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 | 70a9883c | 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) |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 58 | inode_lock(VFS_I(ip)); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 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) |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 69 | inode_unlock(VFS_I(ip)); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 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) |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 79 | inode_unlock(VFS_I(ip)); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 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 | /* |
Dave Chinner | fc0561c | 2015-11-03 13:14:59 +1100 | [diff] [blame] | 245 | * All metadata updates are logged, which means that we just have to |
| 246 | * flush the log up to the latest LSN that touched the inode. If we have |
| 247 | * concurrent fsync/fdatasync() calls, we need them to all block on the |
| 248 | * log force before we clear the ili_fsync_fields field. This ensures |
| 249 | * that we don't get a racing sync operation that does not wait for the |
| 250 | * metadata to hit the journal before returning. If we race with |
| 251 | * clearing the ili_fsync_fields, then all that will happen is the log |
| 252 | * force will do nothing as the lsn will already be on disk. We can't |
| 253 | * race with setting ili_fsync_fields because that is done under |
| 254 | * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared |
| 255 | * until after the ili_fsync_fields is cleared. |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 256 | */ |
| 257 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | 8f639dd | 2012-02-29 09:53:55 +0000 | [diff] [blame] | 258 | if (xfs_ipincount(ip)) { |
| 259 | if (!datasync || |
Dave Chinner | fc0561c | 2015-11-03 13:14:59 +1100 | [diff] [blame] | 260 | (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP)) |
Christoph Hellwig | 8f639dd | 2012-02-29 09:53:55 +0000 | [diff] [blame] | 261 | lsn = ip->i_itemp->ili_last_lsn; |
| 262 | } |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 263 | |
Dave Chinner | fc0561c | 2015-11-03 13:14:59 +1100 | [diff] [blame] | 264 | if (lsn) { |
Christoph Hellwig | b103705 | 2011-09-19 14:55:51 +0000 | [diff] [blame] | 265 | error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed); |
Dave Chinner | fc0561c | 2015-11-03 13:14:59 +1100 | [diff] [blame] | 266 | ip->i_itemp->ili_fsync_fields = 0; |
| 267 | } |
| 268 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | b103705 | 2011-09-19 14:55:51 +0000 | [diff] [blame] | 269 | |
Christoph Hellwig | a27a263 | 2011-06-16 12:02:23 +0000 | [diff] [blame] | 270 | /* |
| 271 | * If we only have a single device, and the log force about was |
| 272 | * a no-op we might have to flush the data device cache here. |
| 273 | * This can only happen for fdatasync/O_DSYNC if we were overwriting |
| 274 | * an already allocated file and thus do not have any metadata to |
| 275 | * commit. |
| 276 | */ |
| 277 | if ((mp->m_flags & XFS_MOUNT_BARRIER) && |
| 278 | mp->m_logdev_targp == mp->m_ddev_targp && |
| 279 | !XFS_IS_REALTIME_INODE(ip) && |
| 280 | !log_flushed) |
| 281 | xfs_blkdev_issue_flush(mp->m_ddev_targp); |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 282 | |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 283 | return error; |
Christoph Hellwig | fd3200b | 2010-02-15 09:44:48 +0000 | [diff] [blame] | 284 | } |
| 285 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 286 | STATIC ssize_t |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 287 | xfs_file_read_iter( |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 288 | struct kiocb *iocb, |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 289 | struct iov_iter *to) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 290 | { |
| 291 | struct file *file = iocb->ki_filp; |
| 292 | struct inode *inode = file->f_mapping->host; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 293 | struct xfs_inode *ip = XFS_I(inode); |
| 294 | struct xfs_mount *mp = ip->i_mount; |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 295 | size_t size = iov_iter_count(to); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 296 | ssize_t ret = 0; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 297 | int ioflags = 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 298 | xfs_fsize_t n; |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 299 | loff_t pos = iocb->ki_pos; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 300 | |
Bill O'Donnell | ff6d6af | 2015-10-12 18:21:22 +1100 | [diff] [blame] | 301 | XFS_STATS_INC(mp, xs_read_calls); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 302 | |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 303 | if (unlikely(iocb->ki_flags & IOCB_DIRECT)) |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 304 | ioflags |= XFS_IO_ISDIRECT; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 305 | if (file->f_mode & FMODE_NOCMTIME) |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 306 | ioflags |= XFS_IO_INVIS; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 307 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 308 | if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) { |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 309 | xfs_buftarg_t *target = |
| 310 | XFS_IS_REALTIME_INODE(ip) ? |
| 311 | mp->m_rtdev_targp : mp->m_ddev_targp; |
Eric Sandeen | 7c71ee7 | 2014-01-21 16:46:23 -0600 | [diff] [blame] | 312 | /* DIO must be aligned to device logical sector size */ |
| 313 | if ((pos | size) & target->bt_logical_sectormask) { |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 314 | if (pos == i_size_read(inode)) |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 315 | return 0; |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 316 | return -EINVAL; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 317 | } |
| 318 | } |
| 319 | |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 320 | n = mp->m_super->s_maxbytes - pos; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 321 | if (n <= 0 || size == 0) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 322 | return 0; |
| 323 | |
| 324 | if (n < size) |
| 325 | size = n; |
| 326 | |
| 327 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 328 | return -EIO; |
| 329 | |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 330 | /* |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 331 | * Locking is a bit tricky here. If we take an exclusive lock for direct |
| 332 | * IO, we effectively serialise all new concurrent read IO to this file |
| 333 | * and block it behind IO that is currently in progress because IO in |
| 334 | * progress holds the IO lock shared. We only need to hold the lock |
| 335 | * exclusive to blow away the page cache, so only take lock exclusively |
| 336 | * if the page cache needs invalidation. This allows the normal direct |
| 337 | * IO case of no page cache pages to proceeed concurrently without |
| 338 | * serialisation. |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 339 | */ |
| 340 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 341 | if ((ioflags & XFS_IO_ISDIRECT) && inode->i_mapping->nrpages) { |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 342 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 343 | xfs_rw_ilock(ip, XFS_IOLOCK_EXCL); |
| 344 | |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 345 | /* |
| 346 | * The generic dio code only flushes the range of the particular |
| 347 | * I/O. Because we take an exclusive lock here, this whole |
| 348 | * sequence is considerably more expensive for us. This has a |
| 349 | * noticeable performance impact for any file with cached pages, |
| 350 | * even when outside of the range of the particular I/O. |
| 351 | * |
| 352 | * Hence, amortize the cost of the lock against a full file |
| 353 | * flush and reduce the chances of repeated iolock cycles going |
| 354 | * forward. |
| 355 | */ |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 356 | if (inode->i_mapping->nrpages) { |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 357 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 358 | if (ret) { |
| 359 | xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL); |
| 360 | return ret; |
| 361 | } |
Chris Mason | 85e584d | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 362 | |
| 363 | /* |
| 364 | * Invalidate whole pages. This can return an error if |
| 365 | * we fail to invalidate a page, but this should never |
| 366 | * happen on XFS. Warn if it does fail. |
| 367 | */ |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 368 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
Chris Mason | 85e584d | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 369 | WARN_ON_ONCE(ret); |
| 370 | ret = 0; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 371 | } |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 372 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
Dave Chinner | 0c38a25 | 2011-08-25 07:17:01 +0000 | [diff] [blame] | 373 | } |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 374 | |
Dave Chinner | fb59581 | 2012-11-12 22:53:57 +1100 | [diff] [blame] | 375 | trace_xfs_file_read(ip, size, pos, ioflags); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 376 | |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 377 | ret = generic_file_read_iter(iocb, to); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 378 | if (ret > 0) |
Bill O'Donnell | ff6d6af | 2015-10-12 18:21:22 +1100 | [diff] [blame] | 379 | XFS_STATS_ADD(mp, xs_read_bytes, ret); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 380 | |
Dave Chinner | 487f84f | 2011-01-12 11:37:10 +1100 | [diff] [blame] | 381 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 382 | return ret; |
| 383 | } |
| 384 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 385 | STATIC ssize_t |
| 386 | xfs_file_splice_read( |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 387 | struct file *infilp, |
| 388 | loff_t *ppos, |
| 389 | struct pipe_inode_info *pipe, |
| 390 | size_t count, |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 391 | unsigned int flags) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 392 | { |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 393 | struct xfs_inode *ip = XFS_I(infilp->f_mapping->host); |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 394 | int ioflags = 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 395 | ssize_t ret; |
| 396 | |
Bill O'Donnell | ff6d6af | 2015-10-12 18:21:22 +1100 | [diff] [blame] | 397 | XFS_STATS_INC(ip->i_mount, xs_read_calls); |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 398 | |
| 399 | if (infilp->f_mode & FMODE_NOCMTIME) |
Dave Chinner | b92cc59f6 | 2014-08-04 13:28:20 +1000 | [diff] [blame] | 400 | ioflags |= XFS_IO_INVIS; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 401 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 402 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 403 | return -EIO; |
| 404 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 405 | trace_xfs_file_splice_read(ip, count, *ppos, ioflags); |
| 406 | |
Dave Chinner | a6d7636 | 2016-01-04 16:28:25 +1100 | [diff] [blame] | 407 | /* |
| 408 | * DAX inodes cannot ues the page cache for splice, so we have to push |
| 409 | * them through the VFS IO path. This means it goes through |
| 410 | * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we |
| 411 | * cannot lock the splice operation at this level for DAX inodes. |
| 412 | */ |
| 413 | if (IS_DAX(VFS_I(ip))) { |
| 414 | ret = default_file_splice_read(infilp, ppos, pipe, count, |
| 415 | flags); |
| 416 | goto out; |
| 417 | } |
| 418 | |
| 419 | xfs_rw_ilock(ip, XFS_IOLOCK_SHARED); |
| 420 | ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); |
| 421 | xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED); |
| 422 | out: |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 423 | if (ret > 0) |
Bill O'Donnell | ff6d6af | 2015-10-12 18:21:22 +1100 | [diff] [blame] | 424 | XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 425 | return ret; |
| 426 | } |
| 427 | |
Dave Chinner | 4c5cfd1 | 2011-01-11 10:14:16 +1100 | [diff] [blame] | 428 | /* |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 429 | * This routine is called to handle zeroing any space in the last block of the |
| 430 | * file that is beyond the EOF. We do this since the size is being increased |
| 431 | * without writing anything to that block and we don't want to read the |
| 432 | * garbage on the disk. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 433 | */ |
| 434 | STATIC int /* error (positive) */ |
| 435 | xfs_zero_last_block( |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 436 | struct xfs_inode *ip, |
| 437 | xfs_fsize_t offset, |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 438 | xfs_fsize_t isize, |
| 439 | bool *did_zeroing) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 440 | { |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 441 | struct xfs_mount *mp = ip->i_mount; |
| 442 | xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize); |
| 443 | int zero_offset = XFS_B_FSB_OFFSET(mp, isize); |
| 444 | int zero_len; |
| 445 | int nimaps = 1; |
| 446 | int error = 0; |
| 447 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 448 | |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 449 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 450 | error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 451 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 452 | if (error) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 453 | return error; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 454 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 455 | ASSERT(nimaps > 0); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 456 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 457 | /* |
| 458 | * If the block underlying isize is just a hole, then there |
| 459 | * is nothing to zero. |
| 460 | */ |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 461 | if (imap.br_startblock == HOLESTARTBLOCK) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 462 | return 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 463 | |
| 464 | zero_len = mp->m_sb.sb_blocksize - zero_offset; |
| 465 | if (isize + zero_len > offset) |
| 466 | zero_len = offset - isize; |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 467 | *did_zeroing = true; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 468 | return xfs_iozero(ip, isize, zero_len); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 469 | } |
| 470 | |
| 471 | /* |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 472 | * Zero any on disk space between the current EOF and the new, larger EOF. |
| 473 | * |
| 474 | * This handles the normal case of zeroing the remainder of the last block in |
| 475 | * the file and the unusual case of zeroing blocks out beyond the size of the |
| 476 | * file. This second case only happens with fixed size extents and when the |
| 477 | * system crashes before the inode size was updated but after blocks were |
| 478 | * allocated. |
| 479 | * |
| 480 | * Expects the iolock to be held exclusive, and will take the ilock internally. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 481 | */ |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 482 | int /* error (positive) */ |
| 483 | xfs_zero_eof( |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 484 | struct xfs_inode *ip, |
| 485 | xfs_off_t offset, /* starting I/O offset */ |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 486 | xfs_fsize_t isize, /* current inode size */ |
| 487 | bool *did_zeroing) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 488 | { |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 489 | struct xfs_mount *mp = ip->i_mount; |
| 490 | xfs_fileoff_t start_zero_fsb; |
| 491 | xfs_fileoff_t end_zero_fsb; |
| 492 | xfs_fileoff_t zero_count_fsb; |
| 493 | xfs_fileoff_t last_fsb; |
| 494 | xfs_fileoff_t zero_off; |
| 495 | xfs_fsize_t zero_len; |
| 496 | int nimaps; |
| 497 | int error = 0; |
| 498 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 499 | |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 500 | ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 501 | ASSERT(offset > isize); |
| 502 | |
Brian Foster | 0a50f16 | 2015-10-12 16:02:08 +1100 | [diff] [blame] | 503 | trace_xfs_zero_eof(ip, isize, offset - isize); |
| 504 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 505 | /* |
| 506 | * First handle zeroing the block on which isize resides. |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 507 | * |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 508 | * We only zero a part of that block so it is handled specially. |
| 509 | */ |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 510 | if (XFS_B_FSB_OFFSET(mp, isize) != 0) { |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 511 | error = xfs_zero_last_block(ip, offset, isize, did_zeroing); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 512 | if (error) |
| 513 | return error; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 514 | } |
| 515 | |
| 516 | /* |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 517 | * Calculate the range between the new size and the old where blocks |
| 518 | * needing to be zeroed may exist. |
| 519 | * |
| 520 | * To get the block where the last byte in the file currently resides, |
| 521 | * we need to subtract one from the size and truncate back to a block |
| 522 | * boundary. We subtract 1 in case the size is exactly on a block |
| 523 | * boundary. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 524 | */ |
| 525 | last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1; |
| 526 | start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize); |
| 527 | end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1); |
| 528 | ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb); |
| 529 | if (last_fsb == end_zero_fsb) { |
| 530 | /* |
| 531 | * The size was only incremented on its last block. |
| 532 | * We took care of that above, so just return. |
| 533 | */ |
| 534 | return 0; |
| 535 | } |
| 536 | |
| 537 | ASSERT(start_zero_fsb <= end_zero_fsb); |
| 538 | while (start_zero_fsb <= end_zero_fsb) { |
| 539 | nimaps = 1; |
| 540 | zero_count_fsb = end_zero_fsb - start_zero_fsb + 1; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 541 | |
| 542 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 543 | error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb, |
| 544 | &imap, &nimaps, 0); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 545 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 546 | if (error) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 547 | return error; |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 548 | |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 549 | ASSERT(nimaps > 0); |
| 550 | |
| 551 | if (imap.br_state == XFS_EXT_UNWRITTEN || |
| 552 | imap.br_startblock == HOLESTARTBLOCK) { |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 553 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
| 554 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); |
| 555 | continue; |
| 556 | } |
| 557 | |
| 558 | /* |
| 559 | * There are blocks we need to zero. |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 560 | */ |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 561 | zero_off = XFS_FSB_TO_B(mp, start_zero_fsb); |
| 562 | zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount); |
| 563 | |
| 564 | if ((zero_off + zero_len) > offset) |
| 565 | zero_len = offset - zero_off; |
| 566 | |
| 567 | error = xfs_iozero(ip, zero_off, zero_len); |
Christoph Hellwig | 193aec1 | 2012-03-27 10:34:49 -0400 | [diff] [blame] | 568 | if (error) |
| 569 | return error; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 570 | |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 571 | *did_zeroing = true; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 572 | start_zero_fsb = imap.br_startoff + imap.br_blockcount; |
| 573 | ASSERT(start_zero_fsb <= (end_zero_fsb + 1)); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 574 | } |
| 575 | |
| 576 | return 0; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 577 | } |
| 578 | |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 579 | /* |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 580 | * Common pre-write limit and setup checks. |
| 581 | * |
Christoph Hellwig | 5bf1f26 | 2011-12-18 20:00:13 +0000 | [diff] [blame] | 582 | * Called with the iolocked held either shared and exclusive according to |
| 583 | * @iolock, and returns with it held. Might upgrade the iolock to exclusive |
| 584 | * if called for a direct write beyond i_size. |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 585 | */ |
| 586 | STATIC ssize_t |
| 587 | xfs_file_aio_write_checks( |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 588 | struct kiocb *iocb, |
| 589 | struct iov_iter *from, |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 590 | int *iolock) |
| 591 | { |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 592 | struct file *file = iocb->ki_filp; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 593 | struct inode *inode = file->f_mapping->host; |
| 594 | struct xfs_inode *ip = XFS_I(inode); |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 595 | ssize_t error = 0; |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 596 | size_t count = iov_iter_count(from); |
Brian Foster | 3136e8b | 2015-10-12 16:02:05 +1100 | [diff] [blame] | 597 | bool drained_dio = false; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 598 | |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 599 | restart: |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 600 | error = generic_write_checks(iocb, from); |
| 601 | if (error <= 0) |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 602 | return error; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 603 | |
Christoph Hellwig | 21c3ea1 | 2015-04-13 11:38:29 +1000 | [diff] [blame] | 604 | error = xfs_break_layouts(inode, iolock, true); |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 605 | if (error) |
| 606 | return error; |
| 607 | |
Jan Kara | a6de82c | 2015-05-21 16:05:56 +0200 | [diff] [blame] | 608 | /* For changing security info in file_remove_privs() we need i_mutex */ |
| 609 | if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) { |
| 610 | xfs_rw_iunlock(ip, *iolock); |
| 611 | *iolock = XFS_IOLOCK_EXCL; |
| 612 | xfs_rw_ilock(ip, *iolock); |
| 613 | goto restart; |
| 614 | } |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 615 | /* |
| 616 | * If the offset is beyond the size of the file, we need to zero any |
| 617 | * blocks that fall between the existing EOF and the start of this |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 618 | * write. If zeroing is needed and we are currently holding the |
Christoph Hellwig | 467f789 | 2012-03-27 10:34:47 -0400 | [diff] [blame] | 619 | * iolock shared, we need to update it to exclusive which implies |
| 620 | * having to redo all checks before. |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 621 | * |
| 622 | * We need to serialise against EOF updates that occur in IO |
| 623 | * completions here. We want to make sure that nobody is changing the |
| 624 | * size while we do this check until we have placed an IO barrier (i.e. |
| 625 | * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. |
| 626 | * The spinlock effectively forms a memory barrier once we have the |
| 627 | * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value |
| 628 | * and hence be able to correctly determine if we need to run zeroing. |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 629 | */ |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 630 | spin_lock(&ip->i_flags_lock); |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 631 | if (iocb->ki_pos > i_size_read(inode)) { |
Dave Chinner | 5885ebd | 2015-02-23 22:37:08 +1100 | [diff] [blame] | 632 | bool zero = false; |
| 633 | |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 634 | spin_unlock(&ip->i_flags_lock); |
Brian Foster | 3136e8b | 2015-10-12 16:02:05 +1100 | [diff] [blame] | 635 | if (!drained_dio) { |
| 636 | if (*iolock == XFS_IOLOCK_SHARED) { |
| 637 | xfs_rw_iunlock(ip, *iolock); |
| 638 | *iolock = XFS_IOLOCK_EXCL; |
| 639 | xfs_rw_ilock(ip, *iolock); |
| 640 | iov_iter_reexpand(from, count); |
| 641 | } |
Dave Chinner | 40c63fb | 2015-04-16 22:03:17 +1000 | [diff] [blame] | 642 | /* |
| 643 | * We now have an IO submission barrier in place, but |
| 644 | * AIO can do EOF updates during IO completion and hence |
| 645 | * we now need to wait for all of them to drain. Non-AIO |
| 646 | * DIO will have drained before we are given the |
| 647 | * XFS_IOLOCK_EXCL, and so for most cases this wait is a |
| 648 | * no-op. |
| 649 | */ |
| 650 | inode_dio_wait(inode); |
Brian Foster | 3136e8b | 2015-10-12 16:02:05 +1100 | [diff] [blame] | 651 | drained_dio = true; |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 652 | goto restart; |
| 653 | } |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 654 | 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] | 655 | if (error) |
| 656 | return error; |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 657 | } else |
| 658 | spin_unlock(&ip->i_flags_lock); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 659 | |
| 660 | /* |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 661 | * Updating the timestamps will grab the ilock again from |
| 662 | * xfs_fs_dirty_inode, so we have to call it after dropping the |
| 663 | * lock above. Eventually we should look into a way to avoid |
| 664 | * the pointless lock roundtrip. |
| 665 | */ |
Josef Bacik | c3b2da3 | 2012-03-26 09:59:21 -0400 | [diff] [blame] | 666 | if (likely(!(file->f_mode & FMODE_NOCMTIME))) { |
| 667 | error = file_update_time(file); |
| 668 | if (error) |
| 669 | return error; |
| 670 | } |
Christoph Hellwig | 8a9c998 | 2012-02-29 09:53:52 +0000 | [diff] [blame] | 671 | |
| 672 | /* |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 673 | * If we're writing the file then make sure to clear the setuid and |
| 674 | * setgid bits if the process is not being run by root. This keeps |
| 675 | * people from modifying setuid and setgid binaries. |
| 676 | */ |
Jan Kara | a6de82c | 2015-05-21 16:05:56 +0200 | [diff] [blame] | 677 | if (!IS_NOSEC(inode)) |
| 678 | return file_remove_privs(file); |
| 679 | return 0; |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 680 | } |
| 681 | |
| 682 | /* |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 683 | * xfs_file_dio_aio_write - handle direct IO writes |
| 684 | * |
| 685 | * Lock the inode appropriately to prepare for and issue a direct IO write. |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 686 | * 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] | 687 | * follow locking changes and looping. |
| 688 | * |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 689 | * If there are cached pages or we're extending the file, we need IOLOCK_EXCL |
| 690 | * until we're sure the bytes at the new EOF have been zeroed and/or the cached |
| 691 | * pages are flushed out. |
| 692 | * |
| 693 | * In most cases the direct IO writes will be done holding IOLOCK_SHARED |
| 694 | * allowing them to be done in parallel with reads and other direct IO writes. |
| 695 | * However, if the IO is not aligned to filesystem blocks, the direct IO layer |
| 696 | * needs to do sub-block zeroing and that requires serialisation against other |
| 697 | * direct IOs to the same block. In this case we need to serialise the |
| 698 | * submission of the unaligned IOs so that we don't get racing block zeroing in |
| 699 | * the dio layer. To avoid the problem with aio, we also need to wait for |
| 700 | * outstanding IOs to complete so that unwritten extent conversion is completed |
| 701 | * before we try to map the overlapping block. This is currently implemented by |
Christoph Hellwig | 4a06fd2 | 2011-08-23 08:28:13 +0000 | [diff] [blame] | 702 | * hitting it with a big hammer (i.e. inode_dio_wait()). |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 703 | * |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 704 | * Returns with locks held indicated by @iolock and errors indicated by |
| 705 | * negative return values. |
| 706 | */ |
| 707 | STATIC ssize_t |
| 708 | xfs_file_dio_aio_write( |
| 709 | struct kiocb *iocb, |
Al Viro | b318891 | 2014-04-02 07:06:30 -0400 | [diff] [blame] | 710 | struct iov_iter *from) |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 711 | { |
| 712 | struct file *file = iocb->ki_filp; |
| 713 | struct address_space *mapping = file->f_mapping; |
| 714 | struct inode *inode = mapping->host; |
| 715 | struct xfs_inode *ip = XFS_I(inode); |
| 716 | struct xfs_mount *mp = ip->i_mount; |
| 717 | ssize_t ret = 0; |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 718 | int unaligned_io = 0; |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 719 | int iolock; |
Al Viro | b318891 | 2014-04-02 07:06:30 -0400 | [diff] [blame] | 720 | size_t count = iov_iter_count(from); |
| 721 | loff_t pos = iocb->ki_pos; |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 722 | loff_t end; |
| 723 | struct iov_iter data; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 724 | struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ? |
| 725 | mp->m_rtdev_targp : mp->m_ddev_targp; |
| 726 | |
Eric Sandeen | 7c71ee7 | 2014-01-21 16:46:23 -0600 | [diff] [blame] | 727 | /* DIO must be aligned to device logical sector size */ |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 728 | if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 729 | return -EINVAL; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 730 | |
Eric Sandeen | 7c71ee7 | 2014-01-21 16:46:23 -0600 | [diff] [blame] | 731 | /* "unaligned" here means not aligned to a filesystem block */ |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 732 | if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask)) |
| 733 | unaligned_io = 1; |
| 734 | |
Dave Chinner | 7271d24 | 2011-08-25 07:17:02 +0000 | [diff] [blame] | 735 | /* |
| 736 | * We don't need to take an exclusive lock unless there page cache needs |
| 737 | * to be invalidated or unaligned IO is being executed. We don't need to |
| 738 | * consider the EOF extension case here because |
| 739 | * xfs_file_aio_write_checks() will relock the inode as necessary for |
| 740 | * EOF zeroing cases and fill out the new inode size as appropriate. |
| 741 | */ |
| 742 | if (unaligned_io || mapping->nrpages) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 743 | iolock = XFS_IOLOCK_EXCL; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 744 | else |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 745 | iolock = XFS_IOLOCK_SHARED; |
| 746 | xfs_rw_ilock(ip, iolock); |
Christoph Hellwig | c58cb16 | 2011-08-27 14:42:53 +0000 | [diff] [blame] | 747 | |
| 748 | /* |
| 749 | * Recheck if there are cached pages that need invalidate after we got |
| 750 | * the iolock to protect against other threads adding new pages while |
| 751 | * we were waiting for the iolock. |
| 752 | */ |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 753 | if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) { |
| 754 | xfs_rw_iunlock(ip, iolock); |
| 755 | iolock = XFS_IOLOCK_EXCL; |
| 756 | xfs_rw_ilock(ip, iolock); |
Christoph Hellwig | c58cb16 | 2011-08-27 14:42:53 +0000 | [diff] [blame] | 757 | } |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 758 | |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 759 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 760 | if (ret) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 761 | goto out; |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 762 | count = iov_iter_count(from); |
| 763 | pos = iocb->ki_pos; |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 764 | end = pos + count - 1; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 765 | |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 766 | /* |
| 767 | * See xfs_file_read_iter() for why we do a full-file flush here. |
| 768 | */ |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 769 | if (mapping->nrpages) { |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 770 | ret = filemap_write_and_wait(VFS_I(ip)->i_mapping); |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 771 | if (ret) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 772 | goto out; |
Dave Chinner | 834ffca | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 773 | /* |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 774 | * Invalidate whole pages. This can return an error if we fail |
| 775 | * to invalidate a page, but this should never happen on XFS. |
| 776 | * Warn if it does fail. |
Dave Chinner | 834ffca | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 777 | */ |
Brian Foster | 3d751af | 2015-08-19 10:35:04 +1000 | [diff] [blame] | 778 | ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping); |
Dave Chinner | 834ffca | 2014-09-02 12:12:52 +1000 | [diff] [blame] | 779 | WARN_ON_ONCE(ret); |
| 780 | ret = 0; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 781 | } |
| 782 | |
Dave Chinner | eda7798 | 2011-01-11 10:22:40 +1100 | [diff] [blame] | 783 | /* |
| 784 | * If we are doing unaligned IO, wait for all other IO to drain, |
| 785 | * otherwise demote the lock if we had to flush cached pages |
| 786 | */ |
| 787 | if (unaligned_io) |
Christoph Hellwig | 4a06fd2 | 2011-08-23 08:28:13 +0000 | [diff] [blame] | 788 | inode_dio_wait(inode); |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 789 | else if (iolock == XFS_IOLOCK_EXCL) { |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 790 | xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL); |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 791 | iolock = XFS_IOLOCK_SHARED; |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 792 | } |
| 793 | |
| 794 | trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0); |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 795 | |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 796 | data = *from; |
Linus Torvalds | 1aef882 | 2015-04-24 07:08:41 -0700 | [diff] [blame] | 797 | ret = mapping->a_ops->direct_IO(iocb, &data, pos); |
Dave Chinner | 0cefb29 | 2015-04-16 22:03:27 +1000 | [diff] [blame] | 798 | |
| 799 | /* see generic_file_direct_write() for why this is necessary */ |
| 800 | if (mapping->nrpages) { |
| 801 | invalidate_inode_pages2_range(mapping, |
| 802 | pos >> PAGE_CACHE_SHIFT, |
| 803 | end >> PAGE_CACHE_SHIFT); |
| 804 | } |
| 805 | |
| 806 | if (ret > 0) { |
| 807 | pos += ret; |
| 808 | iov_iter_advance(from, ret); |
| 809 | iocb->ki_pos = pos; |
| 810 | } |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 811 | out: |
| 812 | xfs_rw_iunlock(ip, iolock); |
| 813 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 814 | /* |
| 815 | * No fallback to buffered IO on errors for XFS. DAX can result in |
| 816 | * partial writes, but direct IO will either complete fully or fail. |
| 817 | */ |
| 818 | ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip))); |
Dave Chinner | f0d26e8 | 2011-01-11 10:15:36 +1100 | [diff] [blame] | 819 | return ret; |
| 820 | } |
| 821 | |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 822 | STATIC ssize_t |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 823 | xfs_file_buffered_aio_write( |
| 824 | struct kiocb *iocb, |
Al Viro | b318891 | 2014-04-02 07:06:30 -0400 | [diff] [blame] | 825 | struct iov_iter *from) |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 826 | { |
| 827 | struct file *file = iocb->ki_filp; |
| 828 | struct address_space *mapping = file->f_mapping; |
| 829 | struct inode *inode = mapping->host; |
| 830 | struct xfs_inode *ip = XFS_I(inode); |
| 831 | ssize_t ret; |
| 832 | int enospc = 0; |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 833 | int iolock = XFS_IOLOCK_EXCL; |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 834 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 835 | xfs_rw_ilock(ip, iolock); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 836 | |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 837 | ret = xfs_file_aio_write_checks(iocb, from, &iolock); |
Dave Chinner | 4d8d158 | 2011-01-11 10:23:42 +1100 | [diff] [blame] | 838 | if (ret) |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 839 | goto out; |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 840 | |
| 841 | /* We can write back this queue in page reclaim */ |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 842 | current->backing_dev_info = inode_to_bdi(inode); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 843 | |
| 844 | write_retry: |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 845 | trace_xfs_file_buffered_write(ip, iov_iter_count(from), |
| 846 | iocb->ki_pos, 0); |
| 847 | ret = generic_perform_write(file, from, iocb->ki_pos); |
Al Viro | 0a64bc2 | 2014-02-11 22:25:22 -0500 | [diff] [blame] | 848 | if (likely(ret >= 0)) |
Al Viro | 99733fa | 2015-04-07 14:25:18 -0400 | [diff] [blame] | 849 | iocb->ki_pos += ret; |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 850 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 851 | /* |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 852 | * If we hit a space limit, try to free up some lingering preallocated |
| 853 | * space before returning an error. In the case of ENOSPC, first try to |
| 854 | * write back all dirty inodes to free up some of the excess reserved |
| 855 | * metadata space. This reduces the chances that the eofblocks scan |
| 856 | * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this |
| 857 | * also behaves as a filter to prevent too many eofblocks scans from |
| 858 | * running at the same time. |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 859 | */ |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 860 | if (ret == -EDQUOT && !enospc) { |
| 861 | enospc = xfs_inode_free_quota_eofblocks(ip); |
| 862 | if (enospc) |
| 863 | goto write_retry; |
| 864 | } else if (ret == -ENOSPC && !enospc) { |
| 865 | struct xfs_eofblocks eofb = {0}; |
| 866 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 867 | enospc = 1; |
Dave Chinner | 9aa0500 | 2012-10-08 21:56:04 +1100 | [diff] [blame] | 868 | xfs_flush_inodes(ip->i_mount); |
Brian Foster | dc06f398 | 2014-07-24 19:49:28 +1000 | [diff] [blame] | 869 | eofb.eof_scan_owner = ip->i_ino; /* for locking */ |
| 870 | eofb.eof_flags = XFS_EOF_FLAGS_SYNC; |
| 871 | xfs_icache_free_eofblocks(ip->i_mount, &eofb); |
Dave Chinner | 9aa0500 | 2012-10-08 21:56:04 +1100 | [diff] [blame] | 872 | goto write_retry; |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 873 | } |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 874 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 875 | current->backing_dev_info = NULL; |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 876 | out: |
| 877 | xfs_rw_iunlock(ip, iolock); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 878 | return ret; |
| 879 | } |
| 880 | |
| 881 | STATIC ssize_t |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 882 | xfs_file_write_iter( |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 883 | struct kiocb *iocb, |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 884 | struct iov_iter *from) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 885 | { |
| 886 | struct file *file = iocb->ki_filp; |
| 887 | struct address_space *mapping = file->f_mapping; |
| 888 | struct inode *inode = mapping->host; |
Christoph Hellwig | 00258e3 | 2010-02-15 09:44:47 +0000 | [diff] [blame] | 889 | struct xfs_inode *ip = XFS_I(inode); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 890 | ssize_t ret; |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 891 | size_t ocount = iov_iter_count(from); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 892 | |
Bill O'Donnell | ff6d6af | 2015-10-12 18:21:22 +1100 | [diff] [blame] | 893 | XFS_STATS_INC(ip->i_mount, xs_write_calls); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 894 | |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 895 | if (ocount == 0) |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 896 | return 0; |
| 897 | |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 898 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 899 | return -EIO; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 900 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 901 | if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode)) |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 902 | ret = xfs_file_dio_aio_write(iocb, from); |
Dave Chinner | 637bbc7 | 2011-01-11 10:17:30 +1100 | [diff] [blame] | 903 | else |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 904 | ret = xfs_file_buffered_aio_write(iocb, from); |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 905 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 906 | if (ret > 0) { |
| 907 | ssize_t err; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 908 | |
Bill O'Donnell | ff6d6af | 2015-10-12 18:21:22 +1100 | [diff] [blame] | 909 | XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret); |
Christoph Hellwig | ce7ae151 | 2011-12-18 20:00:11 +0000 | [diff] [blame] | 910 | |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 911 | /* Handle various SYNC-type writes */ |
Al Viro | d311d79 | 2014-02-09 15:18:09 -0500 | [diff] [blame] | 912 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
Christoph Hellwig | d060646 | 2011-12-18 20:00:14 +0000 | [diff] [blame] | 913 | if (err < 0) |
| 914 | ret = err; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 915 | } |
Dave Chinner | a363f0c | 2011-01-11 10:13:53 +1100 | [diff] [blame] | 916 | return ret; |
Christoph Hellwig | dda35b8 | 2010-02-15 09:44:46 +0000 | [diff] [blame] | 917 | } |
| 918 | |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 919 | #define XFS_FALLOC_FL_SUPPORTED \ |
| 920 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ |
| 921 | FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \ |
| 922 | FALLOC_FL_INSERT_RANGE) |
| 923 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 924 | STATIC long |
| 925 | xfs_file_fallocate( |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 926 | struct file *file, |
| 927 | int mode, |
| 928 | loff_t offset, |
| 929 | loff_t len) |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 930 | { |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 931 | struct inode *inode = file_inode(file); |
| 932 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 933 | long error; |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 934 | enum xfs_prealloc_flags flags = 0; |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 935 | uint iolock = XFS_IOLOCK_EXCL; |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 936 | loff_t new_size = 0; |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 937 | bool do_file_insert = 0; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 938 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 939 | if (!S_ISREG(inode->i_mode)) |
| 940 | return -EINVAL; |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 941 | if (mode & ~XFS_FALLOC_FL_SUPPORTED) |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 942 | return -EOPNOTSUPP; |
| 943 | |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 944 | xfs_ilock(ip, iolock); |
Christoph Hellwig | 21c3ea1 | 2015-04-13 11:38:29 +1000 | [diff] [blame] | 945 | error = xfs_break_layouts(inode, &iolock, false); |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 946 | if (error) |
| 947 | goto out_unlock; |
| 948 | |
Dave Chinner | e8e9ad4 | 2015-02-23 21:45:32 +1100 | [diff] [blame] | 949 | xfs_ilock(ip, XFS_MMAPLOCK_EXCL); |
| 950 | iolock |= XFS_MMAPLOCK_EXCL; |
| 951 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 952 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
| 953 | error = xfs_free_file_space(ip, offset, len); |
| 954 | if (error) |
| 955 | goto out_unlock; |
Namjae Jeon | e1d8fb8 | 2014-02-24 10:58:19 +1100 | [diff] [blame] | 956 | } else if (mode & FALLOC_FL_COLLAPSE_RANGE) { |
| 957 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; |
| 958 | |
| 959 | if (offset & blksize_mask || len & blksize_mask) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 960 | error = -EINVAL; |
Namjae Jeon | e1d8fb8 | 2014-02-24 10:58:19 +1100 | [diff] [blame] | 961 | goto out_unlock; |
| 962 | } |
| 963 | |
Lukas Czerner | 23fffa9 | 2014-04-12 09:56:41 -0400 | [diff] [blame] | 964 | /* |
| 965 | * There is no need to overlap collapse range with EOF, |
| 966 | * in which case it is effectively a truncate operation |
| 967 | */ |
| 968 | if (offset + len >= i_size_read(inode)) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 969 | error = -EINVAL; |
Lukas Czerner | 23fffa9 | 2014-04-12 09:56:41 -0400 | [diff] [blame] | 970 | goto out_unlock; |
| 971 | } |
| 972 | |
Namjae Jeon | e1d8fb8 | 2014-02-24 10:58:19 +1100 | [diff] [blame] | 973 | new_size = i_size_read(inode) - len; |
| 974 | |
| 975 | error = xfs_collapse_file_space(ip, offset, len); |
| 976 | if (error) |
| 977 | goto out_unlock; |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 978 | } else if (mode & FALLOC_FL_INSERT_RANGE) { |
| 979 | unsigned blksize_mask = (1 << inode->i_blkbits) - 1; |
| 980 | |
| 981 | new_size = i_size_read(inode) + len; |
| 982 | if (offset & blksize_mask || len & blksize_mask) { |
| 983 | error = -EINVAL; |
| 984 | goto out_unlock; |
| 985 | } |
| 986 | |
| 987 | /* check the new inode size does not wrap through zero */ |
| 988 | if (new_size > inode->i_sb->s_maxbytes) { |
| 989 | error = -EFBIG; |
| 990 | goto out_unlock; |
| 991 | } |
| 992 | |
| 993 | /* Offset should be less than i_size */ |
| 994 | if (offset >= i_size_read(inode)) { |
| 995 | error = -EINVAL; |
| 996 | goto out_unlock; |
| 997 | } |
| 998 | do_file_insert = 1; |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 999 | } else { |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 1000 | flags |= XFS_PREALLOC_SET; |
| 1001 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 1002 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
| 1003 | offset + len > i_size_read(inode)) { |
| 1004 | new_size = offset + len; |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1005 | error = inode_newsize_ok(inode, new_size); |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 1006 | if (error) |
| 1007 | goto out_unlock; |
| 1008 | } |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1009 | |
Lukas Czerner | 376ba31 | 2014-03-13 19:07:58 +1100 | [diff] [blame] | 1010 | if (mode & FALLOC_FL_ZERO_RANGE) |
| 1011 | error = xfs_zero_file_space(ip, offset, len); |
| 1012 | else |
| 1013 | error = xfs_alloc_file_space(ip, offset, len, |
| 1014 | XFS_BMAPI_PREALLOC); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1015 | if (error) |
| 1016 | goto out_unlock; |
| 1017 | } |
| 1018 | |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 1019 | if (file->f_flags & O_DSYNC) |
Christoph Hellwig | 8add71c | 2015-02-02 09:53:56 +1100 | [diff] [blame] | 1020 | flags |= XFS_PREALLOC_SYNC; |
| 1021 | |
| 1022 | error = xfs_update_prealloc_flags(ip, flags); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1023 | if (error) |
| 1024 | goto out_unlock; |
| 1025 | |
| 1026 | /* Change file size if needed */ |
| 1027 | if (new_size) { |
| 1028 | struct iattr iattr; |
| 1029 | |
| 1030 | iattr.ia_valid = ATTR_SIZE; |
| 1031 | iattr.ia_size = new_size; |
Christoph Hellwig | 83aee9e | 2013-10-12 00:55:07 -0700 | [diff] [blame] | 1032 | error = xfs_setattr_size(ip, &iattr); |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 1033 | if (error) |
| 1034 | goto out_unlock; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1035 | } |
| 1036 | |
Namjae Jeon | a904b1c | 2015-03-25 15:08:56 +1100 | [diff] [blame] | 1037 | /* |
| 1038 | * Perform hole insertion now that the file size has been |
| 1039 | * updated so that if we crash during the operation we don't |
| 1040 | * leave shifted extents past EOF and hence losing access to |
| 1041 | * the data that is contained within them. |
| 1042 | */ |
| 1043 | if (do_file_insert) |
| 1044 | error = xfs_insert_file_space(ip, offset, len); |
| 1045 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1046 | out_unlock: |
Christoph Hellwig | 781355c | 2015-02-16 11:59:50 +1100 | [diff] [blame] | 1047 | xfs_iunlock(ip, iolock); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1048 | return error; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1049 | } |
| 1050 | |
| 1051 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1052 | STATIC int |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1053 | xfs_file_open( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | struct inode *inode, |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1055 | struct file *file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | { |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1057 | 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] | 1058 | return -EFBIG; |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1059 | if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb))) |
| 1060 | return -EIO; |
| 1061 | return 0; |
| 1062 | } |
| 1063 | |
| 1064 | STATIC int |
| 1065 | xfs_dir_open( |
| 1066 | struct inode *inode, |
| 1067 | struct file *file) |
| 1068 | { |
| 1069 | struct xfs_inode *ip = XFS_I(inode); |
| 1070 | int mode; |
| 1071 | int error; |
| 1072 | |
| 1073 | error = xfs_file_open(inode, file); |
| 1074 | if (error) |
| 1075 | return error; |
| 1076 | |
| 1077 | /* |
| 1078 | * If there are any blocks, read-ahead block 0 as we're almost |
| 1079 | * certain to have the next operation be a read there. |
| 1080 | */ |
Christoph Hellwig | 309ecac | 2013-12-06 12:30:09 -0800 | [diff] [blame] | 1081 | mode = xfs_ilock_data_map_shared(ip); |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1082 | if (ip->i_d.di_nextents > 0) |
Eric Sandeen | 9df2dd0 | 2014-04-14 19:01:59 +1000 | [diff] [blame] | 1083 | xfs_dir3_data_readahead(ip, 0, -1); |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1084 | xfs_iunlock(ip, mode); |
| 1085 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1086 | } |
| 1087 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1088 | STATIC int |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1089 | xfs_file_release( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | struct inode *inode, |
| 1091 | struct file *filp) |
| 1092 | { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1093 | return xfs_release(XFS_I(inode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | } |
| 1095 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | STATIC int |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1097 | xfs_file_readdir( |
Al Viro | b822755 | 2013-05-22 17:07:56 -0400 | [diff] [blame] | 1098 | struct file *file, |
| 1099 | struct dir_context *ctx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | { |
Al Viro | b822755 | 2013-05-22 17:07:56 -0400 | [diff] [blame] | 1101 | struct inode *inode = file_inode(file); |
Christoph Hellwig | 739bfb2 | 2007-08-29 10:58:01 +1000 | [diff] [blame] | 1102 | xfs_inode_t *ip = XFS_I(inode); |
Christoph Hellwig | 051e7cd | 2007-08-28 13:58:24 +1000 | [diff] [blame] | 1103 | size_t bufsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | |
Christoph Hellwig | 051e7cd | 2007-08-28 13:58:24 +1000 | [diff] [blame] | 1105 | /* |
| 1106 | * The Linux API doesn't pass down the total size of the buffer |
| 1107 | * we read into down to the filesystem. With the filldir concept |
| 1108 | * it's not needed for correct information, but the XFS dir2 leaf |
| 1109 | * code wants an estimate of the buffer size to calculate it's |
| 1110 | * readahead window and size the buffers used for mapping to |
| 1111 | * physical blocks. |
| 1112 | * |
| 1113 | * Try to give it an estimate that's good enough, maybe at some |
| 1114 | * point we can change the ->readdir prototype to include the |
Eric Sandeen | a9cc799 | 2010-02-03 17:50:13 +0000 | [diff] [blame] | 1115 | * buffer size. For now we use the current glibc buffer size. |
Christoph Hellwig | 051e7cd | 2007-08-28 13:58:24 +1000 | [diff] [blame] | 1116 | */ |
Eric Sandeen | a9cc799 | 2010-02-03 17:50:13 +0000 | [diff] [blame] | 1117 | 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] | 1118 | |
kbuild test robot | 8300475 | 2014-12-01 08:25:28 +1100 | [diff] [blame] | 1119 | return xfs_readdir(ip, ctx, bufsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | } |
| 1121 | |
David Chinner | 4f57dbc | 2007-07-19 16:28:17 +1000 | [diff] [blame] | 1122 | /* |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 1123 | * 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] | 1124 | * to search from page cache for xfs_seek_hole_data(). |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 1125 | */ |
| 1126 | enum { |
| 1127 | HOLE_OFF = 0, |
| 1128 | DATA_OFF, |
| 1129 | }; |
| 1130 | |
| 1131 | /* |
| 1132 | * Lookup the desired type of offset from the given page. |
| 1133 | * |
| 1134 | * On success, return true and the offset argument will point to the |
| 1135 | * start of the region that was found. Otherwise this function will |
| 1136 | * return false and keep the offset argument unchanged. |
| 1137 | */ |
| 1138 | STATIC bool |
| 1139 | xfs_lookup_buffer_offset( |
| 1140 | struct page *page, |
| 1141 | loff_t *offset, |
| 1142 | unsigned int type) |
| 1143 | { |
| 1144 | loff_t lastoff = page_offset(page); |
| 1145 | bool found = false; |
| 1146 | struct buffer_head *bh, *head; |
| 1147 | |
| 1148 | bh = head = page_buffers(page); |
| 1149 | do { |
| 1150 | /* |
| 1151 | * Unwritten extents that have data in the page |
| 1152 | * cache covering them can be identified by the |
| 1153 | * BH_Unwritten state flag. Pages with multiple |
| 1154 | * buffers might have a mix of holes, data and |
| 1155 | * unwritten extents - any buffer with valid |
| 1156 | * data in it should have BH_Uptodate flag set |
| 1157 | * on it. |
| 1158 | */ |
| 1159 | if (buffer_unwritten(bh) || |
| 1160 | buffer_uptodate(bh)) { |
| 1161 | if (type == DATA_OFF) |
| 1162 | found = true; |
| 1163 | } else { |
| 1164 | if (type == HOLE_OFF) |
| 1165 | found = true; |
| 1166 | } |
| 1167 | |
| 1168 | if (found) { |
| 1169 | *offset = lastoff; |
| 1170 | break; |
| 1171 | } |
| 1172 | lastoff += bh->b_size; |
| 1173 | } while ((bh = bh->b_this_page) != head); |
| 1174 | |
| 1175 | return found; |
| 1176 | } |
| 1177 | |
| 1178 | /* |
| 1179 | * This routine is called to find out and return a data or hole offset |
| 1180 | * from the page cache for unwritten extents according to the desired |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1181 | * type for xfs_seek_hole_data(). |
Jeff Liu | d126d43 | 2012-08-21 17:11:57 +0800 | [diff] [blame] | 1182 | * |
| 1183 | * The argument offset is used to tell where we start to search from the |
| 1184 | * page cache. Map is used to figure out the end points of the range to |
| 1185 | * lookup pages. |
| 1186 | * |
| 1187 | * Return true if the desired type of offset was found, and the argument |
| 1188 | * offset is filled with that address. Otherwise, return false and keep |
| 1189 | * offset unchanged. |
| 1190 | */ |
| 1191 | STATIC bool |
| 1192 | xfs_find_get_desired_pgoff( |
| 1193 | struct inode *inode, |
| 1194 | struct xfs_bmbt_irec *map, |
| 1195 | unsigned int type, |
| 1196 | loff_t *offset) |
| 1197 | { |
| 1198 | struct xfs_inode *ip = XFS_I(inode); |
| 1199 | struct xfs_mount *mp = ip->i_mount; |
| 1200 | struct pagevec pvec; |
| 1201 | pgoff_t index; |
| 1202 | pgoff_t end; |
| 1203 | loff_t endoff; |
| 1204 | loff_t startoff = *offset; |
| 1205 | loff_t lastoff = startoff; |
| 1206 | bool found = false; |
| 1207 | |
| 1208 | pagevec_init(&pvec, 0); |
| 1209 | |
| 1210 | index = startoff >> PAGE_CACHE_SHIFT; |
| 1211 | endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount); |
| 1212 | end = endoff >> PAGE_CACHE_SHIFT; |
| 1213 | do { |
| 1214 | int want; |
| 1215 | unsigned nr_pages; |
| 1216 | unsigned int i; |
| 1217 | |
| 1218 | want = min_t(pgoff_t, end - index, PAGEVEC_SIZE); |
| 1219 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index, |
| 1220 | want); |
| 1221 | /* |
| 1222 | * No page mapped into given range. If we are searching holes |
| 1223 | * and if this is the first time we got into the loop, it means |
| 1224 | * that the given offset is landed in a hole, return it. |
| 1225 | * |
| 1226 | * If we have already stepped through some block buffers to find |
| 1227 | * holes but they all contains data. In this case, the last |
| 1228 | * offset is already updated and pointed to the end of the last |
| 1229 | * mapped page, if it does not reach the endpoint to search, |
| 1230 | * that means there should be a hole between them. |
| 1231 | */ |
| 1232 | if (nr_pages == 0) { |
| 1233 | /* Data search found nothing */ |
| 1234 | if (type == DATA_OFF) |
| 1235 | break; |
| 1236 | |
| 1237 | ASSERT(type == HOLE_OFF); |
| 1238 | if (lastoff == startoff || lastoff < endoff) { |
| 1239 | found = true; |
| 1240 | *offset = lastoff; |
| 1241 | } |
| 1242 | break; |
| 1243 | } |
| 1244 | |
| 1245 | /* |
| 1246 | * At lease we found one page. If this is the first time we |
| 1247 | * step into the loop, and if the first page index offset is |
| 1248 | * greater than the given search offset, a hole was found. |
| 1249 | */ |
| 1250 | if (type == HOLE_OFF && lastoff == startoff && |
| 1251 | lastoff < page_offset(pvec.pages[0])) { |
| 1252 | found = true; |
| 1253 | break; |
| 1254 | } |
| 1255 | |
| 1256 | for (i = 0; i < nr_pages; i++) { |
| 1257 | struct page *page = pvec.pages[i]; |
| 1258 | loff_t b_offset; |
| 1259 | |
| 1260 | /* |
| 1261 | * At this point, the page may be truncated or |
| 1262 | * invalidated (changing page->mapping to NULL), |
| 1263 | * or even swizzled back from swapper_space to tmpfs |
| 1264 | * file mapping. However, page->index will not change |
| 1265 | * because we have a reference on the page. |
| 1266 | * |
| 1267 | * Searching done if the page index is out of range. |
| 1268 | * If the current offset is not reaches the end of |
| 1269 | * the specified search range, there should be a hole |
| 1270 | * between them. |
| 1271 | */ |
| 1272 | if (page->index > end) { |
| 1273 | if (type == HOLE_OFF && lastoff < endoff) { |
| 1274 | *offset = lastoff; |
| 1275 | found = true; |
| 1276 | } |
| 1277 | goto out; |
| 1278 | } |
| 1279 | |
| 1280 | lock_page(page); |
| 1281 | /* |
| 1282 | * Page truncated or invalidated(page->mapping == NULL). |
| 1283 | * We can freely skip it and proceed to check the next |
| 1284 | * page. |
| 1285 | */ |
| 1286 | if (unlikely(page->mapping != inode->i_mapping)) { |
| 1287 | unlock_page(page); |
| 1288 | continue; |
| 1289 | } |
| 1290 | |
| 1291 | if (!page_has_buffers(page)) { |
| 1292 | unlock_page(page); |
| 1293 | continue; |
| 1294 | } |
| 1295 | |
| 1296 | found = xfs_lookup_buffer_offset(page, &b_offset, type); |
| 1297 | if (found) { |
| 1298 | /* |
| 1299 | * The found offset may be less than the start |
| 1300 | * point to search if this is the first time to |
| 1301 | * come here. |
| 1302 | */ |
| 1303 | *offset = max_t(loff_t, startoff, b_offset); |
| 1304 | unlock_page(page); |
| 1305 | goto out; |
| 1306 | } |
| 1307 | |
| 1308 | /* |
| 1309 | * We either searching data but nothing was found, or |
| 1310 | * searching hole but found a data buffer. In either |
| 1311 | * case, probably the next page contains the desired |
| 1312 | * things, update the last offset to it so. |
| 1313 | */ |
| 1314 | lastoff = page_offset(page) + PAGE_SIZE; |
| 1315 | unlock_page(page); |
| 1316 | } |
| 1317 | |
| 1318 | /* |
| 1319 | * The number of returned pages less than our desired, search |
| 1320 | * done. In this case, nothing was found for searching data, |
| 1321 | * but we found a hole behind the last offset. |
| 1322 | */ |
| 1323 | if (nr_pages < want) { |
| 1324 | if (type == HOLE_OFF) { |
| 1325 | *offset = lastoff; |
| 1326 | found = true; |
| 1327 | } |
| 1328 | break; |
| 1329 | } |
| 1330 | |
| 1331 | index = pvec.pages[i - 1]->index + 1; |
| 1332 | pagevec_release(&pvec); |
| 1333 | } while (index <= end); |
| 1334 | |
| 1335 | out: |
| 1336 | pagevec_release(&pvec); |
| 1337 | return found; |
| 1338 | } |
| 1339 | |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1340 | STATIC loff_t |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1341 | xfs_seek_hole_data( |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1342 | struct file *file, |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1343 | loff_t start, |
| 1344 | int whence) |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1345 | { |
| 1346 | struct inode *inode = file->f_mapping->host; |
| 1347 | struct xfs_inode *ip = XFS_I(inode); |
| 1348 | struct xfs_mount *mp = ip->i_mount; |
| 1349 | loff_t uninitialized_var(offset); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1350 | xfs_fsize_t isize; |
| 1351 | xfs_fileoff_t fsbno; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1352 | xfs_filblks_t end; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1353 | uint lock; |
| 1354 | int error; |
| 1355 | |
| 1356 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 1357 | return -EIO; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1358 | |
Christoph Hellwig | 309ecac | 2013-12-06 12:30:09 -0800 | [diff] [blame] | 1359 | lock = xfs_ilock_data_map_shared(ip); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1360 | |
| 1361 | isize = i_size_read(inode); |
| 1362 | if (start >= isize) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1363 | error = -ENXIO; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1364 | goto out_unlock; |
| 1365 | } |
| 1366 | |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1367 | /* |
| 1368 | * Try to read extents from the first block indicated |
| 1369 | * by fsbno to the end block of the file. |
| 1370 | */ |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1371 | fsbno = XFS_B_TO_FSBT(mp, start); |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1372 | end = XFS_B_TO_FSB(mp, isize); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1373 | |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1374 | for (;;) { |
| 1375 | struct xfs_bmbt_irec map[2]; |
| 1376 | int nmap = 2; |
| 1377 | unsigned int i; |
| 1378 | |
| 1379 | error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap, |
| 1380 | XFS_BMAPI_ENTIRE); |
| 1381 | if (error) |
| 1382 | goto out_unlock; |
| 1383 | |
| 1384 | /* No extents at given offset, must be beyond EOF */ |
| 1385 | if (nmap == 0) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1386 | error = -ENXIO; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1387 | goto out_unlock; |
| 1388 | } |
| 1389 | |
| 1390 | for (i = 0; i < nmap; i++) { |
| 1391 | offset = max_t(loff_t, start, |
| 1392 | XFS_FSB_TO_B(mp, map[i].br_startoff)); |
| 1393 | |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1394 | /* Landed in the hole we wanted? */ |
| 1395 | if (whence == SEEK_HOLE && |
| 1396 | map[i].br_startblock == HOLESTARTBLOCK) |
| 1397 | goto out; |
| 1398 | |
| 1399 | /* Landed in the data extent we wanted? */ |
| 1400 | if (whence == SEEK_DATA && |
| 1401 | (map[i].br_startblock == DELAYSTARTBLOCK || |
| 1402 | (map[i].br_state == XFS_EXT_NORM && |
| 1403 | !isnullstartblock(map[i].br_startblock)))) |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1404 | goto out; |
| 1405 | |
| 1406 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1407 | * Landed in an unwritten extent, try to search |
| 1408 | * for hole or data from page cache. |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1409 | */ |
| 1410 | if (map[i].br_state == XFS_EXT_UNWRITTEN) { |
| 1411 | if (xfs_find_get_desired_pgoff(inode, &map[i], |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1412 | whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF, |
| 1413 | &offset)) |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1414 | goto out; |
| 1415 | } |
| 1416 | } |
| 1417 | |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1418 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1419 | * We only received one extent out of the two requested. This |
| 1420 | * means we've hit EOF and didn't find what we are looking for. |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1421 | */ |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1422 | if (nmap == 1) { |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1423 | /* |
| 1424 | * If we were looking for a hole, set offset to |
| 1425 | * the end of the file (i.e., there is an implicit |
| 1426 | * hole at the end of any file). |
| 1427 | */ |
| 1428 | if (whence == SEEK_HOLE) { |
| 1429 | offset = isize; |
| 1430 | break; |
| 1431 | } |
| 1432 | /* |
| 1433 | * If we were looking for data, it's nowhere to be found |
| 1434 | */ |
| 1435 | ASSERT(whence == SEEK_DATA); |
| 1436 | error = -ENXIO; |
| 1437 | goto out_unlock; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1438 | } |
| 1439 | |
| 1440 | ASSERT(i > 1); |
| 1441 | |
| 1442 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1443 | * Nothing was found, proceed to the next round of search |
| 1444 | * if the next reading offset is not at or beyond EOF. |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1445 | */ |
| 1446 | fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount; |
| 1447 | start = XFS_FSB_TO_B(mp, fsbno); |
| 1448 | if (start >= isize) { |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1449 | if (whence == SEEK_HOLE) { |
| 1450 | offset = isize; |
| 1451 | break; |
| 1452 | } |
| 1453 | ASSERT(whence == SEEK_DATA); |
| 1454 | error = -ENXIO; |
| 1455 | goto out_unlock; |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1456 | } |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1457 | } |
| 1458 | |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1459 | out: |
| 1460 | /* |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1461 | * 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] | 1462 | * 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] | 1463 | * page boundary for unwritten extents. We need to deal with this |
Jeff Liu | b686d1f | 2012-08-21 17:12:18 +0800 | [diff] [blame] | 1464 | * situation in particular. |
| 1465 | */ |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1466 | if (whence == SEEK_HOLE) |
| 1467 | offset = min_t(loff_t, offset, isize); |
Jie Liu | 46a1c2c | 2013-06-25 12:02:13 +0800 | [diff] [blame] | 1468 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1469 | |
| 1470 | out_unlock: |
Christoph Hellwig | 01f4f32 | 2013-12-06 12:30:08 -0800 | [diff] [blame] | 1471 | xfs_iunlock(ip, lock); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1472 | |
| 1473 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1474 | return error; |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1475 | return offset; |
| 1476 | } |
| 1477 | |
| 1478 | STATIC loff_t |
| 1479 | xfs_file_llseek( |
| 1480 | struct file *file, |
| 1481 | loff_t offset, |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1482 | int whence) |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1483 | { |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1484 | switch (whence) { |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1485 | case SEEK_END: |
| 1486 | case SEEK_CUR: |
| 1487 | case SEEK_SET: |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1488 | return generic_file_llseek(file, offset, whence); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1489 | case SEEK_HOLE: |
Eric Sandeen | 49c6959 | 2014-09-09 11:56:48 +1000 | [diff] [blame] | 1490 | case SEEK_DATA: |
Eric Sandeen | 59f9c00 | 2014-09-09 11:57:10 +1000 | [diff] [blame] | 1491 | return xfs_seek_hole_data(file, offset, whence); |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1492 | default: |
| 1493 | return -EINVAL; |
| 1494 | } |
| 1495 | } |
| 1496 | |
Dave Chinner | de0e8c2 | 2015-02-23 21:44:19 +1100 | [diff] [blame] | 1497 | /* |
| 1498 | * Locking for serialisation of IO during page faults. This results in a lock |
| 1499 | * ordering of: |
| 1500 | * |
| 1501 | * mmap_sem (MM) |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1502 | * sb_start_pagefault(vfs, freeze) |
Dave Chinner | 13ad4fe | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1503 | * i_mmaplock (XFS - truncate serialisation) |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1504 | * page_lock (MM) |
| 1505 | * i_lock (XFS - extent map serialisation) |
Dave Chinner | de0e8c2 | 2015-02-23 21:44:19 +1100 | [diff] [blame] | 1506 | */ |
Dave Chinner | de0e8c2 | 2015-02-23 21:44:19 +1100 | [diff] [blame] | 1507 | |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1508 | /* |
| 1509 | * mmap()d file has taken write protection fault and is being made writable. We |
| 1510 | * can set the page state up correctly for a writable page, which means we can |
| 1511 | * do correct delalloc accounting (ENOSPC checking!) and unwritten extent |
| 1512 | * mapping. |
| 1513 | */ |
| 1514 | STATIC int |
| 1515 | xfs_filemap_page_mkwrite( |
| 1516 | struct vm_area_struct *vma, |
| 1517 | struct vm_fault *vmf) |
| 1518 | { |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1519 | struct inode *inode = file_inode(vma->vm_file); |
Dave Chinner | ec56b1f | 2015-06-04 09:18:18 +1000 | [diff] [blame] | 1520 | int ret; |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1521 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1522 | trace_xfs_filemap_page_mkwrite(XFS_I(inode)); |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1523 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1524 | sb_start_pagefault(inode->i_sb); |
Dave Chinner | ec56b1f | 2015-06-04 09:18:18 +1000 | [diff] [blame] | 1525 | file_update_time(vma->vm_file); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1526 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
| 1527 | |
| 1528 | if (IS_DAX(inode)) { |
Dave Chinner | 01a155e | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1529 | ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_dax_fault, NULL); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1530 | } else { |
Ross Zwisler | 5c50002 | 2015-10-13 16:51:02 -0600 | [diff] [blame] | 1531 | ret = block_page_mkwrite(vma, vmf, xfs_get_blocks); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1532 | ret = block_page_mkwrite_return(ret); |
| 1533 | } |
| 1534 | |
| 1535 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
| 1536 | sb_end_pagefault(inode->i_sb); |
| 1537 | |
| 1538 | return ret; |
| 1539 | } |
| 1540 | |
| 1541 | STATIC int |
| 1542 | xfs_filemap_fault( |
| 1543 | struct vm_area_struct *vma, |
| 1544 | struct vm_fault *vmf) |
| 1545 | { |
Dave Chinner | b2442c5 | 2015-07-29 11:48:00 +1000 | [diff] [blame] | 1546 | struct inode *inode = file_inode(vma->vm_file); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1547 | int ret; |
| 1548 | |
Dave Chinner | b2442c5 | 2015-07-29 11:48:00 +1000 | [diff] [blame] | 1549 | trace_xfs_filemap_fault(XFS_I(inode)); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1550 | |
| 1551 | /* DAX can shortcut the normal fault path on write faults! */ |
Dave Chinner | b2442c5 | 2015-07-29 11:48:00 +1000 | [diff] [blame] | 1552 | if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode)) |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1553 | return xfs_filemap_page_mkwrite(vma, vmf); |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1554 | |
Dave Chinner | b2442c5 | 2015-07-29 11:48:00 +1000 | [diff] [blame] | 1555 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
| 1556 | if (IS_DAX(inode)) { |
| 1557 | /* |
| 1558 | * we do not want to trigger unwritten extent conversion on read |
| 1559 | * faults - that is unnecessary overhead and would also require |
| 1560 | * changes to xfs_get_blocks_direct() to map unwritten extent |
| 1561 | * ioend for conversion on read-only mappings. |
| 1562 | */ |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1563 | ret = __dax_fault(vma, vmf, xfs_get_blocks_dax_fault, NULL); |
Dave Chinner | b2442c5 | 2015-07-29 11:48:00 +1000 | [diff] [blame] | 1564 | } else |
| 1565 | ret = filemap_fault(vma, vmf); |
| 1566 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1567 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1568 | return ret; |
| 1569 | } |
| 1570 | |
Dave Chinner | 13ad4fe | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1571 | /* |
| 1572 | * Similar to xfs_filemap_fault(), the DAX fault path can call into here on |
| 1573 | * both read and write faults. Hence we need to handle both cases. There is no |
| 1574 | * ->pmd_mkwrite callout for huge pages, so we have a single function here to |
| 1575 | * handle both cases here. @flags carries the information on the type of fault |
| 1576 | * occuring. |
| 1577 | */ |
Matthew Wilcox | acd76e7 | 2015-09-08 14:59:06 -0700 | [diff] [blame] | 1578 | STATIC int |
| 1579 | xfs_filemap_pmd_fault( |
| 1580 | struct vm_area_struct *vma, |
| 1581 | unsigned long addr, |
| 1582 | pmd_t *pmd, |
| 1583 | unsigned int flags) |
| 1584 | { |
| 1585 | struct inode *inode = file_inode(vma->vm_file); |
| 1586 | struct xfs_inode *ip = XFS_I(inode); |
| 1587 | int ret; |
| 1588 | |
| 1589 | if (!IS_DAX(inode)) |
| 1590 | return VM_FAULT_FALLBACK; |
| 1591 | |
| 1592 | trace_xfs_filemap_pmd_fault(ip); |
| 1593 | |
Dave Chinner | 13ad4fe | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1594 | if (flags & FAULT_FLAG_WRITE) { |
| 1595 | sb_start_pagefault(inode->i_sb); |
| 1596 | file_update_time(vma->vm_file); |
| 1597 | } |
| 1598 | |
Matthew Wilcox | acd76e7 | 2015-09-08 14:59:06 -0700 | [diff] [blame] | 1599 | xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1600 | ret = __dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault, |
Dave Chinner | 01a155e | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1601 | NULL); |
Matthew Wilcox | acd76e7 | 2015-09-08 14:59:06 -0700 | [diff] [blame] | 1602 | xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); |
Dave Chinner | 13ad4fe | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1603 | |
| 1604 | if (flags & FAULT_FLAG_WRITE) |
| 1605 | sb_end_pagefault(inode->i_sb); |
Matthew Wilcox | acd76e7 | 2015-09-08 14:59:06 -0700 | [diff] [blame] | 1606 | |
| 1607 | return ret; |
| 1608 | } |
| 1609 | |
Dave Chinner | 3af4928 | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1610 | /* |
| 1611 | * pfn_mkwrite was originally inteneded to ensure we capture time stamp |
| 1612 | * updates on write faults. In reality, it's need to serialise against |
Ross Zwisler | 5eb88dc | 2016-01-22 15:10:56 -0800 | [diff] [blame] | 1613 | * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED |
| 1614 | * to ensure we serialise the fault barrier in place. |
Dave Chinner | 3af4928 | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1615 | */ |
| 1616 | static int |
| 1617 | xfs_filemap_pfn_mkwrite( |
| 1618 | struct vm_area_struct *vma, |
| 1619 | struct vm_fault *vmf) |
| 1620 | { |
| 1621 | |
| 1622 | struct inode *inode = file_inode(vma->vm_file); |
| 1623 | struct xfs_inode *ip = XFS_I(inode); |
| 1624 | int ret = VM_FAULT_NOPAGE; |
| 1625 | loff_t size; |
| 1626 | |
| 1627 | trace_xfs_filemap_pfn_mkwrite(ip); |
| 1628 | |
| 1629 | sb_start_pagefault(inode->i_sb); |
| 1630 | file_update_time(vma->vm_file); |
| 1631 | |
| 1632 | /* check if the faulting page hasn't raced with truncate */ |
| 1633 | xfs_ilock(ip, XFS_MMAPLOCK_SHARED); |
| 1634 | size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 1635 | if (vmf->pgoff >= size) |
| 1636 | ret = VM_FAULT_SIGBUS; |
Ross Zwisler | 5eb88dc | 2016-01-22 15:10:56 -0800 | [diff] [blame] | 1637 | else if (IS_DAX(inode)) |
| 1638 | ret = dax_pfn_mkwrite(vma, vmf); |
Dave Chinner | 3af4928 | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1639 | xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); |
| 1640 | sb_end_pagefault(inode->i_sb); |
| 1641 | return ret; |
| 1642 | |
| 1643 | } |
| 1644 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1645 | static const struct vm_operations_struct xfs_file_vm_ops = { |
| 1646 | .fault = xfs_filemap_fault, |
Matthew Wilcox | acd76e7 | 2015-09-08 14:59:06 -0700 | [diff] [blame] | 1647 | .pmd_fault = xfs_filemap_pmd_fault, |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1648 | .map_pages = filemap_map_pages, |
| 1649 | .page_mkwrite = xfs_filemap_page_mkwrite, |
Dave Chinner | 3af4928 | 2015-11-03 12:37:02 +1100 | [diff] [blame] | 1650 | .pfn_mkwrite = xfs_filemap_pfn_mkwrite, |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1651 | }; |
| 1652 | |
| 1653 | STATIC int |
| 1654 | xfs_file_mmap( |
| 1655 | struct file *filp, |
| 1656 | struct vm_area_struct *vma) |
| 1657 | { |
| 1658 | file_accessed(filp); |
| 1659 | vma->vm_ops = &xfs_file_vm_ops; |
| 1660 | if (IS_DAX(file_inode(filp))) |
Matthew Wilcox | acd76e7 | 2015-09-08 14:59:06 -0700 | [diff] [blame] | 1661 | vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE; |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1662 | return 0; |
Dave Chinner | 075a924 | 2015-02-23 21:44:54 +1100 | [diff] [blame] | 1663 | } |
| 1664 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 1665 | const struct file_operations xfs_file_operations = { |
Jeff Liu | 3fe3e6b | 2012-05-10 21:29:17 +0800 | [diff] [blame] | 1666 | .llseek = xfs_file_llseek, |
Al Viro | b4f5d2c | 2014-04-02 14:37:59 -0400 | [diff] [blame] | 1667 | .read_iter = xfs_file_read_iter, |
Al Viro | bf97f3bc | 2014-04-03 14:20:23 -0400 | [diff] [blame] | 1668 | .write_iter = xfs_file_write_iter, |
Nathan Scott | 1b89584 | 2006-03-31 13:08:59 +1000 | [diff] [blame] | 1669 | .splice_read = xfs_file_splice_read, |
Al Viro | 8d02076 | 2014-04-05 04:27:08 -0400 | [diff] [blame] | 1670 | .splice_write = iter_file_splice_write, |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1671 | .unlocked_ioctl = xfs_file_ioctl, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1672 | #ifdef CONFIG_COMPAT |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1673 | .compat_ioctl = xfs_file_compat_ioctl, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | #endif |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1675 | .mmap = xfs_file_mmap, |
| 1676 | .open = xfs_file_open, |
| 1677 | .release = xfs_file_release, |
| 1678 | .fsync = xfs_file_fsync, |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 1679 | .fallocate = xfs_file_fallocate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | }; |
| 1681 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 1682 | const struct file_operations xfs_dir_file_operations = { |
Christoph Hellwig | f999a5b | 2008-11-28 14:23:32 +1100 | [diff] [blame] | 1683 | .open = xfs_dir_open, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | .read = generic_read_dir, |
Al Viro | b822755 | 2013-05-22 17:07:56 -0400 | [diff] [blame] | 1685 | .iterate = xfs_file_readdir, |
Al Viro | 59af158 | 2008-08-24 07:24:41 -0400 | [diff] [blame] | 1686 | .llseek = generic_file_llseek, |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1687 | .unlocked_ioctl = xfs_file_ioctl, |
Nathan Scott | d387039 | 2005-05-06 06:44:46 -0700 | [diff] [blame] | 1688 | #ifdef CONFIG_COMPAT |
Nathan Scott | 3562fd4 | 2006-03-14 14:00:35 +1100 | [diff] [blame] | 1689 | .compat_ioctl = xfs_file_compat_ioctl, |
Nathan Scott | d387039 | 2005-05-06 06:44:46 -0700 | [diff] [blame] | 1690 | #endif |
Christoph Hellwig | 1da2f2d | 2011-10-02 14:25:16 +0000 | [diff] [blame] | 1691 | .fsync = xfs_dir_fsync, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1692 | }; |