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" |
1da177e4 LT |
20 | #include "xfs_log.h" |
21 | #include "xfs_sb.h" | |
a844f451 | 22 | #include "xfs_ag.h" |
1da177e4 | 23 | #include "xfs_trans.h" |
1da177e4 LT |
24 | #include "xfs_mount.h" |
25 | #include "xfs_bmap_btree.h" | |
1da177e4 | 26 | #include "xfs_alloc.h" |
1da177e4 LT |
27 | #include "xfs_dinode.h" |
28 | #include "xfs_inode.h" | |
fd3200be | 29 | #include "xfs_inode_item.h" |
dda35b8f | 30 | #include "xfs_bmap.h" |
1da177e4 | 31 | #include "xfs_error.h" |
739bfb2a | 32 | #include "xfs_vnodeops.h" |
f999a5bf | 33 | #include "xfs_da_btree.h" |
da6958c8 DC |
34 | #include "xfs_dir2_format.h" |
35 | #include "xfs_dir2_priv.h" | |
ddcd856d | 36 | #include "xfs_ioctl.h" |
dda35b8f | 37 | #include "xfs_trace.h" |
1da177e4 | 38 | |
a27bb332 | 39 | #include <linux/aio.h> |
1da177e4 | 40 | #include <linux/dcache.h> |
2fe17c10 | 41 | #include <linux/falloc.h> |
d126d43f | 42 | #include <linux/pagevec.h> |
1da177e4 | 43 | |
f0f37e2f | 44 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 45 | |
487f84f3 DC |
46 | /* |
47 | * Locking primitives for read and write IO paths to ensure we consistently use | |
48 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
49 | */ | |
50 | static inline void | |
51 | xfs_rw_ilock( | |
52 | struct xfs_inode *ip, | |
53 | int type) | |
54 | { | |
55 | if (type & XFS_IOLOCK_EXCL) | |
56 | mutex_lock(&VFS_I(ip)->i_mutex); | |
57 | xfs_ilock(ip, type); | |
58 | } | |
59 | ||
60 | static inline void | |
61 | xfs_rw_iunlock( | |
62 | struct xfs_inode *ip, | |
63 | int type) | |
64 | { | |
65 | xfs_iunlock(ip, type); | |
66 | if (type & XFS_IOLOCK_EXCL) | |
67 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
68 | } | |
69 | ||
70 | static inline void | |
71 | xfs_rw_ilock_demote( | |
72 | struct xfs_inode *ip, | |
73 | int type) | |
74 | { | |
75 | xfs_ilock_demote(ip, type); | |
76 | if (type & XFS_IOLOCK_EXCL) | |
77 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
78 | } | |
79 | ||
dda35b8f CH |
80 | /* |
81 | * xfs_iozero | |
82 | * | |
83 | * xfs_iozero clears the specified range of buffer supplied, | |
84 | * and marks all the affected blocks as valid and modified. If | |
85 | * an affected block is not allocated, it will be allocated. If | |
86 | * an affected block is not completely overwritten, and is not | |
87 | * valid before the operation, it will be read from disk before | |
88 | * being partially zeroed. | |
89 | */ | |
ef9d8733 | 90 | int |
dda35b8f CH |
91 | xfs_iozero( |
92 | struct xfs_inode *ip, /* inode */ | |
93 | loff_t pos, /* offset in file */ | |
94 | size_t count) /* size of data to zero */ | |
95 | { | |
96 | struct page *page; | |
97 | struct address_space *mapping; | |
98 | int status; | |
99 | ||
100 | mapping = VFS_I(ip)->i_mapping; | |
101 | do { | |
102 | unsigned offset, bytes; | |
103 | void *fsdata; | |
104 | ||
105 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
106 | bytes = PAGE_CACHE_SIZE - offset; | |
107 | if (bytes > count) | |
108 | bytes = count; | |
109 | ||
110 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
111 | AOP_FLAG_UNINTERRUPTIBLE, | |
112 | &page, &fsdata); | |
113 | if (status) | |
114 | break; | |
115 | ||
116 | zero_user(page, offset, bytes); | |
117 | ||
118 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
119 | page, fsdata); | |
120 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
121 | pos += bytes; | |
122 | count -= bytes; | |
123 | status = 0; | |
124 | } while (count); | |
125 | ||
126 | return (-status); | |
127 | } | |
128 | ||
1da2f2db CH |
129 | /* |
130 | * Fsync operations on directories are much simpler than on regular files, | |
131 | * as there is no file data to flush, and thus also no need for explicit | |
132 | * cache flush operations, and there are no non-transaction metadata updates | |
133 | * on directories either. | |
134 | */ | |
135 | STATIC int | |
136 | xfs_dir_fsync( | |
137 | struct file *file, | |
138 | loff_t start, | |
139 | loff_t end, | |
140 | int datasync) | |
141 | { | |
142 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
143 | struct xfs_mount *mp = ip->i_mount; | |
144 | xfs_lsn_t lsn = 0; | |
145 | ||
146 | trace_xfs_dir_fsync(ip); | |
147 | ||
148 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
149 | if (xfs_ipincount(ip)) | |
150 | lsn = ip->i_itemp->ili_last_lsn; | |
151 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
152 | ||
153 | if (!lsn) | |
154 | return 0; | |
155 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); | |
156 | } | |
157 | ||
fd3200be CH |
158 | STATIC int |
159 | xfs_file_fsync( | |
160 | struct file *file, | |
02c24a82 JB |
161 | loff_t start, |
162 | loff_t end, | |
fd3200be CH |
163 | int datasync) |
164 | { | |
7ea80859 CH |
165 | struct inode *inode = file->f_mapping->host; |
166 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 167 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
168 | int error = 0; |
169 | int log_flushed = 0; | |
b1037058 | 170 | xfs_lsn_t lsn = 0; |
fd3200be | 171 | |
cca28fb8 | 172 | trace_xfs_file_fsync(ip); |
fd3200be | 173 | |
02c24a82 JB |
174 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
175 | if (error) | |
176 | return error; | |
177 | ||
a27a263b | 178 | if (XFS_FORCED_SHUTDOWN(mp)) |
fd3200be CH |
179 | return -XFS_ERROR(EIO); |
180 | ||
181 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
182 | ||
a27a263b CH |
183 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
184 | /* | |
185 | * If we have an RT and/or log subvolume we need to make sure | |
186 | * to flush the write cache the device used for file data | |
187 | * first. This is to ensure newly written file data make | |
188 | * it to disk before logging the new inode size in case of | |
189 | * an extending write. | |
190 | */ | |
191 | if (XFS_IS_REALTIME_INODE(ip)) | |
192 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
193 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
194 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
195 | } | |
196 | ||
fd3200be | 197 | /* |
8a9c9980 CH |
198 | * All metadata updates are logged, which means that we just have |
199 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
200 | */ |
201 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
202 | if (xfs_ipincount(ip)) { |
203 | if (!datasync || | |
204 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
205 | lsn = ip->i_itemp->ili_last_lsn; | |
206 | } | |
8a9c9980 | 207 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 208 | |
8a9c9980 | 209 | if (lsn) |
b1037058 CH |
210 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
211 | ||
a27a263b CH |
212 | /* |
213 | * If we only have a single device, and the log force about was | |
214 | * a no-op we might have to flush the data device cache here. | |
215 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
216 | * an already allocated file and thus do not have any metadata to | |
217 | * commit. | |
218 | */ | |
219 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
220 | mp->m_logdev_targp == mp->m_ddev_targp && | |
221 | !XFS_IS_REALTIME_INODE(ip) && | |
222 | !log_flushed) | |
223 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be CH |
224 | |
225 | return -error; | |
226 | } | |
227 | ||
00258e36 CH |
228 | STATIC ssize_t |
229 | xfs_file_aio_read( | |
dda35b8f CH |
230 | struct kiocb *iocb, |
231 | const struct iovec *iovp, | |
00258e36 CH |
232 | unsigned long nr_segs, |
233 | loff_t pos) | |
dda35b8f CH |
234 | { |
235 | struct file *file = iocb->ki_filp; | |
236 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
237 | struct xfs_inode *ip = XFS_I(inode); |
238 | struct xfs_mount *mp = ip->i_mount; | |
dda35b8f CH |
239 | size_t size = 0; |
240 | ssize_t ret = 0; | |
00258e36 | 241 | int ioflags = 0; |
dda35b8f | 242 | xfs_fsize_t n; |
dda35b8f | 243 | |
dda35b8f CH |
244 | XFS_STATS_INC(xs_read_calls); |
245 | ||
00258e36 CH |
246 | BUG_ON(iocb->ki_pos != pos); |
247 | ||
248 | if (unlikely(file->f_flags & O_DIRECT)) | |
249 | ioflags |= IO_ISDIRECT; | |
250 | if (file->f_mode & FMODE_NOCMTIME) | |
251 | ioflags |= IO_INVIS; | |
252 | ||
52764329 DC |
253 | ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE); |
254 | if (ret < 0) | |
255 | return ret; | |
dda35b8f CH |
256 | |
257 | if (unlikely(ioflags & IO_ISDIRECT)) { | |
258 | xfs_buftarg_t *target = | |
259 | XFS_IS_REALTIME_INODE(ip) ? | |
260 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
fb595814 DC |
261 | if ((pos & target->bt_smask) || (size & target->bt_smask)) { |
262 | if (pos == i_size_read(inode)) | |
00258e36 | 263 | return 0; |
dda35b8f CH |
264 | return -XFS_ERROR(EINVAL); |
265 | } | |
266 | } | |
267 | ||
fb595814 | 268 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 269 | if (n <= 0 || size == 0) |
dda35b8f CH |
270 | return 0; |
271 | ||
272 | if (n < size) | |
273 | size = n; | |
274 | ||
275 | if (XFS_FORCED_SHUTDOWN(mp)) | |
276 | return -EIO; | |
277 | ||
0c38a251 DC |
278 | /* |
279 | * Locking is a bit tricky here. If we take an exclusive lock | |
280 | * for direct IO, we effectively serialise all new concurrent | |
281 | * read IO to this file and block it behind IO that is currently in | |
282 | * progress because IO in progress holds the IO lock shared. We only | |
283 | * need to hold the lock exclusive to blow away the page cache, so | |
284 | * only take lock exclusively if the page cache needs invalidation. | |
285 | * This allows the normal direct IO case of no page cache pages to | |
286 | * proceeed concurrently without serialisation. | |
287 | */ | |
288 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
289 | if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) { | |
290 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); | |
487f84f3 DC |
291 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
292 | ||
00258e36 | 293 | if (inode->i_mapping->nrpages) { |
fb595814 DC |
294 | ret = -filemap_write_and_wait_range( |
295 | VFS_I(ip)->i_mapping, | |
296 | pos, -1); | |
487f84f3 DC |
297 | if (ret) { |
298 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
299 | return ret; | |
300 | } | |
d96dbb06 CM |
301 | |
302 | /* | |
303 | * Invalidate whole pages. This can return an error if | |
304 | * we fail to invalidate a page, but this should never | |
305 | * happen on XFS. Warn if it does fail. | |
306 | */ | |
307 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
308 | pos >> PAGE_CACHE_SHIFT, -1); | |
309 | WARN_ON_ONCE(ret); | |
310 | ret = 0; | |
00258e36 | 311 | } |
487f84f3 | 312 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 313 | } |
dda35b8f | 314 | |
fb595814 | 315 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 316 | |
fb595814 | 317 | ret = generic_file_aio_read(iocb, iovp, nr_segs, pos); |
dda35b8f CH |
318 | if (ret > 0) |
319 | XFS_STATS_ADD(xs_read_bytes, ret); | |
320 | ||
487f84f3 | 321 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
322 | return ret; |
323 | } | |
324 | ||
00258e36 CH |
325 | STATIC ssize_t |
326 | xfs_file_splice_read( | |
dda35b8f CH |
327 | struct file *infilp, |
328 | loff_t *ppos, | |
329 | struct pipe_inode_info *pipe, | |
330 | size_t count, | |
00258e36 | 331 | unsigned int flags) |
dda35b8f | 332 | { |
00258e36 | 333 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 334 | int ioflags = 0; |
dda35b8f CH |
335 | ssize_t ret; |
336 | ||
337 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
338 | |
339 | if (infilp->f_mode & FMODE_NOCMTIME) | |
340 | ioflags |= IO_INVIS; | |
341 | ||
dda35b8f CH |
342 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
343 | return -EIO; | |
344 | ||
487f84f3 | 345 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 346 | |
dda35b8f CH |
347 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
348 | ||
349 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
350 | if (ret > 0) | |
351 | XFS_STATS_ADD(xs_read_bytes, ret); | |
352 | ||
487f84f3 | 353 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
354 | return ret; |
355 | } | |
356 | ||
487f84f3 DC |
357 | /* |
358 | * xfs_file_splice_write() does not use xfs_rw_ilock() because | |
359 | * generic_file_splice_write() takes the i_mutex itself. This, in theory, | |
360 | * couuld cause lock inversions between the aio_write path and the splice path | |
361 | * if someone is doing concurrent splice(2) based writes and write(2) based | |
362 | * writes to the same inode. The only real way to fix this is to re-implement | |
363 | * the generic code here with correct locking orders. | |
364 | */ | |
00258e36 CH |
365 | STATIC ssize_t |
366 | xfs_file_splice_write( | |
dda35b8f CH |
367 | struct pipe_inode_info *pipe, |
368 | struct file *outfilp, | |
369 | loff_t *ppos, | |
370 | size_t count, | |
00258e36 | 371 | unsigned int flags) |
dda35b8f | 372 | { |
dda35b8f | 373 | struct inode *inode = outfilp->f_mapping->host; |
00258e36 | 374 | struct xfs_inode *ip = XFS_I(inode); |
00258e36 CH |
375 | int ioflags = 0; |
376 | ssize_t ret; | |
dda35b8f CH |
377 | |
378 | XFS_STATS_INC(xs_write_calls); | |
00258e36 CH |
379 | |
380 | if (outfilp->f_mode & FMODE_NOCMTIME) | |
381 | ioflags |= IO_INVIS; | |
382 | ||
dda35b8f CH |
383 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
384 | return -EIO; | |
385 | ||
386 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
387 | ||
dda35b8f CH |
388 | trace_xfs_file_splice_write(ip, count, *ppos, ioflags); |
389 | ||
390 | ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags); | |
ce7ae151 CH |
391 | if (ret > 0) |
392 | XFS_STATS_ADD(xs_write_bytes, ret); | |
dda35b8f | 393 | |
dda35b8f CH |
394 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
395 | return ret; | |
396 | } | |
397 | ||
398 | /* | |
193aec10 CH |
399 | * This routine is called to handle zeroing any space in the last block of the |
400 | * file that is beyond the EOF. We do this since the size is being increased | |
401 | * without writing anything to that block and we don't want to read the | |
402 | * garbage on the disk. | |
dda35b8f CH |
403 | */ |
404 | STATIC int /* error (positive) */ | |
405 | xfs_zero_last_block( | |
193aec10 CH |
406 | struct xfs_inode *ip, |
407 | xfs_fsize_t offset, | |
408 | xfs_fsize_t isize) | |
dda35b8f | 409 | { |
193aec10 CH |
410 | struct xfs_mount *mp = ip->i_mount; |
411 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
412 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
413 | int zero_len; | |
414 | int nimaps = 1; | |
415 | int error = 0; | |
416 | struct xfs_bmbt_irec imap; | |
dda35b8f | 417 | |
193aec10 | 418 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 419 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 420 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 421 | if (error) |
dda35b8f | 422 | return error; |
193aec10 | 423 | |
dda35b8f | 424 | ASSERT(nimaps > 0); |
193aec10 | 425 | |
dda35b8f CH |
426 | /* |
427 | * If the block underlying isize is just a hole, then there | |
428 | * is nothing to zero. | |
429 | */ | |
193aec10 | 430 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 431 | return 0; |
dda35b8f CH |
432 | |
433 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
434 | if (isize + zero_len > offset) | |
435 | zero_len = offset - isize; | |
193aec10 | 436 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
437 | } |
438 | ||
439 | /* | |
193aec10 CH |
440 | * Zero any on disk space between the current EOF and the new, larger EOF. |
441 | * | |
442 | * This handles the normal case of zeroing the remainder of the last block in | |
443 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
444 | * file. This second case only happens with fixed size extents and when the | |
445 | * system crashes before the inode size was updated but after blocks were | |
446 | * allocated. | |
447 | * | |
448 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 449 | */ |
dda35b8f CH |
450 | int /* error (positive) */ |
451 | xfs_zero_eof( | |
193aec10 CH |
452 | struct xfs_inode *ip, |
453 | xfs_off_t offset, /* starting I/O offset */ | |
454 | xfs_fsize_t isize) /* current inode size */ | |
dda35b8f | 455 | { |
193aec10 CH |
456 | struct xfs_mount *mp = ip->i_mount; |
457 | xfs_fileoff_t start_zero_fsb; | |
458 | xfs_fileoff_t end_zero_fsb; | |
459 | xfs_fileoff_t zero_count_fsb; | |
460 | xfs_fileoff_t last_fsb; | |
461 | xfs_fileoff_t zero_off; | |
462 | xfs_fsize_t zero_len; | |
463 | int nimaps; | |
464 | int error = 0; | |
465 | struct xfs_bmbt_irec imap; | |
466 | ||
467 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
468 | ASSERT(offset > isize); |
469 | ||
470 | /* | |
471 | * First handle zeroing the block on which isize resides. | |
193aec10 | 472 | * |
dda35b8f CH |
473 | * We only zero a part of that block so it is handled specially. |
474 | */ | |
193aec10 CH |
475 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
476 | error = xfs_zero_last_block(ip, offset, isize); | |
477 | if (error) | |
478 | return error; | |
dda35b8f CH |
479 | } |
480 | ||
481 | /* | |
193aec10 CH |
482 | * Calculate the range between the new size and the old where blocks |
483 | * needing to be zeroed may exist. | |
484 | * | |
485 | * To get the block where the last byte in the file currently resides, | |
486 | * we need to subtract one from the size and truncate back to a block | |
487 | * boundary. We subtract 1 in case the size is exactly on a block | |
488 | * boundary. | |
dda35b8f CH |
489 | */ |
490 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
491 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
492 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
493 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
494 | if (last_fsb == end_zero_fsb) { | |
495 | /* | |
496 | * The size was only incremented on its last block. | |
497 | * We took care of that above, so just return. | |
498 | */ | |
499 | return 0; | |
500 | } | |
501 | ||
502 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
503 | while (start_zero_fsb <= end_zero_fsb) { | |
504 | nimaps = 1; | |
505 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
506 | |
507 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
508 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
509 | &imap, &nimaps, 0); | |
193aec10 CH |
510 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
511 | if (error) | |
dda35b8f | 512 | return error; |
193aec10 | 513 | |
dda35b8f CH |
514 | ASSERT(nimaps > 0); |
515 | ||
516 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
517 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
518 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
519 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
520 | continue; | |
521 | } | |
522 | ||
523 | /* | |
524 | * There are blocks we need to zero. | |
dda35b8f | 525 | */ |
dda35b8f CH |
526 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
527 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
528 | ||
529 | if ((zero_off + zero_len) > offset) | |
530 | zero_len = offset - zero_off; | |
531 | ||
532 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
533 | if (error) |
534 | return error; | |
dda35b8f CH |
535 | |
536 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
537 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
538 | } |
539 | ||
540 | return 0; | |
dda35b8f CH |
541 | } |
542 | ||
4d8d1581 DC |
543 | /* |
544 | * Common pre-write limit and setup checks. | |
545 | * | |
5bf1f262 CH |
546 | * Called with the iolocked held either shared and exclusive according to |
547 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
548 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
549 | */ |
550 | STATIC ssize_t | |
551 | xfs_file_aio_write_checks( | |
552 | struct file *file, | |
553 | loff_t *pos, | |
554 | size_t *count, | |
555 | int *iolock) | |
556 | { | |
557 | struct inode *inode = file->f_mapping->host; | |
558 | struct xfs_inode *ip = XFS_I(inode); | |
4d8d1581 DC |
559 | int error = 0; |
560 | ||
7271d243 | 561 | restart: |
4d8d1581 | 562 | error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); |
467f7899 | 563 | if (error) |
4d8d1581 | 564 | return error; |
4d8d1581 | 565 | |
4d8d1581 DC |
566 | /* |
567 | * If the offset is beyond the size of the file, we need to zero any | |
568 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 569 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
570 | * iolock shared, we need to update it to exclusive which implies |
571 | * having to redo all checks before. | |
4d8d1581 | 572 | */ |
2813d682 | 573 | if (*pos > i_size_read(inode)) { |
7271d243 | 574 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 575 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 576 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 577 | xfs_rw_ilock(ip, *iolock); |
7271d243 DC |
578 | goto restart; |
579 | } | |
ce7ae151 | 580 | error = -xfs_zero_eof(ip, *pos, i_size_read(inode)); |
467f7899 CH |
581 | if (error) |
582 | return error; | |
7271d243 | 583 | } |
4d8d1581 | 584 | |
8a9c9980 CH |
585 | /* |
586 | * Updating the timestamps will grab the ilock again from | |
587 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
588 | * lock above. Eventually we should look into a way to avoid | |
589 | * the pointless lock roundtrip. | |
590 | */ | |
c3b2da31 JB |
591 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
592 | error = file_update_time(file); | |
593 | if (error) | |
594 | return error; | |
595 | } | |
8a9c9980 | 596 | |
4d8d1581 DC |
597 | /* |
598 | * If we're writing the file then make sure to clear the setuid and | |
599 | * setgid bits if the process is not being run by root. This keeps | |
600 | * people from modifying setuid and setgid binaries. | |
601 | */ | |
602 | return file_remove_suid(file); | |
4d8d1581 DC |
603 | } |
604 | ||
f0d26e86 DC |
605 | /* |
606 | * xfs_file_dio_aio_write - handle direct IO writes | |
607 | * | |
608 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 609 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
610 | * follow locking changes and looping. |
611 | * | |
eda77982 DC |
612 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
613 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
614 | * pages are flushed out. | |
615 | * | |
616 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
617 | * allowing them to be done in parallel with reads and other direct IO writes. | |
618 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
619 | * needs to do sub-block zeroing and that requires serialisation against other | |
620 | * direct IOs to the same block. In this case we need to serialise the | |
621 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
622 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
623 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
624 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 625 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 626 | * |
f0d26e86 DC |
627 | * Returns with locks held indicated by @iolock and errors indicated by |
628 | * negative return values. | |
629 | */ | |
630 | STATIC ssize_t | |
631 | xfs_file_dio_aio_write( | |
632 | struct kiocb *iocb, | |
633 | const struct iovec *iovp, | |
634 | unsigned long nr_segs, | |
635 | loff_t pos, | |
d0606464 | 636 | size_t ocount) |
f0d26e86 DC |
637 | { |
638 | struct file *file = iocb->ki_filp; | |
639 | struct address_space *mapping = file->f_mapping; | |
640 | struct inode *inode = mapping->host; | |
641 | struct xfs_inode *ip = XFS_I(inode); | |
642 | struct xfs_mount *mp = ip->i_mount; | |
643 | ssize_t ret = 0; | |
f0d26e86 | 644 | size_t count = ocount; |
eda77982 | 645 | int unaligned_io = 0; |
d0606464 | 646 | int iolock; |
f0d26e86 DC |
647 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
648 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
649 | ||
f0d26e86 DC |
650 | if ((pos & target->bt_smask) || (count & target->bt_smask)) |
651 | return -XFS_ERROR(EINVAL); | |
652 | ||
eda77982 DC |
653 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
654 | unaligned_io = 1; | |
655 | ||
7271d243 DC |
656 | /* |
657 | * We don't need to take an exclusive lock unless there page cache needs | |
658 | * to be invalidated or unaligned IO is being executed. We don't need to | |
659 | * consider the EOF extension case here because | |
660 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
661 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
662 | */ | |
663 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 664 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 665 | else |
d0606464 CH |
666 | iolock = XFS_IOLOCK_SHARED; |
667 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
668 | |
669 | /* | |
670 | * Recheck if there are cached pages that need invalidate after we got | |
671 | * the iolock to protect against other threads adding new pages while | |
672 | * we were waiting for the iolock. | |
673 | */ | |
d0606464 CH |
674 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
675 | xfs_rw_iunlock(ip, iolock); | |
676 | iolock = XFS_IOLOCK_EXCL; | |
677 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 678 | } |
f0d26e86 | 679 | |
d0606464 | 680 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 681 | if (ret) |
d0606464 | 682 | goto out; |
f0d26e86 DC |
683 | |
684 | if (mapping->nrpages) { | |
fb595814 DC |
685 | ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
686 | pos, -1); | |
f0d26e86 | 687 | if (ret) |
d0606464 | 688 | goto out; |
1025b461 DC |
689 | /* |
690 | * Invalidate whole pages. This can return an error if | |
691 | * we fail to invalidate a page, but this should never | |
692 | * happen on XFS. Warn if it does fail. | |
693 | */ | |
694 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, | |
695 | pos >> PAGE_CACHE_SHIFT, -1); | |
696 | WARN_ON_ONCE(ret); | |
697 | ret = 0; | |
f0d26e86 DC |
698 | } |
699 | ||
eda77982 DC |
700 | /* |
701 | * If we are doing unaligned IO, wait for all other IO to drain, | |
702 | * otherwise demote the lock if we had to flush cached pages | |
703 | */ | |
704 | if (unaligned_io) | |
4a06fd26 | 705 | inode_dio_wait(inode); |
d0606464 | 706 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 707 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 708 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
709 | } |
710 | ||
711 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
712 | ret = generic_file_direct_write(iocb, iovp, | |
713 | &nr_segs, pos, &iocb->ki_pos, count, ocount); | |
714 | ||
d0606464 CH |
715 | out: |
716 | xfs_rw_iunlock(ip, iolock); | |
717 | ||
f0d26e86 DC |
718 | /* No fallback to buffered IO on errors for XFS. */ |
719 | ASSERT(ret < 0 || ret == count); | |
720 | return ret; | |
721 | } | |
722 | ||
00258e36 | 723 | STATIC ssize_t |
637bbc75 | 724 | xfs_file_buffered_aio_write( |
dda35b8f CH |
725 | struct kiocb *iocb, |
726 | const struct iovec *iovp, | |
00258e36 | 727 | unsigned long nr_segs, |
637bbc75 | 728 | loff_t pos, |
d0606464 | 729 | size_t ocount) |
dda35b8f CH |
730 | { |
731 | struct file *file = iocb->ki_filp; | |
732 | struct address_space *mapping = file->f_mapping; | |
733 | struct inode *inode = mapping->host; | |
00258e36 | 734 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
735 | ssize_t ret; |
736 | int enospc = 0; | |
d0606464 | 737 | int iolock = XFS_IOLOCK_EXCL; |
637bbc75 | 738 | size_t count = ocount; |
dda35b8f | 739 | |
d0606464 | 740 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 741 | |
d0606464 | 742 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 743 | if (ret) |
d0606464 | 744 | goto out; |
dda35b8f CH |
745 | |
746 | /* We can write back this queue in page reclaim */ | |
747 | current->backing_dev_info = mapping->backing_dev_info; | |
748 | ||
dda35b8f | 749 | write_retry: |
637bbc75 DC |
750 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); |
751 | ret = generic_file_buffered_write(iocb, iovp, nr_segs, | |
9aa05000 DC |
752 | pos, &iocb->ki_pos, count, 0); |
753 | ||
637bbc75 | 754 | /* |
9aa05000 DC |
755 | * If we just got an ENOSPC, try to write back all dirty inodes to |
756 | * convert delalloc space to free up some of the excess reserved | |
757 | * metadata space. | |
637bbc75 DC |
758 | */ |
759 | if (ret == -ENOSPC && !enospc) { | |
637bbc75 | 760 | enospc = 1; |
9aa05000 DC |
761 | xfs_flush_inodes(ip->i_mount); |
762 | goto write_retry; | |
dda35b8f | 763 | } |
d0606464 | 764 | |
dda35b8f | 765 | current->backing_dev_info = NULL; |
d0606464 CH |
766 | out: |
767 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
768 | return ret; |
769 | } | |
770 | ||
771 | STATIC ssize_t | |
772 | xfs_file_aio_write( | |
773 | struct kiocb *iocb, | |
774 | const struct iovec *iovp, | |
775 | unsigned long nr_segs, | |
776 | loff_t pos) | |
777 | { | |
778 | struct file *file = iocb->ki_filp; | |
779 | struct address_space *mapping = file->f_mapping; | |
780 | struct inode *inode = mapping->host; | |
781 | struct xfs_inode *ip = XFS_I(inode); | |
782 | ssize_t ret; | |
637bbc75 DC |
783 | size_t ocount = 0; |
784 | ||
785 | XFS_STATS_INC(xs_write_calls); | |
786 | ||
787 | BUG_ON(iocb->ki_pos != pos); | |
788 | ||
789 | ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ); | |
790 | if (ret) | |
791 | return ret; | |
792 | ||
793 | if (ocount == 0) | |
794 | return 0; | |
795 | ||
d9457dc0 JK |
796 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
797 | ret = -EIO; | |
798 | goto out; | |
799 | } | |
637bbc75 DC |
800 | |
801 | if (unlikely(file->f_flags & O_DIRECT)) | |
d0606464 | 802 | ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount); |
637bbc75 DC |
803 | else |
804 | ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos, | |
d0606464 | 805 | ocount); |
dda35b8f | 806 | |
d0606464 CH |
807 | if (ret > 0) { |
808 | ssize_t err; | |
dda35b8f | 809 | |
d0606464 | 810 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 811 | |
d0606464 CH |
812 | /* Handle various SYNC-type writes */ |
813 | err = generic_write_sync(file, pos, ret); | |
814 | if (err < 0) | |
815 | ret = err; | |
dda35b8f CH |
816 | } |
817 | ||
d9457dc0 | 818 | out: |
a363f0c2 | 819 | return ret; |
dda35b8f CH |
820 | } |
821 | ||
2fe17c10 CH |
822 | STATIC long |
823 | xfs_file_fallocate( | |
824 | struct file *file, | |
825 | int mode, | |
826 | loff_t offset, | |
827 | loff_t len) | |
828 | { | |
496ad9aa | 829 | struct inode *inode = file_inode(file); |
2fe17c10 CH |
830 | long error; |
831 | loff_t new_size = 0; | |
832 | xfs_flock64_t bf; | |
833 | xfs_inode_t *ip = XFS_I(inode); | |
834 | int cmd = XFS_IOC_RESVSP; | |
82878897 | 835 | int attr_flags = XFS_ATTR_NOLOCK; |
2fe17c10 CH |
836 | |
837 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
838 | return -EOPNOTSUPP; | |
839 | ||
840 | bf.l_whence = 0; | |
841 | bf.l_start = offset; | |
842 | bf.l_len = len; | |
843 | ||
844 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
845 | ||
846 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
847 | cmd = XFS_IOC_UNRESVSP; | |
848 | ||
849 | /* check the new inode size is valid before allocating */ | |
850 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
851 | offset + len > i_size_read(inode)) { | |
852 | new_size = offset + len; | |
853 | error = inode_newsize_ok(inode, new_size); | |
854 | if (error) | |
855 | goto out_unlock; | |
856 | } | |
857 | ||
82878897 DC |
858 | if (file->f_flags & O_DSYNC) |
859 | attr_flags |= XFS_ATTR_SYNC; | |
860 | ||
861 | error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags); | |
2fe17c10 CH |
862 | if (error) |
863 | goto out_unlock; | |
864 | ||
865 | /* Change file size if needed */ | |
866 | if (new_size) { | |
867 | struct iattr iattr; | |
868 | ||
869 | iattr.ia_valid = ATTR_SIZE; | |
870 | iattr.ia_size = new_size; | |
c4ed4243 | 871 | error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK); |
2fe17c10 CH |
872 | } |
873 | ||
874 | out_unlock: | |
875 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
876 | return error; | |
877 | } | |
878 | ||
879 | ||
1da177e4 | 880 | STATIC int |
3562fd45 | 881 | xfs_file_open( |
1da177e4 | 882 | struct inode *inode, |
f999a5bf | 883 | struct file *file) |
1da177e4 | 884 | { |
f999a5bf | 885 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 886 | return -EFBIG; |
f999a5bf CH |
887 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
888 | return -EIO; | |
889 | return 0; | |
890 | } | |
891 | ||
892 | STATIC int | |
893 | xfs_dir_open( | |
894 | struct inode *inode, | |
895 | struct file *file) | |
896 | { | |
897 | struct xfs_inode *ip = XFS_I(inode); | |
898 | int mode; | |
899 | int error; | |
900 | ||
901 | error = xfs_file_open(inode, file); | |
902 | if (error) | |
903 | return error; | |
904 | ||
905 | /* | |
906 | * If there are any blocks, read-ahead block 0 as we're almost | |
907 | * certain to have the next operation be a read there. | |
908 | */ | |
909 | mode = xfs_ilock_map_shared(ip); | |
910 | if (ip->i_d.di_nextents > 0) | |
33363fee | 911 | xfs_dir3_data_readahead(NULL, ip, 0, -1); |
f999a5bf CH |
912 | xfs_iunlock(ip, mode); |
913 | return 0; | |
1da177e4 LT |
914 | } |
915 | ||
1da177e4 | 916 | STATIC int |
3562fd45 | 917 | xfs_file_release( |
1da177e4 LT |
918 | struct inode *inode, |
919 | struct file *filp) | |
920 | { | |
739bfb2a | 921 | return -xfs_release(XFS_I(inode)); |
1da177e4 LT |
922 | } |
923 | ||
1da177e4 | 924 | STATIC int |
3562fd45 | 925 | xfs_file_readdir( |
1da177e4 LT |
926 | struct file *filp, |
927 | void *dirent, | |
928 | filldir_t filldir) | |
929 | { | |
496ad9aa | 930 | struct inode *inode = file_inode(filp); |
739bfb2a | 931 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
932 | int error; |
933 | size_t bufsize; | |
934 | ||
935 | /* | |
936 | * The Linux API doesn't pass down the total size of the buffer | |
937 | * we read into down to the filesystem. With the filldir concept | |
938 | * it's not needed for correct information, but the XFS dir2 leaf | |
939 | * code wants an estimate of the buffer size to calculate it's | |
940 | * readahead window and size the buffers used for mapping to | |
941 | * physical blocks. | |
942 | * | |
943 | * Try to give it an estimate that's good enough, maybe at some | |
944 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 945 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 946 | */ |
a9cc799e | 947 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 948 | |
739bfb2a | 949 | error = xfs_readdir(ip, dirent, bufsize, |
051e7cd4 CH |
950 | (xfs_off_t *)&filp->f_pos, filldir); |
951 | if (error) | |
952 | return -error; | |
953 | return 0; | |
1da177e4 LT |
954 | } |
955 | ||
1da177e4 | 956 | STATIC int |
3562fd45 | 957 | xfs_file_mmap( |
1da177e4 LT |
958 | struct file *filp, |
959 | struct vm_area_struct *vma) | |
960 | { | |
3562fd45 | 961 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 962 | |
fbc1462b | 963 | file_accessed(filp); |
1da177e4 LT |
964 | return 0; |
965 | } | |
966 | ||
4f57dbc6 DC |
967 | /* |
968 | * mmap()d file has taken write protection fault and is being made | |
969 | * writable. We can set the page state up correctly for a writable | |
970 | * page, which means we can do correct delalloc accounting (ENOSPC | |
971 | * checking!) and unwritten extent mapping. | |
972 | */ | |
973 | STATIC int | |
974 | xfs_vm_page_mkwrite( | |
975 | struct vm_area_struct *vma, | |
c2ec175c | 976 | struct vm_fault *vmf) |
4f57dbc6 | 977 | { |
c2ec175c | 978 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
979 | } |
980 | ||
d126d43f JL |
981 | /* |
982 | * This type is designed to indicate the type of offset we would like | |
983 | * to search from page cache for either xfs_seek_data() or xfs_seek_hole(). | |
984 | */ | |
985 | enum { | |
986 | HOLE_OFF = 0, | |
987 | DATA_OFF, | |
988 | }; | |
989 | ||
990 | /* | |
991 | * Lookup the desired type of offset from the given page. | |
992 | * | |
993 | * On success, return true and the offset argument will point to the | |
994 | * start of the region that was found. Otherwise this function will | |
995 | * return false and keep the offset argument unchanged. | |
996 | */ | |
997 | STATIC bool | |
998 | xfs_lookup_buffer_offset( | |
999 | struct page *page, | |
1000 | loff_t *offset, | |
1001 | unsigned int type) | |
1002 | { | |
1003 | loff_t lastoff = page_offset(page); | |
1004 | bool found = false; | |
1005 | struct buffer_head *bh, *head; | |
1006 | ||
1007 | bh = head = page_buffers(page); | |
1008 | do { | |
1009 | /* | |
1010 | * Unwritten extents that have data in the page | |
1011 | * cache covering them can be identified by the | |
1012 | * BH_Unwritten state flag. Pages with multiple | |
1013 | * buffers might have a mix of holes, data and | |
1014 | * unwritten extents - any buffer with valid | |
1015 | * data in it should have BH_Uptodate flag set | |
1016 | * on it. | |
1017 | */ | |
1018 | if (buffer_unwritten(bh) || | |
1019 | buffer_uptodate(bh)) { | |
1020 | if (type == DATA_OFF) | |
1021 | found = true; | |
1022 | } else { | |
1023 | if (type == HOLE_OFF) | |
1024 | found = true; | |
1025 | } | |
1026 | ||
1027 | if (found) { | |
1028 | *offset = lastoff; | |
1029 | break; | |
1030 | } | |
1031 | lastoff += bh->b_size; | |
1032 | } while ((bh = bh->b_this_page) != head); | |
1033 | ||
1034 | return found; | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * This routine is called to find out and return a data or hole offset | |
1039 | * from the page cache for unwritten extents according to the desired | |
1040 | * type for xfs_seek_data() or xfs_seek_hole(). | |
1041 | * | |
1042 | * The argument offset is used to tell where we start to search from the | |
1043 | * page cache. Map is used to figure out the end points of the range to | |
1044 | * lookup pages. | |
1045 | * | |
1046 | * Return true if the desired type of offset was found, and the argument | |
1047 | * offset is filled with that address. Otherwise, return false and keep | |
1048 | * offset unchanged. | |
1049 | */ | |
1050 | STATIC bool | |
1051 | xfs_find_get_desired_pgoff( | |
1052 | struct inode *inode, | |
1053 | struct xfs_bmbt_irec *map, | |
1054 | unsigned int type, | |
1055 | loff_t *offset) | |
1056 | { | |
1057 | struct xfs_inode *ip = XFS_I(inode); | |
1058 | struct xfs_mount *mp = ip->i_mount; | |
1059 | struct pagevec pvec; | |
1060 | pgoff_t index; | |
1061 | pgoff_t end; | |
1062 | loff_t endoff; | |
1063 | loff_t startoff = *offset; | |
1064 | loff_t lastoff = startoff; | |
1065 | bool found = false; | |
1066 | ||
1067 | pagevec_init(&pvec, 0); | |
1068 | ||
1069 | index = startoff >> PAGE_CACHE_SHIFT; | |
1070 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1071 | end = endoff >> PAGE_CACHE_SHIFT; | |
1072 | do { | |
1073 | int want; | |
1074 | unsigned nr_pages; | |
1075 | unsigned int i; | |
1076 | ||
1077 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1078 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1079 | want); | |
1080 | /* | |
1081 | * No page mapped into given range. If we are searching holes | |
1082 | * and if this is the first time we got into the loop, it means | |
1083 | * that the given offset is landed in a hole, return it. | |
1084 | * | |
1085 | * If we have already stepped through some block buffers to find | |
1086 | * holes but they all contains data. In this case, the last | |
1087 | * offset is already updated and pointed to the end of the last | |
1088 | * mapped page, if it does not reach the endpoint to search, | |
1089 | * that means there should be a hole between them. | |
1090 | */ | |
1091 | if (nr_pages == 0) { | |
1092 | /* Data search found nothing */ | |
1093 | if (type == DATA_OFF) | |
1094 | break; | |
1095 | ||
1096 | ASSERT(type == HOLE_OFF); | |
1097 | if (lastoff == startoff || lastoff < endoff) { | |
1098 | found = true; | |
1099 | *offset = lastoff; | |
1100 | } | |
1101 | break; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * At lease we found one page. If this is the first time we | |
1106 | * step into the loop, and if the first page index offset is | |
1107 | * greater than the given search offset, a hole was found. | |
1108 | */ | |
1109 | if (type == HOLE_OFF && lastoff == startoff && | |
1110 | lastoff < page_offset(pvec.pages[0])) { | |
1111 | found = true; | |
1112 | break; | |
1113 | } | |
1114 | ||
1115 | for (i = 0; i < nr_pages; i++) { | |
1116 | struct page *page = pvec.pages[i]; | |
1117 | loff_t b_offset; | |
1118 | ||
1119 | /* | |
1120 | * At this point, the page may be truncated or | |
1121 | * invalidated (changing page->mapping to NULL), | |
1122 | * or even swizzled back from swapper_space to tmpfs | |
1123 | * file mapping. However, page->index will not change | |
1124 | * because we have a reference on the page. | |
1125 | * | |
1126 | * Searching done if the page index is out of range. | |
1127 | * If the current offset is not reaches the end of | |
1128 | * the specified search range, there should be a hole | |
1129 | * between them. | |
1130 | */ | |
1131 | if (page->index > end) { | |
1132 | if (type == HOLE_OFF && lastoff < endoff) { | |
1133 | *offset = lastoff; | |
1134 | found = true; | |
1135 | } | |
1136 | goto out; | |
1137 | } | |
1138 | ||
1139 | lock_page(page); | |
1140 | /* | |
1141 | * Page truncated or invalidated(page->mapping == NULL). | |
1142 | * We can freely skip it and proceed to check the next | |
1143 | * page. | |
1144 | */ | |
1145 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1146 | unlock_page(page); | |
1147 | continue; | |
1148 | } | |
1149 | ||
1150 | if (!page_has_buffers(page)) { | |
1151 | unlock_page(page); | |
1152 | continue; | |
1153 | } | |
1154 | ||
1155 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1156 | if (found) { | |
1157 | /* | |
1158 | * The found offset may be less than the start | |
1159 | * point to search if this is the first time to | |
1160 | * come here. | |
1161 | */ | |
1162 | *offset = max_t(loff_t, startoff, b_offset); | |
1163 | unlock_page(page); | |
1164 | goto out; | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * We either searching data but nothing was found, or | |
1169 | * searching hole but found a data buffer. In either | |
1170 | * case, probably the next page contains the desired | |
1171 | * things, update the last offset to it so. | |
1172 | */ | |
1173 | lastoff = page_offset(page) + PAGE_SIZE; | |
1174 | unlock_page(page); | |
1175 | } | |
1176 | ||
1177 | /* | |
1178 | * The number of returned pages less than our desired, search | |
1179 | * done. In this case, nothing was found for searching data, | |
1180 | * but we found a hole behind the last offset. | |
1181 | */ | |
1182 | if (nr_pages < want) { | |
1183 | if (type == HOLE_OFF) { | |
1184 | *offset = lastoff; | |
1185 | found = true; | |
1186 | } | |
1187 | break; | |
1188 | } | |
1189 | ||
1190 | index = pvec.pages[i - 1]->index + 1; | |
1191 | pagevec_release(&pvec); | |
1192 | } while (index <= end); | |
1193 | ||
1194 | out: | |
1195 | pagevec_release(&pvec); | |
1196 | return found; | |
1197 | } | |
1198 | ||
3fe3e6b1 JL |
1199 | STATIC loff_t |
1200 | xfs_seek_data( | |
1201 | struct file *file, | |
834ab122 | 1202 | loff_t start) |
3fe3e6b1 JL |
1203 | { |
1204 | struct inode *inode = file->f_mapping->host; | |
1205 | struct xfs_inode *ip = XFS_I(inode); | |
1206 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1207 | loff_t uninitialized_var(offset); |
1208 | xfs_fsize_t isize; | |
1209 | xfs_fileoff_t fsbno; | |
1210 | xfs_filblks_t end; | |
1211 | uint lock; | |
1212 | int error; | |
1213 | ||
1214 | lock = xfs_ilock_map_shared(ip); | |
1215 | ||
1216 | isize = i_size_read(inode); | |
1217 | if (start >= isize) { | |
1218 | error = ENXIO; | |
1219 | goto out_unlock; | |
1220 | } | |
1221 | ||
3fe3e6b1 JL |
1222 | /* |
1223 | * Try to read extents from the first block indicated | |
1224 | * by fsbno to the end block of the file. | |
1225 | */ | |
52f1acc8 | 1226 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1227 | end = XFS_B_TO_FSB(mp, isize); |
52f1acc8 JL |
1228 | for (;;) { |
1229 | struct xfs_bmbt_irec map[2]; | |
1230 | int nmap = 2; | |
1231 | unsigned int i; | |
3fe3e6b1 | 1232 | |
52f1acc8 JL |
1233 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1234 | XFS_BMAPI_ENTIRE); | |
1235 | if (error) | |
1236 | goto out_unlock; | |
3fe3e6b1 | 1237 | |
52f1acc8 JL |
1238 | /* No extents at given offset, must be beyond EOF */ |
1239 | if (nmap == 0) { | |
1240 | error = ENXIO; | |
1241 | goto out_unlock; | |
1242 | } | |
1243 | ||
1244 | for (i = 0; i < nmap; i++) { | |
1245 | offset = max_t(loff_t, start, | |
1246 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1247 | ||
1248 | /* Landed in a data extent */ | |
1249 | if (map[i].br_startblock == DELAYSTARTBLOCK || | |
1250 | (map[i].br_state == XFS_EXT_NORM && | |
1251 | !isnullstartblock(map[i].br_startblock))) | |
1252 | goto out; | |
1253 | ||
1254 | /* | |
1255 | * Landed in an unwritten extent, try to search data | |
1256 | * from page cache. | |
1257 | */ | |
1258 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1259 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
1260 | DATA_OFF, &offset)) | |
1261 | goto out; | |
1262 | } | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * map[0] is hole or its an unwritten extent but | |
1267 | * without data in page cache. Probably means that | |
1268 | * we are reading after EOF if nothing in map[1]. | |
1269 | */ | |
3fe3e6b1 JL |
1270 | if (nmap == 1) { |
1271 | error = ENXIO; | |
1272 | goto out_unlock; | |
1273 | } | |
1274 | ||
52f1acc8 JL |
1275 | ASSERT(i > 1); |
1276 | ||
1277 | /* | |
1278 | * Nothing was found, proceed to the next round of search | |
1279 | * if reading offset not beyond or hit EOF. | |
1280 | */ | |
1281 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1282 | start = XFS_FSB_TO_B(mp, fsbno); | |
1283 | if (start >= isize) { | |
1284 | error = ENXIO; | |
1285 | goto out_unlock; | |
1286 | } | |
3fe3e6b1 JL |
1287 | } |
1288 | ||
52f1acc8 | 1289 | out: |
3fe3e6b1 JL |
1290 | if (offset != file->f_pos) |
1291 | file->f_pos = offset; | |
1292 | ||
1293 | out_unlock: | |
1294 | xfs_iunlock_map_shared(ip, lock); | |
1295 | ||
1296 | if (error) | |
1297 | return -error; | |
1298 | return offset; | |
1299 | } | |
1300 | ||
1301 | STATIC loff_t | |
1302 | xfs_seek_hole( | |
1303 | struct file *file, | |
834ab122 | 1304 | loff_t start) |
3fe3e6b1 JL |
1305 | { |
1306 | struct inode *inode = file->f_mapping->host; | |
1307 | struct xfs_inode *ip = XFS_I(inode); | |
1308 | struct xfs_mount *mp = ip->i_mount; | |
1309 | loff_t uninitialized_var(offset); | |
3fe3e6b1 JL |
1310 | xfs_fsize_t isize; |
1311 | xfs_fileoff_t fsbno; | |
b686d1f7 | 1312 | xfs_filblks_t end; |
3fe3e6b1 JL |
1313 | uint lock; |
1314 | int error; | |
1315 | ||
1316 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1317 | return -XFS_ERROR(EIO); | |
1318 | ||
1319 | lock = xfs_ilock_map_shared(ip); | |
1320 | ||
1321 | isize = i_size_read(inode); | |
1322 | if (start >= isize) { | |
1323 | error = ENXIO; | |
1324 | goto out_unlock; | |
1325 | } | |
1326 | ||
1327 | fsbno = XFS_B_TO_FSBT(mp, start); | |
b686d1f7 JL |
1328 | end = XFS_B_TO_FSB(mp, isize); |
1329 | ||
1330 | for (;;) { | |
1331 | struct xfs_bmbt_irec map[2]; | |
1332 | int nmap = 2; | |
1333 | unsigned int i; | |
1334 | ||
1335 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, | |
1336 | XFS_BMAPI_ENTIRE); | |
1337 | if (error) | |
1338 | goto out_unlock; | |
1339 | ||
1340 | /* No extents at given offset, must be beyond EOF */ | |
1341 | if (nmap == 0) { | |
1342 | error = ENXIO; | |
1343 | goto out_unlock; | |
1344 | } | |
1345 | ||
1346 | for (i = 0; i < nmap; i++) { | |
1347 | offset = max_t(loff_t, start, | |
1348 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1349 | ||
1350 | /* Landed in a hole */ | |
1351 | if (map[i].br_startblock == HOLESTARTBLOCK) | |
1352 | goto out; | |
1353 | ||
1354 | /* | |
1355 | * Landed in an unwritten extent, try to search hole | |
1356 | * from page cache. | |
1357 | */ | |
1358 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1359 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
1360 | HOLE_OFF, &offset)) | |
1361 | goto out; | |
1362 | } | |
1363 | } | |
3fe3e6b1 | 1364 | |
3fe3e6b1 | 1365 | /* |
b686d1f7 JL |
1366 | * map[0] contains data or its unwritten but contains |
1367 | * data in page cache, probably means that we are | |
1368 | * reading after EOF. We should fix offset to point | |
1369 | * to the end of the file(i.e., there is an implicit | |
1370 | * hole at the end of any file). | |
3fe3e6b1 | 1371 | */ |
b686d1f7 JL |
1372 | if (nmap == 1) { |
1373 | offset = isize; | |
1374 | break; | |
1375 | } | |
1376 | ||
1377 | ASSERT(i > 1); | |
1378 | ||
1379 | /* | |
1380 | * Both mappings contains data, proceed to the next round of | |
1381 | * search if the current reading offset not beyond or hit EOF. | |
1382 | */ | |
1383 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1384 | start = XFS_FSB_TO_B(mp, fsbno); | |
1385 | if (start >= isize) { | |
1386 | offset = isize; | |
1387 | break; | |
1388 | } | |
3fe3e6b1 JL |
1389 | } |
1390 | ||
b686d1f7 JL |
1391 | out: |
1392 | /* | |
1393 | * At this point, we must have found a hole. However, the returned | |
1394 | * offset may be bigger than the file size as it may be aligned to | |
1395 | * page boundary for unwritten extents, we need to deal with this | |
1396 | * situation in particular. | |
1397 | */ | |
1398 | offset = min_t(loff_t, offset, isize); | |
3fe3e6b1 JL |
1399 | if (offset != file->f_pos) |
1400 | file->f_pos = offset; | |
1401 | ||
1402 | out_unlock: | |
1403 | xfs_iunlock_map_shared(ip, lock); | |
1404 | ||
1405 | if (error) | |
1406 | return -error; | |
1407 | return offset; | |
1408 | } | |
1409 | ||
1410 | STATIC loff_t | |
1411 | xfs_file_llseek( | |
1412 | struct file *file, | |
1413 | loff_t offset, | |
1414 | int origin) | |
1415 | { | |
1416 | switch (origin) { | |
1417 | case SEEK_END: | |
1418 | case SEEK_CUR: | |
1419 | case SEEK_SET: | |
1420 | return generic_file_llseek(file, offset, origin); | |
1421 | case SEEK_DATA: | |
834ab122 | 1422 | return xfs_seek_data(file, offset); |
3fe3e6b1 | 1423 | case SEEK_HOLE: |
834ab122 | 1424 | return xfs_seek_hole(file, offset); |
3fe3e6b1 JL |
1425 | default: |
1426 | return -EINVAL; | |
1427 | } | |
1428 | } | |
1429 | ||
4b6f5d20 | 1430 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1431 | .llseek = xfs_file_llseek, |
1da177e4 | 1432 | .read = do_sync_read, |
bb3f724e | 1433 | .write = do_sync_write, |
3562fd45 NS |
1434 | .aio_read = xfs_file_aio_read, |
1435 | .aio_write = xfs_file_aio_write, | |
1b895840 NS |
1436 | .splice_read = xfs_file_splice_read, |
1437 | .splice_write = xfs_file_splice_write, | |
3562fd45 | 1438 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1439 | #ifdef CONFIG_COMPAT |
3562fd45 | 1440 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1441 | #endif |
3562fd45 NS |
1442 | .mmap = xfs_file_mmap, |
1443 | .open = xfs_file_open, | |
1444 | .release = xfs_file_release, | |
1445 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1446 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1447 | }; |
1448 | ||
4b6f5d20 | 1449 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1450 | .open = xfs_dir_open, |
1da177e4 | 1451 | .read = generic_read_dir, |
3562fd45 | 1452 | .readdir = xfs_file_readdir, |
59af1584 | 1453 | .llseek = generic_file_llseek, |
3562fd45 | 1454 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1455 | #ifdef CONFIG_COMPAT |
3562fd45 | 1456 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1457 | #endif |
1da2f2db | 1458 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1459 | }; |
1460 | ||
f0f37e2f | 1461 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1462 | .fault = filemap_fault, |
4f57dbc6 | 1463 | .page_mkwrite = xfs_vm_page_mkwrite, |
0b173bc4 | 1464 | .remap_pages = generic_file_remap_pages, |
6fac0cb4 | 1465 | }; |