X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_handler.c;h=b59d920b8ce0f0cf23d7ed1489560c2377b44db1;hb=bafa12c06d87c05bb263eed37ad8af1a2df99894;hp=65d69a749bfd8d45f7dc884847356b8b585c16db;hpb=2a4a7af3993bd06afdb3c6499f8cb4edbb9564b2;p=fs%2Flustre-release.git diff --git a/lustre/osd-zfs/osd_handler.c b/lustre/osd-zfs/osd_handler.c index 65d69a7..b59d920 100644 --- a/lustre/osd-zfs/osd_handler.c +++ b/lustre/osd-zfs/osd_handler.c @@ -28,7 +28,7 @@ * Use is subject to license terms. */ /* - * Copyright (c) 2011, 2012 Whamcloud, Inc. + * Copyright (c) 2012, 2013, Intel Corporation. * Use is subject to license terms. * */ @@ -44,9 +44,6 @@ * Author: Johann Lombardi */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_OSD #include @@ -57,6 +54,8 @@ #include #include #include +#include +#include #include "osd_internal.h" @@ -78,7 +77,7 @@ struct lu_context_key osd_key; static char *root_tag = "osd_mount, rootdb"; /* Slab for OSD object allocation */ -cfs_mem_cache_t *osd_object_kmem; +struct kmem_cache *osd_object_kmem; static struct lu_kmem_descr osd_caches[] = { { @@ -149,7 +148,7 @@ static void osd_trans_commit_cb(void *cb_data, int error) dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ - cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) + list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) dcb->dcb_func(NULL, th, dcb, error); /* Unlike ldiskfs, zfs updates space accounting at commit time. @@ -164,7 +163,7 @@ static void osd_trans_commit_cb(void *cb_data, int error) th->th_dev = NULL; lu_context_exit(&th->th_ctx); lu_context_fini(&th->th_ctx); - OBD_FREE_PTR(oh); + thandle_put(&oh->ot_super); EXIT; } @@ -174,7 +173,7 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) struct osd_thandle *oh; oh = container_of0(th, struct osd_thandle, ot_super); - cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); + list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); return 0; } @@ -228,7 +227,8 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d, /* * Concurrency: shouldn't matter. */ -static int osd_trans_stop(const struct lu_env *env, struct thandle *th) +static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, + struct thandle *th) { struct osd_device *osd = osd_dt_dev(th->th_dev); struct osd_thandle *oh; @@ -245,7 +245,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) /* there won't be any commit, release reserved quota space now, * if any */ qsd_op_end(env, osd->od_quota_slave, &oh->ot_quota_trans); - OBD_FREE_PTR(oh); + thandle_put(&oh->ot_super); RETURN(0); } @@ -297,14 +297,16 @@ static struct thandle *osd_trans_create(const struct lu_env *env, } oh->ot_tx = tx; - CFS_INIT_LIST_HEAD(&oh->ot_dcb_list); - CFS_INIT_LIST_HEAD(&oh->ot_sa_list); - cfs_sema_init(&oh->ot_sa_lock, 1); + INIT_LIST_HEAD(&oh->ot_dcb_list); + INIT_LIST_HEAD(&oh->ot_sa_list); + sema_init(&oh->ot_sa_lock, 1); memset(&oh->ot_quota_trans, 0, sizeof(oh->ot_quota_trans)); th = &oh->ot_super; th->th_dev = dt; th->th_result = 0; th->th_tags = LCT_TX_HANDLE; + atomic_set(&th->th_refc, 1); + th->th_alloc_size = sizeof(*oh); RETURN(th); } @@ -319,7 +321,7 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, ENTRY; rc = udmu_objset_statfs(&osd->od_objset, osfs); - if (rc) + if (unlikely(rc)) RETURN(rc); osfs->os_bavail -= min_t(obd_size, OSD_GRANT_FOR_LOCAL_OIDS / osfs->os_bsize, @@ -334,19 +336,23 @@ static void osd_conf_get(const struct lu_env *env, const struct dt_device *dev, struct dt_device_param *param) { + struct osd_device *osd = osd_dt_dev(dev); + /* * XXX should be taken from not-yet-existing fs abstraction layer. */ - param->ddp_max_name_len = MAXNAMELEN; - param->ddp_max_nlink = 1 << 31; /* it's 8byte on a disk */ - param->ddp_block_shift = 12; /* XXX */ - param->ddp_mount_type = LDD_MT_ZFS; + param->ddp_max_name_len = MAXNAMELEN; + param->ddp_max_nlink = 1 << 31; /* it's 8byte on a disk */ + param->ddp_block_shift = 12; /* XXX */ + param->ddp_mount_type = LDD_MT_ZFS; - param->ddp_mntopts = MNTOPT_USERXATTR | MNTOPT_ACL; - param->ddp_max_ea_size = DXATTR_MAX_ENTRY_SIZE; + param->ddp_mntopts = MNTOPT_USERXATTR; + if (osd->od_posix_acl) + param->ddp_mntopts |= MNTOPT_ACL; + param->ddp_max_ea_size = DXATTR_MAX_ENTRY_SIZE; /* for maxbytes, report same value as ZPL */ - param->ddp_maxbytes = MAX_LFS_FILESIZE; + param->ddp_maxbytes = MAX_LFS_FILESIZE; /* Default reserved fraction of the available space that should be kept * for error margin. Unfortunately, there are many factors that can @@ -362,8 +368,6 @@ static void osd_conf_get(const struct lu_env *env, param->ddp_inodespace = OSD_DNODE_EST_COUNT; /* per-fragment overhead to be used by the client code */ param->ddp_grant_frag = udmu_blk_insert_cost(); - - param->ddp_mnt = NULL; } /* @@ -383,6 +387,7 @@ static int osd_commit_async(const struct lu_env *env, struct dt_device *dev) tx_state_t *tx = &dmu_objset_pool(osd->od_objset.os)->dp_tx; uint64_t txg; + mutex_enter(&tx->tx_sync_lock); txg = tx->tx_open_txg + 1; if (tx->tx_quiesce_txg_waiting < txg) { tx->tx_quiesce_txg_waiting = txg; @@ -514,21 +519,31 @@ static void osd_xattr_changed_cb(void *arg, uint64_t newval) static int osd_mount(const struct lu_env *env, struct osd_device *o, struct lustre_cfg *cfg) { - struct dsl_dataset *ds; - char *dev = lustre_cfg_string(cfg, 1); - dmu_buf_t *rootdb; - int rc; + struct dsl_dataset *ds; + char *mntdev = lustre_cfg_string(cfg, 1); + char *svname = lustre_cfg_string(cfg, 4); + dmu_buf_t *rootdb; + dsl_pool_t *dp; + const char *opts; + int rc; ENTRY; if (o->od_objset.os != NULL) RETURN(0); - if (strlen(dev) >= sizeof(o->od_mntdev)) + if (mntdev == NULL || svname == NULL) + RETURN(-EINVAL); + + rc = strlcpy(o->od_mntdev, mntdev, sizeof(o->od_mntdev)); + if (rc >= sizeof(o->od_mntdev)) + RETURN(-E2BIG); + + rc = strlcpy(o->od_svname, svname, sizeof(o->od_svname)); + if (rc >= sizeof(o->od_svname)) RETURN(-E2BIG); - strcpy(o->od_mntdev, dev); - strncpy(o->od_svname, lustre_cfg_string(cfg, 4), - sizeof(o->od_svname) - 1); + if (server_name_is_ost(o->od_svname)) + o->od_is_ost = 1; rc = -udmu_objset_open(o->od_mntdev, &o->od_objset); if (rc) { @@ -537,8 +552,12 @@ static int osd_mount(const struct lu_env *env, } ds = dmu_objset_ds(o->od_objset.os); + dp = dmu_objset_pool(o->od_objset.os); LASSERT(ds); + LASSERT(dp); + dsl_pool_config_enter(dp, FTAG); rc = dsl_prop_register(ds, "xattr", osd_xattr_changed_cb, o); + dsl_pool_config_exit(dp, FTAG); if (rc) CERROR("%s: cat not register xattr callback, ignore: %d\n", o->od_svname, rc); @@ -568,6 +587,10 @@ static int osd_mount(const struct lu_env *env, if (rc) GOTO(err, rc); + rc = osd_convert_root_to_new_seq(env, o); + if (rc) + GOTO(err, rc); + /* Use our own ZAP for inode accounting by default, this can be changed * via procfs to estimate the inode usage from the block usage */ o->od_quota_iused_est = 0; @@ -586,6 +609,12 @@ static int osd_mount(const struct lu_env *env, o->od_quota_slave = NULL; GOTO(err, rc); } + + /* parse mount option "noacl", and enable ACL by default */ + opts = lustre_cfg_string(cfg, 3); + if (opts == NULL || strstr(opts, "noacl") == NULL) + o->od_posix_acl = 1; + err: RETURN(rc); } @@ -594,15 +623,15 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o) { ENTRY; - if (cfs_atomic_read(&o->od_zerocopy_alloc)) + if (atomic_read(&o->od_zerocopy_alloc)) CERROR("%s: lost %d allocated page(s)\n", o->od_svname, - cfs_atomic_read(&o->od_zerocopy_alloc)); - if (cfs_atomic_read(&o->od_zerocopy_loan)) + atomic_read(&o->od_zerocopy_alloc)); + if (atomic_read(&o->od_zerocopy_loan)) CERROR("%s: lost %d loaned abuf(s)\n", o->od_svname, - cfs_atomic_read(&o->od_zerocopy_loan)); - if (cfs_atomic_read(&o->od_zerocopy_pin)) + atomic_read(&o->od_zerocopy_loan)); + if (atomic_read(&o->od_zerocopy_pin)) CERROR("%s: lost %d pinned dbuf(s)\n", o->od_svname, - cfs_atomic_read(&o->od_zerocopy_pin)); + atomic_read(&o->od_zerocopy_pin)); if (o->od_objset.os != NULL) udmu_objset_close(&o->od_objset); @@ -704,8 +733,10 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, if (rc) CERROR("%s: dsl_prop_unregister xattr error %d\n", o->od_svname, rc); - arc_remove_prune_callback(o->arc_prune_cb); - o->arc_prune_cb = NULL; + if (o->arc_prune_cb != NULL) { + arc_remove_prune_callback(o->arc_prune_cb); + o->arc_prune_cb = NULL; + } osd_sync(env, lu2dt_dev(d)); txg_wait_callbacks(spa_get_dsl(dmu_objset_spa(o->od_objset.os))); } @@ -736,21 +767,31 @@ static int osd_process_config(const struct lu_env *env, struct lu_device *d, struct lustre_cfg *cfg) { struct osd_device *o = osd_dev(d); - int err; + int rc; ENTRY; switch(cfg->lcfg_command) { case LCFG_SETUP: - err = osd_mount(env, o, cfg); + rc = osd_mount(env, o, cfg); break; case LCFG_CLEANUP: - err = osd_shutdown(env, o); + rc = osd_shutdown(env, o); + break; + case LCFG_PARAM: { + LASSERT(&o->od_dt_dev); + rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + if (rc > 0 || rc == -ENOSYS) + rc = class_process_proc_param(PARAM_OST, + lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); break; + } default: - err = -ENOTTY; + rc = -ENOTTY; } - RETURN(err); + RETURN(rc); } static int osd_recovery_complete(const struct lu_env *env, struct lu_device *d) @@ -788,9 +829,9 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp, *exp = class_conn2export(&conn); - cfs_spin_lock(&osd->od_objset.lock); + spin_lock(&osd->od_objset.lock); osd->od_connects++; - cfs_spin_unlock(&osd->od_objset.lock); + spin_unlock(&osd->od_objset.lock); RETURN(0); } @@ -807,11 +848,11 @@ static int osd_obd_disconnect(struct obd_export *exp) ENTRY; /* Only disconnect the underlying layers on the final disconnect. */ - cfs_spin_lock(&osd->od_objset.lock); + spin_lock(&osd->od_objset.lock); osd->od_connects--; if (osd->od_connects == 0) release = 1; - cfs_spin_unlock(&osd->od_objset.lock); + spin_unlock(&osd->od_objset.lock); rc = class_disconnect(exp); /* bz 9811 */ @@ -827,15 +868,6 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, int rc = 0; ENTRY; - if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) { - /* MDT/MDD still use old infrastructure to create - * special files */ - rc = llo_local_objects_setup(env, lu2md_dev(pdev), - lu2dt_dev(dev)); - if (rc) - RETURN(rc); - } - if (osd->od_quota_slave != NULL) /* set up quota slave objects */ rc = qsd_prepare(env, osd->od_quota_slave); @@ -858,6 +890,14 @@ static void osd_type_stop(struct lu_device_type *t) { } +int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, + struct lu_fid *fid, struct md_op_data *op_data) +{ + struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev); + + return seq_client_alloc_fid(env, osd->od_cl_seq, fid); +} + static struct lu_device_type_operations osd_device_type_ops = { .ldto_init = osd_type_init, .ldto_fini = osd_type_fini, @@ -883,7 +923,8 @@ static struct lu_device_type osd_device_type = { static struct obd_ops osd_obd_device_ops = { .o_owner = THIS_MODULE, .o_connect = osd_obd_connect, - .o_disconnect = osd_obd_disconnect + .o_disconnect = osd_obd_disconnect, + .o_fid_alloc = osd_fid_alloc }; int __init osd_init(void) @@ -898,8 +939,10 @@ int __init osd_init(void) if (rc) return rc; - rc = class_register_type(&osd_obd_device_ops, NULL, - lprocfs_osd_module_vars, + rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL, +#ifndef HAVE_ONLY_PROCFS_SEQ + NULL, +#endif LUSTRE_OSD_ZFS_NAME, &osd_device_type); if (rc) lu_kmem_fini(osd_caches);