Fix hot spares

The issue with hot spares in ZoL is because it opens all leaf
vdevs exclusively (O_EXCL).  On Linux, exclusive opens cause
subsequent exclusive opens to fail with EBUSY.

This could be resolved by not opening any of the devices
exclusively, which is what Illumos does, but the additional
protection offered by exclusive opens is desirable.  It cleanly
prevents you from accidentally adding an in-use non-ZFS device
to your pool.

To fix this we very slightly relaxed the usage of O_EXCL in
the following ways.

1) Functions which open the device but only read had the
   O_EXCL flag removed and were updated to use O_RDONLY.

2) A common holder was added to the vdev disk code.  This
   allow the ZFS code to internally open the device multiple
   times but non-ZFS callers may not.

3) An exception was added to make_disks() for hot spare when
   creating partition tables.  For hot spare devices which
   are already opened exclusively we skip creating the partition
   table because this must already have been done when the disk
   was originally added as a hot spare.

Additional minor changes include fixing check_in_use() to use
a partition instead of a slice suffix.  And is_spare() was moved
above make_disks() to avoid adding a forward reference.

Signed-off-by: Brian Behlendorf <[email protected]>
Closes #250

3 files changed, 115 insertions, 82 deletions

diff --git a/cmd/zpool/zpool_vdev.c b/cmd/zpool/zpool_vdev.c
index da121f06f..b0b17b1a5 100644
--- a/cmd/zpool/zpool_vdev.c
+++ b/cmd/zpool/zpool_vdev.c
@@ -247,7 +247,7 @@ check_disk(const char *path, blkid_cache cache, int force,
 	 * not easily decode the MBR return a failure and prompt to the
 	 * user to use force option since we cannot check the partitions.
 	 */
-	if ((fd = open(path, O_RDWR|O_DIRECT|O_EXCL)) < 0) {
+	if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
 		check_error(errno);
 		return -1;
 	}
@@ -306,7 +306,7 @@ check_disk(const char *path, blkid_cache cache, int force,
 	efi_free(vtoc);
 	(void) close(fd);
 
-        return (err);
+	return (err);
 }
 
 static int
@@ -353,7 +353,7 @@ is_whole_disk(const char *path)
 	struct dk_gpt *label;
 	int	fd;
 
-	if ((fd = open(path, O_RDWR|O_DIRECT|O_EXCL)) < 0)
+	if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 		return (B_FALSE);
 	if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 		(void) close(fd);
@@ -391,6 +391,58 @@ is_shorthand_path(const char *arg, char *path,
 }
 
 /*
+ * Determine if the given path is a hot spare within the given configuration.
+ * If no configuration is given we rely solely on the label.
+ */
+static boolean_t
+is_spare(nvlist_t *config, const char *path)
+{
+	int fd;
+	pool_state_t state;
+	char *name = NULL;
+	nvlist_t *label;
+	uint64_t guid, spareguid;
+	nvlist_t *nvroot;
+	nvlist_t **spares;
+	uint_t i, nspares;
+	boolean_t inuse;
+
+	if ((fd = open(path, O_RDONLY)) < 0)
+		return (B_FALSE);
+
+	if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
+	    !inuse ||
+	    state != POOL_STATE_SPARE ||
+	    zpool_read_label(fd, &label) != 0) {
+		free(name);
+		(void) close(fd);
+		return (B_FALSE);
+	}
+	free(name);
+	(void) close(fd);
+
+	if (config == NULL)
+		return (B_TRUE);
+
+	verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
+	nvlist_free(label);
+
+	verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
+	    &nvroot) == 0);
+	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
+	    &spares, &nspares) == 0) {
+		for (i = 0; i < nspares; i++) {
+			verify(nvlist_lookup_uint64(spares[i],
+			    ZPOOL_CONFIG_GUID, &spareguid) == 0);
+			if (spareguid == guid)
+				return (B_TRUE);
+		}
+	}
+
+	return (B_FALSE);
+}
+
+/*
  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
  * device, fill in the device id to make a complete nvlist.  Valid forms for a
  * leaf vdev are:
@@ -914,11 +966,13 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 {
 	nvlist_t **child;
 	uint_t c, children;
-	char *type, *path, *diskname;
+	char *type, *path;
 	char devpath[MAXPATHLEN];
 	char udevpath[MAXPATHLEN];
 	uint64_t wholedisk;
 	struct stat64 statbuf;
+	int is_exclusive = 0;
+	int fd;
 	int ret;
 
 	verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
@@ -941,8 +995,8 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 		    &wholedisk));
 
 		if (!wholedisk) {
-			ret = zero_label(path);
-			return (ret);
+			(void) zero_label(path);
+			return (0);
 		}
 
 		if (realpath(path, devpath) == NULL) {
@@ -964,26 +1018,44 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 		strncpy(udevpath, path, MAXPATHLEN);
 		(void) zfs_append_partition(udevpath, MAXPATHLEN);
 
-		if ((strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT)) == 0) &&
-		    (lstat64(udevpath, &statbuf) == 0) &&
-		    S_ISLNK(statbuf.st_mode))
-			(void) unlink(udevpath);
-
-		diskname = strrchr(devpath, '/');
-		assert(diskname != NULL);
-		diskname++;
-		if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
-			return (-1);
+		fd = open(devpath, O_RDWR|O_EXCL);
+		if (fd == -1) {
+			if (errno == EBUSY)
+				is_exclusive = 1;
+		} else {
+			(void) close(fd);
+		}
 
 		/*
-		 * Now we've labeled the disk and the partitions have been
-		 * created.  We still need to wait for udev to create the
-		 * symlinks to those partitions.
+		 * If the partition exists, contains a valid spare label,
+		 * and is opened exclusively there is no need to partition
+		 * it.  Hot spares have already been partitioned and are
+		 * held open exclusively by the kernel as a safety measure.
+		 *
+		 * If the provided path is for a /dev/disk/ device its
+		 * symbolic link will be removed, partition table created,
+		 * and then block until udev creates the new link.
 		 */
-		if ((ret = zpool_label_disk_wait(udevpath, 1000)) != 0) {
-			(void) fprintf(stderr,
-			    gettext( "cannot resolve path '%s'\n"), udevpath);
-			return (-1);
+		if (!is_exclusive || !is_spare(NULL, udevpath)) {
+			ret = strncmp(udevpath,UDISK_ROOT,strlen(UDISK_ROOT));
+			if (ret == 0) {
+				ret = lstat64(udevpath, &statbuf);
+				if (ret == 0 && S_ISLNK(statbuf.st_mode))
+					(void) unlink(udevpath);
+			}
+
+			if (zpool_label_disk(g_zfs, zhp,
+			    strrchr(devpath, '/') + 1) == -1)
+				return (-1);
+
+			ret = zpool_label_disk_wait(udevpath, 1000);
+			if (ret) {
+				(void) fprintf(stderr, gettext("cannot "
+				    "resolve path '%s': %d\n"), udevpath, ret);
+				return (-1);
+			}
+
+			(void) zero_label(udevpath);
 		}
 
 		/*
@@ -994,9 +1066,6 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 		 */
 		verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
 
-		/* Just in case this partition already existed. */
-		(void) zero_label(udevpath);
-
 		return (0);
 	}
 
@@ -1020,54 +1089,6 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 }
 
 /*
- * Determine if the given path is a hot spare within the given configuration.
- */
-static boolean_t
-is_spare(nvlist_t *config, const char *path)
-{
-	int fd;
-	pool_state_t state;
-	char *name = NULL;
-	nvlist_t *label;
-	uint64_t guid, spareguid;
-	nvlist_t *nvroot;
-	nvlist_t **spares;
-	uint_t i, nspares;
-	boolean_t inuse;
-
-	if ((fd = open(path, O_RDONLY|O_EXCL)) < 0)
-		return (B_FALSE);
-
-	if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
-	    !inuse ||
-	    state != POOL_STATE_SPARE ||
-	    zpool_read_label(fd, &label) != 0) {
-		free(name);
-		(void) close(fd);
-		return (B_FALSE);
-	}
-	free(name);
-	(void) close(fd);
-
-	verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
-	nvlist_free(label);
-
-	verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
-	    &nvroot) == 0);
-	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
-	    &spares, &nspares) == 0) {
-		for (i = 0; i < nspares; i++) {
-			verify(nvlist_lookup_uint64(spares[i],
-			    ZPOOL_CONFIG_GUID, &spareguid) == 0);
-			if (spareguid == guid)
-				return (B_TRUE);
-		}
-	}
-
-	return (B_FALSE);
-}
-
-/*
  * Go through and find any devices that are in use.  We rely on libdiskmgt for
  * the majority of this task.
  */
@@ -1098,11 +1119,12 @@ check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
 		 * regardless of what libblkid or zpool_in_use() says.
 		 */
 		if (replacing) {
-			if (wholedisk)
-				(void) snprintf(buf, sizeof (buf), "%ss0",
-				    path);
-			else
-				(void) strlcpy(buf, path, sizeof (buf));
+			(void) strlcpy(buf, path, sizeof (buf));
+			if (wholedisk) {
+				ret = zfs_append_partition(buf,  sizeof (buf));
+				if (ret == -1)
+					return (-1);
+			}
 
 			if (is_spare(config, buf))
 				return (0);
diff --git a/include/linux/blkdev_compat.h b/include/linux/blkdev_compat.h
index 9d3e6f07c..47f569bbd 100644
--- a/include/linux/blkdev_compat.h
+++ b/include/linux/blkdev_compat.h
@@ -478,4 +478,13 @@ blk_queue_discard_granularity(struct request_queue *q, unsigned int dg)
  */
 #define	VDEV_SCHEDULER			"noop"
 
+/*
+ * A common holder for vdev_bdev_open() is used to relax the exclusive open
+ * semantics slightly.  Internal vdev disk callers may pass VDEV_HOLDER to
+ * allow them to open the device multiple times.  Other kernel callers and
+ * user space processes which don't pass this value will get EBUSY.  This is
+ * currently required for the correct operation of hot spares.
+ */
+#define VDEV_HOLDER			((void *)0x2f5401de7)
+
 #endif /* _ZFS_BLKDEV_H */
diff --git a/module/zfs/vdev_disk.c b/module/zfs/vdev_disk.c
index f93b7bcc8..1f6507f2f 100644
--- a/module/zfs/vdev_disk.c
+++ b/module/zfs/vdev_disk.c
@@ -34,6 +34,7 @@
 #include <sys/sunldi.h>
 
 char *zfs_vdev_scheduler = VDEV_SCHEDULER;
+static void *zfs_vdev_holder = VDEV_HOLDER;
 
 /*
  * Virtual device vector for disks.
@@ -203,7 +204,7 @@ vdev_disk_rrpart(const char *path, int mode, vdev_disk_t *vd)
 	struct gendisk *disk;
 	int error, partno;
 
-	bdev = vdev_bdev_open(path, vdev_bdev_mode(mode), vd);
+	bdev = vdev_bdev_open(path, vdev_bdev_mode(mode), zfs_vdev_holder);
 	if (IS_ERR(bdev))
 		return bdev;
 
@@ -281,7 +282,8 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
 	if (v->vdev_wholedisk && v->vdev_expanding)
 		bdev = vdev_disk_rrpart(v->vdev_path, mode, vd);
 	if (IS_ERR(bdev))
-		bdev = vdev_bdev_open(v->vdev_path, vdev_bdev_mode(mode), vd);
+		bdev = vdev_bdev_open(v->vdev_path,
+		    vdev_bdev_mode(mode), zfs_vdev_holder);
 	if (IS_ERR(bdev)) {
 		kmem_free(vd, sizeof(vdev_disk_t));
 		return -PTR_ERR(bdev);
@@ -783,7 +785,7 @@ vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
 	uint64_t s, size;
 	int i;
 
-	bdev = vdev_bdev_open(devpath, vdev_bdev_mode(FREAD), NULL);
+	bdev = vdev_bdev_open(devpath, vdev_bdev_mode(FREAD), zfs_vdev_holder);
 	if (IS_ERR(bdev))
 		return -PTR_ERR(bdev);