Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
dda35b8f | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
a4fbe6ab | 21 | #include "xfs_format.h" |
239880ef DC |
22 | #include "xfs_log_format.h" |
23 | #include "xfs_trans_resv.h" | |
1da177e4 | 24 | #include "xfs_mount.h" |
57062787 DC |
25 | #include "xfs_da_format.h" |
26 | #include "xfs_da_btree.h" | |
1da177e4 | 27 | #include "xfs_inode.h" |
239880ef | 28 | #include "xfs_trans.h" |
fd3200be | 29 | #include "xfs_inode_item.h" |
dda35b8f | 30 | #include "xfs_bmap.h" |
c24b5dfa | 31 | #include "xfs_bmap_util.h" |
1da177e4 | 32 | #include "xfs_error.h" |
2b9ab5ab | 33 | #include "xfs_dir2.h" |
c24b5dfa | 34 | #include "xfs_dir2_priv.h" |
ddcd856d | 35 | #include "xfs_ioctl.h" |
dda35b8f | 36 | #include "xfs_trace.h" |
239880ef | 37 | #include "xfs_log.h" |
dc06f398 | 38 | #include "xfs_icache.h" |
781355c6 | 39 | #include "xfs_pnfs.h" |
1da177e4 | 40 | |
a27bb332 | 41 | #include <linux/aio.h> |
1da177e4 | 42 | #include <linux/dcache.h> |
2fe17c10 | 43 | #include <linux/falloc.h> |
d126d43f | 44 | #include <linux/pagevec.h> |
1da177e4 | 45 | |
f0f37e2f | 46 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 47 | |
487f84f3 DC |
48 | /* |
49 | * Locking primitives for read and write IO paths to ensure we consistently use | |
50 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
51 | */ | |
52 | static inline void | |
53 | xfs_rw_ilock( | |
54 | struct xfs_inode *ip, | |
55 | int type) | |
56 | { | |
57 | if (type & XFS_IOLOCK_EXCL) | |
58 | mutex_lock(&VFS_I(ip)->i_mutex); | |
59 | xfs_ilock(ip, type); | |
60 | } | |
61 | ||
62 | static inline void | |
63 | xfs_rw_iunlock( | |
64 | struct xfs_inode *ip, | |
65 | int type) | |
66 | { | |
67 | xfs_iunlock(ip, type); | |
68 | if (type & XFS_IOLOCK_EXCL) | |
69 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
70 | } | |
71 | ||
72 | static inline void | |
73 | xfs_rw_ilock_demote( | |
74 | struct xfs_inode *ip, | |
75 | int type) | |
76 | { | |
77 | xfs_ilock_demote(ip, type); | |
78 | if (type & XFS_IOLOCK_EXCL) | |
79 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
80 | } | |
81 | ||
dda35b8f CH |
82 | /* |
83 | * xfs_iozero | |
84 | * | |
85 | * xfs_iozero clears the specified range of buffer supplied, | |
86 | * and marks all the affected blocks as valid and modified. If | |
87 | * an affected block is not allocated, it will be allocated. If | |
88 | * an affected block is not completely overwritten, and is not | |
89 | * valid before the operation, it will be read from disk before | |
90 | * being partially zeroed. | |
91 | */ | |
ef9d8733 | 92 | int |
dda35b8f CH |
93 | xfs_iozero( |
94 | struct xfs_inode *ip, /* inode */ | |
95 | loff_t pos, /* offset in file */ | |
96 | size_t count) /* size of data to zero */ | |
97 | { | |
98 | struct page *page; | |
99 | struct address_space *mapping; | |
100 | int status; | |
101 | ||
102 | mapping = VFS_I(ip)->i_mapping; | |
103 | do { | |
104 | unsigned offset, bytes; | |
105 | void *fsdata; | |
106 | ||
107 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
108 | bytes = PAGE_CACHE_SIZE - offset; | |
109 | if (bytes > count) | |
110 | bytes = count; | |
111 | ||
112 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
113 | AOP_FLAG_UNINTERRUPTIBLE, | |
114 | &page, &fsdata); | |
115 | if (status) | |
116 | break; | |
117 | ||
118 | zero_user(page, offset, bytes); | |
119 | ||
120 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
121 | page, fsdata); | |
122 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
123 | pos += bytes; | |
124 | count -= bytes; | |
125 | status = 0; | |
126 | } while (count); | |
127 | ||
128 | return (-status); | |
129 | } | |
130 | ||
8add71ca CH |
131 | int |
132 | xfs_update_prealloc_flags( | |
133 | struct xfs_inode *ip, | |
134 | enum xfs_prealloc_flags flags) | |
135 | { | |
136 | struct xfs_trans *tp; | |
137 | int error; | |
138 | ||
139 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); | |
140 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
141 | if (error) { | |
142 | xfs_trans_cancel(tp, 0); | |
143 | return error; | |
144 | } | |
145 | ||
146 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
147 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
148 | ||
149 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
150 | ip->i_d.di_mode &= ~S_ISUID; | |
151 | if (ip->i_d.di_mode & S_IXGRP) | |
152 | ip->i_d.di_mode &= ~S_ISGID; | |
153 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
154 | } | |
155 | ||
156 | if (flags & XFS_PREALLOC_SET) | |
157 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
158 | if (flags & XFS_PREALLOC_CLEAR) | |
159 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
160 | ||
161 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
162 | if (flags & XFS_PREALLOC_SYNC) | |
163 | xfs_trans_set_sync(tp); | |
164 | return xfs_trans_commit(tp, 0); | |
165 | } | |
166 | ||
1da2f2db CH |
167 | /* |
168 | * Fsync operations on directories are much simpler than on regular files, | |
169 | * as there is no file data to flush, and thus also no need for explicit | |
170 | * cache flush operations, and there are no non-transaction metadata updates | |
171 | * on directories either. | |
172 | */ | |
173 | STATIC int | |
174 | xfs_dir_fsync( | |
175 | struct file *file, | |
176 | loff_t start, | |
177 | loff_t end, | |
178 | int datasync) | |
179 | { | |
180 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
181 | struct xfs_mount *mp = ip->i_mount; | |
182 | xfs_lsn_t lsn = 0; | |
183 | ||
184 | trace_xfs_dir_fsync(ip); | |
185 | ||
186 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
187 | if (xfs_ipincount(ip)) | |
188 | lsn = ip->i_itemp->ili_last_lsn; | |
189 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
190 | ||
191 | if (!lsn) | |
192 | return 0; | |
2451337d | 193 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
194 | } |
195 | ||
fd3200be CH |
196 | STATIC int |
197 | xfs_file_fsync( | |
198 | struct file *file, | |
02c24a82 JB |
199 | loff_t start, |
200 | loff_t end, | |
fd3200be CH |
201 | int datasync) |
202 | { | |
7ea80859 CH |
203 | struct inode *inode = file->f_mapping->host; |
204 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 205 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
206 | int error = 0; |
207 | int log_flushed = 0; | |
b1037058 | 208 | xfs_lsn_t lsn = 0; |
fd3200be | 209 | |
cca28fb8 | 210 | trace_xfs_file_fsync(ip); |
fd3200be | 211 | |
02c24a82 JB |
212 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
213 | if (error) | |
214 | return error; | |
215 | ||
a27a263b | 216 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 217 | return -EIO; |
fd3200be CH |
218 | |
219 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
220 | ||
a27a263b CH |
221 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
222 | /* | |
223 | * If we have an RT and/or log subvolume we need to make sure | |
224 | * to flush the write cache the device used for file data | |
225 | * first. This is to ensure newly written file data make | |
226 | * it to disk before logging the new inode size in case of | |
227 | * an extending write. | |
228 | */ | |
229 | if (XFS_IS_REALTIME_INODE(ip)) | |
230 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
231 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
232 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
233 | } | |
234 | ||
fd3200be | 235 | /* |
8a9c9980 CH |
236 | * All metadata updates are logged, which means that we just have |
237 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
238 | */ |
239 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
240 | if (xfs_ipincount(ip)) { |
241 | if (!datasync || | |
242 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
243 | lsn = ip->i_itemp->ili_last_lsn; | |
244 | } | |
8a9c9980 | 245 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 246 | |
8a9c9980 | 247 | if (lsn) |
b1037058 CH |
248 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
249 | ||
a27a263b CH |
250 | /* |
251 | * If we only have a single device, and the log force about was | |
252 | * a no-op we might have to flush the data device cache here. | |
253 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
254 | * an already allocated file and thus do not have any metadata to | |
255 | * commit. | |
256 | */ | |
257 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
258 | mp->m_logdev_targp == mp->m_ddev_targp && | |
259 | !XFS_IS_REALTIME_INODE(ip) && | |
260 | !log_flushed) | |
261 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 262 | |
2451337d | 263 | return error; |
fd3200be CH |
264 | } |
265 | ||
00258e36 | 266 | STATIC ssize_t |
b4f5d2c6 | 267 | xfs_file_read_iter( |
dda35b8f | 268 | struct kiocb *iocb, |
b4f5d2c6 | 269 | struct iov_iter *to) |
dda35b8f CH |
270 | { |
271 | struct file *file = iocb->ki_filp; | |
272 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
273 | struct xfs_inode *ip = XFS_I(inode); |
274 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 275 | size_t size = iov_iter_count(to); |
dda35b8f | 276 | ssize_t ret = 0; |
00258e36 | 277 | int ioflags = 0; |
dda35b8f | 278 | xfs_fsize_t n; |
b4f5d2c6 | 279 | loff_t pos = iocb->ki_pos; |
dda35b8f | 280 | |
dda35b8f CH |
281 | XFS_STATS_INC(xs_read_calls); |
282 | ||
00258e36 | 283 | if (unlikely(file->f_flags & O_DIRECT)) |
b92cc59f | 284 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 285 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 286 | ioflags |= XFS_IO_INVIS; |
00258e36 | 287 | |
b92cc59f | 288 | if (unlikely(ioflags & XFS_IO_ISDIRECT)) { |
dda35b8f CH |
289 | xfs_buftarg_t *target = |
290 | XFS_IS_REALTIME_INODE(ip) ? | |
291 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
292 | /* DIO must be aligned to device logical sector size */ |
293 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 294 | if (pos == i_size_read(inode)) |
00258e36 | 295 | return 0; |
b474c7ae | 296 | return -EINVAL; |
dda35b8f CH |
297 | } |
298 | } | |
299 | ||
fb595814 | 300 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 301 | if (n <= 0 || size == 0) |
dda35b8f CH |
302 | return 0; |
303 | ||
304 | if (n < size) | |
305 | size = n; | |
306 | ||
307 | if (XFS_FORCED_SHUTDOWN(mp)) | |
308 | return -EIO; | |
309 | ||
0c38a251 DC |
310 | /* |
311 | * Locking is a bit tricky here. If we take an exclusive lock | |
312 | * for direct IO, we effectively serialise all new concurrent | |
313 | * read IO to this file and block it behind IO that is currently in | |
314 | * progress because IO in progress holds the IO lock shared. We only | |
315 | * need to hold the lock exclusive to blow away the page cache, so | |
316 | * only take lock exclusively if the page cache needs invalidation. | |
317 | * This allows the normal direct IO case of no page cache pages to | |
318 | * proceeed concurrently without serialisation. | |
319 | */ | |
320 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 321 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 322 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
323 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
324 | ||
00258e36 | 325 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 326 | ret = filemap_write_and_wait_range( |
fb595814 | 327 | VFS_I(ip)->i_mapping, |
7d4ea3ce | 328 | pos, pos + size - 1); |
487f84f3 DC |
329 | if (ret) { |
330 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
331 | return ret; | |
332 | } | |
85e584da CM |
333 | |
334 | /* | |
335 | * Invalidate whole pages. This can return an error if | |
336 | * we fail to invalidate a page, but this should never | |
337 | * happen on XFS. Warn if it does fail. | |
338 | */ | |
339 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
340 | pos >> PAGE_CACHE_SHIFT, |
341 | (pos + size - 1) >> PAGE_CACHE_SHIFT); | |
85e584da CM |
342 | WARN_ON_ONCE(ret); |
343 | ret = 0; | |
00258e36 | 344 | } |
487f84f3 | 345 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 346 | } |
dda35b8f | 347 | |
fb595814 | 348 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 349 | |
b4f5d2c6 | 350 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
351 | if (ret > 0) |
352 | XFS_STATS_ADD(xs_read_bytes, ret); | |
353 | ||
487f84f3 | 354 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
355 | return ret; |
356 | } | |
357 | ||
00258e36 CH |
358 | STATIC ssize_t |
359 | xfs_file_splice_read( | |
dda35b8f CH |
360 | struct file *infilp, |
361 | loff_t *ppos, | |
362 | struct pipe_inode_info *pipe, | |
363 | size_t count, | |
00258e36 | 364 | unsigned int flags) |
dda35b8f | 365 | { |
00258e36 | 366 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 367 | int ioflags = 0; |
dda35b8f CH |
368 | ssize_t ret; |
369 | ||
370 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
371 | |
372 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 373 | ioflags |= XFS_IO_INVIS; |
00258e36 | 374 | |
dda35b8f CH |
375 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
376 | return -EIO; | |
377 | ||
487f84f3 | 378 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 379 | |
dda35b8f CH |
380 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
381 | ||
382 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
383 | if (ret > 0) | |
384 | XFS_STATS_ADD(xs_read_bytes, ret); | |
385 | ||
487f84f3 | 386 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
387 | return ret; |
388 | } | |
389 | ||
dda35b8f | 390 | /* |
193aec10 CH |
391 | * This routine is called to handle zeroing any space in the last block of the |
392 | * file that is beyond the EOF. We do this since the size is being increased | |
393 | * without writing anything to that block and we don't want to read the | |
394 | * garbage on the disk. | |
dda35b8f CH |
395 | */ |
396 | STATIC int /* error (positive) */ | |
397 | xfs_zero_last_block( | |
193aec10 CH |
398 | struct xfs_inode *ip, |
399 | xfs_fsize_t offset, | |
400 | xfs_fsize_t isize) | |
dda35b8f | 401 | { |
193aec10 CH |
402 | struct xfs_mount *mp = ip->i_mount; |
403 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
404 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
405 | int zero_len; | |
406 | int nimaps = 1; | |
407 | int error = 0; | |
408 | struct xfs_bmbt_irec imap; | |
dda35b8f | 409 | |
193aec10 | 410 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 411 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 412 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 413 | if (error) |
dda35b8f | 414 | return error; |
193aec10 | 415 | |
dda35b8f | 416 | ASSERT(nimaps > 0); |
193aec10 | 417 | |
dda35b8f CH |
418 | /* |
419 | * If the block underlying isize is just a hole, then there | |
420 | * is nothing to zero. | |
421 | */ | |
193aec10 | 422 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 423 | return 0; |
dda35b8f CH |
424 | |
425 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
426 | if (isize + zero_len > offset) | |
427 | zero_len = offset - isize; | |
193aec10 | 428 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
429 | } |
430 | ||
431 | /* | |
193aec10 CH |
432 | * Zero any on disk space between the current EOF and the new, larger EOF. |
433 | * | |
434 | * This handles the normal case of zeroing the remainder of the last block in | |
435 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
436 | * file. This second case only happens with fixed size extents and when the | |
437 | * system crashes before the inode size was updated but after blocks were | |
438 | * allocated. | |
439 | * | |
440 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 441 | */ |
dda35b8f CH |
442 | int /* error (positive) */ |
443 | xfs_zero_eof( | |
193aec10 CH |
444 | struct xfs_inode *ip, |
445 | xfs_off_t offset, /* starting I/O offset */ | |
446 | xfs_fsize_t isize) /* current inode size */ | |
dda35b8f | 447 | { |
193aec10 CH |
448 | struct xfs_mount *mp = ip->i_mount; |
449 | xfs_fileoff_t start_zero_fsb; | |
450 | xfs_fileoff_t end_zero_fsb; | |
451 | xfs_fileoff_t zero_count_fsb; | |
452 | xfs_fileoff_t last_fsb; | |
453 | xfs_fileoff_t zero_off; | |
454 | xfs_fsize_t zero_len; | |
455 | int nimaps; | |
456 | int error = 0; | |
457 | struct xfs_bmbt_irec imap; | |
458 | ||
459 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
460 | ASSERT(offset > isize); |
461 | ||
462 | /* | |
463 | * First handle zeroing the block on which isize resides. | |
193aec10 | 464 | * |
dda35b8f CH |
465 | * We only zero a part of that block so it is handled specially. |
466 | */ | |
193aec10 CH |
467 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
468 | error = xfs_zero_last_block(ip, offset, isize); | |
469 | if (error) | |
470 | return error; | |
dda35b8f CH |
471 | } |
472 | ||
473 | /* | |
193aec10 CH |
474 | * Calculate the range between the new size and the old where blocks |
475 | * needing to be zeroed may exist. | |
476 | * | |
477 | * To get the block where the last byte in the file currently resides, | |
478 | * we need to subtract one from the size and truncate back to a block | |
479 | * boundary. We subtract 1 in case the size is exactly on a block | |
480 | * boundary. | |
dda35b8f CH |
481 | */ |
482 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
483 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
484 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
485 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
486 | if (last_fsb == end_zero_fsb) { | |
487 | /* | |
488 | * The size was only incremented on its last block. | |
489 | * We took care of that above, so just return. | |
490 | */ | |
491 | return 0; | |
492 | } | |
493 | ||
494 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
495 | while (start_zero_fsb <= end_zero_fsb) { | |
496 | nimaps = 1; | |
497 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
498 | |
499 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
500 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
501 | &imap, &nimaps, 0); | |
193aec10 CH |
502 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
503 | if (error) | |
dda35b8f | 504 | return error; |
193aec10 | 505 | |
dda35b8f CH |
506 | ASSERT(nimaps > 0); |
507 | ||
508 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
509 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
510 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
511 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
512 | continue; | |
513 | } | |
514 | ||
515 | /* | |
516 | * There are blocks we need to zero. | |
dda35b8f | 517 | */ |
dda35b8f CH |
518 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
519 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
520 | ||
521 | if ((zero_off + zero_len) > offset) | |
522 | zero_len = offset - zero_off; | |
523 | ||
524 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
525 | if (error) |
526 | return error; | |
dda35b8f CH |
527 | |
528 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
529 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
530 | } |
531 | ||
532 | return 0; | |
dda35b8f CH |
533 | } |
534 | ||
4d8d1581 DC |
535 | /* |
536 | * Common pre-write limit and setup checks. | |
537 | * | |
5bf1f262 CH |
538 | * Called with the iolocked held either shared and exclusive according to |
539 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
540 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
541 | */ |
542 | STATIC ssize_t | |
543 | xfs_file_aio_write_checks( | |
544 | struct file *file, | |
545 | loff_t *pos, | |
546 | size_t *count, | |
547 | int *iolock) | |
548 | { | |
549 | struct inode *inode = file->f_mapping->host; | |
550 | struct xfs_inode *ip = XFS_I(inode); | |
4d8d1581 DC |
551 | int error = 0; |
552 | ||
7271d243 | 553 | restart: |
4d8d1581 | 554 | error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); |
467f7899 | 555 | if (error) |
4d8d1581 | 556 | return error; |
4d8d1581 | 557 | |
781355c6 CH |
558 | error = xfs_break_layouts(inode, iolock); |
559 | if (error) | |
560 | return error; | |
561 | ||
4d8d1581 DC |
562 | /* |
563 | * If the offset is beyond the size of the file, we need to zero any | |
564 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 565 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
566 | * iolock shared, we need to update it to exclusive which implies |
567 | * having to redo all checks before. | |
4d8d1581 | 568 | */ |
2813d682 | 569 | if (*pos > i_size_read(inode)) { |
7271d243 | 570 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 571 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 572 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 573 | xfs_rw_ilock(ip, *iolock); |
7271d243 DC |
574 | goto restart; |
575 | } | |
2451337d | 576 | error = xfs_zero_eof(ip, *pos, i_size_read(inode)); |
467f7899 CH |
577 | if (error) |
578 | return error; | |
7271d243 | 579 | } |
4d8d1581 | 580 | |
8a9c9980 CH |
581 | /* |
582 | * Updating the timestamps will grab the ilock again from | |
583 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
584 | * lock above. Eventually we should look into a way to avoid | |
585 | * the pointless lock roundtrip. | |
586 | */ | |
c3b2da31 JB |
587 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
588 | error = file_update_time(file); | |
589 | if (error) | |
590 | return error; | |
591 | } | |
8a9c9980 | 592 | |
4d8d1581 DC |
593 | /* |
594 | * If we're writing the file then make sure to clear the setuid and | |
595 | * setgid bits if the process is not being run by root. This keeps | |
596 | * people from modifying setuid and setgid binaries. | |
597 | */ | |
598 | return file_remove_suid(file); | |
4d8d1581 DC |
599 | } |
600 | ||
f0d26e86 DC |
601 | /* |
602 | * xfs_file_dio_aio_write - handle direct IO writes | |
603 | * | |
604 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 605 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
606 | * follow locking changes and looping. |
607 | * | |
eda77982 DC |
608 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
609 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
610 | * pages are flushed out. | |
611 | * | |
612 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
613 | * allowing them to be done in parallel with reads and other direct IO writes. | |
614 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
615 | * needs to do sub-block zeroing and that requires serialisation against other | |
616 | * direct IOs to the same block. In this case we need to serialise the | |
617 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
618 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
619 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
620 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 621 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 622 | * |
f0d26e86 DC |
623 | * Returns with locks held indicated by @iolock and errors indicated by |
624 | * negative return values. | |
625 | */ | |
626 | STATIC ssize_t | |
627 | xfs_file_dio_aio_write( | |
628 | struct kiocb *iocb, | |
b3188919 | 629 | struct iov_iter *from) |
f0d26e86 DC |
630 | { |
631 | struct file *file = iocb->ki_filp; | |
632 | struct address_space *mapping = file->f_mapping; | |
633 | struct inode *inode = mapping->host; | |
634 | struct xfs_inode *ip = XFS_I(inode); | |
635 | struct xfs_mount *mp = ip->i_mount; | |
636 | ssize_t ret = 0; | |
eda77982 | 637 | int unaligned_io = 0; |
d0606464 | 638 | int iolock; |
b3188919 AV |
639 | size_t count = iov_iter_count(from); |
640 | loff_t pos = iocb->ki_pos; | |
f0d26e86 DC |
641 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
642 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
643 | ||
7c71ee78 ES |
644 | /* DIO must be aligned to device logical sector size */ |
645 | if ((pos | count) & target->bt_logical_sectormask) | |
b474c7ae | 646 | return -EINVAL; |
f0d26e86 | 647 | |
7c71ee78 | 648 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
649 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
650 | unaligned_io = 1; | |
651 | ||
7271d243 DC |
652 | /* |
653 | * We don't need to take an exclusive lock unless there page cache needs | |
654 | * to be invalidated or unaligned IO is being executed. We don't need to | |
655 | * consider the EOF extension case here because | |
656 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
657 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
658 | */ | |
659 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 660 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 661 | else |
d0606464 CH |
662 | iolock = XFS_IOLOCK_SHARED; |
663 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
664 | |
665 | /* | |
666 | * Recheck if there are cached pages that need invalidate after we got | |
667 | * the iolock to protect against other threads adding new pages while | |
668 | * we were waiting for the iolock. | |
669 | */ | |
d0606464 CH |
670 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
671 | xfs_rw_iunlock(ip, iolock); | |
672 | iolock = XFS_IOLOCK_EXCL; | |
673 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 674 | } |
f0d26e86 | 675 | |
d0606464 | 676 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 677 | if (ret) |
d0606464 | 678 | goto out; |
b3188919 | 679 | iov_iter_truncate(from, count); |
f0d26e86 DC |
680 | |
681 | if (mapping->nrpages) { | |
07d5035a | 682 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
7d4ea3ce | 683 | pos, pos + count - 1); |
f0d26e86 | 684 | if (ret) |
d0606464 | 685 | goto out; |
834ffca6 DC |
686 | /* |
687 | * Invalidate whole pages. This can return an error if | |
688 | * we fail to invalidate a page, but this should never | |
689 | * happen on XFS. Warn if it does fail. | |
690 | */ | |
691 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
7d4ea3ce DC |
692 | pos >> PAGE_CACHE_SHIFT, |
693 | (pos + count - 1) >> PAGE_CACHE_SHIFT); | |
834ffca6 DC |
694 | WARN_ON_ONCE(ret); |
695 | ret = 0; | |
f0d26e86 DC |
696 | } |
697 | ||
eda77982 DC |
698 | /* |
699 | * If we are doing unaligned IO, wait for all other IO to drain, | |
700 | * otherwise demote the lock if we had to flush cached pages | |
701 | */ | |
702 | if (unaligned_io) | |
4a06fd26 | 703 | inode_dio_wait(inode); |
d0606464 | 704 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 705 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 706 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
707 | } |
708 | ||
709 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
b3188919 | 710 | ret = generic_file_direct_write(iocb, from, pos); |
f0d26e86 | 711 | |
d0606464 CH |
712 | out: |
713 | xfs_rw_iunlock(ip, iolock); | |
714 | ||
f0d26e86 DC |
715 | /* No fallback to buffered IO on errors for XFS. */ |
716 | ASSERT(ret < 0 || ret == count); | |
717 | return ret; | |
718 | } | |
719 | ||
00258e36 | 720 | STATIC ssize_t |
637bbc75 | 721 | xfs_file_buffered_aio_write( |
dda35b8f | 722 | struct kiocb *iocb, |
b3188919 | 723 | struct iov_iter *from) |
dda35b8f CH |
724 | { |
725 | struct file *file = iocb->ki_filp; | |
726 | struct address_space *mapping = file->f_mapping; | |
727 | struct inode *inode = mapping->host; | |
00258e36 | 728 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
729 | ssize_t ret; |
730 | int enospc = 0; | |
d0606464 | 731 | int iolock = XFS_IOLOCK_EXCL; |
b3188919 AV |
732 | loff_t pos = iocb->ki_pos; |
733 | size_t count = iov_iter_count(from); | |
dda35b8f | 734 | |
d0606464 | 735 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 736 | |
d0606464 | 737 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 738 | if (ret) |
d0606464 | 739 | goto out; |
dda35b8f | 740 | |
b3188919 | 741 | iov_iter_truncate(from, count); |
dda35b8f | 742 | /* We can write back this queue in page reclaim */ |
de1414a6 | 743 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 744 | |
dda35b8f | 745 | write_retry: |
637bbc75 | 746 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); |
b3188919 | 747 | ret = generic_perform_write(file, from, pos); |
0a64bc2c AV |
748 | if (likely(ret >= 0)) |
749 | iocb->ki_pos = pos + ret; | |
dc06f398 | 750 | |
637bbc75 | 751 | /* |
dc06f398 BF |
752 | * If we hit a space limit, try to free up some lingering preallocated |
753 | * space before returning an error. In the case of ENOSPC, first try to | |
754 | * write back all dirty inodes to free up some of the excess reserved | |
755 | * metadata space. This reduces the chances that the eofblocks scan | |
756 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
757 | * also behaves as a filter to prevent too many eofblocks scans from | |
758 | * running at the same time. | |
637bbc75 | 759 | */ |
dc06f398 BF |
760 | if (ret == -EDQUOT && !enospc) { |
761 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
762 | if (enospc) | |
763 | goto write_retry; | |
764 | } else if (ret == -ENOSPC && !enospc) { | |
765 | struct xfs_eofblocks eofb = {0}; | |
766 | ||
637bbc75 | 767 | enospc = 1; |
9aa05000 | 768 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
769 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
770 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
771 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 772 | goto write_retry; |
dda35b8f | 773 | } |
d0606464 | 774 | |
dda35b8f | 775 | current->backing_dev_info = NULL; |
d0606464 CH |
776 | out: |
777 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
778 | return ret; |
779 | } | |
780 | ||
781 | STATIC ssize_t | |
bf97f3bc | 782 | xfs_file_write_iter( |
637bbc75 | 783 | struct kiocb *iocb, |
bf97f3bc | 784 | struct iov_iter *from) |
637bbc75 DC |
785 | { |
786 | struct file *file = iocb->ki_filp; | |
787 | struct address_space *mapping = file->f_mapping; | |
788 | struct inode *inode = mapping->host; | |
789 | struct xfs_inode *ip = XFS_I(inode); | |
790 | ssize_t ret; | |
bf97f3bc | 791 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
792 | |
793 | XFS_STATS_INC(xs_write_calls); | |
794 | ||
637bbc75 DC |
795 | if (ocount == 0) |
796 | return 0; | |
797 | ||
bf97f3bc AV |
798 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
799 | return -EIO; | |
637bbc75 DC |
800 | |
801 | if (unlikely(file->f_flags & O_DIRECT)) | |
bf97f3bc | 802 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 803 | else |
bf97f3bc | 804 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 805 | |
d0606464 CH |
806 | if (ret > 0) { |
807 | ssize_t err; | |
dda35b8f | 808 | |
d0606464 | 809 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 810 | |
d0606464 | 811 | /* Handle various SYNC-type writes */ |
d311d79d | 812 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
813 | if (err < 0) |
814 | ret = err; | |
dda35b8f | 815 | } |
a363f0c2 | 816 | return ret; |
dda35b8f CH |
817 | } |
818 | ||
2fe17c10 CH |
819 | STATIC long |
820 | xfs_file_fallocate( | |
83aee9e4 CH |
821 | struct file *file, |
822 | int mode, | |
823 | loff_t offset, | |
824 | loff_t len) | |
2fe17c10 | 825 | { |
83aee9e4 CH |
826 | struct inode *inode = file_inode(file); |
827 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 828 | long error; |
8add71ca | 829 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 830 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 831 | loff_t new_size = 0; |
2fe17c10 | 832 | |
83aee9e4 CH |
833 | if (!S_ISREG(inode->i_mode)) |
834 | return -EINVAL; | |
e1d8fb88 | 835 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
376ba313 | 836 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
2fe17c10 CH |
837 | return -EOPNOTSUPP; |
838 | ||
781355c6 CH |
839 | xfs_ilock(ip, iolock); |
840 | error = xfs_break_layouts(inode, &iolock); | |
841 | if (error) | |
842 | goto out_unlock; | |
843 | ||
83aee9e4 CH |
844 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
845 | error = xfs_free_file_space(ip, offset, len); | |
846 | if (error) | |
847 | goto out_unlock; | |
e1d8fb88 NJ |
848 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
849 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
850 | ||
851 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 852 | error = -EINVAL; |
e1d8fb88 NJ |
853 | goto out_unlock; |
854 | } | |
855 | ||
23fffa92 LC |
856 | /* |
857 | * There is no need to overlap collapse range with EOF, | |
858 | * in which case it is effectively a truncate operation | |
859 | */ | |
860 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 861 | error = -EINVAL; |
23fffa92 LC |
862 | goto out_unlock; |
863 | } | |
864 | ||
e1d8fb88 NJ |
865 | new_size = i_size_read(inode) - len; |
866 | ||
867 | error = xfs_collapse_file_space(ip, offset, len); | |
868 | if (error) | |
869 | goto out_unlock; | |
83aee9e4 | 870 | } else { |
8add71ca CH |
871 | flags |= XFS_PREALLOC_SET; |
872 | ||
83aee9e4 CH |
873 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
874 | offset + len > i_size_read(inode)) { | |
875 | new_size = offset + len; | |
2451337d | 876 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
877 | if (error) |
878 | goto out_unlock; | |
879 | } | |
2fe17c10 | 880 | |
376ba313 LC |
881 | if (mode & FALLOC_FL_ZERO_RANGE) |
882 | error = xfs_zero_file_space(ip, offset, len); | |
883 | else | |
884 | error = xfs_alloc_file_space(ip, offset, len, | |
885 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
886 | if (error) |
887 | goto out_unlock; | |
888 | } | |
889 | ||
83aee9e4 | 890 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
891 | flags |= XFS_PREALLOC_SYNC; |
892 | ||
893 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
894 | if (error) |
895 | goto out_unlock; | |
896 | ||
897 | /* Change file size if needed */ | |
898 | if (new_size) { | |
899 | struct iattr iattr; | |
900 | ||
901 | iattr.ia_valid = ATTR_SIZE; | |
902 | iattr.ia_size = new_size; | |
83aee9e4 | 903 | error = xfs_setattr_size(ip, &iattr); |
2fe17c10 CH |
904 | } |
905 | ||
906 | out_unlock: | |
781355c6 | 907 | xfs_iunlock(ip, iolock); |
2451337d | 908 | return error; |
2fe17c10 CH |
909 | } |
910 | ||
911 | ||
1da177e4 | 912 | STATIC int |
3562fd45 | 913 | xfs_file_open( |
1da177e4 | 914 | struct inode *inode, |
f999a5bf | 915 | struct file *file) |
1da177e4 | 916 | { |
f999a5bf | 917 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 918 | return -EFBIG; |
f999a5bf CH |
919 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
920 | return -EIO; | |
921 | return 0; | |
922 | } | |
923 | ||
924 | STATIC int | |
925 | xfs_dir_open( | |
926 | struct inode *inode, | |
927 | struct file *file) | |
928 | { | |
929 | struct xfs_inode *ip = XFS_I(inode); | |
930 | int mode; | |
931 | int error; | |
932 | ||
933 | error = xfs_file_open(inode, file); | |
934 | if (error) | |
935 | return error; | |
936 | ||
937 | /* | |
938 | * If there are any blocks, read-ahead block 0 as we're almost | |
939 | * certain to have the next operation be a read there. | |
940 | */ | |
309ecac8 | 941 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 942 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 943 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
944 | xfs_iunlock(ip, mode); |
945 | return 0; | |
1da177e4 LT |
946 | } |
947 | ||
1da177e4 | 948 | STATIC int |
3562fd45 | 949 | xfs_file_release( |
1da177e4 LT |
950 | struct inode *inode, |
951 | struct file *filp) | |
952 | { | |
2451337d | 953 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
954 | } |
955 | ||
1da177e4 | 956 | STATIC int |
3562fd45 | 957 | xfs_file_readdir( |
b8227554 AV |
958 | struct file *file, |
959 | struct dir_context *ctx) | |
1da177e4 | 960 | { |
b8227554 | 961 | struct inode *inode = file_inode(file); |
739bfb2a | 962 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
963 | size_t bufsize; |
964 | ||
965 | /* | |
966 | * The Linux API doesn't pass down the total size of the buffer | |
967 | * we read into down to the filesystem. With the filldir concept | |
968 | * it's not needed for correct information, but the XFS dir2 leaf | |
969 | * code wants an estimate of the buffer size to calculate it's | |
970 | * readahead window and size the buffers used for mapping to | |
971 | * physical blocks. | |
972 | * | |
973 | * Try to give it an estimate that's good enough, maybe at some | |
974 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 975 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 976 | */ |
a9cc799e | 977 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 978 | |
8300475e | 979 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
980 | } |
981 | ||
1da177e4 | 982 | STATIC int |
3562fd45 | 983 | xfs_file_mmap( |
1da177e4 LT |
984 | struct file *filp, |
985 | struct vm_area_struct *vma) | |
986 | { | |
3562fd45 | 987 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 988 | |
fbc1462b | 989 | file_accessed(filp); |
1da177e4 LT |
990 | return 0; |
991 | } | |
992 | ||
4f57dbc6 DC |
993 | /* |
994 | * mmap()d file has taken write protection fault and is being made | |
995 | * writable. We can set the page state up correctly for a writable | |
996 | * page, which means we can do correct delalloc accounting (ENOSPC | |
997 | * checking!) and unwritten extent mapping. | |
998 | */ | |
999 | STATIC int | |
1000 | xfs_vm_page_mkwrite( | |
1001 | struct vm_area_struct *vma, | |
c2ec175c | 1002 | struct vm_fault *vmf) |
4f57dbc6 | 1003 | { |
c2ec175c | 1004 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
1005 | } |
1006 | ||
d126d43f JL |
1007 | /* |
1008 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1009 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1010 | */ |
1011 | enum { | |
1012 | HOLE_OFF = 0, | |
1013 | DATA_OFF, | |
1014 | }; | |
1015 | ||
1016 | /* | |
1017 | * Lookup the desired type of offset from the given page. | |
1018 | * | |
1019 | * On success, return true and the offset argument will point to the | |
1020 | * start of the region that was found. Otherwise this function will | |
1021 | * return false and keep the offset argument unchanged. | |
1022 | */ | |
1023 | STATIC bool | |
1024 | xfs_lookup_buffer_offset( | |
1025 | struct page *page, | |
1026 | loff_t *offset, | |
1027 | unsigned int type) | |
1028 | { | |
1029 | loff_t lastoff = page_offset(page); | |
1030 | bool found = false; | |
1031 | struct buffer_head *bh, *head; | |
1032 | ||
1033 | bh = head = page_buffers(page); | |
1034 | do { | |
1035 | /* | |
1036 | * Unwritten extents that have data in the page | |
1037 | * cache covering them can be identified by the | |
1038 | * BH_Unwritten state flag. Pages with multiple | |
1039 | * buffers might have a mix of holes, data and | |
1040 | * unwritten extents - any buffer with valid | |
1041 | * data in it should have BH_Uptodate flag set | |
1042 | * on it. | |
1043 | */ | |
1044 | if (buffer_unwritten(bh) || | |
1045 | buffer_uptodate(bh)) { | |
1046 | if (type == DATA_OFF) | |
1047 | found = true; | |
1048 | } else { | |
1049 | if (type == HOLE_OFF) | |
1050 | found = true; | |
1051 | } | |
1052 | ||
1053 | if (found) { | |
1054 | *offset = lastoff; | |
1055 | break; | |
1056 | } | |
1057 | lastoff += bh->b_size; | |
1058 | } while ((bh = bh->b_this_page) != head); | |
1059 | ||
1060 | return found; | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * This routine is called to find out and return a data or hole offset | |
1065 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1066 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1067 | * |
1068 | * The argument offset is used to tell where we start to search from the | |
1069 | * page cache. Map is used to figure out the end points of the range to | |
1070 | * lookup pages. | |
1071 | * | |
1072 | * Return true if the desired type of offset was found, and the argument | |
1073 | * offset is filled with that address. Otherwise, return false and keep | |
1074 | * offset unchanged. | |
1075 | */ | |
1076 | STATIC bool | |
1077 | xfs_find_get_desired_pgoff( | |
1078 | struct inode *inode, | |
1079 | struct xfs_bmbt_irec *map, | |
1080 | unsigned int type, | |
1081 | loff_t *offset) | |
1082 | { | |
1083 | struct xfs_inode *ip = XFS_I(inode); | |
1084 | struct xfs_mount *mp = ip->i_mount; | |
1085 | struct pagevec pvec; | |
1086 | pgoff_t index; | |
1087 | pgoff_t end; | |
1088 | loff_t endoff; | |
1089 | loff_t startoff = *offset; | |
1090 | loff_t lastoff = startoff; | |
1091 | bool found = false; | |
1092 | ||
1093 | pagevec_init(&pvec, 0); | |
1094 | ||
1095 | index = startoff >> PAGE_CACHE_SHIFT; | |
1096 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1097 | end = endoff >> PAGE_CACHE_SHIFT; | |
1098 | do { | |
1099 | int want; | |
1100 | unsigned nr_pages; | |
1101 | unsigned int i; | |
1102 | ||
1103 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1104 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1105 | want); | |
1106 | /* | |
1107 | * No page mapped into given range. If we are searching holes | |
1108 | * and if this is the first time we got into the loop, it means | |
1109 | * that the given offset is landed in a hole, return it. | |
1110 | * | |
1111 | * If we have already stepped through some block buffers to find | |
1112 | * holes but they all contains data. In this case, the last | |
1113 | * offset is already updated and pointed to the end of the last | |
1114 | * mapped page, if it does not reach the endpoint to search, | |
1115 | * that means there should be a hole between them. | |
1116 | */ | |
1117 | if (nr_pages == 0) { | |
1118 | /* Data search found nothing */ | |
1119 | if (type == DATA_OFF) | |
1120 | break; | |
1121 | ||
1122 | ASSERT(type == HOLE_OFF); | |
1123 | if (lastoff == startoff || lastoff < endoff) { | |
1124 | found = true; | |
1125 | *offset = lastoff; | |
1126 | } | |
1127 | break; | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * At lease we found one page. If this is the first time we | |
1132 | * step into the loop, and if the first page index offset is | |
1133 | * greater than the given search offset, a hole was found. | |
1134 | */ | |
1135 | if (type == HOLE_OFF && lastoff == startoff && | |
1136 | lastoff < page_offset(pvec.pages[0])) { | |
1137 | found = true; | |
1138 | break; | |
1139 | } | |
1140 | ||
1141 | for (i = 0; i < nr_pages; i++) { | |
1142 | struct page *page = pvec.pages[i]; | |
1143 | loff_t b_offset; | |
1144 | ||
1145 | /* | |
1146 | * At this point, the page may be truncated or | |
1147 | * invalidated (changing page->mapping to NULL), | |
1148 | * or even swizzled back from swapper_space to tmpfs | |
1149 | * file mapping. However, page->index will not change | |
1150 | * because we have a reference on the page. | |
1151 | * | |
1152 | * Searching done if the page index is out of range. | |
1153 | * If the current offset is not reaches the end of | |
1154 | * the specified search range, there should be a hole | |
1155 | * between them. | |
1156 | */ | |
1157 | if (page->index > end) { | |
1158 | if (type == HOLE_OFF && lastoff < endoff) { | |
1159 | *offset = lastoff; | |
1160 | found = true; | |
1161 | } | |
1162 | goto out; | |
1163 | } | |
1164 | ||
1165 | lock_page(page); | |
1166 | /* | |
1167 | * Page truncated or invalidated(page->mapping == NULL). | |
1168 | * We can freely skip it and proceed to check the next | |
1169 | * page. | |
1170 | */ | |
1171 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1172 | unlock_page(page); | |
1173 | continue; | |
1174 | } | |
1175 | ||
1176 | if (!page_has_buffers(page)) { | |
1177 | unlock_page(page); | |
1178 | continue; | |
1179 | } | |
1180 | ||
1181 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1182 | if (found) { | |
1183 | /* | |
1184 | * The found offset may be less than the start | |
1185 | * point to search if this is the first time to | |
1186 | * come here. | |
1187 | */ | |
1188 | *offset = max_t(loff_t, startoff, b_offset); | |
1189 | unlock_page(page); | |
1190 | goto out; | |
1191 | } | |
1192 | ||
1193 | /* | |
1194 | * We either searching data but nothing was found, or | |
1195 | * searching hole but found a data buffer. In either | |
1196 | * case, probably the next page contains the desired | |
1197 | * things, update the last offset to it so. | |
1198 | */ | |
1199 | lastoff = page_offset(page) + PAGE_SIZE; | |
1200 | unlock_page(page); | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * The number of returned pages less than our desired, search | |
1205 | * done. In this case, nothing was found for searching data, | |
1206 | * but we found a hole behind the last offset. | |
1207 | */ | |
1208 | if (nr_pages < want) { | |
1209 | if (type == HOLE_OFF) { | |
1210 | *offset = lastoff; | |
1211 | found = true; | |
1212 | } | |
1213 | break; | |
1214 | } | |
1215 | ||
1216 | index = pvec.pages[i - 1]->index + 1; | |
1217 | pagevec_release(&pvec); | |
1218 | } while (index <= end); | |
1219 | ||
1220 | out: | |
1221 | pagevec_release(&pvec); | |
1222 | return found; | |
1223 | } | |
1224 | ||
3fe3e6b1 | 1225 | STATIC loff_t |
49c69591 | 1226 | xfs_seek_hole_data( |
3fe3e6b1 | 1227 | struct file *file, |
49c69591 ES |
1228 | loff_t start, |
1229 | int whence) | |
3fe3e6b1 JL |
1230 | { |
1231 | struct inode *inode = file->f_mapping->host; | |
1232 | struct xfs_inode *ip = XFS_I(inode); | |
1233 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1234 | loff_t uninitialized_var(offset); |
1235 | xfs_fsize_t isize; | |
1236 | xfs_fileoff_t fsbno; | |
1237 | xfs_filblks_t end; | |
1238 | uint lock; | |
1239 | int error; | |
1240 | ||
49c69591 ES |
1241 | if (XFS_FORCED_SHUTDOWN(mp)) |
1242 | return -EIO; | |
1243 | ||
309ecac8 | 1244 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1245 | |
1246 | isize = i_size_read(inode); | |
1247 | if (start >= isize) { | |
2451337d | 1248 | error = -ENXIO; |
3fe3e6b1 JL |
1249 | goto out_unlock; |
1250 | } | |
1251 | ||
3fe3e6b1 JL |
1252 | /* |
1253 | * Try to read extents from the first block indicated | |
1254 | * by fsbno to the end block of the file. | |
1255 | */ | |
52f1acc8 | 1256 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1257 | end = XFS_B_TO_FSB(mp, isize); |
49c69591 | 1258 | |
52f1acc8 JL |
1259 | for (;;) { |
1260 | struct xfs_bmbt_irec map[2]; | |
1261 | int nmap = 2; | |
1262 | unsigned int i; | |
3fe3e6b1 | 1263 | |
52f1acc8 JL |
1264 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1265 | XFS_BMAPI_ENTIRE); | |
1266 | if (error) | |
1267 | goto out_unlock; | |
3fe3e6b1 | 1268 | |
52f1acc8 JL |
1269 | /* No extents at given offset, must be beyond EOF */ |
1270 | if (nmap == 0) { | |
2451337d | 1271 | error = -ENXIO; |
52f1acc8 JL |
1272 | goto out_unlock; |
1273 | } | |
1274 | ||
1275 | for (i = 0; i < nmap; i++) { | |
1276 | offset = max_t(loff_t, start, | |
1277 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1278 | ||
49c69591 ES |
1279 | /* Landed in the hole we wanted? */ |
1280 | if (whence == SEEK_HOLE && | |
1281 | map[i].br_startblock == HOLESTARTBLOCK) | |
1282 | goto out; | |
1283 | ||
1284 | /* Landed in the data extent we wanted? */ | |
1285 | if (whence == SEEK_DATA && | |
1286 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1287 | (map[i].br_state == XFS_EXT_NORM && | |
1288 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1289 | goto out; |
1290 | ||
1291 | /* | |
49c69591 ES |
1292 | * Landed in an unwritten extent, try to search |
1293 | * for hole or data from page cache. | |
52f1acc8 JL |
1294 | */ |
1295 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1296 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1297 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1298 | &offset)) | |
52f1acc8 JL |
1299 | goto out; |
1300 | } | |
1301 | } | |
1302 | ||
1303 | /* | |
49c69591 ES |
1304 | * We only received one extent out of the two requested. This |
1305 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1306 | */ |
3fe3e6b1 | 1307 | if (nmap == 1) { |
49c69591 ES |
1308 | /* |
1309 | * If we were looking for a hole, set offset to | |
1310 | * the end of the file (i.e., there is an implicit | |
1311 | * hole at the end of any file). | |
1312 | */ | |
1313 | if (whence == SEEK_HOLE) { | |
1314 | offset = isize; | |
1315 | break; | |
1316 | } | |
1317 | /* | |
1318 | * If we were looking for data, it's nowhere to be found | |
1319 | */ | |
1320 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1321 | error = -ENXIO; |
3fe3e6b1 JL |
1322 | goto out_unlock; |
1323 | } | |
1324 | ||
52f1acc8 JL |
1325 | ASSERT(i > 1); |
1326 | ||
1327 | /* | |
1328 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1329 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1330 | */ |
1331 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1332 | start = XFS_FSB_TO_B(mp, fsbno); | |
1333 | if (start >= isize) { | |
49c69591 ES |
1334 | if (whence == SEEK_HOLE) { |
1335 | offset = isize; | |
1336 | break; | |
1337 | } | |
1338 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1339 | error = -ENXIO; |
52f1acc8 JL |
1340 | goto out_unlock; |
1341 | } | |
3fe3e6b1 JL |
1342 | } |
1343 | ||
b686d1f7 JL |
1344 | out: |
1345 | /* | |
49c69591 | 1346 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1347 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1348 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1349 | * situation in particular. |
1350 | */ | |
49c69591 ES |
1351 | if (whence == SEEK_HOLE) |
1352 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1353 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1354 | |
1355 | out_unlock: | |
01f4f327 | 1356 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1357 | |
1358 | if (error) | |
2451337d | 1359 | return error; |
3fe3e6b1 JL |
1360 | return offset; |
1361 | } | |
1362 | ||
1363 | STATIC loff_t | |
1364 | xfs_file_llseek( | |
1365 | struct file *file, | |
1366 | loff_t offset, | |
59f9c004 | 1367 | int whence) |
3fe3e6b1 | 1368 | { |
59f9c004 | 1369 | switch (whence) { |
3fe3e6b1 JL |
1370 | case SEEK_END: |
1371 | case SEEK_CUR: | |
1372 | case SEEK_SET: | |
59f9c004 | 1373 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1374 | case SEEK_HOLE: |
49c69591 | 1375 | case SEEK_DATA: |
59f9c004 | 1376 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1377 | default: |
1378 | return -EINVAL; | |
1379 | } | |
1380 | } | |
1381 | ||
4b6f5d20 | 1382 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1383 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1384 | .read = new_sync_read, |
bf97f3bc | 1385 | .write = new_sync_write, |
b4f5d2c6 | 1386 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1387 | .write_iter = xfs_file_write_iter, |
1b895840 | 1388 | .splice_read = xfs_file_splice_read, |
8d020765 | 1389 | .splice_write = iter_file_splice_write, |
3562fd45 | 1390 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1391 | #ifdef CONFIG_COMPAT |
3562fd45 | 1392 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1393 | #endif |
3562fd45 NS |
1394 | .mmap = xfs_file_mmap, |
1395 | .open = xfs_file_open, | |
1396 | .release = xfs_file_release, | |
1397 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1398 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1399 | }; |
1400 | ||
4b6f5d20 | 1401 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1402 | .open = xfs_dir_open, |
1da177e4 | 1403 | .read = generic_read_dir, |
b8227554 | 1404 | .iterate = xfs_file_readdir, |
59af1584 | 1405 | .llseek = generic_file_llseek, |
3562fd45 | 1406 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1407 | #ifdef CONFIG_COMPAT |
3562fd45 | 1408 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1409 | #endif |
1da2f2db | 1410 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1411 | }; |
1412 | ||
f0f37e2f | 1413 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1414 | .fault = filemap_fault, |
f1820361 | 1415 | .map_pages = filemap_map_pages, |
4f57dbc6 | 1416 | .page_mkwrite = xfs_vm_page_mkwrite, |
6fac0cb4 | 1417 | }; |