]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
15 | * Goal-directed block allocation by Stephen Tweedie | |
16 | * (sct@redhat.com), 1993, 1998 | |
17 | * Big-endian to little-endian byte-swapping/bitmaps by | |
18 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
20 | * (jj@sunsite.ms.mff.cuni.cz) | |
21 | * | |
617ba13b | 22 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
23 | */ |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/time.h> | |
dab291af | 28 | #include <linux/jbd2.h> |
ac27a0ec DK |
29 | #include <linux/highuid.h> |
30 | #include <linux/pagemap.h> | |
31 | #include <linux/quotaops.h> | |
32 | #include <linux/string.h> | |
33 | #include <linux/buffer_head.h> | |
34 | #include <linux/writeback.h> | |
64769240 | 35 | #include <linux/pagevec.h> |
ac27a0ec | 36 | #include <linux/mpage.h> |
e83c1397 | 37 | #include <linux/namei.h> |
ac27a0ec DK |
38 | #include <linux/uio.h> |
39 | #include <linux/bio.h> | |
4c0425ff | 40 | #include <linux/workqueue.h> |
744692dc | 41 | #include <linux/kernel.h> |
6db26ffc | 42 | #include <linux/printk.h> |
5a0e3ad6 | 43 | #include <linux/slab.h> |
a8901d34 | 44 | #include <linux/ratelimit.h> |
9bffad1e | 45 | |
3dcf5451 | 46 | #include "ext4_jbd2.h" |
ac27a0ec DK |
47 | #include "xattr.h" |
48 | #include "acl.h" | |
d2a17637 | 49 | #include "ext4_extents.h" |
ac27a0ec | 50 | |
9bffad1e TT |
51 | #include <trace/events/ext4.h> |
52 | ||
a1d6cc56 AK |
53 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
54 | ||
678aaf48 JK |
55 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
56 | loff_t new_size) | |
57 | { | |
7ff9c073 | 58 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
59 | /* |
60 | * If jinode is zero, then we never opened the file for | |
61 | * writing, so there's no need to call | |
62 | * jbd2_journal_begin_ordered_truncate() since there's no | |
63 | * outstanding writes we need to flush. | |
64 | */ | |
65 | if (!EXT4_I(inode)->jinode) | |
66 | return 0; | |
67 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
68 | EXT4_I(inode)->jinode, | |
69 | new_size); | |
678aaf48 JK |
70 | } |
71 | ||
64769240 | 72 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
73 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, |
74 | struct buffer_head *bh_result, int create); | |
75 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); | |
76 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
77 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); | |
78 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
64769240 | 79 | |
ac27a0ec DK |
80 | /* |
81 | * Test whether an inode is a fast symlink. | |
82 | */ | |
617ba13b | 83 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 84 | { |
617ba13b | 85 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
86 | (inode->i_sb->s_blocksize >> 9) : 0; |
87 | ||
88 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
89 | } | |
90 | ||
ac27a0ec DK |
91 | /* |
92 | * Work out how many blocks we need to proceed with the next chunk of a | |
93 | * truncate transaction. | |
94 | */ | |
95 | static unsigned long blocks_for_truncate(struct inode *inode) | |
96 | { | |
725d26d3 | 97 | ext4_lblk_t needed; |
ac27a0ec DK |
98 | |
99 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); | |
100 | ||
101 | /* Give ourselves just enough room to cope with inodes in which | |
102 | * i_blocks is corrupt: we've seen disk corruptions in the past | |
103 | * which resulted in random data in an inode which looked enough | |
617ba13b | 104 | * like a regular file for ext4 to try to delete it. Things |
ac27a0ec DK |
105 | * will go a bit crazy if that happens, but at least we should |
106 | * try not to panic the whole kernel. */ | |
107 | if (needed < 2) | |
108 | needed = 2; | |
109 | ||
110 | /* But we need to bound the transaction so we don't overflow the | |
111 | * journal. */ | |
617ba13b MC |
112 | if (needed > EXT4_MAX_TRANS_DATA) |
113 | needed = EXT4_MAX_TRANS_DATA; | |
ac27a0ec | 114 | |
617ba13b | 115 | return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
ac27a0ec DK |
116 | } |
117 | ||
118 | /* | |
119 | * Truncate transactions can be complex and absolutely huge. So we need to | |
120 | * be able to restart the transaction at a conventient checkpoint to make | |
121 | * sure we don't overflow the journal. | |
122 | * | |
123 | * start_transaction gets us a new handle for a truncate transaction, | |
124 | * and extend_transaction tries to extend the existing one a bit. If | |
125 | * extend fails, we need to propagate the failure up and restart the | |
126 | * transaction in the top-level truncate loop. --sct | |
127 | */ | |
128 | static handle_t *start_transaction(struct inode *inode) | |
129 | { | |
130 | handle_t *result; | |
131 | ||
617ba13b | 132 | result = ext4_journal_start(inode, blocks_for_truncate(inode)); |
ac27a0ec DK |
133 | if (!IS_ERR(result)) |
134 | return result; | |
135 | ||
617ba13b | 136 | ext4_std_error(inode->i_sb, PTR_ERR(result)); |
ac27a0ec DK |
137 | return result; |
138 | } | |
139 | ||
140 | /* | |
141 | * Try to extend this transaction for the purposes of truncation. | |
142 | * | |
143 | * Returns 0 if we managed to create more room. If we can't create more | |
144 | * room, and the transaction must be restarted we return 1. | |
145 | */ | |
146 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) | |
147 | { | |
0390131b FM |
148 | if (!ext4_handle_valid(handle)) |
149 | return 0; | |
150 | if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1)) | |
ac27a0ec | 151 | return 0; |
617ba13b | 152 | if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) |
ac27a0ec DK |
153 | return 0; |
154 | return 1; | |
155 | } | |
156 | ||
157 | /* | |
158 | * Restart the transaction associated with *handle. This does a commit, | |
159 | * so before we call here everything must be consistently dirtied against | |
160 | * this transaction. | |
161 | */ | |
fa5d1113 | 162 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 163 | int nblocks) |
ac27a0ec | 164 | { |
487caeef JK |
165 | int ret; |
166 | ||
167 | /* | |
e35fd660 | 168 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
169 | * moment, get_block can be called only for blocks inside i_size since |
170 | * page cache has been already dropped and writes are blocked by | |
171 | * i_mutex. So we can safely drop the i_data_sem here. | |
172 | */ | |
0390131b | 173 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 174 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 175 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 176 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 177 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 178 | ext4_discard_preallocations(inode); |
487caeef JK |
179 | |
180 | return ret; | |
ac27a0ec DK |
181 | } |
182 | ||
183 | /* | |
184 | * Called at the last iput() if i_nlink is zero. | |
185 | */ | |
0930fcc1 | 186 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
187 | { |
188 | handle_t *handle; | |
bc965ab3 | 189 | int err; |
ac27a0ec | 190 | |
7ff9c073 | 191 | trace_ext4_evict_inode(inode); |
0930fcc1 AV |
192 | if (inode->i_nlink) { |
193 | truncate_inode_pages(&inode->i_data, 0); | |
194 | goto no_delete; | |
195 | } | |
196 | ||
907f4554 | 197 | if (!is_bad_inode(inode)) |
871a2931 | 198 | dquot_initialize(inode); |
907f4554 | 199 | |
678aaf48 JK |
200 | if (ext4_should_order_data(inode)) |
201 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
202 | truncate_inode_pages(&inode->i_data, 0); |
203 | ||
204 | if (is_bad_inode(inode)) | |
205 | goto no_delete; | |
206 | ||
bc965ab3 | 207 | handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3); |
ac27a0ec | 208 | if (IS_ERR(handle)) { |
bc965ab3 | 209 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
210 | /* |
211 | * If we're going to skip the normal cleanup, we still need to | |
212 | * make sure that the in-core orphan linked list is properly | |
213 | * cleaned up. | |
214 | */ | |
617ba13b | 215 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
216 | goto no_delete; |
217 | } | |
218 | ||
219 | if (IS_SYNC(inode)) | |
0390131b | 220 | ext4_handle_sync(handle); |
ac27a0ec | 221 | inode->i_size = 0; |
bc965ab3 TT |
222 | err = ext4_mark_inode_dirty(handle, inode); |
223 | if (err) { | |
12062ddd | 224 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
225 | "couldn't mark inode dirty (err %d)", err); |
226 | goto stop_handle; | |
227 | } | |
ac27a0ec | 228 | if (inode->i_blocks) |
617ba13b | 229 | ext4_truncate(inode); |
bc965ab3 TT |
230 | |
231 | /* | |
232 | * ext4_ext_truncate() doesn't reserve any slop when it | |
233 | * restarts journal transactions; therefore there may not be | |
234 | * enough credits left in the handle to remove the inode from | |
235 | * the orphan list and set the dtime field. | |
236 | */ | |
0390131b | 237 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
238 | err = ext4_journal_extend(handle, 3); |
239 | if (err > 0) | |
240 | err = ext4_journal_restart(handle, 3); | |
241 | if (err != 0) { | |
12062ddd | 242 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
243 | "couldn't extend journal (err %d)", err); |
244 | stop_handle: | |
245 | ext4_journal_stop(handle); | |
45388219 | 246 | ext4_orphan_del(NULL, inode); |
bc965ab3 TT |
247 | goto no_delete; |
248 | } | |
249 | } | |
250 | ||
ac27a0ec | 251 | /* |
617ba13b | 252 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 253 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 254 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 255 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 256 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
257 | * (Well, we could do this if we need to, but heck - it works) |
258 | */ | |
617ba13b MC |
259 | ext4_orphan_del(handle, inode); |
260 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
261 | |
262 | /* | |
263 | * One subtle ordering requirement: if anything has gone wrong | |
264 | * (transaction abort, IO errors, whatever), then we can still | |
265 | * do these next steps (the fs will already have been marked as | |
266 | * having errors), but we can't free the inode if the mark_dirty | |
267 | * fails. | |
268 | */ | |
617ba13b | 269 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 270 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 271 | ext4_clear_inode(inode); |
ac27a0ec | 272 | else |
617ba13b MC |
273 | ext4_free_inode(handle, inode); |
274 | ext4_journal_stop(handle); | |
ac27a0ec DK |
275 | return; |
276 | no_delete: | |
0930fcc1 | 277 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
278 | } |
279 | ||
280 | typedef struct { | |
281 | __le32 *p; | |
282 | __le32 key; | |
283 | struct buffer_head *bh; | |
284 | } Indirect; | |
285 | ||
286 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) | |
287 | { | |
288 | p->key = *(p->p = v); | |
289 | p->bh = bh; | |
290 | } | |
291 | ||
ac27a0ec | 292 | /** |
617ba13b | 293 | * ext4_block_to_path - parse the block number into array of offsets |
ac27a0ec DK |
294 | * @inode: inode in question (we are only interested in its superblock) |
295 | * @i_block: block number to be parsed | |
296 | * @offsets: array to store the offsets in | |
8c55e204 DK |
297 | * @boundary: set this non-zero if the referred-to block is likely to be |
298 | * followed (on disk) by an indirect block. | |
ac27a0ec | 299 | * |
617ba13b | 300 | * To store the locations of file's data ext4 uses a data structure common |
ac27a0ec DK |
301 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
302 | * data blocks at leaves and indirect blocks in intermediate nodes. | |
303 | * This function translates the block number into path in that tree - | |
304 | * return value is the path length and @offsets[n] is the offset of | |
305 | * pointer to (n+1)th node in the nth one. If @block is out of range | |
306 | * (negative or too large) warning is printed and zero returned. | |
307 | * | |
308 | * Note: function doesn't find node addresses, so no IO is needed. All | |
309 | * we need to know is the capacity of indirect blocks (taken from the | |
310 | * inode->i_sb). | |
311 | */ | |
312 | ||
313 | /* | |
314 | * Portability note: the last comparison (check that we fit into triple | |
315 | * indirect block) is spelled differently, because otherwise on an | |
316 | * architecture with 32-bit longs and 8Kb pages we might get into trouble | |
317 | * if our filesystem had 8Kb blocks. We might use long long, but that would | |
318 | * kill us on x86. Oh, well, at least the sign propagation does not matter - | |
319 | * i_block would have to be negative in the very beginning, so we would not | |
320 | * get there at all. | |
321 | */ | |
322 | ||
617ba13b | 323 | static int ext4_block_to_path(struct inode *inode, |
de9a55b8 TT |
324 | ext4_lblk_t i_block, |
325 | ext4_lblk_t offsets[4], int *boundary) | |
ac27a0ec | 326 | { |
617ba13b MC |
327 | int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
328 | int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb); | |
329 | const long direct_blocks = EXT4_NDIR_BLOCKS, | |
ac27a0ec DK |
330 | indirect_blocks = ptrs, |
331 | double_blocks = (1 << (ptrs_bits * 2)); | |
332 | int n = 0; | |
333 | int final = 0; | |
334 | ||
c333e073 | 335 | if (i_block < direct_blocks) { |
ac27a0ec DK |
336 | offsets[n++] = i_block; |
337 | final = direct_blocks; | |
af5bc92d | 338 | } else if ((i_block -= direct_blocks) < indirect_blocks) { |
617ba13b | 339 | offsets[n++] = EXT4_IND_BLOCK; |
ac27a0ec DK |
340 | offsets[n++] = i_block; |
341 | final = ptrs; | |
342 | } else if ((i_block -= indirect_blocks) < double_blocks) { | |
617ba13b | 343 | offsets[n++] = EXT4_DIND_BLOCK; |
ac27a0ec DK |
344 | offsets[n++] = i_block >> ptrs_bits; |
345 | offsets[n++] = i_block & (ptrs - 1); | |
346 | final = ptrs; | |
347 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { | |
617ba13b | 348 | offsets[n++] = EXT4_TIND_BLOCK; |
ac27a0ec DK |
349 | offsets[n++] = i_block >> (ptrs_bits * 2); |
350 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); | |
351 | offsets[n++] = i_block & (ptrs - 1); | |
352 | final = ptrs; | |
353 | } else { | |
12062ddd | 354 | ext4_warning(inode->i_sb, "block %lu > max in inode %lu", |
de9a55b8 TT |
355 | i_block + direct_blocks + |
356 | indirect_blocks + double_blocks, inode->i_ino); | |
ac27a0ec DK |
357 | } |
358 | if (boundary) | |
359 | *boundary = final - 1 - (i_block & (ptrs - 1)); | |
360 | return n; | |
361 | } | |
362 | ||
c398eda0 TT |
363 | static int __ext4_check_blockref(const char *function, unsigned int line, |
364 | struct inode *inode, | |
6fd058f7 TT |
365 | __le32 *p, unsigned int max) |
366 | { | |
1c13d5c0 | 367 | struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; |
f73953c0 | 368 | __le32 *bref = p; |
6fd058f7 TT |
369 | unsigned int blk; |
370 | ||
fe2c8191 | 371 | while (bref < p+max) { |
6fd058f7 | 372 | blk = le32_to_cpu(*bref++); |
de9a55b8 TT |
373 | if (blk && |
374 | unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), | |
6fd058f7 | 375 | blk, 1))) { |
1c13d5c0 | 376 | es->s_last_error_block = cpu_to_le64(blk); |
c398eda0 TT |
377 | ext4_error_inode(inode, function, line, blk, |
378 | "invalid block"); | |
de9a55b8 TT |
379 | return -EIO; |
380 | } | |
381 | } | |
382 | return 0; | |
fe2c8191 TN |
383 | } |
384 | ||
385 | ||
386 | #define ext4_check_indirect_blockref(inode, bh) \ | |
c398eda0 TT |
387 | __ext4_check_blockref(__func__, __LINE__, inode, \ |
388 | (__le32 *)(bh)->b_data, \ | |
fe2c8191 TN |
389 | EXT4_ADDR_PER_BLOCK((inode)->i_sb)) |
390 | ||
391 | #define ext4_check_inode_blockref(inode) \ | |
c398eda0 TT |
392 | __ext4_check_blockref(__func__, __LINE__, inode, \ |
393 | EXT4_I(inode)->i_data, \ | |
fe2c8191 TN |
394 | EXT4_NDIR_BLOCKS) |
395 | ||
ac27a0ec | 396 | /** |
617ba13b | 397 | * ext4_get_branch - read the chain of indirect blocks leading to data |
ac27a0ec DK |
398 | * @inode: inode in question |
399 | * @depth: depth of the chain (1 - direct pointer, etc.) | |
400 | * @offsets: offsets of pointers in inode/indirect blocks | |
401 | * @chain: place to store the result | |
402 | * @err: here we store the error value | |
403 | * | |
404 | * Function fills the array of triples <key, p, bh> and returns %NULL | |
405 | * if everything went OK or the pointer to the last filled triple | |
406 | * (incomplete one) otherwise. Upon the return chain[i].key contains | |
407 | * the number of (i+1)-th block in the chain (as it is stored in memory, | |
408 | * i.e. little-endian 32-bit), chain[i].p contains the address of that | |
409 | * number (it points into struct inode for i==0 and into the bh->b_data | |
410 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect | |
411 | * block for i>0 and NULL for i==0. In other words, it holds the block | |
412 | * numbers of the chain, addresses they were taken from (and where we can | |
413 | * verify that chain did not change) and buffer_heads hosting these | |
414 | * numbers. | |
415 | * | |
416 | * Function stops when it stumbles upon zero pointer (absent block) | |
417 | * (pointer to last triple returned, *@err == 0) | |
418 | * or when it gets an IO error reading an indirect block | |
419 | * (ditto, *@err == -EIO) | |
ac27a0ec DK |
420 | * or when it reads all @depth-1 indirect blocks successfully and finds |
421 | * the whole chain, all way to the data (returns %NULL, *err == 0). | |
c278bfec AK |
422 | * |
423 | * Need to be called with | |
0e855ac8 | 424 | * down_read(&EXT4_I(inode)->i_data_sem) |
ac27a0ec | 425 | */ |
725d26d3 AK |
426 | static Indirect *ext4_get_branch(struct inode *inode, int depth, |
427 | ext4_lblk_t *offsets, | |
ac27a0ec DK |
428 | Indirect chain[4], int *err) |
429 | { | |
430 | struct super_block *sb = inode->i_sb; | |
431 | Indirect *p = chain; | |
432 | struct buffer_head *bh; | |
433 | ||
434 | *err = 0; | |
435 | /* i_data is not going away, no lock needed */ | |
af5bc92d | 436 | add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets); |
ac27a0ec DK |
437 | if (!p->key) |
438 | goto no_block; | |
439 | while (--depth) { | |
fe2c8191 TN |
440 | bh = sb_getblk(sb, le32_to_cpu(p->key)); |
441 | if (unlikely(!bh)) | |
ac27a0ec | 442 | goto failure; |
de9a55b8 | 443 | |
fe2c8191 TN |
444 | if (!bh_uptodate_or_lock(bh)) { |
445 | if (bh_submit_read(bh) < 0) { | |
446 | put_bh(bh); | |
447 | goto failure; | |
448 | } | |
449 | /* validate block references */ | |
450 | if (ext4_check_indirect_blockref(inode, bh)) { | |
451 | put_bh(bh); | |
452 | goto failure; | |
453 | } | |
454 | } | |
de9a55b8 | 455 | |
af5bc92d | 456 | add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets); |
ac27a0ec DK |
457 | /* Reader: end */ |
458 | if (!p->key) | |
459 | goto no_block; | |
460 | } | |
461 | return NULL; | |
462 | ||
ac27a0ec DK |
463 | failure: |
464 | *err = -EIO; | |
465 | no_block: | |
466 | return p; | |
467 | } | |
468 | ||
469 | /** | |
617ba13b | 470 | * ext4_find_near - find a place for allocation with sufficient locality |
ac27a0ec DK |
471 | * @inode: owner |
472 | * @ind: descriptor of indirect block. | |
473 | * | |
1cc8dcf5 | 474 | * This function returns the preferred place for block allocation. |
ac27a0ec DK |
475 | * It is used when heuristic for sequential allocation fails. |
476 | * Rules are: | |
477 | * + if there is a block to the left of our position - allocate near it. | |
478 | * + if pointer will live in indirect block - allocate near that block. | |
479 | * + if pointer will live in inode - allocate in the same | |
480 | * cylinder group. | |
481 | * | |
482 | * In the latter case we colour the starting block by the callers PID to | |
483 | * prevent it from clashing with concurrent allocations for a different inode | |
484 | * in the same block group. The PID is used here so that functionally related | |
485 | * files will be close-by on-disk. | |
486 | * | |
487 | * Caller must make sure that @ind is valid and will stay that way. | |
488 | */ | |
617ba13b | 489 | static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind) |
ac27a0ec | 490 | { |
617ba13b | 491 | struct ext4_inode_info *ei = EXT4_I(inode); |
af5bc92d | 492 | __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data; |
ac27a0ec | 493 | __le32 *p; |
617ba13b | 494 | ext4_fsblk_t bg_start; |
74d3487f | 495 | ext4_fsblk_t last_block; |
617ba13b | 496 | ext4_grpblk_t colour; |
a4912123 TT |
497 | ext4_group_t block_group; |
498 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
ac27a0ec DK |
499 | |
500 | /* Try to find previous block */ | |
501 | for (p = ind->p - 1; p >= start; p--) { | |
502 | if (*p) | |
503 | return le32_to_cpu(*p); | |
504 | } | |
505 | ||
506 | /* No such thing, so let's try location of indirect block */ | |
507 | if (ind->bh) | |
508 | return ind->bh->b_blocknr; | |
509 | ||
510 | /* | |
511 | * It is going to be referred to from the inode itself? OK, just put it | |
512 | * into the same cylinder group then. | |
513 | */ | |
a4912123 TT |
514 | block_group = ei->i_block_group; |
515 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
516 | block_group &= ~(flex_size-1); | |
517 | if (S_ISREG(inode->i_mode)) | |
518 | block_group++; | |
519 | } | |
520 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
74d3487f VC |
521 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; |
522 | ||
a4912123 TT |
523 | /* |
524 | * If we are doing delayed allocation, we don't need take | |
525 | * colour into account. | |
526 | */ | |
527 | if (test_opt(inode->i_sb, DELALLOC)) | |
528 | return bg_start; | |
529 | ||
74d3487f VC |
530 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) |
531 | colour = (current->pid % 16) * | |
617ba13b | 532 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
74d3487f VC |
533 | else |
534 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
ac27a0ec DK |
535 | return bg_start + colour; |
536 | } | |
537 | ||
538 | /** | |
1cc8dcf5 | 539 | * ext4_find_goal - find a preferred place for allocation. |
ac27a0ec DK |
540 | * @inode: owner |
541 | * @block: block we want | |
ac27a0ec | 542 | * @partial: pointer to the last triple within a chain |
ac27a0ec | 543 | * |
1cc8dcf5 | 544 | * Normally this function find the preferred place for block allocation, |
fb01bfda | 545 | * returns it. |
fb0a387d ES |
546 | * Because this is only used for non-extent files, we limit the block nr |
547 | * to 32 bits. | |
ac27a0ec | 548 | */ |
725d26d3 | 549 | static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block, |
de9a55b8 | 550 | Indirect *partial) |
ac27a0ec | 551 | { |
fb0a387d ES |
552 | ext4_fsblk_t goal; |
553 | ||
ac27a0ec | 554 | /* |
c2ea3fde | 555 | * XXX need to get goal block from mballoc's data structures |
ac27a0ec | 556 | */ |
ac27a0ec | 557 | |
fb0a387d ES |
558 | goal = ext4_find_near(inode, partial); |
559 | goal = goal & EXT4_MAX_BLOCK_FILE_PHYS; | |
560 | return goal; | |
ac27a0ec DK |
561 | } |
562 | ||
563 | /** | |
225db7d3 | 564 | * ext4_blks_to_allocate - Look up the block map and count the number |
ac27a0ec DK |
565 | * of direct blocks need to be allocated for the given branch. |
566 | * | |
567 | * @branch: chain of indirect blocks | |
568 | * @k: number of blocks need for indirect blocks | |
569 | * @blks: number of data blocks to be mapped. | |
570 | * @blocks_to_boundary: the offset in the indirect block | |
571 | * | |
572 | * return the total number of blocks to be allocate, including the | |
573 | * direct and indirect blocks. | |
574 | */ | |
498e5f24 | 575 | static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks, |
de9a55b8 | 576 | int blocks_to_boundary) |
ac27a0ec | 577 | { |
498e5f24 | 578 | unsigned int count = 0; |
ac27a0ec DK |
579 | |
580 | /* | |
581 | * Simple case, [t,d]Indirect block(s) has not allocated yet | |
582 | * then it's clear blocks on that path have not allocated | |
583 | */ | |
584 | if (k > 0) { | |
585 | /* right now we don't handle cross boundary allocation */ | |
586 | if (blks < blocks_to_boundary + 1) | |
587 | count += blks; | |
588 | else | |
589 | count += blocks_to_boundary + 1; | |
590 | return count; | |
591 | } | |
592 | ||
593 | count++; | |
594 | while (count < blks && count <= blocks_to_boundary && | |
595 | le32_to_cpu(*(branch[0].p + count)) == 0) { | |
596 | count++; | |
597 | } | |
598 | return count; | |
599 | } | |
600 | ||
601 | /** | |
617ba13b | 602 | * ext4_alloc_blocks: multiple allocate blocks needed for a branch |
225db7d3 TT |
603 | * @handle: handle for this transaction |
604 | * @inode: inode which needs allocated blocks | |
605 | * @iblock: the logical block to start allocated at | |
606 | * @goal: preferred physical block of allocation | |
ac27a0ec DK |
607 | * @indirect_blks: the number of blocks need to allocate for indirect |
608 | * blocks | |
225db7d3 | 609 | * @blks: number of desired blocks |
ac27a0ec DK |
610 | * @new_blocks: on return it will store the new block numbers for |
611 | * the indirect blocks(if needed) and the first direct block, | |
225db7d3 TT |
612 | * @err: on return it will store the error code |
613 | * | |
614 | * This function will return the number of blocks allocated as | |
615 | * requested by the passed-in parameters. | |
ac27a0ec | 616 | */ |
617ba13b | 617 | static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
618 | ext4_lblk_t iblock, ext4_fsblk_t goal, |
619 | int indirect_blks, int blks, | |
620 | ext4_fsblk_t new_blocks[4], int *err) | |
ac27a0ec | 621 | { |
815a1130 | 622 | struct ext4_allocation_request ar; |
ac27a0ec | 623 | int target, i; |
7061eba7 | 624 | unsigned long count = 0, blk_allocated = 0; |
ac27a0ec | 625 | int index = 0; |
617ba13b | 626 | ext4_fsblk_t current_block = 0; |
ac27a0ec DK |
627 | int ret = 0; |
628 | ||
629 | /* | |
630 | * Here we try to allocate the requested multiple blocks at once, | |
631 | * on a best-effort basis. | |
632 | * To build a branch, we should allocate blocks for | |
633 | * the indirect blocks(if not allocated yet), and at least | |
634 | * the first direct block of this branch. That's the | |
635 | * minimum number of blocks need to allocate(required) | |
636 | */ | |
7061eba7 AK |
637 | /* first we try to allocate the indirect blocks */ |
638 | target = indirect_blks; | |
639 | while (target > 0) { | |
ac27a0ec DK |
640 | count = target; |
641 | /* allocating blocks for indirect blocks and direct blocks */ | |
7061eba7 AK |
642 | current_block = ext4_new_meta_blocks(handle, inode, |
643 | goal, &count, err); | |
ac27a0ec DK |
644 | if (*err) |
645 | goto failed_out; | |
646 | ||
273df556 FM |
647 | if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) { |
648 | EXT4_ERROR_INODE(inode, | |
649 | "current_block %llu + count %lu > %d!", | |
650 | current_block, count, | |
651 | EXT4_MAX_BLOCK_FILE_PHYS); | |
652 | *err = -EIO; | |
653 | goto failed_out; | |
654 | } | |
fb0a387d | 655 | |
ac27a0ec DK |
656 | target -= count; |
657 | /* allocate blocks for indirect blocks */ | |
658 | while (index < indirect_blks && count) { | |
659 | new_blocks[index++] = current_block++; | |
660 | count--; | |
661 | } | |
7061eba7 AK |
662 | if (count > 0) { |
663 | /* | |
664 | * save the new block number | |
665 | * for the first direct block | |
666 | */ | |
667 | new_blocks[index] = current_block; | |
668 | printk(KERN_INFO "%s returned more blocks than " | |
669 | "requested\n", __func__); | |
670 | WARN_ON(1); | |
ac27a0ec | 671 | break; |
7061eba7 | 672 | } |
ac27a0ec DK |
673 | } |
674 | ||
7061eba7 AK |
675 | target = blks - count ; |
676 | blk_allocated = count; | |
677 | if (!target) | |
678 | goto allocated; | |
679 | /* Now allocate data blocks */ | |
815a1130 TT |
680 | memset(&ar, 0, sizeof(ar)); |
681 | ar.inode = inode; | |
682 | ar.goal = goal; | |
683 | ar.len = target; | |
684 | ar.logical = iblock; | |
685 | if (S_ISREG(inode->i_mode)) | |
686 | /* enable in-core preallocation only for regular files */ | |
687 | ar.flags = EXT4_MB_HINT_DATA; | |
688 | ||
689 | current_block = ext4_mb_new_blocks(handle, &ar, err); | |
273df556 FM |
690 | if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) { |
691 | EXT4_ERROR_INODE(inode, | |
692 | "current_block %llu + ar.len %d > %d!", | |
693 | current_block, ar.len, | |
694 | EXT4_MAX_BLOCK_FILE_PHYS); | |
695 | *err = -EIO; | |
696 | goto failed_out; | |
697 | } | |
815a1130 | 698 | |
7061eba7 AK |
699 | if (*err && (target == blks)) { |
700 | /* | |
701 | * if the allocation failed and we didn't allocate | |
702 | * any blocks before | |
703 | */ | |
704 | goto failed_out; | |
705 | } | |
706 | if (!*err) { | |
707 | if (target == blks) { | |
de9a55b8 TT |
708 | /* |
709 | * save the new block number | |
710 | * for the first direct block | |
711 | */ | |
7061eba7 AK |
712 | new_blocks[index] = current_block; |
713 | } | |
815a1130 | 714 | blk_allocated += ar.len; |
7061eba7 AK |
715 | } |
716 | allocated: | |
ac27a0ec | 717 | /* total number of blocks allocated for direct blocks */ |
7061eba7 | 718 | ret = blk_allocated; |
ac27a0ec DK |
719 | *err = 0; |
720 | return ret; | |
721 | failed_out: | |
af5bc92d | 722 | for (i = 0; i < index; i++) |
7dc57615 | 723 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); |
ac27a0ec DK |
724 | return ret; |
725 | } | |
726 | ||
727 | /** | |
617ba13b | 728 | * ext4_alloc_branch - allocate and set up a chain of blocks. |
225db7d3 | 729 | * @handle: handle for this transaction |
ac27a0ec DK |
730 | * @inode: owner |
731 | * @indirect_blks: number of allocated indirect blocks | |
732 | * @blks: number of allocated direct blocks | |
225db7d3 | 733 | * @goal: preferred place for allocation |
ac27a0ec DK |
734 | * @offsets: offsets (in the blocks) to store the pointers to next. |
735 | * @branch: place to store the chain in. | |
736 | * | |
737 | * This function allocates blocks, zeroes out all but the last one, | |
738 | * links them into chain and (if we are synchronous) writes them to disk. | |
739 | * In other words, it prepares a branch that can be spliced onto the | |
740 | * inode. It stores the information about that chain in the branch[], in | |
617ba13b | 741 | * the same format as ext4_get_branch() would do. We are calling it after |
ac27a0ec DK |
742 | * we had read the existing part of chain and partial points to the last |
743 | * triple of that (one with zero ->key). Upon the exit we have the same | |
617ba13b | 744 | * picture as after the successful ext4_get_block(), except that in one |
ac27a0ec DK |
745 | * place chain is disconnected - *branch->p is still zero (we did not |
746 | * set the last link), but branch->key contains the number that should | |
747 | * be placed into *branch->p to fill that gap. | |
748 | * | |
749 | * If allocation fails we free all blocks we've allocated (and forget | |
750 | * their buffer_heads) and return the error value the from failed | |
617ba13b | 751 | * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
ac27a0ec DK |
752 | * as described above and return 0. |
753 | */ | |
617ba13b | 754 | static int ext4_alloc_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
755 | ext4_lblk_t iblock, int indirect_blks, |
756 | int *blks, ext4_fsblk_t goal, | |
757 | ext4_lblk_t *offsets, Indirect *branch) | |
ac27a0ec DK |
758 | { |
759 | int blocksize = inode->i_sb->s_blocksize; | |
760 | int i, n = 0; | |
761 | int err = 0; | |
762 | struct buffer_head *bh; | |
763 | int num; | |
617ba13b MC |
764 | ext4_fsblk_t new_blocks[4]; |
765 | ext4_fsblk_t current_block; | |
ac27a0ec | 766 | |
7061eba7 | 767 | num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks, |
ac27a0ec DK |
768 | *blks, new_blocks, &err); |
769 | if (err) | |
770 | return err; | |
771 | ||
772 | branch[0].key = cpu_to_le32(new_blocks[0]); | |
773 | /* | |
774 | * metadata blocks and data blocks are allocated. | |
775 | */ | |
776 | for (n = 1; n <= indirect_blks; n++) { | |
777 | /* | |
778 | * Get buffer_head for parent block, zero it out | |
779 | * and set the pointer to new one, then send | |
780 | * parent to disk. | |
781 | */ | |
782 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); | |
87783690 NK |
783 | if (unlikely(!bh)) { |
784 | err = -EIO; | |
785 | goto failed; | |
786 | } | |
787 | ||
ac27a0ec DK |
788 | branch[n].bh = bh; |
789 | lock_buffer(bh); | |
790 | BUFFER_TRACE(bh, "call get_create_access"); | |
617ba13b | 791 | err = ext4_journal_get_create_access(handle, bh); |
ac27a0ec | 792 | if (err) { |
6487a9d3 CW |
793 | /* Don't brelse(bh) here; it's done in |
794 | * ext4_journal_forget() below */ | |
ac27a0ec | 795 | unlock_buffer(bh); |
ac27a0ec DK |
796 | goto failed; |
797 | } | |
798 | ||
799 | memset(bh->b_data, 0, blocksize); | |
800 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; | |
801 | branch[n].key = cpu_to_le32(new_blocks[n]); | |
802 | *branch[n].p = branch[n].key; | |
af5bc92d | 803 | if (n == indirect_blks) { |
ac27a0ec DK |
804 | current_block = new_blocks[n]; |
805 | /* | |
806 | * End of chain, update the last new metablock of | |
807 | * the chain to point to the new allocated | |
808 | * data blocks numbers | |
809 | */ | |
de9a55b8 | 810 | for (i = 1; i < num; i++) |
ac27a0ec DK |
811 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
812 | } | |
813 | BUFFER_TRACE(bh, "marking uptodate"); | |
814 | set_buffer_uptodate(bh); | |
815 | unlock_buffer(bh); | |
816 | ||
0390131b FM |
817 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
818 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
ac27a0ec DK |
819 | if (err) |
820 | goto failed; | |
821 | } | |
822 | *blks = num; | |
823 | return err; | |
824 | failed: | |
825 | /* Allocation failed, free what we already allocated */ | |
7dc57615 | 826 | ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0); |
ac27a0ec | 827 | for (i = 1; i <= n ; i++) { |
60e6679e | 828 | /* |
e6362609 TT |
829 | * branch[i].bh is newly allocated, so there is no |
830 | * need to revoke the block, which is why we don't | |
831 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
b7e57e7c | 832 | */ |
7dc57615 | 833 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, |
e6362609 | 834 | EXT4_FREE_BLOCKS_FORGET); |
ac27a0ec | 835 | } |
e6362609 | 836 | for (i = n+1; i < indirect_blks; i++) |
7dc57615 | 837 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); |
ac27a0ec | 838 | |
7dc57615 | 839 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0); |
ac27a0ec DK |
840 | |
841 | return err; | |
842 | } | |
843 | ||
844 | /** | |
617ba13b | 845 | * ext4_splice_branch - splice the allocated branch onto inode. |
225db7d3 | 846 | * @handle: handle for this transaction |
ac27a0ec DK |
847 | * @inode: owner |
848 | * @block: (logical) number of block we are adding | |
849 | * @chain: chain of indirect blocks (with a missing link - see | |
617ba13b | 850 | * ext4_alloc_branch) |
ac27a0ec DK |
851 | * @where: location of missing link |
852 | * @num: number of indirect blocks we are adding | |
853 | * @blks: number of direct blocks we are adding | |
854 | * | |
855 | * This function fills the missing link and does all housekeeping needed in | |
856 | * inode (->i_blocks, etc.). In case of success we end up with the full | |
857 | * chain to new block and return 0. | |
858 | */ | |
617ba13b | 859 | static int ext4_splice_branch(handle_t *handle, struct inode *inode, |
de9a55b8 TT |
860 | ext4_lblk_t block, Indirect *where, int num, |
861 | int blks) | |
ac27a0ec DK |
862 | { |
863 | int i; | |
864 | int err = 0; | |
617ba13b | 865 | ext4_fsblk_t current_block; |
ac27a0ec | 866 | |
ac27a0ec DK |
867 | /* |
868 | * If we're splicing into a [td]indirect block (as opposed to the | |
869 | * inode) then we need to get write access to the [td]indirect block | |
870 | * before the splice. | |
871 | */ | |
872 | if (where->bh) { | |
873 | BUFFER_TRACE(where->bh, "get_write_access"); | |
617ba13b | 874 | err = ext4_journal_get_write_access(handle, where->bh); |
ac27a0ec DK |
875 | if (err) |
876 | goto err_out; | |
877 | } | |
878 | /* That's it */ | |
879 | ||
880 | *where->p = where->key; | |
881 | ||
882 | /* | |
883 | * Update the host buffer_head or inode to point to more just allocated | |
884 | * direct blocks blocks | |
885 | */ | |
886 | if (num == 0 && blks > 1) { | |
887 | current_block = le32_to_cpu(where->key) + 1; | |
888 | for (i = 1; i < blks; i++) | |
af5bc92d | 889 | *(where->p + i) = cpu_to_le32(current_block++); |
ac27a0ec DK |
890 | } |
891 | ||
ac27a0ec | 892 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
ac27a0ec DK |
893 | /* had we spliced it onto indirect block? */ |
894 | if (where->bh) { | |
895 | /* | |
896 | * If we spliced it onto an indirect block, we haven't | |
897 | * altered the inode. Note however that if it is being spliced | |
898 | * onto an indirect block at the very end of the file (the | |
899 | * file is growing) then we *will* alter the inode to reflect | |
900 | * the new i_size. But that is not done here - it is done in | |
617ba13b | 901 | * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. |
ac27a0ec DK |
902 | */ |
903 | jbd_debug(5, "splicing indirect only\n"); | |
0390131b FM |
904 | BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata"); |
905 | err = ext4_handle_dirty_metadata(handle, inode, where->bh); | |
ac27a0ec DK |
906 | if (err) |
907 | goto err_out; | |
908 | } else { | |
909 | /* | |
910 | * OK, we spliced it into the inode itself on a direct block. | |
ac27a0ec | 911 | */ |
41591750 | 912 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
913 | jbd_debug(5, "splicing direct\n"); |
914 | } | |
915 | return err; | |
916 | ||
917 | err_out: | |
918 | for (i = 1; i <= num; i++) { | |
60e6679e | 919 | /* |
e6362609 TT |
920 | * branch[i].bh is newly allocated, so there is no |
921 | * need to revoke the block, which is why we don't | |
922 | * need to set EXT4_FREE_BLOCKS_METADATA. | |
b7e57e7c | 923 | */ |
e6362609 TT |
924 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, |
925 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec | 926 | } |
7dc57615 | 927 | ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key), |
e6362609 | 928 | blks, 0); |
ac27a0ec DK |
929 | |
930 | return err; | |
931 | } | |
932 | ||
933 | /* | |
e35fd660 | 934 | * The ext4_ind_map_blocks() function handles non-extents inodes |
b920c755 | 935 | * (i.e., using the traditional indirect/double-indirect i_blocks |
e35fd660 | 936 | * scheme) for ext4_map_blocks(). |
b920c755 | 937 | * |
ac27a0ec DK |
938 | * Allocation strategy is simple: if we have to allocate something, we will |
939 | * have to go the whole way to leaf. So let's do it before attaching anything | |
940 | * to tree, set linkage between the newborn blocks, write them if sync is | |
941 | * required, recheck the path, free and repeat if check fails, otherwise | |
942 | * set the last missing link (that will protect us from any truncate-generated | |
943 | * removals - all blocks on the path are immune now) and possibly force the | |
944 | * write on the parent block. | |
945 | * That has a nice additional property: no special recovery from the failed | |
946 | * allocations is needed - we simply release blocks and do not touch anything | |
947 | * reachable from inode. | |
948 | * | |
949 | * `handle' can be NULL if create == 0. | |
950 | * | |
ac27a0ec DK |
951 | * return > 0, # of blocks mapped or allocated. |
952 | * return = 0, if plain lookup failed. | |
953 | * return < 0, error case. | |
c278bfec | 954 | * |
b920c755 TT |
955 | * The ext4_ind_get_blocks() function should be called with |
956 | * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem | |
957 | * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or | |
958 | * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system | |
959 | * blocks. | |
ac27a0ec | 960 | */ |
e35fd660 TT |
961 | static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, |
962 | struct ext4_map_blocks *map, | |
de9a55b8 | 963 | int flags) |
ac27a0ec DK |
964 | { |
965 | int err = -EIO; | |
725d26d3 | 966 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
967 | Indirect chain[4]; |
968 | Indirect *partial; | |
617ba13b | 969 | ext4_fsblk_t goal; |
ac27a0ec DK |
970 | int indirect_blks; |
971 | int blocks_to_boundary = 0; | |
972 | int depth; | |
ac27a0ec | 973 | int count = 0; |
617ba13b | 974 | ext4_fsblk_t first_block = 0; |
ac27a0ec | 975 | |
0562e0ba | 976 | trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); |
12e9b892 | 977 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); |
c2177057 | 978 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
e35fd660 | 979 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, |
de9a55b8 | 980 | &blocks_to_boundary); |
ac27a0ec DK |
981 | |
982 | if (depth == 0) | |
983 | goto out; | |
984 | ||
617ba13b | 985 | partial = ext4_get_branch(inode, depth, offsets, chain, &err); |
ac27a0ec DK |
986 | |
987 | /* Simplest case - block found, no allocation needed */ | |
988 | if (!partial) { | |
989 | first_block = le32_to_cpu(chain[depth - 1].key); | |
ac27a0ec DK |
990 | count++; |
991 | /*map more blocks*/ | |
e35fd660 | 992 | while (count < map->m_len && count <= blocks_to_boundary) { |
617ba13b | 993 | ext4_fsblk_t blk; |
ac27a0ec | 994 | |
ac27a0ec DK |
995 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
996 | ||
997 | if (blk == first_block + count) | |
998 | count++; | |
999 | else | |
1000 | break; | |
1001 | } | |
c278bfec | 1002 | goto got_it; |
ac27a0ec DK |
1003 | } |
1004 | ||
1005 | /* Next simple case - plain lookup or failed read of indirect block */ | |
c2177057 | 1006 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO) |
ac27a0ec DK |
1007 | goto cleanup; |
1008 | ||
ac27a0ec | 1009 | /* |
c2ea3fde | 1010 | * Okay, we need to do block allocation. |
ac27a0ec | 1011 | */ |
e35fd660 | 1012 | goal = ext4_find_goal(inode, map->m_lblk, partial); |
ac27a0ec DK |
1013 | |
1014 | /* the number of blocks need to allocate for [d,t]indirect blocks */ | |
1015 | indirect_blks = (chain + depth) - partial - 1; | |
1016 | ||
1017 | /* | |
1018 | * Next look up the indirect map to count the totoal number of | |
1019 | * direct blocks to allocate for this branch. | |
1020 | */ | |
617ba13b | 1021 | count = ext4_blks_to_allocate(partial, indirect_blks, |
e35fd660 | 1022 | map->m_len, blocks_to_boundary); |
ac27a0ec | 1023 | /* |
617ba13b | 1024 | * Block out ext4_truncate while we alter the tree |
ac27a0ec | 1025 | */ |
e35fd660 | 1026 | err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks, |
de9a55b8 TT |
1027 | &count, goal, |
1028 | offsets + (partial - chain), partial); | |
ac27a0ec DK |
1029 | |
1030 | /* | |
617ba13b | 1031 | * The ext4_splice_branch call will free and forget any buffers |
ac27a0ec DK |
1032 | * on the new chain if there is a failure, but that risks using |
1033 | * up transaction credits, especially for bitmaps where the | |
1034 | * credits cannot be returned. Can we handle this somehow? We | |
1035 | * may need to return -EAGAIN upwards in the worst case. --sct | |
1036 | */ | |
1037 | if (!err) | |
e35fd660 | 1038 | err = ext4_splice_branch(handle, inode, map->m_lblk, |
de9a55b8 | 1039 | partial, indirect_blks, count); |
2bba702d | 1040 | if (err) |
ac27a0ec DK |
1041 | goto cleanup; |
1042 | ||
e35fd660 | 1043 | map->m_flags |= EXT4_MAP_NEW; |
b436b9be JK |
1044 | |
1045 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
ac27a0ec | 1046 | got_it: |
e35fd660 TT |
1047 | map->m_flags |= EXT4_MAP_MAPPED; |
1048 | map->m_pblk = le32_to_cpu(chain[depth-1].key); | |
1049 | map->m_len = count; | |
ac27a0ec | 1050 | if (count > blocks_to_boundary) |
e35fd660 | 1051 | map->m_flags |= EXT4_MAP_BOUNDARY; |
ac27a0ec DK |
1052 | err = count; |
1053 | /* Clean up and exit */ | |
1054 | partial = chain + depth - 1; /* the whole chain */ | |
1055 | cleanup: | |
1056 | while (partial > chain) { | |
1057 | BUFFER_TRACE(partial->bh, "call brelse"); | |
1058 | brelse(partial->bh); | |
1059 | partial--; | |
1060 | } | |
ac27a0ec | 1061 | out: |
0562e0ba JZ |
1062 | trace_ext4_ind_map_blocks_exit(inode, map->m_lblk, |
1063 | map->m_pblk, map->m_len, err); | |
ac27a0ec DK |
1064 | return err; |
1065 | } | |
1066 | ||
a9e7f447 DM |
1067 | #ifdef CONFIG_QUOTA |
1068 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 1069 | { |
a9e7f447 | 1070 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 1071 | } |
a9e7f447 | 1072 | #endif |
9d0be502 | 1073 | |
12219aea AK |
1074 | /* |
1075 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 1076 | * to allocate a new block at @lblocks for non extent file based file |
12219aea | 1077 | */ |
9d0be502 TT |
1078 | static int ext4_indirect_calc_metadata_amount(struct inode *inode, |
1079 | sector_t lblock) | |
12219aea | 1080 | { |
9d0be502 | 1081 | struct ext4_inode_info *ei = EXT4_I(inode); |
d330a5be | 1082 | sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1); |
9d0be502 | 1083 | int blk_bits; |
12219aea | 1084 | |
9d0be502 TT |
1085 | if (lblock < EXT4_NDIR_BLOCKS) |
1086 | return 0; | |
12219aea | 1087 | |
9d0be502 | 1088 | lblock -= EXT4_NDIR_BLOCKS; |
12219aea | 1089 | |
9d0be502 TT |
1090 | if (ei->i_da_metadata_calc_len && |
1091 | (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) { | |
1092 | ei->i_da_metadata_calc_len++; | |
1093 | return 0; | |
1094 | } | |
1095 | ei->i_da_metadata_calc_last_lblock = lblock & dind_mask; | |
1096 | ei->i_da_metadata_calc_len = 1; | |
d330a5be | 1097 | blk_bits = order_base_2(lblock); |
9d0be502 | 1098 | return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1; |
12219aea AK |
1099 | } |
1100 | ||
1101 | /* | |
1102 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 1103 | * to allocate a block located at @lblock |
12219aea | 1104 | */ |
01f49d0b | 1105 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 1106 | { |
12e9b892 | 1107 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 1108 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 1109 | |
9d0be502 | 1110 | return ext4_indirect_calc_metadata_amount(inode, lblock); |
12219aea AK |
1111 | } |
1112 | ||
0637c6f4 TT |
1113 | /* |
1114 | * Called with i_data_sem down, which is important since we can call | |
1115 | * ext4_discard_preallocations() from here. | |
1116 | */ | |
5f634d06 AK |
1117 | void ext4_da_update_reserve_space(struct inode *inode, |
1118 | int used, int quota_claim) | |
12219aea AK |
1119 | { |
1120 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1121 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
1122 | |
1123 | spin_lock(&ei->i_block_reservation_lock); | |
f8ec9d68 | 1124 | trace_ext4_da_update_reserve_space(inode, used); |
0637c6f4 TT |
1125 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
1126 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d " | |
1127 | "with only %d reserved data blocks\n", | |
1128 | __func__, inode->i_ino, used, | |
1129 | ei->i_reserved_data_blocks); | |
1130 | WARN_ON(1); | |
1131 | used = ei->i_reserved_data_blocks; | |
1132 | } | |
12219aea | 1133 | |
0637c6f4 TT |
1134 | /* Update per-inode reservations */ |
1135 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 1136 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
72b8ab9d ES |
1137 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1138 | used + ei->i_allocated_meta_blocks); | |
0637c6f4 | 1139 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 1140 | |
0637c6f4 TT |
1141 | if (ei->i_reserved_data_blocks == 0) { |
1142 | /* | |
1143 | * We can release all of the reserved metadata blocks | |
1144 | * only when we have written all of the delayed | |
1145 | * allocation blocks. | |
1146 | */ | |
72b8ab9d ES |
1147 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1148 | ei->i_reserved_meta_blocks); | |
ee5f4d9c | 1149 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1150 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 1151 | } |
12219aea | 1152 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1153 | |
72b8ab9d ES |
1154 | /* Update quota subsystem for data blocks */ |
1155 | if (quota_claim) | |
5dd4056d | 1156 | dquot_claim_block(inode, used); |
72b8ab9d | 1157 | else { |
5f634d06 AK |
1158 | /* |
1159 | * We did fallocate with an offset that is already delayed | |
1160 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 1161 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 1162 | */ |
72b8ab9d | 1163 | dquot_release_reservation_block(inode, used); |
5f634d06 | 1164 | } |
d6014301 AK |
1165 | |
1166 | /* | |
1167 | * If we have done all the pending block allocations and if | |
1168 | * there aren't any writers on the inode, we can discard the | |
1169 | * inode's preallocations. | |
1170 | */ | |
0637c6f4 TT |
1171 | if ((ei->i_reserved_data_blocks == 0) && |
1172 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 1173 | ext4_discard_preallocations(inode); |
12219aea AK |
1174 | } |
1175 | ||
e29136f8 | 1176 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
1177 | unsigned int line, |
1178 | struct ext4_map_blocks *map) | |
6fd058f7 | 1179 | { |
24676da4 TT |
1180 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
1181 | map->m_len)) { | |
c398eda0 TT |
1182 | ext4_error_inode(inode, func, line, map->m_pblk, |
1183 | "lblock %lu mapped to illegal pblock " | |
1184 | "(length %d)", (unsigned long) map->m_lblk, | |
1185 | map->m_len); | |
6fd058f7 TT |
1186 | return -EIO; |
1187 | } | |
1188 | return 0; | |
1189 | } | |
1190 | ||
e29136f8 | 1191 | #define check_block_validity(inode, map) \ |
c398eda0 | 1192 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 1193 | |
55138e0b | 1194 | /* |
1f94533d TT |
1195 | * Return the number of contiguous dirty pages in a given inode |
1196 | * starting at page frame idx. | |
55138e0b TT |
1197 | */ |
1198 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
1199 | unsigned int max_pages) | |
1200 | { | |
1201 | struct address_space *mapping = inode->i_mapping; | |
1202 | pgoff_t index; | |
1203 | struct pagevec pvec; | |
1204 | pgoff_t num = 0; | |
1205 | int i, nr_pages, done = 0; | |
1206 | ||
1207 | if (max_pages == 0) | |
1208 | return 0; | |
1209 | pagevec_init(&pvec, 0); | |
1210 | while (!done) { | |
1211 | index = idx; | |
1212 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
1213 | PAGECACHE_TAG_DIRTY, | |
1214 | (pgoff_t)PAGEVEC_SIZE); | |
1215 | if (nr_pages == 0) | |
1216 | break; | |
1217 | for (i = 0; i < nr_pages; i++) { | |
1218 | struct page *page = pvec.pages[i]; | |
1219 | struct buffer_head *bh, *head; | |
1220 | ||
1221 | lock_page(page); | |
1222 | if (unlikely(page->mapping != mapping) || | |
1223 | !PageDirty(page) || | |
1224 | PageWriteback(page) || | |
1225 | page->index != idx) { | |
1226 | done = 1; | |
1227 | unlock_page(page); | |
1228 | break; | |
1229 | } | |
1f94533d TT |
1230 | if (page_has_buffers(page)) { |
1231 | bh = head = page_buffers(page); | |
1232 | do { | |
1233 | if (!buffer_delay(bh) && | |
1234 | !buffer_unwritten(bh)) | |
1235 | done = 1; | |
1236 | bh = bh->b_this_page; | |
1237 | } while (!done && (bh != head)); | |
1238 | } | |
55138e0b TT |
1239 | unlock_page(page); |
1240 | if (done) | |
1241 | break; | |
1242 | idx++; | |
1243 | num++; | |
659c6009 ES |
1244 | if (num >= max_pages) { |
1245 | done = 1; | |
55138e0b | 1246 | break; |
659c6009 | 1247 | } |
55138e0b TT |
1248 | } |
1249 | pagevec_release(&pvec); | |
1250 | } | |
1251 | return num; | |
1252 | } | |
1253 | ||
f5ab0d1f | 1254 | /* |
e35fd660 | 1255 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 1256 | * and returns if the blocks are already mapped. |
f5ab0d1f | 1257 | * |
f5ab0d1f MC |
1258 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
1259 | * and store the allocated blocks in the result buffer head and mark it | |
1260 | * mapped. | |
1261 | * | |
e35fd660 TT |
1262 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
1263 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
1264 | * based files |
1265 | * | |
1266 | * On success, it returns the number of blocks being mapped or allocate. | |
1267 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
1268 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
1269 | * the buffer head is mapped. | |
1270 | * | |
1271 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
1272 | * that casem, buffer head is unmapped | |
1273 | * | |
1274 | * It returns the error in case of allocation failure. | |
1275 | */ | |
e35fd660 TT |
1276 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
1277 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
1278 | { |
1279 | int retval; | |
f5ab0d1f | 1280 | |
e35fd660 TT |
1281 | map->m_flags = 0; |
1282 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
1283 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
1284 | (unsigned long) map->m_lblk); | |
4df3d265 | 1285 | /* |
b920c755 TT |
1286 | * Try to see if we can get the block without requesting a new |
1287 | * file system block. | |
4df3d265 AK |
1288 | */ |
1289 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 1290 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 1291 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 1292 | } else { |
e35fd660 | 1293 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 1294 | } |
4df3d265 | 1295 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 1296 | |
e35fd660 | 1297 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 1298 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
1299 | if (ret != 0) |
1300 | return ret; | |
1301 | } | |
1302 | ||
f5ab0d1f | 1303 | /* If it is only a block(s) look up */ |
c2177057 | 1304 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
1305 | return retval; |
1306 | ||
1307 | /* | |
1308 | * Returns if the blocks have already allocated | |
1309 | * | |
1310 | * Note that if blocks have been preallocated | |
1311 | * ext4_ext_get_block() returns th create = 0 | |
1312 | * with buffer head unmapped. | |
1313 | */ | |
e35fd660 | 1314 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
1315 | return retval; |
1316 | ||
2a8964d6 AK |
1317 | /* |
1318 | * When we call get_blocks without the create flag, the | |
1319 | * BH_Unwritten flag could have gotten set if the blocks | |
1320 | * requested were part of a uninitialized extent. We need to | |
1321 | * clear this flag now that we are committed to convert all or | |
1322 | * part of the uninitialized extent to be an initialized | |
1323 | * extent. This is because we need to avoid the combination | |
1324 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
1325 | * set on the buffer_head. | |
1326 | */ | |
e35fd660 | 1327 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 1328 | |
4df3d265 | 1329 | /* |
f5ab0d1f MC |
1330 | * New blocks allocate and/or writing to uninitialized extent |
1331 | * will possibly result in updating i_data, so we take | |
1332 | * the write lock of i_data_sem, and call get_blocks() | |
1333 | * with create == 1 flag. | |
4df3d265 AK |
1334 | */ |
1335 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
1336 | |
1337 | /* | |
1338 | * if the caller is from delayed allocation writeout path | |
1339 | * we have already reserved fs blocks for allocation | |
1340 | * let the underlying get_block() function know to | |
1341 | * avoid double accounting | |
1342 | */ | |
c2177057 | 1343 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 1344 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
1345 | /* |
1346 | * We need to check for EXT4 here because migrate | |
1347 | * could have changed the inode type in between | |
1348 | */ | |
12e9b892 | 1349 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 1350 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 1351 | } else { |
e35fd660 | 1352 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 1353 | |
e35fd660 | 1354 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
1355 | /* |
1356 | * We allocated new blocks which will result in | |
1357 | * i_data's format changing. Force the migrate | |
1358 | * to fail by clearing migrate flags | |
1359 | */ | |
19f5fb7a | 1360 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 1361 | } |
d2a17637 | 1362 | |
5f634d06 AK |
1363 | /* |
1364 | * Update reserved blocks/metadata blocks after successful | |
1365 | * block allocation which had been deferred till now. We don't | |
1366 | * support fallocate for non extent files. So we can update | |
1367 | * reserve space here. | |
1368 | */ | |
1369 | if ((retval > 0) && | |
1296cc85 | 1370 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
1371 | ext4_da_update_reserve_space(inode, retval, 1); |
1372 | } | |
2ac3b6e0 | 1373 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 1374 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 1375 | |
4df3d265 | 1376 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 1377 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 1378 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
1379 | if (ret != 0) |
1380 | return ret; | |
1381 | } | |
0e855ac8 AK |
1382 | return retval; |
1383 | } | |
1384 | ||
f3bd1f3f MC |
1385 | /* Maximum number of blocks we map for direct IO at once. */ |
1386 | #define DIO_MAX_BLOCKS 4096 | |
1387 | ||
2ed88685 TT |
1388 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
1389 | struct buffer_head *bh, int flags) | |
ac27a0ec | 1390 | { |
3e4fdaf8 | 1391 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 1392 | struct ext4_map_blocks map; |
7fb5409d | 1393 | int ret = 0, started = 0; |
f3bd1f3f | 1394 | int dio_credits; |
ac27a0ec | 1395 | |
2ed88685 TT |
1396 | map.m_lblk = iblock; |
1397 | map.m_len = bh->b_size >> inode->i_blkbits; | |
1398 | ||
1399 | if (flags && !handle) { | |
7fb5409d | 1400 | /* Direct IO write... */ |
2ed88685 TT |
1401 | if (map.m_len > DIO_MAX_BLOCKS) |
1402 | map.m_len = DIO_MAX_BLOCKS; | |
1403 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 1404 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 1405 | if (IS_ERR(handle)) { |
ac27a0ec | 1406 | ret = PTR_ERR(handle); |
2ed88685 | 1407 | return ret; |
ac27a0ec | 1408 | } |
7fb5409d | 1409 | started = 1; |
ac27a0ec DK |
1410 | } |
1411 | ||
2ed88685 | 1412 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 1413 | if (ret > 0) { |
2ed88685 TT |
1414 | map_bh(bh, inode->i_sb, map.m_pblk); |
1415 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1416 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 1417 | ret = 0; |
ac27a0ec | 1418 | } |
7fb5409d JK |
1419 | if (started) |
1420 | ext4_journal_stop(handle); | |
ac27a0ec DK |
1421 | return ret; |
1422 | } | |
1423 | ||
2ed88685 TT |
1424 | int ext4_get_block(struct inode *inode, sector_t iblock, |
1425 | struct buffer_head *bh, int create) | |
1426 | { | |
1427 | return _ext4_get_block(inode, iblock, bh, | |
1428 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
1429 | } | |
1430 | ||
ac27a0ec DK |
1431 | /* |
1432 | * `handle' can be NULL if create is zero | |
1433 | */ | |
617ba13b | 1434 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 1435 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 1436 | { |
2ed88685 TT |
1437 | struct ext4_map_blocks map; |
1438 | struct buffer_head *bh; | |
ac27a0ec DK |
1439 | int fatal = 0, err; |
1440 | ||
1441 | J_ASSERT(handle != NULL || create == 0); | |
1442 | ||
2ed88685 TT |
1443 | map.m_lblk = block; |
1444 | map.m_len = 1; | |
1445 | err = ext4_map_blocks(handle, inode, &map, | |
1446 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 1447 | |
2ed88685 TT |
1448 | if (err < 0) |
1449 | *errp = err; | |
1450 | if (err <= 0) | |
1451 | return NULL; | |
1452 | *errp = 0; | |
1453 | ||
1454 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
1455 | if (!bh) { | |
1456 | *errp = -EIO; | |
1457 | return NULL; | |
ac27a0ec | 1458 | } |
2ed88685 TT |
1459 | if (map.m_flags & EXT4_MAP_NEW) { |
1460 | J_ASSERT(create != 0); | |
1461 | J_ASSERT(handle != NULL); | |
ac27a0ec | 1462 | |
2ed88685 TT |
1463 | /* |
1464 | * Now that we do not always journal data, we should | |
1465 | * keep in mind whether this should always journal the | |
1466 | * new buffer as metadata. For now, regular file | |
1467 | * writes use ext4_get_block instead, so it's not a | |
1468 | * problem. | |
1469 | */ | |
1470 | lock_buffer(bh); | |
1471 | BUFFER_TRACE(bh, "call get_create_access"); | |
1472 | fatal = ext4_journal_get_create_access(handle, bh); | |
1473 | if (!fatal && !buffer_uptodate(bh)) { | |
1474 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
1475 | set_buffer_uptodate(bh); | |
ac27a0ec | 1476 | } |
2ed88685 TT |
1477 | unlock_buffer(bh); |
1478 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1479 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
1480 | if (!fatal) | |
1481 | fatal = err; | |
1482 | } else { | |
1483 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 1484 | } |
2ed88685 TT |
1485 | if (fatal) { |
1486 | *errp = fatal; | |
1487 | brelse(bh); | |
1488 | bh = NULL; | |
1489 | } | |
1490 | return bh; | |
ac27a0ec DK |
1491 | } |
1492 | ||
617ba13b | 1493 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 1494 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 1495 | { |
af5bc92d | 1496 | struct buffer_head *bh; |
ac27a0ec | 1497 | |
617ba13b | 1498 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
1499 | if (!bh) |
1500 | return bh; | |
1501 | if (buffer_uptodate(bh)) | |
1502 | return bh; | |
1503 | ll_rw_block(READ_META, 1, &bh); | |
1504 | wait_on_buffer(bh); | |
1505 | if (buffer_uptodate(bh)) | |
1506 | return bh; | |
1507 | put_bh(bh); | |
1508 | *err = -EIO; | |
1509 | return NULL; | |
1510 | } | |
1511 | ||
af5bc92d TT |
1512 | static int walk_page_buffers(handle_t *handle, |
1513 | struct buffer_head *head, | |
1514 | unsigned from, | |
1515 | unsigned to, | |
1516 | int *partial, | |
1517 | int (*fn)(handle_t *handle, | |
1518 | struct buffer_head *bh)) | |
ac27a0ec DK |
1519 | { |
1520 | struct buffer_head *bh; | |
1521 | unsigned block_start, block_end; | |
1522 | unsigned blocksize = head->b_size; | |
1523 | int err, ret = 0; | |
1524 | struct buffer_head *next; | |
1525 | ||
af5bc92d TT |
1526 | for (bh = head, block_start = 0; |
1527 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1528 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1529 | next = bh->b_this_page; |
1530 | block_end = block_start + blocksize; | |
1531 | if (block_end <= from || block_start >= to) { | |
1532 | if (partial && !buffer_uptodate(bh)) | |
1533 | *partial = 1; | |
1534 | continue; | |
1535 | } | |
1536 | err = (*fn)(handle, bh); | |
1537 | if (!ret) | |
1538 | ret = err; | |
1539 | } | |
1540 | return ret; | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * To preserve ordering, it is essential that the hole instantiation and | |
1545 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1546 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1547 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1548 | * prepare_write() is the right place. |
1549 | * | |
617ba13b MC |
1550 | * Also, this function can nest inside ext4_writepage() -> |
1551 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
1552 | * has generated enough buffer credits to do the whole page. So we won't |
1553 | * block on the journal in that case, which is good, because the caller may | |
1554 | * be PF_MEMALLOC. | |
1555 | * | |
617ba13b | 1556 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1557 | * quota file writes. If we were to commit the transaction while thus |
1558 | * reentered, there can be a deadlock - we would be holding a quota | |
1559 | * lock, and the commit would never complete if another thread had a | |
1560 | * transaction open and was blocking on the quota lock - a ranking | |
1561 | * violation. | |
1562 | * | |
dab291af | 1563 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1564 | * will _not_ run commit under these circumstances because handle->h_ref |
1565 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1566 | * write. | |
1567 | */ | |
1568 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 1569 | struct buffer_head *bh) |
ac27a0ec | 1570 | { |
56d35a4c JK |
1571 | int dirty = buffer_dirty(bh); |
1572 | int ret; | |
1573 | ||
ac27a0ec DK |
1574 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1575 | return 0; | |
56d35a4c | 1576 | /* |
ebdec241 | 1577 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1578 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1579 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1580 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1581 | * the bit before releasing a page lock and thus writeback cannot |
1582 | * ever write the buffer. | |
1583 | */ | |
1584 | if (dirty) | |
1585 | clear_buffer_dirty(bh); | |
1586 | ret = ext4_journal_get_write_access(handle, bh); | |
1587 | if (!ret && dirty) | |
1588 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1589 | return ret; | |
ac27a0ec DK |
1590 | } |
1591 | ||
b9a4207d JK |
1592 | /* |
1593 | * Truncate blocks that were not used by write. We have to truncate the | |
1594 | * pagecache as well so that corresponding buffers get properly unmapped. | |
1595 | */ | |
1596 | static void ext4_truncate_failed_write(struct inode *inode) | |
1597 | { | |
1598 | truncate_inode_pages(inode->i_mapping, inode->i_size); | |
1599 | ext4_truncate(inode); | |
1600 | } | |
1601 | ||
744692dc JZ |
1602 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
1603 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 1604 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1605 | loff_t pos, unsigned len, unsigned flags, |
1606 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1607 | { |
af5bc92d | 1608 | struct inode *inode = mapping->host; |
1938a150 | 1609 | int ret, needed_blocks; |
ac27a0ec DK |
1610 | handle_t *handle; |
1611 | int retries = 0; | |
af5bc92d | 1612 | struct page *page; |
de9a55b8 | 1613 | pgoff_t index; |
af5bc92d | 1614 | unsigned from, to; |
bfc1af65 | 1615 | |
9bffad1e | 1616 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1617 | /* |
1618 | * Reserve one block more for addition to orphan list in case | |
1619 | * we allocate blocks but write fails for some reason | |
1620 | */ | |
1621 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1622 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1623 | from = pos & (PAGE_CACHE_SIZE - 1); |
1624 | to = from + len; | |
ac27a0ec DK |
1625 | |
1626 | retry: | |
af5bc92d TT |
1627 | handle = ext4_journal_start(inode, needed_blocks); |
1628 | if (IS_ERR(handle)) { | |
1629 | ret = PTR_ERR(handle); | |
1630 | goto out; | |
7479d2b9 | 1631 | } |
ac27a0ec | 1632 | |
ebd3610b JK |
1633 | /* We cannot recurse into the filesystem as the transaction is already |
1634 | * started */ | |
1635 | flags |= AOP_FLAG_NOFS; | |
1636 | ||
54566b2c | 1637 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
1638 | if (!page) { |
1639 | ext4_journal_stop(handle); | |
1640 | ret = -ENOMEM; | |
1641 | goto out; | |
1642 | } | |
1643 | *pagep = page; | |
1644 | ||
744692dc | 1645 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1646 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1647 | else |
6e1db88d | 1648 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1649 | |
1650 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
1651 | ret = walk_page_buffers(handle, page_buffers(page), |
1652 | from, to, NULL, do_journal_get_write_access); | |
1653 | } | |
bfc1af65 NP |
1654 | |
1655 | if (ret) { | |
af5bc92d | 1656 | unlock_page(page); |
af5bc92d | 1657 | page_cache_release(page); |
ae4d5372 | 1658 | /* |
6e1db88d | 1659 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1660 | * outside i_size. Trim these off again. Don't need |
1661 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1662 | * |
1663 | * Add inode to orphan list in case we crash before | |
1664 | * truncate finishes | |
ae4d5372 | 1665 | */ |
ffacfa7a | 1666 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1667 | ext4_orphan_add(handle, inode); |
1668 | ||
1669 | ext4_journal_stop(handle); | |
1670 | if (pos + len > inode->i_size) { | |
b9a4207d | 1671 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1672 | /* |
ffacfa7a | 1673 | * If truncate failed early the inode might |
1938a150 AK |
1674 | * still be on the orphan list; we need to |
1675 | * make sure the inode is removed from the | |
1676 | * orphan list in that case. | |
1677 | */ | |
1678 | if (inode->i_nlink) | |
1679 | ext4_orphan_del(NULL, inode); | |
1680 | } | |
bfc1af65 NP |
1681 | } |
1682 | ||
617ba13b | 1683 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 1684 | goto retry; |
7479d2b9 | 1685 | out: |
ac27a0ec DK |
1686 | return ret; |
1687 | } | |
1688 | ||
bfc1af65 NP |
1689 | /* For write_end() in data=journal mode */ |
1690 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
1691 | { |
1692 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
1693 | return 0; | |
1694 | set_buffer_uptodate(bh); | |
0390131b | 1695 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
1696 | } |
1697 | ||
f8514083 | 1698 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
1699 | struct address_space *mapping, |
1700 | loff_t pos, unsigned len, unsigned copied, | |
1701 | struct page *page, void *fsdata) | |
f8514083 AK |
1702 | { |
1703 | int i_size_changed = 0; | |
1704 | struct inode *inode = mapping->host; | |
1705 | handle_t *handle = ext4_journal_current_handle(); | |
1706 | ||
1707 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1708 | ||
1709 | /* | |
1710 | * No need to use i_size_read() here, the i_size | |
1711 | * cannot change under us because we hold i_mutex. | |
1712 | * | |
1713 | * But it's important to update i_size while still holding page lock: | |
1714 | * page writeout could otherwise come in and zero beyond i_size. | |
1715 | */ | |
1716 | if (pos + copied > inode->i_size) { | |
1717 | i_size_write(inode, pos + copied); | |
1718 | i_size_changed = 1; | |
1719 | } | |
1720 | ||
1721 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
1722 | /* We need to mark inode dirty even if | |
1723 | * new_i_size is less that inode->i_size | |
1724 | * bu greater than i_disksize.(hint delalloc) | |
1725 | */ | |
1726 | ext4_update_i_disksize(inode, (pos + copied)); | |
1727 | i_size_changed = 1; | |
1728 | } | |
1729 | unlock_page(page); | |
1730 | page_cache_release(page); | |
1731 | ||
1732 | /* | |
1733 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1734 | * makes the holding time of page lock longer. Second, it forces lock | |
1735 | * ordering of page lock and transaction start for journaling | |
1736 | * filesystems. | |
1737 | */ | |
1738 | if (i_size_changed) | |
1739 | ext4_mark_inode_dirty(handle, inode); | |
1740 | ||
1741 | return copied; | |
1742 | } | |
1743 | ||
ac27a0ec DK |
1744 | /* |
1745 | * We need to pick up the new inode size which generic_commit_write gave us | |
1746 | * `file' can be NULL - eg, when called from page_symlink(). | |
1747 | * | |
617ba13b | 1748 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1749 | * buffers are managed internally. |
1750 | */ | |
bfc1af65 | 1751 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1752 | struct address_space *mapping, |
1753 | loff_t pos, unsigned len, unsigned copied, | |
1754 | struct page *page, void *fsdata) | |
ac27a0ec | 1755 | { |
617ba13b | 1756 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1757 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1758 | int ret = 0, ret2; |
1759 | ||
9bffad1e | 1760 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1761 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1762 | |
1763 | if (ret == 0) { | |
f8514083 | 1764 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1765 | page, fsdata); |
f8a87d89 | 1766 | copied = ret2; |
ffacfa7a | 1767 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1768 | /* if we have allocated more blocks and copied |
1769 | * less. We will have blocks allocated outside | |
1770 | * inode->i_size. So truncate them | |
1771 | */ | |
1772 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1773 | if (ret2 < 0) |
1774 | ret = ret2; | |
ac27a0ec | 1775 | } |
617ba13b | 1776 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1777 | if (!ret) |
1778 | ret = ret2; | |
bfc1af65 | 1779 | |
f8514083 | 1780 | if (pos + len > inode->i_size) { |
b9a4207d | 1781 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1782 | /* |
ffacfa7a | 1783 | * If truncate failed early the inode might still be |
f8514083 AK |
1784 | * on the orphan list; we need to make sure the inode |
1785 | * is removed from the orphan list in that case. | |
1786 | */ | |
1787 | if (inode->i_nlink) | |
1788 | ext4_orphan_del(NULL, inode); | |
1789 | } | |
1790 | ||
1791 | ||
bfc1af65 | 1792 | return ret ? ret : copied; |
ac27a0ec DK |
1793 | } |
1794 | ||
bfc1af65 | 1795 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1796 | struct address_space *mapping, |
1797 | loff_t pos, unsigned len, unsigned copied, | |
1798 | struct page *page, void *fsdata) | |
ac27a0ec | 1799 | { |
617ba13b | 1800 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1801 | struct inode *inode = mapping->host; |
ac27a0ec | 1802 | int ret = 0, ret2; |
ac27a0ec | 1803 | |
9bffad1e | 1804 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1805 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1806 | page, fsdata); |
f8a87d89 | 1807 | copied = ret2; |
ffacfa7a | 1808 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1809 | /* if we have allocated more blocks and copied |
1810 | * less. We will have blocks allocated outside | |
1811 | * inode->i_size. So truncate them | |
1812 | */ | |
1813 | ext4_orphan_add(handle, inode); | |
1814 | ||
f8a87d89 RK |
1815 | if (ret2 < 0) |
1816 | ret = ret2; | |
ac27a0ec | 1817 | |
617ba13b | 1818 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1819 | if (!ret) |
1820 | ret = ret2; | |
bfc1af65 | 1821 | |
f8514083 | 1822 | if (pos + len > inode->i_size) { |
b9a4207d | 1823 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1824 | /* |
ffacfa7a | 1825 | * If truncate failed early the inode might still be |
f8514083 AK |
1826 | * on the orphan list; we need to make sure the inode |
1827 | * is removed from the orphan list in that case. | |
1828 | */ | |
1829 | if (inode->i_nlink) | |
1830 | ext4_orphan_del(NULL, inode); | |
1831 | } | |
1832 | ||
bfc1af65 | 1833 | return ret ? ret : copied; |
ac27a0ec DK |
1834 | } |
1835 | ||
bfc1af65 | 1836 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1837 | struct address_space *mapping, |
1838 | loff_t pos, unsigned len, unsigned copied, | |
1839 | struct page *page, void *fsdata) | |
ac27a0ec | 1840 | { |
617ba13b | 1841 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1842 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1843 | int ret = 0, ret2; |
1844 | int partial = 0; | |
bfc1af65 | 1845 | unsigned from, to; |
cf17fea6 | 1846 | loff_t new_i_size; |
ac27a0ec | 1847 | |
9bffad1e | 1848 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1849 | from = pos & (PAGE_CACHE_SIZE - 1); |
1850 | to = from + len; | |
1851 | ||
1852 | if (copied < len) { | |
1853 | if (!PageUptodate(page)) | |
1854 | copied = 0; | |
1855 | page_zero_new_buffers(page, from+copied, to); | |
1856 | } | |
ac27a0ec DK |
1857 | |
1858 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1859 | to, &partial, write_end_fn); |
ac27a0ec DK |
1860 | if (!partial) |
1861 | SetPageUptodate(page); | |
cf17fea6 AK |
1862 | new_i_size = pos + copied; |
1863 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1864 | i_size_write(inode, pos+copied); |
19f5fb7a | 1865 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
cf17fea6 AK |
1866 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1867 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1868 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1869 | if (!ret) |
1870 | ret = ret2; | |
1871 | } | |
bfc1af65 | 1872 | |
cf108bca | 1873 | unlock_page(page); |
f8514083 | 1874 | page_cache_release(page); |
ffacfa7a | 1875 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1876 | /* if we have allocated more blocks and copied |
1877 | * less. We will have blocks allocated outside | |
1878 | * inode->i_size. So truncate them | |
1879 | */ | |
1880 | ext4_orphan_add(handle, inode); | |
1881 | ||
617ba13b | 1882 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1883 | if (!ret) |
1884 | ret = ret2; | |
f8514083 | 1885 | if (pos + len > inode->i_size) { |
b9a4207d | 1886 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1887 | /* |
ffacfa7a | 1888 | * If truncate failed early the inode might still be |
f8514083 AK |
1889 | * on the orphan list; we need to make sure the inode |
1890 | * is removed from the orphan list in that case. | |
1891 | */ | |
1892 | if (inode->i_nlink) | |
1893 | ext4_orphan_del(NULL, inode); | |
1894 | } | |
bfc1af65 NP |
1895 | |
1896 | return ret ? ret : copied; | |
ac27a0ec | 1897 | } |
d2a17637 | 1898 | |
9d0be502 TT |
1899 | /* |
1900 | * Reserve a single block located at lblock | |
1901 | */ | |
01f49d0b | 1902 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1903 | { |
030ba6bc | 1904 | int retries = 0; |
60e58e0f | 1905 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1906 | struct ext4_inode_info *ei = EXT4_I(inode); |
72b8ab9d | 1907 | unsigned long md_needed; |
5dd4056d | 1908 | int ret; |
d2a17637 MC |
1909 | |
1910 | /* | |
1911 | * recalculate the amount of metadata blocks to reserve | |
1912 | * in order to allocate nrblocks | |
1913 | * worse case is one extent per block | |
1914 | */ | |
030ba6bc | 1915 | repeat: |
0637c6f4 | 1916 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1917 | md_needed = ext4_calc_metadata_amount(inode, lblock); |
f8ec9d68 | 1918 | trace_ext4_da_reserve_space(inode, md_needed); |
0637c6f4 | 1919 | spin_unlock(&ei->i_block_reservation_lock); |
d2a17637 | 1920 | |
60e58e0f | 1921 | /* |
72b8ab9d ES |
1922 | * We will charge metadata quota at writeout time; this saves |
1923 | * us from metadata over-estimation, though we may go over by | |
1924 | * a small amount in the end. Here we just reserve for data. | |
60e58e0f | 1925 | */ |
72b8ab9d | 1926 | ret = dquot_reserve_block(inode, 1); |
5dd4056d CH |
1927 | if (ret) |
1928 | return ret; | |
72b8ab9d ES |
1929 | /* |
1930 | * We do still charge estimated metadata to the sb though; | |
1931 | * we cannot afford to run out of free blocks. | |
1932 | */ | |
9d0be502 | 1933 | if (ext4_claim_free_blocks(sbi, md_needed + 1)) { |
72b8ab9d | 1934 | dquot_release_reservation_block(inode, 1); |
030ba6bc AK |
1935 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1936 | yield(); | |
1937 | goto repeat; | |
1938 | } | |
d2a17637 MC |
1939 | return -ENOSPC; |
1940 | } | |
0637c6f4 | 1941 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1942 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1943 | ei->i_reserved_meta_blocks += md_needed; |
1944 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1945 | |
d2a17637 MC |
1946 | return 0; /* success */ |
1947 | } | |
1948 | ||
12219aea | 1949 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1950 | { |
1951 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1952 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1953 | |
cd213226 MC |
1954 | if (!to_free) |
1955 | return; /* Nothing to release, exit */ | |
1956 | ||
d2a17637 | 1957 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1958 | |
5a58ec87 | 1959 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1960 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1961 | /* |
0637c6f4 TT |
1962 | * if there aren't enough reserved blocks, then the |
1963 | * counter is messed up somewhere. Since this | |
1964 | * function is called from invalidate page, it's | |
1965 | * harmless to return without any action. | |
cd213226 | 1966 | */ |
0637c6f4 TT |
1967 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1968 | "ino %lu, to_free %d with only %d reserved " | |
1969 | "data blocks\n", inode->i_ino, to_free, | |
1970 | ei->i_reserved_data_blocks); | |
1971 | WARN_ON(1); | |
1972 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1973 | } |
0637c6f4 | 1974 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1975 | |
0637c6f4 TT |
1976 | if (ei->i_reserved_data_blocks == 0) { |
1977 | /* | |
1978 | * We can release all of the reserved metadata blocks | |
1979 | * only when we have written all of the delayed | |
1980 | * allocation blocks. | |
1981 | */ | |
72b8ab9d ES |
1982 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, |
1983 | ei->i_reserved_meta_blocks); | |
ee5f4d9c | 1984 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1985 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1986 | } |
d2a17637 | 1987 | |
72b8ab9d | 1988 | /* update fs dirty data blocks counter */ |
0637c6f4 | 1989 | percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free); |
d2a17637 | 1990 | |
d2a17637 | 1991 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1992 | |
5dd4056d | 1993 | dquot_release_reservation_block(inode, to_free); |
d2a17637 MC |
1994 | } |
1995 | ||
1996 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1997 | unsigned long offset) |
d2a17637 MC |
1998 | { |
1999 | int to_release = 0; | |
2000 | struct buffer_head *head, *bh; | |
2001 | unsigned int curr_off = 0; | |
2002 | ||
2003 | head = page_buffers(page); | |
2004 | bh = head; | |
2005 | do { | |
2006 | unsigned int next_off = curr_off + bh->b_size; | |
2007 | ||
2008 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
2009 | to_release++; | |
2010 | clear_buffer_delay(bh); | |
2011 | } | |
2012 | curr_off = next_off; | |
2013 | } while ((bh = bh->b_this_page) != head); | |
12219aea | 2014 | ext4_da_release_space(page->mapping->host, to_release); |
d2a17637 | 2015 | } |
ac27a0ec | 2016 | |
64769240 AT |
2017 | /* |
2018 | * Delayed allocation stuff | |
2019 | */ | |
2020 | ||
64769240 AT |
2021 | /* |
2022 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 2023 | * them with writepage() call back |
64769240 AT |
2024 | * |
2025 | * @mpd->inode: inode | |
2026 | * @mpd->first_page: first page of the extent | |
2027 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
2028 | * |
2029 | * By the time mpage_da_submit_io() is called we expect all blocks | |
2030 | * to be allocated. this may be wrong if allocation failed. | |
2031 | * | |
2032 | * As pages are already locked by write_cache_pages(), we can't use it | |
2033 | */ | |
1de3e3df TT |
2034 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
2035 | struct ext4_map_blocks *map) | |
64769240 | 2036 | { |
791b7f08 AK |
2037 | struct pagevec pvec; |
2038 | unsigned long index, end; | |
2039 | int ret = 0, err, nr_pages, i; | |
2040 | struct inode *inode = mpd->inode; | |
2041 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 2042 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
2043 | unsigned int len, block_start; |
2044 | struct buffer_head *bh, *page_bufs = NULL; | |
cb20d518 | 2045 | int journal_data = ext4_should_journal_data(inode); |
1de3e3df | 2046 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 2047 | struct ext4_io_submit io_submit; |
64769240 AT |
2048 | |
2049 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 2050 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
2051 | /* |
2052 | * We need to start from the first_page to the next_page - 1 | |
2053 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 2054 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
2055 | * at the currently mapped buffer_heads. |
2056 | */ | |
64769240 AT |
2057 | index = mpd->first_page; |
2058 | end = mpd->next_page - 1; | |
2059 | ||
791b7f08 | 2060 | pagevec_init(&pvec, 0); |
64769240 | 2061 | while (index <= end) { |
791b7f08 | 2062 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
2063 | if (nr_pages == 0) |
2064 | break; | |
2065 | for (i = 0; i < nr_pages; i++) { | |
97498956 | 2066 | int commit_write = 0, skip_page = 0; |
64769240 AT |
2067 | struct page *page = pvec.pages[i]; |
2068 | ||
791b7f08 AK |
2069 | index = page->index; |
2070 | if (index > end) | |
2071 | break; | |
cb20d518 TT |
2072 | |
2073 | if (index == size >> PAGE_CACHE_SHIFT) | |
2074 | len = size & ~PAGE_CACHE_MASK; | |
2075 | else | |
2076 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
2077 | if (map) { |
2078 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
2079 | inode->i_blkbits); | |
2080 | pblock = map->m_pblk + (cur_logical - | |
2081 | map->m_lblk); | |
2082 | } | |
791b7f08 AK |
2083 | index++; |
2084 | ||
2085 | BUG_ON(!PageLocked(page)); | |
2086 | BUG_ON(PageWriteback(page)); | |
2087 | ||
64769240 | 2088 | /* |
cb20d518 TT |
2089 | * If the page does not have buffers (for |
2090 | * whatever reason), try to create them using | |
a107e5a3 | 2091 | * __block_write_begin. If this fails, |
97498956 | 2092 | * skip the page and move on. |
64769240 | 2093 | */ |
cb20d518 | 2094 | if (!page_has_buffers(page)) { |
a107e5a3 | 2095 | if (__block_write_begin(page, 0, len, |
cb20d518 | 2096 | noalloc_get_block_write)) { |
97498956 | 2097 | skip_page: |
cb20d518 TT |
2098 | unlock_page(page); |
2099 | continue; | |
2100 | } | |
2101 | commit_write = 1; | |
2102 | } | |
64769240 | 2103 | |
3ecdb3a1 TT |
2104 | bh = page_bufs = page_buffers(page); |
2105 | block_start = 0; | |
64769240 | 2106 | do { |
1de3e3df | 2107 | if (!bh) |
97498956 | 2108 | goto skip_page; |
1de3e3df TT |
2109 | if (map && (cur_logical >= map->m_lblk) && |
2110 | (cur_logical <= (map->m_lblk + | |
2111 | (map->m_len - 1)))) { | |
29fa89d0 AK |
2112 | if (buffer_delay(bh)) { |
2113 | clear_buffer_delay(bh); | |
2114 | bh->b_blocknr = pblock; | |
29fa89d0 | 2115 | } |
1de3e3df TT |
2116 | if (buffer_unwritten(bh) || |
2117 | buffer_mapped(bh)) | |
2118 | BUG_ON(bh->b_blocknr != pblock); | |
2119 | if (map->m_flags & EXT4_MAP_UNINIT) | |
2120 | set_buffer_uninit(bh); | |
2121 | clear_buffer_unwritten(bh); | |
2122 | } | |
29fa89d0 | 2123 | |
97498956 | 2124 | /* skip page if block allocation undone */ |
1de3e3df | 2125 | if (buffer_delay(bh) || buffer_unwritten(bh)) |
97498956 | 2126 | skip_page = 1; |
3ecdb3a1 TT |
2127 | bh = bh->b_this_page; |
2128 | block_start += bh->b_size; | |
64769240 AT |
2129 | cur_logical++; |
2130 | pblock++; | |
1de3e3df TT |
2131 | } while (bh != page_bufs); |
2132 | ||
97498956 TT |
2133 | if (skip_page) |
2134 | goto skip_page; | |
cb20d518 TT |
2135 | |
2136 | if (commit_write) | |
2137 | /* mark the buffer_heads as dirty & uptodate */ | |
2138 | block_commit_write(page, 0, len); | |
2139 | ||
97498956 | 2140 | clear_page_dirty_for_io(page); |
bd2d0210 TT |
2141 | /* |
2142 | * Delalloc doesn't support data journalling, | |
2143 | * but eventually maybe we'll lift this | |
2144 | * restriction. | |
2145 | */ | |
2146 | if (unlikely(journal_data && PageChecked(page))) | |
cb20d518 | 2147 | err = __ext4_journalled_writepage(page, len); |
1449032b | 2148 | else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT)) |
bd2d0210 TT |
2149 | err = ext4_bio_write_page(&io_submit, page, |
2150 | len, mpd->wbc); | |
1449032b TT |
2151 | else |
2152 | err = block_write_full_page(page, | |
2153 | noalloc_get_block_write, mpd->wbc); | |
cb20d518 TT |
2154 | |
2155 | if (!err) | |
a1d6cc56 | 2156 | mpd->pages_written++; |
64769240 AT |
2157 | /* |
2158 | * In error case, we have to continue because | |
2159 | * remaining pages are still locked | |
64769240 AT |
2160 | */ |
2161 | if (ret == 0) | |
2162 | ret = err; | |
64769240 AT |
2163 | } |
2164 | pagevec_release(&pvec); | |
2165 | } | |
bd2d0210 | 2166 | ext4_io_submit(&io_submit); |
64769240 | 2167 | return ret; |
64769240 AT |
2168 | } |
2169 | ||
c7f5938a | 2170 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
2171 | { |
2172 | int nr_pages, i; | |
2173 | pgoff_t index, end; | |
2174 | struct pagevec pvec; | |
2175 | struct inode *inode = mpd->inode; | |
2176 | struct address_space *mapping = inode->i_mapping; | |
2177 | ||
c7f5938a CW |
2178 | index = mpd->first_page; |
2179 | end = mpd->next_page - 1; | |
c4a0c46e AK |
2180 | while (index <= end) { |
2181 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
2182 | if (nr_pages == 0) | |
2183 | break; | |
2184 | for (i = 0; i < nr_pages; i++) { | |
2185 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 2186 | if (page->index > end) |
c4a0c46e | 2187 | break; |
c4a0c46e AK |
2188 | BUG_ON(!PageLocked(page)); |
2189 | BUG_ON(PageWriteback(page)); | |
2190 | block_invalidatepage(page, 0); | |
2191 | ClearPageUptodate(page); | |
2192 | unlock_page(page); | |
2193 | } | |
9b1d0998 JK |
2194 | index = pvec.pages[nr_pages - 1]->index + 1; |
2195 | pagevec_release(&pvec); | |
c4a0c46e AK |
2196 | } |
2197 | return; | |
2198 | } | |
2199 | ||
df22291f AK |
2200 | static void ext4_print_free_blocks(struct inode *inode) |
2201 | { | |
2202 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1693918e TT |
2203 | printk(KERN_CRIT "Total free blocks count %lld\n", |
2204 | ext4_count_free_blocks(inode->i_sb)); | |
2205 | printk(KERN_CRIT "Free/Dirty block details\n"); | |
2206 | printk(KERN_CRIT "free_blocks=%lld\n", | |
2207 | (long long) percpu_counter_sum(&sbi->s_freeblocks_counter)); | |
2208 | printk(KERN_CRIT "dirty_blocks=%lld\n", | |
2209 | (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter)); | |
2210 | printk(KERN_CRIT "Block reservation details\n"); | |
2211 | printk(KERN_CRIT "i_reserved_data_blocks=%u\n", | |
2212 | EXT4_I(inode)->i_reserved_data_blocks); | |
2213 | printk(KERN_CRIT "i_reserved_meta_blocks=%u\n", | |
2214 | EXT4_I(inode)->i_reserved_meta_blocks); | |
df22291f AK |
2215 | return; |
2216 | } | |
2217 | ||
64769240 | 2218 | /* |
5a87b7a5 TT |
2219 | * mpage_da_map_and_submit - go through given space, map them |
2220 | * if necessary, and then submit them for I/O | |
64769240 | 2221 | * |
8dc207c0 | 2222 | * @mpd - bh describing space |
64769240 AT |
2223 | * |
2224 | * The function skips space we know is already mapped to disk blocks. | |
2225 | * | |
64769240 | 2226 | */ |
5a87b7a5 | 2227 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 2228 | { |
2ac3b6e0 | 2229 | int err, blks, get_blocks_flags; |
1de3e3df | 2230 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
2231 | sector_t next = mpd->b_blocknr; |
2232 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
2233 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
2234 | handle_t *handle = NULL; | |
64769240 AT |
2235 | |
2236 | /* | |
5a87b7a5 TT |
2237 | * If the blocks are mapped already, or we couldn't accumulate |
2238 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 2239 | */ |
5a87b7a5 TT |
2240 | if ((mpd->b_size == 0) || |
2241 | ((mpd->b_state & (1 << BH_Mapped)) && | |
2242 | !(mpd->b_state & (1 << BH_Delay)) && | |
2243 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
2244 | goto submit_io; | |
2fa3cdfb TT |
2245 | |
2246 | handle = ext4_journal_current_handle(); | |
2247 | BUG_ON(!handle); | |
2248 | ||
79ffab34 | 2249 | /* |
79e83036 | 2250 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
2251 | * blocks, or to convert an uninitialized extent to be |
2252 | * initialized (in the case where we have written into | |
2253 | * one or more preallocated blocks). | |
2254 | * | |
2255 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
2256 | * indicate that we are on the delayed allocation path. This | |
2257 | * affects functions in many different parts of the allocation | |
2258 | * call path. This flag exists primarily because we don't | |
79e83036 | 2259 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 2260 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
2261 | * inode's allocation semaphore is taken. |
2262 | * | |
2263 | * If the blocks in questions were delalloc blocks, set | |
2264 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
2265 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 2266 | */ |
2ed88685 TT |
2267 | map.m_lblk = next; |
2268 | map.m_len = max_blocks; | |
1296cc85 | 2269 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
2270 | if (ext4_should_dioread_nolock(mpd->inode)) |
2271 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 2272 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
2273 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2274 | ||
2ed88685 | 2275 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 2276 | if (blks < 0) { |
e3570639 ES |
2277 | struct super_block *sb = mpd->inode->i_sb; |
2278 | ||
2fa3cdfb | 2279 | err = blks; |
ed5bde0b | 2280 | /* |
5a87b7a5 | 2281 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
2282 | * appears to be free blocks we will just let |
2283 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
2284 | */ |
2285 | if (err == -EAGAIN) | |
5a87b7a5 | 2286 | goto submit_io; |
df22291f AK |
2287 | |
2288 | if (err == -ENOSPC && | |
e3570639 | 2289 | ext4_count_free_blocks(sb)) { |
df22291f | 2290 | mpd->retval = err; |
5a87b7a5 | 2291 | goto submit_io; |
df22291f AK |
2292 | } |
2293 | ||
c4a0c46e | 2294 | /* |
ed5bde0b TT |
2295 | * get block failure will cause us to loop in |
2296 | * writepages, because a_ops->writepage won't be able | |
2297 | * to make progress. The page will be redirtied by | |
2298 | * writepage and writepages will again try to write | |
2299 | * the same. | |
c4a0c46e | 2300 | */ |
e3570639 ES |
2301 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2302 | ext4_msg(sb, KERN_CRIT, | |
2303 | "delayed block allocation failed for inode %lu " | |
2304 | "at logical offset %llu with max blocks %zd " | |
2305 | "with error %d", mpd->inode->i_ino, | |
2306 | (unsigned long long) next, | |
2307 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
2308 | ext4_msg(sb, KERN_CRIT, | |
2309 | "This should not happen!! Data will be lost\n"); | |
2310 | if (err == -ENOSPC) | |
2311 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 2312 | } |
2fa3cdfb | 2313 | /* invalidate all the pages */ |
c7f5938a | 2314 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
2315 | |
2316 | /* Mark this page range as having been completed */ | |
2317 | mpd->io_done = 1; | |
5a87b7a5 | 2318 | return; |
c4a0c46e | 2319 | } |
2fa3cdfb TT |
2320 | BUG_ON(blks == 0); |
2321 | ||
1de3e3df | 2322 | mapp = ↦ |
2ed88685 TT |
2323 | if (map.m_flags & EXT4_MAP_NEW) { |
2324 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
2325 | int i; | |
64769240 | 2326 | |
2ed88685 TT |
2327 | for (i = 0; i < map.m_len; i++) |
2328 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2329 | } | |
64769240 | 2330 | |
2fa3cdfb TT |
2331 | if (ext4_should_order_data(mpd->inode)) { |
2332 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
2333 | if (err) | |
5a87b7a5 TT |
2334 | /* This only happens if the journal is aborted */ |
2335 | return; | |
2fa3cdfb TT |
2336 | } |
2337 | ||
2338 | /* | |
03f5d8bc | 2339 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
2340 | */ |
2341 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
2342 | if (disksize > i_size_read(mpd->inode)) | |
2343 | disksize = i_size_read(mpd->inode); | |
2344 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
2345 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
2346 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
2347 | if (err) | |
2348 | ext4_error(mpd->inode->i_sb, | |
2349 | "Failed to mark inode %lu dirty", | |
2350 | mpd->inode->i_ino); | |
2fa3cdfb TT |
2351 | } |
2352 | ||
5a87b7a5 | 2353 | submit_io: |
1de3e3df | 2354 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 2355 | mpd->io_done = 1; |
64769240 AT |
2356 | } |
2357 | ||
bf068ee2 AK |
2358 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
2359 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
2360 | |
2361 | /* | |
2362 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
2363 | * | |
2364 | * @mpd->lbh - extent of blocks | |
2365 | * @logical - logical number of the block in the file | |
2366 | * @bh - bh of the block (used to access block's state) | |
2367 | * | |
2368 | * the function is used to collect contig. blocks in same state | |
2369 | */ | |
2370 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
2371 | sector_t logical, size_t b_size, |
2372 | unsigned long b_state) | |
64769240 | 2373 | { |
64769240 | 2374 | sector_t next; |
8dc207c0 | 2375 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 2376 | |
c445e3e0 ES |
2377 | /* |
2378 | * XXX Don't go larger than mballoc is willing to allocate | |
2379 | * This is a stopgap solution. We eventually need to fold | |
2380 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 2381 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
2382 | */ |
2383 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
2384 | goto flush_it; | |
2385 | ||
525f4ed8 | 2386 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 2387 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
2388 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
2389 | /* | |
2390 | * With non-extent format we are limited by the journal | |
2391 | * credit available. Total credit needed to insert | |
2392 | * nrblocks contiguous blocks is dependent on the | |
2393 | * nrblocks. So limit nrblocks. | |
2394 | */ | |
2395 | goto flush_it; | |
2396 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
2397 | EXT4_MAX_TRANS_DATA) { | |
2398 | /* | |
2399 | * Adding the new buffer_head would make it cross the | |
2400 | * allowed limit for which we have journal credit | |
2401 | * reserved. So limit the new bh->b_size | |
2402 | */ | |
2403 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
2404 | mpd->inode->i_blkbits; | |
2405 | /* we will do mpage_da_submit_io in the next loop */ | |
2406 | } | |
2407 | } | |
64769240 AT |
2408 | /* |
2409 | * First block in the extent | |
2410 | */ | |
8dc207c0 TT |
2411 | if (mpd->b_size == 0) { |
2412 | mpd->b_blocknr = logical; | |
2413 | mpd->b_size = b_size; | |
2414 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
2415 | return; |
2416 | } | |
2417 | ||
8dc207c0 | 2418 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
2419 | /* |
2420 | * Can we merge the block to our big extent? | |
2421 | */ | |
8dc207c0 TT |
2422 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
2423 | mpd->b_size += b_size; | |
64769240 AT |
2424 | return; |
2425 | } | |
2426 | ||
525f4ed8 | 2427 | flush_it: |
64769240 AT |
2428 | /* |
2429 | * We couldn't merge the block to our extent, so we | |
2430 | * need to flush current extent and start new one | |
2431 | */ | |
5a87b7a5 | 2432 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 2433 | return; |
64769240 AT |
2434 | } |
2435 | ||
c364b22c | 2436 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 2437 | { |
c364b22c | 2438 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
2439 | } |
2440 | ||
64769240 | 2441 | /* |
b920c755 TT |
2442 | * This is a special get_blocks_t callback which is used by |
2443 | * ext4_da_write_begin(). It will either return mapped block or | |
2444 | * reserve space for a single block. | |
29fa89d0 AK |
2445 | * |
2446 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2447 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2448 | * | |
2449 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2450 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2451 | * initialized properly. | |
64769240 AT |
2452 | */ |
2453 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 2454 | struct buffer_head *bh, int create) |
64769240 | 2455 | { |
2ed88685 | 2456 | struct ext4_map_blocks map; |
64769240 | 2457 | int ret = 0; |
33b9817e AK |
2458 | sector_t invalid_block = ~((sector_t) 0xffff); |
2459 | ||
2460 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
2461 | invalid_block = ~0; | |
64769240 AT |
2462 | |
2463 | BUG_ON(create == 0); | |
2ed88685 TT |
2464 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2465 | ||
2466 | map.m_lblk = iblock; | |
2467 | map.m_len = 1; | |
64769240 AT |
2468 | |
2469 | /* | |
2470 | * first, we need to know whether the block is allocated already | |
2471 | * preallocated blocks are unmapped but should treated | |
2472 | * the same as allocated blocks. | |
2473 | */ | |
2ed88685 TT |
2474 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
2475 | if (ret < 0) | |
2476 | return ret; | |
2477 | if (ret == 0) { | |
2478 | if (buffer_delay(bh)) | |
2479 | return 0; /* Not sure this could or should happen */ | |
64769240 | 2480 | /* |
ebdec241 | 2481 | * XXX: __block_write_begin() unmaps passed block, is it OK? |
64769240 | 2482 | */ |
9d0be502 | 2483 | ret = ext4_da_reserve_space(inode, iblock); |
d2a17637 MC |
2484 | if (ret) |
2485 | /* not enough space to reserve */ | |
2486 | return ret; | |
2487 | ||
2ed88685 TT |
2488 | map_bh(bh, inode->i_sb, invalid_block); |
2489 | set_buffer_new(bh); | |
2490 | set_buffer_delay(bh); | |
2491 | return 0; | |
64769240 AT |
2492 | } |
2493 | ||
2ed88685 TT |
2494 | map_bh(bh, inode->i_sb, map.m_pblk); |
2495 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2496 | ||
2497 | if (buffer_unwritten(bh)) { | |
2498 | /* A delayed write to unwritten bh should be marked | |
2499 | * new and mapped. Mapped ensures that we don't do | |
2500 | * get_block multiple times when we write to the same | |
2501 | * offset and new ensures that we do proper zero out | |
2502 | * for partial write. | |
2503 | */ | |
2504 | set_buffer_new(bh); | |
c8205636 | 2505 | set_buffer_mapped(bh); |
2ed88685 TT |
2506 | } |
2507 | return 0; | |
64769240 | 2508 | } |
61628a3f | 2509 | |
b920c755 TT |
2510 | /* |
2511 | * This function is used as a standard get_block_t calback function | |
2512 | * when there is no desire to allocate any blocks. It is used as a | |
ebdec241 | 2513 | * callback function for block_write_begin() and block_write_full_page(). |
206f7ab4 | 2514 | * These functions should only try to map a single block at a time. |
b920c755 TT |
2515 | * |
2516 | * Since this function doesn't do block allocations even if the caller | |
2517 | * requests it by passing in create=1, it is critically important that | |
2518 | * any caller checks to make sure that any buffer heads are returned | |
2519 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
2520 | * delayed allocation before calling block_write_full_page(). Otherwise, |
2521 | * b_blocknr could be left unitialized, and the page write functions will | |
2522 | * be taken by surprise. | |
b920c755 TT |
2523 | */ |
2524 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
2525 | struct buffer_head *bh_result, int create) |
2526 | { | |
a2dc52b5 | 2527 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 2528 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
2529 | } |
2530 | ||
62e086be AK |
2531 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2532 | { | |
2533 | get_bh(bh); | |
2534 | return 0; | |
2535 | } | |
2536 | ||
2537 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2538 | { | |
2539 | put_bh(bh); | |
2540 | return 0; | |
2541 | } | |
2542 | ||
2543 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2544 | unsigned int len) |
2545 | { | |
2546 | struct address_space *mapping = page->mapping; | |
2547 | struct inode *inode = mapping->host; | |
2548 | struct buffer_head *page_bufs; | |
2549 | handle_t *handle = NULL; | |
2550 | int ret = 0; | |
2551 | int err; | |
2552 | ||
cb20d518 | 2553 | ClearPageChecked(page); |
62e086be AK |
2554 | page_bufs = page_buffers(page); |
2555 | BUG_ON(!page_bufs); | |
2556 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
2557 | /* As soon as we unlock the page, it can go away, but we have | |
2558 | * references to buffers so we are safe */ | |
2559 | unlock_page(page); | |
2560 | ||
2561 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
2562 | if (IS_ERR(handle)) { | |
2563 | ret = PTR_ERR(handle); | |
2564 | goto out; | |
2565 | } | |
2566 | ||
2567 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2568 | do_journal_get_write_access); | |
2569 | ||
2570 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2571 | write_end_fn); | |
2572 | if (ret == 0) | |
2573 | ret = err; | |
2574 | err = ext4_journal_stop(handle); | |
2575 | if (!ret) | |
2576 | ret = err; | |
2577 | ||
2578 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 2579 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
2580 | out: |
2581 | return ret; | |
2582 | } | |
2583 | ||
744692dc JZ |
2584 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); |
2585 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
2586 | ||
61628a3f | 2587 | /* |
43ce1d23 AK |
2588 | * Note that we don't need to start a transaction unless we're journaling data |
2589 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2590 | * need to file the inode to the transaction's list in ordered mode because if | |
2591 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2592 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2593 | * transaction the data will hit the disk. In case we are journaling data, we |
2594 | * cannot start transaction directly because transaction start ranks above page | |
2595 | * lock so we have to do some magic. | |
2596 | * | |
b920c755 TT |
2597 | * This function can get called via... |
2598 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2599 | * - journal_submit_inode_data_buffers (no journal handle) | |
2600 | * - shrink_page_list via pdflush (no journal handle) | |
2601 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
2602 | * |
2603 | * We don't do any block allocation in this function. If we have page with | |
2604 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2605 | * is important for mmaped based write. So if we do with blocksize 1K | |
2606 | * truncate(f, 1024); | |
2607 | * a = mmap(f, 0, 4096); | |
2608 | * a[0] = 'a'; | |
2609 | * truncate(f, 4096); | |
2610 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
2611 | * but other bufer_heads would be unmapped but dirty(dirty done via the | |
2612 | * do_wp_page). So writepage should write the first block. If we modify | |
2613 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2614 | * page_mkwrite callback will do the block allocation and mark the | |
2615 | * buffer_heads mapped. | |
2616 | * | |
2617 | * We redirty the page if we have any buffer_heads that is either delay or | |
2618 | * unwritten in the page. | |
2619 | * | |
2620 | * We can get recursively called as show below. | |
2621 | * | |
2622 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2623 | * ext4_writepage() | |
2624 | * | |
2625 | * But since we don't do any block allocation we should not deadlock. | |
2626 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2627 | */ |
43ce1d23 | 2628 | static int ext4_writepage(struct page *page, |
62e086be | 2629 | struct writeback_control *wbc) |
64769240 | 2630 | { |
a42afc5f | 2631 | int ret = 0, commit_write = 0; |
61628a3f | 2632 | loff_t size; |
498e5f24 | 2633 | unsigned int len; |
744692dc | 2634 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
2635 | struct inode *inode = page->mapping->host; |
2636 | ||
43ce1d23 | 2637 | trace_ext4_writepage(inode, page); |
f0e6c985 AK |
2638 | size = i_size_read(inode); |
2639 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2640 | len = size & ~PAGE_CACHE_MASK; | |
2641 | else | |
2642 | len = PAGE_CACHE_SIZE; | |
64769240 | 2643 | |
a42afc5f TT |
2644 | /* |
2645 | * If the page does not have buffers (for whatever reason), | |
a107e5a3 | 2646 | * try to create them using __block_write_begin. If this |
a42afc5f TT |
2647 | * fails, redirty the page and move on. |
2648 | */ | |
b1142e8f | 2649 | if (!page_has_buffers(page)) { |
a107e5a3 | 2650 | if (__block_write_begin(page, 0, len, |
a42afc5f TT |
2651 | noalloc_get_block_write)) { |
2652 | redirty_page: | |
f0e6c985 AK |
2653 | redirty_page_for_writepage(wbc, page); |
2654 | unlock_page(page); | |
2655 | return 0; | |
2656 | } | |
a42afc5f TT |
2657 | commit_write = 1; |
2658 | } | |
2659 | page_bufs = page_buffers(page); | |
2660 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2661 | ext4_bh_delay_or_unwritten)) { | |
f0e6c985 | 2662 | /* |
b1142e8f TT |
2663 | * We don't want to do block allocation, so redirty |
2664 | * the page and return. We may reach here when we do | |
2665 | * a journal commit via journal_submit_inode_data_buffers. | |
2666 | * We can also reach here via shrink_page_list | |
f0e6c985 | 2667 | */ |
a42afc5f TT |
2668 | goto redirty_page; |
2669 | } | |
2670 | if (commit_write) | |
ed9b3e33 | 2671 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 2672 | block_commit_write(page, 0, len); |
64769240 | 2673 | |
cb20d518 | 2674 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2675 | /* |
2676 | * It's mmapped pagecache. Add buffers and journal it. There | |
2677 | * doesn't seem much point in redirtying the page here. | |
2678 | */ | |
3f0ca309 | 2679 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2680 | |
a42afc5f | 2681 | if (buffer_uninit(page_bufs)) { |
744692dc JZ |
2682 | ext4_set_bh_endio(page_bufs, inode); |
2683 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
2684 | wbc, ext4_end_io_buffer_write); | |
2685 | } else | |
b920c755 TT |
2686 | ret = block_write_full_page(page, noalloc_get_block_write, |
2687 | wbc); | |
64769240 | 2688 | |
64769240 AT |
2689 | return ret; |
2690 | } | |
2691 | ||
61628a3f | 2692 | /* |
525f4ed8 | 2693 | * This is called via ext4_da_writepages() to |
25985edc | 2694 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2695 | * a single extent allocation into a single transaction, |
2696 | * ext4_da_writpeages() will loop calling this before | |
2697 | * the block allocation. | |
61628a3f | 2698 | */ |
525f4ed8 MC |
2699 | |
2700 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2701 | { | |
2702 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2703 | ||
2704 | /* | |
2705 | * With non-extent format the journal credit needed to | |
2706 | * insert nrblocks contiguous block is dependent on | |
2707 | * number of contiguous block. So we will limit | |
2708 | * number of contiguous block to a sane value | |
2709 | */ | |
12e9b892 | 2710 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2711 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2712 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2713 | ||
2714 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2715 | } | |
61628a3f | 2716 | |
8e48dcfb TT |
2717 | /* |
2718 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2719 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2720 | * mpage_da_map_and_submit to map a single contiguous memory region |
2721 | * and then write them. | |
8e48dcfb TT |
2722 | */ |
2723 | static int write_cache_pages_da(struct address_space *mapping, | |
2724 | struct writeback_control *wbc, | |
72f84e65 ES |
2725 | struct mpage_da_data *mpd, |
2726 | pgoff_t *done_index) | |
8e48dcfb | 2727 | { |
4f01b02c | 2728 | struct buffer_head *bh, *head; |
168fc022 | 2729 | struct inode *inode = mapping->host; |
4f01b02c TT |
2730 | struct pagevec pvec; |
2731 | unsigned int nr_pages; | |
2732 | sector_t logical; | |
2733 | pgoff_t index, end; | |
2734 | long nr_to_write = wbc->nr_to_write; | |
2735 | int i, tag, ret = 0; | |
8e48dcfb | 2736 | |
168fc022 TT |
2737 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2738 | mpd->wbc = wbc; | |
2739 | mpd->inode = inode; | |
8e48dcfb TT |
2740 | pagevec_init(&pvec, 0); |
2741 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2742 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2743 | ||
5b41d924 ES |
2744 | if (wbc->sync_mode == WB_SYNC_ALL) |
2745 | tag = PAGECACHE_TAG_TOWRITE; | |
2746 | else | |
2747 | tag = PAGECACHE_TAG_DIRTY; | |
2748 | ||
72f84e65 | 2749 | *done_index = index; |
4f01b02c | 2750 | while (index <= end) { |
5b41d924 | 2751 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2752 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2753 | if (nr_pages == 0) | |
4f01b02c | 2754 | return 0; |
8e48dcfb TT |
2755 | |
2756 | for (i = 0; i < nr_pages; i++) { | |
2757 | struct page *page = pvec.pages[i]; | |
2758 | ||
2759 | /* | |
2760 | * At this point, the page may be truncated or | |
2761 | * invalidated (changing page->mapping to NULL), or | |
2762 | * even swizzled back from swapper_space to tmpfs file | |
2763 | * mapping. However, page->index will not change | |
2764 | * because we have a reference on the page. | |
2765 | */ | |
4f01b02c TT |
2766 | if (page->index > end) |
2767 | goto out; | |
8e48dcfb | 2768 | |
72f84e65 ES |
2769 | *done_index = page->index + 1; |
2770 | ||
78aaced3 TT |
2771 | /* |
2772 | * If we can't merge this page, and we have | |
2773 | * accumulated an contiguous region, write it | |
2774 | */ | |
2775 | if ((mpd->next_page != page->index) && | |
2776 | (mpd->next_page != mpd->first_page)) { | |
2777 | mpage_da_map_and_submit(mpd); | |
2778 | goto ret_extent_tail; | |
2779 | } | |
2780 | ||
8e48dcfb TT |
2781 | lock_page(page); |
2782 | ||
2783 | /* | |
4f01b02c TT |
2784 | * If the page is no longer dirty, or its |
2785 | * mapping no longer corresponds to inode we | |
2786 | * are writing (which means it has been | |
2787 | * truncated or invalidated), or the page is | |
2788 | * already under writeback and we are not | |
2789 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2790 | */ |
4f01b02c TT |
2791 | if (!PageDirty(page) || |
2792 | (PageWriteback(page) && | |
2793 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2794 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2795 | unlock_page(page); |
2796 | continue; | |
2797 | } | |
2798 | ||
4f01b02c TT |
2799 | if (PageWriteback(page)) |
2800 | wait_on_page_writeback(page); | |
8e48dcfb TT |
2801 | |
2802 | BUG_ON(PageWriteback(page)); | |
8e48dcfb | 2803 | |
168fc022 | 2804 | if (mpd->next_page != page->index) |
8eb9e5ce | 2805 | mpd->first_page = page->index; |
8eb9e5ce TT |
2806 | mpd->next_page = page->index + 1; |
2807 | logical = (sector_t) page->index << | |
2808 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2809 | ||
2810 | if (!page_has_buffers(page)) { | |
4f01b02c TT |
2811 | mpage_add_bh_to_extent(mpd, logical, |
2812 | PAGE_CACHE_SIZE, | |
8eb9e5ce | 2813 | (1 << BH_Dirty) | (1 << BH_Uptodate)); |
4f01b02c TT |
2814 | if (mpd->io_done) |
2815 | goto ret_extent_tail; | |
8eb9e5ce TT |
2816 | } else { |
2817 | /* | |
4f01b02c TT |
2818 | * Page with regular buffer heads, |
2819 | * just add all dirty ones | |
8eb9e5ce TT |
2820 | */ |
2821 | head = page_buffers(page); | |
2822 | bh = head; | |
2823 | do { | |
2824 | BUG_ON(buffer_locked(bh)); | |
2825 | /* | |
2826 | * We need to try to allocate | |
2827 | * unmapped blocks in the same page. | |
2828 | * Otherwise we won't make progress | |
2829 | * with the page in ext4_writepage | |
2830 | */ | |
2831 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | |
2832 | mpage_add_bh_to_extent(mpd, logical, | |
2833 | bh->b_size, | |
2834 | bh->b_state); | |
4f01b02c TT |
2835 | if (mpd->io_done) |
2836 | goto ret_extent_tail; | |
8eb9e5ce TT |
2837 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2838 | /* | |
4f01b02c TT |
2839 | * mapped dirty buffer. We need |
2840 | * to update the b_state | |
2841 | * because we look at b_state | |
2842 | * in mpage_da_map_blocks. We | |
2843 | * don't update b_size because | |
2844 | * if we find an unmapped | |
2845 | * buffer_head later we need to | |
2846 | * use the b_state flag of that | |
2847 | * buffer_head. | |
8eb9e5ce TT |
2848 | */ |
2849 | if (mpd->b_size == 0) | |
2850 | mpd->b_state = bh->b_state & BH_FLAGS; | |
2851 | } | |
2852 | logical++; | |
2853 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2854 | } |
2855 | ||
2856 | if (nr_to_write > 0) { | |
2857 | nr_to_write--; | |
2858 | if (nr_to_write == 0 && | |
4f01b02c | 2859 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2860 | /* |
2861 | * We stop writing back only if we are | |
2862 | * not doing integrity sync. In case of | |
2863 | * integrity sync we have to keep going | |
2864 | * because someone may be concurrently | |
2865 | * dirtying pages, and we might have | |
2866 | * synced a lot of newly appeared dirty | |
2867 | * pages, but have not synced all of the | |
2868 | * old dirty pages. | |
2869 | */ | |
4f01b02c | 2870 | goto out; |
8e48dcfb TT |
2871 | } |
2872 | } | |
2873 | pagevec_release(&pvec); | |
2874 | cond_resched(); | |
2875 | } | |
4f01b02c TT |
2876 | return 0; |
2877 | ret_extent_tail: | |
2878 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2879 | out: |
2880 | pagevec_release(&pvec); | |
2881 | cond_resched(); | |
8e48dcfb TT |
2882 | return ret; |
2883 | } | |
2884 | ||
2885 | ||
64769240 | 2886 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2887 | struct writeback_control *wbc) |
64769240 | 2888 | { |
22208ded AK |
2889 | pgoff_t index; |
2890 | int range_whole = 0; | |
61628a3f | 2891 | handle_t *handle = NULL; |
df22291f | 2892 | struct mpage_da_data mpd; |
5e745b04 | 2893 | struct inode *inode = mapping->host; |
498e5f24 | 2894 | int pages_written = 0; |
55138e0b | 2895 | unsigned int max_pages; |
2acf2c26 | 2896 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2897 | int needed_blocks, ret = 0; |
2898 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2899 | loff_t range_start = wbc->range_start; |
5e745b04 | 2900 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2901 | pgoff_t done_index = 0; |
5b41d924 | 2902 | pgoff_t end; |
61628a3f | 2903 | |
9bffad1e | 2904 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2905 | |
61628a3f MC |
2906 | /* |
2907 | * No pages to write? This is mainly a kludge to avoid starting | |
2908 | * a transaction for special inodes like journal inode on last iput() | |
2909 | * because that could violate lock ordering on umount | |
2910 | */ | |
a1d6cc56 | 2911 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2912 | return 0; |
2a21e37e TT |
2913 | |
2914 | /* | |
2915 | * If the filesystem has aborted, it is read-only, so return | |
2916 | * right away instead of dumping stack traces later on that | |
2917 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2918 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2919 | * the latter could be true if the filesystem is mounted |
2920 | * read-only, and in that case, ext4_da_writepages should | |
2921 | * *never* be called, so if that ever happens, we would want | |
2922 | * the stack trace. | |
2923 | */ | |
4ab2f15b | 2924 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2925 | return -EROFS; |
2926 | ||
22208ded AK |
2927 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2928 | range_whole = 1; | |
61628a3f | 2929 | |
2acf2c26 AK |
2930 | range_cyclic = wbc->range_cyclic; |
2931 | if (wbc->range_cyclic) { | |
22208ded | 2932 | index = mapping->writeback_index; |
2acf2c26 AK |
2933 | if (index) |
2934 | cycled = 0; | |
2935 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2936 | wbc->range_end = LLONG_MAX; | |
2937 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2938 | end = -1; |
2939 | } else { | |
22208ded | 2940 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2941 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2942 | } | |
a1d6cc56 | 2943 | |
55138e0b TT |
2944 | /* |
2945 | * This works around two forms of stupidity. The first is in | |
2946 | * the writeback code, which caps the maximum number of pages | |
2947 | * written to be 1024 pages. This is wrong on multiple | |
2948 | * levels; different architectues have a different page size, | |
2949 | * which changes the maximum amount of data which gets | |
2950 | * written. Secondly, 4 megabytes is way too small. XFS | |
2951 | * forces this value to be 16 megabytes by multiplying | |
2952 | * nr_to_write parameter by four, and then relies on its | |
2953 | * allocator to allocate larger extents to make them | |
2954 | * contiguous. Unfortunately this brings us to the second | |
2955 | * stupidity, which is that ext4's mballoc code only allocates | |
2956 | * at most 2048 blocks. So we force contiguous writes up to | |
2957 | * the number of dirty blocks in the inode, or | |
2958 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2959 | */ | |
2960 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2961 | if (!range_cyclic && range_whole) { |
2962 | if (wbc->nr_to_write == LONG_MAX) | |
2963 | desired_nr_to_write = wbc->nr_to_write; | |
2964 | else | |
2965 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2966 | } else | |
55138e0b TT |
2967 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2968 | max_pages); | |
2969 | if (desired_nr_to_write > max_pages) | |
2970 | desired_nr_to_write = max_pages; | |
2971 | ||
2972 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2973 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2974 | wbc->nr_to_write = desired_nr_to_write; | |
2975 | } | |
2976 | ||
2acf2c26 | 2977 | retry: |
5b41d924 ES |
2978 | if (wbc->sync_mode == WB_SYNC_ALL) |
2979 | tag_pages_for_writeback(mapping, index, end); | |
2980 | ||
22208ded | 2981 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2982 | |
2983 | /* | |
2984 | * we insert one extent at a time. So we need | |
2985 | * credit needed for single extent allocation. | |
2986 | * journalled mode is currently not supported | |
2987 | * by delalloc | |
2988 | */ | |
2989 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2990 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2991 | |
61628a3f MC |
2992 | /* start a new transaction*/ |
2993 | handle = ext4_journal_start(inode, needed_blocks); | |
2994 | if (IS_ERR(handle)) { | |
2995 | ret = PTR_ERR(handle); | |
1693918e | 2996 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2997 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2998 | wbc->nr_to_write, inode->i_ino, ret); |
61628a3f MC |
2999 | goto out_writepages; |
3000 | } | |
f63e6005 TT |
3001 | |
3002 | /* | |
8eb9e5ce | 3003 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 3004 | * contiguous region of logical blocks that need |
8eb9e5ce | 3005 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 3006 | */ |
72f84e65 | 3007 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); |
f63e6005 | 3008 | /* |
af901ca1 | 3009 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
3010 | * haven't done the I/O yet, map the blocks and submit |
3011 | * them for I/O. | |
3012 | */ | |
3013 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 3014 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
3015 | ret = MPAGE_DA_EXTENT_TAIL; |
3016 | } | |
b3a3ca8c | 3017 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 3018 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 3019 | |
61628a3f | 3020 | ext4_journal_stop(handle); |
df22291f | 3021 | |
8f64b32e | 3022 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
3023 | /* commit the transaction which would |
3024 | * free blocks released in the transaction | |
3025 | * and try again | |
3026 | */ | |
df22291f | 3027 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
3028 | ret = 0; |
3029 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 AK |
3030 | /* |
3031 | * got one extent now try with | |
3032 | * rest of the pages | |
3033 | */ | |
22208ded | 3034 | pages_written += mpd.pages_written; |
a1d6cc56 | 3035 | ret = 0; |
2acf2c26 | 3036 | io_done = 1; |
22208ded | 3037 | } else if (wbc->nr_to_write) |
61628a3f MC |
3038 | /* |
3039 | * There is no more writeout needed | |
3040 | * or we requested for a noblocking writeout | |
3041 | * and we found the device congested | |
3042 | */ | |
61628a3f | 3043 | break; |
a1d6cc56 | 3044 | } |
2acf2c26 AK |
3045 | if (!io_done && !cycled) { |
3046 | cycled = 1; | |
3047 | index = 0; | |
3048 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
3049 | wbc->range_end = mapping->writeback_index - 1; | |
3050 | goto retry; | |
3051 | } | |
22208ded AK |
3052 | |
3053 | /* Update index */ | |
2acf2c26 | 3054 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
3055 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
3056 | /* | |
3057 | * set the writeback_index so that range_cyclic | |
3058 | * mode will write it back later | |
3059 | */ | |
72f84e65 | 3060 | mapping->writeback_index = done_index; |
a1d6cc56 | 3061 | |
61628a3f | 3062 | out_writepages: |
2faf2e19 | 3063 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 3064 | wbc->range_start = range_start; |
9bffad1e | 3065 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 3066 | return ret; |
64769240 AT |
3067 | } |
3068 | ||
79f0be8d AK |
3069 | #define FALL_BACK_TO_NONDELALLOC 1 |
3070 | static int ext4_nonda_switch(struct super_block *sb) | |
3071 | { | |
3072 | s64 free_blocks, dirty_blocks; | |
3073 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
3074 | ||
3075 | /* | |
3076 | * switch to non delalloc mode if we are running low | |
3077 | * on free block. The free block accounting via percpu | |
179f7ebf | 3078 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
3079 | * accumulated on each CPU without updating global counters |
3080 | * Delalloc need an accurate free block accounting. So switch | |
3081 | * to non delalloc when we are near to error range. | |
3082 | */ | |
3083 | free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3084 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter); | |
3085 | if (2 * free_blocks < 3 * dirty_blocks || | |
3086 | free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) { | |
3087 | /* | |
c8afb446 ES |
3088 | * free block count is less than 150% of dirty blocks |
3089 | * or free blocks is less than watermark | |
79f0be8d AK |
3090 | */ |
3091 | return 1; | |
3092 | } | |
c8afb446 ES |
3093 | /* |
3094 | * Even if we don't switch but are nearing capacity, | |
3095 | * start pushing delalloc when 1/2 of free blocks are dirty. | |
3096 | */ | |
3097 | if (free_blocks < 2 * dirty_blocks) | |
3098 | writeback_inodes_sb_if_idle(sb); | |
3099 | ||
79f0be8d AK |
3100 | return 0; |
3101 | } | |
3102 | ||
64769240 | 3103 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
3104 | loff_t pos, unsigned len, unsigned flags, |
3105 | struct page **pagep, void **fsdata) | |
64769240 | 3106 | { |
72b8ab9d | 3107 | int ret, retries = 0; |
64769240 AT |
3108 | struct page *page; |
3109 | pgoff_t index; | |
64769240 AT |
3110 | struct inode *inode = mapping->host; |
3111 | handle_t *handle; | |
3112 | ||
3113 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
3114 | |
3115 | if (ext4_nonda_switch(inode->i_sb)) { | |
3116 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
3117 | return ext4_write_begin(file, mapping, pos, | |
3118 | len, flags, pagep, fsdata); | |
3119 | } | |
3120 | *fsdata = (void *)0; | |
9bffad1e | 3121 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 3122 | retry: |
64769240 AT |
3123 | /* |
3124 | * With delayed allocation, we don't log the i_disksize update | |
3125 | * if there is delayed block allocation. But we still need | |
3126 | * to journalling the i_disksize update if writes to the end | |
3127 | * of file which has an already mapped buffer. | |
3128 | */ | |
3129 | handle = ext4_journal_start(inode, 1); | |
3130 | if (IS_ERR(handle)) { | |
3131 | ret = PTR_ERR(handle); | |
3132 | goto out; | |
3133 | } | |
ebd3610b JK |
3134 | /* We cannot recurse into the filesystem as the transaction is already |
3135 | * started */ | |
3136 | flags |= AOP_FLAG_NOFS; | |
64769240 | 3137 | |
54566b2c | 3138 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
3139 | if (!page) { |
3140 | ext4_journal_stop(handle); | |
3141 | ret = -ENOMEM; | |
3142 | goto out; | |
3143 | } | |
64769240 AT |
3144 | *pagep = page; |
3145 | ||
6e1db88d | 3146 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
3147 | if (ret < 0) { |
3148 | unlock_page(page); | |
3149 | ext4_journal_stop(handle); | |
3150 | page_cache_release(page); | |
ae4d5372 AK |
3151 | /* |
3152 | * block_write_begin may have instantiated a few blocks | |
3153 | * outside i_size. Trim these off again. Don't need | |
3154 | * i_size_read because we hold i_mutex. | |
3155 | */ | |
3156 | if (pos + len > inode->i_size) | |
b9a4207d | 3157 | ext4_truncate_failed_write(inode); |
64769240 AT |
3158 | } |
3159 | ||
d2a17637 MC |
3160 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3161 | goto retry; | |
64769240 AT |
3162 | out: |
3163 | return ret; | |
3164 | } | |
3165 | ||
632eaeab MC |
3166 | /* |
3167 | * Check if we should update i_disksize | |
3168 | * when write to the end of file but not require block allocation | |
3169 | */ | |
3170 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3171 | unsigned long offset) |
632eaeab MC |
3172 | { |
3173 | struct buffer_head *bh; | |
3174 | struct inode *inode = page->mapping->host; | |
3175 | unsigned int idx; | |
3176 | int i; | |
3177 | ||
3178 | bh = page_buffers(page); | |
3179 | idx = offset >> inode->i_blkbits; | |
3180 | ||
af5bc92d | 3181 | for (i = 0; i < idx; i++) |
632eaeab MC |
3182 | bh = bh->b_this_page; |
3183 | ||
29fa89d0 | 3184 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3185 | return 0; |
3186 | return 1; | |
3187 | } | |
3188 | ||
64769240 | 3189 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3190 | struct address_space *mapping, |
3191 | loff_t pos, unsigned len, unsigned copied, | |
3192 | struct page *page, void *fsdata) | |
64769240 AT |
3193 | { |
3194 | struct inode *inode = mapping->host; | |
3195 | int ret = 0, ret2; | |
3196 | handle_t *handle = ext4_journal_current_handle(); | |
3197 | loff_t new_i_size; | |
632eaeab | 3198 | unsigned long start, end; |
79f0be8d AK |
3199 | int write_mode = (int)(unsigned long)fsdata; |
3200 | ||
3201 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
3202 | if (ext4_should_order_data(inode)) { | |
3203 | return ext4_ordered_write_end(file, mapping, pos, | |
3204 | len, copied, page, fsdata); | |
3205 | } else if (ext4_should_writeback_data(inode)) { | |
3206 | return ext4_writeback_write_end(file, mapping, pos, | |
3207 | len, copied, page, fsdata); | |
3208 | } else { | |
3209 | BUG(); | |
3210 | } | |
3211 | } | |
632eaeab | 3212 | |
9bffad1e | 3213 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 3214 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 3215 | end = start + copied - 1; |
64769240 AT |
3216 | |
3217 | /* | |
3218 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3219 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3220 | * into that. | |
3221 | */ | |
3222 | ||
3223 | new_i_size = pos + copied; | |
632eaeab MC |
3224 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
3225 | if (ext4_da_should_update_i_disksize(page, end)) { | |
3226 | down_write(&EXT4_I(inode)->i_data_sem); | |
3227 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
3228 | /* | |
3229 | * Updating i_disksize when extending file | |
3230 | * without needing block allocation | |
3231 | */ | |
3232 | if (ext4_should_order_data(inode)) | |
3233 | ret = ext4_jbd2_file_inode(handle, | |
3234 | inode); | |
64769240 | 3235 | |
632eaeab MC |
3236 | EXT4_I(inode)->i_disksize = new_i_size; |
3237 | } | |
3238 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
3239 | /* We need to mark inode dirty even if |
3240 | * new_i_size is less that inode->i_size | |
3241 | * bu greater than i_disksize.(hint delalloc) | |
3242 | */ | |
3243 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3244 | } |
632eaeab | 3245 | } |
64769240 AT |
3246 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
3247 | page, fsdata); | |
3248 | copied = ret2; | |
3249 | if (ret2 < 0) | |
3250 | ret = ret2; | |
3251 | ret2 = ext4_journal_stop(handle); | |
3252 | if (!ret) | |
3253 | ret = ret2; | |
3254 | ||
3255 | return ret ? ret : copied; | |
3256 | } | |
3257 | ||
3258 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
3259 | { | |
64769240 AT |
3260 | /* |
3261 | * Drop reserved blocks | |
3262 | */ | |
3263 | BUG_ON(!PageLocked(page)); | |
3264 | if (!page_has_buffers(page)) | |
3265 | goto out; | |
3266 | ||
d2a17637 | 3267 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
3268 | |
3269 | out: | |
3270 | ext4_invalidatepage(page, offset); | |
3271 | ||
3272 | return; | |
3273 | } | |
3274 | ||
ccd2506b TT |
3275 | /* |
3276 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3277 | */ | |
3278 | int ext4_alloc_da_blocks(struct inode *inode) | |
3279 | { | |
fb40ba0d TT |
3280 | trace_ext4_alloc_da_blocks(inode); |
3281 | ||
ccd2506b TT |
3282 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
3283 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
3284 | return 0; | |
3285 | ||
3286 | /* | |
3287 | * We do something simple for now. The filemap_flush() will | |
3288 | * also start triggering a write of the data blocks, which is | |
3289 | * not strictly speaking necessary (and for users of | |
3290 | * laptop_mode, not even desirable). However, to do otherwise | |
3291 | * would require replicating code paths in: | |
de9a55b8 | 3292 | * |
ccd2506b TT |
3293 | * ext4_da_writepages() -> |
3294 | * write_cache_pages() ---> (via passed in callback function) | |
3295 | * __mpage_da_writepage() --> | |
3296 | * mpage_add_bh_to_extent() | |
3297 | * mpage_da_map_blocks() | |
3298 | * | |
3299 | * The problem is that write_cache_pages(), located in | |
3300 | * mm/page-writeback.c, marks pages clean in preparation for | |
3301 | * doing I/O, which is not desirable if we're not planning on | |
3302 | * doing I/O at all. | |
3303 | * | |
3304 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3305 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3306 | * would be ugly in the extreme. So instead we would need to |
3307 | * replicate parts of the code in the above functions, | |
25985edc | 3308 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3309 | * write out the pages, but rather only collect contiguous |
3310 | * logical block extents, call the multi-block allocator, and | |
3311 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3312 | * |
ccd2506b TT |
3313 | * For now, though, we'll cheat by calling filemap_flush(), |
3314 | * which will map the blocks, and start the I/O, but not | |
3315 | * actually wait for the I/O to complete. | |
3316 | */ | |
3317 | return filemap_flush(inode->i_mapping); | |
3318 | } | |
64769240 | 3319 | |
ac27a0ec DK |
3320 | /* |
3321 | * bmap() is special. It gets used by applications such as lilo and by | |
3322 | * the swapper to find the on-disk block of a specific piece of data. | |
3323 | * | |
3324 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3325 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3326 | * filesystem and enables swap, then they may get a nasty shock when the |
3327 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3328 | * the original zero's written out previously to the journal and | |
3329 | * awaiting writeback in the kernel's buffer cache. | |
3330 | * | |
3331 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3332 | * take extra steps to flush any blocks which might be in the cache. | |
3333 | */ | |
617ba13b | 3334 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3335 | { |
3336 | struct inode *inode = mapping->host; | |
3337 | journal_t *journal; | |
3338 | int err; | |
3339 | ||
64769240 AT |
3340 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3341 | test_opt(inode->i_sb, DELALLOC)) { | |
3342 | /* | |
3343 | * With delalloc we want to sync the file | |
3344 | * so that we can make sure we allocate | |
3345 | * blocks for file | |
3346 | */ | |
3347 | filemap_write_and_wait(mapping); | |
3348 | } | |
3349 | ||
19f5fb7a TT |
3350 | if (EXT4_JOURNAL(inode) && |
3351 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3352 | /* |
3353 | * This is a REALLY heavyweight approach, but the use of | |
3354 | * bmap on dirty files is expected to be extremely rare: | |
3355 | * only if we run lilo or swapon on a freshly made file | |
3356 | * do we expect this to happen. | |
3357 | * | |
3358 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3359 | * represent an unprivileged user DOS attack --- we'd be | |
3360 | * in trouble if mortal users could trigger this path at | |
3361 | * will.) | |
3362 | * | |
617ba13b | 3363 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3364 | * regular files. If somebody wants to bmap a directory |
3365 | * or symlink and gets confused because the buffer | |
3366 | * hasn't yet been flushed to disk, they deserve | |
3367 | * everything they get. | |
3368 | */ | |
3369 | ||
19f5fb7a | 3370 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3371 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3372 | jbd2_journal_lock_updates(journal); |
3373 | err = jbd2_journal_flush(journal); | |
3374 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3375 | |
3376 | if (err) | |
3377 | return 0; | |
3378 | } | |
3379 | ||
af5bc92d | 3380 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3381 | } |
3382 | ||
617ba13b | 3383 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3384 | { |
0562e0ba | 3385 | trace_ext4_readpage(page); |
617ba13b | 3386 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
3387 | } |
3388 | ||
3389 | static int | |
617ba13b | 3390 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3391 | struct list_head *pages, unsigned nr_pages) |
3392 | { | |
617ba13b | 3393 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3394 | } |
3395 | ||
744692dc JZ |
3396 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) |
3397 | { | |
3398 | struct buffer_head *head, *bh; | |
3399 | unsigned int curr_off = 0; | |
3400 | ||
3401 | if (!page_has_buffers(page)) | |
3402 | return; | |
3403 | head = bh = page_buffers(page); | |
3404 | do { | |
3405 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
3406 | && bh->b_private) { | |
3407 | ext4_free_io_end(bh->b_private); | |
3408 | bh->b_private = NULL; | |
3409 | bh->b_end_io = NULL; | |
3410 | } | |
3411 | curr_off = curr_off + bh->b_size; | |
3412 | bh = bh->b_this_page; | |
3413 | } while (bh != head); | |
3414 | } | |
3415 | ||
617ba13b | 3416 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3417 | { |
617ba13b | 3418 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3419 | |
0562e0ba JZ |
3420 | trace_ext4_invalidatepage(page, offset); |
3421 | ||
744692dc JZ |
3422 | /* |
3423 | * free any io_end structure allocated for buffers to be discarded | |
3424 | */ | |
3425 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
3426 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
3427 | /* |
3428 | * If it's a full truncate we just forget about the pending dirtying | |
3429 | */ | |
3430 | if (offset == 0) | |
3431 | ClearPageChecked(page); | |
3432 | ||
0390131b FM |
3433 | if (journal) |
3434 | jbd2_journal_invalidatepage(journal, page, offset); | |
3435 | else | |
3436 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
3437 | } |
3438 | ||
617ba13b | 3439 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3440 | { |
617ba13b | 3441 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3442 | |
0562e0ba JZ |
3443 | trace_ext4_releasepage(page); |
3444 | ||
ac27a0ec DK |
3445 | WARN_ON(PageChecked(page)); |
3446 | if (!page_has_buffers(page)) | |
3447 | return 0; | |
0390131b FM |
3448 | if (journal) |
3449 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3450 | else | |
3451 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3452 | } |
3453 | ||
3454 | /* | |
4c0425ff MC |
3455 | * O_DIRECT for ext3 (or indirect map) based files |
3456 | * | |
ac27a0ec DK |
3457 | * If the O_DIRECT write will extend the file then add this inode to the |
3458 | * orphan list. So recovery will truncate it back to the original size | |
3459 | * if the machine crashes during the write. | |
3460 | * | |
3461 | * If the O_DIRECT write is intantiating holes inside i_size and the machine | |
7fb5409d JK |
3462 | * crashes then stale disk data _may_ be exposed inside the file. But current |
3463 | * VFS code falls back into buffered path in that case so we are safe. | |
ac27a0ec | 3464 | */ |
4c0425ff | 3465 | static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb, |
de9a55b8 TT |
3466 | const struct iovec *iov, loff_t offset, |
3467 | unsigned long nr_segs) | |
ac27a0ec DK |
3468 | { |
3469 | struct file *file = iocb->ki_filp; | |
3470 | struct inode *inode = file->f_mapping->host; | |
617ba13b | 3471 | struct ext4_inode_info *ei = EXT4_I(inode); |
7fb5409d | 3472 | handle_t *handle; |
ac27a0ec DK |
3473 | ssize_t ret; |
3474 | int orphan = 0; | |
3475 | size_t count = iov_length(iov, nr_segs); | |
fbbf6945 | 3476 | int retries = 0; |
ac27a0ec DK |
3477 | |
3478 | if (rw == WRITE) { | |
3479 | loff_t final_size = offset + count; | |
3480 | ||
ac27a0ec | 3481 | if (final_size > inode->i_size) { |
7fb5409d JK |
3482 | /* Credits for sb + inode write */ |
3483 | handle = ext4_journal_start(inode, 2); | |
3484 | if (IS_ERR(handle)) { | |
3485 | ret = PTR_ERR(handle); | |
3486 | goto out; | |
3487 | } | |
617ba13b | 3488 | ret = ext4_orphan_add(handle, inode); |
7fb5409d JK |
3489 | if (ret) { |
3490 | ext4_journal_stop(handle); | |
3491 | goto out; | |
3492 | } | |
ac27a0ec DK |
3493 | orphan = 1; |
3494 | ei->i_disksize = inode->i_size; | |
7fb5409d | 3495 | ext4_journal_stop(handle); |
ac27a0ec DK |
3496 | } |
3497 | } | |
3498 | ||
fbbf6945 | 3499 | retry: |
b7adc1f3 | 3500 | if (rw == READ && ext4_should_dioread_nolock(inode)) |
eafdc7d1 | 3501 | ret = __blockdev_direct_IO(rw, iocb, inode, |
b7adc1f3 JZ |
3502 | inode->i_sb->s_bdev, iov, |
3503 | offset, nr_segs, | |
eafdc7d1 CH |
3504 | ext4_get_block, NULL, NULL, 0); |
3505 | else { | |
b7adc1f3 JZ |
3506 | ret = blockdev_direct_IO(rw, iocb, inode, |
3507 | inode->i_sb->s_bdev, iov, | |
ac27a0ec | 3508 | offset, nr_segs, |
617ba13b | 3509 | ext4_get_block, NULL); |
eafdc7d1 CH |
3510 | |
3511 | if (unlikely((rw & WRITE) && ret < 0)) { | |
3512 | loff_t isize = i_size_read(inode); | |
3513 | loff_t end = offset + iov_length(iov, nr_segs); | |
3514 | ||
3515 | if (end > isize) | |
3516 | vmtruncate(inode, isize); | |
3517 | } | |
3518 | } | |
fbbf6945 ES |
3519 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
3520 | goto retry; | |
ac27a0ec | 3521 | |
7fb5409d | 3522 | if (orphan) { |
ac27a0ec DK |
3523 | int err; |
3524 | ||
7fb5409d JK |
3525 | /* Credits for sb + inode write */ |
3526 | handle = ext4_journal_start(inode, 2); | |
3527 | if (IS_ERR(handle)) { | |
3528 | /* This is really bad luck. We've written the data | |
3529 | * but cannot extend i_size. Bail out and pretend | |
3530 | * the write failed... */ | |
3531 | ret = PTR_ERR(handle); | |
da1dafca DM |
3532 | if (inode->i_nlink) |
3533 | ext4_orphan_del(NULL, inode); | |
3534 | ||
7fb5409d JK |
3535 | goto out; |
3536 | } | |
3537 | if (inode->i_nlink) | |
617ba13b | 3538 | ext4_orphan_del(handle, inode); |
7fb5409d | 3539 | if (ret > 0) { |
ac27a0ec DK |
3540 | loff_t end = offset + ret; |
3541 | if (end > inode->i_size) { | |
3542 | ei->i_disksize = end; | |
3543 | i_size_write(inode, end); | |
3544 | /* | |
3545 | * We're going to return a positive `ret' | |
3546 | * here due to non-zero-length I/O, so there's | |
3547 | * no way of reporting error returns from | |
617ba13b | 3548 | * ext4_mark_inode_dirty() to userspace. So |
ac27a0ec DK |
3549 | * ignore it. |
3550 | */ | |
617ba13b | 3551 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
3552 | } |
3553 | } | |
617ba13b | 3554 | err = ext4_journal_stop(handle); |
ac27a0ec DK |
3555 | if (ret == 0) |
3556 | ret = err; | |
3557 | } | |
3558 | out: | |
3559 | return ret; | |
3560 | } | |
3561 | ||
2ed88685 TT |
3562 | /* |
3563 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3564 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3565 | * The extent will be converted to initialized after the IO is complete. | |
3566 | */ | |
c7064ef1 | 3567 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3568 | struct buffer_head *bh_result, int create) |
3569 | { | |
c7064ef1 | 3570 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3571 | inode->i_ino, create); |
2ed88685 TT |
3572 | return _ext4_get_block(inode, iblock, bh_result, |
3573 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3574 | } |
3575 | ||
4c0425ff | 3576 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
3577 | ssize_t size, void *private, int ret, |
3578 | bool is_async) | |
4c0425ff MC |
3579 | { |
3580 | ext4_io_end_t *io_end = iocb->private; | |
3581 | struct workqueue_struct *wq; | |
744692dc JZ |
3582 | unsigned long flags; |
3583 | struct ext4_inode_info *ei; | |
4c0425ff | 3584 | |
4b70df18 M |
3585 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3586 | if (!io_end || !size) | |
552ef802 | 3587 | goto out; |
4b70df18 | 3588 | |
8d5d02e6 MC |
3589 | ext_debug("ext4_end_io_dio(): io_end 0x%p" |
3590 | "for inode %lu, iocb 0x%p, offset %llu, size %llu\n", | |
3591 | iocb->private, io_end->inode->i_ino, iocb, offset, | |
3592 | size); | |
8d5d02e6 MC |
3593 | |
3594 | /* if not aio dio with unwritten extents, just free io and return */ | |
bd2d0210 | 3595 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 MC |
3596 | ext4_free_io_end(io_end); |
3597 | iocb->private = NULL; | |
5b3ff237 JZ |
3598 | out: |
3599 | if (is_async) | |
3600 | aio_complete(iocb, ret, 0); | |
3601 | return; | |
8d5d02e6 MC |
3602 | } |
3603 | ||
4c0425ff MC |
3604 | io_end->offset = offset; |
3605 | io_end->size = size; | |
5b3ff237 JZ |
3606 | if (is_async) { |
3607 | io_end->iocb = iocb; | |
3608 | io_end->result = ret; | |
3609 | } | |
4c0425ff MC |
3610 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; |
3611 | ||
8d5d02e6 | 3612 | /* Add the io_end to per-inode completed aio dio list*/ |
744692dc JZ |
3613 | ei = EXT4_I(io_end->inode); |
3614 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
3615 | list_add_tail(&io_end->list, &ei->i_completed_io_list); | |
3616 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
c999af2b ES |
3617 | |
3618 | /* queue the work to convert unwritten extents to written */ | |
3619 | queue_work(wq, &io_end->work); | |
4c0425ff MC |
3620 | iocb->private = NULL; |
3621 | } | |
c7064ef1 | 3622 | |
744692dc JZ |
3623 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
3624 | { | |
3625 | ext4_io_end_t *io_end = bh->b_private; | |
3626 | struct workqueue_struct *wq; | |
3627 | struct inode *inode; | |
3628 | unsigned long flags; | |
3629 | ||
3630 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
3631 | goto out; | |
3632 | ||
3633 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
3634 | printk("sb umounted, discard end_io request for inode %lu\n", | |
3635 | io_end->inode->i_ino); | |
3636 | ext4_free_io_end(io_end); | |
3637 | goto out; | |
3638 | } | |
3639 | ||
bd2d0210 | 3640 | io_end->flag = EXT4_IO_END_UNWRITTEN; |
744692dc JZ |
3641 | inode = io_end->inode; |
3642 | ||
3643 | /* Add the io_end to per-inode completed io list*/ | |
3644 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
3645 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
3646 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
3647 | ||
3648 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
3649 | /* queue the work to convert unwritten extents to written */ | |
3650 | queue_work(wq, &io_end->work); | |
3651 | out: | |
3652 | bh->b_private = NULL; | |
3653 | bh->b_end_io = NULL; | |
3654 | clear_buffer_uninit(bh); | |
3655 | end_buffer_async_write(bh, uptodate); | |
3656 | } | |
3657 | ||
3658 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
3659 | { | |
3660 | ext4_io_end_t *io_end; | |
3661 | struct page *page = bh->b_page; | |
3662 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
3663 | size_t size = bh->b_size; | |
3664 | ||
3665 | retry: | |
3666 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
3667 | if (!io_end) { | |
6db26ffc | 3668 | pr_warn_ratelimited("%s: allocation fail\n", __func__); |
744692dc JZ |
3669 | schedule(); |
3670 | goto retry; | |
3671 | } | |
3672 | io_end->offset = offset; | |
3673 | io_end->size = size; | |
3674 | /* | |
3675 | * We need to hold a reference to the page to make sure it | |
3676 | * doesn't get evicted before ext4_end_io_work() has a chance | |
3677 | * to convert the extent from written to unwritten. | |
3678 | */ | |
3679 | io_end->page = page; | |
3680 | get_page(io_end->page); | |
3681 | ||
3682 | bh->b_private = io_end; | |
3683 | bh->b_end_io = ext4_end_io_buffer_write; | |
3684 | return 0; | |
3685 | } | |
3686 | ||
4c0425ff MC |
3687 | /* |
3688 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3689 | * preallocated extents, and those write extend the file, no need to | |
3690 | * fall back to buffered IO. | |
3691 | * | |
b595076a | 3692 | * For holes, we fallocate those blocks, mark them as uninitialized |
4c0425ff | 3693 | * If those blocks were preallocated, we mark sure they are splited, but |
b595076a | 3694 | * still keep the range to write as uninitialized. |
4c0425ff | 3695 | * |
8d5d02e6 MC |
3696 | * The unwrritten extents will be converted to written when DIO is completed. |
3697 | * For async direct IO, since the IO may still pending when return, we | |
25985edc | 3698 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3699 | * when async direct IO completed. |
4c0425ff MC |
3700 | * |
3701 | * If the O_DIRECT write will extend the file then add this inode to the | |
3702 | * orphan list. So recovery will truncate it back to the original size | |
3703 | * if the machine crashes during the write. | |
3704 | * | |
3705 | */ | |
3706 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3707 | const struct iovec *iov, loff_t offset, | |
3708 | unsigned long nr_segs) | |
3709 | { | |
3710 | struct file *file = iocb->ki_filp; | |
3711 | struct inode *inode = file->f_mapping->host; | |
3712 | ssize_t ret; | |
3713 | size_t count = iov_length(iov, nr_segs); | |
3714 | ||
3715 | loff_t final_size = offset + count; | |
3716 | if (rw == WRITE && final_size <= inode->i_size) { | |
3717 | /* | |
8d5d02e6 MC |
3718 | * We could direct write to holes and fallocate. |
3719 | * | |
3720 | * Allocated blocks to fill the hole are marked as uninitialized | |
25985edc | 3721 | * to prevent parallel buffered read to expose the stale data |
4c0425ff | 3722 | * before DIO complete the data IO. |
8d5d02e6 MC |
3723 | * |
3724 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
3725 | * will just simply mark the buffer mapped but still |
3726 | * keep the extents uninitialized. | |
3727 | * | |
8d5d02e6 MC |
3728 | * for non AIO case, we will convert those unwritten extents |
3729 | * to written after return back from blockdev_direct_IO. | |
3730 | * | |
3731 | * for async DIO, the conversion needs to be defered when | |
3732 | * the IO is completed. The ext4 end_io callback function | |
3733 | * will be called to take care of the conversion work. | |
3734 | * Here for async case, we allocate an io_end structure to | |
3735 | * hook to the iocb. | |
4c0425ff | 3736 | */ |
8d5d02e6 MC |
3737 | iocb->private = NULL; |
3738 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3739 | if (!is_sync_kiocb(iocb)) { | |
744692dc | 3740 | iocb->private = ext4_init_io_end(inode, GFP_NOFS); |
8d5d02e6 MC |
3741 | if (!iocb->private) |
3742 | return -ENOMEM; | |
3743 | /* | |
3744 | * we save the io structure for current async | |
79e83036 | 3745 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
3746 | * could flag the io structure whether there |
3747 | * is a unwritten extents needs to be converted | |
3748 | * when IO is completed. | |
3749 | */ | |
3750 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
3751 | } | |
3752 | ||
4c0425ff MC |
3753 | ret = blockdev_direct_IO(rw, iocb, inode, |
3754 | inode->i_sb->s_bdev, iov, | |
3755 | offset, nr_segs, | |
c7064ef1 | 3756 | ext4_get_block_write, |
4c0425ff | 3757 | ext4_end_io_dio); |
8d5d02e6 MC |
3758 | if (iocb->private) |
3759 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3760 | /* | |
3761 | * The io_end structure takes a reference to the inode, | |
3762 | * that structure needs to be destroyed and the | |
3763 | * reference to the inode need to be dropped, when IO is | |
3764 | * complete, even with 0 byte write, or failed. | |
3765 | * | |
3766 | * In the successful AIO DIO case, the io_end structure will be | |
3767 | * desctroyed and the reference to the inode will be dropped | |
3768 | * after the end_io call back function is called. | |
3769 | * | |
3770 | * In the case there is 0 byte write, or error case, since | |
3771 | * VFS direct IO won't invoke the end_io call back function, | |
3772 | * we need to free the end_io structure here. | |
3773 | */ | |
3774 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3775 | ext4_free_io_end(iocb->private); | |
3776 | iocb->private = NULL; | |
19f5fb7a TT |
3777 | } else if (ret > 0 && ext4_test_inode_state(inode, |
3778 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 3779 | int err; |
8d5d02e6 MC |
3780 | /* |
3781 | * for non AIO case, since the IO is already | |
25985edc | 3782 | * completed, we could do the conversion right here |
8d5d02e6 | 3783 | */ |
109f5565 M |
3784 | err = ext4_convert_unwritten_extents(inode, |
3785 | offset, ret); | |
3786 | if (err < 0) | |
3787 | ret = err; | |
19f5fb7a | 3788 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 3789 | } |
4c0425ff MC |
3790 | return ret; |
3791 | } | |
8d5d02e6 MC |
3792 | |
3793 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
3794 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
3795 | } | |
3796 | ||
3797 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3798 | const struct iovec *iov, loff_t offset, | |
3799 | unsigned long nr_segs) | |
3800 | { | |
3801 | struct file *file = iocb->ki_filp; | |
3802 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3803 | ssize_t ret; |
4c0425ff | 3804 | |
0562e0ba | 3805 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3806 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3807 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3808 | else | |
3809 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3810 | trace_ext4_direct_IO_exit(inode, offset, | |
3811 | iov_length(iov, nr_segs), rw, ret); | |
3812 | return ret; | |
4c0425ff MC |
3813 | } |
3814 | ||
ac27a0ec | 3815 | /* |
617ba13b | 3816 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3817 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3818 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3819 | * not necessarily locked. | |
3820 | * | |
3821 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3822 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3823 | * or jbddirty because all the journalling code will explode. | |
3824 | * | |
3825 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3826 | * is called, propagate that into the buffers appropriately. | |
3827 | */ | |
617ba13b | 3828 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3829 | { |
3830 | SetPageChecked(page); | |
3831 | return __set_page_dirty_nobuffers(page); | |
3832 | } | |
3833 | ||
617ba13b | 3834 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3835 | .readpage = ext4_readpage, |
3836 | .readpages = ext4_readpages, | |
43ce1d23 | 3837 | .writepage = ext4_writepage, |
8ab22b9a HH |
3838 | .write_begin = ext4_write_begin, |
3839 | .write_end = ext4_ordered_write_end, | |
3840 | .bmap = ext4_bmap, | |
3841 | .invalidatepage = ext4_invalidatepage, | |
3842 | .releasepage = ext4_releasepage, | |
3843 | .direct_IO = ext4_direct_IO, | |
3844 | .migratepage = buffer_migrate_page, | |
3845 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3846 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3847 | }; |
3848 | ||
617ba13b | 3849 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3850 | .readpage = ext4_readpage, |
3851 | .readpages = ext4_readpages, | |
43ce1d23 | 3852 | .writepage = ext4_writepage, |
8ab22b9a HH |
3853 | .write_begin = ext4_write_begin, |
3854 | .write_end = ext4_writeback_write_end, | |
3855 | .bmap = ext4_bmap, | |
3856 | .invalidatepage = ext4_invalidatepage, | |
3857 | .releasepage = ext4_releasepage, | |
3858 | .direct_IO = ext4_direct_IO, | |
3859 | .migratepage = buffer_migrate_page, | |
3860 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3861 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3862 | }; |
3863 | ||
617ba13b | 3864 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3865 | .readpage = ext4_readpage, |
3866 | .readpages = ext4_readpages, | |
43ce1d23 | 3867 | .writepage = ext4_writepage, |
8ab22b9a HH |
3868 | .write_begin = ext4_write_begin, |
3869 | .write_end = ext4_journalled_write_end, | |
3870 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3871 | .bmap = ext4_bmap, | |
3872 | .invalidatepage = ext4_invalidatepage, | |
3873 | .releasepage = ext4_releasepage, | |
3874 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3875 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3876 | }; |
3877 | ||
64769240 | 3878 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3879 | .readpage = ext4_readpage, |
3880 | .readpages = ext4_readpages, | |
43ce1d23 | 3881 | .writepage = ext4_writepage, |
8ab22b9a | 3882 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3883 | .write_begin = ext4_da_write_begin, |
3884 | .write_end = ext4_da_write_end, | |
3885 | .bmap = ext4_bmap, | |
3886 | .invalidatepage = ext4_da_invalidatepage, | |
3887 | .releasepage = ext4_releasepage, | |
3888 | .direct_IO = ext4_direct_IO, | |
3889 | .migratepage = buffer_migrate_page, | |
3890 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3891 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3892 | }; |
3893 | ||
617ba13b | 3894 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3895 | { |
cd1aac32 AK |
3896 | if (ext4_should_order_data(inode) && |
3897 | test_opt(inode->i_sb, DELALLOC)) | |
3898 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3899 | else if (ext4_should_order_data(inode)) | |
617ba13b | 3900 | inode->i_mapping->a_ops = &ext4_ordered_aops; |
64769240 AT |
3901 | else if (ext4_should_writeback_data(inode) && |
3902 | test_opt(inode->i_sb, DELALLOC)) | |
3903 | inode->i_mapping->a_ops = &ext4_da_aops; | |
617ba13b MC |
3904 | else if (ext4_should_writeback_data(inode)) |
3905 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
ac27a0ec | 3906 | else |
617ba13b | 3907 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
ac27a0ec DK |
3908 | } |
3909 | ||
3910 | /* | |
617ba13b | 3911 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' |
ac27a0ec DK |
3912 | * up to the end of the block which corresponds to `from'. |
3913 | * This required during truncate. We need to physically zero the tail end | |
3914 | * of that block so it doesn't yield old data if the file is later grown. | |
3915 | */ | |
cf108bca | 3916 | int ext4_block_truncate_page(handle_t *handle, |
ac27a0ec DK |
3917 | struct address_space *mapping, loff_t from) |
3918 | { | |
617ba13b | 3919 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
ac27a0ec | 3920 | unsigned offset = from & (PAGE_CACHE_SIZE-1); |
725d26d3 AK |
3921 | unsigned blocksize, length, pos; |
3922 | ext4_lblk_t iblock; | |
ac27a0ec DK |
3923 | struct inode *inode = mapping->host; |
3924 | struct buffer_head *bh; | |
cf108bca | 3925 | struct page *page; |
ac27a0ec | 3926 | int err = 0; |
ac27a0ec | 3927 | |
f4a01017 TT |
3928 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
3929 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
cf108bca JK |
3930 | if (!page) |
3931 | return -EINVAL; | |
3932 | ||
ac27a0ec DK |
3933 | blocksize = inode->i_sb->s_blocksize; |
3934 | length = blocksize - (offset & (blocksize - 1)); | |
3935 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3936 | ||
ac27a0ec DK |
3937 | if (!page_has_buffers(page)) |
3938 | create_empty_buffers(page, blocksize, 0); | |
3939 | ||
3940 | /* Find the buffer that contains "offset" */ | |
3941 | bh = page_buffers(page); | |
3942 | pos = blocksize; | |
3943 | while (offset >= pos) { | |
3944 | bh = bh->b_this_page; | |
3945 | iblock++; | |
3946 | pos += blocksize; | |
3947 | } | |
3948 | ||
3949 | err = 0; | |
3950 | if (buffer_freed(bh)) { | |
3951 | BUFFER_TRACE(bh, "freed: skip"); | |
3952 | goto unlock; | |
3953 | } | |
3954 | ||
3955 | if (!buffer_mapped(bh)) { | |
3956 | BUFFER_TRACE(bh, "unmapped"); | |
617ba13b | 3957 | ext4_get_block(inode, iblock, bh, 0); |
ac27a0ec DK |
3958 | /* unmapped? It's a hole - nothing to do */ |
3959 | if (!buffer_mapped(bh)) { | |
3960 | BUFFER_TRACE(bh, "still unmapped"); | |
3961 | goto unlock; | |
3962 | } | |
3963 | } | |
3964 | ||
3965 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3966 | if (PageUptodate(page)) | |
3967 | set_buffer_uptodate(bh); | |
3968 | ||
3969 | if (!buffer_uptodate(bh)) { | |
3970 | err = -EIO; | |
3971 | ll_rw_block(READ, 1, &bh); | |
3972 | wait_on_buffer(bh); | |
3973 | /* Uhhuh. Read error. Complain and punt. */ | |
3974 | if (!buffer_uptodate(bh)) | |
3975 | goto unlock; | |
3976 | } | |
3977 | ||
617ba13b | 3978 | if (ext4_should_journal_data(inode)) { |
ac27a0ec | 3979 | BUFFER_TRACE(bh, "get write access"); |
617ba13b | 3980 | err = ext4_journal_get_write_access(handle, bh); |
ac27a0ec DK |
3981 | if (err) |
3982 | goto unlock; | |
3983 | } | |
3984 | ||
eebd2aa3 | 3985 | zero_user(page, offset, length); |
ac27a0ec DK |
3986 | |
3987 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3988 | ||
3989 | err = 0; | |
617ba13b | 3990 | if (ext4_should_journal_data(inode)) { |
0390131b | 3991 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
ac27a0ec | 3992 | } else { |
8aefcd55 | 3993 | if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode) |
678aaf48 | 3994 | err = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
3995 | mark_buffer_dirty(bh); |
3996 | } | |
3997 | ||
3998 | unlock: | |
3999 | unlock_page(page); | |
4000 | page_cache_release(page); | |
4001 | return err; | |
4002 | } | |
4003 | ||
4004 | /* | |
4005 | * Probably it should be a library function... search for first non-zero word | |
4006 | * or memcmp with zero_page, whatever is better for particular architecture. | |
4007 | * Linus? | |
4008 | */ | |
4009 | static inline int all_zeroes(__le32 *p, __le32 *q) | |
4010 | { | |
4011 | while (p < q) | |
4012 | if (*p++) | |
4013 | return 0; | |
4014 | return 1; | |
4015 | } | |
4016 | ||
4017 | /** | |
617ba13b | 4018 | * ext4_find_shared - find the indirect blocks for partial truncation. |
ac27a0ec DK |
4019 | * @inode: inode in question |
4020 | * @depth: depth of the affected branch | |
617ba13b | 4021 | * @offsets: offsets of pointers in that branch (see ext4_block_to_path) |
ac27a0ec DK |
4022 | * @chain: place to store the pointers to partial indirect blocks |
4023 | * @top: place to the (detached) top of branch | |
4024 | * | |
617ba13b | 4025 | * This is a helper function used by ext4_truncate(). |
ac27a0ec DK |
4026 | * |
4027 | * When we do truncate() we may have to clean the ends of several | |
4028 | * indirect blocks but leave the blocks themselves alive. Block is | |
25985edc | 4029 | * partially truncated if some data below the new i_size is referred |
ac27a0ec DK |
4030 | * from it (and it is on the path to the first completely truncated |
4031 | * data block, indeed). We have to free the top of that path along | |
4032 | * with everything to the right of the path. Since no allocation | |
617ba13b | 4033 | * past the truncation point is possible until ext4_truncate() |
ac27a0ec DK |
4034 | * finishes, we may safely do the latter, but top of branch may |
4035 | * require special attention - pageout below the truncation point | |
4036 | * might try to populate it. | |
4037 | * | |
4038 | * We atomically detach the top of branch from the tree, store the | |
4039 | * block number of its root in *@top, pointers to buffer_heads of | |
4040 | * partially truncated blocks - in @chain[].bh and pointers to | |
4041 | * their last elements that should not be removed - in | |
4042 | * @chain[].p. Return value is the pointer to last filled element | |
4043 | * of @chain. | |
4044 | * | |
4045 | * The work left to caller to do the actual freeing of subtrees: | |
4046 | * a) free the subtree starting from *@top | |
4047 | * b) free the subtrees whose roots are stored in | |
4048 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) | |
4049 | * c) free the subtrees growing from the inode past the @chain[0]. | |
4050 | * (no partially truncated stuff there). */ | |
4051 | ||
617ba13b | 4052 | static Indirect *ext4_find_shared(struct inode *inode, int depth, |
de9a55b8 TT |
4053 | ext4_lblk_t offsets[4], Indirect chain[4], |
4054 | __le32 *top) | |
ac27a0ec DK |
4055 | { |
4056 | Indirect *partial, *p; | |
4057 | int k, err; | |
4058 | ||
4059 | *top = 0; | |
bf48aabb | 4060 | /* Make k index the deepest non-null offset + 1 */ |
ac27a0ec DK |
4061 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
4062 | ; | |
617ba13b | 4063 | partial = ext4_get_branch(inode, k, offsets, chain, &err); |
ac27a0ec DK |
4064 | /* Writer: pointers */ |
4065 | if (!partial) | |
4066 | partial = chain + k-1; | |
4067 | /* | |
4068 | * If the branch acquired continuation since we've looked at it - | |
4069 | * fine, it should all survive and (new) top doesn't belong to us. | |
4070 | */ | |
4071 | if (!partial->key && *partial->p) | |
4072 | /* Writer: end */ | |
4073 | goto no_top; | |
af5bc92d | 4074 | for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--) |
ac27a0ec DK |
4075 | ; |
4076 | /* | |
4077 | * OK, we've found the last block that must survive. The rest of our | |
4078 | * branch should be detached before unlocking. However, if that rest | |
4079 | * of branch is all ours and does not grow immediately from the inode | |
4080 | * it's easier to cheat and just decrement partial->p. | |
4081 | */ | |
4082 | if (p == chain + k - 1 && p > chain) { | |
4083 | p->p--; | |
4084 | } else { | |
4085 | *top = *p->p; | |
617ba13b | 4086 | /* Nope, don't do this in ext4. Must leave the tree intact */ |
ac27a0ec DK |
4087 | #if 0 |
4088 | *p->p = 0; | |
4089 | #endif | |
4090 | } | |
4091 | /* Writer: end */ | |
4092 | ||
af5bc92d | 4093 | while (partial > p) { |
ac27a0ec DK |
4094 | brelse(partial->bh); |
4095 | partial--; | |
4096 | } | |
4097 | no_top: | |
4098 | return partial; | |
4099 | } | |
4100 | ||
4101 | /* | |
4102 | * Zero a number of block pointers in either an inode or an indirect block. | |
4103 | * If we restart the transaction we must again get write access to the | |
4104 | * indirect block for further modification. | |
4105 | * | |
4106 | * We release `count' blocks on disk, but (last - first) may be greater | |
4107 | * than `count' because there can be holes in there. | |
d67d1218 AG |
4108 | * |
4109 | * Return 0 on success, 1 on invalid block range | |
4110 | * and < 0 on fatal error. | |
ac27a0ec | 4111 | */ |
1f2acb60 TT |
4112 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, |
4113 | struct buffer_head *bh, | |
4114 | ext4_fsblk_t block_to_free, | |
4115 | unsigned long count, __le32 *first, | |
4116 | __le32 *last) | |
ac27a0ec DK |
4117 | { |
4118 | __le32 *p; | |
1f2acb60 | 4119 | int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED; |
b4097142 | 4120 | int err; |
e6362609 TT |
4121 | |
4122 | if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) | |
4123 | flags |= EXT4_FREE_BLOCKS_METADATA; | |
50689696 | 4124 | |
1f2acb60 TT |
4125 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free, |
4126 | count)) { | |
24676da4 TT |
4127 | EXT4_ERROR_INODE(inode, "attempt to clear invalid " |
4128 | "blocks %llu len %lu", | |
4129 | (unsigned long long) block_to_free, count); | |
1f2acb60 TT |
4130 | return 1; |
4131 | } | |
4132 | ||
ac27a0ec DK |
4133 | if (try_to_extend_transaction(handle, inode)) { |
4134 | if (bh) { | |
0390131b | 4135 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
b4097142 | 4136 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
d67d1218 AG |
4137 | if (unlikely(err)) |
4138 | goto out_err; | |
b4097142 TT |
4139 | } |
4140 | err = ext4_mark_inode_dirty(handle, inode); | |
d67d1218 AG |
4141 | if (unlikely(err)) |
4142 | goto out_err; | |
b4097142 TT |
4143 | err = ext4_truncate_restart_trans(handle, inode, |
4144 | blocks_for_truncate(inode)); | |
d67d1218 AG |
4145 | if (unlikely(err)) |
4146 | goto out_err; | |
ac27a0ec DK |
4147 | if (bh) { |
4148 | BUFFER_TRACE(bh, "retaking write access"); | |
d67d1218 AG |
4149 | err = ext4_journal_get_write_access(handle, bh); |
4150 | if (unlikely(err)) | |
4151 | goto out_err; | |
ac27a0ec DK |
4152 | } |
4153 | } | |
4154 | ||
e6362609 TT |
4155 | for (p = first; p < last; p++) |
4156 | *p = 0; | |
ac27a0ec | 4157 | |
7dc57615 | 4158 | ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); |
1f2acb60 | 4159 | return 0; |
d67d1218 AG |
4160 | out_err: |
4161 | ext4_std_error(inode->i_sb, err); | |
4162 | return err; | |
ac27a0ec DK |
4163 | } |
4164 | ||
4165 | /** | |
617ba13b | 4166 | * ext4_free_data - free a list of data blocks |
ac27a0ec DK |
4167 | * @handle: handle for this transaction |
4168 | * @inode: inode we are dealing with | |
4169 | * @this_bh: indirect buffer_head which contains *@first and *@last | |
4170 | * @first: array of block numbers | |
4171 | * @last: points immediately past the end of array | |
4172 | * | |
25985edc | 4173 | * We are freeing all blocks referred from that array (numbers are stored as |
ac27a0ec DK |
4174 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. |
4175 | * | |
4176 | * We accumulate contiguous runs of blocks to free. Conveniently, if these | |
4177 | * blocks are contiguous then releasing them at one time will only affect one | |
4178 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't | |
4179 | * actually use a lot of journal space. | |
4180 | * | |
4181 | * @this_bh will be %NULL if @first and @last point into the inode's direct | |
4182 | * block pointers. | |
4183 | */ | |
617ba13b | 4184 | static void ext4_free_data(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4185 | struct buffer_head *this_bh, |
4186 | __le32 *first, __le32 *last) | |
4187 | { | |
617ba13b | 4188 | ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
ac27a0ec DK |
4189 | unsigned long count = 0; /* Number of blocks in the run */ |
4190 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind | |
4191 | corresponding to | |
4192 | block_to_free */ | |
617ba13b | 4193 | ext4_fsblk_t nr; /* Current block # */ |
ac27a0ec DK |
4194 | __le32 *p; /* Pointer into inode/ind |
4195 | for current block */ | |
d67d1218 | 4196 | int err = 0; |
ac27a0ec DK |
4197 | |
4198 | if (this_bh) { /* For indirect block */ | |
4199 | BUFFER_TRACE(this_bh, "get_write_access"); | |
617ba13b | 4200 | err = ext4_journal_get_write_access(handle, this_bh); |
ac27a0ec DK |
4201 | /* Important: if we can't update the indirect pointers |
4202 | * to the blocks, we can't free them. */ | |
4203 | if (err) | |
4204 | return; | |
4205 | } | |
4206 | ||
4207 | for (p = first; p < last; p++) { | |
4208 | nr = le32_to_cpu(*p); | |
4209 | if (nr) { | |
4210 | /* accumulate blocks to free if they're contiguous */ | |
4211 | if (count == 0) { | |
4212 | block_to_free = nr; | |
4213 | block_to_free_p = p; | |
4214 | count = 1; | |
4215 | } else if (nr == block_to_free + count) { | |
4216 | count++; | |
4217 | } else { | |
d67d1218 AG |
4218 | err = ext4_clear_blocks(handle, inode, this_bh, |
4219 | block_to_free, count, | |
4220 | block_to_free_p, p); | |
4221 | if (err) | |
1f2acb60 | 4222 | break; |
ac27a0ec DK |
4223 | block_to_free = nr; |
4224 | block_to_free_p = p; | |
4225 | count = 1; | |
4226 | } | |
4227 | } | |
4228 | } | |
4229 | ||
d67d1218 AG |
4230 | if (!err && count > 0) |
4231 | err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, | |
4232 | count, block_to_free_p, p); | |
4233 | if (err < 0) | |
4234 | /* fatal error */ | |
4235 | return; | |
ac27a0ec DK |
4236 | |
4237 | if (this_bh) { | |
0390131b | 4238 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
71dc8fbc DG |
4239 | |
4240 | /* | |
4241 | * The buffer head should have an attached journal head at this | |
4242 | * point. However, if the data is corrupted and an indirect | |
4243 | * block pointed to itself, it would have been detached when | |
4244 | * the block was cleared. Check for this instead of OOPSing. | |
4245 | */ | |
e7f07968 | 4246 | if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh)) |
0390131b | 4247 | ext4_handle_dirty_metadata(handle, inode, this_bh); |
71dc8fbc | 4248 | else |
24676da4 TT |
4249 | EXT4_ERROR_INODE(inode, |
4250 | "circular indirect block detected at " | |
4251 | "block %llu", | |
4252 | (unsigned long long) this_bh->b_blocknr); | |
ac27a0ec DK |
4253 | } |
4254 | } | |
4255 | ||
4256 | /** | |
617ba13b | 4257 | * ext4_free_branches - free an array of branches |
ac27a0ec DK |
4258 | * @handle: JBD handle for this transaction |
4259 | * @inode: inode we are dealing with | |
4260 | * @parent_bh: the buffer_head which contains *@first and *@last | |
4261 | * @first: array of block numbers | |
4262 | * @last: pointer immediately past the end of array | |
4263 | * @depth: depth of the branches to free | |
4264 | * | |
25985edc | 4265 | * We are freeing all blocks referred from these branches (numbers are |
ac27a0ec DK |
4266 | * stored as little-endian 32-bit) and updating @inode->i_blocks |
4267 | * appropriately. | |
4268 | */ | |
617ba13b | 4269 | static void ext4_free_branches(handle_t *handle, struct inode *inode, |
ac27a0ec DK |
4270 | struct buffer_head *parent_bh, |
4271 | __le32 *first, __le32 *last, int depth) | |
4272 | { | |
617ba13b | 4273 | ext4_fsblk_t nr; |
ac27a0ec DK |
4274 | __le32 *p; |
4275 | ||
0390131b | 4276 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4277 | return; |
4278 | ||
4279 | if (depth--) { | |
4280 | struct buffer_head *bh; | |
617ba13b | 4281 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec DK |
4282 | p = last; |
4283 | while (--p >= first) { | |
4284 | nr = le32_to_cpu(*p); | |
4285 | if (!nr) | |
4286 | continue; /* A hole */ | |
4287 | ||
1f2acb60 TT |
4288 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), |
4289 | nr, 1)) { | |
24676da4 TT |
4290 | EXT4_ERROR_INODE(inode, |
4291 | "invalid indirect mapped " | |
4292 | "block %lu (level %d)", | |
4293 | (unsigned long) nr, depth); | |
1f2acb60 TT |
4294 | break; |
4295 | } | |
4296 | ||
ac27a0ec DK |
4297 | /* Go read the buffer for the next level down */ |
4298 | bh = sb_bread(inode->i_sb, nr); | |
4299 | ||
4300 | /* | |
4301 | * A read failure? Report error and clear slot | |
4302 | * (should be rare). | |
4303 | */ | |
4304 | if (!bh) { | |
c398eda0 TT |
4305 | EXT4_ERROR_INODE_BLOCK(inode, nr, |
4306 | "Read failure"); | |
ac27a0ec DK |
4307 | continue; |
4308 | } | |
4309 | ||
4310 | /* This zaps the entire block. Bottom up. */ | |
4311 | BUFFER_TRACE(bh, "free child branches"); | |
617ba13b | 4312 | ext4_free_branches(handle, inode, bh, |
af5bc92d TT |
4313 | (__le32 *) bh->b_data, |
4314 | (__le32 *) bh->b_data + addr_per_block, | |
4315 | depth); | |
1c5b9e90 | 4316 | brelse(bh); |
ac27a0ec | 4317 | |
ac27a0ec DK |
4318 | /* |
4319 | * Everything below this this pointer has been | |
4320 | * released. Now let this top-of-subtree go. | |
4321 | * | |
4322 | * We want the freeing of this indirect block to be | |
4323 | * atomic in the journal with the updating of the | |
4324 | * bitmap block which owns it. So make some room in | |
4325 | * the journal. | |
4326 | * | |
4327 | * We zero the parent pointer *after* freeing its | |
4328 | * pointee in the bitmaps, so if extend_transaction() | |
4329 | * for some reason fails to put the bitmap changes and | |
4330 | * the release into the same transaction, recovery | |
4331 | * will merely complain about releasing a free block, | |
4332 | * rather than leaking blocks. | |
4333 | */ | |
0390131b | 4334 | if (ext4_handle_is_aborted(handle)) |
ac27a0ec DK |
4335 | return; |
4336 | if (try_to_extend_transaction(handle, inode)) { | |
617ba13b | 4337 | ext4_mark_inode_dirty(handle, inode); |
487caeef JK |
4338 | ext4_truncate_restart_trans(handle, inode, |
4339 | blocks_for_truncate(inode)); | |
ac27a0ec DK |
4340 | } |
4341 | ||
40389687 A |
4342 | /* |
4343 | * The forget flag here is critical because if | |
4344 | * we are journaling (and not doing data | |
4345 | * journaling), we have to make sure a revoke | |
4346 | * record is written to prevent the journal | |
4347 | * replay from overwriting the (former) | |
4348 | * indirect block if it gets reallocated as a | |
4349 | * data block. This must happen in the same | |
4350 | * transaction where the data blocks are | |
4351 | * actually freed. | |
4352 | */ | |
7dc57615 | 4353 | ext4_free_blocks(handle, inode, NULL, nr, 1, |
40389687 A |
4354 | EXT4_FREE_BLOCKS_METADATA| |
4355 | EXT4_FREE_BLOCKS_FORGET); | |
ac27a0ec DK |
4356 | |
4357 | if (parent_bh) { | |
4358 | /* | |
4359 | * The block which we have just freed is | |
4360 | * pointed to by an indirect block: journal it | |
4361 | */ | |
4362 | BUFFER_TRACE(parent_bh, "get_write_access"); | |
617ba13b | 4363 | if (!ext4_journal_get_write_access(handle, |
ac27a0ec DK |
4364 | parent_bh)){ |
4365 | *p = 0; | |
4366 | BUFFER_TRACE(parent_bh, | |
0390131b FM |
4367 | "call ext4_handle_dirty_metadata"); |
4368 | ext4_handle_dirty_metadata(handle, | |
4369 | inode, | |
4370 | parent_bh); | |
ac27a0ec DK |
4371 | } |
4372 | } | |
4373 | } | |
4374 | } else { | |
4375 | /* We have reached the bottom of the tree. */ | |
4376 | BUFFER_TRACE(parent_bh, "free data blocks"); | |
617ba13b | 4377 | ext4_free_data(handle, inode, parent_bh, first, last); |
ac27a0ec DK |
4378 | } |
4379 | } | |
4380 | ||
91ef4caf DG |
4381 | int ext4_can_truncate(struct inode *inode) |
4382 | { | |
4383 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
4384 | return 0; | |
4385 | if (S_ISREG(inode->i_mode)) | |
4386 | return 1; | |
4387 | if (S_ISDIR(inode->i_mode)) | |
4388 | return 1; | |
4389 | if (S_ISLNK(inode->i_mode)) | |
4390 | return !ext4_inode_is_fast_symlink(inode); | |
4391 | return 0; | |
4392 | } | |
4393 | ||
ac27a0ec | 4394 | /* |
617ba13b | 4395 | * ext4_truncate() |
ac27a0ec | 4396 | * |
617ba13b MC |
4397 | * We block out ext4_get_block() block instantiations across the entire |
4398 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4399 | * simultaneously on behalf of the same inode. |
4400 | * | |
4401 | * As we work through the truncate and commmit bits of it to the journal there | |
4402 | * is one core, guiding principle: the file's tree must always be consistent on | |
4403 | * disk. We must be able to restart the truncate after a crash. | |
4404 | * | |
4405 | * The file's tree may be transiently inconsistent in memory (although it | |
4406 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4407 | * the contents of (the filesystem + the journal) must be consistent and | |
4408 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4409 | * left-to-right works OK too). | |
4410 | * | |
4411 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4412 | * truncate against the orphan inode list. | |
4413 | * | |
4414 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4415 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4416 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4417 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4418 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4419 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4420 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4421 | */ |
617ba13b | 4422 | void ext4_truncate(struct inode *inode) |
ac27a0ec DK |
4423 | { |
4424 | handle_t *handle; | |
617ba13b | 4425 | struct ext4_inode_info *ei = EXT4_I(inode); |
ac27a0ec | 4426 | __le32 *i_data = ei->i_data; |
617ba13b | 4427 | int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb); |
ac27a0ec | 4428 | struct address_space *mapping = inode->i_mapping; |
725d26d3 | 4429 | ext4_lblk_t offsets[4]; |
ac27a0ec DK |
4430 | Indirect chain[4]; |
4431 | Indirect *partial; | |
4432 | __le32 nr = 0; | |
f80da1e7 KM |
4433 | int n = 0; |
4434 | ext4_lblk_t last_block, max_block; | |
ac27a0ec | 4435 | unsigned blocksize = inode->i_sb->s_blocksize; |
ac27a0ec | 4436 | |
0562e0ba JZ |
4437 | trace_ext4_truncate_enter(inode); |
4438 | ||
91ef4caf | 4439 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
4440 | return; |
4441 | ||
12e9b892 | 4442 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4443 | |
5534fb5b | 4444 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4445 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4446 | |
12e9b892 | 4447 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
cf108bca | 4448 | ext4_ext_truncate(inode); |
0562e0ba | 4449 | trace_ext4_truncate_exit(inode); |
1d03ec98 AK |
4450 | return; |
4451 | } | |
a86c6181 | 4452 | |
ac27a0ec | 4453 | handle = start_transaction(inode); |
cf108bca | 4454 | if (IS_ERR(handle)) |
ac27a0ec | 4455 | return; /* AKPM: return what? */ |
ac27a0ec DK |
4456 | |
4457 | last_block = (inode->i_size + blocksize-1) | |
617ba13b | 4458 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); |
f80da1e7 KM |
4459 | max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1) |
4460 | >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); | |
ac27a0ec | 4461 | |
cf108bca JK |
4462 | if (inode->i_size & (blocksize - 1)) |
4463 | if (ext4_block_truncate_page(handle, mapping, inode->i_size)) | |
4464 | goto out_stop; | |
ac27a0ec | 4465 | |
f80da1e7 KM |
4466 | if (last_block != max_block) { |
4467 | n = ext4_block_to_path(inode, last_block, offsets, NULL); | |
4468 | if (n == 0) | |
4469 | goto out_stop; /* error */ | |
4470 | } | |
ac27a0ec DK |
4471 | |
4472 | /* | |
4473 | * OK. This truncate is going to happen. We add the inode to the | |
4474 | * orphan list, so that if this truncate spans multiple transactions, | |
4475 | * and we crash, we will resume the truncate when the filesystem | |
4476 | * recovers. It also marks the inode dirty, to catch the new size. | |
4477 | * | |
4478 | * Implication: the file must always be in a sane, consistent | |
4479 | * truncatable state while each transaction commits. | |
4480 | */ | |
617ba13b | 4481 | if (ext4_orphan_add(handle, inode)) |
ac27a0ec DK |
4482 | goto out_stop; |
4483 | ||
632eaeab MC |
4484 | /* |
4485 | * From here we block out all ext4_get_block() callers who want to | |
4486 | * modify the block allocation tree. | |
4487 | */ | |
4488 | down_write(&ei->i_data_sem); | |
b4df2030 | 4489 | |
c2ea3fde | 4490 | ext4_discard_preallocations(inode); |
b4df2030 | 4491 | |
ac27a0ec DK |
4492 | /* |
4493 | * The orphan list entry will now protect us from any crash which | |
4494 | * occurs before the truncate completes, so it is now safe to propagate | |
4495 | * the new, shorter inode size (held for now in i_size) into the | |
4496 | * on-disk inode. We do this via i_disksize, which is the value which | |
617ba13b | 4497 | * ext4 *really* writes onto the disk inode. |
ac27a0ec DK |
4498 | */ |
4499 | ei->i_disksize = inode->i_size; | |
4500 | ||
f80da1e7 KM |
4501 | if (last_block == max_block) { |
4502 | /* | |
4503 | * It is unnecessary to free any data blocks if last_block is | |
4504 | * equal to the indirect block limit. | |
4505 | */ | |
4506 | goto out_unlock; | |
4507 | } else if (n == 1) { /* direct blocks */ | |
617ba13b MC |
4508 | ext4_free_data(handle, inode, NULL, i_data+offsets[0], |
4509 | i_data + EXT4_NDIR_BLOCKS); | |
ac27a0ec DK |
4510 | goto do_indirects; |
4511 | } | |
4512 | ||
617ba13b | 4513 | partial = ext4_find_shared(inode, n, offsets, chain, &nr); |
ac27a0ec DK |
4514 | /* Kill the top of shared branch (not detached) */ |
4515 | if (nr) { | |
4516 | if (partial == chain) { | |
4517 | /* Shared branch grows from the inode */ | |
617ba13b | 4518 | ext4_free_branches(handle, inode, NULL, |
ac27a0ec DK |
4519 | &nr, &nr+1, (chain+n-1) - partial); |
4520 | *partial->p = 0; | |
4521 | /* | |
4522 | * We mark the inode dirty prior to restart, | |
4523 | * and prior to stop. No need for it here. | |
4524 | */ | |
4525 | } else { | |
4526 | /* Shared branch grows from an indirect block */ | |
4527 | BUFFER_TRACE(partial->bh, "get_write_access"); | |
617ba13b | 4528 | ext4_free_branches(handle, inode, partial->bh, |
ac27a0ec DK |
4529 | partial->p, |
4530 | partial->p+1, (chain+n-1) - partial); | |
4531 | } | |
4532 | } | |
4533 | /* Clear the ends of indirect blocks on the shared branch */ | |
4534 | while (partial > chain) { | |
617ba13b | 4535 | ext4_free_branches(handle, inode, partial->bh, partial->p + 1, |
ac27a0ec DK |
4536 | (__le32*)partial->bh->b_data+addr_per_block, |
4537 | (chain+n-1) - partial); | |
4538 | BUFFER_TRACE(partial->bh, "call brelse"); | |
de9a55b8 | 4539 | brelse(partial->bh); |
ac27a0ec DK |
4540 | partial--; |
4541 | } | |
4542 | do_indirects: | |
4543 | /* Kill the remaining (whole) subtrees */ | |
4544 | switch (offsets[0]) { | |
4545 | default: | |
617ba13b | 4546 | nr = i_data[EXT4_IND_BLOCK]; |
ac27a0ec | 4547 | if (nr) { |
617ba13b MC |
4548 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
4549 | i_data[EXT4_IND_BLOCK] = 0; | |
ac27a0ec | 4550 | } |
617ba13b MC |
4551 | case EXT4_IND_BLOCK: |
4552 | nr = i_data[EXT4_DIND_BLOCK]; | |
ac27a0ec | 4553 | if (nr) { |
617ba13b MC |
4554 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
4555 | i_data[EXT4_DIND_BLOCK] = 0; | |
ac27a0ec | 4556 | } |
617ba13b MC |
4557 | case EXT4_DIND_BLOCK: |
4558 | nr = i_data[EXT4_TIND_BLOCK]; | |
ac27a0ec | 4559 | if (nr) { |
617ba13b MC |
4560 | ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
4561 | i_data[EXT4_TIND_BLOCK] = 0; | |
ac27a0ec | 4562 | } |
617ba13b | 4563 | case EXT4_TIND_BLOCK: |
ac27a0ec DK |
4564 | ; |
4565 | } | |
4566 | ||
f80da1e7 | 4567 | out_unlock: |
0e855ac8 | 4568 | up_write(&ei->i_data_sem); |
ef7f3835 | 4569 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
617ba13b | 4570 | ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
4571 | |
4572 | /* | |
4573 | * In a multi-transaction truncate, we only make the final transaction | |
4574 | * synchronous | |
4575 | */ | |
4576 | if (IS_SYNC(inode)) | |
0390131b | 4577 | ext4_handle_sync(handle); |
ac27a0ec DK |
4578 | out_stop: |
4579 | /* | |
4580 | * If this was a simple ftruncate(), and the file will remain alive | |
4581 | * then we need to clear up the orphan record which we created above. | |
4582 | * However, if this was a real unlink then we were called by | |
617ba13b | 4583 | * ext4_delete_inode(), and we allow that function to clean up the |
ac27a0ec DK |
4584 | * orphan info for us. |
4585 | */ | |
4586 | if (inode->i_nlink) | |
617ba13b | 4587 | ext4_orphan_del(handle, inode); |
ac27a0ec | 4588 | |
617ba13b | 4589 | ext4_journal_stop(handle); |
0562e0ba | 4590 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
4591 | } |
4592 | ||
ac27a0ec | 4593 | /* |
617ba13b | 4594 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4595 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4596 | * data in memory that is needed to recreate the on-disk version of this | |
4597 | * inode. | |
4598 | */ | |
617ba13b MC |
4599 | static int __ext4_get_inode_loc(struct inode *inode, |
4600 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4601 | { |
240799cd TT |
4602 | struct ext4_group_desc *gdp; |
4603 | struct buffer_head *bh; | |
4604 | struct super_block *sb = inode->i_sb; | |
4605 | ext4_fsblk_t block; | |
4606 | int inodes_per_block, inode_offset; | |
4607 | ||
3a06d778 | 4608 | iloc->bh = NULL; |
240799cd TT |
4609 | if (!ext4_valid_inum(sb, inode->i_ino)) |
4610 | return -EIO; | |
ac27a0ec | 4611 | |
240799cd TT |
4612 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4613 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4614 | if (!gdp) | |
ac27a0ec DK |
4615 | return -EIO; |
4616 | ||
240799cd TT |
4617 | /* |
4618 | * Figure out the offset within the block group inode table | |
4619 | */ | |
00d09882 | 4620 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4621 | inode_offset = ((inode->i_ino - 1) % |
4622 | EXT4_INODES_PER_GROUP(sb)); | |
4623 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4624 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4625 | ||
4626 | bh = sb_getblk(sb, block); | |
ac27a0ec | 4627 | if (!bh) { |
c398eda0 TT |
4628 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4629 | "unable to read itable block"); | |
ac27a0ec DK |
4630 | return -EIO; |
4631 | } | |
4632 | if (!buffer_uptodate(bh)) { | |
4633 | lock_buffer(bh); | |
9c83a923 HK |
4634 | |
4635 | /* | |
4636 | * If the buffer has the write error flag, we have failed | |
4637 | * to write out another inode in the same block. In this | |
4638 | * case, we don't have to read the block because we may | |
4639 | * read the old inode data successfully. | |
4640 | */ | |
4641 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4642 | set_buffer_uptodate(bh); | |
4643 | ||
ac27a0ec DK |
4644 | if (buffer_uptodate(bh)) { |
4645 | /* someone brought it uptodate while we waited */ | |
4646 | unlock_buffer(bh); | |
4647 | goto has_buffer; | |
4648 | } | |
4649 | ||
4650 | /* | |
4651 | * If we have all information of the inode in memory and this | |
4652 | * is the only valid inode in the block, we need not read the | |
4653 | * block. | |
4654 | */ | |
4655 | if (in_mem) { | |
4656 | struct buffer_head *bitmap_bh; | |
240799cd | 4657 | int i, start; |
ac27a0ec | 4658 | |
240799cd | 4659 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4660 | |
240799cd TT |
4661 | /* Is the inode bitmap in cache? */ |
4662 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
4663 | if (!bitmap_bh) |
4664 | goto make_io; | |
4665 | ||
4666 | /* | |
4667 | * If the inode bitmap isn't in cache then the | |
4668 | * optimisation may end up performing two reads instead | |
4669 | * of one, so skip it. | |
4670 | */ | |
4671 | if (!buffer_uptodate(bitmap_bh)) { | |
4672 | brelse(bitmap_bh); | |
4673 | goto make_io; | |
4674 | } | |
240799cd | 4675 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4676 | if (i == inode_offset) |
4677 | continue; | |
617ba13b | 4678 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4679 | break; |
4680 | } | |
4681 | brelse(bitmap_bh); | |
240799cd | 4682 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4683 | /* all other inodes are free, so skip I/O */ |
4684 | memset(bh->b_data, 0, bh->b_size); | |
4685 | set_buffer_uptodate(bh); | |
4686 | unlock_buffer(bh); | |
4687 | goto has_buffer; | |
4688 | } | |
4689 | } | |
4690 | ||
4691 | make_io: | |
240799cd TT |
4692 | /* |
4693 | * If we need to do any I/O, try to pre-readahead extra | |
4694 | * blocks from the inode table. | |
4695 | */ | |
4696 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4697 | ext4_fsblk_t b, end, table; | |
4698 | unsigned num; | |
4699 | ||
4700 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4701 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
4702 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
4703 | if (table > b) | |
4704 | b = table; | |
4705 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
4706 | num = EXT4_INODES_PER_GROUP(sb); | |
4707 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4708 | EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) | |
560671a0 | 4709 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4710 | table += num / inodes_per_block; |
4711 | if (end > table) | |
4712 | end = table; | |
4713 | while (b <= end) | |
4714 | sb_breadahead(sb, b++); | |
4715 | } | |
4716 | ||
ac27a0ec DK |
4717 | /* |
4718 | * There are other valid inodes in the buffer, this inode | |
4719 | * has in-inode xattrs, or we don't have this inode in memory. | |
4720 | * Read the block from disk. | |
4721 | */ | |
0562e0ba | 4722 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4723 | get_bh(bh); |
4724 | bh->b_end_io = end_buffer_read_sync; | |
4725 | submit_bh(READ_META, bh); | |
4726 | wait_on_buffer(bh); | |
4727 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4728 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4729 | "unable to read itable block"); | |
ac27a0ec DK |
4730 | brelse(bh); |
4731 | return -EIO; | |
4732 | } | |
4733 | } | |
4734 | has_buffer: | |
4735 | iloc->bh = bh; | |
4736 | return 0; | |
4737 | } | |
4738 | ||
617ba13b | 4739 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4740 | { |
4741 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4742 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4743 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4744 | } |
4745 | ||
617ba13b | 4746 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4747 | { |
617ba13b | 4748 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
4749 | |
4750 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 4751 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 4752 | inode->i_flags |= S_SYNC; |
617ba13b | 4753 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 4754 | inode->i_flags |= S_APPEND; |
617ba13b | 4755 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 4756 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 4757 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 4758 | inode->i_flags |= S_NOATIME; |
617ba13b | 4759 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
4760 | inode->i_flags |= S_DIRSYNC; |
4761 | } | |
4762 | ||
ff9ddf7e JK |
4763 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4764 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4765 | { | |
84a8dce2 DM |
4766 | unsigned int vfs_fl; |
4767 | unsigned long old_fl, new_fl; | |
4768 | ||
4769 | do { | |
4770 | vfs_fl = ei->vfs_inode.i_flags; | |
4771 | old_fl = ei->i_flags; | |
4772 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4773 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4774 | EXT4_DIRSYNC_FL); | |
4775 | if (vfs_fl & S_SYNC) | |
4776 | new_fl |= EXT4_SYNC_FL; | |
4777 | if (vfs_fl & S_APPEND) | |
4778 | new_fl |= EXT4_APPEND_FL; | |
4779 | if (vfs_fl & S_IMMUTABLE) | |
4780 | new_fl |= EXT4_IMMUTABLE_FL; | |
4781 | if (vfs_fl & S_NOATIME) | |
4782 | new_fl |= EXT4_NOATIME_FL; | |
4783 | if (vfs_fl & S_DIRSYNC) | |
4784 | new_fl |= EXT4_DIRSYNC_FL; | |
4785 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4786 | } |
de9a55b8 | 4787 | |
0fc1b451 | 4788 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4789 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4790 | { |
4791 | blkcnt_t i_blocks ; | |
8180a562 AK |
4792 | struct inode *inode = &(ei->vfs_inode); |
4793 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4794 | |
4795 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4796 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4797 | /* we are using combined 48 bit field */ | |
4798 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4799 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4800 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4801 | /* i_blocks represent file system block size */ |
4802 | return i_blocks << (inode->i_blkbits - 9); | |
4803 | } else { | |
4804 | return i_blocks; | |
4805 | } | |
0fc1b451 AK |
4806 | } else { |
4807 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4808 | } | |
4809 | } | |
ff9ddf7e | 4810 | |
1d1fe1ee | 4811 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4812 | { |
617ba13b MC |
4813 | struct ext4_iloc iloc; |
4814 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4815 | struct ext4_inode_info *ei; |
1d1fe1ee | 4816 | struct inode *inode; |
b436b9be | 4817 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4818 | long ret; |
ac27a0ec DK |
4819 | int block; |
4820 | ||
1d1fe1ee DH |
4821 | inode = iget_locked(sb, ino); |
4822 | if (!inode) | |
4823 | return ERR_PTR(-ENOMEM); | |
4824 | if (!(inode->i_state & I_NEW)) | |
4825 | return inode; | |
4826 | ||
4827 | ei = EXT4_I(inode); | |
7dc57615 | 4828 | iloc.bh = NULL; |
ac27a0ec | 4829 | |
1d1fe1ee DH |
4830 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4831 | if (ret < 0) | |
ac27a0ec | 4832 | goto bad_inode; |
617ba13b | 4833 | raw_inode = ext4_raw_inode(&iloc); |
ac27a0ec DK |
4834 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
4835 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
4836 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4837 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
4838 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4839 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
4840 | } | |
4841 | inode->i_nlink = le16_to_cpu(raw_inode->i_links_count); | |
ac27a0ec | 4842 | |
353eb83c | 4843 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
ac27a0ec DK |
4844 | ei->i_dir_start_lookup = 0; |
4845 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4846 | /* We now have enough fields to check if the inode was active or not. | |
4847 | * This is needed because nfsd might try to access dead inodes | |
4848 | * the test is that same one that e2fsck uses | |
4849 | * NeilBrown 1999oct15 | |
4850 | */ | |
4851 | if (inode->i_nlink == 0) { | |
4852 | if (inode->i_mode == 0 || | |
617ba13b | 4853 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 4854 | /* this inode is deleted */ |
1d1fe1ee | 4855 | ret = -ESTALE; |
ac27a0ec DK |
4856 | goto bad_inode; |
4857 | } | |
4858 | /* The only unlinked inodes we let through here have | |
4859 | * valid i_mode and are being read by the orphan | |
4860 | * recovery code: that's fine, we're about to complete | |
4861 | * the process of deleting those. */ | |
4862 | } | |
ac27a0ec | 4863 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4864 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4865 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4866 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4867 | ei->i_file_acl |= |
4868 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4869 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4870 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4871 | #ifdef CONFIG_QUOTA |
4872 | ei->i_reserved_quota = 0; | |
4873 | #endif | |
ac27a0ec DK |
4874 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4875 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4876 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4877 | /* |
4878 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4879 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4880 | */ | |
617ba13b | 4881 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4882 | ei->i_data[block] = raw_inode->i_block[block]; |
4883 | INIT_LIST_HEAD(&ei->i_orphan); | |
4884 | ||
b436b9be JK |
4885 | /* |
4886 | * Set transaction id's of transactions that have to be committed | |
4887 | * to finish f[data]sync. We set them to currently running transaction | |
4888 | * as we cannot be sure that the inode or some of its metadata isn't | |
4889 | * part of the transaction - the inode could have been reclaimed and | |
4890 | * now it is reread from disk. | |
4891 | */ | |
4892 | if (journal) { | |
4893 | transaction_t *transaction; | |
4894 | tid_t tid; | |
4895 | ||
a931da6a | 4896 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4897 | if (journal->j_running_transaction) |
4898 | transaction = journal->j_running_transaction; | |
4899 | else | |
4900 | transaction = journal->j_committing_transaction; | |
4901 | if (transaction) | |
4902 | tid = transaction->t_tid; | |
4903 | else | |
4904 | tid = journal->j_commit_sequence; | |
a931da6a | 4905 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4906 | ei->i_sync_tid = tid; |
4907 | ei->i_datasync_tid = tid; | |
4908 | } | |
4909 | ||
0040d987 | 4910 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec | 4911 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
617ba13b | 4912 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
e5d2861f | 4913 | EXT4_INODE_SIZE(inode->i_sb)) { |
1d1fe1ee | 4914 | ret = -EIO; |
ac27a0ec | 4915 | goto bad_inode; |
e5d2861f | 4916 | } |
ac27a0ec DK |
4917 | if (ei->i_extra_isize == 0) { |
4918 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4919 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4920 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
4921 | } else { |
4922 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 4923 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 4924 | ei->i_extra_isize; |
617ba13b | 4925 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 4926 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec DK |
4927 | } |
4928 | } else | |
4929 | ei->i_extra_isize = 0; | |
4930 | ||
ef7f3835 KS |
4931 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4932 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4933 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4934 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4935 | ||
25ec56b5 JNC |
4936 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4937 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4938 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4939 | inode->i_version |= | |
4940 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4941 | } | |
4942 | ||
c4b5a614 | 4943 | ret = 0; |
485c26ec | 4944 | if (ei->i_file_acl && |
1032988c | 4945 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4946 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4947 | ei->i_file_acl); | |
485c26ec TT |
4948 | ret = -EIO; |
4949 | goto bad_inode; | |
07a03824 | 4950 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
4951 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
4952 | (S_ISLNK(inode->i_mode) && | |
4953 | !ext4_inode_is_fast_symlink(inode))) | |
4954 | /* Validate extent which is part of inode */ | |
4955 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 4956 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
4957 | (S_ISLNK(inode->i_mode) && |
4958 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 4959 | /* Validate block references which are part of inode */ |
fe2c8191 TN |
4960 | ret = ext4_check_inode_blockref(inode); |
4961 | } | |
567f3e9a | 4962 | if (ret) |
de9a55b8 | 4963 | goto bad_inode; |
7a262f7c | 4964 | |
ac27a0ec | 4965 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4966 | inode->i_op = &ext4_file_inode_operations; |
4967 | inode->i_fop = &ext4_file_operations; | |
4968 | ext4_set_aops(inode); | |
ac27a0ec | 4969 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4970 | inode->i_op = &ext4_dir_inode_operations; |
4971 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4972 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4973 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4974 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4975 | nd_terminate_link(ei->i_data, inode->i_size, |
4976 | sizeof(ei->i_data) - 1); | |
4977 | } else { | |
617ba13b MC |
4978 | inode->i_op = &ext4_symlink_inode_operations; |
4979 | ext4_set_aops(inode); | |
ac27a0ec | 4980 | } |
563bdd61 TT |
4981 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4982 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4983 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4984 | if (raw_inode->i_block[0]) |
4985 | init_special_inode(inode, inode->i_mode, | |
4986 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4987 | else | |
4988 | init_special_inode(inode, inode->i_mode, | |
4989 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 4990 | } else { |
563bdd61 | 4991 | ret = -EIO; |
24676da4 | 4992 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4993 | goto bad_inode; |
ac27a0ec | 4994 | } |
af5bc92d | 4995 | brelse(iloc.bh); |
617ba13b | 4996 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4997 | unlock_new_inode(inode); |
4998 | return inode; | |
ac27a0ec DK |
4999 | |
5000 | bad_inode: | |
567f3e9a | 5001 | brelse(iloc.bh); |
1d1fe1ee DH |
5002 | iget_failed(inode); |
5003 | return ERR_PTR(ret); | |
ac27a0ec DK |
5004 | } |
5005 | ||
0fc1b451 AK |
5006 | static int ext4_inode_blocks_set(handle_t *handle, |
5007 | struct ext4_inode *raw_inode, | |
5008 | struct ext4_inode_info *ei) | |
5009 | { | |
5010 | struct inode *inode = &(ei->vfs_inode); | |
5011 | u64 i_blocks = inode->i_blocks; | |
5012 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
5013 | |
5014 | if (i_blocks <= ~0U) { | |
5015 | /* | |
5016 | * i_blocks can be represnted in a 32 bit variable | |
5017 | * as multiple of 512 bytes | |
5018 | */ | |
8180a562 | 5019 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5020 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 5021 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
5022 | return 0; |
5023 | } | |
5024 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
5025 | return -EFBIG; | |
5026 | ||
5027 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
5028 | /* |
5029 | * i_blocks can be represented in a 48 bit variable | |
5030 | * as multiple of 512 bytes | |
5031 | */ | |
8180a562 | 5032 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5033 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 5034 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 5035 | } else { |
84a8dce2 | 5036 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
5037 | /* i_block is stored in file system block size */ |
5038 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
5039 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
5040 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 5041 | } |
f287a1a5 | 5042 | return 0; |
0fc1b451 AK |
5043 | } |
5044 | ||
ac27a0ec DK |
5045 | /* |
5046 | * Post the struct inode info into an on-disk inode location in the | |
5047 | * buffer-cache. This gobbles the caller's reference to the | |
5048 | * buffer_head in the inode location struct. | |
5049 | * | |
5050 | * The caller must have write access to iloc->bh. | |
5051 | */ | |
617ba13b | 5052 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5053 | struct inode *inode, |
830156c7 | 5054 | struct ext4_iloc *iloc) |
ac27a0ec | 5055 | { |
617ba13b MC |
5056 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5057 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
5058 | struct buffer_head *bh = iloc->bh; |
5059 | int err = 0, rc, block; | |
5060 | ||
5061 | /* For fields not not tracking in the in-memory inode, | |
5062 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 5063 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5064 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 5065 | |
ff9ddf7e | 5066 | ext4_get_inode_flags(ei); |
ac27a0ec | 5067 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 5068 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
5069 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
5070 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
5071 | /* | |
5072 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5073 | * re-used with the upper 16 bits of the uid/gid intact | |
5074 | */ | |
af5bc92d | 5075 | if (!ei->i_dtime) { |
ac27a0ec DK |
5076 | raw_inode->i_uid_high = |
5077 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
5078 | raw_inode->i_gid_high = | |
5079 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
5080 | } else { | |
5081 | raw_inode->i_uid_high = 0; | |
5082 | raw_inode->i_gid_high = 0; | |
5083 | } | |
5084 | } else { | |
5085 | raw_inode->i_uid_low = | |
5086 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
5087 | raw_inode->i_gid_low = | |
5088 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
5089 | raw_inode->i_uid_high = 0; | |
5090 | raw_inode->i_gid_high = 0; | |
5091 | } | |
5092 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5093 | |
5094 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5095 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5096 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5097 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5098 | ||
0fc1b451 AK |
5099 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
5100 | goto out_brelse; | |
ac27a0ec | 5101 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5102 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
5103 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
5104 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
5105 | raw_inode->i_file_acl_high = |
5106 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5107 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
5108 | ext4_isize_set(raw_inode, ei->i_disksize); |
5109 | if (ei->i_disksize > 0x7fffffffULL) { | |
5110 | struct super_block *sb = inode->i_sb; | |
5111 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
5112 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
5113 | EXT4_SB(sb)->s_es->s_rev_level == | |
5114 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
5115 | /* If this is the first large file | |
5116 | * created, add a flag to the superblock. | |
5117 | */ | |
5118 | err = ext4_journal_get_write_access(handle, | |
5119 | EXT4_SB(sb)->s_sbh); | |
5120 | if (err) | |
5121 | goto out_brelse; | |
5122 | ext4_update_dynamic_rev(sb); | |
5123 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 5124 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
a48380f7 | 5125 | sb->s_dirt = 1; |
0390131b | 5126 | ext4_handle_sync(handle); |
73b50c1c | 5127 | err = ext4_handle_dirty_metadata(handle, NULL, |
a48380f7 | 5128 | EXT4_SB(sb)->s_sbh); |
ac27a0ec DK |
5129 | } |
5130 | } | |
5131 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5132 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5133 | if (old_valid_dev(inode->i_rdev)) { | |
5134 | raw_inode->i_block[0] = | |
5135 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5136 | raw_inode->i_block[1] = 0; | |
5137 | } else { | |
5138 | raw_inode->i_block[0] = 0; | |
5139 | raw_inode->i_block[1] = | |
5140 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5141 | raw_inode->i_block[2] = 0; | |
5142 | } | |
de9a55b8 TT |
5143 | } else |
5144 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
5145 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 5146 | |
25ec56b5 JNC |
5147 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
5148 | if (ei->i_extra_isize) { | |
5149 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5150 | raw_inode->i_version_hi = | |
5151 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 5152 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
5153 | } |
5154 | ||
830156c7 | 5155 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5156 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5157 | if (!err) |
5158 | err = rc; | |
19f5fb7a | 5159 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 5160 | |
b436b9be | 5161 | ext4_update_inode_fsync_trans(handle, inode, 0); |
ac27a0ec | 5162 | out_brelse: |
af5bc92d | 5163 | brelse(bh); |
617ba13b | 5164 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5165 | return err; |
5166 | } | |
5167 | ||
5168 | /* | |
617ba13b | 5169 | * ext4_write_inode() |
ac27a0ec DK |
5170 | * |
5171 | * We are called from a few places: | |
5172 | * | |
5173 | * - Within generic_file_write() for O_SYNC files. | |
5174 | * Here, there will be no transaction running. We wait for any running | |
5175 | * trasnaction to commit. | |
5176 | * | |
5177 | * - Within sys_sync(), kupdate and such. | |
5178 | * We wait on commit, if tol to. | |
5179 | * | |
5180 | * - Within prune_icache() (PF_MEMALLOC == true) | |
5181 | * Here we simply return. We can't afford to block kswapd on the | |
5182 | * journal commit. | |
5183 | * | |
5184 | * In all cases it is actually safe for us to return without doing anything, | |
5185 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 5186 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
5187 | * knfsd. |
5188 | * | |
5189 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5190 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5191 | * which we are interested. | |
5192 | * | |
5193 | * It would be a bug for them to not do this. The code: | |
5194 | * | |
5195 | * mark_inode_dirty(inode) | |
5196 | * stuff(); | |
5197 | * inode->i_size = expr; | |
5198 | * | |
5199 | * is in error because a kswapd-driven write_inode() could occur while | |
5200 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
5201 | * will no longer be on the superblock's dirty inode list. | |
5202 | */ | |
a9185b41 | 5203 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5204 | { |
91ac6f43 FM |
5205 | int err; |
5206 | ||
ac27a0ec DK |
5207 | if (current->flags & PF_MEMALLOC) |
5208 | return 0; | |
5209 | ||
91ac6f43 FM |
5210 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5211 | if (ext4_journal_current_handle()) { | |
5212 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5213 | dump_stack(); | |
5214 | return -EIO; | |
5215 | } | |
ac27a0ec | 5216 | |
a9185b41 | 5217 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
5218 | return 0; |
5219 | ||
5220 | err = ext4_force_commit(inode->i_sb); | |
5221 | } else { | |
5222 | struct ext4_iloc iloc; | |
ac27a0ec | 5223 | |
8b472d73 | 5224 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5225 | if (err) |
5226 | return err; | |
a9185b41 | 5227 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
5228 | sync_dirty_buffer(iloc.bh); |
5229 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5230 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5231 | "IO error syncing inode"); | |
830156c7 FM |
5232 | err = -EIO; |
5233 | } | |
fd2dd9fb | 5234 | brelse(iloc.bh); |
91ac6f43 FM |
5235 | } |
5236 | return err; | |
ac27a0ec DK |
5237 | } |
5238 | ||
5239 | /* | |
617ba13b | 5240 | * ext4_setattr() |
ac27a0ec DK |
5241 | * |
5242 | * Called from notify_change. | |
5243 | * | |
5244 | * We want to trap VFS attempts to truncate the file as soon as | |
5245 | * possible. In particular, we want to make sure that when the VFS | |
5246 | * shrinks i_size, we put the inode on the orphan list and modify | |
5247 | * i_disksize immediately, so that during the subsequent flushing of | |
5248 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5249 | * commit will leave the blocks being flushed in an unused state on | |
5250 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5251 | * be freed, so we have a strong guarantee that no future commit will | |
5252 | * leave these blocks visible to the user.) | |
5253 | * | |
678aaf48 JK |
5254 | * Another thing we have to assure is that if we are in ordered mode |
5255 | * and inode is still attached to the committing transaction, we must | |
5256 | * we start writeout of all the dirty pages which are being truncated. | |
5257 | * This way we are sure that all the data written in the previous | |
5258 | * transaction are already on disk (truncate waits for pages under | |
5259 | * writeback). | |
5260 | * | |
5261 | * Called with inode->i_mutex down. | |
ac27a0ec | 5262 | */ |
617ba13b | 5263 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
5264 | { |
5265 | struct inode *inode = dentry->d_inode; | |
5266 | int error, rc = 0; | |
3d287de3 | 5267 | int orphan = 0; |
ac27a0ec DK |
5268 | const unsigned int ia_valid = attr->ia_valid; |
5269 | ||
5270 | error = inode_change_ok(inode, attr); | |
5271 | if (error) | |
5272 | return error; | |
5273 | ||
12755627 | 5274 | if (is_quota_modification(inode, attr)) |
871a2931 | 5275 | dquot_initialize(inode); |
ac27a0ec DK |
5276 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
5277 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
5278 | handle_t *handle; | |
5279 | ||
5280 | /* (user+group)*(old+new) structure, inode write (sb, | |
5281 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 5282 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 5283 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
5284 | if (IS_ERR(handle)) { |
5285 | error = PTR_ERR(handle); | |
5286 | goto err_out; | |
5287 | } | |
b43fa828 | 5288 | error = dquot_transfer(inode, attr); |
ac27a0ec | 5289 | if (error) { |
617ba13b | 5290 | ext4_journal_stop(handle); |
ac27a0ec DK |
5291 | return error; |
5292 | } | |
5293 | /* Update corresponding info in inode so that everything is in | |
5294 | * one transaction */ | |
5295 | if (attr->ia_valid & ATTR_UID) | |
5296 | inode->i_uid = attr->ia_uid; | |
5297 | if (attr->ia_valid & ATTR_GID) | |
5298 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5299 | error = ext4_mark_inode_dirty(handle, inode); |
5300 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5301 | } |
5302 | ||
e2b46574 | 5303 | if (attr->ia_valid & ATTR_SIZE) { |
12e9b892 | 5304 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5305 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5306 | ||
0c095c7f TT |
5307 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5308 | return -EFBIG; | |
e2b46574 ES |
5309 | } |
5310 | } | |
5311 | ||
ac27a0ec | 5312 | if (S_ISREG(inode->i_mode) && |
c8d46e41 JZ |
5313 | attr->ia_valid & ATTR_SIZE && |
5314 | (attr->ia_size < inode->i_size || | |
12e9b892 | 5315 | (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) { |
ac27a0ec DK |
5316 | handle_t *handle; |
5317 | ||
617ba13b | 5318 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
5319 | if (IS_ERR(handle)) { |
5320 | error = PTR_ERR(handle); | |
5321 | goto err_out; | |
5322 | } | |
3d287de3 DM |
5323 | if (ext4_handle_valid(handle)) { |
5324 | error = ext4_orphan_add(handle, inode); | |
5325 | orphan = 1; | |
5326 | } | |
617ba13b MC |
5327 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5328 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
5329 | if (!error) |
5330 | error = rc; | |
617ba13b | 5331 | ext4_journal_stop(handle); |
678aaf48 JK |
5332 | |
5333 | if (ext4_should_order_data(inode)) { | |
5334 | error = ext4_begin_ordered_truncate(inode, | |
5335 | attr->ia_size); | |
5336 | if (error) { | |
5337 | /* Do as much error cleanup as possible */ | |
5338 | handle = ext4_journal_start(inode, 3); | |
5339 | if (IS_ERR(handle)) { | |
5340 | ext4_orphan_del(NULL, inode); | |
5341 | goto err_out; | |
5342 | } | |
5343 | ext4_orphan_del(handle, inode); | |
3d287de3 | 5344 | orphan = 0; |
678aaf48 JK |
5345 | ext4_journal_stop(handle); |
5346 | goto err_out; | |
5347 | } | |
5348 | } | |
c8d46e41 | 5349 | /* ext4_truncate will clear the flag */ |
12e9b892 | 5350 | if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))) |
c8d46e41 | 5351 | ext4_truncate(inode); |
ac27a0ec DK |
5352 | } |
5353 | ||
1025774c CH |
5354 | if ((attr->ia_valid & ATTR_SIZE) && |
5355 | attr->ia_size != i_size_read(inode)) | |
5356 | rc = vmtruncate(inode, attr->ia_size); | |
ac27a0ec | 5357 | |
1025774c CH |
5358 | if (!rc) { |
5359 | setattr_copy(inode, attr); | |
5360 | mark_inode_dirty(inode); | |
5361 | } | |
5362 | ||
5363 | /* | |
5364 | * If the call to ext4_truncate failed to get a transaction handle at | |
5365 | * all, we need to clean up the in-core orphan list manually. | |
5366 | */ | |
3d287de3 | 5367 | if (orphan && inode->i_nlink) |
617ba13b | 5368 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
5369 | |
5370 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 5371 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
5372 | |
5373 | err_out: | |
617ba13b | 5374 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5375 | if (!error) |
5376 | error = rc; | |
5377 | return error; | |
5378 | } | |
5379 | ||
3e3398a0 MC |
5380 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
5381 | struct kstat *stat) | |
5382 | { | |
5383 | struct inode *inode; | |
5384 | unsigned long delalloc_blocks; | |
5385 | ||
5386 | inode = dentry->d_inode; | |
5387 | generic_fillattr(inode, stat); | |
5388 | ||
5389 | /* | |
5390 | * We can't update i_blocks if the block allocation is delayed | |
5391 | * otherwise in the case of system crash before the real block | |
5392 | * allocation is done, we will have i_blocks inconsistent with | |
5393 | * on-disk file blocks. | |
5394 | * We always keep i_blocks updated together with real | |
5395 | * allocation. But to not confuse with user, stat | |
5396 | * will return the blocks that include the delayed allocation | |
5397 | * blocks for this file. | |
5398 | */ | |
3e3398a0 | 5399 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; |
3e3398a0 MC |
5400 | |
5401 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
5402 | return 0; | |
5403 | } | |
ac27a0ec | 5404 | |
a02908f1 MC |
5405 | static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks, |
5406 | int chunk) | |
5407 | { | |
5408 | int indirects; | |
5409 | ||
5410 | /* if nrblocks are contiguous */ | |
5411 | if (chunk) { | |
5412 | /* | |
5b41395f YY |
5413 | * With N contiguous data blocks, we need at most |
5414 | * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks, | |
5415 | * 2 dindirect blocks, and 1 tindirect block | |
a02908f1 | 5416 | */ |
5b41395f YY |
5417 | return DIV_ROUND_UP(nrblocks, |
5418 | EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4; | |
a02908f1 MC |
5419 | } |
5420 | /* | |
5421 | * if nrblocks are not contiguous, worse case, each block touch | |
5422 | * a indirect block, and each indirect block touch a double indirect | |
5423 | * block, plus a triple indirect block | |
5424 | */ | |
5425 | indirects = nrblocks * 2 + 1; | |
5426 | return indirects; | |
5427 | } | |
5428 | ||
5429 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) | |
5430 | { | |
12e9b892 | 5431 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
ac51d837 TT |
5432 | return ext4_indirect_trans_blocks(inode, nrblocks, chunk); |
5433 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); | |
a02908f1 | 5434 | } |
ac51d837 | 5435 | |
ac27a0ec | 5436 | /* |
a02908f1 MC |
5437 | * Account for index blocks, block groups bitmaps and block group |
5438 | * descriptor blocks if modify datablocks and index blocks | |
5439 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5440 | * |
a02908f1 | 5441 | * If datablocks are discontiguous, they are possible to spread over |
af901ca1 | 5442 | * different block groups too. If they are contiuguous, with flexbg, |
a02908f1 | 5443 | * they could still across block group boundary. |
ac27a0ec | 5444 | * |
a02908f1 MC |
5445 | * Also account for superblock, inode, quota and xattr blocks |
5446 | */ | |
1f109d5a | 5447 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 5448 | { |
8df9675f TT |
5449 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5450 | int gdpblocks; | |
a02908f1 MC |
5451 | int idxblocks; |
5452 | int ret = 0; | |
5453 | ||
5454 | /* | |
5455 | * How many index blocks need to touch to modify nrblocks? | |
5456 | * The "Chunk" flag indicating whether the nrblocks is | |
5457 | * physically contiguous on disk | |
5458 | * | |
5459 | * For Direct IO and fallocate, they calls get_block to allocate | |
5460 | * one single extent at a time, so they could set the "Chunk" flag | |
5461 | */ | |
5462 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
5463 | ||
5464 | ret = idxblocks; | |
5465 | ||
5466 | /* | |
5467 | * Now let's see how many group bitmaps and group descriptors need | |
5468 | * to account | |
5469 | */ | |
5470 | groups = idxblocks; | |
5471 | if (chunk) | |
5472 | groups += 1; | |
5473 | else | |
5474 | groups += nrblocks; | |
5475 | ||
5476 | gdpblocks = groups; | |
8df9675f TT |
5477 | if (groups > ngroups) |
5478 | groups = ngroups; | |
a02908f1 MC |
5479 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5480 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5481 | ||
5482 | /* bitmaps and block group descriptor blocks */ | |
5483 | ret += groups + gdpblocks; | |
5484 | ||
5485 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5486 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5487 | ||
5488 | return ret; | |
5489 | } | |
5490 | ||
5491 | /* | |
25985edc | 5492 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5493 | * the modification of a single pages into a single transaction, |
5494 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5495 | * |
525f4ed8 | 5496 | * This could be called via ext4_write_begin() |
ac27a0ec | 5497 | * |
525f4ed8 | 5498 | * We need to consider the worse case, when |
a02908f1 | 5499 | * one new block per extent. |
ac27a0ec | 5500 | */ |
a86c6181 | 5501 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5502 | { |
617ba13b | 5503 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5504 | int ret; |
5505 | ||
a02908f1 | 5506 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 5507 | |
a02908f1 | 5508 | /* Account for data blocks for journalled mode */ |
617ba13b | 5509 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5510 | ret += bpp; |
ac27a0ec DK |
5511 | return ret; |
5512 | } | |
f3bd1f3f MC |
5513 | |
5514 | /* | |
5515 | * Calculate the journal credits for a chunk of data modification. | |
5516 | * | |
5517 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5518 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5519 | * |
5520 | * journal buffers for data blocks are not included here, as DIO | |
5521 | * and fallocate do no need to journal data buffers. | |
5522 | */ | |
5523 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5524 | { | |
5525 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5526 | } | |
5527 | ||
ac27a0ec | 5528 | /* |
617ba13b | 5529 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5530 | * Give this, we know that the caller already has write access to iloc->bh. |
5531 | */ | |
617ba13b | 5532 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5533 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5534 | { |
5535 | int err = 0; | |
5536 | ||
25ec56b5 JNC |
5537 | if (test_opt(inode->i_sb, I_VERSION)) |
5538 | inode_inc_iversion(inode); | |
5539 | ||
ac27a0ec DK |
5540 | /* the do_update_inode consumes one bh->b_count */ |
5541 | get_bh(iloc->bh); | |
5542 | ||
dab291af | 5543 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5544 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5545 | put_bh(iloc->bh); |
5546 | return err; | |
5547 | } | |
5548 | ||
5549 | /* | |
5550 | * On success, We end up with an outstanding reference count against | |
5551 | * iloc->bh. This _must_ be cleaned up later. | |
5552 | */ | |
5553 | ||
5554 | int | |
617ba13b MC |
5555 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5556 | struct ext4_iloc *iloc) | |
ac27a0ec | 5557 | { |
0390131b FM |
5558 | int err; |
5559 | ||
5560 | err = ext4_get_inode_loc(inode, iloc); | |
5561 | if (!err) { | |
5562 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5563 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5564 | if (err) { | |
5565 | brelse(iloc->bh); | |
5566 | iloc->bh = NULL; | |
ac27a0ec DK |
5567 | } |
5568 | } | |
617ba13b | 5569 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5570 | return err; |
5571 | } | |
5572 | ||
6dd4ee7c KS |
5573 | /* |
5574 | * Expand an inode by new_extra_isize bytes. | |
5575 | * Returns 0 on success or negative error number on failure. | |
5576 | */ | |
1d03ec98 AK |
5577 | static int ext4_expand_extra_isize(struct inode *inode, |
5578 | unsigned int new_extra_isize, | |
5579 | struct ext4_iloc iloc, | |
5580 | handle_t *handle) | |
6dd4ee7c KS |
5581 | { |
5582 | struct ext4_inode *raw_inode; | |
5583 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5584 | |
5585 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5586 | return 0; | |
5587 | ||
5588 | raw_inode = ext4_raw_inode(&iloc); | |
5589 | ||
5590 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5591 | |
5592 | /* No extended attributes present */ | |
19f5fb7a TT |
5593 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5594 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5595 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5596 | new_extra_isize); | |
5597 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5598 | return 0; | |
5599 | } | |
5600 | ||
5601 | /* try to expand with EAs present */ | |
5602 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5603 | raw_inode, handle); | |
5604 | } | |
5605 | ||
ac27a0ec DK |
5606 | /* |
5607 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5608 | * dirtiness (it may still be data-dirty). | |
5609 | * This means that the in-core inode may be reaped by prune_icache | |
5610 | * without having to perform any I/O. This is a very good thing, | |
5611 | * because *any* task may call prune_icache - even ones which | |
5612 | * have a transaction open against a different journal. | |
5613 | * | |
5614 | * Is this cheating? Not really. Sure, we haven't written the | |
5615 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5616 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5617 | * we start and wait on commits. | |
5618 | * | |
5619 | * Is this efficient/effective? Well, we're being nice to the system | |
5620 | * by cleaning up our inodes proactively so they can be reaped | |
5621 | * without I/O. But we are potentially leaving up to five seconds' | |
5622 | * worth of inodes floating about which prune_icache wants us to | |
5623 | * write out. One way to fix that would be to get prune_icache() | |
5624 | * to do a write_super() to free up some memory. It has the desired | |
5625 | * effect. | |
5626 | */ | |
617ba13b | 5627 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5628 | { |
617ba13b | 5629 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5630 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5631 | static unsigned int mnt_count; | |
5632 | int err, ret; | |
ac27a0ec DK |
5633 | |
5634 | might_sleep(); | |
7ff9c073 | 5635 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5636 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5637 | if (ext4_handle_valid(handle) && |
5638 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5639 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5640 | /* |
5641 | * We need extra buffer credits since we may write into EA block | |
5642 | * with this same handle. If journal_extend fails, then it will | |
5643 | * only result in a minor loss of functionality for that inode. | |
5644 | * If this is felt to be critical, then e2fsck should be run to | |
5645 | * force a large enough s_min_extra_isize. | |
5646 | */ | |
5647 | if ((jbd2_journal_extend(handle, | |
5648 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5649 | ret = ext4_expand_extra_isize(inode, | |
5650 | sbi->s_want_extra_isize, | |
5651 | iloc, handle); | |
5652 | if (ret) { | |
19f5fb7a TT |
5653 | ext4_set_inode_state(inode, |
5654 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5655 | if (mnt_count != |
5656 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5657 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5658 | "Unable to expand inode %lu. Delete" |
5659 | " some EAs or run e2fsck.", | |
5660 | inode->i_ino); | |
c1bddad9 AK |
5661 | mnt_count = |
5662 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5663 | } |
5664 | } | |
5665 | } | |
5666 | } | |
ac27a0ec | 5667 | if (!err) |
617ba13b | 5668 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5669 | return err; |
5670 | } | |
5671 | ||
5672 | /* | |
617ba13b | 5673 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5674 | * |
5675 | * We're really interested in the case where a file is being extended. | |
5676 | * i_size has been changed by generic_commit_write() and we thus need | |
5677 | * to include the updated inode in the current transaction. | |
5678 | * | |
5dd4056d | 5679 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5680 | * are allocated to the file. |
5681 | * | |
5682 | * If the inode is marked synchronous, we don't honour that here - doing | |
5683 | * so would cause a commit on atime updates, which we don't bother doing. | |
5684 | * We handle synchronous inodes at the highest possible level. | |
5685 | */ | |
617ba13b | 5686 | void ext4_dirty_inode(struct inode *inode) |
ac27a0ec | 5687 | { |
ac27a0ec DK |
5688 | handle_t *handle; |
5689 | ||
617ba13b | 5690 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
5691 | if (IS_ERR(handle)) |
5692 | goto out; | |
f3dc272f | 5693 | |
f3dc272f CW |
5694 | ext4_mark_inode_dirty(handle, inode); |
5695 | ||
617ba13b | 5696 | ext4_journal_stop(handle); |
ac27a0ec DK |
5697 | out: |
5698 | return; | |
5699 | } | |
5700 | ||
5701 | #if 0 | |
5702 | /* | |
5703 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5704 | * it from being flushed to disk early. Unlike | |
617ba13b | 5705 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5706 | * returns no iloc structure, so the caller needs to repeat the iloc |
5707 | * lookup to mark the inode dirty later. | |
5708 | */ | |
617ba13b | 5709 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5710 | { |
617ba13b | 5711 | struct ext4_iloc iloc; |
ac27a0ec DK |
5712 | |
5713 | int err = 0; | |
5714 | if (handle) { | |
617ba13b | 5715 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5716 | if (!err) { |
5717 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5718 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5719 | if (!err) |
0390131b | 5720 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5721 | NULL, |
0390131b | 5722 | iloc.bh); |
ac27a0ec DK |
5723 | brelse(iloc.bh); |
5724 | } | |
5725 | } | |
617ba13b | 5726 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5727 | return err; |
5728 | } | |
5729 | #endif | |
5730 | ||
617ba13b | 5731 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5732 | { |
5733 | journal_t *journal; | |
5734 | handle_t *handle; | |
5735 | int err; | |
5736 | ||
5737 | /* | |
5738 | * We have to be very careful here: changing a data block's | |
5739 | * journaling status dynamically is dangerous. If we write a | |
5740 | * data block to the journal, change the status and then delete | |
5741 | * that block, we risk forgetting to revoke the old log record | |
5742 | * from the journal and so a subsequent replay can corrupt data. | |
5743 | * So, first we make sure that the journal is empty and that | |
5744 | * nobody is changing anything. | |
5745 | */ | |
5746 | ||
617ba13b | 5747 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5748 | if (!journal) |
5749 | return 0; | |
d699594d | 5750 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5751 | return -EROFS; |
5752 | ||
dab291af MC |
5753 | jbd2_journal_lock_updates(journal); |
5754 | jbd2_journal_flush(journal); | |
ac27a0ec DK |
5755 | |
5756 | /* | |
5757 | * OK, there are no updates running now, and all cached data is | |
5758 | * synced to disk. We are now in a completely consistent state | |
5759 | * which doesn't have anything in the journal, and we know that | |
5760 | * no filesystem updates are running, so it is safe to modify | |
5761 | * the inode's in-core data-journaling state flag now. | |
5762 | */ | |
5763 | ||
5764 | if (val) | |
12e9b892 | 5765 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
ac27a0ec | 5766 | else |
12e9b892 | 5767 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
617ba13b | 5768 | ext4_set_aops(inode); |
ac27a0ec | 5769 | |
dab291af | 5770 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
5771 | |
5772 | /* Finally we can mark the inode as dirty. */ | |
5773 | ||
617ba13b | 5774 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
5775 | if (IS_ERR(handle)) |
5776 | return PTR_ERR(handle); | |
5777 | ||
617ba13b | 5778 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5779 | ext4_handle_sync(handle); |
617ba13b MC |
5780 | ext4_journal_stop(handle); |
5781 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5782 | |
5783 | return err; | |
5784 | } | |
2e9ee850 AK |
5785 | |
5786 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5787 | { | |
5788 | return !buffer_mapped(bh); | |
5789 | } | |
5790 | ||
c2ec175c | 5791 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5792 | { |
c2ec175c | 5793 | struct page *page = vmf->page; |
2e9ee850 AK |
5794 | loff_t size; |
5795 | unsigned long len; | |
5796 | int ret = -EINVAL; | |
79f0be8d | 5797 | void *fsdata; |
2e9ee850 AK |
5798 | struct file *file = vma->vm_file; |
5799 | struct inode *inode = file->f_path.dentry->d_inode; | |
5800 | struct address_space *mapping = inode->i_mapping; | |
5801 | ||
5802 | /* | |
5803 | * Get i_alloc_sem to stop truncates messing with the inode. We cannot | |
5804 | * get i_mutex because we are already holding mmap_sem. | |
5805 | */ | |
5806 | down_read(&inode->i_alloc_sem); | |
5807 | size = i_size_read(inode); | |
5808 | if (page->mapping != mapping || size <= page_offset(page) | |
5809 | || !PageUptodate(page)) { | |
5810 | /* page got truncated from under us? */ | |
5811 | goto out_unlock; | |
5812 | } | |
5813 | ret = 0; | |
5814 | if (PageMappedToDisk(page)) | |
5815 | goto out_unlock; | |
5816 | ||
5817 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5818 | len = size & ~PAGE_CACHE_MASK; | |
5819 | else | |
5820 | len = PAGE_CACHE_SIZE; | |
5821 | ||
a827eaff AK |
5822 | lock_page(page); |
5823 | /* | |
5824 | * return if we have all the buffers mapped. This avoid | |
5825 | * the need to call write_begin/write_end which does a | |
5826 | * journal_start/journal_stop which can block and take | |
5827 | * long time | |
5828 | */ | |
2e9ee850 | 5829 | if (page_has_buffers(page)) { |
2e9ee850 | 5830 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff AK |
5831 | ext4_bh_unmapped)) { |
5832 | unlock_page(page); | |
2e9ee850 | 5833 | goto out_unlock; |
a827eaff | 5834 | } |
2e9ee850 | 5835 | } |
a827eaff | 5836 | unlock_page(page); |
2e9ee850 AK |
5837 | /* |
5838 | * OK, we need to fill the hole... Do write_begin write_end | |
5839 | * to do block allocation/reservation.We are not holding | |
5840 | * inode.i__mutex here. That allow * parallel write_begin, | |
5841 | * write_end call. lock_page prevent this from happening | |
5842 | * on the same page though | |
5843 | */ | |
5844 | ret = mapping->a_ops->write_begin(file, mapping, page_offset(page), | |
79f0be8d | 5845 | len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); |
2e9ee850 AK |
5846 | if (ret < 0) |
5847 | goto out_unlock; | |
5848 | ret = mapping->a_ops->write_end(file, mapping, page_offset(page), | |
79f0be8d | 5849 | len, len, page, fsdata); |
2e9ee850 AK |
5850 | if (ret < 0) |
5851 | goto out_unlock; | |
5852 | ret = 0; | |
5853 | out_unlock: | |
c2ec175c NP |
5854 | if (ret) |
5855 | ret = VM_FAULT_SIGBUS; | |
2e9ee850 AK |
5856 | up_read(&inode->i_alloc_sem); |
5857 | return ret; | |
5858 | } |