X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=58a1e1a9239b9ec78446bcb8e5e020659cbfb225;hb=4d408c9aed9adaf1f4e2ea87851728a1cf662594;hp=7ad3f8cab2bfe4ba3d7c5f77446da70b3d549e5a;hpb=d35997e6e100acca3effb75ca402c9df7f6252ef;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 7ad3f8c..58a1e1a 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/ @@ -86,6 +86,9 @@ 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, @@ -93,6 +96,11 @@ 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 } }; @@ -349,6 +357,8 @@ 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); @@ -370,10 +380,22 @@ 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) @@ -473,15 +495,14 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) 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); } } @@ -508,6 +529,8 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } 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; @@ -522,7 +545,7 @@ 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); @@ -531,9 +554,13 @@ 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); @@ -633,10 +660,14 @@ 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 @@ -972,8 +1003,8 @@ static struct thandle *osd_trans_create(const struct lu_env *env, /* * 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); @@ -1030,7 +1061,8 @@ 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; @@ -1069,7 +1101,7 @@ 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); @@ -1355,8 +1387,16 @@ 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; } /** @@ -1711,17 +1751,20 @@ static void osd_inode_getattr(const struct lu_env *env, } static int osd_attr_get(const struct lu_env *env, - struct dt_object *dt, - struct lu_attr *attr, - struct lustre_capa *capa) + struct dt_object *dt, + struct lu_attr *attr, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LINVRNT(osd_invariant(obj)); + 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)); + + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) + return -EACCES; spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); @@ -1950,21 +1993,24 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) } static int osd_attr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_attr *attr, - struct thandle *handle, - struct lustre_capa *capa) + struct dt_object *dt, + const struct lu_attr *attr, + struct thandle *handle, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode; + int rc; - LASSERT(handle != NULL); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(osd_invariant(obj)); + 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)); + + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) + return -EACCES; osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET); @@ -2008,9 +2054,9 @@ 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); } @@ -2046,14 +2092,13 @@ 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 { @@ -2280,13 +2325,16 @@ 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 (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); - } + } + + if (obj->oo_inode != NULL) { + LASSERT(obj->oo_inode->i_state & I_NEW); + + unlock_new_inode(obj->oo_inode); + } /* restore previous umask value */ current->fs->umask = umask; @@ -2317,7 +2365,7 @@ 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); @@ -2391,22 +2439,23 @@ 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, @@ -2573,8 +2622,8 @@ 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 lu_fid *fid) +static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param, + const struct lu_fid *fid) { if (!fid_is_namespace_visible(fid) || OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) { @@ -2657,6 +2706,9 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, 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) { @@ -2719,22 +2771,25 @@ 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, @@ -2744,12 +2799,24 @@ 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); @@ -2782,20 +2849,23 @@ 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); @@ -2830,16 +2900,16 @@ 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]); @@ -2851,20 +2921,23 @@ 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); @@ -2923,16 +2996,26 @@ 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); + + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) + return -EACCES; return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len); } @@ -3023,7 +3106,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) return -EACCES; - CDEBUG(D_INODE, DFID" set xattr '%s' with size %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); @@ -3049,6 +3132,10 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, 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); } @@ -3057,36 +3144,40 @@ 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 lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL); + 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); + + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) + return -EACCES; - dentry->d_inode = inode; + dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); + return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len); } static int osd_declare_xattr_del(const struct lu_env *env, - struct dt_object *dt, const char *name, - struct thandle *handle) + struct dt_object *dt, const char *name, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + LASSERT(!dt_object_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]); @@ -3103,29 +3194,33 @@ 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 lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - int rc; + 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); + + if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE)) + return -EACCES; osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); ll_vfs_dq_init(inode); - dentry->d_inode = inode; + dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - rc = inode->i_op->removexattr(dentry, name); - return rc; + rc = inode->i_op->removexattr(dentry, name); + return rc; } static struct obd_capa *osd_capa_get(const struct lu_env *env, @@ -3146,7 +3241,10 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, if (!osd->od_fl_capa) RETURN(ERR_PTR(-ENOENT)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + if (!dt_object_exists(dt)) + RETURN(ERR_PTR(-ENOENT)); + + LASSERT(!dt_object_remote(dt)); LINVRNT(osd_invariant(obj)); /* renewal sanity check */ @@ -3233,17 +3331,7 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt, file->f_op = inode->i_fop; set_file_inode(file, inode); -#ifdef HAVE_FILE_FSYNC_4ARGS - rc = file->f_op->fsync(file, start, end, 0); -#elif defined(HAVE_FILE_FSYNC_2ARGS) - mutex_lock(&inode->i_mutex); - rc = file->f_op->fsync(file, 0); - mutex_unlock(&inode->i_mutex); -#else - mutex_lock(&inode->i_mutex); - rc = file->f_op->fsync(file, dentry, 0); - mutex_unlock(&inode->i_mutex); -#endif + rc = ll_vfs_fsync_range(file, start, end, 0); RETURN(rc); } @@ -3496,28 +3584,29 @@ static int osd_index_declare_iam_delete(const struct lu_env *env, * \retval 0 success * \retval -ve failure */ - static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) + const struct dt_key *key, + struct thandle *handle, + struct lustre_capa *capa) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thandle *oh; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - int rc; + 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; - 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(handle != NULL); + LINVRNT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(handle != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) - RETURN(-EACCES); + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) + RETURN(-EACCES); osd_trans_exec_op(env, handle, OSD_OT_DELETE); @@ -3551,7 +3640,7 @@ 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); @@ -3615,25 +3704,27 @@ static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, * \retval -ve, on error */ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle, - struct lustre_capa *capa) + const struct dt_key *key, struct thandle *handle, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *dir = obj->oo_inode; - struct dentry *dentry; - struct osd_thandle *oh; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *dir = obj->oo_inode; + struct dentry *dentry; + struct osd_thandle *oh; struct ldiskfs_dir_entry_2 *de = NULL; - struct buffer_head *bh; - struct htree_lock *hlock = NULL; + struct buffer_head *bh; + struct htree_lock *hlock = NULL; struct lu_fid *fid = &osd_oti_get(env)->oti_fid; struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); int rc; - ENTRY; + ENTRY; - LINVRNT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(handle != NULL); + 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); @@ -3772,22 +3863,24 @@ out: * \retval -ve failure */ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, - struct dt_rec *rec, const struct dt_key *key, - struct lustre_capa *capa) + struct dt_rec *rec, const struct dt_key *key, + struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - struct osd_thread_info *oti = osd_oti_get(env); - struct iam_iterator *it = &oti->oti_idx_it; - struct iam_rec *iam_rec; - int rc; + 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; - ENTRY; + if (!dt_object_exists(dt)) + RETURN(-ENOENT); - LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); - LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(osd_invariant(obj)); + LASSERT(!dt_object_remote(dt)); + LASSERT(bag->ic_object == obj->oo_inode); if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP)) RETURN(-EACCES); @@ -3861,26 +3954,30 @@ 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, + 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; + 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)); + + bag = &obj->oo_dir->od_container; + LASSERT(bag->ic_object == obj->oo_inode); + LASSERT(th != NULL); - if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) + if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT)) RETURN(-EACCES); osd_trans_exec_op(env, th, OSD_OT_INSERT); @@ -3946,6 +4043,29 @@ 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); } @@ -3994,8 +4114,7 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, dot_dot_fid, NULL, th); } - if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT) || - OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT2)) { + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) { struct lu_fid tfid = *dot_dot_fid; tfid.f_oid--; @@ -4084,6 +4203,9 @@ 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) @@ -4105,12 +4227,13 @@ 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; } @@ -4142,16 +4265,14 @@ 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; } /** @@ -4184,7 +4305,7 @@ static int osd_get_pfid_from_linkea(const struct lu_env *env, struct lu_buf *buf = &oti->oti_big_buf; struct dentry *dentry = &oti->oti_obj_dentry; struct inode *inode = obj->oo_inode; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; int rc; ENTRY; @@ -4240,39 +4361,38 @@ again: * \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)) { @@ -4286,13 +4406,19 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, /* done with de, release bh */ brelse(bh); if (rc != 0) { - if (unlikely(ino == osd_remote_parent_ino(dev))) + if (unlikely(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. */ - rc = osd_get_pfid_from_linkea(env, obj, fid); - else + 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); } @@ -4305,16 +4431,8 @@ 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; } @@ -4337,9 +4455,9 @@ 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; @@ -4478,8 +4596,11 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, int rc; ENTRY; + if (!dt_object_exists(dt)) + RETURN(-ENOENT); + LASSERT(osd_invariant(obj)); - LASSERT(dt_object_exists(dt) && !dt_object_remote(dt)); + LASSERT(!dt_object_remote(dt)); LASSERT(th != NULL); osd_trans_exec_op(env, th, OSD_OT_INSERT); @@ -4556,28 +4677,33 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, __u32 unused, struct lustre_capa *capa) { - 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); + + if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ)) + return ERR_PTR(-EACCES); + + 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); + } } /** @@ -4586,12 +4712,13 @@ 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); } /** @@ -4829,32 +4956,45 @@ static const struct dt_index_operations osd_index_iam_ops = { * */ static struct dt_it *osd_it_ea_init(const struct lu_env *env, - struct dt_object *dt, - __u32 attr, - struct lustre_capa *capa) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thread_info *info = osd_oti_get(env); - struct osd_it_ea *it = &info->oti_it_ea; - struct file *file = &it->oie_file; - struct lu_object *lo = &dt->do_lu; - struct dentry *obj_dentry = &info->oti_it_dentry; - ENTRY; - LASSERT(lu_object_exists(lo)); + struct dt_object *dt, + __u32 attr, + struct lustre_capa *capa) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thread_info *info = osd_oti_get(env); + struct osd_it_ea *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; @@ -4866,7 +5006,7 @@ 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); } /** @@ -4876,14 +5016,20 @@ 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; } /** @@ -5133,7 +5279,7 @@ osd_dirent_update(handle_t *jh, struct super_block *sb, 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); + rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); RETURN(rc); } @@ -5191,7 +5337,7 @@ 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); + rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh); RETURN(rc); } @@ -5271,7 +5417,7 @@ 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); @@ -5320,16 +5466,14 @@ 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; - } else { + 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); } @@ -5349,7 +5493,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5381,7 +5525,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5420,7 +5564,7 @@ again: GOTO(out_inode, rc = 0); } - if (!dev->od_dirent_journal) { + if (dev->od_dirent_journal == 0) { iput(inode); brelse(bh); if (hlock != NULL) @@ -5473,7 +5617,7 @@ 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); @@ -5502,11 +5646,8 @@ 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; @@ -5521,7 +5662,11 @@ static inline int osd_it_ea_rec(const struct lu_env *env, /* If the parent is on remote MDT, and there * is no FID-in-dirent, then we have to get * the parent FID from the linkEA. */ - osd_get_pfid_from_linkea(env, obj, fid); + 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); @@ -5529,19 +5674,21 @@ 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) && - likely(it->oie_dirent->oied_namelen != 2 || - it->oie_dirent->oied_name[0] != '.' || - it->oie_dirent->oied_name[1] != '.')) - RETURN(-ENOENT); - - if (unlikely(ino == osd_remote_parent_ino(dev))) + 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. */ - rc = osd_get_pfid_from_linkea(env, obj, fid); - else + 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); + } } else { osd_id_gen(id, ino, OSD_OII_NOGEN); } @@ -5559,17 +5706,10 @@ 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 (!(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); + if (likely(!(attr & LUDA_IGNORE) && rc == 0)) + osd_add_oi_cache(oti, dev, id, fid); - RETURN(rc); + RETURN(rc > 0 ? 0 : rc); } /** @@ -5998,6 +6138,8 @@ static int osd_device_init0(const struct lu_env *env, 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_capa, rc); @@ -6015,9 +6157,6 @@ 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; - INIT_LIST_HEAD(&o->od_ios_list); /* setup scrub, including OI files initialization */ rc = osd_scrub_setup(env, o); @@ -6031,7 +6170,8 @@ 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, @@ -6207,10 +6347,12 @@ 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) { @@ -6220,13 +6362,28 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, 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); } -int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp, - struct lu_fid *fid, struct md_op_data *op_data) +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); @@ -6263,13 +6420,21 @@ 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. */ @@ -6278,6 +6443,7 @@ static struct obd_ops osd_obd_device_ops = { .o_connect = osd_obd_connect, .o_disconnect = osd_obd_disconnect, .o_fid_alloc = osd_fid_alloc, + .o_health_check = osd_health_check, }; static int __init osd_mod_init(void)