X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdfilter%2Ffilter.c;h=530762dd8b6b72dc1d987730514efa9a2e6d4ae5;hp=1c485bbae2c1d042acc5bf0785e9a13406a207b6;hb=d750891e478804bc495ffa075d771d1816369958;hpb=d22222efddb2e2611b72d1e695e91bfd7215bf6d diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c index 1c485bb..530762d 100644 --- a/lustre/obdfilter/filter.c +++ b/lustre/obdfilter/filter.c @@ -290,7 +290,6 @@ static int filter_export_stats_init(struct obd_device *obd, RETURN(0); clean: - lprocfs_exp_cleanup(exp); return rc; } @@ -1196,7 +1195,7 @@ static int filter_prep_groups(struct obd_device *obd) loff_t off = 0; ENTRY; - O_dentry = simple_mkdir(current->fs->pwd, obd->u.obt.obt_vfsmnt, + O_dentry = simple_mkdir(cfs_fs_pwd(current->fs), obd->u.obt.obt_vfsmnt, "O", 0700, 1); CDEBUG(D_INODE, "got/created O: %p\n", O_dentry); if (IS_ERR(O_dentry)) { @@ -1562,7 +1561,7 @@ int filter_vfs_unlink(struct inode *dir, struct dentry *dentry, GOTO(out, rc = -EPERM); /* check_sticky() */ - if ((dentry->d_inode->i_uid != current->fsuid && + if ((dentry->d_inode->i_uid != cfs_curproc_fsuid() && !cfs_capable(CFS_CAP_FOWNER)) || IS_APPEND(dentry->d_inode) || IS_IMMUTABLE(dentry->d_inode)) GOTO(out, rc = -EPERM); @@ -1570,7 +1569,7 @@ int filter_vfs_unlink(struct inode *dir, struct dentry *dentry, /* NOTE: This might need to go outside i_mutex, though it isn't clear if * that was done because of journal_start (which is already done * here) or some other ordering issue. */ - DQUOT_INIT(dir); + ll_vfs_dq_init(dir); rc = ll_security_inode_unlink(dir, dentry, mnt); if (rc) @@ -1596,7 +1595,10 @@ static int filter_destroy_internal(struct obd_device *obd, obd_id objid, struct inode *inode = dchild->d_inode; int rc; - if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 1) { + /* There should be 2 references to the inode: + * 1) taken by filter_prepare_destroy + * 2) taken by filter_destroy */ + if (inode->i_nlink != 1 || atomic_read(&inode->i_count) != 2) { CERROR("destroying objid %.*s ino %lu nlink %lu count %d\n", dchild->d_name.len, dchild->d_name.name, inode->i_ino, (unsigned long)inode->i_nlink, @@ -1700,7 +1702,7 @@ static int filter_intent_policy(struct ldlm_namespace *ns, * lock, and should not be granted if the lock will be blocked. */ - LASSERT(ns == res->lr_namespace); + LASSERT(ns == ldlm_res_to_ns(res)); lock_res(res); rc = policy(lock, &tmpflags, 0, &err, &rpc_list); check_res_locked(res); @@ -1722,7 +1724,7 @@ static int filter_intent_policy(struct ldlm_namespace *ns, if (rc == LDLM_ITER_CONTINUE) { /* do not grant locks to the liblustre clients: they cannot * handle ASTs robustly. We need to do this while still - * holding ns_lock to avoid the lock remaining on the res_link + * holding lr_lock to avoid the lock remaining on the res_link * list (and potentially being added to l_pending_list by an * AST) when we are going to drop this lock ASAP. */ if (lock->l_export->exp_libclient || @@ -1745,7 +1747,7 @@ static int filter_intent_policy(struct ldlm_namespace *ns, *reply_lvb = *res_lvb; /* - * ->ns_lock guarantees that no new locks are granted, and, + * lr_lock guarantees that no new locks are granted, and, * therefore, that res->lr_lvb_data cannot increase beyond the * end of already granted lock. As a result, it is safe to * check against "stale" reply_lvb->lvb_size value without @@ -1798,13 +1800,6 @@ static int filter_intent_policy(struct ldlm_namespace *ns, LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l); rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */ - /* Update the LVB from disk if the AST failed (this is a legal race) */ - /* - * XXX nikita: situation when ldlm_server_glimpse_ast() failed before - * sending ast is not handled. This can result in lost client writes. - */ - if (rc != 0) - ldlm_res_lvbo_update(res, NULL, 1); lock_res(res); *reply_lvb = *res_lvb; @@ -1953,7 +1948,7 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, __u8 *uuid_ptr; char *str, *label; char ns_name[48]; - request_queue_t *q; + struct request_queue *q; int rc, i; ENTRY; @@ -2064,8 +2059,10 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, GOTO(err_post, rc = -ENOMEM); sprintf(ns_name, "filter-%s", obd->obd_uuid.uuid); - obd->obd_namespace = ldlm_namespace_new(obd, ns_name, LDLM_NAMESPACE_SERVER, - LDLM_NAMESPACE_GREEDY); + obd->obd_namespace = ldlm_namespace_new(obd, ns_name, + LDLM_NAMESPACE_SERVER, + LDLM_NAMESPACE_GREEDY, + LDLM_NS_TYPE_OST); if (obd->obd_namespace == NULL) GOTO(err_post, rc = -ENOMEM); obd->obd_namespace->ns_lvbp = obd; @@ -2091,13 +2088,13 @@ int filter_common_setup(struct obd_device *obd, struct lustre_cfg* lcfg, GOTO(err_post, rc); q = bdev_get_queue(mnt->mnt_sb->s_bdev); - if (q->max_sectors < q->max_hw_sectors && - q->max_sectors < PTLRPC_MAX_BRW_SIZE >> 9) + if (queue_max_sectors(q) < queue_max_hw_sectors(q) && + queue_max_sectors(q) < PTLRPC_MAX_BRW_SIZE >> 9) LCONSOLE_INFO("%s: underlying device %s should be tuned " "for larger I/O requests: max_sectors = %u " "could be up to max_hw_sectors=%u\n", obd->obd_name, mnt->mnt_sb->s_id, - q->max_sectors, q->max_hw_sectors); + queue_max_sectors(q), queue_max_hw_sectors(q)); uuid_ptr = fsfilt_uuid(obd, obd->u.obt.obt_sb); if (uuid_ptr != NULL) { @@ -2605,7 +2602,7 @@ static int filter_cleanup(struct obd_device *obd) filter_post(obd); - LL_DQUOT_OFF(obd->u.obt.obt_sb); + ll_vfs_dq_off(obd->u.obt.obt_sb, 0); shrink_dcache_sb(obd->u.obt.obt_sb); server_put_mount(obd->obd_name, obd->u.obt.obt_vfsmnt); @@ -2808,7 +2805,6 @@ static int filter_connect(const struct lu_env *env, cleanup: if (rc) { class_disconnect(lexp); - lprocfs_exp_cleanup(lexp); *exp = NULL; } else { *exp = lexp; @@ -3192,13 +3188,15 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, *fcc = oa->o_lcookie; } if (ia_valid & (ATTR_SIZE | ATTR_UID | ATTR_GID)) { - DQUOT_INIT(inode); + unsigned long now = jiffies; + ll_vfs_dq_init(inode); /* Filter truncates and writes are serialized by * i_alloc_sem, see the comment in * filter_preprw_write.*/ if (ia_valid & ATTR_SIZE) down_write(&inode->i_alloc_sem); LOCK_INODE_MUTEX(inode); + fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex"); old_size = i_size_read(inode); } @@ -3280,7 +3278,10 @@ int filter_setattr_internal(struct obd_export *exp, struct dentry *dentry, * we have two left for the last_rcvd and VBR inode version updates. */ err = fsfilt_extend(exp->exp_obd, inode, 2, handle); - rc = filter_finish_transno(exp, inode, oti, rc, sync); + /* Update inode version only if data has changed => size has changed */ + rc = filter_finish_transno(exp, ia_valid & ATTR_SIZE ? inode : NULL, + oti, rc, sync); + if (sync) { filter_cancel_cookies_cb(exp->exp_obd, 0, fcc, rc); fcc = NULL; @@ -3395,7 +3396,9 @@ int filter_setattr(struct obd_export *exp, struct obd_info *oinfo, */ if (oa->o_valid & (OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME)) { + unsigned long now = jiffies; down_write(&dentry->d_inode->i_alloc_sem); + fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem"); fmd = filter_fmd_get(exp, oa->o_id, oa->o_seq); if (fmd && fmd->fmd_mactime_xid < oti->oti_xid) fmd->fmd_mactime_xid = oti->oti_xid; @@ -3809,10 +3812,15 @@ static int filter_precreate(struct obd_device *obd, struct obdo *oa, } else next_id = filter_last_id(filter, group) + 1; - /* Temporary solution for oid in CMD before fid-on-OST */ - if ((fid_seq_is_mdt0(oa->o_seq) && next_id >= IDIF_MAX_OID) && - (fid_seq_is_cmd(oa->o_seq) && next_id >= OBIF_MAX_OID)) { - CERROR("%s:"POSTID" hit the max IDIF_MAX_OID(1<<48)!\n", + /* Don't create objects beyond the valid range for this SEQ */ + if (unlikely(fid_seq_is_mdt0(group) && + next_id >= IDIF_MAX_OID)) { + CERROR("%s:"POSTID" hit the IDIF_MAX_OID (1<<48)!\n", + obd->obd_name, next_id, group); + GOTO(cleanup, rc = -ENOSPC); + } else if (unlikely(!fid_seq_is_mdt0(group) && + next_id >= OBIF_MAX_OID)) { + CERROR("%s:"POSTID" hit the OBIF_MAX_OID (1<<32)!\n", obd->obd_name, next_id, group); GOTO(cleanup, rc = -ENOSPC); } @@ -4015,6 +4023,7 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, struct llog_cookie *fcc = NULL; int rc, rc2, cleanup_phase = 0, sync = 0; struct iattr iattr; + unsigned long now; ENTRY; rc = filter_auth_capa(exp, NULL, oa->o_seq, @@ -4069,7 +4078,7 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, if (fcc != NULL) *fcc = oa->o_lcookie; } - DQUOT_INIT(dchild->d_inode); + ll_vfs_dq_init(dchild->d_inode); /* we're gonna truncate it first in order to avoid possible deadlock: * P1 P2 @@ -4083,8 +4092,10 @@ int filter_destroy(struct obd_export *exp, struct obdo *oa, * between page lock, i_mutex & starting new journal handle. * (see bug 20321) -johann */ + now = jiffies; down_write(&dchild->d_inode->i_alloc_sem); LOCK_INODE_MUTEX(dchild->d_inode); + fsfilt_check_slow(exp->exp_obd, now, "i_alloc_sem and i_mutex"); /* VBR: version recovery check */ rc = filter_version_get_check(exp, oti, dchild->d_inode);