/*
* Given a list of directories to search, find all pools stored on disk. This
* includes partial pools which are not available to import. If no args are
* given (argc is 0), then the default directory (/dev/dsk) is searched.
* poolname or guid (but not both) are provided by the caller when trying
* to import a specific pool.
*/
static nvlist_t *
zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
{
int i, num_labels, dirs = iarg->paths;
DIR *dirp = NULL;
struct dirent *dp;
char path[MAXPATHLEN];
char *end, **dir = iarg->path;
size_t pathleft;
struct stat statbuf;
nvlist_t *ret = NULL, *config;
int fd;
pool_list_t pools = { 0 };
pool_entry_t *pe, *penext;
vdev_entry_t *ve, *venext;
config_entry_t *ce, *cenext;
name_entry_t *ne, *nenext;
verify(iarg->poolname == NULL || iarg->guid == 0);
if (dirs == 0) {
#ifdef HAVE_LIBBLKID
/* Use libblkid to scan all device for their type */
if (zpool_find_import_blkid(hdl, &pools) == 0)
goto skip_scanning;
(void) zfs_error_fmt(hdl, EZFS_BADCACHE,
dgettext(TEXT_DOMAIN, "blkid failure falling back "
"to manual probing"));
#endif /* HAVE_LIBBLKID */
dir = zpool_default_import_path;
dirs = DEFAULT_IMPORT_PATH_SIZE;
}
/*
* Go through and read the label configuration information from every
* possible device, organizing the information according to pool GUID
* and toplevel GUID.
*/
for (i = 0; i < dirs; i++) {
char *rdsk;
int dfd;
/* use realpath to normalize the path */
if (realpath(dir[i], path) == 0) {
/* it is safe to skip missing search paths */
if (errno == ENOENT)
continue;
zfs_error_aux(hdl, strerror(errno));
(void) zfs_error_fmt(hdl, EZFS_BADPATH,
dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
goto error;
}
end = &path[strlen(path)];
*end++ = '/';
*end = 0;
pathleft = &path[sizeof (path)] - end;
/*
* Using raw devices instead of block devices when we're
* reading the labels skips a bunch of slow operations during
* close(2) processing, so we replace /dev/dsk with /dev/rdsk.
*/
if (strcmp(path, "/dev/dsk/") == 0)
rdsk = "/dev/rdsk/";
else
rdsk = path;
if ((dfd = open(rdsk, O_RDONLY)) < 0 ||
(dirp = fdopendir(dfd)) == NULL) {
zfs_error_aux(hdl, strerror(errno));
(void) zfs_error_fmt(hdl, EZFS_BADPATH,
dgettext(TEXT_DOMAIN, "cannot open '%s'"),
rdsk);
goto error;
}
/*
* This is not MT-safe, but we have no MT consumers of libzfs
*/
while ((dp = readdir(dirp)) != NULL) {
const char *name = dp->d_name;
if (name[0] == '.' &&
(name[1] == 0 || (name[1] == '.' && name[2] == 0)))
continue;
/*
* Skip checking devices with well known prefixes:
* watchdog - A special close is required to avoid
* triggering it and resetting the system.
* fuse - Fuse control device.
* ppp - Generic PPP driver.
* tty* - Generic serial interface.
* vcs* - Virtual console memory.
* parport* - Parallel port interface.
* lp* - Printer interface.
* fd* - Floppy interface.
* hpet - High Precision Event Timer, crashes qemu
* when accessed from a virtual machine.
* core - Symlink to /proc/kcore, causes a crash
* when access from Xen dom0.
*/
if ((strncmp(name, "watchdog", 8) == 0) ||
(strncmp(name, "fuse", 4) == 0) ||
(strncmp(name, "ppp", 3) == 0) ||
(strncmp(name, "tty", 3) == 0) ||
(strncmp(name, "vcs", 3) == 0) ||
(strncmp(name, "parport", 7) == 0) ||
(strncmp(name, "lp", 2) == 0) ||
(strncmp(name, "fd", 2) == 0) ||
(strncmp(name, "hpet", 4) == 0) ||
#ifdef __APPLE__
(strncmp(name, "pty", 3) == 0) || // lots, skip for speed
(strncmp(name, "com", 3) == 0) || // /dev/com_digidesign_semiface
#endif
(strncmp(name, "core", 4) == 0))
continue;
/*
* Ignore failed stats. We only want regular
* files and block devices.
*/
if ((fstatat64(dfd, name, &statbuf, 0) != 0) ||
(!S_ISREG(statbuf.st_mode) &&
!S_ISBLK(statbuf.st_mode)))
continue;
#ifdef __APPLE__
/* It is desirable to skip optical media as well, as they are
* also called /dev/diskX
*/
if (is_optical_media((char *)name))
continue;
#endif
if ((fd = openat64(dfd, name, O_RDONLY)) < 0)
continue;
int32_t blksz = 0;
if (S_ISBLK(statbuf.st_mode) &&
(ioctl(fd, DKIOCGETBLOCKSIZE, &blksz) || blksz == 0)) {
if (strncmp(name, "vn", 2) != 0)
fprintf(stderr, "device '%s' failed to report blocksize -- skipping\r\n",
name);
close(fd);
continue;
}
#ifdef __APPLE__
struct sigaction sact;
sigemptyset(&sact.sa_mask);
sact.sa_flags = 0;
sact.sa_handler = signal_alarm;
sigaction(SIGALRM, &sact, NULL);
if (setjmp(buffer) != 0) {
printf("ZFS: Warning, timeout reading device '%s'\n", name);
close(fd);
continue;
}
alarm(20);
#endif
if ((zpool_read_label(fd, &config, NULL)) != 0) {
#ifdef __APPLE__
alarm(0);
#endif
(void) close(fd);
(void) no_memory(hdl);
goto error;
}
#ifdef __APPLE__
alarm(0);
#endif
(void) close(fd);
if (config != NULL) {
boolean_t matched = B_TRUE;
char *pname;
if ((iarg->poolname != NULL) &&
(nvlist_lookup_string(config,
ZPOOL_CONFIG_POOL_NAME, &pname) == 0)) {
if (strcmp(iarg->poolname, pname))
matched = B_FALSE;
} else if (iarg->guid != 0) {
uint64_t this_guid;
matched = nvlist_lookup_uint64(config,
ZPOOL_CONFIG_POOL_GUID,
&this_guid) == 0 &&
iarg->guid == this_guid;
}
if (!matched) {
nvlist_free(config);
config = NULL;
continue;
}
/* use the non-raw path for the config */
(void) strlcpy(end, name, pathleft);
if (add_config(hdl, &pools, path, i+1,
num_labels, config))
goto error;
}
}
(void) closedir(dirp);
dirp = NULL;
}
#ifdef HAVE_LIBBLKID
skip_scanning:
#endif
ret = get_configs(hdl, &pools, iarg->can_be_active);
error:
for (pe = pools.pools; pe != NULL; pe = penext) {
penext = pe->pe_next;
for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
venext = ve->ve_next;
for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
cenext = ce->ce_next;
if (ce->ce_config)
nvlist_free(ce->ce_config);
free(ce);
}
free(ve);
}
free(pe);
}
for (ne = pools.names; ne != NULL; ne = nenext) {
nenext = ne->ne_next;
if (ne->ne_name)
free(ne->ne_name);
free(ne);
}
if (dirp)
(void) closedir(dirp);
return (ret);
}
/*
* Given a list of directories to search, find all pools stored on disk. This
* includes partial pools which are not available to import. If no args are
* given (argc is 0), then the default directory (/dev/dsk) is searched.
* poolname or guid (but not both) are provided by the caller when trying
* to import a specific pool.
*/
static nvlist_t *
zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
{
int i, dirs = iarg->paths;
struct dirent *dp;
char path[MAXPATHLEN];
char *end, **dir = iarg->path;
size_t pathleft;
nvlist_t *ret = NULL;
pool_list_t pools = { 0 };
pool_entry_t *pe, *penext;
vdev_entry_t *ve, *venext;
config_entry_t *ce, *cenext;
name_entry_t *ne, *nenext;
avl_tree_t slice_cache;
rdsk_node_t *slice;
void *cookie;
verify(iarg->poolname == NULL || iarg->guid == 0);
if (dirs == 0) {
#ifdef HAVE_LIBBLKID
/* Use libblkid to scan all device for their type */
if (zpool_find_import_blkid(hdl, &pools) == 0)
goto skip_scanning;
(void) zfs_error_fmt(hdl, EZFS_BADCACHE,
dgettext(TEXT_DOMAIN, "blkid failure falling back "
"to manual probing"));
#endif /* HAVE_LIBBLKID */
dir = zpool_default_import_path;
dirs = DEFAULT_IMPORT_PATH_SIZE;
}
/*
* Go through and read the label configuration information from every
* possible device, organizing the information according to pool GUID
* and toplevel GUID.
*/
for (i = 0; i < dirs; i++) {
taskq_t *t;
char rdsk[MAXPATHLEN];
int dfd;
boolean_t config_failed = B_FALSE;
DIR *dirp;
/* use realpath to normalize the path */
if (realpath(dir[i], path) == 0) {
/* it is safe to skip missing search paths */
if (errno == ENOENT)
continue;
zfs_error_aux(hdl, strerror(errno));
(void) zfs_error_fmt(hdl, EZFS_BADPATH,
dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
goto error;
}
end = &path[strlen(path)];
*end++ = '/';
*end = 0;
pathleft = &path[sizeof (path)] - end;
/*
* Using raw devices instead of block devices when we're
* reading the labels skips a bunch of slow operations during
* close(2) processing, so we replace /dev/dsk with /dev/rdsk.
*/
if (strcmp(path, ZFS_DISK_ROOTD) == 0)
(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
else
(void) strlcpy(rdsk, path, sizeof (rdsk));
if ((dfd = open(rdsk, O_RDONLY)) < 0 ||
(dirp = fdopendir(dfd)) == NULL) {
if (dfd >= 0)
(void) close(dfd);
zfs_error_aux(hdl, strerror(errno));
(void) zfs_error_fmt(hdl, EZFS_BADPATH,
dgettext(TEXT_DOMAIN, "cannot open '%s'"),
rdsk);
goto error;
}
avl_create(&slice_cache, slice_cache_compare,
sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
/*
* This is not MT-safe, but we have no MT consumers of libzfs
*/
while ((dp = readdir(dirp)) != NULL) {
const char *name = dp->d_name;
if (name[0] == '.' &&
(name[1] == 0 || (name[1] == '.' && name[2] == 0)))
continue;
slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
slice->rn_name = zfs_strdup(hdl, name);
slice->rn_avl = &slice_cache;
slice->rn_dfd = dfd;
slice->rn_hdl = hdl;
slice->rn_nozpool = B_FALSE;
avl_add(&slice_cache, slice);
}
/*
* create a thread pool to do all of this in parallel;
* rn_nozpool is not protected, so this is racy in that
* multiple tasks could decide that the same slice can
* not hold a zpool, which is benign. Also choose
* double the number of processors; we hold a lot of
* locks in the kernel, so going beyond this doesn't
* buy us much.
*/
t = taskq_create("z_import", 2 * max_ncpus, defclsyspri,
2 * max_ncpus, INT_MAX, TASKQ_PREPOPULATE);
for (slice = avl_first(&slice_cache); slice;
(slice = avl_walk(&slice_cache, slice,
AVL_AFTER)))
(void) taskq_dispatch(t, zpool_open_func, slice,
TQ_SLEEP);
taskq_wait(t);
taskq_destroy(t);
cookie = NULL;
while ((slice = avl_destroy_nodes(&slice_cache,
&cookie)) != NULL) {
if (slice->rn_config != NULL && !config_failed) {
nvlist_t *config = slice->rn_config;
boolean_t matched = B_TRUE;
if (iarg->poolname != NULL) {
char *pname;
matched = nvlist_lookup_string(config,
ZPOOL_CONFIG_POOL_NAME,
&pname) == 0 &&
strcmp(iarg->poolname, pname) == 0;
} else if (iarg->guid != 0) {
uint64_t this_guid;
matched = nvlist_lookup_uint64(config,
ZPOOL_CONFIG_POOL_GUID,
&this_guid) == 0 &&
iarg->guid == this_guid;
}
if (!matched) {
nvlist_free(config);
} else {
/*
* use the non-raw path for the config
*/
(void) strlcpy(end, slice->rn_name,
pathleft);
if (add_config(hdl, &pools, path, i+1,
slice->rn_num_labels, config) != 0)
config_failed = B_TRUE;
}
}
free(slice->rn_name);
free(slice);
}
avl_destroy(&slice_cache);
(void) closedir(dirp);
if (config_failed)
goto error;
}
#ifdef HAVE_LIBBLKID
skip_scanning:
#endif
ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
error:
for (pe = pools.pools; pe != NULL; pe = penext) {
penext = pe->pe_next;
for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
venext = ve->ve_next;
for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
cenext = ce->ce_next;
if (ce->ce_config)
nvlist_free(ce->ce_config);
free(ce);
}
free(ve);
}
free(pe);
}
for (ne = pools.names; ne != NULL; ne = nenext) {
nenext = ne->ne_next;
free(ne->ne_name);
free(ne);
}
return (ret);
}
/*
* Given a list of directories to search, find all pools stored on disk. This
* includes partial pools which are not available to import. If no args are
* given (argc is 0), then the default directory (/dev/dsk) is searched.
* poolname or guid (but not both) are provided by the caller when trying
* to import a specific pool.
*/
static nvlist_t *
zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
{
int i, dirs = iarg->paths;
DIR *dirp = NULL;
struct dirent64 *dp;
char path[MAXPATHLEN];
char *end, **dir = iarg->path;
size_t pathleft;
struct stat64 statbuf;
nvlist_t *ret = NULL, *config;
static char *default_dir = DISK_ROOT;
int fd;
pool_list_t pools = { 0 };
pool_entry_t *pe, *penext;
vdev_entry_t *ve, *venext;
config_entry_t *ce, *cenext;
name_entry_t *ne, *nenext;
verify(iarg->poolname == NULL || iarg->guid == 0);
if (dirs == 0) {
#ifdef HAVE_LIBBLKID
/* Use libblkid to scan all device for their type */
if (zpool_find_import_blkid(hdl, &pools) == 0)
goto skip_scanning;
(void) zfs_error_fmt(hdl, EZFS_BADCACHE,
dgettext(TEXT_DOMAIN, "blkid failure falling back "
"to manual probing"));
#endif /* HAVE_LIBBLKID */
dirs = 1;
dir = &default_dir;
}
/*
* Go through and read the label configuration information from every
* possible device, organizing the information according to pool GUID
* and toplevel GUID.
*/
for (i = 0; i < dirs; i++) {
char *rdsk;
int dfd;
/* use realpath to normalize the path */
if (realpath(dir[i], path) == 0) {
(void) zfs_error_fmt(hdl, EZFS_BADPATH,
dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
goto error;
}
end = &path[strlen(path)];
*end++ = '/';
*end = 0;
pathleft = &path[sizeof (path)] - end;
/*
* Using raw devices instead of block devices when we're
* reading the labels skips a bunch of slow operations during
* close(2) processing, so we replace /dev/dsk with /dev/rdsk.
*/
if (strcmp(path, "/dev/dsk/") == 0)
rdsk = "/dev/rdsk/";
else
rdsk = path;
if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
(dirp = fdopendir(dfd)) == NULL) {
zfs_error_aux(hdl, strerror(errno));
(void) zfs_error_fmt(hdl, EZFS_BADPATH,
dgettext(TEXT_DOMAIN, "cannot open '%s'"),
rdsk);
goto error;
}
/*
* This is not MT-safe, but we have no MT consumers of libzfs
*/
while ((dp = readdir64(dirp)) != NULL) {
const char *name = dp->d_name;
if (name[0] == '.' &&
(name[1] == 0 || (name[1] == '.' && name[2] == 0)))
continue;
/*
* Skip checking devices with well known prefixes:
* watchdog - A special close is required to avoid
* triggering it and resetting the system.
* fuse - Fuse control device.
* ppp - Generic PPP driver.
* tty* - Generic serial interface.
* vcs* - Virtual console memory.
* parport* - Parallel port interface.
* lp* - Printer interface.
* fd* - Floppy interface.
* hpet - High Precision Event Timer, crashes qemu
* when accessed from a virtual machine.
* core - Symlink to /proc/kcore, causes a crash
* when access from Xen dom0.
*/
if ((strncmp(name, "watchdog", 8) == 0) ||
(strncmp(name, "fuse", 4) == 0) ||
(strncmp(name, "ppp", 3) == 0) ||
(strncmp(name, "tty", 3) == 0) ||
(strncmp(name, "vcs", 3) == 0) ||
(strncmp(name, "parport", 7) == 0) ||
(strncmp(name, "lp", 2) == 0) ||
(strncmp(name, "fd", 2) == 0) ||
(strncmp(name, "hpet", 4) == 0) ||
(strncmp(name, "core", 4) == 0))
continue;
/*
* Ignore failed stats. We only want regular
* files and block devices.
*/
if ((fstatat64(dfd, name, &statbuf, 0) != 0) ||
(!S_ISREG(statbuf.st_mode) &&
!S_ISBLK(statbuf.st_mode)))
continue;
if ((fd = openat64(dfd, name, O_RDONLY)) < 0)
continue;
if ((zpool_read_label(fd, &config)) != 0) {
(void) close(fd);
(void) no_memory(hdl);
goto error;
}
(void) close(fd);
if (config != NULL) {
boolean_t matched = B_TRUE;
if (iarg->poolname != NULL) {
char *pname;
matched = nvlist_lookup_string(config,
ZPOOL_CONFIG_POOL_NAME,
&pname) == 0 &&
strcmp(iarg->poolname, pname) == 0;
} else if (iarg->guid != 0) {
uint64_t this_guid;
matched = nvlist_lookup_uint64(config,
ZPOOL_CONFIG_POOL_GUID,
&this_guid) == 0 &&
iarg->guid == this_guid;
}
if (!matched) {
nvlist_free(config);
config = NULL;
continue;
}
/* use the non-raw path for the config */
(void) strlcpy(end, name, pathleft);
if (add_config(hdl, &pools, path, config) != 0)
goto error;
}
}
(void) closedir(dirp);
dirp = NULL;
}
#ifdef HAVE_LIBBLKID
skip_scanning:
#endif
ret = get_configs(hdl, &pools, iarg->can_be_active);
error:
for (pe = pools.pools; pe != NULL; pe = penext) {
penext = pe->pe_next;
for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
venext = ve->ve_next;
for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
cenext = ce->ce_next;
if (ce->ce_config)
nvlist_free(ce->ce_config);
free(ce);
}
free(ve);
}
free(pe);
}
for (ne = pools.names; ne != NULL; ne = nenext) {
nenext = ne->ne_next;
if (ne->ne_name)
free(ne->ne_name);
free(ne);
}
if (dirp)
(void) closedir(dirp);
return (ret);
}