X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_compat.c;h=d58b6c87ba7e432b3c37200b726db9bec9f5247c;hp=8965c314f53c9c1af08ff79f22b43ecc395a055a;hb=07660ad33a7d109cced29b6400f99f25adab3f54;hpb=809d6a120a238ec64830ab16969728c8433dbeba diff --git a/lustre/osd-ldiskfs/osd_compat.c b/lustre/osd-ldiskfs/osd_compat.c index 8965c31..d58b6c8 100644 --- a/lustre/osd-ldiskfs/osd_compat.c +++ b/lustre/osd-ldiskfs/osd_compat.c @@ -27,7 +27,7 @@ * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, Intel Corporation. + * Copyright (c) 2012, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -62,19 +62,14 @@ static void osd_push_ctxt(const struct osd_device *dev, struct lvfs_run_ctxt *newctxt, struct lvfs_run_ctxt *save) { - OBD_SET_CTXT_MAGIC(newctxt); - newctxt->pwdmnt = dev->od_mnt; - newctxt->pwd = dev->od_mnt->mnt_root; - newctxt->fs = get_ds(); - - push_ctxt(save, newctxt, NULL); -} - -static void osd_pop_ctxt(const struct osd_device *dev, - struct lvfs_run_ctxt *new, - struct lvfs_run_ctxt *save) -{ - pop_ctxt(save, new, NULL); + OBD_SET_CTXT_MAGIC(newctxt); + newctxt->pwdmnt = dev->od_mnt; + newctxt->pwd = dev->od_mnt->mnt_root; + newctxt->fs = get_ds(); + newctxt->umask = current_umask(); + newctxt->dt = NULL; + + push_ctxt(save, newctxt); } /* utility to make a directory */ @@ -112,7 +107,7 @@ static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt, GOTO(out_up, dchild); } - err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode); + err = vfs_mkdir(dir->d_inode, dchild, mode); if (err) GOTO(out_err, err); @@ -125,13 +120,13 @@ out_up: return dchild; } -int osd_last_rcvd_subdir_count(struct osd_device *osd) +static int osd_last_rcvd_subdir_count(struct osd_device *osd) { struct lr_server_data lsd; struct dentry *dlast; loff_t off; int rc = 0; - int count = FILTER_SUBDIR_COUNT; + int count = OBJ_SUBDIR_COUNT; ENTRY; @@ -172,7 +167,7 @@ static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev) struct osd_mdobj_map *omm; struct dentry *d; struct osd_thread_info *info = osd_oti_get(env); - struct lu_fid *fid = &info->oti_fid; + struct lu_fid *fid = &info->oti_fid3; int rc = 0; ENTRY; @@ -182,8 +177,6 @@ static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev) omm = dev->od_mdt_map; - LASSERT(dev->od_fsops); - parent = osd_sb(dev)->s_root; osd_push_ctxt(dev, &new, &save); @@ -196,18 +189,19 @@ static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev) /* Set LMA for remote parent inode */ lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID); - rc = osd_ea_fid_set(info, d->d_inode, fid, 0); - if (rc != 0) - GOTO(cleanup, rc); + rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0); + + GOTO(cleanup, rc); + cleanup: - pop_ctxt(&save, &new, NULL); + pop_ctxt(&save, &new); if (rc) { if (omm->omm_remote_parent != NULL) dput(omm->omm_remote_parent); OBD_FREE_PTR(omm); dev->od_mdt_map = NULL; } - RETURN(rc); + return rc; } static void osd_mdt_fini(struct osd_device *osd) @@ -231,6 +225,7 @@ int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd, struct osd_thread_info *oti = osd_oti_get(env); struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs; char *name = oti->oti_name; + struct osd_thread_info *info = osd_oti_get(env); struct dentry *dentry; struct dentry *parent; int rc; @@ -253,14 +248,11 @@ int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd, dentry = osd_child_dentry_by_inode(env, parent->d_inode, name, strlen(name)); mutex_lock(&parent->d_inode->i_mutex); - rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode, - NULL); + rc = osd_ldiskfs_add_entry(info, oh->ot_handle, dentry, + obj->oo_inode, NULL); CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd), name, obj->oo_inode->i_ino, parent->d_inode->i_ino); - LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d", - osd_name(osd), parent->d_inode->i_ino, - parent->d_inode->i_nlink); - parent->d_inode->i_nlink++; + ldiskfs_inc_count(oh->ot_handle, parent->d_inode); mark_inode_dirty(parent->d_inode); mutex_unlock(&parent->d_inode->i_mutex); RETURN(rc); @@ -294,7 +286,8 @@ int osd_delete_from_remote_parent(const struct lu_env *env, dentry = osd_child_dentry_by_inode(env, parent->d_inode, name, strlen(name)); mutex_lock(&parent->d_inode->i_mutex); - bh = osd_ldiskfs_find_entry(parent->d_inode, dentry, &de, NULL); + bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de, + NULL, NULL); if (bh == NULL) { mutex_unlock(&parent->d_inode->i_mutex); RETURN(-ENOENT); @@ -302,10 +295,7 @@ int osd_delete_from_remote_parent(const struct lu_env *env, CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd), name, obj->oo_inode->i_ino, parent->d_inode->i_ino); rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh); - LASSERTF(parent->d_inode->i_nlink > 1, "%s: %lu nlink %d", - osd_name(osd), parent->d_inode->i_ino, - parent->d_inode->i_nlink); - parent->d_inode->i_nlink--; + ldiskfs_dec_count(oh->ot_handle, parent->d_inode); mark_inode_dirty(parent->d_inode); mutex_unlock(&parent->d_inode->i_mutex); brelse(bh); @@ -318,6 +308,40 @@ int osd_delete_from_remote_parent(const struct lu_env *env, RETURN(rc); } +int osd_lookup_in_remote_parent(struct osd_thread_info *oti, + struct osd_device *osd, + const struct lu_fid *fid, + struct osd_inode_id *id) +{ + struct osd_mdobj_map *omm = osd->od_mdt_map; + char *name = oti->oti_name; + struct dentry *parent; + struct dentry *dentry; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + int rc; + ENTRY; + + parent = omm->omm_remote_parent; + sprintf(name, DFID_NOBRACE, PFID(fid)); + dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode, + name, strlen(name)); + mutex_lock(&parent->d_inode->i_mutex); + bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de, + NULL, NULL); + if (bh == NULL) { + rc = -ENOENT; + } else { + rc = 0; + osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN); + brelse(bh); + } + mutex_unlock(&parent->d_inode->i_mutex); + if (rc == 0) + osd_add_oi_cache(oti, osd, id, fid); + RETURN(rc); +} + /* * directory structure on legacy OST: * @@ -328,13 +352,16 @@ int osd_delete_from_remote_parent(const struct lu_env *env, * CONFIGS * */ -static int osd_ost_init(struct osd_device *dev) +static int osd_ost_init(const struct lu_env *env, struct osd_device *dev) { - struct lvfs_run_ctxt new; - struct lvfs_run_ctxt save; - struct dentry *rootd = osd_sb(dev)->s_root; - struct dentry *d; - int rc; + struct lvfs_run_ctxt new; + struct lvfs_run_ctxt save; + struct dentry *rootd = osd_sb(dev)->s_root; + struct dentry *d; + struct osd_thread_info *info = osd_oti_get(env); + struct inode *inode; + struct lu_fid *fid = &info->oti_fid3; + int rc; ENTRY; OBD_ALLOC_PTR(dev->od_ost_map); @@ -343,36 +370,55 @@ static int osd_ost_init(struct osd_device *dev) /* to get subdir count from last_rcvd */ rc = osd_last_rcvd_subdir_count(dev); - if (rc < 0) { - OBD_FREE_PTR(dev->od_ost_map); - RETURN(rc); - } + if (rc < 0) + GOTO(cleanup_alloc, rc); dev->od_ost_map->om_subdir_count = rc; rc = 0; - CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list); + INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list); rwlock_init(&dev->od_ost_map->om_seq_list_lock); - sema_init(&dev->od_ost_map->om_dir_init_sem, 1); + mutex_init(&dev->od_ost_map->om_dir_init_mutex); - LASSERT(dev->od_fsops); osd_push_ctxt(dev, &new, &save); - d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1); + d = ll_lookup_one_len("O", rootd, strlen("O")); if (IS_ERR(d)) - GOTO(cleanup, rc = PTR_ERR(d)); + GOTO(cleanup_ctxt, rc = PTR_ERR(d)); + if (d->d_inode == NULL) { + dput(d); + /* The lookup() may be called again inside simple_mkdir(). + * Since the repeated lookup() only be called for "/O" at + * mount time, it will not affect the whole performance. */ + d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1); + if (IS_ERR(d)) + GOTO(cleanup_ctxt, rc = PTR_ERR(d)); + + /* It is quite probably that the device is new formatted. */ + dev->od_maybe_new = 1; + } - ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI); + inode = d->d_inode; dev->od_ost_map->om_root = d; -cleanup: - osd_pop_ctxt(dev, &new, &save); - if (IS_ERR(d)) { - OBD_FREE_PTR(dev->od_ost_map); - RETURN(PTR_ERR(d)); - } + /* 'What the @fid is' is not imporatant, because the object + * has no OI mapping, and only is visible inside the OSD.*/ + lu_igif_build(fid, inode->i_ino, inode->i_generation); + rc = osd_ea_fid_set(info, inode, fid, + LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0); + if (rc) + GOTO(cleanup_dentry, rc); - RETURN(rc); + pop_ctxt(&save, &new); + RETURN(0); + +cleanup_dentry: + dput(d); +cleanup_ctxt: + pop_ctxt(&save, &new); +cleanup_alloc: + OBD_FREE_PTR(dev->od_ost_map); + return rc; } static void osd_seq_free(struct osd_obj_map *map, @@ -380,7 +426,7 @@ static void osd_seq_free(struct osd_obj_map *map, { int j; - cfs_list_del_init(&osd_seq->oos_seq_list); + list_del_init(&osd_seq->oos_seq_list); if (osd_seq->oos_dirs) { for (j = 0; j < osd_seq->oos_subdir_count; j++) { @@ -408,9 +454,8 @@ static void osd_ost_fini(struct osd_device *osd) return; write_lock(&map->om_seq_list_lock); - cfs_list_for_each_entry_safe(osd_seq, tmp, - &map->om_seq_list, - oos_seq_list) { + list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list, + oos_seq_list) { osd_seq_free(map, osd_seq); } write_unlock(&map->om_seq_list_lock); @@ -427,28 +472,30 @@ int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev) ENTRY; /* prepare structures for OST */ - rc = osd_ost_init(dev); + rc = osd_ost_init(env, dev); if (rc) RETURN(rc); /* prepare structures for MDS */ rc = osd_mdt_init(env, dev); + if (rc) + osd_ost_fini(dev); RETURN(rc); } -struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq) +static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq) { struct osd_obj_seq *osd_seq; - cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) { + list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) { if (osd_seq->oos_seq == seq) return osd_seq; } return NULL; } -struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq) +static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq) { struct osd_obj_seq *osd_seq; @@ -464,22 +511,166 @@ void osd_obj_map_fini(struct osd_device *dev) osd_mdt_fini(dev); } +/** + * Update the specified OI mapping. + * + * \retval 1, changed nothing + * \retval 0, changed successfully + * \retval -ve, on error + */ +static int osd_obj_update_entry(struct osd_thread_info *info, + struct osd_device *osd, + struct dentry *dir, const char *name, + const struct lu_fid *fid, + const struct osd_inode_id *id, + handle_t *th) +{ + struct inode *parent = dir->d_inode; + struct dentry *child; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + struct inode *inode; + struct dentry *dentry = &info->oti_obj_dentry; + struct osd_inode_id *oi_id = &info->oti_id3; + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct lu_fid *oi_fid = &lma->lma_self_fid; + int rc; + ENTRY; + + LASSERT(th != NULL); + LASSERT(th->h_transaction != NULL); + + child = &info->oti_child_dentry; + child->d_parent = dir; + child->d_name.hash = 0; + child->d_name.name = name; + child->d_name.len = strlen(name); + + ll_vfs_dq_init(parent); + mutex_lock(&parent->i_mutex); + bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL); + if (bh == NULL) + GOTO(out, rc = -ENOENT); + + if (le32_to_cpu(de->inode) == id->oii_ino) + GOTO(out, rc = 1); + + osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN); + inode = osd_iget(info, osd, oi_id); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + if (rc == -ENOENT || rc == -ESTALE) + goto update; + GOTO(out, rc); + } + + /* The EA inode should NOT be in OI, old OI scrub may added + * such OI mapping by wrong, replace it. */ + if (unlikely(osd_is_ea_inode(inode))) { + iput(inode); + goto update; + } + + rc = osd_get_lma(info, inode, dentry, lma); + if (rc == -ENODATA) { + rc = osd_get_idif(info, inode, dentry, oi_fid); + if (rc > 0 || rc == -ENODATA) { + oi_fid = NULL; + rc = 0; + } + } + iput(inode); + + if (rc != 0) + GOTO(out, rc); + + /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is + * either a crashed object or a uninitialized one. Replace it. */ + if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) { + CERROR("%s: the FID "DFID" is used by two objects: " + "%u/%u %u/%u\n", osd_name(osd), PFID(fid), + oi_id->oii_ino, oi_id->oii_gen, + id->oii_ino, id->oii_gen); + GOTO(out, rc = -EEXIST); + } + + if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) { + __u32 idx1 = fid_idif_ost_idx(fid); + __u32 idx2 = fid_idif_ost_idx(oi_fid); + struct ost_id *ostid = &info->oti_ostid; + struct lu_fid *tfid = &info->oti_fid3; + + LASSERTF(idx1 == 0 || idx1 == osd->od_index, + "invalid given FID "DFID", not match the " + "device index %u\n", PFID(fid), osd->od_index); + + if (idx1 != idx2) { + if (idx1 == 0 && idx2 == osd->od_index) { + fid_to_ostid(fid, ostid); + ostid_to_fid(tfid, ostid, idx2); + if (lu_fid_eq(tfid, oi_fid)) { + CERROR("%s: the FID "DFID" is used by " + "two objects(2): %u/%u %u/%u\n", + osd_name(osd), PFID(fid), + oi_id->oii_ino, oi_id->oii_gen, + id->oii_ino, id->oii_gen); + + GOTO(out, rc = -EEXIST); + } + } else if (idx2 == 0 && idx1 == osd->od_index) { + fid_to_ostid(oi_fid, ostid); + ostid_to_fid(tfid, ostid, idx1); + if (lu_fid_eq(tfid, fid)) { + CERROR("%s: the FID "DFID" is used by " + "two objects(2): %u/%u %u/%u\n", + osd_name(osd), PFID(fid), + oi_id->oii_ino, oi_id->oii_gen, + id->oii_ino, id->oii_gen); + + GOTO(out, rc = -EEXIST); + } + } + } + } + +update: + /* There may be temporary inconsistency: On one hand, the new + * object may be referenced by multiple entries, which is out + * of our control unless we traverse the whole /O completely, + * which is non-flat order and inefficient, should be avoided; + * On the other hand, the old object may become orphan if it + * is still valid. Since it was referenced by an invalid entry, + * making it as invisible temporary may be not worse. OI scrub + * will process it later. */ + rc = ldiskfs_journal_get_write_access(th, bh); + if (rc != 0) + GOTO(out, rc); + + de->inode = cpu_to_le32(id->oii_ino); + rc = ldiskfs_handle_dirty_metadata(th, NULL, bh); + + GOTO(out, rc); + +out: + brelse(bh); + mutex_unlock(&parent->i_mutex); + return rc; +} + static int osd_obj_del_entry(struct osd_thread_info *info, struct osd_device *osd, struct dentry *dird, char *name, - struct thandle *th) + handle_t *th) { struct ldiskfs_dir_entry_2 *de; struct buffer_head *bh; - struct osd_thandle *oh; struct dentry *child; struct inode *dir = dird->d_inode; int rc; ENTRY; - oh = container_of(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + LASSERT(th != NULL); + LASSERT(th->h_transaction != NULL); child = &info->oti_child_dentry; @@ -492,9 +683,9 @@ static int osd_obj_del_entry(struct osd_thread_info *info, ll_vfs_dq_init(dir); mutex_lock(&dir->i_mutex); rc = -ENOENT; - bh = osd_ldiskfs_find_entry(dir, child, &de, NULL); + bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL); if (bh) { - rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); + rc = ldiskfs_delete_entry(th, dir, de, bh); brelse(bh); } mutex_unlock(&dir->i_mutex); @@ -502,38 +693,50 @@ static int osd_obj_del_entry(struct osd_thread_info *info, RETURN(rc); } -int osd_obj_add_entry(struct osd_thread_info *info, - struct osd_device *osd, - struct dentry *dir, char *name, - const struct osd_inode_id *id, - struct thandle *th) +static int osd_obj_add_entry(struct osd_thread_info *info, + struct osd_device *osd, + struct dentry *dir, char *name, + const struct osd_inode_id *id, + handle_t *th) { - struct osd_thandle *oh; - struct dentry *child; - struct inode *inode; - int rc; + struct dentry *child; + struct inode *inode; + int rc; - ENTRY; + ENTRY; + + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)) + RETURN(0); + + LASSERT(th != NULL); + LASSERT(th->h_transaction != NULL); - oh = container_of(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + inode = info->oti_inode; + if (unlikely(inode == NULL)) { + struct ldiskfs_inode_info *lii; + OBD_ALLOC_PTR(lii); + if (lii == NULL) + RETURN(-ENOMEM); + inode = info->oti_inode = &lii->vfs_inode; + } - inode = &info->oti_inode; - inode->i_sb = osd_sb(osd); + inode->i_sb = osd_sb(osd); osd_id_to_inode(inode, id); inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */ - child = &info->oti_child_dentry; - child->d_name.hash = 0; - child->d_name.name = name; - child->d_name.len = strlen(name); - child->d_parent = dir; - child->d_inode = inode; + child = &info->oti_child_dentry; + child->d_name.hash = 0; + child->d_name.name = name; + child->d_name.len = strlen(name); + child->d_parent = dir; + child->d_inode = inode; + + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY)) + inode->i_ino++; ll_vfs_dq_init(dir->d_inode); mutex_lock(&dir->d_inode->i_mutex); - rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL); + rc = osd_ldiskfs_add_entry(info, th, child, inode, NULL); mutex_unlock(&dir->d_inode->i_mutex); RETURN(rc); @@ -546,29 +749,35 @@ int osd_obj_add_entry(struct osd_thread_info *info, * debug messages to objects in the future, and the legacy space * of FID_SEQ_OST_MDT0 will be unused in the future. **/ -static inline void osd_seq_name(char *seq_name, obd_seq seq) +static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq) { - sprintf(seq_name, (fid_seq_is_rsvd(seq) || - fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i, - fid_seq_is_idif(seq) ? 0 : seq); + snprintf(seq_name, name_size, + (fid_seq_is_rsvd(seq) || + fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i, + fid_seq_is_idif(seq) ? 0 : seq); } -static inline void osd_oid_name(char *name, const struct lu_fid *fid, obd_id id) +static inline void osd_oid_name(char *name, size_t name_size, + const struct lu_fid *fid, u64 id) { - sprintf(name, (fid_seq_is_rsvd(fid_seq(fid)) || - fid_seq_is_mdt0(fid_seq(fid)) || - fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id); + snprintf(name, name_size, + (fid_seq_is_rsvd(fid_seq(fid)) || + fid_seq_is_mdt0(fid_seq(fid)) || + fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id); } /* external locking is required */ -static int osd_seq_load_locked(struct osd_device *osd, +static int osd_seq_load_locked(struct osd_thread_info *info, + struct osd_device *osd, struct osd_obj_seq *osd_seq) { struct osd_obj_map *map = osd->od_ost_map; struct dentry *seq_dir; + struct inode *inode; + struct lu_fid *fid = &info->oti_fid3; int rc = 0; int i; - char seq_name[32]; + char dir_name[32]; ENTRY; if (osd_seq->oos_root != NULL) @@ -577,17 +786,25 @@ static int osd_seq_load_locked(struct osd_device *osd, LASSERT(map); LASSERT(map->om_root); - osd_seq_name(seq_name, osd_seq->oos_seq); + osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq); - seq_dir = simple_mkdir(map->om_root, osd->od_mnt, seq_name, 0755, 1); + seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1); if (IS_ERR(seq_dir)) GOTO(out_err, rc = PTR_ERR(seq_dir)); else if (seq_dir->d_inode == NULL) GOTO(out_put, rc = -EFAULT); - ldiskfs_set_inode_state(seq_dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI); + inode = seq_dir->d_inode; osd_seq->oos_root = seq_dir; + /* 'What the @fid is' is not imporatant, because the object + * has no OI mapping, and only is visible inside the OSD.*/ + lu_igif_build(fid, inode->i_ino, inode->i_generation); + rc = osd_ea_fid_set(info, inode, fid, + LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0); + if (rc != 0) + GOTO(out_put, rc); + LASSERT(osd_seq->oos_dirs == NULL); OBD_ALLOC(osd_seq->oos_dirs, sizeof(seq_dir) * osd_seq->oos_subdir_count); @@ -596,27 +813,38 @@ static int osd_seq_load_locked(struct osd_device *osd, for (i = 0; i < osd_seq->oos_subdir_count; i++) { struct dentry *dir; - char name[32]; - sprintf(name, "d%u", i); - dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, name, + snprintf(dir_name, sizeof(dir_name), "d%u", i); + dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name, 0700, 1); if (IS_ERR(dir)) { - rc = PTR_ERR(dir); - } else if (dir->d_inode) { - ldiskfs_set_inode_state(dir->d_inode, - LDISKFS_STATE_LUSTRE_NO_OI); - osd_seq->oos_dirs[i] = dir; - rc = 0; - } else { - LBUG(); + GOTO(out_free, rc = PTR_ERR(dir)); + } else if (dir->d_inode == NULL) { + dput(dir); + GOTO(out_free, rc = -EFAULT); } + + inode = dir->d_inode; + osd_seq->oos_dirs[i] = dir; + + /* 'What the @fid is' is not imporatant, because the object + * has no OI mapping, and only is visible inside the OSD.*/ + lu_igif_build(fid, inode->i_ino, inode->i_generation); + rc = osd_ea_fid_set(info, inode, fid, + LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0); + if (rc != 0) + GOTO(out_free, rc); } - if (rc != 0) - osd_seq_free(map, osd_seq); -out_put: if (rc != 0) { +out_free: + for (i = 0; i < osd_seq->oos_subdir_count; i++) { + if (osd_seq->oos_dirs[i] != NULL) + dput(osd_seq->oos_dirs[i]); + } + OBD_FREE(osd_seq->oos_dirs, + sizeof(seq_dir) * osd_seq->oos_subdir_count); +out_put: dput(seq_dir); osd_seq->oos_root = NULL; } @@ -624,7 +852,8 @@ out_err: RETURN(rc); } -struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq) +static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info, + struct osd_device *osd, u64 seq) { struct osd_obj_map *map; struct osd_obj_seq *osd_seq; @@ -640,7 +869,7 @@ struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq) RETURN(osd_seq); /* Serializing init process */ - down(&map->om_dir_init_sem); + mutex_lock(&map->om_dir_init_mutex); /* Check whether the seq has been added */ read_lock(&map->om_seq_list_lock); @@ -655,21 +884,21 @@ struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq) if (osd_seq == NULL) GOTO(cleanup, rc = -ENOMEM); - CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list); + INIT_LIST_HEAD(&osd_seq->oos_seq_list); osd_seq->oos_seq = seq; /* Init subdir count to be 32, but each seq can have * different subdir count */ osd_seq->oos_subdir_count = map->om_subdir_count; - rc = osd_seq_load_locked(osd, osd_seq); + rc = osd_seq_load_locked(info, osd, osd_seq); if (rc != 0) GOTO(cleanup, rc); write_lock(&map->om_seq_list_lock); - cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list); + list_add(&osd_seq->oos_seq_list, &map->om_seq_list); write_unlock(&map->om_seq_list_lock); cleanup: - up(&map->om_dir_init_sem); + mutex_unlock(&map->om_dir_init_mutex); if (rc != 0) { if (osd_seq != NULL) OBD_FREE_PTR(osd_seq); @@ -701,16 +930,16 @@ int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev, LASSERT(map); LASSERT(map->om_root); - fid_ostid_pack(fid, ostid); - osd_seq = osd_seq_load(dev, ostid->oi_seq); + fid_to_ostid(fid, ostid); + osd_seq = osd_seq_load(info, dev, ostid_seq(ostid)); if (IS_ERR(osd_seq)) RETURN(PTR_ERR(osd_seq)); - dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1); + dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1); d_seq = osd_seq->oos_dirs[dirn]; LASSERT(d_seq); - osd_oid_name(name, fid, ostid->oi_id); + osd_oid_name(name, sizeof(name), fid, ostid_id(ostid)); child = &info->oti_child_dentry; child->d_parent = d_seq; @@ -721,7 +950,7 @@ int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev, dir = d_seq->d_inode; mutex_lock(&dir->i_mutex); - bh = osd_ldiskfs_find_entry(dir, child, &de, NULL); + bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL); mutex_unlock(&dir->i_mutex); if (bh == NULL) @@ -731,8 +960,11 @@ int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev, brelse(bh); inode = osd_iget(info, dev, id); - if (IS_ERR(inode)) - RETURN(PTR_ERR(inode)); + if (IS_ERR(inode)) { + int rc = PTR_ERR(inode); + + RETURN(rc == -ENOENT ? -ESTALE : rc); + } iput(inode); RETURN(0); @@ -742,38 +974,50 @@ int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd, const struct lu_fid *fid, const struct osd_inode_id *id, - struct thandle *th) + handle_t *th) { struct osd_obj_map *map; struct osd_obj_seq *osd_seq; struct dentry *d; struct ost_id *ostid = &info->oti_ostid; + u64 oid; int dirn, rc = 0; char name[32]; - ENTRY; + ENTRY; - map = osd->od_ost_map; - LASSERT(map); + map = osd->od_ost_map; + LASSERT(map); /* map fid to seq:objid */ - fid_ostid_pack(fid, ostid); + fid_to_ostid(fid, ostid); - osd_seq = osd_seq_load(osd, ostid->oi_seq); + oid = ostid_id(ostid); + osd_seq = osd_seq_load(info, osd, ostid_seq(ostid)); if (IS_ERR(osd_seq)) RETURN(PTR_ERR(osd_seq)); - dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1); + dirn = oid & (osd_seq->oos_subdir_count - 1); d = osd_seq->oos_dirs[dirn]; - LASSERT(d); + LASSERT(d); - osd_oid_name(name, fid, ostid->oi_id); + osd_oid_name(name, sizeof(name), fid, oid); + +again: rc = osd_obj_add_entry(info, osd, d, name, id, th); + if (rc == -EEXIST) { + rc = osd_obj_update_entry(info, osd, d, name, fid, id, th); + if (unlikely(rc == -ENOENT)) + goto again; - RETURN(rc); + if (unlikely(rc == 1)) + RETURN(0); + } + + RETURN(rc); } int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd, - const struct lu_fid *fid, struct thandle *th) + const struct lu_fid *fid, handle_t *th) { struct osd_obj_map *map; struct osd_obj_seq *osd_seq; @@ -787,49 +1031,230 @@ int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd, LASSERT(map); /* map fid to seq:objid */ - fid_ostid_pack(fid, ostid); + fid_to_ostid(fid, ostid); - osd_seq = osd_seq_load(osd, ostid->oi_seq); + osd_seq = osd_seq_load(info, osd, ostid_seq(ostid)); if (IS_ERR(osd_seq)) GOTO(cleanup, rc = PTR_ERR(osd_seq)); - dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1); + dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1); d = osd_seq->oos_dirs[dirn]; LASSERT(d); - osd_oid_name(name, fid, ostid->oi_id); + osd_oid_name(name, sizeof(name), fid, ostid_id(ostid)); rc = osd_obj_del_entry(info, osd, d, name, th); cleanup: RETURN(rc); } -int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd, - const struct lu_fid *fid, - const struct osd_inode_id *id, - struct thandle *th) +int osd_obj_map_update(struct osd_thread_info *info, + struct osd_device *osd, + const struct lu_fid *fid, + const struct osd_inode_id *id, + handle_t *th) { - struct osd_obj_map *map = osd->od_ost_map; - struct dentry *root = osd_sb(osd)->s_root; - char *name; - int rc = 0; + struct osd_obj_seq *osd_seq; + struct dentry *d; + struct ost_id *ostid = &info->oti_ostid; + int dirn, rc = 0; + char name[32]; + ENTRY; + + fid_to_ostid(fid, ostid); + osd_seq = osd_seq_load(info, osd, ostid_seq(ostid)); + if (IS_ERR(osd_seq)) + RETURN(PTR_ERR(osd_seq)); + + dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1); + d = osd_seq->oos_dirs[dirn]; + LASSERT(d); + + osd_oid_name(name, sizeof(name), fid, ostid_id(ostid)); + rc = osd_obj_update_entry(info, osd, d, name, fid, id, th); + + RETURN(rc); +} + +int osd_obj_map_recover(struct osd_thread_info *info, + struct osd_device *osd, + struct inode *src_parent, + struct dentry *src_child, + const struct lu_fid *fid) +{ + struct osd_obj_seq *osd_seq; + struct dentry *tgt_parent; + struct dentry *tgt_child = &info->oti_child_dentry; + struct inode *dir; + struct inode *inode = src_child->d_inode; + struct ost_id *ostid = &info->oti_ostid; + handle_t *jh; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + char name[32]; + int dirn; + int rc = 0; ENTRY; if (fid_is_last_id(fid)) { - struct osd_obj_seq *osd_seq; + osd_seq = osd_seq_load(info, osd, fid_seq(fid)); + if (IS_ERR(osd_seq)) + RETURN(PTR_ERR(osd_seq)); - /* on creation of LAST_ID we create O/ hierarchy */ - LASSERT(map); - osd_seq = osd_seq_load(osd, fid_seq(fid)); + tgt_parent = osd_seq->oos_root; + tgt_child->d_name.name = "LAST_ID"; + tgt_child->d_name.len = strlen("LAST_ID"); + } else { + fid_to_ostid(fid, ostid); + osd_seq = osd_seq_load(info, osd, ostid_seq(ostid)); if (IS_ERR(osd_seq)) RETURN(PTR_ERR(osd_seq)); - rc = osd_obj_add_entry(info, osd, osd_seq->oos_root, - "LAST_ID", id, th); + + dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1); + tgt_parent = osd_seq->oos_dirs[dirn]; + osd_oid_name(name, sizeof(name), fid, ostid_id(ostid)); + tgt_child->d_name.name = name; + tgt_child->d_name.len = strlen(name); + } + LASSERT(tgt_parent != NULL); + + dir = tgt_parent->d_inode; + tgt_child->d_name.hash = 0; + tgt_child->d_parent = tgt_parent; + tgt_child->d_inode = inode; + + /* The non-initialized src_child may be destroyed. */ + jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, + osd_dto_credits_noquota[DTO_INDEX_DELETE] + + osd_dto_credits_noquota[DTO_INDEX_INSERT] + + osd_dto_credits_noquota[DTO_OBJECT_DELETE]); + if (IS_ERR(jh)) + RETURN(PTR_ERR(jh)); + + ll_vfs_dq_init(src_parent); + ll_vfs_dq_init(dir); + + mutex_lock(&src_parent->i_mutex); + mutex_lock(&dir->i_mutex); + bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL); + if (bh != NULL) { + /* XXX: If some other object occupied the same slot. And If such + * inode is zero-sized and with SUID+SGID, then means it is + * a new created one. Maybe we can remove it and insert the + * original one back to the /O//d. But there are + * something to be considered: + * + * 1) The OST-object under /lost+found has crashed LMA. + * So it should not conflict with the current one. + * + * 2) There are race conditions that: someone may just want + * to modify the current one. Even if the OI scrub takes + * the object lock when remove the current one, it still + * cause the modification to be lost becasue the target + * has been removed when the RPC service thread waiting + * for the lock. + * + * So keep it there before we have suitable solution. */ + brelse(bh); + mutex_unlock(&dir->i_mutex); + mutex_unlock(&src_parent->i_mutex); + ldiskfs_journal_stop(jh); + + rc = -EEXIST; + /* If the src object has never been modified, then remove it. */ + if (inode->i_size == 0 && inode->i_mode & S_ISUID && + inode->i_mode & S_ISGID) { + rc = ll_vfs_unlink(src_parent, src_child); + if (unlikely(rc == -ENOENT)) + rc = 0; + } + RETURN(rc); + } + + bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de, + NULL, NULL); + if (unlikely(bh == NULL)) + GOTO(unlock, rc = -ENOENT); + + rc = ldiskfs_delete_entry(jh, src_parent, de, bh); + brelse(bh); + if (rc != 0) + GOTO(unlock, rc); + + rc = osd_ldiskfs_add_entry(info, jh, tgt_child, inode, NULL); + + GOTO(unlock, rc); + +unlock: + mutex_unlock(&dir->i_mutex); + mutex_unlock(&src_parent->i_mutex); + ldiskfs_journal_stop(jh); + return rc; +} + +static struct dentry * +osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd, + const struct lu_fid *fid, char **name) +{ + struct dentry *root = ERR_PTR(-ENOENT); + + if (fid_is_last_id(fid)) { + struct osd_obj_seq *osd_seq; + + /* on creation of LAST_ID we create O/ hierarchy */ + osd_seq = osd_seq_load(info, osd, fid_seq(fid)); + if (IS_ERR(osd_seq)) + RETURN((struct dentry *)osd_seq); + + *name = "LAST_ID"; + root = osd_seq->oos_root; } else { - name = osd_lf_fid2name(fid); - if (name == NULL) + *name = osd_lf_fid2name(fid); + if (*name == NULL) CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid)); - else if (name[0]) - rc = osd_obj_add_entry(info, osd, root, name, id, th); + else if ((*name)[0]) + root = osd_sb(osd)->s_root; + } + + return root; +} + +int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd, + const struct lu_fid *fid, const struct osd_inode_id *id, + handle_t *th) +{ + struct dentry *root; + char *name = NULL; + int rc; + ENTRY; + + root = osd_object_spec_find(info, osd, fid, &name); + if (!IS_ERR(root)) { + rc = osd_obj_update_entry(info, osd, root, name, fid, id, th); + } else { + rc = PTR_ERR(root); + if (rc == -ENOENT) + rc = 1; + } + + RETURN(rc); +} + +int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd, + const struct lu_fid *fid, const struct osd_inode_id *id, + handle_t *th) +{ + struct dentry *root; + char *name = NULL; + int rc; + ENTRY; + + root = osd_object_spec_find(info, osd, fid, &name); + if (!IS_ERR(root)) { + rc = osd_obj_add_entry(info, osd, root, name, id, th); + } else { + rc = PTR_ERR(root); + if (rc == -ENOENT) + rc = 0; } RETURN(rc); @@ -841,14 +1266,14 @@ int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd, struct dentry *root; struct dentry *dentry; struct inode *inode; - char *name; + char *name = NULL; int rc = -ENOENT; ENTRY; if (fid_is_last_id(fid)) { struct osd_obj_seq *osd_seq; - osd_seq = osd_seq_load(osd, fid_seq(fid)); + osd_seq = osd_seq_load(info, osd, fid_seq(fid)); if (IS_ERR(osd_seq)) RETURN(PTR_ERR(osd_seq)); root = osd_seq->oos_root;