X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lock.c;h=e7a507292e466c81110b79776aaca2c3559a8c97;hp=165fe683d5452dd1a065bd582463b832ebb1e286;hb=98c2e6b446166ba8f89e60a0d4f38683b920f506;hpb=2ad695f40cf1920033d77723311ed5ed7a3b07cc diff --git a/lustre/ldlm/ldlm_lock.c b/lustre/ldlm/ldlm_lock.c index 165fe68..e7a5072 100644 --- a/lustre/ldlm/ldlm_lock.c +++ b/lustre/ldlm/ldlm_lock.c @@ -27,7 +27,7 @@ * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2010, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -112,20 +112,20 @@ 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(struct obd_export *exp, ldlm_type_t type, - const ldlm_wire_policy_data_t *wpolicy, - ldlm_policy_data_t *lpolicy) + const ldlm_wire_policy_data_t *wpolicy, + ldlm_policy_data_t *lpolicy) { - ldlm_policy_wire_to_local_t convert; - int new_client; + ldlm_policy_wire_to_local_t convert; + int new_client; - /** 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]; + /** some badness for 2.0.0 clients, but 2.0.0 isn't supported */ + new_client = (exp_connect_flags(exp) & 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); + convert(wpolicy, lpolicy); } char *ldlm_it2str(int it) @@ -156,7 +156,7 @@ char *ldlm_it2str(int it) } EXPORT_SYMBOL(ldlm_it2str); -extern cfs_mem_cache_t *ldlm_lock_slab; +extern struct kmem_cache *ldlm_lock_slab; #ifdef HAVE_SERVER_SUPPORT static ldlm_processing_policy ldlm_processing_policy_table[] = { @@ -186,7 +186,9 @@ EXPORT_SYMBOL(ldlm_register_intent); */ -/* +/** + * Get a reference on a lock. + * * Lock refcounts, during creation: * - one special one for allocation, dec'd only once in destroy * - one for being a lock that's in-use @@ -194,25 +196,30 @@ EXPORT_SYMBOL(ldlm_register_intent); */ struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock) { - cfs_atomic_inc(&lock->l_refc); + atomic_inc(&lock->l_refc); return lock; } EXPORT_SYMBOL(ldlm_lock_get); +/** + * Release lock reference. + * + * Also frees the lock if it was last reference. + */ void ldlm_lock_put(struct ldlm_lock *lock) { ENTRY; LASSERT(lock->l_resource != LP_POISON); - LASSERT(cfs_atomic_read(&lock->l_refc) > 0); - if (cfs_atomic_dec_and_test(&lock->l_refc)) { + LASSERT(atomic_read(&lock->l_refc) > 0); + if (atomic_dec_and_test(&lock->l_refc)) { struct ldlm_resource *res; LDLM_DEBUG(lock, "final lock_put on destroyed lock, freeing it."); res = lock->l_resource; - LASSERT(lock->l_destroyed); + LASSERT(ldlm_is_destroyed(lock)); LASSERT(cfs_list_empty(&lock->l_res_link)); LASSERT(cfs_list_empty(&lock->l_pending_chain)); @@ -227,7 +234,7 @@ void ldlm_lock_put(struct ldlm_lock *lock) } if (lock->l_lvb_data != NULL) - OBD_FREE(lock->l_lvb_data, lock->l_lvb_len); + OBD_FREE_LARGE(lock->l_lvb_data, lock->l_lvb_len); ldlm_interval_free(ldlm_interval_detach(lock)); lu_ref_fini(&lock->l_reference); @@ -238,89 +245,119 @@ void ldlm_lock_put(struct ldlm_lock *lock) } EXPORT_SYMBOL(ldlm_lock_put); +/** + * Removes LDLM lock \a lock from LRU. Assumes LRU is already locked. + */ int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock) { - int rc = 0; - if (!cfs_list_empty(&lock->l_lru)) { - struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); - - LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); - cfs_list_del_init(&lock->l_lru); - if (lock->l_flags & LDLM_FL_SKIPPED) - lock->l_flags &= ~LDLM_FL_SKIPPED; - LASSERT(ns->ns_nr_unused > 0); - ns->ns_nr_unused--; - rc = 1; - } - return rc; + int rc = 0; + if (!cfs_list_empty(&lock->l_lru)) { + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + + LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); + cfs_list_del_init(&lock->l_lru); + LASSERT(ns->ns_nr_unused > 0); + ns->ns_nr_unused--; + rc = 1; + } + return rc; } +/** + * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first. + */ int ldlm_lock_remove_from_lru(struct ldlm_lock *lock) { - struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); - int rc; + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + int rc; - ENTRY; - if (lock->l_ns_srv) { - LASSERT(cfs_list_empty(&lock->l_lru)); - RETURN(0); - } + ENTRY; + if (ldlm_is_ns_srv(lock)) { + LASSERT(cfs_list_empty(&lock->l_lru)); + RETURN(0); + } - cfs_spin_lock(&ns->ns_lock); - rc = ldlm_lock_remove_from_lru_nolock(lock); - cfs_spin_unlock(&ns->ns_lock); - EXIT; - return rc; + spin_lock(&ns->ns_lock); + rc = ldlm_lock_remove_from_lru_nolock(lock); + spin_unlock(&ns->ns_lock); + EXIT; + return rc; } +/** + * Adds LDLM lock \a lock to namespace LRU. Assumes LRU is already locked. + */ void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock) { - struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); - lock->l_last_used = cfs_time_current(); - LASSERT(cfs_list_empty(&lock->l_lru)); - LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); - cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list); - LASSERT(ns->ns_nr_unused >= 0); - ns->ns_nr_unused++; + lock->l_last_used = cfs_time_current(); + LASSERT(cfs_list_empty(&lock->l_lru)); + LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); + cfs_list_add_tail(&lock->l_lru, &ns->ns_unused_list); + ldlm_clear_skipped(lock); + LASSERT(ns->ns_nr_unused >= 0); + ns->ns_nr_unused++; } +/** + * Adds LDLM lock \a lock to namespace LRU. Obtains necessary LRU locks + * first. + */ void ldlm_lock_add_to_lru(struct ldlm_lock *lock) { - struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); - ENTRY; - cfs_spin_lock(&ns->ns_lock); - ldlm_lock_add_to_lru_nolock(lock); - cfs_spin_unlock(&ns->ns_lock); - EXIT; + ENTRY; + spin_lock(&ns->ns_lock); + ldlm_lock_add_to_lru_nolock(lock); + spin_unlock(&ns->ns_lock); + EXIT; } +/** + * Moves LDLM lock \a lock that is already in namespace LRU to the tail of + * the LRU. Performs necessary LRU locking + */ void ldlm_lock_touch_in_lru(struct ldlm_lock *lock) { - struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); - ENTRY; - if (lock->l_ns_srv) { - LASSERT(cfs_list_empty(&lock->l_lru)); - EXIT; - return; - } + ENTRY; + if (ldlm_is_ns_srv(lock)) { + LASSERT(cfs_list_empty(&lock->l_lru)); + EXIT; + return; + } - cfs_spin_lock(&ns->ns_lock); - if (!cfs_list_empty(&lock->l_lru)) { - ldlm_lock_remove_from_lru_nolock(lock); - ldlm_lock_add_to_lru_nolock(lock); - } - cfs_spin_unlock(&ns->ns_lock); - EXIT; + spin_lock(&ns->ns_lock); + if (!cfs_list_empty(&lock->l_lru)) { + ldlm_lock_remove_from_lru_nolock(lock); + ldlm_lock_add_to_lru_nolock(lock); + } + spin_unlock(&ns->ns_lock); + EXIT; } -/* This used to have a 'strict' flag, which recovery would use to mark an +/** + * Helper to destroy a locked lock. + * + * Used by ldlm_lock_destroy and ldlm_lock_destroy_nolock + * Must be called with l_lock and lr_lock held. + * + * Does not actually free the lock data, but rather marks the lock as + * destroyed by setting l_destroyed field in the lock to 1. Destroys a + * handle->lock association too, so that the lock can no longer be found + * and removes the lock from LRU list. Actual lock freeing occurs when + * last lock reference goes away. + * + * Original comment (of some historical value): + * This used to have a 'strict' flag, which recovery would use to mark an * in-use lock as needing-to-die. Lest I am ever tempted to put it back, I * shall explain why it's gone: with the new hash table scheme, once you call * ldlm_lock_destroy, you can never drop your final references on this lock. - * Because it's not in the hash table anymore. -phil */ + * Because it's not in the hash table anymore. -phil + */ int ldlm_lock_destroy_internal(struct ldlm_lock *lock) { ENTRY; @@ -335,12 +372,12 @@ int ldlm_lock_destroy_internal(struct ldlm_lock *lock) LBUG(); } - if (lock->l_destroyed) { - LASSERT(cfs_list_empty(&lock->l_lru)); - EXIT; - return 0; - } - lock->l_destroyed = 1; + if (ldlm_is_destroyed(lock)) { + LASSERT(cfs_list_empty(&lock->l_lru)); + EXIT; + return 0; + } + ldlm_set_destroyed(lock); if (lock->l_export && lock->l_export->exp_lock_hash) { /* NB: it's safe to call cfs_hash_del() even lock isn't @@ -369,6 +406,9 @@ int ldlm_lock_destroy_internal(struct ldlm_lock *lock) return 1; } +/** + * Destroys a LDLM lock \a lock. Performs necessary locking first. + */ void ldlm_lock_destroy(struct ldlm_lock *lock) { int first; @@ -385,6 +425,9 @@ void ldlm_lock_destroy(struct ldlm_lock *lock) EXIT; } +/** + * Destroys a LDLM lock \a lock that is already locked. + */ void ldlm_lock_destroy_nolock(struct ldlm_lock *lock) { int first; @@ -415,7 +458,10 @@ struct portals_handle_ops lock_handle_ops = { .hop_free = lock_handle_free, }; -/* +/** + * + * Allocate and initialize new lock structure. + * * usage: pass in a resource on which you have done ldlm_resource_get * new lock will take over the refcount. * returns: lock with refcount 2 - one for current caller and one for remote @@ -428,26 +474,26 @@ static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource) if (resource == NULL) LBUG(); - OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, CFS_ALLOC_IO); + OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, __GFP_IO); if (lock == NULL) RETURN(NULL); - cfs_spin_lock_init(&lock->l_lock); + spin_lock_init(&lock->l_lock); lock->l_resource = resource; lu_ref_add(&resource->lr_reference, "lock", lock); - cfs_atomic_set(&lock->l_refc, 2); - CFS_INIT_LIST_HEAD(&lock->l_res_link); - CFS_INIT_LIST_HEAD(&lock->l_lru); - CFS_INIT_LIST_HEAD(&lock->l_pending_chain); - CFS_INIT_LIST_HEAD(&lock->l_bl_ast); - CFS_INIT_LIST_HEAD(&lock->l_cp_ast); - CFS_INIT_LIST_HEAD(&lock->l_rk_ast); - cfs_waitq_init(&lock->l_waitq); - lock->l_blocking_lock = NULL; - CFS_INIT_LIST_HEAD(&lock->l_sl_mode); - CFS_INIT_LIST_HEAD(&lock->l_sl_policy); - CFS_INIT_HLIST_NODE(&lock->l_exp_hash); + atomic_set(&lock->l_refc, 2); + CFS_INIT_LIST_HEAD(&lock->l_res_link); + CFS_INIT_LIST_HEAD(&lock->l_lru); + CFS_INIT_LIST_HEAD(&lock->l_pending_chain); + CFS_INIT_LIST_HEAD(&lock->l_bl_ast); + CFS_INIT_LIST_HEAD(&lock->l_cp_ast); + CFS_INIT_LIST_HEAD(&lock->l_rk_ast); + init_waitqueue_head(&lock->l_waitq); + lock->l_blocking_lock = NULL; + 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, @@ -469,6 +515,11 @@ static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource) RETURN(lock); } +/** + * Moves LDLM lock \a lock to another resource. + * This is used on client when server returns some other lock than requested + * (typically as a result of intent operation) + */ int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock, const struct ldlm_res_id *new_resid) { @@ -506,7 +557,7 @@ int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock, * lock->l_lock, and are taken in the memory address order to avoid * dead-locks. */ - cfs_spin_lock(&lock->l_lock); + spin_lock(&lock->l_lock); oldres = lock->l_resource; if (oldres < newres) { lock_res(oldres); @@ -529,73 +580,87 @@ int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock, } EXPORT_SYMBOL(ldlm_lock_change_resource); -/* - * HANDLES +/** \defgroup ldlm_handles LDLM HANDLES + * Ways to get hold of locks without any addresses. + * @{ */ +/** + * Fills in handle for LDLM lock \a lock into supplied \a lockh + * Does not take any references. + */ void ldlm_lock2handle(const struct ldlm_lock *lock, struct lustre_handle *lockh) { - lockh->cookie = lock->l_handle.h_cookie; + lockh->cookie = lock->l_handle.h_cookie; } EXPORT_SYMBOL(ldlm_lock2handle); -/* if flags: atomically get the lock and set the flags. - * Return NULL if flag already set +/** + * Obtain a lock reference by handle. + * + * if \a flags: atomically get the lock and set the flags. + * Return NULL if flag already set */ - struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *handle, - int flags) + __u64 flags) { - struct ldlm_lock *lock; - ENTRY; + struct ldlm_lock *lock; + ENTRY; - LASSERT(handle); + LASSERT(handle); - lock = class_handle2object(handle->cookie); - if (lock == NULL) - RETURN(NULL); + lock = class_handle2object(handle->cookie, NULL); + if (lock == NULL) + RETURN(NULL); - /* It's unlikely but possible that someone marked the lock as - * destroyed after we did handle2object on it */ - if (flags == 0 && !lock->l_destroyed) { - lu_ref_add(&lock->l_reference, "handle", cfs_current()); - RETURN(lock); - } + /* It's unlikely but possible that someone marked the lock as + * destroyed after we did handle2object on it */ + if ((flags == 0) && !ldlm_is_destroyed(lock)) { + lu_ref_add(&lock->l_reference, "handle", current); + RETURN(lock); + } - lock_res_and_lock(lock); + lock_res_and_lock(lock); - LASSERT(lock->l_resource != NULL); + LASSERT(lock->l_resource != NULL); - lu_ref_add_atomic(&lock->l_reference, "handle", cfs_current()); - if (unlikely(lock->l_destroyed)) { - unlock_res_and_lock(lock); - CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock); - LDLM_LOCK_PUT(lock); - RETURN(NULL); - } + lu_ref_add_atomic(&lock->l_reference, "handle", current); + if (unlikely(ldlm_is_destroyed(lock))) { + unlock_res_and_lock(lock); + CDEBUG(D_INFO, "lock already destroyed: lock %p\n", lock); + LDLM_LOCK_PUT(lock); + RETURN(NULL); + } - if (flags && (lock->l_flags & flags)) { - unlock_res_and_lock(lock); - LDLM_LOCK_PUT(lock); - RETURN(NULL); - } + /* If we're setting flags, make sure none of them are already set. */ + if (flags != 0) { + if ((lock->l_flags & flags) != 0) { + unlock_res_and_lock(lock); + LDLM_LOCK_PUT(lock); + RETURN(NULL); + } - if (flags) - lock->l_flags |= flags; + lock->l_flags |= flags; + } - unlock_res_and_lock(lock); - RETURN(lock); + unlock_res_and_lock(lock); + RETURN(lock); } EXPORT_SYMBOL(__ldlm_handle2lock); +/** @} ldlm_handles */ +/** + * Fill in "on the wire" representation for given LDLM lock into supplied + * lock descriptor \a desc structure. + */ void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc) { - struct obd_export *exp = lock->l_export?:lock->l_conn_export; - /* INODEBITS_INTEROP: If the other side does not support - * inodebits, reply with a plain lock descriptor. - */ - if ((lock->l_resource->lr_type == LDLM_IBITS) && - (exp && !(exp->exp_connect_flags & OBD_CONNECT_IBITS))) { + struct obd_export *exp = lock->l_export ?: lock->l_conn_export; + + /* INODEBITS_INTEROP: If the other side does not support + * inodebits, reply with a plain lock descriptor. */ + if ((lock->l_resource->lr_type == LDLM_IBITS) && + (exp && !(exp_connect_flags(exp) & OBD_CONNECT_IBITS))) { /* Make sure all the right bits are set in this lock we are going to pass to client */ LASSERTF(lock->l_policy_data.l_inodebits.bits == @@ -638,16 +703,21 @@ void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc) } EXPORT_SYMBOL(ldlm_lock2desc); +/** + * Add a lock to list of conflicting locks to send AST to. + * + * Only add if we have not sent a blocking AST to the lock yet. + */ void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, cfs_list_t *work_list) { - if ((lock->l_flags & LDLM_FL_AST_SENT) == 0) { - LDLM_DEBUG(lock, "lock incompatible; sending blocking AST."); - lock->l_flags |= LDLM_FL_AST_SENT; - /* If the enqueuing client said so, tell the AST recipient to - * discard dirty data, rather than writing back. */ - if (new->l_flags & LDLM_AST_DISCARD_DATA) - lock->l_flags |= LDLM_FL_DISCARD_DATA; + if (!ldlm_is_ast_sent(lock)) { + LDLM_DEBUG(lock, "lock incompatible; sending blocking AST."); + ldlm_set_ast_sent(lock); + /* If the enqueuing client said so, tell the AST recipient to + * discard dirty data, rather than writing back. */ + if (ldlm_is_ast_discard_data(new)) + ldlm_set_discard_data(lock); LASSERT(cfs_list_empty(&lock->l_bl_ast)); cfs_list_add(&lock->l_bl_ast, work_list); LDLM_LOCK_GET(lock); @@ -656,10 +726,13 @@ void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, } } +/** + * Add a lock to list of just granted locks to send completion AST to. + */ void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list) { - if ((lock->l_flags & LDLM_FL_CP_REQD) == 0) { - lock->l_flags |= LDLM_FL_CP_REQD; + if (!ldlm_is_cp_reqd(lock)) { + ldlm_set_cp_reqd(lock); LDLM_DEBUG(lock, "lock granted; sending completion AST."); LASSERT(cfs_list_empty(&lock->l_cp_ast)); cfs_list_add(&lock->l_cp_ast, work_list); @@ -667,7 +740,12 @@ void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list) } } -/* must be called with lr_lock held */ +/** + * Aggregator function to add AST work items into a list. Determines + * what sort of an AST work needs to be done and calls the proper + * adding function. + * Must be called with lr_lock held. + */ void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, cfs_list_t *work_list) { @@ -680,6 +758,11 @@ void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, EXIT; } +/** + * Add specified reader/writer reference to LDLM lock with handle \a lockh. + * r/w reference type is determined by \a mode + * Calls ldlm_lock_addref_internal. + */ void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode) { struct ldlm_lock *lock; @@ -691,6 +774,13 @@ void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode) } EXPORT_SYMBOL(ldlm_lock_addref); +/** + * Helper function. + * Add specified reader/writer reference to LDLM lock \a lock. + * r/w reference type is determined by \a mode + * Removes lock from LRU if it is there. + * Assumes the LDLM lock is already locked. + */ void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode) { ldlm_lock_remove_from_lru(lock); @@ -708,8 +798,8 @@ void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode) } /** - * Attempts to addref a lock, and fails if lock is already LDLM_FL_CBPENDING - * or destroyed. + * Attempts to add reader/writer reference to a lock with handle \a lockh, and + * fails if lock is already LDLM_FL_CBPENDING or destroyed. * * \retval 0 success, lock was addref-ed * @@ -725,7 +815,7 @@ int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode) if (lock != NULL) { lock_res_and_lock(lock); if (lock->l_readers != 0 || lock->l_writers != 0 || - !(lock->l_flags & LDLM_FL_CBPENDING)) { + !ldlm_is_cbpending(lock)) { ldlm_lock_addref_internal_nolock(lock, mode); result = 0; } @@ -736,7 +826,11 @@ int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode) } EXPORT_SYMBOL(ldlm_lock_addref_try); -/* only called for local locks */ +/** + * Add specified reader/writer reference to LDLM lock \a lock. + * Locks LDLM lock and calls ldlm_lock_addref_internal_nolock to do the work. + * Only called for local locks. + */ void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode) { lock_res_and_lock(lock); @@ -744,10 +838,13 @@ void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode) unlock_res_and_lock(lock); } -/* only called in ldlm_flock_destroy and for local locks. - * * for LDLM_FLOCK type locks, l_blocking_ast is null, and - * * ldlm_lock_remove_from_lru() does nothing, it is safe - * * for ldlm_flock_destroy usage by dropping some code */ +/** + * Removes reader/writer reference for LDLM lock \a lock. + * Assumes LDLM lock is already locked. + * only called in ldlm_flock_destroy and for local locks. + * Does NOT add lock to LRU if no r/w references left to accomodate flock locks + * that cannot be placed in LRU. + */ void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode) { LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]); @@ -766,6 +863,14 @@ void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode) LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */ } +/** + * Removes reader/writer reference for LDLM lock \a lock. + * Locks LDLM lock first. + * If the lock is determined to be client lock on a client and r/w refcount + * drops to zero and the lock is not blocked, the lock is added to LRU lock + * on the namespace. + * For blocked LDLM locks if r/w count drops to zero, blocking_ast is called. + */ void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) { struct ldlm_namespace *ns; @@ -777,19 +882,19 @@ void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) ldlm_lock_decref_internal_nolock(lock, mode); - if (lock->l_flags & LDLM_FL_LOCAL && + if (ldlm_is_local(lock) && !lock->l_readers && !lock->l_writers) { /* If this is a local lock on a server namespace and this was * the last reference, cancel the lock. */ CDEBUG(D_INFO, "forcing cancel of local lock\n"); - lock->l_flags |= LDLM_FL_CBPENDING; + ldlm_set_cbpending(lock); } if (!lock->l_readers && !lock->l_writers && - (lock->l_flags & LDLM_FL_CBPENDING)) { + ldlm_is_cbpending(lock)) { /* If we received a blocked AST and this was the last reference, * run the callback. */ - if (lock->l_ns_srv && lock->l_export) + if (ldlm_is_ns_srv(lock) && lock->l_export) CERROR("FL_CBPENDING set on non-local lock--just a " "warning\n"); @@ -799,16 +904,16 @@ void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) ldlm_lock_remove_from_lru(lock); unlock_res_and_lock(lock); - if (lock->l_flags & LDLM_FL_FAIL_LOC) + if (ldlm_is_fail_loc(lock)) OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE); - if ((lock->l_flags & LDLM_FL_ATOMIC_CB) || + if (ldlm_is_atomic_cb(lock) || ldlm_bl_to_thread_lock(ns, NULL, lock) != 0) ldlm_handle_bl_callback(ns, NULL, lock); } else if (ns_is_client(ns) && !lock->l_readers && !lock->l_writers && - !(lock->l_flags & LDLM_FL_NO_LRU) && - !(lock->l_flags & LDLM_FL_BL_AST)) { + !ldlm_is_no_lru(lock) && + !ldlm_is_bl_ast(lock)) { LDLM_DEBUG(lock, "add lock into lru list"); @@ -817,7 +922,7 @@ void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) ldlm_lock_add_to_lru(lock); unlock_res_and_lock(lock); - if (lock->l_flags & LDLM_FL_FAIL_LOC) + if (ldlm_is_fail_loc(lock)) OBD_RACE(OBD_FAIL_LDLM_CP_BL_RACE); /* Call ldlm_cancel_lru() only if EARLY_CANCEL and LRU RESIZE @@ -825,7 +930,7 @@ void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) * enqueue. */ if (!exp_connect_cancelset(lock->l_conn_export) && !ns_connect_lru_resize(ns)) - ldlm_cancel_lru(ns, 0, LDLM_ASYNC, 0); + ldlm_cancel_lru(ns, 0, LCF_ASYNC, 0); } else { LDLM_DEBUG(lock, "do not add lock into lru list"); unlock_res_and_lock(lock); @@ -834,6 +939,9 @@ void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) EXIT; } +/** + * Decrease reader/writer refcount for LDLM lock with handle \a lockh + */ void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode) { struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0); @@ -843,8 +951,13 @@ void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode) } EXPORT_SYMBOL(ldlm_lock_decref); -/* This will drop a lock reference and mark it for destruction, but will not - * necessarily cancel the lock before returning. */ +/** + * Decrease reader/writer refcount for LDLM lock with handle + * \a lockh and mark it for subsequent cancellation once r/w refcount + * drops to zero instead of putting into LRU. + * + * Typical usage is for GROUP locks which we cannot allow to be cached. + */ void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode) { struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0); @@ -854,7 +967,7 @@ void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode) LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]); lock_res_and_lock(lock); - lock->l_flags |= LDLM_FL_CBPENDING; + ldlm_set_cbpending(lock); unlock_res_and_lock(lock); ldlm_lock_decref_internal(lock, mode); LDLM_LOCK_PUT(lock); @@ -867,11 +980,11 @@ struct sl_insert_point { cfs_list_t *policy_link; }; -/* - * search_granted_lock +/** + * Finds a position to insert the new lock into granted lock list. + * + * Used for locks eligible for skiplist optimization. * - * Description: - * Finds a position to insert the new lock. * Parameters: * queue [input]: the granted list where search acts on; * req [input]: the lock whose position to be located; @@ -962,6 +1075,10 @@ static void search_granted_lock(cfs_list_t *queue, return; } +/** + * Add a lock into resource granted list after a position described by + * \a prev. + */ static void ldlm_granted_list_add_lock(struct ldlm_lock *lock, struct sl_insert_point *prev) { @@ -973,7 +1090,7 @@ static void ldlm_granted_list_add_lock(struct ldlm_lock *lock, ldlm_resource_dump(D_INFO, res); LDLM_DEBUG(lock, "About to add lock:"); - if (lock->l_destroyed) { + if (ldlm_is_destroyed(lock)) { CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n"); return; } @@ -982,13 +1099,24 @@ static void ldlm_granted_list_add_lock(struct ldlm_lock *lock, LASSERT(cfs_list_empty(&lock->l_sl_mode)); LASSERT(cfs_list_empty(&lock->l_sl_policy)); - cfs_list_add(&lock->l_res_link, prev->res_link); - cfs_list_add(&lock->l_sl_mode, prev->mode_link); - cfs_list_add(&lock->l_sl_policy, prev->policy_link); + /* + * lock->link == prev->link means lock is first starting the group. + * Don't re-add to itself to suppress kernel warnings. + */ + if (&lock->l_res_link != prev->res_link) + cfs_list_add(&lock->l_res_link, prev->res_link); + if (&lock->l_sl_mode != prev->mode_link) + cfs_list_add(&lock->l_sl_mode, prev->mode_link); + if (&lock->l_sl_policy != prev->policy_link) + cfs_list_add(&lock->l_sl_policy, prev->policy_link); EXIT; } +/** + * Add a lock to granted list on a resource maintaining skiplist + * correctness. + */ static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock) { struct sl_insert_point prev; @@ -1001,7 +1129,11 @@ static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock) EXIT; } -/* NOTE: called by +/** + * Perform lock granting bookkeeping. + * + * Includes putting the lock into granted list and updating lock mode. + * NOTE: called by * - ldlm_lock_enqueue * - ldlm_reprocess_queue * - ldlm_lock_convert @@ -1033,13 +1165,17 @@ void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list) EXIT; } -/* returns a referenced lock or NULL. See the flag descriptions below, in the - * comment above ldlm_lock_match */ +/** + * Search for a lock with given properties in a queue. + * + * \retval a referenced lock or NULL. See the flag descriptions below, in the + * comment above ldlm_lock_match + */ static struct ldlm_lock *search_queue(cfs_list_t *queue, ldlm_mode_t *mode, ldlm_policy_data_t *policy, struct ldlm_lock *old_lock, - int flags, int unref) + __u64 flags, int unref) { struct ldlm_lock *lock; cfs_list_t *tmp; @@ -1052,16 +1188,21 @@ static struct ldlm_lock *search_queue(cfs_list_t *queue, if (lock == old_lock) break; + /* Check if this lock can be matched. + * Used by LU-2919(exclusive open) for open lease lock */ + if (ldlm_is_excl(lock)) + continue; + /* llite sometimes wants to match locks that will be * canceled when their users drop, but we allow it to match * if it passes in CBPENDING and the lock still has users. * this is generally only going to be used by children * whose parents already hold a lock so forward progress * can still happen. */ - if (lock->l_flags & LDLM_FL_CBPENDING && + if (ldlm_is_cbpending(lock) && !(flags & LDLM_FL_CBPENDING)) continue; - if (!unref && lock->l_flags & LDLM_FL_CBPENDING && + if (!unref && ldlm_is_cbpending(lock) && lock->l_readers == 0 && lock->l_writers == 0) continue; @@ -1088,13 +1229,11 @@ static struct ldlm_lock *search_queue(cfs_list_t *queue, policy->l_inodebits.bits)) continue; - if (!unref && - (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED || - lock->l_failed)) + if (!unref && LDLM_HAVE_MASK(lock, GONE)) continue; if ((flags & LDLM_FL_LOCAL_ONLY) && - !(lock->l_flags & LDLM_FL_LOCAL)) + !ldlm_is_local(lock)) continue; if (flags & LDLM_FL_TEST_LOCK) { @@ -1112,10 +1251,10 @@ static struct ldlm_lock *search_queue(cfs_list_t *queue, void ldlm_lock_fail_match_locked(struct ldlm_lock *lock) { - if (!lock->l_failed) { - lock->l_failed = 1; - cfs_waitq_broadcast(&lock->l_waitq); - } + if ((lock->l_flags & LDLM_FL_FAIL_NOTIFIED) == 0) { + lock->l_flags |= LDLM_FL_FAIL_NOTIFIED; + wake_up_all(&lock->l_waitq); + } } EXPORT_SYMBOL(ldlm_lock_fail_match_locked); @@ -1127,13 +1266,24 @@ void ldlm_lock_fail_match(struct ldlm_lock *lock) } EXPORT_SYMBOL(ldlm_lock_fail_match); +/** + * Mark lock as "matchable" by OST. + * + * Used to prevent certain races in LOV/OSC where the lock is granted, but LVB + * is not yet valid. + * Assumes LDLM lock is already locked. + */ void ldlm_lock_allow_match_locked(struct ldlm_lock *lock) { - lock->l_flags |= LDLM_FL_LVB_READY; - cfs_waitq_broadcast(&lock->l_waitq); + ldlm_set_lvb_ready(lock); + wake_up_all(&lock->l_waitq); } EXPORT_SYMBOL(ldlm_lock_allow_match_locked); +/** + * Mark lock as "matchable" by OST. + * Locks the lock and then \see ldlm_lock_allow_match_locked + */ void ldlm_lock_allow_match(struct ldlm_lock *lock) { lock_res_and_lock(lock); @@ -1142,7 +1292,13 @@ void ldlm_lock_allow_match(struct ldlm_lock *lock) } EXPORT_SYMBOL(ldlm_lock_allow_match); -/* Can be called in two ways: +/** + * Attempt to find a lock with specified properties. + * + * Typically returns a reference to matched lock unless LDLM_FL_TEST_LOCK is + * set in \a flags + * + * Can be called in two ways: * * If 'ns' is NULL, then lockh describes an existing lock that we want to look * for a duplicate of. @@ -1159,14 +1315,14 @@ EXPORT_SYMBOL(ldlm_lock_allow_match); * If 'flags' contains LDLM_FL_TEST_LOCK, then don't actually reference a lock, * just tell us if we would have matched. * - * Returns 1 if it finds an already-existing lock that is compatible; in this + * \retval 1 if it finds an already-existing lock that is compatible; in this * case, lockh is filled in with a addref()ed lock * - * we also check security context, if that failed we simply return 0 (to keep - * caller code unchanged), the context failure will be discovered by caller - * sometime later. + * We also check security context, and if that fails we simply return 0 (to + * keep caller code unchanged), the context failure will be discovered by + * caller sometime later. */ -ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags, +ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags, const struct ldlm_res_id *res_id, ldlm_type_t type, ldlm_policy_data_t *policy, ldlm_mode_t mode, struct lustre_handle *lockh, int unref) @@ -1219,7 +1375,9 @@ ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags, if (lock) { ldlm_lock2handle(lock, lockh); if ((flags & LDLM_FL_LVB_READY) && - (!(lock->l_flags & LDLM_FL_LVB_READY))) { + (!ldlm_is_lvb_ready(lock))) { + __u64 wait_flags = LDLM_FL_LVB_READY | + LDLM_FL_DESTROYED | LDLM_FL_FAIL_NOTIFIED; struct l_wait_info lwi; if (lock->l_completion_ast) { int err = lock->l_completion_ast(lock, @@ -1239,12 +1397,11 @@ ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags, lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(obd_timeout), NULL, LWI_ON_SIGNAL_NOOP, NULL); - /* 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_failed, - &lwi); - if (!(lock->l_flags & LDLM_FL_LVB_READY)) { + /* XXX FIXME see comment on CAN_MATCH in lustre_dlm.h */ + l_wait_event(lock->l_waitq, + lock->l_flags & wait_flags, + &lwi); + if (!ldlm_is_lvb_ready(lock)) { if (flags & LDLM_FL_TEST_LOCK) LDLM_LOCK_RELEASE(lock); else @@ -1299,11 +1456,10 @@ ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh, lock = ldlm_handle2lock(lockh); if (lock != NULL) { lock_res_and_lock(lock); - if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED || - lock->l_failed) + if (LDLM_HAVE_MASK(lock, GONE)) GOTO(out, mode); - if (lock->l_flags & LDLM_FL_CBPENDING && + if (ldlm_is_cbpending(lock) && lock->l_readers == 0 && lock->l_writers == 0) GOTO(out, mode); @@ -1324,13 +1480,116 @@ out: } EXPORT_SYMBOL(ldlm_revalidate_lock_handle); -/* Returns a referenced lock */ +/** The caller must guarantee that the buffer is large enough. */ +int ldlm_fill_lvb(struct ldlm_lock *lock, struct req_capsule *pill, + enum req_location loc, void *data, int size) +{ + void *lvb; + ENTRY; + + LASSERT(data != NULL); + LASSERT(size >= 0); + + switch (lock->l_lvb_type) { + case LVB_T_OST: + if (size == sizeof(struct ost_lvb)) { + if (loc == RCL_CLIENT) + lvb = req_capsule_client_swab_get(pill, + &RMF_DLM_LVB, + lustre_swab_ost_lvb); + else + lvb = req_capsule_server_swab_get(pill, + &RMF_DLM_LVB, + lustre_swab_ost_lvb); + if (unlikely(lvb == NULL)) { + LDLM_ERROR(lock, "no LVB"); + RETURN(-EPROTO); + } + + memcpy(data, lvb, size); + } else if (size == sizeof(struct ost_lvb_v1)) { + struct ost_lvb *olvb = data; + + if (loc == RCL_CLIENT) + lvb = req_capsule_client_swab_get(pill, + &RMF_DLM_LVB, + lustre_swab_ost_lvb_v1); + else + lvb = req_capsule_server_sized_swab_get(pill, + &RMF_DLM_LVB, size, + lustre_swab_ost_lvb_v1); + if (unlikely(lvb == NULL)) { + LDLM_ERROR(lock, "no LVB"); + RETURN(-EPROTO); + } + + memcpy(data, lvb, size); + olvb->lvb_mtime_ns = 0; + olvb->lvb_atime_ns = 0; + olvb->lvb_ctime_ns = 0; + } else { + LDLM_ERROR(lock, "Replied unexpected ost LVB size %d", + size); + RETURN(-EINVAL); + } + break; + case LVB_T_LQUOTA: + if (size == sizeof(struct lquota_lvb)) { + if (loc == RCL_CLIENT) + lvb = req_capsule_client_swab_get(pill, + &RMF_DLM_LVB, + lustre_swab_lquota_lvb); + else + lvb = req_capsule_server_swab_get(pill, + &RMF_DLM_LVB, + lustre_swab_lquota_lvb); + if (unlikely(lvb == NULL)) { + LDLM_ERROR(lock, "no LVB"); + RETURN(-EPROTO); + } + + memcpy(data, lvb, size); + } else { + LDLM_ERROR(lock, "Replied unexpected lquota LVB size %d", + size); + RETURN(-EINVAL); + } + break; + case LVB_T_LAYOUT: + if (size == 0) + break; + + if (loc == RCL_CLIENT) + lvb = req_capsule_client_get(pill, &RMF_DLM_LVB); + else + lvb = req_capsule_server_get(pill, &RMF_DLM_LVB); + if (unlikely(lvb == NULL)) { + LDLM_ERROR(lock, "no LVB"); + RETURN(-EPROTO); + } + + memcpy(data, lvb, size); + break; + default: + LDLM_ERROR(lock, "Unknown LVB type: %d\n", lock->l_lvb_type); + libcfs_debug_dumpstack(NULL); + RETURN(-EINVAL); + } + + RETURN(0); +} + +/** + * Create and fill in new LDLM lock with specified properties. + * Returns a referenced lock + */ struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns, const struct ldlm_res_id *res_id, ldlm_type_t type, ldlm_mode_t mode, const struct ldlm_callback_suite *cbs, - void *data, __u32 lvb_len) + void *data, __u32 lvb_len, + enum lvb_type lvb_type) { struct ldlm_lock *lock; struct ldlm_resource *res; @@ -1345,15 +1604,15 @@ struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns, if (lock == NULL) RETURN(NULL); - 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_req_mode = mode; + lock->l_ast_data = data; + lock->l_pid = current_pid(); + if (ns_is_server(ns)) + ldlm_set_ns_srv(lock); if (cbs) { lock->l_blocking_ast = cbs->lcs_blocking; lock->l_completion_ast = cbs->lcs_completion; lock->l_glimpse_ast = cbs->lcs_glimpse; - lock->l_weigh_ast = cbs->lcs_weigh; } lock->l_tree_node = NULL; @@ -1365,11 +1624,12 @@ struct ldlm_lock *ldlm_lock_create(struct ldlm_namespace *ns, if (lvb_len) { lock->l_lvb_len = lvb_len; - OBD_ALLOC(lock->l_lvb_data, lvb_len); + OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len); if (lock->l_lvb_data == NULL) GOTO(out, 0); } + lock->l_lvb_type = lvb_type; if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK)) GOTO(out, 0); @@ -1381,9 +1641,19 @@ out: return NULL; } +/** + * Enqueue (request) a lock. + * + * Does not block. As a result of enqueue the lock would be put + * into granted or waiting list. + * + * If namespace has intent policy sent and the lock has LDLM_FL_HAS_INTENT flag + * set, skip all the enqueueing and delegate lock processing to intent policy + * function. + */ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, struct ldlm_lock **lockp, - void *cookie, int *flags) + void *cookie, __u64 *flags) { struct ldlm_lock *lock = *lockp; struct ldlm_resource *res = lock->l_resource; @@ -1424,7 +1694,7 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, * have to allocate the interval node early otherwise we can't regrant * this lock in the future. - jay */ if (!local && (*flags & LDLM_FL_REPLAY) && res->lr_type == LDLM_EXTENT) - OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO); + OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, __GFP_IO); lock_res_and_lock(lock); if (local && lock->l_req_mode == lock->l_granted_mode) { @@ -1433,7 +1703,7 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, * need to do anything else. */ *flags &= ~(LDLM_FL_BLOCK_GRANTED | LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT); - GOTO(out, ELDLM_OK); + GOTO(out, rc = ELDLM_OK); } ldlm_resource_unlink_lock(lock); @@ -1448,21 +1718,22 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, node = NULL; } - /* Some flags from the enqueue want to make it into the AST, via the - * lock's l_flags. */ - lock->l_flags |= *flags & LDLM_AST_DISCARD_DATA; - - /* This distinction between local lock trees is very important; a client - * namespace only has information about locks taken by that client, and - * thus doesn't have enough information to decide for itself if it can - * be granted (below). In this case, we do exactly what the server - * tells us to do, as dictated by the 'flags'. - * - * We do exactly the same thing during recovery, when the server is - * more or less trusting the clients not to lie. - * - * FIXME (bug 268): Detect obvious lies by checking compatibility in - * granted/converting queues. */ + /* Some flags from the enqueue want to make it into the AST, via the + * lock's l_flags. */ + if (*flags & LDLM_FL_AST_DISCARD_DATA) + ldlm_set_ast_discard_data(lock); + + /* This distinction between local lock trees is very important; a client + * namespace only has information about locks taken by that client, and + * thus doesn't have enough information to decide for itself if it can + * be granted (below). In this case, we do exactly what the server + * tells us to do, as dictated by the 'flags'. + * + * We do exactly the same thing during recovery, when the server is + * more or less trusting the clients not to lie. + * + * FIXME (bug 268): Detect obvious lies by checking compatibility in + * granted/converting queues. */ if (local) { if (*flags & LDLM_FL_BLOCK_CONV) ldlm_resource_add_lock(res, &res->lr_converting, lock); @@ -1470,18 +1741,18 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, ldlm_resource_add_lock(res, &res->lr_waiting, lock); else ldlm_grant_lock(lock, NULL); - GOTO(out, ELDLM_OK); + GOTO(out, rc = 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); - GOTO(out, ELDLM_OK); + GOTO(out, rc = ELDLM_OK); } else if (*flags & LDLM_FL_BLOCK_WAIT) { ldlm_resource_add_lock(res, &res->lr_waiting, lock); - GOTO(out, ELDLM_OK); + GOTO(out, rc = ELDLM_OK); } else if (*flags & LDLM_FL_BLOCK_GRANTED) { ldlm_grant_lock(lock, NULL); - GOTO(out, ELDLM_OK); + GOTO(out, rc = ELDLM_OK); } /* If no flags, fall through to normal enqueue path. */ } @@ -1505,13 +1776,18 @@ out: } #ifdef HAVE_SERVER_SUPPORT -/* Must be called with namespace taken: queue is waiting or converting. */ +/** + * Iterate through all waiting locks on a given resource queue and attempt to + * grant them. + * + * Must be called with resource lock held. + */ int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue, cfs_list_t *work_list) { cfs_list_t *tmp, *pos; ldlm_processing_policy policy; - int flags; + __u64 flags; int rc = LDLM_ITER_CONTINUE; ldlm_error_t err; ENTRY; @@ -1537,6 +1813,9 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue, } #endif +/** + * Process a call to blocking AST callback for a lock in ast_work list + */ static int ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { @@ -1555,7 +1834,7 @@ ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) lock_res_and_lock(lock); cfs_list_del_init(&lock->l_bl_ast); - LASSERT(lock->l_flags & LDLM_FL_AST_SENT); + LASSERT(ldlm_is_ast_sent(lock)); LASSERT(lock->l_bl_ast_run == 0); LASSERT(lock->l_blocking_lock); lock->l_bl_ast_run++; @@ -1571,6 +1850,9 @@ ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) RETURN(rc); } +/** + * Process a call to completion AST callback for a lock in ast_work list + */ static int ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { @@ -1599,11 +1881,11 @@ ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) /* 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); + LASSERT(ldlm_is_cp_reqd(lock)); /* 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; + ldlm_clear_cp_reqd(lock); unlock_res_and_lock(lock); if (completion_callback != NULL) @@ -1613,6 +1895,9 @@ ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) RETURN(rc); } +/** + * Process a call to revocation AST callback for a lock in ast_work list + */ static int ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { @@ -1639,6 +1924,9 @@ ldlm_work_revoke_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) RETURN(rc); } +/** + * Process a call to glimpse AST callback for a lock in ast_work list + */ int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { struct ldlm_cb_set_arg *arg = opaq; @@ -1671,6 +1959,12 @@ int ldlm_work_gl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) RETURN(rc); } +/** + * Process list of locks in need of ASTs being sent. + * + * Used on server to send multiple ASTs together instead of sending one by + * one. + */ int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list, ldlm_desc_ast_t ast_type) { @@ -1685,7 +1979,7 @@ int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list, if (arg == NULL) RETURN(-ENOMEM); - cfs_atomic_set(&arg->restart, 0); + atomic_set(&arg->restart, 0); arg->list = rpc_list; switch (ast_type) { @@ -1721,7 +2015,7 @@ int ldlm_run_ast_work(struct ldlm_namespace *ns, cfs_list_t *rpc_list, ptlrpc_set_wait(arg->set); ptlrpc_set_destroy(arg->set); - rc = cfs_atomic_read(&arg->restart) ? -ERESTART : 0; + rc = atomic_read(&arg->restart) ? -ERESTART : 0; GOTO(out, rc); out: OBD_FREE_PTR(arg); @@ -1745,6 +2039,10 @@ static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd, return rc == LDLM_ITER_STOP; } +/** + * Iterate through all resources on a namespace attempting to grant waiting + * locks. + */ void ldlm_reprocess_all_ns(struct ldlm_namespace *ns) { ENTRY; @@ -1757,6 +2055,14 @@ void ldlm_reprocess_all_ns(struct ldlm_namespace *ns) } EXPORT_SYMBOL(ldlm_reprocess_all_ns); +/** + * Try to grant all waiting locks on a resource. + * + * Calls ldlm_reprocess_queue on converting and waiting queues. + * + * Typically called after some resource locks are cancelled to see + * if anything could be granted as a result of the cancellation. + */ void ldlm_reprocess_all(struct ldlm_resource *res) { CFS_LIST_HEAD(rpc_list); @@ -1793,14 +2099,18 @@ restart: #endif EXIT; } +EXPORT_SYMBOL(ldlm_reprocess_all); +/** + * Helper function to call blocking AST for LDLM lock \a lock in a + * "cancelling" mode. + */ void ldlm_cancel_callback(struct ldlm_lock *lock) { - check_res_locked(lock->l_resource); - if (!(lock->l_flags & LDLM_FL_CANCEL)) { - lock->l_flags |= LDLM_FL_CANCEL; - if (lock->l_blocking_ast) { - // l_check_no_ns_lock(ns); + check_res_locked(lock->l_resource); + if (!ldlm_is_cancel(lock)) { + ldlm_set_cancel(lock); + if (lock->l_blocking_ast) { unlock_res_and_lock(lock); lock->l_blocking_ast(lock, NULL, lock->l_ast_data, LDLM_CB_CANCELING); @@ -1809,9 +2119,12 @@ void ldlm_cancel_callback(struct ldlm_lock *lock) LDLM_DEBUG(lock, "no blocking ast"); } } - lock->l_flags |= LDLM_FL_BL_DONE; + ldlm_set_bl_done(lock); } +/** + * Remove skiplist-enabled LDLM lock \a req from granted list + */ void ldlm_unlink_lock_skiplist(struct ldlm_lock *req) { if (req->l_resource->lr_type != LDLM_PLAIN && @@ -1822,6 +2135,9 @@ void ldlm_unlink_lock_skiplist(struct ldlm_lock *req) cfs_list_del_init(&req->l_sl_mode); } +/** + * Attempts to cancel LDLM lock \a lock that has no reader/writer references. + */ void ldlm_lock_cancel(struct ldlm_lock *lock) { struct ldlm_resource *res; @@ -1840,15 +2156,15 @@ void ldlm_lock_cancel(struct ldlm_lock *lock) LBUG(); } - if (lock->l_waited) + if (ldlm_is_waited(lock)) 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 */ - if (lock->l_waited) + /* Yes, second time, just in case it was added again while we were + * running with no res lock in ldlm_cancel_callback */ + if (ldlm_is_waited(lock)) ldlm_del_waiting_lock(lock); ldlm_resource_unlink_lock(lock); @@ -1866,6 +2182,9 @@ void ldlm_lock_cancel(struct ldlm_lock *lock) } EXPORT_SYMBOL(ldlm_lock_cancel); +/** + * Set opaque data into the lock that only makes sense to upper layer. + */ int ldlm_lock_set_data(struct lustre_handle *lockh, void *data) { struct ldlm_lock *lock = ldlm_handle2lock(lockh); @@ -1888,6 +2207,10 @@ struct export_cl_data { int ecl_loop; }; +/** + * Iterator function for ldlm_cancel_locks_for_export. + * Cancels passed locks. + */ int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd, cfs_hlist_node_t *hnode, void *data) @@ -1913,12 +2236,17 @@ int ldlm_cancel_locks_for_export_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd, "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)); + atomic_read(&hs->hs_count)); } return 0; } +/** + * Cancel all locks for given export. + * + * Typically called on client disconnection/eviction + */ void ldlm_cancel_locks_for_export(struct obd_export *exp) { struct export_cl_data ecl = { @@ -1935,6 +2263,7 @@ void ldlm_cancel_locks_for_export(struct obd_export *exp) * * A fast variant of ldlm_lock_convert for convertion of exclusive * locks. The convertion is always successful. + * Used by Commit on Sharing (COS) code. * * \param lock A lock to convert * \param new_mode new lock mode @@ -1963,6 +2292,13 @@ void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode) } EXPORT_SYMBOL(ldlm_lock_downgrade); +/** + * Attempt to convert already granted lock to a different mode. + * + * While lock conversion is not currently used, future client-side + * optimizations could take advantage of it to avoid discarding cached + * pages on a file. + */ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, __u32 *flags) { @@ -1977,15 +2313,16 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, struct ldlm_interval *node; ENTRY; - if (new_mode == lock->l_granted_mode) { // No changes? Just return. + /* Just return if mode is unchanged. */ + if (new_mode == lock->l_granted_mode) { *flags |= LDLM_FL_BLOCK_GRANTED; RETURN(lock->l_resource); } /* I can't check the type of lock here because the bitlock of lock * is not held here, so do the allocation blindly. -jay */ - OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, CFS_ALLOC_IO); - if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */ + OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, __GFP_IO); + if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */ RETURN(NULL); LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR), @@ -2035,7 +2372,7 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, } else { /* This should never happen, because of the way the * server handles conversions. */ - LDLM_ERROR(lock, "Erroneous flags %d on local lock\n", + LDLM_ERROR(lock, "Erroneous flags %x on local lock\n", *flags); LBUG(); @@ -2049,7 +2386,7 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, } else { int rc; ldlm_error_t err; - int pflags = 0; + __u64 pflags = 0; ldlm_processing_policy policy; policy = ldlm_processing_policy_table[res->lr_type]; rc = policy(lock, &pflags, 0, &err, &rpc_list); @@ -2083,6 +2420,11 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, } EXPORT_SYMBOL(ldlm_lock_convert); +/** + * Print lock with lock handle \a lockh description into debug log. + * + * Used when printing all locks on a resource for debug purposes. + */ void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh) { struct ldlm_lock *lock; @@ -2100,6 +2442,10 @@ void ldlm_lock_dump_handle(int level, struct lustre_handle *lockh) } EXPORT_SYMBOL(ldlm_lock_dump_handle); +/** + * Print lock information with custom message into debug log. + * Helper function. + */ void _ldlm_lock_debug(struct ldlm_lock *lock, struct libcfs_debug_msg_data *msgdata, const char *fmt, ...) @@ -2122,109 +2468,109 @@ void _ldlm_lock_debug(struct ldlm_lock *lock, libcfs_debug_vmsg2(msgdata, fmt, args, " ns: \?\? lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s " - "remote: "LPX64" expref: %d pid: %u timeout: %lu\n", + "remote: "LPX64" expref: %d pid: %u timeout: %lu " + "lvb_type: %d\n", lock, - lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc), + lock->l_handle.h_cookie, 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, nid, lock->l_remote_handle.cookie, - exp ? cfs_atomic_read(&exp->exp_refcount) : -99, - lock->l_pid, lock->l_callback_timeout); + exp ? atomic_read(&exp->exp_refcount) : -99, + lock->l_pid, lock->l_callback_timeout, lock->l_lvb_type); va_end(args); return; } - switch (resource->lr_type) { - case LDLM_EXTENT: - 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" 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], - 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, 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(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" 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], - 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, 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(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" 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], - resource->lr_name.name[0], - resource->lr_name.name[1], - lock->l_policy_data.l_inodebits.bits, - 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(msgdata, fmt, args, - " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " - "res: "LPU64"/"LPU64" rrc: %d type: %s 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], - 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; - } - va_end(args); + switch (resource->lr_type) { + case LDLM_EXTENT: + libcfs_debug_vmsg2(msgdata, fmt, args, + " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " + "res: "DLDLMRES" rrc: %d type: %s ["LPU64"->"LPU64"] " + "(req "LPU64"->"LPU64") flags: "LPX64" nid: %s remote: " + LPX64" expref: %d pid: %u timeout: %lu lvb_type: %d\n", + ldlm_lock_to_ns_name(lock), lock, + lock->l_handle.h_cookie, atomic_read(&lock->l_refc), + lock->l_readers, lock->l_writers, + ldlm_lockname[lock->l_granted_mode], + ldlm_lockname[lock->l_req_mode], + PLDLMRES(resource), + 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, nid, lock->l_remote_handle.cookie, + exp ? atomic_read(&exp->exp_refcount) : -99, + lock->l_pid, lock->l_callback_timeout, + lock->l_lvb_type); + break; + + case LDLM_FLOCK: + libcfs_debug_vmsg2(msgdata, fmt, args, + " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " + "res: "DLDLMRES" rrc: %d type: %s pid: %d " + "["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, atomic_read(&lock->l_refc), + lock->l_readers, lock->l_writers, + ldlm_lockname[lock->l_granted_mode], + ldlm_lockname[lock->l_req_mode], + PLDLMRES(resource), + 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, nid, lock->l_remote_handle.cookie, + exp ? atomic_read(&exp->exp_refcount) : -99, + lock->l_pid, lock->l_callback_timeout); + break; + + case LDLM_IBITS: + libcfs_debug_vmsg2(msgdata, fmt, args, + " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " + "res: "DLDLMRES" bits "LPX64" rrc: %d type: %s " + "flags: "LPX64" nid: %s remote: "LPX64" expref: %d " + "pid: %u timeout: %lu lvb_type: %d\n", + ldlm_lock_to_ns_name(lock), + lock, lock->l_handle.h_cookie, + atomic_read(&lock->l_refc), + lock->l_readers, lock->l_writers, + ldlm_lockname[lock->l_granted_mode], + ldlm_lockname[lock->l_req_mode], + PLDLMRES(resource), + lock->l_policy_data.l_inodebits.bits, + atomic_read(&resource->lr_refcount), + ldlm_typename[resource->lr_type], + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? atomic_read(&exp->exp_refcount) : -99, + lock->l_pid, lock->l_callback_timeout, + lock->l_lvb_type); + break; + + default: + libcfs_debug_vmsg2(msgdata, fmt, args, + " ns: %s lock: %p/"LPX64" lrc: %d/%d,%d mode: %s/%s " + "res: "DLDLMRES" rrc: %d type: %s flags: "LPX64" " + "nid: %s remote: "LPX64" expref: %d pid: %u " + "timeout: %lu lvb_type: %d\n", + ldlm_lock_to_ns_name(lock), + lock, lock->l_handle.h_cookie, + atomic_read(&lock->l_refc), + lock->l_readers, lock->l_writers, + ldlm_lockname[lock->l_granted_mode], + ldlm_lockname[lock->l_req_mode], + PLDLMRES(resource), + atomic_read(&resource->lr_refcount), + ldlm_typename[resource->lr_type], + lock->l_flags, nid, lock->l_remote_handle.cookie, + exp ? atomic_read(&exp->exp_refcount) : -99, + lock->l_pid, lock->l_callback_timeout, + lock->l_lvb_type); + break; + } + va_end(args); } EXPORT_SYMBOL(_ldlm_lock_debug);