buffer_uptodate(bitmap_bh)) { | |
3708 | brelse(bitmap_bh); | |
3709 | goto make_io; | |
3710 | } | |
3711 | for (i = start; i < start + inodes_per_buffer; i++) { | |
3712 | if (i == inode_offset) | |
3713 | continue; | |
617ba13b | 3714 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3715 | break; |
3716 | } | |
3717 | brelse(bitmap_bh); | |
3718 | if (i == start + inodes_per_buffer) { | |
3719 | /* all other inodes are free, so skip I/O */ | |
3720 | memset(bh->b_data, 0, bh->b_size); | |
3721 | set_buffer_uptodate(bh); | |
3722 | unlock_buffer(bh); | |
3723 | goto has_buffer; | |
3724 | } | |
3725 | } | |
3726 | ||
3727 | make_io: | |
3728 | /* | |
3729 | * There are other valid inodes in the buffer, this inode | |
3730 | * has in-inode xattrs, or we don't have this inode in memory. | |
3731 | * Read the block from disk. | |
3732 | */ | |
3733 | get_bh(bh); | |
3734 | bh->b_end_io = end_buffer_read_sync; | |
3735 | submit_bh(READ_META, bh); | |
3736 | wait_on_buffer(bh); | |
3737 | if (!buffer_uptodate(bh)) { | |
617ba13b | 3738 | ext4_error(inode->i_sb, "ext4_get_inode_loc", |
ac27a0ec | 3739 | "unable to read inode block - " |
2ae02107 | 3740 | "inode=%lu, block=%llu", |
ac27a0ec DK |
3741 | inode->i_ino, block); |
3742 | brelse(bh); | |
3743 | return -EIO; | |
3744 | } | |
3745 | } | |
3746 | has_buffer: | |
3747 | iloc->bh = bh; | |
3748 | return 0; | |
3749 | } | |
3750 | ||
617ba13b | 3751 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3752 | { |
3753 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b MC |
3754 | return __ext4_get_inode_loc(inode, iloc, |
3755 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); | |
ac27a0ec DK |
3756 | } |
3757 | ||
617ba13b | 3758 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3759 | { |
617ba13b | 3760 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
3761 | |
3762 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 3763 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 3764 | inode->i_flags |= S_SYNC; |
617ba13b | 3765 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 3766 | inode->i_flags |= S_APPEND; |
617ba13b | 3767 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 3768 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 3769 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 3770 | inode->i_flags |= S_NOATIME; |
617ba13b | 3771 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
3772 | inode->i_flags |= S_DIRSYNC; |
3773 | } | |
3774 | ||
ff9ddf7e JK |
3775 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3776 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3777 | { | |
3778 | unsigned int flags = ei->vfs_inode.i_flags; | |
3779 | ||
3780 | ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3781 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL); | |
3782 | if (flags & S_SYNC) | |
3783 | ei->i_flags |= EXT4_SYNC_FL; | |
3784 | if (flags & S_APPEND) | |
3785 | ei->i_flags |= EXT4_APPEND_FL; | |
3786 | if (flags & S_IMMUTABLE) | |
3787 | ei->i_flags |= EXT4_IMMUTABLE_FL; | |
3788 | if (flags & S_NOATIME) | |
3789 | ei->i_flags |= EXT4_NOATIME_FL; | |
3790 | if (flags & S_DIRSYNC) | |
3791 | ei->i_flags |= EXT4_DIRSYNC_FL; | |
3792 | } | |
0fc1b451 AK |
3793 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
3794 | struct ext4_inode_info *ei) | |
3795 | { | |
3796 | blkcnt_t i_blocks ; | |
8180a562 AK |
3797 | struct inode *inode = &(ei->vfs_inode); |
3798 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3799 | |
3800 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3801 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3802 | /* we are using combined 48 bit field */ | |
3803 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3804 | le32_to_cpu(raw_inode->i_blocks_lo); | |
8180a562 AK |
3805 | if (ei->i_flags & EXT4_HUGE_FILE_FL) { |
3806 | /* i_blocks represent file system block size */ | |
3807 | return i_blocks << (inode->i_blkbits - 9); | |
3808 | } else { | |
3809 | return i_blocks; | |
3810 | } | |
0fc1b451 AK |
3811 | } else { |
3812 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
3813 | } | |
3814 | } | |
ff9ddf7e | 3815 | |
1d1fe1ee | 3816 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 3817 | { |
617ba13b MC |
3818 | struct ext4_iloc iloc; |
3819 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 3820 | struct ext4_inode_info *ei; |
ac27a0ec | 3821 | struct buffer_head *bh; |
1d1fe1ee DH |
3822 | struct inode *inode; |
3823 | long ret; | |
ac27a0ec DK |
3824 | int block; |
3825 | ||
1d1fe1ee DH |
3826 | inode = iget_locked(sb, ino); |
3827 | if (!inode) | |
3828 | return ERR_PTR(-ENOMEM); | |
3829 | if (!(inode->i_state & I_NEW)) | |
3830 | return inode; | |
3831 | ||
3832 | ei = EXT4_I(inode); | |
617ba13b MC |
3833 | #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL |
3834 | ei->i_acl = EXT4_ACL_NOT_CACHED; | |
3835 | ei->i_default_acl = EXT4_ACL_NOT_CACHED; | |
ac27a0ec DK |
3836 | #endif |
3837 | ei->i_block_alloc_info = NULL; | |
3838 | ||
1d1fe1ee DH |
3839 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
3840 | if (ret < 0) | |
ac27a0ec DK |
3841 | goto bad_inode; |
3842 | bh = iloc.bh; | |
617ba13b | 3843 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
3844 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
3845 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
3846 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
3847 | if(!(test_opt (inode->i_sb, NO_UID32))) { | |
3848 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; | |
3849 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
3850 | } | |
3851 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec DK |
3852 | |
3853 | ei->i_state = 0; | |
3854 | ei->i_dir_start_lookup = 0; | |
3855 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
3856 | /* We now have enough fields to check if the inode was active or not. | |
3857 | * This is needed because nfsd might try to access dead inodes | |
3858 | * the test is that same one that e2fsck uses | |
3859 | * NeilBrown 1999oct15 | |
3860 | */ | |
3861 | if (inode->i_nlink == 0) { | |
3862 | if (inode->i_mode == 0 || | |
617ba13b | 3863 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec DK |
3864 | /* this inode is deleted */ |
3865 | brelse (bh); | |
1d1fe1ee | 3866 | ret = -ESTALE; |
ac27a0ec DK |
3867 | goto bad_inode; |
3868 | } | |
3869 | /* The only unlinked inodes we let through here have | |
3870 | * valid i_mode and are being read by the orphan | |
3871 | * recovery code: that's fine, we're about to complete | |
3872 | * the process of deleting those. */ | |
3873 | } | |
ac27a0ec | 3874 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 3875 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 3876 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
9b8f1f01 | 3877 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
a48380f7 | 3878 | cpu_to_le32(EXT4_OS_HURD)) { |
a1ddeb7e BP |
3879 | ei->i_file_acl |= |
3880 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
ac27a0ec | 3881 | } |
a48380f7 | 3882 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec DK |
3883 | ei->i_disksize = inode->i_size; |
3884 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
3885 | ei->i_block_group = iloc.block_group; | |
3886 | /* | |
3887 | * NOTE! The in-memory inode i_data array is in little-endian order | |
3888 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
3889 | */ | |
617ba13b | 3890 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
3891 | ei->i_data[block] = raw_inode->i_block[block]; |
3892 | INIT_LIST_HEAD(&ei->i_orphan); | |
3893 | ||
0040d987 | 3894 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 3895 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 3896 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f KK |
3897 | EXT4_INODE_SIZE(inode->i_sb)) { |
3898 | brelse (bh); | |
1d1fe1ee | 3899 | ret = -EIO; |
ac27a0ec | 3900 | goto bad_inode; |
e5d2861f | 3901 | } |
ac27a0ec DK |
3902 | if (ei->i_extra_isize == 0) { |
3903 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
3904 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
3905 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
3906 | } else { |
3907 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 3908 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 3909 | ei->i_extra_isize; |
617ba13b MC |
3910 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
3911 | ei->i_state |= EXT4_STATE_XATTR; | |
ac27a0ec DK |
3912 | } |
3913 | } else | |
3914 | ei->i_extra_isize = 0; | |
3915 | ||
ef7f3835 KS |
3916 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
3917 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
3918 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
3919 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
3920 | ||
25ec56b5 JNC |
3921 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
3922 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3923 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
3924 | inode->i_version |= | |
3925 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
3926 | } | |
3927 | ||
ac27a0ec | 3928 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
3929 | inode->i_op = &ext4_file_inode_operations; |
3930 | inode->i_fop = &ext4_file_operations; | |
3931 | ext4_set_aops(inode); | |
ac27a0ec | 3932 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
3933 | inode->i_op = &ext4_dir_inode_operations; |
3934 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 3935 | } else if (S_ISLNK(inode->i_mode)) { |
617ba13b MC |
3936 | if (ext4_inode_is_fast_symlink(inode)) |
3937 | inode->i_op = &ext4_fast_symlink_inode_operations; | |
ac27a0ec | 3938 | else { |
617ba13b MC |
3939 | inode->i_op = &ext4_symlink_inode_operations; |
3940 | ext4_set_aops(inode); | |
ac27a0ec DK |
3941 | } |
3942 | } else { | |
617ba13b | 3943 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
3944 | if (raw_inode->i_block[0]) |
3945 | init_special_inode(inode, inode->i_mode, | |
3946 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
3947 | else | |
3948 | init_special_inode(inode, inode->i_mode, | |
3949 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
3950 | } | |
3951 | brelse (iloc.bh); | |
617ba13b | 3952 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
3953 | unlock_new_inode(inode); |
3954 | return inode; | |
ac27a0ec DK |
3955 | |
3956 | bad_inode: | |
1d1fe1ee DH |
3957 | iget_failed(inode); |
3958 | return ERR_PTR(ret); | |
ac27a0ec DK |
3959 | } |
3960 | ||
0fc1b451 AK |
3961 | static int ext4_inode_blocks_set(handle_t *handle, |
3962 | struct ext4_inode *raw_inode, | |
3963 | struct ext4_inode_info *ei) | |
3964 | { | |
3965 | struct inode *inode = &(ei->vfs_inode); | |
3966 | u64 i_blocks = inode->i_blocks; | |
3967 | struct super_block *sb = inode->i_sb; | |
3968 | int err = 0; | |
3969 | ||
3970 | if (i_blocks <= ~0U) { | |
3971 | /* | |
3972 | * i_blocks can be represnted in a 32 bit variable | |
3973 | * as multiple of 512 bytes | |
3974 | */ | |
8180a562 | 3975 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3976 | raw_inode->i_blocks_high = 0; |
8180a562 | 3977 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
0fc1b451 AK |
3978 | } else if (i_blocks <= 0xffffffffffffULL) { |
3979 | /* | |
3980 | * i_blocks can be represented in a 48 bit variable | |
3981 | * as multiple of 512 bytes | |
3982 | */ | |
3983 | err = ext4_update_rocompat_feature(handle, sb, | |
3984 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE); | |
3985 | if (err) | |
3986 | goto err_out; | |
3987 | /* i_block is stored in the split 48 bit fields */ | |
8180a562 | 3988 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3989 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
8180a562 | 3990 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
0fc1b451 | 3991 | } else { |
8180a562 AK |
3992 | /* |
3993 | * i_blocks should be represented in a 48 bit variable | |
3994 | * as multiple of file system block size | |
3995 | */ | |
3996 | err = ext4_update_rocompat_feature(handle, sb, | |
3997 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE); | |
3998 | if (err) | |
3999 | goto err_out; | |
4000 | ei->i_flags |= EXT4_HUGE_FILE_FL; | |
4001 | /* i_block is stored in file system block size */ | |
4002 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4003 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4004 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 AK |
4005 | } |
4006 | err_out: | |
4007 | return err; | |
4008 | } | |
4009 | ||
ac27a0ec DK |
4010 | /* |
4011 | * Post the struct inode info into an on-disk inode location in the | |
4012 | * buffer-cache. This gobbles the caller's reference to the | |
4013 | * buffer_head in the inode location struct. | |
4014 | * | |
4015 | * The caller must have write access to iloc->bh. | |
4016 | */ | |
617ba13b | 4017 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4018 | struct inode *inode, |
617ba13b | 4019 | struct ext4_iloc *iloc) |
ac27a0ec | 4020 | { |
617ba13b MC |
4021 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4022 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4023 | struct buffer_head *bh = iloc->bh; |
4024 | int err = 0, rc, block; | |
4025 | ||
4026 | /* For fields not not tracking in the in-memory inode, | |
4027 | * initialise them to zero for new inodes. */ | |
617ba13b MC |
4028 | if (ei->i_state & EXT4_STATE_NEW) |
4029 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
ac27a0ec | 4030 | |
ff9ddf7e | 4031 | ext4_get_inode_flags(ei); |
ac27a0ec DK |
4032 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
4033 | if(!(test_opt(inode->i_sb, NO_UID32))) { | |
4034 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); | |
4035 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
4036 | /* | |
4037 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4038 | * re-used with the upper 16 bits of the uid/gid intact | |
4039 | */ | |
4040 | if(!ei->i_dtime) { | |
4041 | raw_inode->i_uid_high = | |
4042 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
4043 | raw_inode->i_gid_high = | |
4044 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
4045 | } else { | |
4046 | raw_inode->i_uid_high = 0; | |
4047 | raw_inode->i_gid_high = 0; | |
4048 | } | |
4049 | } else { | |
4050 | raw_inode->i_uid_low = | |
4051 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
4052 | raw_inode->i_gid_low = | |
4053 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
4054 | raw_inode->i_uid_high = 0; | |
4055 | raw_inode->i_gid_high = 0; | |
4056 | } | |
4057 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4058 | |
4059 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4060 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4061 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4062 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4063 | ||
0fc1b451 AK |
4064 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4065 | goto out_brelse; | |
ac27a0ec | 4066 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
267e4db9 AK |
4067 | /* clear the migrate flag in the raw_inode */ |
4068 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE); | |
9b8f1f01 MC |
4069 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4070 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4071 | raw_inode->i_file_acl_high = |
4072 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4073 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
4074 | ext4_isize_set(raw_inode, ei->i_disksize); |
4075 | if (ei->i_disksize > 0x7fffffffULL) { | |
4076 | struct super_block *sb = inode->i_sb; | |
4077 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4078 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4079 | EXT4_SB(sb)->s_es->s_rev_level == | |
4080 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4081 | /* If this is the first large file | |
4082 | * created, add a flag to the superblock. | |
4083 | */ | |
4084 | err = ext4_journal_get_write_access(handle, | |
4085 | EXT4_SB(sb)->s_sbh); | |
4086 | if (err) | |
4087 | goto out_brelse; | |
4088 | ext4_update_dynamic_rev(sb); | |
4089 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4090 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 AK |
4091 | sb->s_dirt = 1; |
4092 | handle->h_sync = 1; | |
4093 | err = ext4_journal_dirty_metadata(handle, | |
4094 | EXT4_SB(sb)->s_sbh); | |
ac27a0ec DK |
4095 | } |
4096 | } | |
4097 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4098 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4099 | if (old_valid_dev(inode->i_rdev)) { | |
4100 | raw_inode->i_block[0] = | |
4101 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4102 | raw_inode->i_block[1] = 0; | |
4103 | } else { | |
4104 | raw_inode->i_block[0] = 0; | |
4105 | raw_inode->i_block[1] = | |
4106 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4107 | raw_inode->i_block[2] = 0; | |
4108 | } | |
617ba13b | 4109 | } else for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4110 | raw_inode->i_block[block] = ei->i_data[block]; |
4111 | ||
25ec56b5 JNC |
4112 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4113 | if (ei->i_extra_isize) { | |
4114 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4115 | raw_inode->i_version_hi = | |
4116 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4117 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4118 | } |
4119 | ||
ac27a0ec | 4120 | |
617ba13b MC |
4121 | BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); |
4122 | rc = ext4_journal_dirty_metadata(handle, bh); | |
ac27a0ec DK |
4123 | if (!err) |
4124 | err = rc; | |
617ba13b | 4125 | ei->i_state &= ~EXT4_STATE_NEW; |
ac27a0ec DK |
4126 | |
4127 | out_brelse: | |
4128 | brelse (bh); | |
617ba13b | 4129 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4130 | return err; |
4131 | } | |
4132 | ||
4133 | /* | |
617ba13b | 4134 | * ext4_write_inode() |
ac27a0ec DK |
4135 | * |
4136 | * We are called from a few places: | |
4137 | * | |
4138 | * - Within generic_file_write() for O_SYNC files. | |
4139 | * Here, there will be no transaction running. We wait for any running | |
4140 | * trasnaction to commit. | |
4141 | * | |
4142 | * - Within sys_sync(), kupdate and such. | |
4143 | * We wait on commit, if tol to. | |
4144 | * | |
4145 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4146 | * Here we simply return. We can't afford to block kswapd on the | |
4147 | * journal commit. | |
4148 | * | |
4149 | * In all cases it is actually safe for us to return without doing anything, | |
4150 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4151 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4152 | * knfsd. |
4153 | * | |
4154 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4155 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4156 | * which we are interested. | |
4157 | * | |
4158 | * It would be a bug for them to not do this. The code: | |
4159 | * | |
4160 | * mark_inode_dirty(inode) | |
4161 | * stuff(); | |
4162 | * inode->i_size = expr; | |
4163 | * | |
4164 | * is in error because a kswapd-driven write_inode() could occur while | |
4165 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4166 | * will no longer be on the superblock's dirty inode list. | |
4167 | */ | |
617ba13b | 4168 | int ext4_write_inode(struct inode *inode, int wait) |
ac27a0ec DK |
4169 | { |
4170 | if (current->flags & PF_MEMALLOC) | |
4171 | return 0; | |
4172 | ||
617ba13b | 4173 | if (ext4_journal_current_handle()) { |
b38bd33a | 4174 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
ac27a0ec DK |
4175 | dump_stack(); |
4176 | return -EIO; | |
4177 | } | |
4178 | ||
4179 | if (!wait) | |
4180 | return 0; | |
4181 | ||
617ba13b | 4182 | return ext4_force_commit(inode->i_sb); |
ac27a0ec DK |
4183 | } |
4184 | ||
4185 | /* | |
617ba13b | 4186 | * ext4_setattr() |
ac27a0ec DK |
4187 | * |
4188 | * Called from notify_change. | |
4189 | * | |
4190 | * We want to trap VFS attempts to truncate the file as soon as | |
4191 | * possible. In particular, we want to make sure that when the VFS | |
4192 | * shrinks i_size, we put the inode on the orphan list and modify | |
4193 | * i_disksize immediately, so that during the subsequent flushing of | |
4194 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4195 | * commit will leave the blocks being flushed in an unused state on | |
4196 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4197 | * be freed, so we have a strong guarantee that no future commit will | |
4198 | * leave these blocks visible to the user.) | |
4199 | * | |
678aaf48 JK |
4200 | * Another thing we have to assure is that if we are in ordered mode |
4201 | * and inode is still attached to the committing transaction, we must | |
4202 | * we start writeout of all the dirty pages which are being truncated. | |
4203 | * This way we are sure that all the data written in the previous | |
4204 | * transaction are already on disk (truncate waits for pages under | |
4205 | * writeback). | |
4206 | * | |
4207 | * Called with inode->i_mutex down. | |
ac27a0ec | 4208 | */ |
617ba13b | 4209 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4210 | { |
4211 | struct inode *inode = dentry->d_inode; | |
4212 | int error, rc = 0; | |
4213 | const unsigned int ia_valid = attr->ia_valid; | |
4214 | ||
4215 | error = inode_change_ok(inode, attr); | |
4216 | if (error) | |
4217 | return error; | |
4218 | ||
4219 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
4220 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
4221 | handle_t *handle; | |
4222 | ||
4223 | /* (user+group)*(old+new) structure, inode write (sb, | |
4224 | * inode block, ? - but truncate inode update has it) */ | |
617ba13b MC |
4225 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
4226 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
ac27a0ec DK |
4227 | if (IS_ERR(handle)) { |
4228 | error = PTR_ERR(handle); | |
4229 | goto err_out; | |
4230 | } | |
4231 | error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0; | |
4232 | if (error) { | |
617ba13b | 4233 | ext4_journal_stop(handle); |
ac27a0ec DK |
4234 | return error; |
4235 | } | |
4236 | /* Update corresponding info in inode so that everything is in | |
4237 | * one transaction */ | |
4238 | if (attr->ia_valid & ATTR_UID) | |
4239 | inode->i_uid = attr->ia_uid; | |
4240 | if (attr->ia_valid & ATTR_GID) | |
4241 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4242 | error = ext4_mark_inode_dirty(handle, inode); |
4243 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4244 | } |
4245 | ||
e2b46574 ES |
4246 | if (attr->ia_valid & ATTR_SIZE) { |
4247 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { | |
4248 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4249 | ||
4250 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { | |
4251 | error = -EFBIG; | |
4252 | goto err_out; | |
4253 | } | |
4254 | } | |
4255 | } | |
4256 | ||
ac27a0ec DK |
4257 | if (S_ISREG(inode->i_mode) && |
4258 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
4259 | handle_t *handle; | |
4260 | ||
617ba13b | 4261 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4262 | if (IS_ERR(handle)) { |
4263 | error = PTR_ERR(handle); | |
4264 | goto err_out; | |
4265 | } | |
4266 | ||
617ba13b MC |
4267 | error = ext4_orphan_add(handle, inode); |
4268 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
4269 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4270 | if (!error) |
4271 | error = rc; | |
617ba13b | 4272 | ext4_journal_stop(handle); |
678aaf48 JK |
4273 | |
4274 | if (ext4_should_order_data(inode)) { | |
4275 | error = ext4_begin_ordered_truncate(inode, | |
4276 | attr->ia_size); | |
4277 | if (error) { | |
4278 | /* Do as much error cleanup as possible */ | |
4279 | handle = ext4_journal_start(inode, 3); | |
4280 | if (IS_ERR(handle)) { | |
4281 | ext4_orphan_del(NULL, inode); | |
4282 | goto err_out; | |
4283 | } | |
4284 | ext4_orphan_del(handle, inode); | |
4285 | ext4_journal_stop(handle); | |
4286 | goto err_out; | |
4287 | } | |
4288 | } | |
ac27a0ec DK |
4289 | } |
4290 | ||
4291 | rc = inode_setattr(inode, attr); | |
4292 | ||
617ba13b | 4293 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
4294 | * transaction handle at all, we need to clean up the in-core |
4295 | * orphan list manually. */ | |
4296 | if (inode->i_nlink) | |
617ba13b | 4297 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4298 | |
4299 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4300 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4301 | |
4302 | err_out: | |
617ba13b | 4303 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4304 | if (!error) |
4305 | error = rc; | |
4306 | return error; | |
4307 | } | |
4308 | ||
3e3398a0 MC |
4309 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4310 | struct kstat *stat) | |
4311 | { | |
4312 | struct inode *inode; | |
4313 | unsigned long delalloc_blocks; | |
4314 | ||
4315 | inode = dentry->d_inode; | |
4316 | generic_fillattr(inode, stat); | |
4317 | ||
4318 | /* | |
4319 | * We can't update i_blocks if the block allocation is delayed | |
4320 | * otherwise in the case of system crash before the real block | |
4321 | * allocation is done, we will have i_blocks inconsistent with | |
4322 | * on-disk file blocks. | |
4323 | * We always keep i_blocks updated together with real | |
4324 | * allocation. But to not confuse with user, stat | |
4325 | * will return the blocks that include the delayed allocation | |
4326 | * blocks for this file. | |
4327 | */ | |
4328 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
4329 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
4330 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
4331 | ||
4332 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4333 | return 0; | |
4334 | } | |
ac27a0ec DK |
4335 | |
4336 | /* | |
4337 | * How many blocks doth make a writepage()? | |
4338 | * | |
4339 | * With N blocks per page, it may be: | |
4340 | * N data blocks | |
4341 | * 2 indirect block | |
4342 | * 2 dindirect | |
4343 | * 1 tindirect | |
4344 | * N+5 bitmap blocks (from the above) | |
4345 | * N+5 group descriptor summary blocks | |
4346 | * 1 inode block | |
4347 | * 1 superblock. | |
617ba13b | 4348 | * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files |
ac27a0ec | 4349 | * |
617ba13b | 4350 | * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS |
ac27a0ec DK |
4351 | * |
4352 | * With ordered or writeback data it's the same, less the N data blocks. | |
4353 | * | |
4354 | * If the inode's direct blocks can hold an integral number of pages then a | |
4355 | * page cannot straddle two indirect blocks, and we can only touch one indirect | |
4356 | * and dindirect block, and the "5" above becomes "3". | |
4357 | * | |
4358 | * This still overestimates under most circumstances. If we were to pass the | |
4359 | * start and end offsets in here as well we could do block_to_path() on each | |
4360 | * block and work out the exact number of indirects which are touched. Pah. | |
4361 | */ | |
4362 | ||
a86c6181 | 4363 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4364 | { |
617ba13b MC |
4365 | int bpp = ext4_journal_blocks_per_page(inode); |
4366 | int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3; | |
ac27a0ec DK |
4367 | int ret; |
4368 | ||
a86c6181 AT |
4369 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
4370 | return ext4_ext_writepage_trans_blocks(inode, bpp); | |
4371 | ||
617ba13b | 4372 | if (ext4_should_journal_data(inode)) |
ac27a0ec DK |
4373 | ret = 3 * (bpp + indirects) + 2; |
4374 | else | |
4375 | ret = 2 * (bpp + indirects) + 2; | |
4376 | ||
4377 | #ifdef CONFIG_QUOTA | |
4378 | /* We know that structure was already allocated during DQUOT_INIT so | |
4379 | * we will be updating only the data blocks + inodes */ | |
617ba13b | 4380 | ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); |
ac27a0ec DK |
4381 | #endif |
4382 | ||
4383 | return ret; | |
4384 | } | |
4385 | ||
4386 | /* | |
617ba13b | 4387 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4388 | * Give this, we know that the caller already has write access to iloc->bh. |
4389 | */ | |
617ba13b MC |
4390 | int ext4_mark_iloc_dirty(handle_t *handle, |
4391 | struct inode *inode, struct ext4_iloc *iloc) | |
ac27a0ec DK |
4392 | { |
4393 | int err = 0; | |
4394 | ||
25ec56b5 JNC |
4395 | if (test_opt(inode->i_sb, I_VERSION)) |
4396 | inode_inc_iversion(inode); | |
4397 | ||
ac27a0ec DK |
4398 | /* the do_update_inode consumes one bh->b_count */ |
4399 | get_bh(iloc->bh); | |
4400 | ||
dab291af | 4401 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
617ba13b | 4402 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4403 | put_bh(iloc->bh); |
4404 | return err; | |
4405 | } | |
4406 | ||
4407 | /* | |
4408 | * On success, We end up with an outstanding reference count against | |
4409 | * iloc->bh. This _must_ be cleaned up later. | |
4410 | */ | |
4411 | ||
4412 | int | |
617ba13b MC |
4413 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4414 | struct ext4_iloc *iloc) | |
ac27a0ec DK |
4415 | { |
4416 | int err = 0; | |
4417 | if (handle) { | |
617ba13b | 4418 | err = ext4_get_inode_loc(inode, iloc); |
ac27a0ec DK |
4419 | if (!err) { |
4420 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
617ba13b | 4421 | err = ext4_journal_get_write_access(handle, iloc->bh); |
ac27a0ec DK |
4422 | if (err) { |
4423 | brelse(iloc->bh); | |
4424 | iloc->bh = NULL; | |
4425 | } | |
4426 | } | |
4427 | } | |
617ba13b | 4428 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4429 | return err; |
4430 | } | |
4431 | ||
6dd4ee7c KS |
4432 | /* |
4433 | * Expand an inode by new_extra_isize bytes. | |
4434 | * Returns 0 on success or negative error number on failure. | |
4435 | */ | |
1d03ec98 AK |
4436 | static int ext4_expand_extra_isize(struct inode *inode, |
4437 | unsigned int new_extra_isize, | |
4438 | struct ext4_iloc iloc, | |
4439 | handle_t *handle) | |
6dd4ee7c KS |
4440 | { |
4441 | struct ext4_inode *raw_inode; | |
4442 | struct ext4_xattr_ibody_header *header; | |
4443 | struct ext4_xattr_entry *entry; | |
4444 | ||
4445 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4446 | return 0; | |
4447 | ||
4448 | raw_inode = ext4_raw_inode(&iloc); | |
4449 | ||
4450 | header = IHDR(inode, raw_inode); | |
4451 | entry = IFIRST(header); | |
4452 | ||
4453 | /* No extended attributes present */ | |
4454 | if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) || | |
4455 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
4456 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, | |
4457 | new_extra_isize); | |
4458 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4459 | return 0; | |
4460 | } | |
4461 | ||
4462 | /* try to expand with EAs present */ | |
4463 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4464 | raw_inode, handle); | |
4465 | } | |
4466 | ||
ac27a0ec DK |
4467 | /* |
4468 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4469 | * dirtiness (it may still be data-dirty). | |
4470 | * This means that the in-core inode may be reaped by prune_icache | |
4471 | * without having to perform any I/O. This is a very good thing, | |
4472 | * because *any* task may call prune_icache - even ones which | |
4473 | * have a transaction open against a different journal. | |
4474 | * | |
4475 | * Is this cheating? Not really. Sure, we haven't written the | |
4476 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4477 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4478 | * we start and wait on commits. | |
4479 | * | |
4480 | * Is this efficient/effective? Well, we're being nice to the system | |
4481 | * by cleaning up our inodes proactively so they can be reaped | |
4482 | * without I/O. But we are potentially leaving up to five seconds' | |
4483 | * worth of inodes floating about which prune_icache wants us to | |
4484 | * write out. One way to fix that would be to get prune_icache() | |
4485 | * to do a write_super() to free up some memory. It has the desired | |
4486 | * effect. | |
4487 | */ | |
617ba13b | 4488 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4489 | { |
617ba13b | 4490 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4491 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4492 | static unsigned int mnt_count; | |
4493 | int err, ret; | |
ac27a0ec DK |
4494 | |
4495 | might_sleep(); | |
617ba13b | 4496 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
6dd4ee7c KS |
4497 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
4498 | !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { | |
4499 | /* | |
4500 | * We need extra buffer credits since we may write into EA block | |
4501 | * with this same handle. If journal_extend fails, then it will | |
4502 | * only result in a minor loss of functionality for that inode. | |
4503 | * If this is felt to be critical, then e2fsck should be run to | |
4504 | * force a large enough s_min_extra_isize. | |
4505 | */ | |
4506 | if ((jbd2_journal_extend(handle, | |
4507 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4508 | ret = ext4_expand_extra_isize(inode, | |
4509 | sbi->s_want_extra_isize, | |
4510 | iloc, handle); | |
4511 | if (ret) { | |
4512 | EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND; | |
c1bddad9 AK |
4513 | if (mnt_count != |
4514 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
46e665e9 | 4515 | ext4_warning(inode->i_sb, __func__, |
6dd4ee7c KS |
4516 | "Unable to expand inode %lu. Delete" |
4517 | " some EAs or run e2fsck.", | |
4518 | inode->i_ino); | |
c1bddad9 AK |
4519 | mnt_count = |
4520 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4521 | } |
4522 | } | |
4523 | } | |
4524 | } | |
ac27a0ec | 4525 | if (!err) |
617ba13b | 4526 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4527 | return err; |
4528 | } | |
4529 | ||
4530 | /* | |
617ba13b | 4531 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4532 | * |
4533 | * We're really interested in the case where a file is being extended. | |
4534 | * i_size has been changed by generic_commit_write() and we thus need | |
4535 | * to include the updated inode in the current transaction. | |
4536 | * | |
4537 | * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks | |
4538 | * are allocated to the file. | |
4539 | * | |
4540 | * If the inode is marked synchronous, we don't honour that here - doing | |
4541 | * so would cause a commit on atime updates, which we don't bother doing. | |
4542 | * We handle synchronous inodes at the highest possible level. | |
4543 | */ | |
617ba13b | 4544 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 4545 | { |
617ba13b | 4546 | handle_t *current_handle = ext4_journal_current_handle(); |
ac27a0ec DK |
4547 | handle_t *handle; |
4548 | ||
617ba13b | 4549 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
4550 | if (IS_ERR(handle)) |
4551 | goto out; | |
4552 | if (current_handle && | |
4553 | current_handle->h_transaction != handle->h_transaction) { | |
4554 | /* This task has a transaction open against a different fs */ | |
4555 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
46e665e9 | 4556 | __func__); |
ac27a0ec DK |
4557 | } else { |
4558 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
4559 | current_handle); | |
617ba13b | 4560 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 4561 | } |
617ba13b | 4562 | ext4_journal_stop(handle); |
ac27a0ec DK |
4563 | out: |
4564 | return; | |
4565 | } | |
4566 | ||
4567 | #if 0 | |
4568 | /* | |
4569 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4570 | * it from being flushed to disk early. Unlike | |
617ba13b | 4571 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4572 | * returns no iloc structure, so the caller needs to repeat the iloc |
4573 | * lookup to mark the inode dirty later. | |
4574 | */ | |
617ba13b | 4575 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 4576 | { |
617ba13b | 4577 | struct ext4_iloc iloc; |
ac27a0ec DK |
4578 | |
4579 | int err = 0; | |
4580 | if (handle) { | |
617ba13b | 4581 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
4582 | if (!err) { |
4583 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 4584 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 4585 | if (!err) |
617ba13b | 4586 | err = ext4_journal_dirty_metadata(handle, |
ac27a0ec DK |
4587 | iloc.bh); |
4588 | brelse(iloc.bh); | |
4589 | } | |
4590 | } | |
617ba13b | 4591 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4592 | return err; |
4593 | } | |
4594 | #endif | |
4595 | ||
617ba13b | 4596 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
4597 | { |
4598 | journal_t *journal; | |
4599 | handle_t *handle; | |
4600 | int err; | |
4601 | ||
4602 | /* | |
4603 | * We have to be very careful here: changing a data block's | |
4604 | * journaling status dynamically is dangerous. If we write a | |
4605 | * data block to the journal, change the status and then delete | |
4606 | * that block, we risk forgetting to revoke the old log record | |
4607 | * from the journal and so a subsequent replay can corrupt data. | |
4608 | * So, first we make sure that the journal is empty and that | |
4609 | * nobody is changing anything. | |
4610 | */ | |
4611 | ||
617ba13b | 4612 | journal = EXT4_JOURNAL(inode); |
d699594d | 4613 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
4614 | return -EROFS; |
4615 | ||
dab291af MC |
4616 | jbd2_journal_lock_updates(journal); |
4617 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
4618 | |
4619 | /* | |
4620 | * OK, there are no updates running now, and all cached data is | |
4621 | * synced to disk. We are now in a completely consistent state | |
4622 | * which doesn't have anything in the journal, and we know that | |
4623 | * no filesystem updates are running, so it is safe to modify | |
4624 | * the inode's in-core data-journaling state flag now. | |
4625 | */ | |
4626 | ||
4627 | if (val) | |
617ba13b | 4628 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
ac27a0ec | 4629 | else |
617ba13b MC |
4630 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
4631 | ext4_set_aops(inode); | |
ac27a0ec | 4632 | |
dab291af | 4633 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
4634 | |
4635 | /* Finally we can mark the inode as dirty. */ | |
4636 | ||
617ba13b | 4637 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
4638 | if (IS_ERR(handle)) |
4639 | return PTR_ERR(handle); | |
4640 | ||
617ba13b | 4641 | err = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 4642 | handle->h_sync = 1; |
617ba13b MC |
4643 | ext4_journal_stop(handle); |
4644 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
4645 | |
4646 | return err; | |
4647 | } | |
2e9ee850 AK |
4648 | |
4649 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
4650 | { | |
4651 | return !buffer_mapped(bh); | |
4652 | } | |
4653 | ||
4654 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page) | |
4655 | { | |
4656 | loff_t size; | |
4657 | unsigned long len; | |
4658 | int ret = -EINVAL; | |
4659 | struct file *file = vma->vm_file; | |
4660 | struct inode *inode = file->f_path.dentry->d_inode; | |
4661 | struct address_space *mapping = inode->i_mapping; | |
4662 | ||
4663 | /* | |
4664 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
4665 | * get i_mutex because we are already holding mmap_sem. | |
4666 | */ | |
4667 | down_read(&inode->i_alloc_sem); | |
4668 | size = i_size_read(inode); | |
4669 | if (page->mapping != mapping || size <= page_offset(page) | |
4670 | || !PageUptodate(page)) { | |
4671 | /* page got truncated from under us? */ | |
4672 | goto out_unlock; | |
4673 | } | |
4674 | ret = 0; | |
4675 | if (PageMappedToDisk(page)) | |
4676 | goto out_unlock; | |
4677 | ||
4678 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
4679 | len = size & ~PAGE_CACHE_MASK; | |
4680 | else | |
4681 | len = PAGE_CACHE_SIZE; | |
4682 | ||
4683 | if (page_has_buffers(page)) { | |
4684 | /* return if we have all the buffers mapped */ | |
4685 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
4686 | ext4_bh_unmapped)) | |
4687 | goto out_unlock; | |
4688 | } | |
4689 | /* | |
4690 | * OK, we need to fill the hole... Do write_begin write_end | |
4691 | * to do block allocation/reservation.We are not holding | |
4692 | * inode.i__mutex here. That allow * parallel write_begin, | |
4693 | * write_end call. lock_page prevent this from happening | |
4694 | * on the same page though | |
4695 | */ | |
4696 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
4697 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL); | |
4698 | if (ret < 0) | |
4699 | goto out_unlock; | |
4700 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
4701 | len, len, page, NULL); | |
4702 | if (ret < 0) | |
4703 | goto out_unlock; | |
4704 | ret = 0; | |
4705 | out_unlock: | |
4706 | up_read(&inode->i_alloc_sem); | |
4707 | return ret; | |
4708 | } |