X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=3c09617e33fd2ab0c2bd4996b3403c35849b8ef5;hb=26995a3588e72d65f609b12772d24a879c9deb7f;hp=2288b642741682f8883aa3aaf9d5e56d803becd2;hpb=add1417e6f95a63cd3ed90c968b7b0c260168ce4;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 2288b64..3c09617 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -41,7 +41,7 @@ * Pravin Shelar : Added fid in dirent */ -#define DEBUG_SUBSYSTEM S_MDS +#define DEBUG_SUBSYSTEM S_OSD #include @@ -72,6 +72,9 @@ #include #include +#include +#include + int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, "ldiskfs with parallel directory operations"); @@ -133,10 +136,6 @@ static int osd_object_invariant(const struct lu_object *l) /* * Concurrency: doesn't matter */ -static int osd_read_locked(const struct lu_env *env, struct osd_object *o) -{ - return osd_oti_get(env)->oti_r_locks > 0; -} /* * Concurrency: doesn't matter @@ -353,7 +352,7 @@ check_oi: LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); - /* XXX: There are three possible cases: + /* XXX: There are some possible cases: * 1. rc = 0. * Backup/restore caused the OI invalid. * 2. rc = 0. @@ -468,9 +467,6 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) int rc; ENTRY; - if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY)) - RETURN(0); - CLASSERT(LMA_OLD_SIZE >= sizeof(*lma)); rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, info->oti_mdt_attrs_old, LMA_OLD_SIZE); @@ -563,7 +559,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid)); + LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid)); dev = osd_dev(ldev); scrub = &dev->od_scrub; @@ -591,7 +587,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, } id = &info->oti_id; - if (!cfs_list_empty(&scrub->os_inconsistent_items)) { + if (!list_empty(&scrub->os_inconsistent_items)) { /* Search order: 2. OI scrub pending list. */ result = osd_oii_lookup(dev, fid, id); if (result == 0) @@ -895,7 +891,7 @@ static int osd_param_is_not_sane(const struct osd_device *dev, { struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super); - return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers; + return oh->ot_credits > osd_transaction_size(dev); } /* @@ -917,11 +913,11 @@ static void osd_trans_commit_cb(struct super_block *sb, dt_txn_hook_commit(th); /* call per-transaction callbacks if any */ - cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { + list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC, "commit callback entry: magic=%x name='%s'\n", dcb->dcb_magic, dcb->dcb_name); - cfs_list_del_init(&dcb->dcb_linkage); + list_del_init(&dcb->dcb_linkage); dcb->dcb_func(NULL, th, dcb, error); } @@ -931,7 +927,7 @@ static void osd_trans_commit_cb(struct super_block *sb, lu_context_exit(&th->th_ctx); lu_context_fini(&th->th_ctx); - OBD_FREE_PTR(oh); + thandle_put(th); } static struct thandle *osd_trans_create(const struct lu_env *env, @@ -946,19 +942,21 @@ static struct thandle *osd_trans_create(const struct lu_env *env, /* on pending IO in this thread should left from prev. request */ LASSERT(atomic_read(&iobuf->dr_numreqs) == 0); - th = ERR_PTR(-ENOMEM); - OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO); - if (oh != NULL) { + th = ERR_PTR(-ENOMEM); + OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); + if (oh != NULL) { oh->ot_quota_trans = &oti->oti_quota_trans; memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); - th = &oh->ot_super; - th->th_dev = d; - th->th_result = 0; - th->th_tags = LCT_TX_HANDLE; - oh->ot_credits = 0; - oti->oti_dev = osd_dt_dev(d); - CFS_INIT_LIST_HEAD(&oh->ot_dcb_list); - osd_th_alloced(oh); + th = &oh->ot_super; + th->th_dev = d; + th->th_result = 0; + th->th_tags = LCT_TX_HANDLE; + oh->ot_credits = 0; + atomic_set(&th->th_refc, 1); + th->th_alloc_size = sizeof(*oh); + oti->oti_dev = osd_dt_dev(d); + INIT_LIST_HEAD(&oh->ot_dcb_list); + osd_th_alloced(oh); memset(oti->oti_declare_ops, 0, sizeof(oti->oti_declare_ops)); @@ -967,8 +965,8 @@ static struct thandle *osd_trans_create(const struct lu_env *env, memset(oti->oti_declare_ops_cred, 0, sizeof(oti->oti_declare_ops_cred)); oti->oti_rollback = false; - } - RETURN(th); + } + RETURN(th); } /* @@ -1003,11 +1001,9 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, oh->ot_credits, osd_journal(dev)->j_max_transaction_buffers); - CWARN(" create: %u/%u, delete: %u/%u, destroy: %u/%u\n", + CWARN(" create: %u/%u, destroy: %u/%u\n", oti->oti_declare_ops[OSD_OT_CREATE], oti->oti_declare_ops_cred[OSD_OT_CREATE], - oti->oti_declare_ops[OSD_OT_DELETE], - oti->oti_declare_ops_cred[OSD_OT_DELETE], oti->oti_declare_ops[OSD_OT_DESTROY], oti->oti_declare_ops_cred[OSD_OT_DESTROY]); CWARN(" attr_set: %u/%u, xattr_set: %u/%u\n", @@ -1025,8 +1021,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, CWARN(" insert: %u/%u, delete: %u/%u\n", oti->oti_declare_ops[OSD_OT_INSERT], oti->oti_declare_ops_cred[OSD_OT_INSERT], - oti->oti_declare_ops[OSD_OT_DESTROY], - oti->oti_declare_ops_cred[OSD_OT_DESTROY]); + oti->oti_declare_ops[OSD_OT_DELETE], + oti->oti_declare_ops_cred[OSD_OT_DELETE]); CWARN(" ref_add: %u/%u, ref_del: %u/%u\n", oti->oti_declare_ops[OSD_OT_REF_ADD], oti->oti_declare_ops_cred[OSD_OT_REF_ADD], @@ -1034,7 +1030,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, oti->oti_declare_ops_cred[OSD_OT_REF_DEL]); if (last_credits != oh->ot_credits && - time_after(jiffies, last_printed + 60 * HZ)) { + time_after(jiffies, last_printed + + msecs_to_jiffies(60 * MSEC_PER_SEC))) { libcfs_debug_dumpstack(NULL); last_credits = oh->ot_credits; last_printed = jiffies; @@ -1045,7 +1042,7 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, * * This should be removed when we can calculate the * credits precisely. */ - oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers; + oh->ot_credits = osd_transaction_size(dev); } /* @@ -1080,8 +1077,8 @@ static int osd_seq_exists(const struct lu_env *env, int rc; ENTRY; - if (ss == NULL) - RETURN(1); + LASSERT(ss != NULL); + LASSERT(ss->ss_server_fld != NULL); rc = osd_fld_lookup(env, osd, seq, range); if (rc != 0) { @@ -1097,21 +1094,20 @@ static int osd_seq_exists(const struct lu_env *env, /* * Concurrency: shouldn't matter. */ -static int osd_trans_stop(const struct lu_env *env, struct thandle *th) +static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, + struct thandle *th) { - int rc = 0; - struct osd_thandle *oh; - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; + int rc = 0; + struct osd_thandle *oh; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; struct qsd_instance *qsd = oti->oti_dev->od_quota_slave; - ENTRY; + struct lquota_trans *qtrans; + ENTRY; - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of0(th, struct osd_thandle, ot_super); - if (qsd != NULL) - /* inform the quota slave device that the transaction is - * stopping */ - qsd_op_end(env, qsd, oh->ot_quota_trans); + qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; if (oh->ot_handle != NULL) { @@ -1140,9 +1136,12 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) if (rc != 0) CERROR("Failure to stop transaction: %d\n", rc); } else { - OBD_FREE_PTR(oh); + thandle_put(&oh->ot_super); } + /* inform the quota slave device that the transaction is stopping */ + qsd_op_end(env, qsd, qtrans); + /* as we want IO to journal and data IO be concurrent, we don't block * awaiting data IO completion in osd_do_bio(), instead we wait here * once transaction is submitted to the journal. all reqular requests @@ -1154,6 +1153,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) */ wait_event(iobuf->dr_wait, atomic_read(&iobuf->dr_numreqs) == 0); + osd_fini_iobuf(oti->oti_dev, iobuf); if (!rc) rc = iobuf->dr_error; @@ -1167,7 +1167,7 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC); LASSERT(&dcb->dcb_func != NULL); - cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); + list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); return 0; } @@ -1193,8 +1193,8 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) osd_index_fini(obj); if (inode != NULL) { struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; - qid_t uid = inode->i_uid; - qid_t gid = inode->i_gid; + qid_t uid = i_uid_read(inode); + qid_t gid = i_gid_read(inode); iput(inode); obj->oo_inode = NULL; @@ -1242,6 +1242,8 @@ static int osd_object_print(const struct lu_env *env, void *cookie, d ? d->id_ops->id_name : "plain"); } +#define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */ + /* * Concurrency: shouldn't matter. */ @@ -1266,24 +1268,28 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, } spin_lock(&osd->od_osfs_lock); - /* cache 1 second */ - if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) { - result = sb->s_op->statfs(sb->s_root, ksfs); - if (likely(result == 0)) { /* N.B. statfs can't really fail */ - osd->od_osfs_age = cfs_time_current_64(); - statfs_pack(&osd->od_statfs, ksfs); - if (sb->s_flags & MS_RDONLY) - sfs->os_state = OS_STATE_READONLY; - } + result = sb->s_op->statfs(sb->s_root, ksfs); + if (likely(result == 0)) { /* N.B. statfs can't really fail */ + statfs_pack(sfs, ksfs); + if (sb->s_flags & MS_RDONLY) + sfs->os_state = OS_STATE_READONLY; } - if (likely(result == 0)) - *sfs = osd->od_statfs; spin_unlock(&osd->od_osfs_lock); - if (unlikely(env == NULL)) + if (unlikely(env == NULL)) OBD_FREE_PTR(ksfs); + /* Reserve a small amount of space for local objects like last_rcvd, + * llog, quota files, ... */ + if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) { + sfs->os_bavail = 0; + } else { + sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS; + /** Take out metadata overhead for indirect blocks */ + sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3); + } + return result; } @@ -1306,6 +1312,7 @@ static void osd_conf_get(const struct lu_env *env, struct dt_device_param *param) { struct super_block *sb = osd_sb(osd_dt_dev(dev)); + int ea_overhead; /* * XXX should be taken from not-yet-existing fs abstraction layer. @@ -1329,13 +1336,19 @@ static void osd_conf_get(const struct lu_env *env, if (test_opt(sb, POSIX_ACL)) param->ddp_mntopts |= MNTOPT_ACL; + /* LOD might calculate the max stripe count based on max_ea_size, + * so we need take account in the overhead as well, + * xattr_header + magic + xattr_entry_head */ + ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) + + LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN); + #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE) - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) - param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE; - else + if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)) + param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE - + ea_overhead; + else #endif - param->ddp_max_ea_size = sb->s_blocksize; - + param->ddp_max_ea_size = sb->s_blocksize - ea_overhead; } /* @@ -1424,55 +1437,57 @@ static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d, * If we mount with --data_journal we may need more. */ const int osd_dto_credits_noquota[DTO_NR] = { - /** - * Insert/Delete. - * INDEX_EXTRA_TRANS_BLOCKS(8) + - * SINGLEDATA_TRANS_BLOCKS(8) - * XXX Note: maybe iam need more, since iam have more level than - * EXT3 htree. - */ - [DTO_INDEX_INSERT] = 16, - [DTO_INDEX_DELETE] = 16, - /** + /** + * Insert. + * INDEX_EXTRA_TRANS_BLOCKS(8) + + * SINGLEDATA_TRANS_BLOCKS(8) + * XXX Note: maybe iam need more, since iam have more level than + * EXT3 htree. + */ + [DTO_INDEX_INSERT] = 16, + /** + * Delete + * just modify a single entry, probably merge few within a block + */ + [DTO_INDEX_DELETE] = 1, + /** * Used for OI scrub - */ - [DTO_INDEX_UPDATE] = 16, - /** - * Create a object. The same as create object in EXT3. - * DATA_TRANS_BLOCKS(14) + - * INDEX_EXTRA_BLOCKS(8) + - * 3(inode bits, groups, GDT) - */ - [DTO_OBJECT_CREATE] = 25, - /** - * XXX: real credits to be fixed - */ - [DTO_OBJECT_DELETE] = 25, - /** - * Attr set credits (inode) - */ - [DTO_ATTR_SET_BASE] = 1, - /** - * Xattr set. The same as xattr of EXT3. - * DATA_TRANS_BLOCKS(14) - * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS - * are also counted in. Do not know why? - */ - [DTO_XATTR_SET] = 14, - [DTO_LOG_REC] = 14, - /** - * credits for inode change during write. - */ - [DTO_WRITE_BASE] = 3, - /** - * credits for single block write. - */ - [DTO_WRITE_BLOCK] = 14, - /** - * Attr set credits for chown. - * This is extra credits for setattr, and it is null without quota - */ - [DTO_ATTR_SET_CHOWN]= 0 + */ + [DTO_INDEX_UPDATE] = 16, + /** + * 4(inode, inode bits, groups, GDT) + * notice: OI updates are counted separately with DTO_INDEX_INSERT + */ + [DTO_OBJECT_CREATE] = 4, + /** + * 4(inode, inode bits, groups, GDT) + * notice: OI updates are counted separately with DTO_INDEX_DELETE + */ + [DTO_OBJECT_DELETE] = 4, + /** + * Attr set credits (inode) + */ + [DTO_ATTR_SET_BASE] = 1, + /** + * Xattr set. The same as xattr of EXT3. + * DATA_TRANS_BLOCKS(14) + * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS + * are also counted in. Do not know why? + */ + [DTO_XATTR_SET] = 14, + /** + * credits for inode change during write. + */ + [DTO_WRITE_BASE] = 3, + /** + * credits for single block write. + */ + [DTO_WRITE_BLOCK] = 14, + /** + * Attr set credits for chown. + * This is extra credits for setattr, and it is null without quota + */ + [DTO_ATTR_SET_CHOWN] = 0 }; static const struct dt_device_operations osd_dt_ops = { @@ -1674,40 +1689,41 @@ static struct timespec *osd_inode_time(const struct lu_env *env, static void osd_inode_getattr(const struct lu_env *env, - struct inode *inode, struct lu_attr *attr) -{ - attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | - LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | - LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | - LA_TYPE; - - attr->la_atime = LTIME_S(inode->i_atime); - attr->la_mtime = LTIME_S(inode->i_mtime); - attr->la_ctime = LTIME_S(inode->i_ctime); - attr->la_mode = inode->i_mode; - attr->la_size = i_size_read(inode); - attr->la_blocks = inode->i_blocks; - attr->la_uid = inode->i_uid; - attr->la_gid = inode->i_gid; - attr->la_flags = LDISKFS_I(inode)->i_flags; - attr->la_nlink = inode->i_nlink; - attr->la_rdev = inode->i_rdev; - attr->la_blksize = 1 << inode->i_blkbits; - attr->la_blkbits = inode->i_blkbits; + struct inode *inode, struct lu_attr *attr) +{ + attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | + LA_SIZE | LA_BLOCKS | LA_UID | LA_GID | + LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE | + LA_TYPE; + + attr->la_atime = LTIME_S(inode->i_atime); + attr->la_mtime = LTIME_S(inode->i_mtime); + attr->la_ctime = LTIME_S(inode->i_ctime); + attr->la_mode = inode->i_mode; + attr->la_size = i_size_read(inode); + attr->la_blocks = inode->i_blocks; + attr->la_uid = i_uid_read(inode); + attr->la_gid = i_gid_read(inode); + attr->la_flags = LDISKFS_I(inode)->i_flags; + attr->la_nlink = inode->i_nlink; + attr->la_rdev = inode->i_rdev; + attr->la_blksize = 1 << inode->i_blkbits; + attr->la_blkbits = inode->i_blkbits; } static int osd_attr_get(const struct lu_env *env, - struct dt_object *dt, - struct lu_attr *attr, - struct lustre_capa *capa) + struct dt_object *dt, + struct lu_attr *attr, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LINVRNT(osd_invariant(obj)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) + return -EACCES; spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); @@ -1724,9 +1740,11 @@ static int osd_declare_attr_set(const struct lu_env *env, struct osd_object *obj; struct osd_thread_info *info = osd_oti_get(env); struct lquota_id_info *qi = &info->oti_qi; + qid_t uid; + qid_t gid; long long bspace; int rc = 0; - bool allocated; + bool enforce; ENTRY; LASSERT(dt != NULL); @@ -1754,27 +1772,28 @@ static int osd_declare_attr_set(const struct lu_env *env, * We still need to call the osd_declare_qid() to calculate the journal * credits for updating quota accounting files and to trigger quota * space adjustment once the operation is completed.*/ - if ((attr->la_valid & LA_UID) != 0 && - attr->la_uid != obj->oo_inode->i_uid) { + if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { + /* USERQUOTA */ + uid = i_uid_read(obj->oo_inode); qi->lqi_type = USRQUOTA; - + enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid); /* inode accounting */ qi->lqi_is_blk = false; - /* one more inode for the new owner ... */ + /* one more inode for the new uid ... */ qi->lqi_id.qid_uid = attr->la_uid; qi->lqi_space = 1; - allocated = (attr->la_uid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + /* Reserve credits for the new uid */ + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); /* and one less inode for the current uid */ - qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + qi->lqi_id.qid_uid = uid; qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1783,47 +1802,49 @@ static int osd_declare_attr_set(const struct lu_env *env, /* block accounting */ qi->lqi_is_blk = true; - /* more blocks for the new owner ... */ + /* more blocks for the new uid ... */ qi->lqi_id.qid_uid = attr->la_uid; qi->lqi_space = bspace; - allocated = (attr->la_uid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + /* + * Credits for the new uid has been reserved, re-use "obj" + * to save credit reservation. + */ + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - /* and finally less blocks for the current owner */ - qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + /* and finally less blocks for the current uid */ + qi->lqi_id.qid_uid = uid; qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - } - if (attr->la_valid & LA_GID && - attr->la_gid != obj->oo_inode->i_gid) { + /* GROUP QUOTA */ + gid = i_gid_read(obj->oo_inode); qi->lqi_type = GRPQUOTA; + enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid); /* inode accounting */ qi->lqi_is_blk = false; - /* one more inode for the new group owner ... */ + /* one more inode for the new gid ... */ qi->lqi_id.qid_gid = attr->la_gid; qi->lqi_space = 1; - allocated = (attr->la_gid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); /* and one less inode for the current gid */ - qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + qi->lqi_id.qid_gid = gid; qi->lqi_space = -1; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1832,20 +1853,19 @@ static int osd_declare_attr_set(const struct lu_env *env, /* block accounting */ qi->lqi_is_blk = true; - /* more blocks for the new owner ... */ + /* more blocks for the new gid ... */ qi->lqi_id.qid_gid = attr->la_gid; qi->lqi_space = bspace; - allocated = (attr->la_gid == 0) ? true : false; - rc = osd_declare_qid(env, oh, qi, allocated, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) RETURN(rc); - /* and finally less blocks for the current owner */ - qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + /* and finally less blocks for the current gid */ + qi->lqi_id.qid_gid = gid; qi->lqi_space = -bspace; - rc = osd_declare_qid(env, oh, qi, true, NULL); + rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (rc == -EDQUOT || rc == -EINPROGRESS) rc = 0; if (rc) @@ -1856,11 +1876,16 @@ static int osd_declare_attr_set(const struct lu_env *env, } static int osd_inode_setattr(const struct lu_env *env, - struct inode *inode, const struct lu_attr *attr) + struct inode *inode, const struct lu_attr *attr) { - __u64 bits; + __u64 bits = attr->la_valid; - bits = attr->la_valid; + /* Only allow set size for regular file */ + if (!S_ISREG(inode->i_mode)) + bits &= ~(LA_SIZE | LA_BLOCKS); + + if (bits == 0) + return 0; if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); @@ -1879,17 +1904,17 @@ static int osd_inode_setattr(const struct lu_env *env, if (bits & LA_BLOCKS) inode->i_blocks = attr->la_blocks; #endif - if (bits & LA_MODE) - inode->i_mode = (inode->i_mode & S_IFMT) | - (attr->la_mode & ~S_IFMT); - if (bits & LA_UID) - inode->i_uid = attr->la_uid; - if (bits & LA_GID) - inode->i_gid = attr->la_gid; - if (bits & LA_NLINK) + if (bits & LA_MODE) + inode->i_mode = (inode->i_mode & S_IFMT) | + (attr->la_mode & ~S_IFMT); + if (bits & LA_UID) + i_uid_write(inode, attr->la_uid); + if (bits & LA_GID) + i_gid_write(inode, attr->la_gid); + if (bits & LA_NLINK) set_nlink(inode, attr->la_nlink); - if (bits & LA_RDEV) - inode->i_rdev = attr->la_rdev; + if (bits & LA_RDEV) + inode->i_rdev = attr->la_rdev; if (bits & LA_FLAGS) { /* always keep S_NOCMTIME */ @@ -1901,24 +1926,25 @@ static int osd_inode_setattr(const struct lu_env *env, static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) { - if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) || - (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) { + if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) || + (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) { struct iattr iattr; int rc; + ll_vfs_dq_init(inode); iattr.ia_valid = 0; if (attr->la_valid & LA_UID) iattr.ia_valid |= ATTR_UID; if (attr->la_valid & LA_GID) iattr.ia_valid |= ATTR_GID; - iattr.ia_uid = attr->la_uid; - iattr.ia_gid = attr->la_gid; + iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid); + iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid); rc = ll_vfs_dq_transfer(inode, &iattr); if (rc) { CERROR("%s: quota transfer failed: rc = %d. Is quota " - "enforcement enabled on the ldiskfs filesystem?", - inode->i_sb->s_id, rc); + "enforcement enabled on the ldiskfs " + "filesystem?\n", inode->i_sb->s_id, rc); return rc; } } @@ -1926,21 +1952,22 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) } static int osd_attr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_attr *attr, - struct thandle *handle, - struct lustre_capa *capa) + struct dt_object *dt, + const struct lu_attr *attr, + struct thandle *handle, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode; + int rc; - LASSERT(handle != NULL); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_invariant(obj)); + LASSERT(handle != NULL); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_invariant(obj)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) + return -EACCES; osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); @@ -1970,7 +1997,6 @@ static int osd_attr_set(const struct lu_env *env, } inode = obj->oo_inode; - ll_vfs_dq_init(inode); rc = osd_quota_transfer(inode, attr); if (rc) @@ -2031,7 +2057,6 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, /* 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); - ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI); obj->oo_inode = inode; result = 0; } else { @@ -2196,7 +2221,6 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, { LASSERT(ah); - memset(ah, 0, sizeof(*ah)); ah->dah_parent = parent; ah->dah_mode = child_mode; } @@ -2232,7 +2256,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, * should not happen since quota enforcement is no longer * enabled on ldiskfs (lquota takes care of it). */ - LASSERTF(result == 0, "%d", result); + LASSERTF(result == 0, "%d\n", result); ll_dirty_inode(inode, I_DIRTY_DATASYNC); } @@ -2342,12 +2366,10 @@ 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]); - if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev), - lu_object_fid(&dt->do_lu), OI_CHECK_FLD)) - /* Reuse idle OI block may cause additional one OI block - * to be changed. */ - osd_trans_declare_op(env, oh, OSD_OT_INSERT, - osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + /* 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 @@ -2364,7 +2386,7 @@ static int osd_declare_object_create(const struct lu_env *env, RETURN(0); rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh, - false, false, NULL, false); + osd_dt_obj(dt), false, NULL, false); if (rc != 0) RETURN(rc); @@ -2372,22 +2394,21 @@ static int osd_declare_object_create(const struct lu_env *env, } static int osd_object_create(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, struct thandle *th) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - int result; - - ENTRY; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + int result; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) /* Quota files can't be created from the kernel any more, @@ -2434,13 +2455,13 @@ 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_INDEX_DELETE] + 3); /* one less inode */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh, - false, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + -1, oh, obj, false, NULL, false); if (rc) RETURN(rc); /* data to be truncated */ - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + 0, oh, obj, true, NULL, false); RETURN(rc); } @@ -2555,9 +2576,9 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, * its inmemory API. */ void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) + const struct lu_fid *fid) { - if (!fid_is_namespace_visible((const struct lu_fid *)fid) || + if (!fid_is_namespace_visible(fid) || OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) { param->edp_magic = 0; return; @@ -2594,9 +2615,9 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, 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 inode *parent_dir, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, struct osd_thandle *oth) { struct ldiskfs_dentry_param *dot_ldp; @@ -2618,6 +2639,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, struct osd_device *osd, struct osd_object *pobj, const struct lu_fid *fid, + __u32 type, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); @@ -2630,16 +2652,13 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, 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); + local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type); if (IS_ERR(local)) { CERROR("%s: create local error %d\n", osd_name(osd), (int)PTR_ERR(local)); RETURN(local); } - ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NO_OI); /* Set special LMA flag for local agent inode */ rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT); if (rc != 0) { @@ -2648,9 +2667,12 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, RETURN(ERR_PTR(rc)); } + if (!S_ISDIR(type)) + 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); + lu_object_fid(&pobj->oo_dt.do_lu), + fid, oh); if (rc != 0) { CERROR("%s: "DFID" add dot dotdot error: rc = %d\n", osd_name(osd), PFID(fid), rc); @@ -2699,22 +2721,23 @@ static int osd_delete_local_agent_inode(const struct lu_env *env, * \retval -ve, on error */ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) { - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - int result; + const struct lu_fid *fid = lu_object_fid(&dt->do_lu); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + int result; - ENTRY; + ENTRY; - LASSERT(osd_invariant(obj)); - LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) /* Quota files can't be created from the kernel any more, @@ -2724,7 +2747,7 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_CREATE); osd_trans_declare_rb(env, th, OSD_OT_REF_ADD); - result = __osd_object_create(info, obj, attr, hint, dof, th); + result = __osd_object_create(info, obj, attr, hint, dof, th); if (result == 0) result = osd_ea_fid_set(info, obj->oo_inode, fid, fid_is_on_ost(info, osd_obj2dev(obj), @@ -2762,23 +2785,26 @@ static int osd_declare_object_ref_add(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_object_ref_add(const struct lu_env *env, - struct dt_object *dt, struct thandle *th) + struct dt_object *dt, struct thandle *th) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; struct osd_thandle *oh; int rc = 0; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_write_locked(env, obj)); - LASSERT(th != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_ADD); + CDEBUG(D_INODE, DFID" increase nlink %d\n", + PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink); /* * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX * (65000) subdirectories by storing "1" in i_nlink if the link count @@ -2791,8 +2817,14 @@ static int osd_object_ref_add(const struct lu_env *env, * in case they are being linked into the PENDING directory */ spin_lock(&obj->oo_guard); - ldiskfs_inc_count(oh->ot_handle, inode); - LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + if (unlikely(inode->i_nlink == 0)) + /* inc_nlink from 0 may cause WARN_ON */ + set_nlink(inode, 1); + else { + ldiskfs_inc_count(oh->ot_handle, inode); + if (!S_ISDIR(inode->i_mode)) + LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX); + } spin_unlock(&obj->oo_guard); ll_dirty_inode(inode, I_DIRTY_DATASYNC); @@ -2802,16 +2834,17 @@ static int osd_object_ref_add(const struct lu_env *env, } static int osd_declare_object_ref_del(const struct lu_env *env, - struct dt_object *dt, - struct thandle *handle) + struct dt_object *dt, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_REF_DEL, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); @@ -2823,7 +2856,7 @@ static int osd_declare_object_ref_del(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, - struct thandle *th) + struct thandle *th) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -2831,12 +2864,13 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, struct osd_thandle *oh; LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(osd_write_locked(env, obj)); LASSERT(th != NULL); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); + oh = container_of0(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_DEL); @@ -2853,6 +2887,9 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt, return 0; } + CDEBUG(D_INODE, DFID" decrease nlink %d\n", + PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink); + ldiskfs_dec_count(oh->ot_handle, inode); spin_unlock(&obj->oo_guard); @@ -2892,16 +2929,24 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, if (strcmp(name, XATTR_NAME_VERSION) == 0) { /* for version we are just using xattr API but change inode * field instead */ - LASSERT(buf->lb_len == sizeof(dt_obj_version_t)); - osd_object_version_get(env, dt, buf->lb_buf); - return sizeof(dt_obj_version_t); + if (buf->lb_len == 0) + return sizeof(dt_obj_version_t); + + if (buf->lb_len < sizeof(dt_obj_version_t)) + return -ERANGE; + + osd_object_version_get(env, dt, buf->lb_buf); + + return sizeof(dt_obj_version_t); } - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->getxattr != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) + return -EACCES; return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -2913,16 +2958,36 @@ static int osd_declare_xattr_set(const struct lu_env *env, int fl, struct thandle *handle) { struct osd_thandle *oh; + int credits; + struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, - strcmp(name, XATTR_NAME_VERSION) == 0 ? - osd_dto_credits_noquota[DTO_ATTR_SET_BASE] : - osd_dto_credits_noquota[DTO_XATTR_SET]); + /* optimistic optimization: LMA is set first and usually fit inode */ + if (strcmp(name, XATTR_NAME_LMA) == 0) { + if (dt_object_exists(dt)) + credits = 0; + else + credits = 1; + } else if (strcmp(name, XATTR_NAME_VERSION) == 0) { + credits = 1; + } else { + credits = osd_dto_credits_noquota[DTO_XATTR_SET]; + if (buf && buf->lb_len > sb->s_blocksize) { + credits *= (buf->lb_len + sb->s_blocksize - 1) >> + sb->s_blocksize_bits; + } + /* + * xattr set may involve inode quota change, reserve credits for + * dquot_initialize() + */ + oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); + } + + osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits); return 0; } @@ -2972,6 +3037,9 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) return -EACCES; + CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n", + PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len); + osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); if (fl & LU_XATTR_REPLACE) fs_flags |= XATTR_REPLACE; @@ -2979,6 +3047,26 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (fl & LU_XATTR_CREATE) fs_flags |= XATTR_CREATE; + if (strcmp(name, XATTR_NAME_LMV) == 0) { + struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs; + int rc; + + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); + if (rc != 0) + RETURN(rc); + + lma->lma_incompat |= LMAI_STRIPED; + lustre_lma_swab(lma); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, + sizeof(*lma), XATTR_REPLACE); + if (rc != 0) + RETURN(rc); + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) && + strcmp(name, XATTR_NAME_LINK) == 0) + return -ENOSPC; + return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len, fs_flags); } @@ -2987,39 +3075,47 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, * Concurrency: @dt is read locked. */ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, struct lustre_capa *capa) + struct lu_buf *buf, 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); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; - 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)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->listxattr != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) + return -EACCES; - dentry->d_inode = inode; + dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); + return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); } static int osd_declare_xattr_del(const struct lu_env *env, - struct dt_object *dt, const char *name, - struct thandle *handle) + struct dt_object *dt, const char *name, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; + struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, osd_dto_credits_noquota[DTO_XATTR_SET]); + /* + * xattr del may involve inode quota change, reserve credits for + * dquot_initialize() + */ + oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb); return 0; } @@ -3028,29 +3124,31 @@ static int osd_declare_xattr_del(const struct lu_env *env, * Concurrency: @dt is write locked. */ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, - const char *name, 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); - struct dentry *dentry = &info->oti_obj_dentry; - int rc; - - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL); - LASSERT(handle != NULL); + const char *name, 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); + struct dentry *dentry = &info->oti_obj_dentry; + int rc; + + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->removexattr != NULL); + LASSERT(handle != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) + return -EACCES; osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); ll_vfs_dq_init(inode); - dentry->d_inode = inode; + dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - rc = inode->i_op->removexattr(dentry, name); - return rc; + rc = inode->i_op->removexattr(dentry, name); + return rc; } static struct obd_capa *osd_capa_get(const struct lu_env *env, @@ -3071,7 +3169,8 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, if (!osd->od_fl_capa) RETURN(ERR_PTR(-ENOENT)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); LINVRNT(osd_invariant(obj)); /* renewal sanity check */ @@ -3086,16 +3185,16 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, case LC_ID_NONE: RETURN(NULL); case LC_ID_PLAIN: - capa->lc_uid = obj->oo_inode->i_uid; - capa->lc_gid = obj->oo_inode->i_gid; + capa->lc_uid = i_uid_read(obj->oo_inode); + capa->lc_gid = i_gid_read(obj->oo_inode); capa->lc_flags = LC_ID_PLAIN; break; case LC_ID_CONVERT: { __u32 d[4], s[4]; - s[0] = obj->oo_inode->i_uid; + s[0] = i_uid_read(obj->oo_inode); cfs_get_random_bytes(&(s[1]), sizeof(__u32)); - s[2] = obj->oo_inode->i_gid; + s[2] = i_uid_read(obj->oo_inode); cfs_get_random_bytes(&(s[3]), sizeof(__u32)); rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN); if (unlikely(rc)) @@ -3139,7 +3238,8 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, RETURN(oc); } -static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) +static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, + __u64 start, __u64 end) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -3156,13 +3256,19 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) file->f_mapping = inode->i_mapping; file->f_op = inode->i_fop; set_file_inode(file, inode); -#ifndef HAVE_FILE_FSYNC_4ARGS + +#ifdef HAVE_FILE_FSYNC_4ARGS + rc = file->f_op->fsync(file, start, end, 0); +#elif defined(HAVE_FILE_FSYNC_2ARGS) mutex_lock(&inode->i_mutex); -#endif - rc = do_fsync(file, 0); -#ifndef HAVE_FILE_FSYNC_4ARGS + rc = file->f_op->fsync(file, 0); + mutex_unlock(&inode->i_mutex); +#else + mutex_lock(&inode->i_mutex); + rc = file->f_op->fsync(file, dentry, 0); mutex_unlock(&inode->i_mutex); #endif + RETURN(rc); } @@ -3284,7 +3390,8 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, * recheck under lock. */ if (!osd_has_index(obj)) - result = osd_iam_container_init(env, obj, dir); + result = osd_iam_container_init(env, obj, + obj->oo_dir); else result = 0; up_write(&obj->oo_ext_idx_sem); @@ -3413,28 +3520,27 @@ static int osd_index_declare_iam_delete(const struct lu_env *env, * \retval 0 success * \retval -ve failure */ - static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) + const struct dt_key *key, + struct thandle *handle, + struct lustre_capa *capa) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thandle *oh; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - int rc; - - ENTRY; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thandle *oh; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + int rc; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); - LASSERT(handle != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(handle != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) + RETURN(-EACCES); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3468,7 +3574,8 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -3480,8 +3587,8 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, inode = osd_dt_obj(dt)->oo_inode; LASSERT(inode); - rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, - true, true, NULL, false); + rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), + 0, oh, osd_dt_obj(dt), true, NULL, false); RETURN(rc); } @@ -3499,14 +3606,22 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, } static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, - struct lu_fid *fid) + const struct lu_fid *fid) { + struct seq_server_site *ss = osd_seq_site(osd); ENTRY; /* FID seqs not in FLDB, must be local seq */ if (unlikely(!fid_seq_in_fldb(fid_seq(fid)))) RETURN(0); + /* If FLD is not being initialized yet, it only happens during the + * initialization, likely during mgs initialization, and we assume + * this is local FID. */ + if (ss == NULL || ss->ss_server_fld == NULL) + RETURN(0); + + /* Only check the local FLDB here */ if (osd_seq_exists(env, osd, fid_seq(fid))) RETURN(0); @@ -3524,25 +3639,25 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, * \retval -ve, on error */ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) + const struct dt_key *key, struct thandle *handle, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct osd_thandle *oh; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct osd_thandle *oh; struct ldiskfs_dir_entry_2 *de = NULL; - struct buffer_head *bh; - struct htree_lock *hlock = NULL; + struct buffer_head *bh; + struct htree_lock *hlock = NULL; struct lu_fid *fid = &osd_oti_get(env)->oti_fid; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); int rc; - ENTRY; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3681,22 +3796,22 @@ out: * \retval -ve failure */ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, - struct dt_rec *rec, const struct dt_key *key, - struct lustre_capa *capa) + struct dt_rec *rec, const struct dt_key *key, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - struct osd_thread_info *oti = osd_oti_get(env); - struct iam_iterator *it = &oti->oti_idx_it; - struct iam_rec *iam_rec; - int rc; - - ENTRY; + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_thread_info *oti = osd_oti_get(env); + struct iam_iterator *it = &oti->oti_idx_it; + struct iam_rec *iam_rec; + int rc; + ENTRY; - LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) RETURN(-EACCES); @@ -3770,26 +3885,26 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, * \retval -ve failure */ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, struct thandle *th, - struct lustre_capa *capa, int ignore_quota) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct osd_thandle *oh; - struct iam_container *bag = &obj->oo_dir->od_container; - struct osd_thread_info *oti = osd_oti_get(env); - struct iam_rec *iam_rec; - int rc; - - ENTRY; + const struct dt_rec *rec, + const struct dt_key *key, struct thandle *th, + struct lustre_capa *capa, int ignore_quota) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct osd_thandle *oh; + struct iam_container *bag = &obj->oo_dir->od_container; + struct osd_thread_info *oti = osd_oti_get(env); + struct iam_rec *iam_rec; + int rc; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); - LASSERT(th != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(th != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) RETURN(-EACCES); osd_trans_exec_op(env, th, OSD_OT_INSERT); @@ -3831,9 +3946,9 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, * \retval -ve, on error */ static int __osd_ea_add_rec(struct osd_thread_info *info, - struct osd_object *pobj, struct inode *cinode, - const char *name, const struct dt_rec *fid, - struct htree_lock *hlock, struct thandle *th) + struct osd_object *pobj, struct inode *cinode, + const char *name, const struct lu_fid *fid, + struct htree_lock *hlock, struct thandle *th) { struct ldiskfs_dentry_param *ldp; struct dentry *child; @@ -3855,6 +3970,29 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, child->d_fsdata = (void *)ldp; ll_vfs_dq_init(pobj->oo_inode); rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); + if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) { + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + int rc1; + + bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de, + NULL, hlock); + if (bh != NULL) { + rc1 = ldiskfs_journal_get_write_access(oth->ot_handle, + bh); + if (rc1 == 0) { + if (S_ISDIR(cinode->i_mode)) + de->file_type = LDISKFS_DIRENT_LUFID | + LDISKFS_FT_REG_FILE; + else + de->file_type = LDISKFS_DIRENT_LUFID | + LDISKFS_FT_DIR; + ldiskfs_journal_dirty_metadata(oth->ot_handle, + bh); + brelse(bh); + } + } + } RETURN(rc); } @@ -3872,11 +4010,11 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, * \retval -ve, on error */ static int osd_add_dot_dotdot(struct osd_thread_info *info, - struct osd_object *dir, - struct inode *parent_dir, const char *name, - const struct dt_rec *dot_fid, - const struct dt_rec *dot_dot_fid, - struct thandle *th) + struct osd_object *dir, + struct inode *parent_dir, const char *name, + const struct lu_fid *dot_fid, + const struct lu_fid *dot_dot_fid, + struct thandle *th) { struct inode *inode = dir->oo_inode; struct osd_thandle *oth; @@ -3903,9 +4041,19 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, dot_dot_fid, NULL, th); } - result = osd_add_dot_dotdot_internal(info, dir->oo_inode, - parent_dir, dot_fid, - dot_dot_fid, oth); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) { + struct lu_fid tfid = *dot_dot_fid; + + tfid.f_oid--; + result = osd_add_dot_dotdot_internal(info, + dir->oo_inode, parent_dir, dot_fid, + &tfid, oth); + } else { + 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; } @@ -3919,8 +4067,8 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, * value, return by respective functions. */ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, - struct inode *cinode, const char *name, - const struct dt_rec *fid, struct thandle *th) + struct inode *cinode, const char *name, + const struct lu_fid *fid, struct thandle *th) { struct osd_thread_info *info = osd_oti_get(env); struct htree_lock *hlock; @@ -3935,9 +4083,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, pobj->oo_inode, 0); } else { down_write(&pobj->oo_ext_idx_sem); - } - rc = osd_add_dot_dotdot(info, pobj, cinode, name, - (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu), + } + + rc = osd_add_dot_dotdot(info, pobj, cinode, name, + lu_object_fid(&pobj->oo_dt.do_lu), fid, th); } else { if (hlock != NULL) { @@ -3950,11 +4099,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; - *tfid = *(const struct lu_fid *)fid; + *tfid = *fid; tfid->f_ver = ~0; rc = __osd_ea_add_rec(info, pobj, cinode, name, - (const struct dt_rec *)tfid, - hlock, th); + tfid, hlock, th); } else { rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); @@ -4053,6 +4201,83 @@ int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, } /** + * Get parent FID from the linkEA. + * + * For a directory which parent resides on remote MDT, to satisfy the + * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On + * the other hand, to make the lookup(..) on the directory can return + * the real parent FID, we append the real parent FID after its ".." + * name entry in the /REMOTE_PARENT_DIR. + * + * Unfortunately, such PFID-in-dirent cannot be preserved via file-level + * backup. So after the restore, we cannot get the right parent FID from + * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since + * we have stored the real parent FID in the directory object's linkEA, + * we can parse the linkEA for the real parent FID. + * + * \param[in] env pointer to the thread context + * \param[in] obj pointer to the object to be handled + * \param[out]fid pointer to the buffer to hold the parent FID + * + * \retval 0 for getting the real parent FID successfully + * \retval negative error number on failure + */ +static int osd_get_pfid_from_linkea(const struct lu_env *env, + struct osd_object *obj, + struct lu_fid *fid) +{ + struct osd_thread_info *oti = osd_oti_get(env); + struct lu_buf *buf = &oti->oti_big_buf; + struct dentry *dentry = &oti->oti_obj_dentry; + struct inode *inode = obj->oo_inode; + struct linkea_data ldata = { 0 }; + int rc; + ENTRY; + + fid_zero(fid); + if (!S_ISDIR(inode->i_mode)) + RETURN(-EIO); + +again: + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + buf->lb_buf, buf->lb_len); + if (rc == -ERANGE) { + rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, + NULL, 0); + if (rc > 0) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + } + + if (unlikely(rc == 0)) + RETURN(-ENODATA); + + if (rc < 0) + RETURN(rc); + + if (unlikely(buf->lb_buf == NULL)) { + lu_buf_realloc(buf, rc); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + goto again; + } + + ldata.ld_buf = buf; + rc = linkea_init(&ldata); + if (rc == 0) { + linkea_first_entry(&ldata); + linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, 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 * mode (b11826) @@ -4061,33 +4286,34 @@ int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, * \retval -ve, on error */ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, - struct dt_rec *rec, const struct dt_key *key) -{ - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct ldiskfs_dir_entry_2 *de; - struct buffer_head *bh; - struct lu_fid *fid = (struct lu_fid *) rec; - struct htree_lock *hlock = NULL; - int ino; - int rc; + struct dt_rec *rec, const struct dt_key *key) +{ + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct ldiskfs_dir_entry_2 *de; + struct buffer_head *bh; + struct lu_fid *fid = (struct lu_fid *) rec; + struct htree_lock *hlock = NULL; + int ino; + int rc; ENTRY; - LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL); + LASSERT(dir->i_op != NULL); + LASSERT(dir->i_op->lookup != NULL); - dentry = osd_child_dentry_get(env, obj, - (char *)key, strlen((char *)key)); + dentry = osd_child_dentry_get(env, obj, + (char *)key, strlen((char *)key)); - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - dir, LDISKFS_HLOCK_LOOKUP); - } else { + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + dir, LDISKFS_HLOCK_LOOKUP); + } else { down_read(&obj->oo_ext_idx_sem); - } + } - bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); - if (bh) { + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + if (bh) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; struct osd_idmap_cache *oic = &oti->oti_cache; @@ -4106,10 +4332,18 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, /* done with de, release bh */ brelse(bh); - if (rc != 0) - rc = osd_ea_fid_get(env, obj, ino, fid, id); - else + if (rc != 0) { + if (unlikely(ino == osd_remote_parent_ino(dev))) + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + rc = osd_get_pfid_from_linkea(env, obj, fid); + else + rc = osd_ea_fid_get(env, obj, ino, fid, id); + } else { osd_id_gen(id, ino, OSD_OII_NOGEN); + } + if (rc != 0) { fid_zero(&oic->oic_fid); GOTO(out, rc); @@ -4169,15 +4403,17 @@ struct osd_object *osd_object_find(const struct lu_env *env, * in the cache, otherwise lu_object_alloc() crashes * -bzzz */ - luch = lu_object_find_at(env, ludev, fid, NULL); - if (!IS_ERR(luch)) { - if (lu_object_exists(luch)) { - lo = lu_object_locate(luch->lo_header, ludev->ld_type); - if (lo != NULL) - child = osd_obj(lo); - else - LU_OBJECT_DEBUG(D_ERROR, env, luch, - "lu_object can't be located" + luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ? + ludev : ludev->ld_site->ls_top_dev, + fid, NULL); + if (!IS_ERR(luch)) { + if (lu_object_exists(luch)) { + lo = lu_object_locate(luch->lo_header, ludev->ld_type); + if (lo != NULL) + child = osd_obj(lo); + else + LU_OBJECT_DEBUG(D_ERROR, env, luch, + "lu_object can't be located" DFID"\n", PFID(fid)); if (child == NULL) { @@ -4222,7 +4458,7 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, int rc; ENTRY; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -4231,22 +4467,15 @@ 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]); - 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 { + if (osd_dt_obj(dt)->oo_inode != NULL) { struct inode *inode = osd_dt_obj(dt)->oo_inode; /* We ignore block quota on meta pool (MDTs), so needn't * 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); + rc = osd_declare_inode_qid(env, i_uid_read(inode), + i_gid_read(inode), 0, oh, + osd_dt_obj(dt), true, NULL, false); } if (fid == NULL) @@ -4286,7 +4515,8 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, { 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; + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; const char *name = (const char *)key; struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; @@ -4295,16 +4525,17 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, int rc; ENTRY; - LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(th != NULL); + LASSERT(osd_invariant(obj)); + LASSERT(dt_object_exists(dt)); + LASSERT(!dt_object_remote(dt)); + LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) + RETURN(-EACCES); - LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid)); + LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid)); rc = osd_remote_fid(env, osd, fid); if (rc < 0) { @@ -4333,8 +4564,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = igrab(omm->omm_remote_parent->d_inode); } else { child_inode = osd_create_local_agent_inode(env, osd, - obj, fid, - th); + obj, fid, rec1->rec_type & S_IFMT, th); if (IS_ERR(child_inode)) RETURN(PTR_ERR(child_inode)); } @@ -4351,7 +4581,10 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = igrab(child->oo_inode); } - rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th); + rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th); + + CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n", + obj->oo_inode->i_ino, name, child_inode->i_ino, rc); iput(child_inode); if (child != NULL) @@ -4383,16 +4616,30 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ)) return ERR_PTR(-EACCES); - it = &oti->oti_it; - ipd = osd_it_ipd_get(env, bag); - if (likely(ipd != NULL)) { - it->oi_obj = obj; - it->oi_ipd = ipd; - lu_object_get(lo); - iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd); - return (struct dt_it *)it; - } - return ERR_PTR(-ENOMEM); + if (oti->oti_it_inline) { + OBD_ALLOC_PTR(it); + if (it == NULL) + return ERR_PTR(-ENOMEM); + } else { + it = &oti->oti_it; + oti->oti_it_inline = 1; + } + + ipd = osd_it_ipd_get(env, bag); + if (likely(ipd != NULL)) { + it->oi_obj = obj; + it->oi_ipd = ipd; + lu_object_get(lo); + iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd); + return (struct dt_it *)it; + } else { + if (it != &oti->oti_it) + OBD_FREE_PTR(it); + else + oti->oti_it_inline = 0; + + return ERR_PTR(-ENOMEM); + } } /** @@ -4401,12 +4648,17 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; - struct osd_object *obj = it->oi_obj; - - iam_it_fini(&it->oi_it); - osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); - lu_object_put(env, &obj->oo_dt.do_lu); + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; + + iam_it_fini(&it->oi_it); + osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd); + lu_object_put(env, &obj->oo_dt.do_lu); + if (it != &oti->oti_it) + OBD_FREE_PTR(it); + else + oti->oti_it_inline = 0; } /** @@ -4644,29 +4896,40 @@ static const struct dt_index_operations osd_index_iam_ops = { * */ static struct dt_it *osd_it_ea_init(const struct lu_env *env, - struct dt_object *dt, - __u32 attr, - struct lustre_capa *capa) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *it = &info->oti_it_ea; - struct file *file = &it->oie_file; - struct lu_object *lo = &dt->do_lu; - struct dentry *obj_dentry = &info->oti_it_dentry; - ENTRY; - LASSERT(lu_object_exists(lo)); + struct dt_object *dt, + __u32 attr, + struct lustre_capa *capa) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *it; + struct file *file; + struct lu_object *lo = &dt->do_lu; + struct dentry *obj_dentry = &info->oti_it_dentry; + ENTRY; - obj_dentry->d_inode = obj->oo_inode; - obj_dentry->d_sb = osd_sb(osd_obj2dev(obj)); - obj_dentry->d_name.hash = 0; + LASSERT(lu_object_exists(lo)); - it->oie_rd_dirent = 0; - it->oie_it_dirent = 0; - it->oie_dirent = NULL; - it->oie_buf = info->oti_it_ea_buf; - it->oie_obj = obj; + if (info->oti_it_inline) { + OBD_ALLOC_PTR(it); + if (it == NULL) + RETURN(ERR_PTR(-ENOMEM)); + } else { + it = &info->oti_it_ea; + info->oti_it_inline = 1; + } + + obj_dentry->d_inode = obj->oo_inode; + obj_dentry->d_sb = osd_sb(osd_obj2dev(obj)); + obj_dentry->d_name.hash = 0; + it->oie_rd_dirent = 0; + it->oie_it_dirent = 0; + it->oie_dirent = NULL; + it->oie_buf = info->oti_it_ea_buf; + it->oie_obj = obj; + + file = &it->oie_file; /* Reset the "file" totally to avoid to reuse any old value from * former readdir handling, the "file->f_pos" should be zero. */ memset(file, 0, sizeof(*file)); @@ -4691,14 +4954,19 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, */ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - struct osd_object *obj = it->oie_obj; - struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_object *obj = it->oie_obj; + struct inode *inode = obj->oo_inode; ENTRY; it->oie_file.f_op->release(inode, &it->oie_file); lu_object_put(env, &obj->oo_dt.do_lu); - EXIT; + if (it != &info->oti_it_ea) + OBD_FREE_PTR(it); + else + info->oti_it_inline = 0; + EXIT; } /** @@ -4733,6 +5001,12 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di) { } +struct osd_filldir_cbs { +#ifdef HAVE_DIR_CONTEXT + struct dir_context ctx; +#endif + struct osd_it_ea *it; +}; /** * It is called internally by ->readdir(). It fills the * iterator's in-memory data structure with required @@ -4744,11 +5018,11 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di) * \retval 0 on success * \retval 1 on buffer full */ -static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, +static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen, loff_t offset, __u64 ino, unsigned d_type) { - struct osd_it_ea *it = (struct osd_it_ea *)buf; + struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it; struct osd_object *obj = it->oie_obj; struct osd_it_ea_dirent *ent = it->oie_dirent; struct lu_fid *fid = &ent->oied_fid; @@ -4803,6 +5077,7 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen, * * \retval 0 on success * \retval -ve on error + * \retval +1 reach the end of entry */ static int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di) @@ -4811,7 +5086,14 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, struct osd_object *obj = it->oie_obj; struct inode *inode = obj->oo_inode; struct htree_lock *hlock = NULL; - int result = 0; + struct file *filp = &it->oie_file; + int rc = 0; + struct osd_filldir_cbs buf = { +#ifdef HAVE_DIR_CONTEXT + .ctx.actor = osd_ldiskfs_filldir, +#endif + .it = it + }; ENTRY; it->oie_dirent = it->oie_buf; @@ -4825,22 +5107,31 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, down_read(&obj->oo_ext_idx_sem); } - result = inode->i_fop->readdir(&it->oie_file, it, - (filldir_t) osd_ldiskfs_filldir); +#ifdef HAVE_DIR_CONTEXT + buf.ctx.pos = filp->f_pos; + rc = inode->i_fop->iterate(filp, &buf.ctx); + filp->f_pos = buf.ctx.pos; +#else + rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir); +#endif if (hlock != NULL) ldiskfs_htree_unlock(hlock); else up_read(&obj->oo_ext_idx_sem); - if (it->oie_rd_dirent == 0) { - result = -EIO; - } else { - it->oie_dirent = it->oie_buf; - it->oie_it_dirent = 1; - } + if (it->oie_rd_dirent == 0) { + /*If it does not get any dirent, it means it has been reached + *to the end of the dir */ + it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file); + if (rc == 0) + rc = 1; + } else { + it->oie_dirent = it->oie_buf; + it->oie_it_dirent = 1; + } - RETURN(result); + RETURN(rc); } /** @@ -4920,22 +5211,12 @@ osd_dirent_update(handle_t *jh, struct super_block *sb, 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); + if (rc != 0) 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); } @@ -4984,13 +5265,8 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, 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); + if (rc != 0) RETURN(rc); - } de->name[de->name_len] = 0; rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); @@ -4999,38 +5275,29 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, 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); + if (rc != 0) 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); + osd_get_ldiskfs_dirent_param(ldp, fid); dentry->d_fsdata = (void *)ldp; ll_vfs_dq_init(dir); rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock); /* 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", + CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, " + "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - ent->oied_namelen, ent->oied_name, rc); + dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, PFID(fid), rc); RETURN(rc); } @@ -5091,10 +5358,11 @@ again: jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits); if (IS_ERR(jh)) { rc = PTR_ERR(jh); - CERROR("%.16s: fail to start trans for dirent " - "check_repair: credits %d, name %.*s, rc %d\n", - devname, credits, ent->oied_namelen, - ent->oied_name, rc); + CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent " + "check_repair, dir = %lu/%u, credits = %d, " + "name = %.*s: rc = %d\n", + devname, dir->i_ino, dir->i_generation, credits, + ent->oied_namelen, ent->oied_name, rc); RETURN(rc); } @@ -5135,18 +5403,18 @@ again: inode = osd_iget(info, dev, id); if (IS_ERR(inode)) { rc = PTR_ERR(inode); - if (rc == -ENOENT || rc == -ESTALE) { - *attr |= LUDA_IGNORE; - rc = 0; - } + if (rc == -ENOENT || rc == -ESTALE) + rc = 1; + else + CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent " + "check_repair, dir = %lu/%u, name = %.*s: " + "rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, rc); GOTO(out_journal, rc); } - /* skip the REMOTE_PARENT_DIR. */ - if (inode == dev->od_mdt_map->omm_remote_parent->d_inode) - GOTO(out_inode, rc = 0); - rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma); if (rc == 0) { LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI)); @@ -5179,6 +5447,13 @@ again: rc = osd_dirent_update(jh, sb, ent, fid, bh, de); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to update FID " + "in the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { /* Do not repair under dryrun mode. */ if (*attr & LUDA_VERIFY_DRYRUN) { @@ -5205,6 +5480,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to append FID " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } else if (rc == -ENODATA) { /* Do not repair under dryrun mode. */ @@ -5237,6 +5519,13 @@ again: rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; + else + CDEBUG(D_LFSCK, "%.16s: fail to set LMA for " + "update dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } else { lu_igif_build(fid, inode->i_ino, inode->i_generation); /* It is probably IGIF object. Only aappend the @@ -5245,6 +5534,13 @@ again: fid, bh, de, hlock); if (rc == 0) *attr |= LUDA_UPGRADE; + else + CDEBUG(D_LFSCK, "%.16s: fail to append IGIF " + "after the dirent, dir = %lu/%u, " + "name = %.*s, "DFID": rc = %d\n", + devname, dir->i_ino, dir->i_generation, + ent->oied_namelen, ent->oied_name, + PFID(fid), rc); } } @@ -5298,11 +5594,15 @@ static inline int osd_it_ea_rec(const struct lu_env *env, int rc = 0; ENTRY; + LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode); + if (attr & LUDA_VERIFY) { - attr |= LUDA_TYPE; - if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) { + if (unlikely(ino == osd_remote_parent_ino(dev))) { attr |= LUDA_IGNORE; - rc = 0; + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + osd_get_pfid_from_linkea(env, obj, fid); } else { rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); @@ -5310,10 +5610,19 @@ static inline int osd_it_ea_rec(const struct lu_env *env, } else { attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { - if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) && + likely(it->oie_dirent->oied_namelen != 2 || + it->oie_dirent->oied_name[0] != '.' || + it->oie_dirent->oied_name[1] != '.')) RETURN(-ENOENT); - rc = osd_ea_fid_get(env, obj, ino, fid, id); + if (unlikely(ino == osd_remote_parent_ino(dev))) + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + rc = osd_get_pfid_from_linkea(env, obj, fid); + else + rc = osd_ea_fid_get(env, obj, ino, fid, id); } else { osd_id_gen(id, ino, OSD_OII_NOGEN); } @@ -5331,7 +5640,7 @@ static inline int osd_it_ea_rec(const struct lu_env *env, if (osd_remote_fid(env, dev, fid)) RETURN(0); - if (likely(!(attr & LUDA_IGNORE))) + if (likely(!(attr & LUDA_IGNORE) && rc == 0)) rc = osd_add_oi_cache(oti, dev, id, fid); if (!(attr & LUDA_VERIFY) && @@ -5341,7 +5650,28 @@ static inline int osd_it_ea_rec(const struct lu_env *env, ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) osd_consistency_check(oti, dev, oic); - RETURN(rc); + RETURN(rc > 0 ? 0 : rc); +} + +/** + * Returns the record size size at current position. + * + * This function will return record(lu_dirent) size in bytes. + * + * \param[in] env execution environment + * \param[in] di iterator's in memory structure + * \param[in] attr attribute of the entry, only requires LUDA_TYPE to + * calculate the lu_dirent size. + * + * \retval record size(in bytes & in memory) of the current lu_dirent + * entry. + */ +static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di, + __u32 attr) +{ + struct osd_it_ea *it = (struct osd_it_ea *)di; + + return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr); } /** @@ -5379,6 +5709,9 @@ static int osd_it_ea_load(const struct lu_env *env, it->oie_file.f_pos = hash; rc = osd_ldiskfs_it_fill(env, di); + if (rc > 0) + rc = -ENODATA; + if (rc == 0) rc = +1; @@ -5419,23 +5752,24 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826) */ static const struct dt_index_operations osd_index_ea_ops = { - .dio_lookup = osd_index_ea_lookup, - .dio_declare_insert = osd_index_declare_ea_insert, - .dio_insert = osd_index_ea_insert, - .dio_declare_delete = osd_index_declare_ea_delete, - .dio_delete = osd_index_ea_delete, - .dio_it = { - .init = osd_it_ea_init, - .fini = osd_it_ea_fini, - .get = osd_it_ea_get, - .put = osd_it_ea_put, - .next = osd_it_ea_next, - .key = osd_it_ea_key, - .key_size = osd_it_ea_key_size, - .rec = osd_it_ea_rec, - .store = osd_it_ea_store, - .load = osd_it_ea_load - } + .dio_lookup = osd_index_ea_lookup, + .dio_declare_insert = osd_index_declare_ea_insert, + .dio_insert = osd_index_ea_insert, + .dio_declare_delete = osd_index_declare_ea_delete, + .dio_delete = osd_index_ea_delete, + .dio_it = { + .init = osd_it_ea_init, + .fini = osd_it_ea_fini, + .get = osd_it_ea_get, + .put = osd_it_ea_put, + .next = osd_it_ea_next, + .key = osd_it_ea_key, + .key_size = osd_it_ea_key_size, + .rec = osd_it_ea_rec, + .rec_size = osd_it_ea_rec_size, + .store = osd_it_ea_store, + .load = osd_it_ea_load + } }; static void *osd_key_init(const struct lu_context *ctx, @@ -5476,6 +5810,7 @@ static void osd_key_fini(const struct lu_context *ctx, OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); lu_buf_free(&info->oti_iobuf.dr_pg_buf); lu_buf_free(&info->oti_iobuf.dr_bl_buf); + lu_buf_free(&info->oti_big_buf); OBD_FREE_PTR(info); } @@ -5511,6 +5846,43 @@ static int osd_device_init(const struct lu_env *env, struct lu_device *d, return osd_procfs_init(osd, name); } +static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) +{ + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; + + if (osd->od_is_ost || osd->od_cl_seq != NULL) + RETURN(0); + + if (unlikely(ss == NULL)) + RETURN(-ENODEV); + + OBD_ALLOC_PTR(osd->od_cl_seq); + if (osd->od_cl_seq == NULL) + RETURN(-ENOMEM); + + rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA, + osd->od_svname, ss->ss_server_seq); + + if (rc != 0) { + OBD_FREE_PTR(osd->od_cl_seq); + osd->od_cl_seq = NULL; + } + + RETURN(rc); +} + +static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd) +{ + if (osd->od_cl_seq == NULL) + return; + + seq_client_fini(osd->od_cl_seq); + OBD_FREE_PTR(osd->od_cl_seq); + osd->od_cl_seq = NULL; +} + static int osd_shutdown(const struct lu_env *env, struct osd_device *o) { ENTRY; @@ -5521,6 +5893,8 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) o->od_quota_slave = NULL; } + osd_fid_fini(env, o); + RETURN(0); } @@ -5620,9 +5994,13 @@ static int osd_mount(const struct lu_env *env, if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb, LDISKFS_FEATURE_INCOMPAT_DIRDATA)) LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA; + else if (!o->od_is_ost) + CWARN("%s: device %s was upgraded from Lustre-1.x without " + "enabling the dirdata feature. If you do not want to " + "downgrade to Lustre-1.x again, you can enable it via " + "'tune2fs -O dirdata device'\n", name, dev); #endif inode = osd_sb(o)->s_root->d_inode; - ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI); lu_local_obj_fid(fid, OSD_FS_ROOT_OID); rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0); if (rc != 0) { @@ -5682,7 +6060,6 @@ static int osd_device_init0(const struct lu_env *env, spin_lock_init(&o->od_osfs_lock); mutex_init(&o->od_otable_mutex); - o->od_osfs_age = cfs_time_shift_64(-1000); o->od_capa_hash = init_capa_hash(); if (o->od_capa_hash == NULL) @@ -5692,20 +6069,20 @@ static int osd_device_init0(const struct lu_env *env, o->od_writethrough_cache = 1; o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; - rc = osd_mount(env, o, cfg); - if (rc) - GOTO(out_capa, rc); - 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); + GOTO(out_capa, rc); } if (server_name_is_ost(o->od_svname)) o->od_is_ost = 1; + rc = osd_mount(env, o, cfg); + if (rc != 0) + GOTO(out_capa, rc); + rc = osd_obj_map_init(env, o); if (rc != 0) GOTO(out_mnt, rc); @@ -5722,7 +6099,7 @@ static int osd_device_init0(const struct lu_env *env, /* self-repair LMA by default */ o->od_lma_self_repair = 1; - CFS_INIT_LIST_HEAD(&o->od_ios_list); + INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ rc = osd_scrub_setup(env, o); if (rc < 0) @@ -5735,7 +6112,8 @@ static int osd_device_init0(const struct lu_env *env, GOTO(out_scrub, rc); } - LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev); + LASSERT(l->ld_site->ls_linkage.next != NULL); + LASSERT(l->ld_site->ls_linkage.prev != NULL); /* initialize quota slave instance */ o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev, @@ -5917,13 +6295,26 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, int result = 0; ENTRY; - if (osd->od_quota_slave != NULL) + if (osd->od_quota_slave != NULL) { /* set up quota slave objects */ result = qsd_prepare(env, osd->od_quota_slave); + if (result != 0) + RETURN(result); + } + + result = osd_fid_init(env, osd); RETURN(result); } +int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, + struct lu_fid *fid, struct md_op_data *op_data) +{ + struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev); + + return seq_client_alloc_fid(env, osd->od_cl_seq, fid); +} + static const struct lu_object_operations osd_lu_obj_ops = { .loo_object_init = osd_object_init, .loo_object_delete = osd_object_delete, @@ -5967,7 +6358,8 @@ struct lu_device_type osd_device_type = { static struct obd_ops osd_obd_device_ops = { .o_owner = THIS_MODULE, .o_connect = osd_obd_connect, - .o_disconnect = osd_obd_disconnect + .o_disconnect = osd_obd_disconnect, + .o_fid_alloc = osd_fid_alloc, }; static int __init osd_mod_init(void) @@ -5980,11 +6372,9 @@ static int __init osd_mod_init(void) if (rc) return rc; - rc = class_register_type(&osd_obd_device_ops, NULL, NULL, -#ifndef HAVE_ONLY_PROCFS_SEQ - lprocfs_osd_module_vars, -#endif - LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); + rc = class_register_type(&osd_obd_device_ops, NULL, true, + lprocfs_osd_module_vars, + LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); if (rc) lu_kmem_fini(ldiskfs_caches); return rc;