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 | 40 | |
a27bb332 | 41 | #include <linux/aio.h> |
1da177e4 | 42 | #include <linux/dcache.h> |
2fe17c10 | 43 | #include <linux/falloc.h> |
d126d43f | 44 | #include <linux/pagevec.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) | |
58 | mutex_lock(&VFS_I(ip)->i_mutex); | |
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) | |
69 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
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) | |
79 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
80 | } | |
81 | ||
dda35b8f CH |
82 | /* |
83 | * xfs_iozero | |
84 | * | |
85 | * xfs_iozero clears the specified range of buffer supplied, | |
86 | * and marks all the affected blocks as valid and modified. If | |
87 | * an affected block is not allocated, it will be allocated. If | |
88 | * an affected block is not completely overwritten, and is not | |
89 | * valid before the operation, it will be read from disk before | |
90 | * being partially zeroed. | |
91 | */ | |
ef9d8733 | 92 | int |
dda35b8f CH |
93 | xfs_iozero( |
94 | struct xfs_inode *ip, /* inode */ | |
95 | loff_t pos, /* offset in file */ | |
96 | size_t count) /* size of data to zero */ | |
97 | { | |
98 | struct page *page; | |
99 | struct address_space *mapping; | |
100 | int status; | |
101 | ||
102 | mapping = VFS_I(ip)->i_mapping; | |
103 | do { | |
104 | unsigned offset, bytes; | |
105 | void *fsdata; | |
106 | ||
107 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
108 | bytes = PAGE_CACHE_SIZE - offset; | |
109 | if (bytes > count) | |
110 | bytes = count; | |
111 | ||
112 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
113 | AOP_FLAG_UNINTERRUPTIBLE, | |
114 | &page, &fsdata); | |
115 | if (status) | |
116 | break; | |
117 | ||
118 | zero_user(page, offset, bytes); | |
119 | ||
120 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
121 | page, fsdata); | |
122 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
123 | pos += bytes; | |
124 | count -= bytes; | |
125 | status = 0; | |
126 | } while (count); | |
127 | ||
128 | return (-status); | |
129 | } | |
130 | ||
8add71ca CH |
131 | int |
132 | xfs_update_prealloc_flags( | |
133 | struct xfs_inode *ip, | |
134 | enum xfs_prealloc_flags flags) | |
135 | { | |
136 | struct xfs_trans *tp; | |
137 | int error; | |
138 | ||
139 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); | |
140 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
141 | if (error) { | |
142 | xfs_trans_cancel(tp, 0); | |
143 | return error; | |
144 | } | |
145 | ||
146 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
147 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
148 | ||
149 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
150 | ip->i_d.di_mode &= ~S_ISUID; | |
151 | if (ip->i_d.di_mode & S_IXGRP) | |
152 | ip->i_d.di_mode &= ~S_ISGID; | |
153 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
154 | } | |
155 | ||
156 | if (flags & XFS_PREALLOC_SET) | |
157 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
158 | if (flags & XFS_PREALLOC_CLEAR) | |
159 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
160 | ||
161 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
162 | if (flags & XFS_PREALLOC_SYNC) | |
163 | xfs_trans_set_sync(tp); | |
164 | return xfs_trans_commit(tp, 0); | |
165 | } | |
166 | ||
1da2f2db CH |
167 | /* |
168 | * Fsync operations on directories are much simpler than on regular files, | |
169 | * as there is no file data to flush, and thus also no need for explicit | |
170 | * cache flush operations, and there are no non-transaction metadata updates | |
171 | * on directories either. | |
172 | */ | |
173 | STATIC int | |
174 | xfs_dir_fsync( | |
175 | struct file *file, | |
176 | loff_t start, | |
177 | loff_t end, | |
178 | int datasync) | |
179 | { | |
180 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
181 | struct xfs_mount *mp = ip->i_mount; | |
182 | xfs_lsn_t lsn = 0; | |
183 | ||
184 | trace_xfs_dir_fsync(ip); | |
185 | ||
186 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
187 | if (xfs_ipincount(ip)) | |
188 | lsn = ip->i_itemp->ili_last_lsn; | |
189 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
190 | ||
191 | if (!lsn) | |
192 | return 0; | |
2451337d | 193 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
194 | } |
195 | ||
fd3200be CH |
196 | STATIC int |
197 | xfs_file_fsync( | |
198 | struct file *file, | |
02c24a82 JB |
199 | loff_t start, |
200 | loff_t end, | |
fd3200be CH |
201 | int datasync) |
202 | { | |
7ea80859 CH |
203 | struct inode *inode = file->f_mapping->host; |
204 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 205 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
206 | int error = 0; |
207 | int log_flushed = 0; | |
b1037058 | 208 | xfs_lsn_t lsn = 0; |
fd3200be | 209 | |
cca28fb8 | 210 | trace_xfs_file_fsync(ip); |
fd3200be | 211 | |
02c24a82 JB |
212 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
213 | if (error) | |
214 | return error; | |
215 | ||
a27a263b | 216 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 217 | return -EIO; |
fd3200be CH |
218 | |
219 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
220 | ||
a27a263b CH |
221 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
222 | /* | |
223 | * If we have an RT and/or log subvolume we need to make sure | |
224 | * to flush the write cache the device used for file data | |
225 | * first. This is to ensure newly written file data make | |
226 | * it to disk before logging the new inode size in case of | |
227 | * an extending write. | |
228 | */ | |
229 | if (XFS_IS_REALTIME_INODE(ip)) | |
230 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
231 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
232 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
233 | } | |
234 | ||
fd3200be | 235 | /* |
8a9c9980 CH |
236 | * All metadata updates are logged, which means that we just have |
237 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
238 | */ |
239 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
240 | if (xfs_ipincount(ip)) { |
241 | if (!datasync || | |
242 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
243 | lsn = ip->i_itemp->ili_last_lsn; | |
244 | } | |
8a9c9980 | 245 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 246 | |
8a9c9980 | 247 | if (lsn) |
b1037058 CH |
248 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
249 | ||
a27a263b CH |
250 | /* |
251 | * If we only have a single device, and the log force about was | |
252 | * a no-op we might have to flush the data device cache here. | |
253 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
254 | * an already allocated file and thus do not have any metadata to | |
255 | * commit. | |
256 | */ | |
257 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
258 | mp->m_logdev_targp == mp->m_ddev_targp && | |
259 | !XFS_IS_REALTIME_INODE(ip) && | |
260 | !log_flushed) | |
261 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 262 | |
2451337d | 263 | return error; |
fd3200be CH |
264 | } |
265 | ||
00258e36 | 266 | STATIC ssize_t |
b4f5d2c6 | 267 | xfs_file_read_iter( |
dda35b8f | 268 | struct kiocb *iocb, |
b4f5d2c6 | 269 | struct iov_iter *to) |
dda35b8f CH |
270 | { |
271 | struct file *file = iocb->ki_filp; | |
272 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
273 | struct xfs_inode *ip = XFS_I(inode); |
274 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 275 | size_t size = iov_iter_count(to); |
dda35b8f | 276 | ssize_t ret = 0; |
00258e36 | 277 | int ioflags = 0; |
dda35b8f | 278 | xfs_fsize_t n; |
b4f5d2c6 | 279 | loff_t pos = iocb->ki_pos; |
dda35b8f | 280 | |
dda35b8f CH |
281 | XFS_STATS_INC(xs_read_calls); |
282 | ||
00258e36 | 283 | if (unlikely(file->f_flags & O_DIRECT)) |
b92cc59f | 284 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 285 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 286 | ioflags |= XFS_IO_INVIS; |
00258e36 | 287 | |
b92cc59f | 288 | if (unlikely(ioflags & XFS_IO_ISDIRECT)) { |
dda35b8f CH |
289 | xfs_buftarg_t *target = |
290 | XFS_IS_REALTIME_INODE(ip) ? | |
291 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
292 | /* DIO must be aligned to device logical sector size */ |
293 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 294 | if (pos == i_size_read(inode)) |
00258e36 | 295 | return 0; |
b474c7ae | 296 | return -EINVAL; |
dda35b8f CH |
297 | } |
298 | } | |
299 | ||
fb595814 | 300 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 301 | if (n <= 0 || size == 0) |
dda35b8f CH |
302 | return 0; |
303 | ||
304 | if (n < size) | |
305 | size = n; | |
306 | ||
307 | if (XFS_FORCED_SHUTDOWN(mp)) | |
308 | return -EIO; | |
309 | ||
0c38a251 DC |
310 | /* |
311 | * Locking is a bit tricky here. If we take an exclusive lock | |
312 | * for direct IO, we effectively serialise all new concurrent | |
313 | * read IO to this file and block it behind IO that is currently in | |
314 | * progress because IO in progress holds the IO lock shared. We only | |
315 | * need to hold the lock exclusive to blow away the page cache, so | |
316 | * only take lock exclusively if the page cache needs invalidation. | |
317 | * This allows the normal direct IO case of no page cache pages to | |
318 | * proceeed concurrently without serialisation. | |
319 | */ | |
320 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 321 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 322 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
323 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
324 | ||
00258e36 | 325 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 326 | ret = filemap_write_and_wait_range( |
fb595814 | 327 | VFS_I(ip)->i_mapping, |
7d4ea3ce | 328 | pos, pos + size - 1); |
487f84f3 DC |
329 | if (ret) { |
330 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
331 | return ret; | |
332 | } | |
85e584da CM |
333 | |
334 | /* | |
335 | * Invalidate whole pages. This can return an error if | |
336 | * we fail to invalidate a page, but this should never | |
337 | * happen on XFS. Warn if it does fail. | |
338 | */ | |
339 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
340 | pos >> PAGE_CACHE_SHIFT, |
341 | (pos + size - 1) >> PAGE_CACHE_SHIFT); | |
85e584da CM |
342 | WARN_ON_ONCE(ret); |
343 | ret = 0; | |
00258e36 | 344 | } |
487f84f3 | 345 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 346 | } |
dda35b8f | 347 | |
fb595814 | 348 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 349 | |
b4f5d2c6 | 350 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
351 | if (ret > 0) |
352 | XFS_STATS_ADD(xs_read_bytes, ret); | |
353 | ||
487f84f3 | 354 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
355 | return ret; |
356 | } | |
357 | ||
00258e36 CH |
358 | STATIC ssize_t |
359 | xfs_file_splice_read( | |
dda35b8f CH |
360 | struct file *infilp, |
361 | loff_t *ppos, | |
362 | struct pipe_inode_info *pipe, | |
363 | size_t count, | |
00258e36 | 364 | unsigned int flags) |
dda35b8f | 365 | { |
00258e36 | 366 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 367 | int ioflags = 0; |
dda35b8f CH |
368 | ssize_t ret; |
369 | ||
370 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
371 | |
372 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 373 | ioflags |= XFS_IO_INVIS; |
00258e36 | 374 | |
dda35b8f CH |
375 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
376 | return -EIO; | |
377 | ||
487f84f3 | 378 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 379 | |
dda35b8f CH |
380 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
381 | ||
382 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
383 | if (ret > 0) | |
384 | XFS_STATS_ADD(xs_read_bytes, ret); | |
385 | ||
487f84f3 | 386 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
387 | return ret; |
388 | } | |
389 | ||
dda35b8f | 390 | /* |
193aec10 CH |
391 | * This routine is called to handle zeroing any space in the last block of the |
392 | * file that is beyond the EOF. We do this since the size is being increased | |
393 | * without writing anything to that block and we don't want to read the | |
394 | * garbage on the disk. | |
dda35b8f CH |
395 | */ |
396 | STATIC int /* error (positive) */ | |
397 | xfs_zero_last_block( | |
193aec10 CH |
398 | struct xfs_inode *ip, |
399 | xfs_fsize_t offset, | |
400 | xfs_fsize_t isize) | |
dda35b8f | 401 | { |
193aec10 CH |
402 | struct xfs_mount *mp = ip->i_mount; |
403 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
404 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
405 | int zero_len; | |
406 | int nimaps = 1; | |
407 | int error = 0; | |
408 | struct xfs_bmbt_irec imap; | |
dda35b8f | 409 | |
193aec10 | 410 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 411 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 412 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 413 | if (error) |
dda35b8f | 414 | return error; |
193aec10 | 415 | |
dda35b8f | 416 | ASSERT(nimaps > 0); |
193aec10 | 417 | |
dda35b8f CH |
418 | /* |
419 | * If the block underlying isize is just a hole, then there | |
420 | * is nothing to zero. | |
421 | */ | |
193aec10 | 422 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 423 | return 0; |
dda35b8f CH |
424 | |
425 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
426 | if (isize + zero_len > offset) | |
427 | zero_len = offset - isize; | |
193aec10 | 428 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
429 | } |
430 | ||
431 | /* | |
193aec10 CH |
432 | * Zero any on disk space between the current EOF and the new, larger EOF. |
433 | * | |
434 | * This handles the normal case of zeroing the remainder of the last block in | |
435 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
436 | * file. This second case only happens with fixed size extents and when the | |
437 | * system crashes before the inode size was updated but after blocks were | |
438 | * allocated. | |
439 | * | |
440 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 441 | */ |
dda35b8f CH |
442 | int /* error (positive) */ |
443 | xfs_zero_eof( | |
193aec10 CH |
444 | struct xfs_inode *ip, |
445 | xfs_off_t offset, /* starting I/O offset */ | |
446 | xfs_fsize_t isize) /* current inode size */ | |
dda35b8f | 447 | { |
193aec10 CH |
448 | struct xfs_mount *mp = ip->i_mount; |
449 | xfs_fileoff_t start_zero_fsb; | |
450 | xfs_fileoff_t end_zero_fsb; | |
451 | xfs_fileoff_t zero_count_fsb; | |
452 | xfs_fileoff_t last_fsb; | |
453 | xfs_fileoff_t zero_off; | |
454 | xfs_fsize_t zero_len; | |
455 | int nimaps; | |
456 | int error = 0; | |
457 | struct xfs_bmbt_irec imap; | |
458 | ||
459 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
460 | ASSERT(offset > isize); |
461 | ||
462 | /* | |
463 | * First handle zeroing the block on which isize resides. | |
193aec10 | 464 | * |
dda35b8f CH |
465 | * We only zero a part of that block so it is handled specially. |
466 | */ | |
193aec10 CH |
467 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
468 | error = xfs_zero_last_block(ip, offset, isize); | |
469 | if (error) | |
470 | return error; | |
dda35b8f CH |
471 | } |
472 | ||
473 | /* | |
193aec10 CH |
474 | * Calculate the range between the new size and the old where blocks |
475 | * needing to be zeroed may exist. | |
476 | * | |
477 | * To get the block where the last byte in the file currently resides, | |
478 | * we need to subtract one from the size and truncate back to a block | |
479 | * boundary. We subtract 1 in case the size is exactly on a block | |
480 | * boundary. | |
dda35b8f CH |
481 | */ |
482 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
483 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
484 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
485 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
486 | if (last_fsb == end_zero_fsb) { | |
487 | /* | |
488 | * The size was only incremented on its last block. | |
489 | * We took care of that above, so just return. | |
490 | */ | |
491 | return 0; | |
492 | } | |
493 | ||
494 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
495 | while (start_zero_fsb <= end_zero_fsb) { | |
496 | nimaps = 1; | |
497 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
498 | |
499 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
500 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
501 | &imap, &nimaps, 0); | |
193aec10 CH |
502 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
503 | if (error) | |
dda35b8f | 504 | return error; |
193aec10 | 505 | |
dda35b8f CH |
506 | ASSERT(nimaps > 0); |
507 | ||
508 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
509 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
510 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
511 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
512 | continue; | |
513 | } | |
514 | ||
515 | /* | |
516 | * There are blocks we need to zero. | |
dda35b8f | 517 | */ |
dda35b8f CH |
518 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
519 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
520 | ||
521 | if ((zero_off + zero_len) > offset) | |
522 | zero_len = offset - zero_off; | |
523 | ||
524 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
525 | if (error) |
526 | return error; | |
dda35b8f CH |
527 | |
528 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
529 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
530 | } |
531 | ||
532 | return 0; | |
dda35b8f CH |
533 | } |
534 | ||
4d8d1581 DC |
535 | /* |
536 | * Common pre-write limit and setup checks. | |
537 | * | |
5bf1f262 CH |
538 | * Called with the iolocked held either shared and exclusive according to |
539 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
540 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
541 | */ |
542 | STATIC ssize_t | |
543 | xfs_file_aio_write_checks( | |
544 | struct file *file, | |
545 | loff_t *pos, | |
546 | size_t *count, | |
547 | int *iolock) | |
548 | { | |
549 | struct inode *inode = file->f_mapping->host; | |
550 | struct xfs_inode *ip = XFS_I(inode); | |
4d8d1581 DC |
551 | int error = 0; |
552 | ||
7271d243 | 553 | restart: |
4d8d1581 | 554 | error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); |
467f7899 | 555 | if (error) |
4d8d1581 | 556 | return error; |
4d8d1581 | 557 | |
781355c6 CH |
558 | error = xfs_break_layouts(inode, iolock); |
559 | if (error) | |
560 | return error; | |
561 | ||
4d8d1581 DC |
562 | /* |
563 | * If the offset is beyond the size of the file, we need to zero any | |
564 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 565 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
566 | * iolock shared, we need to update it to exclusive which implies |
567 | * having to redo all checks before. | |
b9d59846 DC |
568 | * |
569 | * We need to serialise against EOF updates that occur in IO | |
570 | * completions here. We want to make sure that nobody is changing the | |
571 | * size while we do this check until we have placed an IO barrier (i.e. | |
572 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
573 | * The spinlock effectively forms a memory barrier once we have the | |
574 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
575 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 576 | */ |
b9d59846 | 577 | spin_lock(&ip->i_flags_lock); |
2813d682 | 578 | if (*pos > i_size_read(inode)) { |
b9d59846 | 579 | spin_unlock(&ip->i_flags_lock); |
7271d243 | 580 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 581 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 582 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 583 | xfs_rw_ilock(ip, *iolock); |
7271d243 DC |
584 | goto restart; |
585 | } | |
2451337d | 586 | error = xfs_zero_eof(ip, *pos, i_size_read(inode)); |
467f7899 CH |
587 | if (error) |
588 | return error; | |
b9d59846 DC |
589 | } else |
590 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 591 | |
8a9c9980 CH |
592 | /* |
593 | * Updating the timestamps will grab the ilock again from | |
594 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
595 | * lock above. Eventually we should look into a way to avoid | |
596 | * the pointless lock roundtrip. | |
597 | */ | |
c3b2da31 JB |
598 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
599 | error = file_update_time(file); | |
600 | if (error) | |
601 | return error; | |
602 | } | |
8a9c9980 | 603 | |
4d8d1581 DC |
604 | /* |
605 | * If we're writing the file then make sure to clear the setuid and | |
606 | * setgid bits if the process is not being run by root. This keeps | |
607 | * people from modifying setuid and setgid binaries. | |
608 | */ | |
609 | return file_remove_suid(file); | |
4d8d1581 DC |
610 | } |
611 | ||
f0d26e86 DC |
612 | /* |
613 | * xfs_file_dio_aio_write - handle direct IO writes | |
614 | * | |
615 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 616 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
617 | * follow locking changes and looping. |
618 | * | |
eda77982 DC |
619 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
620 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
621 | * pages are flushed out. | |
622 | * | |
623 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
624 | * allowing them to be done in parallel with reads and other direct IO writes. | |
625 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
626 | * needs to do sub-block zeroing and that requires serialisation against other | |
627 | * direct IOs to the same block. In this case we need to serialise the | |
628 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
629 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
630 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
631 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 632 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 633 | * |
f0d26e86 DC |
634 | * Returns with locks held indicated by @iolock and errors indicated by |
635 | * negative return values. | |
636 | */ | |
637 | STATIC ssize_t | |
638 | xfs_file_dio_aio_write( | |
639 | struct kiocb *iocb, | |
b3188919 | 640 | struct iov_iter *from) |
f0d26e86 DC |
641 | { |
642 | struct file *file = iocb->ki_filp; | |
643 | struct address_space *mapping = file->f_mapping; | |
644 | struct inode *inode = mapping->host; | |
645 | struct xfs_inode *ip = XFS_I(inode); | |
646 | struct xfs_mount *mp = ip->i_mount; | |
647 | ssize_t ret = 0; | |
eda77982 | 648 | int unaligned_io = 0; |
d0606464 | 649 | int iolock; |
b3188919 AV |
650 | size_t count = iov_iter_count(from); |
651 | loff_t pos = iocb->ki_pos; | |
f0d26e86 DC |
652 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
653 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
654 | ||
7c71ee78 ES |
655 | /* DIO must be aligned to device logical sector size */ |
656 | if ((pos | count) & target->bt_logical_sectormask) | |
b474c7ae | 657 | return -EINVAL; |
f0d26e86 | 658 | |
7c71ee78 | 659 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
660 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
661 | unaligned_io = 1; | |
662 | ||
7271d243 DC |
663 | /* |
664 | * We don't need to take an exclusive lock unless there page cache needs | |
665 | * to be invalidated or unaligned IO is being executed. We don't need to | |
666 | * consider the EOF extension case here because | |
667 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
668 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
669 | */ | |
670 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 671 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 672 | else |
d0606464 CH |
673 | iolock = XFS_IOLOCK_SHARED; |
674 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
675 | |
676 | /* | |
677 | * Recheck if there are cached pages that need invalidate after we got | |
678 | * the iolock to protect against other threads adding new pages while | |
679 | * we were waiting for the iolock. | |
680 | */ | |
d0606464 CH |
681 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
682 | xfs_rw_iunlock(ip, iolock); | |
683 | iolock = XFS_IOLOCK_EXCL; | |
684 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 685 | } |
f0d26e86 | 686 | |
d0606464 | 687 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 688 | if (ret) |
d0606464 | 689 | goto out; |
b3188919 | 690 | iov_iter_truncate(from, count); |
f0d26e86 DC |
691 | |
692 | if (mapping->nrpages) { | |
07d5035a | 693 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
7d4ea3ce | 694 | pos, pos + count - 1); |
f0d26e86 | 695 | if (ret) |
d0606464 | 696 | goto out; |
834ffca6 DC |
697 | /* |
698 | * Invalidate whole pages. This can return an error if | |
699 | * we fail to invalidate a page, but this should never | |
700 | * happen on XFS. Warn if it does fail. | |
701 | */ | |
702 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
703 | pos >> PAGE_CACHE_SHIFT, |
704 | (pos + count - 1) >> PAGE_CACHE_SHIFT); | |
834ffca6 DC |
705 | WARN_ON_ONCE(ret); |
706 | ret = 0; | |
f0d26e86 DC |
707 | } |
708 | ||
eda77982 DC |
709 | /* |
710 | * If we are doing unaligned IO, wait for all other IO to drain, | |
711 | * otherwise demote the lock if we had to flush cached pages | |
712 | */ | |
713 | if (unaligned_io) | |
4a06fd26 | 714 | inode_dio_wait(inode); |
d0606464 | 715 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 716 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 717 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
718 | } |
719 | ||
720 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
b3188919 | 721 | ret = generic_file_direct_write(iocb, from, pos); |
f0d26e86 | 722 | |
d0606464 CH |
723 | out: |
724 | xfs_rw_iunlock(ip, iolock); | |
725 | ||
f0d26e86 DC |
726 | /* No fallback to buffered IO on errors for XFS. */ |
727 | ASSERT(ret < 0 || ret == count); | |
728 | return ret; | |
729 | } | |
730 | ||
00258e36 | 731 | STATIC ssize_t |
637bbc75 | 732 | xfs_file_buffered_aio_write( |
dda35b8f | 733 | struct kiocb *iocb, |
b3188919 | 734 | struct iov_iter *from) |
dda35b8f CH |
735 | { |
736 | struct file *file = iocb->ki_filp; | |
737 | struct address_space *mapping = file->f_mapping; | |
738 | struct inode *inode = mapping->host; | |
00258e36 | 739 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
740 | ssize_t ret; |
741 | int enospc = 0; | |
d0606464 | 742 | int iolock = XFS_IOLOCK_EXCL; |
b3188919 AV |
743 | loff_t pos = iocb->ki_pos; |
744 | size_t count = iov_iter_count(from); | |
dda35b8f | 745 | |
d0606464 | 746 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 747 | |
d0606464 | 748 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 749 | if (ret) |
d0606464 | 750 | goto out; |
dda35b8f | 751 | |
b3188919 | 752 | iov_iter_truncate(from, count); |
dda35b8f | 753 | /* We can write back this queue in page reclaim */ |
de1414a6 | 754 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 755 | |
dda35b8f | 756 | write_retry: |
637bbc75 | 757 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); |
b3188919 | 758 | ret = generic_perform_write(file, from, pos); |
0a64bc2c AV |
759 | if (likely(ret >= 0)) |
760 | iocb->ki_pos = pos + ret; | |
dc06f398 | 761 | |
637bbc75 | 762 | /* |
dc06f398 BF |
763 | * If we hit a space limit, try to free up some lingering preallocated |
764 | * space before returning an error. In the case of ENOSPC, first try to | |
765 | * write back all dirty inodes to free up some of the excess reserved | |
766 | * metadata space. This reduces the chances that the eofblocks scan | |
767 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
768 | * also behaves as a filter to prevent too many eofblocks scans from | |
769 | * running at the same time. | |
637bbc75 | 770 | */ |
dc06f398 BF |
771 | if (ret == -EDQUOT && !enospc) { |
772 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
773 | if (enospc) | |
774 | goto write_retry; | |
775 | } else if (ret == -ENOSPC && !enospc) { | |
776 | struct xfs_eofblocks eofb = {0}; | |
777 | ||
637bbc75 | 778 | enospc = 1; |
9aa05000 | 779 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
780 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
781 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
782 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 783 | goto write_retry; |
dda35b8f | 784 | } |
d0606464 | 785 | |
dda35b8f | 786 | current->backing_dev_info = NULL; |
d0606464 CH |
787 | out: |
788 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
789 | return ret; |
790 | } | |
791 | ||
792 | STATIC ssize_t | |
bf97f3bc | 793 | xfs_file_write_iter( |
637bbc75 | 794 | struct kiocb *iocb, |
bf97f3bc | 795 | struct iov_iter *from) |
637bbc75 DC |
796 | { |
797 | struct file *file = iocb->ki_filp; | |
798 | struct address_space *mapping = file->f_mapping; | |
799 | struct inode *inode = mapping->host; | |
800 | struct xfs_inode *ip = XFS_I(inode); | |
801 | ssize_t ret; | |
bf97f3bc | 802 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
803 | |
804 | XFS_STATS_INC(xs_write_calls); | |
805 | ||
637bbc75 DC |
806 | if (ocount == 0) |
807 | return 0; | |
808 | ||
bf97f3bc AV |
809 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
810 | return -EIO; | |
637bbc75 DC |
811 | |
812 | if (unlikely(file->f_flags & O_DIRECT)) | |
bf97f3bc | 813 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 814 | else |
bf97f3bc | 815 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 816 | |
d0606464 CH |
817 | if (ret > 0) { |
818 | ssize_t err; | |
dda35b8f | 819 | |
d0606464 | 820 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 821 | |
d0606464 | 822 | /* Handle various SYNC-type writes */ |
d311d79d | 823 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
824 | if (err < 0) |
825 | ret = err; | |
dda35b8f | 826 | } |
a363f0c2 | 827 | return ret; |
dda35b8f CH |
828 | } |
829 | ||
2fe17c10 CH |
830 | STATIC long |
831 | xfs_file_fallocate( | |
83aee9e4 CH |
832 | struct file *file, |
833 | int mode, | |
834 | loff_t offset, | |
835 | loff_t len) | |
2fe17c10 | 836 | { |
83aee9e4 CH |
837 | struct inode *inode = file_inode(file); |
838 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 839 | long error; |
8add71ca | 840 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 841 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 842 | loff_t new_size = 0; |
2fe17c10 | 843 | |
83aee9e4 CH |
844 | if (!S_ISREG(inode->i_mode)) |
845 | return -EINVAL; | |
e1d8fb88 | 846 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
376ba313 | 847 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
2fe17c10 CH |
848 | return -EOPNOTSUPP; |
849 | ||
781355c6 CH |
850 | xfs_ilock(ip, iolock); |
851 | error = xfs_break_layouts(inode, &iolock); | |
852 | if (error) | |
853 | goto out_unlock; | |
854 | ||
83aee9e4 CH |
855 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
856 | error = xfs_free_file_space(ip, offset, len); | |
857 | if (error) | |
858 | goto out_unlock; | |
e1d8fb88 NJ |
859 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
860 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
861 | ||
862 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 863 | error = -EINVAL; |
e1d8fb88 NJ |
864 | goto out_unlock; |
865 | } | |
866 | ||
23fffa92 LC |
867 | /* |
868 | * There is no need to overlap collapse range with EOF, | |
869 | * in which case it is effectively a truncate operation | |
870 | */ | |
871 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 872 | error = -EINVAL; |
23fffa92 LC |
873 | goto out_unlock; |
874 | } | |
875 | ||
e1d8fb88 NJ |
876 | new_size = i_size_read(inode) - len; |
877 | ||
878 | error = xfs_collapse_file_space(ip, offset, len); | |
879 | if (error) | |
880 | goto out_unlock; | |
83aee9e4 | 881 | } else { |
8add71ca CH |
882 | flags |= XFS_PREALLOC_SET; |
883 | ||
83aee9e4 CH |
884 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
885 | offset + len > i_size_read(inode)) { | |
886 | new_size = offset + len; | |
2451337d | 887 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
888 | if (error) |
889 | goto out_unlock; | |
890 | } | |
2fe17c10 | 891 | |
376ba313 LC |
892 | if (mode & FALLOC_FL_ZERO_RANGE) |
893 | error = xfs_zero_file_space(ip, offset, len); | |
894 | else | |
895 | error = xfs_alloc_file_space(ip, offset, len, | |
896 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
897 | if (error) |
898 | goto out_unlock; | |
899 | } | |
900 | ||
83aee9e4 | 901 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
902 | flags |= XFS_PREALLOC_SYNC; |
903 | ||
904 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
905 | if (error) |
906 | goto out_unlock; | |
907 | ||
908 | /* Change file size if needed */ | |
909 | if (new_size) { | |
910 | struct iattr iattr; | |
911 | ||
912 | iattr.ia_valid = ATTR_SIZE; | |
913 | iattr.ia_size = new_size; | |
83aee9e4 | 914 | error = xfs_setattr_size(ip, &iattr); |
2fe17c10 CH |
915 | } |
916 | ||
917 | out_unlock: | |
781355c6 | 918 | xfs_iunlock(ip, iolock); |
2451337d | 919 | return error; |
2fe17c10 CH |
920 | } |
921 | ||
922 | ||
1da177e4 | 923 | STATIC int |
3562fd45 | 924 | xfs_file_open( |
1da177e4 | 925 | struct inode *inode, |
f999a5bf | 926 | struct file *file) |
1da177e4 | 927 | { |
f999a5bf | 928 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 929 | return -EFBIG; |
f999a5bf CH |
930 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
931 | return -EIO; | |
932 | return 0; | |
933 | } | |
934 | ||
935 | STATIC int | |
936 | xfs_dir_open( | |
937 | struct inode *inode, | |
938 | struct file *file) | |
939 | { | |
940 | struct xfs_inode *ip = XFS_I(inode); | |
941 | int mode; | |
942 | int error; | |
943 | ||
944 | error = xfs_file_open(inode, file); | |
945 | if (error) | |
946 | return error; | |
947 | ||
948 | /* | |
949 | * If there are any blocks, read-ahead block 0 as we're almost | |
950 | * certain to have the next operation be a read there. | |
951 | */ | |
309ecac8 | 952 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 953 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 954 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
955 | xfs_iunlock(ip, mode); |
956 | return 0; | |
1da177e4 LT |
957 | } |
958 | ||
1da177e4 | 959 | STATIC int |
3562fd45 | 960 | xfs_file_release( |
1da177e4 LT |
961 | struct inode *inode, |
962 | struct file *filp) | |
963 | { | |
2451337d | 964 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
965 | } |
966 | ||
1da177e4 | 967 | STATIC int |
3562fd45 | 968 | xfs_file_readdir( |
b8227554 AV |
969 | struct file *file, |
970 | struct dir_context *ctx) | |
1da177e4 | 971 | { |
b8227554 | 972 | struct inode *inode = file_inode(file); |
739bfb2a | 973 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
974 | size_t bufsize; |
975 | ||
976 | /* | |
977 | * The Linux API doesn't pass down the total size of the buffer | |
978 | * we read into down to the filesystem. With the filldir concept | |
979 | * it's not needed for correct information, but the XFS dir2 leaf | |
980 | * code wants an estimate of the buffer size to calculate it's | |
981 | * readahead window and size the buffers used for mapping to | |
982 | * physical blocks. | |
983 | * | |
984 | * Try to give it an estimate that's good enough, maybe at some | |
985 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 986 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 987 | */ |
a9cc799e | 988 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 989 | |
8300475e | 990 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
991 | } |
992 | ||
1da177e4 | 993 | STATIC int |
3562fd45 | 994 | xfs_file_mmap( |
1da177e4 LT |
995 | struct file *filp, |
996 | struct vm_area_struct *vma) | |
997 | { | |
3562fd45 | 998 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 999 | |
fbc1462b | 1000 | file_accessed(filp); |
1da177e4 LT |
1001 | return 0; |
1002 | } | |
1003 | ||
4f57dbc6 DC |
1004 | /* |
1005 | * mmap()d file has taken write protection fault and is being made | |
1006 | * writable. We can set the page state up correctly for a writable | |
1007 | * page, which means we can do correct delalloc accounting (ENOSPC | |
1008 | * checking!) and unwritten extent mapping. | |
1009 | */ | |
1010 | STATIC int | |
1011 | xfs_vm_page_mkwrite( | |
1012 | struct vm_area_struct *vma, | |
c2ec175c | 1013 | struct vm_fault *vmf) |
4f57dbc6 | 1014 | { |
c2ec175c | 1015 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
1016 | } |
1017 | ||
d126d43f JL |
1018 | /* |
1019 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1020 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1021 | */ |
1022 | enum { | |
1023 | HOLE_OFF = 0, | |
1024 | DATA_OFF, | |
1025 | }; | |
1026 | ||
1027 | /* | |
1028 | * Lookup the desired type of offset from the given page. | |
1029 | * | |
1030 | * On success, return true and the offset argument will point to the | |
1031 | * start of the region that was found. Otherwise this function will | |
1032 | * return false and keep the offset argument unchanged. | |
1033 | */ | |
1034 | STATIC bool | |
1035 | xfs_lookup_buffer_offset( | |
1036 | struct page *page, | |
1037 | loff_t *offset, | |
1038 | unsigned int type) | |
1039 | { | |
1040 | loff_t lastoff = page_offset(page); | |
1041 | bool found = false; | |
1042 | struct buffer_head *bh, *head; | |
1043 | ||
1044 | bh = head = page_buffers(page); | |
1045 | do { | |
1046 | /* | |
1047 | * Unwritten extents that have data in the page | |
1048 | * cache covering them can be identified by the | |
1049 | * BH_Unwritten state flag. Pages with multiple | |
1050 | * buffers might have a mix of holes, data and | |
1051 | * unwritten extents - any buffer with valid | |
1052 | * data in it should have BH_Uptodate flag set | |
1053 | * on it. | |
1054 | */ | |
1055 | if (buffer_unwritten(bh) || | |
1056 | buffer_uptodate(bh)) { | |
1057 | if (type == DATA_OFF) | |
1058 | found = true; | |
1059 | } else { | |
1060 | if (type == HOLE_OFF) | |
1061 | found = true; | |
1062 | } | |
1063 | ||
1064 | if (found) { | |
1065 | *offset = lastoff; | |
1066 | break; | |
1067 | } | |
1068 | lastoff += bh->b_size; | |
1069 | } while ((bh = bh->b_this_page) != head); | |
1070 | ||
1071 | return found; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * This routine is called to find out and return a data or hole offset | |
1076 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1077 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1078 | * |
1079 | * The argument offset is used to tell where we start to search from the | |
1080 | * page cache. Map is used to figure out the end points of the range to | |
1081 | * lookup pages. | |
1082 | * | |
1083 | * Return true if the desired type of offset was found, and the argument | |
1084 | * offset is filled with that address. Otherwise, return false and keep | |
1085 | * offset unchanged. | |
1086 | */ | |
1087 | STATIC bool | |
1088 | xfs_find_get_desired_pgoff( | |
1089 | struct inode *inode, | |
1090 | struct xfs_bmbt_irec *map, | |
1091 | unsigned int type, | |
1092 | loff_t *offset) | |
1093 | { | |
1094 | struct xfs_inode *ip = XFS_I(inode); | |
1095 | struct xfs_mount *mp = ip->i_mount; | |
1096 | struct pagevec pvec; | |
1097 | pgoff_t index; | |
1098 | pgoff_t end; | |
1099 | loff_t endoff; | |
1100 | loff_t startoff = *offset; | |
1101 | loff_t lastoff = startoff; | |
1102 | bool found = false; | |
1103 | ||
1104 | pagevec_init(&pvec, 0); | |
1105 | ||
1106 | index = startoff >> PAGE_CACHE_SHIFT; | |
1107 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1108 | end = endoff >> PAGE_CACHE_SHIFT; | |
1109 | do { | |
1110 | int want; | |
1111 | unsigned nr_pages; | |
1112 | unsigned int i; | |
1113 | ||
1114 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1115 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1116 | want); | |
1117 | /* | |
1118 | * No page mapped into given range. If we are searching holes | |
1119 | * and if this is the first time we got into the loop, it means | |
1120 | * that the given offset is landed in a hole, return it. | |
1121 | * | |
1122 | * If we have already stepped through some block buffers to find | |
1123 | * holes but they all contains data. In this case, the last | |
1124 | * offset is already updated and pointed to the end of the last | |
1125 | * mapped page, if it does not reach the endpoint to search, | |
1126 | * that means there should be a hole between them. | |
1127 | */ | |
1128 | if (nr_pages == 0) { | |
1129 | /* Data search found nothing */ | |
1130 | if (type == DATA_OFF) | |
1131 | break; | |
1132 | ||
1133 | ASSERT(type == HOLE_OFF); | |
1134 | if (lastoff == startoff || lastoff < endoff) { | |
1135 | found = true; | |
1136 | *offset = lastoff; | |
1137 | } | |
1138 | break; | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * At lease we found one page. If this is the first time we | |
1143 | * step into the loop, and if the first page index offset is | |
1144 | * greater than the given search offset, a hole was found. | |
1145 | */ | |
1146 | if (type == HOLE_OFF && lastoff == startoff && | |
1147 | lastoff < page_offset(pvec.pages[0])) { | |
1148 | found = true; | |
1149 | break; | |
1150 | } | |
1151 | ||
1152 | for (i = 0; i < nr_pages; i++) { | |
1153 | struct page *page = pvec.pages[i]; | |
1154 | loff_t b_offset; | |
1155 | ||
1156 | /* | |
1157 | * At this point, the page may be truncated or | |
1158 | * invalidated (changing page->mapping to NULL), | |
1159 | * or even swizzled back from swapper_space to tmpfs | |
1160 | * file mapping. However, page->index will not change | |
1161 | * because we have a reference on the page. | |
1162 | * | |
1163 | * Searching done if the page index is out of range. | |
1164 | * If the current offset is not reaches the end of | |
1165 | * the specified search range, there should be a hole | |
1166 | * between them. | |
1167 | */ | |
1168 | if (page->index > end) { | |
1169 | if (type == HOLE_OFF && lastoff < endoff) { | |
1170 | *offset = lastoff; | |
1171 | found = true; | |
1172 | } | |
1173 | goto out; | |
1174 | } | |
1175 | ||
1176 | lock_page(page); | |
1177 | /* | |
1178 | * Page truncated or invalidated(page->mapping == NULL). | |
1179 | * We can freely skip it and proceed to check the next | |
1180 | * page. | |
1181 | */ | |
1182 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1183 | unlock_page(page); | |
1184 | continue; | |
1185 | } | |
1186 | ||
1187 | if (!page_has_buffers(page)) { | |
1188 | unlock_page(page); | |
1189 | continue; | |
1190 | } | |
1191 | ||
1192 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1193 | if (found) { | |
1194 | /* | |
1195 | * The found offset may be less than the start | |
1196 | * point to search if this is the first time to | |
1197 | * come here. | |
1198 | */ | |
1199 | *offset = max_t(loff_t, startoff, b_offset); | |
1200 | unlock_page(page); | |
1201 | goto out; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * We either searching data but nothing was found, or | |
1206 | * searching hole but found a data buffer. In either | |
1207 | * case, probably the next page contains the desired | |
1208 | * things, update the last offset to it so. | |
1209 | */ | |
1210 | lastoff = page_offset(page) + PAGE_SIZE; | |
1211 | unlock_page(page); | |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * The number of returned pages less than our desired, search | |
1216 | * done. In this case, nothing was found for searching data, | |
1217 | * but we found a hole behind the last offset. | |
1218 | */ | |
1219 | if (nr_pages < want) { | |
1220 | if (type == HOLE_OFF) { | |
1221 | *offset = lastoff; | |
1222 | found = true; | |
1223 | } | |
1224 | break; | |
1225 | } | |
1226 | ||
1227 | index = pvec.pages[i - 1]->index + 1; | |
1228 | pagevec_release(&pvec); | |
1229 | } while (index <= end); | |
1230 | ||
1231 | out: | |
1232 | pagevec_release(&pvec); | |
1233 | return found; | |
1234 | } | |
1235 | ||
3fe3e6b1 | 1236 | STATIC loff_t |
49c69591 | 1237 | xfs_seek_hole_data( |
3fe3e6b1 | 1238 | struct file *file, |
49c69591 ES |
1239 | loff_t start, |
1240 | int whence) | |
3fe3e6b1 JL |
1241 | { |
1242 | struct inode *inode = file->f_mapping->host; | |
1243 | struct xfs_inode *ip = XFS_I(inode); | |
1244 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1245 | loff_t uninitialized_var(offset); |
1246 | xfs_fsize_t isize; | |
1247 | xfs_fileoff_t fsbno; | |
1248 | xfs_filblks_t end; | |
1249 | uint lock; | |
1250 | int error; | |
1251 | ||
49c69591 ES |
1252 | if (XFS_FORCED_SHUTDOWN(mp)) |
1253 | return -EIO; | |
1254 | ||
309ecac8 | 1255 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1256 | |
1257 | isize = i_size_read(inode); | |
1258 | if (start >= isize) { | |
2451337d | 1259 | error = -ENXIO; |
3fe3e6b1 JL |
1260 | goto out_unlock; |
1261 | } | |
1262 | ||
3fe3e6b1 JL |
1263 | /* |
1264 | * Try to read extents from the first block indicated | |
1265 | * by fsbno to the end block of the file. | |
1266 | */ | |
52f1acc8 | 1267 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1268 | end = XFS_B_TO_FSB(mp, isize); |
49c69591 | 1269 | |
52f1acc8 JL |
1270 | for (;;) { |
1271 | struct xfs_bmbt_irec map[2]; | |
1272 | int nmap = 2; | |
1273 | unsigned int i; | |
3fe3e6b1 | 1274 | |
52f1acc8 JL |
1275 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1276 | XFS_BMAPI_ENTIRE); | |
1277 | if (error) | |
1278 | goto out_unlock; | |
3fe3e6b1 | 1279 | |
52f1acc8 JL |
1280 | /* No extents at given offset, must be beyond EOF */ |
1281 | if (nmap == 0) { | |
2451337d | 1282 | error = -ENXIO; |
52f1acc8 JL |
1283 | goto out_unlock; |
1284 | } | |
1285 | ||
1286 | for (i = 0; i < nmap; i++) { | |
1287 | offset = max_t(loff_t, start, | |
1288 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1289 | ||
49c69591 ES |
1290 | /* Landed in the hole we wanted? */ |
1291 | if (whence == SEEK_HOLE && | |
1292 | map[i].br_startblock == HOLESTARTBLOCK) | |
1293 | goto out; | |
1294 | ||
1295 | /* Landed in the data extent we wanted? */ | |
1296 | if (whence == SEEK_DATA && | |
1297 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1298 | (map[i].br_state == XFS_EXT_NORM && | |
1299 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1300 | goto out; |
1301 | ||
1302 | /* | |
49c69591 ES |
1303 | * Landed in an unwritten extent, try to search |
1304 | * for hole or data from page cache. | |
52f1acc8 JL |
1305 | */ |
1306 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1307 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1308 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1309 | &offset)) | |
52f1acc8 JL |
1310 | goto out; |
1311 | } | |
1312 | } | |
1313 | ||
1314 | /* | |
49c69591 ES |
1315 | * We only received one extent out of the two requested. This |
1316 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1317 | */ |
3fe3e6b1 | 1318 | if (nmap == 1) { |
49c69591 ES |
1319 | /* |
1320 | * If we were looking for a hole, set offset to | |
1321 | * the end of the file (i.e., there is an implicit | |
1322 | * hole at the end of any file). | |
1323 | */ | |
1324 | if (whence == SEEK_HOLE) { | |
1325 | offset = isize; | |
1326 | break; | |
1327 | } | |
1328 | /* | |
1329 | * If we were looking for data, it's nowhere to be found | |
1330 | */ | |
1331 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1332 | error = -ENXIO; |
3fe3e6b1 JL |
1333 | goto out_unlock; |
1334 | } | |
1335 | ||
52f1acc8 JL |
1336 | ASSERT(i > 1); |
1337 | ||
1338 | /* | |
1339 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1340 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1341 | */ |
1342 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1343 | start = XFS_FSB_TO_B(mp, fsbno); | |
1344 | if (start >= isize) { | |
49c69591 ES |
1345 | if (whence == SEEK_HOLE) { |
1346 | offset = isize; | |
1347 | break; | |
1348 | } | |
1349 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1350 | error = -ENXIO; |
52f1acc8 JL |
1351 | goto out_unlock; |
1352 | } | |
3fe3e6b1 JL |
1353 | } |
1354 | ||
b686d1f7 JL |
1355 | out: |
1356 | /* | |
49c69591 | 1357 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1358 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1359 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1360 | * situation in particular. |
1361 | */ | |
49c69591 ES |
1362 | if (whence == SEEK_HOLE) |
1363 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1364 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1365 | |
1366 | out_unlock: | |
01f4f327 | 1367 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1368 | |
1369 | if (error) | |
2451337d | 1370 | return error; |
3fe3e6b1 JL |
1371 | return offset; |
1372 | } | |
1373 | ||
1374 | STATIC loff_t | |
1375 | xfs_file_llseek( | |
1376 | struct file *file, | |
1377 | loff_t offset, | |
59f9c004 | 1378 | int whence) |
3fe3e6b1 | 1379 | { |
59f9c004 | 1380 | switch (whence) { |
3fe3e6b1 JL |
1381 | case SEEK_END: |
1382 | case SEEK_CUR: | |
1383 | case SEEK_SET: | |
59f9c004 | 1384 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1385 | case SEEK_HOLE: |
49c69591 | 1386 | case SEEK_DATA: |
59f9c004 | 1387 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1388 | default: |
1389 | return -EINVAL; | |
1390 | } | |
1391 | } | |
1392 | ||
4b6f5d20 | 1393 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1394 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1395 | .read = new_sync_read, |
bf97f3bc | 1396 | .write = new_sync_write, |
b4f5d2c6 | 1397 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1398 | .write_iter = xfs_file_write_iter, |
1b895840 | 1399 | .splice_read = xfs_file_splice_read, |
8d020765 | 1400 | .splice_write = iter_file_splice_write, |
3562fd45 | 1401 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1402 | #ifdef CONFIG_COMPAT |
3562fd45 | 1403 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1404 | #endif |
3562fd45 NS |
1405 | .mmap = xfs_file_mmap, |
1406 | .open = xfs_file_open, | |
1407 | .release = xfs_file_release, | |
1408 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1409 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1410 | }; |
1411 | ||
4b6f5d20 | 1412 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1413 | .open = xfs_dir_open, |
1da177e4 | 1414 | .read = generic_read_dir, |
b8227554 | 1415 | .iterate = xfs_file_readdir, |
59af1584 | 1416 | .llseek = generic_file_llseek, |
3562fd45 | 1417 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1418 | #ifdef CONFIG_COMPAT |
3562fd45 | 1419 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1420 | #endif |
1da2f2db | 1421 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1422 | }; |
1423 | ||
f0f37e2f | 1424 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1425 | .fault = filemap_fault, |
f1820361 | 1426 | .map_pages = filemap_map_pages, |
4f57dbc6 | 1427 | .page_mkwrite = xfs_vm_page_mkwrite, |
6fac0cb4 | 1428 | }; |