X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=584c8c147c49de53bb2014ad9bee82fce730797d;hb=6df76d3357fc5896b6902399ed7ce6d7c7835f58;hp=13d8b51169ccb65c7b207f0becab27ec008e950d;hpb=cf6b8f530fdd9a494207f59dc8d08c345d1a7570;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index 13d8b51..584c8c1 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -39,12 +39,11 @@ #define DEBUG_SUBSYSTEM S_OSD +#include #include #include #include -#ifdef HAVE_UIDGID_HEADER -# include -#endif +#include /* prerequisite for linux/xattr.h */ #include @@ -140,12 +139,12 @@ int osd_trans_declare_op2rb[] = { static int osd_has_index(const struct osd_object *obj) { - return obj->oo_dt.do_index_ops != NULL; + return obj->oo_dt.do_index_ops != NULL; } static int osd_object_invariant(const struct lu_object *l) { - return osd_invariant(osd_obj(l)); + return osd_invariant(osd_obj(l)); } /* @@ -153,18 +152,19 @@ static int osd_object_invariant(const struct lu_object *l) */ static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o) { - struct osd_thread_info *oti = osd_oti_get(env); - return oti->oti_w_locks > 0 && o->oo_owner == env; + struct osd_thread_info *oti = osd_oti_get(env); + + return oti->oti_w_locks > 0 && o->oo_owner == env; } /* * Concurrency: doesn't access mutable data */ static int osd_root_get(const struct lu_env *env, - struct dt_device *dev, struct lu_fid *f) + struct dt_device *dev, struct lu_fid *f) { - lu_local_obj_fid(f, OSD_FS_ROOT_OID); - return 0; + lu_local_obj_fid(f, OSD_FS_ROOT_OID); + return 0; } /* @@ -182,9 +182,10 @@ static struct osd_idmap_cache * osd_idc_find(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_idmap_cache *idc = oti->oti_ins_cache; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc = oti->oti_ins_cache; int i; + for (i = 0; i < oti->oti_ins_cache_used; i++) { if (!lu_fid_eq(&idc[i].oic_fid, fid)) continue; @@ -201,22 +202,22 @@ static struct osd_idmap_cache * osd_idc_add(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_idmap_cache *idc; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_idmap_cache *idc; int i; if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) { i = oti->oti_ins_cache_size * 2; if (i == 0) i = OSD_INS_CACHE_SIZE; - OBD_ALLOC(idc, sizeof(*idc) * i); + OBD_ALLOC_PTR_ARRAY(idc, i); if (idc == NULL) return ERR_PTR(-ENOMEM); if (oti->oti_ins_cache != NULL) { memcpy(idc, oti->oti_ins_cache, oti->oti_ins_cache_used * sizeof(*idc)); - OBD_FREE(oti->oti_ins_cache, - oti->oti_ins_cache_used * sizeof(*idc)); + OBD_FREE_PTR_ARRAY(oti->oti_ins_cache, + oti->oti_ins_cache_used); } oti->oti_ins_cache = idc; oti->oti_ins_cache_size = i; @@ -284,6 +285,76 @@ osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd, return idc; } +static void osd_idc_dump_lma(const struct lu_env *env, + struct osd_device *osd, + unsigned long ino, + bool check_in_oi) +{ + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + const struct lu_fid *fid; + struct osd_inode_id lid; + struct inode *inode; + int rc; + + inode = osd_ldiskfs_iget(osd_sb(osd), ino); + if (IS_ERR(inode)) { + CERROR("%s: can't get inode %lu: rc = %d\n", + osd->od_svname, ino, (int)PTR_ERR(inode)); + return; + } + if (is_bad_inode(inode)) { + CERROR("%s: bad inode %lu\n", osd->od_svname, ino); + goto put; + } + rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa); + if (rc) { + CERROR("%s: can't get LMA for %lu: rc = %d\n", + osd->od_svname, ino, rc); + goto put; + } + fid = &loa->loa_lma.lma_self_fid; + LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname, + PFID(fid), ino, (unsigned)inode->i_generation); + if (!check_in_oi) + goto put; + rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0); + if (rc) { + CERROR("%s: can't lookup "DFID": rc = %d\n", + osd->od_svname, PFID(fid), rc); + goto put; + } + LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname, + PFID(fid), lid.oii_ino, lid.oii_gen); +put: + iput(inode); +} + +static void osd_idc_dump_debug(const struct lu_env *env, + struct osd_device *osd, + const struct lu_fid *fid, + unsigned long ino1, + unsigned long ino2) +{ + struct osd_inode_id lid; + + int rc; + + rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0); + if (!rc) { + LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", + osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen); + osd_idc_dump_lma(env, osd, lid.oii_ino, false); + } else { + CERROR("%s: can't lookup "DFID": rc = %d\n", + osd->od_svname, PFID(fid), rc); + } + if (ino1) + osd_idc_dump_lma(env, osd, ino1, true); + if (ino2) + osd_idc_dump_lma(env, osd, ino2, true); +} + /* * lookup mapping for given FID and fill it from the given object. * the object is lolcal by definition. @@ -292,8 +363,8 @@ static int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd, struct osd_object *obj) { - const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); - struct osd_idmap_cache *idc; + const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu); + struct osd_idmap_cache *idc; idc = osd_idc_find(env, osd, fid); LASSERT(!IS_ERR(idc)); @@ -301,7 +372,12 @@ static int osd_idc_find_and_init(const struct lu_env *env, if (obj->oo_inode == NULL) return 0; if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) { - LASSERT(idc->oic_lid.oii_ino == 0); + if (idc->oic_lid.oii_ino) { + osd_idc_dump_debug(env, osd, fid, + idc->oic_lid.oii_ino, + obj->oo_inode->i_ino); + return -EINVAL; + } idc->oic_lid.oii_ino = obj->oo_inode->i_ino; idc->oic_lid.oii_gen = obj->oo_inode->i_generation; } @@ -335,14 +411,14 @@ static int osd_idc_find_and_init(const struct lu_env *env, * life-cycle. */ static struct lu_object *osd_object_alloc(const struct lu_env *env, - const struct lu_object_header *hdr, - struct lu_device *d) + const struct lu_object_header *hdr, + struct lu_device *d) { - struct osd_object *mo; + struct osd_object *mo; - OBD_ALLOC_PTR(mo); - if (mo != NULL) { - struct lu_object *l; + OBD_ALLOC_PTR(mo); + if (mo != NULL) { + struct lu_object *l; struct lu_object_header *h; struct osd_device *o = osd_dev(d); @@ -369,10 +445,9 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env, init_rwsem(&mo->oo_ext_idx_sem); spin_lock_init(&mo->oo_guard); INIT_LIST_HEAD(&mo->oo_xattr_list); - return l; - } else { - return NULL; - } + return l; + } + return NULL; } int osd_get_lma(struct osd_thread_info *info, struct inode *inode, @@ -392,12 +467,11 @@ int osd_get_lma(struct osd_thread_info *info, struct inode *inode, lustre_loa_swab(loa, true); /* Check LMA compatibility */ if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) { - CWARN("%s: unsupported incompat LMA feature(s) %#x " - "for fid = "DFID", ino = %lu\n", + rc = -EOPNOTSUPP; + CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n", osd_ino2name(inode), lma->lma_incompat & ~LMA_INCOMPAT_SUPP, - PFID(&lma->lma_self_fid), inode->i_ino); - rc = -EOPNOTSUPP; + PFID(&lma->lma_self_fid), inode->i_ino, rc); } } else if (rc == 0) { rc = -ENODATA; @@ -415,12 +489,14 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, int rc; struct inode *inode = NULL; - /* if we look for an inode withing a running - * transaction, then we risk to deadlock */ - /* osd_dirent_check_repair() breaks this */ - /*LASSERT(current->journal_info == NULL);*/ + /* + * if we look for an inode withing a running + * transaction, then we risk to deadlock + * osd_dirent_check_repair() breaks this + */ + /* LASSERT(current->journal_info == NULL); */ - inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); + inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n", id->oii_ino, PTR_ERR(inode)); @@ -432,16 +508,19 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, iput(inode); inode = ERR_PTR(-ESTALE); } else if (inode->i_nlink == 0) { - /* due to parallel readdir and unlink, - * we can have dead inode here. */ + /* + * due to parallel readdir and unlink, + * we can have dead inode here. + */ CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino); iput(inode); inode = ERR_PTR(-ESTALE); } else if (is_bad_inode(inode)) { - CWARN("%s: bad inode: ino = %u\n", - osd_dev2name(dev), id->oii_ino); + rc = -ENOENT; + CWARN("%s: bad inode: ino = %u: rc = %d\n", + osd_dev2name(dev), id->oii_ino, rc); iput(inode); - inode = ERR_PTR(-ENOENT); + inode = ERR_PTR(rc); } else if ((rc = osd_attach_jinode(inode))) { iput(inode); inode = ERR_PTR(rc); @@ -450,11 +529,13 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, if (id->oii_gen == OSD_OII_NOGEN) osd_id_gen(id, inode->i_ino, inode->i_generation); - /* Do not update file c/mtime in ldiskfs. + /* + * Do not update file c/mtime in ldiskfs. * NB: we don't have any lock to protect this because we don't * have reference on osd_object now, but contention with * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ + * between if (...) and set S_NOCMTIME. + */ if (!(inode->i_flags & S_NOCMTIME)) inode->i_flags |= S_NOCMTIME; } @@ -472,6 +553,7 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd, struct lustre_ost_attrs *loa = &info->oti_ost_attrs; struct inode *parent = child->d_parent->d_inode; struct lu_fid *fid = NULL; + char fidstr[FID_LEN + 1] = "unknown"; rc2 = osd_get_lma(info, parent, child->d_parent, loa); if (!rc2) { @@ -488,19 +570,18 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd, } if (fid != NULL) - CWARN("%s: directory (inode: %lu, FID: "DFID") %s " - "maximum entry limit\n", - osd_name(osd), parent->i_ino, PFID(fid), - rc == -ENOSPC ? "has reached" : "is approaching"); - else - CWARN("%s: directory (inode: %lu, FID: unknown) %s " - "maximum entry limit\n", - osd_name(osd), parent->i_ino, - rc == -ENOSPC ? "has reached" : "is approaching"); + snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid)); + + /* below message is checked in sanity.sh test_129 */ + if (rc == -ENOSPC) { + CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n", + osd_name(osd), parent->i_ino, fidstr); + } else { + rc = 0; /* ignore such error now */ + CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n", + osd_name(osd), parent->i_ino, fidstr); + } - /* ignore such error now */ - if (rc == -ENOBUFS) - rc = 0; } return rc; @@ -542,14 +623,17 @@ static struct inode *osd_iget_check(struct osd_thread_info *info, { struct inode *inode; int rc = 0; + ENTRY; - /* The cached OI mapping is trustable. If we cannot locate the inode + /* + * The cached OI mapping is trustable. If we cannot locate the inode * via the cached OI mapping, then return the failure to the caller - * directly without further OI checking. */ + * directly without further OI checking. + */ again: - inode = ldiskfs_iget(osd_sb(dev), id->oii_ino); + inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino); if (IS_ERR(inode)) { rc = PTR_ERR(inode); if (!trusted && (rc == -ENOENT || rc == -ESTALE)) @@ -603,7 +687,8 @@ check_oi: LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc); rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); - /* XXX: There are four possible cases: + /* + * XXX: There are four possible cases: * 1. rc = 0. * Backup/restore caused the OI invalid. * 2. rc = 0. @@ -622,16 +707,21 @@ check_oi: * if the OI files are consistent but may contain * stale OI mappings because of case 2, if iget() * returns -ENOENT or -ESTALE, then it should be - * the case 2. */ + * the case 2. + */ if (rc != 0) - /* If the OI mapping was in OI file before the + /* + * If the OI mapping was in OI file before the * osd_iget_check(), but now, it is disappear, * then it must be removed by race. That is a - * normal race case. */ + * normal race case. + */ GOTO(put, rc); - /* It is the OI scrub updated the OI mapping by race. - * The new OI mapping must be valid. */ + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ if (saved_ino != id->oii_ino || (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) { if (!IS_ERR(inode)) @@ -644,10 +734,12 @@ check_oi: if (IS_ERR(inode)) { if (dev->od_scrub.os_scrub.os_file.sf_flags & SF_INCONSISTENT) - /* It still can be the case 2, but we cannot + /* + * It still can be the case 2, but we cannot * distinguish it from the case 1. So return * -EREMCHG to block current operation until - * OI scrub rebuilt the OI mappings. */ + * OI scrub rebuilt the OI mappings. + */ rc = -EREMCHG; else rc = -ENOENT; @@ -663,11 +755,13 @@ check_oi: if (id->oii_gen == OSD_OII_NOGEN) osd_id_gen(id, inode->i_ino, inode->i_generation); - /* Do not update file c/mtime in ldiskfs. + /* + * Do not update file c/mtime in ldiskfs. * NB: we don't have any lock to protect this because we don't * have reference on osd_object now, but contention with * another lookup + attr_set can't happen in the tiny window - * between if (...) and set S_NOCMTIME. */ + * between if (...) and set S_NOCMTIME. + */ if (!(inode->i_flags & S_NOCMTIME)) inode->i_flags |= S_NOCMTIME; } @@ -686,29 +780,30 @@ put: } /** - * \retval +v: new filter_fid, does not contain self-fid - * \retval 0: filter_fid_old, contains self-fid + * \retval +v: new filter_fid does not contain self-fid + * \retval 0: filter_fid_18_23, contains self-fid * \retval -v: other failure cases */ int osd_get_idif(struct osd_thread_info *info, struct inode *inode, struct dentry *dentry, struct lu_fid *fid) { - struct filter_fid_old *ff = &info->oti_ff; - struct ost_id *ostid = &info->oti_ostid; - int rc; + struct filter_fid *ff = &info->oti_ff; + struct ost_id *ostid = &info->oti_ostid; + int rc; rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff)); - if (rc == sizeof(*ff)) { - rc = 0; - ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq)); - rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid)); + if (rc == sizeof(struct filter_fid_18_23)) { + struct filter_fid_18_23 *ff_old = (void *)ff; + + ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq)); + rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid)); /* * XXX: use 0 as the index for compatibility, the caller will - * handle index related issues when necessary. + * handle index related issues when necessary. */ if (!rc) ostid_to_fid(fid, ostid, 0); - } else if (rc == sizeof(struct filter_fid)) { + } else if (rc >= (int)sizeof(struct filter_fid_24_29)) { rc = 1; } else if (rc >= 0) { rc = -EINVAL; @@ -722,7 +817,7 @@ static int osd_lma_self_repair(struct osd_thread_info *info, const struct lu_fid *fid, __u32 compat) { handle_t *jh; - int rc; + int rc; LASSERT(current->journal_info == NULL); @@ -745,15 +840,16 @@ static int osd_lma_self_repair(struct osd_thread_info *info, static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) { - struct osd_thread_info *info = osd_oti_get(env); - struct osd_device *osd = osd_obj2dev(obj); - struct lustre_ost_attrs *loa = &info->oti_ost_attrs; - struct lustre_mdt_attrs *lma = &loa->loa_lma; - struct inode *inode = obj->oo_inode; - struct dentry *dentry = &info->oti_obj_dentry; - struct lu_fid *fid = NULL; - const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct osd_device *osd = osd_obj2dev(obj); + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; + struct inode *inode = obj->oo_inode; + struct dentry *dentry = &info->oti_obj_dentry; + struct lu_fid *fid = NULL; + const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu); + int rc; + ENTRY; rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, @@ -761,12 +857,14 @@ 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_index_in_idif)) { - /* For the given OST-object, if it has neither LMA nor + 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. */ + * the LMA. + */ osd_lma_self_repair(info, osd, inode, rfid, LMAC_FID_ON_OST); RETURN(0); @@ -804,10 +902,12 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) struct lu_fid *fid1 = &info->oti_fid3; __u32 idx = fid_idif_ost_idx(rfid); - /* For old IDIF, the OST index is not part of the IDIF, + /* + * For old IDIF, the OST index is not part of the IDIF, * Means that different OSTs may have the same IDIFs. * Under such case, we need to make some compatible - * check to make sure to trigger OI scrub properly. */ + * check to make sure to trigger OI scrub properly. + */ if (idx != 0 && fid_idif_ost_idx(fid) == 0) { /* Given @rfid is new, LMA is old. */ fid_to_ostid(fid, oi); @@ -829,19 +929,17 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj) } struct osd_check_lmv_buf { -#ifdef HAVE_DIR_CONTEXT /* please keep it as first member */ - struct dir_context ctx; -#endif - struct osd_thread_info *oclb_info; - struct osd_device *oclb_dev; - struct osd_idmap_cache *oclb_oic; - int oclb_items; - bool oclb_found; + struct dir_context ctx; + struct osd_thread_info *oclb_info; + struct osd_device *oclb_dev; + struct osd_idmap_cache *oclb_oic; + int oclb_items; + bool oclb_found; }; /** - * It is called internally by ->readdir() to filter out the + * It is called internally by ->iterate*() to filter out the * local slave object's FID of the striped directory. * * \retval 1 found the local slave's FID @@ -854,7 +952,7 @@ static int osd_stripe_dir_filldir(struct dir_context *buf, static int osd_stripe_dir_filldir(void *buf, #endif const char *name, int namelen, - loff_t offset, __u64 ino, unsigned d_type) + loff_t offset, __u64 ino, unsigned int d_type) { struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf; struct osd_thread_info *oti = oclb->oclb_info; @@ -893,7 +991,8 @@ static int osd_stripe_dir_filldir(void *buf, return 1; } -/* When lookup item under striped directory, we need to locate the master +/* + * When lookup item under striped directory, we need to locate the master * MDT-object of the striped directory firstly, then the client will send * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID * and the item's name. If the system is restored from MDT file level backup, @@ -926,25 +1025,24 @@ static int osd_stripe_dir_filldir(void *buf, * slave MDT-objects information are stored as sub-directories with the name * "${FID}:${index}". Then when find the local slave MDT-object, its OI * mapping will be recorded. Then subsequent osd_fid_lookup() will know - * the correct OI mapping for the slave MDT-object. */ + * the correct OI mapping for the slave MDT-object. + */ static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev, struct inode *inode, struct osd_idmap_cache *oic) { struct lu_buf *buf = &oti->oti_big_buf; struct dentry *dentry = &oti->oti_obj_dentry; - struct file *filp = &oti->oti_file; - const struct file_operations *fops; + struct file *filp; struct lmv_mds_md_v1 *lmv1; struct osd_check_lmv_buf oclb = { -#ifdef HAVE_DIR_CONTEXT .ctx.actor = osd_stripe_dir_filldir, -#endif .oclb_info = oti, .oclb_dev = dev, .oclb_oic = oic, .oclb_found = false, }; int rc = 0; + ENTRY; again: @@ -979,35 +1077,23 @@ again: if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1) GOTO(out, rc = 0); - fops = inode->i_fop; - dentry->d_inode = inode; - dentry->d_sb = inode->i_sb; - filp->f_pos = 0; - filp->f_path.dentry = dentry; - filp->f_mode = FMODE_64BITHASH; - filp->f_mapping = inode->i_mapping; - filp->f_op = fops; - filp->private_data = NULL; - set_file_inode(filp, inode); + filp = osd_quasi_file(oti->oti_env, inode); + rc = osd_security_file_alloc(filp); + if (rc) + goto out; do { oclb.oclb_items = 0; -#ifdef HAVE_DIR_CONTEXT - oclb.ctx.pos = filp->f_pos; - rc = fops->iterate(filp, &oclb.ctx); - filp->f_pos = oclb.ctx.pos; -#else - rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir); -#endif + rc = iterate_dir(filp, &oclb.ctx); } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found && filp->f_pos != LDISKFS_HTREE_EOF_64BIT); - fops->release(inode, filp); + inode->i_fop->release(inode, filp); out: if (rc < 0) - CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u," - DFID": rc = %d\n", osd_ino2name(inode), - inode->i_ino, inode->i_generation, + CDEBUG(D_LFSCK, + "%s: cannot check LMV, ino = %lu/%u "DFID": rc = %d\n", + osd_ino2name(inode), inode->i_ino, inode->i_generation, PFID(&oic->oic_fid), rc); else rc = 0; @@ -1036,11 +1122,18 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, bool trusted = true; bool updated = false; bool checked = false; + ENTRY; LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid)); + + if (fid_is_sane(fid) == 0) { + CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name, + PFID(fid)); + dump_stack(); + RETURN(-EINVAL); + } dev = osd_dev(ldev); scrub = &dev->od_scrub.os_scrub; @@ -1052,11 +1145,13 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT)) RETURN(-ENOENT); - /* For the object is created as locking anchor, or for the object to + /* + * For the object is created as locking anchor, or for the object to * be created on disk. No need to osd_oi_lookup() at here because FID * shouldn't never be re-used, if it's really a duplicate FID from * unexpected reason, we should be able to detect it later by calling - * do_create->osd_oi_insert(). */ + * do_create->osd_oi_insert(). + */ if (conf && conf->loc_flags & LOC_F_NEW) GOTO(out, result = 0); @@ -1074,8 +1169,10 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, goto iget; } - /* The OI mapping in the OI file can be updated by the OI scrub - * when we locate the inode via FID. So it may be not trustable. */ + /* + * The OI mapping in the OI file can be updated by the OI scrub + * when we locate the inode via FID. So it may be not trustable. + */ trusted = false; /* Search order: 3. OI files. */ @@ -1083,13 +1180,19 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (result == -ENOENT) { if (!(fid_is_norm(fid) || fid_is_igif(fid)) || fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) || - !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), + !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)) GOTO(out, result = 0); goto trigger; } + /* -ESTALE is returned if inode of OST object doesn't exist */ + if (result == -ESTALE && + fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) { + GOTO(out, result = 0); + } + if (result) GOTO(out, result); @@ -1117,7 +1220,8 @@ iget: GOTO(out, result); trigger: - /* We still have chance to get the valid inode: for the + /* + * We still have chance to get the valid inode: for the * object which is referenced by remote name entry, the * object on the local MDT will be linked under the dir * of "/REMOTE_PARENT_DIR" with its FID string as name. @@ -1131,8 +1235,9 @@ trigger: * more, just lookup "/REMOTE_PARENT_DIR". Usually, it * only happened for the RPC from other MDT during the * OI scrub, or for the client side RPC with FID only, - * such as FID to path, or from old connected client. */ - if (!remote && !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) { + * such as FID to path, or from old connected client. + */ + if (!remote) { rc1 = osd_lookup_in_remote_parent(info, dev, fid, id); if (!rc1) { remote = true; @@ -1143,17 +1248,19 @@ trigger: } } - if (thread_is_running(&scrub->os_thread)) { + if (scrub->os_running) { if (scrub->os_partial_scan && !scrub->os_in_join) goto join; - if (IS_ERR_OR_NULL(inode) || result) + osd_add_oi_cache(info, dev, id, fid); + if (IS_ERR_OR_NULL(inode) || result) { + osd_oii_insert(dev, oic, result == -ENOENT); GOTO(out, result = -EINPROGRESS); + } LASSERT(remote); LASSERT(obj->oo_inode == inode); - osd_add_oi_cache(info, dev, id, fid); osd_oii_insert(dev, oic, true); goto found; } @@ -1176,13 +1283,15 @@ join: if (rc1 && rc1 != -EALREADY) GOTO(out, result = -EREMCHG); - if (IS_ERR_OR_NULL(inode) || result) + osd_add_oi_cache(info, dev, id, fid); + if (IS_ERR_OR_NULL(inode) || result) { + osd_oii_insert(dev, oic, result == -ENOENT); GOTO(out, result = -EINPROGRESS); + } LASSERT(remote); LASSERT(obj->oo_inode == inode); - osd_add_oi_cache(info, dev, id, fid); osd_oii_insert(dev, oic, true); goto found; @@ -1205,30 +1314,38 @@ check_lma: saved_gen = inode->i_generation; if (unlikely(result == -ENODATA)) { - /* If the OI scrub updated the OI mapping by race, it - * must be valid. Trust the inode that has no LMA EA. */ + /* + * If the OI scrub updated the OI mapping by race, it + * must be valid. Trust the inode that has no LMA EA. + */ if (updated) goto found; result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); if (!result) { - /* The OI mapping is still there, the inode is still - * valid. It is just becaues the inode has no LMA EA. */ + /* + * The OI mapping is still there, the inode is still + * valid. It is just becaues the inode has no LMA EA. + */ if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) goto found; - /* It is the OI scrub updated the OI mapping by race. - * The new OI mapping must be valid. */ + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ trusted = true; updated = true; goto iget; } - /* "result == -ENOENT" means that the OI mappinghas been + /* + * "result == -ENOENT" means that the OI mappinghas been * removed by race, so the inode belongs to other object. * - * Others error can be returned directly. */ + * Others error can be returned directly. + */ if (result == -ENOENT) { LASSERT(trusted); @@ -1242,9 +1359,24 @@ check_lma: LASSERT(!updated); + /* + * if two OST objects map to the same inode, and inode mode is + * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's + * reserved by precreate, and not written yet, in this case, don't + * set inode for the object whose FID mismatch, so that it can create + * inode and not block precreate. + */ + if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) && + inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) { + obj->oo_inode = NULL; + GOTO(out, result = 0); + } + result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD); - /* "result == -ENOENT" means the cached OI mapping has been removed - * from the OI file by race, above inode belongs to other object. */ + /* + * "result == -ENOENT" means the cached OI mapping has been removed + * from the OI file by race, above inode belongs to other object. + */ if (result == -ENOENT) { LASSERT(trusted); @@ -1260,8 +1392,10 @@ check_lma: goto trigger; } - /* It is the OI scrub updated the OI mapping by race. - * The new OI mapping must be valid. */ + /* + * It is the OI scrub updated the OI mapping by race. + * The new OI mapping must be valid. + */ trusted = true; updated = true; goto iget; @@ -1322,10 +1456,10 @@ out: */ static void osd_object_init0(struct osd_object *obj) { - LASSERT(obj->oo_inode != NULL); - obj->oo_dt.do_body_ops = &osd_body_ops; - obj->oo_dt.do_lu.lo_header->loh_attr |= - (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); + LASSERT(obj->oo_inode != NULL); + obj->oo_dt.do_body_ops = &osd_body_ops; + obj->oo_dt.do_lu.lo_header->loh_attr |= + (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT)); } /* @@ -1340,6 +1474,13 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, LINVRNT(osd_invariant(obj)); + if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) && + cfs_fail_val == 2) { + struct osd_thread_info *info = osd_oti_get(env); + struct osd_idmap_cache *oic = &info->oti_cache; + /* invalidate thread cache */ + memset(&oic->oic_fid, 0, sizeof(oic->oic_fid)); + } if (fid_is_otable_it(&l->lo_header->loh_fid)) { obj->oo_dt.do_ops = &osd_obj_otable_it_ops; l->lo_header->loh_attr |= LOHA_EXISTS; @@ -1359,21 +1500,26 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, result = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, loa); if (!result) { - /* Convert LMAI flags to lustre LMA flags - * and cache it to oo_lma_flags */ + /* + * Convert LMAI flags to lustre LMA flags + * and cache it to oo_lma_flags + */ obj->oo_lma_flags = lma_to_lustre_flags(loa->loa_lma.lma_incompat); } else if (result == -ENODATA) { result = 0; } } + obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET; LINVRNT(osd_invariant(obj)); return result; } -/* The first part of oxe_buf is xattr name, and is '\0' terminated. - * The left part is for value, binary mode. */ +/* + * The first part of oxe_buf is xattr name, and is '\0' terminated. + * The left part is for value, binary mode. + */ struct osd_xattr_entry { struct list_head oxe_list; size_t oxe_len; @@ -1390,7 +1536,6 @@ static int osd_oxc_get(struct osd_object *obj, const char *name, struct osd_xattr_entry *oxe = NULL; size_t namelen = strlen(name); int rc; - ENTRY; rcu_read_lock(); list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) { @@ -1418,7 +1563,6 @@ static int osd_oxc_get(struct osd_object *obj, const char *name, GOTO(out, rc = -ERANGE); memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc); - EXIT; out: rcu_read_unlock(); @@ -1518,11 +1662,10 @@ static void osd_object_free(const struct lu_env *env, struct lu_object *l) dt_object_fini(&obj->oo_dt); if (obj->oo_hl_head != NULL) ldiskfs_htree_lock_head_free(obj->oo_hl_head); + /* obj doesn't contain an lu_object_header, so we don't need call_rcu */ OBD_FREE_PTR(obj); - if (unlikely(h)) { - lu_object_header_fini(h); - OBD_FREE_PTR(h); - } + if (unlikely(h)) + lu_object_header_free(h); } /* @@ -1531,32 +1674,22 @@ static void osd_object_free(const struct lu_env *env, struct lu_object *l) */ static void osd_index_fini(struct osd_object *o) { - struct iam_container *bag; - - if (o->oo_dir != NULL) { - bag = &o->oo_dir->od_container; - if (o->oo_inode != NULL) { - if (bag->ic_object == o->oo_inode) - iam_container_fini(bag); - } - OBD_FREE_PTR(o->oo_dir); - o->oo_dir = NULL; - } -} + struct iam_container *bag; -/* - * Concurrency: no concurrent access is possible that late in object - * life-cycle (for all existing callers, that is. New callers have to provide - * their own locking.) - */ -static int osd_inode_unlinked(const struct inode *inode) -{ - return inode->i_nlink == 0; + if (o->oo_dir != NULL) { + bag = &o->oo_dir->od_container; + if (o->oo_inode != NULL) { + if (bag->ic_object == o->oo_inode) + iam_container_fini(bag); + } + OBD_FREE_PTR(o->oo_dir); + o->oo_dir = NULL; + } } enum { - OSD_TXN_OI_DELETE_CREDITS = 20, - OSD_TXN_INODE_DELETE_CREDITS = 20 + OSD_TXN_OI_DELETE_CREDITS = 20, + OSD_TXN_INODE_DELETE_CREDITS = 20 }; /* @@ -1600,9 +1733,9 @@ static void __osd_th_check_slow(void *oth, struct osd_device *dev, if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) { CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n", - oth, now, alloced, started, closed); - libcfs_debug_dumpstack(NULL); - } + oth, now, alloced, started, closed); + libcfs_debug_dumpstack(NULL); + } } #define OSD_CHECK_SLOW_TH(oth, dev, expr) \ @@ -1638,20 +1771,20 @@ static int osd_param_is_not_sane(const struct osd_device *dev, * Concurrency: shouldn't matter. */ static void osd_trans_commit_cb(struct super_block *sb, - struct ldiskfs_journal_cb_entry *jcb, int error) + struct ldiskfs_journal_cb_entry *jcb, int error) { - struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb); - struct thandle *th = &oh->ot_super; - struct lu_device *lud = &th->th_dev->dd_lu_dev; - struct dt_txn_commit_cb *dcb, *tmp; - - LASSERT(oh->ot_handle == NULL); + struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb); + struct thandle *th = &oh->ot_super; + struct lu_device *lud = &th->th_dev->dd_lu_dev; + struct osd_device *osd = osd_dev(lud); + struct dt_txn_commit_cb *dcb, *tmp; - if (error) - CERROR("transaction @0x%p commit error: %d\n", th, error); + LASSERT(oh->ot_handle == NULL); - dt_txn_hook_commit(th); + if (error) + CERROR("transaction @0x%p commit error: %d\n", th, error); + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40); /* call per-transaction callbacks if any */ list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list, dcb_linkage) { @@ -1663,24 +1796,21 @@ static void osd_trans_commit_cb(struct super_block *sb, } lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th); - lu_device_put(lud); - th->th_dev = NULL; + if (atomic_dec_and_test(&osd->od_commit_cb_in_flight)) + wake_up(&osd->od_commit_cb_done); + th->th_dev = NULL; OBD_FREE_PTR(oh); } -#ifndef HAVE_SB_START_WRITE -# define sb_start_write(sb) do {} while (0) -# define sb_end_write(sb) do {} while (0) -#endif - static struct thandle *osd_trans_create(const struct lu_env *env, struct dt_device *d) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_thandle *oh; - struct thandle *th; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_thandle *oh; + struct thandle *th; + ENTRY; if (d->dd_rdonly) { @@ -1696,7 +1826,7 @@ static struct thandle *osd_trans_create(const struct lu_env *env, sb_start_write(osd_sb(osd_dt_dev(d))); - OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS); + OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS); if (!oh) { sb_end_write(osd_sb(osd_dt_dev(d))); RETURN(ERR_PTR(-ENOMEM)); @@ -1708,8 +1838,10 @@ static struct thandle *osd_trans_create(const struct lu_env *env, th->th_dev = d; th->th_result = 0; oh->ot_credits = 0; + oh->oh_declared_ext = 0; INIT_LIST_HEAD(&oh->ot_commit_dcb_list); INIT_LIST_HEAD(&oh->ot_stop_dcb_list); + INIT_LIST_HEAD(&oh->ot_trunc_locks); osd_th_alloced(oh); memset(oti->oti_declare_ops, 0, @@ -1726,10 +1858,10 @@ static struct thandle *osd_trans_create(const struct lu_env *env, void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_thandle *oh; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_thandle *oh; - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh != NULL); CWARN(" create: %u/%u/%u, destroy: %u/%u/%u\n", @@ -1778,33 +1910,35 @@ void osd_trans_dump_creds(const struct lu_env *env, 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); - handle_t *jh; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_device *dev = osd_dt_dev(d); + handle_t *jh; + struct osd_thandle *oh; + int rc; - ENTRY; + ENTRY; - LASSERT(current->journal_info == NULL); + LASSERT(current->journal_info == NULL); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh != NULL); - LASSERT(oh->ot_handle == NULL); + oh = container_of(th, struct osd_thandle, ot_super); + LASSERT(oh != NULL); + LASSERT(oh->ot_handle == NULL); - rc = dt_txn_hook_start(env, d, th); - if (rc != 0) - GOTO(out, rc); + rc = dt_txn_hook_start(env, d, th); + if (rc != 0) + GOTO(out, rc); if (unlikely(osd_param_is_not_sane(dev, th))) { static unsigned long last_printed; static int last_credits; - /* don't make noise on a tiny testing systems - * actual credits misuse will be caught anyway */ + /* + * don't make noise on a tiny testing systems + * actual credits misuse will be caught anyway + */ if (last_credits != oh->ot_credits && time_after(jiffies, last_printed + - msecs_to_jiffies(60 * MSEC_PER_SEC)) && + cfs_time_seconds(60)) && osd_transaction_size(dev) > 512) { CWARN("%s: credits %u > trans_max %u\n", osd_name(dev), oh->ot_credits, osd_transaction_size(dev)); @@ -1813,51 +1947,59 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d, last_credits = oh->ot_credits; last_printed = jiffies; } - /* XXX Limit the credits to 'max_transaction_buffers', and + /* + * XXX Limit the credits to 'max_transaction_buffers', and * let the underlying filesystem to catch the error if * we really need so many credits. * * This should be removed when we can calculate the - * credits precisely. */ + * credits precisely. + */ oh->ot_credits = osd_transaction_size(dev); } else if (ldiskfs_track_declares_assert != 0) { - /* reserve few credits to prevent an assertion in JBD + /* + * reserve few credits to prevent an assertion in JBD * our debugging mechanism will be able to detected * overuse. this can help to debug single-update - * transactions */ + * transactions + */ oh->ot_credits += 10; if (unlikely(osd_param_is_not_sane(dev, th))) oh->ot_credits = osd_transaction_size(dev); } - /* - * XXX temporary stuff. Some abstraction layer should - * be used. - */ - jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); - osd_th_started(oh); - if (!IS_ERR(jh)) { - oh->ot_handle = jh; - LASSERT(oti->oti_txns == 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START)) + GOTO(out, rc = -EIO); + + /* + * XXX temporary stuff. Some abstraction layer should + * be used. + */ + jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits); + osd_th_started(oh); + if (!IS_ERR(jh)) { + oh->ot_handle = jh; + LASSERT(oti->oti_txns == 0); - lu_device_get(&d->dd_lu_dev); + atomic_inc(&dev->od_commit_cb_in_flight); lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link, "osd-tx", th); - oti->oti_txns++; - rc = 0; - } else { - rc = PTR_ERR(jh); - } + oti->oti_txns++; + rc = 0; + } else { + rc = PTR_ERR(jh); + } out: - RETURN(rc); + RETURN(rc); } static int osd_seq_exists(const struct lu_env *env, 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; + struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range; + struct seq_server_site *ss = osd_seq_site(osd); + int rc; + ENTRY; LASSERT(ss != NULL); @@ -1876,8 +2018,8 @@ static int osd_seq_exists(const struct lu_env *env, static void osd_trans_stop_cb(struct osd_thandle *oth, int result) { - struct dt_txn_commit_cb *dcb; - struct dt_txn_commit_cb *tmp; + struct dt_txn_commit_cb *dcb; + struct dt_txn_commit_cb *tmp; /* call per-transaction stop callbacks if any */ list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list, @@ -1896,39 +2038,45 @@ static void osd_trans_stop_cb(struct osd_thandle *oth, int result) static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, struct thandle *th) { - int rc = 0, remove_agents = 0; - struct osd_thandle *oh; struct osd_thread_info *oti = osd_oti_get(env); - struct osd_iobuf *iobuf = &oti->oti_iobuf; - struct osd_device *osd = osd_dt_dev(th->th_dev); - struct qsd_instance *qsd = osd->od_quota_slave; - struct lquota_trans *qtrans; + struct osd_thandle *oh; + struct osd_iobuf *iobuf = &oti->oti_iobuf; + struct osd_device *osd = osd_dt_dev(th->th_dev); + struct qsd_instance *qsd = osd_def_qsd(osd); + struct lquota_trans *qtrans; + LIST_HEAD(truncates); + int rc = 0, remove_agents = 0; + ENTRY; - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of(th, struct osd_thandle, ot_super); remove_agents = oh->ot_remove_agents; qtrans = oh->ot_quota_trans; oh->ot_quota_trans = NULL; + /* move locks to local list, stop tx, execute truncates */ + list_splice(&oh->ot_trunc_locks, &truncates); + if (oh->ot_handle != NULL) { int rc2; - handle_t *hdl = oh->ot_handle; - /* - * add commit callback - * notice we don't do this in osd_trans_start() - * as underlying transaction can change during truncate - */ - ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, - &oh->ot_jcb); + handle_t *hdl = oh->ot_handle; + + /* + * add commit callback + * notice we don't do this in osd_trans_start() + * as underlying transaction can change during truncate + */ + ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb, + &oh->ot_jcb); - LASSERT(oti->oti_txns == 1); - oti->oti_txns--; + LASSERT(oti->oti_txns == 1); + oti->oti_txns--; - rc = dt_txn_hook_stop(env, th); - if (rc != 0) + rc = dt_txn_hook_stop(env, th); + if (rc != 0) CERROR("%s: failed in transaction hook: rc = %d\n", osd_name(osd), rc); @@ -1943,15 +2091,20 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, osd_name(osd), rc2); if (!rc) rc = rc2; + + osd_process_truncates(&truncates); } else { osd_trans_stop_cb(oh, th->th_result); OBD_FREE_PTR(oh); } + osd_trunc_unlock_all(env, &truncates); + /* inform the quota slave device that the transaction is stopping */ qsd_op_end(env, qsd, qtrans); - /* as we want IO to journal and data IO be concurrent, we don't block + /* + * as we want IO to journal and data IO be concurrent, we don't block * awaiting data IO completion in osd_do_bio(), instead we wait here * once transaction is submitted to the journal. all reqular requests * don't do direct IO (except read/write), thus this wait_event becomes @@ -1969,6 +2122,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, if (unlikely(remove_agents != 0)) osd_process_scheduled_agent_removals(env, osd); + LASSERT(oti->oti_ins_cache_depth > 0); oti->oti_ins_cache_depth--; /* reset OI cache for safety */ if (oti->oti_ins_cache_depth == 0) @@ -1981,8 +2135,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt, static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) { - struct osd_thandle *oh = container_of0(th, struct osd_thandle, - ot_super); + struct osd_thandle *oh = container_of(th, struct osd_thandle, + ot_super); LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC); LASSERT(&dcb->dcb_func != NULL); @@ -2003,108 +2157,129 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb) */ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) { - struct osd_object *obj = osd_obj(l); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_obj(l); + struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj)); + struct inode *inode = obj->oo_inode; + __u64 projid; + qid_t uid; + qid_t gid; - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - /* - * If object is unlinked remove fid->ino mapping from object index. - */ + /* + * If object is unlinked remove fid->ino mapping from object index. + */ - osd_index_fini(obj); - if (inode != NULL) { - struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; - qid_t uid = i_uid_read(inode); - qid_t gid = i_gid_read(inode); + osd_index_fini(obj); - obj->oo_inode = NULL; - iput(inode); - if (!obj->oo_header && qsd) { - struct osd_thread_info *info = osd_oti_get(env); - struct lquota_id_info *qi = &info->oti_qi; + if (!inode) + return; - /* Release granted quota to master if necessary */ - qi->lqi_id.qid_uid = uid; - qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA); + if (osd_has_index(obj) && obj->oo_dt.do_index_ops == &osd_index_iam_ops) + ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA); - qi->lqi_id.qid_uid = gid; - qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA); + uid = i_uid_read(inode); + gid = i_gid_read(inode); + projid = i_projid_read(inode); - qi->lqi_id.qid_uid = i_projid_read(inode); - qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA); - } - } + obj->oo_inode = NULL; + iput(inode); + + /* do not rebalance quota if the caller needs to release memory + * otherwise qsd_refresh_usage() may went into a new ldiskfs + * transaction and risk to deadlock - LU-12178 */ + if (current->flags & (PF_MEMALLOC | PF_KSWAPD)) + return; + + if (!obj->oo_header && qsd) { + struct osd_thread_info *info = osd_oti_get(env); + struct lquota_id_info *qi = &info->oti_qi; + + /* Release granted quota to master if necessary */ + qi->lqi_id.qid_uid = uid; + qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA); + + qi->lqi_id.qid_uid = gid; + qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA); + + qi->lqi_id.qid_uid = projid; + qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA); + } } /* * Concurrency: ->loo_object_release() is called under site spin-lock. */ static void osd_object_release(const struct lu_env *env, - struct lu_object *l) + struct lu_object *l) { struct osd_object *o = osd_obj(l); - /* nobody should be releasing a non-destroyed object with nlink=0 + + /* + * nobody should be releasing a non-destroyed object with nlink=0 * the API allows this, but ldiskfs doesn't like and then report - * this inode as deleted */ - if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0)) - LBUG(); + * this inode as deleted + */ + LASSERT(!(o->oo_destroyed == 0 && o->oo_inode && + o->oo_inode->i_nlink == 0)); } /* * Concurrency: shouldn't matter. */ static int osd_object_print(const struct lu_env *env, void *cookie, - lu_printer_t p, const struct lu_object *l) + lu_printer_t p, const struct lu_object *l) { - struct osd_object *o = osd_obj(l); - struct iam_descr *d; + struct osd_object *o = osd_obj(l); + struct iam_descr *d; - if (o->oo_dir != NULL) - d = o->oo_dir->od_container.ic_descr; - else - d = NULL; + if (o->oo_dir != NULL) + d = o->oo_dir->od_container.ic_descr; + else + d = NULL; return (*p)(env, cookie, LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]", - o, o->oo_inode, - o->oo_inode ? o->oo_inode->i_ino : 0UL, - o->oo_inode ? o->oo_inode->i_generation : 0, - d ? d->id_ops->id_name : "plain"); + o, o->oo_inode, + o->oo_inode ? o->oo_inode->i_ino : 0UL, + o->oo_inode ? o->oo_inode->i_generation : 0, + d ? d->id_ops->id_name : "plain"); } /* * Concurrency: shouldn't matter. */ int osd_statfs(const struct lu_env *env, struct dt_device *d, - struct obd_statfs *sfs) + struct obd_statfs *sfs, struct obd_statfs_info *info) { - struct osd_device *osd = osd_dt_dev(d); - struct super_block *sb = osd_sb(osd); - struct kstatfs *ksfs; - __u64 reserved; - int result = 0; + struct osd_device *osd = osd_dt_dev(d); + struct super_block *sb = osd_sb(osd); + struct kstatfs *ksfs; + __u64 reserved; + int result = 0; if (unlikely(osd->od_mnt == NULL)) return -EINPROGRESS; - /* osd_lproc.c call this without env, allocate ksfs for that case */ - if (unlikely(env == NULL)) { - OBD_ALLOC_PTR(ksfs); - if (ksfs == NULL) - return -ENOMEM; - } else { - ksfs = &osd_oti_get(env)->oti_ksfs; - } + /* osd_lproc.c call this without env, allocate ksfs for that case */ + if (unlikely(env == NULL)) { + OBD_ALLOC_PTR(ksfs); + if (ksfs == NULL) + return -ENOMEM; + } else { + ksfs = &osd_oti_get(env)->oti_ksfs; + } result = sb->s_op->statfs(sb->s_root, ksfs); if (result) goto out; statfs_pack(sfs, ksfs); - if (unlikely(sb->s_flags & MS_RDONLY)) - sfs->os_state |= OS_STATE_READONLY; - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, - LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + if (unlikely(sb->s_flags & SB_RDONLY)) + sfs->os_state |= OS_STATFS_READONLY; + + sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0; + + if (ldiskfs_has_feature_extents(sb)) sfs->os_maxbytes = sb->s_maxbytes; else sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; @@ -2117,7 +2292,7 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, * * Reserve 0.78% of total space, at least 8MB for small filesystems. */ - CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE); + BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE); reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits; if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT)) reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT; @@ -2147,65 +2322,130 @@ out: * Concurrency: doesn't access mutable data. */ static void osd_conf_get(const struct lu_env *env, - const struct dt_device *dev, - struct dt_device_param *param) + const struct dt_device *dev, + struct dt_device_param *param) { - struct super_block *sb = osd_sb(osd_dt_dev(dev)); - int ea_overhead; + struct osd_device *d = osd_dt_dev(dev); + struct super_block *sb = osd_sb(d); + struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev); + const char *name; + int ea_overhead; - /* - * XXX should be taken from not-yet-existing fs abstraction layer. - */ - param->ddp_max_name_len = LDISKFS_NAME_LEN; - param->ddp_max_nlink = LDISKFS_LINK_MAX; + /* + * XXX should be taken from not-yet-existing fs abstraction layer. + */ + param->ddp_max_name_len = LDISKFS_NAME_LEN; + param->ddp_max_nlink = LDISKFS_LINK_MAX; param->ddp_symlink_max = sb->s_blocksize; - param->ddp_mount_type = LDD_MT_LDISKFS; - if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)) + param->ddp_mount_type = LDD_MT_LDISKFS; + if (ldiskfs_has_feature_extents(sb)) param->ddp_maxbytes = sb->s_maxbytes; else param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes; - /* inode are statically allocated, so per-inode space consumption - * is the space consumed by the directory entry */ + /* + * inode are statically allocated, so per-inode space consumption + * is the space consumed by the directory entry + */ param->ddp_inodespace = PER_OBJ_USAGE; - /* EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks - * = 128MB) which is unlikely to be hit in real life. Report a smaller - * maximum length to not under count the actual number of extents - * needed for writing a file. */ - param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2; + /* + * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks + * is 128MB) which is unlikely to be hit in real life. Report a smaller + * maximum length to not under-count the actual number of extents + * needed for writing a file if there are sub-optimal block allocations. + */ + param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1; /* worst-case extent insertion metadata overhead */ param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb); - param->ddp_mntopts = 0; - if (test_opt(sb, XATTR_USER)) - param->ddp_mntopts |= MNTOPT_USERXATTR; - if (test_opt(sb, POSIX_ACL)) - param->ddp_mntopts |= MNTOPT_ACL; + param->ddp_mntopts = 0; + if (test_opt(sb, XATTR_USER)) + param->ddp_mntopts |= MNTOPT_USERXATTR; + if (test_opt(sb, POSIX_ACL)) + param->ddp_mntopts |= MNTOPT_ACL; - /* LOD might calculate the max stripe count based on max_ea_size, + /* + * 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 */ + * 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)) + if (ldiskfs_has_feature_ea_inode(sb)) param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE - ea_overhead; else #endif param->ddp_max_ea_size = sb->s_blocksize - ea_overhead; - if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE) - param->ddp_max_ea_size = OSD_MAX_EA_SIZE; + if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE) + param->ddp_max_ea_size = OBD_MAX_EA_SIZE; - /* Preferred RPC size for efficient disk IO. 4MB shows good + /* + * Preferred RPC size for efficient disk IO. 4MB shows good * all-around performance for ldiskfs, but use bigalloc chunk size - * by default if larger. */ + * by default if larger. + */ #if defined(LDISKFS_CLUSTER_SIZE) if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE) param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb); else #endif param->ddp_brw_size = DT_DEF_BRW_SIZE; + + param->ddp_t10_cksum_type = 0; + if (bi) { + unsigned short interval = blk_integrity_interval(bi); + name = blk_integrity_name(bi); + /* + * Expected values: + * T10-DIF-TYPE1-CRC + * T10-DIF-TYPE3-CRC + * T10-DIF-TYPE1-IP + * T10-DIF-TYPE3-IP + */ + if (strncmp(name, "T10-DIF-TYPE", + sizeof("T10-DIF-TYPE") - 1) == 0) { + /* also skip "1/3-" at end */ + const int type_off = sizeof("T10-DIF-TYPE."); + char type_number = name[type_off - 2]; + + if (interval != 512 && interval != 4096) { + CERROR("%s: unsupported T10PI sector size %u\n", + d->od_svname, interval); + } else if (type_number != '1' && type_number != '3') { + CERROR("%s: unsupported T10PI type %s\n", + d->od_svname, name); + } else if (strcmp(name + type_off, "CRC") == 0) { + d->od_t10_type = type_number == '1' ? + OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC; + param->ddp_t10_cksum_type = interval == 512 ? + OBD_CKSUM_T10CRC512 : + OBD_CKSUM_T10CRC4K; + } else if (strcmp(name + type_off, "IP") == 0) { + d->od_t10_type = type_number == '1' ? + OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP; + param->ddp_t10_cksum_type = interval == 512 ? + OBD_CKSUM_T10IP512 : + OBD_CKSUM_T10IP4K; + } else { + CERROR("%s: unsupported checksum type of " + "T10PI type '%s'", + d->od_svname, name); + } + + } else { + CERROR("%s: unsupported T10PI type '%s'", + d->od_svname, name); + } + } + + param->ddp_has_lseek_data_hole = true; +} + +static struct super_block *osd_mnt_sb_get(const struct dt_device *d) +{ + return osd_sb(osd_dt_dev(d)); } /* @@ -2214,10 +2454,12 @@ static void osd_conf_get(const struct lu_env *env, static int osd_sync(const struct lu_env *env, struct dt_device *d) { int rc; + struct super_block *s = osd_sb(osd_dt_dev(d)); + ENTRY; - CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname); - - rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); + down_read(&s->s_umount); + rc = s->s_op->sync_fs(s, 1); + up_read(&s->s_umount); CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname, rc); @@ -2238,19 +2480,29 @@ static int osd_sync(const struct lu_env *env, struct dt_device *d) * \see dt_device_operations */ static int osd_commit_async(const struct lu_env *env, - struct dt_device *d) + struct dt_device *d) { - struct super_block *s = osd_sb(osd_dt_dev(d)); - ENTRY; + struct super_block *s = osd_sb(osd_dt_dev(d)); + int rc; + + ENTRY; CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname); - RETURN(s->s_op->sync_fs(s, 0)); + down_read(&s->s_umount); + rc = s->s_op->sync_fs(s, 0); + up_read(&s->s_umount); + + RETURN(rc); } -/* Our own copy of the set readonly functions if present, or NU if not. */ -static int (*priv_dev_set_rdonly)(struct block_device *bdev); -static int (*priv_dev_check_rdonly)(struct block_device *bdev); -/* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */ +static int (*priv_security_file_alloc)(struct file *file); + +int osd_security_file_alloc(struct file *file) +{ + if (priv_security_file_alloc) + return priv_security_file_alloc(file); + return 0; +} /* * Concurrency: shouldn't matter. @@ -2260,37 +2512,11 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) struct super_block *sb = osd_sb(osd_dt_dev(d)); struct block_device *dev = sb->s_bdev; int rc = -EOPNOTSUPP; - ENTRY; - - if (priv_dev_set_rdonly) { - struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev; - rc = 0; - CERROR("*** setting %s read-only ***\n", - osd_dt_dev(d)->od_svname); - - if (sb->s_op->freeze_fs) { - rc = sb->s_op->freeze_fs(sb); - if (rc) - goto out; - } - - if (jdev && (jdev != dev)) { - CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n", - (long)jdev); - priv_dev_set_rdonly(jdev); - } - CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev); - priv_dev_set_rdonly(dev); - - if (sb->s_op->unfreeze_fs) - sb->s_op->unfreeze_fs(sb); - } + ENTRY; -out: - if (rc) - CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", - osd_dt_dev(d)->od_svname, (long)dev, rc); + CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n", + osd_dt_dev(d)->od_svname, (long)dev, rc); RETURN(rc); } @@ -2354,20 +2580,21 @@ const int osd_dto_credits_noquota[DTO_NR] = { }; static const struct dt_device_operations osd_dt_ops = { - .dt_root_get = osd_root_get, - .dt_statfs = osd_statfs, - .dt_trans_create = osd_trans_create, - .dt_trans_start = osd_trans_start, - .dt_trans_stop = osd_trans_stop, - .dt_trans_cb_add = osd_trans_cb_add, - .dt_conf_get = osd_conf_get, - .dt_sync = osd_sync, - .dt_ro = osd_ro, - .dt_commit_async = osd_commit_async, + .dt_root_get = osd_root_get, + .dt_statfs = osd_statfs, + .dt_trans_create = osd_trans_create, + .dt_trans_start = osd_trans_start, + .dt_trans_stop = osd_trans_stop, + .dt_trans_cb_add = osd_trans_cb_add, + .dt_conf_get = osd_conf_get, + .dt_mnt_sb_get = osd_mnt_sb_get, + .dt_sync = osd_sync, + .dt_ro = osd_ro, + .dt_commit_async = osd_commit_async, }; static void osd_read_lock(const struct lu_env *env, struct dt_object *dt, - unsigned role) + unsigned int role) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *oti = osd_oti_get(env); @@ -2382,7 +2609,7 @@ static void osd_read_lock(const struct lu_env *env, struct dt_object *dt, } static void osd_write_lock(const struct lu_env *env, struct dt_object *dt, - unsigned role) + unsigned int role) { struct osd_object *obj = osd_dt_obj(dt); struct osd_thread_info *oti = osd_oti_get(env); @@ -2432,29 +2659,18 @@ static int osd_write_locked(const struct lu_env *env, struct dt_object *dt) return obj->oo_owner == env; } -static struct timespec *osd_inode_time(const struct lu_env *env, - struct inode *inode, __u64 seconds) -{ - struct osd_thread_info *oti = osd_oti_get(env); - struct timespec *t = &oti->oti_time; - - t->tv_sec = seconds; - t->tv_nsec = 0; - *t = timespec_trunc(*t, inode->i_sb->s_time_gran); - return t; -} - 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_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV | - LA_BLKSIZE | LA_TYPE; + LA_BLKSIZE | LA_TYPE | LA_BTIME; - 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_atime = inode->i_atime.tv_sec; + attr->la_mtime = inode->i_mtime.tv_sec; + attr->la_ctime = inode->i_ctime.tv_sec; + attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec; attr->la_mode = inode->i_mode; attr->la_size = i_size_read(inode); attr->la_blocks = inode->i_blocks; @@ -2475,11 +2691,71 @@ static void osd_inode_getattr(const struct lu_env *env, attr->la_flags |= LUSTRE_PROJINHERIT_FL; } -static int osd_attr_get(const struct lu_env *env, - struct dt_object *dt, +static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt, + u64 *count) +{ + struct osd_object *obj = osd_dt_obj(dt); + const struct dt_it_ops *iops; + struct dt_it *it; + int rc; + + ENTRY; + + LASSERT(S_ISDIR(obj->oo_inode->i_mode)); + LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu))); + + if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) { + *count = obj->oo_dirent_count; + RETURN(0); + } + + /* directory not initialized yet */ + if (!dt->do_index_ops) { + *count = 0; + RETURN(0); + } + + iops = &dt->do_index_ops->dio_it; + it = iops->init(env, dt, LUDA_64BITHASH); + if (IS_ERR(it)) + RETURN(PTR_ERR(it)); + + rc = iops->load(env, it, 0); + if (rc < 0) { + if (rc == -ENODATA) { + rc = 0; + *count = 0; + } + GOTO(out, rc); + } + if (rc > 0) + rc = iops->next(env, it); + + for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) { + if (rc == -ESTALE) + continue; + + if (iops->key_size(env, it) == 0) + continue; + + (*count)++; + } + if (rc == 1) { + obj->oo_dirent_count = *count; + rc = 0; + } +out: + iops->put(env, it); + iops->fini(env, it); + + RETURN(rc); +} + +static int osd_attr_get(const struct lu_env *env, struct dt_object *dt, struct lu_attr *attr) { struct osd_object *obj = osd_dt_obj(dt); + int rc = 0; if (unlikely(!dt_object_exists(dt))) return -ENOENT; @@ -2491,18 +2767,28 @@ static int osd_attr_get(const struct lu_env *env, spin_lock(&obj->oo_guard); osd_inode_getattr(env, obj->oo_inode, attr); - if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) + if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) { + attr->la_valid |= LA_FLAGS; attr->la_flags |= LUSTRE_ORPHAN_FL; + } + if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) { + attr->la_valid |= LA_FLAGS; + attr->la_flags |= LUSTRE_ENCRYPT_FL; + } spin_unlock(&obj->oo_guard); - return 0; + if (S_ISDIR(obj->oo_inode->i_mode) && + fid_is_namespace_visible(lu_object_fid(&dt->do_lu))) + rc = osd_dirent_count(env, dt, &attr->la_dirent_count); + + return rc; } static int osd_declare_attr_qid(const struct lu_env *env, struct osd_object *obj, struct osd_thandle *oh, long long bspace, qid_t old_id, qid_t new_id, bool enforce, - unsigned type, bool ignore_edquot) + unsigned int type, bool ignore_edquot) { int rc; struct osd_thread_info *info = osd_oti_get(env); @@ -2524,7 +2810,7 @@ static int osd_declare_attr_qid(const struct lu_env *env, /* and one less inode for the current id */ qi->lqi_id.qid_uid = old_id; - qi->lqi_space = -1; + qi->lqi_space = -1; rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL); if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS)) rc = 0; @@ -2558,17 +2844,18 @@ static int osd_declare_attr_qid(const struct lu_env *env, } static int osd_declare_attr_set(const struct lu_env *env, - struct dt_object *dt, - const struct lu_attr *attr, - struct thandle *handle) + struct dt_object *dt, + const struct lu_attr *attr, + struct thandle *handle) { - struct osd_thandle *oh; - struct osd_object *obj; - qid_t uid; - qid_t gid; - long long bspace; - int rc = 0; - bool enforce; + struct osd_thandle *oh; + struct osd_object *obj; + qid_t uid; + qid_t gid; + long long bspace; + int rc = 0; + bool enforce; + ENTRY; LASSERT(dt != NULL); @@ -2577,7 +2864,7 @@ static int osd_declare_attr_set(const struct lu_env *env, obj = osd_dt_obj(dt); LASSERT(osd_invariant(obj)); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET, @@ -2592,19 +2879,21 @@ static int osd_declare_attr_set(const struct lu_env *env, bspace = obj->oo_inode->i_blocks << 9; bspace = toqb(bspace); - /* Changing ownership is always preformed by super user, it should not + /* + * Changing ownership is always preformed by super user, it should not * fail with EDQUOT unless required explicitly. * * 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.*/ + * space adjustment once the operation is completed. + */ if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) { bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL); if (!ignore_edquot) - CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u" - "(the quota space is %lld)\n", - obj->oo_inode->i_sb->s_id, attr->la_uid, + CDEBUG(D_QUOTA, + "%s: enforce quota on UID %u, GID %u (quota space is %lld)\n", + osd_ino2name(obj->oo_inode), attr->la_uid, attr->la_gid, bspace); /* USERQUOTA */ @@ -2617,9 +2906,10 @@ static int osd_declare_attr_set(const struct lu_env *env, RETURN(rc); gid = i_gid_read(obj->oo_inode); + CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid, + attr->la_uid, gid, attr->la_gid); enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid); - rc = osd_declare_attr_qid(env, obj, oh, bspace, - i_gid_read(obj->oo_inode), + rc = osd_declare_attr_qid(env, obj, oh, bspace, gid, attr->la_gid, enforce, GRPQUOTA, ignore_edquot); if (rc) @@ -2629,6 +2919,7 @@ static int osd_declare_attr_set(const struct lu_env *env, #ifdef HAVE_PROJECT_QUOTA if (attr->la_valid & LA_PROJID) { __u32 projid = i_projid_read(obj->oo_inode); + enforce = (attr->la_valid & LA_PROJID) && (attr->la_projid != projid); rc = osd_declare_attr_qid(env, obj, oh, bspace, @@ -2654,11 +2945,11 @@ static int osd_inode_setattr(const struct lu_env *env, return 0; if (bits & LA_ATIME) - inode->i_atime = *osd_inode_time(env, inode, attr->la_atime); + inode->i_atime = osd_inode_time(inode, attr->la_atime); if (bits & LA_CTIME) - inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime); + inode->i_ctime = osd_inode_time(inode, attr->la_ctime); if (bits & LA_MTIME) - inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime); + inode->i_mtime = osd_inode_time(inode, attr->la_mtime); if (bits & LA_SIZE) { spin_lock(&inode->i_lock); LDISKFS_I(inode)->i_disksize = attr->la_size; @@ -2666,8 +2957,10 @@ static int osd_inode_setattr(const struct lu_env *env, spin_unlock(&inode->i_lock); } - /* OSD should not change "i_blocks" which is used by quota. - * "i_blocks" should be changed by ldiskfs only. */ + /* + * OSD should not change "i_blocks" which is used by quota. + * "i_blocks" should be changed by ldiskfs only. + */ if (bits & LA_MODE) inode->i_mode = (inode->i_mode & S_IFMT) | (attr->la_mode & ~S_IFMT); @@ -2686,6 +2979,13 @@ static int osd_inode_setattr(const struct lu_env *env, /* always keep S_NOCMTIME */ inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) | S_NOCMTIME; +#if defined(S_ENCRYPTED) + /* Always remove S_ENCRYPTED, because ldiskfs must not be + * aware of encryption status. It is just stored into LMA + * so that it can be forwared to client side. + */ + inode->i_flags &= ~S_ENCRYPTED; +#endif /* * Ext4 did not transfer inherit flags from * @inode->i_flags to raw inode i_flags when writing @@ -2699,15 +2999,83 @@ static int osd_inode_setattr(const struct lu_env *env, return 0; } -static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) +#ifdef HAVE_PROJECT_QUOTA +static int osd_transfer_project(struct inode *inode, __u32 projid, + struct thandle *handle) +{ + struct super_block *sb = inode->i_sb; + struct ldiskfs_inode_info *ei = LDISKFS_I(inode); + int err; + kprojid_t kprojid; + struct ldiskfs_iloc iloc; + struct ldiskfs_inode *raw_inode; + struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { }; + + if (!ldiskfs_has_feature_project(sb)) { + LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid) + == LDISKFS_DEF_PROJID); + if (projid != LDISKFS_DEF_PROJID) + return -EOPNOTSUPP; + else + return 0; + } + + if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE) + return -EOPNOTSUPP; + + kprojid = make_kprojid(&init_user_ns, (projid_t)projid); + if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid)) + return 0; + + err = ldiskfs_get_inode_loc(inode, &iloc); + if (err) + return err; + + raw_inode = ldiskfs_raw_inode(&iloc); + if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) { + struct osd_thandle *oh = container_of(handle, + struct osd_thandle, + ot_super); + /** + * try to expand inode size automatically. + */ + ldiskfs_mark_inode_dirty(oh->ot_handle, inode); + if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) { + err = -EOVERFLOW; + brelse(iloc.bh); + return err; + } + } + brelse(iloc.bh); + + dquot_initialize(inode); + transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid)); + if (transfer_to[PRJQUOTA]) { + err = __dquot_transfer(inode, transfer_to); + dqput(transfer_to[PRJQUOTA]); + if (err) + return err; + } + + return err; +} +#endif + +static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr, + struct thandle *handle) { int rc; 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; + struct iattr iattr; + + CDEBUG(D_QUOTA, + "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n", + inode->i_ino, i_uid_read(inode), attr->la_uid, + i_gid_read(inode), attr->la_gid); - ll_vfs_dq_init(inode); + dquot_initialize(inode); iattr.ia_valid = 0; if (attr->la_valid & LA_UID) iattr.ia_valid |= ATTR_UID; @@ -2716,11 +3084,10 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) 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); + rc = dquot_transfer(inode, &iattr); if (rc) { - CERROR("%s: quota transfer failed: rc = %d. Is quota " - "enforcement enabled on the ldiskfs " - "filesystem?\n", inode->i_sb->s_id, rc); + CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n", + osd_ino2name(inode), rc); return rc; } } @@ -2729,14 +3096,13 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr) if (attr->la_valid & LA_PROJID && attr->la_projid != i_projid_read(inode)) { #ifdef HAVE_PROJECT_QUOTA - rc = ldiskfs_transfer_project(inode, attr->la_projid); + rc = osd_transfer_project(inode, attr->la_projid, handle); #else rc = -ENOTSUPP; #endif if (rc) { - CERROR("%s: quota transfer failed: rc = %d. Is project " - "enforcement enabled on the ldiskfs " - "filesystem?\n", inode->i_sb->s_id, rc); + CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n", + osd_ino2name(inode), rc); return rc; } } @@ -2749,7 +3115,7 @@ static int osd_attr_set(const struct lu_env *env, struct thandle *handle) { struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode; + struct inode *inode; int rc; if (!dt_object_exists(dt)) @@ -2763,14 +3129,14 @@ static int osd_attr_set(const struct lu_env *env, if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd_obj2dev(obj)->od_is_ost) { - struct osd_thread_info *oti = osd_oti_get(env); - const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); - struct lu_fid *fid1 = &oti->oti_fid; - struct osd_inode_id *id = &oti->oti_id; - struct iam_path_descr *ipd; - struct iam_container *bag; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu); + struct lu_fid *fid1 = &oti->oti_fid; + struct osd_inode_id *id = &oti->oti_id; + struct iam_path_descr *ipd; + struct iam_container *bag; + struct osd_thandle *oh; + int rc; fid_cpu_to_be(fid1, fid0); memset(id, 1, sizeof(*id)); @@ -2780,16 +3146,17 @@ static int osd_attr_set(const struct lu_env *env, if (unlikely(ipd == NULL)) RETURN(-ENOMEM); - oh = container_of0(handle, struct osd_thandle, ot_super); - rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1, + oh = container_of(handle, struct osd_thandle, ot_super); + rc = iam_update(oh->ot_handle, bag, + (const struct iam_key *)fid1, (const struct iam_rec *)id, ipd); osd_ipd_put(env, bag, ipd); return(rc > 0 ? 0 : rc); } - inode = obj->oo_inode; + inode = obj->oo_inode; - rc = osd_quota_transfer(inode, attr); + rc = osd_quota_transfer(inode, attr, handle); if (rc) return rc; @@ -2799,7 +3166,9 @@ static int osd_attr_set(const struct lu_env *env, if (rc != 0) GOTO(out, rc); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + osd_dirty_inode(inode, I_DIRTY_DATASYNC); + + osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); if (!(attr->la_valid & LA_FLAGS)) GOTO(out, rc); @@ -2819,6 +3188,9 @@ static int osd_attr_set(const struct lu_env *env, lma->lma_incompat |= lustre_to_lma_flags(attr->la_flags); lustre_lma_swab(lma); + + osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); + rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma), XATTR_REPLACE); if (rc != 0) { @@ -2834,72 +3206,80 @@ static int osd_attr_set(const struct lu_env *env, osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); } out: - osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET); - return rc; + return rc; } 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); + return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen); } static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, umode_t mode, struct dt_allocation_hint *hint, - struct thandle *th) + struct thandle *th, struct lu_attr *attr) { - int result; - struct osd_device *osd = osd_obj2dev(obj); - struct osd_thandle *oth; - struct dt_object *parent = NULL; - struct inode *inode; + int result; + struct osd_device *osd = osd_obj2dev(obj); + struct osd_thandle *oth; + struct dt_object *parent = NULL; + struct inode *inode; + uid_t owner[2] = {0, 0}; - LINVRNT(osd_invariant(obj)); - LASSERT(obj->oo_inode == NULL); - LASSERT(obj->oo_hl_head == NULL); + if (attr->la_valid & LA_UID) + owner[0] = attr->la_uid; + if (attr->la_valid & LA_GID) + owner[1] = attr->la_gid; + + LINVRNT(osd_invariant(obj)); + LASSERT(obj->oo_inode == NULL); + LASSERT(obj->oo_hl_head == NULL); - if (S_ISDIR(mode) && ldiskfs_pdo) { - obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); - if (obj->oo_hl_head == NULL) - return -ENOMEM; - } + if (S_ISDIR(mode) && ldiskfs_pdo) { + obj->oo_hl_head = + ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF); + if (obj->oo_hl_head == NULL) + return -ENOMEM; + } - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); if (hint != NULL && hint->dah_parent != NULL && !dt_object_remote(hint->dah_parent)) parent = hint->dah_parent; - inode = ldiskfs_create_inode(oth->ot_handle, - parent ? osd_dt_obj(parent)->oo_inode : - osd_sb(osd)->s_root->d_inode, - mode); - if (!IS_ERR(inode)) { + inode = ldiskfs_create_inode(oth->ot_handle, + parent ? osd_dt_obj(parent)->oo_inode : + osd_sb(osd)->s_root->d_inode, + mode, owner); + if (!IS_ERR(inode)) { /* 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. */ + /* + * 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 { - if (obj->oo_hl_head != NULL) { - ldiskfs_htree_lock_head_free(obj->oo_hl_head); - obj->oo_hl_head = NULL; - } - result = PTR_ERR(inode); - } - LINVRNT(osd_invariant(obj)); - return result; + obj->oo_inode = inode; + result = 0; + } else { + if (obj->oo_hl_head != NULL) { + ldiskfs_htree_lock_head_free(obj->oo_hl_head); + obj->oo_hl_head = NULL; + } + result = PTR_ERR(inode); + } + LINVRNT(osd_invariant(obj)); + return result; } enum { - OSD_NAME_LEN = 255 + OSD_NAME_LEN = 255 }; static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj, @@ -2916,68 +3296,71 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj, oth = container_of(th, struct osd_thandle, ot_super); LASSERT(oth->ot_handle->h_transaction != NULL); - result = osd_mkfile(info, obj, mode, hint, th); + if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu))) + obj->oo_dirent_count = 0; + result = osd_mkfile(info, obj, mode, hint, th, attr); return result; } static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *th) -{ - int result; - struct osd_thandle *oth; - const struct dt_index_features *feat = dof->u.dof_idx.di_feat; - - __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX)); - - LASSERT(S_ISREG(attr->la_mode)); - - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); - - result = osd_mkfile(info, obj, mode, hint, th); - if (result == 0) { - LASSERT(obj->oo_inode != NULL); - if (feat->dif_flags & DT_IND_VARKEY) - result = iam_lvar_create(obj->oo_inode, - feat->dif_keysize_max, - feat->dif_ptrsize, - feat->dif_recsize_max, - oth->ot_handle); - else - result = iam_lfix_create(obj->oo_inode, - feat->dif_keysize_max, - feat->dif_ptrsize, - feat->dif_recsize_max, - oth->ot_handle); - - } - return result; + struct lu_attr *attr, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *th) +{ + int result; + struct osd_thandle *oth; + const struct dt_index_features *feat = dof->u.dof_idx.di_feat; + + __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX)); + + LASSERT(S_ISREG(attr->la_mode)); + + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); + + result = osd_mkfile(info, obj, mode, hint, th, attr); + if (result == 0) { + LASSERT(obj->oo_inode != NULL); + if (feat->dif_flags & DT_IND_VARKEY) + result = iam_lvar_create(obj->oo_inode, + feat->dif_keysize_max, + feat->dif_ptrsize, + feat->dif_recsize_max, + oth->ot_handle); + else + result = iam_lfix_create(obj->oo_inode, + feat->dif_keysize_max, + feat->dif_ptrsize, + feat->dif_recsize_max, + oth->ot_handle); + } + return result; } static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj, - 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) { - LASSERT(S_ISREG(attr->la_mode)); - return osd_mkfile(info, obj, (attr->la_mode & - (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); + LASSERT(S_ISREG(attr->la_mode)); + return osd_mkfile(info, obj, (attr->la_mode & + (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th, + attr); } static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj, - 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) { - LASSERT(S_ISLNK(attr->la_mode)); - return osd_mkfile(info, obj, (attr->la_mode & - (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th); + LASSERT(S_ISLNK(attr->la_mode)); + return osd_mkfile(info, obj, (attr->la_mode & + (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th, + attr); } static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, @@ -2991,55 +3374,55 @@ static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj, LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); - LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || - S_ISFIFO(mode) || S_ISSOCK(mode)); + LASSERT(S_ISCHR(mode) || S_ISBLK(mode) || + S_ISFIFO(mode) || S_ISSOCK(mode)); - result = osd_mkfile(info, obj, mode, hint, th); - if (result == 0) { - LASSERT(obj->oo_inode != NULL); + result = osd_mkfile(info, obj, mode, hint, th, attr); + if (result == 0) { + LASSERT(obj->oo_inode != NULL); /* * This inode should be marked dirty for i_rdev. Currently * that is done in the osd_attr_init(). */ init_special_inode(obj->oo_inode, obj->oo_inode->i_mode, attr->la_rdev); - } - LINVRNT(osd_invariant(obj)); - return result; + } + LINVRNT(osd_invariant(obj)); + return result; } typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *, - struct lu_attr *, - struct dt_allocation_hint *hint, - struct dt_object_format *dof, - struct thandle *); + struct lu_attr *, + struct dt_allocation_hint *hint, + struct dt_object_format *dof, + struct thandle *); static osd_obj_type_f osd_create_type_f(enum dt_format_type type) { - osd_obj_type_f result; - - switch (type) { - case DFT_DIR: - result = osd_mkdir; - break; - case DFT_REGULAR: - result = osd_mkreg; - break; - case DFT_SYM: - result = osd_mksym; - break; - case DFT_NODE: - result = osd_mknod; - break; - case DFT_INDEX: - result = osd_mk_index; - break; - - default: - LBUG(); - break; - } - return result; + osd_obj_type_f result; + + switch (type) { + case DFT_DIR: + result = osd_mkdir; + break; + case DFT_REGULAR: + result = osd_mkreg; + break; + case DFT_SYM: + result = osd_mksym; + break; + case DFT_NODE: + result = osd_mknod; + break; + case DFT_INDEX: + result = osd_mk_index; + break; + + default: + LBUG(); + break; + } + return result; } @@ -3047,37 +3430,39 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, struct dt_object *parent, struct dt_object *child, umode_t child_mode) { - LASSERT(ah); + LASSERT(ah); - ah->dah_parent = parent; - ah->dah_mode = child_mode; + ah->dah_parent = parent; + ah->dah_mode = child_mode; if (parent != NULL && !dt_object_remote(parent)) { /* will help to find FID->ino at dt_insert("..") */ struct osd_object *pobj = osd_dt_obj(parent); + osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj); } } static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, - struct lu_attr *attr, struct dt_object_format *dof) + struct lu_attr *attr, struct dt_object_format *dof, + struct thandle *handle) { - struct inode *inode = obj->oo_inode; - __u64 valid = attr->la_valid; - int result; + struct inode *inode = obj->oo_inode; + __u64 valid = attr->la_valid; + int result; attr->la_valid &= ~(LA_TYPE | LA_MODE); - if (dof->dof_type != DFT_NODE) - attr->la_valid &= ~LA_RDEV; - if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime))) - attr->la_valid &= ~LA_ATIME; - if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime))) - attr->la_valid &= ~LA_CTIME; - if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime))) - attr->la_valid &= ~LA_MTIME; + if (dof->dof_type != DFT_NODE) + attr->la_valid &= ~LA_RDEV; + if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec)) + attr->la_valid &= ~LA_ATIME; + if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec)) + attr->la_valid &= ~LA_CTIME; + if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec)) + attr->la_valid &= ~LA_MTIME; - result = osd_quota_transfer(inode, attr); + result = osd_quota_transfer(inode, attr, handle); if (result) return; @@ -3091,7 +3476,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, * enabled on ldiskfs (lquota takes care of it). */ LASSERTF(result == 0, "%d\n", result); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + osd_dirty_inode(inode, I_DIRTY_DATASYNC); } attr->la_valid = valid; @@ -3106,8 +3491,8 @@ static int __osd_create(struct osd_thread_info *info, struct osd_object *obj, struct lu_attr *attr, struct dt_allocation_hint *hint, struct dt_object_format *dof, struct thandle *th) { - int result; - __u32 umask; + int result; + __u32 umask; osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE); @@ -3120,13 +3505,15 @@ static int __osd_create(struct osd_thread_info *info, struct osd_object *obj, 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 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_attr_init(info, obj, attr, dof, th); osd_object_init0(obj); } @@ -3150,17 +3537,33 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id = &info->oti_id; struct osd_device *osd = osd_obj2dev(obj); struct osd_thandle *oh; - int rc; + int rc; LASSERT(obj->oo_inode != NULL); - oh = container_of0(th, struct osd_thandle, ot_super); + if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC)) + return -ENOSPC; + + oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle); osd_trans_exec_op(env, th, OSD_OT_INSERT); osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation); rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD, NULL); + if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) { + struct lu_fid next_fid = *fid; + + /* insert next object in advance, and map to the same inode */ + next_fid.f_oid++; + if (next_fid.f_oid != 0) { + osd_trans_exec_op(env, th, OSD_OT_INSERT); + osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle, + OI_CHECK_FLD, NULL); + osd_trans_exec_check(env, th, OSD_OT_INSERT); + } + } + osd_trans_exec_check(env, th, OSD_OT_INSERT); return rc; @@ -3169,7 +3572,7 @@ 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, u64 seq, struct lu_seq_range *range) { - struct seq_server_site *ss = osd_seq_site(osd); + struct seq_server_site *ss = osd_seq_site(osd); if (fid_seq_is_idif(seq)) { fld_range_set_ost(range); @@ -3180,8 +3583,10 @@ int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, if (!fid_seq_in_fldb(seq)) { fld_range_set_mdt(range); if (ss != NULL) - /* FIXME: If ss is NULL, it suppose not get lsr_index - * at all */ + /* + * FIXME: If ss is NULL, it suppose not get lsr_index + * at all + */ range->lsr_index = ss->ss_node_id; return 0; } @@ -3200,24 +3605,38 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, { struct osd_thandle *oh; int rc; + ENTRY; LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); - /* EA object consumes more credits than regular object: osd_mk_index + /* + * EA object consumes more credits than regular object: osd_mk_index * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and * leaf_node, could involves the block, block bitmap, groups, GDT - * change for each block, so add 4 * 2 credits in that case. */ + * change for each block, so add 4 * 2 credits in that case. + */ osd_trans_declare_op(env, oh, OSD_OT_CREATE, osd_dto_credits_noquota[DTO_OBJECT_CREATE] + (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0); - /* Reuse idle OI block may cause additional one OI block - * to be changed. */ + /* + * 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 (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP)) + osd_trans_declare_op(env, oh, OSD_OT_INSERT, + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1); + + /* will help to find FID->ino mapping at dt_insert() */ + rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)), + osd_dt_obj(dt)); + if (rc != 0) + RETURN(rc); if (!attr) RETURN(0); @@ -3228,10 +3647,6 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, if (rc != 0) RETURN(rc); - /* will help to find FID->ino mapping at dt_insert() */ - rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)), - osd_dt_obj(dt)); - RETURN(rc); } @@ -3243,16 +3658,17 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt, static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - int rc; + int rc; + ENTRY; if (inode == NULL) RETURN(-ENOENT); - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_DESTROY, @@ -3262,8 +3678,10 @@ static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE]; - /* Recycle idle OI leaf may cause additional three OI blocks - * to be changed. */ + /* + * Recycle idle OI leaf may cause additional three OI blocks + * to be changed. + */ if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2)) osd_trans_declare_op(env, oh, OSD_OT_DELETE, osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3); @@ -3280,8 +3698,10 @@ static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt, if (rc) RETURN(rc); - /* will help to find FID->ino when this object is being - * added to PENDING/ */ + /* + * will help to find FID->ino when this object is being + * added to PENDING + */ rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj); RETURN(rc); @@ -3296,9 +3716,10 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt, struct osd_device *osd = osd_obj2dev(obj); struct osd_thandle *oh; int result; + ENTRY; - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle); LASSERT(inode); LASSERT(!lu_object_is_dying(dt->do_lu.lo_header)); @@ -3314,13 +3735,15 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt, } if (S_ISDIR(inode->i_mode)) { - LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 || - inode->i_nlink == 2); + if (inode->i_nlink > 2) + CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n", + osd_name(osd), PFID(fid), inode->i_ino, + inode->i_nlink); spin_lock(&obj->oo_guard); clear_nlink(inode); spin_unlock(&obj->oo_guard); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + osd_dirty_inode(inode, I_DIRTY_DATASYNC); } osd_trans_exec_op(env, th, OSD_OT_DESTROY); @@ -3355,9 +3778,10 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt, int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, const struct lu_fid *fid, __u32 compat, __u32 incompat) { - struct lustre_ost_attrs *loa = &info->oti_ost_attrs; - struct lustre_mdt_attrs *lma = &loa->loa_lma; + struct lustre_ost_attrs *loa = &info->oti_ost_attrs; + struct lustre_mdt_attrs *lma = &loa->loa_lma; int rc; + ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA)) @@ -3369,10 +3793,12 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, lustre_loa_init(loa, fid, compat, incompat); lustre_loa_swab(loa, false); - /* For the OST device with 256 bytes inode size by default, + /* + * For the OST device with 256 bytes inode size by default, * the PFID EA will be stored together with LMA EA to avoid * performance trouble. Otherwise the PFID EA can be stored - * independently. LU-8998 */ + * independently. LU-8998 + */ if ((compat & LMAC_FID_ON_OST) && LDISKFS_INODE_SIZE(inode->i_sb) <= 256) rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, @@ -3380,8 +3806,10 @@ int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode, else rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma), XATTR_CREATE); - /* LMA may already exist, but we need to check that all the - * desired compat/incompat flags have been added. */ + /* + * LMA may already exist, but we need to check that all the + * desired compat/incompat flags have been added. + */ if (unlikely(rc == -EEXIST)) { rc = __osd_xattr_get(inode, &info->oti_obj_dentry, XATTR_NAME_LMA, (void *)loa, sizeof(*loa)); @@ -3449,7 +3877,8 @@ static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj, struct osd_inode_id *id) { struct osd_thread_info *info = osd_oti_get(env); - struct inode *inode; + struct inode *inode; + ENTRY; osd_id_gen(id, ino, OSD_OII_NOGEN); @@ -3473,6 +3902,9 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, __u32 saved_nlink = dir->i_nlink; int rc; + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC)) + return -ENOSPC; + dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2; osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid); @@ -3481,7 +3913,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir, dir, dot_ldp, dot_dot_ldp); - /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then + /* + * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then * the subseqent ref_add() will increase the dir->i_nlink * as 3. That is incorrect for new created directory. * @@ -3493,7 +3926,8 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info, * becuase if the nlink attr is not 2, then there will be * ref_add() called following the dt_insert(..), such call * will make both the in-RAM and on-disk dir-inode's nlink - * attr to be set as 2. LU-7447 */ + * attr to be set as 2. LU-7447 + */ set_nlink(dir, saved_nlink); return rc; } @@ -3508,38 +3942,54 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env, __u32 type, struct thandle *th) { - struct osd_thread_info *info = osd_oti_get(env); - struct inode *local; - struct osd_thandle *oh; - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct inode *local; + struct osd_thandle *oh; + uid_t own[2] = {0, 0}; + int rc; + ENTRY; LASSERT(th); oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle->h_transaction != NULL); - local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type); + local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type, own); if (IS_ERR(local)) { CERROR("%s: create local error %d\n", osd_name(osd), (int)PTR_ERR(local)); RETURN(local); } - /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set - * correct gid on remote file, not agent here */ + /* + * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set + * correct gid on remote file, not agent here + */ local->i_gid = current_fsgid(); ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB); + + /* e2fsck doesn't like empty symlinks. Store remote FID as symlink. + * That gives e2fsck something to look at and be happy, and allows + * debugging if we need to determine where this symlink came from. + */ + if (S_ISLNK(type)) { + BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1); + rc = snprintf((char *)LDISKFS_I(local)->i_data, + LDISKFS_N_BLOCKS * 4, DFID, PFID(fid)); + + i_size_write(local, rc); + LDISKFS_I(local)->i_disksize = rc; + } unlock_new_inode(local); /* Agent inode should not have project ID */ #ifdef HAVE_PROJECT_QUOTA if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL && i_projid_read(pobj->oo_inode) != 0) { - rc = ldiskfs_transfer_project(local, 0); + rc = osd_transfer_project(local, 0, th); if (rc) { - CERROR("%s: quota transfer failed: rc = %d. Is project " - "quota enforcement enabled on the ldiskfs " - "filesystem?\n", local->i_sb->s_id, rc); + CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n", + osd_ino2name(local), rc); RETURN(ERR_PTR(rc)); } } @@ -3581,7 +4031,7 @@ static int osd_schedule_agent_inode_removal(const struct lu_env *env, struct osd_thandle *oh, __u32 ino) { - struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); + struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev); struct osd_obj_orphan *oor; OBD_ALLOC_PTR(oor); @@ -3606,20 +4056,16 @@ static int osd_process_scheduled_agent_removals(const struct lu_env *env, struct osd_thread_info *info = osd_oti_get(env); struct osd_obj_orphan *oor, *tmp; struct osd_inode_id id; - struct list_head list; + LIST_HEAD(list); struct inode *inode; struct lu_fid fid; handle_t *jh; __u32 ino; - INIT_LIST_HEAD(&list); - spin_lock(&osd->od_osfs_lock); list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) { - if (oor->oor_env == env) { - list_del(&oor->oor_list); - list_add(&oor->oor_list, &list); - } + if (oor->oor_env == env) + list_move(&oor->oor_list, &list); } spin_unlock(&osd->od_osfs_lock); @@ -3662,10 +4108,10 @@ static int osd_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) { - 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, on_ost = 0; + 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, on_ost = 0; ENTRY; @@ -3678,8 +4124,10 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt, LASSERT(th != NULL); if (unlikely(fid_is_acct(fid))) - /* Quota files can't be created from the kernel any more, - * 'tune2fs -O quota' will take care of creating them */ + /* + * Quota files can't be created from the kernel any more, + * 'tune2fs -O quota' will take care of creating them + */ RETURN(-EPERM); result = __osd_create(info, obj, attr, hint, dof, th); @@ -3705,14 +4153,30 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt, obj->oo_dt.do_body_ops = &osd_body_ops; } - if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) + if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) { + struct inode *inode = obj->oo_inode; + result = __osd_oi_insert(env, obj, fid, th); + if (result && inode) { + spin_lock(&obj->oo_guard); + clear_nlink(inode); + spin_unlock(&obj->oo_guard); + osd_dirty_inode(inode, I_DIRTY_DATASYNC); + ldiskfs_set_inode_state(inode, + LDISKFS_STATE_LUSTRE_DESTROY); + iput(inode); + obj->oo_inode = NULL; + } + } - /* a small optimization - dt_insert() isn't usually applied + /* + * a small optimization - dt_insert() isn't usually applied * to OST objects, so we don't need to cache OI mapping for - * OST objects */ + * OST objects + */ if (result == 0 && on_ost == 0) { struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + result = osd_idc_find_and_init(env, osd, obj); LASSERT(result == 0); } @@ -3727,19 +4191,21 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *handle) { struct osd_thandle *oh; + int rc; /* it's possible that object doesn't exist yet */ LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_REF_ADD, osd_dto_credits_noquota[DTO_ATTR_SET_BASE]); - osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt)); + rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), + osd_dt_obj(dt)); - return 0; + return rc; } /* @@ -3748,10 +4214,10 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt, static int osd_ref_add(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - int rc = 0; + int rc = 0; if (!dt_object_exists(dt) || obj->oo_destroyed) return -ENOENT; @@ -3761,7 +4227,7 @@ static int osd_ref_add(const struct lu_env *env, struct dt_object *dt, LASSERT(osd_is_write_locked(env, obj)); LASSERT(th != NULL); - oh = container_of0(th, struct osd_thandle, ot_super); + oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_ADD); @@ -3790,7 +4256,7 @@ static int osd_ref_add(const struct lu_env *env, struct dt_object *dt, } spin_unlock(&obj->oo_guard); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + osd_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_REF_ADD); @@ -3809,7 +4275,7 @@ static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_REF_DEL, @@ -3824,10 +4290,10 @@ static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt, static int osd_ref_del(const struct lu_env *env, struct dt_object *dt, 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; + 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; @@ -3837,15 +4303,20 @@ static int osd_ref_del(const struct lu_env *env, struct dt_object *dt, LASSERT(osd_is_write_locked(env, obj)); LASSERT(th != NULL); - oh = container_of0(th, struct osd_thandle, ot_super); + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL)) + return -EIO; + + oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle != NULL); osd_trans_exec_op(env, th, OSD_OT_REF_DEL); spin_lock(&obj->oo_guard); - /* That can be result of upgrade from old Lustre version and + /* + * That can be result of upgrade from old Lustre version and * applied only to local files. Just skip this ref_del call. - * ext4_unlink() only treats this as a warning, don't LASSERT here.*/ + * ext4_unlink() only treats this as a warning, don't LASSERT here. + */ if (inode->i_nlink == 0) { CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ? D_ERROR : D_INODE, "%s: nlink == 0 on "DFID @@ -3861,7 +4332,7 @@ static int osd_ref_del(const struct lu_env *env, struct dt_object *dt, ldiskfs_dec_count(oh->ot_handle, inode); spin_unlock(&obj->oo_guard); - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + osd_dirty_inode(inode, I_DIRTY_DATASYNC); LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_REF_DEL); @@ -3879,8 +4350,8 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, struct inode *inode = obj->oo_inode; struct osd_thread_info *info = osd_oti_get(env); struct dentry *dentry = &info->oti_obj_dentry; - bool cache_xattr = false; - int rc; + bool cache_xattr = false; + int rc; LASSERT(buf); @@ -3888,8 +4359,10 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, if (strcmp(name, XATTR_NAME_VERSION) == 0) { dt_obj_version_t *ver = buf->lb_buf; - /* for version we are just using xattr API but change inode - * field instead */ + /* + * for version we are just using xattr API but change inode + * field instead + */ if (buf->lb_len == 0) return sizeof(dt_obj_version_t); @@ -3909,7 +4382,9 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(inode->i_op != NULL); +#ifdef HAVE_IOP_XATTR LASSERT(inode->i_op->getxattr != NULL); +#endif if (strcmp(name, XATTR_NAME_LOV) == 0 || strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) @@ -3972,9 +4447,9 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, } static int osd_declare_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 dt_object *dt, + const struct lu_buf *buf, const char *name, + int fl, struct thandle *handle) { struct osd_thandle *oh; int credits = 0; @@ -3982,13 +4457,15 @@ static int osd_declare_xattr_set(const struct lu_env *env, LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); if (strcmp(name, XATTR_NAME_LMA) == 0) { - /* For non-upgrading case, the LMA is set first and + /* + * For non-upgrading case, the LMA is set first and * usually fit inode. But for upgrade case, the LMA - * may be in another separated EA block. */ + * may be in another separated EA block. + */ if (dt_object_exists(dt)) { if (fl == LU_XATTR_REPLACE) credits = 1; @@ -4005,12 +4482,14 @@ static int osd_declare_xattr_set(const struct lu_env *env, else goto upgrade; } else { - /* If some name entry resides on remote MDT, then will create + /* + * If some name entry resides on remote MDT, then will create * agent entry under remote parent. On the other hand, if the * remote entry will be removed, then related agent entry may * need to be removed from the remote parent. So there may be * kinds of cases, let's declare enough credits. The credits - * for create agent entry is enough for remove case. */ + * for create agent entry is enough for remove case. + */ if (strcmp(name, XATTR_NAME_LINK) == 0) { credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; if (dt_object_exists(dt)) @@ -4024,8 +4503,10 @@ upgrade: ssize_t buflen; if (buf->lb_buf == NULL && dt_object_exists(dt)) { - /* learn xattr size from osd_xattr_get if - attribute has not been read yet */ + /* + * learn xattr size from osd_xattr_get if + * attribute has not been read yet + */ buflen = __osd_xattr_get( osd_dt_obj(dt)->oo_inode, &osd_oti_get(env)->oti_obj_dentry, @@ -4068,6 +4549,7 @@ static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj, struct ost_layout *ol = &ff->ff_layout; int flags = XATTR_REPLACE; int rc; + ENTRY; if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid)) @@ -4104,15 +4586,17 @@ static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj, if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) RETURN(fl); - /* Old client does not send stripe information, - * then store the PFID EA on disk separatedly. */ + /* + * Old client does not send stripe information, + * then store the PFID EA on disk separatedly. + */ if (unlikely(buf->lb_len == sizeof(struct lu_fid) || ol->ol_stripe_size == 0)) RETURN(fl); /* Remove old PFID EA entry firstly. */ - ll_vfs_dq_init(inode); - rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID); + dquot_initialize(inode); + rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID); if (rc == -ENODATA) { if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE)) RETURN(rc); @@ -4189,9 +4673,10 @@ static int osd_xattr_handle_linkea(const struct lu_env *env, struct osd_thandle *oh; int rc; bool remote = false; + ENTRY; - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle != NULL); rc = linkea_init_with_rec(&ldata); @@ -4240,6 +4725,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, int fs_flags = 0; int len; int rc; + ENTRY; LASSERT(handle); @@ -4249,17 +4735,21 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (strcmp(name, XATTR_NAME_VERSION) == 0) { dt_obj_version_t *version = buf->lb_buf; - /* for version we are just using xattr API but change inode - * field instead */ + /* + * for version we are just using xattr API but change inode + * field instead + */ LASSERT(buf->lb_len == sizeof(dt_obj_version_t)); CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n", *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino); LDISKFS_I(inode)->i_fs_version = *version; - /* Version is set after all inode operations are finished, - * so we should mark it dirty here */ - ll_dirty_inode(inode, I_DIRTY_DATASYNC); + /* + * Version is set after all inode operations are finished, + * so we should mark it dirty here + */ + osd_dirty_inode(inode, I_DIRTY_DATASYNC); RETURN(0); } @@ -4270,10 +4760,12 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, len = buf->lb_len; osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); - /* For the OST device with 256 bytes inode size by default, + /* + * For the OST device with 256 bytes inode size by default, * the PFID EA will be stored together with LMA EA to avoid * performance trouble. Otherwise the PFID EA can be stored - * independently. LU-8998 */ + * independently. LU-8998 + */ if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost && (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) { LASSERT(buf->lb_buf); @@ -4326,10 +4818,10 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt, static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt, 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; if (!dt_object_exists(dt)) return -ENOENT; @@ -4353,7 +4845,7 @@ static int osd_declare_xattr_del(const struct lu_env *env, LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, @@ -4373,19 +4865,21 @@ static int osd_declare_xattr_del(const struct lu_env *env, static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, 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; if (!dt_object_exists(dt)) return -ENOENT; LASSERT(!dt_object_remote(dt)); LASSERT(inode->i_op != NULL); - LASSERT(inode->i_op->removexattr != NULL); LASSERT(handle != NULL); +#ifdef HAVE_IOP_XATTR + LASSERT(inode->i_op->removexattr != NULL); +#endif osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET); @@ -4405,10 +4899,10 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, obj->oo_pfid_in_lma = 0; } } else { - ll_vfs_dq_init(inode); + dquot_initialize(inode); dentry->d_inode = inode; dentry->d_sb = inode->i_sb; - rc = inode->i_op->removexattr(dentry, name); + rc = ll_vfs_removexattr(dentry, inode, name); } osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET); @@ -4424,23 +4918,14 @@ static int osd_xattr_del(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; - struct osd_thread_info *info = osd_oti_get(env); - struct dentry *dentry = &info->oti_obj_dentry; - struct file *file = &info->oti_file; - int rc; + struct osd_object *obj = osd_dt_obj(dt); + struct inode *inode = obj->oo_inode; + struct file *file = osd_quasi_file(env, inode); + int rc; ENTRY; - dentry->d_inode = inode; - dentry->d_sb = inode->i_sb; - file->f_path.dentry = dentry; - file->f_mapping = inode->i_mapping; - file->f_op = inode->i_fop; - set_file_inode(file, inode); - - rc = ll_vfs_fsync_range(file, start, end, 0); + rc = vfs_fsync_range(file, start, end, 0); RETURN(rc); } @@ -4455,53 +4940,51 @@ static int osd_invalidate(const struct lu_env *env, struct dt_object *dt) */ static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o, - const struct dt_index_features *feat) -{ - struct iam_descr *descr; - - if (osd_object_is_root(o)) - return feat == &dt_directory_features; - - LASSERT(o->oo_dir != NULL); - - descr = o->oo_dir->od_container.ic_descr; - if (feat == &dt_directory_features) { - if (descr->id_rec_size == sizeof(struct osd_fid_pack)) - return 1; - else - return 0; - } else { - return - feat->dif_keysize_min <= descr->id_key_size && - descr->id_key_size <= feat->dif_keysize_max && - feat->dif_recsize_min <= descr->id_rec_size && - descr->id_rec_size <= feat->dif_recsize_max && - !(feat->dif_flags & (DT_IND_VARKEY | - DT_IND_VARREC | DT_IND_NONUNQ)) && - ergo(feat->dif_flags & DT_IND_UPDATE, - 1 /* XXX check that object (and file system) is - * writable */); - } + const struct dt_index_features *feat) +{ + struct iam_descr *descr; + + if (osd_object_is_root(o)) + return feat == &dt_directory_features; + + LASSERT(o->oo_dir != NULL); + + descr = o->oo_dir->od_container.ic_descr; + if (feat == &dt_directory_features) { + if (descr->id_rec_size == sizeof(struct osd_fid_pack)) + return 1; + else + return 0; + } else { + return feat->dif_keysize_min <= descr->id_key_size && + descr->id_key_size <= feat->dif_keysize_max && + feat->dif_recsize_min <= descr->id_rec_size && + descr->id_rec_size <= feat->dif_recsize_max && + !(feat->dif_flags & (DT_IND_VARKEY | + DT_IND_VARREC | DT_IND_NONUNQ)) && + ergo(feat->dif_flags & DT_IND_UPDATE, + 1 /* XXX check that object (and fs) is writable */); + } } static int osd_iam_container_init(const struct lu_env *env, - struct osd_object *obj, - struct osd_directory *dir) + struct osd_object *obj, + struct osd_directory *dir) { - struct iam_container *bag = &dir->od_container; - int result; + struct iam_container *bag = &dir->od_container; + int result; - result = iam_container_init(bag, &dir->od_descr, obj->oo_inode); - if (result != 0) - return result; + result = iam_container_init(bag, &dir->od_descr, obj->oo_inode); + if (result != 0) + return result; - result = iam_container_setup(bag); - if (result == 0) - obj->oo_dt.do_index_ops = &osd_index_iam_ops; - else - iam_container_fini(bag); + result = iam_container_setup(bag); + if (result == 0) + obj->oo_dt.do_index_ops = &osd_index_iam_ops; + else + iam_container_fini(bag); - return result; + return result; } @@ -4509,23 +4992,23 @@ static int osd_iam_container_init(const struct lu_env *env, * Concurrency: no external locking is necessary. */ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, - const struct dt_index_features *feat) + const struct dt_index_features *feat) { - int result; - int skip_iam = 0; - struct osd_object *obj = osd_dt_obj(dt); + int result; + int skip_iam = 0; + struct osd_object *obj = osd_dt_obj(dt); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - if (osd_object_is_root(obj)) { - dt->do_index_ops = &osd_index_ea_ops; - result = 0; + if (osd_object_is_root(obj)) { + dt->do_index_ops = &osd_index_ea_ops; + result = 0; } else if (feat == &dt_directory_features) { - dt->do_index_ops = &osd_index_ea_ops; + dt->do_index_ops = &osd_index_ea_ops; if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode)) - result = 0; - else - result = -ENOTDIR; + result = 0; + else + result = -ENOTDIR; skip_iam = 1; } else if (unlikely(feat == &dt_otable_features)) { dt->do_index_ops = &osd_otable_ops; @@ -4589,20 +5072,20 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, unlock: up_write(&obj->oo_ext_idx_sem); - } else { - result = -ENOMEM; - } - } else { - result = 0; - } + } else { + result = -ENOMEM; + } + } else { + result = 0; + } if (result == 0 && skip_iam == 0) { - if (!osd_iam_index_probe(env, obj, feat)) - result = -ENOTDIR; - } - LINVRNT(osd_invariant(obj)); + if (!osd_iam_index_probe(env, obj, feat)) + result = -ENOTDIR; + } + LINVRNT(osd_invariant(obj)); - return result; + return result; } static int osd_otable_it_attr_get(const struct lu_env *env, @@ -4648,14 +5131,14 @@ static const struct dt_object_operations osd_obj_otable_it_ops = { }; static int osd_index_declare_iam_delete(const struct lu_env *env, - struct dt_object *dt, - const struct dt_key *key, - struct thandle *handle) + struct dt_object *dt, + const struct dt_key *key, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); /* Recycle may cause additional three blocks to be changed. */ osd_trans_declare_op(env, oh, OSD_OT_DELETE, @@ -4680,11 +5163,12 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, 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; + 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; if (!dt_object_exists(dt)) @@ -4697,13 +5181,13 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, handle, OSD_OT_DELETE); - ipd = osd_idx_ipd_get(env, bag); - if (unlikely(ipd == NULL)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* swab quota uid/gid provided by caller */ @@ -4711,11 +5195,11 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, key = (const struct dt_key *)&oti->oti_quota_id; } - rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); - osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd); + osd_ipd_put(env, bag, ipd); + LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); - RETURN(rc); + RETURN(rc); } static int osd_index_declare_ea_delete(const struct lu_env *env, @@ -4724,14 +5208,15 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; - struct inode *inode; - int rc, credits; + struct inode *inode; + int rc, credits; + ENTRY; LASSERT(!dt_object_remote(dt)); LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); credits = osd_dto_credits_noquota[DTO_INDEX_DELETE]; @@ -4748,33 +5233,36 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, } static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, - struct dt_rec *fid) + struct dt_rec *fid) { - struct osd_fid_pack *rec; - int rc = -ENODATA; + struct osd_fid_pack *rec; + int rc = -ENODATA; - if (de->file_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack *) (de->name + de->name_len + 1); - rc = osd_fid_unpack((struct lu_fid *)fid, rec); + if (de->file_type & LDISKFS_DIRENT_LUFID) { + rec = (struct osd_fid_pack *)(de->name + de->name_len + 1); + rc = osd_fid_unpack((struct lu_fid *)fid, rec); if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid))) rc = -EINVAL; - } + } return rc; } static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd, const struct lu_fid *fid) { - struct seq_server_site *ss = osd_seq_site(osd); + 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 + /* + * If FLD is not being initialized yet, it only happens during the * initialization, likely during mgs initialization, and we assume - * this is local FID. */ + * this is local FID. + */ if (ss == NULL || ss->ss_server_fld == NULL) RETURN(0); @@ -4802,8 +5290,10 @@ static void osd_take_care_of_agent(const struct lu_env *env, if (IS_ERR(idc) || idc->oic_remote) schedule = 1; } else if (rc == -ENODATA) { - /* can't get FID, postpone to the end of the - * transaction when iget() is safe */ + /* + * can't get FID, postpone to the end of the + * transaction when iget() is safe + */ schedule = 1; } else { CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc); @@ -4826,15 +5316,16 @@ static void osd_take_care_of_agent(const struct lu_env *env, static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, 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 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 osd_device *osd = osd_dev(dt->do_lu.lo_dev); - int rc; + struct buffer_head *bh; + struct htree_lock *hlock = NULL; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4846,25 +5337,26 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, handle, OSD_OT_DELETE); - oh = container_of(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); - ll_vfs_dq_init(dir); - dentry = osd_child_dentry_get(env, obj, - (char *)key, strlen((char *)key)); + dquot_initialize(dir); + 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_DEL); - } 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_DEL); + } else { down_write(&obj->oo_ext_idx_sem); - } + } - bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); + bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock); if (!IS_ERR(bh)) { - /* If this is not the ".." entry, it might be a remote DNE + /* + * If this is not the ".." entry, it might be a remote DNE * entry and we need to check if the FID is for a remote * MDT. If the FID is not in the directory entry (e.g. * upgraded 1.8 filesystem without dirdata enabled) then @@ -4877,7 +5369,8 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, * the entry first might leak the agent inode afterward. The * reverse would need filesystem abort in case of error deleting * the entry after the agent had been removed, or leave a - * dangling entry pointing at a random inode. */ + * dangling entry pointing at a random inode. + */ if (strcmp((char *)key, dotdot) != 0) osd_take_care_of_agent(env, osd, oh, de); rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh); @@ -4885,17 +5378,26 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, } else { rc = PTR_ERR(bh); } - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else - up_write(&obj->oo_ext_idx_sem); + if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) && + obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) { + /* NB, dirent count may not be accurate, because it's counted + * without lock. + */ + if (obj->oo_dirent_count) + obj->oo_dirent_count--; + else + obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET; + } + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_write(&obj->oo_ext_idx_sem); GOTO(out, rc); - out: - LASSERT(osd_invariant(obj)); + LASSERT(osd_invariant(obj)); osd_trans_exec_check(env, handle, OSD_OT_DELETE); - RETURN(rc); + RETURN(rc); } /** @@ -4912,13 +5414,14 @@ out: 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 osd_object *obj = osd_dt_obj(dt); - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; + 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 iam_iterator *it = &oti->oti_idx_it; + struct iam_rec *iam_rec; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -4928,12 +5431,12 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, LASSERT(!dt_object_remote(dt)); LASSERT(bag->ic_object == obj->oo_inode); - ipd = osd_idx_ipd_get(env, bag); - if (IS_ERR(ipd)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (IS_ERR(ipd)) + RETURN(-ENOMEM); - /* got ipd now we can start iterator. */ - iam_it_init(it, bag, 0, ipd); + /* got ipd now we can start iterator. */ + iam_it_init(it, bag, 0, ipd); if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* swab quota uid/gid provided by caller */ @@ -4941,43 +5444,43 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, key = (const struct dt_key *)&oti->oti_quota_id; } - rc = iam_it_get(it, (struct iam_key *)key); - if (rc >= 0) { - if (S_ISDIR(obj->oo_inode->i_mode)) - iam_rec = (struct iam_rec *)oti->oti_ldp; - else - iam_rec = (struct iam_rec *) rec; + rc = iam_it_get(it, (struct iam_key *)key); + if (rc >= 0) { + if (S_ISDIR(obj->oo_inode->i_mode)) + iam_rec = (struct iam_rec *)oti->oti_ldp; + else + iam_rec = (struct iam_rec *)rec; - iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec); + iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec); - if (S_ISDIR(obj->oo_inode->i_mode)) - osd_fid_unpack((struct lu_fid *) rec, - (struct osd_fid_pack *)iam_rec); + if (S_ISDIR(obj->oo_inode->i_mode)) + osd_fid_unpack((struct lu_fid *)rec, + (struct osd_fid_pack *)iam_rec); else if (fid_is_quota(lu_object_fid(&dt->do_lu))) osd_quota_unpack(obj, rec); - } + } - iam_it_put(it); - iam_it_fini(it); - osd_ipd_put(env, bag, ipd); + iam_it_put(it); + iam_it_fini(it); + osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + LINVRNT(osd_invariant(obj)); - RETURN(rc); + RETURN(rc); } static int osd_index_declare_iam_insert(const struct lu_env *env, - struct dt_object *dt, - const struct dt_rec *rec, - const struct dt_key *key, - struct thandle *handle) + struct dt_object *dt, + const struct dt_rec *rec, + const struct dt_key *key, + struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; - LASSERT(handle != NULL); + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); osd_trans_declare_op(env, oh, OSD_OT_INSERT, osd_dto_credits_noquota[DTO_INDEX_INSERT]); @@ -4998,16 +5501,16 @@ static int osd_index_declare_iam_insert(const struct lu_env *env, */ 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, - int ignore_quota) + const struct dt_key *key, struct thandle *th) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); struct iam_path_descr *ipd; - struct osd_thandle *oh; - struct iam_container *bag; + struct osd_thandle *oh; + struct iam_container *bag; struct osd_thread_info *oti = osd_oti_get(env); - struct iam_rec *iam_rec; - int rc; + struct iam_rec *iam_rec; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -5022,16 +5525,17 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, osd_trans_exec_op(env, th, OSD_OT_INSERT); - ipd = osd_idx_ipd_get(env, bag); - if (unlikely(ipd == NULL)) - RETURN(-ENOMEM); + ipd = osd_idx_ipd_get(env, bag); + if (unlikely(ipd == NULL)) + RETURN(-ENOMEM); - oh = container_of0(th, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle != NULL); - LASSERT(oh->ot_handle->h_transaction != NULL); + oh = container_of(th, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle != NULL); + LASSERT(oh->ot_handle->h_transaction != NULL); if (S_ISDIR(obj->oo_inode->i_mode)) { iam_rec = (struct iam_rec *)oti->oti_ldp; - osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid); + osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, + &oti->oti_fid); } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) { /* pack quota uid/gid */ oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); @@ -5043,12 +5547,12 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, iam_rec = (struct iam_rec *)rec; } - rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, - iam_rec, ipd); - osd_ipd_put(env, bag, ipd); - LINVRNT(osd_invariant(obj)); + rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, + iam_rec, ipd); + osd_ipd_put(env, bag, ipd); + LINVRNT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_INSERT); - RETURN(rc); + RETURN(rc); } /** @@ -5064,14 +5568,14 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, const char *name, const struct lu_fid *fid, struct htree_lock *hlock, struct thandle *th) { - struct ldiskfs_dentry_param *ldp; - struct dentry *child; - struct osd_thandle *oth; - int rc; + struct ldiskfs_dentry_param *ldp; + struct dentry *child; + struct osd_thandle *oth; + int rc; - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle != NULL); - LASSERT(oth->ot_handle->h_transaction != NULL); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle != NULL); + LASSERT(oth->ot_handle->h_transaction != NULL); LASSERT(pobj->oo_inode); ldp = (struct ldiskfs_dentry_param *)info->oti_ldp; @@ -5082,7 +5586,7 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, osd_get_ldiskfs_dirent_param(ldp, fid); child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name)); child->d_fsdata = (void *)ldp; - ll_vfs_dq_init(pobj->oo_inode); + dquot_initialize(pobj->oo_inode); rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle, child, cinode, hlock); if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) { @@ -5131,22 +5635,22 @@ static int osd_add_dot_dotdot(struct osd_thread_info *info, const struct lu_fid *dot_dot_fid, struct thandle *th) { - struct inode *inode = dir->oo_inode; - struct osd_thandle *oth; - int result = 0; + struct inode *inode = dir->oo_inode; + struct osd_thandle *oth; + int result = 0; - oth = container_of(th, struct osd_thandle, ot_super); - LASSERT(oth->ot_handle->h_transaction != NULL); - LASSERT(S_ISDIR(dir->oo_inode->i_mode)); + oth = container_of(th, struct osd_thandle, ot_super); + LASSERT(oth->ot_handle->h_transaction != NULL); + LASSERT(S_ISDIR(dir->oo_inode->i_mode)); - if (strcmp(name, dot) == 0) { - if (dir->oo_compat_dot_created) { - result = -EEXIST; - } else { + if (strcmp(name, dot) == 0) { + if (dir->oo_compat_dot_created) { + result = -EEXIST; + } else { LASSERT(inode->i_ino == parent_dir->i_ino); - dir->oo_compat_dot_created = 1; - result = 0; - } + dir->oo_compat_dot_created = 1; + result = 0; + } } else if (strcmp(name, dotdot) == 0) { if (!dir->oo_compat_dot_created) return -EINVAL; @@ -5185,31 +5689,31 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, 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; - int rc; + struct osd_thread_info *info = osd_oti_get(env); + struct htree_lock *hlock; + int rc; - hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL; + hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL; - if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' && - name[2] =='\0'))) { - if (hlock != NULL) { - ldiskfs_htree_lock(hlock, pobj->oo_hl_head, - pobj->oo_inode, 0); - } else { + if (name[0] == '.' && (name[1] == '\0' || + (name[1] == '.' && name[2] == '\0'))) { + if (hlock != NULL) { + ldiskfs_htree_lock(hlock, pobj->oo_hl_head, + pobj->oo_inode, 0); + } else { down_write(&pobj->oo_ext_idx_sem); } rc = osd_add_dot_dotdot(info, pobj, cinode, name, lu_object_fid(&pobj->oo_dt.do_lu), - fid, th); - } else { - if (hlock != NULL) { - ldiskfs_htree_lock(hlock, pobj->oo_hl_head, - pobj->oo_inode, LDISKFS_HLOCK_ADD); - } else { + fid, th); + } else { + if (hlock != NULL) { + ldiskfs_htree_lock(hlock, pobj->oo_hl_head, + pobj->oo_inode, LDISKFS_HLOCK_ADD); + } else { down_write(&pobj->oo_ext_idx_sem); - } + } if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) { struct lu_fid *tfid = &info->oti_fid; @@ -5222,13 +5726,17 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj, rc = __osd_ea_add_rec(info, pobj, cinode, name, fid, hlock, th); } - } - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else + } + if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu)) + && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) + pobj->oo_dirent_count++; + + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else up_write(&pobj->oo_ext_idx_sem); - return rc; + return rc; } static int @@ -5242,13 +5750,13 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, int once = 0; bool insert; int rc; + ENTRY; if (!fid_is_norm(fid) && !fid_is_igif(fid)) RETURN(0); - if (thread_is_running(&scrub->os_thread) && - scrub->os_pos_current > id->oii_ino) + if (scrub->os_running && scrub->os_pos_current > id->oii_ino) RETURN(0); if (dev->od_auto_scrub_interval == AS_NEVER || @@ -5277,8 +5785,10 @@ again: if (gen != OSD_OII_NOGEN) goto trigger; - /* The inode may has been reused by others, we do not know, - * leave it to be handled by subsequent osd_fid_lookup(). */ + /* + * The inode may has been reused by others, we do not know, + * leave it to be handled by subsequent osd_fid_lookup(). + */ GOTO(out, rc = 0); } else if (rc || osd_id_eq(id, &oti->oti_id)) { GOTO(out, rc); @@ -5287,7 +5797,7 @@ again: insert = false; trigger: - if (thread_is_running(&scrub->os_thread)) { + if (scrub->os_running) { if (inode == NULL) { inode = osd_iget(oti, dev, id); /* The inode has been removed (by race maybe). */ @@ -5299,10 +5809,12 @@ trigger: } rc = osd_oii_insert(dev, oic, insert); - /* There is race condition between osd_oi_lookup and OI scrub. + /* + * There is race condition between osd_oi_lookup and OI scrub. * The OI scrub finished just after osd_oi_lookup() failure. * Under such case, it is unnecessary to trigger OI scrub again, - * but try to call osd_oi_lookup() again. */ + * but try to call osd_oi_lookup() again. + */ if (unlikely(rc == -EAGAIN)) goto again; @@ -5318,8 +5830,7 @@ trigger: rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT); CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING, - "%s: trigger partial OI scrub for RPC inconsistency " - "checking FID "DFID": rc = %d\n", + "%s: trigger partial OI scrub for RPC inconsistency checking FID "DFID": rc = %d\n", osd_dev2name(dev), PFID(fid), rc); if (rc == 0 || rc == -EALREADY) goto again; @@ -5395,12 +5906,13 @@ 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; + 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); @@ -5458,6 +5970,7 @@ static int osd_verify_ent_by_linkea(const struct lu_env *env, struct lu_name cname = { .ln_name = name, .ln_namelen = namelen }; int rc; + ENTRY; again: @@ -5499,14 +6012,15 @@ again: 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 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); @@ -5542,12 +6056,14 @@ 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(is_remote_parent_ino(dev, ino))) { const char *name = (const char *)key; - /* If the parent is on remote MDT, and there + /* + * 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. */ + * the parent FID from the linkEA. + */ if (likely(strlen(name) == 2 && name[0] == '.' && name[1] == '.')) rc = osd_get_pfid_from_linkea(env, obj, @@ -5592,12 +6108,13 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, const struct dt_key *key, struct thandle *handle) { - struct osd_thandle *oh; - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; - const struct lu_fid *fid = rec1->rec_fid; - int credits, rc = 0; - struct osd_idmap_cache *idc; + struct osd_thandle *oh; + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec; + const struct lu_fid *fid = rec1->rec_fid; + int credits, rc = 0; + struct osd_idmap_cache *idc; + ENTRY; LASSERT(!dt_object_remote(dt)); @@ -5605,21 +6122,25 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, LASSERT(fid != NULL); LASSERT(rec1->rec_type != 0); - oh = container_of0(handle, struct osd_thandle, ot_super); + oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle == NULL); credits = osd_dto_credits_noquota[DTO_INDEX_INSERT]; - /* we can't call iget() while a transactions is running + /* + * we can't call iget() while a transactions is running * (this can lead to a deadlock), but we need to know * inum and object type. so we find this information at - * declaration and cache in per-thread info */ + * declaration and cache in per-thread info + */ idc = osd_idc_find_or_init(env, osd, fid); if (IS_ERR(idc)) RETURN(PTR_ERR(idc)); if (idc->oic_remote) { - /* a reference to remote inode is represented by an - * agent inode which we have to create */ + /* + * a reference to remote inode is represented by an + * agent inode which we have to create + */ credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE]; credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; } @@ -5629,9 +6150,11 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, 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 + /* + * We ignore block quota on meta pool (MDTs), so needn't * calculate how many blocks will be consumed by this index - * insert */ + * insert + */ rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), i_projid_read(inode), 0, @@ -5641,7 +6164,8 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, RETURN(rc); #ifdef HAVE_PROJECT_QUOTA - /* Reserve credits for local agent inode to transfer + /* + * Reserve credits for local agent inode to transfer * to 0, quota enforcement is ignored in this case. */ if (idc->oic_remote && @@ -5653,7 +6177,6 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, #endif } - RETURN(rc); } @@ -5670,18 +6193,18 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, */ 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, - int ignore_quota) -{ - struct osd_object *obj = osd_dt_obj(dt); - struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); - 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 inode *child_inode = NULL; - struct osd_idmap_cache *idc; - int rc; + const struct dt_key *key, struct thandle *th) +{ + struct osd_object *obj = osd_dt_obj(dt); + struct osd_device *osd = osd_dev(dt->do_lu.lo_dev); + 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 inode *child_inode = NULL; + struct osd_idmap_cache *idc; + int rc; + ENTRY; if (!dt_object_exists(dt)) @@ -5697,17 +6220,24 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, idc = osd_idc_find(env, osd, fid); if (unlikely(idc == NULL)) { - /* this dt_insert() wasn't declared properly, so - * FID is missing in OI cache. we better do not - * lookup FID in FLDB/OI and don't risk to deadlock, - * but in some special cases (lfsck testing, etc) - * it's much simpler than fixing a caller */ - CERROR("%s: "DFID" wasn't declared for insert\n", - osd_name(osd), PFID(fid)); - dump_stack(); idc = osd_idc_find_or_init(env, osd, fid); - if (IS_ERR(idc)) + if (IS_ERR(idc)) { + /* + * this dt_insert() wasn't declared properly, so + * FID is missing in OI cache. we better do not + * lookup FID in FLDB/OI and don't risk to deadlock, + * but in some special cases (lfsck testing, etc) + * it's much simpler than fixing a caller. + * + * normally this error should be placed after the first + * find, but migrate may attach source stripes to + * target, which doesn't create stripes. + */ + CERROR("%s: "DFID" wasn't declared for insert\n", + osd_name(osd), PFID(fid)); + dump_stack(); RETURN(PTR_ERR(idc)); + } } if (idc->oic_remote) { @@ -5732,6 +6262,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child_inode = oti->oti_inode; if (unlikely(child_inode == NULL)) { struct ldiskfs_inode_info *lii; + OBD_ALLOC_PTR(lii); if (lii == NULL) RETURN(-ENOMEM); @@ -5751,6 +6282,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, iput(child_inode); LASSERT(osd_invariant(obj)); osd_trans_exec_check(env, th, OSD_OT_INSERT); + RETURN(rc); } @@ -5764,11 +6296,11 @@ static struct dt_it *osd_it_iam_init(const struct lu_env *env, struct dt_object *dt, __u32 unused) { - 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; + 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); @@ -5795,8 +6327,8 @@ 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); @@ -5816,10 +6348,10 @@ static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di) */ static int osd_it_iam_get(const struct lu_env *env, - struct dt_it *di, const struct dt_key *key) + struct dt_it *di, const struct dt_key *key) { - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_it_iam *it = (struct osd_it_iam *)di; if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) { /* swab quota uid/gid */ @@ -5827,7 +6359,7 @@ static int osd_it_iam_get(const struct lu_env *env, key = (struct dt_key *)&oti->oti_quota_id; } - return iam_it_get(&it->oi_it, (const struct iam_key *)key); + return iam_it_get(&it->oi_it, (const struct iam_key *)key); } /** @@ -5837,9 +6369,9 @@ static int osd_it_iam_get(const struct lu_env *env, */ static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - iam_it_put(&it->oi_it); + iam_it_put(&it->oi_it); } /** @@ -5854,9 +6386,9 @@ static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di) static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_next(&it->oi_it); + return iam_it_next(&it->oi_it); } /** @@ -5864,12 +6396,12 @@ static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di) */ static struct dt_key *osd_it_iam_key(const struct lu_env *env, - const struct dt_it *di) + const struct dt_it *di) { struct osd_thread_info *oti = osd_oti_get(env); - struct osd_it_iam *it = (struct osd_it_iam *)di; - struct osd_object *obj = it->oi_obj; - struct dt_key *key; + struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_object *obj = it->oi_obj; + struct dt_key *key; key = (struct dt_key *)iam_it_key_get(&it->oi_it); @@ -5888,9 +6420,9 @@ static struct dt_key *osd_it_iam_key(const struct lu_env *env, static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_key_size(&it->oi_it); + return iam_it_key_size(&it->oi_it); } static inline void @@ -5935,21 +6467,22 @@ osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset, * Return pointer to the record under iterator. */ static int osd_it_iam_rec(const struct lu_env *env, - const struct dt_it *di, - struct dt_rec *dtrec, __u32 attr) + const struct dt_it *di, + struct dt_rec *dtrec, __u32 attr) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; struct osd_thread_info *info = osd_oti_get(env); + ENTRY; if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) { const struct osd_fid_pack *rec; - struct lu_fid *fid = &info->oti_fid; - struct lu_dirent *lde = (struct lu_dirent *)dtrec; - char *name; - int namelen; - __u64 hash; - int rc; + struct lu_fid *fid = &info->oti_fid; + struct lu_dirent *lde = (struct lu_dirent *)dtrec; + char *name; + int namelen; + __u64 hash; + int rc; name = (char *)iam_it_key_get(&it->oi_it); if (IS_ERR(name)) @@ -5987,9 +6520,9 @@ static int osd_it_iam_rec(const struct lu_env *env, */ static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_store(&it->oi_it); + return iam_it_store(&it->oi_it); } /** @@ -6004,31 +6537,31 @@ static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di) */ static int osd_it_iam_load(const struct lu_env *env, - const struct dt_it *di, __u64 hash) + const struct dt_it *di, __u64 hash) { - struct osd_it_iam *it = (struct osd_it_iam *)di; + struct osd_it_iam *it = (struct osd_it_iam *)di; - return iam_it_load(&it->oi_it, hash); + return iam_it_load(&it->oi_it, hash); } static const struct dt_index_operations osd_index_iam_ops = { - .dio_lookup = osd_index_iam_lookup, - .dio_declare_insert = osd_index_declare_iam_insert, - .dio_insert = osd_index_iam_insert, - .dio_declare_delete = osd_index_declare_iam_delete, - .dio_delete = osd_index_iam_delete, - .dio_it = { - .init = osd_it_iam_init, - .fini = osd_it_iam_fini, - .get = osd_it_iam_get, - .put = osd_it_iam_put, - .next = osd_it_iam_next, - .key = osd_it_iam_key, - .key_size = osd_it_iam_key_size, - .rec = osd_it_iam_rec, - .store = osd_it_iam_store, - .load = osd_it_iam_load - } + .dio_lookup = osd_index_iam_lookup, + .dio_declare_insert = osd_index_declare_iam_insert, + .dio_insert = osd_index_iam_insert, + .dio_declare_delete = osd_index_declare_iam_delete, + .dio_delete = osd_index_iam_delete, + .dio_it = { + .init = osd_it_iam_init, + .fini = osd_it_iam_fini, + .get = osd_it_iam_get, + .put = osd_it_iam_put, + .next = osd_it_iam_next, + .key = osd_it_iam_key, + .key_size = osd_it_iam_key_size, + .rec = osd_it_iam_rec, + .store = osd_it_iam_store, + .load = osd_it_iam_load + } }; @@ -6042,12 +6575,13 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, 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; + 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; if (!dt_object_exists(dt) || obj->oo_destroyed) @@ -6073,7 +6607,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, if (oie->oie_buf == NULL) RETURN(ERR_PTR(-ENOMEM)); } - oie->oie_obj = obj; + oie->oie_obj = obj; file = &oie->oie_file; @@ -6085,10 +6619,10 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, file->f_path.dentry = obj_dentry; file->f_mapping = obj->oo_inode->i_mapping; file->f_op = obj->oo_inode->i_fop; - set_file_inode(file, obj->oo_inode); + file->f_inode = obj->oo_inode; lu_object_get(lo); - RETURN((struct dt_it *) oie); + RETURN((struct dt_it *)oie); } /** @@ -6098,10 +6632,10 @@ 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_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; + 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; oie->oie_file.f_op->release(inode, &oie->oie_file); @@ -6125,18 +6659,18 @@ static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di) * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty(). */ static int osd_it_ea_get(const struct lu_env *env, - struct dt_it *di, const struct dt_key *key) + struct dt_it *di, const struct dt_key *key) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - ENTRY; - LASSERT(((const char *)key)[0] == '\0'); - it->oie_file.f_pos = 0; - it->oie_rd_dirent = 0; - it->oie_it_dirent = 0; - it->oie_dirent = NULL; + ENTRY; + LASSERT(((const char *)key)[0] == '\0'); + it->oie_file.f_pos = 0; + it->oie_rd_dirent = 0; + it->oie_it_dirent = 0; + it->oie_dirent = NULL; - RETURN(+1); + RETURN(+1); } /** @@ -6147,13 +6681,11 @@ 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 + * It is called internally by ->iterate*(). It fills the * iterator's in-memory data structure with required * information i.e. name, namelen, rec_size etc. * @@ -6169,31 +6701,31 @@ static int osd_ldiskfs_filldir(struct dir_context *buf, static int osd_ldiskfs_filldir(void *buf, #endif const char *name, int namelen, - loff_t offset, __u64 ino, unsigned d_type) -{ - 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; - struct osd_fid_pack *rec; - ENTRY; - - /* this should never happen */ - if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) { - CERROR("ldiskfs return invalid namelen %d\n", namelen); - RETURN(-EIO); - } - - if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen > - OSD_IT_EA_BUFSIZE) - RETURN(1); + loff_t offset, __u64 ino, unsigned int d_type) +{ + 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; + struct osd_fid_pack *rec; + + ENTRY; + +/* this should never happen */ + if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) { + CERROR("ldiskfs return invalid namelen %d\n", namelen); + RETURN(-EIO); + } + + if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen > + OSD_IT_EA_BUFSIZE) + RETURN(1); /* "." is just the object itself. */ if (namelen == 1 && name[0] == '.') { *fid = obj->oo_dt.do_lu.lo_header->loh_fid; } else if (d_type & LDISKFS_DIRENT_LUFID) { - rec = (struct osd_fid_pack*) (name + namelen + 1); + rec = (struct osd_fid_pack *)(name + namelen + 1); if (osd_fid_unpack(fid, rec) != 0) fid_zero(fid); } else { @@ -6207,20 +6739,20 @@ static int osd_ldiskfs_filldir(void *buf, *fid = obj->oo_dt.do_lu.lo_header->loh_fid; } - ent->oied_ino = ino; - ent->oied_off = offset; - ent->oied_namelen = namelen; - ent->oied_type = d_type; + ent->oied_ino = ino; + ent->oied_off = offset; + ent->oied_namelen = namelen; + ent->oied_type = d_type; - memcpy(ent->oied_name, name, namelen); + memcpy(ent->oied_name, name, namelen); - it->oie_rd_dirent++; - it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen); - RETURN(0); + it->oie_rd_dirent++; + it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen); + RETURN(0); } /** - * Calls ->readdir() to load a directory entry at a time + * Calls ->iterate*() to load a directory entry at a time * and stored it in iterator's in-memory data structure. * * \param di iterator's in memory structure @@ -6230,49 +6762,47 @@ static int osd_ldiskfs_filldir(void *buf, * \retval +1 reach the end of entry */ static int osd_ldiskfs_it_fill(const struct lu_env *env, - const 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 htree_lock *hlock = NULL; - struct file *filp = &it->oie_file; - int rc = 0; + const 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 htree_lock *hlock = NULL; + 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; - it->oie_rd_dirent = 0; + ENTRY; + it->oie_dirent = it->oie_buf; + it->oie_rd_dirent = 0; - if (obj->oo_hl_head != NULL) { - hlock = osd_oti_get(env)->oti_hlock; - ldiskfs_htree_lock(hlock, obj->oo_hl_head, - inode, LDISKFS_HLOCK_READDIR); - } else { + if (obj->oo_hl_head != NULL) { + hlock = osd_oti_get(env)->oti_hlock; + ldiskfs_htree_lock(hlock, obj->oo_hl_head, + inode, LDISKFS_HLOCK_READDIR); + } else { down_read(&obj->oo_ext_idx_sem); - } + } -#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 + filp->f_cred = current_cred(); + rc = osd_security_file_alloc(filp); + if (rc) + GOTO(unlock, rc); - if (hlock != NULL) - ldiskfs_htree_unlock(hlock); - else - up_read(&obj->oo_ext_idx_sem); + filp->f_flags |= O_NOATIME; + filp->f_mode |= FMODE_NONOTIFY; + rc = iterate_dir(filp, &buf.ctx); + if (rc) + GOTO(unlock, rc); if (it->oie_rd_dirent == 0) { - /*If it does not get any dirent, it means it has been reached - *to the end of the dir */ + /* + * 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; @@ -6280,12 +6810,17 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, it->oie_dirent = it->oie_buf; it->oie_it_dirent = 1; } +unlock: + if (hlock != NULL) + ldiskfs_htree_unlock(hlock); + else + up_read(&obj->oo_ext_idx_sem); RETURN(rc); } /** - * It calls osd_ldiskfs_it_fill() which will use ->readdir() + * It calls osd_ldiskfs_it_fill() which will use ->iterate*() * to load a directory entry at a time and stored it in * iterator's in-memory data structure. * @@ -6297,26 +6832,26 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env, */ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - int rc; + struct osd_it_ea *it = (struct osd_it_ea *)di; + int rc; - ENTRY; + ENTRY; - if (it->oie_it_dirent < it->oie_rd_dirent) { - it->oie_dirent = - (void *) it->oie_dirent + - cfs_size_round(sizeof(struct osd_it_ea_dirent) + - it->oie_dirent->oied_namelen); - it->oie_it_dirent++; - RETURN(0); - } else { + if (it->oie_it_dirent < it->oie_rd_dirent) { + it->oie_dirent = + (void *)it->oie_dirent + + cfs_size_round(sizeof(struct osd_it_ea_dirent) + + it->oie_dirent->oied_namelen); + it->oie_it_dirent++; + rc = 0; + } else { if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file)) - rc = +1; - else - rc = osd_ldiskfs_it_fill(env, di); - } + rc = 1; + else + rc = osd_ldiskfs_it_fill(env, di); + } - RETURN(rc); + RETURN(rc); } /** @@ -6327,15 +6862,15 @@ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) * \retval key i.e. struct dt_key on success */ static struct dt_key *osd_it_ea_key(const struct lu_env *env, - const struct dt_it *di) + const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return (struct dt_key *)it->oie_dirent->oied_name; + return (struct dt_key *)it->oie_dirent->oied_name; } /** - * Returns the key's size at current position from iterator's in memory structure. + * Returns key's size at current position from iterator's in memory structure. * * \param di iterator's in memory structure * @@ -6343,11 +6878,16 @@ static struct dt_key *osd_it_ea_key(const struct lu_env *env, */ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return it->oie_dirent->oied_namelen; + return it->oie_dirent->oied_namelen; } +#if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_ +#undef LDISKFS_DIR_REC_LEN +#define LDISKFS_DIR_REC_LEN(de) LDISKFS_DIR_ENTRY_LEN_((de)) +#endif + static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de) { if (LDISKFS_DIR_REC_LEN(de) >= @@ -6359,7 +6899,7 @@ static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de) static inline bool osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen, - unsigned blocksize, bool dotdot) + unsigned int blocksize, bool dotdot) { if (dotdot) return osd_dotdot_has_space(de); @@ -6378,17 +6918,17 @@ osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev, struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock, bool dotdot) { - struct inode *dir = dentry->d_parent->d_inode; - struct inode *inode = dentry->d_inode; - struct osd_fid_pack *rec; + struct inode *dir = dentry->d_parent->d_inode; + struct inode *inode = dentry->d_inode; + struct osd_fid_pack *rec; struct ldiskfs_dentry_param *ldp; - int namelen = dentry->d_name.len; - int rc; - struct osd_thread_info *info = osd_oti_get(env); + int namelen = dentry->d_name.len; + int rc; + struct osd_thread_info *info = osd_oti_get(env); + ENTRY; - if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb, - LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + if (!ldiskfs_has_feature_dirdata(inode->i_sb)) RETURN(0); /* There is enough space to hold the FID-in-dirent. */ @@ -6416,16 +6956,17 @@ osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev, ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp; osd_get_ldiskfs_dirent_param(ldp, fid); dentry->d_fsdata = (void *)ldp; - ll_vfs_dq_init(dir); + dquot_initialize(dir); rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock); - /* It is too bad, we cannot reinsert the name entry back. - * That means we lose it! */ + /* + * It is too bad, we cannot reinsert the name entry back. + * That means we lose it! + */ if (rc != 0) - CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, " - "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", - osd_ino2name(inode), - dir->i_ino, dir->i_generation, namelen, - dentry->d_name.name, PFID(fid), rc); + CDEBUG(D_LFSCK, + "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n", + osd_ino2name(inode), dir->i_ino, dir->i_generation, + namelen, dentry->d_name.name, PFID(fid), rc); RETURN(rc); } @@ -6435,24 +6976,25 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, struct osd_it_ea *it, struct lu_fid *fid, struct osd_inode_id *id, __u32 *attr) { - struct osd_thread_info *info = osd_oti_get(env); - struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; - struct osd_device *dev = osd_obj2dev(obj); - struct super_block *sb = osd_sb(dev); - const char *devname = osd_name(dev); - struct osd_it_ea_dirent *ent = it->oie_dirent; - struct inode *dir = obj->oo_inode; - struct htree_lock *hlock = NULL; - struct buffer_head *bh = NULL; - handle_t *jh = NULL; + struct osd_thread_info *info = osd_oti_get(env); + struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma; + struct osd_device *dev = osd_obj2dev(obj); + struct super_block *sb = osd_sb(dev); + const char *devname = osd_name(dev); + struct osd_it_ea_dirent *ent = it->oie_dirent; + struct inode *dir = obj->oo_inode; + struct htree_lock *hlock = NULL; + struct buffer_head *bh = NULL; + handle_t *jh = NULL; struct ldiskfs_dir_entry_2 *de; - struct dentry *dentry; - struct inode *inode; + struct dentry *dentry; + struct inode *inode; const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu); - int credits; - int rc; - bool dotdot = false; - bool dirty = false; + int credits; + int rc; + bool dotdot = false; + bool dirty = false; + ENTRY; if (ent->oied_name[0] == '.') { @@ -6468,8 +7010,10 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, if (IS_ERR(inode)) { rc = PTR_ERR(inode); if (rc == -ENOENT || rc == -ESTALE) { - /* Maybe dangling name entry, or - * corrupted directory entry. */ + /* + * Maybe dangling name entry, or + * corrupted directory entry. + */ *attr |= LUDA_UNKNOWN; rc = 0; } else { @@ -6492,7 +7036,8 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, else if (rc != 0) GOTO(out, rc); - /* We need to ensure that the name entry is still valid. + /* + * We need to ensure that the name entry is still valid. * Because it may be removed or renamed by other already. * * The unlink or rename operation will start journal before PDO lock, @@ -6504,7 +7049,8 @@ osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj, * We may need to remove the name entry firstly, then insert back. * One credit is for user quota file update. * One credit is for group quota file update. - * Two credits are for dirty inode. */ + * Two credits are for dirty inode. + */ credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] + osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2; @@ -6526,10 +7072,12 @@ again: if (obj->oo_hl_head != NULL) { hlock = osd_oti_get(env)->oti_hlock; - /* "0" means exclusive lock for the whole directory. + /* + * "0" means exclusive lock for the whole directory. * We need to prevent others access such name entry * during the delete + insert. Neither HLOCK_ADD nor - * HLOCK_DEL cannot guarantee the atomicity. */ + * HLOCK_DEL cannot guarantee the atomicity. + */ ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0); } else { down_write(&obj->oo_ext_idx_sem); @@ -6551,12 +7099,14 @@ again: GOTO(out, rc = 0); } - /* For dotdot entry, if there is not enough space to hold the + /* + * For dotdot entry, if there is not enough space to hold the * FID-in-dirent, just keep them there. It only happens when the * device upgraded from 1.8 or restored from MDT file-level backup. * For the whole directory, only dotdot entry have no FID-in-dirent * and needs to get FID from LMA when readdir, it will not affect the - * performance much. */ + * performance much. + */ if (dotdot && !osd_dotdot_has_space(de)) { *attr |= LUDA_UNKNOWN; @@ -6573,8 +7123,10 @@ again: if (likely(dotdot && fid_seq(tfid) == FID_SEQ_LOCAL_FILE && fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) { - /* It must be REMOTE_PARENT_DIR and as the - * 'dotdot' entry of remote directory */ + /* + * It must be REMOTE_PARENT_DIR and as the + * 'dotdot' entry of remote directory + */ *attr |= LUDA_IGNORE; } else { CDEBUG(D_LFSCK, "%s: expect remote agent " @@ -6597,9 +7149,11 @@ again: if (rc == -ENOENT || (rc == -ENODATA && !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) { - /* linkEA does not recognize the dirent entry, + /* + * linkEA does not recognize the dirent entry, * it may because the dirent entry corruption - * and points to other's inode. */ + * and points to other's inode. + */ CDEBUG(D_LFSCK, "%s: the target inode does not " "recognize the dirent, dir = %lu/%u, " " name = %.*s, ino = %llu, " @@ -6624,8 +7178,10 @@ again: } if (lma != NULL) { - /* linkEA recognizes the dirent entry, the FID-in-LMA is - * valid, trusted, in spite of fid_is_sane(fid) or not. */ + /* + * linkEA recognizes the dirent entry, the FID-in-LMA is + * valid, trusted, in spite of fid_is_sane(fid) or not. + */ if (*attr & LUDA_VERIFY_DRYRUN) { *fid = lma->lma_self_fid; *attr |= LUDA_REPAIR; @@ -6689,8 +7245,10 @@ again: dirty = true; if (unlikely(fid_is_sane(fid))) { - /* FID-in-dirent exists, but FID-in-LMA is lost. - * Trust the FID-in-dirent, and add FID-in-LMA. */ + /* + * FID-in-dirent exists, but FID-in-LMA is lost. + * Trust the FID-in-dirent, and add FID-in-LMA. + */ rc = osd_ea_fid_set(info, inode, fid, 0, 0); if (rc == 0) *attr |= LUDA_REPAIR; @@ -6704,8 +7262,10 @@ again: ent->oied_ino, PFID(fid), rc); } else if (dev->od_index == 0) { lu_igif_build(fid, inode->i_ino, inode->i_generation); - /* It is probably IGIF object. Only aappend the - * FID-in-dirent. OI scrub will process FID-in-LMA. */ + /* + * It is probably IGIF object. Only aappend the + * FID-in-dirent. OI scrub will process FID-in-LMA. + */ rc = osd_dirent_reinsert(env, dev, jh, dentry, fid, bh, de, hlock, dotdot); if (rc == 0) @@ -6765,20 +7325,23 @@ static inline int osd_it_ea_rec(const struct lu_env *env, struct osd_device *dev = osd_obj2dev(obj); struct osd_thread_info *oti = osd_oti_get(env); struct osd_inode_id *id = &oti->oti_id; - struct lu_fid *fid = &it->oie_dirent->oied_fid; + 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; + __u32 ino = it->oie_dirent->oied_ino; + int rc = 0; + ENTRY; - LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode); + LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino)); if (attr & LUDA_VERIFY) { - if (unlikely(ino == osd_remote_parent_ino(dev))) { + if (unlikely(is_remote_parent_ino(dev, ino))) { attr |= LUDA_IGNORE; - /* If the parent is on remote MDT, and there + /* + * 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. */ + * the parent FID from the linkEA. + */ if (!fid_is_sane(fid) && it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && @@ -6795,14 +7358,17 @@ static inline int osd_it_ea_rec(const struct lu_env *env, attr &= ~LU_DIRENT_ATTRS_MASK; if (!fid_is_sane(fid)) { bool is_dotdot = false; + if (it->oie_dirent->oied_namelen == 2 && it->oie_dirent->oied_name[0] == '.' && it->oie_dirent->oied_name[1] == '.') is_dotdot = true; - /* If the parent is on remote MDT, and there + /* + * 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) && is_dotdot) { + * the parent FID from the linkEA. + */ + if (is_remote_parent_ino(dev, ino) && is_dotdot) { rc = osd_get_pfid_from_linkea(env, obj, fid); } else { if (is_dotdot == false && @@ -6811,8 +7377,6 @@ static inline int osd_it_ea_rec(const struct lu_env *env, rc = osd_ea_fid_get(env, obj, ino, fid, id); } - } else { - osd_id_gen(id, ino, OSD_OII_NOGEN); } } @@ -6822,15 +7386,6 @@ static inline int osd_it_ea_rec(const struct lu_env *env, it->oie_dirent->oied_namelen, it->oie_dirent->oied_type, attr); - if (rc < 0) - RETURN(rc); - - if (osd_remote_fid(env, dev, fid)) - RETURN(0); - - if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0)) - osd_add_oi_cache(oti, dev, id, fid); - RETURN(rc > 0 ? 0 : rc); } @@ -6865,13 +7420,13 @@ static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di, */ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di) { - struct osd_it_ea *it = (struct osd_it_ea *)di; + struct osd_it_ea *it = (struct osd_it_ea *)di; - return it->oie_dirent->oied_off; + return it->oie_dirent->oied_off; } /** - * It calls osd_ldiskfs_it_fill() which will use ->readdir() + * It calls osd_ldiskfs_it_fill() which will use ->iterate*() * to load a directory entry at a time and stored it i inn, * in iterator's in-memory data structure. * @@ -6881,22 +7436,22 @@ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di) * \retval -ve on error */ static int osd_it_ea_load(const struct lu_env *env, - const struct dt_it *di, __u64 hash) + const struct dt_it *di, __u64 hash) { - struct osd_it_ea *it = (struct osd_it_ea *)di; - int rc; + struct osd_it_ea *it = (struct osd_it_ea *)di; + int rc; - ENTRY; - it->oie_file.f_pos = hash; + ENTRY; + it->oie_file.f_pos = hash; - rc = osd_ldiskfs_it_fill(env, di); + rc = osd_ldiskfs_it_fill(env, di); if (rc > 0) rc = -ENODATA; - if (rc == 0) - rc = +1; + if (rc == 0) + rc = 1; - RETURN(rc); + RETURN(rc); } /** @@ -6910,18 +7465,18 @@ static int osd_it_ea_load(const struct lu_env *env, 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 osd_object *obj = osd_dt_obj(dt); - int rc = 0; + struct osd_object *obj = osd_dt_obj(dt); + int rc = 0; - ENTRY; + ENTRY; - LASSERT(S_ISDIR(obj->oo_inode->i_mode)); - LINVRNT(osd_invariant(obj)); + LASSERT(S_ISDIR(obj->oo_inode->i_mode)); + LINVRNT(osd_invariant(obj)); - rc = osd_ea_lookup_rec(env, obj, rec, key); - if (rc == 0) - rc = +1; - RETURN(rc); + rc = osd_ea_lookup_rec(env, obj, rec, key); + if (rc == 0) + rc = 1; + RETURN(rc); } /** @@ -6950,39 +7505,55 @@ static const struct dt_index_operations osd_index_ea_ops = { }; static void *osd_key_init(const struct lu_context *ctx, - struct lu_context_key *key) + struct lu_context_key *key) { - struct osd_thread_info *info; + struct osd_thread_info *info; - OBD_ALLOC_PTR(info); - if (info == NULL) - return ERR_PTR(-ENOMEM); + OBD_ALLOC_PTR(info); + if (info == NULL) + return ERR_PTR(-ENOMEM); - OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - if (info->oti_it_ea_buf == NULL) - goto out_free_info; + OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); + if (info->oti_it_ea_buf == NULL) + goto out_free_info; - info->oti_env = container_of(ctx, struct lu_env, le_ctx); + info->oti_env = container_of(ctx, struct lu_env, le_ctx); - info->oti_hlock = ldiskfs_htree_lock_alloc(); - if (info->oti_hlock == NULL) - goto out_free_ea; + info->oti_hlock = ldiskfs_htree_lock_alloc(); + if (info->oti_hlock == NULL) + goto out_free_ea; - return info; + return info; - out_free_ea: - OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); - out_free_info: - OBD_FREE_PTR(info); - return ERR_PTR(-ENOMEM); +out_free_ea: + OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); +out_free_info: + OBD_FREE_PTR(info); + return ERR_PTR(-ENOMEM); } static void osd_key_fini(const struct lu_context *ctx, - struct lu_context_key *key, void* data) + struct lu_context_key *key, void *data) { struct osd_thread_info *info = data; struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode); - struct osd_idmap_cache *idc = info->oti_ins_cache; + struct osd_idmap_cache *idc = info->oti_ins_cache; + + if (info->oti_dio_pages) { + int i; + for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) { + struct page *page = info->oti_dio_pages[i]; + if (page) { + LASSERT(PagePrivate2(page)); + LASSERT(PageLocked(page)); + ClearPagePrivate2(page); + unlock_page(page); + __free_page(page); + } + } + OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages, + PTLRPC_MAX_BRW_PAGES); + } if (info->oti_inode != NULL) OBD_FREE_PTR(lli); @@ -6991,10 +7562,11 @@ static void osd_key_fini(const struct lu_context *ctx, OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE); lu_buf_free(&info->oti_iobuf.dr_pg_buf); lu_buf_free(&info->oti_iobuf.dr_bl_buf); + lu_buf_free(&info->oti_iobuf.dr_lnb_buf); lu_buf_free(&info->oti_big_buf); if (idc != NULL) { LASSERT(info->oti_ins_cache_size > 0); - OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size); + OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size); info->oti_ins_cache = NULL; info->oti_ins_cache_size = 0; } @@ -7002,41 +7574,42 @@ static void osd_key_fini(const struct lu_context *ctx, } static void osd_key_exit(const struct lu_context *ctx, - struct lu_context_key *key, void *data) + struct lu_context_key *key, void *data) { - struct osd_thread_info *info = data; + struct osd_thread_info *info = data; - LASSERT(info->oti_r_locks == 0); - LASSERT(info->oti_w_locks == 0); - LASSERT(info->oti_txns == 0); + LASSERT(info->oti_r_locks == 0); + LASSERT(info->oti_w_locks == 0); + LASSERT(info->oti_txns == 0); } /* type constructor/destructor: osd_type_init, osd_type_fini */ LU_TYPE_INIT_FINI(osd, &osd_key); struct lu_context_key osd_key = { - .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL, - .lct_init = osd_key_init, - .lct_fini = osd_key_fini, - .lct_exit = osd_key_exit + .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL, + .lct_init = osd_key_init, + .lct_fini = osd_key_fini, + .lct_exit = osd_key_exit }; static int osd_device_init(const struct lu_env *env, struct lu_device *d, - const char *name, struct lu_device *next) + const char *name, struct lu_device *next) { struct osd_device *osd = osd_dev(d); - if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) - >= sizeof(osd->od_svname)) + if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >= + sizeof(osd->od_svname)) return -E2BIG; 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; + struct seq_server_site *ss = osd_seq_site(osd); + int rc = 0; + ENTRY; if (osd->od_is_ost || osd->od_cl_seq != NULL) @@ -7049,18 +7622,15 @@ static int osd_fid_init(const struct lu_env *env, struct osd_device *osd) 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); - } + seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA, + osd->od_svname, ss->ss_server_seq); if (ss->ss_node_id == 0) { - /* If the OSD on the sequence controller(MDT0), then allocate + /* + * If the OSD on the sequence controller(MDT0), then allocate * sequence here, otherwise allocate sequence after connected - * to MDT0 (see mdt_register_lwp_callback()). */ + * to MDT0 (see mdt_register_lwp_callback()). + */ rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv, &osd->od_cl_seq->lcs_space, env); } @@ -7083,10 +7653,17 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o) ENTRY; /* shutdown quota slave instance associated with the device */ - if (o->od_quota_slave != NULL) { - struct qsd_instance *qsd = o->od_quota_slave; + if (o->od_quota_slave_md != NULL) { + struct qsd_instance *qsd = o->od_quota_slave_md; - o->od_quota_slave = NULL; + o->od_quota_slave_md = NULL; + qsd_fini(env, qsd); + } + + if (o->od_quota_slave_dt != NULL) { + struct qsd_instance *qsd = o->od_quota_slave_dt; + + o->od_quota_slave_dt = NULL; qsd_fini(env, qsd); } @@ -7103,6 +7680,8 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o) if (o->od_mnt != NULL) { shrink_dcache_sb(osd_sb(o)); osd_sync(env, &o->od_dt_dev); + wait_event(o->od_commit_cb_done, + !atomic_read(&o->od_commit_cb_in_flight)); mntput(o->od_mnt); o->od_mnt = NULL; @@ -7112,21 +7691,22 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o) } static int osd_mount(const struct lu_env *env, - struct osd_device *o, struct lustre_cfg *cfg) + struct osd_device *o, struct lustre_cfg *cfg) { - const char *name = lustre_cfg_string(cfg, 0); - const char *dev = lustre_cfg_string(cfg, 1); - const char *opts; - unsigned long page, s_flags, lmd_flags = 0; - struct page *__page; + const char *name = lustre_cfg_string(cfg, 0); + const char *dev = lustre_cfg_string(cfg, 1); + const char *opts; + unsigned long page, s_flags, lmd_flags = 0; + struct page *__page; struct file_system_type *type; - char *options = NULL; - char *str; - struct osd_thread_info *info = osd_oti_get(env); - struct lu_fid *fid = &info->oti_fid; - struct inode *inode; - int rc = 0, force_over_512tb = 0; - ENTRY; + char *options = NULL; + char *str; + struct osd_thread_info *info = osd_oti_get(env); + struct lu_fid *fid = &info->oti_fid; + struct inode *inode; + int rc = 0, force_over_1024tb = 0; + + ENTRY; if (o->od_mnt != NULL) RETURN(0); @@ -7151,23 +7731,22 @@ static int osd_mount(const struct lu_env *env, #endif #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0) if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) { - CWARN("force_over_128tb option is deprecated. " - "Filesystems less than 512TB can be created without any " - "force options. Use force_over_512tb option for " - "filesystems greater than 512TB.\n"); + CWARN("force_over_128tb option is deprecated. Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n"); } #endif #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0) if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) { - CWARN("force_over_256tb option is deprecated. " - "Filesystems less than 512TB can be created without any " - "force options. Use force_over_512tb option for " - "filesystems greater than 512TB.\n"); + CWARN("force_over_256tb option is deprecated. Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n"); + } +#endif +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0) + if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) { + CWARN("force_over_512tb option is deprecated. Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n"); } #endif - if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) - force_over_512tb = 1; + if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL) + force_over_1024tb = 1; __page = alloc_page(GFP_KERNEL); if (__page == NULL) @@ -7177,7 +7756,7 @@ static int osd_mount(const struct lu_env *env, *options = '\0'; if (opts != NULL) { /* strip out the options for back compatiblity */ - static char *sout[] = { + static const char * const sout[] = { "mballoc", "iopen", "noiopen", @@ -7189,11 +7768,13 @@ static int osd_mount(const struct lu_env *env, "force_over_128tb", "force_over_256tb", "force_over_512tb", + "force_over_1024tb", NULL }; - strcat(options, opts); + strncat(options, opts, PAGE_SIZE); for (rc = 0, str = options; sout[rc]; ) { char *op = strstr(str, sout[rc]); + if (op == NULL) { rc++; str = options; @@ -7210,13 +7791,13 @@ static int osd_mount(const struct lu_env *env, ; } } else { - strncat(options, "user_xattr,acl", 14); + strncat(options, "user_xattr,acl", PAGE_SIZE); } /* Glom up mount options */ if (*options != '\0') - strcat(options, ","); - strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE); + strncat(options, ",", PAGE_SIZE); + strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE); type = get_fs_type("ldiskfs"); if (!type) { @@ -7235,44 +7816,27 @@ static int osd_mount(const struct lu_env *env, } if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) << - osd_sb(o)->s_blocksize_bits > 512ULL << 40 && - force_over_512tb == 0) { - CERROR("%s: device %s LDISKFS does not support filesystems " - "greater than 512TB and can cause data corruption. " - "Use \"force_over_512tb\" mount option to override.\n", + osd_sb(o)->s_blocksize_bits > 1024ULL << 40 && + force_over_1024tb == 0) { + CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n", name, dev); GOTO(out_mnt, rc = -EINVAL); } if (lmd_flags & LMD_FLG_DEV_RDONLY) { - if (priv_dev_set_rdonly) { - priv_dev_set_rdonly(osd_sb(o)->s_bdev); - o->od_dt_dev.dd_rdonly = 1; - LCONSOLE_WARN("%s: set dev_rdonly on this device\n", - name); - } else { - LCONSOLE_WARN("%s: not support dev_rdonly on this device", - name); + LCONSOLE_WARN("%s: not support dev_rdonly on this device", + name); - GOTO(out_mnt, rc = -EOPNOTSUPP); - } - } else if (priv_dev_check_rdonly && - priv_dev_check_rdonly(osd_sb(o)->s_bdev)) { - CERROR("%s: underlying device %s is marked as " - "read-only. Setup failed\n", name, dev); - - GOTO(out_mnt, rc = -EROFS); + GOTO(out_mnt, rc = -EOPNOTSUPP); } - if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { + if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) { CERROR("%s: device %s is mounted w/o journal\n", name, dev); GOTO(out_mnt, rc = -EINVAL); } #ifdef LDISKFS_MOUNT_DIRDATA - if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb, - LDISKFS_FEATURE_INCOMPAT_DIRDATA)) + if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb)) LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA; else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */ CWARN("%s: device %s was upgraded from Lustre-1.x without " @@ -7291,6 +7855,12 @@ static int osd_mount(const struct lu_env *env, if (lmd_flags & LMD_FLG_NOSCRUB) o->od_auto_scrub_interval = AS_NEVER; + if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) { + /* do not use pagecache with flash-backed storage */ + o->od_writethrough_cache = 0; + o->od_read_cache = 0; + } + GOTO(out, rc = 0); out_mnt: @@ -7308,11 +7878,16 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, struct lu_device *d) { struct osd_device *o = osd_dev(d); + ENTRY; osd_index_backup(env, o, false); osd_shutdown(env, o); osd_procfs_fini(o); + if (o->od_oi_table != NULL) + osd_oi_fini(osd_oti_get(env), o); + if (o->od_extent_bytes_percpu) + free_percpu(o->od_extent_bytes_percpu); osd_obj_map_fini(o); osd_umount(env, o); @@ -7323,13 +7898,13 @@ static int osd_device_init0(const struct lu_env *env, struct osd_device *o, struct lustre_cfg *cfg) { - struct lu_device *l = osd2lu_dev(o); + struct lu_device *l = osd2lu_dev(o); struct osd_thread_info *info; - int rc; - int cplen = 0; + int rc; + int cplen = 0; /* if the module was re-loaded, env can loose its keys */ - rc = lu_env_refill((struct lu_env *) env); + rc = lu_env_refill((struct lu_env *)env); if (rc) GOTO(out, rc); info = osd_oti_get(env); @@ -7345,11 +7920,17 @@ static int osd_device_init0(const struct lu_env *env, INIT_LIST_HEAD(&o->od_index_restore_list); spin_lock_init(&o->od_lock); o->od_index_backup_policy = LIBP_NONE; + o->od_t10_type = 0; + init_waitqueue_head(&o->od_commit_cb_done); o->od_read_cache = 1; o->od_writethrough_cache = 1; o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; + o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20; + o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20; o->od_auto_scrub_interval = AS_DEFAULT; + /* default fallocate to unwritten extents: LU-14326/LU-14333 */ + o->od_fallocate_zero_blocks = 0; cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4), sizeof(o->od_svname)); @@ -7370,6 +7951,10 @@ static int osd_device_init0(const struct lu_env *env, if (rc != 0) GOTO(out, rc); + /* Can only check block device after mount */ + o->od_nonrotational = + blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev)); + rc = osd_obj_map_init(env, o); if (rc != 0) GOTO(out_mnt, rc); @@ -7402,11 +7987,35 @@ static int osd_device_init0(const struct lu_env *env, 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, - o->od_proc_entry); - if (IS_ERR(o->od_quota_slave)) { - rc = PTR_ERR(o->od_quota_slave); - o->od_quota_slave = NULL; + /* currently it's no need to prepare qsd_instance_md for OST */ + if (!o->od_is_ost) { + o->od_quota_slave_md = qsd_init(env, o->od_svname, + &o->od_dt_dev, + o->od_proc_entry, true); + if (IS_ERR(o->od_quota_slave_md)) { + rc = PTR_ERR(o->od_quota_slave_md); + o->od_quota_slave_md = NULL; + GOTO(out_procfs, rc); + } + } + + o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev, + o->od_proc_entry, false); + + if (IS_ERR(o->od_quota_slave_dt)) { + if (o->od_quota_slave_md != NULL) { + qsd_fini(env, o->od_quota_slave_md); + o->od_quota_slave_md = NULL; + } + + rc = PTR_ERR(o->od_quota_slave_dt); + o->od_quota_slave_dt = NULL; + GOTO(out_procfs, rc); + } + + o->od_extent_bytes_percpu = alloc_percpu(unsigned int); + if (!o->od_extent_bytes_percpu) { + rc = -ENOMEM; GOTO(out_procfs, rc); } @@ -7427,11 +8036,11 @@ out: } static struct lu_device *osd_device_alloc(const struct lu_env *env, - struct lu_device_type *t, - struct lustre_cfg *cfg) + struct lu_device_type *t, + struct lustre_cfg *cfg) { struct osd_device *o; - int rc; + int rc; OBD_ALLOC_PTR(o); if (o == NULL) @@ -7439,8 +8048,10 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, rc = dt_device_init(&o->od_dt_dev, t); if (rc == 0) { - /* Because the ctx might be revived in dt_device_init, - * refill the env here */ + /* + * Because the ctx might be revived in dt_device_init, + * refill the env here + */ lu_env_refill((struct lu_env *)env); rc = osd_device_init0(env, o, cfg); if (rc) @@ -7454,29 +8065,30 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env, } static struct lu_device *osd_device_free(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - struct osd_device *o = osd_dev(d); - ENTRY; + struct osd_device *o = osd_dev(d); + + ENTRY; /* 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); - if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) { - LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL); - lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer); - } + lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems, + D_ERROR, lu_cdebug_printer); lu_site_fini(&o->od_site); - dt_device_fini(&o->od_dt_dev); - OBD_FREE_PTR(o); - RETURN(NULL); + dt_device_fini(&o->od_dt_dev); + OBD_FREE_PTR(o); + RETURN(NULL); } static int osd_process_config(const struct lu_env *env, - struct lu_device *d, struct lustre_cfg *cfg) + struct lu_device *d, struct lustre_cfg *cfg) { - struct osd_device *o = osd_dev(d); - int rc; + struct osd_device *o = osd_dev(d); + ssize_t count; + int rc; + ENTRY; switch (cfg->lcfg_command) { @@ -7484,23 +8096,22 @@ static int osd_process_config(const struct lu_env *env, rc = osd_mount(env, o, cfg); break; case LCFG_CLEANUP: - /* For the case LCFG_PRE_CLEANUP is not called in advance, - * that may happend if hit failure during mount process. */ + /* + * For the case LCFG_PRE_CLEANUP is not called in advance, + * that may happend if hit failure during mount process. + */ osd_index_backup(env, o, false); lu_dev_del_linkage(d->ld_site, d); rc = osd_shutdown(env, o); break; case LCFG_PARAM: LASSERT(&o->od_dt_dev); - rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars, - cfg, &o->od_dt_dev); - if (rc > 0 || rc == -ENOSYS) { - rc = class_process_proc_param(PARAM_OST, - lprocfs_osd_obd_vars, - cfg, &o->od_dt_dev); - if (rc > 0) - rc = 0; - } + count = class_modify_config(cfg, PARAM_OSD, + &o->od_dt_dev.dd_kobj); + if (count < 0) + count = class_modify_config(cfg, PARAM_OST, + &o->od_dt_dev.dd_kobj); + rc = count > 0 ? 0 : count; break; case LCFG_PRE_CLEANUP: osd_scrub_stop(o); @@ -7516,18 +8127,24 @@ static int osd_process_config(const struct lu_env *env, } static int osd_recovery_complete(const struct lu_env *env, - struct lu_device *d) + struct lu_device *d) { - struct osd_device *osd = osd_dev(d); - int rc = 0; + struct osd_device *osd = osd_dev(d); + int rc = 0; + ENTRY; - if (osd->od_quota_slave == NULL) + if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL) RETURN(0); - /* start qsd instance on recovery completion, this notifies the quota - * slave code that we are about to process new requests now */ - rc = qsd_start(env, osd->od_quota_slave); + /* + * start qsd instance on recovery completion, this notifies the quota + * slave code that we are about to process new requests now + */ + rc = qsd_start(env, osd->od_quota_slave_dt); + if (rc == 0 && osd->od_quota_slave_md != NULL) + rc = qsd_start(env, osd->od_quota_slave_md); + RETURN(rc); } @@ -7538,9 +8155,10 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { - struct osd_device *osd = osd_dev(obd->obd_lu_dev); - struct lustre_handle conn; - int rc; + struct osd_device *osd = osd_dev(obd->obd_lu_dev); + struct lustre_handle conn; + int rc; + ENTRY; CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects); @@ -7566,7 +8184,8 @@ static int osd_obd_disconnect(struct obd_export *exp) { struct obd_device *obd = exp->exp_obd; struct osd_device *osd = osd_dev(obd->obd_lu_dev); - int rc, release = 0; + int rc, release = 0; + ENTRY; /* Only disconnect the underlying layers on the final disconnect. */ @@ -7586,19 +8205,28 @@ 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 osd_device *osd = osd_dev(dev); - struct lr_server_data *lsd = + 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; + int result = 0; + ENTRY; - if (osd->od_quota_slave != NULL) { - /* set up quota slave objects */ - result = qsd_prepare(env, osd->od_quota_slave); + if (osd->od_quota_slave_md != NULL) { + /* set up quota slave objects for inode */ + result = qsd_prepare(env, osd->od_quota_slave_md); if (result != 0) RETURN(result); } + if (osd->od_quota_slave_dt != NULL) { + /* set up quota slave objects for block */ + result = qsd_prepare(env, osd->od_quota_slave_dt); + 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) { @@ -7619,49 +8247,58 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, 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) +/** + * Implementation of lu_device_operations::ldo_fid_alloc() for OSD + * + * Allocate FID. + * + * see include/lu_object.h for the details. + */ +static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d, + struct lu_fid *fid, struct lu_object *parent, + const struct lu_name *name) { - struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev); + struct osd_device *osd = osd_dev(d); 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, - .loo_object_release = osd_object_release, - .loo_object_free = osd_object_free, - .loo_object_print = osd_object_print, - .loo_object_invariant = osd_object_invariant + .loo_object_init = osd_object_init, + .loo_object_delete = osd_object_delete, + .loo_object_release = osd_object_release, + .loo_object_free = osd_object_free, + .loo_object_print = osd_object_print, + .loo_object_invariant = osd_object_invariant }; const struct lu_device_operations osd_lu_ops = { - .ldo_object_alloc = osd_object_alloc, - .ldo_process_config = osd_process_config, - .ldo_recovery_complete = osd_recovery_complete, - .ldo_prepare = osd_prepare, + .ldo_object_alloc = osd_object_alloc, + .ldo_process_config = osd_process_config, + .ldo_recovery_complete = osd_recovery_complete, + .ldo_prepare = osd_prepare, + .ldo_fid_alloc = osd_fid_alloc, }; static const struct lu_device_type_operations osd_device_type_ops = { - .ldto_init = osd_type_init, - .ldto_fini = osd_type_fini, + .ldto_init = osd_type_init, + .ldto_fini = osd_type_fini, - .ldto_start = osd_type_start, - .ldto_stop = osd_type_stop, + .ldto_start = osd_type_start, + .ldto_stop = osd_type_stop, - .ldto_device_alloc = osd_device_alloc, - .ldto_device_free = osd_device_free, + .ldto_device_alloc = osd_device_alloc, + .ldto_device_free = osd_device_free, - .ldto_device_init = osd_device_init, - .ldto_device_fini = osd_device_fini + .ldto_device_init = osd_device_init, + .ldto_device_fini = osd_device_fini }; static struct lu_device_type osd_device_type = { - .ldt_tags = LU_DEVICE_DT, + .ldt_tags = LU_DEVICE_DT, .ldt_name = LUSTRE_OSD_LDISKFS_NAME, - .ldt_ops = &osd_device_type_ops, - .ldt_ctx_tags = LCT_LOCAL, + .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) @@ -7669,28 +8306,53 @@ 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); + return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY); } /* * lprocfs legacy support. */ -static struct obd_ops osd_obd_device_ops = { +static const struct obd_ops osd_obd_device_ops = { .o_owner = THIS_MODULE, .o_connect = osd_obd_connect, .o_disconnect = osd_obd_disconnect, - .o_fid_alloc = osd_fid_alloc, .o_health_check = osd_health_check, }; +static ssize_t track_declares_assert_show(struct kobject *kobj, + struct attribute *attr, + char *buf) +{ + return sprintf(buf, "%d\n", ldiskfs_track_declares_assert); +} + +static ssize_t track_declares_assert_store(struct kobject *kobj, + struct attribute *attr, + const char *buffer, size_t count) +{ + bool track_declares_assert; + int rc; + + rc = kstrtobool(buffer, &track_declares_assert); + if (rc) + return rc; + + ldiskfs_track_declares_assert = track_declares_assert; + + return count; +} +LUSTRE_RW_ATTR(track_declares_assert); + static int __init osd_init(void) { + struct kobject *kobj; int rc; - CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8); + BUILD_BUG_ON(BH_DXLock >= + sizeof(((struct buffer_head *)0)->b_state) * 8); #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); + BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE); #endif osd_oi_mod_init(); @@ -7700,22 +8362,40 @@ static int __init osd_init(void) return rc; #ifdef CONFIG_KALLSYMS - priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly"); - priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly"); - /* Clear readonly is unused at this time */ - /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/ + priv_security_file_alloc = + (void *)kallsyms_lookup_name("security_file_alloc"); #endif - rc = class_register_type(&osd_obd_device_ops, NULL, true, - lprocfs_osd_module_vars, + rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL, LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); - if (rc) + if (rc) { lu_kmem_fini(ldiskfs_caches); + return rc; + } + + kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME); + if (kobj) { + rc = sysfs_create_file(kobj, + &lustre_attr_track_declares_assert.attr); + kobject_put(kobj); + if (rc) { + CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n"); + rc = 0; + } + } return rc; } static void __exit osd_exit(void) { + struct kobject *kobj; + + kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME); + if (kobj) { + sysfs_remove_file(kobj, + &lustre_attr_track_declares_assert.attr); + kobject_put(kobj); + } class_unregister_type(LUSTRE_OSD_LDISKFS_NAME); lu_kmem_fini(ldiskfs_caches); }