X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fofd%2Fofd_dev.c;h=cb7a1b562891833b3516cc60770395add0cbec4e;hp=28bbf03a492ffb4d1a8ba3624b8da58ade24b8bc;hb=1711e26ae861c28829870c2433caf7ee232909cf;hpb=34ba26ea9884df0758f8ec8dd1047b4cf474af09;ds=sidebyside diff --git a/lustre/ofd/ofd_dev.c b/lustre/ofd/ofd_dev.c index 28bbf03..cb7a1b5 100644 --- a/lustre/ofd/ofd_dev.c +++ b/lustre/ofd/ofd_dev.c @@ -23,11 +23,10 @@ * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2015, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lustre/ofd/ofd_dev.c * @@ -69,19 +68,20 @@ #define DEBUG_SUBSYSTEM S_FILTER #include -#include +#include +#include #include #include -#include #include #include #include +#include +#include #include "ofd_internal.h" /* Slab for OFD object allocation */ static struct kmem_cache *ofd_object_kmem; - static struct lu_kmem_descr ofd_caches[] = { { .ckd_cache = &ofd_object_kmem, @@ -161,7 +161,8 @@ out: * \retval negative value on error */ static int ofd_stack_init(const struct lu_env *env, - struct ofd_device *m, struct lustre_cfg *cfg) + struct ofd_device *m, struct lustre_cfg *cfg, + u32 *lmd_flags) { const char *dev = lustre_cfg_string(cfg, 0); struct lu_device *d; @@ -180,8 +181,13 @@ static int ofd_stack_init(const struct lu_env *env, } lmd = s2lsi(lmi->lmi_sb)->lsi_lmd; - if (lmd != NULL && lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) - m->ofd_skip_lfsck = 1; + if (lmd) { + if (lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) + m->ofd_skip_lfsck = 1; + if (lmd->lmd_flags & LMD_FLG_NO_PRECREATE) + m->ofd_no_precreate = 1; + *lmd_flags = lmd->lmd_flags; + } /* find bottom osd */ OBD_ALLOC(osdname, MTI_NAME_MAXLEN); @@ -235,26 +241,53 @@ static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m, if (obd->obd_fail) strcat(flags, "A"); lustre_cfg_bufs_set_string(&bufs, 1, flags); - lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs); - if (lcfg == NULL) + OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen)); + if (!lcfg) RETURN_EXIT; + lustre_cfg_init(lcfg, LCFG_CLEANUP, &bufs); LASSERT(top); top->ld_ops->ldo_process_config(env, top, lcfg); - lustre_cfg_free(lcfg); + OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens)); - lu_site_purge(env, top->ld_site, ~0); - if (!cfs_hash_is_empty(top->ld_site->ls_obj_hash)) { - LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL); - lu_site_print(env, top->ld_site, &msgdata, lu_cdebug_printer); + if (m->ofd_los != NULL) { + local_oid_storage_fini(env, m->ofd_los); + m->ofd_los = NULL; } + lu_site_purge(env, top->ld_site, ~0); + lu_site_print(env, top->ld_site, &top->ld_site->ls_obj_hash.nelems, + D_OTHER, lu_cdebug_printer); LASSERT(m->ofd_osd_exp); obd_disconnect(m->ofd_osd_exp); EXIT; } +static void ofd_stack_pre_fini(const struct lu_env *env, struct ofd_device *m, + struct lu_device *top) +{ + struct lustre_cfg_bufs bufs; + struct lustre_cfg *lcfg; + ENTRY; + + LASSERT(top); + + lustre_cfg_bufs_reset(&bufs, ofd_name(m)); + lustre_cfg_bufs_set_string(&bufs, 1, NULL); + OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen)); + if (!lcfg) { + CERROR("%s: failed to trigger LCFG_PRE_CLEANUP\n", ofd_name(m)); + } else { + lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, &bufs); + top->ld_ops->ldo_process_config(env, top, lcfg); + OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, + lcfg->lcfg_buflens)); + } + + EXIT; +} + /* For interoperability, see mdt_interop_param[]. */ static struct cfg_interop_param ofd_interop_param[] = { { "ost.quota_type", NULL }, @@ -289,13 +322,7 @@ static bool match_symlink_param(char *param) sval = strchr(param, '='); if (sval != NULL) { paramlen = sval - param; - if (strncmp(param, "writethrough_cache_enable", - paramlen) == 0 || - strncmp(param, "readcache_max_filesize", - paramlen) == 0 || - strncmp(param, "read_cache_enable", - paramlen) == 0 || - strncmp(param, "brw_stats", paramlen) == 0) + if (strncmp(param, "brw_stats", paramlen) == 0) return true; } } @@ -328,11 +355,11 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d, switch (cfg->lcfg_command) { case LCFG_PARAM: { - struct obd_device *obd = ofd_obd(m); /* For interoperability */ - struct cfg_interop_param *ptr = NULL; - struct lustre_cfg *old_cfg = NULL; - char *param = NULL; + struct cfg_interop_param *ptr = NULL; + struct lustre_cfg *old_cfg = NULL; + char *param = NULL; + ssize_t count; param = lustre_cfg_string(cfg, 1); if (param == NULL) { @@ -366,14 +393,16 @@ static int ofd_process_config(const struct lu_env *env, struct lu_device *d, break; } - rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg, - d->ld_obd); - if (rc > 0 || rc == -ENOSYS) { - CDEBUG(D_CONFIG, "pass param %s down the stack.\n", - param); - /* we don't understand; pass it on */ - rc = next->ld_ops->ldo_process_config(env, next, cfg); + count = class_modify_config(cfg, PARAM_OST, + &d->ld_obd->obd_kset.kobj); + if (count > 0) { + rc = 0; + break; } + CDEBUG(D_CONFIG, "pass param %s down the stack.\n", + param); + /* we don't understand; pass it on */ + rc = next->ld_ops->ldo_process_config(env, next, cfg); break; } case LCFG_SPTLRPC_CONF: { @@ -423,6 +452,14 @@ static int ofd_object_init(const struct lu_env *env, struct lu_object *o, RETURN(rc); } +static void ofd_object_free_rcu(struct rcu_head *head) +{ + struct ofd_object *of = container_of(head, struct ofd_object, + ofo_header.loh_rcu); + + kmem_cache_free(ofd_object_kmem, of); +} + /** * Implementation of lu_object_operations::loo_object_free. * @@ -444,7 +481,8 @@ static void ofd_object_free(const struct lu_env *env, struct lu_object *o) lu_object_fini(o); lu_object_header_fini(h); - OBD_SLAB_FREE_PTR(of, ofd_object_kmem); + OBD_FREE_PRE(of, sizeof(*of), "slab-freed"); + call_rcu(&of->ofo_header.loh_rcu, ofd_object_free_rcu); EXIT; } @@ -468,7 +506,7 @@ static int ofd_object_print(const struct lu_env *env, void *cookie, return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o); } -static struct lu_object_operations ofd_obj_ops = { +static const struct lu_object_operations ofd_obj_ops = { .loo_object_init = ofd_object_init, .loo_object_free = ofd_object_free, .loo_object_print = ofd_object_print @@ -506,6 +544,7 @@ static struct lu_object *ofd_object_alloc(const struct lu_env *env, lu_object_init(o, h, d); lu_object_add_top(h, o); o->lo_ops = &ofd_obj_ops; + range_lock_tree_init(&of->ofo_write_tree); RETURN(o); } else { RETURN(NULL); @@ -607,6 +646,8 @@ static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev, LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc); target_recovery_init(&ofd->ofd_lut, tgt_request_handle); + OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_OST_PREPARE_DELAY, OBD_FAIL_ONCE, + (OBD_TIMEOUT_DEFAULT + 1) / 4); LASSERT(obd->obd_no_conn); spin_lock(&obd->obd_dev_lock); obd->obd_no_conn = 0; @@ -645,14 +686,14 @@ static int ofd_recovery_complete(const struct lu_env *env, * Grant space for object precreation on the self export. * The initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs) * is enough to create 10k objects. More space is then acquired for - * precreation in ofd_grant_create(). + * precreation in tgt_grant_create(). */ memset(&oti->fti_ocd, 0, sizeof(oti->fti_ocd)); oti->fti_ocd.ocd_grant = OST_MAX_PRECREATE / 2; - oti->fti_ocd.ocd_grant *= ofd->ofd_dt_conf.ddp_inodespace; + oti->fti_ocd.ocd_grant *= ofd->ofd_lut.lut_dt_conf.ddp_inodespace; oti->fti_ocd.ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_GRANT_PARAM; - ofd_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd, + tgt_grant_connect(env, dev->ld_obd->obd_self_export, &oti->fti_ocd, true); rc = next->ld_ops->ldo_recovery_complete(env, next); RETURN(rc); @@ -661,82 +702,14 @@ static int ofd_recovery_complete(const struct lu_env *env, /** * lu_device_operations matrix for OFD device. */ -static struct lu_device_operations ofd_lu_ops = { +static const struct lu_device_operations ofd_lu_ops = { .ldo_object_alloc = ofd_object_alloc, .ldo_process_config = ofd_process_config, .ldo_recovery_complete = ofd_recovery_complete, .ldo_prepare = ofd_prepare, }; -LPROC_SEQ_FOPS(lprocfs_nid_stats_clear); - -/** - * Initialize all needed procfs entries for OFD device. - * - * \param[in] ofd OFD device - * - * \retval 0 if successful - * \retval negative value on error - */ -static int ofd_procfs_init(struct ofd_device *ofd) -{ - struct obd_device *obd = ofd_obd(ofd); - struct proc_dir_entry *entry; - int rc = 0; - - ENTRY; - - /* lprocfs must be setup before the ofd so state can be safely added - * to /proc incrementally as the ofd is setup */ - obd->obd_vars = lprocfs_ofd_obd_vars; - rc = lprocfs_obd_setup(obd); - if (rc) { - CERROR("%s: lprocfs_obd_setup failed: %d.\n", - obd->obd_name, rc); - RETURN(rc); - } - - rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST); - if (rc) { - CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n", - obd->obd_name, rc); - GOTO(obd_cleanup, rc); - } - - obd->obd_uses_nid_stats = 1; - - entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL); - if (IS_ERR(entry)) { - rc = PTR_ERR(entry); - CERROR("%s: error %d setting up lprocfs for %s\n", - obd->obd_name, rc, "exports"); - GOTO(obd_cleanup, rc); - } - obd->obd_proc_exports_entry = entry; - - entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear", - obd, &lprocfs_nid_stats_clear_fops); - if (IS_ERR(entry)) { - rc = PTR_ERR(entry); - CERROR("%s: add proc entry 'clear' failed: %d.\n", - obd->obd_name, rc); - GOTO(obd_cleanup, rc); - } - - ofd_stats_counter_init(obd->obd_stats); - - rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST, - ofd_stats_counter_init); - if (rc) - GOTO(obd_cleanup, rc); - RETURN(0); -obd_cleanup: - lprocfs_obd_cleanup(obd); - lprocfs_free_obd_stats(obd); - - return rc; -} - +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0) /** * Expose OSD statistics to OFD layer. * @@ -750,8 +723,31 @@ obd_cleanup: */ static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd) { - struct obd_device *obd = ofd_obd(ofd); - struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd; + struct obd_device *obd = ofd_obd(ofd); + struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd; + struct kobj_type *osd_type; + int i; + + osd_type = get_ktype(&ofd->ofd_osd->dd_kobj); + for (i = 0; osd_type->default_attrs[i]; i++) { + if (strcmp(osd_type->default_attrs[i]->name, + "read_cache_enable") == 0) { + ofd->ofd_read_cache_enable = + osd_type->default_attrs[i]; + } + + if (strcmp(osd_type->default_attrs[i]->name, + "readcache_max_filesize") == 0) { + ofd->ofd_read_cache_max_filesize = + osd_type->default_attrs[i]; + } + + if (strcmp(osd_type->default_attrs[i]->name, + "writethrough_cache_enable") == 0) { + ofd->ofd_write_cache_enable = + osd_type->default_attrs[i]; + } + } if (obd->obd_proc_entry == NULL) return; @@ -759,21 +755,8 @@ static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd) lprocfs_add_symlink("brw_stats", obd->obd_proc_entry, "../../%s/%s/brw_stats", osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry, - "../../%s/%s/read_cache_enable", - osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("readcache_max_filesize", - obd->obd_proc_entry, - "../../%s/%s/readcache_max_filesize", - osd_obd->obd_type->typ_name, obd->obd_name); - - lprocfs_add_symlink("writethrough_cache_enable", - obd->obd_proc_entry, - "../../%s/%s/writethrough_cache_enable", - osd_obd->obd_type->typ_name, obd->obd_name); } +#endif /** * Cleanup all procfs entries in OFD. @@ -784,6 +767,7 @@ static void ofd_procfs_fini(struct ofd_device *ofd) { struct obd_device *obd = ofd_obd(ofd); + tgt_tunables_fini(&ofd->ofd_lut); lprocfs_free_per_client_stats(obd); lprocfs_obd_cleanup(obd); lprocfs_free_obd_stats(obd); @@ -823,18 +807,19 @@ int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd) */ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) { - struct seq_server_site *ss = &ofd->ofd_seq_site; - struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev; - char *obd_name = ofd_name(ofd); - char *name = NULL; - int rc = 0; + struct seq_server_site *ss = &ofd->ofd_seq_site; + struct lu_device *lu = &ofd->ofd_dt_dev.dd_lu_dev; + char *obd_name = ofd_name(ofd); + char *name = NULL; + int len = strlen(obd_name) + 7; + int rc = 0; ss = &ofd->ofd_seq_site; lu->ld_site->ld_seq_site = ss; ss->ss_lu = lu->ld_site; ss->ss_node_id = ofd->ofd_lut.lut_lsd.lsd_osd_index; - OBD_ALLOC(name, sizeof(obd_name) * 2 + 10); + OBD_ALLOC(name, len); if (name == NULL) return -ENOMEM; @@ -845,7 +830,7 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) rc = seq_server_init(env, ss->ss_server_seq, ofd->ofd_osd, obd_name, LUSTRE_SEQ_SERVER, ss); if (rc) { - CERROR("%s : seq server init error %d\n", obd_name, rc); + CERROR("%s: seq server init error: rc = %d\n", obd_name, rc); GOTO(out_server, rc); } ss->ss_server_seq->lss_space.lsr_index = ss->ss_node_id; @@ -854,22 +839,13 @@ int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd) if (ss->ss_client_seq == NULL) GOTO(out_server, rc = -ENOMEM); - /* - * It always printed as "%p", so that the name is unique in the kernel, - * even if the filesystem is mounted twice. So sizeof(.) * 2 is enough. - */ - snprintf(name, sizeof(obd_name) * 2 + 7, "%p-super", obd_name); - rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, - name, NULL); - if (rc) { - CERROR("%s : seq client init error %d\n", obd_name, rc); - GOTO(out_client, rc); - } + snprintf(name, len, "%s-super", obd_name); + seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_DATA, + name, NULL); rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq); if (rc) { -out_client: seq_client_fini(ss->ss_client_seq); OBD_FREE_PTR(ss->ss_client_seq); ss->ss_client_seq = NULL; @@ -879,7 +855,7 @@ out_server: ss->ss_server_seq = NULL; } out_name: - OBD_FREE(name, sizeof(obd_name) * 2 + 10); + OBD_FREE(name, len); return rc; } @@ -901,6 +877,7 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) void *key, *val = NULL; int keylen, vallen, rc = 0; bool is_grant_shrink; + ktime_t kstart = ktime_get(); ENTRY; @@ -933,25 +910,36 @@ static int ofd_set_info_hdl(struct tgt_session_info *tsi) if (is_grant_shrink) { body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY); + /* + * Because we already sync grant info with client when + * reconnect, grant info will be cleared for resent + * req, otherwise, outdated grant count in the rpc + * would de-sync grant counters + */ + if (lustre_msg_get_flags(req->rq_reqmsg) & + (MSG_RESENT | MSG_REPLAY)) { + DEBUG_REQ(D_CACHE, req, + "clear resent/replay req grant info"); + body->oa.o_valid &= ~OBD_MD_FLGRANT; + } + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); *repbody = *body; /** handle grant shrink, similar to a read request */ - ofd_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp, + tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp, &repbody->oa); } else if (KEY_IS(KEY_EVICT_BY_NID)) { if (vallen > 0) obd_export_evict_by_nid(tsi->tsi_exp->exp_obd, val); rc = 0; - } else if (KEY_IS(KEY_SPTLRPC_CONF)) { - rc = tgt_adapt_sptlrpc_conf(tsi->tsi_tgt, 0); } else { CERROR("%s: Unsupported key %s\n", tgt_name(tsi->tsi_tgt), (char *)key); rc = -EOPNOTSUPP; } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SET_INFO, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -993,48 +981,25 @@ int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd, return rc; } -struct locked_region { - struct list_head list; - struct lustre_handle lh; -}; -/** - * Lock single extent and save lock handle in the list. - * - * This is supplemental function for lock_zero_regions(). It allocates - * new locked_region structure and locks it with extent lock, then adds - * it to the list of all such regions. - * - * \param[in] ns LDLM namespace - * \param[in] res_id resource ID - * \param[in] begin start of region - * \param[in] end end of region - * \param[in] locked list head of regions list - * - * \retval 0 if successful locking - * \retval negative value on error - */ -static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, - unsigned long long begin, unsigned long long end, - struct list_head *locked) +static int ofd_lock_unlock_region(const struct lu_env *env, + struct ldlm_namespace *ns, + struct ldlm_res_id *res_id, + unsigned long long begin, + unsigned long long end) { - struct locked_region *region = NULL; __u64 flags = 0; int rc; + struct lustre_handle lh = { 0 }; LASSERT(begin <= end); - OBD_ALLOC_PTR(region); - if (region == NULL) - return -ENOMEM; - rc = tgt_extent_lock(ns, res_id, begin, end, ®ion->lh, - LCK_PR, &flags); + rc = tgt_extent_lock(env, ns, res_id, begin, end, &lh, LCK_PR, &flags); if (rc != 0) return rc; - CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, - ®ion->lh); - list_add(®ion->list, locked); + CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end, &lh); + tgt_data_unlock(&lh, LCK_PR); return 0; } @@ -1057,10 +1022,10 @@ static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, * \retval 0 if successful * \retval negative value on error */ -static int lock_zero_regions(struct ldlm_namespace *ns, +static int lock_zero_regions(const struct lu_env *env, + struct ldlm_namespace *ns, struct ldlm_res_id *res_id, - struct fiemap *fiemap, - struct list_head *locked) + struct fiemap *fiemap) { __u64 begin = fiemap->fm_start; unsigned int i; @@ -1074,8 +1039,8 @@ static int lock_zero_regions(struct ldlm_namespace *ns, if (fiemap_start[i].fe_logical > begin) { CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n", begin, fiemap_start[i].fe_logical); - rc = lock_region(ns, res_id, begin, - fiemap_start[i].fe_logical, locked); + rc = ofd_lock_unlock_region(env, ns, res_id, begin, + fiemap_start[i].fe_logical); if (rc) RETURN(rc); } @@ -1086,34 +1051,13 @@ static int lock_zero_regions(struct ldlm_namespace *ns, if (begin < (fiemap->fm_start + fiemap->fm_length)) { CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n", begin, fiemap->fm_start + fiemap->fm_length); - rc = lock_region(ns, res_id, begin, - fiemap->fm_start + fiemap->fm_length, locked); + rc = ofd_lock_unlock_region(env, ns, res_id, begin, + fiemap->fm_start + fiemap->fm_length); } RETURN(rc); } -/** - * Unlock all previously locked sparse areas for given resource. - * - * This function goes through list of locked regions, unlocking and freeing - * them one-by-one. - * - * \param[in] ns LDLM namespace - * \param[in] locked list head of regions list - */ -static void -unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked) -{ - struct locked_region *entry, *temp; - - list_for_each_entry_safe(entry, temp, locked, list) { - CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh); - tgt_extent_unlock(&entry->lh, LCK_PR); - list_del(&entry->list); - OBD_FREE_PTR(entry); - } -} /** * OFD request handler for OST_GET_INFO RPC. @@ -1141,6 +1085,7 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) void *key; int keylen; int replylen, rc = 0; + ktime_t kstart = ktime_get(); ENTRY; @@ -1207,19 +1152,12 @@ static int ofd_get_info_hdl(struct tgt_session_info *tsi) * flushed back from client, then call fiemap again. */ if (fm_key->lfik_oa.o_valid & OBD_MD_FLFLAGS && fm_key->lfik_oa.o_flags & OBD_FL_SRVLOCK) { - struct list_head locked; - - INIT_LIST_HEAD(&locked); ost_fid_build_resid(fid, &fti->fti_resid); - rc = lock_zero_regions(ofd->ofd_namespace, - &fti->fti_resid, fiemap, - &locked); - if (rc == 0 && !list_empty(&locked)) { + rc = lock_zero_regions(tsi->tsi_env, ofd->ofd_namespace, + &fti->fti_resid, fiemap); + if (rc == 0) rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid, fiemap); - unlock_zero_regions(ofd->ofd_namespace, - &locked); - } } } else if (KEY_IS(KEY_LAST_FID)) { struct ofd_device *ofd = ofd_exp(exp); @@ -1262,7 +1200,7 @@ out_put: rc = -EOPNOTSUPP; } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GET_INFO, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1287,6 +1225,7 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) struct ofd_object *fo; __u64 flags = 0; enum ldlm_mode lock_mode = LCK_PR; + ktime_t kstart = ktime_get(); bool srvlock; int rc; ENTRY; @@ -1307,7 +1246,8 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) if (unlikely(tsi->tsi_ost_body->oa.o_flags & OBD_FL_FLUSH)) lock_mode = LCK_PW; - rc = tgt_extent_lock(tsi->tsi_tgt->lut_obd->obd_namespace, + rc = tgt_extent_lock(tsi->tsi_env, + tsi->tsi_tgt->lut_obd->obd_namespace, &tsi->tsi_resid, 0, OBD_OBJECT_EOF, &lh, lock_mode, &flags); if (rc != 0) @@ -1323,8 +1263,7 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) __u64 curr_version; obdo_from_la(&repbody->oa, &fti->fti_attr, - OFD_VALID_FLAGS | LA_UID | LA_GID); - tgt_drop_id(tsi->tsi_exp, &repbody->oa); + OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID); /* Store object version in reply */ curr_version = dt_version_get(tsi->tsi_env, @@ -1333,15 +1272,25 @@ static int ofd_getattr_hdl(struct tgt_session_info *tsi) repbody->oa.o_valid |= OBD_MD_FLDATAVERSION; repbody->oa.o_data_version = curr_version; } + + if (fo->ofo_ff.ff_layout_version > 0) { + repbody->oa.o_valid |= OBD_MD_LAYOUT_VERSION; + repbody->oa.o_layout_version = + fo->ofo_ff.ff_layout_version + fo->ofo_ff.ff_range; + + CDEBUG(D_INODE, DFID": get layout version: %u\n", + PFID(&tsi->tsi_fid), + repbody->oa.o_layout_version); + } } ofd_object_put(tsi->tsi_env, fo); out: if (srvlock) - tgt_extent_unlock(&lh, lock_mode); + tgt_data_unlock(&lh, lock_mode); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_GETATTR, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); repbody->oa.o_valid |= OBD_MD_FLFLAGS; repbody->oa.o_flags = OBD_FL_FLUSH; @@ -1368,7 +1317,7 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi) struct ost_body *repbody; struct ldlm_resource *res; struct ofd_object *fo; - struct filter_fid *ff = NULL; + ktime_t kstart = ktime_get(); int rc = 0; ENTRY; @@ -1408,22 +1357,16 @@ static int ofd_setattr_hdl(struct tgt_session_info *tsi) la_from_obdo(&fti->fti_attr, &body->oa, body->oa.o_valid); fti->fti_attr.la_valid &= ~LA_TYPE; - if (body->oa.o_valid & OBD_MD_FLFID) { - ff = &fti->fti_mds_fid; - ofd_prepare_fidea(ff, &body->oa); - } - /* setting objects attributes (including owner/group) */ - rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, ff); + rc = ofd_attr_set(tsi->tsi_env, fo, &fti->fti_attr, &body->oa); if (rc != 0) GOTO(out_put, rc); obdo_from_la(&repbody->oa, &fti->fti_attr, - OFD_VALID_FLAGS | LA_UID | LA_GID); - tgt_drop_id(tsi->tsi_exp, &repbody->oa); + OFD_VALID_FLAGS | LA_UID | LA_GID | LA_PROJID); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SETATTR, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); EXIT; out_put: ofd_object_put(tsi->tsi_env, fo); @@ -1528,8 +1471,7 @@ done: rc = ofd_seq_last_oid_write(env, ofd, oseq); } else { /* don't reuse orphan object, return last used objid */ - ostid_set_id(oi, last); - rc = 0; + rc = ostid_set_id(oi, last); } GOTO(out_put, rc); @@ -1561,15 +1503,20 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) u64 seq = ostid_seq(&oa->o_oi); u64 oid = ostid_id(&oa->o_oi); struct ofd_seq *oseq; - int rc = 0, diff; int sync_trans = 0; long granted = 0; + ktime_t kstart = ktime_get(); + s64 diff; + int rc = 0; ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) RETURN(-EROFS); + if (ofd->ofd_no_precreate) + return -EPERM; + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); if (repbody == NULL) RETURN(-ENOMEM); @@ -1584,14 +1531,13 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) rep_oa = &repbody->oa; rep_oa->o_oi = oa->o_oi; - LASSERT(seq >= FID_SEQ_OST_MDT0); LASSERT(oa->o_valid & OBD_MD_FLGROUP); CDEBUG(D_INFO, "ofd_create("DOSTID")\n", POSTID(&oa->o_oi)); oseq = ofd_seq_load(tsi->tsi_env, ofd, seq); if (IS_ERR(oseq)) { - CERROR("%s: Can't find FID Sequence "LPX64": rc = %ld\n", + CERROR("%s: Can't find FID Sequence %#llx: rc = %ld\n", ofd_name(ofd), seq, PTR_ERR(oseq)); GOTO(out_sem, rc = -EINVAL); } @@ -1600,7 +1546,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) (oa->o_flags & OBD_FL_RECREATE_OBJS)) { if (!ofd_obd(ofd)->obd_recovering || oid > ofd_seq_last_oid(oseq)) { - CERROR("%s: recreate objid "DOSTID" > last id "LPU64 + CERROR("%s: recreate objid "DOSTID" > last id %llu" "\n", ofd_name(ofd), POSTID(&oa->o_oi), ofd_seq_last_oid(oseq)); GOTO(out_nolock, rc = -EINVAL); @@ -1625,18 +1571,27 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) oseq->os_destroys_in_progress = 1; mutex_lock(&oseq->os_create_lock); if (!oseq->os_destroys_in_progress) { - CERROR("%s:["LPU64"] destroys_in_progress already" - " cleared\n", ofd_name(ofd), seq); - ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); - GOTO(out, rc = 0); + CDEBUG(D_HA, + "%s:[%llu] destroys_in_progress already cleared\n", + ofd_name(ofd), seq); + rc = ostid_set_id(&rep_oa->o_oi, + ofd_seq_last_oid(oseq)); + GOTO(out, rc); } diff = oid - ofd_seq_last_oid(oseq); - CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n", - ofd_seq_last_oid(oseq), diff); + CDEBUG(D_HA, "ofd_last_id() = %llu -> diff = %lld\n", + ofd_seq_last_oid(oseq), diff); if (-diff > OST_MAX_PRECREATE) { + LCONSOLE(D_INFO, "%s: too large difference between MDS " + "LAST_ID "DFID" (%llu) and OST LAST_ID "DFID" " + "(%llu), trust the OST\n", + ofd_name(ofd), PFID(&oa->o_oi.oi_fid), oid, + PFID(&oseq->os_oi.oi_fid), + ofd_seq_last_oid(oseq)); + /* Let MDS know that we are so far ahead. */ - ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq) + 1); - rc = 0; + rc = ostid_set_id(&rep_oa->o_oi, + ofd_seq_last_oid(oseq) + 1); } else if (diff < 0) { rc = ofd_orphans_destroy(tsi->tsi_env, exp, ofd, rep_oa); @@ -1681,34 +1636,44 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) GOTO(out, rc = -EINVAL); } - if (diff < 0) { + if (diff <= -OST_MAX_PRECREATE) { /* LU-5648 */ CERROR("%s: invalid precreate request for " - DOSTID", last_id " LPU64 ". " + DOSTID", last_id %llu. " "Likely MDS last_id corruption\n", ofd_name(ofd), POSTID(&oa->o_oi), ofd_seq_last_oid(oseq)); GOTO(out, rc = -EINVAL); + } else if (diff < 0) { + LCONSOLE(D_INFO, + "%s: MDS LAST_ID "DFID" (%llu) is %lld behind OST LAST_ID "DFID" (%llu), trust the OST\n", + ofd_name(ofd), PFID(&oa->o_oi.oi_fid), + oid, -diff, PFID(&oseq->os_oi.oi_fid), + ofd_seq_last_oid(oseq)); + /* Let MDS know that we are so far ahead. */ + rc = ostid_set_id(&rep_oa->o_oi, + ofd_seq_last_oid(oseq) + 1); } } } if (diff > 0) { - cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT); - u64 next_id; - int created = 0; - int count; + time64_t enough_time = ktime_get_seconds() + DISK_TIMEOUT; + u64 next_id; + int created = 0; + int count; + int rc2; if (!(oa->o_valid & OBD_MD_FLFLAGS) || !(oa->o_flags & OBD_FL_DELORPHAN)) { /* don't enforce grant during orphan recovery */ - granted = ofd_grant_create(tsi->tsi_env, - ofd_obd(ofd)->obd_self_export, - &diff); + granted = tgt_grant_create(tsi->tsi_env, + ofd_obd(ofd)->obd_self_export, + &diff); if (granted < 0) { rc = granted; granted = 0; CDEBUG(D_HA, "%s: failed to acquire grant " - "space for precreate (%d): rc = %d\n", + "space for precreate (%lld): rc = %d\n", ofd_name(ofd), diff, rc); diff = 0; } @@ -1720,32 +1685,26 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) * (possibly filling the OST), only precreate the last batch. * LFSCK will eventually clean up any orphans. LU-14 */ if (diff > 5 * OST_MAX_PRECREATE) { + /* Message below is checked in conf-sanity test_122b */ + LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %lld higher than LAST_ID "DOSTID", only precreating the last %u objects. OST replaced or reformatted?\n", + ofd_name(ofd), POSTID(&oa->o_oi), diff, + POSTID(&oseq->os_oi), + OST_MAX_PRECREATE / 2); diff = OST_MAX_PRECREATE / 2; - LCONSOLE_WARN("%s: Too many FIDs to precreate " - "OST replaced or reformatted: " - "LFSCK will clean up", - ofd_name(ofd)); - - CDEBUG(D_HA, "%s: precreate FID "DOSTID" is over " - "%u larger than the LAST_ID "DOSTID", only " - "precreating the last %u objects.\n", - ofd_name(ofd), POSTID(&oa->o_oi), - 5 * OST_MAX_PRECREATE, - POSTID(&oseq->os_oi), diff); ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff); } while (diff > 0) { next_id = ofd_seq_last_oid(oseq) + 1; - count = ofd_precreate_batch(ofd, diff); + count = ofd_precreate_batch(ofd, (int)diff); - CDEBUG(D_HA, "%s: reserve %d objects in group "LPX64 - " at "LPU64"\n", ofd_name(ofd), + CDEBUG(D_HA, "%s: reserve %d objects in group %#llx" + " at %llu\n", ofd_name(ofd), count, seq, next_id); if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) - && cfs_time_after(jiffies, enough_time)) { - CDEBUG(D_HA, "%s: Slow creates, %d/%d objects" + && ktime_get_seconds() > enough_time) { + CDEBUG(D_HA, "%s: Slow creates, %d/%lld objects" " created at a rate of %d/s\n", ofd_name(ofd), created, diff + created, created / DISK_TIMEOUT); @@ -1766,7 +1725,7 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) LCONSOLE_WARN("%s: can't create the same count of" " objects when replaying the request" - " (diff is %d). see LU-4621\n", + " (diff is %lld). see LU-4621\n", ofd_name(ofd), diff); if (created > 0) @@ -1779,31 +1738,24 @@ static int ofd_create_hdl(struct tgt_session_info *tsi) if (!(oa->o_valid & OBD_MD_FLFLAGS) || !(oa->o_flags & OBD_FL_DELORPHAN)) { - ofd_grant_commit(ofd_obd(ofd)->obd_self_export, granted, - rc); + tgt_grant_commit(ofd_obd(ofd)->obd_self_export, + granted, rc); granted = 0; } - ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); + rc2 = ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq)); + rc = rc ? : rc2; } EXIT; ofd_counter_incr(exp, LPROC_OFD_STATS_CREATE, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); + if (unlikely(!oseq->os_last_id_synced)) + oseq->os_last_id_synced = 1; out: mutex_unlock(&oseq->os_create_lock); out_nolock: - if (rc == 0) { -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0) - struct ofd_thread_info *info = ofd_info(tsi->tsi_env); - struct lu_fid *fid = &info->fti_fid; - - /* For compatible purpose, it needs to convert back to - * OST ID before put it on wire. */ - *fid = rep_oa->o_oi.oi_fid; - fid_to_ostid(fid, &rep_oa->o_oi); -#endif + if (rc == 0) rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP; - } ofd_seq_put(tsi->tsi_env, oseq); out_sem: @@ -1829,6 +1781,7 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi) struct ofd_device *ofd = ofd_exp(tsi->tsi_exp); struct ofd_thread_info *fti = tsi2ofd_info(tsi); struct lu_fid *fid = &fti->fti_fid; + ktime_t kstart = ktime_get(); u64 oid; u32 count; int rc = 0; @@ -1890,7 +1843,7 @@ static int ofd_destroy_hdl(struct tgt_session_info *tsi) } ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_DESTROY, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); GOTO(out, rc); @@ -1912,15 +1865,18 @@ out: */ static int ofd_statfs_hdl(struct tgt_session_info *tsi) { + ktime_t kstart = ktime_get(); struct obd_statfs *osfs; int rc; ENTRY; + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_STATFS_DELAY, 10); + osfs = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_STATFS); rc = ofd_statfs(tsi->tsi_env, tsi->tsi_exp, osfs, - cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0); + ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS, 0); if (rc != 0) CERROR("%s: statfs failed: rc = %d\n", tgt_name(tsi->tsi_tgt), rc); @@ -1929,7 +1885,7 @@ static int ofd_statfs_hdl(struct tgt_session_info *tsi) rc = -EINPROGRESS; ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_STATFS, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); RETURN(rc); } @@ -1952,6 +1908,7 @@ static int ofd_sync_hdl(struct tgt_session_info *tsi) struct ofd_thread_info *fti = tsi2ofd_info(tsi); struct ofd_device *ofd = ofd_exp(tsi->tsi_exp); struct ofd_object *fo = NULL; + ktime_t kstart = ktime_get(); int rc = 0; ENTRY; @@ -1972,7 +1929,7 @@ static int ofd_sync_hdl(struct tgt_session_info *tsi) GOTO(put, rc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_SYNC, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (fo == NULL) RETURN(0); @@ -1994,6 +1951,129 @@ put: } /** + * OFD request handler for OST_FALLOCATE RPC. + * + * This is part of request processing. Validate request fields, + * preallocate the given OFD object and pack reply. + * + * \param[in] tsi target session environment for this request + * + * \retval 0 if successful + * \retval negative value on error + */ +static int ofd_fallocate_hdl(struct tgt_session_info *tsi) +{ + struct obdo *oa = &tsi->tsi_ost_body->oa; + struct ost_body *repbody; + struct ofd_thread_info *info = tsi2ofd_info(tsi); + struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace; + struct ldlm_resource *res; + struct ofd_object *fo; + __u64 flags = 0; + __u64 valid; + struct lustre_handle lh = { 0, }; + int rc, mode; + __u64 start, end; + bool srvlock; + ktime_t kstart = ktime_get(); + + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); + if (repbody == NULL) + RETURN(err_serious(-ENOMEM)); + + /* + * fallocate start and end are passed in o_size, o_blocks + * on the wire. + */ + if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != + (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) + RETURN(err_serious(-EPROTO)); + + start = oa->o_size; + end = oa->o_blocks; + mode = oa->o_falloc_mode; + /* + * mode == 0 (which is standard prealloc) and PUNCH is supported + * Rest of mode options are not supported yet. + */ + if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) + RETURN(-EOPNOTSUPP); + + /* PUNCH_HOLE mode should always be accompanied with KEEP_SIZE flag + * Check that and add the missing flag for such invalid call with + * warning. + */ + if (mode & FALLOC_FL_PUNCH_HOLE && !(mode & FALLOC_FL_KEEP_SIZE)) { + CWARN("%s: PUNCH mode misses KEEP_SIZE flag, setting it\n", + tsi->tsi_tgt->lut_obd->obd_name); + mode |= FALLOC_FL_KEEP_SIZE; + } + + repbody->oa.o_oi = oa->o_oi; + repbody->oa.o_valid = OBD_MD_FLID; + + srvlock = oa->o_valid & OBD_MD_FLFLAGS && + oa->o_flags & OBD_FL_SRVLOCK; + + if (srvlock) { + rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, + start, end, &lh, LCK_PW, &flags); + if (rc != 0) + RETURN(rc); + } + + fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp), + &tsi->tsi_fid); + if (IS_ERR(fo)) + GOTO(out, rc = PTR_ERR(fo)); + + valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID | + OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME; + la_from_obdo(&info->fti_attr, oa, valid); + + rc = ofd_object_fallocate(tsi->tsi_env, fo, start, end, mode, + &info->fti_attr, oa); + if (rc) + GOTO(out_put, rc); + + rc = ofd_attr_get(tsi->tsi_env, fo, &info->fti_attr); + if (rc == 0) + obdo_from_la(&repbody->oa, &info->fti_attr, OFD_VALID_FLAGS); + else + rc = 0; + + ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PREALLOC, + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); + + EXIT; +out_put: + ofd_object_put(tsi->tsi_env, fo); +out: + if (srvlock) + tgt_data_unlock(&lh, LCK_PW); + if (rc == 0) { + res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid, + LDLM_EXTENT, 0); + if (!IS_ERR(res)) { + struct ost_lvb *res_lvb; + + ldlm_res_lvbo_update(res, NULL, 0); + res_lvb = res->lr_lvb_data; + /* Blocks */ + repbody->oa.o_valid |= OBD_MD_FLBLOCKS; + repbody->oa.o_blocks = res_lvb->lvb_blocks; + /* Size */ + repbody->oa.o_valid |= OBD_MD_FLSIZE; + repbody->oa.o_size = res_lvb->lvb_size; + + ldlm_resource_putref(res); + } + } + + RETURN(rc); +} + +/** * OFD request handler for OST_PUNCH RPC. * * This is part of request processing. Validate request fields, @@ -2012,17 +2092,16 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace; struct ldlm_resource *res; struct ofd_object *fo; - struct filter_fid *ff = NULL; __u64 flags = 0; struct lustre_handle lh = { 0, }; - int rc; __u64 start, end; bool srvlock; + ktime_t kstart = ktime_get(); + int rc; ENTRY; - /* check that we do support OBD_CONNECT_TRUNCLOCK. */ - CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK); + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_PAUSE_PUNCH, cfs_fail_val); if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) @@ -2051,14 +2130,14 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) oa->o_flags & OBD_FL_SRVLOCK; if (srvlock) { - rc = tgt_extent_lock(ns, &tsi->tsi_resid, start, end, &lh, - LCK_PW, &flags); + rc = tgt_extent_lock(tsi->tsi_env, ns, &tsi->tsi_resid, start, + end, &lh, LCK_PW, &flags); if (rc != 0) RETURN(rc); } - CDEBUG(D_INODE, "calling punch for object "DFID", valid = "LPX64 - ", start = "LPD64", end = "LPD64"\n", PFID(&tsi->tsi_fid), + CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx" + ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid), oa->o_valid, start, end); fo = ofd_object_find_exists(tsi->tsi_env, ofd_exp(tsi->tsi_exp), @@ -2071,24 +2150,19 @@ static int ofd_punch_hdl(struct tgt_session_info *tsi) info->fti_attr.la_size = start; info->fti_attr.la_valid |= LA_SIZE; - if (oa->o_valid & OBD_MD_FLFID) { - ff = &info->fti_mds_fid; - ofd_prepare_fidea(ff, oa); - } - rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr, - ff, (struct obdo *)oa); + (struct obdo *)oa); if (rc) GOTO(out_put, rc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_PUNCH, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); EXIT; out_put: ofd_object_put(tsi->tsi_env, fo); out: if (srvlock) - tgt_extent_unlock(&lh, LCK_PW); + tgt_data_unlock(&lh, LCK_PW); if (rc == 0) { /* we do not call this before to avoid lu_object_find() in * ->lvbo_update() holding another reference on the object. @@ -2098,13 +2172,73 @@ out: res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid, LDLM_EXTENT, 0); if (!IS_ERR(res)) { + struct ost_lvb *res_lvb; + ldlm_res_lvbo_update(res, NULL, 0); + res_lvb = res->lr_lvb_data; + repbody->oa.o_valid |= OBD_MD_FLBLOCKS; + repbody->oa.o_blocks = res_lvb->lvb_blocks; ldlm_resource_putref(res); } } return rc; } +static int ofd_ladvise_prefetch(const struct lu_env *env, + struct ofd_object *fo, + struct niobuf_local *lnb, + __u64 start, __u64 end, enum dt_bufs_type dbt) +{ + struct ofd_thread_info *info = ofd_info(env); + pgoff_t start_index, end_index, pages; + struct niobuf_remote rnb; + unsigned long nr_local; + int rc = 0; + + if (end <= start) + RETURN(-EINVAL); + + ofd_read_lock(env, fo); + if (!ofd_object_exists(fo)) + GOTO(out_unlock, rc = -ENOENT); + + rc = ofd_attr_get(env, fo, &info->fti_attr); + if (rc) + GOTO(out_unlock, rc); + + if (end > info->fti_attr.la_size) + end = info->fti_attr.la_size; + + if (end <= start) + GOTO(out_unlock, rc); + + /* We need page aligned offset and length */ + start_index = start >> PAGE_SHIFT; + end_index = (end - 1) >> PAGE_SHIFT; + pages = end_index - start_index + 1; + while (pages > 0) { + nr_local = pages <= PTLRPC_MAX_BRW_PAGES ? pages : + PTLRPC_MAX_BRW_PAGES; + rnb.rnb_offset = start_index << PAGE_SHIFT; + rnb.rnb_len = nr_local << PAGE_SHIFT; + rc = dt_bufs_get(env, ofd_object_child(fo), &rnb, lnb, + PTLRPC_MAX_BRW_PAGES, dbt); + if (unlikely(rc < 0)) + break; + nr_local = rc; + rc = dt_read_prep(env, ofd_object_child(fo), lnb, nr_local); + dt_bufs_put(env, ofd_object_child(fo), lnb, nr_local); + if (unlikely(rc)) + break; + start_index += nr_local; + pages -= nr_local; + } + +out_unlock: + ofd_read_unlock(env, fo); + RETURN(rc); +} + /** * OFD request handler for OST_LADVISE RPC. * @@ -2117,20 +2251,30 @@ out: */ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) { - struct ptlrpc_request *req = tgt_ses_req(tsi); - struct obd_export *exp = tsi->tsi_exp; - struct ofd_device *ofd = ofd_exp(exp); - struct ost_body *body, *repbody; - struct ofd_thread_info *info; - struct ofd_object *fo; - const struct lu_env *env = req->rq_svc_thread->t_env; - int rc = 0; - struct lu_ladvise *ladvise; - int num_advise; - struct ladvise_hdr *ladvise_hdr; - int i; + struct ptlrpc_request *req = tgt_ses_req(tsi); + struct obd_export *exp = tsi->tsi_exp; + struct ofd_device *ofd = ofd_exp(exp); + struct ost_body *body, *repbody; + struct ofd_thread_info *info; + struct ofd_object *fo; + struct ptlrpc_thread *svc_thread = req->rq_svc_thread; + const struct lu_env *env = svc_thread->t_env; + struct tgt_thread_big_cache *tbc = svc_thread->t_data; + enum dt_bufs_type dbt = DT_BUFS_TYPE_READAHEAD; + struct lu_ladvise *ladvise; + int num_advise; + struct ladvise_hdr *ladvise_hdr; + struct obd_ioobj ioo; + struct lustre_handle lockh = { 0 }; + __u64 flags = 0; + int i; + struct dt_object *dob; + __u64 start; + __u64 end; + int rc = 0; ENTRY; + CFS_FAIL_TIMEOUT(OBD_FAIL_OST_LADVISE_PAUSE, cfs_fail_val); body = tsi->tsi_ost_body; if ((body->oa.o_valid & OBD_MD_FLID) != OBD_MD_FLID) @@ -2154,7 +2298,7 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) num_advise = req_capsule_get_size(&req->rq_pill, &RMF_OST_LADVISE, RCL_CLIENT) / - sizeof(*ladvise); + sizeof(*ladvise); if (num_advise < ladvise_hdr->lah_count) RETURN(err_serious(-EPROTO)); @@ -2174,9 +2318,15 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) RETURN(rc); } LASSERT(fo != NULL); + dob = ofd_object_child(fo); + + if (ptlrpc_connection_is_local(exp->exp_connection)) + dbt |= DT_BUFS_TYPE_LOCAL; for (i = 0; i < num_advise; i++, ladvise++) { - if (ladvise->lla_end <= ladvise->lla_start) { + start = ladvise->lla_start; + end = ladvise->lla_end; + if (end <= start) { rc = err_serious(-EPROTO); break; } @@ -2186,6 +2336,27 @@ static int ofd_ladvise_hdl(struct tgt_session_info *tsi) default: rc = -ENOTSUPP; break; + case LU_LADVISE_WILLREAD: + if (tbc == NULL) + RETURN(-ENOMEM); + + ioo.ioo_oid = body->oa.o_oi; + ioo.ioo_bufcnt = 1; + rc = tgt_extent_lock(env, exp->exp_obd->obd_namespace, + &tsi->tsi_resid, start, end - 1, + &lockh, LCK_PR, &flags); + if (rc != 0) + break; + + req->rq_status = ofd_ladvise_prefetch(env, fo, + tbc->local, + start, end, dbt); + tgt_data_unlock(&lockh, LCK_PR); + break; + case LU_LADVISE_DONTNEED: + rc = dt_ladvise(env, dob, ladvise->lla_start, + ladvise->lla_end, LU_LADVISE_DONTNEED); + break; } if (rc != 0) break; @@ -2211,6 +2382,7 @@ static int ofd_quotactl(struct tgt_session_info *tsi) { struct obd_quotactl *oqctl, *repoqc; struct lu_nodemap *nodemap; + ktime_t kstart = ktime_get(); int id; int rc; @@ -2239,6 +2411,10 @@ static int ofd_quotactl(struct tgt_session_info *tsi) id = nodemap_map_id(nodemap, NODEMAP_GID, NODEMAP_CLIENT_TO_FS, repoqc->qc_id); + else if (oqctl->qc_type == PRJQUOTA) + id = nodemap_map_id(nodemap, NODEMAP_PROJID, + NODEMAP_CLIENT_TO_FS, + repoqc->qc_id); nodemap_putref(nodemap); @@ -2248,7 +2424,7 @@ static int ofd_quotactl(struct tgt_session_info *tsi) rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc); ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL, - tsi->tsi_jobid, 1); + tsi->tsi_jobid, ktime_us_delta(ktime_get(), kstart)); if (repoqc->qc_id != id) swap(repoqc->qc_id, id); @@ -2257,71 +2433,6 @@ static int ofd_quotactl(struct tgt_session_info *tsi) } /** - * Calculate the amount of time for lock prolongation. - * - * This is helper for ofd_prolong_extent_locks() function to get - * the timeout extra time. - * - * \param[in] req current request - * - * \retval amount of time to extend the timeout with - */ -static inline int prolong_timeout(struct ptlrpc_request *req, - struct ldlm_lock *lock) -{ - struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt; - - if (AT_OFF) - return obd_timeout / 2; - - /* We are in the middle of the process - BL AST is sent, CANCEL - is ahead. Take half of AT + IO process time. */ - return at_est2timeout(at_get(&svcpt->scp_at_estimate)) + - (ldlm_bl_timeout(lock) >> 1); -} - -/** - * Prolong single lock timeout. - * - * This is supplemental function to the ofd_prolong_locks(). It prolongs - * a single lock. - * - * \param[in] tsi target session environment for this request - * \param[in] lock LDLM lock to prolong - * \param[in] extent related extent - * \param[in] timeout timeout value to add - * - * \retval 0 if lock is not suitable for prolongation - * \retval 1 if lock was prolonged successfully - */ -static int ofd_prolong_one_lock(struct tgt_session_info *tsi, - struct ldlm_lock *lock, - struct ldlm_extent *extent) -{ - int timeout = prolong_timeout(tgt_ses_req(tsi), lock); - - if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */ - return 0; - - /* XXX: never try to grab resource lock here because we're inside - * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take - * res lock and then exp_bl_list_lock. */ - - if (!(lock->l_flags & LDLM_FL_AST_SENT)) - /* ignore locks not being cancelled */ - return 0; - - LDLM_DEBUG(lock, "refreshed for req x"LPU64" ext("LPU64"->"LPU64") " - "to %ds.\n", tgt_ses_req(tsi)->rq_xid, extent->start, - extent->end, timeout); - - /* OK. this is a possible lock the user holds doing I/O - * let's refresh eviction timer for it */ - ldlm_refresh_waiting_lock(lock, timeout); - return 1; -} - -/** * Prolong lock timeout for the given extent. * * This function finds all locks related with incoming request and @@ -2341,25 +2452,26 @@ static int ofd_prolong_one_lock(struct tgt_session_info *tsi, * request may cover multiple locks. * * \param[in] tsi target session environment for this request - * \param[in] start start of extent - * \param[in] end end of extent + * \param[in] data struct of data to prolong locks * - * \retval number of prolonged locks */ -static int ofd_prolong_extent_locks(struct tgt_session_info *tsi, - __u64 start, __u64 end) +static void ofd_prolong_extent_locks(struct tgt_session_info *tsi, + struct ldlm_prolong_args *data) { - struct obd_export *exp = tsi->tsi_exp; struct obdo *oa = &tsi->tsi_ost_body->oa; - struct ldlm_extent extent = { - .start = start, - .end = end - }; struct ldlm_lock *lock; - int lock_count = 0; ENTRY; + data->lpa_timeout = prolong_timeout(tgt_ses_req(tsi)); + data->lpa_export = tsi->tsi_exp; + data->lpa_resid = tsi->tsi_resid; + + CDEBUG(D_RPCTRACE, "Prolong locks for req %p with x%llu" + " ext(%llu->%llu)\n", tgt_ses_req(tsi), + tgt_ses_req(tsi)->rq_xid, data->lpa_extent.start, + data->lpa_extent.end); + if (oa->o_valid & OBD_MD_FLHANDLE) { /* mostly a request should be covered by only one lock, try * fast path. */ @@ -2367,42 +2479,25 @@ static int ofd_prolong_extent_locks(struct tgt_session_info *tsi, if (lock != NULL) { /* Fast path to check if the lock covers the whole IO * region exclusively. */ - if (lock->l_granted_mode == LCK_PW && - ldlm_extent_contain(&lock->l_policy_data.l_extent, - &extent)) { + if (ldlm_extent_contain(&lock->l_policy_data.l_extent, + &data->lpa_extent)) { /* bingo */ - LASSERT(lock->l_export == exp); - lock_count = ofd_prolong_one_lock(tsi, lock, - &extent); + LASSERT(lock->l_export == data->lpa_export); + ldlm_lock_prolong_one(lock, data); + LDLM_LOCK_PUT(lock); + if (data->lpa_locks_cnt > 0) + RETURN_EXIT; + /* The lock was destroyed probably lets try + * resource tree. */ + } else { + lock->l_last_used = ktime_get(); LDLM_LOCK_PUT(lock); - RETURN(lock_count); } - lock->l_last_used = cfs_time_current(); - LDLM_LOCK_PUT(lock); } } - spin_lock_bh(&exp->exp_bl_list_lock); - list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) { - LASSERT(lock->l_flags & LDLM_FL_AST_SENT); - LASSERT(lock->l_resource->lr_type == LDLM_EXTENT); - - /* ignore waiting locks, no more granted locks in the list */ - if (lock->l_granted_mode != lock->l_req_mode) - break; - - if (!ldlm_res_eq(&tsi->tsi_resid, &lock->l_resource->lr_name)) - continue; - - if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent, - &extent)) - continue; - - lock_count += ofd_prolong_one_lock(tsi, lock, &extent); - } - spin_unlock_bh(&exp->exp_bl_list_lock); - - RETURN(lock_count); + ldlm_resource_prolong(data); + EXIT; } /** @@ -2449,8 +2544,10 @@ static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req, if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name)) RETURN(0); - /* a bulk write can only hold a reference on a PW extent lock */ - mode = LCK_PW; + /* a bulk write can only hold a reference on a PW extent lock + * or GROUP lock. + */ + mode = LCK_PW | LCK_GROUP; if (opc == OST_READ) /* whereas a bulk read can be protected by either a PR or PW * extent lock */ @@ -2466,20 +2563,22 @@ static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req, * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests. * * Check for whether the given PTLRPC request (\a req) is blocking - * an LDLM lock cancel. + * an LDLM lock cancel. Also checks whether the request is covered by an LDLM + * lock. * * \param[in] req the incoming request * * \retval 1 if \a req is blocking an LDLM lock cancel * \retval 0 if it is not + * \retval -ESTALE if lock is not found */ static int ofd_rw_hpreq_check(struct ptlrpc_request *req) { struct tgt_session_info *tsi; struct obd_ioobj *ioo; struct niobuf_remote *rnb; - __u64 start, end; - int lock_count; + int opc; + struct ldlm_prolong_args pa = { 0 }; ENTRY; @@ -2491,6 +2590,9 @@ static int ofd_rw_hpreq_check(struct ptlrpc_request *req) * Use LASSERT below because malformed RPCs should have * been filtered out in tgt_hpreq_handler(). */ + opc = lustre_msg_get_opc(req->rq_reqmsg); + LASSERT(opc == OST_READ || opc == OST_WRITE); + ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ); LASSERT(ioo != NULL); @@ -2498,21 +2600,28 @@ static int ofd_rw_hpreq_check(struct ptlrpc_request *req) LASSERT(rnb != NULL); LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK)); - start = rnb->rnb_offset; + pa.lpa_mode = LCK_PW | LCK_GROUP; + if (opc == OST_READ) + pa.lpa_mode |= LCK_PR; + + pa.lpa_extent.start = rnb->rnb_offset; rnb += ioo->ioo_bufcnt - 1; - end = rnb->rnb_offset + rnb->rnb_len - 1; + pa.lpa_extent.end = rnb->rnb_offset + rnb->rnb_len - 1; - DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID - " ("LPU64"->"LPU64")\n", - tgt_name(tsi->tsi_tgt), current->comm, - PFID(&tsi->tsi_fid), start, end); + DEBUG_REQ(D_RPCTRACE, req, + "%s %s: refresh rw locks for "DFID" (%llu->%llu)", + tgt_name(tsi->tsi_tgt), current->comm, PFID(&tsi->tsi_fid), + pa.lpa_extent.start, pa.lpa_extent.end); - lock_count = ofd_prolong_extent_locks(tsi, start, end); + ofd_prolong_extent_locks(tsi, &pa); - CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n", - tgt_name(tsi->tsi_tgt), lock_count, req); + CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p\n", + tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req); + + if (pa.lpa_blocks_cnt > 0) + RETURN(1); - RETURN(lock_count > 0); + RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE); } /** @@ -2546,6 +2655,10 @@ static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req, struct ldlm_lock *lock) { struct tgt_session_info *tsi; + struct obdo *oa; + struct ldlm_extent ext; + + ENTRY; /* Don't use tgt_ses_info() to get session info, because lock_match() * can be called while request has no processing thread yet. */ @@ -2558,27 +2671,42 @@ static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req, LASSERT(tsi->tsi_ost_body != NULL); if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE && tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie) - return 1; + RETURN(1); - return 0; + oa = &tsi->tsi_ost_body->oa; + ext.start = oa->o_size; + ext.end = oa->o_blocks; + + LASSERT(lock->l_resource != NULL); + if (!ostid_res_name_eq(&oa->o_oi, &lock->l_resource->lr_name)) + RETURN(0); + + if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP))) + RETURN(0); + + RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext)); } /** * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request. * * High-priority queue request check for whether the given punch request - * (\a req) is blocking an LDLM lock cancel. + * (\a req) is blocking an LDLM lock cancel. Also checks whether the request is + * covered by an LDLM lock. + * + * * \param[in] req the incoming request * * \retval 1 if \a req is blocking an LDLM lock cancel * \retval 0 if it is not + * \retval -ESTALE if lock is not found */ static int ofd_punch_hpreq_check(struct ptlrpc_request *req) { struct tgt_session_info *tsi; struct obdo *oa; - int lock_count; + struct ldlm_prolong_args pa = { 0 }; ENTRY; @@ -2591,17 +2719,24 @@ static int ofd_punch_hpreq_check(struct ptlrpc_request *req) LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & OBD_FL_SRVLOCK)); + pa.lpa_mode = LCK_PW | LCK_GROUP; + pa.lpa_extent.start = oa->o_size; + pa.lpa_extent.end = oa->o_blocks; + CDEBUG(D_DLMTRACE, - "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n", + "%s: refresh locks: %llu/%llu (%llu->%llu)\n", tgt_name(tsi->tsi_tgt), tsi->tsi_resid.name[0], - tsi->tsi_resid.name[1], oa->o_size, oa->o_blocks); + tsi->tsi_resid.name[1], pa.lpa_extent.start, pa.lpa_extent.end); - lock_count = ofd_prolong_extent_locks(tsi, oa->o_size, oa->o_blocks); + ofd_prolong_extent_locks(tsi, &pa); CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n", - tgt_name(tsi->tsi_tgt), lock_count, req); + tgt_name(tsi->tsi_tgt), pa.lpa_blocks_cnt, req); - RETURN(lock_count > 0); + if (pa.lpa_blocks_cnt > 0) + RETURN(1); + + RETURN(pa.lpa_locks_cnt > 0 ? 0 : -ESTALE); } /** @@ -2652,7 +2787,9 @@ static void ofd_hp_brw(struct tgt_session_info *tsi) LASSERT(rnb != NULL); /* must exist after request preprocessing */ /* no high priority if server lock is needed */ - if (rnb->rnb_flags & OBD_BRW_SRVLOCK) + if (rnb->rnb_flags & OBD_BRW_SRVLOCK || + (lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) + & MSG_REPLAY)) return; } tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw; @@ -2671,8 +2808,10 @@ static void ofd_hp_punch(struct tgt_session_info *tsi) { LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */ /* no high-priority if server lock is needed */ - if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS && - tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) + if ((tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS && + tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK) || + tgt_conn_flags(tsi) & OBD_CONNECT_MDS || + lustre_msg_get_flags(tgt_ses_req(tsi)->rq_reqmsg) & MSG_REPLAY) return; tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_punch; } @@ -2701,26 +2840,27 @@ TGT_RPC_HANDLER(OST_FIRST_OPC, 0, OST_SET_INFO, ofd_set_info_hdl, &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION), TGT_OST_HDL(0, OST_GET_INFO, ofd_get_info_hdl), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_GETATTR, ofd_getattr_hdl), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_GETATTR, ofd_getattr_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_SETATTR, ofd_setattr_hdl), -TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_CREATE, ofd_create_hdl), -TGT_OST_HDL(0 | HABEO_REFERO | MUTABOR, +TGT_OST_HDL(HAS_REPLY | IS_MUTABLE, OST_DESTROY, ofd_destroy_hdl), -TGT_OST_HDL(0 | HABEO_REFERO, OST_STATFS, ofd_statfs_hdl), -TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO, - OST_BRW_READ, tgt_brw_read, +TGT_OST_HDL(HAS_REPLY, OST_STATFS, ofd_statfs_hdl), +TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read, ofd_hp_brw), /* don't set CORPUS flag for brw_write because -ENOENT may be valid case */ -TGT_OST_HDL_HP(HABEO_CORPUS| MUTABOR, OST_BRW_WRITE, tgt_brw_write, +TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE, OST_BRW_WRITE, tgt_brw_write, ofd_hp_brw), -TGT_OST_HDL_HP(HABEO_CORPUS| HABEO_REFERO | MUTABOR, +TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_PUNCH, ofd_punch_hdl, ofd_hp_punch), -TGT_OST_HDL(HABEO_CORPUS| HABEO_REFERO, OST_SYNC, ofd_sync_hdl), -TGT_OST_HDL(0 | HABEO_REFERO, OST_QUOTACTL, ofd_quotactl), -TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_LADVISE, ofd_ladvise_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC, ofd_sync_hdl), +TGT_OST_HDL(HAS_REPLY, OST_QUOTACTL, ofd_quotactl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_LADVISE, ofd_ladvise_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE, ofd_fallocate_hdl), +TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek), }; static struct tgt_opc_slice ofd_common_slice[] = { @@ -2789,7 +2929,6 @@ static void ofd_key_exit(const struct lu_context *ctx, info->fti_xid = 0; info->fti_pre_version = 0; - info->fti_used = 0; memset(&info->fti_attr, 0, sizeof info->fti_attr); } @@ -2819,11 +2958,15 @@ struct lu_context_key ofd_thread_key = { static int ofd_init0(const struct lu_env *env, struct ofd_device *m, struct lu_device_type *ldt, struct lustre_cfg *cfg) { - const char *dev = lustre_cfg_string(cfg, 0); - struct ofd_thread_info *info = NULL; - struct obd_device *obd; - struct obd_statfs *osfs; - int rc; + const char *dev = lustre_cfg_string(cfg, 0); + struct ofd_thread_info *info = NULL; + struct obd_device *obd; + struct tg_grants_data *tgd = &m->ofd_lut.lut_tgd; + struct lu_fid fid; + struct nm_config_file *nodemap_config; + struct obd_device_target *obt; + u32 lmd_flags = 0; + int rc; ENTRY; @@ -2837,36 +2980,21 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, if (rc != 0) RETURN(rc); - obd->u.obt.obt_magic = OBT_MAGIC; - - m->ofd_fmd_max_num = OFD_FMD_MAX_NUM_DEFAULT; - m->ofd_fmd_max_age = OFD_FMD_MAX_AGE_DEFAULT; + obt = &obd->u.obt; + obt->obt_magic = OBT_MAGIC; spin_lock_init(&m->ofd_flags_lock); m->ofd_raid_degraded = 0; - m->ofd_syncjournal = 0; + m->ofd_sync_journal = 0; ofd_slc_set(m); - m->ofd_grant_compat_disable = 0; m->ofd_soft_sync_limit = OFD_SOFT_SYNC_LIMIT_DEFAULT; - m->ofd_brw_size = ONE_MB_BRW_SIZE; - - /* statfs data */ - spin_lock_init(&m->ofd_osfs_lock); - m->ofd_osfs_age = cfs_time_shift_64(-1000); - m->ofd_osfs_unstable = 0; - m->ofd_statfs_inflight = 0; - m->ofd_osfs_inflight = 0; - - /* grant data */ - spin_lock_init(&m->ofd_grant_lock); - m->ofd_tot_dirty = 0; - m->ofd_tot_granted = 0; - m->ofd_tot_pending = 0; + m->ofd_seq_count = 0; - init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq); INIT_LIST_HEAD(&m->ofd_inconsistency_list); spin_lock_init(&m->ofd_inconsistency_lock); + m->ofd_access_log_mask = -1; /* Log all accesses if enabled. */ + spin_lock_init(&m->ofd_batch_lock); init_rwsem(&m->ofd_lastid_rwsem); @@ -2875,12 +3003,6 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, /* set this lu_device to obd, because error handling need it */ obd->obd_lu_dev = &m->ofd_dt_dev.dd_lu_dev; - rc = ofd_procfs_init(m); - if (rc) { - CERROR("Can't init ofd lprocfs, rc %d\n", rc); - RETURN(rc); - } - /* No connection accepted until configurations will finish */ spin_lock(&obd->obd_dev_lock); obd->obd_no_conn = 1; @@ -2897,33 +3019,18 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, info = ofd_info_init(env, NULL); if (info == NULL) - GOTO(err_fini_proc, rc = -EFAULT); + RETURN(-EFAULT); - rc = ofd_stack_init(env, m, cfg); + rc = ofd_stack_init(env, m, cfg, &lmd_flags); if (rc) { - CERROR("Can't init device stack, rc %d\n", rc); - GOTO(err_fini_proc, rc); + CERROR("%s: can't init device stack, rc %d\n", + obd->obd_name, rc); + RETURN(rc); } +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0) ofd_procfs_add_brw_stats_symlink(m); - - /* populate cached statfs data */ - osfs = &ofd_info(env)->fti_u.osfs; - rc = ofd_statfs_internal(env, m, osfs, 0, NULL); - if (rc != 0) { - CERROR("%s: can't get statfs data, rc %d\n", obd->obd_name, rc); - GOTO(err_fini_stack, rc); - } - if (!IS_PO2(osfs->os_bsize)) { - CERROR("%s: blocksize (%d) is not a power of 2\n", - obd->obd_name, osfs->os_bsize); - GOTO(err_fini_stack, rc = -EPROTO); - } - m->ofd_blockbits = fls(osfs->os_bsize) - 1; - - m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT; - if (osfs->os_bsize * osfs->os_blocks < OFD_PRECREATE_SMALL_FS) - m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL; +#endif snprintf(info->fti_u.name, sizeof(info->fti_u.name), "%s-%s", "filter"/*LUSTRE_OST_NAME*/, obd->obd_uuid.uuid); @@ -2931,8 +3038,13 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, LDLM_NAMESPACE_SERVER, LDLM_NAMESPACE_GREEDY, LDLM_NS_TYPE_OST); - if (m->ofd_namespace == NULL) - GOTO(err_fini_stack, rc = -ENOMEM); + if (IS_ERR(m->ofd_namespace)) { + rc = PTR_ERR(m->ofd_namespace); + CERROR("%s: unable to create server namespace: rc = %d\n", + obd->obd_name, rc); + m->ofd_namespace = NULL; + GOTO(err_fini_stack, rc); + } /* set obd_namespace for compatibility with old code */ obd->obd_namespace = m->ofd_namespace; ldlm_register_intent(m->ofd_namespace, ofd_intent_policy); @@ -2942,28 +3054,70 @@ static int ofd_init0(const struct lu_env *env, struct ofd_device *m, ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL, "filter_ldlm_cb_client", &obd->obd_ldlm_client); - dt_conf_get(env, m->ofd_osd, &m->ofd_dt_conf); - rc = tgt_init(env, &m->ofd_lut, obd, m->ofd_osd, ofd_common_slice, OBD_FAIL_OST_ALL_REQUEST_NET, OBD_FAIL_OST_ALL_REPLY_NET); if (rc) GOTO(err_free_ns, rc); - rc = ofd_fs_setup(env, m, obd); + if (lmd_flags & LMD_FLG_SKIP_LFSCK) + m->ofd_skip_lfsck = 1; + if (lmd_flags & LMD_FLG_LOCAL_RECOV) + m->ofd_lut.lut_local_recovery = 1; + + rc = ofd_tunables_init(m); if (rc) GOTO(err_fini_lut, rc); - rc = ofd_start_inconsistency_verification_thread(m); + tgd->tgd_reserved_pcnt = 0; + + m->ofd_brw_size = m->ofd_lut.lut_dt_conf.ddp_brw_size; + m->ofd_precreate_batch = OFD_PRECREATE_BATCH_DEFAULT; + if (tgd->tgd_osfs.os_bsize * tgd->tgd_osfs.os_blocks < + OFD_PRECREATE_SMALL_FS) + m->ofd_precreate_batch = OFD_PRECREATE_BATCH_SMALL; + m->ofd_atime_diff = OFD_DEF_ATIME_DIFF; + + rc = ofd_fs_setup(env, m, obd); + if (rc) + GOTO(err_fini_proc, rc); + + fid.f_seq = FID_SEQ_LOCAL_NAME; + fid.f_oid = 1; + fid.f_ver = 0; + rc = local_oid_storage_init(env, m->ofd_osd, &fid, + &m->ofd_los); if (rc != 0) GOTO(err_fini_fs, rc); - tgt_adapt_sptlrpc_conf(&m->ofd_lut, 1); + nodemap_config = nm_config_file_register_tgt(env, m->ofd_osd, + m->ofd_los); + if (IS_ERR(nodemap_config)) { + rc = PTR_ERR(nodemap_config); + if (rc != -EROFS) + GOTO(err_fini_los, rc); + } else { + obt->obt_nodemap_config_file = nodemap_config; + } + + rc = ofd_start_inconsistency_verification_thread(m); + if (rc != 0) + GOTO(err_fini_nm, rc); + + tgt_adapt_sptlrpc_conf(&m->ofd_lut); RETURN(0); +err_fini_nm: + nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file); + obt->obt_nodemap_config_file = NULL; +err_fini_los: + local_oid_storage_fini(env, m->ofd_los); + m->ofd_los = NULL; err_fini_fs: ofd_fs_cleanup(env, m); +err_fini_proc: + ofd_procfs_fini(m); err_fini_lut: tgt_fini(env, &m->ofd_lut); err_free_ns: @@ -2971,8 +3125,6 @@ err_free_ns: obd->obd_namespace = m->ofd_namespace = NULL; err_fini_stack: ofd_stack_fini(env, m, &m->ofd_osd->dd_lu_dev); -err_fini_proc: - ofd_procfs_fini(m); return rc; } @@ -2994,6 +3146,7 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) stop.ls_status = LS_PAUSED; stop.ls_flags = 0; lfsck_stop(env, m->ofd_osd, &stop); + ofd_stack_pre_fini(env, m, &m->ofd_dt_dev.dd_lu_dev); target_recovery_fini(obd); if (m->ofd_namespace != NULL) ldlm_namespace_free_prior(m->ofd_namespace, NULL, @@ -3002,18 +3155,24 @@ static void ofd_fini(const struct lu_env *env, struct ofd_device *m) obd_exports_barrier(obd); obd_zombie_barrier(); + ofd_procfs_fini(m); tgt_fini(env, &m->ofd_lut); ofd_stop_inconsistency_verification_thread(m); lfsck_degister(env, m->ofd_osd); ofd_fs_cleanup(env, m); + nm_config_file_deregister_tgt(env, obd->u.obt.obt_nodemap_config_file); + obd->u.obt.obt_nodemap_config_file = NULL; if (m->ofd_namespace != NULL) { ldlm_namespace_free_post(m->ofd_namespace); d->ld_obd->obd_namespace = m->ofd_namespace = NULL; } + ofd_access_log_delete(m->ofd_access_log); + m->ofd_access_log = NULL; + ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev); - ofd_procfs_fini(m); + LASSERT(atomic_read(&d->ld_ref) == 0); server_put_mount(obd->obd_name, true); EXIT; @@ -3097,7 +3256,7 @@ static struct lu_device *ofd_device_alloc(const struct lu_env *env, /* type constructor/destructor: ofd_type_init(), ofd_type_fini() */ LU_TYPE_INIT_FINI(ofd, &ofd_thread_key); -static struct lu_device_type_operations ofd_device_type_ops = { +static const struct lu_device_type_operations ofd_device_type_ops = { .ldto_init = ofd_type_init, .ldto_fini = ofd_type_fini, @@ -3127,20 +3286,28 @@ static struct lu_device_type ofd_device_type = { */ static int __init ofd_init(void) { - int rc; + int rc; rc = lu_kmem_init(ofd_caches); if (rc) return rc; - rc = ofd_fmd_init(); - if (rc) { - lu_kmem_fini(ofd_caches); - return(rc); - } + rc = ofd_access_log_module_init(); + if (rc) + goto out_caches; - rc = class_register_type(&ofd_obd_ops, NULL, true, NULL, + rc = class_register_type(&ofd_obd_ops, NULL, true, LUSTRE_OST_NAME, &ofd_device_type); + if (rc) + goto out_ofd_access_log; + + return 0; + +out_ofd_access_log: + ofd_access_log_module_exit(); +out_caches: + lu_kmem_fini(ofd_caches); + return rc; } @@ -3152,9 +3319,9 @@ static int __init ofd_init(void) */ static void __exit ofd_exit(void) { - ofd_fmd_exit(); - lu_kmem_fini(ofd_caches); class_unregister_type(LUSTRE_OST_NAME); + ofd_access_log_module_exit(); + lu_kmem_fini(ofd_caches); } MODULE_AUTHOR("OpenSFS, Inc. ");