X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flod%2Flod_lov.c;h=a116ae2200cfc266bd9b6555458c20077e4a60a0;hp=e3cd87a92c8c5956e2550ba9164ec4b12ba68cea;hb=dd745f94bf0fac0d4042f46c8faba1b61d3b9fb3;hpb=60e07b972114df24105a3a1bfa7365892f72a4a7 diff --git a/lustre/lod/lod_lov.c b/lustre/lod/lod_lov.c index e3cd87a..a116ae2 100644 --- a/lustre/lod/lod_lov.c +++ b/lustre/lod/lod_lov.c @@ -6,16 +6,16 @@ * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 only, * as published by the Free Software Foundation. - + * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License version 2 for more details. A copy is * included in the COPYING file that accompanied this code. - + * * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * version 2 along with this program; If not, see + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -23,27 +23,33 @@ * Copyright 2009 Sun Microsystems, Inc. All rights reserved * Use is subject to license terms. * - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2012, 2014, Intel Corporation. */ /* * lustre/lod/lod_lov.c * - * Author: Alex Zhuravlev + * A set of helpers to maintain Logical Object Volume (LOV) + * Extended Attribute (EA) and known OST targets + * + * Author: Alex Zhuravlev */ #define DEBUG_SUBSYSTEM S_MDS #include #include +#include #include "lod_internal.h" -/* - * Keep a refcount of lod->ltd_tgts usage to prevent racing with - * addition/deletion. Any function that expects lov_tgts to remain stationary - * must take a ref. +/** + * Increase reference count on the target table. * - * \param lod - is the lod device from which we want to grab a reference + * Increase reference count on the target table usage to prevent racing with + * addition/deletion. Any function that expects the table to remain + * stationary must take a ref. + * + * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs) */ void lod_getref(struct lod_tgt_descs *ltd) { @@ -53,12 +59,15 @@ void lod_getref(struct lod_tgt_descs *ltd) mutex_unlock(<d->ltd_mutex); } -/* - * Companion of lod_getref() to release a reference on the lod table. - * If this is the last reference and the ost entry was scheduled for deletion, - * the descriptor is removed from the array. +/** + * Decrease reference count on the target table. + * + * Companion of lod_getref() to release a reference on the target table. + * If this is the last reference and the OST entry was scheduled for deletion, + * the descriptor is removed from the table. * - * \param lod - is the lod device from which we release a reference + * \param[in] lod LOD device from which we release a reference + * \param[in] ltd target table (lod_ost_descs or lod_mdt_descs) */ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd) { @@ -66,12 +75,14 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd) ltd->ltd_refcount--; if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) { struct lod_tgt_desc *tgt_desc, *tmp; - int idx; - CFS_LIST_HEAD(kill); + struct list_head kill; + unsigned int idx; CDEBUG(D_CONFIG, "destroying %d ltd desc\n", ltd->ltd_death_row); + INIT_LIST_HEAD(&kill); + cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) { tgt_desc = LTD_TGT(ltd, idx); LASSERT(tgt_desc); @@ -79,7 +90,7 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd) if (!tgt_desc->ltd_reap) continue; - cfs_list_add(&tgt_desc->ltd_kill, &kill); + list_add(&tgt_desc->ltd_kill, &kill); LTD_TGT(ltd, idx) = NULL; /*FIXME: only support ost pool for now */ if (ltd == &lod->lod_ost_descs) { @@ -94,9 +105,9 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd) mutex_unlock(<d->ltd_mutex); up_read(<d->ltd_rw_sem); - cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) { + list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) { int rc; - cfs_list_del(&tgt_desc->ltd_kill); + list_del(&tgt_desc->ltd_kill); if (ltd == &lod->lod_ost_descs) { /* remove from QoS structures */ rc = qos_del_tgt(lod, tgt_desc); @@ -120,6 +131,20 @@ void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd) } } +/** + * Expand size of target table. + * + * When the target table is full, we have to extend the table. To do so, + * we allocate new memory with some reserve, move data from the old table + * to the new one and release memory consumed by the old table. + * Notice we take ltd_rw_sem exclusively to ensure atomic switch. + * + * \param[in] ltd target table + * \param[in] newsize new size of the table + * + * \retval 0 on success + * \retval -ENOMEM if reallocation failed + */ static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize) { cfs_bitmap_t *new_bitmap, *old_bitmap = NULL; @@ -160,16 +185,23 @@ out: return rc; } -/* - * Connect LOD to a new OSP and add it to the device table. +/** + * Connect LOD to a new OSP and add it to the target table. + * + * Connect to the OSP device passed, initialize all the internal + * structures related to the device and add it to the target table. + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device to be connected to the new OSP + * \param[in] osp name of OSP device name to be added + * \param[in] index index of the new target + * \param[in] gen target's generation number + * \param[in] tgt_index OSP's group + * \param[in] type type of device (mdc or osc) + * \param[in] active state of OSP: 0 - inactive, 1 - active * - * \param env - is the environment passed by the caller - * \param lod - is the LOD device to be connected to the new OSP - * \param osp - is the name of OSP device name about to be added - * \param index - is the OSP index - * \param gen - is the generation number - * \param tgt_index - is the group of the OSP. - * \param type - is the type of device (mdc or osc) + * \retval 0 if added successfully + * \retval negative error number on failure */ int lod_add_device(const struct lu_env *env, struct lod_device *lod, char *osp, unsigned index, unsigned gen, int tgt_index, @@ -185,6 +217,7 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, struct lod_tgt_descs *ltd; struct obd_uuid obd_uuid; bool for_ost; + bool lock = false; ENTRY; CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen); @@ -221,14 +254,14 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, OBD_CONNECT_LRU_RESIZE | #endif OBD_CONNECT_MDS | - OBD_CONNECT_OSS_CAPA | OBD_CONNECT_REQPORTAL | OBD_CONNECT_SKIP_ORPHAN | OBD_CONNECT_FID | OBD_CONNECT_LVB_TYPE | OBD_CONNECT_VERSION | OBD_CONNECT_PINGLESS | - OBD_CONNECT_LFSCK; + OBD_CONNECT_LFSCK | + OBD_CONNECT_BULK_MBITS; data->ocd_group = tgt_index; ltd = &lod->lod_ost_descs; @@ -238,17 +271,13 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, for_ost = false; data->ocd_ibits_known = MDS_INODELOCK_UPDATE; data->ocd_connect_flags |= OBD_CONNECT_ACL | - OBD_CONNECT_MDS_CAPA | - OBD_CONNECT_OSS_CAPA | OBD_CONNECT_IBITS | OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID | OBD_CONNECT_AT | OBD_CONNECT_FULL20 | - OBD_CONNECT_LFSCK; - /* XXX set MDS-MDS flags, remove this when running this - * on client*/ - data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS; + OBD_CONNECT_LFSCK | + OBD_CONNECT_BULK_MBITS; spin_lock(&imp->imp_lock); imp->imp_server_timeout = 1; spin_unlock(&imp->imp_lock); @@ -305,6 +334,7 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, } mutex_lock(<d->ltd_mutex); + lock = true; if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) { CERROR("%s: device %d is registered already\n", obd->obd_name, index); @@ -320,7 +350,7 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, } } - if (!strcmp(LUSTRE_OSC_NAME, type)) { + if (for_ost) { /* pool and qos are not supported for MDS stack yet */ rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size); @@ -349,21 +379,49 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod, ltd->ltd_tgtnr++; mutex_unlock(<d->ltd_mutex); lod_putref(lod, ltd); + lock = false; if (lod->lod_recovery_completed) ldev->ld_ops->ldo_recovery_complete(env, ldev); + if (!for_ost && lod->lod_initialized) { + rc = lod_sub_init_llog(env, lod, tgt_desc->ltd_tgt); + if (rc != 0) { + CERROR("%s: cannot start llog on %s:rc = %d\n", + lod2obd(lod)->obd_name, osp, rc); + GOTO(out_ltd, rc); + } + } + rc = lfsck_add_target(env, lod->lod_child, d, exp, index, for_ost); - if (rc != 0) + if (rc != 0) { CERROR("Fail to add LFSCK target: name = %s, type = %s, " "index = %u, rc = %d\n", osp, type, index, rc); - + GOTO(out_fini_llog, rc); + } RETURN(rc); +out_fini_llog: + lod_sub_fini_llog(env, tgt_desc->ltd_tgt, + tgt_desc->ltd_recovery_thread); +out_ltd: + lod_getref(ltd); + mutex_lock(<d->ltd_mutex); + lock = true; + if (!for_ost && LTD_TGT(ltd, index)->ltd_recovery_thread != NULL) { + struct ptlrpc_thread *thread; + thread = LTD_TGT(ltd, index)->ltd_recovery_thread; + OBD_FREE_PTR(thread); + } + ltd->ltd_tgtnr--; + cfs_bitmap_clear(ltd->ltd_tgt_bitmap, index); + LTD_TGT(ltd, index) = NULL; out_pool: lod_ost_pool_remove(&lod->lod_pool_info, index); out_mutex: - mutex_unlock(<d->ltd_mutex); - lod_putref(lod, ltd); + if (lock) { + mutex_unlock(<d->ltd_mutex); + lod_putref(lod, ltd); + } out_desc: OBD_FREE_PTR(tgt_desc); out_conn: @@ -372,8 +430,18 @@ out_free: return rc; } -/* - * helper function to schedule OST removal from the device table +/** + * Schedule target removal from the target table. + * + * Mark the device as dead. The device is not removed here because it may + * still be in use. The device will be removed in lod_putref() when the + * last reference is released. + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device the target table belongs to + * \param[in] ltd target table + * \param[in] idx index of the target + * \param[in] for_ost type of the target: 0 - MDT, 1 - OST */ static void __lod_del_device(const struct lu_env *env, struct lod_device *lod, struct lod_tgt_descs *ltd, unsigned idx, @@ -384,16 +452,35 @@ static void __lod_del_device(const struct lu_env *env, struct lod_device *lod, lfsck_del_target(env, lod->lod_child, LTD_TGT(ltd, idx)->ltd_tgt, idx, for_ost); + if (!for_ost && LTD_TGT(ltd, idx)->ltd_recovery_thread != NULL) { + struct ptlrpc_thread *thread; + + thread = LTD_TGT(ltd, idx)->ltd_recovery_thread; + OBD_FREE_PTR(thread); + } + if (LTD_TGT(ltd, idx)->ltd_reap == 0) { LTD_TGT(ltd, idx)->ltd_reap = 1; ltd->ltd_death_row++; } } +/** + * Schedule removal of all the targets from the given target table. + * + * See more details in the description for __lod_del_device() + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device the target table belongs to + * \param[in] ltd target table + * \param[in] for_ost type of the target: MDT or OST + * + * \retval 0 always + */ int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod, struct lod_tgt_descs *ltd, bool for_ost) { - int idx; + unsigned int idx; if (ltd->ltd_tgts_size <= 0) return 0; @@ -412,18 +499,22 @@ int lod_fini_tgt(const struct lu_env *env, struct lod_device *lod, return 0; } -/* - * Add support for administratively disabled OST (through the MGS). - * Schedule a target for deletion. Disconnection and real removal from the - * table takes place in lod_putref() once the last table user release its - * reference. +/** + * Remove device by name. * - * \param env - is the environment passed by the caller - * \param lod - is the lod device currently connected to the OSP about to be - * removed - * \param osp - is the name of OSP device about to be removed - * \param idx - is the OSP index - * \param gen - is the generation number, not used currently + * Remove a device identified by \a osp from the target table. Given + * the device can be in use, the real deletion happens in lod_putref(). + * + * \param[in] env execution environment for this thread + * \param[in] lod LOD device to be connected to the new OSP + * \param[in] ltd target table + * \param[in] osp name of OSP device to be removed + * \param[in] idx index of the target + * \param[in] gen generation number, not used currently + * \param[in] for_ost type of the target: 0 - MDT, 1 - OST + * + * \retval 0 if the device was scheduled for removal + * \retval -EINVAL if no device was found */ int lod_del_device(const struct lu_env *env, struct lod_device *lod, struct lod_tgt_descs *ltd, char *osp, unsigned idx, @@ -478,9 +569,23 @@ out: return(rc); } -int lod_ea_store_resize(struct lod_thread_info *info, int size) +/** + * Resize per-thread storage to hold specified size. + * + * A helper function to resize per-thread temporary storage. This storage + * is used to process LOV/LVM EAs and may be quite large. We do not want to + * allocate/release it every time, so instead we put it into the env and + * reallocate on demand. The memory is released when the correspondent thread + * is finished. + * + * \param[in] info LOD-specific storage in the environment + * \param[in] size new size to grow the buffer to + + * \retval 0 on success, -ENOMEM if reallocation failed + */ +int lod_ea_store_resize(struct lod_thread_info *info, size_t size) { - int round = size_roundup_power2(size); + __u32 round = size_roundup_power2(size); LASSERT(round <= lov_mds_md_size(LOV_MAX_STRIPE_COUNT, LOV_MAGIC_V3)); @@ -501,8 +606,21 @@ int lod_ea_store_resize(struct lod_thread_info *info, int size) RETURN(0); } -/* - * generate and write LOV EA for given striped object +/** + * Make LOV EA for striped object. + * + * Generate striping information and store it in the LOV EA of the given + * object. The caller must ensure nobody else is calling the function + * against the object concurrently. The transaction must be started. + * FLDB service must be running as well; it's used to map FID to the target, + * which is stored in LOV EA. + * + * \param[in] env execution environment for this thread + * \param[in] lo LOD object + * \param[in] th transaction handle + * + * \retval 0 if LOV EA is stored successfully + * \retval negative error number on failure */ int lod_generate_and_set_lovea(const struct lu_env *env, struct lod_object *lo, struct thandle *th) @@ -513,8 +631,8 @@ int lod_generate_and_set_lovea(const struct lu_env *env, struct lov_mds_md_v1 *lmm; struct lov_ost_data_v1 *objs; __u32 magic; - int i, rc, lmm_size; - int cplen = 0; + int i, rc; + size_t lmm_size; ENTRY; LASSERT(lo); @@ -547,7 +665,7 @@ int lod_generate_and_set_lovea(const struct lu_env *env, objs = &lmm->lmm_objects[0]; } else { struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm; - cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool, + size_t cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool, sizeof(v3->lmm_pool_name)); if (cplen >= sizeof(v3->lmm_pool_name)) RETURN(-E2BIG); @@ -588,14 +706,30 @@ int lod_generate_and_set_lovea(const struct lu_env *env, info->lti_buf.lb_buf = lmm; info->lti_buf.lb_len = lmm_size; - rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, - th, BYPASS_CAPA); - if (rc < 0) + rc = lod_sub_object_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, + 0, th); + if (rc < 0) { lod_object_free_striping(env, lo); + RETURN(rc); + } RETURN(rc); } +/** + * Get LOV EA. + * + * Fill lti_ea_store buffer in the environment with a value for the given + * EA. The buffer is reallocated if the value doesn't fit. + * + * \param[in,out] env execution environment for this thread + * .lti_ea_store buffer is filled with EA's value + * \param[in] lo LOD object + * \param[in] name name of the EA + * + * \retval 0 if EA is fetched successfully + * \retval negative error number on failure + */ int lod_get_ea(const struct lu_env *env, struct lod_object *lo, const char *name) { @@ -613,7 +747,7 @@ int lod_get_ea(const struct lu_env *env, struct lod_object *lo, repeat: info->lti_buf.lb_buf = info->lti_ea_store; info->lti_buf.lb_len = info->lti_ea_store_size; - rc = dt_xattr_get(env, next, &info->lti_buf, name, BYPASS_CAPA); + rc = dt_xattr_get(env, next, &info->lti_buf, name); } /* if object is not striped or inaccessible */ @@ -622,8 +756,7 @@ repeat: if (rc == -ERANGE) { /* EA doesn't fit, reallocate new buffer */ - rc = dt_xattr_get(env, next, &LU_BUF_NULL, name, - BYPASS_CAPA); + rc = dt_xattr_get(env, next, &LU_BUF_NULL, name); if (rc == -ENODATA || rc == -ENOENT) RETURN(0); else if (rc < 0) @@ -639,55 +772,16 @@ repeat: RETURN(rc); } -int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt, - struct thandle *th) -{ - struct lod_thread_info *info = lod_env_info(env); - struct lod_object *lo = lod_dt_obj(dt); - struct dt_object *next = dt_object_child(dt); - struct lov_user_md_v3 *v3; - int rc; - ENTRY; - - if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) - RETURN(-ENOTDIR); - /* - * store striping defaults into new directory - * used to implement defaults inheritance - */ - - /* probably nothing to inherite */ - if (lo->ldo_striping_cached == 0) - RETURN(0); - - if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr, - lo->ldo_def_stripe_offset)) - RETURN(0); - - v3 = info->lti_ea_store; - if (info->lti_ea_store_size < sizeof(*v3)) { - rc = lod_ea_store_resize(info, sizeof(*v3)); - if (rc != 0) - RETURN(rc); - v3 = info->lti_ea_store; - } - memset(v3, 0, sizeof(*v3)); - v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3); - v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr); - v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset); - v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size); - if (lo->ldo_pool) - strncpy(v3->lmm_pool_name, lo->ldo_pool, - sizeof(v3->lmm_pool_name)); - info->lti_buf.lb_buf = v3; - info->lti_buf.lb_len = sizeof(*v3); - rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th, - BYPASS_CAPA); - - RETURN(rc); -} - -static int validate_lod_and_idx(struct lod_device *md, int idx) +/** + * Verify the target index is present in the current configuration. + * + * \param[in] md LOD device where the target table is stored + * \param[in] idx target's index + * + * \retval 0 if the index is present + * \retval -EINVAL if not + */ +static int validate_lod_and_idx(struct lod_device *md, __u32 idx) { if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size || !cfs_bitmap_check(md->lod_ost_bitmap, idx))) { @@ -711,9 +805,21 @@ static int validate_lod_and_idx(struct lod_device *md, int idx) return 0; } -/* - * allocate array of objects pointers, find/create objects - * stripenr and other fields should be initialized by this moment +/** + * Instantiate objects for stripes. + * + * Allocate and initialize LU-objects representing the stripes. The number + * of the stripes (ldo_stripenr) must be initialized already. The caller + * must ensure nobody else is calling the function on the object at the same + * time. FLDB service must be running to be able to map a FID to the targets + * and find appropriate device representing that target. + * + * \param[in] env execution environment for this thread + * \param[in,out] lo LOD object + * \param[in] objs an array of IDs to creates the objects from + * + * \retval 0 if the objects are instantiated successfully + * \retval negative error number on failure */ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo, struct lov_ost_data_v1 *objs) @@ -724,7 +830,8 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo, struct lu_device *nd; struct dt_object **stripe; int stripe_len; - int i, idx, rc = 0; + int i, rc = 0; + __u32 idx; ENTRY; LASSERT(lo != NULL); @@ -739,6 +846,9 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo, RETURN(-ENOMEM); for (i = 0; i < lo->ldo_stripenr; i++) { + if (unlikely(lovea_slot_is_dummy(&objs[i]))) + continue; + ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid); idx = le32_to_cpu(objs[i].l_ost_idx); rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx); @@ -777,6 +887,7 @@ out: lu_object_put(env, &stripe[i]->do_lu); OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len); + lo->ldo_stripenr = 0; } else { lo->ldo_stripe = stripe; lo->ldo_stripes_allocated = stripe_len; @@ -785,8 +896,18 @@ out: RETURN(rc); } -/* - * Parse striping information stored in lti_ea_store +/** + * Instantiate objects for striping. + * + * Parse striping information in \a buf and instantiate the objects + * representing the stripes. + * + * \param[in] env execution environment for this thread + * \param[in] lo LOD object + * \param[in] buf buffer storing LOV EA to parse + * + * \retval 0 if parsing and objects creation succeed + * \retval negative error number on failure */ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo, const struct lu_buf *buf) @@ -836,9 +957,23 @@ out: RETURN(rc); } +/** + * Initialize the object representing the stripes. + * + * Unless the stripes are initialized already, fetch LOV (for regular + * objects) or LMV (for directory objects) EA and call lod_parse_striping() + * to instantiate the objects representing the stripes. + * + * \param[in] env execution environment for this thread + * \param[in,out] lo LOD object + * + * \retval 0 if parsing and object creation succeed + * \retval negative error number on failure + */ int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo) { struct lod_thread_info *info = lod_env_info(env); + struct lu_buf *buf = &info->lti_buf; struct dt_object *next = dt_object_child(&lo->ldo_obj); int rc = 0; ENTRY; @@ -854,8 +989,7 @@ int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo) if (lo->ldo_dir_slave_stripe) GOTO(out, rc = 0); - /* only regular files can be striped */ - if (lu_object_attr(lod2lu_obj(lo)) & S_IFREG) { + if (S_ISREG(lu_object_attr(lod2lu_obj(lo)))) { rc = lod_get_lov_ea(env, lo); if (rc <= 0) GOTO(out, rc); @@ -863,28 +997,57 @@ int lod_load_striping_locked(const struct lu_env *env, struct lod_object *lo) * there is LOV EA (striping information) in this object * let's parse it and create in-core objects for the stripes */ - info->lti_buf.lb_buf = info->lti_ea_store; - info->lti_buf.lb_len = info->lti_ea_store_size; - rc = lod_parse_striping(env, lo, &info->lti_buf); - } else if (lu_object_attr(lod2lu_obj(lo)) & S_IFDIR) { + buf->lb_buf = info->lti_ea_store; + buf->lb_len = info->lti_ea_store_size; + rc = lod_parse_striping(env, lo, buf); + } else if (S_ISDIR(lu_object_attr(lod2lu_obj(lo)))) { rc = lod_get_lmv_ea(env, lo); - if (rc <= 0) - GOTO(out, rc); + if (rc < (typeof(rc))sizeof(struct lmv_mds_md_v1)) + GOTO(out, rc = rc > 0 ? -EINVAL : rc); + + buf->lb_buf = info->lti_ea_store; + buf->lb_len = info->lti_ea_store_size; + if (rc == sizeof(struct lmv_mds_md_v1)) { + rc = lod_load_lmv_shards(env, lo, buf, true); + if (buf->lb_buf != info->lti_ea_store) { + OBD_FREE_LARGE(info->lti_ea_store, + info->lti_ea_store_size); + info->lti_ea_store = buf->lb_buf; + info->lti_ea_store_size = buf->lb_len; + } + + if (rc < 0) + GOTO(out, rc); + } + /* * there is LOV EA (striping information) in this object * let's parse it and create in-core objects for the stripes */ - info->lti_buf.lb_buf = info->lti_ea_store; - info->lti_buf.lb_len = info->lti_ea_store_size; - rc = lod_parse_dir_striping(env, lo, &info->lti_buf); + rc = lod_parse_dir_striping(env, lo, buf); } + + if (rc == 0) + lo->ldo_striping_cached = 1; out: RETURN(rc); } /** - * Load and parse striping information, create in-core representation for the - * stripes + * A generic function to initialize the stripe objects. + * + * A protected version of lod_load_striping_locked() - load the striping + * information from storage, parse that and instantiate LU objects to + * represent the stripes. The LOD object \a lo supplies a pointer to the + * next sub-object in the LU stack so we can lock it. Also use \a lo to + * return an array of references to the newly instantiated objects. + * + * \param[in] env execution environment for this thread + * \param[in,out] lo LOD object, where striping is stored and + * which gets an array of references + * + * \retval 0 if parsing and object creation succeed + * \retval negative error number on failure **/ int lod_load_striping(const struct lu_env *env, struct lod_object *lo) { @@ -899,8 +1062,25 @@ int lod_load_striping(const struct lu_env *env, struct lod_object *lo) return rc; } +/** + * Verify striping. + * + * Check the validity of all fields including the magic, stripe size, + * stripe count, stripe offset and that the pool is present. Also check + * that each target index points to an existing target. The additional + * \a is_from_disk turns additional checks. In some cases zero fields + * are allowed (like pattern=0). + * + * \param[in] d LOD device + * \param[in] buf buffer with LOV EA to verify + * \param[in] is_from_disk 0 - from user, allow some fields to be 0 + * 1 - from disk, do not allow + * + * \retval 0 if the striping is valid + * \retval -EINVAL if striping is invalid + */ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, - int specific) + bool is_from_disk) { struct lov_user_md_v1 *lum; struct lov_user_md_v3 *lum3; @@ -918,7 +1098,7 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, LASSERT(sizeof(*lum) < sizeof(*lum3)); if (buf->lb_len < sizeof(*lum)) { - CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n", + CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md\n", buf->lb_len); GOTO(out, rc = -EINVAL); } @@ -932,8 +1112,11 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, GOTO(out, rc = -EINVAL); } - if ((specific && le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) || - (!specific && lum->lmm_pattern != 0)) { + /* the user uses "0" for default stripe pattern normally. */ + if (!is_from_disk && lum->lmm_pattern == 0) + lum->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0); + + if (le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) { CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n", le32_to_cpu(lum->lmm_pattern)); GOTO(out, rc = -EINVAL); @@ -948,9 +1131,8 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, GOTO(out, rc = -EINVAL); } - /* an offset of -1 is treated as a "special" valid offset */ stripe_offset = le16_to_cpu(lum->lmm_stripe_offset); - if (stripe_offset != (typeof(stripe_offset))-1) { + if (stripe_offset != LOV_OFFSET_DEFAULT) { /* if offset is not within valid range [0, osts_size) */ if (stripe_offset >= d->lod_osts_size) { CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n", @@ -966,18 +1148,18 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, } } - stripe_count = le16_to_cpu(lum->lmm_stripe_count); if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF) lum_size = offsetof(struct lov_user_md_v1, - lmm_objects[stripe_count]); + lmm_objects[0]); else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF) lum_size = offsetof(struct lov_user_md_v3, - lmm_objects[stripe_count]); + lmm_objects[0]); else - LBUG(); + GOTO(out, rc = -EINVAL); - if (specific && buf->lb_len != lum_size) { - CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with " + stripe_count = le16_to_cpu(lum->lmm_stripe_count); + if (buf->lb_len != lum_size) { + CDEBUG(D_IOCTL, "invalid buf len %zu for lov_user_md with " "magic %#x and stripe_count %u\n", buf->lb_len, magic, stripe_count); GOTO(out, rc = -EINVAL); @@ -988,7 +1170,7 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, lum3 = buf->lb_buf; if (buf->lb_len < sizeof(*lum3)) { - CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n", + CDEBUG(D_IOCTL, "buf len %zu too small for lov_user_md_v3\n", buf->lb_len); GOTO(out, rc = -EINVAL); } @@ -1000,13 +1182,13 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf, if (pool == NULL) goto out; - if (stripe_offset != (typeof(stripe_offset))-1) { + if (stripe_offset != LOV_OFFSET_DEFAULT) { rc = lod_check_index_in_pool(stripe_offset, pool); if (rc < 0) GOTO(out, rc = -EINVAL); } - if (specific && stripe_count > pool_tgt_count(pool)) { + if (is_from_disk && stripe_count > pool_tgt_count(pool)) { CDEBUG(D_IOCTL, "stripe count %u > # OSTs %u in the pool\n", stripe_count, pool_tgt_count(pool)); @@ -1058,6 +1240,11 @@ void lod_fix_desc_qos_maxage(__u32 *val) *val = LOV_DESC_QOS_MAXAGE_DEFAULT; } +/** + * Used to fix insane default striping. + * + * \param[in] desc striping description + */ void lod_fix_desc(struct lov_desc *desc) { lod_fix_desc_stripe_size(&desc->ld_default_stripe_size); @@ -1066,6 +1253,15 @@ void lod_fix_desc(struct lov_desc *desc) lod_fix_desc_qos_maxage(&desc->ld_qos_maxage); } +/** + * Initialize the structures used to store pools and default striping. + * + * \param[in] lod LOD device + * \param[in] lcfg configuration structure storing default striping. + * + * \retval 0 if initialization succeeds + * \retval negative error number on failure + */ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) { struct obd_device *obd; @@ -1110,7 +1306,7 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) lod->lod_sp_me = LUSTRE_SP_CLI; /* Set up allocation policy (QoS and RR) */ - CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list); + INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list); init_rwsem(&lod->lod_qos.lq_rw_sem); lod->lod_qos.lq_dirty = 1; lod->lod_qos.lq_rr.lqr_dirty = 1; @@ -1131,11 +1327,12 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg) if (lod->lod_pools_hash_body == NULL) RETURN(-ENOMEM); - CFS_INIT_LIST_HEAD(&lod->lod_pool_list); + INIT_LIST_HEAD(&lod->lod_pool_list); lod->lod_pool_count = 0; rc = lod_ost_pool_init(&lod->lod_pool_info, 0); if (rc) GOTO(out_hash, rc); + lod_qos_rr_init(&lod->lod_qos.lq_rr); rc = lod_ost_pool_init(&lod->lod_qos.lq_rr.lqr_pool, 0); if (rc) GOTO(out_pool_info, rc); @@ -1150,15 +1347,20 @@ out_hash: return rc; } +/** + * Release the structures describing the pools. + * + * \param[in] lod LOD device from which we release the structures + * + * \retval 0 always + */ int lod_pools_fini(struct lod_device *lod) { struct obd_device *obd = lod2obd(lod); - cfs_list_t *pos, *tmp; - struct pool_desc *pool; + struct pool_desc *pool, *tmp; ENTRY; - cfs_list_for_each_safe(pos, tmp, &lod->lod_pool_list) { - pool = cfs_list_entry(pos, struct pool_desc, pool_list); + list_for_each_entry_safe(pool, tmp, &lod->lod_pool_list, pool_list) { /* free pool structs */ CDEBUG(D_INFO, "delete pool %p\n", pool); /* In the function below, .hs_keycmp resolves to @@ -1173,4 +1375,3 @@ int lod_pools_fini(struct lod_device *lod) RETURN(0); } -