blob: ed95e5bb04e692b614983ea83bf870ac373a9b0c [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"
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +100040#include "xfs_iomap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010043#include <linux/falloc.h>
Jeff Liud126d432012-08-21 17:11:57 +080044#include <linux/pagevec.h>
Tejun Heo66114ca2015-05-22 17:13:32 -040045#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040047static const struct vm_operations_struct xfs_file_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Christoph Hellwigdda35b82010-02-15 09:44:46 +000049/*
Dave Chinner487f84f2011-01-12 11:37:10 +110050 * Locking primitives for read and write IO paths to ensure we consistently use
51 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
52 */
53static inline void
54xfs_rw_ilock(
55 struct xfs_inode *ip,
56 int type)
57{
58 if (type & XFS_IOLOCK_EXCL)
Al Viro59551022016-01-22 15:40:57 -050059 inode_lock(VFS_I(ip));
Dave Chinner487f84f2011-01-12 11:37:10 +110060 xfs_ilock(ip, type);
61}
62
63static inline void
64xfs_rw_iunlock(
65 struct xfs_inode *ip,
66 int type)
67{
68 xfs_iunlock(ip, type);
69 if (type & XFS_IOLOCK_EXCL)
Al Viro59551022016-01-22 15:40:57 -050070 inode_unlock(VFS_I(ip));
Dave Chinner487f84f2011-01-12 11:37:10 +110071}
72
73static inline void
74xfs_rw_ilock_demote(
75 struct xfs_inode *ip,
76 int type)
77{
78 xfs_ilock_demote(ip, type);
79 if (type & XFS_IOLOCK_EXCL)
Al Viro59551022016-01-22 15:40:57 -050080 inode_unlock(VFS_I(ip));
Dave Chinner487f84f2011-01-12 11:37:10 +110081}
82
83/*
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +100084 * Clear the specified ranges to zero through either the pagecache or DAX.
85 * Holes and unwritten extents will be left as-is as they already are zeroed.
Christoph Hellwigdda35b82010-02-15 09:44:46 +000086 */
Dave Chinneref9d8732012-11-29 15:26:33 +110087int
Christoph Hellwig7bb41db2016-06-21 09:56:26 +100088xfs_zero_range(
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +100089 struct xfs_inode *ip,
Christoph Hellwig7bb41db2016-06-21 09:56:26 +100090 xfs_off_t pos,
91 xfs_off_t count,
92 bool *did_zero)
Christoph Hellwigdda35b82010-02-15 09:44:46 +000093{
Christoph Hellwig459f0fb2016-06-21 09:55:18 +100094 return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops);
Christoph Hellwigdda35b82010-02-15 09:44:46 +000095}
96
Christoph Hellwig8add71c2015-02-02 09:53:56 +110097int
98xfs_update_prealloc_flags(
99 struct xfs_inode *ip,
100 enum xfs_prealloc_flags flags)
101{
102 struct xfs_trans *tp;
103 int error;
104
Christoph Hellwig253f4912016-04-06 09:19:55 +1000105 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
106 0, 0, 0, &tp);
107 if (error)
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100108 return error;
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100109
110 xfs_ilock(ip, XFS_ILOCK_EXCL);
111 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
112
113 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100114 VFS_I(ip)->i_mode &= ~S_ISUID;
115 if (VFS_I(ip)->i_mode & S_IXGRP)
116 VFS_I(ip)->i_mode &= ~S_ISGID;
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100117 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
118 }
119
120 if (flags & XFS_PREALLOC_SET)
121 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
122 if (flags & XFS_PREALLOC_CLEAR)
123 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
124
125 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
126 if (flags & XFS_PREALLOC_SYNC)
127 xfs_trans_set_sync(tp);
Christoph Hellwig70393312015-06-04 13:48:08 +1000128 return xfs_trans_commit(tp);
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100129}
130
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000131/*
132 * Fsync operations on directories are much simpler than on regular files,
133 * as there is no file data to flush, and thus also no need for explicit
134 * cache flush operations, and there are no non-transaction metadata updates
135 * on directories either.
136 */
137STATIC int
138xfs_dir_fsync(
139 struct file *file,
140 loff_t start,
141 loff_t end,
142 int datasync)
143{
144 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
145 struct xfs_mount *mp = ip->i_mount;
146 xfs_lsn_t lsn = 0;
147
148 trace_xfs_dir_fsync(ip);
149
150 xfs_ilock(ip, XFS_ILOCK_SHARED);
151 if (xfs_ipincount(ip))
152 lsn = ip->i_itemp->ili_last_lsn;
153 xfs_iunlock(ip, XFS_ILOCK_SHARED);
154
155 if (!lsn)
156 return 0;
Dave Chinner24513372014-06-25 14:58:08 +1000157 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000158}
159
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000160STATIC int
161xfs_file_fsync(
162 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400163 loff_t start,
164 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000165 int datasync)
166{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200167 struct inode *inode = file->f_mapping->host;
168 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000169 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000170 int error = 0;
171 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000172 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000173
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000174 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000175
Josef Bacik02c24a82011-07-16 20:44:56 -0400176 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
177 if (error)
178 return error;
179
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000180 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000181 return -EIO;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000182
183 xfs_iflags_clear(ip, XFS_ITRUNCATED);
184
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000185 if (mp->m_flags & XFS_MOUNT_BARRIER) {
186 /*
187 * If we have an RT and/or log subvolume we need to make sure
188 * to flush the write cache the device used for file data
189 * first. This is to ensure newly written file data make
190 * it to disk before logging the new inode size in case of
191 * an extending write.
192 */
193 if (XFS_IS_REALTIME_INODE(ip))
194 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
195 else if (mp->m_logdev_targp != mp->m_ddev_targp)
196 xfs_blkdev_issue_flush(mp->m_ddev_targp);
197 }
198
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000199 /*
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100200 * All metadata updates are logged, which means that we just have to
201 * flush the log up to the latest LSN that touched the inode. If we have
202 * concurrent fsync/fdatasync() calls, we need them to all block on the
203 * log force before we clear the ili_fsync_fields field. This ensures
204 * that we don't get a racing sync operation that does not wait for the
205 * metadata to hit the journal before returning. If we race with
206 * clearing the ili_fsync_fields, then all that will happen is the log
207 * force will do nothing as the lsn will already be on disk. We can't
208 * race with setting ili_fsync_fields because that is done under
209 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
210 * until after the ili_fsync_fields is cleared.
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000211 */
212 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000213 if (xfs_ipincount(ip)) {
214 if (!datasync ||
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100215 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000216 lsn = ip->i_itemp->ili_last_lsn;
217 }
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000218
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100219 if (lsn) {
Christoph Hellwigb1037052011-09-19 14:55:51 +0000220 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100221 ip->i_itemp->ili_fsync_fields = 0;
222 }
223 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigb1037052011-09-19 14:55:51 +0000224
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000225 /*
226 * If we only have a single device, and the log force about was
227 * a no-op we might have to flush the data device cache here.
228 * This can only happen for fdatasync/O_DSYNC if we were overwriting
229 * an already allocated file and thus do not have any metadata to
230 * commit.
231 */
232 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
233 mp->m_logdev_targp == mp->m_ddev_targp &&
234 !XFS_IS_REALTIME_INODE(ip) &&
235 !log_flushed)
236 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000237
Dave Chinner24513372014-06-25 14:58:08 +1000238 return error;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000239}
240
Christoph Hellwig00258e32010-02-15 09:44:47 +0000241STATIC ssize_t
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000242xfs_file_dio_aio_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000243 struct kiocb *iocb,
Al Virob4f5d2c2014-04-02 14:37:59 -0400244 struct iov_iter *to)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000245{
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000246 struct address_space *mapping = iocb->ki_filp->f_mapping;
247 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000248 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000249 loff_t isize = i_size_read(inode);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000250 size_t count = iov_iter_count(to);
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000251 struct iov_iter data;
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000252 struct xfs_buftarg *target;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000253 ssize_t ret = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000254
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000255 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000256
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000257 if (!count)
258 return 0; /* skip atime */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000259
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000260 if (XFS_IS_REALTIME_INODE(ip))
261 target = ip->i_mount->m_rtdev_targp;
262 else
263 target = ip->i_mount->m_ddev_targp;
264
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000265 /* DIO must be aligned to device logical sector size */
266 if ((iocb->ki_pos | count) & target->bt_logical_sectormask) {
267 if (iocb->ki_pos == isize)
268 return 0;
269 return -EINVAL;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000270 }
271
Dave Chinner0c38a252011-08-25 07:17:01 +0000272 /*
Brian Foster3d751af2015-08-19 10:35:04 +1000273 * Locking is a bit tricky here. If we take an exclusive lock for direct
274 * IO, we effectively serialise all new concurrent read IO to this file
275 * and block it behind IO that is currently in progress because IO in
276 * progress holds the IO lock shared. We only need to hold the lock
277 * exclusive to blow away the page cache, so only take lock exclusively
278 * if the page cache needs invalidation. This allows the normal direct
279 * IO case of no page cache pages to proceeed concurrently without
280 * serialisation.
Dave Chinner0c38a252011-08-25 07:17:01 +0000281 */
282 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000283 if (mapping->nrpages) {
Dave Chinner0c38a252011-08-25 07:17:01 +0000284 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100285 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
286
Brian Foster3d751af2015-08-19 10:35:04 +1000287 /*
288 * The generic dio code only flushes the range of the particular
289 * I/O. Because we take an exclusive lock here, this whole
290 * sequence is considerably more expensive for us. This has a
291 * noticeable performance impact for any file with cached pages,
292 * even when outside of the range of the particular I/O.
293 *
294 * Hence, amortize the cost of the lock against a full file
295 * flush and reduce the chances of repeated iolock cycles going
296 * forward.
297 */
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000298 if (mapping->nrpages) {
299 ret = filemap_write_and_wait(mapping);
Dave Chinner487f84f2011-01-12 11:37:10 +1100300 if (ret) {
301 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
302 return ret;
303 }
Chris Mason85e584d2014-09-02 12:12:52 +1000304
305 /*
306 * Invalidate whole pages. This can return an error if
307 * we fail to invalidate a page, but this should never
308 * happen on XFS. Warn if it does fail.
309 */
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000310 ret = invalidate_inode_pages2(mapping);
Chris Mason85e584d2014-09-02 12:12:52 +1000311 WARN_ON_ONCE(ret);
312 ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000313 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100314 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000315 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000316
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000317 data = *to;
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000318 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
319 xfs_get_blocks_direct, NULL, NULL, 0);
320 if (ret > 0) {
321 iocb->ki_pos += ret;
322 iov_iter_advance(to, ret);
Christoph Hellwigfa8d9722016-07-20 11:38:01 +1000323 }
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000324 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000325
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000326 file_accessed(iocb->ki_filp);
327 return ret;
328}
329
Arnd Bergmannf021bd02016-07-22 09:50:55 +1000330static noinline ssize_t
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000331xfs_file_dax_read(
332 struct kiocb *iocb,
333 struct iov_iter *to)
334{
335 struct address_space *mapping = iocb->ki_filp->f_mapping;
336 struct inode *inode = mapping->host;
337 struct xfs_inode *ip = XFS_I(inode);
338 struct iov_iter data = *to;
339 size_t count = iov_iter_count(to);
340 ssize_t ret = 0;
341
342 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
343
344 if (!count)
345 return 0; /* skip atime */
346
347 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
348 ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, NULL, 0);
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000349 if (ret > 0) {
350 iocb->ki_pos += ret;
351 iov_iter_advance(to, ret);
352 }
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000353 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
354
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000355 file_accessed(iocb->ki_filp);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000356 return ret;
357}
358
359STATIC ssize_t
360xfs_file_buffered_aio_read(
361 struct kiocb *iocb,
362 struct iov_iter *to)
363{
364 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
365 ssize_t ret;
366
367 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
368
369 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Al Virob4f5d2c2014-04-02 14:37:59 -0400370 ret = generic_file_read_iter(iocb, to);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000371 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
372
373 return ret;
374}
375
376STATIC ssize_t
377xfs_file_read_iter(
378 struct kiocb *iocb,
379 struct iov_iter *to)
380{
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000381 struct inode *inode = file_inode(iocb->ki_filp);
382 struct xfs_mount *mp = XFS_I(inode)->i_mount;
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000383 ssize_t ret = 0;
384
385 XFS_STATS_INC(mp, xs_read_calls);
386
387 if (XFS_FORCED_SHUTDOWN(mp))
388 return -EIO;
389
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000390 if (IS_DAX(inode))
391 ret = xfs_file_dax_read(iocb, to);
392 else if (iocb->ki_flags & IOCB_DIRECT)
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000393 ret = xfs_file_dio_aio_read(iocb, to);
394 else
395 ret = xfs_file_buffered_aio_read(iocb, to);
396
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000397 if (ret > 0)
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100398 XFS_STATS_ADD(mp, xs_read_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000399 return ret;
400}
401
Christoph Hellwig00258e32010-02-15 09:44:47 +0000402STATIC ssize_t
403xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000404 struct file *infilp,
405 loff_t *ppos,
406 struct pipe_inode_info *pipe,
407 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000408 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000409{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000410 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000411 ssize_t ret;
412
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100413 XFS_STATS_INC(ip->i_mount, xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000414
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000415 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
416 return -EIO;
417
Christoph Hellwig3176c3e2016-07-20 11:31:42 +1000418 trace_xfs_file_splice_read(ip, count, *ppos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000419
Dave Chinnera6d76362016-01-04 16:28:25 +1100420 /*
421 * DAX inodes cannot ues the page cache for splice, so we have to push
422 * them through the VFS IO path. This means it goes through
423 * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we
424 * cannot lock the splice operation at this level for DAX inodes.
425 */
426 if (IS_DAX(VFS_I(ip))) {
427 ret = default_file_splice_read(infilp, ppos, pipe, count,
428 flags);
429 goto out;
430 }
431
432 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
433 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
434 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
435out:
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000436 if (ret > 0)
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100437 XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000438 return ret;
439}
440
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100441/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400442 * Zero any on disk space between the current EOF and the new, larger EOF.
443 *
444 * This handles the normal case of zeroing the remainder of the last block in
445 * the file and the unusual case of zeroing blocks out beyond the size of the
446 * file. This second case only happens with fixed size extents and when the
447 * system crashes before the inode size was updated but after blocks were
448 * allocated.
449 *
450 * Expects the iolock to be held exclusive, and will take the ilock internally.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000451 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000452int /* error (positive) */
453xfs_zero_eof(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400454 struct xfs_inode *ip,
455 xfs_off_t offset, /* starting I/O offset */
Dave Chinner5885ebd2015-02-23 22:37:08 +1100456 xfs_fsize_t isize, /* current inode size */
457 bool *did_zeroing)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000458{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400459 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000460 ASSERT(offset > isize);
461
Brian Foster0a50f162015-10-12 16:02:08 +1100462 trace_xfs_zero_eof(ip, isize, offset - isize);
Christoph Hellwig570b6212016-06-21 09:57:26 +1000463 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000464}
465
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100466/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100467 * Common pre-write limit and setup checks.
468 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000469 * Called with the iolocked held either shared and exclusive according to
470 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
471 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100472 */
473STATIC ssize_t
474xfs_file_aio_write_checks(
Al Viro99733fa2015-04-07 14:25:18 -0400475 struct kiocb *iocb,
476 struct iov_iter *from,
Dave Chinner4d8d1582011-01-11 10:23:42 +1100477 int *iolock)
478{
Al Viro99733fa2015-04-07 14:25:18 -0400479 struct file *file = iocb->ki_filp;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100480 struct inode *inode = file->f_mapping->host;
481 struct xfs_inode *ip = XFS_I(inode);
Al Viro3309dd02015-04-09 12:55:47 -0400482 ssize_t error = 0;
Al Viro99733fa2015-04-07 14:25:18 -0400483 size_t count = iov_iter_count(from);
Brian Foster3136e8b2015-10-12 16:02:05 +1100484 bool drained_dio = false;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100485
Dave Chinner7271d242011-08-25 07:17:02 +0000486restart:
Al Viro3309dd02015-04-09 12:55:47 -0400487 error = generic_write_checks(iocb, from);
488 if (error <= 0)
Dave Chinner4d8d1582011-01-11 10:23:42 +1100489 return error;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100490
Christoph Hellwig21c3ea12015-04-13 11:38:29 +1000491 error = xfs_break_layouts(inode, iolock, true);
Christoph Hellwig781355c2015-02-16 11:59:50 +1100492 if (error)
493 return error;
494
Jan Karaa6de82c2015-05-21 16:05:56 +0200495 /* For changing security info in file_remove_privs() we need i_mutex */
496 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
497 xfs_rw_iunlock(ip, *iolock);
498 *iolock = XFS_IOLOCK_EXCL;
499 xfs_rw_ilock(ip, *iolock);
500 goto restart;
501 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100502 /*
503 * If the offset is beyond the size of the file, we need to zero any
504 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000505 * write. If zeroing is needed and we are currently holding the
Christoph Hellwig467f7892012-03-27 10:34:47 -0400506 * iolock shared, we need to update it to exclusive which implies
507 * having to redo all checks before.
Dave Chinnerb9d59842015-04-16 22:03:07 +1000508 *
509 * We need to serialise against EOF updates that occur in IO
510 * completions here. We want to make sure that nobody is changing the
511 * size while we do this check until we have placed an IO barrier (i.e.
512 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
513 * The spinlock effectively forms a memory barrier once we have the
514 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
515 * and hence be able to correctly determine if we need to run zeroing.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100516 */
Dave Chinnerb9d59842015-04-16 22:03:07 +1000517 spin_lock(&ip->i_flags_lock);
Al Viro99733fa2015-04-07 14:25:18 -0400518 if (iocb->ki_pos > i_size_read(inode)) {
Dave Chinner5885ebd2015-02-23 22:37:08 +1100519 bool zero = false;
520
Dave Chinnerb9d59842015-04-16 22:03:07 +1000521 spin_unlock(&ip->i_flags_lock);
Brian Foster3136e8b2015-10-12 16:02:05 +1100522 if (!drained_dio) {
523 if (*iolock == XFS_IOLOCK_SHARED) {
524 xfs_rw_iunlock(ip, *iolock);
525 *iolock = XFS_IOLOCK_EXCL;
526 xfs_rw_ilock(ip, *iolock);
527 iov_iter_reexpand(from, count);
528 }
Dave Chinner40c63fb2015-04-16 22:03:17 +1000529 /*
530 * We now have an IO submission barrier in place, but
531 * AIO can do EOF updates during IO completion and hence
532 * we now need to wait for all of them to drain. Non-AIO
533 * DIO will have drained before we are given the
534 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
535 * no-op.
536 */
537 inode_dio_wait(inode);
Brian Foster3136e8b2015-10-12 16:02:05 +1100538 drained_dio = true;
Dave Chinner7271d242011-08-25 07:17:02 +0000539 goto restart;
540 }
Al Viro99733fa2015-04-07 14:25:18 -0400541 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
Christoph Hellwig467f7892012-03-27 10:34:47 -0400542 if (error)
543 return error;
Dave Chinnerb9d59842015-04-16 22:03:07 +1000544 } else
545 spin_unlock(&ip->i_flags_lock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100546
547 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000548 * Updating the timestamps will grab the ilock again from
549 * xfs_fs_dirty_inode, so we have to call it after dropping the
550 * lock above. Eventually we should look into a way to avoid
551 * the pointless lock roundtrip.
552 */
Josef Bacikc3b2da32012-03-26 09:59:21 -0400553 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
554 error = file_update_time(file);
555 if (error)
556 return error;
557 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000558
559 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100560 * If we're writing the file then make sure to clear the setuid and
561 * setgid bits if the process is not being run by root. This keeps
562 * people from modifying setuid and setgid binaries.
563 */
Jan Karaa6de82c2015-05-21 16:05:56 +0200564 if (!IS_NOSEC(inode))
565 return file_remove_privs(file);
566 return 0;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100567}
568
569/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100570 * xfs_file_dio_aio_write - handle direct IO writes
571 *
572 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100573 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100574 * follow locking changes and looping.
575 *
Dave Chinnereda77982011-01-11 10:22:40 +1100576 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
577 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
578 * pages are flushed out.
579 *
580 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
581 * allowing them to be done in parallel with reads and other direct IO writes.
582 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
583 * needs to do sub-block zeroing and that requires serialisation against other
584 * direct IOs to the same block. In this case we need to serialise the
585 * submission of the unaligned IOs so that we don't get racing block zeroing in
586 * the dio layer. To avoid the problem with aio, we also need to wait for
587 * outstanding IOs to complete so that unwritten extent conversion is completed
588 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000589 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100590 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100591 * Returns with locks held indicated by @iolock and errors indicated by
592 * negative return values.
593 */
594STATIC ssize_t
595xfs_file_dio_aio_write(
596 struct kiocb *iocb,
Al Virob3188912014-04-02 07:06:30 -0400597 struct iov_iter *from)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100598{
599 struct file *file = iocb->ki_filp;
600 struct address_space *mapping = file->f_mapping;
601 struct inode *inode = mapping->host;
602 struct xfs_inode *ip = XFS_I(inode);
603 struct xfs_mount *mp = ip->i_mount;
604 ssize_t ret = 0;
Dave Chinnereda77982011-01-11 10:22:40 +1100605 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000606 int iolock;
Al Virob3188912014-04-02 07:06:30 -0400607 size_t count = iov_iter_count(from);
Dave Chinner0cefb292015-04-16 22:03:27 +1000608 loff_t end;
609 struct iov_iter data;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100610 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
611 mp->m_rtdev_targp : mp->m_ddev_targp;
612
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600613 /* DIO must be aligned to device logical sector size */
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000614 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000615 return -EINVAL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100616
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600617 /* "unaligned" here means not aligned to a filesystem block */
Christoph Hellwig13712712016-04-07 08:51:57 -0700618 if ((iocb->ki_pos & mp->m_blockmask) ||
619 ((iocb->ki_pos + count) & mp->m_blockmask))
Dave Chinnereda77982011-01-11 10:22:40 +1100620 unaligned_io = 1;
621
Dave Chinner7271d242011-08-25 07:17:02 +0000622 /*
623 * We don't need to take an exclusive lock unless there page cache needs
624 * to be invalidated or unaligned IO is being executed. We don't need to
625 * consider the EOF extension case here because
626 * xfs_file_aio_write_checks() will relock the inode as necessary for
627 * EOF zeroing cases and fill out the new inode size as appropriate.
628 */
629 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000630 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100631 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000632 iolock = XFS_IOLOCK_SHARED;
633 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000634
635 /*
636 * Recheck if there are cached pages that need invalidate after we got
637 * the iolock to protect against other threads adding new pages while
638 * we were waiting for the iolock.
639 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000640 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
641 xfs_rw_iunlock(ip, iolock);
642 iolock = XFS_IOLOCK_EXCL;
643 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000644 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100645
Al Viro99733fa2015-04-07 14:25:18 -0400646 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100647 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000648 goto out;
Al Viro99733fa2015-04-07 14:25:18 -0400649 count = iov_iter_count(from);
Christoph Hellwig13712712016-04-07 08:51:57 -0700650 end = iocb->ki_pos + count - 1;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100651
Brian Foster3d751af2015-08-19 10:35:04 +1000652 /*
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000653 * See xfs_file_dio_aio_read() for why we do a full-file flush here.
Brian Foster3d751af2015-08-19 10:35:04 +1000654 */
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100655 if (mapping->nrpages) {
Brian Foster3d751af2015-08-19 10:35:04 +1000656 ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100657 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000658 goto out;
Dave Chinner834ffca2014-09-02 12:12:52 +1000659 /*
Brian Foster3d751af2015-08-19 10:35:04 +1000660 * Invalidate whole pages. This can return an error if we fail
661 * to invalidate a page, but this should never happen on XFS.
662 * Warn if it does fail.
Dave Chinner834ffca2014-09-02 12:12:52 +1000663 */
Brian Foster3d751af2015-08-19 10:35:04 +1000664 ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
Dave Chinner834ffca2014-09-02 12:12:52 +1000665 WARN_ON_ONCE(ret);
666 ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100667 }
668
Dave Chinnereda77982011-01-11 10:22:40 +1100669 /*
670 * If we are doing unaligned IO, wait for all other IO to drain,
671 * otherwise demote the lock if we had to flush cached pages
672 */
673 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000674 inode_dio_wait(inode);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000675 else if (iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100676 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000677 iolock = XFS_IOLOCK_SHARED;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100678 }
679
Christoph Hellwig3176c3e2016-07-20 11:31:42 +1000680 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100681
Dave Chinner0cefb292015-04-16 22:03:27 +1000682 data = *from;
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000683 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
684 xfs_get_blocks_direct, xfs_end_io_direct_write,
685 NULL, DIO_ASYNC_EXTEND);
Dave Chinner0cefb292015-04-16 22:03:27 +1000686
687 /* see generic_file_direct_write() for why this is necessary */
688 if (mapping->nrpages) {
689 invalidate_inode_pages2_range(mapping,
Christoph Hellwig13712712016-04-07 08:51:57 -0700690 iocb->ki_pos >> PAGE_SHIFT,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300691 end >> PAGE_SHIFT);
Dave Chinner0cefb292015-04-16 22:03:27 +1000692 }
693
694 if (ret > 0) {
Christoph Hellwig13712712016-04-07 08:51:57 -0700695 iocb->ki_pos += ret;
Dave Chinner0cefb292015-04-16 22:03:27 +1000696 iov_iter_advance(from, ret);
Dave Chinner0cefb292015-04-16 22:03:27 +1000697 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000698out:
699 xfs_rw_iunlock(ip, iolock);
700
Dave Chinner6b698ed2015-06-04 09:18:53 +1000701 /*
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000702 * No fallback to buffered IO on errors for XFS, direct IO will either
703 * complete fully or fail.
Dave Chinner6b698ed2015-06-04 09:18:53 +1000704 */
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000705 ASSERT(ret < 0 || ret == count);
706 return ret;
707}
708
Arnd Bergmannf021bd02016-07-22 09:50:55 +1000709static noinline ssize_t
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000710xfs_file_dax_write(
711 struct kiocb *iocb,
712 struct iov_iter *from)
713{
714 struct address_space *mapping = iocb->ki_filp->f_mapping;
715 struct inode *inode = mapping->host;
716 struct xfs_inode *ip = XFS_I(inode);
717 struct xfs_mount *mp = ip->i_mount;
718 ssize_t ret = 0;
719 int unaligned_io = 0;
720 int iolock;
721 struct iov_iter data;
722
723 /* "unaligned" here means not aligned to a filesystem block */
724 if ((iocb->ki_pos & mp->m_blockmask) ||
725 ((iocb->ki_pos + iov_iter_count(from)) & mp->m_blockmask)) {
726 unaligned_io = 1;
727 iolock = XFS_IOLOCK_EXCL;
728 } else if (mapping->nrpages) {
729 iolock = XFS_IOLOCK_EXCL;
730 } else {
731 iolock = XFS_IOLOCK_SHARED;
732 }
733 xfs_rw_ilock(ip, iolock);
734
735 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
736 if (ret)
737 goto out;
738
739 /*
740 * Yes, even DAX files can have page cache attached to them: A zeroed
741 * page is inserted into the pagecache when we have to serve a write
742 * fault on a hole. It should never be dirtied and can simply be
743 * dropped from the pagecache once we get real data for the page.
744 */
745 if (mapping->nrpages) {
746 ret = invalidate_inode_pages2(mapping);
747 WARN_ON_ONCE(ret);
748 }
749
750 if (iolock == XFS_IOLOCK_EXCL && !unaligned_io) {
751 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
752 iolock = XFS_IOLOCK_SHARED;
753 }
754
755 trace_xfs_file_dax_write(ip, iov_iter_count(from), iocb->ki_pos);
756
757 data = *from;
758 ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct,
759 xfs_end_io_direct_write, 0);
760 if (ret > 0) {
761 iocb->ki_pos += ret;
762 iov_iter_advance(from, ret);
763 }
764out:
765 xfs_rw_iunlock(ip, iolock);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100766 return ret;
767}
768
Christoph Hellwig00258e32010-02-15 09:44:47 +0000769STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100770xfs_file_buffered_aio_write(
771 struct kiocb *iocb,
Al Virob3188912014-04-02 07:06:30 -0400772 struct iov_iter *from)
Dave Chinner637bbc72011-01-11 10:17:30 +1100773{
774 struct file *file = iocb->ki_filp;
775 struct address_space *mapping = file->f_mapping;
776 struct inode *inode = mapping->host;
777 struct xfs_inode *ip = XFS_I(inode);
778 ssize_t ret;
779 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000780 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100781
Christoph Hellwigd0606462011-12-18 20:00:14 +0000782 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100783
Al Viro99733fa2015-04-07 14:25:18 -0400784 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100785 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000786 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100787
788 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100789 current->backing_dev_info = inode_to_bdi(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100790
791write_retry:
Christoph Hellwig3176c3e2016-07-20 11:31:42 +1000792 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000793 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
Al Viro0a64bc22014-02-11 22:25:22 -0500794 if (likely(ret >= 0))
Al Viro99733fa2015-04-07 14:25:18 -0400795 iocb->ki_pos += ret;
Brian Fosterdc06f3982014-07-24 19:49:28 +1000796
Dave Chinner637bbc72011-01-11 10:17:30 +1100797 /*
Brian Fosterdc06f3982014-07-24 19:49:28 +1000798 * If we hit a space limit, try to free up some lingering preallocated
799 * space before returning an error. In the case of ENOSPC, first try to
800 * write back all dirty inodes to free up some of the excess reserved
801 * metadata space. This reduces the chances that the eofblocks scan
802 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
803 * also behaves as a filter to prevent too many eofblocks scans from
804 * running at the same time.
Dave Chinner637bbc72011-01-11 10:17:30 +1100805 */
Brian Fosterdc06f3982014-07-24 19:49:28 +1000806 if (ret == -EDQUOT && !enospc) {
807 enospc = xfs_inode_free_quota_eofblocks(ip);
808 if (enospc)
809 goto write_retry;
810 } else if (ret == -ENOSPC && !enospc) {
811 struct xfs_eofblocks eofb = {0};
812
Dave Chinner637bbc72011-01-11 10:17:30 +1100813 enospc = 1;
Dave Chinner9aa05002012-10-08 21:56:04 +1100814 xfs_flush_inodes(ip->i_mount);
Brian Fosterdc06f3982014-07-24 19:49:28 +1000815 eofb.eof_scan_owner = ip->i_ino; /* for locking */
816 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
817 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
Dave Chinner9aa05002012-10-08 21:56:04 +1100818 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100819 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000820
Dave Chinner637bbc72011-01-11 10:17:30 +1100821 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000822out:
823 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100824 return ret;
825}
826
827STATIC ssize_t
Al Virobf97f3bc2014-04-03 14:20:23 -0400828xfs_file_write_iter(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000829 struct kiocb *iocb,
Al Virobf97f3bc2014-04-03 14:20:23 -0400830 struct iov_iter *from)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000831{
832 struct file *file = iocb->ki_filp;
833 struct address_space *mapping = file->f_mapping;
834 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000835 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100836 ssize_t ret;
Al Virobf97f3bc2014-04-03 14:20:23 -0400837 size_t ocount = iov_iter_count(from);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000838
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100839 XFS_STATS_INC(ip->i_mount, xs_write_calls);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000840
Dave Chinner637bbc72011-01-11 10:17:30 +1100841 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000842 return 0;
843
Al Virobf97f3bc2014-04-03 14:20:23 -0400844 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
845 return -EIO;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000846
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000847 if (IS_DAX(inode))
848 ret = xfs_file_dax_write(iocb, from);
849 else if (iocb->ki_flags & IOCB_DIRECT)
Al Virobf97f3bc2014-04-03 14:20:23 -0400850 ret = xfs_file_dio_aio_write(iocb, from);
Dave Chinner637bbc72011-01-11 10:17:30 +1100851 else
Al Virobf97f3bc2014-04-03 14:20:23 -0400852 ret = xfs_file_buffered_aio_write(iocb, from);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000853
Christoph Hellwigd0606462011-12-18 20:00:14 +0000854 if (ret > 0) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100855 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000856
Christoph Hellwigd0606462011-12-18 20:00:14 +0000857 /* Handle various SYNC-type writes */
Christoph Hellwige2592212016-04-07 08:52:01 -0700858 ret = generic_write_sync(iocb, ret);
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
David Chinner4f57dbc2007-07-19 16:28:17 +10001066/*
Jeff Liud126d432012-08-21 17:11:57 +08001067 * This type is designed to indicate the type of offset we would like
Eric Sandeen49c69592014-09-09 11:56:48 +10001068 * to search from page cache for xfs_seek_hole_data().
Jeff Liud126d432012-08-21 17:11:57 +08001069 */
1070enum {
1071 HOLE_OFF = 0,
1072 DATA_OFF,
1073};
1074
1075/*
1076 * Lookup the desired type of offset from the given page.
1077 *
1078 * On success, return true and the offset argument will point to the
1079 * start of the region that was found. Otherwise this function will
1080 * return false and keep the offset argument unchanged.
1081 */
1082STATIC bool
1083xfs_lookup_buffer_offset(
1084 struct page *page,
1085 loff_t *offset,
1086 unsigned int type)
1087{
1088 loff_t lastoff = page_offset(page);
1089 bool found = false;
1090 struct buffer_head *bh, *head;
1091
1092 bh = head = page_buffers(page);
1093 do {
1094 /*
1095 * Unwritten extents that have data in the page
1096 * cache covering them can be identified by the
1097 * BH_Unwritten state flag. Pages with multiple
1098 * buffers might have a mix of holes, data and
1099 * unwritten extents - any buffer with valid
1100 * data in it should have BH_Uptodate flag set
1101 * on it.
1102 */
1103 if (buffer_unwritten(bh) ||
1104 buffer_uptodate(bh)) {
1105 if (type == DATA_OFF)
1106 found = true;
1107 } else {
1108 if (type == HOLE_OFF)
1109 found = true;
1110 }
1111
1112 if (found) {
1113 *offset = lastoff;
1114 break;
1115 }
1116 lastoff += bh->b_size;
1117 } while ((bh = bh->b_this_page) != head);
1118
1119 return found;
1120}
1121
1122/*
1123 * This routine is called to find out and return a data or hole offset
1124 * from the page cache for unwritten extents according to the desired
Eric Sandeen49c69592014-09-09 11:56:48 +10001125 * type for xfs_seek_hole_data().
Jeff Liud126d432012-08-21 17:11:57 +08001126 *
1127 * The argument offset is used to tell where we start to search from the
1128 * page cache. Map is used to figure out the end points of the range to
1129 * lookup pages.
1130 *
1131 * Return true if the desired type of offset was found, and the argument
1132 * offset is filled with that address. Otherwise, return false and keep
1133 * offset unchanged.
1134 */
1135STATIC bool
1136xfs_find_get_desired_pgoff(
1137 struct inode *inode,
1138 struct xfs_bmbt_irec *map,
1139 unsigned int type,
1140 loff_t *offset)
1141{
1142 struct xfs_inode *ip = XFS_I(inode);
1143 struct xfs_mount *mp = ip->i_mount;
1144 struct pagevec pvec;
1145 pgoff_t index;
1146 pgoff_t end;
1147 loff_t endoff;
1148 loff_t startoff = *offset;
1149 loff_t lastoff = startoff;
1150 bool found = false;
1151
1152 pagevec_init(&pvec, 0);
1153
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001154 index = startoff >> PAGE_SHIFT;
Jeff Liud126d432012-08-21 17:11:57 +08001155 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001156 end = endoff >> PAGE_SHIFT;
Jeff Liud126d432012-08-21 17:11:57 +08001157 do {
1158 int want;
1159 unsigned nr_pages;
1160 unsigned int i;
1161
1162 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1163 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1164 want);
1165 /*
1166 * No page mapped into given range. If we are searching holes
1167 * and if this is the first time we got into the loop, it means
1168 * that the given offset is landed in a hole, return it.
1169 *
1170 * If we have already stepped through some block buffers to find
1171 * holes but they all contains data. In this case, the last
1172 * offset is already updated and pointed to the end of the last
1173 * mapped page, if it does not reach the endpoint to search,
1174 * that means there should be a hole between them.
1175 */
1176 if (nr_pages == 0) {
1177 /* Data search found nothing */
1178 if (type == DATA_OFF)
1179 break;
1180
1181 ASSERT(type == HOLE_OFF);
1182 if (lastoff == startoff || lastoff < endoff) {
1183 found = true;
1184 *offset = lastoff;
1185 }
1186 break;
1187 }
1188
1189 /*
1190 * At lease we found one page. If this is the first time we
1191 * step into the loop, and if the first page index offset is
1192 * greater than the given search offset, a hole was found.
1193 */
1194 if (type == HOLE_OFF && lastoff == startoff &&
1195 lastoff < page_offset(pvec.pages[0])) {
1196 found = true;
1197 break;
1198 }
1199
1200 for (i = 0; i < nr_pages; i++) {
1201 struct page *page = pvec.pages[i];
1202 loff_t b_offset;
1203
1204 /*
1205 * At this point, the page may be truncated or
1206 * invalidated (changing page->mapping to NULL),
1207 * or even swizzled back from swapper_space to tmpfs
1208 * file mapping. However, page->index will not change
1209 * because we have a reference on the page.
1210 *
1211 * Searching done if the page index is out of range.
1212 * If the current offset is not reaches the end of
1213 * the specified search range, there should be a hole
1214 * between them.
1215 */
1216 if (page->index > end) {
1217 if (type == HOLE_OFF && lastoff < endoff) {
1218 *offset = lastoff;
1219 found = true;
1220 }
1221 goto out;
1222 }
1223
1224 lock_page(page);
1225 /*
1226 * Page truncated or invalidated(page->mapping == NULL).
1227 * We can freely skip it and proceed to check the next
1228 * page.
1229 */
1230 if (unlikely(page->mapping != inode->i_mapping)) {
1231 unlock_page(page);
1232 continue;
1233 }
1234
1235 if (!page_has_buffers(page)) {
1236 unlock_page(page);
1237 continue;
1238 }
1239
1240 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1241 if (found) {
1242 /*
1243 * The found offset may be less than the start
1244 * point to search if this is the first time to
1245 * come here.
1246 */
1247 *offset = max_t(loff_t, startoff, b_offset);
1248 unlock_page(page);
1249 goto out;
1250 }
1251
1252 /*
1253 * We either searching data but nothing was found, or
1254 * searching hole but found a data buffer. In either
1255 * case, probably the next page contains the desired
1256 * things, update the last offset to it so.
1257 */
1258 lastoff = page_offset(page) + PAGE_SIZE;
1259 unlock_page(page);
1260 }
1261
1262 /*
1263 * The number of returned pages less than our desired, search
1264 * done. In this case, nothing was found for searching data,
1265 * but we found a hole behind the last offset.
1266 */
1267 if (nr_pages < want) {
1268 if (type == HOLE_OFF) {
1269 *offset = lastoff;
1270 found = true;
1271 }
1272 break;
1273 }
1274
1275 index = pvec.pages[i - 1]->index + 1;
1276 pagevec_release(&pvec);
1277 } while (index <= end);
1278
1279out:
1280 pagevec_release(&pvec);
1281 return found;
1282}
1283
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001284/*
1285 * caller must lock inode with xfs_ilock_data_map_shared,
1286 * can we craft an appropriate ASSERT?
1287 *
1288 * end is because the VFS-level lseek interface is defined such that any
1289 * offset past i_size shall return -ENXIO, but we use this for quota code
1290 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1291 */
1292loff_t
1293__xfs_seek_hole_data(
1294 struct inode *inode,
Eric Sandeen49c69592014-09-09 11:56:48 +10001295 loff_t start,
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001296 loff_t end,
Eric Sandeen49c69592014-09-09 11:56:48 +10001297 int whence)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001298{
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001299 struct xfs_inode *ip = XFS_I(inode);
1300 struct xfs_mount *mp = ip->i_mount;
1301 loff_t uninitialized_var(offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001302 xfs_fileoff_t fsbno;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001303 xfs_filblks_t lastbno;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001304 int error;
1305
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001306 if (start >= end) {
Dave Chinner24513372014-06-25 14:58:08 +10001307 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001308 goto out_error;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001309 }
1310
Eric Sandeen49c69592014-09-09 11:56:48 +10001311 /*
1312 * Try to read extents from the first block indicated
1313 * by fsbno to the end block of the file.
1314 */
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001315 fsbno = XFS_B_TO_FSBT(mp, start);
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001316 lastbno = XFS_B_TO_FSB(mp, end);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001317
Jeff Liub686d1f2012-08-21 17:12:18 +08001318 for (;;) {
1319 struct xfs_bmbt_irec map[2];
1320 int nmap = 2;
1321 unsigned int i;
1322
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001323 error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap,
Jeff Liub686d1f2012-08-21 17:12:18 +08001324 XFS_BMAPI_ENTIRE);
1325 if (error)
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001326 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001327
1328 /* No extents at given offset, must be beyond EOF */
1329 if (nmap == 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001330 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001331 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001332 }
1333
1334 for (i = 0; i < nmap; i++) {
1335 offset = max_t(loff_t, start,
1336 XFS_FSB_TO_B(mp, map[i].br_startoff));
1337
Eric Sandeen49c69592014-09-09 11:56:48 +10001338 /* Landed in the hole we wanted? */
1339 if (whence == SEEK_HOLE &&
1340 map[i].br_startblock == HOLESTARTBLOCK)
1341 goto out;
1342
1343 /* Landed in the data extent we wanted? */
1344 if (whence == SEEK_DATA &&
1345 (map[i].br_startblock == DELAYSTARTBLOCK ||
1346 (map[i].br_state == XFS_EXT_NORM &&
1347 !isnullstartblock(map[i].br_startblock))))
Jeff Liub686d1f2012-08-21 17:12:18 +08001348 goto out;
1349
1350 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001351 * Landed in an unwritten extent, try to search
1352 * for hole or data from page cache.
Jeff Liub686d1f2012-08-21 17:12:18 +08001353 */
1354 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1355 if (xfs_find_get_desired_pgoff(inode, &map[i],
Eric Sandeen49c69592014-09-09 11:56:48 +10001356 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1357 &offset))
Jeff Liub686d1f2012-08-21 17:12:18 +08001358 goto out;
1359 }
1360 }
1361
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001362 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001363 * We only received one extent out of the two requested. This
1364 * means we've hit EOF and didn't find what we are looking for.
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001365 */
Jeff Liub686d1f2012-08-21 17:12:18 +08001366 if (nmap == 1) {
Eric Sandeen49c69592014-09-09 11:56:48 +10001367 /*
1368 * If we were looking for a hole, set offset to
1369 * the end of the file (i.e., there is an implicit
1370 * hole at the end of any file).
1371 */
1372 if (whence == SEEK_HOLE) {
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001373 offset = end;
Eric Sandeen49c69592014-09-09 11:56:48 +10001374 break;
1375 }
1376 /*
1377 * If we were looking for data, it's nowhere to be found
1378 */
1379 ASSERT(whence == SEEK_DATA);
1380 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001381 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001382 }
1383
1384 ASSERT(i > 1);
1385
1386 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001387 * Nothing was found, proceed to the next round of search
1388 * if the next reading offset is not at or beyond EOF.
Jeff Liub686d1f2012-08-21 17:12:18 +08001389 */
1390 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1391 start = XFS_FSB_TO_B(mp, fsbno);
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001392 if (start >= end) {
Eric Sandeen49c69592014-09-09 11:56:48 +10001393 if (whence == SEEK_HOLE) {
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001394 offset = end;
Eric Sandeen49c69592014-09-09 11:56:48 +10001395 break;
1396 }
1397 ASSERT(whence == SEEK_DATA);
1398 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001399 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001400 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001401 }
1402
Jeff Liub686d1f2012-08-21 17:12:18 +08001403out:
1404 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001405 * If at this point we have found the hole we wanted, the returned
Jeff Liub686d1f2012-08-21 17:12:18 +08001406 * offset may be bigger than the file size as it may be aligned to
Eric Sandeen49c69592014-09-09 11:56:48 +10001407 * page boundary for unwritten extents. We need to deal with this
Jeff Liub686d1f2012-08-21 17:12:18 +08001408 * situation in particular.
1409 */
Eric Sandeen49c69592014-09-09 11:56:48 +10001410 if (whence == SEEK_HOLE)
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001411 offset = min_t(loff_t, offset, end);
1412
1413 return offset;
1414
1415out_error:
1416 return error;
1417}
1418
1419STATIC loff_t
1420xfs_seek_hole_data(
1421 struct file *file,
1422 loff_t start,
1423 int whence)
1424{
1425 struct inode *inode = file->f_mapping->host;
1426 struct xfs_inode *ip = XFS_I(inode);
1427 struct xfs_mount *mp = ip->i_mount;
1428 uint lock;
1429 loff_t offset, end;
1430 int error = 0;
1431
1432 if (XFS_FORCED_SHUTDOWN(mp))
1433 return -EIO;
1434
1435 lock = xfs_ilock_data_map_shared(ip);
1436
1437 end = i_size_read(inode);
1438 offset = __xfs_seek_hole_data(inode, start, end, whence);
1439 if (offset < 0) {
1440 error = offset;
1441 goto out_unlock;
1442 }
1443
Jie Liu46a1c2c2013-06-25 12:02:13 +08001444 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001445
1446out_unlock:
Christoph Hellwig01f4f322013-12-06 12:30:08 -08001447 xfs_iunlock(ip, lock);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001448
1449 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10001450 return error;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001451 return offset;
1452}
1453
1454STATIC loff_t
1455xfs_file_llseek(
1456 struct file *file,
1457 loff_t offset,
Eric Sandeen59f9c002014-09-09 11:57:10 +10001458 int whence)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001459{
Eric Sandeen59f9c002014-09-09 11:57:10 +10001460 switch (whence) {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001461 case SEEK_END:
1462 case SEEK_CUR:
1463 case SEEK_SET:
Eric Sandeen59f9c002014-09-09 11:57:10 +10001464 return generic_file_llseek(file, offset, whence);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001465 case SEEK_HOLE:
Eric Sandeen49c69592014-09-09 11:56:48 +10001466 case SEEK_DATA:
Eric Sandeen59f9c002014-09-09 11:57:10 +10001467 return xfs_seek_hole_data(file, offset, whence);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001468 default:
1469 return -EINVAL;
1470 }
1471}
1472
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001473/*
1474 * Locking for serialisation of IO during page faults. This results in a lock
1475 * ordering of:
1476 *
1477 * mmap_sem (MM)
Dave Chinner6b698ed2015-06-04 09:18:53 +10001478 * sb_start_pagefault(vfs, freeze)
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001479 * i_mmaplock (XFS - truncate serialisation)
Dave Chinner6b698ed2015-06-04 09:18:53 +10001480 * page_lock (MM)
1481 * i_lock (XFS - extent map serialisation)
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001482 */
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001483
Dave Chinner075a9242015-02-23 21:44:54 +11001484/*
1485 * mmap()d file has taken write protection fault and is being made writable. We
1486 * can set the page state up correctly for a writable page, which means we can
1487 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1488 * mapping.
1489 */
1490STATIC int
1491xfs_filemap_page_mkwrite(
1492 struct vm_area_struct *vma,
1493 struct vm_fault *vmf)
1494{
Dave Chinner6b698ed2015-06-04 09:18:53 +10001495 struct inode *inode = file_inode(vma->vm_file);
Dave Chinnerec56b1f2015-06-04 09:18:18 +10001496 int ret;
Dave Chinner075a9242015-02-23 21:44:54 +11001497
Dave Chinner6b698ed2015-06-04 09:18:53 +10001498 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
Dave Chinner075a9242015-02-23 21:44:54 +11001499
Dave Chinner6b698ed2015-06-04 09:18:53 +10001500 sb_start_pagefault(inode->i_sb);
Dave Chinnerec56b1f2015-06-04 09:18:18 +10001501 file_update_time(vma->vm_file);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001502 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1503
1504 if (IS_DAX(inode)) {
Ross Zwisler6b524992016-07-26 15:21:05 -07001505 ret = dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001506 } else {
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001507 ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001508 ret = block_page_mkwrite_return(ret);
1509 }
1510
1511 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1512 sb_end_pagefault(inode->i_sb);
1513
1514 return ret;
1515}
1516
1517STATIC int
1518xfs_filemap_fault(
1519 struct vm_area_struct *vma,
1520 struct vm_fault *vmf)
1521{
Dave Chinnerb2442c52015-07-29 11:48:00 +10001522 struct inode *inode = file_inode(vma->vm_file);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001523 int ret;
1524
Dave Chinnerb2442c52015-07-29 11:48:00 +10001525 trace_xfs_filemap_fault(XFS_I(inode));
Dave Chinner6b698ed2015-06-04 09:18:53 +10001526
1527 /* DAX can shortcut the normal fault path on write faults! */
Dave Chinnerb2442c52015-07-29 11:48:00 +10001528 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
Dave Chinner6b698ed2015-06-04 09:18:53 +10001529 return xfs_filemap_page_mkwrite(vma, vmf);
Dave Chinner075a9242015-02-23 21:44:54 +11001530
Dave Chinnerb2442c52015-07-29 11:48:00 +10001531 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1532 if (IS_DAX(inode)) {
1533 /*
1534 * we do not want to trigger unwritten extent conversion on read
1535 * faults - that is unnecessary overhead and would also require
1536 * changes to xfs_get_blocks_direct() to map unwritten extent
1537 * ioend for conversion on read-only mappings.
1538 */
Ross Zwisler6b524992016-07-26 15:21:05 -07001539 ret = dax_fault(vma, vmf, xfs_get_blocks_dax_fault);
Dave Chinnerb2442c52015-07-29 11:48:00 +10001540 } else
1541 ret = filemap_fault(vma, vmf);
1542 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
Dave Chinner075a9242015-02-23 21:44:54 +11001543
Dave Chinner6b698ed2015-06-04 09:18:53 +10001544 return ret;
1545}
1546
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001547/*
1548 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1549 * both read and write faults. Hence we need to handle both cases. There is no
1550 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1551 * handle both cases here. @flags carries the information on the type of fault
1552 * occuring.
1553 */
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001554STATIC int
1555xfs_filemap_pmd_fault(
1556 struct vm_area_struct *vma,
1557 unsigned long addr,
1558 pmd_t *pmd,
1559 unsigned int flags)
1560{
1561 struct inode *inode = file_inode(vma->vm_file);
1562 struct xfs_inode *ip = XFS_I(inode);
1563 int ret;
1564
1565 if (!IS_DAX(inode))
1566 return VM_FAULT_FALLBACK;
1567
1568 trace_xfs_filemap_pmd_fault(ip);
1569
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001570 if (flags & FAULT_FLAG_WRITE) {
1571 sb_start_pagefault(inode->i_sb);
1572 file_update_time(vma->vm_file);
1573 }
1574
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001575 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
Ross Zwisler6b524992016-07-26 15:21:05 -07001576 ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault);
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001577 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001578
1579 if (flags & FAULT_FLAG_WRITE)
1580 sb_end_pagefault(inode->i_sb);
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001581
1582 return ret;
1583}
1584
Dave Chinner3af49282015-11-03 12:37:02 +11001585/*
1586 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1587 * updates on write faults. In reality, it's need to serialise against
Ross Zwisler5eb88dc2016-01-22 15:10:56 -08001588 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1589 * to ensure we serialise the fault barrier in place.
Dave Chinner3af49282015-11-03 12:37:02 +11001590 */
1591static int
1592xfs_filemap_pfn_mkwrite(
1593 struct vm_area_struct *vma,
1594 struct vm_fault *vmf)
1595{
1596
1597 struct inode *inode = file_inode(vma->vm_file);
1598 struct xfs_inode *ip = XFS_I(inode);
1599 int ret = VM_FAULT_NOPAGE;
1600 loff_t size;
1601
1602 trace_xfs_filemap_pfn_mkwrite(ip);
1603
1604 sb_start_pagefault(inode->i_sb);
1605 file_update_time(vma->vm_file);
1606
1607 /* check if the faulting page hasn't raced with truncate */
1608 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1609 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1610 if (vmf->pgoff >= size)
1611 ret = VM_FAULT_SIGBUS;
Ross Zwisler5eb88dc2016-01-22 15:10:56 -08001612 else if (IS_DAX(inode))
1613 ret = dax_pfn_mkwrite(vma, vmf);
Dave Chinner3af49282015-11-03 12:37:02 +11001614 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1615 sb_end_pagefault(inode->i_sb);
1616 return ret;
1617
1618}
1619
Dave Chinner6b698ed2015-06-04 09:18:53 +10001620static const struct vm_operations_struct xfs_file_vm_ops = {
1621 .fault = xfs_filemap_fault,
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001622 .pmd_fault = xfs_filemap_pmd_fault,
Dave Chinner6b698ed2015-06-04 09:18:53 +10001623 .map_pages = filemap_map_pages,
1624 .page_mkwrite = xfs_filemap_page_mkwrite,
Dave Chinner3af49282015-11-03 12:37:02 +11001625 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
Dave Chinner6b698ed2015-06-04 09:18:53 +10001626};
1627
1628STATIC int
1629xfs_file_mmap(
1630 struct file *filp,
1631 struct vm_area_struct *vma)
1632{
1633 file_accessed(filp);
1634 vma->vm_ops = &xfs_file_vm_ops;
1635 if (IS_DAX(file_inode(filp)))
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001636 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
Dave Chinner6b698ed2015-06-04 09:18:53 +10001637 return 0;
Dave Chinner075a9242015-02-23 21:44:54 +11001638}
1639
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001640const struct file_operations xfs_file_operations = {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001641 .llseek = xfs_file_llseek,
Al Virob4f5d2c2014-04-02 14:37:59 -04001642 .read_iter = xfs_file_read_iter,
Al Virobf97f3bc2014-04-03 14:20:23 -04001643 .write_iter = xfs_file_write_iter,
Nathan Scott1b895842006-03-31 13:08:59 +10001644 .splice_read = xfs_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001645 .splice_write = iter_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001646 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001648 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001650 .mmap = xfs_file_mmap,
1651 .open = xfs_file_open,
1652 .release = xfs_file_release,
1653 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001654 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655};
1656
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001657const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001658 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 .read = generic_read_dir,
Al Viro3b0a3c12016-04-20 23:42:46 -04001660 .iterate_shared = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001661 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001662 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001663#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001664 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001665#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001666 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667};