X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=d03fb6bf65a0d4fcb04aa42b7853d81933ef5c09;hp=633a6eeae6654ae2e4a0df778f5175b2713361a2;hb=381a8cdd527ce4deccfc3f7eb461892f6f2f3fff;hpb=6c9384678f5555f87bcc7311bb5bf73b1e4bf4e4;ds=sidebyside diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 633a6ee..d03fb6b 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, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,14 +39,13 @@ #define DEBUG_SUBSYSTEM S_OSD +#include #include #include #ifdef HAVE_UIDGID_HEADER # include #endif -/* LUSTRE_VERSION_CODE */ -#include /* prerequisite for linux/xattr.h */ #include /* prerequisite for linux/xattr.h */ @@ -71,7 +66,7 @@ #include #include /* process_config */ -#include +#include #include "osd_internal.h" #include "osd_dynlocks.h" @@ -82,13 +77,16 @@ #include +/* Maximum EA size is limited by LNET_MTU for remote objects */ +#define OSD_MAX_EA_SIZE 1048364 + 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; @@ -114,11 +112,9 @@ static struct lu_kmem_descr ldiskfs_caches[] = { static const char dot[] = "."; static const char dotdot[] = ".."; -static const char remote_obj_dir[] = "REM_OBJ_DIR"; static const struct lu_object_operations osd_lu_obj_ops; static const struct dt_object_operations osd_obj_ops; -static const struct dt_object_operations osd_obj_ea_ops; static const struct dt_object_operations osd_obj_otable_it_ops; static const struct dt_index_operations osd_index_iam_ops; static const struct dt_index_operations osd_index_ea_ops; @@ -144,35 +140,32 @@ int osd_trans_declare_op2rb[] = { static int osd_has_index(const struct osd_object *obj) { - return obj->oo_dt.do_index_ops != NULL; + return obj->oo_dt.do_index_ops != NULL; } static int osd_object_invariant(const struct lu_object *l) { - return osd_invariant(osd_obj(l)); + return osd_invariant(osd_obj(l)); } /* * Concurrency: doesn't matter */ - -/* - * Concurrency: doesn't matter - */ -static int osd_write_locked(const struct lu_env *env, struct osd_object *o) +static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o) { - struct osd_thread_info *oti = osd_oti_get(env); - return oti->oti_w_locks > 0 && o->oo_owner == env; + struct osd_thread_info *oti = osd_oti_get(env); + + return oti->oti_w_locks > 0 && o->oo_owner == env; } /* * Concurrency: doesn't access mutable data */ static int osd_root_get(const struct lu_env *env, - struct dt_device *dev, struct lu_fid *f) + struct dt_device *dev, struct lu_fid *f) { - lu_local_obj_fid(f, OSD_FS_ROOT_OID); - return 0; + lu_local_obj_fid(f, OSD_FS_ROOT_OID); + return 0; } /* @@ -190,9 +183,10 @@ 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; + 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; @@ -209,8 +203,8 @@ 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; + 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)) { @@ -259,10 +253,16 @@ osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd, if (idc != NULL) return idc; + CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n", + osd->od_svname, PFID(fid)); + /* new mapping is needed */ idc = osd_idc_add(env, osd, fid); - if (IS_ERR(idc)) + if (IS_ERR(idc)) { + CERROR("%s: FID "DFID" add id map cache failed: %ld\n", + osd->od_svname, PFID(fid), PTR_ERR(idc)); return idc; + } /* initialize it */ rc = osd_remote_fid(env, osd, fid); @@ -294,8 +294,8 @@ 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; + 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)); @@ -310,10 +310,16 @@ static int osd_idc_find_and_init(const struct lu_env *env, return 0; } + CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n", + osd->od_svname, PFID(fid)); + /* new mapping is needed */ idc = osd_idc_add(env, osd, fid); - if (IS_ERR(idc)) + if (IS_ERR(idc)) { + CERROR("%s: FID "DFID" add id map cache failed: %ld\n", + osd->od_svname, PFID(fid), PTR_ERR(idc)); return PTR_ERR(idc); + } if (obj->oo_inode != NULL) { idc->oic_lid.oii_ino = obj->oo_inode->i_ino; @@ -331,46 +337,65 @@ static int osd_idc_find_and_init(const struct lu_env *env, * life-cycle. */ static struct lu_object *osd_object_alloc(const struct lu_env *env, - const struct lu_object_header *hdr, - struct lu_device *d) -{ - struct osd_object *mo; + const struct lu_object_header *hdr, + struct lu_device *d) +{ + struct osd_object *mo; + + OBD_ALLOC_PTR(mo); + if (mo != NULL) { + struct lu_object *l; + struct lu_object_header *h; + struct osd_device *o = osd_dev(d); + + l = &mo->oo_dt.do_lu; + if (unlikely(o->od_in_init)) { + OBD_ALLOC_PTR(h); + if (!h) { + OBD_FREE_PTR(mo); + return NULL; + } - OBD_ALLOC_PTR(mo); - if (mo != NULL) { - struct lu_object *l; + lu_object_header_init(h); + lu_object_init(l, h, d); + lu_object_add_top(h, l); + mo->oo_header = h; + } else { + dt_object_init(&mo->oo_dt, NULL, d); + mo->oo_header = NULL; + } - l = &mo->oo_dt.do_lu; - dt_object_init(&mo->oo_dt, NULL, d); - mo->oo_dt.do_ops = &osd_obj_ea_ops; - l->lo_ops = &osd_lu_obj_ops; + mo->oo_dt.do_ops = &osd_obj_ops; + l->lo_ops = &osd_lu_obj_ops; init_rwsem(&mo->oo_sem); init_rwsem(&mo->oo_ext_idx_sem); spin_lock_init(&mo->oo_guard); - return l; - } else { - return NULL; - } + INIT_LIST_HEAD(&mo->oo_xattr_list); + return l; + } + return NULL; } 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 " + CWARN("%s: unsupported incompat LMA feature(s) %#x " "for fid = "DFID", ino = %lu\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + osd_ino2name(inode), lma->lma_incompat & ~LMA_INCOMPAT_SUPP, PFID(&lma->lma_self_fid), inode->i_ino); rc = -EOPNOTSUPP; @@ -388,12 +413,15 @@ 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 - * transaction, then we risk to deadlock */ - /* osd_dirent_check_repair() breaks this */ - /*LASSERT(current->journal_info == NULL);*/ + /* + * if we look for an inode withing a running + * transaction, then we risk to deadlock + * osd_dirent_check_repair() breaks this + */ + /* LASSERT(current->journal_info == NULL); */ inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { @@ -407,33 +435,40 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, iput(inode); inode = ERR_PTR(-ESTALE); } else if (inode->i_nlink == 0) { - /* due to parallel readdir and unlink, - * we can have dead inode here. */ + /* + * due to parallel readdir and unlink, + * we can have dead inode here. + */ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); iput(inode); inode = ERR_PTR(-ESTALE); } else if (is_bad_inode(inode)) { - CWARN("%.16s: bad inode: ino = %u\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino); + CWARN("%s: bad inode: ino = %u\n", + osd_dev2name(dev), 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) osd_id_gen(id, inode->i_ino, inode->i_generation); - /* Do not update file c/mtime in ldiskfs. + /* + * Do not update file c/mtime in ldiskfs. * NB: we don't have any lock to protect this because we don't * have reference on osd_object now, but contention with * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ + * between if (...) and set S_NOCMTIME. + */ if (!(inode->i_flags & S_NOCMTIME)) inode->i_flags |= S_NOCMTIME; } return inode; } -int osd_ldiskfs_add_entry(struct osd_thread_info *info, +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) { @@ -441,60 +476,59 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, rc = __ldiskfs_add_entry(handle, child, inode, hlock); if (rc == -ENOBUFS || rc == -ENOSPC) { - char fidbuf[FID_LEN + 1]; - struct lustre_mdt_attrs lma; - struct lu_fid fid = { }; - char *errstr; - struct dentry *p_dentry = child->d_parent; - - rc2 = osd_get_lma(info, p_dentry->d_inode, p_dentry, - &lma); - if (rc2 == 0) { - fid = lma.lma_self_fid; - snprintf(fidbuf, sizeof(fidbuf), DFID, PFID(&fid)); + 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(p_dentry->d_inode == - inode->i_sb->s_root->d_inode)) - lu_local_obj_fid(&fid, OSD_FS_ROOT_OID); - else if (info->oti_dev && !info->oti_dev->od_is_ost && - fid_seq_is_mdt0(fid_seq(&fid))) - lu_igif_build(&fid, p_dentry->d_inode->i_ino, - p_dentry->d_inode->i_generation); - snprintf(fidbuf, sizeof(fidbuf), DFID, PFID(&fid)); - } else { - snprintf(fidbuf, FID_LEN, "%s", "unknown"); + if (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 (rc == -ENOSPC) - errstr = "has reached"; + 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 - errstr = "is approaching"; - CWARN("%.16s: directory (inode: %lu FID: %s) %s maximum entry limit\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - p_dentry->d_inode->i_ino, fidbuf, errstr); + 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 * +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); @@ -511,73 +545,76 @@ 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 + /* + * 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. */ + * 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); - - GOTO(put, rc); - } + if (!trusted) + goto check_oi; - 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); + if (!trusted) + goto check_oi; - GOTO(put, rc); - } - - 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); - /* XXX: There are four possible cases: + /* + * XXX: There are four possible cases: * 1. rc = 0. * Backup/restore caused the OI invalid. * 2. rc = 0. @@ -592,30 +629,65 @@ 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 { - /* If the OI mapping was in OI file before the + * 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. */ + * 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 && saved_gen != OSD_OII_NOGEN)) { + if (!IS_ERR(inode)) + iput(inode); + + trusted = true; + goto again; + } + + if (IS_ERR(inode)) { + if (dev->od_scrub.os_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); - /* Do not update file c/mtime in ldiskfs. + /* + * Do not update file c/mtime in ldiskfs. * NB: we don't have any lock to protect this because we don't * have reference on osd_object now, but contention with * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ + * between if (...) and set S_NOCMTIME. + */ if (!(inode->i_flags & S_NOCMTIME)) inode->i_flags |= S_NOCMTIME; } @@ -634,26 +706,30 @@ put: } /** - * \retval +v: new filter_fid, does not contain self-fid - * \retval 0: filter_fid_old, contains self-fid + * \retval +v: new filter_fid does not contain self-fid + * \retval 0: filter_fid_18_23, contains self-fid * \retval -v: other failure cases */ int osd_get_idif(struct osd_thread_info *info, struct inode *inode, struct dentry *dentry, struct lu_fid *fid) { - struct filter_fid_old *ff = &info->oti_ff; - struct ost_id *ostid = &info->oti_ostid; - int rc; + struct filter_fid *ff = &info->oti_ff; + struct ost_id *ostid = &info->oti_ostid; + int rc; rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff)); - if (rc == sizeof(*ff)) { - rc = 0; - ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq)); - ostid_set_id(ostid, le64_to_cpu(ff->ff_objid)); - /* XXX: use 0 as the index for compatibility, the caller will - * handle index related issues when necessarry. */ - ostid_to_fid(fid, ostid, 0); - } else if (rc == sizeof(struct filter_fid)) { + if (rc == sizeof(struct filter_fid_18_23)) { + struct filter_fid_18_23 *ff_old = (void *)ff; + + ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq)); + rc = ostid_set_id(ostid, le64_to_cpu(ff_old->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 >= (int)sizeof(struct filter_fid_24_29)) { rc = 1; } else if (rc >= 0) { rc = -EINVAL; @@ -667,7 +743,7 @@ static int osd_lma_self_repair(struct osd_thread_info *info, const struct lu_fid *fid, __u32 compat) { handle_t *jh; - int rc; + int rc; LASSERT(current->journal_info == NULL); @@ -690,28 +766,31 @@ static int osd_lma_self_repair(struct osd_thread_info *info, static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_device *osd = osd_obj2dev(obj); - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct inode *inode = obj->oo_inode; - struct dentry *dentry = &info->oti_obj_dentry; - struct lu_fid *fid = NULL; - const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_device *osd = osd_obj2dev(obj); + 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; + const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); + int rc; + ENTRY; - CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, - info->oti_mdt_attrs_old, LMA_OLD_SIZE); + (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); - if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) { - /* For the given OST-object, if it has neither LMA nor + if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) { + /* + * For the given OST-object, if it has neither LMA nor * FID in XATTR_NAME_FID, then the given FID (which is * contained in the @obj, from client RPC for locating * the OST-object) is trusted. We use it to generate - * the LMA. */ + * the LMA. + */ osd_lma_self_repair(info, osd, inode, rfid, LMAC_FID_ON_OST); RETURN(0); @@ -725,7 +804,8 @@ 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, @@ -733,6 +813,12 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) rc = -EOPNOTSUPP; } else { fid = &lma->lma_self_fid; + if (lma->lma_compat & LMAC_STRIPE_INFO && + osd->od_is_ost) + obj->oo_pfid_in_lma = 1; + if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) && + !osd->od_is_ost) + lu_object_set_agent_entry(&obj->oo_dt.do_lu); } } @@ -742,10 +828,12 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) struct lu_fid *fid1 = &info->oti_fid3; __u32 idx = fid_idif_ost_idx(rfid); - /* For old IDIF, the OST index is not part of the IDIF, + /* + * For old IDIF, the OST index is not part of the IDIF, * Means that different OSTs may have the same IDIFs. * Under such case, we need to make some compatible - * check to make sure to trigger OI scrub properly. */ + * check to make sure to trigger OI scrub properly. + */ if (idx != 0 && fid_idif_ost_idx(fid) == 0) { /* Given @rfid is new, LMA is old. */ fid_to_ostid(fid, oi); @@ -766,22 +854,222 @@ 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; + int oclb_items; + bool oclb_found; +}; + +/** + * 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 int 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; + + oclb->oclb_items++; + + 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; + osd_oii_insert(dev, oic, true); + oclb->oclb_found = true; + + return 1; +} + +/* + * 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, + .oclb_found = false, + }; + 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); + + do { + oclb.oclb_items = 0; +#ifdef HAVE_DIR_CONTEXT + oclb.ctx.pos = filp->f_pos; +#ifdef HAVE_ITERATE_SHARED + rc = fops->iterate_shared(filp, &oclb.ctx); +#else + rc = fops->iterate(filp, &oclb.ctx); +#endif + filp->f_pos = oclb.ctx.pos; +#else + rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir); +#endif + } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found && + filp->f_pos != LDISKFS_HTREE_EOF_64BIT); + fops->release(inode, filp); + +out: + if (rc < 0) + CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u," + DFID": rc = %d\n", osd_ino2name(inode), + 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 lustre_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; + bool checked = false; + ENTRY; LINVRNT(osd_invariant(obj)); @@ -789,7 +1077,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid)); dev = osd_dev(ldev); - scrub = &dev->od_scrub; + scrub = &dev->od_scrub.os_scrub; sf = &scrub->os_file; info = osd_oti_get(env); LASSERT(info); @@ -798,17 +1086,18 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) RETURN(-ENOENT); - /* For the object is created as locking anchor, or for the object to + /* + * For the object is created as locking anchor, or for the object to * be created on disk. No need to osd_oi_lookup() at here because FID * shouldn't never be re-used, if it's really a duplicate FID from * unexpected reason, we should be able to detect it later by calling - * do_create->osd_oi_insert(). */ - if (conf != NULL && conf->loc_flags & LOC_F_NEW) + * do_create->osd_oi_insert(). + */ + if (conf && conf->loc_flags & LOC_F_NEW) GOTO(out, result = 0); /* Search order: 1. per-thread cache. */ - if (lu_fid_eq(fid, &oic->oic_fid) && - likely(oic->oic_dev == dev)) { + if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) { id = &oic->oic_lid; goto iget; } @@ -817,181 +1106,278 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (!list_empty(&scrub->os_inconsistent_items)) { /* Search order: 2. OI scrub pending list. */ result = osd_oii_lookup(dev, fid, id); - if (result == 0) + if (!result) 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) { if (!(fid_is_norm(fid) || fid_is_igif(fid)) || fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || - !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), + !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); goto trigger; } - if (result != 0) + if (result) GOTO(out, result); iget: - inode = osd_iget_check(info, dev, fid, id, cached); - if (IS_ERR(inode)) { - result = PTR_ERR(inode); - if (result == -ENOENT || result == -ESTALE) - GOTO(out, result = -ENOENT); + obj->oo_inode = NULL; + /* for later passes through checks, not true on first pass */ + if (!IS_ERR_OR_NULL(inode)) + iput(inode); - if (result == -EREMCHG) { + inode = osd_iget_check(info, dev, fid, id, trusted); + if (!IS_ERR(inode)) { + obj->oo_inode = inode; + result = 0; + if (remote) + goto trigger; -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; - } else { - result = -EREMCHG; - } + goto check_lma; + } - if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) - GOTO(out, result); + result = PTR_ERR(inode); + if (result == -ENOENT || result == -ESTALE) + GOTO(out, result = 0); - /* We still have chance to get the valid inode: for the - * object which is referenced by remote name entry, the - * object on the local MDT will be linked under the dir - * of "/REMOTE_PARENT_DIR" with its FID string as name. - * - * We do not know whether the object for the given FID - * is referenced by some remote name entry or not, and - * especially for DNE II, a multiple-linked object may - * have many name entries reside on many MDTs. - * - * To simplify the operation, OSD will not distinguish - * more, just lookup "/REMOTE_PARENT_DIR". Usually, it - * only happened for the RPC from other MDT during the - * OI scrub, or for the client side RPC with FID only, - * such as FID to path, or from old connected client. */ - saved = result; - result = osd_lookup_in_remote_parent(info, dev, - fid, id); - if (result == 0) { - cached = true; - goto iget; - } + if (result != -EREMCHG) + GOTO(out, result); - result = saved; +trigger: + /* + * We still have chance to get the valid inode: for the + * object which is referenced by remote name entry, the + * object on the local MDT will be linked under the dir + * of "/REMOTE_PARENT_DIR" with its FID string as name. + * + * We do not know whether the object for the given FID + * is referenced by some remote name entry or not, and + * especially for DNE II, a multiple-linked object may + * have many name entries reside on many MDTs. + * + * To simplify the operation, OSD will not distinguish + * more, just lookup "/REMOTE_PARENT_DIR". Usually, it + * only happened for the RPC from other MDT during the + * OI scrub, or for the client side RPC with FID only, + * such as FID to path, or from old connected client. + */ + if (!remote) { + rc1 = osd_lookup_in_remote_parent(info, dev, fid, id); + if (!rc1) { + remote = true; + trusted = true; + flags |= SS_AUTO_PARTIAL; + flags &= ~SS_AUTO_FULL; + goto iget; } - - GOTO(out, result); } - obj->oo_inode = inode; - LASSERT(obj->oo_inode->i_sb == osd_sb(dev)); + if (thread_is_running(&scrub->os_thread)) { + if (scrub->os_partial_scan && !scrub->os_in_join) + goto join; - 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 (IS_ERR_OR_NULL(inode) || result) + GOTO(out, result = -EINPROGRESS); - goto found; - } + LASSERT(remote); + LASSERT(obj->oo_inode == inode); - iput(inode); - obj->oo_inode = NULL; - if (result != -EREMCHG) - GOTO(out, result); + osd_add_oi_cache(info, dev, id, fid); + osd_oii_insert(dev, oic, true); + 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); + if (dev->od_auto_scrub_interval == AS_NEVER) { + if (!remote) + GOTO(out, result = -EREMCHG); - /* 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. */ - } + LASSERT(!result); + LASSERT(obj->oo_inode == inode); - goto trigger; + osd_add_oi_cache(info, dev, id, fid); + goto found; } -found: - obj->oo_compat_dot_created = 1; - obj->oo_compat_dotdot_created = 1; +join: + rc1 = osd_scrub_start(env, dev, flags); + LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags " + "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1); + if (rc1 && rc1 != -EALREADY) + GOTO(out, result = -EREMCHG); - if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */ - GOTO(out, result = 0); + if (IS_ERR_OR_NULL(inode) || result) + GOTO(out, result = -EINPROGRESS); - LASSERT(obj->oo_hl_head == NULL); - obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); - if (obj->oo_hl_head == NULL) { - obj->oo_inode = NULL; - iput(inode); - GOTO(out, result = -ENOMEM); - } - GOTO(out, result = 0); + LASSERT(remote); + LASSERT(obj->oo_inode == inode); -out: - if (result != 0 && cached) - fid_zero(&oic->oic_fid); + osd_add_oi_cache(info, dev, id, fid); + osd_oii_insert(dev, oic, true); + goto found; - LINVRNT(osd_invariant(obj)); - return result; -} +check_lma: + checked = true; + if (unlikely(obj->oo_header)) + goto found; -/* - * Concurrency: shouldn't matter. + result = osd_check_lma(env, obj); + if (!result) + 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; + + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); + if (!result) { + /* + * 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; + + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ + trusted = true; + updated = true; + goto iget; + } + + /* + * "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); + + obj->oo_inode = NULL; + result = 0; + } + } + + 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); + + obj->oo_inode = NULL; + GOTO(out, result = 0); + } + + if (result) + GOTO(out, result); + + if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) { + result = -EREMCHG; + 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: + if (!checked) { + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; + + result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (!result) { + if (lma->lma_compat & LMAC_STRIPE_INFO && + dev->od_is_ost) + obj->oo_pfid_in_lma = 1; + if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) && + !dev->od_is_ost) + lu_object_set_agent_entry(&obj->oo_dt.do_lu); + } else if (result != -ENODATA) { + GOTO(out, result); + } + } + + obj->oo_compat_dot_created = 1; + obj->oo_compat_dotdot_created = 1; + + 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) + GOTO(out, result); + + if (!ldiskfs_pdo) + GOTO(out, result = 0); + + LASSERT(!obj->oo_hl_head); + obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); + + GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0)); + +out: + if (result || !obj->oo_inode) { + if (!IS_ERR_OR_NULL(inode)) + iput(inode); + + obj->oo_inode = NULL; + if (trusted) + fid_zero(&oic->oic_fid); + } + + LINVRNT(osd_invariant(obj)); + return result; +} + +/* + * Concurrency: shouldn't matter. */ static void osd_object_init0(struct osd_object *obj) { - LASSERT(obj->oo_inode != NULL); - obj->oo_dt.do_body_ops = &osd_body_ops; - obj->oo_dt.do_lu.lo_header->loh_attr |= - (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); + LASSERT(obj->oo_inode != NULL); + obj->oo_dt.do_body_ops = &osd_body_ops; + obj->oo_dt.do_lu.lo_header->loh_attr |= + (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); } /* @@ -1016,16 +1402,21 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, obj->oo_dt.do_body_ops = &osd_body_ops_new; if (result == 0 && obj->oo_inode != NULL) { struct osd_thread_info *oti = osd_oti_get(env); - struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs; + struct lustre_ost_attrs *loa = &oti->oti_ost_attrs; osd_object_init0(obj); + if (unlikely(obj->oo_header)) + return 0; + result = osd_get_lma(oti, obj->oo_inode, - &oti->oti_obj_dentry, lma); - if (result == 0) { - /* Convert LMAI flags to lustre LMA flags - * and cache it to oo_lma_flags */ + &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(lma->lma_incompat); + lma_to_lustre_flags(loa->loa_lma.lma_incompat); } else if (result == -ENODATA) { result = 0; } @@ -1036,19 +1427,159 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, } /* + * 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. */ static void osd_object_free(const struct lu_env *env, struct lu_object *l) { - struct osd_object *obj = osd_obj(l); + struct osd_object *obj = osd_obj(l); + struct lu_object_header *h = obj->oo_header; - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - dt_object_fini(&obj->oo_dt); - if (obj->oo_hl_head != NULL) - ldiskfs_htree_lock_head_free(obj->oo_hl_head); - OBD_FREE_PTR(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); + OBD_FREE_PTR(obj); + if (unlikely(h)) { + lu_object_header_fini(h); + OBD_FREE_PTR(h); + } } /* @@ -1057,17 +1588,17 @@ static void osd_object_free(const struct lu_env *env, struct lu_object *l) */ static void osd_index_fini(struct osd_object *o) { - struct iam_container *bag; + struct iam_container *bag; - if (o->oo_dir != NULL) { - bag = &o->oo_dir->od_container; - if (o->oo_inode != NULL) { - if (bag->ic_object == o->oo_inode) - iam_container_fini(bag); - } - OBD_FREE_PTR(o->oo_dir); - o->oo_dir = NULL; - } + if (o->oo_dir != NULL) { + bag = &o->oo_dir->od_container; + if (o->oo_inode != NULL) { + if (bag->ic_object == o->oo_inode) + iam_container_fini(bag); + } + OBD_FREE_PTR(o->oo_dir); + o->oo_dir = NULL; + } } /* @@ -1077,12 +1608,12 @@ static void osd_index_fini(struct osd_object *o) */ static int osd_inode_unlinked(const struct inode *inode) { - return inode->i_nlink == 0; + return inode->i_nlink == 0; } enum { - OSD_TXN_OI_DELETE_CREDITS = 20, - OSD_TXN_INODE_DELETE_CREDITS = 20 + OSD_TXN_OI_DELETE_CREDITS = 20, + OSD_TXN_INODE_DELETE_CREDITS = 20 }; /* @@ -1095,7 +1626,7 @@ enum { */ static void osd_th_alloced(struct osd_thandle *oth) { - oth->oth_alloced = cfs_time_current(); + oth->oth_alloced = ktime_get(); } /** @@ -1103,58 +1634,42 @@ static void osd_th_alloced(struct osd_thandle *oth) */ static void osd_th_started(struct osd_thandle *oth) { - oth->oth_started = cfs_time_current(); -} - -/** - * Helper function to convert time interval to microseconds packed in - * long int. - */ -static long interval_to_usec(cfs_time_t start, cfs_time_t end) -{ - struct timeval val; - - cfs_duration_usec(cfs_time_sub(end, start), &val); - return val.tv_sec * 1000000 + val.tv_usec; + oth->oth_started = ktime_get(); } /** * 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) + ktime_t alloced, ktime_t started, + ktime_t closed) { - cfs_time_t now = cfs_time_current(); + ktime_t now = ktime_get(); - LASSERT(dev != NULL); + 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)); + 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 (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", - oth, now, alloced, started, closed); - libcfs_debug_dumpstack(NULL); - } + 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 */ @@ -1180,19 +1695,19 @@ static int osd_param_is_not_sane(const struct osd_device *dev, * Concurrency: shouldn't matter. */ static void osd_trans_commit_cb(struct super_block *sb, - struct ldiskfs_journal_cb_entry *jcb, int error) + struct ldiskfs_journal_cb_entry *jcb, int error) { - struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); - struct thandle *th = &oh->ot_super; - struct lu_device *lud = &th->th_dev->dd_lu_dev; - struct dt_txn_commit_cb *dcb, *tmp; + struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); + struct thandle *th = &oh->ot_super; + struct lu_device *lud = &th->th_dev->dd_lu_dev; + struct dt_txn_commit_cb *dcb, *tmp; - LASSERT(oh->ot_handle == NULL); + LASSERT(oh->ot_handle == NULL); - if (error) - CERROR("transaction @0x%p commit error: %d\n", th, error); + if (error) + CERROR("transaction @0x%p commit error: %d\n", th, error); - dt_txn_hook_commit(th); + dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list, @@ -1205,54 +1720,73 @@ static void osd_trans_commit_cb(struct super_block *sb, } lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th); - lu_device_put(lud); - th->th_dev = NULL; + lu_device_put(lud); + th->th_dev = NULL; - lu_context_exit(&th->th_ctx); - lu_context_fini(&th->th_ctx); 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) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_thandle *oh; - struct thandle *th; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_thandle *oh; + 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); - OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); - if (oh != NULL) { - oh->ot_quota_trans = &oti->oti_quota_trans; - memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); - th = &oh->ot_super; - th->th_dev = d; - th->th_result = 0; - th->th_tags = LCT_TX_HANDLE; - oh->ot_credits = 0; - INIT_LIST_HEAD(&oh->ot_commit_dcb_list); - INIT_LIST_HEAD(&oh->ot_stop_dcb_list); - osd_th_alloced(oh); - - memset(oti->oti_declare_ops, 0, - sizeof(oti->oti_declare_ops)); - memset(oti->oti_declare_ops_cred, 0, - sizeof(oti->oti_declare_ops_cred)); - memset(oti->oti_declare_ops_used, 0, - sizeof(oti->oti_declare_ops_used)); + sb_start_write(osd_sb(osd_dt_dev(d))); + + OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS); + if (!oh) { + sb_end_write(osd_sb(osd_dt_dev(d))); + RETURN(ERR_PTR(-ENOMEM)); } + + oh->ot_quota_trans = &oti->oti_quota_trans; + memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); + th = &oh->ot_super; + th->th_dev = d; + th->th_result = 0; + oh->ot_credits = 0; + INIT_LIST_HEAD(&oh->ot_commit_dcb_list); + INIT_LIST_HEAD(&oh->ot_stop_dcb_list); + INIT_LIST_HEAD(&oh->ot_trunc_locks); + osd_th_alloced(oh); + + memset(oti->oti_declare_ops, 0, + sizeof(oti->oti_declare_ops)); + memset(oti->oti_declare_ops_cred, 0, + sizeof(oti->oti_declare_ops_cred)); + memset(oti->oti_declare_ops_used, 0, + sizeof(oti->oti_declare_ops_used)); + + oti->oti_ins_cache_depth++; + RETURN(th); } void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_thandle *oh; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_thandle *oh; oh = container_of0(th, struct osd_thandle, ot_super); LASSERT(oh != NULL); @@ -1303,81 +1837,96 @@ void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th) static int osd_trans_start(const struct lu_env *env, struct dt_device *d, struct thandle *th) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_device *dev = osd_dt_dev(d); - handle_t *jh; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_device *dev = osd_dt_dev(d); + handle_t *jh; + struct osd_thandle *oh; + int rc; - ENTRY; + ENTRY; - LASSERT(current->journal_info == NULL); + LASSERT(current->journal_info == NULL); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh != NULL); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh != NULL); + LASSERT(oh->ot_handle == NULL); - rc = dt_txn_hook_start(env, d, th); - if (rc != 0) - GOTO(out, rc); + rc = dt_txn_hook_start(env, d, th); + if (rc != 0) + GOTO(out, rc); if (unlikely(osd_param_is_not_sane(dev, th))) { 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; } - /* XXX Limit the credits to 'max_transaction_buffers', and + /* + * XXX Limit the credits to 'max_transaction_buffers', and * let the underlying filesystem to catch the error if * we really need so many credits. * * This should be removed when we can calculate the - * credits precisely. */ + * 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); } - /* - * XXX temporary stuff. Some abstraction layer should - * be used. - */ - jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); - osd_th_started(oh); - if (!IS_ERR(jh)) { - oh->ot_handle = jh; - LASSERT(oti->oti_txns == 0); - lu_context_init(&th->th_ctx, th->th_tags); - lu_context_enter(&th->th_ctx); + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START)) + GOTO(out, rc = -EIO); + + /* + * XXX temporary stuff. Some abstraction layer should + * be used. + */ + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); + osd_th_started(oh); + if (!IS_ERR(jh)) { + oh->ot_handle = jh; + LASSERT(oti->oti_txns == 0); - lu_device_get(&d->dd_lu_dev); + lu_device_get(&d->dd_lu_dev); lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link, "osd-tx", th); - oti->oti_txns++; - rc = 0; - } else { - rc = PTR_ERR(jh); - } + oti->oti_txns++; + rc = 0; + } else { + rc = PTR_ERR(jh); + } out: - RETURN(rc); + RETURN(rc); } static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd, u64 seq) { - struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; - struct seq_server_site *ss = osd_seq_site(osd); - int rc; + struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; LASSERT(ss != NULL); @@ -1386,8 +1935,8 @@ 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); } @@ -1396,8 +1945,8 @@ static int osd_seq_exists(const struct lu_env *env, static void osd_trans_stop_cb(struct osd_thandle *oth, int result) { - struct dt_txn_commit_cb *dcb; - struct dt_txn_commit_cb *tmp; + struct dt_txn_commit_cb *dcb; + struct dt_txn_commit_cb *tmp; /* call per-transaction stop callbacks if any */ list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list, @@ -1416,41 +1965,45 @@ static void osd_trans_stop_cb(struct osd_thandle *oth, int result) static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, struct thandle *th) { - int rc = 0, remove_agents = 0; - struct osd_thandle *oh; struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_device *osd = osd_dt_dev(th->th_dev); - struct qsd_instance *qsd = osd->od_quota_slave; - struct lquota_trans *qtrans; + struct osd_thandle *oh; + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_device *osd = osd_dt_dev(th->th_dev); + struct qsd_instance *qsd = osd_def_qsd(osd); + struct lquota_trans *qtrans; + struct list_head truncates = LIST_HEAD_INIT(truncates); + int rc = 0, remove_agents = 0; + ENTRY; 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; + /* move locks to local list, stop tx, execute truncates */ + list_splice(&oh->ot_trunc_locks, &truncates); + if (oh->ot_handle != NULL) { - handle_t *hdl = oh->ot_handle; + int rc2; + + handle_t *hdl = oh->ot_handle; - /* - * add commit callback - * notice we don't do this in osd_trans_start() - * as underlying transaction can change during truncate - */ - ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, - &oh->ot_jcb); + /* + * add commit callback + * notice we don't do this in osd_trans_start() + * as underlying transaction can change during truncate + */ + ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, + &oh->ot_jcb); - LASSERT(oti->oti_txns == 1); - oti->oti_txns--; + LASSERT(oti->oti_txns == 1); + oti->oti_txns--; - rc = dt_txn_hook_stop(env, th); - if (rc != 0) + rc = dt_txn_hook_stop(env, th); + if (rc != 0) CERROR("%s: failed in transaction hook: rc = %d\n", osd_name(osd), rc); @@ -1459,19 +2012,26 @@ 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; + + osd_process_truncates(&truncates); } else { osd_trans_stop_cb(oh, th->th_result); OBD_FREE_PTR(oh); } + osd_trunc_unlock_all(&truncates); + /* inform the quota slave device that the transaction is stopping */ qsd_op_end(env, qsd, qtrans); - /* as we want IO to journal and data IO be concurrent, we don't block + /* + * as we want IO to journal and data IO be concurrent, we don't block * awaiting data IO completion in osd_do_bio(), instead we wait here * once transaction is submitted to the journal. all reqular requests * don't do direct IO (except read/write), thus this wait_event becomes @@ -1489,6 +2049,13 @@ 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); + oti->oti_ins_cache_depth--; + /* reset OI cache for safety */ + if (oti->oti_ins_cache_depth == 0) + oti->oti_ins_cache_used = 0; + + sb_end_write(osd_sb(osd)); + RETURN(rc); } @@ -1516,27 +2083,26 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) */ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) { - struct osd_object *obj = osd_obj(l); - struct inode *inode = obj->oo_inode; - - LINVRNT(osd_invariant(obj)); + struct osd_object *obj = osd_obj(l); + struct inode *inode = obj->oo_inode; - /* - * If object is unlinked remove fid->ino mapping from object index. - */ + LINVRNT(osd_invariant(obj)); - osd_index_fini(obj); - if (inode != NULL) { - struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; - qid_t uid = i_uid_read(inode); - qid_t gid = i_gid_read(inode); + /* + * If object is unlinked remove fid->ino mapping from object index. + */ - iput(inode); - obj->oo_inode = NULL; + osd_index_fini(obj); + if (inode != NULL) { + struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj)); + qid_t uid = i_uid_read(inode); + qid_t gid = i_gid_read(inode); - if (qsd != NULL) { - struct osd_thread_info *info = osd_oti_get(env); - struct lquota_id_info *qi = &info->oti_qi; + obj->oo_inode = NULL; + iput(inode); + if (!obj->oo_header && qsd) { + struct osd_thread_info *info = osd_oti_get(env); + struct lquota_id_info *qi = &info->oti_qi; /* Release granted quota to master if necessary */ qi->lqi_id.qid_uid = uid; @@ -1544,68 +2110,74 @@ 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); } - } + } } /* * Concurrency: ->loo_object_release() is called under site spin-lock. */ static void osd_object_release(const struct lu_env *env, - struct lu_object *l) + struct lu_object *l) { struct osd_object *o = osd_obj(l); - /* nobody should be releasing a non-destroyed object with nlink=0 + + /* + * 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(); + * this inode as deleted + */ + LASSERT(!(o->oo_destroyed == 0 && o->oo_inode && + o->oo_inode->i_nlink == 0)); } /* * Concurrency: shouldn't matter. */ static int osd_object_print(const struct lu_env *env, void *cookie, - lu_printer_t p, const struct lu_object *l) + lu_printer_t p, const struct lu_object *l) { - struct osd_object *o = osd_obj(l); - struct iam_descr *d; + struct osd_object *o = osd_obj(l); + struct iam_descr *d; - if (o->oo_dir != NULL) - d = o->oo_dir->od_container.ic_descr; - else - d = NULL; + if (o->oo_dir != NULL) + d = o->oo_dir->od_container.ic_descr; + else + d = NULL; return (*p)(env, cookie, LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]", - o, o->oo_inode, - o->oo_inode ? o->oo_inode->i_ino : 0UL, - o->oo_inode ? o->oo_inode->i_generation : 0, - d ? d->id_ops->id_name : "plain"); + o, o->oo_inode, + o->oo_inode ? o->oo_inode->i_ino : 0UL, + o->oo_inode ? o->oo_inode->i_generation : 0, + d ? d->id_ops->id_name : "plain"); } /* * Concurrency: shouldn't matter. */ int osd_statfs(const struct lu_env *env, struct dt_device *d, - struct obd_statfs *sfs) + struct obd_statfs *sfs) { - struct osd_device *osd = osd_dt_dev(d); - struct super_block *sb = osd_sb(osd); - struct kstatfs *ksfs; - __u64 reserved; - 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; - /* osd_lproc.c call this without env, allocate ksfs for that case */ - if (unlikely(env == NULL)) { - OBD_ALLOC_PTR(ksfs); - if (ksfs == NULL) - return -ENOMEM; - } else { - ksfs = &osd_oti_get(env)->oti_ksfs; - } + /* osd_lproc.c call this without env, allocate ksfs for that case */ + if (unlikely(env == NULL)) { + OBD_ALLOC_PTR(ksfs); + if (ksfs == NULL) + return -ENOMEM; + } else { + ksfs = &osd_oti_get(env)->oti_ksfs; + } result = sb->s_op->statfs(sb->s_root, ksfs); if (result) @@ -1613,9 +2185,8 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, 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_state |= OS_STATE_READONLY; + if (ldiskfs_has_feature_extents(sb)) sfs->os_maxbytes = sb->s_maxbytes; else sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; @@ -1658,52 +2229,123 @@ out: * Concurrency: doesn't access mutable data. */ static void osd_conf_get(const struct lu_env *env, - const struct dt_device *dev, - struct dt_device_param *param) + const struct dt_device *dev, + struct dt_device_param *param) { - struct super_block *sb = osd_sb(osd_dt_dev(dev)); - int ea_overhead; + struct osd_device *d = osd_dt_dev(dev); + struct super_block *sb = osd_sb(d); + struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev); + const char *name; + int ea_overhead; - /* - * XXX should be taken from not-yet-existing fs abstraction layer. - */ - param->ddp_max_name_len = LDISKFS_NAME_LEN; - param->ddp_max_nlink = LDISKFS_LINK_MAX; + /* + * XXX should be taken from not-yet-existing fs abstraction layer. + */ + param->ddp_max_name_len = LDISKFS_NAME_LEN; + param->ddp_max_nlink = LDISKFS_LINK_MAX; 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_mount_type = LDD_MT_LDISKFS; + if (ldiskfs_has_feature_extents(sb)) param->ddp_maxbytes = sb->s_maxbytes; else param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; - /* inode are statically allocated, so per-inode space consumption - * is the space consumed by the directory entry */ + /* + * inode are statically allocated, so per-inode space consumption + * is the space consumed by the directory entry + */ param->ddp_inodespace = PER_OBJ_USAGE; - /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks + /* + * 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. */ + * 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)) - param->ddp_mntopts |= MNTOPT_ACL; + param->ddp_mntopts = 0; + if (test_opt(sb, XATTR_USER)) + param->ddp_mntopts |= MNTOPT_USERXATTR; + if (test_opt(sb, POSIX_ACL)) + param->ddp_mntopts |= MNTOPT_ACL; - /* LOD might calculate the max stripe count based on max_ea_size, + /* + * LOD might calculate the max stripe count based on max_ea_size, * so we need take account in the overhead as well, - * xattr_header + magic + xattr_entry_head */ + * xattr_header + magic + xattr_entry_head + */ ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) + LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN); #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE) - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) + if (ldiskfs_has_feature_ea_inode(sb)) param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE - ea_overhead; else #endif param->ddp_max_ea_size = sb->s_blocksize - ea_overhead; + + if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE) + param->ddp_max_ea_size = OSD_MAX_EA_SIZE; + + /* + * Preferred RPC size for efficient disk IO. 4MB shows good + * all-around performance for ldiskfs, but use bigalloc chunk size + * by default if larger. + */ +#if defined(LDISKFS_CLUSTER_SIZE) + if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE) + param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb); + else +#endif + param->ddp_brw_size = DT_DEF_BRW_SIZE; + + param->ddp_t10_cksum_type = 0; + if (bi) { + unsigned short interval = blk_integrity_interval(bi); + name = blk_integrity_name(bi); + /* + * Expected values: + * T10-DIF-TYPE1-CRC + * T10-DIF-TYPE3-CRC + * T10-DIF-TYPE1-IP + * T10-DIF-TYPE3-IP + */ + if (strncmp(name, "T10-DIF-TYPE", + sizeof("T10-DIF-TYPE") - 1) == 0) { + /* also skip "1/3-" at end */ + const int type_off = sizeof("T10-DIF-TYPE."); + char type_number = name[type_off - 2]; + + if (interval != 512 && interval != 4096) { + CERROR("%s: unsupported T10PI sector size %u\n", + d->od_svname, interval); + } else if (type_number != '1' && type_number != '3') { + CERROR("%s: unsupported T10PI type %s\n", + d->od_svname, name); + } else if (strcmp(name + type_off, "CRC") == 0) { + d->od_t10_type = type_number == '1' ? + OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC; + param->ddp_t10_cksum_type = interval == 512 ? + OBD_CKSUM_T10CRC512 : + OBD_CKSUM_T10CRC4K; + } else if (strcmp(name + type_off, "IP") == 0) { + d->od_t10_type = type_number == '1' ? + OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP; + param->ddp_t10_cksum_type = interval == 512 ? + OBD_CKSUM_T10IP512 : + OBD_CKSUM_T10IP4K; + } else { + CERROR("%s: unsupported checksum type of " + "T10PI type '%s'", + d->od_svname, name); + } + + } else { + CERROR("%s: unsupported T10PI type '%s'", + d->od_svname, name); + } + } } /* @@ -1712,13 +2354,15 @@ static void osd_conf_get(const struct lu_env *env, static int osd_sync(const struct lu_env *env, struct dt_device *d) { int rc; + struct super_block *s = osd_sb(osd_dt_dev(d)); + ENTRY; - CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); - - rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + down_read(&s->s_umount); + rc = s->s_op->sync_fs(s, 1); + up_read(&s->s_umount); - CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n", - LUSTRE_OSD_LDISKFS_NAME, rc); + CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname, + rc); return rc; } @@ -1736,45 +2380,67 @@ static int osd_sync(const struct lu_env *env, struct dt_device *d) * \see dt_device_operations */ static int osd_commit_async(const struct lu_env *env, - struct dt_device *d) + struct dt_device *d) { - struct super_block *s = osd_sb(osd_dt_dev(d)); - ENTRY; + struct super_block *s = osd_sb(osd_dt_dev(d)); + int rc; + + ENTRY; + + CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname); + down_read(&s->s_umount); + rc = s->s_op->sync_fs(s, 0); + up_read(&s->s_umount); - CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); - RETURN(s->s_op->sync_fs(s, 0)); + RETURN(rc); } +/* Our own copy of the set readonly functions if present, or NU if not. */ +static int (*priv_dev_set_rdonly)(struct block_device *bdev); +static int (*priv_dev_check_rdonly)(struct block_device *bdev); +/* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */ + /* * Concurrency: shouldn't matter. */ - static int osd_ro(const struct lu_env *env, struct dt_device *d) { struct super_block *sb = osd_sb(osd_dt_dev(d)); struct block_device *dev = sb->s_bdev; -#ifdef HAVE_DEV_SET_RDONLY - struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev; - int rc = 0; -#else int rc = -EOPNOTSUPP; -#endif + ENTRY; -#ifdef HAVE_DEV_SET_RDONLY - CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname); + if (priv_dev_set_rdonly) { + struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev; + + rc = 0; + 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); + priv_dev_set_rdonly(jdev); + } + CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev); + priv_dev_set_rdonly(dev); - if (jdev && (jdev != dev)) { - CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n", - (long)jdev); - dev_set_rdonly(jdev); + if (sb->s_op->unfreeze_fs) + sb->s_op->unfreeze_fs(sb); } - CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev); - dev_set_rdonly(dev); -#else - CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", - osd_dt_dev(d)->od_svname, (long)dev, rc); -#endif + +out: + if (rc) + CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", + osd_dt_dev(d)->od_svname, (long)dev, rc); + RETURN(rc); } @@ -1837,92 +2503,89 @@ const int osd_dto_credits_noquota[DTO_NR] = { }; static const struct dt_device_operations osd_dt_ops = { - .dt_root_get = osd_root_get, - .dt_statfs = osd_statfs, - .dt_trans_create = osd_trans_create, - .dt_trans_start = osd_trans_start, - .dt_trans_stop = osd_trans_stop, - .dt_trans_cb_add = osd_trans_cb_add, - .dt_conf_get = osd_conf_get, - .dt_sync = osd_sync, - .dt_ro = osd_ro, - .dt_commit_async = osd_commit_async, + .dt_root_get = osd_root_get, + .dt_statfs = osd_statfs, + .dt_trans_create = osd_trans_create, + .dt_trans_start = osd_trans_start, + .dt_trans_stop = osd_trans_stop, + .dt_trans_cb_add = osd_trans_cb_add, + .dt_conf_get = osd_conf_get, + .dt_sync = osd_sync, + .dt_ro = osd_ro, + .dt_commit_async = osd_commit_async, }; -static void osd_object_read_lock(const struct lu_env *env, - struct dt_object *dt, unsigned role) +static void osd_read_lock(const struct lu_env *env, struct dt_object *dt, + unsigned int role) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *oti = osd_oti_get(env); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_owner != env); + LASSERT(obj->oo_owner != env); down_read_nested(&obj->oo_sem, role); - LASSERT(obj->oo_owner == NULL); - oti->oti_r_locks++; + LASSERT(obj->oo_owner == NULL); + oti->oti_r_locks++; } -static void osd_object_write_lock(const struct lu_env *env, - struct dt_object *dt, unsigned role) +static void osd_write_lock(const struct lu_env *env, struct dt_object *dt, + unsigned int role) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *oti = osd_oti_get(env); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_owner != env); + LASSERT(obj->oo_owner != env); down_write_nested(&obj->oo_sem, role); - LASSERT(obj->oo_owner == NULL); - obj->oo_owner = env; - oti->oti_w_locks++; + LASSERT(obj->oo_owner == NULL); + obj->oo_owner = env; + oti->oti_w_locks++; } -static void osd_object_read_unlock(const struct lu_env *env, - struct dt_object *dt) +static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *oti = osd_oti_get(env); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - LASSERT(oti->oti_r_locks > 0); - oti->oti_r_locks--; + LASSERT(oti->oti_r_locks > 0); + oti->oti_r_locks--; up_read(&obj->oo_sem); } -static void osd_object_write_unlock(const struct lu_env *env, - struct dt_object *dt) +static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *oti = osd_oti_get(env); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_owner == env); - LASSERT(oti->oti_w_locks > 0); - oti->oti_w_locks--; - obj->oo_owner = NULL; + LASSERT(obj->oo_owner == env); + LASSERT(oti->oti_w_locks > 0); + oti->oti_w_locks--; + obj->oo_owner = NULL; up_write(&obj->oo_sem); } -static int osd_object_write_locked(const struct lu_env *env, - struct dt_object *dt) +static int osd_write_locked(const struct lu_env *env, struct dt_object *dt) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - return obj->oo_owner == env; + return obj->oo_owner == env; } static struct timespec *osd_inode_time(const struct lu_env *env, struct inode *inode, __u64 seconds) { - struct osd_thread_info *oti = osd_oti_get(env); - struct timespec *t = &oti->oti_time; + struct osd_thread_info *oti = osd_oti_get(env); + struct timespec *t = &oti->oti_time; t->tv_sec = seconds; t->tv_nsec = 0; @@ -1935,22 +2598,30 @@ static void osd_inode_getattr(const struct lu_env *env, { attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | - LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | - LA_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); - attr->la_ctime = LTIME_S(inode->i_ctime); + attr->la_atime = inode->i_atime.tv_sec; + attr->la_mtime = inode->i_mtime.tv_sec; + attr->la_ctime = inode->i_ctime.tv_sec; attr->la_mode = inode->i_mode; attr->la_size = i_size_read(inode); attr->la_blocks = inode->i_blocks; attr->la_uid = i_uid_read(inode); attr->la_gid = i_gid_read(inode); + attr->la_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, @@ -1976,20 +2647,78 @@ static int osd_attr_get(const struct lu_env *env, return 0; } -static int osd_declare_attr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_attr *attr, - struct thandle *handle) +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 int type, bool ignore_edquot) { - struct osd_thandle *oh; - struct osd_object *obj; + int rc; struct osd_thread_info *info = osd_oti_get(env); struct lquota_id_info *qi = &info->oti_qi; - qid_t uid; - qid_t gid; - long long bspace; - int rc = 0; - bool enforce; + + 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 (ignore_edquot && (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 (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS)) + rc = 0; + if (rc) + RETURN(rc); + + /* block accounting */ + qi->lqi_is_blk = true; + + /* more blocks for the new id ... */ + 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 (ignore_edquot && (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 (ignore_edquot && (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, + struct thandle *handle) +{ + struct osd_thandle *oh; + struct osd_object *obj; + qid_t uid; + qid_t gid; + long long bspace; + int rc = 0; + bool enforce; + ENTRY; LASSERT(dt != NULL); @@ -2010,116 +2739,58 @@ static int osd_declare_attr_set(const struct lu_env *env, if (attr == NULL || obj->oo_inode == NULL) RETURN(rc); - bspace = obj->oo_inode->i_blocks; - bspace <<= obj->oo_inode->i_sb->s_blocksize_bits; + bspace = obj->oo_inode->i_blocks << 9; bspace = toqb(bspace); - /* Changing ownership is always preformed by super user, it should not - * fail with EDQUOT. + /* + * Changing ownership is always preformed by super user, it should not + * fail with EDQUOT unless required explicitly. * * We still need to call the osd_declare_qid() to calculate the journal * credits for updating quota accounting files and to trigger quota - * space adjustment once the operation is completed.*/ + * space adjustment once the operation is completed. + */ if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { + bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL); + + if (!ignore_edquot) + CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u" + "(the quota space is %lld)\n", + obj->oo_inode->i_sb->s_id, attr->la_uid, + attr->la_gid, bspace); + /* 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, + true); 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; - 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; + rc = osd_declare_attr_qid(env, obj, oh, bspace, + i_gid_read(obj->oo_inode), + attr->la_gid, enforce, GRPQUOTA, + ignore_edquot); 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, true); if (rc) RETURN(rc); } - +#endif RETURN(rc); } @@ -2135,23 +2806,23 @@ 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 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 + 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); + } + + /* + * 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); @@ -2159,25 +2830,89 @@ 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; + else + LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL; + } + return 0; +} + +#ifdef HAVE_PROJECT_QUOTA +static int osd_transfer_project(struct inode *inode, __u32 projid) +{ + struct super_block *sb = inode->i_sb; + struct ldiskfs_inode_info *ei = LDISKFS_I(inode); + int err; + kprojid_t kprojid; + struct ldiskfs_iloc iloc; + struct ldiskfs_inode *raw_inode; + struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { }; + + if (!ldiskfs_has_feature_project(sb)) { + LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid) + == LDISKFS_DEF_PROJID); + if (projid != LDISKFS_DEF_PROJID) + return -EOPNOTSUPP; + else + return 0; + } + + if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE) + return -EOPNOTSUPP; + + kprojid = make_kprojid(&init_user_ns, (projid_t)projid); + if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid)) + return 0; + + err = ldiskfs_get_inode_loc(inode, &iloc); + if (err) + return err; + + raw_inode = ldiskfs_raw_inode(&iloc); + if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) { + err = -EOVERFLOW; + brelse(iloc.bh); + return err; + } + brelse(iloc.bh); + + dquot_initialize(inode); + transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid)); + if (transfer_to[PRJQUOTA]) { + err = __dquot_transfer(inode, transfer_to); + dqput(transfer_to[PRJQUOTA]); + if (err) + return err; + } + + return err; } +#endif 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; + struct iattr iattr; ll_vfs_dq_init(inode); iattr.ia_valid = 0; @@ -2196,6 +2931,22 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) return rc; } } + + /* Handle project id transfer here properly */ + if (attr->la_valid & LA_PROJID && + attr->la_projid != i_projid_read(inode)) { +#ifdef HAVE_PROJECT_QUOTA + rc = osd_transfer_project(inode, attr->la_projid); +#else + rc = -ENOTSUPP; +#endif + if (rc) { + CERROR("%s: quota transfer failed: rc = %d. Is project " + "enforcement enabled on the ldiskfs " + "filesystem?\n", inode->i_sb->s_id, rc); + return rc; + } + } return 0; } @@ -2205,7 +2956,7 @@ static int osd_attr_set(const struct lu_env *env, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode; + struct inode *inode; int rc; if (!dt_object_exists(dt)) @@ -2217,15 +2968,16 @@ static int osd_attr_set(const struct lu_env *env, osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); - if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) { - struct osd_thread_info *oti = osd_oti_get(env); - const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); - struct lu_fid *fid1 = &oti->oti_fid; - struct osd_inode_id *id = &oti->oti_id; - struct iam_path_descr *ipd; - struct iam_container *bag; - struct osd_thandle *oh; - int rc; + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && + !osd_obj2dev(obj)->od_is_ost) { + struct osd_thread_info *oti = osd_oti_get(env); + const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); + struct lu_fid *fid1 = &oti->oti_fid; + struct osd_inode_id *id = &oti->oti_id; + struct iam_path_descr *ipd; + struct iam_container *bag; + struct osd_thandle *oh; + int rc; fid_cpu_to_be(fid1, fid0); memset(id, 1, sizeof(*id)); @@ -2236,13 +2988,14 @@ static int osd_attr_set(const struct lu_env *env, RETURN(-ENOMEM); oh = container_of0(handle, struct osd_thandle, ot_super); - rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1, + rc = iam_update(oh->ot_handle, bag, + (const struct iam_key *)fid1, (const struct iam_rec *)id, ipd); osd_ipd_put(env, bag, ipd); return(rc > 0 ? 0 : rc); } - inode = obj->oo_inode; + inode = obj->oo_inode; rc = osd_quota_transfer(inode, attr); if (rc) @@ -2262,10 +3015,13 @@ static int osd_attr_set(const struct lu_env *env, /* Let's check if there are extra flags need to be set into LMA */ if (attr->la_flags & LUSTRE_LMA_FL_MASKS) { struct osd_thread_info *info = osd_oti_get(env); - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc != 0) + LASSERT(!obj->oo_pfid_in_lma); + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, + &info->oti_ost_attrs); + if (rc) GOTO(out, rc); lma->lma_incompat |= @@ -2288,148 +3044,150 @@ static int osd_attr_set(const struct lu_env *env, out: osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); - return rc; + return rc; } static struct dentry *osd_child_dentry_get(const struct lu_env *env, struct osd_object *obj, const char *name, const int namelen) { - return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); + return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); } static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, umode_t mode, struct dt_allocation_hint *hint, struct thandle *th) { - int result; - struct osd_device *osd = osd_obj2dev(obj); - struct osd_thandle *oth; - struct dt_object *parent = NULL; - struct inode *inode; + int result; + struct osd_device *osd = osd_obj2dev(obj); + struct osd_thandle *oth; + struct dt_object *parent = NULL; + struct inode *inode; - LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_inode == NULL); - LASSERT(obj->oo_hl_head == NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(obj->oo_inode == NULL); + LASSERT(obj->oo_hl_head == NULL); - if (S_ISDIR(mode) && ldiskfs_pdo) { - obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); - if (obj->oo_hl_head == NULL) - return -ENOMEM; - } + if (S_ISDIR(mode) && ldiskfs_pdo) { + obj->oo_hl_head = + ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); + if (obj->oo_hl_head == NULL) + return -ENOMEM; + } - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); if (hint != NULL && hint->dah_parent != NULL && !dt_object_remote(hint->dah_parent)) parent = hint->dah_parent; - inode = ldiskfs_create_inode(oth->ot_handle, - parent ? osd_dt_obj(parent)->oo_inode : - osd_sb(osd)->s_root->d_inode, - mode); - if (!IS_ERR(inode)) { + inode = ldiskfs_create_inode(oth->ot_handle, + parent ? osd_dt_obj(parent)->oo_inode : + osd_sb(osd)->s_root->d_inode, + mode); + if (!IS_ERR(inode)) { /* Do not update file c/mtime in ldiskfs. */ inode->i_flags |= S_NOCMTIME; - /* For new created object, it must be consistent, - * and it is unnecessary to scrub against it. */ + /* + * For new created object, it must be consistent, + * and it is unnecessary to scrub against it. + */ ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); - obj->oo_inode = inode; - result = 0; - } else { - if (obj->oo_hl_head != NULL) { - ldiskfs_htree_lock_head_free(obj->oo_hl_head); - obj->oo_hl_head = NULL; - } - result = PTR_ERR(inode); - } - LINVRNT(osd_invariant(obj)); - return result; + obj->oo_inode = inode; + result = 0; + } else { + if (obj->oo_hl_head != NULL) { + ldiskfs_htree_lock_head_free(obj->oo_hl_head); + obj->oo_hl_head = NULL; + } + result = PTR_ERR(inode); + } + LINVRNT(osd_invariant(obj)); + return result; } enum { - OSD_NAME_LEN = 255 + OSD_NAME_LEN = 255 }; 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, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) -{ - int result; - struct osd_thandle *oth; - const struct dt_index_features *feat = dof->u.dof_idx.di_feat; - - __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX)); - - LASSERT(S_ISREG(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); - if (result == 0) { - LASSERT(obj->oo_inode != NULL); - if (feat->dif_flags & DT_IND_VARKEY) - result = iam_lvar_create(obj->oo_inode, - feat->dif_keysize_max, - feat->dif_ptrsize, - feat->dif_recsize_max, - oth->ot_handle); - else - result = iam_lfix_create(obj->oo_inode, - feat->dif_keysize_max, - feat->dif_ptrsize, - feat->dif_recsize_max, - oth->ot_handle); - - } - return result; + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) +{ + int result; + struct osd_thandle *oth; + const struct dt_index_features *feat = dof->u.dof_idx.di_feat; + + __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX)); + + LASSERT(S_ISREG(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); + if (result == 0) { + LASSERT(obj->oo_inode != NULL); + if (feat->dif_flags & DT_IND_VARKEY) + result = iam_lvar_create(obj->oo_inode, + feat->dif_keysize_max, + feat->dif_ptrsize, + feat->dif_recsize_max, + oth->ot_handle); + else + result = iam_lfix_create(obj->oo_inode, + feat->dif_keysize_max, + feat->dif_ptrsize, + feat->dif_recsize_max, + oth->ot_handle); + } + return result; } static int osd_mkreg(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) { - LASSERT(S_ISREG(attr->la_mode)); - return osd_mkfile(info, obj, (attr->la_mode & - (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); + LASSERT(S_ISREG(attr->la_mode)); + return osd_mkfile(info, obj, (attr->la_mode & + (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); } static int osd_mksym(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) { - LASSERT(S_ISLNK(attr->la_mode)); - return osd_mkfile(info, obj, (attr->la_mode & - (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); + LASSERT(S_ISLNK(attr->la_mode)); + return osd_mkfile(info, obj, (attr->la_mode & + (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); } static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, @@ -2443,55 +3201,55 @@ static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || - S_ISFIFO(mode) || S_ISSOCK(mode)); + LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || + S_ISFIFO(mode) || S_ISSOCK(mode)); - result = osd_mkfile(info, obj, mode, hint, th); - if (result == 0) { - LASSERT(obj->oo_inode != NULL); + result = osd_mkfile(info, obj, mode, hint, th); + if (result == 0) { + LASSERT(obj->oo_inode != NULL); /* * This inode should be marked dirty for i_rdev. Currently * that is done in the osd_attr_init(). */ init_special_inode(obj->oo_inode, obj->oo_inode->i_mode, attr->la_rdev); - } - LINVRNT(osd_invariant(obj)); - return result; + } + LINVRNT(osd_invariant(obj)); + return result; } typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *, - struct lu_attr *, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *); + struct lu_attr *, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *); static osd_obj_type_f osd_create_type_f(enum dt_format_type type) { - osd_obj_type_f result; - - switch (type) { - case DFT_DIR: - result = osd_mkdir; - break; - case DFT_REGULAR: - result = osd_mkreg; - break; - case DFT_SYM: - result = osd_mksym; - break; - case DFT_NODE: - result = osd_mknod; - break; - case DFT_INDEX: - result = osd_mk_index; - break; - - default: - LBUG(); - break; - } - return result; + osd_obj_type_f result; + + switch (type) { + case DFT_DIR: + result = osd_mkdir; + break; + case DFT_REGULAR: + result = osd_mkreg; + break; + case DFT_SYM: + result = osd_mksym; + break; + case DFT_NODE: + result = osd_mknod; + break; + case DFT_INDEX: + result = osd_mk_index; + break; + + default: + LBUG(); + break; + } + return result; } @@ -2499,14 +3257,15 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, struct dt_object *parent, struct dt_object *child, umode_t child_mode) { - LASSERT(ah); + LASSERT(ah); - ah->dah_parent = parent; - ah->dah_mode = child_mode; + 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); } } @@ -2514,20 +3273,20 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, struct lu_attr *attr, struct dt_object_format *dof) { - struct inode *inode = obj->oo_inode; - __u64 valid = attr->la_valid; - int result; + struct inode *inode = obj->oo_inode; + __u64 valid = attr->la_valid; + int result; attr->la_valid &= ~(LA_TYPE | LA_MODE); - if (dof->dof_type != DFT_NODE) - attr->la_valid &= ~LA_RDEV; - if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime))) - attr->la_valid &= ~LA_ATIME; - if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime))) - attr->la_valid &= ~LA_CTIME; - if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime))) - attr->la_valid &= ~LA_MTIME; + if (dof->dof_type != DFT_NODE) + attr->la_valid &= ~LA_RDEV; + if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec)) + attr->la_valid &= ~LA_ATIME; + if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec)) + attr->la_valid &= ~LA_CTIME; + if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec)) + attr->la_valid &= ~LA_MTIME; result = osd_quota_transfer(inode, attr); if (result) @@ -2550,18 +3309,16 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, } /** - * Helper function for osd_object_create() + * Helper function for osd_create() * * \retval 0, on success */ -static int __osd_object_create(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) +static int __osd_create(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) { - int result; - __u32 umask; + int result; + __u32 umask; osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE); @@ -2574,8 +3331,10 @@ static int __osd_object_create(struct osd_thread_info *info, if (likely(obj->oo_inode != NULL)) { LASSERT(obj->oo_inode->i_state & I_NEW); - /* Unlock the inode before attr initialization to avoid - * unnecessary dqget operations. LU-6378 */ + /* + * Unlock the inode before attr initialization to avoid + * unnecessary dqget operations. LU-6378 + */ unlock_new_inode(obj->oo_inode); } @@ -2593,7 +3352,7 @@ static int __osd_object_create(struct osd_thread_info *info, } /** - * Helper function for osd_object_create() + * Helper function for osd_create() * * \retval 0, on success */ @@ -2604,7 +3363,7 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id = &info->oti_id; struct osd_device *osd = osd_obj2dev(obj); struct osd_thandle *oh; - int rc; + int rc; LASSERT(obj->oo_inode != NULL); @@ -2613,7 +3372,8 @@ 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; @@ -2622,7 +3382,7 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, u64 seq, struct lu_seq_range *range) { - struct seq_server_site *ss = osd_seq_site(osd); + struct seq_server_site *ss = osd_seq_site(osd); if (fid_seq_is_idif(seq)) { fld_range_set_ost(range); @@ -2633,8 +3393,10 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, if (!fid_seq_in_fldb(seq)) { fld_range_set_mdt(range); if (ss != NULL) - /* FIXME: If ss is NULL, it suppose not get lsr_index - * at all */ + /* + * FIXME: If ss is NULL, it suppose not get lsr_index + * at all + */ range->lsr_index = ss->ss_node_id; return 0; } @@ -2645,18 +3407,15 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, return fld_local_lookup(env, ss->ss_server_fld, seq, range); } -/* - * Concurrency: no external locking is necessary. - */ -static int osd_declare_object_create(const struct lu_env *env, - struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *handle) +static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *handle) { - struct osd_thandle *oh; - int rc; + struct osd_thandle *oh; + int rc; + ENTRY; LASSERT(handle != NULL); @@ -2664,83 +3423,53 @@ static int osd_declare_object_create(const struct lu_env *env, oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - /* EA object consumes more credits than regular object: osd_mk_index + /* + * EA object consumes more credits than regular object: osd_mk_index * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and * leaf_node, could involves the block, block bitmap, groups, GDT - * change for each block, so add 4 * 2 credits in that case. */ + * change for each block, so add 4 * 2 credits in that case. + */ osd_trans_declare_op(env, oh, OSD_OT_CREATE, osd_dto_credits_noquota[DTO_OBJECT_CREATE] + (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0); - /* Reuse idle OI block may cause additional one OI block - * to be changed. */ + /* + * Reuse idle OI block may cause additional one OI block + * to be changed. + */ osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + /* 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)); + if (rc != 0) + RETURN(rc); + 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), + NULL, OSD_QID_INODE); 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); } -static int osd_object_create(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, struct thandle *th) -{ - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - int result; - ENTRY; - - if (dt_object_exists(dt)) - return -EEXIST; - - LINVRNT(osd_invariant(obj)); - LASSERT(!dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); - - if (unlikely(fid_is_acct(fid))) - /* Quota files can't be created from the kernel any more, - * 'tune2fs -O quota' will take care of creating them */ - RETURN(-EPERM); - - result = __osd_object_create(info, obj, attr, hint, dof, th); - if (result == 0) { - result = __osd_oi_insert(env, obj, fid, th); - if (obj->oo_dt.do_body_ops == &osd_body_ops_new) - obj->oo_dt.do_body_ops = &osd_body_ops; - } - LASSERT(ergo(result == 0, - dt_object_exists(dt) && !dt_object_remote(dt))); - - LASSERT(osd_invariant(obj)); - RETURN(result); -} - /** * Called to destroy on-disk representation of the object * * Concurrency: must be locked */ -static int osd_declare_object_destroy(const struct lu_env *env, - struct dt_object *dt, - struct thandle *th) +static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, + struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - int rc; + int rc; + ENTRY; if (inode == NULL) @@ -2751,58 +3480,71 @@ static int osd_declare_object_destroy(const struct lu_env *env, 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, + + /* For removing agent entry */ + if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) + oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE]; + + /* + * Recycle idle OI leaf may cause additional three OI blocks + * to be changed. + */ + 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, NULL, + OSD_QID_INODE); 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, NULL, + OSD_QID_BLK); if (rc) RETURN(rc); - /* will help to find FID->ino when this object is being - * added to PENDING/ */ + /* + * 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); } -static int osd_object_destroy(const struct lu_env *env, - struct dt_object *dt, - struct thandle *th) +static int osd_destroy(const struct lu_env *env, struct dt_object *dt, + struct thandle *th) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_device *osd = osd_obj2dev(obj); - struct osd_thandle *oh; - int result; - ENTRY; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_device *osd = osd_obj2dev(obj); + struct osd_thandle *oh; + int result; - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle); - LASSERT(inode); - LASSERT(!lu_object_is_dying(dt->do_lu.lo_header)); + ENTRY; + + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle); + LASSERT(inode); + LASSERT(!lu_object_is_dying(dt->do_lu.lo_header)); if (unlikely(fid_is_acct(fid))) RETURN(-EPERM); + if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) { + result = osd_delete_from_remote_parent(env, osd, obj, oh, true); + if (result != 0) + CERROR("%s: remove agent entry "DFID": rc = %d\n", + osd_name(osd), PFID(fid), result); + } + if (S_ISDIR(inode->i_mode)) { LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 || inode->i_nlink == 2); - /* it will check/delete the inode from remote parent, - * how to optimize it? unlink performance impaction XXX */ - result = osd_delete_from_remote_parent(env, osd, obj, oh); - if (result != 0 && result != -ENOENT) { - CERROR("%s: delete inode "DFID": rc = %d\n", - osd_name(osd), PFID(fid), result); - } + spin_lock(&obj->oo_guard); clear_nlink(inode); spin_unlock(&obj->oo_guard); @@ -2812,8 +3554,10 @@ 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 */ @@ -2839,8 +3583,10 @@ 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)) @@ -2849,33 +3595,54 @@ 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); + lustre_loa_init(loa, fid, compat, incompat); + lustre_loa_swab(loa, false); - 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. */ + /* + * 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); @@ -2915,7 +3682,8 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id) { struct osd_thread_info *info = osd_oti_get(env); - struct inode *inode; + struct inode *inode; + ENTRY; osd_id_gen(id, ino, OSD_OII_NOGEN); @@ -2947,7 +3715,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, 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 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. * @@ -2959,7 +3728,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, * 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 */ + * attr to be set as 2. LU-7447 + */ set_nlink(dir, saved_nlink); return rc; } @@ -2974,10 +3744,11 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, __u32 type, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct inode *local; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct inode *local; + struct osd_thandle *oh; + int rc; + ENTRY; LASSERT(th); @@ -2991,9 +3762,40 @@ 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); + + /* e2fsck doesn't like empty symlinks. Store remote FID as symlink. + * That gives e2fsck something to look at and be happy, and allows + * debugging if we need to determine where this symlink came from. + */ + if (S_ISLNK(type)) { + CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1); + rc = snprintf((char *)LDISKFS_I(local)->i_data, + LDISKFS_N_BLOCKS * 4, DFID, PFID(fid)); + + i_size_write(local, rc); + LDISKFS_I(local)->i_disksize = rc; + } 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 && + i_projid_read(pobj->oo_inode) != 0) { + rc = osd_transfer_project(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) { @@ -3031,7 +3833,7 @@ static int osd_schedule_agent_inode_removal(const struct lu_env *env, struct osd_thandle *oh, __u32 ino) { - struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); + struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); struct osd_obj_orphan *oor; OBD_ALLOC_PTR(oor); @@ -3101,40 +3903,40 @@ static int osd_process_scheduled_agent_removals(const struct lu_env *env, } /** - * OSD layer object create function for interoperability mode (b11826). - * This is mostly similar to osd_object_create(). Only difference being, fid is - * inserted into inode ea here. + * OSD layer object create function for OST objects (b=11826). + * + * The FID is inserted into inode xattr here. * * \retval 0, on success * \retval -ve, on error */ -static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) +static int osd_create(const struct lu_env *env, struct dt_object *dt, + struct lu_attr *attr, struct dt_allocation_hint *hint, + struct dt_object_format *dof, struct thandle *th) { - 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, on_ost = 0; + 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, on_ost = 0; ENTRY; if (dt_object_exists(dt)) RETURN(-EEXIST); - LASSERT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); LASSERT(!dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); + LASSERT(osd_is_write_locked(env, obj)); LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) - /* Quota files can't be created from the kernel any more, - * 'tune2fs -O quota' will take care of creating them */ + /* + * Quota files can't be created from the kernel any more, + * 'tune2fs -O quota' will take care of creating them + */ RETURN(-EPERM); - result = __osd_object_create(info, obj, attr, hint, dof, th); + result = __osd_create(info, obj, attr, hint, dof, th); if (result == 0) { if (fid_is_idif(fid) && !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) { @@ -3157,59 +3959,63 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, obj->oo_dt.do_body_ops = &osd_body_ops; } - if (result == 0) + if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) result = __osd_oi_insert(env, obj, fid, th); - /* a small optimization - dt_insert() isn't usually applied + /* + * a small optimization - dt_insert() isn't usually applied * to OST objects, so we don't need to cache OI mapping for - * OST objects */ + * 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)); - RETURN(result); + LINVRNT(osd_invariant(obj)); + RETURN(result); } -static int osd_declare_object_ref_add(const struct lu_env *env, - struct dt_object *dt, - struct thandle *handle) +static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - /* it's possible that object doesn't exist yet */ - LASSERT(handle != NULL); + /* it's possible that object doesn't exist yet */ + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_REF_ADD, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); + osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt)); + return 0; } /* * Concurrency: @dt is write locked. */ -static int osd_object_ref_add(const struct lu_env *env, - struct dt_object *dt, struct thandle *th) +static int osd_ref_add(const struct lu_env *env, struct dt_object *dt, + struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - int rc = 0; + int rc = 0; if (!dt_object_exists(dt) || obj->oo_destroyed) return -ENOENT; LINVRNT(osd_invariant(obj)); LASSERT(!dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); + LASSERT(osd_is_write_locked(env, obj)); LASSERT(th != NULL); oh = container_of0(th, struct osd_thandle, ot_super); @@ -3249,9 +4055,8 @@ static int osd_object_ref_add(const struct lu_env *env, return rc; } -static int osd_declare_object_ref_del(const struct lu_env *env, - struct dt_object *dt, - struct thandle *handle) +static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt, + struct thandle *handle) { struct osd_thandle *oh; @@ -3273,20 +4078,20 @@ static int osd_declare_object_ref_del(const struct lu_env *env, /* * Concurrency: @dt is write locked. */ -static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, - struct thandle *th) +static int osd_ref_del(const struct lu_env *env, struct dt_object *dt, + struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct osd_thandle *oh; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct osd_thandle *oh; if (!dt_object_exists(dt)) return -ENOENT; LINVRNT(osd_invariant(obj)); LASSERT(!dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); + LASSERT(osd_is_write_locked(env, obj)); LASSERT(th != NULL); oh = container_of0(th, struct osd_thandle, ot_super); @@ -3295,9 +4100,11 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_REF_DEL); spin_lock(&obj->oo_guard); - /* That can be result of upgrade from old Lustre version and + /* + * That can be result of upgrade from old Lustre version and * applied only to local files. Just skip this ref_del call. - * ext4_unlink() only treats this as a warning, don't LASSERT here.*/ + * ext4_unlink() only treats this as a warning, don't LASSERT here. + */ if (inode->i_nlink == 0) { CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ? D_ERROR : D_INODE, "%s: nlink == 0 on "DFID @@ -3322,63 +4129,118 @@ 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; - /* 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 */ + LASSERT(buf); + + /* 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; LASSERT(!dt_object_remote(dt)); LASSERT(inode->i_op != NULL); +#ifdef HAVE_IOP_XATTR LASSERT(inode->i_op->getxattr != NULL); +#endif - 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; + } + + if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) { + 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; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (rc) + return rc; + + LASSERT(lma->lma_compat & LMAC_STRIPE_INFO); + + 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; + } + } else { + rc = __osd_xattr_get(inode, dentry, name, + buf->lb_buf, buf->lb_len); + } + + 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, - const struct lu_buf *buf, const char *name, - int fl, struct thandle *handle) + struct dt_object *dt, + const struct lu_buf *buf, const char *name, + 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); @@ -3387,27 +4249,52 @@ static int osd_declare_xattr_set(const struct lu_env *env, LASSERT(oh->ot_handle == NULL); if (strcmp(name, XATTR_NAME_LMA) == 0) { - /* For non-upgrading case, the LMA is set first and + /* + * 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; + * may be in another separated EA block. + */ + 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 { + /* + * If some name entry resides on remote MDT, then will create + * agent entry under remote parent. On the other hand, if the + * remote entry will be removed, then related agent entry may + * need to be removed from the remote parent. So there may be + * kinds of cases, let's declare enough credits. The credits + * for create agent entry is enough for remove case. + */ + if (strcmp(name, XATTR_NAME_LINK) == 0) { + credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; + if (dt_object_exists(dt)) + credits += 1; /* For updating LMA */ + } + upgrade: - credits = osd_dto_credits_noquota[DTO_XATTR_SET]; + credits += osd_dto_credits_noquota[DTO_XATTR_SET]; if (buf != NULL) { ssize_t buflen; if (buf->lb_buf == NULL && dt_object_exists(dt)) { - /* learn xattr size from osd_xattr_get if - attribute has not been read yet */ + /* + * learn xattr size from osd_xattr_get if + * attribute has not been read yet + */ buflen = __osd_xattr_get( osd_dt_obj(dt)->oo_inode, &osd_oti_get(env)->oti_obj_dentry, @@ -3437,22 +4324,179 @@ upgrade: return 0; } +static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj, + const struct lu_buf *buf, int fl, + struct thandle *handle) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + struct inode *inode = obj->oo_inode; + struct filter_fid *ff = buf->lb_buf; + struct ost_layout *ol = &ff->ff_layout; + int flags = XATTR_REPLACE; + int rc; + + ENTRY; + + if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid)) + RETURN(-EINVAL); + + rc = osd_get_lma(info, inode, dentry, loa); + if (rc == -ENODATA) { + /* Usually for upgarding from old device */ + lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu), + LMAC_FID_ON_OST, 0); + flags = XATTR_CREATE; + } else if (rc) { + RETURN(rc); + } + + if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) { + if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE)) + RETURN(-EEXIST); + + if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) { + /* Separate PFID EA from LMA */ + 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); + if (!rc) { + obj->oo_pfid_in_lma = 0; + rc = LU_XATTR_CREATE; + } + + RETURN(rc); + } + } else { + if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) + RETURN(fl); + + /* + * Old client does not send stripe information, + * then store the PFID EA on disk separatedly. + */ + if (unlikely(buf->lb_len == sizeof(struct lu_fid) || + ol->ol_stripe_size == 0)) + RETURN(fl); + + /* Remove old PFID EA entry firstly. */ + ll_vfs_dq_init(inode); + rc = osd_removexattr(dentry, inode, XATTR_NAME_FID); + if (rc == -ENODATA) { + if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE)) + RETURN(rc); + } else if (rc) { + RETURN(rc); + } + } + + fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent); + if (likely(ol->ol_stripe_size != 0)) { + 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); + + /* Store the PFID EA inside LMA. */ + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa), + flags); + if (!rc) + obj->oo_pfid_in_lma = 1; + + RETURN(rc); +} + /* - * Set the 64-bit version for object + * In DNE environment, the object (in spite of regular file or directory) + * and its name entry may reside on different MDTs. Under such case, we will + * create an agent entry on the MDT where the object resides. The agent entry + * references the object locally, that makes the object to be visible to the + * userspace when mounted as 'ldiskfs' directly. Then the userspace tools, + * such as 'tar' can handle the object properly. + * + * We handle the agent entry during set linkEA that is the common interface + * for both regular file and directroy, can handle kinds of cases, such as + * create/link/unlink/rename, and so on. + * + * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory. + * + * XXX: There are two known issues: + * 1. For one object, we will create at most one agent entry even if there + * may be more than one cross-MDTs hard links on the object. So the local + * e2fsck may claim that the object's nlink is larger than the name entries + * that reference such inode. And in further, the e2fsck will fix the nlink + * attribute to match the local references. Then it will cause the object's + * nlink attribute to be inconsistent with the global references. it is bad + * but not fatal. The ref_del() can handle the zero-referenced case. On the + * other hand, the global namespace LFSCK can repair the object's attribute + * according to the linkEA. + * 2. There may be too many hard links on the object as to its linkEA overflow, + * then the linkEA entry for cross-MDTs reference may be discarded. If such + * case happened, then at this point, we do not know whether there are some + * cross-MDTs reference. But there are local references, it guarantees that + * object is visible to userspace when mounted as 'ldiskfs'. That is enough. */ -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; +static int osd_xattr_handle_linkea(const struct lu_env *env, + struct osd_device *osd, + struct osd_object *obj, + const struct lu_buf *buf, + struct thandle *handle) +{ + const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); + struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3; + struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf }; + struct lu_name tmpname; + struct osd_thandle *oh; + int rc; + bool remote = false; + + ENTRY; - CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n", - *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); - 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); + rc = linkea_init_with_rec(&ldata); + if (!rc) { + linkea_first_entry(&ldata); + while (ldata.ld_lee != NULL && !remote) { + linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, + &tmpname, tfid); + if (osd_remote_fid(env, osd, tfid) > 0) + remote = true; + else + linkea_next_entry(&ldata); + } + } else if (rc == -ENODATA) { + rc = 0; + } else { + RETURN(rc); + } + + if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) { + rc = osd_delete_from_remote_parent(env, osd, obj, oh, false); + if (rc) + CERROR("%s: failed to remove agent entry for "DFID + ": rc = %d\n", osd_name(osd), PFID(fid), rc); + } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) { + rc = osd_add_to_remote_parent(env, osd, obj, oh); + if (rc) + CERROR("%s: failed to create agent entry for "DFID + ": rc = %d\n", osd_name(osd), PFID(fid), rc); + } + + RETURN(rc); } /* @@ -3462,40 +4506,67 @@ 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 osd_device *osd = osd_obj2dev(obj); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + 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) { + 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); - /* 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); - } + 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; - - if (fl & LU_XATTR_CREATE) - fs_flags |= XATTR_CREATE; - if (strcmp(name, XATTR_NAME_LMV) == 0) { - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - int rc; - - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); - if (rc != 0) + /* + * 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 && osd->od_is_ost && + (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) { + LASSERT(buf->lb_buf); + + fl = osd_xattr_set_pfid(env, obj, buf, fl, handle); + if (fl <= 0) + RETURN(fl); + } else if (strcmp(name, XATTR_NAME_LMV) == 0) { + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (rc) RETURN(rc); lma->lma_incompat |= LMAI_STRIPED; @@ -3504,16 +4575,28 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, sizeof(*lma), XATTR_REPLACE); if (rc != 0) RETURN(rc); + } else if (strcmp(name, XATTR_NAME_LINK) == 0) { + LASSERT(!osd->od_is_ost); + + rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle); + if (rc) + RETURN(rc); } - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) && - strcmp(name, XATTR_NAME_LINK) == 0) - return -ENOSPC; + if (fl & LU_XATTR_REPLACE) + fs_flags |= XATTR_REPLACE; + + if (fl & LU_XATTR_CREATE) + fs_flags |= XATTR_CREATE; - rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, - fs_flags); + 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; } @@ -3523,10 +4606,10 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; if (!dt_object_exists(dt)) return -ENOENT; @@ -3570,39 +4653,65 @@ static int osd_declare_xattr_del(const struct lu_env *env, static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, const char *name, struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + int rc; if (!dt_object_exists(dt)) return -ENOENT; LASSERT(!dt_object_remote(dt)); LASSERT(inode->i_op != NULL); - LASSERT(inode->i_op->removexattr != NULL); LASSERT(handle != NULL); +#ifdef HAVE_IOP_XATTR + LASSERT(inode->i_op->removexattr != NULL); +#endif osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); - ll_vfs_dq_init(inode); - dentry->d_inode = inode; - dentry->d_sb = inode->i_sb; - rc = inode->i_op->removexattr(dentry, name); + if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) { + 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) { + LASSERT(lma->lma_compat & LMAC_STRIPE_INFO); + + 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); + if (!rc) + obj->oo_pfid_in_lma = 0; + } + } else { + ll_vfs_dq_init(inode); + dentry->d_inode = inode; + dentry->d_sb = inode->i_sb; + rc = osd_removexattr(dentry, inode, name); + } + 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; } static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, __u64 start, __u64 end) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - struct file *file = &info->oti_file; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + struct file *file = &info->oti_file; + int rc; ENTRY; @@ -3618,58 +4727,61 @@ 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. */ static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o, - const struct dt_index_features *feat) -{ - struct iam_descr *descr; - - if (osd_object_is_root(o)) - return feat == &dt_directory_features; - - LASSERT(o->oo_dir != NULL); - - descr = o->oo_dir->od_container.ic_descr; - if (feat == &dt_directory_features) { - if (descr->id_rec_size == sizeof(struct osd_fid_pack)) - return 1; - else - return 0; - } else { - return - feat->dif_keysize_min <= descr->id_key_size && - descr->id_key_size <= feat->dif_keysize_max && - feat->dif_recsize_min <= descr->id_rec_size && - descr->id_rec_size <= feat->dif_recsize_max && - !(feat->dif_flags & (DT_IND_VARKEY | - DT_IND_VARREC | DT_IND_NONUNQ)) && - ergo(feat->dif_flags & DT_IND_UPDATE, - 1 /* XXX check that object (and file system) is - * writable */); - } + const struct dt_index_features *feat) +{ + struct iam_descr *descr; + + if (osd_object_is_root(o)) + return feat == &dt_directory_features; + + LASSERT(o->oo_dir != NULL); + + descr = o->oo_dir->od_container.ic_descr; + if (feat == &dt_directory_features) { + if (descr->id_rec_size == sizeof(struct osd_fid_pack)) + return 1; + else + return 0; + } else { + return feat->dif_keysize_min <= descr->id_key_size && + descr->id_key_size <= feat->dif_keysize_max && + feat->dif_recsize_min <= descr->id_rec_size && + descr->id_rec_size <= feat->dif_recsize_max && + !(feat->dif_flags & (DT_IND_VARKEY | + DT_IND_VARREC | DT_IND_NONUNQ)) && + ergo(feat->dif_flags & DT_IND_UPDATE, + 1 /* XXX check that object (and fs) is writable */); + } } static int osd_iam_container_init(const struct lu_env *env, - struct osd_object *obj, - struct osd_directory *dir) + struct osd_object *obj, + struct osd_directory *dir) { - struct iam_container *bag = &dir->od_container; - int result; + struct iam_container *bag = &dir->od_container; + int result; - result = iam_container_init(bag, &dir->od_descr, obj->oo_inode); - if (result != 0) - return result; + result = iam_container_init(bag, &dir->od_descr, obj->oo_inode); + if (result != 0) + return result; - result = iam_container_setup(bag); - if (result == 0) - obj->oo_dt.do_index_ops = &osd_index_iam_ops; - else - iam_container_fini(bag); + result = iam_container_setup(bag); + if (result == 0) + obj->oo_dt.do_index_ops = &osd_index_iam_ops; + else + iam_container_fini(bag); - return result; + return result; } @@ -3677,23 +4789,23 @@ static int osd_iam_container_init(const struct lu_env *env, * Concurrency: no external locking is necessary. */ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, - const struct dt_index_features *feat) + const struct dt_index_features *feat) { - int result; - int skip_iam = 0; - struct osd_object *obj = osd_dt_obj(dt); + int result; + int skip_iam = 0; + struct osd_object *obj = osd_dt_obj(dt); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - if (osd_object_is_root(obj)) { - dt->do_index_ops = &osd_index_ea_ops; - result = 0; + if (osd_object_is_root(obj)) { + dt->do_index_ops = &osd_index_ea_ops; + result = 0; } else if (feat == &dt_directory_features) { - dt->do_index_ops = &osd_index_ea_ops; + dt->do_index_ops = &osd_index_ea_ops; if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode)) - result = 0; - else - result = -ENOTDIR; + result = 0; + else + result = -ENOTDIR; skip_iam = 1; } else if (unlikely(feat == &dt_otable_features)) { dt->do_index_ops = &osd_otable_ops; @@ -3702,11 +4814,13 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, dt->do_index_ops = &osd_acct_index_ops; result = 0; skip_iam = 1; - } else if (!osd_has_index(obj)) { - struct osd_directory *dir; + } else if (!osd_has_index(obj)) { + struct osd_directory *dir; + struct osd_device *osd = osd_obj2dev(obj); + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - OBD_ALLOC_PTR(dir); - if (dir != NULL) { + OBD_ALLOC_PTR(dir); + if (dir) { spin_lock(&obj->oo_guard); if (obj->oo_dir == NULL) @@ -3725,30 +4839,50 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, /* * recheck under lock. */ - if (!osd_has_index(obj)) - result = osd_iam_container_init(env, obj, - obj->oo_dir); - else + + if (osd_has_index(obj)) { + result = 0; + goto unlock; + } + + result = osd_iam_container_init(env, obj, obj->oo_dir); + if (result || feat == &dt_lfsck_namespace_features || + feat == &dt_lfsck_layout_orphan_features || + feat == &dt_lfsck_layout_dangling_features) + goto unlock; + + result = osd_index_register(osd, fid, + feat->dif_keysize_max, + feat->dif_recsize_max); + if (result < 0) + CWARN("%s: failed to register index " + DFID": rc = %d\n", + osd_name(osd), PFID(fid), result); + else if (result > 0) result = 0; + else + CDEBUG(D_LFSCK, "%s: index object "DFID + " (%d/%d) registered\n", + osd_name(osd), PFID(fid), + (int)feat->dif_keysize_max, + (int)feat->dif_recsize_max); + +unlock: up_write(&obj->oo_ext_idx_sem); - } else { - result = -ENOMEM; - } - } else { - result = 0; - } + } else { + result = -ENOMEM; + } + } else { + result = 0; + } if (result == 0 && skip_iam == 0) { - if (!osd_iam_index_probe(env, obj, feat)) - result = -ENOTDIR; - } - 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); + if (!osd_iam_index_probe(env, obj, feat)) + result = -ENOTDIR; + } + LINVRNT(osd_invariant(obj)); - return result; + return result; } static int osd_otable_it_attr_get(const struct lu_env *env, @@ -3760,63 +4894,32 @@ static int osd_otable_it_attr_get(const struct lu_env *env, } static const struct dt_object_operations osd_obj_ops = { - .do_read_lock = osd_object_read_lock, - .do_write_lock = osd_object_write_lock, - .do_read_unlock = osd_object_read_unlock, - .do_write_unlock = osd_object_write_unlock, - .do_write_locked = osd_object_write_locked, - .do_attr_get = osd_attr_get, - .do_declare_attr_set = osd_declare_attr_set, - .do_attr_set = osd_attr_set, - .do_ah_init = osd_ah_init, - .do_declare_create = osd_declare_object_create, - .do_create = osd_object_create, - .do_declare_destroy = osd_declare_object_destroy, - .do_destroy = osd_object_destroy, - .do_index_try = osd_index_try, - .do_declare_ref_add = osd_declare_object_ref_add, - .do_ref_add = osd_object_ref_add, - .do_declare_ref_del = osd_declare_object_ref_del, - .do_ref_del = osd_object_ref_del, - .do_xattr_get = osd_xattr_get, - .do_declare_xattr_set = osd_declare_xattr_set, - .do_xattr_set = osd_xattr_set, - .do_declare_xattr_del = osd_declare_xattr_del, - .do_xattr_del = osd_xattr_del, - .do_xattr_list = osd_xattr_list, - .do_object_sync = osd_object_sync, -}; - -/** - * dt_object_operations for interoperability mode - * (i.e. to run 2.0 mds on 1.8 disk) (b11826) - */ -static const struct dt_object_operations osd_obj_ea_ops = { - .do_read_lock = osd_object_read_lock, - .do_write_lock = osd_object_write_lock, - .do_read_unlock = osd_object_read_unlock, - .do_write_unlock = osd_object_write_unlock, - .do_write_locked = osd_object_write_locked, - .do_attr_get = osd_attr_get, - .do_declare_attr_set = osd_declare_attr_set, - .do_attr_set = osd_attr_set, - .do_ah_init = osd_ah_init, - .do_declare_create = osd_declare_object_create, - .do_create = osd_object_ea_create, - .do_declare_destroy = osd_declare_object_destroy, - .do_destroy = osd_object_destroy, - .do_index_try = osd_index_try, - .do_declare_ref_add = osd_declare_object_ref_add, - .do_ref_add = osd_object_ref_add, - .do_declare_ref_del = osd_declare_object_ref_del, - .do_ref_del = osd_object_ref_del, - .do_xattr_get = osd_xattr_get, - .do_declare_xattr_set = osd_declare_xattr_set, - .do_xattr_set = osd_xattr_set, - .do_declare_xattr_del = osd_declare_xattr_del, - .do_xattr_del = osd_xattr_del, - .do_xattr_list = osd_xattr_list, - .do_object_sync = osd_object_sync, + .do_read_lock = osd_read_lock, + .do_write_lock = osd_write_lock, + .do_read_unlock = osd_read_unlock, + .do_write_unlock = osd_write_unlock, + .do_write_locked = osd_write_locked, + .do_attr_get = osd_attr_get, + .do_declare_attr_set = osd_declare_attr_set, + .do_attr_set = osd_attr_set, + .do_ah_init = osd_ah_init, + .do_declare_create = osd_declare_create, + .do_create = osd_create, + .do_declare_destroy = osd_declare_destroy, + .do_destroy = osd_destroy, + .do_index_try = osd_index_try, + .do_declare_ref_add = osd_declare_ref_add, + .do_ref_add = osd_ref_add, + .do_declare_ref_del = osd_declare_ref_del, + .do_ref_del = osd_ref_del, + .do_xattr_get = osd_xattr_get, + .do_declare_xattr_set = osd_declare_xattr_set, + .do_xattr_set = osd_xattr_set, + .do_declare_xattr_del = osd_declare_xattr_del, + .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 = { @@ -3825,14 +4928,14 @@ static const struct dt_object_operations osd_obj_otable_it_ops = { }; static int osd_index_declare_iam_delete(const struct lu_env *env, - struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle) + struct dt_object *dt, + const struct dt_key *key, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); /* Recycle may cause additional three blocks to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_DELETE, @@ -3857,11 +4960,12 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, struct thandle *handle) { struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thandle *oh; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thandle *oh; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -3874,13 +4978,13 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, handle, OSD_OT_DELETE); - ipd = osd_idx_ipd_get(env, bag); - if (unlikely(ipd == NULL)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* swab quota uid/gid provided by caller */ @@ -3888,11 +4992,11 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, key = (const struct dt_key *)&oti->oti_quota_id; } - rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); - osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); + osd_ipd_put(env, bag, ipd); + LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); - RETURN(rc); + RETURN(rc); } static int osd_index_declare_ea_delete(const struct lu_env *env, @@ -3901,8 +5005,9 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; - struct inode *inode; - int rc; + struct inode *inode; + int rc, credits; + ENTRY; LASSERT(!dt_object_remote(dt)); @@ -3911,48 +5016,50 @@ 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]; + osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits); inode = osd_dt_obj(dt)->oo_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), + NULL, OSD_QID_BLK); RETURN(rc); } static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, - struct dt_rec *fid) + struct dt_rec *fid) { - struct osd_fid_pack *rec; - int rc = -ENODATA; + struct osd_fid_pack *rec; + int rc = -ENODATA; - if (de->file_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack *) (de->name + de->name_len + 1); - rc = osd_fid_unpack((struct lu_fid *)fid, rec); + if (de->file_type & LDISKFS_DIRENT_LUFID) { + rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + rc = osd_fid_unpack((struct lu_fid *)fid, rec); if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid))) rc = -EINVAL; - } + } return rc; } static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct seq_server_site *ss = osd_seq_site(osd); + struct seq_server_site *ss = osd_seq_site(osd); + ENTRY; /* FID seqs not in FLDB, must be local seq */ if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) RETURN(0); - /* If FLD is not being initialized yet, it only happens during the + /* + * If FLD is not being initialized yet, it only happens during the * initialization, likely during mgs initialization, and we assume - * this is local FID. */ + * this is local FID. + */ if (ss == NULL || ss->ss_server_fld == NULL) RETURN(0); @@ -3963,6 +5070,36 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, RETURN(1); } +static void osd_take_care_of_agent(const struct lu_env *env, + struct osd_device *osd, + struct osd_thandle *oh, + struct ldiskfs_dir_entry_2 *de) +{ + struct lu_fid *fid = &osd_oti_get(env)->oti_fid; + struct osd_idmap_cache *idc; + int rc, schedule = 0; + + LASSERT(de != NULL); + + rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); + if (likely(rc == 0)) { + idc = osd_idc_find_or_init(env, osd, fid); + if (IS_ERR(idc) || idc->oic_remote) + schedule = 1; + } else if (rc == -ENODATA) { + /* + * can't get FID, postpone to the end of the + * transaction when iget() is safe + */ + schedule = 1; + } else { + CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc); + } + if (schedule) + osd_schedule_agent_inode_removal(env, oh, + le32_to_cpu(de->inode)); +} + /** * Index delete function for interoperability mode (b11826). * It will remove the directory entry added by osd_index_ea_insert(). @@ -3976,16 +5113,16 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct osd_thandle *oh; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct osd_thandle *oh; struct ldiskfs_dir_entry_2 *de = NULL; - struct buffer_head *bh; - struct htree_lock *hlock = NULL; - struct lu_fid *fid = &osd_oti_get(env)->oti_fid; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - int rc; + struct buffer_head *bh; + struct htree_lock *hlock = NULL; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -3997,25 +5134,26 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, handle, OSD_OT_DELETE); - oh = container_of(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); ll_vfs_dq_init(dir); - dentry = osd_child_dentry_get(env, obj, - (char *)key, strlen((char *)key)); - - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - dir, LDISKFS_HLOCK_DEL); - } else { + dentry = osd_child_dentry_get(env, obj, + (char *)key, strlen((char *)key)); + + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + dir, LDISKFS_HLOCK_DEL); + } else { down_write(&obj->oo_ext_idx_sem); - } + } - bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); - if (bh) { - /* If this is not the ".." entry, it might be a remote DNE + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + 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. * upgraded 1.8 filesystem without dirdata enabled) then @@ -4028,55 +5166,25 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, * the entry first might leak the agent inode afterward. The * reverse would need filesystem abort in case of error deleting * the entry after the agent had been removed, or leave a - * dangling entry pointing at a random inode. */ - if (strcmp((char *)key, dotdot) != 0) { - LASSERT(de != NULL); - rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); - if (rc == -ENODATA) { - /* can't get FID, postpone to the end of the - * transaction when iget() is safe */ - osd_schedule_agent_inode_removal(env, oh, - le32_to_cpu(de->inode)); - } else if (rc == 0 && - unlikely(osd_remote_fid(env, osd, fid))) { - osd_schedule_agent_inode_removal(env, oh, - le32_to_cpu(de->inode)); - } - } - rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); - brelse(bh); - } else { - rc = -ENOENT; - } - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + * dangling entry pointing at a random inode. + */ + if (strcmp((char *)key, dotdot) != 0) + osd_take_care_of_agent(env, osd, oh, de); + 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 up_write(&obj->oo_ext_idx_sem); - if (rc != 0) - GOTO(out, rc); - - /* For inode on the remote MDT, .. will point to - * /Agent directory, Check whether it needs to delete - * from agent directory */ - if (unlikely(strcmp((char *)key, dotdot) == 0)) { - rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj, - oh); - if (rc != 0 && rc != -ENOENT) { - CERROR("%s: delete agent inode "DFID": rc = %d\n", - osd_name(osd), PFID(fid), rc); - } - - if (rc == -ENOENT) - rc = 0; - - GOTO(out, rc); - } + GOTO(out, rc); out: - - LASSERT(osd_invariant(obj)); + LASSERT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); - RETURN(rc); + RETURN(rc); } /** @@ -4093,13 +5201,14 @@ out: static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, struct dt_rec *rec, const struct dt_key *key) { - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; struct osd_thread_info *oti = osd_oti_get(env); - struct iam_iterator *it = &oti->oti_idx_it; - struct iam_rec *iam_rec; - int rc; + struct iam_iterator *it = &oti->oti_idx_it; + struct iam_rec *iam_rec; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4109,12 +5218,12 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); - ipd = osd_idx_ipd_get(env, bag); - if (IS_ERR(ipd)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (IS_ERR(ipd)) + RETURN(-ENOMEM); - /* got ipd now we can start iterator. */ - iam_it_init(it, bag, 0, ipd); + /* got ipd now we can start iterator. */ + iam_it_init(it, bag, 0, ipd); if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* swab quota uid/gid provided by caller */ @@ -4122,43 +5231,43 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, key = (const struct dt_key *)&oti->oti_quota_id; } - rc = iam_it_get(it, (struct iam_key *)key); - if (rc >= 0) { - if (S_ISDIR(obj->oo_inode->i_mode)) - iam_rec = (struct iam_rec *)oti->oti_ldp; - else - iam_rec = (struct iam_rec *) rec; + rc = iam_it_get(it, (struct iam_key *)key); + if (rc >= 0) { + if (S_ISDIR(obj->oo_inode->i_mode)) + iam_rec = (struct iam_rec *)oti->oti_ldp; + else + iam_rec = (struct iam_rec *)rec; - iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec); + iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec); - if (S_ISDIR(obj->oo_inode->i_mode)) - osd_fid_unpack((struct lu_fid *) rec, - (struct osd_fid_pack *)iam_rec); + if (S_ISDIR(obj->oo_inode->i_mode)) + osd_fid_unpack((struct lu_fid *)rec, + (struct osd_fid_pack *)iam_rec); else if (fid_is_quota(lu_object_fid(&dt->do_lu))) osd_quota_unpack(obj, rec); - } + } - iam_it_put(it); - iam_it_fini(it); - osd_ipd_put(env, bag, ipd); + iam_it_put(it); + iam_it_fini(it); + osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - RETURN(rc); + RETURN(rc); } static int osd_index_declare_iam_insert(const struct lu_env *env, - struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, - struct thandle *handle) + struct dt_object *dt, + const struct dt_rec *rec, + const struct dt_key *key, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - LASSERT(handle != NULL); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT]); @@ -4179,16 +5288,16 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, */ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_rec *rec, - const struct dt_key *key, struct thandle *th, - int ignore_quota) + const struct dt_key *key, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); struct iam_path_descr *ipd; - struct osd_thandle *oh; - struct iam_container *bag; + struct osd_thandle *oh; + struct iam_container *bag; struct osd_thread_info *oti = osd_oti_get(env); - struct iam_rec *iam_rec; - int rc; + struct iam_rec *iam_rec; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4203,16 +5312,17 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_INSERT); - ipd = osd_idx_ipd_get(env, bag); - if (unlikely(ipd == NULL)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); if (S_ISDIR(obj->oo_inode->i_mode)) { iam_rec = (struct iam_rec *)oti->oti_ldp; - osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid); + osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, + &oti->oti_fid); } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* pack quota uid/gid */ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); @@ -4224,12 +5334,12 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, iam_rec = (struct iam_rec *)rec; } - rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, - iam_rec, ipd); - osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, + iam_rec, ipd); + osd_ipd_put(env, bag, ipd); + LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_INSERT); - RETURN(rc); + RETURN(rc); } /** @@ -4245,14 +5355,14 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, const char *name, const struct lu_fid *fid, struct htree_lock *hlock, struct thandle *th) { - struct ldiskfs_dentry_param *ldp; - struct dentry *child; - struct osd_thandle *oth; - int rc; + struct ldiskfs_dentry_param *ldp; + struct dentry *child; + struct osd_thandle *oth; + int rc; - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle != NULL); - LASSERT(oth->ot_handle->h_transaction != NULL); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle != NULL); + LASSERT(oth->ot_handle->h_transaction != NULL); LASSERT(pobj->oo_inode); ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; @@ -4264,8 +5374,8 @@ 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(info, 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; @@ -4273,7 +5383,7 @@ 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) { @@ -4285,8 +5395,8 @@ 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); } } @@ -4312,22 +5422,22 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, const struct lu_fid *dot_dot_fid, struct thandle *th) { - struct inode *inode = dir->oo_inode; - struct osd_thandle *oth; - int result = 0; + struct inode *inode = dir->oo_inode; + struct osd_thandle *oth; + int result = 0; - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); - LASSERT(S_ISDIR(dir->oo_inode->i_mode)); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); + LASSERT(S_ISDIR(dir->oo_inode->i_mode)); - if (strcmp(name, dot) == 0) { - if (dir->oo_compat_dot_created) { - result = -EEXIST; - } else { + if (strcmp(name, dot) == 0) { + if (dir->oo_compat_dot_created) { + result = -EEXIST; + } else { LASSERT(inode->i_ino == parent_dir->i_ino); - dir->oo_compat_dot_created = 1; - result = 0; - } + dir->oo_compat_dot_created = 1; + result = 0; + } } else if (strcmp(name, dotdot) == 0) { if (!dir->oo_compat_dot_created) return -EINVAL; @@ -4366,31 +5476,31 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, struct inode *cinode, const char *name, const struct lu_fid *fid, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct htree_lock *hlock; - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct htree_lock *hlock; + int rc; - hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL; + hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL; - if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' && - name[2] =='\0'))) { - if (hlock != NULL) { - ldiskfs_htree_lock(hlock, pobj->oo_hl_head, - pobj->oo_inode, 0); - } else { + if (name[0] == '.' && (name[1] == '\0' || + (name[1] == '.' && name[2] == '\0'))) { + if (hlock != NULL) { + ldiskfs_htree_lock(hlock, pobj->oo_hl_head, + pobj->oo_inode, 0); + } else { down_write(&pobj->oo_ext_idx_sem); } rc = osd_add_dot_dotdot(info, pobj, cinode, name, lu_object_fid(&pobj->oo_dt.do_lu), - fid, th); - } else { - if (hlock != NULL) { - ldiskfs_htree_lock(hlock, pobj->oo_hl_head, - pobj->oo_inode, LDISKFS_HLOCK_ADD); - } else { + fid, th); + } else { + if (hlock != NULL) { + ldiskfs_htree_lock(hlock, pobj->oo_hl_head, + pobj->oo_inode, LDISKFS_HLOCK_ADD); + } else { down_write(&pobj->oo_ext_idx_sem); - } + } if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; @@ -4403,40 +5513,51 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); } - } - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + } + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else up_write(&pobj->oo_ext_idx_sem); - return rc; + return rc; } 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 lustre_scrub *scrub = &dev->od_scrub.os_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)) RETURN(0); - if (scrub->os_pos_current > id->oii_ino) + if (thread_is_running(&scrub->os_thread) && + scrub->os_pos_current > id->oii_ino) + RETURN(0); + + if (dev->od_auto_scrub_interval == AS_NEVER || + ktime_get_real_seconds() < + scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval) 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); @@ -4444,43 +5565,69 @@ 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; - /* The inode may has been reused by others, we do not know, - * leave it to be handled by subsequent osd_fid_lookup(). */ - RETURN(0); - } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) { - RETURN(rc); + /* + * The inode may has been reused by others, we do not know, + * leave it to be handled by subsequent osd_fid_lookup(). + */ + GOTO(out, rc = 0); + } else if (rc || osd_id_eq(id, &oti->oti_id)) { + GOTO(out, rc); } + insert = false; + trigger: if (thread_is_running(&scrub->os_thread)) { - rc = osd_oii_insert(dev, oic, rc == -ENOENT); - /* There is race condition between osd_oi_lookup and OI scrub. + 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, - * but try to call osd_oi_lookup() again. */ + * but try to call osd_oi_lookup() again. + */ 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); + + GOTO(out, 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", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, - PFID(fid), rc); + if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) { + rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL | + SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT); + CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING, + "%s: trigger partial OI scrub for RPC inconsistency " + "checking FID "DFID": rc = %d\n", + osd_dev2name(dev), PFID(fid), rc); if (rc == 0 || rc == -EALREADY) goto again; } - RETURN(0); + GOTO(out, rc); + +out: + if (inode) + iput(inode); + + RETURN(rc); } static int osd_fail_fid_lookup(struct osd_thread_info *oti, @@ -4488,9 +5635,9 @@ 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); @@ -4499,12 +5646,12 @@ 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; } @@ -4544,12 +5691,13 @@ static int osd_get_pfid_from_linkea(const struct lu_env *env, struct osd_object *obj, struct lu_fid *fid) { - struct osd_thread_info *oti = osd_oti_get(env); - struct lu_buf *buf = &oti->oti_big_buf; - struct dentry *dentry = &oti->oti_obj_dentry; - struct inode *inode = obj->oo_inode; - struct linkea_data ldata = { NULL }; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct inode *inode = obj->oo_inode; + struct linkea_data ldata = { NULL }; + int rc; + ENTRY; fid_zero(fid); @@ -4586,8 +5734,8 @@ 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); } @@ -4595,6 +5743,49 @@ 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 @@ -4606,14 +5797,15 @@ again: static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, struct dt_rec *rec, const struct dt_key *key) { - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct ldiskfs_dir_entry_2 *de; - struct buffer_head *bh; - struct lu_fid *fid = (struct lu_fid *) rec; - struct htree_lock *hlock = NULL; - int ino; - int rc; + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + struct lu_fid *fid = (struct lu_fid *)rec; + struct htree_lock *hlock = NULL; + int ino; + int rc; + ENTRY; LASSERT(dir->i_op != NULL); @@ -4631,7 +5823,7 @@ 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; @@ -4649,12 +5841,14 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, /* done with de, release bh */ brelse(bh); if (rc != 0) { - if (unlikely(ino == osd_remote_parent_ino(dev))) { + if (unlikely(is_remote_parent_ino(dev, ino))) { const char *name = (const char *)key; - /* If the parent is on remote MDT, and there + /* + * If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get - * the parent FID from the linkEA. */ + * the parent FID from the linkEA. + */ if (likely(strlen(name) == 2 && name[0] == '.' && name[1] == '.')) rc = osd_get_pfid_from_linkea(env, obj, @@ -4674,10 +5868,13 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid); rc = osd_consistency_check(oti, dev, oic); - if (rc != 0) + if (rc == -ENOENT) fid_zero(&oic->oic_fid); + else + /* Other error should not affect lookup result. */ + rc = 0; } else { - rc = -ENOENT; + rc = PTR_ERR(bh); } GOTO(out, rc); @@ -4690,29 +5887,19 @@ out: return rc; } -/** - * 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, const struct dt_key *key, struct thandle *handle) { - struct osd_thandle *oh; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - 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; + struct osd_thandle *oh; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + 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)); @@ -4725,16 +5912,20 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, credits = osd_dto_credits_noquota[DTO_INDEX_INSERT]; - /* we can't call iget() while a transactions is running + /* + * 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 */ + * 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 */ + /* + * 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]; } @@ -4744,12 +5935,31 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, if (osd_dt_obj(dt)->oo_inode != NULL) { struct inode *inode = osd_dt_obj(dt)->oo_inode; - /* We ignore block quota on meta pool (MDTs), so needn't + /* + * We ignore block quota on meta pool (MDTs), so needn't * calculate how many blocks will be consumed by this index - * insert */ + * 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), NULL, + OSD_QID_BLK); + if (rc) + RETURN(rc); + +#ifdef HAVE_PROJECT_QUOTA + /* + * Reserve credits for local agent inode to transfer + * to 0, quota enforcement is ignored in this case. + */ + if (idc->oic_remote && + LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL && + i_projid_read(inode) != 0) + rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh, + 0, i_projid_read(inode), + 0, false, PRJQUOTA, true); +#endif } RETURN(rc); @@ -4768,18 +5978,18 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, */ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_rec *rec, - const struct dt_key *key, struct thandle *th, - int ignore_quota) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; - const struct lu_fid *fid = rec1->rec_fid; - const char *name = (const char *)key; - struct osd_thread_info *oti = osd_oti_get(env); - struct inode *child_inode = NULL; - struct osd_idmap_cache *idc; - int rc; + const struct dt_key *key, struct thandle *th) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; + const char *name = (const char *)key; + struct osd_thread_info *oti = osd_oti_get(env); + struct inode *child_inode = NULL; + struct osd_idmap_cache *idc; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4795,37 +6005,31 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, 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)) + if (IS_ERR(idc)) { + /* + * 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. + * + * normally this error should be placed after the first + * find, but migrate may attach source stripes to + * target, which doesn't create stripes. + */ + CERROR("%s: "DFID" wasn't declared for insert\n", + osd_name(osd), PFID(fid)); + dump_stack(); RETURN(PTR_ERR(idc)); + } } if (idc->oic_remote) { /* Insert remote entry */ if (strcmp(name, dotdot) == 0 && strlen(name) == 2) { - struct osd_mdobj_map *omm = osd->od_mdt_map; - struct osd_thandle *oh; - - /* If parent on remote MDT, we need put this object - * under AGENT */ - oh = container_of(th, typeof(*oh), ot_super); - rc = osd_add_to_remote_parent(env, osd, obj, oh); - if (rc != 0) { - CERROR("%s: add "DFID" error: rc = %d\n", - osd_name(osd), - PFID(lu_object_fid(&dt->do_lu)), rc); - RETURN(rc); - } - - child_inode = igrab(omm->omm_remote_parent->d_inode); + child_inode = + igrab(osd->od_mdt_map->omm_remote_parent->d_inode); } else { child_inode = osd_create_local_agent_inode(env, osd, obj, fid, rec1->rec_type & S_IFMT, th); @@ -4843,6 +6047,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = oti->oti_inode; if (unlikely(child_inode == NULL)) { struct ldiskfs_inode_info *lii; + OBD_ALLOC_PTR(lii); if (lii == NULL) RETURN(-ENOMEM); @@ -4875,11 +6080,11 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, struct dt_object *dt, __u32 unused) { - struct osd_it_iam *it; - struct osd_object *obj = osd_dt_obj(dt); - struct lu_object *lo = &dt->do_lu; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_it_iam *it; + struct osd_object *obj = osd_dt_obj(dt); + struct lu_object *lo = &dt->do_lu; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; if (!dt_object_exists(dt)) return ERR_PTR(-ENOENT); @@ -4904,15 +6109,14 @@ 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; - struct osd_object *obj = it->oi_obj; + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; iam_it_fini(&it->oi_it); osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); - lu_object_put(env, &obj->oo_dt.do_lu); + osd_object_put(env, obj); OBD_FREE_PTR(it); } @@ -4928,10 +6132,10 @@ static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) */ static int osd_it_iam_get(const struct lu_env *env, - struct dt_it *di, const struct dt_key *key) + struct dt_it *di, const struct dt_key *key) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_it_iam *it = (struct osd_it_iam *)di; if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) { /* swab quota uid/gid */ @@ -4939,7 +6143,7 @@ static int osd_it_iam_get(const struct lu_env *env, key = (struct dt_key *)&oti->oti_quota_id; } - return iam_it_get(&it->oi_it, (const struct iam_key *)key); + return iam_it_get(&it->oi_it, (const struct iam_key *)key); } /** @@ -4949,9 +6153,9 @@ static int osd_it_iam_get(const struct lu_env *env, */ static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - iam_it_put(&it->oi_it); + iam_it_put(&it->oi_it); } /** @@ -4966,9 +6170,9 @@ static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_next(&it->oi_it); + return iam_it_next(&it->oi_it); } /** @@ -4976,12 +6180,12 @@ static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di) */ static struct dt_key *osd_it_iam_key(const struct lu_env *env, - const struct dt_it *di) + const struct dt_it *di) { struct osd_thread_info *oti = osd_oti_get(env); - struct osd_it_iam *it = (struct osd_it_iam *)di; - struct osd_object *obj = it->oi_obj; - struct dt_key *key; + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; + struct dt_key *key; key = (struct dt_key *)iam_it_key_get(&it->oi_it); @@ -5000,9 +6204,9 @@ static struct dt_key *osd_it_iam_key(const struct lu_env *env, static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_key_size(&it->oi_it); + return iam_it_key_size(&it->oi_it); } static inline void @@ -5047,21 +6251,22 @@ osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset, * Return pointer to the record under iterator. */ static int osd_it_iam_rec(const struct lu_env *env, - const struct dt_it *di, - struct dt_rec *dtrec, __u32 attr) + const struct dt_it *di, + struct dt_rec *dtrec, __u32 attr) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; struct osd_thread_info *info = osd_oti_get(env); + ENTRY; if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) { const struct osd_fid_pack *rec; - struct lu_fid *fid = &info->oti_fid; - struct lu_dirent *lde = (struct lu_dirent *)dtrec; - char *name; - int namelen; - __u64 hash; - int rc; + struct lu_fid *fid = &info->oti_fid; + struct lu_dirent *lde = (struct lu_dirent *)dtrec; + char *name; + int namelen; + __u64 hash; + int rc; name = (char *)iam_it_key_get(&it->oi_it); if (IS_ERR(name)) @@ -5099,9 +6304,9 @@ static int osd_it_iam_rec(const struct lu_env *env, */ static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_store(&it->oi_it); + return iam_it_store(&it->oi_it); } /** @@ -5116,31 +6321,31 @@ static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di) */ static int osd_it_iam_load(const struct lu_env *env, - const struct dt_it *di, __u64 hash) + const struct dt_it *di, __u64 hash) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_load(&it->oi_it, hash); + return iam_it_load(&it->oi_it, hash); } static const struct dt_index_operations osd_index_iam_ops = { - .dio_lookup = osd_index_iam_lookup, - .dio_declare_insert = osd_index_declare_iam_insert, - .dio_insert = osd_index_iam_insert, - .dio_declare_delete = osd_index_declare_iam_delete, - .dio_delete = osd_index_iam_delete, - .dio_it = { - .init = osd_it_iam_init, - .fini = osd_it_iam_fini, - .get = osd_it_iam_get, - .put = osd_it_iam_put, - .next = osd_it_iam_next, - .key = osd_it_iam_key, - .key_size = osd_it_iam_key_size, - .rec = osd_it_iam_rec, - .store = osd_it_iam_store, - .load = osd_it_iam_load - } + .dio_lookup = osd_index_iam_lookup, + .dio_declare_insert = osd_index_declare_iam_insert, + .dio_insert = osd_index_iam_insert, + .dio_declare_delete = osd_index_declare_iam_delete, + .dio_delete = osd_index_iam_delete, + .dio_it = { + .init = osd_it_iam_init, + .fini = osd_it_iam_fini, + .get = osd_it_iam_get, + .put = osd_it_iam_put, + .next = osd_it_iam_next, + .key = osd_it_iam_key, + .key_size = osd_it_iam_key_size, + .rec = osd_it_iam_rec, + .store = osd_it_iam_store, + .load = osd_it_iam_load + } }; @@ -5154,12 +6359,13 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, struct dt_object *dt, __u32 attr) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *oie; - struct file *file; - struct lu_object *lo = &dt->do_lu; - struct dentry *obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie; + struct file *file; + struct lu_object *lo = &dt->do_lu; + struct dentry *obj_dentry; + ENTRY; if (!dt_object_exists(dt) || obj->oo_destroyed) @@ -5185,7 +6391,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, if (oie->oie_buf == NULL) RETURN(ERR_PTR(-ENOMEM)); } - oie->oie_obj = obj; + oie->oie_obj = obj; file = &oie->oie_file; @@ -5200,7 +6406,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, set_file_inode(file, obj->oo_inode); lu_object_get(lo); - RETURN((struct dt_it *) oie); + RETURN((struct dt_it *)oie); } /** @@ -5210,14 +6416,14 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, */ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *oie = (struct osd_it_ea *)di; - struct osd_object *obj = oie->oie_obj; - struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie = (struct osd_it_ea *)di; + struct osd_object *obj = oie->oie_obj; + struct inode *inode = obj->oo_inode; ENTRY; 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 @@ -5237,18 +6443,18 @@ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty(). */ static int osd_it_ea_get(const struct lu_env *env, - struct dt_it *di, const struct dt_key *key) + struct dt_it *di, const struct dt_key *key) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - ENTRY; - LASSERT(((const char *)key)[0] == '\0'); - it->oie_file.f_pos = 0; - it->oie_rd_dirent = 0; - it->oie_it_dirent = 0; - it->oie_dirent = NULL; + ENTRY; + LASSERT(((const char *)key)[0] == '\0'); + it->oie_file.f_pos = 0; + it->oie_rd_dirent = 0; + it->oie_it_dirent = 0; + it->oie_dirent = NULL; - RETURN(+1); + RETURN(+1); } /** @@ -5276,37 +6482,36 @@ struct osd_filldir_cbs { * \retval 1 on buffer full */ #ifdef HAVE_FILLDIR_USE_CTX -static int osd_ldiskfs_filldir(struct dir_context *buf, - const char *name, int namelen, +static int osd_ldiskfs_filldir(struct dir_context *buf, #else -static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, +static int osd_ldiskfs_filldir(void *buf, #endif - loff_t offset, __u64 ino, - unsigned d_type) -{ - 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; - struct osd_fid_pack *rec; - ENTRY; - - /* this should never happen */ - if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) { - CERROR("ldiskfs return invalid namelen %d\n", namelen); - RETURN(-EIO); - } - - if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen > - OSD_IT_EA_BUFSIZE) - RETURN(1); + const char *name, int namelen, + loff_t offset, __u64 ino, unsigned int d_type) +{ + 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; + struct osd_fid_pack *rec; + + ENTRY; + +/* this should never happen */ + if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) { + CERROR("ldiskfs return invalid namelen %d\n", namelen); + RETURN(-EIO); + } + + if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen > + OSD_IT_EA_BUFSIZE) + RETURN(1); /* "." is just the object itself. */ if (namelen == 1 && name[0] == '.') { *fid = obj->oo_dt.do_lu.lo_header->loh_fid; } else if (d_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack*) (name + namelen + 1); + rec = (struct osd_fid_pack *)(name + namelen + 1); if (osd_fid_unpack(fid, rec) != 0) fid_zero(fid); } else { @@ -5320,16 +6525,16 @@ static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, *fid = obj->oo_dt.do_lu.lo_header->loh_fid; } - ent->oied_ino = ino; - ent->oied_off = offset; - ent->oied_namelen = namelen; - ent->oied_type = d_type; + ent->oied_ino = ino; + ent->oied_off = offset; + ent->oied_namelen = namelen; + ent->oied_type = d_type; - memcpy(ent->oied_name, name, namelen); + memcpy(ent->oied_name, name, namelen); - it->oie_rd_dirent++; - it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen); - RETURN(0); + it->oie_rd_dirent++; + it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen); + RETURN(0); } /** @@ -5343,14 +6548,14 @@ static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, * \retval +1 reach the end of entry */ static int osd_ldiskfs_it_fill(const struct lu_env *env, - const struct dt_it *di) -{ - struct osd_it_ea *it = (struct osd_it_ea *)di; - struct osd_object *obj = it->oie_obj; - struct inode *inode = obj->oo_inode; - struct htree_lock *hlock = NULL; - struct file *filp = &it->oie_file; - int rc = 0; + const struct dt_it *di) +{ + struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_object *obj = it->oie_obj; + struct inode *inode = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct file *filp = &it->oie_file; + int rc = 0; struct osd_filldir_cbs buf = { #ifdef HAVE_DIR_CONTEXT .ctx.actor = osd_ldiskfs_filldir, @@ -5358,34 +6563,40 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, .it = it }; - ENTRY; - it->oie_dirent = it->oie_buf; - it->oie_rd_dirent = 0; + ENTRY; + it->oie_dirent = it->oie_buf; + it->oie_rd_dirent = 0; - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - inode, LDISKFS_HLOCK_READDIR); - } else { + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + inode, LDISKFS_HLOCK_READDIR); + } else { down_read(&obj->oo_ext_idx_sem); - } + } #ifdef HAVE_DIR_CONTEXT buf.ctx.pos = filp->f_pos; +#ifdef HAVE_ITERATE_SHARED + rc = inode->i_fop->iterate_shared(filp, &buf.ctx); +#else rc = inode->i_fop->iterate(filp, &buf.ctx); +#endif filp->f_pos = buf.ctx.pos; #else rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir); #endif - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else up_read(&obj->oo_ext_idx_sem); if (it->oie_rd_dirent == 0) { - /*If it does not get any dirent, it means it has been reached - *to the end of the dir */ + /* + * If it does not get any dirent, it means it has been reached + * to the end of the dir + */ it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file); if (rc == 0) rc = 1; @@ -5410,26 +6621,26 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, */ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - int rc; + struct osd_it_ea *it = (struct osd_it_ea *)di; + int rc; - ENTRY; + ENTRY; - if (it->oie_it_dirent < it->oie_rd_dirent) { - it->oie_dirent = - (void *) it->oie_dirent + - cfs_size_round(sizeof(struct osd_it_ea_dirent) + - it->oie_dirent->oied_namelen); - it->oie_it_dirent++; - RETURN(0); - } else { + if (it->oie_it_dirent < it->oie_rd_dirent) { + it->oie_dirent = + (void *)it->oie_dirent + + cfs_size_round(sizeof(struct osd_it_ea_dirent) + + it->oie_dirent->oied_namelen); + it->oie_it_dirent++; + rc = 0; + } else { if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file)) - rc = +1; - else - rc = osd_ldiskfs_it_fill(env, di); - } + rc = 1; + else + rc = osd_ldiskfs_it_fill(env, di); + } - RETURN(rc); + RETURN(rc); } /** @@ -5440,15 +6651,15 @@ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) * \retval key i.e. struct dt_key on success */ static struct dt_key *osd_it_ea_key(const struct lu_env *env, - const struct dt_it *di) + const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return (struct dt_key *)it->oie_dirent->oied_name; + return (struct dt_key *)it->oie_dirent->oied_name; } /** - * Returns the key's size at current position from iterator's in memory structure. + * Returns key's size at current position from iterator's in memory structure. * * \param di iterator's in memory structure * @@ -5456,19 +6667,15 @@ static struct dt_key *osd_it_ea_key(const struct lu_env *env, */ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return it->oie_dirent->oied_namelen; + 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; @@ -5476,10 +6683,10 @@ 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 int 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))) @@ -5489,27 +6696,27 @@ 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) -{ - struct inode *dir = dentry->d_parent->d_inode; - struct inode *inode = dentry->d_inode; - struct osd_fid_pack *rec; +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; + struct osd_fid_pack *rec; struct ldiskfs_dentry_param *ldp; - int namelen = dentry->d_name.len; - int rc; - struct osd_thread_info *info = osd_oti_get(env); + 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, - LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + if (!ldiskfs_has_feature_dirdata(inode->i_sb)) 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); @@ -5524,7 +6731,7 @@ 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) @@ -5534,13 +6741,15 @@ 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(info, jh, dentry, inode, hlock); - /* It is too bad, we cannot reinsert the name entry back. - * That means we lose it! */ + 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) - CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, " + CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, " "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + osd_ino2name(inode), dir->i_ino, dir->i_generation, namelen, dentry->d_name.name, PFID(fid), rc); @@ -5552,39 +6761,53 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, struct osd_it_ea *it, struct lu_fid *fid, struct osd_inode_id *id, __u32 *attr) { - struct osd_thread_info *info = osd_oti_get(env); - struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; - struct osd_device *dev = osd_obj2dev(obj); - struct super_block *sb = osd_sb(dev); - const char *devname = - LDISKFS_SB(sb)->s_es->s_volume_name; - struct osd_it_ea_dirent *ent = it->oie_dirent; - struct inode *dir = obj->oo_inode; - struct htree_lock *hlock = NULL; - struct buffer_head *bh = NULL; - handle_t *jh = NULL; + struct osd_thread_info *info = osd_oti_get(env); + 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 = osd_name(dev); + struct osd_it_ea_dirent *ent = it->oie_dirent; + struct inode *dir = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct buffer_head *bh = NULL; + handle_t *jh = NULL; struct ldiskfs_dir_entry_2 *de; - struct dentry *dentry; - struct inode *inode; - int credits; - int rc; - int dot_dotdot = 0; - bool dirty = false; + struct dentry *dentry; + struct inode *inode; + const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu); + int credits; + int rc; + 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); @@ -5592,20 +6815,14 @@ 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. + /* + * We need to ensure that the name entry is still valid. * Because it may be removed or renamed by other already. * * The unlink or rename operation will start journal before PDO lock, @@ -5617,7 +6834,8 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, * We may need to remove the name entry firstly, then insert back. * One credit is for user quota file update. * One credit is for group quota file update. - * Two credits are for dirty inode. */ + * Two credits are for dirty inode. + */ credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; @@ -5627,21 +6845,24 @@ 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); } if (obj->oo_hl_head != NULL) { hlock = osd_oti_get(env)->oti_hlock; - /* "0" means exclusive lock for the whole directory. + /* + * "0" means exclusive lock for the whole directory. * We need to prevent others access such name entry * during the delete + insert. Neither HLOCK_ADD nor - * HLOCK_DEL cannot guarantee the atomicity. */ + * HLOCK_DEL cannot guarantee the atomicity. + */ ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0); } else { down_write(&obj->oo_ext_idx_sem); @@ -5657,124 +6878,135 @@ 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; + * performance much. + */ + 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_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); - dev->od_dirent_journal = 1; - if (hlock != NULL) { - ldiskfs_htree_unlock(hlock); - hlock = NULL; - } else { - up_read(&obj->oo_ext_idx_sem); - } + 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); - dev->od_dirent_journal = 1; - if (hlock != NULL) { - ldiskfs_htree_unlock(hlock); - hlock = NULL; - } else { - up_read(&obj->oo_ext_idx_sem); - } + 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; @@ -5798,40 +7030,47 @@ again: dirty = true; if (unlikely(fid_is_sane(fid))) { - /* FID-in-dirent exists, but FID-in-LMA is lost. - * Trust the FID-in-dirent, and add FID-in-LMA. */ + /* + * FID-in-dirent exists, but FID-in-LMA is lost. + * Trust the FID-in-dirent, and add FID-in-LMA. + */ rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; else - 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); + /* + * It is probably IGIF object. Only aappend the + * FID-in-dirent. OI scrub will process FID-in-LMA. + */ + 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 { @@ -5871,20 +7110,23 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_device *dev = osd_obj2dev(obj); struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; - struct lu_fid *fid = &it->oie_dirent->oied_fid; + struct lu_fid *fid = &it->oie_dirent->oied_fid; struct lu_dirent *lde = (struct lu_dirent *)dtrec; - __u32 ino = it->oie_dirent->oied_ino; - int rc = 0; + __u32 ino = it->oie_dirent->oied_ino; + int rc = 0; + ENTRY; - LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode); + LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino)); if (attr & LUDA_VERIFY) { - if (unlikely(ino == osd_remote_parent_ino(dev))) { + if (unlikely(is_remote_parent_ino(dev, ino))) { attr |= LUDA_IGNORE; - /* If the parent is on remote MDT, and there + /* + * If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get - * the parent FID from the linkEA. */ + * the parent FID from the linkEA. + */ if (!fid_is_sane(fid) && it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && @@ -5895,22 +7137,23 @@ 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)) { bool is_dotdot = false; + if (it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && it->oie_dirent->oied_name[1] == '.') is_dotdot = true; - /* If the parent is on remote MDT, and there + /* + * If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get - * the parent FID from the linkEA. */ - if (ino == osd_remote_parent_ino(dev) && is_dotdot) { + * the parent FID from the linkEA. + */ + if (is_remote_parent_ino(dev, ino) && is_dotdot) { rc = osd_get_pfid_from_linkea(env, obj, fid); } else { if (is_dotdot == false && @@ -5973,9 +7216,9 @@ static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di, */ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return it->oie_dirent->oied_off; + return it->oie_dirent->oied_off; } /** @@ -5989,22 +7232,22 @@ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di) * \retval -ve on error */ static int osd_it_ea_load(const struct lu_env *env, - const struct dt_it *di, __u64 hash) + const struct dt_it *di, __u64 hash) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - int rc; + struct osd_it_ea *it = (struct osd_it_ea *)di; + int rc; - ENTRY; - it->oie_file.f_pos = hash; + ENTRY; + it->oie_file.f_pos = hash; - rc = osd_ldiskfs_it_fill(env, di); + rc = osd_ldiskfs_it_fill(env, di); if (rc > 0) rc = -ENODATA; - if (rc == 0) - rc = +1; + if (rc == 0) + rc = 1; - RETURN(rc); + RETURN(rc); } /** @@ -6018,18 +7261,18 @@ static int osd_it_ea_load(const struct lu_env *env, static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, struct dt_rec *rec, const struct dt_key *key) { - struct osd_object *obj = osd_dt_obj(dt); - int rc = 0; + struct osd_object *obj = osd_dt_obj(dt); + int rc = 0; - ENTRY; + ENTRY; - LASSERT(S_ISDIR(obj->oo_inode->i_mode)); - LINVRNT(osd_invariant(obj)); + LASSERT(S_ISDIR(obj->oo_inode->i_mode)); + LINVRNT(osd_invariant(obj)); - rc = osd_ea_lookup_rec(env, obj, rec, key); - if (rc == 0) - rc = +1; - RETURN(rc); + rc = osd_ea_lookup_rec(env, obj, rec, key); + if (rc == 0) + rc = 1; + RETURN(rc); } /** @@ -6058,39 +7301,49 @@ static const struct dt_index_operations osd_index_ea_ops = { }; static void *osd_key_init(const struct lu_context *ctx, - struct lu_context_key *key) + struct lu_context_key *key) { - struct osd_thread_info *info; + struct osd_thread_info *info; - OBD_ALLOC_PTR(info); - if (info == NULL) - return ERR_PTR(-ENOMEM); + OBD_ALLOC_PTR(info); + if (info == NULL) + return ERR_PTR(-ENOMEM); - OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - if (info->oti_it_ea_buf == NULL) - goto out_free_info; + OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); + if (info->oti_it_ea_buf == NULL) + goto out_free_info; - info->oti_env = container_of(ctx, struct lu_env, le_ctx); + info->oti_env = container_of(ctx, struct lu_env, le_ctx); - info->oti_hlock = ldiskfs_htree_lock_alloc(); - if (info->oti_hlock == NULL) - goto out_free_ea; + info->oti_hlock = ldiskfs_htree_lock_alloc(); + if (info->oti_hlock == NULL) + goto out_free_ea; - return info; + return info; - out_free_ea: - OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - out_free_info: - OBD_FREE_PTR(info); - return ERR_PTR(-ENOMEM); +out_free_ea: + OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); +out_free_info: + OBD_FREE_PTR(info); + return ERR_PTR(-ENOMEM); } static void osd_key_fini(const struct lu_context *ctx, - struct lu_context_key *key, void* data) + 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; + struct osd_idmap_cache *idc = info->oti_ins_cache; + + if (info->oti_dio_pages) { + int i; + for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) { + if (info->oti_dio_pages[i]) + __free_page(info->oti_dio_pages[i]); + } + OBD_FREE(info->oti_dio_pages, + sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES); + } if (info->oti_inode != NULL) OBD_FREE_PTR(lli); @@ -6099,6 +7352,7 @@ static void osd_key_fini(const struct lu_context *ctx, 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_iobuf.dr_lnb_buf); lu_buf_free(&info->oti_big_buf); if (idc != NULL) { LASSERT(info->oti_ins_cache_size > 0); @@ -6110,41 +7364,42 @@ static void osd_key_fini(const struct lu_context *ctx, } static void osd_key_exit(const struct lu_context *ctx, - struct lu_context_key *key, void *data) + struct lu_context_key *key, void *data) { - struct osd_thread_info *info = data; + struct osd_thread_info *info = data; - LASSERT(info->oti_r_locks == 0); - LASSERT(info->oti_w_locks == 0); - LASSERT(info->oti_txns == 0); + LASSERT(info->oti_r_locks == 0); + LASSERT(info->oti_w_locks == 0); + LASSERT(info->oti_txns == 0); } /* type constructor/destructor: osd_type_init, osd_type_fini */ LU_TYPE_INIT_FINI(osd, &osd_key); struct lu_context_key osd_key = { - .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL, - .lct_init = osd_key_init, - .lct_fini = osd_key_fini, - .lct_exit = osd_key_exit + .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL, + .lct_init = osd_key_init, + .lct_fini = osd_key_fini, + .lct_exit = osd_key_exit }; static int osd_device_init(const struct lu_env *env, struct lu_device *d, - const char *name, struct lu_device *next) + const char *name, struct lu_device *next) { struct osd_device *osd = osd_dev(d); - if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) - >= sizeof(osd->od_svname)) + if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >= + sizeof(osd->od_svname)) return -E2BIG; return osd_procfs_init(osd, name); } static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) { - struct seq_server_site *ss = osd_seq_site(osd); - int rc; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; if (osd->od_is_ost || osd->od_cl_seq != NULL) @@ -6166,9 +7421,11 @@ static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) } if (ss->ss_node_id == 0) { - /* If the OSD on the sequence controller(MDT0), then allocate + /* + * If the OSD on the sequence controller(MDT0), then allocate * sequence here, otherwise allocate sequence after connected - * to MDT0 (see mdt_register_lwp_callback()). */ + * to MDT0 (see mdt_register_lwp_callback()). + */ rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv, &osd->od_cl_seq->lcs_space, env); } @@ -6191,12 +7448,22 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) ENTRY; /* shutdown quota slave instance associated with the device */ - if (o->od_quota_slave != NULL) { - qsd_fini(env, o->od_quota_slave); - o->od_quota_slave = NULL; + if (o->od_quota_slave_md != NULL) { + struct qsd_instance *qsd = o->od_quota_slave_md; + + o->od_quota_slave_md = NULL; + qsd_fini(env, qsd); + } + + if (o->od_quota_slave_dt != NULL) { + struct qsd_instance *qsd = o->od_quota_slave_dt; + + o->od_quota_slave_dt = NULL; + qsd_fini(env, qsd); } osd_fid_fini(env, o); + osd_scrub_cleanup(env, o); RETURN(0); } @@ -6217,21 +7484,22 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o) } static int osd_mount(const struct lu_env *env, - struct osd_device *o, struct lustre_cfg *cfg) + struct osd_device *o, struct lustre_cfg *cfg) { - const char *name = lustre_cfg_string(cfg, 0); - const char *dev = lustre_cfg_string(cfg, 1); - const char *opts; - unsigned long page, s_flags, lmd_flags = 0; - struct page *__page; + const char *name = lustre_cfg_string(cfg, 0); + const char *dev = lustre_cfg_string(cfg, 1); + const char *opts; + unsigned long page, s_flags, lmd_flags = 0; + struct page *__page; struct file_system_type *type; - char *options = NULL; - char *str; - 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; - ENTRY; + char *options = NULL; + char *str; + struct osd_thread_info *info = osd_oti_get(env); + struct lu_fid *fid = &info->oti_fid; + struct inode *inode; + int rc = 0, force_over_512tb = 0; + + ENTRY; if (o->od_mnt != NULL) RETURN(0); @@ -6254,10 +7522,27 @@ 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); @@ -6275,11 +7560,14 @@ 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); for (rc = 0, str = options; sout[rc]; ) { char *op = strstr(str, sout[rc]); + if (op == NULL) { rc++; str = options; @@ -6302,7 +7590,7 @@ 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) { @@ -6320,34 +7608,45 @@ 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); } -#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); + if (lmd_flags & LMD_FLG_DEV_RDONLY) { + if (priv_dev_set_rdonly) { + priv_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); + } + } else if (priv_dev_check_rdonly && + priv_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)) { + if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) { CERROR("%s: device %s is mounted w/o journal\n", name, dev); GOTO(out_mnt, rc = -EINVAL); } #ifdef LDISKFS_MOUNT_DIRDATA - if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb)) LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA; - else if (!o->od_is_ost) + else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */ CWARN("%s: device %s was upgraded from Lustre-1.x without " "enabling the dirdata feature. If you do not want to " "downgrade to Lustre-1.x again, you can enable it via " @@ -6362,7 +7661,13 @@ static int osd_mount(const struct lu_env *env, } if (lmd_flags & LMD_FLG_NOSCRUB) - o->od_noscrub = 1; + o->od_auto_scrub_interval = AS_NEVER; + + if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) { + /* do not use pagecache with flash-backed storage */ + o->od_writethrough_cache = 0; + o->od_read_cache = 0; + } GOTO(out, rc = 0); @@ -6381,11 +7686,12 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, struct lu_device *d) { struct osd_device *o = osd_dev(d); + ENTRY; + osd_index_backup(env, o, false); osd_shutdown(env, o); osd_procfs_fini(o); - osd_scrub_cleanup(env, o); osd_obj_map_fini(o); osd_umount(env, o); @@ -6396,13 +7702,13 @@ static int osd_device_init0(const struct lu_env *env, struct osd_device *o, struct lustre_cfg *cfg) { - struct lu_device *l = osd2lu_dev(o); + struct lu_device *l = osd2lu_dev(o); struct osd_thread_info *info; - int rc; - int cplen = 0; + int rc; + int cplen = 0; /* if the module was re-loaded, env can loose its keys */ - rc = lu_env_refill((struct lu_env *) env); + rc = lu_env_refill((struct lu_env *)env); if (rc) GOTO(out, rc); info = osd_oti_get(env); @@ -6414,10 +7720,16 @@ static int osd_device_init0(const struct lu_env *env, spin_lock_init(&o->od_osfs_lock); mutex_init(&o->od_otable_mutex); INIT_LIST_HEAD(&o->od_orphan_list); + INIT_LIST_HEAD(&o->od_index_backup_list); + INIT_LIST_HEAD(&o->od_index_restore_list); + spin_lock_init(&o->od_lock); + o->od_index_backup_policy = LIBP_NONE; + o->od_t10_type = 0; o->od_read_cache = 1; o->od_writethrough_cache = 1; o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; + o->od_auto_scrub_interval = AS_DEFAULT; cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), sizeof(o->od_svname)); @@ -6426,6 +7738,7 @@ static int osd_device_init0(const struct lu_env *env, GOTO(out, rc); } + o->od_index_backup_stop = 0; o->od_index = -1; /* -1 means index is invalid */ rc = server_name2index(o->od_svname, &o->od_index, NULL); if (rc == LDD_F_SV_TYPE_OST) @@ -6452,7 +7765,9 @@ static int osd_device_init0(const struct lu_env *env, INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ + o->od_in_init = 1; rc = osd_scrub_setup(env, o); + o->od_in_init = 0; if (rc < 0) GOTO(out_site, rc); @@ -6467,11 +7782,29 @@ static int osd_device_init0(const struct lu_env *env, LASSERT(l->ld_site->ls_linkage.prev != NULL); /* initialize quota slave instance */ - o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev, - o->od_proc_entry); - if (IS_ERR(o->od_quota_slave)) { - rc = PTR_ERR(o->od_quota_slave); - o->od_quota_slave = NULL; + /* currently it's no need to prepare qsd_instance_md for OST */ + if (!o->od_is_ost) { + o->od_quota_slave_md = qsd_init(env, o->od_svname, + &o->od_dt_dev, + o->od_proc_entry, true); + if (IS_ERR(o->od_quota_slave_md)) { + rc = PTR_ERR(o->od_quota_slave_md); + o->od_quota_slave_md = NULL; + GOTO(out_procfs, rc); + } + } + + o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev, + o->od_proc_entry, false); + + if (IS_ERR(o->od_quota_slave_dt)) { + if (o->od_quota_slave_md != NULL) { + qsd_fini(env, o->od_quota_slave_md); + o->od_quota_slave_md = NULL; + } + + rc = PTR_ERR(o->od_quota_slave_dt); + o->od_quota_slave_dt = NULL; GOTO(out_procfs, rc); } @@ -6492,11 +7825,11 @@ out: } static struct lu_device *osd_device_alloc(const struct lu_env *env, - struct lu_device_type *t, - struct lustre_cfg *cfg) + struct lu_device_type *t, + struct lustre_cfg *cfg) { struct osd_device *o; - int rc; + int rc; OBD_ALLOC_PTR(o); if (o == NULL) @@ -6504,8 +7837,10 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, rc = dt_device_init(&o->od_dt_dev, t); if (rc == 0) { - /* Because the ctx might be revived in dt_device_init, - * refill the env here */ + /* + * Because the ctx might be revived in dt_device_init, + * refill the env here + */ lu_env_refill((struct lu_env *)env); rc = osd_device_init0(env, o, cfg); if (rc) @@ -6519,10 +7854,11 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, } static struct lu_device *osd_device_free(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - struct osd_device *o = osd_dev(d); - ENTRY; + struct osd_device *o = osd_dev(d); + + ENTRY; /* XXX: make osd top device in order to release reference */ d->ld_site->ls_top_dev = d; @@ -6532,16 +7868,17 @@ static struct lu_device *osd_device_free(const struct lu_env *env, lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer); } lu_site_fini(&o->od_site); - dt_device_fini(&o->od_dt_dev); - OBD_FREE_PTR(o); - RETURN(NULL); + dt_device_fini(&o->od_dt_dev); + OBD_FREE_PTR(o); + RETURN(NULL); } static int osd_process_config(const struct lu_env *env, - struct lu_device *d, struct lustre_cfg *cfg) + struct lu_device *d, struct lustre_cfg *cfg) { - struct osd_device *o = osd_dev(d); - int rc; + struct osd_device *o = osd_dev(d); + int rc; + ENTRY; switch (cfg->lcfg_command) { @@ -6549,6 +7886,11 @@ static int osd_process_config(const struct lu_env *env, rc = osd_mount(env, o, cfg); break; case LCFG_CLEANUP: + /* + * For the case LCFG_PRE_CLEANUP is not called in advance, + * that may happend if hit failure during mount process. + */ + osd_index_backup(env, o, false); lu_dev_del_linkage(d->ld_site, d); rc = osd_shutdown(env, o); break; @@ -6556,10 +7898,19 @@ static int osd_process_config(const struct lu_env *env, LASSERT(&o->od_dt_dev); rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars, cfg, &o->od_dt_dev); - if (rc > 0 || rc == -ENOSYS) + if (rc > 0 || rc == -ENOSYS) { rc = class_process_proc_param(PARAM_OST, lprocfs_osd_obd_vars, cfg, &o->od_dt_dev); + if (rc > 0) + rc = 0; + } + break; + case LCFG_PRE_CLEANUP: + osd_scrub_stop(o); + osd_index_backup(env, o, + o->od_index_backup_policy != LIBP_NONE); + rc = 0; break; default: rc = -ENOSYS; @@ -6569,18 +7920,24 @@ static int osd_process_config(const struct lu_env *env, } static int osd_recovery_complete(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - struct osd_device *osd = osd_dev(d); - int rc = 0; + struct osd_device *osd = osd_dev(d); + int rc = 0; + ENTRY; - if (osd->od_quota_slave == NULL) + if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL) RETURN(0); - /* start qsd instance on recovery completion, this notifies the quota - * slave code that we are about to process new requests now */ - rc = qsd_start(env, osd->od_quota_slave); + /* + * start qsd instance on recovery completion, this notifies the quota + * slave code that we are about to process new requests now + */ + rc = qsd_start(env, osd->od_quota_slave_dt); + if (rc == 0 && osd->od_quota_slave_md != NULL) + rc = qsd_start(env, osd->od_quota_slave_md); + RETURN(rc); } @@ -6591,9 +7948,10 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { - struct osd_device *osd = osd_dev(obd->obd_lu_dev); - struct lustre_handle conn; - int rc; + struct osd_device *osd = osd_dev(obd->obd_lu_dev); + struct lustre_handle conn; + int rc; + ENTRY; CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects); @@ -6619,7 +7977,8 @@ static int osd_obd_disconnect(struct obd_export *exp) { struct obd_device *obd = exp->exp_obd; struct osd_device *osd = osd_dev(obd->obd_lu_dev); - int rc, release = 0; + int rc, release = 0; + ENTRY; /* Only disconnect the underlying layers on the final disconnect. */ @@ -6639,19 +7998,28 @@ static int osd_obd_disconnect(struct obd_export *exp) static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, struct lu_device *dev) { - struct osd_device *osd = osd_dev(dev); - struct lr_server_data *lsd = + struct osd_device *osd = osd_dev(dev); + struct lr_server_data *lsd = &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd; - int result = 0; + int result = 0; + ENTRY; - if (osd->od_quota_slave != NULL) { - /* set up quota slave objects */ - result = qsd_prepare(env, osd->od_quota_slave); + if (osd->od_quota_slave_md != NULL) { + /* set up quota slave objects for inode */ + result = qsd_prepare(env, osd->od_quota_slave_md); + if (result != 0) + RETURN(result); + } + + if (osd->od_quota_slave_dt != NULL) { + /* set up quota slave objects for block */ + result = qsd_prepare(env, osd->od_quota_slave_dt); if (result != 0) RETURN(result); } + if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) { #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0) if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) { @@ -6681,40 +8049,40 @@ static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, } static const struct lu_object_operations osd_lu_obj_ops = { - .loo_object_init = osd_object_init, - .loo_object_delete = osd_object_delete, - .loo_object_release = osd_object_release, - .loo_object_free = osd_object_free, - .loo_object_print = osd_object_print, - .loo_object_invariant = osd_object_invariant + .loo_object_init = osd_object_init, + .loo_object_delete = osd_object_delete, + .loo_object_release = osd_object_release, + .loo_object_free = osd_object_free, + .loo_object_print = osd_object_print, + .loo_object_invariant = osd_object_invariant }; const struct lu_device_operations osd_lu_ops = { - .ldo_object_alloc = osd_object_alloc, - .ldo_process_config = osd_process_config, - .ldo_recovery_complete = osd_recovery_complete, - .ldo_prepare = osd_prepare, + .ldo_object_alloc = osd_object_alloc, + .ldo_process_config = osd_process_config, + .ldo_recovery_complete = osd_recovery_complete, + .ldo_prepare = osd_prepare, }; static const struct lu_device_type_operations osd_device_type_ops = { - .ldto_init = osd_type_init, - .ldto_fini = osd_type_fini, + .ldto_init = osd_type_init, + .ldto_fini = osd_type_fini, - .ldto_start = osd_type_start, - .ldto_stop = osd_type_stop, + .ldto_start = osd_type_start, + .ldto_stop = osd_type_stop, - .ldto_device_alloc = osd_device_alloc, - .ldto_device_free = osd_device_free, + .ldto_device_alloc = osd_device_alloc, + .ldto_device_free = osd_device_free, - .ldto_device_init = osd_device_init, - .ldto_device_fini = osd_device_fini + .ldto_device_init = osd_device_init, + .ldto_device_fini = osd_device_fini }; static struct lu_device_type osd_device_type = { - .ldt_tags = LU_DEVICE_DT, + .ldt_tags = LU_DEVICE_DT, .ldt_name = LUSTRE_OSD_LDISKFS_NAME, - .ldt_ops = &osd_device_type_ops, - .ldt_ctx_tags = LCT_LOCAL, + .ldt_ops = &osd_device_type_ops, + .ldt_ctx_tags = LCT_LOCAL, }; static int osd_health_check(const struct lu_env *env, struct obd_device *obd) @@ -6740,7 +8108,7 @@ static int __init osd_init(void) { int rc; - LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8); + CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8); #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK) /* please, try to keep osd_thread_info smaller than a page */ CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE); @@ -6752,6 +8120,12 @@ static int __init osd_init(void) if (rc) return rc; +#ifdef CONFIG_KALLSYMS + priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly"); + priv_dev_check_rdonly = + (void *)kallsyms_lookup_name("dev_check_rdonly"); +#endif + rc = class_register_type(&osd_obd_device_ops, NULL, true, lprocfs_osd_module_vars, LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);