X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=e1bc36f0d994723066f43638ac25de0f809ea515;hp=8278c6b2d6d866f65a7dcf8231186474a24b8f15;hb=9857b9d2bcf39c5bced8f4f7eedbe3170331a9ae;hpb=28fc8fcc8bfe092c9a1a8c192ab6fe22d92820e7 diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 8278c6b..e1bc36f 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -53,8 +53,6 @@ #include /* XATTR_{REPLACE,CREATE} */ #include -/* simple_mkdir() */ -#include /* * struct OBD_{ALLOC,FREE}*() @@ -63,12 +61,9 @@ #include /* struct ptlrpc_thread */ #include - -/* fid_is_local() */ #include #include "osd_internal.h" -#include "osd_igif.h" /* llo_* api support */ #include @@ -173,14 +168,20 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, } } -static int osd_get_lma(struct osd_thread_info *info, struct inode *inode, - struct dentry *dentry, struct lustre_mdt_attrs *lma) +static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry, + const char *name, void *buf, int len) +{ + dentry->d_inode = inode; + return inode->i_op->getxattr(dentry, name, buf, len); +} + +int osd_get_lma(struct osd_thread_info *info, struct inode *inode, + struct dentry *dentry, struct lustre_mdt_attrs *lma) { int rc; - dentry->d_inode = inode; - rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma, - sizeof(*lma)); + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma, + sizeof(*lma)); if (rc == -ERANGE) { /* try with old lma size */ rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, @@ -255,8 +256,9 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, return inode; } -struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, - struct osd_inode_id *id, struct lu_fid *fid) +static struct inode * +osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, + struct osd_inode_id *id, struct lu_fid *fid) { struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; struct inode *inode; @@ -270,7 +272,10 @@ struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, if (rc == 0) { *fid = lma->lma_self_fid; } else if (rc == -ENODATA) { - LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation); + if (unlikely(inode == osd_sb(dev)->s_root->d_inode)) + lu_local_obj_fid(fid, OSD_FS_ROOT_OID); + else + lu_igif_build(fid, inode->i_ino, inode->i_generation); } else { iput(inode); inode = ERR_PTR(rc); @@ -363,13 +368,13 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, * unexpected reason, we should be able to detect it later by calling * do_create->osd_oi_insert() */ - if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0) + if (conf != NULL && conf->loc_flags & LOC_F_NEW) GOTO(out, result = 0); /* Search order: 3. OI files. */ - result = osd_oi_lookup(info, dev, fid, id); + result = osd_oi_lookup(info, dev, fid, id, true); if (result == -ENOENT) { - if (!fid_is_norm(fid) || + if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); @@ -457,17 +462,17 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, LINVRNT(osd_invariant(obj)); + if (fid_is_otable_it(&l->lo_header->loh_fid)) { + obj->oo_dt.do_ops = &osd_obj_otable_it_ops; + l->lo_header->loh_attr |= LOHA_EXISTS; + return 0; + } + result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); obj->oo_dt.do_body_ops = &osd_body_ops_new; - if (result == 0) { - if (obj->oo_inode != NULL) { - osd_object_init0(obj); - } else if (fid_is_otable_it(&l->lo_header->loh_fid)) { - obj->oo_dt.do_ops = &osd_obj_otable_it_ops; - /* LFSCK iterator object is special without inode */ - l->lo_header->loh_attr |= LOHA_EXISTS; - } - } + if (result == 0 && obj->oo_inode != NULL) + osd_object_init0(obj); + LINVRNT(osd_invariant(obj)); return result; } @@ -616,12 +621,8 @@ static int osd_param_is_not_sane(const struct osd_device *dev, /* * Concurrency: shouldn't matter. */ -#ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD static void osd_trans_commit_cb(struct super_block *sb, - struct journal_callback *jcb, int error) -#else -static void osd_trans_commit_cb(struct journal_callback *jcb, int error) -#endif + struct ldiskfs_journal_cb_entry *jcb, int error) { struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); struct thandle *th = &oh->ot_super; @@ -818,7 +819,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) * notice we don't do this in osd_trans_start() * as underlying transaction can change during truncate */ - osd_journal_callback_set(hdl, osd_trans_commit_cb, + ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, &oh->ot_jcb); LASSERT(oti->oti_txns == 1); @@ -1035,19 +1036,6 @@ static void osd_conf_get(const struct lu_env *env, } -/** - * Helper function to get and fill the buffer with input values. - */ -static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len) -{ - struct lu_buf *buf; - - buf = &osd_oti_get(env)->oti_buf; - buf->lb_buf = area; - buf->lb_len = len; - return buf; -} - /* * Concurrency: shouldn't matter. */ @@ -1085,14 +1073,31 @@ static int osd_commit_async(const struct lu_env *env, static int osd_ro(const struct lu_env *env, struct dt_device *d) { - struct super_block *sb = osd_sb(osd_dt_dev(d)); - int rc; - ENTRY; + struct super_block *sb = osd_sb(osd_dt_dev(d)); + struct block_device *dev = sb->s_bdev; +#ifdef HAVE_DEV_SET_RDONLY + struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev; + int rc = 0; +#else + int rc = -EOPNOTSUPP; +#endif + ENTRY; +#ifdef HAVE_DEV_SET_RDONLY CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname); - rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev); - RETURN(rc); + if (jdev && (jdev != dev)) { + CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n", + (long)jdev); + dev_set_rdonly(jdev); + } + CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev); + dev_set_rdonly(dev); +#else + CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", + osd_dt_dev(d)->od_svname, (long)dev, rc); +#endif + RETURN(rc); } /* @@ -1369,7 +1374,8 @@ static void osd_inode_getattr(const struct lu_env *env, { attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | - LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE; + LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | + LA_TYPE; attr->la_atime = LTIME_S(inode->i_atime); attr->la_mtime = LTIME_S(inode->i_mtime); @@ -1393,7 +1399,7 @@ static int osd_attr_get(const struct lu_env *env, { struct osd_object *obj = osd_dt_obj(dt); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LINVRNT(osd_invariant(obj)); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) @@ -1552,8 +1558,6 @@ static int osd_inode_setattr(const struct lu_env *env, bits = attr->la_valid; - LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */ - if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); if (bits & LA_CTIME) @@ -1628,7 +1632,7 @@ static int osd_attr_set(const struct lu_env *env, int rc; LASSERT(handle != NULL); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_invariant(obj)); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) @@ -1636,6 +1640,31 @@ 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; + + fid_cpu_to_be(fid1, fid0); + memset(id, 1, sizeof(*id)); + bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev), + fid0)->oi_dir.od_container; + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); + + oh = container_of0(handle, struct osd_thandle, ot_super); + rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1, + (const struct iam_rec *)id, ipd); + osd_ipd_put(env, bag, ipd); + return(rc > 0 ? 0 : rc); + } + inode = obj->oo_inode; ll_vfs_dq_init(inode); @@ -1648,7 +1677,7 @@ static int osd_attr_set(const struct lu_env *env, spin_unlock(&obj->oo_guard); if (!rc) - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); return rc; } @@ -1695,7 +1724,10 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, * NB: don't need any lock because no contention at this * early stage */ inode->i_flags |= S_NOCMTIME; - inode->i_state |= I_LUSTRE_NOSCRUB; + + /* For new created object, it must be consistent, + * and it is unnecessary to scrub against it. */ + ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); obj->oo_inode = inode; result = 0; } else { @@ -1897,7 +1929,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, * enabled on ldiskfs (lquota takes care of it). */ LASSERTF(result == 0, "%d", result); - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); } attr->la_valid = valid; @@ -1915,6 +1947,11 @@ static int __osd_object_create(struct osd_thread_info *info, struct thandle *th) { int result; + __u32 umask; + + /* we drop umask so that permissions we pass are not affected */ + umask = current->fs->umask; + current->fs->umask = 0; result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof, th); @@ -1926,7 +1963,10 @@ static int __osd_object_create(struct osd_thread_info *info, unlock_new_inode(obj->oo_inode); } - return result; + /* restore previous umask value */ + current->fs->umask = umask; + + return result; } /** @@ -1953,19 +1993,13 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, struct seq_server_site *ss = osd_seq_site(osd); int rc; - if (fid_is_igif(fid)) { - range->lsr_flags = LU_SEQ_RANGE_MDT; - range->lsr_index = 0; - return 0; - } - if (fid_is_idif(fid)) { range->lsr_flags = LU_SEQ_RANGE_OST; range->lsr_index = fid_idif_ost_idx(fid); return 0; } - if (!fid_is_norm(fid)) { + if (!fid_seq_in_fldb(fid_seq(fid))) { range->lsr_flags = LU_SEQ_RANGE_MDT; if (ss != NULL) /* FIXME: If ss is NULL, it suppose not get lsr_index @@ -1979,12 +2013,14 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range); if (rc != 0) { CERROR("%s can not find "DFID": rc = %d\n", - osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc); + osd_name(osd), PFID(fid), rc); } return rc; } - +/* + * 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, @@ -2004,16 +2040,13 @@ static int osd_declare_object_create(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_CREATE, osd_dto_credits_noquota[DTO_OBJECT_CREATE]); - /* XXX: So far, only normal fid needs be inserted into the oi, - * things could be changed later. Revise following code then. */ - if (fid_is_norm(lu_object_fid(&dt->do_lu)) && - !fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev), - lu_object_fid(&dt->do_lu))) { + if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev), + lu_object_fid(&dt->do_lu))) /* Reuse idle OI block may cause additional one OI block * to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); - } + /* If this is directory, then we expect . and .. to be inserted as * well. The one directory block always needs to be created for the * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap, @@ -2060,7 +2093,7 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt, ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt)); + LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); @@ -2076,7 +2109,9 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt, if (result == 0) result = __osd_oi_insert(env, obj, fid, th); - LASSERT(ergo(result == 0, dt_object_exists(dt))); + LASSERT(ergo(result == 0, + dt_object_exists(dt) && !dt_object_remote(dt))); + LASSERT(osd_invariant(obj)); RETURN(result); } @@ -2102,13 +2137,10 @@ static int osd_declare_object_destroy(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_DELETE, osd_dto_credits_noquota[DTO_OBJECT_DELETE]); - /* XXX: So far, only normal fid needs to be inserted into the OI, - * so only normal fid needs to be removed from the OI also. - * Recycle idle OI leaf may cause additional three OI blocks + /* Recycle idle OI leaf may cause additional three OI blocks * to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_DESTROY, - fid_is_norm(lu_object_fid(&dt->do_lu)) ? - osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0); + osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); /* one less inode */ rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh, @@ -2145,12 +2177,19 @@ static int osd_object_destroy(const struct lu_env *env, * lock contention. So it will not affect unlink performance. */ mutex_lock(&inode->i_mutex); if (S_ISDIR(inode->i_mode)) { - LASSERT(osd_inode_unlinked(inode) || - inode->i_nlink == 1); + LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1); + /* it will check/delete the agent inode for every dir + * destory, how to optimize it? unlink performance + * impaction XXX */ + result = osd_delete_from_agent(env, osd, obj, oh); + if (result != 0 && result != -ENOENT) { + CERROR("%s: delete agent inode "DFID": rc = %d\n", + osd_name(osd), PFID(fid), result); + } spin_lock(&obj->oo_guard); clear_nlink(inode); spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); } osd_trans_exec_op(env, th, OSD_OT_DESTROY); @@ -2167,33 +2206,15 @@ static int osd_object_destroy(const struct lu_env *env, RETURN(0); } -/** - * Helper function for osd_xattr_set() - */ -static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_buf *buf, const char *name, int fl) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_child_dentry; - int fs_flags = 0; - int rc; - - LASSERT(dt_object_exists(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL); - - if (fl & LU_XATTR_REPLACE) - fs_flags |= XATTR_REPLACE; - - if (fl & LU_XATTR_CREATE) - fs_flags |= XATTR_CREATE; +static inline int __osd_xattr_set(struct osd_thread_info *info, + struct inode *inode, const char *name, + const void *buf, int buflen, int fl) +{ + struct dentry *dentry = &info->oti_child_dentry; ll_vfs_dq_init(inode); - dentry->d_inode = inode; - rc = inode->i_op->setxattr(dentry, name, buf->lb_buf, - buf->lb_len, fs_flags); - return rc; + dentry->d_inode = inode; + return inode->i_op->setxattr(dentry, name, buf, buflen, fl); } /** @@ -2206,18 +2227,27 @@ static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt, * * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here */ -static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_fid *fid) +int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, + const struct lu_fid *fid) { - struct osd_thread_info *info = osd_oti_get(env); - struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs; + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + int rc; - lustre_lma_init(mdt_attrs, fid); - lustre_lma_swab(mdt_attrs); - return __osd_xattr_set(env, dt, - osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs), - XATTR_NAME_LMA, LU_XATTR_CREATE); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) + return 0; + + if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid)) + return 0; + lustre_lma_init(lma, fid); + lustre_lma_swab(lma); + + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma), + XATTR_CREATE); + /* Someone may created the EA by race. */ + if (unlikely(rc == -EEXIST)) + rc = 0; + return rc; } /** @@ -2229,18 +2259,20 @@ static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt, * its inmemory API. */ void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) + const struct dt_rec *fid) { - param->edp_magic = LDISKFS_LUFID_MAGIC; - param->edp_len = sizeof(struct lu_fid) + 1; + if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) { + param->edp_magic = 0; + return; + } - fid_cpu_to_be((struct lu_fid *)param->edp_data, - (struct lu_fid *)fid); + param->edp_magic = LDISKFS_LUFID_MAGIC; + param->edp_len = sizeof(struct lu_fid) + 1; + fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid); } /** - * Try to read the fid from inode ea into dt_rec, if return value - * i.e. rc is +ve, then we got fid, otherwise we will have to form igif + * Try to read the fid from inode ea into dt_rec. * * \param fid object fid. * @@ -2263,6 +2295,95 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, RETURN(0); } +static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, + struct inode *dir, + struct inode *parent_dir, + const struct dt_rec *dot_fid, + const struct dt_rec *dot_dot_fid, + struct osd_thandle *oth) +{ + struct ldiskfs_dentry_param *dot_ldp; + struct ldiskfs_dentry_param *dot_dot_ldp; + + dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; + osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); + + dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; + dot_ldp->edp_magic = 0; + return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, + dir, dot_ldp, dot_dot_ldp); +} + +/** + * Create an local inode for remote entry + */ +static struct inode *osd_create_remote_inode(const struct lu_env *env, + struct osd_device *osd, + struct osd_object *pobj, + const struct lu_fid *fid, + struct thandle *th) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct inode *local; + struct osd_thandle *oh; + int rc; + ENTRY; + + LASSERT(th); + oh = container_of(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle->h_transaction != NULL); + + /* FIXME: Insert index api needs to know the mode of + * the remote object. Just use S_IFDIR for now */ + local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR); + if (IS_ERR(local)) { + CERROR("%s: create local error %d\n", osd_name(osd), + (int)PTR_ERR(local)); + RETURN(local); + } + + rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode, + (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), + (const struct dt_rec *)fid, oh); + if (rc != 0) { + CERROR("%s: "DFID" add dot dotdot error: rc = %d\n", + osd_name(osd), PFID(fid), rc); + RETURN(ERR_PTR(rc)); + } + + RETURN(local); +} + +/** + * Delete local inode for remote entry + */ +static int osd_delete_remote_inode(const struct lu_env *env, + struct osd_device *osd, + const struct lu_fid *fid, + __u32 ino, struct osd_thandle *oh) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_inode_id *id = &oti->oti_id; + struct inode *inode; + ENTRY; + + id->oii_ino = le32_to_cpu(ino); + id->oii_gen = OSD_OII_NOGEN; + inode = osd_iget(oti, osd, id); + if (IS_ERR(inode)) { + CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd), + PFID(fid), id->oii_ino, id->oii_gen); + RETURN(PTR_ERR(inode)); + } + + clear_nlink(inode); + mark_inode_dirty(inode); + CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n", + osd_name(osd), PFID(fid), inode->i_ino); + iput(inode); + RETURN(0); +} + /** * OSD layer object create function for interoperability mode (b11826). * This is mostly similar to osd_object_create(). Only difference being, fid is @@ -2285,7 +2406,7 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, ENTRY; LASSERT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt)); + LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); @@ -2298,16 +2419,16 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); result = __osd_object_create(info, obj, attr, hint, dof, th); - /* objects under osd root shld have igif fid, so dont add fid EA */ - /* For ost object, the fid will be stored during first write */ - if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL && - !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)) - result = osd_ea_fid_set(env, dt, fid); + if ((result == 0) && + (fid_is_last_id(fid) || + !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid))) + result = osd_ea_fid_set(info, obj->oo_inode, fid); if (result == 0) result = __osd_oi_insert(env, obj, fid, th); - LASSERT(ergo(result == 0, dt_object_exists(dt))); + LASSERT(ergo(result == 0, + dt_object_exists(dt) && !dt_object_remote(dt))); LINVRNT(osd_invariant(obj)); RETURN(result); } @@ -2340,7 +2461,7 @@ static int osd_object_ref_add(const struct lu_env *env, struct inode *inode = obj->oo_inode; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); @@ -2369,7 +2490,7 @@ static int osd_object_ref_add(const struct lu_env *env, } LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); return 0; @@ -2381,7 +2502,7 @@ static int osd_declare_object_ref_del(const struct lu_env *env, { struct osd_thandle *oh; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -2403,7 +2524,7 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, struct inode *inode = obj->oo_inode; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); @@ -2418,7 +2539,7 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0) set_nlink(inode, 1); spin_unlock(&obj->oo_guard); - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); return 0; @@ -2459,14 +2580,13 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, return sizeof(dt_obj_version_t); } - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) return -EACCES; - dentry->d_inode = inode; - return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len); + return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -2505,7 +2625,7 @@ static void osd_object_version_set(const struct lu_env *env, LDISKFS_I(inode)->i_fs_version = *new_version; /** Version is set after all inode operations are finished, * so we should mark it dirty here */ - inode->i_sb->s_op->dirty_inode(inode); + ll_dirty_inode(inode, I_DIRTY_DATASYNC); } /* @@ -2515,6 +2635,12 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int fl, struct thandle *handle, struct lustre_capa *capa) { + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + int fs_flags = 0; + ENTRY; + LASSERT(handle != NULL); /* version set is not real XATTR */ @@ -2530,7 +2656,14 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, return -EACCES; osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); - return __osd_xattr_set(env, dt, buf, name, fl); + if (fl & LU_XATTR_REPLACE) + fs_flags |= XATTR_REPLACE; + + if (fl & LU_XATTR_CREATE) + fs_flags |= XATTR_CREATE; + + return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, + fs_flags); } /* @@ -2544,7 +2677,7 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, struct osd_thread_info *info = osd_oti_get(env); struct dentry *dentry = &info->oti_obj_dentry; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL); LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj)); @@ -2561,7 +2694,7 @@ static int osd_declare_xattr_del(const struct lu_env *env, { struct osd_thandle *oh; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -2586,9 +2719,8 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, struct dentry *dentry = &info->oti_obj_dentry; int rc; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL); - LASSERT(osd_write_locked(env, obj)); LASSERT(handle != NULL); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) @@ -2621,7 +2753,7 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, if (!dev->od_fl_capa) RETURN(ERR_PTR(-ENOENT)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LINVRNT(osd_invariant(obj)); /* renewal sanity check */ @@ -2704,9 +2836,13 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) file->f_dentry = dentry; file->f_mapping = inode->i_mapping; file->f_op = inode->i_fop; +#ifndef HAVE_FILE_FSYNC_4ARGS mutex_lock(&inode->i_mutex); - rc = file->f_op->fsync(file, dentry, 0); +#endif + rc = do_fsync(file, 0); +#ifndef HAVE_FILE_FSYNC_4ARGS mutex_unlock(&inode->i_mutex); +#endif RETURN(rc); } @@ -2786,14 +2922,13 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, struct osd_object *obj = osd_dt_obj(dt); LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); if (osd_object_is_root(obj)) { dt->do_index_ops = &osd_index_ea_ops; result = 0; } else if (feat == &dt_directory_features) { dt->do_index_ops = &osd_index_ea_ops; - if (S_ISDIR(obj->oo_inode->i_mode)) + if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode)) result = 0; else result = -ENOTDIR; @@ -2801,7 +2936,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } else if (unlikely(feat == &dt_otable_features)) { dt->do_index_ops = &osd_otable_ops; return 0; - } else if (feat == &dt_acct_features) { + } else if (unlikely(feat == &dt_acct_features)) { dt->do_index_ops = &osd_acct_index_ops; result = 0; skip_iam = 1; @@ -2846,7 +2981,8 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } LINVRNT(osd_invariant(obj)); - if (is_quota_glb_feat(feat)) + if (result == 0 && is_quota_glb_feat(feat) && + fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB) result = osd_quota_migration(env, dt, feat); return result; @@ -2973,7 +3109,7 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); LASSERT(handle != NULL); @@ -3012,7 +3148,7 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -3039,7 +3175,29 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, rec = (struct osd_fid_pack *) (de->name + de->name_len + 1); rc = osd_fid_unpack((struct lu_fid *)fid, rec); } - RETURN(rc); + return rc; +} + +static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, + struct lu_fid *fid) +{ + struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; + + /* Those FID seqs, which are not in FLDB, must be local seq */ + if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL)) + RETURN(0); + + rc = osd_fld_lookup(env, osd, fid, range); + if (rc != 0) { + CERROR("%s: Can not lookup fld for "DFID"\n", + osd_name(osd), PFID(fid)); + RETURN(rc); + } + + RETURN(ss->ss_node_id != range->lsr_index); } /** @@ -3061,15 +3219,16 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, struct inode *dir = obj->oo_inode; struct dentry *dentry; struct osd_thandle *oh; - struct ldiskfs_dir_entry_2 *de; + struct ldiskfs_dir_entry_2 *de = NULL; struct buffer_head *bh; struct htree_lock *hlock = NULL; - int rc; - + struct lu_fid *fid = &osd_oti_get(env)->oti_fid; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + int rc; ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3106,6 +3265,30 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, 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. So do not try to lookup/delete + * remote inode for .. */ + if (strcmp((char *)key, dotdot) == 0) + GOTO(out, rc = 0); + + LASSERT(de != NULL); + rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid); + if (rc == 0 && osd_remote_fid(env, osd, fid)) { + __u32 ino = le32_to_cpu(de->inode); + + rc = osd_delete_remote_inode(env, osd, fid, ino, oh); + if (rc != 0) + CERROR("%s: del local inode "DFID": rc = %d\n", + osd_name(osd), PFID(fid), rc); + } else { + if (rc == -ENODATA) + rc = 0; + } +out: + LASSERT(osd_invariant(obj)); RETURN(rc); } @@ -3136,7 +3319,7 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, ENTRY; LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) @@ -3188,7 +3371,6 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, { struct osd_thandle *oh; - LASSERT(dt_object_exists(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -3227,7 +3409,7 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, ENTRY; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); LASSERT(th != NULL); @@ -3285,27 +3467,20 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, oth = container_of(th, struct osd_thandle, ot_super); LASSERT(oth->ot_handle != NULL); LASSERT(oth->ot_handle->h_transaction != NULL); - - child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); - - /* XXX: remove fid_is_igif() check here. - * IGIF check is just to handle insertion of .. when it is 'ROOT', - * it is IGIF now but needs FID in dir entry as well for readdir - * to work. - * LU-838 should fix that and remove fid_is_igif() check */ - if (fid_is_igif((struct lu_fid *)fid) || - fid_is_norm((struct lu_fid *)fid)) { - ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; - osd_get_ldiskfs_dirent_param(ldp, fid); - child->d_fsdata = (void *)ldp; - } else { - child->d_fsdata = NULL; - } LASSERT(pobj->oo_inode); + + ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; + if (unlikely(pobj->oo_inode == + osd_sb(osd_obj2dev(pobj))->s_root->d_inode)) + ldp->edp_magic = 0; + else + osd_get_ldiskfs_dirent_param(ldp, fid); + child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); + child->d_fsdata = (void *)ldp; ll_vfs_dq_init(pobj->oo_inode); - rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); + rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); - RETURN(rc); + RETURN(rc); } /** @@ -3328,8 +3503,6 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, struct thandle *th) { struct inode *inode = dir->oo_inode; - struct ldiskfs_dentry_param *dot_ldp; - struct ldiskfs_dentry_param *dot_dot_ldp; struct osd_thandle *oth; int result = 0; @@ -3345,32 +3518,23 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, dir->oo_compat_dot_created = 1; result = 0; } - } else if(strcmp(name, dotdot) == 0) { - dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; - dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; - - if (!dir->oo_compat_dot_created) - return -EINVAL; - if (!fid_is_igif((struct lu_fid *)dot_fid)) { - osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid); - osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); - } else { - dot_ldp = NULL; - dot_dot_ldp = NULL; - } - /* in case of rename, dotdot is already created */ - if (dir->oo_compat_dotdot_created) { - return __osd_ea_add_rec(info, dir, parent_dir, name, - dot_dot_fid, NULL, th); - } + } else if (strcmp(name, dotdot) == 0) { + if (!dir->oo_compat_dot_created) + return -EINVAL; + /* in case of rename, dotdot is already created */ + if (dir->oo_compat_dotdot_created) { + return __osd_ea_add_rec(info, dir, parent_dir, name, + dot_dot_fid, NULL, th); + } - result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, - inode, dot_ldp, dot_dot_ldp); - if (result == 0) - dir->oo_compat_dotdot_created = 1; - } + result = osd_add_dot_dotdot_internal(info, dir->oo_inode, + parent_dir, dot_fid, + dot_dot_fid, oth); + if (result == 0) + dir->oo_compat_dotdot_created = 1; + } - return result; + return result; } @@ -3407,8 +3571,18 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, down_write(&pobj->oo_ext_idx_sem); } - rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, - hlock, th); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { + struct lu_fid *tfid = &info->oti_fid; + + *tfid = *(const struct lu_fid *)fid; + tfid->f_ver = ~0; + rc = __osd_ea_add_rec(info, pobj, cinode, name, + (const struct dt_rec *)tfid, + hlock, th); + } else { + rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, + hlock, th); + } } if (hlock != NULL) ldiskfs_htree_unlock(hlock); @@ -3433,7 +3607,7 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, RETURN_EXIT; again: - rc = osd_oi_lookup(oti, dev, fid, id); + rc = osd_oi_lookup(oti, dev, fid, id, true); if (rc != 0 && rc != -ENOENT) RETURN_EXIT; @@ -3467,6 +3641,31 @@ again: EXIT; } +static int osd_fail_fid_lookup(struct osd_thread_info *oti, + struct osd_device *dev, + struct osd_idmap_cache *oic, + struct lu_fid *fid, __u32 ino) +{ + struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs; + struct inode *inode; + int rc; + + osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); + inode = osd_iget(oti, dev, &oic->oic_lid); + if (IS_ERR(inode)) { + fid_zero(&oic->oic_fid); + return PTR_ERR(inode); + } + + rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma); + iput(inode); + if (rc != 0) + fid_zero(&oic->oic_fid); + else + *fid = oic->oic_fid = lma->lma_self_fid; + return rc; +} + /** * Calls ->lookup() to find dentry. From dentry get inode and * read inode's ea to get fid. This is required for interoperability @@ -3486,6 +3685,7 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, struct htree_lock *hlock = NULL; int ino; int rc; + ENTRY; LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL); @@ -3503,28 +3703,37 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock); if (bh) { struct osd_thread_info *oti = osd_oti_get(env); + struct osd_inode_id *id = &oti->oti_id; struct osd_idmap_cache *oic = &oti->oti_cache; struct osd_device *dev = osd_obj2dev(obj); struct osd_scrub *scrub = &dev->od_scrub; struct scrub_file *sf = &scrub->os_file; ino = le32_to_cpu(de->inode); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) { + brelse(bh); + rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino); + GOTO(out, rc); + } + rc = osd_get_fid_from_dentry(de, rec); /* done with de, release bh */ brelse(bh); if (rc != 0) - rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid); + rc = osd_ea_fid_get(env, obj, ino, fid, id); else - osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); - if (rc != 0) { + osd_id_gen(id, ino, OSD_OII_NOGEN); + if (rc != 0 || osd_remote_fid(env, dev, fid)) { fid_zero(&oic->oic_fid); GOTO(out, rc); } + oic->oic_lid = *id; oic->oic_fid = *fid; if ((scrub->os_pos_current <= ino) && - (sf->sf_flags & SF_INCONSISTENT || + ((sf->sf_flags & SF_INCONSISTENT) || + (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) osd_consistency_check(oti, dev, oic); @@ -3616,12 +3825,12 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; - struct inode *inode; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); struct lu_fid *fid = (struct lu_fid *)rec; int rc; ENTRY; - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -3630,24 +3839,39 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT]); - inode = osd_dt_obj(dt)->oo_inode; - LASSERT(inode); + if (osd_dt_obj(dt)->oo_inode == NULL) { + const char *name = (const char *)key; + /* Object is not being created yet. Only happens when + * 1. declare directory create + * 2. declare insert . + * 3. declare insert .. + */ + LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0); + } else { + struct inode *inode = osd_dt_obj(dt)->oo_inode; + + /* We ignore block quota on meta pool (MDTs), so needn't + * calculate how many blocks will be consumed by this index + * insert */ + rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, + oh, true, true, NULL, false); + } - /* We ignore block quota on meta pool (MDTs), so needn't - * calculate how many blocks will be consumed by this index - * insert */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); if (fid == NULL) RETURN(0); - /* It does fld look up inside declare, and the result will be - * added to fld cache, so the following fld lookup inside insert - * does not need send RPC anymore, so avoid send rpc with holding - * transaction */ - LASSERTF(fid_is_sane(fid), "fid is insane"DFID"\n", PFID(fid)); - osd_fld_lookup(env, osd_dt_dev(handle->th_dev), fid, - &osd_oti_get(env)->oti_seq_range); + rc = osd_remote_fid(env, osd, fid); + if (rc <= 0) + RETURN(rc); + + rc = 0; + + osd_trans_declare_op(env, oh, OSD_OT_CREATE, + osd_dto_credits_noquota[DTO_OBJECT_CREATE]); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); RETURN(rc); } @@ -3668,16 +3892,19 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, struct thandle *th, struct lustre_capa *capa, int ignore_quota) { - struct osd_object *obj = osd_dt_obj(dt); - struct lu_fid *fid = (struct lu_fid *) rec; - const char *name = (const char *)key; - struct osd_object *child; - int rc; - - ENTRY; + struct osd_object *obj = osd_dt_obj(dt); + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct lu_fid *fid = (struct lu_fid *) rec; + const char *name = (const char *)key; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_inode_id *id = &oti->oti_id; + struct inode *child_inode = NULL; + struct osd_object *child = NULL; + int rc; + ENTRY; LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt)); + LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); @@ -3685,16 +3912,59 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) RETURN(-EACCES); - child = osd_object_find(env, dt, fid); - if (!IS_ERR(child)) { - rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th); - osd_object_put(env, child); - } else { - rc = PTR_ERR(child); - } + LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid)); - LASSERT(osd_invariant(obj)); - RETURN(rc); + rc = osd_remote_fid(env, osd, fid); + if (rc < 0) { + CERROR("%s: Can not find object "DFID" rc %d\n", + osd_name(osd), PFID(fid), rc); + RETURN(rc); + } + + if (rc == 1) { + /* Insert remote entry */ + if (strcmp(name, dotdot) == 0 && strlen(name) == 2) { + struct osd_mdobj_map *omm = osd->od_mdt_map; + struct osd_thandle *oh; + + /* If parent on remote MDT, we need put this object + * under AGENT */ + oh = container_of(th, typeof(*oh), ot_super); + rc = osd_add_to_agent(env, osd, obj, oh); + if (rc != 0) { + CERROR("%s: add agent "DFID" error: rc = %d\n", + osd_name(osd), + PFID(lu_object_fid(&dt->do_lu)), rc); + RETURN(rc); + } + + child_inode = igrab(omm->omm_agent_dentry->d_inode); + } else { + child_inode = osd_create_remote_inode(env, osd, obj, + fid, th); + if (IS_ERR(child_inode)) + RETURN(PTR_ERR(child_inode)); + } + } else { + /* Insert local entry */ + child = osd_object_find(env, dt, fid); + if (IS_ERR(child)) { + CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n", + osd_name(osd), PFID(fid), + id->oii_ino, id->oii_gen, + (int)PTR_ERR(child_inode)); + RETURN(PTR_ERR(child_inode)); + } + child_inode = igrab(child->oo_inode); + } + + rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th); + + iput(child_inode); + if (child != NULL) + osd_object_put(env, child); + LASSERT(osd_invariant(obj)); + RETURN(rc); } /** @@ -3777,7 +4047,6 @@ static int osd_it_iam_get(const struct lu_env *env, * * \param di osd iterator */ - static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) { struct osd_it_iam *it = (struct osd_it_iam *)di; @@ -3836,44 +4105,42 @@ static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di) return iam_it_key_size(&it->oi_it); } -static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr, - int len, __u16 type) +static inline void +osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type) { - struct luda_type *lt; - const unsigned align = sizeof(struct luda_type) - 1; - - /* check if file type is required */ - if (attr & LUDA_TYPE) { - len = (len + align) & ~align; + /* check if file type is required */ + if (ent->lde_attrs & LUDA_TYPE) { + struct luda_type *lt; + int align = sizeof(*lt) - 1; - lt = (void *) ent->lde_name + len; - lt->lt_type = cpu_to_le16(CFS_DTTOIF(type)); - ent->lde_attrs |= LUDA_TYPE; - } + len = (len + align) & ~align; + lt = (struct luda_type *)(ent->lde_name + len); + lt->lt_type = cpu_to_le16(DTTOIF(type)); + } - ent->lde_attrs = cpu_to_le32(ent->lde_attrs); + ent->lde_attrs = cpu_to_le32(ent->lde_attrs); } /** * build lu direct from backend fs dirent. */ -static inline void osd_it_pack_dirent(struct lu_dirent *ent, - struct lu_fid *fid, __u64 offset, - char *name, __u16 namelen, - __u16 type, __u32 attr) +static inline void +osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset, + char *name, __u16 namelen, __u16 type, __u32 attr) { - fid_cpu_to_le(&ent->lde_fid, fid); - ent->lde_attrs = LUDA_FID; + ent->lde_attrs = attr | LUDA_FID; + fid_cpu_to_le(&ent->lde_fid, fid); - ent->lde_hash = cpu_to_le64(offset); - ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr)); + ent->lde_hash = cpu_to_le64(offset); + ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr)); - strncpy(ent->lde_name, name, namelen); - ent->lde_namelen = cpu_to_le16(namelen); + strncpy(ent->lde_name, name, namelen); + ent->lde_name[namelen] = '\0'; + ent->lde_namelen = cpu_to_le16(namelen); - /* append lustre attributes */ - osd_it_append_attrs(ent, attr, namelen, type); + /* append lustre attributes */ + osd_it_append_attrs(ent, namelen, type); } /** @@ -4083,6 +4350,7 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, unsigned d_type) { struct osd_it_ea *it = (struct osd_it_ea *)buf; + struct osd_object *obj = it->oie_obj; struct osd_it_ea_dirent *ent = it->oie_dirent; struct lu_fid *fid = &ent->oied_fid; struct osd_fid_pack *rec; @@ -4098,16 +4366,23 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, OSD_IT_EA_BUFSIZE) RETURN(1); - if (d_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack*) (name + namelen + 1); - - if (osd_fid_unpack(fid, rec) != 0) - fid_zero(fid); + /* "." is just the object itself. */ + if (namelen == 1 && name[0] == '.') { + *fid = obj->oo_dt.do_lu.lo_header->loh_fid; + } else if (d_type & LDISKFS_DIRENT_LUFID) { + rec = (struct osd_fid_pack*) (name + namelen + 1); + if (osd_fid_unpack(fid, rec) != 0) + fid_zero(fid); + } else { + fid_zero(fid); + } + d_type &= ~LDISKFS_DIRENT_LUFID; - d_type &= ~LDISKFS_DIRENT_LUFID; - } else { - fid_zero(fid); - } + /* NOT export local root. */ + if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) { + ino = obj->oo_inode->i_ino; + *fid = obj->oo_dt.do_lu.lo_header->loh_fid; + } ent->oied_ino = ino; ent->oied_off = offset; @@ -4233,10 +4508,337 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) return it->oie_dirent->oied_namelen; } +static int +osd_dirent_update(handle_t *jh, struct super_block *sb, + struct osd_it_ea_dirent *ent, struct lu_fid *fid, + struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de) +{ + struct osd_fid_pack *rec; + int rc; + ENTRY; + + LASSERT(de->file_type & LDISKFS_DIRENT_LUFID); + LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack)); + + rc = ldiskfs_journal_get_write_access(jh, bh); + if (rc != 0) { + CERROR("%.16s: fail to write access for update dirent: " + "name = %.*s, rc = %d\n", + LDISKFS_SB(sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + RETURN(rc); + } + + rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); + rc = ldiskfs_journal_dirty_metadata(jh, bh); + if (rc != 0) + CERROR("%.16s: fail to dirty metadata for update dirent: " + "name = %.*s, rc = %d\n", + LDISKFS_SB(sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + + RETURN(rc); +} + +static inline int +osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize) +{ + if (ldiskfs_rec_len_from_disk(reclen, blocksize) >= + __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack))) + return 1; + else + return 0; +} + +static int +osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, + struct inode *dir, struct inode *inode, + struct osd_it_ea_dirent *ent, struct lu_fid *fid, + struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de, + struct htree_lock *hlock) +{ + struct dentry *dentry; + struct osd_fid_pack *rec; + struct ldiskfs_dentry_param *ldp; + int rc; + ENTRY; + + if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb, + LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + RETURN(0); + + /* There is enough space to hold the FID-in-dirent. */ + if (osd_dirent_has_space(de->rec_len, ent->oied_namelen, + dir->i_sb->s_blocksize)) { + rc = ldiskfs_journal_get_write_access(jh, bh); + if (rc != 0) { + CERROR("%.16s: fail to write access for reinsert " + "dirent: name = %.*s, rc = %d\n", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + RETURN(rc); + } + + de->name[de->name_len] = 0; + rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + rec->fp_len = sizeof(struct lu_fid) + 1; + fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid); + de->file_type |= LDISKFS_DIRENT_LUFID; + + rc = ldiskfs_journal_dirty_metadata(jh, bh); + if (rc != 0) + CERROR("%.16s: fail to dirty metadata for reinsert " + "dirent: name = %.*s, rc = %d\n", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + + RETURN(rc); + } + + rc = ldiskfs_delete_entry(jh, dir, de, bh); + if (rc != 0) { + CERROR("%.16s: fail to delete entry for reinsert dirent: " + "name = %.*s, rc = %d\n", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + RETURN(rc); + } + + dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name, + ent->oied_namelen); + ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp; + osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid); + dentry->d_fsdata = (void *)ldp; + ll_vfs_dq_init(dir); + rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); + /* It is too bad, we cannot reinsert the name entry back. + * That means we lose it! */ + if (rc != 0) + CERROR("%.16s: fail to insert entry for reinsert dirent: " + "name = %.*s, rc = %d\n", + LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, + ent->oied_namelen, ent->oied_name, rc); + + RETURN(rc); +} + +static int +osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, + struct osd_it_ea *it, struct lu_fid *fid, + struct osd_inode_id *id, __u32 *attr) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + struct osd_device *dev = osd_obj2dev(obj); + struct super_block *sb = osd_sb(dev); + const char *devname = + LDISKFS_SB(sb)->s_es->s_volume_name; + struct osd_it_ea_dirent *ent = it->oie_dirent; + struct inode *dir = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct buffer_head *bh = NULL; + handle_t *jh = NULL; + struct ldiskfs_dir_entry_2 *de; + struct dentry *dentry; + struct inode *inode; + int credits; + int rc; + bool dirty = false; + bool is_dotdot = false; + ENTRY; + + if (ent->oied_name[0] == '.') { + /* Skip dot entry, even if it has stale FID-in-dirent, because + * we do not use such FID-in-dirent anymore, it is harmless. */ + if (ent->oied_namelen == 1) + RETURN(0); + + if (ent->oied_namelen == 2 && ent->oied_name[1] == '.') + is_dotdot = true; + } + + dentry = osd_child_dentry_get(env, obj, ent->oied_name, + ent->oied_namelen); + + /* We need to ensure that the name entry is still valid. + * Because it may be removed or renamed by other already. + * + * The unlink or rename operation will start journal before PDO lock, + * so to avoid deadlock, here we need to start journal handle before + * related PDO lock also. But because we do not know whether there + * will be something to be repaired before PDO lock, we just start + * journal without conditions. + * + * We may need to remove the name entry firstly, then insert back. + * One credit is for user quota file update. + * One credit is for group quota file update. + * Two credits are for dirty inode. */ + credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] + + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; + +again: + if (dev->od_dirent_journal) { + jh = ldiskfs_journal_start_sb(sb, credits); + if (IS_ERR(jh)) { + rc = PTR_ERR(jh); + CERROR("%.16s: fail to start trans for dirent " + "check_repair: credits %d, name %.*s, rc %d\n", + devname, credits, ent->oied_namelen, + ent->oied_name, rc); + RETURN(rc); + } + } + + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + 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, &de, hlock); + /* For dotdot entry, if there is not enough space to hold FID-in-dirent, + * just keep it there. It only happens when the device upgraded from 1.8 + * or restored from MDT file-level backup. For the whole directory, only + * dotdot entry has no FID-in-dirent and needs to get FID from LMA when + * readdir, it will not affect the performance much. */ + if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) || + (is_dotdot && !osd_dirent_has_space(de->rec_len, + ent->oied_namelen, + sb->s_blocksize))) { + *attr |= LUDA_IGNORE; + GOTO(out_journal, rc = 0); + } + + osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN); + inode = osd_iget(info, dev, id); + if (IS_ERR(inode)) { + rc = PTR_ERR(inode); + if (rc == -ENOENT || rc == -ESTALE) { + *attr |= LUDA_IGNORE; + rc = 0; + } + + GOTO(out_journal, rc); + } + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + if (rc == 0) { + if (fid_is_sane(fid)) { + /* FID-in-dirent is valid. */ + if (lu_fid_eq(fid, &lma->lma_self_fid)) + GOTO(out_inode, rc = 0); + + /* Do not repair under dryrun mode. */ + if (*attr & LUDA_VERIFY_DRYRUN) { + *attr |= LUDA_REPAIR; + GOTO(out_inode, rc = 0); + } + + if (!dev->od_dirent_journal) { + iput(inode); + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_write(&obj->oo_ext_idx_sem); + dev->od_dirent_journal = 1; + goto again; + } + + *fid = lma->lma_self_fid; + dirty = true; + /* Update the FID-in-dirent. */ + rc = osd_dirent_update(jh, sb, ent, fid, bh, de); + if (rc == 0) + *attr |= LUDA_REPAIR; + } else { + /* Do not repair under dryrun mode. */ + if (*attr & LUDA_VERIFY_DRYRUN) { + *attr |= LUDA_REPAIR; + GOTO(out_inode, rc = 0); + } + + if (!dev->od_dirent_journal) { + iput(inode); + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_write(&obj->oo_ext_idx_sem); + dev->od_dirent_journal = 1; + goto again; + } + + *fid = lma->lma_self_fid; + dirty = true; + /* Append the FID-in-dirent. */ + rc = osd_dirent_reinsert(env, jh, dir, inode, ent, + fid, bh, de, hlock); + if (rc == 0) + *attr |= LUDA_REPAIR; + } + } else if (rc == -ENODATA) { + /* Do not repair under dryrun mode. */ + if (*attr & LUDA_VERIFY_DRYRUN) { + if (fid_is_sane(fid)) + *attr |= LUDA_REPAIR; + else + *attr |= LUDA_UPGRADE; + GOTO(out_inode, rc = 0); + } + + if (!dev->od_dirent_journal) { + iput(inode); + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_write(&obj->oo_ext_idx_sem); + dev->od_dirent_journal = 1; + goto again; + } + + dirty = true; + if (unlikely(fid_is_sane(fid))) { + /* FID-in-dirent exists, but FID-in-LMA is lost. + * Trust the FID-in-dirent, and add FID-in-LMA. */ + rc = osd_ea_fid_set(info, inode, fid); + if (rc == 0) + *attr |= LUDA_REPAIR; + } else { + lu_igif_build(fid, inode->i_ino, inode->i_generation); + /* It is probably IGIF object. Only aappend the + * FID-in-dirent. OI scrub will process FID-in-LMA. */ + rc = osd_dirent_reinsert(env, jh, dir, inode, ent, + fid, bh, de, hlock); + if (rc == 0) + *attr |= LUDA_UPGRADE; + } + } + + GOTO(out_inode, rc); + +out_inode: + iput(inode); + +out_journal: + brelse(bh); + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_write(&obj->oo_ext_idx_sem); + if (jh != NULL) + ldiskfs_journal_stop(jh); + if (rc >= 0 && !dirty) + dev->od_dirent_journal = 0; + return rc; +} /** - * Returns the value (i.e. fid/igif) at current position from iterator's - * in memory structure. + * Returns the value at current position from iterator's in memory structure. * * \param di struct osd_it_ea, iterator's in memory structure * \param attr attr requested for dirent. @@ -4255,6 +4857,7 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_scrub *scrub = &dev->od_scrub; struct scrub_file *sf = &scrub->os_file; struct osd_thread_info *oti = osd_oti_get(env); + struct osd_inode_id *id = &oti->oti_id; struct osd_idmap_cache *oic = &oti->oti_cache; struct lu_fid *fid = &it->oie_dirent->oied_fid; struct lu_dirent *lde = (struct lu_dirent *)dtrec; @@ -4262,23 +4865,48 @@ static inline int osd_it_ea_rec(const struct lu_env *env, int rc = 0; ENTRY; - if (!fid_is_sane(fid)) { - rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid); - if (rc != 0) { - fid_zero(&oic->oic_fid); - RETURN(rc); + if (attr & LUDA_VERIFY) { + attr |= LUDA_TYPE; + if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { + attr |= LUDA_IGNORE; + rc = 0; + goto pack; } + + rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); } else { - osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); + attr &= ~LU_DIRENT_ATTRS_MASK; + if (!fid_is_sane(fid)) { + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + RETURN(-ENOENT); + + rc = osd_ea_fid_get(env, obj, ino, fid, id); + } else { + osd_id_gen(id, ino, OSD_OII_NOGEN); + } } + if (rc < 0) + RETURN(rc); + +pack: osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off, it->oie_dirent->oied_name, it->oie_dirent->oied_namelen, it->oie_dirent->oied_type, attr); - oic->oic_fid = *fid; - if ((scrub->os_pos_current <= ino) && - (sf->sf_flags & SF_INCONSISTENT || + + if (osd_remote_fid(env, dev, fid)) + RETURN(0); + + if (likely(!(attr & LUDA_IGNORE))) { + oic->oic_lid = *id; + oic->oic_fid = *fid; + } + + if (!(attr & LUDA_VERIFY) && + (scrub->os_pos_current <= ino) && + ((sf->sf_flags & SF_INCONSISTENT) || + (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) || ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) osd_consistency_check(oti, dev, oic); @@ -4410,12 +5038,14 @@ static void *osd_key_init(const struct lu_context *ctx, static void osd_key_fini(const struct lu_context *ctx, struct lu_context_key *key, void* data) { - struct osd_thread_info *info = data; + struct osd_thread_info *info = data; - if (info->oti_hlock != NULL) - ldiskfs_htree_lock_free(info->oti_hlock); - OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - OBD_FREE_PTR(info); + if (info->oti_hlock != NULL) + ldiskfs_htree_lock_free(info->oti_hlock); + OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); + lu_buf_free(&info->oti_iobuf.dr_pg_buf); + lu_buf_free(&info->oti_iobuf.dr_bl_buf); + OBD_FREE_PTR(info); } static void osd_key_exit(const struct lu_context *ctx, @@ -4444,7 +5074,9 @@ static int osd_device_init(const struct lu_env *env, struct lu_device *d, { struct osd_device *osd = osd_dev(d); - strncpy(osd->od_svname, name, MAX_OBD_NAME); + if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) + >= sizeof(osd->od_svname)) + return -E2BIG; return osd_procfs_init(osd, name); } @@ -4534,13 +5166,15 @@ static int osd_mount(const struct lu_env *env, GOTO(out, rc); } - if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { +#ifdef HAVE_DEV_SET_RDONLY + if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { CERROR("%s: underlying device %s is marked as read-only. " "Setup failed\n", name, dev); mntput(o->od_mnt); o->od_mnt = NULL; GOTO(out, rc = -EROFS); } +#endif if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { @@ -4550,6 +5184,8 @@ static int osd_mount(const struct lu_env *env, GOTO(out, rc = -EINVAL); } + ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode, + LDISKFS_STATE_LUSTRE_NO_OI); if (lmd_flags & LMD_FLG_NOSCRUB) o->od_noscrub = 1; @@ -4596,6 +5232,7 @@ static int osd_device_init0(const struct lu_env *env, struct lu_device *l = osd2lu_dev(o); struct osd_thread_info *info; int rc; + int cplen = 0; /* if the module was re-loaded, env can loose its keys */ rc = lu_env_refill((struct lu_env *) env); @@ -4623,13 +5260,18 @@ static int osd_device_init0(const struct lu_env *env, if (rc) GOTO(out_capa, rc); + CFS_INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ rc = osd_scrub_setup(env, o); if (rc < 0) GOTO(out_mnt, rc); - strncpy(o->od_svname, lustre_cfg_string(cfg, 4), - sizeof(o->od_svname) - 1); + cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), + sizeof(o->od_svname)); + if (cplen >= sizeof(o->od_svname)) { + rc = -E2BIG; + GOTO(out_mnt, rc); + } rc = osd_obj_map_init(o); if (rc != 0)