X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdclass%2Fgenops.c;h=436aa4d587e5fb99ac9f923f01791c9222a5e17a;hp=d61104e0f437f025a444545d5966780b417069f4;hb=0607e01af74a81d0fe12ceec79bd22810a5dfe92;hpb=36e8719a5d9684f1e1883656d28e56612d10f757 diff --git a/lustre/obdclass/genops.c b/lustre/obdclass/genops.c index d61104e..436aa4d 100644 --- a/lustre/obdclass/genops.c +++ b/lustre/obdclass/genops.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,6 +26,8 @@ /* * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -48,7 +48,7 @@ #include extern cfs_list_t obd_types; -cfs_spinlock_t obd_types_lock; +spinlock_t obd_types_lock; cfs_mem_cache_t *obd_device_cachep; cfs_mem_cache_t *obdo_cachep; @@ -57,7 +57,7 @@ cfs_mem_cache_t *import_cachep; cfs_list_t obd_zombie_imports; cfs_list_t obd_zombie_exports; -cfs_spinlock_t obd_zombie_impexp_lock; +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); @@ -65,6 +65,7 @@ static void print_export_data(struct obd_export *exp, const char *status, int locks); int (*ptlrpc_put_connection_superhack)(struct ptlrpc_connection *c); +EXPORT_SYMBOL(ptlrpc_put_connection_superhack); /* * support functions: we could use inter-module communication, but this @@ -80,7 +81,6 @@ static struct obd_device *obd_device_alloc(void) } return obd; } -EXPORT_SYMBOL(obd_device_alloc); static void obd_device_free(struct obd_device *obd) { @@ -98,20 +98,21 @@ static void obd_device_free(struct obd_device *obd) struct obd_type *class_search_type(const char *name) { - cfs_list_t *tmp; - struct obd_type *type; - - cfs_spin_lock(&obd_types_lock); - cfs_list_for_each(tmp, &obd_types) { - type = cfs_list_entry(tmp, struct obd_type, typ_chain); - if (strcmp(type->typ_name, name) == 0) { - cfs_spin_unlock(&obd_types_lock); - return type; - } - } - cfs_spin_unlock(&obd_types_lock); - return NULL; -} + cfs_list_t *tmp; + struct obd_type *type; + + spin_lock(&obd_types_lock); + cfs_list_for_each(tmp, &obd_types) { + type = cfs_list_entry(tmp, struct obd_type, typ_chain); + if (strcmp(type->typ_name, name) == 0) { + spin_unlock(&obd_types_lock); + return type; + } + } + spin_unlock(&obd_types_lock); + return NULL; +} +EXPORT_SYMBOL(class_search_type); struct obd_type *class_get_type(const char *name) { @@ -120,6 +121,16 @@ struct obd_type *class_get_type(const char *name) #ifdef HAVE_MODULE_LOADING_SUPPORT if (!type) { const char *modname = name; + + if (strcmp(modname, "obdfilter") == 0) + modname = "ofd"; + + if (strcmp(modname, LUSTRE_LWP_NAME) == 0) + modname = LUSTRE_OSP_NAME; + + if (!strncmp(modname, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME))) + modname = LUSTRE_MDT_NAME; + if (!cfs_request_module("%s", modname)) { CDEBUG(D_INFO, "Loaded module '%s'\n", modname); type = class_search_type(name); @@ -130,22 +141,24 @@ struct obd_type *class_get_type(const char *name) } #endif if (type) { - cfs_spin_lock(&type->obd_type_lock); - type->typ_refcnt++; - cfs_try_module_get(type->typ_dt_ops->o_owner); - cfs_spin_unlock(&type->obd_type_lock); - } - return type; + spin_lock(&type->obd_type_lock); + type->typ_refcnt++; + cfs_try_module_get(type->typ_dt_ops->o_owner); + spin_unlock(&type->obd_type_lock); + } + return type; } +EXPORT_SYMBOL(class_get_type); void class_put_type(struct obd_type *type) { - LASSERT(type); - cfs_spin_lock(&type->obd_type_lock); - type->typ_refcnt--; - cfs_module_put(type->typ_dt_ops->o_owner); - cfs_spin_unlock(&type->obd_type_lock); + LASSERT(type); + spin_lock(&type->obd_type_lock); + type->typ_refcnt--; + cfs_module_put(type->typ_dt_ops->o_owner); + spin_unlock(&type->obd_type_lock); } +EXPORT_SYMBOL(class_put_type); #define CLASS_MAX_NAME 1024 @@ -184,7 +197,7 @@ int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops, if (md_ops) *(type->typ_md_ops) = *md_ops; strcpy(type->typ_name, name); - cfs_spin_lock_init(&type->obd_type_lock); + spin_lock_init(&type->obd_type_lock); #ifdef LPROCFS type->typ_procroot = lprocfs_register(type->typ_name, proc_lustre_root, @@ -202,9 +215,9 @@ int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops, GOTO (failed, rc); } - cfs_spin_lock(&obd_types_lock); - cfs_list_add(&type->typ_chain, &obd_types); - cfs_spin_unlock(&obd_types_lock); + spin_lock(&obd_types_lock); + cfs_list_add(&type->typ_chain, &obd_types); + spin_unlock(&obd_types_lock); RETURN (0); @@ -218,6 +231,7 @@ int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops, OBD_FREE(type, sizeof(*type)); RETURN(rc); } +EXPORT_SYMBOL(class_register_type); int class_unregister_type(const char *name) { @@ -238,16 +252,18 @@ int class_unregister_type(const char *name) RETURN(-EBUSY); } - if (type->typ_procroot) { - lprocfs_remove(&type->typ_procroot); - } + /* 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 */ + lprocfs_try_remove_proc_entry(type->typ_name, proc_lustre_root); if (type->typ_lu) lu_device_type_fini(type->typ_lu); - cfs_spin_lock(&obd_types_lock); - cfs_list_del(&type->typ_chain); - cfs_spin_unlock(&obd_types_lock); + spin_lock(&obd_types_lock); + cfs_list_del(&type->typ_chain); + spin_unlock(&obd_types_lock); OBD_FREE(type->typ_name, strlen(name) + 1); if (type->typ_dt_ops != NULL) OBD_FREE_PTR(type->typ_dt_ops); @@ -256,6 +272,7 @@ int class_unregister_type(const char *name) OBD_FREE(type, sizeof(*type)); RETURN(0); } /* class_unregister_type */ +EXPORT_SYMBOL(class_unregister_type); /** * Create a new obd device. @@ -275,6 +292,7 @@ struct obd_device *class_newdev(const char *type_name, const char *name) struct obd_type *type = NULL; int i; int new_obd_minor = 0; + ENTRY; if (strlen(name) >= MAX_OBD_NAME) { CERROR("name/uuid must be < %u bytes long\n", MAX_OBD_NAME); @@ -288,18 +306,18 @@ struct obd_device *class_newdev(const char *type_name, const char *name) } newdev = obd_device_alloc(); - if (newdev == NULL) { - class_put_type(type); - RETURN(ERR_PTR(-ENOMEM)); - } + if (newdev == NULL) + GOTO(out_type, result = ERR_PTR(-ENOMEM)); + LASSERT(newdev->obd_magic == OBD_DEVICE_MAGIC); - cfs_spin_lock(&obd_dev_lock); + write_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); - if (obd && obd->obd_name && - (strcmp(name, obd->obd_name) == 0)) { - CERROR("Device %s already exists, won't add\n", name); + + 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, @@ -324,22 +342,26 @@ struct obd_device *class_newdev(const char *type_name, const char *name) obd_devs[i] = result; } } - cfs_spin_unlock(&obd_dev_lock); + write_unlock(&obd_dev_lock); if (result == NULL && i >= class_devno_max()) { CERROR("all %u OBD devices used, increase MAX_OBD_DEVICES\n", class_devno_max()); - result = ERR_PTR(-EOVERFLOW); + GOTO(out, result = ERR_PTR(-EOVERFLOW)); } - if (IS_ERR(result)) { - obd_device_free(newdev); - class_put_type(type); - } else { - CDEBUG(D_IOCTL, "Adding new device %s (%p)\n", - result->obd_name, result); - } - return result; + if (IS_ERR(result)) + GOTO(out, result); + + CDEBUG(D_IOCTL, "Adding new device %s (%p)\n", + result->obd_name, result); + + RETURN(result); +out: + obd_device_free(newdev); +out_type: + class_put_type(type); + return result; } void class_release_dev(struct obd_device *obd) @@ -352,12 +374,12 @@ void class_release_dev(struct obd_device *obd) obd, obd->obd_minor, obd_devs[obd->obd_minor]); LASSERT(obd_type != NULL); - CDEBUG(D_INFO, "Release obd device %s obd_type name =%s\n", - obd->obd_name,obd->obd_type->typ_name); + 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); - cfs_spin_lock(&obd_dev_lock); + write_lock(&obd_dev_lock); obd_devs[obd->obd_minor] = NULL; - cfs_spin_unlock(&obd_dev_lock); + write_unlock(&obd_dev_lock); obd_device_free(obd); class_put_type(obd_type); @@ -370,24 +392,26 @@ int class_name2dev(const char *name) if (!name) return -1; - cfs_spin_lock(&obd_dev_lock); + read_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); - if (obd && obd->obd_name && strcmp(name, obd->obd_name) == 0) { + + if (obd && strcmp(name, obd->obd_name) == 0) { /* Make sure we finished attaching before we give out any references */ LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC); if (obd->obd_attached) { - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return i; } break; } } - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return -1; } +EXPORT_SYMBOL(class_name2dev); struct obd_device *class_name2obd(const char *name) { @@ -397,24 +421,27 @@ struct obd_device *class_name2obd(const char *name) return NULL; return class_num2obd(dev); } +EXPORT_SYMBOL(class_name2obd); int class_uuid2dev(struct obd_uuid *uuid) { int i; - cfs_spin_lock(&obd_dev_lock); + 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); - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return i; } } - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return -1; } +EXPORT_SYMBOL(class_uuid2dev); struct obd_device *class_uuid2obd(struct obd_uuid *uuid) { @@ -423,6 +450,7 @@ struct obd_device *class_uuid2obd(struct obd_uuid *uuid) return NULL; return class_num2obd(dev); } +EXPORT_SYMBOL(class_uuid2obd); /** * Get obd device from ::obd_devs[] @@ -451,15 +479,38 @@ struct obd_device *class_num2obd(int num) return obd; } +EXPORT_SYMBOL(class_num2obd); + +/** + * Get obd devices count. Device in any + * state are counted + * \retval obd device count + */ +int get_devices_count(void) +{ + int index, max_index = class_devno_max(), dev_count = 0; + + read_lock(&obd_dev_lock); + for (index = 0; index <= max_index; index++) { + struct obd_device *obd = class_num2obd(index); + if (obd != NULL) + dev_count++; + } + read_unlock(&obd_dev_lock); + + return dev_count; +} +EXPORT_SYMBOL(get_devices_count); void class_obd_list(void) { char *status; int i; - cfs_spin_lock(&obd_dev_lock); + read_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); + if (obd == NULL) continue; if (obd->obd_stopping) @@ -475,7 +526,7 @@ void class_obd_list(void) obd->obd_name, obd->obd_uuid.uuid, cfs_atomic_read(&obd->obd_refcount)); } - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return; } @@ -488,9 +539,10 @@ struct obd_device * class_find_client_obd(struct obd_uuid *tgt_uuid, { int i; - cfs_spin_lock(&obd_dev_lock); + read_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); + if (obd == NULL) continue; if ((strncmp(obd->obd_type->typ_name, typ_name, @@ -499,15 +551,16 @@ struct obd_device * class_find_client_obd(struct obd_uuid *tgt_uuid, &obd->u.cli.cl_target_uuid) && ((grp_uuid)? obd_uuid_equals(grp_uuid, &obd->obd_uuid) : 1)) { - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return obd; } } } - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return NULL; } +EXPORT_SYMBOL(class_find_client_obd); /* Iterate the obd_device list looking devices have grp_uuid. Start searching at *next, and if a device is found, the next index to look @@ -524,22 +577,24 @@ struct obd_device * class_devices_in_group(struct obd_uuid *grp_uuid, int *next) else return NULL; - cfs_spin_lock(&obd_dev_lock); + read_lock(&obd_dev_lock); for (; i < class_devno_max(); i++) { struct obd_device *obd = class_num2obd(i); + if (obd == NULL) continue; if (obd_uuid_equals(grp_uuid, &obd->obd_uuid)) { if (next != NULL) *next = i+1; - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return obd; } } - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return NULL; } +EXPORT_SYMBOL(class_devices_in_group); /** * to notify sptlrpc log for \a fsname has changed, let every relevant OBD @@ -553,7 +608,7 @@ int class_notify_sptlrpc_conf(const char *fsname, int namelen) LASSERT(namelen > 0); - cfs_spin_lock(&obd_dev_lock); + read_lock(&obd_dev_lock); for (i = 0; i < class_devno_max(); i++) { obd = class_num2obd(i); @@ -572,15 +627,15 @@ int class_notify_sptlrpc_conf(const char *fsname, int namelen) continue; class_incref(obd, __FUNCTION__, obd); - cfs_spin_unlock(&obd_dev_lock); - rc2 = obd_set_info_async(obd->obd_self_export, + read_unlock(&obd_dev_lock); + rc2 = obd_set_info_async(NULL, obd->obd_self_export, sizeof(KEY_SPTLRPC_CONF), KEY_SPTLRPC_CONF, 0, NULL, NULL); rc = rc ? rc : rc2; class_decref(obd, __FUNCTION__, obd); - cfs_spin_lock(&obd_dev_lock); + read_lock(&obd_dev_lock); } - cfs_spin_unlock(&obd_dev_lock); + read_unlock(&obd_dev_lock); return rc; } EXPORT_SYMBOL(class_notify_sptlrpc_conf); @@ -670,6 +725,7 @@ struct obd_export *class_conn2export(struct lustre_handle *conn) export = class_handle2object(conn->cookie); RETURN(export); } +EXPORT_SYMBOL(class_conn2export); struct obd_device *class_exp2obd(struct obd_export *exp) { @@ -677,6 +733,7 @@ struct obd_device *class_exp2obd(struct obd_export *exp) return exp->exp_obd; return NULL; } +EXPORT_SYMBOL(class_exp2obd); struct obd_device *class_conn2obd(struct lustre_handle *conn) { @@ -689,6 +746,7 @@ struct obd_device *class_conn2obd(struct lustre_handle *conn) } return NULL; } +EXPORT_SYMBOL(class_conn2obd); struct obd_import *class_exp2cliimp(struct obd_export *exp) { @@ -697,6 +755,7 @@ struct obd_import *class_exp2cliimp(struct obd_export *exp) return NULL; return obd->u.cli.cl_import; } +EXPORT_SYMBOL(class_exp2cliimp); struct obd_import *class_conn2cliimp(struct lustre_handle *conn) { @@ -705,6 +764,7 @@ 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) @@ -712,13 +772,12 @@ static void class_export_destroy(struct obd_export *exp) struct obd_device *obd = exp->exp_obd; ENTRY; - LASSERT (cfs_atomic_read(&exp->exp_refcount) == 0); + LASSERT_ATOMIC_ZERO(&exp->exp_refcount); + LASSERT(obd != NULL); CDEBUG(D_IOCTL, "destroying export %p/%s for %s\n", exp, exp->exp_client_uuid.uuid, obd->obd_name); - LASSERT(obd != NULL); - /* "Local" exports (lctl, LOV->{mdc,osc}) have no connection. */ if (exp->exp_connection) ptlrpc_put_connection_superhack(exp->exp_connection); @@ -726,7 +785,7 @@ static void class_export_destroy(struct obd_export *exp) LASSERT(cfs_list_empty(&exp->exp_outstanding_replies)); LASSERT(cfs_list_empty(&exp->exp_uncommitted_replies)); LASSERT(cfs_list_empty(&exp->exp_req_replay_queue)); - LASSERT(cfs_list_empty(&exp->exp_queued_rpc)); + LASSERT(cfs_list_empty(&exp->exp_hp_rpcs)); obd_destroy_export(exp); class_decref(obd, "export", exp); @@ -739,6 +798,11 @@ static void export_handle_addref(void *export) class_export_get(export); } +static struct portals_handle_ops export_handle_ops = { + .hop_addref = export_handle_addref, + .hop_free = NULL, +}; + struct obd_export *class_export_get(struct obd_export *exp) { cfs_atomic_inc(&exp->exp_refcount); @@ -751,15 +815,18 @@ EXPORT_SYMBOL(class_export_get); void class_export_put(struct obd_export *exp) { LASSERT(exp != NULL); + LASSERT_ATOMIC_GT_LT(&exp->exp_refcount, 0, LI_POISON); CDEBUG(D_INFO, "PUTting export %p : new refcount %d\n", exp, cfs_atomic_read(&exp->exp_refcount) - 1); - LASSERT(cfs_atomic_read(&exp->exp_refcount) > 0); - LASSERT(cfs_atomic_read(&exp->exp_refcount) < 0x5a5a5a); if (cfs_atomic_dec_and_test(&exp->exp_refcount)) { LASSERT(!cfs_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); + obd_zombie_export_add(exp); } } @@ -782,43 +849,46 @@ 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; cfs_atomic_set(&export->exp_refcount, 2); cfs_atomic_set(&export->exp_rpc_count, 0); cfs_atomic_set(&export->exp_cb_count, 0); cfs_atomic_set(&export->exp_locks_count, 0); #if LUSTRE_TRACKS_LOCK_EXP_REFS CFS_INIT_LIST_HEAD(&export->exp_locks_list); - cfs_spin_lock_init(&export->exp_locks_list_guard); + spin_lock_init(&export->exp_locks_list_guard); #endif - cfs_atomic_set(&export->exp_replay_count, 0); - export->exp_obd = obd; - CFS_INIT_LIST_HEAD(&export->exp_outstanding_replies); - cfs_spin_lock_init(&export->exp_uncommitted_replies_lock); - CFS_INIT_LIST_HEAD(&export->exp_uncommitted_replies); - CFS_INIT_LIST_HEAD(&export->exp_req_replay_queue); - CFS_INIT_LIST_HEAD(&export->exp_handle.h_link); - CFS_INIT_LIST_HEAD(&export->exp_queued_rpc); - class_handle_hash(&export->exp_handle, export_handle_addref); - export->exp_last_request_time = cfs_time_current_sec(); - cfs_spin_lock_init(&export->exp_lock); - cfs_spin_lock_init(&export->exp_rpc_lock); - CFS_INIT_HLIST_NODE(&export->exp_uuid_hash); - CFS_INIT_HLIST_NODE(&export->exp_nid_hash); - - export->exp_sp_peer = LUSTRE_SP_ANY; - export->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID; - export->exp_client_uuid = *cluuid; - obd_init_export(export); - - cfs_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); - cfs_spin_unlock(&obd->obd_dev_lock); + cfs_atomic_set(&export->exp_replay_count, 0); + export->exp_obd = obd; + CFS_INIT_LIST_HEAD(&export->exp_outstanding_replies); + spin_lock_init(&export->exp_uncommitted_replies_lock); + CFS_INIT_LIST_HEAD(&export->exp_uncommitted_replies); + CFS_INIT_LIST_HEAD(&export->exp_req_replay_queue); + CFS_INIT_LIST_HEAD(&export->exp_handle.h_link); + CFS_INIT_LIST_HEAD(&export->exp_hp_rpcs); + class_handle_hash(&export->exp_handle, &export_handle_ops); + export->exp_last_request_time = cfs_time_current_sec(); + spin_lock_init(&export->exp_lock); + spin_lock_init(&export->exp_rpc_lock); + CFS_INIT_HLIST_NODE(&export->exp_uuid_hash); + CFS_INIT_HLIST_NODE(&export->exp_nid_hash); + spin_lock_init(&export->exp_bl_list_lock); + CFS_INIT_LIST_HEAD(&export->exp_bl_list); + + 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); + + 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); @@ -829,7 +899,7 @@ struct obd_export *class_new_export(struct obd_device *obd, } } - cfs_spin_lock(&obd->obd_dev_lock); + spin_lock(&obd->obd_dev_lock); if (obd->obd_stopping) { cfs_hash_del(hash, cluuid, &export->exp_uuid_hash); GOTO(exit_unlock, rc = -ENODEV); @@ -840,12 +910,12 @@ struct obd_export *class_new_export(struct obd_device *obd, cfs_list_add_tail(&export->exp_obd_chain_timed, &export->exp_obd->obd_exports_timed); export->exp_obd->obd_num_exports++; - cfs_spin_unlock(&obd->obd_dev_lock); - cfs_hash_putref(hash); - RETURN(export); + spin_unlock(&obd->obd_dev_lock); + cfs_hash_putref(hash); + RETURN(export); exit_unlock: - cfs_spin_unlock(&obd->obd_dev_lock); + spin_unlock(&obd->obd_dev_lock); exit_err: if (hash) cfs_hash_putref(hash); @@ -859,20 +929,20 @@ EXPORT_SYMBOL(class_new_export); void class_unlink_export(struct obd_export *exp) { - class_handle_unhash(&exp->exp_handle); + class_handle_unhash(&exp->exp_handle); - cfs_spin_lock(&exp->exp_obd->obd_dev_lock); - /* delete an uuid-export hashitem from hashtables */ - if (!cfs_hlist_unhashed(&exp->exp_uuid_hash)) - cfs_hash_del(exp->exp_obd->obd_uuid_hash, - &exp->exp_client_uuid, - &exp->exp_uuid_hash); + spin_lock(&exp->exp_obd->obd_dev_lock); + /* delete an uuid-export hashitem from hashtables */ + if (!cfs_hlist_unhashed(&exp->exp_uuid_hash)) + cfs_hash_del(exp->exp_obd->obd_uuid_hash, + &exp->exp_client_uuid, + &exp->exp_uuid_hash); - cfs_list_move(&exp->exp_obd_chain, &exp->exp_obd->obd_unlinked_exports); - cfs_list_del_init(&exp->exp_obd_chain_timed); - exp->exp_obd->obd_num_exports--; - cfs_spin_unlock(&exp->exp_obd->obd_dev_lock); - class_export_put(exp); + cfs_list_move(&exp->exp_obd_chain, &exp->exp_obd->obd_unlinked_exports); + cfs_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); } EXPORT_SYMBOL(class_unlink_export); @@ -884,7 +954,7 @@ void class_import_destroy(struct obd_import *imp) CDEBUG(D_IOCTL, "destroying import %p for %s\n", imp, imp->imp_obd->obd_name); - LASSERT(cfs_atomic_read(&imp->imp_refcount) == 0); + LASSERT_ATOMIC_ZERO(&imp->imp_refcount); ptlrpc_put_connection_superhack(imp->imp_connection); @@ -909,10 +979,13 @@ static void import_handle_addref(void *import) class_import_get(import); } +static struct portals_handle_ops import_handle_ops = { + .hop_addref = import_handle_addref, + .hop_free = NULL, +}; + struct obd_import *class_import_get(struct obd_import *import) { - LASSERT(cfs_atomic_read(&import->imp_refcount) >= 0); - LASSERT(cfs_atomic_read(&import->imp_refcount) < 0x5a5a5a); cfs_atomic_inc(&import->imp_refcount); CDEBUG(D_INFO, "import %p refcount=%d obd=%s\n", import, cfs_atomic_read(&import->imp_refcount), @@ -925,9 +998,8 @@ void class_import_put(struct obd_import *imp) { ENTRY; - LASSERT(cfs_atomic_read(&imp->imp_refcount) > 0); - LASSERT(cfs_atomic_read(&imp->imp_refcount) < 0x5a5a5a); LASSERT(cfs_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, cfs_atomic_read(&imp->imp_refcount) - 1, @@ -938,7 +1010,9 @@ void class_import_put(struct obd_import *imp) obd_zombie_import_add(imp); } - EXIT; + /* catch possible import put race */ + LASSERT_ATOMIC_GE_LT(&imp->imp_refcount, 0, LI_POISON); + EXIT; } EXPORT_SYMBOL(class_import_put); @@ -962,15 +1036,16 @@ struct obd_import *class_new_import(struct obd_device *obd) if (imp == NULL) return NULL; + CFS_INIT_LIST_HEAD(&imp->imp_pinger_chain); CFS_INIT_LIST_HEAD(&imp->imp_zombie_chain); CFS_INIT_LIST_HEAD(&imp->imp_replay_list); CFS_INIT_LIST_HEAD(&imp->imp_sending_list); CFS_INIT_LIST_HEAD(&imp->imp_delayed_list); - cfs_spin_lock_init(&imp->imp_lock); - imp->imp_last_success_conn = 0; - imp->imp_state = LUSTRE_IMP_NEW; - imp->imp_obd = class_incref(obd, "import", imp); - cfs_sema_init(&imp->imp_sec_mutex, 1); + spin_lock_init(&imp->imp_lock); + imp->imp_last_success_conn = 0; + imp->imp_state = LUSTRE_IMP_NEW; + imp->imp_obd = class_incref(obd, "import", imp); + mutex_init(&imp->imp_sec_mutex); cfs_waitq_init(&imp->imp_recovery_waitq); cfs_atomic_set(&imp->imp_refcount, 2); @@ -980,7 +1055,7 @@ struct obd_import *class_new_import(struct obd_device *obd) cfs_atomic_set(&imp->imp_inval_count, 0); CFS_INIT_LIST_HEAD(&imp->imp_conn_list); CFS_INIT_LIST_HEAD(&imp->imp_handle.h_link); - class_handle_hash(&imp->imp_handle, import_handle_addref); + class_handle_hash(&imp->imp_handle, &import_handle_ops); init_imp_at(&imp->imp_at); /* the default magic is V2, will be used in connect RPC, and @@ -993,15 +1068,15 @@ EXPORT_SYMBOL(class_new_import); void class_destroy_import(struct obd_import *import) { - LASSERT(import != NULL); - LASSERT(import != LP_POISON); + LASSERT(import != NULL); + LASSERT(import != LP_POISON); - class_handle_unhash(&import->imp_handle); + class_handle_unhash(&import->imp_handle); - cfs_spin_lock(&import->imp_lock); - import->imp_generation++; - cfs_spin_unlock(&import->imp_lock); - class_import_put(import); + spin_lock(&import->imp_lock); + import->imp_generation++; + spin_unlock(&import->imp_lock); + class_import_put(import); } EXPORT_SYMBOL(class_destroy_import); @@ -1009,7 +1084,7 @@ EXPORT_SYMBOL(class_destroy_import); void __class_export_add_lock_ref(struct obd_export *exp, struct ldlm_lock *lock) { - cfs_spin_lock(&exp->exp_locks_list_guard); + spin_lock(&exp->exp_locks_list_guard); LASSERT(lock->l_exp_refs_nr >= 0); @@ -1024,13 +1099,13 @@ void __class_export_add_lock_ref(struct obd_export *exp, struct ldlm_lock *lock) } CDEBUG(D_INFO, "lock = %p, export = %p, refs = %u\n", lock, exp, lock->l_exp_refs_nr); - cfs_spin_unlock(&exp->exp_locks_list_guard); + spin_unlock(&exp->exp_locks_list_guard); } EXPORT_SYMBOL(__class_export_add_lock_ref); void __class_export_del_lock_ref(struct obd_export *exp, struct ldlm_lock *lock) { - cfs_spin_lock(&exp->exp_locks_list_guard); + spin_lock(&exp->exp_locks_list_guard); LASSERT(lock->l_exp_refs_nr > 0); if (lock->l_exp_refs_target != exp) { LCONSOLE_WARN("lock %p, " @@ -1043,7 +1118,7 @@ void __class_export_del_lock_ref(struct obd_export *exp, struct ldlm_lock *lock) } CDEBUG(D_INFO, "lock = %p, export = %p, refs = %u\n", lock, exp, lock->l_exp_refs_nr); - cfs_spin_unlock(&exp->exp_locks_list_guard); + spin_unlock(&exp->exp_locks_list_guard); } EXPORT_SYMBOL(__class_export_del_lock_ref); #endif @@ -1077,35 +1152,44 @@ EXPORT_SYMBOL(class_connect); /* if export is involved in recovery then clean up related things */ void class_export_recovery_cleanup(struct obd_export *exp) { - struct obd_device *obd = exp->exp_obd; - - cfs_spin_lock_bh(&obd->obd_processing_task_lock); - if (exp->exp_delayed) - obd->obd_delayed_clients--; - if (obd->obd_recovering && exp->exp_in_recovery) { - cfs_spin_lock(&exp->exp_lock); - exp->exp_in_recovery = 0; - cfs_spin_unlock(&exp->exp_lock); - LASSERT(obd->obd_connected_clients); - obd->obd_connected_clients--; - } - /** Cleanup req replay fields */ - if (exp->exp_req_replay_needed) { - cfs_spin_lock(&exp->exp_lock); - exp->exp_req_replay_needed = 0; - cfs_spin_unlock(&exp->exp_lock); - LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients)); - cfs_atomic_dec(&obd->obd_req_replay_clients); - } - /** Cleanup lock replay data */ - if (exp->exp_lock_replay_needed) { - cfs_spin_lock(&exp->exp_lock); - exp->exp_lock_replay_needed = 0; - cfs_spin_unlock(&exp->exp_lock); - LASSERT(cfs_atomic_read(&obd->obd_lock_replay_clients)); - cfs_atomic_dec(&obd->obd_lock_replay_clients); - } - cfs_spin_unlock_bh(&obd->obd_processing_task_lock); + struct obd_device *obd = exp->exp_obd; + + spin_lock(&obd->obd_recovery_task_lock); + if (exp->exp_delayed) + obd->obd_delayed_clients--; + if (obd->obd_recovering) { + if (exp->exp_in_recovery) { + spin_lock(&exp->exp_lock); + exp->exp_in_recovery = 0; + spin_unlock(&exp->exp_lock); + LASSERT_ATOMIC_POS(&obd->obd_connected_clients); + cfs_atomic_dec(&obd->obd_connected_clients); + } + + /* if called during recovery then should update + * obd_stale_clients counter, + * lightweight exports are not counted */ + if (exp->exp_failed && + (exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) == 0) + exp->exp_obd->obd_stale_clients++; + } + spin_unlock(&obd->obd_recovery_task_lock); + /** Cleanup req replay fields */ + if (exp->exp_req_replay_needed) { + spin_lock(&exp->exp_lock); + exp->exp_req_replay_needed = 0; + spin_unlock(&exp->exp_lock); + LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients)); + cfs_atomic_dec(&obd->obd_req_replay_clients); + } + /** Cleanup lock replay data */ + if (exp->exp_lock_replay_needed) { + spin_lock(&exp->exp_lock); + exp->exp_lock_replay_needed = 0; + spin_unlock(&exp->exp_lock); + LASSERT(cfs_atomic_read(&obd->obd_lock_replay_clients)); + cfs_atomic_dec(&obd->obd_lock_replay_clients); + } } /* This function removes 1-3 references from the export: @@ -1120,15 +1204,14 @@ int class_disconnect(struct obd_export *export) ENTRY; if (export == NULL) { - fixme(); - CDEBUG(D_IOCTL, "attempting to free NULL export %p\n", export); + CWARN("attempting to free NULL export %p\n", export); RETURN(-EINVAL); } - cfs_spin_lock(&export->exp_lock); - already_disconnected = export->exp_disconnected; - export->exp_disconnected = 1; - cfs_spin_unlock(&export->exp_lock); + spin_lock(&export->exp_lock); + already_disconnected = export->exp_disconnected; + export->exp_disconnected = 1; + spin_unlock(&export->exp_lock); /* class_cleanup(), abort_recovery(), and class_fail_export() * all end up in here, and if any of them race we shouldn't @@ -1152,18 +1235,19 @@ no_disconn: class_export_put(export); RETURN(0); } +EXPORT_SYMBOL(class_disconnect); /* Return non-zero for a fully connected export */ int class_connected_export(struct obd_export *exp) { - if (exp) { - int connected; - cfs_spin_lock(&exp->exp_lock); - connected = (exp->exp_conn_cnt > 0); - cfs_spin_unlock(&exp->exp_lock); - return connected; - } - return 0; + if (exp) { + int connected; + spin_lock(&exp->exp_lock); + connected = (exp->exp_conn_cnt > 0); + spin_unlock(&exp->exp_lock); + return connected; + } + return 0; } EXPORT_SYMBOL(class_connected_export); @@ -1182,9 +1266,9 @@ static void class_disconnect_export_list(cfs_list_t *list, /* need for safe call CDEBUG after obd_disconnect */ class_export_get(exp); - cfs_spin_lock(&exp->exp_lock); - exp->exp_flags = flags; - cfs_spin_unlock(&exp->exp_lock); + spin_lock(&exp->exp_lock); + exp->exp_flags = flags; + spin_unlock(&exp->exp_lock); if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid)) { @@ -1215,15 +1299,15 @@ static void class_disconnect_export_list(cfs_list_t *list, void class_disconnect_exports(struct obd_device *obd) { - cfs_list_t work_list; - ENTRY; + cfs_list_t work_list; + ENTRY; - /* Move all of the exports from obd_exports to a work list, en masse. */ - CFS_INIT_LIST_HEAD(&work_list); - cfs_spin_lock(&obd->obd_dev_lock); - cfs_list_splice_init(&obd->obd_exports, &work_list); - cfs_list_splice_init(&obd->obd_delayed_exports, &work_list); - cfs_spin_unlock(&obd->obd_dev_lock); + /* Move all of the exports from obd_exports to a work list, en masse. */ + CFS_INIT_LIST_HEAD(&work_list); + spin_lock(&obd->obd_dev_lock); + cfs_list_splice_init(&obd->obd_exports, &work_list); + cfs_list_splice_init(&obd->obd_delayed_exports, &work_list); + spin_unlock(&obd->obd_dev_lock); if (!cfs_list_empty(&work_list)) { CDEBUG(D_HA, "OBD device %d (%p) has exports, " @@ -1243,52 +1327,60 @@ void class_disconnect_stale_exports(struct obd_device *obd, int (*test_export)(struct obd_export *)) { cfs_list_t work_list; - cfs_list_t *pos, *n; - struct obd_export *exp; + struct obd_export *exp, *n; int evicted = 0; ENTRY; CFS_INIT_LIST_HEAD(&work_list); - cfs_spin_lock(&obd->obd_dev_lock); - cfs_list_for_each_safe(pos, n, &obd->obd_exports) { - exp = cfs_list_entry(pos, struct obd_export, exp_obd_chain); - if (test_export(exp)) - continue; - + spin_lock(&obd->obd_dev_lock); + cfs_list_for_each_entry_safe(exp, n, &obd->obd_exports, + exp_obd_chain) { /* don't count self-export as client */ if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid)) continue; + /* don't evict clients which have no slot in last_rcvd + * (e.g. lightweight connection) */ + if (exp->exp_target_data.ted_lr_idx == -1) + continue; + + spin_lock(&exp->exp_lock); + if (exp->exp_failed || test_export(exp)) { + spin_unlock(&exp->exp_lock); + continue; + } + exp->exp_failed = 1; + spin_unlock(&exp->exp_lock); + cfs_list_move(&exp->exp_obd_chain, &work_list); evicted++; - CDEBUG(D_ERROR, "%s: disconnect stale client %s@%s\n", + 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); } - cfs_spin_unlock(&obd->obd_dev_lock); + spin_unlock(&obd->obd_dev_lock); - if (evicted) { - CDEBUG(D_HA, "%s: disconnecting %d stale clients\n", - obd->obd_name, evicted); - obd->obd_stale_clients += evicted; - } - class_disconnect_export_list(&work_list, exp_flags_from_obd(obd) | - OBD_OPT_ABORT_RECOV); - EXIT; + if (evicted) + LCONSOLE_WARN("%s: disconnecting %d stale clients\n", + obd->obd_name, evicted); + + class_disconnect_export_list(&work_list, exp_flags_from_obd(obd) | + OBD_OPT_ABORT_RECOV); + EXIT; } EXPORT_SYMBOL(class_disconnect_stale_exports); void class_fail_export(struct obd_export *exp) { - int rc, already_failed; + int rc, already_failed; - cfs_spin_lock(&exp->exp_lock); - already_failed = exp->exp_failed; - exp->exp_failed = 1; - cfs_spin_unlock(&exp->exp_lock); + spin_lock(&exp->exp_lock); + already_failed = exp->exp_failed; + exp->exp_failed = 1; + spin_unlock(&exp->exp_lock); if (already_failed) { CDEBUG(D_HA, "disconnecting dead export %p/%s; skipping\n", @@ -1302,6 +1394,9 @@ void class_fail_export(struct obd_export *exp) if (obd_dump_on_timeout) libcfs_debug_dumplog(); + /* need for safe call CDEBUG after obd_disconnect */ + class_export_get(exp); + /* Most callers into obd_disconnect are removing their own reference * (request, for example) in addition to the one from the hash table. * We don't have such a reference here, so make one. */ @@ -1312,6 +1407,7 @@ void class_fail_export(struct obd_export *exp) else CDEBUG(D_HA, "disconnected export %p/%s\n", exp, exp->exp_client_uuid.uuid); + class_export_put(exp); } EXPORT_SYMBOL(class_fail_export); @@ -1326,13 +1422,25 @@ EXPORT_SYMBOL(obd_export_nid2str); int obd_export_evict_by_nid(struct obd_device *obd, const char *nid) { - struct obd_export *doomed_exp = NULL; - int exports_evicted = 0; - - lnet_nid_t nid_key = libcfs_str2nid((char *)nid); - - do { - doomed_exp = cfs_hash_lookup(obd->obd_nid_hash, &nid_key); + cfs_hash_t *nid_hash; + struct obd_export *doomed_exp = NULL; + int exports_evicted = 0; + + lnet_nid_t nid_key = libcfs_str2nid((char *)nid); + + spin_lock(&obd->obd_dev_lock); + /* umount has run already, so evict thread should leave + * its task to umount thread now */ + if (obd->obd_stopping) { + spin_unlock(&obd->obd_dev_lock); + return exports_evicted; + } + nid_hash = obd->obd_nid_hash; + cfs_hash_getref(nid_hash); + spin_unlock(&obd->obd_dev_lock); + + do { + doomed_exp = cfs_hash_lookup(nid_hash, &nid_key); if (doomed_exp == NULL) break; @@ -1343,13 +1451,16 @@ int obd_export_evict_by_nid(struct obd_device *obd, const char *nid) LASSERTF(doomed_exp != obd->obd_self_export, "self-export is hashed by NID?\n"); exports_evicted++; - CWARN("%s: evict NID '%s' (%s) #%d at adminstrative request\n", - obd->obd_name, nid, doomed_exp->exp_client_uuid.uuid, - exports_evicted); + LCONSOLE_WARN("%s: evicting %s (at %s) by administrative " + "request\n", obd->obd_name, + obd_uuid2str(&doomed_exp->exp_client_uuid), + obd_export_nid2str(doomed_exp)); class_fail_export(doomed_exp); class_export_put(doomed_exp); } while (1); + cfs_hash_putref(nid_hash); + if (!exports_evicted) CDEBUG(D_HA,"%s: can't disconnect NID '%s': no exports found\n", obd->obd_name, nid); @@ -1359,17 +1470,28 @@ EXPORT_SYMBOL(obd_export_evict_by_nid); int obd_export_evict_by_uuid(struct obd_device *obd, const char *uuid) { - struct obd_export *doomed_exp = NULL; - struct obd_uuid doomed_uuid; - int exports_evicted = 0; + cfs_hash_t *uuid_hash; + struct obd_export *doomed_exp = NULL; + struct obd_uuid doomed_uuid; + int exports_evicted = 0; + + spin_lock(&obd->obd_dev_lock); + if (obd->obd_stopping) { + spin_unlock(&obd->obd_dev_lock); + return exports_evicted; + } + uuid_hash = obd->obd_uuid_hash; + cfs_hash_getref(uuid_hash); + spin_unlock(&obd->obd_dev_lock); obd_str2uuid(&doomed_uuid, uuid); if (obd_uuid_equals(&doomed_uuid, &obd->obd_uuid)) { CERROR("%s: can't evict myself\n", obd->obd_name); + cfs_hash_putref(uuid_hash); return exports_evicted; } - doomed_exp = cfs_hash_lookup(obd->obd_uuid_hash, &doomed_uuid); + doomed_exp = cfs_hash_lookup(uuid_hash, &doomed_uuid); if (doomed_exp == NULL) { CERROR("%s: can't disconnect %s: no exports found\n", @@ -1381,6 +1503,7 @@ int obd_export_evict_by_uuid(struct obd_device *obd, const char *uuid) class_export_put(doomed_exp); exports_evicted++; } + cfs_hash_putref(uuid_hash); return exports_evicted; } @@ -1392,20 +1515,20 @@ EXPORT_SYMBOL(class_export_dump_hook); #endif static void print_export_data(struct obd_export *exp, const char *status, - int locks) -{ - struct ptlrpc_reply_state *rs; - struct ptlrpc_reply_state *first_reply = NULL; - int nreplies = 0; - - cfs_spin_lock(&exp->exp_lock); - cfs_list_for_each_entry(rs, &exp->exp_outstanding_replies, - rs_exp_list) { - if (nreplies == 0) - first_reply = rs; - nreplies++; - } - cfs_spin_unlock(&exp->exp_lock); + int locks) +{ + struct ptlrpc_reply_state *rs; + struct ptlrpc_reply_state *first_reply = NULL; + int nreplies = 0; + + spin_lock(&exp->exp_lock); + cfs_list_for_each_entry(rs, &exp->exp_outstanding_replies, + rs_exp_list) { + if (nreplies == 0) + first_reply = rs; + nreplies++; + } + 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, @@ -1426,28 +1549,28 @@ void dump_exports(struct obd_device *obd, int locks) { struct obd_export *exp; - cfs_spin_lock(&obd->obd_dev_lock); - cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) - print_export_data(exp, "ACTIVE", locks); - cfs_list_for_each_entry(exp, &obd->obd_unlinked_exports, exp_obd_chain) - print_export_data(exp, "UNLINKED", locks); - cfs_list_for_each_entry(exp, &obd->obd_delayed_exports, exp_obd_chain) - print_export_data(exp, "DELAYED", locks); - cfs_spin_unlock(&obd->obd_dev_lock); - cfs_spin_lock(&obd_zombie_impexp_lock); - cfs_list_for_each_entry(exp, &obd_zombie_exports, exp_obd_chain) - print_export_data(exp, "ZOMBIE", locks); - cfs_spin_unlock(&obd_zombie_impexp_lock); + spin_lock(&obd->obd_dev_lock); + cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) + print_export_data(exp, "ACTIVE", locks); + cfs_list_for_each_entry(exp, &obd->obd_unlinked_exports, exp_obd_chain) + print_export_data(exp, "UNLINKED", locks); + cfs_list_for_each_entry(exp, &obd->obd_delayed_exports, exp_obd_chain) + print_export_data(exp, "DELAYED", locks); + spin_unlock(&obd->obd_dev_lock); + spin_lock(&obd_zombie_impexp_lock); + cfs_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) { - int waited = 2; - LASSERT(cfs_list_empty(&obd->obd_exports)); - cfs_spin_lock(&obd->obd_dev_lock); - while (!cfs_list_empty(&obd->obd_unlinked_exports)) { - cfs_spin_unlock(&obd->obd_dev_lock); + int waited = 2; + LASSERT(cfs_list_empty(&obd->obd_exports)); + spin_lock(&obd->obd_dev_lock); + while (!cfs_list_empty(&obd->obd_unlinked_exports)) { + spin_unlock(&obd->obd_dev_lock); cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT, cfs_time_seconds(waited)); if (waited > 5 && IS_PO2(waited)) { @@ -1459,23 +1582,26 @@ void obd_exports_barrier(struct obd_device *obd) dump_exports(obd, 1); } waited *= 2; - cfs_spin_lock(&obd->obd_dev_lock); - } - cfs_spin_unlock(&obd->obd_dev_lock); + spin_lock(&obd->obd_dev_lock); + } + spin_unlock(&obd->obd_dev_lock); } 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; + struct obd_import *import; + struct obd_export *export; + ENTRY; - do { - cfs_spin_lock(&obd_zombie_impexp_lock); + do { + spin_lock(&obd_zombie_impexp_lock); import = NULL; if (!cfs_list_empty(&obd_zombie_imports)) { @@ -1493,27 +1619,35 @@ void obd_zombie_impexp_cull(void) cfs_list_del_init(&export->exp_obd_chain); } - cfs_spin_unlock(&obd_zombie_impexp_lock); + spin_unlock(&obd_zombie_impexp_lock); - if (import != NULL) - class_import_destroy(import); + 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); + if (export != NULL) { + class_export_destroy(export); + spin_lock(&obd_zombie_impexp_lock); + zombies_count--; + spin_unlock(&obd_zombie_impexp_lock); + } - cfs_cond_resched(); - } while (import != NULL || export != NULL); - EXIT; + cfs_cond_resched(); + } while (import != NULL || export != NULL); + EXIT; } -static cfs_completion_t obd_zombie_start; -static cfs_completion_t obd_zombie_stop; -static unsigned long obd_zombie_flags; -static cfs_waitq_t obd_zombie_waitq; -static pid_t obd_zombie_pid; +static struct completion obd_zombie_start; +static struct completion obd_zombie_stop; +static unsigned long obd_zombie_flags; +static cfs_waitq_t obd_zombie_waitq; +static pid_t obd_zombie_pid; enum { - OBD_ZOMBIE_STOP = 1 << 1 + OBD_ZOMBIE_STOP = 0x0001, }; /** @@ -1521,46 +1655,45 @@ enum { */ static int obd_zombie_impexp_check(void *arg) { - int rc; + int rc; - cfs_spin_lock(&obd_zombie_impexp_lock); - rc = cfs_list_empty(&obd_zombie_imports) && - cfs_list_empty(&obd_zombie_exports) && - !cfs_test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags); + spin_lock(&obd_zombie_impexp_lock); + rc = (zombies_count == 0) && + !test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags); + spin_unlock(&obd_zombie_impexp_lock); - cfs_spin_unlock(&obd_zombie_impexp_lock); - - RETURN(rc); + RETURN(rc); } /** * Add export to the obd_zombe thread and notify it. */ static void obd_zombie_export_add(struct obd_export *exp) { - cfs_spin_lock(&exp->exp_obd->obd_dev_lock); - LASSERT(!cfs_list_empty(&exp->exp_obd_chain)); - cfs_list_del_init(&exp->exp_obd_chain); - cfs_spin_unlock(&exp->exp_obd->obd_dev_lock); - cfs_spin_lock(&obd_zombie_impexp_lock); - cfs_list_add(&exp->exp_obd_chain, &obd_zombie_exports); - cfs_spin_unlock(&obd_zombie_impexp_lock); + spin_lock(&exp->exp_obd->obd_dev_lock); + LASSERT(!cfs_list_empty(&exp->exp_obd_chain)); + cfs_list_del_init(&exp->exp_obd_chain); + spin_unlock(&exp->exp_obd->obd_dev_lock); + spin_lock(&obd_zombie_impexp_lock); + zombies_count++; + cfs_list_add(&exp->exp_obd_chain, &obd_zombie_exports); + spin_unlock(&obd_zombie_impexp_lock); - if (obd_zombie_impexp_notify != NULL) - obd_zombie_impexp_notify(); + obd_zombie_impexp_notify(); } /** * Add import to the obd_zombe thread and notify it. */ static void obd_zombie_import_add(struct obd_import *imp) { - LASSERT(imp->imp_sec == NULL); - cfs_spin_lock(&obd_zombie_impexp_lock); - LASSERT(cfs_list_empty(&imp->imp_zombie_chain)); - cfs_list_add(&imp->imp_zombie_chain, &obd_zombie_imports); - cfs_spin_unlock(&obd_zombie_impexp_lock); + LASSERT(imp->imp_sec == NULL); + LASSERT(imp->imp_rq_pool == NULL); + spin_lock(&obd_zombie_impexp_lock); + LASSERT(cfs_list_empty(&imp->imp_zombie_chain)); + zombies_count++; + cfs_list_add(&imp->imp_zombie_chain, &obd_zombie_imports); + spin_unlock(&obd_zombie_impexp_lock); - if (obd_zombie_impexp_notify != NULL) - obd_zombie_impexp_notify(); + obd_zombie_impexp_notify(); } /** @@ -1568,7 +1701,12 @@ static void obd_zombie_import_add(struct obd_import *imp) { */ static void obd_zombie_impexp_notify(void) { - cfs_waitq_signal(&obd_zombie_waitq); + /* + * 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 + */ + cfs_waitq_broadcast(&obd_zombie_waitq); } /** @@ -1576,14 +1714,13 @@ static void obd_zombie_impexp_notify(void) */ static int obd_zombie_is_idle(void) { - int rc; + int rc; - LASSERT(!cfs_test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags)); - cfs_spin_lock(&obd_zombie_impexp_lock); - rc = cfs_list_empty(&obd_zombie_imports) && - cfs_list_empty(&obd_zombie_exports); - cfs_spin_unlock(&obd_zombie_impexp_lock); - return 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; } /** @@ -1607,18 +1744,19 @@ EXPORT_SYMBOL(obd_zombie_barrier); */ static int obd_zombie_impexp_thread(void *unused) { - int rc; + int rc; - if ((rc = cfs_daemonize_ctxt("obd_zombid"))) { - cfs_complete(&obd_zombie_start); - RETURN(rc); - } + rc = cfs_daemonize_ctxt("obd_zombid"); + if (rc != 0) { + complete(&obd_zombie_start); + RETURN(rc); + } - cfs_complete(&obd_zombie_start); + complete(&obd_zombie_start); - obd_zombie_pid = cfs_curproc_pid(); + obd_zombie_pid = cfs_curproc_pid(); - while(!cfs_test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags)) { + while (!test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags)) { struct l_wait_info lwi = { 0 }; l_wait_event(obd_zombie_waitq, @@ -1632,9 +1770,9 @@ static int obd_zombie_impexp_thread(void *unused) cfs_waitq_signal(&obd_zombie_waitq); } - cfs_complete(&obd_zombie_stop); + complete(&obd_zombie_stop); - RETURN(0); + RETURN(0); } #else /* ! KERNEL */ @@ -1662,22 +1800,22 @@ int obd_zombie_impexp_kill(void *arg) */ int obd_zombie_impexp_init(void) { - int rc; + int rc; - CFS_INIT_LIST_HEAD(&obd_zombie_imports); - CFS_INIT_LIST_HEAD(&obd_zombie_exports); - cfs_spin_lock_init(&obd_zombie_impexp_lock); - cfs_init_completion(&obd_zombie_start); - cfs_init_completion(&obd_zombie_stop); - cfs_waitq_init(&obd_zombie_waitq); - obd_zombie_pid = 0; + CFS_INIT_LIST_HEAD(&obd_zombie_imports); + CFS_INIT_LIST_HEAD(&obd_zombie_exports); + spin_lock_init(&obd_zombie_impexp_lock); + init_completion(&obd_zombie_start); + init_completion(&obd_zombie_stop); + cfs_waitq_init(&obd_zombie_waitq); + obd_zombie_pid = 0; #ifdef __KERNEL__ - rc = cfs_kernel_thread(obd_zombie_impexp_thread, NULL, 0); - if (rc < 0) - RETURN(rc); + rc = cfs_create_thread(obd_zombie_impexp_thread, NULL, 0); + if (rc < 0) + RETURN(rc); - cfs_wait_for_completion(&obd_zombie_start); + wait_for_completion(&obd_zombie_start); #else obd_zombie_impexp_work_cb = @@ -1696,13 +1834,80 @@ int obd_zombie_impexp_init(void) */ void obd_zombie_impexp_stop(void) { - cfs_set_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags); + set_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags); obd_zombie_impexp_notify(); #ifdef __KERNEL__ - cfs_wait_for_completion(&obd_zombie_stop); + wait_for_completion(&obd_zombie_stop); #else liblustre_deregister_wait_callback(obd_zombie_impexp_work_cb); liblustre_deregister_idle_callback(obd_zombie_impexp_idle_cb); #endif } +/***** Kernel-userspace comm helpers *******/ + +/* Get length of entire message, including header */ +int kuc_len(int payload_len) +{ + return sizeof(struct kuc_hdr) + payload_len; +} +EXPORT_SYMBOL(kuc_len); + +/* Get a pointer to kuc header, given a ptr to the payload + * @param p Pointer to payload area + * @returns Pointer to kuc header + */ +struct kuc_hdr * kuc_ptr(void *p) +{ + struct kuc_hdr *lh = ((struct kuc_hdr *)p) - 1; + LASSERT(lh->kuc_magic == KUC_MAGIC); + return lh; +} +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 + */ +void *kuc_alloc(int payload_len, int transport, int type) +{ + struct kuc_hdr *lh; + int len = kuc_len(payload_len); + + OBD_ALLOC(lh, len); + if (lh == NULL) + return ERR_PTR(-ENOMEM); + + lh->kuc_magic = KUC_MAGIC; + lh->kuc_transport = transport; + lh->kuc_msgtype = type; + lh->kuc_msglen = len; + + return (void *)(lh + 1); +} +EXPORT_SYMBOL(kuc_alloc); + +/* Takes pointer to payload area */ +inline void kuc_free(void *p, int payload_len) +{ + struct kuc_hdr *lh = kuc_ptr(p); + OBD_FREE(lh, kuc_len(payload_len)); +} +EXPORT_SYMBOL(kuc_free); + + +