X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=e4412fb66bf0150bbe72c403729f3a5c6967cf34;hp=1b3c323ffa46fc2a2af29b5dff23fb840b484284;hb=b69b7de30c3977cb69a741099218bc4a81752717;hpb=c52d5f2913c33eb7efe0d8a9073c90091e169b8c diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 1b3c323..e4412fb 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, 2012, Intel Corporation. + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,7 +41,7 @@ * Pravin Shelar : Added fid in dirent */ -#define DEBUG_SUBSYSTEM S_MDS +#define DEBUG_SUBSYSTEM S_OSD #include @@ -53,8 +53,6 @@ #include /* XATTR_{REPLACE,CREATE} */ #include -/* simple_mkdir() */ -#include /* * struct OBD_{ALLOC,FREE}*() @@ -64,17 +62,40 @@ /* struct ptlrpc_thread */ #include #include +/* process_config */ +#include #include "osd_internal.h" +#include "osd_dynlocks.h" /* llo_* api support */ #include #include +#include + int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, "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"); + +/* Slab to allocate dynlocks */ +struct kmem_cache *dynlock_cachep; + +static struct lu_kmem_descr ldiskfs_caches[] = { + { + .ckd_cache = &dynlock_cachep, + .ckd_name = "dynlock_cache", + .ckd_size = sizeof(struct dynlock_handle) + }, + { + .ckd_cache = NULL + } +}; + static const char dot[] = "."; static const char dotdot[] = ".."; static const char remote_obj_dir[] = "REM_OBJ_DIR"; @@ -86,7 +107,6 @@ 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; -#ifdef OSD_TRACK_DECLARES int osd_trans_declare_op2rb[] = { [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET, [OSD_OT_PUNCH] = OSD_OT_MAX, @@ -98,9 +118,9 @@ int osd_trans_declare_op2rb[] = { [OSD_OT_WRITE] = OSD_OT_WRITE, [OSD_OT_INSERT] = OSD_OT_DELETE, [OSD_OT_DELETE] = OSD_OT_INSERT, + [OSD_OT_UPDATE] = OSD_OT_MAX, [OSD_OT_QUOTA] = OSD_OT_MAX, }; -#endif static int osd_has_index(const struct osd_object *obj) { @@ -115,10 +135,6 @@ static int osd_object_invariant(const struct lu_object *l) /* * Concurrency: doesn't matter */ -static int osd_read_locked(const struct lu_env *env, struct osd_object *o) -{ - return osd_oti_get(env)->oti_r_locks > 0; -} /* * Concurrency: doesn't matter @@ -170,40 +186,27 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, } } -static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry, - const char *name, void *buf, int len) -{ - dentry->d_inode = inode; - return inode->i_op->getxattr(dentry, name, buf, len); -} - int osd_get_lma(struct osd_thread_info *info, struct inode *inode, struct dentry *dentry, struct lustre_mdt_attrs *lma) { int rc; - rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma, - sizeof(*lma)); - if (rc == -ERANGE) { - /* try with old lma size */ - rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, - info->oti_mdt_attrs_old, - LMA_OLD_SIZE); - if (rc > 0) - memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma)); - } + CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, + info->oti_mdt_attrs_old, LMA_OLD_SIZE); if (rc > 0) { + if ((void *)lma != (void *)info->oti_mdt_attrs_old) + memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma)); + rc = 0; + lustre_lma_swab(lma); /* Check LMA compatibility */ - if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) { - CWARN("%.16s: unsupported incompat LMA feature(s) " - "%lx/%#x\n", + if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) { + CWARN("%.16s: unsupported incompat LMA feature(s) %#x " + "for fid = "DFID", ino = %lu\n", LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - inode->i_ino, le32_to_cpu(lma->lma_incompat) & - ~LMA_INCOMPAT_SUPP); - rc = -ENOSYS; - } else { - lustre_lma_swab(lma); - rc = 0; + lma->lma_incompat & ~LMA_INCOMPAT_SUPP, + PFID(&lma->lma_self_fid), inode->i_ino); + rc = -EOPNOTSUPP; } } else if (rc == 0) { rc = -ENODATA; @@ -226,8 +229,8 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, id->oii_ino, PTR_ERR(inode)); } else if (id->oii_gen != OSD_OII_NOGEN && inode->i_generation != id->oii_gen) { - CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, " - "gen1 = %u\n", + CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, " + "i_generation = %u\n", id->oii_ino, id->oii_gen, inode->i_generation); iput(inode); inode = ERR_PTR(-ESTALE); @@ -235,7 +238,6 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, /* due to parallel readdir and unlink, * we can have dead inode here. */ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); - make_bad_inode(inode); iput(inode); inode = ERR_PTR(-ESTALE); } else if (is_bad_inode(inode)) { @@ -285,37 +287,257 @@ osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, return inode; } -static struct inode * -osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev, - struct osd_inode_id *id, const struct lu_fid *fid) +static struct inode *osd_iget_check(struct osd_thread_info *info, + struct osd_device *dev, + const struct lu_fid *fid, + struct osd_inode_id *id, + bool in_oi) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct inode *inode; - int rc; + struct inode *inode; + int rc = 0; + ENTRY; - inode = osd_iget(info, dev, id); - if (IS_ERR(inode)) - return inode; + inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) { + CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n", + id->oii_ino, rc); - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc == -ENODATA) - return inode; + GOTO(put, rc); + } + + goto check_oi; + } + + if (is_bad_inode(inode)) { + rc = -ENOENT; + if (!in_oi) { + CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino); + + GOTO(put, rc); + } + + goto check_oi; + } + + if (id->oii_gen != OSD_OII_NOGEN && + inode->i_generation != id->oii_gen) { + rc = -ESTALE; + if (!in_oi) { + CDEBUG(D_INODE, "unmatched inode: ino = %u, " + "oii_gen = %u, i_generation = %u\n", + id->oii_ino, id->oii_gen, inode->i_generation); + GOTO(put, rc); + } + + goto check_oi; + } + + if (inode->i_nlink == 0) { + rc = -ENOENT; + if (!in_oi) { + CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); + + GOTO(put, rc); + } + + goto check_oi; + } + +check_oi: if (rc != 0) { - iput(inode); - return ERR_PTR(rc); + LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); + + rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + /* XXX: There are some possible cases: + * 1. rc = 0. + * Backup/restore caused the OI invalid. + * 2. rc = 0. + * Someone unlinked the object but NOT removed + * the OI mapping, such as mount target device + * as ldiskfs, and modify something directly. + * 3. rc = -ENOENT. + * Someone just removed the object between the + * former oi_lookup and the iget. It is normal. + * 4. Other failure cases. + * + * Generally, when the device is mounted, it will + * auto check whether the system is restored from + * file-level backup or not. We trust such detect + * to distinguish the 1st case from the 2nd case. */ + if (rc == 0) { + if (!IS_ERR(inode) && inode->i_generation != 0 && + inode->i_generation == id->oii_gen) + rc = -ENOENT; + else + rc = -EREMCHG; + } + } else { + if (id->oii_gen == OSD_OII_NOGEN) + osd_id_gen(id, inode->i_ino, inode->i_generation); + + /* Do not update file c/mtime in ldiskfs. + * NB: we don't have any lock to protect this because we don't + * have reference on osd_object now, but contention with + * another lookup + attr_set can't happen in the tiny window + * between if (...) and set S_NOCMTIME. */ + if (!(inode->i_flags & S_NOCMTIME)) + inode->i_flags |= S_NOCMTIME; } - if (!lu_fid_eq(fid, &lma->lma_self_fid)) { - CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n", - PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid)); - iput(inode); - return ERR_PTR(-EREMCHG); + GOTO(put, rc); + +put: + if (rc != 0) { + if (!IS_ERR(inode)) + iput(inode); + + inode = ERR_PTR(rc); } return inode; } +/** + * \retval +v: new filter_fid, does not contain self-fid + * \retval 0: filter_fid_old, contains self-fid + * \retval -v: other failure cases + */ +int osd_get_idif(struct osd_thread_info *info, struct inode *inode, + struct dentry *dentry, struct lu_fid *fid) +{ + struct filter_fid_old *ff = &info->oti_ff; + struct ost_id *ostid = &info->oti_ostid; + int rc; + + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff)); + if (rc == sizeof(*ff)) { + 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 */ + ostid_to_fid(fid, ostid, 0); + } else if (rc == sizeof(struct filter_fid)) { + rc = 1; + } else if (rc >= 0) { + rc = -EINVAL; + } + + return rc; +} + +static int osd_lma_self_repair(struct osd_thread_info *info, + struct osd_device *osd, struct inode *inode, + const struct lu_fid *fid, __u32 compat) +{ + handle_t *jh; + int rc; + + LASSERT(current->journal_info == NULL); + + jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, + osd_dto_credits_noquota[DTO_XATTR_SET]); + if (IS_ERR(jh)) { + rc = PTR_ERR(jh); + CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n", + osd_name(osd), rc); + return rc; + } + + rc = osd_ea_fid_set(info, inode, fid, compat, 0); + if (rc != 0) + CWARN("%s: cannot self repair the LMA: rc = %d\n", + osd_name(osd), rc); + ldiskfs_journal_stop(jh); + return rc; +} + +static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct osd_device *osd = osd_obj2dev(obj); + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct inode *inode = obj->oo_inode; + struct dentry *dentry = &info->oti_obj_dentry; + struct lu_fid *fid = NULL; + const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); + int rc; + ENTRY; + + CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, + info->oti_mdt_attrs_old, LMA_OLD_SIZE); + if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) { + fid = &lma->lma_self_fid; + rc = osd_get_idif(info, inode, dentry, fid); + if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) { + /* For the given OST-object, if it has neither LMA nor + * FID in XATTR_NAME_FID, then the given FID (which is + * contained in the @obj, from client RPC for locating + * the OST-object) is trusted. We use it to generate + * the LMA. */ + osd_lma_self_repair(info, osd, inode, rfid, + fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ? + LMAC_FID_ON_OST : 0); + RETURN(0); + } + } + + if (unlikely(rc == -ENODATA)) + RETURN(0); + + if (rc < 0) + RETURN(rc); + + if (rc > 0) { + rc = 0; + lustre_lma_swab(lma); + if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) || + CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) { + CWARN("%s: unsupported incompat LMA feature(s) %#x for " + "fid = "DFID", ino = %lu\n", osd_name(osd), + lma->lma_incompat & ~LMA_INCOMPAT_SUPP, + PFID(rfid), inode->i_ino); + rc = -EOPNOTSUPP; + } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) { + fid = &lma->lma_self_fid; + } + } + + if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) { + if (fid_is_idif(rfid) && fid_is_idif(fid)) { + struct ost_id *oi = &info->oti_ostid; + struct lu_fid *fid1 = &info->oti_fid3; + __u32 idx = fid_idif_ost_idx(rfid); + + /* For old IDIF, the OST index is not part of the IDIF, + * Means that different OSTs may have the same IDIFs. + * Under such case, we need to make some compatible + * check to make sure to trigger OI scrub properly. */ + if (idx != 0 && fid_idif_ost_idx(fid) == 0) { + /* Given @rfid is new, LMA is old. */ + fid_to_ostid(fid, oi); + ostid_to_fid(fid1, oi, idx); + if (lu_fid_eq(fid1, rfid)) { + if (osd->od_lma_self_repair) + osd_lma_self_repair(info, osd, + inode, rfid, + LMAC_FID_ON_OST); + RETURN(0); + } + } + } + + CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n", + osd_name(osd), PFID(rfid), PFID(fid)); + rc = -EREMCHG; + } + + RETURN(rc); +} + static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, const struct lu_fid *fid, const struct lu_object_conf *conf) @@ -329,7 +551,9 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, struct osd_scrub *scrub; struct scrub_file *sf; int result; - int verify = 0; + int saved = 0; + bool in_oi = false; + bool triggered = false; ENTRY; LINVRNT(osd_invariant(obj)); @@ -346,37 +570,34 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) RETURN(-ENOENT); + /* For the object is created as locking anchor, or for the object to + * be created on disk. No need to osd_oi_lookup() at here because FID + * shouldn't never be re-used, if it's really a duplicate FID from + * unexpected reason, we should be able to detect it later by calling + * do_create->osd_oi_insert(). */ + if (conf != NULL && conf->loc_flags & LOC_F_NEW) + GOTO(out, result = 0); + /* Search order: 1. per-thread cache. */ - if (lu_fid_eq(fid, &oic->oic_fid)) { + if (lu_fid_eq(fid, &oic->oic_fid) && + likely(oic->oic_dev == dev)) { id = &oic->oic_lid; goto iget; } id = &info->oti_id; - if (!cfs_list_empty(&scrub->os_inconsistent_items)) { + if (!list_empty(&scrub->os_inconsistent_items)) { /* Search order: 2. OI scrub pending list. */ result = osd_oii_lookup(dev, fid, id); if (result == 0) goto iget; } - if (sf->sf_flags & SF_INCONSISTENT) - verify = 1; - - /* - * Objects are created as locking anchors or place holders for objects - * yet to be created. No need to osd_oi_lookup() at here because FID - * shouldn't never be re-used, if it's really a duplicate FID from - * unexpected reason, we should be able to detect it later by calling - * do_create->osd_oi_insert() - */ - if (conf != NULL && conf->loc_flags & LOC_F_NEW) - GOTO(out, result = 0); - /* Search order: 3. OI files. */ - result = osd_oi_lookup(info, dev, fid, id, true); + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result == -ENOENT) { - if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) || + if (!fid_is_norm(fid) || + fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); @@ -387,32 +608,64 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (result != 0) GOTO(out, result); + in_oi = true; + iget: - if (verify == 0) - inode = osd_iget(info, dev, id); - else - inode = osd_iget_verify(info, dev, id, fid); + inode = osd_iget_check(info, dev, fid, id, in_oi); if (IS_ERR(inode)) { result = PTR_ERR(inode); if (result == -ENOENT || result == -ESTALE) { - fid_zero(&oic->oic_fid); - result = 0; + if (!in_oi) + fid_zero(&oic->oic_fid); + + GOTO(out, result = -ENOENT); } else if (result == -EREMCHG) { trigger: + if (!in_oi) + fid_zero(&oic->oic_fid); + + if (unlikely(triggered)) + GOTO(out, result = saved); + + triggered = true; if (thread_is_running(&scrub->os_thread)) { result = -EINPROGRESS; } else if (!dev->od_noscrub) { result = osd_scrub_start(dev); - LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC " - "for "DFID", rc = %d [1]\n", - LDISKFS_SB(osd_sb(dev))->s_es->\ - s_volume_name,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; } + + /* We still have chance to get the valid inode: for the + * object which is referenced by remote name entry, the + * object on the local MDT will be linked under the dir + * of "/REMOTE_PARENT_DIR" with its FID string as name. + * + * We do not know whether the object for the given FID + * is referenced by some remote name entry or not, and + * especially for DNE II, a multiple-linked object may + * have many name entries reside on many MDTs. + * + * To simplify the operation, OSD will not distinguish + * more, just lookup "/REMOTE_PARENT_DIR". Usually, it + * only happened for the RPC from other MDT during the + * OI scrub, or for the client side RPC with FID only, + * such as FID to path, or from old connected client. */ + saved = result; + result = osd_lookup_in_remote_parent(info, dev, + fid, id); + if (result == 0) { + in_oi = false; + goto iget; + } + + result = saved; } GOTO(out, result); @@ -421,6 +674,26 @@ trigger: obj->oo_inode = inode; LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); + result = osd_check_lma(env, obj); + if (result != 0) { + iput(inode); + obj->oo_inode = NULL; + if (result == -EREMCHG) { + if (!in_oi) { + result = osd_oi_lookup(info, dev, fid, id, + OI_CHECK_FLD); + if (result != 0) { + fid_zero(&oic->oic_fid); + GOTO(out, result); + } + } + + goto trigger; + } + + GOTO(out, result); + } + obj->oo_compat_dot_created = 1; obj->oo_compat_dotdot_created = 1; @@ -639,55 +912,60 @@ static void osd_trans_commit_cb(struct super_block *sb, dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ - cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { + list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC, "commit callback entry: magic=%x name='%s'\n", dcb->dcb_magic, dcb->dcb_name); - cfs_list_del_init(&dcb->dcb_linkage); + list_del_init(&dcb->dcb_linkage); dcb->dcb_func(NULL, th, dcb, error); } - lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th); + lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th); lu_device_put(lud); th->th_dev = NULL; lu_context_exit(&th->th_ctx); lu_context_fini(&th->th_ctx); - OBD_FREE_PTR(oh); + thandle_put(th); } static struct thandle *osd_trans_create(const struct lu_env *env, - struct dt_device *d) + struct dt_device *d) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_thandle *oh; - struct thandle *th; - ENTRY; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_thandle *oh; + struct thandle *th; + ENTRY; - /* on pending IO in this thread should left from prev. request */ - LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0); + /* on pending IO in this thread should left from prev. request */ + LASSERT(atomic_read(&iobuf->dr_numreqs) == 0); - th = ERR_PTR(-ENOMEM); - OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO); - if (oh != NULL) { + th = ERR_PTR(-ENOMEM); + OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); + if (oh != NULL) { oh->ot_quota_trans = &oti->oti_quota_trans; memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); - th = &oh->ot_super; - th->th_dev = d; - th->th_result = 0; - th->th_tags = LCT_TX_HANDLE; - oh->ot_credits = 0; - oti->oti_dev = osd_dt_dev(d); - CFS_INIT_LIST_HEAD(&oh->ot_dcb_list); - osd_th_alloced(oh); - - memset(oti->oti_declare_ops, 0, OSD_OT_MAX); - memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX); - memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX); + th = &oh->ot_super; + th->th_dev = d; + th->th_result = 0; + th->th_tags = LCT_TX_HANDLE; + oh->ot_credits = 0; + atomic_set(&th->th_refc, 1); + th->th_alloc_size = sizeof(*oh); + oti->oti_dev = osd_dt_dev(d); + INIT_LIST_HEAD(&oh->ot_dcb_list); + 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)); + memset(oti->oti_declare_ops_cred, 0, + sizeof(oti->oti_declare_ops_cred)); oti->oti_rollback = false; - } - RETURN(th); + } + RETURN(th); } /* @@ -722,12 +1000,9 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, oh->ot_credits, osd_journal(dev)->j_max_transaction_buffers); -#ifdef OSD_TRACK_DECLARES - CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n", + 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_DELETE], - oti->oti_declare_ops_cred[OSD_OT_DELETE], 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", @@ -745,8 +1020,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, 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_DESTROY], - oti->oti_declare_ops_cred[OSD_OT_DESTROY]); + 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], @@ -759,7 +1034,6 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, last_credits = oh->ot_credits; last_printed = jiffies; } -#endif /* XXX Limit the credits to 'max_transaction_buffers', and * let the underlying filesystem to catch the error if * we really need so many credits. @@ -773,7 +1047,7 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, * XXX temporary stuff. Some abstraction layer should * be used. */ - jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits); + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); osd_th_started(oh); if (!IS_ERR(jh)) { oh->ot_handle = jh; @@ -782,8 +1056,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, lu_context_enter(&th->th_ctx); lu_device_get(&d->dd_lu_dev); - oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference, - "osd-tx", th); + lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link, + "osd-tx", th); oti->oti_txns++; rc = 0; } else { @@ -793,24 +1067,45 @@ out: RETURN(rc); } +static int osd_seq_exists(const struct lu_env *env, + struct osd_device *osd, obd_seq seq) +{ + struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; + + LASSERT(ss != NULL); + LASSERT(ss->ss_server_fld != NULL); + + rc = osd_fld_lookup(env, osd, seq, range); + if (rc != 0) { + if (rc != -ENOENT) + CERROR("%s: can't lookup FLD sequence "LPX64 + ": rc = %d\n", osd_name(osd), seq, rc); + RETURN(0); + } + + RETURN(ss->ss_node_id == range->lsr_index); +} + /* * Concurrency: shouldn't matter. */ -static int osd_trans_stop(const struct lu_env *env, struct thandle *th) +static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, + struct thandle *th) { - int rc = 0; - struct osd_thandle *oh; - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; + int rc = 0; + struct osd_thandle *oh; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; struct qsd_instance *qsd = oti->oti_dev->od_quota_slave; - ENTRY; + struct lquota_trans *qtrans; + ENTRY; - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of0(th, struct osd_thandle, ot_super); - if (qsd != NULL) - /* inform the quota slave device that the transaction is - * stopping */ - qsd_op_end(env, qsd, oh->ot_quota_trans); + qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; if (oh->ot_handle != NULL) { @@ -839,24 +1134,28 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) if (rc != 0) CERROR("Failure to stop transaction: %d\n", rc); } else { - OBD_FREE_PTR(oh); + thandle_put(&oh->ot_super); } - /* as we want IO to journal and data IO be concurrent, we don't block - * awaiting data IO completion in osd_do_bio(), instead we wait here - * once transaction is submitted to the journal. all reqular requests - * don't do direct IO (except read/write), thus this wait_event becomes - * no-op for them. - * - * IMPORTANT: we have to wait till any IO submited by the thread is - * completed otherwise iobuf may be corrupted by different request - */ - cfs_wait_event(iobuf->dr_wait, - cfs_atomic_read(&iobuf->dr_numreqs) == 0); - if (!rc) - rc = iobuf->dr_error; + /* inform the quota slave device that the transaction is stopping */ + qsd_op_end(env, qsd, qtrans); - RETURN(rc); + /* as we want IO to journal and data IO be concurrent, we don't block + * awaiting data IO completion in osd_do_bio(), instead we wait here + * once transaction is submitted to the journal. all reqular requests + * don't do direct IO (except read/write), thus this wait_event becomes + * no-op for them. + * + * IMPORTANT: we have to wait till any IO submited by the thread is + * completed otherwise iobuf may be corrupted by different request + */ + wait_event(iobuf->dr_wait, + atomic_read(&iobuf->dr_numreqs) == 0); + osd_fini_iobuf(oti->oti_dev, iobuf); + if (!rc) + rc = iobuf->dr_error; + + RETURN(rc); } static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) @@ -866,7 +1165,7 @@ 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); - cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); + list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); return 0; } @@ -892,8 +1191,8 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) osd_index_fini(obj); if (inode != NULL) { struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; - qid_t uid = inode->i_uid; - qid_t gid = inode->i_gid; + qid_t uid = i_uid_read(inode); + qid_t gid = i_gid_read(inode); iput(inode); obj->oo_inode = NULL; @@ -941,6 +1240,8 @@ 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. */ @@ -965,24 +1266,28 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, } spin_lock(&osd->od_osfs_lock); - /* cache 1 second */ - if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) { - result = sb->s_op->statfs(sb->s_root, ksfs); - if (likely(result == 0)) { /* N.B. statfs can't really fail */ - osd->od_osfs_age = cfs_time_current_64(); - statfs_pack(&osd->od_statfs, ksfs); - if (sb->s_flags & MS_RDONLY) - sfs->os_state = OS_STATE_READONLY; - } + 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 (likely(result == 0)) - *sfs = osd->od_statfs; spin_unlock(&osd->od_osfs_lock); - if (unlikely(env == NULL)) + if (unlikely(env == NULL)) OBD_FREE_PTR(ksfs); + /* 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; } @@ -1005,11 +1310,11 @@ static void osd_conf_get(const struct lu_env *env, struct dt_device_param *param) { struct super_block *sb = osd_sb(osd_dt_dev(dev)); + int ea_overhead; /* * XXX should be taken from not-yet-existing fs abstraction layer. */ - param->ddp_mnt = osd_dt_dev(dev)->od_mnt; param->ddp_max_name_len = LDISKFS_NAME_LEN; param->ddp_max_nlink = LDISKFS_LINK_MAX; param->ddp_block_shift = sb->s_blocksize_bits; @@ -1029,13 +1334,19 @@ static void osd_conf_get(const struct lu_env *env, if (test_opt(sb, POSIX_ACL)) param->ddp_mntopts |= MNTOPT_ACL; + /* LOD might calculate the max stripe count based on max_ea_size, + * so we need take account in the overhead as well, + * xattr_header + magic + xattr_entry_head */ + ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) + + LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN); + #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE) - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) - param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE; - else + if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) + param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE - + ea_overhead; + else #endif - param->ddp_max_ea_size = sb->s_blocksize; - + param->ddp_max_ea_size = sb->s_blocksize - ea_overhead; } /* @@ -1075,14 +1386,31 @@ static int osd_commit_async(const struct lu_env *env, static int osd_ro(const struct lu_env *env, struct dt_device *d) { - struct super_block *sb = osd_sb(osd_dt_dev(d)); - int rc; - ENTRY; + struct super_block *sb = osd_sb(osd_dt_dev(d)); + struct block_device *dev = sb->s_bdev; +#ifdef HAVE_DEV_SET_RDONLY + struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev; + int rc = 0; +#else + int rc = -EOPNOTSUPP; +#endif + ENTRY; +#ifdef HAVE_DEV_SET_RDONLY CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname); - rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev); - RETURN(rc); + if (jdev && (jdev != dev)) { + CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n", + (long)jdev); + dev_set_rdonly(jdev); + } + CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev); + dev_set_rdonly(dev); +#else + CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", + osd_dt_dev(d)->od_svname, (long)dev, rc); +#endif + RETURN(rc); } /* @@ -1107,55 +1435,57 @@ static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d, * If we mount with --data_journal we may need more. */ const int osd_dto_credits_noquota[DTO_NR] = { - /** - * Insert/Delete. - * INDEX_EXTRA_TRANS_BLOCKS(8) + - * SINGLEDATA_TRANS_BLOCKS(8) - * XXX Note: maybe iam need more, since iam have more level than - * EXT3 htree. - */ - [DTO_INDEX_INSERT] = 16, - [DTO_INDEX_DELETE] = 16, - /** + /** + * Insert. + * INDEX_EXTRA_TRANS_BLOCKS(8) + + * SINGLEDATA_TRANS_BLOCKS(8) + * XXX Note: maybe iam need more, since iam have more level than + * EXT3 htree. + */ + [DTO_INDEX_INSERT] = 16, + /** + * Delete + * just modify a single entry, probably merge few within a block + */ + [DTO_INDEX_DELETE] = 1, + /** * Used for OI scrub - */ - [DTO_INDEX_UPDATE] = 16, - /** - * Create a object. The same as create object in EXT3. - * DATA_TRANS_BLOCKS(14) + - * INDEX_EXTRA_BLOCKS(8) + - * 3(inode bits, groups, GDT) - */ - [DTO_OBJECT_CREATE] = 25, - /** - * XXX: real credits to be fixed - */ - [DTO_OBJECT_DELETE] = 25, - /** - * Attr set credits (inode) - */ - [DTO_ATTR_SET_BASE] = 1, - /** - * Xattr set. The same as xattr of EXT3. - * DATA_TRANS_BLOCKS(14) - * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS - * are also counted in. Do not know why? - */ - [DTO_XATTR_SET] = 14, - [DTO_LOG_REC] = 14, - /** - * credits for inode change during write. - */ - [DTO_WRITE_BASE] = 3, - /** - * credits for single block write. - */ - [DTO_WRITE_BLOCK] = 14, - /** - * Attr set credits for chown. - * This is extra credits for setattr, and it is null without quota - */ - [DTO_ATTR_SET_CHOWN]= 0 + */ + [DTO_INDEX_UPDATE] = 16, + /** + * 4(inode, inode bits, groups, GDT) + * notice: OI updates are counted separately with DTO_INDEX_INSERT + */ + [DTO_OBJECT_CREATE] = 4, + /** + * 4(inode, inode bits, groups, GDT) + * notice: OI updates are counted separately with DTO_INDEX_DELETE + */ + [DTO_OBJECT_DELETE] = 4, + /** + * Attr set credits (inode) + */ + [DTO_ATTR_SET_BASE] = 1, + /** + * Xattr set. The same as xattr of EXT3. + * DATA_TRANS_BLOCKS(14) + * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS + * are also counted in. Do not know why? + */ + [DTO_XATTR_SET] = 14, + /** + * credits for inode change during write. + */ + [DTO_WRITE_BASE] = 3, + /** + * credits for single block write. + */ + [DTO_WRITE_BLOCK] = 14, + /** + * Attr set credits for chown. + * This is extra credits for setattr, and it is null without quota + */ + [DTO_ATTR_SET_CHOWN] = 0 }; static const struct dt_device_operations osd_dt_ops = { @@ -1297,48 +1627,50 @@ static int capa_is_sane(const struct lu_env *env, } int osd_object_auth(const struct lu_env *env, struct dt_object *dt, - struct lustre_capa *capa, __u64 opc) + struct lustre_capa *capa, __u64 opc) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_device *dev = osd_dev(dt->do_lu.lo_dev); - struct md_capainfo *ci; - int rc; + 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 (!dev->od_fl_capa) - return 0; + if (!osd->od_fl_capa) + return 0; - if (capa == BYPASS_CAPA) - return 0; + if (capa == BYPASS_CAPA) + return 0; - ci = md_capainfo(env); - if (unlikely(!ci)) - return 0; + lci = lu_capainfo_get(env); + if (unlikely(lci == NULL)) + return 0; - if (ci->mc_auth == LC_ID_NONE) - return 0; + if (lci->lci_auth == LC_ID_NONE) + return 0; - if (!capa) { - CERROR("no capability is provided for fid "DFID"\n", PFID(fid)); - return -EACCES; - } + 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 (!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; - } + if (!capa_opc_supported(capa, opc)) { + DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc); + return -EACCES; + } - if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) { - DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc); - 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; + return 0; } static struct timespec *osd_inode_time(const struct lu_env *env, @@ -1355,26 +1687,26 @@ static struct timespec *osd_inode_time(const struct lu_env *env, static void osd_inode_getattr(const struct lu_env *env, - struct inode *inode, struct lu_attr *attr) -{ - attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | - LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | - LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | - LA_TYPE; - - attr->la_atime = LTIME_S(inode->i_atime); - attr->la_mtime = LTIME_S(inode->i_mtime); - attr->la_ctime = LTIME_S(inode->i_ctime); - attr->la_mode = inode->i_mode; - attr->la_size = i_size_read(inode); - attr->la_blocks = inode->i_blocks; - attr->la_uid = inode->i_uid; - attr->la_gid = inode->i_gid; - attr->la_flags = LDISKFS_I(inode)->i_flags; - attr->la_nlink = inode->i_nlink; - attr->la_rdev = inode->i_rdev; - attr->la_blksize = 1 << inode->i_blkbits; - attr->la_blkbits = inode->i_blkbits; + struct inode *inode, struct lu_attr *attr) +{ + attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | + LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | + LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | + LA_TYPE; + + attr->la_atime = LTIME_S(inode->i_atime); + attr->la_mtime = LTIME_S(inode->i_mtime); + attr->la_ctime = LTIME_S(inode->i_ctime); + attr->la_mode = inode->i_mode; + attr->la_size = i_size_read(inode); + 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_nlink = inode->i_nlink; + attr->la_rdev = inode->i_rdev; + attr->la_blksize = 1 << inode->i_blkbits; + attr->la_blkbits = inode->i_blkbits; } static int osd_attr_get(const struct lu_env *env, @@ -1405,6 +1737,8 @@ static int osd_declare_attr_set(const struct lu_env *env, struct osd_object *obj; struct osd_thread_info *info = osd_oti_get(env); struct lquota_id_info *qi = &info->oti_qi; + qid_t uid; + qid_t gid; long long bspace; int rc = 0; bool allocated; @@ -1436,7 +1770,7 @@ static int osd_declare_attr_set(const struct lu_env *env, * credits for updating quota accounting files and to trigger quota * space adjustment once the operation is completed.*/ if ((attr->la_valid & LA_UID) != 0 && - attr->la_uid != obj->oo_inode->i_uid) { + attr->la_uid != (uid = i_uid_read(obj->oo_inode))) { qi->lqi_type = USRQUOTA; /* inode accounting */ @@ -1453,7 +1787,7 @@ static int osd_declare_attr_set(const struct lu_env *env, RETURN(rc); /* and one less inode for the current uid */ - qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + qi->lqi_id.qid_uid = uid; qi->lqi_space = -1; rc = osd_declare_qid(env, oh, qi, true, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) @@ -1475,7 +1809,7 @@ static int osd_declare_attr_set(const struct lu_env *env, RETURN(rc); /* and finally less blocks for the current owner */ - qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + qi->lqi_id.qid_uid = uid; qi->lqi_space = -bspace; rc = osd_declare_qid(env, oh, qi, true, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) @@ -1485,7 +1819,7 @@ static int osd_declare_attr_set(const struct lu_env *env, } if (attr->la_valid & LA_GID && - attr->la_gid != obj->oo_inode->i_gid) { + attr->la_gid != (gid = i_gid_read(obj->oo_inode))) { qi->lqi_type = GRPQUOTA; /* inode accounting */ @@ -1502,7 +1836,7 @@ static int osd_declare_attr_set(const struct lu_env *env, RETURN(rc); /* and one less inode for the current gid */ - qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + qi->lqi_id.qid_gid = gid; qi->lqi_space = -1; rc = osd_declare_qid(env, oh, qi, true, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) @@ -1524,7 +1858,7 @@ static int osd_declare_attr_set(const struct lu_env *env, RETURN(rc); /* and finally less blocks for the current owner */ - qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + qi->lqi_id.qid_gid = gid; qi->lqi_space = -bspace; rc = osd_declare_qid(env, oh, qi, true, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) @@ -1537,11 +1871,16 @@ static int osd_declare_attr_set(const struct lu_env *env, } static int osd_inode_setattr(const struct lu_env *env, - struct inode *inode, const struct lu_attr *attr) + struct inode *inode, const struct lu_attr *attr) { - __u64 bits; + __u64 bits = attr->la_valid; - bits = attr->la_valid; + /* Only allow set size for regular file */ + if (!S_ISREG(inode->i_mode)) + bits &= ~(LA_SIZE | LA_BLOCKS); + + if (bits == 0) + return 0; if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); @@ -1560,17 +1899,17 @@ static int osd_inode_setattr(const struct lu_env *env, if (bits & LA_BLOCKS) inode->i_blocks = attr->la_blocks; #endif - if (bits & LA_MODE) - inode->i_mode = (inode->i_mode & S_IFMT) | - (attr->la_mode & ~S_IFMT); - if (bits & LA_UID) - inode->i_uid = attr->la_uid; - if (bits & LA_GID) - inode->i_gid = attr->la_gid; - if (bits & LA_NLINK) + if (bits & LA_MODE) + inode->i_mode = (inode->i_mode & S_IFMT) | + (attr->la_mode & ~S_IFMT); + if (bits & LA_UID) + i_uid_write(inode, attr->la_uid); + if (bits & LA_GID) + i_gid_write(inode, attr->la_gid); + if (bits & LA_NLINK) set_nlink(inode, attr->la_nlink); - if (bits & LA_RDEV) - inode->i_rdev = attr->la_rdev; + if (bits & LA_RDEV) + inode->i_rdev = attr->la_rdev; if (bits & LA_FLAGS) { /* always keep S_NOCMTIME */ @@ -1582,8 +1921,8 @@ static int osd_inode_setattr(const struct lu_env *env, static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) { - if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) || - (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) { + if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) || + (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) { struct iattr iattr; int rc; @@ -1592,8 +1931,8 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) iattr.ia_valid |= ATTR_UID; if (attr->la_valid & LA_GID) iattr.ia_valid |= ATTR_GID; - iattr.ia_uid = attr->la_uid; - iattr.ia_gid = attr->la_gid; + iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid); + iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid); rc = ll_vfs_dq_transfer(inode, &iattr); if (rc) { @@ -1662,7 +2001,7 @@ static int osd_attr_set(const struct lu_env *env, spin_unlock(&obj->oo_guard); if (!rc) - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); return rc; } @@ -1674,9 +2013,8 @@ struct dentry *osd_child_dentry_get(const struct lu_env *env, } static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, - cfs_umode_t mode, - struct dt_allocation_hint *hint, - struct thandle *th) + umode_t mode, struct dt_allocation_hint *hint, + struct thandle *th) { int result; struct osd_device *osd = osd_obj2dev(obj); @@ -1809,16 +2147,16 @@ static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj, } static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) { - cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX); - int result; + umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX); + int result; - LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_inode == NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(obj->oo_inode == NULL); LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)); @@ -1873,11 +2211,10 @@ static osd_obj_type_f osd_create_type_f(enum dt_format_type type) static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, struct dt_object *parent, struct dt_object *child, - cfs_umode_t child_mode) + umode_t child_mode) { LASSERT(ah); - memset(ah, 0, sizeof(*ah)); ah->dah_parent = parent; ah->dah_mode = child_mode; } @@ -1914,7 +2251,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, * enabled on ldiskfs (lquota takes care of it). */ LASSERTF(result == 0, "%d", result); - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); } attr->la_valid = valid; @@ -1932,6 +2269,11 @@ static int __osd_object_create(struct osd_thread_info *info, struct thandle *th) { int result; + __u32 umask; + + /* 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); @@ -1943,7 +2285,10 @@ static int __osd_object_create(struct osd_thread_info *info, unlock_new_inode(obj->oo_inode); } - return result; + /* restore previous umask value */ + current->fs->umask = umask; + + return result; } /** @@ -1952,32 +2297,35 @@ static int __osd_object_create(struct osd_thread_info *info, * \retval 0, on success */ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, - const struct lu_fid *fid, struct thandle *th) + const struct lu_fid *fid, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_inode_id *id = &info->oti_id; - struct osd_device *osd = osd_obj2dev(obj); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_inode_id *id = &info->oti_id; + struct osd_device *osd = osd_obj2dev(obj); + struct osd_thandle *oh; - LASSERT(obj->oo_inode != NULL); + LASSERT(obj->oo_inode != NULL); + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle); osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation); - return osd_oi_insert(info, osd, fid, id, th); + return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD); } int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, - const struct lu_fid *fid, struct lu_seq_range *range) + obd_seq seq, struct lu_seq_range *range) { struct seq_server_site *ss = osd_seq_site(osd); - int rc; - if (fid_is_idif(fid)) { - range->lsr_flags = LU_SEQ_RANGE_OST; - range->lsr_index = fid_idif_ost_idx(fid); + if (fid_seq_is_idif(seq)) { + fld_range_set_ost(range); + range->lsr_index = idif_ost_idx(seq); return 0; } - if (!fid_seq_in_fldb(fid_seq(fid))) { - range->lsr_flags = LU_SEQ_RANGE_MDT; + if (!fid_seq_in_fldb(seq)) { + fld_range_set_mdt(range); if (ss != NULL) /* FIXME: If ss is NULL, it suppose not get lsr_index * at all */ @@ -1986,13 +2334,9 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, } LASSERT(ss != NULL); - range->lsr_flags = -1; - rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range); - if (rc != 0) { - CERROR("%s can not find "DFID": rc = %d\n", - osd_name(osd), PFID(fid), rc); - } - return rc; + fld_range_set_any(range); + /* OSD will only do local fld lookup */ + return fld_local_lookup(env, ss->ss_server_fld, seq, range); } /* @@ -2005,7 +2349,6 @@ static int osd_declare_object_create(const struct lu_env *env, struct dt_object_format *dof, struct thandle *handle) { - struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; struct osd_thandle *oh; int rc; ENTRY; @@ -2017,12 +2360,10 @@ static int osd_declare_object_create(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_CREATE, osd_dto_credits_noquota[DTO_OBJECT_CREATE]); - if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev), - lu_object_fid(&dt->do_lu))) - /* 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); + /* 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 @@ -2043,16 +2384,6 @@ static int osd_declare_object_create(const struct lu_env *env, if (rc != 0) RETURN(rc); - /* It does fld look up inside declare, and the result will be - * added to fld cache, so the following fld lookup inside insert - * does not need send RPC anymore, so avoid send rpc with holding - * transaction */ - if (fid_is_norm(lu_object_fid(&dt->do_lu)) && - !fid_is_last_id(lu_object_fid(&dt->do_lu))) - osd_fld_lookup(env, osd_dt_dev(handle->th_dev), - lu_object_fid(&dt->do_lu), range); - - RETURN(rc); } @@ -2112,21 +2443,20 @@ static int osd_declare_object_destroy(const struct lu_env *env, LASSERT(oh->ot_handle == NULL); LASSERT(inode); - osd_trans_declare_op(env, oh, OSD_OT_DELETE, + osd_trans_declare_op(env, oh, OSD_OT_DESTROY, osd_dto_credits_noquota[DTO_OBJECT_DELETE]); /* Recycle idle OI leaf may cause additional three OI blocks * to be changed. */ - osd_trans_declare_op(env, oh, OSD_OT_DESTROY, + osd_trans_declare_op(env, oh, OSD_OT_DELETE, osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); - /* one less inode */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh, - false, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + -1, oh, false, true, NULL, false); if (rc) RETURN(rc); /* data to be truncated */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + 0, oh, true, true, NULL, false); RETURN(rc); } @@ -2150,29 +2480,26 @@ static int osd_object_destroy(const struct lu_env *env, if (unlikely(fid_is_acct(fid))) RETURN(-EPERM); - /* Parallel control for OI scrub. For most of cases, there is no - * lock contention. So it will not affect unlink performance. */ - mutex_lock(&inode->i_mutex); if (S_ISDIR(inode->i_mode)) { - LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1); - /* it will check/delete the agent inode for every dir - * destory, how to optimize it? unlink performance - * impaction XXX */ - result = osd_delete_from_agent(env, osd, obj, oh); + LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 || + inode->i_nlink == 2); + /* it will check/delete the inode from remote parent, + * how to optimize it? unlink performance impaction XXX */ + result = osd_delete_from_remote_parent(env, osd, obj, oh); if (result != 0 && result != -ENOENT) { - CERROR("%s: delete agent inode "DFID": rc = %d\n", + CERROR("%s: delete inode "DFID": rc = %d\n", osd_name(osd), PFID(fid), result); } spin_lock(&obj->oo_guard); clear_nlink(inode); spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); } osd_trans_exec_op(env, th, OSD_OT_DESTROY); - result = osd_oi_delete(osd_oti_get(env), osd, fid, th); - mutex_unlock(&inode->i_mutex); + 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) */ @@ -2183,17 +2510,6 @@ static int osd_object_destroy(const struct lu_env *env, RETURN(0); } -static inline int __osd_xattr_set(struct osd_thread_info *info, - struct inode *inode, const char *name, - const void *buf, int buflen, int fl) -{ - struct dentry *dentry = &info->oti_child_dentry; - - ll_vfs_dq_init(inode); - dentry->d_inode = inode; - return inode->i_op->setxattr(dentry, name, buf, buflen, fl); -} - /** * Put the fid into lustre_mdt_attrs, and then place the structure * inode's ea. This fid should not be altered during the life time @@ -2205,26 +2521,45 @@ static inline int __osd_xattr_set(struct osd_thread_info *info, * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here */ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, - const struct lu_fid *fid) + const struct lu_fid *fid, __u32 compat, __u32 incompat) { struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; int rc; + ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) - return 0; - - if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid)) - return 0; + RETURN(0); - lustre_lma_init(lma, fid); + lustre_lma_init(lma, fid, compat, incompat); lustre_lma_swab(lma); rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma), XATTR_CREATE); - /* Someone may created the EA by race. */ - if (unlikely(rc == -EEXIST)) - rc = 0; - return rc; + /* LMA may already exist, but we need to check that all the + * desired compat/incompat flags have been added. */ + if (unlikely(rc == -EEXIST)) { + if (compat == 0 && incompat == 0) + RETURN(0); + + rc = __osd_xattr_get(inode, &info->oti_obj_dentry, + XATTR_NAME_LMA, info->oti_mdt_attrs_old, + LMA_OLD_SIZE); + if (rc <= 0) + RETURN(-EINVAL); + + lustre_lma_swab(lma); + if (!(~lma->lma_compat & compat) && + !(~lma->lma_incompat & incompat)) + RETURN(0); + + lma->lma_compat |= compat; + lma->lma_incompat |= incompat; + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + } + + RETURN(rc); } /** @@ -2236,9 +2571,10 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, * its inmemory API. */ void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) + const struct lu_fid *fid) { - if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) { + if (!fid_is_namespace_visible(fid) || + OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) { param->edp_magic = 0; return; } @@ -2274,9 +2610,9 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, struct inode *dir, - struct inode *parent_dir, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, + struct inode *parent_dir, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, struct osd_thandle *oth) { struct ldiskfs_dentry_param *dot_ldp; @@ -2292,13 +2628,14 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, } /** - * Create an local inode for remote entry + * Create an local agent inode for remote entry */ -static struct inode *osd_create_remote_inode(const struct lu_env *env, - struct osd_device *osd, - struct osd_object *pobj, - const struct lu_fid *fid, - struct thandle *th) +static struct inode *osd_create_local_agent_inode(const struct lu_env *env, + struct osd_device *osd, + struct osd_object *pobj, + const struct lu_fid *fid, + __u32 type, + struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); struct inode *local; @@ -2310,18 +2647,27 @@ static struct inode *osd_create_remote_inode(const struct lu_env *env, oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle->h_transaction != NULL); - /* FIXME: Insert index api needs to know the mode of - * the remote object. Just use S_IFDIR for now */ - local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR); + local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type); if (IS_ERR(local)) { CERROR("%s: create local error %d\n", osd_name(osd), (int)PTR_ERR(local)); RETURN(local); } + /* Set special LMA flag for local agent inode */ + rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT); + if (rc != 0) { + CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n", + osd_name(osd), PFID(fid), rc); + RETURN(ERR_PTR(rc)); + } + + if (!S_ISDIR(type)) + RETURN(local); + rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode, - (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), - (const struct dt_rec *)fid, oh); + lu_object_fid(&pobj->oo_dt.do_lu), + fid, oh); if (rc != 0) { CERROR("%s: "DFID" add dot dotdot error: rc = %d\n", osd_name(osd), PFID(fid), rc); @@ -2332,12 +2678,12 @@ static struct inode *osd_create_remote_inode(const struct lu_env *env, } /** - * Delete local inode for remote entry + * Delete local agent inode for remote entry */ -static int osd_delete_remote_inode(const struct lu_env *env, - struct osd_device *osd, - const struct lu_fid *fid, - __u32 ino, struct osd_thandle *oh) +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) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -2396,13 +2742,14 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); result = __osd_object_create(info, obj, attr, hint, dof, th); - if ((result == 0) && - (fid_is_last_id(fid) || - !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid))) - result = osd_ea_fid_set(info, obj->oo_inode, fid); + if (result == 0) + result = osd_ea_fid_set(info, obj->oo_inode, fid, + fid_is_on_ost(info, osd_obj2dev(obj), + fid, OI_CHECK_FLD) ? + LMAC_FID_ON_OST : 0, 0); - if (result == 0) - result = __osd_oi_insert(env, obj, fid, th); + if (result == 0) + result = __osd_oi_insert(env, obj, fid, th); LASSERT(ergo(result == 0, dt_object_exists(dt) && !dt_object_remote(dt))); @@ -2434,43 +2781,51 @@ static int osd_declare_object_ref_add(const struct lu_env *env, static int osd_object_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thandle *oh; + int rc = 0; LINVRNT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + osd_trans_exec_op(env, th, OSD_OT_REF_ADD); + CDEBUG(D_INODE, DFID" increase nlink %d\n", + PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink); /* - * DIR_NLINK feature is set for compatibility reasons if: - * 1) nlinks > LDISKFS_LINK_MAX, or - * 2) nlinks == 2, since this indicates i_nlink was previously 1. + * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX + * (65000) subdirectories by storing "1" in i_nlink if the link count + * would otherwise overflow. Directory tranversal tools understand + * that (st_nlink == 1) indicates that the filesystem dose not track + * hard links count on the directory, and will not abort subdirectory + * scanning early once (st_nlink - 2) subdirs have been found. * - * It is easier to always set this flag (rather than check and set), - * since it has less overhead, and the superblock will be dirtied - * at some point. Both e2fsprogs and any Lustre-supported ldiskfs - * do not actually care whether this flag is set or not. + * This also has to properly handle the case of inodes with nlink == 0 + * in case they are being linked into the PENDING directory */ +#ifdef I_LINKABLE + /* This is necessary to increment from i_nlink == 0 */ + spin_lock(&inode->i_lock); + inode->i_state |= I_LINKABLE; + spin_unlock(&inode->i_lock); +#endif + spin_lock(&obj->oo_guard); - /* inc_nlink from 0 may cause WARN_ON */ - if(inode->i_nlink == 0) - set_nlink(inode, 1); - else - inc_nlink(inode); - if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) { - if (inode->i_nlink >= LDISKFS_LINK_MAX || - inode->i_nlink == 2) - set_nlink(inode, 1); - } - LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + ldiskfs_inc_count(oh->ot_handle, inode); + if (!S_ISDIR(inode->i_mode)) + LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); + + ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); - return 0; + return rc; } static int osd_declare_object_ref_del(const struct lu_env *env, @@ -2497,26 +2852,41 @@ static int osd_declare_object_ref_del(const struct lu_env *env, static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct osd_thandle *oh; - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_DEL); spin_lock(&obj->oo_guard); - LASSERT(inode->i_nlink > 0); - drop_nlink(inode); - /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX) - * then the nlink count is 1. Don't let it be set to 0 or the directory - * inode will be deleted incorrectly. */ - if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0) - set_nlink(inode, 1); + /* That can be result of upgrade from old Lustre version and + * applied only to local files. Just skip this ref_del call. + * ext4_unlink() only treats this as a warning, don't LASSERT here.*/ + if (inode->i_nlink == 0) { + CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ? + D_ERROR : D_INODE, "%s: nlink == 0 on "DFID + ", maybe an upgraded file? (LU-3915)\n", + osd_name(osd), PFID(lu_object_fid(&dt->do_lu))); + spin_unlock(&obj->oo_guard); + return 0; + } + + CDEBUG(D_INODE, DFID" decrease nlink %d\n", + PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink); + + ldiskfs_dec_count(oh->ot_handle, inode); spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); + + ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); return 0; @@ -2573,16 +2943,32 @@ static int osd_declare_xattr_set(const struct lu_env *env, int fl, struct thandle *handle) { struct osd_thandle *oh; + int credits; LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, - strcmp(name, XATTR_NAME_VERSION) == 0 ? - osd_dto_credits_noquota[DTO_ATTR_SET_BASE] : - osd_dto_credits_noquota[DTO_XATTR_SET]); + /* optimistic optimization: LMA is set first and usually fit inode */ + if (strcmp(name, XATTR_NAME_LMA) == 0) { + if (dt_object_exists(dt)) + credits = 0; + else + credits = 1; + } else if (strcmp(name, XATTR_NAME_VERSION) == 0) { + credits = 1; + } else { + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct super_block *sb = osd_sb(osd); + 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; + } + } + + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); return 0; } @@ -2602,7 +2988,7 @@ static void osd_object_version_set(const struct lu_env *env, LDISKFS_I(inode)->i_fs_version = *new_version; /** Version is set after all inode operations are finished, * so we should mark it dirty here */ - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); } /* @@ -2616,6 +3002,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, struct inode *inode = obj->oo_inode; struct osd_thread_info *info = osd_oti_get(env); int fs_flags = 0; + ENTRY; LASSERT(handle != NULL); @@ -2631,6 +3018,9 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) return -EACCES; + CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n", + PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); + osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); if (fl & LU_XATTR_REPLACE) fs_flags |= XATTR_REPLACE; @@ -2638,6 +3028,22 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (fl & LU_XATTR_CREATE) fs_flags |= XATTR_CREATE; + if (strcmp(name, XATTR_NAME_LMV) == 0) { + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + int rc; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + if (rc != 0) + RETURN(rc); + + lma->lma_incompat |= LMAI_STRIPED; + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + if (rc != 0) + RETURN(rc); + } + return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, fs_flags); } @@ -2655,12 +3061,12 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL); - LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj)); 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); } @@ -2706,98 +3112,99 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, 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 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 *dev = osd_obj2dev(obj); - struct lustre_capa_key *key = &info->oti_capa_key; - struct lustre_capa *capa = &info->oti_capa; - struct obd_capa *oc; - struct md_capainfo *ci; - int rc; - ENTRY; + 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 (!dev->od_fl_capa) - RETURN(ERR_PTR(-ENOENT)); + if (!osd->od_fl_capa) + RETURN(ERR_PTR(-ENOENT)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - /* renewal sanity check */ - if (old && osd_object_auth(env, dt, old, opc)) - RETURN(ERR_PTR(-EACCES)); - - ci = md_capainfo(env); - if (unlikely(!ci)) - RETURN(ERR_PTR(-ENOENT)); - - switch (ci->mc_auth) { - case LC_ID_NONE: - RETURN(NULL); - case LC_ID_PLAIN: - capa->lc_uid = obj->oo_inode->i_uid; - capa->lc_gid = obj->oo_inode->i_gid; - capa->lc_flags = LC_ID_PLAIN; - break; - case LC_ID_CONVERT: { - __u32 d[4], s[4]; - - s[0] = obj->oo_inode->i_uid; - cfs_get_random_bytes(&(s[1]), sizeof(__u32)); - s[2] = obj->oo_inode->i_gid; - 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)); + /* renewal sanity check */ + if (old && osd_object_auth(env, dt, old, opc)) + RETURN(ERR_PTR(-EACCES)); + + lci = lu_capainfo_get(env); + if (unlikely(lci == NULL)) + RETURN(ERR_PTR(-ENOENT)); + + switch (lci->lci_auth) { + case LC_ID_NONE: + RETURN(NULL); + case LC_ID_PLAIN: + capa->lc_uid = i_uid_read(obj->oo_inode); + capa->lc_gid = i_gid_read(obj->oo_inode); + capa->lc_flags = LC_ID_PLAIN; + break; + case LC_ID_CONVERT: { + __u32 d[4], s[4]; + + s[0] = i_uid_read(obj->oo_inode); + cfs_get_random_bytes(&(s[1]), sizeof(__u32)); + s[2] = i_uid_read(obj->oo_inode); + cfs_get_random_bytes(&(s[3]), sizeof(__u32)); + rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN); + if (unlikely(rc)) + RETURN(ERR_PTR(rc)); + + capa->lc_uid = ((__u64)d[1] << 32) | d[0]; + capa->lc_gid = ((__u64)d[3] << 32) | d[2]; + capa->lc_flags = LC_ID_CONVERT; + break; } + default: + RETURN(ERR_PTR(-EINVAL)); + } - capa->lc_fid = *fid; - capa->lc_opc = opc; - capa->lc_flags |= dev->od_capa_alg << 24; - capa->lc_timeout = dev->od_capa_timeout; - capa->lc_expiry = 0; + capa->lc_fid = *fid; + capa->lc_opc = opc; + capa->lc_flags |= osd->od_capa_alg << 24; + capa->lc_timeout = osd->od_capa_timeout; + capa->lc_expiry = 0; - oc = capa_lookup(dev->od_capa_hash, capa, 1); - if (oc) { - LASSERT(!capa_is_expired(oc)); - RETURN(oc); - } + oc = capa_lookup(osd->od_capa_hash, capa, 1); + if (oc) { + LASSERT(!capa_is_expired(oc)); + RETURN(oc); + } spin_lock(&capa_lock); - *key = dev->od_capa_keys[1]; + *key = osd->od_capa_keys[1]; spin_unlock(&capa_lock); - capa->lc_keyid = key->lk_keyid; - capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout; + capa->lc_keyid = key->lk_keyid; + capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout; - rc = capa_hmac(capa->lc_hmac, capa, key->lk_key); - if (rc) { - DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc); - RETURN(ERR_PTR(rc)); - } + rc = capa_hmac(capa->lc_hmac, capa, key->lk_key); + if (rc) { + DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc); + RETURN(ERR_PTR(rc)); + } - oc = capa_add(dev->od_capa_hash, capa); - RETURN(oc); + oc = capa_add(osd->od_capa_hash, capa); + RETURN(oc); } -static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) +static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, + __u64 start, __u64 end) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -2809,12 +3216,24 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) ENTRY; dentry->d_inode = inode; + dentry->d_sb = inode->i_sb; file->f_dentry = dentry; file->f_mapping = inode->i_mapping; file->f_op = inode->i_fop; + set_file_inode(file, inode); + +#ifdef HAVE_FILE_FSYNC_4ARGS + rc = file->f_op->fsync(file, start, end, 0); +#elif defined(HAVE_FILE_FSYNC_2ARGS) + mutex_lock(&inode->i_mutex); + rc = file->f_op->fsync(file, 0); + mutex_unlock(&inode->i_mutex); +#else mutex_lock(&inode->i_mutex); rc = file->f_op->fsync(file, dentry, 0); mutex_unlock(&inode->i_mutex); +#endif + RETURN(rc); } @@ -2936,7 +3355,8 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, * recheck under lock. */ if (!osd_has_index(obj)) - result = osd_iam_container_init(env, obj, dir); + result = osd_iam_container_init(env, obj, + obj->oo_dir); else result = 0; up_write(&obj->oo_ext_idx_sem); @@ -2953,7 +3373,8 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } LINVRNT(osd_invariant(obj)); - if (is_quota_glb_feat(feat)) + if (result == 0 && is_quota_glb_feat(feat) && + fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB) result = osd_quota_migration(env, dt, feat); return result; @@ -3131,8 +3552,8 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, inode = osd_dt_obj(dt)->oo_inode; LASSERT(inode); - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + 0, oh, true, true, NULL, false); RETURN(rc); } @@ -3150,25 +3571,26 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, } static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, - struct lu_fid *fid) + const struct lu_fid *fid) { - struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; struct seq_server_site *ss = osd_seq_site(osd); - int rc; ENTRY; - /* Those FID seqs, which are not in FLDB, must be local seq */ - if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL)) + /* FID seqs not in FLDB, must be local seq */ + if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) RETURN(0); - rc = osd_fld_lookup(env, osd, fid, range); - if (rc != 0) { - CERROR("%s: Can not lookup fld for "DFID"\n", - osd_name(osd), PFID(fid)); - RETURN(rc); - } + /* If FLD is not being initialized yet, it only happens during the + * initialization, likely during mgs initialization, and we assume + * this is local FID. */ + if (ss == NULL || ss->ss_server_fld == NULL) + RETURN(0); - RETURN(ss->ss_node_id != range->lsr_index); + /* Only check the local FLDB here */ + if (osd_seq_exists(env, osd, fid_seq(fid))) + RETURN(0); + + RETURN(1); } /** @@ -3223,11 +3645,77 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, down_write(&obj->oo_ext_idx_sem); } - bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (bh) { - rc = ldiskfs_delete_entry(oh->ot_handle, - dir, de, 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. + * upgraded 1.8 filesystem without dirdata enabled) then + * we need to get the FID from the LMA. For a remote directory + * there HAS to be an LMA, it cannot be an IGIF inode in this + * case. + * + * Delete the entry before the agent inode in order to + * simplify error handling. At worst an error after deleting + * the entry first might leak the agent inode afterward. The + * reverse would need filesystem abort in case of error deleting + * the entry after the agent had been removed, or leave a + * dangling entry pointing at a random inode. */ + 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); + } + } + 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; } @@ -3240,23 +3728,20 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, GOTO(out, rc); /* For inode on the remote MDT, .. will point to - * /Agent directory. So do not try to lookup/delete - * remote inode for .. */ - if (strcmp((char *)key, dotdot) == 0) - GOTO(out, rc = 0); - - LASSERT(de != NULL); - rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); - if (rc == 0 && osd_remote_fid(env, osd, fid)) { - __u32 ino = le32_to_cpu(de->inode); + * /Agent directory, Check whether it needs to delete + * from agent directory */ + if (unlikely(strcmp((char *)key, dotdot) == 0)) { + rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj, + oh); + if (rc != 0 && rc != -ENOENT) { + CERROR("%s: delete agent inode "DFID": rc = %d\n", + osd_name(osd), PFID(fid), rc); + } - rc = osd_delete_remote_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 { - if (rc == -ENODATA) + if (rc == -ENOENT) rc = 0; + + GOTO(out, rc); } out: @@ -3426,9 +3911,9 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, * \retval -ve, on error */ static int __osd_ea_add_rec(struct osd_thread_info *info, - struct osd_object *pobj, struct inode *cinode, - const char *name, const struct dt_rec *fid, - struct htree_lock *hlock, struct thandle *th) + struct osd_object *pobj, struct inode *cinode, + const char *name, const struct lu_fid *fid, + struct htree_lock *hlock, struct thandle *th) { struct ldiskfs_dentry_param *ldp; struct dentry *child; @@ -3467,11 +3952,11 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, * \retval -ve, on error */ static int osd_add_dot_dotdot(struct osd_thread_info *info, - struct osd_object *dir, - struct inode *parent_dir, const char *name, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, - struct thandle *th) + struct osd_object *dir, + struct inode *parent_dir, const char *name, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, + struct thandle *th) { struct inode *inode = dir->oo_inode; struct osd_thandle *oth; @@ -3499,8 +3984,8 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, } result = osd_add_dot_dotdot_internal(info, dir->oo_inode, - parent_dir, dot_fid, - dot_dot_fid, oth); + parent_dir, dot_fid, + dot_dot_fid, oth); if (result == 0) dir->oo_compat_dotdot_created = 1; } @@ -3514,8 +3999,8 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, * value, return by respective functions. */ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, - struct inode *cinode, const char *name, - const struct dt_rec *fid, struct thandle *th) + struct inode *cinode, const char *name, + const struct lu_fid *fid, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); struct htree_lock *hlock; @@ -3530,9 +4015,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, pobj->oo_inode, 0); } else { down_write(&pobj->oo_ext_idx_sem); - } - rc = osd_add_dot_dotdot(info, pobj, cinode, name, - (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), + } + + rc = osd_add_dot_dotdot(info, pobj, cinode, name, + lu_object_fid(&pobj->oo_dt.do_lu), fid, th); } else { if (hlock != NULL) { @@ -3545,11 +4031,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; - *tfid = *(const struct lu_fid *)fid; + *tfid = *fid; tfid->f_ver = ~0; rc = __osd_ea_add_rec(info, pobj, cinode, name, - (const struct dt_rec *)tfid, - hlock, th); + tfid, hlock, th); } else { rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); @@ -3578,7 +4063,7 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, RETURN_EXIT; again: - rc = osd_oi_lookup(oti, dev, fid, id, true); + rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD); if (rc != 0 && rc != -ENOENT) RETURN_EXIT; @@ -3598,13 +4083,11 @@ again: } if (!dev->od_noscrub && ++once == 1) { - CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n", - PFID(fid)); rc = osd_scrub_start(dev); - LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID - ", rc = %d [2]\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, - PFID(fid), rc); + LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID + ", rc = %d [2]\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, + PFID(fid), rc); if (rc == 0) goto again; } @@ -3637,6 +4120,18 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, return rc; } +int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, + struct osd_inode_id *id, const struct lu_fid *fid) +{ + CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid), + id->oii_ino, id->oii_gen, info); + info->oti_cache.oic_lid = *id; + info->oti_cache.oic_fid = *fid; + info->oti_cache.oic_dev = osd; + + return 0; +} + /** * Calls ->lookup() to find dentry. From dentry get inode and * read inode's ea to get fid. This is required for interoperability @@ -3671,7 +4166,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, down_read(&obj->oo_ext_idx_sem); } - bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (bh) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -3695,13 +4190,18 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, rc = osd_ea_fid_get(env, obj, ino, fid, id); else osd_id_gen(id, ino, OSD_OII_NOGEN); - if (rc != 0 || osd_remote_fid(env, dev, fid)) { + if (rc != 0) { fid_zero(&oic->oic_fid); GOTO(out, rc); } - oic->oic_lid = *id; - oic->oic_fid = *fid; + if (osd_remote_fid(env, dev, fid)) + GOTO(out, rc = 0); + + rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, + fid); + if (rc != 0) + GOTO(out, rc); if ((scrub->os_pos_current <= ino) && ((sf->sf_flags & SF_INCONSISTENT) || (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || @@ -3749,15 +4249,17 @@ struct osd_object *osd_object_find(const struct lu_env *env, * in the cache, otherwise lu_object_alloc() crashes * -bzzz */ - luch = lu_object_find_at(env, ludev, 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" + 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) { @@ -3772,10 +4274,11 @@ struct osd_object *osd_object_find(const struct lu_env *env, lu_object_put(env, luch); child = ERR_PTR(-ENOENT); } - } else - child = (void *)luch; + } else { + child = ERR_CAST(luch); + } - return child; + return child; } /** @@ -3801,7 +4304,7 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -3810,21 +4313,14 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT]); - if (osd_dt_obj(dt)->oo_inode == NULL) { - const char *name = (const char *)key; - /* Object is not being created yet. Only happens when - * 1. declare directory create - * 2. declare insert . - * 3. declare insert .. - */ - LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0); - } else { + if (osd_dt_obj(dt)->oo_inode != NULL) { struct inode *inode = osd_dt_obj(dt)->oo_inode; /* We ignore block quota on meta pool (MDTs), so needn't * calculate how many blocks will be consumed by this index * insert */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, + rc = osd_declare_inode_qid(env, i_uid_read(inode), + i_gid_read(inode), 0, oh, true, true, NULL, false); } @@ -3865,7 +4361,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, { struct osd_object *obj = osd_dt_obj(dt); struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct lu_fid *fid = (struct lu_fid *) rec; + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; const char *name = (const char *)key; struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -3874,14 +4371,14 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, int rc; ENTRY; - LASSERT(osd_invariant(obj)); + LASSERT(osd_invariant(obj)); LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(th != NULL); + LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) + RETURN(-EACCES); LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid)); @@ -3901,18 +4398,18 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, /* If parent on remote MDT, we need put this object * under AGENT */ oh = container_of(th, typeof(*oh), ot_super); - rc = osd_add_to_agent(env, osd, obj, oh); + rc = osd_add_to_remote_parent(env, osd, obj, oh); if (rc != 0) { - CERROR("%s: add agent "DFID" error: rc = %d\n", + CERROR("%s: add "DFID" error: rc = %d\n", osd_name(osd), PFID(lu_object_fid(&dt->do_lu)), rc); RETURN(rc); } - child_inode = igrab(omm->omm_agent_dentry->d_inode); + child_inode = igrab(omm->omm_remote_parent->d_inode); } else { - child_inode = osd_create_remote_inode(env, osd, obj, - fid, th); + child_inode = osd_create_local_agent_inode(env, osd, + obj, fid, rec1->rec_type & S_IFMT, th); if (IS_ERR(child_inode)) RETURN(PTR_ERR(child_inode)); } @@ -3923,13 +4420,16 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, 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_inode)); - RETURN(PTR_ERR(child_inode)); + (int)PTR_ERR(child)); + RETURN(PTR_ERR(child)); } child_inode = igrab(child->oo_inode); } - rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th); + rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th); + + CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n", + obj->oo_inode->i_ino, name, child_inode->i_ino, rc); iput(child_inode); if (child != NULL) @@ -4081,12 +4581,12 @@ osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type) { /* check if file type is required */ if (ent->lde_attrs & LUDA_TYPE) { - int align = sizeof(struct luda_type) - 1; struct luda_type *lt; + int align = sizeof(*lt) - 1; len = (len + align) & ~align; lt = (struct luda_type *)(ent->lde_name + len); - lt->lt_type = cpu_to_le16(CFS_DTTOIF(type)); + lt->lt_type = cpu_to_le16(DTTOIF(type)); } ent->lde_attrs = cpu_to_le32(ent->lde_attrs); @@ -4107,6 +4607,7 @@ osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset, ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr)); strncpy(ent->lde_name, name, namelen); + ent->lde_name[namelen] = '\0'; ent->lde_namelen = cpu_to_le16(namelen); /* append lustre attributes */ @@ -4228,6 +4729,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *info = osd_oti_get(env); struct osd_it_ea *it = &info->oti_it_ea; + struct file *file = &it->oie_file; struct lu_object *lo = &dt->do_lu; struct dentry *obj_dentry = &info->oti_it_dentry; ENTRY; @@ -4242,17 +4744,22 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, it->oie_dirent = NULL; it->oie_buf = info->oti_it_ea_buf; it->oie_obj = obj; - it->oie_file.f_pos = 0; - it->oie_file.f_dentry = obj_dentry; - if (attr & LUDA_64BITHASH) - it->oie_file.f_mode |= FMODE_64BITHASH; - else - it->oie_file.f_mode |= FMODE_32BITHASH; - it->oie_file.f_mapping = obj->oo_inode->i_mapping; - it->oie_file.f_op = obj->oo_inode->i_fop; - it->oie_file.private_data = NULL; - lu_object_get(lo); - RETURN((struct dt_it *) it); + + /* Reset the "file" totally to avoid to reuse any old value from + * former readdir handling, the "file->f_pos" should be zero. */ + memset(file, 0, sizeof(*file)); + /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */ + if (attr & LUDA_64BITHASH) + file->f_mode = FMODE_64BITHASH; + else + file->f_mode = FMODE_32BITHASH; + file->f_dentry = obj_dentry; + file->f_mapping = obj->oo_inode->i_mapping; + file->f_op = obj->oo_inode->i_fop; + set_file_inode(file, obj->oo_inode); + + lu_object_get(lo); + RETURN((struct dt_it *) it); } /** @@ -4304,6 +4811,12 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di) { } +struct osd_filldir_cbs { +#ifdef HAVE_DIR_CONTEXT + struct dir_context ctx; +#endif + struct osd_it_ea *it; +}; /** * It is called internally by ->readdir(). It fills the * iterator's in-memory data structure with required @@ -4315,11 +4828,11 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di) * \retval 0 on success * \retval 1 on buffer full */ -static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, +static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { - struct osd_it_ea *it = (struct osd_it_ea *)buf; + 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; @@ -4374,6 +4887,7 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, * * \retval 0 on success * \retval -ve on error + * \retval +1 reach the end of entry */ static int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di) @@ -4382,7 +4896,14 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, struct osd_object *obj = it->oie_obj; struct inode *inode = obj->oo_inode; struct htree_lock *hlock = NULL; - int result = 0; + struct file *filp = &it->oie_file; + int rc = 0; + struct osd_filldir_cbs buf = { +#ifdef HAVE_DIR_CONTEXT + .ctx.actor = osd_ldiskfs_filldir, +#endif + .it = it + }; ENTRY; it->oie_dirent = it->oie_buf; @@ -4396,22 +4917,31 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, down_read(&obj->oo_ext_idx_sem); } - result = inode->i_fop->readdir(&it->oie_file, it, - (filldir_t) osd_ldiskfs_filldir); +#ifdef HAVE_DIR_CONTEXT + buf.ctx.pos = filp->f_pos; + rc = inode->i_fop->iterate(filp, &buf.ctx); + filp->f_pos = buf.ctx.pos; +#else + rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir); +#endif if (hlock != NULL) ldiskfs_htree_unlock(hlock); else up_read(&obj->oo_ext_idx_sem); - if (it->oie_rd_dirent == 0) { - result = -EIO; - } else { - it->oie_dirent = it->oie_buf; - it->oie_it_dirent = 1; - } + if (it->oie_rd_dirent == 0) { + /*If it does not get any dirent, it means it has been reached + *to the end of the dir */ + it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file); + if (rc == 0) + rc = 1; + } else { + it->oie_dirent = it->oie_buf; + it->oie_it_dirent = 1; + } - RETURN(result); + RETURN(rc); } /** @@ -4491,22 +5021,12 @@ osd_dirent_update(handle_t *jh, struct super_block *sb, LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack)); rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) { - CERROR("%.16s: fail to write access for update dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); rc = ldiskfs_journal_dirty_metadata(jh, bh); - if (rc != 0) - CERROR("%.16s: fail to dirty metadata for update dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } @@ -4521,6 +5041,19 @@ osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize) return 0; } +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_DIR_REC_LEN(de) >= + __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack))) + return 1; + else + return 0; +} + static int osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, struct inode *dir, struct inode *inode, @@ -4542,13 +5075,8 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, if (osd_dirent_has_space(de->rec_len, ent->oied_namelen, dir->i_sb->s_blocksize)) { rc = ldiskfs_journal_get_write_access(jh, bh); - if (rc != 0) { - CERROR("%.16s: fail to write access for reinsert " - "dirent: name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } de->name[de->name_len] = 0; rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); @@ -4557,38 +5085,29 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, de->file_type |= LDISKFS_DIRENT_LUFID; rc = ldiskfs_journal_dirty_metadata(jh, bh); - if (rc != 0) - CERROR("%.16s: fail to dirty metadata for reinsert " - "dirent: name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } rc = ldiskfs_delete_entry(jh, dir, de, bh); - if (rc != 0) { - CERROR("%.16s: fail to delete entry for reinsert dirent: " - "name = %.*s, rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + if (rc != 0) RETURN(rc); - } dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name, ent->oied_namelen); ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp; - osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid); + osd_get_ldiskfs_dirent_param(ldp, fid); dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); /* It is too bad, we cannot reinsert the name entry back. * That means we lose it! */ if (rc != 0) - CERROR("%.16s: fail to insert entry for reinsert dirent: " - "name = %.*s, rc = %d\n", + CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, " + "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, PFID(fid), rc); RETURN(rc); } @@ -4614,18 +5133,15 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, struct inode *inode; int credits; int rc; + int dot_dotdot = 0; bool dirty = false; - bool is_dotdot = false; ENTRY; if (ent->oied_name[0] == '.') { - /* Skip dot entry, even if it has stale FID-in-dirent, because - * we do not use such FID-in-dirent anymore, it is harmless. */ if (ent->oied_namelen == 1) - RETURN(0); - - if (ent->oied_namelen == 2 && ent->oied_name[1] == '.') - is_dotdot = true; + dot_dotdot = 1; + else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.') + dot_dotdot = 2; } dentry = osd_child_dentry_get(env, obj, ent->oied_name, @@ -4649,35 +5165,46 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, again: if (dev->od_dirent_journal) { - jh = ldiskfs_journal_start_sb(sb, credits); + jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); - CERROR("%.16s: fail to start trans for dirent " - "check_repair: credits %d, name %.*s, rc %d\n", - devname, credits, ent->oied_namelen, - ent->oied_name, rc); + CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent " + "check_repair, dir = %lu/%u, credits = %d, " + "name = %.*s: rc = %d\n", + devname, dir->i_ino, dir->i_generation, credits, + ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } - } - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, - LDISKFS_HLOCK_DEL); + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + /* "0" means exclusive lock for the whole directory. + * We need to prevent others access such name entry + * during the delete + insert. Neither HLOCK_ADD nor + * HLOCK_DEL cannot guarantee the atomicity. */ + ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0); + } else { + down_write(&obj->oo_ext_idx_sem); + } } else { - down_write(&obj->oo_ext_idx_sem); + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, + LDISKFS_HLOCK_LOOKUP); + } else { + down_read(&obj->oo_ext_idx_sem); + } } - bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock); - /* For dotdot entry, if there is not enough space to hold FID-in-dirent, - * just keep it there. It only happens when the device upgraded from 1.8 - * or restored from MDT file-level backup. For the whole directory, only - * dotdot entry has no FID-in-dirent and needs to get FID from LMA when - * readdir, it will not affect the performance much. */ + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + /* For dot/dotdot entry, if there is not enough space to hold the + * FID-in-dirent, just keep them there. It only happens when the + * device upgraded from 1.8 or restored from MDT file-level backup. + * 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) || - (is_dotdot && !osd_dirent_has_space(de->rec_len, - ent->oied_namelen, - sb->s_blocksize))) { + (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) { *attr |= LUDA_IGNORE; GOTO(out_journal, rc = 0); } @@ -4689,13 +5216,25 @@ again: if (rc == -ENOENT || rc == -ESTALE) { *attr |= LUDA_IGNORE; 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); } GOTO(out_journal, rc); } + /* skip the REMOTE_PARENT_DIR. */ + if (inode == dev->od_mdt_map->omm_remote_parent->d_inode) + GOTO(out_inode, rc = 0); + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc == 0) { + LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI)); + if (fid_is_sane(fid)) { /* FID-in-dirent is valid. */ if (lu_fid_eq(fid, &lma->lma_self_fid)) @@ -4713,7 +5252,7 @@ again: if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - up_write(&obj->oo_ext_idx_sem); + up_read(&obj->oo_ext_idx_sem); dev->od_dirent_journal = 1; goto again; } @@ -4724,9 +5263,17 @@ again: rc = osd_dirent_update(jh, sb, ent, fid, bh, de); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to update FID " + "in the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { + *fid = lma->lma_self_fid; *attr |= LUDA_REPAIR; GOTO(out_inode, rc = 0); } @@ -4737,7 +5284,7 @@ again: if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - up_write(&obj->oo_ext_idx_sem); + up_read(&obj->oo_ext_idx_sem); dev->od_dirent_journal = 1; goto again; } @@ -4749,14 +5296,24 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to append FID " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } else if (rc == -ENODATA) { /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { - if (fid_is_sane(fid)) + if (fid_is_sane(fid)) { *attr |= LUDA_REPAIR; - else + } else { + lu_igif_build(fid, inode->i_ino, + inode->i_generation); *attr |= LUDA_UPGRADE; + } GOTO(out_inode, rc = 0); } @@ -4766,7 +5323,7 @@ again: if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - up_write(&obj->oo_ext_idx_sem); + up_read(&obj->oo_ext_idx_sem); dev->od_dirent_journal = 1; goto again; } @@ -4775,9 +5332,16 @@ again: if (unlikely(fid_is_sane(fid))) { /* FID-in-dirent exists, but FID-in-LMA is lost. * Trust the FID-in-dirent, and add FID-in-LMA. */ - rc = osd_ea_fid_set(info, inode, fid); + rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to set LMA for " + "update dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { lu_igif_build(fid, inode->i_ino, inode->i_generation); /* It is probably IGIF object. Only aappend the @@ -4786,6 +5350,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_UPGRADE; + else + CDEBUG(D_LFSCK, "%.16s: fail to append IGIF " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } @@ -4796,10 +5367,14 @@ out_inode: out_journal: brelse(bh); - if (hlock != NULL) + if (hlock != NULL) { ldiskfs_htree_unlock(hlock); - else - up_write(&obj->oo_ext_idx_sem); + } else { + if (dev->od_dirent_journal) + up_write(&obj->oo_ext_idx_sem); + else + up_read(&obj->oo_ext_idx_sem); + } if (jh != NULL) ldiskfs_journal_stop(jh); if (rc >= 0 && !dirty) @@ -4840,14 +5415,17 @@ static inline int osd_it_ea_rec(const struct lu_env *env, if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { attr |= LUDA_IGNORE; rc = 0; - goto pack; + } else { + rc = osd_dirent_check_repair(env, obj, it, fid, id, + &attr); } - - rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { - if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) && + likely(it->oie_dirent->oied_namelen != 2 || + it->oie_dirent->oied_name[0] != '.' || + it->oie_dirent->oied_name[1] != '.')) RETURN(-ENOENT); rc = osd_ea_fid_get(env, obj, ino, fid, id); @@ -4856,22 +5434,20 @@ static inline int osd_it_ea_rec(const struct lu_env *env, } } - if (rc < 0) - RETURN(rc); - -pack: + /* Pack the entry anyway, at least the offset is right. */ osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off, it->oie_dirent->oied_name, it->oie_dirent->oied_namelen, it->oie_dirent->oied_type, attr); + if (rc < 0) + RETURN(rc); + if (osd_remote_fid(env, dev, fid)) RETURN(0); - if (likely(!(attr & LUDA_IGNORE))) { - oic->oic_lid = *id; - oic->oic_fid = *fid; - } + if (likely(!(attr & LUDA_IGNORE))) + rc = osd_add_oi_cache(oti, dev, id, fid); if (!(attr & LUDA_VERIFY) && (scrub->os_pos_current <= ino) && @@ -4884,6 +5460,27 @@ pack: } /** + * Returns the record size size at current position. + * + * This function will return record(lu_dirent) size in bytes. + * + * \param[in] env execution environment + * \param[in] di iterator's in memory structure + * \param[in] attr attribute of the entry, only requires LUDA_TYPE to + * calculate the lu_dirent size. + * + * \retval record size(in bytes & in memory) of the current lu_dirent + * entry. + */ +static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di, + __u32 attr) +{ + struct osd_it_ea *it = (struct osd_it_ea *)di; + + return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr); +} + +/** * Returns a cookie for current position of the iterator head, so that * user can use this cookie to load/start the iterator next time. * @@ -4918,6 +5515,9 @@ static int osd_it_ea_load(const struct lu_env *env, it->oie_file.f_pos = hash; rc = osd_ldiskfs_it_fill(env, di); + if (rc > 0) + rc = -ENODATA; + if (rc == 0) rc = +1; @@ -4958,23 +5558,24 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826) */ static const struct dt_index_operations osd_index_ea_ops = { - .dio_lookup = osd_index_ea_lookup, - .dio_declare_insert = osd_index_declare_ea_insert, - .dio_insert = osd_index_ea_insert, - .dio_declare_delete = osd_index_declare_ea_delete, - .dio_delete = osd_index_ea_delete, - .dio_it = { - .init = osd_it_ea_init, - .fini = osd_it_ea_fini, - .get = osd_it_ea_get, - .put = osd_it_ea_put, - .next = osd_it_ea_next, - .key = osd_it_ea_key, - .key_size = osd_it_ea_key_size, - .rec = osd_it_ea_rec, - .store = osd_it_ea_store, - .load = osd_it_ea_load - } + .dio_lookup = osd_index_ea_lookup, + .dio_declare_insert = osd_index_declare_ea_insert, + .dio_insert = osd_index_ea_insert, + .dio_declare_delete = osd_index_declare_ea_delete, + .dio_delete = osd_index_ea_delete, + .dio_it = { + .init = osd_it_ea_init, + .fini = osd_it_ea_fini, + .get = osd_it_ea_get, + .put = osd_it_ea_put, + .next = osd_it_ea_next, + .key = osd_it_ea_key, + .key_size = osd_it_ea_key_size, + .rec = osd_it_ea_rec, + .rec_size = osd_it_ea_rec_size, + .store = osd_it_ea_store, + .load = osd_it_ea_load + } }; static void *osd_key_init(const struct lu_context *ctx, @@ -5008,12 +5609,14 @@ static void *osd_key_init(const struct lu_context *ctx, static void osd_key_fini(const struct lu_context *ctx, struct lu_context_key *key, void* data) { - struct osd_thread_info *info = data; + struct osd_thread_info *info = data; - if (info->oti_hlock != NULL) - ldiskfs_htree_lock_free(info->oti_hlock); - OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - OBD_FREE_PTR(info); + 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); + OBD_FREE_PTR(info); } static void osd_key_exit(const struct lu_context *ctx, @@ -5042,30 +5645,79 @@ static int osd_device_init(const struct lu_env *env, struct lu_device *d, { struct osd_device *osd = osd_dev(d); - strncpy(osd->od_svname, name, MAX_OBD_NAME); + if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) + >= sizeof(osd->od_svname)) + return -E2BIG; return osd_procfs_init(osd, name); } -static int osd_shutdown(const struct lu_env *env, struct osd_device *o) +static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) { + struct seq_server_site *ss = osd_seq_site(osd); + int rc; ENTRY; - osd_scrub_cleanup(env, o); + if (osd->od_is_ost || osd->od_cl_seq != NULL) + RETURN(0); - if (o->od_fsops) { - fsfilt_put_ops(o->od_fsops); - o->od_fsops = NULL; + if (unlikely(ss == NULL)) + RETURN(-ENODEV); + + OBD_ALLOC_PTR(osd->od_cl_seq); + if (osd->od_cl_seq == NULL) + RETURN(-ENOMEM); + + 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); +} + +static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd) +{ + if (osd->od_cl_seq == NULL) + return; + + seq_client_fini(osd->od_cl_seq); + OBD_FREE_PTR(osd->od_cl_seq); + osd->od_cl_seq = NULL; +} + +static int osd_shutdown(const struct lu_env *env, struct osd_device *o) +{ + ENTRY; + /* shutdown quota slave instance associated with the device */ if (o->od_quota_slave != NULL) { qsd_fini(env, o->od_quota_slave); o->od_quota_slave = NULL; } + osd_fid_fini(env, o); + RETURN(0); } +static void osd_umount(const struct lu_env *env, struct osd_device *o) +{ + ENTRY; + + if (o->od_mnt != NULL) { + shrink_dcache_sb(osd_sb(o)); + osd_sync(env, &o->od_dt_dev); + + mntput(o->od_mnt); + o->od_mnt = NULL; + } + + EXIT; +} + static int osd_mount(const struct lu_env *env, struct osd_device *o, struct lustre_cfg *cfg) { @@ -5077,7 +5729,10 @@ static int osd_mount(const struct lu_env *env, struct file_system_type *type; char *options = NULL; char *str; - int rc = 0; + struct osd_thread_info *info = osd_oti_get(env); + struct lu_fid *fid = &info->oti_fid; + struct inode *inode; + int rc = 0; ENTRY; if (o->od_mnt != NULL) @@ -5087,13 +5742,7 @@ static int osd_mount(const struct lu_env *env, RETURN(-E2BIG); strcpy(o->od_mntdev, dev); - o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS)); - if (o->od_fsops == NULL) { - CERROR("Can't find fsfilt_ldiskfs\n"); - RETURN(-ENOTSUPP); - } - - OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD); + OBD_PAGE_ALLOC(__page, GFP_IOFS); if (__page == NULL) GOTO(out, rc = -ENOMEM); @@ -5103,7 +5752,7 @@ static int osd_mount(const struct lu_env *env, if (str) lmd_flags = simple_strtoul(str + 1, NULL, 0); opts = lustre_cfg_string(cfg, 3); - page = (unsigned long)cfs_page_address(__page); + page = (unsigned long)page_address(__page); options = (char *)page; *options = '\0'; if (opts == NULL) @@ -5114,7 +5763,7 @@ static int osd_mount(const struct lu_env *env, /* Glom up mount options */ if (*options != '\0') strcat(options, ","); - strlcat(options, "no_mbcache", CFS_PAGE_SIZE); + strlcat(options, "no_mbcache", PAGE_CACHE_SIZE); type = get_fs_type("ldiskfs"); if (!type) { @@ -5123,70 +5772,76 @@ static int osd_mount(const struct lu_env *env, } o->od_mnt = vfs_kern_mount(type, s_flags, dev, options); - cfs_module_put(type->owner); + module_put(type->owner); if (IS_ERR(o->od_mnt)) { rc = PTR_ERR(o->od_mnt); - CERROR("%s: can't mount %s: %d\n", name, dev, rc); o->od_mnt = NULL; + CERROR("%s: can't mount %s: %d\n", name, dev, rc); GOTO(out, rc); } - if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { +#ifdef HAVE_DEV_SET_RDONLY + if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { CERROR("%s: underlying device %s is marked as read-only. " "Setup failed\n", name, dev); - mntput(o->od_mnt); - o->od_mnt = NULL; - GOTO(out, rc = -EROFS); + GOTO(out_mnt, rc = -EROFS); } +#endif if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { + LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { CERROR("%s: device %s is mounted w/o journal\n", name, dev); - mntput(o->od_mnt); - o->od_mnt = NULL; - GOTO(out, rc = -EINVAL); + GOTO(out_mnt, rc = -EINVAL); + } + +#ifdef LDISKFS_MOUNT_DIRDATA + if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb, + LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA; + else if (!o->od_is_ost) + CWARN("%s: device %s was upgraded from Lustre-1.x without " + "enabling the dirdata feature. If you do not want to " + "downgrade to Lustre-1.x again, you can enable it via " + "'tune2fs -O dirdata device'\n", name, dev); +#endif + inode = osd_sb(o)->s_root->d_inode; + lu_local_obj_fid(fid, OSD_FS_ROOT_OID); + rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0); + if (rc != 0) { + CERROR("%s: failed to set lma on %s root inode\n", name, dev); + GOTO(out_mnt, rc); } - ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode, - LDISKFS_STATE_LUSTRE_NO_OI); if (lmd_flags & LMD_FLG_NOSCRUB) o->od_noscrub = 1; + GOTO(out, rc = 0); + +out_mnt: + mntput(o->od_mnt); + o->od_mnt = NULL; + out: if (__page) OBD_PAGE_FREE(__page); - if (rc) - fsfilt_put_ops(o->od_fsops); - RETURN(rc); + return rc; } static struct lu_device *osd_device_fini(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - int rc; - ENTRY; - - rc = osd_shutdown(env, osd_dev(d)); - - osd_obj_map_fini(osd_dev(d)); - - shrink_dcache_sb(osd_sb(osd_dev(d))); - osd_sync(env, lu2dt_dev(d)); - - rc = osd_procfs_fini(osd_dev(d)); - if (rc) { - CERROR("proc fini error %d \n", rc); - RETURN (ERR_PTR(rc)); - } + struct osd_device *o = osd_dev(d); + ENTRY; - if (osd_dev(d)->od_mnt) { - mntput(osd_dev(d)->od_mnt); - osd_dev(d)->od_mnt = NULL; - } + osd_shutdown(env, o); + osd_procfs_fini(o); + osd_scrub_cleanup(env, o); + osd_obj_map_fini(o); + osd_umount(env, o); - RETURN(NULL); + RETURN(NULL); } static int osd_device_init0(const struct lu_env *env, @@ -5196,6 +5851,7 @@ static int osd_device_init0(const struct lu_env *env, struct lu_device *l = osd2lu_dev(o); struct osd_thread_info *info; int rc; + int cplen = 0; /* if the module was re-loaded, env can loose its keys */ rc = lu_env_refill((struct lu_env *) env); @@ -5209,7 +5865,6 @@ static int osd_device_init0(const struct lu_env *env, spin_lock_init(&o->od_osfs_lock); mutex_init(&o->od_otable_mutex); - o->od_osfs_age = cfs_time_shift_64(-1000); o->od_capa_hash = init_capa_hash(); if (o->od_capa_hash == NULL) @@ -5219,37 +5874,47 @@ static int osd_device_init0(const struct lu_env *env, o->od_writethrough_cache = 1; o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; - rc = osd_mount(env, o, cfg); - if (rc) + cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), + sizeof(o->od_svname)); + if (cplen >= sizeof(o->od_svname)) { + rc = -E2BIG; GOTO(out_capa, rc); + } - CFS_INIT_LIST_HEAD(&o->od_ios_list); - /* setup scrub, including OI files initialization */ - rc = osd_scrub_setup(env, o); - if (rc < 0) - GOTO(out_mnt, rc); + if (server_name_is_ost(o->od_svname)) + o->od_is_ost = 1; - strncpy(o->od_svname, lustre_cfg_string(cfg, 4), - sizeof(o->od_svname) - 1); + rc = osd_mount(env, o, cfg); + if (rc != 0) + GOTO(out_capa, rc); - rc = osd_obj_map_init(o); + rc = osd_obj_map_init(env, o); if (rc != 0) - GOTO(out_scrub, rc); + GOTO(out_mnt, rc); rc = lu_site_init(&o->od_site, l); - if (rc) + if (rc != 0) GOTO(out_compat, rc); o->od_site.ls_bottom_dev = l; rc = lu_site_init_finish(&o->od_site); - if (rc) + if (rc != 0) + GOTO(out_site, rc); + + /* self-repair LMA by default */ + o->od_lma_self_repair = 1; + + INIT_LIST_HEAD(&o->od_ios_list); + /* setup scrub, including OI files initialization */ + rc = osd_scrub_setup(env, o); + if (rc < 0) GOTO(out_site, rc); rc = osd_procfs_init(o, o->od_svname); if (rc != 0) { CERROR("%s: can't initialize procfs: rc = %d\n", o->od_svname, rc); - GOTO(out_site, rc); + GOTO(out_scrub, rc); } LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev); @@ -5264,23 +5929,21 @@ static int osd_device_init0(const struct lu_env *env, } RETURN(0); + out_procfs: osd_procfs_fini(o); +out_scrub: + osd_scrub_cleanup(env, o); out_site: lu_site_fini(&o->od_site); out_compat: osd_obj_map_fini(o); -out_scrub: - osd_scrub_cleanup(env, o); out_mnt: - osd_oi_fini(info, o); - osd_shutdown(env, o); - mntput(o->od_mnt); - o->od_mnt = NULL; + osd_umount(env, o); out_capa: cleanup_capa_hash(o->od_capa_hash); out: - RETURN(rc); + return rc; } static struct lu_device *osd_device_alloc(const struct lu_env *env, @@ -5333,23 +5996,33 @@ static struct lu_device *osd_device_free(const struct lu_env *env, static int osd_process_config(const struct lu_env *env, struct lu_device *d, struct lustre_cfg *cfg) { - struct osd_device *o = osd_dev(d); - int err; - ENTRY; + struct osd_device *o = osd_dev(d); + int rc; + ENTRY; - switch(cfg->lcfg_command) { - case LCFG_SETUP: - err = osd_mount(env, o, cfg); - break; - case LCFG_CLEANUP: + switch (cfg->lcfg_command) { + case LCFG_SETUP: + rc = osd_mount(env, o, cfg); + break; + case LCFG_CLEANUP: lu_dev_del_linkage(d->ld_site, d); - err = osd_shutdown(env, o); + rc = osd_shutdown(env, o); break; - default: - err = -ENOSYS; - } + case LCFG_PARAM: + LASSERT(&o->od_dt_dev); + rc = class_process_proc_seq_param(PARAM_OSD, + lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + if (rc > 0 || rc == -ENOSYS) + rc = class_process_proc_seq_param(PARAM_OST, + lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + break; + default: + rc = -ENOSYS; + } - RETURN(err); + RETURN(rc); } static int osd_recovery_complete(const struct lu_env *env, @@ -5427,22 +6100,26 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, int result = 0; ENTRY; - if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) { - /* MDT/MDD still use old infrastructure to create - * special files */ - result = llo_local_objects_setup(env, lu2md_dev(pdev), - lu2dt_dev(dev)); - if (result) + if (osd->od_quota_slave != NULL) { + /* set up quota slave objects */ + result = qsd_prepare(env, osd->od_quota_slave); + if (result != 0) RETURN(result); } - if (osd->od_quota_slave != NULL) - /* set up quota slave objects */ - result = qsd_prepare(env, osd->od_quota_slave); + result = osd_fid_init(env, osd); RETURN(result); } +int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, + struct lu_fid *fid, struct md_op_data *op_data) +{ + struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev); + + return seq_client_alloc_fid(env, osd->od_cl_seq, fid); +} + static const struct lu_object_operations osd_lu_obj_ops = { .loo_object_init = osd_object_init, .loo_object_delete = osd_object_delete, @@ -5486,22 +6163,35 @@ struct lu_device_type osd_device_type = { static struct obd_ops osd_obd_device_ops = { .o_owner = THIS_MODULE, .o_connect = osd_obd_connect, - .o_disconnect = osd_obd_disconnect + .o_disconnect = osd_obd_disconnect, + .o_fid_alloc = osd_fid_alloc, }; static int __init osd_mod_init(void) { - struct lprocfs_static_vars lvars; + int rc; + + osd_oi_mod_init(); + + rc = lu_kmem_init(ldiskfs_caches); + if (rc) + return rc; - osd_oi_mod_init(); - lprocfs_osd_init_vars(&lvars); - return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars, - LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); + rc = class_register_type(&osd_obd_device_ops, NULL, true, + lprocfs_osd_module_vars, +#ifndef HAVE_ONLY_PROCFS_SEQ + NULL, +#endif + LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); + if (rc) + lu_kmem_fini(ldiskfs_caches); + return rc; } static void __exit osd_mod_exit(void) { class_unregister_type(LUSTRE_OSD_LDISKFS_NAME); + lu_kmem_fini(ldiskfs_caches); } MODULE_AUTHOR("Sun Microsystems, Inc. ");