/*
* Take rctl action when the requested file descriptor is too big.
*/
static void
fd_too_big(proc_t *p)
{
mutex_enter(&p->p_lock);
(void) rctl_action(rctlproc_legacy[RLIMIT_NOFILE],
p->p_rctls, p, RCA_SAFE);
mutex_exit(&p->p_lock);
}
/*
* wrxmem does the real work of write requests for xmemfs.
*/
static int
wrxmem(struct xmount *xm, struct xmemnode *xp, struct uio *uio,
struct cred *cr, struct caller_context *ct)
{
uint_t blockoffset; /* offset in the block */
uint_t blkwr; /* offset in blocks into xmem file */
uint_t blkcnt;
caddr_t base;
ssize_t bytes; /* bytes to uiomove */
struct vnode *vp;
int error = 0;
size_t bsize = xm->xm_bsize;
rlim64_t limit = uio->uio_llimit;
long oresid = uio->uio_resid;
timestruc_t now;
offset_t offset;
/*
* xp->xn_size is incremented before the uiomove
* is done on a write. If the move fails (bad user
* address) reset xp->xn_size.
* The better way would be to increment xp->xn_size
* only if the uiomove succeeds.
*/
long xn_size_changed = 0;
offset_t old_xn_size;
vp = XNTOV(xp);
ASSERT(vp->v_type == VREG);
XMEMPRINTF(1, ("wrxmem: vp %p resid %lx off %llx\n",
(void *)vp, uio->uio_resid, uio->uio_loffset));
ASSERT(RW_WRITE_HELD(&xp->xn_contents));
ASSERT(RW_WRITE_HELD(&xp->xn_rwlock));
if (MANDLOCK(vp, xp->xn_mode)) {
rw_exit(&xp->xn_contents);
/*
* xmem_getattr ends up being called by chklock
*/
error = chklock(vp, FWRITE,
uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct);
rw_enter(&xp->xn_contents, RW_WRITER);
if (error != 0) {
XMEMPRINTF(8, ("wrxmem: vp %p error %x\n",
(void *)vp, error));
return (error);
}
}
if ((offset = uio->uio_loffset) < 0)
return (EINVAL);
if (offset >= limit) {
proc_t *p = ttoproc(curthread);
mutex_enter(&p->p_lock);
(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
p, RCA_UNSAFE_SIGINFO);
mutex_exit(&p->p_lock);
return (EFBIG);
}
if (uio->uio_resid == 0) {
XMEMPRINTF(8, ("wrxmem: vp %p resid %lx\n",
(void *)vp, uio->uio_resid));
return (0);
}
/*
* Get the highest blocknumber and allocate page array if needed.
* Note that if xm_bsize != PAGESIZE, each ppa[] is pointer to
* a page array rather than just a page.
*/
blkcnt = howmany((offset + uio->uio_resid), bsize);
blkwr = offset >> xm->xm_bshift; /* write begins here */
XMEMPRINTF(1, ("wrxmem: vp %p blkcnt %x blkwr %x xn_ppasz %lx\n",
(void *)vp, blkcnt, blkwr, xp->xn_ppasz));
/* file size increase */
if (xp->xn_ppasz < blkcnt) {
page_t ***ppa;
int ppasz;
uint_t blksinfile = howmany(xp->xn_size, bsize);
/*
* check if sufficient blocks available for the given offset.
*/
if (blkcnt - blksinfile > xm->xm_max - xm->xm_mem)
return (ENOSPC);
/*
* to prevent reallocating every time the file grows by a
* single block, double the size of the array.
*/
if (blkcnt < xp->xn_ppasz * 2)
ppasz = xp->xn_ppasz * 2;
else
ppasz = blkcnt;
ppa = kmem_zalloc(ppasz * sizeof (page_t **), KM_SLEEP);
ASSERT(ppa);
if (xp->xn_ppasz) {
bcopy(xp->xn_ppa, ppa, blksinfile * sizeof (*ppa));
kmem_free(xp->xn_ppa, xp->xn_ppasz * sizeof (*ppa));
}
xp->xn_ppa = ppa;
xp->xn_ppasz = ppasz;
/*
* fill in the 'hole' if write offset beyond file size. This
* helps in creating large files quickly; an application can
* lseek to a large offset and perform a single write
* operation to create the large file.
*/
if (blksinfile < blkwr) {
old_xn_size = xp->xn_size;
xp->xn_size = (offset_t)blkwr * bsize;
XMEMPRINTF(4, ("wrxmem: fill vp %p blks %x to %x\n",
(void *)vp, blksinfile, blkcnt - 1));
error = xmem_fillpages(xp, vp,
(offset_t)blksinfile * bsize,
(offset_t)(blkcnt - blksinfile) * bsize, 1);
if (error) {
/* truncate file back to original size */
(void) xmemnode_trunc(xm, xp, old_xn_size);
return (error);
}
/*
* if error on blkwr, this allows truncation of the
* filled hole.
*/
xp->xn_size = old_xn_size;
}
}
do {
offset_t pagestart, pageend;
page_t **ppp;
blockoffset = (uint_t)offset & (bsize - 1);
/*
* A maximum of xm->xm_bsize bytes of data is transferred
* each pass through this loop
*/
bytes = MIN(bsize - blockoffset, uio->uio_resid);
ASSERT(bytes);
if (offset + bytes >= limit) {
if (offset >= limit) {
error = EFBIG;
goto out;
}
bytes = limit - offset;
}
if (!xp->xn_ppa[blkwr]) {
/* zero fill new pages - simplify partial updates */
error = xmem_fillpages(xp, vp, offset, bytes, 1);
if (error)
return (error);
}
/* grow the file to the new length */
if (offset + bytes > xp->xn_size) {
xn_size_changed = 1;
old_xn_size = xp->xn_size;
xp->xn_size = offset + bytes;
}
#ifdef LOCKNEST
xmem_getpage();
#endif
/* xn_ppa[] is a page_t * if ppb == 1 */
if (xm->xm_ppb == 1)
ppp = (page_t **)&xp->xn_ppa[blkwr];
else
ppp = &xp->xn_ppa[blkwr][btop(blockoffset)];
pagestart = offset & ~(offset_t)(PAGESIZE - 1);
/*
* subtract 1 in case (offset + bytes) is mod PAGESIZE
* so that pageend is the actual index of last page.
*/
pageend = (offset + bytes - 1) & ~(offset_t)(PAGESIZE - 1);
base = segxmem_getmap(xm->xm_map, vp,
pagestart, pageend - pagestart + PAGESIZE,
ppp, S_WRITE);
rw_exit(&xp->xn_contents);
error = uiomove(base + (offset - pagestart), bytes,
UIO_WRITE, uio);
segxmem_release(xm->xm_map, base,
pageend - pagestart + PAGESIZE);
/*
* Re-acquire contents lock.
*/
rw_enter(&xp->xn_contents, RW_WRITER);
/*
* If the uiomove failed, fix up xn_size.
*/
if (error) {
if (xn_size_changed) {
/*
* The uiomove failed, and we
* allocated blocks,so get rid
* of them.
*/
(void) xmemnode_trunc(xm, xp, old_xn_size);
}
} else {
if ((xp->xn_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) &&
(xp->xn_mode & (S_ISUID | S_ISGID)) &&
secpolicy_vnode_setid_retain(cr,
(xp->xn_mode & S_ISUID) != 0 && xp->xn_uid == 0)
!= 0) {
/*
* Clear Set-UID & Set-GID bits on
* successful write if not privileged
* and at least one of the execute bits
* is set. If we always clear Set-GID,
* mandatory file and record locking is
* unuseable.
*/
xp->xn_mode &= ~(S_ISUID | S_ISGID);
}
gethrestime(&now);
xp->xn_mtime = now;
xp->xn_ctime = now;
}
offset = uio->uio_loffset; /* uiomove sets uio_loffset */
blkwr++;
} while (error == 0 && uio->uio_resid > 0 && bytes != 0);
out:
/*
* If we've already done a partial-write, terminate
* the write but return no error.
*/
if (oresid != uio->uio_resid)
error = 0;
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);
}
