X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=8c2486cdb685cb1c3655e041a9151386e7413293;hp=a3dff28a69d7a4f2ee282e84be694cd896c30ca4;hb=037759bc49ce7f6adc9acb4a748a25c0f4c4ab64;hpb=f4842b027ee57969fa1b01b90d5f23aed18b5eb5 diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index a3dff28..8c2486c 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,9 +41,13 @@ * Pravin Shelar : Added fid in dirent */ -#define DEBUG_SUBSYSTEM S_MDS +#define DEBUG_SUBSYSTEM S_OSD #include +#include +#ifdef HAVE_UIDGID_HEADER +# include +#endif /* LUSTRE_VERSION_CODE */ #include @@ -54,6 +58,9 @@ /* XATTR_{REPLACE,CREATE} */ #include +#include +#include +#undef ENTRY /* * struct OBD_{ALLOC,FREE}*() * OBD_FAIL_CHECK @@ -72,7 +79,7 @@ #include #include -#include +#include int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, @@ -85,6 +92,9 @@ CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644, /* Slab to allocate dynlocks */ struct kmem_cache *dynlock_cachep; +/* Slab to allocate osd_it_ea */ +struct kmem_cache *osd_itea_cachep; + static struct lu_kmem_descr ldiskfs_caches[] = { { .ckd_cache = &dynlock_cachep, @@ -92,6 +102,11 @@ static struct lu_kmem_descr ldiskfs_caches[] = { .ckd_size = sizeof(struct dynlock_handle) }, { + .ckd_cache = &osd_itea_cachep, + .ckd_name = "osd_itea_cache", + .ckd_size = sizeof(struct osd_it_ea) + }, + { .ckd_cache = NULL } }; @@ -118,7 +133,6 @@ int osd_trans_declare_op2rb[] = { [OSD_OT_WRITE] = OSD_OT_WRITE, [OSD_OT_INSERT] = OSD_OT_DELETE, [OSD_OT_DELETE] = OSD_OT_INSERT, - [OSD_OT_UPDATE] = OSD_OT_MAX, [OSD_OT_QUOTA] = OSD_OT_MAX, }; @@ -287,119 +301,6 @@ osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, return inode; } -static struct inode *osd_iget_check(struct osd_thread_info *info, - struct osd_device *dev, - const struct lu_fid *fid, - struct osd_inode_id *id, - bool in_oi) -{ - struct inode *inode; - int rc = 0; - ENTRY; - - inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); - if (IS_ERR(inode)) { - rc = PTR_ERR(inode); - if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) { - CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n", - id->oii_ino, rc); - - GOTO(put, rc); - } - - goto check_oi; - } - - if (is_bad_inode(inode)) { - rc = -ENOENT; - if (!in_oi) { - CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino); - - GOTO(put, rc); - } - - goto check_oi; - } - - if (id->oii_gen != OSD_OII_NOGEN && - inode->i_generation != id->oii_gen) { - rc = -ESTALE; - if (!in_oi) { - CDEBUG(D_INODE, "unmatched inode: ino = %u, " - "oii_gen = %u, i_generation = %u\n", - id->oii_ino, id->oii_gen, inode->i_generation); - - GOTO(put, rc); - } - - goto check_oi; - } - - if (inode->i_nlink == 0) { - rc = -ENOENT; - if (!in_oi) { - CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); - - GOTO(put, rc); - } - - goto check_oi; - } - -check_oi: - if (rc != 0) { - LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); - - rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); - /* XXX: There are some possible cases: - * 1. rc = 0. - * Backup/restore caused the OI invalid. - * 2. rc = 0. - * Someone unlinked the object but NOT removed - * the OI mapping, such as mount target device - * as ldiskfs, and modify something directly. - * 3. rc = -ENOENT. - * Someone just removed the object between the - * former oi_lookup and the iget. It is normal. - * 4. Other failure cases. - * - * Generally, when the device is mounted, it will - * auto check whether the system is restored from - * file-level backup or not. We trust such detect - * to distinguish the 1st case from the 2nd case. */ - if (rc == 0) { - if (!IS_ERR(inode) && inode->i_generation != 0 && - inode->i_generation == id->oii_gen) - rc = -ENOENT; - else - rc = -EREMCHG; - } - } else { - if (id->oii_gen == OSD_OII_NOGEN) - osd_id_gen(id, inode->i_ino, inode->i_generation); - - /* Do not update file c/mtime in ldiskfs. - * NB: we don't have any lock to protect this because we don't - * have reference on osd_object now, but contention with - * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ - if (!(inode->i_flags & S_NOCMTIME)) - inode->i_flags |= S_NOCMTIME; - } - - GOTO(put, rc); - -put: - if (rc != 0) { - if (!IS_ERR(inode)) - iput(inode); - - inode = ERR_PTR(rc); - } - - return inode; -} - /** * \retval +v: new filter_fid, does not contain self-fid * \retval 0: filter_fid_old, contains self-fid @@ -472,15 +373,14 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) { fid = &lma->lma_self_fid; rc = osd_get_idif(info, inode, dentry, fid); - if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) { + if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) { /* For the given OST-object, if it has neither LMA nor * FID in XATTR_NAME_FID, then the given FID (which is * contained in the @obj, from client RPC for locating * the OST-object) is trusted. We use it to generate * the LMA. */ osd_lma_self_repair(info, osd, inode, rfid, - fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ? - LMAC_FID_ON_OST : 0); + LMAC_FID_ON_OST); RETURN(0); } } @@ -501,7 +401,7 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) lma->lma_incompat & ~LMA_INCOMPAT_SUPP, PFID(rfid), inode->i_ino); rc = -EOPNOTSUPP; - } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) { + } else { fid = &lma->lma_self_fid; } } @@ -521,7 +421,7 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) fid_to_ostid(fid, oi); ostid_to_fid(fid1, oi, idx); if (lu_fid_eq(fid1, rfid)) { - if (osd->od_lma_self_repair) + if (osd->od_index_in_idif) osd_lma_self_repair(info, osd, inode, rfid, LMAC_FID_ON_OST); @@ -530,8 +430,6 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } } - CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n", - osd_name(osd), PFID(rfid), PFID(fid)); rc = -EREMCHG; } @@ -553,12 +451,13 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, int result; int saved = 0; bool in_oi = false; + bool in_cache = false; bool triggered = false; ENTRY; LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid)); + LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid)); dev = osd_dev(ldev); scrub = &dev->od_scrub; @@ -582,6 +481,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) { id = &oic->oic_lid; + in_cache = true; goto iget; } @@ -596,8 +496,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, /* Search order: 3. OI files. */ result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result == -ENOENT) { - if (!fid_is_norm(fid) || - fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || + if (!(fid_is_norm(fid) || fid_is_igif(fid)) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); @@ -611,87 +510,107 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, in_oi = true; iget: - inode = osd_iget_check(info, dev, fid, id, in_oi); + inode = osd_iget(info, dev, id); if (IS_ERR(inode)) { result = PTR_ERR(inode); - if (result == -ENOENT || result == -ESTALE) { - if (!in_oi) - fid_zero(&oic->oic_fid); + if (result != -ENOENT && result != -ESTALE) + GOTO(out, result); - GOTO(out, result = -ENOENT); - } else if (result == -EREMCHG) { + if (in_cache) + fid_zero(&oic->oic_fid); -trigger: - if (!in_oi) - fid_zero(&oic->oic_fid); + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + if (result != 0) + GOTO(out, result = (result == -ENOENT ? 0 : result)); - if (unlikely(triggered)) - GOTO(out, result = saved); + /* The OI mapping is there, but the inode is NOT there. + * Two possible cases for that: + * + * 1) Backup/restore caused the OI invalid. + * 2) Someone unlinked the object but NOT removed + * the OI mapping, such as mount target device + * as ldiskfs, and modify something directly. + * + * Generally, when the device is mounted, it will + * auto check whether the system is restored from + * file-level backup or not. We trust such detect + * to distinguish the 1st case from the 2nd case. */ + if (!(scrub->os_file.sf_flags & SF_INCONSISTENT)) + GOTO(out, result = 0); - triggered = true; - if (thread_is_running(&scrub->os_thread)) { +trigger: + if (unlikely(triggered)) + GOTO(out, result = saved); + + triggered = true; + if (thread_is_running(&scrub->os_thread)) { + result = -EINPROGRESS; + } else if (!dev->od_noscrub) { + /* Since we do not know the right OI mapping, we have + * to trigger OI scrub to scan the whole device. */ + result = osd_scrub_start(dev, SS_AUTO_FULL | + SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT); + CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI " + "scrub by RPC for "DFID", rc = %d [1]\n", + osd_name(dev), PFID(fid), result); + if (result == 0 || result == -EALREADY) result = -EINPROGRESS; - } else if (!dev->od_noscrub) { - result = osd_scrub_start(dev); - LCONSOLE_WARN("%.16s: trigger OI scrub by RPC " - "for "DFID", rc = %d [1]\n", - osd_name(dev), PFID(fid),result); - if (result == 0 || result == -EALREADY) - result = -EINPROGRESS; - else - result = -EREMCHG; - } - - /* We still have chance to get the valid inode: for the - * object which is referenced by remote name entry, the - * object on the local MDT will be linked under the dir - * of "/REMOTE_PARENT_DIR" with its FID string as name. - * - * We do not know whether the object for the given FID - * is referenced by some remote name entry or not, and - * especially for DNE II, a multiple-linked object may - * have many name entries reside on many MDTs. - * - * To simplify the operation, OSD will not distinguish - * more, just lookup "/REMOTE_PARENT_DIR". Usually, it - * only happened for the RPC from other MDT during the - * OI scrub, or for the client side RPC with FID only, - * such as FID to path, or from old connected client. */ - saved = result; - result = osd_lookup_in_remote_parent(info, dev, - fid, id); - if (result == 0) { - in_oi = false; - goto iget; - } + else + result = -EREMCHG; + } - result = saved; + /* We still have chance to get the valid inode: for the + * object which is referenced by remote name entry, the + * object on the local MDT will be linked under the dir + * of "/REMOTE_PARENT_DIR" with its FID string as name. + * + * We do not know whether the object for the given FID + * is referenced by some remote name entry or not, and + * especially for DNE II, a multiple-linked object may + * have many name entries reside on many MDTs. + * + * To simplify the operation, OSD will not distinguish + * more, just lookup "/REMOTE_PARENT_DIR". Usually, it + * only happened for the RPC from other MDT during the + * OI scrub, or for the client side RPC with FID only, + * such as FID to path, or from old connected client. */ + saved = result; + result = osd_lookup_in_remote_parent(info, dev, fid, id); + if (result == 0) { + in_oi = false; + goto iget; } - GOTO(out, result); - } + GOTO(out, result = saved); + } - obj->oo_inode = inode; - LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); + obj->oo_inode = inode; + LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); result = osd_check_lma(env, obj); if (result != 0) { iput(inode); obj->oo_inode = NULL; - if (result == -EREMCHG) { - if (!in_oi) { - result = osd_oi_lookup(info, dev, fid, id, - OI_CHECK_FLD); - if (result != 0) { - fid_zero(&oic->oic_fid); - GOTO(out, result); - } - } + if (result != -EREMCHG) + GOTO(out, result); + + if (in_cache) + fid_zero(&oic->oic_fid); + + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + if (result == 0) goto trigger; - } - GOTO(out, result); + if (result != -ENOENT) + GOTO(out, result); + + if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) && + ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), + sf->sf_oi_bitmap)) + goto trigger; + + GOTO(out, result = 0); } obj->oo_compat_dot_created = 1; @@ -890,7 +809,7 @@ static int osd_param_is_not_sane(const struct osd_device *dev, { struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super); - return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers; + return oh->ot_credits > osd_transaction_size(dev); } /* @@ -912,7 +831,8 @@ static void osd_trans_commit_cb(struct super_block *sb, dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ - list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { + list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list, + dcb_linkage) { LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC, "commit callback entry: magic=%x name='%s'\n", dcb->dcb_magic, dcb->dcb_name); @@ -926,7 +846,7 @@ static void osd_trans_commit_cb(struct super_block *sb, lu_context_exit(&th->th_ctx); lu_context_fini(&th->th_ctx); - thandle_put(th); + OBD_FREE_PTR(oh); } static struct thandle *osd_trans_create(const struct lu_env *env, @@ -951,28 +871,73 @@ static struct thandle *osd_trans_create(const struct lu_env *env, th->th_result = 0; th->th_tags = LCT_TX_HANDLE; oh->ot_credits = 0; - atomic_set(&th->th_refc, 1); - th->th_alloc_size = sizeof(*oh); - oti->oti_dev = osd_dt_dev(d); - INIT_LIST_HEAD(&oh->ot_dcb_list); + INIT_LIST_HEAD(&oh->ot_commit_dcb_list); + INIT_LIST_HEAD(&oh->ot_stop_dcb_list); osd_th_alloced(oh); memset(oti->oti_declare_ops, 0, - sizeof(oti->oti_declare_ops)); - memset(oti->oti_declare_ops_rb, 0, - sizeof(oti->oti_declare_ops_rb)); + sizeof(oti->oti_declare_ops)); memset(oti->oti_declare_ops_cred, 0, - sizeof(oti->oti_declare_ops_cred)); - oti->oti_rollback = false; + sizeof(oti->oti_declare_ops_cred)); + memset(oti->oti_declare_ops_used, 0, + sizeof(oti->oti_declare_ops_used)); } RETURN(th); } +void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_thandle *oh; + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh != NULL); + + CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n", + oti->oti_declare_ops[OSD_OT_CREATE], + oti->oti_declare_ops_cred[OSD_OT_CREATE], + oti->oti_declare_ops_used[OSD_OT_CREATE], + oti->oti_declare_ops[OSD_OT_DESTROY], + oti->oti_declare_ops_cred[OSD_OT_DESTROY], + oti->oti_declare_ops_used[OSD_OT_DESTROY]); + CWARN(" attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n", + oti->oti_declare_ops[OSD_OT_ATTR_SET], + oti->oti_declare_ops_cred[OSD_OT_ATTR_SET], + oti->oti_declare_ops_used[OSD_OT_ATTR_SET], + oti->oti_declare_ops[OSD_OT_XATTR_SET], + oti->oti_declare_ops_cred[OSD_OT_XATTR_SET], + oti->oti_declare_ops_used[OSD_OT_XATTR_SET]); + CWARN(" write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n", + oti->oti_declare_ops[OSD_OT_WRITE], + oti->oti_declare_ops_cred[OSD_OT_WRITE], + oti->oti_declare_ops_used[OSD_OT_WRITE], + oti->oti_declare_ops[OSD_OT_PUNCH], + oti->oti_declare_ops_cred[OSD_OT_PUNCH], + oti->oti_declare_ops_used[OSD_OT_PUNCH], + oti->oti_declare_ops[OSD_OT_QUOTA], + oti->oti_declare_ops_cred[OSD_OT_QUOTA], + oti->oti_declare_ops_used[OSD_OT_QUOTA]); + CWARN(" insert: %u/%u/%u, delete: %u/%u/%u\n", + oti->oti_declare_ops[OSD_OT_INSERT], + oti->oti_declare_ops_cred[OSD_OT_INSERT], + oti->oti_declare_ops_used[OSD_OT_INSERT], + oti->oti_declare_ops[OSD_OT_DELETE], + oti->oti_declare_ops_cred[OSD_OT_DELETE], + oti->oti_declare_ops_used[OSD_OT_DELETE]); + CWARN(" ref_add: %u/%u/%u, ref_del: %u/%u/%u\n", + oti->oti_declare_ops[OSD_OT_REF_ADD], + oti->oti_declare_ops_cred[OSD_OT_REF_ADD], + oti->oti_declare_ops_used[OSD_OT_REF_ADD], + oti->oti_declare_ops[OSD_OT_REF_DEL], + oti->oti_declare_ops_cred[OSD_OT_REF_DEL], + oti->oti_declare_ops_used[OSD_OT_REF_DEL]); +} + /* * Concurrency: shouldn't matter. */ -int osd_trans_start(const struct lu_env *env, struct dt_device *d, - struct thandle *th) +static int osd_trans_start(const struct lu_env *env, struct dt_device *d, + struct thandle *th) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_device *dev = osd_dt_dev(d); @@ -1000,36 +965,12 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, oh->ot_credits, osd_journal(dev)->j_max_transaction_buffers); - CWARN(" create: %u/%u, destroy: %u/%u\n", - oti->oti_declare_ops[OSD_OT_CREATE], - oti->oti_declare_ops_cred[OSD_OT_CREATE], - oti->oti_declare_ops[OSD_OT_DESTROY], - oti->oti_declare_ops_cred[OSD_OT_DESTROY]); - CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n", - oti->oti_declare_ops[OSD_OT_ATTR_SET], - oti->oti_declare_ops_cred[OSD_OT_ATTR_SET], - oti->oti_declare_ops[OSD_OT_XATTR_SET], - oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]); - CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n", - oti->oti_declare_ops[OSD_OT_WRITE], - oti->oti_declare_ops_cred[OSD_OT_WRITE], - oti->oti_declare_ops[OSD_OT_PUNCH], - oti->oti_declare_ops_cred[OSD_OT_PUNCH], - oti->oti_declare_ops[OSD_OT_QUOTA], - oti->oti_declare_ops_cred[OSD_OT_QUOTA]); - CWARN(" insert: %u/%u, delete: %u/%u\n", - oti->oti_declare_ops[OSD_OT_INSERT], - oti->oti_declare_ops_cred[OSD_OT_INSERT], - oti->oti_declare_ops[OSD_OT_DELETE], - oti->oti_declare_ops_cred[OSD_OT_DELETE]); - CWARN(" ref_add: %u/%u, ref_del: %u/%u\n", - oti->oti_declare_ops[OSD_OT_REF_ADD], - oti->oti_declare_ops_cred[OSD_OT_REF_ADD], - oti->oti_declare_ops[OSD_OT_REF_DEL], - oti->oti_declare_ops_cred[OSD_OT_REF_DEL]); + + osd_trans_dump_creds(env, th); if (last_credits != oh->ot_credits && - time_after(jiffies, last_printed + 60 * HZ)) { + time_after(jiffies, last_printed + + msecs_to_jiffies(60 * MSEC_PER_SEC))) { libcfs_debug_dumpstack(NULL); last_credits = oh->ot_credits; last_printed = jiffies; @@ -1040,7 +981,7 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, * * This should be removed when we can calculate the * credits precisely. */ - oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers; + oh->ot_credits = osd_transaction_size(dev); } /* @@ -1068,15 +1009,15 @@ out: } static int osd_seq_exists(const struct lu_env *env, - struct osd_device *osd, obd_seq seq) + struct osd_device *osd, u64 seq) { struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; struct seq_server_site *ss = osd_seq_site(osd); int rc; ENTRY; - if (ss == NULL) - RETURN(1); + LASSERT(ss != NULL); + LASSERT(ss->ss_server_fld != NULL); rc = osd_fld_lookup(env, osd, seq, range); if (rc != 0) { @@ -1089,6 +1030,22 @@ static int osd_seq_exists(const struct lu_env *env, RETURN(ss->ss_node_id == range->lsr_index); } +static void osd_trans_stop_cb(struct osd_thandle *oth, int result) +{ + struct dt_txn_commit_cb *dcb; + struct dt_txn_commit_cb *tmp; + + /* call per-transaction stop callbacks if any */ + list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list, + dcb_linkage) { + LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC, + "commit callback entry: magic=%x name='%s'\n", + dcb->dcb_magic, dcb->dcb_name); + list_del_init(&dcb->dcb_linkage); + dcb->dcb_func(NULL, &oth->ot_super, dcb, result); + } +} + /* * Concurrency: shouldn't matter. */ @@ -1099,7 +1056,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, struct osd_thandle *oh; struct osd_thread_info *oti = osd_oti_get(env); struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct qsd_instance *qsd = oti->oti_dev->od_quota_slave; + struct osd_device *osd = osd_dt_dev(th->th_dev); + struct qsd_instance *qsd = osd->od_quota_slave; struct lquota_trans *qtrans; ENTRY; @@ -1108,7 +1066,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; - if (oh->ot_handle != NULL) { + if (oh->ot_handle != NULL) { handle_t *hdl = oh->ot_handle; /* @@ -1121,21 +1079,25 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, LASSERT(oti->oti_txns == 1); oti->oti_txns--; + rc = dt_txn_hook_stop(env, th); if (rc != 0) - CERROR("Failure in transaction hook: %d\n", rc); + CERROR("%s: failed in transaction hook: rc = %d\n", + osd_name(osd), rc); + osd_trans_stop_cb(oh, rc); /* hook functions might modify th_sync */ hdl->h_sync = th->th_sync; - oh->ot_handle = NULL; - OSD_CHECK_SLOW_TH(oh, oti->oti_dev, - rc = ldiskfs_journal_stop(hdl)); - if (rc != 0) - CERROR("Failure to stop transaction: %d\n", rc); - } else { - thandle_put(&oh->ot_super); - } + oh->ot_handle = NULL; + OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl)); + if (rc != 0) + CERROR("%s: failed to stop transaction: rc = %d\n", + osd_name(osd), rc); + } else { + osd_trans_stop_cb(oh, th->th_result); + OBD_FREE_PTR(oh); + } /* inform the quota slave device that the transaction is stopping */ qsd_op_end(env, qsd, qtrans); @@ -1151,7 +1113,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, */ wait_event(iobuf->dr_wait, atomic_read(&iobuf->dr_numreqs) == 0); - osd_fini_iobuf(oti->oti_dev, iobuf); + osd_fini_iobuf(osd, iobuf); if (!rc) rc = iobuf->dr_error; @@ -1165,7 +1127,10 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC); LASSERT(&dcb->dcb_func != NULL); - list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); + if (dcb->dcb_flags & DCB_TRANS_STOP) + list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list); + else + list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list); return 0; } @@ -1269,10 +1234,14 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, result = sb->s_op->statfs(sb->s_root, ksfs); if (likely(result == 0)) { /* N.B. statfs can't really fail */ statfs_pack(sfs, ksfs); - if (sb->s_flags & MS_RDONLY) + if (unlikely(sb->s_flags & MS_RDONLY)) sfs->os_state = OS_STATE_READONLY; + if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, + LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + sfs->os_maxbytes = sb->s_maxbytes; + else + sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; } - spin_unlock(&osd->od_osfs_lock); if (unlikely(env == NULL)) @@ -1319,7 +1288,10 @@ static void osd_conf_get(const struct lu_env *env, param->ddp_max_nlink = LDISKFS_LINK_MAX; param->ddp_block_shift = sb->s_blocksize_bits; param->ddp_mount_type = LDD_MT_LDISKFS; - param->ddp_maxbytes = sb->s_maxbytes; + if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + param->ddp_maxbytes = sb->s_maxbytes; + else + param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; /* Overhead estimate should be fairly accurate, so we really take a tiny * error margin which also avoids fragmenting the filesystem too much */ param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */ @@ -1354,8 +1326,16 @@ static void osd_conf_get(const struct lu_env *env, */ static int osd_sync(const struct lu_env *env, struct dt_device *d) { - CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); - return ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + int rc; + + CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); + + rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + + CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n", + LUSTRE_OSD_LDISKFS_NAME, rc); + + return rc; } /** @@ -1413,23 +1393,6 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) RETURN(rc); } -/* - * Concurrency: serialization provided by callers. - */ -static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d, - int mode, unsigned long timeout, __u32 alg, - struct lustre_capa_key *keys) -{ - struct osd_device *dev = osd_dt_dev(d); - ENTRY; - - dev->od_fl_capa = mode; - dev->od_capa_timeout = timeout; - dev->od_capa_alg = alg; - dev->od_capa_keys = keys; - RETURN(0); -} - /** * Note: we do not count into QUOTA here. * If we mount with --data_journal we may need more. @@ -1499,7 +1462,6 @@ static const struct dt_device_operations osd_dt_ops = { .dt_sync = osd_sync, .dt_ro = osd_ro, .dt_commit_async = osd_commit_async, - .dt_init_capa_ctxt = osd_init_capa_ctxt, }; static void osd_object_read_lock(const struct lu_env *env, @@ -1571,108 +1533,6 @@ static int osd_object_write_locked(const struct lu_env *env, return obj->oo_owner == env; } -static int capa_is_sane(const struct lu_env *env, - struct osd_device *dev, - struct lustre_capa *capa, - struct lustre_capa_key *keys) -{ - struct osd_thread_info *oti = osd_oti_get(env); - struct lustre_capa *tcapa = &oti->oti_capa; - struct obd_capa *oc; - int i, rc = 0; - ENTRY; - - oc = capa_lookup(dev->od_capa_hash, capa, 0); - if (oc) { - if (capa_is_expired(oc)) { - DEBUG_CAPA(D_ERROR, capa, "expired"); - rc = -ESTALE; - } - capa_put(oc); - RETURN(rc); - } - - if (capa_is_expired_sec(capa)) { - DEBUG_CAPA(D_ERROR, capa, "expired"); - RETURN(-ESTALE); - } - - spin_lock(&capa_lock); - for (i = 0; i < 2; i++) { - if (keys[i].lk_keyid == capa->lc_keyid) { - oti->oti_capa_key = keys[i]; - break; - } - } - spin_unlock(&capa_lock); - - if (i == 2) { - DEBUG_CAPA(D_ERROR, capa, "no matched capa key"); - RETURN(-ESTALE); - } - - rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key); - if (rc) - RETURN(rc); - - if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) { - DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch"); - RETURN(-EACCES); - } - - oc = capa_add(dev->od_capa_hash, capa); - capa_put(oc); - - RETURN(0); -} - -int osd_object_auth(const struct lu_env *env, struct dt_object *dt, - struct lustre_capa *capa, __u64 opc) -{ - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct lu_capainfo *lci; - int rc; - - if (!osd->od_fl_capa) - return 0; - - if (capa == BYPASS_CAPA) - return 0; - - lci = lu_capainfo_get(env); - if (unlikely(lci == NULL)) - return 0; - - if (lci->lci_auth == LC_ID_NONE) - return 0; - - if (capa == NULL) { - CERROR("%s: no capability provided for FID "DFID": rc = %d\n", - osd_name(osd), PFID(fid), -EACCES); - return -EACCES; - } - - if (!lu_fid_eq(fid, &capa->lc_fid)) { - DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with", - PFID(fid)); - return -EACCES; - } - - if (!capa_opc_supported(capa, opc)) { - DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc); - return -EACCES; - } - - rc = capa_is_sane(env, osd, capa, osd->od_capa_keys); - if (rc != 0) { - DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc); - return -EACCES; - } - - return 0; -} - static struct timespec *osd_inode_time(const struct lu_env *env, struct inode *inode, __u64 seconds) { @@ -1710,17 +1570,16 @@ static void osd_inode_getattr(const struct lu_env *env, } static int osd_attr_get(const struct lu_env *env, - struct dt_object *dt, - struct lu_attr *attr, - struct lustre_capa *capa) + struct dt_object *dt, + struct lu_attr *attr) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LINVRNT(osd_invariant(obj)); spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); @@ -1741,7 +1600,7 @@ static int osd_declare_attr_set(const struct lu_env *env, qid_t gid; long long bspace; int rc = 0; - bool allocated; + bool enforce; ENTRY; LASSERT(dt != NULL); @@ -1769,18 +1628,19 @@ static int osd_declare_attr_set(const struct lu_env *env, * We still need to call the osd_declare_qid() to calculate the journal * credits for updating quota accounting files and to trigger quota * space adjustment once the operation is completed.*/ - if ((attr->la_valid & LA_UID) != 0 && - attr->la_uid != (uid = i_uid_read(obj->oo_inode))) { + if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { + /* USERQUOTA */ + uid = i_uid_read(obj->oo_inode); qi->lqi_type = USRQUOTA; - + enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid); /* inode accounting */ qi->lqi_is_blk = false; - /* one more inode for the new owner ... */ + /* one more inode for the new uid ... */ qi->lqi_id.qid_uid = attr->la_uid; qi->lqi_space = 1; - allocated = (attr->la_uid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + /* Reserve credits for the new uid */ + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1789,7 +1649,7 @@ static int osd_declare_attr_set(const struct lu_env *env, /* and one less inode for the current uid */ qi->lqi_id.qid_uid = uid; qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1798,38 +1658,40 @@ static int osd_declare_attr_set(const struct lu_env *env, /* block accounting */ qi->lqi_is_blk = true; - /* more blocks for the new owner ... */ + /* more blocks for the new uid ... */ qi->lqi_id.qid_uid = attr->la_uid; qi->lqi_space = bspace; - allocated = (attr->la_uid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + /* + * Credits for the new uid has been reserved, re-use "obj" + * to save credit reservation. + */ + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - /* and finally less blocks for the current owner */ + /* and finally less blocks for the current uid */ qi->lqi_id.qid_uid = uid; qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - } - if (attr->la_valid & LA_GID && - attr->la_gid != (gid = i_gid_read(obj->oo_inode))) { + /* GROUP QUOTA */ + gid = i_gid_read(obj->oo_inode); qi->lqi_type = GRPQUOTA; + enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid); /* inode accounting */ qi->lqi_is_blk = false; - /* one more inode for the new group owner ... */ + /* one more inode for the new gid ... */ qi->lqi_id.qid_gid = attr->la_gid; qi->lqi_space = 1; - allocated = (attr->la_gid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1838,7 +1700,7 @@ static int osd_declare_attr_set(const struct lu_env *env, /* and one less inode for the current gid */ qi->lqi_id.qid_gid = gid; qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1847,20 +1709,19 @@ static int osd_declare_attr_set(const struct lu_env *env, /* block accounting */ qi->lqi_is_blk = true; - /* more blocks for the new owner ... */ + /* more blocks for the new gid ... */ qi->lqi_id.qid_gid = attr->la_gid; qi->lqi_space = bspace; - allocated = (attr->la_gid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - /* and finally less blocks for the current owner */ + /* and finally less blocks for the current gid */ qi->lqi_id.qid_gid = gid; qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1871,11 +1732,16 @@ static int osd_declare_attr_set(const struct lu_env *env, } static int osd_inode_setattr(const struct lu_env *env, - struct inode *inode, const struct lu_attr *attr) + struct inode *inode, const struct lu_attr *attr) { - __u64 bits; + __u64 bits = attr->la_valid; - bits = attr->la_valid; + /* Only allow set size for regular file */ + if (!S_ISREG(inode->i_mode)) + bits &= ~(LA_SIZE | LA_BLOCKS); + + if (bits == 0) + return 0; if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); @@ -1921,6 +1787,7 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) struct iattr iattr; int rc; + ll_vfs_dq_init(inode); iattr.ia_valid = 0; if (attr->la_valid & LA_UID) iattr.ia_valid |= ATTR_UID; @@ -1932,8 +1799,8 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) rc = ll_vfs_dq_transfer(inode, &iattr); if (rc) { CERROR("%s: quota transfer failed: rc = %d. Is quota " - "enforcement enabled on the ldiskfs filesystem?", - inode->i_sb->s_id, rc); + "enforcement enabled on the ldiskfs " + "filesystem?\n", inode->i_sb->s_id, rc); return rc; } } @@ -1941,21 +1808,20 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) } static int osd_attr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_attr *attr, - struct thandle *handle, - struct lustre_capa *capa) + struct dt_object *dt, + const struct lu_attr *attr, + struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode; + int rc; - LASSERT(handle != NULL); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_invariant(obj)); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; + LASSERT(handle != NULL); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_invariant(obj)); osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); @@ -1985,7 +1851,6 @@ static int osd_attr_set(const struct lu_env *env, } inode = obj->oo_inode; - ll_vfs_dq_init(inode); rc = osd_quota_transfer(inode, attr); if (rc) @@ -1997,12 +1862,15 @@ static int osd_attr_set(const struct lu_env *env, if (!rc) ll_dirty_inode(inode, I_DIRTY_DATASYNC); + + osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); + return rc; } -struct dentry *osd_child_dentry_get(const struct lu_env *env, - struct osd_object *obj, - const char *name, const int namelen) +static struct dentry *osd_child_dentry_get(const struct lu_env *env, + struct osd_object *obj, + const char *name, const int namelen) { return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); } @@ -2030,22 +1898,22 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, oth = container_of(th, struct osd_thandle, ot_super); LASSERT(oth->ot_handle->h_transaction != NULL); - if (hint && hint->dah_parent) - parent = hint->dah_parent; + if (hint != NULL && hint->dah_parent != NULL && + !dt_object_remote(hint->dah_parent)) + parent = hint->dah_parent; inode = ldiskfs_create_inode(oth->ot_handle, parent ? osd_dt_obj(parent)->oo_inode : osd_sb(osd)->s_root->d_inode, mode); if (!IS_ERR(inode)) { - /* Do not update file c/mtime in ldiskfs. - * NB: don't need any lock because no contention at this - * early stage */ - inode->i_flags |= S_NOCMTIME; + /* Do not update file c/mtime in ldiskfs. */ + inode->i_flags |= S_NOCMTIME; /* For new created object, it must be consistent, * and it is unnecessary to scrub against it. */ ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); + obj->oo_inode = inode; result = 0; } else { @@ -2245,7 +2113,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, * should not happen since quota enforcement is no longer * enabled on ldiskfs (lquota takes care of it). */ - LASSERTF(result == 0, "%d", result); + LASSERTF(result == 0, "%d\n", result); ll_dirty_inode(inode, I_DIRTY_DATASYNC); } @@ -2266,23 +2134,32 @@ static int __osd_object_create(struct osd_thread_info *info, int result; __u32 umask; + osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE); + /* we drop umask so that permissions we pass are not affected */ umask = current->fs->umask; current->fs->umask = 0; result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof, th); - if (result == 0) { + if (likely(obj->oo_inode != NULL)) { + LASSERT(obj->oo_inode->i_state & I_NEW); + + /* Unlock the inode before attr initialization to avoid + * unnecessary dqget operations. LU-6378 */ + unlock_new_inode(obj->oo_inode); + } + + if (likely(result == 0)) { osd_attr_init(info, obj, attr, dof); osd_object_init0(obj); - /* bz 24037 */ - if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW)) - unlock_new_inode(obj->oo_inode); - } + } /* restore previous umask value */ current->fs->umask = umask; + osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE); + return result; } @@ -2298,18 +2175,23 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id = &info->oti_id; struct osd_device *osd = osd_obj2dev(obj); struct osd_thandle *oh; + int rc; LASSERT(obj->oo_inode != NULL); oh = container_of0(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle); + osd_trans_exec_op(env, th, OSD_OT_INSERT); osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation); - return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD); + rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD); + osd_trans_exec_check(env, th, OSD_OT_INSERT); + + return rc; } int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, - obd_seq seq, struct lu_seq_range *range) + u64 seq, struct lu_seq_range *range) { struct seq_server_site *ss = osd_seq_site(osd); @@ -2353,29 +2235,23 @@ static int osd_declare_object_create(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); + /* EA object consumes more credits than regular object: osd_mk_index + * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and + * leaf_node, could involves the block, block bitmap, groups, GDT + * change for each block, so add 4 * 2 credits in that case. */ osd_trans_declare_op(env, oh, OSD_OT_CREATE, - osd_dto_credits_noquota[DTO_OBJECT_CREATE]); + osd_dto_credits_noquota[DTO_OBJECT_CREATE] + + (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0); /* Reuse idle OI block may cause additional one OI block * to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); - /* If this is directory, then we expect . and .. to be inserted as - * well. The one directory block always needs to be created for the - * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap, - * block), there is no danger of needing a tree for the first block. - */ - if (attr && S_ISDIR(attr->la_mode)) { - osd_trans_declare_op(env, oh, OSD_OT_INSERT, - osd_dto_credits_noquota[DTO_WRITE_BASE]); - osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0); - } - if (!attr) RETURN(0); rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh, - false, false, NULL, false); + osd_dt_obj(dt), false, NULL, false); if (rc != 0) RETURN(rc); @@ -2383,40 +2259,40 @@ static int osd_declare_object_create(const struct lu_env *env, } static int osd_object_create(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, struct thandle *th) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - int result; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + int result; + ENTRY; - ENTRY; + if (dt_object_exists(dt)) + return -EEXIST; - LINVRNT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) /* Quota files can't be created from the kernel any more, * 'tune2fs -O quota' will take care of creating them */ RETURN(-EPERM); - osd_trans_exec_op(env, th, OSD_OT_CREATE); - osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); - - result = __osd_object_create(info, obj, attr, hint, dof, th); - if (result == 0) - result = __osd_oi_insert(env, obj, fid, th); - + result = __osd_object_create(info, obj, attr, hint, dof, th); + if (result == 0) { + result = __osd_oi_insert(env, obj, fid, th); + if (obj->oo_dt.do_body_ops == &osd_body_ops_new) + obj->oo_dt.do_body_ops = &osd_body_ops; + } LASSERT(ergo(result == 0, - dt_object_exists(dt) && !dt_object_remote(dt))); + dt_object_exists(dt) && !dt_object_remote(dt))); - LASSERT(osd_invariant(obj)); - RETURN(result); + LASSERT(osd_invariant(obj)); + RETURN(result); } /** @@ -2446,12 +2322,12 @@ static int osd_declare_object_destroy(const struct lu_env *env, osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); /* one less inode */ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - -1, oh, false, true, NULL, false); + -1, oh, obj, false, NULL, false); if (rc) RETURN(rc); /* data to be truncated */ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - 0, oh, true, true, NULL, false); + 0, oh, obj, true, NULL, false); RETURN(rc); } @@ -2496,13 +2372,15 @@ static int osd_object_destroy(const struct lu_env *env, result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle, OI_CHECK_FLD); - /* XXX: add to ext3 orphan list */ - /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */ + osd_trans_exec_check(env, th, OSD_OT_DESTROY); + /* XXX: add to ext3 orphan list */ + /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */ - /* not needed in the cache anymore */ - set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags); + /* not needed in the cache anymore */ + set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags); + obj->oo_destroyed = 1; - RETURN(0); + RETURN(0); } /** @@ -2525,6 +2403,9 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) RETURN(0); + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET)) + rc = -ENOMEM; + lustre_lma_init(lma, fid, compat, incompat); lustre_lma_swab(lma); @@ -2565,10 +2446,10 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs. * its inmemory API. */ -void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) +static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, + const struct lu_fid *fid) { - if (!fid_is_namespace_visible((const struct lu_fid *)fid) || + if (!fid_is_namespace_visible(fid) || OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) { param->edp_magic = 0; return; @@ -2605,9 +2486,9 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, struct inode *dir, - struct inode *parent_dir, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, + struct inode *parent_dir, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, struct osd_thandle *oth) { struct ldiskfs_dentry_param *dot_ldp; @@ -2618,8 +2499,9 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; dot_ldp->edp_magic = 0; + return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, - dir, dot_ldp, dot_dot_ldp); + dir, dot_ldp, dot_dot_ldp); } /** @@ -2629,6 +2511,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, struct osd_device *osd, struct osd_object *pobj, const struct lu_fid *fid, + __u32 type, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); @@ -2641,15 +2524,16 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle->h_transaction != NULL); - /* FIXME: Insert index api needs to know the mode of - * the remote object. Just use S_IFDIR for now */ - local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR); + local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type); if (IS_ERR(local)) { CERROR("%s: create local error %d\n", osd_name(osd), (int)PTR_ERR(local)); RETURN(local); } + ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB); + unlock_new_inode(local); + /* Set special LMA flag for local agent inode */ rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT); if (rc != 0) { @@ -2658,9 +2542,12 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, RETURN(ERR_PTR(rc)); } + if (!S_ISDIR(type)) + RETURN(local); + rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode, - (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), - (const struct dt_rec *)fid, oh); + lu_object_fid(&pobj->oo_dt.do_lu), + fid, oh); if (rc != 0) { CERROR("%s: "DFID" add dot dotdot error: rc = %d\n", osd_name(osd), PFID(fid), rc); @@ -2709,37 +2596,51 @@ static int osd_delete_local_agent_inode(const struct lu_env *env, * \retval -ve, on error */ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - int result; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + int result; - ENTRY; + ENTRY; - LASSERT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + if (dt_object_exists(dt)) + RETURN(-EEXIST); + + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) /* Quota files can't be created from the kernel any more, * 'tune2fs -O quota' will take care of creating them */ RETURN(-EPERM); - osd_trans_exec_op(env, th, OSD_OT_CREATE); - osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); + result = __osd_object_create(info, obj, attr, hint, dof, th); + if (result == 0) { + if (fid_is_idif(fid) && + !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) { + struct lu_fid *tfid = &info->oti_fid; + struct ost_id *oi = &info->oti_ostid; - result = __osd_object_create(info, obj, attr, hint, dof, th); - if (result == 0) - result = osd_ea_fid_set(info, obj->oo_inode, fid, + fid_to_ostid(fid, oi); + ostid_to_fid(tfid, oi, 0); + result = osd_ea_fid_set(info, obj->oo_inode, tfid, + LMAC_FID_ON_OST, 0); + } else { + result = osd_ea_fid_set(info, obj->oo_inode, fid, fid_is_on_ost(info, osd_obj2dev(obj), fid, OI_CHECK_FLD) ? LMAC_FID_ON_OST : 0, 0); + } + if (obj->oo_dt.do_body_ops == &osd_body_ops_new) + obj->oo_dt.do_body_ops = &osd_body_ops; + } if (result == 0) result = __osd_oi_insert(env, obj, fid, th); @@ -2772,20 +2673,23 @@ static int osd_declare_object_ref_add(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_object_ref_add(const struct lu_env *env, - struct dt_object *dt, struct thandle *th) + struct dt_object *dt, struct thandle *th) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; struct osd_thandle *oh; int rc = 0; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + if (!dt_object_exists(dt)) + return -ENOENT; - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_ADD); @@ -2803,27 +2707,35 @@ static int osd_object_ref_add(const struct lu_env *env, * in case they are being linked into the PENDING directory */ spin_lock(&obj->oo_guard); - ldiskfs_inc_count(oh->ot_handle, inode); - LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + if (unlikely(inode->i_nlink == 0)) + /* inc_nlink from 0 may cause WARN_ON */ + set_nlink(inode, 1); + else { + ldiskfs_inc_count(oh->ot_handle, inode); + if (!S_ISDIR(inode->i_mode)) + LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + } spin_unlock(&obj->oo_guard); ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); + osd_trans_exec_check(env, th, OSD_OT_REF_ADD); + return rc; } static int osd_declare_object_ref_del(const struct lu_env *env, - struct dt_object *dt, - struct thandle *handle) + struct dt_object *dt, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_REF_DEL, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); @@ -2835,20 +2747,23 @@ static int osd_declare_object_ref_del(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, - struct thandle *th) + struct thandle *th) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); struct osd_thandle *oh; + if (!dt_object_exists(dt)) + return -ENOENT; + LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_DEL); @@ -2874,6 +2789,8 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); + osd_trans_exec_check(env, th, OSD_OT_REF_DEL); + return 0; } @@ -2895,8 +2812,7 @@ static int osd_object_version_get(const struct lu_env *env, * Concurrency: @dt is read locked. */ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, const char *name, - struct lustre_capa *capa) + struct lu_buf *buf, const char *name) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -2907,16 +2823,23 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, if (strcmp(name, XATTR_NAME_VERSION) == 0) { /* for version we are just using xattr API but change inode * field instead */ - LASSERT(buf->lb_len == sizeof(dt_obj_version_t)); - osd_object_version_get(env, dt, buf->lb_buf); - return sizeof(dt_obj_version_t); + if (buf->lb_len == 0) + return sizeof(dt_obj_version_t); + + if (buf->lb_len < sizeof(dt_obj_version_t)) + return -ERANGE; + + osd_object_version_get(env, dt, buf->lb_buf); + + return sizeof(dt_obj_version_t); } - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->getxattr != NULL); return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -2929,28 +2852,37 @@ static int osd_declare_xattr_set(const struct lu_env *env, { struct osd_thandle *oh; int credits; + struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - /* optimistic optimization: LMA is set first and usually fit inode */ if (strcmp(name, XATTR_NAME_LMA) == 0) { - if (dt_object_exists(dt)) + /* For non-upgrading case, the LMA is set first and + * usually fit inode. But for upgrade case, the LMA + * may be in another separated EA block. */ + if (!dt_object_exists(dt)) credits = 0; - else + else if (fl == LU_XATTR_REPLACE) credits = 1; + else + goto upgrade; } else if (strcmp(name, XATTR_NAME_VERSION) == 0) { credits = 1; } else { - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct super_block *sb = osd_sb(osd); +upgrade: credits = osd_dto_credits_noquota[DTO_XATTR_SET]; if (buf && buf->lb_len > sb->s_blocksize) { credits *= (buf->lb_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits; } + /* + * xattr set may involve inode quota change, reserve credits for + * dquot_initialize() + */ + credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); } osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); @@ -2980,13 +2912,14 @@ static void osd_object_version_set(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_buf *buf, const char *name, int fl, - struct thandle *handle, struct lustre_capa *capa) + const struct lu_buf *buf, const char *name, int fl, + struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; struct osd_thread_info *info = osd_oti_get(env); int fs_flags = 0; + int rc; ENTRY; LASSERT(handle != NULL); @@ -3000,10 +2933,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, return sizeof(dt_obj_version_t); } - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; - - CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n", + CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n", PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); @@ -3029,46 +2959,60 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } - return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) && + strcmp(name, XATTR_NAME_LINK) == 0) + return -ENOSPC; + + rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, fs_flags); + osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); + + return rc; } /* * Concurrency: @dt is read locked. */ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, struct lustre_capa *capa) + const struct lu_buf *buf) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->listxattr != NULL); - dentry->d_inode = inode; + dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); + return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); } static int osd_declare_xattr_del(const struct lu_env *env, - struct dt_object *dt, const char *name, - struct thandle *handle) + struct dt_object *dt, const char *name, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; + struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, osd_dto_credits_noquota[DTO_XATTR_SET]); + /* + * xattr del may involve inode quota change, reserve credits for + * dquot_initialize() + */ + oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); return 0; } @@ -3077,115 +3021,30 @@ static int osd_declare_xattr_del(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, - const char *name, struct thandle *handle, - struct lustre_capa *capa) + const char *name, struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + int rc; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL); - LASSERT(handle != NULL); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->removexattr != NULL); + LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); ll_vfs_dq_init(inode); - dentry->d_inode = inode; + dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - rc = inode->i_op->removexattr(dentry, name); - return rc; -} - -static struct obd_capa *osd_capa_get(const struct lu_env *env, - struct dt_object *dt, - struct lustre_capa *old, __u64 opc) -{ - struct osd_thread_info *info = osd_oti_get(env); - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *osd = osd_obj2dev(obj); - struct lustre_capa_key *key = &info->oti_capa_key; - struct lustre_capa *capa = &info->oti_capa; - struct obd_capa *oc; - struct lu_capainfo *lci; - int rc; - ENTRY; - - if (!osd->od_fl_capa) - RETURN(ERR_PTR(-ENOENT)); - - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); - - /* renewal sanity check */ - if (old && osd_object_auth(env, dt, old, opc)) - RETURN(ERR_PTR(-EACCES)); - - lci = lu_capainfo_get(env); - if (unlikely(lci == NULL)) - RETURN(ERR_PTR(-ENOENT)); - - switch (lci->lci_auth) { - case LC_ID_NONE: - RETURN(NULL); - case LC_ID_PLAIN: - capa->lc_uid = i_uid_read(obj->oo_inode); - capa->lc_gid = i_gid_read(obj->oo_inode); - capa->lc_flags = LC_ID_PLAIN; - break; - case LC_ID_CONVERT: { - __u32 d[4], s[4]; - - s[0] = i_uid_read(obj->oo_inode); - cfs_get_random_bytes(&(s[1]), sizeof(__u32)); - s[2] = i_uid_read(obj->oo_inode); - cfs_get_random_bytes(&(s[3]), sizeof(__u32)); - rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN); - if (unlikely(rc)) - RETURN(ERR_PTR(rc)); - - capa->lc_uid = ((__u64)d[1] << 32) | d[0]; - capa->lc_gid = ((__u64)d[3] << 32) | d[2]; - capa->lc_flags = LC_ID_CONVERT; - break; - } - default: - RETURN(ERR_PTR(-EINVAL)); - } - - capa->lc_fid = *fid; - capa->lc_opc = opc; - capa->lc_flags |= osd->od_capa_alg << 24; - capa->lc_timeout = osd->od_capa_timeout; - capa->lc_expiry = 0; - - oc = capa_lookup(osd->od_capa_hash, capa, 1); - if (oc) { - LASSERT(!capa_is_expired(oc)); - RETURN(oc); - } - - spin_lock(&capa_lock); - *key = osd->od_capa_keys[1]; - spin_unlock(&capa_lock); - - capa->lc_keyid = key->lk_keyid; - capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout; - - rc = capa_hmac(capa->lc_hmac, capa, key->lk_key); - if (rc) { - DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc); - RETURN(ERR_PTR(rc)); - } - - oc = capa_add(osd->od_capa_hash, capa); - RETURN(oc); + rc = inode->i_op->removexattr(dentry, name); + osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); + return rc; } static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, @@ -3202,36 +3061,16 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - file->f_dentry = dentry; + file->f_path.dentry = dentry; file->f_mapping = inode->i_mapping; file->f_op = inode->i_fop; set_file_inode(file, inode); -#ifdef HAVE_FILE_FSYNC_4ARGS - rc = file->f_op->fsync(file, start, end, 0); -#elif defined(HAVE_FILE_FSYNC_2ARGS) - mutex_lock(&inode->i_mutex); - rc = file->f_op->fsync(file, 0); - mutex_unlock(&inode->i_mutex); -#else - mutex_lock(&inode->i_mutex); - rc = file->f_op->fsync(file, dentry, 0); - mutex_unlock(&inode->i_mutex); -#endif + rc = ll_vfs_fsync_range(file, start, end, 0); RETURN(rc); } -static int osd_data_get(const struct lu_env *env, struct dt_object *dt, - void **data) -{ - struct osd_object *obj = osd_dt_obj(dt); - ENTRY; - - *data = (void *)obj->oo_inode; - RETURN(0); -} - /* * Index operations. */ @@ -3358,17 +3197,16 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } LINVRNT(osd_invariant(obj)); - if (result == 0 && is_quota_glb_feat(feat) && + if (result == 0 && feat == &dt_quota_glb_features && fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB) - result = osd_quota_migration(env, dt, feat); + result = osd_quota_migration(env, dt); return result; } static int osd_otable_it_attr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct lustre_capa *capa) + struct lu_attr *attr) { attr->la_valid = 0; return 0; @@ -3399,9 +3237,7 @@ static const struct dt_object_operations osd_obj_ops = { .do_declare_xattr_del = osd_declare_xattr_del, .do_xattr_del = osd_xattr_del, .do_xattr_list = osd_xattr_list, - .do_capa_get = osd_capa_get, .do_object_sync = osd_object_sync, - .do_data_get = osd_data_get, }; /** @@ -3433,9 +3269,7 @@ static const struct dt_object_operations osd_obj_ea_ops = { .do_declare_xattr_del = osd_declare_xattr_del, .do_xattr_del = osd_xattr_del, .do_xattr_list = osd_xattr_list, - .do_capa_get = osd_capa_get, .do_object_sync = osd_object_sync, - .do_data_get = osd_data_get, }; static const struct dt_object_operations osd_obj_otable_it_ops = { @@ -3470,28 +3304,25 @@ static int osd_index_declare_iam_delete(const struct lu_env *env, * \retval 0 success * \retval -ve failure */ - static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) -{ - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thandle *oh; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - int rc; - - ENTRY; + const struct dt_key *key, + struct thandle *handle) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thandle *oh; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + int rc; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); - LASSERT(handle != NULL); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3512,6 +3343,7 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); osd_ipd_put(env, bag, ipd); LINVRNT(osd_invariant(obj)); + osd_trans_exec_check(env, handle, OSD_OT_DELETE); RETURN(rc); } @@ -3525,20 +3357,22 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); + /* due to DNE we may need to remove an agent inode */ osd_trans_declare_op(env, oh, OSD_OT_DELETE, - osd_dto_credits_noquota[DTO_INDEX_DELETE]); + osd_dto_credits_noquota[DTO_INDEX_DELETE] + + osd_dto_credits_noquota[DTO_OBJECT_DELETE]); inode = osd_dt_obj(dt)->oo_inode; LASSERT(inode); rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - 0, oh, true, true, NULL, false); + 0, oh, osd_dt_obj(dt), true, NULL, false); RETURN(rc); } @@ -3551,19 +3385,29 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, if (de->file_type & LDISKFS_DIRENT_LUFID) { rec = (struct osd_fid_pack *) (de->name + de->name_len + 1); rc = osd_fid_unpack((struct lu_fid *)fid, rec); + if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid))) + rc = -EINVAL; } return rc; } static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, - struct lu_fid *fid) + const struct lu_fid *fid) { + struct seq_server_site *ss = osd_seq_site(osd); ENTRY; /* FID seqs not in FLDB, must be local seq */ if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) RETURN(0); + /* If FLD is not being initialized yet, it only happens during the + * initialization, likely during mgs initialization, and we assume + * this is local FID. */ + if (ss == NULL || ss->ss_server_fld == NULL) + RETURN(0); + + /* Only check the local FLDB here */ if (osd_seq_exists(env, osd, fid_seq(fid))) RETURN(0); @@ -3581,25 +3425,26 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, * \retval -ve, on error */ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct osd_thandle *oh; + const struct dt_key *key, struct thandle *handle) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct osd_thandle *oh; struct ldiskfs_dir_entry_2 *de = NULL; - struct buffer_head *bh; - struct htree_lock *hlock = NULL; + struct buffer_head *bh; + struct htree_lock *hlock = NULL; struct lu_fid *fid = &osd_oti_get(env)->oti_fid; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); int rc; - ENTRY; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); + + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3607,9 +3452,6 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, LASSERT(oh->ot_handle != NULL); LASSERT(oh->ot_handle->h_transaction != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); - ll_vfs_dq_init(dir); dentry = osd_child_dentry_get(env, obj, (char *)key, strlen((char *)key)); @@ -3723,6 +3565,7 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, out: LASSERT(osd_invariant(obj)); + osd_trans_exec_check(env, handle, OSD_OT_DELETE); RETURN(rc); } @@ -3738,25 +3581,23 @@ out: * \retval -ve failure */ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, - struct dt_rec *rec, const struct dt_key *key, - struct lustre_capa *capa) + struct dt_rec *rec, const struct dt_key *key) { - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - struct osd_thread_info *oti = osd_oti_get(env); - struct iam_iterator *it = &oti->oti_idx_it; - struct iam_rec *iam_rec; - int rc; - - ENTRY; + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_thread_info *oti = osd_oti_get(env); + struct iam_iterator *it = &oti->oti_idx_it; + struct iam_rec *iam_rec; + int rc; + ENTRY; - LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) - RETURN(-EACCES); + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); ipd = osd_idx_ipd_get(env, bag); if (IS_ERR(ipd)) @@ -3827,27 +3668,28 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, * \retval -ve failure */ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, struct thandle *th, - struct lustre_capa *capa, int ignore_quota) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct osd_thandle *oh; - struct iam_container *bag = &obj->oo_dir->od_container; - struct osd_thread_info *oti = osd_oti_get(env); - struct iam_rec *iam_rec; - int rc; + const struct dt_rec *rec, + const struct dt_key *key, struct thandle *th, + int ignore_quota) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct osd_thandle *oh; + struct iam_container *bag; + struct osd_thread_info *oti = osd_oti_get(env); + struct iam_rec *iam_rec; + int rc; + ENTRY; - ENTRY; + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); - LASSERT(th != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + bag = &obj->oo_dir->od_container; + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); @@ -3876,6 +3718,7 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, iam_rec, ipd); osd_ipd_put(env, bag, ipd); LINVRNT(osd_invariant(obj)); + osd_trans_exec_check(env, th, OSD_OT_INSERT); RETURN(rc); } @@ -3888,9 +3731,9 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, * \retval -ve, on error */ static int __osd_ea_add_rec(struct osd_thread_info *info, - struct osd_object *pobj, struct inode *cinode, - const char *name, const struct dt_rec *fid, - struct htree_lock *hlock, struct thandle *th) + struct osd_object *pobj, struct inode *cinode, + const char *name, const struct lu_fid *fid, + struct htree_lock *hlock, struct thandle *th) { struct ldiskfs_dentry_param *ldp; struct dentry *child; @@ -3912,6 +3755,29 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, child->d_fsdata = (void *)ldp; ll_vfs_dq_init(pobj->oo_inode); rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); + if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) { + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + int rc1; + + bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de, + NULL, hlock); + if (bh != NULL) { + rc1 = ldiskfs_journal_get_write_access(oth->ot_handle, + bh); + if (rc1 == 0) { + if (S_ISDIR(cinode->i_mode)) + de->file_type = LDISKFS_DIRENT_LUFID | + LDISKFS_FT_REG_FILE; + else + de->file_type = LDISKFS_DIRENT_LUFID | + LDISKFS_FT_DIR; + ldiskfs_handle_dirty_metadata(oth->ot_handle, + NULL, bh); + brelse(bh); + } + } + } RETURN(rc); } @@ -3929,11 +3795,11 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, * \retval -ve, on error */ static int osd_add_dot_dotdot(struct osd_thread_info *info, - struct osd_object *dir, - struct inode *parent_dir, const char *name, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, - struct thandle *th) + struct osd_object *dir, + struct inode *parent_dir, const char *name, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, + struct thandle *th) { struct inode *inode = dir->oo_inode; struct osd_thandle *oth; @@ -3960,9 +3826,19 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, dot_dot_fid, NULL, th); } - result = osd_add_dot_dotdot_internal(info, dir->oo_inode, - parent_dir, dot_fid, - dot_dot_fid, oth); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) { + struct lu_fid tfid = *dot_dot_fid; + + tfid.f_oid--; + result = osd_add_dot_dotdot_internal(info, + dir->oo_inode, parent_dir, dot_fid, + &tfid, oth); + } else { + result = osd_add_dot_dotdot_internal(info, + dir->oo_inode, parent_dir, dot_fid, + dot_dot_fid, oth); + } + if (result == 0) dir->oo_compat_dotdot_created = 1; } @@ -3976,8 +3852,8 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, * value, return by respective functions. */ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, - struct inode *cinode, const char *name, - const struct dt_rec *fid, struct thandle *th) + struct inode *cinode, const char *name, + const struct lu_fid *fid, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); struct htree_lock *hlock; @@ -3992,9 +3868,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, pobj->oo_inode, 0); } else { down_write(&pobj->oo_ext_idx_sem); - } - rc = osd_add_dot_dotdot(info, pobj, cinode, name, - (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), + } + + rc = osd_add_dot_dotdot(info, pobj, cinode, name, + lu_object_fid(&pobj->oo_dt.do_lu), fid, th); } else { if (hlock != NULL) { @@ -4007,11 +3884,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; - *tfid = *(const struct lu_fid *)fid; + *tfid = *fid; tfid->f_ver = ~0; rc = __osd_ea_add_rec(info, pobj, cinode, name, - (const struct dt_rec *)tfid, - hlock, th); + tfid, hlock, th); } else { rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); @@ -4039,6 +3915,9 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, if (!fid_is_norm(fid) && !fid_is_igif(fid)) RETURN_EXIT; + if (scrub->os_pos_current > id->oii_ino) + RETURN_EXIT; + again: rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD); if (rc != 0 && rc != -ENOENT) @@ -4060,12 +3939,13 @@ again: } if (!dev->od_noscrub && ++once == 1) { - rc = osd_scrub_start(dev); - LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID - ", rc = %d [2]\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, - PFID(fid), rc); - if (rc == 0) + rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN | + SS_CLEAR_FAILOUT); + CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC " + "for "DFID", rc = %d [2]\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, + PFID(fid), rc); + if (rc == 0 || rc == -EALREADY) goto again; } @@ -4097,16 +3977,91 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, return rc; } -int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, - struct osd_inode_id *id, const struct lu_fid *fid) +void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, + struct osd_inode_id *id, const struct lu_fid *fid) { CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid), id->oii_ino, id->oii_gen, info); info->oti_cache.oic_lid = *id; info->oti_cache.oic_fid = *fid; info->oti_cache.oic_dev = osd; +} - return 0; +/** + * Get parent FID from the linkEA. + * + * For a directory which parent resides on remote MDT, to satisfy the + * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On + * the other hand, to make the lookup(..) on the directory can return + * the real parent FID, we append the real parent FID after its ".." + * name entry in the /REMOTE_PARENT_DIR. + * + * Unfortunately, such PFID-in-dirent cannot be preserved via file-level + * backup. So after the restore, we cannot get the right parent FID from + * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since + * we have stored the real parent FID in the directory object's linkEA, + * we can parse the linkEA for the real parent FID. + * + * \param[in] env pointer to the thread context + * \param[in] obj pointer to the object to be handled + * \param[out]fid pointer to the buffer to hold the parent FID + * + * \retval 0 for getting the real parent FID successfully + * \retval negative error number on failure + */ +static int osd_get_pfid_from_linkea(const struct lu_env *env, + struct osd_object *obj, + struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct inode *inode = obj->oo_inode; + struct linkea_data ldata = { NULL }; + int rc; + ENTRY; + + fid_zero(fid); + if (!S_ISDIR(inode->i_mode)) + RETURN(-EIO); + +again: + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + buf->lb_buf, buf->lb_len); + if (rc == -ERANGE) { + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + NULL, 0); + if (rc > 0) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + } + + if (unlikely(rc == 0)) + RETURN(-ENODATA); + + if (rc < 0) + RETURN(rc); + + if (unlikely(buf->lb_buf == NULL)) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + + ldata.ld_buf = buf; + rc = linkea_init(&ldata); + if (rc == 0) { + linkea_first_entry(&ldata); + linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid); + } + + RETURN(rc); } /** @@ -4118,39 +4073,38 @@ int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, * \retval -ve, on error */ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, - struct dt_rec *rec, const struct dt_key *key) -{ - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct ldiskfs_dir_entry_2 *de; - struct buffer_head *bh; - struct lu_fid *fid = (struct lu_fid *) rec; - struct htree_lock *hlock = NULL; - int ino; - int rc; + struct dt_rec *rec, const struct dt_key *key) +{ + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + struct lu_fid *fid = (struct lu_fid *) rec; + struct htree_lock *hlock = NULL; + int ino; + int rc; ENTRY; - LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL); + LASSERT(dir->i_op != NULL); + LASSERT(dir->i_op->lookup != NULL); - dentry = osd_child_dentry_get(env, obj, - (char *)key, strlen((char *)key)); + dentry = osd_child_dentry_get(env, obj, + (char *)key, strlen((char *)key)); - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - dir, LDISKFS_HLOCK_LOOKUP); - } else { + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + dir, LDISKFS_HLOCK_LOOKUP); + } else { down_read(&obj->oo_ext_idx_sem); - } + } - bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); - if (bh) { + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + if (bh) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; struct osd_idmap_cache *oic = &oti->oti_cache; struct osd_device *dev = osd_obj2dev(obj); - struct osd_scrub *scrub = &dev->od_scrub; - struct scrub_file *sf = &scrub->os_file; ino = le32_to_cpu(de->inode); if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) { @@ -4163,10 +4117,24 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, /* done with de, release bh */ brelse(bh); - if (rc != 0) - rc = osd_ea_fid_get(env, obj, ino, fid, id); - else + if (rc != 0) { + if (unlikely(ino == osd_remote_parent_ino(dev))) { + const char *name = (const char *)key; + + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + if (likely(strlen(name) == 2 && + name[0] == '.' && name[1] == '.')) + rc = osd_get_pfid_from_linkea(env, obj, + fid); + } else { + rc = osd_ea_fid_get(env, obj, ino, fid, id); + } + } else { osd_id_gen(id, ino, OSD_OII_NOGEN); + } + if (rc != 0) { fid_zero(&oic->oic_fid); GOTO(out, rc); @@ -4175,16 +4143,8 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, if (osd_remote_fid(env, dev, fid)) GOTO(out, rc = 0); - rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, - fid); - if (rc != 0) - GOTO(out, rc); - if ((scrub->os_pos_current <= ino) && - ((sf->sf_flags & SF_INCONSISTENT) || - (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || - ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), - sf->sf_oi_bitmap))) - osd_consistency_check(oti, dev, oic); + osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid); + osd_consistency_check(oti, dev, oic); } else { rc = -ENOENT; } @@ -4207,9 +4167,9 @@ out: * \retval osd_object on success * \retval -ve on error */ -struct osd_object *osd_object_find(const struct lu_env *env, - struct dt_object *dt, - const struct lu_fid *fid) +static struct osd_object *osd_object_find(const struct lu_env *env, + struct dt_object *dt, + const struct lu_fid *fid) { struct lu_device *ludev = dt->do_lu.lo_dev; struct osd_object *child = NULL; @@ -4278,7 +4238,7 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, struct osd_thandle *oh; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); struct lu_fid *fid = (struct lu_fid *)rec; - int rc; + int credits, rc = 0; ENTRY; LASSERT(!dt_object_remote(dt)); @@ -4287,8 +4247,21 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - osd_trans_declare_op(env, oh, OSD_OT_INSERT, - osd_dto_credits_noquota[DTO_INDEX_INSERT]); + credits = osd_dto_credits_noquota[DTO_INDEX_INSERT]; + if (fid != NULL) { + rc = osd_remote_fid(env, osd, fid); + if (unlikely(rc < 0)) + RETURN(rc); + if (rc > 0) { + /* a reference to remote inode is represented by an + * agent inode which we have to create */ + credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE]; + credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; + } + rc = 0; + } + + osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits); if (osd_dt_obj(dt)->oo_inode != NULL) { struct inode *inode = osd_dt_obj(dt)->oo_inode; @@ -4297,26 +4270,10 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, * calculate how many blocks will be consumed by this index * insert */ rc = osd_declare_inode_qid(env, i_uid_read(inode), - i_gid_read(inode), 0, - oh, true, true, NULL, false); + i_gid_read(inode), 0, oh, + osd_dt_obj(dt), true, NULL, false); } - if (fid == NULL) - RETURN(0); - - rc = osd_remote_fid(env, osd, fid); - if (rc <= 0) - RETURN(rc); - - rc = 0; - - osd_trans_declare_op(env, oh, OSD_OT_CREATE, - osd_dto_credits_noquota[DTO_OBJECT_CREATE]); - osd_trans_declare_op(env, oh, OSD_OT_INSERT, - osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); - osd_trans_declare_op(env, oh, OSD_OT_INSERT, - osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); - RETURN(rc); } @@ -4332,13 +4289,14 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, * \retval -ve, on error */ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, struct thandle *th, - struct lustre_capa *capa, int ignore_quota) + const struct dt_rec *rec, + const struct dt_key *key, struct thandle *th, + int ignore_quota) { struct osd_object *obj = osd_dt_obj(dt); struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct lu_fid *fid = (struct lu_fid *) rec; + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; const char *name = (const char *)key; struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -4347,16 +4305,16 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, int rc; ENTRY; - LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(th != NULL); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - osd_trans_exec_op(env, th, OSD_OT_INSERT); + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(th != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + osd_trans_exec_op(env, th, OSD_OT_INSERT); - LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid)); + LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid)); rc = osd_remote_fid(env, osd, fid); if (rc < 0) { @@ -4385,8 +4343,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = igrab(omm->omm_remote_parent->d_inode); } else { child_inode = osd_create_local_agent_inode(env, osd, - obj, fid, - th); + obj, fid, rec1->rec_type & S_IFMT, th); if (IS_ERR(child_inode)) RETURN(PTR_ERR(child_inode)); } @@ -4403,7 +4360,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = igrab(child->oo_inode); } - rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th); + rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th); CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n", obj->oo_inode->i_ino, name, child_inode->i_ino, rc); @@ -4412,6 +4369,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, if (child != NULL) osd_object_put(env, child); LASSERT(osd_invariant(obj)); + osd_trans_exec_check(env, th, OSD_OT_INSERT); RETURN(rc); } @@ -4422,32 +4380,33 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, */ static struct dt_it *osd_it_iam_init(const struct lu_env *env, - struct dt_object *dt, - __u32 unused, - struct lustre_capa *capa) + struct dt_object *dt, + __u32 unused) { - struct osd_it_iam *it; - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct lu_object *lo = &dt->do_lu; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - - LASSERT(lu_object_exists(lo)); - - if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ)) - return ERR_PTR(-EACCES); - - it = &oti->oti_it; - ipd = osd_it_ipd_get(env, bag); - if (likely(ipd != NULL)) { - it->oi_obj = obj; - it->oi_ipd = ipd; - lu_object_get(lo); - iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd); - return (struct dt_it *)it; - } - return ERR_PTR(-ENOMEM); + struct osd_it_iam *it; + struct osd_object *obj = osd_dt_obj(dt); + struct lu_object *lo = &dt->do_lu; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + + if (!dt_object_exists(dt)) + return ERR_PTR(-ENOENT); + + OBD_ALLOC_PTR(it); + if (it == NULL) + return ERR_PTR(-ENOMEM); + + ipd = osd_it_ipd_get(env, bag); + if (likely(ipd != NULL)) { + it->oi_obj = obj; + it->oi_ipd = ipd; + lu_object_get(lo); + iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd); + return (struct dt_it *)it; + } else { + OBD_FREE_PTR(it); + return ERR_PTR(-ENOMEM); + } } /** @@ -4456,12 +4415,13 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; - struct osd_object *obj = it->oi_obj; + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; - iam_it_fini(&it->oi_it); - osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); - lu_object_put(env, &obj->oo_dt.do_lu); + iam_it_fini(&it->oi_it); + osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); + lu_object_put(env, &obj->oo_dt.do_lu); + OBD_FREE_PTR(it); } /** @@ -4699,44 +4659,56 @@ static const struct dt_index_operations osd_index_iam_ops = { * */ static struct dt_it *osd_it_ea_init(const struct lu_env *env, - struct dt_object *dt, - __u32 attr, - struct lustre_capa *capa) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *it = &info->oti_it_ea; - struct file *file = &it->oie_file; - struct lu_object *lo = &dt->do_lu; - struct dentry *obj_dentry = &info->oti_it_dentry; - ENTRY; - LASSERT(lu_object_exists(lo)); + struct dt_object *dt, + __u32 attr) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie; + struct file *file; + struct lu_object *lo = &dt->do_lu; + struct dentry *obj_dentry; + ENTRY; - obj_dentry->d_inode = obj->oo_inode; - obj_dentry->d_sb = osd_sb(osd_obj2dev(obj)); - obj_dentry->d_name.hash = 0; + if (!dt_object_exists(dt) || obj->oo_destroyed) + RETURN(ERR_PTR(-ENOENT)); - it->oie_rd_dirent = 0; - it->oie_it_dirent = 0; - it->oie_dirent = NULL; - it->oie_buf = info->oti_it_ea_buf; - it->oie_obj = obj; + OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS); + if (oie == NULL) + RETURN(ERR_PTR(-ENOMEM)); + obj_dentry = &oie->oie_dentry; + + obj_dentry->d_inode = obj->oo_inode; + obj_dentry->d_sb = osd_sb(osd_obj2dev(obj)); + obj_dentry->d_name.hash = 0; + + oie->oie_rd_dirent = 0; + oie->oie_it_dirent = 0; + oie->oie_dirent = NULL; + if (unlikely(!info->oti_it_ea_buf_used)) { + oie->oie_buf = info->oti_it_ea_buf; + info->oti_it_ea_buf_used = 1; + } else { + OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE); + if (oie->oie_buf == NULL) + RETURN(ERR_PTR(-ENOMEM)); + } + oie->oie_obj = obj; + + file = &oie->oie_file; - /* Reset the "file" totally to avoid to reuse any old value from - * former readdir handling, the "file->f_pos" should be zero. */ - memset(file, 0, sizeof(*file)); /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */ if (attr & LUDA_64BITHASH) file->f_mode = FMODE_64BITHASH; else file->f_mode = FMODE_32BITHASH; - file->f_dentry = obj_dentry; - file->f_mapping = obj->oo_inode->i_mapping; + file->f_path.dentry = obj_dentry; + file->f_mapping = obj->oo_inode->i_mapping; file->f_op = obj->oo_inode->i_fop; set_file_inode(file, obj->oo_inode); lu_object_get(lo); - RETURN((struct dt_it *) it); + RETURN((struct dt_it *) oie); } /** @@ -4746,14 +4718,20 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, */ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - struct osd_object *obj = it->oie_obj; - struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie = (struct osd_it_ea *)di; + struct osd_object *obj = oie->oie_obj; + struct inode *inode = obj->oo_inode; - ENTRY; - it->oie_file.f_op->release(inode, &it->oie_file); - lu_object_put(env, &obj->oo_dt.do_lu); - EXIT; + ENTRY; + oie->oie_file.f_op->release(inode, &oie->oie_file); + lu_object_put(env, &obj->oo_dt.do_lu); + if (unlikely(oie->oie_buf != info->oti_it_ea_buf)) + OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE); + else + info->oti_it_ea_buf_used = 0; + OBD_SLAB_FREE_PTR(oie, osd_itea_cachep); + EXIT; } /** @@ -4998,22 +4976,12 @@ osd_dirent_update(handle_t *jh, struct super_block *sb, LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack)); rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) { - CERROR("%.16s: fail to write access for update dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); - rc = ldiskfs_journal_dirty_metadata(jh, bh); - if (rc != 0) - CERROR("%.16s: fail to dirty metadata for update dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); RETURN(rc); } @@ -5062,13 +5030,8 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, if (osd_dirent_has_space(de->rec_len, ent->oied_namelen, dir->i_sb->s_blocksize)) { rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) { - CERROR("%.16s: fail to write access for reinsert " - "dirent: name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } de->name[de->name_len] = 0; rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); @@ -5076,39 +5039,30 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); de->file_type |= LDISKFS_DIRENT_LUFID; - rc = ldiskfs_journal_dirty_metadata(jh, bh); - if (rc != 0) - CERROR("%.16s: fail to dirty metadata for reinsert " - "dirent: name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); RETURN(rc); } rc = ldiskfs_delete_entry(jh, dir, de, bh); - if (rc != 0) { - CERROR("%.16s: fail to delete entry for reinsert dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name, ent->oied_namelen); ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp; - osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid); + osd_get_ldiskfs_dirent_param(ldp, fid); dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); /* It is too bad, we cannot reinsert the name entry back. * That means we lose it! */ if (rc != 0) - CERROR("%.16s: fail to insert entry for reinsert dirent: " - "name = %.*s, rc = %d\n", + CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, " + "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, PFID(fid), rc); RETURN(rc); } @@ -5165,14 +5119,15 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; again: - if (dev->od_dirent_journal) { + if (dev->od_dirent_journal != 0) { jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); - CERROR("%.16s: fail to start trans for dirent " - "check_repair: credits %d, name %.*s, rc %d\n", - devname, credits, ent->oied_namelen, - ent->oied_name, rc); + CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent " + "check_repair, dir = %lu/%u, credits = %d, " + "name = %.*s: rc = %d\n", + devname, dir->i_ino, dir->i_generation, credits, + ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } @@ -5213,18 +5168,18 @@ again: inode = osd_iget(info, dev, id); if (IS_ERR(inode)) { rc = PTR_ERR(inode); - if (rc == -ENOENT || rc == -ESTALE) { - *attr |= LUDA_IGNORE; - rc = 0; - } + if (rc == -ENOENT || rc == -ESTALE) + rc = 1; + else + CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent " + "check_repair, dir = %lu/%u, name = %.*s: " + "rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, rc); GOTO(out_journal, rc); } - /* skip the REMOTE_PARENT_DIR. */ - if (inode == dev->od_mdt_map->omm_remote_parent->d_inode) - GOTO(out_inode, rc = 0); - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc == 0) { LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI)); @@ -5240,7 +5195,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5257,6 +5212,13 @@ again: rc = osd_dirent_update(jh, sb, ent, fid, bh, de); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to update FID " + "in the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { @@ -5265,7 +5227,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5283,6 +5245,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to append FID " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } else if (rc == -ENODATA) { /* Do not repair under dryrun mode. */ @@ -5297,7 +5266,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5315,6 +5284,13 @@ again: rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to set LMA for " + "update dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { lu_igif_build(fid, inode->i_ino, inode->i_generation); /* It is probably IGIF object. Only aappend the @@ -5323,6 +5299,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_UPGRADE; + else + CDEBUG(D_LFSCK, "%.16s: fail to append IGIF " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } @@ -5336,7 +5319,7 @@ out_journal: if (hlock != NULL) { ldiskfs_htree_unlock(hlock); } else { - if (dev->od_dirent_journal) + if (dev->od_dirent_journal != 0) up_write(&obj->oo_ext_idx_sem); else up_read(&obj->oo_ext_idx_sem); @@ -5365,22 +5348,27 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_it_ea *it = (struct osd_it_ea *)di; struct osd_object *obj = it->oie_obj; struct osd_device *dev = osd_obj2dev(obj); - struct osd_scrub *scrub = &dev->od_scrub; - struct scrub_file *sf = &scrub->os_file; struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; - struct osd_idmap_cache *oic = &oti->oti_cache; struct lu_fid *fid = &it->oie_dirent->oied_fid; struct lu_dirent *lde = (struct lu_dirent *)dtrec; __u32 ino = it->oie_dirent->oied_ino; int rc = 0; ENTRY; + LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode); + if (attr & LUDA_VERIFY) { - attr |= LUDA_TYPE; - if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { + if (unlikely(ino == osd_remote_parent_ino(dev))) { attr |= LUDA_IGNORE; - rc = 0; + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + if (!fid_is_sane(fid) && + it->oie_dirent->oied_namelen == 2 && + it->oie_dirent->oied_name[0] == '.' && + it->oie_dirent->oied_name[1] == '.') + osd_get_pfid_from_linkea(env, obj, fid); } else { rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); @@ -5388,13 +5376,23 @@ static inline int osd_it_ea_rec(const struct lu_env *env, } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { - if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) && - likely(it->oie_dirent->oied_namelen != 2 || - it->oie_dirent->oied_name[0] != '.' || - it->oie_dirent->oied_name[1] != '.')) - RETURN(-ENOENT); + bool is_dotdot = false; + if (it->oie_dirent->oied_namelen == 2 && + it->oie_dirent->oied_name[0] == '.' && + it->oie_dirent->oied_name[1] == '.') + is_dotdot = true; + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + if (ino == osd_remote_parent_ino(dev) && is_dotdot) { + rc = osd_get_pfid_from_linkea(env, obj, fid); + } else { + if (is_dotdot == false && + OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + RETURN(-ENOENT); - rc = osd_ea_fid_get(env, obj, ino, fid, id); + rc = osd_ea_fid_get(env, obj, ino, fid, id); + } } else { osd_id_gen(id, ino, OSD_OII_NOGEN); } @@ -5412,17 +5410,10 @@ static inline int osd_it_ea_rec(const struct lu_env *env, if (osd_remote_fid(env, dev, fid)) RETURN(0); - if (likely(!(attr & LUDA_IGNORE))) - rc = osd_add_oi_cache(oti, dev, id, fid); - - if (!(attr & LUDA_VERIFY) && - (scrub->os_pos_current <= ino) && - ((sf->sf_flags & SF_INCONSISTENT) || - (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || - ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) - osd_consistency_check(oti, dev, oic); + if (likely(!(attr & LUDA_IGNORE) && rc == 0)) + osd_add_oi_cache(oti, dev, id, fid); - RETURN(rc); + RETURN(rc > 0 ? 0 : rc); } /** @@ -5499,8 +5490,7 @@ static int osd_it_ea_load(const struct lu_env *env, * \retval -ve, on error */ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, - struct dt_rec *rec, const struct dt_key *key, - struct lustre_capa *capa) + struct dt_rec *rec, const struct dt_key *key) { struct osd_object *obj = osd_dt_obj(dt); int rc = 0; @@ -5510,9 +5500,6 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, LASSERT(S_ISDIR(obj->oo_inode->i_mode)); LINVRNT(osd_invariant(obj)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) - return -EACCES; - rc = osd_ea_lookup_rec(env, obj, rec, key); if (rc == 0) rc = +1; @@ -5577,11 +5564,14 @@ static void osd_key_fini(const struct lu_context *ctx, { struct osd_thread_info *info = data; + if (info->oti_inode != NULL) + OBD_FREE_PTR(info->oti_inode); if (info->oti_hlock != NULL) ldiskfs_htree_lock_free(info->oti_hlock); OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); lu_buf_free(&info->oti_iobuf.dr_pg_buf); lu_buf_free(&info->oti_iobuf.dr_bl_buf); + lu_buf_free(&info->oti_big_buf); OBD_FREE_PTR(info); } @@ -5635,10 +5625,18 @@ static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA, osd->od_svname, ss->ss_server_seq); - if (rc != 0) { OBD_FREE_PTR(osd->od_cl_seq); osd->od_cl_seq = NULL; + RETURN(rc); + } + + if (ss->ss_node_id == 0) { + /* If the OSD on the sequence controller(MDT0), then allocate + * sequence here, otherwise allocate sequence after connected + * to MDT0 (see mdt_register_lwp_callback()). */ + rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv, + &osd->od_cl_seq->lcs_space, env); } RETURN(rc); @@ -5698,7 +5696,7 @@ static int osd_mount(const struct lu_env *env, struct osd_thread_info *info = osd_oti_get(env); struct lu_fid *fid = &info->oti_fid; struct inode *inode; - int rc = 0; + int rc = 0, force_over_128tb = 0; ENTRY; if (o->od_mnt != NULL) @@ -5708,23 +5706,64 @@ static int osd_mount(const struct lu_env *env, RETURN(-E2BIG); strcpy(o->od_mntdev, dev); - OBD_PAGE_ALLOC(__page, GFP_IOFS); - if (__page == NULL) - GOTO(out, rc = -ENOMEM); - str = lustre_cfg_string(cfg, 2); s_flags = simple_strtoul(str, NULL, 0); str = strstr(str, ":"); if (str) lmd_flags = simple_strtoul(str + 1, NULL, 0); opts = lustre_cfg_string(cfg, 3); +#ifdef __BIG_ENDIAN + if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) { + CERROR("%s: device %s extents feature is not guaranteed to " + "work on big-endian systems. Use \"bigendian_extents\" " + "mount option to override.\n", name, dev); + RETURN(-EINVAL); + } +#endif + if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) + force_over_128tb = 1; + + __page = alloc_page(GFP_IOFS); + if (__page == NULL) + GOTO(out, rc = -ENOMEM); page = (unsigned long)page_address(__page); options = (char *)page; *options = '\0'; - if (opts == NULL) - strcat(options, "user_xattr,acl"); - else + if (opts != NULL) { + /* strip out the options for back compatiblity */ + static char *sout[] = { + "mballoc", + "iopen", + "noiopen", + "iopen_nopriv", + "extents", + "noextents", + /* strip out option we processed in osd */ + "bigendian_extents", + "force_over_128tb", + NULL + }; strcat(options, opts); + for (rc = 0, str = options; sout[rc]; ) { + char *op = strstr(str, sout[rc]); + if (op == NULL) { + rc++; + str = options; + continue; + } + if (op == options || *(op - 1) == ',') { + str = op + strlen(sout[rc]); + if (*str == ',' || *str == '\0') { + *str == ',' ? str++ : str; + memmove(op, str, strlen(str) + 1); + } + } + for (str = op; *str != ',' && *str != '\0'; str++) + ; + } + } else { + strncat(options, "user_xattr,acl", 14); + } /* Glom up mount options */ if (*options != '\0') @@ -5747,6 +5786,15 @@ static int osd_mount(const struct lu_env *env, GOTO(out, rc); } + if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) && + force_over_128tb == 0) { + CERROR("%s: device %s LDISKFS does not support filesystems " + "greater than 128TB and can cause data corruption. " + "Use \"force_over_128tb\" mount option to override.\n", + name, dev); + GOTO(out, rc = -EINVAL); + } + #ifdef HAVE_DEV_SET_RDONLY if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { CERROR("%s: underlying device %s is marked as read-only. " @@ -5790,7 +5838,7 @@ out_mnt: out: if (__page) - OBD_PAGE_FREE(__page); + __free_page(__page); return rc; } @@ -5832,10 +5880,6 @@ static int osd_device_init0(const struct lu_env *env, spin_lock_init(&o->od_osfs_lock); mutex_init(&o->od_otable_mutex); - o->od_capa_hash = init_capa_hash(); - if (o->od_capa_hash == NULL) - GOTO(out, rc = -ENOMEM); - o->od_read_cache = 1; o->od_writethrough_cache = 1; o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; @@ -5844,15 +5888,19 @@ static int osd_device_init0(const struct lu_env *env, sizeof(o->od_svname)); if (cplen >= sizeof(o->od_svname)) { rc = -E2BIG; - GOTO(out_capa, rc); + GOTO(out, rc); } - if (server_name_is_ost(o->od_svname)) + o->od_index = -1; /* -1 means index is invalid */ + rc = server_name2index(o->od_svname, &o->od_index, NULL); + if (rc == LDD_F_SV_TYPE_OST) o->od_is_ost = 1; + o->od_full_scrub_ratio = OFSR_DEFAULT; + o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT; rc = osd_mount(env, o, cfg); if (rc != 0) - GOTO(out_capa, rc); + GOTO(out, rc); rc = osd_obj_map_init(env, o); if (rc != 0) @@ -5867,9 +5915,6 @@ static int osd_device_init0(const struct lu_env *env, if (rc != 0) GOTO(out_site, rc); - /* self-repair LMA by default */ - o->od_lma_self_repair = 1; - INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ rc = osd_scrub_setup(env, o); @@ -5883,7 +5928,8 @@ static int osd_device_init0(const struct lu_env *env, GOTO(out_scrub, rc); } - LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev); + LASSERT(l->ld_site->ls_linkage.next != NULL); + LASSERT(l->ld_site->ls_linkage.prev != NULL); /* initialize quota slave instance */ o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev, @@ -5906,8 +5952,6 @@ out_compat: osd_obj_map_fini(o); out_mnt: osd_umount(env, o); -out_capa: - cleanup_capa_hash(o->od_capa_hash); out: return rc; } @@ -5945,7 +5989,6 @@ static struct lu_device *osd_device_free(const struct lu_env *env, struct osd_device *o = osd_dev(d); ENTRY; - cleanup_capa_hash(o->od_capa_hash); /* XXX: make osd top device in order to release reference */ d->ld_site->ls_top_dev = d; lu_site_purge(env, d->ld_site, -1); @@ -5976,13 +6019,12 @@ static int osd_process_config(const struct lu_env *env, break; case LCFG_PARAM: LASSERT(&o->od_dt_dev); - rc = class_process_proc_seq_param(PARAM_OSD, - lprocfs_osd_obd_vars, - cfg, &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_seq_param(PARAM_OST, - lprocfs_osd_obd_vars, - cfg, &o->od_dt_dev); + rc = class_process_proc_param(PARAM_OST, + lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); break; default: rc = -ENOSYS; @@ -6060,10 +6102,12 @@ static int osd_obd_disconnect(struct obd_export *exp) } static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, - struct lu_device *dev) + struct lu_device *dev) { - struct osd_device *osd = osd_dev(dev); - int result = 0; + struct osd_device *osd = osd_dev(dev); + struct lr_server_data *lsd = + &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd; + int result = 0; ENTRY; if (osd->od_quota_slave != NULL) { @@ -6073,13 +6117,28 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, RETURN(result); } + if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) { +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0) + if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) { + osd->od_index_in_idif = 1; + } else { + osd->od_index_in_idif = 0; + result = osd_register_proc_index_in_idif(osd); + if (result != 0) + RETURN(result); + } +#else + osd->od_index_in_idif = 1; +#endif + } + result = osd_fid_init(env, osd); RETURN(result); } -int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, - struct lu_fid *fid, struct md_op_data *op_data) +static 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); @@ -6116,13 +6175,21 @@ static const struct lu_device_type_operations osd_device_type_ops = { .ldto_device_fini = osd_device_fini }; -struct lu_device_type osd_device_type = { +static struct lu_device_type osd_device_type = { .ldt_tags = LU_DEVICE_DT, .ldt_name = LUSTRE_OSD_LDISKFS_NAME, .ldt_ops = &osd_device_type_ops, .ldt_ctx_tags = LCT_LOCAL, }; +static int osd_health_check(const struct lu_env *env, struct obd_device *obd) +{ + struct osd_device *osd = osd_dev(obd->obd_lu_dev); + struct super_block *sb = osd_sb(osd); + + return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY); +} + /* * lprocfs legacy support. */ @@ -6131,12 +6198,18 @@ static struct obd_ops osd_obd_device_ops = { .o_connect = osd_obd_connect, .o_disconnect = osd_obd_disconnect, .o_fid_alloc = osd_fid_alloc, + .o_health_check = osd_health_check, }; static int __init osd_mod_init(void) { int rc; +#if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK) + /* please, try to keep osd_thread_info smaller than a page */ + CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE); +#endif + osd_oi_mod_init(); rc = lu_kmem_init(ldiskfs_caches); @@ -6145,9 +6218,6 @@ static int __init osd_mod_init(void) rc = class_register_type(&osd_obd_device_ops, NULL, true, lprocfs_osd_module_vars, -#ifndef HAVE_ONLY_PROCFS_SEQ - NULL, -#endif LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); if (rc) lu_kmem_fini(ldiskfs_caches); @@ -6162,6 +6232,8 @@ static void __exit osd_mod_exit(void) MODULE_AUTHOR("Sun Microsystems, Inc. "); MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit); +module_init(osd_mod_init); +module_exit(osd_mod_exit);