Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * journal.c
0004  *
0005  * Defines functions of journalling api
0006  *
0007  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
0008  */
0009 
0010 #include <linux/fs.h>
0011 #include <linux/types.h>
0012 #include <linux/slab.h>
0013 #include <linux/highmem.h>
0014 #include <linux/kthread.h>
0015 #include <linux/time.h>
0016 #include <linux/random.h>
0017 #include <linux/delay.h>
0018 
0019 #include <cluster/masklog.h>
0020 
0021 #include "ocfs2.h"
0022 
0023 #include "alloc.h"
0024 #include "blockcheck.h"
0025 #include "dir.h"
0026 #include "dlmglue.h"
0027 #include "extent_map.h"
0028 #include "heartbeat.h"
0029 #include "inode.h"
0030 #include "journal.h"
0031 #include "localalloc.h"
0032 #include "slot_map.h"
0033 #include "super.h"
0034 #include "sysfile.h"
0035 #include "uptodate.h"
0036 #include "quota.h"
0037 #include "file.h"
0038 #include "namei.h"
0039 
0040 #include "buffer_head_io.h"
0041 #include "ocfs2_trace.h"
0042 
0043 DEFINE_SPINLOCK(trans_inc_lock);
0044 
0045 #define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
0046 
0047 static int ocfs2_force_read_journal(struct inode *inode);
0048 static int ocfs2_recover_node(struct ocfs2_super *osb,
0049                   int node_num, int slot_num);
0050 static int __ocfs2_recovery_thread(void *arg);
0051 static int ocfs2_commit_cache(struct ocfs2_super *osb);
0052 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
0053 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
0054                       int dirty, int replayed);
0055 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
0056                  int slot_num);
0057 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
0058                  int slot,
0059                  enum ocfs2_orphan_reco_type orphan_reco_type);
0060 static int ocfs2_commit_thread(void *arg);
0061 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
0062                         int slot_num,
0063                         struct ocfs2_dinode *la_dinode,
0064                         struct ocfs2_dinode *tl_dinode,
0065                         struct ocfs2_quota_recovery *qrec,
0066                         enum ocfs2_orphan_reco_type orphan_reco_type);
0067 
0068 static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
0069 {
0070     return __ocfs2_wait_on_mount(osb, 0);
0071 }
0072 
0073 static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
0074 {
0075     return __ocfs2_wait_on_mount(osb, 1);
0076 }
0077 
0078 /*
0079  * This replay_map is to track online/offline slots, so we could recover
0080  * offline slots during recovery and mount
0081  */
0082 
0083 enum ocfs2_replay_state {
0084     REPLAY_UNNEEDED = 0,    /* Replay is not needed, so ignore this map */
0085     REPLAY_NEEDED,      /* Replay slots marked in rm_replay_slots */
0086     REPLAY_DONE         /* Replay was already queued */
0087 };
0088 
0089 struct ocfs2_replay_map {
0090     unsigned int rm_slots;
0091     enum ocfs2_replay_state rm_state;
0092     unsigned char rm_replay_slots[];
0093 };
0094 
0095 static void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
0096 {
0097     if (!osb->replay_map)
0098         return;
0099 
0100     /* If we've already queued the replay, we don't have any more to do */
0101     if (osb->replay_map->rm_state == REPLAY_DONE)
0102         return;
0103 
0104     osb->replay_map->rm_state = state;
0105 }
0106 
0107 int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
0108 {
0109     struct ocfs2_replay_map *replay_map;
0110     int i, node_num;
0111 
0112     /* If replay map is already set, we don't do it again */
0113     if (osb->replay_map)
0114         return 0;
0115 
0116     replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
0117                  (osb->max_slots * sizeof(char)), GFP_KERNEL);
0118 
0119     if (!replay_map) {
0120         mlog_errno(-ENOMEM);
0121         return -ENOMEM;
0122     }
0123 
0124     spin_lock(&osb->osb_lock);
0125 
0126     replay_map->rm_slots = osb->max_slots;
0127     replay_map->rm_state = REPLAY_UNNEEDED;
0128 
0129     /* set rm_replay_slots for offline slot(s) */
0130     for (i = 0; i < replay_map->rm_slots; i++) {
0131         if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
0132             replay_map->rm_replay_slots[i] = 1;
0133     }
0134 
0135     osb->replay_map = replay_map;
0136     spin_unlock(&osb->osb_lock);
0137     return 0;
0138 }
0139 
0140 static void ocfs2_queue_replay_slots(struct ocfs2_super *osb,
0141         enum ocfs2_orphan_reco_type orphan_reco_type)
0142 {
0143     struct ocfs2_replay_map *replay_map = osb->replay_map;
0144     int i;
0145 
0146     if (!replay_map)
0147         return;
0148 
0149     if (replay_map->rm_state != REPLAY_NEEDED)
0150         return;
0151 
0152     for (i = 0; i < replay_map->rm_slots; i++)
0153         if (replay_map->rm_replay_slots[i])
0154             ocfs2_queue_recovery_completion(osb->journal, i, NULL,
0155                             NULL, NULL,
0156                             orphan_reco_type);
0157     replay_map->rm_state = REPLAY_DONE;
0158 }
0159 
0160 static void ocfs2_free_replay_slots(struct ocfs2_super *osb)
0161 {
0162     struct ocfs2_replay_map *replay_map = osb->replay_map;
0163 
0164     if (!osb->replay_map)
0165         return;
0166 
0167     kfree(replay_map);
0168     osb->replay_map = NULL;
0169 }
0170 
0171 int ocfs2_recovery_init(struct ocfs2_super *osb)
0172 {
0173     struct ocfs2_recovery_map *rm;
0174 
0175     mutex_init(&osb->recovery_lock);
0176     osb->disable_recovery = 0;
0177     osb->recovery_thread_task = NULL;
0178     init_waitqueue_head(&osb->recovery_event);
0179 
0180     rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
0181              osb->max_slots * sizeof(unsigned int),
0182              GFP_KERNEL);
0183     if (!rm) {
0184         mlog_errno(-ENOMEM);
0185         return -ENOMEM;
0186     }
0187 
0188     rm->rm_entries = (unsigned int *)((char *)rm +
0189                       sizeof(struct ocfs2_recovery_map));
0190     osb->recovery_map = rm;
0191 
0192     return 0;
0193 }
0194 
0195 /* we can't grab the goofy sem lock from inside wait_event, so we use
0196  * memory barriers to make sure that we'll see the null task before
0197  * being woken up */
0198 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
0199 {
0200     mb();
0201     return osb->recovery_thread_task != NULL;
0202 }
0203 
0204 void ocfs2_recovery_exit(struct ocfs2_super *osb)
0205 {
0206     struct ocfs2_recovery_map *rm;
0207 
0208     /* disable any new recovery threads and wait for any currently
0209      * running ones to exit. Do this before setting the vol_state. */
0210     mutex_lock(&osb->recovery_lock);
0211     osb->disable_recovery = 1;
0212     mutex_unlock(&osb->recovery_lock);
0213     wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
0214 
0215     /* At this point, we know that no more recovery threads can be
0216      * launched, so wait for any recovery completion work to
0217      * complete. */
0218     if (osb->ocfs2_wq)
0219         flush_workqueue(osb->ocfs2_wq);
0220 
0221     /*
0222      * Now that recovery is shut down, and the osb is about to be
0223      * freed,  the osb_lock is not taken here.
0224      */
0225     rm = osb->recovery_map;
0226     /* XXX: Should we bug if there are dirty entries? */
0227 
0228     kfree(rm);
0229 }
0230 
0231 static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
0232                      unsigned int node_num)
0233 {
0234     int i;
0235     struct ocfs2_recovery_map *rm = osb->recovery_map;
0236 
0237     assert_spin_locked(&osb->osb_lock);
0238 
0239     for (i = 0; i < rm->rm_used; i++) {
0240         if (rm->rm_entries[i] == node_num)
0241             return 1;
0242     }
0243 
0244     return 0;
0245 }
0246 
0247 /* Behaves like test-and-set.  Returns the previous value */
0248 static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
0249                   unsigned int node_num)
0250 {
0251     struct ocfs2_recovery_map *rm = osb->recovery_map;
0252 
0253     spin_lock(&osb->osb_lock);
0254     if (__ocfs2_recovery_map_test(osb, node_num)) {
0255         spin_unlock(&osb->osb_lock);
0256         return 1;
0257     }
0258 
0259     /* XXX: Can this be exploited? Not from o2dlm... */
0260     BUG_ON(rm->rm_used >= osb->max_slots);
0261 
0262     rm->rm_entries[rm->rm_used] = node_num;
0263     rm->rm_used++;
0264     spin_unlock(&osb->osb_lock);
0265 
0266     return 0;
0267 }
0268 
0269 static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
0270                      unsigned int node_num)
0271 {
0272     int i;
0273     struct ocfs2_recovery_map *rm = osb->recovery_map;
0274 
0275     spin_lock(&osb->osb_lock);
0276 
0277     for (i = 0; i < rm->rm_used; i++) {
0278         if (rm->rm_entries[i] == node_num)
0279             break;
0280     }
0281 
0282     if (i < rm->rm_used) {
0283         /* XXX: be careful with the pointer math */
0284         memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
0285             (rm->rm_used - i - 1) * sizeof(unsigned int));
0286         rm->rm_used--;
0287     }
0288 
0289     spin_unlock(&osb->osb_lock);
0290 }
0291 
0292 static int ocfs2_commit_cache(struct ocfs2_super *osb)
0293 {
0294     int status = 0;
0295     unsigned int flushed;
0296     struct ocfs2_journal *journal = NULL;
0297 
0298     journal = osb->journal;
0299 
0300     /* Flush all pending commits and checkpoint the journal. */
0301     down_write(&journal->j_trans_barrier);
0302 
0303     flushed = atomic_read(&journal->j_num_trans);
0304     trace_ocfs2_commit_cache_begin(flushed);
0305     if (flushed == 0) {
0306         up_write(&journal->j_trans_barrier);
0307         goto finally;
0308     }
0309 
0310     jbd2_journal_lock_updates(journal->j_journal);
0311     status = jbd2_journal_flush(journal->j_journal, 0);
0312     jbd2_journal_unlock_updates(journal->j_journal);
0313     if (status < 0) {
0314         up_write(&journal->j_trans_barrier);
0315         mlog_errno(status);
0316         goto finally;
0317     }
0318 
0319     ocfs2_inc_trans_id(journal);
0320 
0321     flushed = atomic_read(&journal->j_num_trans);
0322     atomic_set(&journal->j_num_trans, 0);
0323     up_write(&journal->j_trans_barrier);
0324 
0325     trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
0326 
0327     ocfs2_wake_downconvert_thread(osb);
0328     wake_up(&journal->j_checkpointed);
0329 finally:
0330     return status;
0331 }
0332 
0333 handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
0334 {
0335     journal_t *journal = osb->journal->j_journal;
0336     handle_t *handle;
0337 
0338     BUG_ON(!osb || !osb->journal->j_journal);
0339 
0340     if (ocfs2_is_hard_readonly(osb))
0341         return ERR_PTR(-EROFS);
0342 
0343     BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
0344     BUG_ON(max_buffs <= 0);
0345 
0346     /* Nested transaction? Just return the handle... */
0347     if (journal_current_handle())
0348         return jbd2_journal_start(journal, max_buffs);
0349 
0350     sb_start_intwrite(osb->sb);
0351 
0352     down_read(&osb->journal->j_trans_barrier);
0353 
0354     handle = jbd2_journal_start(journal, max_buffs);
0355     if (IS_ERR(handle)) {
0356         up_read(&osb->journal->j_trans_barrier);
0357         sb_end_intwrite(osb->sb);
0358 
0359         mlog_errno(PTR_ERR(handle));
0360 
0361         if (is_journal_aborted(journal)) {
0362             ocfs2_abort(osb->sb, "Detected aborted journal\n");
0363             handle = ERR_PTR(-EROFS);
0364         }
0365     } else {
0366         if (!ocfs2_mount_local(osb))
0367             atomic_inc(&(osb->journal->j_num_trans));
0368     }
0369 
0370     return handle;
0371 }
0372 
0373 int ocfs2_commit_trans(struct ocfs2_super *osb,
0374                handle_t *handle)
0375 {
0376     int ret, nested;
0377     struct ocfs2_journal *journal = osb->journal;
0378 
0379     BUG_ON(!handle);
0380 
0381     nested = handle->h_ref > 1;
0382     ret = jbd2_journal_stop(handle);
0383     if (ret < 0)
0384         mlog_errno(ret);
0385 
0386     if (!nested) {
0387         up_read(&journal->j_trans_barrier);
0388         sb_end_intwrite(osb->sb);
0389     }
0390 
0391     return ret;
0392 }
0393 
0394 /*
0395  * 'nblocks' is what you want to add to the current transaction.
0396  *
0397  * This might call jbd2_journal_restart() which will commit dirty buffers
0398  * and then restart the transaction. Before calling
0399  * ocfs2_extend_trans(), any changed blocks should have been
0400  * dirtied. After calling it, all blocks which need to be changed must
0401  * go through another set of journal_access/journal_dirty calls.
0402  *
0403  * WARNING: This will not release any semaphores or disk locks taken
0404  * during the transaction, so make sure they were taken *before*
0405  * start_trans or we'll have ordering deadlocks.
0406  *
0407  * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
0408  * good because transaction ids haven't yet been recorded on the
0409  * cluster locks associated with this handle.
0410  */
0411 int ocfs2_extend_trans(handle_t *handle, int nblocks)
0412 {
0413     int status, old_nblocks;
0414 
0415     BUG_ON(!handle);
0416     BUG_ON(nblocks < 0);
0417 
0418     if (!nblocks)
0419         return 0;
0420 
0421     old_nblocks = jbd2_handle_buffer_credits(handle);
0422 
0423     trace_ocfs2_extend_trans(old_nblocks, nblocks);
0424 
0425 #ifdef CONFIG_OCFS2_DEBUG_FS
0426     status = 1;
0427 #else
0428     status = jbd2_journal_extend(handle, nblocks, 0);
0429     if (status < 0) {
0430         mlog_errno(status);
0431         goto bail;
0432     }
0433 #endif
0434 
0435     if (status > 0) {
0436         trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
0437         status = jbd2_journal_restart(handle,
0438                           old_nblocks + nblocks);
0439         if (status < 0) {
0440             mlog_errno(status);
0441             goto bail;
0442         }
0443     }
0444 
0445     status = 0;
0446 bail:
0447     return status;
0448 }
0449 
0450 /*
0451  * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
0452  * If that fails, restart the transaction & regain write access for the
0453  * buffer head which is used for metadata modifications.
0454  * Taken from Ext4: extend_or_restart_transaction()
0455  */
0456 int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
0457 {
0458     int status, old_nblks;
0459 
0460     BUG_ON(!handle);
0461 
0462     old_nblks = jbd2_handle_buffer_credits(handle);
0463     trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
0464 
0465     if (old_nblks < thresh)
0466         return 0;
0467 
0468     status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA, 0);
0469     if (status < 0) {
0470         mlog_errno(status);
0471         goto bail;
0472     }
0473 
0474     if (status > 0) {
0475         status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
0476         if (status < 0)
0477             mlog_errno(status);
0478     }
0479 
0480 bail:
0481     return status;
0482 }
0483 
0484 
0485 struct ocfs2_triggers {
0486     struct jbd2_buffer_trigger_type ot_triggers;
0487     int             ot_offset;
0488 };
0489 
0490 static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
0491 {
0492     return container_of(triggers, struct ocfs2_triggers, ot_triggers);
0493 }
0494 
0495 static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
0496                  struct buffer_head *bh,
0497                  void *data, size_t size)
0498 {
0499     struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
0500 
0501     /*
0502      * We aren't guaranteed to have the superblock here, so we
0503      * must unconditionally compute the ecc data.
0504      * __ocfs2_journal_access() will only set the triggers if
0505      * metaecc is enabled.
0506      */
0507     ocfs2_block_check_compute(data, size, data + ot->ot_offset);
0508 }
0509 
0510 /*
0511  * Quota blocks have their own trigger because the struct ocfs2_block_check
0512  * offset depends on the blocksize.
0513  */
0514 static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
0515                  struct buffer_head *bh,
0516                  void *data, size_t size)
0517 {
0518     struct ocfs2_disk_dqtrailer *dqt =
0519         ocfs2_block_dqtrailer(size, data);
0520 
0521     /*
0522      * We aren't guaranteed to have the superblock here, so we
0523      * must unconditionally compute the ecc data.
0524      * __ocfs2_journal_access() will only set the triggers if
0525      * metaecc is enabled.
0526      */
0527     ocfs2_block_check_compute(data, size, &dqt->dq_check);
0528 }
0529 
0530 /*
0531  * Directory blocks also have their own trigger because the
0532  * struct ocfs2_block_check offset depends on the blocksize.
0533  */
0534 static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
0535                  struct buffer_head *bh,
0536                  void *data, size_t size)
0537 {
0538     struct ocfs2_dir_block_trailer *trailer =
0539         ocfs2_dir_trailer_from_size(size, data);
0540 
0541     /*
0542      * We aren't guaranteed to have the superblock here, so we
0543      * must unconditionally compute the ecc data.
0544      * __ocfs2_journal_access() will only set the triggers if
0545      * metaecc is enabled.
0546      */
0547     ocfs2_block_check_compute(data, size, &trailer->db_check);
0548 }
0549 
0550 static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
0551                 struct buffer_head *bh)
0552 {
0553     mlog(ML_ERROR,
0554          "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
0555          "bh->b_blocknr = %llu\n",
0556          (unsigned long)bh,
0557          (unsigned long long)bh->b_blocknr);
0558 
0559     ocfs2_error(bh->b_bdev->bd_super,
0560             "JBD2 has aborted our journal, ocfs2 cannot continue\n");
0561 }
0562 
0563 static struct ocfs2_triggers di_triggers = {
0564     .ot_triggers = {
0565         .t_frozen = ocfs2_frozen_trigger,
0566         .t_abort = ocfs2_abort_trigger,
0567     },
0568     .ot_offset  = offsetof(struct ocfs2_dinode, i_check),
0569 };
0570 
0571 static struct ocfs2_triggers eb_triggers = {
0572     .ot_triggers = {
0573         .t_frozen = ocfs2_frozen_trigger,
0574         .t_abort = ocfs2_abort_trigger,
0575     },
0576     .ot_offset  = offsetof(struct ocfs2_extent_block, h_check),
0577 };
0578 
0579 static struct ocfs2_triggers rb_triggers = {
0580     .ot_triggers = {
0581         .t_frozen = ocfs2_frozen_trigger,
0582         .t_abort = ocfs2_abort_trigger,
0583     },
0584     .ot_offset  = offsetof(struct ocfs2_refcount_block, rf_check),
0585 };
0586 
0587 static struct ocfs2_triggers gd_triggers = {
0588     .ot_triggers = {
0589         .t_frozen = ocfs2_frozen_trigger,
0590         .t_abort = ocfs2_abort_trigger,
0591     },
0592     .ot_offset  = offsetof(struct ocfs2_group_desc, bg_check),
0593 };
0594 
0595 static struct ocfs2_triggers db_triggers = {
0596     .ot_triggers = {
0597         .t_frozen = ocfs2_db_frozen_trigger,
0598         .t_abort = ocfs2_abort_trigger,
0599     },
0600 };
0601 
0602 static struct ocfs2_triggers xb_triggers = {
0603     .ot_triggers = {
0604         .t_frozen = ocfs2_frozen_trigger,
0605         .t_abort = ocfs2_abort_trigger,
0606     },
0607     .ot_offset  = offsetof(struct ocfs2_xattr_block, xb_check),
0608 };
0609 
0610 static struct ocfs2_triggers dq_triggers = {
0611     .ot_triggers = {
0612         .t_frozen = ocfs2_dq_frozen_trigger,
0613         .t_abort = ocfs2_abort_trigger,
0614     },
0615 };
0616 
0617 static struct ocfs2_triggers dr_triggers = {
0618     .ot_triggers = {
0619         .t_frozen = ocfs2_frozen_trigger,
0620         .t_abort = ocfs2_abort_trigger,
0621     },
0622     .ot_offset  = offsetof(struct ocfs2_dx_root_block, dr_check),
0623 };
0624 
0625 static struct ocfs2_triggers dl_triggers = {
0626     .ot_triggers = {
0627         .t_frozen = ocfs2_frozen_trigger,
0628         .t_abort = ocfs2_abort_trigger,
0629     },
0630     .ot_offset  = offsetof(struct ocfs2_dx_leaf, dl_check),
0631 };
0632 
0633 static int __ocfs2_journal_access(handle_t *handle,
0634                   struct ocfs2_caching_info *ci,
0635                   struct buffer_head *bh,
0636                   struct ocfs2_triggers *triggers,
0637                   int type)
0638 {
0639     int status;
0640     struct ocfs2_super *osb =
0641         OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
0642 
0643     BUG_ON(!ci || !ci->ci_ops);
0644     BUG_ON(!handle);
0645     BUG_ON(!bh);
0646 
0647     trace_ocfs2_journal_access(
0648         (unsigned long long)ocfs2_metadata_cache_owner(ci),
0649         (unsigned long long)bh->b_blocknr, type, bh->b_size);
0650 
0651     /* we can safely remove this assertion after testing. */
0652     if (!buffer_uptodate(bh)) {
0653         mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
0654         mlog(ML_ERROR, "b_blocknr=%llu, b_state=0x%lx\n",
0655              (unsigned long long)bh->b_blocknr, bh->b_state);
0656 
0657         lock_buffer(bh);
0658         /*
0659          * A previous transaction with a couple of buffer heads fail
0660          * to checkpoint, so all the bhs are marked as BH_Write_EIO.
0661          * For current transaction, the bh is just among those error
0662          * bhs which previous transaction handle. We can't just clear
0663          * its BH_Write_EIO and reuse directly, since other bhs are
0664          * not written to disk yet and that will cause metadata
0665          * inconsistency. So we should set fs read-only to avoid
0666          * further damage.
0667          */
0668         if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) {
0669             unlock_buffer(bh);
0670             return ocfs2_error(osb->sb, "A previous attempt to "
0671                     "write this buffer head failed\n");
0672         }
0673         unlock_buffer(bh);
0674     }
0675 
0676     /* Set the current transaction information on the ci so
0677      * that the locking code knows whether it can drop it's locks
0678      * on this ci or not. We're protected from the commit
0679      * thread updating the current transaction id until
0680      * ocfs2_commit_trans() because ocfs2_start_trans() took
0681      * j_trans_barrier for us. */
0682     ocfs2_set_ci_lock_trans(osb->journal, ci);
0683 
0684     ocfs2_metadata_cache_io_lock(ci);
0685     switch (type) {
0686     case OCFS2_JOURNAL_ACCESS_CREATE:
0687     case OCFS2_JOURNAL_ACCESS_WRITE:
0688         status = jbd2_journal_get_write_access(handle, bh);
0689         break;
0690 
0691     case OCFS2_JOURNAL_ACCESS_UNDO:
0692         status = jbd2_journal_get_undo_access(handle, bh);
0693         break;
0694 
0695     default:
0696         status = -EINVAL;
0697         mlog(ML_ERROR, "Unknown access type!\n");
0698     }
0699     if (!status && ocfs2_meta_ecc(osb) && triggers)
0700         jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
0701     ocfs2_metadata_cache_io_unlock(ci);
0702 
0703     if (status < 0)
0704         mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
0705              status, type);
0706 
0707     return status;
0708 }
0709 
0710 int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
0711                 struct buffer_head *bh, int type)
0712 {
0713     return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
0714 }
0715 
0716 int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
0717                 struct buffer_head *bh, int type)
0718 {
0719     return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
0720 }
0721 
0722 int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
0723                 struct buffer_head *bh, int type)
0724 {
0725     return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
0726                       type);
0727 }
0728 
0729 int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
0730                 struct buffer_head *bh, int type)
0731 {
0732     return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
0733 }
0734 
0735 int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
0736                 struct buffer_head *bh, int type)
0737 {
0738     return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
0739 }
0740 
0741 int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
0742                 struct buffer_head *bh, int type)
0743 {
0744     return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
0745 }
0746 
0747 int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
0748                 struct buffer_head *bh, int type)
0749 {
0750     return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
0751 }
0752 
0753 int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
0754                 struct buffer_head *bh, int type)
0755 {
0756     return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
0757 }
0758 
0759 int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
0760                 struct buffer_head *bh, int type)
0761 {
0762     return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
0763 }
0764 
0765 int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
0766              struct buffer_head *bh, int type)
0767 {
0768     return __ocfs2_journal_access(handle, ci, bh, NULL, type);
0769 }
0770 
0771 void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
0772 {
0773     int status;
0774 
0775     trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
0776 
0777     status = jbd2_journal_dirty_metadata(handle, bh);
0778     if (status) {
0779         mlog_errno(status);
0780         if (!is_handle_aborted(handle)) {
0781             journal_t *journal = handle->h_transaction->t_journal;
0782             struct super_block *sb = bh->b_bdev->bd_super;
0783 
0784             mlog(ML_ERROR, "jbd2_journal_dirty_metadata failed. "
0785                     "Aborting transaction and journal.\n");
0786             handle->h_err = status;
0787             jbd2_journal_abort_handle(handle);
0788             jbd2_journal_abort(journal, status);
0789             ocfs2_abort(sb, "Journal already aborted.\n");
0790         }
0791     }
0792 }
0793 
0794 #define OCFS2_DEFAULT_COMMIT_INTERVAL   (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
0795 
0796 void ocfs2_set_journal_params(struct ocfs2_super *osb)
0797 {
0798     journal_t *journal = osb->journal->j_journal;
0799     unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
0800 
0801     if (osb->osb_commit_interval)
0802         commit_interval = osb->osb_commit_interval;
0803 
0804     write_lock(&journal->j_state_lock);
0805     journal->j_commit_interval = commit_interval;
0806     if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
0807         journal->j_flags |= JBD2_BARRIER;
0808     else
0809         journal->j_flags &= ~JBD2_BARRIER;
0810     write_unlock(&journal->j_state_lock);
0811 }
0812 
0813 /*
0814  * alloc & initialize skeleton for journal structure.
0815  * ocfs2_journal_init() will make fs have journal ability.
0816  */
0817 int ocfs2_journal_alloc(struct ocfs2_super *osb)
0818 {
0819     int status = 0;
0820     struct ocfs2_journal *journal;
0821 
0822     journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
0823     if (!journal) {
0824         mlog(ML_ERROR, "unable to alloc journal\n");
0825         status = -ENOMEM;
0826         goto bail;
0827     }
0828     osb->journal = journal;
0829     journal->j_osb = osb;
0830 
0831     atomic_set(&journal->j_num_trans, 0);
0832     init_rwsem(&journal->j_trans_barrier);
0833     init_waitqueue_head(&journal->j_checkpointed);
0834     spin_lock_init(&journal->j_lock);
0835     journal->j_trans_id = 1UL;
0836     INIT_LIST_HEAD(&journal->j_la_cleanups);
0837     INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
0838     journal->j_state = OCFS2_JOURNAL_FREE;
0839 
0840 bail:
0841     return status;
0842 }
0843 
0844 int ocfs2_journal_init(struct ocfs2_super *osb, int *dirty)
0845 {
0846     int status = -1;
0847     struct inode *inode = NULL; /* the journal inode */
0848     journal_t *j_journal = NULL;
0849     struct ocfs2_journal *journal = osb->journal;
0850     struct ocfs2_dinode *di = NULL;
0851     struct buffer_head *bh = NULL;
0852     int inode_lock = 0;
0853 
0854     BUG_ON(!journal);
0855     /* already have the inode for our journal */
0856     inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
0857                         osb->slot_num);
0858     if (inode == NULL) {
0859         status = -EACCES;
0860         mlog_errno(status);
0861         goto done;
0862     }
0863     if (is_bad_inode(inode)) {
0864         mlog(ML_ERROR, "access error (bad inode)\n");
0865         iput(inode);
0866         inode = NULL;
0867         status = -EACCES;
0868         goto done;
0869     }
0870 
0871     SET_INODE_JOURNAL(inode);
0872     OCFS2_I(inode)->ip_open_count++;
0873 
0874     /* Skip recovery waits here - journal inode metadata never
0875      * changes in a live cluster so it can be considered an
0876      * exception to the rule. */
0877     status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
0878     if (status < 0) {
0879         if (status != -ERESTARTSYS)
0880             mlog(ML_ERROR, "Could not get lock on journal!\n");
0881         goto done;
0882     }
0883 
0884     inode_lock = 1;
0885     di = (struct ocfs2_dinode *)bh->b_data;
0886 
0887     if (i_size_read(inode) <  OCFS2_MIN_JOURNAL_SIZE) {
0888         mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
0889              i_size_read(inode));
0890         status = -EINVAL;
0891         goto done;
0892     }
0893 
0894     trace_ocfs2_journal_init(i_size_read(inode),
0895                  (unsigned long long)inode->i_blocks,
0896                  OCFS2_I(inode)->ip_clusters);
0897 
0898     /* call the kernels journal init function now */
0899     j_journal = jbd2_journal_init_inode(inode);
0900     if (j_journal == NULL) {
0901         mlog(ML_ERROR, "Linux journal layer error\n");
0902         status = -EINVAL;
0903         goto done;
0904     }
0905 
0906     trace_ocfs2_journal_init_maxlen(j_journal->j_total_len);
0907 
0908     *dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
0909           OCFS2_JOURNAL_DIRTY_FL);
0910 
0911     journal->j_journal = j_journal;
0912     journal->j_journal->j_submit_inode_data_buffers =
0913         jbd2_journal_submit_inode_data_buffers;
0914     journal->j_journal->j_finish_inode_data_buffers =
0915         jbd2_journal_finish_inode_data_buffers;
0916     journal->j_inode = inode;
0917     journal->j_bh = bh;
0918 
0919     ocfs2_set_journal_params(osb);
0920 
0921     journal->j_state = OCFS2_JOURNAL_LOADED;
0922 
0923     status = 0;
0924 done:
0925     if (status < 0) {
0926         if (inode_lock)
0927             ocfs2_inode_unlock(inode, 1);
0928         brelse(bh);
0929         if (inode) {
0930             OCFS2_I(inode)->ip_open_count--;
0931             iput(inode);
0932         }
0933     }
0934 
0935     return status;
0936 }
0937 
0938 static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
0939 {
0940     le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
0941 }
0942 
0943 static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
0944 {
0945     return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
0946 }
0947 
0948 static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
0949                       int dirty, int replayed)
0950 {
0951     int status;
0952     unsigned int flags;
0953     struct ocfs2_journal *journal = osb->journal;
0954     struct buffer_head *bh = journal->j_bh;
0955     struct ocfs2_dinode *fe;
0956 
0957     fe = (struct ocfs2_dinode *)bh->b_data;
0958 
0959     /* The journal bh on the osb always comes from ocfs2_journal_init()
0960      * and was validated there inside ocfs2_inode_lock_full().  It's a
0961      * code bug if we mess it up. */
0962     BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
0963 
0964     flags = le32_to_cpu(fe->id1.journal1.ij_flags);
0965     if (dirty)
0966         flags |= OCFS2_JOURNAL_DIRTY_FL;
0967     else
0968         flags &= ~OCFS2_JOURNAL_DIRTY_FL;
0969     fe->id1.journal1.ij_flags = cpu_to_le32(flags);
0970 
0971     if (replayed)
0972         ocfs2_bump_recovery_generation(fe);
0973 
0974     ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
0975     status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
0976     if (status < 0)
0977         mlog_errno(status);
0978 
0979     return status;
0980 }
0981 
0982 /*
0983  * If the journal has been kmalloc'd it needs to be freed after this
0984  * call.
0985  */
0986 void ocfs2_journal_shutdown(struct ocfs2_super *osb)
0987 {
0988     struct ocfs2_journal *journal = NULL;
0989     int status = 0;
0990     struct inode *inode = NULL;
0991     int num_running_trans = 0;
0992 
0993     BUG_ON(!osb);
0994 
0995     journal = osb->journal;
0996     if (!journal)
0997         goto done;
0998 
0999     inode = journal->j_inode;
1000 
1001     if (journal->j_state != OCFS2_JOURNAL_LOADED)
1002         goto done;
1003 
1004     /* need to inc inode use count - jbd2_journal_destroy will iput. */
1005     if (!igrab(inode))
1006         BUG();
1007 
1008     num_running_trans = atomic_read(&(osb->journal->j_num_trans));
1009     trace_ocfs2_journal_shutdown(num_running_trans);
1010 
1011     /* Do a commit_cache here. It will flush our journal, *and*
1012      * release any locks that are still held.
1013      * set the SHUTDOWN flag and release the trans lock.
1014      * the commit thread will take the trans lock for us below. */
1015     journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
1016 
1017     /* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
1018      * drop the trans_lock (which we want to hold until we
1019      * completely destroy the journal. */
1020     if (osb->commit_task) {
1021         /* Wait for the commit thread */
1022         trace_ocfs2_journal_shutdown_wait(osb->commit_task);
1023         kthread_stop(osb->commit_task);
1024         osb->commit_task = NULL;
1025     }
1026 
1027     BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
1028 
1029     if (ocfs2_mount_local(osb)) {
1030         jbd2_journal_lock_updates(journal->j_journal);
1031         status = jbd2_journal_flush(journal->j_journal, 0);
1032         jbd2_journal_unlock_updates(journal->j_journal);
1033         if (status < 0)
1034             mlog_errno(status);
1035     }
1036 
1037     /* Shutdown the kernel journal system */
1038     if (!jbd2_journal_destroy(journal->j_journal) && !status) {
1039         /*
1040          * Do not toggle if flush was unsuccessful otherwise
1041          * will leave dirty metadata in a "clean" journal
1042          */
1043         status = ocfs2_journal_toggle_dirty(osb, 0, 0);
1044         if (status < 0)
1045             mlog_errno(status);
1046     }
1047     journal->j_journal = NULL;
1048 
1049     OCFS2_I(inode)->ip_open_count--;
1050 
1051     /* unlock our journal */
1052     ocfs2_inode_unlock(inode, 1);
1053 
1054     brelse(journal->j_bh);
1055     journal->j_bh = NULL;
1056 
1057     journal->j_state = OCFS2_JOURNAL_FREE;
1058 
1059 done:
1060     iput(inode);
1061     kfree(journal);
1062     osb->journal = NULL;
1063 }
1064 
1065 static void ocfs2_clear_journal_error(struct super_block *sb,
1066                       journal_t *journal,
1067                       int slot)
1068 {
1069     int olderr;
1070 
1071     olderr = jbd2_journal_errno(journal);
1072     if (olderr) {
1073         mlog(ML_ERROR, "File system error %d recorded in "
1074              "journal %u.\n", olderr, slot);
1075         mlog(ML_ERROR, "File system on device %s needs checking.\n",
1076              sb->s_id);
1077 
1078         jbd2_journal_ack_err(journal);
1079         jbd2_journal_clear_err(journal);
1080     }
1081 }
1082 
1083 int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
1084 {
1085     int status = 0;
1086     struct ocfs2_super *osb;
1087 
1088     BUG_ON(!journal);
1089 
1090     osb = journal->j_osb;
1091 
1092     status = jbd2_journal_load(journal->j_journal);
1093     if (status < 0) {
1094         mlog(ML_ERROR, "Failed to load journal!\n");
1095         goto done;
1096     }
1097 
1098     ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
1099 
1100     if (replayed) {
1101         jbd2_journal_lock_updates(journal->j_journal);
1102         status = jbd2_journal_flush(journal->j_journal, 0);
1103         jbd2_journal_unlock_updates(journal->j_journal);
1104         if (status < 0)
1105             mlog_errno(status);
1106     }
1107 
1108     status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
1109     if (status < 0) {
1110         mlog_errno(status);
1111         goto done;
1112     }
1113 
1114     /* Launch the commit thread */
1115     if (!local) {
1116         osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
1117                 "ocfs2cmt-%s", osb->uuid_str);
1118         if (IS_ERR(osb->commit_task)) {
1119             status = PTR_ERR(osb->commit_task);
1120             osb->commit_task = NULL;
1121             mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
1122                  "error=%d", status);
1123             goto done;
1124         }
1125     } else
1126         osb->commit_task = NULL;
1127 
1128 done:
1129     return status;
1130 }
1131 
1132 
1133 /* 'full' flag tells us whether we clear out all blocks or if we just
1134  * mark the journal clean */
1135 int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
1136 {
1137     int status;
1138 
1139     BUG_ON(!journal);
1140 
1141     status = jbd2_journal_wipe(journal->j_journal, full);
1142     if (status < 0) {
1143         mlog_errno(status);
1144         goto bail;
1145     }
1146 
1147     status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
1148     if (status < 0)
1149         mlog_errno(status);
1150 
1151 bail:
1152     return status;
1153 }
1154 
1155 static int ocfs2_recovery_completed(struct ocfs2_super *osb)
1156 {
1157     int empty;
1158     struct ocfs2_recovery_map *rm = osb->recovery_map;
1159 
1160     spin_lock(&osb->osb_lock);
1161     empty = (rm->rm_used == 0);
1162     spin_unlock(&osb->osb_lock);
1163 
1164     return empty;
1165 }
1166 
1167 void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
1168 {
1169     wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
1170 }
1171 
1172 /*
1173  * JBD Might read a cached version of another nodes journal file. We
1174  * don't want this as this file changes often and we get no
1175  * notification on those changes. The only way to be sure that we've
1176  * got the most up to date version of those blocks then is to force
1177  * read them off disk. Just searching through the buffer cache won't
1178  * work as there may be pages backing this file which are still marked
1179  * up to date. We know things can't change on this file underneath us
1180  * as we have the lock by now :)
1181  */
1182 static int ocfs2_force_read_journal(struct inode *inode)
1183 {
1184     int status = 0;
1185     int i;
1186     u64 v_blkno, p_blkno, p_blocks, num_blocks;
1187     struct buffer_head *bh = NULL;
1188     struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1189 
1190     num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
1191     v_blkno = 0;
1192     while (v_blkno < num_blocks) {
1193         status = ocfs2_extent_map_get_blocks(inode, v_blkno,
1194                              &p_blkno, &p_blocks, NULL);
1195         if (status < 0) {
1196             mlog_errno(status);
1197             goto bail;
1198         }
1199 
1200         for (i = 0; i < p_blocks; i++, p_blkno++) {
1201             bh = __find_get_block(osb->sb->s_bdev, p_blkno,
1202                     osb->sb->s_blocksize);
1203             /* block not cached. */
1204             if (!bh)
1205                 continue;
1206 
1207             brelse(bh);
1208             bh = NULL;
1209             /* We are reading journal data which should not
1210              * be put in the uptodate cache.
1211              */
1212             status = ocfs2_read_blocks_sync(osb, p_blkno, 1, &bh);
1213             if (status < 0) {
1214                 mlog_errno(status);
1215                 goto bail;
1216             }
1217 
1218             brelse(bh);
1219             bh = NULL;
1220         }
1221 
1222         v_blkno += p_blocks;
1223     }
1224 
1225 bail:
1226     return status;
1227 }
1228 
1229 struct ocfs2_la_recovery_item {
1230     struct list_head    lri_list;
1231     int         lri_slot;
1232     struct ocfs2_dinode *lri_la_dinode;
1233     struct ocfs2_dinode *lri_tl_dinode;
1234     struct ocfs2_quota_recovery *lri_qrec;
1235     enum ocfs2_orphan_reco_type  lri_orphan_reco_type;
1236 };
1237 
1238 /* Does the second half of the recovery process. By this point, the
1239  * node is marked clean and can actually be considered recovered,
1240  * hence it's no longer in the recovery map, but there's still some
1241  * cleanup we can do which shouldn't happen within the recovery thread
1242  * as locking in that context becomes very difficult if we are to take
1243  * recovering nodes into account.
1244  *
1245  * NOTE: This function can and will sleep on recovery of other nodes
1246  * during cluster locking, just like any other ocfs2 process.
1247  */
1248 void ocfs2_complete_recovery(struct work_struct *work)
1249 {
1250     int ret = 0;
1251     struct ocfs2_journal *journal =
1252         container_of(work, struct ocfs2_journal, j_recovery_work);
1253     struct ocfs2_super *osb = journal->j_osb;
1254     struct ocfs2_dinode *la_dinode, *tl_dinode;
1255     struct ocfs2_la_recovery_item *item, *n;
1256     struct ocfs2_quota_recovery *qrec;
1257     enum ocfs2_orphan_reco_type orphan_reco_type;
1258     LIST_HEAD(tmp_la_list);
1259 
1260     trace_ocfs2_complete_recovery(
1261         (unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
1262 
1263     spin_lock(&journal->j_lock);
1264     list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
1265     spin_unlock(&journal->j_lock);
1266 
1267     list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
1268         list_del_init(&item->lri_list);
1269 
1270         ocfs2_wait_on_quotas(osb);
1271 
1272         la_dinode = item->lri_la_dinode;
1273         tl_dinode = item->lri_tl_dinode;
1274         qrec = item->lri_qrec;
1275         orphan_reco_type = item->lri_orphan_reco_type;
1276 
1277         trace_ocfs2_complete_recovery_slot(item->lri_slot,
1278             la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
1279             tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
1280             qrec);
1281 
1282         if (la_dinode) {
1283             ret = ocfs2_complete_local_alloc_recovery(osb,
1284                                   la_dinode);
1285             if (ret < 0)
1286                 mlog_errno(ret);
1287 
1288             kfree(la_dinode);
1289         }
1290 
1291         if (tl_dinode) {
1292             ret = ocfs2_complete_truncate_log_recovery(osb,
1293                                    tl_dinode);
1294             if (ret < 0)
1295                 mlog_errno(ret);
1296 
1297             kfree(tl_dinode);
1298         }
1299 
1300         ret = ocfs2_recover_orphans(osb, item->lri_slot,
1301                 orphan_reco_type);
1302         if (ret < 0)
1303             mlog_errno(ret);
1304 
1305         if (qrec) {
1306             ret = ocfs2_finish_quota_recovery(osb, qrec,
1307                               item->lri_slot);
1308             if (ret < 0)
1309                 mlog_errno(ret);
1310             /* Recovery info is already freed now */
1311         }
1312 
1313         kfree(item);
1314     }
1315 
1316     trace_ocfs2_complete_recovery_end(ret);
1317 }
1318 
1319 /* NOTE: This function always eats your references to la_dinode and
1320  * tl_dinode, either manually on error, or by passing them to
1321  * ocfs2_complete_recovery */
1322 static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
1323                         int slot_num,
1324                         struct ocfs2_dinode *la_dinode,
1325                         struct ocfs2_dinode *tl_dinode,
1326                         struct ocfs2_quota_recovery *qrec,
1327                         enum ocfs2_orphan_reco_type orphan_reco_type)
1328 {
1329     struct ocfs2_la_recovery_item *item;
1330 
1331     item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
1332     if (!item) {
1333         /* Though we wish to avoid it, we are in fact safe in
1334          * skipping local alloc cleanup as fsck.ocfs2 is more
1335          * than capable of reclaiming unused space. */
1336         kfree(la_dinode);
1337         kfree(tl_dinode);
1338 
1339         if (qrec)
1340             ocfs2_free_quota_recovery(qrec);
1341 
1342         mlog_errno(-ENOMEM);
1343         return;
1344     }
1345 
1346     INIT_LIST_HEAD(&item->lri_list);
1347     item->lri_la_dinode = la_dinode;
1348     item->lri_slot = slot_num;
1349     item->lri_tl_dinode = tl_dinode;
1350     item->lri_qrec = qrec;
1351     item->lri_orphan_reco_type = orphan_reco_type;
1352 
1353     spin_lock(&journal->j_lock);
1354     list_add_tail(&item->lri_list, &journal->j_la_cleanups);
1355     queue_work(journal->j_osb->ocfs2_wq, &journal->j_recovery_work);
1356     spin_unlock(&journal->j_lock);
1357 }
1358 
1359 /* Called by the mount code to queue recovery the last part of
1360  * recovery for it's own and offline slot(s). */
1361 void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
1362 {
1363     struct ocfs2_journal *journal = osb->journal;
1364 
1365     if (ocfs2_is_hard_readonly(osb))
1366         return;
1367 
1368     /* No need to queue up our truncate_log as regular cleanup will catch
1369      * that */
1370     ocfs2_queue_recovery_completion(journal, osb->slot_num,
1371                     osb->local_alloc_copy, NULL, NULL,
1372                     ORPHAN_NEED_TRUNCATE);
1373     ocfs2_schedule_truncate_log_flush(osb, 0);
1374 
1375     osb->local_alloc_copy = NULL;
1376 
1377     /* queue to recover orphan slots for all offline slots */
1378     ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1379     ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1380     ocfs2_free_replay_slots(osb);
1381 }
1382 
1383 void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
1384 {
1385     if (osb->quota_rec) {
1386         ocfs2_queue_recovery_completion(osb->journal,
1387                         osb->slot_num,
1388                         NULL,
1389                         NULL,
1390                         osb->quota_rec,
1391                         ORPHAN_NEED_TRUNCATE);
1392         osb->quota_rec = NULL;
1393     }
1394 }
1395 
1396 static int __ocfs2_recovery_thread(void *arg)
1397 {
1398     int status, node_num, slot_num;
1399     struct ocfs2_super *osb = arg;
1400     struct ocfs2_recovery_map *rm = osb->recovery_map;
1401     int *rm_quota = NULL;
1402     int rm_quota_used = 0, i;
1403     struct ocfs2_quota_recovery *qrec;
1404 
1405     /* Whether the quota supported. */
1406     int quota_enabled = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1407             OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
1408         || OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb,
1409             OCFS2_FEATURE_RO_COMPAT_GRPQUOTA);
1410 
1411     status = ocfs2_wait_on_mount(osb);
1412     if (status < 0) {
1413         goto bail;
1414     }
1415 
1416     if (quota_enabled) {
1417         rm_quota = kcalloc(osb->max_slots, sizeof(int), GFP_NOFS);
1418         if (!rm_quota) {
1419             status = -ENOMEM;
1420             goto bail;
1421         }
1422     }
1423 restart:
1424     status = ocfs2_super_lock(osb, 1);
1425     if (status < 0) {
1426         mlog_errno(status);
1427         goto bail;
1428     }
1429 
1430     status = ocfs2_compute_replay_slots(osb);
1431     if (status < 0)
1432         mlog_errno(status);
1433 
1434     /* queue recovery for our own slot */
1435     ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
1436                     NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
1437 
1438     spin_lock(&osb->osb_lock);
1439     while (rm->rm_used) {
1440         /* It's always safe to remove entry zero, as we won't
1441          * clear it until ocfs2_recover_node() has succeeded. */
1442         node_num = rm->rm_entries[0];
1443         spin_unlock(&osb->osb_lock);
1444         slot_num = ocfs2_node_num_to_slot(osb, node_num);
1445         trace_ocfs2_recovery_thread_node(node_num, slot_num);
1446         if (slot_num == -ENOENT) {
1447             status = 0;
1448             goto skip_recovery;
1449         }
1450 
1451         /* It is a bit subtle with quota recovery. We cannot do it
1452          * immediately because we have to obtain cluster locks from
1453          * quota files and we also don't want to just skip it because
1454          * then quota usage would be out of sync until some node takes
1455          * the slot. So we remember which nodes need quota recovery
1456          * and when everything else is done, we recover quotas. */
1457         if (quota_enabled) {
1458             for (i = 0; i < rm_quota_used
1459                     && rm_quota[i] != slot_num; i++)
1460                 ;
1461 
1462             if (i == rm_quota_used)
1463                 rm_quota[rm_quota_used++] = slot_num;
1464         }
1465 
1466         status = ocfs2_recover_node(osb, node_num, slot_num);
1467 skip_recovery:
1468         if (!status) {
1469             ocfs2_recovery_map_clear(osb, node_num);
1470         } else {
1471             mlog(ML_ERROR,
1472                  "Error %d recovering node %d on device (%u,%u)!\n",
1473                  status, node_num,
1474                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1475             mlog(ML_ERROR, "Volume requires unmount.\n");
1476         }
1477 
1478         spin_lock(&osb->osb_lock);
1479     }
1480     spin_unlock(&osb->osb_lock);
1481     trace_ocfs2_recovery_thread_end(status);
1482 
1483     /* Refresh all journal recovery generations from disk */
1484     status = ocfs2_check_journals_nolocks(osb);
1485     status = (status == -EROFS) ? 0 : status;
1486     if (status < 0)
1487         mlog_errno(status);
1488 
1489     /* Now it is right time to recover quotas... We have to do this under
1490      * superblock lock so that no one can start using the slot (and crash)
1491      * before we recover it */
1492     if (quota_enabled) {
1493         for (i = 0; i < rm_quota_used; i++) {
1494             qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
1495             if (IS_ERR(qrec)) {
1496                 status = PTR_ERR(qrec);
1497                 mlog_errno(status);
1498                 continue;
1499             }
1500             ocfs2_queue_recovery_completion(osb->journal,
1501                     rm_quota[i],
1502                     NULL, NULL, qrec,
1503                     ORPHAN_NEED_TRUNCATE);
1504         }
1505     }
1506 
1507     ocfs2_super_unlock(osb, 1);
1508 
1509     /* queue recovery for offline slots */
1510     ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
1511 
1512 bail:
1513     mutex_lock(&osb->recovery_lock);
1514     if (!status && !ocfs2_recovery_completed(osb)) {
1515         mutex_unlock(&osb->recovery_lock);
1516         goto restart;
1517     }
1518 
1519     ocfs2_free_replay_slots(osb);
1520     osb->recovery_thread_task = NULL;
1521     mb(); /* sync with ocfs2_recovery_thread_running */
1522     wake_up(&osb->recovery_event);
1523 
1524     mutex_unlock(&osb->recovery_lock);
1525 
1526     if (quota_enabled)
1527         kfree(rm_quota);
1528 
1529     return status;
1530 }
1531 
1532 void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
1533 {
1534     mutex_lock(&osb->recovery_lock);
1535 
1536     trace_ocfs2_recovery_thread(node_num, osb->node_num,
1537         osb->disable_recovery, osb->recovery_thread_task,
1538         osb->disable_recovery ?
1539         -1 : ocfs2_recovery_map_set(osb, node_num));
1540 
1541     if (osb->disable_recovery)
1542         goto out;
1543 
1544     if (osb->recovery_thread_task)
1545         goto out;
1546 
1547     osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
1548             "ocfs2rec-%s", osb->uuid_str);
1549     if (IS_ERR(osb->recovery_thread_task)) {
1550         mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
1551         osb->recovery_thread_task = NULL;
1552     }
1553 
1554 out:
1555     mutex_unlock(&osb->recovery_lock);
1556     wake_up(&osb->recovery_event);
1557 }
1558 
1559 static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
1560                     int slot_num,
1561                     struct buffer_head **bh,
1562                     struct inode **ret_inode)
1563 {
1564     int status = -EACCES;
1565     struct inode *inode = NULL;
1566 
1567     BUG_ON(slot_num >= osb->max_slots);
1568 
1569     inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1570                         slot_num);
1571     if (!inode || is_bad_inode(inode)) {
1572         mlog_errno(status);
1573         goto bail;
1574     }
1575     SET_INODE_JOURNAL(inode);
1576 
1577     status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
1578     if (status < 0) {
1579         mlog_errno(status);
1580         goto bail;
1581     }
1582 
1583     status = 0;
1584 
1585 bail:
1586     if (inode) {
1587         if (status || !ret_inode)
1588             iput(inode);
1589         else
1590             *ret_inode = inode;
1591     }
1592     return status;
1593 }
1594 
1595 /* Does the actual journal replay and marks the journal inode as
1596  * clean. Will only replay if the journal inode is marked dirty. */
1597 static int ocfs2_replay_journal(struct ocfs2_super *osb,
1598                 int node_num,
1599                 int slot_num)
1600 {
1601     int status;
1602     int got_lock = 0;
1603     unsigned int flags;
1604     struct inode *inode = NULL;
1605     struct ocfs2_dinode *fe;
1606     journal_t *journal = NULL;
1607     struct buffer_head *bh = NULL;
1608     u32 slot_reco_gen;
1609 
1610     status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
1611     if (status) {
1612         mlog_errno(status);
1613         goto done;
1614     }
1615 
1616     fe = (struct ocfs2_dinode *)bh->b_data;
1617     slot_reco_gen = ocfs2_get_recovery_generation(fe);
1618     brelse(bh);
1619     bh = NULL;
1620 
1621     /*
1622      * As the fs recovery is asynchronous, there is a small chance that
1623      * another node mounted (and recovered) the slot before the recovery
1624      * thread could get the lock. To handle that, we dirty read the journal
1625      * inode for that slot to get the recovery generation. If it is
1626      * different than what we expected, the slot has been recovered.
1627      * If not, it needs recovery.
1628      */
1629     if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
1630         trace_ocfs2_replay_journal_recovered(slot_num,
1631              osb->slot_recovery_generations[slot_num], slot_reco_gen);
1632         osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1633         status = -EBUSY;
1634         goto done;
1635     }
1636 
1637     /* Continue with recovery as the journal has not yet been recovered */
1638 
1639     status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
1640     if (status < 0) {
1641         trace_ocfs2_replay_journal_lock_err(status);
1642         if (status != -ERESTARTSYS)
1643             mlog(ML_ERROR, "Could not lock journal!\n");
1644         goto done;
1645     }
1646     got_lock = 1;
1647 
1648     fe = (struct ocfs2_dinode *) bh->b_data;
1649 
1650     flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1651     slot_reco_gen = ocfs2_get_recovery_generation(fe);
1652 
1653     if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
1654         trace_ocfs2_replay_journal_skip(node_num);
1655         /* Refresh recovery generation for the slot */
1656         osb->slot_recovery_generations[slot_num] = slot_reco_gen;
1657         goto done;
1658     }
1659 
1660     /* we need to run complete recovery for offline orphan slots */
1661     ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
1662 
1663     printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
1664            "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1665            MINOR(osb->sb->s_dev));
1666 
1667     OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1668 
1669     status = ocfs2_force_read_journal(inode);
1670     if (status < 0) {
1671         mlog_errno(status);
1672         goto done;
1673     }
1674 
1675     journal = jbd2_journal_init_inode(inode);
1676     if (journal == NULL) {
1677         mlog(ML_ERROR, "Linux journal layer error\n");
1678         status = -EIO;
1679         goto done;
1680     }
1681 
1682     status = jbd2_journal_load(journal);
1683     if (status < 0) {
1684         mlog_errno(status);
1685         BUG_ON(!igrab(inode));
1686         jbd2_journal_destroy(journal);
1687         goto done;
1688     }
1689 
1690     ocfs2_clear_journal_error(osb->sb, journal, slot_num);
1691 
1692     /* wipe the journal */
1693     jbd2_journal_lock_updates(journal);
1694     status = jbd2_journal_flush(journal, 0);
1695     jbd2_journal_unlock_updates(journal);
1696     if (status < 0)
1697         mlog_errno(status);
1698 
1699     /* This will mark the node clean */
1700     flags = le32_to_cpu(fe->id1.journal1.ij_flags);
1701     flags &= ~OCFS2_JOURNAL_DIRTY_FL;
1702     fe->id1.journal1.ij_flags = cpu_to_le32(flags);
1703 
1704     /* Increment recovery generation to indicate successful recovery */
1705     ocfs2_bump_recovery_generation(fe);
1706     osb->slot_recovery_generations[slot_num] =
1707                     ocfs2_get_recovery_generation(fe);
1708 
1709     ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
1710     status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
1711     if (status < 0)
1712         mlog_errno(status);
1713 
1714     BUG_ON(!igrab(inode));
1715 
1716     jbd2_journal_destroy(journal);
1717 
1718     printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
1719            "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
1720            MINOR(osb->sb->s_dev));
1721 done:
1722     /* drop the lock on this nodes journal */
1723     if (got_lock)
1724         ocfs2_inode_unlock(inode, 1);
1725 
1726     iput(inode);
1727     brelse(bh);
1728 
1729     return status;
1730 }
1731 
1732 /*
1733  * Do the most important parts of node recovery:
1734  *  - Replay it's journal
1735  *  - Stamp a clean local allocator file
1736  *  - Stamp a clean truncate log
1737  *  - Mark the node clean
1738  *
1739  * If this function completes without error, a node in OCFS2 can be
1740  * said to have been safely recovered. As a result, failure during the
1741  * second part of a nodes recovery process (local alloc recovery) is
1742  * far less concerning.
1743  */
1744 static int ocfs2_recover_node(struct ocfs2_super *osb,
1745                   int node_num, int slot_num)
1746 {
1747     int status = 0;
1748     struct ocfs2_dinode *la_copy = NULL;
1749     struct ocfs2_dinode *tl_copy = NULL;
1750 
1751     trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
1752 
1753     /* Should not ever be called to recover ourselves -- in that
1754      * case we should've called ocfs2_journal_load instead. */
1755     BUG_ON(osb->node_num == node_num);
1756 
1757     status = ocfs2_replay_journal(osb, node_num, slot_num);
1758     if (status < 0) {
1759         if (status == -EBUSY) {
1760             trace_ocfs2_recover_node_skip(slot_num, node_num);
1761             status = 0;
1762             goto done;
1763         }
1764         mlog_errno(status);
1765         goto done;
1766     }
1767 
1768     /* Stamp a clean local alloc file AFTER recovering the journal... */
1769     status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
1770     if (status < 0) {
1771         mlog_errno(status);
1772         goto done;
1773     }
1774 
1775     /* An error from begin_truncate_log_recovery is not
1776      * serious enough to warrant halting the rest of
1777      * recovery. */
1778     status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
1779     if (status < 0)
1780         mlog_errno(status);
1781 
1782     /* Likewise, this would be a strange but ultimately not so
1783      * harmful place to get an error... */
1784     status = ocfs2_clear_slot(osb, slot_num);
1785     if (status < 0)
1786         mlog_errno(status);
1787 
1788     /* This will kfree the memory pointed to by la_copy and tl_copy */
1789     ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
1790                     tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
1791 
1792     status = 0;
1793 done:
1794 
1795     return status;
1796 }
1797 
1798 /* Test node liveness by trylocking his journal. If we get the lock,
1799  * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
1800  * still alive (we couldn't get the lock) and < 0 on error. */
1801 static int ocfs2_trylock_journal(struct ocfs2_super *osb,
1802                  int slot_num)
1803 {
1804     int status, flags;
1805     struct inode *inode = NULL;
1806 
1807     inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
1808                         slot_num);
1809     if (inode == NULL) {
1810         mlog(ML_ERROR, "access error\n");
1811         status = -EACCES;
1812         goto bail;
1813     }
1814     if (is_bad_inode(inode)) {
1815         mlog(ML_ERROR, "access error (bad inode)\n");
1816         iput(inode);
1817         inode = NULL;
1818         status = -EACCES;
1819         goto bail;
1820     }
1821     SET_INODE_JOURNAL(inode);
1822 
1823     flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
1824     status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
1825     if (status < 0) {
1826         if (status != -EAGAIN)
1827             mlog_errno(status);
1828         goto bail;
1829     }
1830 
1831     ocfs2_inode_unlock(inode, 1);
1832 bail:
1833     iput(inode);
1834 
1835     return status;
1836 }
1837 
1838 /* Call this underneath ocfs2_super_lock. It also assumes that the
1839  * slot info struct has been updated from disk. */
1840 int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
1841 {
1842     unsigned int node_num;
1843     int status, i;
1844     u32 gen;
1845     struct buffer_head *bh = NULL;
1846     struct ocfs2_dinode *di;
1847 
1848     /* This is called with the super block cluster lock, so we
1849      * know that the slot map can't change underneath us. */
1850 
1851     for (i = 0; i < osb->max_slots; i++) {
1852         /* Read journal inode to get the recovery generation */
1853         status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
1854         if (status) {
1855             mlog_errno(status);
1856             goto bail;
1857         }
1858         di = (struct ocfs2_dinode *)bh->b_data;
1859         gen = ocfs2_get_recovery_generation(di);
1860         brelse(bh);
1861         bh = NULL;
1862 
1863         spin_lock(&osb->osb_lock);
1864         osb->slot_recovery_generations[i] = gen;
1865 
1866         trace_ocfs2_mark_dead_nodes(i,
1867                         osb->slot_recovery_generations[i]);
1868 
1869         if (i == osb->slot_num) {
1870             spin_unlock(&osb->osb_lock);
1871             continue;
1872         }
1873 
1874         status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
1875         if (status == -ENOENT) {
1876             spin_unlock(&osb->osb_lock);
1877             continue;
1878         }
1879 
1880         if (__ocfs2_recovery_map_test(osb, node_num)) {
1881             spin_unlock(&osb->osb_lock);
1882             continue;
1883         }
1884         spin_unlock(&osb->osb_lock);
1885 
1886         /* Ok, we have a slot occupied by another node which
1887          * is not in the recovery map. We trylock his journal
1888          * file here to test if he's alive. */
1889         status = ocfs2_trylock_journal(osb, i);
1890         if (!status) {
1891             /* Since we're called from mount, we know that
1892              * the recovery thread can't race us on
1893              * setting / checking the recovery bits. */
1894             ocfs2_recovery_thread(osb, node_num);
1895         } else if ((status < 0) && (status != -EAGAIN)) {
1896             mlog_errno(status);
1897             goto bail;
1898         }
1899     }
1900 
1901     status = 0;
1902 bail:
1903     return status;
1904 }
1905 
1906 /*
1907  * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
1908  * randomness to the timeout to minimize multple nodes firing the timer at the
1909  * same time.
1910  */
1911 static inline unsigned long ocfs2_orphan_scan_timeout(void)
1912 {
1913     unsigned long time;
1914 
1915     get_random_bytes(&time, sizeof(time));
1916     time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
1917     return msecs_to_jiffies(time);
1918 }
1919 
1920 /*
1921  * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
1922  * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
1923  * is done to catch any orphans that are left over in orphan directories.
1924  *
1925  * It scans all slots, even ones that are in use. It does so to handle the
1926  * case described below:
1927  *
1928  *   Node 1 has an inode it was using. The dentry went away due to memory
1929  *   pressure.  Node 1 closes the inode, but it's on the free list. The node
1930  *   has the open lock.
1931  *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
1932  *   but node 1 has no dentry and doesn't get the message. It trylocks the
1933  *   open lock, sees that another node has a PR, and does nothing.
1934  *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
1935  *   open lock, sees the PR still, and does nothing.
1936  *   Basically, we have to trigger an orphan iput on node 1. The only way
1937  *   for this to happen is if node 1 runs node 2's orphan dir.
1938  *
1939  * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
1940  * seconds.  It gets an EX lock on os_lockres and checks sequence number
1941  * stored in LVB. If the sequence number has changed, it means some other
1942  * node has done the scan.  This node skips the scan and tracks the
1943  * sequence number.  If the sequence number didn't change, it means a scan
1944  * hasn't happened.  The node queues a scan and increments the
1945  * sequence number in the LVB.
1946  */
1947 static void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
1948 {
1949     struct ocfs2_orphan_scan *os;
1950     int status, i;
1951     u32 seqno = 0;
1952 
1953     os = &osb->osb_orphan_scan;
1954 
1955     if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1956         goto out;
1957 
1958     trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
1959                         atomic_read(&os->os_state));
1960 
1961     status = ocfs2_orphan_scan_lock(osb, &seqno);
1962     if (status < 0) {
1963         if (status != -EAGAIN)
1964             mlog_errno(status);
1965         goto out;
1966     }
1967 
1968     /* Do no queue the tasks if the volume is being umounted */
1969     if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
1970         goto unlock;
1971 
1972     if (os->os_seqno != seqno) {
1973         os->os_seqno = seqno;
1974         goto unlock;
1975     }
1976 
1977     for (i = 0; i < osb->max_slots; i++)
1978         ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
1979                         NULL, ORPHAN_NO_NEED_TRUNCATE);
1980     /*
1981      * We queued a recovery on orphan slots, increment the sequence
1982      * number and update LVB so other node will skip the scan for a while
1983      */
1984     seqno++;
1985     os->os_count++;
1986     os->os_scantime = ktime_get_seconds();
1987 unlock:
1988     ocfs2_orphan_scan_unlock(osb, seqno);
1989 out:
1990     trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
1991                       atomic_read(&os->os_state));
1992     return;
1993 }
1994 
1995 /* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
1996 static void ocfs2_orphan_scan_work(struct work_struct *work)
1997 {
1998     struct ocfs2_orphan_scan *os;
1999     struct ocfs2_super *osb;
2000 
2001     os = container_of(work, struct ocfs2_orphan_scan,
2002               os_orphan_scan_work.work);
2003     osb = os->os_osb;
2004 
2005     mutex_lock(&os->os_lock);
2006     ocfs2_queue_orphan_scan(osb);
2007     if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
2008         queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2009                       ocfs2_orphan_scan_timeout());
2010     mutex_unlock(&os->os_lock);
2011 }
2012 
2013 void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
2014 {
2015     struct ocfs2_orphan_scan *os;
2016 
2017     os = &osb->osb_orphan_scan;
2018     if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
2019         atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2020         mutex_lock(&os->os_lock);
2021         cancel_delayed_work(&os->os_orphan_scan_work);
2022         mutex_unlock(&os->os_lock);
2023     }
2024 }
2025 
2026 void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
2027 {
2028     struct ocfs2_orphan_scan *os;
2029 
2030     os = &osb->osb_orphan_scan;
2031     os->os_osb = osb;
2032     os->os_count = 0;
2033     os->os_seqno = 0;
2034     mutex_init(&os->os_lock);
2035     INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
2036 }
2037 
2038 void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
2039 {
2040     struct ocfs2_orphan_scan *os;
2041 
2042     os = &osb->osb_orphan_scan;
2043     os->os_scantime = ktime_get_seconds();
2044     if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
2045         atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
2046     else {
2047         atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
2048         queue_delayed_work(osb->ocfs2_wq, &os->os_orphan_scan_work,
2049                    ocfs2_orphan_scan_timeout());
2050     }
2051 }
2052 
2053 struct ocfs2_orphan_filldir_priv {
2054     struct dir_context  ctx;
2055     struct inode        *head;
2056     struct ocfs2_super  *osb;
2057     enum ocfs2_orphan_reco_type orphan_reco_type;
2058 };
2059 
2060 static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
2061                 int name_len, loff_t pos, u64 ino,
2062                 unsigned type)
2063 {
2064     struct ocfs2_orphan_filldir_priv *p =
2065         container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
2066     struct inode *iter;
2067 
2068     if (name_len == 1 && !strncmp(".", name, 1))
2069         return 0;
2070     if (name_len == 2 && !strncmp("..", name, 2))
2071         return 0;
2072 
2073     /* do not include dio entry in case of orphan scan */
2074     if ((p->orphan_reco_type == ORPHAN_NO_NEED_TRUNCATE) &&
2075             (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2076             OCFS2_DIO_ORPHAN_PREFIX_LEN)))
2077         return 0;
2078 
2079     /* Skip bad inodes so that recovery can continue */
2080     iter = ocfs2_iget(p->osb, ino,
2081               OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
2082     if (IS_ERR(iter))
2083         return 0;
2084 
2085     if (!strncmp(name, OCFS2_DIO_ORPHAN_PREFIX,
2086             OCFS2_DIO_ORPHAN_PREFIX_LEN))
2087         OCFS2_I(iter)->ip_flags |= OCFS2_INODE_DIO_ORPHAN_ENTRY;
2088 
2089     /* Skip inodes which are already added to recover list, since dio may
2090      * happen concurrently with unlink/rename */
2091     if (OCFS2_I(iter)->ip_next_orphan) {
2092         iput(iter);
2093         return 0;
2094     }
2095 
2096     trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
2097     /* No locking is required for the next_orphan queue as there
2098      * is only ever a single process doing orphan recovery. */
2099     OCFS2_I(iter)->ip_next_orphan = p->head;
2100     p->head = iter;
2101 
2102     return 0;
2103 }
2104 
2105 static int ocfs2_queue_orphans(struct ocfs2_super *osb,
2106                    int slot,
2107                    struct inode **head,
2108                    enum ocfs2_orphan_reco_type orphan_reco_type)
2109 {
2110     int status;
2111     struct inode *orphan_dir_inode = NULL;
2112     struct ocfs2_orphan_filldir_priv priv = {
2113         .ctx.actor = ocfs2_orphan_filldir,
2114         .osb = osb,
2115         .head = *head,
2116         .orphan_reco_type = orphan_reco_type
2117     };
2118 
2119     orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2120                                ORPHAN_DIR_SYSTEM_INODE,
2121                                slot);
2122     if  (!orphan_dir_inode) {
2123         status = -ENOENT;
2124         mlog_errno(status);
2125         return status;
2126     }
2127 
2128     inode_lock(orphan_dir_inode);
2129     status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
2130     if (status < 0) {
2131         mlog_errno(status);
2132         goto out;
2133     }
2134 
2135     status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
2136     if (status) {
2137         mlog_errno(status);
2138         goto out_cluster;
2139     }
2140 
2141     *head = priv.head;
2142 
2143 out_cluster:
2144     ocfs2_inode_unlock(orphan_dir_inode, 0);
2145 out:
2146     inode_unlock(orphan_dir_inode);
2147     iput(orphan_dir_inode);
2148     return status;
2149 }
2150 
2151 static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
2152                           int slot)
2153 {
2154     int ret;
2155 
2156     spin_lock(&osb->osb_lock);
2157     ret = !osb->osb_orphan_wipes[slot];
2158     spin_unlock(&osb->osb_lock);
2159     return ret;
2160 }
2161 
2162 static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
2163                          int slot)
2164 {
2165     spin_lock(&osb->osb_lock);
2166     /* Mark ourselves such that new processes in delete_inode()
2167      * know to quit early. */
2168     ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2169     while (osb->osb_orphan_wipes[slot]) {
2170         /* If any processes are already in the middle of an
2171          * orphan wipe on this dir, then we need to wait for
2172          * them. */
2173         spin_unlock(&osb->osb_lock);
2174         wait_event_interruptible(osb->osb_wipe_event,
2175                      ocfs2_orphan_recovery_can_continue(osb, slot));
2176         spin_lock(&osb->osb_lock);
2177     }
2178     spin_unlock(&osb->osb_lock);
2179 }
2180 
2181 static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
2182                           int slot)
2183 {
2184     ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
2185 }
2186 
2187 /*
2188  * Orphan recovery. Each mounted node has it's own orphan dir which we
2189  * must run during recovery. Our strategy here is to build a list of
2190  * the inodes in the orphan dir and iget/iput them. The VFS does
2191  * (most) of the rest of the work.
2192  *
2193  * Orphan recovery can happen at any time, not just mount so we have a
2194  * couple of extra considerations.
2195  *
2196  * - We grab as many inodes as we can under the orphan dir lock -
2197  *   doing iget() outside the orphan dir risks getting a reference on
2198  *   an invalid inode.
2199  * - We must be sure not to deadlock with other processes on the
2200  *   system wanting to run delete_inode(). This can happen when they go
2201  *   to lock the orphan dir and the orphan recovery process attempts to
2202  *   iget() inside the orphan dir lock. This can be avoided by
2203  *   advertising our state to ocfs2_delete_inode().
2204  */
2205 static int ocfs2_recover_orphans(struct ocfs2_super *osb,
2206                  int slot,
2207                  enum ocfs2_orphan_reco_type orphan_reco_type)
2208 {
2209     int ret = 0;
2210     struct inode *inode = NULL;
2211     struct inode *iter;
2212     struct ocfs2_inode_info *oi;
2213     struct buffer_head *di_bh = NULL;
2214     struct ocfs2_dinode *di = NULL;
2215 
2216     trace_ocfs2_recover_orphans(slot);
2217 
2218     ocfs2_mark_recovering_orphan_dir(osb, slot);
2219     ret = ocfs2_queue_orphans(osb, slot, &inode, orphan_reco_type);
2220     ocfs2_clear_recovering_orphan_dir(osb, slot);
2221 
2222     /* Error here should be noted, but we want to continue with as
2223      * many queued inodes as we've got. */
2224     if (ret)
2225         mlog_errno(ret);
2226 
2227     while (inode) {
2228         oi = OCFS2_I(inode);
2229         trace_ocfs2_recover_orphans_iput(
2230                     (unsigned long long)oi->ip_blkno);
2231 
2232         iter = oi->ip_next_orphan;
2233         oi->ip_next_orphan = NULL;
2234 
2235         if (oi->ip_flags & OCFS2_INODE_DIO_ORPHAN_ENTRY) {
2236             inode_lock(inode);
2237             ret = ocfs2_rw_lock(inode, 1);
2238             if (ret < 0) {
2239                 mlog_errno(ret);
2240                 goto unlock_mutex;
2241             }
2242             /*
2243              * We need to take and drop the inode lock to
2244              * force read inode from disk.
2245              */
2246             ret = ocfs2_inode_lock(inode, &di_bh, 1);
2247             if (ret) {
2248                 mlog_errno(ret);
2249                 goto unlock_rw;
2250             }
2251 
2252             di = (struct ocfs2_dinode *)di_bh->b_data;
2253 
2254             if (di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)) {
2255                 ret = ocfs2_truncate_file(inode, di_bh,
2256                         i_size_read(inode));
2257                 if (ret < 0) {
2258                     if (ret != -ENOSPC)
2259                         mlog_errno(ret);
2260                     goto unlock_inode;
2261                 }
2262 
2263                 ret = ocfs2_del_inode_from_orphan(osb, inode,
2264                         di_bh, 0, 0);
2265                 if (ret)
2266                     mlog_errno(ret);
2267             }
2268 unlock_inode:
2269             ocfs2_inode_unlock(inode, 1);
2270             brelse(di_bh);
2271             di_bh = NULL;
2272 unlock_rw:
2273             ocfs2_rw_unlock(inode, 1);
2274 unlock_mutex:
2275             inode_unlock(inode);
2276 
2277             /* clear dio flag in ocfs2_inode_info */
2278             oi->ip_flags &= ~OCFS2_INODE_DIO_ORPHAN_ENTRY;
2279         } else {
2280             spin_lock(&oi->ip_lock);
2281             /* Set the proper information to get us going into
2282              * ocfs2_delete_inode. */
2283             oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
2284             spin_unlock(&oi->ip_lock);
2285         }
2286 
2287         iput(inode);
2288         inode = iter;
2289     }
2290 
2291     return ret;
2292 }
2293 
2294 static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
2295 {
2296     /* This check is good because ocfs2 will wait on our recovery
2297      * thread before changing it to something other than MOUNTED
2298      * or DISABLED. */
2299     wait_event(osb->osb_mount_event,
2300           (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
2301            atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
2302            atomic_read(&osb->vol_state) == VOLUME_DISABLED);
2303 
2304     /* If there's an error on mount, then we may never get to the
2305      * MOUNTED flag, but this is set right before
2306      * dismount_volume() so we can trust it. */
2307     if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
2308         trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
2309         mlog(0, "mount error, exiting!\n");
2310         return -EBUSY;
2311     }
2312 
2313     return 0;
2314 }
2315 
2316 static int ocfs2_commit_thread(void *arg)
2317 {
2318     int status;
2319     struct ocfs2_super *osb = arg;
2320     struct ocfs2_journal *journal = osb->journal;
2321 
2322     /* we can trust j_num_trans here because _should_stop() is only set in
2323      * shutdown and nobody other than ourselves should be able to start
2324      * transactions.  committing on shutdown might take a few iterations
2325      * as final transactions put deleted inodes on the list */
2326     while (!(kthread_should_stop() &&
2327          atomic_read(&journal->j_num_trans) == 0)) {
2328 
2329         wait_event_interruptible(osb->checkpoint_event,
2330                      atomic_read(&journal->j_num_trans)
2331                      || kthread_should_stop());
2332 
2333         status = ocfs2_commit_cache(osb);
2334         if (status < 0) {
2335             static unsigned long abort_warn_time;
2336 
2337             /* Warn about this once per minute */
2338             if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
2339                 mlog(ML_ERROR, "status = %d, journal is "
2340                         "already aborted.\n", status);
2341             /*
2342              * After ocfs2_commit_cache() fails, j_num_trans has a
2343              * non-zero value.  Sleep here to avoid a busy-wait
2344              * loop.
2345              */
2346             msleep_interruptible(1000);
2347         }
2348 
2349         if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
2350             mlog(ML_KTHREAD,
2351                  "commit_thread: %u transactions pending on "
2352                  "shutdown\n",
2353                  atomic_read(&journal->j_num_trans));
2354         }
2355     }
2356 
2357     return 0;
2358 }
2359 
2360 /* Reads all the journal inodes without taking any cluster locks. Used
2361  * for hard readonly access to determine whether any journal requires
2362  * recovery. Also used to refresh the recovery generation numbers after
2363  * a journal has been recovered by another node.
2364  */
2365 int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
2366 {
2367     int ret = 0;
2368     unsigned int slot;
2369     struct buffer_head *di_bh = NULL;
2370     struct ocfs2_dinode *di;
2371     int journal_dirty = 0;
2372 
2373     for(slot = 0; slot < osb->max_slots; slot++) {
2374         ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
2375         if (ret) {
2376             mlog_errno(ret);
2377             goto out;
2378         }
2379 
2380         di = (struct ocfs2_dinode *) di_bh->b_data;
2381 
2382         osb->slot_recovery_generations[slot] =
2383                     ocfs2_get_recovery_generation(di);
2384 
2385         if (le32_to_cpu(di->id1.journal1.ij_flags) &
2386             OCFS2_JOURNAL_DIRTY_FL)
2387             journal_dirty = 1;
2388 
2389         brelse(di_bh);
2390         di_bh = NULL;
2391     }
2392 
2393 out:
2394     if (journal_dirty)
2395         ret = -EROFS;
2396     return ret;
2397 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!