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" |
Dave Chinner | 70a9883 | 2013-10-23 10:36:05 +1100 | [diff] [blame] | 19 | #include "xfs_shared.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 20 | #include "xfs_format.h" |
| 21 | #include "xfs_log_format.h" |
| 22 | #include "xfs_trans_resv.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include "xfs_sb.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 24 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include "xfs_mount.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include "xfs_inode.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 27 | #include "xfs_trans.h" |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 28 | #include "xfs_inode_item.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 29 | #include "xfs_alloc.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include "xfs_error.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include "xfs_iomap.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 32 | #include "xfs_trace.h" |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 33 | #include "xfs_bmap.h" |
Dave Chinner | 6898811 | 2013-08-12 20:49:42 +1000 | [diff] [blame] | 34 | #include "xfs_bmap_util.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 35 | #include "xfs_bmap_btree.h" |
| 36 | #include "xfs_dinode.h" |
Kent Overstreet | a27bb33 | 2013-05-07 16:19:08 -0700 | [diff] [blame] | 37 | #include <linux/aio.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 38 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <linux/mpage.h> |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 40 | #include <linux/pagevec.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | #include <linux/writeback.h> |
| 42 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 43 | void |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 44 | xfs_count_page_state( |
| 45 | struct page *page, |
| 46 | int *delalloc, |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 47 | int *unwritten) |
| 48 | { |
| 49 | struct buffer_head *bh, *head; |
| 50 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 51 | *delalloc = *unwritten = 0; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 52 | |
| 53 | bh = head = page_buffers(page); |
| 54 | do { |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 55 | if (buffer_unwritten(bh)) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 56 | (*unwritten) = 1; |
| 57 | else if (buffer_delay(bh)) |
| 58 | (*delalloc) = 1; |
| 59 | } while ((bh = bh->b_this_page) != head); |
| 60 | } |
| 61 | |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 62 | STATIC struct block_device * |
| 63 | xfs_find_bdev_for_inode( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 64 | struct inode *inode) |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 65 | { |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 66 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 67 | struct xfs_mount *mp = ip->i_mount; |
| 68 | |
Eric Sandeen | 71ddabb | 2007-11-23 16:29:42 +1100 | [diff] [blame] | 69 | if (XFS_IS_REALTIME_INODE(ip)) |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 70 | return mp->m_rtdev_targp->bt_bdev; |
| 71 | else |
| 72 | return mp->m_ddev_targp->bt_bdev; |
| 73 | } |
| 74 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 75 | /* |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 76 | * We're now finished for good with this ioend structure. |
| 77 | * Update the page state via the associated buffer_heads, |
| 78 | * release holds on the inode and bio, and finally free |
| 79 | * up memory. Do not use the ioend after this. |
| 80 | */ |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 81 | STATIC void |
| 82 | xfs_destroy_ioend( |
| 83 | xfs_ioend_t *ioend) |
| 84 | { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 85 | struct buffer_head *bh, *next; |
| 86 | |
| 87 | for (bh = ioend->io_buffer_head; bh; bh = next) { |
| 88 | next = bh->b_private; |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 89 | bh->b_end_io(bh, !ioend->io_error); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 90 | } |
Christoph Hellwig | 583fa58 | 2008-12-03 12:20:38 +0100 | [diff] [blame] | 91 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 92 | mempool_free(ioend, xfs_ioend_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | } |
| 94 | |
| 95 | /* |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 96 | * Fast and loose check if this write could update the on-disk inode size. |
| 97 | */ |
| 98 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) |
| 99 | { |
| 100 | return ioend->io_offset + ioend->io_size > |
| 101 | XFS_I(ioend->io_inode)->i_d.di_size; |
| 102 | } |
| 103 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 104 | STATIC int |
| 105 | xfs_setfilesize_trans_alloc( |
| 106 | struct xfs_ioend *ioend) |
| 107 | { |
| 108 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
| 109 | struct xfs_trans *tp; |
| 110 | int error; |
| 111 | |
| 112 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); |
| 113 | |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 114 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 115 | if (error) { |
| 116 | xfs_trans_cancel(tp, 0); |
| 117 | return error; |
| 118 | } |
| 119 | |
| 120 | ioend->io_append_trans = tp; |
| 121 | |
| 122 | /* |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 123 | * We may pass freeze protection with a transaction. So tell lockdep |
Jan Kara | d9457dc | 2012-06-12 16:20:39 +0200 | [diff] [blame] | 124 | * we released it. |
| 125 | */ |
| 126 | rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], |
| 127 | 1, _THIS_IP_); |
| 128 | /* |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 129 | * We hand off the transaction to the completion thread now, so |
| 130 | * clear the flag here. |
| 131 | */ |
| 132 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
| 133 | return 0; |
| 134 | } |
| 135 | |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 136 | /* |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 137 | * Update on-disk file size now that data has been written to disk. |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 138 | */ |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 139 | STATIC int |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 140 | xfs_setfilesize( |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 141 | struct xfs_ioend *ioend) |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 142 | { |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 143 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 144 | struct xfs_trans *tp = ioend->io_append_trans; |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 145 | xfs_fsize_t isize; |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 146 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 147 | /* |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 148 | * The transaction may have been allocated in the I/O submission thread, |
| 149 | * thus we need to mark ourselves as beeing in a transaction manually. |
| 150 | * Similarly for freeze protection. |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 151 | */ |
| 152 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 153 | rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], |
| 154 | 0, 1, _THIS_IP_); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 155 | |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 156 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | 6923e68 | 2012-02-29 09:53:49 +0000 | [diff] [blame] | 157 | isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 158 | if (!isize) { |
| 159 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 160 | xfs_trans_cancel(tp, 0); |
| 161 | return 0; |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 162 | } |
| 163 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 164 | trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); |
| 165 | |
| 166 | ip->i_d.di_size = isize; |
| 167 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
| 168 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 169 | |
| 170 | return xfs_trans_commit(tp, 0); |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 171 | } |
| 172 | |
| 173 | /* |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 174 | * Schedule IO completion handling on the final put of an ioend. |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 175 | * |
| 176 | * If there is no work to do we might as well call it a day and free the |
| 177 | * ioend right now. |
Dave Chinner | c626d17 | 2009-04-06 18:42:11 +0200 | [diff] [blame] | 178 | */ |
| 179 | STATIC void |
| 180 | xfs_finish_ioend( |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 181 | struct xfs_ioend *ioend) |
Dave Chinner | c626d17 | 2009-04-06 18:42:11 +0200 | [diff] [blame] | 182 | { |
| 183 | if (atomic_dec_and_test(&ioend->io_remaining)) { |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 184 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
| 185 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 186 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 187 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 188 | else if (ioend->io_append_trans || |
| 189 | (ioend->io_isdirect && xfs_ioend_is_append(ioend))) |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 190 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 191 | else |
| 192 | xfs_destroy_ioend(ioend); |
Dave Chinner | c626d17 | 2009-04-06 18:42:11 +0200 | [diff] [blame] | 193 | } |
| 194 | } |
| 195 | |
| 196 | /* |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 197 | * IO write completion. |
| 198 | */ |
| 199 | STATIC void |
| 200 | xfs_end_io( |
| 201 | struct work_struct *work) |
| 202 | { |
| 203 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
| 204 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
Dave Chinner | 6941893 | 2010-03-04 00:57:09 +0000 | [diff] [blame] | 205 | int error = 0; |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 206 | |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 207 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Christoph Hellwig | 810627d | 2011-11-08 08:56:15 +0000 | [diff] [blame] | 208 | ioend->io_error = -EIO; |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 209 | goto done; |
| 210 | } |
| 211 | if (ioend->io_error) |
| 212 | goto done; |
| 213 | |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 214 | /* |
| 215 | * For unwritten extents we need to issue transactions to convert a |
| 216 | * range to normal written extens after the data I/O has finished. |
| 217 | */ |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 218 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 219 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 220 | ioend->io_size); |
| 221 | } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) { |
| 222 | /* |
| 223 | * For direct I/O we do not know if we need to allocate blocks |
| 224 | * or not so we can't preallocate an append transaction as that |
| 225 | * results in nested reservations and log space deadlocks. Hence |
| 226 | * allocate the transaction here. While this is sub-optimal and |
| 227 | * can block IO completion for some time, we're stuck with doing |
| 228 | * it this way until we can pass the ioend to the direct IO |
| 229 | * allocation callbacks and avoid nesting that way. |
| 230 | */ |
| 231 | error = xfs_setfilesize_trans_alloc(ioend); |
| 232 | if (error) |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 233 | goto done; |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 234 | error = xfs_setfilesize(ioend); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 235 | } else if (ioend->io_append_trans) { |
| 236 | error = xfs_setfilesize(ioend); |
Christoph Hellwig | 84803fb | 2012-02-29 09:53:50 +0000 | [diff] [blame] | 237 | } else { |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 238 | ASSERT(!xfs_ioend_is_append(ioend)); |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 239 | } |
| 240 | |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 241 | done: |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 242 | if (error) |
| 243 | ioend->io_error = -error; |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 244 | xfs_destroy_ioend(ioend); |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 245 | } |
| 246 | |
| 247 | /* |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 248 | * Call IO completion handling in caller context on the final put of an ioend. |
| 249 | */ |
| 250 | STATIC void |
| 251 | xfs_finish_ioend_sync( |
| 252 | struct xfs_ioend *ioend) |
| 253 | { |
| 254 | if (atomic_dec_and_test(&ioend->io_remaining)) |
| 255 | xfs_end_io(&ioend->io_work); |
| 256 | } |
| 257 | |
| 258 | /* |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 259 | * Allocate and initialise an IO completion structure. |
| 260 | * We need to track unwritten extent write completion here initially. |
| 261 | * We'll need to extend this for updating the ondisk inode size later |
| 262 | * (vs. incore size). |
| 263 | */ |
| 264 | STATIC xfs_ioend_t * |
| 265 | xfs_alloc_ioend( |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 266 | struct inode *inode, |
| 267 | unsigned int type) |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 268 | { |
| 269 | xfs_ioend_t *ioend; |
| 270 | |
| 271 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); |
| 272 | |
| 273 | /* |
| 274 | * Set the count to 1 initially, which will prevent an I/O |
| 275 | * completion callback from happening before we have started |
| 276 | * all the I/O from calling the completion routine too early. |
| 277 | */ |
| 278 | atomic_set(&ioend->io_remaining, 1); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 279 | ioend->io_isdirect = 0; |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 280 | ioend->io_error = 0; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 281 | ioend->io_list = NULL; |
| 282 | ioend->io_type = type; |
Christoph Hellwig | b677c21 | 2007-08-29 11:46:28 +1000 | [diff] [blame] | 283 | ioend->io_inode = inode; |
Christoph Hellwig | c1a073b | 2005-09-05 08:23:35 +1000 | [diff] [blame] | 284 | ioend->io_buffer_head = NULL; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 285 | ioend->io_buffer_tail = NULL; |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 286 | ioend->io_offset = 0; |
| 287 | ioend->io_size = 0; |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 288 | ioend->io_append_trans = NULL; |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 289 | |
Christoph Hellwig | 5ec4fab | 2009-10-30 09:11:47 +0000 | [diff] [blame] | 290 | INIT_WORK(&ioend->io_work, xfs_end_io); |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 291 | return ioend; |
| 292 | } |
| 293 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | STATIC int |
| 295 | xfs_map_blocks( |
| 296 | struct inode *inode, |
| 297 | loff_t offset, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 298 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 299 | int type, |
| 300 | int nonblocking) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 302 | struct xfs_inode *ip = XFS_I(inode); |
| 303 | struct xfs_mount *mp = ip->i_mount; |
Christoph Hellwig | ed1e7b7 | 2010-12-10 08:42:22 +0000 | [diff] [blame] | 304 | ssize_t count = 1 << inode->i_blkbits; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 305 | xfs_fileoff_t offset_fsb, end_fsb; |
| 306 | int error = 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 307 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
| 308 | int nimaps = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 310 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 311 | return -XFS_ERROR(EIO); |
| 312 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 313 | if (type == XFS_IO_UNWRITTEN) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 314 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 315 | |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 316 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { |
| 317 | if (nonblocking) |
| 318 | return -XFS_ERROR(EAGAIN); |
| 319 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 320 | } |
| 321 | |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 322 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
| 323 | (ip->i_df.if_flags & XFS_IFEXTENTS)); |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 324 | ASSERT(offset <= mp->m_super->s_maxbytes); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 325 | |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 326 | if (offset + count > mp->m_super->s_maxbytes) |
| 327 | count = mp->m_super->s_maxbytes - offset; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 328 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
| 329 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 330 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
| 331 | imap, &nimaps, bmapi_flags); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 332 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
| 333 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 334 | if (error) |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 335 | return -XFS_ERROR(error); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 336 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 337 | if (type == XFS_IO_DELALLOC && |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 338 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
Jie Liu | 0799a3e | 2013-09-29 18:56:04 +0800 | [diff] [blame] | 339 | error = xfs_iomap_write_allocate(ip, offset, imap); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 340 | if (!error) |
| 341 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 342 | return -XFS_ERROR(error); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 343 | } |
| 344 | |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 345 | #ifdef DEBUG |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 346 | if (type == XFS_IO_UNWRITTEN) { |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 347 | ASSERT(nimaps); |
| 348 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 349 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
| 350 | } |
| 351 | #endif |
| 352 | if (nimaps) |
| 353 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); |
| 354 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | } |
| 356 | |
Christoph Hellwig | b8f82a4 | 2009-11-14 16:17:22 +0000 | [diff] [blame] | 357 | STATIC int |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 358 | xfs_imap_valid( |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 359 | struct inode *inode, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 360 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 361 | xfs_off_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | { |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 363 | offset >>= inode->i_blkbits; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 364 | |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 365 | return offset >= imap->br_startoff && |
| 366 | offset < imap->br_startoff + imap->br_blockcount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | } |
| 368 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 369 | /* |
| 370 | * BIO completion handler for buffered IO. |
| 371 | */ |
Al Viro | 782e3b3 | 2007-10-12 07:17:47 +0100 | [diff] [blame] | 372 | STATIC void |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 373 | xfs_end_bio( |
| 374 | struct bio *bio, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 375 | int error) |
| 376 | { |
| 377 | xfs_ioend_t *ioend = bio->bi_private; |
| 378 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 379 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 380 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 381 | |
| 382 | /* Toss bio and pass work off to an xfsdatad thread */ |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 383 | bio->bi_private = NULL; |
| 384 | bio->bi_end_io = NULL; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 385 | bio_put(bio); |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 386 | |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 387 | xfs_finish_ioend(ioend); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 388 | } |
| 389 | |
| 390 | STATIC void |
| 391 | xfs_submit_ioend_bio( |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 392 | struct writeback_control *wbc, |
| 393 | xfs_ioend_t *ioend, |
| 394 | struct bio *bio) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 395 | { |
| 396 | atomic_inc(&ioend->io_remaining); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 397 | bio->bi_private = ioend; |
| 398 | bio->bi_end_io = xfs_end_bio; |
Jens Axboe | 721a960 | 2011-03-09 11:56:30 +0100 | [diff] [blame] | 399 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 400 | } |
| 401 | |
| 402 | STATIC struct bio * |
| 403 | xfs_alloc_ioend_bio( |
| 404 | struct buffer_head *bh) |
| 405 | { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 406 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
Christoph Hellwig | 221cb25 | 2010-12-10 08:42:17 +0000 | [diff] [blame] | 407 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 408 | |
| 409 | ASSERT(bio->bi_private == NULL); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 410 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 411 | bio->bi_bdev = bh->b_bdev; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 412 | return bio; |
| 413 | } |
| 414 | |
| 415 | STATIC void |
| 416 | xfs_start_buffer_writeback( |
| 417 | struct buffer_head *bh) |
| 418 | { |
| 419 | ASSERT(buffer_mapped(bh)); |
| 420 | ASSERT(buffer_locked(bh)); |
| 421 | ASSERT(!buffer_delay(bh)); |
| 422 | ASSERT(!buffer_unwritten(bh)); |
| 423 | |
| 424 | mark_buffer_async_write(bh); |
| 425 | set_buffer_uptodate(bh); |
| 426 | clear_buffer_dirty(bh); |
| 427 | } |
| 428 | |
| 429 | STATIC void |
| 430 | xfs_start_page_writeback( |
| 431 | struct page *page, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 432 | int clear_dirty, |
| 433 | int buffers) |
| 434 | { |
| 435 | ASSERT(PageLocked(page)); |
| 436 | ASSERT(!PageWriteback(page)); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 437 | if (clear_dirty) |
David Chinner | 9213202 | 2006-12-21 10:24:01 +1100 | [diff] [blame] | 438 | clear_page_dirty_for_io(page); |
| 439 | set_page_writeback(page); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 440 | unlock_page(page); |
Fengguang Wu | 1f7decf | 2007-10-16 23:30:42 -0700 | [diff] [blame] | 441 | /* If no buffers on the page are to be written, finish it here */ |
| 442 | if (!buffers) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 443 | end_page_writeback(page); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 444 | } |
| 445 | |
Zhi Yong Wu | c7c1a7d | 2013-08-07 10:11:09 +0000 | [diff] [blame] | 446 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 447 | { |
| 448 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
| 449 | } |
| 450 | |
| 451 | /* |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 452 | * Submit all of the bios for all of the ioends we have saved up, covering the |
| 453 | * initial writepage page and also any probed pages. |
| 454 | * |
| 455 | * Because we may have multiple ioends spanning a page, we need to start |
| 456 | * writeback on all the buffers before we submit them for I/O. If we mark the |
| 457 | * buffers as we got, then we can end up with a page that only has buffers |
| 458 | * marked async write and I/O complete on can occur before we mark the other |
| 459 | * buffers async write. |
| 460 | * |
| 461 | * The end result of this is that we trip a bug in end_page_writeback() because |
| 462 | * we call it twice for the one page as the code in end_buffer_async_write() |
| 463 | * assumes that all buffers on the page are started at the same time. |
| 464 | * |
| 465 | * The fix is two passes across the ioend list - one to start writeback on the |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 466 | * buffer_heads, and then submit them for I/O on the second pass. |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 467 | * |
| 468 | * If @fail is non-zero, it means that we have a situation where some part of |
| 469 | * the submission process has failed after we have marked paged for writeback |
| 470 | * and unlocked them. In this situation, we need to fail the ioend chain rather |
| 471 | * than submit it to IO. This typically only happens on a filesystem shutdown. |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 472 | */ |
| 473 | STATIC void |
| 474 | xfs_submit_ioend( |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 475 | struct writeback_control *wbc, |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 476 | xfs_ioend_t *ioend, |
| 477 | int fail) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 478 | { |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 479 | xfs_ioend_t *head = ioend; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 480 | xfs_ioend_t *next; |
| 481 | struct buffer_head *bh; |
| 482 | struct bio *bio; |
| 483 | sector_t lastblock = 0; |
| 484 | |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 485 | /* Pass 1 - start writeback */ |
| 486 | do { |
| 487 | next = ioend->io_list; |
Christoph Hellwig | 221cb25 | 2010-12-10 08:42:17 +0000 | [diff] [blame] | 488 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 489 | xfs_start_buffer_writeback(bh); |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 490 | } while ((ioend = next) != NULL); |
| 491 | |
| 492 | /* Pass 2 - submit I/O */ |
| 493 | ioend = head; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 494 | do { |
| 495 | next = ioend->io_list; |
| 496 | bio = NULL; |
| 497 | |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 498 | /* |
| 499 | * If we are failing the IO now, just mark the ioend with an |
| 500 | * error and finish it. This will run IO completion immediately |
| 501 | * as there is only one reference to the ioend at this point in |
| 502 | * time. |
| 503 | */ |
| 504 | if (fail) { |
| 505 | ioend->io_error = -fail; |
| 506 | xfs_finish_ioend(ioend); |
| 507 | continue; |
| 508 | } |
| 509 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 510 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 511 | |
| 512 | if (!bio) { |
| 513 | retry: |
| 514 | bio = xfs_alloc_ioend_bio(bh); |
| 515 | } else if (bh->b_blocknr != lastblock + 1) { |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 516 | xfs_submit_ioend_bio(wbc, ioend, bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 517 | goto retry; |
| 518 | } |
| 519 | |
Zhi Yong Wu | c7c1a7d | 2013-08-07 10:11:09 +0000 | [diff] [blame] | 520 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 521 | xfs_submit_ioend_bio(wbc, ioend, bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 522 | goto retry; |
| 523 | } |
| 524 | |
| 525 | lastblock = bh->b_blocknr; |
| 526 | } |
| 527 | if (bio) |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 528 | xfs_submit_ioend_bio(wbc, ioend, bio); |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 529 | xfs_finish_ioend(ioend); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 530 | } while ((ioend = next) != NULL); |
| 531 | } |
| 532 | |
| 533 | /* |
| 534 | * Cancel submission of all buffer_heads so far in this endio. |
| 535 | * Toss the endio too. Only ever called for the initial page |
| 536 | * in a writepage request, so only ever one page. |
| 537 | */ |
| 538 | STATIC void |
| 539 | xfs_cancel_ioend( |
| 540 | xfs_ioend_t *ioend) |
| 541 | { |
| 542 | xfs_ioend_t *next; |
| 543 | struct buffer_head *bh, *next_bh; |
| 544 | |
| 545 | do { |
| 546 | next = ioend->io_list; |
| 547 | bh = ioend->io_buffer_head; |
| 548 | do { |
| 549 | next_bh = bh->b_private; |
| 550 | clear_buffer_async_write(bh); |
| 551 | unlock_buffer(bh); |
| 552 | } while ((bh = next_bh) != NULL); |
| 553 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 554 | mempool_free(ioend, xfs_ioend_pool); |
| 555 | } while ((ioend = next) != NULL); |
| 556 | } |
| 557 | |
| 558 | /* |
| 559 | * Test to see if we've been building up a completion structure for |
| 560 | * earlier buffers -- if so, we try to append to this ioend if we |
| 561 | * can, otherwise we finish off any current ioend and start another. |
| 562 | * Return true if we've finished the given ioend. |
| 563 | */ |
| 564 | STATIC void |
| 565 | xfs_add_to_ioend( |
| 566 | struct inode *inode, |
| 567 | struct buffer_head *bh, |
Christoph Hellwig | 7336cea | 2006-01-11 20:49:16 +1100 | [diff] [blame] | 568 | xfs_off_t offset, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 569 | unsigned int type, |
| 570 | xfs_ioend_t **result, |
| 571 | int need_ioend) |
| 572 | { |
| 573 | xfs_ioend_t *ioend = *result; |
| 574 | |
| 575 | if (!ioend || need_ioend || type != ioend->io_type) { |
| 576 | xfs_ioend_t *previous = *result; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 577 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 578 | ioend = xfs_alloc_ioend(inode, type); |
| 579 | ioend->io_offset = offset; |
| 580 | ioend->io_buffer_head = bh; |
| 581 | ioend->io_buffer_tail = bh; |
| 582 | if (previous) |
| 583 | previous->io_list = ioend; |
| 584 | *result = ioend; |
| 585 | } else { |
| 586 | ioend->io_buffer_tail->b_private = bh; |
| 587 | ioend->io_buffer_tail = bh; |
| 588 | } |
| 589 | |
| 590 | bh->b_private = NULL; |
| 591 | ioend->io_size += bh->b_size; |
| 592 | } |
| 593 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 594 | STATIC void |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 595 | xfs_map_buffer( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 596 | struct inode *inode, |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 597 | struct buffer_head *bh, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 598 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 599 | xfs_off_t offset) |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 600 | { |
| 601 | sector_t bn; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 602 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 603 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
| 604 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 605 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 606 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 607 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 608 | |
Christoph Hellwig | e513182 | 2010-04-28 12:28:55 +0000 | [diff] [blame] | 609 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 610 | ((offset - iomap_offset) >> inode->i_blkbits); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 611 | |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 612 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 613 | |
| 614 | bh->b_blocknr = bn; |
| 615 | set_buffer_mapped(bh); |
| 616 | } |
| 617 | |
| 618 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | xfs_map_at_offset( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 620 | struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 621 | struct buffer_head *bh, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 622 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 623 | xfs_off_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | { |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 625 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 626 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 628 | xfs_map_buffer(inode, bh, imap, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | set_buffer_mapped(bh); |
| 630 | clear_buffer_delay(bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 631 | clear_buffer_unwritten(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | } |
| 633 | |
| 634 | /* |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 635 | * Test if a given page contains at least one buffer of a given @type. |
| 636 | * If @check_all_buffers is true, then we walk all the buffers in the page to |
| 637 | * try to find one of the type passed in. If it is not set, then the caller only |
| 638 | * needs to check the first buffer on the page for a match. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | */ |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 640 | STATIC bool |
Dave Chinner | 6ffc4db | 2012-04-23 15:58:43 +1000 | [diff] [blame] | 641 | xfs_check_page_type( |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 642 | struct page *page, |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 643 | unsigned int type, |
| 644 | bool check_all_buffers) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 645 | { |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 646 | struct buffer_head *bh; |
| 647 | struct buffer_head *head; |
| 648 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 649 | if (PageWriteback(page)) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 650 | return false; |
| 651 | if (!page->mapping) |
| 652 | return false; |
| 653 | if (!page_has_buffers(page)) |
| 654 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 656 | bh = head = page_buffers(page); |
| 657 | do { |
| 658 | if (buffer_unwritten(bh)) { |
| 659 | if (type == XFS_IO_UNWRITTEN) |
| 660 | return true; |
| 661 | } else if (buffer_delay(bh)) { |
Dan Carpenter | 805eeb8 | 2014-04-04 06:56:30 +1100 | [diff] [blame] | 662 | if (type == XFS_IO_DELALLOC) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 663 | return true; |
| 664 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { |
Dan Carpenter | 805eeb8 | 2014-04-04 06:56:30 +1100 | [diff] [blame] | 665 | if (type == XFS_IO_OVERWRITE) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 666 | return true; |
| 667 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 669 | /* If we are only checking the first buffer, we are done now. */ |
| 670 | if (!check_all_buffers) |
| 671 | break; |
| 672 | } while ((bh = bh->b_this_page) != head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 674 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | } |
| 676 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | /* |
| 678 | * Allocate & map buffers for page given the extent map. Write it out. |
| 679 | * except for the original page of a writepage, this is called on |
| 680 | * delalloc/unwritten pages only, for the original page it is possible |
| 681 | * that the page has no mapping at all. |
| 682 | */ |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 683 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 684 | xfs_convert_page( |
| 685 | struct inode *inode, |
| 686 | struct page *page, |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 687 | loff_t tindex, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 688 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 689 | xfs_ioend_t **ioendp, |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 690 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 692 | struct buffer_head *bh, *head; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 693 | xfs_off_t end_offset; |
| 694 | unsigned long p_offset; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 695 | unsigned int type; |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 696 | int len, page_dirty; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 697 | int count = 0, done = 0, uptodate = 1; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 698 | xfs_off_t offset = page_offset(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 700 | if (page->index != tindex) |
| 701 | goto fail; |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 702 | if (!trylock_page(page)) |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 703 | goto fail; |
| 704 | if (PageWriteback(page)) |
| 705 | goto fail_unlock_page; |
| 706 | if (page->mapping != inode->i_mapping) |
| 707 | goto fail_unlock_page; |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 708 | if (!xfs_check_page_type(page, (*ioendp)->io_type, false)) |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 709 | goto fail_unlock_page; |
| 710 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 711 | /* |
| 712 | * page_dirty is initially a count of buffers on the page before |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 713 | * EOF and is decremented as we move each into a cleanable state. |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 714 | * |
| 715 | * Derivation: |
| 716 | * |
| 717 | * End offset is the highest offset that this page should represent. |
| 718 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) |
| 719 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and |
| 720 | * hence give us the correct page_dirty count. On any other page, |
| 721 | * it will be zero and in that case we need page_dirty to be the |
| 722 | * count of buffers on the page. |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 723 | */ |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 724 | end_offset = min_t(unsigned long long, |
| 725 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
| 726 | i_size_read(inode)); |
| 727 | |
Dave Chinner | 480d746 | 2013-05-20 09:51:08 +1000 | [diff] [blame] | 728 | /* |
| 729 | * If the current map does not span the entire page we are about to try |
| 730 | * to write, then give up. The only way we can write a page that spans |
| 731 | * multiple mappings in a single writeback iteration is via the |
| 732 | * xfs_vm_writepage() function. Data integrity writeback requires the |
| 733 | * entire page to be written in a single attempt, otherwise the part of |
| 734 | * the page we don't write here doesn't get written as part of the data |
| 735 | * integrity sync. |
| 736 | * |
| 737 | * For normal writeback, we also don't attempt to write partial pages |
| 738 | * here as it simply means that write_cache_pages() will see it under |
| 739 | * writeback and ignore the page until some point in the future, at |
| 740 | * which time this will be the only page in the file that needs |
| 741 | * writeback. Hence for more optimal IO patterns, we should always |
| 742 | * avoid partial page writeback due to multiple mappings on a page here. |
| 743 | */ |
| 744 | if (!xfs_imap_valid(inode, imap, end_offset)) |
| 745 | goto fail_unlock_page; |
| 746 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 747 | len = 1 << inode->i_blkbits; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 748 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
| 749 | PAGE_CACHE_SIZE); |
| 750 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; |
| 751 | page_dirty = p_offset / len; |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 752 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 753 | /* |
| 754 | * The moment we find a buffer that doesn't match our current type |
| 755 | * specification or can't be written, abort the loop and start |
| 756 | * writeback. As per the above xfs_imap_valid() check, only |
| 757 | * xfs_vm_writepage() can handle partial page writeback fully - we are |
| 758 | * limited here to the buffers that are contiguous with the current |
| 759 | * ioend, and hence a buffer we can't write breaks that contiguity and |
| 760 | * we have to defer the rest of the IO to xfs_vm_writepage(). |
| 761 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | bh = head = page_buffers(page); |
| 763 | do { |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 764 | if (offset >= end_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | break; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 766 | if (!buffer_uptodate(bh)) |
| 767 | uptodate = 0; |
| 768 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { |
| 769 | done = 1; |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 770 | break; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 771 | } |
| 772 | |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 773 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
| 774 | buffer_mapped(bh)) { |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 775 | if (buffer_unwritten(bh)) |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 776 | type = XFS_IO_UNWRITTEN; |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 777 | else if (buffer_delay(bh)) |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 778 | type = XFS_IO_DELALLOC; |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 779 | else |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 780 | type = XFS_IO_OVERWRITE; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 781 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 782 | /* |
| 783 | * imap should always be valid because of the above |
| 784 | * partial page end_offset check on the imap. |
| 785 | */ |
| 786 | ASSERT(xfs_imap_valid(inode, imap, offset)); |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 787 | |
Christoph Hellwig | ecff71e | 2010-12-10 08:42:25 +0000 | [diff] [blame] | 788 | lock_buffer(bh); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 789 | if (type != XFS_IO_OVERWRITE) |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 790 | xfs_map_at_offset(inode, bh, imap, offset); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 791 | xfs_add_to_ioend(inode, bh, offset, type, |
| 792 | ioendp, done); |
| 793 | |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 794 | page_dirty--; |
| 795 | count++; |
| 796 | } else { |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 797 | done = 1; |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 798 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | } |
Christoph Hellwig | 7336cea | 2006-01-11 20:49:16 +1100 | [diff] [blame] | 800 | } while (offset += len, (bh = bh->b_this_page) != head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 802 | if (uptodate && bh == head) |
| 803 | SetPageUptodate(page); |
| 804 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 805 | if (count) { |
Dave Chinner | efceab1 | 2010-08-24 11:44:56 +1000 | [diff] [blame] | 806 | if (--wbc->nr_to_write <= 0 && |
| 807 | wbc->sync_mode == WB_SYNC_NONE) |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 808 | done = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | } |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 810 | xfs_start_page_writeback(page, !page_dirty, count); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 811 | |
| 812 | return done; |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 813 | fail_unlock_page: |
| 814 | unlock_page(page); |
| 815 | fail: |
| 816 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | } |
| 818 | |
| 819 | /* |
| 820 | * Convert & write out a cluster of pages in the same extent as defined |
| 821 | * by mp and following the start page. |
| 822 | */ |
| 823 | STATIC void |
| 824 | xfs_cluster_write( |
| 825 | struct inode *inode, |
| 826 | pgoff_t tindex, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 827 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 828 | xfs_ioend_t **ioendp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | struct writeback_control *wbc, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | pgoff_t tlast) |
| 831 | { |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 832 | struct pagevec pvec; |
| 833 | int done = 0, i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 835 | pagevec_init(&pvec, 0); |
| 836 | while (!done && tindex <= tlast) { |
| 837 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); |
| 838 | |
| 839 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | break; |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 841 | |
| 842 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 843 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 844 | imap, ioendp, wbc); |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 845 | if (done) |
| 846 | break; |
| 847 | } |
| 848 | |
| 849 | pagevec_release(&pvec); |
| 850 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | } |
| 852 | } |
| 853 | |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 854 | STATIC void |
| 855 | xfs_vm_invalidatepage( |
| 856 | struct page *page, |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 857 | unsigned int offset, |
| 858 | unsigned int length) |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 859 | { |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 860 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
| 861 | length); |
| 862 | block_invalidatepage(page, offset, length); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 863 | } |
| 864 | |
| 865 | /* |
| 866 | * If the page has delalloc buffers on it, we need to punch them out before we |
| 867 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the |
| 868 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read |
| 869 | * is done on that same region - the delalloc extent is returned when none is |
| 870 | * supposed to be there. |
| 871 | * |
| 872 | * We prevent this by truncating away the delalloc regions on the page before |
| 873 | * invalidating it. Because they are delalloc, we can do this without needing a |
| 874 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this |
| 875 | * truncation without a transaction as there is no space left for block |
| 876 | * reservation (typically why we see a ENOSPC in writeback). |
| 877 | * |
| 878 | * This is not a performance critical path, so for now just do the punching a |
| 879 | * buffer head at a time. |
| 880 | */ |
| 881 | STATIC void |
| 882 | xfs_aops_discard_page( |
| 883 | struct page *page) |
| 884 | { |
| 885 | struct inode *inode = page->mapping->host; |
| 886 | struct xfs_inode *ip = XFS_I(inode); |
| 887 | struct buffer_head *bh, *head; |
| 888 | loff_t offset = page_offset(page); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 889 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 890 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 891 | goto out_invalidate; |
| 892 | |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 893 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 894 | goto out_invalidate; |
| 895 | |
Dave Chinner | 4f10700 | 2011-03-07 10:00:35 +1100 | [diff] [blame] | 896 | xfs_alert(ip->i_mount, |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 897 | "page discard on page %p, inode 0x%llx, offset %llu.", |
| 898 | page, ip->i_ino, offset); |
| 899 | |
| 900 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 901 | bh = head = page_buffers(page); |
| 902 | do { |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 903 | int error; |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 904 | xfs_fileoff_t start_fsb; |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 905 | |
| 906 | if (!buffer_delay(bh)) |
| 907 | goto next_buffer; |
| 908 | |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 909 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
| 910 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 911 | if (error) { |
| 912 | /* something screwed, just bail */ |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 913 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Dave Chinner | 4f10700 | 2011-03-07 10:00:35 +1100 | [diff] [blame] | 914 | xfs_alert(ip->i_mount, |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 915 | "page discard unable to remove delalloc mapping."); |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 916 | } |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 917 | break; |
| 918 | } |
| 919 | next_buffer: |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 920 | offset += 1 << inode->i_blkbits; |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 921 | |
| 922 | } while ((bh = bh->b_this_page) != head); |
| 923 | |
| 924 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 925 | out_invalidate: |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 926 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 927 | return; |
| 928 | } |
| 929 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | /* |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 931 | * Write out a dirty page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | * |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 933 | * For delalloc space on the page we need to allocate space and flush it. |
| 934 | * For unwritten space on the page we need to start the conversion to |
| 935 | * regular allocated space. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 936 | * For any other dirty buffer heads on the page we should flush them. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 937 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 938 | STATIC int |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 939 | xfs_vm_writepage( |
| 940 | struct page *page, |
| 941 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 942 | { |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 943 | struct inode *inode = page->mapping->host; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 944 | struct buffer_head *bh, *head; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 945 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 946 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 947 | loff_t offset; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 948 | unsigned int type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 949 | __uint64_t end_offset; |
Christoph Hellwig | bd1556a | 2010-04-28 12:29:00 +0000 | [diff] [blame] | 950 | pgoff_t end_index, last_index; |
Christoph Hellwig | ed1e7b7 | 2010-12-10 08:42:22 +0000 | [diff] [blame] | 951 | ssize_t len; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 952 | int err, imap_valid = 0, uptodate = 1; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 953 | int count = 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 954 | int nonblocking = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 956 | trace_xfs_writepage(inode, page, 0, 0); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 957 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 958 | ASSERT(page_has_buffers(page)); |
| 959 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 960 | /* |
| 961 | * Refuse to write the page out if we are called from reclaim context. |
| 962 | * |
Christoph Hellwig | d4f7a5c | 2010-06-28 10:34:44 -0400 | [diff] [blame] | 963 | * This avoids stack overflows when called from deeply used stacks in |
| 964 | * random callers for direct reclaim or memcg reclaim. We explicitly |
| 965 | * allow reclaim from kswapd as the stack usage there is relatively low. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 966 | * |
Mel Gorman | 94054fa | 2011-10-31 17:07:45 -0700 | [diff] [blame] | 967 | * This should never happen except in the case of a VM regression so |
| 968 | * warn about it. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 969 | */ |
Mel Gorman | 94054fa | 2011-10-31 17:07:45 -0700 | [diff] [blame] | 970 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
| 971 | PF_MEMALLOC)) |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 972 | goto redirty; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 973 | |
| 974 | /* |
Christoph Hellwig | 680a647 | 2011-07-08 14:34:05 +0200 | [diff] [blame] | 975 | * Given that we do not allow direct reclaim to call us, we should |
| 976 | * never be called while in a filesystem transaction. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 977 | */ |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 978 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 979 | goto redirty; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 980 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 981 | /* Is this page beyond the end of the file? */ |
| 982 | offset = i_size_read(inode); |
| 983 | end_index = offset >> PAGE_CACHE_SHIFT; |
| 984 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 985 | |
| 986 | /* |
| 987 | * The page index is less than the end_index, adjust the end_offset |
| 988 | * to the highest offset that this page should represent. |
| 989 | * ----------------------------------------------------- |
| 990 | * | file mapping | <EOF> | |
| 991 | * ----------------------------------------------------- |
| 992 | * | Page ... | Page N-2 | Page N-1 | Page N | | |
| 993 | * ^--------------------------------^----------|-------- |
| 994 | * | desired writeback range | see else | |
| 995 | * ---------------------------------^------------------| |
| 996 | */ |
| 997 | if (page->index < end_index) |
| 998 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; |
| 999 | else { |
| 1000 | /* |
| 1001 | * Check whether the page to write out is beyond or straddles |
| 1002 | * i_size or not. |
| 1003 | * ------------------------------------------------------- |
| 1004 | * | file mapping | <EOF> | |
| 1005 | * ------------------------------------------------------- |
| 1006 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | |
| 1007 | * ^--------------------------------^-----------|--------- |
| 1008 | * | | Straddles | |
| 1009 | * ---------------------------------^-----------|--------| |
| 1010 | */ |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1011 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
| 1012 | |
| 1013 | /* |
Jan Kara | ff9a28f | 2013-03-14 14:30:54 +0100 | [diff] [blame] | 1014 | * Skip the page if it is fully outside i_size, e.g. due to a |
| 1015 | * truncate operation that is in progress. We must redirty the |
| 1016 | * page so that reclaim stops reclaiming it. Otherwise |
| 1017 | * xfs_vm_releasepage() is called on it and gets confused. |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1018 | * |
| 1019 | * Note that the end_index is unsigned long, it would overflow |
| 1020 | * if the given offset is greater than 16TB on 32-bit system |
| 1021 | * and if we do check the page is fully outside i_size or not |
| 1022 | * via "if (page->index >= end_index + 1)" as "end_index + 1" |
| 1023 | * will be evaluated to 0. Hence this page will be redirtied |
| 1024 | * and be written out repeatedly which would result in an |
| 1025 | * infinite loop, the user program that perform this operation |
| 1026 | * will hang. Instead, we can verify this situation by checking |
| 1027 | * if the page to write is totally beyond the i_size or if it's |
| 1028 | * offset is just equal to the EOF. |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1029 | */ |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1030 | if (page->index > end_index || |
| 1031 | (page->index == end_index && offset_into_page == 0)) |
Jan Kara | ff9a28f | 2013-03-14 14:30:54 +0100 | [diff] [blame] | 1032 | goto redirty; |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1033 | |
| 1034 | /* |
| 1035 | * The page straddles i_size. It must be zeroed out on each |
| 1036 | * and every writepage invocation because it may be mmapped. |
| 1037 | * "A file is mapped in multiples of the page size. For a file |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1038 | * that is not a multiple of the page size, the remaining |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1039 | * memory is zeroed when mapped, and writes to that region are |
| 1040 | * not written out to the file." |
| 1041 | */ |
| 1042 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1043 | |
| 1044 | /* Adjust the end_offset to the end of file */ |
| 1045 | end_offset = offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | } |
| 1047 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 1048 | len = 1 << inode->i_blkbits; |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 1049 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 1050 | bh = head = page_buffers(page); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1051 | offset = page_offset(page); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1052 | type = XFS_IO_OVERWRITE; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1053 | |
Christoph Hellwig | dbcdde3 | 2011-07-08 14:34:14 +0200 | [diff] [blame] | 1054 | if (wbc->sync_mode == WB_SYNC_NONE) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1055 | nonblocking = 1; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1056 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | do { |
Christoph Hellwig | 6ac7248 | 2010-12-10 08:42:18 +0000 | [diff] [blame] | 1058 | int new_ioend = 0; |
| 1059 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | if (offset >= end_offset) |
| 1061 | break; |
| 1062 | if (!buffer_uptodate(bh)) |
| 1063 | uptodate = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | |
Eric Sandeen | 3d9b02e | 2010-06-24 09:45:30 +1000 | [diff] [blame] | 1065 | /* |
Christoph Hellwig | ece413f | 2010-11-10 21:39:11 +0000 | [diff] [blame] | 1066 | * set_page_dirty dirties all buffers in a page, independent |
| 1067 | * of their state. The dirty state however is entirely |
| 1068 | * meaningless for holes (!mapped && uptodate), so skip |
| 1069 | * buffers covering holes here. |
Eric Sandeen | 3d9b02e | 2010-06-24 09:45:30 +1000 | [diff] [blame] | 1070 | */ |
| 1071 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { |
Eric Sandeen | 3d9b02e | 2010-06-24 09:45:30 +1000 | [diff] [blame] | 1072 | imap_valid = 0; |
| 1073 | continue; |
| 1074 | } |
| 1075 | |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1076 | if (buffer_unwritten(bh)) { |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1077 | if (type != XFS_IO_UNWRITTEN) { |
| 1078 | type = XFS_IO_UNWRITTEN; |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1079 | imap_valid = 0; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1080 | } |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1081 | } else if (buffer_delay(bh)) { |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1082 | if (type != XFS_IO_DELALLOC) { |
| 1083 | type = XFS_IO_DELALLOC; |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1084 | imap_valid = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | } |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1086 | } else if (buffer_uptodate(bh)) { |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1087 | if (type != XFS_IO_OVERWRITE) { |
| 1088 | type = XFS_IO_OVERWRITE; |
Christoph Hellwig | 85da94c | 2010-12-10 08:42:16 +0000 | [diff] [blame] | 1089 | imap_valid = 0; |
| 1090 | } |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1091 | } else { |
Alain Renaud | 7d0fa3e | 2012-06-08 15:34:46 -0400 | [diff] [blame] | 1092 | if (PageUptodate(page)) |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1093 | ASSERT(buffer_mapped(bh)); |
Alain Renaud | 7d0fa3e | 2012-06-08 15:34:46 -0400 | [diff] [blame] | 1094 | /* |
| 1095 | * This buffer is not uptodate and will not be |
| 1096 | * written to disk. Ensure that we will put any |
| 1097 | * subsequent writeable buffers into a new |
| 1098 | * ioend. |
| 1099 | */ |
| 1100 | imap_valid = 0; |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1101 | continue; |
| 1102 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1104 | if (imap_valid) |
| 1105 | imap_valid = xfs_imap_valid(inode, &imap, offset); |
| 1106 | if (!imap_valid) { |
| 1107 | /* |
| 1108 | * If we didn't have a valid mapping then we need to |
| 1109 | * put the new mapping into a separate ioend structure. |
| 1110 | * This ensures non-contiguous extents always have |
| 1111 | * separate ioends, which is particularly important |
| 1112 | * for unwritten extent conversion at I/O completion |
| 1113 | * time. |
| 1114 | */ |
| 1115 | new_ioend = 1; |
| 1116 | err = xfs_map_blocks(inode, offset, &imap, type, |
| 1117 | nonblocking); |
| 1118 | if (err) |
| 1119 | goto error; |
| 1120 | imap_valid = xfs_imap_valid(inode, &imap, offset); |
| 1121 | } |
| 1122 | if (imap_valid) { |
Christoph Hellwig | ecff71e | 2010-12-10 08:42:25 +0000 | [diff] [blame] | 1123 | lock_buffer(bh); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1124 | if (type != XFS_IO_OVERWRITE) |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1125 | xfs_map_at_offset(inode, bh, &imap, offset); |
| 1126 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, |
| 1127 | new_ioend); |
| 1128 | count++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | } |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1130 | |
| 1131 | if (!iohead) |
| 1132 | iohead = ioend; |
| 1133 | |
| 1134 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1135 | |
| 1136 | if (uptodate && bh == head) |
| 1137 | SetPageUptodate(page); |
| 1138 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1139 | xfs_start_page_writeback(page, 1, count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1140 | |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 1141 | /* if there is no IO to be submitted for this page, we are done */ |
| 1142 | if (!ioend) |
| 1143 | return 0; |
| 1144 | |
| 1145 | ASSERT(iohead); |
| 1146 | |
| 1147 | /* |
| 1148 | * Any errors from this point onwards need tobe reported through the IO |
| 1149 | * completion path as we have marked the initial page as under writeback |
| 1150 | * and unlocked it. |
| 1151 | */ |
| 1152 | if (imap_valid) { |
Christoph Hellwig | bd1556a | 2010-04-28 12:29:00 +0000 | [diff] [blame] | 1153 | xfs_off_t end_index; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 1154 | |
Christoph Hellwig | bd1556a | 2010-04-28 12:29:00 +0000 | [diff] [blame] | 1155 | end_index = imap.br_startoff + imap.br_blockcount; |
| 1156 | |
| 1157 | /* to bytes */ |
| 1158 | end_index <<= inode->i_blkbits; |
| 1159 | |
| 1160 | /* to pages */ |
| 1161 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; |
| 1162 | |
| 1163 | /* check against file size */ |
| 1164 | if (end_index > last_index) |
| 1165 | end_index = last_index; |
| 1166 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1167 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 1168 | wbc, end_index); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1169 | } |
| 1170 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1171 | |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 1172 | /* |
| 1173 | * Reserve log space if we might write beyond the on-disk inode size. |
| 1174 | */ |
| 1175 | err = 0; |
| 1176 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) |
| 1177 | err = xfs_setfilesize_trans_alloc(ioend); |
| 1178 | |
| 1179 | xfs_submit_ioend(wbc, iohead, err); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1180 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1181 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1182 | |
| 1183 | error: |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1184 | if (iohead) |
| 1185 | xfs_cancel_ioend(iohead); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1186 | |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 1187 | if (err == -EAGAIN) |
| 1188 | goto redirty; |
| 1189 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1190 | xfs_aops_discard_page(page); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1191 | ClearPageUptodate(page); |
| 1192 | unlock_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | return err; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1194 | |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 1195 | redirty: |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1196 | redirty_page_for_writepage(wbc, page); |
| 1197 | unlock_page(page); |
| 1198 | return 0; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1199 | } |
| 1200 | |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1201 | STATIC int |
| 1202 | xfs_vm_writepages( |
| 1203 | struct address_space *mapping, |
| 1204 | struct writeback_control *wbc) |
| 1205 | { |
Christoph Hellwig | b3aea4e | 2007-08-29 11:44:37 +1000 | [diff] [blame] | 1206 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1207 | return generic_writepages(mapping, wbc); |
| 1208 | } |
| 1209 | |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1210 | /* |
| 1211 | * Called to move a page into cleanable state - and from there |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1212 | * to be released. The page should already be clean. We always |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1213 | * have buffer heads in this call. |
| 1214 | * |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1215 | * Returns 1 if the page is ok to release, 0 otherwise. |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1216 | */ |
| 1217 | STATIC int |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1218 | xfs_vm_releasepage( |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1219 | struct page *page, |
| 1220 | gfp_t gfp_mask) |
| 1221 | { |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1222 | int delalloc, unwritten; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1223 | |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 1224 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1225 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1226 | xfs_count_page_state(page, &delalloc, &unwritten); |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1227 | |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 1228 | if (WARN_ON_ONCE(delalloc)) |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1229 | return 0; |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 1230 | if (WARN_ON_ONCE(unwritten)) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1231 | return 0; |
| 1232 | |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1233 | return try_to_free_buffers(page); |
| 1234 | } |
| 1235 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | STATIC int |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1237 | __xfs_get_blocks( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1238 | struct inode *inode, |
| 1239 | sector_t iblock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1240 | struct buffer_head *bh_result, |
| 1241 | int create, |
Christoph Hellwig | f2bde9b | 2010-06-24 11:44:35 +1000 | [diff] [blame] | 1242 | int direct) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1244 | struct xfs_inode *ip = XFS_I(inode); |
| 1245 | struct xfs_mount *mp = ip->i_mount; |
| 1246 | xfs_fileoff_t offset_fsb, end_fsb; |
| 1247 | int error = 0; |
| 1248 | int lockmode = 0; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1249 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1250 | int nimaps = 1; |
Nathan Scott | fdc7ed7 | 2005-11-02 15:13:13 +1100 | [diff] [blame] | 1251 | xfs_off_t offset; |
| 1252 | ssize_t size; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1253 | int new = 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1254 | |
| 1255 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 1256 | return -XFS_ERROR(EIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | |
Nathan Scott | fdc7ed7 | 2005-11-02 15:13:13 +1100 | [diff] [blame] | 1258 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1259 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
| 1260 | size = bh_result->b_size; |
Lachlan McIlroy | 364f358a | 2008-09-17 16:50:14 +1000 | [diff] [blame] | 1261 | |
| 1262 | if (!create && direct && offset >= i_size_read(inode)) |
| 1263 | return 0; |
| 1264 | |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1265 | /* |
| 1266 | * Direct I/O is usually done on preallocated files, so try getting |
| 1267 | * a block mapping without an exclusive lock first. For buffered |
| 1268 | * writes we already have the exclusive iolock anyway, so avoiding |
| 1269 | * a lock roundtrip here by taking the ilock exclusive from the |
| 1270 | * beginning is a useful micro optimization. |
| 1271 | */ |
| 1272 | if (create && !direct) { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1273 | lockmode = XFS_ILOCK_EXCL; |
| 1274 | xfs_ilock(ip, lockmode); |
| 1275 | } else { |
Christoph Hellwig | 309ecac8 | 2013-12-06 12:30:09 -0800 | [diff] [blame] | 1276 | lockmode = xfs_ilock_data_map_shared(ip); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1277 | } |
Christoph Hellwig | f2bde9b | 2010-06-24 11:44:35 +1000 | [diff] [blame] | 1278 | |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 1279 | ASSERT(offset <= mp->m_super->s_maxbytes); |
| 1280 | if (offset + size > mp->m_super->s_maxbytes) |
| 1281 | size = mp->m_super->s_maxbytes - offset; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1282 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
| 1283 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
| 1284 | |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 1285 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
| 1286 | &imap, &nimaps, XFS_BMAPI_ENTIRE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | if (error) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1288 | goto out_unlock; |
| 1289 | |
| 1290 | if (create && |
| 1291 | (!nimaps || |
| 1292 | (imap.br_startblock == HOLESTARTBLOCK || |
| 1293 | imap.br_startblock == DELAYSTARTBLOCK))) { |
Dave Chinner | aff3a9e | 2012-04-23 15:58:44 +1000 | [diff] [blame] | 1294 | if (direct || xfs_get_extsz_hint(ip)) { |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1295 | /* |
| 1296 | * Drop the ilock in preparation for starting the block |
| 1297 | * allocation transaction. It will be retaken |
| 1298 | * exclusively inside xfs_iomap_write_direct for the |
| 1299 | * actual allocation. |
| 1300 | */ |
| 1301 | xfs_iunlock(ip, lockmode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1302 | error = xfs_iomap_write_direct(ip, offset, size, |
| 1303 | &imap, nimaps); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1304 | if (error) |
| 1305 | return -error; |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1306 | new = 1; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1307 | } else { |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1308 | /* |
| 1309 | * Delalloc reservations do not require a transaction, |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1310 | * we can go on without dropping the lock here. If we |
| 1311 | * are allocating a new delalloc block, make sure that |
| 1312 | * we set the new flag so that we mark the buffer new so |
| 1313 | * that we know that it is newly allocated if the write |
| 1314 | * fails. |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1315 | */ |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1316 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
| 1317 | new = 1; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1318 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1319 | if (error) |
| 1320 | goto out_unlock; |
| 1321 | |
| 1322 | xfs_iunlock(ip, lockmode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1323 | } |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1324 | |
| 1325 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); |
| 1326 | } else if (nimaps) { |
| 1327 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1328 | xfs_iunlock(ip, lockmode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1329 | } else { |
| 1330 | trace_xfs_get_blocks_notfound(ip, offset, size); |
| 1331 | goto out_unlock; |
| 1332 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1334 | if (imap.br_startblock != HOLESTARTBLOCK && |
| 1335 | imap.br_startblock != DELAYSTARTBLOCK) { |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 1336 | /* |
| 1337 | * For unwritten extents do not report a disk address on |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | * the read case (treat as if we're reading into a hole). |
| 1339 | */ |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1340 | if (create || !ISUNWRITTEN(&imap)) |
| 1341 | xfs_map_buffer(inode, bh_result, &imap, offset); |
| 1342 | if (create && ISUNWRITTEN(&imap)) { |
Christoph Hellwig | 7b7a866 | 2013-09-04 15:04:39 +0200 | [diff] [blame] | 1343 | if (direct) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | bh_result->b_private = inode; |
Christoph Hellwig | 7b7a866 | 2013-09-04 15:04:39 +0200 | [diff] [blame] | 1345 | set_buffer_defer_completion(bh_result); |
| 1346 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1347 | set_buffer_unwritten(bh_result); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | } |
| 1349 | } |
| 1350 | |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1351 | /* |
| 1352 | * If this is a realtime file, data may be on a different device. |
| 1353 | * to that pointed to from the buffer_head b_bdev currently. |
| 1354 | */ |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 1355 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1357 | /* |
David Chinner | 549054a | 2007-02-10 18:36:35 +1100 | [diff] [blame] | 1358 | * If we previously allocated a block out beyond eof and we are now |
| 1359 | * coming back to use it then we will need to flag it as new even if it |
| 1360 | * has a disk address. |
| 1361 | * |
| 1362 | * With sub-block writes into unwritten extents we also need to mark |
| 1363 | * the buffer as new so that the unwritten parts of the buffer gets |
| 1364 | * correctly zeroed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | */ |
| 1366 | if (create && |
| 1367 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || |
David Chinner | 549054a | 2007-02-10 18:36:35 +1100 | [diff] [blame] | 1368 | (offset >= i_size_read(inode)) || |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1369 | (new || ISUNWRITTEN(&imap)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | set_buffer_new(bh_result); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1371 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1372 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | BUG_ON(direct); |
| 1374 | if (create) { |
| 1375 | set_buffer_uptodate(bh_result); |
| 1376 | set_buffer_mapped(bh_result); |
| 1377 | set_buffer_delay(bh_result); |
| 1378 | } |
| 1379 | } |
| 1380 | |
Christoph Hellwig | 2b8f12b | 2010-04-28 12:28:59 +0000 | [diff] [blame] | 1381 | /* |
| 1382 | * If this is O_DIRECT or the mpage code calling tell them how large |
| 1383 | * the mapping is, so that we can avoid repeated get_blocks calls. |
Dave Chinner | 0e1f789 | 2014-04-17 08:15:19 +1000 | [diff] [blame] | 1384 | * |
| 1385 | * If the mapping spans EOF, then we have to break the mapping up as the |
| 1386 | * mapping for blocks beyond EOF must be marked new so that sub block |
| 1387 | * regions can be correctly zeroed. We can't do this for mappings within |
| 1388 | * EOF unless the mapping was just allocated or is unwritten, otherwise |
| 1389 | * the callers would overwrite existing data with zeros. Hence we have |
| 1390 | * to split the mapping into a range up to and including EOF, and a |
| 1391 | * second mapping for beyond EOF. |
Christoph Hellwig | 2b8f12b | 2010-04-28 12:28:59 +0000 | [diff] [blame] | 1392 | */ |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1393 | if (direct || size > (1 << inode->i_blkbits)) { |
Christoph Hellwig | 2b8f12b | 2010-04-28 12:28:59 +0000 | [diff] [blame] | 1394 | xfs_off_t mapping_size; |
Christoph Hellwig | 9563b3d | 2010-04-28 12:28:53 +0000 | [diff] [blame] | 1395 | |
Christoph Hellwig | 2b8f12b | 2010-04-28 12:28:59 +0000 | [diff] [blame] | 1396 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; |
| 1397 | mapping_size <<= inode->i_blkbits; |
| 1398 | |
| 1399 | ASSERT(mapping_size > 0); |
| 1400 | if (mapping_size > size) |
| 1401 | mapping_size = size; |
Dave Chinner | 0e1f789 | 2014-04-17 08:15:19 +1000 | [diff] [blame] | 1402 | if (offset < i_size_read(inode) && |
| 1403 | offset + mapping_size >= i_size_read(inode)) { |
| 1404 | /* limit mapping to block that spans EOF */ |
| 1405 | mapping_size = roundup_64(i_size_read(inode) - offset, |
| 1406 | 1 << inode->i_blkbits); |
| 1407 | } |
Christoph Hellwig | 2b8f12b | 2010-04-28 12:28:59 +0000 | [diff] [blame] | 1408 | if (mapping_size > LONG_MAX) |
| 1409 | mapping_size = LONG_MAX; |
| 1410 | |
| 1411 | bh_result->b_size = mapping_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1412 | } |
| 1413 | |
| 1414 | return 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1415 | |
| 1416 | out_unlock: |
| 1417 | xfs_iunlock(ip, lockmode); |
| 1418 | return -error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | } |
| 1420 | |
| 1421 | int |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1422 | xfs_get_blocks( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1423 | struct inode *inode, |
| 1424 | sector_t iblock, |
| 1425 | struct buffer_head *bh_result, |
| 1426 | int create) |
| 1427 | { |
Christoph Hellwig | f2bde9b | 2010-06-24 11:44:35 +1000 | [diff] [blame] | 1428 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1429 | } |
| 1430 | |
| 1431 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1432 | xfs_get_blocks_direct( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | struct inode *inode, |
| 1434 | sector_t iblock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1435 | struct buffer_head *bh_result, |
| 1436 | int create) |
| 1437 | { |
Christoph Hellwig | f2bde9b | 2010-06-24 11:44:35 +1000 | [diff] [blame] | 1438 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1439 | } |
| 1440 | |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1441 | /* |
| 1442 | * Complete a direct I/O write request. |
| 1443 | * |
| 1444 | * If the private argument is non-NULL __xfs_get_blocks signals us that we |
| 1445 | * need to issue a transaction to convert the range from unwritten to written |
| 1446 | * extents. In case this is regular synchronous I/O we just call xfs_end_io |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1447 | * to do this and we are done. But in case this was a successful AIO |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1448 | * request this handler is called from interrupt context, from which we |
| 1449 | * can't start transactions. In that case offload the I/O completion to |
| 1450 | * the workqueues we also use for buffered I/O completion. |
| 1451 | */ |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1452 | STATIC void |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1453 | xfs_end_io_direct_write( |
| 1454 | struct kiocb *iocb, |
| 1455 | loff_t offset, |
| 1456 | ssize_t size, |
Christoph Hellwig | 7b7a866 | 2013-09-04 15:04:39 +0200 | [diff] [blame] | 1457 | void *private) |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1458 | { |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1459 | struct xfs_ioend *ioend = iocb->private; |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1460 | |
| 1461 | /* |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 1462 | * While the generic direct I/O code updates the inode size, it does |
| 1463 | * so only after the end_io handler is called, which means our |
| 1464 | * end_io handler thinks the on-disk size is outside the in-core |
| 1465 | * size. To prevent this just update it a little bit earlier here. |
| 1466 | */ |
| 1467 | if (offset + size > i_size_read(ioend->io_inode)) |
| 1468 | i_size_write(ioend->io_inode, offset + size); |
| 1469 | |
| 1470 | /* |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 1471 | * blockdev_direct_IO can return an error even after the I/O |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1472 | * completion handler was called. Thus we need to protect |
| 1473 | * against double-freeing. |
| 1474 | */ |
| 1475 | iocb->private = NULL; |
Christoph Hellwig | 40e2e97 | 2010-07-18 21:17:09 +0000 | [diff] [blame] | 1476 | |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1477 | ioend->io_offset = offset; |
| 1478 | ioend->io_size = size; |
| 1479 | if (private && size > 0) |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1480 | ioend->io_type = XFS_IO_UNWRITTEN; |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1481 | |
Christoph Hellwig | 7b7a866 | 2013-09-04 15:04:39 +0200 | [diff] [blame] | 1482 | xfs_finish_ioend_sync(ioend); |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1483 | } |
| 1484 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | STATIC ssize_t |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1486 | xfs_vm_direct_IO( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1487 | int rw, |
| 1488 | struct kiocb *iocb, |
Al Viro | d8d3d94 | 2014-03-04 21:27:34 -0500 | [diff] [blame] | 1489 | struct iov_iter *iter, |
| 1490 | loff_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | { |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1492 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
| 1493 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1494 | struct xfs_ioend *ioend = NULL; |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1495 | ssize_t ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1496 | |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1497 | if (rw & WRITE) { |
Al Viro | a6cbcd4 | 2014-03-04 22:38:00 -0500 | [diff] [blame] | 1498 | size_t size = iov_iter_count(iter); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1499 | |
| 1500 | /* |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 1501 | * We cannot preallocate a size update transaction here as we |
| 1502 | * don't know whether allocation is necessary or not. Hence we |
| 1503 | * can only tell IO completion that one is necessary if we are |
| 1504 | * not doing unwritten extent conversion. |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1505 | */ |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1506 | iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT); |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 1507 | if (offset + size > XFS_I(inode)->i_d.di_size) |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1508 | ioend->io_isdirect = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1509 | |
Al Viro | 31b1403 | 2014-03-05 01:33:16 -0500 | [diff] [blame] | 1510 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter, |
| 1511 | offset, xfs_get_blocks_direct, |
Christoph Hellwig | 9862f62 | 2014-02-10 10:28:04 +1100 | [diff] [blame] | 1512 | xfs_end_io_direct_write, NULL, |
| 1513 | DIO_ASYNC_EXTEND); |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1514 | if (ret != -EIOCBQUEUED && iocb->private) |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 1515 | goto out_destroy_ioend; |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1516 | } else { |
Al Viro | 31b1403 | 2014-03-05 01:33:16 -0500 | [diff] [blame] | 1517 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter, |
| 1518 | offset, xfs_get_blocks_direct, |
Christoph Hellwig | eafdc7d | 2010-06-04 11:29:53 +0200 | [diff] [blame] | 1519 | NULL, NULL, 0); |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1520 | } |
Christoph Hellwig | 5fe878ae | 2009-12-15 16:47:50 -0800 | [diff] [blame] | 1521 | |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1522 | return ret; |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1523 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1524 | out_destroy_ioend: |
| 1525 | xfs_destroy_ioend(ioend); |
| 1526 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | } |
| 1528 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1529 | /* |
| 1530 | * Punch out the delalloc blocks we have already allocated. |
| 1531 | * |
| 1532 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet |
| 1533 | * as the page is still locked at this point. |
| 1534 | */ |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1535 | STATIC void |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1536 | xfs_vm_kill_delalloc_range( |
| 1537 | struct inode *inode, |
| 1538 | loff_t start, |
| 1539 | loff_t end) |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1540 | { |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1541 | struct xfs_inode *ip = XFS_I(inode); |
| 1542 | xfs_fileoff_t start_fsb; |
| 1543 | xfs_fileoff_t end_fsb; |
| 1544 | int error; |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1545 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1546 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); |
| 1547 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); |
| 1548 | if (end_fsb <= start_fsb) |
| 1549 | return; |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 1550 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1551 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 1552 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, |
| 1553 | end_fsb - start_fsb); |
| 1554 | if (error) { |
| 1555 | /* something screwed, just bail */ |
| 1556 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
| 1557 | xfs_alert(ip->i_mount, |
| 1558 | "xfs_vm_write_failed: unable to clean up ino %lld", |
| 1559 | ip->i_ino); |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 1560 | } |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1561 | } |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1562 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1563 | } |
| 1564 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1565 | STATIC void |
| 1566 | xfs_vm_write_failed( |
| 1567 | struct inode *inode, |
| 1568 | struct page *page, |
| 1569 | loff_t pos, |
| 1570 | unsigned len) |
| 1571 | { |
Jie Liu | 58e5985 | 2013-07-16 13:11:16 +0800 | [diff] [blame] | 1572 | loff_t block_offset; |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1573 | loff_t block_start; |
| 1574 | loff_t block_end; |
| 1575 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); |
| 1576 | loff_t to = from + len; |
| 1577 | struct buffer_head *bh, *head; |
| 1578 | |
Jie Liu | 58e5985 | 2013-07-16 13:11:16 +0800 | [diff] [blame] | 1579 | /* |
| 1580 | * The request pos offset might be 32 or 64 bit, this is all fine |
| 1581 | * on 64-bit platform. However, for 64-bit pos request on 32-bit |
| 1582 | * platform, the high 32-bit will be masked off if we evaluate the |
| 1583 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is |
| 1584 | * 0xfffff000 as an unsigned long, hence the result is incorrect |
| 1585 | * which could cause the following ASSERT failed in most cases. |
| 1586 | * In order to avoid this, we can evaluate the block_offset of the |
| 1587 | * start of the page by using shifts rather than masks the mismatch |
| 1588 | * problem. |
| 1589 | */ |
| 1590 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; |
| 1591 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1592 | ASSERT(block_offset + from == pos); |
| 1593 | |
| 1594 | head = page_buffers(page); |
| 1595 | block_start = 0; |
| 1596 | for (bh = head; bh != head || !block_start; |
| 1597 | bh = bh->b_this_page, block_start = block_end, |
| 1598 | block_offset += bh->b_size) { |
| 1599 | block_end = block_start + bh->b_size; |
| 1600 | |
| 1601 | /* skip buffers before the write */ |
| 1602 | if (block_end <= from) |
| 1603 | continue; |
| 1604 | |
| 1605 | /* if the buffer is after the write, we're done */ |
| 1606 | if (block_start >= to) |
| 1607 | break; |
| 1608 | |
| 1609 | if (!buffer_delay(bh)) |
| 1610 | continue; |
| 1611 | |
| 1612 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) |
| 1613 | continue; |
| 1614 | |
| 1615 | xfs_vm_kill_delalloc_range(inode, block_offset, |
| 1616 | block_offset + bh->b_size); |
Dave Chinner | 4ab9ed5 | 2014-04-14 18:11:58 +1000 | [diff] [blame] | 1617 | |
| 1618 | /* |
| 1619 | * This buffer does not contain data anymore. make sure anyone |
| 1620 | * who finds it knows that for certain. |
| 1621 | */ |
| 1622 | clear_buffer_delay(bh); |
| 1623 | clear_buffer_uptodate(bh); |
| 1624 | clear_buffer_mapped(bh); |
| 1625 | clear_buffer_new(bh); |
| 1626 | clear_buffer_dirty(bh); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1627 | } |
| 1628 | |
| 1629 | } |
| 1630 | |
| 1631 | /* |
| 1632 | * This used to call block_write_begin(), but it unlocks and releases the page |
| 1633 | * on error, and we need that page to be able to punch stale delalloc blocks out |
| 1634 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at |
| 1635 | * the appropriate point. |
| 1636 | */ |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1637 | STATIC int |
Nick Piggin | d79689c | 2007-10-16 01:25:06 -0700 | [diff] [blame] | 1638 | xfs_vm_write_begin( |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1639 | struct file *file, |
Nick Piggin | d79689c | 2007-10-16 01:25:06 -0700 | [diff] [blame] | 1640 | struct address_space *mapping, |
| 1641 | loff_t pos, |
| 1642 | unsigned len, |
| 1643 | unsigned flags, |
| 1644 | struct page **pagep, |
| 1645 | void **fsdata) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1646 | { |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1647 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
| 1648 | struct page *page; |
| 1649 | int status; |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 1650 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1651 | ASSERT(len <= PAGE_CACHE_SIZE); |
| 1652 | |
Dave Chinner | ad22c7a | 2013-10-29 22:11:57 +1100 | [diff] [blame] | 1653 | page = grab_cache_page_write_begin(mapping, index, flags); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1654 | if (!page) |
| 1655 | return -ENOMEM; |
| 1656 | |
| 1657 | status = __block_write_begin(page, pos, len, xfs_get_blocks); |
| 1658 | if (unlikely(status)) { |
| 1659 | struct inode *inode = mapping->host; |
Dave Chinner | 72ab70a | 2014-04-14 18:13:29 +1000 | [diff] [blame] | 1660 | size_t isize = i_size_read(inode); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1661 | |
| 1662 | xfs_vm_write_failed(inode, page, pos, len); |
| 1663 | unlock_page(page); |
| 1664 | |
Dave Chinner | 72ab70a | 2014-04-14 18:13:29 +1000 | [diff] [blame] | 1665 | /* |
| 1666 | * If the write is beyond EOF, we only want to kill blocks |
| 1667 | * allocated in this write, not blocks that were previously |
| 1668 | * written successfully. |
| 1669 | */ |
| 1670 | if (pos + len > isize) { |
| 1671 | ssize_t start = max_t(ssize_t, pos, isize); |
| 1672 | |
| 1673 | truncate_pagecache_range(inode, start, pos + len); |
| 1674 | } |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1675 | |
| 1676 | page_cache_release(page); |
| 1677 | page = NULL; |
| 1678 | } |
| 1679 | |
| 1680 | *pagep = page; |
| 1681 | return status; |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1682 | } |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 1683 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1684 | /* |
Dave Chinner | aad3f37 | 2014-04-14 18:14:11 +1000 | [diff] [blame] | 1685 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
| 1686 | * this specific write because they will never be written. Previous writes |
| 1687 | * beyond EOF where block allocation succeeded do not need to be trashed, so |
| 1688 | * only new blocks from this write should be trashed. For blocks within |
| 1689 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be |
| 1690 | * written with all the other valid data. |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1691 | */ |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1692 | STATIC int |
| 1693 | xfs_vm_write_end( |
| 1694 | struct file *file, |
| 1695 | struct address_space *mapping, |
| 1696 | loff_t pos, |
| 1697 | unsigned len, |
| 1698 | unsigned copied, |
| 1699 | struct page *page, |
| 1700 | void *fsdata) |
| 1701 | { |
| 1702 | int ret; |
| 1703 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1704 | ASSERT(len <= PAGE_CACHE_SIZE); |
| 1705 | |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1706 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1707 | if (unlikely(ret < len)) { |
| 1708 | struct inode *inode = mapping->host; |
| 1709 | size_t isize = i_size_read(inode); |
| 1710 | loff_t to = pos + len; |
| 1711 | |
| 1712 | if (to > isize) { |
Dave Chinner | aad3f37 | 2014-04-14 18:14:11 +1000 | [diff] [blame] | 1713 | /* only kill blocks in this write beyond EOF */ |
| 1714 | if (pos > isize) |
| 1715 | isize = pos; |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1716 | xfs_vm_kill_delalloc_range(inode, isize, to); |
Dave Chinner | aad3f37 | 2014-04-14 18:14:11 +1000 | [diff] [blame] | 1717 | truncate_pagecache_range(inode, isize, to); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1718 | } |
| 1719 | } |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 1720 | return ret; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1721 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1722 | |
| 1723 | STATIC sector_t |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1724 | xfs_vm_bmap( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1725 | struct address_space *mapping, |
| 1726 | sector_t block) |
| 1727 | { |
| 1728 | struct inode *inode = (struct inode *)mapping->host; |
Christoph Hellwig | 739bfb2 | 2007-08-29 10:58:01 +1000 | [diff] [blame] | 1729 | struct xfs_inode *ip = XFS_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1730 | |
Christoph Hellwig | cca28fb | 2010-06-24 11:57:09 +1000 | [diff] [blame] | 1731 | trace_xfs_vm_bmap(XFS_I(inode)); |
Christoph Hellwig | 126468b | 2008-03-06 13:44:57 +1100 | [diff] [blame] | 1732 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
Dave Chinner | 4bc1ea6 | 2012-11-12 22:53:56 +1100 | [diff] [blame] | 1733 | filemap_write_and_wait(mapping); |
Christoph Hellwig | 126468b | 2008-03-06 13:44:57 +1100 | [diff] [blame] | 1734 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1735 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1736 | } |
| 1737 | |
| 1738 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1739 | xfs_vm_readpage( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1740 | struct file *unused, |
| 1741 | struct page *page) |
| 1742 | { |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1743 | return mpage_readpage(page, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1744 | } |
| 1745 | |
| 1746 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1747 | xfs_vm_readpages( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | struct file *unused, |
| 1749 | struct address_space *mapping, |
| 1750 | struct list_head *pages, |
| 1751 | unsigned nr_pages) |
| 1752 | { |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1753 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1754 | } |
| 1755 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1756 | const struct address_space_operations xfs_address_space_operations = { |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1757 | .readpage = xfs_vm_readpage, |
| 1758 | .readpages = xfs_vm_readpages, |
| 1759 | .writepage = xfs_vm_writepage, |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1760 | .writepages = xfs_vm_writepages, |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1761 | .releasepage = xfs_vm_releasepage, |
| 1762 | .invalidatepage = xfs_vm_invalidatepage, |
Nick Piggin | d79689c | 2007-10-16 01:25:06 -0700 | [diff] [blame] | 1763 | .write_begin = xfs_vm_write_begin, |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1764 | .write_end = xfs_vm_write_end, |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1765 | .bmap = xfs_vm_bmap, |
| 1766 | .direct_IO = xfs_vm_direct_IO, |
Christoph Lameter | e965f96 | 2006-02-01 03:05:41 -0800 | [diff] [blame] | 1767 | .migratepage = buffer_migrate_page, |
Hisashi Hifumi | bddaafa | 2009-03-29 09:53:38 +0200 | [diff] [blame] | 1768 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 1769 | .error_remove_page = generic_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1770 | }; |