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" |
1da177e4 LT |
40 | |
41 | #include <linux/dcache.h> | |
2fe17c10 | 42 | #include <linux/falloc.h> |
d126d43f | 43 | #include <linux/pagevec.h> |
66114cad | 44 | #include <linux/backing-dev.h> |
1da177e4 | 45 | |
f0f37e2f | 46 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 47 | |
487f84f3 DC |
48 | /* |
49 | * Locking primitives for read and write IO paths to ensure we consistently use | |
50 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
51 | */ | |
52 | static inline void | |
53 | xfs_rw_ilock( | |
54 | struct xfs_inode *ip, | |
55 | int type) | |
56 | { | |
57 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 58 | inode_lock(VFS_I(ip)); |
487f84f3 DC |
59 | xfs_ilock(ip, type); |
60 | } | |
61 | ||
62 | static inline void | |
63 | xfs_rw_iunlock( | |
64 | struct xfs_inode *ip, | |
65 | int type) | |
66 | { | |
67 | xfs_iunlock(ip, type); | |
68 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 69 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
70 | } |
71 | ||
72 | static inline void | |
73 | xfs_rw_ilock_demote( | |
74 | struct xfs_inode *ip, | |
75 | int type) | |
76 | { | |
77 | xfs_ilock_demote(ip, type); | |
78 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 79 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
80 | } |
81 | ||
dda35b8f | 82 | /* |
4f69f578 DC |
83 | * xfs_iozero clears the specified range supplied via the page cache (except in |
84 | * the DAX case). Writes through the page cache will allocate blocks over holes, | |
85 | * though the callers usually map the holes first and avoid them. If a block is | |
86 | * not completely zeroed, then it will be read from disk before being partially | |
87 | * zeroed. | |
dda35b8f | 88 | * |
4f69f578 DC |
89 | * In the DAX case, we can just directly write to the underlying pages. This |
90 | * will not allocate blocks, but will avoid holes and unwritten extents and so | |
91 | * not do unnecessary work. | |
dda35b8f | 92 | */ |
ef9d8733 | 93 | int |
dda35b8f CH |
94 | xfs_iozero( |
95 | struct xfs_inode *ip, /* inode */ | |
96 | loff_t pos, /* offset in file */ | |
97 | size_t count) /* size of data to zero */ | |
98 | { | |
99 | struct page *page; | |
100 | struct address_space *mapping; | |
4f69f578 DC |
101 | int status = 0; |
102 | ||
dda35b8f CH |
103 | |
104 | mapping = VFS_I(ip)->i_mapping; | |
105 | do { | |
106 | unsigned offset, bytes; | |
107 | void *fsdata; | |
108 | ||
09cbfeaf KS |
109 | offset = (pos & (PAGE_SIZE -1)); /* Within page */ |
110 | bytes = PAGE_SIZE - offset; | |
dda35b8f CH |
111 | if (bytes > count) |
112 | bytes = count; | |
113 | ||
4f69f578 DC |
114 | if (IS_DAX(VFS_I(ip))) { |
115 | status = dax_zero_page_range(VFS_I(ip), pos, bytes, | |
116 | xfs_get_blocks_direct); | |
117 | if (status) | |
118 | break; | |
119 | } else { | |
120 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
121 | AOP_FLAG_UNINTERRUPTIBLE, | |
122 | &page, &fsdata); | |
123 | if (status) | |
124 | break; | |
dda35b8f | 125 | |
4f69f578 | 126 | zero_user(page, offset, bytes); |
dda35b8f | 127 | |
4f69f578 DC |
128 | status = pagecache_write_end(NULL, mapping, pos, bytes, |
129 | bytes, page, fsdata); | |
130 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
131 | status = 0; | |
132 | } | |
dda35b8f CH |
133 | pos += bytes; |
134 | count -= bytes; | |
dda35b8f CH |
135 | } while (count); |
136 | ||
cddc1162 | 137 | return status; |
dda35b8f CH |
138 | } |
139 | ||
8add71ca CH |
140 | int |
141 | xfs_update_prealloc_flags( | |
142 | struct xfs_inode *ip, | |
143 | enum xfs_prealloc_flags flags) | |
144 | { | |
145 | struct xfs_trans *tp; | |
146 | int error; | |
147 | ||
253f4911 CH |
148 | error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid, |
149 | 0, 0, 0, &tp); | |
150 | if (error) | |
8add71ca | 151 | return error; |
8add71ca CH |
152 | |
153 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
154 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
155 | ||
156 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
157 | VFS_I(ip)->i_mode &= ~S_ISUID; |
158 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
159 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
160 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
161 | } | |
162 | ||
163 | if (flags & XFS_PREALLOC_SET) | |
164 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
165 | if (flags & XFS_PREALLOC_CLEAR) | |
166 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
167 | ||
168 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
169 | if (flags & XFS_PREALLOC_SYNC) | |
170 | xfs_trans_set_sync(tp); | |
70393313 | 171 | return xfs_trans_commit(tp); |
8add71ca CH |
172 | } |
173 | ||
1da2f2db CH |
174 | /* |
175 | * Fsync operations on directories are much simpler than on regular files, | |
176 | * as there is no file data to flush, and thus also no need for explicit | |
177 | * cache flush operations, and there are no non-transaction metadata updates | |
178 | * on directories either. | |
179 | */ | |
180 | STATIC int | |
181 | xfs_dir_fsync( | |
182 | struct file *file, | |
183 | loff_t start, | |
184 | loff_t end, | |
185 | int datasync) | |
186 | { | |
187 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
188 | struct xfs_mount *mp = ip->i_mount; | |
189 | xfs_lsn_t lsn = 0; | |
190 | ||
191 | trace_xfs_dir_fsync(ip); | |
192 | ||
193 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
194 | if (xfs_ipincount(ip)) | |
195 | lsn = ip->i_itemp->ili_last_lsn; | |
196 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
197 | ||
198 | if (!lsn) | |
199 | return 0; | |
2451337d | 200 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
201 | } |
202 | ||
fd3200be CH |
203 | STATIC int |
204 | xfs_file_fsync( | |
205 | struct file *file, | |
02c24a82 JB |
206 | loff_t start, |
207 | loff_t end, | |
fd3200be CH |
208 | int datasync) |
209 | { | |
7ea80859 CH |
210 | struct inode *inode = file->f_mapping->host; |
211 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 212 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
213 | int error = 0; |
214 | int log_flushed = 0; | |
b1037058 | 215 | xfs_lsn_t lsn = 0; |
fd3200be | 216 | |
cca28fb8 | 217 | trace_xfs_file_fsync(ip); |
fd3200be | 218 | |
02c24a82 JB |
219 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
220 | if (error) | |
221 | return error; | |
222 | ||
a27a263b | 223 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 224 | return -EIO; |
fd3200be CH |
225 | |
226 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
227 | ||
a27a263b CH |
228 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
229 | /* | |
230 | * If we have an RT and/or log subvolume we need to make sure | |
231 | * to flush the write cache the device used for file data | |
232 | * first. This is to ensure newly written file data make | |
233 | * it to disk before logging the new inode size in case of | |
234 | * an extending write. | |
235 | */ | |
236 | if (XFS_IS_REALTIME_INODE(ip)) | |
237 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
238 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
239 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
240 | } | |
241 | ||
fd3200be | 242 | /* |
fc0561ce DC |
243 | * All metadata updates are logged, which means that we just have to |
244 | * flush the log up to the latest LSN that touched the inode. If we have | |
245 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
246 | * log force before we clear the ili_fsync_fields field. This ensures | |
247 | * that we don't get a racing sync operation that does not wait for the | |
248 | * metadata to hit the journal before returning. If we race with | |
249 | * clearing the ili_fsync_fields, then all that will happen is the log | |
250 | * force will do nothing as the lsn will already be on disk. We can't | |
251 | * race with setting ili_fsync_fields because that is done under | |
252 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
253 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
254 | */ |
255 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
256 | if (xfs_ipincount(ip)) { |
257 | if (!datasync || | |
fc0561ce | 258 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
259 | lsn = ip->i_itemp->ili_last_lsn; |
260 | } | |
fd3200be | 261 | |
fc0561ce | 262 | if (lsn) { |
b1037058 | 263 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
264 | ip->i_itemp->ili_fsync_fields = 0; |
265 | } | |
266 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 267 | |
a27a263b CH |
268 | /* |
269 | * If we only have a single device, and the log force about was | |
270 | * a no-op we might have to flush the data device cache here. | |
271 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
272 | * an already allocated file and thus do not have any metadata to | |
273 | * commit. | |
274 | */ | |
275 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
276 | mp->m_logdev_targp == mp->m_ddev_targp && | |
277 | !XFS_IS_REALTIME_INODE(ip) && | |
278 | !log_flushed) | |
279 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 280 | |
2451337d | 281 | return error; |
fd3200be CH |
282 | } |
283 | ||
00258e36 | 284 | STATIC ssize_t |
bbc5a740 | 285 | xfs_file_dio_aio_read( |
dda35b8f | 286 | struct kiocb *iocb, |
b4f5d2c6 | 287 | struct iov_iter *to) |
dda35b8f | 288 | { |
bbc5a740 CH |
289 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
290 | struct inode *inode = mapping->host; | |
00258e36 | 291 | struct xfs_inode *ip = XFS_I(inode); |
f1285ff0 | 292 | loff_t isize = i_size_read(inode); |
bbc5a740 | 293 | size_t count = iov_iter_count(to); |
f1285ff0 | 294 | struct iov_iter data; |
bbc5a740 | 295 | struct xfs_buftarg *target; |
dda35b8f | 296 | ssize_t ret = 0; |
dda35b8f | 297 | |
bbc5a740 | 298 | trace_xfs_file_direct_read(ip, count, iocb->ki_pos); |
dda35b8f | 299 | |
f1285ff0 CH |
300 | if (!count) |
301 | return 0; /* skip atime */ | |
302 | ||
bbc5a740 CH |
303 | if (XFS_IS_REALTIME_INODE(ip)) |
304 | target = ip->i_mount->m_rtdev_targp; | |
305 | else | |
306 | target = ip->i_mount->m_ddev_targp; | |
307 | ||
16d4d435 CH |
308 | /* DIO must be aligned to device logical sector size */ |
309 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) { | |
310 | if (iocb->ki_pos == isize) | |
311 | return 0; | |
312 | return -EINVAL; | |
dda35b8f CH |
313 | } |
314 | ||
0c38a251 | 315 | /* |
3d751af2 BF |
316 | * Locking is a bit tricky here. If we take an exclusive lock for direct |
317 | * IO, we effectively serialise all new concurrent read IO to this file | |
318 | * and block it behind IO that is currently in progress because IO in | |
319 | * progress holds the IO lock shared. We only need to hold the lock | |
320 | * exclusive to blow away the page cache, so only take lock exclusively | |
321 | * if the page cache needs invalidation. This allows the normal direct | |
322 | * IO case of no page cache pages to proceeed concurrently without | |
323 | * serialisation. | |
0c38a251 DC |
324 | */ |
325 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
bbc5a740 | 326 | if (mapping->nrpages) { |
0c38a251 | 327 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
328 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
329 | ||
3d751af2 BF |
330 | /* |
331 | * The generic dio code only flushes the range of the particular | |
332 | * I/O. Because we take an exclusive lock here, this whole | |
333 | * sequence is considerably more expensive for us. This has a | |
334 | * noticeable performance impact for any file with cached pages, | |
335 | * even when outside of the range of the particular I/O. | |
336 | * | |
337 | * Hence, amortize the cost of the lock against a full file | |
338 | * flush and reduce the chances of repeated iolock cycles going | |
339 | * forward. | |
340 | */ | |
bbc5a740 CH |
341 | if (mapping->nrpages) { |
342 | ret = filemap_write_and_wait(mapping); | |
487f84f3 DC |
343 | if (ret) { |
344 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
345 | return ret; | |
346 | } | |
85e584da CM |
347 | |
348 | /* | |
349 | * Invalidate whole pages. This can return an error if | |
350 | * we fail to invalidate a page, but this should never | |
351 | * happen on XFS. Warn if it does fail. | |
352 | */ | |
bbc5a740 | 353 | ret = invalidate_inode_pages2(mapping); |
85e584da CM |
354 | WARN_ON_ONCE(ret); |
355 | ret = 0; | |
00258e36 | 356 | } |
487f84f3 | 357 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 358 | } |
dda35b8f | 359 | |
f1285ff0 | 360 | data = *to; |
16d4d435 CH |
361 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
362 | xfs_get_blocks_direct, NULL, NULL, 0); | |
363 | if (ret > 0) { | |
364 | iocb->ki_pos += ret; | |
365 | iov_iter_advance(to, ret); | |
fa8d972d | 366 | } |
16d4d435 CH |
367 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
368 | ||
369 | file_accessed(iocb->ki_filp); | |
370 | return ret; | |
371 | } | |
372 | ||
373 | STATIC ssize_t | |
374 | xfs_file_dax_read( | |
375 | struct kiocb *iocb, | |
376 | struct iov_iter *to) | |
377 | { | |
378 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
379 | struct inode *inode = mapping->host; | |
380 | struct xfs_inode *ip = XFS_I(inode); | |
381 | struct iov_iter data = *to; | |
382 | size_t count = iov_iter_count(to); | |
383 | ssize_t ret = 0; | |
384 | ||
385 | trace_xfs_file_dax_read(ip, count, iocb->ki_pos); | |
386 | ||
387 | if (!count) | |
388 | return 0; /* skip atime */ | |
389 | ||
390 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
391 | ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, NULL, 0); | |
f1285ff0 CH |
392 | if (ret > 0) { |
393 | iocb->ki_pos += ret; | |
394 | iov_iter_advance(to, ret); | |
395 | } | |
bbc5a740 CH |
396 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
397 | ||
f1285ff0 | 398 | file_accessed(iocb->ki_filp); |
bbc5a740 CH |
399 | return ret; |
400 | } | |
401 | ||
402 | STATIC ssize_t | |
403 | xfs_file_buffered_aio_read( | |
404 | struct kiocb *iocb, | |
405 | struct iov_iter *to) | |
406 | { | |
407 | struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp)); | |
408 | ssize_t ret; | |
409 | ||
410 | trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos); | |
411 | ||
412 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
413 | ret = generic_file_read_iter(iocb, to); | |
414 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); | |
415 | ||
416 | return ret; | |
417 | } | |
418 | ||
419 | STATIC ssize_t | |
420 | xfs_file_read_iter( | |
421 | struct kiocb *iocb, | |
422 | struct iov_iter *to) | |
423 | { | |
16d4d435 CH |
424 | struct inode *inode = file_inode(iocb->ki_filp); |
425 | struct xfs_mount *mp = XFS_I(inode)->i_mount; | |
bbc5a740 CH |
426 | ssize_t ret = 0; |
427 | ||
428 | XFS_STATS_INC(mp, xs_read_calls); | |
429 | ||
430 | if (XFS_FORCED_SHUTDOWN(mp)) | |
431 | return -EIO; | |
432 | ||
16d4d435 CH |
433 | if (IS_DAX(inode)) |
434 | ret = xfs_file_dax_read(iocb, to); | |
435 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bbc5a740 | 436 | ret = xfs_file_dio_aio_read(iocb, to); |
3176c3e0 | 437 | else |
bbc5a740 | 438 | ret = xfs_file_buffered_aio_read(iocb, to); |
dda35b8f | 439 | |
dda35b8f | 440 | if (ret > 0) |
ff6d6af2 | 441 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f CH |
442 | return ret; |
443 | } | |
444 | ||
00258e36 CH |
445 | STATIC ssize_t |
446 | xfs_file_splice_read( | |
dda35b8f CH |
447 | struct file *infilp, |
448 | loff_t *ppos, | |
449 | struct pipe_inode_info *pipe, | |
450 | size_t count, | |
00258e36 | 451 | unsigned int flags) |
dda35b8f | 452 | { |
00258e36 | 453 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
dda35b8f CH |
454 | ssize_t ret; |
455 | ||
ff6d6af2 | 456 | XFS_STATS_INC(ip->i_mount, xs_read_calls); |
00258e36 | 457 | |
dda35b8f CH |
458 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
459 | return -EIO; | |
460 | ||
3176c3e0 | 461 | trace_xfs_file_splice_read(ip, count, *ppos); |
dda35b8f | 462 | |
a6d7636e DC |
463 | /* |
464 | * DAX inodes cannot ues the page cache for splice, so we have to push | |
465 | * them through the VFS IO path. This means it goes through | |
466 | * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we | |
467 | * cannot lock the splice operation at this level for DAX inodes. | |
468 | */ | |
469 | if (IS_DAX(VFS_I(ip))) { | |
470 | ret = default_file_splice_read(infilp, ppos, pipe, count, | |
471 | flags); | |
472 | goto out; | |
473 | } | |
dda35b8f | 474 | |
a6d7636e DC |
475 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
476 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
487f84f3 | 477 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
a6d7636e DC |
478 | out: |
479 | if (ret > 0) | |
480 | XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret); | |
dda35b8f CH |
481 | return ret; |
482 | } | |
483 | ||
dda35b8f | 484 | /* |
193aec10 CH |
485 | * This routine is called to handle zeroing any space in the last block of the |
486 | * file that is beyond the EOF. We do this since the size is being increased | |
487 | * without writing anything to that block and we don't want to read the | |
488 | * garbage on the disk. | |
dda35b8f CH |
489 | */ |
490 | STATIC int /* error (positive) */ | |
491 | xfs_zero_last_block( | |
193aec10 CH |
492 | struct xfs_inode *ip, |
493 | xfs_fsize_t offset, | |
5885ebda DC |
494 | xfs_fsize_t isize, |
495 | bool *did_zeroing) | |
dda35b8f | 496 | { |
193aec10 CH |
497 | struct xfs_mount *mp = ip->i_mount; |
498 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
499 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
500 | int zero_len; | |
501 | int nimaps = 1; | |
502 | int error = 0; | |
503 | struct xfs_bmbt_irec imap; | |
dda35b8f | 504 | |
193aec10 | 505 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 506 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 507 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 508 | if (error) |
dda35b8f | 509 | return error; |
193aec10 | 510 | |
dda35b8f | 511 | ASSERT(nimaps > 0); |
193aec10 | 512 | |
dda35b8f CH |
513 | /* |
514 | * If the block underlying isize is just a hole, then there | |
515 | * is nothing to zero. | |
516 | */ | |
193aec10 | 517 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 518 | return 0; |
dda35b8f CH |
519 | |
520 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
521 | if (isize + zero_len > offset) | |
522 | zero_len = offset - isize; | |
5885ebda | 523 | *did_zeroing = true; |
193aec10 | 524 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
525 | } |
526 | ||
527 | /* | |
193aec10 CH |
528 | * Zero any on disk space between the current EOF and the new, larger EOF. |
529 | * | |
530 | * This handles the normal case of zeroing the remainder of the last block in | |
531 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
532 | * file. This second case only happens with fixed size extents and when the | |
533 | * system crashes before the inode size was updated but after blocks were | |
534 | * allocated. | |
535 | * | |
536 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 537 | */ |
dda35b8f CH |
538 | int /* error (positive) */ |
539 | xfs_zero_eof( | |
193aec10 CH |
540 | struct xfs_inode *ip, |
541 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
542 | xfs_fsize_t isize, /* current inode size */ |
543 | bool *did_zeroing) | |
dda35b8f | 544 | { |
193aec10 CH |
545 | struct xfs_mount *mp = ip->i_mount; |
546 | xfs_fileoff_t start_zero_fsb; | |
547 | xfs_fileoff_t end_zero_fsb; | |
548 | xfs_fileoff_t zero_count_fsb; | |
549 | xfs_fileoff_t last_fsb; | |
550 | xfs_fileoff_t zero_off; | |
551 | xfs_fsize_t zero_len; | |
552 | int nimaps; | |
553 | int error = 0; | |
554 | struct xfs_bmbt_irec imap; | |
555 | ||
556 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
557 | ASSERT(offset > isize); |
558 | ||
0a50f162 BF |
559 | trace_xfs_zero_eof(ip, isize, offset - isize); |
560 | ||
dda35b8f CH |
561 | /* |
562 | * First handle zeroing the block on which isize resides. | |
193aec10 | 563 | * |
dda35b8f CH |
564 | * We only zero a part of that block so it is handled specially. |
565 | */ | |
193aec10 | 566 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
5885ebda | 567 | error = xfs_zero_last_block(ip, offset, isize, did_zeroing); |
193aec10 CH |
568 | if (error) |
569 | return error; | |
dda35b8f CH |
570 | } |
571 | ||
572 | /* | |
193aec10 CH |
573 | * Calculate the range between the new size and the old where blocks |
574 | * needing to be zeroed may exist. | |
575 | * | |
576 | * To get the block where the last byte in the file currently resides, | |
577 | * we need to subtract one from the size and truncate back to a block | |
578 | * boundary. We subtract 1 in case the size is exactly on a block | |
579 | * boundary. | |
dda35b8f CH |
580 | */ |
581 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
582 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
583 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
584 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
585 | if (last_fsb == end_zero_fsb) { | |
586 | /* | |
587 | * The size was only incremented on its last block. | |
588 | * We took care of that above, so just return. | |
589 | */ | |
590 | return 0; | |
591 | } | |
592 | ||
593 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
594 | while (start_zero_fsb <= end_zero_fsb) { | |
595 | nimaps = 1; | |
596 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
597 | |
598 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
599 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
600 | &imap, &nimaps, 0); | |
193aec10 CH |
601 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
602 | if (error) | |
dda35b8f | 603 | return error; |
193aec10 | 604 | |
dda35b8f CH |
605 | ASSERT(nimaps > 0); |
606 | ||
607 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
608 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
609 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
610 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
611 | continue; | |
612 | } | |
613 | ||
614 | /* | |
615 | * There are blocks we need to zero. | |
dda35b8f | 616 | */ |
dda35b8f CH |
617 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
618 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
619 | ||
620 | if ((zero_off + zero_len) > offset) | |
621 | zero_len = offset - zero_off; | |
622 | ||
623 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
624 | if (error) |
625 | return error; | |
dda35b8f | 626 | |
5885ebda | 627 | *did_zeroing = true; |
dda35b8f CH |
628 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
629 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
630 | } |
631 | ||
632 | return 0; | |
dda35b8f CH |
633 | } |
634 | ||
4d8d1581 DC |
635 | /* |
636 | * Common pre-write limit and setup checks. | |
637 | * | |
5bf1f262 CH |
638 | * Called with the iolocked held either shared and exclusive according to |
639 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
640 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
641 | */ |
642 | STATIC ssize_t | |
643 | xfs_file_aio_write_checks( | |
99733fa3 AV |
644 | struct kiocb *iocb, |
645 | struct iov_iter *from, | |
4d8d1581 DC |
646 | int *iolock) |
647 | { | |
99733fa3 | 648 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
649 | struct inode *inode = file->f_mapping->host; |
650 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 651 | ssize_t error = 0; |
99733fa3 | 652 | size_t count = iov_iter_count(from); |
3136e8bb | 653 | bool drained_dio = false; |
4d8d1581 | 654 | |
7271d243 | 655 | restart: |
3309dd04 AV |
656 | error = generic_write_checks(iocb, from); |
657 | if (error <= 0) | |
4d8d1581 | 658 | return error; |
4d8d1581 | 659 | |
21c3ea18 | 660 | error = xfs_break_layouts(inode, iolock, true); |
781355c6 CH |
661 | if (error) |
662 | return error; | |
663 | ||
a6de82ca JK |
664 | /* For changing security info in file_remove_privs() we need i_mutex */ |
665 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { | |
666 | xfs_rw_iunlock(ip, *iolock); | |
667 | *iolock = XFS_IOLOCK_EXCL; | |
668 | xfs_rw_ilock(ip, *iolock); | |
669 | goto restart; | |
670 | } | |
4d8d1581 DC |
671 | /* |
672 | * If the offset is beyond the size of the file, we need to zero any | |
673 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 674 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
675 | * iolock shared, we need to update it to exclusive which implies |
676 | * having to redo all checks before. | |
b9d59846 DC |
677 | * |
678 | * We need to serialise against EOF updates that occur in IO | |
679 | * completions here. We want to make sure that nobody is changing the | |
680 | * size while we do this check until we have placed an IO barrier (i.e. | |
681 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
682 | * The spinlock effectively forms a memory barrier once we have the | |
683 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
684 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 685 | */ |
b9d59846 | 686 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 687 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
688 | bool zero = false; |
689 | ||
b9d59846 | 690 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
691 | if (!drained_dio) { |
692 | if (*iolock == XFS_IOLOCK_SHARED) { | |
693 | xfs_rw_iunlock(ip, *iolock); | |
694 | *iolock = XFS_IOLOCK_EXCL; | |
695 | xfs_rw_ilock(ip, *iolock); | |
696 | iov_iter_reexpand(from, count); | |
697 | } | |
40c63fbc DC |
698 | /* |
699 | * We now have an IO submission barrier in place, but | |
700 | * AIO can do EOF updates during IO completion and hence | |
701 | * we now need to wait for all of them to drain. Non-AIO | |
702 | * DIO will have drained before we are given the | |
703 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
704 | * no-op. | |
705 | */ | |
706 | inode_dio_wait(inode); | |
3136e8bb | 707 | drained_dio = true; |
7271d243 DC |
708 | goto restart; |
709 | } | |
99733fa3 | 710 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
711 | if (error) |
712 | return error; | |
b9d59846 DC |
713 | } else |
714 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 715 | |
8a9c9980 CH |
716 | /* |
717 | * Updating the timestamps will grab the ilock again from | |
718 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
719 | * lock above. Eventually we should look into a way to avoid | |
720 | * the pointless lock roundtrip. | |
721 | */ | |
c3b2da31 JB |
722 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
723 | error = file_update_time(file); | |
724 | if (error) | |
725 | return error; | |
726 | } | |
8a9c9980 | 727 | |
4d8d1581 DC |
728 | /* |
729 | * If we're writing the file then make sure to clear the setuid and | |
730 | * setgid bits if the process is not being run by root. This keeps | |
731 | * people from modifying setuid and setgid binaries. | |
732 | */ | |
a6de82ca JK |
733 | if (!IS_NOSEC(inode)) |
734 | return file_remove_privs(file); | |
735 | return 0; | |
4d8d1581 DC |
736 | } |
737 | ||
f0d26e86 DC |
738 | /* |
739 | * xfs_file_dio_aio_write - handle direct IO writes | |
740 | * | |
741 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 742 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
743 | * follow locking changes and looping. |
744 | * | |
eda77982 DC |
745 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
746 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
747 | * pages are flushed out. | |
748 | * | |
749 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
750 | * allowing them to be done in parallel with reads and other direct IO writes. | |
751 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
752 | * needs to do sub-block zeroing and that requires serialisation against other | |
753 | * direct IOs to the same block. In this case we need to serialise the | |
754 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
755 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
756 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
757 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 758 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 759 | * |
f0d26e86 DC |
760 | * Returns with locks held indicated by @iolock and errors indicated by |
761 | * negative return values. | |
762 | */ | |
763 | STATIC ssize_t | |
764 | xfs_file_dio_aio_write( | |
765 | struct kiocb *iocb, | |
b3188919 | 766 | struct iov_iter *from) |
f0d26e86 DC |
767 | { |
768 | struct file *file = iocb->ki_filp; | |
769 | struct address_space *mapping = file->f_mapping; | |
770 | struct inode *inode = mapping->host; | |
771 | struct xfs_inode *ip = XFS_I(inode); | |
772 | struct xfs_mount *mp = ip->i_mount; | |
773 | ssize_t ret = 0; | |
eda77982 | 774 | int unaligned_io = 0; |
d0606464 | 775 | int iolock; |
b3188919 | 776 | size_t count = iov_iter_count(from); |
0cefb29e DC |
777 | loff_t end; |
778 | struct iov_iter data; | |
f0d26e86 DC |
779 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
780 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
781 | ||
7c71ee78 | 782 | /* DIO must be aligned to device logical sector size */ |
16d4d435 | 783 | if ((iocb->ki_pos | count) & target->bt_logical_sectormask) |
b474c7ae | 784 | return -EINVAL; |
f0d26e86 | 785 | |
7c71ee78 | 786 | /* "unaligned" here means not aligned to a filesystem block */ |
13712713 CH |
787 | if ((iocb->ki_pos & mp->m_blockmask) || |
788 | ((iocb->ki_pos + count) & mp->m_blockmask)) | |
eda77982 DC |
789 | unaligned_io = 1; |
790 | ||
7271d243 DC |
791 | /* |
792 | * We don't need to take an exclusive lock unless there page cache needs | |
793 | * to be invalidated or unaligned IO is being executed. We don't need to | |
794 | * consider the EOF extension case here because | |
795 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
796 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
797 | */ | |
798 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 799 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 800 | else |
d0606464 CH |
801 | iolock = XFS_IOLOCK_SHARED; |
802 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
803 | |
804 | /* | |
805 | * Recheck if there are cached pages that need invalidate after we got | |
806 | * the iolock to protect against other threads adding new pages while | |
807 | * we were waiting for the iolock. | |
808 | */ | |
d0606464 CH |
809 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
810 | xfs_rw_iunlock(ip, iolock); | |
811 | iolock = XFS_IOLOCK_EXCL; | |
812 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 813 | } |
f0d26e86 | 814 | |
99733fa3 | 815 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 816 | if (ret) |
d0606464 | 817 | goto out; |
99733fa3 | 818 | count = iov_iter_count(from); |
13712713 | 819 | end = iocb->ki_pos + count - 1; |
f0d26e86 | 820 | |
3d751af2 | 821 | /* |
bbc5a740 | 822 | * See xfs_file_dio_aio_read() for why we do a full-file flush here. |
3d751af2 | 823 | */ |
f0d26e86 | 824 | if (mapping->nrpages) { |
3d751af2 | 825 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
f0d26e86 | 826 | if (ret) |
d0606464 | 827 | goto out; |
834ffca6 | 828 | /* |
3d751af2 BF |
829 | * Invalidate whole pages. This can return an error if we fail |
830 | * to invalidate a page, but this should never happen on XFS. | |
831 | * Warn if it does fail. | |
834ffca6 | 832 | */ |
3d751af2 | 833 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
834ffca6 DC |
834 | WARN_ON_ONCE(ret); |
835 | ret = 0; | |
f0d26e86 DC |
836 | } |
837 | ||
eda77982 DC |
838 | /* |
839 | * If we are doing unaligned IO, wait for all other IO to drain, | |
840 | * otherwise demote the lock if we had to flush cached pages | |
841 | */ | |
842 | if (unaligned_io) | |
4a06fd26 | 843 | inode_dio_wait(inode); |
d0606464 | 844 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 845 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 846 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
847 | } |
848 | ||
3176c3e0 | 849 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos); |
f0d26e86 | 850 | |
0cefb29e | 851 | data = *from; |
16d4d435 CH |
852 | ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data, |
853 | xfs_get_blocks_direct, xfs_end_io_direct_write, | |
854 | NULL, DIO_ASYNC_EXTEND); | |
0cefb29e DC |
855 | |
856 | /* see generic_file_direct_write() for why this is necessary */ | |
857 | if (mapping->nrpages) { | |
858 | invalidate_inode_pages2_range(mapping, | |
13712713 | 859 | iocb->ki_pos >> PAGE_SHIFT, |
09cbfeaf | 860 | end >> PAGE_SHIFT); |
0cefb29e DC |
861 | } |
862 | ||
863 | if (ret > 0) { | |
13712713 | 864 | iocb->ki_pos += ret; |
0cefb29e | 865 | iov_iter_advance(from, ret); |
0cefb29e | 866 | } |
d0606464 CH |
867 | out: |
868 | xfs_rw_iunlock(ip, iolock); | |
869 | ||
6b698ede | 870 | /* |
16d4d435 CH |
871 | * No fallback to buffered IO on errors for XFS, direct IO will either |
872 | * complete fully or fail. | |
873 | */ | |
874 | ASSERT(ret < 0 || ret == count); | |
875 | return ret; | |
876 | } | |
877 | ||
878 | STATIC ssize_t | |
879 | xfs_file_dax_write( | |
880 | struct kiocb *iocb, | |
881 | struct iov_iter *from) | |
882 | { | |
883 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
884 | struct inode *inode = mapping->host; | |
885 | struct xfs_inode *ip = XFS_I(inode); | |
886 | struct xfs_mount *mp = ip->i_mount; | |
887 | ssize_t ret = 0; | |
888 | int unaligned_io = 0; | |
889 | int iolock; | |
890 | struct iov_iter data; | |
891 | ||
892 | /* "unaligned" here means not aligned to a filesystem block */ | |
893 | if ((iocb->ki_pos & mp->m_blockmask) || | |
894 | ((iocb->ki_pos + iov_iter_count(from)) & mp->m_blockmask)) { | |
895 | unaligned_io = 1; | |
896 | iolock = XFS_IOLOCK_EXCL; | |
897 | } else if (mapping->nrpages) { | |
898 | iolock = XFS_IOLOCK_EXCL; | |
899 | } else { | |
900 | iolock = XFS_IOLOCK_SHARED; | |
901 | } | |
902 | xfs_rw_ilock(ip, iolock); | |
903 | ||
904 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); | |
905 | if (ret) | |
906 | goto out; | |
907 | ||
908 | /* | |
909 | * Yes, even DAX files can have page cache attached to them: A zeroed | |
910 | * page is inserted into the pagecache when we have to serve a write | |
911 | * fault on a hole. It should never be dirtied and can simply be | |
912 | * dropped from the pagecache once we get real data for the page. | |
6b698ede | 913 | */ |
16d4d435 CH |
914 | if (mapping->nrpages) { |
915 | ret = invalidate_inode_pages2(mapping); | |
916 | WARN_ON_ONCE(ret); | |
917 | } | |
918 | ||
919 | if (iolock == XFS_IOLOCK_EXCL && !unaligned_io) { | |
920 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); | |
921 | iolock = XFS_IOLOCK_SHARED; | |
922 | } | |
923 | ||
924 | trace_xfs_file_dax_write(ip, iov_iter_count(from), iocb->ki_pos); | |
925 | ||
926 | data = *from; | |
927 | ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, | |
928 | xfs_end_io_direct_write, 0); | |
929 | if (ret > 0) { | |
930 | iocb->ki_pos += ret; | |
931 | iov_iter_advance(from, ret); | |
932 | } | |
933 | out: | |
934 | xfs_rw_iunlock(ip, iolock); | |
f0d26e86 DC |
935 | return ret; |
936 | } | |
937 | ||
00258e36 | 938 | STATIC ssize_t |
637bbc75 | 939 | xfs_file_buffered_aio_write( |
dda35b8f | 940 | struct kiocb *iocb, |
b3188919 | 941 | struct iov_iter *from) |
dda35b8f CH |
942 | { |
943 | struct file *file = iocb->ki_filp; | |
944 | struct address_space *mapping = file->f_mapping; | |
945 | struct inode *inode = mapping->host; | |
00258e36 | 946 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
947 | ssize_t ret; |
948 | int enospc = 0; | |
d0606464 | 949 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 950 | |
d0606464 | 951 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 952 | |
99733fa3 | 953 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 954 | if (ret) |
d0606464 | 955 | goto out; |
dda35b8f CH |
956 | |
957 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 958 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 959 | |
dda35b8f | 960 | write_retry: |
3176c3e0 | 961 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos); |
99733fa3 | 962 | ret = generic_perform_write(file, from, iocb->ki_pos); |
0a64bc2c | 963 | if (likely(ret >= 0)) |
99733fa3 | 964 | iocb->ki_pos += ret; |
dc06f398 | 965 | |
637bbc75 | 966 | /* |
dc06f398 BF |
967 | * If we hit a space limit, try to free up some lingering preallocated |
968 | * space before returning an error. In the case of ENOSPC, first try to | |
969 | * write back all dirty inodes to free up some of the excess reserved | |
970 | * metadata space. This reduces the chances that the eofblocks scan | |
971 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
972 | * also behaves as a filter to prevent too many eofblocks scans from | |
973 | * running at the same time. | |
637bbc75 | 974 | */ |
dc06f398 BF |
975 | if (ret == -EDQUOT && !enospc) { |
976 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
977 | if (enospc) | |
978 | goto write_retry; | |
979 | } else if (ret == -ENOSPC && !enospc) { | |
980 | struct xfs_eofblocks eofb = {0}; | |
981 | ||
637bbc75 | 982 | enospc = 1; |
9aa05000 | 983 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
984 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
985 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
986 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 987 | goto write_retry; |
dda35b8f | 988 | } |
d0606464 | 989 | |
dda35b8f | 990 | current->backing_dev_info = NULL; |
d0606464 CH |
991 | out: |
992 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
993 | return ret; |
994 | } | |
995 | ||
996 | STATIC ssize_t | |
bf97f3bc | 997 | xfs_file_write_iter( |
637bbc75 | 998 | struct kiocb *iocb, |
bf97f3bc | 999 | struct iov_iter *from) |
637bbc75 DC |
1000 | { |
1001 | struct file *file = iocb->ki_filp; | |
1002 | struct address_space *mapping = file->f_mapping; | |
1003 | struct inode *inode = mapping->host; | |
1004 | struct xfs_inode *ip = XFS_I(inode); | |
1005 | ssize_t ret; | |
bf97f3bc | 1006 | size_t ocount = iov_iter_count(from); |
637bbc75 | 1007 | |
ff6d6af2 | 1008 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 1009 | |
637bbc75 DC |
1010 | if (ocount == 0) |
1011 | return 0; | |
1012 | ||
bf97f3bc AV |
1013 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
1014 | return -EIO; | |
637bbc75 | 1015 | |
16d4d435 CH |
1016 | if (IS_DAX(inode)) |
1017 | ret = xfs_file_dax_write(iocb, from); | |
1018 | else if (iocb->ki_flags & IOCB_DIRECT) | |
bf97f3bc | 1019 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 1020 | else |
bf97f3bc | 1021 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 1022 | |
d0606464 | 1023 | if (ret > 0) { |
ff6d6af2 | 1024 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 1025 | |
d0606464 | 1026 | /* Handle various SYNC-type writes */ |
e2592217 | 1027 | ret = generic_write_sync(iocb, ret); |
dda35b8f | 1028 | } |
a363f0c2 | 1029 | return ret; |
dda35b8f CH |
1030 | } |
1031 | ||
a904b1ca NJ |
1032 | #define XFS_FALLOC_FL_SUPPORTED \ |
1033 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
1034 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
1035 | FALLOC_FL_INSERT_RANGE) | |
1036 | ||
2fe17c10 CH |
1037 | STATIC long |
1038 | xfs_file_fallocate( | |
83aee9e4 CH |
1039 | struct file *file, |
1040 | int mode, | |
1041 | loff_t offset, | |
1042 | loff_t len) | |
2fe17c10 | 1043 | { |
83aee9e4 CH |
1044 | struct inode *inode = file_inode(file); |
1045 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 1046 | long error; |
8add71ca | 1047 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 1048 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 1049 | loff_t new_size = 0; |
a904b1ca | 1050 | bool do_file_insert = 0; |
2fe17c10 | 1051 | |
83aee9e4 CH |
1052 | if (!S_ISREG(inode->i_mode)) |
1053 | return -EINVAL; | |
a904b1ca | 1054 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
1055 | return -EOPNOTSUPP; |
1056 | ||
781355c6 | 1057 | xfs_ilock(ip, iolock); |
21c3ea18 | 1058 | error = xfs_break_layouts(inode, &iolock, false); |
781355c6 CH |
1059 | if (error) |
1060 | goto out_unlock; | |
1061 | ||
e8e9ad42 DC |
1062 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
1063 | iolock |= XFS_MMAPLOCK_EXCL; | |
1064 | ||
83aee9e4 CH |
1065 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
1066 | error = xfs_free_file_space(ip, offset, len); | |
1067 | if (error) | |
1068 | goto out_unlock; | |
e1d8fb88 NJ |
1069 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
1070 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
1071 | ||
1072 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 1073 | error = -EINVAL; |
e1d8fb88 NJ |
1074 | goto out_unlock; |
1075 | } | |
1076 | ||
23fffa92 LC |
1077 | /* |
1078 | * There is no need to overlap collapse range with EOF, | |
1079 | * in which case it is effectively a truncate operation | |
1080 | */ | |
1081 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 1082 | error = -EINVAL; |
23fffa92 LC |
1083 | goto out_unlock; |
1084 | } | |
1085 | ||
e1d8fb88 NJ |
1086 | new_size = i_size_read(inode) - len; |
1087 | ||
1088 | error = xfs_collapse_file_space(ip, offset, len); | |
1089 | if (error) | |
1090 | goto out_unlock; | |
a904b1ca NJ |
1091 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
1092 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
1093 | ||
1094 | new_size = i_size_read(inode) + len; | |
1095 | if (offset & blksize_mask || len & blksize_mask) { | |
1096 | error = -EINVAL; | |
1097 | goto out_unlock; | |
1098 | } | |
1099 | ||
1100 | /* check the new inode size does not wrap through zero */ | |
1101 | if (new_size > inode->i_sb->s_maxbytes) { | |
1102 | error = -EFBIG; | |
1103 | goto out_unlock; | |
1104 | } | |
1105 | ||
1106 | /* Offset should be less than i_size */ | |
1107 | if (offset >= i_size_read(inode)) { | |
1108 | error = -EINVAL; | |
1109 | goto out_unlock; | |
1110 | } | |
1111 | do_file_insert = 1; | |
83aee9e4 | 1112 | } else { |
8add71ca CH |
1113 | flags |= XFS_PREALLOC_SET; |
1114 | ||
83aee9e4 CH |
1115 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
1116 | offset + len > i_size_read(inode)) { | |
1117 | new_size = offset + len; | |
2451337d | 1118 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
1119 | if (error) |
1120 | goto out_unlock; | |
1121 | } | |
2fe17c10 | 1122 | |
376ba313 LC |
1123 | if (mode & FALLOC_FL_ZERO_RANGE) |
1124 | error = xfs_zero_file_space(ip, offset, len); | |
1125 | else | |
1126 | error = xfs_alloc_file_space(ip, offset, len, | |
1127 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
1128 | if (error) |
1129 | goto out_unlock; | |
1130 | } | |
1131 | ||
83aee9e4 | 1132 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
1133 | flags |= XFS_PREALLOC_SYNC; |
1134 | ||
1135 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
1136 | if (error) |
1137 | goto out_unlock; | |
1138 | ||
1139 | /* Change file size if needed */ | |
1140 | if (new_size) { | |
1141 | struct iattr iattr; | |
1142 | ||
1143 | iattr.ia_valid = ATTR_SIZE; | |
1144 | iattr.ia_size = new_size; | |
83aee9e4 | 1145 | error = xfs_setattr_size(ip, &iattr); |
a904b1ca NJ |
1146 | if (error) |
1147 | goto out_unlock; | |
2fe17c10 CH |
1148 | } |
1149 | ||
a904b1ca NJ |
1150 | /* |
1151 | * Perform hole insertion now that the file size has been | |
1152 | * updated so that if we crash during the operation we don't | |
1153 | * leave shifted extents past EOF and hence losing access to | |
1154 | * the data that is contained within them. | |
1155 | */ | |
1156 | if (do_file_insert) | |
1157 | error = xfs_insert_file_space(ip, offset, len); | |
1158 | ||
2fe17c10 | 1159 | out_unlock: |
781355c6 | 1160 | xfs_iunlock(ip, iolock); |
2451337d | 1161 | return error; |
2fe17c10 CH |
1162 | } |
1163 | ||
1164 | ||
1da177e4 | 1165 | STATIC int |
3562fd45 | 1166 | xfs_file_open( |
1da177e4 | 1167 | struct inode *inode, |
f999a5bf | 1168 | struct file *file) |
1da177e4 | 1169 | { |
f999a5bf | 1170 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1171 | return -EFBIG; |
f999a5bf CH |
1172 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1173 | return -EIO; | |
1174 | return 0; | |
1175 | } | |
1176 | ||
1177 | STATIC int | |
1178 | xfs_dir_open( | |
1179 | struct inode *inode, | |
1180 | struct file *file) | |
1181 | { | |
1182 | struct xfs_inode *ip = XFS_I(inode); | |
1183 | int mode; | |
1184 | int error; | |
1185 | ||
1186 | error = xfs_file_open(inode, file); | |
1187 | if (error) | |
1188 | return error; | |
1189 | ||
1190 | /* | |
1191 | * If there are any blocks, read-ahead block 0 as we're almost | |
1192 | * certain to have the next operation be a read there. | |
1193 | */ | |
309ecac8 | 1194 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1195 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 1196 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
1197 | xfs_iunlock(ip, mode); |
1198 | return 0; | |
1da177e4 LT |
1199 | } |
1200 | ||
1da177e4 | 1201 | STATIC int |
3562fd45 | 1202 | xfs_file_release( |
1da177e4 LT |
1203 | struct inode *inode, |
1204 | struct file *filp) | |
1205 | { | |
2451337d | 1206 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1207 | } |
1208 | ||
1da177e4 | 1209 | STATIC int |
3562fd45 | 1210 | xfs_file_readdir( |
b8227554 AV |
1211 | struct file *file, |
1212 | struct dir_context *ctx) | |
1da177e4 | 1213 | { |
b8227554 | 1214 | struct inode *inode = file_inode(file); |
739bfb2a | 1215 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1216 | size_t bufsize; |
1217 | ||
1218 | /* | |
1219 | * The Linux API doesn't pass down the total size of the buffer | |
1220 | * we read into down to the filesystem. With the filldir concept | |
1221 | * it's not needed for correct information, but the XFS dir2 leaf | |
1222 | * code wants an estimate of the buffer size to calculate it's | |
1223 | * readahead window and size the buffers used for mapping to | |
1224 | * physical blocks. | |
1225 | * | |
1226 | * Try to give it an estimate that's good enough, maybe at some | |
1227 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1228 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1229 | */ |
a9cc799e | 1230 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 1231 | |
8300475e | 1232 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
1233 | } |
1234 | ||
d126d43f JL |
1235 | /* |
1236 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1237 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1238 | */ |
1239 | enum { | |
1240 | HOLE_OFF = 0, | |
1241 | DATA_OFF, | |
1242 | }; | |
1243 | ||
1244 | /* | |
1245 | * Lookup the desired type of offset from the given page. | |
1246 | * | |
1247 | * On success, return true and the offset argument will point to the | |
1248 | * start of the region that was found. Otherwise this function will | |
1249 | * return false and keep the offset argument unchanged. | |
1250 | */ | |
1251 | STATIC bool | |
1252 | xfs_lookup_buffer_offset( | |
1253 | struct page *page, | |
1254 | loff_t *offset, | |
1255 | unsigned int type) | |
1256 | { | |
1257 | loff_t lastoff = page_offset(page); | |
1258 | bool found = false; | |
1259 | struct buffer_head *bh, *head; | |
1260 | ||
1261 | bh = head = page_buffers(page); | |
1262 | do { | |
1263 | /* | |
1264 | * Unwritten extents that have data in the page | |
1265 | * cache covering them can be identified by the | |
1266 | * BH_Unwritten state flag. Pages with multiple | |
1267 | * buffers might have a mix of holes, data and | |
1268 | * unwritten extents - any buffer with valid | |
1269 | * data in it should have BH_Uptodate flag set | |
1270 | * on it. | |
1271 | */ | |
1272 | if (buffer_unwritten(bh) || | |
1273 | buffer_uptodate(bh)) { | |
1274 | if (type == DATA_OFF) | |
1275 | found = true; | |
1276 | } else { | |
1277 | if (type == HOLE_OFF) | |
1278 | found = true; | |
1279 | } | |
1280 | ||
1281 | if (found) { | |
1282 | *offset = lastoff; | |
1283 | break; | |
1284 | } | |
1285 | lastoff += bh->b_size; | |
1286 | } while ((bh = bh->b_this_page) != head); | |
1287 | ||
1288 | return found; | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * This routine is called to find out and return a data or hole offset | |
1293 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1294 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1295 | * |
1296 | * The argument offset is used to tell where we start to search from the | |
1297 | * page cache. Map is used to figure out the end points of the range to | |
1298 | * lookup pages. | |
1299 | * | |
1300 | * Return true if the desired type of offset was found, and the argument | |
1301 | * offset is filled with that address. Otherwise, return false and keep | |
1302 | * offset unchanged. | |
1303 | */ | |
1304 | STATIC bool | |
1305 | xfs_find_get_desired_pgoff( | |
1306 | struct inode *inode, | |
1307 | struct xfs_bmbt_irec *map, | |
1308 | unsigned int type, | |
1309 | loff_t *offset) | |
1310 | { | |
1311 | struct xfs_inode *ip = XFS_I(inode); | |
1312 | struct xfs_mount *mp = ip->i_mount; | |
1313 | struct pagevec pvec; | |
1314 | pgoff_t index; | |
1315 | pgoff_t end; | |
1316 | loff_t endoff; | |
1317 | loff_t startoff = *offset; | |
1318 | loff_t lastoff = startoff; | |
1319 | bool found = false; | |
1320 | ||
1321 | pagevec_init(&pvec, 0); | |
1322 | ||
09cbfeaf | 1323 | index = startoff >> PAGE_SHIFT; |
d126d43f | 1324 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
09cbfeaf | 1325 | end = endoff >> PAGE_SHIFT; |
d126d43f JL |
1326 | do { |
1327 | int want; | |
1328 | unsigned nr_pages; | |
1329 | unsigned int i; | |
1330 | ||
1331 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1332 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1333 | want); | |
1334 | /* | |
1335 | * No page mapped into given range. If we are searching holes | |
1336 | * and if this is the first time we got into the loop, it means | |
1337 | * that the given offset is landed in a hole, return it. | |
1338 | * | |
1339 | * If we have already stepped through some block buffers to find | |
1340 | * holes but they all contains data. In this case, the last | |
1341 | * offset is already updated and pointed to the end of the last | |
1342 | * mapped page, if it does not reach the endpoint to search, | |
1343 | * that means there should be a hole between them. | |
1344 | */ | |
1345 | if (nr_pages == 0) { | |
1346 | /* Data search found nothing */ | |
1347 | if (type == DATA_OFF) | |
1348 | break; | |
1349 | ||
1350 | ASSERT(type == HOLE_OFF); | |
1351 | if (lastoff == startoff || lastoff < endoff) { | |
1352 | found = true; | |
1353 | *offset = lastoff; | |
1354 | } | |
1355 | break; | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * At lease we found one page. If this is the first time we | |
1360 | * step into the loop, and if the first page index offset is | |
1361 | * greater than the given search offset, a hole was found. | |
1362 | */ | |
1363 | if (type == HOLE_OFF && lastoff == startoff && | |
1364 | lastoff < page_offset(pvec.pages[0])) { | |
1365 | found = true; | |
1366 | break; | |
1367 | } | |
1368 | ||
1369 | for (i = 0; i < nr_pages; i++) { | |
1370 | struct page *page = pvec.pages[i]; | |
1371 | loff_t b_offset; | |
1372 | ||
1373 | /* | |
1374 | * At this point, the page may be truncated or | |
1375 | * invalidated (changing page->mapping to NULL), | |
1376 | * or even swizzled back from swapper_space to tmpfs | |
1377 | * file mapping. However, page->index will not change | |
1378 | * because we have a reference on the page. | |
1379 | * | |
1380 | * Searching done if the page index is out of range. | |
1381 | * If the current offset is not reaches the end of | |
1382 | * the specified search range, there should be a hole | |
1383 | * between them. | |
1384 | */ | |
1385 | if (page->index > end) { | |
1386 | if (type == HOLE_OFF && lastoff < endoff) { | |
1387 | *offset = lastoff; | |
1388 | found = true; | |
1389 | } | |
1390 | goto out; | |
1391 | } | |
1392 | ||
1393 | lock_page(page); | |
1394 | /* | |
1395 | * Page truncated or invalidated(page->mapping == NULL). | |
1396 | * We can freely skip it and proceed to check the next | |
1397 | * page. | |
1398 | */ | |
1399 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1400 | unlock_page(page); | |
1401 | continue; | |
1402 | } | |
1403 | ||
1404 | if (!page_has_buffers(page)) { | |
1405 | unlock_page(page); | |
1406 | continue; | |
1407 | } | |
1408 | ||
1409 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1410 | if (found) { | |
1411 | /* | |
1412 | * The found offset may be less than the start | |
1413 | * point to search if this is the first time to | |
1414 | * come here. | |
1415 | */ | |
1416 | *offset = max_t(loff_t, startoff, b_offset); | |
1417 | unlock_page(page); | |
1418 | goto out; | |
1419 | } | |
1420 | ||
1421 | /* | |
1422 | * We either searching data but nothing was found, or | |
1423 | * searching hole but found a data buffer. In either | |
1424 | * case, probably the next page contains the desired | |
1425 | * things, update the last offset to it so. | |
1426 | */ | |
1427 | lastoff = page_offset(page) + PAGE_SIZE; | |
1428 | unlock_page(page); | |
1429 | } | |
1430 | ||
1431 | /* | |
1432 | * The number of returned pages less than our desired, search | |
1433 | * done. In this case, nothing was found for searching data, | |
1434 | * but we found a hole behind the last offset. | |
1435 | */ | |
1436 | if (nr_pages < want) { | |
1437 | if (type == HOLE_OFF) { | |
1438 | *offset = lastoff; | |
1439 | found = true; | |
1440 | } | |
1441 | break; | |
1442 | } | |
1443 | ||
1444 | index = pvec.pages[i - 1]->index + 1; | |
1445 | pagevec_release(&pvec); | |
1446 | } while (index <= end); | |
1447 | ||
1448 | out: | |
1449 | pagevec_release(&pvec); | |
1450 | return found; | |
1451 | } | |
1452 | ||
8aa7d37e ES |
1453 | /* |
1454 | * caller must lock inode with xfs_ilock_data_map_shared, | |
1455 | * can we craft an appropriate ASSERT? | |
1456 | * | |
1457 | * end is because the VFS-level lseek interface is defined such that any | |
1458 | * offset past i_size shall return -ENXIO, but we use this for quota code | |
1459 | * which does not maintain i_size, and we want to SEEK_DATA past i_size. | |
1460 | */ | |
1461 | loff_t | |
1462 | __xfs_seek_hole_data( | |
1463 | struct inode *inode, | |
49c69591 | 1464 | loff_t start, |
8aa7d37e | 1465 | loff_t end, |
49c69591 | 1466 | int whence) |
3fe3e6b1 | 1467 | { |
3fe3e6b1 JL |
1468 | struct xfs_inode *ip = XFS_I(inode); |
1469 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 | 1470 | loff_t uninitialized_var(offset); |
3fe3e6b1 | 1471 | xfs_fileoff_t fsbno; |
8aa7d37e | 1472 | xfs_filblks_t lastbno; |
3fe3e6b1 JL |
1473 | int error; |
1474 | ||
8aa7d37e | 1475 | if (start >= end) { |
2451337d | 1476 | error = -ENXIO; |
8aa7d37e | 1477 | goto out_error; |
3fe3e6b1 JL |
1478 | } |
1479 | ||
3fe3e6b1 JL |
1480 | /* |
1481 | * Try to read extents from the first block indicated | |
1482 | * by fsbno to the end block of the file. | |
1483 | */ | |
52f1acc8 | 1484 | fsbno = XFS_B_TO_FSBT(mp, start); |
8aa7d37e | 1485 | lastbno = XFS_B_TO_FSB(mp, end); |
49c69591 | 1486 | |
52f1acc8 JL |
1487 | for (;;) { |
1488 | struct xfs_bmbt_irec map[2]; | |
1489 | int nmap = 2; | |
1490 | unsigned int i; | |
3fe3e6b1 | 1491 | |
8aa7d37e | 1492 | error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap, |
52f1acc8 JL |
1493 | XFS_BMAPI_ENTIRE); |
1494 | if (error) | |
8aa7d37e | 1495 | goto out_error; |
3fe3e6b1 | 1496 | |
52f1acc8 JL |
1497 | /* No extents at given offset, must be beyond EOF */ |
1498 | if (nmap == 0) { | |
2451337d | 1499 | error = -ENXIO; |
8aa7d37e | 1500 | goto out_error; |
52f1acc8 JL |
1501 | } |
1502 | ||
1503 | for (i = 0; i < nmap; i++) { | |
1504 | offset = max_t(loff_t, start, | |
1505 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1506 | ||
49c69591 ES |
1507 | /* Landed in the hole we wanted? */ |
1508 | if (whence == SEEK_HOLE && | |
1509 | map[i].br_startblock == HOLESTARTBLOCK) | |
1510 | goto out; | |
1511 | ||
1512 | /* Landed in the data extent we wanted? */ | |
1513 | if (whence == SEEK_DATA && | |
1514 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1515 | (map[i].br_state == XFS_EXT_NORM && | |
1516 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1517 | goto out; |
1518 | ||
1519 | /* | |
49c69591 ES |
1520 | * Landed in an unwritten extent, try to search |
1521 | * for hole or data from page cache. | |
52f1acc8 JL |
1522 | */ |
1523 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1524 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1525 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1526 | &offset)) | |
52f1acc8 JL |
1527 | goto out; |
1528 | } | |
1529 | } | |
1530 | ||
1531 | /* | |
49c69591 ES |
1532 | * We only received one extent out of the two requested. This |
1533 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1534 | */ |
3fe3e6b1 | 1535 | if (nmap == 1) { |
49c69591 ES |
1536 | /* |
1537 | * If we were looking for a hole, set offset to | |
1538 | * the end of the file (i.e., there is an implicit | |
1539 | * hole at the end of any file). | |
1540 | */ | |
1541 | if (whence == SEEK_HOLE) { | |
8aa7d37e | 1542 | offset = end; |
49c69591 ES |
1543 | break; |
1544 | } | |
1545 | /* | |
1546 | * If we were looking for data, it's nowhere to be found | |
1547 | */ | |
1548 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1549 | error = -ENXIO; |
8aa7d37e | 1550 | goto out_error; |
3fe3e6b1 JL |
1551 | } |
1552 | ||
52f1acc8 JL |
1553 | ASSERT(i > 1); |
1554 | ||
1555 | /* | |
1556 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1557 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1558 | */ |
1559 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1560 | start = XFS_FSB_TO_B(mp, fsbno); | |
8aa7d37e | 1561 | if (start >= end) { |
49c69591 | 1562 | if (whence == SEEK_HOLE) { |
8aa7d37e | 1563 | offset = end; |
49c69591 ES |
1564 | break; |
1565 | } | |
1566 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1567 | error = -ENXIO; |
8aa7d37e | 1568 | goto out_error; |
52f1acc8 | 1569 | } |
3fe3e6b1 JL |
1570 | } |
1571 | ||
b686d1f7 JL |
1572 | out: |
1573 | /* | |
49c69591 | 1574 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1575 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1576 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1577 | * situation in particular. |
1578 | */ | |
49c69591 | 1579 | if (whence == SEEK_HOLE) |
8aa7d37e ES |
1580 | offset = min_t(loff_t, offset, end); |
1581 | ||
1582 | return offset; | |
1583 | ||
1584 | out_error: | |
1585 | return error; | |
1586 | } | |
1587 | ||
1588 | STATIC loff_t | |
1589 | xfs_seek_hole_data( | |
1590 | struct file *file, | |
1591 | loff_t start, | |
1592 | int whence) | |
1593 | { | |
1594 | struct inode *inode = file->f_mapping->host; | |
1595 | struct xfs_inode *ip = XFS_I(inode); | |
1596 | struct xfs_mount *mp = ip->i_mount; | |
1597 | uint lock; | |
1598 | loff_t offset, end; | |
1599 | int error = 0; | |
1600 | ||
1601 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1602 | return -EIO; | |
1603 | ||
1604 | lock = xfs_ilock_data_map_shared(ip); | |
1605 | ||
1606 | end = i_size_read(inode); | |
1607 | offset = __xfs_seek_hole_data(inode, start, end, whence); | |
1608 | if (offset < 0) { | |
1609 | error = offset; | |
1610 | goto out_unlock; | |
1611 | } | |
1612 | ||
46a1c2c7 | 1613 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1614 | |
1615 | out_unlock: | |
01f4f327 | 1616 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1617 | |
1618 | if (error) | |
2451337d | 1619 | return error; |
3fe3e6b1 JL |
1620 | return offset; |
1621 | } | |
1622 | ||
1623 | STATIC loff_t | |
1624 | xfs_file_llseek( | |
1625 | struct file *file, | |
1626 | loff_t offset, | |
59f9c004 | 1627 | int whence) |
3fe3e6b1 | 1628 | { |
59f9c004 | 1629 | switch (whence) { |
3fe3e6b1 JL |
1630 | case SEEK_END: |
1631 | case SEEK_CUR: | |
1632 | case SEEK_SET: | |
59f9c004 | 1633 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1634 | case SEEK_HOLE: |
49c69591 | 1635 | case SEEK_DATA: |
59f9c004 | 1636 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1637 | default: |
1638 | return -EINVAL; | |
1639 | } | |
1640 | } | |
1641 | ||
de0e8c20 DC |
1642 | /* |
1643 | * Locking for serialisation of IO during page faults. This results in a lock | |
1644 | * ordering of: | |
1645 | * | |
1646 | * mmap_sem (MM) | |
6b698ede | 1647 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1648 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1649 | * page_lock (MM) |
1650 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1651 | */ |
de0e8c20 | 1652 | |
075a924d DC |
1653 | /* |
1654 | * mmap()d file has taken write protection fault and is being made writable. We | |
1655 | * can set the page state up correctly for a writable page, which means we can | |
1656 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1657 | * mapping. | |
de0e8c20 DC |
1658 | */ |
1659 | STATIC int | |
075a924d | 1660 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1661 | struct vm_area_struct *vma, |
1662 | struct vm_fault *vmf) | |
1663 | { | |
6b698ede | 1664 | struct inode *inode = file_inode(vma->vm_file); |
ec56b1f1 | 1665 | int ret; |
de0e8c20 | 1666 | |
6b698ede | 1667 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1668 | |
6b698ede | 1669 | sb_start_pagefault(inode->i_sb); |
ec56b1f1 | 1670 | file_update_time(vma->vm_file); |
6b698ede | 1671 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1672 | |
6b698ede | 1673 | if (IS_DAX(inode)) { |
02fbd139 | 1674 | ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault); |
6b698ede | 1675 | } else { |
5c500029 | 1676 | ret = block_page_mkwrite(vma, vmf, xfs_get_blocks); |
6b698ede DC |
1677 | ret = block_page_mkwrite_return(ret); |
1678 | } | |
1679 | ||
1680 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1681 | sb_end_pagefault(inode->i_sb); | |
1682 | ||
1683 | return ret; | |
de0e8c20 DC |
1684 | } |
1685 | ||
075a924d | 1686 | STATIC int |
6b698ede | 1687 | xfs_filemap_fault( |
075a924d DC |
1688 | struct vm_area_struct *vma, |
1689 | struct vm_fault *vmf) | |
1690 | { | |
b2442c5a | 1691 | struct inode *inode = file_inode(vma->vm_file); |
6b698ede | 1692 | int ret; |
ec56b1f1 | 1693 | |
b2442c5a | 1694 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1695 | |
6b698ede | 1696 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1697 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
6b698ede | 1698 | return xfs_filemap_page_mkwrite(vma, vmf); |
075a924d | 1699 | |
b2442c5a DC |
1700 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1701 | if (IS_DAX(inode)) { | |
1702 | /* | |
1703 | * we do not want to trigger unwritten extent conversion on read | |
1704 | * faults - that is unnecessary overhead and would also require | |
1705 | * changes to xfs_get_blocks_direct() to map unwritten extent | |
1706 | * ioend for conversion on read-only mappings. | |
1707 | */ | |
02fbd139 | 1708 | ret = __dax_fault(vma, vmf, xfs_get_blocks_dax_fault); |
b2442c5a DC |
1709 | } else |
1710 | ret = filemap_fault(vma, vmf); | |
1711 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
075a924d | 1712 | |
6b698ede DC |
1713 | return ret; |
1714 | } | |
1715 | ||
13ad4fe3 DC |
1716 | /* |
1717 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on | |
1718 | * both read and write faults. Hence we need to handle both cases. There is no | |
1719 | * ->pmd_mkwrite callout for huge pages, so we have a single function here to | |
1720 | * handle both cases here. @flags carries the information on the type of fault | |
1721 | * occuring. | |
1722 | */ | |
acd76e74 MW |
1723 | STATIC int |
1724 | xfs_filemap_pmd_fault( | |
1725 | struct vm_area_struct *vma, | |
1726 | unsigned long addr, | |
1727 | pmd_t *pmd, | |
1728 | unsigned int flags) | |
1729 | { | |
1730 | struct inode *inode = file_inode(vma->vm_file); | |
1731 | struct xfs_inode *ip = XFS_I(inode); | |
1732 | int ret; | |
1733 | ||
1734 | if (!IS_DAX(inode)) | |
1735 | return VM_FAULT_FALLBACK; | |
1736 | ||
1737 | trace_xfs_filemap_pmd_fault(ip); | |
1738 | ||
13ad4fe3 DC |
1739 | if (flags & FAULT_FLAG_WRITE) { |
1740 | sb_start_pagefault(inode->i_sb); | |
1741 | file_update_time(vma->vm_file); | |
1742 | } | |
1743 | ||
acd76e74 | 1744 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
02fbd139 | 1745 | ret = __dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault); |
acd76e74 | 1746 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acd76e74 | 1747 | |
13ad4fe3 DC |
1748 | if (flags & FAULT_FLAG_WRITE) |
1749 | sb_end_pagefault(inode->i_sb); | |
acd76e74 MW |
1750 | |
1751 | return ret; | |
1752 | } | |
1753 | ||
3af49285 DC |
1754 | /* |
1755 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1756 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1757 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1758 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1759 | */ |
1760 | static int | |
1761 | xfs_filemap_pfn_mkwrite( | |
1762 | struct vm_area_struct *vma, | |
1763 | struct vm_fault *vmf) | |
1764 | { | |
1765 | ||
1766 | struct inode *inode = file_inode(vma->vm_file); | |
1767 | struct xfs_inode *ip = XFS_I(inode); | |
1768 | int ret = VM_FAULT_NOPAGE; | |
1769 | loff_t size; | |
1770 | ||
1771 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1772 | ||
1773 | sb_start_pagefault(inode->i_sb); | |
1774 | file_update_time(vma->vm_file); | |
1775 | ||
1776 | /* check if the faulting page hasn't raced with truncate */ | |
1777 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1778 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1779 | if (vmf->pgoff >= size) | |
1780 | ret = VM_FAULT_SIGBUS; | |
5eb88dca RZ |
1781 | else if (IS_DAX(inode)) |
1782 | ret = dax_pfn_mkwrite(vma, vmf); | |
3af49285 DC |
1783 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1784 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1785 | return ret; |
3af49285 | 1786 | |
acd76e74 MW |
1787 | } |
1788 | ||
6b698ede DC |
1789 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1790 | .fault = xfs_filemap_fault, | |
acd76e74 | 1791 | .pmd_fault = xfs_filemap_pmd_fault, |
6b698ede DC |
1792 | .map_pages = filemap_map_pages, |
1793 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1794 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1795 | }; |
1796 | ||
1797 | STATIC int | |
1798 | xfs_file_mmap( | |
1799 | struct file *filp, | |
1800 | struct vm_area_struct *vma) | |
1801 | { | |
1802 | file_accessed(filp); | |
1803 | vma->vm_ops = &xfs_file_vm_ops; | |
1804 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1805 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1806 | return 0; |
075a924d DC |
1807 | } |
1808 | ||
4b6f5d20 | 1809 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1810 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1811 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1812 | .write_iter = xfs_file_write_iter, |
1b895840 | 1813 | .splice_read = xfs_file_splice_read, |
8d020765 | 1814 | .splice_write = iter_file_splice_write, |
3562fd45 | 1815 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1816 | #ifdef CONFIG_COMPAT |
3562fd45 | 1817 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1818 | #endif |
3562fd45 NS |
1819 | .mmap = xfs_file_mmap, |
1820 | .open = xfs_file_open, | |
1821 | .release = xfs_file_release, | |
1822 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1823 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1824 | }; |
1825 | ||
4b6f5d20 | 1826 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1827 | .open = xfs_dir_open, |
1da177e4 | 1828 | .read = generic_read_dir, |
3b0a3c1a | 1829 | .iterate_shared = xfs_file_readdir, |
59af1584 | 1830 | .llseek = generic_file_llseek, |
3562fd45 | 1831 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1832 | #ifdef CONFIG_COMPAT |
3562fd45 | 1833 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1834 | #endif |
1da2f2db | 1835 | .fsync = xfs_dir_fsync, |
1da177e4 | 1836 | }; |