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_sb.h" |
a844f451 | 25 | #include "xfs_ag.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
57062787 DC |
27 | #include "xfs_da_format.h" |
28 | #include "xfs_da_btree.h" | |
1da177e4 | 29 | #include "xfs_inode.h" |
239880ef | 30 | #include "xfs_trans.h" |
fd3200be | 31 | #include "xfs_inode_item.h" |
dda35b8f | 32 | #include "xfs_bmap.h" |
c24b5dfa | 33 | #include "xfs_bmap_util.h" |
1da177e4 | 34 | #include "xfs_error.h" |
2b9ab5ab | 35 | #include "xfs_dir2.h" |
c24b5dfa | 36 | #include "xfs_dir2_priv.h" |
ddcd856d | 37 | #include "xfs_ioctl.h" |
dda35b8f | 38 | #include "xfs_trace.h" |
239880ef | 39 | #include "xfs_log.h" |
a4fbe6ab | 40 | #include "xfs_dinode.h" |
dc06f398 | 41 | #include "xfs_icache.h" |
1da177e4 | 42 | |
a27bb332 | 43 | #include <linux/aio.h> |
1da177e4 | 44 | #include <linux/dcache.h> |
2fe17c10 | 45 | #include <linux/falloc.h> |
d126d43f | 46 | #include <linux/pagevec.h> |
1da177e4 | 47 | |
f0f37e2f | 48 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 49 | |
487f84f3 DC |
50 | /* |
51 | * Locking primitives for read and write IO paths to ensure we consistently use | |
52 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
53 | */ | |
54 | static inline void | |
55 | xfs_rw_ilock( | |
56 | struct xfs_inode *ip, | |
57 | int type) | |
58 | { | |
59 | if (type & XFS_IOLOCK_EXCL) | |
60 | mutex_lock(&VFS_I(ip)->i_mutex); | |
61 | xfs_ilock(ip, type); | |
62 | } | |
63 | ||
64 | static inline void | |
65 | xfs_rw_iunlock( | |
66 | struct xfs_inode *ip, | |
67 | int type) | |
68 | { | |
69 | xfs_iunlock(ip, type); | |
70 | if (type & XFS_IOLOCK_EXCL) | |
71 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
72 | } | |
73 | ||
74 | static inline void | |
75 | xfs_rw_ilock_demote( | |
76 | struct xfs_inode *ip, | |
77 | int type) | |
78 | { | |
79 | xfs_ilock_demote(ip, type); | |
80 | if (type & XFS_IOLOCK_EXCL) | |
81 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
82 | } | |
83 | ||
dda35b8f CH |
84 | /* |
85 | * xfs_iozero | |
86 | * | |
87 | * xfs_iozero clears the specified range of buffer supplied, | |
88 | * and marks all the affected blocks as valid and modified. If | |
89 | * an affected block is not allocated, it will be allocated. If | |
90 | * an affected block is not completely overwritten, and is not | |
91 | * valid before the operation, it will be read from disk before | |
92 | * being partially zeroed. | |
93 | */ | |
ef9d8733 | 94 | int |
dda35b8f CH |
95 | xfs_iozero( |
96 | struct xfs_inode *ip, /* inode */ | |
97 | loff_t pos, /* offset in file */ | |
98 | size_t count) /* size of data to zero */ | |
99 | { | |
100 | struct page *page; | |
101 | struct address_space *mapping; | |
102 | int status; | |
103 | ||
104 | mapping = VFS_I(ip)->i_mapping; | |
105 | do { | |
106 | unsigned offset, bytes; | |
107 | void *fsdata; | |
108 | ||
109 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
110 | bytes = PAGE_CACHE_SIZE - offset; | |
111 | if (bytes > count) | |
112 | bytes = count; | |
113 | ||
114 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
115 | AOP_FLAG_UNINTERRUPTIBLE, | |
116 | &page, &fsdata); | |
117 | if (status) | |
118 | break; | |
119 | ||
120 | zero_user(page, offset, bytes); | |
121 | ||
122 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
123 | page, fsdata); | |
124 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
125 | pos += bytes; | |
126 | count -= bytes; | |
127 | status = 0; | |
128 | } while (count); | |
129 | ||
130 | return (-status); | |
131 | } | |
132 | ||
1da2f2db CH |
133 | /* |
134 | * Fsync operations on directories are much simpler than on regular files, | |
135 | * as there is no file data to flush, and thus also no need for explicit | |
136 | * cache flush operations, and there are no non-transaction metadata updates | |
137 | * on directories either. | |
138 | */ | |
139 | STATIC int | |
140 | xfs_dir_fsync( | |
141 | struct file *file, | |
142 | loff_t start, | |
143 | loff_t end, | |
144 | int datasync) | |
145 | { | |
146 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
147 | struct xfs_mount *mp = ip->i_mount; | |
148 | xfs_lsn_t lsn = 0; | |
149 | ||
150 | trace_xfs_dir_fsync(ip); | |
151 | ||
152 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
153 | if (xfs_ipincount(ip)) | |
154 | lsn = ip->i_itemp->ili_last_lsn; | |
155 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
156 | ||
157 | if (!lsn) | |
158 | return 0; | |
2451337d | 159 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
160 | } |
161 | ||
fd3200be CH |
162 | STATIC int |
163 | xfs_file_fsync( | |
164 | struct file *file, | |
02c24a82 JB |
165 | loff_t start, |
166 | loff_t end, | |
fd3200be CH |
167 | int datasync) |
168 | { | |
7ea80859 CH |
169 | struct inode *inode = file->f_mapping->host; |
170 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 171 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
172 | int error = 0; |
173 | int log_flushed = 0; | |
b1037058 | 174 | xfs_lsn_t lsn = 0; |
fd3200be | 175 | |
cca28fb8 | 176 | trace_xfs_file_fsync(ip); |
fd3200be | 177 | |
02c24a82 JB |
178 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
179 | if (error) | |
180 | return error; | |
181 | ||
a27a263b | 182 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 183 | return -EIO; |
fd3200be CH |
184 | |
185 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
186 | ||
a27a263b CH |
187 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
188 | /* | |
189 | * If we have an RT and/or log subvolume we need to make sure | |
190 | * to flush the write cache the device used for file data | |
191 | * first. This is to ensure newly written file data make | |
192 | * it to disk before logging the new inode size in case of | |
193 | * an extending write. | |
194 | */ | |
195 | if (XFS_IS_REALTIME_INODE(ip)) | |
196 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
197 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
198 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
199 | } | |
200 | ||
fd3200be | 201 | /* |
8a9c9980 CH |
202 | * All metadata updates are logged, which means that we just have |
203 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
204 | */ |
205 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
206 | if (xfs_ipincount(ip)) { |
207 | if (!datasync || | |
208 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
209 | lsn = ip->i_itemp->ili_last_lsn; | |
210 | } | |
8a9c9980 | 211 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 212 | |
8a9c9980 | 213 | if (lsn) |
b1037058 CH |
214 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
215 | ||
a27a263b CH |
216 | /* |
217 | * If we only have a single device, and the log force about was | |
218 | * a no-op we might have to flush the data device cache here. | |
219 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
220 | * an already allocated file and thus do not have any metadata to | |
221 | * commit. | |
222 | */ | |
223 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
224 | mp->m_logdev_targp == mp->m_ddev_targp && | |
225 | !XFS_IS_REALTIME_INODE(ip) && | |
226 | !log_flushed) | |
227 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 228 | |
2451337d | 229 | return error; |
fd3200be CH |
230 | } |
231 | ||
00258e36 | 232 | STATIC ssize_t |
b4f5d2c6 | 233 | xfs_file_read_iter( |
dda35b8f | 234 | struct kiocb *iocb, |
b4f5d2c6 | 235 | struct iov_iter *to) |
dda35b8f CH |
236 | { |
237 | struct file *file = iocb->ki_filp; | |
238 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
239 | struct xfs_inode *ip = XFS_I(inode); |
240 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 241 | size_t size = iov_iter_count(to); |
dda35b8f | 242 | ssize_t ret = 0; |
00258e36 | 243 | int ioflags = 0; |
dda35b8f | 244 | xfs_fsize_t n; |
b4f5d2c6 | 245 | loff_t pos = iocb->ki_pos; |
dda35b8f | 246 | |
dda35b8f CH |
247 | XFS_STATS_INC(xs_read_calls); |
248 | ||
00258e36 | 249 | if (unlikely(file->f_flags & O_DIRECT)) |
b92cc59f | 250 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 251 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 252 | ioflags |= XFS_IO_INVIS; |
00258e36 | 253 | |
b92cc59f | 254 | if (unlikely(ioflags & XFS_IO_ISDIRECT)) { |
dda35b8f CH |
255 | xfs_buftarg_t *target = |
256 | XFS_IS_REALTIME_INODE(ip) ? | |
257 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
258 | /* DIO must be aligned to device logical sector size */ |
259 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 260 | if (pos == i_size_read(inode)) |
00258e36 | 261 | return 0; |
b474c7ae | 262 | return -EINVAL; |
dda35b8f CH |
263 | } |
264 | } | |
265 | ||
fb595814 | 266 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 267 | if (n <= 0 || size == 0) |
dda35b8f CH |
268 | return 0; |
269 | ||
270 | if (n < size) | |
271 | size = n; | |
272 | ||
273 | if (XFS_FORCED_SHUTDOWN(mp)) | |
274 | return -EIO; | |
275 | ||
0c38a251 DC |
276 | /* |
277 | * Locking is a bit tricky here. If we take an exclusive lock | |
278 | * for direct IO, we effectively serialise all new concurrent | |
279 | * read IO to this file and block it behind IO that is currently in | |
280 | * progress because IO in progress holds the IO lock shared. We only | |
281 | * need to hold the lock exclusive to blow away the page cache, so | |
282 | * only take lock exclusively if the page cache needs invalidation. | |
283 | * This allows the normal direct IO case of no page cache pages to | |
284 | * proceeed concurrently without serialisation. | |
285 | */ | |
286 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 287 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 288 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
289 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
290 | ||
00258e36 | 291 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 292 | ret = filemap_write_and_wait_range( |
fb595814 | 293 | VFS_I(ip)->i_mapping, |
7d4ea3ce | 294 | pos, pos + size - 1); |
487f84f3 DC |
295 | if (ret) { |
296 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
297 | return ret; | |
298 | } | |
85e584da CM |
299 | |
300 | /* | |
301 | * Invalidate whole pages. This can return an error if | |
302 | * we fail to invalidate a page, but this should never | |
303 | * happen on XFS. Warn if it does fail. | |
304 | */ | |
305 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
306 | pos >> PAGE_CACHE_SHIFT, |
307 | (pos + size - 1) >> PAGE_CACHE_SHIFT); | |
85e584da CM |
308 | WARN_ON_ONCE(ret); |
309 | ret = 0; | |
00258e36 | 310 | } |
487f84f3 | 311 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 312 | } |
dda35b8f | 313 | |
fb595814 | 314 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 315 | |
b4f5d2c6 | 316 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
317 | if (ret > 0) |
318 | XFS_STATS_ADD(xs_read_bytes, ret); | |
319 | ||
487f84f3 | 320 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
321 | return ret; |
322 | } | |
323 | ||
00258e36 CH |
324 | STATIC ssize_t |
325 | xfs_file_splice_read( | |
dda35b8f CH |
326 | struct file *infilp, |
327 | loff_t *ppos, | |
328 | struct pipe_inode_info *pipe, | |
329 | size_t count, | |
00258e36 | 330 | unsigned int flags) |
dda35b8f | 331 | { |
00258e36 | 332 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 333 | int ioflags = 0; |
dda35b8f CH |
334 | ssize_t ret; |
335 | ||
336 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
337 | |
338 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 339 | ioflags |= XFS_IO_INVIS; |
00258e36 | 340 | |
dda35b8f CH |
341 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
342 | return -EIO; | |
343 | ||
487f84f3 | 344 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 345 | |
dda35b8f CH |
346 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
347 | ||
348 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
349 | if (ret > 0) | |
350 | XFS_STATS_ADD(xs_read_bytes, ret); | |
351 | ||
487f84f3 | 352 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
353 | return ret; |
354 | } | |
355 | ||
dda35b8f | 356 | /* |
193aec10 CH |
357 | * This routine is called to handle zeroing any space in the last block of the |
358 | * file that is beyond the EOF. We do this since the size is being increased | |
359 | * without writing anything to that block and we don't want to read the | |
360 | * garbage on the disk. | |
dda35b8f CH |
361 | */ |
362 | STATIC int /* error (positive) */ | |
363 | xfs_zero_last_block( | |
193aec10 CH |
364 | struct xfs_inode *ip, |
365 | xfs_fsize_t offset, | |
366 | xfs_fsize_t isize) | |
dda35b8f | 367 | { |
193aec10 CH |
368 | struct xfs_mount *mp = ip->i_mount; |
369 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
370 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
371 | int zero_len; | |
372 | int nimaps = 1; | |
373 | int error = 0; | |
374 | struct xfs_bmbt_irec imap; | |
dda35b8f | 375 | |
193aec10 | 376 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 377 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 378 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 379 | if (error) |
dda35b8f | 380 | return error; |
193aec10 | 381 | |
dda35b8f | 382 | ASSERT(nimaps > 0); |
193aec10 | 383 | |
dda35b8f CH |
384 | /* |
385 | * If the block underlying isize is just a hole, then there | |
386 | * is nothing to zero. | |
387 | */ | |
193aec10 | 388 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 389 | return 0; |
dda35b8f CH |
390 | |
391 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
392 | if (isize + zero_len > offset) | |
393 | zero_len = offset - isize; | |
193aec10 | 394 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
395 | } |
396 | ||
397 | /* | |
193aec10 CH |
398 | * Zero any on disk space between the current EOF and the new, larger EOF. |
399 | * | |
400 | * This handles the normal case of zeroing the remainder of the last block in | |
401 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
402 | * file. This second case only happens with fixed size extents and when the | |
403 | * system crashes before the inode size was updated but after blocks were | |
404 | * allocated. | |
405 | * | |
406 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 407 | */ |
dda35b8f CH |
408 | int /* error (positive) */ |
409 | xfs_zero_eof( | |
193aec10 CH |
410 | struct xfs_inode *ip, |
411 | xfs_off_t offset, /* starting I/O offset */ | |
412 | xfs_fsize_t isize) /* current inode size */ | |
dda35b8f | 413 | { |
193aec10 CH |
414 | struct xfs_mount *mp = ip->i_mount; |
415 | xfs_fileoff_t start_zero_fsb; | |
416 | xfs_fileoff_t end_zero_fsb; | |
417 | xfs_fileoff_t zero_count_fsb; | |
418 | xfs_fileoff_t last_fsb; | |
419 | xfs_fileoff_t zero_off; | |
420 | xfs_fsize_t zero_len; | |
421 | int nimaps; | |
422 | int error = 0; | |
423 | struct xfs_bmbt_irec imap; | |
424 | ||
425 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
426 | ASSERT(offset > isize); |
427 | ||
428 | /* | |
429 | * First handle zeroing the block on which isize resides. | |
193aec10 | 430 | * |
dda35b8f CH |
431 | * We only zero a part of that block so it is handled specially. |
432 | */ | |
193aec10 CH |
433 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
434 | error = xfs_zero_last_block(ip, offset, isize); | |
435 | if (error) | |
436 | return error; | |
dda35b8f CH |
437 | } |
438 | ||
439 | /* | |
193aec10 CH |
440 | * Calculate the range between the new size and the old where blocks |
441 | * needing to be zeroed may exist. | |
442 | * | |
443 | * To get the block where the last byte in the file currently resides, | |
444 | * we need to subtract one from the size and truncate back to a block | |
445 | * boundary. We subtract 1 in case the size is exactly on a block | |
446 | * boundary. | |
dda35b8f CH |
447 | */ |
448 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
449 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
450 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
451 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
452 | if (last_fsb == end_zero_fsb) { | |
453 | /* | |
454 | * The size was only incremented on its last block. | |
455 | * We took care of that above, so just return. | |
456 | */ | |
457 | return 0; | |
458 | } | |
459 | ||
460 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
461 | while (start_zero_fsb <= end_zero_fsb) { | |
462 | nimaps = 1; | |
463 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
464 | |
465 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
466 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
467 | &imap, &nimaps, 0); | |
193aec10 CH |
468 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
469 | if (error) | |
dda35b8f | 470 | return error; |
193aec10 | 471 | |
dda35b8f CH |
472 | ASSERT(nimaps > 0); |
473 | ||
474 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
475 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
476 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
477 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
478 | continue; | |
479 | } | |
480 | ||
481 | /* | |
482 | * There are blocks we need to zero. | |
dda35b8f | 483 | */ |
dda35b8f CH |
484 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
485 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
486 | ||
487 | if ((zero_off + zero_len) > offset) | |
488 | zero_len = offset - zero_off; | |
489 | ||
490 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
491 | if (error) |
492 | return error; | |
dda35b8f CH |
493 | |
494 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
495 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
496 | } |
497 | ||
498 | return 0; | |
dda35b8f CH |
499 | } |
500 | ||
4d8d1581 DC |
501 | /* |
502 | * Common pre-write limit and setup checks. | |
503 | * | |
5bf1f262 CH |
504 | * Called with the iolocked held either shared and exclusive according to |
505 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
506 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
507 | */ |
508 | STATIC ssize_t | |
509 | xfs_file_aio_write_checks( | |
510 | struct file *file, | |
511 | loff_t *pos, | |
512 | size_t *count, | |
513 | int *iolock) | |
514 | { | |
515 | struct inode *inode = file->f_mapping->host; | |
516 | struct xfs_inode *ip = XFS_I(inode); | |
4d8d1581 DC |
517 | int error = 0; |
518 | ||
7271d243 | 519 | restart: |
4d8d1581 | 520 | error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); |
467f7899 | 521 | if (error) |
4d8d1581 | 522 | return error; |
4d8d1581 | 523 | |
4d8d1581 DC |
524 | /* |
525 | * If the offset is beyond the size of the file, we need to zero any | |
526 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 527 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
528 | * iolock shared, we need to update it to exclusive which implies |
529 | * having to redo all checks before. | |
4d8d1581 | 530 | */ |
2813d682 | 531 | if (*pos > i_size_read(inode)) { |
7271d243 | 532 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 533 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 534 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 535 | xfs_rw_ilock(ip, *iolock); |
7271d243 DC |
536 | goto restart; |
537 | } | |
2451337d | 538 | error = xfs_zero_eof(ip, *pos, i_size_read(inode)); |
467f7899 CH |
539 | if (error) |
540 | return error; | |
7271d243 | 541 | } |
4d8d1581 | 542 | |
8a9c9980 CH |
543 | /* |
544 | * Updating the timestamps will grab the ilock again from | |
545 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
546 | * lock above. Eventually we should look into a way to avoid | |
547 | * the pointless lock roundtrip. | |
548 | */ | |
c3b2da31 JB |
549 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
550 | error = file_update_time(file); | |
551 | if (error) | |
552 | return error; | |
553 | } | |
8a9c9980 | 554 | |
4d8d1581 DC |
555 | /* |
556 | * If we're writing the file then make sure to clear the setuid and | |
557 | * setgid bits if the process is not being run by root. This keeps | |
558 | * people from modifying setuid and setgid binaries. | |
559 | */ | |
560 | return file_remove_suid(file); | |
4d8d1581 DC |
561 | } |
562 | ||
f0d26e86 DC |
563 | /* |
564 | * xfs_file_dio_aio_write - handle direct IO writes | |
565 | * | |
566 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 567 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
568 | * follow locking changes and looping. |
569 | * | |
eda77982 DC |
570 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
571 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
572 | * pages are flushed out. | |
573 | * | |
574 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
575 | * allowing them to be done in parallel with reads and other direct IO writes. | |
576 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
577 | * needs to do sub-block zeroing and that requires serialisation against other | |
578 | * direct IOs to the same block. In this case we need to serialise the | |
579 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
580 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
581 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
582 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 583 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 584 | * |
f0d26e86 DC |
585 | * Returns with locks held indicated by @iolock and errors indicated by |
586 | * negative return values. | |
587 | */ | |
588 | STATIC ssize_t | |
589 | xfs_file_dio_aio_write( | |
590 | struct kiocb *iocb, | |
b3188919 | 591 | struct iov_iter *from) |
f0d26e86 DC |
592 | { |
593 | struct file *file = iocb->ki_filp; | |
594 | struct address_space *mapping = file->f_mapping; | |
595 | struct inode *inode = mapping->host; | |
596 | struct xfs_inode *ip = XFS_I(inode); | |
597 | struct xfs_mount *mp = ip->i_mount; | |
598 | ssize_t ret = 0; | |
eda77982 | 599 | int unaligned_io = 0; |
d0606464 | 600 | int iolock; |
b3188919 AV |
601 | size_t count = iov_iter_count(from); |
602 | loff_t pos = iocb->ki_pos; | |
f0d26e86 DC |
603 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
604 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
605 | ||
7c71ee78 ES |
606 | /* DIO must be aligned to device logical sector size */ |
607 | if ((pos | count) & target->bt_logical_sectormask) | |
b474c7ae | 608 | return -EINVAL; |
f0d26e86 | 609 | |
7c71ee78 | 610 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
611 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
612 | unaligned_io = 1; | |
613 | ||
7271d243 DC |
614 | /* |
615 | * We don't need to take an exclusive lock unless there page cache needs | |
616 | * to be invalidated or unaligned IO is being executed. We don't need to | |
617 | * consider the EOF extension case here because | |
618 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
619 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
620 | */ | |
621 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 622 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 623 | else |
d0606464 CH |
624 | iolock = XFS_IOLOCK_SHARED; |
625 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
626 | |
627 | /* | |
628 | * Recheck if there are cached pages that need invalidate after we got | |
629 | * the iolock to protect against other threads adding new pages while | |
630 | * we were waiting for the iolock. | |
631 | */ | |
d0606464 CH |
632 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
633 | xfs_rw_iunlock(ip, iolock); | |
634 | iolock = XFS_IOLOCK_EXCL; | |
635 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 636 | } |
f0d26e86 | 637 | |
d0606464 | 638 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 639 | if (ret) |
d0606464 | 640 | goto out; |
b3188919 | 641 | iov_iter_truncate(from, count); |
f0d26e86 DC |
642 | |
643 | if (mapping->nrpages) { | |
07d5035a | 644 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
7d4ea3ce | 645 | pos, pos + count - 1); |
f0d26e86 | 646 | if (ret) |
d0606464 | 647 | goto out; |
834ffca6 DC |
648 | /* |
649 | * Invalidate whole pages. This can return an error if | |
650 | * we fail to invalidate a page, but this should never | |
651 | * happen on XFS. Warn if it does fail. | |
652 | */ | |
653 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
654 | pos >> PAGE_CACHE_SHIFT, |
655 | (pos + count - 1) >> PAGE_CACHE_SHIFT); | |
834ffca6 DC |
656 | WARN_ON_ONCE(ret); |
657 | ret = 0; | |
f0d26e86 DC |
658 | } |
659 | ||
eda77982 DC |
660 | /* |
661 | * If we are doing unaligned IO, wait for all other IO to drain, | |
662 | * otherwise demote the lock if we had to flush cached pages | |
663 | */ | |
664 | if (unaligned_io) | |
4a06fd26 | 665 | inode_dio_wait(inode); |
d0606464 | 666 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 667 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 668 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
669 | } |
670 | ||
671 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
b3188919 | 672 | ret = generic_file_direct_write(iocb, from, pos); |
f0d26e86 | 673 | |
d0606464 CH |
674 | out: |
675 | xfs_rw_iunlock(ip, iolock); | |
676 | ||
f0d26e86 DC |
677 | /* No fallback to buffered IO on errors for XFS. */ |
678 | ASSERT(ret < 0 || ret == count); | |
679 | return ret; | |
680 | } | |
681 | ||
00258e36 | 682 | STATIC ssize_t |
637bbc75 | 683 | xfs_file_buffered_aio_write( |
dda35b8f | 684 | struct kiocb *iocb, |
b3188919 | 685 | struct iov_iter *from) |
dda35b8f CH |
686 | { |
687 | struct file *file = iocb->ki_filp; | |
688 | struct address_space *mapping = file->f_mapping; | |
689 | struct inode *inode = mapping->host; | |
00258e36 | 690 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
691 | ssize_t ret; |
692 | int enospc = 0; | |
d0606464 | 693 | int iolock = XFS_IOLOCK_EXCL; |
b3188919 AV |
694 | loff_t pos = iocb->ki_pos; |
695 | size_t count = iov_iter_count(from); | |
dda35b8f | 696 | |
d0606464 | 697 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 698 | |
d0606464 | 699 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 700 | if (ret) |
d0606464 | 701 | goto out; |
dda35b8f | 702 | |
b3188919 | 703 | iov_iter_truncate(from, count); |
dda35b8f CH |
704 | /* We can write back this queue in page reclaim */ |
705 | current->backing_dev_info = mapping->backing_dev_info; | |
706 | ||
dda35b8f | 707 | write_retry: |
637bbc75 | 708 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); |
b3188919 | 709 | ret = generic_perform_write(file, from, pos); |
0a64bc2c AV |
710 | if (likely(ret >= 0)) |
711 | iocb->ki_pos = pos + ret; | |
dc06f398 | 712 | |
637bbc75 | 713 | /* |
dc06f398 BF |
714 | * If we hit a space limit, try to free up some lingering preallocated |
715 | * space before returning an error. In the case of ENOSPC, first try to | |
716 | * write back all dirty inodes to free up some of the excess reserved | |
717 | * metadata space. This reduces the chances that the eofblocks scan | |
718 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
719 | * also behaves as a filter to prevent too many eofblocks scans from | |
720 | * running at the same time. | |
637bbc75 | 721 | */ |
dc06f398 BF |
722 | if (ret == -EDQUOT && !enospc) { |
723 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
724 | if (enospc) | |
725 | goto write_retry; | |
726 | } else if (ret == -ENOSPC && !enospc) { | |
727 | struct xfs_eofblocks eofb = {0}; | |
728 | ||
637bbc75 | 729 | enospc = 1; |
9aa05000 | 730 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
731 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
732 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
733 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 734 | goto write_retry; |
dda35b8f | 735 | } |
d0606464 | 736 | |
dda35b8f | 737 | current->backing_dev_info = NULL; |
d0606464 CH |
738 | out: |
739 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
740 | return ret; |
741 | } | |
742 | ||
743 | STATIC ssize_t | |
bf97f3bc | 744 | xfs_file_write_iter( |
637bbc75 | 745 | struct kiocb *iocb, |
bf97f3bc | 746 | struct iov_iter *from) |
637bbc75 DC |
747 | { |
748 | struct file *file = iocb->ki_filp; | |
749 | struct address_space *mapping = file->f_mapping; | |
750 | struct inode *inode = mapping->host; | |
751 | struct xfs_inode *ip = XFS_I(inode); | |
752 | ssize_t ret; | |
bf97f3bc | 753 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
754 | |
755 | XFS_STATS_INC(xs_write_calls); | |
756 | ||
637bbc75 DC |
757 | if (ocount == 0) |
758 | return 0; | |
759 | ||
bf97f3bc AV |
760 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
761 | return -EIO; | |
637bbc75 DC |
762 | |
763 | if (unlikely(file->f_flags & O_DIRECT)) | |
bf97f3bc | 764 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 765 | else |
bf97f3bc | 766 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 767 | |
d0606464 CH |
768 | if (ret > 0) { |
769 | ssize_t err; | |
dda35b8f | 770 | |
d0606464 | 771 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 772 | |
d0606464 | 773 | /* Handle various SYNC-type writes */ |
d311d79d | 774 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
775 | if (err < 0) |
776 | ret = err; | |
dda35b8f | 777 | } |
a363f0c2 | 778 | return ret; |
dda35b8f CH |
779 | } |
780 | ||
2fe17c10 CH |
781 | STATIC long |
782 | xfs_file_fallocate( | |
83aee9e4 CH |
783 | struct file *file, |
784 | int mode, | |
785 | loff_t offset, | |
786 | loff_t len) | |
2fe17c10 | 787 | { |
83aee9e4 CH |
788 | struct inode *inode = file_inode(file); |
789 | struct xfs_inode *ip = XFS_I(inode); | |
790 | struct xfs_trans *tp; | |
791 | long error; | |
792 | loff_t new_size = 0; | |
2fe17c10 | 793 | |
83aee9e4 CH |
794 | if (!S_ISREG(inode->i_mode)) |
795 | return -EINVAL; | |
e1d8fb88 | 796 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
376ba313 | 797 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
2fe17c10 CH |
798 | return -EOPNOTSUPP; |
799 | ||
2fe17c10 | 800 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
83aee9e4 CH |
801 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
802 | error = xfs_free_file_space(ip, offset, len); | |
803 | if (error) | |
804 | goto out_unlock; | |
e1d8fb88 NJ |
805 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
806 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
807 | ||
808 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 809 | error = -EINVAL; |
e1d8fb88 NJ |
810 | goto out_unlock; |
811 | } | |
812 | ||
23fffa92 LC |
813 | /* |
814 | * There is no need to overlap collapse range with EOF, | |
815 | * in which case it is effectively a truncate operation | |
816 | */ | |
817 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 818 | error = -EINVAL; |
23fffa92 LC |
819 | goto out_unlock; |
820 | } | |
821 | ||
e1d8fb88 NJ |
822 | new_size = i_size_read(inode) - len; |
823 | ||
824 | error = xfs_collapse_file_space(ip, offset, len); | |
825 | if (error) | |
826 | goto out_unlock; | |
83aee9e4 CH |
827 | } else { |
828 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
829 | offset + len > i_size_read(inode)) { | |
830 | new_size = offset + len; | |
2451337d | 831 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
832 | if (error) |
833 | goto out_unlock; | |
834 | } | |
2fe17c10 | 835 | |
376ba313 LC |
836 | if (mode & FALLOC_FL_ZERO_RANGE) |
837 | error = xfs_zero_file_space(ip, offset, len); | |
838 | else | |
839 | error = xfs_alloc_file_space(ip, offset, len, | |
840 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
841 | if (error) |
842 | goto out_unlock; | |
843 | } | |
844 | ||
83aee9e4 CH |
845 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); |
846 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
847 | if (error) { | |
848 | xfs_trans_cancel(tp, 0); | |
849 | goto out_unlock; | |
850 | } | |
851 | ||
852 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
853 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
854 | ip->i_d.di_mode &= ~S_ISUID; | |
855 | if (ip->i_d.di_mode & S_IXGRP) | |
856 | ip->i_d.di_mode &= ~S_ISGID; | |
82878897 | 857 | |
e1d8fb88 | 858 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE))) |
83aee9e4 CH |
859 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; |
860 | ||
861 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
862 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
863 | ||
864 | if (file->f_flags & O_DSYNC) | |
865 | xfs_trans_set_sync(tp); | |
866 | error = xfs_trans_commit(tp, 0); | |
2fe17c10 CH |
867 | if (error) |
868 | goto out_unlock; | |
869 | ||
870 | /* Change file size if needed */ | |
871 | if (new_size) { | |
872 | struct iattr iattr; | |
873 | ||
874 | iattr.ia_valid = ATTR_SIZE; | |
875 | iattr.ia_size = new_size; | |
83aee9e4 | 876 | error = xfs_setattr_size(ip, &iattr); |
2fe17c10 CH |
877 | } |
878 | ||
879 | out_unlock: | |
880 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2451337d | 881 | return error; |
2fe17c10 CH |
882 | } |
883 | ||
884 | ||
1da177e4 | 885 | STATIC int |
3562fd45 | 886 | xfs_file_open( |
1da177e4 | 887 | struct inode *inode, |
f999a5bf | 888 | struct file *file) |
1da177e4 | 889 | { |
f999a5bf | 890 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 891 | return -EFBIG; |
f999a5bf CH |
892 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
893 | return -EIO; | |
894 | return 0; | |
895 | } | |
896 | ||
897 | STATIC int | |
898 | xfs_dir_open( | |
899 | struct inode *inode, | |
900 | struct file *file) | |
901 | { | |
902 | struct xfs_inode *ip = XFS_I(inode); | |
903 | int mode; | |
904 | int error; | |
905 | ||
906 | error = xfs_file_open(inode, file); | |
907 | if (error) | |
908 | return error; | |
909 | ||
910 | /* | |
911 | * If there are any blocks, read-ahead block 0 as we're almost | |
912 | * certain to have the next operation be a read there. | |
913 | */ | |
309ecac8 | 914 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 915 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 916 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
917 | xfs_iunlock(ip, mode); |
918 | return 0; | |
1da177e4 LT |
919 | } |
920 | ||
1da177e4 | 921 | STATIC int |
3562fd45 | 922 | xfs_file_release( |
1da177e4 LT |
923 | struct inode *inode, |
924 | struct file *filp) | |
925 | { | |
2451337d | 926 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
927 | } |
928 | ||
1da177e4 | 929 | STATIC int |
3562fd45 | 930 | xfs_file_readdir( |
b8227554 AV |
931 | struct file *file, |
932 | struct dir_context *ctx) | |
1da177e4 | 933 | { |
b8227554 | 934 | struct inode *inode = file_inode(file); |
739bfb2a | 935 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
936 | int error; |
937 | size_t bufsize; | |
938 | ||
939 | /* | |
940 | * The Linux API doesn't pass down the total size of the buffer | |
941 | * we read into down to the filesystem. With the filldir concept | |
942 | * it's not needed for correct information, but the XFS dir2 leaf | |
943 | * code wants an estimate of the buffer size to calculate it's | |
944 | * readahead window and size the buffers used for mapping to | |
945 | * physical blocks. | |
946 | * | |
947 | * Try to give it an estimate that's good enough, maybe at some | |
948 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 949 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 950 | */ |
a9cc799e | 951 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 952 | |
b8227554 | 953 | error = xfs_readdir(ip, ctx, bufsize); |
051e7cd4 | 954 | if (error) |
2451337d | 955 | return error; |
051e7cd4 | 956 | return 0; |
1da177e4 LT |
957 | } |
958 | ||
1da177e4 | 959 | STATIC int |
3562fd45 | 960 | xfs_file_mmap( |
1da177e4 LT |
961 | struct file *filp, |
962 | struct vm_area_struct *vma) | |
963 | { | |
3562fd45 | 964 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 965 | |
fbc1462b | 966 | file_accessed(filp); |
1da177e4 LT |
967 | return 0; |
968 | } | |
969 | ||
4f57dbc6 DC |
970 | /* |
971 | * mmap()d file has taken write protection fault and is being made | |
972 | * writable. We can set the page state up correctly for a writable | |
973 | * page, which means we can do correct delalloc accounting (ENOSPC | |
974 | * checking!) and unwritten extent mapping. | |
975 | */ | |
976 | STATIC int | |
977 | xfs_vm_page_mkwrite( | |
978 | struct vm_area_struct *vma, | |
c2ec175c | 979 | struct vm_fault *vmf) |
4f57dbc6 | 980 | { |
c2ec175c | 981 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
982 | } |
983 | ||
d126d43f JL |
984 | /* |
985 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 986 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
987 | */ |
988 | enum { | |
989 | HOLE_OFF = 0, | |
990 | DATA_OFF, | |
991 | }; | |
992 | ||
993 | /* | |
994 | * Lookup the desired type of offset from the given page. | |
995 | * | |
996 | * On success, return true and the offset argument will point to the | |
997 | * start of the region that was found. Otherwise this function will | |
998 | * return false and keep the offset argument unchanged. | |
999 | */ | |
1000 | STATIC bool | |
1001 | xfs_lookup_buffer_offset( | |
1002 | struct page *page, | |
1003 | loff_t *offset, | |
1004 | unsigned int type) | |
1005 | { | |
1006 | loff_t lastoff = page_offset(page); | |
1007 | bool found = false; | |
1008 | struct buffer_head *bh, *head; | |
1009 | ||
1010 | bh = head = page_buffers(page); | |
1011 | do { | |
1012 | /* | |
1013 | * Unwritten extents that have data in the page | |
1014 | * cache covering them can be identified by the | |
1015 | * BH_Unwritten state flag. Pages with multiple | |
1016 | * buffers might have a mix of holes, data and | |
1017 | * unwritten extents - any buffer with valid | |
1018 | * data in it should have BH_Uptodate flag set | |
1019 | * on it. | |
1020 | */ | |
1021 | if (buffer_unwritten(bh) || | |
1022 | buffer_uptodate(bh)) { | |
1023 | if (type == DATA_OFF) | |
1024 | found = true; | |
1025 | } else { | |
1026 | if (type == HOLE_OFF) | |
1027 | found = true; | |
1028 | } | |
1029 | ||
1030 | if (found) { | |
1031 | *offset = lastoff; | |
1032 | break; | |
1033 | } | |
1034 | lastoff += bh->b_size; | |
1035 | } while ((bh = bh->b_this_page) != head); | |
1036 | ||
1037 | return found; | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * This routine is called to find out and return a data or hole offset | |
1042 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1043 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1044 | * |
1045 | * The argument offset is used to tell where we start to search from the | |
1046 | * page cache. Map is used to figure out the end points of the range to | |
1047 | * lookup pages. | |
1048 | * | |
1049 | * Return true if the desired type of offset was found, and the argument | |
1050 | * offset is filled with that address. Otherwise, return false and keep | |
1051 | * offset unchanged. | |
1052 | */ | |
1053 | STATIC bool | |
1054 | xfs_find_get_desired_pgoff( | |
1055 | struct inode *inode, | |
1056 | struct xfs_bmbt_irec *map, | |
1057 | unsigned int type, | |
1058 | loff_t *offset) | |
1059 | { | |
1060 | struct xfs_inode *ip = XFS_I(inode); | |
1061 | struct xfs_mount *mp = ip->i_mount; | |
1062 | struct pagevec pvec; | |
1063 | pgoff_t index; | |
1064 | pgoff_t end; | |
1065 | loff_t endoff; | |
1066 | loff_t startoff = *offset; | |
1067 | loff_t lastoff = startoff; | |
1068 | bool found = false; | |
1069 | ||
1070 | pagevec_init(&pvec, 0); | |
1071 | ||
1072 | index = startoff >> PAGE_CACHE_SHIFT; | |
1073 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1074 | end = endoff >> PAGE_CACHE_SHIFT; | |
1075 | do { | |
1076 | int want; | |
1077 | unsigned nr_pages; | |
1078 | unsigned int i; | |
1079 | ||
1080 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1081 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1082 | want); | |
1083 | /* | |
1084 | * No page mapped into given range. If we are searching holes | |
1085 | * and if this is the first time we got into the loop, it means | |
1086 | * that the given offset is landed in a hole, return it. | |
1087 | * | |
1088 | * If we have already stepped through some block buffers to find | |
1089 | * holes but they all contains data. In this case, the last | |
1090 | * offset is already updated and pointed to the end of the last | |
1091 | * mapped page, if it does not reach the endpoint to search, | |
1092 | * that means there should be a hole between them. | |
1093 | */ | |
1094 | if (nr_pages == 0) { | |
1095 | /* Data search found nothing */ | |
1096 | if (type == DATA_OFF) | |
1097 | break; | |
1098 | ||
1099 | ASSERT(type == HOLE_OFF); | |
1100 | if (lastoff == startoff || lastoff < endoff) { | |
1101 | found = true; | |
1102 | *offset = lastoff; | |
1103 | } | |
1104 | break; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * At lease we found one page. If this is the first time we | |
1109 | * step into the loop, and if the first page index offset is | |
1110 | * greater than the given search offset, a hole was found. | |
1111 | */ | |
1112 | if (type == HOLE_OFF && lastoff == startoff && | |
1113 | lastoff < page_offset(pvec.pages[0])) { | |
1114 | found = true; | |
1115 | break; | |
1116 | } | |
1117 | ||
1118 | for (i = 0; i < nr_pages; i++) { | |
1119 | struct page *page = pvec.pages[i]; | |
1120 | loff_t b_offset; | |
1121 | ||
1122 | /* | |
1123 | * At this point, the page may be truncated or | |
1124 | * invalidated (changing page->mapping to NULL), | |
1125 | * or even swizzled back from swapper_space to tmpfs | |
1126 | * file mapping. However, page->index will not change | |
1127 | * because we have a reference on the page. | |
1128 | * | |
1129 | * Searching done if the page index is out of range. | |
1130 | * If the current offset is not reaches the end of | |
1131 | * the specified search range, there should be a hole | |
1132 | * between them. | |
1133 | */ | |
1134 | if (page->index > end) { | |
1135 | if (type == HOLE_OFF && lastoff < endoff) { | |
1136 | *offset = lastoff; | |
1137 | found = true; | |
1138 | } | |
1139 | goto out; | |
1140 | } | |
1141 | ||
1142 | lock_page(page); | |
1143 | /* | |
1144 | * Page truncated or invalidated(page->mapping == NULL). | |
1145 | * We can freely skip it and proceed to check the next | |
1146 | * page. | |
1147 | */ | |
1148 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1149 | unlock_page(page); | |
1150 | continue; | |
1151 | } | |
1152 | ||
1153 | if (!page_has_buffers(page)) { | |
1154 | unlock_page(page); | |
1155 | continue; | |
1156 | } | |
1157 | ||
1158 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1159 | if (found) { | |
1160 | /* | |
1161 | * The found offset may be less than the start | |
1162 | * point to search if this is the first time to | |
1163 | * come here. | |
1164 | */ | |
1165 | *offset = max_t(loff_t, startoff, b_offset); | |
1166 | unlock_page(page); | |
1167 | goto out; | |
1168 | } | |
1169 | ||
1170 | /* | |
1171 | * We either searching data but nothing was found, or | |
1172 | * searching hole but found a data buffer. In either | |
1173 | * case, probably the next page contains the desired | |
1174 | * things, update the last offset to it so. | |
1175 | */ | |
1176 | lastoff = page_offset(page) + PAGE_SIZE; | |
1177 | unlock_page(page); | |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * The number of returned pages less than our desired, search | |
1182 | * done. In this case, nothing was found for searching data, | |
1183 | * but we found a hole behind the last offset. | |
1184 | */ | |
1185 | if (nr_pages < want) { | |
1186 | if (type == HOLE_OFF) { | |
1187 | *offset = lastoff; | |
1188 | found = true; | |
1189 | } | |
1190 | break; | |
1191 | } | |
1192 | ||
1193 | index = pvec.pages[i - 1]->index + 1; | |
1194 | pagevec_release(&pvec); | |
1195 | } while (index <= end); | |
1196 | ||
1197 | out: | |
1198 | pagevec_release(&pvec); | |
1199 | return found; | |
1200 | } | |
1201 | ||
3fe3e6b1 | 1202 | STATIC loff_t |
49c69591 | 1203 | xfs_seek_hole_data( |
3fe3e6b1 | 1204 | struct file *file, |
49c69591 ES |
1205 | loff_t start, |
1206 | int whence) | |
3fe3e6b1 JL |
1207 | { |
1208 | struct inode *inode = file->f_mapping->host; | |
1209 | struct xfs_inode *ip = XFS_I(inode); | |
1210 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1211 | loff_t uninitialized_var(offset); |
1212 | xfs_fsize_t isize; | |
1213 | xfs_fileoff_t fsbno; | |
1214 | xfs_filblks_t end; | |
1215 | uint lock; | |
1216 | int error; | |
1217 | ||
49c69591 ES |
1218 | if (XFS_FORCED_SHUTDOWN(mp)) |
1219 | return -EIO; | |
1220 | ||
309ecac8 | 1221 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1222 | |
1223 | isize = i_size_read(inode); | |
1224 | if (start >= isize) { | |
2451337d | 1225 | error = -ENXIO; |
3fe3e6b1 JL |
1226 | goto out_unlock; |
1227 | } | |
1228 | ||
3fe3e6b1 JL |
1229 | /* |
1230 | * Try to read extents from the first block indicated | |
1231 | * by fsbno to the end block of the file. | |
1232 | */ | |
52f1acc8 | 1233 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1234 | end = XFS_B_TO_FSB(mp, isize); |
49c69591 | 1235 | |
52f1acc8 JL |
1236 | for (;;) { |
1237 | struct xfs_bmbt_irec map[2]; | |
1238 | int nmap = 2; | |
1239 | unsigned int i; | |
3fe3e6b1 | 1240 | |
52f1acc8 JL |
1241 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1242 | XFS_BMAPI_ENTIRE); | |
1243 | if (error) | |
1244 | goto out_unlock; | |
3fe3e6b1 | 1245 | |
52f1acc8 JL |
1246 | /* No extents at given offset, must be beyond EOF */ |
1247 | if (nmap == 0) { | |
2451337d | 1248 | error = -ENXIO; |
52f1acc8 JL |
1249 | goto out_unlock; |
1250 | } | |
1251 | ||
1252 | for (i = 0; i < nmap; i++) { | |
1253 | offset = max_t(loff_t, start, | |
1254 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1255 | ||
49c69591 ES |
1256 | /* Landed in the hole we wanted? */ |
1257 | if (whence == SEEK_HOLE && | |
1258 | map[i].br_startblock == HOLESTARTBLOCK) | |
1259 | goto out; | |
1260 | ||
1261 | /* Landed in the data extent we wanted? */ | |
1262 | if (whence == SEEK_DATA && | |
1263 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1264 | (map[i].br_state == XFS_EXT_NORM && | |
1265 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1266 | goto out; |
1267 | ||
1268 | /* | |
49c69591 ES |
1269 | * Landed in an unwritten extent, try to search |
1270 | * for hole or data from page cache. | |
52f1acc8 JL |
1271 | */ |
1272 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1273 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1274 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1275 | &offset)) | |
52f1acc8 JL |
1276 | goto out; |
1277 | } | |
1278 | } | |
1279 | ||
1280 | /* | |
49c69591 ES |
1281 | * We only received one extent out of the two requested. This |
1282 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1283 | */ |
3fe3e6b1 | 1284 | if (nmap == 1) { |
49c69591 ES |
1285 | /* |
1286 | * If we were looking for a hole, set offset to | |
1287 | * the end of the file (i.e., there is an implicit | |
1288 | * hole at the end of any file). | |
1289 | */ | |
1290 | if (whence == SEEK_HOLE) { | |
1291 | offset = isize; | |
1292 | break; | |
1293 | } | |
1294 | /* | |
1295 | * If we were looking for data, it's nowhere to be found | |
1296 | */ | |
1297 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1298 | error = -ENXIO; |
3fe3e6b1 JL |
1299 | goto out_unlock; |
1300 | } | |
1301 | ||
52f1acc8 JL |
1302 | ASSERT(i > 1); |
1303 | ||
1304 | /* | |
1305 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1306 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1307 | */ |
1308 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1309 | start = XFS_FSB_TO_B(mp, fsbno); | |
1310 | if (start >= isize) { | |
49c69591 ES |
1311 | if (whence == SEEK_HOLE) { |
1312 | offset = isize; | |
1313 | break; | |
1314 | } | |
1315 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1316 | error = -ENXIO; |
52f1acc8 JL |
1317 | goto out_unlock; |
1318 | } | |
3fe3e6b1 JL |
1319 | } |
1320 | ||
b686d1f7 JL |
1321 | out: |
1322 | /* | |
49c69591 | 1323 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1324 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1325 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1326 | * situation in particular. |
1327 | */ | |
49c69591 ES |
1328 | if (whence == SEEK_HOLE) |
1329 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1330 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1331 | |
1332 | out_unlock: | |
01f4f327 | 1333 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1334 | |
1335 | if (error) | |
2451337d | 1336 | return error; |
3fe3e6b1 JL |
1337 | return offset; |
1338 | } | |
1339 | ||
1340 | STATIC loff_t | |
1341 | xfs_file_llseek( | |
1342 | struct file *file, | |
1343 | loff_t offset, | |
59f9c004 | 1344 | int whence) |
3fe3e6b1 | 1345 | { |
59f9c004 | 1346 | switch (whence) { |
3fe3e6b1 JL |
1347 | case SEEK_END: |
1348 | case SEEK_CUR: | |
1349 | case SEEK_SET: | |
59f9c004 | 1350 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1351 | case SEEK_HOLE: |
49c69591 | 1352 | case SEEK_DATA: |
59f9c004 | 1353 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1354 | default: |
1355 | return -EINVAL; | |
1356 | } | |
1357 | } | |
1358 | ||
4b6f5d20 | 1359 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1360 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1361 | .read = new_sync_read, |
bf97f3bc | 1362 | .write = new_sync_write, |
b4f5d2c6 | 1363 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1364 | .write_iter = xfs_file_write_iter, |
1b895840 | 1365 | .splice_read = xfs_file_splice_read, |
8d020765 | 1366 | .splice_write = iter_file_splice_write, |
3562fd45 | 1367 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1368 | #ifdef CONFIG_COMPAT |
3562fd45 | 1369 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1370 | #endif |
3562fd45 NS |
1371 | .mmap = xfs_file_mmap, |
1372 | .open = xfs_file_open, | |
1373 | .release = xfs_file_release, | |
1374 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1375 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1376 | }; |
1377 | ||
4b6f5d20 | 1378 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1379 | .open = xfs_dir_open, |
1da177e4 | 1380 | .read = generic_read_dir, |
b8227554 | 1381 | .iterate = xfs_file_readdir, |
59af1584 | 1382 | .llseek = generic_file_llseek, |
3562fd45 | 1383 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1384 | #ifdef CONFIG_COMPAT |
3562fd45 | 1385 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1386 | #endif |
1da2f2db | 1387 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1388 | }; |
1389 | ||
f0f37e2f | 1390 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1391 | .fault = filemap_fault, |
f1820361 | 1392 | .map_pages = filemap_map_pages, |
4f57dbc6 | 1393 | .page_mkwrite = xfs_vm_page_mkwrite, |
0b173bc4 | 1394 | .remap_pages = generic_file_remap_pages, |
6fac0cb4 | 1395 | }; |