Page MenuHomeFreeBSD
Differential D25512 Diff 74776 stand/libsa/zfs/zfsimpl.c

Changeset View

Standalone View

View Options

stand/libsa/zfs/zfsimpl.c

Show All 25 Lines
#include <sys/cdefs.h> #include <sys/cdefs.h>
__FBSDID("$FreeBSD$"); __FBSDID("$FreeBSD$");
/* /*
* Stand-alone ZFS file reader. * Stand-alone ZFS file reader.
*/ */
#include <stdbool.h>
#include <sys/endian.h> #include <sys/endian.h>
#include <sys/stat.h> #include <sys/stat.h>
#include <sys/stdint.h> #include <sys/stdint.h>
#include <sys/list.h> #include <sys/list.h>
#include <machine/_inttypes.h> #include <machine/_inttypes.h>
#include "zfsimpl.h" #include "zfsimpl.h"
#include "zfssubr.c" #include "zfssubr.c"
▲ Show 20 Lines • Show All 172 Lines • ▼ Show 20 Lines
static int static int
vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf, vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
off_t offset, size_t size) off_t offset, size_t size)
{ {
size_t psize; size_t psize;
int rc; int rc;
if (!vdev->v_phys_read) if (vdev->v_phys_read == NULL)
return (EIO); return (ENOTSUP);
if (bp) { if (bp) {
psize = BP_GET_PSIZE(bp); psize = BP_GET_PSIZE(bp);
} else { } else {
psize = size; psize = size;
} }
rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize); rc = vdev->v_phys_read(vdev, vdev->v_priv, offset, buf, psize);
if (rc == 0) { if (rc == 0) {
if (bp != NULL) if (bp != NULL)
rc = zio_checksum_verify(vdev->v_spa, bp, buf); rc = zio_checksum_verify(vdev->v_spa, bp, buf);
} }
return (rc); return (rc);
} }
static int
vdev_write_phys(vdev_t *vdev, void *buf, off_t offset, size_t size)
{
if (vdev->v_phys_write == NULL)
return (ENOTSUP);
return (vdev->v_phys_write(vdev, offset, buf, size));
}
typedef struct remap_segment { typedef struct remap_segment {
vdev_t *rs_vd; vdev_t *rs_vd;
uint64_t rs_offset; uint64_t rs_offset;
uint64_t rs_asize; uint64_t rs_asize;
uint64_t rs_split_offset; uint64_t rs_split_offset;
list_node_t rs_node; list_node_t rs_node;
} remap_segment_t; } remap_segment_t;
▲ Show 20 Lines • Show All 1,084 Lines • ▼ Show 20 Linesspa_find_by_name(const char *name)
STAILQ_FOREACH(spa, &zfs_pools, spa_link) STAILQ_FOREACH(spa, &zfs_pools, spa_link)
if (strcmp(spa->spa_name, name) == 0) if (strcmp(spa->spa_name, name) == 0)
return (spa); return (spa);
return (NULL); return (NULL);
} }
static spa_t * static spa_t *
spa_find_by_dev(struct zfs_devdesc *dev)
{
if (dev->dd.d_dev->dv_type != DEVT_ZFS)
return (NULL);
if (dev->pool_guid == 0)
return (STAILQ_FIRST(&zfs_pools));
return (spa_find_by_guid(dev->pool_guid));
}
static spa_t *
spa_create(uint64_t guid, const char *name) spa_create(uint64_t guid, const char *name)
{ {
spa_t *spa; spa_t *spa;
if ((spa = calloc(1, sizeof(spa_t))) == NULL) if ((spa = calloc(1, sizeof(spa_t))) == NULL)
return (NULL); return (NULL);
if ((spa->spa_name = strdup(name)) == NULL) { if ((spa->spa_name = strdup(name)) == NULL) {
free(spa); free(spa);
▲ Show 20 Lines • Show All 233 Lines • ▼ Show 20 Linesvdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL); BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF); BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
DVA_SET_OFFSET(BP_IDENTITY(&bp), off); DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0); ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
return (vdev_read_phys(vd, &bp, buf, off, size)); return (vdev_read_phys(vd, &bp, buf, off, size));
} }
/*
* We do need to be sure we write to correct location.
* Our vdev label does consist of 4 fields:
* pad1 (8k), reserved.
* bootenv (8k), checksummed, previously reserved, may contian garbage.
* vdev_phys (112k), checksummed
* uberblock ring (128k), checksummed.
*
* Since bootenv area may contain garbage, we can not reliably read it, as
* we can get checksum errors.
* Next best thing is vdev_phys - it is just after bootenv. It still may
* be corrupted, but in such case we will miss this one write.
*/
static int
vdev_label_write_validate(vdev_t *vd, int l, uint64_t offset)
{
uint64_t off, o_phys;
void *buf;
size_t size = VDEV_PHYS_SIZE;
int rc;
o_phys = offsetof(vdev_label_t, vl_vdev_phys);
off = vdev_label_offset(vd->v_psize, l, o_phys);
/* off should be 8K from bootenv */
if (vdev_label_offset(vd->v_psize, l, offset) + VDEV_PAD_SIZE != off)
return (EINVAL);
buf = malloc(size);
if (buf == NULL)
return (ENOMEM);
/* Read vdev_phys */
rc = vdev_label_read(vd, l, buf, o_phys, size);
free(buf);
return (rc);
}
static int
vdev_label_write(vdev_t *vd, int l, vdev_boot_envblock_t *be, uint64_t offset)
{
zio_checksum_info_t *ci;
zio_cksum_t cksum;
off_t off;
size_t size = VDEV_PAD_SIZE;
int rc;
if (vd->v_phys_write == NULL)
return (ENOTSUP);
off = vdev_label_offset(vd->v_psize, l, offset);
rc = vdev_label_write_validate(vd, l, offset);
if (rc != 0) {
return (rc);
}
ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
be->vbe_zbt.zec_magic = ZEC_MAGIC;
zio_checksum_label_verifier(&be->vbe_zbt.zec_cksum, off);
ci->ci_func[0](be, size, NULL, &cksum);
be->vbe_zbt.zec_cksum = cksum;
glebiusUnsubmitted
Done
Inline Actions

If I read code correct this message will print once again the failure message from vdev_label_write(), except it doesn't check for NULL v_name.

glebius: If I read code correct this message will print once again the failure message from…
return (vdev_write_phys(vd, be, off, size));
}
static int
vdev_write_bootenv_impl(vdev_t *vdev, vdev_boot_envblock_t *be)
{
vdev_t *kid;
int rv = 0, rc;
STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
if (kid->v_state != VDEV_STATE_HEALTHY)
continue;
rc = vdev_write_bootenv_impl(kid, be);
if (rv == 0)
rv = rc;
}
/*
* Non-leaf vdevs do not have v_phys_write.
*/
if (vdev->v_phys_write == NULL)
return (rv);
for (int l = 0; l < VDEV_LABELS; l++) {
rc = vdev_label_write(vdev, l, be,
offsetof(vdev_label_t, vl_be));
if (rc != 0) {
printf("failed to write bootenv to %s label %d: %d\n",
vdev->v_name ? vdev->v_name : "unknown", l, rc);
rv = rc;
}
}
return (rv);
}
int
vdev_write_bootenv(vdev_t *vdev, nvlist_t *nvl)
{
vdev_boot_envblock_t *be;
nvlist_t nv;
int rv;
if (nvl->nv_size > sizeof(be->vbe_nvlist))
return (E2BIG);
be = calloc(1, sizeof(*be));
if (be == NULL)
return (ENOMEM);
nv.nv_header = nvl->nv_header;
nv.nv_asize = nvl->nv_asize;
nv.nv_size = nvl->nv_size;
*(nvs_header_t *)be->vbe_nvlist = nv.nv_header;
nv.nv_data = be->vbe_nvlist + sizeof(nvs_header_t);
bcopy(nvl->nv_data, nv.nv_data, nv.nv_size);
rv = nvlist_export(&nv);
if (rv == 0) {
rv = vdev_write_bootenv_impl(vdev, be);
}
free(be);
return (rv);
}
/*
* Read the bootenv area from pool label, return the nvlist from it.
* We return from first successful read.
*/
nvlist_t *
vdev_read_bootenv(vdev_t *vdev)
{
vdev_t *kid;
nvlist_t *benv;
vdev_boot_envblock_t *be;
int rv;
STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
if (kid->v_state != VDEV_STATE_HEALTHY)
continue;
benv = vdev_read_bootenv(kid);
if (benv != NULL)
return (benv);
}
be = malloc(sizeof (*be));
if (be == NULL)
return (NULL);
rv = 0;
for (int l = 0; l < VDEV_LABELS; l++) {
rv = vdev_label_read(vdev, l, be,
glebiusUnsubmitted
Done
Inline Actions
nvlist_add_string(benv, "command", command);
}
+ free(be);
return (benv);
}
static uint64_t

We leak be here.

glebius: We leak be here.
offsetof(vdev_label_t, vl_be),
sizeof (*be));
if (rv == 0)
break;
}
if (rv != 0) {
free(be);
return (NULL);
}
benv = nvlist_import(be->vbe_nvlist + 4, be->vbe_nvlist[0],
be->vbe_nvlist[1]);
if (benv == NULL) {
char *command = (char *)be;
bool ok = false;
/* Check for legacy zfsbootcfg command string */
for (int i = 0; command[i] != '\0'; i++) {
if (iscntrl(command[i])) {
ok = false;
break;
} else {
ok = true;
}
}
benv = nvlist_create(NV_UNIQUE_NAME);
if (ok)
nvlist_add_string(benv, "command", command);
}
free(be);
return (benv);
}
static uint64_t static uint64_t
vdev_get_label_asize(nvlist_t *nvl) vdev_get_label_asize(nvlist_t *nvl)
{ {
nvlist_t *vdevs; nvlist_t *vdevs;
uint64_t asize; uint64_t asize;
const char *type; const char *type;
int len; int len;
▲ Show 20 Lines • Show All 50 Lines • ▼ Show 20 Linesvdev_label_read_config(vdev_t *vd, uint64_t txg)
nvlist_t *nvl = NULL, *tmp; nvlist_t *nvl = NULL, *tmp;
int error; int error;
label = malloc(sizeof (vdev_phys_t)); label = malloc(sizeof (vdev_phys_t));
if (label == NULL) if (label == NULL)
return (NULL); return (NULL);
for (int l = 0; l < VDEV_LABELS; l++) { for (int l = 0; l < VDEV_LABELS; l++) {
const unsigned char *nvlist;
if (vdev_label_read(vd, l, label, if (vdev_label_read(vd, l, label,
offsetof(vdev_label_t, vl_vdev_phys), offsetof(vdev_label_t, vl_vdev_phys),
sizeof (vdev_phys_t))) sizeof (vdev_phys_t)))
continue; continue;
nvlist = (const unsigned char *) label->vp_nvlist; tmp = nvlist_import(label->vp_nvlist + 4,
tmp = nvlist_import(nvlist + 4, nvlist[0], nvlist[1]); label->vp_nvlist[0], label->vp_nvlist[1]);
if (tmp == NULL) if (tmp == NULL)
continue; continue;
error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG, error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
DATA_TYPE_UINT64, NULL, &label_txg, NULL); DATA_TYPE_UINT64, NULL, &label_txg, NULL);
if (error != 0 || label_txg == 0) { if (error != 0 || label_txg == 0) {
nvlist_destroy(nvl); nvlist_destroy(nvl);
nvl = tmp; nvl = tmp;
▲ Show 20 Lines • Show All 48 Lines • ▼ Show 20 Lines for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
if (vdev_uberblock_compare(buf, ub) > 0) if (vdev_uberblock_compare(buf, ub) > 0)
*ub = *buf; *ub = *buf;
} }
} }
free(buf); free(buf);
} }
static int static int
vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap) vdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv,
spa_t **spap)
{ {
vdev_t vtmp; vdev_t vtmp;
spa_t *spa; spa_t *spa;
vdev_t *vdev; vdev_t *vdev;
nvlist_t *nvl; nvlist_t *nvl;
uint64_t val; uint64_t val;
uint64_t guid, vdev_children; uint64_t guid, vdev_children;
uint64_t pool_txg, pool_guid; uint64_t pool_txg, pool_guid;
const char *pool_name; const char *pool_name;
int rc, namelen; int rc, namelen;
/* /*
* Load the vdev label and figure out which * Load the vdev label and figure out which
* uberblock is most current. * uberblock is most current.
*/ */
memset(&vtmp, 0, sizeof(vtmp)); memset(&vtmp, 0, sizeof(vtmp));
vtmp.v_phys_read = _read; vtmp.v_phys_read = _read;
vtmp.v_read_priv = read_priv; vtmp.v_phys_write = _write;
vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv), vtmp.v_priv = priv;
vtmp.v_psize = P2ALIGN(ldi_get_size(priv),
(uint64_t)sizeof (vdev_label_t)); (uint64_t)sizeof (vdev_label_t));
/* Test for minimum device size. */ /* Test for minimum device size. */
if (vtmp.v_psize < SPA_MINDEVSIZE) if (vtmp.v_psize < SPA_MINDEVSIZE)
return (EIO); return (EIO);
nvl = vdev_label_read_config(&vtmp, UINT64_MAX); nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
if (nvl == NULL) if (nvl == NULL)
▲ Show 20 Lines • Show All 97 Lines • ▼ Show 20 Linesvdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv,
/* /*
* We should already have created an incomplete vdev for this * We should already have created an incomplete vdev for this
* vdev. Find it and initialise it with our read proc. * vdev. Find it and initialise it with our read proc.
*/ */
vdev = vdev_find(guid); vdev = vdev_find(guid);
if (vdev != NULL) { if (vdev != NULL) {
vdev->v_phys_read = _read; vdev->v_phys_read = _read;
vdev->v_read_priv = read_priv; vdev->v_phys_write = _write;
vdev->v_priv = priv;
vdev->v_psize = vtmp.v_psize; vdev->v_psize = vtmp.v_psize;
/* /*
* If no other state is set, mark vdev healthy. * If no other state is set, mark vdev healthy.
*/ */
if (vdev->v_state == VDEV_STATE_UNKNOWN) if (vdev->v_state == VDEV_STATE_UNKNOWN)
vdev->v_state = VDEV_STATE_HEALTHY; vdev->v_state = VDEV_STATE_HEALTHY;
} else { } else {
printf("ZFS: inconsistent nvlist contents\n"); printf("ZFS: inconsistent nvlist contents\n");
▲ Show 20 Lines • Show All 1,253 Lines • ▼ Show 20 Lines
} }
static int static int
load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
{ {
dnode_phys_t dir; dnode_phys_t dir;
size_t size; size_t size;
int rc; int rc;
unsigned char *nv; char *nv;
*value = NULL; *value = NULL;
if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0) if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0)
return (rc); return (rc);
if (dir.dn_type != DMU_OT_PACKED_NVLIST && if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) { dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
return (EIO); return (EIO);
} }
▲ Show 20 Lines • Show All 351 Lines • Show Last 20 Lines