X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;ds=sidebyside;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=2cf70763693502021d463e7f747da5697b595e10;hb=4e9958fb4e6e96f2a7367d734da8f4597fc07243;hp=eced8dba0d094b132b11cb32dfe174431718bcda;hpb=57df6353ca590f32a32c0f0dc4a1d10a86a027ec;p=fs%2Flustre-release.git diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c index eced8db..2cf7076 100644 --- a/lustre/osd-ldiskfs/osd_handler.c +++ b/lustre/osd-ldiskfs/osd_handler.c @@ -72,8 +72,7 @@ /* llo_* api support */ #include -/* dt_acct_features */ -#include +#include #ifdef HAVE_LDISKFS_PDO int ldiskfs_pdo = 1; @@ -90,6 +89,7 @@ static const char remote_obj_dir[] = "REM_OBJ_DIR"; static const struct lu_object_operations osd_lu_obj_ops; static const struct dt_object_operations osd_obj_ops; static const struct dt_object_operations osd_obj_ea_ops; +static const struct dt_object_operations osd_obj_otable_it_ops; static const struct dt_index_operations osd_index_iam_ops; static const struct dt_index_operations osd_index_ea_ops; @@ -103,41 +103,6 @@ static int osd_object_invariant(const struct lu_object *l) return osd_invariant(osd_obj(l)); } -#ifdef HAVE_QUOTA_SUPPORT -static inline void -osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save) -{ - struct md_ucred *uc = md_ucred(env); - struct cred *tc; - - LASSERT(uc != NULL); - - save->oc_uid = current_fsuid(); - save->oc_gid = current_fsgid(); - save->oc_cap = current_cap(); - if ((tc = prepare_creds())) { - tc->fsuid = uc->mu_fsuid; - tc->fsgid = uc->mu_fsgid; - commit_creds(tc); - } - /* XXX not suboptimal */ - cfs_curproc_cap_unpack(uc->mu_cap); -} - -static inline void -osd_pop_ctxt(struct osd_ctxt *save) -{ - struct cred *tc; - - if ((tc = prepare_creds())) { - tc->fsuid = save->oc_uid; - tc->fsgid = save->oc_gid; - tc->cap_effective = save->oc_cap; - commit_creds(tc); - } -} -#endif - /* * Concurrency: doesn't matter */ @@ -165,51 +130,6 @@ static int osd_root_get(const struct lu_env *env, return 0; } -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); -} - -void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh, - int type, uid_t id, struct inode *inode) -{ -#ifdef CONFIG_QUOTA - int i, allocated = 0; - struct osd_object *obj; - - LASSERT(dt != NULL); - LASSERT(oh != NULL); - LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u", - oh->ot_id_cnt); - - /* id entry is allocated in the quota file */ - if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off) - allocated = 1; - - for (i = 0; i < oh->ot_id_cnt; i++) { - if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type) - return; - } - - if (unlikely(i >= OSD_MAX_UGID_CNT)) { - CERROR("more than %d uid/gids for a transaction?\n", i); - return; - } - - oh->ot_id_array[i] = id; - osd_qid_set_type(oh, i, type); - oh->ot_id_cnt++; - obj = osd_dt_obj(dt); - oh->ot_credits += (allocated || id == 0) ? - 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj))); -#endif -} - /* * OSD object methods. */ @@ -300,8 +220,8 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev, iput(inode); inode = ERR_PTR(-ESTALE); } else if (is_bad_inode(inode)) { - CWARN("%s: bad inode: ino = %u\n", - dev->od_dt_dev.dd_lu_dev.ld_obd->obd_name, id->oii_ino); + CWARN("%.16s: bad inode: ino = %u\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino); iput(inode); inode = ERR_PTR(-ENOENT); } else { @@ -355,12 +275,10 @@ osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev, return inode; rc = osd_get_lma(inode, &info->oti_obj_dentry, lma); + if (rc == -ENODATA) + return inode; + if (rc != 0) { - if (rc == -ENODATA) { - CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n", - inode->i_ino, PFID(fid)); - rc = -EREMCHG; - } iput(inode); return ERR_PTR(rc); } @@ -371,6 +289,7 @@ osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev, iput(inode); return ERR_PTR(EREMCHG); } + return inode; } @@ -386,7 +305,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, struct inode *inode; struct osd_scrub *scrub; struct scrub_file *sf; - int result; + int result; int verify = 0; ENTRY; @@ -405,38 +324,31 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT)) RETURN(-ENOENT); - if (fid_is_norm(fid)) { - /* Search order: 1. per-thread cache. */ - if (lu_fid_eq(fid, &oic->oic_fid)) { + /* Search order: 1. per-thread cache. */ + if (lu_fid_eq(fid, &oic->oic_fid)) { + goto iget; + } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) { + /* Search order: 2. OI scrub pending list. */ + result = osd_oii_lookup(dev, fid, id); + if (result == 0) goto iget; - } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) { - /* Search order: 2. OI scrub pending list. */ - result = osd_oii_lookup(dev, fid, id); - if (result == 0) - goto iget; - } - - if (sf->sf_flags & SF_INCONSISTENT) - verify = 1; } - fid_zero(&oic->oic_fid); + if (sf->sf_flags & SF_INCONSISTENT) + verify = 1; + + /* + * Objects are created as locking anchors or place holders for objects + * yet to be created. 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() + */ + if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0) + GOTO(out, result = 0); + /* Search order: 3. OI files. */ result = osd_oi_lookup(info, dev, fid, id); - if (result != 0 && result != -ENOENT) - GOTO(out, result); - - /* If fid wasn't found in oi, inode-less object is created, - * for which lu_object_exists() returns false. This is used - * in a (frequent) case when objects are created as locking - * anchors or place holders for objects yet to be created. */ - if (conf != NULL && conf->loc_flags & LOC_F_NEW) { - if (unlikely(result == 0)) - GOTO(out, result = -EEXIST); - else - GOTO(out, result = 0); - } - if (result == -ENOENT) { if (!fid_is_norm(fid) || !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid), @@ -446,6 +358,9 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj, goto trigger; } + if (result != 0) + GOTO(out, result); + iget: if (verify == 0) inode = osd_iget(info, dev, id); @@ -454,17 +369,19 @@ iget: if (IS_ERR(inode)) { result = PTR_ERR(inode); if (result == -ENOENT || result == -ESTALE) { + fid_zero(&oic->oic_fid); result = 0; } else if (result == -EREMCHG) { trigger: if (thread_is_running(&scrub->os_thread)) { result = -EINPROGRESS; - } else if (!scrub->os_no_scrub) { + } else if (!dev->od_noscrub) { result = osd_scrub_start(dev); - LCONSOLE_ERROR("Trigger OI scrub by RPC for " - DFID", rc = %d\n", - PFID(fid), result); + LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC " + "for "DFID", rc = %d [1]\n", + LDISKFS_SB(osd_sb(dev))->s_es->\ + s_volume_name,PFID(fid), result); if (result == 0 || result == -EALREADY) result = -EINPROGRESS; else @@ -523,13 +440,18 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l, LINVRNT(osd_invariant(obj)); result = osd_fid_lookup(env, obj, lu_object_fid(l), conf); - obj->oo_dt.do_body_ops = &osd_body_ops_new; - if (result == 0) { - if (obj->oo_inode != NULL) - osd_object_init0(obj); - } - LINVRNT(osd_invariant(obj)); - return result; + obj->oo_dt.do_body_ops = &osd_body_ops_new; + if (result == 0) { + if (obj->oo_inode != NULL) { + osd_object_init0(obj); + } else if (fid_is_otable_it(&l->lo_header->loh_fid)) { + obj->oo_dt.do_ops = &osd_obj_otable_it_ops; + /* LFSCK iterator object is special without inode */ + l->lo_header->loh_attr |= LOHA_EXISTS; + } + } + LINVRNT(osd_invariant(obj)); + return result; } /* @@ -698,6 +620,9 @@ static void osd_trans_commit_cb(struct journal_callback *jcb, int error) /* call per-transaction callbacks if any */ cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) { + LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC, + "commit callback entry: magic=%x name='%s'\n", + dcb->dcb_magic, dcb->dcb_name); cfs_list_del_init(&dcb->dcb_linkage); dcb->dcb_func(NULL, th, dcb, error); } @@ -726,6 +651,8 @@ static struct thandle *osd_trans_create(const struct lu_env *env, th = ERR_PTR(-ENOMEM); OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO); if (oh != NULL) { + oh->ot_quota_trans = &oti->oti_quota_trans; + memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans)); th = &oh->ot_super; th->th_dev = d; th->th_result = 0; @@ -763,9 +690,10 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, GOTO(out, rc); if (!osd_param_is_sane(dev, th)) { - CWARN("%s: too many transaction credits (%d > %d)\n", - d->dd_lu_dev.ld_obd->obd_name, oh->ot_credits, - osd_journal(dev)->j_max_transaction_buffers); + CWARN("%.16s: too many transaction credits (%d > %d)\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, + oh->ot_credits, + osd_journal(dev)->j_max_transaction_buffers); /* XXX Limit the credits to 'max_transaction_buffers', and * let the underlying filesystem to catch the error if * we really need so many credits. @@ -801,22 +729,6 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d, lu_device_get(&d->dd_lu_dev); oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference, "osd-tx", th); - - /* - * XXX: current rule is that we first start tx, - * then lock object(s), but we can't use - * this rule for data (due to locking specifics - * in ldiskfs). also in long-term we'd like to - * use usually-used (locks;tx) ordering. so, - * UGLY thing is that we'll use one ordering for - * data (ofd) and reverse ordering for metadata - * (mdd). then at some point we'll fix the latter - */ - if (lu_device_is_md(&d->dd_lu_dev)) { - LASSERT(oti->oti_r_locks == 0); - LASSERT(oti->oti_w_locks == 0); - } - oti->oti_txns++; rc = 0; } else { @@ -835,16 +747,20 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) struct osd_thandle *oh; struct osd_thread_info *oti = osd_oti_get(env); struct osd_iobuf *iobuf = &oti->oti_iobuf; - + struct qsd_instance *qsd = oti->oti_dev->od_quota_slave; ENTRY; oh = container_of0(th, struct osd_thandle, ot_super); + if (qsd != NULL) + /* inform the quota slave device that the transaction is + * stopping */ + qsd_op_end(env, qsd, oh->ot_quota_trans); + oh->ot_quota_trans = NULL; + if (oh->ot_handle != NULL) { handle_t *hdl = oh->ot_handle; - hdl->h_sync = th->th_sync; - /* * add commit callback * notice we don't do this in osd_trans_start() @@ -855,23 +771,13 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) LASSERT(oti->oti_txns == 1); oti->oti_txns--; - /* - * XXX: current rule is that we first start tx, - * then lock object(s), but we can't use - * this rule for data (due to locking specifics - * in ldiskfs). also in long-term we'd like to - * use usually-used (locks;tx) ordering. so, - * UGLY thing is that we'll use one ordering for - * data (ofd) and reverse ordering for metadata - * (mdd). then at some point we'll fix the latter - */ - if (lu_device_is_md(&th->th_dev->dd_lu_dev)) { - LASSERT(oti->oti_r_locks == 0); - LASSERT(oti->oti_w_locks == 0); - } rc = dt_txn_hook_stop(env, th); if (rc != 0) CERROR("Failure in transaction hook: %d\n", rc); + + /* hook functions might modify th_sync */ + hdl->h_sync = th->th_sync; + oh->ot_handle = NULL; OSD_CHECK_SLOW_TH(oh, oti->oti_dev, rc = ldiskfs_journal_stop(hdl)); @@ -900,12 +806,14 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th) 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_of0(th, struct osd_thandle, + ot_super); - cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); + LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC); + LASSERT(&dcb->dcb_func != NULL); + cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list); - return 0; + return 0; } /* @@ -928,8 +836,24 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l) osd_index_fini(obj); if (inode != NULL) { + struct qsd_instance *qsd = osd_obj2dev(obj)->od_quota_slave; + qid_t uid = inode->i_uid; + qid_t gid = inode->i_gid; + iput(inode); obj->oo_inode = NULL; + + if (qsd != NULL) { + 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_adjust_quota(env, qsd, &qi->lqi_id, USRQUOTA); + + qi->lqi_id.qid_uid = gid; + qsd_adjust_quota(env, qsd, &qi->lqi_id, GRPQUOTA); + } } } @@ -954,7 +878,8 @@ static int osd_object_print(const struct lu_env *env, void *cookie, d = o->oo_dir->od_container.ic_descr; else d = NULL; - return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]", + 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, @@ -972,6 +897,9 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, struct kstatfs *ksfs; 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); @@ -981,15 +909,17 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, ksfs = &osd_oti_get(env)->oti_ksfs; } - cfs_spin_lock(&osd->od_osfs_lock); - /* cache 1 second */ - if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) { - result = ll_do_statfs(sb, ksfs); - if (likely(result == 0)) { /* N.B. statfs can't really fail */ - osd->od_osfs_age = cfs_time_current_64(); - statfs_pack(&osd->od_statfs, ksfs); - } - } + cfs_spin_lock(&osd->od_osfs_lock); + /* cache 1 second */ + if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) { + result = sb->s_op->statfs(sb->s_root, ksfs); + if (likely(result == 0)) { /* N.B. statfs can't really fail */ + osd->od_osfs_age = cfs_time_current_64(); + statfs_pack(&osd->od_statfs, ksfs); + if (sb->s_flags & MS_RDONLY) + sfs->os_state = OS_STATE_READONLY; + } + } if (likely(result == 0)) *sfs = osd->od_statfs; @@ -1001,6 +931,17 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d, return result; } +/** + * Estimate space needed for file creations. We assume the largest filename + * which is 2^64 - 1, hence a filename of 20 chars. + * This is 28 bytes per object which is 28MB for 1M objects ... no so bad. + */ +#ifdef __LDISKFS_DIR_REC_LEN +#define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20) +#else +#define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20) +#endif + /* * Concurrency: doesn't access mutable data. */ @@ -1013,9 +954,20 @@ static void osd_conf_get(const struct lu_env *env, /* * XXX should be taken from not-yet-existing fs abstraction layer. */ + param->ddp_mnt = osd_dt_dev(dev)->od_mnt; param->ddp_max_name_len = LDISKFS_NAME_LEN; param->ddp_max_nlink = LDISKFS_LINK_MAX; param->ddp_block_shift = sb->s_blocksize_bits; + param->ddp_mount_type = LDD_MT_LDISKFS; + param->ddp_maxbytes = sb->s_maxbytes; + /* Overhead estimate should be fairly accurate, so we really take a tiny + * error margin which also avoids fragmenting the filesystem too much */ + param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */ + /* inode are statically allocated, so per-inode space consumption + * is the space consumed by the directory entry */ + param->ddp_inodespace = PER_OBJ_USAGE; + /* per-fragment overhead to be used by the client code */ + param->ddp_grant_frag = 6 * LDISKFS_BLOCK_SIZE(sb); param->ddp_mntopts = 0; if (test_opt(sb, XATTR_USER)) param->ddp_mntopts |= MNTOPT_USERXATTR; @@ -1049,7 +1001,7 @@ static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t */ static int osd_sync(const struct lu_env *env, struct dt_device *d) { - CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME); + CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); return ldiskfs_force_commit(osd_sb(osd_dt_dev(d))); } @@ -1071,7 +1023,7 @@ static int osd_commit_async(const struct lu_env *env, struct super_block *s = osd_sb(osd_dt_dev(d)); ENTRY; - CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME); + CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME); RETURN(s->s_op->sync_fs(s, 0)); } @@ -1085,7 +1037,7 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d) int rc; ENTRY; - CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME); + CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname); rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev); RETURN(rc); @@ -1109,25 +1061,6 @@ static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d, } /** - * Concurrency: serialization provided by callers. - */ -static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d, - struct dt_quota_ctxt *ctxt, void *data) -{ - struct obd_device *obd = (void *)ctxt; - struct vfsmount *mnt = (struct vfsmount *)data; - ENTRY; - - obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb; - OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt); - obd->obd_lvfs_ctxt.pwdmnt = mnt; - obd->obd_lvfs_ctxt.pwd = mnt->mnt_root; - obd->obd_lvfs_ctxt.fs = get_ds(); - - EXIT; -} - -/** * Note: we do not count into QUOTA here. * If we mount with --data_journal we may need more. */ @@ -1195,7 +1128,6 @@ static const struct dt_device_operations osd_dt_ops = { .dt_ro = osd_ro, .dt_commit_async = osd_commit_async, .dt_init_capa_ctxt = osd_init_capa_ctxt, - .dt_init_quota_ctxt= osd_init_quota_ctxt, }; static void osd_object_read_lock(const struct lu_env *env, @@ -1368,15 +1300,15 @@ int osd_object_auth(const struct lu_env *env, struct dt_object *dt, } static struct timespec *osd_inode_time(const struct lu_env *env, - struct inode *inode, __u64 seconds) + struct inode *inode, __u64 seconds) { - struct osd_thread_info *oti = osd_oti_get(env); - struct timespec *t = &oti->oti_time; + 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, get_sb_time_gran(inode->i_sb)); - return t; + t->tv_sec = seconds; + t->tv_nsec = 0; + *t = timespec_trunc(*t, inode->i_sb->s_time_gran); + return t; } @@ -1398,8 +1330,8 @@ static void osd_inode_getattr(const struct lu_env *env, attr->la_flags = LDISKFS_I(inode)->i_flags; attr->la_nlink = inode->i_nlink; attr->la_rdev = inode->i_rdev; - attr->la_blksize = ll_inode_blksize(inode); - attr->la_blkbits = inode->i_blkbits; + attr->la_blksize = 1 << inode->i_blkbits; + attr->la_blkbits = inode->i_blkbits; } static int osd_attr_get(const struct lu_env *env, @@ -1426,35 +1358,139 @@ static int osd_declare_attr_set(const struct lu_env *env, const struct lu_attr *attr, struct thandle *handle) { - struct osd_thandle *oh; - struct osd_object *obj; + struct osd_thandle *oh; + struct osd_object *obj; + struct osd_thread_info *info = osd_oti_get(env); + struct lquota_id_info *qi = &info->oti_qi; + long long bspace; + int rc = 0; + bool allocated; + ENTRY; - LASSERT(dt != NULL); - LASSERT(handle != NULL); + LASSERT(dt != NULL); + LASSERT(handle != NULL); - obj = osd_dt_obj(dt); - LASSERT(osd_invariant(obj)); + obj = osd_dt_obj(dt); + LASSERT(osd_invariant(obj)); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, attr_set); - oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE]; + OSD_DECLARE_OP(oh, attr_set); + oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE]; - if (attr && attr->la_valid & LA_UID) { - if (obj->oo_inode) - osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid, - obj->oo_inode); - osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL); - } - if (attr && attr->la_valid & LA_GID) { - if (obj->oo_inode) - osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid, - obj->oo_inode); - osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL); - } + if (attr == NULL || obj->oo_inode == NULL) + RETURN(rc); - return 0; + bspace = obj->oo_inode->i_blocks; + bspace <<= obj->oo_inode->i_sb->s_blocksize_bits; + bspace = toqb(bspace); + + /* Changing ownership is always preformed by super user, it should not + * fail with EDQUOT. + * + * We still need to call the osd_declare_qid() to calculate the journal + * credits for updating quota accounting files and to trigger quota + * space adjustment once the operation is completed.*/ + if ((attr->la_valid & LA_UID) != 0 && + attr->la_uid != obj->oo_inode->i_uid) { + qi->lqi_type = USRQUOTA; + + /* inode accounting */ + qi->lqi_is_blk = false; + + /* one more inode for the new owner ... */ + qi->lqi_id.qid_uid = attr->la_uid; + qi->lqi_space = 1; + allocated = (attr->la_uid == 0) ? true : false; + rc = osd_declare_qid(env, oh, qi, allocated, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* and one less inode for the current uid */ + qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + qi->lqi_space = -1; + rc = osd_declare_qid(env, oh, qi, true, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* block accounting */ + qi->lqi_is_blk = true; + + /* more blocks for the new owner ... */ + qi->lqi_id.qid_uid = attr->la_uid; + qi->lqi_space = bspace; + allocated = (attr->la_uid == 0) ? true : false; + rc = osd_declare_qid(env, oh, qi, allocated, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* and finally less blocks for the current owner */ + qi->lqi_id.qid_uid = obj->oo_inode->i_uid; + qi->lqi_space = -bspace; + rc = osd_declare_qid(env, oh, qi, true, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + } + + if (attr->la_valid & LA_GID && + attr->la_gid != obj->oo_inode->i_gid) { + qi->lqi_type = GRPQUOTA; + + /* inode accounting */ + qi->lqi_is_blk = false; + + /* one more inode for the new group owner ... */ + qi->lqi_id.qid_gid = attr->la_gid; + qi->lqi_space = 1; + allocated = (attr->la_gid == 0) ? true : false; + rc = osd_declare_qid(env, oh, qi, allocated, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* and one less inode for the current gid */ + qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + qi->lqi_space = -1; + rc = osd_declare_qid(env, oh, qi, true, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* block accounting */ + qi->lqi_is_blk = true; + + /* more blocks for the new owner ... */ + qi->lqi_id.qid_gid = attr->la_gid; + qi->lqi_space = bspace; + allocated = (attr->la_gid == 0) ? true : false; + rc = osd_declare_qid(env, oh, qi, allocated, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + + /* and finally less blocks for the current owner */ + qi->lqi_id.qid_gid = obj->oo_inode->i_gid; + qi->lqi_space = -bspace; + rc = osd_declare_qid(env, oh, qi, true, NULL); + if (rc == -EDQUOT || rc == -EINPROGRESS) + rc = 0; + if (rc) + RETURN(rc); + } + + RETURN(rc); } static int osd_inode_setattr(const struct lu_env *env, @@ -1549,34 +1585,12 @@ static int osd_attr_set(const struct lu_env *env, OSD_EXEC_OP(handle, attr_set); inode = obj->oo_inode; - if (LDISKFS_HAS_RO_COMPAT_FEATURE(inode->i_sb, - LDISKFS_FEATURE_RO_COMPAT_QUOTA)) { - rc = osd_quota_transfer(inode, attr); - if (rc) - return rc; - } else { -#ifdef HAVE_QUOTA_SUPPORT - if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) || - (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) { - struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt; - struct iattr iattr; - int rc; - - iattr.ia_valid = 0; - if (attr->la_valid & LA_UID) - iattr.ia_valid |= ATTR_UID; - if (attr->la_valid & LA_GID) - iattr.ia_valid |= ATTR_GID; - iattr.ia_uid = attr->la_uid; - iattr.ia_gid = attr->la_gid; - osd_push_ctxt(env, save); - rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0; - osd_pop_ctxt(save); - if (rc != 0) - return rc; - } -#endif - } + ll_vfs_dq_init(inode); + + rc = osd_quota_transfer(inode, attr); + if (rc) + return rc; + cfs_spin_lock(&obj->oo_guard); rc = osd_inode_setattr(env, inode, attr); cfs_spin_unlock(&obj->oo_guard); @@ -1603,9 +1617,6 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, struct osd_thandle *oth; struct dt_object *parent = NULL; struct inode *inode; -#ifdef HAVE_QUOTA_SUPPORT - struct osd_ctxt *save = &info->oti_ctxt; -#endif LINVRNT(osd_invariant(obj)); LASSERT(obj->oo_inode == NULL); @@ -1623,16 +1634,10 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj, if (hint && hint->dah_parent) parent = hint->dah_parent; -#ifdef HAVE_QUOTA_SUPPORT - osd_push_ctxt(info->oti_env, save); -#endif inode = ldiskfs_create_inode(oth->ot_handle, parent ? osd_dt_obj(parent)->oo_inode : osd_sb(osd)->s_root->d_inode, mode); -#ifdef HAVE_QUOTA_SUPPORT - osd_pop_ctxt(save); -#endif if (!IS_ERR(inode)) { /* Do not update file c/mtime in ldiskfs. * NB: don't need any lock because no contention at this @@ -1807,7 +1812,8 @@ static osd_obj_type_f osd_create_type_f(enum dt_format_type type) static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah, - struct dt_object *parent, cfs_umode_t child_mode) + struct dt_object *parent, struct dt_object *child, + cfs_umode_t child_mode) { LASSERT(ah); @@ -1834,16 +1840,9 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj, if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime))) attr->la_valid &= ~LA_MTIME; - if (LDISKFS_HAS_RO_COMPAT_FEATURE(inode->i_sb, - LDISKFS_FEATURE_RO_COMPAT_QUOTA)) { - result = osd_quota_transfer(inode, attr); - if (result) - return; - } else { -#ifdef HAVE_QUOTA_SUPPORT - attr->la_valid &= ~(LA_UID | LA_GID); -#endif - } + result = osd_quota_transfer(inode, attr); + if (result) + return; if (attr->la_valid != 0) { result = osd_inode_setattr(info->oti_env, inode, attr); @@ -1906,10 +1905,8 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj, struct osd_thread_info *info = osd_oti_get(env); struct osd_inode_id *id = &info->oti_id; struct osd_device *osd = osd_obj2dev(obj); - struct md_ucred *uc = md_ucred(env); LASSERT(obj->oo_inode != NULL); - LASSERT(uc != NULL); osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation); return osd_oi_insert(info, osd, fid, id, th); @@ -1922,7 +1919,9 @@ static int osd_declare_object_create(const struct lu_env *env, struct dt_object_format *dof, struct thandle *handle) { - struct osd_thandle *oh; + struct osd_thandle *oh; + int rc; + ENTRY; LASSERT(handle != NULL); @@ -1936,6 +1935,9 @@ static int osd_declare_object_create(const struct lu_env *env, if (fid_is_norm(lu_object_fid(&dt->do_lu))) { OSD_DECLARE_OP(oh, insert); oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; + /* Reuse idle OI block may cause additional one OI block + * to be changed. */ + oh->ot_credits += 1; } /* If this is directory, then we expect . and .. to be inserted as * well. The one directory block always needs to be created for the @@ -1948,11 +1950,12 @@ static int osd_declare_object_create(const struct lu_env *env, oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE]; } - if (attr) { - osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL); - osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL); - } - return 0; + if (!attr) + RETURN(0); + + rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh, + false, false, NULL, false); + RETURN(rc); } static int osd_object_create(const struct lu_env *env, struct dt_object *dt, @@ -2001,7 +2004,7 @@ static int osd_declare_object_destroy(const struct lu_env *env, struct osd_object *obj = osd_dt_obj(dt); struct inode *inode = obj->oo_inode; struct osd_thandle *oh; - + int rc; ENTRY; oh = container_of0(th, struct osd_thandle, ot_super); @@ -2011,12 +2014,24 @@ static int osd_declare_object_destroy(const struct lu_env *env, OSD_DECLARE_OP(oh, destroy); OSD_DECLARE_OP(oh, delete); oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE]; - oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE]; - - osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode); - osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode); + /* XXX: So far, only normal fid needs to be inserted into the OI, + * so only normal fid needs to be removed from the OI also. */ + if (fid_is_norm(lu_object_fid(&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. */ + oh->ot_credits += 3; + } - RETURN(0); + /* one less inode */ + rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh, + false, true, NULL, false); + if (rc) + RETURN(rc); + /* data to be truncated */ + rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, true, + true, NULL, false); + RETURN(rc); } static int osd_object_destroy(const struct lu_env *env, @@ -2089,6 +2104,7 @@ static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt, if (fl & LU_XATTR_CREATE) fs_flags |= XATTR_CREATE; + ll_vfs_dq_init(inode); dentry->d_inode = inode; rc = inode->i_op->setxattr(dentry, name, buf->lb_buf, buf->lb_len, fs_flags); @@ -2353,7 +2369,6 @@ static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt, LASSERT(dt_object_exists(dt)); LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL); - LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj)); if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ)) return -EACCES; @@ -2368,22 +2383,20 @@ static int osd_declare_xattr_set(const struct lu_env *env, const struct lu_buf *buf, const char *name, int fl, struct thandle *handle) { - struct osd_thandle *oh; - - LASSERT(handle != NULL); + struct osd_thandle *oh; - if (strcmp(name, XATTR_NAME_VERSION) == 0) { - /* no credits for version */ - return 0; - } + LASSERT(handle != NULL); - oh = container_of0(handle, struct osd_thandle, ot_super); - LASSERT(oh->ot_handle == NULL); + oh = container_of0(handle, struct osd_thandle, ot_super); + LASSERT(oh->ot_handle == NULL); - OSD_DECLARE_OP(oh, xattr_set); - oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET]; + OSD_DECLARE_OP(oh, xattr_set); + if (strcmp(name, XATTR_NAME_VERSION) == 0) + oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE]; + else + oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET]; - return 0; + return 0; } /* @@ -2492,6 +2505,7 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt, OSD_EXEC_OP(handle, xattr_set); + ll_vfs_dq_init(inode); dentry->d_inode = inode; rc = inode->i_op->removexattr(dentry, name); return rc; @@ -2586,23 +2600,23 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env, static int osd_object_sync(const struct lu_env *env, struct dt_object *dt) { - 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 osd_thread_info *info = osd_oti_get(env); + struct dentry *dentry = &info->oti_obj_dentry; + struct file *file = &info->oti_file; + int rc; - ENTRY; + ENTRY; - dentry->d_inode = inode; - file->f_dentry = dentry; - file->f_mapping = inode->i_mapping; - file->f_op = inode->i_fop; - LOCK_INODE_MUTEX(inode); - rc = file->f_op->fsync(file, dentry, 0); - UNLOCK_INODE_MUTEX(inode); - RETURN(rc); + dentry->d_inode = inode; + file->f_dentry = dentry; + file->f_mapping = inode->i_mapping; + file->f_op = inode->i_fop; + mutex_lock(&inode->i_mutex); + rc = file->f_op->fsync(file, dentry, 0); + mutex_unlock(&inode->i_mutex); + RETURN(rc); } static int osd_data_get(const struct lu_env *env, struct dt_object *dt, @@ -2661,17 +2675,6 @@ static int osd_iam_container_init(const struct lu_env *env, return result; result = iam_container_setup(bag); - if (result != 0) - goto out; - - if (osd_obj2dev(obj)->od_iop_mode) { - u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag); - - bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode, - ptr, 0, &result); - } - - out: if (result == 0) obj->oo_dt.do_index_ops = &osd_index_iam_ops; else @@ -2753,9 +2756,21 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt, } LINVRNT(osd_invariant(obj)); + if (is_quota_glb_feat(feat)) + result = osd_quota_migration(env, dt, feat); + return result; } +static int osd_otable_it_attr_get(const struct lu_env *env, + struct dt_object *dt, + struct lu_attr *attr, + struct lustre_capa *capa) +{ + attr->la_valid = 0; + return 0; +} + static const struct dt_object_operations osd_obj_ops = { .do_read_lock = osd_object_read_lock, .do_write_lock = osd_object_write_lock, @@ -2820,6 +2835,11 @@ static const struct dt_object_operations osd_obj_ea_ops = { .do_data_get = osd_data_get, }; +static const struct dt_object_operations osd_obj_otable_it_ops = { + .do_attr_get = osd_otable_it_attr_get, + .do_index_try = osd_index_try, +}; + static int osd_index_declare_iam_delete(const struct lu_env *env, struct dt_object *dt, const struct dt_key *key, @@ -2853,11 +2873,12 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, struct thandle *handle, struct lustre_capa *capa) { - struct osd_object *obj = osd_dt_obj(dt); - struct osd_thandle *oh; - struct iam_path_descr *ipd; - struct iam_container *bag = &obj->oo_dir->od_container; - int rc; + struct osd_thread_info *oti = osd_oti_get(env); + struct osd_object *obj = osd_dt_obj(dt); + struct osd_thandle *oh; + struct iam_path_descr *ipd; + struct iam_container *bag = &obj->oo_dir->od_container; + int rc; ENTRY; @@ -2879,6 +2900,12 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt, 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 */ + oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); + 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)); @@ -2891,6 +2918,9 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; + struct inode *inode; + int rc; + ENTRY; LASSERT(dt_object_exists(dt)); LASSERT(handle != NULL); @@ -2901,13 +2931,12 @@ static int osd_index_declare_ea_delete(const struct lu_env *env, OSD_DECLARE_OP(oh, delete); oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE]; - LASSERT(osd_dt_obj(dt)->oo_inode); - osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid, - osd_dt_obj(dt)->oo_inode); - osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid, - osd_dt_obj(dt)->oo_inode); + inode = osd_dt_obj(dt)->oo_inode; + LASSERT(inode); - return 0; + rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, + true, true, NULL, false); + RETURN(rc); } static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de, @@ -2962,6 +2991,7 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt, if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE)) RETURN(-EACCES); + ll_vfs_dq_init(dir); dentry = osd_child_dentry_get(env, obj, (char *)key, strlen((char *)key)); @@ -3029,6 +3059,12 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, /* 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 */ + oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); + 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)) @@ -3037,10 +3073,14 @@ static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt, iam_rec = (struct iam_rec *) 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); + 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); @@ -3090,11 +3130,8 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, struct iam_path_descr *ipd; struct osd_thandle *oh; struct iam_container *bag = &obj->oo_dir->od_container; -#ifdef HAVE_QUOTA_SUPPORT - cfs_cap_t save = cfs_curproc_cap_pack(); -#endif struct osd_thread_info *oti = osd_oti_get(env); - struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_ldp; + struct iam_rec *iam_rec; int rc; ENTRY; @@ -3116,21 +3153,22 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt, oh = container_of0(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle != NULL); LASSERT(oh->ot_handle->h_transaction != NULL); -#ifdef HAVE_QUOTA_SUPPORT - if (ignore_quota) - cfs_cap_raise(CFS_CAP_SYS_RESOURCE); - else - cfs_cap_lower(CFS_CAP_SYS_RESOURCE); -#endif - if (S_ISDIR(obj->oo_inode->i_mode)) - osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid); - else - iam_rec = (struct iam_rec *) rec; + 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); + } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) { + /* pack quota uid/gid */ + oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); + key = (const struct dt_key *)&oti->oti_quota_id; + /* pack quota record */ + rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec); + iam_rec = (struct iam_rec *)rec; + } else { + iam_rec = (struct iam_rec *)rec; + } + rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key, iam_rec, ipd); -#ifdef HAVE_QUOTA_SUPPORT - cfs_curproc_cap_unpack(save); -#endif osd_ipd_put(env, bag, ipd); LINVRNT(osd_invariant(obj)); RETURN(rc); @@ -3173,6 +3211,8 @@ static int __osd_ea_add_rec(struct osd_thread_info *info, } else { child->d_fsdata = NULL; } + LASSERT(pobj->oo_inode); + ll_vfs_dq_init(pobj->oo_inode); rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock); RETURN(rc); @@ -3299,6 +3339,9 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev, int rc; ENTRY; + if (!fid_is_norm(fid) && !fid_is_igif(fid)) + RETURN(0); + again: rc = osd_oi_lookup(oti, dev, fid, id); if (rc != 0 && rc != -ENOENT) @@ -3319,17 +3362,19 @@ again: RETURN(rc); } - if (!scrub->os_no_scrub && ++once == 1) { + if (!dev->od_noscrub && ++once == 1) { CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n", PFID(fid)); rc = osd_scrub_start(dev); - CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID", rc = %d\n", - PFID(fid), rc); + LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID + ", rc = %d [2]\n", + LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, + PFID(fid), rc); if (rc == 0) goto again; } - RETURN(rc = -EREMCHG); + RETURN(0); } /** @@ -3382,16 +3427,17 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj, rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid); else osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN); - - if (rc != 0 || !fid_is_norm(fid)) + if (rc != 0) { + fid_zero(&oic->oic_fid); GOTO(out, rc); + } oic->oic_fid = *fid; if ((scrub->os_pos_current <= ino) && (sf->sf_flags & SF_INCONSISTENT || ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) - rc = osd_consistency_check(oti, dev, oic); + osd_consistency_check(oti, dev, oic); } else { rc = -ENOENT; } @@ -3423,7 +3469,17 @@ struct osd_object *osd_object_find(const struct lu_env *env, struct lu_object *luch; struct lu_object *lo; - luch = lu_object_find(env, ludev, fid, NULL); + /* + * at this point topdev might not exist yet + * (i.e. MGS is preparing profiles). so we can + * not rely on topdev and instead lookup with + * our device passed as topdev. this can't work + * if the object isn't cached yet (as osd doesn't + * allocate lu_header). IOW, the object must be + * in the cache, otherwise lu_object_alloc() crashes + * -bzzz + */ + luch = lu_object_find_at(env, ludev, fid, NULL); if (!IS_ERR(luch)) { if (lu_object_exists(luch)) { lo = lu_object_locate(luch->lo_header, ludev->ld_type); @@ -3470,6 +3526,9 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, struct thandle *handle) { struct osd_thandle *oh; + struct inode *inode; + int rc; + ENTRY; LASSERT(dt_object_exists(dt)); LASSERT(handle != NULL); @@ -3480,13 +3539,15 @@ static int osd_index_declare_ea_insert(const struct lu_env *env, OSD_DECLARE_OP(oh, insert); oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT]; - LASSERT(osd_dt_obj(dt)->oo_inode); - osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid, - osd_dt_obj(dt)->oo_inode); - osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid, - osd_dt_obj(dt)->oo_inode); + inode = osd_dt_obj(dt)->oo_inode; + LASSERT(inode); - return 0; + /* We ignore block quota on meta pool (MDTs), so needn't + * calculate how many blocks will be consumed by this index + * insert */ + rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh, + true, true, NULL, false); + RETURN(rc); } /** @@ -3509,9 +3570,6 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, struct lu_fid *fid = (struct lu_fid *) rec; const char *name = (const char *)key; struct osd_object *child; -#ifdef HAVE_QUOTA_SUPPORT - cfs_cap_t save = cfs_curproc_cap_pack(); -#endif int rc; ENTRY; @@ -3525,16 +3583,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt, child = osd_object_find(env, dt, fid); if (!IS_ERR(child)) { -#ifdef HAVE_QUOTA_SUPPORT - if (ignore_quota) - cfs_cap_raise(CFS_CAP_SYS_RESOURCE); - else - cfs_cap_lower(CFS_CAP_SYS_RESOURCE); -#endif rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th); -#ifdef HAVE_QUOTA_SUPPORT - cfs_curproc_cap_unpack(save); -#endif osd_object_put(env, child); } else { rc = PTR_ERR(child); @@ -3607,7 +3656,14 @@ 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 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 */ + oti->oti_quota_id = cpu_to_le64(*((__u64 *)key)); + key = (struct dt_key *)&oti->oti_quota_id; + } return iam_it_get(&it->oi_it, (const struct iam_key *)key); } @@ -3649,9 +3705,20 @@ 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) { - 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; + struct osd_object *obj = it->oi_obj; + struct dt_key *key; + + key = (struct dt_key *)iam_it_key_get(&it->oi_it); - return (struct dt_key *)iam_it_key_get(&it->oi_it); + if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) { + /* swab quota uid/gid */ + oti->oti_quota_id = le64_to_cpu(*((__u64 *)key)); + key = (struct dt_key *)&oti->oti_quota_id; + } + + return key; } /** @@ -3744,6 +3811,10 @@ static int osd_it_iam_rec(const struct lu_env *env, /* IAM does not store object type in IAM index (dir) */ osd_it_pack_dirent(lde, fid, hash, name, namelen, 0, LUDA_FID); + } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) { + iam_reccpy(&it->oi_it.ii_path.ip_leaf, + (struct iam_rec *)dtrec); + osd_quota_unpack(it->oi_obj, dtrec); } else { iam_reccpy(&it->oi_it.ii_path.ip_leaf, (struct iam_rec *)dtrec); @@ -3832,9 +3903,9 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env, it->oie_file.f_pos = 0; it->oie_file.f_dentry = obj_dentry; if (attr & LUDA_64BITHASH) - it->oie_file.f_flags = O_64BITHASH; + it->oie_file.f_mode |= FMODE_64BITHASH; else - it->oie_file.f_flags = O_32BITHASH; + it->oie_file.f_mode |= FMODE_32BITHASH; it->oie_file.f_mapping = obj->oo_inode->i_mapping; it->oie_file.f_op = obj->oo_inode->i_fop; it->oie_file.private_data = NULL; @@ -4019,7 +4090,7 @@ static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di) it->oie_it_dirent++; RETURN(0); } else { - if (it->oie_file.f_pos == LDISKFS_HTREE_EOF) + if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file)) rc = +1; else rc = osd_ldiskfs_it_fill(env, di); @@ -4098,15 +4169,11 @@ static inline int osd_it_ea_rec(const struct lu_env *env, it->oie_dirent->oied_name, it->oie_dirent->oied_namelen, it->oie_dirent->oied_type, attr); - - if (!fid_is_norm(fid)) - RETURN(0); - oic->oic_fid = *fid; if ((scrub->os_pos_current <= ino) && (sf->sf_flags & SF_INCONSISTENT || ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap))) - rc = osd_consistency_check(oti, dev, oic); + osd_consistency_check(oti, dev, oic); RETURN(rc); } @@ -4176,7 +4243,6 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt, return -EACCES; rc = osd_ea_lookup_rec(env, obj, rec, key); - if (rc == 0) rc = +1; RETURN(rc); @@ -4298,51 +4364,94 @@ static int osd_shutdown(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 lustre_mount_info *lmi; - const char *dev = lustre_cfg_string(cfg, 0); - struct lustre_disk_data *ldd; - struct lustre_sb_info *lsi; - int rc = 0; - + 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; + int rc = 0; ENTRY; + if (o->od_mnt != NULL) + RETURN(0); + o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS)); if (o->od_fsops == NULL) { CERROR("Can't find fsfilt_ldiskfs\n"); RETURN(-ENOTSUPP); } - if (o->od_mount != NULL) { - CERROR("Already mounted (%s)\n", dev); - RETURN(-EEXIST); - } + OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD); + if (__page == NULL) + GOTO(out, rc = -ENOMEM); + + str = lustre_cfg_string(cfg, 2); + s_flags = simple_strtoul(str, NULL, 0); + str = strstr(str, ":"); + if (str) + lmd_flags = simple_strtoul(str + 1, NULL, 0); + opts = lustre_cfg_string(cfg, 3); + page = (unsigned long)cfs_page_address(__page); + options = (char *)page; + *options = '\0'; + if (opts == NULL) + strcat(options, "user_xattr,acl"); + else + strcat(options, opts); - /* get mount */ - lmi = server_get_mount(dev); - if (lmi == NULL) { - CERROR("Cannot get mount info for %s!\n", dev); - RETURN(-EFAULT); - } + /* Glom up mount options */ + if (*options != '\0') + strcat(options, ","); + strlcat(options, "no_mbcache", CFS_PAGE_SIZE); - LASSERT(lmi != NULL); - /* save lustre_mount_info in dt_device */ - o->od_mount = lmi; - o->od_mnt = lmi->lmi_mnt; + type = get_fs_type("ldiskfs"); + if (!type) { + CERROR("%s: cannot find ldiskfs module\n", name); + GOTO(out, rc = -ENODEV); + } + + o->od_mnt = vfs_kern_mount(type, s_flags, dev, options); + cfs_module_put(type->owner); + + if (IS_ERR(o->od_mnt)) { + rc = PTR_ERR(o->od_mnt); + CERROR("%s: can't mount %s: %d\n", name, dev, rc); + o->od_mnt = NULL; + GOTO(out, rc); + } - lsi = s2lsi(lmi->lmi_sb); - ldd = lsi->lsi_ldd; + if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) { + CERROR("%s: underlying device %s is marked as read-only. " + "Setup failed\n", name, dev); + mntput(o->od_mnt); + o->od_mnt = NULL; + GOTO(out, rc = -EROFS); + } - if (ldd->ldd_flags & LDD_F_IAM_DIR) { + if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb, + LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) { + CERROR("%s: device %s is mounted w/o journal\n", name, dev); + mntput(o->od_mnt); + o->od_mnt = NULL; + GOTO(out, rc = -EINVAL); + } + + if (lmd_flags & LMD_FLG_IAM) { o->od_iop_mode = 0; - LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", dev); + LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", name); } else o->od_iop_mode = 1; + if (lmd_flags & LMD_FLG_NOSCRUB) + o->od_noscrub = 1; - if (ldd->ldd_flags & LDD_F_SV_TYPE_OST) { - rc = osd_compat_init(o); - if (rc) - CERROR("%s: can't initialize compats: %d\n", dev, rc); - } +out: + if (__page) + OBD_PAGE_FREE(__page); + if (rc) + fsfilt_put_ops(o->od_fsops); RETURN(rc); } @@ -4353,6 +4462,8 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, int rc; ENTRY; + rc = osd_shutdown(env, osd_dev(d)); + osd_compat_fini(osd_dev(d)); shrink_dcache_sb(osd_sb(osd_dev(d))); @@ -4364,46 +4475,129 @@ static struct lu_device *osd_device_fini(const struct lu_env *env, RETURN (ERR_PTR(rc)); } - if (osd_dev(d)->od_mount) - server_put_mount(osd_dev(d)->od_mount->lmi_name, - osd_dev(d)->od_mount->lmi_mnt); - osd_dev(d)->od_mount = NULL; + if (osd_dev(d)->od_mnt) { + mntput(osd_dev(d)->od_mnt); + osd_dev(d)->od_mnt = NULL; + } RETURN(NULL); } +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 osd_thread_info *info; + int rc; + + /* if the module was re-loaded, env can loose its keys */ + rc = lu_env_refill((struct lu_env *) env); + if (rc) + GOTO(out, rc); + info = osd_oti_get(env); + LASSERT(info); + + l->ld_ops = &osd_lu_ops; + o->od_dt_dev.dd_ops = &osd_dt_ops; + + cfs_spin_lock_init(&o->od_osfs_lock); + cfs_mutex_init(&o->od_otable_mutex); + o->od_osfs_age = cfs_time_shift_64(-1000); + + o->od_capa_hash = init_capa_hash(); + if (o->od_capa_hash == NULL) + GOTO(out, rc = -ENOMEM); + + o->od_read_cache = 1; + o->od_writethrough_cache = 1; + o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE; + + rc = osd_mount(env, o, cfg); + if (rc) + GOTO(out_capa, rc); + + /* setup scrub, including OI files initialization */ + rc = osd_scrub_setup(env, o); + if (rc < 0) + GOTO(out_mnt, rc); + + strncpy(o->od_svname, lustre_cfg_string(cfg, 4), + sizeof(o->od_svname) - 1); + + rc = osd_compat_init(o); + if (rc != 0) + GOTO(out_scrub, rc); + + rc = lu_site_init(&o->od_site, l); + if (rc) + GOTO(out_compat, rc); + o->od_site.ls_bottom_dev = l; + + rc = lu_site_init_finish(&o->od_site); + if (rc) + GOTO(out_site, rc); + + rc = osd_procfs_init(o, o->od_svname); + if (rc != 0) { + CERROR("%s: can't initialize procfs: rc = %d\n", + o->od_svname, rc); + GOTO(out_site, rc); + } + + LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev); + + /* 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; + GOTO(out_procfs, rc); + } + + RETURN(0); +out_procfs: + osd_procfs_fini(o); +out_site: + lu_site_fini(&o->od_site); +out_compat: + osd_compat_fini(o); +out_scrub: + osd_scrub_cleanup(env, o); +out_mnt: + osd_oi_fini(info, o); + osd_shutdown(env, o); + mntput(o->od_mnt); + o->od_mnt = NULL; +out_capa: + cleanup_capa_hash(o->od_capa_hash); +out: + RETURN(rc); +} + static struct lu_device *osd_device_alloc(const struct lu_env *env, struct lu_device_type *t, struct lustre_cfg *cfg) { - struct lu_device *l; - struct osd_device *o; - - OBD_ALLOC_PTR(o); - if (o != NULL) { - int result; - - result = dt_device_init(&o->od_dt_dev, t); - if (result == 0) { - l = osd2lu_dev(o); - l->ld_ops = &osd_lu_ops; - o->od_dt_dev.dd_ops = &osd_dt_ops; - cfs_spin_lock_init(&o->od_osfs_lock); - cfs_mutex_init(&o->od_otable_mutex); - o->od_osfs_age = cfs_time_shift_64(-1000); - o->od_capa_hash = init_capa_hash(); - if (o->od_capa_hash == NULL) { - dt_device_fini(&o->od_dt_dev); - l = ERR_PTR(-ENOMEM); - } - } else - l = ERR_PTR(result); + struct osd_device *o; + int rc; - if (IS_ERR(l)) - OBD_FREE_PTR(o); - } else - l = ERR_PTR(-ENOMEM); - return l; + OBD_ALLOC_PTR(o); + if (o == NULL) + return ERR_PTR(-ENOMEM); + + rc = dt_device_init(&o->od_dt_dev, t); + if (rc == 0) { + rc = osd_device_init0(env, o, cfg); + if (rc) + dt_device_fini(&o->od_dt_dev); + } + + if (unlikely(rc != 0)) + OBD_FREE_PTR(o); + + return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc); } static struct lu_device *osd_device_free(const struct lu_env *env, @@ -4413,6 +4607,14 @@ static struct lu_device *osd_device_free(const struct lu_env *env, ENTRY; cleanup_capa_hash(o->od_capa_hash); + /* XXX: make osd top device in order to release reference */ + d->ld_site->ls_top_dev = d; + lu_site_purge(env, d->ld_site, -1); + 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_fini(&o->od_site); dt_device_fini(&o->od_dt_dev); OBD_FREE_PTR(o); RETURN(NULL); @@ -4430,8 +4632,9 @@ static int osd_process_config(const struct lu_env *env, err = osd_mount(env, o, cfg); break; case LCFG_CLEANUP: - err = osd_shutdown(env, o); - break; + lu_dev_del_linkage(d->ld_site, d); + err = osd_shutdown(env, o); + break; default: err = -ENOSYS; } @@ -4442,36 +4645,92 @@ static int osd_process_config(const struct lu_env *env, static int osd_recovery_complete(const struct lu_env *env, struct lu_device *d) { - RETURN(0); + struct osd_device *osd = osd_dev(d); + int rc = 0; + ENTRY; + + if (osd->od_quota_slave == 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); + RETURN(rc); +} + +/* + * we use exports to track all osd users + */ +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; + ENTRY; + + CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects); + + rc = class_connect(&conn, obd, cluuid); + if (rc) + RETURN(rc); + + *exp = class_conn2export(&conn); + + cfs_spin_lock(&osd->od_osfs_lock); + osd->od_connects++; + cfs_spin_unlock(&osd->od_osfs_lock); + + RETURN(0); +} + +/* + * once last export (we don't count self-export) disappeared + * osd can be released + */ +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; + ENTRY; + + /* Only disconnect the underlying layers on the final disconnect. */ + cfs_spin_lock(&osd->od_osfs_lock); + osd->od_connects--; + if (osd->od_connects == 0) + release = 1; + cfs_spin_unlock(&osd->od_osfs_lock); + + rc = class_disconnect(exp); /* bz 9811 */ + + if (rc == 0 && release) + class_manual_cleanup(obd); + RETURN(rc); } static int osd_prepare(const struct lu_env *env, struct lu_device *pdev, struct lu_device *dev) { struct osd_device *osd = osd_dev(dev); - int result; + int result = 0; ENTRY; - /* 1. setup scrub, including OI files initialization */ - result = osd_scrub_setup(env, osd); - if (result < 0) - RETURN(result); - - /* 2. setup quota slave instance */ - osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev, - osd->od_proc_entry); - if (IS_ERR(osd->od_quota_slave)) { - result = PTR_ERR(osd->od_quota_slave); - osd->od_quota_slave = NULL; - RETURN(result); + if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) { + /* MDT/MDD still use old infrastructure to create + * special files */ + result = llo_local_objects_setup(env, lu2md_dev(pdev), + lu2dt_dev(dev)); + if (result) + RETURN(result); } - if (!lu_device_is_md(pdev)) - RETURN(0); + if (osd->od_quota_slave != NULL) + /* set up quota slave objects */ + result = qsd_prepare(env, osd->od_quota_slave); - /* 3. setup local objects */ - result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev)); - RETURN(result); + RETURN(result); } static const struct lu_object_operations osd_lu_obj_ops = { @@ -4504,9 +4763,9 @@ static const struct lu_device_type_operations osd_device_type_ops = { .ldto_device_fini = osd_device_fini }; -static struct lu_device_type osd_device_type = { +struct lu_device_type osd_device_type = { .ldt_tags = LU_DEVICE_DT, - .ldt_name = LUSTRE_OSD_NAME, + .ldt_name = LUSTRE_OSD_LDISKFS_NAME, .ldt_ops = &osd_device_type_ops, .ldt_ctx_tags = LCT_LOCAL, }; @@ -4515,7 +4774,9 @@ static struct lu_device_type osd_device_type = { * lprocfs legacy support. */ static struct obd_ops osd_obd_device_ops = { - .o_owner = THIS_MODULE + .o_owner = THIS_MODULE, + .o_connect = osd_obd_connect, + .o_disconnect = osd_obd_disconnect }; static int __init osd_mod_init(void) @@ -4525,16 +4786,16 @@ static int __init osd_mod_init(void) osd_oi_mod_init(); lprocfs_osd_init_vars(&lvars); return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars, - LUSTRE_OSD_NAME, &osd_device_type); + LUSTRE_OSD_LDISKFS_NAME, &osd_device_type); } static void __exit osd_mod_exit(void) { - class_unregister_type(LUSTRE_OSD_NAME); + class_unregister_type(LUSTRE_OSD_LDISKFS_NAME); } MODULE_AUTHOR("Sun Microsystems, Inc. "); -MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")"); +MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")"); MODULE_LICENSE("GPL"); cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);