X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_scrub.c;h=628771a0e8bfb714693977d62bbba9c9c376a889;hp=184c83a2160fc2efcf7733b10c26906bfc9fad9b;hb=1cab7fafd016621160f49030ea46f155fe34d70b;hpb=c769ba3025c59f110c38c44552a05142705c94c3

diff --git a/lustre/osd-ldiskfs/osd_scrub.c b/lustre/osd-ldiskfs/osd_scrub.c
index 184c83a..628771a 100644
--- a/lustre/osd-ldiskfs/osd_scrub.c
+++ b/lustre/osd-ldiskfs/osd_scrub.c
@@ -783,16 +783,12 @@ static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 	struct osd_inode_id *lid;
 	int rc;
 
-	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
-		struct l_wait_info lwi;
-
-		lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
-		if (likely(lwi.lwi_timeout > 0))
-			l_wait_event(thread->t_ctl_waitq,
-				!list_empty(&scrub->os_inconsistent_items) ||
-				!thread_is_running(thread),
-				&lwi);
-	}
+	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0)
+		wait_event_idle_timeout(
+			thread->t_ctl_waitq,
+			!list_empty(&scrub->os_inconsistent_items) ||
+			!thread_is_running(thread),
+			cfs_time_seconds(cfs_fail_val));
 
 	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 		spin_lock(&scrub->os_lock);
@@ -882,7 +878,6 @@ static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 			  struct osd_iit_param *param,
 			  struct osd_idmap_cache *oic, bool *noslot, int rc)
 {
-	struct l_wait_info lwi = { 0 };
 	struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 	struct scrub_file *sf = &scrub->os_file;
 	struct ptlrpc_thread *thread = &scrub->os_thread;
@@ -937,8 +932,8 @@ wait:
 	}
 
 	if (it != NULL)
-		l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
-			     &lwi);
+		wait_event_idle(thread->t_ctl_waitq,
+				osd_scrub_wakeup(scrub, it));
 
 	if (!ooc || osd_scrub_has_window(scrub, ooc))
 		*noslot = false;
@@ -1042,7 +1037,6 @@ static int osd_inode_iteration(struct osd_thread_info *info,
 	__u64 *pos;
 	__u64 *count;
 	struct osd_iit_param *param;
-	struct l_wait_info lwi = { 0 };
 	__u32 limit;
 	int rc;
 	bool noslot = true;
@@ -1105,12 +1099,12 @@ wait:
 			sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 					  SF_UPGRADE | SF_AUTO);
 			sf->sf_status = SS_COMPLETED;
-			l_wait_event(thread->t_ctl_waitq,
-				     !thread_is_running(thread) ||
-				     !scrub->os_partial_scan ||
-				     scrub->os_in_join ||
-				     !list_empty(&scrub->os_inconsistent_items),
-				     &lwi);
+			wait_event_idle(
+				thread->t_ctl_waitq,
+				!thread_is_running(thread) ||
+				!scrub->os_partial_scan ||
+				scrub->os_in_join ||
+				!list_empty(&scrub->os_inconsistent_items));
 			sf->sf_flags = saved_flags;
 			sf->sf_status = SS_SCANNING;
 
@@ -1132,9 +1126,9 @@ wait:
 
 full:
 	if (!preload) {
-		l_wait_event(thread->t_ctl_waitq,
-			     !thread_is_running(thread) || !scrub->os_in_join,
-			     &lwi);
+		wait_event_idle(thread->t_ctl_waitq,
+				!thread_is_running(thread) ||
+				!scrub->os_in_join);
 
 		if (unlikely(!thread_is_running(thread)))
 			RETURN(0);
@@ -1288,13 +1282,12 @@ static int osd_scrub_main(void *args)
 	}
 
 	if (!scrub->os_full_speed && !scrub->os_partial_scan) {
-		struct l_wait_info lwi = { 0 };
 		struct osd_otable_it *it = dev->od_otable_it;
 		struct osd_otable_cache *ooc = &it->ooi_cache;
 
-		l_wait_event(thread->t_ctl_waitq,
-			     it->ooi_user_ready || !thread_is_running(thread),
-			     &lwi);
+		wait_event_idle(thread->t_ctl_waitq,
+				it->ooi_user_ready ||
+				!thread_is_running(thread));
 		if (unlikely(!thread_is_running(thread)))
 			GOTO(post, rc = 0);
 
@@ -1731,10 +1724,8 @@ struct osd_ios_item {
 };
 
 struct osd_ios_filldir_buf {
-#ifdef HAVE_DIR_CONTEXT
 	/* please keep it as first member */
 	struct dir_context	 ctx;
-#endif
 	struct osd_thread_info	*oifb_info;
 	struct osd_device	*oifb_dev;
 	struct dentry		*oifb_dentry;
@@ -1959,7 +1950,8 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
 					       inode);
 	}
 
-	rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
+	/* Since this called from iterate_dir() the inode lock will be taken */
+	rc = osd_oi_lookup(info, dev, &tfid, id2, OI_LOCKED);
 	if (rc != 0) {
 		if (rc != -ENOENT)
 			RETURN(rc);
@@ -2022,16 +2014,17 @@ static int osd_ios_lf_fill(void *buf,
 		RETURN(0);
 
 	scrub->os_lf_scanned++;
-	child = osd_ios_lookup_one_len(name, parent, namelen);
+	child = osd_lookup_one_len(dev, name, parent, namelen);
 	if (IS_ERR(child)) {
+		rc = PTR_ERR(child);
 		CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
-		      osd_name(dev), namelen, name, (int)PTR_ERR(child));
-		RETURN(0);
+		      osd_name(dev), namelen, name, rc);
+		RETURN(rc);
 	} else if (!child->d_inode) {
 		dput(child);
 		CDEBUG(D_INODE, "%s: child '%.*s' lacks inode\n",
 		       osd_name(dev), namelen, name);
-		RETURN(0);
+		RETURN(-ENOENT);
 	}
 
 	inode = child->d_inode;
@@ -2096,7 +2089,7 @@ static int osd_ios_varfid_fill(void *buf,
 	if (name[0] == '.')
 		RETURN(0);
 
-	child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
+	child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
 	if (IS_ERR(child))
 		RETURN(PTR_ERR(child));
 
@@ -2144,7 +2137,7 @@ static int osd_ios_dl_fill(void *buf,
 	if (map->olm_name == NULL)
 		RETURN(0);
 
-	child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
+	child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
 	if (IS_ERR(child))
 		RETURN(PTR_ERR(child));
 
@@ -2166,6 +2159,7 @@ static int osd_ios_uld_fill(void *buf,
 {
 	struct osd_ios_filldir_buf *fill_buf =
 		(struct osd_ios_filldir_buf *)buf;
+	struct osd_device *dev = fill_buf->oifb_dev;
 	struct dentry		   *child;
 	struct lu_fid		    tfid;
 	int			    rc       = 0;
@@ -2177,7 +2171,7 @@ static int osd_ios_uld_fill(void *buf,
 	if (name[0] != '[')
 		RETURN(0);
 
-	child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
+	child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
 	if (IS_ERR(child))
 		RETURN(PTR_ERR(child));
 
@@ -2227,7 +2221,7 @@ static int osd_ios_root_fill(void *buf,
 	if (map->olm_name == NULL)
 		RETURN(0);
 
-	child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
+	child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
 	if (IS_ERR(child))
 		RETURN(PTR_ERR(child));
 	else if (!child->d_inode)
@@ -2251,9 +2245,7 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
 		     struct dentry *dentry, filldir_t filldir)
 {
 	struct osd_ios_filldir_buf    buf   = {
-#ifdef HAVE_DIR_CONTEXT
 						.ctx.actor = filldir,
-#endif
 						.oifb_info = info,
 						.oifb_dev = dev,
 						.oifb_dentry = dentry };
@@ -2267,25 +2259,20 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
 
 	filp->f_pos = 0;
 	filp->f_path.dentry = dentry;
-	filp->f_mode = FMODE_64BITHASH;
+	filp->f_flags |= O_NOATIME;
+	filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
 	filp->f_mapping = inode->i_mapping;
 	filp->f_op = fops;
 	filp->private_data = NULL;
+	filp->f_cred = current_cred();
 	set_file_inode(filp, inode);
+	rc = osd_security_file_alloc(filp);
+	if (rc)
+		RETURN(rc);
 
 	do {
 		buf.oifb_items = 0;
-#ifdef HAVE_DIR_CONTEXT
-		buf.ctx.pos = filp->f_pos;
-#ifdef HAVE_ITERATE_SHARED
-		rc = fops->iterate_shared(filp, &buf.ctx);
-#else
-		rc = fops->iterate(filp, &buf.ctx);
-#endif
-		filp->f_pos = buf.ctx.pos;
-#else
-		rc = fops->readdir(filp, &buf, filldir);
-#endif
+		rc = iterate_dir(filp, &buf.ctx);
 	} while (rc >= 0 && buf.oifb_items > 0 &&
 		 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
 	fops->release(inode, filp);
@@ -2317,8 +2304,8 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
 	 *	OI mapping crashed or lost also, then we have to give up under
 	 *	double failure cases. */
 	scrub->os_convert_igif = 1;
-	child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
-				       strlen(dot_lustre_name));
+	child = osd_lookup_one_len_unlocked(dev, dot_lustre_name, dentry,
+					    strlen(dot_lustre_name));
 	if (IS_ERR(child)) {
 		if (PTR_ERR(child) != -ENOENT)
 			RETURN(PTR_ERR(child));
@@ -2349,10 +2336,12 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
 	 * to access the ".lustre" with cached IGIF. So we prefer
 	 * to the solution 2).
 	 */
+	inode_lock(dentry->d_inode);
 	rc = osd_ios_scan_one(info, dev, dentry->d_inode,
 			      child->d_inode, &LU_DOT_LUSTRE_FID,
 			      dot_lustre_name,
 			      strlen(dot_lustre_name), 0);
+	inode_unlock(dentry->d_inode);
 	if (rc == -ENOENT) {
 out_scrub:
 		/* It is 1.8 MDT device. */
@@ -2390,26 +2379,32 @@ osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
 			RETURN(rc);
 	}
 
-	child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
+	child = osd_lookup_one_len_unlocked(dev, ADMIN_USR, dentry,
+					    strlen(ADMIN_USR));
 	if (IS_ERR(child)) {
 		rc = PTR_ERR(child);
 	} else {
+		inode_lock(dentry->d_inode);
 		rc = osd_ios_scan_one(info, dev, dentry->d_inode,
 				      child->d_inode, NULL, ADMIN_USR,
 				      strlen(ADMIN_USR), 0);
+		inode_unlock(dentry->d_inode);
 		dput(child);
 	}
 
 	if (rc != 0 && rc != -ENOENT)
 		GOTO(out, rc);
 
-	child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
+	child = osd_lookup_one_len_unlocked(dev, ADMIN_GRP, dentry,
+					    strlen(ADMIN_GRP));
 	if (IS_ERR(child))
 		GOTO(out, rc = PTR_ERR(child));
 
+	inode_lock(dentry->d_inode);
 	rc = osd_ios_scan_one(info, dev, dentry->d_inode,
 			      child->d_inode, NULL, ADMIN_GRP,
 			      strlen(ADMIN_GRP), 0);
+	inode_unlock(dentry->d_inode);
 	dput(child);
 out:
 	RETURN(rc == -ENOENT ? 0 : rc);
@@ -2429,6 +2424,12 @@ static void osd_initial_OI_scrub(struct osd_thread_info *info,
 	dev->od_igif_inoi = 1;
 
 	while (1) {
+		/* Don't take inode_lock here since scandir() callbacks
+		 * can call VFS functions which may manully take the
+		 * inode lock itself like iterate_dir(). Since this
+		 * is the case it is best to leave the scandir()
+		 * callbacks to managing the inode lock.
+		 */
 		scandir(info, dev, dentry, filldir);
 		if (item != NULL) {
 			dput(item->oii_dentry);
@@ -2459,9 +2460,9 @@ static void osd_initial_OI_scrub(struct osd_thread_info *info,
 			continue;
 		}
 
-		child = osd_ios_lookup_one_len(map->olm_name,
-					       osd_sb(dev)->s_root,
-					       map->olm_namelen);
+		child = osd_lookup_one_len_unlocked(dev, map->olm_name,
+						    osd_sb(dev)->s_root,
+						    map->olm_namelen);
 		if (PTR_ERR(child) == -ENOENT ||
 		    (!IS_ERR(child) && !child->d_inode))
 			osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
@@ -2868,7 +2869,6 @@ static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
 	struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 	struct osd_otable_cache *ooc    = &it->ooi_cache;
 	struct ptlrpc_thread    *thread = &scrub->os_thread;
-	struct l_wait_info       lwi    = { 0 };
 	int			 rc;
 	ENTRY;
 
@@ -2886,9 +2886,8 @@ again:
 	}
 
 	if (it->ooi_all_cached) {
-		l_wait_event(thread->t_ctl_waitq,
-			     !thread_is_running(thread),
-			     &lwi);
+		wait_event_idle(thread->t_ctl_waitq,
+				!thread_is_running(thread));
 		RETURN(1);
 	}
 
@@ -2900,9 +2899,8 @@ again:
 	}
 
 	if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
-		l_wait_event(thread->t_ctl_waitq,
-			     osd_otable_it_wakeup(scrub, it),
-			     &lwi);
+		wait_event_idle(thread->t_ctl_waitq,
+				osd_otable_it_wakeup(scrub, it));
 
 	if (!thread_is_running(thread) && !it->ooi_used_outside)
 		RETURN(1);