X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fldlm%2Fldlm_lock.c;h=39abaefd100a60f451fce64d46c3c0550be2d625;hp=c9f4536e4f8591a1545c8e11dc4dac6397145162;hb=8f01f8b51d114b0d2d54a5ab7db3161782e52447;hpb=d8adb5057d9fd878cc80f881202c96d18eb7a359 diff --git a/lustre/ldlm/ldlm_lock.c b/lustre/ldlm/ldlm_lock.c index c9f4536..39abaef 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. @@ -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, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,38 +41,73 @@ #define DEBUG_SUBSYSTEM S_LDLM -#ifdef __KERNEL__ -# include -# ifndef HAVE_VFS_INTENT_PATCHES -# include -# endif -#else -# include -#endif - +#include #include #include "ldlm_internal.h" /* lock types */ char *ldlm_lockname[] = { - [0] "--", - [LCK_EX] "EX", - [LCK_PW] "PW", - [LCK_PR] "PR", - [LCK_CW] "CW", - [LCK_CR] "CR", - [LCK_NL] "NL", - [LCK_GROUP] "GROUP", - [LCK_COS] "COS" + [0] = "--", + [LCK_EX] = "EX", + [LCK_PW] = "PW", + [LCK_PR] = "PR", + [LCK_CW] = "CW", + [LCK_CR] = "CR", + [LCK_NL] = "NL", + [LCK_GROUP] = "GROUP", + [LCK_COS] = "COS" }; +EXPORT_SYMBOL(ldlm_lockname); char *ldlm_typename[] = { - [LDLM_PLAIN] "PLN", - [LDLM_EXTENT] "EXT", - [LDLM_FLOCK] "FLK", - [LDLM_IBITS] "IBT", + [LDLM_PLAIN] = "PLN", + [LDLM_EXTENT] = "EXT", + [LDLM_FLOCK] = "FLK", + [LDLM_IBITS] = "IBT", +}; + +static ldlm_policy_wire_to_local_t ldlm_policy_wire_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_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_wire_to_local, +}; + +static ldlm_policy_local_to_wire_t ldlm_policy_local_to_wire[] = { + [LDLM_PLAIN - LDLM_MIN_TYPE] = ldlm_plain_policy_local_to_wire, + [LDLM_EXTENT - LDLM_MIN_TYPE] = ldlm_extent_policy_local_to_wire, + [LDLM_FLOCK - LDLM_MIN_TYPE] = ldlm_flock_policy_local_to_wire, + [LDLM_IBITS - LDLM_MIN_TYPE] = ldlm_ibits_policy_local_to_wire, }; +/** + * Converts lock policy from local format to on the wire lock_desc format + */ +void ldlm_convert_policy_to_wire(enum ldlm_type type, + const union ldlm_policy_data *lpolicy, + union ldlm_wire_policy_data *wpolicy) +{ + ldlm_policy_local_to_wire_t convert; + + convert = ldlm_policy_local_to_wire[type - LDLM_MIN_TYPE]; + + convert(lpolicy, wpolicy); +} + +/** + * Converts lock policy from on the wire lock_desc format to local format + */ +void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type, + const union ldlm_wire_policy_data *wpolicy, + union ldlm_policy_data *lpolicy) +{ + ldlm_policy_wire_to_local_t convert; + + convert = ldlm_policy_wire_to_local[type - LDLM_MIN_TYPE]; + + convert(wpolicy, lpolicy); +} + char *ldlm_it2str(int it) { switch (it) { @@ -92,39 +127,46 @@ char *ldlm_it2str(int it) return "unlink"; case IT_GETXATTR: return "getxattr"; + case IT_LAYOUT: + return "layout"; default: CERROR("Unknown intent %d\n", it); return "UNKNOWN"; } } +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[] = { - [LDLM_PLAIN] ldlm_process_plain_lock, - [LDLM_EXTENT] ldlm_process_extent_lock, -#ifdef __KERNEL__ - [LDLM_FLOCK] ldlm_process_flock_lock, -#endif - [LDLM_IBITS] ldlm_process_inodebits_lock, + [LDLM_PLAIN] = ldlm_process_plain_lock, + [LDLM_EXTENT] = ldlm_process_extent_lock, + [LDLM_FLOCK] = ldlm_process_flock_lock, + [LDLM_IBITS] = ldlm_process_inodebits_lock, }; ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res) { return ldlm_processing_policy_table[res->lr_type]; } +EXPORT_SYMBOL(ldlm_get_processing_policy); +#endif /* HAVE_SERVER_SUPPORT */ void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg) { ns->ns_policy = arg; } +EXPORT_SYMBOL(ldlm_register_intent); /* * REFCOUNTED LOCK OBJECTS */ -/* +/** + * 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 @@ -132,32 +174,33 @@ void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg) */ 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); -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)); -} - +/** + * 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(cfs_list_empty(&lock->l_res_link)); - LASSERT(cfs_list_empty(&lock->l_pending_chain)); + LASSERT(ldlm_is_destroyed(lock)); + LASSERT(list_empty(&lock->l_exp_list)); + LASSERT(list_empty(&lock->l_res_link)); + LASSERT(list_empty(&lock->l_pending_chain)); lprocfs_counter_decr(ldlm_res_to_ns(res)->ns_stats, LDLM_NSS_LOCKS); @@ -170,127 +213,169 @@ 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); - OBD_FREE_RCU_CB(lock, sizeof(*lock), &lock->l_handle, - ldlm_lock_free); + OBD_FREE_RCU(lock, sizeof(*lock), &lock->l_handle); } EXIT; } +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 (!list_empty(&lock->l_lru)) { + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + + LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); + list_del_init(&lock->l_lru); + LASSERT(ns->ns_nr_unused > 0); + ns->ns_nr_unused--; + rc = 1; + } + return rc; } -int ldlm_lock_remove_from_lru(struct ldlm_lock *lock) +/** + * Removes LDLM lock \a lock from LRU. Obtains the LRU lock first. + * + * If \a last_use is non-zero, it will remove the lock from LRU only if + * it matches lock's l_last_used. + * + * \retval 0 if \a last_use is set, the lock is not in LRU list or \a last_use + * doesn't match lock's l_last_used; + * otherwise, the lock hasn't been in the LRU list. + * \retval 1 the lock was in LRU list and removed. + */ +int ldlm_lock_remove_from_lru_check(struct ldlm_lock *lock, cfs_time_t last_use) { - struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); - int rc; + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + int rc = 0; - ENTRY; - if (lock->l_ns_srv) { - LASSERT(cfs_list_empty(&lock->l_lru)); - RETURN(0); - } + ENTRY; + if (ldlm_is_ns_srv(lock)) { + LASSERT(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); + if (last_use == 0 || last_use == lock->l_last_used) + rc = ldlm_lock_remove_from_lru_nolock(lock); + spin_unlock(&ns->ns_lock); + + 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); - - 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++; + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + + lock->l_last_used = cfs_time_current(); + LASSERT(list_empty(&lock->l_lru)); + LASSERT(lock->l_resource->lr_type != LDLM_FLOCK); + 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); - - ENTRY; - if (lock->l_ns_srv) { - 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; + struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); + + ENTRY; + if (ldlm_is_ns_srv(lock)) { + LASSERT(list_empty(&lock->l_lru)); + EXIT; + return; + } + + spin_lock(&ns->ns_lock); + if (!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 */ -int ldlm_lock_destroy_internal(struct ldlm_lock *lock) + * Because it's not in the hash table anymore. -phil + */ +static int ldlm_lock_destroy_internal(struct ldlm_lock *lock) { ENTRY; 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)) { + if (!list_empty(&lock->l_res_link)) { LDLM_ERROR(lock, "lock still on resource"); - ldlm_lock_dump(D_ERROR, lock, 0); LBUG(); } - if (lock->l_destroyed) { - LASSERT(cfs_list_empty(&lock->l_lru)); - EXIT; - return 0; - } - 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 (ldlm_is_destroyed(lock)) { + LASSERT(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 + * in exp_lock_hash. */ + /* In the function below, .hs_keycmp resolves to + * ldlm_export_lock_keycmp() */ + /* coverity[overrun-buffer-val] */ + 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); @@ -309,6 +394,9 @@ 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; @@ -325,6 +413,9 @@ 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; @@ -344,58 +435,79 @@ 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); +} + +static struct portals_handle_ops lock_handle_ops = { + .hop_addref = lock_handle_addref, + .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 */ static struct ldlm_lock *ldlm_lock_new(struct ldlm_resource *resource) { - struct ldlm_lock *lock; - ENTRY; - - if (resource == NULL) - LBUG(); - - OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, CFS_ALLOC_IO); - if (lock == NULL) - RETURN(NULL); - - cfs_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); + struct ldlm_lock *lock; + ENTRY; + + if (resource == NULL) + LBUG(); + + OBD_SLAB_ALLOC_PTR_GFP(lock, ldlm_lock_slab, GFP_NOFS); + if (lock == NULL) + RETURN(NULL); + + spin_lock_init(&lock->l_lock); + lock->l_resource = resource; + lu_ref_add(&resource->lr_reference, "lock", lock); + + atomic_set(&lock->l_refc, 2); + INIT_LIST_HEAD(&lock->l_res_link); + INIT_LIST_HEAD(&lock->l_lru); + INIT_LIST_HEAD(&lock->l_pending_chain); + INIT_LIST_HEAD(&lock->l_bl_ast); + INIT_LIST_HEAD(&lock->l_cp_ast); + INIT_LIST_HEAD(&lock->l_rk_ast); + init_waitqueue_head(&lock->l_waitq); + lock->l_blocking_lock = NULL; + INIT_LIST_HEAD(&lock->l_sl_mode); + INIT_LIST_HEAD(&lock->l_sl_policy); + INIT_HLIST_NODE(&lock->l_exp_hash); + 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); + INIT_LIST_HEAD(&lock->l_handle.h_link); + class_handle_hash(&lock->l_handle, &lock_handle_ops); lu_ref_init(&lock->l_reference); lu_ref_add(&lock->l_reference, "hash", lock); lock->l_callback_timeout = 0; #if LUSTRE_TRACKS_LOCK_EXP_REFS - CFS_INIT_LIST_HEAD(&lock->l_exp_refs_link); + INIT_LIST_HEAD(&lock->l_exp_refs_link); lock->l_exp_refs_nr = 0; lock->l_exp_refs_target = NULL; #endif + INIT_LIST_HEAD(&lock->l_exp_list); 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) { @@ -417,14 +529,14 @@ int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock, LASSERT(new_resid->name[0] != 0); /* This function assumes that the lock isn't on any lists */ - LASSERT(cfs_list_empty(&lock->l_res_link)); + LASSERT(list_empty(&lock->l_res_link)); type = oldres->lr_type; unlock_res_and_lock(lock); - newres = ldlm_resource_get(ns, NULL, new_resid, type, 1); - if (newres == NULL) - RETURN(-ENOMEM); + newres = ldlm_resource_get(ns, NULL, new_resid, type, 1); + if (IS_ERR(newres)) + RETURN(PTR_ERR(newres)); lu_ref_add(&newres->lr_reference, "lock", lock); /* @@ -433,7 +545,7 @@ 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); @@ -455,141 +567,135 @@ int ldlm_lock_change_resource(struct ldlm_namespace *ns, struct ldlm_lock *lock, RETURN(0); } -/* - * 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; - - LASSERT(handle); - - lock = class_handle2object(handle->cookie); - 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); - } - - lock_res_and_lock(lock); - - 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); - } - - if (flags && (lock->l_flags & flags)) { - unlock_res_and_lock(lock); - LDLM_LOCK_PUT(lock); - RETURN(NULL); - } - - if (flags) - lock->l_flags |= flags; - - unlock_res_and_lock(lock); - RETURN(lock); + struct ldlm_lock *lock; + ENTRY; + + LASSERT(handle); + + 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) && !ldlm_is_destroyed(lock)) { + lu_ref_add(&lock->l_reference, "handle", current); + RETURN(lock); + } + + lock_res_and_lock(lock); + + LASSERT(lock->l_resource != 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 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); + } + + lock->l_flags |= flags; + } + + 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))) { - /* 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 == - (MDS_INODELOCK_LOOKUP|MDS_INODELOCK_UPDATE), - "Inappropriate inode lock bits during " - "conversion " LPU64 "\n", - lock->l_policy_data.l_inodebits.bits); - - ldlm_res2desc(lock->l_resource, &desc->l_resource); - desc->l_resource.lr_type = LDLM_PLAIN; - - /* Convert "new" lock mode to something old client can - understand */ - if ((lock->l_req_mode == LCK_CR) || - (lock->l_req_mode == LCK_CW)) - desc->l_req_mode = LCK_PR; - else - desc->l_req_mode = lock->l_req_mode; - if ((lock->l_granted_mode == LCK_CR) || - (lock->l_granted_mode == LCK_CW)) { - desc->l_granted_mode = LCK_PR; - } else { - /* We never grant PW/EX locks to clients */ - LASSERT((lock->l_granted_mode != LCK_PW) && - (lock->l_granted_mode != LCK_EX)); - desc->l_granted_mode = lock->l_granted_mode; - } - - /* We do not copy policy here, because there is no - policy for plain locks */ - } else { - ldlm_res2desc(lock->l_resource, &desc->l_resource); - desc->l_req_mode = lock->l_req_mode; - desc->l_granted_mode = lock->l_granted_mode; - desc->l_policy_data = lock->l_policy_data; - } + ldlm_res2desc(lock->l_resource, &desc->l_resource); + desc->l_req_mode = lock->l_req_mode; + desc->l_granted_mode = lock->l_granted_mode; + ldlm_convert_policy_to_wire(lock->l_resource->lr_type, + &lock->l_policy_data, + &desc->l_policy_data); } -void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, - cfs_list_t *work_list) +/** + * 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. + */ +static void ldlm_add_bl_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, + struct list_head *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; - LASSERT(cfs_list_empty(&lock->l_bl_ast)); - cfs_list_add(&lock->l_bl_ast, work_list); + 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(list_empty(&lock->l_bl_ast)); + list_add(&lock->l_bl_ast, work_list); LDLM_LOCK_GET(lock); LASSERT(lock->l_blocking_lock == NULL); lock->l_blocking_lock = LDLM_LOCK_GET(new); } } -void ldlm_add_cp_work_item(struct ldlm_lock *lock, cfs_list_t *work_list) +/** + * Add a lock to list of just granted locks to send completion AST to. + */ +static void ldlm_add_cp_work_item(struct ldlm_lock *lock, + struct list_head *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); + LASSERT(list_empty(&lock->l_cp_ast)); + list_add(&lock->l_cp_ast, work_list); LDLM_LOCK_GET(lock); } } -/* 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) + struct list_head *work_list) { ENTRY; check_res_locked(lock->l_resource); @@ -600,17 +706,31 @@ void ldlm_add_ast_work_item(struct ldlm_lock *lock, struct ldlm_lock *new, EXIT; } -void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode) +/** + * 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, enum ldlm_mode mode) { - struct ldlm_lock *lock; + struct ldlm_lock *lock; - lock = ldlm_handle2lock(lockh); - LASSERT(lock != NULL); - ldlm_lock_addref_internal(lock, mode); - LDLM_LOCK_PUT(lock); + lock = ldlm_handle2lock(lockh); + LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie); + ldlm_lock_addref_internal(lock, mode); + LDLM_LOCK_PUT(lock); } +EXPORT_SYMBOL(ldlm_lock_addref); -void ldlm_lock_addref_internal_nolock(struct ldlm_lock *lock, __u32 mode) +/** + * 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, + enum ldlm_mode mode) { ldlm_lock_remove_from_lru(lock); if (mode & (LCK_NL | LCK_CR | LCK_PR)) { @@ -627,14 +747,14 @@ 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 * * \retval -EAGAIN lock is being canceled. */ -int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode) +int ldlm_lock_addref_try(struct lustre_handle *lockh, enum ldlm_mode mode) { struct ldlm_lock *lock; int result; @@ -644,7 +764,7 @@ 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; } @@ -653,20 +773,29 @@ int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode) } return result; } +EXPORT_SYMBOL(ldlm_lock_addref_try); -/* only called for local locks */ -void ldlm_lock_addref_internal(struct ldlm_lock *lock, __u32 mode) +/** + * 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, enum ldlm_mode mode) { - lock_res_and_lock(lock); - ldlm_lock_addref_internal_nolock(lock, mode); - unlock_res_and_lock(lock); + lock_res_and_lock(lock); + ldlm_lock_addref_internal_nolock(lock, 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 */ -void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode) +/** + * 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, + enum ldlm_mode mode) { LDLM_DEBUG(lock, "ldlm_lock_decref(%s)", ldlm_lockname[mode]); if (mode & (LCK_NL | LCK_CR | LCK_PR)) { @@ -684,7 +813,15 @@ void ldlm_lock_decref_internal_nolock(struct ldlm_lock *lock, __u32 mode) LDLM_LOCK_RELEASE(lock); /* matches the LDLM_LOCK_GET() in addref */ } -void ldlm_lock_decref_internal(struct ldlm_lock *lock, __u32 mode) +/** + * 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, enum ldlm_mode mode) { struct ldlm_namespace *ns; ENTRY; @@ -695,19 +832,25 @@ 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)) { - /* If we received a blocked AST and this was the last reference, - * run the callback. */ - if (lock->l_ns_srv && lock->l_export) + if (!lock->l_readers && !lock->l_writers && + (ldlm_is_cbpending(lock) || lock->l_req_mode == LCK_GROUP)) { + /* If we received a blocked AST and this was the last reference, + * run the callback. + * Group locks are special: + * They must not go in LRU, but they are not called back + * like non-group locks, instead they are manually released. + * They have an l_writers reference which they keep until + * they are manually released, so we remove them when they have + * no more reader or writer references. - LU-6368 */ + if (ldlm_is_ns_srv(lock) && lock->l_export) CERROR("FL_CBPENDING set on non-local lock--just a " "warning\n"); @@ -717,21 +860,25 @@ 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_BL_AST)) { + !ldlm_is_no_lru(lock) && + !ldlm_is_bl_ast(lock)) { + + LDLM_DEBUG(lock, "add lock into lru list"); + /* If this is a client-side namespace and this was the last * reference, put it on the LRU. */ 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 @@ -739,25 +886,35 @@ 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); } EXIT; } -void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode) +/** + * Decrease reader/writer refcount for LDLM lock with handle \a lockh + */ +void ldlm_lock_decref(struct lustre_handle *lockh, enum ldlm_mode mode) { struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0); LASSERTF(lock != NULL, "Non-existing lock: "LPX64"\n", lockh->cookie); ldlm_lock_decref_internal(lock, mode); LDLM_LOCK_PUT(lock); } +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. */ -void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode) +/** + * 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. + * + */ +void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, + enum ldlm_mode mode) { struct ldlm_lock *lock = __ldlm_handle2lock(lockh, 0); ENTRY; @@ -766,23 +923,24 @@ 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); } +EXPORT_SYMBOL(ldlm_lock_decref_and_cancel); struct sl_insert_point { - cfs_list_t *res_link; - cfs_list_t *mode_link; - cfs_list_t *policy_link; + struct list_head *res_link; + struct list_head *mode_link; + struct list_head *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; @@ -792,18 +950,18 @@ struct sl_insert_point { * NOTE: called by * - ldlm_grant_lock_with_skiplist */ -static void search_granted_lock(cfs_list_t *queue, +static void search_granted_lock(struct list_head *queue, struct ldlm_lock *req, struct sl_insert_point *prev) { - cfs_list_t *tmp; + struct list_head *tmp; struct ldlm_lock *lock, *mode_end, *policy_end; ENTRY; - cfs_list_for_each(tmp, queue) { - lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link); + list_for_each(tmp, queue) { + lock = list_entry(tmp, struct ldlm_lock, l_res_link); - mode_end = cfs_list_entry(lock->l_sl_mode.prev, + mode_end = list_entry(lock->l_sl_mode.prev, struct ldlm_lock, l_sl_mode); if (lock->l_req_mode != req->l_req_mode) { @@ -823,7 +981,7 @@ static void search_granted_lock(cfs_list_t *queue, } else if (lock->l_resource->lr_type == LDLM_IBITS) { for (;;) { policy_end = - cfs_list_entry(lock->l_sl_policy.prev, + list_entry(lock->l_sl_policy.prev, struct ldlm_lock, l_sl_policy); @@ -847,7 +1005,7 @@ static void search_granted_lock(cfs_list_t *queue, /* go to next policy group within mode group */ tmp = policy_end->l_res_link.next; - lock = cfs_list_entry(tmp, struct ldlm_lock, + lock = list_entry(tmp, struct ldlm_lock, l_res_link); } /* loop over policy groups within the mode group */ @@ -873,6 +1031,10 @@ 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) { @@ -882,25 +1044,35 @@ 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) { + if (ldlm_is_destroyed(lock)) { CDEBUG(D_OTHER, "Lock destroyed, not adding to resource\n"); return; } - LASSERT(cfs_list_empty(&lock->l_res_link)); - 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); + LASSERT(list_empty(&lock->l_res_link)); + LASSERT(list_empty(&lock->l_sl_mode)); + LASSERT(list_empty(&lock->l_sl_policy)); + + /* + * 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) + list_add(&lock->l_res_link, prev->res_link); + if (&lock->l_sl_mode != prev->mode_link) + list_add(&lock->l_sl_mode, prev->mode_link); + if (&lock->l_sl_policy != prev->policy_link) + 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; @@ -913,14 +1085,18 @@ 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 * * must be called with lr_lock held */ -void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list) +void ldlm_grant_lock(struct ldlm_lock *lock, struct list_head *work_list) { struct ldlm_resource *res = lock->l_resource; ENTRY; @@ -928,6 +1104,21 @@ void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list) check_res_locked(res); lock->l_granted_mode = lock->l_req_mode; + + if (work_list && lock->l_completion_ast != NULL) + ldlm_add_ast_work_item(lock, NULL, work_list); + + /* We should not add locks to granted list in the following cases: + * - this is an UNLOCK but not a real lock; + * - this is a TEST lock; + * - this is a F_CANCELLK lock (async flock has req_mode == 0) + * - this is a deadlock (flock cannot be granted) */ + if (lock->l_req_mode == 0 || + lock->l_req_mode == LCK_NL || + ldlm_is_test_lock(lock) || + ldlm_is_flock_deadlock(lock)) + RETURN_EXIT; + if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) ldlm_grant_lock_with_skiplist(lock); else if (res->lr_type == LDLM_EXTENT) @@ -935,106 +1126,222 @@ void ldlm_grant_lock(struct ldlm_lock *lock, cfs_list_t *work_list) else ldlm_resource_add_lock(res, &res->lr_granted, lock); - if (lock->l_granted_mode < res->lr_most_restr) - res->lr_most_restr = lock->l_granted_mode; - - if (work_list && lock->l_completion_ast != NULL) - ldlm_add_ast_work_item(lock, NULL, work_list); - ldlm_pool_add(&ldlm_res_to_ns(res)->ns_pool, lock); EXIT; } -/* returns 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) -{ - struct ldlm_lock *lock; - cfs_list_t *tmp; - - cfs_list_for_each(tmp, queue) { - ldlm_mode_t match; +/** + * Describe the overlap between two locks. itree_overlap_cb data. + */ +struct lock_match_data { + struct ldlm_lock *lmd_old; + struct ldlm_lock *lmd_lock; + enum ldlm_mode *lmd_mode; + union ldlm_policy_data *lmd_policy; + __u64 lmd_flags; + int lmd_unref; +}; - lock = cfs_list_entry(tmp, struct ldlm_lock, l_res_link); +/** + * Check if the given @lock meets the criteria for a match. + * A reference on the lock is taken if matched. + * + * \param lock test-against this lock + * \param data parameters + */ +static int lock_matches(struct ldlm_lock *lock, struct lock_match_data *data) +{ + union ldlm_policy_data *lpol = &lock->l_policy_data; + enum ldlm_mode match; + + if (lock == data->lmd_old) + return INTERVAL_ITER_STOP; + + /* Check if this lock can be matched. + * Used by LU-2919(exclusive open) for open lease lock */ + if (ldlm_is_excl(lock)) + return INTERVAL_ITER_CONT; + + /* 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 (ldlm_is_cbpending(lock) && + !(data->lmd_flags & LDLM_FL_CBPENDING)) + return INTERVAL_ITER_CONT; + if (!data->lmd_unref && ldlm_is_cbpending(lock) && + lock->l_readers == 0 && lock->l_writers == 0) + return INTERVAL_ITER_CONT; + + if (!(lock->l_req_mode & *data->lmd_mode)) + return INTERVAL_ITER_CONT; + match = lock->l_req_mode; + + switch (lock->l_resource->lr_type) { + case LDLM_EXTENT: + if (lpol->l_extent.start > data->lmd_policy->l_extent.start || + lpol->l_extent.end < data->lmd_policy->l_extent.end) + return INTERVAL_ITER_CONT; + + if (unlikely(match == LCK_GROUP) && + data->lmd_policy->l_extent.gid != LDLM_GID_ANY && + lpol->l_extent.gid != data->lmd_policy->l_extent.gid) + return INTERVAL_ITER_CONT; + break; + case LDLM_IBITS: + /* We match if we have existing lock with same or wider set + of bits. */ + if ((lpol->l_inodebits.bits & + data->lmd_policy->l_inodebits.bits) != + data->lmd_policy->l_inodebits.bits) + return INTERVAL_ITER_CONT; + break; + default: + ; + } + + /* We match if we have existing lock with same or wider set + of bits. */ + if (!data->lmd_unref && LDLM_HAVE_MASK(lock, GONE)) + return INTERVAL_ITER_CONT; + + if (!equi(data->lmd_flags & LDLM_FL_LOCAL_ONLY, ldlm_is_local(lock))) + return INTERVAL_ITER_CONT; + + if (data->lmd_flags & LDLM_FL_TEST_LOCK) { + LDLM_LOCK_GET(lock); + ldlm_lock_touch_in_lru(lock); + } else { + ldlm_lock_addref_internal_nolock(lock, match); + } + + *data->lmd_mode = match; + data->lmd_lock = lock; + + return INTERVAL_ITER_STOP; +} - if (lock == old_lock) - break; +static unsigned int itree_overlap_cb(struct interval_node *in, void *args) +{ + struct ldlm_interval *node = to_ldlm_interval(in); + struct lock_match_data *data = args; + struct ldlm_lock *lock; + int rc; + + list_for_each_entry(lock, &node->li_group, l_sl_policy) { + rc = lock_matches(lock, data); + if (rc == INTERVAL_ITER_STOP) + return INTERVAL_ITER_STOP; + } + return INTERVAL_ITER_CONT; +} - /* 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 && - !(flags & LDLM_FL_CBPENDING)) - continue; - if (!unref && lock->l_flags & LDLM_FL_CBPENDING && - lock->l_readers == 0 && lock->l_writers == 0) - continue; +/** + * Search for a lock with given parameters in interval trees. + * + * \param res search for a lock in this resource + * \param data parameters + * + * \retval a referenced lock or NULL. + */ +static struct ldlm_lock *search_itree(struct ldlm_resource *res, + struct lock_match_data *data) +{ + struct interval_node_extent ext = { + .start = data->lmd_policy->l_extent.start, + .end = data->lmd_policy->l_extent.end + }; + int idx; - if (!(lock->l_req_mode & *mode)) - continue; - match = lock->l_req_mode; + for (idx = 0; idx < LCK_MODE_NUM; idx++) { + struct ldlm_interval_tree *tree = &res->lr_itree[idx]; - if (lock->l_resource->lr_type == LDLM_EXTENT && - (lock->l_policy_data.l_extent.start > - policy->l_extent.start || - lock->l_policy_data.l_extent.end < policy->l_extent.end)) - continue; + if (tree->lit_root == NULL) + continue; - if (unlikely(match == LCK_GROUP) && - lock->l_resource->lr_type == LDLM_EXTENT && - lock->l_policy_data.l_extent.gid != policy->l_extent.gid) - continue; + if (!(tree->lit_mode & *data->lmd_mode)) + continue; - /* We match if we have existing lock with same or wider set - of bits. */ - if (lock->l_resource->lr_type == LDLM_IBITS && - ((lock->l_policy_data.l_inodebits.bits & - policy->l_inodebits.bits) != - policy->l_inodebits.bits)) - continue; + interval_search(tree->lit_root, &ext, + itree_overlap_cb, data); + } + return data->lmd_lock; +} - if (!unref && - (lock->l_destroyed || (lock->l_flags & LDLM_FL_FAILED))) - continue; - if ((flags & LDLM_FL_LOCAL_ONLY) && - !(lock->l_flags & LDLM_FL_LOCAL)) - continue; +/** + * Search for a lock with given properties in a queue. + * + * \param queue search for a lock in this queue + * \param data parameters + * + * \retval a referenced lock or NULL. + */ +static struct ldlm_lock *search_queue(struct list_head *queue, + struct lock_match_data *data) +{ + struct ldlm_lock *lock; + int rc; + + list_for_each_entry(lock, queue, l_res_link) { + rc = lock_matches(lock, data); + if (rc == INTERVAL_ITER_STOP) + return data->lmd_lock; + } + return NULL; +} - if (flags & LDLM_FL_TEST_LOCK) { - LDLM_LOCK_GET(lock); - ldlm_lock_touch_in_lru(lock); - } else { - ldlm_lock_addref_internal_nolock(lock, match); - } - *mode = match; - return lock; - } +void ldlm_lock_fail_match_locked(struct ldlm_lock *lock) +{ + 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); - return NULL; +void ldlm_lock_fail_match(struct ldlm_lock *lock) +{ + lock_res_and_lock(lock); + ldlm_lock_fail_match_locked(lock); + unlock_res_and_lock(lock); } +/** + * 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_signal(&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); ldlm_lock_allow_match_locked(lock); unlock_res_and_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. @@ -1051,56 +1358,66 @@ void ldlm_lock_allow_match(struct ldlm_lock *lock) * 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, - 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) +enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags, + const struct ldlm_res_id *res_id, + enum ldlm_type type, + union ldlm_policy_data *policy, + enum ldlm_mode mode, + struct lustre_handle *lockh, int unref) { - struct ldlm_resource *res; - struct ldlm_lock *lock, *old_lock = NULL; - int rc = 0; - ENTRY; - - if (ns == NULL) { - old_lock = ldlm_handle2lock(lockh); - LASSERT(old_lock); - - ns = ldlm_lock_to_ns(old_lock); - res_id = &old_lock->l_resource->lr_name; - type = old_lock->l_resource->lr_type; - mode = old_lock->l_req_mode; - } - - res = ldlm_resource_get(ns, NULL, res_id, type, 0); - if (res == NULL) { - LASSERT(old_lock == NULL); - RETURN(0); - } - - LDLM_RESOURCE_ADDREF(res); - lock_res(res); - - lock = search_queue(&res->lr_granted, &mode, policy, old_lock, - flags, unref); - if (lock != NULL) - GOTO(out, rc = 1); - if (flags & LDLM_FL_BLOCK_GRANTED) - GOTO(out, rc = 0); - lock = search_queue(&res->lr_converting, &mode, policy, old_lock, - flags, unref); - if (lock != NULL) - GOTO(out, rc = 1); - lock = search_queue(&res->lr_waiting, &mode, policy, old_lock, - flags, unref); - if (lock != NULL) - GOTO(out, rc = 1); + struct lock_match_data data = { + .lmd_old = NULL, + .lmd_lock = NULL, + .lmd_mode = &mode, + .lmd_policy = policy, + .lmd_flags = flags, + .lmd_unref = unref, + }; + struct ldlm_resource *res; + struct ldlm_lock *lock; + int rc = 0; + ENTRY; + + if (ns == NULL) { + data.lmd_old = ldlm_handle2lock(lockh); + LASSERT(data.lmd_old != NULL); + + ns = ldlm_lock_to_ns(data.lmd_old); + res_id = &data.lmd_old->l_resource->lr_name; + type = data.lmd_old->l_resource->lr_type; + *data.lmd_mode = data.lmd_old->l_req_mode; + } + + res = ldlm_resource_get(ns, NULL, res_id, type, 0); + if (IS_ERR(res)) { + LASSERT(data.lmd_old == NULL); + RETURN(0); + } + + LDLM_RESOURCE_ADDREF(res); + lock_res(res); + + if (res->lr_type == LDLM_EXTENT) + lock = search_itree(res, &data); + else + lock = search_queue(&res->lr_granted, &data); + if (lock != NULL) + GOTO(out, rc = 1); + if (flags & LDLM_FL_BLOCK_GRANTED) + GOTO(out, rc = 0); + lock = search_queue(&res->lr_converting, &data); + if (lock != NULL) + GOTO(out, rc = 1); + lock = search_queue(&res->lr_waiting, &data); + if (lock != NULL) + GOTO(out, rc = 1); EXIT; out: @@ -1111,7 +1428,9 @@ 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, @@ -1131,9 +1450,17 @@ 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), &lwi); + /* 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 + ldlm_lock_decref_internal(lock, mode); + rc = 0; + } } } out2: @@ -1165,82 +1492,231 @@ ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags, (type == LDLM_PLAIN || type == LDLM_IBITS) ? res_id->name[3] : policy->l_extent.end); } - if (old_lock) - LDLM_LOCK_PUT(old_lock); + if (data.lmd_old != NULL) + LDLM_LOCK_PUT(data.lmd_old); - return rc ? mode : 0; + return rc ? mode : 0; } +EXPORT_SYMBOL(ldlm_lock_match); -/* 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) +enum ldlm_mode ldlm_revalidate_lock_handle(struct lustre_handle *lockh, + __u64 *bits) { - struct ldlm_lock *lock; - struct ldlm_resource *res; - ENTRY; + struct ldlm_lock *lock; + enum ldlm_mode mode = 0; + ENTRY; - res = ldlm_resource_get(ns, NULL, res_id, type, 1); - if (res == NULL) - RETURN(NULL); + lock = ldlm_handle2lock(lockh); + if (lock != NULL) { + lock_res_and_lock(lock); + if (LDLM_HAVE_MASK(lock, GONE)) + GOTO(out, mode); - lock = ldlm_lock_new(res); + if (ldlm_is_cbpending(lock) && + lock->l_readers == 0 && lock->l_writers == 0) + GOTO(out, mode); - 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); - 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; + if (bits) + *bits = lock->l_policy_data.l_inodebits.bits; + mode = lock->l_granted_mode; + ldlm_lock_addref_internal_nolock(lock, mode); } - lock->l_tree_node = NULL; - /* if this is the extent lock, allocate the interval tree node */ - if (type == LDLM_EXTENT) { - if (ldlm_interval_alloc(lock) == NULL) - GOTO(out, 0); - } + EXIT; - if (lvb_len) { - lock->l_lvb_len = lvb_len; - OBD_ALLOC(lock->l_lvb_data, lvb_len); - if (lock->l_lvb_data == NULL) - GOTO(out, 0); +out: + if (lock != NULL) { + unlock_res_and_lock(lock); + LDLM_LOCK_PUT(lock); } + return mode; +} +EXPORT_SYMBOL(ldlm_revalidate_lock_handle); - if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK)) - GOTO(out, 0); +/** 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); +} - RETURN(lock); +/** + * 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, + enum ldlm_type type, + enum ldlm_mode mode, + const struct ldlm_callback_suite *cbs, + void *data, __u32 lvb_len, + enum lvb_type lvb_type) +{ + struct ldlm_lock *lock; + struct ldlm_resource *res; + int rc; + ENTRY; + + res = ldlm_resource_get(ns, NULL, res_id, type, 1); + if (IS_ERR(res)) + RETURN(ERR_CAST(res)); + + lock = ldlm_lock_new(res); + if (lock == NULL) + RETURN(ERR_PTR(-ENOMEM)); + + 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_tree_node = NULL; + /* if this is the extent lock, allocate the interval tree node */ + if (type == LDLM_EXTENT) + if (ldlm_interval_alloc(lock) == NULL) + GOTO(out, rc = -ENOMEM); + + if (lvb_len) { + lock->l_lvb_len = lvb_len; + OBD_ALLOC_LARGE(lock->l_lvb_data, lvb_len); + if (lock->l_lvb_data == NULL) + GOTO(out, rc = -ENOMEM); + } + + lock->l_lvb_type = lvb_type; + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_NEW_LOCK)) + GOTO(out, rc = -ENOENT); + + RETURN(lock); out: - ldlm_lock_destroy(lock); - LDLM_LOCK_RELEASE(lock); - return NULL; + ldlm_lock_destroy(lock); + LDLM_LOCK_RELEASE(lock); + RETURN(ERR_PTR(rc)); } -ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, - struct ldlm_lock **lockp, - void *cookie, int *flags) +/** + * 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. + */ +enum ldlm_error ldlm_lock_enqueue(struct ldlm_namespace *ns, + struct ldlm_lock **lockp, + void *cookie, __u64 *flags) { - struct ldlm_lock *lock = *lockp; - struct ldlm_resource *res = lock->l_resource; - int local = ns_is_client(ldlm_res_to_ns(res)); - ldlm_processing_policy policy; - ldlm_error_t rc = ELDLM_OK; - struct ldlm_interval *node = NULL; - ENTRY; + 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 + enum ldlm_error rc = ELDLM_OK; + struct ldlm_interval *node = NULL; + ENTRY; - lock->l_last_activity = cfs_time_current_sec(); /* policies are not executed on the client or during replay */ if ((*flags & (LDLM_FL_HAS_INTENT|LDLM_FL_REPLAY)) == LDLM_FL_HAS_INTENT && !local && ns->ns_policy) { @@ -1264,21 +1740,33 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, } } - /* For a replaying lock, it might be already in granted list. So - * unlinking the lock will cause the interval node to be freed, we - * 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); + if (*flags & LDLM_FL_RESENT) { + /* Reconstruct LDLM_FL_SRV_ENQ_MASK @flags for reply. + * Set LOCK_CHANGED always. + * Check if the lock is granted for BLOCK_GRANTED. + * Take NO_TIMEOUT from the lock as it is inherited through + * LDLM_FL_INHERIT_MASK */ + *flags |= LDLM_FL_LOCK_CHANGED; + if (lock->l_req_mode != lock->l_granted_mode) + *flags |= LDLM_FL_BLOCK_GRANTED; + *flags |= lock->l_flags & LDLM_FL_NO_TIMEOUT; + RETURN(ELDLM_OK); + } + + /* For a replaying lock, it might be already in granted list. So + * unlinking the lock will cause the interval node to be freed, we + * 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, GFP_NOFS); lock_res_and_lock(lock); if (local && lock->l_req_mode == lock->l_granted_mode) { /* The server returned a blocked lock, but it was granted * before we got a chance to actually enqueue it. We don't * need to do anything else. */ - *flags &= ~(LDLM_FL_BLOCK_GRANTED | - LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_WAIT); - GOTO(out, ELDLM_OK); + *flags &= ~LDLM_FL_BLOCKED_MASK; + GOTO(out, rc = ELDLM_OK); } ldlm_resource_unlink_lock(lock); @@ -1288,26 +1776,29 @@ ldlm_error_t ldlm_lock_enqueue(struct ldlm_namespace *ns, GOTO(out, rc = -ENOMEM); } - CFS_INIT_LIST_HEAD(&node->li_group); + INIT_LIST_HEAD(&node->li_group); ldlm_interval_attach(node, lock); 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); + if (*flags & LDLM_FL_TEST_LOCK) + ldlm_set_test_lock(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); @@ -1315,17 +1806,18 @@ 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. */ } @@ -1333,6 +1825,14 @@ 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) @@ -1340,25 +1840,32 @@ out: return rc; } -/* 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) +#ifdef HAVE_SERVER_SUPPORT +/** + * 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, struct list_head *queue, + struct list_head *work_list) { - cfs_list_t *tmp, *pos; - ldlm_processing_policy policy; - int flags; - int rc = LDLM_ITER_CONTINUE; - ldlm_error_t err; - ENTRY; + struct list_head *tmp, *pos; + ldlm_processing_policy policy; + __u64 flags; + int rc = LDLM_ITER_CONTINUE; + enum ldlm_error err; + ENTRY; - check_res_locked(res); + check_res_locked(res); - policy = ldlm_processing_policy_table[res->lr_type]; - LASSERT(policy); + policy = ldlm_processing_policy_table[res->lr_type]; + LASSERT(policy); - cfs_list_for_each_safe(tmp, pos, queue) { - struct ldlm_lock *pending; - pending = cfs_list_entry(tmp, struct ldlm_lock, l_res_link); + list_for_each_safe(tmp, pos, queue) { + struct ldlm_lock *pending; + + pending = list_entry(tmp, struct ldlm_lock, l_res_link); CDEBUG(D_INFO, "Reprocessing lock %p\n", pending); @@ -1370,170 +1877,215 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, cfs_list_t *queue, RETURN(rc); } +#endif -/* Helper function for ldlm_run_ast_work(). - * - * Send an existing rpc set specified by @arg->set and then - * destroy it. Create new one if @do_create flag is set. */ -static void -ldlm_send_and_maybe_create_set(struct ldlm_cb_set_arg *arg, int do_create) +/** + * 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) { - ENTRY; + struct ldlm_cb_set_arg *arg = opaq; + struct ldlm_lock_desc d; + int rc; + struct ldlm_lock *lock; + ENTRY; - ptlrpc_set_wait(arg->set); - if (arg->type == LDLM_BL_CALLBACK) - OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_GLIMPSE, 2); - ptlrpc_set_destroy(arg->set); + if (list_empty(arg->list)) + RETURN(-ENOENT); - if (do_create) - arg->set = ptlrpc_prep_set(); - - EXIT; -} - -static int -ldlm_work_bl_ast_lock(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg) -{ - struct ldlm_lock_desc d; - struct ldlm_lock *lock = cfs_list_entry(tmp, struct ldlm_lock, - l_bl_ast); - ENTRY; + lock = list_entry(arg->list->next, struct ldlm_lock, l_bl_ast); - /* nobody should touch l_bl_ast */ - lock_res_and_lock(lock); - cfs_list_del_init(&lock->l_bl_ast); + /* nobody should touch l_bl_ast */ + lock_res_and_lock(lock); + list_del_init(&lock->l_bl_ast); - 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); + LASSERT(ldlm_is_ast_sent(lock)); + LASSERT(lock->l_bl_ast_run == 0); + LASSERT(lock->l_blocking_lock); + lock->l_bl_ast_run++; + unlock_res_and_lock(lock); - ldlm_lock2desc(lock->l_blocking_lock, &d); + ldlm_lock2desc(lock->l_blocking_lock, &d); - 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); + 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); - RETURN(1); + RETURN(rc); } +/** + * Process a call to completion AST callback for a lock in ast_work list + */ static int -ldlm_work_cp_ast_lock(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg) +ldlm_work_cp_ast_lock(struct ptlrpc_request_set *rqset, void *opaq) { - 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. */ - - /* 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) { - completion_callback(lock, 0, (void *)arg); - rc = 1; - } - LDLM_LOCK_RELEASE(lock); - - RETURN(rc); + struct ldlm_cb_set_arg *arg = opaq; + int rc = 0; + struct ldlm_lock *lock; + ldlm_completion_callback completion_callback; + ENTRY; + + if (list_empty(arg->list)) + RETURN(-ENOENT); + + lock = 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); + list_del_init(&lock->l_cp_ast); + 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; + ldlm_clear_cp_reqd(lock); + unlock_res_and_lock(lock); + + if (completion_callback != NULL) + rc = completion_callback(lock, 0, (void *)arg); + LDLM_LOCK_RELEASE(lock); + + RETURN(rc); } +/** + * Process a call to revocation AST callback for a lock in ast_work list + */ 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); - ENTRY; + struct ldlm_cb_set_arg *arg = opaq; + struct ldlm_lock_desc desc; + int rc; + struct ldlm_lock *lock; + ENTRY; + + if (list_empty(arg->list)) + RETURN(-ENOENT); - cfs_list_del_init(&lock->l_rk_ast); + lock = list_entry(arg->list->next, struct ldlm_lock, l_rk_ast); + list_del_init(&lock->l_rk_ast); - /* the desc just pretend to exclusive */ - ldlm_lock2desc(lock, &desc); - desc.l_req_mode = LCK_EX; - desc.l_granted_mode = 0; + /* the desc just pretend to exclusive */ + ldlm_lock2desc(lock, &desc); + desc.l_req_mode = LCK_EX; + desc.l_granted_mode = 0; - lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING); - LDLM_LOCK_RELEASE(lock); + rc = lock->l_blocking_ast(lock, &desc, (void*)arg, LDLM_CB_BLOCKING); + LDLM_LOCK_RELEASE(lock); - RETURN(1); + RETURN(rc); } -int ldlm_run_ast_work(cfs_list_t *rpc_list, ldlm_desc_ast_t ast_type) +/** + * 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; - cfs_list_t *tmp, *pos; - int (*work_ast_lock)(cfs_list_t *tmp, struct ldlm_cb_set_arg *arg); - int ast_count; - ENTRY; + struct ldlm_cb_set_arg *arg = opaq; + struct ldlm_glimpse_work *gl_work; + struct ldlm_lock *lock; + int rc = 0; + ENTRY; - if (cfs_list_empty(rpc_list)) - RETURN(0); + if (list_empty(arg->list)) + RETURN(-ENOENT); - arg.set = ptlrpc_prep_set(); - if (NULL == arg.set) - RETURN(-ERESTART); - cfs_atomic_set(&arg.restart, 0); - 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(); - } + gl_work = list_entry(arg->list->next, struct ldlm_glimpse_work, + gl_list); + list_del_init(&gl_work->gl_list); - ast_count = 0; - cfs_list_for_each_safe(tmp, pos, rpc_list) { - ast_count += work_ast_lock(tmp, &arg); + lock = gl_work->gl_lock; - /* Send the request set if it exceeds the PARALLEL_AST_LIMIT, - * and create a new set for requests that remained in - * @rpc_list */ - if (unlikely(ast_count == PARALLEL_AST_LIMIT)) { - ldlm_send_and_maybe_create_set(&arg, 1); - ast_count = 0; - } - } + /* transfer the glimpse descriptor to ldlm_cb_set_arg */ + arg->gl_desc = gl_work->gl_desc; - if (ast_count > 0) - ldlm_send_and_maybe_create_set(&arg, 0); - else - /* In case when number of ASTs is multiply of - * PARALLEL_AST_LIMIT or @rpc_list was initially empty, - * @arg.set must be destroyed here, otherwise we get - * write memory leaking. */ - ptlrpc_set_destroy(arg.set); + /* invoke the actual glimpse callback */ + if (lock->l_glimpse_ast(lock, (void*)arg) == 0) + rc = 1; - RETURN(cfs_atomic_read(&arg.restart) ? -ERESTART : 0); + LDLM_LOCK_RELEASE(lock); + + if ((gl_work->gl_flags & LDLM_GL_WORK_NOFREE) == 0) + OBD_FREE_PTR(gl_work); + + 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, struct list_head *rpc_list, + ldlm_desc_ast_t ast_type) +{ + struct ldlm_cb_set_arg *arg; + set_producer_func work_ast_lock; + int rc; + + if (list_empty(rpc_list)) + RETURN(0); + + OBD_ALLOC_PTR(arg); + if (arg == NULL) + RETURN(-ENOMEM); + + 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 = 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) @@ -1542,8 +2094,8 @@ static int reprocess_one_queue(struct ldlm_resource *res, void *closure) return LDLM_ITER_CONTINUE; } -static int ldlm_reprocess_res(cfs_hash_t *hs, cfs_hash_bd_t *bd, - cfs_hlist_node_t *hnode, void *arg) +static int ldlm_reprocess_res(struct cfs_hash *hs, struct cfs_hash_bd *bd, + struct hlist_node *hnode, void *arg) { struct ldlm_resource *res = cfs_hash_object(hs, hnode); int rc; @@ -1553,51 +2105,80 @@ 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; + ENTRY; - if (ns != NULL) { - cfs_hash_for_each_nolock(ns->ns_rs_hash, - ldlm_reprocess_res, NULL); - } - EXIT; + if (ns != NULL) { + cfs_hash_for_each_nolock(ns->ns_rs_hash, + ldlm_reprocess_res, NULL, 0); + } + EXIT; } +/** + * 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); + struct list_head rpc_list; +#ifdef HAVE_SERVER_SUPPORT int rc; ENTRY; + INIT_LIST_HEAD(&rpc_list); /* 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) ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list); unlock_res(res); - rc = ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST); + rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list, + LDLM_WORK_CP_AST); if (rc == -ERESTART) { - LASSERT(cfs_list_empty(&rpc_list)); + LASSERT(list_empty(&rpc_list)); goto restart; } +#else + ENTRY; + + INIT_LIST_HEAD(&rpc_list); + 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; } +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); @@ -1606,19 +2187,25 @@ 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 && req->l_resource->lr_type != LDLM_IBITS) return; - cfs_list_del_init(&req->l_sl_policy); - cfs_list_del_init(&req->l_sl_mode); + list_del_init(&req->l_sl_policy); + 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; @@ -1637,14 +2224,14 @@ void ldlm_lock_cancel(struct ldlm_lock *lock) LBUG(); } - ldlm_del_waiting_lock(lock); + 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 */ - ldlm_del_waiting_lock(lock); + LASSERT(!ldlm_is_waited(lock)); + ldlm_resource_unlink_lock(lock); ldlm_lock_destroy_nolock(lock); @@ -1658,44 +2245,134 @@ void ldlm_lock_cancel(struct ldlm_lock *lock) EXIT; } +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); + int rc = -EINVAL; ENTRY; - if (lock == NULL) - RETURN(-EINVAL); + if (lock) { + if (lock->l_ast_data == NULL) + lock->l_ast_data = data; + if (lock->l_ast_data == data) + rc = 0; + LDLM_LOCK_PUT(lock); + } + RETURN(rc); +} +EXPORT_SYMBOL(ldlm_lock_set_data); - lock->l_ast_data = data; - LDLM_LOCK_PUT(lock); - RETURN(0); +struct export_cl_data { + struct obd_export *ecl_exp; + int ecl_loop; +}; + +static void ldlm_cancel_lock_for_export(struct obd_export *exp, + struct ldlm_lock *lock, + struct export_cl_data *ecl) +{ + struct ldlm_resource *res; + + res = ldlm_resource_getref(lock->l_resource); + + ldlm_res_lvbo_update(res, NULL, 1); + ldlm_lock_cancel(lock); + if (!exp->exp_obd->obd_stopping) + ldlm_reprocess_all(res); + ldlm_resource_putref(res); + + ecl->ecl_loop++; + if ((ecl->ecl_loop & -ecl->ecl_loop) == ecl->ecl_loop) { + CDEBUG(D_INFO, "Export %p, %d locks cancelled.\n", + exp, ecl->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) +/** + * Iterator function for ldlm_export_cancel_locks. + * Cancels passed locks. + */ +static int +ldlm_cancel_locks_for_export_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd, + struct hlist_node *hnode, void *data) { - struct obd_export *exp = data; - struct ldlm_lock *lock = cfs_hash_object(hs, hnode); - struct ldlm_resource *res; + 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); - res = ldlm_resource_getref(lock->l_resource); - LDLM_LOCK_GET(lock); + LDLM_LOCK_GET(lock); + ldlm_cancel_lock_for_export(exp, lock, ecl); + LDLM_LOCK_RELEASE(lock); - LDLM_DEBUG(lock, "export %p", exp); - ldlm_res_lvbo_update(res, NULL, 1); - ldlm_lock_cancel(lock); - ldlm_reprocess_all(res); - ldlm_resource_putref(res); - LDLM_LOCK_RELEASE(lock); - return 0; + return 0; +} + +/** + * Cancel all blocked locks for given export. + * + * Typically called on client disconnection/eviction + */ +int ldlm_export_cancel_blocked_locks(struct obd_export *exp) +{ + struct export_cl_data ecl = { + .ecl_exp = exp, + .ecl_loop = 0, + }; + + while (!list_empty(&exp->exp_bl_list)) { + struct ldlm_lock *lock; + + spin_lock_bh(&exp->exp_bl_list_lock); + if (!list_empty(&exp->exp_bl_list)) { + lock = list_entry(exp->exp_bl_list.next, + struct ldlm_lock, l_exp_list); + LDLM_LOCK_GET(lock); + list_del_init(&lock->l_exp_list); + } else { + lock = NULL; + } + spin_unlock_bh(&exp->exp_bl_list_lock); + + if (lock == NULL) + break; + + ldlm_cancel_lock_for_export(exp, lock, &ecl); + LDLM_LOCK_RELEASE(lock); + } + + CDEBUG(D_DLMTRACE, "Export %p, canceled %d locks, " + "left on hash table %d.\n", exp, ecl.ecl_loop, + atomic_read(&exp->exp_lock_hash->hs_count)); + + return ecl.ecl_loop; } -void ldlm_cancel_locks_for_export(struct obd_export *exp) +/** + * Cancel all locks for given export. + * + * Typically called after client disconnection/eviction + */ +int ldlm_export_cancel_locks(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); + + CDEBUG(D_DLMTRACE, "Export %p, canceled %d locks, " + "left on hash table %d.\n", exp, ecl.ecl_loop, + atomic_read(&exp->exp_lock_hash->hs_count)); + + return ecl.ecl_loop; } /** @@ -1703,11 +2380,12 @@ 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 */ -void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode) +void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode) { ENTRY; @@ -1729,48 +2407,63 @@ void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode) EXIT; } +EXPORT_SYMBOL(ldlm_lock_downgrade); -struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, - __u32 *flags) +/** + * 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, + enum ldlm_mode new_mode, __u32 *flags) { - CFS_LIST_HEAD(rpc_list); - struct ldlm_resource *res; - struct ldlm_namespace *ns; - int granted = 0; - int old_mode, rc; - struct sl_insert_point prev; - ldlm_error_t err; - struct ldlm_interval *node; - ENTRY; + struct list_head rpc_list; + struct ldlm_resource *res; + struct ldlm_namespace *ns; + int granted = 0; +#ifdef HAVE_SERVER_SUPPORT + int old_mode; + struct sl_insert_point prev; +#endif + struct ldlm_interval *node; + ENTRY; - if (new_mode == lock->l_granted_mode) { // No changes? Just return. - *flags |= LDLM_FL_BLOCK_GRANTED; - RETURN(lock->l_resource); - } + INIT_LIST_HEAD(&rpc_list); + /* 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 */ - RETURN(NULL); + /* 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, GFP_NOFS); + if (node == NULL) /* Actually, this causes EDEADLOCK to be returned */ + RETURN(NULL); - LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR), - "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode); + LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR), + "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode); - lock_res_and_lock(lock); + lock_res_and_lock(lock); - res = lock->l_resource; - ns = ldlm_res_to_ns(res); + 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); @@ -1778,7 +2471,7 @@ struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode, /* FIXME: ugly code, I have to attach the lock to a * interval node again since perhaps it will be granted * soon */ - CFS_INIT_LIST_HEAD(&node->li_group); + INIT_LIST_HEAD(&node->li_group); ldlm_interval_attach(node, lock); node = NULL; } @@ -1797,7 +2490,7 @@ 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(); @@ -1807,9 +2500,13 @@ 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); } - } else { - int pflags = 0; - ldlm_processing_policy policy; +#ifdef HAVE_SERVER_SUPPORT + } else { + int rc; + enum ldlm_error err; + __u64 pflags = 0; + ldlm_processing_policy policy; + policy = ldlm_processing_policy_table[res->lr_type]; rc = policy(lock, &pflags, 0, &err, &rpc_list); if (rc == LDLM_ITER_STOP) { @@ -1825,70 +2522,27 @@ 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) - ldlm_run_ast_work(&rpc_list, LDLM_WORK_CP_AST); + ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST); if (node) OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node)); 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); -} - +/** + * 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; @@ -1900,135 +2554,141 @@ 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); } +EXPORT_SYMBOL(ldlm_lock_dump_handle); -void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 level, - struct libcfs_debug_msg_data *data, const char *fmt, - ...) +/** + * 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, ...) { 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, - lock->l_handle.h_cookie, cfs_atomic_read(&lock->l_refc), + "res: \?\? rrc=\?\? type: \?\?\? flags: "LPX64" nid: %s " + "remote: "LPX64" expref: %d pid: %u timeout: %lu " + "lvb_type: %d\n", + 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], - lock->l_flags, lock->l_remote_handle.cookie, - lock->l_export ? - cfs_atomic_read(&lock->l_export->exp_refcount) : -99, - lock->l_pid, lock->l_callback_timeout); + 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); va_end(args); return; } - switch (lock->l_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, - " 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", - 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_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_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, - " 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 - " 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_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_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, - " 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 " - "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], - 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, - 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, - " 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", - 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, - 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);