X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdclass%2Flu_object.c;h=8a86fa5ba37daa0b66b5e23a3c99e066d805ad39;hb=711e142d055fda62be482f74f2f73acac5e7e453;hp=c7747e8ce136a883c1af94e52f38b63f1f6e1898;hpb=068d1a3fbd94d4f8cab3d897b6b6ef7c385b4fc1;p=fs%2Flustre-release.git diff --git a/lustre/obdclass/lu_object.c b/lustre/obdclass/lu_object.c index c7747e8..8a86fa5 100644 --- a/lustre/obdclass/lu_object.c +++ b/lustre/obdclass/lu_object.c @@ -1,31 +1,45 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Lustre Object. + * GPL HEADER START + * + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * Copyright (C) 2006 Cluster File Systems, Inc. - * Author: Nikita Danilov + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * This file is part of the Lustre file system, http://www.lustre.org - * Lustre is a trademark of Cluster File Systems, Inc. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * You may have signed or agreed to another license before downloading - * this software. If so, you are bound by the terms and conditions - * of that agreement, and the following does not apply to you. See the - * LICENSE file included with this distribution for more information. + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * If you did not agree to a different license, then this copy of Lustre - * is open source software; you can redistribute it and/or modify it - * under the terms of version 2 of the GNU General Public License as - * published by the Free Software Foundation. + * GPL HEADER END + */ +/* + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. * - * In either case, Lustre is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * license text for more details. + * lustre/obdclass/lu_object.c * + * Lustre Object. * These are the only exported functions, they provide some generic * infrastructure for managing object devices + * + * Author: Nikita Danilov */ #define DEBUG_SUBSYSTEM S_CLASS @@ -33,12 +47,15 @@ # define EXPORT_SYMTAB #endif -#include -#include -/* nr_free_pages() */ -#include +#include + +#ifdef __KERNEL__ +# include +#endif + /* hash_long() */ -#include +#include +#include #include #include #include @@ -49,7 +66,7 @@ static void lu_object_free(const struct lu_env *env, struct lu_object *o); -/* +/** * Decrease reference counter on object. If last reference is freed, return * object to the cache, unless lu_object_is_dying(o) holds. In the latter * case, free object immediately. @@ -65,13 +82,14 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o) site = o->lo_dev->ld_site; orig = o; kill_it = 0; - write_lock(&site->ls_guard); - if (atomic_dec_and_test(&top->loh_ref)) { + cfs_write_lock(&site->ls_guard); + if (cfs_atomic_dec_and_test(&top->loh_ref)) { /* * When last reference is released, iterate over object * layers, and notify them that object is no longer busy. */ - list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) { + cfs_list_for_each_entry_reverse(o, &top->loh_layers, + lo_linkage) { if (o->lo_ops->loo_object_release != NULL) o->lo_ops->loo_object_release(env, o); } @@ -90,13 +108,19 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o) * object lookup is possible and we can safely destroy * object below. */ - hlist_del_init(&top->loh_hash); - list_del_init(&top->loh_lru); + cfs_hlist_del_init(&top->loh_hash); + cfs_list_del_init(&top->loh_lru); -- site->ls_total; kill_it = 1; } + } else if (lu_object_is_dying(top)) { + /* + * somebody may be waiting for this, currently only used + * for cl_object, see cl_object_put_last(). + */ + cfs_waitq_broadcast(&site->ls_marche_funebre); } - write_unlock(&site->ls_guard); + cfs_write_unlock(&site->ls_guard); if (kill_it) /* * Object was already removed from hash and lru above, can @@ -106,30 +130,31 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o) } EXPORT_SYMBOL(lu_object_put); -/* +/** * Allocate new object. * * This follows object creation protocol, described in the comment within * struct lu_device_operations definition. */ static struct lu_object *lu_object_alloc(const struct lu_env *env, - struct lu_site *s, - const struct lu_fid *f) + struct lu_device *dev, + const struct lu_fid *f, + const struct lu_object_conf *conf) { struct lu_object *scan; struct lu_object *top; - struct list_head *layers; + cfs_list_t *layers; int clean; int result; + ENTRY; /* * Create top-level object slice. This will also create * lu_object_header. */ - top = s->ls_top_dev->ld_ops->ldo_object_alloc(env, - NULL, s->ls_top_dev); - if (IS_ERR(top)) - RETURN(top); + top = dev->ld_ops->ldo_object_alloc(env, NULL, dev); + if (top == NULL) + RETURN(ERR_PTR(-ENOMEM)); /* * This is the only place where object fid is assigned. It's constant * after this point. @@ -143,12 +168,12 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, * object slices are created. */ clean = 1; - list_for_each_entry(scan, layers, lo_linkage) { + cfs_list_for_each_entry(scan, layers, lo_linkage) { if (scan->lo_flags & LU_OBJECT_ALLOCATED) continue; clean = 0; scan->lo_header = top->lo_header; - result = scan->lo_ops->loo_object_init(env, scan); + result = scan->lo_ops->loo_object_init(env, scan, conf); if (result != 0) { lu_object_free(env, top); RETURN(ERR_PTR(result)); @@ -157,7 +182,7 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, } } while (!clean); - list_for_each_entry_reverse(scan, layers, lo_linkage) { + cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) { if (scan->lo_ops->loo_object_start != NULL) { result = scan->lo_ops->loo_object_start(env, scan); if (result != 0) { @@ -167,23 +192,26 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, } } - s->ls_stats.s_created ++; + dev->ld_site->ls_stats.s_created ++; RETURN(top); } -/* - * Free object. +/** + * Free an object. */ static void lu_object_free(const struct lu_env *env, struct lu_object *o) { - struct list_head splice; - struct lu_object *scan; + cfs_list_t splice; + struct lu_object *scan; + struct lu_site *site; + cfs_list_t *layers; + site = o->lo_dev->ld_site; + layers = &o->lo_header->loh_layers; /* * First call ->loo_object_delete() method to release all resources. */ - list_for_each_entry_reverse(scan, - &o->lo_header->loh_layers, lo_linkage) { + cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) { if (scan->lo_ops->loo_object_delete != NULL) scan->lo_ops->loo_object_delete(env, scan); } @@ -194,32 +222,38 @@ static void lu_object_free(const struct lu_env *env, struct lu_object *o) * necessary, because lu_object_header is freed together with the * top-level slice. */ - INIT_LIST_HEAD(&splice); - list_splice_init(&o->lo_header->loh_layers, &splice); - while (!list_empty(&splice)) { - o = container_of0(splice.next, struct lu_object, lo_linkage); - list_del_init(&o->lo_linkage); + CFS_INIT_LIST_HEAD(&splice); + cfs_list_splice_init(layers, &splice); + while (!cfs_list_empty(&splice)) { + /* + * Free layers in bottom-to-top order, so that object header + * lives as long as possible and ->loo_object_free() methods + * can look at its contents. + */ + o = container_of0(splice.prev, struct lu_object, lo_linkage); + cfs_list_del_init(&o->lo_linkage); LASSERT(o->lo_ops->loo_object_free != NULL); o->lo_ops->loo_object_free(env, o); } + cfs_waitq_broadcast(&site->ls_marche_funebre); } -/* - * Free @nr objects from the cold end of the site LRU list. +/** + * Free \a nr objects from the cold end of the site LRU list. */ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr) { - struct list_head dispose; + cfs_list_t dispose; struct lu_object_header *h; struct lu_object_header *temp; - INIT_LIST_HEAD(&dispose); + CFS_INIT_LIST_HEAD(&dispose); /* * Under LRU list lock, scan LRU list and move unreferenced objects to * the dispose list, removing them from LRU and hash table. */ - write_lock(&s->ls_guard); - list_for_each_entry_safe(h, temp, &s->ls_lru, loh_lru) { + cfs_write_lock(&s->ls_guard); + cfs_list_for_each_entry_safe(h, temp, &s->ls_lru, loh_lru) { /* * Objects are sorted in lru order, and "busy" objects (ones * with h->loh_ref > 0) naturally tend to live near hot end @@ -232,21 +266,21 @@ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr) */ if (nr-- == 0) break; - if (atomic_read(&h->loh_ref) > 0) + if (cfs_atomic_read(&h->loh_ref) > 0) continue; - hlist_del_init(&h->loh_hash); - list_move(&h->loh_lru, &dispose); + cfs_hlist_del_init(&h->loh_hash); + cfs_list_move(&h->loh_lru, &dispose); s->ls_total --; } - write_unlock(&s->ls_guard); + cfs_write_unlock(&s->ls_guard); /* * Free everything on the dispose list. This is safe against races due * to the reasons described in lu_object_put(). */ - while (!list_empty(&dispose)) { + while (!cfs_list_empty(&dispose)) { h = container_of0(dispose.next, struct lu_object_header, loh_lru); - list_del_init(&h->loh_lru); + cfs_list_del_init(&h->loh_lru); lu_object_free(env, lu_object_top(h)); s->ls_stats.s_lru_purged ++; } @@ -271,7 +305,7 @@ EXPORT_SYMBOL(lu_site_purge); */ enum { - /* + /** * Maximal line size. * * XXX overflow is not handled correctly. @@ -280,20 +314,16 @@ enum { }; struct lu_cdebug_data { - /* + /** * Temporary buffer. */ char lck_area[LU_CDEBUG_LINE]; - /* - * fid staging area used by dt_store_open(). - */ - struct lu_fid_pack lck_pack; }; /* context key constructor/destructor: lu_global_key_init, lu_global_key_fini */ LU_KEY_INIT_FINI(lu_global, struct lu_cdebug_data); -/* +/** * Key, holding temporary buffer. This key is registered very early by * lu_global_init(). */ @@ -303,7 +333,7 @@ struct lu_context_key lu_global_key = { .lct_fini = lu_global_key_fini }; -/* +/** * Printer function emitting messages through libcfs_debug_msg(). */ int lu_cdebug_printer(const struct lu_env *env, @@ -313,7 +343,7 @@ int lu_cdebug_printer(const struct lu_env *env, struct lu_cdebug_data *key; int used; int complete; - va_list args; + va_list args; va_start(args, format); @@ -328,9 +358,10 @@ int lu_cdebug_printer(const struct lu_env *env, vsnprintf(key->lck_area + used, ARRAY_SIZE(key->lck_area) - used, format, args); if (complete) { - libcfs_debug_msg(NULL, info->lpi_subsys, info->lpi_mask, - (char *)info->lpi_file, info->lpi_fn, - info->lpi_line, "%s", key->lck_area); + if (cfs_cdebug_show(info->lpi_mask, info->lpi_subsys)) + libcfs_debug_msg(NULL, info->lpi_subsys, info->lpi_mask, + (char *)info->lpi_file, info->lpi_fn, + info->lpi_line, "%s", key->lck_area); key->lck_area[0] = 0; } va_end(args); @@ -338,23 +369,25 @@ int lu_cdebug_printer(const struct lu_env *env, } EXPORT_SYMBOL(lu_cdebug_printer); -/* +/** * Print object header. */ -static void lu_object_header_print(const struct lu_env *env, - void *cookie, lu_printer_t printer, - const struct lu_object_header *hdr) +void lu_object_header_print(const struct lu_env *env, void *cookie, + lu_printer_t printer, + const struct lu_object_header *hdr) { (*printer)(env, cookie, "header@%p[%#lx, %d, "DFID"%s%s%s]", - hdr, hdr->loh_flags, atomic_read(&hdr->loh_ref), + hdr, hdr->loh_flags, cfs_atomic_read(&hdr->loh_ref), PFID(&hdr->loh_fid), - hlist_unhashed(&hdr->loh_hash) ? "" : " hash", - list_empty(&hdr->loh_lru) ? "" : " lru", + cfs_hlist_unhashed(&hdr->loh_hash) ? "" : " hash", + cfs_list_empty((cfs_list_t *)&hdr->loh_lru) ? \ + "" : " lru", hdr->loh_attr & LOHA_EXISTS ? " exist":""); } +EXPORT_SYMBOL(lu_object_header_print); -/* - * Print human readable representation of the @o to the @printer. +/** + * Print human readable representation of the \a o to the \a printer. */ void lu_object_print(const struct lu_env *env, void *cookie, lu_printer_t printer, const struct lu_object *o) @@ -365,21 +398,24 @@ void lu_object_print(const struct lu_env *env, void *cookie, top = o->lo_header; lu_object_header_print(env, cookie, printer, top); - (*printer)(env, cookie, "\n"); - list_for_each_entry(o, &top->loh_layers, lo_linkage) { + (*printer)(env, cookie, "{ \n"); + cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) { depth = o->lo_depth + 4; - LASSERT(o->lo_ops->loo_object_print != NULL); + /* - * print `.' @depth times. + * print `.' \a depth times followed by type name and address */ - (*printer)(env, cookie, "%*.*s", depth, depth, ruler); - o->lo_ops->loo_object_print(env, cookie, printer, o); + (*printer)(env, cookie, "%*.*s%s@%p", depth, depth, ruler, + o->lo_dev->ld_type->ldt_name, o); + if (o->lo_ops->loo_object_print != NULL) + o->lo_ops->loo_object_print(env, cookie, printer, o); (*printer)(env, cookie, "\n"); } + (*printer)(env, cookie, "} header@%p\n", top); } EXPORT_SYMBOL(lu_object_print); -/* +/** * Check object consistency. */ int lu_object_invariant(const struct lu_object *o) @@ -387,7 +423,7 @@ int lu_object_invariant(const struct lu_object *o) struct lu_object_header *top; top = o->lo_header; - list_for_each_entry(o, &top->loh_layers, lo_linkage) { + cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) { if (o->lo_ops->loo_object_invariant != NULL && !o->lo_ops->loo_object_invariant(o)) return 0; @@ -397,18 +433,32 @@ int lu_object_invariant(const struct lu_object *o) EXPORT_SYMBOL(lu_object_invariant); static struct lu_object *htable_lookup(struct lu_site *s, - const struct hlist_head *bucket, - const struct lu_fid *f) + const cfs_hlist_head_t *bucket, + const struct lu_fid *f, + cfs_waitlink_t *waiter) { struct lu_object_header *h; - struct hlist_node *scan; + cfs_hlist_node_t *scan; - hlist_for_each_entry(h, scan, bucket, loh_hash) { + cfs_hlist_for_each_entry(h, scan, bucket, loh_hash) { s->ls_stats.s_cache_check ++; - if (likely(lu_fid_eq(&h->loh_fid, f) && - !lu_object_is_dying(h))) { + if (likely(lu_fid_eq(&h->loh_fid, f))) { + if (unlikely(lu_object_is_dying(h))) { + /* + * Lookup found an object being destroyed; + * this object cannot be returned (to assure + * that references to dying objects are + * eventually drained), and moreover, lookup + * has to wait until object is freed. + */ + cfs_waitlink_init(waiter); + cfs_waitq_add(&s->ls_marche_funebre, waiter); + cfs_set_current_state(CFS_TASK_UNINT); + s->ls_stats.s_cache_death_race ++; + return ERR_PTR(-EAGAIN); + } /* bump reference count... */ - if (atomic_add_return(1, &h->loh_ref) == 1) + if (cfs_atomic_add_return(1, &h->loh_ref) == 1) ++ s->ls_busy; /* and move to the head of the LRU */ /* @@ -428,20 +478,35 @@ static __u32 fid_hash(const struct lu_fid *f, int bits) { /* all objects with same id and different versions will belong to same * collisions list. */ - return hash_long(fid_flatten(f), bits); + return cfs_hash_long(fid_flatten(f), bits); } -/* - * Search cache for an object with the fid @f. If such object is found, return - * it. Otherwise, create new object, insert it into cache and return it. In - * any case, additional reference is acquired on the returned object. +/** + * Search cache for an object with the fid \a f. If such object is found, + * return it. Otherwise, create new object, insert it into cache and return + * it. In any case, additional reference is acquired on the returned object. */ struct lu_object *lu_object_find(const struct lu_env *env, - struct lu_site *s, const struct lu_fid *f) + struct lu_device *dev, const struct lu_fid *f, + const struct lu_object_conf *conf) +{ + return lu_object_find_at(env, dev->ld_site->ls_top_dev, f, conf); +} +EXPORT_SYMBOL(lu_object_find); + +/** + * Core logic of lu_object_find*() functions. + */ +static struct lu_object *lu_object_find_try(const struct lu_env *env, + struct lu_device *dev, + const struct lu_fid *f, + const struct lu_object_conf *conf, + cfs_waitlink_t *waiter) { - struct lu_object *o; - struct lu_object *shadow; - struct hlist_head *bucket; + struct lu_site *s; + struct lu_object *o; + struct lu_object *shadow; + cfs_hlist_head_t *bucket; /* * This uses standard index maintenance protocol: @@ -455,13 +520,17 @@ struct lu_object *lu_object_find(const struct lu_env *env, * object just allocated. * - unlock index; * - return object. + * + * If dying object is found during index search, add @waiter to the + * site wait-queue and return ERR_PTR(-EAGAIN). */ + s = dev->ld_site; bucket = s->ls_hash + fid_hash(f, s->ls_hash_bits); - read_lock(&s->ls_guard); - o = htable_lookup(s, bucket, f); - read_unlock(&s->ls_guard); + cfs_read_lock(&s->ls_guard); + o = htable_lookup(s, bucket, f, waiter); + cfs_read_unlock(&s->ls_guard); if (o != NULL) return o; @@ -470,43 +539,128 @@ struct lu_object *lu_object_find(const struct lu_env *env, * Allocate new object. This may result in rather complicated * operations, including fld queries, inode loading, etc. */ - o = lu_object_alloc(env, s, f); + o = lu_object_alloc(env, dev, f, conf); if (unlikely(IS_ERR(o))) return o; LASSERT(lu_fid_eq(lu_object_fid(o), f)); - write_lock(&s->ls_guard); - shadow = htable_lookup(s, bucket, f); + cfs_write_lock(&s->ls_guard); + shadow = htable_lookup(s, bucket, f, waiter); if (likely(shadow == NULL)) { - hlist_add_head(&o->lo_header->loh_hash, bucket); - list_add_tail(&o->lo_header->loh_lru, &s->ls_lru); + cfs_hlist_add_head(&o->lo_header->loh_hash, bucket); + cfs_list_add_tail(&o->lo_header->loh_lru, &s->ls_lru); ++ s->ls_busy; ++ s->ls_total; shadow = o; o = NULL; } else s->ls_stats.s_cache_race ++; - write_unlock(&s->ls_guard); + cfs_write_unlock(&s->ls_guard); if (o != NULL) lu_object_free(env, o); return shadow; } -EXPORT_SYMBOL(lu_object_find); -/* +/** + * Much like lu_object_find(), but top level device of object is specifically + * \a dev rather than top level device of the site. This interface allows + * objects of different "stacking" to be created within the same site. + */ +struct lu_object *lu_object_find_at(const struct lu_env *env, + struct lu_device *dev, + const struct lu_fid *f, + const struct lu_object_conf *conf) +{ + struct lu_object *obj; + cfs_waitlink_t wait; + + while (1) { + obj = lu_object_find_try(env, dev, f, conf, &wait); + if (obj == ERR_PTR(-EAGAIN)) { + /* + * lu_object_find_try() already added waiter into the + * wait queue. + */ + cfs_waitq_wait(&wait, CFS_TASK_UNINT); + cfs_waitq_del(&dev->ld_site->ls_marche_funebre, &wait); + } else + break; + } + return obj; +} +EXPORT_SYMBOL(lu_object_find_at); + +/** + * Find object with given fid, and return its slice belonging to given device. + */ +struct lu_object *lu_object_find_slice(const struct lu_env *env, + struct lu_device *dev, + const struct lu_fid *f, + const struct lu_object_conf *conf) +{ + struct lu_object *top; + struct lu_object *obj; + + top = lu_object_find(env, dev, f, conf); + if (!IS_ERR(top)) { + obj = lu_object_locate(top->lo_header, dev->ld_type); + if (obj == NULL) + lu_object_put(env, top); + } else + obj = top; + return obj; +} +EXPORT_SYMBOL(lu_object_find_slice); + +/** + * Global list of all device types. + */ +static CFS_LIST_HEAD(lu_device_types); + +int lu_device_type_init(struct lu_device_type *ldt) +{ + int result; + + CFS_INIT_LIST_HEAD(&ldt->ldt_linkage); + result = ldt->ldt_ops->ldto_init(ldt); + if (result == 0) + cfs_list_add(&ldt->ldt_linkage, &lu_device_types); + return result; +} +EXPORT_SYMBOL(lu_device_type_init); + +void lu_device_type_fini(struct lu_device_type *ldt) +{ + cfs_list_del_init(&ldt->ldt_linkage); + ldt->ldt_ops->ldto_fini(ldt); +} +EXPORT_SYMBOL(lu_device_type_fini); + +void lu_types_stop(void) +{ + struct lu_device_type *ldt; + + cfs_list_for_each_entry(ldt, &lu_device_types, ldt_linkage) { + if (ldt->ldt_device_nr == 0) + ldt->ldt_ops->ldto_stop(ldt); + } +} +EXPORT_SYMBOL(lu_types_stop); + +/** * Global list of all sites on this node */ -static LIST_HEAD(lu_sites); -static DECLARE_MUTEX(lu_sites_guard); +static CFS_LIST_HEAD(lu_sites); +static CFS_DECLARE_MUTEX(lu_sites_guard); -/* +/** * Global environment used by site shrinker. */ static struct lu_env lu_shrink_env; -/* - * Print all objects in @s. +/** + * Print all objects in \a s. */ void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie, lu_printer_t printer) @@ -515,12 +669,12 @@ void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie, for (i = 0; i < s->ls_hash_size; ++i) { struct lu_object_header *h; - struct hlist_node *scan; + cfs_hlist_node_t *scan; - read_lock(&s->ls_guard); - hlist_for_each_entry(h, scan, &s->ls_hash[i], loh_hash) { + cfs_read_lock(&s->ls_guard); + cfs_hlist_for_each_entry(h, scan, &s->ls_hash[i], loh_hash) { - if (!list_empty(&h->loh_layers)) { + if (!cfs_list_empty(&h->loh_layers)) { const struct lu_object *obj; obj = lu_object_top(h); @@ -528,32 +682,40 @@ void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie, } else lu_object_header_print(env, cookie, printer, h); } - read_unlock(&s->ls_guard); + cfs_read_unlock(&s->ls_guard); } } EXPORT_SYMBOL(lu_site_print); enum { - LU_CACHE_PERCENT = 30, + LU_CACHE_PERCENT = 20, }; -/* +/** * Return desired hash table order. */ static int lu_htable_order(void) { - int bits; unsigned long cache_size; + int bits; /* * Calculate hash table size, assuming that we want reasonable - * performance when 30% of available memory is occupied by cache of + * performance when 20% of total memory is occupied by cache of * lu_objects. * * Size of lu_object is (arbitrary) taken as 1K (together with inode). */ - cache_size = nr_free_buffer_pages() / 100 * - LU_CACHE_PERCENT * (CFS_PAGE_SIZE / 1024); + cache_size = cfs_num_physpages; + +#if BITS_PER_LONG == 32 + /* limit hashtable size for lowmem systems to low RAM */ + if (cache_size > 1 << (30 - CFS_PAGE_SHIFT)) + cache_size = 1 << (30 - CFS_PAGE_SHIFT) * 3 / 4; +#endif + + cache_size = cache_size / 100 * LU_CACHE_PERCENT * + (CFS_PAGE_SIZE / 1024); for (bits = 1; (1 << bits) < cache_size; ++bits) { ; @@ -561,8 +723,10 @@ static int lu_htable_order(void) return bits; } -/* - * Initialize site @s, with @d as the top level device. +static cfs_lock_class_key_t lu_site_guard_class; + +/** + * Initialize site \a s, with \a d as the top level device. */ int lu_site_init(struct lu_site *s, struct lu_device *top) { @@ -572,12 +736,15 @@ int lu_site_init(struct lu_site *s, struct lu_device *top) ENTRY; memset(s, 0, sizeof *s); - rwlock_init(&s->ls_guard); + cfs_rwlock_init(&s->ls_guard); + cfs_lockdep_set_class(&s->ls_guard, &lu_site_guard_class); CFS_INIT_LIST_HEAD(&s->ls_lru); CFS_INIT_LIST_HEAD(&s->ls_linkage); + cfs_waitq_init(&s->ls_marche_funebre); s->ls_top_dev = top; top->ld_site = s; lu_device_get(top); + lu_ref_add(&top->ld_reference, "site-top", s); for (bits = lu_htable_order(), size = 1 << bits; (s->ls_hash = @@ -594,101 +761,114 @@ int lu_site_init(struct lu_site *s, struct lu_device *top) s->ls_hash_mask = size - 1; for (i = 0; i < size; i++) - INIT_HLIST_HEAD(&s->ls_hash[i]); + CFS_INIT_HLIST_HEAD(&s->ls_hash[i]); RETURN(0); } EXPORT_SYMBOL(lu_site_init); -/* - * Finalize @s and release its resources. +/** + * Finalize \a s and release its resources. */ void lu_site_fini(struct lu_site *s) { - LASSERT(list_empty(&s->ls_lru)); + LASSERT(cfs_list_empty(&s->ls_lru)); LASSERT(s->ls_total == 0); - down(&lu_sites_guard); - list_del_init(&s->ls_linkage); - up(&lu_sites_guard); + cfs_down(&lu_sites_guard); + cfs_list_del_init(&s->ls_linkage); + cfs_up(&lu_sites_guard); if (s->ls_hash != NULL) { int i; for (i = 0; i < s->ls_hash_size; i++) - LASSERT(hlist_empty(&s->ls_hash[i])); + LASSERT(cfs_hlist_empty(&s->ls_hash[i])); cfs_free_large(s->ls_hash); s->ls_hash = NULL; } if (s->ls_top_dev != NULL) { s->ls_top_dev->ld_site = NULL; + lu_ref_del(&s->ls_top_dev->ld_reference, "site-top", s); lu_device_put(s->ls_top_dev); s->ls_top_dev = NULL; } } EXPORT_SYMBOL(lu_site_fini); -/* +/** * Called when initialization of stack for this site is completed. */ int lu_site_init_finish(struct lu_site *s) { int result; - down(&lu_sites_guard); + cfs_down(&lu_sites_guard); result = lu_context_refill(&lu_shrink_env.le_ctx); if (result == 0) - list_add(&s->ls_linkage, &lu_sites); - up(&lu_sites_guard); + cfs_list_add(&s->ls_linkage, &lu_sites); + cfs_up(&lu_sites_guard); return result; } EXPORT_SYMBOL(lu_site_init_finish); -/* - * Acquire additional reference on device @d +/** + * Acquire additional reference on device \a d */ void lu_device_get(struct lu_device *d) { - atomic_inc(&d->ld_ref); + cfs_atomic_inc(&d->ld_ref); } EXPORT_SYMBOL(lu_device_get); -/* - * Release reference on device @d. +/** + * Release reference on device \a d. */ void lu_device_put(struct lu_device *d) { - atomic_dec(&d->ld_ref); + LASSERT(cfs_atomic_read(&d->ld_ref) > 0); + cfs_atomic_dec(&d->ld_ref); } EXPORT_SYMBOL(lu_device_put); -/* - * Initialize device @d of type @t. +/** + * Initialize device \a d of type \a t. */ int lu_device_init(struct lu_device *d, struct lu_device_type *t) { + if (t->ldt_device_nr++ == 0 && t->ldt_ops->ldto_start != NULL) + t->ldt_ops->ldto_start(t); memset(d, 0, sizeof *d); - atomic_set(&d->ld_ref, 0); + cfs_atomic_set(&d->ld_ref, 0); d->ld_type = t; + lu_ref_init(&d->ld_reference); return 0; } EXPORT_SYMBOL(lu_device_init); -/* - * Finalize device @d. +/** + * Finalize device \a d. */ void lu_device_fini(struct lu_device *d) { - if (d->ld_obd != NULL) - /* finish lprocfs */ - lprocfs_obd_cleanup(d->ld_obd); + struct lu_device_type *t; + + t = d->ld_type; + if (d->ld_obd != NULL) { + d->ld_obd->obd_lu_dev = NULL; + d->ld_obd = NULL; + } - LASSERTF(atomic_read(&d->ld_ref) == 0, - "Refcount is %u\n", atomic_read(&d->ld_ref)); + lu_ref_fini(&d->ld_reference); + LASSERTF(cfs_atomic_read(&d->ld_ref) == 0, + "Refcount is %u\n", cfs_atomic_read(&d->ld_ref)); + LASSERT(t->ldt_device_nr > 0); + if (--t->ldt_device_nr == 0 && t->ldt_ops->ldto_stop != NULL) + t->ldt_ops->ldto_stop(t); } EXPORT_SYMBOL(lu_device_fini); -/* - * Initialize object @o that is part of compound object @h and was created by - * device @d. +/** + * Initialize object \a o that is part of compound object \a h and was created + * by device \a d. */ int lu_object_init(struct lu_object *o, struct lu_object_header *h, struct lu_device *d) @@ -697,84 +877,91 @@ int lu_object_init(struct lu_object *o, o->lo_header = h; o->lo_dev = d; lu_device_get(d); + o->lo_dev_ref = lu_ref_add(&d->ld_reference, "lu_object", o); CFS_INIT_LIST_HEAD(&o->lo_linkage); return 0; } EXPORT_SYMBOL(lu_object_init); -/* +/** * Finalize object and release its resources. */ void lu_object_fini(struct lu_object *o) { - LASSERT(list_empty(&o->lo_linkage)); + struct lu_device *dev = o->lo_dev; + + LASSERT(cfs_list_empty(&o->lo_linkage)); - if (o->lo_dev != NULL) { - lu_device_put(o->lo_dev); + if (dev != NULL) { + lu_ref_del_at(&dev->ld_reference, + o->lo_dev_ref , "lu_object", o); + lu_device_put(dev); o->lo_dev = NULL; } } EXPORT_SYMBOL(lu_object_fini); -/* - * Add object @o as first layer of compound object @h +/** + * Add object \a o as first layer of compound object \a h * * This is typically called by the ->ldo_object_alloc() method of top-level * device. */ void lu_object_add_top(struct lu_object_header *h, struct lu_object *o) { - list_move(&o->lo_linkage, &h->loh_layers); + cfs_list_move(&o->lo_linkage, &h->loh_layers); } EXPORT_SYMBOL(lu_object_add_top); -/* - * Add object @o as a layer of compound object, going after @before.1 +/** + * Add object \a o as a layer of compound object, going after \a before. * - * This is typically called by the ->ldo_object_alloc() method of - * @before->lo_dev. + * This is typically called by the ->ldo_object_alloc() method of \a + * before->lo_dev. */ void lu_object_add(struct lu_object *before, struct lu_object *o) { - list_move(&o->lo_linkage, &before->lo_linkage); + cfs_list_move(&o->lo_linkage, &before->lo_linkage); } EXPORT_SYMBOL(lu_object_add); -/* +/** * Initialize compound object. */ int lu_object_header_init(struct lu_object_header *h) { memset(h, 0, sizeof *h); - atomic_set(&h->loh_ref, 1); - INIT_HLIST_NODE(&h->loh_hash); + cfs_atomic_set(&h->loh_ref, 1); + CFS_INIT_HLIST_NODE(&h->loh_hash); CFS_INIT_LIST_HEAD(&h->loh_lru); CFS_INIT_LIST_HEAD(&h->loh_layers); + lu_ref_init(&h->loh_reference); return 0; } EXPORT_SYMBOL(lu_object_header_init); -/* +/** * Finalize compound object. */ void lu_object_header_fini(struct lu_object_header *h) { - LASSERT(list_empty(&h->loh_layers)); - LASSERT(list_empty(&h->loh_lru)); - LASSERT(hlist_unhashed(&h->loh_hash)); + LASSERT(cfs_list_empty(&h->loh_layers)); + LASSERT(cfs_list_empty(&h->loh_lru)); + LASSERT(cfs_hlist_unhashed(&h->loh_hash)); + lu_ref_fini(&h->loh_reference); } EXPORT_SYMBOL(lu_object_header_fini); -/* +/** * Given a compound object, find its slice, corresponding to the device type - * @dtype. + * \a dtype. */ struct lu_object *lu_object_locate(struct lu_object_header *h, - struct lu_device_type *dtype) + const struct lu_device_type *dtype) { struct lu_object *o; - list_for_each_entry(o, &h->loh_layers, lo_linkage) { + cfs_list_for_each_entry(o, &h->loh_layers, lo_linkage) { if (o->lo_dev->ld_type == dtype) return o; } @@ -782,18 +969,74 @@ struct lu_object *lu_object_locate(struct lu_object_header *h, } EXPORT_SYMBOL(lu_object_locate); + + +/** + * Finalize and free devices in the device stack. + * + * Finalize device stack by purging object cache, and calling + * lu_device_type_operations::ldto_device_fini() and + * lu_device_type_operations::ldto_device_free() on all devices in the stack. + */ +void lu_stack_fini(const struct lu_env *env, struct lu_device *top) +{ + struct lu_site *site = top->ld_site; + struct lu_device *scan; + struct lu_device *next; + + lu_site_purge(env, site, ~0); + for (scan = top; scan != NULL; scan = next) { + next = scan->ld_type->ldt_ops->ldto_device_fini(env, scan); + lu_ref_del(&scan->ld_reference, "lu-stack", &lu_site_init); + lu_device_put(scan); + } + + /* purge again. */ + lu_site_purge(env, site, ~0); + + if (!cfs_list_empty(&site->ls_lru) || site->ls_total != 0) { + /* + * Uh-oh, objects still exist. + */ + static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR); + + lu_site_print(env, site, &cookie, lu_cdebug_printer); + } + + for (scan = top; scan != NULL; scan = next) { + const struct lu_device_type *ldt = scan->ld_type; + struct obd_type *type; + + next = ldt->ldt_ops->ldto_device_free(env, scan); + type = ldt->ldt_obd_type; + if (type != NULL) { + type->typ_refcnt--; + class_put_type(type); + } + } +} +EXPORT_SYMBOL(lu_stack_fini); + enum { - /* + /** * Maximal number of tld slots. */ - LU_CONTEXT_KEY_NR = 16 + LU_CONTEXT_KEY_NR = 32 }; static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, }; -static spinlock_t lu_keys_guard = SPIN_LOCK_UNLOCKED; +static cfs_spinlock_t lu_keys_guard = CFS_SPIN_LOCK_UNLOCKED; -/* +/** + * Global counter incremented whenever key is registered, unregistered, + * revived or quiesced. This is used to void unnecessary calls to + * lu_context_refill(). No locking is provided, as initialization and shutdown + * are supposed to be externally serialized. + */ +static unsigned key_set_version = 0; + +/** * Register new key. */ int lu_context_key_register(struct lu_context_key *key) @@ -807,76 +1050,210 @@ int lu_context_key_register(struct lu_context_key *key) LASSERT(key->lct_owner != NULL); result = -ENFILE; - spin_lock(&lu_keys_guard); + cfs_spin_lock(&lu_keys_guard); for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) { if (lu_keys[i] == NULL) { key->lct_index = i; - atomic_set(&key->lct_used, 1); + cfs_atomic_set(&key->lct_used, 1); lu_keys[i] = key; + lu_ref_init(&key->lct_reference); result = 0; + ++key_set_version; break; } } - spin_unlock(&lu_keys_guard); + cfs_spin_unlock(&lu_keys_guard); return result; } EXPORT_SYMBOL(lu_context_key_register); static void key_fini(struct lu_context *ctx, int index) { - if (ctx->lc_value[index] != NULL) { + if (ctx->lc_value != NULL && ctx->lc_value[index] != NULL) { struct lu_context_key *key; key = lu_keys[index]; LASSERT(key != NULL); LASSERT(key->lct_fini != NULL); - LASSERT(atomic_read(&key->lct_used) > 1); + LASSERT(cfs_atomic_read(&key->lct_used) > 1); key->lct_fini(ctx, key, ctx->lc_value[index]); - atomic_dec(&key->lct_used); + lu_ref_del(&key->lct_reference, "ctx", ctx); + cfs_atomic_dec(&key->lct_used); LASSERT(key->lct_owner != NULL); if (!(ctx->lc_tags & LCT_NOREF)) { - LASSERT(module_refcount(key->lct_owner) > 0); - module_put(key->lct_owner); + LASSERT(cfs_module_refcount(key->lct_owner) > 0); + cfs_module_put(key->lct_owner); } ctx->lc_value[index] = NULL; } } -/* +/** * Deregister key. */ void lu_context_key_degister(struct lu_context_key *key) { - LASSERT(atomic_read(&key->lct_used) >= 1); - LASSERT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys)); + LASSERT(cfs_atomic_read(&key->lct_used) >= 1); + LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys)); + + lu_context_key_quiesce(key); + ++key_set_version; + cfs_spin_lock(&lu_keys_guard); key_fini(&lu_shrink_env.le_ctx, key->lct_index); + if (lu_keys[key->lct_index]) { + lu_keys[key->lct_index] = NULL; + lu_ref_fini(&key->lct_reference); + } + cfs_spin_unlock(&lu_keys_guard); - if (atomic_read(&key->lct_used) > 1) - CERROR("key has instances.\n"); - spin_lock(&lu_keys_guard); - lu_keys[key->lct_index] = NULL; - spin_unlock(&lu_keys_guard); + LASSERTF(cfs_atomic_read(&key->lct_used) == 1, + "key has instances: %d\n", + cfs_atomic_read(&key->lct_used)); } EXPORT_SYMBOL(lu_context_key_degister); -/* - * Return value associated with key @key in context @ctx. +/** + * Register a number of keys. This has to be called after all keys have been + * initialized by a call to LU_CONTEXT_KEY_INIT(). + */ +int lu_context_key_register_many(struct lu_context_key *k, ...) +{ + struct lu_context_key *key = k; + va_list args; + int result; + + va_start(args, k); + do { + result = lu_context_key_register(key); + if (result) + break; + key = va_arg(args, struct lu_context_key *); + } while (key != NULL); + va_end(args); + + if (result != 0) { + va_start(args, k); + while (k != key) { + lu_context_key_degister(k); + k = va_arg(args, struct lu_context_key *); + } + va_end(args); + } + + return result; +} +EXPORT_SYMBOL(lu_context_key_register_many); + +/** + * De-register a number of keys. This is a dual to + * lu_context_key_register_many(). + */ +void lu_context_key_degister_many(struct lu_context_key *k, ...) +{ + va_list args; + + va_start(args, k); + do { + lu_context_key_degister(k); + k = va_arg(args, struct lu_context_key*); + } while (k != NULL); + va_end(args); +} +EXPORT_SYMBOL(lu_context_key_degister_many); + +/** + * Revive a number of keys. + */ +void lu_context_key_revive_many(struct lu_context_key *k, ...) +{ + va_list args; + + va_start(args, k); + do { + lu_context_key_revive(k); + k = va_arg(args, struct lu_context_key*); + } while (k != NULL); + va_end(args); +} +EXPORT_SYMBOL(lu_context_key_revive_many); + +/** + * Quiescent a number of keys. + */ +void lu_context_key_quiesce_many(struct lu_context_key *k, ...) +{ + va_list args; + + va_start(args, k); + do { + lu_context_key_quiesce(k); + k = va_arg(args, struct lu_context_key*); + } while (k != NULL); + va_end(args); +} +EXPORT_SYMBOL(lu_context_key_quiesce_many); + +/** + * Return value associated with key \a key in context \a ctx. */ void *lu_context_key_get(const struct lu_context *ctx, - struct lu_context_key *key) + const struct lu_context_key *key) { - LASSERT(ctx->lc_state == LCS_ENTERED); - LASSERT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys)); + LINVRNT(ctx->lc_state == LCS_ENTERED); + LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys)); + LASSERT(lu_keys[key->lct_index] == key); return ctx->lc_value[key->lct_index]; } EXPORT_SYMBOL(lu_context_key_get); +/** + * List of remembered contexts. XXX document me. + */ +static CFS_LIST_HEAD(lu_context_remembered); + +/** + * Destroy \a key in all remembered contexts. This is used to destroy key + * values in "shared" contexts (like service threads), when a module owning + * the key is about to be unloaded. + */ +void lu_context_key_quiesce(struct lu_context_key *key) +{ + struct lu_context *ctx; + extern unsigned cl_env_cache_purge(unsigned nr); + + if (!(key->lct_tags & LCT_QUIESCENT)) { + /* + * XXX layering violation. + */ + cl_env_cache_purge(~0); + key->lct_tags |= LCT_QUIESCENT; + /* + * XXX memory barrier has to go here. + */ + cfs_spin_lock(&lu_keys_guard); + cfs_list_for_each_entry(ctx, &lu_context_remembered, + lc_remember) + key_fini(ctx, key->lct_index); + cfs_spin_unlock(&lu_keys_guard); + ++key_set_version; + } +} +EXPORT_SYMBOL(lu_context_key_quiesce); + +void lu_context_key_revive(struct lu_context_key *key) +{ + key->lct_tags &= ~LCT_QUIESCENT; + ++key_set_version; +} +EXPORT_SYMBOL(lu_context_key_revive); + static void keys_fini(struct lu_context *ctx) { int i; + cfs_spin_lock(&lu_keys_guard); if (ctx->lc_value != NULL) { for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) key_fini(ctx, i); @@ -884,9 +1261,10 @@ static void keys_fini(struct lu_context *ctx) ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]); ctx->lc_value = NULL; } + cfs_spin_unlock(&lu_keys_guard); } -static int keys_fill(const struct lu_context *ctx) +static int keys_fill(struct lu_context *ctx) { int i; @@ -894,22 +1272,37 @@ static int keys_fill(const struct lu_context *ctx) struct lu_context_key *key; key = lu_keys[i]; - if (ctx->lc_value[i] == NULL && - key != NULL && key->lct_tags & ctx->lc_tags) { + if (ctx->lc_value[i] == NULL && key != NULL && + (key->lct_tags & ctx->lc_tags) && + /* + * Don't create values for a LCT_QUIESCENT key, as this + * will pin module owning a key. + */ + !(key->lct_tags & LCT_QUIESCENT)) { void *value; - LASSERT(key->lct_init != NULL); - LASSERT(key->lct_index == i); + LINVRNT(key->lct_init != NULL); + LINVRNT(key->lct_index == i); value = key->lct_init(ctx, key); if (unlikely(IS_ERR(value))) return PTR_ERR(value); + LASSERT(key->lct_owner != NULL); if (!(ctx->lc_tags & LCT_NOREF)) - try_module_get(key->lct_owner); - atomic_inc(&key->lct_used); + cfs_try_module_get(key->lct_owner); + lu_ref_add_atomic(&key->lct_reference, "ctx", ctx); + cfs_atomic_inc(&key->lct_used); + /* + * This is the only place in the code, where an + * element of ctx->lc_value[] array is set to non-NULL + * value. + */ ctx->lc_value[i] = value; + if (key->lct_exit != NULL) + ctx->lc_tags |= LCT_HAS_EXIT; } + ctx->lc_version = key_set_version; } return 0; } @@ -929,7 +1322,7 @@ static int keys_init(struct lu_context *ctx) return result; } -/* +/** * Initialize context data-structure. Create values for all keys. */ int lu_context_init(struct lu_context *ctx, __u32 tags) @@ -937,41 +1330,50 @@ int lu_context_init(struct lu_context *ctx, __u32 tags) memset(ctx, 0, sizeof *ctx); ctx->lc_state = LCS_INITIALIZED; ctx->lc_tags = tags; + if (tags & LCT_REMEMBER) { + cfs_spin_lock(&lu_keys_guard); + cfs_list_add(&ctx->lc_remember, &lu_context_remembered); + cfs_spin_unlock(&lu_keys_guard); + } else + CFS_INIT_LIST_HEAD(&ctx->lc_remember); return keys_init(ctx); } EXPORT_SYMBOL(lu_context_init); -/* +/** * Finalize context data-structure. Destroy key values. */ void lu_context_fini(struct lu_context *ctx) { - LASSERT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT); + LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT); ctx->lc_state = LCS_FINALIZED; keys_fini(ctx); + cfs_spin_lock(&lu_keys_guard); + cfs_list_del_init(&ctx->lc_remember); + cfs_spin_unlock(&lu_keys_guard); } EXPORT_SYMBOL(lu_context_fini); -/* +/** * Called before entering context. */ void lu_context_enter(struct lu_context *ctx) { - LASSERT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT); + LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT); ctx->lc_state = LCS_ENTERED; } EXPORT_SYMBOL(lu_context_enter); -/* - * Called after exiting from @ctx +/** + * Called after exiting from \a ctx */ void lu_context_exit(struct lu_context *ctx) { int i; - LASSERT(ctx->lc_state == LCS_ENTERED); + LINVRNT(ctx->lc_state == LCS_ENTERED); ctx->lc_state = LCS_LEFT; - if (ctx->lc_value != NULL) { + if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value != NULL) { for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) { if (ctx->lc_value[i] != NULL) { struct lu_context_key *key; @@ -987,41 +1389,27 @@ void lu_context_exit(struct lu_context *ctx) } EXPORT_SYMBOL(lu_context_exit); -/* +/** * Allocate for context all missing keys that were registered after context * creation. */ -int lu_context_refill(const struct lu_context *ctx) +int lu_context_refill(struct lu_context *ctx) { - LASSERT(ctx->lc_value != NULL); - return keys_fill(ctx); + LINVRNT(ctx->lc_value != NULL); + return ctx->lc_version == key_set_version ? 0 : keys_fill(ctx); } EXPORT_SYMBOL(lu_context_refill); -static int lu_env_setup(struct lu_env *env, struct lu_context *ses, - __u32 tags, int noref) +int lu_env_init(struct lu_env *env, __u32 tags) { int result; - LASSERT(ergo(!noref, !(tags & LCT_NOREF))); - - env->le_ses = ses; + env->le_ses = NULL; result = lu_context_init(&env->le_ctx, tags); if (likely(result == 0)) lu_context_enter(&env->le_ctx); return result; } - -static int lu_env_init_noref(struct lu_env *env, struct lu_context *ses, - __u32 tags) -{ - return lu_env_setup(env, ses, tags, 1); -} - -int lu_env_init(struct lu_env *env, struct lu_context *ses, __u32 tags) -{ - return lu_env_setup(env, ses, tags, 0); -} EXPORT_SYMBOL(lu_env_init); void lu_env_fini(struct lu_env *env) @@ -1032,82 +1420,194 @@ void lu_env_fini(struct lu_env *env) } EXPORT_SYMBOL(lu_env_fini); +int lu_env_refill(struct lu_env *env) +{ + int result; + + result = lu_context_refill(&env->le_ctx); + if (result == 0 && env->le_ses != NULL) + result = lu_context_refill(env->le_ses); + return result; +} +EXPORT_SYMBOL(lu_env_refill); + +static struct cfs_shrinker *lu_site_shrinker = NULL; + +#ifdef __KERNEL__ static int lu_cache_shrink(int nr, unsigned int gfp_mask) { struct lu_site *s; struct lu_site *tmp; int cached = 0; int remain = nr; - LIST_HEAD(splice); + CFS_LIST_HEAD(splice); - if (nr != 0 && !(gfp_mask & __GFP_FS)) - return -1; + if (nr != 0) { + if (!(gfp_mask & __GFP_FS)) + return -1; + CDEBUG(D_INODE, "Shrink %d objects\n", nr); + } - down(&lu_sites_guard); - list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) { + cfs_down(&lu_sites_guard); + cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) { if (nr != 0) { remain = lu_site_purge(&lu_shrink_env, s, remain); /* * Move just shrunk site to the tail of site list to * assure shrinking fairness. */ - list_move_tail(&s->ls_linkage, &splice); + cfs_list_move_tail(&s->ls_linkage, &splice); } - read_lock(&s->ls_guard); + cfs_read_lock(&s->ls_guard); cached += s->ls_total - s->ls_busy; - read_unlock(&s->ls_guard); - if (remain <= 0) + cfs_read_unlock(&s->ls_guard); + if (nr && remain <= 0) break; } - list_splice(&splice, lu_sites.prev); - up(&lu_sites_guard); + cfs_list_splice(&splice, lu_sites.prev); + cfs_up(&lu_sites_guard); + + cached = (cached / 100) * sysctl_vfs_cache_pressure; + if (nr == 0) + CDEBUG(D_INODE, "%d objects cached\n", cached); return cached; } -static struct shrinker *lu_site_shrinker = NULL; - /* + * Debugging stuff. + */ + +/** + * Environment to be used in debugger, contains all tags. + */ +struct lu_env lu_debugging_env; + +/** + * Debugging printer function using printk(). + */ +int lu_printk_printer(const struct lu_env *env, + void *unused, const char *format, ...) +{ + va_list args; + + va_start(args, format); + vprintk(format, args); + va_end(args); + return 0; +} + +void lu_debugging_setup(void) +{ + lu_env_init(&lu_debugging_env, ~0); +} + +void lu_context_keys_dump(void) +{ + int i; + + for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) { + struct lu_context_key *key; + + key = lu_keys[i]; + if (key != NULL) { + CERROR("[%i]: %p %x (%p,%p,%p) %i %i \"%s\"@%p\n", + i, key, key->lct_tags, + key->lct_init, key->lct_fini, key->lct_exit, + key->lct_index, cfs_atomic_read(&key->lct_used), + key->lct_owner ? key->lct_owner->name : "", + key->lct_owner); + lu_ref_print(&key->lct_reference); + } + } +} +EXPORT_SYMBOL(lu_context_keys_dump); +#else /* !__KERNEL__ */ +static int lu_cache_shrink(int nr, unsigned int gfp_mask) +{ + return 0; +} +#endif /* __KERNEL__ */ + +int cl_global_init(void); +void cl_global_fini(void); +int lu_ref_global_init(void); +void lu_ref_global_fini(void); + +int dt_global_init(void); +void dt_global_fini(void); + +int llo_global_init(void); +void llo_global_fini(void); + +/** * Initialization of global lu_* data. */ int lu_global_init(void) { int result; + CDEBUG(D_CONSOLE, "Lustre LU module (%p).\n", &lu_keys); + + result = lu_ref_global_init(); + if (result != 0) + return result; + LU_CONTEXT_KEY_INIT(&lu_global_key); result = lu_context_key_register(&lu_global_key); - if (result == 0) { - /* - * At this level, we don't know what tags are needed, so - * allocate them conservatively. This should not be too bad, - * because this environment is global. - */ - down(&lu_sites_guard); - result = lu_env_init_noref(&lu_shrink_env, NULL, LCT_SHRINKER); - up(&lu_sites_guard); - if (result == 0) { - /* - * seeks estimation: 3 seeks to read a record from oi, - * one to read inode, one for ea. Unfortunately - * setting this high value results in lu_object/inode - * cache consuming all the memory. - */ - lu_site_shrinker = set_shrinker(DEFAULT_SEEKS, - lu_cache_shrink); - if (result == 0) - result = lu_time_global_init(); - } - } + if (result != 0) + return result; + /* + * At this level, we don't know what tags are needed, so allocate them + * conservatively. This should not be too bad, because this + * environment is global. + */ + cfs_down(&lu_sites_guard); + result = lu_env_init(&lu_shrink_env, LCT_SHRINKER); + cfs_up(&lu_sites_guard); + if (result != 0) + return result; + + /* + * seeks estimation: 3 seeks to read a record from oi, one to read + * inode, one for ea. Unfortunately setting this high value results in + * lu_object/inode cache consuming all the memory. + */ + lu_site_shrinker = cfs_set_shrinker(CFS_DEFAULT_SEEKS, lu_cache_shrink); + if (lu_site_shrinker == NULL) + return -ENOMEM; + + result = lu_time_global_init(); + if (result) + GOTO(out, result); + +#ifdef __KERNEL__ + result = dt_global_init(); + if (result) + GOTO(out, result); + + result = llo_global_init(); + if (result) + GOTO(out, result); +#endif + result = cl_global_init(); +out: + return result; } -/* +/** * Dual to lu_global_init(). */ void lu_global_fini(void) { + cl_global_fini(); +#ifdef __KERNEL__ + llo_global_fini(); + dt_global_fini(); +#endif lu_time_global_fini(); if (lu_site_shrinker != NULL) { - remove_shrinker(lu_site_shrinker); + cfs_remove_shrinker(lu_site_shrinker); lu_site_shrinker = NULL; } @@ -1117,9 +1617,11 @@ void lu_global_fini(void) * Tear shrinker environment down _after_ de-registering * lu_global_key, because the latter has a value in the former. */ - down(&lu_sites_guard); + cfs_down(&lu_sites_guard); lu_env_fini(&lu_shrink_env); - up(&lu_sites_guard); + cfs_up(&lu_sites_guard); + + lu_ref_global_fini(); } struct lu_buf LU_BUF_NULL = { @@ -1128,84 +1630,79 @@ struct lu_buf LU_BUF_NULL = { }; EXPORT_SYMBOL(LU_BUF_NULL); -/* - * XXX: Functions below logically belong to fid module, but they are used by - * dt_store_open(). Put them here until better place is found. +/** + * Output site statistical counters into a buffer. Suitable for + * lprocfs_rd_*()-style functions. */ - -void fid_pack(struct lu_fid_pack *pack, const struct lu_fid *fid, - struct lu_fid *befider) +int lu_site_stats_print(const struct lu_site *s, char *page, int count) { - int recsize; - __u64 seq; - __u32 oid; - - seq = fid_seq(fid); - oid = fid_oid(fid); - - /* - * Two cases: compact 6 bytes representation for a common case, and - * full 17 byte representation for "unusual" fid. - */ + int i; + int populated; /* - * Check that usual case is really usual. + * How many hash buckets are not-empty? Don't bother with locks: it's + * an estimation anyway. */ - CLASSERT(LUSTRE_SEQ_MAX_WIDTH < 0xffffull); - - if (fid_is_igif(fid) || - seq > 0xffffffull || oid > 0xffff || fid_ver(fid) != 0) { - fid_cpu_to_be(befider, fid); - recsize = sizeof *befider; - } else { - unsigned char *small_befider; - - small_befider = (char *)befider; - - small_befider[0] = seq >> 16; - small_befider[1] = seq >> 8; - small_befider[2] = seq; - - small_befider[3] = oid >> 8; - small_befider[4] = oid; - - recsize = 5; - } - memcpy(pack->fp_area, befider, recsize); - pack->fp_len = recsize + 1; + for (i = 0, populated = 0; i < s->ls_hash_size; i++) + populated += !cfs_hlist_empty(&s->ls_hash[i]); + + return snprintf(page, count, "%d %d %d/%d %d %d %d %d %d %d %d\n", + s->ls_total, + s->ls_busy, + populated, + s->ls_hash_size, + s->ls_stats.s_created, + s->ls_stats.s_cache_hit, + s->ls_stats.s_cache_miss, + s->ls_stats.s_cache_check, + s->ls_stats.s_cache_race, + s->ls_stats.s_cache_death_race, + s->ls_stats.s_lru_purged); } -EXPORT_SYMBOL(fid_pack); +EXPORT_SYMBOL(lu_site_stats_print); + +const char *lu_time_names[LU_TIME_NR] = { + [LU_TIME_FIND_LOOKUP] = "find_lookup", + [LU_TIME_FIND_ALLOC] = "find_alloc", + [LU_TIME_FIND_INSERT] = "find_insert" +}; +EXPORT_SYMBOL(lu_time_names); -int fid_unpack(const struct lu_fid_pack *pack, struct lu_fid *fid) +/** + * Helper function to initialize a number of kmem slab caches at once. + */ +int lu_kmem_init(struct lu_kmem_descr *caches) { int result; - result = 0; - switch (pack->fp_len) { - case sizeof *fid + 1: - memcpy(fid, pack->fp_area, sizeof *fid); - fid_be_to_cpu(fid, fid); - break; - case 6: { - const unsigned char *area; - - area = pack->fp_area; - fid->f_seq = (area[0] << 16) | (area[1] << 8) | area[2]; - fid->f_oid = (area[3] << 8) | area[4]; - fid->f_ver = 0; - break; - } - default: - CERROR("Unexpected packed fid size: %d\n", pack->fp_len); - result = -EIO; + for (result = 0; caches->ckd_cache != NULL; ++caches) { + *caches->ckd_cache = cfs_mem_cache_create(caches->ckd_name, + caches->ckd_size, + 0, 0); + if (*caches->ckd_cache == NULL) { + result = -ENOMEM; + break; + } } return result; } -EXPORT_SYMBOL(fid_unpack); +EXPORT_SYMBOL(lu_kmem_init); -const char *lu_time_names[LU_TIME_NR] = { - [LU_TIME_FIND_LOOKUP] = "find_lookup", - [LU_TIME_FIND_ALLOC] = "find_alloc", - [LU_TIME_FIND_INSERT] = "find_insert" -}; -EXPORT_SYMBOL(lu_time_names); +/** + * Helper function to finalize a number of kmem slab cached at once. Dual to + * lu_kmem_init(). + */ +void lu_kmem_fini(struct lu_kmem_descr *caches) +{ + int rc; + + for (; caches->ckd_cache != NULL; ++caches) { + if (*caches->ckd_cache != NULL) { + rc = cfs_mem_cache_destroy(*caches->ckd_cache); + LASSERTF(rc == 0, "couldn't destroy %s slab\n", + caches->ckd_name); + *caches->ckd_cache = NULL; + } + } +} +EXPORT_SYMBOL(lu_kmem_fini);