X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_quota.c;h=b3dc66acd0f084874a5f3b2081a275e48cf56b36;hb=0c1ae1cb9c19f8a4f6c5a7ff6a1fd54807430795;hp=e426b43eb751369c07d7c4368fba68a6a30e9c11;hpb=dd30f316f25ac70dba8891406d2a107dbcfcd75a;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_quota.c b/lustre/osd-ldiskfs/osd_quota.c index e426b43..b3dc66a 100644 --- a/lustre/osd-ldiskfs/osd_quota.c +++ b/lustre/osd-ldiskfs/osd_quota.c @@ -21,7 +21,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2012, 2015, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. * Use is subject to license terms. * * Author: Johann Lombardi @@ -38,14 +38,17 @@ static inline int fid2type(const struct lu_fid *fid) { LASSERT(fid_is_acct(fid)); - if (fid_oid(fid) == ACCT_GROUP_OID) + switch (fid_oid(fid)) { + case ACCT_USER_OID: + return USRQUOTA; + case ACCT_GROUP_OID: return GRPQUOTA; - return USRQUOTA; -} + case ACCT_PROJECT_OID: + return PRJQUOTA; + } -static inline int obj2type(struct dt_object *obj) -{ - return fid2type(lu_object_fid(&obj->do_lu)); + LASSERTF(0, "invalid fid for quota type: %u\n", fid_oid(fid)); + return USRQUOTA; } /** @@ -65,22 +68,41 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd, const struct lu_fid *fid, struct osd_inode_id *id) { struct super_block *sb = osd_sb(osd); - unsigned long qf_inums[2] = { - le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum), - le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum) - }; ENTRY; LASSERT(fid_is_acct(fid)); - if (!LDISKFS_HAS_RO_COMPAT_FEATURE(sb, - LDISKFS_FEATURE_RO_COMPAT_QUOTA)) + if (!ldiskfs_has_feature_quota(sb)) RETURN(-ENOENT); id->oii_gen = OSD_OII_NOGEN; - id->oii_ino = qf_inums[fid2type(fid)]; - if (!ldiskfs_valid_inum(sb, id->oii_ino)) + switch (fid2type(fid)) { + case USRQUOTA: + id->oii_ino = + le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum); + break; + case GRPQUOTA: + id->oii_ino = + le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum); + break; + case PRJQUOTA: +#ifdef HAVE_PROJECT_QUOTA + if (ldiskfs_has_feature_project(sb)) { + __le32 prj_quota; + + prj_quota = LDISKFS_SB(sb)->s_es->s_prj_quota_inum; + id->oii_ino = le32_to_cpu(prj_quota); + } else +#endif + RETURN(-ENOENT); + break; + } + + if (!ldiskfs_valid_inum(sb, id->oii_ino) && + id->oii_ino != LDISKFS_USR_QUOTA_INO && + id->oii_ino != LDISKFS_GRP_QUOTA_INO) RETURN(-ENOENT); + RETURN(0); } @@ -101,42 +123,33 @@ static int osd_acct_index_lookup(const struct lu_env *env, struct dt_rec *dtrec, const struct dt_key *dtkey) { - struct osd_thread_info *info = osd_oti_get(env); + struct osd_thread_info *info = osd_oti_get(env); #if defined(HAVE_DQUOT_QC_DQBLK) - struct qc_dqblk *dqblk = &info->oti_qdq; -#elif defined(HAVE_DQUOT_FS_DISK_QUOTA) - struct fs_disk_quota *dqblk = &info->oti_fdq; + struct qc_dqblk *dqblk = &info->oti_qdq; #else - struct if_dqblk *dqblk = &info->oti_dqblk; -#endif - struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj))); - struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec; - __u64 id = *((__u64 *)dtkey); - int rc; -#ifdef HAVE_DQUOT_KQID - struct kqid qid; + struct fs_disk_quota *dqblk = &info->oti_fdq; #endif + struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj))); + struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec; + __u64 id = *((__u64 *)dtkey); + int rc; + struct kqid qid; + int type; ENTRY; + type = fid2type(lu_object_fid(&dtobj->do_lu)); memset(dqblk, 0, sizeof(*dqblk)); -#ifdef HAVE_DQUOT_KQID - qid = make_kqid(&init_user_ns, obj2type(dtobj), id); + qid = make_kqid(&init_user_ns, type, id); rc = sb->s_qcop->get_dqblk(sb, qid, dqblk); -#else - rc = sb->s_qcop->get_dqblk(sb, obj2type(dtobj), (qid_t)id, dqblk); -#endif if (rc) RETURN(rc); #if defined(HAVE_DQUOT_QC_DQBLK) rec->bspace = dqblk->d_space; rec->ispace = dqblk->d_ino_count; -#elif defined(HAVE_DQUOT_FS_DISK_QUOTA) +#else rec->bspace = dqblk->d_bcount; rec->ispace = dqblk->d_icount; -#else - rec->bspace = dqblk->dqb_curspace; - rec->ispace = dqblk->dqb_curinodes; #endif RETURN(+1); } @@ -144,7 +157,7 @@ static int osd_acct_index_lookup(const struct lu_env *env, #define QUOTA_IT_READ_ERROR(it, rc) \ CERROR("%s: Error while trying to read quota information, " \ "failed with %d\n", \ - osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \ + osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc) \ /** * Initialize osd Iterator for given osd index object. @@ -156,9 +169,9 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env, struct dt_object *dt, __u32 attr) { - struct osd_it_quota *it; - struct lu_object *lo = &dt->do_lu; - struct osd_object *obj = osd_dt_obj(dt); + struct osd_it_quota *it; + struct lu_object *lo = &dt->do_lu; + struct osd_object *obj = osd_dt_obj(dt); ENTRY; @@ -175,9 +188,11 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env, /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */ it->oiq_blk[0] = LUSTRE_DQTREEOFF; - /* NB: we don't need to store the tree depth since it is always + /* + * NB: we don't need to store the tree depth since it is always * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf - * block. */ + * block. + */ RETURN((struct dt_it *)it); } @@ -190,9 +205,10 @@ static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di) { struct osd_it_quota *it = (struct osd_it_quota *)di; struct osd_quota_leaf *leaf, *tmp; + ENTRY; - lu_object_put(env, &it->oiq_obj->oo_dt.do_lu); + osd_object_put(env, it->oiq_obj); list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) { list_del_init(&leaf->oql_link); @@ -218,15 +234,15 @@ static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di) static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di, const struct dt_key *key) { - struct osd_it_quota *it = (struct osd_it_quota *)di; - const struct lu_fid *fid = - lu_object_fid(&it->oiq_obj->oo_dt.do_lu); - int type = fid2type(fid); - qid_t dqid = *(qid_t *)key; - loff_t offset; - int rc; + struct osd_it_quota *it = (struct osd_it_quota *)di; + const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu); + int type; + qid_t dqid = *(qid_t *)key; + loff_t offset; + int rc; ENTRY; + type = fid2type(fid); offset = find_tree_dqentry(env, it->oiq_obj, type, dqid, LUSTRE_DQTREEOFF, 0, it); @@ -282,16 +298,18 @@ static int osd_it_add_processed(struct osd_it_quota *it, int depth) */ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di) { - struct osd_it_quota *it = (struct osd_it_quota *)di; - const struct lu_fid *fid = - lu_object_fid(&it->oiq_obj->oo_dt.do_lu); - int type = fid2type(fid); - int depth, rc; - uint index; + struct osd_it_quota *it = (struct osd_it_quota *)di; + const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu); + int type; + int depth, rc; + uint index; ENTRY; - /* Let's first check if there are any remaining valid entry in the + type = fid2type(fid); + + /* + * Let's first check if there are any remaining valid entry in the * current leaf block. Start with the next entry after the current one. */ depth = LUSTRE_DQTREEDEPTH; @@ -304,8 +322,10 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di) QUOTA_IT_READ_ERROR(it, rc); RETURN(rc); } else if (rc == 0) { - /* Found on entry, @it is already updated to the - * new position in walk_block_dqentry(). */ + /* + * Found on entry, @it is already updated to the + * new position in walk_block_dqentry(). + */ RETURN(0); } else { rc = osd_it_add_processed(it, depth); @@ -319,12 +339,16 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di) } rc = 1; - /* We have consumed all the entries of the current leaf block, move on - * to the next one. */ + /* + * We have consumed all the entries of the current leaf block, move on + * to the next one. + */ depth--; - /* We keep searching as long as walk_tree_dqentry() returns +1 - * (= no valid entry found). */ + /* + * We keep searching as long as walk_tree_dqentry() returns +1 + * (= no valid entry found). + */ for (; depth >= 0 && rc > 0; depth--) { index = it->oiq_index[depth]; if (++index > 0xff) @@ -376,9 +400,9 @@ static int osd_it_acct_rec(const struct lu_env *env, const struct dt_it *di, struct dt_rec *dtrec, __u32 attr) { - struct osd_it_quota *it = (struct osd_it_quota *)di; - const struct dt_key *key = osd_it_acct_key(env, di); - int rc; + struct osd_it_quota *it = (struct osd_it_quota *)di; + const struct dt_key *key = osd_it_acct_key(env, di); + int rc; ENTRY; @@ -414,7 +438,18 @@ static __u64 osd_it_acct_store(const struct lu_env *env, static int osd_it_acct_load(const struct lu_env *env, const struct dt_it *di, __u64 hash) { + struct osd_it_quota *it = (struct osd_it_quota *)di; + ENTRY; + + /* + * LU-8999 - If it is called to resume the iteration, calling + * osd_it_acct_get could change the block orders in the lower level + * of the quota tree, which are saved in osd_it_quota->oiq_blk. + */ + if (it->oiq_id != 0 && it->oiq_id == hash) + RETURN(1); + RETURN(osd_it_acct_get(env, (struct dt_it *)di, (const struct dt_key *)&hash)); } @@ -423,8 +458,8 @@ static int osd_it_acct_load(const struct lu_env *env, * Index and Iterator operations for accounting objects */ const struct dt_index_operations osd_acct_index_ops = { - .dio_lookup = osd_acct_index_lookup, - .dio_it = { + .dio_lookup = osd_acct_index_lookup, + .dio_it = { .init = osd_it_acct_init, .fini = osd_it_acct_fini, .get = osd_it_acct_get, @@ -445,7 +480,7 @@ static inline void osd_quota_swab(char *ptr, size_t size) LASSERT((size & (sizeof(__u64) - 1)) == 0); for (offset = 0; offset < size; offset += sizeof(__u64)) - __swab64s((__u64 *)(ptr + offset)); + __swab64s((__u64 *)(ptr + offset)); } const struct dt_rec *osd_quota_pack(struct osd_object *obj, @@ -476,19 +511,12 @@ void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec) descr = obj->oo_dir->od_container.ic_descr; osd_quota_swab((char *)rec, descr->id_rec_size); -#else - return; #endif } static inline int osd_qid_type(struct osd_thandle *oh, int i) { - return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA; -} - -static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type) -{ - oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0); + return oh->ot_id_types[i]; } /** @@ -508,22 +536,33 @@ static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type) */ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh, struct lquota_id_info *qi, struct osd_object *obj, - bool enforce, int *flags) + bool enforce, enum osd_quota_local_flags *local_flags) { - struct osd_device *dev = osd_dt_dev(oh->ot_super.th_dev); - struct qsd_instance *qsd = dev->od_quota_slave; - struct inode *inode = NULL; - int i, rc = 0, crd; - bool found = false; + struct osd_device *dev; + struct qsd_instance *qsd; + struct inode *inode = NULL; + int i, rc = 0, crd; + __u8 res = qi->lqi_is_blk ? LQUOTA_RES_DT : LQUOTA_RES_MD; + bool found = false; + ENTRY; LASSERT(oh != NULL); LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n", oh->ot_id_cnt); + dev = osd_dt_dev(oh->ot_super.th_dev); + LASSERT(dev != NULL); + + if (res == LQUOTA_RES_DT) + qsd = dev->od_quota_slave_dt; + else + qsd = dev->od_quota_slave_md; + for (i = 0; i < oh->ot_id_cnt; i++) { if (oh->ot_id_array[i] == qi->lqi_id.qid_uid && - osd_qid_type(oh, i) == qi->lqi_type) { + oh->ot_id_res[i] == res && + oh->ot_id_types[i] == qi->lqi_type) { found = true; break; } @@ -539,28 +578,27 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh, if (obj != NULL) inode = obj->oo_inode; - /* root ID entry should be always present in the quota file */ if (qi->lqi_id.qid_uid == 0) { + /* root ID should be always present in the quota file */ crd = 1; } else { - /* used space for this ID could be dropped to zero, + /* can't rely on the current state as it can change + * by the execution. + * if used space for this ID could be dropped to zero, * reserve extra credits for removing ID entry from - * the quota file */ + * the quota file + */ if (qi->lqi_space < 0) crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev)); - /* reserve credits for adding ID entry to the quota - * file if the i_dquot isn't initialized yet. */ - else if (inode == NULL || - inode->i_dquot[qi->lqi_type] == NULL) - crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev)); else - crd = 1; + crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev)); } osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd); oh->ot_id_array[i] = qi->lqi_id.qid_uid; - osd_qid_set_type(oh, i, qi->lqi_type); + oh->ot_id_types[i] = qi->lqi_type; + oh->ot_id_res[i] = res; oh->ot_id_cnt++; } @@ -570,7 +608,8 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh, /* check quota */ if (enforce) - rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags); + rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, + local_flags); RETURN(rc); } @@ -583,142 +622,73 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh, * \param space - how many blocks/inodes will be consumed/released * \param oh - osd transaction handle * \param obj - osd object, could be NULL when it's under create - * \param is_blk - block quota or inode quota? * \param flags - if the operation is write, return no user quota, no * group quota, or sync commit flags to the caller - * \param force - set to 1 when changes are performed by root user and thus - * can't failed with EDQUOT + * \param osd_qid_flags - indicate this is a inode/block accounting + * and whether changes are performed by root user * * \retval 0 - success * \retval -ve - failure */ int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid, - long long space, struct osd_thandle *oh, - struct osd_object *obj, bool is_blk, int *flags, - bool force) + __u32 projid, long long space, struct osd_thandle *oh, + struct osd_object *obj, + enum osd_quota_local_flags *local_flags, + enum osd_qid_declare_flags osd_qid_declare_flags) { - struct osd_thread_info *info = osd_oti_get(env); - struct lquota_id_info *qi = &info->oti_qi; - int rcu, rcg; /* user & group rc */ + struct osd_thread_info *info = osd_oti_get(env); + struct lquota_id_info *qi = &info->oti_qi; + int rcu, rcg, rcp = 0; /* user & group & project rc */ + struct thandle *th = &oh->ot_super; + bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE) || + th->th_ignore_quota; ENTRY; /* let's start with user quota */ qi->lqi_id.qid_uid = uid; - qi->lqi_type = USRQUOTA; - qi->lqi_space = space; - qi->lqi_is_blk = is_blk; - rcu = osd_declare_qid(env, oh, qi, obj, true, flags); + qi->lqi_type = USRQUOTA; + qi->lqi_space = space; + qi->lqi_is_blk = !!(osd_qid_declare_flags & OSD_QID_BLK); + rcu = osd_declare_qid(env, oh, qi, obj, true, local_flags); if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS)) /* ignore EDQUOT & EINPROGRESS when changes are done by root */ rcu = 0; - /* For non-fatal error, we want to continue to get the noquota flags + /* + * For non-fatal error, we want to continue to get the noquota flags * for group id. This is only for commit write, which has @flags passed * in. See osd_declare_write_commit(). - * When force is set to true, we also want to proceed with the gid */ - if (rcu && (rcu != -EDQUOT || flags == NULL)) + * When force is set to true, we also want to proceed with the gid + */ + if (rcu && (rcu != -EDQUOT || local_flags == NULL)) RETURN(rcu); /* and now group quota */ qi->lqi_id.qid_gid = gid; - qi->lqi_type = GRPQUOTA; - rcg = osd_declare_qid(env, oh, qi, obj, true, flags); + qi->lqi_type = GRPQUOTA; + rcg = osd_declare_qid(env, oh, qi, obj, true, local_flags); if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS)) /* as before, ignore EDQUOT & EINPROGRESS for root */ rcg = 0; - RETURN(rcu ? rcu : rcg); -} - -int osd_quota_migration(const struct lu_env *env, struct dt_object *dt) -{ - struct osd_thread_info *oti = osd_oti_get(env); - struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt)); - struct dt_object *root, *parent = NULL, *admin = NULL; - dt_obj_version_t version; - char *fname, *fnames[] = {ADMIN_USR, ADMIN_GRP}; - int rc, i; - ENTRY; - - /* not newly created global index */ - version = dt_version_get(env, dt); - if (version != 0) - RETURN(0); +#ifdef HAVE_PROJECT_QUOTA + if (rcg && (rcg != -EDQUOT || local_flags == NULL)) + RETURN(rcg); - /* locate root */ - rc = dt_root_get(env, &osd->od_dt_dev, &oti->oti_fid); - if (rc) { - CERROR("%s: Can't get root FID, rc:%d\n", osd->od_svname, rc); - RETURN(rc); - } + /* and now project quota */ + qi->lqi_id.qid_projid = projid; + qi->lqi_type = PRJQUOTA; + rcp = osd_declare_qid(env, oh, qi, obj, true, local_flags); - root = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid); - if (IS_ERR(root)) { - CERROR("%s: Failed to locate root "DFID", rc:%ld\n", - osd->od_svname, PFID(&oti->oti_fid), PTR_ERR(root)); - RETURN(PTR_ERR(root)); - } - - /* locate /OBJECTS */ - rc = dt_lookup_dir(env, root, OBJECTS, &oti->oti_fid); - if (rc == -ENOENT) { - GOTO(out, rc = 0); - } else if (rc) { - CERROR("%s: Failed to lookup %s, rc:%d\n", - osd->od_svname, OBJECTS, rc); - GOTO(out, rc); - } - - parent = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid); - if (IS_ERR(parent)) { - CERROR("%s: Failed to locate %s "DFID", rc:%ld\n", - osd->od_svname, OBJECTS, PFID(&oti->oti_fid), - PTR_ERR(parent)); - GOTO(out, rc = PTR_ERR(parent)); + if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS)) { + CDEBUG(D_QUOTA, "forced to ignore quota flags = %#x\n", + local_flags ? *local_flags : -1); + /* as before, ignore EDQUOT & EINPROGRESS for root */ + rcp = 0; } +#endif - /* locate quota admin files */ - for (i = 0; i < 2; i++) { - fname = fnames[i]; - rc = dt_lookup_dir(env, parent, fname, &oti->oti_fid); - if (rc == -ENOENT) { - rc = 0; - continue; - } else if (rc) { - CERROR("%s: Failed to lookup %s, rc:%d\n", - osd->od_svname, fname, rc); - GOTO(out, rc); - } - - admin = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid); - if (IS_ERR(admin)) { - CERROR("%s: Failed to locate %s "DFID", rc:%d\n", - osd->od_svname, fname, PFID(&oti->oti_fid), rc); - GOTO(out, rc = PTR_ERR(admin)); - } - - if (!dt_object_exists(admin)) { - CERROR("%s: Old admin file %s doesn't exist, but is " - "still referenced in parent directory.\n", - osd->od_svname, fname); - lu_object_put(env, &admin->do_lu); - GOTO(out, rc = -ENOENT); - } - - LCONSOLE_WARN("%s: Detected old quota admin file(%s)! If you " - "want to keep the old quota limits settings, " - "please upgrade to lower version(2.5) first to " - "convert them into new format.\n", - osd->od_svname, fname); - - lu_object_put(env, &admin->do_lu); - GOTO(out, rc = -EINVAL); - } -out: - if (parent && !IS_ERR(parent)) - lu_object_put(env, &parent->do_lu); - lu_object_put(env, &root->do_lu); - RETURN(rc); + RETURN(rcu ? rcu : (rcg ? rcg : rcp)); }