blob: 54a67dd9ac0a5fbe5a7caf4271d14c67798fd872 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000019#include "xfs_fs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110020#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
27#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_dinode.h"
30#include "xfs_inode.h"
Christoph Hellwigfd3200b2010-02-15 09:44:48 +000031#include "xfs_inode_item.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000032#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_error.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100034#include "xfs_vnodeops.h"
Christoph Hellwigf999a5b2008-11-28 14:23:32 +110035#include "xfs_da_btree.h"
Christoph Hellwigddcd8562008-12-03 07:55:34 -050036#include "xfs_ioctl.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000037#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010040#include <linux/falloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040042static const struct vm_operations_struct xfs_file_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Christoph Hellwigdda35b82010-02-15 09:44:46 +000044/*
Dave Chinner487f84f2011-01-12 11:37:10 +110045 * Locking primitives for read and write IO paths to ensure we consistently use
46 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
47 */
48static inline void
49xfs_rw_ilock(
50 struct xfs_inode *ip,
51 int type)
52{
53 if (type & XFS_IOLOCK_EXCL)
54 mutex_lock(&VFS_I(ip)->i_mutex);
55 xfs_ilock(ip, type);
56}
57
58static inline void
59xfs_rw_iunlock(
60 struct xfs_inode *ip,
61 int type)
62{
63 xfs_iunlock(ip, type);
64 if (type & XFS_IOLOCK_EXCL)
65 mutex_unlock(&VFS_I(ip)->i_mutex);
66}
67
68static inline void
69xfs_rw_ilock_demote(
70 struct xfs_inode *ip,
71 int type)
72{
73 xfs_ilock_demote(ip, type);
74 if (type & XFS_IOLOCK_EXCL)
75 mutex_unlock(&VFS_I(ip)->i_mutex);
76}
77
78/*
Christoph Hellwigdda35b82010-02-15 09:44:46 +000079 * xfs_iozero
80 *
81 * xfs_iozero clears the specified range of buffer supplied,
82 * and marks all the affected blocks as valid and modified. If
83 * an affected block is not allocated, it will be allocated. If
84 * an affected block is not completely overwritten, and is not
85 * valid before the operation, it will be read from disk before
86 * being partially zeroed.
87 */
88STATIC int
89xfs_iozero(
90 struct xfs_inode *ip, /* inode */
91 loff_t pos, /* offset in file */
92 size_t count) /* size of data to zero */
93{
94 struct page *page;
95 struct address_space *mapping;
96 int status;
97
98 mapping = VFS_I(ip)->i_mapping;
99 do {
100 unsigned offset, bytes;
101 void *fsdata;
102
103 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
104 bytes = PAGE_CACHE_SIZE - offset;
105 if (bytes > count)
106 bytes = count;
107
108 status = pagecache_write_begin(NULL, mapping, pos, bytes,
109 AOP_FLAG_UNINTERRUPTIBLE,
110 &page, &fsdata);
111 if (status)
112 break;
113
114 zero_user(page, offset, bytes);
115
116 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
117 page, fsdata);
118 WARN_ON(status <= 0); /* can't return less than zero! */
119 pos += bytes;
120 count -= bytes;
121 status = 0;
122 } while (count);
123
124 return (-status);
125}
126
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000127/*
128 * Fsync operations on directories are much simpler than on regular files,
129 * as there is no file data to flush, and thus also no need for explicit
130 * cache flush operations, and there are no non-transaction metadata updates
131 * on directories either.
132 */
133STATIC int
134xfs_dir_fsync(
135 struct file *file,
136 loff_t start,
137 loff_t end,
138 int datasync)
139{
140 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
141 struct xfs_mount *mp = ip->i_mount;
142 xfs_lsn_t lsn = 0;
143
144 trace_xfs_dir_fsync(ip);
145
146 xfs_ilock(ip, XFS_ILOCK_SHARED);
147 if (xfs_ipincount(ip))
148 lsn = ip->i_itemp->ili_last_lsn;
149 xfs_iunlock(ip, XFS_ILOCK_SHARED);
150
151 if (!lsn)
152 return 0;
153 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
154}
155
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000156STATIC int
157xfs_file_fsync(
158 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400159 loff_t start,
160 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000161 int datasync)
162{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200163 struct inode *inode = file->f_mapping->host;
164 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000165 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000166 int error = 0;
167 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000168 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000169
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000170 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000171
Josef Bacik02c24a82011-07-16 20:44:56 -0400172 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
173 if (error)
174 return error;
175
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000176 if (XFS_FORCED_SHUTDOWN(mp))
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000177 return -XFS_ERROR(EIO);
178
179 xfs_iflags_clear(ip, XFS_ITRUNCATED);
180
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000181 if (mp->m_flags & XFS_MOUNT_BARRIER) {
182 /*
183 * If we have an RT and/or log subvolume we need to make sure
184 * to flush the write cache the device used for file data
185 * first. This is to ensure newly written file data make
186 * it to disk before logging the new inode size in case of
187 * an extending write.
188 */
189 if (XFS_IS_REALTIME_INODE(ip))
190 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
191 else if (mp->m_logdev_targp != mp->m_ddev_targp)
192 xfs_blkdev_issue_flush(mp->m_ddev_targp);
193 }
194
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000195 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000196 * All metadata updates are logged, which means that we just have
197 * to flush the log up to the latest LSN that touched the inode.
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000198 */
199 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000200 if (xfs_ipincount(ip)) {
201 if (!datasync ||
202 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
203 lsn = ip->i_itemp->ili_last_lsn;
204 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000205 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000206
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000207 if (lsn)
Christoph Hellwigb1037052011-09-19 14:55:51 +0000208 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
209
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000210 /*
211 * If we only have a single device, and the log force about was
212 * a no-op we might have to flush the data device cache here.
213 * This can only happen for fdatasync/O_DSYNC if we were overwriting
214 * an already allocated file and thus do not have any metadata to
215 * commit.
216 */
217 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
218 mp->m_logdev_targp == mp->m_ddev_targp &&
219 !XFS_IS_REALTIME_INODE(ip) &&
220 !log_flushed)
221 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000222
223 return -error;
224}
225
Christoph Hellwig00258e32010-02-15 09:44:47 +0000226STATIC ssize_t
227xfs_file_aio_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000228 struct kiocb *iocb,
229 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000230 unsigned long nr_segs,
231 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000232{
233 struct file *file = iocb->ki_filp;
234 struct inode *inode = file->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000235 struct xfs_inode *ip = XFS_I(inode);
236 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000237 size_t size = 0;
238 ssize_t ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000239 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000240 xfs_fsize_t n;
241 unsigned long seg;
242
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000243 XFS_STATS_INC(xs_read_calls);
244
Christoph Hellwig00258e32010-02-15 09:44:47 +0000245 BUG_ON(iocb->ki_pos != pos);
246
247 if (unlikely(file->f_flags & O_DIRECT))
248 ioflags |= IO_ISDIRECT;
249 if (file->f_mode & FMODE_NOCMTIME)
250 ioflags |= IO_INVIS;
251
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000252 /* START copy & waste from filemap.c */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000253 for (seg = 0; seg < nr_segs; seg++) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000254 const struct iovec *iv = &iovp[seg];
255
256 /*
257 * If any segment has a negative length, or the cumulative
258 * length ever wraps negative then return -EINVAL.
259 */
260 size += iv->iov_len;
261 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
262 return XFS_ERROR(-EINVAL);
263 }
264 /* END copy & waste from filemap.c */
265
266 if (unlikely(ioflags & IO_ISDIRECT)) {
267 xfs_buftarg_t *target =
268 XFS_IS_REALTIME_INODE(ip) ?
269 mp->m_rtdev_targp : mp->m_ddev_targp;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000270 if ((iocb->ki_pos & target->bt_smask) ||
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000271 (size & target->bt_smask)) {
Christoph Hellwigce7ae152011-12-18 20:00:11 +0000272 if (iocb->ki_pos == i_size_read(inode))
Christoph Hellwig00258e32010-02-15 09:44:47 +0000273 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000274 return -XFS_ERROR(EINVAL);
275 }
276 }
277
Christoph Hellwig00258e32010-02-15 09:44:47 +0000278 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
279 if (n <= 0 || size == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000280 return 0;
281
282 if (n < size)
283 size = n;
284
285 if (XFS_FORCED_SHUTDOWN(mp))
286 return -EIO;
287
Dave Chinner0c38a252011-08-25 07:17:01 +0000288 /*
289 * Locking is a bit tricky here. If we take an exclusive lock
290 * for direct IO, we effectively serialise all new concurrent
291 * read IO to this file and block it behind IO that is currently in
292 * progress because IO in progress holds the IO lock shared. We only
293 * need to hold the lock exclusive to blow away the page cache, so
294 * only take lock exclusively if the page cache needs invalidation.
295 * This allows the normal direct IO case of no page cache pages to
296 * proceeed concurrently without serialisation.
297 */
298 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
299 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
300 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100301 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
302
Christoph Hellwig00258e32010-02-15 09:44:47 +0000303 if (inode->i_mapping->nrpages) {
304 ret = -xfs_flushinval_pages(ip,
305 (iocb->ki_pos & PAGE_CACHE_MASK),
306 -1, FI_REMAPF_LOCKED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100307 if (ret) {
308 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
309 return ret;
310 }
Christoph Hellwig00258e32010-02-15 09:44:47 +0000311 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100312 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000313 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000314
Christoph Hellwig00258e32010-02-15 09:44:47 +0000315 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000316
Christoph Hellwig00258e32010-02-15 09:44:47 +0000317 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000318 if (ret > 0)
319 XFS_STATS_ADD(xs_read_bytes, ret);
320
Dave Chinner487f84f2011-01-12 11:37:10 +1100321 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000322 return ret;
323}
324
Christoph Hellwig00258e32010-02-15 09:44:47 +0000325STATIC ssize_t
326xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000327 struct file *infilp,
328 loff_t *ppos,
329 struct pipe_inode_info *pipe,
330 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000331 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000332{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000333 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000334 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000335 ssize_t ret;
336
337 XFS_STATS_INC(xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000338
339 if (infilp->f_mode & FMODE_NOCMTIME)
340 ioflags |= IO_INVIS;
341
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000342 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
343 return -EIO;
344
Dave Chinner487f84f2011-01-12 11:37:10 +1100345 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000346
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000347 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
348
349 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
350 if (ret > 0)
351 XFS_STATS_ADD(xs_read_bytes, ret);
352
Dave Chinner487f84f2011-01-12 11:37:10 +1100353 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000354 return ret;
355}
356
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100357/*
Dave Chinner487f84f2011-01-12 11:37:10 +1100358 * xfs_file_splice_write() does not use xfs_rw_ilock() because
359 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
360 * couuld cause lock inversions between the aio_write path and the splice path
361 * if someone is doing concurrent splice(2) based writes and write(2) based
362 * writes to the same inode. The only real way to fix this is to re-implement
363 * the generic code here with correct locking orders.
364 */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000365STATIC ssize_t
366xfs_file_splice_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000367 struct pipe_inode_info *pipe,
368 struct file *outfilp,
369 loff_t *ppos,
370 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000371 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000372{
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000373 struct inode *inode = outfilp->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000374 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000375 int ioflags = 0;
376 ssize_t ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000377
378 XFS_STATS_INC(xs_write_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000379
380 if (outfilp->f_mode & FMODE_NOCMTIME)
381 ioflags |= IO_INVIS;
382
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000383 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
384 return -EIO;
385
386 xfs_ilock(ip, XFS_IOLOCK_EXCL);
387
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000388 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
389
390 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
Christoph Hellwigce7ae152011-12-18 20:00:11 +0000391 if (ret > 0)
392 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000393
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000394 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
395 return ret;
396}
397
398/*
399 * This routine is called to handle zeroing any space in the last
400 * block of the file that is beyond the EOF. We do this since the
401 * size is being increased without writing anything to that block
402 * and we don't want anyone to read the garbage on the disk.
403 */
404STATIC int /* error (positive) */
405xfs_zero_last_block(
406 xfs_inode_t *ip,
407 xfs_fsize_t offset,
408 xfs_fsize_t isize)
409{
410 xfs_fileoff_t last_fsb;
411 xfs_mount_t *mp = ip->i_mount;
412 int nimaps;
413 int zero_offset;
414 int zero_len;
415 int error = 0;
416 xfs_bmbt_irec_t imap;
417
418 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
419
420 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
421 if (zero_offset == 0) {
422 /*
423 * There are no extra bytes in the last block on disk to
424 * zero, so return.
425 */
426 return 0;
427 }
428
429 last_fsb = XFS_B_TO_FSBT(mp, isize);
430 nimaps = 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000431 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
432 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000433 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000434 ASSERT(nimaps > 0);
435 /*
436 * If the block underlying isize is just a hole, then there
437 * is nothing to zero.
438 */
439 if (imap.br_startblock == HOLESTARTBLOCK) {
440 return 0;
441 }
442 /*
443 * Zero the part of the last block beyond the EOF, and write it
444 * out sync. We need to drop the ilock while we do this so we
445 * don't deadlock when the buffer cache calls back to us.
446 */
447 xfs_iunlock(ip, XFS_ILOCK_EXCL);
448
449 zero_len = mp->m_sb.sb_blocksize - zero_offset;
450 if (isize + zero_len > offset)
451 zero_len = offset - isize;
452 error = xfs_iozero(ip, isize, zero_len);
453
454 xfs_ilock(ip, XFS_ILOCK_EXCL);
455 ASSERT(error >= 0);
456 return error;
457}
458
459/*
460 * Zero any on disk space between the current EOF and the new,
461 * larger EOF. This handles the normal case of zeroing the remainder
462 * of the last block in the file and the unusual case of zeroing blocks
463 * out beyond the size of the file. This second case only happens
464 * with fixed size extents and when the system crashes before the inode
465 * size was updated but after blocks were allocated. If fill is set,
466 * then any holes in the range are filled and zeroed. If not, the holes
467 * are left alone as holes.
468 */
469
470int /* error (positive) */
471xfs_zero_eof(
472 xfs_inode_t *ip,
473 xfs_off_t offset, /* starting I/O offset */
474 xfs_fsize_t isize) /* current inode size */
475{
476 xfs_mount_t *mp = ip->i_mount;
477 xfs_fileoff_t start_zero_fsb;
478 xfs_fileoff_t end_zero_fsb;
479 xfs_fileoff_t zero_count_fsb;
480 xfs_fileoff_t last_fsb;
481 xfs_fileoff_t zero_off;
482 xfs_fsize_t zero_len;
483 int nimaps;
484 int error = 0;
485 xfs_bmbt_irec_t imap;
486
487 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
488 ASSERT(offset > isize);
489
490 /*
491 * First handle zeroing the block on which isize resides.
492 * We only zero a part of that block so it is handled specially.
493 */
494 error = xfs_zero_last_block(ip, offset, isize);
495 if (error) {
496 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
497 return error;
498 }
499
500 /*
501 * Calculate the range between the new size and the old
502 * where blocks needing to be zeroed may exist. To get the
503 * block where the last byte in the file currently resides,
504 * we need to subtract one from the size and truncate back
505 * to a block boundary. We subtract 1 in case the size is
506 * exactly on a block boundary.
507 */
508 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
509 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
510 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
511 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
512 if (last_fsb == end_zero_fsb) {
513 /*
514 * The size was only incremented on its last block.
515 * We took care of that above, so just return.
516 */
517 return 0;
518 }
519
520 ASSERT(start_zero_fsb <= end_zero_fsb);
521 while (start_zero_fsb <= end_zero_fsb) {
522 nimaps = 1;
523 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
Dave Chinner5c8ed202011-09-18 20:40:45 +0000524 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
525 &imap, &nimaps, 0);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000526 if (error) {
527 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
528 return error;
529 }
530 ASSERT(nimaps > 0);
531
532 if (imap.br_state == XFS_EXT_UNWRITTEN ||
533 imap.br_startblock == HOLESTARTBLOCK) {
534 /*
535 * This loop handles initializing pages that were
536 * partially initialized by the code below this
537 * loop. It basically zeroes the part of the page
538 * that sits on a hole and sets the page as P_HOLE
539 * and calls remapf if it is a mapped file.
540 */
541 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
542 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
543 continue;
544 }
545
546 /*
547 * There are blocks we need to zero.
548 * Drop the inode lock while we're doing the I/O.
549 * We'll still have the iolock to protect us.
550 */
551 xfs_iunlock(ip, XFS_ILOCK_EXCL);
552
553 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
554 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
555
556 if ((zero_off + zero_len) > offset)
557 zero_len = offset - zero_off;
558
559 error = xfs_iozero(ip, zero_off, zero_len);
560 if (error) {
561 goto out_lock;
562 }
563
564 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
565 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
566
567 xfs_ilock(ip, XFS_ILOCK_EXCL);
568 }
569
570 return 0;
571
572out_lock:
573 xfs_ilock(ip, XFS_ILOCK_EXCL);
574 ASSERT(error >= 0);
575 return error;
576}
577
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100578/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100579 * Common pre-write limit and setup checks.
580 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000581 * Called with the iolocked held either shared and exclusive according to
582 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
583 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100584 */
585STATIC ssize_t
586xfs_file_aio_write_checks(
587 struct file *file,
588 loff_t *pos,
589 size_t *count,
590 int *iolock)
591{
592 struct inode *inode = file->f_mapping->host;
593 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100594 int error = 0;
595
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000596 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner7271d242011-08-25 07:17:02 +0000597restart:
Dave Chinner4d8d1582011-01-11 10:23:42 +1100598 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
599 if (error) {
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000600 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100601 return error;
602 }
603
Dave Chinner4d8d1582011-01-11 10:23:42 +1100604 /*
605 * If the offset is beyond the size of the file, we need to zero any
606 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000607 * write. If zeroing is needed and we are currently holding the
608 * iolock shared, we need to update it to exclusive which involves
609 * dropping all locks and relocking to maintain correct locking order.
610 * If we do this, restart the function to ensure all checks and values
611 * are still valid.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100612 */
Christoph Hellwig2813d682011-12-18 20:00:12 +0000613 if (*pos > i_size_read(inode)) {
Dave Chinner7271d242011-08-25 07:17:02 +0000614 if (*iolock == XFS_IOLOCK_SHARED) {
615 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
616 *iolock = XFS_IOLOCK_EXCL;
617 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
618 goto restart;
619 }
Christoph Hellwigce7ae152011-12-18 20:00:11 +0000620 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
Dave Chinner7271d242011-08-25 07:17:02 +0000621 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100622 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
623 if (error)
624 return error;
625
626 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000627 * Updating the timestamps will grab the ilock again from
628 * xfs_fs_dirty_inode, so we have to call it after dropping the
629 * lock above. Eventually we should look into a way to avoid
630 * the pointless lock roundtrip.
631 */
632 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
633 file_update_time(file);
634
635 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100636 * If we're writing the file then make sure to clear the setuid and
637 * setgid bits if the process is not being run by root. This keeps
638 * people from modifying setuid and setgid binaries.
639 */
640 return file_remove_suid(file);
641
642}
643
644/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100645 * xfs_file_dio_aio_write - handle direct IO writes
646 *
647 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100648 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100649 * follow locking changes and looping.
650 *
Dave Chinnereda77982011-01-11 10:22:40 +1100651 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
652 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
653 * pages are flushed out.
654 *
655 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
656 * allowing them to be done in parallel with reads and other direct IO writes.
657 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
658 * needs to do sub-block zeroing and that requires serialisation against other
659 * direct IOs to the same block. In this case we need to serialise the
660 * submission of the unaligned IOs so that we don't get racing block zeroing in
661 * the dio layer. To avoid the problem with aio, we also need to wait for
662 * outstanding IOs to complete so that unwritten extent conversion is completed
663 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000664 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100665 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100666 * Returns with locks held indicated by @iolock and errors indicated by
667 * negative return values.
668 */
669STATIC ssize_t
670xfs_file_dio_aio_write(
671 struct kiocb *iocb,
672 const struct iovec *iovp,
673 unsigned long nr_segs,
674 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000675 size_t ocount)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100676{
677 struct file *file = iocb->ki_filp;
678 struct address_space *mapping = file->f_mapping;
679 struct inode *inode = mapping->host;
680 struct xfs_inode *ip = XFS_I(inode);
681 struct xfs_mount *mp = ip->i_mount;
682 ssize_t ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100683 size_t count = ocount;
Dave Chinnereda77982011-01-11 10:22:40 +1100684 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000685 int iolock;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100686 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
687 mp->m_rtdev_targp : mp->m_ddev_targp;
688
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100689 if ((pos & target->bt_smask) || (count & target->bt_smask))
690 return -XFS_ERROR(EINVAL);
691
Dave Chinnereda77982011-01-11 10:22:40 +1100692 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
693 unaligned_io = 1;
694
Dave Chinner7271d242011-08-25 07:17:02 +0000695 /*
696 * We don't need to take an exclusive lock unless there page cache needs
697 * to be invalidated or unaligned IO is being executed. We don't need to
698 * consider the EOF extension case here because
699 * xfs_file_aio_write_checks() will relock the inode as necessary for
700 * EOF zeroing cases and fill out the new inode size as appropriate.
701 */
702 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000703 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100704 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000705 iolock = XFS_IOLOCK_SHARED;
706 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000707
708 /*
709 * Recheck if there are cached pages that need invalidate after we got
710 * the iolock to protect against other threads adding new pages while
711 * we were waiting for the iolock.
712 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000713 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
714 xfs_rw_iunlock(ip, iolock);
715 iolock = XFS_IOLOCK_EXCL;
716 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000717 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100718
Christoph Hellwigd0606462011-12-18 20:00:14 +0000719 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100720 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000721 goto out;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100722
723 if (mapping->nrpages) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100724 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
725 FI_REMAPF_LOCKED);
726 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000727 goto out;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100728 }
729
Dave Chinnereda77982011-01-11 10:22:40 +1100730 /*
731 * If we are doing unaligned IO, wait for all other IO to drain,
732 * otherwise demote the lock if we had to flush cached pages
733 */
734 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000735 inode_dio_wait(inode);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000736 else if (iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100737 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000738 iolock = XFS_IOLOCK_SHARED;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100739 }
740
741 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
742 ret = generic_file_direct_write(iocb, iovp,
743 &nr_segs, pos, &iocb->ki_pos, count, ocount);
744
Christoph Hellwigd0606462011-12-18 20:00:14 +0000745out:
746 xfs_rw_iunlock(ip, iolock);
747
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100748 /* No fallback to buffered IO on errors for XFS. */
749 ASSERT(ret < 0 || ret == count);
750 return ret;
751}
752
Christoph Hellwig00258e32010-02-15 09:44:47 +0000753STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100754xfs_file_buffered_aio_write(
755 struct kiocb *iocb,
756 const struct iovec *iovp,
757 unsigned long nr_segs,
758 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000759 size_t ocount)
Dave Chinner637bbc72011-01-11 10:17:30 +1100760{
761 struct file *file = iocb->ki_filp;
762 struct address_space *mapping = file->f_mapping;
763 struct inode *inode = mapping->host;
764 struct xfs_inode *ip = XFS_I(inode);
765 ssize_t ret;
766 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000767 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100768 size_t count = ocount;
769
Christoph Hellwigd0606462011-12-18 20:00:14 +0000770 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100771
Christoph Hellwigd0606462011-12-18 20:00:14 +0000772 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100773 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000774 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100775
776 /* We can write back this queue in page reclaim */
777 current->backing_dev_info = mapping->backing_dev_info;
778
779write_retry:
780 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
781 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
782 pos, &iocb->ki_pos, count, ret);
783 /*
784 * if we just got an ENOSPC, flush the inode now we aren't holding any
785 * page locks and retry *once*
786 */
787 if (ret == -ENOSPC && !enospc) {
Dave Chinner637bbc72011-01-11 10:17:30 +1100788 enospc = 1;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000789 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
790 if (!ret)
791 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100792 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000793
Dave Chinner637bbc72011-01-11 10:17:30 +1100794 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000795out:
796 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100797 return ret;
798}
799
800STATIC ssize_t
Christoph Hellwig00258e32010-02-15 09:44:47 +0000801xfs_file_aio_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000802 struct kiocb *iocb,
803 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000804 unsigned long nr_segs,
805 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000806{
807 struct file *file = iocb->ki_filp;
808 struct address_space *mapping = file->f_mapping;
809 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000810 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100811 ssize_t ret;
Dave Chinner637bbc72011-01-11 10:17:30 +1100812 size_t ocount = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000813
814 XFS_STATS_INC(xs_write_calls);
815
Christoph Hellwig00258e32010-02-15 09:44:47 +0000816 BUG_ON(iocb->ki_pos != pos);
817
Dave Chinnera363f0c2011-01-11 10:13:53 +1100818 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
819 if (ret)
820 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000821
Dave Chinner637bbc72011-01-11 10:17:30 +1100822 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000823 return 0;
824
Dave Chinner637bbc72011-01-11 10:17:30 +1100825 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000826
Dave Chinner637bbc72011-01-11 10:17:30 +1100827 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000828 return -EIO;
829
Dave Chinner637bbc72011-01-11 10:17:30 +1100830 if (unlikely(file->f_flags & O_DIRECT))
Christoph Hellwigd0606462011-12-18 20:00:14 +0000831 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
Dave Chinner637bbc72011-01-11 10:17:30 +1100832 else
833 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000834 ocount);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000835
Christoph Hellwigd0606462011-12-18 20:00:14 +0000836 if (ret > 0) {
837 ssize_t err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000838
Christoph Hellwigd0606462011-12-18 20:00:14 +0000839 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigce7ae152011-12-18 20:00:11 +0000840
Christoph Hellwigd0606462011-12-18 20:00:14 +0000841 /* Handle various SYNC-type writes */
842 err = generic_write_sync(file, pos, ret);
843 if (err < 0)
844 ret = err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000845 }
846
Dave Chinnera363f0c2011-01-11 10:13:53 +1100847 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000848}
849
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100850STATIC long
851xfs_file_fallocate(
852 struct file *file,
853 int mode,
854 loff_t offset,
855 loff_t len)
856{
857 struct inode *inode = file->f_path.dentry->d_inode;
858 long error;
859 loff_t new_size = 0;
860 xfs_flock64_t bf;
861 xfs_inode_t *ip = XFS_I(inode);
862 int cmd = XFS_IOC_RESVSP;
Dave Chinner82878892011-03-26 09:13:08 +1100863 int attr_flags = XFS_ATTR_NOLOCK;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100864
865 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
866 return -EOPNOTSUPP;
867
868 bf.l_whence = 0;
869 bf.l_start = offset;
870 bf.l_len = len;
871
872 xfs_ilock(ip, XFS_IOLOCK_EXCL);
873
874 if (mode & FALLOC_FL_PUNCH_HOLE)
875 cmd = XFS_IOC_UNRESVSP;
876
877 /* check the new inode size is valid before allocating */
878 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
879 offset + len > i_size_read(inode)) {
880 new_size = offset + len;
881 error = inode_newsize_ok(inode, new_size);
882 if (error)
883 goto out_unlock;
884 }
885
Dave Chinner82878892011-03-26 09:13:08 +1100886 if (file->f_flags & O_DSYNC)
887 attr_flags |= XFS_ATTR_SYNC;
888
889 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100890 if (error)
891 goto out_unlock;
892
893 /* Change file size if needed */
894 if (new_size) {
895 struct iattr iattr;
896
897 iattr.ia_valid = ATTR_SIZE;
898 iattr.ia_size = new_size;
Christoph Hellwigc4ed4242011-07-08 14:34:23 +0200899 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100900 }
901
902out_unlock:
903 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
904 return error;
905}
906
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100909xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100911 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100913 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100915 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
916 return -EIO;
917 return 0;
918}
919
920STATIC int
921xfs_dir_open(
922 struct inode *inode,
923 struct file *file)
924{
925 struct xfs_inode *ip = XFS_I(inode);
926 int mode;
927 int error;
928
929 error = xfs_file_open(inode, file);
930 if (error)
931 return error;
932
933 /*
934 * If there are any blocks, read-ahead block 0 as we're almost
935 * certain to have the next operation be a read there.
936 */
937 mode = xfs_ilock_map_shared(ip);
938 if (ip->i_d.di_nextents > 0)
939 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
940 xfs_iunlock(ip, mode);
941 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942}
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100945xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 struct inode *inode,
947 struct file *filp)
948{
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000949 return -xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950}
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100953xfs_file_readdir(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 struct file *filp,
955 void *dirent,
956 filldir_t filldir)
957{
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000958 struct inode *inode = filp->f_path.dentry->d_inode;
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000959 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000960 int error;
961 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000963 /*
964 * The Linux API doesn't pass down the total size of the buffer
965 * we read into down to the filesystem. With the filldir concept
966 * it's not needed for correct information, but the XFS dir2 leaf
967 * code wants an estimate of the buffer size to calculate it's
968 * readahead window and size the buffers used for mapping to
969 * physical blocks.
970 *
971 * Try to give it an estimate that's good enough, maybe at some
972 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +0000973 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000974 */
Eric Sandeena9cc7992010-02-03 17:50:13 +0000975 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000977 error = xfs_readdir(ip, dirent, bufsize,
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000978 (xfs_off_t *)&filp->f_pos, filldir);
979 if (error)
980 return -error;
981 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982}
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100985xfs_file_mmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 struct file *filp,
987 struct vm_area_struct *vma)
988{
Nathan Scott3562fd42006-03-14 14:00:35 +1100989 vma->vm_ops = &xfs_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -0700990 vma->vm_flags |= VM_CAN_NONLINEAR;
Dean Roehrich6fac0cb2005-06-21 14:07:45 +1000991
Nathan Scottfbc14622006-06-09 14:52:13 +1000992 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 return 0;
994}
995
David Chinner4f57dbc2007-07-19 16:28:17 +1000996/*
997 * mmap()d file has taken write protection fault and is being made
998 * writable. We can set the page state up correctly for a writable
999 * page, which means we can do correct delalloc accounting (ENOSPC
1000 * checking!) and unwritten extent mapping.
1001 */
1002STATIC int
1003xfs_vm_page_mkwrite(
1004 struct vm_area_struct *vma,
Nick Pigginc2ec1752009-03-31 15:23:21 -07001005 struct vm_fault *vmf)
David Chinner4f57dbc2007-07-19 16:28:17 +10001006{
Nick Pigginc2ec1752009-03-31 15:23:21 -07001007 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
David Chinner4f57dbc2007-07-19 16:28:17 +10001008}
1009
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001010const struct file_operations xfs_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 .llseek = generic_file_llseek,
1012 .read = do_sync_read,
Dean Roehrichbb3f7242005-09-02 15:43:05 +10001013 .write = do_sync_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001014 .aio_read = xfs_file_aio_read,
1015 .aio_write = xfs_file_aio_write,
Nathan Scott1b895842006-03-31 13:08:59 +10001016 .splice_read = xfs_file_splice_read,
1017 .splice_write = xfs_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001018 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001020 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001022 .mmap = xfs_file_mmap,
1023 .open = xfs_file_open,
1024 .release = xfs_file_release,
1025 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001026 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027};
1028
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001029const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001030 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 .read = generic_read_dir,
Nathan Scott3562fd42006-03-14 14:00:35 +11001032 .readdir = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001033 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001034 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001035#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001036 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001037#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001038 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039};
1040
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001041static const struct vm_operations_struct xfs_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001042 .fault = filemap_fault,
David Chinner4f57dbc2007-07-19 16:28:17 +10001043 .page_mkwrite = xfs_vm_page_mkwrite,
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001044};