mirror_zfs/cmd/zhack/zhack.c
Fedor Uporov d04b5c9e87
zhack: Add repair label option
In case if all label checksums will be invalid on any vdev, the pool
will become unimportable. The zhack with newly added cli options could
be used to restore label checksums and make pool importable again.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Fedor Uporov <fuporov.vstack@gmail.com>
Closes #2510
Closes #12686
2021-11-11 11:26:18 -08:00

703 lines
16 KiB
C

/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright (c) 2011, 2015 by Delphix. All rights reserved.
* Copyright (c) 2013 Steven Hartland. All rights reserved.
*/
/*
* zhack is a debugging tool that can write changes to ZFS pool using libzpool
* for testing purposes. Altering pools with zhack is unsupported and may
* result in corrupted pools.
*/
#include <stdio.h>
#include <stdlib.h>
#include <ctype.h>
#include <sys/stat.h>
#include <sys/zfs_context.h>
#include <sys/spa.h>
#include <sys/spa_impl.h>
#include <sys/dmu.h>
#include <sys/zap.h>
#include <sys/zfs_znode.h>
#include <sys/dsl_synctask.h>
#include <sys/vdev.h>
#include <sys/vdev_impl.h>
#include <sys/fs/zfs.h>
#include <sys/dmu_objset.h>
#include <sys/dsl_pool.h>
#include <sys/zio_checksum.h>
#include <sys/zio_compress.h>
#include <sys/zfeature.h>
#include <sys/dmu_tx.h>
#include <zfeature_common.h>
#include <libzutil.h>
const char cmdname[] = "zhack";
static importargs_t g_importargs;
static char *g_pool;
static boolean_t g_readonly;
static __attribute__((noreturn)) void
usage(void)
{
(void) fprintf(stderr,
"Usage: %s [-c cachefile] [-d dir] <subcommand> <args> ...\n"
"where <subcommand> <args> is one of the following:\n"
"\n", cmdname);
(void) fprintf(stderr,
" feature stat <pool>\n"
" print information about enabled features\n"
" feature enable [-r] [-d desc] <pool> <feature>\n"
" add a new enabled feature to the pool\n"
" -d <desc> sets the feature's description\n"
" -r set read-only compatible flag for feature\n"
" feature ref [-md] <pool> <feature>\n"
" change the refcount on the given feature\n"
" -d decrease instead of increase the refcount\n"
" -m add the feature to the label if increasing refcount\n"
"\n"
" <feature> : should be a feature guid\n"
"\n"
" label repair <device>\n"
" repair corrupted label checksums\n"
"\n"
" <device> : path to vdev\n");
exit(1);
}
static __attribute__((noreturn)) __attribute__((format(printf, 3, 4))) void
fatal(spa_t *spa, void *tag, const char *fmt, ...)
{
va_list ap;
if (spa != NULL) {
spa_close(spa, tag);
(void) spa_export(g_pool, NULL, B_TRUE, B_FALSE);
}
va_start(ap, fmt);
(void) fprintf(stderr, "%s: ", cmdname);
(void) vfprintf(stderr, fmt, ap);
va_end(ap);
(void) fprintf(stderr, "\n");
exit(1);
}
/* ARGSUSED */
static int
space_delta_cb(dmu_object_type_t bonustype, const void *data,
zfs_file_info_t *zoi)
{
/*
* Is it a valid type of object to track?
*/
if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
return (ENOENT);
(void) fprintf(stderr, "modifying object that needs user accounting");
abort();
}
/*
* Target is the dataset whose pool we want to open.
*/
static void
zhack_import(char *target, boolean_t readonly)
{
nvlist_t *config;
nvlist_t *props;
int error;
kernel_init(readonly ? SPA_MODE_READ :
(SPA_MODE_READ | SPA_MODE_WRITE));
dmu_objset_register_type(DMU_OST_ZFS, space_delta_cb);
g_readonly = readonly;
g_importargs.can_be_active = readonly;
g_pool = strdup(target);
error = zpool_find_config(NULL, target, &config, &g_importargs,
&libzpool_config_ops);
if (error)
fatal(NULL, FTAG, "cannot import '%s'", target);
props = NULL;
if (readonly) {
VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
VERIFY(nvlist_add_uint64(props,
zpool_prop_to_name(ZPOOL_PROP_READONLY), 1) == 0);
}
zfeature_checks_disable = B_TRUE;
error = spa_import(target, config, props,
(readonly ? ZFS_IMPORT_SKIP_MMP : ZFS_IMPORT_NORMAL));
fnvlist_free(config);
zfeature_checks_disable = B_FALSE;
if (error == EEXIST)
error = 0;
if (error)
fatal(NULL, FTAG, "can't import '%s': %s", target,
strerror(error));
}
static void
zhack_spa_open(char *target, boolean_t readonly, void *tag, spa_t **spa)
{
int err;
zhack_import(target, readonly);
zfeature_checks_disable = B_TRUE;
err = spa_open(target, spa, tag);
zfeature_checks_disable = B_FALSE;
if (err != 0)
fatal(*spa, FTAG, "cannot open '%s': %s", target,
strerror(err));
if (spa_version(*spa) < SPA_VERSION_FEATURES) {
fatal(*spa, FTAG, "'%s' has version %d, features not enabled",
target, (int)spa_version(*spa));
}
}
static void
dump_obj(objset_t *os, uint64_t obj, const char *name)
{
zap_cursor_t zc;
zap_attribute_t za;
(void) printf("%s_obj:\n", name);
for (zap_cursor_init(&zc, os, obj);
zap_cursor_retrieve(&zc, &za) == 0;
zap_cursor_advance(&zc)) {
if (za.za_integer_length == 8) {
ASSERT(za.za_num_integers == 1);
(void) printf("\t%s = %llu\n",
za.za_name, (u_longlong_t)za.za_first_integer);
} else {
ASSERT(za.za_integer_length == 1);
char val[1024];
VERIFY(zap_lookup(os, obj, za.za_name,
1, sizeof (val), val) == 0);
(void) printf("\t%s = %s\n", za.za_name, val);
}
}
zap_cursor_fini(&zc);
}
static void
dump_mos(spa_t *spa)
{
nvlist_t *nv = spa->spa_label_features;
nvpair_t *pair;
(void) printf("label config:\n");
for (pair = nvlist_next_nvpair(nv, NULL);
pair != NULL;
pair = nvlist_next_nvpair(nv, pair)) {
(void) printf("\t%s\n", nvpair_name(pair));
}
}
static void
zhack_do_feature_stat(int argc, char **argv)
{
spa_t *spa;
objset_t *os;
char *target;
argc--;
argv++;
if (argc < 1) {
(void) fprintf(stderr, "error: missing pool name\n");
usage();
}
target = argv[0];
zhack_spa_open(target, B_TRUE, FTAG, &spa);
os = spa->spa_meta_objset;
dump_obj(os, spa->spa_feat_for_read_obj, "for_read");
dump_obj(os, spa->spa_feat_for_write_obj, "for_write");
dump_obj(os, spa->spa_feat_desc_obj, "descriptions");
if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
dump_obj(os, spa->spa_feat_enabled_txg_obj, "enabled_txg");
}
dump_mos(spa);
spa_close(spa, FTAG);
}
static void
zhack_feature_enable_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
feature_enable_sync(spa, feature, tx);
spa_history_log_internal(spa, "zhack enable feature", tx,
"name=%s flags=%u",
feature->fi_guid, feature->fi_flags);
}
static void
zhack_do_feature_enable(int argc, char **argv)
{
int c;
char *desc, *target;
spa_t *spa;
objset_t *mos;
zfeature_info_t feature;
spa_feature_t nodeps[] = { SPA_FEATURE_NONE };
/*
* Features are not added to the pool's label until their refcounts
* are incremented, so fi_mos can just be left as false for now.
*/
desc = NULL;
feature.fi_uname = "zhack";
feature.fi_flags = 0;
feature.fi_depends = nodeps;
feature.fi_feature = SPA_FEATURE_NONE;
optind = 1;
while ((c = getopt(argc, argv, "+rd:")) != -1) {
switch (c) {
case 'r':
feature.fi_flags |= ZFEATURE_FLAG_READONLY_COMPAT;
break;
case 'd':
desc = strdup(optarg);
break;
default:
usage();
break;
}
}
if (desc == NULL)
desc = strdup("zhack injected");
feature.fi_desc = desc;
argc -= optind;
argv += optind;
if (argc < 2) {
(void) fprintf(stderr, "error: missing feature or pool name\n");
usage();
}
target = argv[0];
feature.fi_guid = argv[1];
if (!zfeature_is_valid_guid(feature.fi_guid))
fatal(NULL, FTAG, "invalid feature guid: %s", feature.fi_guid);
zhack_spa_open(target, B_FALSE, FTAG, &spa);
mos = spa->spa_meta_objset;
if (zfeature_is_supported(feature.fi_guid))
fatal(spa, FTAG, "'%s' is a real feature, will not enable",
feature.fi_guid);
if (0 == zap_contains(mos, spa->spa_feat_desc_obj, feature.fi_guid))
fatal(spa, FTAG, "feature already enabled: %s",
feature.fi_guid);
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
zhack_feature_enable_sync, &feature, 5, ZFS_SPACE_CHECK_NORMAL));
spa_close(spa, FTAG);
free(desc);
}
static void
feature_incr_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
uint64_t refcount;
VERIFY0(feature_get_refcount_from_disk(spa, feature, &refcount));
feature_sync(spa, feature, refcount + 1, tx);
spa_history_log_internal(spa, "zhack feature incr", tx,
"name=%s", feature->fi_guid);
}
static void
feature_decr_sync(void *arg, dmu_tx_t *tx)
{
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
zfeature_info_t *feature = arg;
uint64_t refcount;
VERIFY0(feature_get_refcount_from_disk(spa, feature, &refcount));
feature_sync(spa, feature, refcount - 1, tx);
spa_history_log_internal(spa, "zhack feature decr", tx,
"name=%s", feature->fi_guid);
}
static void
zhack_do_feature_ref(int argc, char **argv)
{
int c;
char *target;
boolean_t decr = B_FALSE;
spa_t *spa;
objset_t *mos;
zfeature_info_t feature;
spa_feature_t nodeps[] = { SPA_FEATURE_NONE };
/*
* fi_desc does not matter here because it was written to disk
* when the feature was enabled, but we need to properly set the
* feature for read or write based on the information we read off
* disk later.
*/
feature.fi_uname = "zhack";
feature.fi_flags = 0;
feature.fi_desc = NULL;
feature.fi_depends = nodeps;
feature.fi_feature = SPA_FEATURE_NONE;
optind = 1;
while ((c = getopt(argc, argv, "+md")) != -1) {
switch (c) {
case 'm':
feature.fi_flags |= ZFEATURE_FLAG_MOS;
break;
case 'd':
decr = B_TRUE;
break;
default:
usage();
break;
}
}
argc -= optind;
argv += optind;
if (argc < 2) {
(void) fprintf(stderr, "error: missing feature or pool name\n");
usage();
}
target = argv[0];
feature.fi_guid = argv[1];
if (!zfeature_is_valid_guid(feature.fi_guid))
fatal(NULL, FTAG, "invalid feature guid: %s", feature.fi_guid);
zhack_spa_open(target, B_FALSE, FTAG, &spa);
mos = spa->spa_meta_objset;
if (zfeature_is_supported(feature.fi_guid)) {
fatal(spa, FTAG,
"'%s' is a real feature, will not change refcount",
feature.fi_guid);
}
if (0 == zap_contains(mos, spa->spa_feat_for_read_obj,
feature.fi_guid)) {
feature.fi_flags &= ~ZFEATURE_FLAG_READONLY_COMPAT;
} else if (0 == zap_contains(mos, spa->spa_feat_for_write_obj,
feature.fi_guid)) {
feature.fi_flags |= ZFEATURE_FLAG_READONLY_COMPAT;
} else {
fatal(spa, FTAG, "feature is not enabled: %s", feature.fi_guid);
}
if (decr) {
uint64_t count;
if (feature_get_refcount_from_disk(spa, &feature,
&count) == 0 && count == 0) {
fatal(spa, FTAG, "feature refcount already 0: %s",
feature.fi_guid);
}
}
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
decr ? feature_decr_sync : feature_incr_sync, &feature,
5, ZFS_SPACE_CHECK_NORMAL));
spa_close(spa, FTAG);
}
static int
zhack_do_feature(int argc, char **argv)
{
char *subcommand;
argc--;
argv++;
if (argc == 0) {
(void) fprintf(stderr,
"error: no feature operation specified\n");
usage();
}
subcommand = argv[0];
if (strcmp(subcommand, "stat") == 0) {
zhack_do_feature_stat(argc, argv);
} else if (strcmp(subcommand, "enable") == 0) {
zhack_do_feature_enable(argc, argv);
} else if (strcmp(subcommand, "ref") == 0) {
zhack_do_feature_ref(argc, argv);
} else {
(void) fprintf(stderr, "error: unknown subcommand: %s\n",
subcommand);
usage();
}
return (0);
}
static int
zhack_repair_label_cksum(int argc, char **argv)
{
zio_checksum_info_t *ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
const char *cfg_keys[] = { ZPOOL_CONFIG_VERSION,
ZPOOL_CONFIG_POOL_STATE, ZPOOL_CONFIG_GUID };
boolean_t labels_repaired[VDEV_LABELS];
boolean_t repaired = B_FALSE;
vdev_label_t labels[VDEV_LABELS];
struct stat st;
int fd;
bzero(labels_repaired, sizeof (labels_repaired));
bzero(labels, sizeof (labels));
abd_init();
argc -= 1;
argv += 1;
if (argc < 1) {
(void) fprintf(stderr, "error: missing device\n");
usage();
}
if ((fd = open(argv[0], O_RDWR)) == -1)
fatal(NULL, FTAG, "cannot open '%s': %s", argv[0],
strerror(errno));
if (stat(argv[0], &st) != 0)
fatal(NULL, FTAG, "cannot stat '%s': %s", argv[0],
strerror(errno));
for (int l = 0; l < VDEV_LABELS; l++) {
uint64_t label_offset, offset;
zio_cksum_t expected_cksum;
zio_cksum_t actual_cksum;
zio_cksum_t verifier;
zio_eck_t *eck;
nvlist_t *cfg;
int byteswap;
uint64_t val;
ssize_t err;
vdev_label_t *vl = &labels[l];
label_offset = vdev_label_offset(st.st_size, l, 0);
err = pread64(fd, vl, sizeof (vdev_label_t), label_offset);
if (err == -1) {
(void) fprintf(stderr, "error: cannot read "
"label %d: %s\n", l, strerror(errno));
continue;
} else if (err != sizeof (vdev_label_t)) {
(void) fprintf(stderr, "error: bad label %d read size "
"\n", l);
continue;
}
err = nvlist_unpack(vl->vl_vdev_phys.vp_nvlist,
VDEV_PHYS_SIZE - sizeof (zio_eck_t), &cfg, 0);
if (err) {
(void) fprintf(stderr, "error: cannot unpack nvlist "
"label %d\n", l);
continue;
}
for (int i = 0; i < ARRAY_SIZE(cfg_keys); i++) {
err = nvlist_lookup_uint64(cfg, cfg_keys[i], &val);
if (err) {
(void) fprintf(stderr, "error: label %d: "
"cannot find nvlist key %s\n",
l, cfg_keys[i]);
continue;
}
}
void *data = (char *)vl + offsetof(vdev_label_t, vl_vdev_phys);
eck = (zio_eck_t *)((char *)(data) + VDEV_PHYS_SIZE) - 1;
offset = label_offset + offsetof(vdev_label_t, vl_vdev_phys);
ZIO_SET_CHECKSUM(&verifier, offset, 0, 0, 0);
byteswap = (eck->zec_magic == BSWAP_64(ZEC_MAGIC));
if (byteswap)
byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
expected_cksum = eck->zec_cksum;
eck->zec_cksum = verifier;
abd_t *abd = abd_get_from_buf(data, VDEV_PHYS_SIZE);
ci->ci_func[byteswap](abd, VDEV_PHYS_SIZE, NULL, &actual_cksum);
abd_free(abd);
if (byteswap)
byteswap_uint64_array(&expected_cksum,
sizeof (zio_cksum_t));
if (ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
continue;
eck->zec_cksum = actual_cksum;
err = pwrite64(fd, data, VDEV_PHYS_SIZE, offset);
if (err == -1) {
(void) fprintf(stderr, "error: cannot write "
"label %d: %s\n", l, strerror(errno));
continue;
} else if (err != VDEV_PHYS_SIZE) {
(void) fprintf(stderr, "error: bad write size "
"label %d\n", l);
continue;
}
fsync(fd);
labels_repaired[l] = B_TRUE;
}
close(fd);
abd_fini();
for (int l = 0; l < VDEV_LABELS; l++) {
(void) printf("label %d: %s\n", l,
labels_repaired[l] ? "repaired" : "skipped");
repaired |= labels_repaired[l];
}
if (repaired)
return (0);
return (1);
}
static int
zhack_do_label(int argc, char **argv)
{
char *subcommand;
int err;
argc--;
argv++;
if (argc == 0) {
(void) fprintf(stderr,
"error: no label operation specified\n");
usage();
}
subcommand = argv[0];
if (strcmp(subcommand, "repair") == 0) {
err = zhack_repair_label_cksum(argc, argv);
} else {
(void) fprintf(stderr, "error: unknown subcommand: %s\n",
subcommand);
usage();
}
return (err);
}
#define MAX_NUM_PATHS 1024
int
main(int argc, char **argv)
{
extern void zfs_prop_init(void);
char *path[MAX_NUM_PATHS];
const char *subcommand;
int rv = 0;
int c;
g_importargs.path = path;
dprintf_setup(&argc, argv);
zfs_prop_init();
while ((c = getopt(argc, argv, "+c:d:")) != -1) {
switch (c) {
case 'c':
g_importargs.cachefile = optarg;
break;
case 'd':
assert(g_importargs.paths < MAX_NUM_PATHS);
g_importargs.path[g_importargs.paths++] = optarg;
break;
default:
usage();
break;
}
}
argc -= optind;
argv += optind;
optind = 1;
if (argc == 0) {
(void) fprintf(stderr, "error: no command specified\n");
usage();
}
subcommand = argv[0];
if (strcmp(subcommand, "feature") == 0) {
rv = zhack_do_feature(argc, argv);
} else if (strcmp(subcommand, "label") == 0) {
return (zhack_do_label(argc, argv));
} else {
(void) fprintf(stderr, "error: unknown subcommand: %s\n",
subcommand);
usage();
}
if (!g_readonly && spa_export(g_pool, NULL, B_TRUE, B_FALSE) != 0) {
fatal(NULL, FTAG, "pool export failed; "
"changes may not be committed to disk\n");
}
kernel_fini();
return (rv);
}