Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
| 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
| 3 | * |
| 4 | * journal.c |
| 5 | * |
| 6 | * Defines functions of journalling api |
| 7 | * |
| 8 | * Copyright (C) 2003, 2004 Oracle. All rights reserved. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU General Public |
| 12 | * License as published by the Free Software Foundation; either |
| 13 | * version 2 of the License, or (at your option) any later version. |
| 14 | * |
| 15 | * This program is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 18 | * General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU General Public |
| 21 | * License along with this program; if not, write to the |
| 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 23 | * Boston, MA 021110-1307, USA. |
| 24 | */ |
| 25 | |
| 26 | #include <linux/fs.h> |
| 27 | #include <linux/types.h> |
| 28 | #include <linux/slab.h> |
| 29 | #include <linux/highmem.h> |
| 30 | #include <linux/kthread.h> |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 31 | #include <linux/time.h> |
| 32 | #include <linux/random.h> |
Joseph Qi | 55b465b | 2014-06-04 16:06:09 -0700 | [diff] [blame] | 33 | #include <linux/delay.h> |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 34 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 35 | #include <cluster/masklog.h> |
| 36 | |
| 37 | #include "ocfs2.h" |
| 38 | |
| 39 | #include "alloc.h" |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 40 | #include "blockcheck.h" |
Mark Fasheh | 316f4b9 | 2007-09-07 18:21:26 -0700 | [diff] [blame] | 41 | #include "dir.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 42 | #include "dlmglue.h" |
| 43 | #include "extent_map.h" |
| 44 | #include "heartbeat.h" |
| 45 | #include "inode.h" |
| 46 | #include "journal.h" |
| 47 | #include "localalloc.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 48 | #include "slot_map.h" |
| 49 | #include "super.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 50 | #include "sysfile.h" |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 51 | #include "uptodate.h" |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 52 | #include "quota.h" |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 53 | #include "file.h" |
| 54 | #include "namei.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 55 | |
| 56 | #include "buffer_head_io.h" |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 57 | #include "ocfs2_trace.h" |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 58 | |
Ingo Molnar | 34af946 | 2006-06-27 02:53:55 -0700 | [diff] [blame] | 59 | DEFINE_SPINLOCK(trans_inc_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 60 | |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 61 | #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000 |
| 62 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 63 | static int ocfs2_force_read_journal(struct inode *inode); |
| 64 | static int ocfs2_recover_node(struct ocfs2_super *osb, |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 65 | int node_num, int slot_num); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 66 | static int __ocfs2_recovery_thread(void *arg); |
| 67 | static int ocfs2_commit_cache(struct ocfs2_super *osb); |
Jan Kara | 19ece54 | 2008-08-21 20:13:17 +0200 | [diff] [blame] | 68 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 69 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 70 | int dirty, int replayed); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 71 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
| 72 | int slot_num); |
| 73 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 74 | int slot, |
| 75 | enum ocfs2_orphan_reco_type orphan_reco_type); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 76 | static int ocfs2_commit_thread(void *arg); |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 77 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
| 78 | int slot_num, |
| 79 | struct ocfs2_dinode *la_dinode, |
| 80 | struct ocfs2_dinode *tl_dinode, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 81 | struct ocfs2_quota_recovery *qrec, |
| 82 | enum ocfs2_orphan_reco_type orphan_reco_type); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 83 | |
Jan Kara | 19ece54 | 2008-08-21 20:13:17 +0200 | [diff] [blame] | 84 | static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb) |
| 85 | { |
| 86 | return __ocfs2_wait_on_mount(osb, 0); |
| 87 | } |
| 88 | |
| 89 | static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb) |
| 90 | { |
| 91 | return __ocfs2_wait_on_mount(osb, 1); |
| 92 | } |
| 93 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 94 | /* |
| 95 | * This replay_map is to track online/offline slots, so we could recover |
| 96 | * offline slots during recovery and mount |
| 97 | */ |
| 98 | |
| 99 | enum ocfs2_replay_state { |
| 100 | REPLAY_UNNEEDED = 0, /* Replay is not needed, so ignore this map */ |
| 101 | REPLAY_NEEDED, /* Replay slots marked in rm_replay_slots */ |
| 102 | REPLAY_DONE /* Replay was already queued */ |
| 103 | }; |
| 104 | |
| 105 | struct ocfs2_replay_map { |
| 106 | unsigned int rm_slots; |
| 107 | enum ocfs2_replay_state rm_state; |
| 108 | unsigned char rm_replay_slots[0]; |
| 109 | }; |
| 110 | |
Joseph Qi | b519ea6 | 2015-06-24 16:55:34 -0700 | [diff] [blame] | 111 | static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state) |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 112 | { |
| 113 | if (!osb->replay_map) |
| 114 | return; |
| 115 | |
| 116 | /* If we've already queued the replay, we don't have any more to do */ |
| 117 | if (osb->replay_map->rm_state == REPLAY_DONE) |
| 118 | return; |
| 119 | |
| 120 | osb->replay_map->rm_state = state; |
| 121 | } |
| 122 | |
| 123 | int ocfs2_compute_replay_slots(struct ocfs2_super *osb) |
| 124 | { |
| 125 | struct ocfs2_replay_map *replay_map; |
| 126 | int i, node_num; |
| 127 | |
| 128 | /* If replay map is already set, we don't do it again */ |
| 129 | if (osb->replay_map) |
| 130 | return 0; |
| 131 | |
| 132 | replay_map = kzalloc(sizeof(struct ocfs2_replay_map) + |
| 133 | (osb->max_slots * sizeof(char)), GFP_KERNEL); |
| 134 | |
| 135 | if (!replay_map) { |
| 136 | mlog_errno(-ENOMEM); |
| 137 | return -ENOMEM; |
| 138 | } |
| 139 | |
| 140 | spin_lock(&osb->osb_lock); |
| 141 | |
| 142 | replay_map->rm_slots = osb->max_slots; |
| 143 | replay_map->rm_state = REPLAY_UNNEEDED; |
| 144 | |
| 145 | /* set rm_replay_slots for offline slot(s) */ |
| 146 | for (i = 0; i < replay_map->rm_slots; i++) { |
| 147 | if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT) |
| 148 | replay_map->rm_replay_slots[i] = 1; |
| 149 | } |
| 150 | |
| 151 | osb->replay_map = replay_map; |
| 152 | spin_unlock(&osb->osb_lock); |
| 153 | return 0; |
| 154 | } |
| 155 | |
Joseph Qi | b519ea6 | 2015-06-24 16:55:34 -0700 | [diff] [blame] | 156 | static void ocfs2_queue_replay_slots(struct ocfs2_super *osb, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 157 | enum ocfs2_orphan_reco_type orphan_reco_type) |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 158 | { |
| 159 | struct ocfs2_replay_map *replay_map = osb->replay_map; |
| 160 | int i; |
| 161 | |
| 162 | if (!replay_map) |
| 163 | return; |
| 164 | |
| 165 | if (replay_map->rm_state != REPLAY_NEEDED) |
| 166 | return; |
| 167 | |
| 168 | for (i = 0; i < replay_map->rm_slots; i++) |
| 169 | if (replay_map->rm_replay_slots[i]) |
| 170 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 171 | NULL, NULL, |
| 172 | orphan_reco_type); |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 173 | replay_map->rm_state = REPLAY_DONE; |
| 174 | } |
| 175 | |
Joseph Qi | b519ea6 | 2015-06-24 16:55:34 -0700 | [diff] [blame] | 176 | static void ocfs2_free_replay_slots(struct ocfs2_super *osb) |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 177 | { |
| 178 | struct ocfs2_replay_map *replay_map = osb->replay_map; |
| 179 | |
| 180 | if (!osb->replay_map) |
| 181 | return; |
| 182 | |
| 183 | kfree(replay_map); |
| 184 | osb->replay_map = NULL; |
| 185 | } |
| 186 | |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 187 | int ocfs2_recovery_init(struct ocfs2_super *osb) |
| 188 | { |
| 189 | struct ocfs2_recovery_map *rm; |
| 190 | |
| 191 | mutex_init(&osb->recovery_lock); |
| 192 | osb->disable_recovery = 0; |
| 193 | osb->recovery_thread_task = NULL; |
| 194 | init_waitqueue_head(&osb->recovery_event); |
| 195 | |
| 196 | rm = kzalloc(sizeof(struct ocfs2_recovery_map) + |
| 197 | osb->max_slots * sizeof(unsigned int), |
| 198 | GFP_KERNEL); |
| 199 | if (!rm) { |
| 200 | mlog_errno(-ENOMEM); |
| 201 | return -ENOMEM; |
| 202 | } |
| 203 | |
| 204 | rm->rm_entries = (unsigned int *)((char *)rm + |
| 205 | sizeof(struct ocfs2_recovery_map)); |
| 206 | osb->recovery_map = rm; |
| 207 | |
| 208 | return 0; |
| 209 | } |
| 210 | |
| 211 | /* we can't grab the goofy sem lock from inside wait_event, so we use |
| 212 | * memory barriers to make sure that we'll see the null task before |
| 213 | * being woken up */ |
| 214 | static int ocfs2_recovery_thread_running(struct ocfs2_super *osb) |
| 215 | { |
| 216 | mb(); |
| 217 | return osb->recovery_thread_task != NULL; |
| 218 | } |
| 219 | |
| 220 | void ocfs2_recovery_exit(struct ocfs2_super *osb) |
| 221 | { |
| 222 | struct ocfs2_recovery_map *rm; |
| 223 | |
| 224 | /* disable any new recovery threads and wait for any currently |
| 225 | * running ones to exit. Do this before setting the vol_state. */ |
| 226 | mutex_lock(&osb->recovery_lock); |
| 227 | osb->disable_recovery = 1; |
| 228 | mutex_unlock(&osb->recovery_lock); |
| 229 | wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb)); |
| 230 | |
| 231 | /* At this point, we know that no more recovery threads can be |
| 232 | * launched, so wait for any recovery completion work to |
| 233 | * complete. */ |
jiangyiwen | 35ddf78 | 2016-03-25 14:21:32 -0700 | [diff] [blame] | 234 | flush_workqueue(osb->ocfs2_wq); |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 235 | |
| 236 | /* |
| 237 | * Now that recovery is shut down, and the osb is about to be |
| 238 | * freed, the osb_lock is not taken here. |
| 239 | */ |
| 240 | rm = osb->recovery_map; |
| 241 | /* XXX: Should we bug if there are dirty entries? */ |
| 242 | |
| 243 | kfree(rm); |
| 244 | } |
| 245 | |
| 246 | static int __ocfs2_recovery_map_test(struct ocfs2_super *osb, |
| 247 | unsigned int node_num) |
| 248 | { |
| 249 | int i; |
| 250 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
| 251 | |
| 252 | assert_spin_locked(&osb->osb_lock); |
| 253 | |
| 254 | for (i = 0; i < rm->rm_used; i++) { |
| 255 | if (rm->rm_entries[i] == node_num) |
| 256 | return 1; |
| 257 | } |
| 258 | |
| 259 | return 0; |
| 260 | } |
| 261 | |
| 262 | /* Behaves like test-and-set. Returns the previous value */ |
| 263 | static int ocfs2_recovery_map_set(struct ocfs2_super *osb, |
| 264 | unsigned int node_num) |
| 265 | { |
| 266 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
| 267 | |
| 268 | spin_lock(&osb->osb_lock); |
| 269 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
| 270 | spin_unlock(&osb->osb_lock); |
| 271 | return 1; |
| 272 | } |
| 273 | |
| 274 | /* XXX: Can this be exploited? Not from o2dlm... */ |
| 275 | BUG_ON(rm->rm_used >= osb->max_slots); |
| 276 | |
| 277 | rm->rm_entries[rm->rm_used] = node_num; |
| 278 | rm->rm_used++; |
| 279 | spin_unlock(&osb->osb_lock); |
| 280 | |
| 281 | return 0; |
| 282 | } |
| 283 | |
| 284 | static void ocfs2_recovery_map_clear(struct ocfs2_super *osb, |
| 285 | unsigned int node_num) |
| 286 | { |
| 287 | int i; |
| 288 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
| 289 | |
| 290 | spin_lock(&osb->osb_lock); |
| 291 | |
| 292 | for (i = 0; i < rm->rm_used; i++) { |
| 293 | if (rm->rm_entries[i] == node_num) |
| 294 | break; |
| 295 | } |
| 296 | |
| 297 | if (i < rm->rm_used) { |
| 298 | /* XXX: be careful with the pointer math */ |
| 299 | memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]), |
| 300 | (rm->rm_used - i - 1) * sizeof(unsigned int)); |
| 301 | rm->rm_used--; |
| 302 | } |
| 303 | |
| 304 | spin_unlock(&osb->osb_lock); |
| 305 | } |
| 306 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 307 | static int ocfs2_commit_cache(struct ocfs2_super *osb) |
| 308 | { |
| 309 | int status = 0; |
| 310 | unsigned int flushed; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 311 | struct ocfs2_journal *journal = NULL; |
| 312 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 313 | journal = osb->journal; |
| 314 | |
| 315 | /* Flush all pending commits and checkpoint the journal. */ |
| 316 | down_write(&journal->j_trans_barrier); |
| 317 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 318 | flushed = atomic_read(&journal->j_num_trans); |
| 319 | trace_ocfs2_commit_cache_begin(flushed); |
| 320 | if (flushed == 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 321 | up_write(&journal->j_trans_barrier); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 322 | goto finally; |
| 323 | } |
| 324 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 325 | jbd2_journal_lock_updates(journal->j_journal); |
| 326 | status = jbd2_journal_flush(journal->j_journal); |
| 327 | jbd2_journal_unlock_updates(journal->j_journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 328 | if (status < 0) { |
| 329 | up_write(&journal->j_trans_barrier); |
| 330 | mlog_errno(status); |
| 331 | goto finally; |
| 332 | } |
| 333 | |
Tao Ma | f9c57ad | 2010-08-02 11:02:13 +0800 | [diff] [blame] | 334 | ocfs2_inc_trans_id(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 335 | |
| 336 | flushed = atomic_read(&journal->j_num_trans); |
| 337 | atomic_set(&journal->j_num_trans, 0); |
| 338 | up_write(&journal->j_trans_barrier); |
| 339 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 340 | trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 341 | |
Mark Fasheh | 34d024f | 2007-09-24 15:56:19 -0700 | [diff] [blame] | 342 | ocfs2_wake_downconvert_thread(osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 343 | wake_up(&journal->j_checkpointed); |
| 344 | finally: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 345 | return status; |
| 346 | } |
| 347 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 348 | handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 349 | { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 350 | journal_t *journal = osb->journal->j_journal; |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 351 | handle_t *handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 352 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 353 | BUG_ON(!osb || !osb->journal->j_journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 354 | |
Mark Fasheh | 65eff9c | 2006-10-09 17:26:22 -0700 | [diff] [blame] | 355 | if (ocfs2_is_hard_readonly(osb)) |
| 356 | return ERR_PTR(-EROFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 357 | |
| 358 | BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE); |
| 359 | BUG_ON(max_buffs <= 0); |
| 360 | |
Jan Kara | 90e86a6 | 2008-08-27 22:30:28 +0200 | [diff] [blame] | 361 | /* Nested transaction? Just return the handle... */ |
| 362 | if (journal_current_handle()) |
| 363 | return jbd2_journal_start(journal, max_buffs); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 364 | |
Jan Kara | fef6925 | 2012-06-12 16:20:40 +0200 | [diff] [blame] | 365 | sb_start_intwrite(osb->sb); |
| 366 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 367 | down_read(&osb->journal->j_trans_barrier); |
| 368 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 369 | handle = jbd2_journal_start(journal, max_buffs); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 370 | if (IS_ERR(handle)) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 371 | up_read(&osb->journal->j_trans_barrier); |
Jan Kara | fef6925 | 2012-06-12 16:20:40 +0200 | [diff] [blame] | 372 | sb_end_intwrite(osb->sb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 373 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 374 | mlog_errno(PTR_ERR(handle)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 375 | |
| 376 | if (is_journal_aborted(journal)) { |
Joe Perches | 7ecef14 | 2015-09-04 15:44:51 -0700 | [diff] [blame] | 377 | ocfs2_abort(osb->sb, "Detected aborted journal\n"); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 378 | handle = ERR_PTR(-EROFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 379 | } |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 380 | } else { |
| 381 | if (!ocfs2_mount_local(osb)) |
| 382 | atomic_inc(&(osb->journal->j_num_trans)); |
| 383 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 384 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 385 | return handle; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 386 | } |
| 387 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 388 | int ocfs2_commit_trans(struct ocfs2_super *osb, |
| 389 | handle_t *handle) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 390 | { |
Jan Kara | 90e86a6 | 2008-08-27 22:30:28 +0200 | [diff] [blame] | 391 | int ret, nested; |
Mark Fasheh | 02dc1af | 2006-10-09 16:48:10 -0700 | [diff] [blame] | 392 | struct ocfs2_journal *journal = osb->journal; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 393 | |
| 394 | BUG_ON(!handle); |
| 395 | |
Jan Kara | 90e86a6 | 2008-08-27 22:30:28 +0200 | [diff] [blame] | 396 | nested = handle->h_ref > 1; |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 397 | ret = jbd2_journal_stop(handle); |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 398 | if (ret < 0) |
| 399 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 400 | |
Jan Kara | fef6925 | 2012-06-12 16:20:40 +0200 | [diff] [blame] | 401 | if (!nested) { |
Jan Kara | 90e86a6 | 2008-08-27 22:30:28 +0200 | [diff] [blame] | 402 | up_read(&journal->j_trans_barrier); |
Jan Kara | fef6925 | 2012-06-12 16:20:40 +0200 | [diff] [blame] | 403 | sb_end_intwrite(osb->sb); |
| 404 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 405 | |
Mark Fasheh | 1fabe14 | 2006-10-09 18:11:45 -0700 | [diff] [blame] | 406 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 407 | } |
| 408 | |
| 409 | /* |
Tao Ma | c901fb0 | 2010-04-26 14:34:57 +0800 | [diff] [blame] | 410 | * 'nblocks' is what you want to add to the current transaction. |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 411 | * |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 412 | * This might call jbd2_journal_restart() which will commit dirty buffers |
Mark Fasheh | e8aed34 | 2007-12-03 16:43:01 -0800 | [diff] [blame] | 413 | * and then restart the transaction. Before calling |
| 414 | * ocfs2_extend_trans(), any changed blocks should have been |
| 415 | * dirtied. After calling it, all blocks which need to be changed must |
| 416 | * go through another set of journal_access/journal_dirty calls. |
| 417 | * |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 418 | * WARNING: This will not release any semaphores or disk locks taken |
| 419 | * during the transaction, so make sure they were taken *before* |
| 420 | * start_trans or we'll have ordering deadlocks. |
| 421 | * |
| 422 | * WARNING2: Note that we do *not* drop j_trans_barrier here. This is |
| 423 | * good because transaction ids haven't yet been recorded on the |
| 424 | * cluster locks associated with this handle. |
| 425 | */ |
Mark Fasheh | 1fc5814 | 2006-10-05 14:15:36 -0700 | [diff] [blame] | 426 | int ocfs2_extend_trans(handle_t *handle, int nblocks) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 427 | { |
Tao Ma | c901fb0 | 2010-04-26 14:34:57 +0800 | [diff] [blame] | 428 | int status, old_nblocks; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 429 | |
| 430 | BUG_ON(!handle); |
Tao Ma | c901fb0 | 2010-04-26 14:34:57 +0800 | [diff] [blame] | 431 | BUG_ON(nblocks < 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 432 | |
Tao Ma | c901fb0 | 2010-04-26 14:34:57 +0800 | [diff] [blame] | 433 | if (!nblocks) |
| 434 | return 0; |
| 435 | |
| 436 | old_nblocks = handle->h_buffer_credits; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 437 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 438 | trace_ocfs2_extend_trans(old_nblocks, nblocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 439 | |
Joel Becker | e407e39 | 2008-06-12 22:35:39 -0700 | [diff] [blame] | 440 | #ifdef CONFIG_OCFS2_DEBUG_FS |
Mark Fasheh | 0879c58 | 2007-12-03 16:42:19 -0800 | [diff] [blame] | 441 | status = 1; |
| 442 | #else |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 443 | status = jbd2_journal_extend(handle, nblocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 444 | if (status < 0) { |
| 445 | mlog_errno(status); |
| 446 | goto bail; |
| 447 | } |
Mark Fasheh | 0879c58 | 2007-12-03 16:42:19 -0800 | [diff] [blame] | 448 | #endif |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 449 | |
| 450 | if (status > 0) { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 451 | trace_ocfs2_extend_trans_restart(old_nblocks + nblocks); |
Tao Ma | c901fb0 | 2010-04-26 14:34:57 +0800 | [diff] [blame] | 452 | status = jbd2_journal_restart(handle, |
| 453 | old_nblocks + nblocks); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 454 | if (status < 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 455 | mlog_errno(status); |
| 456 | goto bail; |
| 457 | } |
Mark Fasheh | 01ddf1e | 2006-10-05 13:54:39 -0700 | [diff] [blame] | 458 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 459 | |
| 460 | status = 0; |
| 461 | bail: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 462 | return status; |
| 463 | } |
| 464 | |
Younger Liu | 2b1e55c | 2013-09-11 14:19:44 -0700 | [diff] [blame] | 465 | /* |
| 466 | * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA. |
| 467 | * If that fails, restart the transaction & regain write access for the |
| 468 | * buffer head which is used for metadata modifications. |
| 469 | * Taken from Ext4: extend_or_restart_transaction() |
| 470 | */ |
| 471 | int ocfs2_allocate_extend_trans(handle_t *handle, int thresh) |
| 472 | { |
| 473 | int status, old_nblks; |
| 474 | |
| 475 | BUG_ON(!handle); |
| 476 | |
| 477 | old_nblks = handle->h_buffer_credits; |
| 478 | trace_ocfs2_allocate_extend_trans(old_nblks, thresh); |
| 479 | |
| 480 | if (old_nblks < thresh) |
| 481 | return 0; |
| 482 | |
| 483 | status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA); |
| 484 | if (status < 0) { |
| 485 | mlog_errno(status); |
| 486 | goto bail; |
| 487 | } |
| 488 | |
| 489 | if (status > 0) { |
| 490 | status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA); |
| 491 | if (status < 0) |
| 492 | mlog_errno(status); |
| 493 | } |
| 494 | |
| 495 | bail: |
| 496 | return status; |
| 497 | } |
| 498 | |
| 499 | |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 500 | struct ocfs2_triggers { |
| 501 | struct jbd2_buffer_trigger_type ot_triggers; |
| 502 | int ot_offset; |
| 503 | }; |
| 504 | |
| 505 | static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers) |
| 506 | { |
| 507 | return container_of(triggers, struct ocfs2_triggers, ot_triggers); |
| 508 | } |
| 509 | |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 510 | static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 511 | struct buffer_head *bh, |
| 512 | void *data, size_t size) |
| 513 | { |
| 514 | struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers); |
| 515 | |
| 516 | /* |
| 517 | * We aren't guaranteed to have the superblock here, so we |
| 518 | * must unconditionally compute the ecc data. |
| 519 | * __ocfs2_journal_access() will only set the triggers if |
| 520 | * metaecc is enabled. |
| 521 | */ |
| 522 | ocfs2_block_check_compute(data, size, data + ot->ot_offset); |
| 523 | } |
| 524 | |
| 525 | /* |
| 526 | * Quota blocks have their own trigger because the struct ocfs2_block_check |
| 527 | * offset depends on the blocksize. |
| 528 | */ |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 529 | static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 530 | struct buffer_head *bh, |
| 531 | void *data, size_t size) |
| 532 | { |
| 533 | struct ocfs2_disk_dqtrailer *dqt = |
| 534 | ocfs2_block_dqtrailer(size, data); |
| 535 | |
| 536 | /* |
| 537 | * We aren't guaranteed to have the superblock here, so we |
| 538 | * must unconditionally compute the ecc data. |
| 539 | * __ocfs2_journal_access() will only set the triggers if |
| 540 | * metaecc is enabled. |
| 541 | */ |
| 542 | ocfs2_block_check_compute(data, size, &dqt->dq_check); |
| 543 | } |
| 544 | |
Joel Becker | c175a51 | 2008-12-10 17:58:22 -0800 | [diff] [blame] | 545 | /* |
| 546 | * Directory blocks also have their own trigger because the |
| 547 | * struct ocfs2_block_check offset depends on the blocksize. |
| 548 | */ |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 549 | static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers, |
Joel Becker | c175a51 | 2008-12-10 17:58:22 -0800 | [diff] [blame] | 550 | struct buffer_head *bh, |
| 551 | void *data, size_t size) |
| 552 | { |
| 553 | struct ocfs2_dir_block_trailer *trailer = |
| 554 | ocfs2_dir_trailer_from_size(size, data); |
| 555 | |
| 556 | /* |
| 557 | * We aren't guaranteed to have the superblock here, so we |
| 558 | * must unconditionally compute the ecc data. |
| 559 | * __ocfs2_journal_access() will only set the triggers if |
| 560 | * metaecc is enabled. |
| 561 | */ |
| 562 | ocfs2_block_check_compute(data, size, &trailer->db_check); |
| 563 | } |
| 564 | |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 565 | static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers, |
| 566 | struct buffer_head *bh) |
| 567 | { |
| 568 | mlog(ML_ERROR, |
| 569 | "ocfs2_abort_trigger called by JBD2. bh = 0x%lx, " |
| 570 | "bh->b_blocknr = %llu\n", |
| 571 | (unsigned long)bh, |
| 572 | (unsigned long long)bh->b_blocknr); |
| 573 | |
Xue jiufei | 74e364a | 2015-06-24 16:55:20 -0700 | [diff] [blame] | 574 | ocfs2_error(bh->b_bdev->bd_super, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 575 | "JBD2 has aborted our journal, ocfs2 cannot continue\n"); |
| 576 | } |
| 577 | |
| 578 | static struct ocfs2_triggers di_triggers = { |
| 579 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 580 | .t_frozen = ocfs2_frozen_trigger, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 581 | .t_abort = ocfs2_abort_trigger, |
| 582 | }, |
| 583 | .ot_offset = offsetof(struct ocfs2_dinode, i_check), |
| 584 | }; |
| 585 | |
| 586 | static struct ocfs2_triggers eb_triggers = { |
| 587 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 588 | .t_frozen = ocfs2_frozen_trigger, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 589 | .t_abort = ocfs2_abort_trigger, |
| 590 | }, |
| 591 | .ot_offset = offsetof(struct ocfs2_extent_block, h_check), |
| 592 | }; |
| 593 | |
Tao Ma | 93c9708 | 2009-08-18 11:19:20 +0800 | [diff] [blame] | 594 | static struct ocfs2_triggers rb_triggers = { |
| 595 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 596 | .t_frozen = ocfs2_frozen_trigger, |
Tao Ma | 93c9708 | 2009-08-18 11:19:20 +0800 | [diff] [blame] | 597 | .t_abort = ocfs2_abort_trigger, |
| 598 | }, |
| 599 | .ot_offset = offsetof(struct ocfs2_refcount_block, rf_check), |
| 600 | }; |
| 601 | |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 602 | static struct ocfs2_triggers gd_triggers = { |
| 603 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 604 | .t_frozen = ocfs2_frozen_trigger, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 605 | .t_abort = ocfs2_abort_trigger, |
| 606 | }, |
| 607 | .ot_offset = offsetof(struct ocfs2_group_desc, bg_check), |
| 608 | }; |
| 609 | |
Joel Becker | c175a51 | 2008-12-10 17:58:22 -0800 | [diff] [blame] | 610 | static struct ocfs2_triggers db_triggers = { |
| 611 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 612 | .t_frozen = ocfs2_db_frozen_trigger, |
Joel Becker | c175a51 | 2008-12-10 17:58:22 -0800 | [diff] [blame] | 613 | .t_abort = ocfs2_abort_trigger, |
| 614 | }, |
| 615 | }; |
| 616 | |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 617 | static struct ocfs2_triggers xb_triggers = { |
| 618 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 619 | .t_frozen = ocfs2_frozen_trigger, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 620 | .t_abort = ocfs2_abort_trigger, |
| 621 | }, |
| 622 | .ot_offset = offsetof(struct ocfs2_xattr_block, xb_check), |
| 623 | }; |
| 624 | |
| 625 | static struct ocfs2_triggers dq_triggers = { |
| 626 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 627 | .t_frozen = ocfs2_dq_frozen_trigger, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 628 | .t_abort = ocfs2_abort_trigger, |
| 629 | }, |
| 630 | }; |
| 631 | |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 632 | static struct ocfs2_triggers dr_triggers = { |
| 633 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 634 | .t_frozen = ocfs2_frozen_trigger, |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 635 | .t_abort = ocfs2_abort_trigger, |
| 636 | }, |
| 637 | .ot_offset = offsetof(struct ocfs2_dx_root_block, dr_check), |
| 638 | }; |
| 639 | |
| 640 | static struct ocfs2_triggers dl_triggers = { |
| 641 | .ot_triggers = { |
Jan Kara | 13ceef0 | 2010-07-14 07:56:33 +0200 | [diff] [blame] | 642 | .t_frozen = ocfs2_frozen_trigger, |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 643 | .t_abort = ocfs2_abort_trigger, |
| 644 | }, |
| 645 | .ot_offset = offsetof(struct ocfs2_dx_leaf, dl_check), |
| 646 | }; |
| 647 | |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 648 | static int __ocfs2_journal_access(handle_t *handle, |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 649 | struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 650 | struct buffer_head *bh, |
| 651 | struct ocfs2_triggers *triggers, |
| 652 | int type) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 653 | { |
| 654 | int status; |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 655 | struct ocfs2_super *osb = |
| 656 | OCFS2_SB(ocfs2_metadata_cache_get_super(ci)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 657 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 658 | BUG_ON(!ci || !ci->ci_ops); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 659 | BUG_ON(!handle); |
| 660 | BUG_ON(!bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 661 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 662 | trace_ocfs2_journal_access( |
| 663 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
| 664 | (unsigned long long)bh->b_blocknr, type, bh->b_size); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 665 | |
| 666 | /* we can safely remove this assertion after testing. */ |
| 667 | if (!buffer_uptodate(bh)) { |
| 668 | mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n"); |
| 669 | mlog(ML_ERROR, "b_blocknr=%llu\n", |
| 670 | (unsigned long long)bh->b_blocknr); |
Joseph Qi | acf8fdb | 2015-09-04 15:43:43 -0700 | [diff] [blame] | 671 | |
| 672 | lock_buffer(bh); |
| 673 | /* |
| 674 | * A previous attempt to write this buffer head failed. |
| 675 | * Nothing we can do but to retry the write and hope for |
| 676 | * the best. |
| 677 | */ |
| 678 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) { |
| 679 | clear_buffer_write_io_error(bh); |
| 680 | set_buffer_uptodate(bh); |
| 681 | } |
| 682 | |
| 683 | if (!buffer_uptodate(bh)) { |
| 684 | unlock_buffer(bh); |
| 685 | return -EIO; |
| 686 | } |
| 687 | unlock_buffer(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 688 | } |
| 689 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 690 | /* Set the current transaction information on the ci so |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 691 | * that the locking code knows whether it can drop it's locks |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 692 | * on this ci or not. We're protected from the commit |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 693 | * thread updating the current transaction id until |
| 694 | * ocfs2_commit_trans() because ocfs2_start_trans() took |
| 695 | * j_trans_barrier for us. */ |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 696 | ocfs2_set_ci_lock_trans(osb->journal, ci); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 697 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 698 | ocfs2_metadata_cache_io_lock(ci); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 699 | switch (type) { |
| 700 | case OCFS2_JOURNAL_ACCESS_CREATE: |
| 701 | case OCFS2_JOURNAL_ACCESS_WRITE: |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 702 | status = jbd2_journal_get_write_access(handle, bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 703 | break; |
| 704 | |
| 705 | case OCFS2_JOURNAL_ACCESS_UNDO: |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 706 | status = jbd2_journal_get_undo_access(handle, bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 707 | break; |
| 708 | |
| 709 | default: |
| 710 | status = -EINVAL; |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 711 | mlog(ML_ERROR, "Unknown access type!\n"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 712 | } |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 713 | if (!status && ocfs2_meta_ecc(osb) && triggers) |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 714 | jbd2_journal_set_triggers(bh, &triggers->ot_triggers); |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 715 | ocfs2_metadata_cache_io_unlock(ci); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 716 | |
| 717 | if (status < 0) |
| 718 | mlog(ML_ERROR, "Error %d getting %d access to buffer!\n", |
| 719 | status, type); |
| 720 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 721 | return status; |
| 722 | } |
| 723 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 724 | int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 725 | struct buffer_head *bh, int type) |
| 726 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 727 | return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type); |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 728 | } |
| 729 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 730 | int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 731 | struct buffer_head *bh, int type) |
| 732 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 733 | return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type); |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 734 | } |
| 735 | |
Tao Ma | 93c9708 | 2009-08-18 11:19:20 +0800 | [diff] [blame] | 736 | int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci, |
| 737 | struct buffer_head *bh, int type) |
| 738 | { |
| 739 | return __ocfs2_journal_access(handle, ci, bh, &rb_triggers, |
| 740 | type); |
| 741 | } |
| 742 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 743 | int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 744 | struct buffer_head *bh, int type) |
| 745 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 746 | return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type); |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 747 | } |
| 748 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 749 | int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 750 | struct buffer_head *bh, int type) |
| 751 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 752 | return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type); |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 753 | } |
| 754 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 755 | int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 756 | struct buffer_head *bh, int type) |
| 757 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 758 | return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type); |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 759 | } |
| 760 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 761 | int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci, |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 762 | struct buffer_head *bh, int type) |
| 763 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 764 | return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type); |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 765 | } |
| 766 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 767 | int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci, |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 768 | struct buffer_head *bh, int type) |
| 769 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 770 | return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type); |
Mark Fasheh | 9b7895e | 2008-11-12 16:27:44 -0800 | [diff] [blame] | 771 | } |
| 772 | |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 773 | int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci, |
| 774 | struct buffer_head *bh, int type) |
| 775 | { |
| 776 | return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type); |
| 777 | } |
| 778 | |
| 779 | int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci, |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 780 | struct buffer_head *bh, int type) |
| 781 | { |
Joel Becker | 0cf2f76 | 2009-02-12 16:41:25 -0800 | [diff] [blame] | 782 | return __ocfs2_journal_access(handle, ci, bh, NULL, type); |
Joel Becker | 50655ae | 2008-09-11 15:53:07 -0700 | [diff] [blame] | 783 | } |
| 784 | |
Joel Becker | ec20cec | 2010-03-19 14:13:52 -0700 | [diff] [blame] | 785 | void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 786 | { |
| 787 | int status; |
| 788 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 789 | trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 790 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 791 | status = jbd2_journal_dirty_metadata(handle, bh); |
Joseph Qi | e272e7f | 2015-06-24 16:55:12 -0700 | [diff] [blame] | 792 | if (status) { |
| 793 | mlog_errno(status); |
| 794 | if (!is_handle_aborted(handle)) { |
| 795 | journal_t *journal = handle->h_transaction->t_journal; |
| 796 | struct super_block *sb = bh->b_bdev->bd_super; |
| 797 | |
| 798 | mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. " |
| 799 | "Aborting transaction and journal.\n"); |
| 800 | handle->h_err = status; |
| 801 | jbd2_journal_abort_handle(handle); |
| 802 | jbd2_journal_abort(journal, status); |
| 803 | ocfs2_abort(sb, "Journal already aborted.\n"); |
| 804 | } |
| 805 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 806 | } |
| 807 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 808 | #define OCFS2_DEFAULT_COMMIT_INTERVAL (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 809 | |
| 810 | void ocfs2_set_journal_params(struct ocfs2_super *osb) |
| 811 | { |
| 812 | journal_t *journal = osb->journal->j_journal; |
Mark Fasheh | d147b3d | 2007-11-07 14:40:36 -0800 | [diff] [blame] | 813 | unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL; |
| 814 | |
| 815 | if (osb->osb_commit_interval) |
| 816 | commit_interval = osb->osb_commit_interval; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 817 | |
Theodore Ts'o | a931da6 | 2010-08-03 21:35:12 -0400 | [diff] [blame] | 818 | write_lock(&journal->j_state_lock); |
Mark Fasheh | d147b3d | 2007-11-07 14:40:36 -0800 | [diff] [blame] | 819 | journal->j_commit_interval = commit_interval; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 820 | if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER) |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 821 | journal->j_flags |= JBD2_BARRIER; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 822 | else |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 823 | journal->j_flags &= ~JBD2_BARRIER; |
Theodore Ts'o | a931da6 | 2010-08-03 21:35:12 -0400 | [diff] [blame] | 824 | write_unlock(&journal->j_state_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 825 | } |
| 826 | |
| 827 | int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty) |
| 828 | { |
| 829 | int status = -1; |
| 830 | struct inode *inode = NULL; /* the journal inode */ |
| 831 | journal_t *j_journal = NULL; |
| 832 | struct ocfs2_dinode *di = NULL; |
| 833 | struct buffer_head *bh = NULL; |
| 834 | struct ocfs2_super *osb; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 835 | int inode_lock = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 836 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 837 | BUG_ON(!journal); |
| 838 | |
| 839 | osb = journal->j_osb; |
| 840 | |
| 841 | /* already have the inode for our journal */ |
| 842 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 843 | osb->slot_num); |
| 844 | if (inode == NULL) { |
| 845 | status = -EACCES; |
| 846 | mlog_errno(status); |
| 847 | goto done; |
| 848 | } |
| 849 | if (is_bad_inode(inode)) { |
| 850 | mlog(ML_ERROR, "access error (bad inode)\n"); |
| 851 | iput(inode); |
| 852 | inode = NULL; |
| 853 | status = -EACCES; |
| 854 | goto done; |
| 855 | } |
| 856 | |
| 857 | SET_INODE_JOURNAL(inode); |
| 858 | OCFS2_I(inode)->ip_open_count++; |
| 859 | |
Mark Fasheh | 6eff579 | 2006-01-18 10:31:47 -0800 | [diff] [blame] | 860 | /* Skip recovery waits here - journal inode metadata never |
| 861 | * changes in a live cluster so it can be considered an |
| 862 | * exception to the rule. */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 863 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 864 | if (status < 0) { |
| 865 | if (status != -ERESTARTSYS) |
| 866 | mlog(ML_ERROR, "Could not get lock on journal!\n"); |
| 867 | goto done; |
| 868 | } |
| 869 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 870 | inode_lock = 1; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 871 | di = (struct ocfs2_dinode *)bh->b_data; |
| 872 | |
Junxiao Bi | f17c20d | 2013-09-11 14:19:45 -0700 | [diff] [blame] | 873 | if (i_size_read(inode) < OCFS2_MIN_JOURNAL_SIZE) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 874 | mlog(ML_ERROR, "Journal file size (%lld) is too small!\n", |
Junxiao Bi | f17c20d | 2013-09-11 14:19:45 -0700 | [diff] [blame] | 875 | i_size_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 876 | status = -EINVAL; |
| 877 | goto done; |
| 878 | } |
| 879 | |
Junxiao Bi | f17c20d | 2013-09-11 14:19:45 -0700 | [diff] [blame] | 880 | trace_ocfs2_journal_init(i_size_read(inode), |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 881 | (unsigned long long)inode->i_blocks, |
| 882 | OCFS2_I(inode)->ip_clusters); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 883 | |
| 884 | /* call the kernels journal init function now */ |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 885 | j_journal = jbd2_journal_init_inode(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 886 | if (j_journal == NULL) { |
| 887 | mlog(ML_ERROR, "Linux journal layer error\n"); |
| 888 | status = -EINVAL; |
| 889 | goto done; |
| 890 | } |
| 891 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 892 | trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 893 | |
| 894 | *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) & |
| 895 | OCFS2_JOURNAL_DIRTY_FL); |
| 896 | |
| 897 | journal->j_journal = j_journal; |
| 898 | journal->j_inode = inode; |
| 899 | journal->j_bh = bh; |
| 900 | |
| 901 | ocfs2_set_journal_params(osb); |
| 902 | |
| 903 | journal->j_state = OCFS2_JOURNAL_LOADED; |
| 904 | |
| 905 | status = 0; |
| 906 | done: |
| 907 | if (status < 0) { |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 908 | if (inode_lock) |
| 909 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 910 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 911 | if (inode) { |
| 912 | OCFS2_I(inode)->ip_open_count--; |
| 913 | iput(inode); |
| 914 | } |
| 915 | } |
| 916 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 917 | return status; |
| 918 | } |
| 919 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 920 | static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di) |
| 921 | { |
| 922 | le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1); |
| 923 | } |
| 924 | |
| 925 | static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di) |
| 926 | { |
| 927 | return le32_to_cpu(di->id1.journal1.ij_recovery_generation); |
| 928 | } |
| 929 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 930 | static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb, |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 931 | int dirty, int replayed) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 932 | { |
| 933 | int status; |
| 934 | unsigned int flags; |
| 935 | struct ocfs2_journal *journal = osb->journal; |
| 936 | struct buffer_head *bh = journal->j_bh; |
| 937 | struct ocfs2_dinode *fe; |
| 938 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 939 | fe = (struct ocfs2_dinode *)bh->b_data; |
Joel Becker | 10995aa | 2008-11-13 14:49:12 -0800 | [diff] [blame] | 940 | |
| 941 | /* The journal bh on the osb always comes from ocfs2_journal_init() |
| 942 | * and was validated there inside ocfs2_inode_lock_full(). It's a |
| 943 | * code bug if we mess it up. */ |
| 944 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 945 | |
| 946 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 947 | if (dirty) |
| 948 | flags |= OCFS2_JOURNAL_DIRTY_FL; |
| 949 | else |
| 950 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; |
| 951 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); |
| 952 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 953 | if (replayed) |
| 954 | ocfs2_bump_recovery_generation(fe); |
| 955 | |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 956 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 957 | status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 958 | if (status < 0) |
| 959 | mlog_errno(status); |
| 960 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 961 | return status; |
| 962 | } |
| 963 | |
| 964 | /* |
| 965 | * If the journal has been kmalloc'd it needs to be freed after this |
| 966 | * call. |
| 967 | */ |
| 968 | void ocfs2_journal_shutdown(struct ocfs2_super *osb) |
| 969 | { |
| 970 | struct ocfs2_journal *journal = NULL; |
| 971 | int status = 0; |
| 972 | struct inode *inode = NULL; |
| 973 | int num_running_trans = 0; |
| 974 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 975 | BUG_ON(!osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 976 | |
| 977 | journal = osb->journal; |
| 978 | if (!journal) |
| 979 | goto done; |
| 980 | |
| 981 | inode = journal->j_inode; |
| 982 | |
| 983 | if (journal->j_state != OCFS2_JOURNAL_LOADED) |
| 984 | goto done; |
| 985 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 986 | /* need to inc inode use count - jbd2_journal_destroy will iput. */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 987 | if (!igrab(inode)) |
| 988 | BUG(); |
| 989 | |
| 990 | num_running_trans = atomic_read(&(osb->journal->j_num_trans)); |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 991 | trace_ocfs2_journal_shutdown(num_running_trans); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 992 | |
| 993 | /* Do a commit_cache here. It will flush our journal, *and* |
| 994 | * release any locks that are still held. |
| 995 | * set the SHUTDOWN flag and release the trans lock. |
| 996 | * the commit thread will take the trans lock for us below. */ |
| 997 | journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN; |
| 998 | |
| 999 | /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not |
| 1000 | * drop the trans_lock (which we want to hold until we |
| 1001 | * completely destroy the journal. */ |
| 1002 | if (osb->commit_task) { |
| 1003 | /* Wait for the commit thread */ |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1004 | trace_ocfs2_journal_shutdown_wait(osb->commit_task); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1005 | kthread_stop(osb->commit_task); |
| 1006 | osb->commit_task = NULL; |
| 1007 | } |
| 1008 | |
| 1009 | BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0); |
| 1010 | |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 1011 | if (ocfs2_mount_local(osb)) { |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1012 | jbd2_journal_lock_updates(journal->j_journal); |
| 1013 | status = jbd2_journal_flush(journal->j_journal); |
| 1014 | jbd2_journal_unlock_updates(journal->j_journal); |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 1015 | if (status < 0) |
| 1016 | mlog_errno(status); |
| 1017 | } |
| 1018 | |
| 1019 | if (status == 0) { |
| 1020 | /* |
| 1021 | * Do not toggle if flush was unsuccessful otherwise |
| 1022 | * will leave dirty metadata in a "clean" journal |
| 1023 | */ |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1024 | status = ocfs2_journal_toggle_dirty(osb, 0, 0); |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 1025 | if (status < 0) |
| 1026 | mlog_errno(status); |
| 1027 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1028 | |
| 1029 | /* Shutdown the kernel journal system */ |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1030 | jbd2_journal_destroy(journal->j_journal); |
Sunil Mushran | ae0dff6 | 2008-10-22 13:24:29 -0700 | [diff] [blame] | 1031 | journal->j_journal = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1032 | |
| 1033 | OCFS2_I(inode)->ip_open_count--; |
| 1034 | |
| 1035 | /* unlock our journal */ |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1036 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1037 | |
| 1038 | brelse(journal->j_bh); |
| 1039 | journal->j_bh = NULL; |
| 1040 | |
| 1041 | journal->j_state = OCFS2_JOURNAL_FREE; |
| 1042 | |
| 1043 | // up_write(&journal->j_trans_barrier); |
| 1044 | done: |
Joseph Qi | 72865d9 | 2016-01-14 15:17:27 -0800 | [diff] [blame] | 1045 | iput(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1046 | } |
| 1047 | |
| 1048 | static void ocfs2_clear_journal_error(struct super_block *sb, |
| 1049 | journal_t *journal, |
| 1050 | int slot) |
| 1051 | { |
| 1052 | int olderr; |
| 1053 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1054 | olderr = jbd2_journal_errno(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1055 | if (olderr) { |
| 1056 | mlog(ML_ERROR, "File system error %d recorded in " |
| 1057 | "journal %u.\n", olderr, slot); |
| 1058 | mlog(ML_ERROR, "File system on device %s needs checking.\n", |
| 1059 | sb->s_id); |
| 1060 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1061 | jbd2_journal_ack_err(journal); |
| 1062 | jbd2_journal_clear_err(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1063 | } |
| 1064 | } |
| 1065 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1066 | int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1067 | { |
| 1068 | int status = 0; |
| 1069 | struct ocfs2_super *osb; |
| 1070 | |
Julia Lawall | b1f3550 | 2008-03-04 15:21:05 -0800 | [diff] [blame] | 1071 | BUG_ON(!journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1072 | |
| 1073 | osb = journal->j_osb; |
| 1074 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1075 | status = jbd2_journal_load(journal->j_journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1076 | if (status < 0) { |
| 1077 | mlog(ML_ERROR, "Failed to load journal!\n"); |
| 1078 | goto done; |
| 1079 | } |
| 1080 | |
| 1081 | ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num); |
| 1082 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1083 | status = ocfs2_journal_toggle_dirty(osb, 1, replayed); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1084 | if (status < 0) { |
| 1085 | mlog_errno(status); |
| 1086 | goto done; |
| 1087 | } |
| 1088 | |
| 1089 | /* Launch the commit thread */ |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 1090 | if (!local) { |
| 1091 | osb->commit_task = kthread_run(ocfs2_commit_thread, osb, |
Joseph Qi | 5afc44e | 2015-11-05 18:44:13 -0800 | [diff] [blame] | 1092 | "ocfs2cmt-%s", osb->uuid_str); |
Sunil Mushran | c271c5c | 2006-12-05 17:56:35 -0800 | [diff] [blame] | 1093 | if (IS_ERR(osb->commit_task)) { |
| 1094 | status = PTR_ERR(osb->commit_task); |
| 1095 | osb->commit_task = NULL; |
| 1096 | mlog(ML_ERROR, "unable to launch ocfs2commit thread, " |
| 1097 | "error=%d", status); |
| 1098 | goto done; |
| 1099 | } |
| 1100 | } else |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1101 | osb->commit_task = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1102 | |
| 1103 | done: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1104 | return status; |
| 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | /* 'full' flag tells us whether we clear out all blocks or if we just |
| 1109 | * mark the journal clean */ |
| 1110 | int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full) |
| 1111 | { |
| 1112 | int status; |
| 1113 | |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 1114 | BUG_ON(!journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1115 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1116 | status = jbd2_journal_wipe(journal->j_journal, full); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1117 | if (status < 0) { |
| 1118 | mlog_errno(status); |
| 1119 | goto bail; |
| 1120 | } |
| 1121 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1122 | status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1123 | if (status < 0) |
| 1124 | mlog_errno(status); |
| 1125 | |
| 1126 | bail: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1127 | return status; |
| 1128 | } |
| 1129 | |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1130 | static int ocfs2_recovery_completed(struct ocfs2_super *osb) |
| 1131 | { |
| 1132 | int empty; |
| 1133 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
| 1134 | |
| 1135 | spin_lock(&osb->osb_lock); |
| 1136 | empty = (rm->rm_used == 0); |
| 1137 | spin_unlock(&osb->osb_lock); |
| 1138 | |
| 1139 | return empty; |
| 1140 | } |
| 1141 | |
| 1142 | void ocfs2_wait_for_recovery(struct ocfs2_super *osb) |
| 1143 | { |
| 1144 | wait_event(osb->recovery_event, ocfs2_recovery_completed(osb)); |
| 1145 | } |
| 1146 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1147 | /* |
| 1148 | * JBD Might read a cached version of another nodes journal file. We |
| 1149 | * don't want this as this file changes often and we get no |
| 1150 | * notification on those changes. The only way to be sure that we've |
| 1151 | * got the most up to date version of those blocks then is to force |
| 1152 | * read them off disk. Just searching through the buffer cache won't |
| 1153 | * work as there may be pages backing this file which are still marked |
| 1154 | * up to date. We know things can't change on this file underneath us |
| 1155 | * as we have the lock by now :) |
| 1156 | */ |
| 1157 | static int ocfs2_force_read_journal(struct inode *inode) |
| 1158 | { |
| 1159 | int status = 0; |
Mark Fasheh | 4f902c3 | 2007-03-09 16:26:50 -0800 | [diff] [blame] | 1160 | int i; |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 1161 | u64 v_blkno, p_blkno, p_blocks, num_blocks; |
Junxiao Bi | 0b492f6 | 2016-07-26 15:21:32 -0700 | [diff] [blame] | 1162 | struct buffer_head *bh = NULL; |
| 1163 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1164 | |
Junxiao Bi | f17c20d | 2013-09-11 14:19:45 -0700 | [diff] [blame] | 1165 | num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1166 | v_blkno = 0; |
Mark Fasheh | 8110b07 | 2007-03-22 16:53:23 -0700 | [diff] [blame] | 1167 | while (v_blkno < num_blocks) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1168 | status = ocfs2_extent_map_get_blocks(inode, v_blkno, |
Mark Fasheh | 49cb8d2 | 2007-03-09 16:21:46 -0800 | [diff] [blame] | 1169 | &p_blkno, &p_blocks, NULL); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1170 | if (status < 0) { |
| 1171 | mlog_errno(status); |
| 1172 | goto bail; |
| 1173 | } |
| 1174 | |
Junxiao Bi | 0b492f6 | 2016-07-26 15:21:32 -0700 | [diff] [blame] | 1175 | for (i = 0; i < p_blocks; i++, p_blkno++) { |
| 1176 | bh = __find_get_block(osb->sb->s_bdev, p_blkno, |
| 1177 | osb->sb->s_blocksize); |
| 1178 | /* block not cached. */ |
| 1179 | if (!bh) |
| 1180 | continue; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1181 | |
Junxiao Bi | 0b492f6 | 2016-07-26 15:21:32 -0700 | [diff] [blame] | 1182 | brelse(bh); |
| 1183 | bh = NULL; |
| 1184 | /* We are reading journal data which should not |
| 1185 | * be put in the uptodate cache. |
| 1186 | */ |
| 1187 | status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh); |
| 1188 | if (status < 0) { |
| 1189 | mlog_errno(status); |
| 1190 | goto bail; |
| 1191 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1192 | |
Junxiao Bi | 0b492f6 | 2016-07-26 15:21:32 -0700 | [diff] [blame] | 1193 | brelse(bh); |
| 1194 | bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1195 | } |
| 1196 | |
| 1197 | v_blkno += p_blocks; |
| 1198 | } |
| 1199 | |
| 1200 | bail: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1201 | return status; |
| 1202 | } |
| 1203 | |
| 1204 | struct ocfs2_la_recovery_item { |
| 1205 | struct list_head lri_list; |
| 1206 | int lri_slot; |
| 1207 | struct ocfs2_dinode *lri_la_dinode; |
| 1208 | struct ocfs2_dinode *lri_tl_dinode; |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1209 | struct ocfs2_quota_recovery *lri_qrec; |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1210 | enum ocfs2_orphan_reco_type lri_orphan_reco_type; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1211 | }; |
| 1212 | |
| 1213 | /* Does the second half of the recovery process. By this point, the |
| 1214 | * node is marked clean and can actually be considered recovered, |
| 1215 | * hence it's no longer in the recovery map, but there's still some |
| 1216 | * cleanup we can do which shouldn't happen within the recovery thread |
| 1217 | * as locking in that context becomes very difficult if we are to take |
| 1218 | * recovering nodes into account. |
| 1219 | * |
| 1220 | * NOTE: This function can and will sleep on recovery of other nodes |
| 1221 | * during cluster locking, just like any other ocfs2 process. |
| 1222 | */ |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 1223 | void ocfs2_complete_recovery(struct work_struct *work) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1224 | { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1225 | int ret = 0; |
David Howells | c402895 | 2006-11-22 14:57:56 +0000 | [diff] [blame] | 1226 | struct ocfs2_journal *journal = |
| 1227 | container_of(work, struct ocfs2_journal, j_recovery_work); |
| 1228 | struct ocfs2_super *osb = journal->j_osb; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1229 | struct ocfs2_dinode *la_dinode, *tl_dinode; |
Christoph Hellwig | 800deef | 2007-05-17 16:03:13 +0200 | [diff] [blame] | 1230 | struct ocfs2_la_recovery_item *item, *n; |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1231 | struct ocfs2_quota_recovery *qrec; |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1232 | enum ocfs2_orphan_reco_type orphan_reco_type; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1233 | LIST_HEAD(tmp_la_list); |
| 1234 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1235 | trace_ocfs2_complete_recovery( |
| 1236 | (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1237 | |
| 1238 | spin_lock(&journal->j_lock); |
| 1239 | list_splice_init(&journal->j_la_cleanups, &tmp_la_list); |
| 1240 | spin_unlock(&journal->j_lock); |
| 1241 | |
Christoph Hellwig | 800deef | 2007-05-17 16:03:13 +0200 | [diff] [blame] | 1242 | list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1243 | list_del_init(&item->lri_list); |
| 1244 | |
Jan Kara | 19ece54 | 2008-08-21 20:13:17 +0200 | [diff] [blame] | 1245 | ocfs2_wait_on_quotas(osb); |
| 1246 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1247 | la_dinode = item->lri_la_dinode; |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1248 | tl_dinode = item->lri_tl_dinode; |
| 1249 | qrec = item->lri_qrec; |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1250 | orphan_reco_type = item->lri_orphan_reco_type; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1251 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1252 | trace_ocfs2_complete_recovery_slot(item->lri_slot, |
| 1253 | la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0, |
| 1254 | tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0, |
| 1255 | qrec); |
| 1256 | |
| 1257 | if (la_dinode) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1258 | ret = ocfs2_complete_local_alloc_recovery(osb, |
| 1259 | la_dinode); |
| 1260 | if (ret < 0) |
| 1261 | mlog_errno(ret); |
| 1262 | |
| 1263 | kfree(la_dinode); |
| 1264 | } |
| 1265 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1266 | if (tl_dinode) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1267 | ret = ocfs2_complete_truncate_log_recovery(osb, |
| 1268 | tl_dinode); |
| 1269 | if (ret < 0) |
| 1270 | mlog_errno(ret); |
| 1271 | |
| 1272 | kfree(tl_dinode); |
| 1273 | } |
| 1274 | |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1275 | ret = ocfs2_recover_orphans(osb, item->lri_slot, |
| 1276 | orphan_reco_type); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1277 | if (ret < 0) |
| 1278 | mlog_errno(ret); |
| 1279 | |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1280 | if (qrec) { |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1281 | ret = ocfs2_finish_quota_recovery(osb, qrec, |
| 1282 | item->lri_slot); |
| 1283 | if (ret < 0) |
| 1284 | mlog_errno(ret); |
| 1285 | /* Recovery info is already freed now */ |
| 1286 | } |
| 1287 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1288 | kfree(item); |
| 1289 | } |
| 1290 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1291 | trace_ocfs2_complete_recovery_end(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1292 | } |
| 1293 | |
| 1294 | /* NOTE: This function always eats your references to la_dinode and |
| 1295 | * tl_dinode, either manually on error, or by passing them to |
| 1296 | * ocfs2_complete_recovery */ |
| 1297 | static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal, |
| 1298 | int slot_num, |
| 1299 | struct ocfs2_dinode *la_dinode, |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1300 | struct ocfs2_dinode *tl_dinode, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1301 | struct ocfs2_quota_recovery *qrec, |
| 1302 | enum ocfs2_orphan_reco_type orphan_reco_type) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1303 | { |
| 1304 | struct ocfs2_la_recovery_item *item; |
| 1305 | |
Sunil Mushran | afae00ab | 2006-04-12 14:37:00 -0700 | [diff] [blame] | 1306 | item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1307 | if (!item) { |
| 1308 | /* Though we wish to avoid it, we are in fact safe in |
| 1309 | * skipping local alloc cleanup as fsck.ocfs2 is more |
| 1310 | * than capable of reclaiming unused space. */ |
Tim Gardner | d787ab0 | 2013-02-21 16:42:44 -0800 | [diff] [blame] | 1311 | kfree(la_dinode); |
| 1312 | kfree(tl_dinode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1313 | |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1314 | if (qrec) |
| 1315 | ocfs2_free_quota_recovery(qrec); |
| 1316 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1317 | mlog_errno(-ENOMEM); |
| 1318 | return; |
| 1319 | } |
| 1320 | |
| 1321 | INIT_LIST_HEAD(&item->lri_list); |
| 1322 | item->lri_la_dinode = la_dinode; |
| 1323 | item->lri_slot = slot_num; |
| 1324 | item->lri_tl_dinode = tl_dinode; |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1325 | item->lri_qrec = qrec; |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1326 | item->lri_orphan_reco_type = orphan_reco_type; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1327 | |
| 1328 | spin_lock(&journal->j_lock); |
| 1329 | list_add_tail(&item->lri_list, &journal->j_la_cleanups); |
jiangyiwen | 35ddf78 | 2016-03-25 14:21:32 -0700 | [diff] [blame] | 1330 | queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1331 | spin_unlock(&journal->j_lock); |
| 1332 | } |
| 1333 | |
| 1334 | /* Called by the mount code to queue recovery the last part of |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1335 | * recovery for it's own and offline slot(s). */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1336 | void ocfs2_complete_mount_recovery(struct ocfs2_super *osb) |
| 1337 | { |
| 1338 | struct ocfs2_journal *journal = osb->journal; |
| 1339 | |
Sunil Mushran | 10b3dd7 | 2011-05-04 10:28:02 -0700 | [diff] [blame] | 1340 | if (ocfs2_is_hard_readonly(osb)) |
| 1341 | return; |
| 1342 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1343 | /* No need to queue up our truncate_log as regular cleanup will catch |
| 1344 | * that */ |
| 1345 | ocfs2_queue_recovery_completion(journal, osb->slot_num, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1346 | osb->local_alloc_copy, NULL, NULL, |
| 1347 | ORPHAN_NEED_TRUNCATE); |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1348 | ocfs2_schedule_truncate_log_flush(osb, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1349 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1350 | osb->local_alloc_copy = NULL; |
| 1351 | osb->dirty = 0; |
| 1352 | |
| 1353 | /* queue to recover orphan slots for all offline slots */ |
| 1354 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1355 | ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1356 | ocfs2_free_replay_slots(osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1357 | } |
| 1358 | |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1359 | void ocfs2_complete_quota_recovery(struct ocfs2_super *osb) |
| 1360 | { |
| 1361 | if (osb->quota_rec) { |
| 1362 | ocfs2_queue_recovery_completion(osb->journal, |
| 1363 | osb->slot_num, |
| 1364 | NULL, |
| 1365 | NULL, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1366 | osb->quota_rec, |
| 1367 | ORPHAN_NEED_TRUNCATE); |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1368 | osb->quota_rec = NULL; |
| 1369 | } |
| 1370 | } |
| 1371 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1372 | static int __ocfs2_recovery_thread(void *arg) |
| 1373 | { |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1374 | int status, node_num, slot_num; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1375 | struct ocfs2_super *osb = arg; |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1376 | struct ocfs2_recovery_map *rm = osb->recovery_map; |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1377 | int *rm_quota = NULL; |
| 1378 | int rm_quota_used = 0, i; |
| 1379 | struct ocfs2_quota_recovery *qrec; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1380 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1381 | status = ocfs2_wait_on_mount(osb); |
| 1382 | if (status < 0) { |
| 1383 | goto bail; |
| 1384 | } |
| 1385 | |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1386 | rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS); |
| 1387 | if (!rm_quota) { |
| 1388 | status = -ENOMEM; |
| 1389 | goto bail; |
| 1390 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1391 | restart: |
| 1392 | status = ocfs2_super_lock(osb, 1); |
| 1393 | if (status < 0) { |
| 1394 | mlog_errno(status); |
| 1395 | goto bail; |
| 1396 | } |
| 1397 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1398 | status = ocfs2_compute_replay_slots(osb); |
| 1399 | if (status < 0) |
| 1400 | mlog_errno(status); |
| 1401 | |
| 1402 | /* queue recovery for our own slot */ |
| 1403 | ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1404 | NULL, NULL, ORPHAN_NO_NEED_TRUNCATE); |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1405 | |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1406 | spin_lock(&osb->osb_lock); |
| 1407 | while (rm->rm_used) { |
| 1408 | /* It's always safe to remove entry zero, as we won't |
| 1409 | * clear it until ocfs2_recover_node() has succeeded. */ |
| 1410 | node_num = rm->rm_entries[0]; |
| 1411 | spin_unlock(&osb->osb_lock); |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1412 | slot_num = ocfs2_node_num_to_slot(osb, node_num); |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1413 | trace_ocfs2_recovery_thread_node(node_num, slot_num); |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1414 | if (slot_num == -ENOENT) { |
| 1415 | status = 0; |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1416 | goto skip_recovery; |
| 1417 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1418 | |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1419 | /* It is a bit subtle with quota recovery. We cannot do it |
| 1420 | * immediately because we have to obtain cluster locks from |
| 1421 | * quota files and we also don't want to just skip it because |
| 1422 | * then quota usage would be out of sync until some node takes |
| 1423 | * the slot. So we remember which nodes need quota recovery |
| 1424 | * and when everything else is done, we recover quotas. */ |
| 1425 | for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++); |
| 1426 | if (i == rm_quota_used) |
| 1427 | rm_quota[rm_quota_used++] = slot_num; |
| 1428 | |
| 1429 | status = ocfs2_recover_node(osb, node_num, slot_num); |
| 1430 | skip_recovery: |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1431 | if (!status) { |
| 1432 | ocfs2_recovery_map_clear(osb, node_num); |
| 1433 | } else { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1434 | mlog(ML_ERROR, |
| 1435 | "Error %d recovering node %d on device (%u,%u)!\n", |
| 1436 | status, node_num, |
| 1437 | MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev)); |
| 1438 | mlog(ML_ERROR, "Volume requires unmount.\n"); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1439 | } |
| 1440 | |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1441 | spin_lock(&osb->osb_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1442 | } |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1443 | spin_unlock(&osb->osb_lock); |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1444 | trace_ocfs2_recovery_thread_end(status); |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1445 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1446 | /* Refresh all journal recovery generations from disk */ |
| 1447 | status = ocfs2_check_journals_nolocks(osb); |
| 1448 | status = (status == -EROFS) ? 0 : status; |
| 1449 | if (status < 0) |
| 1450 | mlog_errno(status); |
| 1451 | |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1452 | /* Now it is right time to recover quotas... We have to do this under |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1453 | * superblock lock so that no one can start using the slot (and crash) |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1454 | * before we recover it */ |
| 1455 | for (i = 0; i < rm_quota_used; i++) { |
| 1456 | qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]); |
| 1457 | if (IS_ERR(qrec)) { |
| 1458 | status = PTR_ERR(qrec); |
| 1459 | mlog_errno(status); |
| 1460 | continue; |
| 1461 | } |
| 1462 | ocfs2_queue_recovery_completion(osb->journal, rm_quota[i], |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1463 | NULL, NULL, qrec, |
| 1464 | ORPHAN_NEED_TRUNCATE); |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1465 | } |
| 1466 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1467 | ocfs2_super_unlock(osb, 1); |
| 1468 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1469 | /* queue recovery for offline slots */ |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1470 | ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1471 | |
| 1472 | bail: |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1473 | mutex_lock(&osb->recovery_lock); |
Joel Becker | 553abd0 | 2008-02-01 12:03:57 -0800 | [diff] [blame] | 1474 | if (!status && !ocfs2_recovery_completed(osb)) { |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1475 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1476 | goto restart; |
| 1477 | } |
| 1478 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1479 | ocfs2_free_replay_slots(osb); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1480 | osb->recovery_thread_task = NULL; |
| 1481 | mb(); /* sync with ocfs2_recovery_thread_running */ |
| 1482 | wake_up(&osb->recovery_event); |
| 1483 | |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1484 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1485 | |
Tim Gardner | d787ab0 | 2013-02-21 16:42:44 -0800 | [diff] [blame] | 1486 | kfree(rm_quota); |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1487 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1488 | /* no one is callint kthread_stop() for us so the kthread() api |
| 1489 | * requires that we call do_exit(). And it isn't exported, but |
| 1490 | * complete_and_exit() seems to be a minimal wrapper around it. */ |
| 1491 | complete_and_exit(NULL, status); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1492 | } |
| 1493 | |
| 1494 | void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num) |
| 1495 | { |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1496 | mutex_lock(&osb->recovery_lock); |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1497 | |
| 1498 | trace_ocfs2_recovery_thread(node_num, osb->node_num, |
| 1499 | osb->disable_recovery, osb->recovery_thread_task, |
| 1500 | osb->disable_recovery ? |
| 1501 | -1 : ocfs2_recovery_map_set(osb, node_num)); |
| 1502 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1503 | if (osb->disable_recovery) |
| 1504 | goto out; |
| 1505 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1506 | if (osb->recovery_thread_task) |
| 1507 | goto out; |
| 1508 | |
| 1509 | osb->recovery_thread_task = kthread_run(__ocfs2_recovery_thread, osb, |
Joseph Qi | 5afc44e | 2015-11-05 18:44:13 -0800 | [diff] [blame] | 1510 | "ocfs2rec-%s", osb->uuid_str); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1511 | if (IS_ERR(osb->recovery_thread_task)) { |
| 1512 | mlog_errno((int)PTR_ERR(osb->recovery_thread_task)); |
| 1513 | osb->recovery_thread_task = NULL; |
| 1514 | } |
| 1515 | |
| 1516 | out: |
Arjan van de Ven | c74ec2f | 2006-01-13 21:54:23 -0800 | [diff] [blame] | 1517 | mutex_unlock(&osb->recovery_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1518 | wake_up(&osb->recovery_event); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1519 | } |
| 1520 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1521 | static int ocfs2_read_journal_inode(struct ocfs2_super *osb, |
| 1522 | int slot_num, |
| 1523 | struct buffer_head **bh, |
| 1524 | struct inode **ret_inode) |
| 1525 | { |
| 1526 | int status = -EACCES; |
| 1527 | struct inode *inode = NULL; |
| 1528 | |
| 1529 | BUG_ON(slot_num >= osb->max_slots); |
| 1530 | |
| 1531 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 1532 | slot_num); |
| 1533 | if (!inode || is_bad_inode(inode)) { |
| 1534 | mlog_errno(status); |
| 1535 | goto bail; |
| 1536 | } |
| 1537 | SET_INODE_JOURNAL(inode); |
| 1538 | |
Joel Becker | b657c95 | 2008-11-13 14:49:11 -0800 | [diff] [blame] | 1539 | status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1540 | if (status < 0) { |
| 1541 | mlog_errno(status); |
| 1542 | goto bail; |
| 1543 | } |
| 1544 | |
| 1545 | status = 0; |
| 1546 | |
| 1547 | bail: |
| 1548 | if (inode) { |
| 1549 | if (status || !ret_inode) |
| 1550 | iput(inode); |
| 1551 | else |
| 1552 | *ret_inode = inode; |
| 1553 | } |
| 1554 | return status; |
| 1555 | } |
| 1556 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1557 | /* Does the actual journal replay and marks the journal inode as |
| 1558 | * clean. Will only replay if the journal inode is marked dirty. */ |
| 1559 | static int ocfs2_replay_journal(struct ocfs2_super *osb, |
| 1560 | int node_num, |
| 1561 | int slot_num) |
| 1562 | { |
| 1563 | int status; |
| 1564 | int got_lock = 0; |
| 1565 | unsigned int flags; |
| 1566 | struct inode *inode = NULL; |
| 1567 | struct ocfs2_dinode *fe; |
| 1568 | journal_t *journal = NULL; |
| 1569 | struct buffer_head *bh = NULL; |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1570 | u32 slot_reco_gen; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1571 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1572 | status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode); |
| 1573 | if (status) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1574 | mlog_errno(status); |
| 1575 | goto done; |
| 1576 | } |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1577 | |
| 1578 | fe = (struct ocfs2_dinode *)bh->b_data; |
| 1579 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
| 1580 | brelse(bh); |
| 1581 | bh = NULL; |
| 1582 | |
| 1583 | /* |
| 1584 | * As the fs recovery is asynchronous, there is a small chance that |
| 1585 | * another node mounted (and recovered) the slot before the recovery |
| 1586 | * thread could get the lock. To handle that, we dirty read the journal |
| 1587 | * inode for that slot to get the recovery generation. If it is |
| 1588 | * different than what we expected, the slot has been recovered. |
| 1589 | * If not, it needs recovery. |
| 1590 | */ |
| 1591 | if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1592 | trace_ocfs2_replay_journal_recovered(slot_num, |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1593 | osb->slot_recovery_generations[slot_num], slot_reco_gen); |
| 1594 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; |
| 1595 | status = -EBUSY; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1596 | goto done; |
| 1597 | } |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1598 | |
| 1599 | /* Continue with recovery as the journal has not yet been recovered */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1600 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1601 | status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1602 | if (status < 0) { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1603 | trace_ocfs2_replay_journal_lock_err(status); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1604 | if (status != -ERESTARTSYS) |
| 1605 | mlog(ML_ERROR, "Could not lock journal!\n"); |
| 1606 | goto done; |
| 1607 | } |
| 1608 | got_lock = 1; |
| 1609 | |
| 1610 | fe = (struct ocfs2_dinode *) bh->b_data; |
| 1611 | |
| 1612 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1613 | slot_reco_gen = ocfs2_get_recovery_generation(fe); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1614 | |
| 1615 | if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1616 | trace_ocfs2_replay_journal_skip(node_num); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1617 | /* Refresh recovery generation for the slot */ |
| 1618 | osb->slot_recovery_generations[slot_num] = slot_reco_gen; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1619 | goto done; |
| 1620 | } |
| 1621 | |
Srinivas Eeda | 9140db0 | 2009-03-06 14:21:46 -0800 | [diff] [blame] | 1622 | /* we need to run complete recovery for offline orphan slots */ |
| 1623 | ocfs2_replay_map_set_state(osb, REPLAY_NEEDED); |
| 1624 | |
Sunil Mushran | 619c200 | 2011-07-24 10:34:54 -0700 | [diff] [blame] | 1625 | printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\ |
| 1626 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), |
| 1627 | MINOR(osb->sb->s_dev)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1628 | |
| 1629 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); |
| 1630 | |
| 1631 | status = ocfs2_force_read_journal(inode); |
| 1632 | if (status < 0) { |
| 1633 | mlog_errno(status); |
| 1634 | goto done; |
| 1635 | } |
| 1636 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1637 | journal = jbd2_journal_init_inode(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1638 | if (journal == NULL) { |
| 1639 | mlog(ML_ERROR, "Linux journal layer error\n"); |
| 1640 | status = -EIO; |
| 1641 | goto done; |
| 1642 | } |
| 1643 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1644 | status = jbd2_journal_load(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1645 | if (status < 0) { |
| 1646 | mlog_errno(status); |
| 1647 | if (!igrab(inode)) |
| 1648 | BUG(); |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1649 | jbd2_journal_destroy(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1650 | goto done; |
| 1651 | } |
| 1652 | |
| 1653 | ocfs2_clear_journal_error(osb->sb, journal, slot_num); |
| 1654 | |
| 1655 | /* wipe the journal */ |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1656 | jbd2_journal_lock_updates(journal); |
| 1657 | status = jbd2_journal_flush(journal); |
| 1658 | jbd2_journal_unlock_updates(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1659 | if (status < 0) |
| 1660 | mlog_errno(status); |
| 1661 | |
| 1662 | /* This will mark the node clean */ |
| 1663 | flags = le32_to_cpu(fe->id1.journal1.ij_flags); |
| 1664 | flags &= ~OCFS2_JOURNAL_DIRTY_FL; |
| 1665 | fe->id1.journal1.ij_flags = cpu_to_le32(flags); |
| 1666 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1667 | /* Increment recovery generation to indicate successful recovery */ |
| 1668 | ocfs2_bump_recovery_generation(fe); |
| 1669 | osb->slot_recovery_generations[slot_num] = |
| 1670 | ocfs2_get_recovery_generation(fe); |
| 1671 | |
Joel Becker | 13723d0 | 2008-10-17 19:25:01 -0700 | [diff] [blame] | 1672 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check); |
Joel Becker | 8cb471e | 2009-02-10 20:00:41 -0800 | [diff] [blame] | 1673 | status = ocfs2_write_block(osb, bh, INODE_CACHE(inode)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1674 | if (status < 0) |
| 1675 | mlog_errno(status); |
| 1676 | |
| 1677 | if (!igrab(inode)) |
| 1678 | BUG(); |
| 1679 | |
Joel Becker | 2b4e30f | 2008-09-03 20:03:41 -0700 | [diff] [blame] | 1680 | jbd2_journal_destroy(journal); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1681 | |
Sunil Mushran | 619c200 | 2011-07-24 10:34:54 -0700 | [diff] [blame] | 1682 | printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\ |
| 1683 | "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev), |
| 1684 | MINOR(osb->sb->s_dev)); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1685 | done: |
| 1686 | /* drop the lock on this nodes journal */ |
| 1687 | if (got_lock) |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1688 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1689 | |
Joseph Qi | 72865d9 | 2016-01-14 15:17:27 -0800 | [diff] [blame] | 1690 | iput(inode); |
Mark Fasheh | a81cb88 | 2008-10-07 14:25:16 -0700 | [diff] [blame] | 1691 | brelse(bh); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1692 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1693 | return status; |
| 1694 | } |
| 1695 | |
| 1696 | /* |
| 1697 | * Do the most important parts of node recovery: |
| 1698 | * - Replay it's journal |
| 1699 | * - Stamp a clean local allocator file |
| 1700 | * - Stamp a clean truncate log |
| 1701 | * - Mark the node clean |
| 1702 | * |
| 1703 | * If this function completes without error, a node in OCFS2 can be |
| 1704 | * said to have been safely recovered. As a result, failure during the |
| 1705 | * second part of a nodes recovery process (local alloc recovery) is |
| 1706 | * far less concerning. |
| 1707 | */ |
| 1708 | static int ocfs2_recover_node(struct ocfs2_super *osb, |
Jan Kara | 2205363 | 2008-10-20 23:50:38 +0200 | [diff] [blame] | 1709 | int node_num, int slot_num) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1710 | { |
| 1711 | int status = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1712 | struct ocfs2_dinode *la_copy = NULL; |
| 1713 | struct ocfs2_dinode *tl_copy = NULL; |
| 1714 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1715 | trace_ocfs2_recover_node(node_num, slot_num, osb->node_num); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1716 | |
| 1717 | /* Should not ever be called to recover ourselves -- in that |
| 1718 | * case we should've called ocfs2_journal_load instead. */ |
Eric Sesterhenn / snakebyte | ebdec83 | 2006-01-27 10:32:52 +0100 | [diff] [blame] | 1719 | BUG_ON(osb->node_num == node_num); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1720 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1721 | status = ocfs2_replay_journal(osb, node_num, slot_num); |
| 1722 | if (status < 0) { |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1723 | if (status == -EBUSY) { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1724 | trace_ocfs2_recover_node_skip(slot_num, node_num); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1725 | status = 0; |
| 1726 | goto done; |
| 1727 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1728 | mlog_errno(status); |
| 1729 | goto done; |
| 1730 | } |
| 1731 | |
| 1732 | /* Stamp a clean local alloc file AFTER recovering the journal... */ |
| 1733 | status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy); |
| 1734 | if (status < 0) { |
| 1735 | mlog_errno(status); |
| 1736 | goto done; |
| 1737 | } |
| 1738 | |
| 1739 | /* An error from begin_truncate_log_recovery is not |
| 1740 | * serious enough to warrant halting the rest of |
| 1741 | * recovery. */ |
| 1742 | status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy); |
| 1743 | if (status < 0) |
| 1744 | mlog_errno(status); |
| 1745 | |
| 1746 | /* Likewise, this would be a strange but ultimately not so |
| 1747 | * harmful place to get an error... */ |
Mark Fasheh | 8e8a460 | 2008-02-01 11:59:09 -0800 | [diff] [blame] | 1748 | status = ocfs2_clear_slot(osb, slot_num); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1749 | if (status < 0) |
| 1750 | mlog_errno(status); |
| 1751 | |
| 1752 | /* This will kfree the memory pointed to by la_copy and tl_copy */ |
| 1753 | ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1754 | tl_copy, NULL, ORPHAN_NEED_TRUNCATE); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1755 | |
| 1756 | status = 0; |
| 1757 | done: |
| 1758 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1759 | return status; |
| 1760 | } |
| 1761 | |
| 1762 | /* Test node liveness by trylocking his journal. If we get the lock, |
| 1763 | * we drop it here. Return 0 if we got the lock, -EAGAIN if node is |
| 1764 | * still alive (we couldn't get the lock) and < 0 on error. */ |
| 1765 | static int ocfs2_trylock_journal(struct ocfs2_super *osb, |
| 1766 | int slot_num) |
| 1767 | { |
| 1768 | int status, flags; |
| 1769 | struct inode *inode = NULL; |
| 1770 | |
| 1771 | inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE, |
| 1772 | slot_num); |
| 1773 | if (inode == NULL) { |
| 1774 | mlog(ML_ERROR, "access error\n"); |
| 1775 | status = -EACCES; |
| 1776 | goto bail; |
| 1777 | } |
| 1778 | if (is_bad_inode(inode)) { |
| 1779 | mlog(ML_ERROR, "access error (bad inode)\n"); |
| 1780 | iput(inode); |
| 1781 | inode = NULL; |
| 1782 | status = -EACCES; |
| 1783 | goto bail; |
| 1784 | } |
| 1785 | SET_INODE_JOURNAL(inode); |
| 1786 | |
| 1787 | flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE; |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1788 | status = ocfs2_inode_lock_full(inode, NULL, 1, flags); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1789 | if (status < 0) { |
| 1790 | if (status != -EAGAIN) |
| 1791 | mlog_errno(status); |
| 1792 | goto bail; |
| 1793 | } |
| 1794 | |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 1795 | ocfs2_inode_unlock(inode, 1); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1796 | bail: |
Joseph Qi | 72865d9 | 2016-01-14 15:17:27 -0800 | [diff] [blame] | 1797 | iput(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1798 | |
| 1799 | return status; |
| 1800 | } |
| 1801 | |
| 1802 | /* Call this underneath ocfs2_super_lock. It also assumes that the |
| 1803 | * slot info struct has been updated from disk. */ |
| 1804 | int ocfs2_mark_dead_nodes(struct ocfs2_super *osb) |
| 1805 | { |
Joel Becker | d85b20e | 2008-02-01 12:01:05 -0800 | [diff] [blame] | 1806 | unsigned int node_num; |
| 1807 | int status, i; |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1808 | u32 gen; |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1809 | struct buffer_head *bh = NULL; |
| 1810 | struct ocfs2_dinode *di; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1811 | |
| 1812 | /* This is called with the super block cluster lock, so we |
| 1813 | * know that the slot map can't change underneath us. */ |
| 1814 | |
Joel Becker | d85b20e | 2008-02-01 12:01:05 -0800 | [diff] [blame] | 1815 | for (i = 0; i < osb->max_slots; i++) { |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1816 | /* Read journal inode to get the recovery generation */ |
| 1817 | status = ocfs2_read_journal_inode(osb, i, &bh, NULL); |
| 1818 | if (status) { |
| 1819 | mlog_errno(status); |
| 1820 | goto bail; |
| 1821 | } |
| 1822 | di = (struct ocfs2_dinode *)bh->b_data; |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1823 | gen = ocfs2_get_recovery_generation(di); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1824 | brelse(bh); |
| 1825 | bh = NULL; |
| 1826 | |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1827 | spin_lock(&osb->osb_lock); |
| 1828 | osb->slot_recovery_generations[i] = gen; |
| 1829 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1830 | trace_ocfs2_mark_dead_nodes(i, |
| 1831 | osb->slot_recovery_generations[i]); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 1832 | |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1833 | if (i == osb->slot_num) { |
| 1834 | spin_unlock(&osb->osb_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1835 | continue; |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1836 | } |
Joel Becker | d85b20e | 2008-02-01 12:01:05 -0800 | [diff] [blame] | 1837 | |
| 1838 | status = ocfs2_slot_to_node_num_locked(osb, i, &node_num); |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1839 | if (status == -ENOENT) { |
| 1840 | spin_unlock(&osb->osb_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1841 | continue; |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1842 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1843 | |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1844 | if (__ocfs2_recovery_map_test(osb, node_num)) { |
| 1845 | spin_unlock(&osb->osb_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1846 | continue; |
Mark Fasheh | a1af7d1 | 2008-08-19 17:20:28 -0700 | [diff] [blame] | 1847 | } |
Joel Becker | d85b20e | 2008-02-01 12:01:05 -0800 | [diff] [blame] | 1848 | spin_unlock(&osb->osb_lock); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1849 | |
| 1850 | /* Ok, we have a slot occupied by another node which |
| 1851 | * is not in the recovery map. We trylock his journal |
| 1852 | * file here to test if he's alive. */ |
| 1853 | status = ocfs2_trylock_journal(osb, i); |
| 1854 | if (!status) { |
| 1855 | /* Since we're called from mount, we know that |
| 1856 | * the recovery thread can't race us on |
| 1857 | * setting / checking the recovery bits. */ |
| 1858 | ocfs2_recovery_thread(osb, node_num); |
| 1859 | } else if ((status < 0) && (status != -EAGAIN)) { |
| 1860 | mlog_errno(status); |
| 1861 | goto bail; |
| 1862 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1863 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1864 | |
| 1865 | status = 0; |
| 1866 | bail: |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 1867 | return status; |
| 1868 | } |
| 1869 | |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1870 | /* |
| 1871 | * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some |
| 1872 | * randomness to the timeout to minimize multple nodes firing the timer at the |
| 1873 | * same time. |
| 1874 | */ |
| 1875 | static inline unsigned long ocfs2_orphan_scan_timeout(void) |
| 1876 | { |
| 1877 | unsigned long time; |
| 1878 | |
| 1879 | get_random_bytes(&time, sizeof(time)); |
| 1880 | time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000); |
| 1881 | return msecs_to_jiffies(time); |
| 1882 | } |
| 1883 | |
| 1884 | /* |
| 1885 | * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for |
| 1886 | * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This |
| 1887 | * is done to catch any orphans that are left over in orphan directories. |
| 1888 | * |
Sunil Mushran | a035bff | 2011-07-24 10:35:54 -0700 | [diff] [blame] | 1889 | * It scans all slots, even ones that are in use. It does so to handle the |
| 1890 | * case described below: |
| 1891 | * |
| 1892 | * Node 1 has an inode it was using. The dentry went away due to memory |
| 1893 | * pressure. Node 1 closes the inode, but it's on the free list. The node |
| 1894 | * has the open lock. |
| 1895 | * Node 2 unlinks the inode. It grabs the dentry lock to notify others, |
| 1896 | * but node 1 has no dentry and doesn't get the message. It trylocks the |
| 1897 | * open lock, sees that another node has a PR, and does nothing. |
| 1898 | * Later node 2 runs its orphan dir. It igets the inode, trylocks the |
| 1899 | * open lock, sees the PR still, and does nothing. |
| 1900 | * Basically, we have to trigger an orphan iput on node 1. The only way |
| 1901 | * for this to happen is if node 1 runs node 2's orphan dir. |
| 1902 | * |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1903 | * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT |
| 1904 | * seconds. It gets an EX lock on os_lockres and checks sequence number |
| 1905 | * stored in LVB. If the sequence number has changed, it means some other |
| 1906 | * node has done the scan. This node skips the scan and tracks the |
| 1907 | * sequence number. If the sequence number didn't change, it means a scan |
| 1908 | * hasn't happened. The node queues a scan and increments the |
| 1909 | * sequence number in the LVB. |
| 1910 | */ |
Joseph Qi | b519ea6 | 2015-06-24 16:55:34 -0700 | [diff] [blame] | 1911 | static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb) |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1912 | { |
| 1913 | struct ocfs2_orphan_scan *os; |
| 1914 | int status, i; |
| 1915 | u32 seqno = 0; |
| 1916 | |
| 1917 | os = &osb->osb_orphan_scan; |
| 1918 | |
Sunil Mushran | 692684e | 2009-06-19 16:53:17 -0700 | [diff] [blame] | 1919 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) |
| 1920 | goto out; |
| 1921 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1922 | trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno, |
| 1923 | atomic_read(&os->os_state)); |
| 1924 | |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 1925 | status = ocfs2_orphan_scan_lock(osb, &seqno); |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1926 | if (status < 0) { |
| 1927 | if (status != -EAGAIN) |
| 1928 | mlog_errno(status); |
| 1929 | goto out; |
| 1930 | } |
| 1931 | |
Sunil Mushran | 692684e | 2009-06-19 16:53:17 -0700 | [diff] [blame] | 1932 | /* Do no queue the tasks if the volume is being umounted */ |
| 1933 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE) |
| 1934 | goto unlock; |
| 1935 | |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1936 | if (os->os_seqno != seqno) { |
| 1937 | os->os_seqno = seqno; |
| 1938 | goto unlock; |
| 1939 | } |
| 1940 | |
| 1941 | for (i = 0; i < osb->max_slots; i++) |
| 1942 | ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 1943 | NULL, ORPHAN_NO_NEED_TRUNCATE); |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1944 | /* |
| 1945 | * We queued a recovery on orphan slots, increment the sequence |
| 1946 | * number and update LVB so other node will skip the scan for a while |
| 1947 | */ |
| 1948 | seqno++; |
Srinivas Eeda | 15633a2 | 2009-06-03 17:02:56 -0700 | [diff] [blame] | 1949 | os->os_count++; |
Deepa Dinamani | 395627b | 2016-12-12 16:41:23 -0800 | [diff] [blame] | 1950 | os->os_scantime = ktime_get_seconds(); |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1951 | unlock: |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 1952 | ocfs2_orphan_scan_unlock(osb, seqno); |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1953 | out: |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 1954 | trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno, |
| 1955 | atomic_read(&os->os_state)); |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1956 | return; |
| 1957 | } |
| 1958 | |
| 1959 | /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */ |
Joseph Qi | b519ea6 | 2015-06-24 16:55:34 -0700 | [diff] [blame] | 1960 | static void ocfs2_orphan_scan_work(struct work_struct *work) |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1961 | { |
| 1962 | struct ocfs2_orphan_scan *os; |
| 1963 | struct ocfs2_super *osb; |
| 1964 | |
| 1965 | os = container_of(work, struct ocfs2_orphan_scan, |
| 1966 | os_orphan_scan_work.work); |
| 1967 | osb = os->os_osb; |
| 1968 | |
| 1969 | mutex_lock(&os->os_lock); |
| 1970 | ocfs2_queue_orphan_scan(osb); |
Sunil Mushran | 692684e | 2009-06-19 16:53:17 -0700 | [diff] [blame] | 1971 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) |
jiangyiwen | 35ddf78 | 2016-03-25 14:21:32 -0700 | [diff] [blame] | 1972 | queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, |
Sunil Mushran | 692684e | 2009-06-19 16:53:17 -0700 | [diff] [blame] | 1973 | ocfs2_orphan_scan_timeout()); |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1974 | mutex_unlock(&os->os_lock); |
| 1975 | } |
| 1976 | |
| 1977 | void ocfs2_orphan_scan_stop(struct ocfs2_super *osb) |
| 1978 | { |
| 1979 | struct ocfs2_orphan_scan *os; |
| 1980 | |
| 1981 | os = &osb->osb_orphan_scan; |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 1982 | if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) { |
| 1983 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); |
| 1984 | mutex_lock(&os->os_lock); |
| 1985 | cancel_delayed_work(&os->os_orphan_scan_work); |
| 1986 | mutex_unlock(&os->os_lock); |
| 1987 | } |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1988 | } |
| 1989 | |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 1990 | void ocfs2_orphan_scan_init(struct ocfs2_super *osb) |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1991 | { |
| 1992 | struct ocfs2_orphan_scan *os; |
| 1993 | |
| 1994 | os = &osb->osb_orphan_scan; |
| 1995 | os->os_osb = osb; |
Srinivas Eeda | 15633a2 | 2009-06-03 17:02:56 -0700 | [diff] [blame] | 1996 | os->os_count = 0; |
Sunil Mushran | 3211949 | 2009-06-19 16:53:18 -0700 | [diff] [blame] | 1997 | os->os_seqno = 0; |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 1998 | mutex_init(&os->os_lock); |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 1999 | INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work); |
Jeff Mahoney | 8b712cd | 2009-07-07 17:22:12 -0400 | [diff] [blame] | 2000 | } |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 2001 | |
Jeff Mahoney | 8b712cd | 2009-07-07 17:22:12 -0400 | [diff] [blame] | 2002 | void ocfs2_orphan_scan_start(struct ocfs2_super *osb) |
| 2003 | { |
| 2004 | struct ocfs2_orphan_scan *os; |
| 2005 | |
| 2006 | os = &osb->osb_orphan_scan; |
Deepa Dinamani | 395627b | 2016-12-12 16:41:23 -0800 | [diff] [blame] | 2007 | os->os_scantime = ktime_get_seconds(); |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 2008 | if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) |
| 2009 | atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE); |
| 2010 | else { |
| 2011 | atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE); |
jiangyiwen | 35ddf78 | 2016-03-25 14:21:32 -0700 | [diff] [blame] | 2012 | queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work, |
Tao Ma | 40f165f | 2010-05-28 14:22:59 +0800 | [diff] [blame] | 2013 | ocfs2_orphan_scan_timeout()); |
Sunil Mushran | df152c2 | 2009-06-22 11:40:07 -0700 | [diff] [blame] | 2014 | } |
Srinivas Eeda | 8327393 | 2009-06-03 17:02:55 -0700 | [diff] [blame] | 2015 | } |
| 2016 | |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2017 | struct ocfs2_orphan_filldir_priv { |
Al Viro | 3704412b | 2013-05-22 21:06:00 -0400 | [diff] [blame] | 2018 | struct dir_context ctx; |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2019 | struct inode *head; |
| 2020 | struct ocfs2_super *osb; |
Joseph Qi | 30edc43 | 2015-11-05 18:44:01 -0800 | [diff] [blame] | 2021 | enum ocfs2_orphan_reco_type orphan_reco_type; |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2022 | }; |
| 2023 | |
Miklos Szeredi | ac7576f | 2014-10-30 17:37:34 +0100 | [diff] [blame] | 2024 | static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name, |
| 2025 | int name_len, loff_t pos, u64 ino, |
| 2026 | unsigned type) |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2027 | { |
Miklos Szeredi | ac7576f | 2014-10-30 17:37:34 +0100 | [diff] [blame] | 2028 | struct ocfs2_orphan_filldir_priv *p = |
| 2029 | container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx); |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2030 | struct inode *iter; |
| 2031 | |
| 2032 | if (name_len == 1 && !strncmp(".", name, 1)) |
| 2033 | return 0; |
| 2034 | if (name_len == 2 && !strncmp("..", name, 2)) |
| 2035 | return 0; |
| 2036 | |
Joseph Qi | 30edc43 | 2015-11-05 18:44:01 -0800 | [diff] [blame] | 2037 | /* do not include dio entry in case of orphan scan */ |
| 2038 | if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) && |
| 2039 | (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, |
| 2040 | OCFS2_DIO_ORPHAN_PREFIX_LEN))) |
| 2041 | return 0; |
| 2042 | |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2043 | /* Skip bad inodes so that recovery can continue */ |
| 2044 | iter = ocfs2_iget(p->osb, ino, |
Jan Kara | 5fa0613 | 2008-01-11 00:11:45 +0100 | [diff] [blame] | 2045 | OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0); |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2046 | if (IS_ERR(iter)) |
| 2047 | return 0; |
| 2048 | |
Joseph Qi | 93d911f | 2015-11-05 18:44:04 -0800 | [diff] [blame] | 2049 | if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX, |
| 2050 | OCFS2_DIO_ORPHAN_PREFIX_LEN)) |
| 2051 | OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY; |
| 2052 | |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 2053 | /* Skip inodes which are already added to recover list, since dio may |
| 2054 | * happen concurrently with unlink/rename */ |
| 2055 | if (OCFS2_I(iter)->ip_next_orphan) { |
| 2056 | iput(iter); |
| 2057 | return 0; |
| 2058 | } |
| 2059 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 2060 | trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno); |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2061 | /* No locking is required for the next_orphan queue as there |
| 2062 | * is only ever a single process doing orphan recovery. */ |
| 2063 | OCFS2_I(iter)->ip_next_orphan = p->head; |
| 2064 | p->head = iter; |
| 2065 | |
| 2066 | return 0; |
| 2067 | } |
| 2068 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2069 | static int ocfs2_queue_orphans(struct ocfs2_super *osb, |
| 2070 | int slot, |
Joseph Qi | 30edc43 | 2015-11-05 18:44:01 -0800 | [diff] [blame] | 2071 | struct inode **head, |
| 2072 | enum ocfs2_orphan_reco_type orphan_reco_type) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2073 | { |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2074 | int status; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2075 | struct inode *orphan_dir_inode = NULL; |
Al Viro | 3704412b | 2013-05-22 21:06:00 -0400 | [diff] [blame] | 2076 | struct ocfs2_orphan_filldir_priv priv = { |
| 2077 | .ctx.actor = ocfs2_orphan_filldir, |
| 2078 | .osb = osb, |
Joseph Qi | 30edc43 | 2015-11-05 18:44:01 -0800 | [diff] [blame] | 2079 | .head = *head, |
| 2080 | .orphan_reco_type = orphan_reco_type |
Al Viro | 3704412b | 2013-05-22 21:06:00 -0400 | [diff] [blame] | 2081 | }; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2082 | |
| 2083 | orphan_dir_inode = ocfs2_get_system_file_inode(osb, |
| 2084 | ORPHAN_DIR_SYSTEM_INODE, |
| 2085 | slot); |
| 2086 | if (!orphan_dir_inode) { |
| 2087 | status = -ENOENT; |
| 2088 | mlog_errno(status); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2089 | return status; |
Sunil Mushran | 2bd6321 | 2010-01-25 16:57:38 -0800 | [diff] [blame] | 2090 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2091 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 2092 | inode_lock(orphan_dir_inode); |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2093 | status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2094 | if (status < 0) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2095 | mlog_errno(status); |
| 2096 | goto out; |
| 2097 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2098 | |
Al Viro | 3704412b | 2013-05-22 21:06:00 -0400 | [diff] [blame] | 2099 | status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx); |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2100 | if (status) { |
| 2101 | mlog_errno(status); |
Mark Fasheh | a86370f | 2007-12-03 14:06:23 -0800 | [diff] [blame] | 2102 | goto out_cluster; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2103 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2104 | |
Mark Fasheh | 5eae5b9 | 2007-09-10 17:50:51 -0700 | [diff] [blame] | 2105 | *head = priv.head; |
| 2106 | |
Mark Fasheh | a86370f | 2007-12-03 14:06:23 -0800 | [diff] [blame] | 2107 | out_cluster: |
Mark Fasheh | e63aecb6 | 2007-10-18 15:30:42 -0700 | [diff] [blame] | 2108 | ocfs2_inode_unlock(orphan_dir_inode, 0); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2109 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 2110 | inode_unlock(orphan_dir_inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2111 | iput(orphan_dir_inode); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2112 | return status; |
| 2113 | } |
| 2114 | |
| 2115 | static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb, |
| 2116 | int slot) |
| 2117 | { |
| 2118 | int ret; |
| 2119 | |
| 2120 | spin_lock(&osb->osb_lock); |
| 2121 | ret = !osb->osb_orphan_wipes[slot]; |
| 2122 | spin_unlock(&osb->osb_lock); |
| 2123 | return ret; |
| 2124 | } |
| 2125 | |
| 2126 | static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb, |
| 2127 | int slot) |
| 2128 | { |
| 2129 | spin_lock(&osb->osb_lock); |
| 2130 | /* Mark ourselves such that new processes in delete_inode() |
| 2131 | * know to quit early. */ |
| 2132 | ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot); |
| 2133 | while (osb->osb_orphan_wipes[slot]) { |
| 2134 | /* If any processes are already in the middle of an |
| 2135 | * orphan wipe on this dir, then we need to wait for |
| 2136 | * them. */ |
| 2137 | spin_unlock(&osb->osb_lock); |
| 2138 | wait_event_interruptible(osb->osb_wipe_event, |
| 2139 | ocfs2_orphan_recovery_can_continue(osb, slot)); |
| 2140 | spin_lock(&osb->osb_lock); |
| 2141 | } |
| 2142 | spin_unlock(&osb->osb_lock); |
| 2143 | } |
| 2144 | |
| 2145 | static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb, |
| 2146 | int slot) |
| 2147 | { |
| 2148 | ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot); |
| 2149 | } |
| 2150 | |
| 2151 | /* |
| 2152 | * Orphan recovery. Each mounted node has it's own orphan dir which we |
| 2153 | * must run during recovery. Our strategy here is to build a list of |
| 2154 | * the inodes in the orphan dir and iget/iput them. The VFS does |
| 2155 | * (most) of the rest of the work. |
| 2156 | * |
| 2157 | * Orphan recovery can happen at any time, not just mount so we have a |
| 2158 | * couple of extra considerations. |
| 2159 | * |
| 2160 | * - We grab as many inodes as we can under the orphan dir lock - |
| 2161 | * doing iget() outside the orphan dir risks getting a reference on |
| 2162 | * an invalid inode. |
| 2163 | * - We must be sure not to deadlock with other processes on the |
| 2164 | * system wanting to run delete_inode(). This can happen when they go |
| 2165 | * to lock the orphan dir and the orphan recovery process attempts to |
| 2166 | * iget() inside the orphan dir lock. This can be avoided by |
| 2167 | * advertising our state to ocfs2_delete_inode(). |
| 2168 | */ |
| 2169 | static int ocfs2_recover_orphans(struct ocfs2_super *osb, |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 2170 | int slot, |
| 2171 | enum ocfs2_orphan_reco_type orphan_reco_type) |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2172 | { |
| 2173 | int ret = 0; |
| 2174 | struct inode *inode = NULL; |
| 2175 | struct inode *iter; |
| 2176 | struct ocfs2_inode_info *oi; |
Joseph Qi | cf1776a | 2015-06-24 16:54:59 -0700 | [diff] [blame] | 2177 | struct buffer_head *di_bh = NULL; |
| 2178 | struct ocfs2_dinode *di = NULL; |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2179 | |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 2180 | trace_ocfs2_recover_orphans(slot); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2181 | |
| 2182 | ocfs2_mark_recovering_orphan_dir(osb, slot); |
Joseph Qi | 30edc43 | 2015-11-05 18:44:01 -0800 | [diff] [blame] | 2183 | ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type); |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2184 | ocfs2_clear_recovering_orphan_dir(osb, slot); |
| 2185 | |
| 2186 | /* Error here should be noted, but we want to continue with as |
| 2187 | * many queued inodes as we've got. */ |
| 2188 | if (ret) |
| 2189 | mlog_errno(ret); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2190 | |
| 2191 | while (inode) { |
| 2192 | oi = OCFS2_I(inode); |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 2193 | trace_ocfs2_recover_orphans_iput( |
| 2194 | (unsigned long long)oi->ip_blkno); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2195 | |
| 2196 | iter = oi->ip_next_orphan; |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 2197 | oi->ip_next_orphan = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2198 | |
Joseph Qi | 93d911f | 2015-11-05 18:44:04 -0800 | [diff] [blame] | 2199 | if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) { |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 2200 | inode_lock(inode); |
Joseph Qi | 93d911f | 2015-11-05 18:44:04 -0800 | [diff] [blame] | 2201 | ret = ocfs2_rw_lock(inode, 1); |
| 2202 | if (ret < 0) { |
| 2203 | mlog_errno(ret); |
| 2204 | goto unlock_mutex; |
| 2205 | } |
| 2206 | /* |
| 2207 | * We need to take and drop the inode lock to |
| 2208 | * force read inode from disk. |
| 2209 | */ |
| 2210 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
| 2211 | if (ret) { |
| 2212 | mlog_errno(ret); |
| 2213 | goto unlock_rw; |
| 2214 | } |
Joseph Qi | cf1776a | 2015-06-24 16:54:59 -0700 | [diff] [blame] | 2215 | |
Joseph Qi | 93d911f | 2015-11-05 18:44:04 -0800 | [diff] [blame] | 2216 | di = (struct ocfs2_dinode *)di_bh->b_data; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2217 | |
Joseph Qi | 93d911f | 2015-11-05 18:44:04 -0800 | [diff] [blame] | 2218 | if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) { |
| 2219 | ret = ocfs2_truncate_file(inode, di_bh, |
| 2220 | i_size_read(inode)); |
| 2221 | if (ret < 0) { |
| 2222 | if (ret != -ENOSPC) |
| 2223 | mlog_errno(ret); |
| 2224 | goto unlock_inode; |
| 2225 | } |
| 2226 | |
| 2227 | ret = ocfs2_del_inode_from_orphan(osb, inode, |
| 2228 | di_bh, 0, 0); |
| 2229 | if (ret) |
| 2230 | mlog_errno(ret); |
| 2231 | } |
| 2232 | unlock_inode: |
| 2233 | ocfs2_inode_unlock(inode, 1); |
| 2234 | brelse(di_bh); |
| 2235 | di_bh = NULL; |
| 2236 | unlock_rw: |
| 2237 | ocfs2_rw_unlock(inode, 1); |
| 2238 | unlock_mutex: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 2239 | inode_unlock(inode); |
Joseph Qi | 93d911f | 2015-11-05 18:44:04 -0800 | [diff] [blame] | 2240 | |
| 2241 | /* clear dio flag in ocfs2_inode_info */ |
| 2242 | oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY; |
| 2243 | } else { |
Joseph Qi | ed460cf | 2015-02-16 15:59:57 -0800 | [diff] [blame] | 2244 | spin_lock(&oi->ip_lock); |
| 2245 | /* Set the proper information to get us going into |
| 2246 | * ocfs2_delete_inode. */ |
| 2247 | oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED; |
| 2248 | spin_unlock(&oi->ip_lock); |
Joseph Qi | ad69482 | 2015-09-04 15:44:40 -0700 | [diff] [blame] | 2249 | } |
| 2250 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2251 | iput(inode); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2252 | inode = iter; |
| 2253 | } |
| 2254 | |
Mark Fasheh | b4df6ed | 2006-02-22 17:35:08 -0800 | [diff] [blame] | 2255 | return ret; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2256 | } |
| 2257 | |
Jan Kara | 19ece54 | 2008-08-21 20:13:17 +0200 | [diff] [blame] | 2258 | static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota) |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2259 | { |
| 2260 | /* This check is good because ocfs2 will wait on our recovery |
| 2261 | * thread before changing it to something other than MOUNTED |
| 2262 | * or DISABLED. */ |
| 2263 | wait_event(osb->osb_mount_event, |
Jan Kara | 19ece54 | 2008-08-21 20:13:17 +0200 | [diff] [blame] | 2264 | (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) || |
| 2265 | atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS || |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2266 | atomic_read(&osb->vol_state) == VOLUME_DISABLED); |
| 2267 | |
| 2268 | /* If there's an error on mount, then we may never get to the |
| 2269 | * MOUNTED flag, but this is set right before |
| 2270 | * dismount_volume() so we can trust it. */ |
| 2271 | if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) { |
Tao Ma | b410795 | 2011-02-24 14:15:35 +0800 | [diff] [blame] | 2272 | trace_ocfs2_wait_on_mount(VOLUME_DISABLED); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2273 | mlog(0, "mount error, exiting!\n"); |
| 2274 | return -EBUSY; |
| 2275 | } |
| 2276 | |
| 2277 | return 0; |
| 2278 | } |
| 2279 | |
| 2280 | static int ocfs2_commit_thread(void *arg) |
| 2281 | { |
| 2282 | int status; |
| 2283 | struct ocfs2_super *osb = arg; |
| 2284 | struct ocfs2_journal *journal = osb->journal; |
| 2285 | |
| 2286 | /* we can trust j_num_trans here because _should_stop() is only set in |
| 2287 | * shutdown and nobody other than ourselves should be able to start |
| 2288 | * transactions. committing on shutdown might take a few iterations |
| 2289 | * as final transactions put deleted inodes on the list */ |
| 2290 | while (!(kthread_should_stop() && |
| 2291 | atomic_read(&journal->j_num_trans) == 0)) { |
| 2292 | |
Mark Fasheh | 745ae8ba | 2006-02-09 13:23:39 -0800 | [diff] [blame] | 2293 | wait_event_interruptible(osb->checkpoint_event, |
| 2294 | atomic_read(&journal->j_num_trans) |
| 2295 | || kthread_should_stop()); |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2296 | |
| 2297 | status = ocfs2_commit_cache(osb); |
Joseph Qi | 55b465b | 2014-06-04 16:06:09 -0700 | [diff] [blame] | 2298 | if (status < 0) { |
| 2299 | static unsigned long abort_warn_time; |
| 2300 | |
| 2301 | /* Warn about this once per minute */ |
| 2302 | if (printk_timed_ratelimit(&abort_warn_time, 60*HZ)) |
| 2303 | mlog(ML_ERROR, "status = %d, journal is " |
| 2304 | "already aborted.\n", status); |
| 2305 | /* |
| 2306 | * After ocfs2_commit_cache() fails, j_num_trans has a |
| 2307 | * non-zero value. Sleep here to avoid a busy-wait |
| 2308 | * loop. |
| 2309 | */ |
| 2310 | msleep_interruptible(1000); |
| 2311 | } |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2312 | |
| 2313 | if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){ |
| 2314 | mlog(ML_KTHREAD, |
| 2315 | "commit_thread: %u transactions pending on " |
| 2316 | "shutdown\n", |
| 2317 | atomic_read(&journal->j_num_trans)); |
| 2318 | } |
| 2319 | } |
| 2320 | |
| 2321 | return 0; |
| 2322 | } |
| 2323 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2324 | /* Reads all the journal inodes without taking any cluster locks. Used |
| 2325 | * for hard readonly access to determine whether any journal requires |
| 2326 | * recovery. Also used to refresh the recovery generation numbers after |
| 2327 | * a journal has been recovered by another node. |
| 2328 | */ |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2329 | int ocfs2_check_journals_nolocks(struct ocfs2_super *osb) |
| 2330 | { |
| 2331 | int ret = 0; |
| 2332 | unsigned int slot; |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2333 | struct buffer_head *di_bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2334 | struct ocfs2_dinode *di; |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2335 | int journal_dirty = 0; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2336 | |
| 2337 | for(slot = 0; slot < osb->max_slots; slot++) { |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2338 | ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL); |
| 2339 | if (ret) { |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2340 | mlog_errno(ret); |
| 2341 | goto out; |
| 2342 | } |
| 2343 | |
| 2344 | di = (struct ocfs2_dinode *) di_bh->b_data; |
| 2345 | |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2346 | osb->slot_recovery_generations[slot] = |
| 2347 | ocfs2_get_recovery_generation(di); |
| 2348 | |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2349 | if (le32_to_cpu(di->id1.journal1.ij_flags) & |
| 2350 | OCFS2_JOURNAL_DIRTY_FL) |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2351 | journal_dirty = 1; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2352 | |
| 2353 | brelse(di_bh); |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2354 | di_bh = NULL; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2355 | } |
| 2356 | |
| 2357 | out: |
Sunil Mushran | 539d826 | 2008-07-14 17:31:10 -0700 | [diff] [blame] | 2358 | if (journal_dirty) |
| 2359 | ret = -EROFS; |
Mark Fasheh | ccd979b | 2005-12-15 14:31:24 -0800 | [diff] [blame] | 2360 | return ret; |
| 2361 | } |