Merge branch 'linux-user-disk' into refs/top-bases/linux-zfs-branch

cmd/zpool/zpool_vdev.c

@@ -279,7 +279,7 @@ check_disk(const char *path, blkid_cache cache, int force,
 	 * label at the end of the device is intact.  Rather than use this
 	 * label we should play it safe and treat this as a non efi device.
 	 */
-	if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
+	if (!force && vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 		vdev_error(gettext(
 		    "%s contains a corrupt primary efi partition table.  "
 		    "If you are\nsure you want to use this device use "
@@ -938,12 +938,7 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 	char *type, *path, *diskname;
 	char buf[MAXPATHLEN];
 	uint64_t wholedisk;
-	int fd;
 	int ret;
-#if defined(__sun__) || defined(__sun)
-	ddi_devid_t devid;
-	char *minor = NULL, *devid_str = NULL;
-#endif
 
 	verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 
@@ -983,50 +978,34 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 			return (-1);
 
 		/*
-		 * Fill in the devid, now that we've labeled the disk.  We
+		 * Now the we've labeled the disk and the partitions have
-		 * attempt to open the new zfs slice first by appending the
+		 * been created.  We still need to wait for udev to create
-		 * slice number.  If that fails this may be a Linux udev
+		 * the symlinks to those partitions.  If we are accessing
-		 * path in which case the -part# convention is tried.
+		 * the devices via a udev disk path, /dev/disk, then wait
+		 * for *-part# to be created.  Otherwise just use the normal
+		 * syntax for devices in /dev.
 		 */
-		(void) snprintf(buf, sizeof (buf), "%s%s", path, FIRST_SLICE);
+		if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
-		if ((fd = open(buf, O_RDONLY)) < 0) {
+			(void) snprintf(buf, sizeof (buf),
+			    "%s%s%s", path, "-part", FIRST_SLICE);
+		else
+			(void) snprintf(buf, sizeof (buf),
+			    "%s%s", path, FIRST_SLICE);
 
-			(void) snprintf(buf, sizeof (buf), "%s%s%s",
+		if ((ret = zpool_label_disk_wait(buf, 1000)) != 0) {
-					path, "-part", FIRST_SLICE);
+			(void) fprintf(stderr,
-			if ((fd = open(buf, O_RDONLY)) < 0) {
+			    gettext( "cannot resolve path '%s'\n"), buf);
-				(void) fprintf(stderr,
+			return (-1);
-				    gettext("cannot open '%s': %s\n"),
-				    buf, strerror(errno));
-				return (-1);
-			}
 		}
 
-#if defined(__sun__) || defined(__sun)
-		if (devid_get(fd, &devid) == 0) {
-			if (devid_get_minor_name(fd, &minor) == 0 &&
-			    (devid_str = devid_str_encode(devid, minor)) !=
-			    NULL) {
-				verify(nvlist_add_string(nv,
-				    ZPOOL_CONFIG_DEVID, devid_str) == 0);
-			}
-			if (devid_str != NULL)
-				devid_str_free(devid_str);
-			if (minor != NULL)
-				devid_str_free(minor);
-			devid_free(devid);
-		}
-#endif
-
 		/*
-		 * Update the path to refer to the 's0' slice.  The presence of
+		 * Update the path to refer to FIRST_SLICE.  The presence of
 		 * the 'whole_disk' field indicates to the CLI that we should
 		 * chop off the slice number when displaying the device in
 		 * future output.
 		 */
 		verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
 
-		(void) close(fd);
-
 		/* Just in case this partition already existed. */
 		(void) zero_label(buf);
 

lib/libzfs/include/libzfs.h

@@ -56,6 +56,7 @@ extern "C" {
 #if defined(__sun__) || defined(__sun)
 #define	DISK_ROOT	"/dev/dsk"
 #define	RDISK_ROOT	"/dev/rdsk"
+#define	UDISK_ROOT	RDISK_ROOT
 #define	FIRST_SLICE	"s0"
 #define	BACKUP_SLICE	"s2"
 #endif
@@ -63,6 +64,7 @@ extern "C" {
 #ifdef __linux__
 #define	DISK_ROOT	"/dev"
 #define	RDISK_ROOT	DISK_ROOT
+#define	UDISK_ROOT	"/dev/disk"
 #define	FIRST_SLICE	"1"
 #define	BACKUP_SLICE	""
 #endif
@@ -253,6 +255,7 @@ extern nvlist_t *zpool_find_vdev(zpool_handle_t *, const char *, boolean_t *,
     boolean_t *, boolean_t *);
 extern nvlist_t *zpool_find_vdev_by_physpath(zpool_handle_t *, const char *,
     boolean_t *, boolean_t *, boolean_t *);
+extern int zpool_label_disk_wait(char *, int);
 extern int zpool_label_disk(libzfs_handle_t *, zpool_handle_t *, char *);
 
 /*

lib/libzfs/libzfs_pool.c

@@ -3118,6 +3118,33 @@ find_start_block(nvlist_t *config)
 	return (MAXOFFSET_T);
 }
 
+int
+zpool_label_disk_wait(char *path, int timeout)
+{
+#if defined(__linux__)
+	struct stat64 statbuf;
+	int i;
+
+	/*
+	 * Wait timeout miliseconds for a newly created device to be available
+	 * from the given path.  There is a small window when a /dev/ device
+	 * will exist and the udev link will not, so we must wait for the
+	 * symlink.  Depending on the udev rules this may take a few seconds.
+	 */
+	for (i = 0; i < timeout; i++) {
+		usleep(1000);
+
+		errno = 0;
+		if ((stat64(path, &statbuf) == 0) && (errno == 0))
+			return (0);
+	}
+
+	return (ENOENT);
+#else
+	return (0);
+#endif
+}
+
 /*
  * Label an individual disk.  The name provided is the short name,
  * stripped of any leading /dev path.
@@ -3132,10 +3159,7 @@ zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
 	uint64_t slice_size;
 	diskaddr_t start_block;
 	char errbuf[1024];
-#if defined(__linux__)
+
-	struct stat64 statbuf;
-	int i;
-#endif
 	/* prepare an error message just in case */
 	(void) snprintf(errbuf, sizeof (errbuf),
 	    dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
@@ -3235,23 +3259,13 @@ zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
 	efi_free(vtoc);
 
 #if defined(__linux__)
-	/*
+	/* Wait for the first expected slice to appear */
-	 * The efi partition table has been successfully written and the
+	(void) snprintf(path, sizeof (path), "%s/%s%s",
-	 * kernel notified.  However, it still may take a moment for udev
+	    DISK_ROOT, name, FIRST_SLICE);
-	 * to notice the devfs update and properly populate /dev/.  We will
+	return zpool_label_disk_wait(path, 3000);
-	 * wait up to 3 seconds which is far far far longer than needed.
+#else
-	 */
-	(void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
-	    FIRST_SLICE);
-	for (i = 0; i < 3000; i++) {
-		if (stat64(path, &statbuf) == 0 || errno != ENOENT)
-			break;
-
-		usleep(1000);
-	}
-#endif
-
 	return (0);
+#endif
 }
 
 static boolean_t