/*
* Search the devid cache, returning dev_t list for all
* device paths mapping to the device identified by the
* given devid.
*
* Primary interface used by ddi_lyr_devid_to_devlist()
*/
int
e_devid_cache_to_devt_list(ddi_devid_t devid, char *minor_name,
int *retndevts, dev_t **retdevts)
{
char *path, **paths;
int i, j, n;
dev_t *devts, *udevts;
dev_t tdevt;
int ndevts, undevts, ndevts_alloced;
dev_info_t *devi, **devis;
int ndevis, npaths, nalloced;
ddi_devid_t match_devid;
DEVID_LOG_FIND(("find", devid, NULL));
ASSERT(ddi_devid_valid(devid) == DDI_SUCCESS);
if (ddi_devid_valid(devid) != DDI_SUCCESS) {
DEVID_LOG_ERR(("invalid devid", devid, NULL));
return (DDI_FAILURE);
}
nalloced = 128;
for (;;) {
paths = kmem_zalloc(nalloced * sizeof (char *), KM_SLEEP);
devis = kmem_zalloc(nalloced * sizeof (dev_info_t *), KM_SLEEP);
rw_enter(&dcfd->nvf_lock, RW_READER);
n = e_devid_cache_devi_path_lists(devid, nalloced,
&ndevis, devis, &npaths, paths);
if (n <= nalloced)
break;
rw_exit(&dcfd->nvf_lock);
for (i = 0; i < ndevis; i++)
ndi_rele_devi(devis[i]);
kmem_free(paths, nalloced * sizeof (char *));
kmem_free(devis, nalloced * sizeof (dev_info_t *));
nalloced = n + 128;
}
for (i = 0; i < npaths; i++) {
path = i_ddi_strdup(paths[i], KM_SLEEP);
paths[i] = path;
}
rw_exit(&dcfd->nvf_lock);
if (ndevis == 0 && npaths == 0) {
DEVID_LOG_ERR(("no devid found", devid, NULL));
kmem_free(paths, nalloced * sizeof (char *));
kmem_free(devis, nalloced * sizeof (dev_info_t *));
return (DDI_FAILURE);
}
ndevts_alloced = 128;
restart:
ndevts = 0;
devts = kmem_alloc(ndevts_alloced * sizeof (dev_t), KM_SLEEP);
for (i = 0; i < ndevis; i++) {
ASSERT(!DEVI_IS_ATTACHING(devis[i]));
ASSERT(!DEVI_IS_DETACHING(devis[i]));
e_devid_minor_to_devlist(devis[i], minor_name,
ndevts_alloced, &ndevts, devts);
if (ndevts > ndevts_alloced) {
kmem_free(devts, ndevts_alloced * sizeof (dev_t));
ndevts_alloced += 128;
goto restart;
}
}
for (i = 0; i < npaths; i++) {
DEVID_LOG_LOOKUP((CE_CONT, "lookup %s\n", paths[i]));
devi = e_ddi_hold_devi_by_path(paths[i], 0);
if (devi == NULL) {
DEVID_LOG_STALE(("stale device reference",
devid, paths[i]));
continue;
}
/*
* Verify the newly attached device registered a matching devid
*/
if (i_ddi_devi_get_devid(DDI_DEV_T_ANY, devi,
&match_devid) != DDI_SUCCESS) {
DEVIDERR((CE_CONT,
"%s: no devid registered on attach\n",
paths[i]));
ddi_release_devi(devi);
continue;
}
if (ddi_devid_compare(devid, match_devid) != 0) {
DEVID_LOG_STALE(("new devid registered",
devid, paths[i]));
ddi_release_devi(devi);
ddi_devid_free(match_devid);
continue;
}
ddi_devid_free(match_devid);
e_devid_minor_to_devlist(devi, minor_name,
ndevts_alloced, &ndevts, devts);
ddi_release_devi(devi);
if (ndevts > ndevts_alloced) {
kmem_free(devts,
ndevts_alloced * sizeof (dev_t));
ndevts_alloced += 128;
goto restart;
}
}
/* drop hold from e_devid_cache_devi_path_lists */
for (i = 0; i < ndevis; i++) {
ndi_rele_devi(devis[i]);
}
for (i = 0; i < npaths; i++) {
kmem_free(paths[i], strlen(paths[i]) + 1);
}
kmem_free(paths, nalloced * sizeof (char *));
kmem_free(devis, nalloced * sizeof (dev_info_t *));
if (ndevts == 0) {
DEVID_LOG_ERR(("no devid found", devid, NULL));
kmem_free(devts, ndevts_alloced * sizeof (dev_t));
return (DDI_FAILURE);
}
/*
* Build the final list of sorted dev_t's with duplicates collapsed so
* returned results are consistent. This prevents implementation
* artifacts from causing unnecessary changes in SVM namespace.
*/
/* bubble sort */
for (i = 0; i < (ndevts - 1); i++) {
for (j = 0; j < ((ndevts - 1) - i); j++) {
if (devts[j + 1] < devts[j]) {
tdevt = devts[j];
devts[j] = devts[j + 1];
devts[j + 1] = tdevt;
}
}
}
/* determine number of unique values */
for (undevts = ndevts, i = 1; i < ndevts; i++) {
if (devts[i - 1] == devts[i])
undevts--;
}
/* allocate unique */
udevts = kmem_alloc(undevts * sizeof (dev_t), KM_SLEEP);
/* copy unique */
udevts[0] = devts[0];
for (i = 1, j = 1; i < ndevts; i++) {
if (devts[i - 1] != devts[i])
udevts[j++] = devts[i];
}
ASSERT(j == undevts);
kmem_free(devts, ndevts_alloced * sizeof (dev_t));
*retndevts = undevts;
*retdevts = udevts;
return (DDI_SUCCESS);
}
/*
* Search for cached entries matching a devid
* Return two lists:
* a list of dev_info nodes, for those devices in the attached state
* a list of pathnames whose instances registered the given devid
* If the lists passed in are not sufficient to return the matching
* references, return the size of lists required.
* The dev_info nodes are returned with a hold that the caller must release.
*/
static int
e_devid_cache_devi_path_lists(ddi_devid_t devid, int retmax,
int *retndevis, dev_info_t **retdevis, int *retnpaths, char **retpaths)
{
nvp_devid_t *np;
int ndevis, npaths;
dev_info_t *dip, *pdip;
int circ;
int maxdevis = 0;
int maxpaths = 0;
ndevis = 0;
npaths = 0;
for (np = NVF_DEVID_LIST(dcfd); np; np = NVP_DEVID_NEXT(np)) {
if (np->nvp_devid == NULL)
continue;
if (ddi_devid_valid(np->nvp_devid) != DDI_SUCCESS) {
DEVIDERR((CE_CONT,
"find: invalid devid %s\n",
np->nvp_devpath));
continue;
}
if (ddi_devid_compare(devid, np->nvp_devid) == 0) {
DEVID_DEBUG2((CE_CONT,
"find: devid match: %s 0x%x\n",
np->nvp_devpath, np->nvp_flags));
DEVID_LOG_MATCH(("find", devid, np->nvp_devpath));
DEVID_LOG_PATHS((CE_CONT, "%s\n", np->nvp_devpath));
/*
* Check if we have a cached devinfo reference for this
* devid. Place a hold on it to prevent detach
* Otherwise, use the path instead.
* Note: returns with a hold on each dev_info
* node in the list.
*/
dip = NULL;
if (np->nvp_flags & NVP_DEVID_DIP) {
pdip = ddi_get_parent(np->nvp_dip);
if (ndi_devi_tryenter(pdip, &circ)) {
dip = np->nvp_dip;
ndi_hold_devi(dip);
ndi_devi_exit(pdip, circ);
ASSERT(!DEVI_IS_ATTACHING(dip));
ASSERT(!DEVI_IS_DETACHING(dip));
} else {
DEVID_LOG_DETACH((CE_CONT,
"may be detaching: %s\n",
np->nvp_devpath));
}
}
if (dip) {
if (ndevis < retmax) {
retdevis[ndevis++] = dip;
} else {
ndi_rele_devi(dip);
}
maxdevis++;
} else {
if (npaths < retmax)
retpaths[npaths++] = np->nvp_devpath;
maxpaths++;
}
}
}
*retndevis = ndevis;
*retnpaths = npaths;
return (maxdevis > maxpaths ? maxdevis : maxpaths);
}
int
e_devid_cache_register(dev_info_t *dip, ddi_devid_t devid)
{
nvp_devid_t *np;
nvp_devid_t *new_nvp;
ddi_devid_t new_devid;
int new_devid_size;
char *path, *fullpath;
ddi_devid_t free_devid = NULL;
int pathlen;
ASSERT(ddi_devid_valid(devid) == DDI_SUCCESS);
fullpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
(void) ddi_pathname(dip, fullpath);
pathlen = strlen(fullpath) + 1;
path = kmem_alloc(pathlen, KM_SLEEP);
bcopy(fullpath, path, pathlen);
kmem_free(fullpath, MAXPATHLEN);
DEVID_LOG_REG(("register", devid, path));
new_nvp = kmem_zalloc(sizeof (nvp_devid_t), KM_SLEEP);
new_devid_size = ddi_devid_sizeof(devid);
new_devid = kmem_alloc(new_devid_size, KM_SLEEP);
(void) bcopy(devid, new_devid, new_devid_size);
rw_enter(&dcfd->nvf_lock, RW_WRITER);
for (np = NVF_DEVID_LIST(dcfd); np; np = NVP_DEVID_NEXT(np)) {
if (strcmp(path, np->nvp_devpath) == 0) {
DEVID_DEBUG2((CE_CONT,
"register: %s path match\n", path));
if (np->nvp_devid == NULL) {
replace:
np->nvp_devid = new_devid;
np->nvp_flags |=
NVP_DEVID_DIP | NVP_DEVID_REGISTERED;
np->nvp_dip = dip;
NVF_MARK_DIRTY(dcfd);
rw_exit(&dcfd->nvf_lock);
kmem_free(new_nvp, sizeof (nvp_devid_t));
kmem_free(path, pathlen);
goto exit;
}
if (ddi_devid_valid(np->nvp_devid) != DDI_SUCCESS) {
/* replace invalid devid */
free_devid = np->nvp_devid;
goto replace;
}
/*
* We're registering an already-cached path
* Does the device's devid match the cache?
*/
if (ddi_devid_compare(devid, np->nvp_devid) != 0) {
DEVID_DEBUG((CE_CONT, "devid register: "
"devid %s does not match\n", path));
/*
* Replace cached devid for this path
* with newly registered devid. A devid
* may map to multiple paths but one path
* should only map to one devid.
*/
nfd_nvp_free_and_unlink(dcfd, NVPLIST(np));
np = NULL;
break;
} else {
DEVID_DEBUG2((CE_CONT,
"devid register: %s devid match\n", path));
np->nvp_flags |=
NVP_DEVID_DIP | NVP_DEVID_REGISTERED;
np->nvp_dip = dip;
rw_exit(&dcfd->nvf_lock);
kmem_free(new_nvp, sizeof (nvp_devid_t));
kmem_free(path, pathlen);
kmem_free(new_devid, new_devid_size);
return (DDI_SUCCESS);
}
}
}
/*
* Add newly registered devid to the cache
*/
ASSERT(np == NULL);
new_nvp->nvp_devpath = path;
new_nvp->nvp_flags = NVP_DEVID_DIP | NVP_DEVID_REGISTERED;
new_nvp->nvp_dip = dip;
new_nvp->nvp_devid = new_devid;
NVF_MARK_DIRTY(dcfd);
nfd_nvp_link(dcfd, NVPLIST(new_nvp));
rw_exit(&dcfd->nvf_lock);
exit:
if (free_devid)
kmem_free(free_devid, ddi_devid_sizeof(free_devid));
if (!devid_cache_write_disable)
wake_nvpflush_daemon();
return (DDI_SUCCESS);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd = vd->vdev_tsd;
ldi_ev_cookie_t ecookie;
vdev_disk_ldi_cb_t *lcb;
union {
struct dk_minfo_ext ude;
struct dk_minfo ud;
} dks;
struct dk_minfo_ext *dkmext = &dks.ude;
struct dk_minfo *dkm = &dks.ud;
int error;
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
dev_t dev;
int otyp;
boolean_t validate_devid = B_FALSE;
ddi_devid_t devid;
#endif
uint64_t capacity = 0, blksz = 0, pbsize;
#ifdef __APPLE__
int isssd;
#endif
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
if (dvd->vd_ldi_offline && dvd->vd_lh == NULL) {
/*
* If we are opening a device in its offline notify
* context, the LDI handle was just closed. Clean
* up the LDI event callbacks and free vd->vdev_tsd.
*/
vdev_disk_free(vd);
} else {
ASSERT(vd->vdev_reopening);
goto skip_open;
}
}
/*
* Create vd->vdev_tsd.
*/
vdev_disk_alloc(vd);
dvd = vd->vdev_tsd;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
/*
* XXX We must not set or modify the devid as this check would prevent
* import on Solaris/illumos.
*/
#ifdef illumos
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
vdev_dbgmsg(vd, "vdev_disk_open: invalid "
"vdev_devid '%s'", vd->vdev_devid);
return (SET_ERROR(EINVAL));
}
}
#endif
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
/*
* XXX This assumes that if vdev_path refers to a device path /dev/dsk/cNtNdN,
* then the whole disk can be found by slice 0 at path /dev/dsk/cNtNdNs0.
*/
#ifdef illumos
if (vd->vdev_wholedisk == -1ULL) {
size_t len = strlen(vd->vdev_path) + 3;
char *buf = kmem_alloc(len, KM_SLEEP);
(void) snprintf(buf, len, "%ss0", vd->vdev_path);
error = ldi_open_by_name(buf, spa_mode(spa), kcred,
&dvd->vd_lh, zfs_li);
if (error == 0) {
spa_strfree(vd->vdev_path);
vd->vdev_path = buf;
vd->vdev_wholedisk = 1ULL;
} else {
kmem_free(buf, len);
}
}
#endif
/*
* If we have not yet opened the device, try to open it by the
* specified path.
*/
if (error != 0) {
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
/*
* Compare the devid to the stored value.
*/
if (error == 0 && vd->vdev_devid != NULL &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
error = SET_ERROR(EINVAL);
(void) ldi_close(dvd->vd_lh, spa_mode(spa),
kcred);
dvd->vd_lh = NULL;
}
ddi_devid_free(devid);
}
#endif
/*
* If we succeeded in opening the device, but 'vdev_wholedisk'
* is not yet set, then this must be a slice.
*/
if (error == 0 && vd->vdev_wholedisk == -1ULL)
vd->vdev_wholedisk = 0;
}
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
/*
* If we were unable to open by path, or the devid check fails, open by
* devid instead.
*/
if (error != 0 && vd->vdev_devid != NULL) {
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
}
#endif
/*
* If all else fails, then try opening by physical path (if available)
* or the logical path (if we failed due to the devid check). While not
* as reliable as the devid, this will give us something, and the higher
* level vdev validation will prevent us from opening the wrong device.
*/
if (error) {
/* XXX Apple - must leave devid unchanged */
#ifdef illumos
if (vd->vdev_devid != NULL)
validate_devid = B_TRUE;
#endif
/* XXX Apple to do - make ddi_ interface for this, using IORegistry path */
#ifdef illumos
if (vd->vdev_physpath != NULL &&
(dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
#endif
/*
* Note that we don't support the legacy auto-wholedisk support
* as above. This hasn't been used in a very long time and we
* don't need to propagate its oddities to this edge condition.
*/
if (error && vd->vdev_path != NULL)
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_dbgmsg(vd, "vdev_disk_open: failed to open [error=%d]",
error);
return (error);
}
/*
* XXX Apple - We must not set or modify the devid. Import on Solaris/illumos
* expects a valid devid and fails if it cannot be decoded.
*/
#ifdef illumos
/*
* Now that the device has been successfully opened, update the devid
* if necessary.
*/
if (validate_devid && spa_writeable(spa) &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
char *vd_devid;
vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
vdev_dbgmsg(vd, "vdev_disk_open: update devid from "
"'%s' to '%s'", vd->vdev_devid, vd_devid);
spa_strfree(vd->vdev_devid);
vd->vdev_devid = spa_strdup(vd_devid);
ddi_devid_str_free(vd_devid);
}
ddi_devid_free(devid);
}
#endif
/* XXX Apple to do, needs IORegistry physpath interface */
#ifdef illumos
/*
* Once a device is opened, verify that the physical device path (if
* available) is up to date.
*/
if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
char *physpath, *minorname;
physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
minorname = NULL;
if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
(vd->vdev_physpath == NULL ||
strcmp(vd->vdev_physpath, physpath) != 0)) {
if (vd->vdev_physpath)
spa_strfree(vd->vdev_physpath);
(void) strlcat(physpath, ":", MAXPATHLEN);
(void) strlcat(physpath, minorname, MAXPATHLEN);
vd->vdev_physpath = spa_strdup(physpath);
}
if (minorname)
kmem_free(minorname, strlen(minorname) + 1);
kmem_free(physpath, MAXPATHLEN);
}
#endif
/*
* Register callbacks for the LDI offline event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
}
/* XXX Apple to do - we could support the degrade event, or just no-op */
#ifdef illumos
/*
* Register callbacks for the LDI degrade event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
}
#endif
#if 0
int len = MAXPATHLEN;
if (vn_getpath(devvp, dvd->vd_readlinkname, &len) == 0) {
dprintf("ZFS: '%s' resolved name is '%s'\n",
vd->vdev_path, dvd->vd_readlinkname);
} else {
dvd->vd_readlinkname[0] = 0;
}
#endif
skip_open:
/*
* Determine the actual size of the device.
*/
if (ldi_get_size(dvd->vd_lh, psize) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_dbgmsg(vd, "vdev_disk_open: failed to get size");
return (SET_ERROR(EINVAL));
}
*max_psize = *psize;
/*
* Determine the device's minimum transfer size.
* If the ioctl isn't supported, assume DEV_BSIZE.
*/
if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
(intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
capacity = dkmext->dki_capacity - 1;
blksz = dkmext->dki_lbsize;
pbsize = dkmext->dki_pbsize;
} else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
(intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
VDEV_DEBUG(
"vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
vd->vdev_path);
capacity = dkm->dki_capacity - 1;
blksz = dkm->dki_lbsize;
pbsize = blksz;
} else {
VDEV_DEBUG("vdev_disk_open(\"%s\"): "
"both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
vd->vdev_path, error);
pbsize = DEV_BSIZE;
}
*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
/* XXX Now that we opened the device, determine if it is a whole disk. */
#ifdef __APPLE__
/*
* XXX Apple to do - provide an ldi_ mechanism
* to report whether this is a whole disk or a
* partition.
* Return 0 (no), 1 (yes), or -1 (error).
*/
// vd->vdev_wholedisk = ldi_is_wholedisk(vd->vd_lh);
#endif
if (vd->vdev_wholedisk == 1) {
int wce = 1;
/* Gets information about the disk if it has GPT partitions */
#ifdef illumos
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize = capacity * blksz;
}
#endif
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
#ifdef __APPLE__
/* Inform the ZIO pipeline that we are non-rotational */
vd->vdev_nonrot = B_FALSE;
if (ldi_ioctl(dvd->vd_lh, DKIOCISSOLIDSTATE, (intptr_t)&isssd,
FKIOCTL, kcred, NULL) == 0) {
vd->vdev_nonrot = (isssd ? B_TRUE : B_FALSE);
}
#endif //__APPLE__
return (0);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd;
ldi_ev_cookie_t ecookie;
vdev_disk_ldi_cb_t *lcb;
union {
struct dk_minfo_ext ude;
struct dk_minfo ud;
} dks;
struct dk_minfo_ext *dkmext = &dks.ude;
struct dk_minfo *dkm = &dks.ud;
int error;
dev_t dev;
int otyp;
boolean_t validate_devid = B_FALSE;
ddi_devid_t devid;
uint64_t capacity = 0, blksz = 0, pbsize;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
rw_enter(&vd->vdev_tsd_lock, RW_WRITER);
dvd = vd->vdev_tsd;
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
ASSERT(vd->vdev_reopening);
/*
* Here vd_lh is protected by vdev_tsd_lock
*/
ASSERT(dvd->vd_lh != NULL);
/* This should not happen, but let's be safe */
if (dvd->vd_lh == NULL) {
/* What are we going to do here??? */
rw_exit(&vd->vdev_tsd_lock);
return (SET_ERROR(ENXIO));
}
goto skip_open;
}
/*
* Create dvd to be used as vd->vdev_tsd.
*/
vd->vdev_tsd = dvd = vdev_disk_alloc();
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
vdev_disk_free_locked(vd);
rw_exit(&vd->vdev_tsd_lock);
return (SET_ERROR(EINVAL));
}
}
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
if (vd->vdev_wholedisk == -1ULL) {
size_t len = strlen(vd->vdev_path) + 3;
char *buf = kmem_alloc(len, KM_SLEEP);
(void) snprintf(buf, len, "%ss0", vd->vdev_path);
error = ldi_open_by_name(buf, spa_mode(spa), kcred,
&dvd->vd_lh, zfs_li);
if (error == 0) {
spa_strfree(vd->vdev_path);
vd->vdev_path = buf;
vd->vdev_wholedisk = 1ULL;
} else {
kmem_free(buf, len);
}
}
/*
* If we have not yet opened the device, try to open it by the
* specified path.
*/
if (error != 0) {
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
/*
* Compare the devid to the stored value.
*/
if (error == 0 && vd->vdev_devid != NULL &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
error = SET_ERROR(EINVAL);
(void) ldi_close(dvd->vd_lh, spa_mode(spa),
kcred);
dvd->vd_lh = NULL;
}
ddi_devid_free(devid);
}
/*
* If we succeeded in opening the device, but 'vdev_wholedisk'
* is not yet set, then this must be a slice.
*/
if (error == 0 && vd->vdev_wholedisk == -1ULL)
vd->vdev_wholedisk = 0;
}
/*
* If we were unable to open by path, or the devid check fails, open by
* devid instead.
*/
if (error != 0 && vd->vdev_devid != NULL) {
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
}
/*
* If all else fails, then try opening by physical path (if available)
* or the logical path (if we failed due to the devid check). While not
* as reliable as the devid, this will give us something, and the higher
* level vdev validation will prevent us from opening the wrong device.
*/
if (error) {
if (vd->vdev_devid != NULL)
validate_devid = B_TRUE;
if (vd->vdev_physpath != NULL &&
(dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
/*
* Note that we don't support the legacy auto-wholedisk support
* as above. This hasn't been used in a very long time and we
* don't need to propagate its oddities to this edge condition.
*/
if (error && vd->vdev_path != NULL)
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_disk_free_locked(vd);
rw_exit(&vd->vdev_tsd_lock);
return (error);
}
/*
* Now that the device has been successfully opened, update the devid
* if necessary.
*/
if (validate_devid && spa_writeable(spa) &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
char *vd_devid;
vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
zfs_dbgmsg("vdev %s: update devid from %s, "
"to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
spa_strfree(vd->vdev_devid);
vd->vdev_devid = spa_strdup(vd_devid);
ddi_devid_str_free(vd_devid);
}
ddi_devid_free(devid);
}
/*
* Once a device is opened, verify that the physical device path (if
* available) is up to date.
*/
if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
char *physpath, *minorname;
physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
minorname = NULL;
if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
(vd->vdev_physpath == NULL ||
strcmp(vd->vdev_physpath, physpath) != 0)) {
if (vd->vdev_physpath)
spa_strfree(vd->vdev_physpath);
(void) strlcat(physpath, ":", MAXPATHLEN);
(void) strlcat(physpath, minorname, MAXPATHLEN);
vd->vdev_physpath = spa_strdup(physpath);
}
if (minorname)
kmem_free(minorname, strlen(minorname) + 1);
kmem_free(physpath, MAXPATHLEN);
}
/*
* Register callbacks for the LDI offline event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_OFFLINE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_off_callb, (void *) vd, &lcb->lcb_id);
}
/*
* Register callbacks for the LDI degrade event.
*/
if (ldi_ev_get_cookie(dvd->vd_lh, LDI_EV_DEGRADE, &ecookie) ==
LDI_EV_SUCCESS) {
lcb = kmem_zalloc(sizeof (vdev_disk_ldi_cb_t), KM_SLEEP);
list_insert_tail(&dvd->vd_ldi_cbs, lcb);
(void) ldi_ev_register_callbacks(dvd->vd_lh, ecookie,
&vdev_disk_dgrd_callb, (void *) vd, &lcb->lcb_id);
}
/* Reset TRIM flag, as underlying device support may have changed */
vd->vdev_notrim = B_FALSE;
skip_open:
ASSERT(dvd != NULL);
/*
* Determine the actual size of the device.
*/
if (ldi_get_size(dvd->vd_lh, psize) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
vdev_disk_free_locked(vd);
rw_exit(&vd->vdev_tsd_lock);
return (SET_ERROR(EINVAL));
}
*max_psize = *psize;
/*
* Determine the device's minimum transfer size.
* If the ioctl isn't supported, assume DEV_BSIZE.
*/
if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
(intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
capacity = dkmext->dki_capacity - 1;
blksz = dkmext->dki_lbsize;
pbsize = dkmext->dki_pbsize;
} else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
(intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
VDEV_DEBUG(
"vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
vd->vdev_path);
capacity = dkm->dki_capacity - 1;
blksz = dkm->dki_lbsize;
pbsize = blksz;
} else {
VDEV_DEBUG("vdev_disk_open(\"%s\"): "
"both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
vd->vdev_path, error);
pbsize = DEV_BSIZE;
}
*ashift = highbit64(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
if (vd->vdev_wholedisk == 1) {
int wce = 1;
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize += vdev_disk_get_space(vd, capacity, blksz);
zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
"max_psize %llu", vd->vdev_path, *psize,
*max_psize);
}
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
/*
* We are done with vd_lh and vdev_tsd, release the vdev_tsd_lock
*/
rw_exit(&vd->vdev_tsd_lock);
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
return (0);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd;
struct dk_minfo_ext dkmext;
int error;
dev_t dev;
int otyp;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (vd->vdev_tsd != NULL) {
ASSERT(vd->vdev_reopening);
dvd = vd->vdev_tsd;
goto skip_open;
}
dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the save device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
}
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
ddi_devid_t devid;
if (vd->vdev_wholedisk == -1ULL) {
size_t len = strlen(vd->vdev_path) + 3;
char *buf = kmem_alloc(len, KM_SLEEP);
ldi_handle_t lh;
(void) snprintf(buf, len, "%ss0", vd->vdev_path);
if (ldi_open_by_name(buf, spa_mode(spa), kcred,
&lh, zfs_li) == 0) {
spa_strfree(vd->vdev_path);
vd->vdev_path = buf;
vd->vdev_wholedisk = 1ULL;
(void) ldi_close(lh, spa_mode(spa), kcred);
} else {
kmem_free(buf, len);
}
}
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
&dvd->vd_lh, zfs_li);
/*
* Compare the devid to the stored value.
*/
if (error == 0 && vd->vdev_devid != NULL &&
ldi_get_devid(dvd->vd_lh, &devid) == 0) {
if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
error = EINVAL;
(void) ldi_close(dvd->vd_lh, spa_mode(spa),
kcred);
dvd->vd_lh = NULL;
}
ddi_devid_free(devid);
}
/*
* If we succeeded in opening the device, but 'vdev_wholedisk'
* is not yet set, then this must be a slice.
*/
if (error == 0 && vd->vdev_wholedisk == -1ULL)
vd->vdev_wholedisk = 0;
}
/*
* If we were unable to open by path, or the devid check fails, open by
* devid instead.
*/
if (error != 0 && vd->vdev_devid != NULL)
error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
/*
* If all else fails, then try opening by physical path (if available)
* or the logical path (if we failed due to the devid check). While not
* as reliable as the devid, this will give us something, and the higher
* level vdev validation will prevent us from opening the wrong device.
*/
if (error) {
if (vd->vdev_physpath != NULL &&
(dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
/*
* Note that we don't support the legacy auto-wholedisk support
* as above. This hasn't been used in a very long time and we
* don't need to propagate its oddities to this edge condition.
*/
if (error && vd->vdev_path != NULL)
error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
kcred, &dvd->vd_lh, zfs_li);
}
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (error);
}
/*
* Once a device is opened, verify that the physical device path (if
* available) is up to date.
*/
if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
char *physpath, *minorname;
physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
minorname = NULL;
if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
(vd->vdev_physpath == NULL ||
strcmp(vd->vdev_physpath, physpath) != 0)) {
if (vd->vdev_physpath)
spa_strfree(vd->vdev_physpath);
(void) strlcat(physpath, ":", MAXPATHLEN);
(void) strlcat(physpath, minorname, MAXPATHLEN);
vd->vdev_physpath = spa_strdup(physpath);
}
if (minorname)
kmem_free(minorname, strlen(minorname) + 1);
kmem_free(physpath, MAXPATHLEN);
}
skip_open:
/*
* Determine the actual size of the device.
*/
if (ldi_get_size(dvd->vd_lh, psize) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (EINVAL);
}
/*
* If we own the whole disk, try to enable disk write caching.
* We ignore errors because it's OK if we can't do it.
*/
if (vd->vdev_wholedisk == 1) {
int wce = 1;
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
/*
* Determine the device's minimum transfer size.
* If the ioctl isn't supported, assume DEV_BSIZE.
*/
if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
FKIOCTL, kcred, NULL) != 0)
dkmext.dki_pbsize = DEV_BSIZE;
*ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
return (0);
}
