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" |
1da177e4 | 39 | |
a27bb332 | 40 | #include <linux/aio.h> |
1da177e4 | 41 | #include <linux/dcache.h> |
2fe17c10 | 42 | #include <linux/falloc.h> |
d126d43f | 43 | #include <linux/pagevec.h> |
1da177e4 | 44 | |
f0f37e2f | 45 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 46 | |
487f84f3 DC |
47 | /* |
48 | * Locking primitives for read and write IO paths to ensure we consistently use | |
49 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
50 | */ | |
51 | static inline void | |
52 | xfs_rw_ilock( | |
53 | struct xfs_inode *ip, | |
54 | int type) | |
55 | { | |
56 | if (type & XFS_IOLOCK_EXCL) | |
57 | mutex_lock(&VFS_I(ip)->i_mutex); | |
58 | xfs_ilock(ip, type); | |
59 | } | |
60 | ||
61 | static inline void | |
62 | xfs_rw_iunlock( | |
63 | struct xfs_inode *ip, | |
64 | int type) | |
65 | { | |
66 | xfs_iunlock(ip, type); | |
67 | if (type & XFS_IOLOCK_EXCL) | |
68 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
69 | } | |
70 | ||
71 | static inline void | |
72 | xfs_rw_ilock_demote( | |
73 | struct xfs_inode *ip, | |
74 | int type) | |
75 | { | |
76 | xfs_ilock_demote(ip, type); | |
77 | if (type & XFS_IOLOCK_EXCL) | |
78 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
79 | } | |
80 | ||
dda35b8f CH |
81 | /* |
82 | * xfs_iozero | |
83 | * | |
84 | * xfs_iozero clears the specified range of buffer supplied, | |
85 | * and marks all the affected blocks as valid and modified. If | |
86 | * an affected block is not allocated, it will be allocated. If | |
87 | * an affected block is not completely overwritten, and is not | |
88 | * valid before the operation, it will be read from disk before | |
89 | * being partially zeroed. | |
90 | */ | |
ef9d8733 | 91 | int |
dda35b8f CH |
92 | xfs_iozero( |
93 | struct xfs_inode *ip, /* inode */ | |
94 | loff_t pos, /* offset in file */ | |
95 | size_t count) /* size of data to zero */ | |
96 | { | |
97 | struct page *page; | |
98 | struct address_space *mapping; | |
99 | int status; | |
100 | ||
101 | mapping = VFS_I(ip)->i_mapping; | |
102 | do { | |
103 | unsigned offset, bytes; | |
104 | void *fsdata; | |
105 | ||
106 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
107 | bytes = PAGE_CACHE_SIZE - offset; | |
108 | if (bytes > count) | |
109 | bytes = count; | |
110 | ||
111 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
112 | AOP_FLAG_UNINTERRUPTIBLE, | |
113 | &page, &fsdata); | |
114 | if (status) | |
115 | break; | |
116 | ||
117 | zero_user(page, offset, bytes); | |
118 | ||
119 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
120 | page, fsdata); | |
121 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
122 | pos += bytes; | |
123 | count -= bytes; | |
124 | status = 0; | |
125 | } while (count); | |
126 | ||
127 | return (-status); | |
128 | } | |
129 | ||
1da2f2db CH |
130 | /* |
131 | * Fsync operations on directories are much simpler than on regular files, | |
132 | * as there is no file data to flush, and thus also no need for explicit | |
133 | * cache flush operations, and there are no non-transaction metadata updates | |
134 | * on directories either. | |
135 | */ | |
136 | STATIC int | |
137 | xfs_dir_fsync( | |
138 | struct file *file, | |
139 | loff_t start, | |
140 | loff_t end, | |
141 | int datasync) | |
142 | { | |
143 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
144 | struct xfs_mount *mp = ip->i_mount; | |
145 | xfs_lsn_t lsn = 0; | |
146 | ||
147 | trace_xfs_dir_fsync(ip); | |
148 | ||
149 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
150 | if (xfs_ipincount(ip)) | |
151 | lsn = ip->i_itemp->ili_last_lsn; | |
152 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
153 | ||
154 | if (!lsn) | |
155 | return 0; | |
2451337d | 156 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
157 | } |
158 | ||
fd3200be CH |
159 | STATIC int |
160 | xfs_file_fsync( | |
161 | struct file *file, | |
02c24a82 JB |
162 | loff_t start, |
163 | loff_t end, | |
fd3200be CH |
164 | int datasync) |
165 | { | |
7ea80859 CH |
166 | struct inode *inode = file->f_mapping->host; |
167 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 168 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
169 | int error = 0; |
170 | int log_flushed = 0; | |
b1037058 | 171 | xfs_lsn_t lsn = 0; |
fd3200be | 172 | |
cca28fb8 | 173 | trace_xfs_file_fsync(ip); |
fd3200be | 174 | |
02c24a82 JB |
175 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
176 | if (error) | |
177 | return error; | |
178 | ||
a27a263b | 179 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 180 | return -EIO; |
fd3200be CH |
181 | |
182 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
183 | ||
a27a263b CH |
184 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
185 | /* | |
186 | * If we have an RT and/or log subvolume we need to make sure | |
187 | * to flush the write cache the device used for file data | |
188 | * first. This is to ensure newly written file data make | |
189 | * it to disk before logging the new inode size in case of | |
190 | * an extending write. | |
191 | */ | |
192 | if (XFS_IS_REALTIME_INODE(ip)) | |
193 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
194 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
195 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
196 | } | |
197 | ||
fd3200be | 198 | /* |
8a9c9980 CH |
199 | * All metadata updates are logged, which means that we just have |
200 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
201 | */ |
202 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
203 | if (xfs_ipincount(ip)) { |
204 | if (!datasync || | |
205 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
206 | lsn = ip->i_itemp->ili_last_lsn; | |
207 | } | |
8a9c9980 | 208 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 209 | |
8a9c9980 | 210 | if (lsn) |
b1037058 CH |
211 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
212 | ||
a27a263b CH |
213 | /* |
214 | * If we only have a single device, and the log force about was | |
215 | * a no-op we might have to flush the data device cache here. | |
216 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
217 | * an already allocated file and thus do not have any metadata to | |
218 | * commit. | |
219 | */ | |
220 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
221 | mp->m_logdev_targp == mp->m_ddev_targp && | |
222 | !XFS_IS_REALTIME_INODE(ip) && | |
223 | !log_flushed) | |
224 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 225 | |
2451337d | 226 | return error; |
fd3200be CH |
227 | } |
228 | ||
00258e36 | 229 | STATIC ssize_t |
b4f5d2c6 | 230 | xfs_file_read_iter( |
dda35b8f | 231 | struct kiocb *iocb, |
b4f5d2c6 | 232 | struct iov_iter *to) |
dda35b8f CH |
233 | { |
234 | struct file *file = iocb->ki_filp; | |
235 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
236 | struct xfs_inode *ip = XFS_I(inode); |
237 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 238 | size_t size = iov_iter_count(to); |
dda35b8f | 239 | ssize_t ret = 0; |
00258e36 | 240 | int ioflags = 0; |
dda35b8f | 241 | xfs_fsize_t n; |
b4f5d2c6 | 242 | loff_t pos = iocb->ki_pos; |
dda35b8f | 243 | |
dda35b8f CH |
244 | XFS_STATS_INC(xs_read_calls); |
245 | ||
00258e36 | 246 | if (unlikely(file->f_flags & O_DIRECT)) |
b92cc59f | 247 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 248 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 249 | ioflags |= XFS_IO_INVIS; |
00258e36 | 250 | |
b92cc59f | 251 | if (unlikely(ioflags & XFS_IO_ISDIRECT)) { |
dda35b8f CH |
252 | xfs_buftarg_t *target = |
253 | XFS_IS_REALTIME_INODE(ip) ? | |
254 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
255 | /* DIO must be aligned to device logical sector size */ |
256 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 257 | if (pos == i_size_read(inode)) |
00258e36 | 258 | return 0; |
b474c7ae | 259 | return -EINVAL; |
dda35b8f CH |
260 | } |
261 | } | |
262 | ||
fb595814 | 263 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 264 | if (n <= 0 || size == 0) |
dda35b8f CH |
265 | return 0; |
266 | ||
267 | if (n < size) | |
268 | size = n; | |
269 | ||
270 | if (XFS_FORCED_SHUTDOWN(mp)) | |
271 | return -EIO; | |
272 | ||
0c38a251 DC |
273 | /* |
274 | * Locking is a bit tricky here. If we take an exclusive lock | |
275 | * for direct IO, we effectively serialise all new concurrent | |
276 | * read IO to this file and block it behind IO that is currently in | |
277 | * progress because IO in progress holds the IO lock shared. We only | |
278 | * need to hold the lock exclusive to blow away the page cache, so | |
279 | * only take lock exclusively if the page cache needs invalidation. | |
280 | * This allows the normal direct IO case of no page cache pages to | |
281 | * proceeed concurrently without serialisation. | |
282 | */ | |
283 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 284 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 285 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
286 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
287 | ||
00258e36 | 288 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 289 | ret = filemap_write_and_wait_range( |
fb595814 | 290 | VFS_I(ip)->i_mapping, |
7d4ea3ce | 291 | pos, pos + size - 1); |
487f84f3 DC |
292 | if (ret) { |
293 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
294 | return ret; | |
295 | } | |
85e584da CM |
296 | |
297 | /* | |
298 | * Invalidate whole pages. This can return an error if | |
299 | * we fail to invalidate a page, but this should never | |
300 | * happen on XFS. Warn if it does fail. | |
301 | */ | |
302 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
303 | pos >> PAGE_CACHE_SHIFT, |
304 | (pos + size - 1) >> PAGE_CACHE_SHIFT); | |
85e584da CM |
305 | WARN_ON_ONCE(ret); |
306 | ret = 0; | |
00258e36 | 307 | } |
487f84f3 | 308 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 309 | } |
dda35b8f | 310 | |
fb595814 | 311 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 312 | |
b4f5d2c6 | 313 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
314 | if (ret > 0) |
315 | XFS_STATS_ADD(xs_read_bytes, ret); | |
316 | ||
487f84f3 | 317 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
318 | return ret; |
319 | } | |
320 | ||
00258e36 CH |
321 | STATIC ssize_t |
322 | xfs_file_splice_read( | |
dda35b8f CH |
323 | struct file *infilp, |
324 | loff_t *ppos, | |
325 | struct pipe_inode_info *pipe, | |
326 | size_t count, | |
00258e36 | 327 | unsigned int flags) |
dda35b8f | 328 | { |
00258e36 | 329 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 330 | int ioflags = 0; |
dda35b8f CH |
331 | ssize_t ret; |
332 | ||
333 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
334 | |
335 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 336 | ioflags |= XFS_IO_INVIS; |
00258e36 | 337 | |
dda35b8f CH |
338 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
339 | return -EIO; | |
340 | ||
487f84f3 | 341 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 342 | |
dda35b8f CH |
343 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
344 | ||
345 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
346 | if (ret > 0) | |
347 | XFS_STATS_ADD(xs_read_bytes, ret); | |
348 | ||
487f84f3 | 349 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
350 | return ret; |
351 | } | |
352 | ||
dda35b8f | 353 | /* |
193aec10 CH |
354 | * This routine is called to handle zeroing any space in the last block of the |
355 | * file that is beyond the EOF. We do this since the size is being increased | |
356 | * without writing anything to that block and we don't want to read the | |
357 | * garbage on the disk. | |
dda35b8f CH |
358 | */ |
359 | STATIC int /* error (positive) */ | |
360 | xfs_zero_last_block( | |
193aec10 CH |
361 | struct xfs_inode *ip, |
362 | xfs_fsize_t offset, | |
363 | xfs_fsize_t isize) | |
dda35b8f | 364 | { |
193aec10 CH |
365 | struct xfs_mount *mp = ip->i_mount; |
366 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
367 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
368 | int zero_len; | |
369 | int nimaps = 1; | |
370 | int error = 0; | |
371 | struct xfs_bmbt_irec imap; | |
dda35b8f | 372 | |
193aec10 | 373 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 374 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 375 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 376 | if (error) |
dda35b8f | 377 | return error; |
193aec10 | 378 | |
dda35b8f | 379 | ASSERT(nimaps > 0); |
193aec10 | 380 | |
dda35b8f CH |
381 | /* |
382 | * If the block underlying isize is just a hole, then there | |
383 | * is nothing to zero. | |
384 | */ | |
193aec10 | 385 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 386 | return 0; |
dda35b8f CH |
387 | |
388 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
389 | if (isize + zero_len > offset) | |
390 | zero_len = offset - isize; | |
193aec10 | 391 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
392 | } |
393 | ||
394 | /* | |
193aec10 CH |
395 | * Zero any on disk space between the current EOF and the new, larger EOF. |
396 | * | |
397 | * This handles the normal case of zeroing the remainder of the last block in | |
398 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
399 | * file. This second case only happens with fixed size extents and when the | |
400 | * system crashes before the inode size was updated but after blocks were | |
401 | * allocated. | |
402 | * | |
403 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 404 | */ |
dda35b8f CH |
405 | int /* error (positive) */ |
406 | xfs_zero_eof( | |
193aec10 CH |
407 | struct xfs_inode *ip, |
408 | xfs_off_t offset, /* starting I/O offset */ | |
409 | xfs_fsize_t isize) /* current inode size */ | |
dda35b8f | 410 | { |
193aec10 CH |
411 | struct xfs_mount *mp = ip->i_mount; |
412 | xfs_fileoff_t start_zero_fsb; | |
413 | xfs_fileoff_t end_zero_fsb; | |
414 | xfs_fileoff_t zero_count_fsb; | |
415 | xfs_fileoff_t last_fsb; | |
416 | xfs_fileoff_t zero_off; | |
417 | xfs_fsize_t zero_len; | |
418 | int nimaps; | |
419 | int error = 0; | |
420 | struct xfs_bmbt_irec imap; | |
421 | ||
422 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
423 | ASSERT(offset > isize); |
424 | ||
425 | /* | |
426 | * First handle zeroing the block on which isize resides. | |
193aec10 | 427 | * |
dda35b8f CH |
428 | * We only zero a part of that block so it is handled specially. |
429 | */ | |
193aec10 CH |
430 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
431 | error = xfs_zero_last_block(ip, offset, isize); | |
432 | if (error) | |
433 | return error; | |
dda35b8f CH |
434 | } |
435 | ||
436 | /* | |
193aec10 CH |
437 | * Calculate the range between the new size and the old where blocks |
438 | * needing to be zeroed may exist. | |
439 | * | |
440 | * To get the block where the last byte in the file currently resides, | |
441 | * we need to subtract one from the size and truncate back to a block | |
442 | * boundary. We subtract 1 in case the size is exactly on a block | |
443 | * boundary. | |
dda35b8f CH |
444 | */ |
445 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
446 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
447 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
448 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
449 | if (last_fsb == end_zero_fsb) { | |
450 | /* | |
451 | * The size was only incremented on its last block. | |
452 | * We took care of that above, so just return. | |
453 | */ | |
454 | return 0; | |
455 | } | |
456 | ||
457 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
458 | while (start_zero_fsb <= end_zero_fsb) { | |
459 | nimaps = 1; | |
460 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
461 | |
462 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
463 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
464 | &imap, &nimaps, 0); | |
193aec10 CH |
465 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
466 | if (error) | |
dda35b8f | 467 | return error; |
193aec10 | 468 | |
dda35b8f CH |
469 | ASSERT(nimaps > 0); |
470 | ||
471 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
472 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
473 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
474 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
475 | continue; | |
476 | } | |
477 | ||
478 | /* | |
479 | * There are blocks we need to zero. | |
dda35b8f | 480 | */ |
dda35b8f CH |
481 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
482 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
483 | ||
484 | if ((zero_off + zero_len) > offset) | |
485 | zero_len = offset - zero_off; | |
486 | ||
487 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
488 | if (error) |
489 | return error; | |
dda35b8f CH |
490 | |
491 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
492 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
493 | } |
494 | ||
495 | return 0; | |
dda35b8f CH |
496 | } |
497 | ||
4d8d1581 DC |
498 | /* |
499 | * Common pre-write limit and setup checks. | |
500 | * | |
5bf1f262 CH |
501 | * Called with the iolocked held either shared and exclusive according to |
502 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
503 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
504 | */ |
505 | STATIC ssize_t | |
506 | xfs_file_aio_write_checks( | |
507 | struct file *file, | |
508 | loff_t *pos, | |
509 | size_t *count, | |
510 | int *iolock) | |
511 | { | |
512 | struct inode *inode = file->f_mapping->host; | |
513 | struct xfs_inode *ip = XFS_I(inode); | |
4d8d1581 DC |
514 | int error = 0; |
515 | ||
7271d243 | 516 | restart: |
4d8d1581 | 517 | error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); |
467f7899 | 518 | if (error) |
4d8d1581 | 519 | return error; |
4d8d1581 | 520 | |
4d8d1581 DC |
521 | /* |
522 | * If the offset is beyond the size of the file, we need to zero any | |
523 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 524 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
525 | * iolock shared, we need to update it to exclusive which implies |
526 | * having to redo all checks before. | |
4d8d1581 | 527 | */ |
2813d682 | 528 | if (*pos > i_size_read(inode)) { |
7271d243 | 529 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 530 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 531 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 532 | xfs_rw_ilock(ip, *iolock); |
7271d243 DC |
533 | goto restart; |
534 | } | |
2451337d | 535 | error = xfs_zero_eof(ip, *pos, i_size_read(inode)); |
467f7899 CH |
536 | if (error) |
537 | return error; | |
7271d243 | 538 | } |
4d8d1581 | 539 | |
8a9c9980 CH |
540 | /* |
541 | * Updating the timestamps will grab the ilock again from | |
542 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
543 | * lock above. Eventually we should look into a way to avoid | |
544 | * the pointless lock roundtrip. | |
545 | */ | |
c3b2da31 JB |
546 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
547 | error = file_update_time(file); | |
548 | if (error) | |
549 | return error; | |
550 | } | |
8a9c9980 | 551 | |
4d8d1581 DC |
552 | /* |
553 | * If we're writing the file then make sure to clear the setuid and | |
554 | * setgid bits if the process is not being run by root. This keeps | |
555 | * people from modifying setuid and setgid binaries. | |
556 | */ | |
557 | return file_remove_suid(file); | |
4d8d1581 DC |
558 | } |
559 | ||
f0d26e86 DC |
560 | /* |
561 | * xfs_file_dio_aio_write - handle direct IO writes | |
562 | * | |
563 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 564 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
565 | * follow locking changes and looping. |
566 | * | |
eda77982 DC |
567 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
568 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
569 | * pages are flushed out. | |
570 | * | |
571 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
572 | * allowing them to be done in parallel with reads and other direct IO writes. | |
573 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
574 | * needs to do sub-block zeroing and that requires serialisation against other | |
575 | * direct IOs to the same block. In this case we need to serialise the | |
576 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
577 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
578 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
579 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 580 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 581 | * |
f0d26e86 DC |
582 | * Returns with locks held indicated by @iolock and errors indicated by |
583 | * negative return values. | |
584 | */ | |
585 | STATIC ssize_t | |
586 | xfs_file_dio_aio_write( | |
587 | struct kiocb *iocb, | |
b3188919 | 588 | struct iov_iter *from) |
f0d26e86 DC |
589 | { |
590 | struct file *file = iocb->ki_filp; | |
591 | struct address_space *mapping = file->f_mapping; | |
592 | struct inode *inode = mapping->host; | |
593 | struct xfs_inode *ip = XFS_I(inode); | |
594 | struct xfs_mount *mp = ip->i_mount; | |
595 | ssize_t ret = 0; | |
eda77982 | 596 | int unaligned_io = 0; |
d0606464 | 597 | int iolock; |
b3188919 AV |
598 | size_t count = iov_iter_count(from); |
599 | loff_t pos = iocb->ki_pos; | |
f0d26e86 DC |
600 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
601 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
602 | ||
7c71ee78 ES |
603 | /* DIO must be aligned to device logical sector size */ |
604 | if ((pos | count) & target->bt_logical_sectormask) | |
b474c7ae | 605 | return -EINVAL; |
f0d26e86 | 606 | |
7c71ee78 | 607 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
608 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
609 | unaligned_io = 1; | |
610 | ||
7271d243 DC |
611 | /* |
612 | * We don't need to take an exclusive lock unless there page cache needs | |
613 | * to be invalidated or unaligned IO is being executed. We don't need to | |
614 | * consider the EOF extension case here because | |
615 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
616 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
617 | */ | |
618 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 619 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 620 | else |
d0606464 CH |
621 | iolock = XFS_IOLOCK_SHARED; |
622 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
623 | |
624 | /* | |
625 | * Recheck if there are cached pages that need invalidate after we got | |
626 | * the iolock to protect against other threads adding new pages while | |
627 | * we were waiting for the iolock. | |
628 | */ | |
d0606464 CH |
629 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
630 | xfs_rw_iunlock(ip, iolock); | |
631 | iolock = XFS_IOLOCK_EXCL; | |
632 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 633 | } |
f0d26e86 | 634 | |
d0606464 | 635 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 636 | if (ret) |
d0606464 | 637 | goto out; |
b3188919 | 638 | iov_iter_truncate(from, count); |
f0d26e86 DC |
639 | |
640 | if (mapping->nrpages) { | |
07d5035a | 641 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
7d4ea3ce | 642 | pos, pos + count - 1); |
f0d26e86 | 643 | if (ret) |
d0606464 | 644 | goto out; |
834ffca6 DC |
645 | /* |
646 | * Invalidate whole pages. This can return an error if | |
647 | * we fail to invalidate a page, but this should never | |
648 | * happen on XFS. Warn if it does fail. | |
649 | */ | |
650 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
651 | pos >> PAGE_CACHE_SHIFT, |
652 | (pos + count - 1) >> PAGE_CACHE_SHIFT); | |
834ffca6 DC |
653 | WARN_ON_ONCE(ret); |
654 | ret = 0; | |
f0d26e86 DC |
655 | } |
656 | ||
eda77982 DC |
657 | /* |
658 | * If we are doing unaligned IO, wait for all other IO to drain, | |
659 | * otherwise demote the lock if we had to flush cached pages | |
660 | */ | |
661 | if (unaligned_io) | |
4a06fd26 | 662 | inode_dio_wait(inode); |
d0606464 | 663 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 664 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 665 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
666 | } |
667 | ||
668 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
b3188919 | 669 | ret = generic_file_direct_write(iocb, from, pos); |
f0d26e86 | 670 | |
d0606464 CH |
671 | out: |
672 | xfs_rw_iunlock(ip, iolock); | |
673 | ||
f0d26e86 DC |
674 | /* No fallback to buffered IO on errors for XFS. */ |
675 | ASSERT(ret < 0 || ret == count); | |
676 | return ret; | |
677 | } | |
678 | ||
00258e36 | 679 | STATIC ssize_t |
637bbc75 | 680 | xfs_file_buffered_aio_write( |
dda35b8f | 681 | struct kiocb *iocb, |
b3188919 | 682 | struct iov_iter *from) |
dda35b8f CH |
683 | { |
684 | struct file *file = iocb->ki_filp; | |
685 | struct address_space *mapping = file->f_mapping; | |
686 | struct inode *inode = mapping->host; | |
00258e36 | 687 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
688 | ssize_t ret; |
689 | int enospc = 0; | |
d0606464 | 690 | int iolock = XFS_IOLOCK_EXCL; |
b3188919 AV |
691 | loff_t pos = iocb->ki_pos; |
692 | size_t count = iov_iter_count(from); | |
dda35b8f | 693 | |
d0606464 | 694 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 695 | |
d0606464 | 696 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 697 | if (ret) |
d0606464 | 698 | goto out; |
dda35b8f | 699 | |
b3188919 | 700 | iov_iter_truncate(from, count); |
dda35b8f CH |
701 | /* We can write back this queue in page reclaim */ |
702 | current->backing_dev_info = mapping->backing_dev_info; | |
703 | ||
dda35b8f | 704 | write_retry: |
637bbc75 | 705 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); |
b3188919 | 706 | ret = generic_perform_write(file, from, pos); |
0a64bc2c AV |
707 | if (likely(ret >= 0)) |
708 | iocb->ki_pos = pos + ret; | |
dc06f398 | 709 | |
637bbc75 | 710 | /* |
dc06f398 BF |
711 | * If we hit a space limit, try to free up some lingering preallocated |
712 | * space before returning an error. In the case of ENOSPC, first try to | |
713 | * write back all dirty inodes to free up some of the excess reserved | |
714 | * metadata space. This reduces the chances that the eofblocks scan | |
715 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
716 | * also behaves as a filter to prevent too many eofblocks scans from | |
717 | * running at the same time. | |
637bbc75 | 718 | */ |
dc06f398 BF |
719 | if (ret == -EDQUOT && !enospc) { |
720 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
721 | if (enospc) | |
722 | goto write_retry; | |
723 | } else if (ret == -ENOSPC && !enospc) { | |
724 | struct xfs_eofblocks eofb = {0}; | |
725 | ||
637bbc75 | 726 | enospc = 1; |
9aa05000 | 727 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
728 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
729 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
730 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 731 | goto write_retry; |
dda35b8f | 732 | } |
d0606464 | 733 | |
dda35b8f | 734 | current->backing_dev_info = NULL; |
d0606464 CH |
735 | out: |
736 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
737 | return ret; |
738 | } | |
739 | ||
740 | STATIC ssize_t | |
bf97f3bc | 741 | xfs_file_write_iter( |
637bbc75 | 742 | struct kiocb *iocb, |
bf97f3bc | 743 | struct iov_iter *from) |
637bbc75 DC |
744 | { |
745 | struct file *file = iocb->ki_filp; | |
746 | struct address_space *mapping = file->f_mapping; | |
747 | struct inode *inode = mapping->host; | |
748 | struct xfs_inode *ip = XFS_I(inode); | |
749 | ssize_t ret; | |
bf97f3bc | 750 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
751 | |
752 | XFS_STATS_INC(xs_write_calls); | |
753 | ||
637bbc75 DC |
754 | if (ocount == 0) |
755 | return 0; | |
756 | ||
bf97f3bc AV |
757 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
758 | return -EIO; | |
637bbc75 DC |
759 | |
760 | if (unlikely(file->f_flags & O_DIRECT)) | |
bf97f3bc | 761 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 762 | else |
bf97f3bc | 763 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 764 | |
d0606464 CH |
765 | if (ret > 0) { |
766 | ssize_t err; | |
dda35b8f | 767 | |
d0606464 | 768 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 769 | |
d0606464 | 770 | /* Handle various SYNC-type writes */ |
d311d79d | 771 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
772 | if (err < 0) |
773 | ret = err; | |
dda35b8f | 774 | } |
a363f0c2 | 775 | return ret; |
dda35b8f CH |
776 | } |
777 | ||
2fe17c10 CH |
778 | STATIC long |
779 | xfs_file_fallocate( | |
83aee9e4 CH |
780 | struct file *file, |
781 | int mode, | |
782 | loff_t offset, | |
783 | loff_t len) | |
2fe17c10 | 784 | { |
83aee9e4 CH |
785 | struct inode *inode = file_inode(file); |
786 | struct xfs_inode *ip = XFS_I(inode); | |
787 | struct xfs_trans *tp; | |
788 | long error; | |
789 | loff_t new_size = 0; | |
2fe17c10 | 790 | |
83aee9e4 CH |
791 | if (!S_ISREG(inode->i_mode)) |
792 | return -EINVAL; | |
e1d8fb88 | 793 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
376ba313 | 794 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
2fe17c10 CH |
795 | return -EOPNOTSUPP; |
796 | ||
2fe17c10 | 797 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
83aee9e4 CH |
798 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
799 | error = xfs_free_file_space(ip, offset, len); | |
800 | if (error) | |
801 | goto out_unlock; | |
e1d8fb88 NJ |
802 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
803 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
804 | ||
805 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 806 | error = -EINVAL; |
e1d8fb88 NJ |
807 | goto out_unlock; |
808 | } | |
809 | ||
23fffa92 LC |
810 | /* |
811 | * There is no need to overlap collapse range with EOF, | |
812 | * in which case it is effectively a truncate operation | |
813 | */ | |
814 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 815 | error = -EINVAL; |
23fffa92 LC |
816 | goto out_unlock; |
817 | } | |
818 | ||
e1d8fb88 NJ |
819 | new_size = i_size_read(inode) - len; |
820 | ||
821 | error = xfs_collapse_file_space(ip, offset, len); | |
822 | if (error) | |
823 | goto out_unlock; | |
83aee9e4 CH |
824 | } else { |
825 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
826 | offset + len > i_size_read(inode)) { | |
827 | new_size = offset + len; | |
2451337d | 828 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
829 | if (error) |
830 | goto out_unlock; | |
831 | } | |
2fe17c10 | 832 | |
376ba313 LC |
833 | if (mode & FALLOC_FL_ZERO_RANGE) |
834 | error = xfs_zero_file_space(ip, offset, len); | |
835 | else | |
836 | error = xfs_alloc_file_space(ip, offset, len, | |
837 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
838 | if (error) |
839 | goto out_unlock; | |
840 | } | |
841 | ||
83aee9e4 CH |
842 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); |
843 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
844 | if (error) { | |
845 | xfs_trans_cancel(tp, 0); | |
846 | goto out_unlock; | |
847 | } | |
848 | ||
849 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
850 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
851 | ip->i_d.di_mode &= ~S_ISUID; | |
852 | if (ip->i_d.di_mode & S_IXGRP) | |
853 | ip->i_d.di_mode &= ~S_ISGID; | |
82878897 | 854 | |
e1d8fb88 | 855 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE))) |
83aee9e4 CH |
856 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; |
857 | ||
858 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
859 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
860 | ||
861 | if (file->f_flags & O_DSYNC) | |
862 | xfs_trans_set_sync(tp); | |
863 | error = xfs_trans_commit(tp, 0); | |
2fe17c10 CH |
864 | if (error) |
865 | goto out_unlock; | |
866 | ||
867 | /* Change file size if needed */ | |
868 | if (new_size) { | |
869 | struct iattr iattr; | |
870 | ||
871 | iattr.ia_valid = ATTR_SIZE; | |
872 | iattr.ia_size = new_size; | |
83aee9e4 | 873 | error = xfs_setattr_size(ip, &iattr); |
2fe17c10 CH |
874 | } |
875 | ||
876 | out_unlock: | |
877 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2451337d | 878 | return error; |
2fe17c10 CH |
879 | } |
880 | ||
881 | ||
1da177e4 | 882 | STATIC int |
3562fd45 | 883 | xfs_file_open( |
1da177e4 | 884 | struct inode *inode, |
f999a5bf | 885 | struct file *file) |
1da177e4 | 886 | { |
f999a5bf | 887 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 888 | return -EFBIG; |
f999a5bf CH |
889 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
890 | return -EIO; | |
891 | return 0; | |
892 | } | |
893 | ||
894 | STATIC int | |
895 | xfs_dir_open( | |
896 | struct inode *inode, | |
897 | struct file *file) | |
898 | { | |
899 | struct xfs_inode *ip = XFS_I(inode); | |
900 | int mode; | |
901 | int error; | |
902 | ||
903 | error = xfs_file_open(inode, file); | |
904 | if (error) | |
905 | return error; | |
906 | ||
907 | /* | |
908 | * If there are any blocks, read-ahead block 0 as we're almost | |
909 | * certain to have the next operation be a read there. | |
910 | */ | |
309ecac8 | 911 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 912 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 913 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
914 | xfs_iunlock(ip, mode); |
915 | return 0; | |
1da177e4 LT |
916 | } |
917 | ||
1da177e4 | 918 | STATIC int |
3562fd45 | 919 | xfs_file_release( |
1da177e4 LT |
920 | struct inode *inode, |
921 | struct file *filp) | |
922 | { | |
2451337d | 923 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
924 | } |
925 | ||
1da177e4 | 926 | STATIC int |
3562fd45 | 927 | xfs_file_readdir( |
b8227554 AV |
928 | struct file *file, |
929 | struct dir_context *ctx) | |
1da177e4 | 930 | { |
b8227554 | 931 | struct inode *inode = file_inode(file); |
739bfb2a | 932 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
933 | size_t bufsize; |
934 | ||
935 | /* | |
936 | * The Linux API doesn't pass down the total size of the buffer | |
937 | * we read into down to the filesystem. With the filldir concept | |
938 | * it's not needed for correct information, but the XFS dir2 leaf | |
939 | * code wants an estimate of the buffer size to calculate it's | |
940 | * readahead window and size the buffers used for mapping to | |
941 | * physical blocks. | |
942 | * | |
943 | * Try to give it an estimate that's good enough, maybe at some | |
944 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 945 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 946 | */ |
a9cc799e | 947 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 948 | |
8300475e | 949 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
950 | } |
951 | ||
1da177e4 | 952 | STATIC int |
3562fd45 | 953 | xfs_file_mmap( |
1da177e4 LT |
954 | struct file *filp, |
955 | struct vm_area_struct *vma) | |
956 | { | |
3562fd45 | 957 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 958 | |
fbc1462b | 959 | file_accessed(filp); |
1da177e4 LT |
960 | return 0; |
961 | } | |
962 | ||
4f57dbc6 DC |
963 | /* |
964 | * mmap()d file has taken write protection fault and is being made | |
965 | * writable. We can set the page state up correctly for a writable | |
966 | * page, which means we can do correct delalloc accounting (ENOSPC | |
967 | * checking!) and unwritten extent mapping. | |
968 | */ | |
969 | STATIC int | |
970 | xfs_vm_page_mkwrite( | |
971 | struct vm_area_struct *vma, | |
c2ec175c | 972 | struct vm_fault *vmf) |
4f57dbc6 | 973 | { |
c2ec175c | 974 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
975 | } |
976 | ||
d126d43f JL |
977 | /* |
978 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 979 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
980 | */ |
981 | enum { | |
982 | HOLE_OFF = 0, | |
983 | DATA_OFF, | |
984 | }; | |
985 | ||
986 | /* | |
987 | * Lookup the desired type of offset from the given page. | |
988 | * | |
989 | * On success, return true and the offset argument will point to the | |
990 | * start of the region that was found. Otherwise this function will | |
991 | * return false and keep the offset argument unchanged. | |
992 | */ | |
993 | STATIC bool | |
994 | xfs_lookup_buffer_offset( | |
995 | struct page *page, | |
996 | loff_t *offset, | |
997 | unsigned int type) | |
998 | { | |
999 | loff_t lastoff = page_offset(page); | |
1000 | bool found = false; | |
1001 | struct buffer_head *bh, *head; | |
1002 | ||
1003 | bh = head = page_buffers(page); | |
1004 | do { | |
1005 | /* | |
1006 | * Unwritten extents that have data in the page | |
1007 | * cache covering them can be identified by the | |
1008 | * BH_Unwritten state flag. Pages with multiple | |
1009 | * buffers might have a mix of holes, data and | |
1010 | * unwritten extents - any buffer with valid | |
1011 | * data in it should have BH_Uptodate flag set | |
1012 | * on it. | |
1013 | */ | |
1014 | if (buffer_unwritten(bh) || | |
1015 | buffer_uptodate(bh)) { | |
1016 | if (type == DATA_OFF) | |
1017 | found = true; | |
1018 | } else { | |
1019 | if (type == HOLE_OFF) | |
1020 | found = true; | |
1021 | } | |
1022 | ||
1023 | if (found) { | |
1024 | *offset = lastoff; | |
1025 | break; | |
1026 | } | |
1027 | lastoff += bh->b_size; | |
1028 | } while ((bh = bh->b_this_page) != head); | |
1029 | ||
1030 | return found; | |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * This routine is called to find out and return a data or hole offset | |
1035 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1036 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1037 | * |
1038 | * The argument offset is used to tell where we start to search from the | |
1039 | * page cache. Map is used to figure out the end points of the range to | |
1040 | * lookup pages. | |
1041 | * | |
1042 | * Return true if the desired type of offset was found, and the argument | |
1043 | * offset is filled with that address. Otherwise, return false and keep | |
1044 | * offset unchanged. | |
1045 | */ | |
1046 | STATIC bool | |
1047 | xfs_find_get_desired_pgoff( | |
1048 | struct inode *inode, | |
1049 | struct xfs_bmbt_irec *map, | |
1050 | unsigned int type, | |
1051 | loff_t *offset) | |
1052 | { | |
1053 | struct xfs_inode *ip = XFS_I(inode); | |
1054 | struct xfs_mount *mp = ip->i_mount; | |
1055 | struct pagevec pvec; | |
1056 | pgoff_t index; | |
1057 | pgoff_t end; | |
1058 | loff_t endoff; | |
1059 | loff_t startoff = *offset; | |
1060 | loff_t lastoff = startoff; | |
1061 | bool found = false; | |
1062 | ||
1063 | pagevec_init(&pvec, 0); | |
1064 | ||
1065 | index = startoff >> PAGE_CACHE_SHIFT; | |
1066 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1067 | end = endoff >> PAGE_CACHE_SHIFT; | |
1068 | do { | |
1069 | int want; | |
1070 | unsigned nr_pages; | |
1071 | unsigned int i; | |
1072 | ||
1073 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1074 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1075 | want); | |
1076 | /* | |
1077 | * No page mapped into given range. If we are searching holes | |
1078 | * and if this is the first time we got into the loop, it means | |
1079 | * that the given offset is landed in a hole, return it. | |
1080 | * | |
1081 | * If we have already stepped through some block buffers to find | |
1082 | * holes but they all contains data. In this case, the last | |
1083 | * offset is already updated and pointed to the end of the last | |
1084 | * mapped page, if it does not reach the endpoint to search, | |
1085 | * that means there should be a hole between them. | |
1086 | */ | |
1087 | if (nr_pages == 0) { | |
1088 | /* Data search found nothing */ | |
1089 | if (type == DATA_OFF) | |
1090 | break; | |
1091 | ||
1092 | ASSERT(type == HOLE_OFF); | |
1093 | if (lastoff == startoff || lastoff < endoff) { | |
1094 | found = true; | |
1095 | *offset = lastoff; | |
1096 | } | |
1097 | break; | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * At lease we found one page. If this is the first time we | |
1102 | * step into the loop, and if the first page index offset is | |
1103 | * greater than the given search offset, a hole was found. | |
1104 | */ | |
1105 | if (type == HOLE_OFF && lastoff == startoff && | |
1106 | lastoff < page_offset(pvec.pages[0])) { | |
1107 | found = true; | |
1108 | break; | |
1109 | } | |
1110 | ||
1111 | for (i = 0; i < nr_pages; i++) { | |
1112 | struct page *page = pvec.pages[i]; | |
1113 | loff_t b_offset; | |
1114 | ||
1115 | /* | |
1116 | * At this point, the page may be truncated or | |
1117 | * invalidated (changing page->mapping to NULL), | |
1118 | * or even swizzled back from swapper_space to tmpfs | |
1119 | * file mapping. However, page->index will not change | |
1120 | * because we have a reference on the page. | |
1121 | * | |
1122 | * Searching done if the page index is out of range. | |
1123 | * If the current offset is not reaches the end of | |
1124 | * the specified search range, there should be a hole | |
1125 | * between them. | |
1126 | */ | |
1127 | if (page->index > end) { | |
1128 | if (type == HOLE_OFF && lastoff < endoff) { | |
1129 | *offset = lastoff; | |
1130 | found = true; | |
1131 | } | |
1132 | goto out; | |
1133 | } | |
1134 | ||
1135 | lock_page(page); | |
1136 | /* | |
1137 | * Page truncated or invalidated(page->mapping == NULL). | |
1138 | * We can freely skip it and proceed to check the next | |
1139 | * page. | |
1140 | */ | |
1141 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1142 | unlock_page(page); | |
1143 | continue; | |
1144 | } | |
1145 | ||
1146 | if (!page_has_buffers(page)) { | |
1147 | unlock_page(page); | |
1148 | continue; | |
1149 | } | |
1150 | ||
1151 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1152 | if (found) { | |
1153 | /* | |
1154 | * The found offset may be less than the start | |
1155 | * point to search if this is the first time to | |
1156 | * come here. | |
1157 | */ | |
1158 | *offset = max_t(loff_t, startoff, b_offset); | |
1159 | unlock_page(page); | |
1160 | goto out; | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * We either searching data but nothing was found, or | |
1165 | * searching hole but found a data buffer. In either | |
1166 | * case, probably the next page contains the desired | |
1167 | * things, update the last offset to it so. | |
1168 | */ | |
1169 | lastoff = page_offset(page) + PAGE_SIZE; | |
1170 | unlock_page(page); | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * The number of returned pages less than our desired, search | |
1175 | * done. In this case, nothing was found for searching data, | |
1176 | * but we found a hole behind the last offset. | |
1177 | */ | |
1178 | if (nr_pages < want) { | |
1179 | if (type == HOLE_OFF) { | |
1180 | *offset = lastoff; | |
1181 | found = true; | |
1182 | } | |
1183 | break; | |
1184 | } | |
1185 | ||
1186 | index = pvec.pages[i - 1]->index + 1; | |
1187 | pagevec_release(&pvec); | |
1188 | } while (index <= end); | |
1189 | ||
1190 | out: | |
1191 | pagevec_release(&pvec); | |
1192 | return found; | |
1193 | } | |
1194 | ||
3fe3e6b1 | 1195 | STATIC loff_t |
49c69591 | 1196 | xfs_seek_hole_data( |
3fe3e6b1 | 1197 | struct file *file, |
49c69591 ES |
1198 | loff_t start, |
1199 | int whence) | |
3fe3e6b1 JL |
1200 | { |
1201 | struct inode *inode = file->f_mapping->host; | |
1202 | struct xfs_inode *ip = XFS_I(inode); | |
1203 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1204 | loff_t uninitialized_var(offset); |
1205 | xfs_fsize_t isize; | |
1206 | xfs_fileoff_t fsbno; | |
1207 | xfs_filblks_t end; | |
1208 | uint lock; | |
1209 | int error; | |
1210 | ||
49c69591 ES |
1211 | if (XFS_FORCED_SHUTDOWN(mp)) |
1212 | return -EIO; | |
1213 | ||
309ecac8 | 1214 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1215 | |
1216 | isize = i_size_read(inode); | |
1217 | if (start >= isize) { | |
2451337d | 1218 | error = -ENXIO; |
3fe3e6b1 JL |
1219 | goto out_unlock; |
1220 | } | |
1221 | ||
3fe3e6b1 JL |
1222 | /* |
1223 | * Try to read extents from the first block indicated | |
1224 | * by fsbno to the end block of the file. | |
1225 | */ | |
52f1acc8 | 1226 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1227 | end = XFS_B_TO_FSB(mp, isize); |
49c69591 | 1228 | |
52f1acc8 JL |
1229 | for (;;) { |
1230 | struct xfs_bmbt_irec map[2]; | |
1231 | int nmap = 2; | |
1232 | unsigned int i; | |
3fe3e6b1 | 1233 | |
52f1acc8 JL |
1234 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1235 | XFS_BMAPI_ENTIRE); | |
1236 | if (error) | |
1237 | goto out_unlock; | |
3fe3e6b1 | 1238 | |
52f1acc8 JL |
1239 | /* No extents at given offset, must be beyond EOF */ |
1240 | if (nmap == 0) { | |
2451337d | 1241 | error = -ENXIO; |
52f1acc8 JL |
1242 | goto out_unlock; |
1243 | } | |
1244 | ||
1245 | for (i = 0; i < nmap; i++) { | |
1246 | offset = max_t(loff_t, start, | |
1247 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1248 | ||
49c69591 ES |
1249 | /* Landed in the hole we wanted? */ |
1250 | if (whence == SEEK_HOLE && | |
1251 | map[i].br_startblock == HOLESTARTBLOCK) | |
1252 | goto out; | |
1253 | ||
1254 | /* Landed in the data extent we wanted? */ | |
1255 | if (whence == SEEK_DATA && | |
1256 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1257 | (map[i].br_state == XFS_EXT_NORM && | |
1258 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1259 | goto out; |
1260 | ||
1261 | /* | |
49c69591 ES |
1262 | * Landed in an unwritten extent, try to search |
1263 | * for hole or data from page cache. | |
52f1acc8 JL |
1264 | */ |
1265 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1266 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1267 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1268 | &offset)) | |
52f1acc8 JL |
1269 | goto out; |
1270 | } | |
1271 | } | |
1272 | ||
1273 | /* | |
49c69591 ES |
1274 | * We only received one extent out of the two requested. This |
1275 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1276 | */ |
3fe3e6b1 | 1277 | if (nmap == 1) { |
49c69591 ES |
1278 | /* |
1279 | * If we were looking for a hole, set offset to | |
1280 | * the end of the file (i.e., there is an implicit | |
1281 | * hole at the end of any file). | |
1282 | */ | |
1283 | if (whence == SEEK_HOLE) { | |
1284 | offset = isize; | |
1285 | break; | |
1286 | } | |
1287 | /* | |
1288 | * If we were looking for data, it's nowhere to be found | |
1289 | */ | |
1290 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1291 | error = -ENXIO; |
3fe3e6b1 JL |
1292 | goto out_unlock; |
1293 | } | |
1294 | ||
52f1acc8 JL |
1295 | ASSERT(i > 1); |
1296 | ||
1297 | /* | |
1298 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1299 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1300 | */ |
1301 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1302 | start = XFS_FSB_TO_B(mp, fsbno); | |
1303 | if (start >= isize) { | |
49c69591 ES |
1304 | if (whence == SEEK_HOLE) { |
1305 | offset = isize; | |
1306 | break; | |
1307 | } | |
1308 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1309 | error = -ENXIO; |
52f1acc8 JL |
1310 | goto out_unlock; |
1311 | } | |
3fe3e6b1 JL |
1312 | } |
1313 | ||
b686d1f7 JL |
1314 | out: |
1315 | /* | |
49c69591 | 1316 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1317 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1318 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1319 | * situation in particular. |
1320 | */ | |
49c69591 ES |
1321 | if (whence == SEEK_HOLE) |
1322 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1323 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1324 | |
1325 | out_unlock: | |
01f4f327 | 1326 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1327 | |
1328 | if (error) | |
2451337d | 1329 | return error; |
3fe3e6b1 JL |
1330 | return offset; |
1331 | } | |
1332 | ||
1333 | STATIC loff_t | |
1334 | xfs_file_llseek( | |
1335 | struct file *file, | |
1336 | loff_t offset, | |
59f9c004 | 1337 | int whence) |
3fe3e6b1 | 1338 | { |
59f9c004 | 1339 | switch (whence) { |
3fe3e6b1 JL |
1340 | case SEEK_END: |
1341 | case SEEK_CUR: | |
1342 | case SEEK_SET: | |
59f9c004 | 1343 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1344 | case SEEK_HOLE: |
49c69591 | 1345 | case SEEK_DATA: |
59f9c004 | 1346 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1347 | default: |
1348 | return -EINVAL; | |
1349 | } | |
1350 | } | |
1351 | ||
4b6f5d20 | 1352 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1353 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1354 | .read = new_sync_read, |
bf97f3bc | 1355 | .write = new_sync_write, |
b4f5d2c6 | 1356 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1357 | .write_iter = xfs_file_write_iter, |
1b895840 | 1358 | .splice_read = xfs_file_splice_read, |
8d020765 | 1359 | .splice_write = iter_file_splice_write, |
3562fd45 | 1360 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1361 | #ifdef CONFIG_COMPAT |
3562fd45 | 1362 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1363 | #endif |
3562fd45 NS |
1364 | .mmap = xfs_file_mmap, |
1365 | .open = xfs_file_open, | |
1366 | .release = xfs_file_release, | |
1367 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1368 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1369 | }; |
1370 | ||
4b6f5d20 | 1371 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1372 | .open = xfs_dir_open, |
1da177e4 | 1373 | .read = generic_read_dir, |
b8227554 | 1374 | .iterate = xfs_file_readdir, |
59af1584 | 1375 | .llseek = generic_file_llseek, |
3562fd45 | 1376 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1377 | #ifdef CONFIG_COMPAT |
3562fd45 | 1378 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1379 | #endif |
1da2f2db | 1380 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1381 | }; |
1382 | ||
f0f37e2f | 1383 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1384 | .fault = filemap_fault, |
f1820361 | 1385 | .map_pages = filemap_map_pages, |
4f57dbc6 | 1386 | .page_mkwrite = xfs_vm_page_mkwrite, |
0b173bc4 | 1387 | .remap_pages = generic_file_remap_pages, |
6fac0cb4 | 1388 | }; |