X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_resource.c;h=2174271ff73f13cf13b76d1c7267ddc51f605b8c;hp=0d34e4080d567238261c6407489f7f48e361aab1;hb=702258c220a8d9224413f259af4a21dda9950278;hpb=864669f7853a3325b1433832541d47d71fe0d19d diff --git a/lustre/ldlm/ldlm_resource.c b/lustre/ldlm/ldlm_resource.c index 0d34e40..2174271 100644 --- a/lustre/ldlm/ldlm_resource.c +++ b/lustre/ldlm/ldlm_resource.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. @@ -28,6 +26,8 @@ /* * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2010, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -55,10 +55,10 @@ cfs_mem_cache_t *ldlm_resource_slab, *ldlm_lock_slab; cfs_atomic_t ldlm_srv_namespace_nr = CFS_ATOMIC_INIT(0); cfs_atomic_t ldlm_cli_namespace_nr = CFS_ATOMIC_INIT(0); -cfs_semaphore_t ldlm_srv_namespace_lock; +struct mutex ldlm_srv_namespace_lock; CFS_LIST_HEAD(ldlm_srv_namespace_list); -cfs_semaphore_t ldlm_cli_namespace_lock; +struct mutex ldlm_cli_namespace_lock; CFS_LIST_HEAD(ldlm_cli_namespace_list); cfs_proc_dir_entry_t *ldlm_type_proc_dir = NULL; @@ -67,6 +67,10 @@ cfs_proc_dir_entry_t *ldlm_svc_proc_dir = NULL; extern unsigned int ldlm_cancel_unused_locks_before_replay; +/* during debug dump certain amount of granted locks for one resource to avoid + * DDOS. */ +unsigned int ldlm_dump_granted_max = 256; + #ifdef LPROCFS static int ldlm_proc_dump_ns(struct file *file, const char *buffer, unsigned long count, void *data) @@ -81,6 +85,9 @@ int ldlm_proc_setup(void) int rc; struct lprocfs_vars list[] = { { "dump_namespaces", NULL, ldlm_proc_dump_ns, NULL }, + { "dump_granted_max", + lprocfs_rd_uint, lprocfs_wr_uint, + &ldlm_dump_granted_max, NULL }, { "cancel_unused_locks_before_replay", lprocfs_rd_uint, lprocfs_wr_uint, &ldlm_cancel_unused_locks_before_replay, NULL }, @@ -249,8 +256,8 @@ static int lprocfs_wr_lru_size(struct file *file, const char *buffer, ns->ns_max_unused = (unsigned int)tmp; ldlm_cancel_lru(ns, 0, LDLM_ASYNC, LDLM_CANCEL_PASSED); - /* Make sure that originally lru resize was supported before - * turning it on here. */ + /* Make sure that LRU resize was originally supported before + * turning it on here. */ if (lru_resize && (ns->ns_orig_connect_flags & OBD_CONNECT_LRU_RESIZE)) { CDEBUG(D_DLMTRACE, @@ -263,6 +270,33 @@ static int lprocfs_wr_lru_size(struct file *file, const char *buffer, return count; } +static int lprocfs_rd_elc(char *page, char **start, off_t off, + int count, int *eof, void *data) +{ + struct ldlm_namespace *ns = data; + unsigned int supp = ns_connect_cancelset(ns); + + return lprocfs_rd_uint(page, start, off, count, eof, &supp); +} + +static int lprocfs_wr_elc(struct file *file, const char *buffer, + unsigned long count, void *data) +{ + struct ldlm_namespace *ns = data; + unsigned int supp = -1; + int rc; + + rc = lprocfs_wr_uint(file, buffer, count, &supp); + if (rc < 0) + return rc; + + if (supp == 0) + ns->ns_connect_flags &= ~OBD_CONNECT_CANCELSET; + else if (ns->ns_orig_connect_flags & OBD_CONNECT_CANCELSET) + ns->ns_connect_flags |= OBD_CONNECT_CANCELSET; + return count; +} + void ldlm_namespace_proc_unregister(struct ldlm_namespace *ns) { struct proc_dir_entry *dir; @@ -331,6 +365,13 @@ int ldlm_namespace_proc_register(struct ldlm_namespace *ns) lock_vars[0].read_fptr = lprocfs_rd_uint; lock_vars[0].write_fptr = lprocfs_wr_uint; lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0); + + snprintf(lock_name, MAX_STRING_SIZE, "%s/early_lock_cancel", + ldlm_ns_name(ns)); + lock_vars[0].data = ns; + lock_vars[0].read_fptr = lprocfs_rd_elc; + lock_vars[0].write_fptr = lprocfs_wr_elc; + lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0); } else { snprintf(lock_name, MAX_STRING_SIZE, "%s/ctime_age_limit", ldlm_ns_name(ns)); @@ -365,6 +406,13 @@ int ldlm_namespace_proc_register(struct ldlm_namespace *ns) lock_vars[0].read_fptr = lprocfs_rd_uint; lock_vars[0].write_fptr = lprocfs_wr_uint; lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0); + + snprintf(lock_name, MAX_STRING_SIZE, "%s/max_parallel_ast", + ldlm_ns_name(ns)); + lock_vars[0].data = &ns->ns_max_parallel_ast; + lock_vars[0].read_fptr = lprocfs_rd_uint; + lock_vars[0].write_fptr = lprocfs_wr_uint; + lprocfs_add_vars(ldlm_ns_proc_dir, lock_vars, 0); } return 0; } @@ -376,9 +424,10 @@ int ldlm_namespace_proc_register(struct ldlm_namespace *ns) #endif /* LPROCFS */ -static unsigned ldlm_res_hop_hash(cfs_hash_t *hs, void *key, unsigned mask) +static unsigned ldlm_res_hop_hash(cfs_hash_t *hs, + const void *key, unsigned mask) { - struct ldlm_res_id *id = key; + const struct ldlm_res_id *id = key; unsigned val = 0; unsigned i; @@ -387,23 +436,30 @@ static unsigned ldlm_res_hop_hash(cfs_hash_t *hs, void *key, unsigned mask) return val & mask; } -static unsigned ldlm_res_hop_fid_hash(cfs_hash_t *hs, void *key, unsigned mask) +static unsigned ldlm_res_hop_fid_hash(cfs_hash_t *hs, + const void *key, unsigned mask) { - struct ldlm_res_id *id = key; + const struct ldlm_res_id *id = key; struct lu_fid fid; - __u64 hash; + __u32 hash; + __u32 val; fid.f_seq = id->name[LUSTRE_RES_ID_SEQ_OFF]; - fid.f_oid = (__u32)id->name[LUSTRE_RES_ID_OID_OFF]; - fid.f_ver = (__u32)id->name[LUSTRE_RES_ID_VER_OFF]; - - hash = fid_flatten(&fid); + fid.f_oid = (__u32)id->name[LUSTRE_RES_ID_VER_OID_OFF]; + fid.f_ver = (__u32)(id->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32); + + hash = fid_flatten32(&fid); + hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */ + if (id->name[LUSTRE_RES_ID_HSH_OFF] != 0) { + val = id->name[LUSTRE_RES_ID_HSH_OFF]; + hash += (val >> 5) + (val << 11); + } else { + val = fid_oid(&fid); + } hash = cfs_hash_long(hash, hs->hs_bkt_bits); - /* ignore a few low bits */ - if (id->name[LUSTRE_RES_ID_HSH_OFF] != 0) - hash += id->name[LUSTRE_RES_ID_HSH_OFF] >> 5; - else - hash = hash >> 5; + /* give me another random factor */ + hash -= cfs_hash_long((unsigned long)hs, val % 11 + 3); + hash <<= hs->hs_cur_bits - hs->hs_bkt_bits; hash |= ldlm_res_hop_hash(hs, key, CFS_HASH_NBKT(hs) - 1); @@ -418,13 +474,7 @@ static void *ldlm_res_hop_key(cfs_hlist_node_t *hnode) return &res->lr_name; } -static int ldlm_res_eq(const struct ldlm_res_id *res0, - const struct ldlm_res_id *res1) -{ - return !memcmp(res0, res1, sizeof(*res0)); -} - -static int ldlm_res_hop_keycmp(void *key, cfs_hlist_node_t *hnode) +static int ldlm_res_hop_keycmp(const void *key, cfs_hlist_node_t *hnode) { struct ldlm_resource *res; @@ -444,7 +494,6 @@ static void ldlm_res_hop_get_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode) res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash); ldlm_resource_getref(res); - LDLM_RESOURCE_ADDREF(res); } static void ldlm_res_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode) @@ -453,7 +502,6 @@ static void ldlm_res_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode) res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash); /* cfs_hash_for_each_nolock is the only chance we call it */ - LDLM_RESOURCE_DELREF(res); ldlm_resource_putref_locked(res); } @@ -462,7 +510,6 @@ static void ldlm_res_hop_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode) struct ldlm_resource *res; res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash); - LDLM_RESOURCE_DELREF(res); ldlm_resource_putref(res); } @@ -503,7 +550,7 @@ ldlm_ns_hash_def_t ldlm_ns_hash_defs[] = { .nsd_type = LDLM_NS_TYPE_MDC, .nsd_bkt_bits = 11, - .nsd_all_bits = 15, + .nsd_all_bits = 16, .nsd_hops = &ldlm_ns_fid_hash_ops, }, { @@ -541,6 +588,9 @@ ldlm_ns_hash_def_t ldlm_ns_hash_defs[] = }, }; +/** + * Create and initialize new empty namespace. + */ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name, ldlm_side_t client, ldlm_appetite_t apt, @@ -602,7 +652,7 @@ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name, CFS_INIT_LIST_HEAD(&ns->ns_list_chain); CFS_INIT_LIST_HEAD(&ns->ns_unused_list); - cfs_spin_lock_init(&ns->ns_lock); + spin_lock_init(&ns->ns_lock); cfs_atomic_set(&ns->ns_bref, 0); cfs_waitq_init(&ns->ns_waitq); @@ -610,6 +660,7 @@ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name, ns->ns_contention_time = NS_DEFAULT_CONTENTION_SECONDS; ns->ns_contended_locks = NS_DEFAULT_CONTENDED_LOCKS; + ns->ns_max_parallel_ast = LDLM_DEFAULT_PARALLEL_AST_LIMIT; ns->ns_nr_unused = 0; ns->ns_max_unused = LDLM_DEFAULT_LRU_SIZE; ns->ns_max_age = LDLM_DEFAULT_MAX_ALIVE; @@ -617,6 +668,7 @@ struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name, ns->ns_timeouts = 0; ns->ns_orig_connect_flags = 0; ns->ns_connect_flags = 0; + ns->ns_stopping = 0; rc = ldlm_namespace_proc_register(ns); if (rc != 0) { CERROR("Can't initialize ns proc, rc %d\n", rc); @@ -643,24 +695,30 @@ out_ref: ldlm_put_ref(); RETURN(NULL); } +EXPORT_SYMBOL(ldlm_namespace_new); extern struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock); -/* If flags contains FL_LOCAL_ONLY, don't try to tell the server, just cleanup. - * This is currently only used for recovery, and we make certain assumptions - * as a result--notably, that we shouldn't cancel locks with refs. -phil */ +/** + * Cancel and destroy all locks on a resource. + * + * If flags contains FL_LOCAL_ONLY, don't try to tell the server, just + * clean up. This is currently only used for recovery, and we make + * certain assumptions as a result--notably, that we shouldn't cancel + * locks with refs. + */ static void cleanup_resource(struct ldlm_resource *res, cfs_list_t *q, - int flags) + __u64 flags) { - cfs_list_t *tmp; - int rc = 0, client = ns_is_client(ldlm_res_to_ns(res)); - int local_only = (flags & LDLM_FL_LOCAL_ONLY); + cfs_list_t *tmp; + int rc = 0, client = ns_is_client(ldlm_res_to_ns(res)); + bool local_only = !!(flags & LDLM_FL_LOCAL_ONLY); do { struct ldlm_lock *lock = NULL; - /* first, we look for non-cleaned-yet lock - * all cleaned locks are marked by CLEANED flag */ + /* First, we look for non-cleaned-yet lock + * all cleaned locks are marked by CLEANED flag. */ lock_res(res); cfs_list_for_each(tmp, q) { lock = cfs_list_entry(tmp, struct ldlm_lock, @@ -680,7 +738,7 @@ static void cleanup_resource(struct ldlm_resource *res, cfs_list_t *q, } /* Set CBPENDING so nothing in the cancellation path - * can match this lock */ + * can match this lock. */ lock->l_flags |= LDLM_FL_CBPENDING; lock->l_flags |= LDLM_FL_FAILED; lock->l_flags |= flags; @@ -725,7 +783,7 @@ static int ldlm_resource_clean(cfs_hash_t *hs, cfs_hash_bd_t *bd, cfs_hlist_node_t *hnode, void *arg) { struct ldlm_resource *res = cfs_hash_object(hs, hnode); - int flags = (int)(unsigned long)arg; + __u64 flags = *(__u64 *)arg; cleanup_resource(res, &res->lr_granted, flags); cleanup_resource(res, &res->lr_converting, flags); @@ -739,6 +797,7 @@ static int ldlm_resource_complain(cfs_hash_t *hs, cfs_hash_bd_t *bd, { struct ldlm_resource *res = cfs_hash_object(hs, hnode); + lock_res(res); CERROR("Namespace %s resource refcount nonzero " "(%d) after lock cleanup; forcing " "cleanup.\n", @@ -750,22 +809,37 @@ static int ldlm_resource_complain(cfs_hash_t *hs, cfs_hash_bd_t *bd, res->lr_name.name[0], res->lr_name.name[1], res->lr_name.name[2], res->lr_name.name[3], cfs_atomic_read(&res->lr_refcount) - 1); + + ldlm_resource_dump(D_ERROR, res); + unlock_res(res); return 0; } -int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags) +/** + * Cancel and destroy all locks in the namespace. + * + * Typically used during evictions when server notified client that it was + * evicted and all of its state needs to be destroyed. + * Also used during shutdown. + */ +int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags) { if (ns == NULL) { CDEBUG(D_INFO, "NULL ns, skipping cleanup\n"); return ELDLM_OK; } - cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_clean, - (void *)(unsigned long)flags); + cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_clean, &flags); cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_resource_complain, NULL); return ELDLM_OK; } +EXPORT_SYMBOL(ldlm_namespace_cleanup); +/** + * Attempts to free namespace. + * + * Only used when namespace goes away, like during an unmount. + */ static int __ldlm_namespace_free(struct ldlm_namespace *ns, int force) { ENTRY; @@ -811,12 +885,12 @@ force_wait: } /** - * Performs various cleanups for passed \a ns to make it drop refc and be ready - * for freeing. Waits for refc == 0. + * Performs various cleanups for passed \a ns to make it drop refc and be + * ready for freeing. Waits for refc == 0. * * The following is done: - * (0) Unregister \a ns from its list to make inaccessible for potential users - * like pools thread and others; + * (0) Unregister \a ns from its list to make inaccessible for potential + * users like pools thread and others; * (1) Clear all locks in \a ns. */ void ldlm_namespace_free_prior(struct ldlm_namespace *ns, @@ -830,6 +904,9 @@ void ldlm_namespace_free_prior(struct ldlm_namespace *ns, return; } + spin_lock(&ns->ns_lock); + ns->ns_stopping = 1; + spin_unlock(&ns->ns_lock); /* * Can fail with -EINTR when force == 0 in which case try harder. @@ -852,9 +929,9 @@ void ldlm_namespace_free_prior(struct ldlm_namespace *ns, } /** - * Performs freeing memory structures related to \a ns. This is only done when - * ldlm_namespce_free_prior() successfully removed all resources referencing - * \a ns and its refc == 0. + * Performs freeing memory structures related to \a ns. This is only done + * when ldlm_namespce_free_prior() successfully removed all resources + * referencing \a ns and its refc == 0. */ void ldlm_namespace_free_post(struct ldlm_namespace *ns) { @@ -864,32 +941,26 @@ void ldlm_namespace_free_post(struct ldlm_namespace *ns) return; } - - /* - * Make sure that nobody can find this ns in its list. - */ - ldlm_namespace_unregister(ns, ns->ns_client); - /* - * Fini pool _before_ parent proc dir is removed. This is important as - * ldlm_pool_fini() removes own proc dir which is child to @dir. Removing - * it after @dir may cause oops. - */ - ldlm_pool_fini(&ns->ns_pool); - - ldlm_namespace_proc_unregister(ns); - cfs_hash_putref(ns->ns_rs_hash); - /* - * Namespace \a ns should be not on list in this time, otherwise this - * will cause issues realted to using freed \a ns in pools thread. - */ - LASSERT(cfs_list_empty(&ns->ns_list_chain)); - OBD_FREE_PTR(ns); - ldlm_put_ref(); - EXIT; + /* Make sure that nobody can find this ns in its list. */ + ldlm_namespace_unregister(ns, ns->ns_client); + /* Fini pool _before_ parent proc dir is removed. This is important as + * ldlm_pool_fini() removes own proc dir which is child to @dir. + * Removing it after @dir may cause oops. */ + ldlm_pool_fini(&ns->ns_pool); + + ldlm_namespace_proc_unregister(ns); + cfs_hash_putref(ns->ns_rs_hash); + /* Namespace \a ns should be not on list at this time, otherwise + * this will cause issues related to using freed \a ns in poold + * thread. */ + LASSERT(cfs_list_empty(&ns->ns_list_chain)); + OBD_FREE_PTR(ns); + ldlm_put_ref(); + EXIT; } - -/* Cleanup the resource, and free namespace. +/** + * Cleanup the resource, and free namespace. * bug 12864: * Deadlock issue: * proc1: destroy import @@ -913,62 +984,64 @@ void ldlm_namespace_free(struct ldlm_namespace *ns, ldlm_namespace_free_prior(ns, imp, force); ldlm_namespace_free_post(ns); } +EXPORT_SYMBOL(ldlm_namespace_free); void ldlm_namespace_get(struct ldlm_namespace *ns) { cfs_atomic_inc(&ns->ns_bref); } +EXPORT_SYMBOL(ldlm_namespace_get); void ldlm_namespace_put(struct ldlm_namespace *ns) { - if (cfs_atomic_dec_and_lock(&ns->ns_bref, &ns->ns_lock)) { - cfs_waitq_signal(&ns->ns_waitq); - cfs_spin_unlock(&ns->ns_lock); - } + if (cfs_atomic_dec_and_lock(&ns->ns_bref, &ns->ns_lock)) { + cfs_waitq_signal(&ns->ns_waitq); + spin_unlock(&ns->ns_lock); + } } +EXPORT_SYMBOL(ldlm_namespace_put); -/* Register @ns in the list of namespaces */ +/** Register \a ns in the list of namespaces */ void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client) { - cfs_mutex_down(ldlm_namespace_lock(client)); - LASSERT(cfs_list_empty(&ns->ns_list_chain)); - cfs_list_add(&ns->ns_list_chain, ldlm_namespace_list(client)); - cfs_atomic_inc(ldlm_namespace_nr(client)); - cfs_mutex_up(ldlm_namespace_lock(client)); + mutex_lock(ldlm_namespace_lock(client)); + LASSERT(cfs_list_empty(&ns->ns_list_chain)); + cfs_list_add(&ns->ns_list_chain, ldlm_namespace_list(client)); + cfs_atomic_inc(ldlm_namespace_nr(client)); + mutex_unlock(ldlm_namespace_lock(client)); } -/* Unregister @ns from the list of namespaces */ +/** Unregister \a ns from the list of namespaces. */ void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client) { - cfs_mutex_down(ldlm_namespace_lock(client)); - LASSERT(!cfs_list_empty(&ns->ns_list_chain)); - /* - * Some asserts and possibly other parts of code still using - * list_empty(&ns->ns_list_chain). This is why it is important - * to use list_del_init() here. - */ - cfs_list_del_init(&ns->ns_list_chain); - cfs_atomic_dec(ldlm_namespace_nr(client)); - cfs_mutex_up(ldlm_namespace_lock(client)); + mutex_lock(ldlm_namespace_lock(client)); + LASSERT(!cfs_list_empty(&ns->ns_list_chain)); + /* Some asserts and possibly other parts of the code are still + * using list_empty(&ns->ns_list_chain). This is why it is + * important to use list_del_init() here. */ + cfs_list_del_init(&ns->ns_list_chain); + cfs_atomic_dec(ldlm_namespace_nr(client)); + mutex_unlock(ldlm_namespace_lock(client)); } -/* Should be called under ldlm_namespace_lock(client) taken */ +/** Should be called with ldlm_namespace_lock(client) taken. */ void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client) { LASSERT(!cfs_list_empty(&ns->ns_list_chain)); - LASSERT_SEM_LOCKED(ldlm_namespace_lock(client)); + LASSERT_MUTEX_LOCKED(ldlm_namespace_lock(client)); cfs_list_move_tail(&ns->ns_list_chain, ldlm_namespace_list(client)); } -/* Should be called under ldlm_namespace_lock(client) taken */ +/** Should be called with ldlm_namespace_lock(client) taken. */ struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client) { - LASSERT_SEM_LOCKED(ldlm_namespace_lock(client)); + LASSERT_MUTEX_LOCKED(ldlm_namespace_lock(client)); LASSERT(!cfs_list_empty(ldlm_namespace_list(client))); return container_of(ldlm_namespace_list(client)->next, struct ldlm_namespace, ns_list_chain); } +/** Create and initialize new resource. */ static struct ldlm_resource *ldlm_resource_new(void) { struct ldlm_resource *res; @@ -982,7 +1055,7 @@ static struct ldlm_resource *ldlm_resource_new(void) CFS_INIT_LIST_HEAD(&res->lr_converting); CFS_INIT_LIST_HEAD(&res->lr_waiting); - /* initialize interval trees for each lock mode*/ + /* Initialize interval trees for each lock mode. */ for (idx = 0; idx < LCK_MODE_NUM; idx++) { res->lr_itree[idx].lit_size = 0; res->lr_itree[idx].lit_mode = 1 << idx; @@ -990,19 +1063,23 @@ static struct ldlm_resource *ldlm_resource_new(void) } cfs_atomic_set(&res->lr_refcount, 1); - cfs_spin_lock_init(&res->lr_lock); - lu_ref_init(&res->lr_reference); + spin_lock_init(&res->lr_lock); + lu_ref_init(&res->lr_reference); - /* one who creates the resource must unlock - * the semaphore after lvb initialization */ - cfs_init_mutex_locked(&res->lr_lvb_sem); + /* The creator of the resource must unlock the mutex after LVB + * initialization. */ + mutex_init(&res->lr_lvb_mutex); + mutex_lock(&res->lr_lvb_mutex); - return res; + return res; } -/* Args: unlocked namespace - * * Locks: takes and releases NS hash-lock and res->lr_lock - * * Returns: referenced, unlocked ldlm_resource or NULL */ +/** + * Return a reference to resource with given name, creating it if necessary. + * Args: namespace with ns_lock unlocked + * Locks: takes and releases NS hash-lock and res->lr_lock + * Returns: referenced, unlocked ldlm_resource or NULL + */ struct ldlm_resource * ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent, const struct ldlm_res_id *name, ldlm_type_t type, int create) @@ -1022,10 +1099,10 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent, if (hnode != NULL) { cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 0); res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash); - /* synchronize WRT resource creation */ + /* Synchronize with regard to resource creation. */ if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) { - cfs_down(&res->lr_lvb_sem); - cfs_up(&res->lr_lvb_sem); + mutex_lock(&res->lr_lvb_mutex); + mutex_unlock(&res->lr_lvb_mutex); } return res; } @@ -1052,21 +1129,23 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent, cfs_hash_bd_lookup_locked(ns->ns_rs_hash, &bd, (void *)name); if (hnode != NULL) { - /* someone won the race and added the resource before */ - cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1); - /* clean lu_ref for failed resource */ - lu_ref_fini(&res->lr_reference); - OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res); - - res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash); - /* synchronize WRT resource creation */ - if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) { - cfs_down(&res->lr_lvb_sem); - cfs_up(&res->lr_lvb_sem); - } - return res; - } - /* we won! let's add the resource */ + /* Someone won the race and already added the resource. */ + cfs_hash_bd_unlock(ns->ns_rs_hash, &bd, 1); + /* Clean lu_ref for failed resource. */ + lu_ref_fini(&res->lr_reference); + /* We have taken lr_lvb_mutex. Drop it. */ + mutex_unlock(&res->lr_lvb_mutex); + OBD_SLAB_FREE(res, ldlm_resource_slab, sizeof *res); + + res = cfs_hlist_entry(hnode, struct ldlm_resource, lr_hash); + /* Synchronize with regard to resource creation. */ + if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) { + mutex_lock(&res->lr_lvb_mutex); + mutex_unlock(&res->lr_lvb_mutex); + } + return res; + } + /* We won! Let's add the resource. */ cfs_hash_bd_add_locked(ns->ns_rs_hash, &bd, &res->lr_hash); if (cfs_hash_bd_count_get(&bd) == 1) ldlm_namespace_get(ns); @@ -1080,12 +1159,14 @@ ldlm_resource_get(struct ldlm_namespace *ns, struct ldlm_resource *parent, if (rc) CERROR("lvbo_init failed for resource " LPU64": rc %d\n", name->name[0], rc); - /* we create resource with locked lr_lvb_sem */ - cfs_up(&res->lr_lvb_sem); - } + } - return res; + /* We create resource with locked lr_lvb_mutex. */ + mutex_unlock(&res->lr_lvb_mutex); + + return res; } +EXPORT_SYMBOL(ldlm_resource_get); struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res) { @@ -1128,11 +1209,12 @@ static void __ldlm_resource_putref_final(cfs_hash_bd_t *bd, int ldlm_resource_putref(struct ldlm_resource *res) { struct ldlm_namespace *ns = ldlm_res_to_ns(res); - int ref = cfs_atomic_read(&res->lr_refcount); cfs_hash_bd_t bd; - CDEBUG(D_INFO, "putref res: %p count: %d\n", res, ref - 1); - LASSERTF(ref > 0 && ref < LI_POISON, "%d", ref); + LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON); + CDEBUG(D_INFO, "putref res: %p count: %d\n", + res, cfs_atomic_read(&res->lr_refcount) - 1); + cfs_hash_bd_get(ns->ns_rs_hash, &res->lr_name, &bd); if (cfs_hash_bd_dec_and_lock(ns->ns_rs_hash, &bd, &res->lr_refcount)) { __ldlm_resource_putref_final(&bd, res); @@ -1144,15 +1226,17 @@ int ldlm_resource_putref(struct ldlm_resource *res) } return 0; } +EXPORT_SYMBOL(ldlm_resource_putref); /* Returns 1 if the resource was freed, 0 if it remains. */ int ldlm_resource_putref_locked(struct ldlm_resource *res) { struct ldlm_namespace *ns = ldlm_res_to_ns(res); - int ref = cfs_atomic_read(&res->lr_refcount); - CDEBUG(D_INFO, "putref res: %p count: %d\n", res, ref - 1); - LASSERTF(ref > 0 && ref < LI_POISON, "%d", ref); + LASSERT_ATOMIC_GT_LT(&res->lr_refcount, 0, LI_POISON); + CDEBUG(D_INFO, "putref res: %p count: %d\n", + res, cfs_atomic_read(&res->lr_refcount) - 1); + if (cfs_atomic_dec_and_test(&res->lr_refcount)) { cfs_hash_bd_t bd; @@ -1175,14 +1259,15 @@ int ldlm_resource_putref_locked(struct ldlm_resource *res) return 0; } +/** + * Add a lock into a given resource into specified lock list. + */ void ldlm_resource_add_lock(struct ldlm_resource *res, cfs_list_t *head, 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 this lock:\n"); if (lock->l_destroyed) { CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n"); @@ -1194,6 +1279,11 @@ void ldlm_resource_add_lock(struct ldlm_resource *res, cfs_list_t *head, cfs_list_add_tail(&lock->l_res_link, head); } +/** + * Insert a lock into resource after specified lock. + * + * Obtain resource description from the lock we are inserting after. + */ void ldlm_resource_insert_lock_after(struct ldlm_lock *original, struct ldlm_lock *new) { @@ -1202,8 +1292,7 @@ void ldlm_resource_insert_lock_after(struct ldlm_lock *original, check_res_locked(res); ldlm_resource_dump(D_INFO, res); - CDEBUG(D_OTHER, "About to insert this lock after %p:\n", original); - ldlm_lock_dump(D_OTHER, new, 0); + LDLM_DEBUG(new, "About to insert this lock after %p:\n", original); if (new->l_destroyed) { CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n"); @@ -1227,6 +1316,7 @@ void ldlm_resource_unlink_lock(struct ldlm_lock *lock) ldlm_extent_unlink_lock(lock); cfs_list_del_init(&lock->l_res_link); } +EXPORT_SYMBOL(ldlm_resource_unlink_lock); void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc) { @@ -1234,6 +1324,10 @@ void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc) desc->lr_name = res->lr_name; } +/** + * Print information about all locks in all namespaces on this node to debug + * log. + */ void ldlm_dump_all_namespaces(ldlm_side_t client, int level) { cfs_list_t *tmp; @@ -1241,7 +1335,7 @@ void ldlm_dump_all_namespaces(ldlm_side_t client, int level) if (!((libcfs_debug | D_ERROR) & level)) return; - cfs_mutex_down(ldlm_namespace_lock(client)); + mutex_lock(ldlm_namespace_lock(client)); cfs_list_for_each(tmp, ldlm_namespace_list(client)) { struct ldlm_namespace *ns; @@ -1249,8 +1343,9 @@ void ldlm_dump_all_namespaces(ldlm_side_t client, int level) ldlm_namespace_dump(level, ns); } - cfs_mutex_up(ldlm_namespace_lock(client)); + mutex_unlock(ldlm_namespace_lock(client)); } +EXPORT_SYMBOL(ldlm_dump_all_namespaces); static int ldlm_res_hash_dump(cfs_hash_t *hs, cfs_hash_bd_t *bd, cfs_hlist_node_t *hnode, void *arg) @@ -1265,6 +1360,10 @@ static int ldlm_res_hash_dump(cfs_hash_t *hs, cfs_hash_bd_t *bd, return 0; } +/** + * Print information about all locks in this namespace on this node to debug + * log. + */ void ldlm_namespace_dump(int level, struct ldlm_namespace *ns) { if (!((libcfs_debug | D_ERROR) & level)) @@ -1280,15 +1379,19 @@ void ldlm_namespace_dump(int level, struct ldlm_namespace *ns) cfs_hash_for_each_nolock(ns->ns_rs_hash, ldlm_res_hash_dump, (void *)(unsigned long)level); - cfs_spin_lock(&ns->ns_lock); - ns->ns_next_dump = cfs_time_shift(10); - cfs_spin_unlock(&ns->ns_lock); + spin_lock(&ns->ns_lock); + ns->ns_next_dump = cfs_time_shift(10); + spin_unlock(&ns->ns_lock); } +EXPORT_SYMBOL(ldlm_namespace_dump); +/** + * Print information about all locks in this resource to debug log. + */ void ldlm_resource_dump(int level, struct ldlm_resource *res) { - cfs_list_t *tmp; - int pos; + struct ldlm_lock *lock; + unsigned int granted = 0; CLASSERT(RES_NAME_SIZE == 4); @@ -1301,33 +1404,26 @@ void ldlm_resource_dump(int level, struct ldlm_resource *res) cfs_atomic_read(&res->lr_refcount)); if (!cfs_list_empty(&res->lr_granted)) { - pos = 0; - CDEBUG(level, "Granted locks:\n"); - cfs_list_for_each(tmp, &res->lr_granted) { - struct ldlm_lock *lock; - lock = cfs_list_entry(tmp, struct ldlm_lock, - l_res_link); - ldlm_lock_dump(level, lock, ++pos); + CDEBUG(level, "Granted locks (in reverse order):\n"); + cfs_list_for_each_entry_reverse(lock, &res->lr_granted, + l_res_link) { + LDLM_DEBUG_LIMIT(level, lock, "###"); + if (!(level & D_CANTMASK) && + ++granted > ldlm_dump_granted_max) { + CDEBUG(level, "only dump %d granted locks to " + "avoid DDOS.\n", granted); + break; + } } } if (!cfs_list_empty(&res->lr_converting)) { - pos = 0; CDEBUG(level, "Converting locks:\n"); - cfs_list_for_each(tmp, &res->lr_converting) { - struct ldlm_lock *lock; - lock = cfs_list_entry(tmp, struct ldlm_lock, - l_res_link); - ldlm_lock_dump(level, lock, ++pos); - } + cfs_list_for_each_entry(lock, &res->lr_converting, l_res_link) + LDLM_DEBUG_LIMIT(level, lock, "###"); } if (!cfs_list_empty(&res->lr_waiting)) { - pos = 0; CDEBUG(level, "Waiting locks:\n"); - cfs_list_for_each(tmp, &res->lr_waiting) { - struct ldlm_lock *lock; - lock = cfs_list_entry(tmp, struct ldlm_lock, - l_res_link); - ldlm_lock_dump(level, lock, ++pos); - } + cfs_list_for_each_entry(lock, &res->lr_waiting, l_res_link) + LDLM_DEBUG_LIMIT(level, lock, "###"); } }