*
* 1. Construct the vdev specification. Performs syntax validation and
* makes sure each device is valid.
- * 2. Check for devices in use. Using libdiskmgt, makes sure that no
+ * 2. Check for devices in use. Using libblkid to make sure that no
* devices are also in use. Some can be overridden using the 'force'
* flag, others cannot.
* 3. Check for replication errors if the 'force' flag is not specified.
*/
#include <assert.h>
+#include <ctype.h>
#include <devid.h>
#include <errno.h>
#include <fcntl.h>
-#include <libdiskmgt.h>
#include <libintl.h>
#include <libnvpair.h>
#include <limits.h>
+#include <scsi/scsi.h>
+#include <scsi/sg.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/vtoc.h>
#include <sys/mntent.h>
+#include <uuid/uuid.h>
+#ifdef HAVE_LIBBLKID
+#include <blkid/blkid.h>
+#else
+#define blkid_cache void *
+#endif /* HAVE_LIBBLKID */
#include "zpool_util.h"
-
-#define DISK_ROOT "/dev/dsk"
-#define RDISK_ROOT "/dev/rdsk"
-#define BACKUP_SLICE "s2"
+#include <sys/zfs_context.h>
/*
* For any given vdev specification, we can have multiple errors. The
boolean_t error_seen;
boolean_t is_force;
-/*PRINTFLIKE1*/
-static void
-vdev_error(const char *fmt, ...)
+typedef struct vdev_disk_db_entry
{
- va_list ap;
-
- if (!error_seen) {
- (void) fprintf(stderr, gettext("invalid vdev specification\n"));
- if (!is_force)
- (void) fprintf(stderr, gettext("use '-f' to override "
- "the following errors:\n"));
- else
- (void) fprintf(stderr, gettext("the following errors "
- "must be manually repaired:\n"));
- error_seen = B_TRUE;
- }
-
- va_start(ap, fmt);
- (void) vfprintf(stderr, fmt, ap);
- va_end(ap);
-}
-
-static void
-libdiskmgt_error(int error)
-{
- /*
- * ENXIO/ENODEV is a valid error message if the device doesn't live in
- * /dev/dsk. Don't bother printing an error message in this case.
- */
- if (error == ENXIO || error == ENODEV)
- return;
-
- (void) fprintf(stderr, gettext("warning: device in use checking "
- "failed: %s\n"), strerror(error));
-}
+ char id[24];
+ int sector_size;
+} vdev_disk_db_entry_t;
/*
- * Validate a device, passing the bulk of the work off to libdiskmgt.
+ * Database of block devices that lie about physical sector sizes. The
+ * identification string must be precisely 24 characters to avoid false
+ * negatives
*/
-static int
-check_slice(const char *path, int force, boolean_t wholedisk, boolean_t isspare)
-{
- char *msg;
- int error = 0;
- dm_who_type_t who;
-
- if (force)
- who = DM_WHO_ZPOOL_FORCE;
- else if (isspare)
- who = DM_WHO_ZPOOL_SPARE;
- else
- who = DM_WHO_ZPOOL;
-
- if (dm_inuse((char *)path, &msg, who, &error) || error) {
- if (error != 0) {
- libdiskmgt_error(error);
- return (0);
- } else {
- vdev_error("%s", msg);
- free(msg);
- return (-1);
- }
- }
-
- /*
- * If we're given a whole disk, ignore overlapping slices since we're
- * about to label it anyway.
- */
- error = 0;
- if (!wholedisk && !force &&
- (dm_isoverlapping((char *)path, &msg, &error) || error)) {
- if (error == 0) {
- /* dm_isoverlapping returned -1 */
- vdev_error(gettext("%s overlaps with %s\n"), path, msg);
- free(msg);
- return (-1);
- } else if (error != ENODEV) {
- /* libdiskmgt's devcache only handles physical drives */
- libdiskmgt_error(error);
- return (0);
- }
- }
-
- return (0);
-}
-
+static vdev_disk_db_entry_t vdev_disk_database[] = {
+ {"ATA Corsair Force 3 ", 8192},
+ {"ATA INTEL SSDSA2CT04", 8192},
+ {"ATA INTEL SSDSA2CW16", 8192},
+ {"ATA INTEL SSDSC2CT18", 8192},
+ {"ATA INTEL SSDSC2CW12", 8192},
+ {"ATA KINGSTON SH100S3", 8192},
+ {"ATA M4-CT064M4SSD2 ", 8192},
+ {"ATA M4-CT128M4SSD2 ", 8192},
+ {"ATA M4-CT256M4SSD2 ", 8192},
+ {"ATA M4-CT512M4SSD2 ", 8192},
+ {"ATA OCZ-AGILITY2 ", 8192},
+ {"ATA OCZ-VERTEX2 3.5 ", 8192},
+ {"ATA OCZ-VERTEX3 ", 8192},
+ {"ATA OCZ-VERTEX3 LT ", 8192},
+ {"ATA OCZ-VERTEX3 MI ", 8192},
+ {"ATA SAMSUNG SSD 830 ", 8192},
+ {"ATA Samsung SSD 840 ", 8192},
+ {"ATA INTEL SSDSA2M040", 4096},
+ {"ATA INTEL SSDSA2M080", 4096},
+ {"ATA INTEL SSDSA2M160", 4096},
+ /* Imported from Open Solaris*/
+ {"ATA MARVELL SD88SA02", 4096},
+ /* Advanced format Hard drives */
+ {"ATA Hitachi HDS5C303", 4096},
+ {"ATA SAMSUNG HD204UI ", 4096},
+ {"ATA ST2000DL004 HD20", 4096},
+ {"ATA WDC WD10EARS-00M", 4096},
+ {"ATA WDC WD10EARS-00S", 4096},
+ {"ATA WDC WD10EARS-00Z", 4096},
+ {"ATA WDC WD15EARS-00M", 4096},
+ {"ATA WDC WD15EARS-00S", 4096},
+ {"ATA WDC WD15EARS-00Z", 4096},
+ {"ATA WDC WD20EARS-00M", 4096},
+ {"ATA WDC WD20EARS-00S", 4096},
+ {"ATA WDC WD20EARS-00Z", 4096},
+ /* Virtual disks: Assume zvols with default volblocksize */
+#if 0
+ {"ATA QEMU HARDDISK ", 8192},
+ {"IET VIRTUAL-DISK ", 8192},
+ {"OI COMSTAR ", 8192},
+#endif
+};
+
+static const int vdev_disk_database_size =
+ sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
+
+#define INQ_REPLY_LEN 96
+#define INQ_CMD_LEN 6
-/*
- * Validate a whole disk. Iterate over all slices on the disk and make sure
- * that none is in use by calling check_slice().
- */
-static int
-check_disk(const char *name, dm_descriptor_t disk, int force, int isspare)
+static boolean_t
+check_sector_size_database(char *path, int *sector_size)
{
- dm_descriptor_t *drive, *media, *slice;
- int err = 0;
+ unsigned char inq_buff[INQ_REPLY_LEN];
+ unsigned char sense_buffer[32];
+ unsigned char inq_cmd_blk[INQ_CMD_LEN] =
+ {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
+ sg_io_hdr_t io_hdr;
+ int error;
+ int fd;
int i;
- int ret;
-
- /*
- * Get the drive associated with this disk. This should never fail,
- * because we already have an alias handle open for the device.
- */
- if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE,
- &err)) == NULL || *drive == NULL) {
- if (err)
- libdiskmgt_error(err);
- return (0);
- }
- if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA,
- &err)) == NULL) {
- dm_free_descriptors(drive);
- if (err)
- libdiskmgt_error(err);
- return (0);
- }
-
- dm_free_descriptors(drive);
-
- /*
- * It is possible that the user has specified a removable media drive,
- * and the media is not present.
- */
- if (*media == NULL) {
- dm_free_descriptors(media);
- vdev_error(gettext("'%s' has no media in drive\n"), name);
- return (-1);
- }
+ /* Prepare INQUIRY command */
+ memset(&io_hdr, 0, sizeof(sg_io_hdr_t));
+ io_hdr.interface_id = 'S';
+ io_hdr.cmd_len = sizeof(inq_cmd_blk);
+ io_hdr.mx_sb_len = sizeof(sense_buffer);
+ io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
+ io_hdr.dxfer_len = INQ_REPLY_LEN;
+ io_hdr.dxferp = inq_buff;
+ io_hdr.cmdp = inq_cmd_blk;
+ io_hdr.sbp = sense_buffer;
+ io_hdr.timeout = 10; /* 10 milliseconds is ample time */
+
+ if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
+ return (B_FALSE);
- if ((slice = dm_get_associated_descriptors(*media, DM_SLICE,
- &err)) == NULL) {
- dm_free_descriptors(media);
- if (err)
- libdiskmgt_error(err);
- return (0);
- }
+ error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
- dm_free_descriptors(media);
+ (void) close(fd);
- ret = 0;
+ if (error < 0)
+ return (B_FALSE);
- /*
- * Iterate over all slices and report any errors. We don't care about
- * overlapping slices because we are using the whole disk.
- */
- for (i = 0; slice[i] != NULL; i++) {
- char *name = dm_get_name(slice[i], &err);
+ if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
+ return (B_FALSE);
- if (check_slice(name, force, B_TRUE, isspare) != 0)
- ret = -1;
+ for (i = 0; i < vdev_disk_database_size; i++) {
+ if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
+ continue;
- dm_free_name(name);
+ *sector_size = vdev_disk_database[i].sector_size;
+ return (B_TRUE);
}
- dm_free_descriptors(slice);
- return (ret);
+ return (B_FALSE);
}
-/*
- * Validate a device.
- */
-static int
-check_device(const char *path, boolean_t force, boolean_t isspare)
+/*PRINTFLIKE1*/
+static void
+vdev_error(const char *fmt, ...)
{
- dm_descriptor_t desc;
- int err;
- char *dev;
+ va_list ap;
- /*
- * For whole disks, libdiskmgt does not include the leading dev path.
- */
- dev = strrchr(path, '/');
- assert(dev != NULL);
- dev++;
- if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != NULL) {
- err = check_disk(path, desc, force, isspare);
- dm_free_descriptor(desc);
- return (err);
+ if (!error_seen) {
+ (void) fprintf(stderr, gettext("invalid vdev specification\n"));
+ if (!is_force)
+ (void) fprintf(stderr, gettext("use '-f' to override "
+ "the following errors:\n"));
+ else
+ (void) fprintf(stderr, gettext("the following errors "
+ "must be manually repaired:\n"));
+ error_seen = B_TRUE;
}
- return (check_slice(path, force, B_FALSE, isspare));
+ va_start(ap, fmt);
+ (void) vfprintf(stderr, fmt, ap);
+ va_end(ap);
}
/*
char *name;
int fd;
int ret = 0;
- int err;
pool_state_t state;
boolean_t inuse;
- if (dm_inuse_swap(file, &err)) {
- if (err)
- libdiskmgt_error(err);
- else
- vdev_error(gettext("%s is currently used by swap. "
- "Please see swap(1M).\n"), file);
- return (-1);
- }
-
if ((fd = open(file, O_RDONLY)) < 0)
return (0);
return (ret);
}
+static void
+check_error(int err)
+{
+ (void) fprintf(stderr, gettext("warning: device in use checking "
+ "failed: %s\n"), strerror(err));
+}
+
+static int
+check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
+{
+ int err;
+#ifdef HAVE_LIBBLKID
+ char *value;
+
+ /* No valid type detected device is safe to use */
+ value = blkid_get_tag_value(cache, "TYPE", path);
+ if (value == NULL)
+ return (0);
+
+ /*
+ * If libblkid detects a ZFS device, we check the device
+ * using check_file() to see if it's safe. The one safe
+ * case is a spare device shared between multiple pools.
+ */
+ if (strcmp(value, "zfs") == 0) {
+ err = check_file(path, force, isspare);
+ } else {
+ if (force) {
+ err = 0;
+ } else {
+ err = -1;
+ vdev_error(gettext("%s contains a filesystem of "
+ "type '%s'\n"), path, value);
+ }
+ }
+
+ free(value);
+#else
+ err = check_file(path, force, isspare);
+#endif /* HAVE_LIBBLKID */
+
+ return (err);
+}
+
+/*
+ * Validate a whole disk. Iterate over all slices on the disk and make sure
+ * that none is in use by calling check_slice().
+ */
+static int
+check_disk(const char *path, blkid_cache cache, int force,
+ boolean_t isspare, boolean_t iswholedisk)
+{
+ struct dk_gpt *vtoc;
+ char slice_path[MAXPATHLEN];
+ int err = 0;
+ int fd, i;
+
+ /* This is not a wholedisk we only check the given partition */
+ if (!iswholedisk)
+ return check_slice(path, cache, force, isspare);
+
+ /*
+ * When the device is a whole disk try to read the efi partition
+ * label. If this is successful we safely check the all of the
+ * partitions. However, when it fails it may simply be because
+ * the disk is partitioned via the MBR. Since we currently can
+ * 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_RDONLY|O_DIRECT)) < 0) {
+ check_error(errno);
+ return -1;
+ }
+
+ if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
+ (void) close(fd);
+
+ if (force) {
+ return 0;
+ } else {
+ vdev_error(gettext("%s does not contain an EFI "
+ "label but it may contain partition\n"
+ "information in the MBR.\n"), path);
+ return -1;
+ }
+ }
+
+ /*
+ * The primary efi partition label is damaged however the secondary
+ * 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) {
+ efi_free(vtoc);
+ (void) close(fd);
+
+ if (force) {
+ /* Partitions will no be created using the backup */
+ return 0;
+ } else {
+ vdev_error(gettext("%s contains a corrupt primary "
+ "EFI label.\n"), path);
+ return -1;
+ }
+ }
+
+ for (i = 0; i < vtoc->efi_nparts; i++) {
+
+ if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
+ uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
+ continue;
+
+ if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
+ (void) snprintf(slice_path, sizeof (slice_path),
+ "%s%s%d", path, "-part", i+1);
+ else
+ (void) snprintf(slice_path, sizeof (slice_path),
+ "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
+ "p" : "", i+1);
+
+ err = check_slice(slice_path, cache, force, isspare);
+ if (err)
+ break;
+ }
+
+ efi_free(vtoc);
+ (void) close(fd);
+
+ return (err);
+}
+
+static int
+check_device(const char *path, boolean_t force,
+ boolean_t isspare, boolean_t iswholedisk)
+{
+ static blkid_cache cache = NULL;
+
+#ifdef HAVE_LIBBLKID
+ /*
+ * There is no easy way to add a correct blkid_put_cache() call,
+ * memory will be reclaimed when the command exits.
+ */
+ if (cache == NULL) {
+ int err;
+
+ if ((err = blkid_get_cache(&cache, NULL)) != 0) {
+ check_error(err);
+ return -1;
+ }
+
+ if ((err = blkid_probe_all(cache)) != 0) {
+ blkid_put_cache(cache);
+ check_error(err);
+ return -1;
+ }
+ }
+#endif /* HAVE_LIBBLKID */
+
+ return check_disk(path, cache, force, isspare, iswholedisk);
+}
/*
* By "whole disk" we mean an entire physical disk (something we can
* it isn't.
*/
static boolean_t
-is_whole_disk(const char *arg)
+is_whole_disk(const char *path)
{
struct dk_gpt *label;
int fd;
- char path[MAXPATHLEN];
- (void) snprintf(path, sizeof (path), "%s%s%s",
- RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE);
- if ((fd = open(path, O_RDWR | O_NDELAY)) < 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);
}
/*
+ * This may be a shorthand device path or it could be total gibberish.
+ * Check to see if it is a known device available in zfs_vdev_paths.
+ * As part of this check, see if we've been given an entire disk
+ * (minus the slice number).
+ */
+static int
+is_shorthand_path(const char *arg, char *path,
+ struct stat64 *statbuf, boolean_t *wholedisk)
+{
+ int error;
+
+ error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
+ if (error == 0) {
+ *wholedisk = is_whole_disk(path);
+ if (*wholedisk || (stat64(path, statbuf) == 0))
+ return (0);
+ }
+
+ strlcpy(path, arg, sizeof(path));
+ memset(statbuf, 0, sizeof(*statbuf));
+ *wholedisk = B_FALSE;
+
+ return (error);
+}
+
+/*
+ * 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:
*
- * /dev/dsk/xxx Complete disk path
- * /xxx Full path to file
- * xxx Shorthand for /dev/dsk/xxx
+ * /dev/xxx Complete disk path
+ * /xxx Full path to file
+ * xxx Shorthand for <zfs_vdev_paths>/xxx
*/
static nvlist_t *
-make_leaf_vdev(const char *arg, uint64_t is_log)
+make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
{
char path[MAXPATHLEN];
struct stat64 statbuf;
nvlist_t *vdev = NULL;
char *type = NULL;
boolean_t wholedisk = B_FALSE;
+ uint64_t ashift = 0;
+ int err;
/*
* Determine what type of vdev this is, and put the full path into
if (arg[0] == '/') {
/*
* Complete device or file path. Exact type is determined by
- * examining the file descriptor afterwards.
+ * examining the file descriptor afterwards. Symbolic links
+ * are resolved to their real paths for the is_whole_disk()
+ * and S_ISBLK/S_ISREG type checks. However, we are careful
+ * to store the given path as ZPOOL_CONFIG_PATH to ensure we
+ * can leverage udev's persistent device labels.
*/
- wholedisk = is_whole_disk(arg);
- if (!wholedisk && (stat64(arg, &statbuf) != 0)) {
+ if (realpath(arg, path) == NULL) {
(void) fprintf(stderr,
- gettext("cannot open '%s': %s\n"),
- arg, strerror(errno));
+ gettext("cannot resolve path '%s'\n"), arg);
return (NULL);
}
- (void) strlcpy(path, arg, sizeof (path));
- } else {
- /*
- * This may be a short path for a device, or it could be total
- * gibberish. Check to see if it's a known device in
- * /dev/dsk/. As part of this check, see if we've been given a
- * an entire disk (minus the slice number).
- */
- (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT,
- arg);
wholedisk = is_whole_disk(path);
if (!wholedisk && (stat64(path, &statbuf) != 0)) {
+ (void) fprintf(stderr,
+ gettext("cannot open '%s': %s\n"),
+ path, strerror(errno));
+ return (NULL);
+ }
+
+ /* After is_whole_disk() check restore original passed path */
+ strlcpy(path, arg, MAXPATHLEN);
+ } else {
+ err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
+ if (err != 0) {
/*
* If we got ENOENT, then the user gave us
* gibberish, so try to direct them with a
* regurgitate strerror() since it's the best we
* can do.
*/
- if (errno == ENOENT) {
+ if (err == ENOENT) {
(void) fprintf(stderr,
gettext("cannot open '%s': no such "
"device in %s\n"), arg, DISK_ROOT);
(uint64_t)wholedisk) == 0);
/*
- * For a whole disk, defer getting its devid until after labeling it.
+ * Override defaults if custom properties are provided.
*/
- if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
- /*
- * Get the devid for the device.
- */
- int fd;
- ddi_devid_t devid;
- char *minor = NULL, *devid_str = NULL;
-
- if ((fd = open(path, O_RDONLY)) < 0) {
- (void) fprintf(stderr, gettext("cannot open '%s': "
- "%s\n"), path, strerror(errno));
- nvlist_free(vdev);
- return (NULL);
- }
+ if (props != NULL) {
+ char *value = NULL;
- 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(vdev,
- 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);
- }
+ if (nvlist_lookup_string(props,
+ zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
+ zfs_nicestrtonum(NULL, value, &ashift);
+ }
- (void) close(fd);
+ /*
+ * If the device is known to incorrectly report its physical sector
+ * size explicitly provide the known correct value.
+ */
+ if (ashift == 0) {
+ int sector_size;
+
+ if (check_sector_size_database(path, §or_size) == B_TRUE)
+ ashift = highbit(sector_size) - 1;
}
+ if (ashift > 0)
+ nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
+
return (vdev);
}
return (ret);
}
+static int
+zero_label(char *path)
+{
+ const int size = 4096;
+ char buf[size];
+ int err, fd;
+
+ if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
+ (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
+ path, strerror(errno));
+ return (-1);
+ }
+
+ memset(buf, 0, size);
+ err = write(fd, buf, size);
+ (void) fdatasync(fd);
+ (void) close(fd);
+
+ if (err == -1) {
+ (void) fprintf(stderr, gettext("cannot zero first %d bytes "
+ "of '%s': %s\n"), size, path, strerror(errno));
+ return (-1);
+ }
+
+ if (err != size) {
+ (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
+ "of '%s'\n"), err, size, path);
+ return (-1);
+ }
+
+ return 0;
+}
+
/*
* Go through and find any whole disks in the vdev specification, labelling them
* as appropriate. When constructing the vdev spec, we were unable to open this
{
nvlist_t **child;
uint_t c, children;
- char *type, *path, *diskname;
- char buf[MAXPATHLEN];
+ char *type, *path;
+ char devpath[MAXPATHLEN];
+ char udevpath[MAXPATHLEN];
uint64_t wholedisk;
+ struct stat64 statbuf;
+ int is_exclusive = 0;
int fd;
int ret;
- ddi_devid_t devid;
- char *minor = NULL, *devid_str = NULL;
verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
return (0);
/*
- * We have a disk device. Get the path to the device
- * and see if it's a whole disk by appending the backup
- * slice and stat()ing the device.
+ * We have a disk device. If this is a whole disk write
+ * out the efi partition table, otherwise write zero's to
+ * the first 4k of the partition. This is to ensure that
+ * libblkid will not misidentify the partition due to a
+ * magic value left by the previous filesystem.
*/
- verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
- if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
- &wholedisk) != 0 || !wholedisk)
+ verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
+ verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
+ &wholedisk));
+
+ if (!wholedisk) {
+ (void) zero_label(path);
return (0);
+ }
- diskname = strrchr(path, '/');
- assert(diskname != NULL);
- diskname++;
- if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
- return (-1);
+ if (realpath(path, devpath) == NULL) {
+ ret = errno;
+ (void) fprintf(stderr,
+ gettext("cannot resolve path '%s'\n"), path);
+ return (ret);
+ }
/*
- * Fill in the devid, now that we've labeled the disk.
+ * Remove any previously existing symlink from a udev path to
+ * the device before labeling the disk. This makes
+ * zpool_label_disk_wait() truly wait for the new link to show
+ * up instead of returning if it finds an old link still in
+ * place. Otherwise there is a window between when udev
+ * deletes and recreates the link during which access attempts
+ * will fail with ENOENT.
*/
- (void) snprintf(buf, sizeof (buf), "%ss0", path);
- if ((fd = open(buf, O_RDONLY)) < 0) {
- (void) fprintf(stderr,
- gettext("cannot open '%s': %s\n"),
- buf, strerror(errno));
- return (-1);
+ strncpy(udevpath, path, MAXPATHLEN);
+ (void) zfs_append_partition(udevpath, MAXPATHLEN);
+
+ fd = open(devpath, O_RDWR|O_EXCL);
+ if (fd == -1) {
+ if (errno == EBUSY)
+ is_exclusive = 1;
+ } else {
+ (void) close(fd);
}
- 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 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 (!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 (devid_str != NULL)
- devid_str_free(devid_str);
- if (minor != NULL)
- devid_str_free(minor);
- devid_free(devid);
+
+ 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);
}
/*
- * Update the path to refer to the 's0' slice. The presence of
+ * Update the path to refer to the partition. The presence of
* the 'whole_disk' field indicates to the CLI that we should
- * chop off the slice number when displaying the device in
+ * chop off the partition number when displaying the device in
* future output.
*/
- verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
-
- (void) close(fd);
+ verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
return (0);
}
}
/*
- * 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)) < 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.
*/
nvlist_t **child;
uint_t c, children;
char *type, *path;
- int ret;
+ int ret = 0;
char buf[MAXPATHLEN];
- uint64_t wholedisk;
+ uint64_t wholedisk = B_FALSE;
verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
&child, &children) != 0) {
- verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
+ verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
+ if (strcmp(type, VDEV_TYPE_DISK) == 0)
+ verify(!nvlist_lookup_uint64(nv,
+ ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
/*
* As a generic check, we look to see if this is a replace of a
* hot spare within the same pool. If so, we allow it
- * regardless of what libdiskmgt or zpool_in_use() says.
+ * regardless of what libblkid or zpool_in_use() says.
*/
if (replacing) {
- if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
- &wholedisk) == 0 && 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);
}
if (strcmp(type, VDEV_TYPE_DISK) == 0)
- ret = check_device(path, force, isspare);
+ ret = check_device(path, force, isspare, wholedisk);
if (strcmp(type, VDEV_TYPE_FILE) == 0)
ret = check_file(path, force, isspare);
* because the program is just going to exit anyway.
*/
nvlist_t *
-construct_spec(int argc, char **argv)
+construct_spec(nvlist_t *props, int argc, char **argv)
{
nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
children * sizeof (nvlist_t *));
if (child == NULL)
zpool_no_memory();
- if ((nv = make_leaf_vdev(argv[c], B_FALSE))
+ if ((nv = make_leaf_vdev(props, argv[c], B_FALSE))
== NULL)
return (NULL);
child[children - 1] = nv;
* We have a device. Pass off to make_leaf_vdev() to
* construct the appropriate nvlist describing the vdev.
*/
- if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
+ if ((nv = make_leaf_vdev(props, argv[0], is_log)) == NULL)
return (NULL);
if (is_log)
nlogs++;
uint_t c, children;
if (argc > 0) {
- if ((newroot = construct_spec(argc, argv)) == NULL) {
+ if ((newroot = construct_spec(props, argc, argv)) == NULL) {
(void) fprintf(stderr, gettext("Unable to build a "
"pool from the specified devices\n"));
return (NULL);
* added, even if they appear in use.
*/
nvlist_t *
-make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
+make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
boolean_t replacing, boolean_t dryrun, int argc, char **argv)
{
nvlist_t *newroot;
* that we have a valid specification, and that all devices can be
* opened.
*/
- if ((newroot = construct_spec(argc, argv)) == NULL)
+ if ((newroot = construct_spec(props, argc, argv)) == NULL)
return (NULL);
if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
git://git.camperquake.de / zfs.git / blobdiff