* Copyright (c) 2021 Allan Jude
* Copyright (c) 2021 Toomas Soome <tsoome@me.com>
* Copyright (c) 2023, Klara Inc.
+ * Copyright (c) 2023, Rob Norris <robn@despairlabs.com>
*/
#include <stdio.h>
#include <sys/dsl_scan.h>
#include <sys/btree.h>
#include <sys/brt.h>
+#include <sys/brt_impl.h>
#include <zfs_comutil.h>
#include <sys/zstd/zstd.h>
#include <libnvpair.h>
#include <libzutil.h>
+#include <libzdb.h>
+
#include "zdb.h"
-#define ZDB_COMPRESS_NAME(idx) ((idx) < ZIO_COMPRESS_FUNCTIONS ? \
- zio_compress_table[(idx)].ci_name : "UNKNOWN")
-#define ZDB_CHECKSUM_NAME(idx) ((idx) < ZIO_CHECKSUM_FUNCTIONS ? \
- zio_checksum_table[(idx)].ci_name : "UNKNOWN")
-#define ZDB_OT_TYPE(idx) ((idx) < DMU_OT_NUMTYPES ? (idx) : \
- (idx) == DMU_OTN_ZAP_DATA || (idx) == DMU_OTN_ZAP_METADATA ? \
- DMU_OT_ZAP_OTHER : \
- (idx) == DMU_OTN_UINT64_DATA || (idx) == DMU_OTN_UINT64_METADATA ? \
- DMU_OT_UINT64_OTHER : DMU_OT_NUMTYPES)
-
-/* Some platforms require part of inode IDs to be remapped */
-#ifdef __APPLE__
-#define ZDB_MAP_OBJECT_ID(obj) INO_XNUTOZFS(obj, 2)
-#else
-#define ZDB_MAP_OBJECT_ID(obj) (obj)
-#endif
-
-static const char *
-zdb_ot_name(dmu_object_type_t type)
-{
- if (type < DMU_OT_NUMTYPES)
- return (dmu_ot[type].ot_name);
- else if ((type & DMU_OT_NEWTYPE) &&
- ((type & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS))
- return (dmu_ot_byteswap[type & DMU_OT_BYTESWAP_MASK].ob_name);
- else
- return ("UNKNOWN");
-}
extern int reference_tracking_enable;
extern int zfs_recover;
static uint64_t *zopt_metaslab = NULL;
static unsigned zopt_metaslab_args = 0;
-typedef struct zopt_object_range {
- uint64_t zor_obj_start;
- uint64_t zor_obj_end;
- uint64_t zor_flags;
-} zopt_object_range_t;
static zopt_object_range_t *zopt_object_ranges = NULL;
static unsigned zopt_object_args = 0;
static int flagbits[256];
-#define ZOR_FLAG_PLAIN_FILE 0x0001
-#define ZOR_FLAG_DIRECTORY 0x0002
-#define ZOR_FLAG_SPACE_MAP 0x0004
-#define ZOR_FLAG_ZAP 0x0008
-#define ZOR_FLAG_ALL_TYPES -1
-#define ZOR_SUPPORTED_FLAGS (ZOR_FLAG_PLAIN_FILE | \
- ZOR_FLAG_DIRECTORY | \
- ZOR_FLAG_SPACE_MAP | \
- ZOR_FLAG_ZAP)
-
-#define ZDB_FLAG_CHECKSUM 0x0001
-#define ZDB_FLAG_DECOMPRESS 0x0002
-#define ZDB_FLAG_BSWAP 0x0004
-#define ZDB_FLAG_GBH 0x0008
-#define ZDB_FLAG_INDIRECT 0x0010
-#define ZDB_FLAG_RAW 0x0020
-#define ZDB_FLAG_PRINT_BLKPTR 0x0040
-#define ZDB_FLAG_VERBOSE 0x0080
static uint64_t max_inflight_bytes = 256 * 1024 * 1024; /* 256MB */
static int leaked_objects = 0;
static int dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t free,
dmu_tx_t *tx);
-typedef struct sublivelist_verify {
- /* FREE's that haven't yet matched to an ALLOC, in one sub-livelist */
- zfs_btree_t sv_pair;
-
- /* ALLOC's without a matching FREE, accumulates across sub-livelists */
- zfs_btree_t sv_leftover;
-} sublivelist_verify_t;
-
-static int
-livelist_compare(const void *larg, const void *rarg)
-{
- const blkptr_t *l = larg;
- const blkptr_t *r = rarg;
-
- /* Sort them according to dva[0] */
- uint64_t l_dva0_vdev, r_dva0_vdev;
- l_dva0_vdev = DVA_GET_VDEV(&l->blk_dva[0]);
- r_dva0_vdev = DVA_GET_VDEV(&r->blk_dva[0]);
- if (l_dva0_vdev < r_dva0_vdev)
- return (-1);
- else if (l_dva0_vdev > r_dva0_vdev)
- return (+1);
-
- /* if vdevs are equal, sort by offsets. */
- uint64_t l_dva0_offset;
- uint64_t r_dva0_offset;
- l_dva0_offset = DVA_GET_OFFSET(&l->blk_dva[0]);
- r_dva0_offset = DVA_GET_OFFSET(&r->blk_dva[0]);
- if (l_dva0_offset < r_dva0_offset) {
- return (-1);
- } else if (l_dva0_offset > r_dva0_offset) {
- return (+1);
- }
-
- /*
- * Since we're storing blkptrs without cancelling FREE/ALLOC pairs,
- * it's possible the offsets are equal. In that case, sort by txg
- */
- if (l->blk_birth < r->blk_birth) {
- return (-1);
- } else if (l->blk_birth > r->blk_birth) {
- return (+1);
- }
- return (0);
-}
-typedef struct sublivelist_verify_block {
- dva_t svb_dva;
-
- /*
- * We need this to check if the block marked as allocated
- * in the livelist was freed (and potentially reallocated)
- * in the metaslab spacemaps at a later TXG.
- */
- uint64_t svb_allocated_txg;
-} sublivelist_verify_block_t;
static void zdb_print_blkptr(const blkptr_t *bp, int flags);
"don't print label contents\n");
(void) fprintf(stderr, " -t --txg=INTEGER "
"highest txg to use when searching for uberblocks\n");
+ (void) fprintf(stderr, " -T --brt-stats "
+ "BRT statistics\n");
(void) fprintf(stderr, " -u --uberblock "
"uberblock\n");
(void) fprintf(stderr, " -U --cachefile=PATH "
nicenum(num, buf, buflen);
}
+static void
+zdb_nicebytes(uint64_t bytes, char *buf, size_t buflen)
+{
+ if (dump_opt['P'])
+ (void) snprintf(buf, buflen, "%llu", (longlong_t)bytes);
+ else
+ zfs_nicebytes(bytes, buf, buflen);
+}
+
static const char histo_stars[] = "****************************************";
static const uint64_t histo_width = sizeof (histo_stars) - 1;
dump_dedup_ratio(&dds_total);
}
+static void
+dump_brt(spa_t *spa)
+{
+ if (!spa_feature_is_enabled(spa, SPA_FEATURE_BLOCK_CLONING)) {
+ printf("BRT: unsupported on this pool\n");
+ return;
+ }
+
+ if (!spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
+ printf("BRT: empty\n");
+ return;
+ }
+
+ brt_t *brt = spa->spa_brt;
+ VERIFY(brt);
+
+ char count[32], used[32], saved[32];
+ zdb_nicebytes(brt_get_used(spa), used, sizeof (used));
+ zdb_nicebytes(brt_get_saved(spa), saved, sizeof (saved));
+ uint64_t ratio = brt_get_ratio(spa);
+ printf("BRT: used %s; saved %s; ratio %llu.%02llux\n", used, saved,
+ (u_longlong_t)(ratio / 100), (u_longlong_t)(ratio % 100));
+
+ if (dump_opt['T'] < 2)
+ return;
+
+ for (uint64_t vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) {
+ brt_vdev_t *brtvd = &brt->brt_vdevs[vdevid];
+ if (brtvd == NULL)
+ continue;
+
+ if (!brtvd->bv_initiated) {
+ printf("BRT: vdev %" PRIu64 ": empty\n", vdevid);
+ continue;
+ }
+
+ zdb_nicenum(brtvd->bv_totalcount, count, sizeof (count));
+ zdb_nicebytes(brtvd->bv_usedspace, used, sizeof (used));
+ zdb_nicebytes(brtvd->bv_savedspace, saved, sizeof (saved));
+ printf("BRT: vdev %" PRIu64 ": refcnt %s; used %s; saved %s\n",
+ vdevid, count, used, saved);
+ }
+
+ if (dump_opt['T'] < 3)
+ return;
+
+ char dva[64];
+ printf("\n%-16s %-10s\n", "DVA", "REFCNT");
+
+ for (uint64_t vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) {
+ brt_vdev_t *brtvd = &brt->brt_vdevs[vdevid];
+ if (brtvd == NULL || !brtvd->bv_initiated)
+ continue;
+
+ zap_cursor_t zc;
+ zap_attribute_t za;
+ for (zap_cursor_init(&zc, brt->brt_mos, brtvd->bv_mos_entries);
+ zap_cursor_retrieve(&zc, &za) == 0;
+ zap_cursor_advance(&zc)) {
+ uint64_t offset = *(uint64_t *)za.za_name;
+ uint64_t refcnt = za.za_first_integer;
+
+ snprintf(dva, sizeof (dva), "%" PRIu64 ":%llx", vdevid,
+ (u_longlong_t)offset);
+ printf("%-16s %-10llu\n", dva, (u_longlong_t)refcnt);
+ }
+ zap_cursor_fini(&zc);
+ }
+}
+
static void
dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
{
snprintf_zstd_header(spa_t *spa, char *blkbuf, size_t buflen,
const blkptr_t *bp)
{
- abd_t *pabd;
+ static abd_t *pabd = NULL;
void *buf;
zio_t *zio;
zfs_zstdhdr_t zstd_hdr;
return;
}
- pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
+ if (!pabd)
+ pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
zio = zio_root(spa, NULL, NULL, 0);
/* Decrypt but don't decompress so we can read the compression header */
}
typedef struct zdb_ddt_entry {
+ /* key must be first for ddt_key_compare */
ddt_key_t zdde_key;
uint64_t zdde_ref_blocks;
uint64_t zdde_ref_lsize;
ddt_histogram_t ddh_total = {{{0}}};
ddt_stat_t dds_total = {0};
- avl_create(&t, ddt_entry_compare,
+ avl_create(&t, ddt_key_compare,
sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
}
}
+ if (spa->spa_brt != NULL) {
+ brt_t *brt = spa->spa_brt;
+ for (uint64_t vdevid = 0; vdevid < brt->brt_nvdevs; vdevid++) {
+ brt_vdev_t *brtvd = &brt->brt_vdevs[vdevid];
+ if (brtvd != NULL && brtvd->bv_initiated) {
+ mos_obj_refd(brtvd->bv_mos_brtvdev);
+ mos_obj_refd(brtvd->bv_mos_entries);
+ }
+ }
+ }
+
/*
* Visit all allocated objects and make sure they are referenced.
*/
if (dump_opt['D'])
dump_all_ddts(spa);
+ if (dump_opt['T'])
+ dump_brt(spa);
+
if (dump_opt['d'] > 2 || dump_opt['m'])
dump_metaslabs(spa);
if (dump_opt['M'])
#define ZIO_COMPRESS_MASK(alg) (1ULL << (ZIO_COMPRESS_##alg))
static boolean_t
+try_decompress_block(abd_t *pabd, uint64_t lsize, uint64_t psize,
+ int flags, int cfunc, void *lbuf, void *lbuf2)
+{
+ if (flags & ZDB_FLAG_VERBOSE) {
+ (void) fprintf(stderr,
+ "Trying %05llx -> %05llx (%s)\n",
+ (u_longlong_t)psize,
+ (u_longlong_t)lsize,
+ zio_compress_table[cfunc].ci_name);
+ }
+
+ /*
+ * We set lbuf to all zeros and lbuf2 to all
+ * ones, then decompress to both buffers and
+ * compare their contents. This way we can
+ * know if decompression filled exactly to
+ * lsize or if it left some bytes unwritten.
+ */
+
+ memset(lbuf, 0x00, lsize);
+ memset(lbuf2, 0xff, lsize);
+
+ if (zio_decompress_data(cfunc, pabd,
+ lbuf, psize, lsize, NULL) == 0 &&
+ zio_decompress_data(cfunc, pabd,
+ lbuf2, psize, lsize, NULL) == 0 &&
+ memcmp(lbuf, lbuf2, lsize) == 0)
+ return (B_TRUE);
+ return (B_FALSE);
+}
+
+static uint64_t
zdb_decompress_block(abd_t *pabd, void *buf, void *lbuf, uint64_t lsize,
uint64_t psize, int flags)
{
(void) buf;
- boolean_t exceeded = B_FALSE;
+ uint64_t orig_lsize = lsize;
+ boolean_t tryzle = ((getenv("ZDB_NO_ZLE") == NULL));
+ boolean_t found = B_FALSE;
/*
* We don't know how the data was compressed, so just try
* every decompress function at every inflated blocksize.
uint64_t maxlsize = SPA_MAXBLOCKSIZE;
uint64_t mask = ZIO_COMPRESS_MASK(ON) | ZIO_COMPRESS_MASK(OFF) |
ZIO_COMPRESS_MASK(INHERIT) | ZIO_COMPRESS_MASK(EMPTY) |
- (getenv("ZDB_NO_ZLE") ? ZIO_COMPRESS_MASK(ZLE) : 0);
+ ZIO_COMPRESS_MASK(ZLE);
*cfuncp++ = ZIO_COMPRESS_LZ4;
*cfuncp++ = ZIO_COMPRESS_LZJB;
mask |= ZIO_COMPRESS_MASK(LZ4) | ZIO_COMPRESS_MASK(LZJB);
+ /*
+ * Every gzip level has the same decompressor, no need to
+ * run it 9 times per bruteforce attempt.
+ */
+ mask |= ZIO_COMPRESS_MASK(GZIP_2) | ZIO_COMPRESS_MASK(GZIP_3);
+ mask |= ZIO_COMPRESS_MASK(GZIP_4) | ZIO_COMPRESS_MASK(GZIP_5);
+ mask |= ZIO_COMPRESS_MASK(GZIP_6) | ZIO_COMPRESS_MASK(GZIP_7);
+ mask |= ZIO_COMPRESS_MASK(GZIP_8) | ZIO_COMPRESS_MASK(GZIP_9);
for (int c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++)
if (((1ULL << c) & mask) == 0)
*cfuncp++ = c;
lsize += SPA_MINBLOCKSIZE;
else
maxlsize = lsize;
+
for (; lsize <= maxlsize; lsize += SPA_MINBLOCKSIZE) {
for (cfuncp = cfuncs; *cfuncp; cfuncp++) {
- if (flags & ZDB_FLAG_VERBOSE) {
- (void) fprintf(stderr,
- "Trying %05llx -> %05llx (%s)\n",
- (u_longlong_t)psize,
- (u_longlong_t)lsize,
- zio_compress_table[*cfuncp].\
- ci_name);
- }
-
- /*
- * We randomize lbuf2, and decompress to both
- * lbuf and lbuf2. This way, we will know if
- * decompression fill exactly to lsize.
- */
- VERIFY0(random_get_pseudo_bytes(lbuf2, lsize));
-
- if (zio_decompress_data(*cfuncp, pabd,
- lbuf, psize, lsize, NULL) == 0 &&
- zio_decompress_data(*cfuncp, pabd,
- lbuf2, psize, lsize, NULL) == 0 &&
- memcmp(lbuf, lbuf2, lsize) == 0)
+ if (try_decompress_block(pabd, lsize, psize, flags,
+ *cfuncp, lbuf, lbuf2)) {
+ found = B_TRUE;
break;
+ }
}
if (*cfuncp != 0)
break;
}
+ if (!found && tryzle) {
+ for (lsize = orig_lsize; lsize <= maxlsize;
+ lsize += SPA_MINBLOCKSIZE) {
+ if (try_decompress_block(pabd, lsize, psize, flags,
+ ZIO_COMPRESS_ZLE, lbuf, lbuf2)) {
+ *cfuncp = ZIO_COMPRESS_ZLE;
+ found = B_TRUE;
+ break;
+ }
+ }
+ }
umem_free(lbuf2, SPA_MAXBLOCKSIZE);
- if (lsize > maxlsize) {
- exceeded = B_TRUE;
- }
if (*cfuncp == ZIO_COMPRESS_ZLE) {
printf("\nZLE decompression was selected. If you "
"suspect the results are wrong,\ntry avoiding ZLE "
"by setting and exporting ZDB_NO_ZLE=\"true\"\n");
}
- return (exceeded);
+ return (lsize > maxlsize ? -1 : lsize);
}
/*
uint64_t orig_lsize = lsize;
buf = lbuf;
if (flags & ZDB_FLAG_DECOMPRESS) {
- boolean_t failed = zdb_decompress_block(pabd, buf, lbuf,
+ lsize = zdb_decompress_block(pabd, buf, lbuf,
lsize, psize, flags);
- if (failed) {
+ if (lsize == -1) {
(void) printf("Decompress of %s failed\n", thing);
goto out;
}
abd_return_buf_copy(pabd, buf, lsize);
borrowed = B_FALSE;
buf = lbuf;
- boolean_t failed = zdb_decompress_block(pabd, buf,
+ lsize = zdb_decompress_block(pabd, buf,
lbuf, lsize, psize, flags);
b = (const blkptr_t *)(void *)
((uintptr_t)buf + (uintptr_t)blkptr_offset);
- if (failed || zfs_blkptr_verify(spa, b,
+ if (lsize == -1 || zfs_blkptr_verify(spa, b,
BLK_CONFIG_NEEDED, BLK_VERIFY_LOG) == B_FALSE) {
printf("invalid block pointer at this DVA\n");
goto out;
{"io-stats", no_argument, NULL, 's'},
{"simulate-dedup", no_argument, NULL, 'S'},
{"txg", required_argument, NULL, 't'},
+ {"brt-stats", no_argument, NULL, 'T'},
{"uberblock", no_argument, NULL, 'u'},
{"cachefile", required_argument, NULL, 'U'},
{"verbose", no_argument, NULL, 'v'},
};
while ((c = getopt_long(argc, argv,
- "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:uU:vVx:XYyZ",
+ "AbBcCdDeEFGhiI:kK:lLmMNo:Op:PqrRsSt:TuU:vVx:XYyZ",
long_options, NULL)) != -1) {
switch (c) {
case 'b':
case 'R':
case 's':
case 'S':
+ case 'T':
case 'u':
case 'y':
case 'Z':
if (dump_opt['l'])
return (dump_label(argv[0]));
- if (dump_opt['O']) {
- if (argc != 2)
- usage();
- dump_opt['v'] = verbose + 3;
- return (dump_path(argv[0], argv[1], NULL));
- }
- if (dump_opt['r']) {
- target_is_spa = B_FALSE;
- if (argc != 3)
- usage();
- dump_opt['v'] = verbose;
- error = dump_path(argv[0], argv[1], &object);
- if (error != 0)
- fatal("internal error: %s", strerror(error));
- }
-
if (dump_opt['X'] || dump_opt['F'])
rewind = ZPOOL_DO_REWIND |
(dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
searchdirs = NULL;
}
+ /*
+ * We need to make sure to process -O option or call
+ * dump_path after the -e option has been processed,
+ * which imports the pool to the namespace if it's
+ * not in the cachefile.
+ */
+ if (dump_opt['O']) {
+ if (argc != 2)
+ usage();
+ dump_opt['v'] = verbose + 3;
+ return (dump_path(argv[0], argv[1], NULL));
+ }
+
+ if (dump_opt['r']) {
+ target_is_spa = B_FALSE;
+ if (argc != 3)
+ usage();
+ dump_opt['v'] = verbose;
+ error = dump_path(argv[0], argv[1], &object);
+ if (error != 0)
+ fatal("internal error: %s", strerror(error));
+ }
+
/*
* import_checkpointed_state makes the assumption that the
* target pool that we pass it is already part of the spa
projects / mirror_zfs.git / blobdiff