Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * dlmthread.c
0004  *
0005  * standalone DLM module
0006  *
0007  * Copyright (C) 2004 Oracle.  All rights reserved.
0008  */
0009 
0010 
0011 #include <linux/module.h>
0012 #include <linux/fs.h>
0013 #include <linux/types.h>
0014 #include <linux/highmem.h>
0015 #include <linux/init.h>
0016 #include <linux/sysctl.h>
0017 #include <linux/random.h>
0018 #include <linux/blkdev.h>
0019 #include <linux/socket.h>
0020 #include <linux/inet.h>
0021 #include <linux/timer.h>
0022 #include <linux/kthread.h>
0023 #include <linux/delay.h>
0024 
0025 
0026 #include "../cluster/heartbeat.h"
0027 #include "../cluster/nodemanager.h"
0028 #include "../cluster/tcp.h"
0029 
0030 #include "dlmapi.h"
0031 #include "dlmcommon.h"
0032 #include "dlmdomain.h"
0033 
0034 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
0035 #include "../cluster/masklog.h"
0036 
0037 static int dlm_thread(void *data);
0038 static void dlm_flush_asts(struct dlm_ctxt *dlm);
0039 
0040 /* will exit holding res->spinlock, but may drop in function */
0041 /* waits until flags are cleared on res->state */
0042 void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
0043 {
0044     DECLARE_WAITQUEUE(wait, current);
0045 
0046     assert_spin_locked(&res->spinlock);
0047 
0048     add_wait_queue(&res->wq, &wait);
0049 repeat:
0050     set_current_state(TASK_UNINTERRUPTIBLE);
0051     if (res->state & flags) {
0052         spin_unlock(&res->spinlock);
0053         schedule();
0054         spin_lock(&res->spinlock);
0055         goto repeat;
0056     }
0057     remove_wait_queue(&res->wq, &wait);
0058     __set_current_state(TASK_RUNNING);
0059 }
0060 
0061 int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
0062 {
0063     if (list_empty(&res->granted) &&
0064         list_empty(&res->converting) &&
0065         list_empty(&res->blocked))
0066         return 0;
0067     return 1;
0068 }
0069 
0070 /* "unused": the lockres has no locks, is not on the dirty list,
0071  * has no inflight locks (in the gap between mastery and acquiring
0072  * the first lock), and has no bits in its refmap.
0073  * truly ready to be freed. */
0074 int __dlm_lockres_unused(struct dlm_lock_resource *res)
0075 {
0076     int bit;
0077 
0078     assert_spin_locked(&res->spinlock);
0079 
0080     if (__dlm_lockres_has_locks(res))
0081         return 0;
0082 
0083     /* Locks are in the process of being created */
0084     if (res->inflight_locks)
0085         return 0;
0086 
0087     if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY)
0088         return 0;
0089 
0090     if (res->state & (DLM_LOCK_RES_RECOVERING|
0091             DLM_LOCK_RES_RECOVERY_WAITING))
0092         return 0;
0093 
0094     /* Another node has this resource with this node as the master */
0095     bit = find_first_bit(res->refmap, O2NM_MAX_NODES);
0096     if (bit < O2NM_MAX_NODES)
0097         return 0;
0098 
0099     return 1;
0100 }
0101 
0102 
0103 /* Call whenever you may have added or deleted something from one of
0104  * the lockres queue's. This will figure out whether it belongs on the
0105  * unused list or not and does the appropriate thing. */
0106 void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
0107                   struct dlm_lock_resource *res)
0108 {
0109     assert_spin_locked(&dlm->spinlock);
0110     assert_spin_locked(&res->spinlock);
0111 
0112     if (__dlm_lockres_unused(res)){
0113         if (list_empty(&res->purge)) {
0114             mlog(0, "%s: Adding res %.*s to purge list\n",
0115                  dlm->name, res->lockname.len, res->lockname.name);
0116 
0117             res->last_used = jiffies;
0118             dlm_lockres_get(res);
0119             list_add_tail(&res->purge, &dlm->purge_list);
0120             dlm->purge_count++;
0121         }
0122     } else if (!list_empty(&res->purge)) {
0123         mlog(0, "%s: Removing res %.*s from purge list\n",
0124              dlm->name, res->lockname.len, res->lockname.name);
0125 
0126         list_del_init(&res->purge);
0127         dlm_lockres_put(res);
0128         dlm->purge_count--;
0129     }
0130 }
0131 
0132 void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
0133                 struct dlm_lock_resource *res)
0134 {
0135     spin_lock(&dlm->spinlock);
0136     spin_lock(&res->spinlock);
0137 
0138     __dlm_lockres_calc_usage(dlm, res);
0139 
0140     spin_unlock(&res->spinlock);
0141     spin_unlock(&dlm->spinlock);
0142 }
0143 
0144 /*
0145  * Do the real purge work:
0146  *     unhash the lockres, and
0147  *     clear flag DLM_LOCK_RES_DROPPING_REF.
0148  * It requires dlm and lockres spinlock to be taken.
0149  */
0150 void __dlm_do_purge_lockres(struct dlm_ctxt *dlm,
0151         struct dlm_lock_resource *res)
0152 {
0153     assert_spin_locked(&dlm->spinlock);
0154     assert_spin_locked(&res->spinlock);
0155 
0156     if (!list_empty(&res->purge)) {
0157         mlog(0, "%s: Removing res %.*s from purgelist\n",
0158              dlm->name, res->lockname.len, res->lockname.name);
0159         list_del_init(&res->purge);
0160         dlm_lockres_put(res);
0161         dlm->purge_count--;
0162     }
0163 
0164     if (!__dlm_lockres_unused(res)) {
0165         mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
0166              dlm->name, res->lockname.len, res->lockname.name);
0167         __dlm_print_one_lock_resource(res);
0168         BUG();
0169     }
0170 
0171     __dlm_unhash_lockres(dlm, res);
0172 
0173     spin_lock(&dlm->track_lock);
0174     if (!list_empty(&res->tracking))
0175         list_del_init(&res->tracking);
0176     else {
0177         mlog(ML_ERROR, "%s: Resource %.*s not on the Tracking list\n",
0178              dlm->name, res->lockname.len, res->lockname.name);
0179         __dlm_print_one_lock_resource(res);
0180     }
0181     spin_unlock(&dlm->track_lock);
0182 
0183     /*
0184      * lockres is not in the hash now. drop the flag and wake up
0185      * any processes waiting in dlm_get_lock_resource.
0186      */
0187     res->state &= ~DLM_LOCK_RES_DROPPING_REF;
0188 }
0189 
0190 static void dlm_purge_lockres(struct dlm_ctxt *dlm,
0191                  struct dlm_lock_resource *res)
0192 {
0193     int master;
0194     int ret = 0;
0195 
0196     assert_spin_locked(&dlm->spinlock);
0197     assert_spin_locked(&res->spinlock);
0198 
0199     master = (res->owner == dlm->node_num);
0200 
0201     mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name,
0202          res->lockname.len, res->lockname.name, master);
0203 
0204     if (!master) {
0205         if (res->state & DLM_LOCK_RES_DROPPING_REF) {
0206             mlog(ML_NOTICE, "%s: res %.*s already in DLM_LOCK_RES_DROPPING_REF state\n",
0207                 dlm->name, res->lockname.len, res->lockname.name);
0208             spin_unlock(&res->spinlock);
0209             return;
0210         }
0211 
0212         res->state |= DLM_LOCK_RES_DROPPING_REF;
0213         /* drop spinlock...  retake below */
0214         spin_unlock(&res->spinlock);
0215         spin_unlock(&dlm->spinlock);
0216 
0217         spin_lock(&res->spinlock);
0218         /* This ensures that clear refmap is sent after the set */
0219         __dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
0220         spin_unlock(&res->spinlock);
0221 
0222         /* clear our bit from the master's refmap, ignore errors */
0223         ret = dlm_drop_lockres_ref(dlm, res);
0224         if (ret < 0) {
0225             if (!dlm_is_host_down(ret))
0226                 BUG();
0227         }
0228         spin_lock(&dlm->spinlock);
0229         spin_lock(&res->spinlock);
0230     }
0231 
0232     if (!list_empty(&res->purge)) {
0233         mlog(0, "%s: Removing res %.*s from purgelist, master %d\n",
0234              dlm->name, res->lockname.len, res->lockname.name, master);
0235         list_del_init(&res->purge);
0236         dlm_lockres_put(res);
0237         dlm->purge_count--;
0238     }
0239 
0240     if (!master && ret == DLM_DEREF_RESPONSE_INPROG) {
0241         mlog(0, "%s: deref %.*s in progress\n",
0242             dlm->name, res->lockname.len, res->lockname.name);
0243         spin_unlock(&res->spinlock);
0244         return;
0245     }
0246 
0247     if (!__dlm_lockres_unused(res)) {
0248         mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
0249              dlm->name, res->lockname.len, res->lockname.name);
0250         __dlm_print_one_lock_resource(res);
0251         BUG();
0252     }
0253 
0254     __dlm_unhash_lockres(dlm, res);
0255 
0256     spin_lock(&dlm->track_lock);
0257     if (!list_empty(&res->tracking))
0258         list_del_init(&res->tracking);
0259     else {
0260         mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n",
0261                 res->lockname.len, res->lockname.name);
0262         __dlm_print_one_lock_resource(res);
0263     }
0264     spin_unlock(&dlm->track_lock);
0265 
0266     /* lockres is not in the hash now.  drop the flag and wake up
0267      * any processes waiting in dlm_get_lock_resource. */
0268     if (!master) {
0269         res->state &= ~DLM_LOCK_RES_DROPPING_REF;
0270         spin_unlock(&res->spinlock);
0271         wake_up(&res->wq);
0272     } else
0273         spin_unlock(&res->spinlock);
0274 }
0275 
0276 static void dlm_run_purge_list(struct dlm_ctxt *dlm,
0277                    int purge_now)
0278 {
0279     unsigned int run_max, unused;
0280     unsigned long purge_jiffies;
0281     struct dlm_lock_resource *lockres;
0282 
0283     spin_lock(&dlm->spinlock);
0284     run_max = dlm->purge_count;
0285 
0286     while(run_max && !list_empty(&dlm->purge_list)) {
0287         run_max--;
0288 
0289         lockres = list_entry(dlm->purge_list.next,
0290                      struct dlm_lock_resource, purge);
0291 
0292         spin_lock(&lockres->spinlock);
0293 
0294         purge_jiffies = lockres->last_used +
0295             msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
0296 
0297         /* Make sure that we want to be processing this guy at
0298          * this time. */
0299         if (!purge_now && time_after(purge_jiffies, jiffies)) {
0300             /* Since resources are added to the purge list
0301              * in tail order, we can stop at the first
0302              * unpurgable resource -- anyone added after
0303              * him will have a greater last_used value */
0304             spin_unlock(&lockres->spinlock);
0305             break;
0306         }
0307 
0308         /* Status of the lockres *might* change so double
0309          * check. If the lockres is unused, holding the dlm
0310          * spinlock will prevent people from getting and more
0311          * refs on it. */
0312         unused = __dlm_lockres_unused(lockres);
0313         if (!unused ||
0314             (lockres->state & DLM_LOCK_RES_MIGRATING) ||
0315             (lockres->inflight_assert_workers != 0)) {
0316             mlog(0, "%s: res %.*s is in use or being remastered, "
0317                  "used %d, state %d, assert master workers %u\n",
0318                  dlm->name, lockres->lockname.len,
0319                  lockres->lockname.name,
0320                  !unused, lockres->state,
0321                  lockres->inflight_assert_workers);
0322             list_move_tail(&lockres->purge, &dlm->purge_list);
0323             spin_unlock(&lockres->spinlock);
0324             continue;
0325         }
0326 
0327         dlm_lockres_get(lockres);
0328 
0329         dlm_purge_lockres(dlm, lockres);
0330 
0331         dlm_lockres_put(lockres);
0332 
0333         /* Avoid adding any scheduling latencies */
0334         cond_resched_lock(&dlm->spinlock);
0335     }
0336 
0337     spin_unlock(&dlm->spinlock);
0338 }
0339 
0340 static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
0341                   struct dlm_lock_resource *res)
0342 {
0343     struct dlm_lock *lock, *target;
0344     int can_grant = 1;
0345 
0346     /*
0347      * Because this function is called with the lockres
0348      * spinlock, and because we know that it is not migrating/
0349      * recovering/in-progress, it is fine to reserve asts and
0350      * basts right before queueing them all throughout
0351      */
0352     assert_spin_locked(&dlm->ast_lock);
0353     assert_spin_locked(&res->spinlock);
0354     BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
0355                   DLM_LOCK_RES_RECOVERING|
0356                   DLM_LOCK_RES_IN_PROGRESS)));
0357 
0358 converting:
0359     if (list_empty(&res->converting))
0360         goto blocked;
0361     mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name,
0362          res->lockname.len, res->lockname.name);
0363 
0364     target = list_entry(res->converting.next, struct dlm_lock, list);
0365     if (target->ml.convert_type == LKM_IVMODE) {
0366         mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n",
0367              dlm->name, res->lockname.len, res->lockname.name);
0368         BUG();
0369     }
0370     list_for_each_entry(lock, &res->granted, list) {
0371         if (lock==target)
0372             continue;
0373         if (!dlm_lock_compatible(lock->ml.type,
0374                      target->ml.convert_type)) {
0375             can_grant = 0;
0376             /* queue the BAST if not already */
0377             if (lock->ml.highest_blocked == LKM_IVMODE) {
0378                 __dlm_lockres_reserve_ast(res);
0379                 __dlm_queue_bast(dlm, lock);
0380             }
0381             /* update the highest_blocked if needed */
0382             if (lock->ml.highest_blocked < target->ml.convert_type)
0383                 lock->ml.highest_blocked =
0384                     target->ml.convert_type;
0385         }
0386     }
0387 
0388     list_for_each_entry(lock, &res->converting, list) {
0389         if (lock==target)
0390             continue;
0391         if (!dlm_lock_compatible(lock->ml.type,
0392                      target->ml.convert_type)) {
0393             can_grant = 0;
0394             if (lock->ml.highest_blocked == LKM_IVMODE) {
0395                 __dlm_lockres_reserve_ast(res);
0396                 __dlm_queue_bast(dlm, lock);
0397             }
0398             if (lock->ml.highest_blocked < target->ml.convert_type)
0399                 lock->ml.highest_blocked =
0400                     target->ml.convert_type;
0401         }
0402     }
0403 
0404     /* we can convert the lock */
0405     if (can_grant) {
0406         spin_lock(&target->spinlock);
0407         BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
0408 
0409         mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type "
0410              "%d => %d, node %u\n", dlm->name, res->lockname.len,
0411              res->lockname.name,
0412              dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
0413              dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
0414              target->ml.type,
0415              target->ml.convert_type, target->ml.node);
0416 
0417         target->ml.type = target->ml.convert_type;
0418         target->ml.convert_type = LKM_IVMODE;
0419         list_move_tail(&target->list, &res->granted);
0420 
0421         BUG_ON(!target->lksb);
0422         target->lksb->status = DLM_NORMAL;
0423 
0424         spin_unlock(&target->spinlock);
0425 
0426         __dlm_lockres_reserve_ast(res);
0427         __dlm_queue_ast(dlm, target);
0428         /* go back and check for more */
0429         goto converting;
0430     }
0431 
0432 blocked:
0433     if (list_empty(&res->blocked))
0434         goto leave;
0435     target = list_entry(res->blocked.next, struct dlm_lock, list);
0436 
0437     list_for_each_entry(lock, &res->granted, list) {
0438         if (lock==target)
0439             continue;
0440         if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
0441             can_grant = 0;
0442             if (lock->ml.highest_blocked == LKM_IVMODE) {
0443                 __dlm_lockres_reserve_ast(res);
0444                 __dlm_queue_bast(dlm, lock);
0445             }
0446             if (lock->ml.highest_blocked < target->ml.type)
0447                 lock->ml.highest_blocked = target->ml.type;
0448         }
0449     }
0450 
0451     list_for_each_entry(lock, &res->converting, list) {
0452         if (lock==target)
0453             continue;
0454         if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
0455             can_grant = 0;
0456             if (lock->ml.highest_blocked == LKM_IVMODE) {
0457                 __dlm_lockres_reserve_ast(res);
0458                 __dlm_queue_bast(dlm, lock);
0459             }
0460             if (lock->ml.highest_blocked < target->ml.type)
0461                 lock->ml.highest_blocked = target->ml.type;
0462         }
0463     }
0464 
0465     /* we can grant the blocked lock (only
0466      * possible if converting list empty) */
0467     if (can_grant) {
0468         spin_lock(&target->spinlock);
0469         BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
0470 
0471         mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, "
0472              "node %u\n", dlm->name, res->lockname.len,
0473              res->lockname.name,
0474              dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
0475              dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
0476              target->ml.type, target->ml.node);
0477 
0478         /* target->ml.type is already correct */
0479         list_move_tail(&target->list, &res->granted);
0480 
0481         BUG_ON(!target->lksb);
0482         target->lksb->status = DLM_NORMAL;
0483 
0484         spin_unlock(&target->spinlock);
0485 
0486         __dlm_lockres_reserve_ast(res);
0487         __dlm_queue_ast(dlm, target);
0488         /* go back and check for more */
0489         goto converting;
0490     }
0491 
0492 leave:
0493     return;
0494 }
0495 
0496 /* must have NO locks when calling this with res !=NULL * */
0497 void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
0498 {
0499     if (res) {
0500         spin_lock(&dlm->spinlock);
0501         spin_lock(&res->spinlock);
0502         __dlm_dirty_lockres(dlm, res);
0503         spin_unlock(&res->spinlock);
0504         spin_unlock(&dlm->spinlock);
0505     }
0506     wake_up(&dlm->dlm_thread_wq);
0507 }
0508 
0509 void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
0510 {
0511     assert_spin_locked(&dlm->spinlock);
0512     assert_spin_locked(&res->spinlock);
0513 
0514     /* don't shuffle secondary queues */
0515     if (res->owner == dlm->node_num) {
0516         if (res->state & (DLM_LOCK_RES_MIGRATING |
0517                   DLM_LOCK_RES_BLOCK_DIRTY))
0518             return;
0519 
0520         if (list_empty(&res->dirty)) {
0521             /* ref for dirty_list */
0522             dlm_lockres_get(res);
0523             list_add_tail(&res->dirty, &dlm->dirty_list);
0524             res->state |= DLM_LOCK_RES_DIRTY;
0525         }
0526     }
0527 
0528     mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
0529          res->lockname.name);
0530 }
0531 
0532 
0533 /* Launch the NM thread for the mounted volume */
0534 int dlm_launch_thread(struct dlm_ctxt *dlm)
0535 {
0536     mlog(0, "Starting dlm_thread...\n");
0537 
0538     dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm-%s",
0539             dlm->name);
0540     if (IS_ERR(dlm->dlm_thread_task)) {
0541         mlog_errno(PTR_ERR(dlm->dlm_thread_task));
0542         dlm->dlm_thread_task = NULL;
0543         return -EINVAL;
0544     }
0545 
0546     return 0;
0547 }
0548 
0549 void dlm_complete_thread(struct dlm_ctxt *dlm)
0550 {
0551     if (dlm->dlm_thread_task) {
0552         mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n");
0553         kthread_stop(dlm->dlm_thread_task);
0554         dlm->dlm_thread_task = NULL;
0555     }
0556 }
0557 
0558 static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
0559 {
0560     int empty;
0561 
0562     spin_lock(&dlm->spinlock);
0563     empty = list_empty(&dlm->dirty_list);
0564     spin_unlock(&dlm->spinlock);
0565 
0566     return empty;
0567 }
0568 
0569 static void dlm_flush_asts(struct dlm_ctxt *dlm)
0570 {
0571     int ret;
0572     struct dlm_lock *lock;
0573     struct dlm_lock_resource *res;
0574     u8 hi;
0575 
0576     spin_lock(&dlm->ast_lock);
0577     while (!list_empty(&dlm->pending_asts)) {
0578         lock = list_entry(dlm->pending_asts.next,
0579                   struct dlm_lock, ast_list);
0580         /* get an extra ref on lock */
0581         dlm_lock_get(lock);
0582         res = lock->lockres;
0583         mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, "
0584              "node %u\n", dlm->name, res->lockname.len,
0585              res->lockname.name,
0586              dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
0587              dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
0588              lock->ml.type, lock->ml.node);
0589 
0590         BUG_ON(!lock->ast_pending);
0591 
0592         /* remove from list (including ref) */
0593         list_del_init(&lock->ast_list);
0594         dlm_lock_put(lock);
0595         spin_unlock(&dlm->ast_lock);
0596 
0597         if (lock->ml.node != dlm->node_num) {
0598             ret = dlm_do_remote_ast(dlm, res, lock);
0599             if (ret < 0)
0600                 mlog_errno(ret);
0601         } else
0602             dlm_do_local_ast(dlm, res, lock);
0603 
0604         spin_lock(&dlm->ast_lock);
0605 
0606         /* possible that another ast was queued while
0607          * we were delivering the last one */
0608         if (!list_empty(&lock->ast_list)) {
0609             mlog(0, "%s: res %.*s, AST queued while flushing last "
0610                  "one\n", dlm->name, res->lockname.len,
0611                  res->lockname.name);
0612         } else
0613             lock->ast_pending = 0;
0614 
0615         /* drop the extra ref.
0616          * this may drop it completely. */
0617         dlm_lock_put(lock);
0618         dlm_lockres_release_ast(dlm, res);
0619     }
0620 
0621     while (!list_empty(&dlm->pending_basts)) {
0622         lock = list_entry(dlm->pending_basts.next,
0623                   struct dlm_lock, bast_list);
0624         /* get an extra ref on lock */
0625         dlm_lock_get(lock);
0626         res = lock->lockres;
0627 
0628         BUG_ON(!lock->bast_pending);
0629 
0630         /* get the highest blocked lock, and reset */
0631         spin_lock(&lock->spinlock);
0632         BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
0633         hi = lock->ml.highest_blocked;
0634         lock->ml.highest_blocked = LKM_IVMODE;
0635         spin_unlock(&lock->spinlock);
0636 
0637         /* remove from list (including ref) */
0638         list_del_init(&lock->bast_list);
0639         dlm_lock_put(lock);
0640         spin_unlock(&dlm->ast_lock);
0641 
0642         mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, "
0643              "blocked %d, node %u\n",
0644              dlm->name, res->lockname.len, res->lockname.name,
0645              dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
0646              dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
0647              hi, lock->ml.node);
0648 
0649         if (lock->ml.node != dlm->node_num) {
0650             ret = dlm_send_proxy_bast(dlm, res, lock, hi);
0651             if (ret < 0)
0652                 mlog_errno(ret);
0653         } else
0654             dlm_do_local_bast(dlm, res, lock, hi);
0655 
0656         spin_lock(&dlm->ast_lock);
0657 
0658         /* possible that another bast was queued while
0659          * we were delivering the last one */
0660         if (!list_empty(&lock->bast_list)) {
0661             mlog(0, "%s: res %.*s, BAST queued while flushing last "
0662                  "one\n", dlm->name, res->lockname.len,
0663                  res->lockname.name);
0664         } else
0665             lock->bast_pending = 0;
0666 
0667         /* drop the extra ref.
0668          * this may drop it completely. */
0669         dlm_lock_put(lock);
0670         dlm_lockres_release_ast(dlm, res);
0671     }
0672     wake_up(&dlm->ast_wq);
0673     spin_unlock(&dlm->ast_lock);
0674 }
0675 
0676 
0677 #define DLM_THREAD_TIMEOUT_MS (4 * 1000)
0678 #define DLM_THREAD_MAX_DIRTY  100
0679 
0680 static int dlm_thread(void *data)
0681 {
0682     struct dlm_lock_resource *res;
0683     struct dlm_ctxt *dlm = data;
0684     unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
0685 
0686     mlog(0, "dlm thread running for %s...\n", dlm->name);
0687 
0688     while (!kthread_should_stop()) {
0689         int n = DLM_THREAD_MAX_DIRTY;
0690 
0691         /* dlm_shutting_down is very point-in-time, but that
0692          * doesn't matter as we'll just loop back around if we
0693          * get false on the leading edge of a state
0694          * transition. */
0695         dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
0696 
0697         /* We really don't want to hold dlm->spinlock while
0698          * calling dlm_shuffle_lists on each lockres that
0699          * needs to have its queues adjusted and AST/BASTs
0700          * run.  So let's pull each entry off the dirty_list
0701          * and drop dlm->spinlock ASAP.  Once off the list,
0702          * res->spinlock needs to be taken again to protect
0703          * the queues while calling dlm_shuffle_lists.  */
0704         spin_lock(&dlm->spinlock);
0705         while (!list_empty(&dlm->dirty_list)) {
0706             int delay = 0;
0707             res = list_entry(dlm->dirty_list.next,
0708                      struct dlm_lock_resource, dirty);
0709 
0710             /* peel a lockres off, remove it from the list,
0711              * unset the dirty flag and drop the dlm lock */
0712             BUG_ON(!res);
0713             dlm_lockres_get(res);
0714 
0715             spin_lock(&res->spinlock);
0716             /* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */
0717             list_del_init(&res->dirty);
0718             spin_unlock(&res->spinlock);
0719             spin_unlock(&dlm->spinlock);
0720             /* Drop dirty_list ref */
0721             dlm_lockres_put(res);
0722 
0723             /* lockres can be re-dirtied/re-added to the
0724              * dirty_list in this gap, but that is ok */
0725 
0726             spin_lock(&dlm->ast_lock);
0727             spin_lock(&res->spinlock);
0728             if (res->owner != dlm->node_num) {
0729                 __dlm_print_one_lock_resource(res);
0730                 mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d,"
0731                      " dirty %d\n", dlm->name,
0732                      !!(res->state & DLM_LOCK_RES_IN_PROGRESS),
0733                      !!(res->state & DLM_LOCK_RES_MIGRATING),
0734                      !!(res->state & DLM_LOCK_RES_RECOVERING),
0735                      !!(res->state & DLM_LOCK_RES_DIRTY));
0736             }
0737             BUG_ON(res->owner != dlm->node_num);
0738 
0739             /* it is now ok to move lockreses in these states
0740              * to the dirty list, assuming that they will only be
0741              * dirty for a short while. */
0742             BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
0743             if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
0744                       DLM_LOCK_RES_RECOVERING |
0745                       DLM_LOCK_RES_RECOVERY_WAITING)) {
0746                 /* move it to the tail and keep going */
0747                 res->state &= ~DLM_LOCK_RES_DIRTY;
0748                 spin_unlock(&res->spinlock);
0749                 spin_unlock(&dlm->ast_lock);
0750                 mlog(0, "%s: res %.*s, inprogress, delay list "
0751                      "shuffle, state %d\n", dlm->name,
0752                      res->lockname.len, res->lockname.name,
0753                      res->state);
0754                 delay = 1;
0755                 goto in_progress;
0756             }
0757 
0758             /* at this point the lockres is not migrating/
0759              * recovering/in-progress.  we have the lockres
0760              * spinlock and do NOT have the dlm lock.
0761              * safe to reserve/queue asts and run the lists. */
0762 
0763             /* called while holding lockres lock */
0764             dlm_shuffle_lists(dlm, res);
0765             res->state &= ~DLM_LOCK_RES_DIRTY;
0766             spin_unlock(&res->spinlock);
0767             spin_unlock(&dlm->ast_lock);
0768 
0769             dlm_lockres_calc_usage(dlm, res);
0770 
0771 in_progress:
0772 
0773             spin_lock(&dlm->spinlock);
0774             /* if the lock was in-progress, stick
0775              * it on the back of the list */
0776             if (delay) {
0777                 spin_lock(&res->spinlock);
0778                 __dlm_dirty_lockres(dlm, res);
0779                 spin_unlock(&res->spinlock);
0780             }
0781             dlm_lockres_put(res);
0782 
0783             /* unlikely, but we may need to give time to
0784              * other tasks */
0785             if (!--n) {
0786                 mlog(0, "%s: Throttling dlm thread\n",
0787                      dlm->name);
0788                 break;
0789             }
0790         }
0791 
0792         spin_unlock(&dlm->spinlock);
0793         dlm_flush_asts(dlm);
0794 
0795         /* yield and continue right away if there is more work to do */
0796         if (!n) {
0797             cond_resched();
0798             continue;
0799         }
0800 
0801         wait_event_interruptible_timeout(dlm->dlm_thread_wq,
0802                          !dlm_dirty_list_empty(dlm) ||
0803                          kthread_should_stop(),
0804                          timeout);
0805     }
0806 
0807     mlog(0, "quitting DLM thread\n");
0808     return 0;
0809 }
[ 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!