X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fobdclass%2Flu_object.c;h=3023736502887887691f60a1c27db0077dbbdc07;hb=aaf5a7adcc310aa09f0b0265d3dc64742099633b;hp=30c4fe28f1b71a616f34daba101170b90f107f56;hpb=2b5ed846625fb0cd0c1241afd82ddcd18df52d8e;p=fs%2Flustre-release.git diff --git a/lustre/obdclass/lu_object.c b/lustre/obdclass/lu_object.c index 30c4fe2..3023736 100644 --- a/lustre/obdclass/lu_object.c +++ b/lustre/obdclass/lu_object.c @@ -26,7 +26,7 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* @@ -73,53 +73,66 @@ static void lu_object_free(const struct lu_env *env, struct lu_object *o); */ void lu_object_put(const struct lu_env *env, struct lu_object *o) { + struct lu_site_bkt_data *bkt; struct lu_object_header *top; struct lu_site *site; struct lu_object *orig; - int kill_it; + cfs_hash_bd_t bd; - top = o->lo_header; + top = o->lo_header; site = o->lo_dev->ld_site; orig = o; - kill_it = 0; - write_lock(&site->ls_guard); - if (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) { - if (o->lo_ops->loo_object_release != NULL) - o->lo_ops->loo_object_release(env, o); - } - -- site->ls_busy; + + cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd); + bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd); + + if (!cfs_hash_bd_dec_and_lock(site->ls_obj_hash, &bd, &top->loh_ref)) { if (lu_object_is_dying(top)) { + /* - * If object is dying (will not be cached), removed it - * from hash table and LRU. - * - * This is done with hash table and LRU lists - * locked. As the only way to acquire first reference - * to previously unreferenced object is through - * hash-table lookup (lu_object_find()), or LRU - * scanning (lu_site_purge()), that are done under - * hash-table and LRU lock, no race with concurrent - * object lookup is possible and we can safely destroy - * object below. + * somebody may be waiting for this, currently only + * used for cl_object, see cl_object_put_last(). */ - hlist_del_init(&top->loh_hash); - list_del_init(&top->loh_lru); - -- site->ls_total; - kill_it = 1; + cfs_waitq_broadcast(&bkt->lsb_marche_funebre); } + return; } - write_unlock(&site->ls_guard); - if (kill_it) - /* - * Object was already removed from hash and lru above, can - * kill it. - */ - lu_object_free(env, orig); + + LASSERT(bkt->lsb_busy > 0); + bkt->lsb_busy--; + /* + * When last reference is released, iterate over object + * layers, and notify them that object is no longer busy. + */ + 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); + } + + if (!lu_object_is_dying(top)) { + cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1); + return; + } + + /* + * If object is dying (will not be cached), removed it + * from hash table and LRU. + * + * This is done with hash table and LRU lists locked. As the only + * way to acquire first reference to previously unreferenced + * object is through hash-table lookup (lu_object_find()), + * or LRU scanning (lu_site_purge()), that are done under hash-table + * and LRU lock, no race with concurrent object lookup is possible + * and we can safely destroy object below. + */ + cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash); + cfs_list_del_init(&top->loh_lru); + cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1); + /* + * Object was already removed from hash and lru above, can + * kill it. + */ + lu_object_free(env, orig); } EXPORT_SYMBOL(lu_object_put); @@ -136,7 +149,7 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, { struct lu_object *scan; struct lu_object *top; - struct list_head *layers; + cfs_list_t *layers; int clean; int result; ENTRY; @@ -153,7 +166,7 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, * after this point. */ LASSERT(fid_is_igif(f) || fid_ver(f) == 0); - top->lo_header->loh_fid = *f; + top->lo_header->loh_fid = *f; layers = &top->lo_header->loh_layers; do { /* @@ -161,7 +174,7 @@ 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; @@ -175,7 +188,7 @@ 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) { @@ -185,7 +198,7 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, } } - dev->ld_site->ls_stats.s_created ++; + lprocfs_counter_incr(dev->ld_site->ls_stats, LU_SS_CREATED); RETURN(top); } @@ -194,17 +207,19 @@ static struct lu_object *lu_object_alloc(const struct lu_env *env, */ static void lu_object_free(const struct lu_env *env, struct lu_object *o) { - struct list_head splice; - struct lu_object *scan; + struct lu_site_bkt_data *bkt; struct lu_site *site; - struct list_head *layers; + struct lu_object *scan; + cfs_list_t *layers; + cfs_list_t splice; site = o->lo_dev->ld_site; layers = &o->lo_header->loh_layers; + bkt = lu_site_bkt_from_fid(site, &o->lo_header->loh_fid); /* * First call ->loo_object_delete() method to release all resources. */ - 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_delete != NULL) scan->lo_ops->loo_object_delete(env, scan); } @@ -216,19 +231,21 @@ static void lu_object_free(const struct lu_env *env, struct lu_object *o) * top-level slice. */ CFS_INIT_LIST_HEAD(&splice); - list_splice_init(layers, &splice); - while (!list_empty(&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); - list_del_init(&o->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); + + if (cfs_waitq_active(&bkt->lsb_marche_funebre)) + cfs_waitq_broadcast(&bkt->lsb_marche_funebre); } /** @@ -236,47 +253,91 @@ static void lu_object_free(const struct lu_env *env, struct lu_object *o) */ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr) { - struct list_head dispose; struct lu_object_header *h; struct lu_object_header *temp; + struct lu_site_bkt_data *bkt; + cfs_hash_bd_t bd; + cfs_hash_bd_t bd2; + cfs_list_t dispose; + int did_sth; + int start; + int count; + int bnr; + int i; 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) { + start = s->ls_purge_start; + bnr = (nr == ~0) ? -1 : nr / CFS_HASH_NBKT(s->ls_obj_hash) + 1; + again: + did_sth = 0; + cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) { + if (i < start) + continue; + count = bnr; + cfs_hash_bd_lock(s->ls_obj_hash, &bd, 1); + bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd); + + cfs_list_for_each_entry_safe(h, temp, &bkt->lsb_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 that we scan last. Unfortunately, + * sites usually have small (less then ten) number of + * busy yet rarely accessed objects (some global + * objects, accessed directly through pointers, + * bypassing hash table). + * Currently algorithm scans them over and over again. + * Probably we should move busy objects out of LRU, + * or we can live with that. + */ + if (cfs_atomic_read(&h->loh_ref) > 0) + continue; + + cfs_hash_bd_get(s->ls_obj_hash, &h->loh_fid, &bd2); + LASSERT(bd.bd_bucket == bd2.bd_bucket); + + cfs_hash_bd_del_locked(s->ls_obj_hash, + &bd2, &h->loh_hash); + cfs_list_move(&h->loh_lru, &dispose); + if (did_sth == 0) + did_sth = 1; + + if (nr != ~0 && --nr == 0) + break; + + if (count > 0 && --count == 0) + break; + + } + cfs_hash_bd_unlock(s->ls_obj_hash, &bd, 1); + cfs_cond_resched(); /* - * Objects are sorted in lru order, and "busy" objects (ones - * with h->loh_ref > 0) naturally tend to live near hot end - * that we scan last. Unfortunately, sites usually have small - * (less then ten) number of busy yet rarely accessed objects - * (some global objects, accessed directly through pointers, - * bypassing hash table). Currently algorithm scans them over - * and over again. Probably we should move busy objects out of - * LRU, or we can live with that. + * Free everything on the dispose list. This is safe against + * races due to the reasons described in lu_object_put(). */ - if (nr-- == 0) + while (!cfs_list_empty(&dispose)) { + h = container_of0(dispose.next, + struct lu_object_header, loh_lru); + cfs_list_del_init(&h->loh_lru); + lu_object_free(env, lu_object_top(h)); + lprocfs_counter_incr(s->ls_stats, LU_SS_LRU_PURGED); + } + + if (nr == 0) break; - if (atomic_read(&h->loh_ref) > 0) - continue; - hlist_del_init(&h->loh_hash); - list_move(&h->loh_lru, &dispose); - s->ls_total --; } - 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)) { - h = container_of0(dispose.next, - struct lu_object_header, loh_lru); - list_del_init(&h->loh_lru); - lu_object_free(env, lu_object_top(h)); - s->ls_stats.s_lru_purged ++; + + if (nr != 0 && did_sth && start != 0) { + start = 0; /* restart from the first bucket */ + goto again; } + /* race on s->ls_purge_start, but nobody cares */ + s->ls_purge_start = i % CFS_HASH_NBKT(s->ls_obj_hash); + return nr; } EXPORT_SYMBOL(lu_site_purge); @@ -336,7 +397,7 @@ 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); @@ -351,9 +412,10 @@ 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); @@ -361,23 +423,25 @@ 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) @@ -388,21 +452,24 @@ 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) @@ -410,7 +477,7 @@ 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; @@ -420,51 +487,77 @@ 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) + cfs_hash_bd_t *bd, + const struct lu_fid *f, + cfs_waitlink_t *waiter, + __u64 *version) { + struct lu_site_bkt_data *bkt; struct lu_object_header *h; - struct hlist_node *scan; - - 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))) { - /* bump reference count... */ - if (atomic_add_return(1, &h->loh_ref) == 1) - ++ s->ls_busy; - /* and move to the head of the LRU */ - /* - * XXX temporary disable this to measure effects of - * read-write locking. - */ - /* list_move_tail(&h->loh_lru, &s->ls_lru); */ - s->ls_stats.s_cache_hit ++; - return lu_object_top(h); - } + cfs_hlist_node_t *hnode; + __u64 ver = cfs_hash_bd_version_get(bd); + + if (*version == ver) + return NULL; + + *version = ver; + bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd); + /* cfs_hash_bd_lookup_intent is a somehow "internal" function + * of cfs_hash, but we don't want refcount on object right now */ + hnode = cfs_hash_bd_lookup_locked(s->ls_obj_hash, bd, (void *)f); + if (hnode == NULL) { + lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS); + return NULL; } - s->ls_stats.s_cache_miss ++; - return NULL; + + h = container_of0(hnode, struct lu_object_header, loh_hash); + if (likely(!lu_object_is_dying(h))) { + lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT); + return lu_object_top(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_atomic_dec(&h->loh_ref); + + cfs_waitlink_init(waiter); + cfs_waitq_add(&bkt->lsb_marche_funebre, waiter); + cfs_set_current_state(CFS_TASK_UNINT); + lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_DEATH_RACE); + return ERR_PTR(-EAGAIN); } -static __u32 fid_hash(const struct lu_fid *f, int bits) +/** + * 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_device *dev, const struct lu_fid *f, + const struct lu_object_conf *conf) { - /* all objects with same id and different versions will belong to same - * collisions list. */ - return hash_long(fid_flatten(f), bits); + return lu_object_find_at(env, dev->ld_site->ls_top_dev, f, conf); } +EXPORT_SYMBOL(lu_object_find); -/* - * 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. +/** + * Core logic of lu_object_find*() functions. */ -struct lu_object *lu_object_find(const struct lu_env *env, - struct lu_site *s, const struct lu_fid *f) +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_object *o; + struct lu_object *shadow; + struct lu_site *s; + cfs_hash_t *hs; + cfs_hash_bd_t bd; + __u64 version = 0; /* * This uses standard index maintenance protocol: @@ -478,14 +571,15 @@ 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). */ - - 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); - + s = dev->ld_site; + hs = s->ls_obj_hash; + cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1); + o = htable_lookup(s, &bd, f, waiter, &version); + cfs_hash_bd_unlock(hs, &bd, 1); if (o != NULL) return o; @@ -493,66 +587,169 @@ 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_hash_bd_lock(hs, &bd, 1); + + shadow = htable_lookup(s, &bd, f, waiter, &version); if (likely(shadow == NULL)) { - hlist_add_head(&o->lo_header->loh_hash, bucket); - 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); - if (o != NULL) - lu_object_free(env, o); + struct lu_site_bkt_data *bkt; + + bkt = cfs_hash_bd_extra_get(hs, &bd); + cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash); + cfs_list_add_tail(&o->lo_header->loh_lru, &bkt->lsb_lru); + bkt->lsb_busy++; + cfs_hash_bd_unlock(hs, &bd, 1); + return o; + } + + lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_RACE); + cfs_hash_bd_unlock(hs, &bd, 1); + 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_site_bkt_data *bkt; + 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)) + return obj; + /* + * lu_object_find_try() already added waiter into the + * wait queue. + */ + cfs_waitq_wait(&wait, CFS_TASK_UNINT); + bkt = lu_site_bkt_from_fid(dev->ld_site, (void *)f); + cfs_waitq_del(&bkt->lsb_marche_funebre, &wait); + } +} +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 CFS_LIST_HEAD(lu_sites); -static DECLARE_MUTEX(lu_sites_guard); +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. - */ -void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie, - lu_printer_t printer) +struct lu_site_print_arg { + struct lu_env *lsp_env; + void *lsp_cookie; + lu_printer_t lsp_printer; +}; + +static int +lu_site_obj_print(cfs_hash_t *hs, cfs_hash_bd_t *bd, + cfs_hlist_node_t *hnode, void *data) { - int i; + struct lu_site_print_arg *arg = (struct lu_site_print_arg *)data; + struct lu_object_header *h; - for (i = 0; i < s->ls_hash_size; ++i) { - struct lu_object_header *h; - struct hlist_node *scan; + h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash); + if (!cfs_list_empty(&h->loh_layers)) { + const struct lu_object *o; - read_lock(&s->ls_guard); - hlist_for_each_entry(h, scan, &s->ls_hash[i], loh_hash) { + o = lu_object_top(h); + lu_object_print(arg->lsp_env, arg->lsp_cookie, + arg->lsp_printer, o); + } else { + lu_object_header_print(arg->lsp_env, arg->lsp_cookie, + arg->lsp_printer, h); + } + return 0; +} - if (!list_empty(&h->loh_layers)) { - const struct lu_object *obj; +/** + * 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) +{ + struct lu_site_print_arg arg = { + .lsp_env = (struct lu_env *)env, + .lsp_cookie = cookie, + .lsp_printer = printer, + }; - obj = lu_object_top(h); - lu_object_print(env, cookie, printer, obj); - } else - lu_object_header_print(env, cookie, printer, h); - } - read_unlock(&s->ls_guard); - } + cfs_hash_for_each(s->ls_obj_hash, lu_site_obj_print, &arg); } EXPORT_SYMBOL(lu_site_print); @@ -560,7 +757,7 @@ enum { LU_CACHE_PERCENT = 20, }; -/* +/** * Return desired hash table order. */ static int lu_htable_order(void) @@ -575,7 +772,7 @@ static int lu_htable_order(void) * * Size of lu_object is (arbitrary) taken as 1K (together with inode). */ - cache_size = num_physpages; + cache_size = cfs_num_physpages; #if BITS_PER_LONG == 32 /* limit hashtable size for lowmem systems to low RAM */ @@ -592,134 +789,243 @@ static int lu_htable_order(void) return bits; } -/* - * Initialize site @s, with @d as the top level device. +static unsigned lu_obj_hop_hash(cfs_hash_t *hs, void *key, unsigned mask) +{ + struct lu_fid *fid = (struct lu_fid *)key; + unsigned hash; + + hash = (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1); + hash += fid_hash(fid, hs->hs_bkt_bits) << hs->hs_bkt_bits; + return hash & mask; +} + +static void *lu_obj_hop_object(cfs_hlist_node_t *hnode) +{ + return cfs_hlist_entry(hnode, struct lu_object_header, loh_hash); +} + +static void *lu_obj_hop_key(cfs_hlist_node_t *hnode) +{ + struct lu_object_header *h; + + h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash); + return &h->loh_fid; +} + +static int lu_obj_hop_keycmp(void *key, cfs_hlist_node_t *hnode) +{ + struct lu_object_header *h; + + h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash); + return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key); +} + +static void lu_obj_hop_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode) +{ + struct lu_object_header *h; + + h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash); + if (cfs_atomic_add_return(1, &h->loh_ref) == 1) { + struct lu_site_bkt_data *bkt; + cfs_hash_bd_t bd; + + cfs_hash_bd_get(hs, &h->loh_fid, &bd); + bkt = cfs_hash_bd_extra_get(hs, &bd); + bkt->lsb_busy++; + } +} + +static void lu_obj_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode) +{ + LBUG(); /* we should never called it */ +} + +cfs_hash_ops_t lu_site_hash_ops = { + .hs_hash = lu_obj_hop_hash, + .hs_key = lu_obj_hop_key, + .hs_keycmp = lu_obj_hop_keycmp, + .hs_object = lu_obj_hop_object, + .hs_get = lu_obj_hop_get, + .hs_put_locked = lu_obj_hop_put_locked, +}; + +/** + * Initialize site \a s, with \a d as the top level device. + */ +#define LU_SITE_BITS_MIN 12 +#define LU_SITE_BITS_MAX 23 +/** + * total 128 buckets, we don't want too many buckets because: + * - consume too much memory + * - avoid unbalanced LRU list */ +#define LU_SITE_BKT_BITS 7 + int lu_site_init(struct lu_site *s, struct lu_device *top) { + struct lu_site_bkt_data *bkt; + cfs_hash_bd_t bd; int bits; - int size; int i; ENTRY; memset(s, 0, sizeof *s); - rwlock_init(&s->ls_guard); - CFS_INIT_LIST_HEAD(&s->ls_lru); - CFS_INIT_LIST_HEAD(&s->ls_linkage); - s->ls_top_dev = top; - top->ld_site = s; - lu_device_get(top); + bits = lu_htable_order(); + for (bits = min(max(LU_SITE_BITS_MIN, bits), LU_SITE_BITS_MAX); + bits >= LU_SITE_BITS_MIN; bits--) { + s->ls_obj_hash = cfs_hash_create("lu_site", bits, bits, + bits - LU_SITE_BKT_BITS, + sizeof(*bkt), 0, 0, + &lu_site_hash_ops, + CFS_HASH_SPIN_BKTLOCK | + CFS_HASH_NO_ITEMREF | + CFS_HASH_DEPTH | + CFS_HASH_ASSERT_EMPTY); + if (s->ls_obj_hash != NULL) + break; + } - for (bits = lu_htable_order(), size = 1 << bits; - (s->ls_hash = - cfs_alloc_large(size * sizeof s->ls_hash[0])) == NULL; - --bits, size >>= 1) { - /* - * Scale hash table down, until allocation succeeds. - */ - ; + if (s->ls_obj_hash == NULL) { + CERROR("failed to create lu_site hash with bits: %d\n", bits); + return -ENOMEM; + } + + cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) { + bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd); + CFS_INIT_LIST_HEAD(&bkt->lsb_lru); + cfs_waitq_init(&bkt->lsb_marche_funebre); } - s->ls_hash_size = size; - s->ls_hash_bits = bits; - s->ls_hash_mask = size - 1; + s->ls_stats = lprocfs_alloc_stats(LU_SS_LAST_STAT, 0); + if (s->ls_stats == NULL) { + cfs_hash_putref(s->ls_obj_hash); + s->ls_obj_hash = NULL; + return -ENOMEM; + } - for (i = 0; i < size; i++) - INIT_HLIST_HEAD(&s->ls_hash[i]); + lprocfs_counter_init(s->ls_stats, LU_SS_CREATED, + 0, "created", "created"); + lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_HIT, + 0, "cache_hit", "cache_hit"); + lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_MISS, + 0, "cache_miss", "cache_miss"); + lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_RACE, + 0, "cache_race", "cache_race"); + lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_DEATH_RACE, + 0, "cache_death_race", "cache_death_race"); + lprocfs_counter_init(s->ls_stats, LU_SS_LRU_PURGED, + 0, "lru_purged", "lru_purged"); + + CFS_INIT_LIST_HEAD(&s->ls_linkage); + s->ls_top_dev = top; + top->ld_site = s; + lu_device_get(top); + lu_ref_add(&top->ld_reference, "site-top", s); 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(s->ls_total == 0); + cfs_down(&lu_sites_guard); + cfs_list_del_init(&s->ls_linkage); + cfs_up(&lu_sites_guard); - down(&lu_sites_guard); - list_del_init(&s->ls_linkage); - 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])); - cfs_free_large(s->ls_hash); - s->ls_hash = NULL; + if (s->ls_obj_hash != NULL) { + cfs_hash_putref(s->ls_obj_hash); + s->ls_obj_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; } + + if (s->ls_stats != NULL) + lprocfs_free_stats(&s->ls_stats); } 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; - LASSERTF(atomic_read(&d->ld_ref) == 0, - "Refcount is %u\n", atomic_read(&d->ld_ref)); + t = d->ld_type; + if (d->ld_obd != NULL) { + d->ld_obd->obd_lu_dev = NULL; + d->ld_obd = NULL; + } + + 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) @@ -728,34 +1034,39 @@ 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); @@ -767,7 +1078,7 @@ EXPORT_SYMBOL(lu_object_add_top); */ 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); @@ -777,35 +1088,37 @@ EXPORT_SYMBOL(lu_object_add); 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; } @@ -815,9 +1128,9 @@ 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. @@ -831,13 +1144,14 @@ void lu_stack_fini(const struct lu_env *env, struct lu_device *top) 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 (!list_empty(&site->ls_lru) || site->ls_total != 0) { + if (!cfs_hash_is_empty(site->ls_obj_hash)) { /* * Uh-oh, objects still exist. */ @@ -853,9 +1167,9 @@ void lu_stack_fini(const struct lu_env *env, struct lu_device *top) next = ldt->ldt_ops->ldto_device_free(env, scan); type = ldt->ldt_obd_type; if (type != NULL) { - type->typ_refcnt--; - class_put_type(type); - } + type->typ_refcnt--; + class_put_type(type); + } } } EXPORT_SYMBOL(lu_stack_fini); @@ -869,7 +1183,7 @@ enum { 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, @@ -893,11 +1207,11 @@ 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; @@ -905,28 +1219,28 @@ int lu_context_key_register(struct lu_context_key *key) 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]); lu_ref_del(&key->lct_reference, "ctx", ctx); - atomic_dec(&key->lct_used); + 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; } @@ -937,17 +1251,23 @@ static void key_fini(struct lu_context *ctx, int index) */ void lu_context_key_degister(struct lu_context_key *key) { - LASSERT(atomic_read(&key->lct_used) >= 1); + 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); @@ -1040,6 +1360,7 @@ void *lu_context_key_get(const struct lu_context *ctx, { 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); @@ -1057,16 +1378,22 @@ static CFS_LIST_HEAD(lu_context_remembered); 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. */ - spin_lock(&lu_keys_guard); - list_for_each_entry(ctx, &lu_context_remembered, lc_remember) + cfs_spin_lock(&lu_keys_guard); + cfs_list_for_each_entry(ctx, &lu_context_remembered, + lc_remember) key_fini(ctx, key->lct_index); - spin_unlock(&lu_keys_guard); + cfs_spin_unlock(&lu_keys_guard); ++key_set_version; } } @@ -1083,6 +1410,7 @@ 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); @@ -1090,6 +1418,7 @@ 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(struct lu_context *ctx) @@ -1115,11 +1444,12 @@ static int keys_fill(struct lu_context *ctx) 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); + cfs_try_module_get(key->lct_owner); lu_ref_add_atomic(&key->lct_reference, "ctx", ctx); - atomic_inc(&key->lct_used); + 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 @@ -1158,9 +1488,9 @@ int lu_context_init(struct lu_context *ctx, __u32 tags) ctx->lc_state = LCS_INITIALIZED; ctx->lc_tags = tags; if (tags & LCT_REMEMBER) { - spin_lock(&lu_keys_guard); - list_add(&ctx->lc_remember, &lu_context_remembered); - spin_unlock(&lu_keys_guard); + 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); @@ -1175,9 +1505,9 @@ void lu_context_fini(struct lu_context *ctx) LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT); ctx->lc_state = LCS_FINALIZED; keys_fini(ctx); - spin_lock(&lu_keys_guard); - list_del_init(&ctx->lc_remember); - spin_unlock(&lu_keys_guard); + cfs_spin_lock(&lu_keys_guard); + cfs_list_del_init(&ctx->lc_remember); + cfs_spin_unlock(&lu_keys_guard); } EXPORT_SYMBOL(lu_context_fini); @@ -1227,30 +1557,16 @@ int lu_context_refill(struct lu_context *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; - LINVRNT(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) @@ -1272,41 +1588,134 @@ int lu_env_refill(struct lu_env *env) } EXPORT_SYMBOL(lu_env_refill); -static struct shrinker *lu_site_shrinker = NULL; +static struct cfs_shrinker *lu_site_shrinker = NULL; + +typedef struct lu_site_stats{ + unsigned lss_populated; + unsigned lss_max_search; + unsigned lss_total; + unsigned lss_busy; +} lu_site_stats_t; + +static void lu_site_stats_get(cfs_hash_t *hs, + lu_site_stats_t *stats, int populated) +{ + cfs_hash_bd_t bd; + int i; + + cfs_hash_for_each_bucket(hs, &bd, i) { + struct lu_site_bkt_data *bkt = cfs_hash_bd_extra_get(hs, &bd); + cfs_hlist_head_t *hhead; + + cfs_hash_bd_lock(hs, &bd, 1); + stats->lss_busy += bkt->lsb_busy; + stats->lss_total += cfs_hash_bd_count_get(&bd); + stats->lss_max_search = max((int)stats->lss_max_search, + cfs_hash_bd_depmax_get(&bd)); + if (!populated) { + cfs_hash_bd_unlock(hs, &bd, 1); + continue; + } + + cfs_hash_bd_for_each_hlist(hs, &bd, hhead) { + if (!cfs_hlist_empty(hhead)) + stats->lss_populated++; + } + cfs_hash_bd_unlock(hs, &bd, 1); + } +} #ifdef __KERNEL__ -static int lu_cache_shrink(int nr, unsigned int gfp_mask) +static int lu_cache_shrink(SHRINKER_FIRST_ARG int nr_to_scan, + unsigned int gfp_mask) { + lu_site_stats_t stats; struct lu_site *s; struct lu_site *tmp; int cached = 0; - int remain = nr; + int remain = nr_to_scan; CFS_LIST_HEAD(splice); - if (nr != 0 && !(gfp_mask & __GFP_FS)) - return -1; + if (nr_to_scan != 0) { + if (!(gfp_mask & __GFP_FS)) + return -1; + CDEBUG(D_INODE, "Shrink %d objects\n", nr_to_scan); + } - down(&lu_sites_guard); - list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) { - if (nr != 0) { + cfs_down(&lu_sites_guard); + cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) { + if (nr_to_scan != 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); - cached += s->ls_total - s->ls_busy; - read_unlock(&s->ls_guard); - if (remain <= 0) + + memset(&stats, 0, sizeof(stats)); + lu_site_stats_get(s->ls_obj_hash, &stats, 0); + cached += stats.lss_total - stats.lss_busy; + if (nr_to_scan && 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_to_scan == 0) + CDEBUG(D_INODE, "%d objects cached\n", cached); return cached; } +/* + * 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("[%d]: %p %x (%p,%p,%p) %d %d \"%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) { @@ -1314,9 +1723,17 @@ static int lu_cache_shrink(int nr, unsigned int gfp_mask) } #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. */ @@ -1326,34 +1743,50 @@ int lu_global_init(void) 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) 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. - */ - down(&lu_sites_guard); - result = lu_env_init_noref(&lu_shrink_env, NULL, LCT_SHRINKER); - up(&lu_sites_guard); + */ + cfs_down(&lu_sites_guard); + result = lu_env_init(&lu_shrink_env, LCT_SHRINKER); + cfs_up(&lu_sites_guard); if (result != 0) return result; - result = lu_ref_global_init(); - 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 = set_shrinker(DEFAULT_SEEKS, lu_cache_shrink); + lu_site_shrinker = cfs_set_shrinker(CFS_DEFAULT_SEEKS, lu_cache_shrink); if (lu_site_shrinker == NULL) return -ENOMEM; - result = lu_time_global_init(); + 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; } @@ -1362,9 +1795,14 @@ int lu_global_init(void) */ 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; } @@ -1374,9 +1812,9 @@ 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(); } @@ -1387,80 +1825,43 @@ 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. - */ - -void fid_pack(struct lu_fid_pack *pack, const struct lu_fid *fid, - struct lu_fid *befider) +static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx) { - 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. - */ - - /* - * Check that usual case is really usual. - */ - 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; +#ifdef LPROCFS + struct lprocfs_counter ret; - 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; + lprocfs_stats_collect(stats, idx, &ret); + return (__u32)ret.lc_count; +#else + return 0; +#endif } -EXPORT_SYMBOL(fid_pack); -int fid_unpack(const struct lu_fid_pack *pack, struct lu_fid *fid) +/** + * Output site statistical counters into a buffer. Suitable for + * lprocfs_rd_*()-style functions. + */ +int lu_site_stats_print(const struct lu_site *s, char *page, int count) { - 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; - } - return result; + lu_site_stats_t stats; + + memset(&stats, 0, sizeof(stats)); + lu_site_stats_get(s->ls_obj_hash, &stats, 1); + + return snprintf(page, count, "%d/%d %d/%d %d %d %d %d %d %d %d\n", + stats.lss_busy, + stats.lss_total, + stats.lss_populated, + CFS_HASH_NHLIST(s->ls_obj_hash), + stats.lss_max_search, + ls_stats_read(s->ls_stats, LU_SS_CREATED), + ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT), + ls_stats_read(s->ls_stats, LU_SS_CACHE_MISS), + ls_stats_read(s->ls_stats, LU_SS_CACHE_RACE), + ls_stats_read(s->ls_stats, LU_SS_CACHE_DEATH_RACE), + ls_stats_read(s->ls_stats, LU_SS_LRU_PURGED)); } -EXPORT_SYMBOL(fid_unpack); +EXPORT_SYMBOL(lu_site_stats_print); const char *lu_time_names[LU_TIME_NR] = { [LU_TIME_FIND_LOOKUP] = "find_lookup", @@ -1468,3 +1869,42 @@ const char *lu_time_names[LU_TIME_NR] = { [LU_TIME_FIND_INSERT] = "find_insert" }; EXPORT_SYMBOL(lu_time_names); + +/** + * Helper function to initialize a number of kmem slab caches at once. + */ +int lu_kmem_init(struct lu_kmem_descr *caches) +{ + int result; + + 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(lu_kmem_init); + +/** + * 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);