David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
| 4 | * |
| 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 |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 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. |
| 13 | * |
| 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 |
| 17 | */ |
| 18 | #include "xfs.h" |
| 19 | #include "xfs_fs.h" |
| 20 | #include "xfs_types.h" |
| 21 | #include "xfs_bit.h" |
| 22 | #include "xfs_log.h" |
| 23 | #include "xfs_inum.h" |
| 24 | #include "xfs_trans.h" |
| 25 | #include "xfs_sb.h" |
| 26 | #include "xfs_ag.h" |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 27 | #include "xfs_mount.h" |
| 28 | #include "xfs_bmap_btree.h" |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 29 | #include "xfs_inode.h" |
| 30 | #include "xfs_dinode.h" |
| 31 | #include "xfs_error.h" |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 32 | #include "xfs_filestream.h" |
| 33 | #include "xfs_vnodeops.h" |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 34 | #include "xfs_inode_item.h" |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 35 | #include "xfs_quota.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 36 | #include "xfs_trace.h" |
Dave Chinner | 1a387d3 | 2010-08-24 11:46:31 +1000 | [diff] [blame] | 37 | #include "xfs_fsops.h" |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 38 | |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 39 | #include <linux/kthread.h> |
| 40 | #include <linux/freezer.h> |
| 41 | |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 42 | /* |
| 43 | * The inode lookup is done in batches to keep the amount of lock traffic and |
| 44 | * radix tree lookups to a minimum. The batch size is a trade off between |
| 45 | * lookup reduction and stack usage. This is in the reclaim path, so we can't |
| 46 | * be too greedy. |
| 47 | */ |
| 48 | #define XFS_LOOKUP_BATCH 32 |
| 49 | |
Dave Chinner | e13de95 | 2010-09-28 12:28:06 +1000 | [diff] [blame] | 50 | STATIC int |
| 51 | xfs_inode_ag_walk_grab( |
| 52 | struct xfs_inode *ip) |
| 53 | { |
| 54 | struct inode *inode = VFS_I(ip); |
| 55 | |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 56 | ASSERT(rcu_read_lock_held()); |
| 57 | |
| 58 | /* |
| 59 | * check for stale RCU freed inode |
| 60 | * |
| 61 | * If the inode has been reallocated, it doesn't matter if it's not in |
| 62 | * the AG we are walking - we are walking for writeback, so if it |
| 63 | * passes all the "valid inode" checks and is dirty, then we'll write |
| 64 | * it back anyway. If it has been reallocated and still being |
| 65 | * initialised, the XFS_INEW check below will catch it. |
| 66 | */ |
| 67 | spin_lock(&ip->i_flags_lock); |
| 68 | if (!ip->i_ino) |
| 69 | goto out_unlock_noent; |
| 70 | |
| 71 | /* avoid new or reclaimable inodes. Leave for reclaim code to flush */ |
| 72 | if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM)) |
| 73 | goto out_unlock_noent; |
| 74 | spin_unlock(&ip->i_flags_lock); |
| 75 | |
Dave Chinner | e13de95 | 2010-09-28 12:28:06 +1000 | [diff] [blame] | 76 | /* nothing to sync during shutdown */ |
| 77 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
| 78 | return EFSCORRUPTED; |
| 79 | |
Dave Chinner | e13de95 | 2010-09-28 12:28:06 +1000 | [diff] [blame] | 80 | /* If we can't grab the inode, it must on it's way to reclaim. */ |
| 81 | if (!igrab(inode)) |
| 82 | return ENOENT; |
| 83 | |
| 84 | if (is_bad_inode(inode)) { |
| 85 | IRELE(ip); |
| 86 | return ENOENT; |
| 87 | } |
| 88 | |
| 89 | /* inode is valid */ |
| 90 | return 0; |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 91 | |
| 92 | out_unlock_noent: |
| 93 | spin_unlock(&ip->i_flags_lock); |
| 94 | return ENOENT; |
Dave Chinner | e13de95 | 2010-09-28 12:28:06 +1000 | [diff] [blame] | 95 | } |
| 96 | |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 97 | STATIC int |
| 98 | xfs_inode_ag_walk( |
| 99 | struct xfs_mount *mp, |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 100 | struct xfs_perag *pag, |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 101 | int (*execute)(struct xfs_inode *ip, |
| 102 | struct xfs_perag *pag, int flags), |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 103 | int flags) |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 104 | { |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 105 | uint32_t first_index; |
| 106 | int last_error = 0; |
| 107 | int skipped; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 108 | int done; |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 109 | int nr_found; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 110 | |
| 111 | restart: |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 112 | done = 0; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 113 | skipped = 0; |
| 114 | first_index = 0; |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 115 | nr_found = 0; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 116 | do { |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 117 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 118 | int error = 0; |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 119 | int i; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 120 | |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 121 | rcu_read_lock(); |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 122 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 123 | (void **)batch, first_index, |
| 124 | XFS_LOOKUP_BATCH); |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 125 | if (!nr_found) { |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 126 | rcu_read_unlock(); |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 127 | break; |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 128 | } |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 129 | |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 130 | /* |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 131 | * Grab the inodes before we drop the lock. if we found |
| 132 | * nothing, nr == 0 and the loop will be skipped. |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 133 | */ |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 134 | for (i = 0; i < nr_found; i++) { |
| 135 | struct xfs_inode *ip = batch[i]; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 136 | |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 137 | if (done || xfs_inode_ag_walk_grab(ip)) |
| 138 | batch[i] = NULL; |
| 139 | |
| 140 | /* |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 141 | * Update the index for the next lookup. Catch |
| 142 | * overflows into the next AG range which can occur if |
| 143 | * we have inodes in the last block of the AG and we |
| 144 | * are currently pointing to the last inode. |
| 145 | * |
| 146 | * Because we may see inodes that are from the wrong AG |
| 147 | * due to RCU freeing and reallocation, only update the |
| 148 | * index if it lies in this AG. It was a race that lead |
| 149 | * us to see this inode, so another lookup from the |
| 150 | * same index will not find it again. |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 151 | */ |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 152 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno) |
| 153 | continue; |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 154 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
| 155 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) |
| 156 | done = 1; |
Dave Chinner | e13de95 | 2010-09-28 12:28:06 +1000 | [diff] [blame] | 157 | } |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 158 | |
| 159 | /* unlock now we've grabbed the inodes. */ |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 160 | rcu_read_unlock(); |
Dave Chinner | e13de95 | 2010-09-28 12:28:06 +1000 | [diff] [blame] | 161 | |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 162 | for (i = 0; i < nr_found; i++) { |
| 163 | if (!batch[i]) |
| 164 | continue; |
| 165 | error = execute(batch[i], pag, flags); |
| 166 | IRELE(batch[i]); |
| 167 | if (error == EAGAIN) { |
| 168 | skipped++; |
| 169 | continue; |
| 170 | } |
| 171 | if (error && last_error != EFSCORRUPTED) |
| 172 | last_error = error; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 173 | } |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 174 | |
| 175 | /* bail out if the filesystem is corrupted. */ |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 176 | if (error == EFSCORRUPTED) |
| 177 | break; |
| 178 | |
Dave Chinner | 78ae525 | 2010-09-28 12:28:19 +1000 | [diff] [blame] | 179 | } while (nr_found && !done); |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 180 | |
| 181 | if (skipped) { |
| 182 | delay(1); |
| 183 | goto restart; |
| 184 | } |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 185 | return last_error; |
| 186 | } |
| 187 | |
Christoph Hellwig | fe588ed | 2009-06-08 15:35:27 +0200 | [diff] [blame] | 188 | int |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 189 | xfs_inode_ag_iterator( |
| 190 | struct xfs_mount *mp, |
| 191 | int (*execute)(struct xfs_inode *ip, |
| 192 | struct xfs_perag *pag, int flags), |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 193 | int flags) |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 194 | { |
Dave Chinner | 16fd536 | 2010-07-20 09:43:39 +1000 | [diff] [blame] | 195 | struct xfs_perag *pag; |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 196 | int error = 0; |
| 197 | int last_error = 0; |
| 198 | xfs_agnumber_t ag; |
| 199 | |
Dave Chinner | 16fd536 | 2010-07-20 09:43:39 +1000 | [diff] [blame] | 200 | ag = 0; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 201 | while ((pag = xfs_perag_get(mp, ag))) { |
| 202 | ag = pag->pag_agno + 1; |
| 203 | error = xfs_inode_ag_walk(mp, pag, execute, flags); |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 204 | xfs_perag_put(pag); |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 205 | if (error) { |
| 206 | last_error = error; |
| 207 | if (error == EFSCORRUPTED) |
| 208 | break; |
| 209 | } |
| 210 | } |
| 211 | return XFS_ERROR(last_error); |
| 212 | } |
| 213 | |
Dave Chinner | 5a34d5c | 2009-06-08 15:35:03 +0200 | [diff] [blame] | 214 | STATIC int |
| 215 | xfs_sync_inode_data( |
| 216 | struct xfs_inode *ip, |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 217 | struct xfs_perag *pag, |
Dave Chinner | 5a34d5c | 2009-06-08 15:35:03 +0200 | [diff] [blame] | 218 | int flags) |
| 219 | { |
| 220 | struct inode *inode = VFS_I(ip); |
| 221 | struct address_space *mapping = inode->i_mapping; |
| 222 | int error = 0; |
| 223 | |
| 224 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 225 | goto out_wait; |
| 226 | |
| 227 | if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED)) { |
| 228 | if (flags & SYNC_TRYLOCK) |
| 229 | goto out_wait; |
| 230 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
| 231 | } |
| 232 | |
| 233 | error = xfs_flush_pages(ip, 0, -1, (flags & SYNC_WAIT) ? |
Christoph Hellwig | 0cadda1 | 2010-01-19 09:56:44 +0000 | [diff] [blame] | 234 | 0 : XBF_ASYNC, FI_NONE); |
Dave Chinner | 5a34d5c | 2009-06-08 15:35:03 +0200 | [diff] [blame] | 235 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
| 236 | |
| 237 | out_wait: |
Christoph Hellwig | b0710cc | 2009-06-08 15:37:11 +0200 | [diff] [blame] | 238 | if (flags & SYNC_WAIT) |
Dave Chinner | 5a34d5c | 2009-06-08 15:35:03 +0200 | [diff] [blame] | 239 | xfs_ioend_wait(ip); |
| 240 | return error; |
| 241 | } |
| 242 | |
Christoph Hellwig | 845b6d0 | 2009-06-08 15:35:05 +0200 | [diff] [blame] | 243 | STATIC int |
| 244 | xfs_sync_inode_attr( |
| 245 | struct xfs_inode *ip, |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 246 | struct xfs_perag *pag, |
Christoph Hellwig | 845b6d0 | 2009-06-08 15:35:05 +0200 | [diff] [blame] | 247 | int flags) |
| 248 | { |
| 249 | int error = 0; |
| 250 | |
| 251 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
| 252 | if (xfs_inode_clean(ip)) |
| 253 | goto out_unlock; |
| 254 | if (!xfs_iflock_nowait(ip)) { |
| 255 | if (!(flags & SYNC_WAIT)) |
| 256 | goto out_unlock; |
| 257 | xfs_iflock(ip); |
| 258 | } |
| 259 | |
| 260 | if (xfs_inode_clean(ip)) { |
| 261 | xfs_ifunlock(ip); |
| 262 | goto out_unlock; |
| 263 | } |
| 264 | |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 265 | error = xfs_iflush(ip, flags); |
Christoph Hellwig | 845b6d0 | 2009-06-08 15:35:05 +0200 | [diff] [blame] | 266 | |
| 267 | out_unlock: |
| 268 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
| 269 | return error; |
| 270 | } |
| 271 | |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 272 | /* |
| 273 | * Write out pagecache data for the whole filesystem. |
| 274 | */ |
Christoph Hellwig | 64c8614 | 2010-06-24 11:45:34 +1000 | [diff] [blame] | 275 | STATIC int |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 276 | xfs_sync_data( |
| 277 | struct xfs_mount *mp, |
| 278 | int flags) |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 279 | { |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 280 | int error; |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 281 | |
Christoph Hellwig | b0710cc | 2009-06-08 15:37:11 +0200 | [diff] [blame] | 282 | ASSERT((flags & ~(SYNC_TRYLOCK|SYNC_WAIT)) == 0); |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 283 | |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 284 | error = xfs_inode_ag_iterator(mp, xfs_sync_inode_data, flags); |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 285 | if (error) |
| 286 | return XFS_ERROR(error); |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 287 | |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 288 | xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0); |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 289 | return 0; |
| 290 | } |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 291 | |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 292 | /* |
| 293 | * Write out inode metadata (attributes) for the whole filesystem. |
| 294 | */ |
Christoph Hellwig | 64c8614 | 2010-06-24 11:45:34 +1000 | [diff] [blame] | 295 | STATIC int |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 296 | xfs_sync_attr( |
| 297 | struct xfs_mount *mp, |
| 298 | int flags) |
| 299 | { |
| 300 | ASSERT((flags & ~SYNC_WAIT) == 0); |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 301 | |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 302 | return xfs_inode_ag_iterator(mp, xfs_sync_inode_attr, flags); |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 303 | } |
| 304 | |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 305 | STATIC int |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 306 | xfs_sync_fsdata( |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 307 | struct xfs_mount *mp) |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 308 | { |
| 309 | struct xfs_buf *bp; |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 310 | |
| 311 | /* |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 312 | * If the buffer is pinned then push on the log so we won't get stuck |
| 313 | * waiting in the write for someone, maybe ourselves, to flush the log. |
| 314 | * |
| 315 | * Even though we just pushed the log above, we did not have the |
| 316 | * superblock buffer locked at that point so it can become pinned in |
| 317 | * between there and here. |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 318 | */ |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 319 | bp = xfs_getsb(mp, 0); |
| 320 | if (XFS_BUF_ISPINNED(bp)) |
| 321 | xfs_log_force(mp, 0); |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 322 | |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 323 | return xfs_bwrite(mp, bp); |
Christoph Hellwig | 2af75df | 2008-10-30 17:14:53 +1100 | [diff] [blame] | 324 | } |
| 325 | |
David Chinner | fe4fa4b | 2008-10-30 17:06:08 +1100 | [diff] [blame] | 326 | /* |
David Chinner | a4e4c4f | 2008-10-30 17:16:11 +1100 | [diff] [blame] | 327 | * When remounting a filesystem read-only or freezing the filesystem, we have |
| 328 | * two phases to execute. This first phase is syncing the data before we |
| 329 | * quiesce the filesystem, and the second is flushing all the inodes out after |
| 330 | * we've waited for all the transactions created by the first phase to |
| 331 | * complete. The second phase ensures that the inodes are written to their |
| 332 | * location on disk rather than just existing in transactions in the log. This |
| 333 | * means after a quiesce there is no log replay required to write the inodes to |
| 334 | * disk (this is the main difference between a sync and a quiesce). |
| 335 | */ |
| 336 | /* |
| 337 | * First stage of freeze - no writers will make progress now we are here, |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 338 | * so we flush delwri and delalloc buffers here, then wait for all I/O to |
| 339 | * complete. Data is frozen at that point. Metadata is not frozen, |
David Chinner | a4e4c4f | 2008-10-30 17:16:11 +1100 | [diff] [blame] | 340 | * transactions can still occur here so don't bother flushing the buftarg |
| 341 | * because it'll just get dirty again. |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 342 | */ |
| 343 | int |
| 344 | xfs_quiesce_data( |
| 345 | struct xfs_mount *mp) |
| 346 | { |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 347 | int error, error2 = 0; |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 348 | |
| 349 | /* push non-blocking */ |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 350 | xfs_sync_data(mp, 0); |
Christoph Hellwig | 8b5403a | 2009-06-08 15:37:16 +0200 | [diff] [blame] | 351 | xfs_qm_sync(mp, SYNC_TRYLOCK); |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 352 | |
Dave Chinner | c90b07e | 2009-10-06 20:29:27 +0000 | [diff] [blame] | 353 | /* push and block till complete */ |
Christoph Hellwig | b0710cc | 2009-06-08 15:37:11 +0200 | [diff] [blame] | 354 | xfs_sync_data(mp, SYNC_WAIT); |
Christoph Hellwig | 7d09525 | 2009-06-08 15:33:32 +0200 | [diff] [blame] | 355 | xfs_qm_sync(mp, SYNC_WAIT); |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 356 | |
David Chinner | a4e4c4f | 2008-10-30 17:16:11 +1100 | [diff] [blame] | 357 | /* write superblock and hoover up shutdown errors */ |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 358 | error = xfs_sync_fsdata(mp); |
| 359 | |
| 360 | /* make sure all delwri buffers are written out */ |
| 361 | xfs_flush_buftarg(mp->m_ddev_targp, 1); |
| 362 | |
| 363 | /* mark the log as covered if needed */ |
| 364 | if (xfs_log_need_covered(mp)) |
Dave Chinner | 1a387d3 | 2010-08-24 11:46:31 +1000 | [diff] [blame] | 365 | error2 = xfs_fs_log_dummy(mp, SYNC_WAIT); |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 366 | |
David Chinner | a4e4c4f | 2008-10-30 17:16:11 +1100 | [diff] [blame] | 367 | /* flush data-only devices */ |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 368 | if (mp->m_rtdev_targp) |
| 369 | XFS_bflush(mp->m_rtdev_targp); |
| 370 | |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 371 | return error ? error : error2; |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 372 | } |
| 373 | |
David Chinner | 76bf105 | 2008-10-30 17:16:21 +1100 | [diff] [blame] | 374 | STATIC void |
| 375 | xfs_quiesce_fs( |
| 376 | struct xfs_mount *mp) |
| 377 | { |
| 378 | int count = 0, pincount; |
| 379 | |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 380 | xfs_reclaim_inodes(mp, 0); |
David Chinner | 76bf105 | 2008-10-30 17:16:21 +1100 | [diff] [blame] | 381 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
David Chinner | 76bf105 | 2008-10-30 17:16:21 +1100 | [diff] [blame] | 382 | |
| 383 | /* |
| 384 | * This loop must run at least twice. The first instance of the loop |
| 385 | * will flush most meta data but that will generate more meta data |
| 386 | * (typically directory updates). Which then must be flushed and |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 387 | * logged before we can write the unmount record. We also so sync |
| 388 | * reclaim of inodes to catch any that the above delwri flush skipped. |
David Chinner | 76bf105 | 2008-10-30 17:16:21 +1100 | [diff] [blame] | 389 | */ |
| 390 | do { |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 391 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 392 | xfs_sync_attr(mp, SYNC_WAIT); |
David Chinner | 76bf105 | 2008-10-30 17:16:21 +1100 | [diff] [blame] | 393 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
| 394 | if (!pincount) { |
| 395 | delay(50); |
| 396 | count++; |
| 397 | } |
| 398 | } while (count < 2); |
| 399 | } |
| 400 | |
| 401 | /* |
| 402 | * Second stage of a quiesce. The data is already synced, now we have to take |
| 403 | * care of the metadata. New transactions are already blocked, so we need to |
| 404 | * wait for any remaining transactions to drain out before proceding. |
| 405 | */ |
| 406 | void |
| 407 | xfs_quiesce_attr( |
| 408 | struct xfs_mount *mp) |
| 409 | { |
| 410 | int error = 0; |
| 411 | |
| 412 | /* wait for all modifications to complete */ |
| 413 | while (atomic_read(&mp->m_active_trans) > 0) |
| 414 | delay(100); |
| 415 | |
| 416 | /* flush inodes and push all remaining buffers out to disk */ |
| 417 | xfs_quiesce_fs(mp); |
| 418 | |
Felix Blyakher | 5e10657 | 2009-01-22 21:34:05 -0600 | [diff] [blame] | 419 | /* |
| 420 | * Just warn here till VFS can correctly support |
| 421 | * read-only remount without racing. |
| 422 | */ |
| 423 | WARN_ON(atomic_read(&mp->m_active_trans) != 0); |
David Chinner | 76bf105 | 2008-10-30 17:16:21 +1100 | [diff] [blame] | 424 | |
| 425 | /* Push the superblock and write an unmount record */ |
| 426 | error = xfs_log_sbcount(mp, 1); |
| 427 | if (error) |
| 428 | xfs_fs_cmn_err(CE_WARN, mp, |
| 429 | "xfs_attr_quiesce: failed to log sb changes. " |
| 430 | "Frozen image may not be consistent."); |
| 431 | xfs_log_unmount_write(mp); |
| 432 | xfs_unmountfs_writesb(mp); |
| 433 | } |
| 434 | |
David Chinner | e9f1c6e | 2008-10-30 17:15:50 +1100 | [diff] [blame] | 435 | /* |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 436 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. |
| 437 | * Doing this has two advantages: |
| 438 | * - It saves on stack space, which is tight in certain situations |
| 439 | * - It can be used (with care) as a mechanism to avoid deadlocks. |
| 440 | * Flushing while allocating in a full filesystem requires both. |
| 441 | */ |
| 442 | STATIC void |
| 443 | xfs_syncd_queue_work( |
| 444 | struct xfs_mount *mp, |
| 445 | void *data, |
Dave Chinner | e43afd7 | 2009-04-06 18:47:27 +0200 | [diff] [blame] | 446 | void (*syncer)(struct xfs_mount *, void *), |
| 447 | struct completion *completion) |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 448 | { |
Dave Chinner | a8d770d | 2009-04-06 18:44:54 +0200 | [diff] [blame] | 449 | struct xfs_sync_work *work; |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 450 | |
Dave Chinner | a8d770d | 2009-04-06 18:44:54 +0200 | [diff] [blame] | 451 | work = kmem_alloc(sizeof(struct xfs_sync_work), KM_SLEEP); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 452 | INIT_LIST_HEAD(&work->w_list); |
| 453 | work->w_syncer = syncer; |
| 454 | work->w_data = data; |
| 455 | work->w_mount = mp; |
Dave Chinner | e43afd7 | 2009-04-06 18:47:27 +0200 | [diff] [blame] | 456 | work->w_completion = completion; |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 457 | spin_lock(&mp->m_sync_lock); |
| 458 | list_add_tail(&work->w_list, &mp->m_sync_list); |
| 459 | spin_unlock(&mp->m_sync_lock); |
| 460 | wake_up_process(mp->m_sync_task); |
| 461 | } |
| 462 | |
| 463 | /* |
| 464 | * Flush delayed allocate data, attempting to free up reserved space |
| 465 | * from existing allocations. At this point a new allocation attempt |
| 466 | * has failed with ENOSPC and we are in the process of scratching our |
| 467 | * heads, looking about for more room... |
| 468 | */ |
| 469 | STATIC void |
Dave Chinner | a8d770d | 2009-04-06 18:44:54 +0200 | [diff] [blame] | 470 | xfs_flush_inodes_work( |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 471 | struct xfs_mount *mp, |
| 472 | void *arg) |
| 473 | { |
| 474 | struct inode *inode = arg; |
Christoph Hellwig | 075fe10 | 2009-06-08 15:35:48 +0200 | [diff] [blame] | 475 | xfs_sync_data(mp, SYNC_TRYLOCK); |
Christoph Hellwig | b0710cc | 2009-06-08 15:37:11 +0200 | [diff] [blame] | 476 | xfs_sync_data(mp, SYNC_TRYLOCK | SYNC_WAIT); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 477 | iput(inode); |
| 478 | } |
| 479 | |
| 480 | void |
Dave Chinner | a8d770d | 2009-04-06 18:44:54 +0200 | [diff] [blame] | 481 | xfs_flush_inodes( |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 482 | xfs_inode_t *ip) |
| 483 | { |
| 484 | struct inode *inode = VFS_I(ip); |
Dave Chinner | e43afd7 | 2009-04-06 18:47:27 +0200 | [diff] [blame] | 485 | DECLARE_COMPLETION_ONSTACK(completion); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 486 | |
| 487 | igrab(inode); |
Dave Chinner | e43afd7 | 2009-04-06 18:47:27 +0200 | [diff] [blame] | 488 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion); |
| 489 | wait_for_completion(&completion); |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 490 | xfs_log_force(ip->i_mount, XFS_LOG_SYNC); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 491 | } |
| 492 | |
David Chinner | aacaa88 | 2008-10-30 17:15:29 +1100 | [diff] [blame] | 493 | /* |
Christoph Hellwig | df308bc | 2010-03-12 10:59:16 +0000 | [diff] [blame] | 494 | * Every sync period we need to unpin all items, reclaim inodes and sync |
| 495 | * disk quotas. We might need to cover the log to indicate that the |
Dave Chinner | 1a387d3 | 2010-08-24 11:46:31 +1000 | [diff] [blame] | 496 | * filesystem is idle and not frozen. |
David Chinner | aacaa88 | 2008-10-30 17:15:29 +1100 | [diff] [blame] | 497 | */ |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 498 | STATIC void |
| 499 | xfs_sync_worker( |
| 500 | struct xfs_mount *mp, |
| 501 | void *unused) |
| 502 | { |
| 503 | int error; |
| 504 | |
David Chinner | aacaa88 | 2008-10-30 17:15:29 +1100 | [diff] [blame] | 505 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 506 | xfs_log_force(mp, 0); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 507 | xfs_reclaim_inodes(mp, 0); |
David Chinner | aacaa88 | 2008-10-30 17:15:29 +1100 | [diff] [blame] | 508 | /* dgc: errors ignored here */ |
Christoph Hellwig | 8b5403a | 2009-06-08 15:37:16 +0200 | [diff] [blame] | 509 | error = xfs_qm_sync(mp, SYNC_TRYLOCK); |
Dave Chinner | 1a387d3 | 2010-08-24 11:46:31 +1000 | [diff] [blame] | 510 | if (mp->m_super->s_frozen == SB_UNFROZEN && |
| 511 | xfs_log_need_covered(mp)) |
| 512 | error = xfs_fs_log_dummy(mp, 0); |
David Chinner | aacaa88 | 2008-10-30 17:15:29 +1100 | [diff] [blame] | 513 | } |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 514 | mp->m_sync_seq++; |
| 515 | wake_up(&mp->m_wait_single_sync_task); |
| 516 | } |
| 517 | |
| 518 | STATIC int |
| 519 | xfssyncd( |
| 520 | void *arg) |
| 521 | { |
| 522 | struct xfs_mount *mp = arg; |
| 523 | long timeleft; |
Dave Chinner | a8d770d | 2009-04-06 18:44:54 +0200 | [diff] [blame] | 524 | xfs_sync_work_t *work, *n; |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 525 | LIST_HEAD (tmp); |
| 526 | |
| 527 | set_freezable(); |
| 528 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); |
| 529 | for (;;) { |
Dave Chinner | 20f6b2c | 2010-03-04 01:46:23 +0000 | [diff] [blame] | 530 | if (list_empty(&mp->m_sync_list)) |
| 531 | timeleft = schedule_timeout_interruptible(timeleft); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 532 | /* swsusp */ |
| 533 | try_to_freeze(); |
| 534 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) |
| 535 | break; |
| 536 | |
| 537 | spin_lock(&mp->m_sync_lock); |
| 538 | /* |
| 539 | * We can get woken by laptop mode, to do a sync - |
| 540 | * that's the (only!) case where the list would be |
| 541 | * empty with time remaining. |
| 542 | */ |
| 543 | if (!timeleft || list_empty(&mp->m_sync_list)) { |
| 544 | if (!timeleft) |
| 545 | timeleft = xfs_syncd_centisecs * |
| 546 | msecs_to_jiffies(10); |
| 547 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); |
| 548 | list_add_tail(&mp->m_sync_work.w_list, |
| 549 | &mp->m_sync_list); |
| 550 | } |
Dave Chinner | 20f6b2c | 2010-03-04 01:46:23 +0000 | [diff] [blame] | 551 | list_splice_init(&mp->m_sync_list, &tmp); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 552 | spin_unlock(&mp->m_sync_lock); |
| 553 | |
| 554 | list_for_each_entry_safe(work, n, &tmp, w_list) { |
| 555 | (*work->w_syncer)(mp, work->w_data); |
| 556 | list_del(&work->w_list); |
| 557 | if (work == &mp->m_sync_work) |
| 558 | continue; |
Dave Chinner | e43afd7 | 2009-04-06 18:47:27 +0200 | [diff] [blame] | 559 | if (work->w_completion) |
| 560 | complete(work->w_completion); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 561 | kmem_free(work); |
| 562 | } |
| 563 | } |
| 564 | |
| 565 | return 0; |
| 566 | } |
| 567 | |
| 568 | int |
| 569 | xfs_syncd_init( |
| 570 | struct xfs_mount *mp) |
| 571 | { |
| 572 | mp->m_sync_work.w_syncer = xfs_sync_worker; |
| 573 | mp->m_sync_work.w_mount = mp; |
Dave Chinner | e43afd7 | 2009-04-06 18:47:27 +0200 | [diff] [blame] | 574 | mp->m_sync_work.w_completion = NULL; |
Jan Engelhardt | e2a0781 | 2010-03-23 09:52:55 +1100 | [diff] [blame] | 575 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd/%s", mp->m_fsname); |
David Chinner | a167b17 | 2008-10-30 17:06:18 +1100 | [diff] [blame] | 576 | if (IS_ERR(mp->m_sync_task)) |
| 577 | return -PTR_ERR(mp->m_sync_task); |
| 578 | return 0; |
| 579 | } |
| 580 | |
| 581 | void |
| 582 | xfs_syncd_stop( |
| 583 | struct xfs_mount *mp) |
| 584 | { |
| 585 | kthread_stop(mp->m_sync_task); |
| 586 | } |
| 587 | |
Christoph Hellwig | bc990f5 | 2009-08-16 20:36:34 -0400 | [diff] [blame] | 588 | void |
| 589 | __xfs_inode_set_reclaim_tag( |
| 590 | struct xfs_perag *pag, |
| 591 | struct xfs_inode *ip) |
| 592 | { |
| 593 | radix_tree_tag_set(&pag->pag_ici_root, |
| 594 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), |
| 595 | XFS_ICI_RECLAIM_TAG); |
Dave Chinner | 16fd536 | 2010-07-20 09:43:39 +1000 | [diff] [blame] | 596 | |
| 597 | if (!pag->pag_ici_reclaimable) { |
| 598 | /* propagate the reclaim tag up into the perag radix tree */ |
| 599 | spin_lock(&ip->i_mount->m_perag_lock); |
| 600 | radix_tree_tag_set(&ip->i_mount->m_perag_tree, |
| 601 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), |
| 602 | XFS_ICI_RECLAIM_TAG); |
| 603 | spin_unlock(&ip->i_mount->m_perag_lock); |
| 604 | trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno, |
| 605 | -1, _RET_IP_); |
| 606 | } |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 607 | pag->pag_ici_reclaimable++; |
Christoph Hellwig | bc990f5 | 2009-08-16 20:36:34 -0400 | [diff] [blame] | 608 | } |
| 609 | |
David Chinner | 1165451 | 2008-10-30 17:37:49 +1100 | [diff] [blame] | 610 | /* |
| 611 | * We set the inode flag atomically with the radix tree tag. |
| 612 | * Once we get tag lookups on the radix tree, this inode flag |
| 613 | * can go away. |
| 614 | */ |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 615 | void |
| 616 | xfs_inode_set_reclaim_tag( |
| 617 | xfs_inode_t *ip) |
| 618 | { |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 619 | struct xfs_mount *mp = ip->i_mount; |
| 620 | struct xfs_perag *pag; |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 621 | |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 622 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
Christoph Hellwig | f1f724e | 2010-03-01 11:30:31 +0000 | [diff] [blame] | 623 | write_lock(&pag->pag_ici_lock); |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 624 | spin_lock(&ip->i_flags_lock); |
Christoph Hellwig | bc990f5 | 2009-08-16 20:36:34 -0400 | [diff] [blame] | 625 | __xfs_inode_set_reclaim_tag(pag, ip); |
David Chinner | 1165451 | 2008-10-30 17:37:49 +1100 | [diff] [blame] | 626 | __xfs_iflags_set(ip, XFS_IRECLAIMABLE); |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 627 | spin_unlock(&ip->i_flags_lock); |
Christoph Hellwig | f1f724e | 2010-03-01 11:30:31 +0000 | [diff] [blame] | 628 | write_unlock(&pag->pag_ici_lock); |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 629 | xfs_perag_put(pag); |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 630 | } |
| 631 | |
Johannes Weiner | 081003f | 2010-10-01 07:43:54 +0000 | [diff] [blame] | 632 | STATIC void |
| 633 | __xfs_inode_clear_reclaim( |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 634 | xfs_perag_t *pag, |
| 635 | xfs_inode_t *ip) |
| 636 | { |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 637 | pag->pag_ici_reclaimable--; |
Dave Chinner | 16fd536 | 2010-07-20 09:43:39 +1000 | [diff] [blame] | 638 | if (!pag->pag_ici_reclaimable) { |
| 639 | /* clear the reclaim tag from the perag radix tree */ |
| 640 | spin_lock(&ip->i_mount->m_perag_lock); |
| 641 | radix_tree_tag_clear(&ip->i_mount->m_perag_tree, |
| 642 | XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino), |
| 643 | XFS_ICI_RECLAIM_TAG); |
| 644 | spin_unlock(&ip->i_mount->m_perag_lock); |
| 645 | trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno, |
| 646 | -1, _RET_IP_); |
| 647 | } |
David Chinner | 396beb8 | 2008-10-30 17:37:26 +1100 | [diff] [blame] | 648 | } |
| 649 | |
Johannes Weiner | 081003f | 2010-10-01 07:43:54 +0000 | [diff] [blame] | 650 | void |
| 651 | __xfs_inode_clear_reclaim_tag( |
| 652 | xfs_mount_t *mp, |
| 653 | xfs_perag_t *pag, |
| 654 | xfs_inode_t *ip) |
| 655 | { |
| 656 | radix_tree_tag_clear(&pag->pag_ici_root, |
| 657 | XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG); |
| 658 | __xfs_inode_clear_reclaim(pag, ip); |
| 659 | } |
| 660 | |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 661 | /* |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 662 | * Grab the inode for reclaim exclusively. |
| 663 | * Return 0 if we grabbed it, non-zero otherwise. |
| 664 | */ |
| 665 | STATIC int |
| 666 | xfs_reclaim_inode_grab( |
| 667 | struct xfs_inode *ip, |
| 668 | int flags) |
| 669 | { |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 670 | ASSERT(rcu_read_lock_held()); |
| 671 | |
| 672 | /* quick check for stale RCU freed inode */ |
| 673 | if (!ip->i_ino) |
| 674 | return 1; |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 675 | |
| 676 | /* |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 677 | * do some unlocked checks first to avoid unnecessary lock traffic. |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 678 | * The first is a flush lock check, the second is a already in reclaim |
| 679 | * check. Only do these checks if we are not going to block on locks. |
| 680 | */ |
| 681 | if ((flags & SYNC_TRYLOCK) && |
| 682 | (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) { |
| 683 | return 1; |
| 684 | } |
| 685 | |
| 686 | /* |
| 687 | * The radix tree lock here protects a thread in xfs_iget from racing |
| 688 | * with us starting reclaim on the inode. Once we have the |
| 689 | * XFS_IRECLAIM flag set it will not touch us. |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 690 | * |
| 691 | * Due to RCU lookup, we may find inodes that have been freed and only |
| 692 | * have XFS_IRECLAIM set. Indeed, we may see reallocated inodes that |
| 693 | * aren't candidates for reclaim at all, so we must check the |
| 694 | * XFS_IRECLAIMABLE is set first before proceeding to reclaim. |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 695 | */ |
| 696 | spin_lock(&ip->i_flags_lock); |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 697 | if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) || |
| 698 | __xfs_iflags_test(ip, XFS_IRECLAIM)) { |
| 699 | /* not a reclaim candidate. */ |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 700 | spin_unlock(&ip->i_flags_lock); |
| 701 | return 1; |
| 702 | } |
| 703 | __xfs_iflags_set(ip, XFS_IRECLAIM); |
| 704 | spin_unlock(&ip->i_flags_lock); |
| 705 | return 0; |
| 706 | } |
| 707 | |
| 708 | /* |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 709 | * Inodes in different states need to be treated differently, and the return |
| 710 | * value of xfs_iflush is not sufficient to get this right. The following table |
| 711 | * lists the inode states and the reclaim actions necessary for non-blocking |
| 712 | * reclaim: |
| 713 | * |
| 714 | * |
| 715 | * inode state iflush ret required action |
| 716 | * --------------- ---------- --------------- |
| 717 | * bad - reclaim |
| 718 | * shutdown EIO unpin and reclaim |
| 719 | * clean, unpinned 0 reclaim |
| 720 | * stale, unpinned 0 reclaim |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 721 | * clean, pinned(*) 0 requeue |
| 722 | * stale, pinned EAGAIN requeue |
| 723 | * dirty, delwri ok 0 requeue |
| 724 | * dirty, delwri blocked EAGAIN requeue |
| 725 | * dirty, sync flush 0 reclaim |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 726 | * |
| 727 | * (*) dgc: I don't think the clean, pinned state is possible but it gets |
| 728 | * handled anyway given the order of checks implemented. |
| 729 | * |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 730 | * As can be seen from the table, the return value of xfs_iflush() is not |
| 731 | * sufficient to correctly decide the reclaim action here. The checks in |
| 732 | * xfs_iflush() might look like duplicates, but they are not. |
| 733 | * |
| 734 | * Also, because we get the flush lock first, we know that any inode that has |
| 735 | * been flushed delwri has had the flush completed by the time we check that |
| 736 | * the inode is clean. The clean inode check needs to be done before flushing |
| 737 | * the inode delwri otherwise we would loop forever requeuing clean inodes as |
| 738 | * we cannot tell apart a successful delwri flush and a clean inode from the |
| 739 | * return value of xfs_iflush(). |
| 740 | * |
| 741 | * Note that because the inode is flushed delayed write by background |
| 742 | * writeback, the flush lock may already be held here and waiting on it can |
| 743 | * result in very long latencies. Hence for sync reclaims, where we wait on the |
| 744 | * flush lock, the caller should push out delayed write inodes first before |
| 745 | * trying to reclaim them to minimise the amount of time spent waiting. For |
| 746 | * background relaim, we just requeue the inode for the next pass. |
| 747 | * |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 748 | * Hence the order of actions after gaining the locks should be: |
| 749 | * bad => reclaim |
| 750 | * shutdown => unpin and reclaim |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 751 | * pinned, delwri => requeue |
| 752 | * pinned, sync => unpin |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 753 | * stale => reclaim |
| 754 | * clean => reclaim |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 755 | * dirty, delwri => flush and requeue |
| 756 | * dirty, sync => flush, wait and reclaim |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 757 | */ |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 758 | STATIC int |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 759 | xfs_reclaim_inode( |
Dave Chinner | 75f3cb1 | 2009-06-08 15:35:14 +0200 | [diff] [blame] | 760 | struct xfs_inode *ip, |
| 761 | struct xfs_perag *pag, |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 762 | int sync_mode) |
David Chinner | 7a3be02 | 2008-10-30 17:37:37 +1100 | [diff] [blame] | 763 | { |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 764 | int error = 0; |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 765 | |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 766 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 767 | if (!xfs_iflock_nowait(ip)) { |
| 768 | if (!(sync_mode & SYNC_WAIT)) |
| 769 | goto out; |
| 770 | xfs_iflock(ip); |
| 771 | } |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 772 | |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 773 | if (is_bad_inode(VFS_I(ip))) |
| 774 | goto reclaim; |
| 775 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
| 776 | xfs_iunpin_wait(ip); |
| 777 | goto reclaim; |
| 778 | } |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 779 | if (xfs_ipincount(ip)) { |
| 780 | if (!(sync_mode & SYNC_WAIT)) { |
| 781 | xfs_ifunlock(ip); |
| 782 | goto out; |
| 783 | } |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 784 | xfs_iunpin_wait(ip); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 785 | } |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 786 | if (xfs_iflags_test(ip, XFS_ISTALE)) |
| 787 | goto reclaim; |
| 788 | if (xfs_inode_clean(ip)) |
| 789 | goto reclaim; |
| 790 | |
| 791 | /* Now we have an inode that needs flushing */ |
| 792 | error = xfs_iflush(ip, sync_mode); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 793 | if (sync_mode & SYNC_WAIT) { |
| 794 | xfs_iflock(ip); |
| 795 | goto reclaim; |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 796 | } |
| 797 | |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 798 | /* |
| 799 | * When we have to flush an inode but don't have SYNC_WAIT set, we |
| 800 | * flush the inode out using a delwri buffer and wait for the next |
| 801 | * call into reclaim to find it in a clean state instead of waiting for |
| 802 | * it now. We also don't return errors here - if the error is transient |
| 803 | * then the next reclaim pass will flush the inode, and if the error |
Dave Chinner | f1d486a | 2010-04-13 15:06:45 +1000 | [diff] [blame] | 804 | * is permanent then the next sync reclaim will reclaim the inode and |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 805 | * pass on the error. |
| 806 | */ |
Dave Chinner | f1d486a | 2010-04-13 15:06:45 +1000 | [diff] [blame] | 807 | if (error && error != EAGAIN && !XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 808 | xfs_fs_cmn_err(CE_WARN, ip->i_mount, |
| 809 | "inode 0x%llx background reclaim flush failed with %d", |
| 810 | (long long)ip->i_ino, error); |
| 811 | } |
| 812 | out: |
| 813 | xfs_iflags_clear(ip, XFS_IRECLAIM); |
| 814 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 815 | /* |
| 816 | * We could return EAGAIN here to make reclaim rescan the inode tree in |
| 817 | * a short while. However, this just burns CPU time scanning the tree |
| 818 | * waiting for IO to complete and xfssyncd never goes back to the idle |
| 819 | * state. Instead, return 0 to let the next scheduled background reclaim |
| 820 | * attempt to reclaim the inode again. |
| 821 | */ |
| 822 | return 0; |
| 823 | |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 824 | reclaim: |
| 825 | xfs_ifunlock(ip); |
Dave Chinner | c8e20be | 2010-01-10 23:51:45 +0000 | [diff] [blame] | 826 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Dave Chinner | 2f11fea | 2010-07-20 17:53:25 +1000 | [diff] [blame] | 827 | |
| 828 | XFS_STATS_INC(xs_ig_reclaims); |
| 829 | /* |
| 830 | * Remove the inode from the per-AG radix tree. |
| 831 | * |
| 832 | * Because radix_tree_delete won't complain even if the item was never |
| 833 | * added to the tree assert that it's been there before to catch |
| 834 | * problems with the inode life time early on. |
| 835 | */ |
| 836 | write_lock(&pag->pag_ici_lock); |
| 837 | if (!radix_tree_delete(&pag->pag_ici_root, |
| 838 | XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino))) |
| 839 | ASSERT(0); |
Johannes Weiner | 081003f | 2010-10-01 07:43:54 +0000 | [diff] [blame] | 840 | __xfs_inode_clear_reclaim(pag, ip); |
Dave Chinner | 2f11fea | 2010-07-20 17:53:25 +1000 | [diff] [blame] | 841 | write_unlock(&pag->pag_ici_lock); |
| 842 | |
| 843 | /* |
| 844 | * Here we do an (almost) spurious inode lock in order to coordinate |
| 845 | * with inode cache radix tree lookups. This is because the lookup |
| 846 | * can reference the inodes in the cache without taking references. |
| 847 | * |
| 848 | * We make that OK here by ensuring that we wait until the inode is |
| 849 | * unlocked after the lookup before we go ahead and free it. We get |
| 850 | * both the ilock and the iolock because the code may need to drop the |
| 851 | * ilock one but will still hold the iolock. |
| 852 | */ |
| 853 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
| 854 | xfs_qm_dqdetach(ip); |
| 855 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
| 856 | |
| 857 | xfs_inode_free(ip); |
Dave Chinner | c854363 | 2010-02-06 12:39:36 +1100 | [diff] [blame] | 858 | return error; |
| 859 | |
David Chinner | 7a3be02 | 2008-10-30 17:37:37 +1100 | [diff] [blame] | 860 | } |
| 861 | |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 862 | /* |
| 863 | * Walk the AGs and reclaim the inodes in them. Even if the filesystem is |
| 864 | * corrupted, we still want to try to reclaim all the inodes. If we don't, |
| 865 | * then a shut down during filesystem unmount reclaim walk leak all the |
| 866 | * unreclaimed inodes. |
| 867 | */ |
| 868 | int |
| 869 | xfs_reclaim_inodes_ag( |
| 870 | struct xfs_mount *mp, |
| 871 | int flags, |
| 872 | int *nr_to_scan) |
| 873 | { |
| 874 | struct xfs_perag *pag; |
| 875 | int error = 0; |
| 876 | int last_error = 0; |
| 877 | xfs_agnumber_t ag; |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 878 | int trylock = flags & SYNC_TRYLOCK; |
| 879 | int skipped; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 880 | |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 881 | restart: |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 882 | ag = 0; |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 883 | skipped = 0; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 884 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
| 885 | unsigned long first_index = 0; |
| 886 | int done = 0; |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 887 | int nr_found = 0; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 888 | |
| 889 | ag = pag->pag_agno + 1; |
| 890 | |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 891 | if (trylock) { |
| 892 | if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) { |
| 893 | skipped++; |
Dave Chinner | f83282a | 2010-11-08 08:55:04 +0000 | [diff] [blame] | 894 | xfs_perag_put(pag); |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 895 | continue; |
| 896 | } |
| 897 | first_index = pag->pag_ici_reclaim_cursor; |
| 898 | } else |
| 899 | mutex_lock(&pag->pag_ici_reclaim_lock); |
| 900 | |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 901 | do { |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 902 | struct xfs_inode *batch[XFS_LOOKUP_BATCH]; |
| 903 | int i; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 904 | |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 905 | rcu_read_lock(); |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 906 | nr_found = radix_tree_gang_lookup_tag( |
| 907 | &pag->pag_ici_root, |
| 908 | (void **)batch, first_index, |
| 909 | XFS_LOOKUP_BATCH, |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 910 | XFS_ICI_RECLAIM_TAG); |
| 911 | if (!nr_found) { |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 912 | rcu_read_unlock(); |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 913 | break; |
| 914 | } |
| 915 | |
| 916 | /* |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 917 | * Grab the inodes before we drop the lock. if we found |
| 918 | * nothing, nr == 0 and the loop will be skipped. |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 919 | */ |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 920 | for (i = 0; i < nr_found; i++) { |
| 921 | struct xfs_inode *ip = batch[i]; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 922 | |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 923 | if (done || xfs_reclaim_inode_grab(ip, flags)) |
| 924 | batch[i] = NULL; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 925 | |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 926 | /* |
| 927 | * Update the index for the next lookup. Catch |
| 928 | * overflows into the next AG range which can |
| 929 | * occur if we have inodes in the last block of |
| 930 | * the AG and we are currently pointing to the |
| 931 | * last inode. |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 932 | * |
| 933 | * Because we may see inodes that are from the |
| 934 | * wrong AG due to RCU freeing and |
| 935 | * reallocation, only update the index if it |
| 936 | * lies in this AG. It was a race that lead us |
| 937 | * to see this inode, so another lookup from |
| 938 | * the same index will not find it again. |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 939 | */ |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 940 | if (XFS_INO_TO_AGNO(mp, ip->i_ino) != |
| 941 | pag->pag_agno) |
| 942 | continue; |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 943 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); |
| 944 | if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) |
| 945 | done = 1; |
| 946 | } |
| 947 | |
| 948 | /* unlock now we've grabbed the inodes. */ |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame^] | 949 | rcu_read_unlock(); |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 950 | |
| 951 | for (i = 0; i < nr_found; i++) { |
| 952 | if (!batch[i]) |
| 953 | continue; |
| 954 | error = xfs_reclaim_inode(batch[i], pag, flags); |
| 955 | if (error && last_error != EFSCORRUPTED) |
| 956 | last_error = error; |
| 957 | } |
| 958 | |
| 959 | *nr_to_scan -= XFS_LOOKUP_BATCH; |
| 960 | |
| 961 | } while (nr_found && !done && *nr_to_scan > 0); |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 962 | |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 963 | if (trylock && !done) |
| 964 | pag->pag_ici_reclaim_cursor = first_index; |
| 965 | else |
| 966 | pag->pag_ici_reclaim_cursor = 0; |
| 967 | mutex_unlock(&pag->pag_ici_reclaim_lock); |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 968 | xfs_perag_put(pag); |
| 969 | } |
Dave Chinner | 69b491c | 2010-09-27 11:09:51 +1000 | [diff] [blame] | 970 | |
| 971 | /* |
| 972 | * if we skipped any AG, and we still have scan count remaining, do |
| 973 | * another pass this time using blocking reclaim semantics (i.e |
| 974 | * waiting on the reclaim locks and ignoring the reclaim cursors). This |
| 975 | * ensure that when we get more reclaimers than AGs we block rather |
| 976 | * than spin trying to execute reclaim. |
| 977 | */ |
| 978 | if (trylock && skipped && *nr_to_scan > 0) { |
| 979 | trylock = 0; |
| 980 | goto restart; |
| 981 | } |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 982 | return XFS_ERROR(last_error); |
| 983 | } |
| 984 | |
David Chinner | fce08f2 | 2008-10-30 17:37:03 +1100 | [diff] [blame] | 985 | int |
David Chinner | 1dc3318 | 2008-10-30 17:37:15 +1100 | [diff] [blame] | 986 | xfs_reclaim_inodes( |
David Chinner | fce08f2 | 2008-10-30 17:37:03 +1100 | [diff] [blame] | 987 | xfs_mount_t *mp, |
David Chinner | fce08f2 | 2008-10-30 17:37:03 +1100 | [diff] [blame] | 988 | int mode) |
| 989 | { |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 990 | int nr_to_scan = INT_MAX; |
| 991 | |
| 992 | return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan); |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 993 | } |
| 994 | |
| 995 | /* |
| 996 | * Shrinker infrastructure. |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 997 | */ |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 998 | static int |
| 999 | xfs_reclaim_inode_shrink( |
Dave Chinner | 7f8275d | 2010-07-19 14:56:17 +1000 | [diff] [blame] | 1000 | struct shrinker *shrink, |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1001 | int nr_to_scan, |
| 1002 | gfp_t gfp_mask) |
| 1003 | { |
| 1004 | struct xfs_mount *mp; |
| 1005 | struct xfs_perag *pag; |
| 1006 | xfs_agnumber_t ag; |
Dave Chinner | 16fd536 | 2010-07-20 09:43:39 +1000 | [diff] [blame] | 1007 | int reclaimable; |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1008 | |
Dave Chinner | 70e60ce | 2010-07-20 08:07:02 +1000 | [diff] [blame] | 1009 | mp = container_of(shrink, struct xfs_mount, m_inode_shrink); |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1010 | if (nr_to_scan) { |
| 1011 | if (!(gfp_mask & __GFP_FS)) |
| 1012 | return -1; |
| 1013 | |
Dave Chinner | e3a20c0 | 2010-09-24 19:51:50 +1000 | [diff] [blame] | 1014 | xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK, &nr_to_scan); |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 1015 | /* terminate if we don't exhaust the scan */ |
Dave Chinner | 70e60ce | 2010-07-20 08:07:02 +1000 | [diff] [blame] | 1016 | if (nr_to_scan > 0) |
| 1017 | return -1; |
| 1018 | } |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1019 | |
Dave Chinner | 16fd536 | 2010-07-20 09:43:39 +1000 | [diff] [blame] | 1020 | reclaimable = 0; |
| 1021 | ag = 0; |
Dave Chinner | 65d0f20 | 2010-09-24 18:40:15 +1000 | [diff] [blame] | 1022 | while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { |
| 1023 | ag = pag->pag_agno + 1; |
Dave Chinner | 70e60ce | 2010-07-20 08:07:02 +1000 | [diff] [blame] | 1024 | reclaimable += pag->pag_ici_reclaimable; |
| 1025 | xfs_perag_put(pag); |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1026 | } |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1027 | return reclaimable; |
| 1028 | } |
| 1029 | |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1030 | void |
| 1031 | xfs_inode_shrinker_register( |
| 1032 | struct xfs_mount *mp) |
| 1033 | { |
Dave Chinner | 70e60ce | 2010-07-20 08:07:02 +1000 | [diff] [blame] | 1034 | mp->m_inode_shrink.shrink = xfs_reclaim_inode_shrink; |
| 1035 | mp->m_inode_shrink.seeks = DEFAULT_SEEKS; |
| 1036 | register_shrinker(&mp->m_inode_shrink); |
Dave Chinner | 9bf729c | 2010-04-29 09:55:50 +1000 | [diff] [blame] | 1037 | } |
| 1038 | |
| 1039 | void |
| 1040 | xfs_inode_shrinker_unregister( |
| 1041 | struct xfs_mount *mp) |
| 1042 | { |
Dave Chinner | 70e60ce | 2010-07-20 08:07:02 +1000 | [diff] [blame] | 1043 | unregister_shrinker(&mp->m_inode_shrink); |
David Chinner | fce08f2 | 2008-10-30 17:37:03 +1100 | [diff] [blame] | 1044 | } |