static void
msgringent_free (msgringent * ktmp)
{
int i;
for(i=0; i<ktmp->body.n; i++)
ppage_unpin (kva2pp ((u_long) ktmp->body.r[i].data));
if (ktmp->owner != NULL)
ppage_unpin (kva2pp ((u_long) ktmp->owner));
free (ktmp);
}
void
disk_buf_free (struct buf *bp)
{
u_int datalen;
u_int data, next;
/* unpin pages */
if (bp->b_flags & B_SCSICMD) {
struct scsicmd *scsicmd = (struct scsicmd *) bp->b_memaddr;
data = (u_int) scsicmd->data_addr;
datalen = scsicmd->datalen;
free (scsicmd->scsi_cmd);
free (scsicmd);
} else {
data = (u_int)bp->b_memaddr;
datalen = bp->b_bcount;
}
while (datalen > 0) {
if (bp->b_flags & B_BC_REQ) {
ppage_unpin (kva2pp (data));
} else {
struct Env *e;
int r;
e = env_id2env (bp->b_envid, &r);
assert (e); /* XXX - what if an env with an outstanding request dies? */
ppage_unpin (&ppages[PGNO (*env_va2ptep (e, data))]);
}
/* go to start of next page of data */
next = (data & ~PGMASK) + NBPG;
if (next - data >= datalen) break;
assert (next > data);
datalen -= next - data;
data = next;
}
if (bp->b_resptr) {
ppage_unpin (kva2pp ((u_int) bp->b_resptr));
}
free (bp);
}
int
disk_prepare_bc_request (u_int devno, u_quad_t blkno, void *vaddr, u_int flags,
int *resptr, struct buf **headbpp)
{
struct buf *bp;
/* XXX - test for big blkno wraparound */
if ((devno >= si->si_ndisks) ||
(((blkno * NBPG) / si->si_disks[devno].d_bsize) >=
si->si_disks[devno].d_size)) {
warn ("disk_prepare_bc_request: invalid devno (%u) or blkno (%qu)",
devno, blkno);
return (-E_INVAL);
}
if (headbpp == NULL) {
warn ("disk_prepare_bc_request: headbpp == NULL");
return (-E_INVAL);
}
bp = disk_buf_alloc();
if (!bp)
return (-E_NO_MEM);
bp->b_next = NULL;
bp->b_sgnext = NULL;
bp->b_flags = flags;
bp->b_dev = devno;
bp->b_blkno = (blkno * NBPG) / si->si_disks[devno].d_bsize;
bp->b_bcount = NBPG;
bp->b_sgtot = 0;
bp->b_memaddr = vaddr;
bp->b_envid = curenv->env_id;
bp->b_resid = 0;
bp->b_resptr = NULL;
ppage_pin (kva2pp((u_int) vaddr));
if (*headbpp) {
struct buf *tmpbp = *headbpp;
while (tmpbp->b_flags & B_SCATGATH) {
tmpbp = (struct buf *) tmpbp->b_sgnext;
if (tmpbp == NULL) {
warn ("disk_prepare_bc_request: bad scatter/gather list");
ppage_unpin (kva2pp((u_int) vaddr));
free (bp);
return (-E_INVAL);
}
}
/* XXX - test for big blkno wraparound */
if (bp->b_blkno != (tmpbp->b_blkno +
(tmpbp->b_bcount /
si->si_disks[bp->b_dev].d_bsize))) {
warn ("disk_prepare_bc_request: noncontiguous requests "
"(prevblk %qu, size %u, curblk %qu) can't be merged",
tmpbp->b_blkno, tmpbp->b_bcount, bp->b_blkno);
ppage_unpin (kva2pp((u_int) vaddr));
free (bp);
return (-E_INVAL);
}
(*headbpp)->b_sgtot += NBPG;
tmpbp->b_flags |= B_SCATGATH;
tmpbp->b_sgnext = bp;
} else {
*headbpp = bp;
bp->b_sgtot = NBPG;
}
if (resptr) {
ppage_pin (kva2pp(((u_int) resptr)));
}
bp->b_resptr = resptr;
return (0);
}
/*
* sys_disk_request
*
* Disk I/O without going through the buffer cache.
*
* xn_user is the name of a pxn that grants access to the disk
* reqbp is a list of scatter/gather requests
* k is which capability in the env should be checked
*
* permission is granted to perform the operation if:
* 1) the blocks in reqbp are covered by the pxn
* 2) the capability gives access to the pxn
*
*/
int
sys_disk_request (u_int sn, struct Xn_name *xn_user, struct buf *reqbp,
u_int k)
{
struct Xn_name xn;
struct Xn_xtnt xtnt;
struct Pxn *pxn;
cap c;
int ret;
int access;
struct disk *di;
int *resptr = NULL;
u_int bcount = 0;
struct buf *bp, *segbp, *nsegbp;
int noncontigs = 0, nctemp;
#ifdef MEASURE_DISK_TIMES
disk_pctr_start = rdtsc();
#endif
/* XXX - use PFM or copyin instead of isreadable_* */
/* bypass for direct scsi commands */
if (reqbp->b_flags & B_SCSICMD) {
return sys_disk_scsicmd (sn, k, reqbp);
}
/* get the capability */
if ((ret = env_getcap (curenv, k, &c)) < 0)
return ret;
/* and the pxn */
copyin (xn_user, &xn, sizeof (xn));
if (! (pxn = lookup_pxn (&xn))) {
warn ("sys_disk_request: no pxn found");
return (-E_NOT_FOUND);
}
/* XXX - do we need to check that this is a physical disk? */
/* get a refernce to the disk unit for this command */
di = &(si->si_disks[xn.xa_dev]);
/* Iterate over the request list checking:
-- if the request is transfering data to/from
memory that this user can read/write.
-- if the pxn and capability specified give
access to these blocks */
for (segbp = reqbp; ; segbp = (struct buf *) segbp->b_sgnext) {
if (! (isreadable_varange ((u_int)segbp, sizeof(struct buf)))) {
warn ("sys_disk_request: bad reqbp (%p)", segbp);
return (-E_FAULT);
}
if (segbp->b_flags & B_READ) {
access = ACL_R;
} else {
access = ACL_W;
}
xtnt.xtnt_block = segbp->b_blkno;
xtnt.xtnt_size = segbp->b_bcount / di->d_bsize;
bcount += segbp->b_bcount;
if (! pxn_authorizes_xtnt (pxn, &c, &xtnt, access, &ret)) {
warn ("sys_disk_request: pxn/cap does not grant access to block(s)");
return ret;
}
if (! ((reqbp->b_flags & B_READ) ? iswriteable_varange
: isreadable_varange) ((u_int) segbp->b_memaddr, segbp->b_bcount)) {
warn ("sys_disk_request: bad b_memaddr: %p (b_bcount %d)",
segbp->b_memaddr, segbp->b_bcount);
return (-E_FAULT);
}
if (! (segbp->b_flags & B_SCATGATH)) {
if (segbp->b_resptr) {
resptr = segbp->b_resptr;
if ((((u_int) resptr) % sizeof(u_int)) ||
!(isvawriteable (resptr))) {
warn ("sys_disk_request: bad resptr (%p)", resptr);
return (-E_FAULT);
}
resptr = (int *) pa2kva (va2pa (resptr));
}
break;
}
}
if ((reqbp->b_flags & B_SCATGATH) && bcount != reqbp->b_sgtot) {
warn ("sys_disk_request: invalid scatter/gather, with total (%u) unequal "
"to sum of parts (%u)", reqbp->b_sgtot, bcount);
return (-E_INVAL);
}
/* are we done before we've started? */
if (bcount == 0) {
if (resptr)
(*resptr)--;
return (0);
}
if (bcount & di->d_bmod) {
warn ("sys_disk_request: bad bcount %u", bcount);
return (-E_INVAL);
}
/* copy request into kernel buffer */
segbp = reqbp;
nsegbp = NULL;
reqbp = NULL;
do {
segbp->b_dev = di->d_id;
bp = copy_and_pin(segbp, segbp->b_bcount, &nctemp);
if (!bp) {
warn ("sys_disk_request: could not copy_and_pin");
/* XXX - cleanup before returning */
return (-E_NO_MEM);
}
noncontigs += nctemp;
if (nsegbp) nsegbp->b_sgnext = bp;
if (!reqbp) reqbp = bp;
if (noncontigs >= DISK_MAX_SCATTER) {
warn ("sys_disk_request: would require too many scatter/gather entries "
"(%d)", noncontigs);
/* XXX - cleanup before returning */
return (-E_INVAL);
}
nsegbp = bp;
segbp = segbp->b_sgnext;
} while (nsegbp->b_flags & B_SCATGATH);
nsegbp->b_resptr = resptr;
if (resptr) ppage_pin (kva2pp((u_int) resptr));
/* call appropriate strategy routine */
di->d_strategy (reqbp);
#ifdef MEASURE_DISK_TIMES
disk_pctr_return = rdtsc();
#endif
return (0);
}
/* XXX - we should use copyin, etc, instead of isreadable_* so that user will
get pagefaults he can handle transparently */
static int
sys_disk_scsicmd (u_int sn, u_int k, struct buf *reqbp)
{
struct buf *bp;
struct scsicmd *scsicmd = (struct scsicmd *) reqbp->b_memaddr;
struct scsicmd *scsicmd2;
int noncontigs;
struct disk *di;
/* must have root capability for system to do a raw SCSI command!! */
/* XXX -- later, if desired, deeper checking of validity can reduce */
/* this restriction... */
if (k >= curenv->env_clen || ! curenv->env_clist[k].c_valid) {
warn ("sys_disk_scsicmd: bad capability number %u\n", k);
return (-E_CAP_INVALID);
}
if (! cap_isroot(&curenv->env_clist[k])) {
warn ("sys_disk_scsicmd: cap %u is not root capability for system\n", k);
return (-E_CAP_INSUFF);
}
/* must be able to read the reqbp ... */
if (! (isreadable_varange ((u_int) reqbp, sizeof (struct buf)))) {
warn ("sys_disk_scsicmd: bad reqbp (%p)", reqbp);
return (-E_FAULT);
}
/* Should be a SCSICMD */
if (! (reqbp->b_flags & B_SCSICMD)) {
warn ("sys_disk_scsicmd: not a B_SCSICMD\n");
return (-E_INVAL);
}
/* Must be proper environment */
if (reqbp->b_envid != curenv->env_id) {
warn ("sys_disk_scsicmd: bad envid\n");
return (-E_INVAL);
}
/* no scatter/gather support for raw SCSI commands */
if (reqbp->b_flags & B_SCATGATH) {
warn ("sys_disk_scsicmd: B_SCATGATH not allowed with B_SCSICMD\n");
return (-E_INVAL);
}
/* can't send request to non-existent disk... */
if (reqbp->b_dev >= si->si_ndevs) {
warn ("sys_disk_scsicmd: there is no disk %u in system\n", reqbp->b_dev);
return (-E_NOT_FOUND);
}
/* check that everything is readable */
if (! isreadable_varange ((u_int) reqbp->b_memaddr,
sizeof (struct scsicmd))) {
warn ("sys_disk_scsicmd: SCSI command description is not readable\n");
return (-E_FAULT);
}
if (! isreadable_varange ((u_int) scsicmd->scsi_cmd, scsicmd->cmdlen) ) {
warn ("sys_disk_scsicmd: SCSI command itself is not readable\n");
return (-E_FAULT);
}
if (! isreadable_varange ((u_int)scsicmd->data_addr, scsicmd->datalen) ) {
warn ("sys_disk_scsicmd: data area for SCSI command is not readable\n");
return (-E_FAULT);
}
/* length of SCSI command must not be greater than B_SCSICMD_MAXLEN */
if (scsicmd->cmdlen > B_SCSICMD_MAXLEN) {
/* XXX - why do we compare scsicmd->cmdlen, but we print out
reqbp->b_bcount? */
warn ("sys_disk_scsicmd: specified SCSI command too large (%d > %d)\n",
reqbp->b_bcount, B_SCSICMD_MAXLEN);
return (-E_INVAL);
}
/* copy the SCSI command to avoid sharing it with app */
bp = bp_copy (reqbp);
if (bp == NULL) {
warn ("sys_disk_scsicmd: kernel malloc for bp failed\n");
return (-E_NO_MEM);
}
bp->b_memaddr = malloc (sizeof (struct scsicmd));
if (bp->b_memaddr == NULL) {
warn ("sys_disk_scsicmd: kernel malloc for scsicmd failed\n");
free (bp);
return (-E_NO_MEM);
}
scsicmd2 = (struct scsicmd *) bp->b_memaddr;
bcopy (scsicmd, scsicmd2, sizeof (struct scsicmd));
scsicmd2->scsi_cmd = (struct scsi_generic *) malloc (scsicmd->cmdlen);
if (scsicmd2->scsi_cmd == NULL) {
warn ("sys_disk_scsicmd: second kernel malloc failed\n");
free (bp->b_memaddr);
free (bp);
return (-E_NO_MEM);
}
bcopy (scsicmd->scsi_cmd, scsicmd2->scsi_cmd, scsicmd->cmdlen);
scsicmd2->bp = bp;
bp->b_resid = scsicmd->datalen;
bp->b_resptr = (int *) pa2kva (va2pa (reqbp->b_resptr));
/* pin down the app pages that will later be used by the driver */
ppage_pin (kva2pp ((u_int) bp->b_resptr));
noncontigs = pin_and_count_noncontigs (scsicmd2->data_addr,
scsicmd2->datalen);
if (noncontigs >= DISK_MAX_SCATTER) {
warn ("sys_disk_scsicmd: will require too many scatter/gather entries "
"(%d)", noncontigs);
disk_buf_free (bp);
return (-E_TOO_BIG);
}
/* XXX */
/* call down to the low-level driver. GROK -- since the partition stuff */
/* creates and abstract disk that is separate from the real one, a hack */
/* is needed to get the actual disk strategy routine for raw SCSI commands */
/* This is fine as long as all disks actually go to the same strategy */
/* routine. */
di = &(si->si_disks[0]);
di->d_strategy (bp);
return (0);
}
static msgringent *
msgringent_setup (msgringent * u_msgringent)
{
msgringent *ktmp;
Pte *pte = NULL;
int scatptr = 0;
int total_len = 0;
ktmp = (msgringent *) malloc (sizeof (msgringent));
if (ktmp == NULL)
{
warn ("msgringent_setup: failed malloc");
return (NULL);
}
ktmp->appaddr = u_msgringent;
ktmp->owner = NULL;
ktmp->body.n = 0;
/* Verify and translate owner field */
if ((((u_int) u_msgringent->owner % sizeof (int)) ||
! (pte = va2ptep ((u_int) u_msgringent->owner)) ||
((*pte & WRITE_MASK) != WRITE_MASK)))
{
warn ("msgringent_setup: owner field failed\n");
msgringent_free (ktmp);
return (NULL);
}
ktmp->owner = (u_int *) pa2kva (va2pa (u_msgringent->owner));
ppage_pin (kva2pp ((u_long) ktmp->owner));
/* Verify and translate data field */
if (u_msgringent->body.n > 1)
{
warn ("msgringent_setup: not allowed to setup disjoint message body\n");
msgringent_free (ktmp);
return (NULL);
}
scatptr = 0;
total_len = 0;
{
int len = u_msgringent->body.r[0].sz;
caddr_t addr = u_msgringent->body.r[0].data;
u_int pagebound = NBPG-(((u_long)addr)&(NBPG - 1));
while (len > 0)
{
u_int slen = min (len, pagebound);
if (!(pte = va2ptep ((u_int) addr)) ||
((*pte & READ_MASK) != READ_MASK))
{
/* physical page is not accessible */
warn ("msgringent_setup: can't read scatter ptr\n");
msgringent_free (ktmp);
return (NULL);
}
ktmp->body.r[scatptr].data = (char *) pa2kva (va2pa (addr));
ktmp->body.r[scatptr].sz = slen;
ktmp->body.n++;
/* pin the page to prevent re-allocation */
ppage_pin (kva2pp ((u_long) ktmp->body.r[scatptr].data));
len -= slen;
addr += slen;
total_len += slen;
pagebound = NBPG;
scatptr++;
if (scatptr > IPC_MAX_SCATTER_PTR || total_len > IPC_MAX_MSG_SIZE)
{
msgringent_free (ktmp);
warn ("msgringent_setup: message body too big\n");
return (NULL);
}
}
}
return (ktmp);
}
