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 | { |
459f0fbc | 61 | return iomap_zero_range(VFS_I(ip), pos, count, NULL, &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 | |
02c24a82 JB |
143 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
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 | ||
65523218 | 240 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
11c59c92 | 241 | ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops); |
65523218 | 242 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 | 243 | |
f1285ff0 | 244 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
245 | return ret; |
246 | } | |
247 | ||
248 | STATIC ssize_t | |
249 | xfs_file_buffered_aio_read( | |
250 | struct kiocb *iocb, | |
251 | struct iov_iter *to) | |
252 | { | |
253 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
254 | ssize_t ret; | |
255 | ||
256 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
dda35b8f | 257 | |
65523218 | 258 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
b4f5d2c6 | 259 | ret = generic_file_read_iter(iocb, to); |
65523218 | 260 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
bbc5a740 CH |
261 | |
262 | return ret; | |
263 | } | |
264 | ||
265 | STATIC ssize_t | |
266 | xfs_file_read_iter( | |
267 | struct kiocb *iocb, | |
268 | struct iov_iter *to) | |
269 | { | |
16d4d435 CH |
270 | struct inode *inode = file_inode(iocb->ki_filp); |
271 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
272 | ssize_t ret = 0; |
273 | ||
274 | XFS_STATS_INC(mp, xs_read_calls); | |
275 | ||
276 | if (XFS_FORCED_SHUTDOWN(mp)) | |
277 | return -EIO; | |
278 | ||
16d4d435 CH |
279 | if (IS_DAX(inode)) |
280 | ret = xfs_file_dax_read(iocb, to); | |
281 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 282 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 283 | else |
bbc5a740 | 284 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 285 | |
dda35b8f | 286 | if (ret > 0) |
ff6d6af2 | 287 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
288 | return ret; |
289 | } | |
290 | ||
dda35b8f | 291 | /* |
193aec10 CH |
292 | * Zero any on disk space between the current EOF and the new, larger EOF. |
293 | * | |
294 | * This handles the normal case of zeroing the remainder of the last block in | |
295 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
296 | * file. This second case only happens with fixed size extents and when the | |
297 | * system crashes before the inode size was updated but after blocks were | |
298 | * allocated. | |
299 | * | |
300 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 301 | */ |
dda35b8f CH |
302 | int /* error (positive) */ |
303 | xfs_zero_eof( | |
193aec10 CH |
304 | struct xfs_inode *ip, |
305 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
306 | xfs_fsize_t isize, /* current inode size */ |
307 | bool *did_zeroing) | |
dda35b8f | 308 | { |
193aec10 | 309 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
dda35b8f CH |
310 | ASSERT(offset > isize); |
311 | ||
0a50f162 | 312 | trace_xfs_zero_eof(ip, isize, offset - isize); |
570b6211 | 313 | return xfs_zero_range(ip, isize, offset - isize, did_zeroing); |
dda35b8f CH |
314 | } |
315 | ||
4d8d1581 DC |
316 | /* |
317 | * Common pre-write limit and setup checks. | |
318 | * | |
5bf1f262 CH |
319 | * Called with the iolocked held either shared and exclusive according to |
320 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
321 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
322 | */ |
323 | STATIC ssize_t | |
324 | xfs_file_aio_write_checks( | |
99733fa3 AV |
325 | struct kiocb *iocb, |
326 | struct iov_iter *from, | |
4d8d1581 DC |
327 | int *iolock) |
328 | { | |
99733fa3 | 329 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
330 | struct inode *inode = file->f_mapping->host; |
331 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 332 | ssize_t error = 0; |
99733fa3 | 333 | size_t count = iov_iter_count(from); |
3136e8bb | 334 | bool drained_dio = false; |
4d8d1581 | 335 | |
7271d243 | 336 | restart: |
3309dd04 AV |
337 | error = generic_write_checks(iocb, from); |
338 | if (error <= 0) | |
4d8d1581 | 339 | return error; |
4d8d1581 | 340 | |
65523218 | 341 | error = xfs_break_layouts(inode, iolock); |
781355c6 CH |
342 | if (error) |
343 | return error; | |
344 | ||
65523218 CH |
345 | /* |
346 | * For changing security info in file_remove_privs() we need i_rwsem | |
347 | * exclusively. | |
348 | */ | |
a6de82ca | 349 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
65523218 | 350 | xfs_iunlock(ip, *iolock); |
a6de82ca | 351 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 352 | xfs_ilock(ip, *iolock); |
a6de82ca JK |
353 | goto restart; |
354 | } | |
4d8d1581 DC |
355 | /* |
356 | * If the offset is beyond the size of the file, we need to zero any | |
357 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 358 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
359 | * iolock shared, we need to update it to exclusive which implies |
360 | * having to redo all checks before. | |
b9d59846 DC |
361 | * |
362 | * We need to serialise against EOF updates that occur in IO | |
363 | * completions here. We want to make sure that nobody is changing the | |
364 | * size while we do this check until we have placed an IO barrier (i.e. | |
365 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
366 | * The spinlock effectively forms a memory barrier once we have the | |
367 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
368 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 369 | */ |
b9d59846 | 370 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 371 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
372 | bool zero = false; |
373 | ||
b9d59846 | 374 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
375 | if (!drained_dio) { |
376 | if (*iolock == XFS_IOLOCK_SHARED) { | |
65523218 | 377 | xfs_iunlock(ip, *iolock); |
3136e8bb | 378 | *iolock = XFS_IOLOCK_EXCL; |
65523218 | 379 | xfs_ilock(ip, *iolock); |
3136e8bb BF |
380 | iov_iter_reexpand(from, count); |
381 | } | |
40c63fbc DC |
382 | /* |
383 | * We now have an IO submission barrier in place, but | |
384 | * AIO can do EOF updates during IO completion and hence | |
385 | * we now need to wait for all of them to drain. Non-AIO | |
386 | * DIO will have drained before we are given the | |
387 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
388 | * no-op. | |
389 | */ | |
390 | inode_dio_wait(inode); | |
3136e8bb | 391 | drained_dio = true; |
7271d243 DC |
392 | goto restart; |
393 | } | |
99733fa3 | 394 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
395 | if (error) |
396 | return error; | |
b9d59846 DC |
397 | } else |
398 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 399 | |
8a9c9980 CH |
400 | /* |
401 | * Updating the timestamps will grab the ilock again from | |
402 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
403 | * lock above. Eventually we should look into a way to avoid | |
404 | * the pointless lock roundtrip. | |
405 | */ | |
c3b2da31 JB |
406 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
407 | error = file_update_time(file); | |
408 | if (error) | |
409 | return error; | |
410 | } | |
8a9c9980 | 411 | |
4d8d1581 DC |
412 | /* |
413 | * If we're writing the file then make sure to clear the setuid and | |
414 | * setgid bits if the process is not being run by root. This keeps | |
415 | * people from modifying setuid and setgid binaries. | |
416 | */ | |
a6de82ca JK |
417 | if (!IS_NOSEC(inode)) |
418 | return file_remove_privs(file); | |
419 | return 0; | |
4d8d1581 DC |
420 | } |
421 | ||
acdda3aa CH |
422 | static int |
423 | xfs_dio_write_end_io( | |
424 | struct kiocb *iocb, | |
425 | ssize_t size, | |
426 | unsigned flags) | |
427 | { | |
428 | struct inode *inode = file_inode(iocb->ki_filp); | |
429 | struct xfs_inode *ip = XFS_I(inode); | |
430 | loff_t offset = iocb->ki_pos; | |
431 | bool update_size = false; | |
432 | int error = 0; | |
433 | ||
434 | trace_xfs_end_io_direct_write(ip, offset, size); | |
435 | ||
436 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
437 | return -EIO; | |
438 | ||
439 | if (size <= 0) | |
440 | return size; | |
441 | ||
442 | /* | |
443 | * We need to update the in-core inode size here so that we don't end up | |
444 | * with the on-disk inode size being outside the in-core inode size. We | |
445 | * have no other method of updating EOF for AIO, so always do it here | |
446 | * if necessary. | |
447 | * | |
448 | * We need to lock the test/set EOF update as we can be racing with | |
449 | * other IO completions here to update the EOF. Failing to serialise | |
450 | * here can result in EOF moving backwards and Bad Things Happen when | |
451 | * that occurs. | |
452 | */ | |
453 | spin_lock(&ip->i_flags_lock); | |
454 | if (offset + size > i_size_read(inode)) { | |
455 | i_size_write(inode, offset + size); | |
456 | update_size = true; | |
457 | } | |
458 | spin_unlock(&ip->i_flags_lock); | |
459 | ||
460 | if (flags & IOMAP_DIO_COW) { | |
461 | error = xfs_reflink_end_cow(ip, offset, size); | |
462 | if (error) | |
463 | return error; | |
464 | } | |
465 | ||
466 | if (flags & IOMAP_DIO_UNWRITTEN) | |
467 | error = xfs_iomap_write_unwritten(ip, offset, size); | |
468 | else if (update_size) | |
469 | error = xfs_setfilesize(ip, offset, size); | |
470 | ||
471 | return error; | |
472 | } | |
473 | ||
f0d26e86 DC |
474 | /* |
475 | * xfs_file_dio_aio_write - handle direct IO writes | |
476 | * | |
477 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 478 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
479 | * follow locking changes and looping. |
480 | * | |
eda77982 DC |
481 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
482 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
483 | * pages are flushed out. | |
484 | * | |
485 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
486 | * allowing them to be done in parallel with reads and other direct IO writes. | |
487 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
488 | * needs to do sub-block zeroing and that requires serialisation against other | |
489 | * direct IOs to the same block. In this case we need to serialise the | |
490 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
491 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
492 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
493 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 494 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 495 | * |
f0d26e86 DC |
496 | * Returns with locks held indicated by @iolock and errors indicated by |
497 | * negative return values. | |
498 | */ | |
499 | STATIC ssize_t | |
500 | xfs_file_dio_aio_write( | |
501 | struct kiocb *iocb, | |
b3188919 | 502 | struct iov_iter *from) |
f0d26e86 DC |
503 | { |
504 | struct file *file = iocb->ki_filp; | |
505 | struct address_space *mapping = file->f_mapping; | |
506 | struct inode *inode = mapping->host; | |
507 | struct xfs_inode *ip = XFS_I(inode); | |
508 | struct xfs_mount *mp = ip->i_mount; | |
509 | ssize_t ret = 0; | |
eda77982 | 510 | int unaligned_io = 0; |
d0606464 | 511 | int iolock; |
b3188919 | 512 | size_t count = iov_iter_count(from); |
acdda3aa | 513 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
f0d26e86 DC |
514 | mp->m_rtdev_targp : mp->m_ddev_targp; |
515 | ||
7c71ee78 | 516 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 517 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 518 | return -EINVAL; |
f0d26e86 | 519 | |
7271d243 | 520 | /* |
0ee7a3f6 CH |
521 | * Don't take the exclusive iolock here unless the I/O is unaligned to |
522 | * the file system block size. We don't need to consider the EOF | |
523 | * extension case here because xfs_file_aio_write_checks() will relock | |
524 | * the inode as necessary for EOF zeroing cases and fill out the new | |
525 | * inode size as appropriate. | |
7271d243 | 526 | */ |
0ee7a3f6 CH |
527 | if ((iocb->ki_pos & mp->m_blockmask) || |
528 | ((iocb->ki_pos + count) & mp->m_blockmask)) { | |
529 | unaligned_io = 1; | |
54a4ef8a CH |
530 | |
531 | /* | |
532 | * We can't properly handle unaligned direct I/O to reflink | |
533 | * files yet, as we can't unshare a partial block. | |
534 | */ | |
535 | if (xfs_is_reflink_inode(ip)) { | |
536 | trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count); | |
537 | return -EREMCHG; | |
538 | } | |
d0606464 | 539 | iolock = XFS_IOLOCK_EXCL; |
0ee7a3f6 | 540 | } else { |
d0606464 | 541 | iolock = XFS_IOLOCK_SHARED; |
c58cb165 | 542 | } |
f0d26e86 | 543 | |
65523218 | 544 | xfs_ilock(ip, iolock); |
0ee7a3f6 | 545 | |
99733fa3 | 546 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 547 | if (ret) |
d0606464 | 548 | goto out; |
99733fa3 | 549 | count = iov_iter_count(from); |
f0d26e86 | 550 | |
eda77982 DC |
551 | /* |
552 | * If we are doing unaligned IO, wait for all other IO to drain, | |
0ee7a3f6 CH |
553 | * otherwise demote the lock if we had to take the exclusive lock |
554 | * for other reasons in xfs_file_aio_write_checks. | |
eda77982 DC |
555 | */ |
556 | if (unaligned_io) | |
4a06fd26 | 557 | inode_dio_wait(inode); |
d0606464 | 558 | else if (iolock == XFS_IOLOCK_EXCL) { |
65523218 | 559 | xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 560 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
561 | } |
562 | ||
3176c3e0 | 563 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
acdda3aa | 564 | ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, xfs_dio_write_end_io); |
d0606464 | 565 | out: |
65523218 | 566 | xfs_iunlock(ip, iolock); |
d0606464 | 567 | |
6b698ede | 568 | /* |
16d4d435 CH |
569 | * No fallback to buffered IO on errors for XFS, direct IO will either |
570 | * complete fully or fail. | |
6b698ede | 571 | */ |
16d4d435 CH |
572 | ASSERT(ret < 0 || ret == count); |
573 | return ret; | |
574 | } | |
575 | ||
f021bd07 | 576 | static noinline ssize_t |
16d4d435 CH |
577 | xfs_file_dax_write( |
578 | struct kiocb *iocb, | |
579 | struct iov_iter *from) | |
580 | { | |
6c31f495 | 581 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
16d4d435 | 582 | struct xfs_inode *ip = XFS_I(inode); |
17879e8f | 583 | int iolock = XFS_IOLOCK_EXCL; |
6c31f495 CH |
584 | ssize_t ret, error = 0; |
585 | size_t count; | |
586 | loff_t pos; | |
16d4d435 | 587 | |
65523218 | 588 | xfs_ilock(ip, iolock); |
16d4d435 CH |
589 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
590 | if (ret) | |
591 | goto out; | |
592 | ||
6c31f495 CH |
593 | pos = iocb->ki_pos; |
594 | count = iov_iter_count(from); | |
8b2180b3 | 595 | |
6c31f495 | 596 | trace_xfs_file_dax_write(ip, count, pos); |
11c59c92 | 597 | ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops); |
6c31f495 CH |
598 | if (ret > 0 && iocb->ki_pos > i_size_read(inode)) { |
599 | i_size_write(inode, iocb->ki_pos); | |
600 | error = xfs_setfilesize(ip, pos, ret); | |
16d4d435 | 601 | } |
16d4d435 | 602 | out: |
65523218 | 603 | xfs_iunlock(ip, iolock); |
6c31f495 | 604 | return error ? error : ret; |
f0d26e86 DC |
605 | } |
606 | ||
00258e36 | 607 | STATIC ssize_t |
637bbc75 | 608 | xfs_file_buffered_aio_write( |
dda35b8f | 609 | struct kiocb *iocb, |
b3188919 | 610 | struct iov_iter *from) |
dda35b8f CH |
611 | { |
612 | struct file *file = iocb->ki_filp; | |
613 | struct address_space *mapping = file->f_mapping; | |
614 | struct inode *inode = mapping->host; | |
00258e36 | 615 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
616 | ssize_t ret; |
617 | int enospc = 0; | |
c3155097 | 618 | int iolock; |
dda35b8f | 619 | |
c3155097 BF |
620 | write_retry: |
621 | iolock = XFS_IOLOCK_EXCL; | |
65523218 | 622 | xfs_ilock(ip, iolock); |
dda35b8f | 623 | |
99733fa3 | 624 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 625 | if (ret) |
d0606464 | 626 | goto out; |
dda35b8f CH |
627 | |
628 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 629 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 630 | |
3176c3e0 | 631 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
68a9f5e7 | 632 | ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops); |
0a64bc2c | 633 | if (likely(ret >= 0)) |
99733fa3 | 634 | iocb->ki_pos += ret; |
dc06f398 | 635 | |
637bbc75 | 636 | /* |
dc06f398 BF |
637 | * If we hit a space limit, try to free up some lingering preallocated |
638 | * space before returning an error. In the case of ENOSPC, first try to | |
639 | * write back all dirty inodes to free up some of the excess reserved | |
640 | * metadata space. This reduces the chances that the eofblocks scan | |
641 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
642 | * also behaves as a filter to prevent too many eofblocks scans from | |
643 | * running at the same time. | |
637bbc75 | 644 | */ |
dc06f398 | 645 | if (ret == -EDQUOT && !enospc) { |
c3155097 | 646 | xfs_iunlock(ip, iolock); |
dc06f398 BF |
647 | enospc = xfs_inode_free_quota_eofblocks(ip); |
648 | if (enospc) | |
649 | goto write_retry; | |
83104d44 DW |
650 | enospc = xfs_inode_free_quota_cowblocks(ip); |
651 | if (enospc) | |
652 | goto write_retry; | |
c3155097 | 653 | iolock = 0; |
dc06f398 BF |
654 | } else if (ret == -ENOSPC && !enospc) { |
655 | struct xfs_eofblocks eofb = {0}; | |
656 | ||
637bbc75 | 657 | enospc = 1; |
9aa05000 | 658 | xfs_flush_inodes(ip->i_mount); |
c3155097 BF |
659 | |
660 | xfs_iunlock(ip, iolock); | |
dc06f398 BF |
661 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; |
662 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 663 | goto write_retry; |
dda35b8f | 664 | } |
d0606464 | 665 | |
dda35b8f | 666 | current->backing_dev_info = NULL; |
d0606464 | 667 | out: |
c3155097 BF |
668 | if (iolock) |
669 | xfs_iunlock(ip, iolock); | |
637bbc75 DC |
670 | return ret; |
671 | } | |
672 | ||
673 | STATIC ssize_t | |
bf97f3bc | 674 | xfs_file_write_iter( |
637bbc75 | 675 | struct kiocb *iocb, |
bf97f3bc | 676 | struct iov_iter *from) |
637bbc75 DC |
677 | { |
678 | struct file *file = iocb->ki_filp; | |
679 | struct address_space *mapping = file->f_mapping; | |
680 | struct inode *inode = mapping->host; | |
681 | struct xfs_inode *ip = XFS_I(inode); | |
682 | ssize_t ret; | |
bf97f3bc | 683 | size_t ocount = iov_iter_count(from); |
637bbc75 | 684 | |
ff6d6af2 | 685 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 686 | |
637bbc75 DC |
687 | if (ocount == 0) |
688 | return 0; | |
689 | ||
bf97f3bc AV |
690 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
691 | return -EIO; | |
637bbc75 | 692 | |
16d4d435 CH |
693 | if (IS_DAX(inode)) |
694 | ret = xfs_file_dax_write(iocb, from); | |
0613f16c DW |
695 | else if (iocb->ki_flags & IOCB_DIRECT) { |
696 | /* | |
697 | * Allow a directio write to fall back to a buffered | |
698 | * write *only* in the case that we're doing a reflink | |
699 | * CoW. In all other directio scenarios we do not | |
700 | * allow an operation to fall back to buffered mode. | |
701 | */ | |
bf97f3bc | 702 | ret = xfs_file_dio_aio_write(iocb, from); |
0613f16c DW |
703 | if (ret == -EREMCHG) |
704 | goto buffered; | |
705 | } else { | |
706 | buffered: | |
bf97f3bc | 707 | ret = xfs_file_buffered_aio_write(iocb, from); |
0613f16c | 708 | } |
dda35b8f | 709 | |
d0606464 | 710 | if (ret > 0) { |
ff6d6af2 | 711 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 712 | |
d0606464 | 713 | /* Handle various SYNC-type writes */ |
e2592217 | 714 | ret = generic_write_sync(iocb, ret); |
dda35b8f | 715 | } |
a363f0c2 | 716 | return ret; |
dda35b8f CH |
717 | } |
718 | ||
a904b1ca NJ |
719 | #define XFS_FALLOC_FL_SUPPORTED \ |
720 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
721 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
98cc2db5 | 722 | FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE) |
a904b1ca | 723 | |
2fe17c10 CH |
724 | STATIC long |
725 | xfs_file_fallocate( | |
83aee9e4 CH |
726 | struct file *file, |
727 | int mode, | |
728 | loff_t offset, | |
729 | loff_t len) | |
2fe17c10 | 730 | { |
83aee9e4 CH |
731 | struct inode *inode = file_inode(file); |
732 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 733 | long error; |
8add71ca | 734 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 735 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 736 | loff_t new_size = 0; |
a904b1ca | 737 | bool do_file_insert = 0; |
2fe17c10 | 738 | |
83aee9e4 CH |
739 | if (!S_ISREG(inode->i_mode)) |
740 | return -EINVAL; | |
a904b1ca | 741 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
742 | return -EOPNOTSUPP; |
743 | ||
781355c6 | 744 | xfs_ilock(ip, iolock); |
65523218 | 745 | error = xfs_break_layouts(inode, &iolock); |
781355c6 CH |
746 | if (error) |
747 | goto out_unlock; | |
748 | ||
e8e9ad42 DC |
749 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
750 | iolock |= XFS_MMAPLOCK_EXCL; | |
751 | ||
83aee9e4 CH |
752 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
753 | error = xfs_free_file_space(ip, offset, len); | |
754 | if (error) | |
755 | goto out_unlock; | |
e1d8fb88 | 756 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
93407472 | 757 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
e1d8fb88 NJ |
758 | |
759 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 760 | error = -EINVAL; |
e1d8fb88 NJ |
761 | goto out_unlock; |
762 | } | |
763 | ||
23fffa92 LC |
764 | /* |
765 | * There is no need to overlap collapse range with EOF, | |
766 | * in which case it is effectively a truncate operation | |
767 | */ | |
768 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 769 | error = -EINVAL; |
23fffa92 LC |
770 | goto out_unlock; |
771 | } | |
772 | ||
e1d8fb88 NJ |
773 | new_size = i_size_read(inode) - len; |
774 | ||
775 | error = xfs_collapse_file_space(ip, offset, len); | |
776 | if (error) | |
777 | goto out_unlock; | |
a904b1ca | 778 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
93407472 | 779 | unsigned int blksize_mask = i_blocksize(inode) - 1; |
a904b1ca NJ |
780 | |
781 | new_size = i_size_read(inode) + len; | |
782 | if (offset & blksize_mask || len & blksize_mask) { | |
783 | error = -EINVAL; | |
784 | goto out_unlock; | |
785 | } | |
786 | ||
787 | /* check the new inode size does not wrap through zero */ | |
788 | if (new_size > inode->i_sb->s_maxbytes) { | |
789 | error = -EFBIG; | |
790 | goto out_unlock; | |
791 | } | |
792 | ||
793 | /* Offset should be less than i_size */ | |
794 | if (offset >= i_size_read(inode)) { | |
795 | error = -EINVAL; | |
796 | goto out_unlock; | |
797 | } | |
798 | do_file_insert = 1; | |
83aee9e4 | 799 | } else { |
8add71ca CH |
800 | flags |= XFS_PREALLOC_SET; |
801 | ||
83aee9e4 CH |
802 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
803 | offset + len > i_size_read(inode)) { | |
804 | new_size = offset + len; | |
2451337d | 805 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
806 | if (error) |
807 | goto out_unlock; | |
808 | } | |
2fe17c10 | 809 | |
376ba313 LC |
810 | if (mode & FALLOC_FL_ZERO_RANGE) |
811 | error = xfs_zero_file_space(ip, offset, len); | |
98cc2db5 DW |
812 | else { |
813 | if (mode & FALLOC_FL_UNSHARE_RANGE) { | |
814 | error = xfs_reflink_unshare(ip, offset, len); | |
815 | if (error) | |
816 | goto out_unlock; | |
817 | } | |
376ba313 LC |
818 | error = xfs_alloc_file_space(ip, offset, len, |
819 | XFS_BMAPI_PREALLOC); | |
98cc2db5 | 820 | } |
2fe17c10 CH |
821 | if (error) |
822 | goto out_unlock; | |
823 | } | |
824 | ||
83aee9e4 | 825 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
826 | flags |= XFS_PREALLOC_SYNC; |
827 | ||
828 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
829 | if (error) |
830 | goto out_unlock; | |
831 | ||
832 | /* Change file size if needed */ | |
833 | if (new_size) { | |
834 | struct iattr iattr; | |
835 | ||
836 | iattr.ia_valid = ATTR_SIZE; | |
837 | iattr.ia_size = new_size; | |
69bca807 | 838 | error = xfs_vn_setattr_size(file_dentry(file), &iattr); |
a904b1ca NJ |
839 | if (error) |
840 | goto out_unlock; | |
2fe17c10 CH |
841 | } |
842 | ||
a904b1ca NJ |
843 | /* |
844 | * Perform hole insertion now that the file size has been | |
845 | * updated so that if we crash during the operation we don't | |
846 | * leave shifted extents past EOF and hence losing access to | |
847 | * the data that is contained within them. | |
848 | */ | |
849 | if (do_file_insert) | |
850 | error = xfs_insert_file_space(ip, offset, len); | |
851 | ||
2fe17c10 | 852 | out_unlock: |
781355c6 | 853 | xfs_iunlock(ip, iolock); |
2451337d | 854 | return error; |
2fe17c10 CH |
855 | } |
856 | ||
9fe26045 DW |
857 | STATIC int |
858 | xfs_file_clone_range( | |
859 | struct file *file_in, | |
860 | loff_t pos_in, | |
861 | struct file *file_out, | |
862 | loff_t pos_out, | |
863 | u64 len) | |
864 | { | |
5faaf4fa | 865 | return xfs_reflink_remap_range(file_in, pos_in, file_out, pos_out, |
cc714660 DW |
866 | len, false); |
867 | } | |
868 | ||
cc714660 DW |
869 | STATIC ssize_t |
870 | xfs_file_dedupe_range( | |
871 | struct file *src_file, | |
872 | u64 loff, | |
873 | u64 len, | |
874 | struct file *dst_file, | |
875 | u64 dst_loff) | |
876 | { | |
877 | int error; | |
878 | ||
5faaf4fa | 879 | error = xfs_reflink_remap_range(src_file, loff, dst_file, dst_loff, |
cc714660 DW |
880 | len, true); |
881 | if (error) | |
882 | return error; | |
883 | return len; | |
9fe26045 | 884 | } |
2fe17c10 | 885 | |
1da177e4 | 886 | STATIC int |
3562fd45 | 887 | xfs_file_open( |
1da177e4 | 888 | struct inode *inode, |
f999a5bf | 889 | struct file *file) |
1da177e4 | 890 | { |
f999a5bf | 891 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 892 | return -EFBIG; |
f999a5bf CH |
893 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
894 | return -EIO; | |
895 | return 0; | |
896 | } | |
897 | ||
898 | STATIC int | |
899 | xfs_dir_open( | |
900 | struct inode *inode, | |
901 | struct file *file) | |
902 | { | |
903 | struct xfs_inode *ip = XFS_I(inode); | |
904 | int mode; | |
905 | int error; | |
906 | ||
907 | error = xfs_file_open(inode, file); | |
908 | if (error) | |
909 | return error; | |
910 | ||
911 | /* | |
912 | * If there are any blocks, read-ahead block 0 as we're almost | |
913 | * certain to have the next operation be a read there. | |
914 | */ | |
309ecac8 | 915 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 916 | if (ip->i_d.di_nextents > 0) |
7a652bbe | 917 | error = xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf | 918 | xfs_iunlock(ip, mode); |
7a652bbe | 919 | return error; |
1da177e4 LT |
920 | } |
921 | ||
1da177e4 | 922 | STATIC int |
3562fd45 | 923 | xfs_file_release( |
1da177e4 LT |
924 | struct inode *inode, |
925 | struct file *filp) | |
926 | { | |
2451337d | 927 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
928 | } |
929 | ||
1da177e4 | 930 | STATIC int |
3562fd45 | 931 | xfs_file_readdir( |
b8227554 AV |
932 | struct file *file, |
933 | struct dir_context *ctx) | |
1da177e4 | 934 | { |
b8227554 | 935 | struct inode *inode = file_inode(file); |
739bfb2a | 936 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
937 | size_t bufsize; |
938 | ||
939 | /* | |
940 | * The Linux API doesn't pass down the total size of the buffer | |
941 | * we read into down to the filesystem. With the filldir concept | |
942 | * it's not needed for correct information, but the XFS dir2 leaf | |
943 | * code wants an estimate of the buffer size to calculate it's | |
944 | * readahead window and size the buffers used for mapping to | |
945 | * physical blocks. | |
946 | * | |
947 | * Try to give it an estimate that's good enough, maybe at some | |
948 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 949 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 950 | */ |
a9cc799e | 951 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 952 | |
8300475e | 953 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
954 | } |
955 | ||
d126d43f JL |
956 | /* |
957 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 958 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
959 | */ |
960 | enum { | |
961 | HOLE_OFF = 0, | |
962 | DATA_OFF, | |
963 | }; | |
964 | ||
965 | /* | |
966 | * Lookup the desired type of offset from the given page. | |
967 | * | |
968 | * On success, return true and the offset argument will point to the | |
969 | * start of the region that was found. Otherwise this function will | |
970 | * return false and keep the offset argument unchanged. | |
971 | */ | |
972 | STATIC bool | |
973 | xfs_lookup_buffer_offset( | |
974 | struct page *page, | |
975 | loff_t *offset, | |
976 | unsigned int type) | |
977 | { | |
978 | loff_t lastoff = page_offset(page); | |
979 | bool found = false; | |
980 | struct buffer_head *bh, *head; | |
981 | ||
982 | bh = head = page_buffers(page); | |
983 | do { | |
984 | /* | |
985 | * Unwritten extents that have data in the page | |
986 | * cache covering them can be identified by the | |
987 | * BH_Unwritten state flag. Pages with multiple | |
988 | * buffers might have a mix of holes, data and | |
989 | * unwritten extents - any buffer with valid | |
990 | * data in it should have BH_Uptodate flag set | |
991 | * on it. | |
992 | */ | |
993 | if (buffer_unwritten(bh) || | |
994 | buffer_uptodate(bh)) { | |
995 | if (type == DATA_OFF) | |
996 | found = true; | |
997 | } else { | |
998 | if (type == HOLE_OFF) | |
999 | found = true; | |
1000 | } | |
1001 | ||
1002 | if (found) { | |
1003 | *offset = lastoff; | |
1004 | break; | |
1005 | } | |
1006 | lastoff += bh->b_size; | |
1007 | } while ((bh = bh->b_this_page) != head); | |
1008 | ||
1009 | return found; | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * This routine is called to find out and return a data or hole offset | |
1014 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1015 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1016 | * |
1017 | * The argument offset is used to tell where we start to search from the | |
1018 | * page cache. Map is used to figure out the end points of the range to | |
1019 | * lookup pages. | |
1020 | * | |
1021 | * Return true if the desired type of offset was found, and the argument | |
1022 | * offset is filled with that address. Otherwise, return false and keep | |
1023 | * offset unchanged. | |
1024 | */ | |
1025 | STATIC bool | |
1026 | xfs_find_get_desired_pgoff( | |
1027 | struct inode *inode, | |
1028 | struct xfs_bmbt_irec *map, | |
1029 | unsigned int type, | |
1030 | loff_t *offset) | |
1031 | { | |
1032 | struct xfs_inode *ip = XFS_I(inode); | |
1033 | struct xfs_mount *mp = ip->i_mount; | |
1034 | struct pagevec pvec; | |
1035 | pgoff_t index; | |
1036 | pgoff_t end; | |
1037 | loff_t endoff; | |
1038 | loff_t startoff = *offset; | |
1039 | loff_t lastoff = startoff; | |
1040 | bool found = false; | |
1041 | ||
1042 | pagevec_init(&pvec, 0); | |
1043 | ||
09cbfeaf | 1044 | index = startoff >> PAGE_SHIFT; |
d126d43f | 1045 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
09cbfeaf | 1046 | end = endoff >> PAGE_SHIFT; |
d126d43f JL |
1047 | do { |
1048 | int want; | |
1049 | unsigned nr_pages; | |
1050 | unsigned int i; | |
1051 | ||
1052 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1053 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1054 | want); | |
1055 | /* | |
1056 | * No page mapped into given range. If we are searching holes | |
1057 | * and if this is the first time we got into the loop, it means | |
1058 | * that the given offset is landed in a hole, return it. | |
1059 | * | |
1060 | * If we have already stepped through some block buffers to find | |
1061 | * holes but they all contains data. In this case, the last | |
1062 | * offset is already updated and pointed to the end of the last | |
1063 | * mapped page, if it does not reach the endpoint to search, | |
1064 | * that means there should be a hole between them. | |
1065 | */ | |
1066 | if (nr_pages == 0) { | |
1067 | /* Data search found nothing */ | |
1068 | if (type == DATA_OFF) | |
1069 | break; | |
1070 | ||
1071 | ASSERT(type == HOLE_OFF); | |
1072 | if (lastoff == startoff || lastoff < endoff) { | |
1073 | found = true; | |
1074 | *offset = lastoff; | |
1075 | } | |
1076 | break; | |
1077 | } | |
1078 | ||
1079 | /* | |
1080 | * At lease we found one page. If this is the first time we | |
1081 | * step into the loop, and if the first page index offset is | |
1082 | * greater than the given search offset, a hole was found. | |
1083 | */ | |
1084 | if (type == HOLE_OFF && lastoff == startoff && | |
1085 | lastoff < page_offset(pvec.pages[0])) { | |
1086 | found = true; | |
1087 | break; | |
1088 | } | |
1089 | ||
1090 | for (i = 0; i < nr_pages; i++) { | |
1091 | struct page *page = pvec.pages[i]; | |
1092 | loff_t b_offset; | |
1093 | ||
1094 | /* | |
1095 | * At this point, the page may be truncated or | |
1096 | * invalidated (changing page->mapping to NULL), | |
1097 | * or even swizzled back from swapper_space to tmpfs | |
1098 | * file mapping. However, page->index will not change | |
1099 | * because we have a reference on the page. | |
1100 | * | |
1101 | * Searching done if the page index is out of range. | |
1102 | * If the current offset is not reaches the end of | |
1103 | * the specified search range, there should be a hole | |
1104 | * between them. | |
1105 | */ | |
1106 | if (page->index > end) { | |
1107 | if (type == HOLE_OFF && lastoff < endoff) { | |
1108 | *offset = lastoff; | |
1109 | found = true; | |
1110 | } | |
1111 | goto out; | |
1112 | } | |
1113 | ||
1114 | lock_page(page); | |
1115 | /* | |
1116 | * Page truncated or invalidated(page->mapping == NULL). | |
1117 | * We can freely skip it and proceed to check the next | |
1118 | * page. | |
1119 | */ | |
1120 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1121 | unlock_page(page); | |
1122 | continue; | |
1123 | } | |
1124 | ||
1125 | if (!page_has_buffers(page)) { | |
1126 | unlock_page(page); | |
1127 | continue; | |
1128 | } | |
1129 | ||
1130 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1131 | if (found) { | |
1132 | /* | |
1133 | * The found offset may be less than the start | |
1134 | * point to search if this is the first time to | |
1135 | * come here. | |
1136 | */ | |
1137 | *offset = max_t(loff_t, startoff, b_offset); | |
1138 | unlock_page(page); | |
1139 | goto out; | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * We either searching data but nothing was found, or | |
1144 | * searching hole but found a data buffer. In either | |
1145 | * case, probably the next page contains the desired | |
1146 | * things, update the last offset to it so. | |
1147 | */ | |
1148 | lastoff = page_offset(page) + PAGE_SIZE; | |
1149 | unlock_page(page); | |
1150 | } | |
1151 | ||
1152 | /* | |
1153 | * The number of returned pages less than our desired, search | |
1154 | * done. In this case, nothing was found for searching data, | |
1155 | * but we found a hole behind the last offset. | |
1156 | */ | |
1157 | if (nr_pages < want) { | |
1158 | if (type == HOLE_OFF) { | |
1159 | *offset = lastoff; | |
1160 | found = true; | |
1161 | } | |
1162 | break; | |
1163 | } | |
1164 | ||
1165 | index = pvec.pages[i - 1]->index + 1; | |
1166 | pagevec_release(&pvec); | |
1167 | } while (index <= end); | |
1168 | ||
1169 | out: | |
1170 | pagevec_release(&pvec); | |
1171 | return found; | |
1172 | } | |
1173 | ||
8aa7d37e ES |
1174 | /* |
1175 | * caller must lock inode with xfs_ilock_data_map_shared, | |
1176 | * can we craft an appropriate ASSERT? | |
1177 | * | |
1178 | * end is because the VFS-level lseek interface is defined such that any | |
1179 | * offset past i_size shall return -ENXIO, but we use this for quota code | |
1180 | * which does not maintain i_size, and we want to SEEK_DATA past i_size. | |
1181 | */ | |
1182 | loff_t | |
1183 | __xfs_seek_hole_data( | |
1184 | struct inode *inode, | |
49c69591 | 1185 | loff_t start, |
8aa7d37e | 1186 | loff_t end, |
49c69591 | 1187 | int whence) |
3fe3e6b1 | 1188 | { |
3fe3e6b1 JL |
1189 | struct xfs_inode *ip = XFS_I(inode); |
1190 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 | 1191 | loff_t uninitialized_var(offset); |
3fe3e6b1 | 1192 | xfs_fileoff_t fsbno; |
8aa7d37e | 1193 | xfs_filblks_t lastbno; |
3fe3e6b1 JL |
1194 | int error; |
1195 | ||
8aa7d37e | 1196 | if (start >= end) { |
2451337d | 1197 | error = -ENXIO; |
8aa7d37e | 1198 | goto out_error; |
3fe3e6b1 JL |
1199 | } |
1200 | ||
3fe3e6b1 JL |
1201 | /* |
1202 | * Try to read extents from the first block indicated | |
1203 | * by fsbno to the end block of the file. | |
1204 | */ | |
52f1acc8 | 1205 | fsbno = XFS_B_TO_FSBT(mp, start); |
8aa7d37e | 1206 | lastbno = XFS_B_TO_FSB(mp, end); |
49c69591 | 1207 | |
52f1acc8 JL |
1208 | for (;;) { |
1209 | struct xfs_bmbt_irec map[2]; | |
1210 | int nmap = 2; | |
1211 | unsigned int i; | |
3fe3e6b1 | 1212 | |
8aa7d37e | 1213 | error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap, |
52f1acc8 JL |
1214 | XFS_BMAPI_ENTIRE); |
1215 | if (error) | |
8aa7d37e | 1216 | goto out_error; |
3fe3e6b1 | 1217 | |
52f1acc8 JL |
1218 | /* No extents at given offset, must be beyond EOF */ |
1219 | if (nmap == 0) { | |
2451337d | 1220 | error = -ENXIO; |
8aa7d37e | 1221 | goto out_error; |
52f1acc8 JL |
1222 | } |
1223 | ||
1224 | for (i = 0; i < nmap; i++) { | |
1225 | offset = max_t(loff_t, start, | |
1226 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1227 | ||
49c69591 ES |
1228 | /* Landed in the hole we wanted? */ |
1229 | if (whence == SEEK_HOLE && | |
1230 | map[i].br_startblock == HOLESTARTBLOCK) | |
1231 | goto out; | |
1232 | ||
1233 | /* Landed in the data extent we wanted? */ | |
1234 | if (whence == SEEK_DATA && | |
1235 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1236 | (map[i].br_state == XFS_EXT_NORM && | |
1237 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1238 | goto out; |
1239 | ||
1240 | /* | |
49c69591 ES |
1241 | * Landed in an unwritten extent, try to search |
1242 | * for hole or data from page cache. | |
52f1acc8 JL |
1243 | */ |
1244 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1245 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1246 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1247 | &offset)) | |
52f1acc8 JL |
1248 | goto out; |
1249 | } | |
1250 | } | |
1251 | ||
1252 | /* | |
49c69591 ES |
1253 | * We only received one extent out of the two requested. This |
1254 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1255 | */ |
3fe3e6b1 | 1256 | if (nmap == 1) { |
49c69591 ES |
1257 | /* |
1258 | * If we were looking for a hole, set offset to | |
1259 | * the end of the file (i.e., there is an implicit | |
1260 | * hole at the end of any file). | |
1261 | */ | |
1262 | if (whence == SEEK_HOLE) { | |
8aa7d37e | 1263 | offset = end; |
49c69591 ES |
1264 | break; |
1265 | } | |
1266 | /* | |
1267 | * If we were looking for data, it's nowhere to be found | |
1268 | */ | |
1269 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1270 | error = -ENXIO; |
8aa7d37e | 1271 | goto out_error; |
3fe3e6b1 JL |
1272 | } |
1273 | ||
52f1acc8 JL |
1274 | ASSERT(i > 1); |
1275 | ||
1276 | /* | |
1277 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1278 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1279 | */ |
1280 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1281 | start = XFS_FSB_TO_B(mp, fsbno); | |
8aa7d37e | 1282 | if (start >= end) { |
49c69591 | 1283 | if (whence == SEEK_HOLE) { |
8aa7d37e | 1284 | offset = end; |
49c69591 ES |
1285 | break; |
1286 | } | |
1287 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1288 | error = -ENXIO; |
8aa7d37e | 1289 | goto out_error; |
52f1acc8 | 1290 | } |
3fe3e6b1 JL |
1291 | } |
1292 | ||
b686d1f7 JL |
1293 | out: |
1294 | /* | |
49c69591 | 1295 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1296 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1297 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1298 | * situation in particular. |
1299 | */ | |
49c69591 | 1300 | if (whence == SEEK_HOLE) |
8aa7d37e ES |
1301 | offset = min_t(loff_t, offset, end); |
1302 | ||
1303 | return offset; | |
1304 | ||
1305 | out_error: | |
1306 | return error; | |
1307 | } | |
1308 | ||
1309 | STATIC loff_t | |
1310 | xfs_seek_hole_data( | |
1311 | struct file *file, | |
1312 | loff_t start, | |
1313 | int whence) | |
1314 | { | |
1315 | struct inode *inode = file->f_mapping->host; | |
1316 | struct xfs_inode *ip = XFS_I(inode); | |
1317 | struct xfs_mount *mp = ip->i_mount; | |
1318 | uint lock; | |
1319 | loff_t offset, end; | |
1320 | int error = 0; | |
1321 | ||
1322 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1323 | return -EIO; | |
1324 | ||
1325 | lock = xfs_ilock_data_map_shared(ip); | |
1326 | ||
1327 | end = i_size_read(inode); | |
1328 | offset = __xfs_seek_hole_data(inode, start, end, whence); | |
1329 | if (offset < 0) { | |
1330 | error = offset; | |
1331 | goto out_unlock; | |
1332 | } | |
1333 | ||
46a1c2c7 | 1334 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1335 | |
1336 | out_unlock: | |
01f4f327 | 1337 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1338 | |
1339 | if (error) | |
2451337d | 1340 | return error; |
3fe3e6b1 JL |
1341 | return offset; |
1342 | } | |
1343 | ||
1344 | STATIC loff_t | |
1345 | xfs_file_llseek( | |
1346 | struct file *file, | |
1347 | loff_t offset, | |
59f9c004 | 1348 | int whence) |
3fe3e6b1 | 1349 | { |
59f9c004 | 1350 | switch (whence) { |
3fe3e6b1 JL |
1351 | case SEEK_END: |
1352 | case SEEK_CUR: | |
1353 | case SEEK_SET: | |
59f9c004 | 1354 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1355 | case SEEK_HOLE: |
49c69591 | 1356 | case SEEK_DATA: |
59f9c004 | 1357 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1358 | default: |
1359 | return -EINVAL; | |
1360 | } | |
1361 | } | |
1362 | ||
de0e8c20 DC |
1363 | /* |
1364 | * Locking for serialisation of IO during page faults. This results in a lock | |
1365 | * ordering of: | |
1366 | * | |
1367 | * mmap_sem (MM) | |
6b698ede | 1368 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1369 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1370 | * page_lock (MM) |
1371 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1372 | */ |
de0e8c20 | 1373 | |
075a924d DC |
1374 | /* |
1375 | * mmap()d file has taken write protection fault and is being made writable. We | |
1376 | * can set the page state up correctly for a writable page, which means we can | |
1377 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1378 | * mapping. | |
de0e8c20 DC |
1379 | */ |
1380 | STATIC int | |
075a924d | 1381 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1382 | struct vm_fault *vmf) |
1383 | { | |
11bac800 | 1384 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ec56b1f1 | 1385 | int ret; |
de0e8c20 | 1386 | |
6b698ede | 1387 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1388 | |
6b698ede | 1389 | sb_start_pagefault(inode->i_sb); |
11bac800 | 1390 | file_update_time(vmf->vma->vm_file); |
6b698ede | 1391 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1392 | |
6b698ede | 1393 | if (IS_DAX(inode)) { |
c791ace1 | 1394 | ret = dax_iomap_fault(vmf, PE_SIZE_PTE, &xfs_iomap_ops); |
6b698ede | 1395 | } else { |
11bac800 | 1396 | ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops); |
6b698ede DC |
1397 | ret = block_page_mkwrite_return(ret); |
1398 | } | |
1399 | ||
1400 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1401 | sb_end_pagefault(inode->i_sb); | |
1402 | ||
1403 | return ret; | |
de0e8c20 DC |
1404 | } |
1405 | ||
075a924d | 1406 | STATIC int |
6b698ede | 1407 | xfs_filemap_fault( |
075a924d DC |
1408 | struct vm_fault *vmf) |
1409 | { | |
11bac800 | 1410 | struct inode *inode = file_inode(vmf->vma->vm_file); |
6b698ede | 1411 | int ret; |
ec56b1f1 | 1412 | |
b2442c5a | 1413 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1414 | |
6b698ede | 1415 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1416 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
11bac800 | 1417 | return xfs_filemap_page_mkwrite(vmf); |
075a924d | 1418 | |
b2442c5a | 1419 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acdda3aa | 1420 | if (IS_DAX(inode)) |
c791ace1 | 1421 | ret = dax_iomap_fault(vmf, PE_SIZE_PTE, &xfs_iomap_ops); |
acdda3aa | 1422 | else |
11bac800 | 1423 | ret = filemap_fault(vmf); |
b2442c5a | 1424 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
075a924d | 1425 | |
6b698ede DC |
1426 | return ret; |
1427 | } | |
1428 | ||
13ad4fe3 DC |
1429 | /* |
1430 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on | |
1431 | * both read and write faults. Hence we need to handle both cases. There is no | |
a2d58167 | 1432 | * ->huge_mkwrite callout for huge pages, so we have a single function here to |
13ad4fe3 DC |
1433 | * handle both cases here. @flags carries the information on the type of fault |
1434 | * occuring. | |
1435 | */ | |
acd76e74 | 1436 | STATIC int |
a2d58167 | 1437 | xfs_filemap_huge_fault( |
c791ace1 DJ |
1438 | struct vm_fault *vmf, |
1439 | enum page_entry_size pe_size) | |
acd76e74 | 1440 | { |
f4200391 | 1441 | struct inode *inode = file_inode(vmf->vma->vm_file); |
acd76e74 MW |
1442 | struct xfs_inode *ip = XFS_I(inode); |
1443 | int ret; | |
1444 | ||
1445 | if (!IS_DAX(inode)) | |
1446 | return VM_FAULT_FALLBACK; | |
1447 | ||
a2d58167 | 1448 | trace_xfs_filemap_huge_fault(ip); |
acd76e74 | 1449 | |
d8a849e1 | 1450 | if (vmf->flags & FAULT_FLAG_WRITE) { |
13ad4fe3 | 1451 | sb_start_pagefault(inode->i_sb); |
f4200391 | 1452 | file_update_time(vmf->vma->vm_file); |
13ad4fe3 DC |
1453 | } |
1454 | ||
acd76e74 | 1455 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
c791ace1 | 1456 | ret = dax_iomap_fault(vmf, pe_size, &xfs_iomap_ops); |
acd76e74 | 1457 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acd76e74 | 1458 | |
d8a849e1 | 1459 | if (vmf->flags & FAULT_FLAG_WRITE) |
13ad4fe3 | 1460 | sb_end_pagefault(inode->i_sb); |
acd76e74 MW |
1461 | |
1462 | return ret; | |
1463 | } | |
1464 | ||
3af49285 DC |
1465 | /* |
1466 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1467 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1468 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1469 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1470 | */ |
1471 | static int | |
1472 | xfs_filemap_pfn_mkwrite( | |
3af49285 DC |
1473 | struct vm_fault *vmf) |
1474 | { | |
1475 | ||
11bac800 | 1476 | struct inode *inode = file_inode(vmf->vma->vm_file); |
3af49285 DC |
1477 | struct xfs_inode *ip = XFS_I(inode); |
1478 | int ret = VM_FAULT_NOPAGE; | |
1479 | loff_t size; | |
1480 | ||
1481 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1482 | ||
1483 | sb_start_pagefault(inode->i_sb); | |
11bac800 | 1484 | file_update_time(vmf->vma->vm_file); |
3af49285 DC |
1485 | |
1486 | /* check if the faulting page hasn't raced with truncate */ | |
1487 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1488 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1489 | if (vmf->pgoff >= size) | |
1490 | ret = VM_FAULT_SIGBUS; | |
5eb88dca | 1491 | else if (IS_DAX(inode)) |
11bac800 | 1492 | ret = dax_pfn_mkwrite(vmf); |
3af49285 DC |
1493 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1494 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1495 | return ret; |
3af49285 | 1496 | |
acd76e74 MW |
1497 | } |
1498 | ||
6b698ede DC |
1499 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1500 | .fault = xfs_filemap_fault, | |
a2d58167 | 1501 | .huge_fault = xfs_filemap_huge_fault, |
6b698ede DC |
1502 | .map_pages = filemap_map_pages, |
1503 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1504 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1505 | }; |
1506 | ||
1507 | STATIC int | |
1508 | xfs_file_mmap( | |
1509 | struct file *filp, | |
1510 | struct vm_area_struct *vma) | |
1511 | { | |
1512 | file_accessed(filp); | |
1513 | vma->vm_ops = &xfs_file_vm_ops; | |
1514 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1515 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1516 | return 0; |
075a924d DC |
1517 | } |
1518 | ||
4b6f5d20 | 1519 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1520 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1521 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1522 | .write_iter = xfs_file_write_iter, |
82c156f8 | 1523 | .splice_read = generic_file_splice_read, |
8d020765 | 1524 | .splice_write = iter_file_splice_write, |
3562fd45 | 1525 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1526 | #ifdef CONFIG_COMPAT |
3562fd45 | 1527 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1528 | #endif |
3562fd45 NS |
1529 | .mmap = xfs_file_mmap, |
1530 | .open = xfs_file_open, | |
1531 | .release = xfs_file_release, | |
1532 | .fsync = xfs_file_fsync, | |
dbe6ec81 | 1533 | .get_unmapped_area = thp_get_unmapped_area, |
2fe17c10 | 1534 | .fallocate = xfs_file_fallocate, |
9fe26045 | 1535 | .clone_file_range = xfs_file_clone_range, |
cc714660 | 1536 | .dedupe_file_range = xfs_file_dedupe_range, |
1da177e4 LT |
1537 | }; |
1538 | ||
4b6f5d20 | 1539 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1540 | .open = xfs_dir_open, |
1da177e4 | 1541 | .read = generic_read_dir, |
3b0a3c1a | 1542 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1543 | .llseek = generic_file_llseek, |
3562fd45 | 1544 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1545 | #ifdef CONFIG_COMPAT |
3562fd45 | 1546 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1547 | #endif |
1da2f2db | 1548 | .fsync = xfs_dir_fsync, |
1da177e4 | 1549 | }; |