X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lockd.c;h=08b2eadfb790f93dc321cf977ebaed8f94308240;hp=64dfb52a008dfbefa25d2235f85ab585082898be;hb=ec295cad998d4d5e7fc915491da5ac646dbc6af6;hpb=ea6598377a832b594e09a286eae8b3571145188c diff --git a/lustre/ldlm/ldlm_lockd.c b/lustre/ldlm/ldlm_lockd.c index 64dfb52..08b2ead 100644 --- a/lustre/ldlm/ldlm_lockd.c +++ b/lustre/ldlm/ldlm_lockd.c @@ -1,24 +1,27 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (C) 2002, 2003 Cluster File Systems, Inc. + * Copyright (C) 2002-2004 Cluster File Systems, Inc. * Author: Peter Braam * Author: Phil Schwan * - * This file is part of Lustre, http://www.lustre.org. + * This file is part of the Lustre file system, http://www.lustre.org + * Lustre is a trademark of Cluster File Systems, Inc. * - * 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. + * You may have signed or agreed to another license before downloading + * this software. If so, you are bound by the terms and conditions + * of that agreement, and the following does not apply to you. See the + * LICENSE file included with this distribution for more information. * - * 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. + * If you did not agree to a different license, then this copy of Lustre + * is open source 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. * - * 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. + * In either case, 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 + * license text for more details. */ #ifndef EXPORT_SYMTAB @@ -27,48 +30,50 @@ #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 lustre_lock ldlm_handle_lock; -extern struct list_head ldlm_namespace_list; -extern int (*mds_reint_p)(int offset, struct ptlrpc_request *req); -extern int (*mds_getattr_name_p)(int offset, struct ptlrpc_request *req); - -static DECLARE_MUTEX(ldlm_ref_sem); -static int ldlm_refcount = 0; +extern cfs_mem_cache_t *ldlm_resource_slab; +extern cfs_mem_cache_t *ldlm_lock_slab; +static struct semaphore ldlm_ref_sem; +static int ldlm_refcount; /* LDLM state */ -static struct ldlm_state *ldlm ; +static struct ldlm_state *ldlm_state; -inline unsigned long round_timeout(unsigned long timeout) +inline cfs_time_t round_timeout(cfs_time_t timeout) { - return ((timeout / HZ) + 1) * HZ; + return cfs_time_seconds((int)cfs_duration_sec(cfs_time_sub(timeout, 0)) + 1); +} + +/* timeout for initial callback (AST) reply */ +static inline unsigned int ldlm_get_rq_timeout(unsigned int ldlm_timeout, + unsigned int obd_timeout) +{ + unsigned int timeout = min(ldlm_timeout, obd_timeout / 3); + + return timeout < 1 ? 1 : timeout; } #ifdef __KERNEL__ -/* XXX should this be per-ldlm? */ +/* w_l_spinlock protects both waiting_locks_list and expired_lock_thread */ +static spinlock_t waiting_locks_spinlock; /* BH lock (timer) */ static struct list_head waiting_locks_list; -static spinlock_t waiting_locks_spinlock; -static struct timer_list waiting_locks_timer; +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; + int elt_dump; struct list_head elt_expired_locks; - spinlock_t elt_lock; } expired_lock_thread; #endif @@ -79,7 +84,7 @@ static struct expired_lock_thread { struct ldlm_bl_pool { spinlock_t blp_lock; struct list_head blp_list; - wait_queue_head_t blp_waitq; + cfs_waitq_t blp_waitq; atomic_t blp_num_threads; struct completion blp_comp; }; @@ -89,6 +94,8 @@ struct ldlm_bl_work_item { struct ldlm_namespace *blwi_ns; struct ldlm_lock_desc blwi_ld; struct ldlm_lock *blwi_lock; + struct list_head blwi_head; + int blwi_count; }; #ifdef __KERNEL__ @@ -97,9 +104,10 @@ static inline int have_expired_locks(void) { int need_to_run; - spin_lock_bh(&expired_lock_thread.elt_lock); + ENTRY; + spin_lock_bh(&waiting_locks_spinlock); need_to_run = !list_empty(&expired_lock_thread.elt_expired_locks); - spin_unlock_bh(&expired_lock_thread.elt_lock); + spin_unlock_bh(&waiting_locks_spinlock); RETURN(need_to_run); } @@ -108,99 +116,149 @@ static int expired_lock_main(void *arg) { struct list_head *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) { - struct list_head *tmp, *n, work_list; l_wait_event(expired_lock_thread.elt_waitq, have_expired_locks() || expired_lock_thread.elt_state == ELT_TERMINATE, &lwi); - spin_lock_bh(&expired_lock_thread.elt_lock); - while (!list_empty(expired)) { - struct ldlm_lock *lock; + spin_lock_bh(&waiting_locks_spinlock); + if (expired_lock_thread.elt_dump) { + spin_unlock_bh(&waiting_locks_spinlock); - list_add(&work_list, expired); - list_del_init(expired); + /* from waiting_locks_callback, but not in timer */ + libcfs_debug_dumplog(); + libcfs_run_lbug_upcall(__FILE__, + "waiting_locks_callback", + expired_lock_thread.elt_dump); - list_for_each_entry(lock, &work_list, l_pending_chain) { - LDLM_DEBUG(lock, "moving to work list"); - } + spin_lock_bh(&waiting_locks_spinlock); + expired_lock_thread.elt_dump = 0; + } - spin_unlock_bh(&expired_lock_thread.elt_lock); + do_dump = 0; + while (!list_empty(expired)) { + struct obd_export *export; + struct ldlm_lock *lock; - list_for_each_safe(tmp, n, &work_list) { - lock = list_entry(tmp, struct ldlm_lock, - l_pending_chain); - ptlrpc_fail_export(lock->l_export); + lock = list_entry(expired->next, struct ldlm_lock, + l_pending_chain); + if ((void *)lock < LP_POISON + CFS_PAGE_SIZE && + (void *)lock >= LP_POISON) { + spin_unlock_bh(&waiting_locks_spinlock); + CERROR("free lock on elt list %p\n", lock); + LBUG(); } - - - if (!list_empty(&work_list)) { - list_for_each_entry(lock, &work_list, l_pending_chain) { - LDLM_ERROR(lock, "still on work list!"); - } + 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", + lock->l_export); + lock->l_export = NULL; + LDLM_ERROR(lock, "free export"); + continue; } - LASSERTF (list_empty(&work_list), - "some exports not failed properly\n"); + export = class_export_get(lock->l_export); + spin_unlock_bh(&waiting_locks_spinlock); - spin_lock_bh(&expired_lock_thread.elt_lock); + do_dump++; + class_fail_export(export); + class_export_put(export); + spin_lock_bh(&waiting_locks_spinlock); + } + 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 int ldlm_add_waiting_lock(struct ldlm_lock *lock); + +/* This is called from within a timer interrupt and cannot schedule */ static void waiting_locks_callback(unsigned long unused) { - struct ldlm_lock *lock; - char str[PTL_NALFMT_SIZE]; + struct ldlm_lock *lock, *last = NULL; +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); - if (lock->l_callback_timeout > jiffies) + if (cfs_time_after(lock->l_callback_timeout, cfs_time_current()) || + (lock->l_req_mode == LCK_GROUP)) break; + if (ptlrpc_check_suspend()) { + /* there is a case when we talk to one mds, holding + * lock from another mds. this way we easily can get + * here, if second mds is being recovered. so, we + * suspend timeouts. bug 6019 */ + + 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, + libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid)); + + list_del_init(&lock->l_pending_chain); + 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)); + + list_del_init(&lock->l_pending_chain); + 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 "LPX64" (%s) ", + "%s@%s nid %s\n", lock->l_export->exp_client_uuid.uuid, lock->l_export->exp_connection->c_remote_uuid.uuid, - lock->l_export->exp_connection->c_peer.peer_nid, - portals_nid2str(lock->l_export->exp_connection->c_peer.peer_ni->pni_number, - lock->l_export->exp_connection->c_peer.peer_nid, - str)); + libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid)); + + last = lock; - 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); + } + + if (!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); } /* @@ -208,11 +266,11 @@ static void waiting_locks_callback(unsigned long unused) * left. */ if (!list_empty(&waiting_locks_list)) { - unsigned long timeout_rounded; + cfs_time_t 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); + timeout_rounded = (cfs_time_t)round_timeout(lock->l_callback_timeout); + cfs_timer_arm(&waiting_locks_timer, timeout_rounded); } spin_unlock_bh(&waiting_locks_spinlock); } @@ -222,76 +280,133 @@ static void waiting_locks_callback(unsigned long unused) * 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). + * + * 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) { - unsigned long timeout_rounded; + 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 (!list_empty(&lock->l_pending_chain)) return 0; - } - LDLM_DEBUG(lock, "adding to wait list"); - lock->l_callback_timeout = jiffies + (obd_timeout * HZ / 2); + lock->l_callback_timeout =cfs_time_add(cfs_time_current(), + cfs_time_seconds(obd_timeout)/2); 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); return 1; } +static int ldlm_add_waiting_lock(struct ldlm_lock *lock) +{ + int ret; + + LASSERT(!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)); + + spin_lock_bh(&waiting_locks_spinlock); + if (lock->l_destroyed) { + static cfs_time_t next; + 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); + spin_unlock_bh(&waiting_locks_spinlock); + + LDLM_DEBUG(lock, "%sadding to wait list", + ret == 0 ? "not re-" : ""); + 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. + * + * Called with namespace lock held. */ -int ldlm_del_waiting_lock(struct ldlm_lock *lock) +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); - - if (list_empty(&lock->l_pending_chain)) { - spin_unlock_bh(&waiting_locks_spinlock); - LDLM_DEBUG(lock, "wasn't waiting"); + if (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)); + cfs_timer_arm(&waiting_locks_timer, + round_timeout(next->l_callback_timeout)); } } - - spin_lock_bh(&expired_lock_thread.elt_lock); list_del_init(&lock->l_pending_chain); - spin_unlock_bh(&expired_lock_thread.elt_lock); + 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. */ + LDLM_DEBUG(lock, "client lock: no-op"); + return 0; + } + + spin_lock_bh(&waiting_locks_spinlock); + ret = __ldlm_del_waiting_lock(lock); spin_unlock_bh(&waiting_locks_spinlock); - LDLM_DEBUG(lock, "removed"); + + 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) +{ + 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); + + if (list_empty(&lock->l_pending_chain)) { + spin_unlock_bh(&waiting_locks_spinlock); + LDLM_DEBUG(lock, "wasn't waiting"); + return 0; + } + + __ldlm_del_waiting_lock(lock); + __ldlm_add_waiting_lock(lock); + spin_unlock_bh(&waiting_locks_spinlock); + + LDLM_DEBUG(lock, "refreshed"); return 1; } @@ -299,6 +414,7 @@ int ldlm_del_waiting_lock(struct ldlm_lock *lock) static int ldlm_add_waiting_lock(struct ldlm_lock *lock) { + LASSERT(!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)); RETURN(1); } @@ -307,33 +423,94 @@ int ldlm_del_waiting_lock(struct ldlm_lock *lock) RETURN(0); } +int ldlm_refresh_waiting_lock(struct ldlm_lock *lock) +{ + RETURN(0); +} #endif /* __KERNEL__ */ -static void ldlm_failed_ast(struct ldlm_lock *lock, int rc, char *ast_type) +static void ldlm_failed_ast(struct ldlm_lock *lock, int rc, + const char *ast_type) { - const struct ptlrpc_connection *conn = lock->l_export->exp_connection; - char str[PTL_NALFMT_SIZE]; - - CERROR("%s AST failed (%d) for res "LPU64"/"LPU64 - ", mode %s: evicting client %s@%s NID "LPX64" (%s)\n", - ast_type, rc, - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], - ldlm_lockname[lock->l_granted_mode], - lock->l_export->exp_client_uuid.uuid, - conn->c_remote_uuid.uuid, conn->c_peer.peer_nid, - portals_nid2str(conn->c_peer.peer_ni->pni_number, - conn->c_peer.peer_nid, str)); - ptlrpc_fail_export(lock->l_export); + struct ptlrpc_connection *conn = lock->l_export->exp_connection; + char *str = libcfs_nid2str(conn->c_peer.nid); + + LCONSOLE_ERROR_MSG(0x138, "A client on nid %s was evicted from " + "service %s.\n", str, + lock->l_export->exp_obd->obd_name); + + LCONSOLE_ERROR_MSG(0x012, "Lock %s callback to %s timed out for " + "resource %d\n", ast_type, + obd_export_nid2str(lock->l_export), rc); + + if (obd_dump_on_timeout) + libcfs_debug_dumplog(); + class_fail_export(lock->l_export); } +static int ldlm_handle_ast_error(struct ldlm_lock *lock, + struct ptlrpc_request *req, int rc, + const char *ast_type) +{ + 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, + 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 { + ldlm_del_waiting_lock(lock); + ldlm_failed_ast(lock, rc, ast_type); + } + } else if (rc) { + if (rc == -EINVAL) + LDLM_DEBUG(lock, "client (nid %s) returned %d" + " 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", libcfs_nid2str(peer.nid), + (req->rq_repmsg != NULL) ? + 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. */ + rc = -ERESTART; + } + + return rc; +} + +/* + * ->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); + int size[] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body), + [DLM_LOCKREQ_OFF] = sizeof(*body) }; + int instant_cancel = 0, rc = 0; ENTRY; if (flag == LDLM_CB_CANCELING) { @@ -342,107 +519,83 @@ int ldlm_server_blocking_ast(struct ldlm_lock *lock, } LASSERT(lock); + 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); + } - l_lock(&lock->l_resource->lr_namespace->ns_lock); + req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, + LUSTRE_DLM_VERSION, LDLM_BL_CALLBACK, 2, size, + NULL); + if (req == NULL) + RETURN(-ENOMEM); + + 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 */ - l_unlock(&lock->l_resource->lr_namespace->ns_lock); - LDLM_DEBUG(lock, "lock not granted, not sending blocking AST"); RETURN(0); + 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? */ - l_unlock(&lock->l_resource->lr_namespace->ns_lock); + unlock_res(lock->l_resource); + ptlrpc_req_finished(req); RETURN(0); } -#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); - } -#endif + if (lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK) + instant_cancel = 1; - req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, - LDLM_BL_CALLBACK, 1, &size, NULL); - if (req == NULL) { - l_unlock(&lock->l_resource->lr_namespace->ns_lock); - RETURN(-ENOMEM); - } - - 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)); + body = lustre_msg_buf(req->rq_reqmsg, DLM_LOCKREQ_OFF, sizeof(*body)); + 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_req_set_repsize(req, 1, NULL); + 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); - l_unlock(&lock->l_resource->lr_namespace->ns_lock); + unlock_res(lock->l_resource); + } req->rq_send_state = LUSTRE_IMP_FULL; - req->rq_timeout = 2; /* 2 second timeout for initial AST reply */ - rc = ptlrpc_queue_wait(req); - if (rc == -ETIMEDOUT || rc == -EINTR) { -#ifdef __KERNEL__ - ldlm_del_waiting_lock(lock); - ldlm_failed_ast(lock, rc, "blocking"); -#else - /* XXX - * Here we treat all clients as liblustre. When BLOCKING AST - * timeout we don't evicting the client and only cancel - * the lock. - * restore to orignial implementation later!!! - * XXX - */ - CERROR("BLOCKING AST to client (nid "LPU64") timeout, " - "simply cancel lock 0x%p\n", - req->rq_peer.peer_nid, lock); - ldlm_lock_cancel(lock); - rc = -ERESTART; -#endif - } else if (rc) { + req->rq_timeout = ldlm_get_rq_timeout(ldlm_timeout, obd_timeout); + + if (lock->l_export && lock->l_export->exp_ldlm_stats) + lprocfs_counter_incr(lock->l_export->exp_ldlm_stats, + LDLM_BL_CALLBACK - LDLM_FIRST_OPC); + + if (unlikely(instant_cancel)) { + rc = ptl_send_rpc(req, 1); + } else { + rc = ptlrpc_queue_wait(req); + OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2); + } + if (rc != 0) { + /* If client canceled the lock but the cancel has not been + * recieved yet, we need to update lvbo to have the proper + * attributes cached. */ if (rc == -EINVAL) - CDEBUG(D_DLMTRACE, "client (nid "LPU64") returned %d " - "from blocking AST for lock %p--normal race\n", - req->rq_peer.peer_nid, - req->rq_repmsg->status, lock); - else if (rc == -ENOTCONN) - CDEBUG(D_DLMTRACE, "client (nid "LPU64") returned %d " - "from blocking AST for lock %p--this client was " - "probably rebooted while it held a lock, nothing" - " serious\n",req->rq_peer.peer_nid, - req->rq_repmsg->status, lock); - else - CDEBUG(D_ERROR, "client (nid "LPU64") returned %d " - "from blocking AST for lock %p\n", - req->rq_peer.peer_nid, - (req->rq_repmsg != NULL)? - req->rq_repmsg->status : 0, - lock); - LDLM_DEBUG(lock, "client sent rc %d rq_status %d from blocking " - "AST", rc, req->rq_status); - ldlm_lock_cancel(lock); - /* Server-side AST functions are called from ldlm_reprocess_all, - * which needs to be told to please restart its reprocessing. */ - rc = -ERESTART; + ldlm_res_lvbo_update(lock->l_resource, NULL, 0, 1); + rc = ldlm_handle_ast_error(lock, req, rc, "blocking"); } ptlrpc_req_finished(req); - RETURN(rc); -} + /* If we cancelled the lock, we need to restart ldlm_reprocess_queue */ + if (!rc && instant_cancel) + rc = -ERESTART; -/* 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) @@ -451,211 +604,504 @@ int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data) struct ptlrpc_request *req; struct timeval granted_time; long total_enqueue_wait; - int rc = 0, size = sizeof(*body); + int size[3] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body), + [DLM_LOCKREQ_OFF] = sizeof(*body) }; + int rc = 0, buffers = 2, instant_cancel = 0; ENTRY; - if (lock == NULL) { - LBUG(); - RETURN(-EINVAL); - } + LASSERT(lock != NULL); do_gettimeofday(&granted_time); - total_enqueue_wait = timeval_sub(&granted_time, &lock->l_enqueued_time); + total_enqueue_wait = cfs_timeval_sub(&granted_time, + &lock->l_enqueued_time, NULL); if (total_enqueue_wait / 1000000 > obd_timeout) - LDLM_ERROR(lock, "enqueue wait took %ldus", total_enqueue_wait); + LDLM_ERROR(lock, "enqueue wait took %luus from %lu", + total_enqueue_wait, lock->l_enqueued_time.tv_sec); + + lock_res_and_lock(lock); + if (lock->l_resource->lr_lvb_len) { + size[DLM_REQ_REC_OFF] = lock->l_resource->lr_lvb_len; + buffers = 3; + } + unlock_res_and_lock(lock); req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, - LDLM_CP_CALLBACK, 1, &size, NULL); - if (!req) + 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)); + body = lustre_msg_buf(req->rq_reqmsg, DLM_LOCKREQ_OFF, sizeof(*body)); + body->lock_handle[0] = lock->l_remote_handle; body->lock_flags = flags; ldlm_lock2desc(lock, &body->lock_desc); + if (buffers == 3) { + void *lvb; + + lvb = lustre_msg_buf(req->rq_reqmsg, DLM_REQ_REC_OFF, + 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)", total_enqueue_wait); - req->rq_replen = lustre_msg_size(0, NULL); + + ptlrpc_req_set_repsize(req, 1, NULL); req->rq_send_state = LUSTRE_IMP_FULL; - req->rq_timeout = 2; /* 2 second timeout for initial AST reply */ + req->rq_timeout = ldlm_get_rq_timeout(ldlm_timeout, obd_timeout); /* We only send real blocking ASTs after the lock is granted */ - l_lock(&lock->l_resource->lr_namespace->ns_lock); + 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 */ + + /* 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); + } } - l_unlock(&lock->l_resource->lr_namespace->ns_lock); + unlock_res_and_lock(lock); + + if (lock->l_export && lock->l_export->exp_ldlm_stats) + lprocfs_counter_incr(lock->l_export->exp_ldlm_stats, + LDLM_CP_CALLBACK - LDLM_FIRST_OPC); rc = ptlrpc_queue_wait(req); - if (rc == -ETIMEDOUT || rc == -EINTR) { - ldlm_del_waiting_lock(lock); - ldlm_failed_ast(lock, rc, "completion"); - } else if (rc) { - LDLM_ERROR(lock, "client sent rc %d rq_status %d from " - "completion AST\n", rc, req->rq_status); - ldlm_lock_cancel(lock); - /* Server-side AST functions are called from ldlm_reprocess_all, - * which needs to be told to please restart its reprocessing. */ - rc = -ERESTART; - } + if (rc != 0) + rc = ldlm_handle_ast_error(lock, req, rc, "completion"); + ptlrpc_req_finished(req); + /* If we cancelled the lock, we need to restart ldlm_reprocess_queue */ + if (!rc && instant_cancel) + rc = -ERESTART; + RETURN(rc); } -int ldlm_handle_enqueue(struct ptlrpc_request *req, - ldlm_completion_callback completion_callback, - ldlm_blocking_callback blocking_callback) +int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data) +{ + struct ldlm_resource *res = lock->l_resource; + struct ldlm_request *body; + struct ptlrpc_request *req; + int size[] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body), + [DLM_LOCKREQ_OFF] = sizeof(*body) }; + int rc = 0; + ENTRY; + + LASSERT(lock != NULL); + + req = ptlrpc_prep_req(lock->l_export->exp_imp_reverse, + LUSTRE_DLM_VERSION, LDLM_GL_CALLBACK, 2, size, + NULL); + if (req == NULL) + RETURN(-ENOMEM); + + body = lustre_msg_buf(req->rq_reqmsg, DLM_LOCKREQ_OFF, sizeof(*body)); + body->lock_handle[0] = lock->l_remote_handle; + ldlm_lock2desc(lock, &body->lock_desc); + + lock_res_and_lock(lock); + size[REPLY_REC_OFF] = lock->l_resource->lr_lvb_len; + unlock_res_and_lock(lock); + res = lock->l_resource; + ptlrpc_req_set_repsize(req, 2, size); + + req->rq_send_state = LUSTRE_IMP_FULL; + req->rq_timeout = ldlm_get_rq_timeout(ldlm_timeout, obd_timeout); + + if (lock->l_export && lock->l_export->exp_ldlm_stats) + lprocfs_counter_incr(lock->l_export->exp_ldlm_stats, + LDLM_GL_CALLBACK - LDLM_FIRST_OPC); + + rc = ptlrpc_queue_wait(req); + if (rc == -ELDLM_NO_LOCK_DATA) + LDLM_DEBUG(lock, "lost race - client has a lock but no inode"); + else if (rc != 0) + rc = ldlm_handle_ast_error(lock, req, rc, "glimpse"); + else + rc = ldlm_res_lvbo_update(res, req->rq_repmsg, + REPLY_REC_OFF, 1); + ptlrpc_req_finished(req); + RETURN(rc); +} + +static struct ldlm_lock * +find_existing_lock(struct obd_export *exp, + const struct lustre_handle *remote_hdl) +{ + struct list_head *iter; + + 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; + } + } + spin_unlock(&exp->exp_ldlm_data.led_lock); + return NULL; +} + +#ifdef __KERNEL__ +extern unsigned long long lu_time_stamp_get(void); +#else +#define lu_time_stamp_get() time(NULL) +#endif + +/* + * 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, size = sizeof(*dlm_rep), cookielen = 0; + int size[3] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body), + [DLM_LOCKREPLY_OFF] = sizeof(*dlm_rep) }; + int rc = 0; __u32 flags; - ldlm_error_t err; + ldlm_error_t err = ELDLM_OK; struct ldlm_lock *lock = NULL; void *cookie = NULL; ENTRY; LDLM_DEBUG_NOLOCK("server-side enqueue handler START"); - 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); + ldlm_request_cancel(req, dlm_req, LDLM_ENQUEUE_CANCEL_OFF); + flags = dlm_req->lock_flags; + + LASSERT(req->rq_export); + + if (req->rq_export->exp_ldlm_stats) + lprocfs_counter_incr(req->rq_export->exp_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 (dlm_req->lock_desc.l_resource.lr_type == LDLM_PLAIN && - (flags & LDLM_FL_HAS_INTENT)) { - /* In this case, the reply buffer is allocated deep in - * local_lock_enqueue by the policy function. */ - cookie = req; - cookielen = sizeof(*req); - } else { - rc = lustre_pack_reply(req, 1, &size, NULL); - if (rc) { - CERROR("out of memory\n"); - RETURN(-ENOMEM); + 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); + } + + 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); } - if (dlm_req->lock_desc.l_resource.lr_type != LDLM_PLAIN) { - cookie = &dlm_req->lock_desc.l_policy_data; - cookielen = sizeof(ldlm_policy_data_t); + } 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 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)) { + lock = find_existing_lock(req->rq_export, + &dlm_req->lock_handle[0]); + if (lock != NULL) { + 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, NULL); + cbs->lcs_blocking, cbs->lcs_completion, + cbs->lcs_glimpse, NULL, 0); + if (!lock) - GOTO(out, err = -ENOMEM); + 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_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", req->rq_export); + GOTO(out, rc = -ENOTCONN); + } lock->l_export = class_export_get(req->rq_export); - l_lock(&lock->l_resource->lr_namespace->ns_lock); + 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); - l_unlock(&lock->l_resource->lr_namespace->ns_lock); + spin_unlock(&lock->l_export->exp_ldlm_data.led_lock); - err = ldlm_lock_enqueue(obddev->obd_namespace, &lock, cookie, cookielen, - &flags); +existing_lock: + + if (flags & LDLM_FL_HAS_INTENT) { + /* In this case, the reply buffer is allocated deep in + * local_lock_enqueue by the policy function. */ + cookie = req; + } else { + int buffers = 2; + + lock_res_and_lock(lock); + if (lock->l_resource->lr_lvb_len) { + size[DLM_REPLY_REC_OFF] = lock->l_resource->lr_lvb_len; + buffers = 3; + } + unlock_res_and_lock(lock); + + if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR)) + GOTO(out, rc = -ENOMEM); + + rc = lustre_pack_reply(req, buffers, size, NULL); + if (rc) + GOTO(out, rc); + } + + if (dlm_req->lock_desc.l_resource.lr_type != LDLM_PLAIN) + lock->l_policy_data = dlm_req->lock_desc.l_policy_data; + if (dlm_req->lock_desc.l_resource.lr_type == LDLM_EXTENT) + lock->l_req_extent = lock->l_policy_data.l_extent; + + err = ldlm_lock_enqueue(ns, &lock, cookie, &flags); if (err) GOTO(out, err); - dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep)); + dlm_rep = lustre_msg_buf(req->rq_repmsg, DLM_LOCKREPLY_OFF, + sizeof(*dlm_rep)); dlm_rep->lock_flags = flags; - ldlm_lock2handle(lock, &dlm_rep->lock_handle); - if (dlm_req->lock_desc.l_resource.lr_type != LDLM_PLAIN) { - memcpy(&dlm_rep->lock_policy_data, &lock->l_policy_data, - cookielen); - } - if (dlm_rep->lock_flags & LDLM_FL_LOCK_CHANGED) { - memcpy(&dlm_rep->lock_resource_name, &lock->l_resource->lr_name, - sizeof(dlm_rep->lock_resource_name)); - dlm_rep->lock_mode = lock->l_req_mode; - } + ldlm_lock2desc(lock, &dlm_rep->lock_desc); + ldlm_lock2handle(lock, &dlm_rep->lock_handle); /* We never send a blocking AST until the lock is granted, but * we can tell it right now */ - l_lock(&lock->l_resource->lr_namespace->ns_lock); - if (lock->l_flags & LDLM_FL_AST_SENT) { + lock_res_and_lock(lock); + + /* 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_ONCE(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 immediatelly and would never be granted + * in the future, causing timeouts on client. Not + * granted lock will be cancelled immediatelly 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 %d\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 = lustre_msg_buf(req->rq_reqmsg, + DLM_INTENT_IT_OFF, + sizeof(*it)); + if (it != NULL) { + CERROR("This is intent %s ("LPU64")\n", + ldlm_it2str(it->opc), it->opc); + } + } + } } - l_unlock(&lock->l_resource->lr_namespace->ns_lock); + + unlock_res_and_lock(lock); EXIT; out: req->rq_status = err; + if (req->rq_reply_state == NULL) { + 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 * ldlm_reprocess_all. If this moves, revisit that code. -phil */ if (lock) { LDLM_DEBUG(lock, "server-side enqueue handler, sending reply" - "(err=%d)", err); + "(err=%d, rc=%d)", err, rc); + + lock_res_and_lock(lock); + if (rc == 0) { + size[DLM_REPLY_REC_OFF] = lock->l_resource->lr_lvb_len; + if (size[DLM_REPLY_REC_OFF] > 0) { + void *lvb = lustre_msg_buf(req->rq_repmsg, + DLM_REPLY_REC_OFF, + size[DLM_REPLY_REC_OFF]); + LASSERTF(lvb != NULL, "req %p, lock %p\n", + req, lock); + + memcpy(lvb, lock->l_resource->lr_lvb_data, + size[DLM_REPLY_REC_OFF]); + } + } else { + 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_DEBUG_NOLOCK("server-side enqueue handler END (lock %p)", lock); - return 0; + 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) { + int rc; struct ldlm_request *dlm_req; + struct ldlm_callback_suite cbs = { + .lcs_completion = completion_callback, + .lcs_blocking = blocking_callback, + .lcs_glimpse = glimpse_callback + }; + + + dlm_req = lustre_swab_reqbuf(req, DLM_LOCKREQ_OFF, + sizeof *dlm_req, lustre_swab_ldlm_request); + if (dlm_req != NULL) { + rc = ldlm_handle_enqueue0(req->rq_export->exp_obd->obd_namespace, + req, dlm_req, &cbs); + } else { + CERROR ("Can't unpack dlm_req\n"); + 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; + int size[2] = { [MSG_PTLRPC_BODY_OFF] = sizeof(struct ptlrpc_body), + [DLM_LOCKREPLY_OFF] = sizeof(*dlm_rep) }; 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_ldlm_stats) + lprocfs_counter_incr(req->rq_export->exp_ldlm_stats, + LDLM_CONVERT - LDLM_FIRST_OPC); - rc = lustre_pack_reply(req, 1, &size, NULL); + rc = lustre_pack_reply(req, 2, size, NULL); if (rc) { CERROR("out of memory\n"); RETURN(-ENOMEM); } - dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep)); + dlm_rep = lustre_msg_buf(req->rq_repmsg, DLM_LOCKREPLY_OFF, + sizeof(*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 { + void *res = NULL; + LDLM_DEBUG(lock, "server-side convert handler START"); - ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode, - &dlm_rep->lock_flags); - if (ldlm_del_waiting_lock(lock)) - CDEBUG(D_DLMTRACE, "converted waiting lock %p\n", lock); - req->rq_status = 0; + + do_gettimeofday(&lock->l_enqueued_time); + res = ldlm_lock_convert(lock, dlm_req->lock_desc.l_req_mode, + &dlm_rep->lock_flags); + if (res) { + if (ldlm_del_waiting_lock(lock)) + LDLM_DEBUG(lock, "converted waiting lock"); + req->rq_status = 0; + } else { + req->rq_status = EDEADLOCK; + } } if (lock) { - ldlm_reprocess_all(lock->l_resource); + if (!req->rq_status) + ldlm_reprocess_all(lock->l_resource); LDLM_DEBUG(lock, "server-side convert handler END"); LDLM_LOCK_PUT(lock); } else @@ -664,88 +1110,138 @@ 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 = lustre_swab_reqbuf(req, DLM_LOCKREQ_OFF, sizeof *dlm_req, + lustre_swab_ldlm_request); + 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, which 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; - char str[PTL_NALFMT_SIZE]; + 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_putref(pres); + } + if (res != NULL) { + ldlm_resource_getref(res); + ldlm_res_lvbo_update(res, NULL, 0, 1); + } + pres = res; + } + ldlm_lock_cancel(lock); + LDLM_LOCK_PUT(lock); + } + if (pres != NULL) { + ldlm_reprocess_all(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 = lustre_swab_reqbuf(req, DLM_LOCKREQ_OFF, sizeof(*dlm_req), + lustre_swab_ldlm_request); if (dlm_req == NULL) { CERROR("bad request buffer for cancel\n"); RETURN(-EFAULT); } - rc = lustre_pack_reply(req, 0, NULL, NULL); + if (req->rq_export && req->rq_export->exp_ldlm_stats) + lprocfs_counter_incr(req->rq_export->exp_ldlm_stats, + LDLM_CANCEL - LDLM_FIRST_OPC); + + rc = lustre_pack_reply(req, 1, NULL, NULL); if (rc) { CERROR("out of memory\n"); RETURN(-ENOMEM); } - lock = ldlm_handle2lock(&dlm_req->lock_handle1); - if (!lock) { - CERROR("received cancel for unknown lock cookie "LPX64 - " from nid "LPX64" (%s)\n", dlm_req->lock_handle1.cookie, - req->rq_peer.peer_nid, - portals_nid2str(req->rq_peer.peer_ni->pni_number, - req->rq_peer.peer_nid, str)); - LDLM_DEBUG_NOLOCK("server-side cancel handler stale lock " - "(cookie "LPU64")", - dlm_req->lock_handle1.cookie); + if (!ldlm_request_cancel(req, dlm_req, 0)) req->rq_status = ESTALE; - } else { - LDLM_DEBUG(lock, "server-side cancel handler START"); - ldlm_lock_cancel(lock); - if (ldlm_del_waiting_lock(lock)) - CDEBUG(D_DLMTRACE, "cancelled waiting lock %p\n", lock); - req->rq_status = 0; - } 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); } -static void ldlm_handle_bl_callback(struct ldlm_namespace *ns, - struct ldlm_lock_desc *ld, - struct ldlm_lock *lock) +void ldlm_handle_bl_callback(struct ldlm_namespace *ns, + struct ldlm_lock_desc *ld, struct ldlm_lock *lock) { int do_ast; ENTRY; - l_lock(&ns->ns_lock); LDLM_DEBUG(lock, "client blocking AST callback handler START"); + 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); - if (lock->l_blocking_ast != NULL) { - l_unlock(&ns->ns_lock); - l_check_no_ns_lock(ns); + if (lock->l_blocking_ast != NULL) lock->l_blocking_ast(lock, ld, lock->l_ast_data, LDLM_CB_BLOCKING); - l_lock(&ns->ns_lock); - } } else { LDLM_DEBUG(lock, "Lock still has references, will be" " cancelled later"); } LDLM_DEBUG(lock, "client blocking callback handler END"); - l_unlock(&ns->ns_lock); LDLM_LOCK_PUT(lock); EXIT; } @@ -755,29 +1251,35 @@ 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; - l_lock(&ns->ns_lock); LDLM_DEBUG(lock, "client completion callback handler START"); + lock_res_and_lock(lock); + /* If we receive the completion AST before the actual enqueue returned, * then we might need to switch lock modes, resources, or extents. */ if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) { lock->l_req_mode = dlm_req->lock_desc.l_granted_mode; LDLM_DEBUG(lock, "completion AST, new lock mode"); } - 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)); + + if (lock->l_resource->lr_type != LDLM_PLAIN) { + lock->l_policy_data = dlm_req->lock_desc.l_policy_data; + LDLM_DEBUG(lock, "completion AST, new policy data"); + } ldlm_resource_unlink_lock(lock); if (memcmp(&dlm_req->lock_desc.l_resource.lr_name, &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); + &dlm_req->lock_desc.l_resource.lr_name); 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) { @@ -785,60 +1287,140 @@ static void ldlm_handle_cp_callback(struct ptlrpc_request *req, LDLM_DEBUG(lock, "completion AST includes blocking AST"); } - lock->l_resource->lr_tmp = &ast_list; - ldlm_grant_lock(lock, req, sizeof(*req), 1); - lock->l_resource->lr_tmp = NULL; + if (lock->l_lvb_len) { + void *lvb; + lvb = lustre_swab_reqbuf(req, DLM_REQ_REC_OFF, lock->l_lvb_len, + lock->l_lvb_swabber); + if (lvb == NULL) { + LDLM_ERROR(lock, "completion AST did not contain " + "expected LVB!"); + } else { + memcpy(lock->l_lvb_data, lvb, lock->l_lvb_len); + } + } + + ldlm_grant_lock(lock, &ast_list); + unlock_res_and_lock(lock); + LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work"); - l_unlock(&ns->ns_lock); - LDLM_LOCK_PUT(lock); - ldlm_run_ast_work(ns, &ast_list); + ldlm_run_cp_ast_work(&ast_list); LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)", lock); + LDLM_LOCK_PUT(lock); + EXIT; +} + +static void ldlm_handle_gl_callback(struct ptlrpc_request *req, + struct ldlm_namespace *ns, + struct ldlm_request *dlm_req, + struct ldlm_lock *lock) +{ + int rc = -ENOSYS; + ENTRY; + + LDLM_DEBUG(lock, "client glimpse AST callback handler"); + + if (lock->l_glimpse_ast != NULL) + rc = lock->l_glimpse_ast(lock, req); + + if (req->rq_repmsg != NULL) { + ptlrpc_reply(req); + } else { + req->rq_status = rc; + ptlrpc_error(req); + } + + lock_res_and_lock(lock); + if (lock->l_granted_mode == LCK_PW && + !lock->l_readers && !lock->l_writers && + 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_lock(ns, NULL, lock)) + ldlm_handle_bl_callback(ns, NULL, lock); + + EXIT; + return; + } + unlock_res_and_lock(lock); + LDLM_LOCK_PUT(lock); EXIT; } static int ldlm_callback_reply(struct ptlrpc_request *req, int rc) { req->rq_status = rc; - rc = lustre_pack_reply(req, 0, NULL, NULL); - if (rc) - return rc; + if (req->rq_reply_state == NULL) { + rc = lustre_pack_reply(req, 1, NULL, NULL); + if (rc) + return rc; + } return ptlrpc_reply(req); } #ifdef __KERNEL__ -static int ldlm_bl_to_thread(struct ldlm_state *ldlm, struct ldlm_namespace *ns, - struct ldlm_lock_desc *ld, struct ldlm_lock *lock) +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) { - struct ldlm_bl_pool *blp = ldlm->ldlm_bl_pool; + struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool; struct ldlm_bl_work_item *blwi; ENTRY; + if (cancels && count == 0) + RETURN(0); + OBD_ALLOC(blwi, sizeof(*blwi)); if (blwi == NULL) RETURN(-ENOMEM); blwi->blwi_ns = ns; - blwi->blwi_ld = *ld; - blwi->blwi_lock = lock; - + if (ld != NULL) + blwi->blwi_ld = *ld; + if (count) { + list_add(&blwi->blwi_head, cancels); + list_del_init(cancels); + blwi->blwi_count = count; + } else { + blwi->blwi_lock = lock; + } spin_lock(&blp->blp_lock); list_add_tail(&blwi->blwi_entry, &blp->blp_list); - wake_up(&blp->blp_waitq); + cfs_waitq_signal(&blp->blp_waitq); spin_unlock(&blp->blp_lock); RETURN(0); } #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)); +#else + RETURN(-ENOSYS); +#endif +} + +int ldlm_bl_to_thread_list(struct ldlm_namespace *ns, struct ldlm_lock_desc *ld, + struct list_head *cancels, int count) +{ +#ifdef __KERNEL__ + RETURN(ldlm_bl_to_thread(ns, ld, NULL, cancels, count)); +#else + RETURN(-ENOSYS); +#endif +} + static int ldlm_callback_handler(struct ptlrpc_request *req) { struct ldlm_namespace *ns; struct ldlm_request *dlm_req; struct ldlm_lock *lock; - char str[PTL_NALFMT_SIZE]; int rc; ENTRY; @@ -850,98 +1432,121 @@ static int ldlm_callback_handler(struct ptlrpc_request *req) if (req->rq_export == NULL) { struct ldlm_request *dlm_req; - CDEBUG(D_RPCTRACE, "operation %d from nid "LPX64" (%s) with bad " - "export cookie "LPX64" (ptl req %d/rep %d); this is " + 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_peer.peer_nid, - portals_nid2str(req->rq_peer.peer_ni->pni_number, - req->rq_peer.peer_nid, str), - req->rq_reqmsg->handle.cookie, - req->rq_request_portal, req->rq_reply_portal); + lustre_msg_get_opc(req->rq_reqmsg), + libcfs_id2str(req->rq_peer), + lustre_msg_get_handle(req->rq_reqmsg)->cookie); - dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req), + dlm_req = lustre_swab_reqbuf(req, DLM_LOCKREQ_OFF, + sizeof(*dlm_req), lustre_swab_ldlm_request); if (dlm_req != NULL) CDEBUG(D_RPCTRACE, "--> lock cookie: "LPX64"\n", - dlm_req->lock_handle1.cookie); + dlm_req->lock_handle[0].cookie); ldlm_callback_reply(req, -ENOTCONN); RETURN(0); } - if (req->rq_reqmsg->opc == LDLM_BL_CALLBACK) { + LASSERT(req->rq_export != NULL); + LASSERT(req->rq_export->exp_obd != NULL); + + switch (lustre_msg_get_opc(req->rq_reqmsg)) { + case LDLM_BL_CALLBACK: OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0); - } else if (req->rq_reqmsg->opc == LDLM_CP_CALLBACK) { + break; + case LDLM_CP_CALLBACK: OBD_FAIL_RETURN(OBD_FAIL_LDLM_CP_CALLBACK, 0); - } else if (req->rq_reqmsg->opc == OBD_LOG_CANCEL) { + break; + case LDLM_GL_CALLBACK: + OBD_FAIL_RETURN(OBD_FAIL_LDLM_GL_CALLBACK, 0); + break; + case OBD_LOG_CANCEL: /* remove this eventually - for 1.4.0 compat */ OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0); - } else if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_CREATE) { - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - } else if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_NEXT_BLOCK) { - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - } else if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_READ_HEADER) { - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - } else if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_CLOSE) { - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); - } else { - ldlm_callback_reply(req, -EPROTO); + rc = llog_origin_handle_cancel(req); + ldlm_callback_reply(req, rc); RETURN(0); - } - - LASSERT(req->rq_export != NULL); - LASSERT(req->rq_export->exp_obd != NULL); - - /* FIXME - how to send reply */ - if (req->rq_reqmsg->opc == OBD_LOG_CANCEL) { - int rc = llog_origin_handle_cancel(req); + case OBD_QC_CALLBACK: + OBD_FAIL_RETURN(OBD_FAIL_OBD_QC_CALLBACK_NET, 0); + rc = target_handle_qc_callback(req); ldlm_callback_reply(req, rc); RETURN(0); - } - if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_CREATE) { - int rc = llog_origin_handle_create(req); - req->rq_status = rc; - ptlrpc_reply(req); + case QUOTA_DQACQ: + case QUOTA_DQREL: + /* reply in handler */ + rc = target_handle_dqacq_callback(req); RETURN(0); - } - if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_NEXT_BLOCK) { - int rc = llog_origin_handle_next_block(req); - req->rq_status = rc; - ptlrpc_reply(req); + case LLOG_ORIGIN_HANDLE_CREATE: + OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); + rc = llog_origin_handle_create(req); + ldlm_callback_reply(req, rc); RETURN(0); - } - if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_READ_HEADER) { - int rc = llog_origin_handle_read_header(req); - req->rq_status = rc; - ptlrpc_reply(req); + case LLOG_ORIGIN_HANDLE_NEXT_BLOCK: + OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 0); + rc = llog_origin_handle_next_block(req); + ldlm_callback_reply(req, rc); RETURN(0); - } - if (req->rq_reqmsg->opc == LLOG_ORIGIN_HANDLE_CLOSE) { - int rc = llog_origin_handle_close(req); + case LLOG_ORIGIN_HANDLE_READ_HEADER: + OBD_FAIL_RETURN(OBD_FAIL_OBD_LOGD_NET, 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); + rc = llog_origin_handle_close(req); ldlm_callback_reply(req, rc); RETURN(0); + case SEC_CTX_FINI: + /* do nothing */ + RETURN(0); + default: + CERROR("unknown opcode %u\n", + lustre_msg_get_opc(req->rq_reqmsg)); + ldlm_callback_reply(req, -EPROTO); + RETURN(0); } 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); + dlm_req = lustre_swab_reqbuf(req, DLM_LOCKREQ_OFF, sizeof(*dlm_req), + lustre_swab_ldlm_request); if (dlm_req == NULL) { CERROR ("can't unpack dlm_req\n"); - ldlm_callback_reply (req, -EPROTO); + ldlm_callback_reply(req, -EPROTO); RETURN (0); } - lock = ldlm_handle2lock_ns(ns, &dlm_req->lock_handle1); + lock = ldlm_handle2lock_ns(ns, &dlm_req->lock_handle[0]); if (!lock) { - CDEBUG(D_INODE, "callback on lock "LPX64" - lock disappeared\n", - dlm_req->lock_handle1.cookie); + CDEBUG(D_DLMTRACE, "callback on lock "LPX64" - lock " + "disappeared\n", dlm_req->lock_handle[0].cookie); ldlm_callback_reply(req, -EINVAL); RETURN(0); } /* 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 locks and cache is already droped, + * 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)) { + LDLM_DEBUG(lock, "callback on lock " + LPX64" - lock disappeared\n", + dlm_req->lock_handle[0].cookie); + unlock_res_and_lock(lock); + LDLM_LOCK_PUT(lock); + ldlm_callback_reply(req, -EINVAL); + RETURN(0); + } + 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 * to ost requests (write cache writeback) that might be triggered @@ -951,25 +1556,24 @@ static int ldlm_callback_handler(struct ptlrpc_request *req) * cancelling right now, because it's unused, or have an intent result * in the reply, so we might have to push the responsibility for sending * the reply down into the AST handlers, alas. */ - if (req->rq_reqmsg->opc != LDLM_BL_CALLBACK) - ldlm_callback_reply(req, 0); - switch (req->rq_reqmsg->opc) { + switch (lustre_msg_get_opc(req->rq_reqmsg)) { case LDLM_BL_CALLBACK: CDEBUG(D_INODE, "blocking ast\n"); -#ifdef __KERNEL__ - rc = ldlm_bl_to_thread(ldlm, ns, &dlm_req->lock_desc, lock); - ldlm_callback_reply(req, rc); -#else - rc = 0; - ldlm_callback_reply(req, rc); - ldlm_handle_bl_callback(ns, &dlm_req->lock_desc, lock); -#endif + if (!(lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK)) + ldlm_callback_reply(req, 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"); + 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"); + ldlm_handle_gl_callback(req, ns, dlm_req, lock); + break; default: LBUG(); /* checked above */ } @@ -989,19 +1593,23 @@ static int ldlm_cancel_handler(struct ptlrpc_request *req) if (req->rq_export == NULL) { struct ldlm_request *dlm_req; - CERROR("operation %d with bad export (ptl req %d/rep %d)\n", - req->rq_reqmsg->opc, req->rq_request_portal, - req->rq_reply_portal); - CERROR("--> export cookie: "LPX64"\n", - req->rq_reqmsg->handle.cookie); - dlm_req = lustre_swab_reqbuf(req, 0, sizeof (*dlm_req), + + 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); + + dlm_req = lustre_swab_reqbuf(req, DLM_LOCKREQ_OFF, + sizeof(*dlm_req), lustre_swab_ldlm_request); if (dlm_req != NULL) - ldlm_lock_dump_handle(D_ERROR, &dlm_req->lock_handle1); - RETURN(-ENOTCONN); + 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: @@ -1011,15 +1619,79 @@ static int ldlm_cancel_handler(struct ptlrpc_request *req) if (rc) break; RETURN(0); - + case OBD_LOG_CANCEL: + OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0); + rc = llog_origin_handle_cancel(req); + ldlm_callback_reply(req, rc); + RETURN(0); default: - CERROR("invalid opcode %d\n", req->rq_reqmsg->opc); - RETURN(-EINVAL); + CERROR("invalid opcode %d\n", + lustre_msg_get_opc(req->rq_reqmsg)); + ldlm_callback_reply(req, -EINVAL); } RETURN(0); } +void ldlm_revoke_export_locks(struct obd_export *exp) +{ + struct list_head *locklist = &exp->exp_ldlm_data.led_held_locks; + struct list_head rpc_list; + struct ldlm_lock *lock, *next; + struct ldlm_lock_desc desc; + + ENTRY; + INIT_LIST_HEAD(&rpc_list); + + spin_lock(&exp->exp_ldlm_data.led_lock); + list_for_each_entry_safe(lock, next, locklist, l_export_chain) { + lock_res_and_lock(lock); + + if (lock->l_req_mode != lock->l_granted_mode) { + unlock_res_and_lock(lock); + continue; + } + + LASSERT(lock->l_resource); + if (lock->l_resource->lr_type != LDLM_IBITS && + lock->l_resource->lr_type != LDLM_PLAIN) { + unlock_res_and_lock(lock); + continue; + } + + if (lock->l_flags & LDLM_FL_AST_SENT) { + unlock_res_and_lock(lock); + continue; + } + + LASSERT(lock->l_blocking_ast); + LASSERT(!lock->l_blocking_lock); + + lock->l_flags |= LDLM_FL_AST_SENT; + list_move(&lock->l_export_chain, &rpc_list); + + unlock_res_and_lock(lock); + } + spin_unlock(&exp->exp_ldlm_data.led_lock); + + while (!list_empty(&rpc_list)) { + lock = list_entry(rpc_list.next, struct ldlm_lock, + l_export_chain); + list_del_init(&lock->l_export_chain); + + /* the desc just pretend to exclusive */ + ldlm_lock2desc(lock, &desc); + desc.l_req_mode = LCK_EX; + desc.l_granted_mode = 0; + + LDLM_LOCK_GET(lock); + lock->l_blocking_ast(lock, &desc, lock->l_ast_data, + LDLM_CB_BLOCKING); + LDLM_LOCK_PUT(lock); + } + EXIT; +} + #ifdef __KERNEL__ static struct ldlm_bl_work_item *ldlm_bl_get_work(struct ldlm_bl_pool *blp) { @@ -1045,20 +1717,14 @@ 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; ENTRY; - /* XXX boiler-plate */ { - char name[sizeof(current->comm)]; + char name[CFS_CURPROC_COMM_MAX]; snprintf(name, sizeof(name) - 1, "ldlm_bl_%02d", bltd->bltd_num); - kportal_daemonize(name); + cfs_daemonize(name); } - SIGNAL_MASK_LOCK(current, flags); - sigfillset(¤t->blocked); - RECALC_SIGPENDING; - SIGNAL_MASK_UNLOCK(current, flags); atomic_inc(&blp->blp_num_threads); complete(&blp->blp_comp); @@ -1074,8 +1740,17 @@ static int ldlm_bl_thread_main(void *arg) if (blwi->blwi_ns == NULL) break; - ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld, - blwi->blwi_lock); + if (blwi->blwi_count) { + /* The special case when we cancel locks in lru + * asynchronously, we pass the list of locks here. + * Thus lock is marked LDLM_FL_CANCELING, and already + * canceled locally. */ + ldlm_cli_cancel_list(&blwi->blwi_head, + blwi->blwi_count, NULL, 0, 0); + } else { + ldlm_handle_bl_callback(blwi->blwi_ns, &blwi->blwi_ld, + blwi->blwi_lock); + } OBD_FREE(blwi, sizeof(*blwi)); } @@ -1092,20 +1767,22 @@ static int ldlm_cleanup(int force); int ldlm_get_ref(void) { int rc = 0; - down(&ldlm_ref_sem); + ENTRY; + mutex_down(&ldlm_ref_sem); if (++ldlm_refcount == 1) { rc = ldlm_setup(); if (rc) ldlm_refcount--; } - up(&ldlm_ref_sem); + mutex_up(&ldlm_ref_sem); RETURN(rc); } void ldlm_put_ref(int force) { - down(&ldlm_ref_sem); + ENTRY; + mutex_down(&ldlm_ref_sem); if (ldlm_refcount == 1) { int rc = ldlm_cleanup(force); if (rc) @@ -1115,7 +1792,7 @@ void ldlm_put_ref(int force) } else { ldlm_refcount--; } - up(&ldlm_ref_sem); + mutex_up(&ldlm_ref_sem); EXIT; } @@ -1129,39 +1806,45 @@ static int ldlm_setup(void) #endif ENTRY; - if (ldlm != NULL) + if (ldlm_state != NULL) RETURN(-EALREADY); - OBD_ALLOC(ldlm, sizeof(*ldlm)); - if (ldlm == NULL) + OBD_ALLOC(ldlm_state, sizeof(*ldlm_state)); + if (ldlm_state == NULL) RETURN(-ENOMEM); -#ifdef __KERNEL__ +#ifdef LPROCFS rc = ldlm_proc_setup(); if (rc != 0) GOTO(out_free, rc); #endif - ldlm->ldlm_cb_service = - ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE, - LDLM_MAXREQSIZE, LDLM_CB_REQUEST_PORTAL, - LDLM_CB_REPLY_PORTAL, + ldlm_state->ldlm_cb_service = + ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE, + LDLM_MAXREPSIZE, LDLM_CB_REQUEST_PORTAL, + LDLM_CB_REPLY_PORTAL, ldlm_timeout * 900, ldlm_callback_handler, "ldlm_cbd", - ldlm_svc_proc_dir); + ldlm_svc_proc_dir, NULL, + LDLM_THREADS_AUTO_MIN, LDLM_THREADS_AUTO_MAX, + "ldlm_cb", + LCT_MD_THREAD|LCT_DT_THREAD); - if (!ldlm->ldlm_cb_service) { + if (!ldlm_state->ldlm_cb_service) { CERROR("failed to start service\n"); GOTO(out_proc, rc = -ENOMEM); } - ldlm->ldlm_cancel_service = - ptlrpc_init_svc(LDLM_NEVENTS, LDLM_NBUFS, LDLM_BUFSIZE, - LDLM_MAXREQSIZE, LDLM_CANCEL_REQUEST_PORTAL, - LDLM_CANCEL_REPLY_PORTAL, + ldlm_state->ldlm_cancel_service = + ptlrpc_init_svc(LDLM_NBUFS, LDLM_BUFSIZE, LDLM_MAXREQSIZE, + LDLM_MAXREPSIZE, LDLM_CANCEL_REQUEST_PORTAL, + LDLM_CANCEL_REPLY_PORTAL, ldlm_timeout * 6000, ldlm_cancel_handler, "ldlm_canceld", - ldlm_svc_proc_dir); + ldlm_svc_proc_dir, NULL, + LDLM_THREADS_AUTO_MIN, LDLM_THREADS_AUTO_MAX, + "ldlm_cn", + LCT_MD_THREAD|LCT_DT_THREAD|LCT_CL_THREAD); - if (!ldlm->ldlm_cancel_service) { + if (!ldlm_state->ldlm_cancel_service) { CERROR("failed to start service\n"); GOTO(out_proc, rc = -ENOMEM); } @@ -1169,50 +1852,46 @@ static int ldlm_setup(void) OBD_ALLOC(blp, sizeof(*blp)); if (blp == NULL) GOTO(out_proc, rc = -ENOMEM); - ldlm->ldlm_bl_pool = blp; + ldlm_state->ldlm_bl_pool = blp; atomic_set(&blp->blp_num_threads, 0); - init_waitqueue_head(&blp->blp_waitq); + cfs_waitq_init(&blp->blp_waitq); spin_lock_init(&blp->blp_lock); - INIT_LIST_HEAD(&blp->blp_list); + CFS_INIT_LIST_HEAD(&blp->blp_list); #ifdef __KERNEL__ - for (i = 0; i < LDLM_NUM_THREADS; i++) { + for (i = 0; i < LDLM_BL_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); + rc = cfs_kernel_thread(ldlm_bl_thread_main, &bltd, 0); if (rc < 0) { CERROR("cannot start LDLM thread #%d: rc %d\n", i, rc); - LBUG(); GOTO(out_thread, rc); } wait_for_completion(&blp->blp_comp); } - rc = ptlrpc_start_n_threads(NULL, ldlm->ldlm_cancel_service, - LDLM_NUM_THREADS, "ldlm_cn"); - if (rc) { - LBUG(); + rc = ptlrpc_start_threads(NULL, ldlm_state->ldlm_cancel_service); + if (rc) GOTO(out_thread, rc); - } - rc = ptlrpc_start_n_threads(NULL, ldlm->ldlm_cb_service, - LDLM_NUM_THREADS, "ldlm_cb"); - if (rc) { - LBUG(); + 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); - rc = kernel_thread(expired_lock_main, NULL, CLONE_VM | CLONE_FS); + CFS_INIT_LIST_HEAD(&waiting_locks_list); + spin_lock_init(&waiting_locks_spinlock); + cfs_timer_init(&waiting_locks_timer, waiting_locks_callback, 0); + + 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); @@ -1220,46 +1899,51 @@ static int ldlm_setup(void) 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); #endif +#ifdef __KERNEL__ + rc = ldlm_pools_init(); + if (rc) + GOTO(out_thread, rc); +#endif RETURN(0); #ifdef __KERNEL__ out_thread: - ptlrpc_unregister_service(ldlm->ldlm_cancel_service); - ptlrpc_unregister_service(ldlm->ldlm_cb_service); + ptlrpc_unregister_service(ldlm_state->ldlm_cancel_service); + ptlrpc_unregister_service(ldlm_state->ldlm_cb_service); #endif out_proc: -#ifdef __KERNEL__ +#ifdef LPROCFS ldlm_proc_cleanup(); out_free: #endif - OBD_FREE(ldlm, sizeof(*ldlm)); - ldlm = NULL; + OBD_FREE(ldlm_state, sizeof(*ldlm_state)); + ldlm_state = NULL; return rc; } static int ldlm_cleanup(int force) { #ifdef __KERNEL__ - struct ldlm_bl_pool *blp = ldlm->ldlm_bl_pool; + struct ldlm_bl_pool *blp = ldlm_state->ldlm_bl_pool; #endif ENTRY; - if (!list_empty(&ldlm_namespace_list)) { + if (!list_empty(ldlm_namespace_list(LDLM_NAMESPACE_SERVER)) || + !list_empty(ldlm_namespace_list(LDLM_NAMESPACE_CLIENT))) { CERROR("ldlm still has namespaces; clean these up first.\n"); - ldlm_dump_all_namespaces(); + ldlm_dump_all_namespaces(LDLM_NAMESPACE_SERVER, D_DLMTRACE); + ldlm_dump_all_namespaces(LDLM_NAMESPACE_CLIENT, D_DLMTRACE); RETURN(-EBUSY); } #ifdef __KERNEL__ + ldlm_pools_fini(); +#endif + +#ifdef __KERNEL__ while (atomic_read(&blp->blp_num_threads) > 0) { struct ldlm_bl_work_item blwi = { .blwi_ns = NULL }; @@ -1267,74 +1951,77 @@ static int ldlm_cleanup(int force) spin_lock(&blp->blp_lock); list_add_tail(&blwi.blwi_entry, &blp->blp_list); - wake_up(&blp->blp_waitq); + cfs_waitq_signal(&blp->blp_waitq); spin_unlock(&blp->blp_lock); wait_for_completion(&blp->blp_comp); } OBD_FREE(blp, sizeof(*blp)); - ptlrpc_stop_all_threads(ldlm->ldlm_cb_service); - ptlrpc_unregister_service(ldlm->ldlm_cb_service); - ptlrpc_stop_all_threads(ldlm->ldlm_cancel_service); - ptlrpc_unregister_service(ldlm->ldlm_cancel_service); + ptlrpc_unregister_service(ldlm_state->ldlm_cb_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); + cfs_waitq_signal(&expired_lock_thread.elt_waitq); 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); #endif - OBD_FREE(ldlm, sizeof(*ldlm)); - ldlm = NULL; + OBD_FREE(ldlm_state, sizeof(*ldlm_state)); + ldlm_state = NULL; RETURN(0); } int __init ldlm_init(void) { - ldlm_resource_slab = kmem_cache_create("ldlm_resources", + init_mutex(&ldlm_ref_sem); + init_mutex(ldlm_namespace_lock(LDLM_NAMESPACE_SERVER)); + 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); + SLAB_HWCACHE_ALIGN); if (ldlm_resource_slab == NULL) return -ENOMEM; - ldlm_lock_slab = kmem_cache_create("ldlm_locks", + ldlm_lock_slab = cfs_mem_cache_create("ldlm_locks", sizeof(struct ldlm_lock), 0, - SLAB_HWCACHE_ALIGN, NULL, NULL); + SLAB_HWCACHE_ALIGN); if (ldlm_lock_slab == NULL) { - kmem_cache_destroy(ldlm_resource_slab); + cfs_mem_cache_destroy(ldlm_resource_slab); return -ENOMEM; } - l_lock_init(&ldlm_handle_lock); - 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); - if (kmem_cache_destroy(ldlm_resource_slab) != 0) - CERROR("couldn't free ldlm resource slab\n"); - if (kmem_cache_destroy(ldlm_lock_slab) != 0) - CERROR("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"); + rc = cfs_mem_cache_destroy(ldlm_lock_slab); + LASSERTF(rc == 0, "couldn't free ldlm lock slab\n"); } -/* ldlm_flock.c */ -EXPORT_SYMBOL(ldlm_flock_completion_ast); +/* ldlm_extent.c */ +EXPORT_SYMBOL(ldlm_extent_shift_kms); /* ldlm_lock.c */ +EXPORT_SYMBOL(ldlm_get_processing_policy); EXPORT_SYMBOL(ldlm_lock2desc); EXPORT_SYMBOL(ldlm_register_intent); -EXPORT_SYMBOL(ldlm_unregister_intent); EXPORT_SYMBOL(ldlm_lockname); EXPORT_SYMBOL(ldlm_typename); EXPORT_SYMBOL(ldlm_lock2handle); EXPORT_SYMBOL(__ldlm_handle2lock); +EXPORT_SYMBOL(ldlm_lock_get); EXPORT_SYMBOL(ldlm_lock_put); EXPORT_SYMBOL(ldlm_lock_match); EXPORT_SYMBOL(ldlm_lock_cancel); @@ -1348,58 +2035,77 @@ 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); /* ldlm_request.c */ 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_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_cli_join_lru); 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); /* 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); EXPORT_SYMBOL(ldlm_namespace_cleanup); EXPORT_SYMBOL(ldlm_namespace_free); - -/* l_lock.c */ -EXPORT_SYMBOL(l_lock); -EXPORT_SYMBOL(l_unlock); +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); +EXPORT_SYMBOL(client_import_del_conn); EXPORT_SYMBOL(client_obd_setup); EXPORT_SYMBOL(client_obd_cleanup); EXPORT_SYMBOL(client_connect_import); EXPORT_SYMBOL(client_disconnect_export); -EXPORT_SYMBOL(target_abort_recovery); +EXPORT_SYMBOL(target_start_recovery_thread); +EXPORT_SYMBOL(target_stop_recovery_thread); EXPORT_SYMBOL(target_handle_connect); +EXPORT_SYMBOL(target_cleanup_recovery); EXPORT_SYMBOL(target_destroy_export); 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); -EXPORT_SYMBOL(target_queue_final_reply); -EXPORT_SYMBOL(ldlm_put_lock_into_req); + +/* l_lock.c */ +EXPORT_SYMBOL(lock_res_and_lock); +EXPORT_SYMBOL(unlock_res_and_lock); +