X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=2105a26b2ac8487830f74af52abaae557f5e9548;hp=3ffec1064ef9ebd9a231a1216a29eaee67332178;hb=cef4a7fdb1183deaf2cbb34849a3f9f7b9037da1;hpb=92dfc25dc760bba8664dc480204a72da5f5e215d diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 3ffec10..2105a26 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -41,9 +41,13 @@ * Pravin Shelar : Added fid in dirent */ -#define DEBUG_SUBSYSTEM S_MDS +#define DEBUG_SUBSYSTEM S_OSD #include +#include +#ifdef HAVE_UIDGID_HEADER +# include +#endif /* LUSTRE_VERSION_CODE */ #include @@ -54,6 +58,9 @@ /* XATTR_{REPLACE,CREATE} */ #include +#include +#include +#undef ENTRY /* * struct OBD_{ALLOC,FREE}*() * OBD_FAIL_CHECK @@ -72,6 +79,8 @@ #include #include +#include + int ldiskfs_pdo = 1; CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644, "ldiskfs with parallel directory operations"); @@ -83,6 +92,9 @@ CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644, /* Slab to allocate dynlocks */ struct kmem_cache *dynlock_cachep; +/* Slab to allocate osd_it_ea */ +struct kmem_cache *osd_itea_cachep; + static struct lu_kmem_descr ldiskfs_caches[] = { { .ckd_cache = &dynlock_cachep, @@ -90,6 +102,11 @@ static struct lu_kmem_descr ldiskfs_caches[] = { .ckd_size = sizeof(struct dynlock_handle) }, { + .ckd_cache = &osd_itea_cachep, + .ckd_name = "osd_itea_cache", + .ckd_size = sizeof(struct osd_it_ea) + }, + { .ckd_cache = NULL } }; @@ -116,7 +133,6 @@ int osd_trans_declare_op2rb[] = { [OSD_OT_WRITE] = OSD_OT_WRITE, [OSD_OT_INSERT] = OSD_OT_DELETE, [OSD_OT_DELETE] = OSD_OT_INSERT, - [OSD_OT_UPDATE] = OSD_OT_MAX, [OSD_OT_QUOTA] = OSD_OT_MAX, }; @@ -133,10 +149,6 @@ 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 @@ -350,6 +362,8 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, check_oi: if (rc != 0) { + struct osd_inode_id saved_id = *id; + LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); @@ -371,10 +385,22 @@ check_oi: * to distinguish the 1st case from the 2nd case. */ if (rc == 0) { if (!IS_ERR(inode) && inode->i_generation != 0 && - inode->i_generation == id->oii_gen) + inode->i_generation == id->oii_gen) { rc = -ENOENT; - else + } else { + __u32 level = D_LFSCK; + rc = -EREMCHG; + if (!thread_is_running(&dev->od_scrub.os_thread)) + level |= D_CONSOLE; + + CDEBUG(level, "%s: the OI mapping for the FID " + DFID" become inconsistent, the given ID " + "%u/%u, the ID in OI mapping %u/%u\n", + osd_name(dev), PFID(fid), + saved_id.oii_ino, saved_id.oii_gen, + id->oii_ino, id->oii_ino); + } } } else { if (id->oii_gen == OSD_OII_NOGEN) @@ -468,24 +494,20 @@ 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); if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) { fid = &lma->lma_self_fid; rc = osd_get_idif(info, inode, dentry, fid); - if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) { + if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) { /* For the given OST-object, if it has neither LMA nor * FID in XATTR_NAME_FID, then the given FID (which is * contained in the @obj, from client RPC for locating * the OST-object) is trusted. We use it to generate * the LMA. */ osd_lma_self_repair(info, osd, inode, rfid, - fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ? - LMAC_FID_ON_OST : 0); + LMAC_FID_ON_OST); RETURN(0); } } @@ -506,12 +528,14 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) lma->lma_incompat & ~LMA_INCOMPAT_SUPP, PFID(rfid), inode->i_ino); rc = -EOPNOTSUPP; - } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) { + } else { fid = &lma->lma_self_fid; } } if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) { + __u32 level = D_LFSCK; + if (fid_is_idif(rfid) && fid_is_idif(fid)) { struct ost_id *oi = &info->oti_ostid; struct lu_fid *fid1 = &info->oti_fid3; @@ -526,7 +550,7 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) fid_to_ostid(fid, oi); ostid_to_fid(fid1, oi, idx); if (lu_fid_eq(fid1, rfid)) { - if (osd->od_lma_self_repair) + if (osd->od_index_in_idif) osd_lma_self_repair(info, osd, inode, rfid, LMAC_FID_ON_OST); @@ -535,9 +559,13 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } } - CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n", - osd_name(osd), PFID(rfid), PFID(fid)); + rc = -EREMCHG; + if (!thread_is_running(&osd->od_scrub.os_thread)) + level |= D_CONSOLE; + + CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n", + osd_name(osd), PFID(rfid), PFID(fid)); } RETURN(rc); @@ -563,7 +591,7 @@ 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 +619,7 @@ 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) @@ -637,10 +665,14 @@ trigger: if (thread_is_running(&scrub->os_thread)) { result = -EINPROGRESS; } else if (!dev->od_noscrub) { - result = osd_scrub_start(dev); - LCONSOLE_WARN("%.16s: trigger OI scrub by RPC " - "for "DFID", rc = %d [1]\n", - osd_name(dev), PFID(fid),result); + /* Since we do not know the right OI mapping, + * we have to trigger OI scrub to scan the + * whole device. */ + result = osd_scrub_start(dev, SS_AUTO_FULL | + SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT); + CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI " + "scrub by RPC for "DFID", rc = %d [1]\n", + osd_name(dev), PFID(fid),result); if (result == 0 || result == -EALREADY) result = -EINPROGRESS; else @@ -895,7 +927,7 @@ 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 +949,11 @@ 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 +963,7 @@ static void osd_trans_commit_cb(struct super_block *sb, lu_context_exit(&th->th_ctx); lu_context_fini(&th->th_ctx); - thandle_put(th); + OBD_FREE_PTR(oh); } static struct thandle *osd_trans_create(const struct lu_env *env, @@ -946,21 +978,18 @@ 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; - atomic_set(&th->th_refc, 1); - th->th_alloc_size = sizeof(*oh); - 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; + INIT_LIST_HEAD(&oh->ot_dcb_list); + osd_th_alloced(oh); memset(oti->oti_declare_ops, 0, sizeof(oti->oti_declare_ops)); @@ -969,15 +998,15 @@ 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); } /* * Concurrency: shouldn't matter. */ -int osd_trans_start(const struct lu_env *env, struct dt_device *d, - struct thandle *th) +static int osd_trans_start(const struct lu_env *env, struct dt_device *d, + struct thandle *th) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_device *dev = osd_dt_dev(d); @@ -1005,11 +1034,9 @@ 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", @@ -1027,8 +1054,8 @@ 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], @@ -1036,7 +1063,8 @@ 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; @@ -1047,7 +1075,7 @@ 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); } /* @@ -1075,15 +1103,15 @@ out: } static int osd_seq_exists(const struct lu_env *env, - struct osd_device *osd, obd_seq seq) + struct osd_device *osd, u64 seq) { struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; struct seq_server_site *ss = osd_seq_site(osd); int rc; 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) { @@ -1106,7 +1134,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, 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; + struct osd_device *osd = osd_dt_dev(th->th_dev); + struct qsd_instance *qsd = osd->od_quota_slave; struct lquota_trans *qtrans; ENTRY; @@ -1130,19 +1159,20 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, oti->oti_txns--; rc = dt_txn_hook_stop(env, th); if (rc != 0) - CERROR("Failure in transaction hook: %d\n", rc); + CERROR("%s: failed in transaction hook: rc = %d\n", + osd_name(osd), rc); /* hook functions might modify th_sync */ hdl->h_sync = th->th_sync; - oh->ot_handle = NULL; - OSD_CHECK_SLOW_TH(oh, oti->oti_dev, - rc = ldiskfs_journal_stop(hdl)); - if (rc != 0) - CERROR("Failure to stop transaction: %d\n", rc); - } else { - thandle_put(&oh->ot_super); - } + oh->ot_handle = NULL; + OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl)); + if (rc != 0) + CERROR("%s: failed to stop transaction: rc = %d\n", + osd_name(osd), rc); + } else { + OBD_FREE_PTR(oh); + } /* inform the quota slave device that the transaction is stopping */ qsd_op_end(env, qsd, qtrans); @@ -1158,6 +1188,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, */ wait_event(iobuf->dr_wait, atomic_read(&iobuf->dr_numreqs) == 0); + osd_fini_iobuf(osd, iobuf); if (!rc) rc = iobuf->dr_error; @@ -1171,7 +1202,7 @@ 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; } @@ -1197,8 +1228,8 @@ 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; @@ -1316,6 +1347,7 @@ 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. @@ -1339,13 +1371,19 @@ 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; } /* @@ -1353,8 +1391,16 @@ static void osd_conf_get(const struct lu_env *env, */ static int osd_sync(const struct lu_env *env, struct dt_device *d) { - CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); - return ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + int rc; + + CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); + + rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + + CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n", + LUSTRE_OSD_LDISKFS_NAME, rc); + + return rc; } /** @@ -1412,77 +1458,62 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) RETURN(rc); } -/* - * Concurrency: serialization provided by callers. - */ -static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d, - int mode, unsigned long timeout, __u32 alg, - struct lustre_capa_key *keys) -{ - struct osd_device *dev = osd_dt_dev(d); - ENTRY; - - dev->od_fl_capa = mode; - dev->od_capa_timeout = timeout; - dev->od_capa_alg = alg; - dev->od_capa_keys = keys; - RETURN(0); -} - /** * Note: we do not count into QUOTA here. * 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 = { @@ -1496,7 +1527,6 @@ static const struct dt_device_operations osd_dt_ops = { .dt_sync = osd_sync, .dt_ro = osd_ro, .dt_commit_async = osd_commit_async, - .dt_init_capa_ctxt = osd_init_capa_ctxt, }; static void osd_object_read_lock(const struct lu_env *env, @@ -1568,108 +1598,6 @@ static int osd_object_write_locked(const struct lu_env *env, return obj->oo_owner == env; } -static int capa_is_sane(const struct lu_env *env, - struct osd_device *dev, - struct lustre_capa *capa, - struct lustre_capa_key *keys) -{ - struct osd_thread_info *oti = osd_oti_get(env); - struct lustre_capa *tcapa = &oti->oti_capa; - struct obd_capa *oc; - int i, rc = 0; - ENTRY; - - oc = capa_lookup(dev->od_capa_hash, capa, 0); - if (oc) { - if (capa_is_expired(oc)) { - DEBUG_CAPA(D_ERROR, capa, "expired"); - rc = -ESTALE; - } - capa_put(oc); - RETURN(rc); - } - - if (capa_is_expired_sec(capa)) { - DEBUG_CAPA(D_ERROR, capa, "expired"); - RETURN(-ESTALE); - } - - spin_lock(&capa_lock); - for (i = 0; i < 2; i++) { - if (keys[i].lk_keyid == capa->lc_keyid) { - oti->oti_capa_key = keys[i]; - break; - } - } - spin_unlock(&capa_lock); - - if (i == 2) { - DEBUG_CAPA(D_ERROR, capa, "no matched capa key"); - RETURN(-ESTALE); - } - - rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key); - if (rc) - RETURN(rc); - - if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) { - DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch"); - RETURN(-EACCES); - } - - oc = capa_add(dev->od_capa_hash, capa); - capa_put(oc); - - RETURN(0); -} - -int osd_object_auth(const struct lu_env *env, struct dt_object *dt, - struct lustre_capa *capa, __u64 opc) -{ - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct lu_capainfo *lci; - int rc; - - if (!osd->od_fl_capa) - return 0; - - if (capa == BYPASS_CAPA) - return 0; - - lci = lu_capainfo_get(env); - if (unlikely(lci == NULL)) - return 0; - - if (lci->lci_auth == LC_ID_NONE) - return 0; - - if (capa == NULL) { - CERROR("%s: no capability provided for FID "DFID": rc = %d\n", - osd_name(osd), PFID(fid), -EACCES); - return -EACCES; - } - - if (!lu_fid_eq(fid, &capa->lc_fid)) { - DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with", - PFID(fid)); - return -EACCES; - } - - if (!capa_opc_supported(capa, opc)) { - DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc); - return -EACCES; - } - - rc = capa_is_sane(env, osd, capa, osd->od_capa_keys); - if (rc != 0) { - DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc); - return -EACCES; - } - - return 0; -} - static struct timespec *osd_inode_time(const struct lu_env *env, struct inode *inode, __u64 seconds) { @@ -1684,40 +1612,39 @@ 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 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)); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LINVRNT(osd_invariant(obj)); spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); @@ -1734,9 +1661,11 @@ 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); @@ -1764,27 +1693,28 @@ 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) @@ -1793,47 +1723,49 @@ 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) @@ -1842,20 +1774,19 @@ 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) @@ -1866,11 +1797,16 @@ 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; + + /* Only allow set size for regular file */ + if (!S_ISREG(inode->i_mode)) + bits &= ~(LA_SIZE | LA_BLOCKS); - bits = attr->la_valid; + if (bits == 0) + return 0; if (bits & LA_ATIME) inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); @@ -1889,17 +1825,17 @@ 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 */ @@ -1911,24 +1847,25 @@ 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; } } @@ -1936,21 +1873,20 @@ 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 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)); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; + LASSERT(handle != NULL); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_invariant(obj)); osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); @@ -1980,7 +1916,6 @@ 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) @@ -1995,9 +1930,9 @@ static int osd_attr_set(const struct lu_env *env, return rc; } -struct dentry *osd_child_dentry_get(const struct lu_env *env, - struct osd_object *obj, - const char *name, const int namelen) +static struct dentry *osd_child_dentry_get(const struct lu_env *env, + struct osd_object *obj, + const char *name, const int namelen) { return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); } @@ -2033,14 +1968,13 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, osd_sb(osd)->s_root->d_inode, mode); if (!IS_ERR(inode)) { - /* Do not update file c/mtime in ldiskfs. - * NB: don't need any lock because no contention at this - * early stage */ - inode->i_flags |= S_NOCMTIME; + /* Do not update file c/mtime in ldiskfs. */ + inode->i_flags |= S_NOCMTIME; /* For new created object, it must be consistent, * and it is unnecessary to scrub against it. */ ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB); + obj->oo_inode = inode; result = 0; } else { @@ -2240,7 +2174,7 @@ 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); } @@ -2267,13 +2201,18 @@ static int __osd_object_create(struct osd_thread_info *info, result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof, th); - if (result == 0) { + if (likely(obj->oo_inode != NULL)) { + LASSERT(obj->oo_inode->i_state & I_NEW); + + /* Unlock the inode before attr initialization to avoid + * unnecessary dqget operations. LU-6378 */ + unlock_new_inode(obj->oo_inode); + } + + if (likely(result == 0)) { osd_attr_init(info, obj, attr, dof); osd_object_init0(obj); - /* bz 24037 */ - if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW)) - unlock_new_inode(obj->oo_inode); - } + } /* restore previous umask value */ current->fs->umask = umask; @@ -2304,7 +2243,7 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, } int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, - obd_seq seq, struct lu_seq_range *range) + u64 seq, struct lu_seq_range *range) { struct seq_server_site *ss = osd_seq_site(osd); @@ -2350,12 +2289,10 @@ 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 @@ -2372,7 +2309,7 @@ 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); @@ -2380,22 +2317,23 @@ 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; + 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; + if (dt_object_exists(dt)) + return -EEXIST; - 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_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, @@ -2442,13 +2380,13 @@ 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); } @@ -2562,10 +2500,10 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs. * its inmemory API. */ -void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, - const struct dt_rec *fid) +static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, + 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; @@ -2602,9 +2540,9 @@ 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; @@ -2626,6 +2564,7 @@ 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); @@ -2638,15 +2577,16 @@ 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_NOSCRUB); + unlock_new_inode(local); + /* Set special LMA flag for local agent inode */ rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT); if (rc != 0) { @@ -2655,9 +2595,12 @@ 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); @@ -2706,22 +2649,25 @@ 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); + if (dt_object_exists(dt)) + RETURN(-EEXIST); + + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(osd_write_locked(env, obj)); + LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) /* Quota files can't be created from the kernel any more, @@ -2731,12 +2677,24 @@ 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); - if (result == 0) - result = osd_ea_fid_set(info, obj->oo_inode, fid, + result = __osd_object_create(info, obj, attr, hint, dof, th); + if (result == 0) { + if (fid_is_idif(fid) && + !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) { + struct lu_fid *tfid = &info->oti_fid; + struct ost_id *oi = &info->oti_ostid; + + fid_to_ostid(fid, oi); + ostid_to_fid(tfid, oi, 0); + result = osd_ea_fid_set(info, obj->oo_inode, tfid, + LMAC_FID_ON_OST, 0); + } else { + result = osd_ea_fid_set(info, obj->oo_inode, fid, fid_is_on_ost(info, osd_obj2dev(obj), fid, OI_CHECK_FLD) ? LMAC_FID_ON_OST : 0, 0); + } + } if (result == 0) result = __osd_oi_insert(env, obj, fid, th); @@ -2769,23 +2727,28 @@ 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); + if (!dt_object_exists(dt)) + return -ENOENT; - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); + LINVRNT(osd_invariant(obj)); + 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); 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 @@ -2798,8 +2761,14 @@ 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); @@ -2809,16 +2778,16 @@ 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_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]); @@ -2830,20 +2799,23 @@ 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; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); struct osd_thandle *oh; + if (!dt_object_exists(dt)) + return -ENOENT; + LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(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); @@ -2860,6 +2832,9 @@ 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); @@ -2887,8 +2862,7 @@ static int osd_object_version_get(const struct lu_env *env, * Concurrency: @dt is read locked. */ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, const char *name, - struct lustre_capa *capa) + struct lu_buf *buf, const char *name) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -2899,16 +2873,23 @@ 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); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->getxattr != NULL); return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -2920,16 +2901,36 @@ 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; } @@ -2956,8 +2957,8 @@ static void osd_object_version_set(const struct lu_env *env, * Concurrency: @dt is write locked. */ 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) + const struct lu_buf *buf, const char *name, int fl, + struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; @@ -2976,10 +2977,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, return sizeof(dt_obj_version_t); } - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; - - CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n", + 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); @@ -2989,6 +2987,26 @@ 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); } @@ -2997,39 +3015,45 @@ 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) + const struct lu_buf *buf) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; - 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)); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->listxattr != NULL); - 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_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; } @@ -3038,118 +3062,33 @@ 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) + const char *name, struct thandle *handle) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + int rc; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL); - LASSERT(handle != NULL); + if (!dt_object_exists(dt)) + return -ENOENT; - if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) - return -EACCES; + LASSERT(!dt_object_remote(dt)); + LASSERT(inode->i_op != NULL); + LASSERT(inode->i_op->removexattr != NULL); + LASSERT(handle != NULL); 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; -} - -static struct obd_capa *osd_capa_get(const struct lu_env *env, - struct dt_object *dt, - struct lustre_capa *old, __u64 opc) -{ - struct osd_thread_info *info = osd_oti_get(env); - const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *osd = osd_obj2dev(obj); - struct lustre_capa_key *key = &info->oti_capa_key; - struct lustre_capa *capa = &info->oti_capa; - struct obd_capa *oc; - struct lu_capainfo *lci; - int rc; - ENTRY; - - if (!osd->od_fl_capa) - RETURN(ERR_PTR(-ENOENT)); - - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); - - /* renewal sanity check */ - if (old && osd_object_auth(env, dt, old, opc)) - RETURN(ERR_PTR(-EACCES)); - - lci = lu_capainfo_get(env); - if (unlikely(lci == NULL)) - RETURN(ERR_PTR(-ENOENT)); - - switch (lci->lci_auth) { - 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_flags = LC_ID_PLAIN; - break; - case LC_ID_CONVERT: { - __u32 d[4], s[4]; - - s[0] = obj->oo_inode->i_uid; - cfs_get_random_bytes(&(s[1]), sizeof(__u32)); - s[2] = obj->oo_inode->i_gid; - 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)) - RETURN(ERR_PTR(rc)); - - capa->lc_uid = ((__u64)d[1] << 32) | d[0]; - capa->lc_gid = ((__u64)d[3] << 32) | d[2]; - capa->lc_flags = LC_ID_CONVERT; - break; - } - default: - RETURN(ERR_PTR(-EINVAL)); - } - - capa->lc_fid = *fid; - capa->lc_opc = opc; - capa->lc_flags |= osd->od_capa_alg << 24; - capa->lc_timeout = osd->od_capa_timeout; - capa->lc_expiry = 0; - - oc = capa_lookup(osd->od_capa_hash, capa, 1); - if (oc) { - LASSERT(!capa_is_expired(oc)); - RETURN(oc); - } - - spin_lock(&capa_lock); - *key = osd->od_capa_keys[1]; - spin_unlock(&capa_lock); - - capa->lc_keyid = key->lk_keyid; - capa->lc_expiry = cfs_time_current_sec() + osd->od_capa_timeout; - - rc = capa_hmac(capa->lc_hmac, capa, key->lk_key); - if (rc) { - DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc); - RETURN(ERR_PTR(rc)); - } - - oc = capa_add(osd->od_capa_hash, capa); - RETURN(oc); + rc = inode->i_op->removexattr(dentry, name); + return rc; } -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; @@ -3166,24 +3105,10 @@ 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 - mutex_lock(&inode->i_mutex); -#endif - rc = do_fsync(file, 0); -#ifndef HAVE_FILE_FSYNC_4ARGS - mutex_unlock(&inode->i_mutex); -#endif - RETURN(rc); -} -static int osd_data_get(const struct lu_env *env, struct dt_object *dt, - void **data) -{ - struct osd_object *obj = osd_dt_obj(dt); - ENTRY; + rc = ll_vfs_fsync_range(file, start, end, 0); - *data = (void *)obj->oo_inode; - RETURN(0); + RETURN(rc); } /* @@ -3294,7 +3219,8 @@ 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); @@ -3320,8 +3246,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, static int osd_otable_it_attr_get(const struct lu_env *env, struct dt_object *dt, - struct lu_attr *attr, - struct lustre_capa *capa) + struct lu_attr *attr) { attr->la_valid = 0; return 0; @@ -3352,9 +3277,7 @@ static const struct dt_object_operations osd_obj_ops = { .do_declare_xattr_del = osd_declare_xattr_del, .do_xattr_del = osd_xattr_del, .do_xattr_list = osd_xattr_list, - .do_capa_get = osd_capa_get, .do_object_sync = osd_object_sync, - .do_data_get = osd_data_get, }; /** @@ -3386,9 +3309,7 @@ static const struct dt_object_operations osd_obj_ea_ops = { .do_declare_xattr_del = osd_declare_xattr_del, .do_xattr_del = osd_xattr_del, .do_xattr_list = osd_xattr_list, - .do_capa_get = osd_capa_get, .do_object_sync = osd_object_sync, - .do_data_get = osd_data_get, }; static const struct dt_object_operations osd_obj_otable_it_ops = { @@ -3423,28 +3344,25 @@ 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 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); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3478,7 +3396,7 @@ 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_remote(dt)); LASSERT(handle != NULL); oh = container_of0(handle, struct osd_thandle, ot_super); @@ -3490,8 +3408,8 @@ 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); } @@ -3509,14 +3427,22 @@ 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); @@ -3534,25 +3460,26 @@ 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) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct osd_thandle *oh; + const struct dt_key *key, struct thandle *handle) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct osd_thandle *oh; struct 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); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); + + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(handle != NULL); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3560,9 +3487,6 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, LASSERT(oh->ot_handle != NULL); LASSERT(oh->ot_handle->h_transaction != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); - ll_vfs_dq_init(dir); dentry = osd_child_dentry_get(env, obj, (char *)key, strlen((char *)key)); @@ -3691,25 +3615,23 @@ 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 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); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) - RETURN(-EACCES); + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); ipd = osd_idx_ipd_get(env, bag); if (IS_ERR(ipd)) @@ -3780,27 +3702,28 @@ 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; + const struct dt_rec *rec, + const struct dt_key *key, struct thandle *th, + int ignore_quota) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct iam_path_descr *ipd; + struct osd_thandle *oh; + struct iam_container *bag; + struct osd_thread_info *oti = osd_oti_get(env); + struct iam_rec *iam_rec; + int rc; + ENTRY; - ENTRY; + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - 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_remote(dt)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + bag = &obj->oo_dir->od_container; + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); @@ -3841,9 +3764,9 @@ 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; @@ -3865,6 +3788,29 @@ 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_handle_dirty_metadata(oth->ot_handle, + NULL, bh); + brelse(bh); + } + } + } RETURN(rc); } @@ -3882,11 +3828,11 @@ 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; @@ -3913,9 +3859,19 @@ 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; } @@ -3929,8 +3885,8 @@ 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; @@ -3945,9 +3901,10 @@ 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) { @@ -3960,11 +3917,10 @@ 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); @@ -3992,6 +3948,9 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, if (!fid_is_norm(fid) && !fid_is_igif(fid)) RETURN_EXIT; + if (scrub->os_pos_current > id->oii_ino) + RETURN_EXIT; + again: rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD); if (rc != 0 && rc != -ENOENT) @@ -4013,12 +3972,13 @@ again: } if (!dev->od_noscrub && ++once == 1) { - rc = osd_scrub_start(dev); - LCONSOLE_WARN("%.16s: trigger OI scrub by RPC for "DFID - ", rc = %d [2]\n", - LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, - PFID(fid), rc); - if (rc == 0) + rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN | + SS_CLEAR_FAILOUT); + CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC " + "for "DFID", rc = %d [2]\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, + PFID(fid), rc); + if (rc == 0 || rc == -EALREADY) goto again; } @@ -4050,16 +4010,91 @@ static int osd_fail_fid_lookup(struct osd_thread_info *oti, return rc; } -int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, - struct osd_inode_id *id, const struct lu_fid *fid) +void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd, + struct osd_inode_id *id, const struct lu_fid *fid) { CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid), id->oii_ino, id->oii_gen, info); info->oti_cache.oic_lid = *id; info->oti_cache.oic_fid = *fid; info->oti_cache.oic_dev = osd; +} - return 0; +/** + * 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 = { NULL }; + 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); } /** @@ -4071,39 +4106,38 @@ 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; 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)) { @@ -4116,10 +4150,24 @@ 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))) { + const char *name = (const char *)key; + + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + if (likely(strlen(name) == 2 && + name[0] == '.' && name[1] == '.')) + 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); @@ -4128,16 +4176,8 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, if (osd_remote_fid(env, dev, fid)) GOTO(out, rc = 0); - rc = osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, - fid); - if (rc != 0) - GOTO(out, rc); - if ((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); + osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid); + osd_consistency_check(oti, dev, oic); } else { rc = -ENOENT; } @@ -4160,9 +4200,9 @@ out: * \retval osd_object on success * \retval -ve on error */ -struct osd_object *osd_object_find(const struct lu_env *env, - struct dt_object *dt, - const struct lu_fid *fid) +static struct osd_object *osd_object_find(const struct lu_env *env, + struct dt_object *dt, + const struct lu_fid *fid) { struct lu_device *ludev = dt->do_lu.lo_dev; struct osd_object *child = NULL; @@ -4179,15 +4219,17 @@ 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) { @@ -4232,7 +4274,7 @@ 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); @@ -4241,22 +4283,15 @@ 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) @@ -4290,13 +4325,14 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, * \retval -ve, on error */ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, struct thandle *th, - struct lustre_capa *capa, int ignore_quota) + const struct dt_rec *rec, + const struct dt_key *key, struct thandle *th, + int ignore_quota) { struct osd_object *obj = osd_dt_obj(dt); struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct 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; @@ -4305,16 +4341,16 @@ 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); + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - osd_trans_exec_op(env, th, OSD_OT_INSERT); + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(th != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) - RETURN(-EACCES); + osd_trans_exec_op(env, th, OSD_OT_INSERT); - 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) { @@ -4343,8 +4379,7 @@ 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)); } @@ -4361,7 +4396,10 @@ 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) @@ -4377,32 +4415,33 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, */ static struct dt_it *osd_it_iam_init(const struct lu_env *env, - struct dt_object *dt, - __u32 unused, - struct lustre_capa *capa) + struct dt_object *dt, + __u32 unused) { - struct osd_it_iam *it; - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct lu_object *lo = &dt->do_lu; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - - LASSERT(lu_object_exists(lo)); - - 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); + struct osd_it_iam *it; + struct osd_object *obj = osd_dt_obj(dt); + struct lu_object *lo = &dt->do_lu; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + + if (!dt_object_exists(dt)) + return ERR_PTR(-ENOENT); + + OBD_ALLOC_PTR(it); + if (it == NULL) + return ERR_PTR(-ENOMEM); + + 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 { + OBD_FREE_PTR(it); + return ERR_PTR(-ENOMEM); + } } /** @@ -4411,12 +4450,13 @@ 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; + 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); + 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); + OBD_FREE_PTR(it); } /** @@ -4654,32 +4694,44 @@ 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 osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *oie; + struct file *file; + struct lu_object *lo = &dt->do_lu; + struct dentry *obj_dentry; + ENTRY; - obj_dentry->d_inode = obj->oo_inode; - obj_dentry->d_sb = osd_sb(osd_obj2dev(obj)); - obj_dentry->d_name.hash = 0; + if (!dt_object_exists(dt)) + RETURN(ERR_PTR(-ENOENT)); - 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; + OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS); + if (oie == NULL) + RETURN(ERR_PTR(-ENOMEM)); + obj_dentry = &oie->oie_dentry; + + obj_dentry->d_inode = obj->oo_inode; + obj_dentry->d_sb = osd_sb(osd_obj2dev(obj)); + obj_dentry->d_name.hash = 0; + + oie->oie_rd_dirent = 0; + oie->oie_it_dirent = 0; + oie->oie_dirent = NULL; + if (unlikely(!info->oti_it_ea_buf_used)) { + oie->oie_buf = info->oti_it_ea_buf; + info->oti_it_ea_buf_used = 1; + } else { + OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE); + if (oie->oie_buf == NULL) + RETURN(ERR_PTR(-ENOMEM)); + } + oie->oie_obj = obj; + + file = &oie->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)); /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */ if (attr & LUDA_64BITHASH) file->f_mode = FMODE_64BITHASH; @@ -4691,7 +4743,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, set_file_inode(file, obj->oo_inode); lu_object_get(lo); - RETURN((struct dt_it *) it); + RETURN((struct dt_it *) oie); } /** @@ -4701,14 +4753,20 @@ 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 *oie = (struct osd_it_ea *)di; + struct osd_object *obj = oie->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; + ENTRY; + oie->oie_file.f_op->release(inode, &oie->oie_file); + lu_object_put(env, &obj->oo_dt.do_lu); + if (unlikely(oie->oie_buf != info->oti_it_ea_buf)) + OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE); + else + info->oti_it_ea_buf_used = 0; + OBD_SLAB_FREE_PTR(oie, osd_itea_cachep); + EXIT; } /** @@ -4743,6 +4801,12 @@ 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 @@ -4754,11 +4818,11 @@ 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; @@ -4813,6 +4877,7 @@ 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) @@ -4821,7 +4886,14 @@ 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; @@ -4835,8 +4907,13 @@ 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); @@ -4847,12 +4924,14 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, /*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); } /** @@ -4932,22 +5011,12 @@ 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); + rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); RETURN(rc); } @@ -4996,13 +5065,8 @@ 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); @@ -5010,39 +5074,30 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh, 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); + rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); 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); } @@ -5099,14 +5154,15 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; again: - if (dev->od_dirent_journal) { + if (dev->od_dirent_journal != 0) { 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); } @@ -5147,18 +5203,18 @@ 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)); @@ -5174,7 +5230,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5191,6 +5247,13 @@ 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) { @@ -5199,7 +5262,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5217,6 +5280,13 @@ 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. */ @@ -5231,7 +5301,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5249,6 +5319,13 @@ 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 @@ -5257,6 +5334,13 @@ 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); } } @@ -5270,7 +5354,7 @@ out_journal: if (hlock != NULL) { ldiskfs_htree_unlock(hlock); } else { - if (dev->od_dirent_journal) + if (dev->od_dirent_journal != 0) up_write(&obj->oo_ext_idx_sem); else up_read(&obj->oo_ext_idx_sem); @@ -5299,22 +5383,27 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_it_ea *it = (struct osd_it_ea *)di; struct osd_object *obj = it->oie_obj; struct osd_device *dev = osd_obj2dev(obj); - 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; __u32 ino = it->oie_dirent->oied_ino; 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. */ + if (!fid_is_sane(fid) && + it->oie_dirent->oied_namelen == 2 && + it->oie_dirent->oied_name[0] == '.' && + it->oie_dirent->oied_name[1] == '.') + osd_get_pfid_from_linkea(env, obj, fid); } else { rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr); @@ -5322,10 +5411,21 @@ 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)) - RETURN(-ENOENT); + if (it->oie_dirent->oied_namelen == 2 && + it->oie_dirent->oied_name[0] == '.' && + it->oie_dirent->oied_name[1] == '.') { + /* If the parent is on remote MDT, and there + * is no FID-in-dirent, then we have to get + * the parent FID from the linkEA. */ + if (ino == osd_remote_parent_ino(dev)) + rc = osd_get_pfid_from_linkea(env, obj, + fid); + } else { + if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) + RETURN(-ENOENT); - rc = osd_ea_fid_get(env, obj, ino, fid, id); + rc = osd_ea_fid_get(env, obj, ino, fid, id); + } } else { osd_id_gen(id, ino, OSD_OII_NOGEN); } @@ -5343,17 +5443,31 @@ 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))) - rc = osd_add_oi_cache(oti, dev, id, fid); + if (likely(!(attr & LUDA_IGNORE) && rc == 0)) + osd_add_oi_cache(oti, dev, id, 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); + RETURN(rc > 0 ? 0 : rc); +} - RETURN(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); } /** @@ -5391,6 +5505,9 @@ 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; @@ -5406,8 +5523,7 @@ static int osd_it_ea_load(const struct lu_env *env, * \retval -ve, on error */ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, - struct dt_rec *rec, const struct dt_key *key, - struct lustre_capa *capa) + struct dt_rec *rec, const struct dt_key *key) { struct osd_object *obj = osd_dt_obj(dt); int rc = 0; @@ -5417,9 +5533,6 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, LASSERT(S_ISDIR(obj->oo_inode->i_mode)); LINVRNT(osd_invariant(obj)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) - return -EACCES; - rc = osd_ea_lookup_rec(env, obj, rec, key); if (rc == 0) rc = +1; @@ -5431,23 +5544,24 @@ 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, @@ -5483,11 +5597,14 @@ static void osd_key_fini(const struct lu_context *ctx, { struct osd_thread_info *info = data; + if (info->oti_inode != NULL) + OBD_FREE_PTR(info->oti_inode); if (info->oti_hlock != NULL) ldiskfs_htree_lock_free(info->oti_hlock); OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); lu_buf_free(&info->oti_iobuf.dr_pg_buf); lu_buf_free(&info->oti_iobuf.dr_bl_buf); + lu_buf_free(&info->oti_big_buf); OBD_FREE_PTR(info); } @@ -5523,6 +5640,43 @@ 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; @@ -5533,6 +5687,8 @@ 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); } @@ -5565,7 +5721,7 @@ static int osd_mount(const struct lu_env *env, struct osd_thread_info *info = osd_oti_get(env); struct lu_fid *fid = &info->oti_fid; struct inode *inode; - int rc = 0; + int rc = 0, force_over_128tb = 0; ENTRY; if (o->od_mnt != NULL) @@ -5575,23 +5731,64 @@ static int osd_mount(const struct lu_env *env, RETURN(-E2BIG); strcpy(o->od_mntdev, dev); - OBD_PAGE_ALLOC(__page, GFP_IOFS); - if (__page == NULL) - GOTO(out, rc = -ENOMEM); - str = lustre_cfg_string(cfg, 2); s_flags = simple_strtoul(str, NULL, 0); str = strstr(str, ":"); if (str) lmd_flags = simple_strtoul(str + 1, NULL, 0); opts = lustre_cfg_string(cfg, 3); +#ifdef __BIG_ENDIAN + if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) { + CERROR("%s: device %s extents feature is not guaranteed to " + "work on big-endian systems. Use \"bigendian_extents\" " + "mount option to override.\n", name, dev); + RETURN(-EINVAL); + } +#endif + if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) + force_over_128tb = 1; + + OBD_PAGE_ALLOC(__page, GFP_IOFS); + if (__page == NULL) + GOTO(out, rc = -ENOMEM); page = (unsigned long)page_address(__page); options = (char *)page; *options = '\0'; - if (opts == NULL) - strcat(options, "user_xattr,acl"); - else + if (opts != NULL) { + /* strip out the options for back compatiblity */ + static char *sout[] = { + "mballoc", + "iopen", + "noiopen", + "iopen_nopriv", + "extents", + "noextents", + /* strip out option we processed in osd */ + "bigendian_extents", + "force_over_128tb", + NULL + }; strcat(options, opts); + for (rc = 0, str = options; sout[rc]; ) { + char *op = strstr(str, sout[rc]); + if (op == NULL) { + rc++; + str = options; + continue; + } + if (op == options || *(op - 1) == ',') { + str = op + strlen(sout[rc]); + if (*str == ',' || *str == '\0') { + *str == ',' ? str++ : str; + memmove(op, str, strlen(str) + 1); + } + } + for (str = op; *str != ',' && *str != '\0'; str++) + ; + } + } else { + strncat(options, "user_xattr,acl", 14); + } /* Glom up mount options */ if (*options != '\0') @@ -5614,6 +5811,15 @@ static int osd_mount(const struct lu_env *env, GOTO(out, rc); } + if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) && + force_over_128tb == 0) { + CERROR("%s: device %s LDISKFS does not support filesystems " + "greater than 128TB and can cause data corruption. " + "Use \"force_over_128tb\" mount option to override.\n", + name, dev); + GOTO(out, rc = -EINVAL); + } + #ifdef HAVE_DEV_SET_RDONLY if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { CERROR("%s: underlying device %s is marked as read-only. " @@ -5632,6 +5838,11 @@ 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; lu_local_obj_fid(fid, OSD_FS_ROOT_OID); @@ -5694,28 +5905,26 @@ 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_capa_hash = init_capa_hash(); - if (o->od_capa_hash == NULL) - GOTO(out, rc = -ENOMEM); - o->od_read_cache = 1; 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, rc); } if (server_name_is_ost(o->od_svname)) o->od_is_ost = 1; + o->od_full_scrub_ratio = OFSR_DEFAULT; + o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT; + rc = osd_mount(env, o, cfg); + if (rc != 0) + GOTO(out, rc); + rc = osd_obj_map_init(env, o); if (rc != 0) GOTO(out_mnt, rc); @@ -5729,10 +5938,7 @@ static int osd_device_init0(const struct lu_env *env, if (rc != 0) GOTO(out_site, rc); - /* 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) @@ -5745,7 +5951,8 @@ 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, @@ -5768,8 +5975,6 @@ out_compat: osd_obj_map_fini(o); out_mnt: osd_umount(env, o); -out_capa: - cleanup_capa_hash(o->od_capa_hash); out: return rc; } @@ -5807,7 +6012,6 @@ static struct lu_device *osd_device_free(const struct lu_env *env, struct osd_device *o = osd_dev(d); ENTRY; - cleanup_capa_hash(o->od_capa_hash); /* XXX: make osd top device in order to release reference */ d->ld_site->ls_top_dev = d; lu_site_purge(env, d->ld_site, -1); @@ -5921,19 +6125,49 @@ static int osd_obd_disconnect(struct obd_export *exp) } static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, - struct lu_device *dev) + struct lu_device *dev) { - struct osd_device *osd = osd_dev(dev); - int result = 0; + struct osd_device *osd = osd_dev(dev); + struct lr_server_data *lsd = + &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd; + int result = 0; 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); + } + + if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) { +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0) + if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) { + osd->od_index_in_idif = 1; + } else { + osd->od_index_in_idif = 0; + result = osd_register_proc_index_in_idif(osd); + if (result != 0) + RETURN(result); + } +#else + osd->od_index_in_idif = 1; +#endif + } + + result = osd_fid_init(env, osd); RETURN(result); } +static 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, @@ -5964,37 +6198,50 @@ static const struct lu_device_type_operations osd_device_type_ops = { .ldto_device_fini = osd_device_fini }; -struct lu_device_type osd_device_type = { +static struct lu_device_type osd_device_type = { .ldt_tags = LU_DEVICE_DT, .ldt_name = LUSTRE_OSD_LDISKFS_NAME, .ldt_ops = &osd_device_type_ops, .ldt_ctx_tags = LCT_LOCAL, }; +static int osd_health_check(const struct lu_env *env, struct obd_device *obd) +{ + struct osd_device *osd = osd_dev(obd->obd_lu_dev); + struct super_block *sb = osd_sb(osd); + + return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY); +} + /* * lprocfs legacy support. */ 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, + .o_health_check = osd_health_check, }; static int __init osd_mod_init(void) { int rc; +#if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK) + /* please, try to keep osd_thread_info smaller than a page */ + CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE); +#endif + osd_oi_mod_init(); rc = lu_kmem_init(ldiskfs_caches); 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; @@ -6008,6 +6255,8 @@ static void __exit osd_mod_exit(void) MODULE_AUTHOR("Sun Microsystems, Inc. "); MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit); +module_init(osd_mod_init); +module_exit(osd_mod_exit);