X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lockd.c;h=3de44bb0e7baf71bf25d7d7508e356ba5b28089b;hb=675dd06e429ee9551d0f874f3461ac3e5091c039;hp=6c459fe77c245a36f62204fbb77254316e4bcf4e;hpb=16cd8ac1f7833a04fff591677006656286f5fa76;p=fs%2Flustre-release.git diff --git a/lustre/ldlm/ldlm_lockd.c b/lustre/ldlm/ldlm_lockd.c index 6c459fe..3de44bb 100644 --- a/lustre/ldlm/ldlm_lockd.c +++ b/lustre/ldlm/ldlm_lockd.c @@ -1,24 +1,42 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (C) 2002, 2003 Cluster File Systems, Inc. - * Author: Peter Braam - * Author: Phil Schwan + * GPL HEADER START * - * This file is part of Lustre, http://www.lustre.org. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * Lustre is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General Public - * License as published by the Free Software Foundation. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * Lustre is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + * GPL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/ldlm/ldlm_lockd.c + * + * Author: Peter Braam + * Author: Phil Schwan */ #ifndef EXPORT_SYMTAB @@ -27,106 +45,133 @@ #define DEBUG_SUBSYSTEM S_LDLM #ifdef __KERNEL__ -# include -# include -# include -# include +# include #else # include #endif -#include -#include +#include +#include #include #include "ldlm_internal.h" -extern kmem_cache_t *ldlm_resource_slab; -extern kmem_cache_t *ldlm_lock_slab; -extern struct list_head ldlm_namespace_list; +#ifdef __KERNEL__ +static int ldlm_num_threads; +CFS_MODULE_PARM(ldlm_num_threads, "i", int, 0444, + "number of DLM service threads to start"); +#endif -static DECLARE_MUTEX(ldlm_ref_sem); +extern cfs_mem_cache_t *ldlm_resource_slab; +extern cfs_mem_cache_t *ldlm_lock_slab; +static cfs_semaphore_t ldlm_ref_sem; static int ldlm_refcount; + /* LDLM state */ static struct ldlm_state *ldlm_state; -static inline unsigned long -round_timeout(unsigned long timeout) +inline cfs_time_t round_timeout(cfs_time_t timeout) +{ + return cfs_time_seconds((int)cfs_duration_sec(cfs_time_sub(timeout, 0)) + 1); +} + +/* timeout for initial callback (AST) reply (bz10399) */ +static inline unsigned int ldlm_get_rq_timeout(void) { - return ((timeout / HZ) + 1) * HZ; + /* Non-AT value */ + unsigned int timeout = min(ldlm_timeout, obd_timeout / 3); + + return timeout < 1 ? 1 : timeout; } #ifdef __KERNEL__ -/* XXX should this be per-ldlm? */ -static struct list_head waiting_locks_list; -static spinlock_t waiting_locks_spinlock; -static struct timer_list waiting_locks_timer; +/* w_l_spinlock protects both waiting_locks_list and expired_lock_thread */ +static cfs_spinlock_t waiting_locks_spinlock; /* BH lock (timer) */ +static cfs_list_t waiting_locks_list; +static cfs_timer_t waiting_locks_timer; static struct expired_lock_thread { - wait_queue_head_t elt_waitq; + cfs_waitq_t elt_waitq; int elt_state; - struct list_head elt_expired_locks; - spinlock_t elt_lock; + int elt_dump; + cfs_list_t elt_expired_locks; } expired_lock_thread; #endif -#if !defined(ENOTSUPP) -# define ENOTSUPP 524 -#endif - #define ELT_STOPPED 0 #define ELT_READY 1 #define ELT_TERMINATE 2 struct ldlm_bl_pool { - spinlock_t blp_lock; - struct list_head blp_list; - wait_queue_head_t blp_waitq; - atomic_t blp_num_threads; - struct completion blp_comp; + cfs_spinlock_t blp_lock; + + /* + * blp_prio_list is used for callbacks that should be handled + * as a priority. It is used for LDLM_FL_DISCARD_DATA requests. + * see bug 13843 + */ + cfs_list_t blp_prio_list; + + /* + * blp_list is used for all other callbacks which are likely + * to take longer to process. + */ + cfs_list_t blp_list; + + cfs_waitq_t blp_waitq; + cfs_completion_t blp_comp; + cfs_atomic_t blp_num_threads; + cfs_atomic_t blp_busy_threads; + int blp_min_threads; + int blp_max_threads; }; struct ldlm_bl_work_item { - struct list_head blwi_entry; - struct ldlm_namespace *blwi_ns; + cfs_list_t blwi_entry; + struct ldlm_namespace *blwi_ns; struct ldlm_lock_desc blwi_ld; - struct ldlm_lock *blwi_lock; + struct ldlm_lock *blwi_lock; + cfs_list_t blwi_head; + int blwi_count; + cfs_completion_t blwi_comp; + cfs_atomic_t blwi_ref_count; }; #ifdef __KERNEL__ +static inline void ldlm_bl_work_item_get(struct ldlm_bl_work_item *blwi) +{ + cfs_atomic_inc(&blwi->blwi_ref_count); +} -static int ldlm_add_waiting_lock(struct ldlm_lock *lock); +static inline void ldlm_bl_work_item_put(struct ldlm_bl_work_item *blwi) +{ + if (cfs_atomic_dec_and_test(&blwi->blwi_ref_count)) + OBD_FREE(blwi, sizeof(*blwi)); +} static inline int have_expired_locks(void) { int need_to_run; - spin_lock_bh(&expired_lock_thread.elt_lock); - need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks); - spin_unlock_bh(&expired_lock_thread.elt_lock); + ENTRY; + cfs_spin_lock_bh(&waiting_locks_spinlock); + need_to_run = !cfs_list_empty(&expired_lock_thread.elt_expired_locks); + cfs_spin_unlock_bh(&waiting_locks_spinlock); RETURN(need_to_run); } static int expired_lock_main(void *arg) { - struct list_head *expired = &expired_lock_thread.elt_expired_locks; + cfs_list_t *expired = &expired_lock_thread.elt_expired_locks; struct l_wait_info lwi = { 0 }; - unsigned long flags; + int do_dump; ENTRY; - lock_kernel(); - kportal_daemonize("ldlm_elt"); - - SIGNAL_MASK_LOCK(current, flags); - sigfillset(¤t->blocked); - RECALC_SIGPENDING; - SIGNAL_MASK_UNLOCK(current, flags); - - unlock_kernel(); + cfs_daemonize("ldlm_elt"); expired_lock_thread.elt_state = ELT_READY; - wake_up(&expired_lock_thread.elt_waitq); + cfs_waitq_signal(&expired_lock_thread.elt_waitq); while (1) { l_wait_event(expired_lock_thread.elt_waitq, @@ -134,60 +179,116 @@ static int expired_lock_main(void *arg) expired_lock_thread.elt_state == ELT_TERMINATE, &lwi); - spin_lock_bh(&expired_lock_thread.elt_lock); - while (!list_empty(expired)) { + cfs_spin_lock_bh(&waiting_locks_spinlock); + if (expired_lock_thread.elt_dump) { + cfs_spin_unlock_bh(&waiting_locks_spinlock); + + /* from waiting_locks_callback, but not in timer */ + libcfs_debug_dumplog(); + libcfs_run_lbug_upcall(__FILE__, + "waiting_locks_callback", + expired_lock_thread.elt_dump); + + cfs_spin_lock_bh(&waiting_locks_spinlock); + expired_lock_thread.elt_dump = 0; + } + + do_dump = 0; + + while (!cfs_list_empty(expired)) { struct obd_export *export; struct ldlm_lock *lock; - lock = list_entry(expired->next, struct ldlm_lock, + lock = cfs_list_entry(expired->next, struct ldlm_lock, l_pending_chain); - if ((void *)lock < LP_POISON + PAGE_SIZE && + if ((void *)lock < LP_POISON + CFS_PAGE_SIZE && (void *)lock >= LP_POISON) { + cfs_spin_unlock_bh(&waiting_locks_spinlock); CERROR("free lock on elt list %p\n", lock); LBUG(); } - list_del_init(&lock->l_pending_chain); - if ((void *)lock->l_export < LP_POISON + PAGE_SIZE && - (void *)lock->l_export >= LP_POISON + PAGE_SIZE) { + cfs_list_del_init(&lock->l_pending_chain); + if ((void *)lock->l_export < LP_POISON + CFS_PAGE_SIZE && + (void *)lock->l_export >= LP_POISON) { CERROR("lock with free export on elt list %p\n", - export); + lock->l_export); lock->l_export = NULL; - LDLM_ERROR(lock, "free export\n"); + LDLM_ERROR(lock, "free export"); + /* release extra ref grabbed by + * ldlm_add_waiting_lock() or + * ldlm_failed_ast() */ + LDLM_LOCK_RELEASE(lock); continue; } - export = class_export_get(lock->l_export); - spin_unlock_bh(&expired_lock_thread.elt_lock); + export = class_export_lock_get(lock->l_export, lock); + cfs_spin_unlock_bh(&waiting_locks_spinlock); + + do_dump++; + class_fail_export(export); + class_export_lock_put(export, lock); + + /* release extra ref grabbed by ldlm_add_waiting_lock() + * or ldlm_failed_ast() */ + LDLM_LOCK_RELEASE(lock); - ptlrpc_fail_export(export); - class_export_put(export); - spin_lock_bh(&expired_lock_thread.elt_lock); + cfs_spin_lock_bh(&waiting_locks_spinlock); + } + cfs_spin_unlock_bh(&waiting_locks_spinlock); + + if (do_dump && obd_dump_on_eviction) { + CERROR("dump the log upon eviction\n"); + libcfs_debug_dumplog(); } - spin_unlock_bh(&expired_lock_thread.elt_lock); if (expired_lock_thread.elt_state == ELT_TERMINATE) break; } expired_lock_thread.elt_state = ELT_STOPPED; - wake_up(&expired_lock_thread.elt_waitq); + cfs_waitq_signal(&expired_lock_thread.elt_waitq); RETURN(0); } -static void waiting_locks_callback(unsigned long unused) +static int ldlm_add_waiting_lock(struct ldlm_lock *lock); + +/** + * Check if there is a request in the export request list + * which prevents the lock canceling. + */ +static int ldlm_lock_busy(struct ldlm_lock *lock) { - struct ldlm_lock *lock; - char str[PTL_NALFMT_SIZE]; + struct ptlrpc_request *req; + int match = 0; + ENTRY; - if (obd_dump_on_timeout) - portals_debug_dumplog(); + if (lock->l_export == NULL) + return 0; -repeat: - spin_lock_bh(&waiting_locks_spinlock); - while (!list_empty(&waiting_locks_list)) { - lock = list_entry(waiting_locks_list.next, struct ldlm_lock, - l_pending_chain); + cfs_spin_lock(&lock->l_export->exp_lock); + cfs_list_for_each_entry(req, &lock->l_export->exp_queued_rpc, + rq_exp_list) { + if (req->rq_ops->hpreq_lock_match) { + match = req->rq_ops->hpreq_lock_match(req, lock); + if (match) + break; + } + } + cfs_spin_unlock(&lock->l_export->exp_lock); + RETURN(match); +} + +/* This is called from within a timer interrupt and cannot schedule */ +static void waiting_locks_callback(unsigned long unused) +{ + struct ldlm_lock *lock, *last = NULL; - if ((lock->l_callback_timeout > jiffies) || +repeat: + cfs_spin_lock_bh(&waiting_locks_spinlock); + while (!cfs_list_empty(&waiting_locks_list)) { + lock = cfs_list_entry(waiting_locks_list.next, struct ldlm_lock, + l_pending_chain); + if (cfs_time_after(lock->l_callback_timeout, + cfs_time_current()) || (lock->l_req_mode == LCK_GROUP)) break; @@ -200,41 +301,90 @@ repeat: LDLM_ERROR(lock, "recharge timeout: %s@%s nid %s ", lock->l_export->exp_client_uuid.uuid, lock->l_export->exp_connection->c_remote_uuid.uuid, - ptlrpc_peernid2str(&lock->l_export->exp_connection->c_peer, str)); + libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid)); - list_del_init(&lock->l_pending_chain); - spin_unlock_bh(&waiting_locks_spinlock); + cfs_list_del_init(&lock->l_pending_chain); + cfs_spin_unlock_bh(&waiting_locks_spinlock); ldlm_add_waiting_lock(lock); + goto repeat; + } + + /* if timeout overlaps the activation time of suspended timeouts + * then extend it to give a chance for client to reconnect */ + if (cfs_time_before(cfs_time_sub(lock->l_callback_timeout, + cfs_time_seconds(obd_timeout)/2), + ptlrpc_suspend_wakeup_time())) { + LDLM_ERROR(lock, "extend timeout due to recovery: %s@%s nid %s ", + lock->l_export->exp_client_uuid.uuid, + lock->l_export->exp_connection->c_remote_uuid.uuid, + libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid)); + cfs_list_del_init(&lock->l_pending_chain); + cfs_spin_unlock_bh(&waiting_locks_spinlock); + ldlm_add_waiting_lock(lock); goto repeat; } - LDLM_ERROR(lock, "lock callback timer expired: evicting client " - "%s@%s nid %s ", - lock->l_export->exp_client_uuid.uuid, - lock->l_export->exp_connection->c_remote_uuid.uuid, - ptlrpc_peernid2str(&lock->l_export->exp_connection->c_peer, str)); + /* Check if we need to prolong timeout */ + if (!OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_TIMEOUT) && + ldlm_lock_busy(lock)) { + int cont = 1; + + if (lock->l_pending_chain.next == &waiting_locks_list) + cont = 0; - spin_lock_bh(&expired_lock_thread.elt_lock); - list_del(&lock->l_pending_chain); - list_add(&lock->l_pending_chain, - &expired_lock_thread.elt_expired_locks); - spin_unlock_bh(&expired_lock_thread.elt_lock); - wake_up(&expired_lock_thread.elt_waitq); + LDLM_LOCK_GET(lock); + + cfs_spin_unlock_bh(&waiting_locks_spinlock); + LDLM_DEBUG(lock, "prolong the busy lock"); + ldlm_refresh_waiting_lock(lock, + ldlm_get_enq_timeout(lock)); + cfs_spin_lock_bh(&waiting_locks_spinlock); + + if (!cont) { + LDLM_LOCK_RELEASE(lock); + break; + } + + LDLM_LOCK_RELEASE(lock); + continue; + } + lock->l_resource->lr_namespace->ns_timeouts++; + LDLM_ERROR(lock, "lock callback timer expired after %lds: " + "evicting client at %s ", + cfs_time_current_sec()- lock->l_last_activity, + libcfs_nid2str( + lock->l_export->exp_connection->c_peer.nid)); + + last = lock; + + /* no needs to take an extra ref on the lock since it was in + * the waiting_locks_list and ldlm_add_waiting_lock() + * already grabbed a ref */ + cfs_list_del(&lock->l_pending_chain); + cfs_list_add(&lock->l_pending_chain, + &expired_lock_thread.elt_expired_locks); + } + + if (!cfs_list_empty(&expired_lock_thread.elt_expired_locks)) { + if (obd_dump_on_timeout) + expired_lock_thread.elt_dump = __LINE__; + + cfs_waitq_signal(&expired_lock_thread.elt_waitq); } /* * Make sure the timer will fire again if we have any locks * left. */ - if (!list_empty(&waiting_locks_list)) { - unsigned long timeout_rounded; - lock = list_entry(waiting_locks_list.next, struct ldlm_lock, - l_pending_chain); - timeout_rounded = round_timeout(lock->l_callback_timeout); - mod_timer(&waiting_locks_timer, timeout_rounded); - } - spin_unlock_bh(&waiting_locks_spinlock); + if (!cfs_list_empty(&waiting_locks_list)) { + cfs_time_t timeout_rounded; + lock = cfs_list_entry(waiting_locks_list.next, struct ldlm_lock, + l_pending_chain); + timeout_rounded = (cfs_time_t)round_timeout(lock->l_callback_timeout); + cfs_timer_arm(&waiting_locks_timer, timeout_rounded); + } + cfs_spin_unlock_bh(&waiting_locks_spinlock); } /* @@ -242,83 +392,165 @@ repeat: * lock. We add it to the pending-callback chain, and schedule the lock-timeout * timer to fire appropriately. (We round up to the next second, to avoid * floods of timer firings during periods of high lock contention and traffic). + * As done by ldlm_add_waiting_lock(), the caller must grab a lock reference + * if it has been added to the waiting list (1 is returned). + * + * Called with the namespace lock held. */ -static int ldlm_add_waiting_lock(struct ldlm_lock *lock) +static int __ldlm_add_waiting_lock(struct ldlm_lock *lock, int seconds) { - unsigned long timeout_rounded; + cfs_time_t timeout; + cfs_time_t timeout_rounded; - spin_lock_bh(&waiting_locks_spinlock); - if (!list_empty(&lock->l_pending_chain)) { - LDLM_DEBUG(lock, "not re-adding to wait list"); - spin_unlock_bh(&waiting_locks_spinlock); + if (!cfs_list_empty(&lock->l_pending_chain)) return 0; - } - LDLM_DEBUG(lock, "adding to wait list"); - lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2); + if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_NOTIMEOUT) || + OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_HPREQ_TIMEOUT)) + seconds = 1; + + timeout = cfs_time_shift(seconds); + if (likely(cfs_time_after(timeout, lock->l_callback_timeout))) + lock->l_callback_timeout = timeout; timeout_rounded = round_timeout(lock->l_callback_timeout); - if (timeout_rounded < waiting_locks_timer.expires || - !timer_pending(&waiting_locks_timer)) { - mod_timer(&waiting_locks_timer, timeout_rounded); + if (cfs_time_before(timeout_rounded, + cfs_timer_deadline(&waiting_locks_timer)) || + !cfs_timer_is_armed(&waiting_locks_timer)) { + cfs_timer_arm(&waiting_locks_timer, timeout_rounded); } - list_add_tail(&lock->l_pending_chain, &waiting_locks_list); /* FIFO */ - spin_unlock_bh(&waiting_locks_spinlock); + /* if the new lock has a shorter timeout than something earlier on + the list, we'll wait the longer amount of time; no big deal. */ + /* FIFO */ + cfs_list_add_tail(&lock->l_pending_chain, &waiting_locks_list); return 1; } +static int ldlm_add_waiting_lock(struct ldlm_lock *lock) +{ + int ret; + int timeout = ldlm_get_enq_timeout(lock); + + LASSERT(!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)); + + cfs_spin_lock_bh(&waiting_locks_spinlock); + if (lock->l_destroyed) { + static cfs_time_t next; + cfs_spin_unlock_bh(&waiting_locks_spinlock); + LDLM_ERROR(lock, "not waiting on destroyed lock (bug 5653)"); + if (cfs_time_after(cfs_time_current(), next)) { + next = cfs_time_shift(14400); + libcfs_debug_dumpstack(NULL); + } + return 0; + } + + ret = __ldlm_add_waiting_lock(lock, timeout); + if (ret) + /* grab ref on the lock if it has been added to the + * waiting list */ + LDLM_LOCK_GET(lock); + cfs_spin_unlock_bh(&waiting_locks_spinlock); + + LDLM_DEBUG(lock, "%sadding to wait list(timeout: %d, AT: %s)", + ret == 0 ? "not re-" : "", timeout, + AT_OFF ? "off" : "on"); + return ret; +} + /* * Remove a lock from the pending list, likely because it had its cancellation * callback arrive without incident. This adjusts the lock-timeout timer if * needed. Returns 0 if the lock wasn't pending after all, 1 if it was. + * As done by ldlm_del_waiting_lock(), the caller must release the lock + * reference when the lock is removed from any list (1 is returned). + * + * Called with namespace lock held. */ -int ldlm_del_waiting_lock(struct ldlm_lock *lock) +static int __ldlm_del_waiting_lock(struct ldlm_lock *lock) { - struct list_head *list_next; - - if (lock->l_export == NULL) { - /* We don't have a "waiting locks list" on clients. */ - LDLM_DEBUG(lock, "client lock: no-op"); - return 0; - } - - spin_lock_bh(&waiting_locks_spinlock); + cfs_list_t *list_next; - if (list_empty(&lock->l_pending_chain)) { - spin_unlock_bh(&waiting_locks_spinlock); - LDLM_DEBUG(lock, "wasn't waiting"); + if (cfs_list_empty(&lock->l_pending_chain)) return 0; - } list_next = lock->l_pending_chain.next; if (lock->l_pending_chain.prev == &waiting_locks_list) { /* Removing the head of the list, adjust timer. */ if (list_next == &waiting_locks_list) { /* No more, just cancel. */ - del_timer(&waiting_locks_timer); + cfs_timer_disarm(&waiting_locks_timer); } else { struct ldlm_lock *next; - next = list_entry(list_next, struct ldlm_lock, - l_pending_chain); - mod_timer(&waiting_locks_timer, - round_timeout(next->l_callback_timeout)); + next = cfs_list_entry(list_next, struct ldlm_lock, + l_pending_chain); + cfs_timer_arm(&waiting_locks_timer, + round_timeout(next->l_callback_timeout)); } } + cfs_list_del_init(&lock->l_pending_chain); - spin_lock_bh(&expired_lock_thread.elt_lock); - list_del_init(&lock->l_pending_chain); - spin_unlock_bh(&expired_lock_thread.elt_lock); - - spin_unlock_bh(&waiting_locks_spinlock); - LDLM_DEBUG(lock, "removed"); return 1; } +int ldlm_del_waiting_lock(struct ldlm_lock *lock) +{ + int ret; + + if (lock->l_export == NULL) { + /* We don't have a "waiting locks list" on clients. */ + CDEBUG(D_DLMTRACE, "Client lock %p : no-op\n", lock); + return 0; + } + + cfs_spin_lock_bh(&waiting_locks_spinlock); + ret = __ldlm_del_waiting_lock(lock); + cfs_spin_unlock_bh(&waiting_locks_spinlock); + if (ret) + /* release lock ref if it has indeed been removed + * from a list */ + LDLM_LOCK_RELEASE(lock); + + LDLM_DEBUG(lock, "%s", ret == 0 ? "wasn't waiting" : "removed"); + return ret; +} + +/* + * Prolong the lock + * + * Called with namespace lock held. + */ +int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout) +{ + if (lock->l_export == NULL) { + /* We don't have a "waiting locks list" on clients. */ + LDLM_DEBUG(lock, "client lock: no-op"); + return 0; + } + + cfs_spin_lock_bh(&waiting_locks_spinlock); + + if (cfs_list_empty(&lock->l_pending_chain)) { + cfs_spin_unlock_bh(&waiting_locks_spinlock); + LDLM_DEBUG(lock, "wasn't waiting"); + return 0; + } + + /* we remove/add the lock to the waiting list, so no needs to + * release/take a lock reference */ + __ldlm_del_waiting_lock(lock); + __ldlm_add_waiting_lock(lock, timeout); + cfs_spin_unlock_bh(&waiting_locks_spinlock); + + LDLM_DEBUG(lock, "refreshed"); + return 1; +} #else /* !__KERNEL__ */ static int ldlm_add_waiting_lock(struct ldlm_lock *lock) { + LASSERT(!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)); RETURN(1); } @@ -327,36 +559,55 @@ int ldlm_del_waiting_lock(struct ldlm_lock *lock) RETURN(0); } +int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout) +{ + RETURN(0); +} #endif /* __KERNEL__ */ -static void ldlm_failed_ast(struct ldlm_lock *lock, int rc,const char *ast_type) +static void ldlm_failed_ast(struct ldlm_lock *lock, int rc, + const char *ast_type) { - struct ptlrpc_connection *conn = lock->l_export->exp_connection; - char str[PTL_NALFMT_SIZE]; - - LDLM_ERROR(lock, "%s AST failed (%d): evicting client %s@%s NID "LPX64 - " (%s)", ast_type, rc, lock->l_export->exp_client_uuid.uuid, - conn->c_remote_uuid.uuid, conn->c_peer.peer_id.nid, - ptlrpc_peernid2str(&conn->c_peer, str)); + LCONSOLE_ERROR_MSG(0x138, "%s: A client on nid %s was evicted due " + "to a lock %s callback time out: rc %d\n", + lock->l_export->exp_obd->obd_name, + obd_export_nid2str(lock->l_export), ast_type, rc); if (obd_dump_on_timeout) - portals_debug_dumplog(); - ptlrpc_fail_export(lock->l_export); + libcfs_debug_dumplog(); +#ifdef __KERNEL__ + cfs_spin_lock_bh(&waiting_locks_spinlock); + if (__ldlm_del_waiting_lock(lock) == 0) + /* the lock was not in any list, grab an extra ref before adding + * the lock to the expired list */ + LDLM_LOCK_GET(lock); + cfs_list_add(&lock->l_pending_chain, + &expired_lock_thread.elt_expired_locks); + cfs_waitq_signal(&expired_lock_thread.elt_waitq); + cfs_spin_unlock_bh(&waiting_locks_spinlock); +#else + class_fail_export(lock->l_export); +#endif } static int ldlm_handle_ast_error(struct ldlm_lock *lock, struct ptlrpc_request *req, int rc, const char *ast_type) { - struct ptlrpc_peer *peer = &req->rq_import->imp_connection->c_peer; - char str[PTL_NALFMT_SIZE]; + lnet_process_id_t peer = req->rq_import->imp_connection->c_peer; if (rc == -ETIMEDOUT || rc == -EINTR || rc == -ENOTCONN) { LASSERT(lock->l_export); if (lock->l_export->exp_libclient) { LDLM_DEBUG(lock, "%s AST to liblustre client (nid %s)" " timeout, just cancelling lock", ast_type, - ptlrpc_peernid2str(peer, str)); + libcfs_nid2str(peer.nid)); + ldlm_lock_cancel(lock); + rc = -ERESTART; + } else if (lock->l_flags & LDLM_FL_CANCEL) { + LDLM_DEBUG(lock, "%s AST timeout from nid %s, but " + "cancel was received (AST reply lost?)", + ast_type, libcfs_nid2str(peer.nid)); ldlm_lock_cancel(lock); rc = -ERESTART; } else { @@ -366,14 +617,17 @@ static int ldlm_handle_ast_error(struct ldlm_lock *lock, } else if (rc) { if (rc == -EINVAL) LDLM_DEBUG(lock, "client (nid %s) returned %d" - " from %s AST - normal race", - ptlrpc_peernid2str(peer, str), - req->rq_repmsg->status, ast_type); + " from %s AST - normal race", + libcfs_nid2str(peer.nid), + req->rq_repmsg ? + lustre_msg_get_status(req->rq_repmsg) : -1, + ast_type); else LDLM_ERROR(lock, "client (nid %s) returned %d " - "from %s AST", ptlrpc_peernid2str(peer, str), + "from %s AST", libcfs_nid2str(peer.nid), (req->rq_repmsg != NULL) ? - req->rq_repmsg->status : 0, ast_type); + lustre_msg_get_status(req->rq_repmsg) : 0, + ast_type); ldlm_lock_cancel(lock); /* Server-side AST functions are called from ldlm_reprocess_all, * which needs to be told to please restart its reprocessing. */ @@ -383,187 +637,328 @@ static int ldlm_handle_ast_error(struct ldlm_lock *lock, return rc; } +static int ldlm_cb_interpret(const struct lu_env *env, + struct ptlrpc_request *req, void *data, int rc) +{ + struct ldlm_cb_set_arg *arg; + struct ldlm_lock *lock; + ENTRY; + + LASSERT(data != NULL); + + arg = req->rq_async_args.pointer_arg[0]; + lock = req->rq_async_args.pointer_arg[1]; + LASSERT(lock != NULL); + if (rc != 0) { + /* If client canceled the lock but the cancel has not + * been received yet, we need to update lvbo to have the + * proper attributes cached. */ + if (rc == -EINVAL && arg->type == LDLM_BL_CALLBACK) + ldlm_res_lvbo_update(lock->l_resource, NULL, 1); + rc = ldlm_handle_ast_error(lock, req, rc, + arg->type == LDLM_BL_CALLBACK + ? "blocking" : "completion"); + } + + LDLM_LOCK_RELEASE(lock); + + if (rc == -ERESTART) + cfs_atomic_set(&arg->restart, 1); + + RETURN(0); +} + +static inline int ldlm_bl_and_cp_ast_fini(struct ptlrpc_request *req, + struct ldlm_cb_set_arg *arg, + struct ldlm_lock *lock, + int instant_cancel) +{ + int rc = 0; + ENTRY; + + if (unlikely(instant_cancel)) { + rc = ptl_send_rpc(req, 1); + ptlrpc_req_finished(req); + if (rc == 0) + /* If we cancelled the lock, we need to restart + * ldlm_reprocess_queue */ + cfs_atomic_set(&arg->restart, 1); + } else { + LDLM_LOCK_GET(lock); + ptlrpc_set_add_req(arg->set, req); + } + + RETURN(rc); +} + +/** + * Check if there are requests in the export request list which prevent + * the lock canceling and make these requests high priority ones. + */ +static void ldlm_lock_reorder_req(struct ldlm_lock *lock) +{ + struct ptlrpc_request *req; + ENTRY; + + if (lock->l_export == NULL) { + LDLM_DEBUG(lock, "client lock: no-op"); + RETURN_EXIT; + } + + cfs_spin_lock(&lock->l_export->exp_lock); + cfs_list_for_each_entry(req, &lock->l_export->exp_queued_rpc, + rq_exp_list) { + if (!req->rq_hp && req->rq_ops->hpreq_lock_match && + req->rq_ops->hpreq_lock_match(req, lock)) + ptlrpc_hpreq_reorder(req); + } + cfs_spin_unlock(&lock->l_export->exp_lock); + EXIT; +} + +/* + * ->l_blocking_ast() method for server-side locks. This is invoked when newly + * enqueued server lock conflicts with given one. + * + * Sends blocking ast rpc to the client owning that lock; arms timeout timer + * to wait for client response. + */ int ldlm_server_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, void *data, int flag) { - struct ldlm_request *body; - struct ptlrpc_request *req; - int rc = 0, size = sizeof(*body); + struct ldlm_cb_set_arg *arg = data; + struct ldlm_request *body; + struct ptlrpc_request *req; + int instant_cancel = 0; + int rc = 0; ENTRY; - if (flag == LDLM_CB_CANCELING) { + if (flag == LDLM_CB_CANCELING) /* Don't need to do anything here. */ RETURN(0); - } - - LASSERTF(lock->l_export->exp_obd->obd_recovering == 0, - "BUG 6063: lock collide during recovery"); LASSERT(lock); - -#if 0 - if (LTIME_S(CURRENT_TIME) - lock->l_export->exp_last_request_time > 30){ - ldlm_failed_ast(lock, -ETIMEDOUT, "Not-attempted blocking"); - l_unlock(&lock->l_resource->lr_namespace->ns_lock); - RETURN(-ETIMEDOUT); + LASSERT(data != NULL); + if (lock->l_export->exp_obd->obd_recovering != 0) { + LDLM_ERROR(lock, "BUG 6063: lock collide during recovery"); + ldlm_lock_dump(D_ERROR, lock, 0); } -#endif - req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, - LUSTRE_DLM_VERSION, LDLM_BL_CALLBACK, - 1, &size, NULL); + ldlm_lock_reorder_req(lock); + + req = ptlrpc_request_alloc_pack(lock->l_export->exp_imp_reverse, + &RQF_LDLM_BL_CALLBACK, + LUSTRE_DLM_VERSION, LDLM_BL_CALLBACK); if (req == NULL) RETURN(-ENOMEM); - lock_res_and_lock(lock); + req->rq_async_args.pointer_arg[0] = arg; + req->rq_async_args.pointer_arg[1] = lock; + req->rq_interpret_reply = ldlm_cb_interpret; + req->rq_no_resend = 1; + + lock_res(lock->l_resource); if (lock->l_granted_mode != lock->l_req_mode) { /* this blocking AST will be communicated as part of the * completion AST instead */ - unlock_res_and_lock(lock); - LDLM_DEBUG(lock, "lock not granted, not sending blocking AST"); + unlock_res(lock->l_resource); ptlrpc_req_finished(req); + LDLM_DEBUG(lock, "lock not granted, not sending blocking AST"); RETURN(0); } if (lock->l_destroyed) { /* What's the point? */ - unlock_res_and_lock(lock); + unlock_res(lock->l_resource); ptlrpc_req_finished(req); RETURN(0); } - body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body)); - memcpy(&body->lock_handle1, &lock->l_remote_handle, - sizeof(body->lock_handle1)); - memcpy(&body->lock_desc, desc, sizeof(*desc)); + if (lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK) + instant_cancel = 1; + + body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); + body->lock_handle[0] = lock->l_remote_handle; + body->lock_desc = *desc; body->lock_flags |= (lock->l_flags & LDLM_AST_FLAGS); LDLM_DEBUG(lock, "server preparing blocking AST"); - req->rq_replen = lustre_msg_size(0, NULL); - if (lock->l_granted_mode == lock->l_req_mode) + ptlrpc_request_set_replen(req); + if (instant_cancel) { + unlock_res(lock->l_resource); + ldlm_lock_cancel(lock); + } else { + LASSERT(lock->l_granted_mode == lock->l_req_mode); ldlm_add_waiting_lock(lock); - unlock_res_and_lock(lock); + unlock_res(lock->l_resource); + } req->rq_send_state = LUSTRE_IMP_FULL; - req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */ - rc = ptlrpc_queue_wait(req); - if (rc != 0) - rc = ldlm_handle_ast_error(lock, req, rc, "blocking"); + /* ptlrpc_request_alloc_pack already set timeout */ + if (AT_OFF) + req->rq_timeout = ldlm_get_rq_timeout(); - ptlrpc_req_finished(req); + if (lock->l_export && lock->l_export->exp_nid_stats && + lock->l_export->exp_nid_stats->nid_ldlm_stats) + lprocfs_counter_incr(lock->l_export->exp_nid_stats->nid_ldlm_stats, + LDLM_BL_CALLBACK - LDLM_FIRST_OPC); - RETURN(rc); -} + rc = ldlm_bl_and_cp_ast_fini(req, arg, lock, instant_cancel); -/* XXX copied from ptlrpc/service.c */ -static long timeval_sub(struct timeval *large, struct timeval *small) -{ - return (large->tv_sec - small->tv_sec) * 1000000 + - (large->tv_usec - small->tv_usec); + RETURN(rc); } int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data) { - struct ldlm_request *body; - struct ptlrpc_request *req; - struct timeval granted_time; - long total_enqueue_wait; - int rc = 0, size[2] = {sizeof(*body)}, buffers = 1; + struct ldlm_cb_set_arg *arg = data; + struct ldlm_request *body; + struct ptlrpc_request *req; + long total_enqueue_wait; + int instant_cancel = 0; + int rc = 0; ENTRY; LASSERT(lock != NULL); + LASSERT(data != NULL); - do_gettimeofday(&granted_time); - total_enqueue_wait = timeval_sub(&granted_time, &lock->l_enqueued_time); + total_enqueue_wait = cfs_time_sub(cfs_time_current_sec(), + lock->l_last_activity); - if (total_enqueue_wait / 1000000 > obd_timeout) - LDLM_ERROR(lock, "enqueue wait took %ldus", total_enqueue_wait); + req = ptlrpc_request_alloc(lock->l_export->exp_imp_reverse, + &RQF_LDLM_CP_CALLBACK); + if (req == NULL) + RETURN(-ENOMEM); lock_res_and_lock(lock); - if (lock->l_resource->lr_lvb_len) { - buffers = 2; - size[1] = lock->l_resource->lr_lvb_len; - } + if (lock->l_resource->lr_lvb_len) + req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_CLIENT, + lock->l_resource->lr_lvb_len); unlock_res_and_lock(lock); - - req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, - LUSTRE_DLM_VERSION, LDLM_CP_CALLBACK, - buffers, size, NULL); - if (req == NULL) - RETURN(-ENOMEM); - body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body)); - memcpy(&body->lock_handle1, &lock->l_remote_handle, - sizeof(body->lock_handle1)); + rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_CP_CALLBACK); + if (rc) { + ptlrpc_request_free(req); + RETURN(rc); + } + + req->rq_async_args.pointer_arg[0] = arg; + req->rq_async_args.pointer_arg[1] = lock; + req->rq_interpret_reply = ldlm_cb_interpret; + req->rq_no_resend = 1; + body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); + + body->lock_handle[0] = lock->l_remote_handle; body->lock_flags = flags; ldlm_lock2desc(lock, &body->lock_desc); + if (lock->l_resource->lr_lvb_len) { + void *lvb = req_capsule_client_get(&req->rq_pill, &RMF_DLM_LVB); - if (buffers == 2) { - void *lvb; - - lvb = lustre_msg_buf(req->rq_reqmsg, 1, - lock->l_resource->lr_lvb_len); lock_res_and_lock(lock); memcpy(lvb, lock->l_resource->lr_lvb_data, lock->l_resource->lr_lvb_len); unlock_res_and_lock(lock); } - LDLM_DEBUG(lock, "server preparing completion AST (after %ldus wait)", + LDLM_DEBUG(lock, "server preparing completion AST (after %lds wait)", total_enqueue_wait); - req->rq_replen = lustre_msg_size(0, NULL); + + /* Server-side enqueue wait time estimate, used in + __ldlm_add_waiting_lock to set future enqueue timers */ + if (total_enqueue_wait < ldlm_get_enq_timeout(lock)) + at_measured(&lock->l_resource->lr_namespace->ns_at_estimate, + total_enqueue_wait); + else + /* bz18618. Don't add lock enqueue time we spend waiting for a + previous callback to fail. Locks waiting legitimately will + get extended by ldlm_refresh_waiting_lock regardless of the + estimate, so it's okay to underestimate here. */ + LDLM_DEBUG(lock, "lock completed after %lus; estimate was %ds. " + "It is likely that a previous callback timed out.", + total_enqueue_wait, + at_get(&lock->l_resource->lr_namespace->ns_at_estimate)); + + ptlrpc_request_set_replen(req); req->rq_send_state = LUSTRE_IMP_FULL; - req->rq_timeout = ldlm_timeout; /* timeout for initial AST reply */ + /* ptlrpc_request_pack already set timeout */ + if (AT_OFF) + req->rq_timeout = ldlm_get_rq_timeout(); /* We only send real blocking ASTs after the lock is granted */ lock_res_and_lock(lock); if (lock->l_flags & LDLM_FL_AST_SENT) { body->lock_flags |= LDLM_FL_AST_SENT; - ldlm_add_waiting_lock(lock); /* start the lock-timeout clock */ + /* copy ast flags like LDLM_FL_DISCARD_DATA */ + body->lock_flags |= (lock->l_flags & LDLM_AST_FLAGS); + + /* We might get here prior to ldlm_handle_enqueue setting + * LDLM_FL_CANCEL_ON_BLOCK flag. Then we will put this lock + * into waiting list, but this is safe and similar code in + * ldlm_handle_enqueue will call ldlm_lock_cancel() still, + * that would not only cancel the lock, but will also remove + * it from waiting list */ + if (lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK) { + unlock_res_and_lock(lock); + ldlm_lock_cancel(lock); + instant_cancel = 1; + lock_res_and_lock(lock); + } else { + /* start the lock-timeout clock */ + ldlm_add_waiting_lock(lock); + } } unlock_res_and_lock(lock); - rc = ptlrpc_queue_wait(req); - if (rc != 0) - rc = ldlm_handle_ast_error(lock, req, rc, "completion"); + if (lock->l_export && lock->l_export->exp_nid_stats && + lock->l_export->exp_nid_stats->nid_ldlm_stats) + lprocfs_counter_incr(lock->l_export->exp_nid_stats->nid_ldlm_stats, + LDLM_CP_CALLBACK - LDLM_FIRST_OPC); - ptlrpc_req_finished(req); + rc = ldlm_bl_and_cp_ast_fini(req, arg, lock, instant_cancel); RETURN(rc); } int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data) { - struct ldlm_resource *res; - struct ldlm_request *body; + struct ldlm_resource *res = lock->l_resource; + struct ldlm_request *body; struct ptlrpc_request *req; - int rc = 0, size = sizeof(*body); + int rc; ENTRY; LASSERT(lock != NULL); - req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, - LUSTRE_DLM_VERSION, LDLM_GL_CALLBACK, - 1, &size, NULL); + req = ptlrpc_request_alloc_pack(lock->l_export->exp_imp_reverse, + &RQF_LDLM_GL_CALLBACK, + LUSTRE_DLM_VERSION, LDLM_GL_CALLBACK); + if (req == NULL) RETURN(-ENOMEM); - body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body)); - memcpy(&body->lock_handle1, &lock->l_remote_handle, - sizeof(body->lock_handle1)); + body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); + body->lock_handle[0] = lock->l_remote_handle; ldlm_lock2desc(lock, &body->lock_desc); lock_res_and_lock(lock); - size = lock->l_resource->lr_lvb_len; + req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, + lock->l_resource->lr_lvb_len); unlock_res_and_lock(lock); res = lock->l_resource; - req->rq_replen = lustre_msg_size(1, &size); + ptlrpc_request_set_replen(req); + req->rq_send_state = LUSTRE_IMP_FULL; - req->rq_timeout = 2; /* 2 second timeout for initial AST reply */ + /* ptlrpc_request_alloc_pack already set timeout */ + if (AT_OFF) + req->rq_timeout = ldlm_get_rq_timeout(); + + if (lock->l_export && lock->l_export->exp_nid_stats && + lock->l_export->exp_nid_stats->nid_ldlm_stats) + lprocfs_counter_incr(lock->l_export->exp_nid_stats->nid_ldlm_stats, + LDLM_GL_CALLBACK - LDLM_FIRST_OPC); rc = ptlrpc_queue_wait(req); if (rc == -ELDLM_NO_LOCK_DATA) @@ -571,97 +966,172 @@ int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data) else if (rc != 0) rc = ldlm_handle_ast_error(lock, req, rc, "glimpse"); else - rc = res->lr_namespace->ns_lvbo->lvbo_update - (res, req->rq_repmsg, 0, 1); + rc = ldlm_res_lvbo_update(res, req, 1); + ptlrpc_req_finished(req); + if (rc == -ERESTART) + ldlm_reprocess_all(res); + RETURN(rc); } -static struct ldlm_lock * -find_existing_lock(struct obd_export *exp, struct lustre_handle *remote_hdl) +#ifdef __KERNEL__ +extern unsigned long long lu_time_stamp_get(void); +#else +#define lu_time_stamp_get() time(NULL) +#endif + +static void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req, + struct lprocfs_stats *srv_stats) { - struct list_head *iter; + int lock_type = 0, op = 0; - spin_lock(&exp->exp_ldlm_data.led_lock); - list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) { - struct ldlm_lock *lock; - lock = list_entry(iter, struct ldlm_lock, l_export_chain); - if (lock->l_remote_handle.cookie == remote_hdl->cookie) { - LDLM_LOCK_GET(lock); - spin_unlock(&exp->exp_ldlm_data.led_lock); - return lock; - } + lock_type = dlm_req->lock_desc.l_resource.lr_type; + + switch (lock_type) { + case LDLM_PLAIN: + op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE; + break; + case LDLM_EXTENT: + if (dlm_req->lock_flags & LDLM_FL_HAS_INTENT) + op = PTLRPC_LAST_CNTR + LDLM_GLIMPSE_ENQUEUE; + else + op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE; + break; + case LDLM_FLOCK: + op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE; + break; + case LDLM_IBITS: + op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE; + break; + default: + op = 0; + break; } - spin_unlock(&exp->exp_ldlm_data.led_lock); - return NULL; + + if (op) + lprocfs_counter_incr(srv_stats, op); + + return; } -int ldlm_handle_enqueue(struct ptlrpc_request *req, - ldlm_completion_callback completion_callback, - ldlm_blocking_callback blocking_callback, - ldlm_glimpse_callback glimpse_callback) +/* + * Main server-side entry point into LDLM. This is called by ptlrpc service + * threads to carry out client lock enqueueing requests. + */ +int ldlm_handle_enqueue0(struct ldlm_namespace *ns, + struct ptlrpc_request *req, + const struct ldlm_request *dlm_req, + const struct ldlm_callback_suite *cbs) { - struct obd_device *obddev = req->rq_export->exp_obd; struct ldlm_reply *dlm_rep; - struct ldlm_request *dlm_req; - int rc = 0, size[2] = {sizeof(*dlm_rep)}; __u32 flags; ldlm_error_t err = ELDLM_OK; struct ldlm_lock *lock = NULL; void *cookie = NULL; + int rc = 0; ENTRY; LDLM_DEBUG_NOLOCK("server-side enqueue handler START"); - dlm_req = lustre_swab_reqbuf(req, MDS_REQ_INTENT_LOCKREQ_OFF, - sizeof (*dlm_req), - lustre_swab_ldlm_request); - if (dlm_req == NULL) { - CERROR ("Can't unpack dlm_req\n"); + ldlm_request_cancel(req, dlm_req, LDLM_ENQUEUE_CANCEL_OFF); + flags = dlm_req->lock_flags; + + LASSERT(req->rq_export); + + if (req->rq_rqbd->rqbd_service->srv_stats) + ldlm_svc_get_eopc(dlm_req, + req->rq_rqbd->rqbd_service->srv_stats); + + if (req->rq_export && req->rq_export->exp_nid_stats && + req->rq_export->exp_nid_stats->nid_ldlm_stats) + lprocfs_counter_incr(req->rq_export->exp_nid_stats->nid_ldlm_stats, + LDLM_ENQUEUE - LDLM_FIRST_OPC); + + if (unlikely(dlm_req->lock_desc.l_resource.lr_type < LDLM_MIN_TYPE || + dlm_req->lock_desc.l_resource.lr_type >= LDLM_MAX_TYPE)) { + DEBUG_REQ(D_ERROR, req, "invalid lock request type %d", + dlm_req->lock_desc.l_resource.lr_type); GOTO(out, rc = -EFAULT); } - flags = dlm_req->lock_flags; + if (unlikely(dlm_req->lock_desc.l_req_mode <= LCK_MINMODE || + dlm_req->lock_desc.l_req_mode >= LCK_MAXMODE || + dlm_req->lock_desc.l_req_mode & + (dlm_req->lock_desc.l_req_mode-1))) { + DEBUG_REQ(D_ERROR, req, "invalid lock request mode %d", + dlm_req->lock_desc.l_req_mode); + GOTO(out, rc = -EFAULT); + } - LASSERT(req->rq_export); + if (req->rq_export->exp_connect_flags & OBD_CONNECT_IBITS) { + if (unlikely(dlm_req->lock_desc.l_resource.lr_type == + LDLM_PLAIN)) { + DEBUG_REQ(D_ERROR, req, + "PLAIN lock request from IBITS client?"); + GOTO(out, rc = -EPROTO); + } + } else if (unlikely(dlm_req->lock_desc.l_resource.lr_type == + LDLM_IBITS)) { + DEBUG_REQ(D_ERROR, req, + "IBITS lock request from unaware client?"); + GOTO(out, rc = -EPROTO); + } - if (flags & LDLM_FL_REPLAY) { - lock = find_existing_lock(req->rq_export, - &dlm_req->lock_handle1); +#if 0 + /* FIXME this makes it impossible to use LDLM_PLAIN locks -- check + against server's _CONNECT_SUPPORTED flags? (I don't want to use + ibits for mgc/mgs) */ + + /* INODEBITS_INTEROP: Perform conversion from plain lock to + * inodebits lock if client does not support them. */ + if (!(req->rq_export->exp_connect_flags & OBD_CONNECT_IBITS) && + (dlm_req->lock_desc.l_resource.lr_type == LDLM_PLAIN)) { + dlm_req->lock_desc.l_resource.lr_type = LDLM_IBITS; + dlm_req->lock_desc.l_policy_data.l_inodebits.bits = + MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE; + if (dlm_req->lock_desc.l_req_mode == LCK_PR) + dlm_req->lock_desc.l_req_mode = LCK_CR; + } +#endif + + if (unlikely(flags & LDLM_FL_REPLAY)) { + /* Find an existing lock in the per-export lock hash */ + lock = cfs_hash_lookup(req->rq_export->exp_lock_hash, + (void *)&dlm_req->lock_handle[0]); if (lock != NULL) { - DEBUG_REQ(D_HA, req, "found existing lock cookie "LPX64, - lock->l_handle.h_cookie); + DEBUG_REQ(D_DLMTRACE, req, "found existing lock cookie " + LPX64, lock->l_handle.h_cookie); GOTO(existing_lock, rc = 0); } } /* The lock's callback data might be set in the policy function */ - lock = ldlm_lock_create(obddev->obd_namespace, &dlm_req->lock_handle2, - dlm_req->lock_desc.l_resource.lr_name, + lock = ldlm_lock_create(ns, &dlm_req->lock_desc.l_resource.lr_name, dlm_req->lock_desc.l_resource.lr_type, dlm_req->lock_desc.l_req_mode, - blocking_callback, completion_callback, - glimpse_callback, NULL, 0); + cbs, NULL, 0); + if (!lock) GOTO(out, rc = -ENOMEM); - do_gettimeofday(&lock->l_enqueued_time); - memcpy(&lock->l_remote_handle, &dlm_req->lock_handle1, - sizeof(lock->l_remote_handle)); + lock->l_last_activity = cfs_time_current_sec(); + lock->l_remote_handle = dlm_req->lock_handle[0]; LDLM_DEBUG(lock, "server-side enqueue handler, new lock created"); - LASSERT(req->rq_export); OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_BLOCKED, obd_timeout * 2); + /* Don't enqueue a lock onto the export if it has already + * been evicted. Cancel it now instead. (bug 3822) */ if (req->rq_export->exp_failed) { - LDLM_ERROR(lock,"lock on destroyed export %p\n",req->rq_export); - GOTO(out, err = -ENOTCONN); + LDLM_ERROR(lock, "lock on destroyed export %p", req->rq_export); + GOTO(out, rc = -ENOTCONN); } - lock->l_export = class_export_get(req->rq_export); - spin_lock(&lock->l_export->exp_ldlm_data.led_lock); - list_add(&lock->l_export_chain, - &lock->l_export->exp_ldlm_data.led_held_locks); - spin_unlock(&lock->l_export->exp_ldlm_data.led_lock); + lock->l_export = class_export_lock_get(req->rq_export, lock); + if (lock->l_export->exp_lock_hash) + cfs_hash_add(lock->l_export->exp_lock_hash, + &lock->l_remote_handle, + &lock->l_exp_hash); existing_lock: @@ -670,34 +1140,32 @@ existing_lock: * local_lock_enqueue by the policy function. */ cookie = req; } else { - int buffers = 1; lock_res_and_lock(lock); if (lock->l_resource->lr_lvb_len) { - size[1] = lock->l_resource->lr_lvb_len; - buffers = 2; + req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, + RCL_SERVER, + lock->l_resource->lr_lvb_len); } unlock_res_and_lock(lock); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR)) + + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR)) GOTO(out, rc = -ENOMEM); - rc = lustre_pack_reply(req, buffers, size, NULL); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) GOTO(out, rc); } if (dlm_req->lock_desc.l_resource.lr_type != LDLM_PLAIN) - memcpy(&lock->l_policy_data, &dlm_req->lock_desc.l_policy_data, - sizeof(ldlm_policy_data_t)); + lock->l_policy_data = dlm_req->lock_desc.l_policy_data; if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT) - memcpy(&lock->l_req_extent, &lock->l_policy_data.l_extent, - sizeof(lock->l_req_extent)); + lock->l_req_extent = lock->l_policy_data.l_extent; - err = ldlm_lock_enqueue(obddev->obd_namespace, &lock, - cookie, (int *)&flags); + err = ldlm_lock_enqueue(ns, &lock, cookie, (int *)&flags); if (err) GOTO(out, err); - dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*dlm_rep)); + dlm_rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP); dlm_rep->lock_flags = flags; ldlm_lock2desc(lock, &dlm_rep->lock_desc); @@ -706,21 +1174,70 @@ existing_lock: /* We never send a blocking AST until the lock is granted, but * we can tell it right now */ lock_res_and_lock(lock); - if (lock->l_flags & LDLM_FL_AST_SENT) { + + /* Now take into account flags to be inherited from original lock + request both in reply to client and in our own lock flags. */ + dlm_rep->lock_flags |= dlm_req->lock_flags & LDLM_INHERIT_FLAGS; + lock->l_flags |= dlm_req->lock_flags & LDLM_INHERIT_FLAGS; + + /* Don't move a pending lock onto the export if it has already + * been evicted. Cancel it now instead. (bug 5683) */ + if (unlikely(req->rq_export->exp_failed || + OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT))) { + LDLM_ERROR(lock, "lock on destroyed export %p", req->rq_export); + rc = -ENOTCONN; + } else if (lock->l_flags & LDLM_FL_AST_SENT) { dlm_rep->lock_flags |= LDLM_FL_AST_SENT; - if (lock->l_granted_mode == lock->l_req_mode) - ldlm_add_waiting_lock(lock); + if (lock->l_granted_mode == lock->l_req_mode) { + /* + * Only cancel lock if it was granted, because it would + * be destroyed immediately and would never be granted + * in the future, causing timeouts on client. Not + * granted lock will be cancelled immediately after + * sending completion AST. + */ + if (dlm_rep->lock_flags & LDLM_FL_CANCEL_ON_BLOCK) { + unlock_res_and_lock(lock); + ldlm_lock_cancel(lock); + lock_res_and_lock(lock); + } else + ldlm_add_waiting_lock(lock); + } + } + /* Make sure we never ever grant usual metadata locks to liblustre + clients */ + if ((dlm_req->lock_desc.l_resource.lr_type == LDLM_PLAIN || + dlm_req->lock_desc.l_resource.lr_type == LDLM_IBITS) && + req->rq_export->exp_libclient) { + if (unlikely(!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK) || + !(dlm_rep->lock_flags & LDLM_FL_CANCEL_ON_BLOCK))){ + CERROR("Granting sync lock to libclient. " + "req fl %d, rep fl %d, lock fl "LPX64"\n", + dlm_req->lock_flags, dlm_rep->lock_flags, + lock->l_flags); + LDLM_ERROR(lock, "sync lock"); + if (dlm_req->lock_flags & LDLM_FL_HAS_INTENT) { + struct ldlm_intent *it; + + it = req_capsule_client_get(&req->rq_pill, + &RMF_LDLM_INTENT); + if (it != NULL) { + CERROR("This is intent %s ("LPU64")\n", + ldlm_it2str(it->opc), it->opc); + } + } + } } + unlock_res_and_lock(lock); EXIT; out: - req->rq_status = err; - if (req->rq_reply_state == NULL) { - err = lustre_pack_reply(req, 0, NULL, NULL); + req->rq_status = rc ?: err; /* return either error - bug 11190 */ + if (!req->rq_packed_final) { + err = lustre_pack_reply(req, 1, NULL, NULL); if (rc == 0) rc = err; - req->rq_status = rc; } /* The LOCK_CHANGED code in ldlm_lock_enqueue depends on this @@ -729,57 +1246,81 @@ existing_lock: LDLM_DEBUG(lock, "server-side enqueue handler, sending reply" "(err=%d, rc=%d)", err, rc); + lock_res_and_lock(lock); if (rc == 0) { - lock_res_and_lock(lock); - size[1] = lock->l_resource->lr_lvb_len; - if (size[1] > 0) { - void *lvb = lustre_msg_buf(req->rq_repmsg, - 1, size[1]); + if (lock->l_resource->lr_lvb_len > 0) { + void *lvb; + + lvb = req_capsule_server_get(&req->rq_pill, + &RMF_DLM_LVB); LASSERTF(lvb != NULL, "req %p, lock %p\n", req, lock); memcpy(lvb, lock->l_resource->lr_lvb_data, - size[1]); + lock->l_resource->lr_lvb_len); } - unlock_res_and_lock(lock); } else { - ldlm_lock_destroy(lock); + ldlm_resource_unlink_lock(lock); + ldlm_lock_destroy_nolock(lock); } + unlock_res_and_lock(lock); if (!err && dlm_req->lock_desc.l_resource.lr_type != LDLM_FLOCK) ldlm_reprocess_all(lock->l_resource); - LDLM_LOCK_PUT(lock); + + LDLM_LOCK_RELEASE(lock); } + LDLM_DEBUG_NOLOCK("server-side enqueue handler END (lock %p, rc %d)", lock, rc); return rc; } -int ldlm_handle_convert(struct ptlrpc_request *req) +int ldlm_handle_enqueue(struct ptlrpc_request *req, + ldlm_completion_callback completion_callback, + ldlm_blocking_callback blocking_callback, + ldlm_glimpse_callback glimpse_callback) { struct ldlm_request *dlm_req; + struct ldlm_callback_suite cbs = { + .lcs_completion = completion_callback, + .lcs_blocking = blocking_callback, + .lcs_glimpse = glimpse_callback + }; + int rc; + + dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); + if (dlm_req != NULL) { + rc = ldlm_handle_enqueue0(req->rq_export->exp_obd->obd_namespace, + req, dlm_req, &cbs); + } else { + rc = -EFAULT; + } + return rc; +} + +int ldlm_handle_convert0(struct ptlrpc_request *req, + const struct ldlm_request *dlm_req) +{ struct ldlm_reply *dlm_rep; struct ldlm_lock *lock; - int rc, size = sizeof(*dlm_rep); + int rc; ENTRY; - dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req), - lustre_swab_ldlm_request); - if (dlm_req == NULL) { - CERROR ("Can't unpack dlm_req\n"); - RETURN (-EFAULT); - } + if (req->rq_export && req->rq_export->exp_nid_stats && + req->rq_export->exp_nid_stats->nid_ldlm_stats) + lprocfs_counter_incr(req->rq_export->exp_nid_stats->nid_ldlm_stats, + LDLM_CONVERT - LDLM_FIRST_OPC); - rc = lustre_pack_reply(req, 1, &size, NULL); - if (rc) { - CERROR("out of memory\n"); - RETURN(-ENOMEM); - } - dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep)); + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) + RETURN(rc); + + dlm_rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP); dlm_rep->lock_flags = dlm_req->lock_flags; - lock = ldlm_handle2lock(&dlm_req->lock_handle1); + lock = ldlm_handle2lock(&dlm_req->lock_handle[0]); if (!lock) { req->rq_status = EINVAL; } else { @@ -787,9 +1328,9 @@ int ldlm_handle_convert(struct ptlrpc_request *req) LDLM_DEBUG(lock, "server-side convert handler START"); - do_gettimeofday(&lock->l_enqueued_time); + lock->l_last_activity = cfs_time_current_sec(); res = ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode, - (int *)&dlm_rep->lock_flags); + &dlm_rep->lock_flags); if (res) { if (ldlm_del_waiting_lock(lock)) LDLM_DEBUG(lock, "converted waiting lock"); @@ -810,63 +1351,106 @@ int ldlm_handle_convert(struct ptlrpc_request *req) RETURN(0); } -int ldlm_handle_cancel(struct ptlrpc_request *req) +int ldlm_handle_convert(struct ptlrpc_request *req) { + int rc; struct ldlm_request *dlm_req; + + dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); + if (dlm_req != NULL) { + rc = ldlm_handle_convert0(req, dlm_req); + } else { + CERROR ("Can't unpack dlm_req\n"); + rc = -EFAULT; + } + return rc; +} + +/* Cancel all the locks whos handles are packed into ldlm_request */ +int ldlm_request_cancel(struct ptlrpc_request *req, + const struct ldlm_request *dlm_req, int first) +{ + struct ldlm_resource *res, *pres = NULL; struct ldlm_lock *lock; - struct ldlm_resource *res; + int i, count, done = 0; + ENTRY; + + count = dlm_req->lock_count ? dlm_req->lock_count : 1; + if (first >= count) + RETURN(0); + + /* There is no lock on the server at the replay time, + * skip lock cancelling to make replay tests to pass. */ + if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) + RETURN(0); + + LDLM_DEBUG_NOLOCK("server-side cancel handler START: %d locks, " + "starting at %d", count, first); + + for (i = first; i < count; i++) { + lock = ldlm_handle2lock(&dlm_req->lock_handle[i]); + if (!lock) { + LDLM_DEBUG_NOLOCK("server-side cancel handler stale " + "lock (cookie "LPU64")", + dlm_req->lock_handle[i].cookie); + continue; + } + + res = lock->l_resource; + done++; + + if (res != pres) { + if (pres != NULL) { + ldlm_reprocess_all(pres); + LDLM_RESOURCE_DELREF(pres); + ldlm_resource_putref(pres); + } + if (res != NULL) { + ldlm_resource_getref(res); + LDLM_RESOURCE_ADDREF(res); + ldlm_res_lvbo_update(res, NULL, 1); + } + pres = res; + } + ldlm_lock_cancel(lock); + LDLM_LOCK_PUT(lock); + } + if (pres != NULL) { + ldlm_reprocess_all(pres); + LDLM_RESOURCE_DELREF(pres); + ldlm_resource_putref(pres); + } + LDLM_DEBUG_NOLOCK("server-side cancel handler END"); + RETURN(done); +} + +int ldlm_handle_cancel(struct ptlrpc_request *req) +{ + struct ldlm_request *dlm_req; int rc; ENTRY; - dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req), - lustre_swab_ldlm_request); + dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); if (dlm_req == NULL) { - CERROR("bad request buffer for cancel\n"); + CDEBUG(D_INFO, "bad request buffer for cancel\n"); RETURN(-EFAULT); } - rc = lustre_pack_reply(req, 0, NULL, NULL); - if (rc) { - CERROR("out of memory\n"); - RETURN(-ENOMEM); - } + if (req->rq_export && req->rq_export->exp_nid_stats && + req->rq_export->exp_nid_stats->nid_ldlm_stats) + lprocfs_counter_incr(req->rq_export->exp_nid_stats->nid_ldlm_stats, + LDLM_CANCEL - LDLM_FIRST_OPC); - lock = ldlm_handle2lock(&dlm_req->lock_handle1); - if (!lock) { - CERROR("received cancel for unknown lock cookie "LPX64 - " from client %s id %s\n", - dlm_req->lock_handle1.cookie, - req->rq_export->exp_client_uuid.uuid, - req->rq_peerstr); - LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock " - "(cookie "LPU64")", - dlm_req->lock_handle1.cookie); - req->rq_status = ESTALE; - } else { - LDLM_DEBUG(lock, "server-side cancel handler START"); - res = lock->l_resource; - if (res && res->lr_namespace->ns_lvbo && - res->lr_namespace->ns_lvbo->lvbo_update) { - (void)res->lr_namespace->ns_lvbo->lvbo_update - (res, NULL, 0, 0); - //(res, req->rq_reqmsg, 1); - } + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) + RETURN(rc); - ldlm_lock_cancel(lock); - if (ldlm_del_waiting_lock(lock)) - CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock); - req->rq_status = rc; - } + if (!ldlm_request_cancel(req, dlm_req, 0)) + req->rq_status = ESTALE; if (ptlrpc_reply(req) != 0) LBUG(); - if (lock) { - ldlm_reprocess_all(lock->l_resource); - LDLM_DEBUG(lock, "server-side cancel handler END"); - LDLM_LOCK_PUT(lock); - } - RETURN(0); } @@ -876,26 +1460,30 @@ void ldlm_handle_bl_callback(struct ldlm_namespace *ns, int do_ast; ENTRY; - LDLM_DEBUG(lock, "client blocking AST callback handler START"); - + LDLM_DEBUG(lock, "client blocking AST callback handler"); + lock_res_and_lock(lock); lock->l_flags |= LDLM_FL_CBPENDING; + + if (lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK) + lock->l_flags |= LDLM_FL_CANCEL; + do_ast = (!lock->l_readers && !lock->l_writers); unlock_res_and_lock(lock); if (do_ast) { - LDLM_DEBUG(lock, "already unused, calling " - "callback (%p)", lock->l_blocking_ast); + CDEBUG(D_DLMTRACE, "Lock %p already unused, calling callback (%p)\n", + lock, lock->l_blocking_ast); if (lock->l_blocking_ast != NULL) lock->l_blocking_ast(lock, ld, lock->l_ast_data, LDLM_CB_BLOCKING); } else { - LDLM_DEBUG(lock, "Lock still has references, will be" - " cancelled later"); + CDEBUG(D_DLMTRACE, "Lock %p is referenced, will be cancelled later\n", + lock); } LDLM_DEBUG(lock, "client blocking callback handler END"); - LDLM_LOCK_PUT(lock); + LDLM_LOCK_RELEASE(lock); EXIT; } @@ -904,12 +1492,32 @@ static void ldlm_handle_cp_callback(struct ptlrpc_request *req, struct ldlm_request *dlm_req, struct ldlm_lock *lock) { - LIST_HEAD(ast_list); + CFS_LIST_HEAD(ast_list); ENTRY; LDLM_DEBUG(lock, "client completion callback handler START"); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) { + int to = cfs_time_seconds(1); + while (to > 0) { + cfs_schedule_timeout_and_set_state( + CFS_TASK_INTERRUPTIBLE, to); + if (lock->l_granted_mode == lock->l_req_mode || + lock->l_destroyed) + break; + } + } + lock_res_and_lock(lock); + if (lock->l_destroyed || + lock->l_granted_mode == lock->l_req_mode) { + /* bug 11300: the lock has already been granted */ + unlock_res_and_lock(lock); + LDLM_DEBUG(lock, "Double grant race happened"); + LDLM_LOCK_RELEASE(lock); + EXIT; + return; + } /* If we receive the completion AST before the actual enqueue returned, * then we might need to switch lock modes, resources, or extents. */ @@ -919,8 +1527,7 @@ static void ldlm_handle_cp_callback(struct ptlrpc_request *req, } if (lock->l_resource->lr_type != LDLM_PLAIN) { - memcpy(&lock->l_policy_data, &dlm_req->lock_desc.l_policy_data, - sizeof(lock->l_policy_data)); + lock->l_policy_data = dlm_req->lock_desc.l_policy_data; LDLM_DEBUG(lock, "completion AST, new policy data"); } @@ -929,26 +1536,34 @@ static void ldlm_handle_cp_callback(struct ptlrpc_request *req, &lock->l_resource->lr_name, sizeof(lock->l_resource->lr_name)) != 0) { unlock_res_and_lock(lock); - ldlm_lock_change_resource(ns, lock, - dlm_req->lock_desc.l_resource.lr_name); + if (ldlm_lock_change_resource(ns, lock, + &dlm_req->lock_desc.l_resource.lr_name) != 0) { + LDLM_ERROR(lock, "Failed to allocate resource"); + LDLM_LOCK_RELEASE(lock); + EXIT; + return; + } LDLM_DEBUG(lock, "completion AST, new resource"); CERROR("change resource!\n"); lock_res_and_lock(lock); } if (dlm_req->lock_flags & LDLM_FL_AST_SENT) { - lock->l_flags |= LDLM_FL_CBPENDING; + /* BL_AST locks are not needed in lru. + * let ldlm_cancel_lru() be fast. */ + ldlm_lock_remove_from_lru(lock); + lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST; LDLM_DEBUG(lock, "completion AST includes blocking AST"); } if (lock->l_lvb_len) { - void *lvb; - lvb = lustre_swab_reqbuf(req, 1, lock->l_lvb_len, - lock->l_lvb_swabber); - if (lvb == NULL) { + if (req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB, + RCL_CLIENT) < lock->l_lvb_len) { LDLM_ERROR(lock, "completion AST did not contain " "expected LVB!"); } else { + void *lvb = req_capsule_client_get(&req->rq_pill, + &RMF_DLM_LVB); memcpy(lock->l_lvb_data, lvb, lock->l_lvb_len); } } @@ -957,12 +1572,12 @@ static void ldlm_handle_cp_callback(struct ptlrpc_request *req, unlock_res_and_lock(lock); LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work"); - LDLM_LOCK_PUT(lock); - ldlm_run_cp_ast_work(&ast_list); + ldlm_run_ast_work(&ast_list, LDLM_WORK_CP_AST); LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)", lock); + LDLM_LOCK_RELEASE(lock); EXIT; } @@ -989,96 +1604,191 @@ static void ldlm_handle_gl_callback(struct ptlrpc_request *req, lock_res_and_lock(lock); if (lock->l_granted_mode == LCK_PW && !lock->l_readers && !lock->l_writers && - time_after(jiffies, lock->l_last_used + 10 * HZ)) { + cfs_time_after(cfs_time_current(), + cfs_time_add(lock->l_last_used, + cfs_time_seconds(10)))) { unlock_res_and_lock(lock); - if (ldlm_bl_to_thread(ns, NULL, lock)) + if (ldlm_bl_to_thread_lock(ns, NULL, lock)) ldlm_handle_bl_callback(ns, NULL, lock); EXIT; return; } unlock_res_and_lock(lock); - LDLM_LOCK_PUT(lock); + LDLM_LOCK_RELEASE(lock); EXIT; } static int ldlm_callback_reply(struct ptlrpc_request *req, int rc) { + if (req->rq_no_reply) + return 0; + req->rq_status = rc; - if (req->rq_reply_state == NULL) { - rc = lustre_pack_reply(req, 0, NULL, NULL); + if (!req->rq_packed_final) { + rc = lustre_pack_reply(req, 1, NULL, NULL); if (rc) return rc; } return ptlrpc_reply(req); } -int ldlm_bl_to_thread(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld, - struct ldlm_lock *lock) -{ #ifdef __KERNEL__ +static int __ldlm_bl_to_thread(struct ldlm_namespace *ns, struct ldlm_bl_work_item *blwi, + struct ldlm_lock_desc *ld, struct ldlm_lock *lock, + cfs_list_t *cancels, int count, int mode) +{ struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool; - struct ldlm_bl_work_item *blwi; ENTRY; - OBD_ALLOC(blwi, sizeof(*blwi)); - if (blwi == NULL) - RETURN(-ENOMEM); + if (cancels && count == 0) { + if (mode == LDLM_ASYNC) + OBD_FREE(blwi, sizeof(*blwi)); + RETURN(0); + } + + cfs_init_completion(&blwi->blwi_comp); + cfs_atomic_set(&blwi->blwi_ref_count, 1); blwi->blwi_ns = ns; if (ld != NULL) blwi->blwi_ld = *ld; - blwi->blwi_lock = lock; + if (count) { + cfs_list_add(&blwi->blwi_head, cancels); + cfs_list_del_init(cancels); + blwi->blwi_count = count; + } else { + blwi->blwi_lock = lock; + } + + cfs_spin_lock(&blp->blp_lock); + if (lock && lock->l_flags & LDLM_FL_DISCARD_DATA) { + /* add LDLM_FL_DISCARD_DATA requests to the priority list */ + cfs_list_add_tail(&blwi->blwi_entry, &blp->blp_prio_list); + } else { + /* other blocking callbacks are added to the regular list */ + cfs_list_add_tail(&blwi->blwi_entry, &blp->blp_list); + } + cfs_spin_unlock(&blp->blp_lock); - spin_lock(&blp->blp_lock); - list_add_tail(&blwi->blwi_entry, &blp->blp_list); - wake_up(&blp->blp_waitq); - spin_unlock(&blp->blp_lock); + if (mode == LDLM_SYNC) { + /* keep ref count as object is on this stack for SYNC call */ + ldlm_bl_work_item_get(blwi); + cfs_waitq_signal(&blp->blp_waitq); + cfs_wait_for_completion(&blwi->blwi_comp); + } else { + cfs_waitq_signal(&blp->blp_waitq); + } RETURN(0); +} + +static int ldlm_bl_to_thread(struct ldlm_namespace *ns, + struct ldlm_lock_desc *ld, struct ldlm_lock *lock, + struct list_head *cancels, int count, int mode) +{ + ENTRY; + + if (mode == LDLM_SYNC) { + /* if it is synchronous call do minimum mem alloc, as it could + * be triggered from kernel shrinker + */ + struct ldlm_bl_work_item blwi; + memset(&blwi, 0, sizeof(blwi)); + /* have extra ref as this obj is on stack */ + RETURN(__ldlm_bl_to_thread(ns, &blwi, ld, lock, cancels, count, mode)); + } else { + struct ldlm_bl_work_item *blwi; + OBD_ALLOC(blwi, sizeof(*blwi)); + if (blwi == NULL) + RETURN(-ENOMEM); + + RETURN(__ldlm_bl_to_thread(ns, blwi, ld, lock, cancels, count, mode)); + } +} + +#endif + +int ldlm_bl_to_thread_lock(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld, + struct ldlm_lock *lock) +{ +#ifdef __KERNEL__ + RETURN(ldlm_bl_to_thread(ns, ld, lock, NULL, 0, LDLM_ASYNC)); #else RETURN(-ENOSYS); #endif +} +int ldlm_bl_to_thread_list(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld, + cfs_list_t *cancels, int count, int mode) +{ +#ifdef __KERNEL__ + RETURN(ldlm_bl_to_thread(ns, ld, NULL, cancels, count, mode)); +#else + RETURN(-ENOSYS); +#endif } -static int ldlm_msg_check_version(struct lustre_msg *msg) +/* Setinfo coming from Server (eg MDT) to Client (eg MDC)! */ +static int ldlm_handle_setinfo(struct ptlrpc_request *req) { - int rc; + struct obd_device *obd = req->rq_export->exp_obd; + char *key; + void *val; + int keylen, vallen; + int rc = -ENOSYS; + ENTRY; - switch (msg->opc) { - case LDLM_ENQUEUE: - case LDLM_CONVERT: - case LDLM_CANCEL: - case LDLM_BL_CALLBACK: - case LDLM_CP_CALLBACK: - case LDLM_GL_CALLBACK: - rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION); - if (rc) - CERROR("bad opc %u version %08x, expecting %08x\n", - msg->opc, msg->version, LUSTRE_DLM_VERSION); - break; - case OBD_LOG_CANCEL: - case LLOG_ORIGIN_HANDLE_OPEN: - case LLOG_ORIGIN_HANDLE_NEXT_BLOCK: - case LLOG_ORIGIN_HANDLE_PREV_BLOCK: - case LLOG_ORIGIN_HANDLE_READ_HEADER: - case LLOG_ORIGIN_HANDLE_CLOSE: - case LLOG_CATINFO: - rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION); - if (rc) - CERROR("bad opc %u version %08x, expecting %08x\n", - msg->opc, msg->version, LUSTRE_LOG_VERSION); - break; - default: - CERROR("LDLM unknown opcode %d\n", msg->opc); - rc = -ENOTSUPP; - break; + DEBUG_REQ(D_HSM, req, "%s: handle setinfo\n", obd->obd_name); + + req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO); + + key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY); + if (key == NULL) { + DEBUG_REQ(D_IOCTL, req, "no set_info key"); + RETURN(-EFAULT); } + keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY, + RCL_CLIENT); + val = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL); + if (val == NULL) { + DEBUG_REQ(D_IOCTL, req, "no set_info val"); + RETURN(-EFAULT); + } + vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL, + RCL_CLIENT); + + /* We are responsible for swabbing contents of val */ + + if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) + /* Pass it on to mdc (the "export" in this case) */ + rc = obd_set_info_async(req->rq_export, + sizeof(KEY_HSM_COPYTOOL_SEND), + KEY_HSM_COPYTOOL_SEND, + vallen, val, NULL); + else + DEBUG_REQ(D_WARNING, req, "ignoring unknown key %s", key); return rc; } +static inline void ldlm_callback_errmsg(struct ptlrpc_request *req, + const char *msg, int rc, + struct lustre_handle *handle) +{ + CWARN("%s: [pid %d] [xid x"LPU64"] [nid %s] [opc %d] [rc %d] " + "[lock "LPX64"].\n", + msg, lustre_msg_get_status(req->rq_reqmsg), + req->rq_xid, libcfs_id2str(req->rq_peer), + lustre_msg_get_opc(req->rq_reqmsg), rc, + handle ? handle->cookie : 0); + if (req->rq_no_reply) + CWARN("No reply was sent, maybe cause bug 21636.\n"); + else if (rc) + CWARN("Send reply failed, maybe cause bug 21636.\n"); +} + +/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */ static int ldlm_callback_handler(struct ptlrpc_request *req) { struct ldlm_namespace *ns; @@ -1087,81 +1797,99 @@ static int ldlm_callback_handler(struct ptlrpc_request *req) int rc; ENTRY; - rc = ldlm_msg_check_version(req->rq_reqmsg); - if (rc) { - CERROR("LDLM_CB drop mal-formed request\n"); - RETURN(rc); - } - /* Requests arrive in sender's byte order. The ptlrpc service * handler has already checked and, if necessary, byte-swapped the * incoming request message body, but I am responsible for the * message buffers. */ - if (req->rq_export == NULL) { - struct ldlm_request *dlm_req; + /* do nothing for sec context finalize */ + if (lustre_msg_get_opc(req->rq_reqmsg) == SEC_CTX_FINI) + RETURN(0); - CDEBUG(D_RPCTRACE, "operation %d from %s with bad " - "export cookie "LPX64"; this is " - "normal if this node rebooted with a lock held\n", - req->rq_reqmsg->opc, - req->rq_peerstr, - req->rq_reqmsg->handle.cookie); - dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req), - lustre_swab_ldlm_request); - if (dlm_req != NULL) - CDEBUG(D_RPCTRACE, "--> lock cookie: "LPX64"\n", - dlm_req->lock_handle1.cookie); + req_capsule_init(&req->rq_pill, req, RCL_SERVER); - ldlm_callback_reply(req, -ENOTCONN); + if (req->rq_export == NULL) { + rc = ldlm_callback_reply(req, -ENOTCONN); + ldlm_callback_errmsg(req, "Operate on unconnected server", + rc, NULL); RETURN(0); } LASSERT(req->rq_export != NULL); LASSERT(req->rq_export->exp_obd != NULL); - switch(req->rq_reqmsg->opc) { + switch (lustre_msg_get_opc(req->rq_reqmsg)) { case LDLM_BL_CALLBACK: - OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK)) + RETURN(0); break; case LDLM_CP_CALLBACK: - OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CP_CALLBACK)) + RETURN(0); break; case LDLM_GL_CALLBACK: - OBD_FAIL_RETURN(OBD_FAIL_LDLM_GL_CALLBACK, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_GL_CALLBACK)) + RETURN(0); break; - case OBD_LOG_CANCEL: - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0); + case LDLM_SET_INFO: + rc = ldlm_handle_setinfo(req); + ldlm_callback_reply(req, rc); + RETURN(0); + case OBD_LOG_CANCEL: /* remove this eventually - for 1.4.0 compat */ + CERROR("shouldn't be handling OBD_LOG_CANCEL on DLM thread\n"); + req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET)) + RETURN(0); rc = llog_origin_handle_cancel(req); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP)) + RETURN(0); ldlm_callback_reply(req, rc); RETURN(0); - case LLOG_ORIGIN_HANDLE_OPEN: - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - rc = llog_origin_handle_open(req); + case OBD_QC_CALLBACK: + req_capsule_set(&req->rq_pill, &RQF_QC_CALLBACK); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_QC_CALLBACK_NET)) + RETURN(0); + rc = target_handle_qc_callback(req); ldlm_callback_reply(req, rc); RETURN(0); - case LLOG_ORIGIN_HANDLE_NEXT_BLOCK: - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - rc = llog_origin_handle_next_block(req); + case QUOTA_DQACQ: + case QUOTA_DQREL: + /* reply in handler */ + req_capsule_set(&req->rq_pill, &RQF_MDS_QUOTA_DQACQ); + rc = target_handle_dqacq_callback(req); + RETURN(0); + case LLOG_ORIGIN_HANDLE_CREATE: + req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET)) + RETURN(0); + rc = llog_origin_handle_create(req); ldlm_callback_reply(req, rc); RETURN(0); - case LLOG_ORIGIN_HANDLE_PREV_BLOCK: - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - rc = llog_origin_handle_prev_block(req); + case LLOG_ORIGIN_HANDLE_NEXT_BLOCK: + req_capsule_set(&req->rq_pill, + &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET)) + RETURN(0); + rc = llog_origin_handle_next_block(req); ldlm_callback_reply(req, rc); RETURN(0); case LLOG_ORIGIN_HANDLE_READ_HEADER: - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); + req_capsule_set(&req->rq_pill, + &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET)) + RETURN(0); rc = llog_origin_handle_read_header(req); ldlm_callback_reply(req, rc); RETURN(0); case LLOG_ORIGIN_HANDLE_CLOSE: - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOGD_NET)) + RETURN(0); rc = llog_origin_handle_close(req); ldlm_callback_reply(req, rc); RETURN(0); default: - CERROR("unknown opcode %u\n", req->rq_reqmsg->opc); + CERROR("unknown opcode %u\n", + lustre_msg_get_opc(req->rq_reqmsg)); ldlm_callback_reply(req, -EPROTO); RETURN(0); } @@ -1169,25 +1897,65 @@ static int ldlm_callback_handler(struct ptlrpc_request *req) ns = req->rq_export->exp_obd->obd_namespace; LASSERT(ns != NULL); - dlm_req = lustre_swab_reqbuf (req, 0, sizeof (*dlm_req), - lustre_swab_ldlm_request); + req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK); + + dlm_req = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ); if (dlm_req == NULL) { - CERROR ("can't unpack dlm_req\n"); - ldlm_callback_reply (req, -EPROTO); - RETURN (0); + rc = ldlm_callback_reply(req, -EPROTO); + ldlm_callback_errmsg(req, "Operate without parameter", rc, + NULL); + RETURN(0); + } + + /* Force a known safe race, send a cancel to the server for a lock + * which the server has already started a blocking callback on. */ + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE) && + lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) { + rc = ldlm_cli_cancel(&dlm_req->lock_handle[0]); + if (rc < 0) + CERROR("ldlm_cli_cancel: %d\n", rc); } - lock = ldlm_handle2lock_ns(ns, &dlm_req->lock_handle1); + lock = ldlm_handle2lock_long(&dlm_req->lock_handle[0], 0); if (!lock) { - LDLM_DEBUG_NOLOCK("callback on lock "LPX64" - lock " - "disappeared\n",dlm_req->lock_handle1.cookie); - ldlm_callback_reply(req, -EINVAL); + CDEBUG(D_DLMTRACE, "callback on lock "LPX64" - lock " + "disappeared\n", dlm_req->lock_handle[0].cookie); + rc = ldlm_callback_reply(req, -EINVAL); + ldlm_callback_errmsg(req, "Operate with invalid parameter", rc, + &dlm_req->lock_handle[0]); RETURN(0); } + if ((lock->l_flags & LDLM_FL_FAIL_LOC) && + lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) + OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE); + /* Copy hints/flags (e.g. LDLM_FL_DISCARD_DATA) from AST. */ lock_res_and_lock(lock); lock->l_flags |= (dlm_req->lock_flags & LDLM_AST_FLAGS); + if (lustre_msg_get_opc(req->rq_reqmsg) == LDLM_BL_CALLBACK) { + /* If somebody cancels lock and cache is already dropped, + * or lock is failed before cp_ast received on client, + * we can tell the server we have no lock. Otherwise, we + * should send cancel after dropping the cache. */ + if (((lock->l_flags & LDLM_FL_CANCELING) && + (lock->l_flags & LDLM_FL_BL_DONE)) || + (lock->l_flags & LDLM_FL_FAILED)) { + LDLM_DEBUG(lock, "callback on lock " + LPX64" - lock disappeared\n", + dlm_req->lock_handle[0].cookie); + unlock_res_and_lock(lock); + LDLM_LOCK_RELEASE(lock); + rc = ldlm_callback_reply(req, -EINVAL); + ldlm_callback_errmsg(req, "Operate on stale lock", rc, + &dlm_req->lock_handle[0]); + RETURN(0); + } + /* BL_AST locks are not needed in lru. + * let ldlm_cancel_lru() be fast. */ + ldlm_lock_remove_from_lru(lock); + lock->l_flags |= LDLM_FL_BL_AST; + } unlock_res_and_lock(lock); /* We want the ost thread to get this reply so that it can respond @@ -1199,21 +1967,28 @@ static int ldlm_callback_handler(struct ptlrpc_request *req) * in the reply, so we might have to push the responsibility for sending * the reply down into the AST handlers, alas. */ - switch (req->rq_reqmsg->opc) { + switch (lustre_msg_get_opc(req->rq_reqmsg)) { case LDLM_BL_CALLBACK: CDEBUG(D_INODE, "blocking ast\n"); - ldlm_callback_reply(req, 0); - if (ldlm_bl_to_thread(ns, &dlm_req->lock_desc, lock)) + req_capsule_extend(&req->rq_pill, &RQF_LDLM_BL_CALLBACK); + if (!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)) { + rc = ldlm_callback_reply(req, 0); + if (req->rq_no_reply || rc) + ldlm_callback_errmsg(req, "Normal process", rc, + &dlm_req->lock_handle[0]); + } + if (ldlm_bl_to_thread_lock(ns, &dlm_req->lock_desc, lock)) ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock); - break; case LDLM_CP_CALLBACK: CDEBUG(D_INODE, "completion ast\n"); + req_capsule_extend(&req->rq_pill, &RQF_LDLM_CP_CALLBACK); ldlm_callback_reply(req, 0); ldlm_handle_cp_callback(req, ns, dlm_req, lock); break; case LDLM_GL_CALLBACK: CDEBUG(D_INODE, "glimpse ast\n"); + req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK); ldlm_handle_gl_callback(req, ns, dlm_req, lock); break; default: @@ -1225,139 +2000,274 @@ static int ldlm_callback_handler(struct ptlrpc_request *req) static int ldlm_cancel_handler(struct ptlrpc_request *req) { - int rc = 0; + int rc; ENTRY; - rc = ldlm_msg_check_version(req->rq_reqmsg); - if (rc) { - CERROR("LDLM_CL drop mal-formed request\n"); - RETURN(rc); - } - /* Requests arrive in sender's byte order. The ptlrpc service * handler has already checked and, if necessary, byte-swapped the * incoming request message body, but I am responsible for the * message buffers. */ + req_capsule_init(&req->rq_pill, req, RCL_SERVER); + if (req->rq_export == NULL) { struct ldlm_request *dlm_req; - CERROR("operation %d with bad export from %s\n", - req->rq_reqmsg->opc, - req->rq_peerstr); - CERROR("--> export cookie: "LPX64"\n", - req->rq_reqmsg->handle.cookie); - dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req), - lustre_swab_ldlm_request); - if (dlm_req != NULL) - ldlm_lock_dump_handle(D_ERROR, &dlm_req->lock_handle1); + + CERROR("operation %d from %s with bad export cookie "LPU64"\n", + lustre_msg_get_opc(req->rq_reqmsg), + libcfs_id2str(req->rq_peer), + lustre_msg_get_handle(req->rq_reqmsg)->cookie); + + if (lustre_msg_get_opc(req->rq_reqmsg) == LDLM_CANCEL) { + req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK); + dlm_req = req_capsule_client_get(&req->rq_pill, + &RMF_DLM_REQ); + if (dlm_req != NULL) + ldlm_lock_dump_handle(D_ERROR, + &dlm_req->lock_handle[0]); + } ldlm_callback_reply(req, -ENOTCONN); RETURN(0); } - switch (req->rq_reqmsg->opc) { + switch (lustre_msg_get_opc(req->rq_reqmsg)) { /* XXX FIXME move this back to mds/handler.c, bug 249 */ case LDLM_CANCEL: + req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL); CDEBUG(D_INODE, "cancel\n"); - OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL)) + RETURN(0); rc = ldlm_handle_cancel(req); - break; + if (rc) + break; + RETURN(0); + case OBD_LOG_CANCEL: + req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET)) + RETURN(0); + rc = llog_origin_handle_cancel(req); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP)) + RETURN(0); + ldlm_callback_reply(req, rc); + RETURN(0); default: - CERROR("invalid opcode %d\n", req->rq_reqmsg->opc); + CERROR("invalid opcode %d\n", + lustre_msg_get_opc(req->rq_reqmsg)); + req_capsule_set(&req->rq_pill, &RQF_LDLM_CALLBACK); ldlm_callback_reply(req, -EINVAL); } - RETURN(rc); + RETURN(0); +} + +void ldlm_revoke_lock_cb(void *obj, void *data) +{ + cfs_list_t *rpc_list = data; + struct ldlm_lock *lock = obj; + + lock_res_and_lock(lock); + + if (lock->l_req_mode != lock->l_granted_mode) { + unlock_res_and_lock(lock); + return; + } + + LASSERT(lock->l_resource); + if (lock->l_resource->lr_type != LDLM_IBITS && + lock->l_resource->lr_type != LDLM_PLAIN) { + unlock_res_and_lock(lock); + return; + } + + if (lock->l_flags & LDLM_FL_AST_SENT) { + unlock_res_and_lock(lock); + return; + } + + LASSERT(lock->l_blocking_ast); + LASSERT(!lock->l_blocking_lock); + + lock->l_flags |= LDLM_FL_AST_SENT; + if (lock->l_export && lock->l_export->exp_lock_hash && + !cfs_hlist_unhashed(&lock->l_exp_hash)) + cfs_hash_del(lock->l_export->exp_lock_hash, + &lock->l_remote_handle, &lock->l_exp_hash); + cfs_list_add_tail(&lock->l_rk_ast, rpc_list); + LDLM_LOCK_GET(lock); + + unlock_res_and_lock(lock); +} + +void ldlm_revoke_export_locks(struct obd_export *exp) +{ + cfs_list_t rpc_list; + ENTRY; + + CFS_INIT_LIST_HEAD(&rpc_list); + cfs_hash_for_each_empty(exp->exp_lock_hash, + ldlm_revoke_lock_cb, &rpc_list); + ldlm_run_ast_work(&rpc_list, LDLM_WORK_REVOKE_AST); + + EXIT; } #ifdef __KERNEL__ static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp) { struct ldlm_bl_work_item *blwi = NULL; - - spin_lock(&blp->blp_lock); - if (!list_empty(&blp->blp_list)) { - blwi = list_entry(blp->blp_list.next, struct ldlm_bl_work_item, - blwi_entry); - list_del(&blwi->blwi_entry); + static unsigned int num_bl = 0; + + cfs_spin_lock(&blp->blp_lock); + /* process a request from the blp_list at least every blp_num_threads */ + if (!cfs_list_empty(&blp->blp_list) && + (cfs_list_empty(&blp->blp_prio_list) || num_bl == 0)) + blwi = cfs_list_entry(blp->blp_list.next, + struct ldlm_bl_work_item, blwi_entry); + else + if (!cfs_list_empty(&blp->blp_prio_list)) + blwi = cfs_list_entry(blp->blp_prio_list.next, + struct ldlm_bl_work_item, + blwi_entry); + + if (blwi) { + if (++num_bl >= cfs_atomic_read(&blp->blp_num_threads)) + num_bl = 0; + cfs_list_del(&blwi->blwi_entry); } - spin_unlock(&blp->blp_lock); + cfs_spin_unlock(&blp->blp_lock); return blwi; } +/* This only contains temporary data until the thread starts */ struct ldlm_bl_thread_data { - int bltd_num; + char bltd_name[CFS_CURPROC_COMM_MAX]; struct ldlm_bl_pool *bltd_blp; + cfs_completion_t bltd_comp; + int bltd_num; }; +static int ldlm_bl_thread_main(void *arg); + +static int ldlm_bl_thread_start(struct ldlm_bl_pool *blp) +{ + struct ldlm_bl_thread_data bltd = { .bltd_blp = blp }; + int rc; + + cfs_init_completion(&bltd.bltd_comp); + rc = cfs_kernel_thread(ldlm_bl_thread_main, &bltd, 0); + if (rc < 0) { + CERROR("cannot start LDLM thread ldlm_bl_%02d: rc %d\n", + cfs_atomic_read(&blp->blp_num_threads), rc); + return rc; + } + cfs_wait_for_completion(&bltd.bltd_comp); + + return 0; +} + static int ldlm_bl_thread_main(void *arg) { - struct ldlm_bl_thread_data *bltd = arg; - struct ldlm_bl_pool *blp = bltd->bltd_blp; - unsigned long flags; + struct ldlm_bl_pool *blp; ENTRY; - /* XXX boiler-plate */ { - char name[sizeof(current->comm)]; - snprintf(name, sizeof(name) - 1, "ldlm_bl_%02d", - bltd->bltd_num); - kportal_daemonize(name); - } - SIGNAL_MASK_LOCK(current, flags); - sigfillset(¤t->blocked); - RECALC_SIGPENDING; - SIGNAL_MASK_UNLOCK(current, flags); + struct ldlm_bl_thread_data *bltd = arg; + + blp = bltd->bltd_blp; + + bltd->bltd_num = + cfs_atomic_inc_return(&blp->blp_num_threads) - 1; + cfs_atomic_inc(&blp->blp_busy_threads); + + snprintf(bltd->bltd_name, sizeof(bltd->bltd_name) - 1, + "ldlm_bl_%02d", bltd->bltd_num); + cfs_daemonize(bltd->bltd_name); - atomic_inc(&blp->blp_num_threads); - complete(&blp->blp_comp); + cfs_complete(&bltd->bltd_comp); + /* cannot use bltd after this, it is only on caller's stack */ + } - while(1) { + while (1) { struct l_wait_info lwi = { 0 }; struct ldlm_bl_work_item *blwi = NULL; - l_wait_event_exclusive(blp->blp_waitq, - (blwi = ldlm_bl_get_work(blp)) != NULL, - &lwi); + blwi = ldlm_bl_get_work(blp); - if (blwi->blwi_ns == NULL) - break; + if (blwi == NULL) { + int busy; - ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld, - blwi->blwi_lock); - OBD_FREE(blwi, sizeof(*blwi)); + cfs_atomic_dec(&blp->blp_busy_threads); + l_wait_event_exclusive(blp->blp_waitq, + (blwi = ldlm_bl_get_work(blp)) != NULL, + &lwi); + busy = cfs_atomic_inc_return(&blp->blp_busy_threads); + + if (blwi->blwi_ns == NULL) + /* added by ldlm_cleanup() */ + break; + + /* Not fatal if racy and have a few too many threads */ + if (unlikely(busy < blp->blp_max_threads && + busy >= cfs_atomic_read(&blp->blp_num_threads))) + /* discard the return value, we tried */ + ldlm_bl_thread_start(blp); + } else { + if (blwi->blwi_ns == NULL) + /* added by ldlm_cleanup() */ + break; + } + + if (blwi->blwi_count) { + /* The special case when we cancel locks in lru + * asynchronously, we pass the list of locks here. + * Thus locks are marked LDLM_FL_CANCELING, but NOT + * canceled locally yet. */ + ldlm_cli_cancel_list_local(&blwi->blwi_head, + blwi->blwi_count, 0); + ldlm_cli_cancel_list(&blwi->blwi_head, + blwi->blwi_count, NULL, 0); + } else { + ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld, + blwi->blwi_lock); + } + cfs_complete(&blwi->blwi_comp); + ldlm_bl_work_item_put(blwi); } - atomic_dec(&blp->blp_num_threads); - complete(&blp->blp_comp); + cfs_atomic_dec(&blp->blp_busy_threads); + cfs_atomic_dec(&blp->blp_num_threads); + cfs_complete(&blp->blp_comp); RETURN(0); } #endif static int ldlm_setup(void); -static int ldlm_cleanup(int force); +static int ldlm_cleanup(void); int ldlm_get_ref(void) { int rc = 0; - down(&ldlm_ref_sem); + ENTRY; + cfs_mutex_down(&ldlm_ref_sem); if (++ldlm_refcount == 1) { rc = ldlm_setup(); if (rc) ldlm_refcount--; } - up(&ldlm_ref_sem); + cfs_mutex_up(&ldlm_ref_sem); RETURN(rc); } -void ldlm_put_ref(int force) +void ldlm_put_ref(void) { - down(&ldlm_ref_sem); + ENTRY; + cfs_mutex_down(&ldlm_ref_sem); if (ldlm_refcount == 1) { - int rc = ldlm_cleanup(force); + int rc = ldlm_cleanup(); if (rc) CERROR("ldlm_cleanup failed: %d\n", rc); else @@ -1365,15 +2275,100 @@ void ldlm_put_ref(int force) } else { ldlm_refcount--; } - up(&ldlm_ref_sem); + cfs_mutex_up(&ldlm_ref_sem); + + EXIT; +} + +/* + * Export handle<->lock hash operations. + */ +static unsigned +ldlm_export_lock_hash(cfs_hash_t *hs, void *key, unsigned mask) +{ + return cfs_hash_u64_hash(((struct lustre_handle *)key)->cookie, mask); +} + +static void * +ldlm_export_lock_key(cfs_hlist_node_t *hnode) +{ + struct ldlm_lock *lock; + ENTRY; + + lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash); + RETURN(&lock->l_remote_handle); +} + +static int +ldlm_export_lock_compare(void *key, cfs_hlist_node_t *hnode) +{ + ENTRY; + RETURN(lustre_handle_equal(ldlm_export_lock_key(hnode), key)); +} + +static void * +ldlm_export_lock_get(cfs_hlist_node_t *hnode) +{ + struct ldlm_lock *lock; + ENTRY; + + lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash); + LDLM_LOCK_GET(lock); + + RETURN(lock); +} + +static void * +ldlm_export_lock_put(cfs_hlist_node_t *hnode) +{ + struct ldlm_lock *lock; + ENTRY; + + lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_hash); + LDLM_LOCK_RELEASE(lock); + + RETURN(lock); +} + +static cfs_hash_ops_t ldlm_export_lock_ops = { + .hs_hash = ldlm_export_lock_hash, + .hs_key = ldlm_export_lock_key, + .hs_compare = ldlm_export_lock_compare, + .hs_get = ldlm_export_lock_get, + .hs_put = ldlm_export_lock_put +}; + +int ldlm_init_export(struct obd_export *exp) +{ + ENTRY; + + exp->exp_lock_hash = + cfs_hash_create(obd_uuid2str(&exp->exp_client_uuid), + HASH_EXP_LOCK_CUR_BITS, HASH_EXP_LOCK_MAX_BITS, + &ldlm_export_lock_ops, CFS_HASH_REHASH); + + if (!exp->exp_lock_hash) + RETURN(-ENOMEM); + + RETURN(0); +} +EXPORT_SYMBOL(ldlm_init_export); +void ldlm_destroy_export(struct obd_export *exp) +{ + ENTRY; + cfs_hash_putref(exp->exp_lock_hash); + exp->exp_lock_hash = NULL; EXIT; } +EXPORT_SYMBOL(ldlm_destroy_export); static int ldlm_setup(void) { struct ldlm_bl_pool *blp; int rc = 0; + int ldlm_min_threads = LDLM_THREADS_AUTO_MIN; + int ldlm_max_threads = LDLM_THREADS_AUTO_MAX; #ifdef __KERNEL__ int i; #endif @@ -1386,17 +2381,32 @@ static int ldlm_setup(void) if (ldlm_state == NULL) RETURN(-ENOMEM); -#ifdef __KERNEL__ +#ifdef LPROCFS rc = ldlm_proc_setup(); if (rc != 0) GOTO(out_free, rc); #endif +#ifdef __KERNEL__ + if (ldlm_num_threads) { + /* If ldlm_num_threads is set, it is the min and the max. */ + if (ldlm_num_threads > LDLM_THREADS_AUTO_MAX) + ldlm_num_threads = LDLM_THREADS_AUTO_MAX; + if (ldlm_num_threads < LDLM_THREADS_AUTO_MIN) + ldlm_num_threads = LDLM_THREADS_AUTO_MIN; + ldlm_min_threads = ldlm_max_threads = ldlm_num_threads; + } +#endif + ldlm_state->ldlm_cb_service = ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE, - LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL, - 1500, ldlm_callback_handler, "ldlm_cbd", - ldlm_svc_proc_dir); + LDLM_MAXREPSIZE, LDLM_CB_REQUEST_PORTAL, + LDLM_CB_REPLY_PORTAL, 2, + ldlm_callback_handler, "ldlm_cbd", + ldlm_svc_proc_dir, NULL, + ldlm_min_threads, ldlm_max_threads, + "ldlm_cb", + LCT_MD_THREAD|LCT_DT_THREAD, NULL); if (!ldlm_state->ldlm_cb_service) { CERROR("failed to start service\n"); @@ -1405,10 +2415,14 @@ static int ldlm_setup(void) ldlm_state->ldlm_cancel_service = ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE, - LDLM_CANCEL_REQUEST_PORTAL, - LDLM_CANCEL_REPLY_PORTAL, 30000, + LDLM_MAXREPSIZE, LDLM_CANCEL_REQUEST_PORTAL, + LDLM_CANCEL_REPLY_PORTAL, 6, ldlm_cancel_handler, "ldlm_canceld", - ldlm_svc_proc_dir); + ldlm_svc_proc_dir, NULL, + ldlm_min_threads, ldlm_max_threads, + "ldlm_cn", + LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD, + NULL); if (!ldlm_state->ldlm_cancel_service) { CERROR("failed to start service\n"); @@ -1420,58 +2434,53 @@ static int ldlm_setup(void) GOTO(out_proc, rc = -ENOMEM); ldlm_state->ldlm_bl_pool = blp; - atomic_set(&blp->blp_num_threads, 0); - init_waitqueue_head(&blp->blp_waitq); - spin_lock_init(&blp->blp_lock); - - INIT_LIST_HEAD(&blp->blp_list); + cfs_spin_lock_init(&blp->blp_lock); + CFS_INIT_LIST_HEAD(&blp->blp_list); + CFS_INIT_LIST_HEAD(&blp->blp_prio_list); + cfs_waitq_init(&blp->blp_waitq); + cfs_atomic_set(&blp->blp_num_threads, 0); + cfs_atomic_set(&blp->blp_busy_threads, 0); + blp->blp_min_threads = ldlm_min_threads; + blp->blp_max_threads = ldlm_max_threads; #ifdef __KERNEL__ - for (i = 0; i < LDLM_NUM_THREADS; i++) { - struct ldlm_bl_thread_data bltd = { - .bltd_num = i, - .bltd_blp = blp, - }; - init_completion(&blp->blp_comp); - rc = kernel_thread(ldlm_bl_thread_main, &bltd, 0); - if (rc < 0) { - CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc); + for (i = 0; i < blp->blp_min_threads; i++) { + rc = ldlm_bl_thread_start(blp); + if (rc < 0) GOTO(out_thread, rc); - } - wait_for_completion(&blp->blp_comp); } - rc = ptlrpc_start_n_threads(NULL, ldlm_state->ldlm_cancel_service, - LDLM_NUM_THREADS, "ldlm_cn"); + rc = ptlrpc_start_threads(NULL, ldlm_state->ldlm_cancel_service); if (rc) GOTO(out_thread, rc); - rc = ptlrpc_start_n_threads(NULL, ldlm_state->ldlm_cb_service, - LDLM_NUM_THREADS, "ldlm_cb"); + rc = ptlrpc_start_threads(NULL, ldlm_state->ldlm_cb_service); if (rc) GOTO(out_thread, rc); - INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks); - spin_lock_init(&expired_lock_thread.elt_lock); + CFS_INIT_LIST_HEAD(&expired_lock_thread.elt_expired_locks); expired_lock_thread.elt_state = ELT_STOPPED; - init_waitqueue_head(&expired_lock_thread.elt_waitq); + cfs_waitq_init(&expired_lock_thread.elt_waitq); + + CFS_INIT_LIST_HEAD(&waiting_locks_list); + cfs_spin_lock_init(&waiting_locks_spinlock); + cfs_timer_init(&waiting_locks_timer, waiting_locks_callback, 0); - rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS); + rc = cfs_kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FILES); if (rc < 0) { CERROR("Cannot start ldlm expired-lock thread: %d\n", rc); GOTO(out_thread, rc); } - wait_event(expired_lock_thread.elt_waitq, - expired_lock_thread.elt_state == ELT_READY); - - INIT_LIST_HEAD(&waiting_locks_list); - spin_lock_init(&waiting_locks_spinlock); - waiting_locks_timer.function = waiting_locks_callback; - waiting_locks_timer.data = 0; - init_timer(&waiting_locks_timer); + cfs_wait_event(expired_lock_thread.elt_waitq, + expired_lock_thread.elt_state == ELT_READY); #endif +#ifdef __KERNEL__ + rc = ldlm_pools_init(); + if (rc) + GOTO(out_thread, rc); +#endif RETURN(0); #ifdef __KERNEL__ @@ -1481,7 +2490,7 @@ static int ldlm_setup(void) #endif out_proc: -#ifdef __KERNEL__ +#ifdef LPROCFS ldlm_proc_cleanup(); out_free: #endif @@ -1490,44 +2499,48 @@ static int ldlm_setup(void) return rc; } -static int ldlm_cleanup(int force) +static int ldlm_cleanup(void) { #ifdef __KERNEL__ struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool; #endif ENTRY; - if (!list_empty(&ldlm_namespace_list)) { + if (!cfs_list_empty(ldlm_namespace_list(LDLM_NAMESPACE_SERVER)) || + !cfs_list_empty(ldlm_namespace_list(LDLM_NAMESPACE_CLIENT))) { CERROR("ldlm still has namespaces; clean these up first.\n"); - ldlm_dump_all_namespaces(D_DLMTRACE); + ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE); + ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE); RETURN(-EBUSY); } #ifdef __KERNEL__ - while (atomic_read(&blp->blp_num_threads) > 0) { + ldlm_pools_fini(); +#endif + +#ifdef __KERNEL__ + while (cfs_atomic_read(&blp->blp_num_threads) > 0) { struct ldlm_bl_work_item blwi = { .blwi_ns = NULL }; - init_completion(&blp->blp_comp); + cfs_init_completion(&blp->blp_comp); - spin_lock(&blp->blp_lock); - list_add_tail(&blwi.blwi_entry, &blp->blp_list); - wake_up(&blp->blp_waitq); - spin_unlock(&blp->blp_lock); + cfs_spin_lock(&blp->blp_lock); + cfs_list_add_tail(&blwi.blwi_entry, &blp->blp_list); + cfs_waitq_signal(&blp->blp_waitq); + cfs_spin_unlock(&blp->blp_lock); - wait_for_completion(&blp->blp_comp); + cfs_wait_for_completion(&blp->blp_comp); } OBD_FREE(blp, sizeof(*blp)); - ptlrpc_stop_all_threads(ldlm_state->ldlm_cb_service); ptlrpc_unregister_service(ldlm_state->ldlm_cb_service); - ptlrpc_stop_all_threads(ldlm_state->ldlm_cancel_service); ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service); ldlm_proc_cleanup(); expired_lock_thread.elt_state = ELT_TERMINATE; - wake_up(&expired_lock_thread.elt_waitq); - wait_event(expired_lock_thread.elt_waitq, - expired_lock_thread.elt_state == ELT_STOPPED); + cfs_waitq_signal(&expired_lock_thread.elt_waitq); + cfs_wait_event(expired_lock_thread.elt_waitq, + expired_lock_thread.elt_state == ELT_STOPPED); #else ptlrpc_unregister_service(ldlm_state->ldlm_cb_service); ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service); @@ -1541,37 +2554,56 @@ static int ldlm_cleanup(int force) int __init ldlm_init(void) { - ldlm_resource_slab = kmem_cache_create("ldlm_resources", + cfs_init_mutex(&ldlm_ref_sem); + cfs_init_mutex(ldlm_namespace_lock(LDLM_NAMESPACE_SERVER)); + cfs_init_mutex(ldlm_namespace_lock(LDLM_NAMESPACE_CLIENT)); + ldlm_resource_slab = cfs_mem_cache_create("ldlm_resources", sizeof(struct ldlm_resource), 0, - SLAB_HWCACHE_ALIGN, NULL, NULL); + CFS_SLAB_HWCACHE_ALIGN); if (ldlm_resource_slab == NULL) return -ENOMEM; - ldlm_lock_slab = kmem_cache_create("ldlm_locks", - sizeof(struct ldlm_lock), 0, - SLAB_HWCACHE_ALIGN, NULL, NULL); + ldlm_lock_slab = cfs_mem_cache_create("ldlm_locks", + sizeof(struct ldlm_lock), 0, + CFS_SLAB_HWCACHE_ALIGN | CFS_SLAB_DESTROY_BY_RCU); if (ldlm_lock_slab == NULL) { - kmem_cache_destroy(ldlm_resource_slab); + cfs_mem_cache_destroy(ldlm_resource_slab); return -ENOMEM; } + ldlm_interval_slab = cfs_mem_cache_create("interval_node", + sizeof(struct ldlm_interval), + 0, CFS_SLAB_HWCACHE_ALIGN); + if (ldlm_interval_slab == NULL) { + cfs_mem_cache_destroy(ldlm_resource_slab); + cfs_mem_cache_destroy(ldlm_lock_slab); + return -ENOMEM; + } +#if LUSTRE_TRACKS_LOCK_EXP_REFS + class_export_dump_hook = ldlm_dump_export_locks; +#endif return 0; } void __exit ldlm_exit(void) { - if ( ldlm_refcount ) + int rc; + if (ldlm_refcount) CERROR("ldlm_refcount is %d in ldlm_exit!\n", ldlm_refcount); - LASSERTF(kmem_cache_destroy(ldlm_resource_slab) == 0, - "couldn't free ldlm resource slab\n"); - LASSERTF(kmem_cache_destroy(ldlm_lock_slab) == 0, - "couldn't free ldlm lock slab\n"); - + rc = cfs_mem_cache_destroy(ldlm_resource_slab); + LASSERTF(rc == 0, "couldn't free ldlm resource slab\n"); +#ifdef __KERNEL__ + /* ldlm_lock_put() use RCU to call ldlm_lock_free, so need call + * synchronize_rcu() to wait a grace period elapsed, so that + * ldlm_lock_free() get a chance to be called. */ + synchronize_rcu(); +#endif + rc = cfs_mem_cache_destroy(ldlm_lock_slab); + LASSERTF(rc == 0, "couldn't free ldlm lock slab\n"); + rc = cfs_mem_cache_destroy(ldlm_interval_slab); + LASSERTF(rc == 0, "couldn't free interval node slab\n"); } -/* ldlm_flock.c */ -EXPORT_SYMBOL(ldlm_flock_completion_ast); - /* ldlm_extent.c */ EXPORT_SYMBOL(ldlm_extent_shift_kms); @@ -1588,47 +2620,60 @@ EXPORT_SYMBOL(ldlm_lock_put); EXPORT_SYMBOL(ldlm_lock_match); EXPORT_SYMBOL(ldlm_lock_cancel); EXPORT_SYMBOL(ldlm_lock_addref); +EXPORT_SYMBOL(ldlm_lock_addref_try); EXPORT_SYMBOL(ldlm_lock_decref); EXPORT_SYMBOL(ldlm_lock_decref_and_cancel); EXPORT_SYMBOL(ldlm_lock_change_resource); -EXPORT_SYMBOL(ldlm_lock_set_data); EXPORT_SYMBOL(ldlm_it2str); EXPORT_SYMBOL(ldlm_lock_dump); EXPORT_SYMBOL(ldlm_lock_dump_handle); -EXPORT_SYMBOL(ldlm_cancel_locks_for_export); EXPORT_SYMBOL(ldlm_reprocess_all_ns); +EXPORT_SYMBOL(ldlm_lock_allow_match_locked); EXPORT_SYMBOL(ldlm_lock_allow_match); +EXPORT_SYMBOL(ldlm_lock_downgrade); +EXPORT_SYMBOL(ldlm_lock_convert); /* ldlm_request.c */ +EXPORT_SYMBOL(ldlm_completion_ast_async); +EXPORT_SYMBOL(ldlm_blocking_ast_nocheck); EXPORT_SYMBOL(ldlm_completion_ast); +EXPORT_SYMBOL(ldlm_blocking_ast); +EXPORT_SYMBOL(ldlm_glimpse_ast); EXPORT_SYMBOL(ldlm_expired_completion_wait); +EXPORT_SYMBOL(ldlm_prep_enqueue_req); +EXPORT_SYMBOL(ldlm_prep_elc_req); EXPORT_SYMBOL(ldlm_cli_convert); EXPORT_SYMBOL(ldlm_cli_enqueue); +EXPORT_SYMBOL(ldlm_cli_enqueue_fini); +EXPORT_SYMBOL(ldlm_cli_enqueue_local); EXPORT_SYMBOL(ldlm_cli_cancel); EXPORT_SYMBOL(ldlm_cli_cancel_unused); +EXPORT_SYMBOL(ldlm_cli_cancel_unused_resource); +EXPORT_SYMBOL(ldlm_cli_cancel_req); EXPORT_SYMBOL(ldlm_replay_locks); EXPORT_SYMBOL(ldlm_resource_foreach); EXPORT_SYMBOL(ldlm_namespace_foreach); EXPORT_SYMBOL(ldlm_namespace_foreach_res); -EXPORT_SYMBOL(ldlm_change_cbdata); +EXPORT_SYMBOL(ldlm_resource_iterate); +EXPORT_SYMBOL(ldlm_cancel_resource_local); +EXPORT_SYMBOL(ldlm_cli_cancel_list_local); +EXPORT_SYMBOL(ldlm_cli_cancel_list); /* ldlm_lockd.c */ EXPORT_SYMBOL(ldlm_server_blocking_ast); EXPORT_SYMBOL(ldlm_server_completion_ast); EXPORT_SYMBOL(ldlm_server_glimpse_ast); EXPORT_SYMBOL(ldlm_handle_enqueue); +EXPORT_SYMBOL(ldlm_handle_enqueue0); EXPORT_SYMBOL(ldlm_handle_cancel); +EXPORT_SYMBOL(ldlm_request_cancel); EXPORT_SYMBOL(ldlm_handle_convert); +EXPORT_SYMBOL(ldlm_handle_convert0); EXPORT_SYMBOL(ldlm_del_waiting_lock); EXPORT_SYMBOL(ldlm_get_ref); EXPORT_SYMBOL(ldlm_put_ref); - -#if 0 -/* ldlm_test.c */ -EXPORT_SYMBOL(ldlm_test); -EXPORT_SYMBOL(ldlm_regression_start); -EXPORT_SYMBOL(ldlm_regression_stop); -#endif +EXPORT_SYMBOL(ldlm_refresh_waiting_lock); +EXPORT_SYMBOL(ldlm_revoke_export_locks); /* ldlm_resource.c */ EXPORT_SYMBOL(ldlm_namespace_new); @@ -1638,6 +2683,7 @@ EXPORT_SYMBOL(ldlm_namespace_dump); EXPORT_SYMBOL(ldlm_dump_all_namespaces); EXPORT_SYMBOL(ldlm_resource_get); EXPORT_SYMBOL(ldlm_resource_putref); +EXPORT_SYMBOL(ldlm_resource_unlink_lock); /* ldlm_lib.c */ EXPORT_SYMBOL(client_import_add_conn); @@ -1646,7 +2692,7 @@ EXPORT_SYMBOL(client_obd_setup); EXPORT_SYMBOL(client_obd_cleanup); EXPORT_SYMBOL(client_connect_import); EXPORT_SYMBOL(client_disconnect_export); -EXPORT_SYMBOL(target_start_recovery_thread); +EXPORT_SYMBOL(server_disconnect_export); EXPORT_SYMBOL(target_stop_recovery_thread); EXPORT_SYMBOL(target_handle_connect); EXPORT_SYMBOL(target_cleanup_recovery); @@ -1655,10 +2701,9 @@ EXPORT_SYMBOL(target_cancel_recovery_timer); EXPORT_SYMBOL(target_send_reply); EXPORT_SYMBOL(target_queue_recovery_request); EXPORT_SYMBOL(target_handle_ping); +EXPORT_SYMBOL(target_pack_pool_reply); EXPORT_SYMBOL(target_handle_disconnect); /* l_lock.c */ EXPORT_SYMBOL(lock_res_and_lock); EXPORT_SYMBOL(unlock_res_and_lock); - -