]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Goal-directed block allocation by Stephen Tweedie | |
16 | * (sct@redhat.com), 1993, 1998 | |
17 | * Big-endian to little-endian byte-swapping/bitmaps by | |
18 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
20 | * (jj@sunsite.ms.mff.cuni.cz) | |
21 | * | |
617ba13b | 22 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
23 | */ |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
dab291af | 28 | #include <linux/jbd2.h> |
ac27a0ec DK |
29 | #include <linux/highuid.h> |
30 | #include <linux/pagemap.h> | |
31 | #include <linux/quotaops.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/buffer_head.h> | |
34 | #include <linux/writeback.h> | |
64769240 | 35 | #include <linux/pagevec.h> |
ac27a0ec | 36 | #include <linux/mpage.h> |
e83c1397 | 37 | #include <linux/namei.h> |
ac27a0ec DK |
38 | #include <linux/uio.h> |
39 | #include <linux/bio.h> | |
9bffad1e | 40 | |
3dcf5451 | 41 | #include "ext4_jbd2.h" |
ac27a0ec DK |
42 | #include "xattr.h" |
43 | #include "acl.h" | |
d2a17637 | 44 | #include "ext4_extents.h" |
ac27a0ec | 45 | |
9bffad1e TT |
46 | #include <trace/events/ext4.h> |
47 | ||
a1d6cc56 AK |
48 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
49 | ||
678aaf48 JK |
50 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
51 | loff_t new_size) | |
52 | { | |
7f5aa215 JK |
53 | return jbd2_journal_begin_ordered_truncate( |
54 | EXT4_SB(inode->i_sb)->s_journal, | |
55 | &EXT4_I(inode)->jinode, | |
56 | new_size); | |
678aaf48 JK |
57 | } |
58 | ||
64769240 AT |
59 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
60 | ||
ac27a0ec DK |
61 | /* |
62 | * Test whether an inode is a fast symlink. | |
63 | */ | |
617ba13b | 64 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 65 | { |
617ba13b | 66 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
67 | (inode->i_sb->s_blocksize >> 9) : 0; |
68 | ||
69 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
70 | } | |
71 | ||
72 | /* | |
617ba13b | 73 | * The ext4 forget function must perform a revoke if we are freeing data |
ac27a0ec DK |
74 | * which has been journaled. Metadata (eg. indirect blocks) must be |
75 | * revoked in all cases. | |
76 | * | |
77 | * "bh" may be NULL: a metadata block may have been freed from memory | |
78 | * but there may still be a record of it in the journal, and that record | |
79 | * still needs to be revoked. | |
0390131b FM |
80 | * |
81 | * If the handle isn't valid we're not journaling so there's nothing to do. | |
ac27a0ec | 82 | */ |
617ba13b | 83 | int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, |
de9a55b8 | 84 | struct buffer_head *bh, ext4_fsblk_t blocknr) |
ac27a0ec DK |
85 | { |
86 | int err; | |
87 | ||
0390131b FM |
88 | if (!ext4_handle_valid(handle)) |
89 | return 0; | |
90 | ||
ac27a0ec DK |
91 | might_sleep(); |
92 | ||
93 | BUFFER_TRACE(bh, "enter"); | |
94 | ||
95 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " | |
7f4520cc | 96 | "data mode %x\n", |
ac27a0ec DK |
97 | bh, is_metadata, inode->i_mode, |
98 | test_opt(inode->i_sb, DATA_FLAGS)); | |
99 | ||
100 | /* Never use the revoke function if we are doing full data | |
101 | * journaling: there is no need to, and a V1 superblock won't | |
102 | * support it. Otherwise, only skip the revoke on un-journaled | |
103 | * data blocks. */ | |
104 | ||
617ba13b MC |
105 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA || |
106 | (!is_metadata && !ext4_should_journal_data(inode))) { | |
ac27a0ec | 107 | if (bh) { |
dab291af | 108 | BUFFER_TRACE(bh, "call jbd2_journal_forget"); |
617ba13b | 109 | return ext4_journal_forget(handle, bh); |
ac27a0ec DK |
110 | } |
111 | return 0; | |
112 | } | |
113 | ||
114 | /* | |
115 | * data!=journal && (is_metadata || should_journal_data(inode)) | |
116 | */ | |
617ba13b MC |
117 | BUFFER_TRACE(bh, "call ext4_journal_revoke"); |
118 | err = ext4_journal_revoke(handle, blocknr, bh); | |
ac27a0ec | 119 | if (err) |
46e665e9 | 120 | ext4_abort(inode->i_sb, __func__, |
ac27a0ec DK |
121 | "error %d when attempting revoke", err); |
122 | BUFFER_TRACE(bh, "exit"); | |
123 | return err; | |
124 | } | |
125 | ||
126 | /* | |
127 | * Work out how many blocks we need to proceed with the next chunk of a | |
128 | * truncate transaction. | |
129 | */ | |
130 | static unsigned long blocks_for_truncate(struct inode *inode) | |
131 | { | |
725d26d3 | 132 | ext4_lblk_t needed; |
ac27a0ec DK |
133 | |
134 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
135 | ||
136 | /* Give ourselves just enough room to cope with inodes in which | |
137 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
138 | * which resulted in random data in an inode which looked enough | |
617ba13b | 139 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
140 | * will go a bit crazy if that happens, but at least we should |
141 | * try not to panic the whole kernel. */ | |
142 | if (needed < 2) | |
143 | needed = 2; | |
144 | ||
145 | /* But we need to bound the transaction so we don't overflow the | |
146 | * journal. */ | |
617ba13b MC |
147 | if (needed > EXT4_MAX_TRANS_DATA) |
148 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 149 | |
617ba13b | 150 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
151 | } |
152 | ||
153 | /* | |
154 | * Truncate transactions can be complex and absolutely huge. So we need to | |
155 | * be able to restart the transaction at a conventient checkpoint to make | |
156 | * sure we don't overflow the journal. | |
157 | * | |
158 | * start_transaction gets us a new handle for a truncate transaction, | |
159 | * and extend_transaction tries to extend the existing one a bit. If | |
160 | * extend fails, we need to propagate the failure up and restart the | |
161 | * transaction in the top-level truncate loop. --sct | |
162 | */ | |
163 | static handle_t *start_transaction(struct inode *inode) | |
164 | { | |
165 | handle_t *result; | |
166 | ||
617ba13b | 167 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
168 | if (!IS_ERR(result)) |
169 | return result; | |
170 | ||
617ba13b | 171 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
172 | return result; |
173 | } | |
174 | ||
175 | /* | |
176 | * Try to extend this transaction for the purposes of truncation. | |
177 | * | |
178 | * Returns 0 if we managed to create more room. If we can't create more | |
179 | * room, and the transaction must be restarted we return 1. | |
180 | */ | |
181 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
182 | { | |
0390131b FM |
183 | if (!ext4_handle_valid(handle)) |
184 | return 0; | |
185 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
ac27a0ec | 186 | return 0; |
617ba13b | 187 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
188 | return 0; |
189 | return 1; | |
190 | } | |
191 | ||
192 | /* | |
193 | * Restart the transaction associated with *handle. This does a commit, | |
194 | * so before we call here everything must be consistently dirtied against | |
195 | * this transaction. | |
196 | */ | |
617ba13b | 197 | static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) |
ac27a0ec | 198 | { |
0390131b | 199 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 200 | jbd_debug(2, "restarting handle %p\n", handle); |
617ba13b | 201 | return ext4_journal_restart(handle, blocks_for_truncate(inode)); |
ac27a0ec DK |
202 | } |
203 | ||
204 | /* | |
205 | * Called at the last iput() if i_nlink is zero. | |
206 | */ | |
af5bc92d | 207 | void ext4_delete_inode(struct inode *inode) |
ac27a0ec DK |
208 | { |
209 | handle_t *handle; | |
bc965ab3 | 210 | int err; |
ac27a0ec | 211 | |
678aaf48 JK |
212 | if (ext4_should_order_data(inode)) |
213 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
214 | truncate_inode_pages(&inode->i_data, 0); |
215 | ||
216 | if (is_bad_inode(inode)) | |
217 | goto no_delete; | |
218 | ||
bc965ab3 | 219 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 220 | if (IS_ERR(handle)) { |
bc965ab3 | 221 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
222 | /* |
223 | * If we're going to skip the normal cleanup, we still need to | |
224 | * make sure that the in-core orphan linked list is properly | |
225 | * cleaned up. | |
226 | */ | |
617ba13b | 227 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
228 | goto no_delete; |
229 | } | |
230 | ||
231 | if (IS_SYNC(inode)) | |
0390131b | 232 | ext4_handle_sync(handle); |
ac27a0ec | 233 | inode->i_size = 0; |
bc965ab3 TT |
234 | err = ext4_mark_inode_dirty(handle, inode); |
235 | if (err) { | |
236 | ext4_warning(inode->i_sb, __func__, | |
237 | "couldn't mark inode dirty (err %d)", err); | |
238 | goto stop_handle; | |
239 | } | |
ac27a0ec | 240 | if (inode->i_blocks) |
617ba13b | 241 | ext4_truncate(inode); |
bc965ab3 TT |
242 | |
243 | /* | |
244 | * ext4_ext_truncate() doesn't reserve any slop when it | |
245 | * restarts journal transactions; therefore there may not be | |
246 | * enough credits left in the handle to remove the inode from | |
247 | * the orphan list and set the dtime field. | |
248 | */ | |
0390131b | 249 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
250 | err = ext4_journal_extend(handle, 3); |
251 | if (err > 0) | |
252 | err = ext4_journal_restart(handle, 3); | |
253 | if (err != 0) { | |
254 | ext4_warning(inode->i_sb, __func__, | |
255 | "couldn't extend journal (err %d)", err); | |
256 | stop_handle: | |
257 | ext4_journal_stop(handle); | |
258 | goto no_delete; | |
259 | } | |
260 | } | |
261 | ||
ac27a0ec | 262 | /* |
617ba13b | 263 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 264 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 265 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 266 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 267 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
268 | * (Well, we could do this if we need to, but heck - it works) |
269 | */ | |
617ba13b MC |
270 | ext4_orphan_del(handle, inode); |
271 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
272 | |
273 | /* | |
274 | * One subtle ordering requirement: if anything has gone wrong | |
275 | * (transaction abort, IO errors, whatever), then we can still | |
276 | * do these next steps (the fs will already have been marked as | |
277 | * having errors), but we can't free the inode if the mark_dirty | |
278 | * fails. | |
279 | */ | |
617ba13b | 280 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec DK |
281 | /* If that failed, just do the required in-core inode clear. */ |
282 | clear_inode(inode); | |
283 | else | |
617ba13b MC |
284 | ext4_free_inode(handle, inode); |
285 | ext4_journal_stop(handle); | |
ac27a0ec DK |
286 | return; |
287 | no_delete: | |
288 | clear_inode(inode); /* We must guarantee clearing of inode... */ | |
289 | } | |
290 | ||
291 | typedef struct { | |
292 | __le32 *p; | |
293 | __le32 key; | |
294 | struct buffer_head *bh; | |
295 | } Indirect; | |
296 | ||
297 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
298 | { | |
299 | p->key = *(p->p = v); | |
300 | p->bh = bh; | |
301 | } | |
302 | ||
ac27a0ec | 303 | /** |
617ba13b | 304 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
305 | * @inode: inode in question (we are only interested in its superblock) |
306 | * @i_block: block number to be parsed | |
307 | * @offsets: array to store the offsets in | |
8c55e204 DK |
308 | * @boundary: set this non-zero if the referred-to block is likely to be |
309 | * followed (on disk) by an indirect block. | |
ac27a0ec | 310 | * |
617ba13b | 311 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
312 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
313 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
314 | * This function translates the block number into path in that tree - | |
315 | * return value is the path length and @offsets[n] is the offset of | |
316 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
317 | * (negative or too large) warning is printed and zero returned. | |
318 | * | |
319 | * Note: function doesn't find node addresses, so no IO is needed. All | |
320 | * we need to know is the capacity of indirect blocks (taken from the | |
321 | * inode->i_sb). | |
322 | */ | |
323 | ||
324 | /* | |
325 | * Portability note: the last comparison (check that we fit into triple | |
326 | * indirect block) is spelled differently, because otherwise on an | |
327 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
328 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
329 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
330 | * i_block would have to be negative in the very beginning, so we would not | |
331 | * get there at all. | |
332 | */ | |
333 | ||
617ba13b | 334 | static int ext4_block_to_path(struct inode *inode, |
de9a55b8 TT |
335 | ext4_lblk_t i_block, |
336 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 337 | { |
617ba13b MC |
338 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
339 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
340 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
341 | indirect_blocks = ptrs, |
342 | double_blocks = (1 << (ptrs_bits * 2)); | |
343 | int n = 0; | |
344 | int final = 0; | |
345 | ||
346 | if (i_block < 0) { | |
af5bc92d | 347 | ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0"); |
ac27a0ec DK |
348 | } else if (i_block < direct_blocks) { |
349 | offsets[n++] = i_block; | |
350 | final = direct_blocks; | |
af5bc92d | 351 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
617ba13b | 352 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
353 | offsets[n++] = i_block; |
354 | final = ptrs; | |
355 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 356 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
357 | offsets[n++] = i_block >> ptrs_bits; |
358 | offsets[n++] = i_block & (ptrs - 1); | |
359 | final = ptrs; | |
360 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 361 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
362 | offsets[n++] = i_block >> (ptrs_bits * 2); |
363 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
364 | offsets[n++] = i_block & (ptrs - 1); | |
365 | final = ptrs; | |
366 | } else { | |
e2b46574 | 367 | ext4_warning(inode->i_sb, "ext4_block_to_path", |
de9a55b8 TT |
368 | "block %lu > max in inode %lu", |
369 | i_block + direct_blocks + | |
370 | indirect_blocks + double_blocks, inode->i_ino); | |
ac27a0ec DK |
371 | } |
372 | if (boundary) | |
373 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
374 | return n; | |
375 | } | |
376 | ||
fe2c8191 | 377 | static int __ext4_check_blockref(const char *function, struct inode *inode, |
6fd058f7 TT |
378 | __le32 *p, unsigned int max) |
379 | { | |
f73953c0 | 380 | __le32 *bref = p; |
6fd058f7 TT |
381 | unsigned int blk; |
382 | ||
fe2c8191 | 383 | while (bref < p+max) { |
6fd058f7 | 384 | blk = le32_to_cpu(*bref++); |
de9a55b8 TT |
385 | if (blk && |
386 | unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
6fd058f7 | 387 | blk, 1))) { |
fe2c8191 | 388 | ext4_error(inode->i_sb, function, |
6fd058f7 TT |
389 | "invalid block reference %u " |
390 | "in inode #%lu", blk, inode->i_ino); | |
de9a55b8 TT |
391 | return -EIO; |
392 | } | |
393 | } | |
394 | return 0; | |
fe2c8191 TN |
395 | } |
396 | ||
397 | ||
398 | #define ext4_check_indirect_blockref(inode, bh) \ | |
de9a55b8 | 399 | __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \ |
fe2c8191 TN |
400 | EXT4_ADDR_PER_BLOCK((inode)->i_sb)) |
401 | ||
402 | #define ext4_check_inode_blockref(inode) \ | |
de9a55b8 | 403 | __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \ |
fe2c8191 TN |
404 | EXT4_NDIR_BLOCKS) |
405 | ||
ac27a0ec | 406 | /** |
617ba13b | 407 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
408 | * @inode: inode in question |
409 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
410 | * @offsets: offsets of pointers in inode/indirect blocks | |
411 | * @chain: place to store the result | |
412 | * @err: here we store the error value | |
413 | * | |
414 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
415 | * if everything went OK or the pointer to the last filled triple | |
416 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
417 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
418 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
419 | * number (it points into struct inode for i==0 and into the bh->b_data | |
420 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
421 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
422 | * numbers of the chain, addresses they were taken from (and where we can | |
423 | * verify that chain did not change) and buffer_heads hosting these | |
424 | * numbers. | |
425 | * | |
426 | * Function stops when it stumbles upon zero pointer (absent block) | |
427 | * (pointer to last triple returned, *@err == 0) | |
428 | * or when it gets an IO error reading an indirect block | |
429 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
430 | * or when it reads all @depth-1 indirect blocks successfully and finds |
431 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
432 | * |
433 | * Need to be called with | |
0e855ac8 | 434 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 435 | */ |
725d26d3 AK |
436 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
437 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
438 | Indirect chain[4], int *err) |
439 | { | |
440 | struct super_block *sb = inode->i_sb; | |
441 | Indirect *p = chain; | |
442 | struct buffer_head *bh; | |
443 | ||
444 | *err = 0; | |
445 | /* i_data is not going away, no lock needed */ | |
af5bc92d | 446 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
447 | if (!p->key) |
448 | goto no_block; | |
449 | while (--depth) { | |
fe2c8191 TN |
450 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
451 | if (unlikely(!bh)) | |
ac27a0ec | 452 | goto failure; |
de9a55b8 | 453 | |
fe2c8191 TN |
454 | if (!bh_uptodate_or_lock(bh)) { |
455 | if (bh_submit_read(bh) < 0) { | |
456 | put_bh(bh); | |
457 | goto failure; | |
458 | } | |
459 | /* validate block references */ | |
460 | if (ext4_check_indirect_blockref(inode, bh)) { | |
461 | put_bh(bh); | |
462 | goto failure; | |
463 | } | |
464 | } | |
de9a55b8 | 465 | |
af5bc92d | 466 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
ac27a0ec DK |
467 | /* Reader: end */ |
468 | if (!p->key) | |
469 | goto no_block; | |
470 | } | |
471 | return NULL; | |
472 | ||
ac27a0ec DK |
473 | failure: |
474 | *err = -EIO; | |
475 | no_block: | |
476 | return p; | |
477 | } | |
478 | ||
479 | /** | |
617ba13b | 480 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
481 | * @inode: owner |
482 | * @ind: descriptor of indirect block. | |
483 | * | |
1cc8dcf5 | 484 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
485 | * It is used when heuristic for sequential allocation fails. |
486 | * Rules are: | |
487 | * + if there is a block to the left of our position - allocate near it. | |
488 | * + if pointer will live in indirect block - allocate near that block. | |
489 | * + if pointer will live in inode - allocate in the same | |
490 | * cylinder group. | |
491 | * | |
492 | * In the latter case we colour the starting block by the callers PID to | |
493 | * prevent it from clashing with concurrent allocations for a different inode | |
494 | * in the same block group. The PID is used here so that functionally related | |
495 | * files will be close-by on-disk. | |
496 | * | |
497 | * Caller must make sure that @ind is valid and will stay that way. | |
498 | */ | |
617ba13b | 499 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 500 | { |
617ba13b | 501 | struct ext4_inode_info *ei = EXT4_I(inode); |
af5bc92d | 502 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
ac27a0ec | 503 | __le32 *p; |
617ba13b | 504 | ext4_fsblk_t bg_start; |
74d3487f | 505 | ext4_fsblk_t last_block; |
617ba13b | 506 | ext4_grpblk_t colour; |
a4912123 TT |
507 | ext4_group_t block_group; |
508 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
ac27a0ec DK |
509 | |
510 | /* Try to find previous block */ | |
511 | for (p = ind->p - 1; p >= start; p--) { | |
512 | if (*p) | |
513 | return le32_to_cpu(*p); | |
514 | } | |
515 | ||
516 | /* No such thing, so let's try location of indirect block */ | |
517 | if (ind->bh) | |
518 | return ind->bh->b_blocknr; | |
519 | ||
520 | /* | |
521 | * It is going to be referred to from the inode itself? OK, just put it | |
522 | * into the same cylinder group then. | |
523 | */ | |
a4912123 TT |
524 | block_group = ei->i_block_group; |
525 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
526 | block_group &= ~(flex_size-1); | |
527 | if (S_ISREG(inode->i_mode)) | |
528 | block_group++; | |
529 | } | |
530 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
74d3487f VC |
531 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
532 | ||
a4912123 TT |
533 | /* |
534 | * If we are doing delayed allocation, we don't need take | |
535 | * colour into account. | |
536 | */ | |
537 | if (test_opt(inode->i_sb, DELALLOC)) | |
538 | return bg_start; | |
539 | ||
74d3487f VC |
540 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
541 | colour = (current->pid % 16) * | |
617ba13b | 542 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
543 | else |
544 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
545 | return bg_start + colour; |
546 | } | |
547 | ||
548 | /** | |
1cc8dcf5 | 549 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
550 | * @inode: owner |
551 | * @block: block we want | |
ac27a0ec | 552 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 553 | * |
1cc8dcf5 | 554 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 555 | * returns it. |
ac27a0ec | 556 | */ |
725d26d3 | 557 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
de9a55b8 | 558 | Indirect *partial) |
ac27a0ec | 559 | { |
ac27a0ec | 560 | /* |
c2ea3fde | 561 | * XXX need to get goal block from mballoc's data structures |
ac27a0ec | 562 | */ |
ac27a0ec | 563 | |
617ba13b | 564 | return ext4_find_near(inode, partial); |
ac27a0ec DK |
565 | } |
566 | ||
567 | /** | |
617ba13b | 568 | * ext4_blks_to_allocate: Look up the block map and count the number |
ac27a0ec DK |
569 | * of direct blocks need to be allocated for the given branch. |
570 | * | |
571 | * @branch: chain of indirect blocks | |
572 | * @k: number of blocks need for indirect blocks | |
573 | * @blks: number of data blocks to be mapped. | |
574 | * @blocks_to_boundary: the offset in the indirect block | |
575 | * | |
576 | * return the total number of blocks to be allocate, including the | |
577 | * direct and indirect blocks. | |
578 | */ | |
498e5f24 | 579 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
de9a55b8 | 580 | int blocks_to_boundary) |
ac27a0ec | 581 | { |
498e5f24 | 582 | unsigned int count = 0; |
ac27a0ec DK |
583 | |
584 | /* | |
585 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
586 | * then it's clear blocks on that path have not allocated | |
587 | */ | |
588 | if (k > 0) { | |
589 | /* right now we don't handle cross boundary allocation */ | |
590 | if (blks < blocks_to_boundary + 1) | |
591 | count += blks; | |
592 | else | |
593 | count += blocks_to_boundary + 1; | |
594 | return count; | |
595 | } | |
596 | ||
597 | count++; | |
598 | while (count < blks && count <= blocks_to_boundary && | |
599 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
600 | count++; | |
601 | } | |
602 | return count; | |
603 | } | |
604 | ||
605 | /** | |
617ba13b | 606 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
ac27a0ec DK |
607 | * @indirect_blks: the number of blocks need to allocate for indirect |
608 | * blocks | |
609 | * | |
610 | * @new_blocks: on return it will store the new block numbers for | |
611 | * the indirect blocks(if needed) and the first direct block, | |
612 | * @blks: on return it will store the total number of allocated | |
613 | * direct blocks | |
614 | */ | |
617ba13b | 615 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
616 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
617 | int indirect_blks, int blks, | |
618 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec | 619 | { |
815a1130 | 620 | struct ext4_allocation_request ar; |
ac27a0ec | 621 | int target, i; |
7061eba7 | 622 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 623 | int index = 0; |
617ba13b | 624 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
625 | int ret = 0; |
626 | ||
627 | /* | |
628 | * Here we try to allocate the requested multiple blocks at once, | |
629 | * on a best-effort basis. | |
630 | * To build a branch, we should allocate blocks for | |
631 | * the indirect blocks(if not allocated yet), and at least | |
632 | * the first direct block of this branch. That's the | |
633 | * minimum number of blocks need to allocate(required) | |
634 | */ | |
7061eba7 AK |
635 | /* first we try to allocate the indirect blocks */ |
636 | target = indirect_blks; | |
637 | while (target > 0) { | |
ac27a0ec DK |
638 | count = target; |
639 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
640 | current_block = ext4_new_meta_blocks(handle, inode, |
641 | goal, &count, err); | |
ac27a0ec DK |
642 | if (*err) |
643 | goto failed_out; | |
644 | ||
645 | target -= count; | |
646 | /* allocate blocks for indirect blocks */ | |
647 | while (index < indirect_blks && count) { | |
648 | new_blocks[index++] = current_block++; | |
649 | count--; | |
650 | } | |
7061eba7 AK |
651 | if (count > 0) { |
652 | /* | |
653 | * save the new block number | |
654 | * for the first direct block | |
655 | */ | |
656 | new_blocks[index] = current_block; | |
657 | printk(KERN_INFO "%s returned more blocks than " | |
658 | "requested\n", __func__); | |
659 | WARN_ON(1); | |
ac27a0ec | 660 | break; |
7061eba7 | 661 | } |
ac27a0ec DK |
662 | } |
663 | ||
7061eba7 AK |
664 | target = blks - count ; |
665 | blk_allocated = count; | |
666 | if (!target) | |
667 | goto allocated; | |
668 | /* Now allocate data blocks */ | |
815a1130 TT |
669 | memset(&ar, 0, sizeof(ar)); |
670 | ar.inode = inode; | |
671 | ar.goal = goal; | |
672 | ar.len = target; | |
673 | ar.logical = iblock; | |
674 | if (S_ISREG(inode->i_mode)) | |
675 | /* enable in-core preallocation only for regular files */ | |
676 | ar.flags = EXT4_MB_HINT_DATA; | |
677 | ||
678 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
679 | ||
7061eba7 AK |
680 | if (*err && (target == blks)) { |
681 | /* | |
682 | * if the allocation failed and we didn't allocate | |
683 | * any blocks before | |
684 | */ | |
685 | goto failed_out; | |
686 | } | |
687 | if (!*err) { | |
688 | if (target == blks) { | |
de9a55b8 TT |
689 | /* |
690 | * save the new block number | |
691 | * for the first direct block | |
692 | */ | |
7061eba7 AK |
693 | new_blocks[index] = current_block; |
694 | } | |
815a1130 | 695 | blk_allocated += ar.len; |
7061eba7 AK |
696 | } |
697 | allocated: | |
ac27a0ec | 698 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 699 | ret = blk_allocated; |
ac27a0ec DK |
700 | *err = 0; |
701 | return ret; | |
702 | failed_out: | |
af5bc92d | 703 | for (i = 0; i < index; i++) |
c9de560d | 704 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec DK |
705 | return ret; |
706 | } | |
707 | ||
708 | /** | |
617ba13b | 709 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
ac27a0ec DK |
710 | * @inode: owner |
711 | * @indirect_blks: number of allocated indirect blocks | |
712 | * @blks: number of allocated direct blocks | |
713 | * @offsets: offsets (in the blocks) to store the pointers to next. | |
714 | * @branch: place to store the chain in. | |
715 | * | |
716 | * This function allocates blocks, zeroes out all but the last one, | |
717 | * links them into chain and (if we are synchronous) writes them to disk. | |
718 | * In other words, it prepares a branch that can be spliced onto the | |
719 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 720 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
721 | * we had read the existing part of chain and partial points to the last |
722 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 723 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
724 | * place chain is disconnected - *branch->p is still zero (we did not |
725 | * set the last link), but branch->key contains the number that should | |
726 | * be placed into *branch->p to fill that gap. | |
727 | * | |
728 | * If allocation fails we free all blocks we've allocated (and forget | |
729 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 730 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
731 | * as described above and return 0. |
732 | */ | |
617ba13b | 733 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
734 | ext4_lblk_t iblock, int indirect_blks, |
735 | int *blks, ext4_fsblk_t goal, | |
736 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
737 | { |
738 | int blocksize = inode->i_sb->s_blocksize; | |
739 | int i, n = 0; | |
740 | int err = 0; | |
741 | struct buffer_head *bh; | |
742 | int num; | |
617ba13b MC |
743 | ext4_fsblk_t new_blocks[4]; |
744 | ext4_fsblk_t current_block; | |
ac27a0ec | 745 | |
7061eba7 | 746 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
747 | *blks, new_blocks, &err); |
748 | if (err) | |
749 | return err; | |
750 | ||
751 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
752 | /* | |
753 | * metadata blocks and data blocks are allocated. | |
754 | */ | |
755 | for (n = 1; n <= indirect_blks; n++) { | |
756 | /* | |
757 | * Get buffer_head for parent block, zero it out | |
758 | * and set the pointer to new one, then send | |
759 | * parent to disk. | |
760 | */ | |
761 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
762 | branch[n].bh = bh; | |
763 | lock_buffer(bh); | |
764 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 765 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec DK |
766 | if (err) { |
767 | unlock_buffer(bh); | |
768 | brelse(bh); | |
769 | goto failed; | |
770 | } | |
771 | ||
772 | memset(bh->b_data, 0, blocksize); | |
773 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
774 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
775 | *branch[n].p = branch[n].key; | |
af5bc92d | 776 | if (n == indirect_blks) { |
ac27a0ec DK |
777 | current_block = new_blocks[n]; |
778 | /* | |
779 | * End of chain, update the last new metablock of | |
780 | * the chain to point to the new allocated | |
781 | * data blocks numbers | |
782 | */ | |
de9a55b8 | 783 | for (i = 1; i < num; i++) |
ac27a0ec DK |
784 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
785 | } | |
786 | BUFFER_TRACE(bh, "marking uptodate"); | |
787 | set_buffer_uptodate(bh); | |
788 | unlock_buffer(bh); | |
789 | ||
0390131b FM |
790 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
791 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
792 | if (err) |
793 | goto failed; | |
794 | } | |
795 | *blks = num; | |
796 | return err; | |
797 | failed: | |
798 | /* Allocation failed, free what we already allocated */ | |
799 | for (i = 1; i <= n ; i++) { | |
dab291af | 800 | BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget"); |
617ba13b | 801 | ext4_journal_forget(handle, branch[i].bh); |
ac27a0ec | 802 | } |
af5bc92d | 803 | for (i = 0; i < indirect_blks; i++) |
c9de560d | 804 | ext4_free_blocks(handle, inode, new_blocks[i], 1, 0); |
ac27a0ec | 805 | |
c9de560d | 806 | ext4_free_blocks(handle, inode, new_blocks[i], num, 0); |
ac27a0ec DK |
807 | |
808 | return err; | |
809 | } | |
810 | ||
811 | /** | |
617ba13b | 812 | * ext4_splice_branch - splice the allocated branch onto inode. |
ac27a0ec DK |
813 | * @inode: owner |
814 | * @block: (logical) number of block we are adding | |
815 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 816 | * ext4_alloc_branch) |
ac27a0ec DK |
817 | * @where: location of missing link |
818 | * @num: number of indirect blocks we are adding | |
819 | * @blks: number of direct blocks we are adding | |
820 | * | |
821 | * This function fills the missing link and does all housekeeping needed in | |
822 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
823 | * chain to new block and return 0. | |
824 | */ | |
617ba13b | 825 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
826 | ext4_lblk_t block, Indirect *where, int num, |
827 | int blks) | |
ac27a0ec DK |
828 | { |
829 | int i; | |
830 | int err = 0; | |
617ba13b | 831 | ext4_fsblk_t current_block; |
ac27a0ec | 832 | |
ac27a0ec DK |
833 | /* |
834 | * If we're splicing into a [td]indirect block (as opposed to the | |
835 | * inode) then we need to get write access to the [td]indirect block | |
836 | * before the splice. | |
837 | */ | |
838 | if (where->bh) { | |
839 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 840 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
841 | if (err) |
842 | goto err_out; | |
843 | } | |
844 | /* That's it */ | |
845 | ||
846 | *where->p = where->key; | |
847 | ||
848 | /* | |
849 | * Update the host buffer_head or inode to point to more just allocated | |
850 | * direct blocks blocks | |
851 | */ | |
852 | if (num == 0 && blks > 1) { | |
853 | current_block = le32_to_cpu(where->key) + 1; | |
854 | for (i = 1; i < blks; i++) | |
af5bc92d | 855 | *(where->p + i) = cpu_to_le32(current_block++); |
ac27a0ec DK |
856 | } |
857 | ||
ac27a0ec | 858 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
ac27a0ec DK |
859 | /* had we spliced it onto indirect block? */ |
860 | if (where->bh) { | |
861 | /* | |
862 | * If we spliced it onto an indirect block, we haven't | |
863 | * altered the inode. Note however that if it is being spliced | |
864 | * onto an indirect block at the very end of the file (the | |
865 | * file is growing) then we *will* alter the inode to reflect | |
866 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 867 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
868 | */ |
869 | jbd_debug(5, "splicing indirect only\n"); | |
0390131b FM |
870 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
871 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
ac27a0ec DK |
872 | if (err) |
873 | goto err_out; | |
874 | } else { | |
875 | /* | |
876 | * OK, we spliced it into the inode itself on a direct block. | |
ac27a0ec | 877 | */ |
41591750 | 878 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
879 | jbd_debug(5, "splicing direct\n"); |
880 | } | |
881 | return err; | |
882 | ||
883 | err_out: | |
884 | for (i = 1; i <= num; i++) { | |
dab291af | 885 | BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget"); |
617ba13b | 886 | ext4_journal_forget(handle, where[i].bh); |
c9de560d AT |
887 | ext4_free_blocks(handle, inode, |
888 | le32_to_cpu(where[i-1].key), 1, 0); | |
ac27a0ec | 889 | } |
c9de560d | 890 | ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0); |
ac27a0ec DK |
891 | |
892 | return err; | |
893 | } | |
894 | ||
895 | /* | |
b920c755 TT |
896 | * The ext4_ind_get_blocks() function handles non-extents inodes |
897 | * (i.e., using the traditional indirect/double-indirect i_blocks | |
898 | * scheme) for ext4_get_blocks(). | |
899 | * | |
ac27a0ec DK |
900 | * Allocation strategy is simple: if we have to allocate something, we will |
901 | * have to go the whole way to leaf. So let's do it before attaching anything | |
902 | * to tree, set linkage between the newborn blocks, write them if sync is | |
903 | * required, recheck the path, free and repeat if check fails, otherwise | |
904 | * set the last missing link (that will protect us from any truncate-generated | |
905 | * removals - all blocks on the path are immune now) and possibly force the | |
906 | * write on the parent block. | |
907 | * That has a nice additional property: no special recovery from the failed | |
908 | * allocations is needed - we simply release blocks and do not touch anything | |
909 | * reachable from inode. | |
910 | * | |
911 | * `handle' can be NULL if create == 0. | |
912 | * | |
ac27a0ec DK |
913 | * return > 0, # of blocks mapped or allocated. |
914 | * return = 0, if plain lookup failed. | |
915 | * return < 0, error case. | |
c278bfec | 916 | * |
b920c755 TT |
917 | * The ext4_ind_get_blocks() function should be called with |
918 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
919 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
920 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
921 | * blocks. | |
ac27a0ec | 922 | */ |
e4d996ca | 923 | static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
924 | ext4_lblk_t iblock, unsigned int maxblocks, |
925 | struct buffer_head *bh_result, | |
926 | int flags) | |
ac27a0ec DK |
927 | { |
928 | int err = -EIO; | |
725d26d3 | 929 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
930 | Indirect chain[4]; |
931 | Indirect *partial; | |
617ba13b | 932 | ext4_fsblk_t goal; |
ac27a0ec DK |
933 | int indirect_blks; |
934 | int blocks_to_boundary = 0; | |
935 | int depth; | |
ac27a0ec | 936 | int count = 0; |
617ba13b | 937 | ext4_fsblk_t first_block = 0; |
ac27a0ec | 938 | |
a86c6181 | 939 | J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); |
c2177057 | 940 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
725d26d3 | 941 | depth = ext4_block_to_path(inode, iblock, offsets, |
de9a55b8 | 942 | &blocks_to_boundary); |
ac27a0ec DK |
943 | |
944 | if (depth == 0) | |
945 | goto out; | |
946 | ||
617ba13b | 947 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
948 | |
949 | /* Simplest case - block found, no allocation needed */ | |
950 | if (!partial) { | |
951 | first_block = le32_to_cpu(chain[depth - 1].key); | |
952 | clear_buffer_new(bh_result); | |
953 | count++; | |
954 | /*map more blocks*/ | |
955 | while (count < maxblocks && count <= blocks_to_boundary) { | |
617ba13b | 956 | ext4_fsblk_t blk; |
ac27a0ec | 957 | |
ac27a0ec DK |
958 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
959 | ||
960 | if (blk == first_block + count) | |
961 | count++; | |
962 | else | |
963 | break; | |
964 | } | |
c278bfec | 965 | goto got_it; |
ac27a0ec DK |
966 | } |
967 | ||
968 | /* Next simple case - plain lookup or failed read of indirect block */ | |
c2177057 | 969 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
ac27a0ec DK |
970 | goto cleanup; |
971 | ||
ac27a0ec | 972 | /* |
c2ea3fde | 973 | * Okay, we need to do block allocation. |
ac27a0ec | 974 | */ |
fb01bfda | 975 | goal = ext4_find_goal(inode, iblock, partial); |
ac27a0ec DK |
976 | |
977 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
978 | indirect_blks = (chain + depth) - partial - 1; | |
979 | ||
980 | /* | |
981 | * Next look up the indirect map to count the totoal number of | |
982 | * direct blocks to allocate for this branch. | |
983 | */ | |
617ba13b | 984 | count = ext4_blks_to_allocate(partial, indirect_blks, |
ac27a0ec DK |
985 | maxblocks, blocks_to_boundary); |
986 | /* | |
617ba13b | 987 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 988 | */ |
7061eba7 | 989 | err = ext4_alloc_branch(handle, inode, iblock, indirect_blks, |
de9a55b8 TT |
990 | &count, goal, |
991 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
992 | |
993 | /* | |
617ba13b | 994 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
995 | * on the new chain if there is a failure, but that risks using |
996 | * up transaction credits, especially for bitmaps where the | |
997 | * credits cannot be returned. Can we handle this somehow? We | |
998 | * may need to return -EAGAIN upwards in the worst case. --sct | |
999 | */ | |
1000 | if (!err) | |
617ba13b | 1001 | err = ext4_splice_branch(handle, inode, iblock, |
de9a55b8 TT |
1002 | partial, indirect_blks, count); |
1003 | else | |
ac27a0ec DK |
1004 | goto cleanup; |
1005 | ||
1006 | set_buffer_new(bh_result); | |
1007 | got_it: | |
1008 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); | |
1009 | if (count > blocks_to_boundary) | |
1010 | set_buffer_boundary(bh_result); | |
1011 | err = count; | |
1012 | /* Clean up and exit */ | |
1013 | partial = chain + depth - 1; /* the whole chain */ | |
1014 | cleanup: | |
1015 | while (partial > chain) { | |
1016 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1017 | brelse(partial->bh); | |
1018 | partial--; | |
1019 | } | |
1020 | BUFFER_TRACE(bh_result, "returned"); | |
1021 | out: | |
1022 | return err; | |
1023 | } | |
1024 | ||
60e58e0f MC |
1025 | qsize_t ext4_get_reserved_space(struct inode *inode) |
1026 | { | |
1027 | unsigned long long total; | |
1028 | ||
1029 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1030 | total = EXT4_I(inode)->i_reserved_data_blocks + | |
1031 | EXT4_I(inode)->i_reserved_meta_blocks; | |
1032 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1033 | ||
1034 | return total; | |
1035 | } | |
12219aea AK |
1036 | /* |
1037 | * Calculate the number of metadata blocks need to reserve | |
1038 | * to allocate @blocks for non extent file based file | |
1039 | */ | |
1040 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks) | |
1041 | { | |
1042 | int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
1043 | int ind_blks, dind_blks, tind_blks; | |
1044 | ||
1045 | /* number of new indirect blocks needed */ | |
1046 | ind_blks = (blocks + icap - 1) / icap; | |
1047 | ||
1048 | dind_blks = (ind_blks + icap - 1) / icap; | |
1049 | ||
1050 | tind_blks = 1; | |
1051 | ||
1052 | return ind_blks + dind_blks + tind_blks; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Calculate the number of metadata blocks need to reserve | |
1057 | * to allocate given number of blocks | |
1058 | */ | |
1059 | static int ext4_calc_metadata_amount(struct inode *inode, int blocks) | |
1060 | { | |
cd213226 MC |
1061 | if (!blocks) |
1062 | return 0; | |
1063 | ||
12219aea AK |
1064 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) |
1065 | return ext4_ext_calc_metadata_amount(inode, blocks); | |
1066 | ||
1067 | return ext4_indirect_calc_metadata_amount(inode, blocks); | |
1068 | } | |
1069 | ||
1070 | static void ext4_da_update_reserve_space(struct inode *inode, int used) | |
1071 | { | |
1072 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1073 | int total, mdb, mdb_free; | |
1074 | ||
1075 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
1076 | /* recalculate the number of metablocks still need to be reserved */ | |
1077 | total = EXT4_I(inode)->i_reserved_data_blocks - used; | |
1078 | mdb = ext4_calc_metadata_amount(inode, total); | |
1079 | ||
1080 | /* figure out how many metablocks to release */ | |
1081 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1082 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1083 | ||
6bc6e63f AK |
1084 | if (mdb_free) { |
1085 | /* Account for allocated meta_blocks */ | |
1086 | mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks; | |
1087 | ||
1088 | /* update fs dirty blocks counter */ | |
1089 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free); | |
1090 | EXT4_I(inode)->i_allocated_meta_blocks = 0; | |
1091 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
1092 | } | |
12219aea AK |
1093 | |
1094 | /* update per-inode reservations */ | |
1095 | BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks); | |
1096 | EXT4_I(inode)->i_reserved_data_blocks -= used; | |
12219aea | 1097 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f MC |
1098 | |
1099 | /* | |
1100 | * free those over-booking quota for metadata blocks | |
1101 | */ | |
60e58e0f MC |
1102 | if (mdb_free) |
1103 | vfs_dq_release_reservation_block(inode, mdb_free); | |
d6014301 AK |
1104 | |
1105 | /* | |
1106 | * If we have done all the pending block allocations and if | |
1107 | * there aren't any writers on the inode, we can discard the | |
1108 | * inode's preallocations. | |
1109 | */ | |
1110 | if (!total && (atomic_read(&inode->i_writecount) == 0)) | |
1111 | ext4_discard_preallocations(inode); | |
12219aea AK |
1112 | } |
1113 | ||
6fd058f7 TT |
1114 | static int check_block_validity(struct inode *inode, sector_t logical, |
1115 | sector_t phys, int len) | |
1116 | { | |
1117 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) { | |
1118 | ext4_error(inode->i_sb, "check_block_validity", | |
1119 | "inode #%lu logical block %llu mapped to %llu " | |
1120 | "(size %d)", inode->i_ino, | |
1121 | (unsigned long long) logical, | |
1122 | (unsigned long long) phys, len); | |
1123 | WARN_ON(1); | |
1124 | return -EIO; | |
1125 | } | |
1126 | return 0; | |
1127 | } | |
1128 | ||
f5ab0d1f | 1129 | /* |
12b7ac17 | 1130 | * The ext4_get_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 1131 | * and returns if the blocks are already mapped. |
f5ab0d1f | 1132 | * |
f5ab0d1f MC |
1133 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1134 | * and store the allocated blocks in the result buffer head and mark it | |
1135 | * mapped. | |
1136 | * | |
1137 | * If file type is extents based, it will call ext4_ext_get_blocks(), | |
e4d996ca | 1138 | * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping |
f5ab0d1f MC |
1139 | * based files |
1140 | * | |
1141 | * On success, it returns the number of blocks being mapped or allocate. | |
1142 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1143 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1144 | * the buffer head is mapped. | |
1145 | * | |
1146 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1147 | * that casem, buffer head is unmapped | |
1148 | * | |
1149 | * It returns the error in case of allocation failure. | |
1150 | */ | |
12b7ac17 TT |
1151 | int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block, |
1152 | unsigned int max_blocks, struct buffer_head *bh, | |
c2177057 | 1153 | int flags) |
0e855ac8 AK |
1154 | { |
1155 | int retval; | |
f5ab0d1f MC |
1156 | |
1157 | clear_buffer_mapped(bh); | |
2a8964d6 | 1158 | clear_buffer_unwritten(bh); |
f5ab0d1f | 1159 | |
4df3d265 | 1160 | /* |
b920c755 TT |
1161 | * Try to see if we can get the block without requesting a new |
1162 | * file system block. | |
4df3d265 AK |
1163 | */ |
1164 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1165 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { | |
1166 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
c2177057 | 1167 | bh, 0); |
0e855ac8 | 1168 | } else { |
e4d996ca | 1169 | retval = ext4_ind_get_blocks(handle, inode, block, max_blocks, |
c2177057 | 1170 | bh, 0); |
0e855ac8 | 1171 | } |
4df3d265 | 1172 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 1173 | |
6fd058f7 | 1174 | if (retval > 0 && buffer_mapped(bh)) { |
de9a55b8 | 1175 | int ret = check_block_validity(inode, block, |
6fd058f7 TT |
1176 | bh->b_blocknr, retval); |
1177 | if (ret != 0) | |
1178 | return ret; | |
1179 | } | |
1180 | ||
f5ab0d1f | 1181 | /* If it is only a block(s) look up */ |
c2177057 | 1182 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
1183 | return retval; |
1184 | ||
1185 | /* | |
1186 | * Returns if the blocks have already allocated | |
1187 | * | |
1188 | * Note that if blocks have been preallocated | |
1189 | * ext4_ext_get_block() returns th create = 0 | |
1190 | * with buffer head unmapped. | |
1191 | */ | |
1192 | if (retval > 0 && buffer_mapped(bh)) | |
4df3d265 AK |
1193 | return retval; |
1194 | ||
2a8964d6 AK |
1195 | /* |
1196 | * When we call get_blocks without the create flag, the | |
1197 | * BH_Unwritten flag could have gotten set if the blocks | |
1198 | * requested were part of a uninitialized extent. We need to | |
1199 | * clear this flag now that we are committed to convert all or | |
1200 | * part of the uninitialized extent to be an initialized | |
1201 | * extent. This is because we need to avoid the combination | |
1202 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
1203 | * set on the buffer_head. | |
1204 | */ | |
1205 | clear_buffer_unwritten(bh); | |
1206 | ||
4df3d265 | 1207 | /* |
f5ab0d1f MC |
1208 | * New blocks allocate and/or writing to uninitialized extent |
1209 | * will possibly result in updating i_data, so we take | |
1210 | * the write lock of i_data_sem, and call get_blocks() | |
1211 | * with create == 1 flag. | |
4df3d265 AK |
1212 | */ |
1213 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1214 | |
1215 | /* | |
1216 | * if the caller is from delayed allocation writeout path | |
1217 | * we have already reserved fs blocks for allocation | |
1218 | * let the underlying get_block() function know to | |
1219 | * avoid double accounting | |
1220 | */ | |
c2177057 | 1221 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1222 | EXT4_I(inode)->i_delalloc_reserved_flag = 1; |
4df3d265 AK |
1223 | /* |
1224 | * We need to check for EXT4 here because migrate | |
1225 | * could have changed the inode type in between | |
1226 | */ | |
0e855ac8 AK |
1227 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
1228 | retval = ext4_ext_get_blocks(handle, inode, block, max_blocks, | |
c2177057 | 1229 | bh, flags); |
0e855ac8 | 1230 | } else { |
e4d996ca | 1231 | retval = ext4_ind_get_blocks(handle, inode, block, |
c2177057 | 1232 | max_blocks, bh, flags); |
267e4db9 AK |
1233 | |
1234 | if (retval > 0 && buffer_new(bh)) { | |
1235 | /* | |
1236 | * We allocated new blocks which will result in | |
1237 | * i_data's format changing. Force the migrate | |
1238 | * to fail by clearing migrate flags | |
1239 | */ | |
1240 | EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags & | |
1241 | ~EXT4_EXT_MIGRATE; | |
1242 | } | |
0e855ac8 | 1243 | } |
d2a17637 | 1244 | |
2ac3b6e0 | 1245 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
d2a17637 | 1246 | EXT4_I(inode)->i_delalloc_reserved_flag = 0; |
2ac3b6e0 TT |
1247 | |
1248 | /* | |
1249 | * Update reserved blocks/metadata blocks after successful | |
1250 | * block allocation which had been deferred till now. | |
1251 | */ | |
1252 | if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE)) | |
1253 | ext4_da_update_reserve_space(inode, retval); | |
d2a17637 | 1254 | |
4df3d265 | 1255 | up_write((&EXT4_I(inode)->i_data_sem)); |
6fd058f7 | 1256 | if (retval > 0 && buffer_mapped(bh)) { |
de9a55b8 | 1257 | int ret = check_block_validity(inode, block, |
6fd058f7 TT |
1258 | bh->b_blocknr, retval); |
1259 | if (ret != 0) | |
1260 | return ret; | |
1261 | } | |
0e855ac8 AK |
1262 | return retval; |
1263 | } | |
1264 | ||
f3bd1f3f MC |
1265 | /* Maximum number of blocks we map for direct IO at once. */ |
1266 | #define DIO_MAX_BLOCKS 4096 | |
1267 | ||
6873fa0d ES |
1268 | int ext4_get_block(struct inode *inode, sector_t iblock, |
1269 | struct buffer_head *bh_result, int create) | |
ac27a0ec | 1270 | { |
3e4fdaf8 | 1271 | handle_t *handle = ext4_journal_current_handle(); |
7fb5409d | 1272 | int ret = 0, started = 0; |
ac27a0ec | 1273 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
f3bd1f3f | 1274 | int dio_credits; |
ac27a0ec | 1275 | |
7fb5409d JK |
1276 | if (create && !handle) { |
1277 | /* Direct IO write... */ | |
1278 | if (max_blocks > DIO_MAX_BLOCKS) | |
1279 | max_blocks = DIO_MAX_BLOCKS; | |
f3bd1f3f MC |
1280 | dio_credits = ext4_chunk_trans_blocks(inode, max_blocks); |
1281 | handle = ext4_journal_start(inode, dio_credits); | |
7fb5409d | 1282 | if (IS_ERR(handle)) { |
ac27a0ec | 1283 | ret = PTR_ERR(handle); |
7fb5409d | 1284 | goto out; |
ac27a0ec | 1285 | } |
7fb5409d | 1286 | started = 1; |
ac27a0ec DK |
1287 | } |
1288 | ||
12b7ac17 | 1289 | ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result, |
c2177057 | 1290 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
7fb5409d JK |
1291 | if (ret > 0) { |
1292 | bh_result->b_size = (ret << inode->i_blkbits); | |
1293 | ret = 0; | |
ac27a0ec | 1294 | } |
7fb5409d JK |
1295 | if (started) |
1296 | ext4_journal_stop(handle); | |
1297 | out: | |
ac27a0ec DK |
1298 | return ret; |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * `handle' can be NULL if create is zero | |
1303 | */ | |
617ba13b | 1304 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1305 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec DK |
1306 | { |
1307 | struct buffer_head dummy; | |
1308 | int fatal = 0, err; | |
03f5d8bc | 1309 | int flags = 0; |
ac27a0ec DK |
1310 | |
1311 | J_ASSERT(handle != NULL || create == 0); | |
1312 | ||
1313 | dummy.b_state = 0; | |
1314 | dummy.b_blocknr = -1000; | |
1315 | buffer_trace_init(&dummy.b_history); | |
c2177057 TT |
1316 | if (create) |
1317 | flags |= EXT4_GET_BLOCKS_CREATE; | |
1318 | err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags); | |
ac27a0ec | 1319 | /* |
c2177057 TT |
1320 | * ext4_get_blocks() returns number of blocks mapped. 0 in |
1321 | * case of a HOLE. | |
ac27a0ec DK |
1322 | */ |
1323 | if (err > 0) { | |
1324 | if (err > 1) | |
1325 | WARN_ON(1); | |
1326 | err = 0; | |
1327 | } | |
1328 | *errp = err; | |
1329 | if (!err && buffer_mapped(&dummy)) { | |
1330 | struct buffer_head *bh; | |
1331 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); | |
1332 | if (!bh) { | |
1333 | *errp = -EIO; | |
1334 | goto err; | |
1335 | } | |
1336 | if (buffer_new(&dummy)) { | |
1337 | J_ASSERT(create != 0); | |
ac39849d | 1338 | J_ASSERT(handle != NULL); |
ac27a0ec DK |
1339 | |
1340 | /* | |
1341 | * Now that we do not always journal data, we should | |
1342 | * keep in mind whether this should always journal the | |
1343 | * new buffer as metadata. For now, regular file | |
617ba13b | 1344 | * writes use ext4_get_block instead, so it's not a |
ac27a0ec DK |
1345 | * problem. |
1346 | */ | |
1347 | lock_buffer(bh); | |
1348 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 1349 | fatal = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 1350 | if (!fatal && !buffer_uptodate(bh)) { |
af5bc92d | 1351 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
ac27a0ec DK |
1352 | set_buffer_uptodate(bh); |
1353 | } | |
1354 | unlock_buffer(bh); | |
0390131b FM |
1355 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
1356 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
1357 | if (!fatal) |
1358 | fatal = err; | |
1359 | } else { | |
1360 | BUFFER_TRACE(bh, "not a new buffer"); | |
1361 | } | |
1362 | if (fatal) { | |
1363 | *errp = fatal; | |
1364 | brelse(bh); | |
1365 | bh = NULL; | |
1366 | } | |
1367 | return bh; | |
1368 | } | |
1369 | err: | |
1370 | return NULL; | |
1371 | } | |
1372 | ||
617ba13b | 1373 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1374 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 1375 | { |
af5bc92d | 1376 | struct buffer_head *bh; |
ac27a0ec | 1377 | |
617ba13b | 1378 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1379 | if (!bh) |
1380 | return bh; | |
1381 | if (buffer_uptodate(bh)) | |
1382 | return bh; | |
1383 | ll_rw_block(READ_META, 1, &bh); | |
1384 | wait_on_buffer(bh); | |
1385 | if (buffer_uptodate(bh)) | |
1386 | return bh; | |
1387 | put_bh(bh); | |
1388 | *err = -EIO; | |
1389 | return NULL; | |
1390 | } | |
1391 | ||
af5bc92d TT |
1392 | static int walk_page_buffers(handle_t *handle, |
1393 | struct buffer_head *head, | |
1394 | unsigned from, | |
1395 | unsigned to, | |
1396 | int *partial, | |
1397 | int (*fn)(handle_t *handle, | |
1398 | struct buffer_head *bh)) | |
ac27a0ec DK |
1399 | { |
1400 | struct buffer_head *bh; | |
1401 | unsigned block_start, block_end; | |
1402 | unsigned blocksize = head->b_size; | |
1403 | int err, ret = 0; | |
1404 | struct buffer_head *next; | |
1405 | ||
af5bc92d TT |
1406 | for (bh = head, block_start = 0; |
1407 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1408 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1409 | next = bh->b_this_page; |
1410 | block_end = block_start + blocksize; | |
1411 | if (block_end <= from || block_start >= to) { | |
1412 | if (partial && !buffer_uptodate(bh)) | |
1413 | *partial = 1; | |
1414 | continue; | |
1415 | } | |
1416 | err = (*fn)(handle, bh); | |
1417 | if (!ret) | |
1418 | ret = err; | |
1419 | } | |
1420 | return ret; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * To preserve ordering, it is essential that the hole instantiation and | |
1425 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1426 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1427 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1428 | * prepare_write() is the right place. |
1429 | * | |
617ba13b MC |
1430 | * Also, this function can nest inside ext4_writepage() -> |
1431 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1432 | * has generated enough buffer credits to do the whole page. So we won't |
1433 | * block on the journal in that case, which is good, because the caller may | |
1434 | * be PF_MEMALLOC. | |
1435 | * | |
617ba13b | 1436 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1437 | * quota file writes. If we were to commit the transaction while thus |
1438 | * reentered, there can be a deadlock - we would be holding a quota | |
1439 | * lock, and the commit would never complete if another thread had a | |
1440 | * transaction open and was blocking on the quota lock - a ranking | |
1441 | * violation. | |
1442 | * | |
dab291af | 1443 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1444 | * will _not_ run commit under these circumstances because handle->h_ref |
1445 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1446 | * write. | |
1447 | */ | |
1448 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 1449 | struct buffer_head *bh) |
ac27a0ec DK |
1450 | { |
1451 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1452 | return 0; | |
617ba13b | 1453 | return ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
1454 | } |
1455 | ||
bfc1af65 | 1456 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1457 | loff_t pos, unsigned len, unsigned flags, |
1458 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1459 | { |
af5bc92d | 1460 | struct inode *inode = mapping->host; |
1938a150 | 1461 | int ret, needed_blocks; |
ac27a0ec DK |
1462 | handle_t *handle; |
1463 | int retries = 0; | |
af5bc92d | 1464 | struct page *page; |
de9a55b8 | 1465 | pgoff_t index; |
af5bc92d | 1466 | unsigned from, to; |
bfc1af65 | 1467 | |
9bffad1e | 1468 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1469 | /* |
1470 | * Reserve one block more for addition to orphan list in case | |
1471 | * we allocate blocks but write fails for some reason | |
1472 | */ | |
1473 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1474 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1475 | from = pos & (PAGE_CACHE_SIZE - 1); |
1476 | to = from + len; | |
ac27a0ec DK |
1477 | |
1478 | retry: | |
af5bc92d TT |
1479 | handle = ext4_journal_start(inode, needed_blocks); |
1480 | if (IS_ERR(handle)) { | |
1481 | ret = PTR_ERR(handle); | |
1482 | goto out; | |
7479d2b9 | 1483 | } |
ac27a0ec | 1484 | |
ebd3610b JK |
1485 | /* We cannot recurse into the filesystem as the transaction is already |
1486 | * started */ | |
1487 | flags |= AOP_FLAG_NOFS; | |
1488 | ||
54566b2c | 1489 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
1490 | if (!page) { |
1491 | ext4_journal_stop(handle); | |
1492 | ret = -ENOMEM; | |
1493 | goto out; | |
1494 | } | |
1495 | *pagep = page; | |
1496 | ||
bfc1af65 | 1497 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, |
ebd3610b | 1498 | ext4_get_block); |
bfc1af65 NP |
1499 | |
1500 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1501 | ret = walk_page_buffers(handle, page_buffers(page), |
1502 | from, to, NULL, do_journal_get_write_access); | |
1503 | } | |
bfc1af65 NP |
1504 | |
1505 | if (ret) { | |
af5bc92d | 1506 | unlock_page(page); |
af5bc92d | 1507 | page_cache_release(page); |
ae4d5372 AK |
1508 | /* |
1509 | * block_write_begin may have instantiated a few blocks | |
1510 | * outside i_size. Trim these off again. Don't need | |
1511 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1512 | * |
1513 | * Add inode to orphan list in case we crash before | |
1514 | * truncate finishes | |
ae4d5372 AK |
1515 | */ |
1516 | if (pos + len > inode->i_size) | |
1938a150 AK |
1517 | ext4_orphan_add(handle, inode); |
1518 | ||
1519 | ext4_journal_stop(handle); | |
1520 | if (pos + len > inode->i_size) { | |
ae4d5372 | 1521 | vmtruncate(inode, inode->i_size); |
de9a55b8 | 1522 | /* |
1938a150 AK |
1523 | * If vmtruncate failed early the inode might |
1524 | * still be on the orphan list; we need to | |
1525 | * make sure the inode is removed from the | |
1526 | * orphan list in that case. | |
1527 | */ | |
1528 | if (inode->i_nlink) | |
1529 | ext4_orphan_del(NULL, inode); | |
1530 | } | |
bfc1af65 NP |
1531 | } |
1532 | ||
617ba13b | 1533 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1534 | goto retry; |
7479d2b9 | 1535 | out: |
ac27a0ec DK |
1536 | return ret; |
1537 | } | |
1538 | ||
bfc1af65 NP |
1539 | /* For write_end() in data=journal mode */ |
1540 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1541 | { |
1542 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1543 | return 0; | |
1544 | set_buffer_uptodate(bh); | |
0390131b | 1545 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1546 | } |
1547 | ||
f8514083 | 1548 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
1549 | struct address_space *mapping, |
1550 | loff_t pos, unsigned len, unsigned copied, | |
1551 | struct page *page, void *fsdata) | |
f8514083 AK |
1552 | { |
1553 | int i_size_changed = 0; | |
1554 | struct inode *inode = mapping->host; | |
1555 | handle_t *handle = ext4_journal_current_handle(); | |
1556 | ||
1557 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1558 | ||
1559 | /* | |
1560 | * No need to use i_size_read() here, the i_size | |
1561 | * cannot change under us because we hold i_mutex. | |
1562 | * | |
1563 | * But it's important to update i_size while still holding page lock: | |
1564 | * page writeout could otherwise come in and zero beyond i_size. | |
1565 | */ | |
1566 | if (pos + copied > inode->i_size) { | |
1567 | i_size_write(inode, pos + copied); | |
1568 | i_size_changed = 1; | |
1569 | } | |
1570 | ||
1571 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
1572 | /* We need to mark inode dirty even if | |
1573 | * new_i_size is less that inode->i_size | |
1574 | * bu greater than i_disksize.(hint delalloc) | |
1575 | */ | |
1576 | ext4_update_i_disksize(inode, (pos + copied)); | |
1577 | i_size_changed = 1; | |
1578 | } | |
1579 | unlock_page(page); | |
1580 | page_cache_release(page); | |
1581 | ||
1582 | /* | |
1583 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1584 | * makes the holding time of page lock longer. Second, it forces lock | |
1585 | * ordering of page lock and transaction start for journaling | |
1586 | * filesystems. | |
1587 | */ | |
1588 | if (i_size_changed) | |
1589 | ext4_mark_inode_dirty(handle, inode); | |
1590 | ||
1591 | return copied; | |
1592 | } | |
1593 | ||
ac27a0ec DK |
1594 | /* |
1595 | * We need to pick up the new inode size which generic_commit_write gave us | |
1596 | * `file' can be NULL - eg, when called from page_symlink(). | |
1597 | * | |
617ba13b | 1598 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1599 | * buffers are managed internally. |
1600 | */ | |
bfc1af65 | 1601 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1602 | struct address_space *mapping, |
1603 | loff_t pos, unsigned len, unsigned copied, | |
1604 | struct page *page, void *fsdata) | |
ac27a0ec | 1605 | { |
617ba13b | 1606 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1607 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1608 | int ret = 0, ret2; |
1609 | ||
9bffad1e | 1610 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1611 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1612 | |
1613 | if (ret == 0) { | |
f8514083 | 1614 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1615 | page, fsdata); |
f8a87d89 | 1616 | copied = ret2; |
f8514083 AK |
1617 | if (pos + len > inode->i_size) |
1618 | /* if we have allocated more blocks and copied | |
1619 | * less. We will have blocks allocated outside | |
1620 | * inode->i_size. So truncate them | |
1621 | */ | |
1622 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1623 | if (ret2 < 0) |
1624 | ret = ret2; | |
ac27a0ec | 1625 | } |
617ba13b | 1626 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1627 | if (!ret) |
1628 | ret = ret2; | |
bfc1af65 | 1629 | |
f8514083 AK |
1630 | if (pos + len > inode->i_size) { |
1631 | vmtruncate(inode, inode->i_size); | |
de9a55b8 | 1632 | /* |
f8514083 AK |
1633 | * If vmtruncate failed early the inode might still be |
1634 | * on the orphan list; we need to make sure the inode | |
1635 | * is removed from the orphan list in that case. | |
1636 | */ | |
1637 | if (inode->i_nlink) | |
1638 | ext4_orphan_del(NULL, inode); | |
1639 | } | |
1640 | ||
1641 | ||
bfc1af65 | 1642 | return ret ? ret : copied; |
ac27a0ec DK |
1643 | } |
1644 | ||
bfc1af65 | 1645 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1646 | struct address_space *mapping, |
1647 | loff_t pos, unsigned len, unsigned copied, | |
1648 | struct page *page, void *fsdata) | |
ac27a0ec | 1649 | { |
617ba13b | 1650 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1651 | struct inode *inode = mapping->host; |
ac27a0ec | 1652 | int ret = 0, ret2; |
ac27a0ec | 1653 | |
9bffad1e | 1654 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1655 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1656 | page, fsdata); |
f8a87d89 | 1657 | copied = ret2; |
f8514083 AK |
1658 | if (pos + len > inode->i_size) |
1659 | /* if we have allocated more blocks and copied | |
1660 | * less. We will have blocks allocated outside | |
1661 | * inode->i_size. So truncate them | |
1662 | */ | |
1663 | ext4_orphan_add(handle, inode); | |
1664 | ||
f8a87d89 RK |
1665 | if (ret2 < 0) |
1666 | ret = ret2; | |
ac27a0ec | 1667 | |
617ba13b | 1668 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1669 | if (!ret) |
1670 | ret = ret2; | |
bfc1af65 | 1671 | |
f8514083 AK |
1672 | if (pos + len > inode->i_size) { |
1673 | vmtruncate(inode, inode->i_size); | |
de9a55b8 | 1674 | /* |
f8514083 AK |
1675 | * If vmtruncate failed early the inode might still be |
1676 | * on the orphan list; we need to make sure the inode | |
1677 | * is removed from the orphan list in that case. | |
1678 | */ | |
1679 | if (inode->i_nlink) | |
1680 | ext4_orphan_del(NULL, inode); | |
1681 | } | |
1682 | ||
bfc1af65 | 1683 | return ret ? ret : copied; |
ac27a0ec DK |
1684 | } |
1685 | ||
bfc1af65 | 1686 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1687 | struct address_space *mapping, |
1688 | loff_t pos, unsigned len, unsigned copied, | |
1689 | struct page *page, void *fsdata) | |
ac27a0ec | 1690 | { |
617ba13b | 1691 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1692 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1693 | int ret = 0, ret2; |
1694 | int partial = 0; | |
bfc1af65 | 1695 | unsigned from, to; |
cf17fea6 | 1696 | loff_t new_i_size; |
ac27a0ec | 1697 | |
9bffad1e | 1698 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1699 | from = pos & (PAGE_CACHE_SIZE - 1); |
1700 | to = from + len; | |
1701 | ||
1702 | if (copied < len) { | |
1703 | if (!PageUptodate(page)) | |
1704 | copied = 0; | |
1705 | page_zero_new_buffers(page, from+copied, to); | |
1706 | } | |
ac27a0ec DK |
1707 | |
1708 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1709 | to, &partial, write_end_fn); |
ac27a0ec DK |
1710 | if (!partial) |
1711 | SetPageUptodate(page); | |
cf17fea6 AK |
1712 | new_i_size = pos + copied; |
1713 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1714 | i_size_write(inode, pos+copied); |
617ba13b | 1715 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; |
cf17fea6 AK |
1716 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1717 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1718 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1719 | if (!ret) |
1720 | ret = ret2; | |
1721 | } | |
bfc1af65 | 1722 | |
cf108bca | 1723 | unlock_page(page); |
f8514083 AK |
1724 | page_cache_release(page); |
1725 | if (pos + len > inode->i_size) | |
1726 | /* if we have allocated more blocks and copied | |
1727 | * less. We will have blocks allocated outside | |
1728 | * inode->i_size. So truncate them | |
1729 | */ | |
1730 | ext4_orphan_add(handle, inode); | |
1731 | ||
617ba13b | 1732 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1733 | if (!ret) |
1734 | ret = ret2; | |
f8514083 AK |
1735 | if (pos + len > inode->i_size) { |
1736 | vmtruncate(inode, inode->i_size); | |
de9a55b8 | 1737 | /* |
f8514083 AK |
1738 | * If vmtruncate failed early the inode might still be |
1739 | * on the orphan list; we need to make sure the inode | |
1740 | * is removed from the orphan list in that case. | |
1741 | */ | |
1742 | if (inode->i_nlink) | |
1743 | ext4_orphan_del(NULL, inode); | |
1744 | } | |
bfc1af65 NP |
1745 | |
1746 | return ret ? ret : copied; | |
ac27a0ec | 1747 | } |
d2a17637 MC |
1748 | |
1749 | static int ext4_da_reserve_space(struct inode *inode, int nrblocks) | |
1750 | { | |
030ba6bc | 1751 | int retries = 0; |
60e58e0f MC |
1752 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1753 | unsigned long md_needed, mdblocks, total = 0; | |
d2a17637 MC |
1754 | |
1755 | /* | |
1756 | * recalculate the amount of metadata blocks to reserve | |
1757 | * in order to allocate nrblocks | |
1758 | * worse case is one extent per block | |
1759 | */ | |
030ba6bc | 1760 | repeat: |
d2a17637 MC |
1761 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
1762 | total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks; | |
1763 | mdblocks = ext4_calc_metadata_amount(inode, total); | |
1764 | BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks); | |
1765 | ||
1766 | md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks; | |
1767 | total = md_needed + nrblocks; | |
1768 | ||
60e58e0f MC |
1769 | /* |
1770 | * Make quota reservation here to prevent quota overflow | |
1771 | * later. Real quota accounting is done at pages writeout | |
1772 | * time. | |
1773 | */ | |
1774 | if (vfs_dq_reserve_block(inode, total)) { | |
1775 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1776 | return -EDQUOT; | |
1777 | } | |
1778 | ||
a30d542a | 1779 | if (ext4_claim_free_blocks(sbi, total)) { |
d2a17637 | 1780 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
030ba6bc AK |
1781 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1782 | yield(); | |
1783 | goto repeat; | |
1784 | } | |
60e58e0f | 1785 | vfs_dq_release_reservation_block(inode, total); |
d2a17637 MC |
1786 | return -ENOSPC; |
1787 | } | |
d2a17637 MC |
1788 | EXT4_I(inode)->i_reserved_data_blocks += nrblocks; |
1789 | EXT4_I(inode)->i_reserved_meta_blocks = mdblocks; | |
1790 | ||
1791 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1792 | return 0; /* success */ | |
1793 | } | |
1794 | ||
12219aea | 1795 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1796 | { |
1797 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1798 | int total, mdb, mdb_free, release; | |
1799 | ||
cd213226 MC |
1800 | if (!to_free) |
1801 | return; /* Nothing to release, exit */ | |
1802 | ||
d2a17637 | 1803 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 MC |
1804 | |
1805 | if (!EXT4_I(inode)->i_reserved_data_blocks) { | |
1806 | /* | |
1807 | * if there is no reserved blocks, but we try to free some | |
1808 | * then the counter is messed up somewhere. | |
1809 | * but since this function is called from invalidate | |
1810 | * page, it's harmless to return without any action | |
1811 | */ | |
1812 | printk(KERN_INFO "ext4 delalloc try to release %d reserved " | |
1813 | "blocks for inode %lu, but there is no reserved " | |
1814 | "data blocks\n", to_free, inode->i_ino); | |
1815 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
1816 | return; | |
1817 | } | |
1818 | ||
d2a17637 | 1819 | /* recalculate the number of metablocks still need to be reserved */ |
12219aea | 1820 | total = EXT4_I(inode)->i_reserved_data_blocks - to_free; |
d2a17637 MC |
1821 | mdb = ext4_calc_metadata_amount(inode, total); |
1822 | ||
1823 | /* figure out how many metablocks to release */ | |
1824 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1825 | mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb; | |
1826 | ||
d2a17637 MC |
1827 | release = to_free + mdb_free; |
1828 | ||
6bc6e63f AK |
1829 | /* update fs dirty blocks counter for truncate case */ |
1830 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, release); | |
d2a17637 MC |
1831 | |
1832 | /* update per-inode reservations */ | |
12219aea AK |
1833 | BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks); |
1834 | EXT4_I(inode)->i_reserved_data_blocks -= to_free; | |
d2a17637 MC |
1835 | |
1836 | BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks); | |
1837 | EXT4_I(inode)->i_reserved_meta_blocks = mdb; | |
d2a17637 | 1838 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f MC |
1839 | |
1840 | vfs_dq_release_reservation_block(inode, release); | |
d2a17637 MC |
1841 | } |
1842 | ||
1843 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1844 | unsigned long offset) |
d2a17637 MC |
1845 | { |
1846 | int to_release = 0; | |
1847 | struct buffer_head *head, *bh; | |
1848 | unsigned int curr_off = 0; | |
1849 | ||
1850 | head = page_buffers(page); | |
1851 | bh = head; | |
1852 | do { | |
1853 | unsigned int next_off = curr_off + bh->b_size; | |
1854 | ||
1855 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1856 | to_release++; | |
1857 | clear_buffer_delay(bh); | |
1858 | } | |
1859 | curr_off = next_off; | |
1860 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 1861 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 1862 | } |
ac27a0ec | 1863 | |
64769240 AT |
1864 | /* |
1865 | * Delayed allocation stuff | |
1866 | */ | |
1867 | ||
1868 | struct mpage_da_data { | |
1869 | struct inode *inode; | |
8dc207c0 TT |
1870 | sector_t b_blocknr; /* start block number of extent */ |
1871 | size_t b_size; /* size of extent */ | |
1872 | unsigned long b_state; /* state of the extent */ | |
64769240 | 1873 | unsigned long first_page, next_page; /* extent of pages */ |
64769240 | 1874 | struct writeback_control *wbc; |
a1d6cc56 | 1875 | int io_done; |
498e5f24 | 1876 | int pages_written; |
df22291f | 1877 | int retval; |
64769240 AT |
1878 | }; |
1879 | ||
1880 | /* | |
1881 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1882 | * them with writepage() call back |
64769240 AT |
1883 | * |
1884 | * @mpd->inode: inode | |
1885 | * @mpd->first_page: first page of the extent | |
1886 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1887 | * |
1888 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1889 | * to be allocated. this may be wrong if allocation failed. | |
1890 | * | |
1891 | * As pages are already locked by write_cache_pages(), we can't use it | |
1892 | */ | |
1893 | static int mpage_da_submit_io(struct mpage_da_data *mpd) | |
1894 | { | |
22208ded | 1895 | long pages_skipped; |
791b7f08 AK |
1896 | struct pagevec pvec; |
1897 | unsigned long index, end; | |
1898 | int ret = 0, err, nr_pages, i; | |
1899 | struct inode *inode = mpd->inode; | |
1900 | struct address_space *mapping = inode->i_mapping; | |
64769240 AT |
1901 | |
1902 | BUG_ON(mpd->next_page <= mpd->first_page); | |
791b7f08 AK |
1903 | /* |
1904 | * We need to start from the first_page to the next_page - 1 | |
1905 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1906 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1907 | * at the currently mapped buffer_heads. |
1908 | */ | |
64769240 AT |
1909 | index = mpd->first_page; |
1910 | end = mpd->next_page - 1; | |
1911 | ||
791b7f08 | 1912 | pagevec_init(&pvec, 0); |
64769240 | 1913 | while (index <= end) { |
791b7f08 | 1914 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1915 | if (nr_pages == 0) |
1916 | break; | |
1917 | for (i = 0; i < nr_pages; i++) { | |
1918 | struct page *page = pvec.pages[i]; | |
1919 | ||
791b7f08 AK |
1920 | index = page->index; |
1921 | if (index > end) | |
1922 | break; | |
1923 | index++; | |
1924 | ||
1925 | BUG_ON(!PageLocked(page)); | |
1926 | BUG_ON(PageWriteback(page)); | |
1927 | ||
22208ded | 1928 | pages_skipped = mpd->wbc->pages_skipped; |
a1d6cc56 | 1929 | err = mapping->a_ops->writepage(page, mpd->wbc); |
22208ded AK |
1930 | if (!err && (pages_skipped == mpd->wbc->pages_skipped)) |
1931 | /* | |
1932 | * have successfully written the page | |
1933 | * without skipping the same | |
1934 | */ | |
a1d6cc56 | 1935 | mpd->pages_written++; |
64769240 AT |
1936 | /* |
1937 | * In error case, we have to continue because | |
1938 | * remaining pages are still locked | |
1939 | * XXX: unlock and re-dirty them? | |
1940 | */ | |
1941 | if (ret == 0) | |
1942 | ret = err; | |
1943 | } | |
1944 | pagevec_release(&pvec); | |
1945 | } | |
64769240 AT |
1946 | return ret; |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers | |
1951 | * | |
1952 | * @mpd->inode - inode to walk through | |
1953 | * @exbh->b_blocknr - first block on a disk | |
1954 | * @exbh->b_size - amount of space in bytes | |
1955 | * @logical - first logical block to start assignment with | |
1956 | * | |
1957 | * the function goes through all passed space and put actual disk | |
29fa89d0 | 1958 | * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten |
64769240 AT |
1959 | */ |
1960 | static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical, | |
1961 | struct buffer_head *exbh) | |
1962 | { | |
1963 | struct inode *inode = mpd->inode; | |
1964 | struct address_space *mapping = inode->i_mapping; | |
1965 | int blocks = exbh->b_size >> inode->i_blkbits; | |
1966 | sector_t pblock = exbh->b_blocknr, cur_logical; | |
1967 | struct buffer_head *head, *bh; | |
a1d6cc56 | 1968 | pgoff_t index, end; |
64769240 AT |
1969 | struct pagevec pvec; |
1970 | int nr_pages, i; | |
1971 | ||
1972 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1973 | end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1974 | cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1975 | ||
1976 | pagevec_init(&pvec, 0); | |
1977 | ||
1978 | while (index <= end) { | |
1979 | /* XXX: optimize tail */ | |
1980 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1981 | if (nr_pages == 0) | |
1982 | break; | |
1983 | for (i = 0; i < nr_pages; i++) { | |
1984 | struct page *page = pvec.pages[i]; | |
1985 | ||
1986 | index = page->index; | |
1987 | if (index > end) | |
1988 | break; | |
1989 | index++; | |
1990 | ||
1991 | BUG_ON(!PageLocked(page)); | |
1992 | BUG_ON(PageWriteback(page)); | |
1993 | BUG_ON(!page_has_buffers(page)); | |
1994 | ||
1995 | bh = page_buffers(page); | |
1996 | head = bh; | |
1997 | ||
1998 | /* skip blocks out of the range */ | |
1999 | do { | |
2000 | if (cur_logical >= logical) | |
2001 | break; | |
2002 | cur_logical++; | |
2003 | } while ((bh = bh->b_this_page) != head); | |
2004 | ||
2005 | do { | |
2006 | if (cur_logical >= logical + blocks) | |
2007 | break; | |
29fa89d0 AK |
2008 | |
2009 | if (buffer_delay(bh) || | |
2010 | buffer_unwritten(bh)) { | |
2011 | ||
2012 | BUG_ON(bh->b_bdev != inode->i_sb->s_bdev); | |
2013 | ||
2014 | if (buffer_delay(bh)) { | |
2015 | clear_buffer_delay(bh); | |
2016 | bh->b_blocknr = pblock; | |
2017 | } else { | |
2018 | /* | |
2019 | * unwritten already should have | |
2020 | * blocknr assigned. Verify that | |
2021 | */ | |
2022 | clear_buffer_unwritten(bh); | |
2023 | BUG_ON(bh->b_blocknr != pblock); | |
2024 | } | |
2025 | ||
61628a3f | 2026 | } else if (buffer_mapped(bh)) |
64769240 | 2027 | BUG_ON(bh->b_blocknr != pblock); |
64769240 AT |
2028 | |
2029 | cur_logical++; | |
2030 | pblock++; | |
2031 | } while ((bh = bh->b_this_page) != head); | |
2032 | } | |
2033 | pagevec_release(&pvec); | |
2034 | } | |
2035 | } | |
2036 | ||
2037 | ||
2038 | /* | |
2039 | * __unmap_underlying_blocks - just a helper function to unmap | |
2040 | * set of blocks described by @bh | |
2041 | */ | |
2042 | static inline void __unmap_underlying_blocks(struct inode *inode, | |
2043 | struct buffer_head *bh) | |
2044 | { | |
2045 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2046 | int blocks, i; | |
2047 | ||
2048 | blocks = bh->b_size >> inode->i_blkbits; | |
2049 | for (i = 0; i < blocks; i++) | |
2050 | unmap_underlying_metadata(bdev, bh->b_blocknr + i); | |
2051 | } | |
2052 | ||
c4a0c46e AK |
2053 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, |
2054 | sector_t logical, long blk_cnt) | |
2055 | { | |
2056 | int nr_pages, i; | |
2057 | pgoff_t index, end; | |
2058 | struct pagevec pvec; | |
2059 | struct inode *inode = mpd->inode; | |
2060 | struct address_space *mapping = inode->i_mapping; | |
2061 | ||
2062 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2063 | end = (logical + blk_cnt - 1) >> | |
2064 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2065 | while (index <= end) { | |
2066 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
2067 | if (nr_pages == 0) | |
2068 | break; | |
2069 | for (i = 0; i < nr_pages; i++) { | |
2070 | struct page *page = pvec.pages[i]; | |
2071 | index = page->index; | |
2072 | if (index > end) | |
2073 | break; | |
2074 | index++; | |
2075 | ||
2076 | BUG_ON(!PageLocked(page)); | |
2077 | BUG_ON(PageWriteback(page)); | |
2078 | block_invalidatepage(page, 0); | |
2079 | ClearPageUptodate(page); | |
2080 | unlock_page(page); | |
2081 | } | |
2082 | } | |
2083 | return; | |
2084 | } | |
2085 | ||
df22291f AK |
2086 | static void ext4_print_free_blocks(struct inode *inode) |
2087 | { | |
2088 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
2089 | printk(KERN_EMERG "Total free blocks count %lld\n", | |
2090 | ext4_count_free_blocks(inode->i_sb)); | |
2091 | printk(KERN_EMERG "Free/Dirty block details\n"); | |
2092 | printk(KERN_EMERG "free_blocks=%lld\n", | |
8f72fbdf | 2093 | (long long)percpu_counter_sum(&sbi->s_freeblocks_counter)); |
df22291f | 2094 | printk(KERN_EMERG "dirty_blocks=%lld\n", |
8f72fbdf | 2095 | (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter)); |
df22291f | 2096 | printk(KERN_EMERG "Block reservation details\n"); |
498e5f24 | 2097 | printk(KERN_EMERG "i_reserved_data_blocks=%u\n", |
df22291f | 2098 | EXT4_I(inode)->i_reserved_data_blocks); |
498e5f24 | 2099 | printk(KERN_EMERG "i_reserved_meta_blocks=%u\n", |
df22291f AK |
2100 | EXT4_I(inode)->i_reserved_meta_blocks); |
2101 | return; | |
2102 | } | |
2103 | ||
64769240 AT |
2104 | /* |
2105 | * mpage_da_map_blocks - go through given space | |
2106 | * | |
8dc207c0 | 2107 | * @mpd - bh describing space |
64769240 AT |
2108 | * |
2109 | * The function skips space we know is already mapped to disk blocks. | |
2110 | * | |
64769240 | 2111 | */ |
ed5bde0b | 2112 | static int mpage_da_map_blocks(struct mpage_da_data *mpd) |
64769240 | 2113 | { |
2ac3b6e0 | 2114 | int err, blks, get_blocks_flags; |
030ba6bc | 2115 | struct buffer_head new; |
2fa3cdfb TT |
2116 | sector_t next = mpd->b_blocknr; |
2117 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
2118 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
2119 | handle_t *handle = NULL; | |
64769240 AT |
2120 | |
2121 | /* | |
2122 | * We consider only non-mapped and non-allocated blocks | |
2123 | */ | |
8dc207c0 | 2124 | if ((mpd->b_state & (1 << BH_Mapped)) && |
29fa89d0 AK |
2125 | !(mpd->b_state & (1 << BH_Delay)) && |
2126 | !(mpd->b_state & (1 << BH_Unwritten))) | |
c4a0c46e | 2127 | return 0; |
2fa3cdfb TT |
2128 | |
2129 | /* | |
2130 | * If we didn't accumulate anything to write simply return | |
2131 | */ | |
2132 | if (!mpd->b_size) | |
2133 | return 0; | |
2134 | ||
2135 | handle = ext4_journal_current_handle(); | |
2136 | BUG_ON(!handle); | |
2137 | ||
79ffab34 | 2138 | /* |
2ac3b6e0 TT |
2139 | * Call ext4_get_blocks() to allocate any delayed allocation |
2140 | * blocks, or to convert an uninitialized extent to be | |
2141 | * initialized (in the case where we have written into | |
2142 | * one or more preallocated blocks). | |
2143 | * | |
2144 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
2145 | * indicate that we are on the delayed allocation path. This | |
2146 | * affects functions in many different parts of the allocation | |
2147 | * call path. This flag exists primarily because we don't | |
2148 | * want to change *many* call functions, so ext4_get_blocks() | |
2149 | * will set the magic i_delalloc_reserved_flag once the | |
2150 | * inode's allocation semaphore is taken. | |
2151 | * | |
2152 | * If the blocks in questions were delalloc blocks, set | |
2153 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
2154 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 2155 | */ |
2ac3b6e0 TT |
2156 | new.b_state = 0; |
2157 | get_blocks_flags = (EXT4_GET_BLOCKS_CREATE | | |
2158 | EXT4_GET_BLOCKS_DELALLOC_RESERVE); | |
2159 | if (mpd->b_state & (1 << BH_Delay)) | |
2160 | get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE; | |
2fa3cdfb | 2161 | blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks, |
2ac3b6e0 | 2162 | &new, get_blocks_flags); |
2fa3cdfb TT |
2163 | if (blks < 0) { |
2164 | err = blks; | |
ed5bde0b TT |
2165 | /* |
2166 | * If get block returns with error we simply | |
2167 | * return. Later writepage will redirty the page and | |
2168 | * writepages will find the dirty page again | |
c4a0c46e AK |
2169 | */ |
2170 | if (err == -EAGAIN) | |
2171 | return 0; | |
df22291f AK |
2172 | |
2173 | if (err == -ENOSPC && | |
ed5bde0b | 2174 | ext4_count_free_blocks(mpd->inode->i_sb)) { |
df22291f AK |
2175 | mpd->retval = err; |
2176 | return 0; | |
2177 | } | |
2178 | ||
c4a0c46e | 2179 | /* |
ed5bde0b TT |
2180 | * get block failure will cause us to loop in |
2181 | * writepages, because a_ops->writepage won't be able | |
2182 | * to make progress. The page will be redirtied by | |
2183 | * writepage and writepages will again try to write | |
2184 | * the same. | |
c4a0c46e AK |
2185 | */ |
2186 | printk(KERN_EMERG "%s block allocation failed for inode %lu " | |
2187 | "at logical offset %llu with max blocks " | |
2188 | "%zd with error %d\n", | |
2189 | __func__, mpd->inode->i_ino, | |
2190 | (unsigned long long)next, | |
8dc207c0 | 2191 | mpd->b_size >> mpd->inode->i_blkbits, err); |
c4a0c46e AK |
2192 | printk(KERN_EMERG "This should not happen.!! " |
2193 | "Data will be lost\n"); | |
030ba6bc | 2194 | if (err == -ENOSPC) { |
df22291f | 2195 | ext4_print_free_blocks(mpd->inode); |
030ba6bc | 2196 | } |
2fa3cdfb | 2197 | /* invalidate all the pages */ |
c4a0c46e | 2198 | ext4_da_block_invalidatepages(mpd, next, |
8dc207c0 | 2199 | mpd->b_size >> mpd->inode->i_blkbits); |
c4a0c46e AK |
2200 | return err; |
2201 | } | |
2fa3cdfb TT |
2202 | BUG_ON(blks == 0); |
2203 | ||
2204 | new.b_size = (blks << mpd->inode->i_blkbits); | |
64769240 | 2205 | |
a1d6cc56 AK |
2206 | if (buffer_new(&new)) |
2207 | __unmap_underlying_blocks(mpd->inode, &new); | |
64769240 | 2208 | |
a1d6cc56 AK |
2209 | /* |
2210 | * If blocks are delayed marked, we need to | |
2211 | * put actual blocknr and drop delayed bit | |
2212 | */ | |
8dc207c0 TT |
2213 | if ((mpd->b_state & (1 << BH_Delay)) || |
2214 | (mpd->b_state & (1 << BH_Unwritten))) | |
a1d6cc56 | 2215 | mpage_put_bnr_to_bhs(mpd, next, &new); |
64769240 | 2216 | |
2fa3cdfb TT |
2217 | if (ext4_should_order_data(mpd->inode)) { |
2218 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
2219 | if (err) | |
2220 | return err; | |
2221 | } | |
2222 | ||
2223 | /* | |
03f5d8bc | 2224 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
2225 | */ |
2226 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
2227 | if (disksize > i_size_read(mpd->inode)) | |
2228 | disksize = i_size_read(mpd->inode); | |
2229 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
2230 | ext4_update_i_disksize(mpd->inode, disksize); | |
2231 | return ext4_mark_inode_dirty(handle, mpd->inode); | |
2232 | } | |
2233 | ||
c4a0c46e | 2234 | return 0; |
64769240 AT |
2235 | } |
2236 | ||
bf068ee2 AK |
2237 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
2238 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
2239 | |
2240 | /* | |
2241 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
2242 | * | |
2243 | * @mpd->lbh - extent of blocks | |
2244 | * @logical - logical number of the block in the file | |
2245 | * @bh - bh of the block (used to access block's state) | |
2246 | * | |
2247 | * the function is used to collect contig. blocks in same state | |
2248 | */ | |
2249 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
2250 | sector_t logical, size_t b_size, |
2251 | unsigned long b_state) | |
64769240 | 2252 | { |
64769240 | 2253 | sector_t next; |
8dc207c0 | 2254 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 2255 | |
525f4ed8 MC |
2256 | /* check if thereserved journal credits might overflow */ |
2257 | if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) { | |
2258 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { | |
2259 | /* | |
2260 | * With non-extent format we are limited by the journal | |
2261 | * credit available. Total credit needed to insert | |
2262 | * nrblocks contiguous blocks is dependent on the | |
2263 | * nrblocks. So limit nrblocks. | |
2264 | */ | |
2265 | goto flush_it; | |
2266 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
2267 | EXT4_MAX_TRANS_DATA) { | |
2268 | /* | |
2269 | * Adding the new buffer_head would make it cross the | |
2270 | * allowed limit for which we have journal credit | |
2271 | * reserved. So limit the new bh->b_size | |
2272 | */ | |
2273 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
2274 | mpd->inode->i_blkbits; | |
2275 | /* we will do mpage_da_submit_io in the next loop */ | |
2276 | } | |
2277 | } | |
64769240 AT |
2278 | /* |
2279 | * First block in the extent | |
2280 | */ | |
8dc207c0 TT |
2281 | if (mpd->b_size == 0) { |
2282 | mpd->b_blocknr = logical; | |
2283 | mpd->b_size = b_size; | |
2284 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
2285 | return; |
2286 | } | |
2287 | ||
8dc207c0 | 2288 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
2289 | /* |
2290 | * Can we merge the block to our big extent? | |
2291 | */ | |
8dc207c0 TT |
2292 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
2293 | mpd->b_size += b_size; | |
64769240 AT |
2294 | return; |
2295 | } | |
2296 | ||
525f4ed8 | 2297 | flush_it: |
64769240 AT |
2298 | /* |
2299 | * We couldn't merge the block to our extent, so we | |
2300 | * need to flush current extent and start new one | |
2301 | */ | |
c4a0c46e AK |
2302 | if (mpage_da_map_blocks(mpd) == 0) |
2303 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2304 | mpd->io_done = 1; |
2305 | return; | |
64769240 AT |
2306 | } |
2307 | ||
29fa89d0 AK |
2308 | static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh) |
2309 | { | |
2310 | /* | |
2311 | * unmapped buffer is possible for holes. | |
2312 | * delay buffer is possible with delayed allocation. | |
2313 | * We also need to consider unwritten buffer as unmapped. | |
2314 | */ | |
2315 | return (!buffer_mapped(bh) || buffer_delay(bh) || | |
2316 | buffer_unwritten(bh)) && buffer_dirty(bh); | |
2317 | } | |
2318 | ||
64769240 AT |
2319 | /* |
2320 | * __mpage_da_writepage - finds extent of pages and blocks | |
2321 | * | |
2322 | * @page: page to consider | |
2323 | * @wbc: not used, we just follow rules | |
2324 | * @data: context | |
2325 | * | |
2326 | * The function finds extents of pages and scan them for all blocks. | |
2327 | */ | |
2328 | static int __mpage_da_writepage(struct page *page, | |
2329 | struct writeback_control *wbc, void *data) | |
2330 | { | |
2331 | struct mpage_da_data *mpd = data; | |
2332 | struct inode *inode = mpd->inode; | |
8dc207c0 | 2333 | struct buffer_head *bh, *head; |
64769240 AT |
2334 | sector_t logical; |
2335 | ||
a1d6cc56 AK |
2336 | if (mpd->io_done) { |
2337 | /* | |
2338 | * Rest of the page in the page_vec | |
2339 | * redirty then and skip then. We will | |
2340 | * try to to write them again after | |
2341 | * starting a new transaction | |
2342 | */ | |
2343 | redirty_page_for_writepage(wbc, page); | |
2344 | unlock_page(page); | |
2345 | return MPAGE_DA_EXTENT_TAIL; | |
2346 | } | |
64769240 AT |
2347 | /* |
2348 | * Can we merge this page to current extent? | |
2349 | */ | |
2350 | if (mpd->next_page != page->index) { | |
2351 | /* | |
2352 | * Nope, we can't. So, we map non-allocated blocks | |
a1d6cc56 | 2353 | * and start IO on them using writepage() |
64769240 AT |
2354 | */ |
2355 | if (mpd->next_page != mpd->first_page) { | |
c4a0c46e AK |
2356 | if (mpage_da_map_blocks(mpd) == 0) |
2357 | mpage_da_submit_io(mpd); | |
a1d6cc56 AK |
2358 | /* |
2359 | * skip rest of the page in the page_vec | |
2360 | */ | |
2361 | mpd->io_done = 1; | |
2362 | redirty_page_for_writepage(wbc, page); | |
2363 | unlock_page(page); | |
2364 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2365 | } |
2366 | ||
2367 | /* | |
2368 | * Start next extent of pages ... | |
2369 | */ | |
2370 | mpd->first_page = page->index; | |
2371 | ||
2372 | /* | |
2373 | * ... and blocks | |
2374 | */ | |
8dc207c0 TT |
2375 | mpd->b_size = 0; |
2376 | mpd->b_state = 0; | |
2377 | mpd->b_blocknr = 0; | |
64769240 AT |
2378 | } |
2379 | ||
2380 | mpd->next_page = page->index + 1; | |
2381 | logical = (sector_t) page->index << | |
2382 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2383 | ||
2384 | if (!page_has_buffers(page)) { | |
8dc207c0 TT |
2385 | mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE, |
2386 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | |
a1d6cc56 AK |
2387 | if (mpd->io_done) |
2388 | return MPAGE_DA_EXTENT_TAIL; | |
64769240 AT |
2389 | } else { |
2390 | /* | |
2391 | * Page with regular buffer heads, just add all dirty ones | |
2392 | */ | |
2393 | head = page_buffers(page); | |
2394 | bh = head; | |
2395 | do { | |
2396 | BUG_ON(buffer_locked(bh)); | |
791b7f08 AK |
2397 | /* |
2398 | * We need to try to allocate | |
2399 | * unmapped blocks in the same page. | |
2400 | * Otherwise we won't make progress | |
2401 | * with the page in ext4_da_writepage | |
2402 | */ | |
29fa89d0 | 2403 | if (ext4_bh_unmapped_or_delay(NULL, bh)) { |
8dc207c0 TT |
2404 | mpage_add_bh_to_extent(mpd, logical, |
2405 | bh->b_size, | |
2406 | bh->b_state); | |
a1d6cc56 AK |
2407 | if (mpd->io_done) |
2408 | return MPAGE_DA_EXTENT_TAIL; | |
791b7f08 AK |
2409 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2410 | /* | |
2411 | * mapped dirty buffer. We need to update | |
2412 | * the b_state because we look at | |
2413 | * b_state in mpage_da_map_blocks. We don't | |
2414 | * update b_size because if we find an | |
2415 | * unmapped buffer_head later we need to | |
2416 | * use the b_state flag of that buffer_head. | |
2417 | */ | |
8dc207c0 TT |
2418 | if (mpd->b_size == 0) |
2419 | mpd->b_state = bh->b_state & BH_FLAGS; | |
a1d6cc56 | 2420 | } |
64769240 AT |
2421 | logical++; |
2422 | } while ((bh = bh->b_this_page) != head); | |
2423 | } | |
2424 | ||
2425 | return 0; | |
2426 | } | |
2427 | ||
64769240 | 2428 | /* |
b920c755 TT |
2429 | * This is a special get_blocks_t callback which is used by |
2430 | * ext4_da_write_begin(). It will either return mapped block or | |
2431 | * reserve space for a single block. | |
29fa89d0 AK |
2432 | * |
2433 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2434 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2435 | * | |
2436 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2437 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2438 | * initialized properly. | |
64769240 AT |
2439 | */ |
2440 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2441 | struct buffer_head *bh_result, int create) | |
2442 | { | |
2443 | int ret = 0; | |
33b9817e AK |
2444 | sector_t invalid_block = ~((sector_t) 0xffff); |
2445 | ||
2446 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
2447 | invalid_block = ~0; | |
64769240 AT |
2448 | |
2449 | BUG_ON(create == 0); | |
2450 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); | |
2451 | ||
2452 | /* | |
2453 | * first, we need to know whether the block is allocated already | |
2454 | * preallocated blocks are unmapped but should treated | |
2455 | * the same as allocated blocks. | |
2456 | */ | |
c2177057 | 2457 | ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0); |
d2a17637 MC |
2458 | if ((ret == 0) && !buffer_delay(bh_result)) { |
2459 | /* the block isn't (pre)allocated yet, let's reserve space */ | |
64769240 AT |
2460 | /* |
2461 | * XXX: __block_prepare_write() unmaps passed block, | |
2462 | * is it OK? | |
2463 | */ | |
d2a17637 MC |
2464 | ret = ext4_da_reserve_space(inode, 1); |
2465 | if (ret) | |
2466 | /* not enough space to reserve */ | |
2467 | return ret; | |
2468 | ||
33b9817e | 2469 | map_bh(bh_result, inode->i_sb, invalid_block); |
64769240 AT |
2470 | set_buffer_new(bh_result); |
2471 | set_buffer_delay(bh_result); | |
2472 | } else if (ret > 0) { | |
2473 | bh_result->b_size = (ret << inode->i_blkbits); | |
29fa89d0 AK |
2474 | if (buffer_unwritten(bh_result)) { |
2475 | /* A delayed write to unwritten bh should | |
2476 | * be marked new and mapped. Mapped ensures | |
2477 | * that we don't do get_block multiple times | |
2478 | * when we write to the same offset and new | |
2479 | * ensures that we do proper zero out for | |
2480 | * partial write. | |
2481 | */ | |
9c1ee184 | 2482 | set_buffer_new(bh_result); |
29fa89d0 AK |
2483 | set_buffer_mapped(bh_result); |
2484 | } | |
64769240 AT |
2485 | ret = 0; |
2486 | } | |
2487 | ||
2488 | return ret; | |
2489 | } | |
61628a3f | 2490 | |
b920c755 TT |
2491 | /* |
2492 | * This function is used as a standard get_block_t calback function | |
2493 | * when there is no desire to allocate any blocks. It is used as a | |
2494 | * callback function for block_prepare_write(), nobh_writepage(), and | |
2495 | * block_write_full_page(). These functions should only try to map a | |
2496 | * single block at a time. | |
2497 | * | |
2498 | * Since this function doesn't do block allocations even if the caller | |
2499 | * requests it by passing in create=1, it is critically important that | |
2500 | * any caller checks to make sure that any buffer heads are returned | |
2501 | * by this function are either all already mapped or marked for | |
2502 | * delayed allocation before calling nobh_writepage() or | |
2503 | * block_write_full_page(). Otherwise, b_blocknr could be left | |
2504 | * unitialized, and the page write functions will be taken by | |
2505 | * surprise. | |
2506 | */ | |
2507 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
2508 | struct buffer_head *bh_result, int create) |
2509 | { | |
2510 | int ret = 0; | |
2511 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; | |
2512 | ||
a2dc52b5 TT |
2513 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2514 | ||
f0e6c985 AK |
2515 | /* |
2516 | * we don't want to do block allocation in writepage | |
2517 | * so call get_block_wrap with create = 0 | |
2518 | */ | |
c2177057 | 2519 | ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0); |
a2dc52b5 | 2520 | BUG_ON(create && ret == 0); |
f0e6c985 AK |
2521 | if (ret > 0) { |
2522 | bh_result->b_size = (ret << inode->i_blkbits); | |
2523 | ret = 0; | |
2524 | } | |
2525 | return ret; | |
61628a3f MC |
2526 | } |
2527 | ||
61628a3f | 2528 | /* |
b920c755 TT |
2529 | * This function can get called via... |
2530 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2531 | * - journal_submit_inode_data_buffers (no journal handle) | |
2532 | * - shrink_page_list via pdflush (no journal handle) | |
2533 | * - grab_page_cache when doing write_begin (have journal handle) | |
61628a3f | 2534 | */ |
64769240 AT |
2535 | static int ext4_da_writepage(struct page *page, |
2536 | struct writeback_control *wbc) | |
2537 | { | |
64769240 | 2538 | int ret = 0; |
61628a3f | 2539 | loff_t size; |
498e5f24 | 2540 | unsigned int len; |
61628a3f MC |
2541 | struct buffer_head *page_bufs; |
2542 | struct inode *inode = page->mapping->host; | |
2543 | ||
9bffad1e | 2544 | trace_ext4_da_writepage(inode, page); |
f0e6c985 AK |
2545 | size = i_size_read(inode); |
2546 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2547 | len = size & ~PAGE_CACHE_MASK; | |
2548 | else | |
2549 | len = PAGE_CACHE_SIZE; | |
64769240 | 2550 | |
f0e6c985 | 2551 | if (page_has_buffers(page)) { |
61628a3f | 2552 | page_bufs = page_buffers(page); |
f0e6c985 AK |
2553 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2554 | ext4_bh_unmapped_or_delay)) { | |
61628a3f | 2555 | /* |
f0e6c985 AK |
2556 | * We don't want to do block allocation |
2557 | * So redirty the page and return | |
cd1aac32 AK |
2558 | * We may reach here when we do a journal commit |
2559 | * via journal_submit_inode_data_buffers. | |
2560 | * If we don't have mapping block we just ignore | |
f0e6c985 AK |
2561 | * them. We can also reach here via shrink_page_list |
2562 | */ | |
2563 | redirty_page_for_writepage(wbc, page); | |
2564 | unlock_page(page); | |
2565 | return 0; | |
2566 | } | |
2567 | } else { | |
2568 | /* | |
2569 | * The test for page_has_buffers() is subtle: | |
2570 | * We know the page is dirty but it lost buffers. That means | |
2571 | * that at some moment in time after write_begin()/write_end() | |
2572 | * has been called all buffers have been clean and thus they | |
2573 | * must have been written at least once. So they are all | |
2574 | * mapped and we can happily proceed with mapping them | |
2575 | * and writing the page. | |
2576 | * | |
2577 | * Try to initialize the buffer_heads and check whether | |
2578 | * all are mapped and non delay. We don't want to | |
2579 | * do block allocation here. | |
2580 | */ | |
2581 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, | |
b920c755 | 2582 | noalloc_get_block_write); |
f0e6c985 AK |
2583 | if (!ret) { |
2584 | page_bufs = page_buffers(page); | |
2585 | /* check whether all are mapped and non delay */ | |
2586 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2587 | ext4_bh_unmapped_or_delay)) { | |
2588 | redirty_page_for_writepage(wbc, page); | |
2589 | unlock_page(page); | |
2590 | return 0; | |
2591 | } | |
2592 | } else { | |
2593 | /* | |
2594 | * We can't do block allocation here | |
2595 | * so just redity the page and unlock | |
2596 | * and return | |
61628a3f | 2597 | */ |
61628a3f MC |
2598 | redirty_page_for_writepage(wbc, page); |
2599 | unlock_page(page); | |
2600 | return 0; | |
2601 | } | |
ed9b3e33 AK |
2602 | /* now mark the buffer_heads as dirty and uptodate */ |
2603 | block_commit_write(page, 0, PAGE_CACHE_SIZE); | |
64769240 AT |
2604 | } |
2605 | ||
2606 | if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) | |
b920c755 | 2607 | ret = nobh_writepage(page, noalloc_get_block_write, wbc); |
64769240 | 2608 | else |
b920c755 TT |
2609 | ret = block_write_full_page(page, noalloc_get_block_write, |
2610 | wbc); | |
64769240 | 2611 | |
64769240 AT |
2612 | return ret; |
2613 | } | |
2614 | ||
61628a3f | 2615 | /* |
525f4ed8 MC |
2616 | * This is called via ext4_da_writepages() to |
2617 | * calulate the total number of credits to reserve to fit | |
2618 | * a single extent allocation into a single transaction, | |
2619 | * ext4_da_writpeages() will loop calling this before | |
2620 | * the block allocation. | |
61628a3f | 2621 | */ |
525f4ed8 MC |
2622 | |
2623 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2624 | { | |
2625 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2626 | ||
2627 | /* | |
2628 | * With non-extent format the journal credit needed to | |
2629 | * insert nrblocks contiguous block is dependent on | |
2630 | * number of contiguous block. So we will limit | |
2631 | * number of contiguous block to a sane value | |
2632 | */ | |
2633 | if (!(inode->i_flags & EXT4_EXTENTS_FL) && | |
2634 | (max_blocks > EXT4_MAX_TRANS_DATA)) | |
2635 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2636 | ||
2637 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2638 | } | |
61628a3f | 2639 | |
64769240 | 2640 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2641 | struct writeback_control *wbc) |
64769240 | 2642 | { |
22208ded AK |
2643 | pgoff_t index; |
2644 | int range_whole = 0; | |
61628a3f | 2645 | handle_t *handle = NULL; |
df22291f | 2646 | struct mpage_da_data mpd; |
5e745b04 | 2647 | struct inode *inode = mapping->host; |
22208ded | 2648 | int no_nrwrite_index_update; |
498e5f24 TT |
2649 | int pages_written = 0; |
2650 | long pages_skipped; | |
2acf2c26 | 2651 | int range_cyclic, cycled = 1, io_done = 0; |
5e745b04 | 2652 | int needed_blocks, ret = 0, nr_to_writebump = 0; |
5e745b04 | 2653 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
61628a3f | 2654 | |
9bffad1e | 2655 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2656 | |
61628a3f MC |
2657 | /* |
2658 | * No pages to write? This is mainly a kludge to avoid starting | |
2659 | * a transaction for special inodes like journal inode on last iput() | |
2660 | * because that could violate lock ordering on umount | |
2661 | */ | |
a1d6cc56 | 2662 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2663 | return 0; |
2a21e37e TT |
2664 | |
2665 | /* | |
2666 | * If the filesystem has aborted, it is read-only, so return | |
2667 | * right away instead of dumping stack traces later on that | |
2668 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2669 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2670 | * the latter could be true if the filesystem is mounted |
2671 | * read-only, and in that case, ext4_da_writepages should | |
2672 | * *never* be called, so if that ever happens, we would want | |
2673 | * the stack trace. | |
2674 | */ | |
4ab2f15b | 2675 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2676 | return -EROFS; |
2677 | ||
5e745b04 AK |
2678 | /* |
2679 | * Make sure nr_to_write is >= sbi->s_mb_stream_request | |
2680 | * This make sure small files blocks are allocated in | |
2681 | * single attempt. This ensure that small files | |
2682 | * get less fragmented. | |
2683 | */ | |
2684 | if (wbc->nr_to_write < sbi->s_mb_stream_request) { | |
2685 | nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write; | |
2686 | wbc->nr_to_write = sbi->s_mb_stream_request; | |
2687 | } | |
22208ded AK |
2688 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2689 | range_whole = 1; | |
61628a3f | 2690 | |
2acf2c26 AK |
2691 | range_cyclic = wbc->range_cyclic; |
2692 | if (wbc->range_cyclic) { | |
22208ded | 2693 | index = mapping->writeback_index; |
2acf2c26 AK |
2694 | if (index) |
2695 | cycled = 0; | |
2696 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2697 | wbc->range_end = LLONG_MAX; | |
2698 | wbc->range_cyclic = 0; | |
2699 | } else | |
22208ded | 2700 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
a1d6cc56 | 2701 | |
df22291f AK |
2702 | mpd.wbc = wbc; |
2703 | mpd.inode = mapping->host; | |
2704 | ||
22208ded AK |
2705 | /* |
2706 | * we don't want write_cache_pages to update | |
2707 | * nr_to_write and writeback_index | |
2708 | */ | |
2709 | no_nrwrite_index_update = wbc->no_nrwrite_index_update; | |
2710 | wbc->no_nrwrite_index_update = 1; | |
2711 | pages_skipped = wbc->pages_skipped; | |
2712 | ||
2acf2c26 | 2713 | retry: |
22208ded | 2714 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2715 | |
2716 | /* | |
2717 | * we insert one extent at a time. So we need | |
2718 | * credit needed for single extent allocation. | |
2719 | * journalled mode is currently not supported | |
2720 | * by delalloc | |
2721 | */ | |
2722 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2723 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2724 | |
61628a3f MC |
2725 | /* start a new transaction*/ |
2726 | handle = ext4_journal_start(inode, needed_blocks); | |
2727 | if (IS_ERR(handle)) { | |
2728 | ret = PTR_ERR(handle); | |
2a21e37e | 2729 | printk(KERN_CRIT "%s: jbd2_start: " |
a1d6cc56 AK |
2730 | "%ld pages, ino %lu; err %d\n", __func__, |
2731 | wbc->nr_to_write, inode->i_ino, ret); | |
2732 | dump_stack(); | |
61628a3f MC |
2733 | goto out_writepages; |
2734 | } | |
f63e6005 TT |
2735 | |
2736 | /* | |
2737 | * Now call __mpage_da_writepage to find the next | |
2738 | * contiguous region of logical blocks that need | |
2739 | * blocks to be allocated by ext4. We don't actually | |
2740 | * submit the blocks for I/O here, even though | |
2741 | * write_cache_pages thinks it will, and will set the | |
2742 | * pages as clean for write before calling | |
2743 | * __mpage_da_writepage(). | |
2744 | */ | |
2745 | mpd.b_size = 0; | |
2746 | mpd.b_state = 0; | |
2747 | mpd.b_blocknr = 0; | |
2748 | mpd.first_page = 0; | |
2749 | mpd.next_page = 0; | |
2750 | mpd.io_done = 0; | |
2751 | mpd.pages_written = 0; | |
2752 | mpd.retval = 0; | |
2753 | ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, | |
2754 | &mpd); | |
2755 | /* | |
2756 | * If we have a contigous extent of pages and we | |
2757 | * haven't done the I/O yet, map the blocks and submit | |
2758 | * them for I/O. | |
2759 | */ | |
2760 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
2761 | if (mpage_da_map_blocks(&mpd) == 0) | |
2762 | mpage_da_submit_io(&mpd); | |
2763 | mpd.io_done = 1; | |
2764 | ret = MPAGE_DA_EXTENT_TAIL; | |
2765 | } | |
2766 | wbc->nr_to_write -= mpd.pages_written; | |
df22291f | 2767 | |
61628a3f | 2768 | ext4_journal_stop(handle); |
df22291f | 2769 | |
8f64b32e | 2770 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2771 | /* commit the transaction which would |
2772 | * free blocks released in the transaction | |
2773 | * and try again | |
2774 | */ | |
df22291f | 2775 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2776 | wbc->pages_skipped = pages_skipped; |
2777 | ret = 0; | |
2778 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
2779 | /* |
2780 | * got one extent now try with | |
2781 | * rest of the pages | |
2782 | */ | |
22208ded AK |
2783 | pages_written += mpd.pages_written; |
2784 | wbc->pages_skipped = pages_skipped; | |
a1d6cc56 | 2785 | ret = 0; |
2acf2c26 | 2786 | io_done = 1; |
22208ded | 2787 | } else if (wbc->nr_to_write) |
61628a3f MC |
2788 | /* |
2789 | * There is no more writeout needed | |
2790 | * or we requested for a noblocking writeout | |
2791 | * and we found the device congested | |
2792 | */ | |
61628a3f | 2793 | break; |
a1d6cc56 | 2794 | } |
2acf2c26 AK |
2795 | if (!io_done && !cycled) { |
2796 | cycled = 1; | |
2797 | index = 0; | |
2798 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2799 | wbc->range_end = mapping->writeback_index - 1; | |
2800 | goto retry; | |
2801 | } | |
22208ded AK |
2802 | if (pages_skipped != wbc->pages_skipped) |
2803 | printk(KERN_EMERG "This should not happen leaving %s " | |
2804 | "with nr_to_write = %ld ret = %d\n", | |
2805 | __func__, wbc->nr_to_write, ret); | |
2806 | ||
2807 | /* Update index */ | |
2808 | index += pages_written; | |
2acf2c26 | 2809 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2810 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2811 | /* | |
2812 | * set the writeback_index so that range_cyclic | |
2813 | * mode will write it back later | |
2814 | */ | |
2815 | mapping->writeback_index = index; | |
a1d6cc56 | 2816 | |
61628a3f | 2817 | out_writepages: |
22208ded AK |
2818 | if (!no_nrwrite_index_update) |
2819 | wbc->no_nrwrite_index_update = 0; | |
2820 | wbc->nr_to_write -= nr_to_writebump; | |
9bffad1e | 2821 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2822 | return ret; |
64769240 AT |
2823 | } |
2824 | ||
79f0be8d AK |
2825 | #define FALL_BACK_TO_NONDELALLOC 1 |
2826 | static int ext4_nonda_switch(struct super_block *sb) | |
2827 | { | |
2828 | s64 free_blocks, dirty_blocks; | |
2829 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2830 | ||
2831 | /* | |
2832 | * switch to non delalloc mode if we are running low | |
2833 | * on free block. The free block accounting via percpu | |
179f7ebf | 2834 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2835 | * accumulated on each CPU without updating global counters |
2836 | * Delalloc need an accurate free block accounting. So switch | |
2837 | * to non delalloc when we are near to error range. | |
2838 | */ | |
2839 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
2840 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter); | |
2841 | if (2 * free_blocks < 3 * dirty_blocks || | |
2842 | free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) { | |
2843 | /* | |
2844 | * free block count is less that 150% of dirty blocks | |
2845 | * or free blocks is less that watermark | |
2846 | */ | |
2847 | return 1; | |
2848 | } | |
2849 | return 0; | |
2850 | } | |
2851 | ||
64769240 | 2852 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2853 | loff_t pos, unsigned len, unsigned flags, |
2854 | struct page **pagep, void **fsdata) | |
64769240 | 2855 | { |
d2a17637 | 2856 | int ret, retries = 0; |
64769240 AT |
2857 | struct page *page; |
2858 | pgoff_t index; | |
2859 | unsigned from, to; | |
2860 | struct inode *inode = mapping->host; | |
2861 | handle_t *handle; | |
2862 | ||
2863 | index = pos >> PAGE_CACHE_SHIFT; | |
2864 | from = pos & (PAGE_CACHE_SIZE - 1); | |
2865 | to = from + len; | |
79f0be8d AK |
2866 | |
2867 | if (ext4_nonda_switch(inode->i_sb)) { | |
2868 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2869 | return ext4_write_begin(file, mapping, pos, | |
2870 | len, flags, pagep, fsdata); | |
2871 | } | |
2872 | *fsdata = (void *)0; | |
9bffad1e | 2873 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 2874 | retry: |
64769240 AT |
2875 | /* |
2876 | * With delayed allocation, we don't log the i_disksize update | |
2877 | * if there is delayed block allocation. But we still need | |
2878 | * to journalling the i_disksize update if writes to the end | |
2879 | * of file which has an already mapped buffer. | |
2880 | */ | |
2881 | handle = ext4_journal_start(inode, 1); | |
2882 | if (IS_ERR(handle)) { | |
2883 | ret = PTR_ERR(handle); | |
2884 | goto out; | |
2885 | } | |
ebd3610b JK |
2886 | /* We cannot recurse into the filesystem as the transaction is already |
2887 | * started */ | |
2888 | flags |= AOP_FLAG_NOFS; | |
64769240 | 2889 | |
54566b2c | 2890 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
2891 | if (!page) { |
2892 | ext4_journal_stop(handle); | |
2893 | ret = -ENOMEM; | |
2894 | goto out; | |
2895 | } | |
64769240 AT |
2896 | *pagep = page; |
2897 | ||
2898 | ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
b920c755 | 2899 | ext4_da_get_block_prep); |
64769240 AT |
2900 | if (ret < 0) { |
2901 | unlock_page(page); | |
2902 | ext4_journal_stop(handle); | |
2903 | page_cache_release(page); | |
ae4d5372 AK |
2904 | /* |
2905 | * block_write_begin may have instantiated a few blocks | |
2906 | * outside i_size. Trim these off again. Don't need | |
2907 | * i_size_read because we hold i_mutex. | |
2908 | */ | |
2909 | if (pos + len > inode->i_size) | |
2910 | vmtruncate(inode, inode->i_size); | |
64769240 AT |
2911 | } |
2912 | ||
d2a17637 MC |
2913 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2914 | goto retry; | |
64769240 AT |
2915 | out: |
2916 | return ret; | |
2917 | } | |
2918 | ||
632eaeab MC |
2919 | /* |
2920 | * Check if we should update i_disksize | |
2921 | * when write to the end of file but not require block allocation | |
2922 | */ | |
2923 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2924 | unsigned long offset) |
632eaeab MC |
2925 | { |
2926 | struct buffer_head *bh; | |
2927 | struct inode *inode = page->mapping->host; | |
2928 | unsigned int idx; | |
2929 | int i; | |
2930 | ||
2931 | bh = page_buffers(page); | |
2932 | idx = offset >> inode->i_blkbits; | |
2933 | ||
af5bc92d | 2934 | for (i = 0; i < idx; i++) |
632eaeab MC |
2935 | bh = bh->b_this_page; |
2936 | ||
29fa89d0 | 2937 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2938 | return 0; |
2939 | return 1; | |
2940 | } | |
2941 | ||
64769240 | 2942 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2943 | struct address_space *mapping, |
2944 | loff_t pos, unsigned len, unsigned copied, | |
2945 | struct page *page, void *fsdata) | |
64769240 AT |
2946 | { |
2947 | struct inode *inode = mapping->host; | |
2948 | int ret = 0, ret2; | |
2949 | handle_t *handle = ext4_journal_current_handle(); | |
2950 | loff_t new_i_size; | |
632eaeab | 2951 | unsigned long start, end; |
79f0be8d AK |
2952 | int write_mode = (int)(unsigned long)fsdata; |
2953 | ||
2954 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
2955 | if (ext4_should_order_data(inode)) { | |
2956 | return ext4_ordered_write_end(file, mapping, pos, | |
2957 | len, copied, page, fsdata); | |
2958 | } else if (ext4_should_writeback_data(inode)) { | |
2959 | return ext4_writeback_write_end(file, mapping, pos, | |
2960 | len, copied, page, fsdata); | |
2961 | } else { | |
2962 | BUG(); | |
2963 | } | |
2964 | } | |
632eaeab | 2965 | |
9bffad1e | 2966 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2967 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2968 | end = start + copied - 1; |
64769240 AT |
2969 | |
2970 | /* | |
2971 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2972 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2973 | * into that. | |
2974 | */ | |
2975 | ||
2976 | new_i_size = pos + copied; | |
632eaeab MC |
2977 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
2978 | if (ext4_da_should_update_i_disksize(page, end)) { | |
2979 | down_write(&EXT4_I(inode)->i_data_sem); | |
2980 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2981 | /* | |
2982 | * Updating i_disksize when extending file | |
2983 | * without needing block allocation | |
2984 | */ | |
2985 | if (ext4_should_order_data(inode)) | |
2986 | ret = ext4_jbd2_file_inode(handle, | |
2987 | inode); | |
64769240 | 2988 | |
632eaeab MC |
2989 | EXT4_I(inode)->i_disksize = new_i_size; |
2990 | } | |
2991 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
2992 | /* We need to mark inode dirty even if |
2993 | * new_i_size is less that inode->i_size | |
2994 | * bu greater than i_disksize.(hint delalloc) | |
2995 | */ | |
2996 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2997 | } |
632eaeab | 2998 | } |
64769240 AT |
2999 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
3000 | page, fsdata); | |
3001 | copied = ret2; | |
3002 | if (ret2 < 0) | |
3003 | ret = ret2; | |
3004 | ret2 = ext4_journal_stop(handle); | |
3005 | if (!ret) | |
3006 | ret = ret2; | |
3007 | ||
3008 | return ret ? ret : copied; | |
3009 | } | |
3010 | ||
3011 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
3012 | { | |
64769240 AT |
3013 | /* |
3014 | * Drop reserved blocks | |
3015 | */ | |
3016 | BUG_ON(!PageLocked(page)); | |
3017 | if (!page_has_buffers(page)) | |
3018 | goto out; | |
3019 | ||
d2a17637 | 3020 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
3021 | |
3022 | out: | |
3023 | ext4_invalidatepage(page, offset); | |
3024 | ||
3025 | return; | |
3026 | } | |
3027 | ||
ccd2506b TT |
3028 | /* |
3029 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3030 | */ | |
3031 | int ext4_alloc_da_blocks(struct inode *inode) | |
3032 | { | |
3033 | if (!EXT4_I(inode)->i_reserved_data_blocks && | |
3034 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
3035 | return 0; | |
3036 | ||
3037 | /* | |
3038 | * We do something simple for now. The filemap_flush() will | |
3039 | * also start triggering a write of the data blocks, which is | |
3040 | * not strictly speaking necessary (and for users of | |
3041 | * laptop_mode, not even desirable). However, to do otherwise | |
3042 | * would require replicating code paths in: | |
de9a55b8 | 3043 | * |
ccd2506b TT |
3044 | * ext4_da_writepages() -> |
3045 | * write_cache_pages() ---> (via passed in callback function) | |
3046 | * __mpage_da_writepage() --> | |
3047 | * mpage_add_bh_to_extent() | |
3048 | * mpage_da_map_blocks() | |
3049 | * | |
3050 | * The problem is that write_cache_pages(), located in | |
3051 | * mm/page-writeback.c, marks pages clean in preparation for | |
3052 | * doing I/O, which is not desirable if we're not planning on | |
3053 | * doing I/O at all. | |
3054 | * | |
3055 | * We could call write_cache_pages(), and then redirty all of | |
3056 | * the pages by calling redirty_page_for_writeback() but that | |
3057 | * would be ugly in the extreme. So instead we would need to | |
3058 | * replicate parts of the code in the above functions, | |
3059 | * simplifying them becuase we wouldn't actually intend to | |
3060 | * write out the pages, but rather only collect contiguous | |
3061 | * logical block extents, call the multi-block allocator, and | |
3062 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3063 | * |
ccd2506b TT |
3064 | * For now, though, we'll cheat by calling filemap_flush(), |
3065 | * which will map the blocks, and start the I/O, but not | |
3066 | * actually wait for the I/O to complete. | |
3067 | */ | |
3068 | return filemap_flush(inode->i_mapping); | |
3069 | } | |
64769240 | 3070 | |
ac27a0ec DK |
3071 | /* |
3072 | * bmap() is special. It gets used by applications such as lilo and by | |
3073 | * the swapper to find the on-disk block of a specific piece of data. | |
3074 | * | |
3075 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3076 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3077 | * filesystem and enables swap, then they may get a nasty shock when the |
3078 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3079 | * the original zero's written out previously to the journal and | |
3080 | * awaiting writeback in the kernel's buffer cache. | |
3081 | * | |
3082 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3083 | * take extra steps to flush any blocks which might be in the cache. | |
3084 | */ | |
617ba13b | 3085 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3086 | { |
3087 | struct inode *inode = mapping->host; | |
3088 | journal_t *journal; | |
3089 | int err; | |
3090 | ||
64769240 AT |
3091 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3092 | test_opt(inode->i_sb, DELALLOC)) { | |
3093 | /* | |
3094 | * With delalloc we want to sync the file | |
3095 | * so that we can make sure we allocate | |
3096 | * blocks for file | |
3097 | */ | |
3098 | filemap_write_and_wait(mapping); | |
3099 | } | |
3100 | ||
0390131b | 3101 | if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { |
ac27a0ec DK |
3102 | /* |
3103 | * This is a REALLY heavyweight approach, but the use of | |
3104 | * bmap on dirty files is expected to be extremely rare: | |
3105 | * only if we run lilo or swapon on a freshly made file | |
3106 | * do we expect this to happen. | |
3107 | * | |
3108 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3109 | * represent an unprivileged user DOS attack --- we'd be | |
3110 | * in trouble if mortal users could trigger this path at | |
3111 | * will.) | |
3112 | * | |
617ba13b | 3113 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3114 | * regular files. If somebody wants to bmap a directory |
3115 | * or symlink and gets confused because the buffer | |
3116 | * hasn't yet been flushed to disk, they deserve | |
3117 | * everything they get. | |
3118 | */ | |
3119 | ||
617ba13b MC |
3120 | EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA; |
3121 | journal = EXT4_JOURNAL(inode); | |
dab291af MC |
3122 | jbd2_journal_lock_updates(journal); |
3123 | err = jbd2_journal_flush(journal); | |
3124 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3125 | |
3126 | if (err) | |
3127 | return 0; | |
3128 | } | |
3129 | ||
af5bc92d | 3130 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3131 | } |
3132 | ||
3133 | static int bget_one(handle_t *handle, struct buffer_head *bh) | |
3134 | { | |
3135 | get_bh(bh); | |
3136 | return 0; | |
3137 | } | |
3138 | ||
3139 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
3140 | { | |
3141 | put_bh(bh); | |
3142 | return 0; | |
3143 | } | |
3144 | ||
ac27a0ec | 3145 | /* |
678aaf48 JK |
3146 | * Note that we don't need to start a transaction unless we're journaling data |
3147 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
3148 | * need to file the inode to the transaction's list in ordered mode because if | |
3149 | * we are writing back data added by write(), the inode is already there and if | |
3150 | * we are writing back data modified via mmap(), noone guarantees in which | |
3151 | * transaction the data will hit the disk. In case we are journaling data, we | |
3152 | * cannot start transaction directly because transaction start ranks above page | |
3153 | * lock so we have to do some magic. | |
ac27a0ec | 3154 | * |
678aaf48 | 3155 | * In all journaling modes block_write_full_page() will start the I/O. |
ac27a0ec DK |
3156 | * |
3157 | * Problem: | |
3158 | * | |
617ba13b MC |
3159 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
3160 | * ext4_writepage() | |
ac27a0ec DK |
3161 | * |
3162 | * Similar for: | |
3163 | * | |
617ba13b | 3164 | * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
ac27a0ec | 3165 | * |
617ba13b | 3166 | * Same applies to ext4_get_block(). We will deadlock on various things like |
0e855ac8 | 3167 | * lock_journal and i_data_sem |
ac27a0ec DK |
3168 | * |
3169 | * Setting PF_MEMALLOC here doesn't work - too many internal memory | |
3170 | * allocations fail. | |
3171 | * | |
3172 | * 16May01: If we're reentered then journal_current_handle() will be | |
3173 | * non-zero. We simply *return*. | |
3174 | * | |
3175 | * 1 July 2001: @@@ FIXME: | |
3176 | * In journalled data mode, a data buffer may be metadata against the | |
3177 | * current transaction. But the same file is part of a shared mapping | |
3178 | * and someone does a writepage() on it. | |
3179 | * | |
3180 | * We will move the buffer onto the async_data list, but *after* it has | |
3181 | * been dirtied. So there's a small window where we have dirty data on | |
3182 | * BJ_Metadata. | |
3183 | * | |
3184 | * Note that this only applies to the last partial page in the file. The | |
3185 | * bit which block_write_full_page() uses prepare/commit for. (That's | |
3186 | * broken code anyway: it's wrong for msync()). | |
3187 | * | |
3188 | * It's a rare case: affects the final partial page, for journalled data | |
3189 | * where the file is subject to bith write() and writepage() in the same | |
3190 | * transction. To fix it we'll need a custom block_write_full_page(). | |
3191 | * We'll probably need that anyway for journalling writepage() output. | |
3192 | * | |
3193 | * We don't honour synchronous mounts for writepage(). That would be | |
3194 | * disastrous. Any write() or metadata operation will sync the fs for | |
3195 | * us. | |
3196 | * | |
ac27a0ec | 3197 | */ |
678aaf48 | 3198 | static int __ext4_normal_writepage(struct page *page, |
de9a55b8 | 3199 | struct writeback_control *wbc) |
cf108bca JK |
3200 | { |
3201 | struct inode *inode = page->mapping->host; | |
3202 | ||
3203 | if (test_opt(inode->i_sb, NOBH)) | |
b920c755 | 3204 | return nobh_writepage(page, noalloc_get_block_write, wbc); |
cf108bca | 3205 | else |
b920c755 TT |
3206 | return block_write_full_page(page, noalloc_get_block_write, |
3207 | wbc); | |
cf108bca JK |
3208 | } |
3209 | ||
678aaf48 | 3210 | static int ext4_normal_writepage(struct page *page, |
de9a55b8 | 3211 | struct writeback_control *wbc) |
ac27a0ec DK |
3212 | { |
3213 | struct inode *inode = page->mapping->host; | |
cf108bca JK |
3214 | loff_t size = i_size_read(inode); |
3215 | loff_t len; | |
3216 | ||
9bffad1e | 3217 | trace_ext4_normal_writepage(inode, page); |
cf108bca | 3218 | J_ASSERT(PageLocked(page)); |
cf108bca JK |
3219 | if (page->index == size >> PAGE_CACHE_SHIFT) |
3220 | len = size & ~PAGE_CACHE_MASK; | |
3221 | else | |
3222 | len = PAGE_CACHE_SIZE; | |
f0e6c985 AK |
3223 | |
3224 | if (page_has_buffers(page)) { | |
3225 | /* if page has buffers it should all be mapped | |
3226 | * and allocated. If there are not buffers attached | |
3227 | * to the page we know the page is dirty but it lost | |
3228 | * buffers. That means that at some moment in time | |
3229 | * after write_begin() / write_end() has been called | |
3230 | * all buffers have been clean and thus they must have been | |
3231 | * written at least once. So they are all mapped and we can | |
3232 | * happily proceed with mapping them and writing the page. | |
3233 | */ | |
3234 | BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
3235 | ext4_bh_unmapped_or_delay)); | |
3236 | } | |
cf108bca JK |
3237 | |
3238 | if (!ext4_journal_current_handle()) | |
678aaf48 | 3239 | return __ext4_normal_writepage(page, wbc); |
cf108bca JK |
3240 | |
3241 | redirty_page_for_writepage(wbc, page); | |
3242 | unlock_page(page); | |
3243 | return 0; | |
3244 | } | |
3245 | ||
3246 | static int __ext4_journalled_writepage(struct page *page, | |
de9a55b8 | 3247 | struct writeback_control *wbc) |
cf108bca JK |
3248 | { |
3249 | struct address_space *mapping = page->mapping; | |
3250 | struct inode *inode = mapping->host; | |
3251 | struct buffer_head *page_bufs; | |
ac27a0ec DK |
3252 | handle_t *handle = NULL; |
3253 | int ret = 0; | |
3254 | int err; | |
3255 | ||
f0e6c985 | 3256 | ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, |
b920c755 | 3257 | noalloc_get_block_write); |
cf108bca JK |
3258 | if (ret != 0) |
3259 | goto out_unlock; | |
3260 | ||
3261 | page_bufs = page_buffers(page); | |
3262 | walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL, | |
3263 | bget_one); | |
3264 | /* As soon as we unlock the page, it can go away, but we have | |
3265 | * references to buffers so we are safe */ | |
3266 | unlock_page(page); | |
ac27a0ec | 3267 | |
617ba13b | 3268 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); |
ac27a0ec DK |
3269 | if (IS_ERR(handle)) { |
3270 | ret = PTR_ERR(handle); | |
cf108bca | 3271 | goto out; |
ac27a0ec DK |
3272 | } |
3273 | ||
cf108bca JK |
3274 | ret = walk_page_buffers(handle, page_bufs, 0, |
3275 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); | |
ac27a0ec | 3276 | |
cf108bca JK |
3277 | err = walk_page_buffers(handle, page_bufs, 0, |
3278 | PAGE_CACHE_SIZE, NULL, write_end_fn); | |
3279 | if (ret == 0) | |
3280 | ret = err; | |
617ba13b | 3281 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3282 | if (!ret) |
3283 | ret = err; | |
ac27a0ec | 3284 | |
cf108bca JK |
3285 | walk_page_buffers(handle, page_bufs, 0, |
3286 | PAGE_CACHE_SIZE, NULL, bput_one); | |
3287 | EXT4_I(inode)->i_state |= EXT4_STATE_JDATA; | |
3288 | goto out; | |
3289 | ||
3290 | out_unlock: | |
ac27a0ec | 3291 | unlock_page(page); |
cf108bca | 3292 | out: |
ac27a0ec DK |
3293 | return ret; |
3294 | } | |
3295 | ||
617ba13b | 3296 | static int ext4_journalled_writepage(struct page *page, |
de9a55b8 | 3297 | struct writeback_control *wbc) |
ac27a0ec DK |
3298 | { |
3299 | struct inode *inode = page->mapping->host; | |
cf108bca JK |
3300 | loff_t size = i_size_read(inode); |
3301 | loff_t len; | |
ac27a0ec | 3302 | |
9bffad1e | 3303 | trace_ext4_journalled_writepage(inode, page); |
cf108bca | 3304 | J_ASSERT(PageLocked(page)); |
cf108bca JK |
3305 | if (page->index == size >> PAGE_CACHE_SHIFT) |
3306 | len = size & ~PAGE_CACHE_MASK; | |
3307 | else | |
3308 | len = PAGE_CACHE_SIZE; | |
f0e6c985 AK |
3309 | |
3310 | if (page_has_buffers(page)) { | |
3311 | /* if page has buffers it should all be mapped | |
3312 | * and allocated. If there are not buffers attached | |
3313 | * to the page we know the page is dirty but it lost | |
3314 | * buffers. That means that at some moment in time | |
3315 | * after write_begin() / write_end() has been called | |
3316 | * all buffers have been clean and thus they must have been | |
3317 | * written at least once. So they are all mapped and we can | |
3318 | * happily proceed with mapping them and writing the page. | |
3319 | */ | |
3320 | BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
3321 | ext4_bh_unmapped_or_delay)); | |
3322 | } | |
ac27a0ec | 3323 | |
cf108bca | 3324 | if (ext4_journal_current_handle()) |
ac27a0ec | 3325 | goto no_write; |
ac27a0ec | 3326 | |
cf108bca | 3327 | if (PageChecked(page)) { |
ac27a0ec DK |
3328 | /* |
3329 | * It's mmapped pagecache. Add buffers and journal it. There | |
3330 | * doesn't seem much point in redirtying the page here. | |
3331 | */ | |
3332 | ClearPageChecked(page); | |
cf108bca | 3333 | return __ext4_journalled_writepage(page, wbc); |
ac27a0ec DK |
3334 | } else { |
3335 | /* | |
3336 | * It may be a page full of checkpoint-mode buffers. We don't | |
3337 | * really know unless we go poke around in the buffer_heads. | |
3338 | * But block_write_full_page will do the right thing. | |
3339 | */ | |
b920c755 TT |
3340 | return block_write_full_page(page, noalloc_get_block_write, |
3341 | wbc); | |
ac27a0ec | 3342 | } |
ac27a0ec DK |
3343 | no_write: |
3344 | redirty_page_for_writepage(wbc, page); | |
ac27a0ec | 3345 | unlock_page(page); |
cf108bca | 3346 | return 0; |
ac27a0ec DK |
3347 | } |
3348 | ||
617ba13b | 3349 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3350 | { |
617ba13b | 3351 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
3352 | } |
3353 | ||
3354 | static int | |
617ba13b | 3355 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3356 | struct list_head *pages, unsigned nr_pages) |
3357 | { | |
617ba13b | 3358 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3359 | } |
3360 | ||
617ba13b | 3361 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3362 | { |
617ba13b | 3363 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3364 | |
3365 | /* | |
3366 | * If it's a full truncate we just forget about the pending dirtying | |
3367 | */ | |
3368 | if (offset == 0) | |
3369 | ClearPageChecked(page); | |
3370 | ||
0390131b FM |
3371 | if (journal) |
3372 | jbd2_journal_invalidatepage(journal, page, offset); | |
3373 | else | |
3374 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
3375 | } |
3376 | ||
617ba13b | 3377 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3378 | { |
617ba13b | 3379 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec DK |
3380 | |
3381 | WARN_ON(PageChecked(page)); | |
3382 | if (!page_has_buffers(page)) | |
3383 | return 0; | |
0390131b FM |
3384 | if (journal) |
3385 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3386 | else | |
3387 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3388 | } |
3389 | ||
3390 | /* | |
3391 | * If the O_DIRECT write will extend the file then add this inode to the | |
3392 | * orphan list. So recovery will truncate it back to the original size | |
3393 | * if the machine crashes during the write. | |
3394 | * | |
3395 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
3396 | * crashes then stale disk data _may_ be exposed inside the file. But current |
3397 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 3398 | */ |
617ba13b | 3399 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
de9a55b8 TT |
3400 | const struct iovec *iov, loff_t offset, |
3401 | unsigned long nr_segs) | |
ac27a0ec DK |
3402 | { |
3403 | struct file *file = iocb->ki_filp; | |
3404 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 3405 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 3406 | handle_t *handle; |
ac27a0ec DK |
3407 | ssize_t ret; |
3408 | int orphan = 0; | |
3409 | size_t count = iov_length(iov, nr_segs); | |
3410 | ||
3411 | if (rw == WRITE) { | |
3412 | loff_t final_size = offset + count; | |
3413 | ||
ac27a0ec | 3414 | if (final_size > inode->i_size) { |
7fb5409d JK |
3415 | /* Credits for sb + inode write */ |
3416 | handle = ext4_journal_start(inode, 2); | |
3417 | if (IS_ERR(handle)) { | |
3418 | ret = PTR_ERR(handle); | |
3419 | goto out; | |
3420 | } | |
617ba13b | 3421 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
3422 | if (ret) { |
3423 | ext4_journal_stop(handle); | |
3424 | goto out; | |
3425 | } | |
ac27a0ec DK |
3426 | orphan = 1; |
3427 | ei->i_disksize = inode->i_size; | |
7fb5409d | 3428 | ext4_journal_stop(handle); |
ac27a0ec DK |
3429 | } |
3430 | } | |
3431 | ||
3432 | ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov, | |
3433 | offset, nr_segs, | |
617ba13b | 3434 | ext4_get_block, NULL); |
ac27a0ec | 3435 | |
7fb5409d | 3436 | if (orphan) { |
ac27a0ec DK |
3437 | int err; |
3438 | ||
7fb5409d JK |
3439 | /* Credits for sb + inode write */ |
3440 | handle = ext4_journal_start(inode, 2); | |
3441 | if (IS_ERR(handle)) { | |
3442 | /* This is really bad luck. We've written the data | |
3443 | * but cannot extend i_size. Bail out and pretend | |
3444 | * the write failed... */ | |
3445 | ret = PTR_ERR(handle); | |
3446 | goto out; | |
3447 | } | |
3448 | if (inode->i_nlink) | |
617ba13b | 3449 | ext4_orphan_del(handle, inode); |
7fb5409d | 3450 | if (ret > 0) { |
ac27a0ec DK |
3451 | loff_t end = offset + ret; |
3452 | if (end > inode->i_size) { | |
3453 | ei->i_disksize = end; | |
3454 | i_size_write(inode, end); | |
3455 | /* | |
3456 | * We're going to return a positive `ret' | |
3457 | * here due to non-zero-length I/O, so there's | |
3458 | * no way of reporting error returns from | |
617ba13b | 3459 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
3460 | * ignore it. |
3461 | */ | |
617ba13b | 3462 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3463 | } |
3464 | } | |
617ba13b | 3465 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3466 | if (ret == 0) |
3467 | ret = err; | |
3468 | } | |
3469 | out: | |
3470 | return ret; | |
3471 | } | |
3472 | ||
3473 | /* | |
617ba13b | 3474 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3475 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3476 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3477 | * not necessarily locked. | |
3478 | * | |
3479 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3480 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3481 | * or jbddirty because all the journalling code will explode. | |
3482 | * | |
3483 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3484 | * is called, propagate that into the buffers appropriately. | |
3485 | */ | |
617ba13b | 3486 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3487 | { |
3488 | SetPageChecked(page); | |
3489 | return __set_page_dirty_nobuffers(page); | |
3490 | } | |
3491 | ||
617ba13b | 3492 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3493 | .readpage = ext4_readpage, |
3494 | .readpages = ext4_readpages, | |
3495 | .writepage = ext4_normal_writepage, | |
3496 | .sync_page = block_sync_page, | |
3497 | .write_begin = ext4_write_begin, | |
3498 | .write_end = ext4_ordered_write_end, | |
3499 | .bmap = ext4_bmap, | |
3500 | .invalidatepage = ext4_invalidatepage, | |
3501 | .releasepage = ext4_releasepage, | |
3502 | .direct_IO = ext4_direct_IO, | |
3503 | .migratepage = buffer_migrate_page, | |
3504 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
3505 | }; |
3506 | ||
617ba13b | 3507 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3508 | .readpage = ext4_readpage, |
3509 | .readpages = ext4_readpages, | |
3510 | .writepage = ext4_normal_writepage, | |
3511 | .sync_page = block_sync_page, | |
3512 | .write_begin = ext4_write_begin, | |
3513 | .write_end = ext4_writeback_write_end, | |
3514 | .bmap = ext4_bmap, | |
3515 | .invalidatepage = ext4_invalidatepage, | |
3516 | .releasepage = ext4_releasepage, | |
3517 | .direct_IO = ext4_direct_IO, | |
3518 | .migratepage = buffer_migrate_page, | |
3519 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
3520 | }; |
3521 | ||
617ba13b | 3522 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3523 | .readpage = ext4_readpage, |
3524 | .readpages = ext4_readpages, | |
3525 | .writepage = ext4_journalled_writepage, | |
3526 | .sync_page = block_sync_page, | |
3527 | .write_begin = ext4_write_begin, | |
3528 | .write_end = ext4_journalled_write_end, | |
3529 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3530 | .bmap = ext4_bmap, | |
3531 | .invalidatepage = ext4_invalidatepage, | |
3532 | .releasepage = ext4_releasepage, | |
3533 | .is_partially_uptodate = block_is_partially_uptodate, | |
ac27a0ec DK |
3534 | }; |
3535 | ||
64769240 | 3536 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3537 | .readpage = ext4_readpage, |
3538 | .readpages = ext4_readpages, | |
3539 | .writepage = ext4_da_writepage, | |
3540 | .writepages = ext4_da_writepages, | |
3541 | .sync_page = block_sync_page, | |
3542 | .write_begin = ext4_da_write_begin, | |
3543 | .write_end = ext4_da_write_end, | |
3544 | .bmap = ext4_bmap, | |
3545 | .invalidatepage = ext4_da_invalidatepage, | |
3546 | .releasepage = ext4_releasepage, | |
3547 | .direct_IO = ext4_direct_IO, | |
3548 | .migratepage = buffer_migrate_page, | |
3549 | .is_partially_uptodate = block_is_partially_uptodate, | |
64769240 AT |
3550 | }; |
3551 | ||
617ba13b | 3552 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3553 | { |
cd1aac32 AK |
3554 | if (ext4_should_order_data(inode) && |
3555 | test_opt(inode->i_sb, DELALLOC)) | |
3556 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3557 | else if (ext4_should_order_data(inode)) | |
617ba13b | 3558 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
3559 | else if (ext4_should_writeback_data(inode) && |
3560 | test_opt(inode->i_sb, DELALLOC)) | |
3561 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
3562 | else if (ext4_should_writeback_data(inode)) |
3563 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 3564 | else |
617ba13b | 3565 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
3566 | } |
3567 | ||
3568 | /* | |
617ba13b | 3569 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
3570 | * up to the end of the block which corresponds to `from'. |
3571 | * This required during truncate. We need to physically zero the tail end | |
3572 | * of that block so it doesn't yield old data if the file is later grown. | |
3573 | */ | |
cf108bca | 3574 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
3575 | struct address_space *mapping, loff_t from) |
3576 | { | |
617ba13b | 3577 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 3578 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
3579 | unsigned blocksize, length, pos; |
3580 | ext4_lblk_t iblock; | |
ac27a0ec DK |
3581 | struct inode *inode = mapping->host; |
3582 | struct buffer_head *bh; | |
cf108bca | 3583 | struct page *page; |
ac27a0ec | 3584 | int err = 0; |
ac27a0ec | 3585 | |
cf108bca JK |
3586 | page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT); |
3587 | if (!page) | |
3588 | return -EINVAL; | |
3589 | ||
ac27a0ec DK |
3590 | blocksize = inode->i_sb->s_blocksize; |
3591 | length = blocksize - (offset & (blocksize - 1)); | |
3592 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3593 | ||
3594 | /* | |
3595 | * For "nobh" option, we can only work if we don't need to | |
3596 | * read-in the page - otherwise we create buffers to do the IO. | |
3597 | */ | |
3598 | if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) && | |
617ba13b | 3599 | ext4_should_writeback_data(inode) && PageUptodate(page)) { |
eebd2aa3 | 3600 | zero_user(page, offset, length); |
ac27a0ec DK |
3601 | set_page_dirty(page); |
3602 | goto unlock; | |
3603 | } | |
3604 | ||
3605 | if (!page_has_buffers(page)) | |
3606 | create_empty_buffers(page, blocksize, 0); | |
3607 | ||
3608 | /* Find the buffer that contains "offset" */ | |
3609 | bh = page_buffers(page); | |
3610 | pos = blocksize; | |
3611 | while (offset >= pos) { | |
3612 | bh = bh->b_this_page; | |
3613 | iblock++; | |
3614 | pos += blocksize; | |
3615 | } | |
3616 | ||
3617 | err = 0; | |
3618 | if (buffer_freed(bh)) { | |
3619 | BUFFER_TRACE(bh, "freed: skip"); | |
3620 | goto unlock; | |
3621 | } | |
3622 | ||
3623 | if (!buffer_mapped(bh)) { | |
3624 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3625 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3626 | /* unmapped? It's a hole - nothing to do */ |
3627 | if (!buffer_mapped(bh)) { | |
3628 | BUFFER_TRACE(bh, "still unmapped"); | |
3629 | goto unlock; | |
3630 | } | |
3631 | } | |
3632 | ||
3633 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3634 | if (PageUptodate(page)) | |
3635 | set_buffer_uptodate(bh); | |
3636 | ||
3637 | if (!buffer_uptodate(bh)) { | |
3638 | err = -EIO; | |
3639 | ll_rw_block(READ, 1, &bh); | |
3640 | wait_on_buffer(bh); | |
3641 | /* Uhhuh. Read error. Complain and punt. */ | |
3642 | if (!buffer_uptodate(bh)) | |
3643 | goto unlock; | |
3644 | } | |
3645 | ||
617ba13b | 3646 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3647 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3648 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3649 | if (err) |
3650 | goto unlock; | |
3651 | } | |
3652 | ||
eebd2aa3 | 3653 | zero_user(page, offset, length); |
ac27a0ec DK |
3654 | |
3655 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3656 | ||
3657 | err = 0; | |
617ba13b | 3658 | if (ext4_should_journal_data(inode)) { |
0390131b | 3659 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
ac27a0ec | 3660 | } else { |
617ba13b | 3661 | if (ext4_should_order_data(inode)) |
678aaf48 | 3662 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
3663 | mark_buffer_dirty(bh); |
3664 | } | |
3665 | ||
3666 | unlock: | |
3667 | unlock_page(page); | |
3668 | page_cache_release(page); | |
3669 | return err; | |
3670 | } | |
3671 | ||
3672 | /* | |
3673 | * Probably it should be a library function... search for first non-zero word | |
3674 | * or memcmp with zero_page, whatever is better for particular architecture. | |
3675 | * Linus? | |
3676 | */ | |
3677 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
3678 | { | |
3679 | while (p < q) | |
3680 | if (*p++) | |
3681 | return 0; | |
3682 | return 1; | |
3683 | } | |
3684 | ||
3685 | /** | |
617ba13b | 3686 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
3687 | * @inode: inode in question |
3688 | * @depth: depth of the affected branch | |
617ba13b | 3689 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
3690 | * @chain: place to store the pointers to partial indirect blocks |
3691 | * @top: place to the (detached) top of branch | |
3692 | * | |
617ba13b | 3693 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
3694 | * |
3695 | * When we do truncate() we may have to clean the ends of several | |
3696 | * indirect blocks but leave the blocks themselves alive. Block is | |
3697 | * partially truncated if some data below the new i_size is refered | |
3698 | * from it (and it is on the path to the first completely truncated | |
3699 | * data block, indeed). We have to free the top of that path along | |
3700 | * with everything to the right of the path. Since no allocation | |
617ba13b | 3701 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
3702 | * finishes, we may safely do the latter, but top of branch may |
3703 | * require special attention - pageout below the truncation point | |
3704 | * might try to populate it. | |
3705 | * | |
3706 | * We atomically detach the top of branch from the tree, store the | |
3707 | * block number of its root in *@top, pointers to buffer_heads of | |
3708 | * partially truncated blocks - in @chain[].bh and pointers to | |
3709 | * their last elements that should not be removed - in | |
3710 | * @chain[].p. Return value is the pointer to last filled element | |
3711 | * of @chain. | |
3712 | * | |
3713 | * The work left to caller to do the actual freeing of subtrees: | |
3714 | * a) free the subtree starting from *@top | |
3715 | * b) free the subtrees whose roots are stored in | |
3716 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
3717 | * c) free the subtrees growing from the inode past the @chain[0]. | |
3718 | * (no partially truncated stuff there). */ | |
3719 | ||
617ba13b | 3720 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
de9a55b8 TT |
3721 | ext4_lblk_t offsets[4], Indirect chain[4], |
3722 | __le32 *top) | |
ac27a0ec DK |
3723 | { |
3724 | Indirect *partial, *p; | |
3725 | int k, err; | |
3726 | ||
3727 | *top = 0; | |
3728 | /* Make k index the deepest non-null offest + 1 */ | |
3729 | for (k = depth; k > 1 && !offsets[k-1]; k--) | |
3730 | ; | |
617ba13b | 3731 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
3732 | /* Writer: pointers */ |
3733 | if (!partial) | |
3734 | partial = chain + k-1; | |
3735 | /* | |
3736 | * If the branch acquired continuation since we've looked at it - | |
3737 | * fine, it should all survive and (new) top doesn't belong to us. | |
3738 | */ | |
3739 | if (!partial->key && *partial->p) | |
3740 | /* Writer: end */ | |
3741 | goto no_top; | |
af5bc92d | 3742 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
ac27a0ec DK |
3743 | ; |
3744 | /* | |
3745 | * OK, we've found the last block that must survive. The rest of our | |
3746 | * branch should be detached before unlocking. However, if that rest | |
3747 | * of branch is all ours and does not grow immediately from the inode | |
3748 | * it's easier to cheat and just decrement partial->p. | |
3749 | */ | |
3750 | if (p == chain + k - 1 && p > chain) { | |
3751 | p->p--; | |
3752 | } else { | |
3753 | *top = *p->p; | |
617ba13b | 3754 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
3755 | #if 0 |
3756 | *p->p = 0; | |
3757 | #endif | |
3758 | } | |
3759 | /* Writer: end */ | |
3760 | ||
af5bc92d | 3761 | while (partial > p) { |
ac27a0ec DK |
3762 | brelse(partial->bh); |
3763 | partial--; | |
3764 | } | |
3765 | no_top: | |
3766 | return partial; | |
3767 | } | |
3768 | ||
3769 | /* | |
3770 | * Zero a number of block pointers in either an inode or an indirect block. | |
3771 | * If we restart the transaction we must again get write access to the | |
3772 | * indirect block for further modification. | |
3773 | * | |
3774 | * We release `count' blocks on disk, but (last - first) may be greater | |
3775 | * than `count' because there can be holes in there. | |
3776 | */ | |
617ba13b | 3777 | static void ext4_clear_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
3778 | struct buffer_head *bh, |
3779 | ext4_fsblk_t block_to_free, | |
3780 | unsigned long count, __le32 *first, | |
3781 | __le32 *last) | |
ac27a0ec DK |
3782 | { |
3783 | __le32 *p; | |
3784 | if (try_to_extend_transaction(handle, inode)) { | |
3785 | if (bh) { | |
0390131b FM |
3786 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
3787 | ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec | 3788 | } |
617ba13b MC |
3789 | ext4_mark_inode_dirty(handle, inode); |
3790 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
3791 | if (bh) { |
3792 | BUFFER_TRACE(bh, "retaking write access"); | |
617ba13b | 3793 | ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3794 | } |
3795 | } | |
3796 | ||
3797 | /* | |
de9a55b8 TT |
3798 | * Any buffers which are on the journal will be in memory. We |
3799 | * find them on the hash table so jbd2_journal_revoke() will | |
3800 | * run jbd2_journal_forget() on them. We've already detached | |
3801 | * each block from the file, so bforget() in | |
3802 | * jbd2_journal_forget() should be safe. | |
ac27a0ec | 3803 | * |
dab291af | 3804 | * AKPM: turn on bforget in jbd2_journal_forget()!!! |
ac27a0ec DK |
3805 | */ |
3806 | for (p = first; p < last; p++) { | |
3807 | u32 nr = le32_to_cpu(*p); | |
3808 | if (nr) { | |
1d03ec98 | 3809 | struct buffer_head *tbh; |
ac27a0ec DK |
3810 | |
3811 | *p = 0; | |
1d03ec98 AK |
3812 | tbh = sb_find_get_block(inode->i_sb, nr); |
3813 | ext4_forget(handle, 0, inode, tbh, nr); | |
ac27a0ec DK |
3814 | } |
3815 | } | |
3816 | ||
c9de560d | 3817 | ext4_free_blocks(handle, inode, block_to_free, count, 0); |
ac27a0ec DK |
3818 | } |
3819 | ||
3820 | /** | |
617ba13b | 3821 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
3822 | * @handle: handle for this transaction |
3823 | * @inode: inode we are dealing with | |
3824 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
3825 | * @first: array of block numbers | |
3826 | * @last: points immediately past the end of array | |
3827 | * | |
3828 | * We are freeing all blocks refered from that array (numbers are stored as | |
3829 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. | |
3830 | * | |
3831 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
3832 | * blocks are contiguous then releasing them at one time will only affect one | |
3833 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
3834 | * actually use a lot of journal space. | |
3835 | * | |
3836 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
3837 | * block pointers. | |
3838 | */ | |
617ba13b | 3839 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
3840 | struct buffer_head *this_bh, |
3841 | __le32 *first, __le32 *last) | |
3842 | { | |
617ba13b | 3843 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
3844 | unsigned long count = 0; /* Number of blocks in the run */ |
3845 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
3846 | corresponding to | |
3847 | block_to_free */ | |
617ba13b | 3848 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
3849 | __le32 *p; /* Pointer into inode/ind |
3850 | for current block */ | |
3851 | int err; | |
3852 | ||
3853 | if (this_bh) { /* For indirect block */ | |
3854 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 3855 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
3856 | /* Important: if we can't update the indirect pointers |
3857 | * to the blocks, we can't free them. */ | |
3858 | if (err) | |
3859 | return; | |
3860 | } | |
3861 | ||
3862 | for (p = first; p < last; p++) { | |
3863 | nr = le32_to_cpu(*p); | |
3864 | if (nr) { | |
3865 | /* accumulate blocks to free if they're contiguous */ | |
3866 | if (count == 0) { | |
3867 | block_to_free = nr; | |
3868 | block_to_free_p = p; | |
3869 | count = 1; | |
3870 | } else if (nr == block_to_free + count) { | |
3871 | count++; | |
3872 | } else { | |
617ba13b | 3873 | ext4_clear_blocks(handle, inode, this_bh, |
ac27a0ec DK |
3874 | block_to_free, |
3875 | count, block_to_free_p, p); | |
3876 | block_to_free = nr; | |
3877 | block_to_free_p = p; | |
3878 | count = 1; | |
3879 | } | |
3880 | } | |
3881 | } | |
3882 | ||
3883 | if (count > 0) | |
617ba13b | 3884 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
ac27a0ec DK |
3885 | count, block_to_free_p, p); |
3886 | ||
3887 | if (this_bh) { | |
0390131b | 3888 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
71dc8fbc DG |
3889 | |
3890 | /* | |
3891 | * The buffer head should have an attached journal head at this | |
3892 | * point. However, if the data is corrupted and an indirect | |
3893 | * block pointed to itself, it would have been detached when | |
3894 | * the block was cleared. Check for this instead of OOPSing. | |
3895 | */ | |
e7f07968 | 3896 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
0390131b | 3897 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
71dc8fbc DG |
3898 | else |
3899 | ext4_error(inode->i_sb, __func__, | |
3900 | "circular indirect block detected, " | |
3901 | "inode=%lu, block=%llu", | |
3902 | inode->i_ino, | |
3903 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
3904 | } |
3905 | } | |
3906 | ||
3907 | /** | |
617ba13b | 3908 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
3909 | * @handle: JBD handle for this transaction |
3910 | * @inode: inode we are dealing with | |
3911 | * @parent_bh: the buffer_head which contains *@first and *@last | |
3912 | * @first: array of block numbers | |
3913 | * @last: pointer immediately past the end of array | |
3914 | * @depth: depth of the branches to free | |
3915 | * | |
3916 | * We are freeing all blocks refered from these branches (numbers are | |
3917 | * stored as little-endian 32-bit) and updating @inode->i_blocks | |
3918 | * appropriately. | |
3919 | */ | |
617ba13b | 3920 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
3921 | struct buffer_head *parent_bh, |
3922 | __le32 *first, __le32 *last, int depth) | |
3923 | { | |
617ba13b | 3924 | ext4_fsblk_t nr; |
ac27a0ec DK |
3925 | __le32 *p; |
3926 | ||
0390131b | 3927 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
3928 | return; |
3929 | ||
3930 | if (depth--) { | |
3931 | struct buffer_head *bh; | |
617ba13b | 3932 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
3933 | p = last; |
3934 | while (--p >= first) { | |
3935 | nr = le32_to_cpu(*p); | |
3936 | if (!nr) | |
3937 | continue; /* A hole */ | |
3938 | ||
3939 | /* Go read the buffer for the next level down */ | |
3940 | bh = sb_bread(inode->i_sb, nr); | |
3941 | ||
3942 | /* | |
3943 | * A read failure? Report error and clear slot | |
3944 | * (should be rare). | |
3945 | */ | |
3946 | if (!bh) { | |
617ba13b | 3947 | ext4_error(inode->i_sb, "ext4_free_branches", |
2ae02107 | 3948 | "Read failure, inode=%lu, block=%llu", |
ac27a0ec DK |
3949 | inode->i_ino, nr); |
3950 | continue; | |
3951 | } | |
3952 | ||
3953 | /* This zaps the entire block. Bottom up. */ | |
3954 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 3955 | ext4_free_branches(handle, inode, bh, |
af5bc92d TT |
3956 | (__le32 *) bh->b_data, |
3957 | (__le32 *) bh->b_data + addr_per_block, | |
3958 | depth); | |
ac27a0ec DK |
3959 | |
3960 | /* | |
3961 | * We've probably journalled the indirect block several | |
3962 | * times during the truncate. But it's no longer | |
3963 | * needed and we now drop it from the transaction via | |
dab291af | 3964 | * jbd2_journal_revoke(). |
ac27a0ec DK |
3965 | * |
3966 | * That's easy if it's exclusively part of this | |
3967 | * transaction. But if it's part of the committing | |
dab291af | 3968 | * transaction then jbd2_journal_forget() will simply |
ac27a0ec | 3969 | * brelse() it. That means that if the underlying |
617ba13b | 3970 | * block is reallocated in ext4_get_block(), |
ac27a0ec DK |
3971 | * unmap_underlying_metadata() will find this block |
3972 | * and will try to get rid of it. damn, damn. | |
3973 | * | |
3974 | * If this block has already been committed to the | |
3975 | * journal, a revoke record will be written. And | |
3976 | * revoke records must be emitted *before* clearing | |
3977 | * this block's bit in the bitmaps. | |
3978 | */ | |
617ba13b | 3979 | ext4_forget(handle, 1, inode, bh, bh->b_blocknr); |
ac27a0ec DK |
3980 | |
3981 | /* | |
3982 | * Everything below this this pointer has been | |
3983 | * released. Now let this top-of-subtree go. | |
3984 | * | |
3985 | * We want the freeing of this indirect block to be | |
3986 | * atomic in the journal with the updating of the | |
3987 | * bitmap block which owns it. So make some room in | |
3988 | * the journal. | |
3989 | * | |
3990 | * We zero the parent pointer *after* freeing its | |
3991 | * pointee in the bitmaps, so if extend_transaction() | |
3992 | * for some reason fails to put the bitmap changes and | |
3993 | * the release into the same transaction, recovery | |
3994 | * will merely complain about releasing a free block, | |
3995 | * rather than leaking blocks. | |
3996 | */ | |
0390131b | 3997 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
3998 | return; |
3999 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b MC |
4000 | ext4_mark_inode_dirty(handle, inode); |
4001 | ext4_journal_test_restart(handle, inode); | |
ac27a0ec DK |
4002 | } |
4003 | ||
c9de560d | 4004 | ext4_free_blocks(handle, inode, nr, 1, 1); |
ac27a0ec DK |
4005 | |
4006 | if (parent_bh) { | |
4007 | /* | |
4008 | * The block which we have just freed is | |
4009 | * pointed to by an indirect block: journal it | |
4010 | */ | |
4011 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 4012 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
4013 | parent_bh)){ |
4014 | *p = 0; | |
4015 | BUFFER_TRACE(parent_bh, | |
0390131b FM |
4016 | "call ext4_handle_dirty_metadata"); |
4017 | ext4_handle_dirty_metadata(handle, | |
4018 | inode, | |
4019 | parent_bh); | |
ac27a0ec DK |
4020 | } |
4021 | } | |
4022 | } | |
4023 | } else { | |
4024 | /* We have reached the bottom of the tree. */ | |
4025 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 4026 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
4027 | } |
4028 | } | |
4029 | ||
91ef4caf DG |
4030 | int ext4_can_truncate(struct inode *inode) |
4031 | { | |
4032 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4033 | return 0; | |
4034 | if (S_ISREG(inode->i_mode)) | |
4035 | return 1; | |
4036 | if (S_ISDIR(inode->i_mode)) | |
4037 | return 1; | |
4038 | if (S_ISLNK(inode->i_mode)) | |
4039 | return !ext4_inode_is_fast_symlink(inode); | |
4040 | return 0; | |
4041 | } | |
4042 | ||
ac27a0ec | 4043 | /* |
617ba13b | 4044 | * ext4_truncate() |
ac27a0ec | 4045 | * |
617ba13b MC |
4046 | * We block out ext4_get_block() block instantiations across the entire |
4047 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4048 | * simultaneously on behalf of the same inode. |
4049 | * | |
4050 | * As we work through the truncate and commmit bits of it to the journal there | |
4051 | * is one core, guiding principle: the file's tree must always be consistent on | |
4052 | * disk. We must be able to restart the truncate after a crash. | |
4053 | * | |
4054 | * The file's tree may be transiently inconsistent in memory (although it | |
4055 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4056 | * the contents of (the filesystem + the journal) must be consistent and | |
4057 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4058 | * left-to-right works OK too). | |
4059 | * | |
4060 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4061 | * truncate against the orphan inode list. | |
4062 | * | |
4063 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4064 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4065 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4066 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4067 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4068 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4069 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4070 | */ |
617ba13b | 4071 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
4072 | { |
4073 | handle_t *handle; | |
617ba13b | 4074 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 4075 | __le32 *i_data = ei->i_data; |
617ba13b | 4076 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 4077 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 4078 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
4079 | Indirect chain[4]; |
4080 | Indirect *partial; | |
4081 | __le32 nr = 0; | |
4082 | int n; | |
725d26d3 | 4083 | ext4_lblk_t last_block; |
ac27a0ec | 4084 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 4085 | |
91ef4caf | 4086 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4087 | return; |
4088 | ||
0eab9282 TT |
4089 | if (ei->i_disksize && inode->i_size == 0 && |
4090 | !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) | |
7d8f9f7d TT |
4091 | ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE; |
4092 | ||
1d03ec98 | 4093 | if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { |
cf108bca | 4094 | ext4_ext_truncate(inode); |
1d03ec98 AK |
4095 | return; |
4096 | } | |
a86c6181 | 4097 | |
ac27a0ec | 4098 | handle = start_transaction(inode); |
cf108bca | 4099 | if (IS_ERR(handle)) |
ac27a0ec | 4100 | return; /* AKPM: return what? */ |
ac27a0ec DK |
4101 | |
4102 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 4103 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
ac27a0ec | 4104 | |
cf108bca JK |
4105 | if (inode->i_size & (blocksize - 1)) |
4106 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
4107 | goto out_stop; | |
ac27a0ec | 4108 | |
617ba13b | 4109 | n = ext4_block_to_path(inode, last_block, offsets, NULL); |
ac27a0ec DK |
4110 | if (n == 0) |
4111 | goto out_stop; /* error */ | |
4112 | ||
4113 | /* | |
4114 | * OK. This truncate is going to happen. We add the inode to the | |
4115 | * orphan list, so that if this truncate spans multiple transactions, | |
4116 | * and we crash, we will resume the truncate when the filesystem | |
4117 | * recovers. It also marks the inode dirty, to catch the new size. | |
4118 | * | |
4119 | * Implication: the file must always be in a sane, consistent | |
4120 | * truncatable state while each transaction commits. | |
4121 | */ | |
617ba13b | 4122 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
4123 | goto out_stop; |
4124 | ||
632eaeab MC |
4125 | /* |
4126 | * From here we block out all ext4_get_block() callers who want to | |
4127 | * modify the block allocation tree. | |
4128 | */ | |
4129 | down_write(&ei->i_data_sem); | |
b4df2030 | 4130 | |
c2ea3fde | 4131 | ext4_discard_preallocations(inode); |
b4df2030 | 4132 | |
ac27a0ec DK |
4133 | /* |
4134 | * The orphan list entry will now protect us from any crash which | |
4135 | * occurs before the truncate completes, so it is now safe to propagate | |
4136 | * the new, shorter inode size (held for now in i_size) into the | |
4137 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 4138 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
4139 | */ |
4140 | ei->i_disksize = inode->i_size; | |
4141 | ||
ac27a0ec | 4142 | if (n == 1) { /* direct blocks */ |
617ba13b MC |
4143 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
4144 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
4145 | goto do_indirects; |
4146 | } | |
4147 | ||
617ba13b | 4148 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
4149 | /* Kill the top of shared branch (not detached) */ |
4150 | if (nr) { | |
4151 | if (partial == chain) { | |
4152 | /* Shared branch grows from the inode */ | |
617ba13b | 4153 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
4154 | &nr, &nr+1, (chain+n-1) - partial); |
4155 | *partial->p = 0; | |
4156 | /* | |
4157 | * We mark the inode dirty prior to restart, | |
4158 | * and prior to stop. No need for it here. | |
4159 | */ | |
4160 | } else { | |
4161 | /* Shared branch grows from an indirect block */ | |
4162 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 4163 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
4164 | partial->p, |
4165 | partial->p+1, (chain+n-1) - partial); | |
4166 | } | |
4167 | } | |
4168 | /* Clear the ends of indirect blocks on the shared branch */ | |
4169 | while (partial > chain) { | |
617ba13b | 4170 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
4171 | (__le32*)partial->bh->b_data+addr_per_block, |
4172 | (chain+n-1) - partial); | |
4173 | BUFFER_TRACE(partial->bh, "call brelse"); | |
de9a55b8 | 4174 | brelse(partial->bh); |
ac27a0ec DK |
4175 | partial--; |
4176 | } | |
4177 | do_indirects: | |
4178 | /* Kill the remaining (whole) subtrees */ | |
4179 | switch (offsets[0]) { | |
4180 | default: | |
617ba13b | 4181 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 4182 | if (nr) { |
617ba13b MC |
4183 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
4184 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 4185 | } |
617ba13b MC |
4186 | case EXT4_IND_BLOCK: |
4187 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 4188 | if (nr) { |
617ba13b MC |
4189 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
4190 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 4191 | } |
617ba13b MC |
4192 | case EXT4_DIND_BLOCK: |
4193 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 4194 | if (nr) { |
617ba13b MC |
4195 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
4196 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 4197 | } |
617ba13b | 4198 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
4199 | ; |
4200 | } | |
4201 | ||
0e855ac8 | 4202 | up_write(&ei->i_data_sem); |
ef7f3835 | 4203 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 4204 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
4205 | |
4206 | /* | |
4207 | * In a multi-transaction truncate, we only make the final transaction | |
4208 | * synchronous | |
4209 | */ | |
4210 | if (IS_SYNC(inode)) | |
0390131b | 4211 | ext4_handle_sync(handle); |
ac27a0ec DK |
4212 | out_stop: |
4213 | /* | |
4214 | * If this was a simple ftruncate(), and the file will remain alive | |
4215 | * then we need to clear up the orphan record which we created above. | |
4216 | * However, if this was a real unlink then we were called by | |
617ba13b | 4217 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
4218 | * orphan info for us. |
4219 | */ | |
4220 | if (inode->i_nlink) | |
617ba13b | 4221 | ext4_orphan_del(handle, inode); |
ac27a0ec | 4222 | |
617ba13b | 4223 | ext4_journal_stop(handle); |
ac27a0ec DK |
4224 | } |
4225 | ||
ac27a0ec | 4226 | /* |
617ba13b | 4227 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4228 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4229 | * data in memory that is needed to recreate the on-disk version of this | |
4230 | * inode. | |
4231 | */ | |
617ba13b MC |
4232 | static int __ext4_get_inode_loc(struct inode *inode, |
4233 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4234 | { |
240799cd TT |
4235 | struct ext4_group_desc *gdp; |
4236 | struct buffer_head *bh; | |
4237 | struct super_block *sb = inode->i_sb; | |
4238 | ext4_fsblk_t block; | |
4239 | int inodes_per_block, inode_offset; | |
4240 | ||
3a06d778 | 4241 | iloc->bh = NULL; |
240799cd TT |
4242 | if (!ext4_valid_inum(sb, inode->i_ino)) |
4243 | return -EIO; | |
ac27a0ec | 4244 | |
240799cd TT |
4245 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4246 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4247 | if (!gdp) | |
ac27a0ec DK |
4248 | return -EIO; |
4249 | ||
240799cd TT |
4250 | /* |
4251 | * Figure out the offset within the block group inode table | |
4252 | */ | |
4253 | inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb)); | |
4254 | inode_offset = ((inode->i_ino - 1) % | |
4255 | EXT4_INODES_PER_GROUP(sb)); | |
4256 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4257 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4258 | ||
4259 | bh = sb_getblk(sb, block); | |
ac27a0ec | 4260 | if (!bh) { |
240799cd TT |
4261 | ext4_error(sb, "ext4_get_inode_loc", "unable to read " |
4262 | "inode block - inode=%lu, block=%llu", | |
4263 | inode->i_ino, block); | |
ac27a0ec DK |
4264 | return -EIO; |
4265 | } | |
4266 | if (!buffer_uptodate(bh)) { | |
4267 | lock_buffer(bh); | |
9c83a923 HK |
4268 | |
4269 | /* | |
4270 | * If the buffer has the write error flag, we have failed | |
4271 | * to write out another inode in the same block. In this | |
4272 | * case, we don't have to read the block because we may | |
4273 | * read the old inode data successfully. | |
4274 | */ | |
4275 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4276 | set_buffer_uptodate(bh); | |
4277 | ||
ac27a0ec DK |
4278 | if (buffer_uptodate(bh)) { |
4279 | /* someone brought it uptodate while we waited */ | |
4280 | unlock_buffer(bh); | |
4281 | goto has_buffer; | |
4282 | } | |
4283 | ||
4284 | /* | |
4285 | * If we have all information of the inode in memory and this | |
4286 | * is the only valid inode in the block, we need not read the | |
4287 | * block. | |
4288 | */ | |
4289 | if (in_mem) { | |
4290 | struct buffer_head *bitmap_bh; | |
240799cd | 4291 | int i, start; |
ac27a0ec | 4292 | |
240799cd | 4293 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4294 | |
240799cd TT |
4295 | /* Is the inode bitmap in cache? */ |
4296 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
4297 | if (!bitmap_bh) |
4298 | goto make_io; | |
4299 | ||
4300 | /* | |
4301 | * If the inode bitmap isn't in cache then the | |
4302 | * optimisation may end up performing two reads instead | |
4303 | * of one, so skip it. | |
4304 | */ | |
4305 | if (!buffer_uptodate(bitmap_bh)) { | |
4306 | brelse(bitmap_bh); | |
4307 | goto make_io; | |
4308 | } | |
240799cd | 4309 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4310 | if (i == inode_offset) |
4311 | continue; | |
617ba13b | 4312 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4313 | break; |
4314 | } | |
4315 | brelse(bitmap_bh); | |
240799cd | 4316 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4317 | /* all other inodes are free, so skip I/O */ |
4318 | memset(bh->b_data, 0, bh->b_size); | |
4319 | set_buffer_uptodate(bh); | |
4320 | unlock_buffer(bh); | |
4321 | goto has_buffer; | |
4322 | } | |
4323 | } | |
4324 | ||
4325 | make_io: | |
240799cd TT |
4326 | /* |
4327 | * If we need to do any I/O, try to pre-readahead extra | |
4328 | * blocks from the inode table. | |
4329 | */ | |
4330 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4331 | ext4_fsblk_t b, end, table; | |
4332 | unsigned num; | |
4333 | ||
4334 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4335 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
4336 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
4337 | if (table > b) | |
4338 | b = table; | |
4339 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
4340 | num = EXT4_INODES_PER_GROUP(sb); | |
4341 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4342 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 4343 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4344 | table += num / inodes_per_block; |
4345 | if (end > table) | |
4346 | end = table; | |
4347 | while (b <= end) | |
4348 | sb_breadahead(sb, b++); | |
4349 | } | |
4350 | ||
ac27a0ec DK |
4351 | /* |
4352 | * There are other valid inodes in the buffer, this inode | |
4353 | * has in-inode xattrs, or we don't have this inode in memory. | |
4354 | * Read the block from disk. | |
4355 | */ | |
4356 | get_bh(bh); | |
4357 | bh->b_end_io = end_buffer_read_sync; | |
4358 | submit_bh(READ_META, bh); | |
4359 | wait_on_buffer(bh); | |
4360 | if (!buffer_uptodate(bh)) { | |
240799cd TT |
4361 | ext4_error(sb, __func__, |
4362 | "unable to read inode block - inode=%lu, " | |
4363 | "block=%llu", inode->i_ino, block); | |
ac27a0ec DK |
4364 | brelse(bh); |
4365 | return -EIO; | |
4366 | } | |
4367 | } | |
4368 | has_buffer: | |
4369 | iloc->bh = bh; | |
4370 | return 0; | |
4371 | } | |
4372 | ||
617ba13b | 4373 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4374 | { |
4375 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b MC |
4376 | return __ext4_get_inode_loc(inode, iloc, |
4377 | !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR)); | |
ac27a0ec DK |
4378 | } |
4379 | ||
617ba13b | 4380 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4381 | { |
617ba13b | 4382 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4383 | |
4384 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4385 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4386 | inode->i_flags |= S_SYNC; |
617ba13b | 4387 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4388 | inode->i_flags |= S_APPEND; |
617ba13b | 4389 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4390 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4391 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4392 | inode->i_flags |= S_NOATIME; |
617ba13b | 4393 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4394 | inode->i_flags |= S_DIRSYNC; |
4395 | } | |
4396 | ||
ff9ddf7e JK |
4397 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4398 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4399 | { | |
4400 | unsigned int flags = ei->vfs_inode.i_flags; | |
4401 | ||
4402 | ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4403 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL); | |
4404 | if (flags & S_SYNC) | |
4405 | ei->i_flags |= EXT4_SYNC_FL; | |
4406 | if (flags & S_APPEND) | |
4407 | ei->i_flags |= EXT4_APPEND_FL; | |
4408 | if (flags & S_IMMUTABLE) | |
4409 | ei->i_flags |= EXT4_IMMUTABLE_FL; | |
4410 | if (flags & S_NOATIME) | |
4411 | ei->i_flags |= EXT4_NOATIME_FL; | |
4412 | if (flags & S_DIRSYNC) | |
4413 | ei->i_flags |= EXT4_DIRSYNC_FL; | |
4414 | } | |
de9a55b8 | 4415 | |
0fc1b451 | 4416 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4417 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4418 | { |
4419 | blkcnt_t i_blocks ; | |
8180a562 AK |
4420 | struct inode *inode = &(ei->vfs_inode); |
4421 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4422 | |
4423 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4424 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4425 | /* we are using combined 48 bit field */ | |
4426 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4427 | le32_to_cpu(raw_inode->i_blocks_lo); | |
8180a562 AK |
4428 | if (ei->i_flags & EXT4_HUGE_FILE_FL) { |
4429 | /* i_blocks represent file system block size */ | |
4430 | return i_blocks << (inode->i_blkbits - 9); | |
4431 | } else { | |
4432 | return i_blocks; | |
4433 | } | |
0fc1b451 AK |
4434 | } else { |
4435 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4436 | } | |
4437 | } | |
ff9ddf7e | 4438 | |
1d1fe1ee | 4439 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4440 | { |
617ba13b MC |
4441 | struct ext4_iloc iloc; |
4442 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4443 | struct ext4_inode_info *ei; |
ac27a0ec | 4444 | struct buffer_head *bh; |
1d1fe1ee DH |
4445 | struct inode *inode; |
4446 | long ret; | |
ac27a0ec DK |
4447 | int block; |
4448 | ||
1d1fe1ee DH |
4449 | inode = iget_locked(sb, ino); |
4450 | if (!inode) | |
4451 | return ERR_PTR(-ENOMEM); | |
4452 | if (!(inode->i_state & I_NEW)) | |
4453 | return inode; | |
4454 | ||
4455 | ei = EXT4_I(inode); | |
ac27a0ec | 4456 | |
1d1fe1ee DH |
4457 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4458 | if (ret < 0) | |
ac27a0ec DK |
4459 | goto bad_inode; |
4460 | bh = iloc.bh; | |
617ba13b | 4461 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
4462 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
4463 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
4464 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4465 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4466 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4467 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
4468 | } | |
4469 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec DK |
4470 | |
4471 | ei->i_state = 0; | |
4472 | ei->i_dir_start_lookup = 0; | |
4473 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4474 | /* We now have enough fields to check if the inode was active or not. | |
4475 | * This is needed because nfsd might try to access dead inodes | |
4476 | * the test is that same one that e2fsck uses | |
4477 | * NeilBrown 1999oct15 | |
4478 | */ | |
4479 | if (inode->i_nlink == 0) { | |
4480 | if (inode->i_mode == 0 || | |
617ba13b | 4481 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 4482 | /* this inode is deleted */ |
af5bc92d | 4483 | brelse(bh); |
1d1fe1ee | 4484 | ret = -ESTALE; |
ac27a0ec DK |
4485 | goto bad_inode; |
4486 | } | |
4487 | /* The only unlinked inodes we let through here have | |
4488 | * valid i_mode and are being read by the orphan | |
4489 | * recovery code: that's fine, we're about to complete | |
4490 | * the process of deleting those. */ | |
4491 | } | |
ac27a0ec | 4492 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4493 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4494 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4495 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4496 | ei->i_file_acl |= |
4497 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4498 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec DK |
4499 | ei->i_disksize = inode->i_size; |
4500 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); | |
4501 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4502 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4503 | /* |
4504 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4505 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4506 | */ | |
617ba13b | 4507 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4508 | ei->i_data[block] = raw_inode->i_block[block]; |
4509 | INIT_LIST_HEAD(&ei->i_orphan); | |
4510 | ||
0040d987 | 4511 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 4512 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 4513 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 4514 | EXT4_INODE_SIZE(inode->i_sb)) { |
af5bc92d | 4515 | brelse(bh); |
1d1fe1ee | 4516 | ret = -EIO; |
ac27a0ec | 4517 | goto bad_inode; |
e5d2861f | 4518 | } |
ac27a0ec DK |
4519 | if (ei->i_extra_isize == 0) { |
4520 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4521 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4522 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
4523 | } else { |
4524 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 4525 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 4526 | ei->i_extra_isize; |
617ba13b | 4527 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
de9a55b8 | 4528 | ei->i_state |= EXT4_STATE_XATTR; |
ac27a0ec DK |
4529 | } |
4530 | } else | |
4531 | ei->i_extra_isize = 0; | |
4532 | ||
ef7f3835 KS |
4533 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4534 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4535 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4536 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4537 | ||
25ec56b5 JNC |
4538 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4539 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4540 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4541 | inode->i_version |= | |
4542 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4543 | } | |
4544 | ||
c4b5a614 | 4545 | ret = 0; |
485c26ec | 4546 | if (ei->i_file_acl && |
de9a55b8 | 4547 | ((ei->i_file_acl < |
485c26ec TT |
4548 | (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) + |
4549 | EXT4_SB(sb)->s_gdb_count)) || | |
4550 | (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) { | |
4551 | ext4_error(sb, __func__, | |
4552 | "bad extended attribute block %llu in inode #%lu", | |
4553 | ei->i_file_acl, inode->i_ino); | |
4554 | ret = -EIO; | |
4555 | goto bad_inode; | |
4556 | } else if (ei->i_flags & EXT4_EXTENTS_FL) { | |
c4b5a614 TT |
4557 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
4558 | (S_ISLNK(inode->i_mode) && | |
4559 | !ext4_inode_is_fast_symlink(inode))) | |
4560 | /* Validate extent which is part of inode */ | |
4561 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 4562 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
4563 | (S_ISLNK(inode->i_mode) && |
4564 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 4565 | /* Validate block references which are part of inode */ |
fe2c8191 TN |
4566 | ret = ext4_check_inode_blockref(inode); |
4567 | } | |
4568 | if (ret) { | |
de9a55b8 TT |
4569 | brelse(bh); |
4570 | goto bad_inode; | |
7a262f7c AK |
4571 | } |
4572 | ||
ac27a0ec | 4573 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4574 | inode->i_op = &ext4_file_inode_operations; |
4575 | inode->i_fop = &ext4_file_operations; | |
4576 | ext4_set_aops(inode); | |
ac27a0ec | 4577 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4578 | inode->i_op = &ext4_dir_inode_operations; |
4579 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4580 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4581 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4582 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4583 | nd_terminate_link(ei->i_data, inode->i_size, |
4584 | sizeof(ei->i_data) - 1); | |
4585 | } else { | |
617ba13b MC |
4586 | inode->i_op = &ext4_symlink_inode_operations; |
4587 | ext4_set_aops(inode); | |
ac27a0ec | 4588 | } |
563bdd61 TT |
4589 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4590 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4591 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4592 | if (raw_inode->i_block[0]) |
4593 | init_special_inode(inode, inode->i_mode, | |
4594 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4595 | else | |
4596 | init_special_inode(inode, inode->i_mode, | |
4597 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 TT |
4598 | } else { |
4599 | brelse(bh); | |
4600 | ret = -EIO; | |
de9a55b8 | 4601 | ext4_error(inode->i_sb, __func__, |
563bdd61 TT |
4602 | "bogus i_mode (%o) for inode=%lu", |
4603 | inode->i_mode, inode->i_ino); | |
4604 | goto bad_inode; | |
ac27a0ec | 4605 | } |
af5bc92d | 4606 | brelse(iloc.bh); |
617ba13b | 4607 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4608 | unlock_new_inode(inode); |
4609 | return inode; | |
ac27a0ec DK |
4610 | |
4611 | bad_inode: | |
1d1fe1ee DH |
4612 | iget_failed(inode); |
4613 | return ERR_PTR(ret); | |
ac27a0ec DK |
4614 | } |
4615 | ||
0fc1b451 AK |
4616 | static int ext4_inode_blocks_set(handle_t *handle, |
4617 | struct ext4_inode *raw_inode, | |
4618 | struct ext4_inode_info *ei) | |
4619 | { | |
4620 | struct inode *inode = &(ei->vfs_inode); | |
4621 | u64 i_blocks = inode->i_blocks; | |
4622 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4623 | |
4624 | if (i_blocks <= ~0U) { | |
4625 | /* | |
4626 | * i_blocks can be represnted in a 32 bit variable | |
4627 | * as multiple of 512 bytes | |
4628 | */ | |
8180a562 | 4629 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4630 | raw_inode->i_blocks_high = 0; |
8180a562 | 4631 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
f287a1a5 TT |
4632 | return 0; |
4633 | } | |
4634 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4635 | return -EFBIG; | |
4636 | ||
4637 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4638 | /* |
4639 | * i_blocks can be represented in a 48 bit variable | |
4640 | * as multiple of 512 bytes | |
4641 | */ | |
8180a562 | 4642 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4643 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
8180a562 | 4644 | ei->i_flags &= ~EXT4_HUGE_FILE_FL; |
0fc1b451 | 4645 | } else { |
8180a562 AK |
4646 | ei->i_flags |= EXT4_HUGE_FILE_FL; |
4647 | /* i_block is stored in file system block size */ | |
4648 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4649 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4650 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4651 | } |
f287a1a5 | 4652 | return 0; |
0fc1b451 AK |
4653 | } |
4654 | ||
ac27a0ec DK |
4655 | /* |
4656 | * Post the struct inode info into an on-disk inode location in the | |
4657 | * buffer-cache. This gobbles the caller's reference to the | |
4658 | * buffer_head in the inode location struct. | |
4659 | * | |
4660 | * The caller must have write access to iloc->bh. | |
4661 | */ | |
617ba13b | 4662 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4663 | struct inode *inode, |
617ba13b | 4664 | struct ext4_iloc *iloc) |
ac27a0ec | 4665 | { |
617ba13b MC |
4666 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4667 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4668 | struct buffer_head *bh = iloc->bh; |
4669 | int err = 0, rc, block; | |
4670 | ||
4671 | /* For fields not not tracking in the in-memory inode, | |
4672 | * initialise them to zero for new inodes. */ | |
617ba13b MC |
4673 | if (ei->i_state & EXT4_STATE_NEW) |
4674 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
ac27a0ec | 4675 | |
ff9ddf7e | 4676 | ext4_get_inode_flags(ei); |
ac27a0ec | 4677 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 4678 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4679 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
4680 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
4681 | /* | |
4682 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4683 | * re-used with the upper 16 bits of the uid/gid intact | |
4684 | */ | |
af5bc92d | 4685 | if (!ei->i_dtime) { |
ac27a0ec DK |
4686 | raw_inode->i_uid_high = |
4687 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
4688 | raw_inode->i_gid_high = | |
4689 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
4690 | } else { | |
4691 | raw_inode->i_uid_high = 0; | |
4692 | raw_inode->i_gid_high = 0; | |
4693 | } | |
4694 | } else { | |
4695 | raw_inode->i_uid_low = | |
4696 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
4697 | raw_inode->i_gid_low = | |
4698 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
4699 | raw_inode->i_uid_high = 0; | |
4700 | raw_inode->i_gid_high = 0; | |
4701 | } | |
4702 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4703 | |
4704 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4705 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4706 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4707 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4708 | ||
0fc1b451 AK |
4709 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4710 | goto out_brelse; | |
ac27a0ec | 4711 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
267e4db9 AK |
4712 | /* clear the migrate flag in the raw_inode */ |
4713 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE); | |
9b8f1f01 MC |
4714 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4715 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4716 | raw_inode->i_file_acl_high = |
4717 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4718 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
4719 | ext4_isize_set(raw_inode, ei->i_disksize); |
4720 | if (ei->i_disksize > 0x7fffffffULL) { | |
4721 | struct super_block *sb = inode->i_sb; | |
4722 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4723 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4724 | EXT4_SB(sb)->s_es->s_rev_level == | |
4725 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4726 | /* If this is the first large file | |
4727 | * created, add a flag to the superblock. | |
4728 | */ | |
4729 | err = ext4_journal_get_write_access(handle, | |
4730 | EXT4_SB(sb)->s_sbh); | |
4731 | if (err) | |
4732 | goto out_brelse; | |
4733 | ext4_update_dynamic_rev(sb); | |
4734 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4735 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 4736 | sb->s_dirt = 1; |
0390131b FM |
4737 | ext4_handle_sync(handle); |
4738 | err = ext4_handle_dirty_metadata(handle, inode, | |
a48380f7 | 4739 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
4740 | } |
4741 | } | |
4742 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4743 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4744 | if (old_valid_dev(inode->i_rdev)) { | |
4745 | raw_inode->i_block[0] = | |
4746 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4747 | raw_inode->i_block[1] = 0; | |
4748 | } else { | |
4749 | raw_inode->i_block[0] = 0; | |
4750 | raw_inode->i_block[1] = | |
4751 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4752 | raw_inode->i_block[2] = 0; | |
4753 | } | |
de9a55b8 TT |
4754 | } else |
4755 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
4756 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 4757 | |
25ec56b5 JNC |
4758 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4759 | if (ei->i_extra_isize) { | |
4760 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4761 | raw_inode->i_version_hi = | |
4762 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4763 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4764 | } |
4765 | ||
0390131b FM |
4766 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
4767 | rc = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
4768 | if (!err) |
4769 | err = rc; | |
617ba13b | 4770 | ei->i_state &= ~EXT4_STATE_NEW; |
ac27a0ec DK |
4771 | |
4772 | out_brelse: | |
af5bc92d | 4773 | brelse(bh); |
617ba13b | 4774 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4775 | return err; |
4776 | } | |
4777 | ||
4778 | /* | |
617ba13b | 4779 | * ext4_write_inode() |
ac27a0ec DK |
4780 | * |
4781 | * We are called from a few places: | |
4782 | * | |
4783 | * - Within generic_file_write() for O_SYNC files. | |
4784 | * Here, there will be no transaction running. We wait for any running | |
4785 | * trasnaction to commit. | |
4786 | * | |
4787 | * - Within sys_sync(), kupdate and such. | |
4788 | * We wait on commit, if tol to. | |
4789 | * | |
4790 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4791 | * Here we simply return. We can't afford to block kswapd on the | |
4792 | * journal commit. | |
4793 | * | |
4794 | * In all cases it is actually safe for us to return without doing anything, | |
4795 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4796 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4797 | * knfsd. |
4798 | * | |
4799 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4800 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4801 | * which we are interested. | |
4802 | * | |
4803 | * It would be a bug for them to not do this. The code: | |
4804 | * | |
4805 | * mark_inode_dirty(inode) | |
4806 | * stuff(); | |
4807 | * inode->i_size = expr; | |
4808 | * | |
4809 | * is in error because a kswapd-driven write_inode() could occur while | |
4810 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4811 | * will no longer be on the superblock's dirty inode list. | |
4812 | */ | |
617ba13b | 4813 | int ext4_write_inode(struct inode *inode, int wait) |
ac27a0ec DK |
4814 | { |
4815 | if (current->flags & PF_MEMALLOC) | |
4816 | return 0; | |
4817 | ||
617ba13b | 4818 | if (ext4_journal_current_handle()) { |
b38bd33a | 4819 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
ac27a0ec DK |
4820 | dump_stack(); |
4821 | return -EIO; | |
4822 | } | |
4823 | ||
4824 | if (!wait) | |
4825 | return 0; | |
4826 | ||
617ba13b | 4827 | return ext4_force_commit(inode->i_sb); |
ac27a0ec DK |
4828 | } |
4829 | ||
4830 | /* | |
617ba13b | 4831 | * ext4_setattr() |
ac27a0ec DK |
4832 | * |
4833 | * Called from notify_change. | |
4834 | * | |
4835 | * We want to trap VFS attempts to truncate the file as soon as | |
4836 | * possible. In particular, we want to make sure that when the VFS | |
4837 | * shrinks i_size, we put the inode on the orphan list and modify | |
4838 | * i_disksize immediately, so that during the subsequent flushing of | |
4839 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4840 | * commit will leave the blocks being flushed in an unused state on | |
4841 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4842 | * be freed, so we have a strong guarantee that no future commit will | |
4843 | * leave these blocks visible to the user.) | |
4844 | * | |
678aaf48 JK |
4845 | * Another thing we have to assure is that if we are in ordered mode |
4846 | * and inode is still attached to the committing transaction, we must | |
4847 | * we start writeout of all the dirty pages which are being truncated. | |
4848 | * This way we are sure that all the data written in the previous | |
4849 | * transaction are already on disk (truncate waits for pages under | |
4850 | * writeback). | |
4851 | * | |
4852 | * Called with inode->i_mutex down. | |
ac27a0ec | 4853 | */ |
617ba13b | 4854 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4855 | { |
4856 | struct inode *inode = dentry->d_inode; | |
4857 | int error, rc = 0; | |
4858 | const unsigned int ia_valid = attr->ia_valid; | |
4859 | ||
4860 | error = inode_change_ok(inode, attr); | |
4861 | if (error) | |
4862 | return error; | |
4863 | ||
4864 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || | |
4865 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
4866 | handle_t *handle; | |
4867 | ||
4868 | /* (user+group)*(old+new) structure, inode write (sb, | |
4869 | * inode block, ? - but truncate inode update has it) */ | |
617ba13b MC |
4870 | handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+ |
4871 | EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3); | |
ac27a0ec DK |
4872 | if (IS_ERR(handle)) { |
4873 | error = PTR_ERR(handle); | |
4874 | goto err_out; | |
4875 | } | |
a269eb18 | 4876 | error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0; |
ac27a0ec | 4877 | if (error) { |
617ba13b | 4878 | ext4_journal_stop(handle); |
ac27a0ec DK |
4879 | return error; |
4880 | } | |
4881 | /* Update corresponding info in inode so that everything is in | |
4882 | * one transaction */ | |
4883 | if (attr->ia_valid & ATTR_UID) | |
4884 | inode->i_uid = attr->ia_uid; | |
4885 | if (attr->ia_valid & ATTR_GID) | |
4886 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4887 | error = ext4_mark_inode_dirty(handle, inode); |
4888 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4889 | } |
4890 | ||
e2b46574 ES |
4891 | if (attr->ia_valid & ATTR_SIZE) { |
4892 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) { | |
4893 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4894 | ||
4895 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { | |
4896 | error = -EFBIG; | |
4897 | goto err_out; | |
4898 | } | |
4899 | } | |
4900 | } | |
4901 | ||
ac27a0ec DK |
4902 | if (S_ISREG(inode->i_mode) && |
4903 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { | |
4904 | handle_t *handle; | |
4905 | ||
617ba13b | 4906 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4907 | if (IS_ERR(handle)) { |
4908 | error = PTR_ERR(handle); | |
4909 | goto err_out; | |
4910 | } | |
4911 | ||
617ba13b MC |
4912 | error = ext4_orphan_add(handle, inode); |
4913 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
4914 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4915 | if (!error) |
4916 | error = rc; | |
617ba13b | 4917 | ext4_journal_stop(handle); |
678aaf48 JK |
4918 | |
4919 | if (ext4_should_order_data(inode)) { | |
4920 | error = ext4_begin_ordered_truncate(inode, | |
4921 | attr->ia_size); | |
4922 | if (error) { | |
4923 | /* Do as much error cleanup as possible */ | |
4924 | handle = ext4_journal_start(inode, 3); | |
4925 | if (IS_ERR(handle)) { | |
4926 | ext4_orphan_del(NULL, inode); | |
4927 | goto err_out; | |
4928 | } | |
4929 | ext4_orphan_del(handle, inode); | |
4930 | ext4_journal_stop(handle); | |
4931 | goto err_out; | |
4932 | } | |
4933 | } | |
ac27a0ec DK |
4934 | } |
4935 | ||
4936 | rc = inode_setattr(inode, attr); | |
4937 | ||
617ba13b | 4938 | /* If inode_setattr's call to ext4_truncate failed to get a |
ac27a0ec DK |
4939 | * transaction handle at all, we need to clean up the in-core |
4940 | * orphan list manually. */ | |
4941 | if (inode->i_nlink) | |
617ba13b | 4942 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4943 | |
4944 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4945 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4946 | |
4947 | err_out: | |
617ba13b | 4948 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4949 | if (!error) |
4950 | error = rc; | |
4951 | return error; | |
4952 | } | |
4953 | ||
3e3398a0 MC |
4954 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4955 | struct kstat *stat) | |
4956 | { | |
4957 | struct inode *inode; | |
4958 | unsigned long delalloc_blocks; | |
4959 | ||
4960 | inode = dentry->d_inode; | |
4961 | generic_fillattr(inode, stat); | |
4962 | ||
4963 | /* | |
4964 | * We can't update i_blocks if the block allocation is delayed | |
4965 | * otherwise in the case of system crash before the real block | |
4966 | * allocation is done, we will have i_blocks inconsistent with | |
4967 | * on-disk file blocks. | |
4968 | * We always keep i_blocks updated together with real | |
4969 | * allocation. But to not confuse with user, stat | |
4970 | * will return the blocks that include the delayed allocation | |
4971 | * blocks for this file. | |
4972 | */ | |
4973 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); | |
4974 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
4975 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); | |
4976 | ||
4977 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4978 | return 0; | |
4979 | } | |
ac27a0ec | 4980 | |
a02908f1 MC |
4981 | static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks, |
4982 | int chunk) | |
4983 | { | |
4984 | int indirects; | |
4985 | ||
4986 | /* if nrblocks are contiguous */ | |
4987 | if (chunk) { | |
4988 | /* | |
4989 | * With N contiguous data blocks, it need at most | |
4990 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks | |
4991 | * 2 dindirect blocks | |
4992 | * 1 tindirect block | |
4993 | */ | |
4994 | indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb); | |
4995 | return indirects + 3; | |
4996 | } | |
4997 | /* | |
4998 | * if nrblocks are not contiguous, worse case, each block touch | |
4999 | * a indirect block, and each indirect block touch a double indirect | |
5000 | * block, plus a triple indirect block | |
5001 | */ | |
5002 | indirects = nrblocks * 2 + 1; | |
5003 | return indirects; | |
5004 | } | |
5005 | ||
5006 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5007 | { | |
5008 | if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) | |
ac51d837 TT |
5009 | return ext4_indirect_trans_blocks(inode, nrblocks, chunk); |
5010 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); | |
a02908f1 | 5011 | } |
ac51d837 | 5012 | |
ac27a0ec | 5013 | /* |
a02908f1 MC |
5014 | * Account for index blocks, block groups bitmaps and block group |
5015 | * descriptor blocks if modify datablocks and index blocks | |
5016 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5017 | * |
a02908f1 MC |
5018 | * If datablocks are discontiguous, they are possible to spread over |
5019 | * different block groups too. If they are contiugous, with flexbg, | |
5020 | * they could still across block group boundary. | |
ac27a0ec | 5021 | * |
a02908f1 MC |
5022 | * Also account for superblock, inode, quota and xattr blocks |
5023 | */ | |
5024 | int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5025 | { | |
8df9675f TT |
5026 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5027 | int gdpblocks; | |
a02908f1 MC |
5028 | int idxblocks; |
5029 | int ret = 0; | |
5030 | ||
5031 | /* | |
5032 | * How many index blocks need to touch to modify nrblocks? | |
5033 | * The "Chunk" flag indicating whether the nrblocks is | |
5034 | * physically contiguous on disk | |
5035 | * | |
5036 | * For Direct IO and fallocate, they calls get_block to allocate | |
5037 | * one single extent at a time, so they could set the "Chunk" flag | |
5038 | */ | |
5039 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
5040 | ||
5041 | ret = idxblocks; | |
5042 | ||
5043 | /* | |
5044 | * Now let's see how many group bitmaps and group descriptors need | |
5045 | * to account | |
5046 | */ | |
5047 | groups = idxblocks; | |
5048 | if (chunk) | |
5049 | groups += 1; | |
5050 | else | |
5051 | groups += nrblocks; | |
5052 | ||
5053 | gdpblocks = groups; | |
8df9675f TT |
5054 | if (groups > ngroups) |
5055 | groups = ngroups; | |
a02908f1 MC |
5056 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5057 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5058 | ||
5059 | /* bitmaps and block group descriptor blocks */ | |
5060 | ret += groups + gdpblocks; | |
5061 | ||
5062 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5063 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5064 | ||
5065 | return ret; | |
5066 | } | |
5067 | ||
5068 | /* | |
5069 | * Calulate the total number of credits to reserve to fit | |
f3bd1f3f MC |
5070 | * the modification of a single pages into a single transaction, |
5071 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5072 | * |
525f4ed8 | 5073 | * This could be called via ext4_write_begin() |
ac27a0ec | 5074 | * |
525f4ed8 | 5075 | * We need to consider the worse case, when |
a02908f1 | 5076 | * one new block per extent. |
ac27a0ec | 5077 | */ |
a86c6181 | 5078 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5079 | { |
617ba13b | 5080 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5081 | int ret; |
5082 | ||
a02908f1 | 5083 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 5084 | |
a02908f1 | 5085 | /* Account for data blocks for journalled mode */ |
617ba13b | 5086 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5087 | ret += bpp; |
ac27a0ec DK |
5088 | return ret; |
5089 | } | |
f3bd1f3f MC |
5090 | |
5091 | /* | |
5092 | * Calculate the journal credits for a chunk of data modification. | |
5093 | * | |
5094 | * This is called from DIO, fallocate or whoever calling | |
12b7ac17 | 5095 | * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks. |
f3bd1f3f MC |
5096 | * |
5097 | * journal buffers for data blocks are not included here, as DIO | |
5098 | * and fallocate do no need to journal data buffers. | |
5099 | */ | |
5100 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5101 | { | |
5102 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5103 | } | |
5104 | ||
ac27a0ec | 5105 | /* |
617ba13b | 5106 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5107 | * Give this, we know that the caller already has write access to iloc->bh. |
5108 | */ | |
617ba13b | 5109 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5110 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5111 | { |
5112 | int err = 0; | |
5113 | ||
25ec56b5 JNC |
5114 | if (test_opt(inode->i_sb, I_VERSION)) |
5115 | inode_inc_iversion(inode); | |
5116 | ||
ac27a0ec DK |
5117 | /* the do_update_inode consumes one bh->b_count */ |
5118 | get_bh(iloc->bh); | |
5119 | ||
dab291af | 5120 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
617ba13b | 5121 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5122 | put_bh(iloc->bh); |
5123 | return err; | |
5124 | } | |
5125 | ||
5126 | /* | |
5127 | * On success, We end up with an outstanding reference count against | |
5128 | * iloc->bh. This _must_ be cleaned up later. | |
5129 | */ | |
5130 | ||
5131 | int | |
617ba13b MC |
5132 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5133 | struct ext4_iloc *iloc) | |
ac27a0ec | 5134 | { |
0390131b FM |
5135 | int err; |
5136 | ||
5137 | err = ext4_get_inode_loc(inode, iloc); | |
5138 | if (!err) { | |
5139 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5140 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5141 | if (err) { | |
5142 | brelse(iloc->bh); | |
5143 | iloc->bh = NULL; | |
ac27a0ec DK |
5144 | } |
5145 | } | |
617ba13b | 5146 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5147 | return err; |
5148 | } | |
5149 | ||
6dd4ee7c KS |
5150 | /* |
5151 | * Expand an inode by new_extra_isize bytes. | |
5152 | * Returns 0 on success or negative error number on failure. | |
5153 | */ | |
1d03ec98 AK |
5154 | static int ext4_expand_extra_isize(struct inode *inode, |
5155 | unsigned int new_extra_isize, | |
5156 | struct ext4_iloc iloc, | |
5157 | handle_t *handle) | |
6dd4ee7c KS |
5158 | { |
5159 | struct ext4_inode *raw_inode; | |
5160 | struct ext4_xattr_ibody_header *header; | |
5161 | struct ext4_xattr_entry *entry; | |
5162 | ||
5163 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5164 | return 0; | |
5165 | ||
5166 | raw_inode = ext4_raw_inode(&iloc); | |
5167 | ||
5168 | header = IHDR(inode, raw_inode); | |
5169 | entry = IFIRST(header); | |
5170 | ||
5171 | /* No extended attributes present */ | |
5172 | if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) || | |
5173 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5174 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, | |
5175 | new_extra_isize); | |
5176 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5177 | return 0; | |
5178 | } | |
5179 | ||
5180 | /* try to expand with EAs present */ | |
5181 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5182 | raw_inode, handle); | |
5183 | } | |
5184 | ||
ac27a0ec DK |
5185 | /* |
5186 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5187 | * dirtiness (it may still be data-dirty). | |
5188 | * This means that the in-core inode may be reaped by prune_icache | |
5189 | * without having to perform any I/O. This is a very good thing, | |
5190 | * because *any* task may call prune_icache - even ones which | |
5191 | * have a transaction open against a different journal. | |
5192 | * | |
5193 | * Is this cheating? Not really. Sure, we haven't written the | |
5194 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5195 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5196 | * we start and wait on commits. | |
5197 | * | |
5198 | * Is this efficient/effective? Well, we're being nice to the system | |
5199 | * by cleaning up our inodes proactively so they can be reaped | |
5200 | * without I/O. But we are potentially leaving up to five seconds' | |
5201 | * worth of inodes floating about which prune_icache wants us to | |
5202 | * write out. One way to fix that would be to get prune_icache() | |
5203 | * to do a write_super() to free up some memory. It has the desired | |
5204 | * effect. | |
5205 | */ | |
617ba13b | 5206 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5207 | { |
617ba13b | 5208 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5209 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5210 | static unsigned int mnt_count; | |
5211 | int err, ret; | |
ac27a0ec DK |
5212 | |
5213 | might_sleep(); | |
617ba13b | 5214 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5215 | if (ext4_handle_valid(handle) && |
5216 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
6dd4ee7c KS |
5217 | !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { |
5218 | /* | |
5219 | * We need extra buffer credits since we may write into EA block | |
5220 | * with this same handle. If journal_extend fails, then it will | |
5221 | * only result in a minor loss of functionality for that inode. | |
5222 | * If this is felt to be critical, then e2fsck should be run to | |
5223 | * force a large enough s_min_extra_isize. | |
5224 | */ | |
5225 | if ((jbd2_journal_extend(handle, | |
5226 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5227 | ret = ext4_expand_extra_isize(inode, | |
5228 | sbi->s_want_extra_isize, | |
5229 | iloc, handle); | |
5230 | if (ret) { | |
5231 | EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND; | |
c1bddad9 AK |
5232 | if (mnt_count != |
5233 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
46e665e9 | 5234 | ext4_warning(inode->i_sb, __func__, |
6dd4ee7c KS |
5235 | "Unable to expand inode %lu. Delete" |
5236 | " some EAs or run e2fsck.", | |
5237 | inode->i_ino); | |
c1bddad9 AK |
5238 | mnt_count = |
5239 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5240 | } |
5241 | } | |
5242 | } | |
5243 | } | |
ac27a0ec | 5244 | if (!err) |
617ba13b | 5245 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5246 | return err; |
5247 | } | |
5248 | ||
5249 | /* | |
617ba13b | 5250 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5251 | * |
5252 | * We're really interested in the case where a file is being extended. | |
5253 | * i_size has been changed by generic_commit_write() and we thus need | |
5254 | * to include the updated inode in the current transaction. | |
5255 | * | |
a269eb18 | 5256 | * Also, vfs_dq_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5257 | * are allocated to the file. |
5258 | * | |
5259 | * If the inode is marked synchronous, we don't honour that here - doing | |
5260 | * so would cause a commit on atime updates, which we don't bother doing. | |
5261 | * We handle synchronous inodes at the highest possible level. | |
5262 | */ | |
617ba13b | 5263 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 5264 | { |
617ba13b | 5265 | handle_t *current_handle = ext4_journal_current_handle(); |
ac27a0ec DK |
5266 | handle_t *handle; |
5267 | ||
0390131b FM |
5268 | if (!ext4_handle_valid(current_handle)) { |
5269 | ext4_mark_inode_dirty(current_handle, inode); | |
5270 | return; | |
5271 | } | |
5272 | ||
617ba13b | 5273 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
5274 | if (IS_ERR(handle)) |
5275 | goto out; | |
5276 | if (current_handle && | |
5277 | current_handle->h_transaction != handle->h_transaction) { | |
5278 | /* This task has a transaction open against a different fs */ | |
5279 | printk(KERN_EMERG "%s: transactions do not match!\n", | |
46e665e9 | 5280 | __func__); |
ac27a0ec DK |
5281 | } else { |
5282 | jbd_debug(5, "marking dirty. outer handle=%p\n", | |
5283 | current_handle); | |
617ba13b | 5284 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 5285 | } |
617ba13b | 5286 | ext4_journal_stop(handle); |
ac27a0ec DK |
5287 | out: |
5288 | return; | |
5289 | } | |
5290 | ||
5291 | #if 0 | |
5292 | /* | |
5293 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5294 | * it from being flushed to disk early. Unlike | |
617ba13b | 5295 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5296 | * returns no iloc structure, so the caller needs to repeat the iloc |
5297 | * lookup to mark the inode dirty later. | |
5298 | */ | |
617ba13b | 5299 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5300 | { |
617ba13b | 5301 | struct ext4_iloc iloc; |
ac27a0ec DK |
5302 | |
5303 | int err = 0; | |
5304 | if (handle) { | |
617ba13b | 5305 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5306 | if (!err) { |
5307 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5308 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5309 | if (!err) |
0390131b FM |
5310 | err = ext4_handle_dirty_metadata(handle, |
5311 | inode, | |
5312 | iloc.bh); | |
ac27a0ec DK |
5313 | brelse(iloc.bh); |
5314 | } | |
5315 | } | |
617ba13b | 5316 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5317 | return err; |
5318 | } | |
5319 | #endif | |
5320 | ||
617ba13b | 5321 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5322 | { |
5323 | journal_t *journal; | |
5324 | handle_t *handle; | |
5325 | int err; | |
5326 | ||
5327 | /* | |
5328 | * We have to be very careful here: changing a data block's | |
5329 | * journaling status dynamically is dangerous. If we write a | |
5330 | * data block to the journal, change the status and then delete | |
5331 | * that block, we risk forgetting to revoke the old log record | |
5332 | * from the journal and so a subsequent replay can corrupt data. | |
5333 | * So, first we make sure that the journal is empty and that | |
5334 | * nobody is changing anything. | |
5335 | */ | |
5336 | ||
617ba13b | 5337 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5338 | if (!journal) |
5339 | return 0; | |
d699594d | 5340 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5341 | return -EROFS; |
5342 | ||
dab291af MC |
5343 | jbd2_journal_lock_updates(journal); |
5344 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
5345 | |
5346 | /* | |
5347 | * OK, there are no updates running now, and all cached data is | |
5348 | * synced to disk. We are now in a completely consistent state | |
5349 | * which doesn't have anything in the journal, and we know that | |
5350 | * no filesystem updates are running, so it is safe to modify | |
5351 | * the inode's in-core data-journaling state flag now. | |
5352 | */ | |
5353 | ||
5354 | if (val) | |
617ba13b | 5355 | EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL; |
ac27a0ec | 5356 | else |
617ba13b MC |
5357 | EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL; |
5358 | ext4_set_aops(inode); | |
ac27a0ec | 5359 | |
dab291af | 5360 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
5361 | |
5362 | /* Finally we can mark the inode as dirty. */ | |
5363 | ||
617ba13b | 5364 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
5365 | if (IS_ERR(handle)) |
5366 | return PTR_ERR(handle); | |
5367 | ||
617ba13b | 5368 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5369 | ext4_handle_sync(handle); |
617ba13b MC |
5370 | ext4_journal_stop(handle); |
5371 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5372 | |
5373 | return err; | |
5374 | } | |
2e9ee850 AK |
5375 | |
5376 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5377 | { | |
5378 | return !buffer_mapped(bh); | |
5379 | } | |
5380 | ||
c2ec175c | 5381 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5382 | { |
c2ec175c | 5383 | struct page *page = vmf->page; |
2e9ee850 AK |
5384 | loff_t size; |
5385 | unsigned long len; | |
5386 | int ret = -EINVAL; | |
79f0be8d | 5387 | void *fsdata; |
2e9ee850 AK |
5388 | struct file *file = vma->vm_file; |
5389 | struct inode *inode = file->f_path.dentry->d_inode; | |
5390 | struct address_space *mapping = inode->i_mapping; | |
5391 | ||
5392 | /* | |
5393 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
5394 | * get i_mutex because we are already holding mmap_sem. | |
5395 | */ | |
5396 | down_read(&inode->i_alloc_sem); | |
5397 | size = i_size_read(inode); | |
5398 | if (page->mapping != mapping || size <= page_offset(page) | |
5399 | || !PageUptodate(page)) { | |
5400 | /* page got truncated from under us? */ | |
5401 | goto out_unlock; | |
5402 | } | |
5403 | ret = 0; | |
5404 | if (PageMappedToDisk(page)) | |
5405 | goto out_unlock; | |
5406 | ||
5407 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5408 | len = size & ~PAGE_CACHE_MASK; | |
5409 | else | |
5410 | len = PAGE_CACHE_SIZE; | |
5411 | ||
5412 | if (page_has_buffers(page)) { | |
5413 | /* return if we have all the buffers mapped */ | |
5414 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, | |
5415 | ext4_bh_unmapped)) | |
5416 | goto out_unlock; | |
5417 | } | |
5418 | /* | |
5419 | * OK, we need to fill the hole... Do write_begin write_end | |
5420 | * to do block allocation/reservation.We are not holding | |
5421 | * inode.i__mutex here. That allow * parallel write_begin, | |
5422 | * write_end call. lock_page prevent this from happening | |
5423 | * on the same page though | |
5424 | */ | |
5425 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
79f0be8d | 5426 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); |
2e9ee850 AK |
5427 | if (ret < 0) |
5428 | goto out_unlock; | |
5429 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
79f0be8d | 5430 | len, len, page, fsdata); |
2e9ee850 AK |
5431 | if (ret < 0) |
5432 | goto out_unlock; | |
5433 | ret = 0; | |
5434 | out_unlock: | |
c2ec175c NP |
5435 | if (ret) |
5436 | ret = VM_FAULT_SIGBUS; | |
2e9ee850 AK |
5437 | up_read(&inode->i_alloc_sem); |
5438 | return ret; | |
5439 | } |