X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fgenops.c;h=5f679de82da536dea02def23813c220c0c917f66;hp=f20f3b1dc886a0dd1f0db9e434c9c4d9f9574463;hb=135fea8fa986d7abf107953b5b9a57170a418eda;hpb=1f1d3a376d488d715dd1b0c94d5b66ea05c1e6ca diff --git a/lustre/obdclass/genops.c b/lustre/obdclass/genops.c index f20f3b1..5f679de 100644 --- a/lustre/obdclass/genops.c +++ b/lustre/obdclass/genops.c @@ -15,11 +15,7 @@ * * 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 - * - * 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. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -40,24 +36,36 @@ */ #define DEBUG_SUBSYSTEM S_CLASS + +#include +#include +#include #include +#include #include +#include +#include -spinlock_t obd_types_lock; +static DEFINE_SPINLOCK(obd_types_lock); +static LIST_HEAD(obd_types); +DEFINE_RWLOCK(obd_dev_lock); +static struct obd_device *obd_devs[MAX_OBD_DEVICES]; -struct kmem_cache *obd_device_cachep; +static struct kmem_cache *obd_device_cachep; struct kmem_cache *obdo_cachep; EXPORT_SYMBOL(obdo_cachep); -struct kmem_cache *import_cachep; +static struct kmem_cache *import_cachep; + +static struct workqueue_struct *zombie_wq; -struct list_head obd_zombie_imports; -struct list_head obd_zombie_exports; -spinlock_t obd_zombie_impexp_lock; -static void obd_zombie_impexp_notify(void); static void obd_zombie_export_add(struct obd_export *exp); static void obd_zombie_import_add(struct obd_import *imp); static void print_export_data(struct obd_export *exp, - const char *status, int locks); + const char *status, int locks, int debug_level); + +static LIST_HEAD(obd_stale_exports); +static DEFINE_SPINLOCK(obd_stale_export_lock); +static atomic_t obd_stale_export_num = ATOMIC_INIT(0); int (*ptlrpc_put_connection_superhack)(struct ptlrpc_connection *c); EXPORT_SYMBOL(ptlrpc_put_connection_superhack); @@ -143,7 +151,6 @@ struct obd_type *class_get_type(const char *name) } return type; } -EXPORT_SYMBOL(class_get_type); void class_put_type(struct obd_type *type) { @@ -153,23 +160,57 @@ void class_put_type(struct obd_type *type) module_put(type->typ_dt_ops->o_owner); spin_unlock(&type->obd_type_lock); } -EXPORT_SYMBOL(class_put_type); + +static void class_sysfs_release(struct kobject *kobj) +{ + OBD_FREE(kobj, sizeof(*kobj)); +} + +static struct kobj_type class_ktype = { + .sysfs_ops = &lustre_sysfs_ops, + .release = class_sysfs_release, +}; + +struct kobject *class_setup_tunables(const char *name) +{ + struct kobject *kobj; + int rc; + +#ifdef HAVE_SERVER_SUPPORT + kobj = kset_find_obj(lustre_kset, name); + if (kobj) + return kobj; +#endif + OBD_ALLOC(kobj, sizeof(*kobj)); + if (!kobj) + return ERR_PTR(-ENOMEM); + + kobj->kset = lustre_kset; + kobject_init(kobj, &class_ktype); + rc = kobject_add(kobj, &lustre_kset->kobj, "%s", name); + if (rc) { + kobject_put(kobj); + return ERR_PTR(rc); + } + return kobj; +} +EXPORT_SYMBOL(class_setup_tunables); #define CLASS_MAX_NAME 1024 int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops, - bool enable_proc, struct lprocfs_seq_vars *module_vars, -#ifndef HAVE_ONLY_PROCFS_SEQ - struct lprocfs_vars *vars, -#endif + bool enable_proc, struct lprocfs_vars *vars, const char *name, struct lu_device_type *ldt) { - struct obd_type *type; - int rc = 0; - ENTRY; + struct obd_type *type; +#ifdef HAVE_SERVER_SUPPORT + struct qstr dname; +#endif /* HAVE_SERVER_SUPPORT */ + int rc = 0; - /* sanity check */ - LASSERT(strnlen(name, CLASS_MAX_NAME) < CLASS_MAX_NAME); + ENTRY; + /* sanity check */ + LASSERT(strnlen(name, CLASS_MAX_NAME) < CLASS_MAX_NAME); if (class_search_type(name)) { CDEBUG(D_IOCTL, "Type %s already registered\n", name); @@ -197,20 +238,11 @@ int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops, strcpy(type->typ_name, name); spin_lock_init(&type->obd_type_lock); -#ifdef LPROCFS +#ifdef CONFIG_PROC_FS if (enable_proc) { -#ifndef HAVE_ONLY_PROCFS_SEQ - if (vars) { - type->typ_procroot = lprocfs_register(type->typ_name, - proc_lustre_root, - vars, type); - } else -#endif - { - type->typ_procroot = lprocfs_seq_register(type->typ_name, - proc_lustre_root, - module_vars, type); - } + type->typ_procroot = lprocfs_register(type->typ_name, + proc_lustre_root, + vars, type); if (IS_ERR(type->typ_procroot)) { rc = PTR_ERR(type->typ_procroot); type->typ_procroot = NULL; @@ -218,31 +250,62 @@ int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops, } } #endif - if (ldt != NULL) { - type->typ_lu = ldt; - rc = lu_device_type_init(ldt); - if (rc != 0) - GOTO (failed, rc); - } +#ifdef HAVE_SERVER_SUPPORT + dname.name = name; + dname.len = strlen(dname.name); + dname.hash = ll_full_name_hash(debugfs_lustre_root, dname.name, + dname.len); + type->typ_debugfs_entry = d_lookup(debugfs_lustre_root, &dname); + if (type->typ_debugfs_entry) { + dput(type->typ_debugfs_entry); + type->typ_sym_filter = true; + goto dir_exist; + } +#endif /* HAVE_SERVER_SUPPORT */ + + type->typ_debugfs_entry = ldebugfs_register(type->typ_name, + debugfs_lustre_root, + NULL, type); + if (IS_ERR_OR_NULL(type->typ_debugfs_entry)) { + rc = type->typ_debugfs_entry ? PTR_ERR(type->typ_debugfs_entry) + : -ENOMEM; + type->typ_debugfs_entry = NULL; + GOTO(failed, rc); + } +#ifdef HAVE_SERVER_SUPPORT +dir_exist: +#endif + type->typ_kobj = class_setup_tunables(type->typ_name); + if (IS_ERR(type->typ_kobj)) + GOTO(failed, rc = PTR_ERR(type->typ_kobj)); + + if (ldt) { + type->typ_lu = ldt; + rc = lu_device_type_init(ldt); + if (rc) { + kobject_put(type->typ_kobj); + GOTO(failed, rc); + } + } spin_lock(&obd_types_lock); list_add(&type->typ_chain, &obd_types); spin_unlock(&obd_types_lock); - RETURN (0); + RETURN(0); failed: +#ifdef HAVE_SERVER_SUPPORT + if (type->typ_sym_filter) + type->typ_debugfs_entry = NULL; +#endif + if (!IS_ERR_OR_NULL(type->typ_debugfs_entry)) + ldebugfs_remove(&type->typ_debugfs_entry); if (type->typ_name != NULL) { -#ifdef LPROCFS - if (type->typ_procroot != NULL) { -#ifndef HAVE_ONLY_PROCFS_SEQ - lprocfs_try_remove_proc_entry(type->typ_name, - proc_lustre_root); -#else +#ifdef CONFIG_PROC_FS + if (type->typ_procroot != NULL) remove_proc_subtree(type->typ_name, proc_lustre_root); #endif - } -#endif OBD_FREE(type->typ_name, strlen(name) + 1); } if (type->typ_md_ops != NULL) @@ -273,22 +336,25 @@ int class_unregister_type(const char *name) RETURN(-EBUSY); } + kobject_put(type->typ_kobj); + /* we do not use type->typ_procroot as for compatibility purposes * other modules can share names (i.e. lod can use lov entry). so * we can't reference pointer as it can get invalided when another * module removes the entry */ -#ifdef LPROCFS - if (type->typ_procroot != NULL) { -#ifndef HAVE_ONLY_PROCFS_SEQ - lprocfs_try_remove_proc_entry(type->typ_name, proc_lustre_root); -#else +#ifdef CONFIG_PROC_FS + if (type->typ_procroot != NULL) remove_proc_subtree(type->typ_name, proc_lustre_root); -#endif - } - if (type->typ_procsym != NULL) lprocfs_remove(&type->typ_procsym); #endif +#ifdef HAVE_SERVER_SUPPORT + if (type->typ_sym_filter) + type->typ_debugfs_entry = NULL; +#endif + if (!IS_ERR_OR_NULL(type->typ_debugfs_entry)) + ldebugfs_remove(&type->typ_debugfs_entry); + if (type->typ_lu) lu_device_type_fini(type->typ_lu); @@ -308,21 +374,20 @@ EXPORT_SYMBOL(class_unregister_type); /** * Create a new obd device. * - * Find an empty slot in ::obd_devs[], create a new obd device in it. + * Allocate the new obd_device and initialize it. * * \param[in] type_name obd device type string. * \param[in] name obd device name. + * \param[in] uuid obd device UUID * - * \retval NULL if create fails, otherwise return the obd device - * pointer created. + * \retval newdev pointer to created obd_device + * \retval ERR_PTR(errno) on error */ -struct obd_device *class_newdev(const char *type_name, const char *name) +struct obd_device *class_newdev(const char *type_name, const char *name, + const char *uuid) { - struct obd_device *result = NULL; struct obd_device *newdev; struct obd_type *type = NULL; - int i; - int new_obd_minor = 0; ENTRY; if (strlen(name) >= MAX_OBD_NAME) { @@ -337,93 +402,198 @@ struct obd_device *class_newdev(const char *type_name, const char *name) } newdev = obd_device_alloc(); - if (newdev == NULL) - GOTO(out_type, result = ERR_PTR(-ENOMEM)); - + if (newdev == NULL) { + class_put_type(type); + RETURN(ERR_PTR(-ENOMEM)); + } LASSERT(newdev->obd_magic == OBD_DEVICE_MAGIC); + strncpy(newdev->obd_name, name, sizeof(newdev->obd_name) - 1); + newdev->obd_type = type; + newdev->obd_minor = -1; + + rwlock_init(&newdev->obd_pool_lock); + newdev->obd_pool_limit = 0; + newdev->obd_pool_slv = 0; + + INIT_LIST_HEAD(&newdev->obd_exports); + INIT_LIST_HEAD(&newdev->obd_unlinked_exports); + INIT_LIST_HEAD(&newdev->obd_delayed_exports); + INIT_LIST_HEAD(&newdev->obd_exports_timed); + INIT_LIST_HEAD(&newdev->obd_nid_stats); + spin_lock_init(&newdev->obd_nid_lock); + spin_lock_init(&newdev->obd_dev_lock); + mutex_init(&newdev->obd_dev_mutex); + spin_lock_init(&newdev->obd_osfs_lock); + /* newdev->obd_osfs_age must be set to a value in the distant + * past to guarantee a fresh statfs is fetched on mount. */ + newdev->obd_osfs_age = ktime_get_seconds() - 1000; + + /* XXX belongs in setup not attach */ + init_rwsem(&newdev->obd_observer_link_sem); + /* recovery data */ + init_timer(&newdev->obd_recovery_timer); + spin_lock_init(&newdev->obd_recovery_task_lock); + init_waitqueue_head(&newdev->obd_next_transno_waitq); + init_waitqueue_head(&newdev->obd_evict_inprogress_waitq); + INIT_LIST_HEAD(&newdev->obd_req_replay_queue); + INIT_LIST_HEAD(&newdev->obd_lock_replay_queue); + INIT_LIST_HEAD(&newdev->obd_final_req_queue); + INIT_LIST_HEAD(&newdev->obd_evict_list); + INIT_LIST_HEAD(&newdev->obd_lwp_list); + + llog_group_init(&newdev->obd_olg); + /* Detach drops this */ + atomic_set(&newdev->obd_refcount, 1); + lu_ref_init(&newdev->obd_reference); + lu_ref_add(&newdev->obd_reference, "newdev", newdev); + + newdev->obd_conn_inprogress = 0; + + strncpy(newdev->obd_uuid.uuid, uuid, strlen(uuid)); + + CDEBUG(D_IOCTL, "Allocate new device %s (%p)\n", + newdev->obd_name, newdev); + + return newdev; +} - write_lock(&obd_dev_lock); - for (i = 0; i < class_devno_max(); i++) { - struct obd_device *obd = class_num2obd(i); +/** + * Free obd device. + * + * \param[in] obd obd_device to be freed + * + * \retval none + */ +void class_free_dev(struct obd_device *obd) +{ + struct obd_type *obd_type = obd->obd_type; - if (obd && (strcmp(name, obd->obd_name) == 0)) { - CERROR("Device %s already exists at %d, won't add\n", - name, i); - if (result) { - LASSERTF(result->obd_magic == OBD_DEVICE_MAGIC, - "%p obd_magic %08x != %08x\n", result, - result->obd_magic, OBD_DEVICE_MAGIC); - LASSERTF(result->obd_minor == new_obd_minor, - "%p obd_minor %d != %d\n", result, - result->obd_minor, new_obd_minor); - - obd_devs[result->obd_minor] = NULL; - result->obd_name[0]='\0'; - } - result = ERR_PTR(-EEXIST); - break; - } - if (!result && !obd) { - result = newdev; - result->obd_minor = i; - new_obd_minor = i; - result->obd_type = type; - strncpy(result->obd_name, name, - sizeof(result->obd_name) - 1); - obd_devs[i] = result; - } - } - write_unlock(&obd_dev_lock); + LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC, "%p obd_magic %08x " + "!= %08x\n", obd, obd->obd_magic, OBD_DEVICE_MAGIC); + LASSERTF(obd->obd_minor == -1 || obd_devs[obd->obd_minor] == obd, + "obd %p != obd_devs[%d] %p\n", + obd, obd->obd_minor, obd_devs[obd->obd_minor]); + LASSERTF(atomic_read(&obd->obd_refcount) == 0, + "obd_refcount should be 0, not %d\n", + atomic_read(&obd->obd_refcount)); + LASSERT(obd_type != NULL); - if (result == NULL && i >= class_devno_max()) { - CERROR("all %u OBD devices used, increase MAX_OBD_DEVICES\n", - class_devno_max()); - GOTO(out, result = ERR_PTR(-EOVERFLOW)); - } + CDEBUG(D_INFO, "Release obd device %s obd_type name = %s\n", + obd->obd_name, obd->obd_type->typ_name); - if (IS_ERR(result)) - GOTO(out, result); + CDEBUG(D_CONFIG, "finishing cleanup of obd %s (%s)\n", + obd->obd_name, obd->obd_uuid.uuid); + if (obd->obd_stopping) { + int err; - CDEBUG(D_IOCTL, "Adding new device %s (%p)\n", - result->obd_name, result); + /* If we're not stopping, we were never set up */ + err = obd_cleanup(obd); + if (err) + CERROR("Cleanup %s returned %d\n", + obd->obd_name, err); + } - RETURN(result); -out: - obd_device_free(newdev); -out_type: - class_put_type(type); - return result; + obd_device_free(obd); + + class_put_type(obd_type); } -void class_release_dev(struct obd_device *obd) +/** + * Unregister obd device. + * + * Free slot in obd_dev[] used by \a obd. + * + * \param[in] new_obd obd_device to be unregistered + * + * \retval none + */ +void class_unregister_device(struct obd_device *obd) { - struct obd_type *obd_type = obd->obd_type; - - LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC, "%p obd_magic %08x != %08x\n", - obd, obd->obd_magic, OBD_DEVICE_MAGIC); - LASSERTF(obd == obd_devs[obd->obd_minor], "obd %p != obd_devs[%d] %p\n", - obd, obd->obd_minor, obd_devs[obd->obd_minor]); - LASSERT(obd_type != NULL); + write_lock(&obd_dev_lock); + if (obd->obd_minor >= 0) { + LASSERT(obd_devs[obd->obd_minor] == obd); + obd_devs[obd->obd_minor] = NULL; + obd->obd_minor = -1; + } + write_unlock(&obd_dev_lock); +} - CDEBUG(D_INFO, "Release obd device %s at %d obd_type name =%s\n", - obd->obd_name, obd->obd_minor, obd->obd_type->typ_name); +/** + * Register obd device. + * + * Find free slot in obd_devs[], fills it with \a new_obd. + * + * \param[in] new_obd obd_device to be registered + * + * \retval 0 success + * \retval -EEXIST device with this name is registered + * \retval -EOVERFLOW obd_devs[] is full + */ +int class_register_device(struct obd_device *new_obd) +{ + int ret = 0; + int i; + int new_obd_minor = 0; + bool minor_assign = false; + bool retried = false; +again: write_lock(&obd_dev_lock); - obd_devs[obd->obd_minor] = NULL; + for (i = 0; i < class_devno_max(); i++) { + struct obd_device *obd = class_num2obd(i); + + if (obd != NULL && + (strcmp(new_obd->obd_name, obd->obd_name) == 0)) { + + if (!retried) { + write_unlock(&obd_dev_lock); + + /* the obd_device could be waited to be + * destroyed by the "obd_zombie_impexp_thread". + */ + obd_zombie_barrier(); + retried = true; + goto again; + } + + CERROR("%s: already exists, won't add\n", + obd->obd_name); + /* in case we found a free slot before duplicate */ + minor_assign = false; + ret = -EEXIST; + break; + } + if (!minor_assign && obd == NULL) { + new_obd_minor = i; + minor_assign = true; + } + } + + if (minor_assign) { + new_obd->obd_minor = new_obd_minor; + LASSERTF(obd_devs[new_obd_minor] == NULL, "obd_devs[%d] " + "%p\n", new_obd_minor, obd_devs[new_obd_minor]); + obd_devs[new_obd_minor] = new_obd; + } else { + if (ret == 0) { + ret = -EOVERFLOW; + CERROR("%s: all %u/%u devices used, increase " + "MAX_OBD_DEVICES: rc = %d\n", new_obd->obd_name, + i, class_devno_max(), ret); + } + } write_unlock(&obd_dev_lock); - obd_device_free(obd); - class_put_type(obd_type); + RETURN(ret); } -int class_name2dev(const char *name) +static int class_name2dev_nolock(const char *name) { int i; if (!name) return -1; - read_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); @@ -432,16 +602,28 @@ int class_name2dev(const char *name) out any references */ LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC); if (obd->obd_attached) { - read_unlock(&obd_dev_lock); return i; } break; } } - read_unlock(&obd_dev_lock); return -1; } + +int class_name2dev(const char *name) +{ + int i; + + if (!name) + return -1; + + read_lock(&obd_dev_lock); + i = class_name2dev_nolock(name); + read_unlock(&obd_dev_lock); + + return i; +} EXPORT_SYMBOL(class_name2dev); struct obd_device *class_name2obd(const char *name) @@ -454,24 +636,32 @@ struct obd_device *class_name2obd(const char *name) } EXPORT_SYMBOL(class_name2obd); -int class_uuid2dev(struct obd_uuid *uuid) +int class_uuid2dev_nolock(struct obd_uuid *uuid) { int i; - read_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); if (obd && obd_uuid_equals(uuid, &obd->obd_uuid)) { LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC); - read_unlock(&obd_dev_lock); return i; } } - read_unlock(&obd_dev_lock); return -1; } + +int class_uuid2dev(struct obd_uuid *uuid) +{ + int i; + + read_lock(&obd_dev_lock); + i = class_uuid2dev_nolock(uuid); + read_unlock(&obd_dev_lock); + + return i; +} EXPORT_SYMBOL(class_uuid2dev); struct obd_device *class_uuid2obd(struct obd_uuid *uuid) @@ -510,7 +700,40 @@ struct obd_device *class_num2obd(int num) return obd; } -EXPORT_SYMBOL(class_num2obd); + +/** + * Find obd in obd_dev[] by name or uuid. + * + * Increment obd's refcount if found. + * + * \param[in] str obd name or uuid + * + * \retval NULL if not found + * \retval target pointer to found obd_device + */ +struct obd_device *class_dev_by_str(const char *str) +{ + struct obd_device *target = NULL; + struct obd_uuid tgtuuid; + int rc; + + obd_str2uuid(&tgtuuid, str); + + read_lock(&obd_dev_lock); + rc = class_uuid2dev_nolock(&tgtuuid); + if (rc < 0) + rc = class_name2dev_nolock(str); + + if (rc >= 0) + target = class_num2obd(rc); + + if (target != NULL) + class_incref(target, "find", current); + read_unlock(&obd_dev_lock); + + RETURN(target); +} +EXPORT_SYMBOL(class_dev_by_str); /** * Get obd devices count. Device in any @@ -640,19 +863,22 @@ int class_notify_sptlrpc_conf(const char *fsname, int namelen) LASSERT(namelen > 0); read_lock(&obd_dev_lock); - for (i = 0; i < class_devno_max(); i++) { - obd = class_num2obd(i); + for (i = 0; i < class_devno_max(); i++) { + obd = class_num2obd(i); - if (obd == NULL || obd->obd_set_up == 0 || obd->obd_stopping) - continue; + if (obd == NULL || obd->obd_set_up == 0 || obd->obd_stopping) + continue; - /* only notify mdc, osc, mdt, ost */ - type = obd->obd_type->typ_name; - if (strcmp(type, LUSTRE_MDC_NAME) != 0 && - strcmp(type, LUSTRE_OSC_NAME) != 0 && - strcmp(type, LUSTRE_MDT_NAME) != 0 && - strcmp(type, LUSTRE_OST_NAME) != 0) - continue; + /* only notify mdc, osc, osp, lwp, mdt, ost + * because only these have a -sptlrpc llog */ + type = obd->obd_type->typ_name; + if (strcmp(type, LUSTRE_MDC_NAME) != 0 && + strcmp(type, LUSTRE_OSC_NAME) != 0 && + strcmp(type, LUSTRE_OSP_NAME) != 0 && + strcmp(type, LUSTRE_LWP_NAME) != 0 && + strcmp(type, LUSTRE_MDT_NAME) != 0 && + strcmp(type, LUSTRE_OST_NAME) != 0) + continue; if (strncmp(obd->obd_name, fsname, namelen)) continue; @@ -686,10 +912,7 @@ void obd_cleanup_caches(void) kmem_cache_destroy(import_cachep); import_cachep = NULL; } - if (capa_cachep) { - kmem_cache_destroy(capa_cachep); - capa_cachep = NULL; - } + EXIT; } @@ -718,12 +941,6 @@ int obd_init_caches(void) if (!import_cachep) GOTO(out, rc = -ENOMEM); - LASSERT(capa_cachep == NULL); - capa_cachep = kmem_cache_create("capa_cache", sizeof(struct obd_capa), - 0, 0, NULL); - if (!capa_cachep) - GOTO(out, rc = -ENOMEM); - RETURN(0); out: obd_cleanup_caches(); @@ -746,7 +963,7 @@ struct obd_export *class_conn2export(struct lustre_handle *conn) RETURN(NULL); } - CDEBUG(D_INFO, "looking for export cookie "LPX64"\n", conn->cookie); + CDEBUG(D_INFO, "looking for export cookie %#llx\n", conn->cookie); export = class_handle2object(conn->cookie, NULL); RETURN(export); } @@ -771,7 +988,6 @@ struct obd_device *class_conn2obd(struct lustre_handle *conn) } return NULL; } -EXPORT_SYMBOL(class_conn2obd); struct obd_import *class_exp2cliimp(struct obd_export *exp) { @@ -789,7 +1005,6 @@ struct obd_import *class_conn2cliimp(struct lustre_handle *conn) return NULL; return obd->u.cli.cl_import; } -EXPORT_SYMBOL(class_conn2cliimp); /* Export management functions */ static void class_export_destroy(struct obd_export *exp) @@ -812,7 +1027,10 @@ static void class_export_destroy(struct obd_export *exp) LASSERT(list_empty(&exp->exp_req_replay_queue)); LASSERT(list_empty(&exp->exp_hp_rpcs)); obd_destroy_export(exp); - class_decref(obd, "export", exp); + /* self export doesn't hold a reference to an obd, although it + * exists until freeing of the obd */ + if (exp != obd->obd_self_export) + class_decref(obd, "export", exp); OBD_FREE_RCU(exp, sizeof(*exp), &exp->exp_handle); EXIT; @@ -845,26 +1063,49 @@ void class_export_put(struct obd_export *exp) atomic_read(&exp->exp_refcount) - 1); if (atomic_dec_and_test(&exp->exp_refcount)) { - LASSERT(!list_empty(&exp->exp_obd_chain)); - CDEBUG(D_IOCTL, "final put %p/%s\n", - exp, exp->exp_client_uuid.uuid); - - /* release nid stat refererence */ - lprocfs_exp_cleanup(exp); + struct obd_device *obd = exp->exp_obd; + + CDEBUG(D_IOCTL, "final put %p/%s\n", + exp, exp->exp_client_uuid.uuid); + + /* release nid stat refererence */ + lprocfs_exp_cleanup(exp); + + if (exp == obd->obd_self_export) { + /* self export should be destroyed without + * zombie thread as it doesn't hold a + * reference to obd and doesn't hold any + * resources */ + class_export_destroy(exp); + /* self export is destroyed, no class + * references exist and it is safe to free + * obd */ + class_free_dev(obd); + } else { + LASSERT(!list_empty(&exp->exp_obd_chain)); + obd_zombie_export_add(exp); + } - obd_zombie_export_add(exp); - } + } } EXPORT_SYMBOL(class_export_put); +static void obd_zombie_exp_cull(struct work_struct *ws) +{ + struct obd_export *export; + + export = container_of(ws, struct obd_export, exp_zombie_work); + class_export_destroy(export); +} + /* Creates a new export, adds it to the hash table, and returns a * pointer to it. The refcount is 2: one for the hash reference, and * one for the pointer returned by this function. */ -struct obd_export *class_new_export(struct obd_device *obd, - struct obd_uuid *cluuid) +struct obd_export *__class_new_export(struct obd_device *obd, + struct obd_uuid *cluuid, bool is_self) { struct obd_export *export; - cfs_hash_t *hash = NULL; + struct cfs_hash *hash = NULL; int rc = 0; ENTRY; @@ -875,6 +1116,7 @@ struct obd_export *class_new_export(struct obd_device *obd, export->exp_conn_cnt = 0; export->exp_lock_hash = NULL; export->exp_flock_hash = NULL; + /* 2 = class_handle_hash + last */ atomic_set(&export->exp_refcount, 2); atomic_set(&export->exp_rpc_count, 0); atomic_set(&export->exp_cb_count, 0); @@ -893,30 +1135,33 @@ struct obd_export *class_new_export(struct obd_device *obd, INIT_LIST_HEAD(&export->exp_hp_rpcs); INIT_LIST_HEAD(&export->exp_reg_rpcs); class_handle_hash(&export->exp_handle, &export_handle_ops); - export->exp_last_request_time = cfs_time_current_sec(); + export->exp_last_request_time = ktime_get_real_seconds(); spin_lock_init(&export->exp_lock); spin_lock_init(&export->exp_rpc_lock); INIT_HLIST_NODE(&export->exp_uuid_hash); INIT_HLIST_NODE(&export->exp_nid_hash); + INIT_HLIST_NODE(&export->exp_gen_hash); spin_lock_init(&export->exp_bl_list_lock); INIT_LIST_HEAD(&export->exp_bl_list); + INIT_LIST_HEAD(&export->exp_stale_list); + INIT_WORK(&export->exp_zombie_work, obd_zombie_exp_cull); export->exp_sp_peer = LUSTRE_SP_ANY; export->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID; export->exp_client_uuid = *cluuid; obd_init_export(export); - spin_lock(&obd->obd_dev_lock); - /* shouldn't happen, but might race */ - if (obd->obd_stopping) - GOTO(exit_unlock, rc = -ENODEV); + if (!obd_uuid_equals(cluuid, &obd->obd_uuid)) { + spin_lock(&obd->obd_dev_lock); + /* shouldn't happen, but might race */ + if (obd->obd_stopping) + GOTO(exit_unlock, rc = -ENODEV); - hash = cfs_hash_getref(obd->obd_uuid_hash); - if (hash == NULL) - GOTO(exit_unlock, rc = -ENODEV); - spin_unlock(&obd->obd_dev_lock); + hash = cfs_hash_getref(obd->obd_uuid_hash); + if (hash == NULL) + GOTO(exit_unlock, rc = -ENODEV); + spin_unlock(&obd->obd_dev_lock); - if (!obd_uuid_equals(cluuid, &obd->obd_uuid)) { rc = cfs_hash_add_unique(hash, cluuid, &export->exp_uuid_hash); if (rc != 0) { LCONSOLE_WARN("%s: denying duplicate export for %s, %d\n", @@ -925,19 +1170,27 @@ struct obd_export *class_new_export(struct obd_device *obd, } } + at_init(&export->exp_bl_lock_at, obd_timeout, 0); spin_lock(&obd->obd_dev_lock); if (obd->obd_stopping) { - cfs_hash_del(hash, cluuid, &export->exp_uuid_hash); - GOTO(exit_unlock, rc = -ENODEV); + if (hash) + cfs_hash_del(hash, cluuid, &export->exp_uuid_hash); + GOTO(exit_unlock, rc = -ESHUTDOWN); } - class_incref(obd, "export", export); - list_add(&export->exp_obd_chain, &export->exp_obd->obd_exports); - list_add_tail(&export->exp_obd_chain_timed, - &export->exp_obd->obd_exports_timed); - export->exp_obd->obd_num_exports++; + if (!is_self) { + class_incref(obd, "export", export); + list_add_tail(&export->exp_obd_chain_timed, + &obd->obd_exports_timed); + list_add(&export->exp_obd_chain, &obd->obd_exports); + obd->obd_num_exports++; + } else { + INIT_LIST_HEAD(&export->exp_obd_chain_timed); + INIT_LIST_HEAD(&export->exp_obd_chain); + } spin_unlock(&obd->obd_dev_lock); - cfs_hash_putref(hash); + if (hash) + cfs_hash_putref(hash); RETURN(export); exit_unlock: @@ -951,12 +1204,29 @@ exit_err: OBD_FREE_PTR(export); return ERR_PTR(rc); } + +struct obd_export *class_new_export(struct obd_device *obd, + struct obd_uuid *uuid) +{ + return __class_new_export(obd, uuid, false); +} EXPORT_SYMBOL(class_new_export); +struct obd_export *class_new_export_self(struct obd_device *obd, + struct obd_uuid *uuid) +{ + return __class_new_export(obd, uuid, true); +} + void class_unlink_export(struct obd_export *exp) { class_handle_unhash(&exp->exp_handle); + if (exp->exp_obd->obd_self_export == exp) { + class_export_put(exp); + return; + } + spin_lock(&exp->exp_obd->obd_dev_lock); /* delete an uuid-export hashitem from hashtables */ if (!hlist_unhashed(&exp->exp_uuid_hash)) @@ -964,16 +1234,34 @@ void class_unlink_export(struct obd_export *exp) &exp->exp_client_uuid, &exp->exp_uuid_hash); +#ifdef HAVE_SERVER_SUPPORT + if (!hlist_unhashed(&exp->exp_gen_hash)) { + struct tg_export_data *ted = &exp->exp_target_data; + struct cfs_hash *hash; + + /* Because obd_gen_hash will not be released until + * class_cleanup(), so hash should never be NULL here */ + hash = cfs_hash_getref(exp->exp_obd->obd_gen_hash); + LASSERT(hash != NULL); + cfs_hash_del(hash, &ted->ted_lcd->lcd_generation, + &exp->exp_gen_hash); + cfs_hash_putref(hash); + } +#endif /* HAVE_SERVER_SUPPORT */ + list_move(&exp->exp_obd_chain, &exp->exp_obd->obd_unlinked_exports); list_del_init(&exp->exp_obd_chain_timed); exp->exp_obd->obd_num_exports--; spin_unlock(&exp->exp_obd->obd_dev_lock); - class_export_put(exp); + atomic_inc(&obd_stale_export_num); + + /* A reference is kept by obd_stale_exports list */ + obd_stale_export_put(exp); } EXPORT_SYMBOL(class_unlink_export); /* Import management functions */ -void class_import_destroy(struct obd_import *imp) +static void class_import_destroy(struct obd_import *imp) { ENTRY; @@ -1024,7 +1312,6 @@ void class_import_put(struct obd_import *imp) { ENTRY; - LASSERT(list_empty(&imp->imp_zombie_chain)); LASSERT_ATOMIC_GT_LT(&imp->imp_refcount, 0, LI_POISON); CDEBUG(D_INFO, "import %p refcount=%d obd=%s\n", imp, @@ -1054,20 +1341,30 @@ static void init_imp_at(struct imp_at *at) { } } +static void obd_zombie_imp_cull(struct work_struct *ws) +{ + struct obd_import *import; + + import = container_of(ws, struct obd_import, imp_zombie_work); + class_import_destroy(import); +} + struct obd_import *class_new_import(struct obd_device *obd) { struct obd_import *imp; + struct pid_namespace *curr_pid_ns = ll_task_pid_ns(current); OBD_ALLOC(imp, sizeof(*imp)); if (imp == NULL) return NULL; INIT_LIST_HEAD(&imp->imp_pinger_chain); - INIT_LIST_HEAD(&imp->imp_zombie_chain); INIT_LIST_HEAD(&imp->imp_replay_list); INIT_LIST_HEAD(&imp->imp_sending_list); INIT_LIST_HEAD(&imp->imp_delayed_list); INIT_LIST_HEAD(&imp->imp_committed_list); + INIT_LIST_HEAD(&imp->imp_unreplied_list); + imp->imp_known_replied_xid = 0; imp->imp_replay_cursor = &imp->imp_committed_list; spin_lock_init(&imp->imp_lock); imp->imp_last_success_conn = 0; @@ -1075,6 +1372,12 @@ struct obd_import *class_new_import(struct obd_device *obd) imp->imp_obd = class_incref(obd, "import", imp); mutex_init(&imp->imp_sec_mutex); init_waitqueue_head(&imp->imp_recovery_waitq); + INIT_WORK(&imp->imp_zombie_work, obd_zombie_imp_cull); + + if (curr_pid_ns->child_reaper) + imp->imp_sec_refpid = curr_pid_ns->child_reaper->pid; + else + imp->imp_sec_refpid = 1; atomic_set(&imp->imp_refcount, 2); atomic_set(&imp->imp_unregistering, 0); @@ -1171,14 +1474,14 @@ int class_connect(struct lustre_handle *conn, struct obd_device *obd, conn->cookie = export->exp_handle.h_cookie; class_export_put(export); - CDEBUG(D_IOCTL, "connect: client %s, cookie "LPX64"\n", + CDEBUG(D_IOCTL, "connect: client %s, cookie %#llx\n", cluuid->uuid, conn->cookie); RETURN(0); } EXPORT_SYMBOL(class_connect); /* if export is involved in recovery then clean up related things */ -void class_export_recovery_cleanup(struct obd_export *exp) +static void class_export_recovery_cleanup(struct obd_export *exp) { struct obd_device *obd = exp->exp_obd; @@ -1238,6 +1541,13 @@ int class_disconnect(struct obd_export *export) spin_lock(&export->exp_lock); already_disconnected = export->exp_disconnected; export->exp_disconnected = 1; + /* We hold references of export for uuid hash + * and nid_hash and export link at least. So + * it is safe to call cfs_hash_del in there. */ + if (!hlist_unhashed(&export->exp_nid_hash)) + cfs_hash_del(export->exp_obd->obd_nid_hash, + &export->exp_connection->c_peer.nid, + &export->exp_nid_hash); spin_unlock(&export->exp_lock); /* class_cleanup(), abort_recovery(), and class_fail_export() @@ -1248,14 +1558,9 @@ int class_disconnect(struct obd_export *export) GOTO(no_disconn, already_disconnected); } - CDEBUG(D_IOCTL, "disconnect: cookie "LPX64"\n", + CDEBUG(D_IOCTL, "disconnect: cookie %#llx\n", export->exp_handle.h_cookie); - if (!hlist_unhashed(&export->exp_nid_hash)) - cfs_hash_del(export->exp_obd->obd_nid_hash, - &export->exp_connection->c_peer.nid, - &export->exp_nid_hash); - class_export_recovery_cleanup(export); class_unlink_export(export); no_disconn: @@ -1311,7 +1616,7 @@ static void class_disconnect_export_list(struct list_head *list, class_export_get(exp); CDEBUG(D_HA, "%s: disconnecting export at %s (%p), " - "last request at "CFS_TIME_T"\n", + "last request at %lld\n", exp->exp_obd->obd_name, obd_export_nid2str(exp), exp, exp->exp_last_request_time); /* release one export reference anyway */ @@ -1381,13 +1686,12 @@ void class_disconnect_stale_exports(struct obd_device *obd, spin_unlock(&exp->exp_lock); list_move(&exp->exp_obd_chain, &work_list); - evicted++; - CDEBUG(D_HA, "%s: disconnect stale client %s@%s\n", - obd->obd_name, exp->exp_client_uuid.uuid, - exp->exp_connection == NULL ? "" : - libcfs_nid2str(exp->exp_connection->c_peer.nid)); - print_export_data(exp, "EVICTING", 0); - } + evicted++; + CDEBUG(D_HA, "%s: disconnect stale client %s@%s\n", + obd->obd_name, exp->exp_client_uuid.uuid, + obd_export_nid2str(exp)); + print_export_data(exp, "EVICTING", 0, D_HA); + } spin_unlock(&obd->obd_dev_lock); if (evicted) @@ -1438,18 +1742,9 @@ void class_fail_export(struct obd_export *exp) } EXPORT_SYMBOL(class_fail_export); -char *obd_export_nid2str(struct obd_export *exp) -{ - if (exp->exp_connection != NULL) - return libcfs_nid2str(exp->exp_connection->c_peer.nid); - - return "(no nid)"; -} -EXPORT_SYMBOL(obd_export_nid2str); - int obd_export_evict_by_nid(struct obd_device *obd, const char *nid) { - cfs_hash_t *nid_hash; + struct cfs_hash *nid_hash; struct obd_export *doomed_exp = NULL; int exports_evicted = 0; @@ -1497,7 +1792,7 @@ EXPORT_SYMBOL(obd_export_evict_by_nid); int obd_export_evict_by_uuid(struct obd_device *obd, const char *uuid) { - cfs_hash_t *uuid_hash; + struct cfs_hash *uuid_hash; struct obd_export *doomed_exp = NULL; struct obd_uuid doomed_uuid; int exports_evicted = 0; @@ -1534,7 +1829,6 @@ int obd_export_evict_by_uuid(struct obd_device *obd, const char *uuid) return exports_evicted; } -EXPORT_SYMBOL(obd_export_evict_by_uuid); #if LUSTRE_TRACKS_LOCK_EXP_REFS void (*class_export_dump_hook)(struct obd_export*) = NULL; @@ -1542,7 +1836,7 @@ EXPORT_SYMBOL(class_export_dump_hook); #endif static void print_export_data(struct obd_export *exp, const char *status, - int locks) + int locks, int debug_level) { struct ptlrpc_reply_state *rs; struct ptlrpc_reply_state *first_reply = NULL; @@ -1557,39 +1851,35 @@ static void print_export_data(struct obd_export *exp, const char *status, } spin_unlock(&exp->exp_lock); - CDEBUG(D_HA, "%s: %s %p %s %s %d (%d %d %d) %d %d %d %d: %p %s "LPU64"\n", - exp->exp_obd->obd_name, status, exp, exp->exp_client_uuid.uuid, + CDEBUG(debug_level, "%s: %s %p %s %s %d (%d %d %d) %d %d %d %d: " + "%p %s %llu stale:%d\n", + exp->exp_obd->obd_name, status, exp, exp->exp_client_uuid.uuid, obd_export_nid2str(exp), atomic_read(&exp->exp_refcount), atomic_read(&exp->exp_rpc_count), atomic_read(&exp->exp_cb_count), atomic_read(&exp->exp_locks_count), - exp->exp_disconnected, exp->exp_delayed, exp->exp_failed, - nreplies, first_reply, nreplies > 3 ? "..." : "", - exp->exp_last_committed); + exp->exp_disconnected, exp->exp_delayed, exp->exp_failed, + nreplies, first_reply, nreplies > 3 ? "..." : "", + exp->exp_last_committed, !list_empty(&exp->exp_stale_list)); #if LUSTRE_TRACKS_LOCK_EXP_REFS - if (locks && class_export_dump_hook != NULL) - class_export_dump_hook(exp); + if (locks && class_export_dump_hook != NULL) + class_export_dump_hook(exp); #endif } -void dump_exports(struct obd_device *obd, int locks) +void dump_exports(struct obd_device *obd, int locks, int debug_level) { struct obd_export *exp; spin_lock(&obd->obd_dev_lock); list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) - print_export_data(exp, "ACTIVE", locks); + print_export_data(exp, "ACTIVE", locks, debug_level); list_for_each_entry(exp, &obd->obd_unlinked_exports, exp_obd_chain) - print_export_data(exp, "UNLINKED", locks); + print_export_data(exp, "UNLINKED", locks, debug_level); list_for_each_entry(exp, &obd->obd_delayed_exports, exp_obd_chain) - print_export_data(exp, "DELAYED", locks); + print_export_data(exp, "DELAYED", locks, debug_level); spin_unlock(&obd->obd_dev_lock); - spin_lock(&obd_zombie_impexp_lock); - list_for_each_entry(exp, &obd_zombie_exports, exp_obd_chain) - print_export_data(exp, "ZOMBIE", locks); - spin_unlock(&obd_zombie_impexp_lock); } -EXPORT_SYMBOL(dump_exports); void obd_exports_barrier(struct obd_device *obd) { @@ -1598,15 +1888,15 @@ void obd_exports_barrier(struct obd_device *obd) spin_lock(&obd->obd_dev_lock); while (!list_empty(&obd->obd_unlinked_exports)) { spin_unlock(&obd->obd_dev_lock); - schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, - cfs_time_seconds(waited)); - if (waited > 5 && IS_PO2(waited)) { + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(cfs_time_seconds(waited)); + if (waited > 5 && is_power_of_2(waited)) { LCONSOLE_WARN("%s is waiting for obd_unlinked_exports " "more than %d seconds. " "The obd refcount = %d. Is it stuck?\n", obd->obd_name, waited, atomic_read(&obd->obd_refcount)); - dump_exports(obd, 1); + dump_exports(obd, 1, D_CONSOLE | D_WARNING); } waited *= 2; spin_lock(&obd->obd_dev_lock); @@ -1615,97 +1905,17 @@ void obd_exports_barrier(struct obd_device *obd) } EXPORT_SYMBOL(obd_exports_barrier); -/* Total amount of zombies to be destroyed */ -static int zombies_count = 0; - -/** - * kill zombie imports and exports - */ -void obd_zombie_impexp_cull(void) -{ - struct obd_import *import; - struct obd_export *export; - ENTRY; - - do { - spin_lock(&obd_zombie_impexp_lock); - - import = NULL; - if (!list_empty(&obd_zombie_imports)) { - import = list_entry(obd_zombie_imports.next, - struct obd_import, - imp_zombie_chain); - list_del_init(&import->imp_zombie_chain); - } - - export = NULL; - if (!list_empty(&obd_zombie_exports)) { - export = list_entry(obd_zombie_exports.next, - struct obd_export, - exp_obd_chain); - list_del_init(&export->exp_obd_chain); - } - - spin_unlock(&obd_zombie_impexp_lock); - - if (import != NULL) { - class_import_destroy(import); - spin_lock(&obd_zombie_impexp_lock); - zombies_count--; - spin_unlock(&obd_zombie_impexp_lock); - } - - if (export != NULL) { - class_export_destroy(export); - spin_lock(&obd_zombie_impexp_lock); - zombies_count--; - spin_unlock(&obd_zombie_impexp_lock); - } - - cond_resched(); - } while (import != NULL || export != NULL); - EXIT; -} - -static struct completion obd_zombie_start; -static struct completion obd_zombie_stop; -static unsigned long obd_zombie_flags; -static wait_queue_head_t obd_zombie_waitq; -static pid_t obd_zombie_pid; - -enum { - OBD_ZOMBIE_STOP = 0x0001, -}; - -/** - * check for work for kill zombie import/export thread. - */ -static int obd_zombie_impexp_check(void *arg) -{ - int rc; - - spin_lock(&obd_zombie_impexp_lock); - rc = (zombies_count == 0) && - !test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags); - spin_unlock(&obd_zombie_impexp_lock); - - RETURN(rc); -} - /** * Add export to the obd_zombe thread and notify it. */ static void obd_zombie_export_add(struct obd_export *exp) { + atomic_dec(&obd_stale_export_num); spin_lock(&exp->exp_obd->obd_dev_lock); LASSERT(!list_empty(&exp->exp_obd_chain)); list_del_init(&exp->exp_obd_chain); spin_unlock(&exp->exp_obd->obd_dev_lock); - spin_lock(&obd_zombie_impexp_lock); - zombies_count++; - list_add(&exp->exp_obd_chain, &obd_zombie_exports); - spin_unlock(&obd_zombie_impexp_lock); - obd_zombie_impexp_notify(); + queue_work(zombie_wq, &exp->exp_zombie_work); } /** @@ -1713,41 +1923,8 @@ static void obd_zombie_export_add(struct obd_export *exp) { */ static void obd_zombie_import_add(struct obd_import *imp) { LASSERT(imp->imp_sec == NULL); - LASSERT(imp->imp_rq_pool == NULL); - spin_lock(&obd_zombie_impexp_lock); - LASSERT(list_empty(&imp->imp_zombie_chain)); - zombies_count++; - list_add(&imp->imp_zombie_chain, &obd_zombie_imports); - spin_unlock(&obd_zombie_impexp_lock); - - obd_zombie_impexp_notify(); -} - -/** - * notify import/export destroy thread about new zombie. - */ -static void obd_zombie_impexp_notify(void) -{ - /* - * Make sure obd_zomebie_impexp_thread get this notification. - * It is possible this signal only get by obd_zombie_barrier, and - * barrier gulps this notification and sleeps away and hangs ensues - */ - wake_up_all(&obd_zombie_waitq); -} - -/** - * check whether obd_zombie is idle - */ -static int obd_zombie_is_idle(void) -{ - int rc; - LASSERT(!test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags)); - spin_lock(&obd_zombie_impexp_lock); - rc = (zombies_count == 0); - spin_unlock(&obd_zombie_impexp_lock); - return rc; + queue_work(zombie_wq, &imp->imp_zombie_work); } /** @@ -1755,77 +1932,100 @@ static int obd_zombie_is_idle(void) */ void obd_zombie_barrier(void) { - struct l_wait_info lwi = { 0 }; - - if (obd_zombie_pid == current_pid()) - /* don't wait for myself */ - return; - l_wait_event(obd_zombie_waitq, obd_zombie_is_idle(), &lwi); + flush_workqueue(zombie_wq); } EXPORT_SYMBOL(obd_zombie_barrier); -/** - * destroy zombie export/import thread. - */ -static int obd_zombie_impexp_thread(void *unused) +struct obd_export *obd_stale_export_get(void) { - unshare_fs_struct(); - complete(&obd_zombie_start); + struct obd_export *exp = NULL; + ENTRY; - obd_zombie_pid = current_pid(); + spin_lock(&obd_stale_export_lock); + if (!list_empty(&obd_stale_exports)) { + exp = list_entry(obd_stale_exports.next, + struct obd_export, exp_stale_list); + list_del_init(&exp->exp_stale_list); + } + spin_unlock(&obd_stale_export_lock); - while (!test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags)) { - struct l_wait_info lwi = { 0 }; + if (exp) { + CDEBUG(D_DLMTRACE, "Get export %p: total %d\n", exp, + atomic_read(&obd_stale_export_num)); + } + RETURN(exp); +} +EXPORT_SYMBOL(obd_stale_export_get); - l_wait_event(obd_zombie_waitq, - !obd_zombie_impexp_check(NULL), &lwi); - obd_zombie_impexp_cull(); +void obd_stale_export_put(struct obd_export *exp) +{ + ENTRY; - /* - * Notify obd_zombie_barrier callers that queues - * may be empty. - */ - wake_up(&obd_zombie_waitq); + LASSERT(list_empty(&exp->exp_stale_list)); + if (exp->exp_lock_hash && + atomic_read(&exp->exp_lock_hash->hs_count)) { + CDEBUG(D_DLMTRACE, "Put export %p: total %d\n", exp, + atomic_read(&obd_stale_export_num)); + + spin_lock_bh(&exp->exp_bl_list_lock); + spin_lock(&obd_stale_export_lock); + /* Add to the tail if there is no blocked locks, + * to the head otherwise. */ + if (list_empty(&exp->exp_bl_list)) + list_add_tail(&exp->exp_stale_list, + &obd_stale_exports); + else + list_add(&exp->exp_stale_list, + &obd_stale_exports); + + spin_unlock(&obd_stale_export_lock); + spin_unlock_bh(&exp->exp_bl_list_lock); + } else { + class_export_put(exp); } + EXIT; +} +EXPORT_SYMBOL(obd_stale_export_put); - complete(&obd_zombie_stop); +/** + * Adjust the position of the export in the stale list, + * i.e. move to the head of the list if is needed. + **/ +void obd_stale_export_adjust(struct obd_export *exp) +{ + LASSERT(exp != NULL); + spin_lock_bh(&exp->exp_bl_list_lock); + spin_lock(&obd_stale_export_lock); - RETURN(0); -} + if (!list_empty(&exp->exp_stale_list) && + !list_empty(&exp->exp_bl_list)) + list_move(&exp->exp_stale_list, &obd_stale_exports); + spin_unlock(&obd_stale_export_lock); + spin_unlock_bh(&exp->exp_bl_list_lock); +} +EXPORT_SYMBOL(obd_stale_export_adjust); /** * start destroy zombie import/export thread */ int obd_zombie_impexp_init(void) { - struct task_struct *task; + zombie_wq = alloc_workqueue("obd_zombid", 0, 0); + if (!zombie_wq) + return -ENOMEM; - INIT_LIST_HEAD(&obd_zombie_imports); - - INIT_LIST_HEAD(&obd_zombie_exports); - spin_lock_init(&obd_zombie_impexp_lock); - init_completion(&obd_zombie_start); - init_completion(&obd_zombie_stop); - init_waitqueue_head(&obd_zombie_waitq); - obd_zombie_pid = 0; - - task = kthread_run(obd_zombie_impexp_thread, NULL, "obd_zombid"); - if (IS_ERR(task)) - RETURN(PTR_ERR(task)); - - wait_for_completion(&obd_zombie_start); - RETURN(0); + return 0; } + /** * stop destroy zombie import/export thread */ void obd_zombie_impexp_stop(void) { - set_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags); - obd_zombie_impexp_notify(); - wait_for_completion(&obd_zombie_stop); + destroy_workqueue(zombie_wq); + LASSERT(list_empty(&obd_stale_exports)); } /***** Kernel-userspace comm helpers *******/ @@ -1849,21 +2049,6 @@ struct kuc_hdr * kuc_ptr(void *p) } EXPORT_SYMBOL(kuc_ptr); -/* Test if payload is part of kuc message - * @param p Pointer to payload area - * @returns boolean - */ -int kuc_ispayload(void *p) -{ - struct kuc_hdr *kh = ((struct kuc_hdr *)p) - 1; - - if (kh->kuc_magic == KUC_MAGIC) - return 1; - else - return 0; -} -EXPORT_SYMBOL(kuc_ispayload); - /* Alloc space for a message, and fill in header * @return Pointer to payload area */ @@ -1886,7 +2071,7 @@ void *kuc_alloc(int payload_len, int transport, int type) EXPORT_SYMBOL(kuc_alloc); /* Takes pointer to payload area */ -inline void kuc_free(void *p, int payload_len) +void kuc_free(void *p, int payload_len) { struct kuc_hdr *lh = kuc_ptr(p); OBD_FREE(lh, kuc_len(payload_len)); @@ -1904,9 +2089,9 @@ static bool obd_request_slot_avail(struct client_obd *cli, { bool avail; - client_obd_list_lock(&cli->cl_loi_list_lock); + spin_lock(&cli->cl_loi_list_lock); avail = !!list_empty(&orsw->orsw_entry); - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_unlock(&cli->cl_loi_list_lock); return avail; }; @@ -1924,17 +2109,17 @@ int obd_get_request_slot(struct client_obd *cli) struct l_wait_info lwi; int rc; - client_obd_list_lock(&cli->cl_loi_list_lock); - if (cli->cl_r_in_flight < cli->cl_max_rpcs_in_flight) { - cli->cl_r_in_flight++; - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_lock(&cli->cl_loi_list_lock); + if (cli->cl_rpcs_in_flight < cli->cl_max_rpcs_in_flight) { + cli->cl_rpcs_in_flight++; + spin_unlock(&cli->cl_loi_list_lock); return 0; } init_waitqueue_head(&orsw.orsw_waitq); - list_add_tail(&orsw.orsw_entry, &cli->cl_loi_read_list); + list_add_tail(&orsw.orsw_entry, &cli->cl_flight_waiters); orsw.orsw_signaled = false; - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_unlock(&cli->cl_loi_list_lock); lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL); rc = l_wait_event(orsw.orsw_waitq, @@ -1944,11 +2129,11 @@ int obd_get_request_slot(struct client_obd *cli) /* Here, we must take the lock to avoid the on-stack 'orsw' to be * freed but other (such as obd_put_request_slot) is using it. */ - client_obd_list_lock(&cli->cl_loi_list_lock); + spin_lock(&cli->cl_loi_list_lock); if (rc != 0) { if (!orsw.orsw_signaled) { if (list_empty(&orsw.orsw_entry)) - cli->cl_r_in_flight--; + cli->cl_rpcs_in_flight--; else list_del(&orsw.orsw_entry); } @@ -1959,7 +2144,7 @@ int obd_get_request_slot(struct client_obd *cli) rc = -EINTR; } - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_unlock(&cli->cl_loi_list_lock); return rc; } @@ -1969,19 +2154,19 @@ void obd_put_request_slot(struct client_obd *cli) { struct obd_request_slot_waiter *orsw; - client_obd_list_lock(&cli->cl_loi_list_lock); - cli->cl_r_in_flight--; + spin_lock(&cli->cl_loi_list_lock); + cli->cl_rpcs_in_flight--; /* If there is free slot, wakeup the first waiter. */ - if (!list_empty(&cli->cl_loi_read_list) && - likely(cli->cl_r_in_flight < cli->cl_max_rpcs_in_flight)) { - orsw = list_entry(cli->cl_loi_read_list.next, + if (!list_empty(&cli->cl_flight_waiters) && + likely(cli->cl_rpcs_in_flight < cli->cl_max_rpcs_in_flight)) { + orsw = list_entry(cli->cl_flight_waiters.next, struct obd_request_slot_waiter, orsw_entry); list_del_init(&orsw->orsw_entry); - cli->cl_r_in_flight++; + cli->cl_rpcs_in_flight++; wake_up(&orsw->orsw_waitq); } - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_unlock(&cli->cl_loi_list_lock); } EXPORT_SYMBOL(obd_put_request_slot); @@ -1997,28 +2182,274 @@ int obd_set_max_rpcs_in_flight(struct client_obd *cli, __u32 max) __u32 old; int diff; int i; + char *typ_name; + int rc; if (max > OBD_MAX_RIF_MAX || max < 1) return -ERANGE; - client_obd_list_lock(&cli->cl_loi_list_lock); + typ_name = cli->cl_import->imp_obd->obd_type->typ_name; + if (strcmp(typ_name, LUSTRE_MDC_NAME) == 0) { + /* adjust max_mod_rpcs_in_flight to ensure it is always + * strictly lower that max_rpcs_in_flight */ + if (max < 2) { + CERROR("%s: cannot set max_rpcs_in_flight to 1 " + "because it must be higher than " + "max_mod_rpcs_in_flight value", + cli->cl_import->imp_obd->obd_name); + return -ERANGE; + } + if (max <= cli->cl_max_mod_rpcs_in_flight) { + rc = obd_set_max_mod_rpcs_in_flight(cli, max - 1); + if (rc != 0) + return rc; + } + } + + spin_lock(&cli->cl_loi_list_lock); old = cli->cl_max_rpcs_in_flight; cli->cl_max_rpcs_in_flight = max; + client_adjust_max_dirty(cli); + diff = max - old; /* We increase the max_rpcs_in_flight, then wakeup some waiters. */ for (i = 0; i < diff; i++) { - if (list_empty(&cli->cl_loi_read_list)) + if (list_empty(&cli->cl_flight_waiters)) break; - orsw = list_entry(cli->cl_loi_read_list.next, + orsw = list_entry(cli->cl_flight_waiters.next, struct obd_request_slot_waiter, orsw_entry); list_del_init(&orsw->orsw_entry); - cli->cl_r_in_flight++; + cli->cl_rpcs_in_flight++; wake_up(&orsw->orsw_waitq); } - client_obd_list_unlock(&cli->cl_loi_list_lock); + spin_unlock(&cli->cl_loi_list_lock); return 0; } EXPORT_SYMBOL(obd_set_max_rpcs_in_flight); + +__u16 obd_get_max_mod_rpcs_in_flight(struct client_obd *cli) +{ + return cli->cl_max_mod_rpcs_in_flight; +} +EXPORT_SYMBOL(obd_get_max_mod_rpcs_in_flight); + +int obd_set_max_mod_rpcs_in_flight(struct client_obd *cli, __u16 max) +{ + struct obd_connect_data *ocd; + __u16 maxmodrpcs; + __u16 prev; + + if (max > OBD_MAX_RIF_MAX || max < 1) + return -ERANGE; + + /* cannot exceed or equal max_rpcs_in_flight */ + if (max >= cli->cl_max_rpcs_in_flight) { + CERROR("%s: can't set max_mod_rpcs_in_flight to a value (%hu) " + "higher or equal to max_rpcs_in_flight value (%u)\n", + cli->cl_import->imp_obd->obd_name, + max, cli->cl_max_rpcs_in_flight); + return -ERANGE; + } + + /* cannot exceed max modify RPCs in flight supported by the server */ + ocd = &cli->cl_import->imp_connect_data; + if (ocd->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) + maxmodrpcs = ocd->ocd_maxmodrpcs; + else + maxmodrpcs = 1; + if (max > maxmodrpcs) { + CERROR("%s: can't set max_mod_rpcs_in_flight to a value (%hu) " + "higher than max_mod_rpcs_per_client value (%hu) " + "returned by the server at connection\n", + cli->cl_import->imp_obd->obd_name, + max, maxmodrpcs); + return -ERANGE; + } + + spin_lock(&cli->cl_mod_rpcs_lock); + + prev = cli->cl_max_mod_rpcs_in_flight; + cli->cl_max_mod_rpcs_in_flight = max; + + /* wakeup waiters if limit has been increased */ + if (cli->cl_max_mod_rpcs_in_flight > prev) + wake_up(&cli->cl_mod_rpcs_waitq); + + spin_unlock(&cli->cl_mod_rpcs_lock); + + return 0; +} +EXPORT_SYMBOL(obd_set_max_mod_rpcs_in_flight); + + +#define pct(a, b) (b ? a * 100 / b : 0) +int obd_mod_rpc_stats_seq_show(struct client_obd *cli, + struct seq_file *seq) +{ + unsigned long mod_tot = 0, mod_cum; + struct timespec64 now; + int i; + + ktime_get_real_ts64(&now); + + spin_lock(&cli->cl_mod_rpcs_lock); + + seq_printf(seq, "snapshot_time: %llu.%9lu (secs.nsecs)\n", + (s64)now.tv_sec, now.tv_nsec); + seq_printf(seq, "modify_RPCs_in_flight: %hu\n", + cli->cl_mod_rpcs_in_flight); + + seq_printf(seq, "\n\t\t\tmodify\n"); + seq_printf(seq, "rpcs in flight rpcs %% cum %%\n"); + + mod_tot = lprocfs_oh_sum(&cli->cl_mod_rpcs_hist); + + mod_cum = 0; + for (i = 0; i < OBD_HIST_MAX; i++) { + unsigned long mod = cli->cl_mod_rpcs_hist.oh_buckets[i]; + mod_cum += mod; + seq_printf(seq, "%d:\t\t%10lu %3lu %3lu\n", + i, mod, pct(mod, mod_tot), + pct(mod_cum, mod_tot)); + if (mod_cum == mod_tot) + break; + } + + spin_unlock(&cli->cl_mod_rpcs_lock); + + return 0; +} +EXPORT_SYMBOL(obd_mod_rpc_stats_seq_show); +#undef pct + + +/* The number of modify RPCs sent in parallel is limited + * because the server has a finite number of slots per client to + * store request result and ensure reply reconstruction when needed. + * On the client, this limit is stored in cl_max_mod_rpcs_in_flight + * that takes into account server limit and cl_max_rpcs_in_flight + * value. + * On the MDC client, to avoid a potential deadlock (see Bugzilla 3462), + * one close request is allowed above the maximum. + */ +static inline bool obd_mod_rpc_slot_avail_locked(struct client_obd *cli, + bool close_req) +{ + bool avail; + + /* A slot is available if + * - number of modify RPCs in flight is less than the max + * - it's a close RPC and no other close request is in flight + */ + avail = cli->cl_mod_rpcs_in_flight < cli->cl_max_mod_rpcs_in_flight || + (close_req && cli->cl_close_rpcs_in_flight == 0); + + return avail; +} + +static inline bool obd_mod_rpc_slot_avail(struct client_obd *cli, + bool close_req) +{ + bool avail; + + spin_lock(&cli->cl_mod_rpcs_lock); + avail = obd_mod_rpc_slot_avail_locked(cli, close_req); + spin_unlock(&cli->cl_mod_rpcs_lock); + return avail; +} + +static inline bool obd_skip_mod_rpc_slot(const struct lookup_intent *it) +{ + if (it != NULL && + (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP || + it->it_op == IT_READDIR || + (it->it_op == IT_LAYOUT && !(it->it_flags & MDS_FMODE_WRITE)))) + return true; + return false; +} + +/* Get a modify RPC slot from the obd client @cli according + * to the kind of operation @opc that is going to be sent + * and the intent @it of the operation if it applies. + * If the maximum number of modify RPCs in flight is reached + * the thread is put to sleep. + * Returns the tag to be set in the request message. Tag 0 + * is reserved for non-modifying requests. + */ +__u16 obd_get_mod_rpc_slot(struct client_obd *cli, __u32 opc, + struct lookup_intent *it) +{ + struct l_wait_info lwi = LWI_INTR(NULL, NULL); + bool close_req = false; + __u16 i, max; + + /* read-only metadata RPCs don't consume a slot on MDT + * for reply reconstruction + */ + if (obd_skip_mod_rpc_slot(it)) + return 0; + + if (opc == MDS_CLOSE) + close_req = true; + + do { + spin_lock(&cli->cl_mod_rpcs_lock); + max = cli->cl_max_mod_rpcs_in_flight; + if (obd_mod_rpc_slot_avail_locked(cli, close_req)) { + /* there is a slot available */ + cli->cl_mod_rpcs_in_flight++; + if (close_req) + cli->cl_close_rpcs_in_flight++; + lprocfs_oh_tally(&cli->cl_mod_rpcs_hist, + cli->cl_mod_rpcs_in_flight); + /* find a free tag */ + i = find_first_zero_bit(cli->cl_mod_tag_bitmap, + max + 1); + LASSERT(i < OBD_MAX_RIF_MAX); + LASSERT(!test_and_set_bit(i, cli->cl_mod_tag_bitmap)); + spin_unlock(&cli->cl_mod_rpcs_lock); + /* tag 0 is reserved for non-modify RPCs */ + return i + 1; + } + spin_unlock(&cli->cl_mod_rpcs_lock); + + CDEBUG(D_RPCTRACE, "%s: sleeping for a modify RPC slot " + "opc %u, max %hu\n", + cli->cl_import->imp_obd->obd_name, opc, max); + + l_wait_event(cli->cl_mod_rpcs_waitq, + obd_mod_rpc_slot_avail(cli, close_req), &lwi); + } while (true); +} +EXPORT_SYMBOL(obd_get_mod_rpc_slot); + +/* Put a modify RPC slot from the obd client @cli according + * to the kind of operation @opc that has been sent and the + * intent @it of the operation if it applies. + */ +void obd_put_mod_rpc_slot(struct client_obd *cli, __u32 opc, + struct lookup_intent *it, __u16 tag) +{ + bool close_req = false; + + if (obd_skip_mod_rpc_slot(it)) + return; + + if (opc == MDS_CLOSE) + close_req = true; + + spin_lock(&cli->cl_mod_rpcs_lock); + cli->cl_mod_rpcs_in_flight--; + if (close_req) + cli->cl_close_rpcs_in_flight--; + /* release the tag in the bitmap */ + LASSERT(tag - 1 < OBD_MAX_RIF_MAX); + LASSERT(test_and_clear_bit(tag - 1, cli->cl_mod_tag_bitmap) != 0); + spin_unlock(&cli->cl_mod_rpcs_lock); + wake_up(&cli->cl_mod_rpcs_waitq); +} +EXPORT_SYMBOL(obd_put_mod_rpc_slot); +