]>
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" |
a844f451 | 20 | #include "xfs_bit.h" |
1da177e4 | 21 | #include "xfs_log.h" |
a844f451 | 22 | #include "xfs_inum.h" |
1da177e4 | 23 | #include "xfs_sb.h" |
a844f451 | 24 | #include "xfs_ag.h" |
1da177e4 | 25 | #include "xfs_trans.h" |
1da177e4 LT |
26 | #include "xfs_mount.h" |
27 | #include "xfs_bmap_btree.h" | |
1da177e4 | 28 | #include "xfs_alloc.h" |
1da177e4 LT |
29 | #include "xfs_dinode.h" |
30 | #include "xfs_inode.h" | |
fd3200be | 31 | #include "xfs_inode_item.h" |
dda35b8f | 32 | #include "xfs_bmap.h" |
1da177e4 | 33 | #include "xfs_error.h" |
739bfb2a | 34 | #include "xfs_vnodeops.h" |
f999a5bf | 35 | #include "xfs_da_btree.h" |
ddcd856d | 36 | #include "xfs_ioctl.h" |
dda35b8f | 37 | #include "xfs_trace.h" |
1da177e4 LT |
38 | |
39 | #include <linux/dcache.h> | |
1da177e4 | 40 | |
f0f37e2f | 41 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 42 | |
dda35b8f CH |
43 | /* |
44 | * xfs_iozero | |
45 | * | |
46 | * xfs_iozero clears the specified range of buffer supplied, | |
47 | * and marks all the affected blocks as valid and modified. If | |
48 | * an affected block is not allocated, it will be allocated. If | |
49 | * an affected block is not completely overwritten, and is not | |
50 | * valid before the operation, it will be read from disk before | |
51 | * being partially zeroed. | |
52 | */ | |
53 | STATIC int | |
54 | xfs_iozero( | |
55 | struct xfs_inode *ip, /* inode */ | |
56 | loff_t pos, /* offset in file */ | |
57 | size_t count) /* size of data to zero */ | |
58 | { | |
59 | struct page *page; | |
60 | struct address_space *mapping; | |
61 | int status; | |
62 | ||
63 | mapping = VFS_I(ip)->i_mapping; | |
64 | do { | |
65 | unsigned offset, bytes; | |
66 | void *fsdata; | |
67 | ||
68 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
69 | bytes = PAGE_CACHE_SIZE - offset; | |
70 | if (bytes > count) | |
71 | bytes = count; | |
72 | ||
73 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
74 | AOP_FLAG_UNINTERRUPTIBLE, | |
75 | &page, &fsdata); | |
76 | if (status) | |
77 | break; | |
78 | ||
79 | zero_user(page, offset, bytes); | |
80 | ||
81 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
82 | page, fsdata); | |
83 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
84 | pos += bytes; | |
85 | count -= bytes; | |
86 | status = 0; | |
87 | } while (count); | |
88 | ||
89 | return (-status); | |
90 | } | |
91 | ||
fd3200be CH |
92 | STATIC int |
93 | xfs_file_fsync( | |
94 | struct file *file, | |
fd3200be CH |
95 | int datasync) |
96 | { | |
7ea80859 CH |
97 | struct inode *inode = file->f_mapping->host; |
98 | struct xfs_inode *ip = XFS_I(inode); | |
fd3200be CH |
99 | struct xfs_trans *tp; |
100 | int error = 0; | |
101 | int log_flushed = 0; | |
102 | ||
cca28fb8 | 103 | trace_xfs_file_fsync(ip); |
fd3200be CH |
104 | |
105 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
106 | return -XFS_ERROR(EIO); | |
107 | ||
108 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
109 | ||
37bc5743 CH |
110 | xfs_ioend_wait(ip); |
111 | ||
fd3200be CH |
112 | /* |
113 | * We always need to make sure that the required inode state is safe on | |
114 | * disk. The inode might be clean but we still might need to force the | |
115 | * log because of committed transactions that haven't hit the disk yet. | |
116 | * Likewise, there could be unflushed non-transactional changes to the | |
117 | * inode core that have to go to disk and this requires us to issue | |
118 | * a synchronous transaction to capture these changes correctly. | |
119 | * | |
120 | * This code relies on the assumption that if the i_update_core field | |
121 | * of the inode is clear and the inode is unpinned then it is clean | |
122 | * and no action is required. | |
123 | */ | |
124 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
125 | ||
66d834ea CH |
126 | /* |
127 | * First check if the VFS inode is marked dirty. All the dirtying | |
128 | * of non-transactional updates no goes through mark_inode_dirty*, | |
129 | * which allows us to distinguish beteeen pure timestamp updates | |
130 | * and i_size updates which need to be caught for fdatasync. | |
131 | * After that also theck for the dirty state in the XFS inode, which | |
132 | * might gets cleared when the inode gets written out via the AIL | |
133 | * or xfs_iflush_cluster. | |
134 | */ | |
7ea80859 CH |
135 | if (((inode->i_state & I_DIRTY_DATASYNC) || |
136 | ((inode->i_state & I_DIRTY_SYNC) && !datasync)) && | |
66d834ea | 137 | ip->i_update_core) { |
fd3200be CH |
138 | /* |
139 | * Kick off a transaction to log the inode core to get the | |
140 | * updates. The sync transaction will also force the log. | |
141 | */ | |
142 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
143 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS); | |
144 | error = xfs_trans_reserve(tp, 0, | |
145 | XFS_FSYNC_TS_LOG_RES(ip->i_mount), 0, 0, 0); | |
146 | if (error) { | |
147 | xfs_trans_cancel(tp, 0); | |
148 | return -error; | |
149 | } | |
150 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
151 | ||
152 | /* | |
153 | * Note - it's possible that we might have pushed ourselves out | |
154 | * of the way during trans_reserve which would flush the inode. | |
155 | * But there's no guarantee that the inode buffer has actually | |
156 | * gone out yet (it's delwri). Plus the buffer could be pinned | |
157 | * anyway if it's part of an inode in another recent | |
158 | * transaction. So we play it safe and fire off the | |
159 | * transaction anyway. | |
160 | */ | |
898621d5 | 161 | xfs_trans_ijoin(tp, ip); |
fd3200be CH |
162 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
163 | xfs_trans_set_sync(tp); | |
164 | error = _xfs_trans_commit(tp, 0, &log_flushed); | |
165 | ||
166 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
167 | } else { | |
168 | /* | |
169 | * Timestamps/size haven't changed since last inode flush or | |
170 | * inode transaction commit. That means either nothing got | |
171 | * written or a transaction committed which caught the updates. | |
172 | * If the latter happened and the transaction hasn't hit the | |
173 | * disk yet, the inode will be still be pinned. If it is, | |
174 | * force the log. | |
175 | */ | |
fd3200be | 176 | if (xfs_ipincount(ip)) { |
024910cb CH |
177 | error = _xfs_log_force_lsn(ip->i_mount, |
178 | ip->i_itemp->ili_last_lsn, | |
179 | XFS_LOG_SYNC, &log_flushed); | |
fd3200be | 180 | } |
024910cb | 181 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be CH |
182 | } |
183 | ||
184 | if (ip->i_mount->m_flags & XFS_MOUNT_BARRIER) { | |
185 | /* | |
186 | * If the log write didn't issue an ordered tag we need | |
187 | * to flush the disk cache for the data device now. | |
188 | */ | |
189 | if (!log_flushed) | |
190 | xfs_blkdev_issue_flush(ip->i_mount->m_ddev_targp); | |
191 | ||
192 | /* | |
193 | * If this inode is on the RT dev we need to flush that | |
194 | * cache as well. | |
195 | */ | |
196 | if (XFS_IS_REALTIME_INODE(ip)) | |
197 | xfs_blkdev_issue_flush(ip->i_mount->m_rtdev_targp); | |
198 | } | |
199 | ||
200 | return -error; | |
201 | } | |
202 | ||
00258e36 CH |
203 | STATIC ssize_t |
204 | xfs_file_aio_read( | |
dda35b8f CH |
205 | struct kiocb *iocb, |
206 | const struct iovec *iovp, | |
00258e36 CH |
207 | unsigned long nr_segs, |
208 | loff_t pos) | |
dda35b8f CH |
209 | { |
210 | struct file *file = iocb->ki_filp; | |
211 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
212 | struct xfs_inode *ip = XFS_I(inode); |
213 | struct xfs_mount *mp = ip->i_mount; | |
dda35b8f CH |
214 | size_t size = 0; |
215 | ssize_t ret = 0; | |
00258e36 | 216 | int ioflags = 0; |
dda35b8f CH |
217 | xfs_fsize_t n; |
218 | unsigned long seg; | |
219 | ||
dda35b8f CH |
220 | XFS_STATS_INC(xs_read_calls); |
221 | ||
00258e36 CH |
222 | BUG_ON(iocb->ki_pos != pos); |
223 | ||
224 | if (unlikely(file->f_flags & O_DIRECT)) | |
225 | ioflags |= IO_ISDIRECT; | |
226 | if (file->f_mode & FMODE_NOCMTIME) | |
227 | ioflags |= IO_INVIS; | |
228 | ||
dda35b8f | 229 | /* START copy & waste from filemap.c */ |
00258e36 | 230 | for (seg = 0; seg < nr_segs; seg++) { |
dda35b8f CH |
231 | const struct iovec *iv = &iovp[seg]; |
232 | ||
233 | /* | |
234 | * If any segment has a negative length, or the cumulative | |
235 | * length ever wraps negative then return -EINVAL. | |
236 | */ | |
237 | size += iv->iov_len; | |
238 | if (unlikely((ssize_t)(size|iv->iov_len) < 0)) | |
239 | return XFS_ERROR(-EINVAL); | |
240 | } | |
241 | /* END copy & waste from filemap.c */ | |
242 | ||
243 | if (unlikely(ioflags & IO_ISDIRECT)) { | |
244 | xfs_buftarg_t *target = | |
245 | XFS_IS_REALTIME_INODE(ip) ? | |
246 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
00258e36 | 247 | if ((iocb->ki_pos & target->bt_smask) || |
dda35b8f | 248 | (size & target->bt_smask)) { |
00258e36 CH |
249 | if (iocb->ki_pos == ip->i_size) |
250 | return 0; | |
dda35b8f CH |
251 | return -XFS_ERROR(EINVAL); |
252 | } | |
253 | } | |
254 | ||
00258e36 CH |
255 | n = XFS_MAXIOFFSET(mp) - iocb->ki_pos; |
256 | if (n <= 0 || size == 0) | |
dda35b8f CH |
257 | return 0; |
258 | ||
259 | if (n < size) | |
260 | size = n; | |
261 | ||
262 | if (XFS_FORCED_SHUTDOWN(mp)) | |
263 | return -EIO; | |
264 | ||
265 | if (unlikely(ioflags & IO_ISDIRECT)) | |
266 | mutex_lock(&inode->i_mutex); | |
267 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
268 | ||
dda35b8f | 269 | if (unlikely(ioflags & IO_ISDIRECT)) { |
00258e36 CH |
270 | if (inode->i_mapping->nrpages) { |
271 | ret = -xfs_flushinval_pages(ip, | |
272 | (iocb->ki_pos & PAGE_CACHE_MASK), | |
273 | -1, FI_REMAPF_LOCKED); | |
274 | } | |
dda35b8f CH |
275 | mutex_unlock(&inode->i_mutex); |
276 | if (ret) { | |
277 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
278 | return ret; | |
279 | } | |
280 | } | |
281 | ||
00258e36 | 282 | trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags); |
dda35b8f | 283 | |
00258e36 | 284 | ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos); |
dda35b8f CH |
285 | if (ret > 0) |
286 | XFS_STATS_ADD(xs_read_bytes, ret); | |
287 | ||
288 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
289 | return ret; | |
290 | } | |
291 | ||
00258e36 CH |
292 | STATIC ssize_t |
293 | xfs_file_splice_read( | |
dda35b8f CH |
294 | struct file *infilp, |
295 | loff_t *ppos, | |
296 | struct pipe_inode_info *pipe, | |
297 | size_t count, | |
00258e36 | 298 | unsigned int flags) |
dda35b8f | 299 | { |
00258e36 | 300 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 301 | int ioflags = 0; |
dda35b8f CH |
302 | ssize_t ret; |
303 | ||
304 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
305 | |
306 | if (infilp->f_mode & FMODE_NOCMTIME) | |
307 | ioflags |= IO_INVIS; | |
308 | ||
dda35b8f CH |
309 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
310 | return -EIO; | |
311 | ||
312 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
313 | ||
dda35b8f CH |
314 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
315 | ||
316 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
317 | if (ret > 0) | |
318 | XFS_STATS_ADD(xs_read_bytes, ret); | |
319 | ||
320 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
321 | return ret; | |
322 | } | |
323 | ||
edafb6da DC |
324 | STATIC void |
325 | xfs_aio_write_isize_update( | |
326 | struct inode *inode, | |
327 | loff_t *ppos, | |
328 | ssize_t bytes_written) | |
329 | { | |
330 | struct xfs_inode *ip = XFS_I(inode); | |
331 | xfs_fsize_t isize = i_size_read(inode); | |
332 | ||
333 | if (bytes_written > 0) | |
334 | XFS_STATS_ADD(xs_write_bytes, bytes_written); | |
335 | ||
336 | if (unlikely(bytes_written < 0 && bytes_written != -EFAULT && | |
337 | *ppos > isize)) | |
338 | *ppos = isize; | |
339 | ||
340 | if (*ppos > ip->i_size) { | |
341 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
342 | if (*ppos > ip->i_size) | |
343 | ip->i_size = *ppos; | |
344 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
345 | } | |
346 | } | |
347 | ||
4c5cfd1b DC |
348 | /* |
349 | * If this was a direct or synchronous I/O that failed (such as ENOSPC) then | |
350 | * part of the I/O may have been written to disk before the error occured. In | |
351 | * this case the on-disk file size may have been adjusted beyond the in-memory | |
352 | * file size and now needs to be truncated back. | |
353 | */ | |
354 | STATIC void | |
355 | xfs_aio_write_newsize_update( | |
356 | struct xfs_inode *ip) | |
357 | { | |
358 | if (ip->i_new_size) { | |
359 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
360 | ip->i_new_size = 0; | |
361 | if (ip->i_d.di_size > ip->i_size) | |
362 | ip->i_d.di_size = ip->i_size; | |
363 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
364 | } | |
365 | } | |
366 | ||
00258e36 CH |
367 | STATIC ssize_t |
368 | xfs_file_splice_write( | |
dda35b8f CH |
369 | struct pipe_inode_info *pipe, |
370 | struct file *outfilp, | |
371 | loff_t *ppos, | |
372 | size_t count, | |
00258e36 | 373 | unsigned int flags) |
dda35b8f | 374 | { |
dda35b8f | 375 | struct inode *inode = outfilp->f_mapping->host; |
00258e36 | 376 | struct xfs_inode *ip = XFS_I(inode); |
edafb6da | 377 | xfs_fsize_t new_size; |
00258e36 CH |
378 | int ioflags = 0; |
379 | ssize_t ret; | |
dda35b8f CH |
380 | |
381 | XFS_STATS_INC(xs_write_calls); | |
00258e36 CH |
382 | |
383 | if (outfilp->f_mode & FMODE_NOCMTIME) | |
384 | ioflags |= IO_INVIS; | |
385 | ||
dda35b8f CH |
386 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
387 | return -EIO; | |
388 | ||
389 | xfs_ilock(ip, XFS_IOLOCK_EXCL); | |
390 | ||
dda35b8f CH |
391 | new_size = *ppos + count; |
392 | ||
393 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
394 | if (new_size > ip->i_size) | |
395 | ip->i_new_size = new_size; | |
396 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
397 | ||
398 | trace_xfs_file_splice_write(ip, count, *ppos, ioflags); | |
399 | ||
400 | ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags); | |
dda35b8f | 401 | |
edafb6da | 402 | xfs_aio_write_isize_update(inode, ppos, ret); |
4c5cfd1b | 403 | xfs_aio_write_newsize_update(ip); |
dda35b8f CH |
404 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
405 | return ret; | |
406 | } | |
407 | ||
408 | /* | |
409 | * This routine is called to handle zeroing any space in the last | |
410 | * block of the file that is beyond the EOF. We do this since the | |
411 | * size is being increased without writing anything to that block | |
412 | * and we don't want anyone to read the garbage on the disk. | |
413 | */ | |
414 | STATIC int /* error (positive) */ | |
415 | xfs_zero_last_block( | |
416 | xfs_inode_t *ip, | |
417 | xfs_fsize_t offset, | |
418 | xfs_fsize_t isize) | |
419 | { | |
420 | xfs_fileoff_t last_fsb; | |
421 | xfs_mount_t *mp = ip->i_mount; | |
422 | int nimaps; | |
423 | int zero_offset; | |
424 | int zero_len; | |
425 | int error = 0; | |
426 | xfs_bmbt_irec_t imap; | |
427 | ||
428 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
429 | ||
430 | zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
431 | if (zero_offset == 0) { | |
432 | /* | |
433 | * There are no extra bytes in the last block on disk to | |
434 | * zero, so return. | |
435 | */ | |
436 | return 0; | |
437 | } | |
438 | ||
439 | last_fsb = XFS_B_TO_FSBT(mp, isize); | |
440 | nimaps = 1; | |
441 | error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap, | |
b4e9181e | 442 | &nimaps, NULL); |
dda35b8f CH |
443 | if (error) { |
444 | return error; | |
445 | } | |
446 | ASSERT(nimaps > 0); | |
447 | /* | |
448 | * If the block underlying isize is just a hole, then there | |
449 | * is nothing to zero. | |
450 | */ | |
451 | if (imap.br_startblock == HOLESTARTBLOCK) { | |
452 | return 0; | |
453 | } | |
454 | /* | |
455 | * Zero the part of the last block beyond the EOF, and write it | |
456 | * out sync. We need to drop the ilock while we do this so we | |
457 | * don't deadlock when the buffer cache calls back to us. | |
458 | */ | |
459 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
460 | ||
461 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
462 | if (isize + zero_len > offset) | |
463 | zero_len = offset - isize; | |
464 | error = xfs_iozero(ip, isize, zero_len); | |
465 | ||
466 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
467 | ASSERT(error >= 0); | |
468 | return error; | |
469 | } | |
470 | ||
471 | /* | |
472 | * Zero any on disk space between the current EOF and the new, | |
473 | * larger EOF. This handles the normal case of zeroing the remainder | |
474 | * of the last block in the file and the unusual case of zeroing blocks | |
475 | * out beyond the size of the file. This second case only happens | |
476 | * with fixed size extents and when the system crashes before the inode | |
477 | * size was updated but after blocks were allocated. If fill is set, | |
478 | * then any holes in the range are filled and zeroed. If not, the holes | |
479 | * are left alone as holes. | |
480 | */ | |
481 | ||
482 | int /* error (positive) */ | |
483 | xfs_zero_eof( | |
484 | xfs_inode_t *ip, | |
485 | xfs_off_t offset, /* starting I/O offset */ | |
486 | xfs_fsize_t isize) /* current inode size */ | |
487 | { | |
488 | xfs_mount_t *mp = ip->i_mount; | |
489 | xfs_fileoff_t start_zero_fsb; | |
490 | xfs_fileoff_t end_zero_fsb; | |
491 | xfs_fileoff_t zero_count_fsb; | |
492 | xfs_fileoff_t last_fsb; | |
493 | xfs_fileoff_t zero_off; | |
494 | xfs_fsize_t zero_len; | |
495 | int nimaps; | |
496 | int error = 0; | |
497 | xfs_bmbt_irec_t imap; | |
498 | ||
499 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | |
500 | ASSERT(offset > isize); | |
501 | ||
502 | /* | |
503 | * First handle zeroing the block on which isize resides. | |
504 | * We only zero a part of that block so it is handled specially. | |
505 | */ | |
506 | error = xfs_zero_last_block(ip, offset, isize); | |
507 | if (error) { | |
508 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | |
509 | return error; | |
510 | } | |
511 | ||
512 | /* | |
513 | * Calculate the range between the new size and the old | |
514 | * where blocks needing to be zeroed may exist. To get the | |
515 | * block where the last byte in the file currently resides, | |
516 | * we need to subtract one from the size and truncate back | |
517 | * to a block boundary. We subtract 1 in case the size is | |
518 | * exactly on a block boundary. | |
519 | */ | |
520 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
521 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
522 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
523 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
524 | if (last_fsb == end_zero_fsb) { | |
525 | /* | |
526 | * The size was only incremented on its last block. | |
527 | * We took care of that above, so just return. | |
528 | */ | |
529 | return 0; | |
530 | } | |
531 | ||
532 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
533 | while (start_zero_fsb <= end_zero_fsb) { | |
534 | nimaps = 1; | |
535 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
536 | error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb, | |
b4e9181e | 537 | 0, NULL, 0, &imap, &nimaps, NULL); |
dda35b8f CH |
538 | if (error) { |
539 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL)); | |
540 | return error; | |
541 | } | |
542 | ASSERT(nimaps > 0); | |
543 | ||
544 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
545 | imap.br_startblock == HOLESTARTBLOCK) { | |
546 | /* | |
547 | * This loop handles initializing pages that were | |
548 | * partially initialized by the code below this | |
549 | * loop. It basically zeroes the part of the page | |
550 | * that sits on a hole and sets the page as P_HOLE | |
551 | * and calls remapf if it is a mapped file. | |
552 | */ | |
553 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
554 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
555 | continue; | |
556 | } | |
557 | ||
558 | /* | |
559 | * There are blocks we need to zero. | |
560 | * Drop the inode lock while we're doing the I/O. | |
561 | * We'll still have the iolock to protect us. | |
562 | */ | |
563 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
564 | ||
565 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); | |
566 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
567 | ||
568 | if ((zero_off + zero_len) > offset) | |
569 | zero_len = offset - zero_off; | |
570 | ||
571 | error = xfs_iozero(ip, zero_off, zero_len); | |
572 | if (error) { | |
573 | goto out_lock; | |
574 | } | |
575 | ||
576 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
577 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
578 | ||
579 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
580 | } | |
581 | ||
582 | return 0; | |
583 | ||
584 | out_lock: | |
585 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
586 | ASSERT(error >= 0); | |
587 | return error; | |
588 | } | |
589 | ||
00258e36 CH |
590 | STATIC ssize_t |
591 | xfs_file_aio_write( | |
dda35b8f CH |
592 | struct kiocb *iocb, |
593 | const struct iovec *iovp, | |
00258e36 CH |
594 | unsigned long nr_segs, |
595 | loff_t pos) | |
dda35b8f CH |
596 | { |
597 | struct file *file = iocb->ki_filp; | |
598 | struct address_space *mapping = file->f_mapping; | |
599 | struct inode *inode = mapping->host; | |
00258e36 CH |
600 | struct xfs_inode *ip = XFS_I(inode); |
601 | struct xfs_mount *mp = ip->i_mount; | |
a363f0c2 | 602 | ssize_t ret = 0; |
00258e36 | 603 | int ioflags = 0; |
edafb6da | 604 | xfs_fsize_t new_size; |
dda35b8f | 605 | int iolock; |
dda35b8f | 606 | size_t ocount = 0, count; |
dda35b8f CH |
607 | int need_i_mutex; |
608 | ||
609 | XFS_STATS_INC(xs_write_calls); | |
610 | ||
00258e36 CH |
611 | BUG_ON(iocb->ki_pos != pos); |
612 | ||
613 | if (unlikely(file->f_flags & O_DIRECT)) | |
614 | ioflags |= IO_ISDIRECT; | |
615 | if (file->f_mode & FMODE_NOCMTIME) | |
616 | ioflags |= IO_INVIS; | |
617 | ||
a363f0c2 DC |
618 | ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ); |
619 | if (ret) | |
620 | return ret; | |
dda35b8f CH |
621 | |
622 | count = ocount; | |
dda35b8f CH |
623 | if (count == 0) |
624 | return 0; | |
625 | ||
dda35b8f CH |
626 | xfs_wait_for_freeze(mp, SB_FREEZE_WRITE); |
627 | ||
628 | if (XFS_FORCED_SHUTDOWN(mp)) | |
629 | return -EIO; | |
630 | ||
631 | relock: | |
632 | if (ioflags & IO_ISDIRECT) { | |
633 | iolock = XFS_IOLOCK_SHARED; | |
634 | need_i_mutex = 0; | |
635 | } else { | |
636 | iolock = XFS_IOLOCK_EXCL; | |
637 | need_i_mutex = 1; | |
638 | mutex_lock(&inode->i_mutex); | |
639 | } | |
640 | ||
00258e36 | 641 | xfs_ilock(ip, XFS_ILOCK_EXCL|iolock); |
dda35b8f CH |
642 | |
643 | start: | |
a363f0c2 | 644 | ret = generic_write_checks(file, &pos, &count, |
dda35b8f | 645 | S_ISBLK(inode->i_mode)); |
a363f0c2 | 646 | if (ret) { |
00258e36 | 647 | xfs_iunlock(ip, XFS_ILOCK_EXCL|iolock); |
dda35b8f CH |
648 | goto out_unlock_mutex; |
649 | } | |
650 | ||
dda35b8f CH |
651 | if (ioflags & IO_ISDIRECT) { |
652 | xfs_buftarg_t *target = | |
00258e36 | 653 | XFS_IS_REALTIME_INODE(ip) ? |
dda35b8f CH |
654 | mp->m_rtdev_targp : mp->m_ddev_targp; |
655 | ||
656 | if ((pos & target->bt_smask) || (count & target->bt_smask)) { | |
00258e36 | 657 | xfs_iunlock(ip, XFS_ILOCK_EXCL|iolock); |
dda35b8f CH |
658 | return XFS_ERROR(-EINVAL); |
659 | } | |
660 | ||
00258e36 CH |
661 | if (!need_i_mutex && (mapping->nrpages || pos > ip->i_size)) { |
662 | xfs_iunlock(ip, XFS_ILOCK_EXCL|iolock); | |
dda35b8f CH |
663 | iolock = XFS_IOLOCK_EXCL; |
664 | need_i_mutex = 1; | |
665 | mutex_lock(&inode->i_mutex); | |
00258e36 | 666 | xfs_ilock(ip, XFS_ILOCK_EXCL|iolock); |
dda35b8f CH |
667 | goto start; |
668 | } | |
669 | } | |
670 | ||
671 | new_size = pos + count; | |
00258e36 CH |
672 | if (new_size > ip->i_size) |
673 | ip->i_new_size = new_size; | |
dda35b8f CH |
674 | |
675 | if (likely(!(ioflags & IO_INVIS))) | |
676 | file_update_time(file); | |
677 | ||
678 | /* | |
679 | * If the offset is beyond the size of the file, we have a couple | |
680 | * of things to do. First, if there is already space allocated | |
681 | * we need to either create holes or zero the disk or ... | |
682 | * | |
683 | * If there is a page where the previous size lands, we need | |
684 | * to zero it out up to the new size. | |
685 | */ | |
686 | ||
00258e36 | 687 | if (pos > ip->i_size) { |
a363f0c2 DC |
688 | ret = -xfs_zero_eof(ip, pos, ip->i_size); |
689 | if (ret) { | |
00258e36 | 690 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
dda35b8f CH |
691 | goto out_unlock_internal; |
692 | } | |
693 | } | |
00258e36 | 694 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
dda35b8f CH |
695 | |
696 | /* | |
697 | * If we're writing the file then make sure to clear the | |
698 | * setuid and setgid bits if the process is not being run | |
699 | * by root. This keeps people from modifying setuid and | |
700 | * setgid binaries. | |
701 | */ | |
a363f0c2 DC |
702 | ret = file_remove_suid(file); |
703 | if (unlikely(ret)) | |
dda35b8f CH |
704 | goto out_unlock_internal; |
705 | ||
706 | /* We can write back this queue in page reclaim */ | |
707 | current->backing_dev_info = mapping->backing_dev_info; | |
708 | ||
709 | if ((ioflags & IO_ISDIRECT)) { | |
710 | if (mapping->nrpages) { | |
711 | WARN_ON(need_i_mutex == 0); | |
a363f0c2 | 712 | ret = -xfs_flushinval_pages(ip, |
dda35b8f CH |
713 | (pos & PAGE_CACHE_MASK), |
714 | -1, FI_REMAPF_LOCKED); | |
a363f0c2 | 715 | if (ret) |
dda35b8f CH |
716 | goto out_unlock_internal; |
717 | } | |
718 | ||
719 | if (need_i_mutex) { | |
720 | /* demote the lock now the cached pages are gone */ | |
00258e36 | 721 | xfs_ilock_demote(ip, XFS_IOLOCK_EXCL); |
dda35b8f CH |
722 | mutex_unlock(&inode->i_mutex); |
723 | ||
724 | iolock = XFS_IOLOCK_SHARED; | |
725 | need_i_mutex = 0; | |
726 | } | |
727 | ||
00258e36 | 728 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, ioflags); |
dda35b8f | 729 | ret = generic_file_direct_write(iocb, iovp, |
00258e36 | 730 | &nr_segs, pos, &iocb->ki_pos, count, ocount); |
dda35b8f CH |
731 | |
732 | /* | |
733 | * direct-io write to a hole: fall through to buffered I/O | |
734 | * for completing the rest of the request. | |
735 | */ | |
736 | if (ret >= 0 && ret != count) { | |
737 | XFS_STATS_ADD(xs_write_bytes, ret); | |
738 | ||
739 | pos += ret; | |
740 | count -= ret; | |
741 | ||
742 | ioflags &= ~IO_ISDIRECT; | |
00258e36 | 743 | xfs_iunlock(ip, iolock); |
dda35b8f CH |
744 | goto relock; |
745 | } | |
746 | } else { | |
747 | int enospc = 0; | |
dda35b8f CH |
748 | |
749 | write_retry: | |
00258e36 | 750 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, ioflags); |
a363f0c2 | 751 | ret = generic_file_buffered_write(iocb, iovp, nr_segs, |
00258e36 | 752 | pos, &iocb->ki_pos, count, ret); |
dda35b8f CH |
753 | /* |
754 | * if we just got an ENOSPC, flush the inode now we | |
755 | * aren't holding any page locks and retry *once* | |
756 | */ | |
a363f0c2 DC |
757 | if (ret == -ENOSPC && !enospc) { |
758 | ret = xfs_flush_pages(ip, 0, -1, 0, FI_NONE); | |
759 | if (ret) | |
dda35b8f CH |
760 | goto out_unlock_internal; |
761 | enospc = 1; | |
762 | goto write_retry; | |
763 | } | |
dda35b8f CH |
764 | } |
765 | ||
766 | current->backing_dev_info = NULL; | |
767 | ||
edafb6da | 768 | xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret); |
dda35b8f | 769 | |
dda35b8f CH |
770 | if (ret <= 0) |
771 | goto out_unlock_internal; | |
772 | ||
dda35b8f CH |
773 | /* Handle various SYNC-type writes */ |
774 | if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) { | |
775 | loff_t end = pos + ret - 1; | |
a363f0c2 | 776 | int error, error2; |
dda35b8f | 777 | |
00258e36 | 778 | xfs_iunlock(ip, iolock); |
dda35b8f CH |
779 | if (need_i_mutex) |
780 | mutex_unlock(&inode->i_mutex); | |
781 | ||
a363f0c2 | 782 | error = filemap_write_and_wait_range(mapping, pos, end); |
dda35b8f CH |
783 | if (need_i_mutex) |
784 | mutex_lock(&inode->i_mutex); | |
00258e36 | 785 | xfs_ilock(ip, iolock); |
dda35b8f | 786 | |
7ea80859 | 787 | error2 = -xfs_file_fsync(file, |
fd3200be | 788 | (file->f_flags & __O_SYNC) ? 0 : 1); |
a363f0c2 DC |
789 | if (error) |
790 | ret = error; | |
791 | else if (error2) | |
792 | ret = error2; | |
dda35b8f CH |
793 | } |
794 | ||
795 | out_unlock_internal: | |
4c5cfd1b | 796 | xfs_aio_write_newsize_update(ip); |
00258e36 | 797 | xfs_iunlock(ip, iolock); |
dda35b8f CH |
798 | out_unlock_mutex: |
799 | if (need_i_mutex) | |
800 | mutex_unlock(&inode->i_mutex); | |
a363f0c2 | 801 | return ret; |
dda35b8f CH |
802 | } |
803 | ||
1da177e4 | 804 | STATIC int |
3562fd45 | 805 | xfs_file_open( |
1da177e4 | 806 | struct inode *inode, |
f999a5bf | 807 | struct file *file) |
1da177e4 | 808 | { |
f999a5bf | 809 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 810 | return -EFBIG; |
f999a5bf CH |
811 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
812 | return -EIO; | |
813 | return 0; | |
814 | } | |
815 | ||
816 | STATIC int | |
817 | xfs_dir_open( | |
818 | struct inode *inode, | |
819 | struct file *file) | |
820 | { | |
821 | struct xfs_inode *ip = XFS_I(inode); | |
822 | int mode; | |
823 | int error; | |
824 | ||
825 | error = xfs_file_open(inode, file); | |
826 | if (error) | |
827 | return error; | |
828 | ||
829 | /* | |
830 | * If there are any blocks, read-ahead block 0 as we're almost | |
831 | * certain to have the next operation be a read there. | |
832 | */ | |
833 | mode = xfs_ilock_map_shared(ip); | |
834 | if (ip->i_d.di_nextents > 0) | |
835 | xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK); | |
836 | xfs_iunlock(ip, mode); | |
837 | return 0; | |
1da177e4 LT |
838 | } |
839 | ||
1da177e4 | 840 | STATIC int |
3562fd45 | 841 | xfs_file_release( |
1da177e4 LT |
842 | struct inode *inode, |
843 | struct file *filp) | |
844 | { | |
739bfb2a | 845 | return -xfs_release(XFS_I(inode)); |
1da177e4 LT |
846 | } |
847 | ||
1da177e4 | 848 | STATIC int |
3562fd45 | 849 | xfs_file_readdir( |
1da177e4 LT |
850 | struct file *filp, |
851 | void *dirent, | |
852 | filldir_t filldir) | |
853 | { | |
051e7cd4 | 854 | struct inode *inode = filp->f_path.dentry->d_inode; |
739bfb2a | 855 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
856 | int error; |
857 | size_t bufsize; | |
858 | ||
859 | /* | |
860 | * The Linux API doesn't pass down the total size of the buffer | |
861 | * we read into down to the filesystem. With the filldir concept | |
862 | * it's not needed for correct information, but the XFS dir2 leaf | |
863 | * code wants an estimate of the buffer size to calculate it's | |
864 | * readahead window and size the buffers used for mapping to | |
865 | * physical blocks. | |
866 | * | |
867 | * Try to give it an estimate that's good enough, maybe at some | |
868 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 869 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 870 | */ |
a9cc799e | 871 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 872 | |
739bfb2a | 873 | error = xfs_readdir(ip, dirent, bufsize, |
051e7cd4 CH |
874 | (xfs_off_t *)&filp->f_pos, filldir); |
875 | if (error) | |
876 | return -error; | |
877 | return 0; | |
1da177e4 LT |
878 | } |
879 | ||
1da177e4 | 880 | STATIC int |
3562fd45 | 881 | xfs_file_mmap( |
1da177e4 LT |
882 | struct file *filp, |
883 | struct vm_area_struct *vma) | |
884 | { | |
3562fd45 | 885 | vma->vm_ops = &xfs_file_vm_ops; |
d0217ac0 | 886 | vma->vm_flags |= VM_CAN_NONLINEAR; |
6fac0cb4 | 887 | |
fbc1462b | 888 | file_accessed(filp); |
1da177e4 LT |
889 | return 0; |
890 | } | |
891 | ||
4f57dbc6 DC |
892 | /* |
893 | * mmap()d file has taken write protection fault and is being made | |
894 | * writable. We can set the page state up correctly for a writable | |
895 | * page, which means we can do correct delalloc accounting (ENOSPC | |
896 | * checking!) and unwritten extent mapping. | |
897 | */ | |
898 | STATIC int | |
899 | xfs_vm_page_mkwrite( | |
900 | struct vm_area_struct *vma, | |
c2ec175c | 901 | struct vm_fault *vmf) |
4f57dbc6 | 902 | { |
c2ec175c | 903 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
904 | } |
905 | ||
4b6f5d20 | 906 | const struct file_operations xfs_file_operations = { |
1da177e4 LT |
907 | .llseek = generic_file_llseek, |
908 | .read = do_sync_read, | |
bb3f724e | 909 | .write = do_sync_write, |
3562fd45 NS |
910 | .aio_read = xfs_file_aio_read, |
911 | .aio_write = xfs_file_aio_write, | |
1b895840 NS |
912 | .splice_read = xfs_file_splice_read, |
913 | .splice_write = xfs_file_splice_write, | |
3562fd45 | 914 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 915 | #ifdef CONFIG_COMPAT |
3562fd45 | 916 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 917 | #endif |
3562fd45 NS |
918 | .mmap = xfs_file_mmap, |
919 | .open = xfs_file_open, | |
920 | .release = xfs_file_release, | |
921 | .fsync = xfs_file_fsync, | |
1da177e4 LT |
922 | }; |
923 | ||
4b6f5d20 | 924 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 925 | .open = xfs_dir_open, |
1da177e4 | 926 | .read = generic_read_dir, |
3562fd45 | 927 | .readdir = xfs_file_readdir, |
59af1584 | 928 | .llseek = generic_file_llseek, |
3562fd45 | 929 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 930 | #ifdef CONFIG_COMPAT |
3562fd45 | 931 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 932 | #endif |
3562fd45 | 933 | .fsync = xfs_file_fsync, |
1da177e4 LT |
934 | }; |
935 | ||
f0f37e2f | 936 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 937 | .fault = filemap_fault, |
4f57dbc6 | 938 | .page_mkwrite = xfs_vm_page_mkwrite, |
6fac0cb4 | 939 | }; |