]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
ac27a0ec | 2 | /* |
617ba13b | 3 | * linux/fs/ext4/file.c |
ac27a0ec DK |
4 | * |
5 | * Copyright (C) 1992, 1993, 1994, 1995 | |
6 | * Remy Card (card@masi.ibp.fr) | |
7 | * Laboratoire MASI - Institut Blaise Pascal | |
8 | * Universite Pierre et Marie Curie (Paris VI) | |
9 | * | |
10 | * from | |
11 | * | |
12 | * linux/fs/minix/file.c | |
13 | * | |
14 | * Copyright (C) 1991, 1992 Linus Torvalds | |
15 | * | |
617ba13b | 16 | * ext4 fs regular file handling primitives |
ac27a0ec DK |
17 | * |
18 | * 64-bit file support on 64-bit platforms by Jakub Jelinek | |
19 | * (jj@sunsite.ms.mff.cuni.cz) | |
20 | */ | |
21 | ||
22 | #include <linux/time.h> | |
23 | #include <linux/fs.h> | |
545052e9 | 24 | #include <linux/iomap.h> |
bc0b0d6d TT |
25 | #include <linux/mount.h> |
26 | #include <linux/path.h> | |
c94c2acf | 27 | #include <linux/dax.h> |
871a2931 | 28 | #include <linux/quotaops.h> |
c8c0df24 | 29 | #include <linux/pagevec.h> |
e2e40f2c | 30 | #include <linux/uio.h> |
b8a6176c | 31 | #include <linux/mman.h> |
378f32ba | 32 | #include <linux/backing-dev.h> |
3dcf5451 CH |
33 | #include "ext4.h" |
34 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
35 | #include "xattr.h" |
36 | #include "acl.h" | |
569342dc | 37 | #include "truncate.h" |
ac27a0ec | 38 | |
b1b4705d MB |
39 | static bool ext4_dio_supported(struct inode *inode) |
40 | { | |
41 | if (IS_ENABLED(CONFIG_FS_ENCRYPTION) && IS_ENCRYPTED(inode)) | |
42 | return false; | |
43 | if (fsverity_active(inode)) | |
44 | return false; | |
45 | if (ext4_should_journal_data(inode)) | |
46 | return false; | |
47 | if (ext4_has_inline_data(inode)) | |
48 | return false; | |
49 | return true; | |
50 | } | |
51 | ||
52 | static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
53 | { | |
54 | ssize_t ret; | |
55 | struct inode *inode = file_inode(iocb->ki_filp); | |
56 | ||
57 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
58 | if (!inode_trylock_shared(inode)) | |
59 | return -EAGAIN; | |
60 | } else { | |
61 | inode_lock_shared(inode); | |
62 | } | |
63 | ||
64 | if (!ext4_dio_supported(inode)) { | |
65 | inode_unlock_shared(inode); | |
66 | /* | |
67 | * Fallback to buffered I/O if the operation being performed on | |
68 | * the inode is not supported by direct I/O. The IOCB_DIRECT | |
69 | * flag needs to be cleared here in order to ensure that the | |
70 | * direct I/O path within generic_file_read_iter() is not | |
71 | * taken. | |
72 | */ | |
73 | iocb->ki_flags &= ~IOCB_DIRECT; | |
74 | return generic_file_read_iter(iocb, to); | |
75 | } | |
76 | ||
2f632965 | 77 | ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL, 0); |
b1b4705d MB |
78 | inode_unlock_shared(inode); |
79 | ||
80 | file_accessed(iocb->ki_filp); | |
81 | return ret; | |
82 | } | |
83 | ||
364443cb JK |
84 | #ifdef CONFIG_FS_DAX |
85 | static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
86 | { | |
87 | struct inode *inode = file_inode(iocb->ki_filp); | |
88 | ssize_t ret; | |
89 | ||
f629afe3 RH |
90 | if (iocb->ki_flags & IOCB_NOWAIT) { |
91 | if (!inode_trylock_shared(inode)) | |
728fbc0e | 92 | return -EAGAIN; |
f629afe3 | 93 | } else { |
728fbc0e GR |
94 | inode_lock_shared(inode); |
95 | } | |
364443cb JK |
96 | /* |
97 | * Recheck under inode lock - at this point we are sure it cannot | |
98 | * change anymore | |
99 | */ | |
100 | if (!IS_DAX(inode)) { | |
101 | inode_unlock_shared(inode); | |
102 | /* Fallback to buffered IO in case we cannot support DAX */ | |
103 | return generic_file_read_iter(iocb, to); | |
104 | } | |
105 | ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops); | |
106 | inode_unlock_shared(inode); | |
107 | ||
108 | file_accessed(iocb->ki_filp); | |
109 | return ret; | |
110 | } | |
111 | #endif | |
112 | ||
113 | static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
114 | { | |
b1b4705d MB |
115 | struct inode *inode = file_inode(iocb->ki_filp); |
116 | ||
117 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
0db1ff22 TT |
118 | return -EIO; |
119 | ||
364443cb JK |
120 | if (!iov_iter_count(to)) |
121 | return 0; /* skip atime */ | |
122 | ||
123 | #ifdef CONFIG_FS_DAX | |
b1b4705d | 124 | if (IS_DAX(inode)) |
364443cb JK |
125 | return ext4_dax_read_iter(iocb, to); |
126 | #endif | |
b1b4705d MB |
127 | if (iocb->ki_flags & IOCB_DIRECT) |
128 | return ext4_dio_read_iter(iocb, to); | |
129 | ||
364443cb JK |
130 | return generic_file_read_iter(iocb, to); |
131 | } | |
132 | ||
ac27a0ec DK |
133 | /* |
134 | * Called when an inode is released. Note that this is different | |
617ba13b | 135 | * from ext4_file_open: open gets called at every open, but release |
ac27a0ec DK |
136 | * gets called only when /all/ the files are closed. |
137 | */ | |
af5bc92d | 138 | static int ext4_release_file(struct inode *inode, struct file *filp) |
ac27a0ec | 139 | { |
19f5fb7a | 140 | if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) { |
7d8f9f7d | 141 | ext4_alloc_da_blocks(inode); |
19f5fb7a | 142 | ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 143 | } |
ac27a0ec DK |
144 | /* if we are the last writer on the inode, drop the block reservation */ |
145 | if ((filp->f_mode & FMODE_WRITE) && | |
d6014301 | 146 | (atomic_read(&inode->i_writecount) == 1) && |
e030a288 | 147 | !EXT4_I(inode)->i_reserved_data_blocks) { |
0e855ac8 | 148 | down_write(&EXT4_I(inode)->i_data_sem); |
27bc446e | 149 | ext4_discard_preallocations(inode, 0); |
0e855ac8 | 150 | up_write(&EXT4_I(inode)->i_data_sem); |
ac27a0ec DK |
151 | } |
152 | if (is_dx(inode) && filp->private_data) | |
617ba13b | 153 | ext4_htree_free_dir_info(filp->private_data); |
ac27a0ec DK |
154 | |
155 | return 0; | |
156 | } | |
157 | ||
e9e3bcec ES |
158 | /* |
159 | * This tests whether the IO in question is block-aligned or not. | |
160 | * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they | |
161 | * are converted to written only after the IO is complete. Until they are | |
162 | * mapped, these blocks appear as holes, so dio_zero_block() will assume that | |
163 | * it needs to zero out portions of the start and/or end block. If 2 AIO | |
164 | * threads are at work on the same unwritten block, they must be synchronized | |
165 | * or one thread will zero the other's data, causing corruption. | |
166 | */ | |
aa9714d0 RH |
167 | static bool |
168 | ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos) | |
e9e3bcec ES |
169 | { |
170 | struct super_block *sb = inode->i_sb; | |
aa9714d0 | 171 | unsigned long blockmask = sb->s_blocksize - 1; |
e9e3bcec | 172 | |
9b884164 | 173 | if ((pos | iov_iter_alignment(from)) & blockmask) |
aa9714d0 | 174 | return true; |
e9e3bcec | 175 | |
aa9714d0 RH |
176 | return false; |
177 | } | |
178 | ||
179 | static bool | |
180 | ext4_extending_io(struct inode *inode, loff_t offset, size_t len) | |
181 | { | |
182 | if (offset + len > i_size_read(inode) || | |
183 | offset + len > EXT4_I(inode)->i_disksize) | |
184 | return true; | |
185 | return false; | |
e9e3bcec ES |
186 | } |
187 | ||
213bcd9c JK |
188 | /* Is IO overwriting allocated and initialized blocks? */ |
189 | static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len) | |
190 | { | |
191 | struct ext4_map_blocks map; | |
192 | unsigned int blkbits = inode->i_blkbits; | |
193 | int err, blklen; | |
194 | ||
195 | if (pos + len > i_size_read(inode)) | |
196 | return false; | |
197 | ||
198 | map.m_lblk = pos >> blkbits; | |
199 | map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits); | |
200 | blklen = map.m_len; | |
201 | ||
202 | err = ext4_map_blocks(NULL, inode, &map, 0); | |
203 | /* | |
204 | * 'err==len' means that all of the blocks have been preallocated, | |
205 | * regardless of whether they have been initialized or not. To exclude | |
206 | * unwritten extents, we need to check m_flags. | |
207 | */ | |
208 | return err == blklen && (map.m_flags & EXT4_MAP_MAPPED); | |
209 | } | |
210 | ||
aa9714d0 RH |
211 | static ssize_t ext4_generic_write_checks(struct kiocb *iocb, |
212 | struct iov_iter *from) | |
213bcd9c JK |
213 | { |
214 | struct inode *inode = file_inode(iocb->ki_filp); | |
215 | ssize_t ret; | |
216 | ||
378f32ba MB |
217 | if (unlikely(IS_IMMUTABLE(inode))) |
218 | return -EPERM; | |
219 | ||
213bcd9c JK |
220 | ret = generic_write_checks(iocb, from); |
221 | if (ret <= 0) | |
222 | return ret; | |
02b016ca | 223 | |
213bcd9c JK |
224 | /* |
225 | * If we have encountered a bitmap-format file, the size limit | |
226 | * is smaller than s_maxbytes, which is for extent-mapped files. | |
227 | */ | |
228 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
229 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
230 | ||
231 | if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) | |
232 | return -EFBIG; | |
233 | iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos); | |
234 | } | |
378f32ba | 235 | |
aa9714d0 RH |
236 | return iov_iter_count(from); |
237 | } | |
238 | ||
239 | static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from) | |
240 | { | |
241 | ssize_t ret, count; | |
242 | ||
243 | count = ext4_generic_write_checks(iocb, from); | |
244 | if (count <= 0) | |
245 | return count; | |
246 | ||
378f32ba MB |
247 | ret = file_modified(iocb->ki_filp); |
248 | if (ret) | |
249 | return ret; | |
aa9714d0 | 250 | return count; |
213bcd9c JK |
251 | } |
252 | ||
378f32ba MB |
253 | static ssize_t ext4_buffered_write_iter(struct kiocb *iocb, |
254 | struct iov_iter *from) | |
255 | { | |
256 | ssize_t ret; | |
257 | struct inode *inode = file_inode(iocb->ki_filp); | |
258 | ||
259 | if (iocb->ki_flags & IOCB_NOWAIT) | |
260 | return -EOPNOTSUPP; | |
261 | ||
aa75f4d3 | 262 | ext4_fc_start_update(inode); |
378f32ba MB |
263 | inode_lock(inode); |
264 | ret = ext4_write_checks(iocb, from); | |
265 | if (ret <= 0) | |
266 | goto out; | |
267 | ||
268 | current->backing_dev_info = inode_to_bdi(inode); | |
269 | ret = generic_perform_write(iocb->ki_filp, from, iocb->ki_pos); | |
270 | current->backing_dev_info = NULL; | |
271 | ||
272 | out: | |
273 | inode_unlock(inode); | |
aa75f4d3 | 274 | ext4_fc_stop_update(inode); |
378f32ba MB |
275 | if (likely(ret > 0)) { |
276 | iocb->ki_pos += ret; | |
277 | ret = generic_write_sync(iocb, ret); | |
278 | } | |
279 | ||
280 | return ret; | |
281 | } | |
282 | ||
569342dc MB |
283 | static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset, |
284 | ssize_t written, size_t count) | |
285 | { | |
286 | handle_t *handle; | |
287 | bool truncate = false; | |
288 | u8 blkbits = inode->i_blkbits; | |
289 | ext4_lblk_t written_blk, end_blk; | |
4209ae12 | 290 | int ret; |
569342dc MB |
291 | |
292 | /* | |
293 | * Note that EXT4_I(inode)->i_disksize can get extended up to | |
294 | * inode->i_size while the I/O was running due to writeback of delalloc | |
295 | * blocks. But, the code in ext4_iomap_alloc() is careful to use | |
296 | * zeroed/unwritten extents if this is possible; thus we won't leave | |
297 | * uninitialized blocks in a file even if we didn't succeed in writing | |
298 | * as much as we intended. | |
299 | */ | |
300 | WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize); | |
301 | if (offset + count <= EXT4_I(inode)->i_disksize) { | |
302 | /* | |
303 | * We need to ensure that the inode is removed from the orphan | |
304 | * list if it has been added prematurely, due to writeback of | |
305 | * delalloc blocks. | |
306 | */ | |
307 | if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) { | |
308 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
309 | ||
310 | if (IS_ERR(handle)) { | |
311 | ext4_orphan_del(NULL, inode); | |
312 | return PTR_ERR(handle); | |
313 | } | |
314 | ||
315 | ext4_orphan_del(handle, inode); | |
316 | ext4_journal_stop(handle); | |
317 | } | |
318 | ||
319 | return written; | |
320 | } | |
321 | ||
322 | if (written < 0) | |
323 | goto truncate; | |
324 | ||
325 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
326 | if (IS_ERR(handle)) { | |
327 | written = PTR_ERR(handle); | |
328 | goto truncate; | |
329 | } | |
330 | ||
4209ae12 HS |
331 | if (ext4_update_inode_size(inode, offset + written)) { |
332 | ret = ext4_mark_inode_dirty(handle, inode); | |
333 | if (unlikely(ret)) { | |
334 | written = ret; | |
335 | ext4_journal_stop(handle); | |
336 | goto truncate; | |
337 | } | |
338 | } | |
569342dc MB |
339 | |
340 | /* | |
341 | * We may need to truncate allocated but not written blocks beyond EOF. | |
342 | */ | |
343 | written_blk = ALIGN(offset + written, 1 << blkbits); | |
344 | end_blk = ALIGN(offset + count, 1 << blkbits); | |
345 | if (written_blk < end_blk && ext4_can_truncate(inode)) | |
346 | truncate = true; | |
347 | ||
348 | /* | |
349 | * Remove the inode from the orphan list if it has been extended and | |
350 | * everything went OK. | |
351 | */ | |
352 | if (!truncate && inode->i_nlink) | |
353 | ext4_orphan_del(handle, inode); | |
354 | ext4_journal_stop(handle); | |
355 | ||
356 | if (truncate) { | |
357 | truncate: | |
358 | ext4_truncate_failed_write(inode); | |
359 | /* | |
360 | * If the truncate operation failed early, then the inode may | |
361 | * still be on the orphan list. In that case, we need to try | |
362 | * remove the inode from the in-memory linked list. | |
363 | */ | |
364 | if (inode->i_nlink) | |
365 | ext4_orphan_del(NULL, inode); | |
366 | } | |
367 | ||
368 | return written; | |
369 | } | |
370 | ||
378f32ba MB |
371 | static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size, |
372 | int error, unsigned int flags) | |
373 | { | |
5899593f | 374 | loff_t pos = iocb->ki_pos; |
378f32ba MB |
375 | struct inode *inode = file_inode(iocb->ki_filp); |
376 | ||
377 | if (error) | |
378 | return error; | |
379 | ||
5899593f JK |
380 | if (size && flags & IOMAP_DIO_UNWRITTEN) { |
381 | error = ext4_convert_unwritten_extents(NULL, inode, pos, size); | |
382 | if (error < 0) | |
383 | return error; | |
384 | } | |
385 | /* | |
386 | * If we are extending the file, we have to update i_size here before | |
387 | * page cache gets invalidated in iomap_dio_rw(). Otherwise racing | |
388 | * buffered reads could zero out too much from page cache pages. Update | |
389 | * of on-disk size will happen later in ext4_dio_write_iter() where | |
390 | * we have enough information to also perform orphan list handling etc. | |
391 | * Note that we perform all extending writes synchronously under | |
392 | * i_rwsem held exclusively so i_size update is safe here in that case. | |
393 | * If the write was not extending, we cannot see pos > i_size here | |
394 | * because operations reducing i_size like truncate wait for all | |
395 | * outstanding DIO before updating i_size. | |
396 | */ | |
397 | pos += size; | |
398 | if (pos > i_size_read(inode)) | |
399 | i_size_write(inode, pos); | |
378f32ba MB |
400 | |
401 | return 0; | |
402 | } | |
403 | ||
404 | static const struct iomap_dio_ops ext4_dio_write_ops = { | |
405 | .end_io = ext4_dio_write_end_io, | |
406 | }; | |
407 | ||
aa9714d0 RH |
408 | /* |
409 | * The intention here is to start with shared lock acquired then see if any | |
410 | * condition requires an exclusive inode lock. If yes, then we restart the | |
411 | * whole operation by releasing the shared lock and acquiring exclusive lock. | |
412 | * | |
413 | * - For unaligned_io we never take shared lock as it may cause data corruption | |
414 | * when two unaligned IO tries to modify the same block e.g. while zeroing. | |
415 | * | |
416 | * - For extending writes case we don't take the shared lock, since it requires | |
417 | * updating inode i_disksize and/or orphan handling with exclusive lock. | |
418 | * | |
bc6385da RH |
419 | * - shared locking will only be true mostly with overwrites. Otherwise we will |
420 | * switch to exclusive i_rwsem lock. | |
aa9714d0 RH |
421 | */ |
422 | static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from, | |
423 | bool *ilock_shared, bool *extend) | |
424 | { | |
425 | struct file *file = iocb->ki_filp; | |
426 | struct inode *inode = file_inode(file); | |
427 | loff_t offset; | |
428 | size_t count; | |
429 | ssize_t ret; | |
430 | ||
431 | restart: | |
432 | ret = ext4_generic_write_checks(iocb, from); | |
433 | if (ret <= 0) | |
434 | goto out; | |
435 | ||
436 | offset = iocb->ki_pos; | |
437 | count = ret; | |
438 | if (ext4_extending_io(inode, offset, count)) | |
439 | *extend = true; | |
440 | /* | |
441 | * Determine whether the IO operation will overwrite allocated | |
bc6385da | 442 | * and initialized blocks. |
aa9714d0 RH |
443 | * We need exclusive i_rwsem for changing security info |
444 | * in file_modified(). | |
445 | */ | |
446 | if (*ilock_shared && (!IS_NOSEC(inode) || *extend || | |
aa9714d0 | 447 | !ext4_overwrite_io(inode, offset, count))) { |
0b3171b6 JK |
448 | if (iocb->ki_flags & IOCB_NOWAIT) { |
449 | ret = -EAGAIN; | |
450 | goto out; | |
451 | } | |
aa9714d0 RH |
452 | inode_unlock_shared(inode); |
453 | *ilock_shared = false; | |
454 | inode_lock(inode); | |
455 | goto restart; | |
456 | } | |
457 | ||
458 | ret = file_modified(file); | |
459 | if (ret < 0) | |
460 | goto out; | |
461 | ||
462 | return count; | |
463 | out: | |
464 | if (*ilock_shared) | |
465 | inode_unlock_shared(inode); | |
466 | else | |
467 | inode_unlock(inode); | |
468 | return ret; | |
469 | } | |
470 | ||
378f32ba MB |
471 | static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) |
472 | { | |
473 | ssize_t ret; | |
378f32ba MB |
474 | handle_t *handle; |
475 | struct inode *inode = file_inode(iocb->ki_filp); | |
aa9714d0 RH |
476 | loff_t offset = iocb->ki_pos; |
477 | size_t count = iov_iter_count(from); | |
8cd115bd | 478 | const struct iomap_ops *iomap_ops = &ext4_iomap_ops; |
aa9714d0 RH |
479 | bool extend = false, unaligned_io = false; |
480 | bool ilock_shared = true; | |
481 | ||
482 | /* | |
483 | * We initially start with shared inode lock unless it is | |
484 | * unaligned IO which needs exclusive lock anyways. | |
485 | */ | |
486 | if (ext4_unaligned_io(inode, from, offset)) { | |
487 | unaligned_io = true; | |
488 | ilock_shared = false; | |
489 | } | |
490 | /* | |
491 | * Quick check here without any i_rwsem lock to see if it is extending | |
492 | * IO. A more reliable check is done in ext4_dio_write_checks() with | |
493 | * proper locking in place. | |
494 | */ | |
495 | if (offset + count > i_size_read(inode)) | |
496 | ilock_shared = false; | |
378f32ba MB |
497 | |
498 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
aa9714d0 RH |
499 | if (ilock_shared) { |
500 | if (!inode_trylock_shared(inode)) | |
501 | return -EAGAIN; | |
502 | } else { | |
503 | if (!inode_trylock(inode)) | |
504 | return -EAGAIN; | |
505 | } | |
378f32ba | 506 | } else { |
aa9714d0 RH |
507 | if (ilock_shared) |
508 | inode_lock_shared(inode); | |
509 | else | |
510 | inode_lock(inode); | |
378f32ba MB |
511 | } |
512 | ||
aa9714d0 | 513 | /* Fallback to buffered I/O if the inode does not support direct I/O. */ |
378f32ba | 514 | if (!ext4_dio_supported(inode)) { |
aa9714d0 RH |
515 | if (ilock_shared) |
516 | inode_unlock_shared(inode); | |
517 | else | |
518 | inode_unlock(inode); | |
378f32ba MB |
519 | return ext4_buffered_write_iter(iocb, from); |
520 | } | |
521 | ||
aa9714d0 RH |
522 | ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend); |
523 | if (ret <= 0) | |
378f32ba | 524 | return ret; |
378f32ba | 525 | |
6e014c62 JA |
526 | /* if we're going to block and IOCB_NOWAIT is set, return -EAGAIN */ |
527 | if ((iocb->ki_flags & IOCB_NOWAIT) && (unaligned_io || extend)) { | |
528 | ret = -EAGAIN; | |
529 | goto out; | |
530 | } | |
531 | ||
378f32ba | 532 | offset = iocb->ki_pos; |
aa9714d0 | 533 | count = ret; |
378f32ba MB |
534 | |
535 | /* | |
aa9714d0 RH |
536 | * Unaligned direct IO must be serialized among each other as zeroing |
537 | * of partial blocks of two competing unaligned IOs can result in data | |
538 | * corruption. | |
539 | * | |
540 | * So we make sure we don't allow any unaligned IO in flight. | |
541 | * For IOs where we need not wait (like unaligned non-AIO DIO), | |
542 | * below inode_dio_wait() may anyway become a no-op, since we start | |
543 | * with exclusive lock. | |
378f32ba | 544 | */ |
aa9714d0 RH |
545 | if (unaligned_io) |
546 | inode_dio_wait(inode); | |
378f32ba | 547 | |
aa9714d0 | 548 | if (extend) { |
378f32ba MB |
549 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
550 | if (IS_ERR(handle)) { | |
551 | ret = PTR_ERR(handle); | |
552 | goto out; | |
553 | } | |
554 | ||
aa75f4d3 | 555 | ext4_fc_start_update(inode); |
378f32ba | 556 | ret = ext4_orphan_add(handle, inode); |
aa75f4d3 | 557 | ext4_fc_stop_update(inode); |
378f32ba MB |
558 | if (ret) { |
559 | ext4_journal_stop(handle); | |
560 | goto out; | |
561 | } | |
562 | ||
378f32ba MB |
563 | ext4_journal_stop(handle); |
564 | } | |
565 | ||
8cd115bd JK |
566 | if (ilock_shared) |
567 | iomap_ops = &ext4_iomap_overwrite_ops; | |
568 | ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops, | |
2f632965 | 569 | (unaligned_io || extend) ? IOMAP_DIO_FORCE_WAIT : 0); |
60263d58 CH |
570 | if (ret == -ENOTBLK) |
571 | ret = 0; | |
378f32ba MB |
572 | |
573 | if (extend) | |
574 | ret = ext4_handle_inode_extension(inode, offset, ret, count); | |
575 | ||
576 | out: | |
aa9714d0 | 577 | if (ilock_shared) |
378f32ba MB |
578 | inode_unlock_shared(inode); |
579 | else | |
580 | inode_unlock(inode); | |
581 | ||
582 | if (ret >= 0 && iov_iter_count(from)) { | |
583 | ssize_t err; | |
584 | loff_t endbyte; | |
585 | ||
586 | offset = iocb->ki_pos; | |
587 | err = ext4_buffered_write_iter(iocb, from); | |
588 | if (err < 0) | |
589 | return err; | |
590 | ||
591 | /* | |
592 | * We need to ensure that the pages within the page cache for | |
593 | * the range covered by this I/O are written to disk and | |
594 | * invalidated. This is in attempt to preserve the expected | |
595 | * direct I/O semantics in the case we fallback to buffered I/O | |
596 | * to complete off the I/O request. | |
597 | */ | |
598 | ret += err; | |
599 | endbyte = offset + err - 1; | |
600 | err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping, | |
601 | offset, endbyte); | |
602 | if (!err) | |
603 | invalidate_mapping_pages(iocb->ki_filp->f_mapping, | |
604 | offset >> PAGE_SHIFT, | |
605 | endbyte >> PAGE_SHIFT); | |
606 | } | |
607 | ||
608 | return ret; | |
609 | } | |
610 | ||
776722e8 JK |
611 | #ifdef CONFIG_FS_DAX |
612 | static ssize_t | |
613 | ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
614 | { | |
776722e8 | 615 | ssize_t ret; |
569342dc MB |
616 | size_t count; |
617 | loff_t offset; | |
0b9f230b MB |
618 | handle_t *handle; |
619 | bool extend = false; | |
569342dc | 620 | struct inode *inode = file_inode(iocb->ki_filp); |
776722e8 | 621 | |
f629afe3 RH |
622 | if (iocb->ki_flags & IOCB_NOWAIT) { |
623 | if (!inode_trylock(inode)) | |
728fbc0e | 624 | return -EAGAIN; |
f629afe3 | 625 | } else { |
728fbc0e GR |
626 | inode_lock(inode); |
627 | } | |
378f32ba | 628 | |
776722e8 JK |
629 | ret = ext4_write_checks(iocb, from); |
630 | if (ret <= 0) | |
631 | goto out; | |
776722e8 | 632 | |
569342dc MB |
633 | offset = iocb->ki_pos; |
634 | count = iov_iter_count(from); | |
0b9f230b MB |
635 | |
636 | if (offset + count > EXT4_I(inode)->i_disksize) { | |
637 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
638 | if (IS_ERR(handle)) { | |
639 | ret = PTR_ERR(handle); | |
640 | goto out; | |
641 | } | |
642 | ||
643 | ret = ext4_orphan_add(handle, inode); | |
644 | if (ret) { | |
645 | ext4_journal_stop(handle); | |
646 | goto out; | |
647 | } | |
648 | ||
649 | extend = true; | |
650 | ext4_journal_stop(handle); | |
651 | } | |
652 | ||
776722e8 | 653 | ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops); |
0b9f230b MB |
654 | |
655 | if (extend) | |
656 | ret = ext4_handle_inode_extension(inode, offset, ret, count); | |
776722e8 | 657 | out: |
ff5462e3 | 658 | inode_unlock(inode); |
776722e8 JK |
659 | if (ret > 0) |
660 | ret = generic_write_sync(iocb, ret); | |
661 | return ret; | |
662 | } | |
663 | #endif | |
664 | ||
ac27a0ec | 665 | static ssize_t |
9b884164 | 666 | ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ac27a0ec | 667 | { |
8ad2850f | 668 | struct inode *inode = file_inode(iocb->ki_filp); |
7608e610 | 669 | |
0db1ff22 TT |
670 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
671 | return -EIO; | |
672 | ||
776722e8 JK |
673 | #ifdef CONFIG_FS_DAX |
674 | if (IS_DAX(inode)) | |
675 | return ext4_dax_write_iter(iocb, from); | |
676 | #endif | |
378f32ba MB |
677 | if (iocb->ki_flags & IOCB_DIRECT) |
678 | return ext4_dio_write_iter(iocb, from); | |
aa75f4d3 HS |
679 | else |
680 | return ext4_buffered_write_iter(iocb, from); | |
ac27a0ec DK |
681 | } |
682 | ||
923ae0ff | 683 | #ifdef CONFIG_FS_DAX |
71fe9899 | 684 | static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf, |
c791ace1 | 685 | enum page_entry_size pe_size) |
923ae0ff | 686 | { |
71fe9899 SJ |
687 | int error = 0; |
688 | vm_fault_t result; | |
22446423 | 689 | int retries = 0; |
fb26a1cb | 690 | handle_t *handle = NULL; |
11bac800 | 691 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ea3d7209 | 692 | struct super_block *sb = inode->i_sb; |
fd96b8da RD |
693 | |
694 | /* | |
695 | * We have to distinguish real writes from writes which will result in a | |
696 | * COW page; COW writes should *not* poke the journal (the file will not | |
697 | * be changed). Doing so would cause unintended failures when mounted | |
698 | * read-only. | |
699 | * | |
700 | * We check for VM_SHARED rather than vmf->cow_page since the latter is | |
701 | * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for | |
702 | * other sizes, dax_iomap_fault will handle splitting / fallback so that | |
703 | * we eventually come back with a COW page. | |
704 | */ | |
705 | bool write = (vmf->flags & FAULT_FLAG_WRITE) && | |
706 | (vmf->vma->vm_flags & VM_SHARED); | |
b8a6176c | 707 | pfn_t pfn; |
01a33b4a MW |
708 | |
709 | if (write) { | |
710 | sb_start_pagefault(sb); | |
11bac800 | 711 | file_update_time(vmf->vma->vm_file); |
fb26a1cb | 712 | down_read(&EXT4_I(inode)->i_mmap_sem); |
22446423 | 713 | retry: |
fb26a1cb JK |
714 | handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE, |
715 | EXT4_DATA_TRANS_BLOCKS(sb)); | |
497f6926 JK |
716 | if (IS_ERR(handle)) { |
717 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
718 | sb_end_pagefault(sb); | |
719 | return VM_FAULT_SIGBUS; | |
720 | } | |
fb26a1cb JK |
721 | } else { |
722 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
1db17542 | 723 | } |
22446423 | 724 | result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops); |
fb26a1cb | 725 | if (write) { |
497f6926 | 726 | ext4_journal_stop(handle); |
22446423 JK |
727 | |
728 | if ((result & VM_FAULT_ERROR) && error == -ENOSPC && | |
729 | ext4_should_retry_alloc(sb, &retries)) | |
730 | goto retry; | |
b8a6176c JK |
731 | /* Handling synchronous page fault? */ |
732 | if (result & VM_FAULT_NEEDDSYNC) | |
733 | result = dax_finish_sync_fault(vmf, pe_size, pfn); | |
fb26a1cb | 734 | up_read(&EXT4_I(inode)->i_mmap_sem); |
01a33b4a | 735 | sb_end_pagefault(sb); |
fb26a1cb JK |
736 | } else { |
737 | up_read(&EXT4_I(inode)->i_mmap_sem); | |
738 | } | |
01a33b4a MW |
739 | |
740 | return result; | |
923ae0ff RZ |
741 | } |
742 | ||
71fe9899 | 743 | static vm_fault_t ext4_dax_fault(struct vm_fault *vmf) |
c791ace1 DJ |
744 | { |
745 | return ext4_dax_huge_fault(vmf, PE_SIZE_PTE); | |
746 | } | |
747 | ||
923ae0ff RZ |
748 | static const struct vm_operations_struct ext4_dax_vm_ops = { |
749 | .fault = ext4_dax_fault, | |
c791ace1 | 750 | .huge_fault = ext4_dax_huge_fault, |
1e9d180b | 751 | .page_mkwrite = ext4_dax_fault, |
91d25ba8 | 752 | .pfn_mkwrite = ext4_dax_fault, |
923ae0ff RZ |
753 | }; |
754 | #else | |
755 | #define ext4_dax_vm_ops ext4_file_vm_ops | |
756 | #endif | |
757 | ||
f0f37e2f | 758 | static const struct vm_operations_struct ext4_file_vm_ops = { |
ea3d7209 | 759 | .fault = ext4_filemap_fault, |
f1820361 | 760 | .map_pages = filemap_map_pages, |
2e9ee850 AK |
761 | .page_mkwrite = ext4_page_mkwrite, |
762 | }; | |
763 | ||
764 | static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma) | |
765 | { | |
c9c7429c | 766 | struct inode *inode = file->f_mapping->host; |
e46bfc3f PG |
767 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
768 | struct dax_device *dax_dev = sbi->s_daxdev; | |
c9c7429c | 769 | |
e46bfc3f | 770 | if (unlikely(ext4_forced_shutdown(sbi))) |
0db1ff22 TT |
771 | return -EIO; |
772 | ||
b8a6176c | 773 | /* |
e46bfc3f PG |
774 | * We don't support synchronous mappings for non-DAX files and |
775 | * for DAX files if underneath dax_device is not synchronous. | |
b8a6176c | 776 | */ |
e46bfc3f | 777 | if (!daxdev_mapping_supported(vma, dax_dev)) |
b8a6176c JK |
778 | return -EOPNOTSUPP; |
779 | ||
2e9ee850 | 780 | file_accessed(file); |
923ae0ff RZ |
781 | if (IS_DAX(file_inode(file))) { |
782 | vma->vm_ops = &ext4_dax_vm_ops; | |
e1fb4a08 | 783 | vma->vm_flags |= VM_HUGEPAGE; |
923ae0ff RZ |
784 | } else { |
785 | vma->vm_ops = &ext4_file_vm_ops; | |
786 | } | |
2e9ee850 AK |
787 | return 0; |
788 | } | |
789 | ||
833a9508 AG |
790 | static int ext4_sample_last_mounted(struct super_block *sb, |
791 | struct vfsmount *mnt) | |
bc0b0d6d | 792 | { |
833a9508 | 793 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
bc0b0d6d TT |
794 | struct path path; |
795 | char buf[64], *cp; | |
833a9508 AG |
796 | handle_t *handle; |
797 | int err; | |
798 | ||
9b5f6c9b | 799 | if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED))) |
833a9508 AG |
800 | return 0; |
801 | ||
db6516a5 | 802 | if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb)) |
833a9508 AG |
803 | return 0; |
804 | ||
9b5f6c9b | 805 | ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED); |
833a9508 AG |
806 | /* |
807 | * Sample where the filesystem has been mounted and | |
808 | * store it in the superblock for sysadmin convenience | |
809 | * when trying to sort through large numbers of block | |
810 | * devices or filesystem images. | |
811 | */ | |
812 | memset(buf, 0, sizeof(buf)); | |
813 | path.mnt = mnt; | |
814 | path.dentry = mnt->mnt_root; | |
815 | cp = d_path(&path, buf, sizeof(buf)); | |
db6516a5 | 816 | err = 0; |
833a9508 | 817 | if (IS_ERR(cp)) |
db6516a5 | 818 | goto out; |
833a9508 AG |
819 | |
820 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); | |
db6516a5 | 821 | err = PTR_ERR(handle); |
833a9508 | 822 | if (IS_ERR(handle)) |
db6516a5 | 823 | goto out; |
833a9508 AG |
824 | BUFFER_TRACE(sbi->s_sbh, "get_write_access"); |
825 | err = ext4_journal_get_write_access(handle, sbi->s_sbh); | |
826 | if (err) | |
db6516a5 | 827 | goto out_journal; |
05c2c00f | 828 | lock_buffer(sbi->s_sbh); |
5a3b590d | 829 | strncpy(sbi->s_es->s_last_mounted, cp, |
833a9508 | 830 | sizeof(sbi->s_es->s_last_mounted)); |
05c2c00f JK |
831 | ext4_superblock_csum_set(sb); |
832 | unlock_buffer(sbi->s_sbh); | |
a3f5cf14 | 833 | ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh); |
db6516a5 | 834 | out_journal: |
833a9508 | 835 | ext4_journal_stop(handle); |
db6516a5 AG |
836 | out: |
837 | sb_end_intwrite(sb); | |
833a9508 AG |
838 | return err; |
839 | } | |
840 | ||
e030a288 | 841 | static int ext4_file_open(struct inode *inode, struct file *filp) |
833a9508 | 842 | { |
c9c7429c | 843 | int ret; |
bc0b0d6d | 844 | |
0db1ff22 TT |
845 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
846 | return -EIO; | |
847 | ||
833a9508 AG |
848 | ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt); |
849 | if (ret) | |
850 | return ret; | |
9dd78d8c | 851 | |
09a5c31c EB |
852 | ret = fscrypt_file_open(inode, filp); |
853 | if (ret) | |
c93d8f88 EB |
854 | return ret; |
855 | ||
856 | ret = fsverity_file_open(inode, filp); | |
857 | if (ret) | |
09a5c31c EB |
858 | return ret; |
859 | ||
8aefcd55 TT |
860 | /* |
861 | * Set up the jbd2_inode if we are opening the inode for | |
862 | * writing and the journal is present | |
863 | */ | |
a361293f | 864 | if (filp->f_mode & FMODE_WRITE) { |
c9c7429c | 865 | ret = ext4_inode_attach_jinode(inode); |
a361293f JK |
866 | if (ret < 0) |
867 | return ret; | |
8aefcd55 | 868 | } |
728fbc0e | 869 | |
766ef1e1 | 870 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC; |
abdd438b | 871 | return dquot_file_open(inode, filp); |
bc0b0d6d TT |
872 | } |
873 | ||
e0d10bfa | 874 | /* |
ec7268ce ES |
875 | * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values |
876 | * by calling generic_file_llseek_size() with the appropriate maxbytes | |
877 | * value for each. | |
e0d10bfa | 878 | */ |
965c8e59 | 879 | loff_t ext4_llseek(struct file *file, loff_t offset, int whence) |
e0d10bfa TO |
880 | { |
881 | struct inode *inode = file->f_mapping->host; | |
882 | loff_t maxbytes; | |
883 | ||
884 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) | |
885 | maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes; | |
886 | else | |
887 | maxbytes = inode->i_sb->s_maxbytes; | |
e0d10bfa | 888 | |
965c8e59 | 889 | switch (whence) { |
545052e9 | 890 | default: |
965c8e59 | 891 | return generic_file_llseek_size(file, offset, whence, |
c8c0df24 | 892 | maxbytes, i_size_read(inode)); |
c8c0df24 | 893 | case SEEK_HOLE: |
545052e9 | 894 | inode_lock_shared(inode); |
09edf4d3 MB |
895 | offset = iomap_seek_hole(inode, offset, |
896 | &ext4_iomap_report_ops); | |
545052e9 CH |
897 | inode_unlock_shared(inode); |
898 | break; | |
899 | case SEEK_DATA: | |
900 | inode_lock_shared(inode); | |
09edf4d3 MB |
901 | offset = iomap_seek_data(inode, offset, |
902 | &ext4_iomap_report_ops); | |
545052e9 CH |
903 | inode_unlock_shared(inode); |
904 | break; | |
c8c0df24 ZL |
905 | } |
906 | ||
545052e9 CH |
907 | if (offset < 0) |
908 | return offset; | |
909 | return vfs_setpos(file, offset, maxbytes); | |
e0d10bfa TO |
910 | } |
911 | ||
617ba13b | 912 | const struct file_operations ext4_file_operations = { |
e0d10bfa | 913 | .llseek = ext4_llseek, |
364443cb | 914 | .read_iter = ext4_file_read_iter, |
9b884164 | 915 | .write_iter = ext4_file_write_iter, |
72f9da1d | 916 | .iopoll = iomap_dio_iopoll, |
5cdd7b2d | 917 | .unlocked_ioctl = ext4_ioctl, |
ac27a0ec | 918 | #ifdef CONFIG_COMPAT |
617ba13b | 919 | .compat_ioctl = ext4_compat_ioctl, |
ac27a0ec | 920 | #endif |
2e9ee850 | 921 | .mmap = ext4_file_mmap, |
b8a6176c | 922 | .mmap_supported_flags = MAP_SYNC, |
bc0b0d6d | 923 | .open = ext4_file_open, |
617ba13b MC |
924 | .release = ext4_release_file, |
925 | .fsync = ext4_sync_file, | |
dbe6ec81 | 926 | .get_unmapped_area = thp_get_unmapped_area, |
ac27a0ec | 927 | .splice_read = generic_file_splice_read, |
8d020765 | 928 | .splice_write = iter_file_splice_write, |
2fe17c10 | 929 | .fallocate = ext4_fallocate, |
ac27a0ec DK |
930 | }; |
931 | ||
754661f1 | 932 | const struct inode_operations ext4_file_inode_operations = { |
617ba13b | 933 | .setattr = ext4_setattr, |
99652ea5 | 934 | .getattr = ext4_file_getattr, |
617ba13b | 935 | .listxattr = ext4_listxattr, |
4e34e719 | 936 | .get_acl = ext4_get_acl, |
64e178a7 | 937 | .set_acl = ext4_set_acl, |
6873fa0d | 938 | .fiemap = ext4_fiemap, |
4db5c2e6 MS |
939 | .fileattr_get = ext4_fileattr_get, |
940 | .fileattr_set = ext4_fileattr_set, | |
ac27a0ec DK |
941 | }; |
942 |