blob: 8121e75352ee9bddd4726ca685d6d3e855256bdd [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"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110021#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110022#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110025#include "xfs_da_format.h"
26#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110028#include "xfs_trans.h"
Christoph Hellwigfd3200b2010-02-15 09:44:48 +000029#include "xfs_inode_item.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000030#include "xfs_bmap.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100031#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_error.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100033#include "xfs_dir2.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100034#include "xfs_dir2_priv.h"
Christoph Hellwigddcd8562008-12-03 07:55:34 -050035#include "xfs_ioctl.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000036#include "xfs_trace.h"
Dave Chinner239880e2013-10-23 10:50:10 +110037#include "xfs_log.h"
Brian Fosterdc06f3982014-07-24 19:49:28 +100038#include "xfs_icache.h"
Christoph Hellwig781355c2015-02-16 11:59:50 +110039#include "xfs_pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010042#include <linux/falloc.h>
Jeff Liud126d432012-08-21 17:11:57 +080043#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040045static const struct vm_operations_struct xfs_file_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Christoph Hellwigdda35b82010-02-15 09:44:46 +000047/*
Dave Chinner487f84f2011-01-12 11:37:10 +110048 * Locking primitives for read and write IO paths to ensure we consistently use
49 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
50 */
51static inline void
52xfs_rw_ilock(
53 struct xfs_inode *ip,
54 int type)
55{
56 if (type & XFS_IOLOCK_EXCL)
57 mutex_lock(&VFS_I(ip)->i_mutex);
58 xfs_ilock(ip, type);
59}
60
61static inline void
62xfs_rw_iunlock(
63 struct xfs_inode *ip,
64 int type)
65{
66 xfs_iunlock(ip, type);
67 if (type & XFS_IOLOCK_EXCL)
68 mutex_unlock(&VFS_I(ip)->i_mutex);
69}
70
71static inline void
72xfs_rw_ilock_demote(
73 struct xfs_inode *ip,
74 int type)
75{
76 xfs_ilock_demote(ip, type);
77 if (type & XFS_IOLOCK_EXCL)
78 mutex_unlock(&VFS_I(ip)->i_mutex);
79}
80
81/*
Christoph Hellwigdda35b82010-02-15 09:44:46 +000082 * xfs_iozero
83 *
84 * xfs_iozero clears the specified range of buffer supplied,
85 * and marks all the affected blocks as valid and modified. If
86 * an affected block is not allocated, it will be allocated. If
87 * an affected block is not completely overwritten, and is not
88 * valid before the operation, it will be read from disk before
89 * being partially zeroed.
90 */
Dave Chinneref9d8732012-11-29 15:26:33 +110091int
Christoph Hellwigdda35b82010-02-15 09:44:46 +000092xfs_iozero(
93 struct xfs_inode *ip, /* inode */
94 loff_t pos, /* offset in file */
95 size_t count) /* size of data to zero */
96{
97 struct page *page;
98 struct address_space *mapping;
99 int status;
100
101 mapping = VFS_I(ip)->i_mapping;
102 do {
103 unsigned offset, bytes;
104 void *fsdata;
105
106 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
107 bytes = PAGE_CACHE_SIZE - offset;
108 if (bytes > count)
109 bytes = count;
110
111 status = pagecache_write_begin(NULL, mapping, pos, bytes,
112 AOP_FLAG_UNINTERRUPTIBLE,
113 &page, &fsdata);
114 if (status)
115 break;
116
117 zero_user(page, offset, bytes);
118
119 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
120 page, fsdata);
121 WARN_ON(status <= 0); /* can't return less than zero! */
122 pos += bytes;
123 count -= bytes;
124 status = 0;
125 } while (count);
126
127 return (-status);
128}
129
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100130int
131xfs_update_prealloc_flags(
132 struct xfs_inode *ip,
133 enum xfs_prealloc_flags flags)
134{
135 struct xfs_trans *tp;
136 int error;
137
138 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
139 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
140 if (error) {
141 xfs_trans_cancel(tp, 0);
142 return error;
143 }
144
145 xfs_ilock(ip, XFS_ILOCK_EXCL);
146 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
147
148 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
149 ip->i_d.di_mode &= ~S_ISUID;
150 if (ip->i_d.di_mode & S_IXGRP)
151 ip->i_d.di_mode &= ~S_ISGID;
152 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
153 }
154
155 if (flags & XFS_PREALLOC_SET)
156 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
157 if (flags & XFS_PREALLOC_CLEAR)
158 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
159
160 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
161 if (flags & XFS_PREALLOC_SYNC)
162 xfs_trans_set_sync(tp);
163 return xfs_trans_commit(tp, 0);
164}
165
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000166/*
167 * Fsync operations on directories are much simpler than on regular files,
168 * as there is no file data to flush, and thus also no need for explicit
169 * cache flush operations, and there are no non-transaction metadata updates
170 * on directories either.
171 */
172STATIC int
173xfs_dir_fsync(
174 struct file *file,
175 loff_t start,
176 loff_t end,
177 int datasync)
178{
179 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
180 struct xfs_mount *mp = ip->i_mount;
181 xfs_lsn_t lsn = 0;
182
183 trace_xfs_dir_fsync(ip);
184
185 xfs_ilock(ip, XFS_ILOCK_SHARED);
186 if (xfs_ipincount(ip))
187 lsn = ip->i_itemp->ili_last_lsn;
188 xfs_iunlock(ip, XFS_ILOCK_SHARED);
189
190 if (!lsn)
191 return 0;
Dave Chinner24513372014-06-25 14:58:08 +1000192 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000193}
194
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000195STATIC int
196xfs_file_fsync(
197 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400198 loff_t start,
199 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000200 int datasync)
201{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200202 struct inode *inode = file->f_mapping->host;
203 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000204 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000205 int error = 0;
206 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000207 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000208
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000209 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000210
Josef Bacik02c24a82011-07-16 20:44:56 -0400211 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
212 if (error)
213 return error;
214
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000215 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000216 return -EIO;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000217
218 xfs_iflags_clear(ip, XFS_ITRUNCATED);
219
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000220 if (mp->m_flags & XFS_MOUNT_BARRIER) {
221 /*
222 * If we have an RT and/or log subvolume we need to make sure
223 * to flush the write cache the device used for file data
224 * first. This is to ensure newly written file data make
225 * it to disk before logging the new inode size in case of
226 * an extending write.
227 */
228 if (XFS_IS_REALTIME_INODE(ip))
229 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
230 else if (mp->m_logdev_targp != mp->m_ddev_targp)
231 xfs_blkdev_issue_flush(mp->m_ddev_targp);
232 }
233
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000234 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000235 * All metadata updates are logged, which means that we just have
236 * to flush the log up to the latest LSN that touched the inode.
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000237 */
238 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000239 if (xfs_ipincount(ip)) {
240 if (!datasync ||
241 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
242 lsn = ip->i_itemp->ili_last_lsn;
243 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000244 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000245
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000246 if (lsn)
Christoph Hellwigb1037052011-09-19 14:55:51 +0000247 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
248
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000249 /*
250 * If we only have a single device, and the log force about was
251 * a no-op we might have to flush the data device cache here.
252 * This can only happen for fdatasync/O_DSYNC if we were overwriting
253 * an already allocated file and thus do not have any metadata to
254 * commit.
255 */
256 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
257 mp->m_logdev_targp == mp->m_ddev_targp &&
258 !XFS_IS_REALTIME_INODE(ip) &&
259 !log_flushed)
260 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000261
Dave Chinner24513372014-06-25 14:58:08 +1000262 return error;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000263}
264
Christoph Hellwig00258e32010-02-15 09:44:47 +0000265STATIC ssize_t
Al Virob4f5d2c2014-04-02 14:37:59 -0400266xfs_file_read_iter(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000267 struct kiocb *iocb,
Al Virob4f5d2c2014-04-02 14:37:59 -0400268 struct iov_iter *to)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000269{
270 struct file *file = iocb->ki_filp;
271 struct inode *inode = file->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000272 struct xfs_inode *ip = XFS_I(inode);
273 struct xfs_mount *mp = ip->i_mount;
Al Virob4f5d2c2014-04-02 14:37:59 -0400274 size_t size = iov_iter_count(to);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000275 ssize_t ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000276 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000277 xfs_fsize_t n;
Al Virob4f5d2c2014-04-02 14:37:59 -0400278 loff_t pos = iocb->ki_pos;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000279
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000280 XFS_STATS_INC(xs_read_calls);
281
Al Viro2ba48ce2015-04-09 13:52:01 -0400282 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
Dave Chinnerb92cc59f62014-08-04 13:28:20 +1000283 ioflags |= XFS_IO_ISDIRECT;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000284 if (file->f_mode & FMODE_NOCMTIME)
Dave Chinnerb92cc59f62014-08-04 13:28:20 +1000285 ioflags |= XFS_IO_INVIS;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000286
Dave Chinnerb92cc59f62014-08-04 13:28:20 +1000287 if (unlikely(ioflags & XFS_IO_ISDIRECT)) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000288 xfs_buftarg_t *target =
289 XFS_IS_REALTIME_INODE(ip) ?
290 mp->m_rtdev_targp : mp->m_ddev_targp;
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600291 /* DIO must be aligned to device logical sector size */
292 if ((pos | size) & target->bt_logical_sectormask) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100293 if (pos == i_size_read(inode))
Christoph Hellwig00258e32010-02-15 09:44:47 +0000294 return 0;
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000295 return -EINVAL;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000296 }
297 }
298
Dave Chinnerfb595812012-11-12 22:53:57 +1100299 n = mp->m_super->s_maxbytes - pos;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000300 if (n <= 0 || size == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000301 return 0;
302
303 if (n < size)
304 size = n;
305
306 if (XFS_FORCED_SHUTDOWN(mp))
307 return -EIO;
308
Dave Chinner0c38a252011-08-25 07:17:01 +0000309 /*
310 * Locking is a bit tricky here. If we take an exclusive lock
311 * for direct IO, we effectively serialise all new concurrent
312 * read IO to this file and block it behind IO that is currently in
313 * progress because IO in progress holds the IO lock shared. We only
314 * need to hold the lock exclusive to blow away the page cache, so
315 * only take lock exclusively if the page cache needs invalidation.
316 * This allows the normal direct IO case of no page cache pages to
317 * proceeed concurrently without serialisation.
318 */
319 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinnerb92cc59f62014-08-04 13:28:20 +1000320 if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) {
Dave Chinner0c38a252011-08-25 07:17:01 +0000321 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100322 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
323
Christoph Hellwig00258e32010-02-15 09:44:47 +0000324 if (inode->i_mapping->nrpages) {
Dave Chinner8ff1e672014-05-15 09:21:37 +1000325 ret = filemap_write_and_wait_range(
Dave Chinnerfb595812012-11-12 22:53:57 +1100326 VFS_I(ip)->i_mapping,
Dave Chinner7d4ea3c2014-09-02 12:12:53 +1000327 pos, pos + size - 1);
Dave Chinner487f84f2011-01-12 11:37:10 +1100328 if (ret) {
329 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
330 return ret;
331 }
Chris Mason85e584d2014-09-02 12:12:52 +1000332
333 /*
334 * Invalidate whole pages. This can return an error if
335 * we fail to invalidate a page, but this should never
336 * happen on XFS. Warn if it does fail.
337 */
338 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
Dave Chinner7d4ea3c2014-09-02 12:12:53 +1000339 pos >> PAGE_CACHE_SHIFT,
340 (pos + size - 1) >> PAGE_CACHE_SHIFT);
Chris Mason85e584d2014-09-02 12:12:52 +1000341 WARN_ON_ONCE(ret);
342 ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000343 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100344 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000345 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000346
Dave Chinnerfb595812012-11-12 22:53:57 +1100347 trace_xfs_file_read(ip, size, pos, ioflags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000348
Al Virob4f5d2c2014-04-02 14:37:59 -0400349 ret = generic_file_read_iter(iocb, to);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000350 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
Christoph Hellwig00258e32010-02-15 09:44:47 +0000357STATIC ssize_t
358xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000359 struct file *infilp,
360 loff_t *ppos,
361 struct pipe_inode_info *pipe,
362 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000363 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000364{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000365 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000366 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000367 ssize_t ret;
368
369 XFS_STATS_INC(xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000370
371 if (infilp->f_mode & FMODE_NOCMTIME)
Dave Chinnerb92cc59f62014-08-04 13:28:20 +1000372 ioflags |= XFS_IO_INVIS;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000373
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000374 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
375 return -EIO;
376
Dave Chinner487f84f2011-01-12 11:37:10 +1100377 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000378
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000379 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
380
381 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
382 if (ret > 0)
383 XFS_STATS_ADD(xs_read_bytes, ret);
384
Dave Chinner487f84f2011-01-12 11:37:10 +1100385 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000386 return ret;
387}
388
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100389/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400390 * This routine is called to handle zeroing any space in the last block of the
391 * file that is beyond the EOF. We do this since the size is being increased
392 * without writing anything to that block and we don't want to read the
393 * garbage on the disk.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000394 */
395STATIC int /* error (positive) */
396xfs_zero_last_block(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400397 struct xfs_inode *ip,
398 xfs_fsize_t offset,
Dave Chinner5885ebd2015-02-23 22:37:08 +1100399 xfs_fsize_t isize,
400 bool *did_zeroing)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000401{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400402 struct xfs_mount *mp = ip->i_mount;
403 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
404 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
405 int zero_len;
406 int nimaps = 1;
407 int error = 0;
408 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000409
Christoph Hellwig193aec12012-03-27 10:34:49 -0400410 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000411 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400412 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000413 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000414 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400415
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000416 ASSERT(nimaps > 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400417
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000418 /*
419 * If the block underlying isize is just a hole, then there
420 * is nothing to zero.
421 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400422 if (imap.br_startblock == HOLESTARTBLOCK)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000423 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000424
425 zero_len = mp->m_sb.sb_blocksize - zero_offset;
426 if (isize + zero_len > offset)
427 zero_len = offset - isize;
Dave Chinner5885ebd2015-02-23 22:37:08 +1100428 *did_zeroing = true;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400429 return xfs_iozero(ip, isize, zero_len);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000430}
431
432/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400433 * Zero any on disk space between the current EOF and the new, larger EOF.
434 *
435 * This handles the normal case of zeroing the remainder of the last block in
436 * the file and the unusual case of zeroing blocks out beyond the size of the
437 * file. This second case only happens with fixed size extents and when the
438 * system crashes before the inode size was updated but after blocks were
439 * allocated.
440 *
441 * Expects the iolock to be held exclusive, and will take the ilock internally.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000442 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000443int /* error (positive) */
444xfs_zero_eof(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400445 struct xfs_inode *ip,
446 xfs_off_t offset, /* starting I/O offset */
Dave Chinner5885ebd2015-02-23 22:37:08 +1100447 xfs_fsize_t isize, /* current inode size */
448 bool *did_zeroing)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000449{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400450 struct xfs_mount *mp = ip->i_mount;
451 xfs_fileoff_t start_zero_fsb;
452 xfs_fileoff_t end_zero_fsb;
453 xfs_fileoff_t zero_count_fsb;
454 xfs_fileoff_t last_fsb;
455 xfs_fileoff_t zero_off;
456 xfs_fsize_t zero_len;
457 int nimaps;
458 int error = 0;
459 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000460
Christoph Hellwig193aec12012-03-27 10:34:49 -0400461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000462 ASSERT(offset > isize);
463
464 /*
465 * First handle zeroing the block on which isize resides.
Christoph Hellwig193aec12012-03-27 10:34:49 -0400466 *
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000467 * We only zero a part of that block so it is handled specially.
468 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400469 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
Dave Chinner5885ebd2015-02-23 22:37:08 +1100470 error = xfs_zero_last_block(ip, offset, isize, did_zeroing);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400471 if (error)
472 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000473 }
474
475 /*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400476 * Calculate the range between the new size and the old where blocks
477 * needing to be zeroed may exist.
478 *
479 * To get the block where the last byte in the file currently resides,
480 * we need to subtract one from the size and truncate back to a block
481 * boundary. We subtract 1 in case the size is exactly on a block
482 * boundary.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000483 */
484 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
485 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
486 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
487 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
488 if (last_fsb == end_zero_fsb) {
489 /*
490 * The size was only incremented on its last block.
491 * We took care of that above, so just return.
492 */
493 return 0;
494 }
495
496 ASSERT(start_zero_fsb <= end_zero_fsb);
497 while (start_zero_fsb <= end_zero_fsb) {
498 nimaps = 1;
499 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400500
501 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000502 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
503 &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
505 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000506 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400507
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000508 ASSERT(nimaps > 0);
509
510 if (imap.br_state == XFS_EXT_UNWRITTEN ||
511 imap.br_startblock == HOLESTARTBLOCK) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000512 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
513 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
514 continue;
515 }
516
517 /*
518 * There are blocks we need to zero.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000519 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000520 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
521 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
522
523 if ((zero_off + zero_len) > offset)
524 zero_len = offset - zero_off;
525
526 error = xfs_iozero(ip, zero_off, zero_len);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400527 if (error)
528 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000529
Dave Chinner5885ebd2015-02-23 22:37:08 +1100530 *did_zeroing = true;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000531 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
532 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000533 }
534
535 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000536}
537
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100538/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100539 * Common pre-write limit and setup checks.
540 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000541 * Called with the iolocked held either shared and exclusive according to
542 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
543 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100544 */
545STATIC ssize_t
546xfs_file_aio_write_checks(
Al Viro99733fa2015-04-07 14:25:18 -0400547 struct kiocb *iocb,
548 struct iov_iter *from,
Dave Chinner4d8d1582011-01-11 10:23:42 +1100549 int *iolock)
550{
Al Viro99733fa2015-04-07 14:25:18 -0400551 struct file *file = iocb->ki_filp;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100552 struct inode *inode = file->f_mapping->host;
553 struct xfs_inode *ip = XFS_I(inode);
Al Viro3309dd02015-04-09 12:55:47 -0400554 ssize_t error = 0;
Al Viro99733fa2015-04-07 14:25:18 -0400555 size_t count = iov_iter_count(from);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100556
Dave Chinner7271d242011-08-25 07:17:02 +0000557restart:
Al Viro3309dd02015-04-09 12:55:47 -0400558 error = generic_write_checks(iocb, from);
559 if (error <= 0)
Dave Chinner4d8d1582011-01-11 10:23:42 +1100560 return error;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100561
Christoph Hellwig21c3ea12015-04-13 11:38:29 +1000562 error = xfs_break_layouts(inode, iolock, true);
Christoph Hellwig781355c2015-02-16 11:59:50 +1100563 if (error)
564 return error;
565
Dave Chinner4d8d1582011-01-11 10:23:42 +1100566 /*
567 * If the offset is beyond the size of the file, we need to zero any
568 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000569 * write. If zeroing is needed and we are currently holding the
Christoph Hellwig467f7892012-03-27 10:34:47 -0400570 * iolock shared, we need to update it to exclusive which implies
571 * having to redo all checks before.
Dave Chinnerb9d59842015-04-16 22:03:07 +1000572 *
573 * We need to serialise against EOF updates that occur in IO
574 * completions here. We want to make sure that nobody is changing the
575 * size while we do this check until we have placed an IO barrier (i.e.
576 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
577 * The spinlock effectively forms a memory barrier once we have the
578 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
579 * and hence be able to correctly determine if we need to run zeroing.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100580 */
Dave Chinnerb9d59842015-04-16 22:03:07 +1000581 spin_lock(&ip->i_flags_lock);
Al Viro99733fa2015-04-07 14:25:18 -0400582 if (iocb->ki_pos > i_size_read(inode)) {
Dave Chinner5885ebd2015-02-23 22:37:08 +1100583 bool zero = false;
584
Dave Chinnerb9d59842015-04-16 22:03:07 +1000585 spin_unlock(&ip->i_flags_lock);
Dave Chinner7271d242011-08-25 07:17:02 +0000586 if (*iolock == XFS_IOLOCK_SHARED) {
Christoph Hellwig467f7892012-03-27 10:34:47 -0400587 xfs_rw_iunlock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000588 *iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig467f7892012-03-27 10:34:47 -0400589 xfs_rw_ilock(ip, *iolock);
Al Viro3309dd02015-04-09 12:55:47 -0400590 iov_iter_reexpand(from, count);
Dave Chinner40c63fb2015-04-16 22:03:17 +1000591
592 /*
593 * We now have an IO submission barrier in place, but
594 * AIO can do EOF updates during IO completion and hence
595 * we now need to wait for all of them to drain. Non-AIO
596 * DIO will have drained before we are given the
597 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
598 * no-op.
599 */
600 inode_dio_wait(inode);
Dave Chinner7271d242011-08-25 07:17:02 +0000601 goto restart;
602 }
Al Viro99733fa2015-04-07 14:25:18 -0400603 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
Christoph Hellwig467f7892012-03-27 10:34:47 -0400604 if (error)
605 return error;
Dave Chinnerb9d59842015-04-16 22:03:07 +1000606 } else
607 spin_unlock(&ip->i_flags_lock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100608
609 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000610 * Updating the timestamps will grab the ilock again from
611 * xfs_fs_dirty_inode, so we have to call it after dropping the
612 * lock above. Eventually we should look into a way to avoid
613 * the pointless lock roundtrip.
614 */
Josef Bacikc3b2da32012-03-26 09:59:21 -0400615 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
616 error = file_update_time(file);
617 if (error)
618 return error;
619 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000620
621 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100622 * If we're writing the file then make sure to clear the setuid and
623 * setgid bits if the process is not being run by root. This keeps
624 * people from modifying setuid and setgid binaries.
625 */
626 return file_remove_suid(file);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100627}
628
629/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100630 * xfs_file_dio_aio_write - handle direct IO writes
631 *
632 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100633 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100634 * follow locking changes and looping.
635 *
Dave Chinnereda77982011-01-11 10:22:40 +1100636 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
637 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
638 * pages are flushed out.
639 *
640 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
641 * allowing them to be done in parallel with reads and other direct IO writes.
642 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
643 * needs to do sub-block zeroing and that requires serialisation against other
644 * direct IOs to the same block. In this case we need to serialise the
645 * submission of the unaligned IOs so that we don't get racing block zeroing in
646 * the dio layer. To avoid the problem with aio, we also need to wait for
647 * outstanding IOs to complete so that unwritten extent conversion is completed
648 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000649 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100650 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100651 * Returns with locks held indicated by @iolock and errors indicated by
652 * negative return values.
653 */
654STATIC ssize_t
655xfs_file_dio_aio_write(
656 struct kiocb *iocb,
Al Virob3188912014-04-02 07:06:30 -0400657 struct iov_iter *from)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100658{
659 struct file *file = iocb->ki_filp;
660 struct address_space *mapping = file->f_mapping;
661 struct inode *inode = mapping->host;
662 struct xfs_inode *ip = XFS_I(inode);
663 struct xfs_mount *mp = ip->i_mount;
664 ssize_t ret = 0;
Dave Chinnereda77982011-01-11 10:22:40 +1100665 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000666 int iolock;
Al Virob3188912014-04-02 07:06:30 -0400667 size_t count = iov_iter_count(from);
668 loff_t pos = iocb->ki_pos;
Dave Chinner0cefb292015-04-16 22:03:27 +1000669 loff_t end;
670 struct iov_iter data;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100671 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
672 mp->m_rtdev_targp : mp->m_ddev_targp;
673
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600674 /* DIO must be aligned to device logical sector size */
675 if ((pos | count) & target->bt_logical_sectormask)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000676 return -EINVAL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100677
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600678 /* "unaligned" here means not aligned to a filesystem block */
Dave Chinnereda77982011-01-11 10:22:40 +1100679 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
680 unaligned_io = 1;
681
Dave Chinner7271d242011-08-25 07:17:02 +0000682 /*
683 * We don't need to take an exclusive lock unless there page cache needs
684 * to be invalidated or unaligned IO is being executed. We don't need to
685 * consider the EOF extension case here because
686 * xfs_file_aio_write_checks() will relock the inode as necessary for
687 * EOF zeroing cases and fill out the new inode size as appropriate.
688 */
689 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000690 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100691 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000692 iolock = XFS_IOLOCK_SHARED;
693 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000694
695 /*
696 * Recheck if there are cached pages that need invalidate after we got
697 * the iolock to protect against other threads adding new pages while
698 * we were waiting for the iolock.
699 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000700 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
701 xfs_rw_iunlock(ip, iolock);
702 iolock = XFS_IOLOCK_EXCL;
703 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000704 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100705
Al Viro99733fa2015-04-07 14:25:18 -0400706 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100707 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000708 goto out;
Al Viro99733fa2015-04-07 14:25:18 -0400709 count = iov_iter_count(from);
710 pos = iocb->ki_pos;
Dave Chinner0cefb292015-04-16 22:03:27 +1000711 end = pos + count - 1;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100712
713 if (mapping->nrpages) {
Dave Chinner07d50352014-04-17 08:15:27 +1000714 ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
Dave Chinner0cefb292015-04-16 22:03:27 +1000715 pos, end);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100716 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000717 goto out;
Dave Chinner834ffca2014-09-02 12:12:52 +1000718 /*
719 * Invalidate whole pages. This can return an error if
720 * we fail to invalidate a page, but this should never
721 * happen on XFS. Warn if it does fail.
722 */
723 ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
Dave Chinner7d4ea3c2014-09-02 12:12:53 +1000724 pos >> PAGE_CACHE_SHIFT,
Dave Chinner0cefb292015-04-16 22:03:27 +1000725 end >> PAGE_CACHE_SHIFT);
Dave Chinner834ffca2014-09-02 12:12:52 +1000726 WARN_ON_ONCE(ret);
727 ret = 0;
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);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100742
Dave Chinner0cefb292015-04-16 22:03:27 +1000743 data = *from;
Linus Torvalds1aef8822015-04-24 07:08:41 -0700744 ret = mapping->a_ops->direct_IO(iocb, &data, pos);
Dave Chinner0cefb292015-04-16 22:03:27 +1000745
746 /* see generic_file_direct_write() for why this is necessary */
747 if (mapping->nrpages) {
748 invalidate_inode_pages2_range(mapping,
749 pos >> PAGE_CACHE_SHIFT,
750 end >> PAGE_CACHE_SHIFT);
751 }
752
753 if (ret > 0) {
754 pos += ret;
755 iov_iter_advance(from, ret);
756 iocb->ki_pos = pos;
757 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000758out:
759 xfs_rw_iunlock(ip, iolock);
760
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100761 /* No fallback to buffered IO on errors for XFS. */
762 ASSERT(ret < 0 || ret == count);
763 return ret;
764}
765
Christoph Hellwig00258e32010-02-15 09:44:47 +0000766STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100767xfs_file_buffered_aio_write(
768 struct kiocb *iocb,
Al Virob3188912014-04-02 07:06:30 -0400769 struct iov_iter *from)
Dave Chinner637bbc72011-01-11 10:17:30 +1100770{
771 struct file *file = iocb->ki_filp;
772 struct address_space *mapping = file->f_mapping;
773 struct inode *inode = mapping->host;
774 struct xfs_inode *ip = XFS_I(inode);
775 ssize_t ret;
776 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000777 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100778
Christoph Hellwigd0606462011-12-18 20:00:14 +0000779 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100780
Al Viro99733fa2015-04-07 14:25:18 -0400781 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100782 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000783 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100784
785 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100786 current->backing_dev_info = inode_to_bdi(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100787
788write_retry:
Al Viro99733fa2015-04-07 14:25:18 -0400789 trace_xfs_file_buffered_write(ip, iov_iter_count(from),
790 iocb->ki_pos, 0);
791 ret = generic_perform_write(file, from, iocb->ki_pos);
Al Viro0a64bc22014-02-11 22:25:22 -0500792 if (likely(ret >= 0))
Al Viro99733fa2015-04-07 14:25:18 -0400793 iocb->ki_pos += ret;
Brian Fosterdc06f3982014-07-24 19:49:28 +1000794
Dave Chinner637bbc72011-01-11 10:17:30 +1100795 /*
Brian Fosterdc06f3982014-07-24 19:49:28 +1000796 * If we hit a space limit, try to free up some lingering preallocated
797 * space before returning an error. In the case of ENOSPC, first try to
798 * write back all dirty inodes to free up some of the excess reserved
799 * metadata space. This reduces the chances that the eofblocks scan
800 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
801 * also behaves as a filter to prevent too many eofblocks scans from
802 * running at the same time.
Dave Chinner637bbc72011-01-11 10:17:30 +1100803 */
Brian Fosterdc06f3982014-07-24 19:49:28 +1000804 if (ret == -EDQUOT && !enospc) {
805 enospc = xfs_inode_free_quota_eofblocks(ip);
806 if (enospc)
807 goto write_retry;
808 } else if (ret == -ENOSPC && !enospc) {
809 struct xfs_eofblocks eofb = {0};
810
Dave Chinner637bbc72011-01-11 10:17:30 +1100811 enospc = 1;
Dave Chinner9aa05002012-10-08 21:56:04 +1100812 xfs_flush_inodes(ip->i_mount);
Brian Fosterdc06f3982014-07-24 19:49:28 +1000813 eofb.eof_scan_owner = ip->i_ino; /* for locking */
814 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
815 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
Dave Chinner9aa05002012-10-08 21:56:04 +1100816 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100817 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000818
Dave Chinner637bbc72011-01-11 10:17:30 +1100819 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000820out:
821 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100822 return ret;
823}
824
825STATIC ssize_t
Al Virobf97f3bc2014-04-03 14:20:23 -0400826xfs_file_write_iter(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000827 struct kiocb *iocb,
Al Virobf97f3bc2014-04-03 14:20:23 -0400828 struct iov_iter *from)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000829{
830 struct file *file = iocb->ki_filp;
831 struct address_space *mapping = file->f_mapping;
832 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000833 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100834 ssize_t ret;
Al Virobf97f3bc2014-04-03 14:20:23 -0400835 size_t ocount = iov_iter_count(from);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000836
837 XFS_STATS_INC(xs_write_calls);
838
Dave Chinner637bbc72011-01-11 10:17:30 +1100839 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000840 return 0;
841
Al Virobf97f3bc2014-04-03 14:20:23 -0400842 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
843 return -EIO;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000844
Al Viro2ba48ce2015-04-09 13:52:01 -0400845 if (unlikely(iocb->ki_flags & IOCB_DIRECT))
Al Virobf97f3bc2014-04-03 14:20:23 -0400846 ret = xfs_file_dio_aio_write(iocb, from);
Dave Chinner637bbc72011-01-11 10:17:30 +1100847 else
Al Virobf97f3bc2014-04-03 14:20:23 -0400848 ret = xfs_file_buffered_aio_write(iocb, from);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000849
Christoph Hellwigd0606462011-12-18 20:00:14 +0000850 if (ret > 0) {
851 ssize_t err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000852
Christoph Hellwigd0606462011-12-18 20:00:14 +0000853 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000854
Christoph Hellwigd0606462011-12-18 20:00:14 +0000855 /* Handle various SYNC-type writes */
Al Virod311d792014-02-09 15:18:09 -0500856 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000857 if (err < 0)
858 ret = err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000859 }
Dave Chinnera363f0c2011-01-11 10:13:53 +1100860 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000861}
862
Namjae Jeona904b1c2015-03-25 15:08:56 +1100863#define XFS_FALLOC_FL_SUPPORTED \
864 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
865 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
866 FALLOC_FL_INSERT_RANGE)
867
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100868STATIC long
869xfs_file_fallocate(
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700870 struct file *file,
871 int mode,
872 loff_t offset,
873 loff_t len)
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100874{
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700875 struct inode *inode = file_inode(file);
876 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700877 long error;
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100878 enum xfs_prealloc_flags flags = 0;
Christoph Hellwig781355c2015-02-16 11:59:50 +1100879 uint iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700880 loff_t new_size = 0;
Namjae Jeona904b1c2015-03-25 15:08:56 +1100881 bool do_file_insert = 0;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100882
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700883 if (!S_ISREG(inode->i_mode))
884 return -EINVAL;
Namjae Jeona904b1c2015-03-25 15:08:56 +1100885 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100886 return -EOPNOTSUPP;
887
Christoph Hellwig781355c2015-02-16 11:59:50 +1100888 xfs_ilock(ip, iolock);
Christoph Hellwig21c3ea12015-04-13 11:38:29 +1000889 error = xfs_break_layouts(inode, &iolock, false);
Christoph Hellwig781355c2015-02-16 11:59:50 +1100890 if (error)
891 goto out_unlock;
892
Dave Chinnere8e9ad42015-02-23 21:45:32 +1100893 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
894 iolock |= XFS_MMAPLOCK_EXCL;
895
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700896 if (mode & FALLOC_FL_PUNCH_HOLE) {
897 error = xfs_free_file_space(ip, offset, len);
898 if (error)
899 goto out_unlock;
Namjae Jeone1d8fb82014-02-24 10:58:19 +1100900 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
901 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
902
903 if (offset & blksize_mask || len & blksize_mask) {
Dave Chinner24513372014-06-25 14:58:08 +1000904 error = -EINVAL;
Namjae Jeone1d8fb82014-02-24 10:58:19 +1100905 goto out_unlock;
906 }
907
Lukas Czerner23fffa92014-04-12 09:56:41 -0400908 /*
909 * There is no need to overlap collapse range with EOF,
910 * in which case it is effectively a truncate operation
911 */
912 if (offset + len >= i_size_read(inode)) {
Dave Chinner24513372014-06-25 14:58:08 +1000913 error = -EINVAL;
Lukas Czerner23fffa92014-04-12 09:56:41 -0400914 goto out_unlock;
915 }
916
Namjae Jeone1d8fb82014-02-24 10:58:19 +1100917 new_size = i_size_read(inode) - len;
918
919 error = xfs_collapse_file_space(ip, offset, len);
920 if (error)
921 goto out_unlock;
Namjae Jeona904b1c2015-03-25 15:08:56 +1100922 } else if (mode & FALLOC_FL_INSERT_RANGE) {
923 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
924
925 new_size = i_size_read(inode) + len;
926 if (offset & blksize_mask || len & blksize_mask) {
927 error = -EINVAL;
928 goto out_unlock;
929 }
930
931 /* check the new inode size does not wrap through zero */
932 if (new_size > inode->i_sb->s_maxbytes) {
933 error = -EFBIG;
934 goto out_unlock;
935 }
936
937 /* Offset should be less than i_size */
938 if (offset >= i_size_read(inode)) {
939 error = -EINVAL;
940 goto out_unlock;
941 }
942 do_file_insert = 1;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700943 } else {
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100944 flags |= XFS_PREALLOC_SET;
945
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700946 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
947 offset + len > i_size_read(inode)) {
948 new_size = offset + len;
Dave Chinner24513372014-06-25 14:58:08 +1000949 error = inode_newsize_ok(inode, new_size);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700950 if (error)
951 goto out_unlock;
952 }
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100953
Lukas Czerner376ba312014-03-13 19:07:58 +1100954 if (mode & FALLOC_FL_ZERO_RANGE)
955 error = xfs_zero_file_space(ip, offset, len);
956 else
957 error = xfs_alloc_file_space(ip, offset, len,
958 XFS_BMAPI_PREALLOC);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100959 if (error)
960 goto out_unlock;
961 }
962
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700963 if (file->f_flags & O_DSYNC)
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100964 flags |= XFS_PREALLOC_SYNC;
965
966 error = xfs_update_prealloc_flags(ip, flags);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100967 if (error)
968 goto out_unlock;
969
970 /* Change file size if needed */
971 if (new_size) {
972 struct iattr iattr;
973
974 iattr.ia_valid = ATTR_SIZE;
975 iattr.ia_size = new_size;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700976 error = xfs_setattr_size(ip, &iattr);
Namjae Jeona904b1c2015-03-25 15:08:56 +1100977 if (error)
978 goto out_unlock;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100979 }
980
Namjae Jeona904b1c2015-03-25 15:08:56 +1100981 /*
982 * Perform hole insertion now that the file size has been
983 * updated so that if we crash during the operation we don't
984 * leave shifted extents past EOF and hence losing access to
985 * the data that is contained within them.
986 */
987 if (do_file_insert)
988 error = xfs_insert_file_space(ip, offset, len);
989
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100990out_unlock:
Christoph Hellwig781355c2015-02-16 11:59:50 +1100991 xfs_iunlock(ip, iolock);
Dave Chinner24513372014-06-25 14:58:08 +1000992 return error;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100993}
994
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100997xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100999 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001001 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001003 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1004 return -EIO;
1005 return 0;
1006}
1007
1008STATIC int
1009xfs_dir_open(
1010 struct inode *inode,
1011 struct file *file)
1012{
1013 struct xfs_inode *ip = XFS_I(inode);
1014 int mode;
1015 int error;
1016
1017 error = xfs_file_open(inode, file);
1018 if (error)
1019 return error;
1020
1021 /*
1022 * If there are any blocks, read-ahead block 0 as we're almost
1023 * certain to have the next operation be a read there.
1024 */
Christoph Hellwig309ecac2013-12-06 12:30:09 -08001025 mode = xfs_ilock_data_map_shared(ip);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001026 if (ip->i_d.di_nextents > 0)
Eric Sandeen9df2dd02014-04-14 19:01:59 +10001027 xfs_dir3_data_readahead(ip, 0, -1);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001028 xfs_iunlock(ip, mode);
1029 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030}
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001033xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 struct inode *inode,
1035 struct file *filp)
1036{
Dave Chinner24513372014-06-25 14:58:08 +10001037 return xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038}
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001041xfs_file_readdir(
Al Virob8227552013-05-22 17:07:56 -04001042 struct file *file,
1043 struct dir_context *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
Al Virob8227552013-05-22 17:07:56 -04001045 struct inode *inode = file_inode(file);
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001046 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001047 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001049 /*
1050 * The Linux API doesn't pass down the total size of the buffer
1051 * we read into down to the filesystem. With the filldir concept
1052 * it's not needed for correct information, but the XFS dir2 leaf
1053 * code wants an estimate of the buffer size to calculate it's
1054 * readahead window and size the buffers used for mapping to
1055 * physical blocks.
1056 *
1057 * Try to give it an estimate that's good enough, maybe at some
1058 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +00001059 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001060 */
Eric Sandeena9cc7992010-02-03 17:50:13 +00001061 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
kbuild test robot83004752014-12-01 08:25:28 +11001063 return xfs_readdir(ip, ctx, bufsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064}
1065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001067xfs_file_mmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 struct file *filp,
1069 struct vm_area_struct *vma)
1070{
Nathan Scott3562fd42006-03-14 14:00:35 +11001071 vma->vm_ops = &xfs_file_vm_ops;
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001072
Nathan Scottfbc14622006-06-09 14:52:13 +10001073 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 return 0;
1075}
1076
David Chinner4f57dbc2007-07-19 16:28:17 +10001077/*
Jeff Liud126d432012-08-21 17:11:57 +08001078 * This type is designed to indicate the type of offset we would like
Eric Sandeen49c69592014-09-09 11:56:48 +10001079 * to search from page cache for xfs_seek_hole_data().
Jeff Liud126d432012-08-21 17:11:57 +08001080 */
1081enum {
1082 HOLE_OFF = 0,
1083 DATA_OFF,
1084};
1085
1086/*
1087 * Lookup the desired type of offset from the given page.
1088 *
1089 * On success, return true and the offset argument will point to the
1090 * start of the region that was found. Otherwise this function will
1091 * return false and keep the offset argument unchanged.
1092 */
1093STATIC bool
1094xfs_lookup_buffer_offset(
1095 struct page *page,
1096 loff_t *offset,
1097 unsigned int type)
1098{
1099 loff_t lastoff = page_offset(page);
1100 bool found = false;
1101 struct buffer_head *bh, *head;
1102
1103 bh = head = page_buffers(page);
1104 do {
1105 /*
1106 * Unwritten extents that have data in the page
1107 * cache covering them can be identified by the
1108 * BH_Unwritten state flag. Pages with multiple
1109 * buffers might have a mix of holes, data and
1110 * unwritten extents - any buffer with valid
1111 * data in it should have BH_Uptodate flag set
1112 * on it.
1113 */
1114 if (buffer_unwritten(bh) ||
1115 buffer_uptodate(bh)) {
1116 if (type == DATA_OFF)
1117 found = true;
1118 } else {
1119 if (type == HOLE_OFF)
1120 found = true;
1121 }
1122
1123 if (found) {
1124 *offset = lastoff;
1125 break;
1126 }
1127 lastoff += bh->b_size;
1128 } while ((bh = bh->b_this_page) != head);
1129
1130 return found;
1131}
1132
1133/*
1134 * This routine is called to find out and return a data or hole offset
1135 * from the page cache for unwritten extents according to the desired
Eric Sandeen49c69592014-09-09 11:56:48 +10001136 * type for xfs_seek_hole_data().
Jeff Liud126d432012-08-21 17:11:57 +08001137 *
1138 * The argument offset is used to tell where we start to search from the
1139 * page cache. Map is used to figure out the end points of the range to
1140 * lookup pages.
1141 *
1142 * Return true if the desired type of offset was found, and the argument
1143 * offset is filled with that address. Otherwise, return false and keep
1144 * offset unchanged.
1145 */
1146STATIC bool
1147xfs_find_get_desired_pgoff(
1148 struct inode *inode,
1149 struct xfs_bmbt_irec *map,
1150 unsigned int type,
1151 loff_t *offset)
1152{
1153 struct xfs_inode *ip = XFS_I(inode);
1154 struct xfs_mount *mp = ip->i_mount;
1155 struct pagevec pvec;
1156 pgoff_t index;
1157 pgoff_t end;
1158 loff_t endoff;
1159 loff_t startoff = *offset;
1160 loff_t lastoff = startoff;
1161 bool found = false;
1162
1163 pagevec_init(&pvec, 0);
1164
1165 index = startoff >> PAGE_CACHE_SHIFT;
1166 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1167 end = endoff >> PAGE_CACHE_SHIFT;
1168 do {
1169 int want;
1170 unsigned nr_pages;
1171 unsigned int i;
1172
1173 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1174 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1175 want);
1176 /*
1177 * No page mapped into given range. If we are searching holes
1178 * and if this is the first time we got into the loop, it means
1179 * that the given offset is landed in a hole, return it.
1180 *
1181 * If we have already stepped through some block buffers to find
1182 * holes but they all contains data. In this case, the last
1183 * offset is already updated and pointed to the end of the last
1184 * mapped page, if it does not reach the endpoint to search,
1185 * that means there should be a hole between them.
1186 */
1187 if (nr_pages == 0) {
1188 /* Data search found nothing */
1189 if (type == DATA_OFF)
1190 break;
1191
1192 ASSERT(type == HOLE_OFF);
1193 if (lastoff == startoff || lastoff < endoff) {
1194 found = true;
1195 *offset = lastoff;
1196 }
1197 break;
1198 }
1199
1200 /*
1201 * At lease we found one page. If this is the first time we
1202 * step into the loop, and if the first page index offset is
1203 * greater than the given search offset, a hole was found.
1204 */
1205 if (type == HOLE_OFF && lastoff == startoff &&
1206 lastoff < page_offset(pvec.pages[0])) {
1207 found = true;
1208 break;
1209 }
1210
1211 for (i = 0; i < nr_pages; i++) {
1212 struct page *page = pvec.pages[i];
1213 loff_t b_offset;
1214
1215 /*
1216 * At this point, the page may be truncated or
1217 * invalidated (changing page->mapping to NULL),
1218 * or even swizzled back from swapper_space to tmpfs
1219 * file mapping. However, page->index will not change
1220 * because we have a reference on the page.
1221 *
1222 * Searching done if the page index is out of range.
1223 * If the current offset is not reaches the end of
1224 * the specified search range, there should be a hole
1225 * between them.
1226 */
1227 if (page->index > end) {
1228 if (type == HOLE_OFF && lastoff < endoff) {
1229 *offset = lastoff;
1230 found = true;
1231 }
1232 goto out;
1233 }
1234
1235 lock_page(page);
1236 /*
1237 * Page truncated or invalidated(page->mapping == NULL).
1238 * We can freely skip it and proceed to check the next
1239 * page.
1240 */
1241 if (unlikely(page->mapping != inode->i_mapping)) {
1242 unlock_page(page);
1243 continue;
1244 }
1245
1246 if (!page_has_buffers(page)) {
1247 unlock_page(page);
1248 continue;
1249 }
1250
1251 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1252 if (found) {
1253 /*
1254 * The found offset may be less than the start
1255 * point to search if this is the first time to
1256 * come here.
1257 */
1258 *offset = max_t(loff_t, startoff, b_offset);
1259 unlock_page(page);
1260 goto out;
1261 }
1262
1263 /*
1264 * We either searching data but nothing was found, or
1265 * searching hole but found a data buffer. In either
1266 * case, probably the next page contains the desired
1267 * things, update the last offset to it so.
1268 */
1269 lastoff = page_offset(page) + PAGE_SIZE;
1270 unlock_page(page);
1271 }
1272
1273 /*
1274 * The number of returned pages less than our desired, search
1275 * done. In this case, nothing was found for searching data,
1276 * but we found a hole behind the last offset.
1277 */
1278 if (nr_pages < want) {
1279 if (type == HOLE_OFF) {
1280 *offset = lastoff;
1281 found = true;
1282 }
1283 break;
1284 }
1285
1286 index = pvec.pages[i - 1]->index + 1;
1287 pagevec_release(&pvec);
1288 } while (index <= end);
1289
1290out:
1291 pagevec_release(&pvec);
1292 return found;
1293}
1294
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001295STATIC loff_t
Eric Sandeen49c69592014-09-09 11:56:48 +10001296xfs_seek_hole_data(
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001297 struct file *file,
Eric Sandeen49c69592014-09-09 11:56:48 +10001298 loff_t start,
1299 int whence)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001300{
1301 struct inode *inode = file->f_mapping->host;
1302 struct xfs_inode *ip = XFS_I(inode);
1303 struct xfs_mount *mp = ip->i_mount;
1304 loff_t uninitialized_var(offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001305 xfs_fsize_t isize;
1306 xfs_fileoff_t fsbno;
Jeff Liub686d1f2012-08-21 17:12:18 +08001307 xfs_filblks_t end;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001308 uint lock;
1309 int error;
1310
1311 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001312 return -EIO;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001313
Christoph Hellwig309ecac2013-12-06 12:30:09 -08001314 lock = xfs_ilock_data_map_shared(ip);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001315
1316 isize = i_size_read(inode);
1317 if (start >= isize) {
Dave Chinner24513372014-06-25 14:58:08 +10001318 error = -ENXIO;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001319 goto out_unlock;
1320 }
1321
Eric Sandeen49c69592014-09-09 11:56:48 +10001322 /*
1323 * Try to read extents from the first block indicated
1324 * by fsbno to the end block of the file.
1325 */
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001326 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liub686d1f2012-08-21 17:12:18 +08001327 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001328
Jeff Liub686d1f2012-08-21 17:12:18 +08001329 for (;;) {
1330 struct xfs_bmbt_irec map[2];
1331 int nmap = 2;
1332 unsigned int i;
1333
1334 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1335 XFS_BMAPI_ENTIRE);
1336 if (error)
1337 goto out_unlock;
1338
1339 /* No extents at given offset, must be beyond EOF */
1340 if (nmap == 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001341 error = -ENXIO;
Jeff Liub686d1f2012-08-21 17:12:18 +08001342 goto out_unlock;
1343 }
1344
1345 for (i = 0; i < nmap; i++) {
1346 offset = max_t(loff_t, start,
1347 XFS_FSB_TO_B(mp, map[i].br_startoff));
1348
Eric Sandeen49c69592014-09-09 11:56:48 +10001349 /* Landed in the hole we wanted? */
1350 if (whence == SEEK_HOLE &&
1351 map[i].br_startblock == HOLESTARTBLOCK)
1352 goto out;
1353
1354 /* Landed in the data extent we wanted? */
1355 if (whence == SEEK_DATA &&
1356 (map[i].br_startblock == DELAYSTARTBLOCK ||
1357 (map[i].br_state == XFS_EXT_NORM &&
1358 !isnullstartblock(map[i].br_startblock))))
Jeff Liub686d1f2012-08-21 17:12:18 +08001359 goto out;
1360
1361 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001362 * Landed in an unwritten extent, try to search
1363 * for hole or data from page cache.
Jeff Liub686d1f2012-08-21 17:12:18 +08001364 */
1365 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1366 if (xfs_find_get_desired_pgoff(inode, &map[i],
Eric Sandeen49c69592014-09-09 11:56:48 +10001367 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1368 &offset))
Jeff Liub686d1f2012-08-21 17:12:18 +08001369 goto out;
1370 }
1371 }
1372
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001373 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001374 * We only received one extent out of the two requested. This
1375 * means we've hit EOF and didn't find what we are looking for.
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001376 */
Jeff Liub686d1f2012-08-21 17:12:18 +08001377 if (nmap == 1) {
Eric Sandeen49c69592014-09-09 11:56:48 +10001378 /*
1379 * If we were looking for a hole, set offset to
1380 * the end of the file (i.e., there is an implicit
1381 * hole at the end of any file).
1382 */
1383 if (whence == SEEK_HOLE) {
1384 offset = isize;
1385 break;
1386 }
1387 /*
1388 * If we were looking for data, it's nowhere to be found
1389 */
1390 ASSERT(whence == SEEK_DATA);
1391 error = -ENXIO;
1392 goto out_unlock;
Jeff Liub686d1f2012-08-21 17:12:18 +08001393 }
1394
1395 ASSERT(i > 1);
1396
1397 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001398 * Nothing was found, proceed to the next round of search
1399 * if the next reading offset is not at or beyond EOF.
Jeff Liub686d1f2012-08-21 17:12:18 +08001400 */
1401 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1402 start = XFS_FSB_TO_B(mp, fsbno);
1403 if (start >= isize) {
Eric Sandeen49c69592014-09-09 11:56:48 +10001404 if (whence == SEEK_HOLE) {
1405 offset = isize;
1406 break;
1407 }
1408 ASSERT(whence == SEEK_DATA);
1409 error = -ENXIO;
1410 goto out_unlock;
Jeff Liub686d1f2012-08-21 17:12:18 +08001411 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001412 }
1413
Jeff Liub686d1f2012-08-21 17:12:18 +08001414out:
1415 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001416 * If at this point we have found the hole we wanted, the returned
Jeff Liub686d1f2012-08-21 17:12:18 +08001417 * offset may be bigger than the file size as it may be aligned to
Eric Sandeen49c69592014-09-09 11:56:48 +10001418 * page boundary for unwritten extents. We need to deal with this
Jeff Liub686d1f2012-08-21 17:12:18 +08001419 * situation in particular.
1420 */
Eric Sandeen49c69592014-09-09 11:56:48 +10001421 if (whence == SEEK_HOLE)
1422 offset = min_t(loff_t, offset, isize);
Jie Liu46a1c2c2013-06-25 12:02:13 +08001423 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001424
1425out_unlock:
Christoph Hellwig01f4f322013-12-06 12:30:08 -08001426 xfs_iunlock(ip, lock);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001427
1428 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10001429 return error;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001430 return offset;
1431}
1432
1433STATIC loff_t
1434xfs_file_llseek(
1435 struct file *file,
1436 loff_t offset,
Eric Sandeen59f9c002014-09-09 11:57:10 +10001437 int whence)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001438{
Eric Sandeen59f9c002014-09-09 11:57:10 +10001439 switch (whence) {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001440 case SEEK_END:
1441 case SEEK_CUR:
1442 case SEEK_SET:
Eric Sandeen59f9c002014-09-09 11:57:10 +10001443 return generic_file_llseek(file, offset, whence);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001444 case SEEK_HOLE:
Eric Sandeen49c69592014-09-09 11:56:48 +10001445 case SEEK_DATA:
Eric Sandeen59f9c002014-09-09 11:57:10 +10001446 return xfs_seek_hole_data(file, offset, whence);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001447 default:
1448 return -EINVAL;
1449 }
1450}
1451
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001452/*
1453 * Locking for serialisation of IO during page faults. This results in a lock
1454 * ordering of:
1455 *
1456 * mmap_sem (MM)
1457 * i_mmap_lock (XFS - truncate serialisation)
1458 * page_lock (MM)
1459 * i_lock (XFS - extent map serialisation)
1460 */
1461STATIC int
1462xfs_filemap_fault(
1463 struct vm_area_struct *vma,
1464 struct vm_fault *vmf)
1465{
1466 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host);
1467 int error;
1468
1469 trace_xfs_filemap_fault(ip);
1470
1471 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1472 error = filemap_fault(vma, vmf);
1473 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1474
1475 return error;
1476}
1477
Dave Chinner075a9242015-02-23 21:44:54 +11001478/*
1479 * mmap()d file has taken write protection fault and is being made writable. We
1480 * can set the page state up correctly for a writable page, which means we can
1481 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1482 * mapping.
1483 */
1484STATIC int
1485xfs_filemap_page_mkwrite(
1486 struct vm_area_struct *vma,
1487 struct vm_fault *vmf)
1488{
1489 struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host);
1490 int error;
1491
1492 trace_xfs_filemap_page_mkwrite(ip);
1493
1494 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1495 error = block_page_mkwrite(vma, vmf, xfs_get_blocks);
1496 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1497
1498 return error;
1499}
1500
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001501const struct file_operations xfs_file_operations = {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001502 .llseek = xfs_file_llseek,
Al Virob4f5d2c2014-04-02 14:37:59 -04001503 .read_iter = xfs_file_read_iter,
Al Virobf97f3bc2014-04-03 14:20:23 -04001504 .write_iter = xfs_file_write_iter,
Nathan Scott1b895842006-03-31 13:08:59 +10001505 .splice_read = xfs_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001506 .splice_write = iter_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001507 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001509 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001511 .mmap = xfs_file_mmap,
1512 .open = xfs_file_open,
1513 .release = xfs_file_release,
1514 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001515 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516};
1517
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001518const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001519 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 .read = generic_read_dir,
Al Virob8227552013-05-22 17:07:56 -04001521 .iterate = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001522 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001523 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001524#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001525 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001526#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001527 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528};
1529
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001530static const struct vm_operations_struct xfs_file_vm_ops = {
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001531 .fault = xfs_filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07001532 .map_pages = filemap_map_pages,
Dave Chinner075a9242015-02-23 21:44:54 +11001533 .page_mkwrite = xfs_filemap_page_mkwrite,
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001534};