Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
dda35b8f | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
a4fbe6ab | 21 | #include "xfs_format.h" |
239880ef DC |
22 | #include "xfs_log_format.h" |
23 | #include "xfs_trans_resv.h" | |
1da177e4 | 24 | #include "xfs_mount.h" |
57062787 DC |
25 | #include "xfs_da_format.h" |
26 | #include "xfs_da_btree.h" | |
1da177e4 | 27 | #include "xfs_inode.h" |
239880ef | 28 | #include "xfs_trans.h" |
fd3200be | 29 | #include "xfs_inode_item.h" |
dda35b8f | 30 | #include "xfs_bmap.h" |
c24b5dfa | 31 | #include "xfs_bmap_util.h" |
1da177e4 | 32 | #include "xfs_error.h" |
2b9ab5ab | 33 | #include "xfs_dir2.h" |
c24b5dfa | 34 | #include "xfs_dir2_priv.h" |
ddcd856d | 35 | #include "xfs_ioctl.h" |
dda35b8f | 36 | #include "xfs_trace.h" |
239880ef | 37 | #include "xfs_log.h" |
dc06f398 | 38 | #include "xfs_icache.h" |
781355c6 | 39 | #include "xfs_pnfs.h" |
68a9f5e7 | 40 | #include "xfs_iomap.h" |
0613f16c | 41 | #include "xfs_reflink.h" |
1da177e4 LT |
42 | |
43 | #include <linux/dcache.h> | |
2fe17c10 | 44 | #include <linux/falloc.h> |
d126d43f | 45 | #include <linux/pagevec.h> |
66114cad | 46 | #include <linux/backing-dev.h> |
1da177e4 | 47 | |
f0f37e2f | 48 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 49 | |
dda35b8f | 50 | /* |
68a9f5e7 CH |
51 | * Clear the specified ranges to zero through either the pagecache or DAX. |
52 | * Holes and unwritten extents will be left as-is as they already are zeroed. | |
dda35b8f | 53 | */ |
ef9d8733 | 54 | int |
7bb41db3 | 55 | xfs_zero_range( |
68a9f5e7 | 56 | struct xfs_inode *ip, |
7bb41db3 CH |
57 | xfs_off_t pos, |
58 | xfs_off_t count, | |
59 | bool *did_zero) | |
dda35b8f | 60 | { |
d20a5e38 | 61 | return iomap_zero_range(VFS_I(ip), pos, count, did_zero, &xfs_iomap_ops); |
dda35b8f CH |
62 | } |
63 | ||
8add71ca CH |
64 | int |
65 | xfs_update_prealloc_flags( | |
66 | struct xfs_inode *ip, | |
67 | enum xfs_prealloc_flags flags) | |
68 | { | |
69 | struct xfs_trans *tp; | |
70 | int error; | |
71 | ||
253f4911 CH |
72 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, |
73 | 0, 0, 0, &tp); | |
74 | if (error) | |
8add71ca | 75 | return error; |
8add71ca CH |
76 | |
77 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
78 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
79 | ||
80 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
81 | VFS_I(ip)->i_mode &= ~S_ISUID; |
82 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
83 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
84 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
85 | } | |
86 | ||
87 | if (flags & XFS_PREALLOC_SET) | |
88 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
89 | if (flags & XFS_PREALLOC_CLEAR) | |
90 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
91 | ||
92 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
93 | if (flags & XFS_PREALLOC_SYNC) | |
94 | xfs_trans_set_sync(tp); | |
70393313 | 95 | return xfs_trans_commit(tp); |
8add71ca CH |
96 | } |
97 | ||
1da2f2db CH |
98 | /* |
99 | * Fsync operations on directories are much simpler than on regular files, | |
100 | * as there is no file data to flush, and thus also no need for explicit | |
101 | * cache flush operations, and there are no non-transaction metadata updates | |
102 | * on directories either. | |
103 | */ | |
104 | STATIC int | |
105 | xfs_dir_fsync( | |
106 | struct file *file, | |
107 | loff_t start, | |
108 | loff_t end, | |
109 | int datasync) | |
110 | { | |
111 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
112 | struct xfs_mount *mp = ip->i_mount; | |
113 | xfs_lsn_t lsn = 0; | |
114 | ||
115 | trace_xfs_dir_fsync(ip); | |
116 | ||
117 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
118 | if (xfs_ipincount(ip)) | |
119 | lsn = ip->i_itemp->ili_last_lsn; | |
120 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
121 | ||
122 | if (!lsn) | |
123 | return 0; | |
2451337d | 124 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
125 | } |
126 | ||
fd3200be CH |
127 | STATIC int |
128 | xfs_file_fsync( | |
129 | struct file *file, | |
02c24a82 JB |
130 | loff_t start, |
131 | loff_t end, | |
fd3200be CH |
132 | int datasync) |
133 | { | |
7ea80859 CH |
134 | struct inode *inode = file->f_mapping->host; |
135 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 136 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
137 | int error = 0; |
138 | int log_flushed = 0; | |
b1037058 | 139 | xfs_lsn_t lsn = 0; |
fd3200be | 140 | |
cca28fb8 | 141 | trace_xfs_file_fsync(ip); |
fd3200be | 142 | |
1b180274 | 143 | error = file_write_and_wait_range(file, start, end); |
02c24a82 JB |
144 | if (error) |
145 | return error; | |
146 | ||
a27a263b | 147 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 148 | return -EIO; |
fd3200be CH |
149 | |
150 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
151 | ||
2291dab2 DC |
152 | /* |
153 | * If we have an RT and/or log subvolume we need to make sure to flush | |
154 | * the write cache the device used for file data first. This is to | |
155 | * ensure newly written file data make it to disk before logging the new | |
156 | * inode size in case of an extending write. | |
157 | */ | |
158 | if (XFS_IS_REALTIME_INODE(ip)) | |
159 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
160 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
161 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
a27a263b | 162 | |
fd3200be | 163 | /* |
fc0561ce DC |
164 | * All metadata updates are logged, which means that we just have to |
165 | * flush the log up to the latest LSN that touched the inode. If we have | |
166 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
167 | * log force before we clear the ili_fsync_fields field. This ensures | |
168 | * that we don't get a racing sync operation that does not wait for the | |
169 | * metadata to hit the journal before returning. If we race with | |
170 | * clearing the ili_fsync_fields, then all that will happen is the log | |
171 | * force will do nothing as the lsn will already be on disk. We can't | |
172 | * race with setting ili_fsync_fields because that is done under | |
173 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
174 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
175 | */ |
176 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
177 | if (xfs_ipincount(ip)) { |
178 | if (!datasync || | |
fc0561ce | 179 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
180 | lsn = ip->i_itemp->ili_last_lsn; |
181 | } | |
fd3200be | 182 | |
fc0561ce | 183 | if (lsn) { |
b1037058 | 184 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
185 | ip->i_itemp->ili_fsync_fields = 0; |
186 | } | |
187 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 188 | |
a27a263b CH |
189 | /* |
190 | * If we only have a single device, and the log force about was | |
191 | * a no-op we might have to flush the data device cache here. | |
192 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
193 | * an already allocated file and thus do not have any metadata to | |
194 | * commit. | |
195 | */ | |
2291dab2 DC |
196 | if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) && |
197 | mp->m_logdev_targp == mp->m_ddev_targp) | |
a27a263b | 198 | xfs_blkdev_issue_flush(mp->m_ddev_targp); |
fd3200be | 199 | |
2451337d | 200 | return error; |
fd3200be CH |
201 | } |
202 | ||
00258e36 | 203 | STATIC ssize_t |
bbc5a740 | 204 | xfs_file_dio_aio_read( |
dda35b8f | 205 | struct kiocb *iocb, |
b4f5d2c6 | 206 | struct iov_iter *to) |
dda35b8f | 207 | { |
acdda3aa | 208 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); |
bbc5a740 | 209 | size_t count = iov_iter_count(to); |
acdda3aa | 210 | ssize_t ret; |
dda35b8f | 211 | |
bbc5a740 | 212 | trace_xfs_file_direct_read(ip, count, iocb->ki_pos); |
dda35b8f | 213 | |
f1285ff0 CH |
214 | if (!count) |
215 | return 0; /* skip atime */ | |
dda35b8f | 216 | |
a447d7cd CH |
217 | file_accessed(iocb->ki_filp); |
218 | ||
65523218 | 219 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 220 | ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL); |
65523218 | 221 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
acdda3aa | 222 | |
16d4d435 CH |
223 | return ret; |
224 | } | |
225 | ||
f021bd07 | 226 | static noinline ssize_t |
16d4d435 CH |
227 | xfs_file_dax_read( |
228 | struct kiocb *iocb, | |
229 | struct iov_iter *to) | |
230 | { | |
6c31f495 | 231 | struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host); |
16d4d435 CH |
232 | size_t count = iov_iter_count(to); |
233 | ssize_t ret = 0; | |
234 | ||
235 | trace_xfs_file_dax_read(ip, count, iocb->ki_pos); | |
236 | ||
237 | if (!count) | |
238 | return 0; /* skip atime */ | |
239 | ||
29a5d29e GR |
240 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { |
241 | if (iocb->ki_flags & IOCB_NOWAIT) | |
242 | return -EAGAIN; | |
243 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
244 | } | |
11c59c92 | 245 | ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops); |
65523218 | 246 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 | 247 | |
f1285ff0 | 248 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
249 | return ret; |
250 | } | |
251 | ||
252 | STATIC ssize_t | |
253 | xfs_file_buffered_aio_read( | |
254 | struct kiocb *iocb, | |
255 | struct iov_iter *to) | |
256 | { | |
257 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
258 | ssize_t ret; | |
259 | ||
260 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
dda35b8f | 261 | |
91f9943e CH |
262 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { |
263 | if (iocb->ki_flags & IOCB_NOWAIT) | |
264 | return -EAGAIN; | |
265 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
266 | } | |
b4f5d2c6 | 267 | ret = generic_file_read_iter(iocb, to); |
65523218 | 268 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 CH |
269 | |
270 | return ret; | |
271 | } | |
272 | ||
273 | STATIC ssize_t | |
274 | xfs_file_read_iter( | |
275 | struct kiocb *iocb, | |
276 | struct iov_iter *to) | |
277 | { | |
16d4d435 CH |
278 | struct inode *inode = file_inode(iocb->ki_filp); |
279 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
280 | ssize_t ret = 0; |
281 | ||
282 | XFS_STATS_INC(mp, xs_read_calls); | |
283 | ||
284 | if (XFS_FORCED_SHUTDOWN(mp)) | |
285 | return -EIO; | |
286 | ||
16d4d435 CH |
287 | if (IS_DAX(inode)) |
288 | ret = xfs_file_dax_read(iocb, to); | |
289 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 290 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 291 | else |
bbc5a740 | 292 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 293 | |
dda35b8f | 294 | if (ret > 0) |
ff6d6af2 | 295 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
296 | return ret; |
297 | } | |
298 | ||
dda35b8f | 299 | /* |
193aec10 CH |
300 | * Zero any on disk space between the current EOF and the new, larger EOF. |
301 | * | |
302 | * This handles the normal case of zeroing the remainder of the last block in | |
303 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
304 | * file. This second case only happens with fixed size extents and when the | |
305 | * system crashes before the inode size was updated but after blocks were | |
306 | * allocated. | |
307 | * | |
308 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 309 | */ |
dda35b8f CH |
310 | int /* error (positive) */ |
311 | xfs_zero_eof( | |
193aec10 CH |
312 | struct xfs_inode *ip, |
313 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
314 | xfs_fsize_t isize, /* current inode size */ |
315 | bool *did_zeroing) | |
dda35b8f | 316 | { |
193aec10 | 317 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
dda35b8f CH |
318 | ASSERT(offset > isize); |
319 | ||
0a50f162 | 320 | trace_xfs_zero_eof(ip, isize, offset - isize); |
570b6211 | 321 | return xfs_zero_range(ip, isize, offset - isize, did_zeroing); |
dda35b8f CH |
322 | } |
323 | ||
4d8d1581 DC |
324 | /* |
325 | * Common pre-write limit and setup checks. | |
326 | * | |
5bf1f262 CH |
327 | * Called with the iolocked held either shared and exclusive according to |
328 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
329 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
330 | */ |
331 | STATIC ssize_t | |
332 | xfs_file_aio_write_checks( | |
99733fa3 AV |
333 | struct kiocb *iocb, |
334 | struct iov_iter *from, | |
4d8d1581 DC |
335 | int *iolock) |
336 | { | |
99733fa3 | 337 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
338 | struct inode *inode = file->f_mapping->host; |
339 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 340 | ssize_t error = 0; |
99733fa3 | 341 | size_t count = iov_iter_count(from); |
3136e8bb | 342 | bool drained_dio = false; |
4d8d1581 | 343 | |
7271d243 | 344 | restart: |
3309dd04 AV |
345 | error = generic_write_checks(iocb, from); |
346 | if (error <= 0) | |
4d8d1581 | 347 | return error; |
4d8d1581 | 348 | |
65523218 | 349 | error = xfs_break_layouts(inode, iolock); |
781355c6 CH |
350 | if (error) |
351 | return error; | |
352 | ||
65523218 CH |
353 | /* |
354 | * For changing security info in file_remove_privs() we need i_rwsem | |
355 | * exclusively. | |
356 | */ | |
a6de82ca | 357 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
65523218 | 358 | xfs_iunlock(ip, *iolock); |
a6de82ca | 359 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 360 | xfs_ilock(ip, *iolock); |
a6de82ca JK |
361 | goto restart; |
362 | } | |
4d8d1581 DC |
363 | /* |
364 | * If the offset is beyond the size of the file, we need to zero any | |
365 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 366 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
367 | * iolock shared, we need to update it to exclusive which implies |
368 | * having to redo all checks before. | |
b9d59846 DC |
369 | * |
370 | * We need to serialise against EOF updates that occur in IO | |
371 | * completions here. We want to make sure that nobody is changing the | |
372 | * size while we do this check until we have placed an IO barrier (i.e. | |
373 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
374 | * The spinlock effectively forms a memory barrier once we have the | |
375 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
376 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 377 | */ |
b9d59846 | 378 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 379 | if (iocb->ki_pos > i_size_read(inode)) { |
b9d59846 | 380 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
381 | if (!drained_dio) { |
382 | if (*iolock == XFS_IOLOCK_SHARED) { | |
65523218 | 383 | xfs_iunlock(ip, *iolock); |
3136e8bb | 384 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 385 | xfs_ilock(ip, *iolock); |
3136e8bb BF |
386 | iov_iter_reexpand(from, count); |
387 | } | |
40c63fbc DC |
388 | /* |
389 | * We now have an IO submission barrier in place, but | |
390 | * AIO can do EOF updates during IO completion and hence | |
391 | * we now need to wait for all of them to drain. Non-AIO | |
392 | * DIO will have drained before we are given the | |
393 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
394 | * no-op. | |
395 | */ | |
396 | inode_dio_wait(inode); | |
3136e8bb | 397 | drained_dio = true; |
7271d243 DC |
398 | goto restart; |
399 | } | |
64671baf | 400 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), NULL); |
467f7899 CH |
401 | if (error) |
402 | return error; | |
b9d59846 DC |
403 | } else |
404 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 405 | |
8a9c9980 CH |
406 | /* |
407 | * Updating the timestamps will grab the ilock again from | |
408 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
409 | * lock above. Eventually we should look into a way to avoid | |
410 | * the pointless lock roundtrip. | |
411 | */ | |
c3b2da31 JB |
412 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
413 | error = file_update_time(file); | |
414 | if (error) | |
415 | return error; | |
416 | } | |
8a9c9980 | 417 | |
4d8d1581 DC |
418 | /* |
419 | * If we're writing the file then make sure to clear the setuid and | |
420 | * setgid bits if the process is not being run by root. This keeps | |
421 | * people from modifying setuid and setgid binaries. | |
422 | */ | |
a6de82ca JK |
423 | if (!IS_NOSEC(inode)) |
424 | return file_remove_privs(file); | |
425 | return 0; | |
4d8d1581 DC |
426 | } |
427 | ||
acdda3aa CH |
428 | static int |
429 | xfs_dio_write_end_io( | |
430 | struct kiocb *iocb, | |
431 | ssize_t size, | |
432 | unsigned flags) | |
433 | { | |
434 | struct inode *inode = file_inode(iocb->ki_filp); | |
435 | struct xfs_inode *ip = XFS_I(inode); | |
436 | loff_t offset = iocb->ki_pos; | |
acdda3aa CH |
437 | int error = 0; |
438 | ||
439 | trace_xfs_end_io_direct_write(ip, offset, size); | |
440 | ||
441 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
442 | return -EIO; | |
443 | ||
444 | if (size <= 0) | |
445 | return size; | |
446 | ||
ee70daab EG |
447 | if (flags & IOMAP_DIO_COW) { |
448 | error = xfs_reflink_end_cow(ip, offset, size); | |
449 | if (error) | |
450 | return error; | |
451 | } | |
452 | ||
453 | /* | |
454 | * Unwritten conversion updates the in-core isize after extent | |
455 | * conversion but before updating the on-disk size. Updating isize any | |
456 | * earlier allows a racing dio read to find unwritten extents before | |
457 | * they are converted. | |
458 | */ | |
459 | if (flags & IOMAP_DIO_UNWRITTEN) | |
460 | return xfs_iomap_write_unwritten(ip, offset, size, true); | |
461 | ||
acdda3aa CH |
462 | /* |
463 | * We need to update the in-core inode size here so that we don't end up | |
464 | * with the on-disk inode size being outside the in-core inode size. We | |
465 | * have no other method of updating EOF for AIO, so always do it here | |
466 | * if necessary. | |
467 | * | |
468 | * We need to lock the test/set EOF update as we can be racing with | |
469 | * other IO completions here to update the EOF. Failing to serialise | |
470 | * here can result in EOF moving backwards and Bad Things Happen when | |
471 | * that occurs. | |
472 | */ | |
473 | spin_lock(&ip->i_flags_lock); | |
474 | if (offset + size > i_size_read(inode)) { | |
475 | i_size_write(inode, offset + size); | |
ee70daab | 476 | spin_unlock(&ip->i_flags_lock); |
acdda3aa | 477 | error = xfs_setfilesize(ip, offset, size); |
ee70daab EG |
478 | } else { |
479 | spin_unlock(&ip->i_flags_lock); | |
480 | } | |
acdda3aa CH |
481 | |
482 | return error; | |
483 | } | |
484 | ||
f0d26e86 DC |
485 | /* |
486 | * xfs_file_dio_aio_write - handle direct IO writes | |
487 | * | |
488 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 489 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
490 | * follow locking changes and looping. |
491 | * | |
eda77982 DC |
492 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
493 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
494 | * pages are flushed out. | |
495 | * | |
496 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
497 | * allowing them to be done in parallel with reads and other direct IO writes. | |
498 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
499 | * needs to do sub-block zeroing and that requires serialisation against other | |
500 | * direct IOs to the same block. In this case we need to serialise the | |
501 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
502 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
503 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
504 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 505 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 506 | * |
f0d26e86 DC |
507 | * Returns with locks held indicated by @iolock and errors indicated by |
508 | * negative return values. | |
509 | */ | |
510 | STATIC ssize_t | |
511 | xfs_file_dio_aio_write( | |
512 | struct kiocb *iocb, | |
b3188919 | 513 | struct iov_iter *from) |
f0d26e86 DC |
514 | { |
515 | struct file *file = iocb->ki_filp; | |
516 | struct address_space *mapping = file->f_mapping; | |
517 | struct inode *inode = mapping->host; | |
518 | struct xfs_inode *ip = XFS_I(inode); | |
519 | struct xfs_mount *mp = ip->i_mount; | |
520 | ssize_t ret = 0; | |
eda77982 | 521 | int unaligned_io = 0; |
d0606464 | 522 | int iolock; |
b3188919 | 523 | size_t count = iov_iter_count(from); |
acdda3aa | 524 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
f0d26e86 DC |
525 | mp->m_rtdev_targp : mp->m_ddev_targp; |
526 | ||
7c71ee78 | 527 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 528 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 529 | return -EINVAL; |
f0d26e86 | 530 | |
7271d243 | 531 | /* |
0ee7a3f6 CH |
532 | * Don't take the exclusive iolock here unless the I/O is unaligned to |
533 | * the file system block size. We don't need to consider the EOF | |
534 | * extension case here because xfs_file_aio_write_checks() will relock | |
535 | * the inode as necessary for EOF zeroing cases and fill out the new | |
536 | * inode size as appropriate. | |
7271d243 | 537 | */ |
0ee7a3f6 CH |
538 | if ((iocb->ki_pos & mp->m_blockmask) || |
539 | ((iocb->ki_pos + count) & mp->m_blockmask)) { | |
540 | unaligned_io = 1; | |
54a4ef8a CH |
541 | |
542 | /* | |
543 | * We can't properly handle unaligned direct I/O to reflink | |
544 | * files yet, as we can't unshare a partial block. | |
545 | */ | |
546 | if (xfs_is_reflink_inode(ip)) { | |
547 | trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count); | |
548 | return -EREMCHG; | |
549 | } | |
d0606464 | 550 | iolock = XFS_IOLOCK_EXCL; |
0ee7a3f6 | 551 | } else { |
d0606464 | 552 | iolock = XFS_IOLOCK_SHARED; |
c58cb165 | 553 | } |
f0d26e86 | 554 | |
29a5d29e GR |
555 | if (!xfs_ilock_nowait(ip, iolock)) { |
556 | if (iocb->ki_flags & IOCB_NOWAIT) | |
557 | return -EAGAIN; | |
558 | xfs_ilock(ip, iolock); | |
559 | } | |
0ee7a3f6 | 560 | |
99733fa3 | 561 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 562 | if (ret) |
d0606464 | 563 | goto out; |
99733fa3 | 564 | count = iov_iter_count(from); |
f0d26e86 | 565 | |
eda77982 DC |
566 | /* |
567 | * If we are doing unaligned IO, wait for all other IO to drain, | |
0ee7a3f6 CH |
568 | * otherwise demote the lock if we had to take the exclusive lock |
569 | * for other reasons in xfs_file_aio_write_checks. | |
eda77982 | 570 | */ |
29a5d29e GR |
571 | if (unaligned_io) { |
572 | /* If we are going to wait for other DIO to finish, bail */ | |
573 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
574 | if (atomic_read(&inode->i_dio_count)) | |
575 | return -EAGAIN; | |
576 | } else { | |
577 | inode_dio_wait(inode); | |
578 | } | |
579 | } else if (iolock == XFS_IOLOCK_EXCL) { | |
65523218 | 580 | xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 581 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
582 | } |
583 | ||
3176c3e0 | 584 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
acdda3aa | 585 | ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io); |
d0606464 | 586 | out: |
65523218 | 587 | xfs_iunlock(ip, iolock); |
d0606464 | 588 | |
6b698ede | 589 | /* |
16d4d435 CH |
590 | * No fallback to buffered IO on errors for XFS, direct IO will either |
591 | * complete fully or fail. | |
6b698ede | 592 | */ |
16d4d435 CH |
593 | ASSERT(ret < 0 || ret == count); |
594 | return ret; | |
595 | } | |
596 | ||
f021bd07 | 597 | static noinline ssize_t |
16d4d435 CH |
598 | xfs_file_dax_write( |
599 | struct kiocb *iocb, | |
600 | struct iov_iter *from) | |
601 | { | |
6c31f495 | 602 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
16d4d435 | 603 | struct xfs_inode *ip = XFS_I(inode); |
17879e8f | 604 | int iolock = XFS_IOLOCK_EXCL; |
6c31f495 CH |
605 | ssize_t ret, error = 0; |
606 | size_t count; | |
607 | loff_t pos; | |
16d4d435 | 608 | |
29a5d29e GR |
609 | if (!xfs_ilock_nowait(ip, iolock)) { |
610 | if (iocb->ki_flags & IOCB_NOWAIT) | |
611 | return -EAGAIN; | |
612 | xfs_ilock(ip, iolock); | |
613 | } | |
614 | ||
16d4d435 CH |
615 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
616 | if (ret) | |
617 | goto out; | |
618 | ||
6c31f495 CH |
619 | pos = iocb->ki_pos; |
620 | count = iov_iter_count(from); | |
8b2180b3 | 621 | |
6c31f495 | 622 | trace_xfs_file_dax_write(ip, count, pos); |
11c59c92 | 623 | ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops); |
6c31f495 CH |
624 | if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { |
625 | i_size_write(inode, iocb->ki_pos); | |
626 | error = xfs_setfilesize(ip, pos, ret); | |
16d4d435 | 627 | } |
16d4d435 | 628 | out: |
65523218 | 629 | xfs_iunlock(ip, iolock); |
6c31f495 | 630 | return error ? error : ret; |
f0d26e86 DC |
631 | } |
632 | ||
00258e36 | 633 | STATIC ssize_t |
637bbc75 | 634 | xfs_file_buffered_aio_write( |
dda35b8f | 635 | struct kiocb *iocb, |
b3188919 | 636 | struct iov_iter *from) |
dda35b8f CH |
637 | { |
638 | struct file *file = iocb->ki_filp; | |
639 | struct address_space *mapping = file->f_mapping; | |
640 | struct inode *inode = mapping->host; | |
00258e36 | 641 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
642 | ssize_t ret; |
643 | int enospc = 0; | |
c3155097 | 644 | int iolock; |
dda35b8f | 645 | |
91f9943e CH |
646 | if (iocb->ki_flags & IOCB_NOWAIT) |
647 | return -EOPNOTSUPP; | |
648 | ||
c3155097 BF |
649 | write_retry: |
650 | iolock = XFS_IOLOCK_EXCL; | |
65523218 | 651 | xfs_ilock(ip, iolock); |
dda35b8f | 652 | |
99733fa3 | 653 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 654 | if (ret) |
d0606464 | 655 | goto out; |
dda35b8f CH |
656 | |
657 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 658 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 659 | |
3176c3e0 | 660 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 661 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 662 | if (likely(ret >= 0)) |
99733fa3 | 663 | iocb->ki_pos += ret; |
dc06f398 | 664 | |
637bbc75 | 665 | /* |
dc06f398 BF |
666 | * If we hit a space limit, try to free up some lingering preallocated |
667 | * space before returning an error. In the case of ENOSPC, first try to | |
668 | * write back all dirty inodes to free up some of the excess reserved | |
669 | * metadata space. This reduces the chances that the eofblocks scan | |
670 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
671 | * also behaves as a filter to prevent too many eofblocks scans from | |
672 | * running at the same time. | |
637bbc75 | 673 | */ |
dc06f398 | 674 | if (ret == -EDQUOT && !enospc) { |
c3155097 | 675 | xfs_iunlock(ip, iolock); |
dc06f398 BF |
676 | enospc = xfs_inode_free_quota_eofblocks(ip); |
677 | if (enospc) | |
678 | goto write_retry; | |
83104d44 DW |
679 | enospc = xfs_inode_free_quota_cowblocks(ip); |
680 | if (enospc) | |
681 | goto write_retry; | |
c3155097 | 682 | iolock = 0; |
dc06f398 BF |
683 | } else if (ret == -ENOSPC && !enospc) { |
684 | struct xfs_eofblocks eofb = {0}; | |
685 | ||
637bbc75 | 686 | enospc = 1; |
9aa05000 | 687 | xfs_flush_inodes(ip->i_mount); |
c3155097 BF |
688 | |
689 | xfs_iunlock(ip, iolock); | |
dc06f398 BF |
690 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; |
691 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
cf2cb784 | 692 | xfs_icache_free_cowblocks(ip->i_mount, &eofb); |
9aa05000 | 693 | goto write_retry; |
dda35b8f | 694 | } |
d0606464 | 695 | |
dda35b8f | 696 | current->backing_dev_info = NULL; |
d0606464 | 697 | out: |
c3155097 BF |
698 | if (iolock) |
699 | xfs_iunlock(ip, iolock); | |
637bbc75 DC |
700 | return ret; |
701 | } | |
702 | ||
703 | STATIC ssize_t | |
bf97f3bc | 704 | xfs_file_write_iter( |
637bbc75 | 705 | struct kiocb *iocb, |
bf97f3bc | 706 | struct iov_iter *from) |
637bbc75 DC |
707 | { |
708 | struct file *file = iocb->ki_filp; | |
709 | struct address_space *mapping = file->f_mapping; | |
710 | struct inode *inode = mapping->host; | |
711 | struct xfs_inode *ip = XFS_I(inode); | |
712 | ssize_t ret; | |
bf97f3bc | 713 | size_t ocount = iov_iter_count(from); |
637bbc75 | 714 | |
ff6d6af2 | 715 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 716 | |
637bbc75 DC |
717 | if (ocount == 0) |
718 | return 0; | |
719 | ||
bf97f3bc AV |
720 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
721 | return -EIO; | |
637bbc75 | 722 | |
16d4d435 CH |
723 | if (IS_DAX(inode)) |
724 | ret = xfs_file_dax_write(iocb, from); | |
0613f16c DW |
725 | else if (iocb->ki_flags & IOCB_DIRECT) { |
726 | /* | |
727 | * Allow a directio write to fall back to a buffered | |
728 | * write *only* in the case that we're doing a reflink | |
729 | * CoW. In all other directio scenarios we do not | |
730 | * allow an operation to fall back to buffered mode. | |
731 | */ | |
bf97f3bc | 732 | ret = xfs_file_dio_aio_write(iocb, from); |
0613f16c DW |
733 | if (ret == -EREMCHG) |
734 | goto buffered; | |
735 | } else { | |
736 | buffered: | |
bf97f3bc | 737 | ret = xfs_file_buffered_aio_write(iocb, from); |
0613f16c | 738 | } |
dda35b8f | 739 | |
d0606464 | 740 | if (ret > 0) { |
ff6d6af2 | 741 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 742 | |
d0606464 | 743 | /* Handle various SYNC-type writes */ |
e2592217 | 744 | ret = generic_write_sync(iocb, ret); |
dda35b8f | 745 | } |
a363f0c2 | 746 | return ret; |
dda35b8f CH |
747 | } |
748 | ||
a904b1ca NJ |
749 | #define XFS_FALLOC_FL_SUPPORTED \ |
750 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
751 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
98cc2db5 | 752 | FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) |
a904b1ca | 753 | |
2fe17c10 CH |
754 | STATIC long |
755 | xfs_file_fallocate( | |
83aee9e4 CH |
756 | struct file *file, |
757 | int mode, | |
758 | loff_t offset, | |
759 | loff_t len) | |
2fe17c10 | 760 | { |
83aee9e4 CH |
761 | struct inode *inode = file_inode(file); |
762 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 763 | long error; |
8add71ca | 764 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 765 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 766 | loff_t new_size = 0; |
a904b1ca | 767 | bool do_file_insert = 0; |
2fe17c10 | 768 | |
83aee9e4 CH |
769 | if (!S_ISREG(inode->i_mode)) |
770 | return -EINVAL; | |
a904b1ca | 771 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
772 | return -EOPNOTSUPP; |
773 | ||
781355c6 | 774 | xfs_ilock(ip, iolock); |
65523218 | 775 | error = xfs_break_layouts(inode, &iolock); |
781355c6 CH |
776 | if (error) |
777 | goto out_unlock; | |
778 | ||
e8e9ad42 DC |
779 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
780 | iolock |= XFS_MMAPLOCK_EXCL; | |
781 | ||
83aee9e4 CH |
782 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
783 | error = xfs_free_file_space(ip, offset, len); | |
784 | if (error) | |
785 | goto out_unlock; | |
e1d8fb88 | 786 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
93407472 | 787 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
e1d8fb88 NJ |
788 | |
789 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 790 | error = -EINVAL; |
e1d8fb88 NJ |
791 | goto out_unlock; |
792 | } | |
793 | ||
23fffa92 LC |
794 | /* |
795 | * There is no need to overlap collapse range with EOF, | |
796 | * in which case it is effectively a truncate operation | |
797 | */ | |
798 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 799 | error = -EINVAL; |
23fffa92 LC |
800 | goto out_unlock; |
801 | } | |
802 | ||
e1d8fb88 NJ |
803 | new_size = i_size_read(inode) - len; |
804 | ||
805 | error = xfs_collapse_file_space(ip, offset, len); | |
806 | if (error) | |
807 | goto out_unlock; | |
a904b1ca | 808 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
93407472 | 809 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
a904b1ca NJ |
810 | |
811 | new_size = i_size_read(inode) + len; | |
812 | if (offset & blksize_mask || len & blksize_mask) { | |
813 | error = -EINVAL; | |
814 | goto out_unlock; | |
815 | } | |
816 | ||
817 | /* check the new inode size does not wrap through zero */ | |
818 | if (new_size > inode->i_sb->s_maxbytes) { | |
819 | error = -EFBIG; | |
820 | goto out_unlock; | |
821 | } | |
822 | ||
823 | /* Offset should be less than i_size */ | |
824 | if (offset >= i_size_read(inode)) { | |
825 | error = -EINVAL; | |
826 | goto out_unlock; | |
827 | } | |
828 | do_file_insert = 1; | |
83aee9e4 | 829 | } else { |
8add71ca CH |
830 | flags |= XFS_PREALLOC_SET; |
831 | ||
83aee9e4 CH |
832 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
833 | offset + len > i_size_read(inode)) { | |
834 | new_size = offset + len; | |
2451337d | 835 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
836 | if (error) |
837 | goto out_unlock; | |
838 | } | |
2fe17c10 | 839 | |
376ba313 LC |
840 | if (mode & FALLOC_FL_ZERO_RANGE) |
841 | error = xfs_zero_file_space(ip, offset, len); | |
98cc2db5 DW |
842 | else { |
843 | if (mode & FALLOC_FL_UNSHARE_RANGE) { | |
844 | error = xfs_reflink_unshare(ip, offset, len); | |
845 | if (error) | |
846 | goto out_unlock; | |
847 | } | |
376ba313 LC |
848 | error = xfs_alloc_file_space(ip, offset, len, |
849 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 850 | } |
2fe17c10 CH |
851 | if (error) |
852 | goto out_unlock; | |
853 | } | |
854 | ||
83aee9e4 | 855 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
856 | flags |= XFS_PREALLOC_SYNC; |
857 | ||
858 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
859 | if (error) |
860 | goto out_unlock; | |
861 | ||
862 | /* Change file size if needed */ | |
863 | if (new_size) { | |
864 | struct iattr iattr; | |
865 | ||
866 | iattr.ia_valid = ATTR_SIZE; | |
867 | iattr.ia_size = new_size; | |
69bca807 | 868 | error = xfs_vn_setattr_size(file_dentry(file), &iattr); |
a904b1ca NJ |
869 | if (error) |
870 | goto out_unlock; | |
2fe17c10 CH |
871 | } |
872 | ||
a904b1ca NJ |
873 | /* |
874 | * Perform hole insertion now that the file size has been | |
875 | * updated so that if we crash during the operation we don't | |
876 | * leave shifted extents past EOF and hence losing access to | |
877 | * the data that is contained within them. | |
878 | */ | |
879 | if (do_file_insert) | |
880 | error = xfs_insert_file_space(ip, offset, len); | |
881 | ||
2fe17c10 | 882 | out_unlock: |
781355c6 | 883 | xfs_iunlock(ip, iolock); |
2451337d | 884 | return error; |
2fe17c10 CH |
885 | } |
886 | ||
9fe26045 DW |
887 | STATIC int |
888 | xfs_file_clone_range( | |
889 | struct file *file_in, | |
890 | loff_t pos_in, | |
891 | struct file *file_out, | |
892 | loff_t pos_out, | |
893 | u64 len) | |
894 | { | |
5faaf4fa | 895 | return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out, |
cc714660 DW |
896 | len, false); |
897 | } | |
898 | ||
cc714660 DW |
899 | STATIC ssize_t |
900 | xfs_file_dedupe_range( | |
901 | struct file *src_file, | |
902 | u64 loff, | |
903 | u64 len, | |
904 | struct file *dst_file, | |
905 | u64 dst_loff) | |
906 | { | |
907 | int error; | |
908 | ||
5faaf4fa | 909 | error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff, |
cc714660 DW |
910 | len, true); |
911 | if (error) | |
912 | return error; | |
913 | return len; | |
9fe26045 | 914 | } |
2fe17c10 | 915 | |
1da177e4 | 916 | STATIC int |
3562fd45 | 917 | xfs_file_open( |
1da177e4 | 918 | struct inode *inode, |
f999a5bf | 919 | struct file *file) |
1da177e4 | 920 | { |
f999a5bf | 921 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 922 | return -EFBIG; |
f999a5bf CH |
923 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
924 | return -EIO; | |
91f9943e | 925 | file->f_mode |= FMODE_NOWAIT; |
f999a5bf CH |
926 | return 0; |
927 | } | |
928 | ||
929 | STATIC int | |
930 | xfs_dir_open( | |
931 | struct inode *inode, | |
932 | struct file *file) | |
933 | { | |
934 | struct xfs_inode *ip = XFS_I(inode); | |
935 | int mode; | |
936 | int error; | |
937 | ||
938 | error = xfs_file_open(inode, file); | |
939 | if (error) | |
940 | return error; | |
941 | ||
942 | /* | |
943 | * If there are any blocks, read-ahead block 0 as we're almost | |
944 | * certain to have the next operation be a read there. | |
945 | */ | |
309ecac8 | 946 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 947 | if (ip->i_d.di_nextents > 0) |
7a652bbe | 948 | error = xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf | 949 | xfs_iunlock(ip, mode); |
7a652bbe | 950 | return error; |
1da177e4 LT |
951 | } |
952 | ||
1da177e4 | 953 | STATIC int |
3562fd45 | 954 | xfs_file_release( |
1da177e4 LT |
955 | struct inode *inode, |
956 | struct file *filp) | |
957 | { | |
2451337d | 958 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
959 | } |
960 | ||
1da177e4 | 961 | STATIC int |
3562fd45 | 962 | xfs_file_readdir( |
b8227554 AV |
963 | struct file *file, |
964 | struct dir_context *ctx) | |
1da177e4 | 965 | { |
b8227554 | 966 | struct inode *inode = file_inode(file); |
739bfb2a | 967 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
968 | size_t bufsize; |
969 | ||
970 | /* | |
971 | * The Linux API doesn't pass down the total size of the buffer | |
972 | * we read into down to the filesystem. With the filldir concept | |
973 | * it's not needed for correct information, but the XFS dir2 leaf | |
974 | * code wants an estimate of the buffer size to calculate it's | |
975 | * readahead window and size the buffers used for mapping to | |
976 | * physical blocks. | |
977 | * | |
978 | * Try to give it an estimate that's good enough, maybe at some | |
979 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 980 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 981 | */ |
a9cc799e | 982 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 983 | |
acb9553c | 984 | return xfs_readdir(NULL, ip, ctx, bufsize); |
3fe3e6b1 JL |
985 | } |
986 | ||
987 | STATIC loff_t | |
988 | xfs_file_llseek( | |
989 | struct file *file, | |
990 | loff_t offset, | |
59f9c004 | 991 | int whence) |
3fe3e6b1 | 992 | { |
9b2970aa CH |
993 | struct inode *inode = file->f_mapping->host; |
994 | ||
995 | if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount)) | |
996 | return -EIO; | |
997 | ||
59f9c004 | 998 | switch (whence) { |
9b2970aa | 999 | default: |
59f9c004 | 1000 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1001 | case SEEK_HOLE: |
9b2970aa CH |
1002 | offset = iomap_seek_hole(inode, offset, &xfs_iomap_ops); |
1003 | break; | |
49c69591 | 1004 | case SEEK_DATA: |
9b2970aa CH |
1005 | offset = iomap_seek_data(inode, offset, &xfs_iomap_ops); |
1006 | break; | |
3fe3e6b1 | 1007 | } |
9b2970aa CH |
1008 | |
1009 | if (offset < 0) | |
1010 | return offset; | |
1011 | return vfs_setpos(file, offset, inode->i_sb->s_maxbytes); | |
3fe3e6b1 JL |
1012 | } |
1013 | ||
de0e8c20 DC |
1014 | /* |
1015 | * Locking for serialisation of IO during page faults. This results in a lock | |
1016 | * ordering of: | |
1017 | * | |
1018 | * mmap_sem (MM) | |
6b698ede | 1019 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1020 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1021 | * page_lock (MM) |
1022 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1023 | */ |
d522d569 CH |
1024 | static int |
1025 | __xfs_filemap_fault( | |
1026 | struct vm_fault *vmf, | |
1027 | enum page_entry_size pe_size, | |
1028 | bool write_fault) | |
de0e8c20 | 1029 | { |
11bac800 | 1030 | struct inode *inode = file_inode(vmf->vma->vm_file); |
d522d569 | 1031 | struct xfs_inode *ip = XFS_I(inode); |
ec56b1f1 | 1032 | int ret; |
de0e8c20 | 1033 | |
d522d569 | 1034 | trace_xfs_filemap_fault(ip, pe_size, write_fault); |
de0e8c20 | 1035 | |
d522d569 CH |
1036 | if (write_fault) { |
1037 | sb_start_pagefault(inode->i_sb); | |
1038 | file_update_time(vmf->vma->vm_file); | |
1039 | } | |
de0e8c20 | 1040 | |
d522d569 | 1041 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1042 | if (IS_DAX(inode)) { |
9a0dd422 | 1043 | ret = dax_iomap_fault(vmf, pe_size, NULL, &xfs_iomap_ops); |
6b698ede | 1044 | } else { |
d522d569 CH |
1045 | if (write_fault) |
1046 | ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops); | |
1047 | else | |
1048 | ret = filemap_fault(vmf); | |
6b698ede | 1049 | } |
6b698ede | 1050 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
6b698ede | 1051 | |
d522d569 CH |
1052 | if (write_fault) |
1053 | sb_end_pagefault(inode->i_sb); | |
6b698ede | 1054 | return ret; |
de0e8c20 DC |
1055 | } |
1056 | ||
d522d569 | 1057 | static int |
6b698ede | 1058 | xfs_filemap_fault( |
075a924d DC |
1059 | struct vm_fault *vmf) |
1060 | { | |
6b698ede | 1061 | /* DAX can shortcut the normal fault path on write faults! */ |
d522d569 CH |
1062 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, |
1063 | IS_DAX(file_inode(vmf->vma->vm_file)) && | |
1064 | (vmf->flags & FAULT_FLAG_WRITE)); | |
6b698ede DC |
1065 | } |
1066 | ||
d522d569 | 1067 | static int |
a2d58167 | 1068 | xfs_filemap_huge_fault( |
c791ace1 DJ |
1069 | struct vm_fault *vmf, |
1070 | enum page_entry_size pe_size) | |
acd76e74 | 1071 | { |
d522d569 | 1072 | if (!IS_DAX(file_inode(vmf->vma->vm_file))) |
acd76e74 MW |
1073 | return VM_FAULT_FALLBACK; |
1074 | ||
d522d569 CH |
1075 | /* DAX can shortcut the normal fault path on write faults! */ |
1076 | return __xfs_filemap_fault(vmf, pe_size, | |
1077 | (vmf->flags & FAULT_FLAG_WRITE)); | |
1078 | } | |
acd76e74 | 1079 | |
d522d569 CH |
1080 | static int |
1081 | xfs_filemap_page_mkwrite( | |
1082 | struct vm_fault *vmf) | |
1083 | { | |
1084 | return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true); | |
acd76e74 MW |
1085 | } |
1086 | ||
3af49285 DC |
1087 | /* |
1088 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1089 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1090 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1091 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1092 | */ |
1093 | static int | |
1094 | xfs_filemap_pfn_mkwrite( | |
3af49285 DC |
1095 | struct vm_fault *vmf) |
1096 | { | |
1097 | ||
11bac800 | 1098 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3af49285 DC |
1099 | struct xfs_inode *ip = XFS_I(inode); |
1100 | int ret = VM_FAULT_NOPAGE; | |
1101 | loff_t size; | |
1102 | ||
1103 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1104 | ||
1105 | sb_start_pagefault(inode->i_sb); | |
11bac800 | 1106 | file_update_time(vmf->vma->vm_file); |
3af49285 DC |
1107 | |
1108 | /* check if the faulting page hasn't raced with truncate */ | |
1109 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1110 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1111 | if (vmf->pgoff >= size) | |
1112 | ret = VM_FAULT_SIGBUS; | |
5eb88dca | 1113 | else if (IS_DAX(inode)) |
9a0dd422 | 1114 | ret = dax_iomap_fault(vmf, PE_SIZE_PTE, NULL, &xfs_iomap_ops); |
3af49285 DC |
1115 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1116 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1117 | return ret; |
3af49285 | 1118 | |
acd76e74 MW |
1119 | } |
1120 | ||
6b698ede DC |
1121 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1122 | .fault = xfs_filemap_fault, | |
a2d58167 | 1123 | .huge_fault = xfs_filemap_huge_fault, |
6b698ede DC |
1124 | .map_pages = filemap_map_pages, |
1125 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1126 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1127 | }; |
1128 | ||
1129 | STATIC int | |
1130 | xfs_file_mmap( | |
1131 | struct file *filp, | |
1132 | struct vm_area_struct *vma) | |
1133 | { | |
1134 | file_accessed(filp); | |
1135 | vma->vm_ops = &xfs_file_vm_ops; | |
1136 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1137 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1138 | return 0; |
075a924d DC |
1139 | } |
1140 | ||
4b6f5d20 | 1141 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1142 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1143 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1144 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1145 | .splice_read = generic_file_splice_read, |
8d020765 | 1146 | .splice_write = iter_file_splice_write, |
3562fd45 | 1147 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1148 | #ifdef CONFIG_COMPAT |
3562fd45 | 1149 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1150 | #endif |
3562fd45 NS |
1151 | .mmap = xfs_file_mmap, |
1152 | .open = xfs_file_open, | |
1153 | .release = xfs_file_release, | |
1154 | .fsync = xfs_file_fsync, | |
dbe6ec81 | 1155 | .get_unmapped_area = thp_get_unmapped_area, |
2fe17c10 | 1156 | .fallocate = xfs_file_fallocate, |
9fe26045 | 1157 | .clone_file_range = xfs_file_clone_range, |
cc714660 | 1158 | .dedupe_file_range = xfs_file_dedupe_range, |
1da177e4 LT |
1159 | }; |
1160 | ||
4b6f5d20 | 1161 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1162 | .open = xfs_dir_open, |
1da177e4 | 1163 | .read = generic_read_dir, |
3b0a3c1a | 1164 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1165 | .llseek = generic_file_llseek, |
3562fd45 | 1166 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1167 | #ifdef CONFIG_COMPAT |
3562fd45 | 1168 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1169 | #endif |
1da2f2db | 1170 | .fsync = xfs_dir_fsync, |
1da177e4 | 1171 | }; |