Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * published by the Free Software Foundation. |
| 8 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include "xfs.h" |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 19 | #include "xfs_fs.h" |
Dave Chinner | 70a9883 | 2013-10-23 10:36:05 +1100 | [diff] [blame] | 20 | #include "xfs_shared.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 21 | #include "xfs_format.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 22 | #include "xfs_log_format.h" |
| 23 | #include "xfs_trans_resv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "xfs_mount.h" |
Dave Chinner | 5706278 | 2013-10-15 09:17:51 +1100 | [diff] [blame] | 25 | #include "xfs_da_format.h" |
| 26 | #include "xfs_da_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include "xfs_inode.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 28 | #include "xfs_trans.h" |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 29 | #include "xfs_inode_item.h" |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 30 | #include "xfs_bmap.h" |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 31 | #include "xfs_bmap_util.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include "xfs_error.h" |
Dave Chinner | 2b9ab5a | 2013-08-12 20:49:37 +1000 | [diff] [blame] | 33 | #include "xfs_dir2.h" |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 34 | #include "xfs_dir2_priv.h" |
Christoph Hellwig | ddcd856 | 2008-12-03 07:55:34 -0500 | [diff] [blame] | 35 | #include "xfs_ioctl.h" |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 36 | #include "xfs_trace.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 37 | #include "xfs_log.h" |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 38 | #include "xfs_icache.h" |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 39 | #include "xfs_pnfs.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | |
| 41 | #include <linux/dcache.h> |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 42 | #include <linux/falloc.h> |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 43 | #include <linux/pagevec.h> |
Tejun Heo | 66114ca | 2015-05-22 17:13:32 -0400 | [diff] [blame] | 44 | #include <linux/backing-dev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | |
Alexey Dobriyan | f0f37e2 | 2009-09-27 22:29:37 +0400 | [diff] [blame] | 46 | static const struct vm_operations_struct xfs_file_vm_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 48 | /* |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 49 | * Locking primitives for read and write IO paths to ensure we consistently use |
| 50 | * and order the inode->i_mutex, ip->i_lock and ip->i_iolock. |
| 51 | */ |
| 52 | static inline void |
| 53 | xfs_rw_ilock( |
| 54 | struct xfs_inode *ip, |
| 55 | int type) |
| 56 | { |
| 57 | if (type & XFS_IOLOCK_EXCL) |
| 58 | mutex_lock(&VFS_I(ip)->i_mutex); |
| 59 | xfs_ilock(ip, type); |
| 60 | } |
| 61 | |
| 62 | static inline void |
| 63 | xfs_rw_iunlock( |
| 64 | struct xfs_inode *ip, |
| 65 | int type) |
| 66 | { |
| 67 | xfs_iunlock(ip, type); |
| 68 | if (type & XFS_IOLOCK_EXCL) |
| 69 | mutex_unlock(&VFS_I(ip)->i_mutex); |
| 70 | } |
| 71 | |
| 72 | static inline void |
| 73 | xfs_rw_ilock_demote( |
| 74 | struct xfs_inode *ip, |
| 75 | int type) |
| 76 | { |
| 77 | xfs_ilock_demote(ip, type); |
| 78 | if (type & XFS_IOLOCK_EXCL) |
| 79 | mutex_unlock(&VFS_I(ip)->i_mutex); |
| 80 | } |
| 81 | |
| 82 | /* |
Dave Chinner | 4f69f57 | 2015-06-04 09:19:08 +1000 | [diff] [blame] | 83 | * xfs_iozero clears the specified range supplied via the page cache (except in |
| 84 | * the DAX case). Writes through the page cache will allocate blocks over holes, |
| 85 | * though the callers usually map the holes first and avoid them. If a block is |
| 86 | * not completely zeroed, then it will be read from disk before being partially |
| 87 | * zeroed. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 88 | * |
Dave Chinner | 4f69f57 | 2015-06-04 09:19:08 +1000 | [diff] [blame] | 89 | * In the DAX case, we can just directly write to the underlying pages. This |
| 90 | * will not allocate blocks, but will avoid holes and unwritten extents and so |
| 91 | * not do unnecessary work. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 92 | */ |
Dave Chinner | ef9d873 | 2012-11-29 15:26:33 +1100 | [diff] [blame] | 93 | int |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 94 | xfs_iozero( |
| 95 | struct xfs_inode *ip, /* inode */ |
| 96 | loff_t pos, /* offset in file */ |
| 97 | size_t count) /* size of data to zero */ |
| 98 | { |
| 99 | struct page *page; |
| 100 | struct address_space *mapping; |
Dave Chinner | 4f69f57 | 2015-06-04 09:19:08 +1000 | [diff] [blame] | 101 | int status = 0; |
| 102 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 103 | |
| 104 | mapping = VFS_I(ip)->i_mapping; |
| 105 | do { |
| 106 | unsigned offset, bytes; |
| 107 | void *fsdata; |
| 108 | |
| 109 | offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ |
| 110 | bytes = PAGE_CACHE_SIZE - offset; |
| 111 | if (bytes > count) |
| 112 | bytes = count; |
| 113 | |
Dave Chinner | 4f69f57 | 2015-06-04 09:19:08 +1000 | [diff] [blame] | 114 | if (IS_DAX(VFS_I(ip))) { |
| 115 | status = dax_zero_page_range(VFS_I(ip), pos, bytes, |
| 116 | xfs_get_blocks_direct); |
| 117 | if (status) |
| 118 | break; |
| 119 | } else { |
| 120 | status = pagecache_write_begin(NULL, mapping, pos, bytes, |
| 121 | AOP_FLAG_UNINTERRUPTIBLE, |
| 122 | &page, &fsdata); |
| 123 | if (status) |
| 124 | break; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 125 | |
Dave Chinner | 4f69f57 | 2015-06-04 09:19:08 +1000 | [diff] [blame] | 126 | zero_user(page, offset, bytes); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 127 | |
Dave Chinner | 4f69f57 | 2015-06-04 09:19:08 +1000 | [diff] [blame] | 128 | status = pagecache_write_end(NULL, mapping, pos, bytes, |
| 129 | bytes, page, fsdata); |
| 130 | WARN_ON(status <= 0); /* can't return less than zero! */ |
| 131 | status = 0; |
| 132 | } |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 133 | pos += bytes; |
| 134 | count -= bytes; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 135 | } while (count); |
| 136 | |
Dave Chinner | cddc116 | 2015-05-29 07:40:32 +1000 | [diff] [blame] | 137 | return status; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 138 | } |
| 139 | |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 140 | int |
| 141 | xfs_update_prealloc_flags( |
| 142 | struct xfs_inode *ip, |
| 143 | enum xfs_prealloc_flags flags) |
| 144 | { |
| 145 | struct xfs_trans *tp; |
| 146 | int error; |
| 147 | |
| 148 | tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID); |
| 149 | error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0); |
| 150 | if (error) { |
Christoph Hellwig | 4906e21 | 2015-06-04 13:47:56 +1000 | [diff] [blame] | 151 | xfs_trans_cancel(tp); |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 152 | return error; |
| 153 | } |
| 154 | |
| 155 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 156 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
| 157 | |
| 158 | if (!(flags & XFS_PREALLOC_INVISIBLE)) { |
| 159 | ip->i_d.di_mode &= ~S_ISUID; |
| 160 | if (ip->i_d.di_mode & S_IXGRP) |
| 161 | ip->i_d.di_mode &= ~S_ISGID; |
| 162 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 163 | } |
| 164 | |
| 165 | if (flags & XFS_PREALLOC_SET) |
| 166 | ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC; |
| 167 | if (flags & XFS_PREALLOC_CLEAR) |
| 168 | ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC; |
| 169 | |
| 170 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 171 | if (flags & XFS_PREALLOC_SYNC) |
| 172 | xfs_trans_set_sync(tp); |
Christoph Hellwig | 7039331 | 2015-06-04 13:48:08 +1000 | [diff] [blame] | 173 | return xfs_trans_commit(tp); |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 174 | } |
| 175 | |
Christoph Hellwig | 1da2f2d | 2011-10-02 14:25:16 +0000 | [diff] [blame] | 176 | /* |
| 177 | * Fsync operations on directories are much simpler than on regular files, |
| 178 | * as there is no file data to flush, and thus also no need for explicit |
| 179 | * cache flush operations, and there are no non-transaction metadata updates |
| 180 | * on directories either. |
| 181 | */ |
| 182 | STATIC int |
| 183 | xfs_dir_fsync( |
| 184 | struct file *file, |
| 185 | loff_t start, |
| 186 | loff_t end, |
| 187 | int datasync) |
| 188 | { |
| 189 | struct xfs_inode *ip = XFS_I(file->f_mapping->host); |
| 190 | struct xfs_mount *mp = ip->i_mount; |
| 191 | xfs_lsn_t lsn = 0; |
| 192 | |
| 193 | trace_xfs_dir_fsync(ip); |
| 194 | |
| 195 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
| 196 | if (xfs_ipincount(ip)) |
| 197 | lsn = ip->i_itemp->ili_last_lsn; |
| 198 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
| 199 | |
| 200 | if (!lsn) |
| 201 | return 0; |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 202 | return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL); |
Christoph Hellwig | 1da2f2d | 2011-10-02 14:25:16 +0000 | [diff] [blame] | 203 | } |
| 204 | |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 205 | STATIC int |
| 206 | xfs_file_fsync( |
| 207 | struct file *file, |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 208 | loff_t start, |
| 209 | loff_t end, |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 210 | int datasync) |
| 211 | { |
Christoph Hellwig | 7ea8085 | 2010-05-26 17:53:25 +0200 | [diff] [blame] | 212 | struct inode *inode = file->f_mapping->host; |
| 213 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | a27a263 | 2011-06-16 12:02:23 +0000 | [diff] [blame] | 214 | struct xfs_mount *mp = ip->i_mount; |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 215 | int error = 0; |
| 216 | int log_flushed = 0; |
Christoph Hellwig | b103705 | 2011-09-19 14:55:51 +0000 | [diff] [blame] | 217 | xfs_lsn_t lsn = 0; |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 218 | |
Christoph Hellwig | cca28fb | 2010-06-24 11:57:09 +1000 | [diff] [blame] | 219 | trace_xfs_file_fsync(ip); |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 220 | |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 221 | error = filemap_write_and_wait_range(inode->i_mapping, start, end); |
| 222 | if (error) |
| 223 | return error; |
| 224 | |
Christoph Hellwig | a27a263 | 2011-06-16 12:02:23 +0000 | [diff] [blame] | 225 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 226 | return -EIO; |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 227 | |
| 228 | xfs_iflags_clear(ip, XFS_ITRUNCATED); |
| 229 | |
Christoph Hellwig | a27a263 | 2011-06-16 12:02:23 +0000 | [diff] [blame] | 230 | if (mp->m_flags & XFS_MOUNT_BARRIER) { |
| 231 | /* |
| 232 | * If we have an RT and/or log subvolume we need to make sure |
| 233 | * to flush the write cache the device used for file data |
| 234 | * first. This is to ensure newly written file data make |
| 235 | * it to disk before logging the new inode size in case of |
| 236 | * an extending write. |
| 237 | */ |
| 238 | if (XFS_IS_REALTIME_INODE(ip)) |
| 239 | xfs_blkdev_issue_flush(mp->m_rtdev_targp); |
| 240 | else if (mp->m_logdev_targp != mp->m_ddev_targp) |
| 241 | xfs_blkdev_issue_flush(mp->m_ddev_targp); |
| 242 | } |
| 243 | |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 244 | /* |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 245 | * All metadata updates are logged, which means that we just have |
| 246 | * to flush the log up to the latest LSN that touched the inode. |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 247 | */ |
| 248 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | 8f639dd | 2012-02-29 09:53:55 +0000 | [diff] [blame] | 249 | if (xfs_ipincount(ip)) { |
| 250 | if (!datasync || |
| 251 | (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP)) |
| 252 | lsn = ip->i_itemp->ili_last_lsn; |
| 253 | } |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 254 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 255 | |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 256 | if (lsn) |
Christoph Hellwig | b103705 | 2011-09-19 14:55:51 +0000 | [diff] [blame] | 257 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
| 258 | |
Christoph Hellwig | a27a263 | 2011-06-16 12:02:23 +0000 | [diff] [blame] | 259 | /* |
| 260 | * If we only have a single device, and the log force about was |
| 261 | * a no-op we might have to flush the data device cache here. |
| 262 | * This can only happen for fdatasync/O_DSYNC if we were overwriting |
| 263 | * an already allocated file and thus do not have any metadata to |
| 264 | * commit. |
| 265 | */ |
| 266 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && |
| 267 | mp->m_logdev_targp == mp->m_ddev_targp && |
| 268 | !XFS_IS_REALTIME_INODE(ip) && |
| 269 | !log_flushed) |
| 270 | xfs_blkdev_issue_flush(mp->m_ddev_targp); |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 271 | |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 272 | return error; |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 273 | } |
| 274 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 275 | STATIC ssize_t |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 276 | xfs_file_read_iter( |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 277 | struct kiocb *iocb, |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 278 | struct iov_iter *to) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 279 | { |
| 280 | struct file *file = iocb->ki_filp; |
| 281 | struct inode *inode = file->f_mapping->host; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 282 | struct xfs_inode *ip = XFS_I(inode); |
| 283 | struct xfs_mount *mp = ip->i_mount; |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 284 | size_t size = iov_iter_count(to); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 285 | ssize_t ret = 0; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 286 | int ioflags = 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 287 | xfs_fsize_t n; |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 288 | loff_t pos = iocb->ki_pos; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 289 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 290 | XFS_STATS_INC(xs_read_calls); |
| 291 | |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 292 | if (unlikely(iocb->ki_flags & IOCB_DIRECT)) |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 293 | ioflags |= XFS_IO_ISDIRECT; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 294 | if (file->f_mode & FMODE_NOCMTIME) |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 295 | ioflags |= XFS_IO_INVIS; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 296 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 297 | if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) { |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 298 | xfs_buftarg_t *target = |
| 299 | XFS_IS_REALTIME_INODE(ip) ? |
| 300 | mp->m_rtdev_targp : mp->m_ddev_targp; |
Eric Sandeen | 7c71ee7 | 2014-01-21 16:46:23 -0600 | [diff] [blame] | 301 | /* DIO must be aligned to device logical sector size */ |
| 302 | if ((pos | size) & target->bt_logical_sectormask) { |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 303 | if (pos == i_size_read(inode)) |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 304 | return 0; |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 305 | return -EINVAL; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 306 | } |
| 307 | } |
| 308 | |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 309 | n = mp->m_super->s_maxbytes - pos; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 310 | if (n <= 0 || size == 0) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 311 | return 0; |
| 312 | |
| 313 | if (n < size) |
| 314 | size = n; |
| 315 | |
| 316 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 317 | return -EIO; |
| 318 | |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 319 | /* |
| 320 | * Locking is a bit tricky here. If we take an exclusive lock |
| 321 | * for direct IO, we effectively serialise all new concurrent |
| 322 | * read IO to this file and block it behind IO that is currently in |
| 323 | * progress because IO in progress holds the IO lock shared. We only |
| 324 | * need to hold the lock exclusive to blow away the page cache, so |
| 325 | * only take lock exclusively if the page cache needs invalidation. |
| 326 | * This allows the normal direct IO case of no page cache pages to |
| 327 | * proceeed concurrently without serialisation. |
| 328 | */ |
| 329 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 330 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 331 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 332 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
| 333 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 334 | if (inode->i_mapping->nrpages) { |
Dave Chinner | 8ff1e67 | 2014-05-15 09:21:37 +1000 | [diff] [blame] | 335 | ret = filemap_write_and_wait_range( |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 336 | VFS_I(ip)->i_mapping, |
Dave Chinner | 7d4ea3c | 2014-09-02 12:12:53 +1000 | [diff] [blame] | 337 | pos, pos + size - 1); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 338 | if (ret) { |
| 339 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); |
| 340 | return ret; |
| 341 | } |
Chris Mason | 85e584d | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 342 | |
| 343 | /* |
| 344 | * Invalidate whole pages. This can return an error if |
| 345 | * we fail to invalidate a page, but this should never |
| 346 | * happen on XFS. Warn if it does fail. |
| 347 | */ |
| 348 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, |
Dave Chinner | 7d4ea3c | 2014-09-02 12:12:53 +1000 | [diff] [blame] | 349 | pos >> PAGE_CACHE_SHIFT, |
| 350 | (pos + size - 1) >> PAGE_CACHE_SHIFT); |
Chris Mason | 85e584d | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 351 | WARN_ON_ONCE(ret); |
| 352 | ret = 0; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 353 | } |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 354 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 355 | } |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 356 | |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 357 | trace_xfs_file_read(ip, size, pos, ioflags); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 358 | |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 359 | ret = generic_file_read_iter(iocb, to); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 360 | if (ret > 0) |
| 361 | XFS_STATS_ADD(xs_read_bytes, ret); |
| 362 | |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 363 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 364 | return ret; |
| 365 | } |
| 366 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 367 | STATIC ssize_t |
| 368 | xfs_file_splice_read( |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 369 | struct file *infilp, |
| 370 | loff_t *ppos, |
| 371 | struct pipe_inode_info *pipe, |
| 372 | size_t count, |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 373 | unsigned int flags) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 374 | { |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 375 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 376 | int ioflags = 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 377 | ssize_t ret; |
| 378 | |
| 379 | XFS_STATS_INC(xs_read_calls); |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 380 | |
| 381 | if (infilp->f_mode & FMODE_NOCMTIME) |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 382 | ioflags |= XFS_IO_INVIS; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 383 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 384 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 385 | return -EIO; |
| 386 | |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 387 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 388 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 389 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
| 390 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 391 | /* for dax, we need to avoid the page cache */ |
| 392 | if (IS_DAX(VFS_I(ip))) |
| 393 | ret = default_file_splice_read(infilp, ppos, pipe, count, flags); |
| 394 | else |
| 395 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 396 | if (ret > 0) |
| 397 | XFS_STATS_ADD(xs_read_bytes, ret); |
| 398 | |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 399 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 400 | return ret; |
| 401 | } |
| 402 | |
Dave Chinner | 4c5cfd1 | 2011-01-11 10:14:16 +1100 | [diff] [blame] | 403 | /* |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 404 | * This routine is called to handle zeroing any space in the last block of the |
| 405 | * file that is beyond the EOF. We do this since the size is being increased |
| 406 | * without writing anything to that block and we don't want to read the |
| 407 | * garbage on the disk. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 408 | */ |
| 409 | STATIC int /* error (positive) */ |
| 410 | xfs_zero_last_block( |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 411 | struct xfs_inode *ip, |
| 412 | xfs_fsize_t offset, |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 413 | xfs_fsize_t isize, |
| 414 | bool *did_zeroing) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 415 | { |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 416 | struct xfs_mount *mp = ip->i_mount; |
| 417 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); |
| 418 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); |
| 419 | int zero_len; |
| 420 | int nimaps = 1; |
| 421 | int error = 0; |
| 422 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 423 | |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 424 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 425 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 426 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 427 | if (error) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 428 | return error; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 429 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 430 | ASSERT(nimaps > 0); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 431 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 432 | /* |
| 433 | * If the block underlying isize is just a hole, then there |
| 434 | * is nothing to zero. |
| 435 | */ |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 436 | if (imap.br_startblock == HOLESTARTBLOCK) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 437 | return 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 438 | |
| 439 | zero_len = mp->m_sb.sb_blocksize - zero_offset; |
| 440 | if (isize + zero_len > offset) |
| 441 | zero_len = offset - isize; |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 442 | *did_zeroing = true; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 443 | return xfs_iozero(ip, isize, zero_len); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 444 | } |
| 445 | |
| 446 | /* |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 447 | * Zero any on disk space between the current EOF and the new, larger EOF. |
| 448 | * |
| 449 | * This handles the normal case of zeroing the remainder of the last block in |
| 450 | * the file and the unusual case of zeroing blocks out beyond the size of the |
| 451 | * file. This second case only happens with fixed size extents and when the |
| 452 | * system crashes before the inode size was updated but after blocks were |
| 453 | * allocated. |
| 454 | * |
| 455 | * Expects the iolock to be held exclusive, and will take the ilock internally. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 456 | */ |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 457 | int /* error (positive) */ |
| 458 | xfs_zero_eof( |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 459 | struct xfs_inode *ip, |
| 460 | xfs_off_t offset, /* starting I/O offset */ |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 461 | xfs_fsize_t isize, /* current inode size */ |
| 462 | bool *did_zeroing) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 463 | { |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 464 | struct xfs_mount *mp = ip->i_mount; |
| 465 | xfs_fileoff_t start_zero_fsb; |
| 466 | xfs_fileoff_t end_zero_fsb; |
| 467 | xfs_fileoff_t zero_count_fsb; |
| 468 | xfs_fileoff_t last_fsb; |
| 469 | xfs_fileoff_t zero_off; |
| 470 | xfs_fsize_t zero_len; |
| 471 | int nimaps; |
| 472 | int error = 0; |
| 473 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 474 | |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 475 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 476 | ASSERT(offset > isize); |
| 477 | |
| 478 | /* |
| 479 | * First handle zeroing the block on which isize resides. |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 480 | * |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 481 | * We only zero a part of that block so it is handled specially. |
| 482 | */ |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 483 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 484 | error = xfs_zero_last_block(ip, offset, isize, did_zeroing); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 485 | if (error) |
| 486 | return error; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 487 | } |
| 488 | |
| 489 | /* |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 490 | * Calculate the range between the new size and the old where blocks |
| 491 | * needing to be zeroed may exist. |
| 492 | * |
| 493 | * To get the block where the last byte in the file currently resides, |
| 494 | * we need to subtract one from the size and truncate back to a block |
| 495 | * boundary. We subtract 1 in case the size is exactly on a block |
| 496 | * boundary. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 497 | */ |
| 498 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; |
| 499 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); |
| 500 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); |
| 501 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); |
| 502 | if (last_fsb == end_zero_fsb) { |
| 503 | /* |
| 504 | * The size was only incremented on its last block. |
| 505 | * We took care of that above, so just return. |
| 506 | */ |
| 507 | return 0; |
| 508 | } |
| 509 | |
| 510 | ASSERT(start_zero_fsb <= end_zero_fsb); |
| 511 | while (start_zero_fsb <= end_zero_fsb) { |
| 512 | nimaps = 1; |
| 513 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 514 | |
| 515 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 516 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
| 517 | &imap, &nimaps, 0); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 518 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 519 | if (error) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 520 | return error; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 521 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 522 | ASSERT(nimaps > 0); |
| 523 | |
| 524 | if (imap.br_state == XFS_EXT_UNWRITTEN || |
| 525 | imap.br_startblock == HOLESTARTBLOCK) { |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 526 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
| 527 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); |
| 528 | continue; |
| 529 | } |
| 530 | |
| 531 | /* |
| 532 | * There are blocks we need to zero. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 533 | */ |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 534 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
| 535 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); |
| 536 | |
| 537 | if ((zero_off + zero_len) > offset) |
| 538 | zero_len = offset - zero_off; |
| 539 | |
| 540 | error = xfs_iozero(ip, zero_off, zero_len); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 541 | if (error) |
| 542 | return error; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 543 | |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 544 | *did_zeroing = true; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 545 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
| 546 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 547 | } |
| 548 | |
| 549 | return 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 550 | } |
| 551 | |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 552 | /* |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 553 | * Common pre-write limit and setup checks. |
| 554 | * |
Christoph Hellwig | 5bf1f26 | 2011-12-18 20:00:13 +0000 | [diff] [blame] | 555 | * Called with the iolocked held either shared and exclusive according to |
| 556 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive |
| 557 | * if called for a direct write beyond i_size. |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 558 | */ |
| 559 | STATIC ssize_t |
| 560 | xfs_file_aio_write_checks( |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 561 | struct kiocb *iocb, |
| 562 | struct iov_iter *from, |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 563 | int *iolock) |
| 564 | { |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 565 | struct file *file = iocb->ki_filp; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 566 | struct inode *inode = file->f_mapping->host; |
| 567 | struct xfs_inode *ip = XFS_I(inode); |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 568 | ssize_t error = 0; |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 569 | size_t count = iov_iter_count(from); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 570 | |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 571 | restart: |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 572 | error = generic_write_checks(iocb, from); |
| 573 | if (error <= 0) |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 574 | return error; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 575 | |
Christoph Hellwig | 21c3ea1 | 2015-04-13 11:38:29 +1000 | [diff] [blame] | 576 | error = xfs_break_layouts(inode, iolock, true); |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 577 | if (error) |
| 578 | return error; |
| 579 | |
Jan Kara | a6de82c | 2015-05-21 16:05:56 +0200 | [diff] [blame] | 580 | /* For changing security info in file_remove_privs() we need i_mutex */ |
| 581 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
| 582 | xfs_rw_iunlock(ip, *iolock); |
| 583 | *iolock = XFS_IOLOCK_EXCL; |
| 584 | xfs_rw_ilock(ip, *iolock); |
| 585 | goto restart; |
| 586 | } |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 587 | /* |
| 588 | * If the offset is beyond the size of the file, we need to zero any |
| 589 | * blocks that fall between the existing EOF and the start of this |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 590 | * write. If zeroing is needed and we are currently holding the |
Christoph Hellwig | 467f789 | 2012-03-27 10:34:47 -0400 | [diff] [blame] | 591 | * iolock shared, we need to update it to exclusive which implies |
| 592 | * having to redo all checks before. |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 593 | * |
| 594 | * We need to serialise against EOF updates that occur in IO |
| 595 | * completions here. We want to make sure that nobody is changing the |
| 596 | * size while we do this check until we have placed an IO barrier (i.e. |
| 597 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. |
| 598 | * The spinlock effectively forms a memory barrier once we have the |
| 599 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value |
| 600 | * and hence be able to correctly determine if we need to run zeroing. |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 601 | */ |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 602 | spin_lock(&ip->i_flags_lock); |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 603 | if (iocb->ki_pos > i_size_read(inode)) { |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 604 | bool zero = false; |
| 605 | |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 606 | spin_unlock(&ip->i_flags_lock); |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 607 | if (*iolock == XFS_IOLOCK_SHARED) { |
Christoph Hellwig | 467f789 | 2012-03-27 10:34:47 -0400 | [diff] [blame] | 608 | xfs_rw_iunlock(ip, *iolock); |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 609 | *iolock = XFS_IOLOCK_EXCL; |
Christoph Hellwig | 467f789 | 2012-03-27 10:34:47 -0400 | [diff] [blame] | 610 | xfs_rw_ilock(ip, *iolock); |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 611 | iov_iter_reexpand(from, count); |
Dave Chinner | 40c63fb | 2015-04-16 22:03:17 +1000 | [diff] [blame] | 612 | |
| 613 | /* |
| 614 | * We now have an IO submission barrier in place, but |
| 615 | * AIO can do EOF updates during IO completion and hence |
| 616 | * we now need to wait for all of them to drain. Non-AIO |
| 617 | * DIO will have drained before we are given the |
| 618 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a |
| 619 | * no-op. |
| 620 | */ |
| 621 | inode_dio_wait(inode); |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 622 | goto restart; |
| 623 | } |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 624 | error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero); |
Christoph Hellwig | 467f789 | 2012-03-27 10:34:47 -0400 | [diff] [blame] | 625 | if (error) |
| 626 | return error; |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 627 | } else |
| 628 | spin_unlock(&ip->i_flags_lock); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 629 | |
| 630 | /* |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 631 | * Updating the timestamps will grab the ilock again from |
| 632 | * xfs_fs_dirty_inode, so we have to call it after dropping the |
| 633 | * lock above. Eventually we should look into a way to avoid |
| 634 | * the pointless lock roundtrip. |
| 635 | */ |
Josef Bacik | c3b2da3 | 2012-03-26 09:59:21 -0400 | [diff] [blame] | 636 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
| 637 | error = file_update_time(file); |
| 638 | if (error) |
| 639 | return error; |
| 640 | } |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 641 | |
| 642 | /* |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 643 | * If we're writing the file then make sure to clear the setuid and |
| 644 | * setgid bits if the process is not being run by root. This keeps |
| 645 | * people from modifying setuid and setgid binaries. |
| 646 | */ |
Jan Kara | a6de82c | 2015-05-21 16:05:56 +0200 | [diff] [blame] | 647 | if (!IS_NOSEC(inode)) |
| 648 | return file_remove_privs(file); |
| 649 | return 0; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 650 | } |
| 651 | |
| 652 | /* |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 653 | * xfs_file_dio_aio_write - handle direct IO writes |
| 654 | * |
| 655 | * Lock the inode appropriately to prepare for and issue a direct IO write. |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 656 | * By separating it from the buffered write path we remove all the tricky to |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 657 | * follow locking changes and looping. |
| 658 | * |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 659 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
| 660 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached |
| 661 | * pages are flushed out. |
| 662 | * |
| 663 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED |
| 664 | * allowing them to be done in parallel with reads and other direct IO writes. |
| 665 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer |
| 666 | * needs to do sub-block zeroing and that requires serialisation against other |
| 667 | * direct IOs to the same block. In this case we need to serialise the |
| 668 | * submission of the unaligned IOs so that we don't get racing block zeroing in |
| 669 | * the dio layer. To avoid the problem with aio, we also need to wait for |
| 670 | * outstanding IOs to complete so that unwritten extent conversion is completed |
| 671 | * before we try to map the overlapping block. This is currently implemented by |
Christoph Hellwig | 4a06fd2 | 2011-08-23 08:28:13 +0000 | [diff] [blame] | 672 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 673 | * |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 674 | * Returns with locks held indicated by @iolock and errors indicated by |
| 675 | * negative return values. |
| 676 | */ |
| 677 | STATIC ssize_t |
| 678 | xfs_file_dio_aio_write( |
| 679 | struct kiocb *iocb, |
Al Viro | b318891 | 2014-04-02 07:06:30 -0400 | [diff] [blame] | 680 | struct iov_iter *from) |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 681 | { |
| 682 | struct file *file = iocb->ki_filp; |
| 683 | struct address_space *mapping = file->f_mapping; |
| 684 | struct inode *inode = mapping->host; |
| 685 | struct xfs_inode *ip = XFS_I(inode); |
| 686 | struct xfs_mount *mp = ip->i_mount; |
| 687 | ssize_t ret = 0; |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 688 | int unaligned_io = 0; |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 689 | int iolock; |
Al Viro | b318891 | 2014-04-02 07:06:30 -0400 | [diff] [blame] | 690 | size_t count = iov_iter_count(from); |
| 691 | loff_t pos = iocb->ki_pos; |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 692 | loff_t end; |
| 693 | struct iov_iter data; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 694 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
| 695 | mp->m_rtdev_targp : mp->m_ddev_targp; |
| 696 | |
Eric Sandeen | 7c71ee7 | 2014-01-21 16:46:23 -0600 | [diff] [blame] | 697 | /* DIO must be aligned to device logical sector size */ |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 698 | if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 699 | return -EINVAL; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 700 | |
Eric Sandeen | 7c71ee7 | 2014-01-21 16:46:23 -0600 | [diff] [blame] | 701 | /* "unaligned" here means not aligned to a filesystem block */ |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 702 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
| 703 | unaligned_io = 1; |
| 704 | |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 705 | /* |
| 706 | * We don't need to take an exclusive lock unless there page cache needs |
| 707 | * to be invalidated or unaligned IO is being executed. We don't need to |
| 708 | * consider the EOF extension case here because |
| 709 | * xfs_file_aio_write_checks() will relock the inode as necessary for |
| 710 | * EOF zeroing cases and fill out the new inode size as appropriate. |
| 711 | */ |
| 712 | if (unaligned_io || mapping->nrpages) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 713 | iolock = XFS_IOLOCK_EXCL; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 714 | else |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 715 | iolock = XFS_IOLOCK_SHARED; |
| 716 | xfs_rw_ilock(ip, iolock); |
Christoph Hellwig | c58cb16 | 2011-08-27 14:42:53 +0000 | [diff] [blame] | 717 | |
| 718 | /* |
| 719 | * Recheck if there are cached pages that need invalidate after we got |
| 720 | * the iolock to protect against other threads adding new pages while |
| 721 | * we were waiting for the iolock. |
| 722 | */ |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 723 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
| 724 | xfs_rw_iunlock(ip, iolock); |
| 725 | iolock = XFS_IOLOCK_EXCL; |
| 726 | xfs_rw_ilock(ip, iolock); |
Christoph Hellwig | c58cb16 | 2011-08-27 14:42:53 +0000 | [diff] [blame] | 727 | } |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 728 | |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 729 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 730 | if (ret) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 731 | goto out; |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 732 | count = iov_iter_count(from); |
| 733 | pos = iocb->ki_pos; |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 734 | end = pos + count - 1; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 735 | |
| 736 | if (mapping->nrpages) { |
Dave Chinner | 07d5035 | 2014-04-17 08:15:27 +1000 | [diff] [blame] | 737 | ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 738 | pos, end); |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 739 | if (ret) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 740 | goto out; |
Dave Chinner | 834ffca | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 741 | /* |
| 742 | * Invalidate whole pages. This can return an error if |
| 743 | * we fail to invalidate a page, but this should never |
| 744 | * happen on XFS. Warn if it does fail. |
| 745 | */ |
| 746 | ret = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, |
Dave Chinner | 7d4ea3c | 2014-09-02 12:12:53 +1000 | [diff] [blame] | 747 | pos >> PAGE_CACHE_SHIFT, |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 748 | end >> PAGE_CACHE_SHIFT); |
Dave Chinner | 834ffca | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 749 | WARN_ON_ONCE(ret); |
| 750 | ret = 0; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 751 | } |
| 752 | |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 753 | /* |
| 754 | * If we are doing unaligned IO, wait for all other IO to drain, |
| 755 | * otherwise demote the lock if we had to flush cached pages |
| 756 | */ |
| 757 | if (unaligned_io) |
Christoph Hellwig | 4a06fd2 | 2011-08-23 08:28:13 +0000 | [diff] [blame] | 758 | inode_dio_wait(inode); |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 759 | else if (iolock == XFS_IOLOCK_EXCL) { |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 760 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 761 | iolock = XFS_IOLOCK_SHARED; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 762 | } |
| 763 | |
| 764 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 765 | |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 766 | data = *from; |
Linus Torvalds | 1aef882 | 2015-04-24 07:08:41 -0700 | [diff] [blame] | 767 | ret = mapping->a_ops->direct_IO(iocb, &data, pos); |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 768 | |
| 769 | /* see generic_file_direct_write() for why this is necessary */ |
| 770 | if (mapping->nrpages) { |
| 771 | invalidate_inode_pages2_range(mapping, |
| 772 | pos >> PAGE_CACHE_SHIFT, |
| 773 | end >> PAGE_CACHE_SHIFT); |
| 774 | } |
| 775 | |
| 776 | if (ret > 0) { |
| 777 | pos += ret; |
| 778 | iov_iter_advance(from, ret); |
| 779 | iocb->ki_pos = pos; |
| 780 | } |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 781 | out: |
| 782 | xfs_rw_iunlock(ip, iolock); |
| 783 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 784 | /* |
| 785 | * No fallback to buffered IO on errors for XFS. DAX can result in |
| 786 | * partial writes, but direct IO will either complete fully or fail. |
| 787 | */ |
| 788 | ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip))); |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 789 | return ret; |
| 790 | } |
| 791 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 792 | STATIC ssize_t |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 793 | xfs_file_buffered_aio_write( |
| 794 | struct kiocb *iocb, |
Al Viro | b318891 | 2014-04-02 07:06:30 -0400 | [diff] [blame] | 795 | struct iov_iter *from) |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 796 | { |
| 797 | struct file *file = iocb->ki_filp; |
| 798 | struct address_space *mapping = file->f_mapping; |
| 799 | struct inode *inode = mapping->host; |
| 800 | struct xfs_inode *ip = XFS_I(inode); |
| 801 | ssize_t ret; |
| 802 | int enospc = 0; |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 803 | int iolock = XFS_IOLOCK_EXCL; |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 804 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 805 | xfs_rw_ilock(ip, iolock); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 806 | |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 807 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 808 | if (ret) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 809 | goto out; |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 810 | |
| 811 | /* We can write back this queue in page reclaim */ |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 812 | current->backing_dev_info = inode_to_bdi(inode); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 813 | |
| 814 | write_retry: |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 815 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), |
| 816 | iocb->ki_pos, 0); |
| 817 | ret = generic_perform_write(file, from, iocb->ki_pos); |
Al Viro | 0a64bc2 | 2014-02-11 22:25:22 -0500 | [diff] [blame] | 818 | if (likely(ret >= 0)) |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 819 | iocb->ki_pos += ret; |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 820 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 821 | /* |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 822 | * If we hit a space limit, try to free up some lingering preallocated |
| 823 | * space before returning an error. In the case of ENOSPC, first try to |
| 824 | * write back all dirty inodes to free up some of the excess reserved |
| 825 | * metadata space. This reduces the chances that the eofblocks scan |
| 826 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this |
| 827 | * also behaves as a filter to prevent too many eofblocks scans from |
| 828 | * running at the same time. |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 829 | */ |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 830 | if (ret == -EDQUOT && !enospc) { |
| 831 | enospc = xfs_inode_free_quota_eofblocks(ip); |
| 832 | if (enospc) |
| 833 | goto write_retry; |
| 834 | } else if (ret == -ENOSPC && !enospc) { |
| 835 | struct xfs_eofblocks eofb = {0}; |
| 836 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 837 | enospc = 1; |
Dave Chinner | 9aa0500 | 2012-10-08 21:56:04 +1100 | [diff] [blame] | 838 | xfs_flush_inodes(ip->i_mount); |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 839 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
| 840 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; |
| 841 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); |
Dave Chinner | 9aa0500 | 2012-10-08 21:56:04 +1100 | [diff] [blame] | 842 | goto write_retry; |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 843 | } |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 844 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 845 | current->backing_dev_info = NULL; |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 846 | out: |
| 847 | xfs_rw_iunlock(ip, iolock); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 848 | return ret; |
| 849 | } |
| 850 | |
| 851 | STATIC ssize_t |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 852 | xfs_file_write_iter( |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 853 | struct kiocb *iocb, |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 854 | struct iov_iter *from) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 855 | { |
| 856 | struct file *file = iocb->ki_filp; |
| 857 | struct address_space *mapping = file->f_mapping; |
| 858 | struct inode *inode = mapping->host; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 859 | struct xfs_inode *ip = XFS_I(inode); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 860 | ssize_t ret; |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 861 | size_t ocount = iov_iter_count(from); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 862 | |
| 863 | XFS_STATS_INC(xs_write_calls); |
| 864 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 865 | if (ocount == 0) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 866 | return 0; |
| 867 | |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 868 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 869 | return -EIO; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 870 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 871 | if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode)) |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 872 | ret = xfs_file_dio_aio_write(iocb, from); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 873 | else |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 874 | ret = xfs_file_buffered_aio_write(iocb, from); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 875 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 876 | if (ret > 0) { |
| 877 | ssize_t err; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 878 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 879 | XFS_STATS_ADD(xs_write_bytes, ret); |
Christoph Hellwig | ce7ae151 | 2011-12-18 20:00:11 +0000 | [diff] [blame] | 880 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 881 | /* Handle various SYNC-type writes */ |
Al Viro | d311d79 | 2014-02-09 15:18:09 -0500 | [diff] [blame] | 882 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 883 | if (err < 0) |
| 884 | ret = err; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 885 | } |
Dave Chinner | a363f0c | 2011-01-11 10:13:53 +1100 | [diff] [blame] | 886 | return ret; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 887 | } |
| 888 | |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 889 | #define XFS_FALLOC_FL_SUPPORTED \ |
| 890 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ |
| 891 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ |
| 892 | FALLOC_FL_INSERT_RANGE) |
| 893 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 894 | STATIC long |
| 895 | xfs_file_fallocate( |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 896 | struct file *file, |
| 897 | int mode, |
| 898 | loff_t offset, |
| 899 | loff_t len) |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 900 | { |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 901 | struct inode *inode = file_inode(file); |
| 902 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 903 | long error; |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 904 | enum xfs_prealloc_flags flags = 0; |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 905 | uint iolock = XFS_IOLOCK_EXCL; |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 906 | loff_t new_size = 0; |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 907 | bool do_file_insert = 0; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 908 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 909 | if (!S_ISREG(inode->i_mode)) |
| 910 | return -EINVAL; |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 911 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 912 | return -EOPNOTSUPP; |
| 913 | |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 914 | xfs_ilock(ip, iolock); |
Christoph Hellwig | 21c3ea1 | 2015-04-13 11:38:29 +1000 | [diff] [blame] | 915 | error = xfs_break_layouts(inode, &iolock, false); |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 916 | if (error) |
| 917 | goto out_unlock; |
| 918 | |
Dave Chinner | e8e9ad4 | 2015-02-23 21:45:32 +1100 | [diff] [blame] | 919 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
| 920 | iolock |= XFS_MMAPLOCK_EXCL; |
| 921 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 922 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
| 923 | error = xfs_free_file_space(ip, offset, len); |
| 924 | if (error) |
| 925 | goto out_unlock; |
Namjae Jeon | e1d8fb8 | 2014-02-24 10:58:19 +1100 | [diff] [blame] | 926 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
| 927 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; |
| 928 | |
| 929 | if (offset & blksize_mask || len & blksize_mask) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 930 | error = -EINVAL; |
Namjae Jeon | e1d8fb8 | 2014-02-24 10:58:19 +1100 | [diff] [blame] | 931 | goto out_unlock; |
| 932 | } |
| 933 | |
Lukas Czerner | 23fffa9 | 2014-04-12 09:56:41 -0400 | [diff] [blame] | 934 | /* |
| 935 | * There is no need to overlap collapse range with EOF, |
| 936 | * in which case it is effectively a truncate operation |
| 937 | */ |
| 938 | if (offset + len >= i_size_read(inode)) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 939 | error = -EINVAL; |
Lukas Czerner | 23fffa9 | 2014-04-12 09:56:41 -0400 | [diff] [blame] | 940 | goto out_unlock; |
| 941 | } |
| 942 | |
Namjae Jeon | e1d8fb8 | 2014-02-24 10:58:19 +1100 | [diff] [blame] | 943 | new_size = i_size_read(inode) - len; |
| 944 | |
| 945 | error = xfs_collapse_file_space(ip, offset, len); |
| 946 | if (error) |
| 947 | goto out_unlock; |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 948 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
| 949 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; |
| 950 | |
| 951 | new_size = i_size_read(inode) + len; |
| 952 | if (offset & blksize_mask || len & blksize_mask) { |
| 953 | error = -EINVAL; |
| 954 | goto out_unlock; |
| 955 | } |
| 956 | |
| 957 | /* check the new inode size does not wrap through zero */ |
| 958 | if (new_size > inode->i_sb->s_maxbytes) { |
| 959 | error = -EFBIG; |
| 960 | goto out_unlock; |
| 961 | } |
| 962 | |
| 963 | /* Offset should be less than i_size */ |
| 964 | if (offset >= i_size_read(inode)) { |
| 965 | error = -EINVAL; |
| 966 | goto out_unlock; |
| 967 | } |
| 968 | do_file_insert = 1; |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 969 | } else { |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 970 | flags |= XFS_PREALLOC_SET; |
| 971 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 972 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
| 973 | offset + len > i_size_read(inode)) { |
| 974 | new_size = offset + len; |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 975 | error = inode_newsize_ok(inode, new_size); |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 976 | if (error) |
| 977 | goto out_unlock; |
| 978 | } |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 979 | |
Lukas Czerner | 376ba31 | 2014-03-13 19:07:58 +1100 | [diff] [blame] | 980 | if (mode & FALLOC_FL_ZERO_RANGE) |
| 981 | error = xfs_zero_file_space(ip, offset, len); |
| 982 | else |
| 983 | error = xfs_alloc_file_space(ip, offset, len, |
| 984 | XFS_BMAPI_PREALLOC); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 985 | if (error) |
| 986 | goto out_unlock; |
| 987 | } |
| 988 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 989 | if (file->f_flags & O_DSYNC) |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 990 | flags |= XFS_PREALLOC_SYNC; |
| 991 | |
| 992 | error = xfs_update_prealloc_flags(ip, flags); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 993 | if (error) |
| 994 | goto out_unlock; |
| 995 | |
| 996 | /* Change file size if needed */ |
| 997 | if (new_size) { |
| 998 | struct iattr iattr; |
| 999 | |
| 1000 | iattr.ia_valid = ATTR_SIZE; |
| 1001 | iattr.ia_size = new_size; |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 1002 | error = xfs_setattr_size(ip, &iattr); |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 1003 | if (error) |
| 1004 | goto out_unlock; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1005 | } |
| 1006 | |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 1007 | /* |
| 1008 | * Perform hole insertion now that the file size has been |
| 1009 | * updated so that if we crash during the operation we don't |
| 1010 | * leave shifted extents past EOF and hence losing access to |
| 1011 | * the data that is contained within them. |
| 1012 | */ |
| 1013 | if (do_file_insert) |
| 1014 | error = xfs_insert_file_space(ip, offset, len); |
| 1015 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1016 | out_unlock: |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 1017 | xfs_iunlock(ip, iolock); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1018 | return error; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1019 | } |
| 1020 | |
| 1021 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | STATIC int |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1023 | xfs_file_open( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | struct inode *inode, |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1025 | struct file *file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | { |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1027 | if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1028 | return -EFBIG; |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1029 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
| 1030 | return -EIO; |
| 1031 | return 0; |
| 1032 | } |
| 1033 | |
| 1034 | STATIC int |
| 1035 | xfs_dir_open( |
| 1036 | struct inode *inode, |
| 1037 | struct file *file) |
| 1038 | { |
| 1039 | struct xfs_inode *ip = XFS_I(inode); |
| 1040 | int mode; |
| 1041 | int error; |
| 1042 | |
| 1043 | error = xfs_file_open(inode, file); |
| 1044 | if (error) |
| 1045 | return error; |
| 1046 | |
| 1047 | /* |
| 1048 | * If there are any blocks, read-ahead block 0 as we're almost |
| 1049 | * certain to have the next operation be a read there. |
| 1050 | */ |
Christoph Hellwig | 309ecac8 | 2013-12-06 12:30:09 -0800 | [diff] [blame] | 1051 | mode = xfs_ilock_data_map_shared(ip); |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1052 | if (ip->i_d.di_nextents > 0) |
Eric Sandeen | 9df2dd0 | 2014-04-14 19:01:59 +1000 | [diff] [blame] | 1053 | xfs_dir3_data_readahead(ip, 0, -1); |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1054 | xfs_iunlock(ip, mode); |
| 1055 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | } |
| 1057 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | STATIC int |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1059 | xfs_file_release( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | struct inode *inode, |
| 1061 | struct file *filp) |
| 1062 | { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1063 | return xfs_release(XFS_I(inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | } |
| 1065 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1066 | STATIC int |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1067 | xfs_file_readdir( |
Al Viro | b822755 | 2013-05-22 17:07:56 -0400 | [diff] [blame] | 1068 | struct file *file, |
| 1069 | struct dir_context *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1070 | { |
Al Viro | b822755 | 2013-05-22 17:07:56 -0400 | [diff] [blame] | 1071 | struct inode *inode = file_inode(file); |
Christoph Hellwig | 739bfb2 | 2007-08-29 10:58:01 +1000 | [diff] [blame] | 1072 | xfs_inode_t *ip = XFS_I(inode); |
Christoph Hellwig | 051e7cd | 2007-08-28 13:58:24 +1000 | [diff] [blame] | 1073 | size_t bufsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1074 | |
Christoph Hellwig | 051e7cd | 2007-08-28 13:58:24 +1000 | [diff] [blame] | 1075 | /* |
| 1076 | * The Linux API doesn't pass down the total size of the buffer |
| 1077 | * we read into down to the filesystem. With the filldir concept |
| 1078 | * it's not needed for correct information, but the XFS dir2 leaf |
| 1079 | * code wants an estimate of the buffer size to calculate it's |
| 1080 | * readahead window and size the buffers used for mapping to |
| 1081 | * physical blocks. |
| 1082 | * |
| 1083 | * Try to give it an estimate that's good enough, maybe at some |
| 1084 | * point we can change the ->readdir prototype to include the |
Eric Sandeen | a9cc799 | 2010-02-03 17:50:13 +0000 | [diff] [blame] | 1085 | * buffer size. For now we use the current glibc buffer size. |
Christoph Hellwig | 051e7cd | 2007-08-28 13:58:24 +1000 | [diff] [blame] | 1086 | */ |
Eric Sandeen | a9cc799 | 2010-02-03 17:50:13 +0000 | [diff] [blame] | 1087 | bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1088 | |
kbuild test robot | 8300475 | 2014-12-01 08:25:28 +1100 | [diff] [blame] | 1089 | return xfs_readdir(ip, ctx, bufsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | } |
| 1091 | |
David Chinner | 4f57dbc | 2007-07-19 16:28:17 +1000 | [diff] [blame] | 1092 | /* |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 1093 | * This type is designed to indicate the type of offset we would like |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1094 | * to search from page cache for xfs_seek_hole_data(). |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 1095 | */ |
| 1096 | enum { |
| 1097 | HOLE_OFF = 0, |
| 1098 | DATA_OFF, |
| 1099 | }; |
| 1100 | |
| 1101 | /* |
| 1102 | * Lookup the desired type of offset from the given page. |
| 1103 | * |
| 1104 | * On success, return true and the offset argument will point to the |
| 1105 | * start of the region that was found. Otherwise this function will |
| 1106 | * return false and keep the offset argument unchanged. |
| 1107 | */ |
| 1108 | STATIC bool |
| 1109 | xfs_lookup_buffer_offset( |
| 1110 | struct page *page, |
| 1111 | loff_t *offset, |
| 1112 | unsigned int type) |
| 1113 | { |
| 1114 | loff_t lastoff = page_offset(page); |
| 1115 | bool found = false; |
| 1116 | struct buffer_head *bh, *head; |
| 1117 | |
| 1118 | bh = head = page_buffers(page); |
| 1119 | do { |
| 1120 | /* |
| 1121 | * Unwritten extents that have data in the page |
| 1122 | * cache covering them can be identified by the |
| 1123 | * BH_Unwritten state flag. Pages with multiple |
| 1124 | * buffers might have a mix of holes, data and |
| 1125 | * unwritten extents - any buffer with valid |
| 1126 | * data in it should have BH_Uptodate flag set |
| 1127 | * on it. |
| 1128 | */ |
| 1129 | if (buffer_unwritten(bh) || |
| 1130 | buffer_uptodate(bh)) { |
| 1131 | if (type == DATA_OFF) |
| 1132 | found = true; |
| 1133 | } else { |
| 1134 | if (type == HOLE_OFF) |
| 1135 | found = true; |
| 1136 | } |
| 1137 | |
| 1138 | if (found) { |
| 1139 | *offset = lastoff; |
| 1140 | break; |
| 1141 | } |
| 1142 | lastoff += bh->b_size; |
| 1143 | } while ((bh = bh->b_this_page) != head); |
| 1144 | |
| 1145 | return found; |
| 1146 | } |
| 1147 | |
| 1148 | /* |
| 1149 | * This routine is called to find out and return a data or hole offset |
| 1150 | * from the page cache for unwritten extents according to the desired |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1151 | * type for xfs_seek_hole_data(). |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 1152 | * |
| 1153 | * The argument offset is used to tell where we start to search from the |
| 1154 | * page cache. Map is used to figure out the end points of the range to |
| 1155 | * lookup pages. |
| 1156 | * |
| 1157 | * Return true if the desired type of offset was found, and the argument |
| 1158 | * offset is filled with that address. Otherwise, return false and keep |
| 1159 | * offset unchanged. |
| 1160 | */ |
| 1161 | STATIC bool |
| 1162 | xfs_find_get_desired_pgoff( |
| 1163 | struct inode *inode, |
| 1164 | struct xfs_bmbt_irec *map, |
| 1165 | unsigned int type, |
| 1166 | loff_t *offset) |
| 1167 | { |
| 1168 | struct xfs_inode *ip = XFS_I(inode); |
| 1169 | struct xfs_mount *mp = ip->i_mount; |
| 1170 | struct pagevec pvec; |
| 1171 | pgoff_t index; |
| 1172 | pgoff_t end; |
| 1173 | loff_t endoff; |
| 1174 | loff_t startoff = *offset; |
| 1175 | loff_t lastoff = startoff; |
| 1176 | bool found = false; |
| 1177 | |
| 1178 | pagevec_init(&pvec, 0); |
| 1179 | |
| 1180 | index = startoff >> PAGE_CACHE_SHIFT; |
| 1181 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
| 1182 | end = endoff >> PAGE_CACHE_SHIFT; |
| 1183 | do { |
| 1184 | int want; |
| 1185 | unsigned nr_pages; |
| 1186 | unsigned int i; |
| 1187 | |
| 1188 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); |
| 1189 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, |
| 1190 | want); |
| 1191 | /* |
| 1192 | * No page mapped into given range. If we are searching holes |
| 1193 | * and if this is the first time we got into the loop, it means |
| 1194 | * that the given offset is landed in a hole, return it. |
| 1195 | * |
| 1196 | * If we have already stepped through some block buffers to find |
| 1197 | * holes but they all contains data. In this case, the last |
| 1198 | * offset is already updated and pointed to the end of the last |
| 1199 | * mapped page, if it does not reach the endpoint to search, |
| 1200 | * that means there should be a hole between them. |
| 1201 | */ |
| 1202 | if (nr_pages == 0) { |
| 1203 | /* Data search found nothing */ |
| 1204 | if (type == DATA_OFF) |
| 1205 | break; |
| 1206 | |
| 1207 | ASSERT(type == HOLE_OFF); |
| 1208 | if (lastoff == startoff || lastoff < endoff) { |
| 1209 | found = true; |
| 1210 | *offset = lastoff; |
| 1211 | } |
| 1212 | break; |
| 1213 | } |
| 1214 | |
| 1215 | /* |
| 1216 | * At lease we found one page. If this is the first time we |
| 1217 | * step into the loop, and if the first page index offset is |
| 1218 | * greater than the given search offset, a hole was found. |
| 1219 | */ |
| 1220 | if (type == HOLE_OFF && lastoff == startoff && |
| 1221 | lastoff < page_offset(pvec.pages[0])) { |
| 1222 | found = true; |
| 1223 | break; |
| 1224 | } |
| 1225 | |
| 1226 | for (i = 0; i < nr_pages; i++) { |
| 1227 | struct page *page = pvec.pages[i]; |
| 1228 | loff_t b_offset; |
| 1229 | |
| 1230 | /* |
| 1231 | * At this point, the page may be truncated or |
| 1232 | * invalidated (changing page->mapping to NULL), |
| 1233 | * or even swizzled back from swapper_space to tmpfs |
| 1234 | * file mapping. However, page->index will not change |
| 1235 | * because we have a reference on the page. |
| 1236 | * |
| 1237 | * Searching done if the page index is out of range. |
| 1238 | * If the current offset is not reaches the end of |
| 1239 | * the specified search range, there should be a hole |
| 1240 | * between them. |
| 1241 | */ |
| 1242 | if (page->index > end) { |
| 1243 | if (type == HOLE_OFF && lastoff < endoff) { |
| 1244 | *offset = lastoff; |
| 1245 | found = true; |
| 1246 | } |
| 1247 | goto out; |
| 1248 | } |
| 1249 | |
| 1250 | lock_page(page); |
| 1251 | /* |
| 1252 | * Page truncated or invalidated(page->mapping == NULL). |
| 1253 | * We can freely skip it and proceed to check the next |
| 1254 | * page. |
| 1255 | */ |
| 1256 | if (unlikely(page->mapping != inode->i_mapping)) { |
| 1257 | unlock_page(page); |
| 1258 | continue; |
| 1259 | } |
| 1260 | |
| 1261 | if (!page_has_buffers(page)) { |
| 1262 | unlock_page(page); |
| 1263 | continue; |
| 1264 | } |
| 1265 | |
| 1266 | found = xfs_lookup_buffer_offset(page, &b_offset, type); |
| 1267 | if (found) { |
| 1268 | /* |
| 1269 | * The found offset may be less than the start |
| 1270 | * point to search if this is the first time to |
| 1271 | * come here. |
| 1272 | */ |
| 1273 | *offset = max_t(loff_t, startoff, b_offset); |
| 1274 | unlock_page(page); |
| 1275 | goto out; |
| 1276 | } |
| 1277 | |
| 1278 | /* |
| 1279 | * We either searching data but nothing was found, or |
| 1280 | * searching hole but found a data buffer. In either |
| 1281 | * case, probably the next page contains the desired |
| 1282 | * things, update the last offset to it so. |
| 1283 | */ |
| 1284 | lastoff = page_offset(page) + PAGE_SIZE; |
| 1285 | unlock_page(page); |
| 1286 | } |
| 1287 | |
| 1288 | /* |
| 1289 | * The number of returned pages less than our desired, search |
| 1290 | * done. In this case, nothing was found for searching data, |
| 1291 | * but we found a hole behind the last offset. |
| 1292 | */ |
| 1293 | if (nr_pages < want) { |
| 1294 | if (type == HOLE_OFF) { |
| 1295 | *offset = lastoff; |
| 1296 | found = true; |
| 1297 | } |
| 1298 | break; |
| 1299 | } |
| 1300 | |
| 1301 | index = pvec.pages[i - 1]->index + 1; |
| 1302 | pagevec_release(&pvec); |
| 1303 | } while (index <= end); |
| 1304 | |
| 1305 | out: |
| 1306 | pagevec_release(&pvec); |
| 1307 | return found; |
| 1308 | } |
| 1309 | |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1310 | STATIC loff_t |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1311 | xfs_seek_hole_data( |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1312 | struct file *file, |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1313 | loff_t start, |
| 1314 | int whence) |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1315 | { |
| 1316 | struct inode *inode = file->f_mapping->host; |
| 1317 | struct xfs_inode *ip = XFS_I(inode); |
| 1318 | struct xfs_mount *mp = ip->i_mount; |
| 1319 | loff_t uninitialized_var(offset); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1320 | xfs_fsize_t isize; |
| 1321 | xfs_fileoff_t fsbno; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1322 | xfs_filblks_t end; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1323 | uint lock; |
| 1324 | int error; |
| 1325 | |
| 1326 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 1327 | return -EIO; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1328 | |
Christoph Hellwig | 309ecac8 | 2013-12-06 12:30:09 -0800 | [diff] [blame] | 1329 | lock = xfs_ilock_data_map_shared(ip); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1330 | |
| 1331 | isize = i_size_read(inode); |
| 1332 | if (start >= isize) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1333 | error = -ENXIO; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1334 | goto out_unlock; |
| 1335 | } |
| 1336 | |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1337 | /* |
| 1338 | * Try to read extents from the first block indicated |
| 1339 | * by fsbno to the end block of the file. |
| 1340 | */ |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1341 | fsbno = XFS_B_TO_FSBT(mp, start); |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1342 | end = XFS_B_TO_FSB(mp, isize); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1343 | |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1344 | for (;;) { |
| 1345 | struct xfs_bmbt_irec map[2]; |
| 1346 | int nmap = 2; |
| 1347 | unsigned int i; |
| 1348 | |
| 1349 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
| 1350 | XFS_BMAPI_ENTIRE); |
| 1351 | if (error) |
| 1352 | goto out_unlock; |
| 1353 | |
| 1354 | /* No extents at given offset, must be beyond EOF */ |
| 1355 | if (nmap == 0) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1356 | error = -ENXIO; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1357 | goto out_unlock; |
| 1358 | } |
| 1359 | |
| 1360 | for (i = 0; i < nmap; i++) { |
| 1361 | offset = max_t(loff_t, start, |
| 1362 | XFS_FSB_TO_B(mp, map[i].br_startoff)); |
| 1363 | |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1364 | /* Landed in the hole we wanted? */ |
| 1365 | if (whence == SEEK_HOLE && |
| 1366 | map[i].br_startblock == HOLESTARTBLOCK) |
| 1367 | goto out; |
| 1368 | |
| 1369 | /* Landed in the data extent we wanted? */ |
| 1370 | if (whence == SEEK_DATA && |
| 1371 | (map[i].br_startblock == DELAYSTARTBLOCK || |
| 1372 | (map[i].br_state == XFS_EXT_NORM && |
| 1373 | !isnullstartblock(map[i].br_startblock)))) |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1374 | goto out; |
| 1375 | |
| 1376 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1377 | * Landed in an unwritten extent, try to search |
| 1378 | * for hole or data from page cache. |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1379 | */ |
| 1380 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { |
| 1381 | if (xfs_find_get_desired_pgoff(inode, &map[i], |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1382 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
| 1383 | &offset)) |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1384 | goto out; |
| 1385 | } |
| 1386 | } |
| 1387 | |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1388 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1389 | * We only received one extent out of the two requested. This |
| 1390 | * means we've hit EOF and didn't find what we are looking for. |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1391 | */ |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1392 | if (nmap == 1) { |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1393 | /* |
| 1394 | * If we were looking for a hole, set offset to |
| 1395 | * the end of the file (i.e., there is an implicit |
| 1396 | * hole at the end of any file). |
| 1397 | */ |
| 1398 | if (whence == SEEK_HOLE) { |
| 1399 | offset = isize; |
| 1400 | break; |
| 1401 | } |
| 1402 | /* |
| 1403 | * If we were looking for data, it's nowhere to be found |
| 1404 | */ |
| 1405 | ASSERT(whence == SEEK_DATA); |
| 1406 | error = -ENXIO; |
| 1407 | goto out_unlock; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1408 | } |
| 1409 | |
| 1410 | ASSERT(i > 1); |
| 1411 | |
| 1412 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1413 | * Nothing was found, proceed to the next round of search |
| 1414 | * if the next reading offset is not at or beyond EOF. |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1415 | */ |
| 1416 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; |
| 1417 | start = XFS_FSB_TO_B(mp, fsbno); |
| 1418 | if (start >= isize) { |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1419 | if (whence == SEEK_HOLE) { |
| 1420 | offset = isize; |
| 1421 | break; |
| 1422 | } |
| 1423 | ASSERT(whence == SEEK_DATA); |
| 1424 | error = -ENXIO; |
| 1425 | goto out_unlock; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1426 | } |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1427 | } |
| 1428 | |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1429 | out: |
| 1430 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1431 | * If at this point we have found the hole we wanted, the returned |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1432 | * offset may be bigger than the file size as it may be aligned to |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1433 | * page boundary for unwritten extents. We need to deal with this |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1434 | * situation in particular. |
| 1435 | */ |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1436 | if (whence == SEEK_HOLE) |
| 1437 | offset = min_t(loff_t, offset, isize); |
Jie Liu | 46a1c2c | 2013-06-25 12:02:13 +0800 | [diff] [blame] | 1438 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1439 | |
| 1440 | out_unlock: |
Christoph Hellwig | 01f4f32 | 2013-12-06 12:30:08 -0800 | [diff] [blame] | 1441 | xfs_iunlock(ip, lock); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1442 | |
| 1443 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1444 | return error; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1445 | return offset; |
| 1446 | } |
| 1447 | |
| 1448 | STATIC loff_t |
| 1449 | xfs_file_llseek( |
| 1450 | struct file *file, |
| 1451 | loff_t offset, |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1452 | int whence) |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1453 | { |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1454 | switch (whence) { |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1455 | case SEEK_END: |
| 1456 | case SEEK_CUR: |
| 1457 | case SEEK_SET: |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1458 | return generic_file_llseek(file, offset, whence); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1459 | case SEEK_HOLE: |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1460 | case SEEK_DATA: |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1461 | return xfs_seek_hole_data(file, offset, whence); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1462 | default: |
| 1463 | return -EINVAL; |
| 1464 | } |
| 1465 | } |
| 1466 | |
Dave Chinner | de0e8c2 | 2015-02-23 21:44:19 +1100 | [diff] [blame] | 1467 | /* |
| 1468 | * Locking for serialisation of IO during page faults. This results in a lock |
| 1469 | * ordering of: |
| 1470 | * |
| 1471 | * mmap_sem (MM) |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1472 | * sb_start_pagefault(vfs, freeze) |
| 1473 | * i_mmap_lock (XFS - truncate serialisation) |
| 1474 | * page_lock (MM) |
| 1475 | * i_lock (XFS - extent map serialisation) |
Dave Chinner | de0e8c2 | 2015-02-23 21:44:19 +1100 | [diff] [blame] | 1476 | */ |
Dave Chinner | de0e8c2 | 2015-02-23 21:44:19 +1100 | [diff] [blame] | 1477 | |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1478 | /* |
| 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 | */ |
| 1484 | STATIC int |
| 1485 | xfs_filemap_page_mkwrite( |
| 1486 | struct vm_area_struct *vma, |
| 1487 | struct vm_fault *vmf) |
| 1488 | { |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1489 | struct inode *inode = file_inode(vma->vm_file); |
Dave Chinner | ec56b1f | 2015-06-04 09:18:18 +1000 | [diff] [blame] | 1490 | int ret; |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1491 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1492 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1493 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1494 | sb_start_pagefault(inode->i_sb); |
Dave Chinner | ec56b1f | 2015-06-04 09:18:18 +1000 | [diff] [blame] | 1495 | file_update_time(vma->vm_file); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1496 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
| 1497 | |
| 1498 | if (IS_DAX(inode)) { |
| 1499 | ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_direct, |
| 1500 | xfs_end_io_dax_write); |
| 1501 | } else { |
| 1502 | ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks); |
| 1503 | ret = block_page_mkwrite_return(ret); |
| 1504 | } |
| 1505 | |
| 1506 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
| 1507 | sb_end_pagefault(inode->i_sb); |
| 1508 | |
| 1509 | return ret; |
| 1510 | } |
| 1511 | |
| 1512 | STATIC int |
| 1513 | xfs_filemap_fault( |
| 1514 | struct vm_area_struct *vma, |
| 1515 | struct vm_fault *vmf) |
| 1516 | { |
| 1517 | struct xfs_inode *ip = XFS_I(file_inode(vma->vm_file)); |
| 1518 | int ret; |
| 1519 | |
| 1520 | trace_xfs_filemap_fault(ip); |
| 1521 | |
| 1522 | /* DAX can shortcut the normal fault path on write faults! */ |
| 1523 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(VFS_I(ip))) |
| 1524 | return xfs_filemap_page_mkwrite(vma, vmf); |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1525 | |
| 1526 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1527 | ret = filemap_fault(vma, vmf); |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1528 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
| 1529 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1530 | return ret; |
| 1531 | } |
| 1532 | |
| 1533 | static const struct vm_operations_struct xfs_file_vm_ops = { |
| 1534 | .fault = xfs_filemap_fault, |
| 1535 | .map_pages = filemap_map_pages, |
| 1536 | .page_mkwrite = xfs_filemap_page_mkwrite, |
| 1537 | }; |
| 1538 | |
| 1539 | STATIC int |
| 1540 | xfs_file_mmap( |
| 1541 | struct file *filp, |
| 1542 | struct vm_area_struct *vma) |
| 1543 | { |
| 1544 | file_accessed(filp); |
| 1545 | vma->vm_ops = &xfs_file_vm_ops; |
| 1546 | if (IS_DAX(file_inode(filp))) |
| 1547 | vma->vm_flags |= VM_MIXEDMAP; |
| 1548 | return 0; |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1549 | } |
| 1550 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 1551 | const struct file_operations xfs_file_operations = { |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1552 | .llseek = xfs_file_llseek, |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 1553 | .read_iter = xfs_file_read_iter, |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 1554 | .write_iter = xfs_file_write_iter, |
Nathan Scott | 1b89584 | 2006-03-31 13:08:59 +1000 | [diff] [blame] | 1555 | .splice_read = xfs_file_splice_read, |
Al Viro | 8d02076 | 2014-04-05 04:27:08 -0400 | [diff] [blame] | 1556 | .splice_write = iter_file_splice_write, |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1557 | .unlocked_ioctl = xfs_file_ioctl, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1558 | #ifdef CONFIG_COMPAT |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1559 | .compat_ioctl = xfs_file_compat_ioctl, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1560 | #endif |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1561 | .mmap = xfs_file_mmap, |
| 1562 | .open = xfs_file_open, |
| 1563 | .release = xfs_file_release, |
| 1564 | .fsync = xfs_file_fsync, |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1565 | .fallocate = xfs_file_fallocate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | }; |
| 1567 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 1568 | const struct file_operations xfs_dir_file_operations = { |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1569 | .open = xfs_dir_open, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | .read = generic_read_dir, |
Al Viro | b822755 | 2013-05-22 17:07:56 -0400 | [diff] [blame] | 1571 | .iterate = xfs_file_readdir, |
Al Viro | 59af158 | 2008-08-24 07:24:41 -0400 | [diff] [blame] | 1572 | .llseek = generic_file_llseek, |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1573 | .unlocked_ioctl = xfs_file_ioctl, |
Nathan Scott | d387039 | 2005-05-06 06:44:46 -0700 | [diff] [blame] | 1574 | #ifdef CONFIG_COMPAT |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1575 | .compat_ioctl = xfs_file_compat_ioctl, |
Nathan Scott | d387039 | 2005-05-06 06:44:46 -0700 | [diff] [blame] | 1576 | #endif |
Christoph Hellwig | 1da2f2d | 2011-10-02 14:25:16 +0000 | [diff] [blame] | 1577 | .fsync = xfs_dir_fsync, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1578 | }; |