X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lock.c;h=93ccbee37c27b0f4d26b6533e888ae364229345d;hp=8e3490470cc09ae9bda1dbbfca6ac67f9c41bbcf;hb=ae0d69437e35961c257f076da6dcc1842a55456d;hpb=7d3723d80078c0852b9511523960ba5f4590643c diff --git a/lustre/ldlm/ldlm_lock.c b/lustre/ldlm/ldlm_lock.c index 8e34904..93ccbee 100644 --- a/lustre/ldlm/ldlm_lock.c +++ b/lustre/ldlm/ldlm_lock.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -29,8 +27,7 @@ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011 Whamcloud, Inc. - * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -74,10 +71,17 @@ char *ldlm_typename[] = { [LDLM_IBITS] "IBT", }; -static ldlm_policy_wire_to_local_t ldlm_policy_wire_to_local[] = { +static ldlm_policy_wire_to_local_t ldlm_policy_wire18_to_local[] = { + [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_wire_to_local, + [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_wire_to_local, + [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire18_to_local, + [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_wire_to_local, +}; + +static ldlm_policy_wire_to_local_t ldlm_policy_wire21_to_local[] = { [LDLM_PLAIN - LDLM_MIN_TYPE] ldlm_plain_policy_wire_to_local, [LDLM_EXTENT - LDLM_MIN_TYPE] ldlm_extent_policy_wire_to_local, - [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire_to_local, + [LDLM_FLOCK - LDLM_MIN_TYPE] ldlm_flock_policy_wire21_to_local, [LDLM_IBITS - LDLM_MIN_TYPE] ldlm_ibits_policy_wire_to_local, }; @@ -105,13 +109,19 @@ void ldlm_convert_policy_to_wire(ldlm_type_t type, /** * Converts lock policy from on the wire lock_desc format to local format */ -void ldlm_convert_policy_to_local(ldlm_type_t type, +void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type, const ldlm_wire_policy_data_t *wpolicy, ldlm_policy_data_t *lpolicy) { ldlm_policy_wire_to_local_t convert; + int new_client; - convert = ldlm_policy_wire_to_local[type - LDLM_MIN_TYPE]; + /** some badnes for 2.0.0 clients, but 2.0.0 isn't supported */ + new_client = (exp->exp_connect_flags & OBD_CONNECT_FULL20) != 0; + if (new_client) + convert = ldlm_policy_wire21_to_local[type - LDLM_MIN_TYPE]; + else + convert = ldlm_policy_wire18_to_local[type - LDLM_MIN_TYPE]; convert(wpolicy, lpolicy); } @@ -145,12 +155,13 @@ char *ldlm_it2str(int it) extern cfs_mem_cache_t *ldlm_lock_slab; +#ifdef HAVE_SERVER_SUPPORT static ldlm_processing_policy ldlm_processing_policy_table[] = { [LDLM_PLAIN] ldlm_process_plain_lock, [LDLM_EXTENT] ldlm_process_extent_lock, -#ifdef __KERNEL__ +# ifdef __KERNEL__ [LDLM_FLOCK] ldlm_process_flock_lock, -#endif +# endif [LDLM_IBITS] ldlm_process_inodebits_lock, }; @@ -158,6 +169,7 @@ ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res) { return ldlm_processing_policy_table[res->lr_type]; } +#endif /* HAVE_SERVER_SUPPORT */ void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg) { @@ -181,12 +193,6 @@ struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock) return lock; } -static void ldlm_lock_free(struct ldlm_lock *lock, size_t size) -{ - LASSERT(size == sizeof(*lock)); - OBD_SLAB_FREE(lock, ldlm_lock_slab, sizeof(*lock)); -} - void ldlm_lock_put(struct ldlm_lock *lock) { ENTRY; @@ -219,8 +225,7 @@ void ldlm_lock_put(struct ldlm_lock *lock) ldlm_interval_free(ldlm_interval_detach(lock)); lu_ref_fini(&lock->l_reference); - OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle, - ldlm_lock_free); + OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle); } EXIT; @@ -315,13 +320,11 @@ int ldlm_lock_destroy_internal(struct ldlm_lock *lock) if (lock->l_readers || lock->l_writers) { LDLM_ERROR(lock, "lock still has references"); - ldlm_lock_dump(D_ERROR, lock, 0); LBUG(); } if (!cfs_list_empty(&lock->l_res_link)) { LDLM_ERROR(lock, "lock still on resource"); - ldlm_lock_dump(D_ERROR, lock, 0); LBUG(); } @@ -332,10 +335,12 @@ int ldlm_lock_destroy_internal(struct ldlm_lock *lock) } lock->l_destroyed = 1; - 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); + if (lock->l_export && lock->l_export->exp_lock_hash) { + /* NB: it's safe to call cfs_hash_del() even lock isn't + * in exp_lock_hash. */ + cfs_hash_del(lock->l_export->exp_lock_hash, + &lock->l_remote_handle, &lock->l_exp_hash); + } ldlm_lock_remove_from_lru(lock); class_handle_unhash(&lock->l_handle); @@ -389,6 +394,17 @@ static void lock_handle_addref(void *lock) LDLM_LOCK_GET((struct ldlm_lock *)lock); } +static void lock_handle_free(void *lock, int size) +{ + LASSERT(size == sizeof(struct ldlm_lock)); + OBD_SLAB_FREE(lock, ldlm_lock_slab, size); +} + +struct portals_handle_ops lock_handle_ops = { + .hop_addref = lock_handle_addref, + .hop_free = lock_handle_free, +}; + /* * usage: pass in a resource on which you have done ldlm_resource_get * new lock will take over the refcount. @@ -422,11 +438,12 @@ static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource) CFS_INIT_LIST_HEAD(&lock->l_sl_mode); CFS_INIT_LIST_HEAD(&lock->l_sl_policy); CFS_INIT_HLIST_NODE(&lock->l_exp_hash); + CFS_INIT_HLIST_NODE(&lock->l_exp_flock_hash); lprocfs_counter_incr(ldlm_res_to_ns(resource)->ns_stats, LDLM_NSS_LOCKS); CFS_INIT_LIST_HEAD(&lock->l_handle.h_link); - class_handle_hash(&lock->l_handle, lock_handle_addref); + class_handle_hash(&lock->l_handle, &lock_handle_ops); lu_ref_init(&lock->l_reference); lu_ref_add(&lock->l_reference, "hash", lock); @@ -437,6 +454,7 @@ static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource) lock->l_exp_refs_nr = 0; lock->l_exp_refs_target = NULL; #endif + CFS_INIT_LIST_HEAD(&lock->l_exp_list); RETURN(lock); } @@ -935,8 +953,7 @@ static void ldlm_granted_list_add_lock(struct ldlm_lock *lock, check_res_locked(res); ldlm_resource_dump(D_INFO, res); - CDEBUG(D_OTHER, "About to add this lock:\n"); - ldlm_lock_dump(D_OTHER, lock, 0); + LDLM_DEBUG(lock, "About to add lock:"); if (lock->l_destroyed) { CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n"); @@ -1055,7 +1072,7 @@ static struct ldlm_lock *search_queue(cfs_list_t *queue, if (!unref && (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED || - lock->l_fail_value != 0)) + lock->l_failed)) continue; if ((flags & LDLM_FL_LOCAL_ONLY) && @@ -1075,19 +1092,19 @@ static struct ldlm_lock *search_queue(cfs_list_t *queue, return NULL; } -void ldlm_lock_fail_match_locked(struct ldlm_lock *lock, int rc) +void ldlm_lock_fail_match_locked(struct ldlm_lock *lock) { - if (lock->l_fail_value == 0) { - lock->l_fail_value = rc; - cfs_waitq_signal(&lock->l_waitq); + if (!lock->l_failed) { + lock->l_failed = 1; + cfs_waitq_broadcast(&lock->l_waitq); } } EXPORT_SYMBOL(ldlm_lock_fail_match_locked); -void ldlm_lock_fail_match(struct ldlm_lock *lock, int rc) +void ldlm_lock_fail_match(struct ldlm_lock *lock) { lock_res_and_lock(lock); - ldlm_lock_fail_match_locked(lock, rc); + ldlm_lock_fail_match_locked(lock); unlock_res_and_lock(lock); } EXPORT_SYMBOL(ldlm_lock_fail_match); @@ -1095,7 +1112,7 @@ EXPORT_SYMBOL(ldlm_lock_fail_match); void ldlm_lock_allow_match_locked(struct ldlm_lock *lock) { lock->l_flags |= LDLM_FL_LVB_READY; - cfs_waitq_signal(&lock->l_waitq); + cfs_waitq_broadcast(&lock->l_waitq); } void ldlm_lock_allow_match(struct ldlm_lock *lock) @@ -1205,7 +1222,7 @@ ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags, /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */ l_wait_event(lock->l_waitq, lock->l_flags & LDLM_FL_LVB_READY || - lock->l_fail_value != 0, + lock->l_failed, &lwi); if (!(lock->l_flags & LDLM_FL_LVB_READY)) { if (flags & LDLM_FL_TEST_LOCK) @@ -1262,7 +1279,7 @@ ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh, if (lock != NULL) { lock_res_and_lock(lock); if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED || - lock->l_fail_value != 0) + lock->l_failed) GOTO(out, mode); if (lock->l_flags & LDLM_FL_CBPENDING && @@ -1310,7 +1327,7 @@ struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns, lock->l_req_mode = mode; lock->l_ast_data = data; lock->l_pid = cfs_curproc_pid(); - lock->l_ns_srv = ns_is_server(ns); + lock->l_ns_srv = !!ns_is_server(ns); if (cbs) { lock->l_blocking_ast = cbs->lcs_blocking; lock->l_completion_ast = cbs->lcs_completion; @@ -1350,7 +1367,9 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, struct ldlm_lock *lock = *lockp; struct ldlm_resource *res = lock->l_resource; int local = ns_is_client(ldlm_res_to_ns(res)); +#ifdef HAVE_SERVER_SUPPORT ldlm_processing_policy policy; +#endif ldlm_error_t rc = ELDLM_OK; struct ldlm_interval *node = NULL; ENTRY; @@ -1431,6 +1450,7 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, else ldlm_grant_lock(lock, NULL); GOTO(out, ELDLM_OK); +#ifdef HAVE_SERVER_SUPPORT } else if (*flags & LDLM_FL_REPLAY) { if (*flags & LDLM_FL_BLOCK_CONV) { ldlm_resource_add_lock(res, &res->lr_converting, lock); @@ -1448,6 +1468,14 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, policy = ldlm_processing_policy_table[res->lr_type]; policy(lock, flags, 1, &rc, NULL); GOTO(out, rc); +#else + } else { + CERROR("This is client-side-only module, cannot handle " + "LDLM_NAMESPACE_SERVER resource type lock.\n"); + LBUG(); + } +#endif + out: unlock_res_and_lock(lock); if (node) @@ -1455,6 +1483,7 @@ out: return rc; } +#ifdef HAVE_SERVER_SUPPORT /* Must be called with namespace taken: queue is waiting or converting. */ int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue, cfs_list_t *work_list) @@ -1485,154 +1514,192 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue, RETURN(rc); } +#endif static int -ldlm_work_bl_ast_lock(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg) +ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { - struct ldlm_lock_desc d; - struct ldlm_lock *lock = cfs_list_entry(tmp, struct ldlm_lock, - l_bl_ast); - int rc; - ENTRY; + struct ldlm_cb_set_arg *arg = opaq; + struct ldlm_lock_desc d; + int rc; + struct ldlm_lock *lock; + ENTRY; - /* nobody should touch l_bl_ast */ - lock_res_and_lock(lock); - cfs_list_del_init(&lock->l_bl_ast); + if (cfs_list_empty(arg->list)) + RETURN(-ENOENT); - LASSERT(lock->l_flags & LDLM_FL_AST_SENT); - LASSERT(lock->l_bl_ast_run == 0); - LASSERT(lock->l_blocking_lock); - lock->l_bl_ast_run++; - unlock_res_and_lock(lock); + lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_bl_ast); - ldlm_lock2desc(lock->l_blocking_lock, &d); + /* nobody should touch l_bl_ast */ + lock_res_and_lock(lock); + cfs_list_del_init(&lock->l_bl_ast); - rc = lock->l_blocking_ast(lock, &d, (void *)arg, - LDLM_CB_BLOCKING); - LDLM_LOCK_RELEASE(lock->l_blocking_lock); - lock->l_blocking_lock = NULL; - LDLM_LOCK_RELEASE(lock); + LASSERT(lock->l_flags & LDLM_FL_AST_SENT); + LASSERT(lock->l_bl_ast_run == 0); + LASSERT(lock->l_blocking_lock); + lock->l_bl_ast_run++; + unlock_res_and_lock(lock); - RETURN(rc); -} + ldlm_lock2desc(lock->l_blocking_lock, &d); -static int -ldlm_work_cp_ast_lock(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg) -{ - struct ldlm_lock *lock = cfs_list_entry(tmp, struct ldlm_lock, l_cp_ast); - ldlm_completion_callback completion_callback; - int rc = 0; - ENTRY; - - /* It's possible to receive a completion AST before we've set - * the l_completion_ast pointer: either because the AST arrived - * before the reply, or simply because there's a small race - * window between receiving the reply and finishing the local - * enqueue. (bug 842) - * - * This can't happen with the blocking_ast, however, because we - * will never call the local blocking_ast until we drop our - * reader/writer reference, which we won't do until we get the - * reply and finish enqueueing. */ + rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING); + LDLM_LOCK_RELEASE(lock->l_blocking_lock); + lock->l_blocking_lock = NULL; + LDLM_LOCK_RELEASE(lock); - /* nobody should touch l_cp_ast */ - lock_res_and_lock(lock); - cfs_list_del_init(&lock->l_cp_ast); - LASSERT(lock->l_flags & LDLM_FL_CP_REQD); - /* save l_completion_ast since it can be changed by - * mds_intent_policy(), see bug 14225 */ - completion_callback = lock->l_completion_ast; - lock->l_flags &= ~LDLM_FL_CP_REQD; - unlock_res_and_lock(lock); - - if (completion_callback != NULL) - rc = completion_callback(lock, 0, (void *)arg); - LDLM_LOCK_RELEASE(lock); + RETURN(rc); +} - RETURN(rc); +static int +ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) +{ + struct ldlm_cb_set_arg *arg = opaq; + int rc = 0; + struct ldlm_lock *lock; + ldlm_completion_callback completion_callback; + ENTRY; + + if (cfs_list_empty(arg->list)) + RETURN(-ENOENT); + + lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_cp_ast); + + /* It's possible to receive a completion AST before we've set + * the l_completion_ast pointer: either because the AST arrived + * before the reply, or simply because there's a small race + * window between receiving the reply and finishing the local + * enqueue. (bug 842) + * + * This can't happen with the blocking_ast, however, because we + * will never call the local blocking_ast until we drop our + * reader/writer reference, which we won't do until we get the + * reply and finish enqueueing. */ + + /* nobody should touch l_cp_ast */ + lock_res_and_lock(lock); + cfs_list_del_init(&lock->l_cp_ast); + LASSERT(lock->l_flags & LDLM_FL_CP_REQD); + /* save l_completion_ast since it can be changed by + * mds_intent_policy(), see bug 14225 */ + completion_callback = lock->l_completion_ast; + lock->l_flags &= ~LDLM_FL_CP_REQD; + unlock_res_and_lock(lock); + + if (completion_callback != NULL) + rc = completion_callback(lock, 0, (void *)arg); + LDLM_LOCK_RELEASE(lock); + + RETURN(rc); } static int -ldlm_work_revoke_ast_lock(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg) +ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { - struct ldlm_lock_desc desc; - struct ldlm_lock *lock = cfs_list_entry(tmp, struct ldlm_lock, - l_rk_ast); - int rc; - ENTRY; + struct ldlm_cb_set_arg *arg = opaq; + struct ldlm_lock_desc desc; + int rc; + struct ldlm_lock *lock; + ENTRY; - cfs_list_del_init(&lock->l_rk_ast); + if (cfs_list_empty(arg->list)) + RETURN(-ENOENT); - /* the desc just pretend to exclusive */ - ldlm_lock2desc(lock, &desc); - desc.l_req_mode = LCK_EX; - desc.l_granted_mode = 0; + lock = cfs_list_entry(arg->list->next, struct ldlm_lock, l_rk_ast); + cfs_list_del_init(&lock->l_rk_ast); - rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING); - LDLM_LOCK_RELEASE(lock); + /* the desc just pretend to exclusive */ + ldlm_lock2desc(lock, &desc); + desc.l_req_mode = LCK_EX; + desc.l_granted_mode = 0; - RETURN(rc); + rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING); + LDLM_LOCK_RELEASE(lock); + + RETURN(rc); } -int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list, - ldlm_desc_ast_t ast_type) +int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { - struct l_wait_info lwi = { 0 }; - struct ldlm_cb_set_arg *arg; - cfs_list_t *tmp, *pos; - int (*work_ast_lock)(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg); - unsigned int max_ast_count; - int rc; - ENTRY; - - if (cfs_list_empty(rpc_list)) - RETURN(0); + struct ldlm_cb_set_arg *arg = opaq; + struct ldlm_glimpse_work *gl_work; + struct ldlm_lock *lock; + int rc = 0; + ENTRY; - OBD_ALLOC_PTR(arg); - if (arg == NULL) - RETURN(-ENOMEM); + if (cfs_list_empty(arg->list)) + RETURN(-ENOENT); - cfs_atomic_set(&arg->restart, 0); - cfs_atomic_set(&arg->rpcs, 0); - cfs_atomic_set(&arg->refcount, 1); - cfs_waitq_init(&arg->waitq); + gl_work = cfs_list_entry(arg->list->next, struct ldlm_glimpse_work, + gl_list); + cfs_list_del_init(&gl_work->gl_list); - switch (ast_type) { - case LDLM_WORK_BL_AST: - arg->type = LDLM_BL_CALLBACK; - work_ast_lock = ldlm_work_bl_ast_lock; - break; - case LDLM_WORK_CP_AST: - arg->type = LDLM_CP_CALLBACK; - work_ast_lock = ldlm_work_cp_ast_lock; - break; - case LDLM_WORK_REVOKE_AST: - arg->type = LDLM_BL_CALLBACK; - work_ast_lock = ldlm_work_revoke_ast_lock; - break; - default: - LBUG(); - } + lock = gl_work->gl_lock; + if (lock->l_glimpse_ast(lock, (void*)arg) == 0) + rc = 1; - max_ast_count = ns->ns_max_parallel_ast ? : UINT_MAX; - arg->threshold = max_ast_count; + LDLM_LOCK_RELEASE(lock); - cfs_list_for_each_safe(tmp, pos, rpc_list) { - (void)work_ast_lock(tmp, arg); - if (cfs_atomic_read(&arg->rpcs) < max_ast_count) - continue; - - l_wait_event(arg->waitq, - cfs_atomic_read(&arg->rpcs) < arg->threshold, - &lwi); - } + if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0) + OBD_FREE_PTR(gl_work); - arg->threshold = 1; - l_wait_event(arg->waitq, cfs_atomic_read(&arg->rpcs) == 0, &lwi); + RETURN(rc); +} - rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0; - ldlm_csa_put(arg); - RETURN(rc); +int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list, + ldlm_desc_ast_t ast_type) +{ + struct ldlm_cb_set_arg *arg; + set_producer_func work_ast_lock; + int rc; + + if (cfs_list_empty(rpc_list)) + RETURN(0); + + OBD_ALLOC_PTR(arg); + if (arg == NULL) + RETURN(-ENOMEM); + + cfs_atomic_set(&arg->restart, 0); + arg->list = rpc_list; + + switch (ast_type) { + case LDLM_WORK_BL_AST: + arg->type = LDLM_BL_CALLBACK; + work_ast_lock = ldlm_work_bl_ast_lock; + break; + case LDLM_WORK_CP_AST: + arg->type = LDLM_CP_CALLBACK; + work_ast_lock = ldlm_work_cp_ast_lock; + break; + case LDLM_WORK_REVOKE_AST: + arg->type = LDLM_BL_CALLBACK; + work_ast_lock = ldlm_work_revoke_ast_lock; + break; + case LDLM_WORK_GL_AST: + arg->type = LDLM_GL_CALLBACK; + work_ast_lock = ldlm_work_gl_ast_lock; + break; + default: + LBUG(); + } + + /* We create a ptlrpc request set with flow control extension. + * This request set will use the work_ast_lock function to produce new + * requests and will send a new request each time one completes in order + * to keep the number of requests in flight to ns_max_parallel_ast */ + arg->set = ptlrpc_prep_fcset(ns->ns_max_parallel_ast ? : UINT_MAX, + work_ast_lock, arg); + if (arg->set == NULL) + GOTO(out, rc = -ENOMEM); + + ptlrpc_set_wait(arg->set); + ptlrpc_set_destroy(arg->set); + + rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0; + GOTO(out, rc); +out: + OBD_FREE_PTR(arg); + return rc; } static int reprocess_one_queue(struct ldlm_resource *res, void *closure) @@ -1666,16 +1733,17 @@ void ldlm_reprocess_all_ns(struct ldlm_namespace *ns) void ldlm_reprocess_all(struct ldlm_resource *res) { CFS_LIST_HEAD(rpc_list); + +#ifdef HAVE_SERVER_SUPPORT int rc; ENTRY; - /* Local lock trees don't get reprocessed. */ if (ns_is_client(ldlm_res_to_ns(res))) { EXIT; return; } - restart: +restart: lock_res(res); rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list); if (rc == LDLM_ITER_CONTINUE) @@ -1688,6 +1756,14 @@ void ldlm_reprocess_all(struct ldlm_resource *res) LASSERT(cfs_list_empty(&rpc_list)); goto restart; } +#else + ENTRY; + if (!ns_is_client(ldlm_res_to_ns(res))) { + CERROR("This is client-side-only module, cannot handle " + "LDLM_NAMESPACE_SERVER resource type lock.\n"); + LBUG(); + } +#endif EXIT; } @@ -1737,14 +1813,17 @@ void ldlm_lock_cancel(struct ldlm_lock *lock) LBUG(); } - ldlm_del_waiting_lock(lock); + if (lock->l_waited) + ldlm_del_waiting_lock(lock); /* Releases cancel callback. */ ldlm_cancel_callback(lock); /* Yes, second time, just in case it was added again while we were running with no res lock in ldlm_cancel_callback */ - ldlm_del_waiting_lock(lock); + if (lock->l_waited) + ldlm_del_waiting_lock(lock); + ldlm_resource_unlink_lock(lock); ldlm_lock_destroy_nolock(lock); @@ -1776,11 +1855,17 @@ int ldlm_lock_set_data(struct lustre_handle *lockh, void *data) } EXPORT_SYMBOL(ldlm_lock_set_data); +struct export_cl_data { + struct obd_export *ecl_exp; + int ecl_loop; +}; + int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd, cfs_hlist_node_t *hnode, void *data) { - struct obd_export *exp = data; + struct export_cl_data *ecl = (struct export_cl_data *)data; + struct obd_export *exp = ecl->ecl_exp; struct ldlm_lock *lock = cfs_hash_object(hs, hnode); struct ldlm_resource *res; @@ -1793,13 +1878,28 @@ int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd, ldlm_reprocess_all(res); ldlm_resource_putref(res); LDLM_LOCK_RELEASE(lock); - return 0; + + ecl->ecl_loop++; + if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) { + CDEBUG(D_INFO, + "Cancel lock %p for export %p (loop %d), still have " + "%d locks left on hash table.\n", + lock, exp, ecl->ecl_loop, + cfs_atomic_read(&hs->hs_count)); + } + + return 0; } void ldlm_cancel_locks_for_export(struct obd_export *exp) { - cfs_hash_for_each_empty(exp->exp_lock_hash, - ldlm_cancel_locks_for_export_cb, exp); + struct export_cl_data ecl = { + .ecl_exp = exp, + .ecl_loop = 0, + }; + + cfs_hash_for_each_empty(exp->exp_lock_hash, + ldlm_cancel_locks_for_export_cb, &ecl); } /** @@ -1841,9 +1941,10 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, struct ldlm_resource *res; struct ldlm_namespace *ns; int granted = 0; - int old_mode, rc; - struct sl_insert_point prev; - ldlm_error_t err; +#ifdef HAVE_SERVER_SUPPORT + int old_mode; + struct sl_insert_point prev; +#endif struct ldlm_interval *node; ENTRY; @@ -1866,15 +1967,19 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, res = lock->l_resource; ns = ldlm_res_to_ns(res); - old_mode = lock->l_req_mode; - lock->l_req_mode = new_mode; - if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) { - /* remember the lock position where the lock might be - * added back to the granted list later and also - * remember the join mode for skiplist fixing. */ - prev.res_link = lock->l_res_link.prev; - prev.mode_link = lock->l_sl_mode.prev; - prev.policy_link = lock->l_sl_policy.prev; +#ifdef HAVE_SERVER_SUPPORT + old_mode = lock->l_req_mode; +#endif + lock->l_req_mode = new_mode; + if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) { +#ifdef HAVE_SERVER_SUPPORT + /* remember the lock position where the lock might be + * added back to the granted list later and also + * remember the join mode for skiplist fixing. */ + prev.res_link = lock->l_res_link.prev; + prev.mode_link = lock->l_sl_mode.prev; + prev.policy_link = lock->l_sl_policy.prev; +#endif ldlm_resource_unlink_lock(lock); } else { ldlm_resource_unlink_lock(lock); @@ -1911,7 +2016,10 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, if (lock->l_completion_ast) lock->l_completion_ast(lock, 0, NULL); } +#ifdef HAVE_SERVER_SUPPORT } else { + int rc; + ldlm_error_t err; int pflags = 0; ldlm_processing_policy policy; policy = ldlm_processing_policy_table[res->lr_type]; @@ -1929,6 +2037,13 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, granted = 1; } } +#else + } else { + CERROR("This is client-side-only module, cannot handle " + "LDLM_NAMESPACE_SERVER resource type lock.\n"); + LBUG(); + } +#endif unlock_res_and_lock(lock); if (granted) @@ -1938,61 +2053,6 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, RETURN(res); } -void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos) -{ - struct obd_device *obd = NULL; - - if (!((libcfs_debug | D_ERROR) & level)) - return; - - if (!lock) { - CDEBUG(level, " NULL LDLM lock\n"); - return; - } - - CDEBUG(level," -- Lock dump: %p/"LPX64" (rc: %d) (pos: %d) (pid: %d)\n", - lock, lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc), - pos, lock->l_pid); - if (lock->l_conn_export != NULL) - obd = lock->l_conn_export->exp_obd; - if (lock->l_export && lock->l_export->exp_connection) { - CDEBUG(level, " Node: NID %s (rhandle: "LPX64")\n", - libcfs_nid2str(lock->l_export->exp_connection->c_peer.nid), - lock->l_remote_handle.cookie); - } else if (obd == NULL) { - CDEBUG(level, " Node: local\n"); - } else { - struct obd_import *imp = obd->u.cli.cl_import; - CDEBUG(level, " Node: NID %s (rhandle: "LPX64")\n", - libcfs_nid2str(imp->imp_connection->c_peer.nid), - lock->l_remote_handle.cookie); - } - CDEBUG(level, " Resource: %p ("LPU64"/"LPU64"/"LPU64")\n", - lock->l_resource, - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], - lock->l_resource->lr_name.name[2]); - CDEBUG(level, " Req mode: %s, grant mode: %s, rc: %u, read: %d, " - "write: %d flags: "LPX64"\n", ldlm_lockname[lock->l_req_mode], - ldlm_lockname[lock->l_granted_mode], - cfs_atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers, - lock->l_flags); - if (lock->l_resource->lr_type == LDLM_EXTENT) - CDEBUG(level, " Extent: "LPU64" -> "LPU64 - " (req "LPU64"-"LPU64")\n", - lock->l_policy_data.l_extent.start, - lock->l_policy_data.l_extent.end, - lock->l_req_extent.start, lock->l_req_extent.end); - else if (lock->l_resource->lr_type == LDLM_FLOCK) - CDEBUG(level, " Pid: %d Extent: "LPU64" -> "LPU64"\n", - lock->l_policy_data.l_flock.pid, - lock->l_policy_data.l_flock.start, - lock->l_policy_data.l_flock.end); - else if (lock->l_resource->lr_type == LDLM_IBITS) - CDEBUG(level, " Bits: "LPX64"\n", - lock->l_policy_data.l_inodebits.bits); -} - void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh) { struct ldlm_lock *lock; @@ -2004,95 +2064,98 @@ void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh) if (lock == NULL) return; - ldlm_lock_dump(D_OTHER, lock, 0); + LDLM_DEBUG_LIMIT(level, lock, "###"); LDLM_LOCK_PUT(lock); } -void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level, - struct libcfs_debug_msg_data *data, const char *fmt, - ...) +void _ldlm_lock_debug(struct ldlm_lock *lock, + struct libcfs_debug_msg_data *msgdata, + const char *fmt, ...) { va_list args; - cfs_debug_limit_state_t *cdls = data->msg_cdls; + struct obd_export *exp = lock->l_export; + struct ldlm_resource *resource = lock->l_resource; + char *nid = "local"; va_start(args, fmt); - if (lock->l_resource == NULL) { - libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file, - data->msg_fn, data->msg_line, fmt, args, + if (exp && exp->exp_connection) { + nid = libcfs_nid2str(exp->exp_connection->c_peer.nid); + } else if (exp && exp->exp_obd != NULL) { + struct obd_import *imp = exp->exp_obd->u.cli.cl_import; + nid = libcfs_nid2str(imp->imp_connection->c_peer.nid); + } + + if (resource == NULL) { + libcfs_debug_vmsg2(msgdata, fmt, args, " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " - "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" remote: " - LPX64" expref: %d pid: %u timeout: %lu\n", lock, + "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s " + "remote: "LPX64" expref: %d pid: %u timeout: %lu\n", + lock, lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers, ldlm_lockname[lock->l_granted_mode], ldlm_lockname[lock->l_req_mode], - lock->l_flags, lock->l_remote_handle.cookie, - lock->l_export ? - cfs_atomic_read(&lock->l_export->exp_refcount) : -99, + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? cfs_atomic_read(&exp->exp_refcount) : -99, lock->l_pid, lock->l_callback_timeout); va_end(args); return; } - switch (lock->l_resource->lr_type) { + switch (resource->lr_type) { case LDLM_EXTENT: - libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file, - data->msg_fn, data->msg_line, fmt, args, + libcfs_debug_vmsg2(msgdata, fmt, args, " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " "res: "LPU64"/"LPU64" rrc: %d type: %s ["LPU64"->"LPU64 - "] (req "LPU64"->"LPU64") flags: "LPX64" remote: "LPX64 - " expref: %d pid: %u timeout %lu\n", + "] (req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote:" + " "LPX64" expref: %d pid: %u timeout %lu\n", ldlm_lock_to_ns_name(lock), lock, lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers, ldlm_lockname[lock->l_granted_mode], ldlm_lockname[lock->l_req_mode], - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], - cfs_atomic_read(&lock->l_resource->lr_refcount), - ldlm_typename[lock->l_resource->lr_type], + resource->lr_name.name[0], + resource->lr_name.name[1], + cfs_atomic_read(&resource->lr_refcount), + ldlm_typename[resource->lr_type], lock->l_policy_data.l_extent.start, lock->l_policy_data.l_extent.end, lock->l_req_extent.start, lock->l_req_extent.end, - lock->l_flags, lock->l_remote_handle.cookie, - lock->l_export ? - cfs_atomic_read(&lock->l_export->exp_refcount) : -99, + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? cfs_atomic_read(&exp->exp_refcount) : -99, lock->l_pid, lock->l_callback_timeout); break; case LDLM_FLOCK: - libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file, - data->msg_fn, data->msg_line, fmt, args, + libcfs_debug_vmsg2(msgdata, fmt, args, " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " "res: "LPU64"/"LPU64" rrc: %d type: %s pid: %d " - "["LPU64"->"LPU64"] flags: "LPX64" remote: "LPX64 + "["LPU64"->"LPU64"] flags: "LPX64" nid: %s remote: "LPX64 " expref: %d pid: %u timeout: %lu\n", ldlm_lock_to_ns_name(lock), lock, lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc), lock->l_readers, lock->l_writers, ldlm_lockname[lock->l_granted_mode], ldlm_lockname[lock->l_req_mode], - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], - cfs_atomic_read(&lock->l_resource->lr_refcount), - ldlm_typename[lock->l_resource->lr_type], + resource->lr_name.name[0], + resource->lr_name.name[1], + cfs_atomic_read(&resource->lr_refcount), + ldlm_typename[resource->lr_type], lock->l_policy_data.l_flock.pid, lock->l_policy_data.l_flock.start, lock->l_policy_data.l_flock.end, - lock->l_flags, lock->l_remote_handle.cookie, - lock->l_export ? - cfs_atomic_read(&lock->l_export->exp_refcount) : -99, + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? cfs_atomic_read(&exp->exp_refcount) : -99, lock->l_pid, lock->l_callback_timeout); break; case LDLM_IBITS: - libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file, - data->msg_fn, data->msg_line, fmt, args, + libcfs_debug_vmsg2(msgdata, fmt, args, " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " "res: "LPU64"/"LPU64" bits "LPX64" rrc: %d type: %s " - "flags: "LPX64" remote: "LPX64" expref: %d " + "flags: "LPX64" nid: %s remote: "LPX64" expref: %d " "pid: %u timeout: %lu\n", ldlm_lock_to_ns_name(lock), lock, lock->l_handle.h_cookie, @@ -2100,36 +2163,34 @@ void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level, lock->l_readers, lock->l_writers, ldlm_lockname[lock->l_granted_mode], ldlm_lockname[lock->l_req_mode], - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], + resource->lr_name.name[0], + resource->lr_name.name[1], lock->l_policy_data.l_inodebits.bits, - cfs_atomic_read(&lock->l_resource->lr_refcount), - ldlm_typename[lock->l_resource->lr_type], - lock->l_flags, lock->l_remote_handle.cookie, - lock->l_export ? - cfs_atomic_read(&lock->l_export->exp_refcount) : -99, + cfs_atomic_read(&resource->lr_refcount), + ldlm_typename[resource->lr_type], + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? cfs_atomic_read(&exp->exp_refcount) : -99, lock->l_pid, lock->l_callback_timeout); break; default: - libcfs_debug_vmsg2(cdls, data->msg_subsys, level,data->msg_file, - data->msg_fn, data->msg_line, fmt, args, + libcfs_debug_vmsg2(msgdata, fmt, args, " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " "res: "LPU64"/"LPU64" rrc: %d type: %s flags: "LPX64" " - "remote: "LPX64" expref: %d pid: %u timeout %lu\n", + "nid: %s remote: "LPX64" expref: %d pid: %u timeout %lu" + "\n", ldlm_lock_to_ns_name(lock), lock, lock->l_handle.h_cookie, cfs_atomic_read (&lock->l_refc), lock->l_readers, lock->l_writers, ldlm_lockname[lock->l_granted_mode], ldlm_lockname[lock->l_req_mode], - lock->l_resource->lr_name.name[0], - lock->l_resource->lr_name.name[1], - cfs_atomic_read(&lock->l_resource->lr_refcount), - ldlm_typename[lock->l_resource->lr_type], - lock->l_flags, lock->l_remote_handle.cookie, - lock->l_export ? - cfs_atomic_read(&lock->l_export->exp_refcount) : -99, + resource->lr_name.name[0], + resource->lr_name.name[1], + cfs_atomic_read(&resource->lr_refcount), + ldlm_typename[resource->lr_type], + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? cfs_atomic_read(&exp->exp_refcount) : -99, lock->l_pid, lock->l_callback_timeout); break; }