uintptr_t process_load_elf(struct process *p, uintptr_t elf)
{
int need_kernel = 0;
Elf32_Ehdr *hdr = (void *)elf;
Elf32_Phdr *phdr = (void *)((char *)hdr + hdr->e_phoff);
struct thread *t = thread_current();
/*
* If the process address space is not the current one we need extra
* mapping in the kernel to write on the pages
*/
if (!t || t->parent->as != p->as)
need_kernel = 1;
for (uint32_t i = 0; i < hdr->e_phnum; ++i)
{
if (phdr[i].p_type != PT_LOAD)
continue;
vaddr_t vaddr;
paddr_t paddr;
size_t page_size = align(phdr[i].p_memsz, PAGE_SIZE) / PAGE_SIZE;
/* TODO: Error handling */
vaddr = region_reserve(p->as, phdr[i].p_vaddr, page_size);
/* TODO: Error handling */
paddr = segment_alloc_address(page_size);
/* TODO: Error handling */
vaddr = as_map(p->as, vaddr, paddr, phdr[i].p_memsz,
AS_MAP_USER | AS_MAP_WRITE);
/* TODO: Error handling */
if (need_kernel)
vaddr = as_map(&kernel_as, 0, paddr, phdr[i].p_memsz,
AS_MAP_WRITE);
memcpy((void *)vaddr, (void *)(elf + phdr[i].p_offset),
phdr[i].p_filesz);
memset((char *)vaddr + phdr[i].p_filesz, 0,
phdr[i].p_memsz - phdr[i].p_filesz);
if (need_kernel)
as_unmap(&kernel_as, vaddr, AS_UNMAP_NORELEASE);
}
return hdr->e_entry;
}
/*
* Used for MAP_ANON - fast way to get anonymous pages
*/
static int
zmap(struct as *as, caddr_t *addrp, size_t len, uint_t uprot, int flags,
offset_t pos)
{
struct segvn_crargs vn_a;
int error;
if (((PROT_ALL & uprot) != uprot))
return (EACCES);
if ((flags & MAP_FIXED) != 0) {
caddr_t userlimit;
/*
* Use the user address. First verify that
* the address to be used is page aligned.
* Then make some simple bounds checks.
*/
if (((uintptr_t)*addrp & PAGEOFFSET) != 0)
return (EINVAL);
userlimit = flags & _MAP_LOW32 ?
(caddr_t)USERLIMIT32 : as->a_userlimit;
switch (valid_usr_range(*addrp, len, uprot, as, userlimit)) {
case RANGE_OKAY:
break;
case RANGE_BADPROT:
return (ENOTSUP);
case RANGE_BADADDR:
default:
return (ENOMEM);
}
}
/*
* No need to worry about vac alignment for anonymous
* pages since this is a "clone" object that doesn't
* yet exist.
*/
error = choose_addr(as, addrp, len, pos, ADDR_NOVACALIGN, flags);
if (error != 0) {
return (error);
}
/*
* Use the seg_vn segment driver; passing in the NULL amp
* gives the desired "cloning" effect.
*/
vn_a.vp = NULL;
vn_a.offset = 0;
vn_a.type = flags & MAP_TYPE;
vn_a.prot = uprot;
vn_a.maxprot = PROT_ALL;
vn_a.flags = flags & ~MAP_TYPE;
vn_a.cred = CRED();
vn_a.amp = NULL;
vn_a.szc = 0;
vn_a.lgrp_mem_policy_flags = 0;
return (as_map(as, *addrp, len, segvn_create, &vn_a));
}
/*ARGSUSED*/
static int
smb_segmap(dev_t dev, off_t off, struct as *as, caddr_t *addrp, off_t len,
uint_t prot, uint_t maxprot, uint_t flags, cred_t *cred)
{
smb_clone_t *cp = &smb_clones[getminor(dev)];
size_t alen = P2ROUNDUP(len, PAGESIZE);
caddr_t addr;
iovec_t iov;
uio_t uio;
int err;
if (len <= 0 || (flags & MAP_FIXED))
return (EINVAL);
if ((prot & PROT_WRITE) && (flags & MAP_SHARED))
return (EACCES);
if (off < 0 || off + len < off || off + len > cp->c_eplen + cp->c_stlen)
return (ENXIO);
as_rangelock(as);
map_addr(&addr, alen, 0, 1, 0);
if (addr != NULL)
err = as_map(as, addr, alen, segvn_create, zfod_argsp);
else
err = ENOMEM;
as_rangeunlock(as);
*addrp = addr;
if (err != 0)
return (err);
iov.iov_base = addr;
iov.iov_len = len;
bzero(&uio, sizeof (uio_t));
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = off;
uio.uio_segflg = UIO_USERSPACE;
uio.uio_extflg = UIO_COPY_DEFAULT;
uio.uio_resid = len;
if ((err = smb_uiomove(cp, &uio)) != 0)
(void) as_unmap(as, addr, alen);
return (err);
}
/*
* rlen is in multiples of PAGESIZE
*/
static char *
ksyms_asmap(struct as *as, size_t rlen)
{
char *addr = NULL;
as_rangelock(as);
map_addr(&addr, rlen, 0, 1, 0);
if (addr == NULL || as_map(as, addr, rlen, segvn_create, zfod_argsp)) {
as_rangeunlock(as);
return (NULL);
}
as_rangeunlock(as);
return (addr);
}
/*ARGSUSED*/
static int
bootfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
caller_context_t *ct)
{
int ret;
segvn_crargs_t vn_a;
#ifdef _ILP32
if (len > MAXOFF_T)
return (ENOMEM);
#endif
if (vp->v_flag & VNOMAP)
return (ENOSYS);
if (off < 0 || off > MAXOFFSET_T - off)
return (ENXIO);
if (vp->v_type != VREG)
return (ENODEV);
if (prot & PROT_WRITE)
return (ENOTSUP);
as_rangelock(as);
ret = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
if (ret != 0) {
as_rangeunlock(as);
return (ret);
}
vn_a.vp = vp;
vn_a.offset = (u_offset_t)off;
vn_a.type = flags & MAP_TYPE;
vn_a.prot = prot;
vn_a.maxprot = maxprot;
vn_a.cred = cr;
vn_a.amp = NULL;
vn_a.flags = flags & ~MAP_TYPE;
vn_a.szc = 0;
vn_a.lgrp_mem_policy_flags = 0;
ret = as_map(as, *addrp, len, segvn_create, &vn_a);
as_rangeunlock(as);
return (ret);
}
/*ARGSUSED8*/
static int
privcmd_segmap(dev_t dev, off_t off, struct as *as, caddr_t *addrp,
off_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
{
struct segmf_crargs a;
int error;
if (secpolicy_xvm_control(cr))
return (EPERM);
as_rangelock(as);
if ((flags & MAP_FIXED) == 0) {
map_addr(addrp, len, (offset_t)off, 0, flags);
if (*addrp == NULL) {
error = ENOMEM;
goto rangeunlock;
}
} else {
/*
* User specified address
*/
(void) as_unmap(as, *addrp, len);
}
/*
* The mapping *must* be MAP_SHARED at offset 0.
*
* (Foreign pages are treated like device memory; the
* ioctl interface allows the backing objects to be
* arbitrarily redefined to point at any machine frame.)
*/
if ((flags & MAP_TYPE) != MAP_SHARED || off != 0) {
error = EINVAL;
goto rangeunlock;
}
a.dev = dev;
a.prot = (uchar_t)prot;
a.maxprot = (uchar_t)maxprot;
error = as_map(as, *addrp, len, segmf_create, &a);
rangeunlock:
as_rangeunlock(as);
return (error);
}
/*
* This function is called when a page needs to be mapped into a
* process's address space. Allocate the user address space and
* set up the mapping to the page. Assumes the page has already
* been allocated and locked in memory via schedctl_getpage.
*/
static int
schedctl_map(struct anon_map *amp, caddr_t *uaddrp, caddr_t kaddr)
{
caddr_t addr = NULL;
struct as *as = curproc->p_as;
struct segvn_crargs vn_a;
int error;
as_rangelock(as);
/* pass address of kernel mapping as offset to avoid VAC conflicts */
map_addr(&addr, PAGESIZE, (offset_t)(uintptr_t)kaddr, 1, 0);
if (addr == NULL) {
as_rangeunlock(as);
return (ENOMEM);
}
/*
* Use segvn to set up the mapping to the page.
*/
vn_a.vp = NULL;
vn_a.offset = 0;
vn_a.cred = NULL;
vn_a.type = MAP_SHARED;
vn_a.prot = vn_a.maxprot = PROT_ALL;
vn_a.flags = 0;
vn_a.amp = amp;
vn_a.szc = 0;
vn_a.lgrp_mem_policy_flags = 0;
error = as_map(as, addr, PAGESIZE, segvn_create, &vn_a);
as_rangeunlock(as);
if (error)
return (error);
*uaddrp = addr;
return (0);
}
static int
xmem_map(struct vnode *vp, offset_t off, struct as *as, caddr_t *addrp,
size_t len, uchar_t prot, uchar_t maxprot, uint_t flags,
struct cred *cred)
{
struct seg *seg;
struct segxmem_crargs xmem_a;
struct xmemnode *xp = (struct xmemnode *)VTOXN(vp);
struct xmount *xm = (struct xmount *)VTOXM(vp);
uint_t blocknumber;
int error;
#ifdef lint
maxprot = maxprot;
#endif
if (vp->v_flag & VNOMAP)
return (ENOSYS);
if (off < 0)
return (EINVAL);
/* offset, length and address has to all be block aligned */
if (off & (xm->xm_bsize - 1) || len & (xm->xm_bsize - 1) ||
((ulong_t)*addrp) & (xm->xm_bsize - 1)) {
return (EINVAL);
}
if (vp->v_type != VREG)
return (ENODEV);
if (flags & MAP_PRIVATE)
return (EINVAL); /* XXX need to be handled */
/*
* Don't allow mapping to locked file
*/
if (vn_has_mandatory_locks(vp, xp->xn_mode)) {
return (EAGAIN);
}
if (error = xmem_fillpages(xp, vp, off, len, 1)) {
return (error);
}
blocknumber = off >> xm->xm_bshift;
if (flags & MAP_FIXED) {
/*
* User specified address - blow away any previous mappings
*/
AS_LOCK_ENTER(as, &as->a_lock, RW_WRITER);
seg = as_findseg(as, *addrp, 0);
/*
* Fast path. segxmem_remap will fail if this is the wrong
* segment or if the len is beyond end of seg. If it fails,
* we do the regular stuff thru as_* routines.
*/
if (seg && (segxmem_remap(seg, vp, *addrp, len,
&xp->xn_ppa[blocknumber], prot) == 0)) {
AS_LOCK_EXIT(as, &as->a_lock);
return (0);
}
AS_LOCK_EXIT(as, &as->a_lock);
if (seg)
(void) as_unmap(as, *addrp, len);
as_rangelock(as);
error = valid_usr_range(*addrp, len, prot, as, as->a_userlimit);
if (error != RANGE_OKAY ||
as_gap(as, len, addrp, &len, AH_CONTAIN, *addrp)) {
as_rangeunlock(as);
return (EINVAL);
}
} else {
as_rangelock(as);
map_addr(addrp, len, (offset_t)off, 1, flags);
}
if (*addrp == NULL) {
as_rangeunlock(as);
return (ENOMEM);
}
xmem_a.xma_vp = vp;
xmem_a.xma_offset = (u_offset_t)off;
xmem_a.xma_prot = prot;
xmem_a.xma_cred = cred;
xmem_a.xma_ppa = &xp->xn_ppa[blocknumber];
xmem_a.xma_bshift = xm->xm_bshift;
error = as_map(as, *addrp, len, segxmem_create, &xmem_a);
as_rangeunlock(as);
return (error);
}
/*
* This function is called when a memory device is mmap'ed.
* Set up the mapping to the correct device driver.
*/
static int
mmsegmap(dev_t dev, off_t off, struct as *as, caddr_t *addrp, off_t len,
uint_t prot, uint_t maxprot, uint_t flags, struct cred *cred)
{
struct segvn_crargs vn_a;
struct segdev_crargs dev_a;
int error;
minor_t minor;
off_t i;
minor = getminor(dev);
as_rangelock(as);
/*
* No need to worry about vac alignment on /dev/zero
* since this is a "clone" object that doesn't yet exist.
*/
error = choose_addr(as, addrp, len, off,
(minor == M_MEM) || (minor == M_KMEM), flags);
if (error != 0) {
as_rangeunlock(as);
return (error);
}
switch (minor) {
case M_MEM:
/* /dev/mem cannot be mmap'ed with MAP_PRIVATE */
if ((flags & MAP_TYPE) != MAP_SHARED) {
as_rangeunlock(as);
return (EINVAL);
}
/*
* Check to ensure that the entire range is
* legal and we are not trying to map in
* more than the device will let us.
*/
for (i = 0; i < len; i += PAGESIZE) {
if (mmmmap(dev, off + i, maxprot) == -1) {
as_rangeunlock(as);
return (ENXIO);
}
}
/*
* Use seg_dev segment driver for /dev/mem mapping.
*/
dev_a.mapfunc = mmmmap;
dev_a.dev = dev;
dev_a.offset = off;
dev_a.type = (flags & MAP_TYPE);
dev_a.prot = (uchar_t)prot;
dev_a.maxprot = (uchar_t)maxprot;
dev_a.hat_attr = 0;
/*
* Make /dev/mem mappings non-consistent since we can't
* alias pages that don't have page structs behind them,
* such as kernel stack pages. If someone mmap()s a kernel
* stack page and if we give him a tte with cv, a line from
* that page can get into both pages of the spitfire d$.
* But snoop from another processor will only invalidate
* the first page. This later caused kernel (xc_attention)
* to go into an infinite loop at pil 13 and no interrupts
* could come in. See 1203630.
*
*/
dev_a.hat_flags = HAT_LOAD_NOCONSIST;
dev_a.devmap_data = NULL;
error = as_map(as, *addrp, len, segdev_create, &dev_a);
break;
case M_ZERO:
/*
* Use seg_vn segment driver for /dev/zero mapping.
* Passing in a NULL amp gives us the "cloning" effect.
*/
vn_a.vp = NULL;
vn_a.offset = 0;
vn_a.type = (flags & MAP_TYPE);
vn_a.prot = prot;
vn_a.maxprot = maxprot;
vn_a.flags = flags & ~MAP_TYPE;
vn_a.cred = cred;
vn_a.amp = NULL;
vn_a.szc = 0;
vn_a.lgrp_mem_policy_flags = 0;
error = as_map(as, *addrp, len, segvn_create, &vn_a);
break;
case M_KMEM:
case M_ALLKMEM:
/* No longer supported with KPR. */
error = ENXIO;
break;
case M_NULL:
/*
* Use seg_dev segment driver for /dev/null mapping.
*/
dev_a.mapfunc = mmmmap;
dev_a.dev = dev;
dev_a.offset = off;
dev_a.type = 0; /* neither PRIVATE nor SHARED */
dev_a.prot = dev_a.maxprot = (uchar_t)PROT_NONE;
dev_a.hat_attr = 0;
dev_a.hat_flags = 0;
error = as_map(as, *addrp, len, segdev_create, &dev_a);
break;
default:
error = ENXIO;
}
as_rangeunlock(as);
return (error);
}
/* ARGSUSED */
int
gfs_vop_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cred,
caller_context_t *ct)
{
int rv;
ssize_t resid = len;
/*
* Check for bad parameters
*/
#ifdef _ILP32
if (len > MAXOFF_T)
return (ENOMEM);
#endif
if (vp->v_flag & VNOMAP)
return (ENOTSUP);
if (off > MAXOFF_T)
return (EFBIG);
if ((long)off < 0 || (long)(off + len) < 0)
return (EINVAL);
if (vp->v_type != VREG)
return (ENODEV);
if ((prot & (PROT_EXEC | PROT_WRITE)) != 0)
return (EACCES);
/*
* Find appropriate address if needed, otherwise clear address range.
*/
as_rangelock(as);
rv = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
if (rv != 0) {
as_rangeunlock(as);
return (rv);
}
/*
* Create mapping
*/
rv = as_map(as, *addrp, len, segvn_create, zfod_argsp);
as_rangeunlock(as);
if (rv != 0)
return (rv);
/*
* Fill with data from read()
*/
rv = vn_rdwr(UIO_READ, vp, *addrp, len, off, UIO_USERSPACE,
0, (rlim64_t)0, cred, &resid);
if (rv == 0 && resid != 0)
rv = ENXIO;
if (rv != 0) {
as_rangelock(as);
(void) as_unmap(as, *addrp, len);
as_rangeunlock(as);
}
return (rv);
}
/*ARGSUSED*/
static int
xpvtap_segmap(dev_t dev, off_t off, struct as *asp, caddr_t *addrp,
off_t len, unsigned int prot, unsigned int maxprot, unsigned int flags,
cred_t *cred_p)
{
struct segmf_crargs a;
xpvtap_state_t *state;
int instance;
int e;
if (secpolicy_xvm_control(cred_p)) {
return (EPERM);
}
instance = getminor(dev);
state = ddi_get_soft_state(xpvtap_statep, instance);
if (state == NULL) {
return (EBADF);
}
/* the user app should be doing a MAP_SHARED mapping */
if ((flags & MAP_TYPE) != MAP_SHARED) {
return (EINVAL);
}
/*
* if this is the user ring (offset = 0), devmap it (which ends up in
* xpvtap_devmap). devmap will alloc and map the ring into the
* app's VA space.
*/
if (off == 0) {
e = devmap_setup(dev, (offset_t)off, asp, addrp, (size_t)len,
prot, maxprot, flags, cred_p);
return (e);
}
/* this should be the mmap for the gref pages (offset = PAGESIZE) */
if (off != PAGESIZE) {
return (EINVAL);
}
/* make sure we get the size we're expecting */
if (len != XPVTAP_GREF_BUFSIZE) {
return (EINVAL);
}
/*
* reserve user app VA space for the gref pages and use segmf to
* manage the backing store for the physical memory. segmf will
* map in/out the grefs and fault them in/out.
*/
ASSERT(asp == state->bt_map.um_as);
as_rangelock(asp);
if ((flags & MAP_FIXED) == 0) {
map_addr(addrp, len, 0, 0, flags);
if (*addrp == NULL) {
as_rangeunlock(asp);
return (ENOMEM);
}
} else {
/* User specified address */
(void) as_unmap(asp, *addrp, len);
}
a.dev = dev;
a.prot = (uchar_t)prot;
a.maxprot = (uchar_t)maxprot;
e = as_map(asp, *addrp, len, segmf_create, &a);
if (e != 0) {
as_rangeunlock(asp);
return (e);
}
as_rangeunlock(asp);
/*
* Stash user base address, and compute address where the request
* array will end up.
*/
state->bt_map.um_guest_pages = (caddr_t)*addrp;
state->bt_map.um_guest_size = (size_t)len;
/* register an as callback so we can cleanup when the app goes away */
e = as_add_callback(asp, xpvtap_segmf_unregister, state,
AS_UNMAP_EVENT, *addrp, len, KM_SLEEP);
if (e != 0) {
(void) as_unmap(asp, *addrp, len);
return (EINVAL);
}
/* wake thread to see if there are requests already queued up */
mutex_enter(&state->bt_thread.ut_mutex);
state->bt_thread.ut_wake = B_TRUE;
cv_signal(&state->bt_thread.ut_wake_cv);
mutex_exit(&state->bt_thread.ut_mutex);
return (0);
}
/*ARGSUSED*/
int
gfxp_ddi_segmap_setup(dev_t dev, off_t offset, struct as *as, caddr_t *addrp,
off_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cred,
ddi_device_acc_attr_t *accattrp, uint_t rnumber)
{
struct segdev_crargs dev_a;
int (*mapfunc)(dev_t dev, off_t off, int prot);
uint_t hat_attr;
pfn_t pfn;
int error, i;
if ((mapfunc = devopsp[getmajor(dev)]->devo_cb_ops->cb_mmap) == nodev)
return (ENODEV);
/*
* Character devices that support the d_mmap
* interface can only be mmap'ed shared.
*/
if ((flags & MAP_TYPE) != MAP_SHARED)
return (EINVAL);
/*
* Check that this region is indeed mappable on this platform.
* Use the mapping function.
*/
if (ddi_device_mapping_check(dev, accattrp, rnumber, &hat_attr) == -1)
return (ENXIO);
if (accattrp != NULL) {
switch (accattrp->devacc_attr_dataorder) {
case DDI_STRICTORDER_ACC:
/* Want UC */
hat_attr &= ~HAT_ORDER_MASK;
hat_attr |= (HAT_STRICTORDER | HAT_PLAT_NOCACHE);
break;
case DDI_MERGING_OK_ACC:
/* Want WC */
hat_attr &= ~HAT_ORDER_MASK;
hat_attr |= (HAT_MERGING_OK | HAT_PLAT_NOCACHE);
break;
}
}
/*
* Check to ensure that the entire range is
* legal and we are not trying to map in
* more than the device will let us.
*/
for (i = 0; i < len; i += PAGESIZE) {
if (i == 0) {
/*
* Save the pfn at offset here. This pfn will be
* used later to get user address.
*/
if ((pfn = (pfn_t)cdev_mmap(mapfunc, dev, offset,
maxprot)) == PFN_INVALID)
return (ENXIO);
} else {
if (cdev_mmap(mapfunc, dev, offset + i, maxprot) ==
PFN_INVALID)
return (ENXIO);
}
}
as_rangelock(as);
if ((flags & MAP_FIXED) == 0) {
/*
* Pick an address w/o worrying about
* any vac alignment constraints.
*/
map_addr(addrp, len, ptob(pfn), 0, flags);
if (*addrp == NULL) {
as_rangeunlock(as);
return (ENOMEM);
}
} else {
/*
* User-specified address; blow away any previous mappings.
*/
(void) as_unmap(as, *addrp, len);
}
dev_a.mapfunc = mapfunc;
dev_a.dev = dev;
dev_a.offset = (offset_t)offset;
dev_a.type = flags & MAP_TYPE;
dev_a.prot = (uchar_t)prot;
dev_a.maxprot = (uchar_t)maxprot;
dev_a.hat_attr = hat_attr;
#if DEBUG
dev_a.hat_flags = 0;
#else
dev_a.hat_flags = HAT_LOAD_LOCK;
#endif
dev_a.devmap_data = NULL;
error = as_map(as, *addrp, len, segdev_create, &dev_a);
as_rangeunlock(as);
return (error);
}
/*
* This routine assumes that the stack grows downward.
* Returns 0 on success, errno on failure.
*/
int
grow_internal(caddr_t sp, uint_t growszc)
{
struct proc *p = curproc;
size_t newsize;
size_t oldsize;
int error;
size_t pgsz;
uint_t szc;
struct segvn_crargs crargs = SEGVN_ZFOD_ARGS(PROT_ZFOD, PROT_ALL);
ASSERT(sp < p->p_usrstack);
sp = (caddr_t)P2ALIGN((uintptr_t)sp, PAGESIZE);
/*
* grow to growszc alignment but use current p->p_stkpageszc for
* the segvn_crargs szc passed to segvn_create. For memcntl to
* increase the szc, this allows the new extension segment to be
* concatenated successfully with the existing stack segment.
*/
if ((szc = growszc) != 0) {
pgsz = page_get_pagesize(szc);
ASSERT(pgsz > PAGESIZE);
newsize = p->p_usrstack - (caddr_t)P2ALIGN((uintptr_t)sp, pgsz);
if (newsize > (size_t)p->p_stk_ctl) {
szc = 0;
pgsz = PAGESIZE;
newsize = p->p_usrstack - sp;
}
} else {
pgsz = PAGESIZE;
newsize = p->p_usrstack - sp;
}
if (newsize > (size_t)p->p_stk_ctl) {
(void) rctl_action(rctlproc_legacy[RLIMIT_STACK], p->p_rctls, p,
RCA_UNSAFE_ALL);
return (ENOMEM);
}
oldsize = p->p_stksize;
ASSERT(P2PHASE(oldsize, PAGESIZE) == 0);
if (newsize <= oldsize) { /* prevent the stack from shrinking */
return (0);
}
if (!(p->p_stkprot & PROT_EXEC)) {
crargs.prot &= ~PROT_EXEC;
}
/*
* extend stack with the proposed new growszc, which is different
* than p_stkpageszc only on a memcntl to increase the stack pagesize.
* AS_MAP_NO_LPOOB means use 0, and don't reapply OOB policies via
* map_pgszcvec(). Use AS_MAP_STACK to get intermediate page sizes
* if not aligned to szc's pgsz.
*/
if (szc > 0) {
caddr_t oldsp = p->p_usrstack - oldsize;
caddr_t austk = (caddr_t)P2ALIGN((uintptr_t)p->p_usrstack,
pgsz);
if (IS_P2ALIGNED(p->p_usrstack, pgsz) || oldsp < austk) {
crargs.szc = p->p_stkpageszc ? p->p_stkpageszc :
AS_MAP_NO_LPOOB;
} else if (oldsp == austk) {
crargs.szc = szc;
} else {
crargs.szc = AS_MAP_STACK;
}
} else {
crargs.szc = AS_MAP_NO_LPOOB;
}
crargs.lgrp_mem_policy_flags = LGRP_MP_FLAG_EXTEND_DOWN;
if ((error = as_map(p->p_as, p->p_usrstack - newsize, newsize - oldsize,
segvn_create, &crargs)) != 0) {
if (error == EAGAIN) {
cmn_err(CE_WARN, "Sorry, no swap space to grow stack "
"for pid %d (%s)", p->p_pid, PTOU(p)->u_comm);
}
return (error);
}
p->p_stksize = newsize;
return (0);
}
/*
* Returns 0 on success.
*/
int
brk_internal(caddr_t nva, uint_t brkszc)
{
caddr_t ova; /* current break address */
size_t size;
int error;
struct proc *p = curproc;
struct as *as = p->p_as;
size_t pgsz;
uint_t szc;
rctl_qty_t as_rctl;
/*
* extend heap to brkszc alignment but use current p->p_brkpageszc
* for the newly created segment. This allows the new extension
* segment to be concatenated successfully with the existing brk
* segment.
*/
if ((szc = brkszc) != 0) {
pgsz = page_get_pagesize(szc);
ASSERT(pgsz > PAGESIZE);
} else {
pgsz = PAGESIZE;
}
mutex_enter(&p->p_lock);
as_rctl = rctl_enforced_value(rctlproc_legacy[RLIMIT_DATA],
p->p_rctls, p);
mutex_exit(&p->p_lock);
/*
* If p_brkbase has not yet been set, the first call
* to brk() will initialize it.
*/
if (p->p_brkbase == 0)
p->p_brkbase = nva;
/*
* Before multiple page size support existed p_brksize was the value
* not rounded to the pagesize (i.e. it stored the exact user request
* for heap size). If pgsz is greater than PAGESIZE calculate the
* heap size as the real new heap size by rounding it up to pgsz.
* This is useful since we may want to know where the heap ends
* without knowing heap pagesize (e.g. some old code) and also if
* heap pagesize changes we can update p_brkpageszc but delay adding
* new mapping yet still know from p_brksize where the heap really
* ends. The user requested heap end is stored in libc variable.
*/
if (pgsz > PAGESIZE) {
caddr_t tnva = (caddr_t)P2ROUNDUP((uintptr_t)nva, pgsz);
size = tnva - p->p_brkbase;
if (tnva < p->p_brkbase || (size > p->p_brksize &&
size > (size_t)as_rctl)) {
szc = 0;
pgsz = PAGESIZE;
size = nva - p->p_brkbase;
}
} else {
size = nva - p->p_brkbase;
}
/*
* use PAGESIZE to roundup ova because we want to know the real value
* of the current heap end in case p_brkpageszc changes since the last
* p_brksize was computed.
*/
nva = (caddr_t)P2ROUNDUP((uintptr_t)nva, pgsz);
ova = (caddr_t)P2ROUNDUP((uintptr_t)(p->p_brkbase + p->p_brksize),
PAGESIZE);
if ((nva < p->p_brkbase) || (size > p->p_brksize &&
size > as_rctl)) {
mutex_enter(&p->p_lock);
(void) rctl_action(rctlproc_legacy[RLIMIT_DATA], p->p_rctls, p,
RCA_SAFE);
mutex_exit(&p->p_lock);
return (ENOMEM);
}
if (nva > ova) {
struct segvn_crargs crargs =
SEGVN_ZFOD_ARGS(PROT_ZFOD, PROT_ALL);
if (!(p->p_datprot & PROT_EXEC)) {
crargs.prot &= ~PROT_EXEC;
}
/*
* Add new zfod mapping to extend UNIX data segment
* AS_MAP_NO_LPOOB means use 0, and don't reapply OOB policies
* via map_pgszcvec(). Use AS_MAP_HEAP to get intermediate
* page sizes if ova is not aligned to szc's pgsz.
*/
if (szc > 0) {
caddr_t rbss;
rbss = (caddr_t)P2ROUNDUP((uintptr_t)p->p_bssbase,
pgsz);
if (IS_P2ALIGNED(p->p_bssbase, pgsz) || ova > rbss) {
crargs.szc = p->p_brkpageszc ? p->p_brkpageszc :
AS_MAP_NO_LPOOB;
} else if (ova == rbss) {
crargs.szc = szc;
} else {
crargs.szc = AS_MAP_HEAP;
}
} else {
crargs.szc = AS_MAP_NO_LPOOB;
}
crargs.lgrp_mem_policy_flags = LGRP_MP_FLAG_EXTEND_UP;
error = as_map(as, ova, (size_t)(nva - ova), segvn_create,
&crargs);
if (error) {
return (error);
}
} else if (nva < ova) {
/*
* Release mapping to shrink UNIX data segment.
*/
(void) as_unmap(as, nva, (size_t)(ova - nva));
}
p->p_brksize = size;
return (0);
}