X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=2876e6ee437a7f4ebad82543c711f3f4c3c4abaa;hp=e3de52a54b4f3512dcb6bb23c29e2f58249e7bf6;hb=2e92c57d71384d50f06baf5b6591d2809c8288b2;hpb=0754bc8f2623bea184111af216f7567608db35b6 diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index e3de52a..2876e6e 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2015, Intel Corporation. + * Copyright (c) 2011, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -60,6 +56,7 @@ #include #include +#include #undef ENTRY /* * struct OBD_{ALLOC,FREE}*() @@ -81,13 +78,16 @@ #include +#define PFID_STRIPE_IDX_BITS 16 +#define PFID_STRIPE_COUNT_MASK ((1 << PFID_STRIPE_IDX_BITS) - 1) + int ldiskfs_pdo = 1; -CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, - "ldiskfs with parallel directory operations"); +module_param(ldiskfs_pdo, int, 0644); +MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations"); int ldiskfs_track_declares_assert; -CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644, - "LBUG during tracking of declares"); +module_param(ldiskfs_track_declares_assert, int, 0644); +MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares"); /* Slab to allocate dynlocks */ struct kmem_cache *dynlock_cachep; @@ -175,6 +175,153 @@ static int osd_root_get(const struct lu_env *env, } /* + * the following set of functions are used to maintain per-thread + * cache of FID->ino mapping. this mechanism is needed to resolve + * FID to inode at dt_insert() which in turn stores ino in the + * directory entries to keep ldiskfs compatible with ext[34]. + * due to locking-originated restrictions we can't lookup ino + * using LU cache (deadlock is possible). lookup using OI is quite + * expensive. so instead we maintain this cache and methods like + * dt_create() fill it. so in the majority of cases dt_insert() is + * able to find needed mapping in lockless manner. + */ +static struct osd_idmap_cache * +osd_idc_find(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc = oti->oti_ins_cache; + int i; + for (i = 0; i < oti->oti_ins_cache_used; i++) { + if (!lu_fid_eq(&idc[i].oic_fid, fid)) + continue; + if (idc[i].oic_dev != osd) + continue; + + return idc + i; + } + + return NULL; +} + +static struct osd_idmap_cache * +osd_idc_add(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc; + int i; + + if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) { + i = oti->oti_ins_cache_size * 2; + if (i == 0) + i = OSD_INS_CACHE_SIZE; + OBD_ALLOC(idc, sizeof(*idc) * i); + if (idc == NULL) + return ERR_PTR(-ENOMEM); + if (oti->oti_ins_cache != NULL) { + memcpy(idc, oti->oti_ins_cache, + oti->oti_ins_cache_used * sizeof(*idc)); + OBD_FREE(oti->oti_ins_cache, + oti->oti_ins_cache_used * sizeof(*idc)); + } + oti->oti_ins_cache = idc; + oti->oti_ins_cache_size = i; + } + + idc = oti->oti_ins_cache + oti->oti_ins_cache_used++; + idc->oic_fid = *fid; + idc->oic_dev = osd; + idc->oic_lid.oii_ino = 0; + idc->oic_lid.oii_gen = 0; + idc->oic_remote = 0; + + return idc; +} + +/* + * lookup mapping for the given fid in the cache, initialize a + * new one if not found. the initialization checks whether the + * object is local or remote. for local objects, OI is used to + * learn ino/generation. the function is used when the caller + * has no information about the object, e.g. at dt_insert(). + */ +static struct osd_idmap_cache * +osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid) +{ + struct osd_idmap_cache *idc; + int rc; + + idc = osd_idc_find(env, osd, fid); + LASSERT(!IS_ERR(idc)); + if (idc != NULL) + return idc; + + /* new mapping is needed */ + idc = osd_idc_add(env, osd, fid); + if (IS_ERR(idc)) + return idc; + + /* initialize it */ + rc = osd_remote_fid(env, osd, fid); + if (unlikely(rc < 0)) + return ERR_PTR(rc); + + if (rc == 0) { + /* the object is local, lookup in OI */ + /* XXX: probably cheaper to lookup in LU first? */ + rc = osd_oi_lookup(osd_oti_get(env), osd, fid, + &idc->oic_lid, 0); + if (unlikely(rc < 0)) { + CERROR("can't lookup: rc = %d\n", rc); + return ERR_PTR(rc); + } + } else { + /* the object is remote */ + idc->oic_remote = 1; + } + + return idc; +} + +/* + * lookup mapping for given FID and fill it from the given object. + * the object is lolcal by definition. + */ +static int osd_idc_find_and_init(const struct lu_env *env, + struct osd_device *osd, + struct osd_object *obj) +{ + const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); + struct osd_idmap_cache *idc; + + idc = osd_idc_find(env, osd, fid); + LASSERT(!IS_ERR(idc)); + if (idc != NULL) { + if (obj->oo_inode == NULL) + return 0; + if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) { + LASSERT(idc->oic_lid.oii_ino == 0); + idc->oic_lid.oii_ino = obj->oo_inode->i_ino; + idc->oic_lid.oii_gen = obj->oo_inode->i_generation; + } + return 0; + } + + /* new mapping is needed */ + idc = osd_idc_add(env, osd, fid); + if (IS_ERR(idc)) + return PTR_ERR(idc); + + if (obj->oo_inode != NULL) { + idc->oic_lid.oii_ino = obj->oo_inode->i_ino; + idc->oic_lid.oii_gen = obj->oo_inode->i_generation; + } + return 0; +} + +/* * OSD object methods. */ @@ -199,6 +346,7 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, init_rwsem(&mo->oo_sem); init_rwsem(&mo->oo_ext_idx_sem); spin_lock_init(&mo->oo_guard); + INIT_LIST_HEAD(&mo->oo_xattr_list); return l; } else { return NULL; @@ -206,18 +354,20 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, } int osd_get_lma(struct osd_thread_info *info, struct inode *inode, - struct dentry *dentry, struct lustre_mdt_attrs *lma) + struct dentry *dentry, struct lustre_ost_attrs *loa) { int rc; - CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, - info->oti_mdt_attrs_old, LMA_OLD_SIZE); + (void *)loa, sizeof(*loa)); if (rc > 0) { - if ((void *)lma != (void *)info->oti_mdt_attrs_old) - memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma)); + struct lustre_mdt_attrs *lma = &loa->loa_lma; + + if (rc < sizeof(*lma)) + return -EINVAL; + rc = 0; - lustre_lma_swab(lma); + lustre_loa_swab(loa, true); /* Check LMA compatibility */ if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) { CWARN("%.16s: unsupported incompat LMA feature(s) %#x " @@ -240,6 +390,7 @@ int osd_get_lma(struct osd_thread_info *info, struct inode *inode, struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, struct osd_inode_id *id) { + int rc; struct inode *inode = NULL; /* if we look for an inode withing a running @@ -269,6 +420,9 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino); iput(inode); inode = ERR_PTR(-ENOENT); + } else if ((rc = osd_attach_jinode(inode))) { + iput(inode); + inode = ERR_PTR(rc); } else { ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY); if (id->oii_gen == OSD_OII_NOGEN) @@ -285,21 +439,67 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, return inode; } +int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd, + handle_t *handle, struct dentry *child, + struct inode *inode, struct htree_lock *hlock) +{ + int rc, rc2; + + rc = __ldiskfs_add_entry(handle, child, inode, hlock); + if (rc == -ENOBUFS || rc == -ENOSPC) { + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct inode *parent = child->d_parent->d_inode; + struct lu_fid *fid = NULL; + + rc2 = osd_get_lma(info, parent, child->d_parent, loa); + if (!rc2) { + fid = &loa->loa_lma.lma_self_fid; + } else if (rc2 == -ENODATA) { + if (unlikely(parent == inode->i_sb->s_root->d_inode)) { + fid = &info->oti_fid3; + lu_local_obj_fid(fid, OSD_FS_ROOT_OID); + } else if (!osd->od_is_ost && osd->od_index == 0) { + fid = &info->oti_fid3; + lu_igif_build(fid, parent->i_ino, + parent->i_generation); + } + } + + if (fid != NULL) + CWARN("%s: directory (inode: %lu, FID: "DFID") %s " + "maximum entry limit\n", + osd_name(osd), parent->i_ino, PFID(fid), + rc == -ENOSPC ? "has reached" : "is approaching"); + else + CWARN("%s: directory (inode: %lu, FID: unknown) %s " + "maximum entry limit\n", + osd_name(osd), parent->i_ino, + rc == -ENOSPC ? "has reached" : "is approaching"); + + /* ignore such error now */ + if (rc == -ENOBUFS) + rc = 0; + } + + return rc; +} + + static struct inode * osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, struct osd_inode_id *id, struct lu_fid *fid) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct inode *inode; - int rc; + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct inode *inode; + int rc; inode = osd_iget(info, dev, id); if (IS_ERR(inode)) return inode; - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc == 0) { - *fid = lma->lma_self_fid; + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (!rc) { + *fid = loa->loa_lma.lma_self_fid; } else if (rc == -ENODATA) { if (unlikely(inode == osd_sb(dev)->s_root->d_inode)) lu_local_obj_fid(fid, OSD_FS_ROOT_OID); @@ -316,69 +516,68 @@ 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 cached) + bool trusted) { - struct inode *inode; - int rc = 0; + struct inode *inode; + int rc = 0; ENTRY; /* The cached OI mapping is trustable. If we cannot locate the inode * via the cached OI mapping, then return the failure to the caller * directly without further OI checking. */ +again: inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { rc = PTR_ERR(inode); - if (cached || (rc != -ENOENT && rc != -ESTALE)) { - CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n", - id->oii_ino, rc); - - GOTO(put, rc); - } + if (!trusted && (rc == -ENOENT || rc == -ESTALE)) + goto check_oi; - goto check_oi; + CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n", + PFID(fid), id->oii_ino, rc); + GOTO(put, rc); } if (is_bad_inode(inode)) { rc = -ENOENT; - if (cached) { - CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino); + if (!trusted) + goto check_oi; - GOTO(put, rc); - } - - goto check_oi; + CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n", + PFID(fid), id->oii_ino); + GOTO(put, rc); } if (id->oii_gen != OSD_OII_NOGEN && inode->i_generation != id->oii_gen) { rc = -ESTALE; - if (cached) { - CDEBUG(D_INODE, "unmatched inode: ino = %u, " - "oii_gen = %u, i_generation = %u\n", - id->oii_ino, id->oii_gen, inode->i_generation); + if (!trusted) + goto check_oi; - GOTO(put, rc); - } - - goto check_oi; + CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, " + "oii_gen = %u, i_generation = %u\n", PFID(fid), + id->oii_ino, id->oii_gen, inode->i_generation); + GOTO(put, rc); } if (inode->i_nlink == 0) { rc = -ENOENT; - if (cached) { - CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); - - GOTO(put, rc); - } + if (!trusted) + goto check_oi; - goto check_oi; + CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n", + PFID(fid), id->oii_ino); + GOTO(put, rc); } ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY); check_oi: if (rc != 0) { + __u32 saved_ino = id->oii_ino; + __u32 saved_gen = id->oii_gen; + + LASSERT(!trusted); LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); @@ -397,21 +596,42 @@ check_oi: * 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) - /* "id->oii_gen != OSD_OII_NOGEN" is for - * "@cached == false" case. */ - rc = -ENOENT; - else - rc = -EREMCHG; - } else { + * to distinguish the 1st case from the 2nd case: + * if the OI files are consistent but may contain + * stale OI mappings because of case 2, if iget() + * returns -ENOENT or -ESTALE, then it should be + * the case 2. */ + if (rc != 0) /* If the OI mapping was in OI file before the * osd_iget_check(), but now, it is disappear, * then it must be removed by race. That is a * normal race case. */ + GOTO(put, rc); + + /* It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. */ + if (saved_ino != id->oii_ino || saved_gen != id->oii_gen) { + trusted = true; + goto again; + } + + if (IS_ERR(inode)) { + if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT) + /* It still can be the case 2, but we cannot + * distinguish it from the case 1. So return + * -EREMCHG to block current operation until + * OI scrub rebuilt the OI mappings. */ + rc = -EREMCHG; + else + rc = -ENOENT; + + GOTO(put, rc); } + + if (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); @@ -454,10 +674,13 @@ int osd_get_idif(struct osd_thread_info *info, struct inode *inode, 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: use 0 as the index for compatibility, the caller will - * handle index related issues when necessarry. */ - ostid_to_fid(fid, ostid, 0); + rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid)); + /* + * XXX: use 0 as the index for compatibility, the caller will + * handle index related issues when necessary. + */ + if (!rc) + ostid_to_fid(fid, ostid, 0); } else if (rc == sizeof(struct filter_fid)) { rc = 1; } else if (rc >= 0) { @@ -497,7 +720,8 @@ 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 lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; struct inode *inode = obj->oo_inode; struct dentry *dentry = &info->oti_obj_dentry; struct lu_fid *fid = NULL; @@ -505,9 +729,8 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) int rc; ENTRY; - CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, - info->oti_mdt_attrs_old, LMA_OLD_SIZE); + (void *)loa, sizeof(*loa)); if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) { fid = &lma->lma_self_fid; rc = osd_get_idif(info, inode, dentry, fid); @@ -530,7 +753,8 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) rc = 0; lustre_lma_swab(lma); if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) || - CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) { + (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) && + S_ISREG(inode->i_mode)))) { CWARN("%s: unsupported incompat LMA feature(s) %#x for " "fid = "DFID", ino = %lu\n", osd_name(osd), lma->lma_incompat & ~LMA_INCOMPAT_SUPP, @@ -571,22 +795,205 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) RETURN(rc); } +struct osd_check_lmv_buf { +#ifdef HAVE_DIR_CONTEXT + /* please keep it as first member */ + struct dir_context ctx; +#endif + struct osd_thread_info *oclb_info; + struct osd_device *oclb_dev; + struct osd_idmap_cache *oclb_oic; +}; + +/** + * It is called internally by ->readdir() to filter out the + * local slave object's FID of the striped directory. + * + * \retval 1 found the local slave's FID + * \retval 0 continue to check next item + * \retval -ve for failure + */ +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_stripe_dir_filldir(struct dir_context *buf, +#else +static int osd_stripe_dir_filldir(void *buf, +#endif + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) +{ + struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf; + struct osd_thread_info *oti = oclb->oclb_info; + struct lu_fid *fid = &oti->oti_fid3; + struct osd_inode_id *id = &oti->oti_id3; + struct osd_device *dev = oclb->oclb_dev; + struct osd_idmap_cache *oic = oclb->oclb_oic; + struct inode *inode; + int rc; + + if (name[0] == '.') + return 0; + + fid_zero(fid); + sscanf(name + 1, SFID, RFID(fid)); + if (!fid_is_sane(fid)) + return 0; + + if (osd_remote_fid(oti->oti_env, dev, fid)) + return 0; + + osd_id_gen(id, ino, OSD_OII_NOGEN); + inode = osd_iget(oti, dev, id); + if (IS_ERR(inode)) + return PTR_ERR(inode); + + iput(inode); + osd_add_oi_cache(oti, dev, id, fid); + oic->oic_fid = *fid; + oic->oic_lid = *id; + oic->oic_dev = dev; + rc = osd_oii_insert(dev, oic, true); + + return rc == 0 ? 1 : rc; +} + +/* When lookup item under striped directory, we need to locate the master + * MDT-object of the striped directory firstly, then the client will send + * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID + * and the item's name. If the system is restored from MDT file level backup, + * then before the OI scrub completely built the OI files, the OI mappings of + * the master MDT-object and slave MDT-object may be invalid. Usually, it is + * not a problem for the master MDT-object. Because when locate the master + * MDT-object, we will do name based lookup (for the striped directory itself) + * firstly, during such process we can setup the correct OI mapping for the + * master MDT-object. But it will be trouble for the slave MDT-object. Because + * the client will not trigger name based lookup on the MDT to locate the slave + * MDT-object before locating item under the striped directory, then when + * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object + * is invalid and does not know what the right OI mapping is, then the MDT has + * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding + * the OI file, related OI mapping is unknown yet, please try again later. And + * then client will re-try the RPC again and again until related OI mapping has + * been updated. That is quite inefficient. + * + * To resolve above trouble, we will handle it as the following two cases: + * + * 1) The slave MDT-object and the master MDT-object are on different MDTs. + * It is relative easy. Be as one of remote MDT-objects, the slave MDT-object + * is linked under /REMOTE_PARENT_DIR with the name of its FID string. + * We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR + * directly. Please check osd_fid_lookup(). + * + * 2) The slave MDT-object and the master MDT-object reside on the same MDT. + * Under such case, during lookup the master MDT-object, we will lookup the + * slave MDT-object via readdir against the master MDT-object, because the + * slave MDT-objects information are stored as sub-directories with the name + * "${FID}:${index}". Then when find the local slave MDT-object, its OI + * mapping will be recorded. Then subsequent osd_fid_lookup() will know + * the correct OI mapping for the slave MDT-object. */ +static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev, + struct inode *inode, struct osd_idmap_cache *oic) +{ + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct file *filp = &oti->oti_file; + const struct file_operations *fops; + struct lmv_mds_md_v1 *lmv1; + struct osd_check_lmv_buf oclb = { +#ifdef HAVE_DIR_CONTEXT + .ctx.actor = osd_stripe_dir_filldir, +#endif + .oclb_info = oti, + .oclb_dev = dev, + .oclb_oic = oic + }; + int rc = 0; + ENTRY; + +again: + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf, + buf->lb_len); + if (rc == -ERANGE) { + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0); + if (rc > 0) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + GOTO(out, rc = -ENOMEM); + + goto again; + } + } + + if (unlikely(rc == 0 || rc == -ENODATA)) + GOTO(out, rc = 0); + + if (rc < 0) + GOTO(out, rc); + + if (unlikely(buf->lb_buf == NULL)) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + GOTO(out, rc = -ENOMEM); + + goto again; + } + + lmv1 = buf->lb_buf; + if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1) + GOTO(out, rc = 0); + + fops = inode->i_fop; + dentry->d_inode = inode; + dentry->d_sb = inode->i_sb; + filp->f_pos = 0; + filp->f_path.dentry = dentry; + filp->f_mode = FMODE_64BITHASH; + filp->f_mapping = inode->i_mapping; + filp->f_op = fops; + filp->private_data = NULL; + set_file_inode(filp, inode); + +#ifdef HAVE_DIR_CONTEXT + oclb.ctx.pos = filp->f_pos; + rc = fops->iterate(filp, &oclb.ctx); + filp->f_pos = oclb.ctx.pos; +#else + rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir); +#endif + fops->release(inode, filp); + +out: + if (rc < 0) + CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u," + DFID": rc = %d\n", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + inode->i_ino, inode->i_generation, + PFID(&oic->oic_fid), rc); + else + rc = 0; + + 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) { struct osd_thread_info *info; - struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev; - struct osd_device *dev; + struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev; + struct osd_device *dev; struct osd_idmap_cache *oic; - struct osd_inode_id *id; - struct inode *inode; - struct osd_scrub *scrub; - struct scrub_file *sf; - int result; - int saved = 0; - bool cached = true; - bool triggered = false; + struct osd_inode_id *id; + struct inode *inode = NULL; + struct osd_scrub *scrub; + struct scrub_file *sf; + __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL; + __u32 saved_ino; + __u32 saved_gen; + int result = 0; + int rc1 = 0; + bool remote = false; + bool trusted = true; + bool updated = false; ENTRY; LINVRNT(osd_invariant(obj)); @@ -600,7 +1007,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, LASSERT(info); oic = &info->oti_cache; - if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) + if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) RETURN(-ENOENT); /* For the object is created as locking anchor, or for the object to @@ -626,7 +1033,10 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, goto iget; } - cached = false; + /* The OI mapping in the OI file can be updated by the OI scrub + * when we locate the inode via FID. So it may be not trustable. */ + trusted = false; + /* Search order: 3. OI files. */ result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (result == -ENOENT) { @@ -643,36 +1053,15 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, GOTO(out, result); iget: - inode = osd_iget_check(info, dev, fid, id, cached); + inode = osd_iget_check(info, dev, fid, id, trusted); if (IS_ERR(inode)) { result = PTR_ERR(inode); if (result == -ENOENT || result == -ESTALE) - GOTO(out, result = -ENOENT); + GOTO(out, result = 0); if (result == -EREMCHG) { trigger: - 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, SS_AUTO_FULL | - SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT); - 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; - } - - if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) - GOTO(out, result); - /* We still have chance to get the valid inode: for the * object which is referenced by remote name entry, the * object on the local MDT will be linked under the dir @@ -688,85 +1077,163 @@ trigger: * 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) { - cached = true; - goto iget; + if (!remote && + !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) { + rc1 = osd_lookup_in_remote_parent(info, dev, + fid, id); + if (rc1 == 0) { + remote = true; + trusted = true; + flags |= SS_AUTO_PARTIAL; + flags &= ~SS_AUTO_FULL; + goto iget; + } } - result = saved; + if (thread_is_running(&scrub->os_thread)) { + if (scrub->os_partial_scan && + !scrub->os_in_join) { + goto join; + } else { + if (inode != NULL && !IS_ERR(inode)) { + LASSERT(remote); + + osd_add_oi_cache(info, dev, id, + fid); + osd_oii_insert(dev, oic, true); + } else { + result = -EINPROGRESS; + } + } + } else if (!dev->od_noscrub) { + +join: + rc1 = osd_scrub_start(dev, flags); + LCONSOLE_WARN("%.16s: trigger OI scrub by RPC " + "for the "DFID" with flags 0x%x," + " rc = %d\n", osd_name(dev), + PFID(fid), flags, rc1); + if (rc1 == 0 || rc1 == -EALREADY) { + if (inode != NULL && !IS_ERR(inode)) { + LASSERT(remote); + + osd_add_oi_cache(info, dev, id, + fid); + osd_oii_insert(dev, oic, true); + } else { + result = -EINPROGRESS; + } + } else { + result = -EREMCHG; + } + } else { + result = -EREMCHG; + } } - GOTO(out, result); + if (inode == NULL || IS_ERR(inode)) + GOTO(out, result); + } else if (remote) { + goto trigger; } obj->oo_inode = inode; LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); result = osd_check_lma(env, obj); - if (result != 0) { - if (result == -ENODATA) { - if (cached) { - result = osd_oi_lookup(info, dev, fid, id, - OI_CHECK_FLD); - if (result != 0) { - /* result == -ENOENT means that the OI - * mapping has been removed by race, - * the target inode belongs to other - * object. - * - * Others error also can be returned - * directly. */ - iput(inode); - obj->oo_inode = NULL; - GOTO(out, result); - } else { - /* result == 0 means the cached OI - * mapping is still in the OI file, - * the target the inode is valid. */ - } - } else { - /* The current OI mapping is from the OI file, - * since the inode has been found via - * osd_iget_check(), no need recheck OI. */ - } + if (result == 0) + goto found; + + LASSERTF(id->oii_ino == inode->i_ino && + id->oii_gen == inode->i_generation, + "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n", + PFID(fid), id->oii_ino, id->oii_gen, + inode->i_ino, inode->i_generation); + + saved_ino = inode->i_ino; + saved_gen = inode->i_generation; + if (unlikely(result == -ENODATA)) { + /* If the OI scrub updated the OI mapping by race, it + * must be valid. Trust the inode that has no LMA EA. */ + if (updated) goto found; - } - iput(inode); - obj->oo_inode = NULL; - if (result != -EREMCHG) - GOTO(out, result); + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + if (result == 0) { + /* The OI mapping is still there, the inode is still + * valid. It is just becaues the inode has no LMA EA. */ + if (saved_ino == id->oii_ino && + saved_gen == id->oii_gen) + goto found; - if (cached) { - result = osd_oi_lookup(info, dev, fid, id, - OI_CHECK_FLD); - /* result == -ENOENT means the cached OI mapping - * has been removed from the OI file by race, - * above target inode belongs to other object. - * - * Others error also can be returned directly. */ - if (result != 0) - GOTO(out, result); + /* It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. */ + iput(inode); + inode = NULL; + obj->oo_inode = NULL; + trusted = true; + updated = true; + goto iget; + } - /* result == 0, goto trigger */ - } else { - /* The current OI mapping is from the OI file, - * since the inode has been found via - * osd_iget_check(), no need recheck OI. */ + /* "result == -ENOENT" means that the OI mappinghas been + * removed by race, so the inode belongs to other object. + * + * Others error can be returned directly. */ + if (result == -ENOENT) { + LASSERT(trusted); + + result = 0; } + } - goto trigger; + iput(inode); + inode = NULL; + obj->oo_inode = NULL; + + if (result != -EREMCHG) + GOTO(out, result); + + LASSERT(!updated); + + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + /* "result == -ENOENT" means the cached OI mapping has been removed + * from the OI file by race, above inode belongs to other object. */ + if (result == -ENOENT) { + LASSERT(trusted); + + GOTO(out, result = 0); } + if (result != 0) + GOTO(out, result); + + if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) + goto trigger; + + /* It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. */ + trusted = true; + updated = true; + goto iget; + found: obj->oo_compat_dot_created = 1; obj->oo_compat_dotdot_created = 1; - if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */ + if (S_ISDIR(inode->i_mode) && + (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING)) + osd_check_lmv(info, dev, inode, oic); + + result = osd_attach_jinode(inode); + if (result) { + obj->oo_inode = NULL; + iput(inode); + GOTO(out, result); + } + + if (!ldiskfs_pdo) GOTO(out, result = 0); LASSERT(obj->oo_hl_head == NULL); @@ -779,7 +1246,7 @@ found: GOTO(out, result = 0); out: - if (result != 0 && cached) + if (result != 0 && trusted) fid_zero(&oic->oic_fid); LINVRNT(osd_invariant(obj)); @@ -817,13 +1284,158 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); obj->oo_dt.do_body_ops = &osd_body_ops_new; - if (result == 0 && obj->oo_inode != NULL) + if (result == 0 && obj->oo_inode != NULL) { + struct osd_thread_info *oti = osd_oti_get(env); + struct lustre_ost_attrs *loa = &oti->oti_ost_attrs; + osd_object_init0(obj); + result = osd_get_lma(oti, obj->oo_inode, + &oti->oti_obj_dentry, loa); + if (!result) { + /* Convert LMAI flags to lustre LMA flags + * and cache it to oo_lma_flags */ + obj->oo_lma_flags = + lma_to_lustre_flags(loa->loa_lma.lma_incompat); + } else if (result == -ENODATA) { + result = 0; + } + } LINVRNT(osd_invariant(obj)); return result; } +/* The first part of oxe_buf is xattr name, and is '\0' terminated. + * The left part is for value, binary mode. */ +struct osd_xattr_entry { + struct list_head oxe_list; + size_t oxe_len; + size_t oxe_namelen; + bool oxe_exist; + struct rcu_head oxe_rcu; + char oxe_buf[0]; +}; + +static int osd_oxc_get(struct osd_object *obj, const char *name, + struct lu_buf *buf) +{ + struct osd_xattr_entry *tmp; + struct osd_xattr_entry *oxe = NULL; + size_t namelen = strlen(name); + int rc; + ENTRY; + + rcu_read_lock(); + list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) { + if (namelen == tmp->oxe_namelen && + strncmp(name, tmp->oxe_buf, namelen) == 0) { + oxe = tmp; + break; + } + } + + if (oxe == NULL) + GOTO(out, rc = -ENOENT); + + if (!oxe->oxe_exist) + GOTO(out, rc = -ENODATA); + + /* vallen */ + rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1; + LASSERT(rc > 0); + + if (buf->lb_buf == NULL) + GOTO(out, rc); + + if (buf->lb_len < rc) + GOTO(out, rc = -ERANGE); + + memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc); + EXIT; +out: + rcu_read_unlock(); + + return rc; +} + +static void osd_oxc_free(struct rcu_head *head) +{ + struct osd_xattr_entry *oxe; + + oxe = container_of(head, struct osd_xattr_entry, oxe_rcu); + OBD_FREE(oxe, oxe->oxe_len); +} + +static void osd_oxc_add(struct osd_object *obj, const char *name, + const char *buf, int buflen) +{ + struct osd_xattr_entry *oxe; + struct osd_xattr_entry *old = NULL; + struct osd_xattr_entry *tmp; + size_t namelen = strlen(name); + size_t len = sizeof(*oxe) + namelen + 1 + buflen; + + OBD_ALLOC(oxe, len); + if (oxe == NULL) + return; + + INIT_LIST_HEAD(&oxe->oxe_list); + oxe->oxe_len = len; + oxe->oxe_namelen = namelen; + memcpy(oxe->oxe_buf, name, namelen); + if (buflen > 0) { + LASSERT(buf != NULL); + memcpy(oxe->oxe_buf + namelen + 1, buf, buflen); + oxe->oxe_exist = true; + } else { + oxe->oxe_exist = false; + } + + /* this should be rarely called, just remove old and add new */ + spin_lock(&obj->oo_guard); + list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) { + if (namelen == tmp->oxe_namelen && + strncmp(name, tmp->oxe_buf, namelen) == 0) { + old = tmp; + break; + } + } + if (old != NULL) { + list_replace_rcu(&old->oxe_list, &oxe->oxe_list); + call_rcu(&old->oxe_rcu, osd_oxc_free); + } else { + list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list); + } + spin_unlock(&obj->oo_guard); +} + +static void osd_oxc_del(struct osd_object *obj, const char *name) +{ + struct osd_xattr_entry *oxe; + size_t namelen = strlen(name); + + spin_lock(&obj->oo_guard); + list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) { + if (namelen == oxe->oxe_namelen && + strncmp(name, oxe->oxe_buf, namelen) == 0) { + list_del_rcu(&oxe->oxe_list); + call_rcu(&oxe->oxe_rcu, osd_oxc_free); + break; + } + } + spin_unlock(&obj->oo_guard); +} + +static void osd_oxc_fini(struct osd_object *obj) +{ + struct osd_xattr_entry *oxe, *next; + + list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) { + list_del(&oxe->oxe_list); + OBD_FREE(oxe, oxe->oxe_len); + } +} + /* * Concurrency: no concurrent access is possible that late in object * life-cycle. @@ -834,6 +1446,7 @@ static void osd_object_free(const struct lu_env *env, struct lu_object *l) LINVRNT(osd_invariant(obj)); + osd_oxc_fini(obj); dt_object_fini(&obj->oo_dt); if (obj->oo_hl_head != NULL) ldiskfs_htree_lock_head_free(obj->oo_hl_head); @@ -884,7 +1497,7 @@ enum { */ static void osd_th_alloced(struct osd_thandle *oth) { - oth->oth_alloced = cfs_time_current(); + oth->oth_alloced = ktime_get(); } /** @@ -892,58 +1505,42 @@ static void osd_th_alloced(struct osd_thandle *oth) */ static void osd_th_started(struct osd_thandle *oth) { - oth->oth_started = cfs_time_current(); + oth->oth_started = ktime_get(); } /** - * Helper function to convert time interval to microseconds packed in - * long int. + * Check whether the we deal with this handle for too long. */ -static long interval_to_usec(cfs_time_t start, cfs_time_t end) +static void __osd_th_check_slow(void *oth, struct osd_device *dev, + ktime_t alloced, ktime_t started, + ktime_t closed) { - struct timeval val; + ktime_t now = ktime_get(); - cfs_duration_usec(cfs_time_sub(end, start), &val); - return val.tv_sec * 1000000 + val.tv_usec; -} + LASSERT(dev != NULL); -/** - * Check whether the we deal with this handle for too long. - */ -static void __osd_th_check_slow(void *oth, struct osd_device *dev, - cfs_time_t alloced, cfs_time_t started, - cfs_time_t closed) -{ - cfs_time_t now = cfs_time_current(); - - LASSERT(dev != NULL); - - lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING, - interval_to_usec(alloced, started)); - lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN, - interval_to_usec(started, closed)); - lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING, - interval_to_usec(closed, now)); - - if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) { - CWARN("transaction handle %p was open for too long: " - "now "CFS_TIME_T" ," - "alloced "CFS_TIME_T" ," - "started "CFS_TIME_T" ," - "closed "CFS_TIME_T"\n", + lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING, + ktime_us_delta(started, alloced)); + lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN, + ktime_us_delta(closed, started)); + lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING, + ktime_us_delta(now, closed)); + + if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) { + CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n", oth, now, alloced, started, closed); libcfs_debug_dumpstack(NULL); } } -#define OSD_CHECK_SLOW_TH(oth, dev, expr) \ -{ \ - cfs_time_t __closed = cfs_time_current(); \ - cfs_time_t __alloced = oth->oth_alloced; \ - cfs_time_t __started = oth->oth_started; \ - \ - expr; \ - __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \ +#define OSD_CHECK_SLOW_TH(oth, dev, expr) \ +{ \ + ktime_t __closed = ktime_get(); \ + ktime_t __alloced = oth->oth_alloced; \ + ktime_t __started = oth->oth_started; \ + \ + expr; \ + __osd_th_check_slow(oth, dev, __alloced, __started, __closed); \ } #else /* OSD_THANDLE_STATS */ @@ -1002,6 +1599,11 @@ static void osd_trans_commit_cb(struct super_block *sb, OBD_FREE_PTR(oh); } +#ifndef HAVE_SB_START_WRITE +# define sb_start_write(sb) do {} while (0) +# define sb_end_write(sb) do {} while (0) +#endif + static struct thandle *osd_trans_create(const struct lu_env *env, struct dt_device *d) { @@ -1011,10 +1613,19 @@ static struct thandle *osd_trans_create(const struct lu_env *env, struct thandle *th; ENTRY; + if (d->dd_rdonly) { + CERROR("%s: someone try to start transaction under " + "readonly mode, should be disabled.\n", + osd_name(osd_dt_dev(d))); + dump_stack(); + RETURN(ERR_PTR(-EROFS)); + } + /* on pending IO in this thread should left from prev. request */ LASSERT(atomic_read(&iobuf->dr_numreqs) == 0); - th = ERR_PTR(-ENOMEM); + sb_start_write(osd_sb(osd_dt_dev(d))); + OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); if (oh != NULL) { oh->ot_quota_trans = &oti->oti_quota_trans; @@ -1034,6 +1645,9 @@ static struct thandle *osd_trans_create(const struct lu_env *env, sizeof(oti->oti_declare_ops_cred)); memset(oti->oti_declare_ops_used, 0, sizeof(oti->oti_declare_ops_used)); + } else { + sb_end_write(osd_sb(osd_dt_dev(d))); + th = ERR_PTR(-ENOMEM); } RETURN(th); } @@ -1114,16 +1728,15 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d, static unsigned long last_printed; static int last_credits; - CWARN("%.16s: too many transaction credits (%d > %d)\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, - oh->ot_credits, - osd_journal(dev)->j_max_transaction_buffers); - - osd_trans_dump_creds(env, th); - + /* don't make noise on a tiny testing systems + * actual credits misuse will be caught anyway */ if (last_credits != oh->ot_credits && time_after(jiffies, last_printed + - msecs_to_jiffies(60 * MSEC_PER_SEC))) { + msecs_to_jiffies(60 * MSEC_PER_SEC)) && + osd_transaction_size(dev) > 512) { + CWARN("%s: credits %u > trans_max %u\n", osd_name(dev), + oh->ot_credits, osd_transaction_size(dev)); + osd_trans_dump_creds(env, th); libcfs_debug_dumpstack(NULL); last_credits = oh->ot_credits; last_printed = jiffies; @@ -1135,6 +1748,14 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d, * This should be removed when we can calculate the * credits precisely. */ oh->ot_credits = osd_transaction_size(dev); + } else if (ldiskfs_track_declares_assert != 0) { + /* reserve few credits to prevent an assertion in JBD + * our debugging mechanism will be able to detected + * overuse. this can help to debug single-update + * transactions */ + oh->ot_credits += 10; + if (unlikely(osd_param_is_not_sane(dev, th))) + oh->ot_credits = osd_transaction_size(dev); } /* @@ -1175,8 +1796,8 @@ static int osd_seq_exists(const struct lu_env *env, 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); + CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n", + osd_name(osd), seq, rc); RETURN(0); } @@ -1216,12 +1837,16 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, oh = container_of0(th, struct osd_thandle, ot_super); + /* reset OI cache for safety */ + oti->oti_ins_cache_used = 0; + remove_agents = oh->ot_remove_agents; qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; if (oh->ot_handle != NULL) { + int rc2; handle_t *hdl = oh->ot_handle; /* @@ -1245,10 +1870,12 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, hdl->h_sync = th->th_sync; oh->ot_handle = NULL; - OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl)); - if (rc != 0) + OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl)); + if (rc2 != 0) CERROR("%s: failed to stop transaction: rc = %d\n", - osd_name(osd), rc); + osd_name(osd), rc2); + if (!rc) + rc = rc2; } else { osd_trans_stop_cb(oh, th->th_result); OBD_FREE_PTR(oh); @@ -1275,6 +1902,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, if (unlikely(remove_agents != 0)) osd_process_scheduled_agent_removals(env, osd); + sb_end_write(osd_sb(osd)); + RETURN(rc); } @@ -1330,6 +1959,9 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) qi->lqi_id.qid_uid = gid; qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA); + + qi->lqi_id.qid_uid = i_projid_read(inode); + qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA); } } } @@ -1340,6 +1972,12 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) static void osd_object_release(const struct lu_env *env, struct lu_object *l) { + struct osd_object *o = osd_obj(l); + /* nobody should be releasing a non-destroyed object with nlink=0 + * the API allows this, but ldiskfs doesn't like and then report + * this inode as deleted */ + if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0)) + LBUG(); } /* @@ -1363,18 +2001,17 @@ static int osd_object_print(const struct lu_env *env, void *cookie, d ? d->id_ops->id_name : "plain"); } -#define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */ - /* * Concurrency: shouldn't matter. */ int osd_statfs(const struct lu_env *env, struct dt_device *d, struct obd_statfs *sfs) { - struct osd_device *osd = osd_dt_dev(d); - struct super_block *sb = osd_sb(osd); - struct kstatfs *ksfs; - int result = 0; + struct osd_device *osd = osd_dt_dev(d); + struct super_block *sb = osd_sb(osd); + struct kstatfs *ksfs; + __u64 reserved; + int result = 0; if (unlikely(osd->od_mnt == NULL)) return -EINPROGRESS; @@ -1388,34 +2025,40 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, ksfs = &osd_oti_get(env)->oti_ksfs; } - spin_lock(&osd->od_osfs_lock); result = sb->s_op->statfs(sb->s_root, ksfs); - if (likely(result == 0)) { /* N.B. statfs can't really fail */ - statfs_pack(sfs, ksfs); - if (unlikely(sb->s_flags & MS_RDONLY)) - sfs->os_state = OS_STATE_READONLY; - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, - LDISKFS_FEATURE_INCOMPAT_EXTENTS)) - sfs->os_maxbytes = sb->s_maxbytes; - else - sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; - } - spin_unlock(&osd->od_osfs_lock); + if (result) + goto out; + + statfs_pack(sfs, ksfs); + if (unlikely(sb->s_flags & MS_RDONLY)) + sfs->os_state |= OS_STATE_READONLY; + if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, + LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + sfs->os_maxbytes = sb->s_maxbytes; + else + sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; - if (unlikely(env == NULL)) - OBD_FREE_PTR(ksfs); + /* + * Reserve some space so to avoid fragmenting the filesystem too much. + * Fragmentation not only impacts performance, but can also increase + * metadata overhead significantly, causing grant calculation to be + * wrong. + * + * Reserve 0.78% of total space, at least 8MB for small filesystems. + */ + CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE); + reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits; + if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT)) + reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT; - /* 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); - } + sfs->os_blocks -= reserved; + sfs->os_bfree -= min(reserved, sfs->os_bfree); + sfs->os_bavail -= min(reserved, sfs->os_bavail); - return result; +out: + if (unlikely(env == NULL)) + OBD_FREE_PTR(ksfs); + return result; } /** @@ -1444,21 +2087,23 @@ static void osd_conf_get(const struct lu_env *env, */ param->ddp_max_name_len = LDISKFS_NAME_LEN; param->ddp_max_nlink = LDISKFS_LINK_MAX; - param->ddp_block_shift = sb->s_blocksize_bits; + param->ddp_symlink_max = sb->s_blocksize; param->ddp_mount_type = LDD_MT_LDISKFS; if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)) param->ddp_maxbytes = sb->s_maxbytes; else param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; - /* Overhead estimate should be fairly accurate, so we really take a tiny - * error margin which also avoids fragmenting the filesystem too much */ - param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */ /* inode are statically allocated, so per-inode space consumption * is the space consumed by the directory entry */ param->ddp_inodespace = PER_OBJ_USAGE; - /* per-fragment overhead to be used by the client code */ - param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb); - param->ddp_mntopts = 0; + /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks + * = 128MB) which is unlikely to be hit in real life. Report a smaller + * maximum length to not under count the actual number of extents + * needed for writing a file. */ + param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2; + /* worst-case extent insertion metadata overhead */ + param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb); + param->ddp_mntopts = 0; if (test_opt(sb, XATTR_USER)) param->ddp_mntopts |= MNTOPT_USERXATTR; if (test_opt(sb, POSIX_ACL)) @@ -1537,6 +2182,12 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) #ifdef HAVE_DEV_SET_RDONLY CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname); + if (sb->s_op->freeze_fs) { + rc = sb->s_op->freeze_fs(sb); + if (rc) + goto out; + } + if (jdev && (jdev != dev)) { CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n", (long)jdev); @@ -1544,10 +2195,16 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) } 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); + + if (sb->s_op->unfreeze_fs) + sb->s_op->unfreeze_fs(sb); + +out: #endif + if (rc) + CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", + osd_dt_dev(d)->od_svname, (long)dev, rc); + RETURN(rc); } @@ -1703,14 +2360,13 @@ static struct timespec *osd_inode_time(const struct lu_env *env, return t; } - static void osd_inode_getattr(const struct lu_env *env, struct inode *inode, struct lu_attr *attr) { attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | - LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | - LA_TYPE; + LA_PROJID | 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); @@ -1720,11 +2376,19 @@ static void osd_inode_getattr(const struct lu_env *env, attr->la_blocks = inode->i_blocks; attr->la_uid = i_uid_read(inode); attr->la_gid = i_gid_read(inode); - attr->la_flags = LDISKFS_I(inode)->i_flags; + attr->la_projid = i_projid_read(inode); + attr->la_flags = ll_inode_to_ext_flags(inode->i_flags); attr->la_nlink = inode->i_nlink; attr->la_rdev = inode->i_rdev; attr->la_blksize = 1 << inode->i_blkbits; attr->la_blkbits = inode->i_blkbits; + /* + * Ext4 did not transfer inherit flags from raw inode + * to inode flags, and ext4 internally test raw inode + * @i_flags directly. Instead of patching ext4, we do it here. + */ + if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL) + attr->la_flags |= LUSTRE_PROJINHERIT_FL; } static int osd_attr_get(const struct lu_env *env, @@ -1733,7 +2397,9 @@ static int osd_attr_get(const struct lu_env *env, { struct osd_object *obj = osd_dt_obj(dt); - if (!dt_object_exists(dt)) + if (unlikely(!dt_object_exists(dt))) + return -ENOENT; + if (unlikely(obj->oo_destroyed)) return -ENOENT; LASSERT(!dt_object_remote(dt)); @@ -1741,10 +2407,72 @@ static int osd_attr_get(const struct lu_env *env, spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); + if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) + attr->la_flags |= LUSTRE_ORPHAN_FL; spin_unlock(&obj->oo_guard); + return 0; } +static int osd_declare_attr_qid(const struct lu_env *env, + struct osd_object *obj, + struct osd_thandle *oh, long long bspace, + qid_t old_id, qid_t new_id, bool enforce, + unsigned type) +{ + int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct lquota_id_info *qi = &info->oti_qi; + + qi->lqi_type = type; + /* inode accounting */ + qi->lqi_is_blk = false; + + /* one more inode for the new id ... */ + qi->lqi_id.qid_uid = new_id; + qi->lqi_space = 1; + /* Reserve credits for the new id */ + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* and one less inode for the current id */ + qi->lqi_id.qid_uid = old_id; + qi->lqi_space = -1; + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* block accounting */ + qi->lqi_is_blk = true; + + /* more blocks for the new uid ... */ + qi->lqi_id.qid_uid = new_id; + qi->lqi_space = bspace; + /* + * Credits for the new uid has been reserved, re-use "obj" + * to save credit reservation. + */ + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* and finally less blocks for the current uid */ + qi->lqi_id.qid_uid = old_id; + qi->lqi_space = -bspace; + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + + RETURN(rc); +} + static int osd_declare_attr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_attr *attr, @@ -1752,8 +2480,6 @@ static int osd_declare_attr_set(const struct lu_env *env, { struct osd_thandle *oh; struct osd_object *obj; - struct osd_thread_info *info = osd_oti_get(env); - struct lquota_id_info *qi = &info->oti_qi; qid_t uid; qid_t gid; long long bspace; @@ -1773,6 +2499,9 @@ static int osd_declare_attr_set(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, + osd_dto_credits_noquota[DTO_XATTR_SET]); + if (attr == NULL || obj->oo_inode == NULL) RETURN(rc); @@ -1789,103 +2518,33 @@ static int osd_declare_attr_set(const struct lu_env *env, if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { /* USERQUOTA */ uid = i_uid_read(obj->oo_inode); - qi->lqi_type = USRQUOTA; enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid); - /* inode accounting */ - qi->lqi_is_blk = false; - - /* one more inode for the new uid ... */ - qi->lqi_id.qid_uid = attr->la_uid; - qi->lqi_space = 1; - /* Reserve credits for the new uid */ - rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; - if (rc) - RETURN(rc); - - /* and one less inode for the current uid */ - qi->lqi_id.qid_uid = uid; - qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; - if (rc) - RETURN(rc); - - /* block accounting */ - qi->lqi_is_blk = true; - - /* more blocks for the new uid ... */ - qi->lqi_id.qid_uid = attr->la_uid; - qi->lqi_space = bspace; - /* - * Credits for the new uid has been reserved, re-use "obj" - * to save credit reservation. - */ - rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; - if (rc) - RETURN(rc); - - /* and finally less blocks for the current uid */ - qi->lqi_id.qid_uid = uid; - qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; + rc = osd_declare_attr_qid(env, obj, oh, bspace, uid, + attr->la_uid, enforce, USRQUOTA); if (rc) RETURN(rc); - /* GROUP QUOTA */ gid = i_gid_read(obj->oo_inode); - qi->lqi_type = GRPQUOTA; enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid); - - /* inode accounting */ - qi->lqi_is_blk = false; - - /* one more inode for the new gid ... */ - qi->lqi_id.qid_gid = attr->la_gid; - qi->lqi_space = 1; - rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; + rc = osd_declare_attr_qid(env, obj, oh, bspace, + i_gid_read(obj->oo_inode), attr->la_gid, + enforce, GRPQUOTA); if (rc) RETURN(rc); - /* and one less inode for the current gid */ - qi->lqi_id.qid_gid = gid; - qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; - if (rc) - RETURN(rc); - - /* block accounting */ - qi->lqi_is_blk = true; - - /* more blocks for the new gid ... */ - qi->lqi_id.qid_gid = attr->la_gid; - qi->lqi_space = bspace; - rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; - if (rc) - RETURN(rc); - - /* and finally less blocks for the current gid */ - qi->lqi_id.qid_gid = gid; - qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); - if (rc == -EDQUOT || rc == -EINPROGRESS) - rc = 0; + } +#ifdef HAVE_PROJECT_QUOTA + if (attr->la_valid & LA_PROJID) { + __u32 projid = i_projid_read(obj->oo_inode); + enforce = (attr->la_valid & LA_PROJID) && + (attr->la_projid != projid); + rc = osd_declare_attr_qid(env, obj, oh, bspace, + (qid_t)projid, (qid_t)attr->la_projid, + enforce, PRJQUOTA); if (rc) RETURN(rc); } - +#endif RETURN(rc); } @@ -1901,23 +2560,21 @@ static int osd_inode_setattr(const struct lu_env *env, if (bits == 0) return 0; - if (bits & LA_ATIME) - inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); - if (bits & LA_CTIME) - inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime); - if (bits & LA_MTIME) - inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime); - if (bits & LA_SIZE) { - LDISKFS_I(inode)->i_disksize = attr->la_size; - i_size_write(inode, attr->la_size); - } + if (bits & LA_ATIME) + inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); + if (bits & LA_CTIME) + inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime); + if (bits & LA_MTIME) + inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime); + if (bits & LA_SIZE) { + spin_lock(&inode->i_lock); + LDISKFS_I(inode)->i_disksize = attr->la_size; + i_size_write(inode, attr->la_size); + spin_unlock(&inode->i_lock); + } -#if 0 - /* OSD should not change "i_blocks" which is used by quota. - * "i_blocks" should be changed by ldiskfs only. */ - if (bits & LA_BLOCKS) - inode->i_blocks = attr->la_blocks; -#endif + /* OSD should not change "i_blocks" which is used by quota. + * "i_blocks" should be changed by ldiskfs only. */ if (bits & LA_MODE) inode->i_mode = (inode->i_mode & S_IFMT) | (attr->la_mode & ~S_IFMT); @@ -1925,25 +2582,35 @@ static int osd_inode_setattr(const struct lu_env *env, i_uid_write(inode, attr->la_uid); if (bits & LA_GID) i_gid_write(inode, attr->la_gid); + if (bits & LA_PROJID) + i_projid_write(inode, attr->la_projid); if (bits & LA_NLINK) set_nlink(inode, attr->la_nlink); if (bits & LA_RDEV) inode->i_rdev = attr->la_rdev; - if (bits & LA_FLAGS) { - /* always keep S_NOCMTIME */ - inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) | - S_NOCMTIME; - } - return 0; + if (bits & LA_FLAGS) { + /* always keep S_NOCMTIME */ + inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) | + S_NOCMTIME; + /* + * Ext4 did not transfer inherit flags from + * @inode->i_flags to raw inode i_flags when writing + * flags, we do it explictly here. + */ + if (attr->la_flags & LUSTRE_PROJINHERIT_FL) + LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL; + } + return 0; } static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) { + int rc; + if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) || (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) { struct iattr iattr; - int rc; ll_vfs_dq_init(inode); iattr.ia_valid = 0; @@ -1962,6 +2629,20 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) return rc; } } + +#ifdef HAVE_PROJECT_QUOTA + /* Handle project id transfer here properly */ + if (attr->la_valid & LA_PROJID && attr->la_projid != + i_projid_read(inode)) { + rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid); + if (rc) { + CERROR("%s: quota transfer failed: rc = %d. Is quota " + "enforcement enabled on the ldiskfs " + "filesystem?\n", inode->i_sb->s_id, rc); + return rc; + } + } +#endif return 0; } @@ -2017,10 +2698,42 @@ static int osd_attr_set(const struct lu_env *env, spin_lock(&obj->oo_guard); rc = osd_inode_setattr(env, inode, attr); spin_unlock(&obj->oo_guard); + if (rc != 0) + GOTO(out, rc); - if (!rc) - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + + if (!(attr->la_valid & LA_FLAGS)) + GOTO(out, rc); + + /* Let's check if there are extra flags need to be set into LMA */ + if (attr->la_flags & LUSTRE_LMA_FL_MASKS) { + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, + &info->oti_ost_attrs); + if (rc) + GOTO(out, rc); + + lma->lma_incompat |= + lustre_to_lma_flags(attr->la_flags); + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, + lma, sizeof(*lma), XATTR_REPLACE); + if (rc != 0) { + struct osd_device *osd = osd_obj2dev(obj); + + CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n", + osd_name(osd), PFID(lu_object_fid(&dt->do_lu)), + lma->lma_incompat, rc); + } else { + obj->oo_lma_flags = + attr->la_flags & LUSTRE_LMA_FL_MASKS; + } + osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); + } +out: osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); return rc; @@ -2090,22 +2803,22 @@ enum { }; static int osd_mkdir(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) { - int result; - struct osd_thandle *oth; - __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX)); + int result; + struct osd_thandle *oth; + __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID)); - LASSERT(S_ISDIR(attr->la_mode)); + LASSERT(S_ISDIR(attr->la_mode)); - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); - result = osd_mkfile(info, obj, mode, hint, th); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); + result = osd_mkfile(info, obj, mode, hint, th); - return result; + return result; } static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj, @@ -2238,6 +2951,12 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, ah->dah_parent = parent; ah->dah_mode = child_mode; + + if (parent != NULL && !dt_object_remote(parent)) { + /* will help to find FID->ino at dt_insert("..") */ + struct osd_object *pobj = osd_dt_obj(parent); + osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj); + } } static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, @@ -2262,20 +2981,20 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, if (result) return; - if (attr->la_valid != 0) { - result = osd_inode_setattr(info->oti_env, inode, attr); - /* - * The osd_inode_setattr() should always succeed here. The - * only error that could be returned is EDQUOT when we are - * trying to change the UID or GID of the inode. However, this - * should not happen since quota enforcement is no longer - * enabled on ldiskfs (lquota takes care of it). - */ + if (attr->la_valid != 0) { + result = osd_inode_setattr(info->oti_env, inode, attr); + /* + * The osd_inode_setattr() should always succeed here. The + * only error that could be returned is EDQUOT when we are + * trying to change the UID or GID of the inode. However, this + * should not happen since quota enforcement is no longer + * enabled on ldiskfs (lquota takes care of it). + */ LASSERTF(result == 0, "%d\n", result); ll_dirty_inode(inode, I_DIRTY_DATASYNC); - } + } - attr->la_valid = valid; + attr->la_valid = valid; } /** @@ -2342,7 +3061,8 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, osd_trans_exec_op(env, th, OSD_OT_INSERT); osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation); - rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD); + rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, + OI_CHECK_FLD, NULL); osd_trans_exec_check(env, th, OSD_OT_INSERT); return rc; @@ -2408,11 +3128,16 @@ static int osd_declare_object_create(const struct lu_env *env, if (!attr) RETURN(0); - rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh, - osd_dt_obj(dt), false, NULL, false); + rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, + attr->la_projid, 1, oh, osd_dt_obj(dt), + false, NULL, false); if (rc != 0) RETURN(rc); + /* will help to find FID->ino mapping at dt_insert() */ + rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)), + osd_dt_obj(dt)); + RETURN(rc); } @@ -2468,24 +3193,36 @@ static int osd_declare_object_destroy(const struct lu_env *env, int rc; ENTRY; + if (inode == NULL) + RETURN(-ENOENT); + oh = container_of0(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - LASSERT(inode); osd_trans_declare_op(env, oh, OSD_OT_DESTROY, osd_dto_credits_noquota[DTO_OBJECT_DELETE]); /* Recycle idle OI leaf may cause additional three OI blocks * to be changed. */ - osd_trans_declare_op(env, oh, OSD_OT_DELETE, + if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2)) + 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, i_uid_read(inode), i_gid_read(inode), - -1, oh, obj, false, NULL, false); + i_projid_read(inode), -1, oh, obj, false, + NULL, false); if (rc) RETURN(rc); /* data to be truncated */ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - 0, oh, obj, true, NULL, false); + i_projid_read(inode), 0, oh, obj, true, + NULL, false); + if (rc) + RETURN(rc); + + /* will help to find FID->ino when this object is being + * added to PENDING/ */ + rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj); + RETURN(rc); } @@ -2515,10 +3252,10 @@ static int osd_object_destroy(const struct lu_env *env, /* 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) { + if (result != 0) 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); @@ -2528,8 +3265,10 @@ static int osd_object_destroy(const struct lu_env *env, osd_trans_exec_op(env, th, OSD_OT_DESTROY); ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY); - result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle, - OI_CHECK_FLD); + + if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2)) + result = osd_oi_delete(osd_oti_get(env), osd, fid, + oh->ot_handle, OI_CHECK_FLD); osd_trans_exec_check(env, th, OSD_OT_DESTROY); /* XXX: add to ext3 orphan list */ @@ -2555,8 +3294,9 @@ static int osd_object_destroy(const struct lu_env *env, int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, const struct lu_fid *fid, __u32 compat, __u32 incompat) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - int rc; + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + int rc; ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) @@ -2565,33 +3305,50 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET)) rc = -ENOMEM; - 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); + lustre_loa_init(loa, fid, compat, incompat); + lustre_loa_swab(loa, false); + + /* For the OST device with 256 bytes inode size by default, + * the PFID EA will be stored together with LMA EA to avoid + * performance trouble. Otherwise the PFID EA can be stored + * independently. LU-8998 */ + if ((compat & LMAC_FID_ON_OST) && + LDISKFS_INODE_SIZE(inode->i_sb) <= 256) + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, + sizeof(*loa), XATTR_CREATE); + else + 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) + XATTR_NAME_LMA, (void *)loa, sizeof(*loa)); + if (rc < 0) + RETURN(rc); + + if (rc < sizeof(*lma)) RETURN(-EINVAL); - lustre_lma_swab(lma); - if (!(~lma->lma_compat & compat) && - !(~lma->lma_incompat & incompat)) + lustre_loa_swab(loa, true); + if (lu_fid_eq(fid, &lma->lma_self_fid) && + ((compat == 0 && incompat == 0) || + (!(~lma->lma_compat & compat) && + !(~lma->lma_incompat & incompat)))) RETURN(0); + lma->lma_self_fid = *fid; 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); + if (rc == sizeof(*lma)) { + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + } else { + lustre_loa_swab(loa, false); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, + sizeof(*loa), XATTR_REPLACE); + } } RETURN(rc); @@ -2652,6 +3409,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, { struct ldiskfs_dentry_param *dot_ldp; struct ldiskfs_dentry_param *dot_dot_ldp; + __u32 saved_nlink = dir->i_nlink; + int rc; dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); @@ -2659,8 +3418,23 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; dot_ldp->edp_magic = 0; - return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, - dir, dot_ldp, dot_dot_ldp); + rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, + dir, dot_ldp, dot_dot_ldp); + /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then + * the subseqent ref_add() will increase the dir->i_nlink + * as 3. That is incorrect for new created directory. + * + * It looks like hack, because we want to make the OSD API + * to be order-independent for new created directory object + * between dt_insert(..) and ref_add() operations. + * + * Here, we only restore the in-RAM dir-inode's nlink attr, + * becuase if the nlink attr is not 2, then there will be + * ref_add() called following the dt_insert(..), such call + * will make both the in-RAM and on-disk dir-inode's nlink + * attr to be set as 2. LU-7447 */ + set_nlink(dir, saved_nlink); + return rc; } /** @@ -2690,9 +3464,25 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, RETURN(local); } + /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set + * correct gid on remote file, not agent here */ + local->i_gid = current_fsgid(); ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB); unlock_new_inode(local); + /* Agent inode should not have project ID*/ +#ifdef HAVE_PROJECT_QUOTA + if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL) { + rc = __ldiskfs_ioctl_setproject(local, 0); + if (rc) { + CERROR("%s: quota transfer failed: rc = %d. Is project " + "quota enforcement enabled on the ldiskfs " + "filesystem?\n", local->i_sb->s_id, rc); + RETURN(ERR_PTR(rc)); + } + } + +#endif /* Set special LMA flag for local agent inode */ rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT); if (rc != 0) { @@ -2816,7 +3606,7 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, const struct lu_fid *fid = lu_object_fid(&dt->do_lu); struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *info = osd_oti_get(env); - int result; + int result, on_ost = 0; ENTRY; @@ -2842,13 +3632,15 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, fid_to_ostid(fid, oi); ostid_to_fid(tfid, oi, 0); + on_ost = 1; result = osd_ea_fid_set(info, obj->oo_inode, tfid, LMAC_FID_ON_OST, 0); } else { + on_ost = fid_is_on_ost(info, osd_obj2dev(obj), + fid, OI_CHECK_FLD); result = osd_ea_fid_set(info, obj->oo_inode, fid, - fid_is_on_ost(info, osd_obj2dev(obj), - fid, OI_CHECK_FLD) ? - LMAC_FID_ON_OST : 0, 0); + on_ost ? LMAC_FID_ON_OST : 0, + 0); } if (obj->oo_dt.do_body_ops == &osd_body_ops_new) obj->oo_dt.do_body_ops = &osd_body_ops; @@ -2857,6 +3649,15 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, if (result == 0) result = __osd_oi_insert(env, obj, fid, th); + /* a small optimization - dt_insert() isn't usually applied + * to OST objects, so we don't need to cache OI mapping for + * OST objects */ + if (result == 0 && on_ost == 0) { + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + result = osd_idc_find_and_init(env, osd, obj); + LASSERT(result == 0); + } + LASSERT(ergo(result == 0, dt_object_exists(dt) && !dt_object_remote(dt))); LINVRNT(osd_invariant(obj)); @@ -2892,7 +3693,7 @@ static int osd_object_ref_add(const struct lu_env *env, struct osd_thandle *oh; int rc = 0; - if (!dt_object_exists(dt)) + if (!dt_object_exists(dt) || obj->oo_destroyed) return -ENOENT; LINVRNT(osd_invariant(obj)); @@ -2943,6 +3744,9 @@ static int osd_declare_object_ref_del(const struct lu_env *env, { struct osd_thandle *oh; + if (!dt_object_exists(dt)) + return -ENOENT; + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); @@ -3007,44 +3811,39 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, } /* - * Get the 64-bit version for an inode. - */ -static int osd_object_version_get(const struct lu_env *env, - struct dt_object *dt, dt_obj_version_t *ver) -{ - struct inode *inode = osd_dt_obj(dt)->oo_inode; - - CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n", - LDISKFS_I(inode)->i_fs_version, inode->i_ino); - *ver = LDISKFS_I(inode)->i_fs_version; - return 0; -} - -/* * Concurrency: @dt is read locked. */ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, struct lu_buf *buf, const char *name) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + bool cache_xattr = false; + int rc; + + LASSERT(buf); - /* version get is not real XATTR but uses xattr API */ - if (strcmp(name, XATTR_NAME_VERSION) == 0) { - /* for version we are just using xattr API but change inode - * field instead */ + /* version get is not real XATTR but uses xattr API */ + if (strcmp(name, XATTR_NAME_VERSION) == 0) { + dt_obj_version_t *ver = buf->lb_buf; + + /* for version we are just using xattr API but change inode + * field instead */ if (buf->lb_len == 0) return sizeof(dt_obj_version_t); if (buf->lb_len < sizeof(dt_obj_version_t)) return -ERANGE; - osd_object_version_get(env, dt, buf->lb_buf); + CDEBUG(D_INODE, "Get version %#llx for inode %lu\n", + LDISKFS_I(inode)->i_fs_version, inode->i_ino); + + *ver = LDISKFS_I(inode)->i_fs_version; return sizeof(dt_obj_version_t); - } + } if (!dt_object_exists(dt)) return -ENOENT; @@ -3053,9 +3852,69 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, LASSERT(inode->i_op != NULL); LASSERT(inode->i_op->getxattr != NULL); - return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); -} + if (strcmp(name, XATTR_NAME_LOV) == 0 || + strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) + cache_xattr = true; + + if (cache_xattr) { + rc = osd_oxc_get(obj, name, buf); + if (rc != -ENOENT) + return rc; + } + + rc = __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); + if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) { + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + struct filter_fid *ff; + struct ost_layout *ol; + + LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost); + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (rc) + return rc; + + if (!(lma->lma_compat & LMAC_STRIPE_INFO)) { + rc = -ENODATA; + goto cache; + } + + rc = sizeof(*ff); + if (buf->lb_len == 0 || !buf->lb_buf) + return rc; + + if (buf->lb_len < rc) + return -ERANGE; + ff = buf->lb_buf; + ol = &ff->ff_layout; + ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >> + PFID_STRIPE_IDX_BITS); + ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size); + loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK; + fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid); + if (lma->lma_compat & LMAC_COMP_INFO) { + ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start); + ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end); + ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id); + } else { + ol->ol_comp_start = 0; + ol->ol_comp_end = 0; + ol->ol_comp_id = 0; + } + } + +cache: + if (cache_xattr) { + if (rc == -ENOENT || rc == -ENODATA) + osd_oxc_add(obj, name, NULL, 0); + else if (rc > 0 && buf->lb_buf != NULL) + osd_oxc_add(obj, name, buf->lb_buf, rc); + } + + return rc; +} static int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, @@ -3063,7 +3922,7 @@ static int osd_declare_xattr_set(const struct lu_env *env, int fl, struct thandle *handle) { struct osd_thandle *oh; - int credits; + int credits = 0; struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); LASSERT(handle != NULL); @@ -3075,17 +3934,25 @@ static int osd_declare_xattr_set(const struct lu_env *env, /* For non-upgrading case, the LMA is set first and * usually fit inode. But for upgrade case, the LMA * may be in another separated EA block. */ - if (!dt_object_exists(dt)) - credits = 0; - else if (fl == LU_XATTR_REPLACE) - credits = 1; - else - goto upgrade; + if (dt_object_exists(dt)) { + if (fl == LU_XATTR_REPLACE) + credits = 1; + else + goto upgrade; + } } else if (strcmp(name, XATTR_NAME_VERSION) == 0) { credits = 1; + } else if (strcmp(name, XATTR_NAME_FID) == 0) { + /* We may need to delete the old PFID EA. */ + credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb); + if (fl == LU_XATTR_REPLACE) + credits += 1; + else + goto upgrade; } else { + upgrade: - credits = osd_dto_credits_noquota[DTO_XATTR_SET]; + credits += osd_dto_credits_noquota[DTO_XATTR_SET]; if (buf != NULL) { ssize_t buflen; @@ -3117,27 +3984,9 @@ upgrade: credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); } - osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); - - return 0; -} - -/* - * Set the 64-bit version for object - */ -static void osd_object_version_set(const struct lu_env *env, - struct dt_object *dt, - dt_obj_version_t *new_version) -{ - struct inode *inode = osd_dt_obj(dt)->oo_inode; - - CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n", - *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino); + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); - LDISKFS_I(inode)->i_fs_version = *new_version; - /** Version is set after all inode operations are finished, - * so we should mark it dirty here */ - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + return 0; } /* @@ -3147,27 +3996,42 @@ 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 osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - int fs_flags = 0; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + int fs_flags = 0; + int len; + int rc; ENTRY; - LASSERT(handle != NULL); + LASSERT(handle); + LASSERT(buf); - /* version set is not real XATTR */ - if (strcmp(name, XATTR_NAME_VERSION) == 0) { - /* for version we are just using xattr API but change inode - * field instead */ - LASSERT(buf->lb_len == sizeof(dt_obj_version_t)); - osd_object_version_set(env, dt, buf->lb_buf); - return sizeof(dt_obj_version_t); - } + /* version set is not real XATTR */ + if (strcmp(name, XATTR_NAME_VERSION) == 0) { + dt_obj_version_t *version = buf->lb_buf; + + /* for version we are just using xattr API but change inode + * field instead */ + LASSERT(buf->lb_len == sizeof(dt_obj_version_t)); + + CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n", + *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino); + + LDISKFS_I(inode)->i_fs_version = *version; + /* Version is set after all inode operations are finished, + * so we should mark it dirty here */ + ll_dirty_inode(inode, I_DIRTY_DATASYNC); + + RETURN(0); + } CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n", PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); + len = buf->lb_len; osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); if (fl & LU_XATTR_REPLACE) fs_flags |= XATTR_REPLACE; @@ -3175,12 +4039,72 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (fl & LU_XATTR_CREATE) fs_flags |= XATTR_CREATE; - if (strcmp(name, XATTR_NAME_LMV) == 0) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - int rc; + /* For the OST device with 256 bytes inode size by default, + * the PFID EA will be stored together with LMA EA to avoid + * performance trouble. Otherwise the PFID EA can be stored + * independently. LU-8998 */ + if (strcmp(name, XATTR_NAME_FID) == 0 && + LDISKFS_INODE_SIZE(inode->i_sb) <= 256) { + struct dentry *dentry = &info->oti_obj_dentry; + struct filter_fid *ff; + struct ost_layout *ol; + int fl; + + LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost); + + ff = buf->lb_buf; + ol = &ff->ff_layout; + /* Old client does not send stripe information, store + * the PFID EA on disk directly. */ + if (buf->lb_len == sizeof(struct lu_fid) || + ol->ol_stripe_size == 0) { + len = sizeof(struct lu_fid); + goto set; + } - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc != 0) + if (buf->lb_len != sizeof(*ff)) + RETURN(-EINVAL); + + rc = osd_get_lma(info, inode, dentry, loa); + if (unlikely(rc == -ENODATA)) { + /* Usually for upgarding from old device */ + lustre_loa_init(loa, lu_object_fid(&dt->do_lu), + LMAC_FID_ON_OST, 0); + fl = XATTR_CREATE; + } else if (rc) { + RETURN(rc); + } else { + fl = XATTR_REPLACE; + } + + fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent); + loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) << + PFID_STRIPE_IDX_BITS; + loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size); + lma->lma_compat |= LMAC_STRIPE_INFO; + if (ol->ol_comp_id != 0) { + loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id); + loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start); + loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end); + lma->lma_compat |= LMAC_COMP_INFO; + } + + lustre_loa_swab(loa, false); + + /* Remove old PFID EA entry firstly. */ + ll_vfs_dq_init(inode); + rc = inode->i_op->removexattr(dentry, name); + if (rc && rc != -ENODATA) + RETURN(rc); + + /* Store the PFID EA inside the LMA EA. */ + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, + sizeof(*loa), fl); + + RETURN(rc); + } else if (strcmp(name, XATTR_NAME_LMV) == 0) { + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (rc) RETURN(rc); lma->lma_incompat |= LMAI_STRIPED; @@ -3191,14 +4115,15 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) && - strcmp(name, XATTR_NAME_LINK) == 0) - return -ENOSPC; - - rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, - fs_flags); +set: + rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags); osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); + if (rc == 0 && + (strcmp(name, XATTR_NAME_LOV) == 0 || + strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)) + osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len); + return rc; } @@ -3275,7 +4200,34 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, dentry->d_inode = inode; dentry->d_sb = inode->i_sb; rc = inode->i_op->removexattr(dentry, name); + if (rc == -ENODATA && strcmp(name, XATTR_NAME_FID) == 0) { + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; + + LASSERT(osd_dev(dt->do_lu.lo_dev)->od_is_ost); + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, + &info->oti_ost_attrs); + if (!rc) { + if (!(lma->lma_compat & LMAC_STRIPE_INFO)) { + rc = -ENODATA; + goto out; + } + + lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO); + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + } + } + +out: osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); + + if (rc == 0 && + (strcmp(name, XATTR_NAME_LOV) == 0 || + strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)) + osd_oxc_del(obj, name); + return rc; } @@ -3303,6 +4255,11 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } +static int osd_invalidate(const struct lu_env *env, struct dt_object *dt) +{ + return 0; +} + /* * Index operations. */ @@ -3375,7 +4332,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, result = 0; } else if (feat == &dt_directory_features) { dt->do_index_ops = &osd_index_ea_ops; - if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode)) + if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode)) result = 0; else result = -ENOTDIR; @@ -3429,10 +4386,6 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } LINVRNT(osd_invariant(obj)); - if (result == 0 && feat == &dt_quota_glb_features && - fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB) - result = osd_quota_migration(env, dt); - return result; } @@ -3470,6 +4423,7 @@ static const struct dt_object_operations osd_obj_ops = { .do_xattr_del = osd_xattr_del, .do_xattr_list = osd_xattr_list, .do_object_sync = osd_object_sync, + .do_invalidate = osd_invalidate, }; /** @@ -3502,6 +4456,7 @@ static const struct dt_object_operations osd_obj_ea_ops = { .do_xattr_del = osd_xattr_del, .do_xattr_list = osd_xattr_list, .do_object_sync = osd_object_sync, + .do_invalidate = osd_invalidate, }; static const struct dt_object_operations osd_obj_otable_it_ops = { @@ -3587,7 +4542,7 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, { struct osd_thandle *oh; struct inode *inode; - int rc; + int rc, credits; ENTRY; LASSERT(!dt_object_remote(dt)); @@ -3596,16 +4551,22 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - /* due to DNE we may need to remove an agent inode */ - osd_trans_declare_op(env, oh, OSD_OT_DELETE, - osd_dto_credits_noquota[DTO_INDEX_DELETE] + - osd_dto_credits_noquota[DTO_OBJECT_DELETE]); + credits = osd_dto_credits_noquota[DTO_INDEX_DELETE]; + if (key != NULL && unlikely(strcmp((char *)key, dotdot) == 0)) { + /* '..' to a remote object has a local representative */ + credits += osd_dto_credits_noquota[DTO_INDEX_DELETE]; + /* to reset LMAI_REMOTE_PARENT */ + credits += 1; + } + osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits); inode = osd_dt_obj(dt)->oo_inode; - LASSERT(inode); + if (inode == NULL) + RETURN(-ENOENT); rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - 0, oh, osd_dt_obj(dt), true, NULL, false); + i_projid_read(inode), 0, oh, osd_dt_obj(dt), + true, NULL, false); RETURN(rc); } @@ -3698,7 +4659,7 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, } bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); - if (bh) { + if (!IS_ERR(bh)) { /* 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. @@ -3727,11 +4688,11 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, le32_to_cpu(de->inode)); } } - rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); - brelse(bh); - } else { - rc = -ENOENT; - } + rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); + brelse(bh); + } else { + rc = PTR_ERR(bh); + } if (hlock != NULL) ldiskfs_htree_unlock(hlock); else @@ -3744,20 +4705,19 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, * /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); + int ret; + + ret = osd_delete_from_remote_parent(env, osd_obj2dev(obj), + obj, oh); + if (ret != 0) + /* Sigh, the entry has been deleted, and + * it is not easy to revert it back, so + * let's keep this error private, and let + * LFSCK fix it. XXX */ + CERROR("%s: delete remote parent "DFID": rc = %d\n", + osd_name(osd), PFID(fid), ret); } out: - LASSERT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); RETURN(rc); @@ -3948,7 +4908,8 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); child->d_fsdata = (void *)ldp; ll_vfs_dq_init(pobj->oo_inode); - rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); + rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle, + child, cinode, hlock); if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) { struct ldiskfs_dir_entry_2 *de; struct buffer_head *bh; @@ -3956,7 +4917,7 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de, NULL, hlock); - if (bh != NULL) { + if (!IS_ERR(bh)) { rc1 = ldiskfs_journal_get_write_access(oth->ot_handle, bh); if (rc1 == 0) { @@ -3968,8 +4929,8 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, LDISKFS_FT_DIR; ldiskfs_handle_dirty_metadata(oth->ot_handle, NULL, bh); - brelse(bh); } + brelse(bh); } } @@ -4007,7 +4968,7 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, if (dir->oo_compat_dot_created) { result = -EEXIST; } else { - LASSERT(inode == parent_dir); + LASSERT(inode->i_ino == parent_dir->i_ino); dir->oo_compat_dot_created = 1; result = 0; } @@ -4099,11 +5060,13 @@ static int osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, struct osd_idmap_cache *oic) { - struct osd_scrub *scrub = &dev->od_scrub; - struct lu_fid *fid = &oic->oic_fid; - struct osd_inode_id *id = &oti->oti_id; - int once = 0; - int rc; + struct osd_scrub *scrub = &dev->od_scrub; + struct lu_fid *fid = &oic->oic_fid; + struct osd_inode_id *id = &oic->oic_lid; + struct inode *inode = NULL; + int once = 0; + bool insert; + int rc; ENTRY; if (!fid_is_norm(fid) && !fid_is_igif(fid)) @@ -4113,13 +5076,15 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, RETURN(0); again: - rc = osd_oi_lookup(oti, dev, fid, id, 0); + rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0); if (rc == -ENOENT) { - struct inode *inode; + __u32 gen = id->oii_gen; - *id = oic->oic_lid; - inode = osd_iget(oti, dev, &oic->oic_lid); + insert = true; + if (inode != NULL) + goto trigger; + inode = osd_iget(oti, dev, id); /* The inode has been removed (by race maybe). */ if (IS_ERR(inode)) { rc = PTR_ERR(inode); @@ -4127,21 +5092,33 @@ again: RETURN(rc == -ESTALE ? -ENOENT : rc); } - iput(inode); /* The OI mapping is lost. */ - if (id->oii_gen != OSD_OII_NOGEN) + if (gen != OSD_OII_NOGEN) goto trigger; + iput(inode); /* The inode may has been reused by others, we do not know, * leave it to be handled by subsequent osd_fid_lookup(). */ RETURN(0); - } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) { + } else if (rc != 0 || osd_id_eq(id, &oti->oti_id)) { RETURN(rc); + } else { + insert = false; } trigger: if (thread_is_running(&scrub->os_thread)) { - rc = osd_oii_insert(dev, oic, rc == -ENOENT); + if (inode == NULL) { + inode = osd_iget(oti, dev, id); + /* The inode has been removed (by race maybe). */ + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + + RETURN(rc == -ESTALE ? -ENOENT : rc); + } + } + + rc = osd_oii_insert(dev, oic, insert); /* There is race condition between osd_oi_lookup and OI scrub. * The OI scrub finished just after osd_oi_lookup() failure. * Under such case, it is unnecessary to trigger OI scrub again, @@ -4149,21 +5126,31 @@ trigger: if (unlikely(rc == -EAGAIN)) goto again; - RETURN(0); + if (!S_ISDIR(inode->i_mode)) + rc = 0; + else + rc = osd_check_lmv(oti, dev, inode, oic); + + iput(inode); + RETURN(rc); } if (!dev->od_noscrub && ++once == 1) { rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT); - CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC " - "for "DFID", rc = %d [2]\n", + CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING, + "%.16s: trigger partial OI scrub for RPC inconsistency " + "checking FID "DFID": rc = %d\n", LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, PFID(fid), rc); if (rc == 0 || rc == -EALREADY) goto again; } - RETURN(0); + if (inode != NULL) + iput(inode); + + RETURN(rc); } static int osd_fail_fid_lookup(struct osd_thread_info *oti, @@ -4171,9 +5158,9 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, 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; + struct lustre_ost_attrs *loa = &oti->oti_ost_attrs; + struct inode *inode; + int rc; osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); inode = osd_iget(oti, dev, &oic->oic_lid); @@ -4182,12 +5169,12 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, return PTR_ERR(inode); } - rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma); + rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa); iput(inode); if (rc != 0) fid_zero(&oic->oic_fid); else - *fid = oic->oic_fid = lma->lma_self_fid; + *fid = oic->oic_fid = loa->loa_lma.lma_self_fid; return rc; } @@ -4269,8 +5256,8 @@ again: } ldata.ld_buf = buf; - rc = linkea_init(&ldata); - if (rc == 0) { + rc = linkea_init_with_rec(&ldata); + if (!rc) { linkea_first_entry(&ldata); linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid); } @@ -4278,6 +5265,48 @@ again: RETURN(rc); } +static int osd_verify_ent_by_linkea(const struct lu_env *env, + struct inode *inode, + const struct lu_fid *pfid, + const char *name, const int namelen) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct linkea_data ldata = { NULL }; + struct lu_name cname = { .ln_name = name, + .ln_namelen = namelen }; + int rc; + ENTRY; + +again: + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + buf->lb_buf, buf->lb_len); + if (rc == -ERANGE) + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0); + + if (rc < 0) + RETURN(rc); + + if (unlikely(rc == 0)) + RETURN(-ENODATA); + + if (buf->lb_len < rc) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + + ldata.ld_buf = buf; + rc = linkea_init_with_rec(&ldata); + if (!rc) + rc = linkea_links_find(&ldata, &cname, pfid); + + RETURN(rc); +} + /** * Calls ->lookup() to find dentry. From dentry get inode and * read inode's ea to get fid. This is required for interoperability @@ -4314,7 +5343,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, } bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); - if (bh) { + if (!IS_ERR(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; @@ -4360,7 +5389,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, if (rc != 0) fid_zero(&oic->oic_fid); } else { - rc = -ENOENT; + rc = PTR_ERR(bh); } GOTO(out, rc); @@ -4373,76 +5402,6 @@ out: return rc; } -/** - * Find the osd object for given fid. - * - * \param fid need to find the osd object having this fid - * - * \retval osd_object on success - * \retval -ve on error - */ -static struct osd_object *osd_object_find(const struct lu_env *env, - struct dt_object *dt, - const struct lu_fid *fid) -{ - struct lu_device *ludev = dt->do_lu.lo_dev; - struct osd_object *child = NULL; - struct lu_object *luch; - struct lu_object *lo; - - /* - * at this point topdev might not exist yet - * (i.e. MGS is preparing profiles). so we can - * not rely on topdev and instead lookup with - * our device passed as topdev. this can't work - * if the object isn't cached yet (as osd doesn't - * allocate lu_header). IOW, the object must be - * in the cache, otherwise lu_object_alloc() crashes - * -bzzz - */ - luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ? - ludev : ludev->ld_site->ls_top_dev, - fid, NULL); - if (!IS_ERR(luch)) { - if (lu_object_exists(luch)) { - lo = lu_object_locate(luch->lo_header, ludev->ld_type); - if (lo != NULL) - child = osd_obj(lo); - else - LU_OBJECT_DEBUG(D_ERROR, env, luch, - "lu_object can't be located" - DFID"\n", PFID(fid)); - - if (child == NULL) { - lu_object_put(env, luch); - CERROR("Unable to get osd_object\n"); - child = ERR_PTR(-ENOENT); - } - } else { - LU_OBJECT_DEBUG(D_ERROR, env, luch, - "lu_object does not exists "DFID"\n", - PFID(fid)); - lu_object_put(env, luch); - child = ERR_PTR(-ENOENT); - } - } else { - child = ERR_CAST(luch); - } - - return child; -} - -/** - * Put the osd object once done with it. - * - * \param obj osd object that needs to be put - */ -static inline void osd_object_put(const struct lu_env *env, - struct osd_object *obj) -{ - lu_object_put(env, &obj->oo_dt.do_lu); -} - static int osd_index_declare_ea_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_rec *rec, @@ -4451,28 +5410,34 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, { struct osd_thandle *oh; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct lu_fid *fid = (struct lu_fid *)rec; + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; int credits, rc = 0; + struct osd_idmap_cache *idc; ENTRY; LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); + LASSERT(fid != NULL); + LASSERT(rec1->rec_type != 0); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); credits = osd_dto_credits_noquota[DTO_INDEX_INSERT]; - if (fid != NULL) { - rc = osd_remote_fid(env, osd, fid); - if (unlikely(rc < 0)) - RETURN(rc); - if (rc > 0) { - /* a reference to remote inode is represented by an - * agent inode which we have to create */ - credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE]; - credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; - } - rc = 0; + + /* we can't call iget() while a transactions is running + * (this can lead to a deadlock), but we need to know + * inum and object type. so we find this information at + * declaration and cache in per-thread info */ + idc = osd_idc_find_or_init(env, osd, fid); + if (IS_ERR(idc)) + RETURN(PTR_ERR(idc)); + if (idc->oic_remote) { + /* a reference to remote inode is represented by an + * agent inode which we have to create */ + credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE]; + credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; } osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits); @@ -4484,8 +5449,10 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, * calculate how many blocks will be consumed by this index * insert */ rc = osd_declare_inode_qid(env, i_uid_read(inode), - i_gid_read(inode), 0, oh, - osd_dt_obj(dt), true, NULL, false); + i_gid_read(inode), + i_projid_read(inode), 0, + oh, osd_dt_obj(dt), true, + NULL, false); } RETURN(rc); @@ -4513,9 +5480,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, const struct lu_fid *fid = rec1->rec_fid; const char *name = (const char *)key; struct osd_thread_info *oti = osd_oti_get(env); - struct osd_inode_id *id = &oti->oti_id; struct inode *child_inode = NULL; - struct osd_object *child = NULL; + struct osd_idmap_cache *idc; int rc; ENTRY; @@ -4530,14 +5496,22 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid)); - rc = osd_remote_fid(env, osd, fid); - if (rc < 0) { - CERROR("%s: Can not find object "DFID" rc %d\n", - osd_name(osd), PFID(fid), rc); - RETURN(rc); + idc = osd_idc_find(env, osd, fid); + if (unlikely(idc == NULL)) { + /* this dt_insert() wasn't declared properly, so + * FID is missing in OI cache. we better do not + * lookup FID in FLDB/OI and don't risk to deadlock, + * but in some special cases (lfsck testing, etc) + * it's much simpler than fixing a caller */ + CERROR("%s: "DFID" wasn't declared for insert\n", + osd_name(osd), PFID(fid)); + dump_stack(); + idc = osd_idc_find_or_init(env, osd, fid); + if (IS_ERR(idc)) + RETURN(PTR_ERR(idc)); } - if (rc == 1) { + if (idc->oic_remote) { /* Insert remote entry */ if (strcmp(name, dotdot) == 0 && strlen(name) == 2) { struct osd_mdobj_map *omm = osd->od_mdt_map; @@ -4563,15 +5537,23 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, } } else { /* Insert local entry */ - child = osd_object_find(env, dt, fid); - if (IS_ERR(child)) { - CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n", - osd_name(osd), PFID(fid), - id->oii_ino, id->oii_gen, - (int)PTR_ERR(child)); - RETURN(PTR_ERR(child)); + if (unlikely(idc->oic_lid.oii_ino == 0)) { + /* for a reason OI cache wasn't filled properly */ + CERROR("%s: OIC for "DFID" isn't filled\n", + osd_name(osd), PFID(fid)); + RETURN(-EINVAL); } - child_inode = igrab(child->oo_inode); + child_inode = oti->oti_inode; + if (unlikely(child_inode == NULL)) { + struct ldiskfs_inode_info *lii; + OBD_ALLOC_PTR(lii); + if (lii == NULL) + RETURN(-ENOMEM); + child_inode = oti->oti_inode = &lii->vfs_inode; + } + child_inode->i_sb = osd_sb(osd); + child_inode->i_ino = idc->oic_lid.oii_ino; + child_inode->i_mode = rec1->rec_type & S_IFMT; } rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th); @@ -4579,9 +5561,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n", obj->oo_inode->i_ino, name, child_inode->i_ino, rc); - iput(child_inode); - if (child != NULL) - osd_object_put(env, child); + if (child_inode && child_inode != oti->oti_inode) + iput(child_inode); LASSERT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_INSERT); RETURN(rc); @@ -4626,7 +5607,6 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, /** * free given Iterator. */ - static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) { struct osd_it_iam *it = (struct osd_it_iam *)di; @@ -4634,7 +5614,7 @@ static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) iam_it_fini(&it->oi_it); osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); - lu_object_put(env, &obj->oo_dt.do_lu); + osd_object_put(env, obj); OBD_FREE_PTR(it); } @@ -4939,7 +5919,7 @@ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) ENTRY; oie->oie_file.f_op->release(inode, &oie->oie_file); - lu_object_put(env, &obj->oo_dt.do_lu); + osd_object_put(env, obj); if (unlikely(oie->oie_buf != info->oti_it_ea_buf)) OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE); else @@ -4997,11 +5977,16 @@ struct osd_filldir_cbs { * \retval 0 on success * \retval 1 on buffer full */ -static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, - loff_t offset, __u64 ino, - unsigned d_type) +#ifdef HAVE_FILLDIR_USE_CTX +static int osd_ldiskfs_filldir(struct dir_context *buf, +#else +static int osd_ldiskfs_filldir(void *buf, +#endif + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned d_type) { - struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it; + struct osd_it_ea *it = + ((struct osd_filldir_cbs *)buf)->it; struct osd_object *obj = it->oie_obj; struct osd_it_ea_dirent *ent = it->oie_dirent; struct lu_fid *fid = &ent->oied_fid; @@ -5177,14 +6162,10 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) return it->oie_dirent->oied_namelen; } -static inline bool -osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot) +static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de) { - 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))) + __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack))) return true; return false; @@ -5192,10 +6173,10 @@ osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot) static inline bool osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen, - unsigned blocksize, int dot_dotdot) + unsigned blocksize, bool dotdot) { - if (dot_dotdot > 0) - return osd_dot_dotdot_has_space(de, dot_dotdot); + if (dotdot) + return osd_dotdot_has_space(de); if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >= __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack))) @@ -5205,10 +6186,11 @@ osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen, } static int -osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, - struct dentry *dentry, const struct lu_fid *fid, - struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de, - struct htree_lock *hlock, int dot_dotdot) +osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev, + handle_t *jh, struct dentry *dentry, + const struct lu_fid *fid, struct buffer_head *bh, + struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock, + bool dotdot) { struct inode *dir = dentry->d_parent->d_inode; struct inode *inode = dentry->d_inode; @@ -5216,6 +6198,7 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, struct ldiskfs_dentry_param *ldp; int namelen = dentry->d_name.len; int rc; + struct osd_thread_info *info = osd_oti_get(env); ENTRY; if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb, @@ -5223,8 +6206,7 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, RETURN(0); /* There is enough space to hold the FID-in-dirent. */ - if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, - dot_dotdot)) { + if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) { rc = ldiskfs_journal_get_write_access(jh, bh); if (rc != 0) RETURN(rc); @@ -5239,7 +6221,7 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, RETURN(rc); } - LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot); + LASSERT(!dotdot); rc = ldiskfs_delete_entry(jh, dir, de, bh); if (rc != 0) @@ -5249,7 +6231,7 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, osd_get_ldiskfs_dirent_param(ldp, fid); dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); - rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); + rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock); /* It is too bad, we cannot reinsert the name entry back. * That means we lose it! */ if (rc != 0) @@ -5268,11 +6250,10 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, 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 lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; 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; + const char *devname = osd_name(dev); struct osd_it_ea_dirent *ent = it->oie_dirent; struct inode *dir = obj->oo_inode; struct htree_lock *hlock = NULL; @@ -5281,25 +6262,37 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, struct ldiskfs_dir_entry_2 *de; struct dentry *dentry; struct inode *inode; + const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu); int credits; int rc; - int dot_dotdot = 0; + bool dotdot = false; bool dirty = false; ENTRY; + if (ent->oied_name[0] == '.') { + if (ent->oied_namelen == 1) + RETURN(0); + + if (ent->oied_namelen == 2 && ent->oied_name[1] == '.') + dotdot = true; + } + 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) { + /* Maybe dangling name entry, or + * corrupted directory entry. */ *attr |= LUDA_UNKNOWN; rc = 0; } else { - CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent " - "check_repair, dir = %lu/%u, name = %.*s: " - "rc = %d\n", + CDEBUG(D_LFSCK, "%s: fail to iget() for dirent " + "check_repair, dir = %lu/%u, name = %.*s, " + "ino = %llu, rc = %d\n", devname, dir->i_ino, dir->i_generation, - ent->oied_namelen, ent->oied_name, rc); + ent->oied_namelen, ent->oied_name, + ent->oied_ino, rc); } RETURN(rc); @@ -5307,19 +6300,12 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name, ent->oied_namelen); - rc = osd_get_lma(info, inode, dentry, lma); - if (rc == -ENODATA) + rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs); + if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid)) lma = NULL; else if (rc != 0) GOTO(out, rc); - 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; - } - /* We need to ensure that the name entry is still valid. * Because it may be removed or renamed by other already. * @@ -5342,11 +6328,12 @@ again: jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); - CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent " + CDEBUG(D_LFSCK, "%s: fail to start trans for dirent " "check_repair, dir = %lu/%u, credits = %d, " - "name = %.*s: rc = %d\n", + "name = %.*s, ino = %llu: rc = %d\n", devname, dir->i_ino, dir->i_generation, credits, - ent->oied_namelen, ent->oied_name, rc); + ent->oied_namelen, ent->oied_name, + ent->oied_ino, rc); GOTO(out_inode, rc); } @@ -5372,120 +6359,127 @@ again: } 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 + if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) { + *attr |= LUDA_IGNORE; + + GOTO(out, rc = 0); + } + + /* For 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 + * For the whole directory, only 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) != inode->i_ino) || - (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) { - *attr |= LUDA_IGNORE; + if (dotdot && !osd_dotdot_has_space(de)) { + *attr |= LUDA_UNKNOWN; GOTO(out, rc = 0); } if (lma != NULL) { + if (lu_fid_eq(fid, &lma->lma_self_fid)) + GOTO(out, rc = 0); + if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) { struct lu_fid *tfid = &lma->lma_self_fid; - *attr |= LUDA_IGNORE; - /* It must be REMOTE_PARENT_DIR and as the - * dotdot entry of remote directory */ - if (unlikely(dot_dotdot != 2 || - fid_seq(tfid) != FID_SEQ_LOCAL_FILE || - fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) { - CDEBUG(D_LFSCK, "%.16s: expect remote agent " + if (likely(dotdot && + fid_seq(tfid) == FID_SEQ_LOCAL_FILE && + fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) { + /* It must be REMOTE_PARENT_DIR and as the + * 'dotdot' entry of remote directory */ + *attr |= LUDA_IGNORE; + } else { + CDEBUG(D_LFSCK, "%s: expect remote agent " "parent directory, but got %.*s under " "dir = %lu/%u with the FID "DFID"\n", devname, ent->oied_namelen, ent->oied_name, dir->i_ino, dir->i_generation, PFID(tfid)); - GOTO(out, rc = -EIO); + *attr |= LUDA_UNKNOWN; } GOTO(out, rc = 0); } + } - if (fid_is_sane(fid)) { - /* FID-in-dirent is valid. */ - if (lu_fid_eq(fid, &lma->lma_self_fid)) - GOTO(out, rc = 0); - - /* Do not repair under dryrun mode. */ - if (*attr & LUDA_VERIFY_DRYRUN) { - *attr |= LUDA_REPAIR; + if (!fid_is_zero(fid)) { + rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name, + ent->oied_namelen); + if (rc == -ENOENT || + (rc == -ENODATA && + !(dev->od_scrub.os_file.sf_flags & SF_UPGRADE))) { + /* linkEA does not recognize the dirent entry, + * it may because the dirent entry corruption + * and points to other's inode. */ + CDEBUG(D_LFSCK, "%s: the target inode does not " + "recognize the dirent, dir = %lu/%u, " + " name = %.*s, ino = %llu, " + DFID": rc = %d\n", devname, dir->i_ino, + dir->i_generation, ent->oied_namelen, + ent->oied_name, ent->oied_ino, PFID(fid), rc); + *attr |= LUDA_UNKNOWN; - GOTO(out, rc = 0); - } + GOTO(out, rc = 0); + } - if (jh == NULL) { - brelse(bh); - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else - up_read(&obj->oo_ext_idx_sem); - dev->od_dirent_journal = 1; + if (rc && rc != -ENODATA) { + CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, " + "dir = %lu/%u, name = %.*s, ino = %llu, " + DFID": rc = %d\n", devname, dir->i_ino, + dir->i_generation, ent->oied_namelen, + ent->oied_name, ent->oied_ino, PFID(fid), rc); + *attr |= LUDA_UNKNOWN; - goto again; - } + GOTO(out, rc = 0); + } + } + if (lma != NULL) { + /* linkEA recognizes the dirent entry, the FID-in-LMA is + * valid, trusted, in spite of fid_is_sane(fid) or not. */ + if (*attr & LUDA_VERIFY_DRYRUN) { *fid = lma->lma_self_fid; - dirty = true; - /* Update the FID-in-dirent. */ - rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de, - hlock, dot_dotdot); - if (rc == 0) - *attr |= LUDA_REPAIR; - else - CDEBUG(D_LFSCK, "%.16s: fail to update FID " - "in the dirent, dir = %lu/%u, " - "name = %.*s, "DFID": rc = %d\n", - devname, dir->i_ino, dir->i_generation, - ent->oied_namelen, ent->oied_name, - PFID(fid), rc); - } else { - /* Do not repair under dryrun mode. */ - if (*attr & LUDA_VERIFY_DRYRUN) { - *fid = lma->lma_self_fid; - *attr |= LUDA_REPAIR; - - GOTO(out, rc = 0); - } + *attr |= LUDA_REPAIR; - if (jh == NULL) { - brelse(bh); - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else - up_read(&obj->oo_ext_idx_sem); - dev->od_dirent_journal = 1; + GOTO(out, rc = 0); + } - goto again; + if (jh == NULL) { + brelse(bh); + dev->od_dirent_journal = 1; + if (hlock != NULL) { + ldiskfs_htree_unlock(hlock); + hlock = NULL; + } else { + up_read(&obj->oo_ext_idx_sem); } - *fid = lma->lma_self_fid; - dirty = true; - /* Append the FID-in-dirent. */ - rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de, - hlock, dot_dotdot); - if (rc == 0) - *attr |= LUDA_REPAIR; - else - CDEBUG(D_LFSCK, "%.16s: fail to append FID " - "after the dirent, dir = %lu/%u, " - "name = %.*s, "DFID": rc = %d\n", - devname, dir->i_ino, dir->i_generation, - ent->oied_namelen, ent->oied_name, - PFID(fid), rc); + goto again; } + + *fid = lma->lma_self_fid; + dirty = true; + /* Update or append the FID-in-dirent. */ + rc = osd_dirent_reinsert(env, dev, jh, dentry, fid, + bh, de, hlock, dotdot); + if (rc == 0) + *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%s: fail to re-insert FID after " + "the dirent, dir = %lu/%u, name = %.*s, " + "ino = %llu, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + ent->oied_ino, PFID(fid), rc); } else { - /* Do not repair under dryrun mode. */ + /* lma is NULL, trust the FID-in-dirent if it is valid. */ if (*attr & LUDA_VERIFY_DRYRUN) { if (fid_is_sane(fid)) { *attr |= LUDA_REPAIR; - } else { + } else if (dev->od_index == 0) { lu_igif_build(fid, inode->i_ino, inode->i_generation); *attr |= LUDA_UPGRADE; @@ -5496,11 +6490,13 @@ again: if (jh == NULL) { brelse(bh); - if (hlock != NULL) + dev->od_dirent_journal = 1; + if (hlock != NULL) { ldiskfs_htree_unlock(hlock); - else + hlock = NULL; + } else { up_read(&obj->oo_ext_idx_sem); - dev->od_dirent_journal = 1; + } goto again; } @@ -5513,34 +6509,37 @@ again: if (rc == 0) *attr |= LUDA_REPAIR; else - CDEBUG(D_LFSCK, "%.16s: fail to set LMA for " + CDEBUG(D_LFSCK, "%s: fail to set LMA for " "update dirent, dir = %lu/%u, " - "name = %.*s, "DFID": rc = %d\n", + "name = %.*s, ino = %llu, " + DFID": rc = %d\n", devname, dir->i_ino, dir->i_generation, ent->oied_namelen, ent->oied_name, - PFID(fid), rc); - } else { + ent->oied_ino, PFID(fid), rc); + } else if (dev->od_index == 0) { 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, dentry, fid, bh, de, - hlock, dot_dotdot); + rc = osd_dirent_reinsert(env, dev, jh, dentry, fid, + bh, de, hlock, dotdot); if (rc == 0) *attr |= LUDA_UPGRADE; else - CDEBUG(D_LFSCK, "%.16s: fail to append IGIF " + CDEBUG(D_LFSCK, "%s: fail to append IGIF " "after the dirent, dir = %lu/%u, " - "name = %.*s, "DFID": rc = %d\n", + "name = %.*s, ino = %llu, " + DFID": rc = %d\n", devname, dir->i_ino, dir->i_generation, ent->oied_namelen, ent->oied_name, - PFID(fid), rc); + ent->oied_ino, PFID(fid), rc); } } GOTO(out, rc); out: - brelse(bh); + if (!IS_ERR(bh)) + brelse(bh); if (hlock != NULL) { ldiskfs_htree_unlock(hlock); } else { @@ -5604,10 +6603,8 @@ static inline int osd_it_ea_rec(const struct lu_env *env, &attr); } - if (!fid_is_sane(fid)) { - attr &= ~LUDA_IGNORE; + if (!fid_is_sane(fid)) attr |= LUDA_UNKNOWN; - } } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { @@ -5798,15 +6795,23 @@ static void osd_key_fini(const struct lu_context *ctx, struct lu_context_key *key, void* data) { struct osd_thread_info *info = data; + struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode); + struct osd_idmap_cache *idc = info->oti_ins_cache; if (info->oti_inode != NULL) - OBD_FREE_PTR(info->oti_inode); + OBD_FREE_PTR(lli); if (info->oti_hlock != NULL) ldiskfs_htree_lock_free(info->oti_hlock); OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); lu_buf_free(&info->oti_iobuf.dr_pg_buf); lu_buf_free(&info->oti_iobuf.dr_bl_buf); lu_buf_free(&info->oti_big_buf); + if (idc != NULL) { + LASSERT(info->oti_ins_cache_size > 0); + OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size); + info->oti_ins_cache = NULL; + info->oti_ins_cache_size = 0; + } OBD_FREE_PTR(info); } @@ -5931,7 +6936,7 @@ static int osd_mount(const struct lu_env *env, struct osd_thread_info *info = osd_oti_get(env); struct lu_fid *fid = &info->oti_fid; struct inode *inode; - int rc = 0, force_over_128tb = 0; + int rc = 0, force_over_512tb = 0; ENTRY; if (o->od_mnt != NULL) @@ -5955,10 +6960,27 @@ static int osd_mount(const struct lu_env *env, RETURN(-EINVAL); } #endif - if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) - force_over_128tb = 1; +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0) + if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) { + CWARN("force_over_128tb option is deprecated. " + "Filesystems less than 512TB can be created without any " + "force options. Use force_over_512tb option for " + "filesystems greater than 512TB.\n"); + } +#endif +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0) + if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) { + CWARN("force_over_256tb option is deprecated. " + "Filesystems less than 512TB can be created without any " + "force options. Use force_over_512tb option for " + "filesystems greater than 512TB.\n"); + } +#endif + + if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) + force_over_512tb = 1; - __page = alloc_page(GFP_IOFS); + __page = alloc_page(GFP_KERNEL); if (__page == NULL) GOTO(out, rc = -ENOMEM); page = (unsigned long)page_address(__page); @@ -5976,6 +6998,8 @@ static int osd_mount(const struct lu_env *env, /* strip out option we processed in osd */ "bigendian_extents", "force_over_128tb", + "force_over_256tb", + "force_over_512tb", NULL }; strcat(options, opts); @@ -6003,7 +7027,7 @@ static int osd_mount(const struct lu_env *env, /* Glom up mount options */ if (*options != '\0') strcat(options, ","); - strlcat(options, "no_mbcache", PAGE_CACHE_SIZE); + strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE); type = get_fs_type("ldiskfs"); if (!type) { @@ -6021,22 +7045,37 @@ static int osd_mount(const struct lu_env *env, GOTO(out, rc); } - if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) && - force_over_128tb == 0) { + if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) << + osd_sb(o)->s_blocksize_bits > 512ULL << 40 && + force_over_512tb == 0) { CERROR("%s: device %s LDISKFS does not support filesystems " - "greater than 128TB and can cause data corruption. " - "Use \"force_over_128tb\" mount option to override.\n", + "greater than 512TB and can cause data corruption. " + "Use \"force_over_512tb\" mount option to override.\n", name, dev); - GOTO(out, rc = -EINVAL); + GOTO(out_mnt, rc = -EINVAL); } + if (lmd_flags & LMD_FLG_DEV_RDONLY) { #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); - GOTO(out_mnt, rc = -EROFS); - } + dev_set_rdonly(osd_sb(o)->s_bdev); + o->od_dt_dev.dd_rdonly = 1; + LCONSOLE_WARN("%s: set dev_rdonly on this device\n", name); +#else + LCONSOLE_WARN("%s: not support dev_rdonly on this device", + name); + + GOTO(out_mnt, rc = -EOPNOTSUPP); +#endif + } else { +#ifdef HAVE_DEV_SET_RDONLY + if (dev_check_rdonly(osd_sb(o)->s_bdev)) { + CERROR("%s: underlying device %s is marked as " + "read-only. Setup failed\n", name, dev); + + GOTO(out_mnt, rc = -EROFS); + } #endif + } if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { @@ -6437,7 +7476,7 @@ static struct obd_ops osd_obd_device_ops = { .o_health_check = osd_health_check, }; -static int __init osd_mod_init(void) +static int __init osd_init(void) { int rc; @@ -6461,7 +7500,7 @@ static int __init osd_mod_init(void) return rc; } -static void __exit osd_mod_exit(void) +static void __exit osd_exit(void) { class_unregister_type(LUSTRE_OSD_LDISKFS_NAME); lu_kmem_fini(ldiskfs_caches); @@ -6472,5 +7511,5 @@ MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")"); MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -module_init(osd_mod_init); -module_exit(osd_mod_exit); +module_init(osd_init); +module_exit(osd_exit);