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_mount.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include "xfs_inode.h" |
Dave Chinner | 239880e | 2013-10-23 10:50:10 +1100 | [diff] [blame] | 25 | #include "xfs_trans.h" |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 26 | #include "xfs_inode_item.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 27 | #include "xfs_alloc.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include "xfs_error.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include "xfs_iomap.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 30 | #include "xfs_trace.h" |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 31 | #include "xfs_bmap.h" |
Dave Chinner | 6898811 | 2013-08-12 20:49:42 +1000 | [diff] [blame] | 32 | #include "xfs_bmap_util.h" |
Dave Chinner | a4fbe6a | 2013-10-23 10:51:50 +1100 | [diff] [blame] | 33 | #include "xfs_bmap_btree.h" |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 34 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <linux/mpage.h> |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 36 | #include <linux/pagevec.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | #include <linux/writeback.h> |
| 38 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 39 | void |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 40 | xfs_count_page_state( |
| 41 | struct page *page, |
| 42 | int *delalloc, |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 43 | int *unwritten) |
| 44 | { |
| 45 | struct buffer_head *bh, *head; |
| 46 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 47 | *delalloc = *unwritten = 0; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 48 | |
| 49 | bh = head = page_buffers(page); |
| 50 | do { |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 51 | if (buffer_unwritten(bh)) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 52 | (*unwritten) = 1; |
| 53 | else if (buffer_delay(bh)) |
| 54 | (*delalloc) = 1; |
| 55 | } while ((bh = bh->b_this_page) != head); |
| 56 | } |
| 57 | |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 58 | STATIC struct block_device * |
| 59 | xfs_find_bdev_for_inode( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 60 | struct inode *inode) |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 61 | { |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 62 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 63 | struct xfs_mount *mp = ip->i_mount; |
| 64 | |
Eric Sandeen | 71ddabb | 2007-11-23 16:29:42 +1100 | [diff] [blame] | 65 | if (XFS_IS_REALTIME_INODE(ip)) |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 66 | return mp->m_rtdev_targp->bt_bdev; |
| 67 | else |
| 68 | return mp->m_ddev_targp->bt_bdev; |
| 69 | } |
| 70 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 71 | /* |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 72 | * We're now finished for good with this ioend structure. |
| 73 | * Update the page state via the associated buffer_heads, |
| 74 | * release holds on the inode and bio, and finally free |
| 75 | * up memory. Do not use the ioend after this. |
| 76 | */ |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 77 | STATIC void |
| 78 | xfs_destroy_ioend( |
| 79 | xfs_ioend_t *ioend) |
| 80 | { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 81 | struct buffer_head *bh, *next; |
| 82 | |
| 83 | for (bh = ioend->io_buffer_head; bh; bh = next) { |
| 84 | next = bh->b_private; |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 85 | bh->b_end_io(bh, !ioend->io_error); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 86 | } |
Christoph Hellwig | 583fa58 | 2008-12-03 12:20:38 +0100 | [diff] [blame] | 87 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 88 | mempool_free(ioend, xfs_ioend_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | } |
| 90 | |
| 91 | /* |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 92 | * Fast and loose check if this write could update the on-disk inode size. |
| 93 | */ |
| 94 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) |
| 95 | { |
| 96 | return ioend->io_offset + ioend->io_size > |
| 97 | XFS_I(ioend->io_inode)->i_d.di_size; |
| 98 | } |
| 99 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 100 | STATIC int |
| 101 | xfs_setfilesize_trans_alloc( |
| 102 | struct xfs_ioend *ioend) |
| 103 | { |
| 104 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
| 105 | struct xfs_trans *tp; |
| 106 | int error; |
| 107 | |
| 108 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); |
| 109 | |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 110 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 111 | if (error) { |
Christoph Hellwig | 4906e21 | 2015-06-04 13:47:56 +1000 | [diff] [blame] | 112 | xfs_trans_cancel(tp); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 113 | return error; |
| 114 | } |
| 115 | |
| 116 | ioend->io_append_trans = tp; |
| 117 | |
| 118 | /* |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 119 | * We may pass freeze protection with a transaction. So tell lockdep |
Jan Kara | d9457dc | 2012-06-12 16:20:39 +0200 | [diff] [blame] | 120 | * we released it. |
| 121 | */ |
Oleg Nesterov | bee9182 | 2015-07-19 23:48:20 +0200 | [diff] [blame] | 122 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
Jan Kara | d9457dc | 2012-06-12 16:20:39 +0200 | [diff] [blame] | 123 | /* |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 124 | * We hand off the transaction to the completion thread now, so |
| 125 | * clear the flag here. |
| 126 | */ |
| 127 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
| 128 | return 0; |
| 129 | } |
| 130 | |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 131 | /* |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 132 | * 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] | 133 | */ |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 134 | STATIC int |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 135 | xfs_setfilesize( |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 136 | struct xfs_inode *ip, |
| 137 | struct xfs_trans *tp, |
| 138 | xfs_off_t offset, |
| 139 | size_t size) |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 140 | { |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 141 | xfs_fsize_t isize; |
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 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 144 | isize = xfs_new_eof(ip, offset + size); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 145 | if (!isize) { |
| 146 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | 4906e21 | 2015-06-04 13:47:56 +1000 | [diff] [blame] | 147 | xfs_trans_cancel(tp); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 148 | return 0; |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 149 | } |
| 150 | |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 151 | trace_xfs_setfilesize(ip, offset, size); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 152 | |
| 153 | ip->i_d.di_size = isize; |
| 154 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
| 155 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 156 | |
Christoph Hellwig | 7039331 | 2015-06-04 13:48:08 +1000 | [diff] [blame] | 157 | return xfs_trans_commit(tp); |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 158 | } |
| 159 | |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 160 | STATIC int |
| 161 | xfs_setfilesize_ioend( |
| 162 | struct xfs_ioend *ioend) |
| 163 | { |
| 164 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
| 165 | struct xfs_trans *tp = ioend->io_append_trans; |
| 166 | |
| 167 | /* |
| 168 | * The transaction may have been allocated in the I/O submission thread, |
| 169 | * thus we need to mark ourselves as being in a transaction manually. |
| 170 | * Similarly for freeze protection. |
| 171 | */ |
| 172 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
Oleg Nesterov | bee9182 | 2015-07-19 23:48:20 +0200 | [diff] [blame] | 173 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 174 | |
Zhaohongjiang | 5cb13dc | 2015-10-12 15:28:39 +1100 | [diff] [blame] | 175 | /* we abort the update if there was an IO error */ |
| 176 | if (ioend->io_error) { |
| 177 | xfs_trans_cancel(tp); |
| 178 | return ioend->io_error; |
| 179 | } |
| 180 | |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 181 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
| 182 | } |
| 183 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 184 | /* |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 185 | * Schedule IO completion handling on the final put of an ioend. |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 186 | * |
| 187 | * If there is no work to do we might as well call it a day and free the |
| 188 | * ioend right now. |
Dave Chinner | c626d17 | 2009-04-06 18:42:11 +0200 | [diff] [blame] | 189 | */ |
| 190 | STATIC void |
| 191 | xfs_finish_ioend( |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 192 | struct xfs_ioend *ioend) |
Dave Chinner | c626d17 | 2009-04-06 18:42:11 +0200 | [diff] [blame] | 193 | { |
| 194 | if (atomic_dec_and_test(&ioend->io_remaining)) { |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 195 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
| 196 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 197 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 198 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 199 | else if (ioend->io_append_trans) |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 200 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 201 | else |
| 202 | xfs_destroy_ioend(ioend); |
Dave Chinner | c626d17 | 2009-04-06 18:42:11 +0200 | [diff] [blame] | 203 | } |
| 204 | } |
| 205 | |
| 206 | /* |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 207 | * IO write completion. |
| 208 | */ |
| 209 | STATIC void |
| 210 | xfs_end_io( |
| 211 | struct work_struct *work) |
| 212 | { |
| 213 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
| 214 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
Dave Chinner | 6941893 | 2010-03-04 00:57:09 +0000 | [diff] [blame] | 215 | int error = 0; |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 216 | |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 217 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Christoph Hellwig | 810627d | 2011-11-08 08:56:15 +0000 | [diff] [blame] | 218 | ioend->io_error = -EIO; |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 219 | goto done; |
| 220 | } |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 221 | |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 222 | /* |
| 223 | * For unwritten extents we need to issue transactions to convert a |
| 224 | * range to normal written extens after the data I/O has finished. |
Zhaohongjiang | 5cb13dc | 2015-10-12 15:28:39 +1100 | [diff] [blame] | 225 | * Detecting and handling completion IO errors is done individually |
| 226 | * for each case as different cleanup operations need to be performed |
| 227 | * on error. |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 228 | */ |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 229 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
Zhaohongjiang | 5cb13dc | 2015-10-12 15:28:39 +1100 | [diff] [blame] | 230 | if (ioend->io_error) |
| 231 | goto done; |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 232 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 233 | ioend->io_size); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 234 | } else if (ioend->io_append_trans) { |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 235 | error = xfs_setfilesize_ioend(ioend); |
Christoph Hellwig | 84803fb | 2012-02-29 09:53:50 +0000 | [diff] [blame] | 236 | } else { |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 237 | ASSERT(!xfs_ioend_is_append(ioend)); |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 238 | } |
| 239 | |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 240 | done: |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 241 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 242 | ioend->io_error = error; |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 243 | xfs_destroy_ioend(ioend); |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 244 | } |
| 245 | |
| 246 | /* |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 247 | * Allocate and initialise an IO completion structure. |
| 248 | * We need to track unwritten extent write completion here initially. |
| 249 | * We'll need to extend this for updating the ondisk inode size later |
| 250 | * (vs. incore size). |
| 251 | */ |
| 252 | STATIC xfs_ioend_t * |
| 253 | xfs_alloc_ioend( |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 254 | struct inode *inode, |
| 255 | unsigned int type) |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 256 | { |
| 257 | xfs_ioend_t *ioend; |
| 258 | |
| 259 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); |
| 260 | |
| 261 | /* |
| 262 | * Set the count to 1 initially, which will prevent an I/O |
| 263 | * completion callback from happening before we have started |
| 264 | * all the I/O from calling the completion routine too early. |
| 265 | */ |
| 266 | atomic_set(&ioend->io_remaining, 1); |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 267 | ioend->io_error = 0; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 268 | ioend->io_list = NULL; |
| 269 | ioend->io_type = type; |
Christoph Hellwig | b677c21 | 2007-08-29 11:46:28 +1000 | [diff] [blame] | 270 | ioend->io_inode = inode; |
Christoph Hellwig | c1a073b | 2005-09-05 08:23:35 +1000 | [diff] [blame] | 271 | ioend->io_buffer_head = NULL; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 272 | ioend->io_buffer_tail = NULL; |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 273 | ioend->io_offset = 0; |
| 274 | ioend->io_size = 0; |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 275 | ioend->io_append_trans = NULL; |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 276 | |
Christoph Hellwig | 5ec4fab | 2009-10-30 09:11:47 +0000 | [diff] [blame] | 277 | INIT_WORK(&ioend->io_work, xfs_end_io); |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 278 | return ioend; |
| 279 | } |
| 280 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | STATIC int |
| 282 | xfs_map_blocks( |
| 283 | struct inode *inode, |
| 284 | loff_t offset, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 285 | struct xfs_bmbt_irec *imap, |
Dave Chinner | 988ef92 | 2016-02-15 17:20:50 +1100 | [diff] [blame^] | 286 | int type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 288 | struct xfs_inode *ip = XFS_I(inode); |
| 289 | struct xfs_mount *mp = ip->i_mount; |
Christoph Hellwig | ed1e7b7 | 2010-12-10 08:42:22 +0000 | [diff] [blame] | 290 | ssize_t count = 1 << inode->i_blkbits; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 291 | xfs_fileoff_t offset_fsb, end_fsb; |
| 292 | int error = 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 293 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
| 294 | int nimaps = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 296 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 297 | return -EIO; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 298 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 299 | if (type == XFS_IO_UNWRITTEN) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 300 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 301 | |
Dave Chinner | 988ef92 | 2016-02-15 17:20:50 +1100 | [diff] [blame^] | 302 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 303 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
| 304 | (ip->i_df.if_flags & XFS_IFEXTENTS)); |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 305 | ASSERT(offset <= mp->m_super->s_maxbytes); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 306 | |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 307 | if (offset + count > mp->m_super->s_maxbytes) |
| 308 | count = mp->m_super->s_maxbytes - offset; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 309 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
| 310 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 311 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
| 312 | imap, &nimaps, bmapi_flags); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 313 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
| 314 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 315 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 316 | return error; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 317 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 318 | if (type == XFS_IO_DELALLOC && |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 319 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
Jie Liu | 0799a3e | 2013-09-29 18:56:04 +0800 | [diff] [blame] | 320 | error = xfs_iomap_write_allocate(ip, offset, imap); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 321 | if (!error) |
| 322 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 323 | return error; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 324 | } |
| 325 | |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 326 | #ifdef DEBUG |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 327 | if (type == XFS_IO_UNWRITTEN) { |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 328 | ASSERT(nimaps); |
| 329 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 330 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
| 331 | } |
| 332 | #endif |
| 333 | if (nimaps) |
| 334 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); |
| 335 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | } |
| 337 | |
Christoph Hellwig | b8f82a4 | 2009-11-14 16:17:22 +0000 | [diff] [blame] | 338 | STATIC int |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 339 | xfs_imap_valid( |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 340 | struct inode *inode, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 341 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 342 | xfs_off_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | { |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 344 | offset >>= inode->i_blkbits; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 345 | |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 346 | return offset >= imap->br_startoff && |
| 347 | offset < imap->br_startoff + imap->br_blockcount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | } |
| 349 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 350 | /* |
| 351 | * BIO completion handler for buffered IO. |
| 352 | */ |
Al Viro | 782e3b3 | 2007-10-12 07:17:47 +0100 | [diff] [blame] | 353 | STATIC void |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 354 | xfs_end_bio( |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 355 | struct bio *bio) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 356 | { |
| 357 | xfs_ioend_t *ioend = bio->bi_private; |
| 358 | |
Linus Torvalds | 77a7880 | 2015-09-07 13:28:32 -0700 | [diff] [blame] | 359 | if (!ioend->io_error) |
| 360 | ioend->io_error = bio->bi_error; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 361 | |
| 362 | /* Toss bio and pass work off to an xfsdatad thread */ |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 363 | bio->bi_private = NULL; |
| 364 | bio->bi_end_io = NULL; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 365 | bio_put(bio); |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 366 | |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 367 | xfs_finish_ioend(ioend); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 368 | } |
| 369 | |
| 370 | STATIC void |
| 371 | xfs_submit_ioend_bio( |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 372 | struct writeback_control *wbc, |
| 373 | xfs_ioend_t *ioend, |
| 374 | struct bio *bio) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 375 | { |
| 376 | atomic_inc(&ioend->io_remaining); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 377 | bio->bi_private = ioend; |
| 378 | bio->bi_end_io = xfs_end_bio; |
Jens Axboe | 721a960 | 2011-03-09 11:56:30 +0100 | [diff] [blame] | 379 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 380 | } |
| 381 | |
| 382 | STATIC struct bio * |
| 383 | xfs_alloc_ioend_bio( |
| 384 | struct buffer_head *bh) |
| 385 | { |
Kent Overstreet | b54ffb7 | 2015-05-19 14:31:01 +0200 | [diff] [blame] | 386 | struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 387 | |
| 388 | ASSERT(bio->bi_private == NULL); |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 389 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 390 | bio->bi_bdev = bh->b_bdev; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 391 | return bio; |
| 392 | } |
| 393 | |
| 394 | STATIC void |
| 395 | xfs_start_buffer_writeback( |
| 396 | struct buffer_head *bh) |
| 397 | { |
| 398 | ASSERT(buffer_mapped(bh)); |
| 399 | ASSERT(buffer_locked(bh)); |
| 400 | ASSERT(!buffer_delay(bh)); |
| 401 | ASSERT(!buffer_unwritten(bh)); |
| 402 | |
| 403 | mark_buffer_async_write(bh); |
| 404 | set_buffer_uptodate(bh); |
| 405 | clear_buffer_dirty(bh); |
| 406 | } |
| 407 | |
| 408 | STATIC void |
| 409 | xfs_start_page_writeback( |
| 410 | struct page *page, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 411 | int clear_dirty, |
| 412 | int buffers) |
| 413 | { |
| 414 | ASSERT(PageLocked(page)); |
| 415 | ASSERT(!PageWriteback(page)); |
Dave Chinner | 0d085a5 | 2014-09-23 15:36:27 +1000 | [diff] [blame] | 416 | |
| 417 | /* |
| 418 | * if the page was not fully cleaned, we need to ensure that the higher |
| 419 | * layers come back to it correctly. That means we need to keep the page |
| 420 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the |
| 421 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to |
| 422 | * write this page in this writeback sweep will be made. |
| 423 | */ |
| 424 | if (clear_dirty) { |
David Chinner | 9213202 | 2006-12-21 10:24:01 +1100 | [diff] [blame] | 425 | clear_page_dirty_for_io(page); |
Dave Chinner | 0d085a5 | 2014-09-23 15:36:27 +1000 | [diff] [blame] | 426 | set_page_writeback(page); |
| 427 | } else |
| 428 | set_page_writeback_keepwrite(page); |
| 429 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 430 | unlock_page(page); |
Dave Chinner | 0d085a5 | 2014-09-23 15:36:27 +1000 | [diff] [blame] | 431 | |
Fengguang Wu | 1f7decf | 2007-10-16 23:30:42 -0700 | [diff] [blame] | 432 | /* If no buffers on the page are to be written, finish it here */ |
| 433 | if (!buffers) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 434 | end_page_writeback(page); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 435 | } |
| 436 | |
Zhi Yong Wu | c7c1a7d | 2013-08-07 10:11:09 +0000 | [diff] [blame] | 437 | 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] | 438 | { |
| 439 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
| 440 | } |
| 441 | |
| 442 | /* |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 443 | * Submit all of the bios for all of the ioends we have saved up, covering the |
| 444 | * initial writepage page and also any probed pages. |
| 445 | * |
| 446 | * Because we may have multiple ioends spanning a page, we need to start |
| 447 | * writeback on all the buffers before we submit them for I/O. If we mark the |
| 448 | * buffers as we got, then we can end up with a page that only has buffers |
| 449 | * marked async write and I/O complete on can occur before we mark the other |
| 450 | * buffers async write. |
| 451 | * |
| 452 | * The end result of this is that we trip a bug in end_page_writeback() because |
| 453 | * we call it twice for the one page as the code in end_buffer_async_write() |
| 454 | * assumes that all buffers on the page are started at the same time. |
| 455 | * |
| 456 | * 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] | 457 | * 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] | 458 | * |
| 459 | * If @fail is non-zero, it means that we have a situation where some part of |
| 460 | * the submission process has failed after we have marked paged for writeback |
| 461 | * and unlocked them. In this situation, we need to fail the ioend chain rather |
| 462 | * 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] | 463 | */ |
| 464 | STATIC void |
| 465 | xfs_submit_ioend( |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 466 | struct writeback_control *wbc, |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 467 | xfs_ioend_t *ioend, |
| 468 | int fail) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 469 | { |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 470 | xfs_ioend_t *head = ioend; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 471 | xfs_ioend_t *next; |
| 472 | struct buffer_head *bh; |
| 473 | struct bio *bio; |
| 474 | sector_t lastblock = 0; |
| 475 | |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 476 | /* Pass 1 - start writeback */ |
| 477 | do { |
| 478 | next = ioend->io_list; |
Christoph Hellwig | 221cb25 | 2010-12-10 08:42:17 +0000 | [diff] [blame] | 479 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 480 | xfs_start_buffer_writeback(bh); |
David Chinner | d88992f | 2006-01-18 13:38:12 +1100 | [diff] [blame] | 481 | } while ((ioend = next) != NULL); |
| 482 | |
| 483 | /* Pass 2 - submit I/O */ |
| 484 | ioend = head; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 485 | do { |
| 486 | next = ioend->io_list; |
| 487 | bio = NULL; |
| 488 | |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 489 | /* |
| 490 | * If we are failing the IO now, just mark the ioend with an |
| 491 | * error and finish it. This will run IO completion immediately |
| 492 | * as there is only one reference to the ioend at this point in |
| 493 | * time. |
| 494 | */ |
| 495 | if (fail) { |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 496 | ioend->io_error = fail; |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 497 | xfs_finish_ioend(ioend); |
| 498 | continue; |
| 499 | } |
| 500 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 501 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 502 | |
| 503 | if (!bio) { |
| 504 | retry: |
| 505 | bio = xfs_alloc_ioend_bio(bh); |
| 506 | } else if (bh->b_blocknr != lastblock + 1) { |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 507 | xfs_submit_ioend_bio(wbc, ioend, bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 508 | goto retry; |
| 509 | } |
| 510 | |
Zhi Yong Wu | c7c1a7d | 2013-08-07 10:11:09 +0000 | [diff] [blame] | 511 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 512 | xfs_submit_ioend_bio(wbc, ioend, bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 513 | goto retry; |
| 514 | } |
| 515 | |
| 516 | lastblock = bh->b_blocknr; |
| 517 | } |
| 518 | if (bio) |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 519 | xfs_submit_ioend_bio(wbc, ioend, bio); |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 520 | xfs_finish_ioend(ioend); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 521 | } while ((ioend = next) != NULL); |
| 522 | } |
| 523 | |
| 524 | /* |
| 525 | * Cancel submission of all buffer_heads so far in this endio. |
| 526 | * Toss the endio too. Only ever called for the initial page |
| 527 | * in a writepage request, so only ever one page. |
| 528 | */ |
| 529 | STATIC void |
| 530 | xfs_cancel_ioend( |
| 531 | xfs_ioend_t *ioend) |
| 532 | { |
| 533 | xfs_ioend_t *next; |
| 534 | struct buffer_head *bh, *next_bh; |
| 535 | |
| 536 | do { |
| 537 | next = ioend->io_list; |
| 538 | bh = ioend->io_buffer_head; |
| 539 | do { |
| 540 | next_bh = bh->b_private; |
| 541 | clear_buffer_async_write(bh); |
Brian Foster | 07d0868 | 2014-10-02 09:42:06 +1000 | [diff] [blame] | 542 | /* |
| 543 | * The unwritten flag is cleared when added to the |
| 544 | * ioend. We're not submitting for I/O so mark the |
| 545 | * buffer unwritten again for next time around. |
| 546 | */ |
| 547 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
| 548 | set_buffer_unwritten(bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 549 | unlock_buffer(bh); |
| 550 | } while ((bh = next_bh) != NULL); |
| 551 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 552 | mempool_free(ioend, xfs_ioend_pool); |
| 553 | } while ((ioend = next) != NULL); |
| 554 | } |
| 555 | |
| 556 | /* |
| 557 | * Test to see if we've been building up a completion structure for |
| 558 | * earlier buffers -- if so, we try to append to this ioend if we |
| 559 | * can, otherwise we finish off any current ioend and start another. |
| 560 | * Return true if we've finished the given ioend. |
| 561 | */ |
| 562 | STATIC void |
| 563 | xfs_add_to_ioend( |
| 564 | struct inode *inode, |
| 565 | struct buffer_head *bh, |
Christoph Hellwig | 7336cea | 2006-01-11 20:49:16 +1100 | [diff] [blame] | 566 | xfs_off_t offset, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 567 | unsigned int type, |
| 568 | xfs_ioend_t **result, |
| 569 | int need_ioend) |
| 570 | { |
| 571 | xfs_ioend_t *ioend = *result; |
| 572 | |
| 573 | if (!ioend || need_ioend || type != ioend->io_type) { |
| 574 | xfs_ioend_t *previous = *result; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 575 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 576 | ioend = xfs_alloc_ioend(inode, type); |
| 577 | ioend->io_offset = offset; |
| 578 | ioend->io_buffer_head = bh; |
| 579 | ioend->io_buffer_tail = bh; |
| 580 | if (previous) |
| 581 | previous->io_list = ioend; |
| 582 | *result = ioend; |
| 583 | } else { |
| 584 | ioend->io_buffer_tail->b_private = bh; |
| 585 | ioend->io_buffer_tail = bh; |
| 586 | } |
| 587 | |
| 588 | bh->b_private = NULL; |
| 589 | ioend->io_size += bh->b_size; |
| 590 | } |
| 591 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | STATIC void |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 593 | xfs_map_buffer( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 594 | struct inode *inode, |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 595 | struct buffer_head *bh, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 596 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 597 | xfs_off_t offset) |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 598 | { |
| 599 | sector_t bn; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 600 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 601 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
| 602 | 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] | 603 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 604 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 605 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 606 | |
Christoph Hellwig | e513182 | 2010-04-28 12:28:55 +0000 | [diff] [blame] | 607 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 608 | ((offset - iomap_offset) >> inode->i_blkbits); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 609 | |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 610 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 611 | |
| 612 | bh->b_blocknr = bn; |
| 613 | set_buffer_mapped(bh); |
| 614 | } |
| 615 | |
| 616 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 617 | xfs_map_at_offset( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 618 | struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | struct buffer_head *bh, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 620 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 621 | xfs_off_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | { |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 623 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 624 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 626 | xfs_map_buffer(inode, bh, imap, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | set_buffer_mapped(bh); |
| 628 | clear_buffer_delay(bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 629 | clear_buffer_unwritten(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | } |
| 631 | |
| 632 | /* |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 633 | * Test if a given page contains at least one buffer of a given @type. |
| 634 | * If @check_all_buffers is true, then we walk all the buffers in the page to |
| 635 | * try to find one of the type passed in. If it is not set, then the caller only |
| 636 | * needs to check the first buffer on the page for a match. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | */ |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 638 | STATIC bool |
Dave Chinner | 6ffc4db | 2012-04-23 15:58:43 +1000 | [diff] [blame] | 639 | xfs_check_page_type( |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 640 | struct page *page, |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 641 | unsigned int type, |
| 642 | bool check_all_buffers) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | { |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 644 | struct buffer_head *bh; |
| 645 | struct buffer_head *head; |
| 646 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 647 | if (PageWriteback(page)) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 648 | return false; |
| 649 | if (!page->mapping) |
| 650 | return false; |
| 651 | if (!page_has_buffers(page)) |
| 652 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 653 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 654 | bh = head = page_buffers(page); |
| 655 | do { |
| 656 | if (buffer_unwritten(bh)) { |
| 657 | if (type == XFS_IO_UNWRITTEN) |
| 658 | return true; |
| 659 | } else if (buffer_delay(bh)) { |
Dan Carpenter | 805eeb8 | 2014-04-04 06:56:30 +1100 | [diff] [blame] | 660 | if (type == XFS_IO_DELALLOC) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 661 | return true; |
| 662 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { |
Dan Carpenter | 805eeb8 | 2014-04-04 06:56:30 +1100 | [diff] [blame] | 663 | if (type == XFS_IO_OVERWRITE) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 664 | return true; |
| 665 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 667 | /* If we are only checking the first buffer, we are done now. */ |
| 668 | if (!check_all_buffers) |
| 669 | break; |
| 670 | } while ((bh = bh->b_this_page) != head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 672 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | } |
| 674 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | /* |
| 676 | * Allocate & map buffers for page given the extent map. Write it out. |
| 677 | * except for the original page of a writepage, this is called on |
| 678 | * delalloc/unwritten pages only, for the original page it is possible |
| 679 | * that the page has no mapping at all. |
| 680 | */ |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 681 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | xfs_convert_page( |
| 683 | struct inode *inode, |
| 684 | struct page *page, |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 685 | loff_t tindex, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 686 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 687 | xfs_ioend_t **ioendp, |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 688 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | { |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 690 | struct buffer_head *bh, *head; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 691 | xfs_off_t end_offset; |
| 692 | unsigned long p_offset; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 693 | unsigned int type; |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 694 | int len, page_dirty; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 695 | int count = 0, done = 0, uptodate = 1; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 696 | xfs_off_t offset = page_offset(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 698 | if (page->index != tindex) |
| 699 | goto fail; |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 700 | if (!trylock_page(page)) |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 701 | goto fail; |
| 702 | if (PageWriteback(page)) |
| 703 | goto fail_unlock_page; |
| 704 | if (page->mapping != inode->i_mapping) |
| 705 | goto fail_unlock_page; |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 706 | if (!xfs_check_page_type(page, (*ioendp)->io_type, false)) |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 707 | goto fail_unlock_page; |
| 708 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 709 | /* |
| 710 | * page_dirty is initially a count of buffers on the page before |
Nathan Scott | c41564b | 2006-03-29 08:55:14 +1000 | [diff] [blame] | 711 | * EOF and is decremented as we move each into a cleanable state. |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 712 | * |
| 713 | * Derivation: |
| 714 | * |
| 715 | * End offset is the highest offset that this page should represent. |
| 716 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) |
| 717 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and |
| 718 | * hence give us the correct page_dirty count. On any other page, |
| 719 | * it will be zero and in that case we need page_dirty to be the |
| 720 | * count of buffers on the page. |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 721 | */ |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 722 | end_offset = min_t(unsigned long long, |
| 723 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
| 724 | i_size_read(inode)); |
| 725 | |
Dave Chinner | 480d746 | 2013-05-20 09:51:08 +1000 | [diff] [blame] | 726 | /* |
| 727 | * If the current map does not span the entire page we are about to try |
| 728 | * to write, then give up. The only way we can write a page that spans |
| 729 | * multiple mappings in a single writeback iteration is via the |
| 730 | * xfs_vm_writepage() function. Data integrity writeback requires the |
| 731 | * entire page to be written in a single attempt, otherwise the part of |
| 732 | * the page we don't write here doesn't get written as part of the data |
| 733 | * integrity sync. |
| 734 | * |
| 735 | * For normal writeback, we also don't attempt to write partial pages |
| 736 | * here as it simply means that write_cache_pages() will see it under |
| 737 | * writeback and ignore the page until some point in the future, at |
| 738 | * which time this will be the only page in the file that needs |
| 739 | * writeback. Hence for more optimal IO patterns, we should always |
| 740 | * avoid partial page writeback due to multiple mappings on a page here. |
| 741 | */ |
| 742 | if (!xfs_imap_valid(inode, imap, end_offset)) |
| 743 | goto fail_unlock_page; |
| 744 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 745 | len = 1 << inode->i_blkbits; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 746 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
| 747 | PAGE_CACHE_SIZE); |
| 748 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; |
| 749 | page_dirty = p_offset / len; |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 750 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 751 | /* |
| 752 | * The moment we find a buffer that doesn't match our current type |
| 753 | * specification or can't be written, abort the loop and start |
| 754 | * writeback. As per the above xfs_imap_valid() check, only |
| 755 | * xfs_vm_writepage() can handle partial page writeback fully - we are |
| 756 | * limited here to the buffers that are contiguous with the current |
| 757 | * ioend, and hence a buffer we can't write breaks that contiguity and |
| 758 | * we have to defer the rest of the IO to xfs_vm_writepage(). |
| 759 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | bh = head = page_buffers(page); |
| 761 | do { |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 762 | if (offset >= end_offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | break; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 764 | if (!buffer_uptodate(bh)) |
| 765 | uptodate = 0; |
| 766 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { |
| 767 | done = 1; |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 768 | break; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 769 | } |
| 770 | |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 771 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
| 772 | buffer_mapped(bh)) { |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 773 | if (buffer_unwritten(bh)) |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 774 | type = XFS_IO_UNWRITTEN; |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 775 | else if (buffer_delay(bh)) |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 776 | type = XFS_IO_DELALLOC; |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 777 | else |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 778 | type = XFS_IO_OVERWRITE; |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 779 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 780 | /* |
| 781 | * imap should always be valid because of the above |
| 782 | * partial page end_offset check on the imap. |
| 783 | */ |
| 784 | ASSERT(xfs_imap_valid(inode, imap, offset)); |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 785 | |
Christoph Hellwig | ecff71e | 2010-12-10 08:42:25 +0000 | [diff] [blame] | 786 | lock_buffer(bh); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 787 | if (type != XFS_IO_OVERWRITE) |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 788 | xfs_map_at_offset(inode, bh, imap, offset); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 789 | xfs_add_to_ioend(inode, bh, offset, type, |
| 790 | ioendp, done); |
| 791 | |
Christoph Hellwig | 9260dc6 | 2006-01-11 20:48:47 +1100 | [diff] [blame] | 792 | page_dirty--; |
| 793 | count++; |
| 794 | } else { |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 795 | done = 1; |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 796 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | } |
Christoph Hellwig | 7336cea | 2006-01-11 20:49:16 +1100 | [diff] [blame] | 798 | } while (offset += len, (bh = bh->b_this_page) != head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 800 | if (uptodate && bh == head) |
| 801 | SetPageUptodate(page); |
| 802 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 803 | if (count) { |
Dave Chinner | efceab1 | 2010-08-24 11:44:56 +1000 | [diff] [blame] | 804 | if (--wbc->nr_to_write <= 0 && |
| 805 | wbc->sync_mode == WB_SYNC_NONE) |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 806 | done = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | } |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 808 | xfs_start_page_writeback(page, !page_dirty, count); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 809 | |
| 810 | return done; |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 811 | fail_unlock_page: |
| 812 | unlock_page(page); |
| 813 | fail: |
| 814 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | } |
| 816 | |
| 817 | /* |
| 818 | * Convert & write out a cluster of pages in the same extent as defined |
| 819 | * by mp and following the start page. |
| 820 | */ |
| 821 | STATIC void |
| 822 | xfs_cluster_write( |
| 823 | struct inode *inode, |
| 824 | pgoff_t tindex, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 825 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 826 | xfs_ioend_t **ioendp, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | struct writeback_control *wbc, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | pgoff_t tlast) |
| 829 | { |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 830 | struct pagevec pvec; |
| 831 | int done = 0, i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 832 | |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 833 | pagevec_init(&pvec, 0); |
| 834 | while (!done && tindex <= tlast) { |
| 835 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); |
| 836 | |
| 837 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | break; |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 839 | |
| 840 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 841 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 842 | imap, ioendp, wbc); |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 843 | if (done) |
| 844 | break; |
| 845 | } |
| 846 | |
| 847 | pagevec_release(&pvec); |
| 848 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | } |
| 850 | } |
| 851 | |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 852 | STATIC void |
| 853 | xfs_vm_invalidatepage( |
| 854 | struct page *page, |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 855 | unsigned int offset, |
| 856 | unsigned int length) |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 857 | { |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 858 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
| 859 | length); |
| 860 | block_invalidatepage(page, offset, length); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 861 | } |
| 862 | |
| 863 | /* |
| 864 | * If the page has delalloc buffers on it, we need to punch them out before we |
| 865 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the |
| 866 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read |
| 867 | * is done on that same region - the delalloc extent is returned when none is |
| 868 | * supposed to be there. |
| 869 | * |
| 870 | * We prevent this by truncating away the delalloc regions on the page before |
| 871 | * invalidating it. Because they are delalloc, we can do this without needing a |
| 872 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this |
| 873 | * truncation without a transaction as there is no space left for block |
| 874 | * reservation (typically why we see a ENOSPC in writeback). |
| 875 | * |
| 876 | * This is not a performance critical path, so for now just do the punching a |
| 877 | * buffer head at a time. |
| 878 | */ |
| 879 | STATIC void |
| 880 | xfs_aops_discard_page( |
| 881 | struct page *page) |
| 882 | { |
| 883 | struct inode *inode = page->mapping->host; |
| 884 | struct xfs_inode *ip = XFS_I(inode); |
| 885 | struct buffer_head *bh, *head; |
| 886 | loff_t offset = page_offset(page); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 887 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 888 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 889 | goto out_invalidate; |
| 890 | |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 891 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 892 | goto out_invalidate; |
| 893 | |
Dave Chinner | 4f10700 | 2011-03-07 10:00:35 +1100 | [diff] [blame] | 894 | xfs_alert(ip->i_mount, |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 895 | "page discard on page %p, inode 0x%llx, offset %llu.", |
| 896 | page, ip->i_ino, offset); |
| 897 | |
| 898 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 899 | bh = head = page_buffers(page); |
| 900 | do { |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 901 | int error; |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 902 | xfs_fileoff_t start_fsb; |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 903 | |
| 904 | if (!buffer_delay(bh)) |
| 905 | goto next_buffer; |
| 906 | |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 907 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
| 908 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 909 | if (error) { |
| 910 | /* something screwed, just bail */ |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 911 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Dave Chinner | 4f10700 | 2011-03-07 10:00:35 +1100 | [diff] [blame] | 912 | xfs_alert(ip->i_mount, |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 913 | "page discard unable to remove delalloc mapping."); |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 914 | } |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 915 | break; |
| 916 | } |
| 917 | next_buffer: |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 918 | offset += 1 << inode->i_blkbits; |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 919 | |
| 920 | } while ((bh = bh->b_this_page) != head); |
| 921 | |
| 922 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 923 | out_invalidate: |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 924 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 925 | return; |
| 926 | } |
| 927 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | /* |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 929 | * Write out a dirty page. |
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 | * For delalloc space on the page we need to allocate space and flush it. |
| 932 | * For unwritten space on the page we need to start the conversion to |
| 933 | * regular allocated space. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 934 | * 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] | 935 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 936 | STATIC int |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 937 | xfs_vm_writepage( |
| 938 | struct page *page, |
| 939 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 940 | { |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 941 | struct inode *inode = page->mapping->host; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 942 | struct buffer_head *bh, *head; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 943 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 944 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 945 | loff_t offset; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 946 | unsigned int type; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 947 | __uint64_t end_offset; |
Christoph Hellwig | bd1556a | 2010-04-28 12:29:00 +0000 | [diff] [blame] | 948 | pgoff_t end_index, last_index; |
Christoph Hellwig | ed1e7b7 | 2010-12-10 08:42:22 +0000 | [diff] [blame] | 949 | ssize_t len; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 950 | int err, imap_valid = 0, uptodate = 1; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 951 | int count = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 953 | trace_xfs_writepage(inode, page, 0, 0); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 954 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 955 | ASSERT(page_has_buffers(page)); |
| 956 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 957 | /* |
| 958 | * Refuse to write the page out if we are called from reclaim context. |
| 959 | * |
Christoph Hellwig | d4f7a5c | 2010-06-28 10:34:44 -0400 | [diff] [blame] | 960 | * This avoids stack overflows when called from deeply used stacks in |
| 961 | * random callers for direct reclaim or memcg reclaim. We explicitly |
| 962 | * allow reclaim from kswapd as the stack usage there is relatively low. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 963 | * |
Mel Gorman | 94054fa | 2011-10-31 17:07:45 -0700 | [diff] [blame] | 964 | * This should never happen except in the case of a VM regression so |
| 965 | * warn about it. |
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 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
| 968 | PF_MEMALLOC)) |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 969 | goto redirty; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 970 | |
| 971 | /* |
Christoph Hellwig | 680a647 | 2011-07-08 14:34:05 +0200 | [diff] [blame] | 972 | * Given that we do not allow direct reclaim to call us, we should |
| 973 | * never be called while in a filesystem transaction. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 974 | */ |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 975 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 976 | goto redirty; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 977 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 978 | /* Is this page beyond the end of the file? */ |
| 979 | offset = i_size_read(inode); |
| 980 | end_index = offset >> PAGE_CACHE_SHIFT; |
| 981 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 982 | |
| 983 | /* |
| 984 | * The page index is less than the end_index, adjust the end_offset |
| 985 | * to the highest offset that this page should represent. |
| 986 | * ----------------------------------------------------- |
| 987 | * | file mapping | <EOF> | |
| 988 | * ----------------------------------------------------- |
| 989 | * | Page ... | Page N-2 | Page N-1 | Page N | | |
| 990 | * ^--------------------------------^----------|-------- |
| 991 | * | desired writeback range | see else | |
| 992 | * ---------------------------------^------------------| |
| 993 | */ |
| 994 | if (page->index < end_index) |
| 995 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; |
| 996 | else { |
| 997 | /* |
| 998 | * Check whether the page to write out is beyond or straddles |
| 999 | * i_size or not. |
| 1000 | * ------------------------------------------------------- |
| 1001 | * | file mapping | <EOF> | |
| 1002 | * ------------------------------------------------------- |
| 1003 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | |
| 1004 | * ^--------------------------------^-----------|--------- |
| 1005 | * | | Straddles | |
| 1006 | * ---------------------------------^-----------|--------| |
| 1007 | */ |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1008 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
| 1009 | |
| 1010 | /* |
Jan Kara | ff9a28f | 2013-03-14 14:30:54 +0100 | [diff] [blame] | 1011 | * Skip the page if it is fully outside i_size, e.g. due to a |
| 1012 | * truncate operation that is in progress. We must redirty the |
| 1013 | * page so that reclaim stops reclaiming it. Otherwise |
| 1014 | * xfs_vm_releasepage() is called on it and gets confused. |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1015 | * |
| 1016 | * Note that the end_index is unsigned long, it would overflow |
| 1017 | * if the given offset is greater than 16TB on 32-bit system |
| 1018 | * and if we do check the page is fully outside i_size or not |
| 1019 | * via "if (page->index >= end_index + 1)" as "end_index + 1" |
| 1020 | * will be evaluated to 0. Hence this page will be redirtied |
| 1021 | * and be written out repeatedly which would result in an |
| 1022 | * infinite loop, the user program that perform this operation |
| 1023 | * will hang. Instead, we can verify this situation by checking |
| 1024 | * if the page to write is totally beyond the i_size or if it's |
| 1025 | * offset is just equal to the EOF. |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1026 | */ |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1027 | if (page->index > end_index || |
| 1028 | (page->index == end_index && offset_into_page == 0)) |
Jan Kara | ff9a28f | 2013-03-14 14:30:54 +0100 | [diff] [blame] | 1029 | goto redirty; |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1030 | |
| 1031 | /* |
| 1032 | * The page straddles i_size. It must be zeroed out on each |
| 1033 | * and every writepage invocation because it may be mmapped. |
| 1034 | * "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] | 1035 | * that is not a multiple of the page size, the remaining |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1036 | * memory is zeroed when mapped, and writes to that region are |
| 1037 | * not written out to the file." |
| 1038 | */ |
| 1039 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1040 | |
| 1041 | /* Adjust the end_offset to the end of file */ |
| 1042 | end_offset = offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1043 | } |
| 1044 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 1045 | len = 1 << inode->i_blkbits; |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 1046 | |
Nathan Scott | 24e17b5 | 2005-05-05 13:33:20 -0700 | [diff] [blame] | 1047 | bh = head = page_buffers(page); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1048 | offset = page_offset(page); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1049 | type = XFS_IO_OVERWRITE; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1050 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | do { |
Christoph Hellwig | 6ac7248 | 2010-12-10 08:42:18 +0000 | [diff] [blame] | 1052 | int new_ioend = 0; |
| 1053 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | if (offset >= end_offset) |
| 1055 | break; |
| 1056 | if (!buffer_uptodate(bh)) |
| 1057 | uptodate = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | |
Eric Sandeen | 3d9b02e | 2010-06-24 09:45:30 +1000 | [diff] [blame] | 1059 | /* |
Christoph Hellwig | ece413f | 2010-11-10 21:39:11 +0000 | [diff] [blame] | 1060 | * set_page_dirty dirties all buffers in a page, independent |
| 1061 | * of their state. The dirty state however is entirely |
| 1062 | * meaningless for holes (!mapped && uptodate), so skip |
| 1063 | * buffers covering holes here. |
Eric Sandeen | 3d9b02e | 2010-06-24 09:45:30 +1000 | [diff] [blame] | 1064 | */ |
| 1065 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { |
Eric Sandeen | 3d9b02e | 2010-06-24 09:45:30 +1000 | [diff] [blame] | 1066 | imap_valid = 0; |
| 1067 | continue; |
| 1068 | } |
| 1069 | |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1070 | if (buffer_unwritten(bh)) { |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1071 | if (type != XFS_IO_UNWRITTEN) { |
| 1072 | type = XFS_IO_UNWRITTEN; |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1073 | imap_valid = 0; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1074 | } |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1075 | } else if (buffer_delay(bh)) { |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1076 | if (type != XFS_IO_DELALLOC) { |
| 1077 | type = XFS_IO_DELALLOC; |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1078 | imap_valid = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1079 | } |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1080 | } else if (buffer_uptodate(bh)) { |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1081 | if (type != XFS_IO_OVERWRITE) { |
| 1082 | type = XFS_IO_OVERWRITE; |
Christoph Hellwig | 85da94c | 2010-12-10 08:42:16 +0000 | [diff] [blame] | 1083 | imap_valid = 0; |
| 1084 | } |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1085 | } else { |
Alain Renaud | 7d0fa3e | 2012-06-08 15:34:46 -0400 | [diff] [blame] | 1086 | if (PageUptodate(page)) |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1087 | ASSERT(buffer_mapped(bh)); |
Alain Renaud | 7d0fa3e | 2012-06-08 15:34:46 -0400 | [diff] [blame] | 1088 | /* |
| 1089 | * This buffer is not uptodate and will not be |
| 1090 | * written to disk. Ensure that we will put any |
| 1091 | * subsequent writeable buffers into a new |
| 1092 | * ioend. |
| 1093 | */ |
| 1094 | imap_valid = 0; |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1095 | continue; |
| 1096 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1098 | if (imap_valid) |
| 1099 | imap_valid = xfs_imap_valid(inode, &imap, offset); |
| 1100 | if (!imap_valid) { |
| 1101 | /* |
| 1102 | * If we didn't have a valid mapping then we need to |
| 1103 | * put the new mapping into a separate ioend structure. |
| 1104 | * This ensures non-contiguous extents always have |
| 1105 | * separate ioends, which is particularly important |
| 1106 | * for unwritten extent conversion at I/O completion |
| 1107 | * time. |
| 1108 | */ |
| 1109 | new_ioend = 1; |
Dave Chinner | 988ef92 | 2016-02-15 17:20:50 +1100 | [diff] [blame^] | 1110 | err = xfs_map_blocks(inode, offset, &imap, type); |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1111 | if (err) |
| 1112 | goto error; |
| 1113 | imap_valid = xfs_imap_valid(inode, &imap, offset); |
| 1114 | } |
| 1115 | if (imap_valid) { |
Christoph Hellwig | ecff71e | 2010-12-10 08:42:25 +0000 | [diff] [blame] | 1116 | lock_buffer(bh); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 1117 | if (type != XFS_IO_OVERWRITE) |
Christoph Hellwig | aeea1b1 | 2010-12-10 08:42:24 +0000 | [diff] [blame] | 1118 | xfs_map_at_offset(inode, bh, &imap, offset); |
| 1119 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, |
| 1120 | new_ioend); |
| 1121 | count++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1122 | } |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1123 | |
| 1124 | if (!iohead) |
| 1125 | iohead = ioend; |
| 1126 | |
| 1127 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1128 | |
| 1129 | if (uptodate && bh == head) |
| 1130 | SetPageUptodate(page); |
| 1131 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1132 | xfs_start_page_writeback(page, 1, count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 1134 | /* if there is no IO to be submitted for this page, we are done */ |
| 1135 | if (!ioend) |
| 1136 | return 0; |
| 1137 | |
| 1138 | ASSERT(iohead); |
| 1139 | |
| 1140 | /* |
| 1141 | * Any errors from this point onwards need tobe reported through the IO |
| 1142 | * completion path as we have marked the initial page as under writeback |
| 1143 | * and unlocked it. |
| 1144 | */ |
| 1145 | if (imap_valid) { |
Christoph Hellwig | bd1556a | 2010-04-28 12:29:00 +0000 | [diff] [blame] | 1146 | xfs_off_t end_index; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 1147 | |
Christoph Hellwig | bd1556a | 2010-04-28 12:29:00 +0000 | [diff] [blame] | 1148 | end_index = imap.br_startoff + imap.br_blockcount; |
| 1149 | |
| 1150 | /* to bytes */ |
| 1151 | end_index <<= inode->i_blkbits; |
| 1152 | |
| 1153 | /* to pages */ |
| 1154 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; |
| 1155 | |
| 1156 | /* check against file size */ |
| 1157 | if (end_index > last_index) |
| 1158 | end_index = last_index; |
| 1159 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1160 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
Christoph Hellwig | 2fa24f9 | 2010-12-10 08:42:23 +0000 | [diff] [blame] | 1161 | wbc, end_index); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | } |
| 1163 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 1164 | |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 1165 | /* |
| 1166 | * Reserve log space if we might write beyond the on-disk inode size. |
| 1167 | */ |
| 1168 | err = 0; |
| 1169 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) |
| 1170 | err = xfs_setfilesize_trans_alloc(ioend); |
| 1171 | |
| 1172 | xfs_submit_ioend(wbc, iohead, err); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1173 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1174 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1175 | |
| 1176 | error: |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 1177 | if (iohead) |
| 1178 | xfs_cancel_ioend(iohead); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1179 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1180 | xfs_aops_discard_page(page); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1181 | ClearPageUptodate(page); |
| 1182 | unlock_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1183 | return err; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1184 | |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 1185 | redirty: |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1186 | redirty_page_for_writepage(wbc, page); |
| 1187 | unlock_page(page); |
| 1188 | return 0; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1189 | } |
| 1190 | |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1191 | STATIC int |
| 1192 | xfs_vm_writepages( |
| 1193 | struct address_space *mapping, |
| 1194 | struct writeback_control *wbc) |
| 1195 | { |
Christoph Hellwig | b3aea4e | 2007-08-29 11:44:37 +1000 | [diff] [blame] | 1196 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1197 | return generic_writepages(mapping, wbc); |
| 1198 | } |
| 1199 | |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1200 | /* |
| 1201 | * Called to move a page into cleanable state - and from there |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1202 | * to be released. The page should already be clean. We always |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1203 | * have buffer heads in this call. |
| 1204 | * |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1205 | * Returns 1 if the page is ok to release, 0 otherwise. |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1206 | */ |
| 1207 | STATIC int |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1208 | xfs_vm_releasepage( |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1209 | struct page *page, |
| 1210 | gfp_t gfp_mask) |
| 1211 | { |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1212 | int delalloc, unwritten; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1213 | |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 1214 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1215 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1216 | xfs_count_page_state(page, &delalloc, &unwritten); |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1217 | |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 1218 | if (WARN_ON_ONCE(delalloc)) |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1219 | return 0; |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 1220 | if (WARN_ON_ONCE(unwritten)) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1221 | return 0; |
| 1222 | |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1223 | return try_to_free_buffers(page); |
| 1224 | } |
| 1225 | |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1226 | /* |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1227 | * When we map a DIO buffer, we may need to attach an ioend that describes the |
| 1228 | * type of write IO we are doing. This passes to the completion function the |
| 1229 | * operations it needs to perform. If the mapping is for an overwrite wholly |
| 1230 | * within the EOF then we don't need an ioend and so we don't allocate one. |
| 1231 | * This avoids the unnecessary overhead of allocating and freeing ioends for |
| 1232 | * workloads that don't require transactions on IO completion. |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1233 | * |
| 1234 | * If we get multiple mappings in a single IO, we might be mapping different |
| 1235 | * types. But because the direct IO can only have a single private pointer, we |
| 1236 | * need to ensure that: |
| 1237 | * |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1238 | * a) i) the ioend spans the entire region of unwritten mappings; or |
| 1239 | * ii) the ioend spans all the mappings that cross or are beyond EOF; and |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1240 | * b) if it contains unwritten extents, it is *permanently* marked as such |
| 1241 | * |
| 1242 | * We could do this by chaining ioends like buffered IO does, but we only |
| 1243 | * actually get one IO completion callback from the direct IO, and that spans |
| 1244 | * the entire IO regardless of how many mappings and IOs are needed to complete |
| 1245 | * the DIO. There is only going to be one reference to the ioend and its life |
| 1246 | * cycle is constrained by the DIO completion code. hence we don't need |
| 1247 | * reference counting here. |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1248 | * |
| 1249 | * Note that for DIO, an IO to the highest supported file block offset (i.e. |
| 1250 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 |
| 1251 | * bit variable. Hence if we see this overflow, we have to assume that the IO is |
| 1252 | * extending the file size. We won't know for sure until IO completion is run |
| 1253 | * and the actual max write offset is communicated to the IO completion |
| 1254 | * routine. |
| 1255 | * |
| 1256 | * For DAX page faults, we are preparing to never see unwritten extents here, |
| 1257 | * nor should we ever extend the inode size. Hence we will soon have nothing to |
| 1258 | * do here for this case, ensuring we don't have to provide an IO completion |
| 1259 | * callback to free an ioend that we don't actually need for a fault into the |
| 1260 | * page at offset (2^63 - 1FSB) bytes. |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1261 | */ |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1262 | |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1263 | static void |
| 1264 | xfs_map_direct( |
| 1265 | struct inode *inode, |
| 1266 | struct buffer_head *bh_result, |
| 1267 | struct xfs_bmbt_irec *imap, |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1268 | xfs_off_t offset, |
| 1269 | bool dax_fault) |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1270 | { |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1271 | struct xfs_ioend *ioend; |
| 1272 | xfs_off_t size = bh_result->b_size; |
| 1273 | int type; |
| 1274 | |
| 1275 | if (ISUNWRITTEN(imap)) |
| 1276 | type = XFS_IO_UNWRITTEN; |
| 1277 | else |
| 1278 | type = XFS_IO_OVERWRITE; |
| 1279 | |
| 1280 | trace_xfs_gbmap_direct(XFS_I(inode), offset, size, type, imap); |
| 1281 | |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1282 | if (dax_fault) { |
| 1283 | ASSERT(type == XFS_IO_OVERWRITE); |
| 1284 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, |
| 1285 | imap); |
| 1286 | return; |
| 1287 | } |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1288 | |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1289 | if (bh_result->b_private) { |
| 1290 | ioend = bh_result->b_private; |
| 1291 | ASSERT(ioend->io_size > 0); |
| 1292 | ASSERT(offset >= ioend->io_offset); |
| 1293 | if (offset + size > ioend->io_offset + ioend->io_size) |
| 1294 | ioend->io_size = offset - ioend->io_offset + size; |
| 1295 | |
| 1296 | if (type == XFS_IO_UNWRITTEN && type != ioend->io_type) |
| 1297 | ioend->io_type = XFS_IO_UNWRITTEN; |
| 1298 | |
| 1299 | trace_xfs_gbmap_direct_update(XFS_I(inode), ioend->io_offset, |
| 1300 | ioend->io_size, ioend->io_type, |
| 1301 | imap); |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1302 | } else if (type == XFS_IO_UNWRITTEN || |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1303 | offset + size > i_size_read(inode) || |
| 1304 | offset + size < 0) { |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1305 | ioend = xfs_alloc_ioend(inode, type); |
| 1306 | ioend->io_offset = offset; |
| 1307 | ioend->io_size = size; |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1308 | |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1309 | bh_result->b_private = ioend; |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1310 | set_buffer_defer_completion(bh_result); |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1311 | |
| 1312 | trace_xfs_gbmap_direct_new(XFS_I(inode), offset, size, type, |
| 1313 | imap); |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1314 | } else { |
| 1315 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, |
| 1316 | imap); |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1317 | } |
| 1318 | } |
| 1319 | |
Dave Chinner | 1fdca9c | 2015-04-16 21:58:21 +1000 | [diff] [blame] | 1320 | /* |
| 1321 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping |
| 1322 | * is, so that we can avoid repeated get_blocks calls. |
| 1323 | * |
| 1324 | * If the mapping spans EOF, then we have to break the mapping up as the mapping |
| 1325 | * for blocks beyond EOF must be marked new so that sub block regions can be |
| 1326 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping |
| 1327 | * was just allocated or is unwritten, otherwise the callers would overwrite |
| 1328 | * existing data with zeros. Hence we have to split the mapping into a range up |
| 1329 | * to and including EOF, and a second mapping for beyond EOF. |
| 1330 | */ |
| 1331 | static void |
| 1332 | xfs_map_trim_size( |
| 1333 | struct inode *inode, |
| 1334 | sector_t iblock, |
| 1335 | struct buffer_head *bh_result, |
| 1336 | struct xfs_bmbt_irec *imap, |
| 1337 | xfs_off_t offset, |
| 1338 | ssize_t size) |
| 1339 | { |
| 1340 | xfs_off_t mapping_size; |
| 1341 | |
| 1342 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; |
| 1343 | mapping_size <<= inode->i_blkbits; |
| 1344 | |
| 1345 | ASSERT(mapping_size > 0); |
| 1346 | if (mapping_size > size) |
| 1347 | mapping_size = size; |
| 1348 | if (offset < i_size_read(inode) && |
| 1349 | offset + mapping_size >= i_size_read(inode)) { |
| 1350 | /* limit mapping to block that spans EOF */ |
| 1351 | mapping_size = roundup_64(i_size_read(inode) - offset, |
| 1352 | 1 << inode->i_blkbits); |
| 1353 | } |
| 1354 | if (mapping_size > LONG_MAX) |
| 1355 | mapping_size = LONG_MAX; |
| 1356 | |
| 1357 | bh_result->b_size = mapping_size; |
| 1358 | } |
| 1359 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1360 | STATIC int |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1361 | __xfs_get_blocks( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | struct inode *inode, |
| 1363 | sector_t iblock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | struct buffer_head *bh_result, |
| 1365 | int create, |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1366 | bool direct, |
| 1367 | bool dax_fault) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1369 | struct xfs_inode *ip = XFS_I(inode); |
| 1370 | struct xfs_mount *mp = ip->i_mount; |
| 1371 | xfs_fileoff_t offset_fsb, end_fsb; |
| 1372 | int error = 0; |
| 1373 | int lockmode = 0; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1374 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1375 | int nimaps = 1; |
Nathan Scott | fdc7ed7 | 2005-11-02 15:13:13 +1100 | [diff] [blame] | 1376 | xfs_off_t offset; |
| 1377 | ssize_t size; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1378 | int new = 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1379 | |
| 1380 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 1381 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | |
Nathan Scott | fdc7ed7 | 2005-11-02 15:13:13 +1100 | [diff] [blame] | 1383 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1384 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
| 1385 | size = bh_result->b_size; |
Lachlan McIlroy | 364f358 | 2008-09-17 16:50:14 +1000 | [diff] [blame] | 1386 | |
| 1387 | if (!create && direct && offset >= i_size_read(inode)) |
| 1388 | return 0; |
| 1389 | |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1390 | /* |
| 1391 | * Direct I/O is usually done on preallocated files, so try getting |
| 1392 | * a block mapping without an exclusive lock first. For buffered |
| 1393 | * writes we already have the exclusive iolock anyway, so avoiding |
| 1394 | * a lock roundtrip here by taking the ilock exclusive from the |
| 1395 | * beginning is a useful micro optimization. |
| 1396 | */ |
| 1397 | if (create && !direct) { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1398 | lockmode = XFS_ILOCK_EXCL; |
| 1399 | xfs_ilock(ip, lockmode); |
| 1400 | } else { |
Christoph Hellwig | 309ecac8 | 2013-12-06 12:30:09 -0800 | [diff] [blame] | 1401 | lockmode = xfs_ilock_data_map_shared(ip); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1402 | } |
Christoph Hellwig | f2bde9b | 2010-06-24 11:44:35 +1000 | [diff] [blame] | 1403 | |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 1404 | ASSERT(offset <= mp->m_super->s_maxbytes); |
| 1405 | if (offset + size > mp->m_super->s_maxbytes) |
| 1406 | size = mp->m_super->s_maxbytes - offset; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1407 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
| 1408 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
| 1409 | |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 1410 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
| 1411 | &imap, &nimaps, XFS_BMAPI_ENTIRE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1412 | if (error) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1413 | goto out_unlock; |
| 1414 | |
Dave Chinner | 1ca1915 | 2015-11-03 12:37:00 +1100 | [diff] [blame] | 1415 | /* for DAX, we convert unwritten extents directly */ |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1416 | if (create && |
| 1417 | (!nimaps || |
| 1418 | (imap.br_startblock == HOLESTARTBLOCK || |
Dave Chinner | 1ca1915 | 2015-11-03 12:37:00 +1100 | [diff] [blame] | 1419 | imap.br_startblock == DELAYSTARTBLOCK) || |
| 1420 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { |
Dave Chinner | aff3a9e | 2012-04-23 15:58:44 +1000 | [diff] [blame] | 1421 | if (direct || xfs_get_extsz_hint(ip)) { |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1422 | /* |
Brian Foster | 009c6e8 | 2015-10-12 15:34:20 +1100 | [diff] [blame] | 1423 | * xfs_iomap_write_direct() expects the shared lock. It |
| 1424 | * is unlocked on return. |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1425 | */ |
Brian Foster | 009c6e8 | 2015-10-12 15:34:20 +1100 | [diff] [blame] | 1426 | if (lockmode == XFS_ILOCK_EXCL) |
| 1427 | xfs_ilock_demote(ip, lockmode); |
| 1428 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1429 | error = xfs_iomap_write_direct(ip, offset, size, |
| 1430 | &imap, nimaps); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1431 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1432 | return error; |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1433 | new = 1; |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1434 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1435 | } else { |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1436 | /* |
| 1437 | * Delalloc reservations do not require a transaction, |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1438 | * we can go on without dropping the lock here. If we |
| 1439 | * are allocating a new delalloc block, make sure that |
| 1440 | * we set the new flag so that we mark the buffer new so |
| 1441 | * that we know that it is newly allocated if the write |
| 1442 | * fails. |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1443 | */ |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1444 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
| 1445 | new = 1; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1446 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1447 | if (error) |
| 1448 | goto out_unlock; |
| 1449 | |
| 1450 | xfs_iunlock(ip, lockmode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1451 | } |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1452 | trace_xfs_get_blocks_alloc(ip, offset, size, |
| 1453 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN |
| 1454 | : XFS_IO_DELALLOC, &imap); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1455 | } else if (nimaps) { |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1456 | trace_xfs_get_blocks_found(ip, offset, size, |
| 1457 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN |
| 1458 | : XFS_IO_OVERWRITE, &imap); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1459 | xfs_iunlock(ip, lockmode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1460 | } else { |
| 1461 | trace_xfs_get_blocks_notfound(ip, offset, size); |
| 1462 | goto out_unlock; |
| 1463 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1464 | |
Dave Chinner | 1ca1915 | 2015-11-03 12:37:00 +1100 | [diff] [blame] | 1465 | if (IS_DAX(inode) && create) { |
| 1466 | ASSERT(!ISUNWRITTEN(&imap)); |
| 1467 | /* zeroing is not needed at a higher layer */ |
| 1468 | new = 0; |
| 1469 | } |
| 1470 | |
Dave Chinner | 1fdca9c | 2015-04-16 21:58:21 +1000 | [diff] [blame] | 1471 | /* trim mapping down to size requested */ |
| 1472 | if (direct || size > (1 << inode->i_blkbits)) |
| 1473 | xfs_map_trim_size(inode, iblock, bh_result, |
| 1474 | &imap, offset, size); |
| 1475 | |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1476 | /* |
| 1477 | * For unwritten extents do not report a disk address in the buffered |
| 1478 | * read case (treat as if we're reading into a hole). |
| 1479 | */ |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1480 | if (imap.br_startblock != HOLESTARTBLOCK && |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1481 | imap.br_startblock != DELAYSTARTBLOCK && |
| 1482 | (create || !ISUNWRITTEN(&imap))) { |
| 1483 | xfs_map_buffer(inode, bh_result, &imap, offset); |
| 1484 | if (ISUNWRITTEN(&imap)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | set_buffer_unwritten(bh_result); |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1486 | /* direct IO needs special help */ |
| 1487 | if (create && direct) |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1488 | xfs_map_direct(inode, bh_result, &imap, offset, |
| 1489 | dax_fault); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | } |
| 1491 | |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1492 | /* |
| 1493 | * If this is a realtime file, data may be on a different device. |
| 1494 | * to that pointed to from the buffer_head b_bdev currently. |
| 1495 | */ |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 1496 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1497 | |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1498 | /* |
David Chinner | 549054a | 2007-02-10 18:36:35 +1100 | [diff] [blame] | 1499 | * If we previously allocated a block out beyond eof and we are now |
| 1500 | * coming back to use it then we will need to flag it as new even if it |
| 1501 | * has a disk address. |
| 1502 | * |
| 1503 | * With sub-block writes into unwritten extents we also need to mark |
| 1504 | * the buffer as new so that the unwritten parts of the buffer gets |
| 1505 | * correctly zeroed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | */ |
| 1507 | if (create && |
| 1508 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || |
David Chinner | 549054a | 2007-02-10 18:36:35 +1100 | [diff] [blame] | 1509 | (offset >= i_size_read(inode)) || |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1510 | (new || ISUNWRITTEN(&imap)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | set_buffer_new(bh_result); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1513 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | BUG_ON(direct); |
| 1515 | if (create) { |
| 1516 | set_buffer_uptodate(bh_result); |
| 1517 | set_buffer_mapped(bh_result); |
| 1518 | set_buffer_delay(bh_result); |
| 1519 | } |
| 1520 | } |
| 1521 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | return 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1523 | |
| 1524 | out_unlock: |
| 1525 | xfs_iunlock(ip, lockmode); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1526 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | } |
| 1528 | |
| 1529 | int |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1530 | xfs_get_blocks( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | struct inode *inode, |
| 1532 | sector_t iblock, |
| 1533 | struct buffer_head *bh_result, |
| 1534 | int create) |
| 1535 | { |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1536 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1537 | } |
| 1538 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1539 | int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1540 | xfs_get_blocks_direct( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1541 | struct inode *inode, |
| 1542 | sector_t iblock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1543 | struct buffer_head *bh_result, |
| 1544 | int create) |
| 1545 | { |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1546 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
| 1547 | } |
| 1548 | |
| 1549 | int |
| 1550 | xfs_get_blocks_dax_fault( |
| 1551 | struct inode *inode, |
| 1552 | sector_t iblock, |
| 1553 | struct buffer_head *bh_result, |
| 1554 | int create) |
| 1555 | { |
| 1556 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | } |
| 1558 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1559 | static void |
| 1560 | __xfs_end_io_direct_write( |
| 1561 | struct inode *inode, |
| 1562 | struct xfs_ioend *ioend, |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1563 | loff_t offset, |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1564 | ssize_t size) |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1565 | { |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1566 | struct xfs_mount *mp = XFS_I(inode)->i_mount; |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 1567 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1568 | if (XFS_FORCED_SHUTDOWN(mp) || ioend->io_error) |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1569 | goto out_end_io; |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1570 | |
| 1571 | /* |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1572 | * dio completion end_io functions are only called on writes if more |
| 1573 | * than 0 bytes was written. |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 1574 | */ |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1575 | ASSERT(size > 0); |
| 1576 | |
| 1577 | /* |
| 1578 | * The ioend only maps whole blocks, while the IO may be sector aligned. |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1579 | * Hence the ioend offset/size may not match the IO offset/size exactly. |
| 1580 | * Because we don't map overwrites within EOF into the ioend, the offset |
| 1581 | * may not match, but only if the endio spans EOF. Either way, write |
| 1582 | * the IO sizes into the ioend so that completion processing does the |
| 1583 | * right thing. |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1584 | */ |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1585 | ASSERT(offset + size <= ioend->io_offset + ioend->io_size); |
| 1586 | ioend->io_size = size; |
| 1587 | ioend->io_offset = offset; |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1588 | |
| 1589 | /* |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1590 | * The ioend tells us whether we are doing unwritten extent conversion |
| 1591 | * or an append transaction that updates the on-disk file size. These |
| 1592 | * cases are the only cases where we should *potentially* be needing |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1593 | * to update the VFS inode size. |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1594 | * |
| 1595 | * We need to update the in-core inode size here so that we don't end up |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1596 | * with the on-disk inode size being outside the in-core inode size. We |
| 1597 | * have no other method of updating EOF for AIO, so always do it here |
| 1598 | * if necessary. |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 1599 | * |
| 1600 | * We need to lock the test/set EOF update as we can be racing with |
| 1601 | * other IO completions here to update the EOF. Failing to serialise |
| 1602 | * here can result in EOF moving backwards and Bad Things Happen when |
| 1603 | * that occurs. |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 1604 | */ |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1605 | spin_lock(&XFS_I(inode)->i_flags_lock); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 1606 | if (offset + size > i_size_read(inode)) |
| 1607 | i_size_write(inode, offset + size); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1608 | spin_unlock(&XFS_I(inode)->i_flags_lock); |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 1609 | |
| 1610 | /* |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1611 | * If we are doing an append IO that needs to update the EOF on disk, |
| 1612 | * do the transaction reserve now so we can use common end io |
| 1613 | * processing. Stashing the error (if there is one) in the ioend will |
| 1614 | * result in the ioend processing passing on the error if it is |
| 1615 | * possible as we can't return it from here. |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1616 | */ |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1617 | if (ioend->io_type == XFS_IO_OVERWRITE) |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1618 | ioend->io_error = xfs_setfilesize_trans_alloc(ioend); |
Christoph Hellwig | 40e2e97 | 2010-07-18 21:17:09 +0000 | [diff] [blame] | 1619 | |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1620 | out_end_io: |
| 1621 | xfs_end_io(&ioend->io_work); |
| 1622 | return; |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1623 | } |
| 1624 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1625 | /* |
| 1626 | * Complete a direct I/O write request. |
| 1627 | * |
| 1628 | * The ioend structure is passed from __xfs_get_blocks() to tell us what to do. |
| 1629 | * If no ioend exists (i.e. @private == NULL) then the write IO is an overwrite |
| 1630 | * wholly within the EOF and so there is nothing for us to do. Note that in this |
| 1631 | * case the completion can be called in interrupt context, whereas if we have an |
| 1632 | * ioend we will always be called in task context (i.e. from a workqueue). |
| 1633 | */ |
| 1634 | STATIC void |
| 1635 | xfs_end_io_direct_write( |
| 1636 | struct kiocb *iocb, |
| 1637 | loff_t offset, |
| 1638 | ssize_t size, |
| 1639 | void *private) |
| 1640 | { |
| 1641 | struct inode *inode = file_inode(iocb->ki_filp); |
| 1642 | struct xfs_ioend *ioend = private; |
| 1643 | |
| 1644 | trace_xfs_gbmap_direct_endio(XFS_I(inode), offset, size, |
| 1645 | ioend ? ioend->io_type : 0, NULL); |
| 1646 | |
| 1647 | if (!ioend) { |
| 1648 | ASSERT(offset + size <= i_size_read(inode)); |
| 1649 | return; |
| 1650 | } |
| 1651 | |
| 1652 | __xfs_end_io_direct_write(inode, ioend, offset, size); |
| 1653 | } |
| 1654 | |
Dave Chinner | 6e1ba0b | 2015-06-04 09:19:15 +1000 | [diff] [blame] | 1655 | static inline ssize_t |
| 1656 | xfs_vm_do_dio( |
| 1657 | struct inode *inode, |
| 1658 | struct kiocb *iocb, |
| 1659 | struct iov_iter *iter, |
| 1660 | loff_t offset, |
| 1661 | void (*endio)(struct kiocb *iocb, |
| 1662 | loff_t offset, |
| 1663 | ssize_t size, |
| 1664 | void *private), |
| 1665 | int flags) |
| 1666 | { |
| 1667 | struct block_device *bdev; |
| 1668 | |
| 1669 | if (IS_DAX(inode)) |
| 1670 | return dax_do_io(iocb, inode, iter, offset, |
| 1671 | xfs_get_blocks_direct, endio, 0); |
| 1672 | |
| 1673 | bdev = xfs_find_bdev_for_inode(inode); |
| 1674 | return __blockdev_direct_IO(iocb, inode, bdev, iter, offset, |
| 1675 | xfs_get_blocks_direct, endio, NULL, flags); |
| 1676 | } |
| 1677 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1678 | STATIC ssize_t |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1679 | xfs_vm_direct_IO( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | struct kiocb *iocb, |
Al Viro | d8d3d94 | 2014-03-04 21:27:34 -0500 | [diff] [blame] | 1681 | struct iov_iter *iter, |
| 1682 | loff_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1683 | { |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1684 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1685 | |
Dave Chinner | 6e1ba0b | 2015-06-04 09:19:15 +1000 | [diff] [blame] | 1686 | if (iov_iter_rw(iter) == WRITE) |
| 1687 | return xfs_vm_do_dio(inode, iocb, iter, offset, |
| 1688 | xfs_end_io_direct_write, DIO_ASYNC_EXTEND); |
| 1689 | return xfs_vm_do_dio(inode, iocb, iter, offset, NULL, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1690 | } |
| 1691 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1692 | /* |
| 1693 | * Punch out the delalloc blocks we have already allocated. |
| 1694 | * |
| 1695 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet |
| 1696 | * as the page is still locked at this point. |
| 1697 | */ |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1698 | STATIC void |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1699 | xfs_vm_kill_delalloc_range( |
| 1700 | struct inode *inode, |
| 1701 | loff_t start, |
| 1702 | loff_t end) |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1703 | { |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1704 | struct xfs_inode *ip = XFS_I(inode); |
| 1705 | xfs_fileoff_t start_fsb; |
| 1706 | xfs_fileoff_t end_fsb; |
| 1707 | int error; |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1708 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1709 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); |
| 1710 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); |
| 1711 | if (end_fsb <= start_fsb) |
| 1712 | return; |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 1713 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1714 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 1715 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, |
| 1716 | end_fsb - start_fsb); |
| 1717 | if (error) { |
| 1718 | /* something screwed, just bail */ |
| 1719 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
| 1720 | xfs_alert(ip->i_mount, |
| 1721 | "xfs_vm_write_failed: unable to clean up ino %lld", |
| 1722 | ip->i_ino); |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 1723 | } |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1724 | } |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1725 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1726 | } |
| 1727 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1728 | STATIC void |
| 1729 | xfs_vm_write_failed( |
| 1730 | struct inode *inode, |
| 1731 | struct page *page, |
| 1732 | loff_t pos, |
| 1733 | unsigned len) |
| 1734 | { |
Jie Liu | 58e5985 | 2013-07-16 13:11:16 +0800 | [diff] [blame] | 1735 | loff_t block_offset; |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1736 | loff_t block_start; |
| 1737 | loff_t block_end; |
| 1738 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); |
| 1739 | loff_t to = from + len; |
| 1740 | struct buffer_head *bh, *head; |
| 1741 | |
Jie Liu | 58e5985 | 2013-07-16 13:11:16 +0800 | [diff] [blame] | 1742 | /* |
| 1743 | * The request pos offset might be 32 or 64 bit, this is all fine |
| 1744 | * on 64-bit platform. However, for 64-bit pos request on 32-bit |
| 1745 | * platform, the high 32-bit will be masked off if we evaluate the |
| 1746 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is |
| 1747 | * 0xfffff000 as an unsigned long, hence the result is incorrect |
| 1748 | * which could cause the following ASSERT failed in most cases. |
| 1749 | * In order to avoid this, we can evaluate the block_offset of the |
| 1750 | * start of the page by using shifts rather than masks the mismatch |
| 1751 | * problem. |
| 1752 | */ |
| 1753 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; |
| 1754 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1755 | ASSERT(block_offset + from == pos); |
| 1756 | |
| 1757 | head = page_buffers(page); |
| 1758 | block_start = 0; |
| 1759 | for (bh = head; bh != head || !block_start; |
| 1760 | bh = bh->b_this_page, block_start = block_end, |
| 1761 | block_offset += bh->b_size) { |
| 1762 | block_end = block_start + bh->b_size; |
| 1763 | |
| 1764 | /* skip buffers before the write */ |
| 1765 | if (block_end <= from) |
| 1766 | continue; |
| 1767 | |
| 1768 | /* if the buffer is after the write, we're done */ |
| 1769 | if (block_start >= to) |
| 1770 | break; |
| 1771 | |
| 1772 | if (!buffer_delay(bh)) |
| 1773 | continue; |
| 1774 | |
| 1775 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) |
| 1776 | continue; |
| 1777 | |
| 1778 | xfs_vm_kill_delalloc_range(inode, block_offset, |
| 1779 | block_offset + bh->b_size); |
Dave Chinner | 4ab9ed5 | 2014-04-14 18:11:58 +1000 | [diff] [blame] | 1780 | |
| 1781 | /* |
| 1782 | * This buffer does not contain data anymore. make sure anyone |
| 1783 | * who finds it knows that for certain. |
| 1784 | */ |
| 1785 | clear_buffer_delay(bh); |
| 1786 | clear_buffer_uptodate(bh); |
| 1787 | clear_buffer_mapped(bh); |
| 1788 | clear_buffer_new(bh); |
| 1789 | clear_buffer_dirty(bh); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1790 | } |
| 1791 | |
| 1792 | } |
| 1793 | |
| 1794 | /* |
| 1795 | * This used to call block_write_begin(), but it unlocks and releases the page |
| 1796 | * on error, and we need that page to be able to punch stale delalloc blocks out |
| 1797 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at |
| 1798 | * the appropriate point. |
| 1799 | */ |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1800 | STATIC int |
Nick Piggin | d79689c | 2007-10-16 01:25:06 -0700 | [diff] [blame] | 1801 | xfs_vm_write_begin( |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1802 | struct file *file, |
Nick Piggin | d79689c | 2007-10-16 01:25:06 -0700 | [diff] [blame] | 1803 | struct address_space *mapping, |
| 1804 | loff_t pos, |
| 1805 | unsigned len, |
| 1806 | unsigned flags, |
| 1807 | struct page **pagep, |
| 1808 | void **fsdata) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1809 | { |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1810 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
| 1811 | struct page *page; |
| 1812 | int status; |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 1813 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1814 | ASSERT(len <= PAGE_CACHE_SIZE); |
| 1815 | |
Dave Chinner | ad22c7a | 2013-10-29 22:11:57 +1100 | [diff] [blame] | 1816 | page = grab_cache_page_write_begin(mapping, index, flags); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1817 | if (!page) |
| 1818 | return -ENOMEM; |
| 1819 | |
| 1820 | status = __block_write_begin(page, pos, len, xfs_get_blocks); |
| 1821 | if (unlikely(status)) { |
| 1822 | struct inode *inode = mapping->host; |
Dave Chinner | 72ab70a | 2014-04-14 18:13:29 +1000 | [diff] [blame] | 1823 | size_t isize = i_size_read(inode); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1824 | |
| 1825 | xfs_vm_write_failed(inode, page, pos, len); |
| 1826 | unlock_page(page); |
| 1827 | |
Dave Chinner | 72ab70a | 2014-04-14 18:13:29 +1000 | [diff] [blame] | 1828 | /* |
| 1829 | * If the write is beyond EOF, we only want to kill blocks |
| 1830 | * allocated in this write, not blocks that were previously |
| 1831 | * written successfully. |
| 1832 | */ |
| 1833 | if (pos + len > isize) { |
| 1834 | ssize_t start = max_t(ssize_t, pos, isize); |
| 1835 | |
| 1836 | truncate_pagecache_range(inode, start, pos + len); |
| 1837 | } |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1838 | |
| 1839 | page_cache_release(page); |
| 1840 | page = NULL; |
| 1841 | } |
| 1842 | |
| 1843 | *pagep = page; |
| 1844 | return status; |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1845 | } |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 1846 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1847 | /* |
Dave Chinner | aad3f37 | 2014-04-14 18:14:11 +1000 | [diff] [blame] | 1848 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
| 1849 | * this specific write because they will never be written. Previous writes |
| 1850 | * beyond EOF where block allocation succeeded do not need to be trashed, so |
| 1851 | * only new blocks from this write should be trashed. For blocks within |
| 1852 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be |
| 1853 | * written with all the other valid data. |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1854 | */ |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1855 | STATIC int |
| 1856 | xfs_vm_write_end( |
| 1857 | struct file *file, |
| 1858 | struct address_space *mapping, |
| 1859 | loff_t pos, |
| 1860 | unsigned len, |
| 1861 | unsigned copied, |
| 1862 | struct page *page, |
| 1863 | void *fsdata) |
| 1864 | { |
| 1865 | int ret; |
| 1866 | |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1867 | ASSERT(len <= PAGE_CACHE_SIZE); |
| 1868 | |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1869 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1870 | if (unlikely(ret < len)) { |
| 1871 | struct inode *inode = mapping->host; |
| 1872 | size_t isize = i_size_read(inode); |
| 1873 | loff_t to = pos + len; |
| 1874 | |
| 1875 | if (to > isize) { |
Dave Chinner | aad3f37 | 2014-04-14 18:14:11 +1000 | [diff] [blame] | 1876 | /* only kill blocks in this write beyond EOF */ |
| 1877 | if (pos > isize) |
| 1878 | isize = pos; |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1879 | xfs_vm_kill_delalloc_range(inode, isize, to); |
Dave Chinner | aad3f37 | 2014-04-14 18:14:11 +1000 | [diff] [blame] | 1880 | truncate_pagecache_range(inode, isize, to); |
Dave Chinner | d3bc815 | 2012-04-27 19:45:21 +1000 | [diff] [blame] | 1881 | } |
| 1882 | } |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 1883 | return ret; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1884 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1885 | |
| 1886 | STATIC sector_t |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1887 | xfs_vm_bmap( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1888 | struct address_space *mapping, |
| 1889 | sector_t block) |
| 1890 | { |
| 1891 | struct inode *inode = (struct inode *)mapping->host; |
Christoph Hellwig | 739bfb2 | 2007-08-29 10:58:01 +1000 | [diff] [blame] | 1892 | struct xfs_inode *ip = XFS_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1893 | |
Christoph Hellwig | cca28fb | 2010-06-24 11:57:09 +1000 | [diff] [blame] | 1894 | trace_xfs_vm_bmap(XFS_I(inode)); |
Christoph Hellwig | 126468b | 2008-03-06 13:44:57 +1100 | [diff] [blame] | 1895 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
Dave Chinner | 4bc1ea6 | 2012-11-12 22:53:56 +1100 | [diff] [blame] | 1896 | filemap_write_and_wait(mapping); |
Christoph Hellwig | 126468b | 2008-03-06 13:44:57 +1100 | [diff] [blame] | 1897 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1898 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1899 | } |
| 1900 | |
| 1901 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1902 | xfs_vm_readpage( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1903 | struct file *unused, |
| 1904 | struct page *page) |
| 1905 | { |
Dave Chinner | 121e213 | 2016-01-08 11:28:35 +1100 | [diff] [blame] | 1906 | trace_xfs_vm_readpage(page->mapping->host, 1); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1907 | return mpage_readpage(page, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1908 | } |
| 1909 | |
| 1910 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1911 | xfs_vm_readpages( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1912 | struct file *unused, |
| 1913 | struct address_space *mapping, |
| 1914 | struct list_head *pages, |
| 1915 | unsigned nr_pages) |
| 1916 | { |
Dave Chinner | 121e213 | 2016-01-08 11:28:35 +1100 | [diff] [blame] | 1917 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1918 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | } |
| 1920 | |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1921 | /* |
| 1922 | * This is basically a copy of __set_page_dirty_buffers() with one |
| 1923 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them |
| 1924 | * dirty, we'll never be able to clean them because we don't write buffers |
| 1925 | * beyond EOF, and that means we can't invalidate pages that span EOF |
| 1926 | * that have been marked dirty. Further, the dirty state can leak into |
| 1927 | * the file interior if the file is extended, resulting in all sorts of |
| 1928 | * bad things happening as the state does not match the underlying data. |
| 1929 | * |
| 1930 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like |
| 1931 | * this only exist because of bufferheads and how the generic code manages them. |
| 1932 | */ |
| 1933 | STATIC int |
| 1934 | xfs_vm_set_page_dirty( |
| 1935 | struct page *page) |
| 1936 | { |
| 1937 | struct address_space *mapping = page->mapping; |
| 1938 | struct inode *inode = mapping->host; |
| 1939 | loff_t end_offset; |
| 1940 | loff_t offset; |
| 1941 | int newly_dirty; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1942 | struct mem_cgroup *memcg; |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1943 | |
| 1944 | if (unlikely(!mapping)) |
| 1945 | return !TestSetPageDirty(page); |
| 1946 | |
| 1947 | end_offset = i_size_read(inode); |
| 1948 | offset = page_offset(page); |
| 1949 | |
| 1950 | spin_lock(&mapping->private_lock); |
| 1951 | if (page_has_buffers(page)) { |
| 1952 | struct buffer_head *head = page_buffers(page); |
| 1953 | struct buffer_head *bh = head; |
| 1954 | |
| 1955 | do { |
| 1956 | if (offset < end_offset) |
| 1957 | set_buffer_dirty(bh); |
| 1958 | bh = bh->b_this_page; |
| 1959 | offset += 1 << inode->i_blkbits; |
| 1960 | } while (bh != head); |
| 1961 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1962 | /* |
| 1963 | * Use mem_group_begin_page_stat() to keep PageDirty synchronized with |
| 1964 | * per-memcg dirty page counters. |
| 1965 | */ |
| 1966 | memcg = mem_cgroup_begin_page_stat(page); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1967 | newly_dirty = !TestSetPageDirty(page); |
| 1968 | spin_unlock(&mapping->private_lock); |
| 1969 | |
| 1970 | if (newly_dirty) { |
| 1971 | /* sigh - __set_page_dirty() is static, so copy it here, too */ |
| 1972 | unsigned long flags; |
| 1973 | |
| 1974 | spin_lock_irqsave(&mapping->tree_lock, flags); |
| 1975 | if (page->mapping) { /* Race with truncate? */ |
| 1976 | WARN_ON_ONCE(!PageUptodate(page)); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1977 | account_page_dirtied(page, mapping, memcg); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1978 | radix_tree_tag_set(&mapping->page_tree, |
| 1979 | page_index(page), PAGECACHE_TAG_DIRTY); |
| 1980 | } |
| 1981 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1982 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1983 | mem_cgroup_end_page_stat(memcg); |
| 1984 | if (newly_dirty) |
| 1985 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1986 | return newly_dirty; |
| 1987 | } |
| 1988 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1989 | const struct address_space_operations xfs_address_space_operations = { |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1990 | .readpage = xfs_vm_readpage, |
| 1991 | .readpages = xfs_vm_readpages, |
| 1992 | .writepage = xfs_vm_writepage, |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1993 | .writepages = xfs_vm_writepages, |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1994 | .set_page_dirty = xfs_vm_set_page_dirty, |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1995 | .releasepage = xfs_vm_releasepage, |
| 1996 | .invalidatepage = xfs_vm_invalidatepage, |
Nick Piggin | d79689c | 2007-10-16 01:25:06 -0700 | [diff] [blame] | 1997 | .write_begin = xfs_vm_write_begin, |
Christoph Hellwig | fa9b227 | 2010-06-14 05:17:31 -0400 | [diff] [blame] | 1998 | .write_end = xfs_vm_write_end, |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1999 | .bmap = xfs_vm_bmap, |
| 2000 | .direct_IO = xfs_vm_direct_IO, |
Christoph Lameter | e965f96 | 2006-02-01 03:05:41 -0800 | [diff] [blame] | 2001 | .migratepage = buffer_migrate_page, |
Hisashi Hifumi | bddaafa | 2009-03-29 09:53:38 +0200 | [diff] [blame] | 2002 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 2003 | .error_remove_page = generic_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2004 | }; |