Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
dda35b8f | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
a4fbe6ab | 21 | #include "xfs_format.h" |
239880ef DC |
22 | #include "xfs_log_format.h" |
23 | #include "xfs_trans_resv.h" | |
1da177e4 | 24 | #include "xfs_mount.h" |
57062787 DC |
25 | #include "xfs_da_format.h" |
26 | #include "xfs_da_btree.h" | |
1da177e4 | 27 | #include "xfs_inode.h" |
239880ef | 28 | #include "xfs_trans.h" |
fd3200be | 29 | #include "xfs_inode_item.h" |
dda35b8f | 30 | #include "xfs_bmap.h" |
c24b5dfa | 31 | #include "xfs_bmap_util.h" |
1da177e4 | 32 | #include "xfs_error.h" |
2b9ab5ab | 33 | #include "xfs_dir2.h" |
c24b5dfa | 34 | #include "xfs_dir2_priv.h" |
ddcd856d | 35 | #include "xfs_ioctl.h" |
dda35b8f | 36 | #include "xfs_trace.h" |
239880ef | 37 | #include "xfs_log.h" |
dc06f398 | 38 | #include "xfs_icache.h" |
781355c6 | 39 | #include "xfs_pnfs.h" |
1da177e4 LT |
40 | |
41 | #include <linux/dcache.h> | |
2fe17c10 | 42 | #include <linux/falloc.h> |
d126d43f | 43 | #include <linux/pagevec.h> |
66114cad | 44 | #include <linux/backing-dev.h> |
1da177e4 | 45 | |
f0f37e2f | 46 | static const struct vm_operations_struct xfs_file_vm_ops; |
1da177e4 | 47 | |
487f84f3 DC |
48 | /* |
49 | * Locking primitives for read and write IO paths to ensure we consistently use | |
50 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. | |
51 | */ | |
52 | static inline void | |
53 | xfs_rw_ilock( | |
54 | struct xfs_inode *ip, | |
55 | int type) | |
56 | { | |
57 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 58 | inode_lock(VFS_I(ip)); |
487f84f3 DC |
59 | xfs_ilock(ip, type); |
60 | } | |
61 | ||
62 | static inline void | |
63 | xfs_rw_iunlock( | |
64 | struct xfs_inode *ip, | |
65 | int type) | |
66 | { | |
67 | xfs_iunlock(ip, type); | |
68 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 69 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
70 | } |
71 | ||
72 | static inline void | |
73 | xfs_rw_ilock_demote( | |
74 | struct xfs_inode *ip, | |
75 | int type) | |
76 | { | |
77 | xfs_ilock_demote(ip, type); | |
78 | if (type & XFS_IOLOCK_EXCL) | |
5955102c | 79 | inode_unlock(VFS_I(ip)); |
487f84f3 DC |
80 | } |
81 | ||
dda35b8f | 82 | /* |
4f69f578 DC |
83 | * xfs_iozero clears the specified range supplied via the page cache (except in |
84 | * the DAX case). Writes through the page cache will allocate blocks over holes, | |
85 | * though the callers usually map the holes first and avoid them. If a block is | |
86 | * not completely zeroed, then it will be read from disk before being partially | |
87 | * zeroed. | |
dda35b8f | 88 | * |
4f69f578 DC |
89 | * In the DAX case, we can just directly write to the underlying pages. This |
90 | * will not allocate blocks, but will avoid holes and unwritten extents and so | |
91 | * not do unnecessary work. | |
dda35b8f | 92 | */ |
ef9d8733 | 93 | int |
dda35b8f CH |
94 | xfs_iozero( |
95 | struct xfs_inode *ip, /* inode */ | |
96 | loff_t pos, /* offset in file */ | |
97 | size_t count) /* size of data to zero */ | |
98 | { | |
99 | struct page *page; | |
100 | struct address_space *mapping; | |
4f69f578 DC |
101 | int status = 0; |
102 | ||
dda35b8f CH |
103 | |
104 | mapping = VFS_I(ip)->i_mapping; | |
105 | do { | |
106 | unsigned offset, bytes; | |
107 | void *fsdata; | |
108 | ||
09cbfeaf KS |
109 | offset = (pos & (PAGE_SIZE -1)); /* Within page */ |
110 | bytes = PAGE_SIZE - offset; | |
dda35b8f CH |
111 | if (bytes > count) |
112 | bytes = count; | |
113 | ||
4f69f578 DC |
114 | if (IS_DAX(VFS_I(ip))) { |
115 | status = dax_zero_page_range(VFS_I(ip), pos, bytes, | |
116 | xfs_get_blocks_direct); | |
117 | if (status) | |
118 | break; | |
119 | } else { | |
120 | status = pagecache_write_begin(NULL, mapping, pos, bytes, | |
121 | AOP_FLAG_UNINTERRUPTIBLE, | |
122 | &page, &fsdata); | |
123 | if (status) | |
124 | break; | |
dda35b8f | 125 | |
4f69f578 | 126 | zero_user(page, offset, bytes); |
dda35b8f | 127 | |
4f69f578 DC |
128 | status = pagecache_write_end(NULL, mapping, pos, bytes, |
129 | bytes, page, fsdata); | |
130 | WARN_ON(status <= 0); /* can't return less than zero! */ | |
131 | status = 0; | |
132 | } | |
dda35b8f CH |
133 | pos += bytes; |
134 | count -= bytes; | |
dda35b8f CH |
135 | } while (count); |
136 | ||
cddc1162 | 137 | return status; |
dda35b8f CH |
138 | } |
139 | ||
8add71ca CH |
140 | int |
141 | xfs_update_prealloc_flags( | |
142 | struct xfs_inode *ip, | |
143 | enum xfs_prealloc_flags flags) | |
144 | { | |
145 | struct xfs_trans *tp; | |
146 | int error; | |
147 | ||
148 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); | |
149 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); | |
150 | if (error) { | |
4906e215 | 151 | xfs_trans_cancel(tp); |
8add71ca CH |
152 | return error; |
153 | } | |
154 | ||
155 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
156 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
157 | ||
158 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { | |
c19b3b05 DC |
159 | VFS_I(ip)->i_mode &= ~S_ISUID; |
160 | if (VFS_I(ip)->i_mode & S_IXGRP) | |
161 | VFS_I(ip)->i_mode &= ~S_ISGID; | |
8add71ca CH |
162 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
163 | } | |
164 | ||
165 | if (flags & XFS_PREALLOC_SET) | |
166 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; | |
167 | if (flags & XFS_PREALLOC_CLEAR) | |
168 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; | |
169 | ||
170 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
171 | if (flags & XFS_PREALLOC_SYNC) | |
172 | xfs_trans_set_sync(tp); | |
70393313 | 173 | return xfs_trans_commit(tp); |
8add71ca CH |
174 | } |
175 | ||
1da2f2db CH |
176 | /* |
177 | * Fsync operations on directories are much simpler than on regular files, | |
178 | * as there is no file data to flush, and thus also no need for explicit | |
179 | * cache flush operations, and there are no non-transaction metadata updates | |
180 | * on directories either. | |
181 | */ | |
182 | STATIC int | |
183 | xfs_dir_fsync( | |
184 | struct file *file, | |
185 | loff_t start, | |
186 | loff_t end, | |
187 | int datasync) | |
188 | { | |
189 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); | |
190 | struct xfs_mount *mp = ip->i_mount; | |
191 | xfs_lsn_t lsn = 0; | |
192 | ||
193 | trace_xfs_dir_fsync(ip); | |
194 | ||
195 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
196 | if (xfs_ipincount(ip)) | |
197 | lsn = ip->i_itemp->ili_last_lsn; | |
198 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
199 | ||
200 | if (!lsn) | |
201 | return 0; | |
2451337d | 202 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
1da2f2db CH |
203 | } |
204 | ||
fd3200be CH |
205 | STATIC int |
206 | xfs_file_fsync( | |
207 | struct file *file, | |
02c24a82 JB |
208 | loff_t start, |
209 | loff_t end, | |
fd3200be CH |
210 | int datasync) |
211 | { | |
7ea80859 CH |
212 | struct inode *inode = file->f_mapping->host; |
213 | struct xfs_inode *ip = XFS_I(inode); | |
a27a263b | 214 | struct xfs_mount *mp = ip->i_mount; |
fd3200be CH |
215 | int error = 0; |
216 | int log_flushed = 0; | |
b1037058 | 217 | xfs_lsn_t lsn = 0; |
fd3200be | 218 | |
cca28fb8 | 219 | trace_xfs_file_fsync(ip); |
fd3200be | 220 | |
02c24a82 JB |
221 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
222 | if (error) | |
223 | return error; | |
224 | ||
a27a263b | 225 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 226 | return -EIO; |
fd3200be CH |
227 | |
228 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
229 | ||
a27a263b CH |
230 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
231 | /* | |
232 | * If we have an RT and/or log subvolume we need to make sure | |
233 | * to flush the write cache the device used for file data | |
234 | * first. This is to ensure newly written file data make | |
235 | * it to disk before logging the new inode size in case of | |
236 | * an extending write. | |
237 | */ | |
238 | if (XFS_IS_REALTIME_INODE(ip)) | |
239 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); | |
240 | else if (mp->m_logdev_targp != mp->m_ddev_targp) | |
241 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
242 | } | |
243 | ||
fd3200be | 244 | /* |
fc0561ce DC |
245 | * All metadata updates are logged, which means that we just have to |
246 | * flush the log up to the latest LSN that touched the inode. If we have | |
247 | * concurrent fsync/fdatasync() calls, we need them to all block on the | |
248 | * log force before we clear the ili_fsync_fields field. This ensures | |
249 | * that we don't get a racing sync operation that does not wait for the | |
250 | * metadata to hit the journal before returning. If we race with | |
251 | * clearing the ili_fsync_fields, then all that will happen is the log | |
252 | * force will do nothing as the lsn will already be on disk. We can't | |
253 | * race with setting ili_fsync_fields because that is done under | |
254 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared | |
255 | * until after the ili_fsync_fields is cleared. | |
fd3200be CH |
256 | */ |
257 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
8f639dde CH |
258 | if (xfs_ipincount(ip)) { |
259 | if (!datasync || | |
fc0561ce | 260 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
8f639dde CH |
261 | lsn = ip->i_itemp->ili_last_lsn; |
262 | } | |
fd3200be | 263 | |
fc0561ce | 264 | if (lsn) { |
b1037058 | 265 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
fc0561ce DC |
266 | ip->i_itemp->ili_fsync_fields = 0; |
267 | } | |
268 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
b1037058 | 269 | |
a27a263b CH |
270 | /* |
271 | * If we only have a single device, and the log force about was | |
272 | * a no-op we might have to flush the data device cache here. | |
273 | * This can only happen for fdatasync/O_DSYNC if we were overwriting | |
274 | * an already allocated file and thus do not have any metadata to | |
275 | * commit. | |
276 | */ | |
277 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && | |
278 | mp->m_logdev_targp == mp->m_ddev_targp && | |
279 | !XFS_IS_REALTIME_INODE(ip) && | |
280 | !log_flushed) | |
281 | xfs_blkdev_issue_flush(mp->m_ddev_targp); | |
fd3200be | 282 | |
2451337d | 283 | return error; |
fd3200be CH |
284 | } |
285 | ||
00258e36 | 286 | STATIC ssize_t |
b4f5d2c6 | 287 | xfs_file_read_iter( |
dda35b8f | 288 | struct kiocb *iocb, |
b4f5d2c6 | 289 | struct iov_iter *to) |
dda35b8f CH |
290 | { |
291 | struct file *file = iocb->ki_filp; | |
292 | struct inode *inode = file->f_mapping->host; | |
00258e36 CH |
293 | struct xfs_inode *ip = XFS_I(inode); |
294 | struct xfs_mount *mp = ip->i_mount; | |
b4f5d2c6 | 295 | size_t size = iov_iter_count(to); |
dda35b8f | 296 | ssize_t ret = 0; |
00258e36 | 297 | int ioflags = 0; |
dda35b8f | 298 | xfs_fsize_t n; |
b4f5d2c6 | 299 | loff_t pos = iocb->ki_pos; |
dda35b8f | 300 | |
ff6d6af2 | 301 | XFS_STATS_INC(mp, xs_read_calls); |
dda35b8f | 302 | |
2ba48ce5 | 303 | if (unlikely(iocb->ki_flags & IOCB_DIRECT)) |
b92cc59f | 304 | ioflags |= XFS_IO_ISDIRECT; |
00258e36 | 305 | if (file->f_mode & FMODE_NOCMTIME) |
b92cc59f | 306 | ioflags |= XFS_IO_INVIS; |
00258e36 | 307 | |
6b698ede | 308 | if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) { |
dda35b8f CH |
309 | xfs_buftarg_t *target = |
310 | XFS_IS_REALTIME_INODE(ip) ? | |
311 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
7c71ee78 ES |
312 | /* DIO must be aligned to device logical sector size */ |
313 | if ((pos | size) & target->bt_logical_sectormask) { | |
fb595814 | 314 | if (pos == i_size_read(inode)) |
00258e36 | 315 | return 0; |
b474c7ae | 316 | return -EINVAL; |
dda35b8f CH |
317 | } |
318 | } | |
319 | ||
fb595814 | 320 | n = mp->m_super->s_maxbytes - pos; |
00258e36 | 321 | if (n <= 0 || size == 0) |
dda35b8f CH |
322 | return 0; |
323 | ||
324 | if (n < size) | |
325 | size = n; | |
326 | ||
327 | if (XFS_FORCED_SHUTDOWN(mp)) | |
328 | return -EIO; | |
329 | ||
0c38a251 | 330 | /* |
3d751af2 BF |
331 | * Locking is a bit tricky here. If we take an exclusive lock for direct |
332 | * IO, we effectively serialise all new concurrent read IO to this file | |
333 | * and block it behind IO that is currently in progress because IO in | |
334 | * progress holds the IO lock shared. We only need to hold the lock | |
335 | * exclusive to blow away the page cache, so only take lock exclusively | |
336 | * if the page cache needs invalidation. This allows the normal direct | |
337 | * IO case of no page cache pages to proceeed concurrently without | |
338 | * serialisation. | |
0c38a251 DC |
339 | */ |
340 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); | |
b92cc59f | 341 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
0c38a251 | 342 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
487f84f3 DC |
343 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
344 | ||
3d751af2 BF |
345 | /* |
346 | * The generic dio code only flushes the range of the particular | |
347 | * I/O. Because we take an exclusive lock here, this whole | |
348 | * sequence is considerably more expensive for us. This has a | |
349 | * noticeable performance impact for any file with cached pages, | |
350 | * even when outside of the range of the particular I/O. | |
351 | * | |
352 | * Hence, amortize the cost of the lock against a full file | |
353 | * flush and reduce the chances of repeated iolock cycles going | |
354 | * forward. | |
355 | */ | |
00258e36 | 356 | if (inode->i_mapping->nrpages) { |
3d751af2 | 357 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
487f84f3 DC |
358 | if (ret) { |
359 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); | |
360 | return ret; | |
361 | } | |
85e584da CM |
362 | |
363 | /* | |
364 | * Invalidate whole pages. This can return an error if | |
365 | * we fail to invalidate a page, but this should never | |
366 | * happen on XFS. Warn if it does fail. | |
367 | */ | |
3d751af2 | 368 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
85e584da CM |
369 | WARN_ON_ONCE(ret); |
370 | ret = 0; | |
00258e36 | 371 | } |
487f84f3 | 372 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
0c38a251 | 373 | } |
dda35b8f | 374 | |
fb595814 | 375 | trace_xfs_file_read(ip, size, pos, ioflags); |
dda35b8f | 376 | |
b4f5d2c6 | 377 | ret = generic_file_read_iter(iocb, to); |
dda35b8f | 378 | if (ret > 0) |
ff6d6af2 | 379 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
dda35b8f | 380 | |
487f84f3 | 381 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
dda35b8f CH |
382 | return ret; |
383 | } | |
384 | ||
00258e36 CH |
385 | STATIC ssize_t |
386 | xfs_file_splice_read( | |
dda35b8f CH |
387 | struct file *infilp, |
388 | loff_t *ppos, | |
389 | struct pipe_inode_info *pipe, | |
390 | size_t count, | |
00258e36 | 391 | unsigned int flags) |
dda35b8f | 392 | { |
00258e36 | 393 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
00258e36 | 394 | int ioflags = 0; |
dda35b8f CH |
395 | ssize_t ret; |
396 | ||
ff6d6af2 | 397 | XFS_STATS_INC(ip->i_mount, xs_read_calls); |
00258e36 CH |
398 | |
399 | if (infilp->f_mode & FMODE_NOCMTIME) | |
b92cc59f | 400 | ioflags |= XFS_IO_INVIS; |
00258e36 | 401 | |
dda35b8f CH |
402 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
403 | return -EIO; | |
404 | ||
dda35b8f CH |
405 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
406 | ||
a6d7636e DC |
407 | /* |
408 | * DAX inodes cannot ues the page cache for splice, so we have to push | |
409 | * them through the VFS IO path. This means it goes through | |
410 | * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we | |
411 | * cannot lock the splice operation at this level for DAX inodes. | |
412 | */ | |
413 | if (IS_DAX(VFS_I(ip))) { | |
414 | ret = default_file_splice_read(infilp, ppos, pipe, count, | |
415 | flags); | |
416 | goto out; | |
417 | } | |
dda35b8f | 418 | |
a6d7636e DC |
419 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
420 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); | |
487f84f3 | 421 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
a6d7636e DC |
422 | out: |
423 | if (ret > 0) | |
424 | XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret); | |
dda35b8f CH |
425 | return ret; |
426 | } | |
427 | ||
dda35b8f | 428 | /* |
193aec10 CH |
429 | * This routine is called to handle zeroing any space in the last block of the |
430 | * file that is beyond the EOF. We do this since the size is being increased | |
431 | * without writing anything to that block and we don't want to read the | |
432 | * garbage on the disk. | |
dda35b8f CH |
433 | */ |
434 | STATIC int /* error (positive) */ | |
435 | xfs_zero_last_block( | |
193aec10 CH |
436 | struct xfs_inode *ip, |
437 | xfs_fsize_t offset, | |
5885ebda DC |
438 | xfs_fsize_t isize, |
439 | bool *did_zeroing) | |
dda35b8f | 440 | { |
193aec10 CH |
441 | struct xfs_mount *mp = ip->i_mount; |
442 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); | |
443 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); | |
444 | int zero_len; | |
445 | int nimaps = 1; | |
446 | int error = 0; | |
447 | struct xfs_bmbt_irec imap; | |
dda35b8f | 448 | |
193aec10 | 449 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 450 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
193aec10 | 451 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
5c8ed202 | 452 | if (error) |
dda35b8f | 453 | return error; |
193aec10 | 454 | |
dda35b8f | 455 | ASSERT(nimaps > 0); |
193aec10 | 456 | |
dda35b8f CH |
457 | /* |
458 | * If the block underlying isize is just a hole, then there | |
459 | * is nothing to zero. | |
460 | */ | |
193aec10 | 461 | if (imap.br_startblock == HOLESTARTBLOCK) |
dda35b8f | 462 | return 0; |
dda35b8f CH |
463 | |
464 | zero_len = mp->m_sb.sb_blocksize - zero_offset; | |
465 | if (isize + zero_len > offset) | |
466 | zero_len = offset - isize; | |
5885ebda | 467 | *did_zeroing = true; |
193aec10 | 468 | return xfs_iozero(ip, isize, zero_len); |
dda35b8f CH |
469 | } |
470 | ||
471 | /* | |
193aec10 CH |
472 | * Zero any on disk space between the current EOF and the new, larger EOF. |
473 | * | |
474 | * This handles the normal case of zeroing the remainder of the last block in | |
475 | * the file and the unusual case of zeroing blocks out beyond the size of the | |
476 | * file. This second case only happens with fixed size extents and when the | |
477 | * system crashes before the inode size was updated but after blocks were | |
478 | * allocated. | |
479 | * | |
480 | * Expects the iolock to be held exclusive, and will take the ilock internally. | |
dda35b8f | 481 | */ |
dda35b8f CH |
482 | int /* error (positive) */ |
483 | xfs_zero_eof( | |
193aec10 CH |
484 | struct xfs_inode *ip, |
485 | xfs_off_t offset, /* starting I/O offset */ | |
5885ebda DC |
486 | xfs_fsize_t isize, /* current inode size */ |
487 | bool *did_zeroing) | |
dda35b8f | 488 | { |
193aec10 CH |
489 | struct xfs_mount *mp = ip->i_mount; |
490 | xfs_fileoff_t start_zero_fsb; | |
491 | xfs_fileoff_t end_zero_fsb; | |
492 | xfs_fileoff_t zero_count_fsb; | |
493 | xfs_fileoff_t last_fsb; | |
494 | xfs_fileoff_t zero_off; | |
495 | xfs_fsize_t zero_len; | |
496 | int nimaps; | |
497 | int error = 0; | |
498 | struct xfs_bmbt_irec imap; | |
499 | ||
500 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
dda35b8f CH |
501 | ASSERT(offset > isize); |
502 | ||
0a50f162 BF |
503 | trace_xfs_zero_eof(ip, isize, offset - isize); |
504 | ||
dda35b8f CH |
505 | /* |
506 | * First handle zeroing the block on which isize resides. | |
193aec10 | 507 | * |
dda35b8f CH |
508 | * We only zero a part of that block so it is handled specially. |
509 | */ | |
193aec10 | 510 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
5885ebda | 511 | error = xfs_zero_last_block(ip, offset, isize, did_zeroing); |
193aec10 CH |
512 | if (error) |
513 | return error; | |
dda35b8f CH |
514 | } |
515 | ||
516 | /* | |
193aec10 CH |
517 | * Calculate the range between the new size and the old where blocks |
518 | * needing to be zeroed may exist. | |
519 | * | |
520 | * To get the block where the last byte in the file currently resides, | |
521 | * we need to subtract one from the size and truncate back to a block | |
522 | * boundary. We subtract 1 in case the size is exactly on a block | |
523 | * boundary. | |
dda35b8f CH |
524 | */ |
525 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; | |
526 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); | |
527 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); | |
528 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); | |
529 | if (last_fsb == end_zero_fsb) { | |
530 | /* | |
531 | * The size was only incremented on its last block. | |
532 | * We took care of that above, so just return. | |
533 | */ | |
534 | return 0; | |
535 | } | |
536 | ||
537 | ASSERT(start_zero_fsb <= end_zero_fsb); | |
538 | while (start_zero_fsb <= end_zero_fsb) { | |
539 | nimaps = 1; | |
540 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; | |
193aec10 CH |
541 | |
542 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
5c8ed202 DC |
543 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
544 | &imap, &nimaps, 0); | |
193aec10 CH |
545 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
546 | if (error) | |
dda35b8f | 547 | return error; |
193aec10 | 548 | |
dda35b8f CH |
549 | ASSERT(nimaps > 0); |
550 | ||
551 | if (imap.br_state == XFS_EXT_UNWRITTEN || | |
552 | imap.br_startblock == HOLESTARTBLOCK) { | |
dda35b8f CH |
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. | |
dda35b8f | 560 | */ |
dda35b8f CH |
561 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
562 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); | |
563 | ||
564 | if ((zero_off + zero_len) > offset) | |
565 | zero_len = offset - zero_off; | |
566 | ||
567 | error = xfs_iozero(ip, zero_off, zero_len); | |
193aec10 CH |
568 | if (error) |
569 | return error; | |
dda35b8f | 570 | |
5885ebda | 571 | *did_zeroing = true; |
dda35b8f CH |
572 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
573 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); | |
dda35b8f CH |
574 | } |
575 | ||
576 | return 0; | |
dda35b8f CH |
577 | } |
578 | ||
4d8d1581 DC |
579 | /* |
580 | * Common pre-write limit and setup checks. | |
581 | * | |
5bf1f262 CH |
582 | * Called with the iolocked held either shared and exclusive according to |
583 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive | |
584 | * if called for a direct write beyond i_size. | |
4d8d1581 DC |
585 | */ |
586 | STATIC ssize_t | |
587 | xfs_file_aio_write_checks( | |
99733fa3 AV |
588 | struct kiocb *iocb, |
589 | struct iov_iter *from, | |
4d8d1581 DC |
590 | int *iolock) |
591 | { | |
99733fa3 | 592 | struct file *file = iocb->ki_filp; |
4d8d1581 DC |
593 | struct inode *inode = file->f_mapping->host; |
594 | struct xfs_inode *ip = XFS_I(inode); | |
3309dd04 | 595 | ssize_t error = 0; |
99733fa3 | 596 | size_t count = iov_iter_count(from); |
3136e8bb | 597 | bool drained_dio = false; |
4d8d1581 | 598 | |
7271d243 | 599 | restart: |
3309dd04 AV |
600 | error = generic_write_checks(iocb, from); |
601 | if (error <= 0) | |
4d8d1581 | 602 | return error; |
4d8d1581 | 603 | |
21c3ea18 | 604 | error = xfs_break_layouts(inode, iolock, true); |
781355c6 CH |
605 | if (error) |
606 | return error; | |
607 | ||
a6de82ca JK |
608 | /* For changing security info in file_remove_privs() we need i_mutex */ |
609 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { | |
610 | xfs_rw_iunlock(ip, *iolock); | |
611 | *iolock = XFS_IOLOCK_EXCL; | |
612 | xfs_rw_ilock(ip, *iolock); | |
613 | goto restart; | |
614 | } | |
4d8d1581 DC |
615 | /* |
616 | * If the offset is beyond the size of the file, we need to zero any | |
617 | * blocks that fall between the existing EOF and the start of this | |
2813d682 | 618 | * write. If zeroing is needed and we are currently holding the |
467f7899 CH |
619 | * iolock shared, we need to update it to exclusive which implies |
620 | * having to redo all checks before. | |
b9d59846 DC |
621 | * |
622 | * We need to serialise against EOF updates that occur in IO | |
623 | * completions here. We want to make sure that nobody is changing the | |
624 | * size while we do this check until we have placed an IO barrier (i.e. | |
625 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. | |
626 | * The spinlock effectively forms a memory barrier once we have the | |
627 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value | |
628 | * and hence be able to correctly determine if we need to run zeroing. | |
4d8d1581 | 629 | */ |
b9d59846 | 630 | spin_lock(&ip->i_flags_lock); |
99733fa3 | 631 | if (iocb->ki_pos > i_size_read(inode)) { |
5885ebda DC |
632 | bool zero = false; |
633 | ||
b9d59846 | 634 | spin_unlock(&ip->i_flags_lock); |
3136e8bb BF |
635 | if (!drained_dio) { |
636 | if (*iolock == XFS_IOLOCK_SHARED) { | |
637 | xfs_rw_iunlock(ip, *iolock); | |
638 | *iolock = XFS_IOLOCK_EXCL; | |
639 | xfs_rw_ilock(ip, *iolock); | |
640 | iov_iter_reexpand(from, count); | |
641 | } | |
40c63fbc DC |
642 | /* |
643 | * We now have an IO submission barrier in place, but | |
644 | * AIO can do EOF updates during IO completion and hence | |
645 | * we now need to wait for all of them to drain. Non-AIO | |
646 | * DIO will have drained before we are given the | |
647 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a | |
648 | * no-op. | |
649 | */ | |
650 | inode_dio_wait(inode); | |
3136e8bb | 651 | drained_dio = true; |
7271d243 DC |
652 | goto restart; |
653 | } | |
99733fa3 | 654 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
467f7899 CH |
655 | if (error) |
656 | return error; | |
b9d59846 DC |
657 | } else |
658 | spin_unlock(&ip->i_flags_lock); | |
4d8d1581 | 659 | |
8a9c9980 CH |
660 | /* |
661 | * Updating the timestamps will grab the ilock again from | |
662 | * xfs_fs_dirty_inode, so we have to call it after dropping the | |
663 | * lock above. Eventually we should look into a way to avoid | |
664 | * the pointless lock roundtrip. | |
665 | */ | |
c3b2da31 JB |
666 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
667 | error = file_update_time(file); | |
668 | if (error) | |
669 | return error; | |
670 | } | |
8a9c9980 | 671 | |
4d8d1581 DC |
672 | /* |
673 | * If we're writing the file then make sure to clear the setuid and | |
674 | * setgid bits if the process is not being run by root. This keeps | |
675 | * people from modifying setuid and setgid binaries. | |
676 | */ | |
a6de82ca JK |
677 | if (!IS_NOSEC(inode)) |
678 | return file_remove_privs(file); | |
679 | return 0; | |
4d8d1581 DC |
680 | } |
681 | ||
f0d26e86 DC |
682 | /* |
683 | * xfs_file_dio_aio_write - handle direct IO writes | |
684 | * | |
685 | * Lock the inode appropriately to prepare for and issue a direct IO write. | |
eda77982 | 686 | * By separating it from the buffered write path we remove all the tricky to |
f0d26e86 DC |
687 | * follow locking changes and looping. |
688 | * | |
eda77982 DC |
689 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
690 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached | |
691 | * pages are flushed out. | |
692 | * | |
693 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED | |
694 | * allowing them to be done in parallel with reads and other direct IO writes. | |
695 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer | |
696 | * needs to do sub-block zeroing and that requires serialisation against other | |
697 | * direct IOs to the same block. In this case we need to serialise the | |
698 | * submission of the unaligned IOs so that we don't get racing block zeroing in | |
699 | * the dio layer. To avoid the problem with aio, we also need to wait for | |
700 | * outstanding IOs to complete so that unwritten extent conversion is completed | |
701 | * before we try to map the overlapping block. This is currently implemented by | |
4a06fd26 | 702 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
eda77982 | 703 | * |
f0d26e86 DC |
704 | * Returns with locks held indicated by @iolock and errors indicated by |
705 | * negative return values. | |
706 | */ | |
707 | STATIC ssize_t | |
708 | xfs_file_dio_aio_write( | |
709 | struct kiocb *iocb, | |
b3188919 | 710 | struct iov_iter *from) |
f0d26e86 DC |
711 | { |
712 | struct file *file = iocb->ki_filp; | |
713 | struct address_space *mapping = file->f_mapping; | |
714 | struct inode *inode = mapping->host; | |
715 | struct xfs_inode *ip = XFS_I(inode); | |
716 | struct xfs_mount *mp = ip->i_mount; | |
717 | ssize_t ret = 0; | |
eda77982 | 718 | int unaligned_io = 0; |
d0606464 | 719 | int iolock; |
b3188919 AV |
720 | size_t count = iov_iter_count(from); |
721 | loff_t pos = iocb->ki_pos; | |
0cefb29e DC |
722 | loff_t end; |
723 | struct iov_iter data; | |
f0d26e86 DC |
724 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
725 | mp->m_rtdev_targp : mp->m_ddev_targp; | |
726 | ||
7c71ee78 | 727 | /* DIO must be aligned to device logical sector size */ |
6b698ede | 728 | if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask)) |
b474c7ae | 729 | return -EINVAL; |
f0d26e86 | 730 | |
7c71ee78 | 731 | /* "unaligned" here means not aligned to a filesystem block */ |
eda77982 DC |
732 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
733 | unaligned_io = 1; | |
734 | ||
7271d243 DC |
735 | /* |
736 | * We don't need to take an exclusive lock unless there page cache needs | |
737 | * to be invalidated or unaligned IO is being executed. We don't need to | |
738 | * consider the EOF extension case here because | |
739 | * xfs_file_aio_write_checks() will relock the inode as necessary for | |
740 | * EOF zeroing cases and fill out the new inode size as appropriate. | |
741 | */ | |
742 | if (unaligned_io || mapping->nrpages) | |
d0606464 | 743 | iolock = XFS_IOLOCK_EXCL; |
f0d26e86 | 744 | else |
d0606464 CH |
745 | iolock = XFS_IOLOCK_SHARED; |
746 | xfs_rw_ilock(ip, iolock); | |
c58cb165 CH |
747 | |
748 | /* | |
749 | * Recheck if there are cached pages that need invalidate after we got | |
750 | * the iolock to protect against other threads adding new pages while | |
751 | * we were waiting for the iolock. | |
752 | */ | |
d0606464 CH |
753 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
754 | xfs_rw_iunlock(ip, iolock); | |
755 | iolock = XFS_IOLOCK_EXCL; | |
756 | xfs_rw_ilock(ip, iolock); | |
c58cb165 | 757 | } |
f0d26e86 | 758 | |
99733fa3 | 759 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 760 | if (ret) |
d0606464 | 761 | goto out; |
99733fa3 AV |
762 | count = iov_iter_count(from); |
763 | pos = iocb->ki_pos; | |
0cefb29e | 764 | end = pos + count - 1; |
f0d26e86 | 765 | |
3d751af2 BF |
766 | /* |
767 | * See xfs_file_read_iter() for why we do a full-file flush here. | |
768 | */ | |
f0d26e86 | 769 | if (mapping->nrpages) { |
3d751af2 | 770 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
f0d26e86 | 771 | if (ret) |
d0606464 | 772 | goto out; |
834ffca6 | 773 | /* |
3d751af2 BF |
774 | * Invalidate whole pages. This can return an error if we fail |
775 | * to invalidate a page, but this should never happen on XFS. | |
776 | * Warn if it does fail. | |
834ffca6 | 777 | */ |
3d751af2 | 778 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
834ffca6 DC |
779 | WARN_ON_ONCE(ret); |
780 | ret = 0; | |
f0d26e86 DC |
781 | } |
782 | ||
eda77982 DC |
783 | /* |
784 | * If we are doing unaligned IO, wait for all other IO to drain, | |
785 | * otherwise demote the lock if we had to flush cached pages | |
786 | */ | |
787 | if (unaligned_io) | |
4a06fd26 | 788 | inode_dio_wait(inode); |
d0606464 | 789 | else if (iolock == XFS_IOLOCK_EXCL) { |
f0d26e86 | 790 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
d0606464 | 791 | iolock = XFS_IOLOCK_SHARED; |
f0d26e86 DC |
792 | } |
793 | ||
794 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); | |
f0d26e86 | 795 | |
0cefb29e | 796 | data = *from; |
1aef882f | 797 | ret = mapping->a_ops->direct_IO(iocb, &data, pos); |
0cefb29e DC |
798 | |
799 | /* see generic_file_direct_write() for why this is necessary */ | |
800 | if (mapping->nrpages) { | |
801 | invalidate_inode_pages2_range(mapping, | |
09cbfeaf KS |
802 | pos >> PAGE_SHIFT, |
803 | end >> PAGE_SHIFT); | |
0cefb29e DC |
804 | } |
805 | ||
806 | if (ret > 0) { | |
807 | pos += ret; | |
808 | iov_iter_advance(from, ret); | |
809 | iocb->ki_pos = pos; | |
810 | } | |
d0606464 CH |
811 | out: |
812 | xfs_rw_iunlock(ip, iolock); | |
813 | ||
6b698ede DC |
814 | /* |
815 | * No fallback to buffered IO on errors for XFS. DAX can result in | |
816 | * partial writes, but direct IO will either complete fully or fail. | |
817 | */ | |
818 | ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip))); | |
f0d26e86 DC |
819 | return ret; |
820 | } | |
821 | ||
00258e36 | 822 | STATIC ssize_t |
637bbc75 | 823 | xfs_file_buffered_aio_write( |
dda35b8f | 824 | struct kiocb *iocb, |
b3188919 | 825 | struct iov_iter *from) |
dda35b8f CH |
826 | { |
827 | struct file *file = iocb->ki_filp; | |
828 | struct address_space *mapping = file->f_mapping; | |
829 | struct inode *inode = mapping->host; | |
00258e36 | 830 | struct xfs_inode *ip = XFS_I(inode); |
637bbc75 DC |
831 | ssize_t ret; |
832 | int enospc = 0; | |
d0606464 | 833 | int iolock = XFS_IOLOCK_EXCL; |
dda35b8f | 834 | |
d0606464 | 835 | xfs_rw_ilock(ip, iolock); |
dda35b8f | 836 | |
99733fa3 | 837 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
4d8d1581 | 838 | if (ret) |
d0606464 | 839 | goto out; |
dda35b8f CH |
840 | |
841 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 842 | current->backing_dev_info = inode_to_bdi(inode); |
dda35b8f | 843 | |
dda35b8f | 844 | write_retry: |
99733fa3 AV |
845 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), |
846 | iocb->ki_pos, 0); | |
847 | ret = generic_perform_write(file, from, iocb->ki_pos); | |
0a64bc2c | 848 | if (likely(ret >= 0)) |
99733fa3 | 849 | iocb->ki_pos += ret; |
dc06f398 | 850 | |
637bbc75 | 851 | /* |
dc06f398 BF |
852 | * If we hit a space limit, try to free up some lingering preallocated |
853 | * space before returning an error. In the case of ENOSPC, first try to | |
854 | * write back all dirty inodes to free up some of the excess reserved | |
855 | * metadata space. This reduces the chances that the eofblocks scan | |
856 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this | |
857 | * also behaves as a filter to prevent too many eofblocks scans from | |
858 | * running at the same time. | |
637bbc75 | 859 | */ |
dc06f398 BF |
860 | if (ret == -EDQUOT && !enospc) { |
861 | enospc = xfs_inode_free_quota_eofblocks(ip); | |
862 | if (enospc) | |
863 | goto write_retry; | |
864 | } else if (ret == -ENOSPC && !enospc) { | |
865 | struct xfs_eofblocks eofb = {0}; | |
866 | ||
637bbc75 | 867 | enospc = 1; |
9aa05000 | 868 | xfs_flush_inodes(ip->i_mount); |
dc06f398 BF |
869 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
870 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; | |
871 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); | |
9aa05000 | 872 | goto write_retry; |
dda35b8f | 873 | } |
d0606464 | 874 | |
dda35b8f | 875 | current->backing_dev_info = NULL; |
d0606464 CH |
876 | out: |
877 | xfs_rw_iunlock(ip, iolock); | |
637bbc75 DC |
878 | return ret; |
879 | } | |
880 | ||
881 | STATIC ssize_t | |
bf97f3bc | 882 | xfs_file_write_iter( |
637bbc75 | 883 | struct kiocb *iocb, |
bf97f3bc | 884 | struct iov_iter *from) |
637bbc75 DC |
885 | { |
886 | struct file *file = iocb->ki_filp; | |
887 | struct address_space *mapping = file->f_mapping; | |
888 | struct inode *inode = mapping->host; | |
889 | struct xfs_inode *ip = XFS_I(inode); | |
890 | ssize_t ret; | |
bf97f3bc | 891 | size_t ocount = iov_iter_count(from); |
637bbc75 | 892 | |
ff6d6af2 | 893 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
637bbc75 | 894 | |
637bbc75 DC |
895 | if (ocount == 0) |
896 | return 0; | |
897 | ||
bf97f3bc AV |
898 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
899 | return -EIO; | |
637bbc75 | 900 | |
6b698ede | 901 | if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode)) |
bf97f3bc | 902 | ret = xfs_file_dio_aio_write(iocb, from); |
637bbc75 | 903 | else |
bf97f3bc | 904 | ret = xfs_file_buffered_aio_write(iocb, from); |
dda35b8f | 905 | |
d0606464 CH |
906 | if (ret > 0) { |
907 | ssize_t err; | |
dda35b8f | 908 | |
ff6d6af2 | 909 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
dda35b8f | 910 | |
d0606464 | 911 | /* Handle various SYNC-type writes */ |
d311d79d | 912 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
d0606464 CH |
913 | if (err < 0) |
914 | ret = err; | |
dda35b8f | 915 | } |
a363f0c2 | 916 | return ret; |
dda35b8f CH |
917 | } |
918 | ||
a904b1ca NJ |
919 | #define XFS_FALLOC_FL_SUPPORTED \ |
920 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
921 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ | |
922 | FALLOC_FL_INSERT_RANGE) | |
923 | ||
2fe17c10 CH |
924 | STATIC long |
925 | xfs_file_fallocate( | |
83aee9e4 CH |
926 | struct file *file, |
927 | int mode, | |
928 | loff_t offset, | |
929 | loff_t len) | |
2fe17c10 | 930 | { |
83aee9e4 CH |
931 | struct inode *inode = file_inode(file); |
932 | struct xfs_inode *ip = XFS_I(inode); | |
83aee9e4 | 933 | long error; |
8add71ca | 934 | enum xfs_prealloc_flags flags = 0; |
781355c6 | 935 | uint iolock = XFS_IOLOCK_EXCL; |
83aee9e4 | 936 | loff_t new_size = 0; |
a904b1ca | 937 | bool do_file_insert = 0; |
2fe17c10 | 938 | |
83aee9e4 CH |
939 | if (!S_ISREG(inode->i_mode)) |
940 | return -EINVAL; | |
a904b1ca | 941 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
2fe17c10 CH |
942 | return -EOPNOTSUPP; |
943 | ||
781355c6 | 944 | xfs_ilock(ip, iolock); |
21c3ea18 | 945 | error = xfs_break_layouts(inode, &iolock, false); |
781355c6 CH |
946 | if (error) |
947 | goto out_unlock; | |
948 | ||
e8e9ad42 DC |
949 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
950 | iolock |= XFS_MMAPLOCK_EXCL; | |
951 | ||
83aee9e4 CH |
952 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
953 | error = xfs_free_file_space(ip, offset, len); | |
954 | if (error) | |
955 | goto out_unlock; | |
e1d8fb88 NJ |
956 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
957 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
958 | ||
959 | if (offset & blksize_mask || len & blksize_mask) { | |
2451337d | 960 | error = -EINVAL; |
e1d8fb88 NJ |
961 | goto out_unlock; |
962 | } | |
963 | ||
23fffa92 LC |
964 | /* |
965 | * There is no need to overlap collapse range with EOF, | |
966 | * in which case it is effectively a truncate operation | |
967 | */ | |
968 | if (offset + len >= i_size_read(inode)) { | |
2451337d | 969 | error = -EINVAL; |
23fffa92 LC |
970 | goto out_unlock; |
971 | } | |
972 | ||
e1d8fb88 NJ |
973 | new_size = i_size_read(inode) - len; |
974 | ||
975 | error = xfs_collapse_file_space(ip, offset, len); | |
976 | if (error) | |
977 | goto out_unlock; | |
a904b1ca NJ |
978 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
979 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; | |
980 | ||
981 | new_size = i_size_read(inode) + len; | |
982 | if (offset & blksize_mask || len & blksize_mask) { | |
983 | error = -EINVAL; | |
984 | goto out_unlock; | |
985 | } | |
986 | ||
987 | /* check the new inode size does not wrap through zero */ | |
988 | if (new_size > inode->i_sb->s_maxbytes) { | |
989 | error = -EFBIG; | |
990 | goto out_unlock; | |
991 | } | |
992 | ||
993 | /* Offset should be less than i_size */ | |
994 | if (offset >= i_size_read(inode)) { | |
995 | error = -EINVAL; | |
996 | goto out_unlock; | |
997 | } | |
998 | do_file_insert = 1; | |
83aee9e4 | 999 | } else { |
8add71ca CH |
1000 | flags |= XFS_PREALLOC_SET; |
1001 | ||
83aee9e4 CH |
1002 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
1003 | offset + len > i_size_read(inode)) { | |
1004 | new_size = offset + len; | |
2451337d | 1005 | error = inode_newsize_ok(inode, new_size); |
83aee9e4 CH |
1006 | if (error) |
1007 | goto out_unlock; | |
1008 | } | |
2fe17c10 | 1009 | |
376ba313 LC |
1010 | if (mode & FALLOC_FL_ZERO_RANGE) |
1011 | error = xfs_zero_file_space(ip, offset, len); | |
1012 | else | |
1013 | error = xfs_alloc_file_space(ip, offset, len, | |
1014 | XFS_BMAPI_PREALLOC); | |
2fe17c10 CH |
1015 | if (error) |
1016 | goto out_unlock; | |
1017 | } | |
1018 | ||
83aee9e4 | 1019 | if (file->f_flags & O_DSYNC) |
8add71ca CH |
1020 | flags |= XFS_PREALLOC_SYNC; |
1021 | ||
1022 | error = xfs_update_prealloc_flags(ip, flags); | |
2fe17c10 CH |
1023 | if (error) |
1024 | goto out_unlock; | |
1025 | ||
1026 | /* Change file size if needed */ | |
1027 | if (new_size) { | |
1028 | struct iattr iattr; | |
1029 | ||
1030 | iattr.ia_valid = ATTR_SIZE; | |
1031 | iattr.ia_size = new_size; | |
83aee9e4 | 1032 | error = xfs_setattr_size(ip, &iattr); |
a904b1ca NJ |
1033 | if (error) |
1034 | goto out_unlock; | |
2fe17c10 CH |
1035 | } |
1036 | ||
a904b1ca NJ |
1037 | /* |
1038 | * Perform hole insertion now that the file size has been | |
1039 | * updated so that if we crash during the operation we don't | |
1040 | * leave shifted extents past EOF and hence losing access to | |
1041 | * the data that is contained within them. | |
1042 | */ | |
1043 | if (do_file_insert) | |
1044 | error = xfs_insert_file_space(ip, offset, len); | |
1045 | ||
2fe17c10 | 1046 | out_unlock: |
781355c6 | 1047 | xfs_iunlock(ip, iolock); |
2451337d | 1048 | return error; |
2fe17c10 CH |
1049 | } |
1050 | ||
1051 | ||
1da177e4 | 1052 | STATIC int |
3562fd45 | 1053 | xfs_file_open( |
1da177e4 | 1054 | struct inode *inode, |
f999a5bf | 1055 | struct file *file) |
1da177e4 | 1056 | { |
f999a5bf | 1057 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
1da177e4 | 1058 | return -EFBIG; |
f999a5bf CH |
1059 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
1060 | return -EIO; | |
1061 | return 0; | |
1062 | } | |
1063 | ||
1064 | STATIC int | |
1065 | xfs_dir_open( | |
1066 | struct inode *inode, | |
1067 | struct file *file) | |
1068 | { | |
1069 | struct xfs_inode *ip = XFS_I(inode); | |
1070 | int mode; | |
1071 | int error; | |
1072 | ||
1073 | error = xfs_file_open(inode, file); | |
1074 | if (error) | |
1075 | return error; | |
1076 | ||
1077 | /* | |
1078 | * If there are any blocks, read-ahead block 0 as we're almost | |
1079 | * certain to have the next operation be a read there. | |
1080 | */ | |
309ecac8 | 1081 | mode = xfs_ilock_data_map_shared(ip); |
f999a5bf | 1082 | if (ip->i_d.di_nextents > 0) |
9df2dd0b | 1083 | xfs_dir3_data_readahead(ip, 0, -1); |
f999a5bf CH |
1084 | xfs_iunlock(ip, mode); |
1085 | return 0; | |
1da177e4 LT |
1086 | } |
1087 | ||
1da177e4 | 1088 | STATIC int |
3562fd45 | 1089 | xfs_file_release( |
1da177e4 LT |
1090 | struct inode *inode, |
1091 | struct file *filp) | |
1092 | { | |
2451337d | 1093 | return xfs_release(XFS_I(inode)); |
1da177e4 LT |
1094 | } |
1095 | ||
1da177e4 | 1096 | STATIC int |
3562fd45 | 1097 | xfs_file_readdir( |
b8227554 AV |
1098 | struct file *file, |
1099 | struct dir_context *ctx) | |
1da177e4 | 1100 | { |
b8227554 | 1101 | struct inode *inode = file_inode(file); |
739bfb2a | 1102 | xfs_inode_t *ip = XFS_I(inode); |
051e7cd4 CH |
1103 | size_t bufsize; |
1104 | ||
1105 | /* | |
1106 | * The Linux API doesn't pass down the total size of the buffer | |
1107 | * we read into down to the filesystem. With the filldir concept | |
1108 | * it's not needed for correct information, but the XFS dir2 leaf | |
1109 | * code wants an estimate of the buffer size to calculate it's | |
1110 | * readahead window and size the buffers used for mapping to | |
1111 | * physical blocks. | |
1112 | * | |
1113 | * Try to give it an estimate that's good enough, maybe at some | |
1114 | * point we can change the ->readdir prototype to include the | |
a9cc799e | 1115 | * buffer size. For now we use the current glibc buffer size. |
051e7cd4 | 1116 | */ |
a9cc799e | 1117 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
051e7cd4 | 1118 | |
8300475e | 1119 | return xfs_readdir(ip, ctx, bufsize); |
1da177e4 LT |
1120 | } |
1121 | ||
d126d43f JL |
1122 | /* |
1123 | * This type is designed to indicate the type of offset we would like | |
49c69591 | 1124 | * to search from page cache for xfs_seek_hole_data(). |
d126d43f JL |
1125 | */ |
1126 | enum { | |
1127 | HOLE_OFF = 0, | |
1128 | DATA_OFF, | |
1129 | }; | |
1130 | ||
1131 | /* | |
1132 | * Lookup the desired type of offset from the given page. | |
1133 | * | |
1134 | * On success, return true and the offset argument will point to the | |
1135 | * start of the region that was found. Otherwise this function will | |
1136 | * return false and keep the offset argument unchanged. | |
1137 | */ | |
1138 | STATIC bool | |
1139 | xfs_lookup_buffer_offset( | |
1140 | struct page *page, | |
1141 | loff_t *offset, | |
1142 | unsigned int type) | |
1143 | { | |
1144 | loff_t lastoff = page_offset(page); | |
1145 | bool found = false; | |
1146 | struct buffer_head *bh, *head; | |
1147 | ||
1148 | bh = head = page_buffers(page); | |
1149 | do { | |
1150 | /* | |
1151 | * Unwritten extents that have data in the page | |
1152 | * cache covering them can be identified by the | |
1153 | * BH_Unwritten state flag. Pages with multiple | |
1154 | * buffers might have a mix of holes, data and | |
1155 | * unwritten extents - any buffer with valid | |
1156 | * data in it should have BH_Uptodate flag set | |
1157 | * on it. | |
1158 | */ | |
1159 | if (buffer_unwritten(bh) || | |
1160 | buffer_uptodate(bh)) { | |
1161 | if (type == DATA_OFF) | |
1162 | found = true; | |
1163 | } else { | |
1164 | if (type == HOLE_OFF) | |
1165 | found = true; | |
1166 | } | |
1167 | ||
1168 | if (found) { | |
1169 | *offset = lastoff; | |
1170 | break; | |
1171 | } | |
1172 | lastoff += bh->b_size; | |
1173 | } while ((bh = bh->b_this_page) != head); | |
1174 | ||
1175 | return found; | |
1176 | } | |
1177 | ||
1178 | /* | |
1179 | * This routine is called to find out and return a data or hole offset | |
1180 | * from the page cache for unwritten extents according to the desired | |
49c69591 | 1181 | * type for xfs_seek_hole_data(). |
d126d43f JL |
1182 | * |
1183 | * The argument offset is used to tell where we start to search from the | |
1184 | * page cache. Map is used to figure out the end points of the range to | |
1185 | * lookup pages. | |
1186 | * | |
1187 | * Return true if the desired type of offset was found, and the argument | |
1188 | * offset is filled with that address. Otherwise, return false and keep | |
1189 | * offset unchanged. | |
1190 | */ | |
1191 | STATIC bool | |
1192 | xfs_find_get_desired_pgoff( | |
1193 | struct inode *inode, | |
1194 | struct xfs_bmbt_irec *map, | |
1195 | unsigned int type, | |
1196 | loff_t *offset) | |
1197 | { | |
1198 | struct xfs_inode *ip = XFS_I(inode); | |
1199 | struct xfs_mount *mp = ip->i_mount; | |
1200 | struct pagevec pvec; | |
1201 | pgoff_t index; | |
1202 | pgoff_t end; | |
1203 | loff_t endoff; | |
1204 | loff_t startoff = *offset; | |
1205 | loff_t lastoff = startoff; | |
1206 | bool found = false; | |
1207 | ||
1208 | pagevec_init(&pvec, 0); | |
1209 | ||
09cbfeaf | 1210 | index = startoff >> PAGE_SHIFT; |
d126d43f | 1211 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
09cbfeaf | 1212 | end = endoff >> PAGE_SHIFT; |
d126d43f JL |
1213 | do { |
1214 | int want; | |
1215 | unsigned nr_pages; | |
1216 | unsigned int i; | |
1217 | ||
1218 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); | |
1219 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, | |
1220 | want); | |
1221 | /* | |
1222 | * No page mapped into given range. If we are searching holes | |
1223 | * and if this is the first time we got into the loop, it means | |
1224 | * that the given offset is landed in a hole, return it. | |
1225 | * | |
1226 | * If we have already stepped through some block buffers to find | |
1227 | * holes but they all contains data. In this case, the last | |
1228 | * offset is already updated and pointed to the end of the last | |
1229 | * mapped page, if it does not reach the endpoint to search, | |
1230 | * that means there should be a hole between them. | |
1231 | */ | |
1232 | if (nr_pages == 0) { | |
1233 | /* Data search found nothing */ | |
1234 | if (type == DATA_OFF) | |
1235 | break; | |
1236 | ||
1237 | ASSERT(type == HOLE_OFF); | |
1238 | if (lastoff == startoff || lastoff < endoff) { | |
1239 | found = true; | |
1240 | *offset = lastoff; | |
1241 | } | |
1242 | break; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * At lease we found one page. If this is the first time we | |
1247 | * step into the loop, and if the first page index offset is | |
1248 | * greater than the given search offset, a hole was found. | |
1249 | */ | |
1250 | if (type == HOLE_OFF && lastoff == startoff && | |
1251 | lastoff < page_offset(pvec.pages[0])) { | |
1252 | found = true; | |
1253 | break; | |
1254 | } | |
1255 | ||
1256 | for (i = 0; i < nr_pages; i++) { | |
1257 | struct page *page = pvec.pages[i]; | |
1258 | loff_t b_offset; | |
1259 | ||
1260 | /* | |
1261 | * At this point, the page may be truncated or | |
1262 | * invalidated (changing page->mapping to NULL), | |
1263 | * or even swizzled back from swapper_space to tmpfs | |
1264 | * file mapping. However, page->index will not change | |
1265 | * because we have a reference on the page. | |
1266 | * | |
1267 | * Searching done if the page index is out of range. | |
1268 | * If the current offset is not reaches the end of | |
1269 | * the specified search range, there should be a hole | |
1270 | * between them. | |
1271 | */ | |
1272 | if (page->index > end) { | |
1273 | if (type == HOLE_OFF && lastoff < endoff) { | |
1274 | *offset = lastoff; | |
1275 | found = true; | |
1276 | } | |
1277 | goto out; | |
1278 | } | |
1279 | ||
1280 | lock_page(page); | |
1281 | /* | |
1282 | * Page truncated or invalidated(page->mapping == NULL). | |
1283 | * We can freely skip it and proceed to check the next | |
1284 | * page. | |
1285 | */ | |
1286 | if (unlikely(page->mapping != inode->i_mapping)) { | |
1287 | unlock_page(page); | |
1288 | continue; | |
1289 | } | |
1290 | ||
1291 | if (!page_has_buffers(page)) { | |
1292 | unlock_page(page); | |
1293 | continue; | |
1294 | } | |
1295 | ||
1296 | found = xfs_lookup_buffer_offset(page, &b_offset, type); | |
1297 | if (found) { | |
1298 | /* | |
1299 | * The found offset may be less than the start | |
1300 | * point to search if this is the first time to | |
1301 | * come here. | |
1302 | */ | |
1303 | *offset = max_t(loff_t, startoff, b_offset); | |
1304 | unlock_page(page); | |
1305 | goto out; | |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * We either searching data but nothing was found, or | |
1310 | * searching hole but found a data buffer. In either | |
1311 | * case, probably the next page contains the desired | |
1312 | * things, update the last offset to it so. | |
1313 | */ | |
1314 | lastoff = page_offset(page) + PAGE_SIZE; | |
1315 | unlock_page(page); | |
1316 | } | |
1317 | ||
1318 | /* | |
1319 | * The number of returned pages less than our desired, search | |
1320 | * done. In this case, nothing was found for searching data, | |
1321 | * but we found a hole behind the last offset. | |
1322 | */ | |
1323 | if (nr_pages < want) { | |
1324 | if (type == HOLE_OFF) { | |
1325 | *offset = lastoff; | |
1326 | found = true; | |
1327 | } | |
1328 | break; | |
1329 | } | |
1330 | ||
1331 | index = pvec.pages[i - 1]->index + 1; | |
1332 | pagevec_release(&pvec); | |
1333 | } while (index <= end); | |
1334 | ||
1335 | out: | |
1336 | pagevec_release(&pvec); | |
1337 | return found; | |
1338 | } | |
1339 | ||
8aa7d37e ES |
1340 | /* |
1341 | * caller must lock inode with xfs_ilock_data_map_shared, | |
1342 | * can we craft an appropriate ASSERT? | |
1343 | * | |
1344 | * end is because the VFS-level lseek interface is defined such that any | |
1345 | * offset past i_size shall return -ENXIO, but we use this for quota code | |
1346 | * which does not maintain i_size, and we want to SEEK_DATA past i_size. | |
1347 | */ | |
1348 | loff_t | |
1349 | __xfs_seek_hole_data( | |
1350 | struct inode *inode, | |
49c69591 | 1351 | loff_t start, |
8aa7d37e | 1352 | loff_t end, |
49c69591 | 1353 | int whence) |
3fe3e6b1 | 1354 | { |
3fe3e6b1 JL |
1355 | struct xfs_inode *ip = XFS_I(inode); |
1356 | struct xfs_mount *mp = ip->i_mount; | |
3fe3e6b1 | 1357 | loff_t uninitialized_var(offset); |
3fe3e6b1 | 1358 | xfs_fileoff_t fsbno; |
8aa7d37e | 1359 | xfs_filblks_t lastbno; |
3fe3e6b1 JL |
1360 | int error; |
1361 | ||
8aa7d37e | 1362 | if (start >= end) { |
2451337d | 1363 | error = -ENXIO; |
8aa7d37e | 1364 | goto out_error; |
3fe3e6b1 JL |
1365 | } |
1366 | ||
3fe3e6b1 JL |
1367 | /* |
1368 | * Try to read extents from the first block indicated | |
1369 | * by fsbno to the end block of the file. | |
1370 | */ | |
52f1acc8 | 1371 | fsbno = XFS_B_TO_FSBT(mp, start); |
8aa7d37e | 1372 | lastbno = XFS_B_TO_FSB(mp, end); |
49c69591 | 1373 | |
52f1acc8 JL |
1374 | for (;;) { |
1375 | struct xfs_bmbt_irec map[2]; | |
1376 | int nmap = 2; | |
1377 | unsigned int i; | |
3fe3e6b1 | 1378 | |
8aa7d37e | 1379 | error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap, |
52f1acc8 JL |
1380 | XFS_BMAPI_ENTIRE); |
1381 | if (error) | |
8aa7d37e | 1382 | goto out_error; |
3fe3e6b1 | 1383 | |
52f1acc8 JL |
1384 | /* No extents at given offset, must be beyond EOF */ |
1385 | if (nmap == 0) { | |
2451337d | 1386 | error = -ENXIO; |
8aa7d37e | 1387 | goto out_error; |
52f1acc8 JL |
1388 | } |
1389 | ||
1390 | for (i = 0; i < nmap; i++) { | |
1391 | offset = max_t(loff_t, start, | |
1392 | XFS_FSB_TO_B(mp, map[i].br_startoff)); | |
1393 | ||
49c69591 ES |
1394 | /* Landed in the hole we wanted? */ |
1395 | if (whence == SEEK_HOLE && | |
1396 | map[i].br_startblock == HOLESTARTBLOCK) | |
1397 | goto out; | |
1398 | ||
1399 | /* Landed in the data extent we wanted? */ | |
1400 | if (whence == SEEK_DATA && | |
1401 | (map[i].br_startblock == DELAYSTARTBLOCK || | |
1402 | (map[i].br_state == XFS_EXT_NORM && | |
1403 | !isnullstartblock(map[i].br_startblock)))) | |
52f1acc8 JL |
1404 | goto out; |
1405 | ||
1406 | /* | |
49c69591 ES |
1407 | * Landed in an unwritten extent, try to search |
1408 | * for hole or data from page cache. | |
52f1acc8 JL |
1409 | */ |
1410 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { | |
1411 | if (xfs_find_get_desired_pgoff(inode, &map[i], | |
49c69591 ES |
1412 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
1413 | &offset)) | |
52f1acc8 JL |
1414 | goto out; |
1415 | } | |
1416 | } | |
1417 | ||
1418 | /* | |
49c69591 ES |
1419 | * We only received one extent out of the two requested. This |
1420 | * means we've hit EOF and didn't find what we are looking for. | |
52f1acc8 | 1421 | */ |
3fe3e6b1 | 1422 | if (nmap == 1) { |
49c69591 ES |
1423 | /* |
1424 | * If we were looking for a hole, set offset to | |
1425 | * the end of the file (i.e., there is an implicit | |
1426 | * hole at the end of any file). | |
1427 | */ | |
1428 | if (whence == SEEK_HOLE) { | |
8aa7d37e | 1429 | offset = end; |
49c69591 ES |
1430 | break; |
1431 | } | |
1432 | /* | |
1433 | * If we were looking for data, it's nowhere to be found | |
1434 | */ | |
1435 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1436 | error = -ENXIO; |
8aa7d37e | 1437 | goto out_error; |
3fe3e6b1 JL |
1438 | } |
1439 | ||
52f1acc8 JL |
1440 | ASSERT(i > 1); |
1441 | ||
1442 | /* | |
1443 | * Nothing was found, proceed to the next round of search | |
49c69591 | 1444 | * if the next reading offset is not at or beyond EOF. |
52f1acc8 JL |
1445 | */ |
1446 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; | |
1447 | start = XFS_FSB_TO_B(mp, fsbno); | |
8aa7d37e | 1448 | if (start >= end) { |
49c69591 | 1449 | if (whence == SEEK_HOLE) { |
8aa7d37e | 1450 | offset = end; |
49c69591 ES |
1451 | break; |
1452 | } | |
1453 | ASSERT(whence == SEEK_DATA); | |
2451337d | 1454 | error = -ENXIO; |
8aa7d37e | 1455 | goto out_error; |
52f1acc8 | 1456 | } |
3fe3e6b1 JL |
1457 | } |
1458 | ||
b686d1f7 JL |
1459 | out: |
1460 | /* | |
49c69591 | 1461 | * If at this point we have found the hole we wanted, the returned |
b686d1f7 | 1462 | * offset may be bigger than the file size as it may be aligned to |
49c69591 | 1463 | * page boundary for unwritten extents. We need to deal with this |
b686d1f7 JL |
1464 | * situation in particular. |
1465 | */ | |
49c69591 | 1466 | if (whence == SEEK_HOLE) |
8aa7d37e ES |
1467 | offset = min_t(loff_t, offset, end); |
1468 | ||
1469 | return offset; | |
1470 | ||
1471 | out_error: | |
1472 | return error; | |
1473 | } | |
1474 | ||
1475 | STATIC loff_t | |
1476 | xfs_seek_hole_data( | |
1477 | struct file *file, | |
1478 | loff_t start, | |
1479 | int whence) | |
1480 | { | |
1481 | struct inode *inode = file->f_mapping->host; | |
1482 | struct xfs_inode *ip = XFS_I(inode); | |
1483 | struct xfs_mount *mp = ip->i_mount; | |
1484 | uint lock; | |
1485 | loff_t offset, end; | |
1486 | int error = 0; | |
1487 | ||
1488 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1489 | return -EIO; | |
1490 | ||
1491 | lock = xfs_ilock_data_map_shared(ip); | |
1492 | ||
1493 | end = i_size_read(inode); | |
1494 | offset = __xfs_seek_hole_data(inode, start, end, whence); | |
1495 | if (offset < 0) { | |
1496 | error = offset; | |
1497 | goto out_unlock; | |
1498 | } | |
1499 | ||
46a1c2c7 | 1500 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
3fe3e6b1 JL |
1501 | |
1502 | out_unlock: | |
01f4f327 | 1503 | xfs_iunlock(ip, lock); |
3fe3e6b1 JL |
1504 | |
1505 | if (error) | |
2451337d | 1506 | return error; |
3fe3e6b1 JL |
1507 | return offset; |
1508 | } | |
1509 | ||
1510 | STATIC loff_t | |
1511 | xfs_file_llseek( | |
1512 | struct file *file, | |
1513 | loff_t offset, | |
59f9c004 | 1514 | int whence) |
3fe3e6b1 | 1515 | { |
59f9c004 | 1516 | switch (whence) { |
3fe3e6b1 JL |
1517 | case SEEK_END: |
1518 | case SEEK_CUR: | |
1519 | case SEEK_SET: | |
59f9c004 | 1520 | return generic_file_llseek(file, offset, whence); |
3fe3e6b1 | 1521 | case SEEK_HOLE: |
49c69591 | 1522 | case SEEK_DATA: |
59f9c004 | 1523 | return xfs_seek_hole_data(file, offset, whence); |
3fe3e6b1 JL |
1524 | default: |
1525 | return -EINVAL; | |
1526 | } | |
1527 | } | |
1528 | ||
de0e8c20 DC |
1529 | /* |
1530 | * Locking for serialisation of IO during page faults. This results in a lock | |
1531 | * ordering of: | |
1532 | * | |
1533 | * mmap_sem (MM) | |
6b698ede | 1534 | * sb_start_pagefault(vfs, freeze) |
13ad4fe3 | 1535 | * i_mmaplock (XFS - truncate serialisation) |
6b698ede DC |
1536 | * page_lock (MM) |
1537 | * i_lock (XFS - extent map serialisation) | |
de0e8c20 | 1538 | */ |
de0e8c20 | 1539 | |
075a924d DC |
1540 | /* |
1541 | * mmap()d file has taken write protection fault and is being made writable. We | |
1542 | * can set the page state up correctly for a writable page, which means we can | |
1543 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent | |
1544 | * mapping. | |
de0e8c20 DC |
1545 | */ |
1546 | STATIC int | |
075a924d | 1547 | xfs_filemap_page_mkwrite( |
de0e8c20 DC |
1548 | struct vm_area_struct *vma, |
1549 | struct vm_fault *vmf) | |
1550 | { | |
6b698ede | 1551 | struct inode *inode = file_inode(vma->vm_file); |
ec56b1f1 | 1552 | int ret; |
de0e8c20 | 1553 | |
6b698ede | 1554 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
de0e8c20 | 1555 | |
6b698ede | 1556 | sb_start_pagefault(inode->i_sb); |
ec56b1f1 | 1557 | file_update_time(vma->vm_file); |
6b698ede | 1558 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
de0e8c20 | 1559 | |
6b698ede | 1560 | if (IS_DAX(inode)) { |
01a155e6 | 1561 | ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault, NULL); |
6b698ede | 1562 | } else { |
5c500029 | 1563 | ret = block_page_mkwrite(vma, vmf, xfs_get_blocks); |
6b698ede DC |
1564 | ret = block_page_mkwrite_return(ret); |
1565 | } | |
1566 | ||
1567 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
1568 | sb_end_pagefault(inode->i_sb); | |
1569 | ||
1570 | return ret; | |
de0e8c20 DC |
1571 | } |
1572 | ||
075a924d | 1573 | STATIC int |
6b698ede | 1574 | xfs_filemap_fault( |
075a924d DC |
1575 | struct vm_area_struct *vma, |
1576 | struct vm_fault *vmf) | |
1577 | { | |
b2442c5a | 1578 | struct inode *inode = file_inode(vma->vm_file); |
6b698ede | 1579 | int ret; |
ec56b1f1 | 1580 | |
b2442c5a | 1581 | trace_xfs_filemap_fault(XFS_I(inode)); |
075a924d | 1582 | |
6b698ede | 1583 | /* DAX can shortcut the normal fault path on write faults! */ |
b2442c5a | 1584 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
6b698ede | 1585 | return xfs_filemap_page_mkwrite(vma, vmf); |
075a924d | 1586 | |
b2442c5a DC |
1587 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
1588 | if (IS_DAX(inode)) { | |
1589 | /* | |
1590 | * we do not want to trigger unwritten extent conversion on read | |
1591 | * faults - that is unnecessary overhead and would also require | |
1592 | * changes to xfs_get_blocks_direct() to map unwritten extent | |
1593 | * ioend for conversion on read-only mappings. | |
1594 | */ | |
3e12dbbd | 1595 | ret = __dax_fault(vma, vmf, xfs_get_blocks_dax_fault, NULL); |
b2442c5a DC |
1596 | } else |
1597 | ret = filemap_fault(vma, vmf); | |
1598 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); | |
075a924d | 1599 | |
6b698ede DC |
1600 | return ret; |
1601 | } | |
1602 | ||
13ad4fe3 DC |
1603 | /* |
1604 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on | |
1605 | * both read and write faults. Hence we need to handle both cases. There is no | |
1606 | * ->pmd_mkwrite callout for huge pages, so we have a single function here to | |
1607 | * handle both cases here. @flags carries the information on the type of fault | |
1608 | * occuring. | |
1609 | */ | |
acd76e74 MW |
1610 | STATIC int |
1611 | xfs_filemap_pmd_fault( | |
1612 | struct vm_area_struct *vma, | |
1613 | unsigned long addr, | |
1614 | pmd_t *pmd, | |
1615 | unsigned int flags) | |
1616 | { | |
1617 | struct inode *inode = file_inode(vma->vm_file); | |
1618 | struct xfs_inode *ip = XFS_I(inode); | |
1619 | int ret; | |
1620 | ||
1621 | if (!IS_DAX(inode)) | |
1622 | return VM_FAULT_FALLBACK; | |
1623 | ||
1624 | trace_xfs_filemap_pmd_fault(ip); | |
1625 | ||
13ad4fe3 DC |
1626 | if (flags & FAULT_FLAG_WRITE) { |
1627 | sb_start_pagefault(inode->i_sb); | |
1628 | file_update_time(vma->vm_file); | |
1629 | } | |
1630 | ||
acd76e74 | 1631 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
3e12dbbd | 1632 | ret = __dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault, |
01a155e6 | 1633 | NULL); |
acd76e74 | 1634 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
acd76e74 | 1635 | |
13ad4fe3 DC |
1636 | if (flags & FAULT_FLAG_WRITE) |
1637 | sb_end_pagefault(inode->i_sb); | |
acd76e74 MW |
1638 | |
1639 | return ret; | |
1640 | } | |
1641 | ||
3af49285 DC |
1642 | /* |
1643 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp | |
1644 | * updates on write faults. In reality, it's need to serialise against | |
5eb88dca RZ |
1645 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
1646 | * to ensure we serialise the fault barrier in place. | |
3af49285 DC |
1647 | */ |
1648 | static int | |
1649 | xfs_filemap_pfn_mkwrite( | |
1650 | struct vm_area_struct *vma, | |
1651 | struct vm_fault *vmf) | |
1652 | { | |
1653 | ||
1654 | struct inode *inode = file_inode(vma->vm_file); | |
1655 | struct xfs_inode *ip = XFS_I(inode); | |
1656 | int ret = VM_FAULT_NOPAGE; | |
1657 | loff_t size; | |
1658 | ||
1659 | trace_xfs_filemap_pfn_mkwrite(ip); | |
1660 | ||
1661 | sb_start_pagefault(inode->i_sb); | |
1662 | file_update_time(vma->vm_file); | |
1663 | ||
1664 | /* check if the faulting page hasn't raced with truncate */ | |
1665 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); | |
1666 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1667 | if (vmf->pgoff >= size) | |
1668 | ret = VM_FAULT_SIGBUS; | |
5eb88dca RZ |
1669 | else if (IS_DAX(inode)) |
1670 | ret = dax_pfn_mkwrite(vma, vmf); | |
3af49285 DC |
1671 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
1672 | sb_end_pagefault(inode->i_sb); | |
acd76e74 | 1673 | return ret; |
3af49285 | 1674 | |
acd76e74 MW |
1675 | } |
1676 | ||
6b698ede DC |
1677 | static const struct vm_operations_struct xfs_file_vm_ops = { |
1678 | .fault = xfs_filemap_fault, | |
acd76e74 | 1679 | .pmd_fault = xfs_filemap_pmd_fault, |
6b698ede DC |
1680 | .map_pages = filemap_map_pages, |
1681 | .page_mkwrite = xfs_filemap_page_mkwrite, | |
3af49285 | 1682 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
6b698ede DC |
1683 | }; |
1684 | ||
1685 | STATIC int | |
1686 | xfs_file_mmap( | |
1687 | struct file *filp, | |
1688 | struct vm_area_struct *vma) | |
1689 | { | |
1690 | file_accessed(filp); | |
1691 | vma->vm_ops = &xfs_file_vm_ops; | |
1692 | if (IS_DAX(file_inode(filp))) | |
acd76e74 | 1693 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
6b698ede | 1694 | return 0; |
075a924d DC |
1695 | } |
1696 | ||
4b6f5d20 | 1697 | const struct file_operations xfs_file_operations = { |
3fe3e6b1 | 1698 | .llseek = xfs_file_llseek, |
b4f5d2c6 | 1699 | .read_iter = xfs_file_read_iter, |
bf97f3bc | 1700 | .write_iter = xfs_file_write_iter, |
1b895840 | 1701 | .splice_read = xfs_file_splice_read, |
8d020765 | 1702 | .splice_write = iter_file_splice_write, |
3562fd45 | 1703 | .unlocked_ioctl = xfs_file_ioctl, |
1da177e4 | 1704 | #ifdef CONFIG_COMPAT |
3562fd45 | 1705 | .compat_ioctl = xfs_file_compat_ioctl, |
1da177e4 | 1706 | #endif |
3562fd45 NS |
1707 | .mmap = xfs_file_mmap, |
1708 | .open = xfs_file_open, | |
1709 | .release = xfs_file_release, | |
1710 | .fsync = xfs_file_fsync, | |
2fe17c10 | 1711 | .fallocate = xfs_file_fallocate, |
1da177e4 LT |
1712 | }; |
1713 | ||
4b6f5d20 | 1714 | const struct file_operations xfs_dir_file_operations = { |
f999a5bf | 1715 | .open = xfs_dir_open, |
1da177e4 | 1716 | .read = generic_read_dir, |
b8227554 | 1717 | .iterate = xfs_file_readdir, |
59af1584 | 1718 | .llseek = generic_file_llseek, |
3562fd45 | 1719 | .unlocked_ioctl = xfs_file_ioctl, |
d3870398 | 1720 | #ifdef CONFIG_COMPAT |
3562fd45 | 1721 | .compat_ioctl = xfs_file_compat_ioctl, |
d3870398 | 1722 | #endif |
1da2f2db | 1723 | .fsync = xfs_dir_fsync, |
1da177e4 | 1724 | }; |