X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=316429bb5dde24fe950f1bbf2aa02d79154827ea;hp=a321a5e96ca4e2cbe8fd6f7d0c29bafcf879938f;hb=5ca1a1e01d456c09d11d8a3409a83e055a7974a1;hpb=61b6ab5915648c423f0174f6cfdce640eee4e7ff diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index a321a5e..316429b 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, 2014, Intel Corporation. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -44,6 +44,10 @@ #define DEBUG_SUBSYSTEM S_OSD #include +#include +#ifdef HAVE_UIDGID_HEADER +# include +#endif /* LUSTRE_VERSION_CODE */ #include @@ -56,6 +60,7 @@ #include #include +#include #undef ENTRY /* * struct OBD_{ALLOC,FREE}*() @@ -78,12 +83,12 @@ #include int ldiskfs_pdo = 1; -CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, - "ldiskfs with parallel directory operations"); +module_param(ldiskfs_pdo, int, 0644); +MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations"); int ldiskfs_track_declares_assert; -CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644, - "LBUG during tracking of declares"); +module_param(ldiskfs_track_declares_assert, int, 0644); +MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares"); /* Slab to allocate dynlocks */ struct kmem_cache *dynlock_cachep; @@ -118,6 +123,11 @@ static const struct dt_object_operations osd_obj_otable_it_ops; static const struct dt_index_operations osd_index_iam_ops; static const struct dt_index_operations osd_index_ea_ops; +static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid); +static int osd_process_scheduled_agent_removals(const struct lu_env *env, + struct osd_device *osd); + int osd_trans_declare_op2rb[] = { [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET, [OSD_OT_PUNCH] = OSD_OT_MAX, @@ -166,6 +176,153 @@ static int osd_root_get(const struct lu_env *env, } /* + * the following set of functions are used to maintain per-thread + * cache of FID->ino mapping. this mechanism is needed to resolve + * FID to inode at dt_insert() which in turn stores ino in the + * directory entries to keep ldiskfs compatible with ext[34]. + * due to locking-originated restrictions we can't lookup ino + * using LU cache (deadlock is possible). lookup using OI is quite + * expensive. so instead we maintain this cache and methods like + * dt_create() fill it. so in the majority of cases dt_insert() is + * able to find needed mapping in lockless manner. + */ +static struct osd_idmap_cache * +osd_idc_find(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc = oti->oti_ins_cache; + int i; + for (i = 0; i < oti->oti_ins_cache_used; i++) { + if (!lu_fid_eq(&idc[i].oic_fid, fid)) + continue; + if (idc[i].oic_dev != osd) + continue; + + return idc + i; + } + + return NULL; +} + +static struct osd_idmap_cache * +osd_idc_add(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc; + int i; + + if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) { + i = oti->oti_ins_cache_size * 2; + if (i == 0) + i = OSD_INS_CACHE_SIZE; + OBD_ALLOC(idc, sizeof(*idc) * i); + if (idc == NULL) + return ERR_PTR(-ENOMEM); + if (oti->oti_ins_cache != NULL) { + memcpy(idc, oti->oti_ins_cache, + oti->oti_ins_cache_used * sizeof(*idc)); + OBD_FREE(oti->oti_ins_cache, + oti->oti_ins_cache_used * sizeof(*idc)); + } + oti->oti_ins_cache = idc; + oti->oti_ins_cache_size = i; + } + + idc = oti->oti_ins_cache + oti->oti_ins_cache_used++; + idc->oic_fid = *fid; + idc->oic_dev = osd; + idc->oic_lid.oii_ino = 0; + idc->oic_lid.oii_gen = 0; + idc->oic_remote = 0; + + return idc; +} + +/* + * lookup mapping for the given fid in the cache, initialize a + * new one if not found. the initialization checks whether the + * object is local or remote. for local objects, OI is used to + * learn ino/generation. the function is used when the caller + * has no information about the object, e.g. at dt_insert(). + */ +static struct osd_idmap_cache * +osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid) +{ + struct osd_idmap_cache *idc; + int rc; + + idc = osd_idc_find(env, osd, fid); + LASSERT(!IS_ERR(idc)); + if (idc != NULL) + return idc; + + /* new mapping is needed */ + idc = osd_idc_add(env, osd, fid); + if (IS_ERR(idc)) + return idc; + + /* initialize it */ + rc = osd_remote_fid(env, osd, fid); + if (unlikely(rc < 0)) + return ERR_PTR(rc); + + if (rc == 0) { + /* the object is local, lookup in OI */ + /* XXX: probably cheaper to lookup in LU first? */ + rc = osd_oi_lookup(osd_oti_get(env), osd, fid, + &idc->oic_lid, 0); + if (unlikely(rc < 0)) { + CERROR("can't lookup: rc = %d\n", rc); + return ERR_PTR(rc); + } + } else { + /* the object is remote */ + idc->oic_remote = 1; + } + + return idc; +} + +/* + * lookup mapping for given FID and fill it from the given object. + * the object is lolcal by definition. + */ +static int osd_idc_find_and_init(const struct lu_env *env, + struct osd_device *osd, + struct osd_object *obj) +{ + const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); + struct osd_idmap_cache *idc; + + idc = osd_idc_find(env, osd, fid); + LASSERT(!IS_ERR(idc)); + if (idc != NULL) { + if (obj->oo_inode == NULL) + return 0; + if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) { + LASSERT(idc->oic_lid.oii_ino == 0); + idc->oic_lid.oii_ino = obj->oo_inode->i_ino; + idc->oic_lid.oii_gen = obj->oo_inode->i_generation; + } + return 0; + } + + /* new mapping is needed */ + idc = osd_idc_add(env, osd, fid); + if (IS_ERR(idc)) + return PTR_ERR(idc); + + if (obj->oo_inode != NULL) { + idc->oic_lid.oii_ino = obj->oo_inode->i_ino; + idc->oic_lid.oii_gen = obj->oo_inode->i_generation; + } + return 0; +} + +/* * OSD object methods. */ @@ -233,6 +390,11 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, { struct inode *inode = NULL; + /* if we look for an inode withing a running + * transaction, then we risk to deadlock */ + /* osd_dirent_check_repair() breaks this */ + /*LASSERT(current->journal_info == NULL);*/ + inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n", @@ -256,6 +418,7 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, iput(inode); inode = ERR_PTR(-ENOENT); } else { + ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY); if (id->oii_gen == OSD_OII_NOGEN) osd_id_gen(id, inode->i_ino, inode->i_generation); @@ -270,6 +433,53 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, return inode; } +int osd_ldiskfs_add_entry(struct osd_thread_info *info, + handle_t *handle, struct dentry *child, + struct inode *inode, struct htree_lock *hlock) +{ + int rc, rc2; + + rc = __ldiskfs_add_entry(handle, child, inode, hlock); + if (rc == -ENOBUFS || rc == -ENOSPC) { + char fidbuf[FID_LEN + 1]; + struct lustre_mdt_attrs lma; + struct lu_fid fid = { }; + char *errstr; + struct dentry *p_dentry = child->d_parent; + + rc2 = osd_get_lma(info, p_dentry->d_inode, p_dentry, + &lma); + if (rc2 == 0) { + fid = lma.lma_self_fid; + snprintf(fidbuf, sizeof(fidbuf), DFID, PFID(&fid)); + } else if (rc2 == -ENODATA) { + if (unlikely(p_dentry->d_inode == + inode->i_sb->s_root->d_inode)) + lu_local_obj_fid(&fid, OSD_FS_ROOT_OID); + else if (info->oti_dev && !info->oti_dev->od_is_ost && + fid_seq_is_mdt0(fid_seq(&fid))) + lu_igif_build(&fid, p_dentry->d_inode->i_ino, + p_dentry->d_inode->i_generation); + snprintf(fidbuf, sizeof(fidbuf), DFID, PFID(&fid)); + } else { + snprintf(fidbuf, FID_LEN, "%s", "unknown"); + } + + if (rc == -ENOSPC) + errstr = "has reached"; + else + errstr = "is approaching"; + CWARN("%.16s: directory (inode: %lu FID: %s) %s maximum entry limit\n", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + p_dentry->d_inode->i_ino, fidbuf, errstr); + /* ignore such error now */ + if (rc == -ENOBUFS) + rc = 0; + } + return rc; +} + + static struct inode * osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, struct osd_inode_id *id, struct lu_fid *fid) @@ -301,16 +511,20 @@ 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) + bool cached) { struct inode *inode; int rc = 0; ENTRY; + /* The cached OI mapping is trustable. If we cannot locate the inode + * via the cached OI mapping, then return the failure to the caller + * directly without further OI checking. */ + inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { rc = PTR_ERR(inode); - if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) { + if (cached || (rc != -ENOENT && rc != -ESTALE)) { CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n", id->oii_ino, rc); @@ -322,7 +536,7 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, if (is_bad_inode(inode)) { rc = -ENOENT; - if (!in_oi) { + if (cached) { CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino); GOTO(put, rc); @@ -334,7 +548,7 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, if (id->oii_gen != OSD_OII_NOGEN && inode->i_generation != id->oii_gen) { rc = -ESTALE; - if (!in_oi) { + if (cached) { CDEBUG(D_INODE, "unmatched inode: ino = %u, " "oii_gen = %u, i_generation = %u\n", id->oii_ino, id->oii_gen, inode->i_generation); @@ -347,7 +561,7 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, if (inode->i_nlink == 0) { rc = -ENOENT; - if (!in_oi) { + if (cached) { CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); GOTO(put, rc); @@ -356,14 +570,14 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, goto check_oi; } + ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY); + check_oi: if (rc != 0) { - struct osd_inode_id saved_id = *id; - 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: + /* XXX: There are four possible cases: * 1. rc = 0. * Backup/restore caused the OI invalid. * 2. rc = 0. @@ -381,22 +595,17 @@ check_oi: * 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) { + inode->i_generation == id->oii_gen) + /* "id->oii_gen != OSD_OII_NOGEN" is for + * "@cached == false" case. */ rc = -ENOENT; - } else { - __u32 level = D_LFSCK; - + else rc = -EREMCHG; - if (!thread_is_running(&dev->od_scrub.os_thread)) - level |= D_CONSOLE; - - CDEBUG(level, "%s: the OI mapping for the FID " - DFID" become inconsistent, the given ID " - "%u/%u, the ID in OI mapping %u/%u\n", - osd_name(dev), PFID(fid), - saved_id.oii_ino, saved_id.oii_gen, - id->oii_ino, id->oii_ino); - } + } else { + /* If the OI mapping was in OI file before the + * osd_iget_check(), but now, it is disappear, + * then it must be removed by race. That is a + * normal race case. */ } } else { if (id->oii_gen == OSD_OII_NOGEN) @@ -441,7 +650,8 @@ int osd_get_idif(struct osd_thread_info *info, struct inode *inode, rc = 0; ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq)); ostid_set_id(ostid, le64_to_cpu(ff->ff_objid)); - /* XXX: should use real OST index in the future. LU-3569 */ + /* XXX: use 0 as the index for compatibility, the caller will + * handle index related issues when necessarry. */ ostid_to_fid(fid, ostid, 0); } else if (rc == sizeof(struct filter_fid)) { rc = 1; @@ -508,9 +718,6 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } } - if (unlikely(rc == -ENODATA)) - RETURN(0); - if (rc < 0) RETURN(rc); @@ -530,8 +737,6 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) { - __u32 level = D_LFSCK; - if (fid_is_idif(rfid) && fid_is_idif(fid)) { struct ost_id *oi = &info->oti_ostid; struct lu_fid *fid1 = &info->oti_fid3; @@ -555,13 +760,7 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } } - rc = -EREMCHG; - if (!thread_is_running(&osd->od_scrub.os_thread)) - level |= D_CONSOLE; - - CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n", - osd_name(osd), PFID(rfid), PFID(fid)); } RETURN(rc); @@ -581,7 +780,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, struct scrub_file *sf; int result; int saved = 0; - bool in_oi = false; + bool cached = true; bool triggered = false; ENTRY; @@ -596,7 +795,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, LASSERT(info); oic = &info->oti_cache; - if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) + if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) RETURN(-ENOENT); /* For the object is created as locking anchor, or for the object to @@ -622,10 +821,11 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, goto iget; } + cached = false; /* Search order: 3. OI files. */ result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result == -ENOENT) { - if (!fid_is_norm(fid) || + if (!(fid_is_norm(fid) || fid_is_igif(fid)) || fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), sf->sf_oi_bitmap)) @@ -637,23 +837,16 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (result != 0) GOTO(out, result); - in_oi = true; - iget: - inode = osd_iget_check(info, dev, fid, id, in_oi); + inode = osd_iget_check(info, dev, fid, id, cached); 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 = -ENOENT); - } else if (result == -EREMCHG) { -trigger: - if (!in_oi) - fid_zero(&oic->oic_fid); + if (result == -EREMCHG) { +trigger: if (unlikely(triggered)) GOTO(out, result = saved); @@ -661,20 +854,22 @@ trigger: 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); + 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; + } else { + result = -EREMCHG; } + if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) + GOTO(out, result); + /* 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 @@ -694,43 +889,81 @@ trigger: result = osd_lookup_in_remote_parent(info, dev, fid, id); if (result == 0) { - in_oi = false; + cached = true; goto iget; } result = saved; } - GOTO(out, result); - } + GOTO(out, result); + } - 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) { + if (result == -ENODATA) { + if (cached) { result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result != 0) { - fid_zero(&oic->oic_fid); + /* result == -ENOENT means that the OI + * mapping has been removed by race, + * the target inode belongs to other + * object. + * + * Others error also can be returned + * directly. */ + iput(inode); + obj->oo_inode = NULL; GOTO(out, result); + } else { + /* result == 0 means the cached OI + * mapping is still in the OI file, + * the target the inode is valid. */ } + } else { + /* The current OI mapping is from the OI file, + * since the inode has been found via + * osd_iget_check(), no need recheck OI. */ } - goto trigger; + goto found; } - GOTO(out, result); + iput(inode); + obj->oo_inode = NULL; + if (result != -EREMCHG) + GOTO(out, result); + + if (cached) { + result = osd_oi_lookup(info, dev, fid, id, + OI_CHECK_FLD); + /* result == -ENOENT means the cached OI mapping + * has been removed from the OI file by race, + * above target inode belongs to other object. + * + * Others error also can be returned directly. */ + if (result != 0) + GOTO(out, result); + + /* result == 0, goto trigger */ + } else { + /* The current OI mapping is from the OI file, + * since the inode has been found via + * osd_iget_check(), no need recheck OI. */ + } + + goto trigger; } +found: obj->oo_compat_dot_created = 1; obj->oo_compat_dotdot_created = 1; - if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */ + if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */ GOTO(out, result = 0); LASSERT(obj->oo_hl_head == NULL); @@ -743,6 +976,9 @@ trigger: GOTO(out, result = 0); out: + if (result != 0 && cached) + fid_zero(&oic->oic_fid); + LINVRNT(osd_invariant(obj)); return result; } @@ -778,8 +1014,22 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); obj->oo_dt.do_body_ops = &osd_body_ops_new; - if (result == 0 && obj->oo_inode != NULL) + if (result == 0 && obj->oo_inode != NULL) { + struct osd_thread_info *oti = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs; + osd_object_init0(obj); + result = osd_get_lma(oti, obj->oo_inode, + &oti->oti_obj_dentry, lma); + if (result == 0) { + /* Convert LMAI flags to lustre LMA flags + * and cache it to oo_lma_flags */ + obj->oo_lma_flags = + lma_to_lustre_flags(lma->lma_incompat); + } else if (result == -ENODATA) { + result = 0; + } + } LINVRNT(osd_invariant(obj)); return result; @@ -945,7 +1195,8 @@ 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); @@ -959,7 +1210,7 @@ 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, @@ -984,22 +1235,68 @@ 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); - 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. */ @@ -1028,41 +1325,13 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d, static unsigned long last_printed; static int last_credits; - CWARN("%.16s: too many transaction credits (%d > %d)\n", - 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]); - + /* don't make noise on a tiny testing systems + * actual credits misuse will be caught anyway */ if (last_credits != oh->ot_credits && time_after(jiffies, last_printed + - msecs_to_jiffies(60 * MSEC_PER_SEC))) { + msecs_to_jiffies(60 * MSEC_PER_SEC)) && + osd_transaction_size(dev) > 512) { + osd_trans_dump_creds(env, th); libcfs_debug_dumpstack(NULL); last_credits = oh->ot_credits; last_printed = jiffies; @@ -1122,13 +1391,29 @@ 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. */ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, struct thandle *th) { - int rc = 0; + int rc = 0, remove_agents = 0; struct osd_thandle *oh; struct osd_thread_info *oti = osd_oti_get(env); struct osd_iobuf *iobuf = &oti->oti_iobuf; @@ -1139,10 +1424,15 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, oh = container_of0(th, struct osd_thandle, ot_super); + /* reset OI cache for safety */ + oti->oti_ins_cache_used = 0; + + remove_agents = oh->ot_remove_agents; + 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; /* @@ -1155,11 +1445,13 @@ 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("%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; @@ -1169,7 +1461,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, CERROR("%s: failed to stop transaction: rc = %d\n", osd_name(osd), rc); } else { - thandle_put(&oh->ot_super); + osd_trans_stop_cb(oh, th->th_result); + OBD_FREE_PTR(oh); } /* inform the quota slave device that the transaction is stopping */ @@ -1190,6 +1483,9 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, if (!rc) rc = iobuf->dr_error; + if (unlikely(remove_agents != 0)) + osd_process_scheduled_agent_removals(env, osd); + RETURN(rc); } @@ -1200,7 +1496,10 @@ 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; } @@ -1252,6 +1551,12 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) static void osd_object_release(const struct lu_env *env, struct lu_object *l) { + struct osd_object *o = osd_obj(l); + /* nobody should be releasing a non-destroyed object with nlink=0 + * the API allows this, but ldiskfs doesn't like and then report + * this inode as deleted */ + if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0)) + LBUG(); } /* @@ -1275,18 +1580,17 @@ static int osd_object_print(const struct lu_env *env, void *cookie, d ? d->id_ops->id_name : "plain"); } -#define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */ - /* * Concurrency: shouldn't matter. */ int osd_statfs(const struct lu_env *env, struct dt_device *d, struct obd_statfs *sfs) { - struct osd_device *osd = osd_dt_dev(d); - struct super_block *sb = osd_sb(osd); - struct kstatfs *ksfs; - int result = 0; + struct osd_device *osd = osd_dt_dev(d); + struct super_block *sb = osd_sb(osd); + struct kstatfs *ksfs; + __u64 reserved; + int result = 0; if (unlikely(osd->od_mnt == NULL)) return -EINPROGRESS; @@ -1300,30 +1604,40 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, ksfs = &osd_oti_get(env)->oti_ksfs; } - spin_lock(&osd->od_osfs_lock); 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) - sfs->os_state = OS_STATE_READONLY; - } + if (result) + goto out; + + statfs_pack(sfs, ksfs); + 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); + /* + * Reserve some space so to avoid fragmenting the filesystem too much. + * Fragmentation not only impacts performance, but can also increase + * metadata overhead significantly, causing grant calculation to be + * wrong. + * + * Reserve 0.78% of total space, at least 8MB for small filesystems. + */ + CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE); + reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits; + if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT)) + reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT; - if (unlikely(env == NULL)) - OBD_FREE_PTR(ksfs); + sfs->os_blocks -= reserved; + sfs->os_bfree -= min(reserved, sfs->os_bfree); + sfs->os_bavail -= min(reserved, sfs->os_bavail); - /* Reserve a small amount of space for local objects like last_rcvd, - * llog, quota files, ... */ - if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) { - sfs->os_bavail = 0; - } else { - sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS; - /** Take out metadata overhead for indirect blocks */ - sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3); - } - - return result; +out: + if (unlikely(env == NULL)) + OBD_FREE_PTR(ksfs); + return result; } /** @@ -1352,18 +1666,23 @@ static void osd_conf_get(const struct lu_env *env, */ param->ddp_max_name_len = LDISKFS_NAME_LEN; param->ddp_max_nlink = LDISKFS_LINK_MAX; - param->ddp_block_shift = sb->s_blocksize_bits; + param->ddp_symlink_max = sb->s_blocksize; param->ddp_mount_type = LDD_MT_LDISKFS; - param->ddp_maxbytes = sb->s_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 */ + 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; /* inode are statically allocated, so per-inode space consumption * is the space consumed by the directory entry */ param->ddp_inodespace = PER_OBJ_USAGE; - /* per-fragment overhead to be used by the client code */ - param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb); - param->ddp_mntopts = 0; + /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks + * = 128MB) which is unlikely to be hit in real life. Report a smaller + * maximum length to not under count the actual number of extents + * needed for writing a file. */ + param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2; + /* worst-case extent insertion metadata overhead */ + param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb); + param->ddp_mntopts = 0; if (test_opt(sb, XATTR_USER)) param->ddp_mntopts |= MNTOPT_USERXATTR; if (test_opt(sb, POSIX_ACL)) @@ -1456,23 +1775,6 @@ 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. @@ -1542,7 +1844,6 @@ 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, @@ -1614,108 +1915,6 @@ 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) { @@ -1728,7 +1927,6 @@ static struct timespec *osd_inode_time(const struct lu_env *env, return t; } - static void osd_inode_getattr(const struct lu_env *env, struct inode *inode, struct lu_attr *attr) { @@ -1745,7 +1943,7 @@ static void osd_inode_getattr(const struct lu_env *env, attr->la_blocks = inode->i_blocks; attr->la_uid = i_uid_read(inode); attr->la_gid = i_gid_read(inode); - attr->la_flags = LDISKFS_I(inode)->i_flags; + attr->la_flags = ll_inode_to_ext_flags(inode->i_flags); attr->la_nlink = inode->i_nlink; attr->la_rdev = inode->i_rdev; attr->la_blksize = 1 << inode->i_blkbits; @@ -1754,23 +1952,24 @@ 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 lu_attr *attr) { struct osd_object *obj = osd_dt_obj(dt); - if (!dt_object_exists(dt)) + if (unlikely(!dt_object_exists(dt))) + return -ENOENT; + if (unlikely(obj->oo_destroyed)) return -ENOENT; LASSERT(!dt_object_remote(dt)); LINVRNT(osd_invariant(obj)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; - spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); + if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) + attr->la_flags |= LUSTRE_ORPHAN_FL; spin_unlock(&obj->oo_guard); + return 0; } @@ -1802,6 +2001,9 @@ static int osd_declare_attr_set(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, + osd_dto_credits_noquota[DTO_XATTR_SET]); + if (attr == NULL || obj->oo_inode == NULL) RETURN(rc); @@ -1997,8 +2199,7 @@ 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 thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode; @@ -2011,9 +2212,6 @@ static int osd_attr_set(const struct lu_env *env, LASSERT(!dt_object_remote(dt)); LASSERT(osd_invariant(obj)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; - osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) { @@ -2050,9 +2248,43 @@ static int osd_attr_set(const struct lu_env *env, spin_lock(&obj->oo_guard); rc = osd_inode_setattr(env, inode, attr); spin_unlock(&obj->oo_guard); + if (rc != 0) + GOTO(out, rc); + + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + + if (!(attr->la_valid & LA_FLAGS)) + GOTO(out, rc); + + /* Let's check if there are extra flags need to be set into LMA */ + if (attr->la_flags & LUSTRE_LMA_FL_MASKS) { + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + if (rc != 0) + GOTO(out, rc); + + lma->lma_incompat |= + lustre_to_lma_flags(attr->la_flags); + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, + lma, sizeof(*lma), XATTR_REPLACE); + if (rc != 0) { + struct osd_device *osd = osd_obj2dev(obj); + + CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n", + osd_name(osd), PFID(lu_object_fid(&dt->do_lu)), + lma->lma_incompat, rc); + } else { + obj->oo_lma_flags = + attr->la_flags & LUSTRE_LMA_FL_MASKS; + } + osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); + } +out: + osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); - if (!rc) - ll_dirty_inode(inode, I_DIRTY_DATASYNC); return rc; } @@ -2086,8 +2318,9 @@ 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 : @@ -2267,6 +2500,12 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, ah->dah_parent = parent; ah->dah_mode = child_mode; + + if (parent != NULL && !dt_object_remote(parent)) { + /* will help to find FID->ino at dt_insert("..") */ + struct osd_object *pobj = osd_dt_obj(parent); + osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj); + } } static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, @@ -2291,20 +2530,20 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, if (result) return; - if (attr->la_valid != 0) { - result = osd_inode_setattr(info->oti_env, inode, attr); - /* - * The osd_inode_setattr() should always succeed here. The - * only error that could be returned is EDQUOT when we are - * trying to change the UID or GID of the inode. However, this - * should not happen since quota enforcement is no longer - * enabled on ldiskfs (lquota takes care of it). - */ + if (attr->la_valid != 0) { + result = osd_inode_setattr(info->oti_env, inode, attr); + /* + * The osd_inode_setattr() should always succeed here. The + * only error that could be returned is EDQUOT when we are + * trying to change the UID or GID of the inode. However, this + * should not happen since quota enforcement is no longer + * enabled on ldiskfs (lquota takes care of it). + */ LASSERTF(result == 0, "%d\n", result); ll_dirty_inode(inode, I_DIRTY_DATASYNC); - } + } - attr->la_valid = valid; + attr->la_valid = valid; } /** @@ -2321,26 +2560,32 @@ 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) { - osd_attr_init(info, obj, attr, dof); - osd_object_init0(obj); - } - - if (obj->oo_inode != NULL) { + 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); + } + /* restore previous umask value */ current->fs->umask = umask; + osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE); + return result; } @@ -2356,14 +2601,19 @@ 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, @@ -2411,24 +2661,18 @@ 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); @@ -2437,6 +2681,10 @@ static int osd_declare_object_create(const struct lu_env *env, if (rc != 0) RETURN(rc); + /* will help to find FID->ino mapping at dt_insert() */ + rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)), + osd_dt_obj(dt)); + RETURN(rc); } @@ -2464,18 +2712,17 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt, * '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); } /** @@ -2493,9 +2740,11 @@ static int osd_declare_object_destroy(const struct lu_env *env, int rc; ENTRY; + if (inode == NULL) + RETURN(-ENOENT); + oh = container_of0(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - LASSERT(inode); osd_trans_declare_op(env, oh, OSD_OT_DESTROY, osd_dto_credits_noquota[DTO_OBJECT_DELETE]); @@ -2511,6 +2760,13 @@ static int osd_declare_object_destroy(const struct lu_env *env, /* data to be truncated */ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), 0, oh, obj, true, NULL, false); + if (rc) + RETURN(rc); + + /* will help to find FID->ino when this object is being + * added to PENDING/ */ + rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj); + RETURN(rc); } @@ -2552,16 +2808,19 @@ static int osd_object_destroy(const struct lu_env *env, osd_trans_exec_op(env, th, OSD_OT_DESTROY); + ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY); 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); } /** @@ -2584,6 +2843,9 @@ 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); @@ -2671,14 +2933,32 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, { struct ldiskfs_dentry_param *dot_ldp; struct ldiskfs_dentry_param *dot_dot_ldp; + __u32 saved_nlink = dir->i_nlink; + int rc; dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); 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); + + rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, + dir, dot_ldp, dot_dot_ldp); + /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then + * the subseqent ref_add() will increase the dir->i_nlink + * as 3. That is incorrect for new created directory. + * + * It looks like hack, because we want to make the OSD API + * to be order-independent for new created directory object + * between dt_insert(..) and ref_add() operations. + * + * Here, we only restore the in-RAM dir-inode's nlink attr, + * becuase if the nlink attr is not 2, then there will be + * ref_add() called following the dt_insert(..), such call + * will make both the in-RAM and on-disk dir-inode's nlink + * attr to be set as 2. LU-7447 */ + set_nlink(dir, saved_nlink); + return rc; } /** @@ -2735,33 +3015,86 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, } /** - * Delete local agent inode for remote entry + * when direntry is deleted, we have to take care of possible agent inode + * referenced by that. unfortunately we can't do this at that point: + * iget() within a running transaction leads to deadlock and we better do + * not call that every delete declaration to save performance. so we put + * a potention agent inode on a list and process that once the transaction + * is over. Notice it's not any worse in terms of real orphans as regular + * object destroy doesn't put inodes on the on-disk orphan list. this should + * be addressed separately */ -static int osd_delete_local_agent_inode(const struct lu_env *env, - struct osd_device *osd, - const struct lu_fid *fid, - __u32 ino, struct osd_thandle *oh) +static int osd_schedule_agent_inode_removal(const struct lu_env *env, + struct osd_thandle *oh, + __u32 ino) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_inode_id *id = &oti->oti_id; - struct inode *inode; - ENTRY; + struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); + struct osd_obj_orphan *oor; - id->oii_ino = le32_to_cpu(ino); - id->oii_gen = OSD_OII_NOGEN; - inode = osd_iget(oti, osd, id); - if (IS_ERR(inode)) { - CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd), - PFID(fid), id->oii_ino, id->oii_gen); - RETURN(PTR_ERR(inode)); + OBD_ALLOC_PTR(oor); + if (oor == NULL) + return -ENOMEM; + + oor->oor_ino = ino; + oor->oor_env = (struct lu_env *)env; + spin_lock(&osd->od_osfs_lock); + list_add(&oor->oor_list, &osd->od_orphan_list); + spin_unlock(&osd->od_osfs_lock); + + oh->ot_remove_agents = 1; + + return 0; + +} + +static int osd_process_scheduled_agent_removals(const struct lu_env *env, + struct osd_device *osd) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct osd_obj_orphan *oor, *tmp; + struct osd_inode_id id; + struct list_head list; + struct inode *inode; + struct lu_fid fid; + handle_t *jh; + __u32 ino; + + INIT_LIST_HEAD(&list); + + spin_lock(&osd->od_osfs_lock); + list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) { + if (oor->oor_env == env) { + list_del(&oor->oor_list); + list_add(&oor->oor_list, &list); + } } + spin_unlock(&osd->od_osfs_lock); - clear_nlink(inode); - mark_inode_dirty(inode); - CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n", - osd_name(osd), PFID(fid), inode->i_ino); - iput(inode); - RETURN(0); + list_for_each_entry_safe(oor, tmp, &list, oor_list) { + + ino = oor->oor_ino; + + list_del(&oor->oor_list); + OBD_FREE_PTR(oor); + + osd_id_gen(&id, ino, OSD_OII_NOGEN); + inode = osd_iget_fid(info, osd, &id, &fid); + if (IS_ERR(inode)) + continue; + + if (!osd_remote_fid(env, osd, &fid)) { + iput(inode); + continue; + } + + jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1); + clear_nlink(inode); + mark_inode_dirty(inode); + ldiskfs_journal_stop(jh); + iput(inode); + } + + return 0; } /** @@ -2781,7 +3114,7 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, 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; + int result, on_ost = 0; ENTRY; @@ -2798,9 +3131,6 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, * '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) && @@ -2810,19 +3140,32 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, fid_to_ostid(fid, oi); ostid_to_fid(tfid, oi, 0); + on_ost = 1; result = osd_ea_fid_set(info, obj->oo_inode, tfid, LMAC_FID_ON_OST, 0); } else { + on_ost = fid_is_on_ost(info, osd_obj2dev(obj), + fid, OI_CHECK_FLD); 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); + on_ost ? 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); + /* a small optimization - dt_insert() isn't usually applied + * to OST objects, so we don't need to cache OI mapping for + * OST objects */ + if (result == 0 && on_ost == 0) { + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + result = osd_idc_find_and_init(env, osd, obj); + LASSERT(result == 0); + } + LASSERT(ergo(result == 0, dt_object_exists(dt) && !dt_object_remote(dt))); LINVRNT(osd_invariant(obj)); @@ -2858,7 +3201,7 @@ static int osd_object_ref_add(const struct lu_env *env, struct osd_thandle *oh; int rc = 0; - if (!dt_object_exists(dt)) + if (!dt_object_exists(dt) || obj->oo_destroyed) return -ENOENT; LINVRNT(osd_invariant(obj)); @@ -2898,6 +3241,8 @@ static int osd_object_ref_add(const struct lu_env *env, ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); + osd_trans_exec_check(env, th, OSD_OT_REF_ADD); + return rc; } @@ -2907,6 +3252,9 @@ static int osd_declare_object_ref_del(const struct lu_env *env, { struct osd_thandle *oh; + if (!dt_object_exists(dt)) + return -ENOENT; + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); @@ -2965,6 +3313,8 @@ 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; } @@ -2986,8 +3336,7 @@ 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; @@ -3016,9 +3365,6 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, LASSERT(inode->i_op != NULL); LASSERT(inode->i_op->getxattr != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; - return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -3037,25 +3383,50 @@ static int osd_declare_xattr_set(const struct lu_env *env, 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 { +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; + + if (buf != NULL) { + ssize_t buflen; + + if (buf->lb_buf == NULL && dt_object_exists(dt)) { + /* learn xattr size from osd_xattr_get if + attribute has not been read yet */ + buflen = __osd_xattr_get( + osd_dt_obj(dt)->oo_inode, + &osd_oti_get(env)->oti_obj_dentry, + name, NULL, 0); + if (buflen < 0) + buflen = 0; + } else { + buflen = buf->lb_len; + } + + if (buflen > sb->s_blocksize) { + credits += osd_calc_bkmap_credits( + sb, NULL, 0, -1, + (buflen + sb->s_blocksize - 1) >> + sb->s_blocksize_bits); + } } /* * xattr set may involve inode quota change, reserve credits for * dquot_initialize() */ - oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); + credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); } osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); @@ -3085,13 +3456,14 @@ 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); @@ -3105,9 +3477,6 @@ 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 %zu\n", PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); @@ -3138,15 +3507,18 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, strcmp(name, XATTR_NAME_LINK) == 0) return -ENOSPC; - return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, + 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, - const 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; @@ -3160,9 +3532,6 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, LASSERT(inode->i_op != NULL); LASSERT(inode->i_op->listxattr != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; - dentry->d_inode = inode; dentry->d_sb = inode->i_sb; return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); @@ -3185,133 +3554,41 @@ static int osd_declare_xattr_del(const struct lu_env *env, 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; -} - -/* - * 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) -{ - 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; - - if (!dt_object_exists(dt)) - return -ENOENT; - - LASSERT(!dt_object_remote(dt)); - LASSERT(inode->i_op != NULL); - LASSERT(inode->i_op->removexattr != NULL); - LASSERT(handle != NULL); - - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; - - osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); - - ll_vfs_dq_init(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)); - - if (!dt_object_exists(dt)) - RETURN(ERR_PTR(-ENOENT)); - - LASSERT(!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)); - } + * dquot_initialize() + */ + oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); - 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; + return 0; +} - oc = capa_lookup(osd->od_capa_hash, capa, 1); - if (oc) { - LASSERT(!capa_is_expired(oc)); - RETURN(oc); - } +/* + * 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 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; - spin_lock(&capa_lock); - *key = osd->od_capa_keys[1]; - spin_unlock(&capa_lock); + if (!dt_object_exists(dt)) + return -ENOENT; - capa->lc_keyid = key->lk_keyid; - capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->removexattr != NULL); + LASSERT(handle != NULL); - 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)); - } + osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); - oc = capa_add(osd->od_capa_hash, capa); - RETURN(oc); + ll_vfs_dq_init(inode); + dentry->d_inode = inode; + dentry->d_sb = inode->i_sb; + 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, @@ -3328,7 +3605,7 @@ 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); @@ -3338,16 +3615,6 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, 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. */ @@ -3420,7 +3687,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, result = 0; } else if (feat == &dt_directory_features) { dt->do_index_ops = &osd_index_ea_ops; - if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode)) + if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode)) result = 0; else result = -ENOTDIR; @@ -3474,17 +3741,16 @@ 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; @@ -3515,9 +3781,7 @@ 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, }; /** @@ -3549,9 +3813,7 @@ 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 = { @@ -3569,8 +3831,9 @@ static int osd_index_declare_iam_delete(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); + /* Recycle may cause additional three blocks to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_DELETE, - osd_dto_credits_noquota[DTO_INDEX_DELETE]); + osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); return 0; } @@ -3588,8 +3851,7 @@ static int osd_index_declare_iam_delete(const struct lu_env *env, */ 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 thandle *handle) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_object *obj = osd_dt_obj(dt); @@ -3607,9 +3869,6 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, LASSERT(bag->ic_object == obj->oo_inode); LASSERT(handle != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); - osd_trans_exec_op(env, handle, OSD_OT_DELETE); ipd = osd_idx_ipd_get(env, bag); @@ -3629,6 +3888,7 @@ 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); } @@ -3648,11 +3908,14 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, 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); + if (inode == NULL) + RETURN(-ENOENT); rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), 0, oh, osd_dt_obj(dt), true, NULL, false); @@ -3668,6 +3931,8 @@ 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; } @@ -3706,8 +3971,7 @@ 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) + const struct dt_key *key, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct inode *dir = obj->oo_inode; @@ -3734,9 +3998,6 @@ 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)); @@ -3751,8 +4012,6 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (bh) { - __u32 ino = 0; - /* If this is not the ".." entry, it might be a remote DNE * entry and we need to check if the FID is for a remote * MDT. If the FID is not in the directory entry (e.g. @@ -3770,56 +4029,19 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, if (strcmp((char *)key, dotdot) != 0) { LASSERT(de != NULL); rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); - /* If Fid is not in dentry, try to get it from LMA */ if (rc == -ENODATA) { - struct osd_inode_id *id; - struct inode *inode; - - /* Before trying to get fid from the inode, - * check whether the inode is valid. - * - * If the inode has been deleted, do not go - * ahead to do osd_ea_fid_get, which will set - * the inode to bad inode, which might cause - * the inode to be deleted uncorrectly */ - inode = ldiskfs_iget(osd_sb(osd), - le32_to_cpu(de->inode)); - if (IS_ERR(inode)) { - CDEBUG(D_INODE, "%s: "DFID"get inode" - "error.\n", osd_name(osd), - PFID(fid)); - rc = PTR_ERR(inode); - } else { - if (likely(inode->i_nlink != 0)) { - id = &osd_oti_get(env)->oti_id; - rc = osd_ea_fid_get(env, obj, - le32_to_cpu(de->inode), - fid, id); - } else { - CDEBUG(D_INFO, "%s: %u "DFID - "deleted.\n", - osd_name(osd), - le32_to_cpu(de->inode), - PFID(fid)); - rc = -ESTALE; - } - iput(inode); - } + /* can't get FID, postpone to the end of the + * transaction when iget() is safe */ + osd_schedule_agent_inode_removal(env, oh, + le32_to_cpu(de->inode)); + } else if (rc == 0 && + unlikely(osd_remote_fid(env, osd, fid))) { + osd_schedule_agent_inode_removal(env, oh, + le32_to_cpu(de->inode)); } - if (rc == 0 && - unlikely(osd_remote_fid(env, osd, fid))) - /* Need to delete agent inode */ - ino = le32_to_cpu(de->inode); } rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); brelse(bh); - if (rc == 0 && unlikely(ino != 0)) { - rc = osd_delete_local_agent_inode(env, osd, fid, ino, - oh); - if (rc != 0) - CERROR("%s: del local inode "DFID": rc = %d\n", - osd_name(osd), PFID(fid), rc); - } } else { rc = -ENOENT; } @@ -3850,6 +4072,7 @@ 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); } @@ -3865,8 +4088,7 @@ 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; @@ -3884,9 +4106,6 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) - RETURN(-EACCES); - ipd = osd_idx_ipd_get(env, bag); if (IS_ERR(ipd)) RETURN(-ENOMEM); @@ -3958,7 +4177,7 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, 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) + int ignore_quota) { struct osd_object *obj = osd_dt_obj(dt); struct iam_path_descr *ipd; @@ -3979,9 +4198,6 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, LASSERT(bag->ic_object == obj->oo_inode); 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); ipd = osd_idx_ipd_get(env, bag); @@ -4009,6 +4225,7 @@ 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); } @@ -4044,7 +4261,8 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); child->d_fsdata = (void *)ldp; ll_vfs_dq_init(pobj->oo_inode); - rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); + rc = osd_ldiskfs_add_entry(info, 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; @@ -4103,7 +4321,7 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, if (dir->oo_compat_dot_created) { result = -EEXIST; } else { - LASSERT(inode == parent_dir); + LASSERT(inode->i_ino == parent_dir->i_ino); dir->oo_compat_dot_created = 1; result = 0; } @@ -4191,7 +4409,7 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, return rc; } -static void +static int osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, struct osd_idmap_cache *oic) { @@ -4203,19 +4421,39 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, ENTRY; if (!fid_is_norm(fid) && !fid_is_igif(fid)) - RETURN_EXIT; + RETURN(0); if (scrub->os_pos_current > id->oii_ino) - RETURN_EXIT; + RETURN(0); again: - rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD); - if (rc != 0 && rc != -ENOENT) - RETURN_EXIT; + rc = osd_oi_lookup(oti, dev, fid, id, 0); + if (rc == -ENOENT) { + struct inode *inode; + + *id = oic->oic_lid; + inode = osd_iget(oti, dev, &oic->oic_lid); + + /* The inode has been removed (by race maybe). */ + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + + RETURN(rc == -ESTALE ? -ENOENT : rc); + } - if (rc == 0 && osd_id_eq(id, &oic->oic_lid)) - RETURN_EXIT; + iput(inode); + /* The OI mapping is lost. */ + if (id->oii_gen != OSD_OII_NOGEN) + goto trigger; + + /* The inode may has been reused by others, we do not know, + * leave it to be handled by subsequent osd_fid_lookup(). */ + RETURN(0); + } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) { + RETURN(rc); + } +trigger: if (thread_is_running(&scrub->os_thread)) { rc = osd_oii_insert(dev, oic, rc == -ENOENT); /* There is race condition between osd_oi_lookup and OI scrub. @@ -4225,7 +4463,7 @@ again: if (unlikely(rc == -EAGAIN)) goto again; - RETURN_EXIT; + RETURN(0); } if (!dev->od_noscrub && ++once == 1) { @@ -4239,7 +4477,7 @@ again: goto again; } - EXIT; + RETURN(0); } static int osd_fail_fid_lookup(struct osd_thread_info *oti, @@ -4425,16 +4663,16 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, osd_id_gen(id, ino, OSD_OII_NOGEN); } - if (rc != 0) { + if (rc != 0 || osd_remote_fid(env, dev, fid)) { fid_zero(&oic->oic_fid); + GOTO(out, rc); } - if (osd_remote_fid(env, dev, fid)) - GOTO(out, rc = 0); - osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid); - osd_consistency_check(oti, dev, oic); + rc = osd_consistency_check(oti, dev, oic); + if (rc != 0) + fid_zero(&oic->oic_fid); } else { rc = -ENOENT; } @@ -4450,65 +4688,6 @@ out: } /** - * Find the osd object for given fid. - * - * \param fid need to find the osd object having this fid - * - * \retval osd_object on success - * \retval -ve on error - */ -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; - struct lu_object *luch; - struct lu_object *lo; - - /* - * at this point topdev might not exist yet - * (i.e. MGS is preparing profiles). so we can - * not rely on topdev and instead lookup with - * our device passed as topdev. this can't work - * if the object isn't cached yet (as osd doesn't - * allocate lu_header). IOW, the object must be - * in the cache, otherwise lu_object_alloc() crashes - * -bzzz - */ - luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ? - ludev : ludev->ld_site->ls_top_dev, - fid, NULL); - if (!IS_ERR(luch)) { - if (lu_object_exists(luch)) { - lo = lu_object_locate(luch->lo_header, ludev->ld_type); - if (lo != NULL) - child = osd_obj(lo); - else - LU_OBJECT_DEBUG(D_ERROR, env, luch, - "lu_object can't be located" - DFID"\n", PFID(fid)); - - if (child == NULL) { - lu_object_put(env, luch); - CERROR("Unable to get osd_object\n"); - child = ERR_PTR(-ENOENT); - } - } else { - LU_OBJECT_DEBUG(D_ERROR, env, luch, - "lu_object does not exists "DFID"\n", - PFID(fid)); - lu_object_put(env, luch); - child = ERR_PTR(-ENOENT); - } - } else { - child = ERR_CAST(luch); - } - - return child; -} - -/** * Put the osd object once done with it. * * \param obj osd object that needs to be put @@ -4527,18 +4706,37 @@ 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; + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; + int credits, rc = 0; + struct osd_idmap_cache *idc; ENTRY; LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); + LASSERT(fid != NULL); + LASSERT(rec1->rec_type != 0); 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]; + + /* we can't call iget() while a transactions is running + * (this can lead to a deadlock), but we need to know + * inum and object type. so we find this information at + * declaration and cache in per-thread info */ + idc = osd_idc_find_or_init(env, osd, fid); + if (IS_ERR(idc)) + RETURN(PTR_ERR(idc)); + if (idc->oic_remote) { + /* 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]; + } + + 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; @@ -4551,22 +4749,6 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, 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); } @@ -4582,9 +4764,9 @@ 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); @@ -4592,9 +4774,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, 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; struct inode *child_inode = NULL; - struct osd_object *child = NULL; + struct osd_idmap_cache *idc; int rc; ENTRY; @@ -4607,19 +4788,24 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_INSERT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); - LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid)); - rc = osd_remote_fid(env, osd, fid); - if (rc < 0) { - CERROR("%s: Can not find object "DFID" rc %d\n", - osd_name(osd), PFID(fid), rc); - RETURN(rc); + idc = osd_idc_find(env, osd, fid); + if (unlikely(idc == NULL)) { + /* this dt_insert() wasn't declared properly, so + * FID is missing in OI cache. we better do not + * lookup FID in FLDB/OI and don't risk to deadlock, + * but in some special cases (lfsck testing, etc) + * it's much simpler than fixing a caller */ + CERROR("%s: "DFID" wasn't declared for insert\n", + osd_name(osd), PFID(fid)); + dump_stack(); + idc = osd_idc_find_or_init(env, osd, fid); + if (IS_ERR(idc)) + RETURN(PTR_ERR(idc)); } - if (rc == 1) { + if (idc->oic_remote) { /* Insert remote entry */ if (strcmp(name, dotdot) == 0 && strlen(name) == 2) { struct osd_mdobj_map *omm = osd->od_mdt_map; @@ -4645,15 +4831,23 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, } } else { /* Insert local entry */ - child = osd_object_find(env, dt, fid); - if (IS_ERR(child)) { - CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n", - osd_name(osd), PFID(fid), - id->oii_ino, id->oii_gen, - (int)PTR_ERR(child)); - RETURN(PTR_ERR(child)); + if (unlikely(idc->oic_lid.oii_ino == 0)) { + /* for a reason OI cache wasn't filled properly */ + CERROR("%s: OIC for "DFID" isn't filled\n", + osd_name(osd), PFID(fid)); + RETURN(-EINVAL); } - child_inode = igrab(child->oo_inode); + child_inode = oti->oti_inode; + if (unlikely(child_inode == NULL)) { + struct ldiskfs_inode_info *lii; + OBD_ALLOC_PTR(lii); + if (lii == NULL) + RETURN(-ENOMEM); + child_inode = oti->oti_inode = &lii->vfs_inode; + } + child_inode->i_sb = osd_sb(osd); + child_inode->i_ino = idc->oic_lid.oii_ino; + child_inode->i_mode = rec1->rec_type & S_IFMT; } rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th); @@ -4661,10 +4855,10 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n", obj->oo_inode->i_ino, name, child_inode->i_ino, rc); - iput(child_inode); - if (child != NULL) - osd_object_put(env, child); + if (child_inode && child_inode != oti->oti_inode) + iput(child_inode); LASSERT(osd_invariant(obj)); + osd_trans_exec_check(env, th, OSD_OT_INSERT); RETURN(rc); } @@ -4675,9 +4869,8 @@ 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_object *obj = osd_dt_obj(dt); @@ -4688,9 +4881,6 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, if (!dt_object_exists(dt)) return ERR_PTR(-ENOENT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ)) - return ERR_PTR(-EACCES); - OBD_ALLOC_PTR(it); if (it == NULL) return ERR_PTR(-ENOMEM); @@ -4959,8 +5149,7 @@ 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) + __u32 attr) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *info = osd_oti_get(env); @@ -4970,7 +5159,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, struct dentry *obj_dentry; ENTRY; - if (!dt_object_exists(dt)) + if (!dt_object_exists(dt) || obj->oo_destroyed) RETURN(ERR_PTR(-ENOENT)); OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS); @@ -5002,8 +5191,8 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, 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); @@ -5083,11 +5272,17 @@ struct osd_filldir_cbs { * \retval 0 on success * \retval 1 on buffer full */ +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ldiskfs_filldir(struct dir_context *buf, + const char *name, int namelen, +#else static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, +#endif loff_t offset, __u64 ino, unsigned d_type) { - struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it; + struct osd_it_ea *it = + ((struct osd_filldir_cbs *)buf)->it; struct osd_object *obj = it->oie_obj; struct osd_it_ea_dirent *ent = it->oie_dirent; struct lu_fid *fid = &ent->oied_fid; @@ -5263,63 +5458,46 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) return it->oie_dirent->oied_namelen; } -static int -osd_dirent_update(handle_t *jh, struct super_block *sb, - struct osd_it_ea_dirent *ent, struct lu_fid *fid, - struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de) +static inline bool +osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot) { - struct osd_fid_pack *rec; - int rc; - ENTRY; - - LASSERT(de->file_type & LDISKFS_DIRENT_LUFID); - LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack)); - - rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) - RETURN(rc); + LASSERTF(dot_dotdot == 1 || dot_dotdot == 2, + "dot_dotdot = %d\n", dot_dotdot); - rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); - fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); - rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); + if (LDISKFS_DIR_REC_LEN(de) >= + __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack))) + return true; - RETURN(rc); + return false; } -static inline int -osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize) +static inline bool +osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen, + unsigned blocksize, int dot_dotdot) { - if (ldiskfs_rec_len_from_disk(reclen, blocksize) >= - __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack))) - return 1; - else - return 0; -} + if (dot_dotdot > 0) + return osd_dot_dotdot_has_space(de, dot_dotdot); -static inline int -osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot) -{ - LASSERTF(dot_dotdot == 1 || dot_dotdot == 2, - "dot_dotdot = %d\n", dot_dotdot); + if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >= + __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack))) + return true; - if (LDISKFS_DIR_REC_LEN(de) >= - __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack))) - return 1; - else - return 0; + return false; } static int osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, - struct inode *dir, struct inode *inode, - struct osd_it_ea_dirent *ent, struct lu_fid *fid, + struct dentry *dentry, const struct lu_fid *fid, struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de, - struct htree_lock *hlock) + struct htree_lock *hlock, int dot_dotdot) { - struct dentry *dentry; + struct inode *dir = dentry->d_parent->d_inode; + struct inode *inode = dentry->d_inode; struct osd_fid_pack *rec; struct ldiskfs_dentry_param *ldp; + int namelen = dentry->d_name.len; int rc; + struct osd_thread_info *info = osd_oti_get(env); ENTRY; if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb, @@ -5327,42 +5505,41 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, RETURN(0); /* There is enough space to hold the FID-in-dirent. */ - if (osd_dirent_has_space(de->rec_len, ent->oied_namelen, - dir->i_sb->s_blocksize)) { + if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, + dot_dotdot)) { rc = ldiskfs_journal_get_write_access(jh, bh); if (rc != 0) RETURN(rc); - de->name[de->name_len] = 0; - rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + de->name[namelen] = 0; + rec = (struct osd_fid_pack *)(de->name + namelen + 1); rec->fp_len = sizeof(struct lu_fid) + 1; fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); de->file_type |= LDISKFS_DIRENT_LUFID; - rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); RETURN(rc); } + LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot); + rc = ldiskfs_delete_entry(jh, dir, de, bh); 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, fid); dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); - rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); + rc = osd_ldiskfs_add_entry(info, jh, dentry, inode, hlock); /* It is too bad, we cannot reinsert the name entry back. * That means we lose it! */ if (rc != 0) 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, - dir->i_ino, dir->i_generation, - ent->oied_namelen, ent->oied_name, PFID(fid), rc); + dir->i_ino, dir->i_generation, namelen, + dentry->d_name.name, PFID(fid), rc); RETURN(rc); } @@ -5392,6 +5569,32 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, bool dirty = false; ENTRY; + osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN); + inode = osd_iget(info, dev, id); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + if (rc == -ENOENT || rc == -ESTALE) { + *attr |= LUDA_UNKNOWN; + rc = 0; + } 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); + } + + RETURN(rc); + } + + dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name, + ent->oied_namelen); + rc = osd_get_lma(info, inode, dentry, lma); + if (rc == -ENODATA) + lma = NULL; + else if (rc != 0) + GOTO(out, rc); + if (ent->oied_name[0] == '.') { if (ent->oied_namelen == 1) dot_dotdot = 1; @@ -5399,9 +5602,6 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, dot_dotdot = 2; } - dentry = osd_child_dentry_get(env, obj, ent->oied_name, - ent->oied_namelen); - /* We need to ensure that the name entry is still valid. * Because it may be removed or renamed by other already. * @@ -5418,8 +5618,9 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; -again: if (dev->od_dirent_journal != 0) { + +again: jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); @@ -5428,7 +5629,8 @@ again: "name = %.*s: rc = %d\n", devname, dir->i_ino, dir->i_generation, credits, ent->oied_namelen, ent->oied_name, rc); - RETURN(rc); + + GOTO(out_inode, rc); } if (obj->oo_hl_head != NULL) { @@ -5458,58 +5660,66 @@ again: * For the whole directory, only dot/dotdot entry have no FID-in-dirent * and needs to get FID from LMA when readdir, it will not affect the * performance much. */ - if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) || + if ((bh == NULL) || (le32_to_cpu(de->inode) != inode->i_ino) || (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) { *attr |= LUDA_IGNORE; - GOTO(out_journal, rc = 0); - } - - osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN); - inode = osd_iget(info, dev, id); - if (IS_ERR(inode)) { - rc = PTR_ERR(inode); - 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); + GOTO(out, rc = 0); } - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc == 0) { - LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI)); + if (lma != NULL) { + if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) { + struct lu_fid *tfid = &lma->lma_self_fid; + + *attr |= LUDA_IGNORE; + /* It must be REMOTE_PARENT_DIR and as the + * dotdot entry of remote directory */ + if (unlikely(dot_dotdot != 2 || + fid_seq(tfid) != FID_SEQ_LOCAL_FILE || + fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) { + CDEBUG(D_LFSCK, "%.16s: expect remote agent " + "parent directory, but got %.*s under " + "dir = %lu/%u with the FID "DFID"\n", + devname, ent->oied_namelen, + ent->oied_name, dir->i_ino, + dir->i_generation, PFID(tfid)); + + GOTO(out, rc = -EIO); + } + + GOTO(out, rc = 0); + } if (fid_is_sane(fid)) { /* FID-in-dirent is valid. */ if (lu_fid_eq(fid, &lma->lma_self_fid)) - GOTO(out_inode, rc = 0); + GOTO(out, rc = 0); /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { *attr |= LUDA_REPAIR; - GOTO(out_inode, rc = 0); + + GOTO(out, rc = 0); } - if (dev->od_dirent_journal == 0) { - iput(inode); + if (jh == NULL) { brelse(bh); - if (hlock != NULL) + dev->od_dirent_journal = 1; + if (hlock != NULL) { ldiskfs_htree_unlock(hlock); - else + hlock = NULL; + } else { up_read(&obj->oo_ext_idx_sem); - dev->od_dirent_journal = 1; + } + goto again; } *fid = lma->lma_self_fid; dirty = true; /* Update the FID-in-dirent. */ - rc = osd_dirent_update(jh, sb, ent, fid, bh, de); + rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de, + hlock, dot_dotdot); if (rc == 0) *attr |= LUDA_REPAIR; else @@ -5524,25 +5734,28 @@ again: if (*attr & LUDA_VERIFY_DRYRUN) { *fid = lma->lma_self_fid; *attr |= LUDA_REPAIR; - GOTO(out_inode, rc = 0); + + GOTO(out, rc = 0); } - if (dev->od_dirent_journal == 0) { - iput(inode); + if (jh == NULL) { brelse(bh); - if (hlock != NULL) + dev->od_dirent_journal = 1; + if (hlock != NULL) { ldiskfs_htree_unlock(hlock); - else + hlock = NULL; + } else { up_read(&obj->oo_ext_idx_sem); - dev->od_dirent_journal = 1; + } + goto again; } *fid = lma->lma_self_fid; dirty = true; /* Append the FID-in-dirent. */ - rc = osd_dirent_reinsert(env, jh, dir, inode, ent, - fid, bh, de, hlock); + rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de, + hlock, dot_dotdot); if (rc == 0) *attr |= LUDA_REPAIR; else @@ -5553,7 +5766,7 @@ again: ent->oied_namelen, ent->oied_name, PFID(fid), rc); } - } else if (rc == -ENODATA) { + } else { /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { if (fid_is_sane(fid)) { @@ -5563,17 +5776,20 @@ again: inode->i_generation); *attr |= LUDA_UPGRADE; } - GOTO(out_inode, rc = 0); + + GOTO(out, rc = 0); } - if (dev->od_dirent_journal == 0) { - iput(inode); + if (jh == NULL) { brelse(bh); - if (hlock != NULL) + dev->od_dirent_journal = 1; + if (hlock != NULL) { ldiskfs_htree_unlock(hlock); - else + hlock = NULL; + } else { up_read(&obj->oo_ext_idx_sem); - dev->od_dirent_journal = 1; + } + goto again; } @@ -5595,8 +5811,8 @@ again: lu_igif_build(fid, inode->i_ino, inode->i_generation); /* It is probably IGIF object. Only aappend the * FID-in-dirent. OI scrub will process FID-in-LMA. */ - rc = osd_dirent_reinsert(env, jh, dir, inode, ent, - fid, bh, de, hlock); + rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de, + hlock, dot_dotdot); if (rc == 0) *attr |= LUDA_UPGRADE; else @@ -5609,12 +5825,9 @@ again: } } - GOTO(out_inode, rc); - -out_inode: - iput(inode); + GOTO(out, rc); -out_journal: +out: brelse(bh); if (hlock != NULL) { ldiskfs_htree_unlock(hlock); @@ -5624,10 +5837,15 @@ out_journal: else up_read(&obj->oo_ext_idx_sem); } + if (jh != NULL) ldiskfs_journal_stop(jh); + +out_inode: + iput(inode); if (rc >= 0 && !dirty) dev->od_dirent_journal = 0; + return rc; } @@ -5673,20 +5891,27 @@ static inline int osd_it_ea_rec(const struct lu_env *env, rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); } + + if (!fid_is_sane(fid)) { + attr &= ~LUDA_IGNORE; + attr |= LUDA_UNKNOWN; + } } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { + bool is_dotdot = false; if (it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && - it->oie_dirent->oied_name[1] == '.') { - /* 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)) - rc = osd_get_pfid_from_linkea(env, obj, - fid); + 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 (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + if (is_dotdot == false && + OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) RETURN(-ENOENT); rc = osd_ea_fid_get(env, obj, ino, fid, id); @@ -5708,7 +5933,7 @@ 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 == 0)) + if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0)) osd_add_oi_cache(oti, dev, id, fid); RETURN(rc > 0 ? 0 : rc); @@ -5788,8 +6013,7 @@ 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; @@ -5799,9 +6023,6 @@ 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; @@ -5865,15 +6086,23 @@ static void osd_key_fini(const struct lu_context *ctx, struct lu_context_key *key, void* data) { struct osd_thread_info *info = data; + struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode); + struct osd_idmap_cache *idc = info->oti_ins_cache; if (info->oti_inode != NULL) - OBD_FREE_PTR(info->oti_inode); + OBD_FREE_PTR(lli); 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); + if (idc != NULL) { + LASSERT(info->oti_ins_cache_size > 0); + OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size); + info->oti_ins_cache = NULL; + info->oti_ins_cache_size = 0; + } OBD_FREE_PTR(info); } @@ -5927,10 +6156,18 @@ 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); @@ -5990,7 +6227,7 @@ 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, force_over_128tb = 0; + int rc = 0, force_over_256tb = 0; ENTRY; if (o->od_mnt != NULL) @@ -6014,10 +6251,17 @@ static int osd_mount(const struct lu_env *env, RETURN(-EINVAL); } #endif - if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) - force_over_128tb = 1; + if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) { + CWARN("force_over_128tb option is depricated." + "Filesystems less then 256TB can be created without any" + "force options. Use force_over_256tb option for" + "filesystems greather then 256TB.\n"); + } + + if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) + force_over_256tb = 1; - OBD_PAGE_ALLOC(__page, GFP_IOFS); + __page = alloc_page(GFP_IOFS); if (__page == NULL) GOTO(out, rc = -ENOMEM); page = (unsigned long)page_address(__page); @@ -6034,7 +6278,12 @@ static int osd_mount(const struct lu_env *env, "noextents", /* strip out option we processed in osd */ "bigendian_extents", - "force_over_128tb", +#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(3,0,53,0) +#warning "remove force_over_128 option" +#else + "force_over_128tb (deprecated)", +#endif + "force_over_256tb", NULL }; strcat(options, opts); @@ -6080,11 +6329,11 @@ 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) { + if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (64ULL << 30) && + force_over_256tb == 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", + "greater than 256TB and can cause data corruption. " + "Use \"force_over_256tb\" mount option to override.\n", name, dev); GOTO(out, rc = -EINVAL); } @@ -6132,7 +6381,7 @@ out_mnt: out: if (__page) - OBD_PAGE_FREE(__page); + __free_page(__page); return rc; } @@ -6173,10 +6422,7 @@ 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); + INIT_LIST_HEAD(&o->od_orphan_list); o->od_read_cache = 1; o->od_writethrough_cache = 1; @@ -6186,17 +6432,19 @@ 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) @@ -6248,8 +6496,6 @@ 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; } @@ -6287,7 +6533,6 @@ 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); @@ -6500,10 +6745,11 @@ static struct obd_ops osd_obd_device_ops = { .o_health_check = osd_health_check, }; -static int __init osd_mod_init(void) +static int __init osd_init(void) { int rc; + LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8); #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); @@ -6523,14 +6769,16 @@ static int __init osd_mod_init(void) return rc; } -static void __exit osd_mod_exit(void) +static void __exit osd_exit(void) { class_unregister_type(LUSTRE_OSD_LDISKFS_NAME); lu_kmem_fini(ldiskfs_caches); } -MODULE_AUTHOR("Sun Microsystems, Inc. "); +MODULE_AUTHOR("OpenSFS, 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_init); +module_exit(osd_exit);