Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
dda35b8f | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
a4fbe6ab | 21 | #include "xfs_format.h" |
239880ef DC |
22 | #include "xfs_log_format.h" |
23 | #include "xfs_trans_resv.h" | |
1da177e4 | 24 | #include "xfs_sb.h" |
a844f451 | 25 | #include "xfs_ag.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
57062787 DC |
27 | #include "xfs_da_format.h" |
28 | #include "xfs_da_btree.h" | |
1da177e4 | 29 | #include "xfs_inode.h" |
239880ef | 30 | #include "xfs_trans.h" |
fd3200be | 31 | #include "xfs_inode_item.h" |
dda35b8f | 32 | #include "xfs_bmap.h" |
c24b5dfa | 33 | #include "xfs_bmap_util.h" |
1da177e4 | 34 | #include "xfs_error.h" |
2b9ab5ab | 35 | #include "xfs_dir2.h" |
c24b5dfa | 36 | #include "xfs_dir2_priv.h" |
ddcd856d | 37 | #include "xfs_ioctl.h" |
dda35b8f | 38 | #include "xfs_trace.h" |
239880ef | 39 | #include "xfs_log.h" |
a4fbe6ab | 40 | #include "xfs_dinode.h" |
dc06f398 | 41 | #include "xfs_icache.h" |
1da177e4 | 42 | |
a27bb332 | 43 | #include <linux/aio.h> |
1da177e4 | 44 | #include <linux/dcache.h> |
2fe17c10 | 45 | #include <linux/falloc.h> |
d126d43f | 46 | #include <linux/pagevec.h> |
1da177e4 | 47 | |
f0f37e2f | 48 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 49 | |
487f84f3 DC |
50 | /* |
51 | * Locking primitives for read and write IO paths to ensure we consistently use | |
52 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
53 | */ | |
54 | static inline void | |
55 | xfs_rw_ilock( | |
56 | struct xfs_inode *ip, | |
57 | int type) | |
58 | { | |
59 | if (type & XFS_IOLOCK_EXCL) | |
60 | mutex_lock(&VFS_I(ip)->i_mutex); | |
61 | xfs_ilock(ip, type); | |
62 | } | |
63 | ||
64 | static inline void | |
65 | xfs_rw_iunlock( | |
66 | struct xfs_inode *ip, | |
67 | int type) | |
68 | { | |
69 | xfs_iunlock(ip, type); | |
70 | if (type & XFS_IOLOCK_EXCL) | |
71 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
72 | } | |
73 | ||
74 | static inline void | |
75 | xfs_rw_ilock_demote( | |
76 | struct xfs_inode *ip, | |
77 | int type) | |
78 | { | |
79 | xfs_ilock_demote(ip, type); | |
80 | if (type & XFS_IOLOCK_EXCL) | |
81 | mutex_unlock(&VFS_I(ip)->i_mutex); | |
82 | } | |
83 | ||
dda35b8f CH |
84 | /* |
85 | * xfs_iozero | |
86 | * | |
87 | * xfs_iozero clears the specified range of buffer supplied, | |
88 | * and marks all the affected blocks as valid and modified. If | |
89 | * an affected block is not allocated, it will be allocated. If | |
90 | * an affected block is not completely overwritten, and is not | |
91 | * valid before the operation, it will be read from disk before | |
92 | * being partially zeroed. | |
93 | */ | |
ef9d8733 | 94 | int |
dda35b8f CH |
95 | xfs_iozero( |
96 | struct xfs_inode *ip, /* inode */ | |
97 | loff_t pos, /* offset in file */ | |
98 | size_t count) /* size of data to zero */ | |
99 | { | |
100 | struct page *page; | |
101 | struct address_space *mapping; | |
102 | int status; | |
103 | ||
104 | mapping = VFS_I(ip)->i_mapping; | |
105 | do { | |
106 | unsigned offset, bytes; | |
107 | void *fsdata; | |
108 | ||
109 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ | |
110 | bytes = PAGE_CACHE_SIZE - offset; | |
111 | if (bytes > count) | |
112 | bytes = count; | |
113 | ||
114 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
115 | AOP_FLAG_UNINTERRUPTIBLE, | |
116 | &page, &fsdata); | |
117 | if (status) | |
118 | break; | |
119 | ||
120 | zero_user(page, offset, bytes); | |
121 | ||
122 | status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, | |
123 | page, fsdata); | |
124 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
125 | pos += bytes; | |
126 | count -= bytes; | |
127 | status = 0; | |
128 | } while (count); | |
129 | ||
130 | return (-status); | |
131 | } | |
132 | ||
1da2f2db CH |
133 | /* |
134 | * Fsync operations on directories are much simpler than on regular files, | |
135 | * as there is no file data to flush, and thus also no need for explicit | |
136 | * cache flush operations, and there are no non-transaction metadata updates | |
137 | * on directories either. | |
138 | */ | |
139 | STATIC int | |
140 | xfs_dir_fsync( | |
141 | struct file *file, | |
142 | loff_t start, | |
143 | loff_t end, | |
144 | int datasync) | |
145 | { | |
146 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
147 | struct xfs_mount *mp = ip->i_mount; | |
148 | xfs_lsn_t lsn = 0; | |
149 | ||
150 | trace_xfs_dir_fsync(ip); | |
151 | ||
152 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
153 | if (xfs_ipincount(ip)) | |
154 | lsn = ip->i_itemp->ili_last_lsn; | |
155 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
156 | ||
157 | if (!lsn) | |
158 | return 0; | |
2451337d | 159 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
160 | } |
161 | ||
fd3200be CH |
162 | STATIC int |
163 | xfs_file_fsync( | |
164 | struct file *file, | |
02c24a82 JB |
165 | loff_t start, |
166 | loff_t end, | |
fd3200be CH |
167 | int datasync) |
168 | { | |
7ea80859 CH |
169 | struct inode *inode = file->f_mapping->host; |
170 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 171 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
172 | int error = 0; |
173 | int log_flushed = 0; | |
b1037058 | 174 | xfs_lsn_t lsn = 0; |
fd3200be | 175 | |
cca28fb8 | 176 | trace_xfs_file_fsync(ip); |
fd3200be | 177 | |
02c24a82 JB |
178 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
179 | if (error) | |
180 | return error; | |
181 | ||
a27a263b | 182 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 183 | return -EIO; |
fd3200be CH |
184 | |
185 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
186 | ||
a27a263b CH |
187 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
188 | /* | |
189 | * If we have an RT and/or log subvolume we need to make sure | |
190 | * to flush the write cache the device used for file data | |
191 | * first. This is to ensure newly written file data make | |
192 | * it to disk before logging the new inode size in case of | |
193 | * an extending write. | |
194 | */ | |
195 | if (XFS_IS_REALTIME_INODE(ip)) | |
196 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
197 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
198 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
199 | } | |
200 | ||
fd3200be | 201 | /* |
8a9c9980 CH |
202 | * All metadata updates are logged, which means that we just have |
203 | * to flush the log up to the latest LSN that touched the inode. | |
fd3200be CH |
204 | */ |
205 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
206 | if (xfs_ipincount(ip)) { |
207 | if (!datasync || | |
208 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) | |
209 | lsn = ip->i_itemp->ili_last_lsn; | |
210 | } | |
8a9c9980 | 211 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
fd3200be | 212 | |
8a9c9980 | 213 | if (lsn) |
b1037058 CH |
214 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
215 | ||
a27a263b CH |
216 | /* |
217 | * If we only have a single device, and the log force about was | |
218 | * a no-op we might have to flush the data device cache here. | |
219 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
220 | * an already allocated file and thus do not have any metadata to | |
221 | * commit. | |
222 | */ | |
223 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
224 | mp->m_logdev_targp == mp->m_ddev_targp && | |
225 | !XFS_IS_REALTIME_INODE(ip) && | |
226 | !log_flushed) | |
227 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 228 | |
2451337d | 229 | return error; |
fd3200be CH |
230 | } |
231 | ||
00258e36 | 232 | STATIC ssize_t |
b4f5d2c6 | 233 | xfs_file_read_iter( |
dda35b8f | 234 | struct kiocb *iocb, |
b4f5d2c6 | 235 | struct iov_iter *to) |
dda35b8f CH |
236 | { |
237 | struct file *file = iocb->ki_filp; | |
238 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
239 | struct xfs_inode *ip = XFS_I(inode); |
240 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 241 | size_t size = iov_iter_count(to); |
dda35b8f | 242 | ssize_t ret = 0; |
00258e36 | 243 | int ioflags = 0; |
dda35b8f | 244 | xfs_fsize_t n; |
b4f5d2c6 | 245 | loff_t pos = iocb->ki_pos; |
dda35b8f | 246 | |
dda35b8f CH |
247 | XFS_STATS_INC(xs_read_calls); |
248 | ||
00258e36 | 249 | if (unlikely(file->f_flags & O_DIRECT)) |
b92cc59f | 250 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 251 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 252 | ioflags |= XFS_IO_INVIS; |
00258e36 | 253 | |
b92cc59f | 254 | if (unlikely(ioflags & XFS_IO_ISDIRECT)) { |
dda35b8f CH |
255 | xfs_buftarg_t *target = |
256 | XFS_IS_REALTIME_INODE(ip) ? | |
257 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
258 | /* DIO must be aligned to device logical sector size */ |
259 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 260 | if (pos == i_size_read(inode)) |
00258e36 | 261 | return 0; |
b474c7ae | 262 | return -EINVAL; |
dda35b8f CH |
263 | } |
264 | } | |
265 | ||
fb595814 | 266 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 267 | if (n <= 0 || size == 0) |
dda35b8f CH |
268 | return 0; |
269 | ||
270 | if (n < size) | |
271 | size = n; | |
272 | ||
273 | if (XFS_FORCED_SHUTDOWN(mp)) | |
274 | return -EIO; | |
275 | ||
0c38a251 DC |
276 | /* |
277 | * Locking is a bit tricky here. If we take an exclusive lock | |
278 | * for direct IO, we effectively serialise all new concurrent | |
279 | * read IO to this file and block it behind IO that is currently in | |
280 | * progress because IO in progress holds the IO lock shared. We only | |
281 | * need to hold the lock exclusive to blow away the page cache, so | |
282 | * only take lock exclusively if the page cache needs invalidation. | |
283 | * This allows the normal direct IO case of no page cache pages to | |
284 | * proceeed concurrently without serialisation. | |
285 | */ | |
286 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 287 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 288 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
289 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
290 | ||
00258e36 | 291 | if (inode->i_mapping->nrpages) { |
8ff1e670 | 292 | ret = filemap_write_and_wait_range( |
fb595814 DC |
293 | VFS_I(ip)->i_mapping, |
294 | pos, -1); | |
487f84f3 DC |
295 | if (ret) { |
296 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
297 | return ret; | |
298 | } | |
fb595814 | 299 | truncate_pagecache_range(VFS_I(ip), pos, -1); |
00258e36 | 300 | } |
487f84f3 | 301 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 302 | } |
dda35b8f | 303 | |
fb595814 | 304 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 305 | |
b4f5d2c6 | 306 | ret = generic_file_read_iter(iocb, to); |
dda35b8f CH |
307 | if (ret > 0) |
308 | XFS_STATS_ADD(xs_read_bytes, ret); | |
309 | ||
487f84f3 | 310 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
311 | return ret; |
312 | } | |
313 | ||
00258e36 CH |
314 | STATIC ssize_t |
315 | xfs_file_splice_read( | |
dda35b8f CH |
316 | struct file *infilp, |
317 | loff_t *ppos, | |
318 | struct pipe_inode_info *pipe, | |
319 | size_t count, | |
00258e36 | 320 | unsigned int flags) |
dda35b8f | 321 | { |
00258e36 | 322 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 323 | int ioflags = 0; |
dda35b8f CH |
324 | ssize_t ret; |
325 | ||
326 | XFS_STATS_INC(xs_read_calls); | |
00258e36 CH |
327 | |
328 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 329 | ioflags |= XFS_IO_INVIS; |
00258e36 | 330 | |
dda35b8f CH |
331 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
332 | return -EIO; | |
333 | ||
487f84f3 | 334 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
dda35b8f | 335 | |
dda35b8f CH |
336 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
337 | ||
338 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
339 | if (ret > 0) | |
340 | XFS_STATS_ADD(xs_read_bytes, ret); | |
341 | ||
487f84f3 | 342 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
343 | return ret; |
344 | } | |
345 | ||
dda35b8f | 346 | /* |
193aec10 CH |
347 | * This routine is called to handle zeroing any space in the last block of the |
348 | * file that is beyond the EOF. We do this since the size is being increased | |
349 | * without writing anything to that block and we don't want to read the | |
350 | * garbage on the disk. | |
dda35b8f CH |
351 | */ |
352 | STATIC int /* error (positive) */ | |
353 | xfs_zero_last_block( | |
193aec10 CH |
354 | struct xfs_inode *ip, |
355 | xfs_fsize_t offset, | |
356 | xfs_fsize_t isize) | |
dda35b8f | 357 | { |
193aec10 CH |
358 | struct xfs_mount *mp = ip->i_mount; |
359 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
360 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
361 | int zero_len; | |
362 | int nimaps = 1; | |
363 | int error = 0; | |
364 | struct xfs_bmbt_irec imap; | |
dda35b8f | 365 | |
193aec10 | 366 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 367 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 368 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 369 | if (error) |
dda35b8f | 370 | return error; |
193aec10 | 371 | |
dda35b8f | 372 | ASSERT(nimaps > 0); |
193aec10 | 373 | |
dda35b8f CH |
374 | /* |
375 | * If the block underlying isize is just a hole, then there | |
376 | * is nothing to zero. | |
377 | */ | |
193aec10 | 378 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 379 | return 0; |
dda35b8f CH |
380 | |
381 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
382 | if (isize + zero_len > offset) | |
383 | zero_len = offset - isize; | |
193aec10 | 384 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
385 | } |
386 | ||
387 | /* | |
193aec10 CH |
388 | * Zero any on disk space between the current EOF and the new, larger EOF. |
389 | * | |
390 | * This handles the normal case of zeroing the remainder of the last block in | |
391 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
392 | * file. This second case only happens with fixed size extents and when the | |
393 | * system crashes before the inode size was updated but after blocks were | |
394 | * allocated. | |
395 | * | |
396 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 397 | */ |
dda35b8f CH |
398 | int /* error (positive) */ |
399 | xfs_zero_eof( | |
193aec10 CH |
400 | struct xfs_inode *ip, |
401 | xfs_off_t offset, /* starting I/O offset */ | |
402 | xfs_fsize_t isize) /* current inode size */ | |
dda35b8f | 403 | { |
193aec10 CH |
404 | struct xfs_mount *mp = ip->i_mount; |
405 | xfs_fileoff_t start_zero_fsb; | |
406 | xfs_fileoff_t end_zero_fsb; | |
407 | xfs_fileoff_t zero_count_fsb; | |
408 | xfs_fileoff_t last_fsb; | |
409 | xfs_fileoff_t zero_off; | |
410 | xfs_fsize_t zero_len; | |
411 | int nimaps; | |
412 | int error = 0; | |
413 | struct xfs_bmbt_irec imap; | |
414 | ||
415 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
416 | ASSERT(offset > isize); |
417 | ||
418 | /* | |
419 | * First handle zeroing the block on which isize resides. | |
193aec10 | 420 | * |
dda35b8f CH |
421 | * We only zero a part of that block so it is handled specially. |
422 | */ | |
193aec10 CH |
423 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
424 | error = xfs_zero_last_block(ip, offset, isize); | |
425 | if (error) | |
426 | return error; | |
dda35b8f CH |
427 | } |
428 | ||
429 | /* | |
193aec10 CH |
430 | * Calculate the range between the new size and the old where blocks |
431 | * needing to be zeroed may exist. | |
432 | * | |
433 | * To get the block where the last byte in the file currently resides, | |
434 | * we need to subtract one from the size and truncate back to a block | |
435 | * boundary. We subtract 1 in case the size is exactly on a block | |
436 | * boundary. | |
dda35b8f CH |
437 | */ |
438 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
439 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
440 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
441 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
442 | if (last_fsb == end_zero_fsb) { | |
443 | /* | |
444 | * The size was only incremented on its last block. | |
445 | * We took care of that above, so just return. | |
446 | */ | |
447 | return 0; | |
448 | } | |
449 | ||
450 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
451 | while (start_zero_fsb <= end_zero_fsb) { | |
452 | nimaps = 1; | |
453 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
454 | |
455 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
456 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
457 | &imap, &nimaps, 0); | |
193aec10 CH |
458 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
459 | if (error) | |
dda35b8f | 460 | return error; |
193aec10 | 461 | |
dda35b8f CH |
462 | ASSERT(nimaps > 0); |
463 | ||
464 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
465 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
466 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
467 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
468 | continue; | |
469 | } | |
470 | ||
471 | /* | |
472 | * There are blocks we need to zero. | |
dda35b8f | 473 | */ |
dda35b8f CH |
474 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
475 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
476 | ||
477 | if ((zero_off + zero_len) > offset) | |
478 | zero_len = offset - zero_off; | |
479 | ||
480 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
481 | if (error) |
482 | return error; | |
dda35b8f CH |
483 | |
484 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; | |
485 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
486 | } |
487 | ||
488 | return 0; | |
dda35b8f CH |
489 | } |
490 | ||
4d8d1581 DC |
491 | /* |
492 | * Common pre-write limit and setup checks. | |
493 | * | |
5bf1f262 CH |
494 | * Called with the iolocked held either shared and exclusive according to |
495 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
496 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
497 | */ |
498 | STATIC ssize_t | |
499 | xfs_file_aio_write_checks( | |
500 | struct file *file, | |
501 | loff_t *pos, | |
502 | size_t *count, | |
503 | int *iolock) | |
504 | { | |
505 | struct inode *inode = file->f_mapping->host; | |
506 | struct xfs_inode *ip = XFS_I(inode); | |
4d8d1581 DC |
507 | int error = 0; |
508 | ||
7271d243 | 509 | restart: |
4d8d1581 | 510 | error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode)); |
467f7899 | 511 | if (error) |
4d8d1581 | 512 | return error; |
4d8d1581 | 513 | |
4d8d1581 DC |
514 | /* |
515 | * If the offset is beyond the size of the file, we need to zero any | |
516 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 517 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
518 | * iolock shared, we need to update it to exclusive which implies |
519 | * having to redo all checks before. | |
4d8d1581 | 520 | */ |
2813d682 | 521 | if (*pos > i_size_read(inode)) { |
7271d243 | 522 | if (*iolock == XFS_IOLOCK_SHARED) { |
467f7899 | 523 | xfs_rw_iunlock(ip, *iolock); |
7271d243 | 524 | *iolock = XFS_IOLOCK_EXCL; |
467f7899 | 525 | xfs_rw_ilock(ip, *iolock); |
7271d243 DC |
526 | goto restart; |
527 | } | |
2451337d | 528 | error = xfs_zero_eof(ip, *pos, i_size_read(inode)); |
467f7899 CH |
529 | if (error) |
530 | return error; | |
7271d243 | 531 | } |
4d8d1581 | 532 | |
8a9c9980 CH |
533 | /* |
534 | * Updating the timestamps will grab the ilock again from | |
535 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
536 | * lock above. Eventually we should look into a way to avoid | |
537 | * the pointless lock roundtrip. | |
538 | */ | |
c3b2da31 JB |
539 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
540 | error = file_update_time(file); | |
541 | if (error) | |
542 | return error; | |
543 | } | |
8a9c9980 | 544 | |
4d8d1581 DC |
545 | /* |
546 | * If we're writing the file then make sure to clear the setuid and | |
547 | * setgid bits if the process is not being run by root. This keeps | |
548 | * people from modifying setuid and setgid binaries. | |
549 | */ | |
550 | return file_remove_suid(file); | |
4d8d1581 DC |
551 | } |
552 | ||
f0d26e86 DC |
553 | /* |
554 | * xfs_file_dio_aio_write - handle direct IO writes | |
555 | * | |
556 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 557 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
558 | * follow locking changes and looping. |
559 | * | |
eda77982 DC |
560 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
561 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
562 | * pages are flushed out. | |
563 | * | |
564 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
565 | * allowing them to be done in parallel with reads and other direct IO writes. | |
566 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
567 | * needs to do sub-block zeroing and that requires serialisation against other | |
568 | * direct IOs to the same block. In this case we need to serialise the | |
569 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
570 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
571 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
572 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 573 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 574 | * |
f0d26e86 DC |
575 | * Returns with locks held indicated by @iolock and errors indicated by |
576 | * negative return values. | |
577 | */ | |
578 | STATIC ssize_t | |
579 | xfs_file_dio_aio_write( | |
580 | struct kiocb *iocb, | |
b3188919 | 581 | struct iov_iter *from) |
f0d26e86 DC |
582 | { |
583 | struct file *file = iocb->ki_filp; | |
584 | struct address_space *mapping = file->f_mapping; | |
585 | struct inode *inode = mapping->host; | |
586 | struct xfs_inode *ip = XFS_I(inode); | |
587 | struct xfs_mount *mp = ip->i_mount; | |
588 | ssize_t ret = 0; | |
eda77982 | 589 | int unaligned_io = 0; |
d0606464 | 590 | int iolock; |
b3188919 AV |
591 | size_t count = iov_iter_count(from); |
592 | loff_t pos = iocb->ki_pos; | |
f0d26e86 DC |
593 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
594 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
595 | ||
7c71ee78 ES |
596 | /* DIO must be aligned to device logical sector size */ |
597 | if ((pos | count) & target->bt_logical_sectormask) | |
b474c7ae | 598 | return -EINVAL; |
f0d26e86 | 599 | |
7c71ee78 | 600 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
601 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
602 | unaligned_io = 1; | |
603 | ||
7271d243 DC |
604 | /* |
605 | * We don't need to take an exclusive lock unless there page cache needs | |
606 | * to be invalidated or unaligned IO is being executed. We don't need to | |
607 | * consider the EOF extension case here because | |
608 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
609 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
610 | */ | |
611 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 612 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 613 | else |
d0606464 CH |
614 | iolock = XFS_IOLOCK_SHARED; |
615 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
616 | |
617 | /* | |
618 | * Recheck if there are cached pages that need invalidate after we got | |
619 | * the iolock to protect against other threads adding new pages while | |
620 | * we were waiting for the iolock. | |
621 | */ | |
d0606464 CH |
622 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
623 | xfs_rw_iunlock(ip, iolock); | |
624 | iolock = XFS_IOLOCK_EXCL; | |
625 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 626 | } |
f0d26e86 | 627 | |
d0606464 | 628 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 629 | if (ret) |
d0606464 | 630 | goto out; |
b3188919 | 631 | iov_iter_truncate(from, count); |
f0d26e86 DC |
632 | |
633 | if (mapping->nrpages) { | |
07d5035a | 634 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
fb595814 | 635 | pos, -1); |
f0d26e86 | 636 | if (ret) |
d0606464 | 637 | goto out; |
fb595814 | 638 | truncate_pagecache_range(VFS_I(ip), pos, -1); |
f0d26e86 DC |
639 | } |
640 | ||
eda77982 DC |
641 | /* |
642 | * If we are doing unaligned IO, wait for all other IO to drain, | |
643 | * otherwise demote the lock if we had to flush cached pages | |
644 | */ | |
645 | if (unaligned_io) | |
4a06fd26 | 646 | inode_dio_wait(inode); |
d0606464 | 647 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 648 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 649 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
650 | } |
651 | ||
652 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
b3188919 | 653 | ret = generic_file_direct_write(iocb, from, pos); |
f0d26e86 | 654 | |
d0606464 CH |
655 | out: |
656 | xfs_rw_iunlock(ip, iolock); | |
657 | ||
f0d26e86 DC |
658 | /* No fallback to buffered IO on errors for XFS. */ |
659 | ASSERT(ret < 0 || ret == count); | |
660 | return ret; | |
661 | } | |
662 | ||
00258e36 | 663 | STATIC ssize_t |
637bbc75 | 664 | xfs_file_buffered_aio_write( |
dda35b8f | 665 | struct kiocb *iocb, |
b3188919 | 666 | struct iov_iter *from) |
dda35b8f CH |
667 | { |
668 | struct file *file = iocb->ki_filp; | |
669 | struct address_space *mapping = file->f_mapping; | |
670 | struct inode *inode = mapping->host; | |
00258e36 | 671 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
672 | ssize_t ret; |
673 | int enospc = 0; | |
d0606464 | 674 | int iolock = XFS_IOLOCK_EXCL; |
b3188919 AV |
675 | loff_t pos = iocb->ki_pos; |
676 | size_t count = iov_iter_count(from); | |
dda35b8f | 677 | |
d0606464 | 678 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 679 | |
d0606464 | 680 | ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock); |
4d8d1581 | 681 | if (ret) |
d0606464 | 682 | goto out; |
dda35b8f | 683 | |
b3188919 | 684 | iov_iter_truncate(from, count); |
dda35b8f CH |
685 | /* We can write back this queue in page reclaim */ |
686 | current->backing_dev_info = mapping->backing_dev_info; | |
687 | ||
dda35b8f | 688 | write_retry: |
637bbc75 | 689 | trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0); |
b3188919 | 690 | ret = generic_perform_write(file, from, pos); |
0a64bc2c AV |
691 | if (likely(ret >= 0)) |
692 | iocb->ki_pos = pos + ret; | |
dc06f398 | 693 | |
637bbc75 | 694 | /* |
dc06f398 BF |
695 | * If we hit a space limit, try to free up some lingering preallocated |
696 | * space before returning an error. In the case of ENOSPC, first try to | |
697 | * write back all dirty inodes to free up some of the excess reserved | |
698 | * metadata space. This reduces the chances that the eofblocks scan | |
699 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
700 | * also behaves as a filter to prevent too many eofblocks scans from | |
701 | * running at the same time. | |
637bbc75 | 702 | */ |
dc06f398 BF |
703 | if (ret == -EDQUOT && !enospc) { |
704 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
705 | if (enospc) | |
706 | goto write_retry; | |
707 | } else if (ret == -ENOSPC && !enospc) { | |
708 | struct xfs_eofblocks eofb = {0}; | |
709 | ||
637bbc75 | 710 | enospc = 1; |
9aa05000 | 711 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
712 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
713 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
714 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 715 | goto write_retry; |
dda35b8f | 716 | } |
d0606464 | 717 | |
dda35b8f | 718 | current->backing_dev_info = NULL; |
d0606464 CH |
719 | out: |
720 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
721 | return ret; |
722 | } | |
723 | ||
724 | STATIC ssize_t | |
bf97f3bc | 725 | xfs_file_write_iter( |
637bbc75 | 726 | struct kiocb *iocb, |
bf97f3bc | 727 | struct iov_iter *from) |
637bbc75 DC |
728 | { |
729 | struct file *file = iocb->ki_filp; | |
730 | struct address_space *mapping = file->f_mapping; | |
731 | struct inode *inode = mapping->host; | |
732 | struct xfs_inode *ip = XFS_I(inode); | |
733 | ssize_t ret; | |
bf97f3bc | 734 | size_t ocount = iov_iter_count(from); |
637bbc75 DC |
735 | |
736 | XFS_STATS_INC(xs_write_calls); | |
737 | ||
637bbc75 DC |
738 | if (ocount == 0) |
739 | return 0; | |
740 | ||
bf97f3bc AV |
741 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
742 | return -EIO; | |
637bbc75 DC |
743 | |
744 | if (unlikely(file->f_flags & O_DIRECT)) | |
bf97f3bc | 745 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 746 | else |
bf97f3bc | 747 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 748 | |
d0606464 CH |
749 | if (ret > 0) { |
750 | ssize_t err; | |
dda35b8f | 751 | |
d0606464 | 752 | XFS_STATS_ADD(xs_write_bytes, ret); |
dda35b8f | 753 | |
d0606464 | 754 | /* Handle various SYNC-type writes */ |
d311d79d | 755 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
756 | if (err < 0) |
757 | ret = err; | |
dda35b8f | 758 | } |
a363f0c2 | 759 | return ret; |
dda35b8f CH |
760 | } |
761 | ||
2fe17c10 CH |
762 | STATIC long |
763 | xfs_file_fallocate( | |
83aee9e4 CH |
764 | struct file *file, |
765 | int mode, | |
766 | loff_t offset, | |
767 | loff_t len) | |
2fe17c10 | 768 | { |
83aee9e4 CH |
769 | struct inode *inode = file_inode(file); |
770 | struct xfs_inode *ip = XFS_I(inode); | |
771 | struct xfs_trans *tp; | |
772 | long error; | |
773 | loff_t new_size = 0; | |
2fe17c10 | 774 | |
83aee9e4 CH |
775 | if (!S_ISREG(inode->i_mode)) |
776 | return -EINVAL; | |
e1d8fb88 | 777 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
376ba313 | 778 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) |
2fe17c10 CH |
779 | return -EOPNOTSUPP; |
780 | ||
2fe17c10 | 781 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
83aee9e4 CH |
782 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
783 | error = xfs_free_file_space(ip, offset, len); | |
784 | if (error) | |
785 | goto out_unlock; | |
e1d8fb88 NJ |
786 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
787 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
788 | ||
789 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 790 | error = -EINVAL; |
e1d8fb88 NJ |
791 | goto out_unlock; |
792 | } | |
793 | ||
23fffa92 LC |
794 | /* |
795 | * There is no need to overlap collapse range with EOF, | |
796 | * in which case it is effectively a truncate operation | |
797 | */ | |
798 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 799 | error = -EINVAL; |
23fffa92 LC |
800 | goto out_unlock; |
801 | } | |
802 | ||
e1d8fb88 NJ |
803 | new_size = i_size_read(inode) - len; |
804 | ||
805 | error = xfs_collapse_file_space(ip, offset, len); | |
806 | if (error) | |
807 | goto out_unlock; | |
83aee9e4 CH |
808 | } else { |
809 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
810 | offset + len > i_size_read(inode)) { | |
811 | new_size = offset + len; | |
2451337d | 812 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
813 | if (error) |
814 | goto out_unlock; | |
815 | } | |
2fe17c10 | 816 | |
376ba313 LC |
817 | if (mode & FALLOC_FL_ZERO_RANGE) |
818 | error = xfs_zero_file_space(ip, offset, len); | |
819 | else | |
820 | error = xfs_alloc_file_space(ip, offset, len, | |
821 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
822 | if (error) |
823 | goto out_unlock; | |
824 | } | |
825 | ||
83aee9e4 CH |
826 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); |
827 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
828 | if (error) { | |
829 | xfs_trans_cancel(tp, 0); | |
830 | goto out_unlock; | |
831 | } | |
832 | ||
833 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
834 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
835 | ip->i_d.di_mode &= ~S_ISUID; | |
836 | if (ip->i_d.di_mode & S_IXGRP) | |
837 | ip->i_d.di_mode &= ~S_ISGID; | |
82878897 | 838 | |
e1d8fb88 | 839 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_COLLAPSE_RANGE))) |
83aee9e4 CH |
840 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; |
841 | ||
842 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
843 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
844 | ||
845 | if (file->f_flags & O_DSYNC) | |
846 | xfs_trans_set_sync(tp); | |
847 | error = xfs_trans_commit(tp, 0); | |
2fe17c10 CH |
848 | if (error) |
849 | goto out_unlock; | |
850 | ||
851 | /* Change file size if needed */ | |
852 | if (new_size) { | |
853 | struct iattr iattr; | |
854 | ||
855 | iattr.ia_valid = ATTR_SIZE; | |
856 | iattr.ia_size = new_size; | |
83aee9e4 | 857 | error = xfs_setattr_size(ip, &iattr); |
2fe17c10 CH |
858 | } |
859 | ||
860 | out_unlock: | |
861 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); | |
2451337d | 862 | return error; |
2fe17c10 CH |
863 | } |
864 | ||
865 | ||
1da177e4 | 866 | STATIC int |
3562fd45 | 867 | xfs_file_open( |
1da177e4 | 868 | struct inode *inode, |
f999a5bf | 869 | struct file *file) |
1da177e4 | 870 | { |
f999a5bf | 871 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 872 | return -EFBIG; |
f999a5bf CH |
873 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
874 | return -EIO; | |
875 | return 0; | |
876 | } | |
877 | ||
878 | STATIC int | |
879 | xfs_dir_open( | |
880 | struct inode *inode, | |
881 | struct file *file) | |
882 | { | |
883 | struct xfs_inode *ip = XFS_I(inode); | |
884 | int mode; | |
885 | int error; | |
886 | ||
887 | error = xfs_file_open(inode, file); | |
888 | if (error) | |
889 | return error; | |
890 | ||
891 | /* | |
892 | * If there are any blocks, read-ahead block 0 as we're almost | |
893 | * certain to have the next operation be a read there. | |
894 | */ | |
309ecac8 | 895 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 896 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 897 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
898 | xfs_iunlock(ip, mode); |
899 | return 0; | |
1da177e4 LT |
900 | } |
901 | ||
1da177e4 | 902 | STATIC int |
3562fd45 | 903 | xfs_file_release( |
1da177e4 LT |
904 | struct inode *inode, |
905 | struct file *filp) | |
906 | { | |
2451337d | 907 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
908 | } |
909 | ||
1da177e4 | 910 | STATIC int |
3562fd45 | 911 | xfs_file_readdir( |
b8227554 AV |
912 | struct file *file, |
913 | struct dir_context *ctx) | |
1da177e4 | 914 | { |
b8227554 | 915 | struct inode *inode = file_inode(file); |
739bfb2a | 916 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
917 | int error; |
918 | size_t bufsize; | |
919 | ||
920 | /* | |
921 | * The Linux API doesn't pass down the total size of the buffer | |
922 | * we read into down to the filesystem. With the filldir concept | |
923 | * it's not needed for correct information, but the XFS dir2 leaf | |
924 | * code wants an estimate of the buffer size to calculate it's | |
925 | * readahead window and size the buffers used for mapping to | |
926 | * physical blocks. | |
927 | * | |
928 | * Try to give it an estimate that's good enough, maybe at some | |
929 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 930 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 931 | */ |
a9cc799e | 932 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 933 | |
b8227554 | 934 | error = xfs_readdir(ip, ctx, bufsize); |
051e7cd4 | 935 | if (error) |
2451337d | 936 | return error; |
051e7cd4 | 937 | return 0; |
1da177e4 LT |
938 | } |
939 | ||
1da177e4 | 940 | STATIC int |
3562fd45 | 941 | xfs_file_mmap( |
1da177e4 LT |
942 | struct file *filp, |
943 | struct vm_area_struct *vma) | |
944 | { | |
3562fd45 | 945 | vma->vm_ops = &xfs_file_vm_ops; |
6fac0cb4 | 946 | |
fbc1462b | 947 | file_accessed(filp); |
1da177e4 LT |
948 | return 0; |
949 | } | |
950 | ||
4f57dbc6 DC |
951 | /* |
952 | * mmap()d file has taken write protection fault and is being made | |
953 | * writable. We can set the page state up correctly for a writable | |
954 | * page, which means we can do correct delalloc accounting (ENOSPC | |
955 | * checking!) and unwritten extent mapping. | |
956 | */ | |
957 | STATIC int | |
958 | xfs_vm_page_mkwrite( | |
959 | struct vm_area_struct *vma, | |
c2ec175c | 960 | struct vm_fault *vmf) |
4f57dbc6 | 961 | { |
c2ec175c | 962 | return block_page_mkwrite(vma, vmf, xfs_get_blocks); |
4f57dbc6 DC |
963 | } |
964 | ||
d126d43f JL |
965 | /* |
966 | * This type is designed to indicate the type of offset we would like | |
967 | * to search from page cache for either xfs_seek_data() or xfs_seek_hole(). | |
968 | */ | |
969 | enum { | |
970 | HOLE_OFF = 0, | |
971 | DATA_OFF, | |
972 | }; | |
973 | ||
974 | /* | |
975 | * Lookup the desired type of offset from the given page. | |
976 | * | |
977 | * On success, return true and the offset argument will point to the | |
978 | * start of the region that was found. Otherwise this function will | |
979 | * return false and keep the offset argument unchanged. | |
980 | */ | |
981 | STATIC bool | |
982 | xfs_lookup_buffer_offset( | |
983 | struct page *page, | |
984 | loff_t *offset, | |
985 | unsigned int type) | |
986 | { | |
987 | loff_t lastoff = page_offset(page); | |
988 | bool found = false; | |
989 | struct buffer_head *bh, *head; | |
990 | ||
991 | bh = head = page_buffers(page); | |
992 | do { | |
993 | /* | |
994 | * Unwritten extents that have data in the page | |
995 | * cache covering them can be identified by the | |
996 | * BH_Unwritten state flag. Pages with multiple | |
997 | * buffers might have a mix of holes, data and | |
998 | * unwritten extents - any buffer with valid | |
999 | * data in it should have BH_Uptodate flag set | |
1000 | * on it. | |
1001 | */ | |
1002 | if (buffer_unwritten(bh) || | |
1003 | buffer_uptodate(bh)) { | |
1004 | if (type == DATA_OFF) | |
1005 | found = true; | |
1006 | } else { | |
1007 | if (type == HOLE_OFF) | |
1008 | found = true; | |
1009 | } | |
1010 | ||
1011 | if (found) { | |
1012 | *offset = lastoff; | |
1013 | break; | |
1014 | } | |
1015 | lastoff += bh->b_size; | |
1016 | } while ((bh = bh->b_this_page) != head); | |
1017 | ||
1018 | return found; | |
1019 | } | |
1020 | ||
1021 | /* | |
1022 | * This routine is called to find out and return a data or hole offset | |
1023 | * from the page cache for unwritten extents according to the desired | |
1024 | * type for xfs_seek_data() or xfs_seek_hole(). | |
1025 | * | |
1026 | * The argument offset is used to tell where we start to search from the | |
1027 | * page cache. Map is used to figure out the end points of the range to | |
1028 | * lookup pages. | |
1029 | * | |
1030 | * Return true if the desired type of offset was found, and the argument | |
1031 | * offset is filled with that address. Otherwise, return false and keep | |
1032 | * offset unchanged. | |
1033 | */ | |
1034 | STATIC bool | |
1035 | xfs_find_get_desired_pgoff( | |
1036 | struct inode *inode, | |
1037 | struct xfs_bmbt_irec *map, | |
1038 | unsigned int type, | |
1039 | loff_t *offset) | |
1040 | { | |
1041 | struct xfs_inode *ip = XFS_I(inode); | |
1042 | struct xfs_mount *mp = ip->i_mount; | |
1043 | struct pagevec pvec; | |
1044 | pgoff_t index; | |
1045 | pgoff_t end; | |
1046 | loff_t endoff; | |
1047 | loff_t startoff = *offset; | |
1048 | loff_t lastoff = startoff; | |
1049 | bool found = false; | |
1050 | ||
1051 | pagevec_init(&pvec, 0); | |
1052 | ||
1053 | index = startoff >> PAGE_CACHE_SHIFT; | |
1054 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); | |
1055 | end = endoff >> PAGE_CACHE_SHIFT; | |
1056 | do { | |
1057 | int want; | |
1058 | unsigned nr_pages; | |
1059 | unsigned int i; | |
1060 | ||
1061 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1062 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1063 | want); | |
1064 | /* | |
1065 | * No page mapped into given range. If we are searching holes | |
1066 | * and if this is the first time we got into the loop, it means | |
1067 | * that the given offset is landed in a hole, return it. | |
1068 | * | |
1069 | * If we have already stepped through some block buffers to find | |
1070 | * holes but they all contains data. In this case, the last | |
1071 | * offset is already updated and pointed to the end of the last | |
1072 | * mapped page, if it does not reach the endpoint to search, | |
1073 | * that means there should be a hole between them. | |
1074 | */ | |
1075 | if (nr_pages == 0) { | |
1076 | /* Data search found nothing */ | |
1077 | if (type == DATA_OFF) | |
1078 | break; | |
1079 | ||
1080 | ASSERT(type == HOLE_OFF); | |
1081 | if (lastoff == startoff || lastoff < endoff) { | |
1082 | found = true; | |
1083 | *offset = lastoff; | |
1084 | } | |
1085 | break; | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * At lease we found one page. If this is the first time we | |
1090 | * step into the loop, and if the first page index offset is | |
1091 | * greater than the given search offset, a hole was found. | |
1092 | */ | |
1093 | if (type == HOLE_OFF && lastoff == startoff && | |
1094 | lastoff < page_offset(pvec.pages[0])) { | |
1095 | found = true; | |
1096 | break; | |
1097 | } | |
1098 | ||
1099 | for (i = 0; i < nr_pages; i++) { | |
1100 | struct page *page = pvec.pages[i]; | |
1101 | loff_t b_offset; | |
1102 | ||
1103 | /* | |
1104 | * At this point, the page may be truncated or | |
1105 | * invalidated (changing page->mapping to NULL), | |
1106 | * or even swizzled back from swapper_space to tmpfs | |
1107 | * file mapping. However, page->index will not change | |
1108 | * because we have a reference on the page. | |
1109 | * | |
1110 | * Searching done if the page index is out of range. | |
1111 | * If the current offset is not reaches the end of | |
1112 | * the specified search range, there should be a hole | |
1113 | * between them. | |
1114 | */ | |
1115 | if (page->index > end) { | |
1116 | if (type == HOLE_OFF && lastoff < endoff) { | |
1117 | *offset = lastoff; | |
1118 | found = true; | |
1119 | } | |
1120 | goto out; | |
1121 | } | |
1122 | ||
1123 | lock_page(page); | |
1124 | /* | |
1125 | * Page truncated or invalidated(page->mapping == NULL). | |
1126 | * We can freely skip it and proceed to check the next | |
1127 | * page. | |
1128 | */ | |
1129 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1130 | unlock_page(page); | |
1131 | continue; | |
1132 | } | |
1133 | ||
1134 | if (!page_has_buffers(page)) { | |
1135 | unlock_page(page); | |
1136 | continue; | |
1137 | } | |
1138 | ||
1139 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1140 | if (found) { | |
1141 | /* | |
1142 | * The found offset may be less than the start | |
1143 | * point to search if this is the first time to | |
1144 | * come here. | |
1145 | */ | |
1146 | *offset = max_t(loff_t, startoff, b_offset); | |
1147 | unlock_page(page); | |
1148 | goto out; | |
1149 | } | |
1150 | ||
1151 | /* | |
1152 | * We either searching data but nothing was found, or | |
1153 | * searching hole but found a data buffer. In either | |
1154 | * case, probably the next page contains the desired | |
1155 | * things, update the last offset to it so. | |
1156 | */ | |
1157 | lastoff = page_offset(page) + PAGE_SIZE; | |
1158 | unlock_page(page); | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * The number of returned pages less than our desired, search | |
1163 | * done. In this case, nothing was found for searching data, | |
1164 | * but we found a hole behind the last offset. | |
1165 | */ | |
1166 | if (nr_pages < want) { | |
1167 | if (type == HOLE_OFF) { | |
1168 | *offset = lastoff; | |
1169 | found = true; | |
1170 | } | |
1171 | break; | |
1172 | } | |
1173 | ||
1174 | index = pvec.pages[i - 1]->index + 1; | |
1175 | pagevec_release(&pvec); | |
1176 | } while (index <= end); | |
1177 | ||
1178 | out: | |
1179 | pagevec_release(&pvec); | |
1180 | return found; | |
1181 | } | |
1182 | ||
3fe3e6b1 JL |
1183 | STATIC loff_t |
1184 | xfs_seek_data( | |
1185 | struct file *file, | |
834ab122 | 1186 | loff_t start) |
3fe3e6b1 JL |
1187 | { |
1188 | struct inode *inode = file->f_mapping->host; | |
1189 | struct xfs_inode *ip = XFS_I(inode); | |
1190 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 JL |
1191 | loff_t uninitialized_var(offset); |
1192 | xfs_fsize_t isize; | |
1193 | xfs_fileoff_t fsbno; | |
1194 | xfs_filblks_t end; | |
1195 | uint lock; | |
1196 | int error; | |
1197 | ||
309ecac8 | 1198 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1199 | |
1200 | isize = i_size_read(inode); | |
1201 | if (start >= isize) { | |
2451337d | 1202 | error = -ENXIO; |
3fe3e6b1 JL |
1203 | goto out_unlock; |
1204 | } | |
1205 | ||
3fe3e6b1 JL |
1206 | /* |
1207 | * Try to read extents from the first block indicated | |
1208 | * by fsbno to the end block of the file. | |
1209 | */ | |
52f1acc8 | 1210 | fsbno = XFS_B_TO_FSBT(mp, start); |
3fe3e6b1 | 1211 | end = XFS_B_TO_FSB(mp, isize); |
52f1acc8 JL |
1212 | for (;;) { |
1213 | struct xfs_bmbt_irec map[2]; | |
1214 | int nmap = 2; | |
1215 | unsigned int i; | |
3fe3e6b1 | 1216 | |
52f1acc8 JL |
1217 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
1218 | XFS_BMAPI_ENTIRE); | |
1219 | if (error) | |
1220 | goto out_unlock; | |
3fe3e6b1 | 1221 | |
52f1acc8 JL |
1222 | /* No extents at given offset, must be beyond EOF */ |
1223 | if (nmap == 0) { | |
2451337d | 1224 | error = -ENXIO; |
52f1acc8 JL |
1225 | goto out_unlock; |
1226 | } | |
1227 | ||
1228 | for (i = 0; i < nmap; i++) { | |
1229 | offset = max_t(loff_t, start, | |
1230 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1231 | ||
1232 | /* Landed in a data extent */ | |
1233 | if (map[i].br_startblock == DELAYSTARTBLOCK || | |
1234 | (map[i].br_state == XFS_EXT_NORM && | |
1235 | !isnullstartblock(map[i].br_startblock))) | |
1236 | goto out; | |
1237 | ||
1238 | /* | |
1239 | * Landed in an unwritten extent, try to search data | |
1240 | * from page cache. | |
1241 | */ | |
1242 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1243 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
1244 | DATA_OFF, &offset)) | |
1245 | goto out; | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * map[0] is hole or its an unwritten extent but | |
1251 | * without data in page cache. Probably means that | |
1252 | * we are reading after EOF if nothing in map[1]. | |
1253 | */ | |
3fe3e6b1 | 1254 | if (nmap == 1) { |
2451337d | 1255 | error = -ENXIO; |
3fe3e6b1 JL |
1256 | goto out_unlock; |
1257 | } | |
1258 | ||
52f1acc8 JL |
1259 | ASSERT(i > 1); |
1260 | ||
1261 | /* | |
1262 | * Nothing was found, proceed to the next round of search | |
1263 | * if reading offset not beyond or hit EOF. | |
1264 | */ | |
1265 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1266 | start = XFS_FSB_TO_B(mp, fsbno); | |
1267 | if (start >= isize) { | |
2451337d | 1268 | error = -ENXIO; |
52f1acc8 JL |
1269 | goto out_unlock; |
1270 | } | |
3fe3e6b1 JL |
1271 | } |
1272 | ||
52f1acc8 | 1273 | out: |
46a1c2c7 | 1274 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1275 | |
1276 | out_unlock: | |
01f4f327 | 1277 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1278 | |
1279 | if (error) | |
2451337d | 1280 | return error; |
3fe3e6b1 JL |
1281 | return offset; |
1282 | } | |
1283 | ||
1284 | STATIC loff_t | |
1285 | xfs_seek_hole( | |
1286 | struct file *file, | |
834ab122 | 1287 | loff_t start) |
3fe3e6b1 JL |
1288 | { |
1289 | struct inode *inode = file->f_mapping->host; | |
1290 | struct xfs_inode *ip = XFS_I(inode); | |
1291 | struct xfs_mount *mp = ip->i_mount; | |
1292 | loff_t uninitialized_var(offset); | |
3fe3e6b1 JL |
1293 | xfs_fsize_t isize; |
1294 | xfs_fileoff_t fsbno; | |
b686d1f7 | 1295 | xfs_filblks_t end; |
3fe3e6b1 JL |
1296 | uint lock; |
1297 | int error; | |
1298 | ||
1299 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1300 | return -EIO; |
3fe3e6b1 | 1301 | |
309ecac8 | 1302 | lock = xfs_ilock_data_map_shared(ip); |
3fe3e6b1 JL |
1303 | |
1304 | isize = i_size_read(inode); | |
1305 | if (start >= isize) { | |
2451337d | 1306 | error = -ENXIO; |
3fe3e6b1 JL |
1307 | goto out_unlock; |
1308 | } | |
1309 | ||
1310 | fsbno = XFS_B_TO_FSBT(mp, start); | |
b686d1f7 JL |
1311 | end = XFS_B_TO_FSB(mp, isize); |
1312 | ||
1313 | for (;;) { | |
1314 | struct xfs_bmbt_irec map[2]; | |
1315 | int nmap = 2; | |
1316 | unsigned int i; | |
1317 | ||
1318 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, | |
1319 | XFS_BMAPI_ENTIRE); | |
1320 | if (error) | |
1321 | goto out_unlock; | |
1322 | ||
1323 | /* No extents at given offset, must be beyond EOF */ | |
1324 | if (nmap == 0) { | |
2451337d | 1325 | error = -ENXIO; |
b686d1f7 JL |
1326 | goto out_unlock; |
1327 | } | |
1328 | ||
1329 | for (i = 0; i < nmap; i++) { | |
1330 | offset = max_t(loff_t, start, | |
1331 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1332 | ||
1333 | /* Landed in a hole */ | |
1334 | if (map[i].br_startblock == HOLESTARTBLOCK) | |
1335 | goto out; | |
1336 | ||
1337 | /* | |
1338 | * Landed in an unwritten extent, try to search hole | |
1339 | * from page cache. | |
1340 | */ | |
1341 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1342 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
1343 | HOLE_OFF, &offset)) | |
1344 | goto out; | |
1345 | } | |
1346 | } | |
3fe3e6b1 | 1347 | |
3fe3e6b1 | 1348 | /* |
b686d1f7 JL |
1349 | * map[0] contains data or its unwritten but contains |
1350 | * data in page cache, probably means that we are | |
1351 | * reading after EOF. We should fix offset to point | |
1352 | * to the end of the file(i.e., there is an implicit | |
1353 | * hole at the end of any file). | |
3fe3e6b1 | 1354 | */ |
b686d1f7 JL |
1355 | if (nmap == 1) { |
1356 | offset = isize; | |
1357 | break; | |
1358 | } | |
1359 | ||
1360 | ASSERT(i > 1); | |
1361 | ||
1362 | /* | |
1363 | * Both mappings contains data, proceed to the next round of | |
1364 | * search if the current reading offset not beyond or hit EOF. | |
1365 | */ | |
1366 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1367 | start = XFS_FSB_TO_B(mp, fsbno); | |
1368 | if (start >= isize) { | |
1369 | offset = isize; | |
1370 | break; | |
1371 | } | |
3fe3e6b1 JL |
1372 | } |
1373 | ||
b686d1f7 JL |
1374 | out: |
1375 | /* | |
1376 | * At this point, we must have found a hole. However, the returned | |
1377 | * offset may be bigger than the file size as it may be aligned to | |
1378 | * page boundary for unwritten extents, we need to deal with this | |
1379 | * situation in particular. | |
1380 | */ | |
1381 | offset = min_t(loff_t, offset, isize); | |
46a1c2c7 | 1382 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1383 | |
1384 | out_unlock: | |
01f4f327 | 1385 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1386 | |
1387 | if (error) | |
2451337d | 1388 | return error; |
3fe3e6b1 JL |
1389 | return offset; |
1390 | } | |
1391 | ||
1392 | STATIC loff_t | |
1393 | xfs_file_llseek( | |
1394 | struct file *file, | |
1395 | loff_t offset, | |
1396 | int origin) | |
1397 | { | |
1398 | switch (origin) { | |
1399 | case SEEK_END: | |
1400 | case SEEK_CUR: | |
1401 | case SEEK_SET: | |
1402 | return generic_file_llseek(file, offset, origin); | |
1403 | case SEEK_DATA: | |
834ab122 | 1404 | return xfs_seek_data(file, offset); |
3fe3e6b1 | 1405 | case SEEK_HOLE: |
834ab122 | 1406 | return xfs_seek_hole(file, offset); |
3fe3e6b1 JL |
1407 | default: |
1408 | return -EINVAL; | |
1409 | } | |
1410 | } | |
1411 | ||
4b6f5d20 | 1412 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1413 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1414 | .read = new_sync_read, |
bf97f3bc | 1415 | .write = new_sync_write, |
b4f5d2c6 | 1416 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1417 | .write_iter = xfs_file_write_iter, |
1b895840 | 1418 | .splice_read = xfs_file_splice_read, |
8d020765 | 1419 | .splice_write = iter_file_splice_write, |
3562fd45 | 1420 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1421 | #ifdef CONFIG_COMPAT |
3562fd45 | 1422 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1423 | #endif |
3562fd45 NS |
1424 | .mmap = xfs_file_mmap, |
1425 | .open = xfs_file_open, | |
1426 | .release = xfs_file_release, | |
1427 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1428 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1429 | }; |
1430 | ||
4b6f5d20 | 1431 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1432 | .open = xfs_dir_open, |
1da177e4 | 1433 | .read = generic_read_dir, |
b8227554 | 1434 | .iterate = xfs_file_readdir, |
59af1584 | 1435 | .llseek = generic_file_llseek, |
3562fd45 | 1436 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1437 | #ifdef CONFIG_COMPAT |
3562fd45 | 1438 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1439 | #endif |
1da2f2db | 1440 | .fsync = xfs_dir_fsync, |
1da177e4 LT |
1441 | }; |
1442 | ||
f0f37e2f | 1443 | static const struct vm_operations_struct xfs_file_vm_ops = { |
54cb8821 | 1444 | .fault = filemap_fault, |
f1820361 | 1445 | .map_pages = filemap_map_pages, |
4f57dbc6 | 1446 | .page_mkwrite = xfs_vm_page_mkwrite, |
0b173bc4 | 1447 | .remap_pages = generic_file_remap_pages, |
6fac0cb4 | 1448 | }; |