X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=3fd2e4cfd5b57ef13a4758c3c6ae656fab264463;hp=d2099561cb7dbf7cd832b8da84770f271422041f;hb=db68565b462d954dc50b0f638fea820991f54209;hpb=1f14f0da5098e9ad2fec1fa283659f48b8f827a7 diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index d209956..3fd2e4c 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, Whamcloud, Inc. + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -53,8 +53,6 @@ #include /* XATTR_{REPLACE,CREATE} */ #include -/* simple_mkdir() */ -#include /* * struct OBD_{ALLOC,FREE}*() @@ -63,24 +61,38 @@ #include /* struct ptlrpc_thread */ #include - -/* fid_is_local() */ #include +/* process_config */ +#include #include "osd_internal.h" -#include "osd_igif.h" +#include "osd_dynlocks.h" /* llo_* api support */ #include #include -#ifdef HAVE_LDISKFS_PDO int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, "ldiskfs with parallel directory operations"); -#else -int ldiskfs_pdo = 0; -#endif + +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[] = ".."; @@ -93,6 +105,21 @@ 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; +int osd_trans_declare_op2rb[] = { + [OSD_OT_ATTR_SET] = OSD_OT_ATTR_SET, + [OSD_OT_PUNCH] = OSD_OT_MAX, + [OSD_OT_XATTR_SET] = OSD_OT_XATTR_SET, + [OSD_OT_CREATE] = OSD_OT_DESTROY, + [OSD_OT_DESTROY] = OSD_OT_CREATE, + [OSD_OT_REF_ADD] = OSD_OT_REF_DEL, + [OSD_OT_REF_DEL] = OSD_OT_REF_ADD, + [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, +}; + static int osd_has_index(const struct osd_object *obj) { return obj->oo_dt.do_index_ops != NULL; @@ -150,41 +177,38 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, l = &mo->oo_dt.do_lu; dt_object_init(&mo->oo_dt, NULL, d); - if (osd_dev(d)->od_iop_mode) - mo->oo_dt.do_ops = &osd_obj_ea_ops; - else - mo->oo_dt.do_ops = &osd_obj_ops; - + mo->oo_dt.do_ops = &osd_obj_ea_ops; l->lo_ops = &osd_lu_obj_ops; - cfs_init_rwsem(&mo->oo_sem); - cfs_init_rwsem(&mo->oo_ext_idx_sem); - cfs_spin_lock_init(&mo->oo_guard); + init_rwsem(&mo->oo_sem); + init_rwsem(&mo->oo_ext_idx_sem); + spin_lock_init(&mo->oo_guard); return l; } else { return NULL; } } -static int osd_get_lma(struct inode *inode, struct dentry *dentry, - struct lustre_mdt_attrs *lma) +int osd_get_lma(struct osd_thread_info *info, struct inode *inode, + struct dentry *dentry, struct lustre_mdt_attrs *lma) { int rc; - dentry->d_inode = inode; - rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma, - 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; @@ -207,8 +231,8 @@ 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); @@ -216,7 +240,6 @@ 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)) { @@ -239,8 +262,9 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, return inode; } -struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, - struct osd_inode_id *id, struct lu_fid *fid) +static struct inode * +osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, + struct osd_inode_id *id, struct lu_fid *fid) { struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; struct inode *inode; @@ -250,11 +274,14 @@ struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, if (IS_ERR(inode)) return inode; - rc = osd_get_lma(inode, &info->oti_obj_dentry, lma); + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc == 0) { *fid = lma->lma_self_fid; } else if (rc == -ENODATA) { - LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation); + if (unlikely(inode == osd_sb(dev)->s_root->d_inode)) + lu_local_obj_fid(fid, OSD_FS_ROOT_OID); + else + lu_igif_build(fid, inode->i_ino, inode->i_generation); } else { iput(inode); inode = ERR_PTR(rc); @@ -262,37 +289,260 @@ struct inode *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(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; + + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY)) + RETURN(0); + + 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) @@ -306,7 +556,9 @@ 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)); @@ -323,8 +575,17 @@ 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; } @@ -337,23 +598,11 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, 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) != 0) - GOTO(out, result = 0); - /* Search order: 3. OI files. */ - result = osd_oi_lookup(info, dev, fid, id); + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result == -ENOENT) { if (!fid_is_norm(fid) || + fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); @@ -364,32 +613,64 @@ 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); @@ -397,10 +678,29 @@ trigger: obj->oo_inode = inode; LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); - if (dev->od_iop_mode) { - obj->oo_compat_dot_created = 1; - obj->oo_compat_dotdot_created = 1; - } + + 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; if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */ GOTO(out, result = 0); @@ -442,17 +742,17 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, LINVRNT(osd_invariant(obj)); + if (fid_is_otable_it(&l->lo_header->loh_fid)) { + obj->oo_dt.do_ops = &osd_obj_otable_it_ops; + l->lo_header->loh_attr |= LOHA_EXISTS; + return 0; + } + result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); obj->oo_dt.do_body_ops = &osd_body_ops_new; - if (result == 0) { - if (obj->oo_inode != NULL) { - osd_object_init0(obj); - } else if (fid_is_otable_it(&l->lo_header->loh_fid)) { - obj->oo_dt.do_ops = &osd_obj_otable_it_ops; - /* LFSCK iterator object is special without inode */ - l->lo_header->loh_attr |= LOHA_EXISTS; - } - } + if (result == 0 && obj->oo_inode != NULL) + osd_object_init0(obj); + LINVRNT(osd_invariant(obj)); return result; } @@ -530,8 +830,7 @@ static void osd_th_started(struct osd_thandle *oth) /** * Helper function to convert time interval to microseconds packed in - * long int (default time units for the counter in "stats" initialized - * by lu_time_init() ) + * long int. */ static long interval_to_usec(cfs_time_t start, cfs_time_t end) { @@ -602,12 +901,8 @@ static int osd_param_is_not_sane(const struct osd_device *dev, /* * Concurrency: shouldn't matter. */ -#ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD static void osd_trans_commit_cb(struct super_block *sb, - struct journal_callback *jcb, int error) -#else -static void osd_trans_commit_cb(struct journal_callback *jcb, int error) -#endif + struct ldiskfs_journal_cb_entry *jcb, int error) { struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); struct thandle *th = &oh->ot_super; @@ -630,29 +925,29 @@ static void osd_trans_commit_cb(struct journal_callback *jcb, int error) 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); + OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO); if (oh != NULL) { oh->ot_quota_trans = &oti->oti_quota_trans; memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); @@ -661,9 +956,19 @@ static struct thandle *osd_trans_create(const struct lu_env *env, th->th_result = 0; th->th_tags = LCT_TX_HANDLE; oh->ot_credits = 0; - oti->oti_dev = osd_dt_dev(d); + atomic_set(&th->th_refc, 1); + th->th_alloc_size = sizeof(*oh); + 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, + 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); } @@ -700,24 +1005,35 @@ 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", - oh->ot_declare_create, oh->ot_declare_create_cred, - oh->ot_declare_delete, oh->ot_declare_delete_cred, - oh->ot_declare_destroy, oh->ot_declare_destroy_cred); + 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", - oh->ot_declare_attr_set, oh->ot_declare_attr_set_cred, - oh->ot_declare_xattr_set, oh->ot_declare_xattr_set_cred); + oti->oti_declare_ops[OSD_OT_ATTR_SET], + oti->oti_declare_ops_cred[OSD_OT_ATTR_SET], + oti->oti_declare_ops[OSD_OT_XATTR_SET], + oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]); CWARN(" write: %u/%u, punch: %u/%u, quota %u/%u\n", - oh->ot_declare_write, oh->ot_declare_write_cred, - oh->ot_declare_punch, oh->ot_declare_punch_cred, - oh->ot_declare_quota, oh->ot_declare_quota_cred); + oti->oti_declare_ops[OSD_OT_WRITE], + oti->oti_declare_ops_cred[OSD_OT_WRITE], + oti->oti_declare_ops[OSD_OT_PUNCH], + oti->oti_declare_ops_cred[OSD_OT_PUNCH], + oti->oti_declare_ops[OSD_OT_QUOTA], + oti->oti_declare_ops_cred[OSD_OT_QUOTA]); CWARN(" insert: %u/%u, delete: %u/%u\n", - oh->ot_declare_insert, oh->ot_declare_insert_cred, - oh->ot_declare_delete, oh->ot_declare_destroy_cred); + 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]); CWARN(" ref_add: %u/%u, ref_del: %u/%u\n", - oh->ot_declare_ref_add, oh->ot_declare_ref_add_cred, - oh->ot_declare_ref_del, oh->ot_declare_ref_del_cred); + oti->oti_declare_ops[OSD_OT_REF_ADD], + oti->oti_declare_ops_cred[OSD_OT_REF_ADD], + oti->oti_declare_ops[OSD_OT_REF_DEL], + oti->oti_declare_ops_cred[OSD_OT_REF_DEL]); if (last_credits != oh->ot_credits && time_after(jiffies, last_printed + 60 * HZ)) { @@ -725,7 +1041,6 @@ 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. @@ -739,7 +1054,7 @@ 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; @@ -748,8 +1063,8 @@ 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 { @@ -759,10 +1074,33 @@ 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; + + if (ss == NULL) + RETURN(1); + + 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; @@ -787,7 +1125,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) * notice we don't do this in osd_trans_start() * as underlying transaction can change during truncate */ - osd_journal_callback_set(hdl, osd_trans_commit_cb, + ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, &oh->ot_jcb); LASSERT(oti->oti_txns == 1); @@ -805,24 +1143,24 @@ 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; + /* 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); + if (!rc) + rc = iobuf->dr_error; - RETURN(rc); + RETURN(rc); } static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) @@ -870,10 +1208,10 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) /* Release granted quota to master if necessary */ qi->lqi_id.qid_uid = uid; - qsd_adjust_quota(env, qsd, &qi->lqi_id, USRQUOTA); + qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA); qi->lqi_id.qid_uid = gid; - qsd_adjust_quota(env, qsd, &qi->lqi_id, GRPQUOTA); + qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA); } } } @@ -907,6 +1245,8 @@ 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. */ @@ -930,25 +1270,29 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, ksfs = &osd_oti_get(env)->oti_ksfs; } - cfs_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; - } + spin_lock(&osd->od_osfs_lock); + result = sb->s_op->statfs(sb->s_root, ksfs); + if (likely(result == 0)) { /* N.B. statfs can't really fail */ + statfs_pack(sfs, ksfs); + if (sb->s_flags & MS_RDONLY) + sfs->os_state = OS_STATE_READONLY; } - if (likely(result == 0)) - *sfs = osd->od_statfs; - cfs_spin_unlock(&osd->od_osfs_lock); + 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; } @@ -975,7 +1319,6 @@ static void osd_conf_get(const struct lu_env *env, /* * 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; @@ -1004,19 +1347,6 @@ static void osd_conf_get(const struct lu_env *env, } -/** - * Helper function to get and fill the buffer with input values. - */ -static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len) -{ - struct lu_buf *buf; - - buf = &osd_oti_get(env)->oti_buf; - buf->lb_buf = area; - buf->lb_len = len; - return buf; -} - /* * Concurrency: shouldn't matter. */ @@ -1054,14 +1384,31 @@ 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); } /* @@ -1160,7 +1507,7 @@ static void osd_object_read_lock(const struct lu_env *env, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_owner != env); - cfs_down_read_nested(&obj->oo_sem, role); + down_read_nested(&obj->oo_sem, role); LASSERT(obj->oo_owner == NULL); oti->oti_r_locks++; @@ -1175,7 +1522,7 @@ static void osd_object_write_lock(const struct lu_env *env, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_owner != env); - cfs_down_write_nested(&obj->oo_sem, role); + down_write_nested(&obj->oo_sem, role); LASSERT(obj->oo_owner == NULL); obj->oo_owner = env; @@ -1192,7 +1539,7 @@ static void osd_object_read_unlock(const struct lu_env *env, LASSERT(oti->oti_r_locks > 0); oti->oti_r_locks--; - cfs_up_read(&obj->oo_sem); + up_read(&obj->oo_sem); } static void osd_object_write_unlock(const struct lu_env *env, @@ -1207,7 +1554,7 @@ static void osd_object_write_unlock(const struct lu_env *env, LASSERT(oti->oti_w_locks > 0); oti->oti_w_locks--; obj->oo_owner = NULL; - cfs_up_write(&obj->oo_sem); + up_write(&obj->oo_sem); } static int osd_object_write_locked(const struct lu_env *env, @@ -1246,14 +1593,14 @@ static int capa_is_sane(const struct lu_env *env, RETURN(-ESTALE); } - cfs_spin_lock(&capa_lock); - for (i = 0; i < 2; i++) { - if (keys[i].lk_keyid == capa->lc_keyid) { - oti->oti_capa_key = keys[i]; - break; - } - } - cfs_spin_unlock(&capa_lock); + spin_lock(&capa_lock); + for (i = 0; i < 2; i++) { + if (keys[i].lk_keyid == capa->lc_keyid) { + oti->oti_capa_key = keys[i]; + break; + } + } + spin_unlock(&capa_lock); if (i == 2) { DEBUG_CAPA(D_ERROR, capa, "no matched capa key"); @@ -1276,48 +1623,50 @@ 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, @@ -1338,7 +1687,8 @@ static void osd_inode_getattr(const struct lu_env *env, { 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_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); @@ -1362,16 +1712,16 @@ static int osd_attr_get(const struct lu_env *env, { struct osd_object *obj = osd_dt_obj(dt); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LINVRNT(osd_invariant(obj)); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) return -EACCES; - cfs_spin_lock(&obj->oo_guard); - osd_inode_getattr(env, obj->oo_inode, attr); - cfs_spin_unlock(&obj->oo_guard); - return 0; + spin_lock(&obj->oo_guard); + osd_inode_getattr(env, obj->oo_inode, attr); + spin_unlock(&obj->oo_guard); + return 0; } static int osd_declare_attr_set(const struct lu_env *env, @@ -1397,8 +1747,8 @@ static int osd_declare_attr_set(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, attr_set, - osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET, + osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); if (attr == NULL || obj->oo_inode == NULL) RETURN(rc); @@ -1521,8 +1871,6 @@ static int osd_inode_setattr(const struct lu_env *env, bits = attr->la_valid; - LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */ - if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); if (bits & LA_CTIME) @@ -1548,7 +1896,7 @@ static int osd_inode_setattr(const struct lu_env *env, if (bits & LA_GID) inode->i_gid = attr->la_gid; if (bits & LA_NLINK) - inode->i_nlink = attr->la_nlink; + set_nlink(inode, attr->la_nlink); if (bits & LA_RDEV) inode->i_rdev = attr->la_rdev; @@ -1597,13 +1945,38 @@ static int osd_attr_set(const struct lu_env *env, int rc; LASSERT(handle != NULL); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_invariant(obj)); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) return -EACCES; - OSD_EXEC_OP(handle, attr_set); + osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); + + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) { + struct osd_thread_info *oti = osd_oti_get(env); + const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); + struct lu_fid *fid1 = &oti->oti_fid; + struct osd_inode_id *id = &oti->oti_id; + struct iam_path_descr *ipd; + struct iam_container *bag; + struct osd_thandle *oh; + int rc; + + fid_cpu_to_be(fid1, fid0); + memset(id, 1, sizeof(*id)); + bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev), + fid0)->oi_dir.od_container; + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); + + oh = container_of0(handle, struct osd_thandle, ot_super); + rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1, + (const struct iam_rec *)id, ipd); + osd_ipd_put(env, bag, ipd); + return(rc > 0 ? 0 : rc); + } inode = obj->oo_inode; ll_vfs_dq_init(inode); @@ -1612,12 +1985,12 @@ static int osd_attr_set(const struct lu_env *env, if (rc) return rc; - cfs_spin_lock(&obj->oo_guard); - rc = osd_inode_setattr(env, inode, attr); - cfs_spin_unlock(&obj->oo_guard); + spin_lock(&obj->oo_guard); + rc = osd_inode_setattr(env, inode, attr); + spin_unlock(&obj->oo_guard); if (!rc) - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); return rc; } @@ -1629,9 +2002,8 @@ 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); @@ -1664,7 +2036,10 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, * NB: don't need any lock because no contention at this * early stage */ inode->i_flags |= S_NOCMTIME; - inode->i_state |= I_LUSTRE_NOSCRUB; + + /* For new created object, it must be consistent, + * and it is unnecessary to scrub against it. */ + ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); obj->oo_inode = inode; result = 0; } else { @@ -1690,7 +2065,6 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj, { int result; struct osd_thandle *oth; - struct osd_device *osd = osd_obj2dev(obj); __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX)); LASSERT(S_ISDIR(attr->la_mode)); @@ -1698,16 +2072,7 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj, oth = container_of(th, struct osd_thandle, ot_super); LASSERT(oth->ot_handle->h_transaction != NULL); result = osd_mkfile(info, obj, mode, hint, th); - if (result == 0 && osd->od_iop_mode == 0) { - LASSERT(obj->oo_inode != NULL); - /* - * XXX uh-oh... call low-level iam function directly. - */ - result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4, - sizeof (struct osd_fid_pack), - oth->ot_handle); - } return result; } @@ -1771,16 +2136,16 @@ 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)); @@ -1791,7 +2156,8 @@ static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, * This inode should be marked dirty for i_rdev. Currently * that is done in the osd_attr_init(). */ - init_special_inode(obj->oo_inode, mode, attr->la_rdev); + init_special_inode(obj->oo_inode, obj->oo_inode->i_mode, + attr->la_rdev); } LINVRNT(osd_invariant(obj)); return result; @@ -1834,11 +2200,10 @@ 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; } @@ -1875,7 +2240,7 @@ 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; @@ -1912,7 +2277,7 @@ static int __osd_object_create(struct osd_thread_info *info, /* restore previous umask value */ current->fs->umask = umask; - return result; + return result; } /** @@ -1921,24 +2286,57 @@ 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); } -static int osd_declare_object_create(const struct lu_env *env, - struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *handle) +int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, + obd_seq seq, struct lu_seq_range *range) +{ + struct seq_server_site *ss = osd_seq_site(osd); + + 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(seq)) { + fld_range_set_mdt(range); + if (ss != NULL) + /* FIXME: If ss is NULL, it suppose not get lsr_index + * at all */ + range->lsr_index = ss->ss_node_id; + return 0; + } + + LASSERT(ss != NULL); + fld_range_set_any(range); + /* OSD will only do local fld lookup */ + return fld_local_lookup(env, ss->ss_server_fld, seq, range); +} + +/* + * Concurrency: no external locking is necessary. + */ +static int osd_declare_object_create(const struct lu_env *env, + struct dt_object *dt, + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *handle) { struct osd_thandle *oh; int rc; @@ -1949,24 +2347,24 @@ static int osd_declare_object_create(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, create, osd_dto_credits_noquota[DTO_OBJECT_CREATE]); - /* XXX: So far, only normal fid needs be inserted into the oi, - * things could be changed later. Revise following code then. */ - if (fid_is_norm(lu_object_fid(&dt->do_lu))) { + 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), OI_CHECK_FLD)) /* Reuse idle OI block may cause additional one OI block * to be changed. */ - OSD_DECLARE_OP(oh, insert, - osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); - } + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + /* If this is directory, then we expect . and .. to be inserted as * well. The one directory block always needs to be created for the * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap, * block), there is no danger of needing a tree for the first block. */ if (attr && S_ISDIR(attr->la_mode)) { - OSD_DECLARE_OP(oh, insert, - osd_dto_credits_noquota[DTO_WRITE_BASE]); - OSD_DECLARE_OP(oh, insert, 0); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_WRITE_BASE]); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0); } if (!attr) @@ -1974,6 +2372,9 @@ static int osd_declare_object_create(const struct lu_env *env, rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh, false, false, NULL, false); + if (rc != 0) + RETURN(rc); + RETURN(rc); } @@ -1991,7 +2392,7 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt, ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt)); + LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); @@ -2000,13 +2401,16 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt, * 'tune2fs -O quota' will take care of creating them */ RETURN(-EPERM); - OSD_EXEC_OP(th, create); + osd_trans_exec_op(env, th, OSD_OT_CREATE); + osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); result = __osd_object_create(info, obj, attr, hint, dof, th); if (result == 0) result = __osd_oi_insert(env, obj, fid, th); - LASSERT(ergo(result == 0, dt_object_exists(dt))); + LASSERT(ergo(result == 0, + dt_object_exists(dt) && !dt_object_remote(dt))); + LASSERT(osd_invariant(obj)); RETURN(result); } @@ -2030,14 +2434,12 @@ static int osd_declare_object_destroy(const struct lu_env *env, LASSERT(oh->ot_handle == NULL); LASSERT(inode); - OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_OBJECT_DELETE]); - /* XXX: So far, only normal fid needs to be inserted into the OI, - * so only normal fid needs to be removed from the OI also. - * Recycle idle OI leaf may cause additional three OI blocks + 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_DECLARE_OP(oh, destroy, fid_is_norm(lu_object_fid(&dt->do_lu)) ? - osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0); - + 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); @@ -2069,24 +2471,26 @@ 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. */ - cfs_mutex_lock(&inode->i_mutex); - if (S_ISDIR(inode->i_mode)) { - LASSERT(osd_inode_unlinked(inode) || - inode->i_nlink == 1); - cfs_spin_lock(&obj->oo_guard); - inode->i_nlink = 0; - cfs_spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); - } else { - LASSERT(osd_inode_unlinked(inode)); - } + if (S_ISDIR(inode->i_mode)) { + 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 inode "DFID": rc = %d\n", + osd_name(osd), PFID(fid), result); + } + spin_lock(&obj->oo_guard); + clear_nlink(inode); + spin_unlock(&obj->oo_guard); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + } - OSD_EXEC_OP(th, destroy); + osd_trans_exec_op(env, th, OSD_OT_DESTROY); - result = osd_oi_delete(osd_oti_get(env), osd, fid, th); - cfs_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) */ @@ -2098,35 +2502,6 @@ static int osd_object_destroy(const struct lu_env *env, } /** - * Helper function for osd_xattr_set() - */ -static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_buf *buf, const char *name, int fl) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_child_dentry; - int fs_flags = 0; - int rc; - - LASSERT(dt_object_exists(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL); - - if (fl & LU_XATTR_REPLACE) - fs_flags |= XATTR_REPLACE; - - if (fl & LU_XATTR_CREATE) - fs_flags |= XATTR_CREATE; - - ll_vfs_dq_init(inode); - dentry->d_inode = inode; - rc = inode->i_op->setxattr(dentry, name, buf->lb_buf, - buf->lb_len, fs_flags); - return rc; -} - -/** * 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 * of the inode. @@ -2136,18 +2511,46 @@ static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt, * * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here */ -static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_fid *fid) +int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, + const struct lu_fid *fid, __u32 compat, __u32 incompat) { - struct osd_thread_info *info = osd_oti_get(env); - struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs; + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + int rc; + ENTRY; - lustre_lma_init(mdt_attrs, fid); - lustre_lma_swab(mdt_attrs); - return __osd_xattr_set(env, dt, - osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs), - XATTR_NAME_LMA, LU_XATTR_CREATE); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) + RETURN(0); + + 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); + /* 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); } /** @@ -2159,18 +2562,21 @@ static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt, * its inmemory API. */ void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) + const struct dt_rec *fid) { - param->edp_magic = LDISKFS_LUFID_MAGIC; - param->edp_len = sizeof(struct lu_fid) + 1; + if (!fid_is_namespace_visible((const struct lu_fid *)fid) || + OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) { + param->edp_magic = 0; + return; + } - fid_cpu_to_be((struct lu_fid *)param->edp_data, - (struct lu_fid *)fid); + param->edp_magic = LDISKFS_LUFID_MAGIC; + param->edp_len = sizeof(struct lu_fid) + 1; + fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid); } /** - * Try to read the fid from inode ea into dt_rec, if return value - * i.e. rc is +ve, then we got fid, otherwise we will have to form igif + * Try to read the fid from inode ea into dt_rec. * * \param fid object fid. * @@ -2193,6 +2599,103 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, RETURN(0); } +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 osd_thandle *oth) +{ + struct ldiskfs_dentry_param *dot_ldp; + struct ldiskfs_dentry_param *dot_dot_ldp; + + dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; + osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); + + dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; + dot_ldp->edp_magic = 0; + return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, + dir, dot_ldp, dot_dot_ldp); +} + +/** + * Create an local agent inode for remote entry + */ +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, + struct thandle *th) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct inode *local; + struct osd_thandle *oh; + int rc; + ENTRY; + + LASSERT(th); + 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); + 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)); + } + + 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); + if (rc != 0) { + CERROR("%s: "DFID" add dot dotdot error: rc = %d\n", + osd_name(osd), PFID(fid), rc); + RETURN(ERR_PTR(rc)); + } + + RETURN(local); +} + +/** + * Delete local agent inode for remote entry + */ +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; + struct inode *inode; + ENTRY; + + id->oii_ino = le32_to_cpu(ino); + id->oii_gen = OSD_OII_NOGEN; + inode = osd_iget(oti, osd, id); + if (IS_ERR(inode)) { + CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd), + PFID(fid), id->oii_ino, id->oii_gen); + RETURN(PTR_ERR(inode)); + } + + clear_nlink(inode); + mark_inode_dirty(inode); + CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n", + osd_name(osd), PFID(fid), inode->i_ino); + iput(inode); + RETURN(0); +} + /** * OSD layer object create function for interoperability mode (b11826). * This is mostly similar to osd_object_create(). Only difference being, fid is @@ -2215,7 +2718,7 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, ENTRY; LASSERT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt)); + LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); @@ -2224,17 +2727,21 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, * 'tune2fs -O quota' will take care of creating them */ RETURN(-EPERM); - OSD_EXEC_OP(th, create); + osd_trans_exec_op(env, th, OSD_OT_CREATE); + osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); result = __osd_object_create(info, obj, attr, hint, dof, th); - /* objects under osd root shld have igif fid, so dont add fid EA */ - if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL) - result = osd_ea_fid_set(env, dt, 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))); + LASSERT(ergo(result == 0, + dt_object_exists(dt) && !dt_object_remote(dt))); LINVRNT(osd_invariant(obj)); RETURN(result); } @@ -2243,7 +2750,7 @@ static int osd_declare_object_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; /* it's possible that object doesn't exist yet */ LASSERT(handle != NULL); @@ -2251,7 +2758,8 @@ static int osd_declare_object_ref_add(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, ref_add, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_trans_declare_op(env, oh, OSD_OT_REF_ADD, + osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); return 0; } @@ -2262,39 +2770,41 @@ 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)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); - OSD_EXEC_OP(th, ref_add); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + + osd_trans_exec_op(env, th, OSD_OT_REF_ADD); - /* - * 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. - * - * 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. - */ - cfs_spin_lock(&obj->oo_guard); - inode->i_nlink++; - if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) { - if (inode->i_nlink >= LDISKFS_LINK_MAX || - inode->i_nlink == 2) - inode->i_nlink = 1; - } - LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); - cfs_spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); - LINVRNT(osd_invariant(obj)); + /* + * 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. + * + * This also has to properly handle the case of inodes with nlink == 0 + * in case they are being linked into the PENDING directory + */ + spin_lock(&obj->oo_guard); + ldiskfs_inc_count(oh->ot_handle, inode); + LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + spin_unlock(&obj->oo_guard); - return 0; + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + LINVRNT(osd_invariant(obj)); + + return rc; } static int osd_declare_object_ref_del(const struct lu_env *env, @@ -2303,13 +2813,14 @@ static int osd_declare_object_ref_del(const struct lu_env *env, { struct osd_thandle *oh; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, ref_del, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_trans_declare_op(env, oh, OSD_OT_REF_DEL, + osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); return 0; } @@ -2320,29 +2831,41 @@ 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)); - LASSERT(dt_object_exists(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); - OSD_EXEC_OP(th, ref_del); - - cfs_spin_lock(&obj->oo_guard); - LASSERT(inode->i_nlink > 0); - inode->i_nlink--; - /* 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) - inode->i_nlink++; - cfs_spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); - LINVRNT(osd_invariant(obj)); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); - return 0; + osd_trans_exec_op(env, th, OSD_OT_REF_DEL); + + spin_lock(&obj->oo_guard); + /* 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; + } + + ldiskfs_dec_count(oh->ot_handle, inode); + spin_unlock(&obj->oo_guard); + + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + LINVRNT(osd_invariant(obj)); + + return 0; } /* @@ -2380,14 +2903,13 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, return sizeof(dt_obj_version_t); } - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) return -EACCES; - dentry->d_inode = inode; - return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len); + return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -2403,9 +2925,10 @@ static int osd_declare_xattr_set(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, xattr_set, strcmp(name, XATTR_NAME_VERSION) == 0 ? - osd_dto_credits_noquota[DTO_ATTR_SET_BASE] : - osd_dto_credits_noquota[DTO_XATTR_SET]); + 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]); return 0; } @@ -2425,7 +2948,7 @@ 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); } /* @@ -2435,6 +2958,12 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int fl, struct thandle *handle, struct lustre_capa *capa) { + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + int fs_flags = 0; + ENTRY; + LASSERT(handle != NULL); /* version set is not real XATTR */ @@ -2449,8 +2978,18 @@ 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; - OSD_EXEC_OP(handle, xattr_set); - return __osd_xattr_set(env, dt, buf, name, fl); + 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; + + if (fl & LU_XATTR_CREATE) + fs_flags |= XATTR_CREATE; + + return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, + fs_flags); } /* @@ -2464,7 +3003,7 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, struct osd_thread_info *info = osd_oti_get(env); struct dentry *dentry = &info->oti_obj_dentry; - LASSERT(dt_object_exists(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)); @@ -2472,6 +3011,7 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, return -EACCES; dentry->d_inode = inode; + dentry->d_sb = inode->i_sb; return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); } @@ -2481,13 +3021,14 @@ static int osd_declare_xattr_del(const struct lu_env *env, { struct osd_thandle *oh; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, xattr_set, osd_dto_credits_noquota[DTO_XATTR_SET]); + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, + osd_dto_credits_noquota[DTO_XATTR_SET]); return 0; } @@ -2505,107 +3046,106 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, struct dentry *dentry = &info->oti_obj_dentry; int rc; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL); - LASSERT(osd_write_locked(env, obj)); LASSERT(handle != NULL); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) return -EACCES; - OSD_EXEC_OP(handle, xattr_set); + osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); ll_vfs_dq_init(inode); dentry->d_inode = inode; + dentry->d_sb = inode->i_sb; rc = inode->i_op->removexattr(dentry, name); return rc; } static struct obd_capa *osd_capa_get(const struct lu_env *env, - struct dt_object *dt, - struct lustre_capa *old, - __u64 opc) + struct 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)); - LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LINVRNT(osd_invariant(obj)); - /* renewal sanity check */ - if (old && osd_object_auth(env, dt, old, opc)) - RETURN(ERR_PTR(-EACCES)); - - 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 = 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)); + } - 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); + } - cfs_spin_lock(&capa_lock); - *key = dev->od_capa_keys[1]; - cfs_spin_unlock(&capa_lock); + spin_lock(&capa_lock); + *key = osd->od_capa_keys[1]; + spin_unlock(&capa_lock); - capa->lc_keyid = key->lk_keyid; - capa->lc_expiry = cfs_time_current_sec() + 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) @@ -2620,12 +3160,18 @@ 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); +#ifndef HAVE_FILE_FSYNC_4ARGS mutex_lock(&inode->i_mutex); - rc = file->f_op->fsync(file, dentry, 0); +#endif + rc = do_fsync(file, 0); +#ifndef HAVE_FILE_FSYNC_4ARGS mutex_unlock(&inode->i_mutex); +#endif RETURN(rc); } @@ -2703,17 +3249,15 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, int result; int skip_iam = 0; struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *osd = osd_obj2dev(obj); LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); if (osd_object_is_root(obj)) { dt->do_index_ops = &osd_index_ea_ops; result = 0; - } else if (feat == &dt_directory_features && osd->od_iop_mode) { + } else if (feat == &dt_directory_features) { dt->do_index_ops = &osd_index_ea_ops; - if (S_ISDIR(obj->oo_inode->i_mode)) + if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode)) result = 0; else result = -ENOTDIR; @@ -2721,7 +3265,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } else if (unlikely(feat == &dt_otable_features)) { dt->do_index_ops = &osd_otable_ops; return 0; - } else if (feat == &dt_acct_features) { + } else if (unlikely(feat == &dt_acct_features)) { dt->do_index_ops = &osd_acct_index_ops; result = 0; skip_iam = 1; @@ -2731,28 +3275,28 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, OBD_ALLOC_PTR(dir); if (dir != NULL) { - cfs_spin_lock(&obj->oo_guard); - if (obj->oo_dir == NULL) - obj->oo_dir = dir; - else - /* - * Concurrent thread allocated container data. - */ - OBD_FREE_PTR(dir); - cfs_spin_unlock(&obj->oo_guard); - /* - * Now, that we have container data, serialize its - * initialization. - */ - cfs_down_write(&obj->oo_ext_idx_sem); - /* - * recheck under lock. - */ - if (!osd_has_index(obj)) - result = osd_iam_container_init(env, obj, dir); - else - result = 0; - cfs_up_write(&obj->oo_ext_idx_sem); + spin_lock(&obj->oo_guard); + if (obj->oo_dir == NULL) + obj->oo_dir = dir; + else + /* + * Concurrent thread allocated container data. + */ + OBD_FREE_PTR(dir); + spin_unlock(&obj->oo_guard); + /* + * Now, that we have container data, serialize its + * initialization. + */ + down_write(&obj->oo_ext_idx_sem); + /* + * recheck under lock. + */ + if (!osd_has_index(obj)) + result = osd_iam_container_init(env, obj, dir); + else + result = 0; + up_write(&obj->oo_ext_idx_sem); } else { result = -ENOMEM; } @@ -2766,7 +3310,8 @@ 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; @@ -2860,7 +3405,8 @@ static int osd_index_declare_iam_delete(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]); + osd_trans_declare_op(env, oh, OSD_OT_DELETE, + osd_dto_credits_noquota[DTO_INDEX_DELETE]); return 0; } @@ -2892,14 +3438,14 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); LASSERT(handle != NULL); if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) RETURN(-EACCES); - OSD_EXEC_OP(handle, delete); + osd_trans_exec_op(env, handle, OSD_OT_DELETE); ipd = osd_idx_ipd_get(env, bag); if (unlikely(ipd == NULL)) @@ -2931,13 +3477,14 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, delete, osd_dto_credits_noquota[DTO_INDEX_DELETE]); + osd_trans_declare_op(env, oh, OSD_OT_DELETE, + osd_dto_credits_noquota[DTO_INDEX_DELETE]); inode = osd_dt_obj(dt)->oo_inode; LASSERT(inode); @@ -2957,7 +3504,22 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, rec = (struct osd_fid_pack *) (de->name + de->name_len + 1); rc = osd_fid_unpack((struct lu_fid *)fid, rec); } - RETURN(rc); + return rc; +} + +static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, + struct lu_fid *fid) +{ + ENTRY; + + /* FID seqs not in FLDB, must be local seq */ + if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) + RETURN(0); + + if (osd_seq_exists(env, osd, fid_seq(fid))) + RETURN(0); + + RETURN(1); } /** @@ -2979,18 +3541,19 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, struct inode *dir = obj->oo_inode; struct dentry *dentry; struct osd_thandle *oh; - struct ldiskfs_dir_entry_2 *de; + struct ldiskfs_dir_entry_2 *de = NULL; struct buffer_head *bh; struct htree_lock *hlock = NULL; - int rc; - + struct lu_fid *fid = &osd_oti_get(env)->oti_fid; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + int rc; ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); - OSD_EXEC_OP(handle, delete); + osd_trans_exec_op(env, handle, OSD_OT_DELETE); oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle != NULL); @@ -3008,21 +3571,108 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, LDISKFS_HLOCK_DEL); } else { - cfs_down_write(&obj->oo_ext_idx_sem); + 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; } if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - cfs_up_write(&obj->oo_ext_idx_sem); + up_write(&obj->oo_ext_idx_sem); + + if (rc != 0) + GOTO(out, rc); + + /* For inode on the remote MDT, .. will point to + * /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); + } + + if (rc == -ENOENT) + rc = 0; + + GOTO(out, rc); + } +out: LASSERT(osd_invariant(obj)); RETURN(rc); @@ -3054,7 +3704,7 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, ENTRY; LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) @@ -3106,13 +3756,13 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, { struct osd_thandle *oh; - LASSERT(dt_object_exists(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT]); return 0; } @@ -3144,14 +3794,14 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); LASSERT(th != NULL); if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) RETURN(-EACCES); - OSD_EXEC_OP(th, insert); + osd_trans_exec_op(env, th, OSD_OT_INSERT); ipd = osd_idx_ipd_get(env, bag); if (unlikely(ipd == NULL)) @@ -3202,27 +3852,20 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, oth = container_of(th, struct osd_thandle, ot_super); LASSERT(oth->ot_handle != NULL); LASSERT(oth->ot_handle->h_transaction != NULL); - - child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); - - /* XXX: remove fid_is_igif() check here. - * IGIF check is just to handle insertion of .. when it is 'ROOT', - * it is IGIF now but needs FID in dir entry as well for readdir - * to work. - * LU-838 should fix that and remove fid_is_igif() check */ - if (fid_is_igif((struct lu_fid *)fid) || - fid_is_norm((struct lu_fid *)fid)) { - ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; - osd_get_ldiskfs_dirent_param(ldp, fid); - child->d_fsdata = (void *)ldp; - } else { - child->d_fsdata = NULL; - } LASSERT(pobj->oo_inode); + + ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; + if (unlikely(pobj->oo_inode == + osd_sb(osd_obj2dev(pobj))->s_root->d_inode)) + ldp->edp_magic = 0; + else + osd_get_ldiskfs_dirent_param(ldp, fid); + child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); + child->d_fsdata = (void *)ldp; ll_vfs_dq_init(pobj->oo_inode); - rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); + rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); - RETURN(rc); + RETURN(rc); } /** @@ -3245,8 +3888,6 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, struct thandle *th) { struct inode *inode = dir->oo_inode; - struct ldiskfs_dentry_param *dot_ldp; - struct ldiskfs_dentry_param *dot_dot_ldp; struct osd_thandle *oth; int result = 0; @@ -3262,32 +3903,23 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, dir->oo_compat_dot_created = 1; result = 0; } - } else if(strcmp(name, dotdot) == 0) { - dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; - dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; - - if (!dir->oo_compat_dot_created) - return -EINVAL; - if (!fid_is_igif((struct lu_fid *)dot_fid)) { - osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid); - osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); - } else { - dot_ldp = NULL; - dot_dot_ldp = NULL; - } - /* in case of rename, dotdot is already created */ - if (dir->oo_compat_dotdot_created) { - return __osd_ea_add_rec(info, dir, parent_dir, name, - dot_dot_fid, NULL, th); - } + } else if (strcmp(name, dotdot) == 0) { + if (!dir->oo_compat_dot_created) + return -EINVAL; + /* in case of rename, dotdot is already created */ + if (dir->oo_compat_dotdot_created) { + return __osd_ea_add_rec(info, dir, parent_dir, name, + dot_dot_fid, NULL, th); + } - result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, - inode, dot_ldp, dot_dot_ldp); - if (result == 0) - dir->oo_compat_dotdot_created = 1; - } + result = osd_add_dot_dotdot_internal(info, dir->oo_inode, + parent_dir, dot_fid, + dot_dot_fid, oth); + if (result == 0) + dir->oo_compat_dotdot_created = 1; + } - return result; + return result; } @@ -3311,7 +3943,7 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, ldiskfs_htree_lock(hlock, pobj->oo_hl_head, pobj->oo_inode, 0); } else { - cfs_down_write(&pobj->oo_ext_idx_sem); + 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), @@ -3321,16 +3953,26 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, ldiskfs_htree_lock(hlock, pobj->oo_hl_head, pobj->oo_inode, LDISKFS_HLOCK_ADD); } else { - cfs_down_write(&pobj->oo_ext_idx_sem); + down_write(&pobj->oo_ext_idx_sem); } - rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, - hlock, th); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { + struct lu_fid *tfid = &info->oti_fid; + + *tfid = *(const struct lu_fid *)fid; + tfid->f_ver = ~0; + rc = __osd_ea_add_rec(info, pobj, cinode, name, + (const struct dt_rec *)tfid, + hlock, th); + } else { + rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, + hlock, th); + } } if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - cfs_up_write(&pobj->oo_ext_idx_sem); + up_write(&pobj->oo_ext_idx_sem); return rc; } @@ -3350,7 +3992,7 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, RETURN_EXIT; again: - rc = osd_oi_lookup(oti, dev, fid, id); + rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD); if (rc != 0 && rc != -ENOENT) RETURN_EXIT; @@ -3370,13 +4012,11 @@ 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; } @@ -3384,6 +4024,43 @@ again: EXIT; } +static int osd_fail_fid_lookup(struct osd_thread_info *oti, + struct osd_device *dev, + struct osd_idmap_cache *oic, + struct lu_fid *fid, __u32 ino) +{ + struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs; + struct inode *inode; + int rc; + + osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); + inode = osd_iget(oti, dev, &oic->oic_lid); + if (IS_ERR(inode)) { + fid_zero(&oic->oic_fid); + return PTR_ERR(inode); + } + + rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma); + iput(inode); + if (rc != 0) + fid_zero(&oic->oic_fid); + else + *fid = oic->oic_fid = lma->lma_self_fid; + 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 @@ -3403,6 +4080,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, struct htree_lock *hlock = NULL; int ino; int rc; + ENTRY; LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL); @@ -3414,34 +4092,48 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, LDISKFS_HLOCK_LOOKUP); } else { - cfs_down_read(&obj->oo_ext_idx_sem); + 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; struct osd_idmap_cache *oic = &oti->oti_cache; struct osd_device *dev = osd_obj2dev(obj); struct osd_scrub *scrub = &dev->od_scrub; struct scrub_file *sf = &scrub->os_file; ino = le32_to_cpu(de->inode); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) { + brelse(bh); + rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino); + GOTO(out, rc); + } + rc = osd_get_fid_from_dentry(de, rec); /* done with de, release bh */ brelse(bh); if (rc != 0) - rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid); + rc = osd_ea_fid_get(env, obj, ino, fid, id); else - osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); + osd_id_gen(id, ino, OSD_OII_NOGEN); if (rc != 0) { fid_zero(&oic->oic_fid); GOTO(out, rc); } - 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_INCONSISTENT) || + (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) osd_consistency_check(oti, dev, oic); @@ -3455,7 +4147,7 @@ out: if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - cfs_up_read(&obj->oo_ext_idx_sem); + up_read(&obj->oo_ext_idx_sem); return rc; } @@ -3509,10 +4201,11 @@ 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; } /** @@ -3532,27 +4225,55 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, const struct dt_key *key, struct thandle *handle) { - struct osd_thandle *oh; - struct inode *inode; - int rc; + struct osd_thandle *oh; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct lu_fid *fid = (struct lu_fid *)rec; + int rc; ENTRY; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, insert, osd_dto_credits_noquota[DTO_INDEX_INSERT]); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT]); - inode = osd_dt_obj(dt)->oo_inode; - LASSERT(inode); + 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 { + 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, + oh, true, true, NULL, false); + } + + if (fid == NULL) + RETURN(0); + + rc = osd_remote_fid(env, osd, fid); + if (rc <= 0) + RETURN(rc); + + rc = 0; + + osd_trans_declare_op(env, oh, OSD_OT_CREATE, + osd_dto_credits_noquota[DTO_OBJECT_CREATE]); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); - /* 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, oh, - true, true, NULL, false); RETURN(rc); } @@ -3572,31 +4293,80 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, struct thandle *th, struct lustre_capa *capa, int ignore_quota) { - struct osd_object *obj = osd_dt_obj(dt); - struct lu_fid *fid = (struct lu_fid *) rec; - const char *name = (const char *)key; - struct osd_object *child; - int rc; - - ENTRY; + 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; + const char *name = (const char *)key; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_inode_id *id = &oti->oti_id; + struct inode *child_inode = NULL; + struct osd_object *child = NULL; + int rc; + ENTRY; LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); 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); - child = osd_object_find(env, dt, fid); - if (!IS_ERR(child)) { - rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th); - osd_object_put(env, child); - } else { - rc = PTR_ERR(child); - } + LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid)); - LASSERT(osd_invariant(obj)); - RETURN(rc); + rc = osd_remote_fid(env, osd, fid); + if (rc < 0) { + CERROR("%s: Can not find object "DFID" rc %d\n", + osd_name(osd), PFID(fid), rc); + RETURN(rc); + } + + if (rc == 1) { + /* Insert remote entry */ + if (strcmp(name, dotdot) == 0 && strlen(name) == 2) { + struct osd_mdobj_map *omm = osd->od_mdt_map; + struct osd_thandle *oh; + + /* If parent on remote MDT, we need put this object + * under AGENT */ + oh = container_of(th, typeof(*oh), ot_super); + rc = osd_add_to_remote_parent(env, osd, obj, oh); + if (rc != 0) { + 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_remote_parent->d_inode); + } else { + child_inode = osd_create_local_agent_inode(env, osd, + obj, fid, + th); + if (IS_ERR(child_inode)) + RETURN(PTR_ERR(child_inode)); + } + } else { + /* Insert local entry */ + child = osd_object_find(env, dt, fid); + if (IS_ERR(child)) { + CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n", + osd_name(osd), PFID(fid), + id->oii_ino, id->oii_gen, + (int)PTR_ERR(child)); + RETURN(PTR_ERR(child)); + } + child_inode = igrab(child->oo_inode); + } + + rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th); + + iput(child_inode); + if (child != NULL) + osd_object_put(env, child); + LASSERT(osd_invariant(obj)); + RETURN(rc); } /** @@ -3679,7 +4449,6 @@ static int osd_it_iam_get(const struct lu_env *env, * * \param di osd iterator */ - static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) { struct osd_it_iam *it = (struct osd_it_iam *)di; @@ -3738,44 +4507,42 @@ static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di) return iam_it_key_size(&it->oi_it); } -static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr, - int len, __u16 type) +static inline void +osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type) { - struct luda_type *lt; - const unsigned align = sizeof(struct luda_type) - 1; + /* check if file type is required */ + if (ent->lde_attrs & LUDA_TYPE) { + struct luda_type *lt; + int align = sizeof(*lt) - 1; - /* check if file type is required */ - if (attr & LUDA_TYPE) { - len = (len + align) & ~align; - - lt = (void *) ent->lde_name + len; - lt->lt_type = cpu_to_le16(CFS_DTTOIF(type)); - ent->lde_attrs |= LUDA_TYPE; - } + len = (len + align) & ~align; + lt = (struct luda_type *)(ent->lde_name + len); + lt->lt_type = cpu_to_le16(DTTOIF(type)); + } - ent->lde_attrs = cpu_to_le32(ent->lde_attrs); + ent->lde_attrs = cpu_to_le32(ent->lde_attrs); } /** * build lu direct from backend fs dirent. */ -static inline void osd_it_pack_dirent(struct lu_dirent *ent, - struct lu_fid *fid, __u64 offset, - char *name, __u16 namelen, - __u16 type, __u32 attr) +static inline void +osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset, + char *name, __u16 namelen, __u16 type, __u32 attr) { - fid_cpu_to_le(&ent->lde_fid, fid); - ent->lde_attrs = LUDA_FID; + ent->lde_attrs = attr | LUDA_FID; + fid_cpu_to_le(&ent->lde_fid, fid); - ent->lde_hash = cpu_to_le64(offset); - ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr)); + ent->lde_hash = cpu_to_le64(offset); + ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr)); - strncpy(ent->lde_name, name, namelen); - ent->lde_namelen = cpu_to_le16(namelen); + strncpy(ent->lde_name, name, namelen); + ent->lde_name[namelen] = '\0'; + ent->lde_namelen = cpu_to_le16(namelen); - /* append lustre attributes */ - osd_it_append_attrs(ent, attr, namelen, type); + /* append lustre attributes */ + osd_it_append_attrs(ent, namelen, type); } /** @@ -3878,6 +4645,7 @@ static const struct dt_index_operations osd_index_iam_ops = { } }; + /** * Creates or initializes iterator context. * @@ -3892,6 +4660,7 @@ 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; @@ -3906,17 +4675,22 @@ 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); } /** @@ -3984,6 +4758,7 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, unsigned d_type) { struct osd_it_ea *it = (struct osd_it_ea *)buf; + struct osd_object *obj = it->oie_obj; struct osd_it_ea_dirent *ent = it->oie_dirent; struct lu_fid *fid = &ent->oied_fid; struct osd_fid_pack *rec; @@ -3999,16 +4774,23 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, OSD_IT_EA_BUFSIZE) RETURN(1); - if (d_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack*) (name + namelen + 1); - - if (osd_fid_unpack(fid, rec) != 0) - fid_zero(fid); + /* "." is just the object itself. */ + if (namelen == 1 && name[0] == '.') { + *fid = obj->oo_dt.do_lu.lo_header->loh_fid; + } else if (d_type & LDISKFS_DIRENT_LUFID) { + rec = (struct osd_fid_pack*) (name + namelen + 1); + if (osd_fid_unpack(fid, rec) != 0) + fid_zero(fid); + } else { + fid_zero(fid); + } + d_type &= ~LDISKFS_DIRENT_LUFID; - d_type &= ~LDISKFS_DIRENT_LUFID; - } else { - fid_zero(fid); - } + /* NOT export local root. */ + if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) { + ino = obj->oo_inode->i_ino; + *fid = obj->oo_dt.do_lu.lo_header->loh_fid; + } ent->oied_ino = ino; ent->oied_off = offset; @@ -4049,7 +4831,7 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, ldiskfs_htree_lock(hlock, obj->oo_hl_head, inode, LDISKFS_HLOCK_READDIR); } else { - cfs_down_read(&obj->oo_ext_idx_sem); + down_read(&obj->oo_ext_idx_sem); } result = inode->i_fop->readdir(&it->oie_file, it, @@ -4058,16 +4840,18 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, if (hlock != NULL) ldiskfs_htree_unlock(hlock); else - cfs_up_read(&obj->oo_ext_idx_sem); + 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); + } else { + it->oie_dirent = it->oie_buf; + it->oie_it_dirent = 1; + } - RETURN(result); + RETURN(result); } /** @@ -4134,10 +4918,371 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) return it->oie_dirent->oied_namelen; } +static int +osd_dirent_update(handle_t *jh, struct super_block *sb, + struct osd_it_ea_dirent *ent, struct lu_fid *fid, + struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de) +{ + struct osd_fid_pack *rec; + int rc; + ENTRY; + + LASSERT(de->file_type & LDISKFS_DIRENT_LUFID); + LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack)); + + rc = ldiskfs_journal_get_write_access(jh, bh); + if (rc != 0) { + 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); + 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); +} + +static inline int +osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize) +{ + if (ldiskfs_rec_len_from_disk(reclen, blocksize) >= + __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack))) + return 1; + else + 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, + struct osd_it_ea_dirent *ent, struct lu_fid *fid, + struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de, + struct htree_lock *hlock) +{ + struct dentry *dentry; + struct osd_fid_pack *rec; + struct ldiskfs_dentry_param *ldp; + int rc; + ENTRY; + + if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb, + LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + RETURN(0); + + /* There is enough space to hold the FID-in-dirent. */ + if (osd_dirent_has_space(de->rec_len, ent->oied_namelen, + dir->i_sb->s_blocksize)) { + 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); + RETURN(rc); + } + + de->name[de->name_len] = 0; + rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + rec->fp_len = sizeof(struct lu_fid) + 1; + fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); + de->file_type |= LDISKFS_DIRENT_LUFID; + + rc = ldiskfs_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); + 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); + 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", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + + RETURN(rc); +} + +static int +osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, + struct osd_it_ea *it, struct lu_fid *fid, + struct osd_inode_id *id, __u32 *attr) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct osd_device *dev = osd_obj2dev(obj); + struct super_block *sb = osd_sb(dev); + const char *devname = + LDISKFS_SB(sb)->s_es->s_volume_name; + struct osd_it_ea_dirent *ent = it->oie_dirent; + struct inode *dir = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct buffer_head *bh = NULL; + handle_t *jh = NULL; + struct ldiskfs_dir_entry_2 *de; + struct dentry *dentry; + struct inode *inode; + int credits; + int rc; + int dot_dotdot = 0; + bool dirty = false; + ENTRY; + + if (ent->oied_name[0] == '.') { + if (ent->oied_namelen == 1) + 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, + ent->oied_namelen); + + /* We need to ensure that the name entry is still valid. + * Because it may be removed or renamed by other already. + * + * The unlink or rename operation will start journal before PDO lock, + * so to avoid deadlock, here we need to start journal handle before + * related PDO lock also. But because we do not know whether there + * will be something to be repaired before PDO lock, we just start + * journal without conditions. + * + * We may need to remove the name entry firstly, then insert back. + * One credit is for user quota file update. + * One credit is for group quota file update. + * Two credits are for dirty inode. */ + credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] + + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; + +again: + if (dev->od_dirent_journal) { + 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); + RETURN(rc); + } + + 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 { + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, + LDISKFS_HLOCK_LOOKUP); + } else { + down_read(&obj->oo_ext_idx_sem); + } + } + + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + /* 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) || + (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) { + *attr |= LUDA_IGNORE; + GOTO(out_journal, rc = 0); + } + + osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN); + inode = osd_iget(info, dev, id); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + if (rc == -ENOENT || rc == -ESTALE) { + *attr |= LUDA_IGNORE; + rc = 0; + } + + 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)) + GOTO(out_inode, rc = 0); + + /* Do not repair under dryrun mode. */ + if (*attr & LUDA_VERIFY_DRYRUN) { + *attr |= LUDA_REPAIR; + GOTO(out_inode, rc = 0); + } + + if (!dev->od_dirent_journal) { + iput(inode); + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_read(&obj->oo_ext_idx_sem); + dev->od_dirent_journal = 1; + goto again; + } + + *fid = lma->lma_self_fid; + dirty = true; + /* Update the FID-in-dirent. */ + rc = osd_dirent_update(jh, sb, ent, fid, bh, de); + if (rc == 0) + *attr |= LUDA_REPAIR; + } 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); + } + + if (!dev->od_dirent_journal) { + iput(inode); + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_read(&obj->oo_ext_idx_sem); + dev->od_dirent_journal = 1; + goto again; + } + + *fid = lma->lma_self_fid; + dirty = true; + /* Append the FID-in-dirent. */ + rc = osd_dirent_reinsert(env, jh, dir, inode, ent, + fid, bh, de, hlock); + if (rc == 0) + *attr |= LUDA_REPAIR; + } + } else if (rc == -ENODATA) { + /* Do not repair under dryrun mode. */ + if (*attr & LUDA_VERIFY_DRYRUN) { + if (fid_is_sane(fid)) { + *attr |= LUDA_REPAIR; + } else { + lu_igif_build(fid, inode->i_ino, + inode->i_generation); + *attr |= LUDA_UPGRADE; + } + GOTO(out_inode, rc = 0); + } + + if (!dev->od_dirent_journal) { + iput(inode); + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_read(&obj->oo_ext_idx_sem); + dev->od_dirent_journal = 1; + goto again; + } + + dirty = true; + 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, 0, 0); + if (rc == 0) + *attr |= LUDA_REPAIR; + } else { + lu_igif_build(fid, inode->i_ino, inode->i_generation); + /* It is probably IGIF object. Only aappend the + * FID-in-dirent. OI scrub will process FID-in-LMA. */ + rc = osd_dirent_reinsert(env, jh, dir, inode, ent, + fid, bh, de, hlock); + if (rc == 0) + *attr |= LUDA_UPGRADE; + } + } + + GOTO(out_inode, rc); + +out_inode: + iput(inode); + +out_journal: + brelse(bh); + if (hlock != NULL) { + ldiskfs_htree_unlock(hlock); + } 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) + dev->od_dirent_journal = 0; + return rc; +} /** - * Returns the value (i.e. fid/igif) at current position from iterator's - * in memory structure. + * Returns the value at current position from iterator's in memory structure. * * \param di struct osd_it_ea, iterator's in memory structure * \param attr attr requested for dirent. @@ -4156,6 +5301,7 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_scrub *scrub = &dev->od_scrub; struct scrub_file *sf = &scrub->os_file; struct osd_thread_info *oti = osd_oti_get(env); + struct osd_inode_id *id = &oti->oti_id; struct osd_idmap_cache *oic = &oti->oti_cache; struct lu_fid *fid = &it->oie_dirent->oied_fid; struct lu_dirent *lde = (struct lu_dirent *)dtrec; @@ -4163,23 +5309,46 @@ static inline int osd_it_ea_rec(const struct lu_env *env, int rc = 0; ENTRY; - if (!fid_is_sane(fid)) { - rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid); - if (rc != 0) { - fid_zero(&oic->oic_fid); - RETURN(rc); + if (attr & LUDA_VERIFY) { + attr |= LUDA_TYPE; + if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { + attr |= LUDA_IGNORE; + rc = 0; + } else { + rc = osd_dirent_check_repair(env, obj, it, fid, id, + &attr); } } else { - osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); + attr &= ~LU_DIRENT_ATTRS_MASK; + if (!fid_is_sane(fid)) { + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + RETURN(-ENOENT); + + rc = osd_ea_fid_get(env, obj, ino, fid, id); + } else { + osd_id_gen(id, ino, OSD_OII_NOGEN); + } } + /* 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); - oic->oic_fid = *fid; - if ((scrub->os_pos_current <= ino) && - (sf->sf_flags & SF_INCONSISTENT || + + if (rc < 0) + RETURN(rc); + + if (osd_remote_fid(env, dev, fid)) + RETURN(0); + + if (likely(!(attr & LUDA_IGNORE))) + rc = osd_add_oi_cache(oti, dev, id, fid); + + if (!(attr & LUDA_VERIFY) && + (scrub->os_pos_current <= ino) && + ((sf->sf_flags & SF_INCONSISTENT) || + (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) osd_consistency_check(oti, dev, oic); @@ -4311,12 +5480,14 @@ 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, @@ -4345,7 +5516,9 @@ 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); } @@ -4353,13 +5526,6 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) { ENTRY; - osd_scrub_cleanup(env, o); - - if (o->od_fsops) { - fsfilt_put_ops(o->od_fsops); - o->od_fsops = NULL; - } - /* shutdown quota slave instance associated with the device */ if (o->od_quota_slave != NULL) { qsd_fini(env, o->od_quota_slave); @@ -4369,6 +5535,21 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *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) { @@ -4380,19 +5561,20 @@ 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) RETURN(0); - 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); - } + if (strlen(dev) >= sizeof(o->od_mntdev)) + RETURN(-E2BIG); + strcpy(o->od_mntdev, dev); - OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD); + OBD_PAGE_ALLOC(__page, GFP_IOFS); if (__page == NULL) GOTO(out, rc = -ENOMEM); @@ -4402,7 +5584,7 @@ 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) @@ -4413,7 +5595,7 @@ 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) { @@ -4422,73 +5604,71 @@ 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; +#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); } - if (lmd_flags & LMD_FLG_IAM) { - o->od_iop_mode = 0; - LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name); - } else - o->od_iop_mode = 1; 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_compat_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, @@ -4498,6 +5678,7 @@ 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); @@ -4509,9 +5690,8 @@ static int osd_device_init0(const struct lu_env *env, l->ld_ops = &osd_lu_ops; o->od_dt_dev.dd_ops = &osd_dt_ops; - cfs_spin_lock_init(&o->od_osfs_lock); - cfs_mutex_init(&o->od_otable_mutex); - o->od_osfs_age = cfs_time_shift_64(-1000); + spin_lock_init(&o->od_osfs_lock); + mutex_init(&o->od_otable_mutex); o->od_capa_hash = init_capa_hash(); if (o->od_capa_hash == NULL) @@ -4525,32 +5705,43 @@ static int osd_device_init0(const struct lu_env *env, if (rc) GOTO(out_capa, rc); - /* setup scrub, including OI files initialization */ - rc = osd_scrub_setup(env, o); - if (rc < 0) + cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), + sizeof(o->od_svname)); + if (cplen >= sizeof(o->od_svname)) { + rc = -E2BIG; GOTO(out_mnt, rc); + } - strncpy(o->od_svname, lustre_cfg_string(cfg, 4), - sizeof(o->od_svname) - 1); + if (server_name_is_ost(o->od_svname)) + o->od_is_ost = 1; - rc = osd_compat_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; + + 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_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); @@ -4565,23 +5756,21 @@ 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_compat_fini(o); -out_scrub: - osd_scrub_cleanup(env, o); + osd_obj_map_fini(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, @@ -4597,6 +5786,9 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, rc = dt_device_init(&o->od_dt_dev, t); if (rc == 0) { + /* Because the ctx might be revived in dt_device_init, + * refill the env here */ + lu_env_refill((struct lu_env *)env); rc = osd_device_init0(env, o, cfg); if (rc) dt_device_fini(&o->od_dt_dev); @@ -4631,23 +5823,32 @@ 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_param(PARAM_OSD, lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + if (rc > 0 || rc == -ENOSYS) + rc = class_process_proc_param(PARAM_OST, + lprocfs_osd_obd_vars, + cfg, &o->od_dt_dev); + break; + default: + rc = -ENOSYS; + } - RETURN(err); + RETURN(rc); } static int osd_recovery_complete(const struct lu_env *env, @@ -4686,9 +5887,9 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp, *exp = class_conn2export(&conn); - cfs_spin_lock(&osd->od_osfs_lock); + spin_lock(&osd->od_osfs_lock); osd->od_connects++; - cfs_spin_unlock(&osd->od_osfs_lock); + spin_unlock(&osd->od_osfs_lock); RETURN(0); } @@ -4705,11 +5906,11 @@ static int osd_obd_disconnect(struct obd_export *exp) ENTRY; /* Only disconnect the underlying layers on the final disconnect. */ - cfs_spin_lock(&osd->od_osfs_lock); + spin_lock(&osd->od_osfs_lock); osd->od_connects--; if (osd->od_connects == 0) release = 1; - cfs_spin_unlock(&osd->od_osfs_lock); + spin_unlock(&osd->od_osfs_lock); rc = class_disconnect(exp); /* bz 9811 */ @@ -4725,15 +5926,6 @@ 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) - RETURN(result); - } - if (osd->od_quota_slave != NULL) /* set up quota slave objects */ result = qsd_prepare(env, osd->od_quota_slave); @@ -4789,17 +5981,28 @@ static struct obd_ops osd_obd_device_ops = { 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, NULL, +#ifndef HAVE_ONLY_PROCFS_SEQ + lprocfs_osd_module_vars, +#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. ");