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 | 70a9883c | 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" |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 34 | #include "xfs_reflink.h" |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 35 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include <linux/mpage.h> |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 37 | #include <linux/pagevec.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include <linux/writeback.h> |
| 39 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 40 | /* flags for direct write completions */ |
| 41 | #define XFS_DIO_FLAG_UNWRITTEN (1 << 0) |
| 42 | #define XFS_DIO_FLAG_APPEND (1 << 1) |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 43 | #define XFS_DIO_FLAG_COW (1 << 2) |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 44 | |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 45 | /* |
| 46 | * structure owned by writepages passed to individual writepage calls |
| 47 | */ |
| 48 | struct xfs_writepage_ctx { |
| 49 | struct xfs_bmbt_irec imap; |
| 50 | bool imap_valid; |
| 51 | unsigned int io_type; |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 52 | struct xfs_ioend *ioend; |
| 53 | sector_t last_block; |
| 54 | }; |
| 55 | |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 56 | void |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 57 | xfs_count_page_state( |
| 58 | struct page *page, |
| 59 | int *delalloc, |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 60 | int *unwritten) |
| 61 | { |
| 62 | struct buffer_head *bh, *head; |
| 63 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 64 | *delalloc = *unwritten = 0; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 65 | |
| 66 | bh = head = page_buffers(page); |
| 67 | do { |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 68 | if (buffer_unwritten(bh)) |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 69 | (*unwritten) = 1; |
| 70 | else if (buffer_delay(bh)) |
| 71 | (*delalloc) = 1; |
| 72 | } while ((bh = bh->b_this_page) != head); |
| 73 | } |
| 74 | |
Ross Zwisler | 20a90f5 | 2016-02-26 15:19:52 -0800 | [diff] [blame] | 75 | struct block_device * |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 76 | xfs_find_bdev_for_inode( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 77 | struct inode *inode) |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 78 | { |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 79 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 80 | struct xfs_mount *mp = ip->i_mount; |
| 81 | |
Eric Sandeen | 71ddabb | 2007-11-23 16:29:42 +1100 | [diff] [blame] | 82 | if (XFS_IS_REALTIME_INODE(ip)) |
Christoph Hellwig | 6214ed4 | 2007-09-14 15:23:17 +1000 | [diff] [blame] | 83 | return mp->m_rtdev_targp->bt_bdev; |
| 84 | else |
| 85 | return mp->m_ddev_targp->bt_bdev; |
| 86 | } |
| 87 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 88 | /* |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 89 | * We're now finished for good with this page. Update the page state via the |
| 90 | * associated buffer_heads, paying attention to the start and end offsets that |
| 91 | * we need to process on the page. |
Dave Chinner | 28b783e | 2016-07-22 09:56:38 +1000 | [diff] [blame] | 92 | * |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 93 | * Note that we open code the action in end_buffer_async_write here so that we |
| 94 | * only have to iterate over the buffers attached to the page once. This is not |
| 95 | * only more efficient, but also ensures that we only calls end_page_writeback |
| 96 | * at the end of the iteration, and thus avoids the pitfall of having the page |
| 97 | * and buffers potentially freed after every call to end_buffer_async_write. |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 98 | */ |
| 99 | static void |
| 100 | xfs_finish_page_writeback( |
| 101 | struct inode *inode, |
| 102 | struct bio_vec *bvec, |
| 103 | int error) |
| 104 | { |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 105 | struct buffer_head *head = page_buffers(bvec->bv_page), *bh = head; |
| 106 | bool busy = false; |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 107 | unsigned int off = 0; |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 108 | unsigned long flags; |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 109 | |
| 110 | ASSERT(bvec->bv_offset < PAGE_SIZE); |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 111 | ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0); |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 112 | ASSERT(bvec->bv_offset + bvec->bv_len <= PAGE_SIZE); |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 113 | ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0); |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 114 | |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 115 | local_irq_save(flags); |
| 116 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 117 | do { |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 118 | if (off >= bvec->bv_offset && |
| 119 | off < bvec->bv_offset + bvec->bv_len) { |
| 120 | ASSERT(buffer_async_write(bh)); |
| 121 | ASSERT(bh->b_end_io == NULL); |
| 122 | |
| 123 | if (error) { |
| 124 | mapping_set_error(bvec->bv_page->mapping, -EIO); |
| 125 | set_buffer_write_io_error(bh); |
| 126 | clear_buffer_uptodate(bh); |
| 127 | SetPageError(bvec->bv_page); |
| 128 | } else { |
| 129 | set_buffer_uptodate(bh); |
| 130 | } |
| 131 | clear_buffer_async_write(bh); |
| 132 | unlock_buffer(bh); |
| 133 | } else if (buffer_async_write(bh)) { |
| 134 | ASSERT(buffer_locked(bh)); |
| 135 | busy = true; |
| 136 | } |
| 137 | off += bh->b_size; |
| 138 | } while ((bh = bh->b_this_page) != head); |
| 139 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); |
| 140 | local_irq_restore(flags); |
| 141 | |
| 142 | if (!busy) |
| 143 | end_page_writeback(bvec->bv_page); |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 144 | } |
| 145 | |
| 146 | /* |
| 147 | * We're now finished for good with this ioend structure. Update the page |
| 148 | * state, release holds on bios, and finally free up memory. Do not use the |
| 149 | * ioend after this. |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 150 | */ |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 151 | STATIC void |
| 152 | xfs_destroy_ioend( |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 153 | struct xfs_ioend *ioend, |
| 154 | int error) |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 155 | { |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 156 | struct inode *inode = ioend->io_inode; |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 157 | struct bio *bio = &ioend->io_inline_bio; |
| 158 | struct bio *last = ioend->io_bio, *next; |
| 159 | u64 start = bio->bi_iter.bi_sector; |
| 160 | bool quiet = bio_flagged(bio, BIO_QUIET); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 161 | |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 162 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 163 | struct bio_vec *bvec; |
| 164 | int i; |
| 165 | |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 166 | /* |
| 167 | * For the last bio, bi_private points to the ioend, so we |
| 168 | * need to explicitly end the iteration here. |
| 169 | */ |
| 170 | if (bio == last) |
| 171 | next = NULL; |
| 172 | else |
| 173 | next = bio->bi_private; |
Dave Chinner | 37992c1 | 2016-04-06 08:12:28 +1000 | [diff] [blame] | 174 | |
| 175 | /* walk each page on bio, ending page IO on them */ |
| 176 | bio_for_each_segment_all(bvec, bio, i) |
| 177 | xfs_finish_page_writeback(inode, bvec, error); |
| 178 | |
| 179 | bio_put(bio); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 180 | } |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 181 | |
| 182 | if (unlikely(error && !quiet)) { |
| 183 | xfs_err_ratelimited(XFS_I(inode)->i_mount, |
| 184 | "writeback error on sector %llu", start); |
| 185 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | } |
| 187 | |
| 188 | /* |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 189 | * Fast and loose check if this write could update the on-disk inode size. |
| 190 | */ |
| 191 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) |
| 192 | { |
| 193 | return ioend->io_offset + ioend->io_size > |
| 194 | XFS_I(ioend->io_inode)->i_d.di_size; |
| 195 | } |
| 196 | |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 197 | STATIC int |
| 198 | xfs_setfilesize_trans_alloc( |
| 199 | struct xfs_ioend *ioend) |
| 200 | { |
| 201 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
| 202 | struct xfs_trans *tp; |
| 203 | int error; |
| 204 | |
Christoph Hellwig | 253f491 | 2016-04-06 09:19:55 +1000 | [diff] [blame] | 205 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
| 206 | if (error) |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 207 | return error; |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 208 | |
| 209 | ioend->io_append_trans = tp; |
| 210 | |
| 211 | /* |
Dave Chinner | 437a255 | 2012-11-28 13:01:00 +1100 | [diff] [blame] | 212 | * We may pass freeze protection with a transaction. So tell lockdep |
Jan Kara | d9457dc | 2012-06-12 16:20:39 +0200 | [diff] [blame] | 213 | * we released it. |
| 214 | */ |
Oleg Nesterov | bee9182 | 2015-07-19 23:48:20 +0200 | [diff] [blame] | 215 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
Jan Kara | d9457dc | 2012-06-12 16:20:39 +0200 | [diff] [blame] | 216 | /* |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 217 | * We hand off the transaction to the completion thread now, so |
| 218 | * clear the flag here. |
| 219 | */ |
| 220 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); |
| 221 | return 0; |
| 222 | } |
| 223 | |
Christoph Hellwig | fc0063c | 2011-08-23 08:28:11 +0000 | [diff] [blame] | 224 | /* |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 225 | * 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] | 226 | */ |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 227 | STATIC int |
Christoph Hellwig | e372843 | 2016-09-19 11:26:41 +1000 | [diff] [blame] | 228 | __xfs_setfilesize( |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 229 | struct xfs_inode *ip, |
| 230 | struct xfs_trans *tp, |
| 231 | xfs_off_t offset, |
| 232 | size_t size) |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 233 | { |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 234 | xfs_fsize_t isize; |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 235 | |
Christoph Hellwig | aa6bf01 | 2012-02-29 09:53:48 +0000 | [diff] [blame] | 236 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 237 | isize = xfs_new_eof(ip, offset + size); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 238 | if (!isize) { |
| 239 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Christoph Hellwig | 4906e21 | 2015-06-04 13:47:56 +1000 | [diff] [blame] | 240 | xfs_trans_cancel(tp); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 241 | return 0; |
Lachlan McIlroy | ba87ea6 | 2007-05-08 13:49:46 +1000 | [diff] [blame] | 242 | } |
| 243 | |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 244 | trace_xfs_setfilesize(ip, offset, size); |
Christoph Hellwig | 281627d | 2012-03-13 08:41:05 +0000 | [diff] [blame] | 245 | |
| 246 | ip->i_d.di_size = isize; |
| 247 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
| 248 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 249 | |
Christoph Hellwig | 7039331 | 2015-06-04 13:48:08 +1000 | [diff] [blame] | 250 | return xfs_trans_commit(tp); |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 251 | } |
| 252 | |
Christoph Hellwig | e372843 | 2016-09-19 11:26:41 +1000 | [diff] [blame] | 253 | int |
| 254 | xfs_setfilesize( |
| 255 | struct xfs_inode *ip, |
| 256 | xfs_off_t offset, |
| 257 | size_t size) |
| 258 | { |
| 259 | struct xfs_mount *mp = ip->i_mount; |
| 260 | struct xfs_trans *tp; |
| 261 | int error; |
| 262 | |
| 263 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
| 264 | if (error) |
| 265 | return error; |
| 266 | |
| 267 | return __xfs_setfilesize(ip, tp, offset, size); |
| 268 | } |
| 269 | |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 270 | STATIC int |
| 271 | xfs_setfilesize_ioend( |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 272 | struct xfs_ioend *ioend, |
| 273 | int error) |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 274 | { |
| 275 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
| 276 | struct xfs_trans *tp = ioend->io_append_trans; |
| 277 | |
| 278 | /* |
| 279 | * The transaction may have been allocated in the I/O submission thread, |
| 280 | * thus we need to mark ourselves as being in a transaction manually. |
| 281 | * Similarly for freeze protection. |
| 282 | */ |
| 283 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); |
Oleg Nesterov | bee9182 | 2015-07-19 23:48:20 +0200 | [diff] [blame] | 284 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 285 | |
Zhaohongjiang | 5cb13dc | 2015-10-12 15:28:39 +1100 | [diff] [blame] | 286 | /* we abort the update if there was an IO error */ |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 287 | if (error) { |
Zhaohongjiang | 5cb13dc | 2015-10-12 15:28:39 +1100 | [diff] [blame] | 288 | xfs_trans_cancel(tp); |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 289 | return error; |
Zhaohongjiang | 5cb13dc | 2015-10-12 15:28:39 +1100 | [diff] [blame] | 290 | } |
| 291 | |
Christoph Hellwig | e372843 | 2016-09-19 11:26:41 +1000 | [diff] [blame] | 292 | return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 293 | } |
| 294 | |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 295 | /* |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 296 | * IO write completion. |
| 297 | */ |
| 298 | STATIC void |
| 299 | xfs_end_io( |
| 300 | struct work_struct *work) |
| 301 | { |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 302 | struct xfs_ioend *ioend = |
| 303 | container_of(work, struct xfs_ioend, io_work); |
| 304 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 305 | xfs_off_t offset = ioend->io_offset; |
| 306 | size_t size = ioend->io_size; |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 307 | int error = ioend->io_bio->bi_error; |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 308 | |
Brian Foster | af055e3 | 2016-02-08 15:00:02 +1100 | [diff] [blame] | 309 | /* |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 310 | * Just clean up the in-memory strutures if the fs has been shut down. |
Brian Foster | af055e3 | 2016-02-08 15:00:02 +1100 | [diff] [blame] | 311 | */ |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 312 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 313 | error = -EIO; |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 314 | goto done; |
Darrick J. Wong | 43caeb1 | 2016-10-03 09:11:35 -0700 | [diff] [blame] | 315 | } |
| 316 | |
| 317 | /* |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 318 | * Clean up any COW blocks on an I/O error. |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 319 | */ |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 320 | if (unlikely(error)) { |
| 321 | switch (ioend->io_type) { |
| 322 | case XFS_IO_COW: |
| 323 | xfs_reflink_cancel_cow_range(ip, offset, size, true); |
| 324 | break; |
| 325 | } |
| 326 | |
| 327 | goto done; |
| 328 | } |
| 329 | |
| 330 | /* |
| 331 | * Success: commit the COW or unwritten blocks if needed. |
| 332 | */ |
| 333 | switch (ioend->io_type) { |
| 334 | case XFS_IO_COW: |
| 335 | error = xfs_reflink_end_cow(ip, offset, size); |
| 336 | break; |
| 337 | case XFS_IO_UNWRITTEN: |
| 338 | error = xfs_iomap_write_unwritten(ip, offset, size); |
| 339 | break; |
| 340 | default: |
| 341 | ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans); |
| 342 | break; |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 343 | } |
| 344 | |
Christoph Hellwig | 04f658e | 2011-08-24 05:59:25 +0000 | [diff] [blame] | 345 | done: |
Christoph Hellwig | d07b585 | 2017-03-02 15:02:51 -0800 | [diff] [blame] | 346 | if (ioend->io_append_trans) |
| 347 | error = xfs_setfilesize_ioend(ioend, error); |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 348 | xfs_destroy_ioend(ioend, error); |
Dave Chinner | 77d7a0c | 2010-02-17 05:36:29 +0000 | [diff] [blame] | 349 | } |
| 350 | |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 351 | STATIC void |
| 352 | xfs_end_bio( |
| 353 | struct bio *bio) |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 354 | { |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 355 | struct xfs_ioend *ioend = bio->bi_private; |
| 356 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 357 | |
Darrick J. Wong | 43caeb1 | 2016-10-03 09:11:35 -0700 | [diff] [blame] | 358 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 359 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
| 360 | else if (ioend->io_append_trans) |
| 361 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
| 362 | else |
| 363 | xfs_destroy_ioend(ioend, bio->bi_error); |
Christoph Hellwig | 0829c36 | 2005-09-02 16:58:49 +1000 | [diff] [blame] | 364 | } |
| 365 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | STATIC int |
| 367 | xfs_map_blocks( |
| 368 | struct inode *inode, |
| 369 | loff_t offset, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 370 | struct xfs_bmbt_irec *imap, |
Dave Chinner | 988ef92 | 2016-02-15 17:20:50 +1100 | [diff] [blame] | 371 | int type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 373 | struct xfs_inode *ip = XFS_I(inode); |
| 374 | struct xfs_mount *mp = ip->i_mount; |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 375 | ssize_t count = i_blocksize(inode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 376 | xfs_fileoff_t offset_fsb, end_fsb; |
| 377 | int error = 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 378 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
| 379 | int nimaps = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 380 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 381 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 382 | return -EIO; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 383 | |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 384 | ASSERT(type != XFS_IO_COW); |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 385 | if (type == XFS_IO_UNWRITTEN) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 386 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 387 | |
Dave Chinner | 988ef92 | 2016-02-15 17:20:50 +1100 | [diff] [blame] | 388 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 389 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
| 390 | (ip->i_df.if_flags & XFS_IFEXTENTS)); |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 391 | ASSERT(offset <= mp->m_super->s_maxbytes); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 392 | |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 393 | if (offset + count > mp->m_super->s_maxbytes) |
| 394 | count = mp->m_super->s_maxbytes - offset; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 395 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
| 396 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
Dave Chinner | 5c8ed20 | 2011-09-18 20:40:45 +0000 | [diff] [blame] | 397 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
| 398 | imap, &nimaps, bmapi_flags); |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 399 | /* |
| 400 | * Truncate an overwrite extent if there's a pending CoW |
| 401 | * reservation before the end of this extent. This forces us |
| 402 | * to come back to writepage to take care of the CoW. |
| 403 | */ |
| 404 | if (nimaps && type == XFS_IO_OVERWRITE) |
| 405 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 406 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
| 407 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 408 | if (error) |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 409 | return error; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 410 | |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 411 | if (type == XFS_IO_DELALLOC && |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 412 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
Darrick J. Wong | 60b4984 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 413 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
| 414 | imap); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 415 | if (!error) |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 416 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 417 | return error; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 418 | } |
| 419 | |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 420 | #ifdef DEBUG |
Alain Renaud | 0d882a3 | 2012-05-22 15:56:21 -0500 | [diff] [blame] | 421 | if (type == XFS_IO_UNWRITTEN) { |
Christoph Hellwig | 8ff2957 | 2010-12-10 08:42:21 +0000 | [diff] [blame] | 422 | ASSERT(nimaps); |
| 423 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 424 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
| 425 | } |
| 426 | #endif |
| 427 | if (nimaps) |
| 428 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); |
| 429 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | } |
| 431 | |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 432 | STATIC bool |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 433 | xfs_imap_valid( |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 434 | struct inode *inode, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 435 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 436 | xfs_off_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | { |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 438 | offset >>= inode->i_blkbits; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 439 | |
Christoph Hellwig | 558e689 | 2010-04-28 12:28:58 +0000 | [diff] [blame] | 440 | return offset >= imap->br_startoff && |
| 441 | offset < imap->br_startoff + imap->br_blockcount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 442 | } |
| 443 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 444 | STATIC void |
| 445 | xfs_start_buffer_writeback( |
| 446 | struct buffer_head *bh) |
| 447 | { |
| 448 | ASSERT(buffer_mapped(bh)); |
| 449 | ASSERT(buffer_locked(bh)); |
| 450 | ASSERT(!buffer_delay(bh)); |
| 451 | ASSERT(!buffer_unwritten(bh)); |
| 452 | |
Christoph Hellwig | 772003c | 2017-09-17 14:07:11 -0700 | [diff] [blame] | 453 | bh->b_end_io = NULL; |
| 454 | set_buffer_async_write(bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 455 | set_buffer_uptodate(bh); |
| 456 | clear_buffer_dirty(bh); |
| 457 | } |
| 458 | |
| 459 | STATIC void |
| 460 | xfs_start_page_writeback( |
| 461 | struct page *page, |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 462 | int clear_dirty) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 463 | { |
| 464 | ASSERT(PageLocked(page)); |
| 465 | ASSERT(!PageWriteback(page)); |
Dave Chinner | 0d085a5 | 2014-09-23 15:36:27 +1000 | [diff] [blame] | 466 | |
| 467 | /* |
| 468 | * if the page was not fully cleaned, we need to ensure that the higher |
| 469 | * layers come back to it correctly. That means we need to keep the page |
| 470 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the |
| 471 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to |
| 472 | * write this page in this writeback sweep will be made. |
| 473 | */ |
| 474 | if (clear_dirty) { |
David Chinner | 9213202 | 2006-12-21 10:24:01 +1100 | [diff] [blame] | 475 | clear_page_dirty_for_io(page); |
Dave Chinner | 0d085a5 | 2014-09-23 15:36:27 +1000 | [diff] [blame] | 476 | set_page_writeback(page); |
| 477 | } else |
| 478 | set_page_writeback_keepwrite(page); |
| 479 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 480 | unlock_page(page); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 481 | } |
| 482 | |
Zhi Yong Wu | c7c1a7d | 2013-08-07 10:11:09 +0000 | [diff] [blame] | 483 | 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] | 484 | { |
| 485 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
| 486 | } |
| 487 | |
| 488 | /* |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 489 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
| 490 | * it, and we submit that bio. The ioend may be used for multiple bio |
| 491 | * submissions, so we only want to allocate an append transaction for the ioend |
| 492 | * once. In the case of multiple bio submission, each bio will take an IO |
| 493 | * reference to the ioend to ensure that the ioend completion is only done once |
| 494 | * all bios have been submitted and the ioend is really done. |
Dave Chinner | 7bf7f35 | 2012-11-12 22:09:45 +1100 | [diff] [blame] | 495 | * |
| 496 | * If @fail is non-zero, it means that we have a situation where some part of |
| 497 | * the submission process has failed after we have marked paged for writeback |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 498 | * and unlocked them. In this situation, we need to fail the bio and ioend |
| 499 | * rather than submit it to IO. This typically only happens on a filesystem |
| 500 | * shutdown. |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 501 | */ |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 502 | STATIC int |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 503 | xfs_submit_ioend( |
Christoph Hellwig | 06342cf | 2009-10-30 09:09:15 +0000 | [diff] [blame] | 504 | struct writeback_control *wbc, |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 505 | struct xfs_ioend *ioend, |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 506 | int status) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 507 | { |
Darrick J. Wong | e02f0ff | 2017-02-02 15:14:02 -0800 | [diff] [blame] | 508 | /* Convert CoW extents to regular */ |
| 509 | if (!status && ioend->io_type == XFS_IO_COW) { |
| 510 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), |
| 511 | ioend->io_offset, ioend->io_size); |
| 512 | } |
| 513 | |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 514 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
| 515 | if (!status && |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 516 | ioend->io_type != XFS_IO_UNWRITTEN && |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 517 | xfs_ioend_is_append(ioend) && |
| 518 | !ioend->io_append_trans) |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 519 | status = xfs_setfilesize_trans_alloc(ioend); |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 520 | |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 521 | ioend->io_bio->bi_private = ioend; |
| 522 | ioend->io_bio->bi_end_io = xfs_end_bio; |
Mike Christie | 50bfcd0 | 2016-06-05 14:31:57 -0500 | [diff] [blame] | 523 | bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE, |
| 524 | (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0); |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 525 | /* |
| 526 | * If we are failing the IO now, just mark the ioend with an |
| 527 | * error and finish it. This will run IO completion immediately |
| 528 | * as there is only one reference to the ioend at this point in |
| 529 | * time. |
| 530 | */ |
| 531 | if (status) { |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 532 | ioend->io_bio->bi_error = status; |
| 533 | bio_endio(ioend->io_bio); |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 534 | return status; |
| 535 | } |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 536 | |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 537 | submit_bio(ioend->io_bio); |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 538 | return 0; |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 539 | } |
| 540 | |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 541 | static void |
| 542 | xfs_init_bio_from_bh( |
| 543 | struct bio *bio, |
| 544 | struct buffer_head *bh) |
| 545 | { |
| 546 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
| 547 | bio->bi_bdev = bh->b_bdev; |
| 548 | } |
| 549 | |
| 550 | static struct xfs_ioend * |
| 551 | xfs_alloc_ioend( |
| 552 | struct inode *inode, |
| 553 | unsigned int type, |
| 554 | xfs_off_t offset, |
| 555 | struct buffer_head *bh) |
| 556 | { |
| 557 | struct xfs_ioend *ioend; |
| 558 | struct bio *bio; |
| 559 | |
| 560 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
| 561 | xfs_init_bio_from_bh(bio, bh); |
| 562 | |
| 563 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); |
| 564 | INIT_LIST_HEAD(&ioend->io_list); |
| 565 | ioend->io_type = type; |
| 566 | ioend->io_inode = inode; |
| 567 | ioend->io_size = 0; |
| 568 | ioend->io_offset = offset; |
| 569 | INIT_WORK(&ioend->io_work, xfs_end_io); |
| 570 | ioend->io_append_trans = NULL; |
| 571 | ioend->io_bio = bio; |
| 572 | return ioend; |
| 573 | } |
| 574 | |
| 575 | /* |
| 576 | * Allocate a new bio, and chain the old bio to the new one. |
| 577 | * |
| 578 | * Note that we have to do perform the chaining in this unintuitive order |
| 579 | * so that the bi_private linkage is set up in the right direction for the |
| 580 | * traversal in xfs_destroy_ioend(). |
| 581 | */ |
| 582 | static void |
| 583 | xfs_chain_bio( |
| 584 | struct xfs_ioend *ioend, |
| 585 | struct writeback_control *wbc, |
| 586 | struct buffer_head *bh) |
| 587 | { |
| 588 | struct bio *new; |
| 589 | |
| 590 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); |
| 591 | xfs_init_bio_from_bh(new, bh); |
| 592 | |
| 593 | bio_chain(ioend->io_bio, new); |
| 594 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ |
Mike Christie | 50bfcd0 | 2016-06-05 14:31:57 -0500 | [diff] [blame] | 595 | bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE, |
| 596 | (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0); |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 597 | submit_bio(ioend->io_bio); |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 598 | ioend->io_bio = new; |
| 599 | } |
| 600 | |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 601 | /* |
| 602 | * Test to see if we've been building up a completion structure for |
| 603 | * earlier buffers -- if so, we try to append to this ioend if we |
| 604 | * can, otherwise we finish off any current ioend and start another. |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 605 | * Return the ioend we finished off so that the caller can submit it |
| 606 | * once it has finished processing the dirty page. |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 607 | */ |
| 608 | STATIC void |
| 609 | xfs_add_to_ioend( |
| 610 | struct inode *inode, |
| 611 | struct buffer_head *bh, |
Christoph Hellwig | 7336cea | 2006-01-11 20:49:16 +1100 | [diff] [blame] | 612 | xfs_off_t offset, |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 613 | struct xfs_writepage_ctx *wpc, |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 614 | struct writeback_control *wbc, |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 615 | struct list_head *iolist) |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 616 | { |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 617 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
Darrick J. Wong | 0df61da | 2016-03-07 09:32:14 +1100 | [diff] [blame] | 618 | bh->b_blocknr != wpc->last_block + 1 || |
| 619 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 620 | if (wpc->ioend) |
| 621 | list_add(&wpc->ioend->io_list, iolist); |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 622 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 623 | } |
| 624 | |
Christoph Hellwig | 0e51a8e | 2016-04-06 08:34:30 +1000 | [diff] [blame] | 625 | /* |
| 626 | * If the buffer doesn't fit into the bio we need to allocate a new |
| 627 | * one. This shouldn't happen more than once for a given buffer. |
| 628 | */ |
| 629 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) |
| 630 | xfs_chain_bio(wpc->ioend, wbc, bh); |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 631 | |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 632 | wpc->ioend->io_size += bh->b_size; |
| 633 | wpc->last_block = bh->b_blocknr; |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 634 | xfs_start_buffer_writeback(bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 635 | } |
| 636 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 637 | STATIC void |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 638 | xfs_map_buffer( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 639 | struct inode *inode, |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 640 | struct buffer_head *bh, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 641 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 642 | xfs_off_t offset) |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 643 | { |
| 644 | sector_t bn; |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 645 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 646 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
| 647 | 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] | 648 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 649 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 650 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 651 | |
Christoph Hellwig | e513182 | 2010-04-28 12:28:55 +0000 | [diff] [blame] | 652 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
Christoph Hellwig | 8699bb0 | 2010-04-28 12:28:54 +0000 | [diff] [blame] | 653 | ((offset - iomap_offset) >> inode->i_blkbits); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 654 | |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 655 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
Nathan Scott | 87cbc49 | 2006-03-14 13:26:43 +1100 | [diff] [blame] | 656 | |
| 657 | bh->b_blocknr = bn; |
| 658 | set_buffer_mapped(bh); |
| 659 | } |
| 660 | |
| 661 | STATIC void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 662 | xfs_map_at_offset( |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 663 | struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | struct buffer_head *bh, |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 665 | struct xfs_bmbt_irec *imap, |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 666 | xfs_off_t offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 667 | { |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 668 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
| 669 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 671 | xfs_map_buffer(inode, bh, imap, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 672 | set_buffer_mapped(bh); |
| 673 | clear_buffer_delay(bh); |
Christoph Hellwig | f6d6d4f | 2006-01-11 15:40:13 +1100 | [diff] [blame] | 674 | clear_buffer_unwritten(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | } |
| 676 | |
| 677 | /* |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 678 | * Test if a given page contains at least one buffer of a given @type. |
| 679 | * If @check_all_buffers is true, then we walk all the buffers in the page to |
| 680 | * try to find one of the type passed in. If it is not set, then the caller only |
| 681 | * needs to check the first buffer on the page for a match. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | */ |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 683 | STATIC bool |
Dave Chinner | 6ffc4db | 2012-04-23 15:58:43 +1000 | [diff] [blame] | 684 | xfs_check_page_type( |
Christoph Hellwig | 10ce444 | 2006-01-11 20:48:14 +1100 | [diff] [blame] | 685 | struct page *page, |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 686 | unsigned int type, |
| 687 | bool check_all_buffers) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | { |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 689 | struct buffer_head *bh; |
| 690 | struct buffer_head *head; |
| 691 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | if (PageWriteback(page)) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 693 | return false; |
| 694 | if (!page->mapping) |
| 695 | return false; |
| 696 | if (!page_has_buffers(page)) |
| 697 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 698 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 699 | bh = head = page_buffers(page); |
| 700 | do { |
| 701 | if (buffer_unwritten(bh)) { |
| 702 | if (type == XFS_IO_UNWRITTEN) |
| 703 | return true; |
| 704 | } else if (buffer_delay(bh)) { |
Dan Carpenter | 805eeb8 | 2014-04-04 06:56:30 +1100 | [diff] [blame] | 705 | if (type == XFS_IO_DELALLOC) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 706 | return true; |
| 707 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { |
Dan Carpenter | 805eeb8 | 2014-04-04 06:56:30 +1100 | [diff] [blame] | 708 | if (type == XFS_IO_OVERWRITE) |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 709 | return true; |
| 710 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 712 | /* If we are only checking the first buffer, we are done now. */ |
| 713 | if (!check_all_buffers) |
| 714 | break; |
| 715 | } while ((bh = bh->b_this_page) != head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 717 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | } |
| 719 | |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 720 | STATIC void |
| 721 | xfs_vm_invalidatepage( |
| 722 | struct page *page, |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 723 | unsigned int offset, |
| 724 | unsigned int length) |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 725 | { |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 726 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
| 727 | length); |
| 728 | block_invalidatepage(page, offset, length); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 729 | } |
| 730 | |
| 731 | /* |
| 732 | * If the page has delalloc buffers on it, we need to punch them out before we |
| 733 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the |
| 734 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read |
| 735 | * is done on that same region - the delalloc extent is returned when none is |
| 736 | * supposed to be there. |
| 737 | * |
| 738 | * We prevent this by truncating away the delalloc regions on the page before |
| 739 | * invalidating it. Because they are delalloc, we can do this without needing a |
| 740 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this |
| 741 | * truncation without a transaction as there is no space left for block |
| 742 | * reservation (typically why we see a ENOSPC in writeback). |
| 743 | * |
| 744 | * This is not a performance critical path, so for now just do the punching a |
| 745 | * buffer head at a time. |
| 746 | */ |
| 747 | STATIC void |
| 748 | xfs_aops_discard_page( |
| 749 | struct page *page) |
| 750 | { |
| 751 | struct inode *inode = page->mapping->host; |
| 752 | struct xfs_inode *ip = XFS_I(inode); |
| 753 | struct buffer_head *bh, *head; |
| 754 | loff_t offset = page_offset(page); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 755 | |
Dave Chinner | a49935f | 2014-03-07 16:19:14 +1100 | [diff] [blame] | 756 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 757 | goto out_invalidate; |
| 758 | |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 759 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 760 | goto out_invalidate; |
| 761 | |
Dave Chinner | 4f10700 | 2011-03-07 10:00:35 +1100 | [diff] [blame] | 762 | xfs_alert(ip->i_mount, |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 763 | "page discard on page %p, inode 0x%llx, offset %llu.", |
| 764 | page, ip->i_ino, offset); |
| 765 | |
| 766 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 767 | bh = head = page_buffers(page); |
| 768 | do { |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 769 | int error; |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 770 | xfs_fileoff_t start_fsb; |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 771 | |
| 772 | if (!buffer_delay(bh)) |
| 773 | goto next_buffer; |
| 774 | |
Dave Chinner | c726de4 | 2010-11-30 15:14:39 +1100 | [diff] [blame] | 775 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
| 776 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 777 | if (error) { |
| 778 | /* something screwed, just bail */ |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 779 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Dave Chinner | 4f10700 | 2011-03-07 10:00:35 +1100 | [diff] [blame] | 780 | xfs_alert(ip->i_mount, |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 781 | "page discard unable to remove delalloc mapping."); |
Dave Chinner | e8c3753 | 2010-03-15 02:36:35 +0000 | [diff] [blame] | 782 | } |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 783 | break; |
| 784 | } |
| 785 | next_buffer: |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 786 | offset += i_blocksize(inode); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 787 | |
| 788 | } while ((bh = bh->b_this_page) != head); |
| 789 | |
| 790 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 791 | out_invalidate: |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 792 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
Dave Chinner | 3ed3a43 | 2010-03-05 02:00:42 +0000 | [diff] [blame] | 793 | return; |
| 794 | } |
| 795 | |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 796 | static int |
| 797 | xfs_map_cow( |
| 798 | struct xfs_writepage_ctx *wpc, |
| 799 | struct inode *inode, |
| 800 | loff_t offset, |
| 801 | unsigned int *new_type) |
| 802 | { |
| 803 | struct xfs_inode *ip = XFS_I(inode); |
| 804 | struct xfs_bmbt_irec imap; |
| 805 | bool is_cow = false, need_alloc = false; |
| 806 | int error; |
| 807 | |
| 808 | /* |
| 809 | * If we already have a valid COW mapping keep using it. |
| 810 | */ |
| 811 | if (wpc->io_type == XFS_IO_COW) { |
| 812 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); |
| 813 | if (wpc->imap_valid) { |
| 814 | *new_type = XFS_IO_COW; |
| 815 | return 0; |
| 816 | } |
| 817 | } |
| 818 | |
| 819 | /* |
| 820 | * Else we need to check if there is a COW mapping at this offset. |
| 821 | */ |
| 822 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
| 823 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap, &need_alloc); |
| 824 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
| 825 | |
| 826 | if (!is_cow) |
| 827 | return 0; |
| 828 | |
| 829 | /* |
| 830 | * And if the COW mapping has a delayed extent here we need to |
| 831 | * allocate real space for it now. |
| 832 | */ |
| 833 | if (need_alloc) { |
| 834 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, |
| 835 | &imap); |
| 836 | if (error) |
| 837 | return error; |
| 838 | } |
| 839 | |
| 840 | wpc->io_type = *new_type = XFS_IO_COW; |
| 841 | wpc->imap_valid = true; |
| 842 | wpc->imap = imap; |
| 843 | return 0; |
| 844 | } |
| 845 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | /* |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 847 | * We implement an immediate ioend submission policy here to avoid needing to |
| 848 | * chain multiple ioends and hence nest mempool allocations which can violate |
| 849 | * forward progress guarantees we need to provide. The current ioend we are |
| 850 | * adding buffers to is cached on the writepage context, and if the new buffer |
| 851 | * does not append to the cached ioend it will create a new ioend and cache that |
| 852 | * instead. |
| 853 | * |
| 854 | * If a new ioend is created and cached, the old ioend is returned and queued |
| 855 | * locally for submission once the entire page is processed or an error has been |
| 856 | * detected. While ioends are submitted immediately after they are completed, |
| 857 | * batching optimisations are provided by higher level block plugging. |
| 858 | * |
| 859 | * At the end of a writeback pass, there will be a cached ioend remaining on the |
| 860 | * writepage context that the caller will need to submit. |
| 861 | */ |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 862 | static int |
| 863 | xfs_writepage_map( |
| 864 | struct xfs_writepage_ctx *wpc, |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 865 | struct writeback_control *wbc, |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 866 | struct inode *inode, |
| 867 | struct page *page, |
| 868 | loff_t offset, |
| 869 | __uint64_t end_offset) |
| 870 | { |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 871 | LIST_HEAD(submit_list); |
| 872 | struct xfs_ioend *ioend, *next; |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 873 | struct buffer_head *bh, *head; |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 874 | ssize_t len = i_blocksize(inode); |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 875 | int error = 0; |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 876 | int count = 0; |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 877 | int uptodate = 1; |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 878 | unsigned int new_type; |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 879 | |
| 880 | bh = head = page_buffers(page); |
| 881 | offset = page_offset(page); |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 882 | do { |
| 883 | if (offset >= end_offset) |
| 884 | break; |
| 885 | if (!buffer_uptodate(bh)) |
| 886 | uptodate = 0; |
| 887 | |
| 888 | /* |
| 889 | * set_page_dirty dirties all buffers in a page, independent |
| 890 | * of their state. The dirty state however is entirely |
| 891 | * meaningless for holes (!mapped && uptodate), so skip |
| 892 | * buffers covering holes here. |
| 893 | */ |
| 894 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { |
| 895 | wpc->imap_valid = false; |
| 896 | continue; |
| 897 | } |
| 898 | |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 899 | if (buffer_unwritten(bh)) |
| 900 | new_type = XFS_IO_UNWRITTEN; |
| 901 | else if (buffer_delay(bh)) |
| 902 | new_type = XFS_IO_DELALLOC; |
| 903 | else if (buffer_uptodate(bh)) |
| 904 | new_type = XFS_IO_OVERWRITE; |
| 905 | else { |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 906 | if (PageUptodate(page)) |
| 907 | ASSERT(buffer_mapped(bh)); |
| 908 | /* |
| 909 | * This buffer is not uptodate and will not be |
| 910 | * written to disk. Ensure that we will put any |
| 911 | * subsequent writeable buffers into a new |
| 912 | * ioend. |
| 913 | */ |
| 914 | wpc->imap_valid = false; |
| 915 | continue; |
| 916 | } |
| 917 | |
Darrick J. Wong | ef47366 | 2016-10-03 09:11:34 -0700 | [diff] [blame] | 918 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
| 919 | error = xfs_map_cow(wpc, inode, offset, &new_type); |
| 920 | if (error) |
| 921 | goto out; |
| 922 | } |
| 923 | |
| 924 | if (wpc->io_type != new_type) { |
| 925 | wpc->io_type = new_type; |
| 926 | wpc->imap_valid = false; |
| 927 | } |
| 928 | |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 929 | if (wpc->imap_valid) |
| 930 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
| 931 | offset); |
| 932 | if (!wpc->imap_valid) { |
| 933 | error = xfs_map_blocks(inode, offset, &wpc->imap, |
| 934 | wpc->io_type); |
| 935 | if (error) |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 936 | goto out; |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 937 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
| 938 | offset); |
| 939 | } |
| 940 | if (wpc->imap_valid) { |
| 941 | lock_buffer(bh); |
| 942 | if (wpc->io_type != XFS_IO_OVERWRITE) |
| 943 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); |
Dave Chinner | bb18782 | 2016-04-06 08:11:25 +1000 | [diff] [blame] | 944 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 945 | count++; |
| 946 | } |
| 947 | |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 948 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
| 949 | |
| 950 | if (uptodate && bh == head) |
| 951 | SetPageUptodate(page); |
| 952 | |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 953 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 954 | |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 955 | out: |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 956 | /* |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 957 | * On error, we have to fail the ioend here because we have locked |
| 958 | * buffers in the ioend. If we don't do this, we'll deadlock |
| 959 | * invalidating the page as that tries to lock the buffers on the page. |
| 960 | * Also, because we may have set pages under writeback, we have to make |
| 961 | * sure we run IO completion to mark the error state of the IO |
| 962 | * appropriately, so we can't cancel the ioend directly here. That means |
| 963 | * we have to mark this page as under writeback if we included any |
| 964 | * buffers from it in the ioend chain so that completion treats it |
| 965 | * correctly. |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 966 | * |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 967 | * If we didn't include the page in the ioend, the on error we can |
| 968 | * simply discard and unlock it as there are no other users of the page |
| 969 | * or it's buffers right now. The caller will still need to trigger |
| 970 | * submission of outstanding ioends on the writepage context so they are |
| 971 | * treated correctly on error. |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 972 | */ |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 973 | if (count) { |
| 974 | xfs_start_page_writeback(page, !error); |
| 975 | |
| 976 | /* |
| 977 | * Preserve the original error if there was one, otherwise catch |
| 978 | * submission errors here and propagate into subsequent ioend |
| 979 | * submissions. |
| 980 | */ |
| 981 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { |
| 982 | int error2; |
| 983 | |
| 984 | list_del_init(&ioend->io_list); |
| 985 | error2 = xfs_submit_ioend(wbc, ioend, error); |
| 986 | if (error2 && !error) |
| 987 | error = error2; |
| 988 | } |
| 989 | } else if (error) { |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 990 | xfs_aops_discard_page(page); |
| 991 | ClearPageUptodate(page); |
| 992 | unlock_page(page); |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 993 | } else { |
| 994 | /* |
| 995 | * We can end up here with no error and nothing to write if we |
| 996 | * race with a partial page truncate on a sub-page block sized |
| 997 | * filesystem. In that case we need to mark the page clean. |
| 998 | */ |
| 999 | xfs_start_page_writeback(page, 1); |
| 1000 | end_page_writeback(page); |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 1001 | } |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 1002 | |
Dave Chinner | bfce7d2 | 2016-02-15 17:21:37 +1100 | [diff] [blame] | 1003 | mapping_set_error(page->mapping, error); |
| 1004 | return error; |
| 1005 | } |
| 1006 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | /* |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1008 | * Write out a dirty page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | * |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1010 | * For delalloc space on the page we need to allocate space and flush it. |
| 1011 | * For unwritten space on the page we need to start the conversion to |
| 1012 | * regular allocated space. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1013 | * 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] | 1014 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | STATIC int |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1016 | xfs_do_writepage( |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1017 | struct page *page, |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1018 | struct writeback_control *wbc, |
| 1019 | void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | { |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1021 | struct xfs_writepage_ctx *wpc = data; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1022 | struct inode *inode = page->mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1023 | loff_t offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1024 | __uint64_t end_offset; |
Dave Chinner | ad68972 | 2016-02-15 17:21:31 +1100 | [diff] [blame] | 1025 | pgoff_t end_index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 1027 | trace_xfs_writepage(inode, page, 0, 0); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1028 | |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1029 | ASSERT(page_has_buffers(page)); |
| 1030 | |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1031 | /* |
| 1032 | * Refuse to write the page out if we are called from reclaim context. |
| 1033 | * |
Christoph Hellwig | d4f7a5c | 2010-06-28 10:34:44 -0400 | [diff] [blame] | 1034 | * This avoids stack overflows when called from deeply used stacks in |
| 1035 | * random callers for direct reclaim or memcg reclaim. We explicitly |
| 1036 | * allow reclaim from kswapd as the stack usage there is relatively low. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1037 | * |
Mel Gorman | 94054fa | 2011-10-31 17:07:45 -0700 | [diff] [blame] | 1038 | * This should never happen except in the case of a VM regression so |
| 1039 | * warn about it. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1040 | */ |
Mel Gorman | 94054fa | 2011-10-31 17:07:45 -0700 | [diff] [blame] | 1041 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
| 1042 | PF_MEMALLOC)) |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 1043 | goto redirty; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1044 | |
| 1045 | /* |
Christoph Hellwig | 680a647 | 2011-07-08 14:34:05 +0200 | [diff] [blame] | 1046 | * Given that we do not allow direct reclaim to call us, we should |
| 1047 | * never be called while in a filesystem transaction. |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1048 | */ |
Christoph Hellwig | 448011e | 2014-06-06 16:05:15 +1000 | [diff] [blame] | 1049 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 1050 | goto redirty; |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1051 | |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1052 | /* |
Dave Chinner | ad68972 | 2016-02-15 17:21:31 +1100 | [diff] [blame] | 1053 | * Is this page beyond the end of the file? |
| 1054 | * |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1055 | * The page index is less than the end_index, adjust the end_offset |
| 1056 | * to the highest offset that this page should represent. |
| 1057 | * ----------------------------------------------------- |
| 1058 | * | file mapping | <EOF> | |
| 1059 | * ----------------------------------------------------- |
| 1060 | * | Page ... | Page N-2 | Page N-1 | Page N | | |
| 1061 | * ^--------------------------------^----------|-------- |
| 1062 | * | desired writeback range | see else | |
| 1063 | * ---------------------------------^------------------| |
| 1064 | */ |
Dave Chinner | ad68972 | 2016-02-15 17:21:31 +1100 | [diff] [blame] | 1065 | offset = i_size_read(inode); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1066 | end_index = offset >> PAGE_SHIFT; |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1067 | if (page->index < end_index) |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1068 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1069 | else { |
| 1070 | /* |
| 1071 | * Check whether the page to write out is beyond or straddles |
| 1072 | * i_size or not. |
| 1073 | * ------------------------------------------------------- |
| 1074 | * | file mapping | <EOF> | |
| 1075 | * ------------------------------------------------------- |
| 1076 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | |
| 1077 | * ^--------------------------------^-----------|--------- |
| 1078 | * | | Straddles | |
| 1079 | * ---------------------------------^-----------|--------| |
| 1080 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1081 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1082 | |
| 1083 | /* |
Jan Kara | ff9a28f | 2013-03-14 14:30:54 +0100 | [diff] [blame] | 1084 | * Skip the page if it is fully outside i_size, e.g. due to a |
| 1085 | * truncate operation that is in progress. We must redirty the |
| 1086 | * page so that reclaim stops reclaiming it. Otherwise |
| 1087 | * xfs_vm_releasepage() is called on it and gets confused. |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1088 | * |
| 1089 | * Note that the end_index is unsigned long, it would overflow |
| 1090 | * if the given offset is greater than 16TB on 32-bit system |
| 1091 | * and if we do check the page is fully outside i_size or not |
| 1092 | * via "if (page->index >= end_index + 1)" as "end_index + 1" |
| 1093 | * will be evaluated to 0. Hence this page will be redirtied |
| 1094 | * and be written out repeatedly which would result in an |
| 1095 | * infinite loop, the user program that perform this operation |
| 1096 | * will hang. Instead, we can verify this situation by checking |
| 1097 | * if the page to write is totally beyond the i_size or if it's |
| 1098 | * offset is just equal to the EOF. |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1099 | */ |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1100 | if (page->index > end_index || |
| 1101 | (page->index == end_index && offset_into_page == 0)) |
Jan Kara | ff9a28f | 2013-03-14 14:30:54 +0100 | [diff] [blame] | 1102 | goto redirty; |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1103 | |
| 1104 | /* |
| 1105 | * The page straddles i_size. It must be zeroed out on each |
| 1106 | * and every writepage invocation because it may be mmapped. |
| 1107 | * "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] | 1108 | * that is not a multiple of the page size, the remaining |
Christoph Hellwig | 6b7a03f | 2012-07-03 12:20:00 -0400 | [diff] [blame] | 1109 | * memory is zeroed when mapped, and writes to that region are |
| 1110 | * not written out to the file." |
| 1111 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1112 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
Jie Liu | 8695d27 | 2014-05-20 08:24:26 +1000 | [diff] [blame] | 1113 | |
| 1114 | /* Adjust the end_offset to the end of file */ |
| 1115 | end_offset = offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | } |
| 1117 | |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 1118 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1119 | |
Christoph Hellwig | b5420f2 | 2010-08-24 11:47:51 +1000 | [diff] [blame] | 1120 | redirty: |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1121 | redirty_page_for_writepage(wbc, page); |
| 1122 | unlock_page(page); |
| 1123 | return 0; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1124 | } |
| 1125 | |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1126 | STATIC int |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1127 | xfs_vm_writepage( |
| 1128 | struct page *page, |
| 1129 | struct writeback_control *wbc) |
| 1130 | { |
| 1131 | struct xfs_writepage_ctx wpc = { |
| 1132 | .io_type = XFS_IO_INVALID, |
| 1133 | }; |
| 1134 | int ret; |
| 1135 | |
| 1136 | ret = xfs_do_writepage(page, wbc, &wpc); |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 1137 | if (wpc.ioend) |
| 1138 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); |
| 1139 | return ret; |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1140 | } |
| 1141 | |
| 1142 | STATIC int |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1143 | xfs_vm_writepages( |
| 1144 | struct address_space *mapping, |
| 1145 | struct writeback_control *wbc) |
| 1146 | { |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1147 | struct xfs_writepage_ctx wpc = { |
| 1148 | .io_type = XFS_IO_INVALID, |
| 1149 | }; |
| 1150 | int ret; |
| 1151 | |
Christoph Hellwig | b3aea4e | 2007-08-29 11:44:37 +1000 | [diff] [blame] | 1152 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
Ross Zwisler | 7f6d5b5 | 2016-02-26 15:19:55 -0800 | [diff] [blame] | 1153 | if (dax_mapping(mapping)) |
| 1154 | return dax_writeback_mapping_range(mapping, |
| 1155 | xfs_find_bdev_for_inode(mapping->host), wbc); |
| 1156 | |
Dave Chinner | fbcc025 | 2016-02-15 17:21:19 +1100 | [diff] [blame] | 1157 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
Dave Chinner | e10de37 | 2016-02-15 17:23:12 +1100 | [diff] [blame] | 1158 | if (wpc.ioend) |
| 1159 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); |
| 1160 | return ret; |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1161 | } |
| 1162 | |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1163 | /* |
| 1164 | * Called to move a page into cleanable state - and from there |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1165 | * to be released. The page should already be clean. We always |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1166 | * have buffer heads in this call. |
| 1167 | * |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1168 | * Returns 1 if the page is ok to release, 0 otherwise. |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1169 | */ |
| 1170 | STATIC int |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1171 | xfs_vm_releasepage( |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1172 | struct page *page, |
| 1173 | gfp_t gfp_mask) |
| 1174 | { |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1175 | int delalloc, unwritten; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1176 | |
Lukas Czerner | 34097df | 2013-05-21 23:58:01 -0400 | [diff] [blame] | 1177 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1178 | |
Brian Foster | 99579cc | 2016-07-22 09:50:38 +1000 | [diff] [blame] | 1179 | /* |
| 1180 | * mm accommodates an old ext3 case where clean pages might not have had |
| 1181 | * the dirty bit cleared. Thus, it can send actual dirty pages to |
| 1182 | * ->releasepage() via shrink_active_list(). Conversely, |
| 1183 | * block_invalidatepage() can send pages that are still marked dirty |
| 1184 | * but otherwise have invalidated buffers. |
| 1185 | * |
Jan Kara | 6ba35da | 2017-01-11 10:20:04 -0800 | [diff] [blame] | 1186 | * We want to release the latter to avoid unnecessary buildup of the |
| 1187 | * LRU, skip the former and warn if we've left any lingering |
| 1188 | * delalloc/unwritten buffers on clean pages. Skip pages with delalloc |
| 1189 | * or unwritten buffers and warn if the page is not dirty. Otherwise |
| 1190 | * try to release the buffers. |
Brian Foster | 99579cc | 2016-07-22 09:50:38 +1000 | [diff] [blame] | 1191 | */ |
Christoph Hellwig | 20cb52e | 2010-06-24 09:46:01 +1000 | [diff] [blame] | 1192 | xfs_count_page_state(page, &delalloc, &unwritten); |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1193 | |
Jan Kara | 6ba35da | 2017-01-11 10:20:04 -0800 | [diff] [blame] | 1194 | if (delalloc) { |
| 1195 | WARN_ON_ONCE(!PageDirty(page)); |
Christoph Hellwig | 89f3b363 | 2010-06-24 09:45:48 +1000 | [diff] [blame] | 1196 | return 0; |
Jan Kara | 6ba35da | 2017-01-11 10:20:04 -0800 | [diff] [blame] | 1197 | } |
| 1198 | if (unwritten) { |
| 1199 | WARN_ON_ONCE(!PageDirty(page)); |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1200 | return 0; |
Jan Kara | 6ba35da | 2017-01-11 10:20:04 -0800 | [diff] [blame] | 1201 | } |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1202 | |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1203 | return try_to_free_buffers(page); |
| 1204 | } |
| 1205 | |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1206 | /* |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1207 | * When we map a DIO buffer, we may need to pass flags to |
| 1208 | * xfs_end_io_direct_write to tell it what kind of write IO we are doing. |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1209 | * |
| 1210 | * Note that for DIO, an IO to the highest supported file block offset (i.e. |
| 1211 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 |
| 1212 | * bit variable. Hence if we see this overflow, we have to assume that the IO is |
| 1213 | * extending the file size. We won't know for sure until IO completion is run |
| 1214 | * and the actual max write offset is communicated to the IO completion |
| 1215 | * routine. |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1216 | */ |
| 1217 | static void |
| 1218 | xfs_map_direct( |
| 1219 | struct inode *inode, |
| 1220 | struct buffer_head *bh_result, |
| 1221 | struct xfs_bmbt_irec *imap, |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1222 | xfs_off_t offset, |
| 1223 | bool is_cow) |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1224 | { |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1225 | uintptr_t *flags = (uintptr_t *)&bh_result->b_private; |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1226 | xfs_off_t size = bh_result->b_size; |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1227 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1228 | trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size, |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1229 | ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : is_cow ? XFS_IO_COW : |
| 1230 | XFS_IO_OVERWRITE, imap); |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1231 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1232 | if (ISUNWRITTEN(imap)) { |
| 1233 | *flags |= XFS_DIO_FLAG_UNWRITTEN; |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1234 | set_buffer_defer_completion(bh_result); |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1235 | } else if (is_cow) { |
| 1236 | *flags |= XFS_DIO_FLAG_COW; |
| 1237 | set_buffer_defer_completion(bh_result); |
| 1238 | } |
| 1239 | if (offset + size > i_size_read(inode) || offset + size < 0) { |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1240 | *flags |= XFS_DIO_FLAG_APPEND; |
| 1241 | set_buffer_defer_completion(bh_result); |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1242 | } |
| 1243 | } |
| 1244 | |
Dave Chinner | 1fdca9c | 2015-04-16 21:58:21 +1000 | [diff] [blame] | 1245 | /* |
| 1246 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping |
| 1247 | * is, so that we can avoid repeated get_blocks calls. |
| 1248 | * |
| 1249 | * If the mapping spans EOF, then we have to break the mapping up as the mapping |
| 1250 | * for blocks beyond EOF must be marked new so that sub block regions can be |
| 1251 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping |
| 1252 | * was just allocated or is unwritten, otherwise the callers would overwrite |
| 1253 | * existing data with zeros. Hence we have to split the mapping into a range up |
| 1254 | * to and including EOF, and a second mapping for beyond EOF. |
| 1255 | */ |
| 1256 | static void |
| 1257 | xfs_map_trim_size( |
| 1258 | struct inode *inode, |
| 1259 | sector_t iblock, |
| 1260 | struct buffer_head *bh_result, |
| 1261 | struct xfs_bmbt_irec *imap, |
| 1262 | xfs_off_t offset, |
| 1263 | ssize_t size) |
| 1264 | { |
| 1265 | xfs_off_t mapping_size; |
| 1266 | |
| 1267 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; |
| 1268 | mapping_size <<= inode->i_blkbits; |
| 1269 | |
| 1270 | ASSERT(mapping_size > 0); |
| 1271 | if (mapping_size > size) |
| 1272 | mapping_size = size; |
| 1273 | if (offset < i_size_read(inode) && |
| 1274 | offset + mapping_size >= i_size_read(inode)) { |
| 1275 | /* limit mapping to block that spans EOF */ |
| 1276 | mapping_size = roundup_64(i_size_read(inode) - offset, |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 1277 | i_blocksize(inode)); |
Dave Chinner | 1fdca9c | 2015-04-16 21:58:21 +1000 | [diff] [blame] | 1278 | } |
| 1279 | if (mapping_size > LONG_MAX) |
| 1280 | mapping_size = LONG_MAX; |
| 1281 | |
| 1282 | bh_result->b_size = mapping_size; |
| 1283 | } |
| 1284 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | STATIC int |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1286 | __xfs_get_blocks( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | struct inode *inode, |
| 1288 | sector_t iblock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | struct buffer_head *bh_result, |
| 1290 | int create, |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1291 | bool direct, |
| 1292 | bool dax_fault) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1293 | { |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1294 | struct xfs_inode *ip = XFS_I(inode); |
| 1295 | struct xfs_mount *mp = ip->i_mount; |
| 1296 | xfs_fileoff_t offset_fsb, end_fsb; |
| 1297 | int error = 0; |
| 1298 | int lockmode = 0; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1299 | struct xfs_bmbt_irec imap; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1300 | int nimaps = 1; |
Nathan Scott | fdc7ed7 | 2005-11-02 15:13:13 +1100 | [diff] [blame] | 1301 | xfs_off_t offset; |
| 1302 | ssize_t size; |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1303 | int new = 0; |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1304 | bool is_cow = false; |
| 1305 | bool need_alloc = false; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1306 | |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1307 | BUG_ON(create && !direct); |
| 1308 | |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1309 | if (XFS_FORCED_SHUTDOWN(mp)) |
Eric Sandeen | b474c7a | 2014-06-22 15:04:54 +1000 | [diff] [blame] | 1310 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1311 | |
Nathan Scott | fdc7ed7 | 2005-11-02 15:13:13 +1100 | [diff] [blame] | 1312 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 1313 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1314 | size = bh_result->b_size; |
Lachlan McIlroy | 364f358 | 2008-09-17 16:50:14 +1000 | [diff] [blame] | 1315 | |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1316 | if (!create && offset >= i_size_read(inode)) |
Lachlan McIlroy | 364f358 | 2008-09-17 16:50:14 +1000 | [diff] [blame] | 1317 | return 0; |
| 1318 | |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1319 | /* |
| 1320 | * Direct I/O is usually done on preallocated files, so try getting |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1321 | * a block mapping without an exclusive lock first. |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1322 | */ |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1323 | lockmode = xfs_ilock_data_map_shared(ip); |
Christoph Hellwig | f2bde9b | 2010-06-24 11:44:35 +1000 | [diff] [blame] | 1324 | |
Dave Chinner | d2c2819 | 2012-06-08 15:44:53 +1000 | [diff] [blame] | 1325 | ASSERT(offset <= mp->m_super->s_maxbytes); |
| 1326 | if (offset + size > mp->m_super->s_maxbytes) |
| 1327 | size = mp->m_super->s_maxbytes - offset; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1328 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
| 1329 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
| 1330 | |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1331 | if (create && direct && xfs_is_reflink_inode(ip)) |
| 1332 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap, |
| 1333 | &need_alloc); |
| 1334 | if (!is_cow) { |
| 1335 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
| 1336 | &imap, &nimaps, XFS_BMAPI_ENTIRE); |
| 1337 | /* |
| 1338 | * Truncate an overwrite extent if there's a pending CoW |
| 1339 | * reservation before the end of this extent. This |
| 1340 | * forces us to come back to get_blocks to take care of |
| 1341 | * the CoW. |
| 1342 | */ |
| 1343 | if (create && direct && nimaps && |
| 1344 | imap.br_startblock != HOLESTARTBLOCK && |
| 1345 | imap.br_startblock != DELAYSTARTBLOCK && |
| 1346 | !ISUNWRITTEN(&imap)) |
| 1347 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, |
| 1348 | &imap); |
| 1349 | } |
| 1350 | ASSERT(!need_alloc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | if (error) |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1352 | goto out_unlock; |
| 1353 | |
Brian Foster | 3978c5bb | 2017-01-09 16:38:35 +0100 | [diff] [blame] | 1354 | /* |
| 1355 | * The only time we can ever safely find delalloc blocks on direct I/O |
| 1356 | * is a dio write to post-eof speculative preallocation. All other |
| 1357 | * scenarios are indicative of a problem or misuse (such as mixing |
| 1358 | * direct and mapped I/O). |
| 1359 | * |
| 1360 | * The file may be unmapped by the time we get here so we cannot |
| 1361 | * reliably fail the I/O based on mapping. Instead, fail the I/O if this |
| 1362 | * is a read or a write within eof. Otherwise, carry on but warn as a |
| 1363 | * precuation if the file happens to be mapped. |
| 1364 | */ |
| 1365 | if (direct && imap.br_startblock == DELAYSTARTBLOCK) { |
| 1366 | if (!create || offset < i_size_read(VFS_I(ip))) { |
| 1367 | WARN_ON_ONCE(1); |
| 1368 | error = -EIO; |
| 1369 | goto out_unlock; |
| 1370 | } |
| 1371 | WARN_ON_ONCE(mapping_mapped(VFS_I(ip)->i_mapping)); |
| 1372 | } |
| 1373 | |
Dave Chinner | 1ca1915 | 2015-11-03 12:37:00 +1100 | [diff] [blame] | 1374 | /* for DAX, we convert unwritten extents directly */ |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1375 | if (create && |
| 1376 | (!nimaps || |
| 1377 | (imap.br_startblock == HOLESTARTBLOCK || |
Dave Chinner | 1ca1915 | 2015-11-03 12:37:00 +1100 | [diff] [blame] | 1378 | imap.br_startblock == DELAYSTARTBLOCK) || |
| 1379 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1380 | /* |
| 1381 | * xfs_iomap_write_direct() expects the shared lock. It |
| 1382 | * is unlocked on return. |
| 1383 | */ |
| 1384 | if (lockmode == XFS_ILOCK_EXCL) |
| 1385 | xfs_ilock_demote(ip, lockmode); |
Brian Foster | 009c6e8 | 2015-10-12 15:34:20 +1100 | [diff] [blame] | 1386 | |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1387 | error = xfs_iomap_write_direct(ip, offset, size, |
| 1388 | &imap, nimaps); |
| 1389 | if (error) |
| 1390 | return error; |
| 1391 | new = 1; |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1392 | |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1393 | trace_xfs_get_blocks_alloc(ip, offset, size, |
| 1394 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN |
| 1395 | : XFS_IO_DELALLOC, &imap); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1396 | } else if (nimaps) { |
Dave Chinner | d5cc2e3 | 2015-04-16 21:59:07 +1000 | [diff] [blame] | 1397 | trace_xfs_get_blocks_found(ip, offset, size, |
| 1398 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN |
| 1399 | : XFS_IO_OVERWRITE, &imap); |
Dave Chinner | 507630b | 2012-03-27 10:34:50 -0400 | [diff] [blame] | 1400 | xfs_iunlock(ip, lockmode); |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1401 | } else { |
| 1402 | trace_xfs_get_blocks_notfound(ip, offset, size); |
| 1403 | goto out_unlock; |
| 1404 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1405 | |
Dave Chinner | 1ca1915 | 2015-11-03 12:37:00 +1100 | [diff] [blame] | 1406 | if (IS_DAX(inode) && create) { |
| 1407 | ASSERT(!ISUNWRITTEN(&imap)); |
| 1408 | /* zeroing is not needed at a higher layer */ |
| 1409 | new = 0; |
| 1410 | } |
| 1411 | |
Dave Chinner | 1fdca9c | 2015-04-16 21:58:21 +1000 | [diff] [blame] | 1412 | /* trim mapping down to size requested */ |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1413 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
Dave Chinner | 1fdca9c | 2015-04-16 21:58:21 +1000 | [diff] [blame] | 1414 | |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1415 | /* |
| 1416 | * For unwritten extents do not report a disk address in the buffered |
| 1417 | * read case (treat as if we're reading into a hole). |
| 1418 | */ |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1419 | if (imap.br_startblock != HOLESTARTBLOCK && |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1420 | imap.br_startblock != DELAYSTARTBLOCK && |
| 1421 | (create || !ISUNWRITTEN(&imap))) { |
| 1422 | xfs_map_buffer(inode, bh_result, &imap, offset); |
| 1423 | if (ISUNWRITTEN(&imap)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1424 | set_buffer_unwritten(bh_result); |
Dave Chinner | a719370 | 2015-04-16 21:57:48 +1000 | [diff] [blame] | 1425 | /* direct IO needs special help */ |
Christoph Hellwig | 6e8a27a | 2016-06-21 09:53:45 +1000 | [diff] [blame] | 1426 | if (create) { |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1427 | if (dax_fault) |
| 1428 | ASSERT(!ISUNWRITTEN(&imap)); |
| 1429 | else |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1430 | xfs_map_direct(inode, bh_result, &imap, offset, |
| 1431 | is_cow); |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1432 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | } |
| 1434 | |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1435 | /* |
| 1436 | * If this is a realtime file, data may be on a different device. |
| 1437 | * to that pointed to from the buffer_head b_bdev currently. |
| 1438 | */ |
Christoph Hellwig | 046f168 | 2010-04-28 12:28:52 +0000 | [diff] [blame] | 1439 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1440 | |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1441 | /* |
David Chinner | 549054a | 2007-02-10 18:36:35 +1100 | [diff] [blame] | 1442 | * If we previously allocated a block out beyond eof and we are now |
| 1443 | * coming back to use it then we will need to flag it as new even if it |
| 1444 | * has a disk address. |
| 1445 | * |
| 1446 | * With sub-block writes into unwritten extents we also need to mark |
| 1447 | * the buffer as new so that the unwritten parts of the buffer gets |
| 1448 | * correctly zeroed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1449 | */ |
| 1450 | if (create && |
| 1451 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || |
David Chinner | 549054a | 2007-02-10 18:36:35 +1100 | [diff] [blame] | 1452 | (offset >= i_size_read(inode)) || |
Christoph Hellwig | 207d041 | 2010-04-28 12:28:56 +0000 | [diff] [blame] | 1453 | (new || ISUNWRITTEN(&imap)))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | set_buffer_new(bh_result); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1456 | return 0; |
Christoph Hellwig | a206c81 | 2010-12-10 08:42:20 +0000 | [diff] [blame] | 1457 | |
| 1458 | out_unlock: |
| 1459 | xfs_iunlock(ip, lockmode); |
Dave Chinner | 2451337 | 2014-06-25 14:58:08 +1000 | [diff] [blame] | 1460 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1461 | } |
| 1462 | |
| 1463 | int |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1464 | xfs_get_blocks( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | struct inode *inode, |
| 1466 | sector_t iblock, |
| 1467 | struct buffer_head *bh_result, |
| 1468 | int create) |
| 1469 | { |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1470 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | } |
| 1472 | |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1473 | int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1474 | xfs_get_blocks_direct( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | struct inode *inode, |
| 1476 | sector_t iblock, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | struct buffer_head *bh_result, |
| 1478 | int create) |
| 1479 | { |
Dave Chinner | 3e12dbb | 2015-11-03 12:27:22 +1100 | [diff] [blame] | 1480 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
| 1481 | } |
| 1482 | |
| 1483 | int |
| 1484 | xfs_get_blocks_dax_fault( |
| 1485 | struct inode *inode, |
| 1486 | sector_t iblock, |
| 1487 | struct buffer_head *bh_result, |
| 1488 | int create) |
| 1489 | { |
| 1490 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | } |
| 1492 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1493 | /* |
| 1494 | * Complete a direct I/O write request. |
| 1495 | * |
| 1496 | * xfs_map_direct passes us some flags in the private data to tell us what to |
| 1497 | * do. If no flags are set, then the write IO is an overwrite wholly within |
| 1498 | * the existing allocated file size and so there is nothing for us to do. |
| 1499 | * |
| 1500 | * Note that in this case the completion can be called in interrupt context, |
| 1501 | * whereas if we have flags set we will always be called in task context |
| 1502 | * (i.e. from a workqueue). |
| 1503 | */ |
Christoph Hellwig | fa8d972 | 2016-07-20 11:38:01 +1000 | [diff] [blame] | 1504 | int |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1505 | xfs_end_io_direct_write( |
| 1506 | struct kiocb *iocb, |
Christoph Hellwig | 209fb87 | 2010-07-18 21:17:11 +0000 | [diff] [blame] | 1507 | loff_t offset, |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1508 | ssize_t size, |
| 1509 | void *private) |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1510 | { |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1511 | struct inode *inode = file_inode(iocb->ki_filp); |
| 1512 | struct xfs_inode *ip = XFS_I(inode); |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1513 | uintptr_t flags = (uintptr_t)private; |
| 1514 | int error = 0; |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 1515 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1516 | trace_xfs_end_io_direct_write(ip, offset, size); |
| 1517 | |
Christoph Hellwig | e372843 | 2016-09-19 11:26:41 +1000 | [diff] [blame] | 1518 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1519 | return -EIO; |
| 1520 | |
| 1521 | if (size <= 0) |
| 1522 | return size; |
Christoph Hellwig | f097386 | 2005-09-05 08:22:52 +1000 | [diff] [blame] | 1523 | |
| 1524 | /* |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1525 | * The flags tell us whether we are doing unwritten extent conversions |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1526 | * or an append transaction that updates the on-disk file size. These |
| 1527 | * cases are the only cases where we should *potentially* be needing |
Dave Chinner | a06c277 | 2015-04-16 22:00:00 +1000 | [diff] [blame] | 1528 | * to update the VFS inode size. |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1529 | */ |
| 1530 | if (flags == 0) { |
| 1531 | ASSERT(offset + size <= i_size_read(inode)); |
| 1532 | return 0; |
| 1533 | } |
| 1534 | |
| 1535 | /* |
Dave Chinner | 6dfa1b6 | 2015-04-16 21:59:34 +1000 | [diff] [blame] | 1536 | * 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] | 1537 | * with the on-disk inode size being outside the in-core inode size. We |
| 1538 | * have no other method of updating EOF for AIO, so always do it here |
| 1539 | * if necessary. |
Dave Chinner | b9d5984 | 2015-04-16 22:03:07 +1000 | [diff] [blame] | 1540 | * |
| 1541 | * We need to lock the test/set EOF update as we can be racing with |
| 1542 | * other IO completions here to update the EOF. Failing to serialise |
| 1543 | * here can result in EOF moving backwards and Bad Things Happen when |
| 1544 | * that occurs. |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 1545 | */ |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1546 | spin_lock(&ip->i_flags_lock); |
Christoph Hellwig | 2ba6623 | 2015-02-02 10:02:09 +1100 | [diff] [blame] | 1547 | if (offset + size > i_size_read(inode)) |
| 1548 | i_size_write(inode, offset + size); |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1549 | spin_unlock(&ip->i_flags_lock); |
Christoph Hellwig | 2813d68 | 2011-12-18 20:00:12 +0000 | [diff] [blame] | 1550 | |
Christoph Hellwig | feac470 | 2016-10-11 09:03:19 +1100 | [diff] [blame] | 1551 | if (flags & XFS_DIO_FLAG_COW) |
| 1552 | error = xfs_reflink_end_cow(ip, offset, size); |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1553 | if (flags & XFS_DIO_FLAG_UNWRITTEN) { |
| 1554 | trace_xfs_end_io_direct_write_unwritten(ip, offset, size); |
Christoph Hellwig | 40e2e97 | 2010-07-18 21:17:09 +0000 | [diff] [blame] | 1555 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1556 | error = xfs_iomap_write_unwritten(ip, offset, size); |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1557 | } |
Darrick J. Wong | 0613f16 | 2016-10-03 09:11:37 -0700 | [diff] [blame] | 1558 | if (flags & XFS_DIO_FLAG_APPEND) { |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1559 | trace_xfs_end_io_direct_write_append(ip, offset, size); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1560 | |
Christoph Hellwig | e372843 | 2016-09-19 11:26:41 +1000 | [diff] [blame] | 1561 | error = xfs_setfilesize(ip, offset, size); |
Dave Chinner | 6b698ed | 2015-06-04 09:18:53 +1000 | [diff] [blame] | 1562 | } |
| 1563 | |
Christoph Hellwig | 273dda7 | 2016-02-08 14:40:51 +1100 | [diff] [blame] | 1564 | return error; |
Dave Chinner | 6e1ba0b | 2015-06-04 09:19:15 +1000 | [diff] [blame] | 1565 | } |
| 1566 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | STATIC ssize_t |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1568 | xfs_vm_direct_IO( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1569 | struct kiocb *iocb, |
Christoph Hellwig | c8b8e32 | 2016-04-07 08:51:58 -0700 | [diff] [blame] | 1570 | struct iov_iter *iter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1571 | { |
Jie Liu | 58e5985 | 2013-07-16 13:11:16 +0800 | [diff] [blame] | 1572 | /* |
Christoph Hellwig | fa8d972 | 2016-07-20 11:38:01 +1000 | [diff] [blame] | 1573 | * We just need the method present so that open/fcntl allow direct I/O. |
Jie Liu | 58e5985 | 2013-07-16 13:11:16 +0800 | [diff] [blame] | 1574 | */ |
Christoph Hellwig | fa8d972 | 2016-07-20 11:38:01 +1000 | [diff] [blame] | 1575 | return -EINVAL; |
Nathan Scott | f51623b | 2006-03-14 13:26:27 +1100 | [diff] [blame] | 1576 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | |
| 1578 | STATIC sector_t |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1579 | xfs_vm_bmap( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1580 | struct address_space *mapping, |
| 1581 | sector_t block) |
| 1582 | { |
| 1583 | struct inode *inode = (struct inode *)mapping->host; |
Christoph Hellwig | 739bfb2 | 2007-08-29 10:58:01 +1000 | [diff] [blame] | 1584 | struct xfs_inode *ip = XFS_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1585 | |
Christoph Hellwig | cca28fb | 2010-06-24 11:57:09 +1000 | [diff] [blame] | 1586 | trace_xfs_vm_bmap(XFS_I(inode)); |
Christoph Hellwig | 126468b | 2008-03-06 13:44:57 +1100 | [diff] [blame] | 1587 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
Darrick J. Wong | db1327b | 2016-10-03 09:11:36 -0700 | [diff] [blame] | 1588 | |
| 1589 | /* |
| 1590 | * The swap code (ab-)uses ->bmap to get a block mapping and then |
| 1591 | * bypasseѕ the file system for actual I/O. We really can't allow |
| 1592 | * that on reflinks inodes, so we have to skip out here. And yes, |
Darrick J. Wong | 621d0b7 | 2017-06-19 13:19:08 -0700 | [diff] [blame] | 1593 | * 0 is the magic code for a bmap error. |
| 1594 | * |
| 1595 | * Since we don't pass back blockdev info, we can't return bmap |
| 1596 | * information for rt files either. |
Darrick J. Wong | db1327b | 2016-10-03 09:11:36 -0700 | [diff] [blame] | 1597 | */ |
Darrick J. Wong | 621d0b7 | 2017-06-19 13:19:08 -0700 | [diff] [blame] | 1598 | if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) { |
Darrick J. Wong | db1327b | 2016-10-03 09:11:36 -0700 | [diff] [blame] | 1599 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
| 1600 | return 0; |
| 1601 | } |
Dave Chinner | 4bc1ea6 | 2012-11-12 22:53:56 +1100 | [diff] [blame] | 1602 | filemap_write_and_wait(mapping); |
Christoph Hellwig | 126468b | 2008-03-06 13:44:57 +1100 | [diff] [blame] | 1603 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1604 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1605 | } |
| 1606 | |
| 1607 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1608 | xfs_vm_readpage( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1609 | struct file *unused, |
| 1610 | struct page *page) |
| 1611 | { |
Dave Chinner | 121e213 | 2016-01-08 11:28:35 +1100 | [diff] [blame] | 1612 | trace_xfs_vm_readpage(page->mapping->host, 1); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1613 | return mpage_readpage(page, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | } |
| 1615 | |
| 1616 | STATIC int |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1617 | xfs_vm_readpages( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | struct file *unused, |
| 1619 | struct address_space *mapping, |
| 1620 | struct list_head *pages, |
| 1621 | unsigned nr_pages) |
| 1622 | { |
Dave Chinner | 121e213 | 2016-01-08 11:28:35 +1100 | [diff] [blame] | 1623 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
Nathan Scott | c253666 | 2006-03-29 10:44:40 +1000 | [diff] [blame] | 1624 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1625 | } |
| 1626 | |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1627 | /* |
| 1628 | * This is basically a copy of __set_page_dirty_buffers() with one |
| 1629 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them |
| 1630 | * dirty, we'll never be able to clean them because we don't write buffers |
| 1631 | * beyond EOF, and that means we can't invalidate pages that span EOF |
| 1632 | * that have been marked dirty. Further, the dirty state can leak into |
| 1633 | * the file interior if the file is extended, resulting in all sorts of |
| 1634 | * bad things happening as the state does not match the underlying data. |
| 1635 | * |
| 1636 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like |
| 1637 | * this only exist because of bufferheads and how the generic code manages them. |
| 1638 | */ |
| 1639 | STATIC int |
| 1640 | xfs_vm_set_page_dirty( |
| 1641 | struct page *page) |
| 1642 | { |
| 1643 | struct address_space *mapping = page->mapping; |
| 1644 | struct inode *inode = mapping->host; |
| 1645 | loff_t end_offset; |
| 1646 | loff_t offset; |
| 1647 | int newly_dirty; |
| 1648 | |
| 1649 | if (unlikely(!mapping)) |
| 1650 | return !TestSetPageDirty(page); |
| 1651 | |
| 1652 | end_offset = i_size_read(inode); |
| 1653 | offset = page_offset(page); |
| 1654 | |
| 1655 | spin_lock(&mapping->private_lock); |
| 1656 | if (page_has_buffers(page)) { |
| 1657 | struct buffer_head *head = page_buffers(page); |
| 1658 | struct buffer_head *bh = head; |
| 1659 | |
| 1660 | do { |
| 1661 | if (offset < end_offset) |
| 1662 | set_buffer_dirty(bh); |
| 1663 | bh = bh->b_this_page; |
Fabian Frederick | 61604a2 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 1664 | offset += i_blocksize(inode); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1665 | } while (bh != head); |
| 1666 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1667 | /* |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1668 | * Lock out page->mem_cgroup migration to keep PageDirty |
| 1669 | * synchronized with per-memcg dirty page counters. |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1670 | */ |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1671 | lock_page_memcg(page); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1672 | newly_dirty = !TestSetPageDirty(page); |
| 1673 | spin_unlock(&mapping->private_lock); |
| 1674 | |
| 1675 | if (newly_dirty) { |
| 1676 | /* sigh - __set_page_dirty() is static, so copy it here, too */ |
| 1677 | unsigned long flags; |
| 1678 | |
| 1679 | spin_lock_irqsave(&mapping->tree_lock, flags); |
| 1680 | if (page->mapping) { /* Race with truncate? */ |
| 1681 | WARN_ON_ONCE(!PageUptodate(page)); |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1682 | account_page_dirtied(page, mapping); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1683 | radix_tree_tag_set(&mapping->page_tree, |
| 1684 | page_index(page), PAGECACHE_TAG_DIRTY); |
| 1685 | } |
| 1686 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1687 | } |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1688 | unlock_page_memcg(page); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1689 | if (newly_dirty) |
| 1690 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1691 | return newly_dirty; |
| 1692 | } |
| 1693 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1694 | const struct address_space_operations xfs_address_space_operations = { |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1695 | .readpage = xfs_vm_readpage, |
| 1696 | .readpages = xfs_vm_readpages, |
| 1697 | .writepage = xfs_vm_writepage, |
Nathan Scott | 7d4fb40 | 2006-06-09 15:27:16 +1000 | [diff] [blame] | 1698 | .writepages = xfs_vm_writepages, |
Dave Chinner | 22e757a | 2014-09-02 12:12:51 +1000 | [diff] [blame] | 1699 | .set_page_dirty = xfs_vm_set_page_dirty, |
Nathan Scott | 238f4c5 | 2006-03-17 17:26:25 +1100 | [diff] [blame] | 1700 | .releasepage = xfs_vm_releasepage, |
| 1701 | .invalidatepage = xfs_vm_invalidatepage, |
Nathan Scott | e4c573b | 2006-03-14 13:54:26 +1100 | [diff] [blame] | 1702 | .bmap = xfs_vm_bmap, |
| 1703 | .direct_IO = xfs_vm_direct_IO, |
Christoph Lameter | e965f96 | 2006-02-01 03:05:41 -0800 | [diff] [blame] | 1704 | .migratepage = buffer_migrate_page, |
Hisashi Hifumi | bddaafa | 2009-03-29 09:53:38 +0200 | [diff] [blame] | 1705 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 1706 | .error_remove_page = generic_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1707 | }; |