/* --------------------------------------------------------------------- */
static int
tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
vm_offset_t offset, size_t tlen, struct uio *uio)
{
vm_page_t m;
int error, rv;
VM_OBJECT_LOCK(tobj);
m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(tobj, &m, 1, 0);
if (rv != VM_PAGER_OK) {
vm_page_lock(m);
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_UNLOCK(tobj);
return (EIO);
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(tobj);
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
/*
* Obtain a page pointer array and lock all pages into system memory. A segmentation violation will
* occur here if the calling user does not have access to the submitted address.
*/
static int
via_lock_all_dma_pages(drm_via_sg_info_t *vsg, drm_via_dmablit_t *xfer)
{
unsigned long first_pfn = VIA_PFN(xfer->mem_addr);
vm_page_t m;
int i;
vsg->num_pages = VIA_PFN(xfer->mem_addr +
(xfer->num_lines * xfer->mem_stride -1)) - first_pfn + 1;
if (NULL == (vsg->pages = malloc(sizeof(vm_page_t) * vsg->num_pages,
DRM_MEM_DRIVER, M_NOWAIT)))
return -ENOMEM;
vsg->state = dr_via_pages_alloc;
if (vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map,
(vm_offset_t)xfer->mem_addr, vsg->num_pages * PAGE_SIZE,
VM_PROT_READ | VM_PROT_WRITE, vsg->pages, vsg->num_pages) < 0)
return -EACCES;
for (i = 0; i < vsg->num_pages; i++) {
m = vsg->pages[i];
vm_page_lock(m);
vm_page_wire(m);
vm_page_unhold(m);
vm_page_unlock(m);
}
vsg->state = dr_via_pages_locked;
DRM_DEBUG("DMA pages locked\n");
return 0;
}
/*
* vm_fault_unwire:
*
* Unwire a range of virtual addresses in a map.
*/
void
vm_fault_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
boolean_t fictitious)
{
vm_paddr_t pa;
vm_offset_t va;
vm_page_t m;
pmap_t pmap;
pmap = vm_map_pmap(map);
/*
* Since the pages are wired down, we must be able to get their
* mappings from the physical map system.
*/
for (va = start; va < end; va += PAGE_SIZE) {
pa = pmap_extract(pmap, va);
if (pa != 0) {
pmap_change_wiring(pmap, va, FALSE);
if (!fictitious) {
m = PHYS_TO_VM_PAGE(pa);
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
}
}
}
}
/*
* Function that frees up all resources for a blit. It is usable even if the
* blit info has only been partially built as long as the status enum is consistent
* with the actual status of the used resources.
*/
static void
via_free_sg_info(drm_via_sg_info_t *vsg)
{
vm_page_t page;
int i;
switch(vsg->state) {
case dr_via_device_mapped:
via_unmap_blit_from_device(vsg);
case dr_via_desc_pages_alloc:
for (i=0; i<vsg->num_desc_pages; ++i) {
if (vsg->desc_pages[i] != NULL)
free(vsg->desc_pages[i], DRM_MEM_PAGES);
}
free(vsg->desc_pages, DRM_MEM_DRIVER);
case dr_via_pages_locked:
for (i=0; i < vsg->num_pages; ++i) {
page = vsg->pages[i];
vm_page_lock(page);
vm_page_unwire(page, 0);
vm_page_unlock(page);
}
case dr_via_pages_alloc:
free(vsg->pages, DRM_MEM_DRIVER);
default:
vsg->state = dr_via_sg_init;
}
free(vsg->bounce_buffer, DRM_MEM_DRIVER);
vsg->bounce_buffer = NULL;
vsg->free_on_sequence = 0;
}
/* --------------------------------------------------------------------- */
static int
tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
vm_offset_t offset, size_t tlen, struct uio *uio)
{
vm_page_t m;
int error;
VM_OBJECT_LOCK(tobj);
vm_object_pip_add(tobj, 1);
m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
error = vm_pager_get_pages(tobj, &m, 1, 0);
if (error != 0) {
printf("tmpfs get pages from pager error [read]\n");
goto out;
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(tobj);
out:
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
vm_object_pip_subtract(tobj, 1);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
/*
* Same as above, but forces the page to be detached from the object
* and go into free pool.
*/
void
sf_ext_free_nocache(void *arg1, void *arg2)
{
struct sf_buf *sf = arg1;
struct sendfile_sync *sfs = arg2;
vm_page_t pg = sf_buf_page(sf);
sf_buf_free(sf);
vm_page_lock(pg);
if (vm_page_unwire(pg, PQ_NONE)) {
vm_object_t obj;
/* Try to free the page, but only if it is cheap to. */
if ((obj = pg->object) == NULL)
vm_page_free(pg);
else if (!vm_page_xbusied(pg) && VM_OBJECT_TRYWLOCK(obj)) {
vm_page_free(pg);
VM_OBJECT_WUNLOCK(obj);
} else
vm_page_deactivate(pg);
}
vm_page_unlock(pg);
if (sfs != NULL) {
mtx_lock(&sfs->mtx);
KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
if (--sfs->count == 0)
cv_signal(&sfs->cv);
mtx_unlock(&sfs->mtx);
}
}
/*
* Detach mapped page and release resources back to the system. Called
* by mbuf(9) code when last reference to a page is freed.
*/
void
sf_ext_free(void *arg1, void *arg2)
{
struct sf_buf *sf = arg1;
struct sendfile_sync *sfs = arg2;
vm_page_t pg = sf_buf_page(sf);
sf_buf_free(sf);
vm_page_lock(pg);
/*
* Check for the object going away on us. This can
* happen since we don't hold a reference to it.
* If so, we're responsible for freeing the page.
*/
if (vm_page_unwire(pg, PQ_INACTIVE) && pg->object == NULL)
vm_page_free(pg);
vm_page_unlock(pg);
if (sfs != NULL) {
mtx_lock(&sfs->mtx);
KASSERT(sfs->count > 0, ("Sendfile sync botchup count == 0"));
if (--sfs->count == 0)
cv_signal(&sfs->cv);
mtx_unlock(&sfs->mtx);
}
}
/*
* Fill as many pages as vm_fault has allocated for us.
*/
static int
phys_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
{
int i;
VM_OBJECT_ASSERT_WLOCKED(object);
for (i = 0; i < count; i++) {
if (m[i]->valid == 0) {
if ((m[i]->flags & PG_ZERO) == 0)
pmap_zero_page(m[i]);
m[i]->valid = VM_PAGE_BITS_ALL;
}
KASSERT(m[i]->valid == VM_PAGE_BITS_ALL,
("phys_pager_getpages: partially valid page %p", m[i]));
KASSERT(m[i]->dirty == 0,
("phys_pager_getpages: dirty page %p", m[i]));
/* The requested page must remain busy, the others not. */
if (i == reqpage) {
vm_page_lock(m[i]);
vm_page_flash(m[i]);
vm_page_unlock(m[i]);
} else
vm_page_xunbusy(m[i]);
}
return (VM_PAGER_OK);
}
/*
* Return the requested word from the user-space address.
* Returns 0 if it isn't mapped. (For what we're using it
* for, if the actual value is 0, that's equivalent to that.)
*/
static caddr_t
GET_WORD(pmap_t map, caddr_t virtual_addr)
{
caddr_t retval = 0;
vm_page_t page;
int err;
caddr_t old_fault;
page = pmap_extract_and_hold(map, (vm_offset_t)virtual_addr, VM_PROT_READ);
if (page == 0) {
return 0;
}
// I do this because copyin/copyout aren't re-entrant.
old_fault = curpcb->pcb_onfault;
err = copyin(virtual_addr, &retval, sizeof(retval));
curpcb->pcb_onfault = old_fault;
if (err != 0) {
#if SAMPLE_DEBUG
printf("%s(%d): copyin(%p, %p, %zd) failed: %d\n", __FUNCTION__, __LINE__, (void*)virtual_addr, &retval, sizeof(retval), err);
#endif
retval = 0;
}
vm_page_lock(page);
vm_page_unhold(page);
vm_page_unlock(page);
return retval;
}
void
vm_gpa_release(void *cookie)
{
vm_page_t m = cookie;
vm_page_lock(m);
vm_page_unhold(m);
vm_page_unlock(m);
}
static inline void
release_page(struct faultstate *fs)
{
vm_page_wakeup(fs->m);
vm_page_lock(fs->m);
vm_page_deactivate(fs->m);
vm_page_unlock(fs->m);
fs->m = NULL;
}
/*
* Release a page we've previously wired.
*/
static void
zbuf_page_free(vm_page_t pp)
{
vm_page_lock(pp);
vm_page_unwire(pp, 0);
if (pp->wire_count == 0 && pp->object == NULL)
vm_page_free(pp);
vm_page_unlock(pp);
}
/*
* Identify the physical page mapped at the given kernel virtual
* address. Insert this physical page into the given address space at
* the given virtual address, replacing the physical page, if any,
* that already exists there.
*/
static int
vm_pgmoveco(vm_map_t mapa, vm_offset_t kaddr, vm_offset_t uaddr)
{
vm_map_t map = mapa;
vm_page_t kern_pg, user_pg;
vm_object_t uobject;
vm_map_entry_t entry;
vm_pindex_t upindex;
vm_prot_t prot;
boolean_t wired;
KASSERT((uaddr & PAGE_MASK) == 0,
("vm_pgmoveco: uaddr is not page aligned"));
/*
* Herein the physical page is validated and dirtied. It is
* unwired in sf_buf_mext().
*/
kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
kern_pg->valid = VM_PAGE_BITS_ALL;
KASSERT(kern_pg->queue == PQ_NONE && kern_pg->wire_count == 1,
("vm_pgmoveco: kern_pg is not correctly wired"));
if ((vm_map_lookup(&map, uaddr,
VM_PROT_WRITE, &entry, &uobject,
&upindex, &prot, &wired)) != KERN_SUCCESS) {
return(EFAULT);
}
VM_OBJECT_LOCK(uobject);
retry:
if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
if (vm_page_sleep_if_busy(user_pg, TRUE, "vm_pgmoveco"))
goto retry;
vm_page_lock(user_pg);
pmap_remove_all(user_pg);
vm_page_free(user_pg);
vm_page_unlock(user_pg);
} else {
/*
* Even if a physical page does not exist in the
* object chain's first object, a physical page from a
* backing object may be mapped read only.
*/
if (uobject->backing_object != NULL)
pmap_remove(map->pmap, uaddr, uaddr + PAGE_SIZE);
}
vm_page_insert(kern_pg, uobject, upindex);
vm_page_dirty(kern_pg);
VM_OBJECT_UNLOCK(uobject);
vm_map_lookup_done(map, entry);
return(KERN_SUCCESS);
}
static void
unwire_ddp_buffer(struct ddp_buffer *db)
{
int i;
vm_page_t p;
for (i = 0; i < db->npages; i++) {
p = db->pages[i];
vm_page_lock(p);
vm_page_unwire(p, PQ_INACTIVE);
vm_page_unlock(p);
}
}
void
exec_unmap_first_page(struct image_params *imgp)
{
vm_page_t m;
if (imgp->firstpage != NULL) {
m = sf_buf_page(imgp->firstpage);
sf_buf_free(imgp->firstpage);
imgp->firstpage = NULL;
vm_page_lock(m);
vm_page_unhold(m);
vm_page_unlock(m);
}
}
void
cdev_pager_free_page(vm_object_t object, vm_page_t m)
{
VM_OBJECT_ASSERT_WLOCKED(object);
if (object->type == OBJT_MGTDEVICE) {
KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
pmap_remove_all(m);
vm_page_lock(m);
vm_page_remove(m);
vm_page_unlock(m);
} else if (object->type == OBJT_DEVICE)
dev_pager_free_page(object, m);
}
static void
wire_ddp_buffer(struct ddp_buffer *db)
{
int i;
vm_page_t p;
for (i = 0; i < db->npages; i++) {
p = db->pages[i];
vm_page_lock(p);
vm_page_wire(p);
vm_page_unhold(p);
vm_page_unlock(p);
}
}
int ttm_tt_swapin(struct ttm_tt *ttm)
{
vm_object_t obj;
vm_page_t from_page, to_page;
int i, ret, rv;
obj = ttm->swap_storage;
VM_OBJECT_WLOCK(obj);
vm_object_pip_add(obj, 1);
for (i = 0; i < ttm->num_pages; ++i) {
from_page = vm_page_grab(obj, i, VM_ALLOC_NOBUSY |
VM_ALLOC_RETRY);
if (from_page->valid != VM_PAGE_BITS_ALL) {
vm_page_busy(from_page);
if (vm_pager_has_page(obj, i, NULL, NULL)) {
rv = vm_pager_get_pages(obj, &from_page, 1, 0);
if (rv != VM_PAGER_OK) {
vm_page_lock(from_page);
vm_page_free(from_page);
vm_page_unlock(from_page);
ret = -EIO;
goto err_ret;
}
} else
vm_page_zero_invalid(from_page, TRUE);
vm_page_wakeup(from_page);
}
to_page = ttm->pages[i];
if (unlikely(to_page == NULL)) {
ret = -ENOMEM;
goto err_ret;
}
pmap_copy_page(from_page, to_page);
}
vm_object_pip_wakeup(obj);
VM_OBJECT_WUNLOCK(obj);
if (!(ttm->page_flags & TTM_PAGE_FLAG_PERSISTENT_SWAP))
vm_object_deallocate(obj);
ttm->swap_storage = NULL;
ttm->page_flags &= ~TTM_PAGE_FLAG_SWAPPED;
return (0);
err_ret:
vm_object_pip_wakeup(obj);
VM_OBJECT_WUNLOCK(obj);
return (ret);
}
/*
* Speed up the reclamation of up to "distance" pages that precede the
* faulting pindex within the first object of the shadow chain.
*/
static void
vm_fault_cache_behind(const struct faultstate *fs, int distance)
{
vm_object_t first_object, object;
vm_page_t m, m_prev;
vm_pindex_t pindex;
object = fs->object;
VM_OBJECT_ASSERT_WLOCKED(object);
first_object = fs->first_object;
if (first_object != object) {
if (!VM_OBJECT_TRYWLOCK(first_object)) {
VM_OBJECT_WUNLOCK(object);
VM_OBJECT_WLOCK(first_object);
VM_OBJECT_WLOCK(object);
}
}
/* Neither fictitious nor unmanaged pages can be cached. */
if ((first_object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0) {
if (fs->first_pindex < distance)
pindex = 0;
else
pindex = fs->first_pindex - distance;
if (pindex < OFF_TO_IDX(fs->entry->offset))
pindex = OFF_TO_IDX(fs->entry->offset);
m = first_object != object ? fs->first_m : fs->m;
KASSERT((m->oflags & VPO_BUSY) != 0,
("vm_fault_cache_behind: page %p is not busy", m));
m_prev = vm_page_prev(m);
while ((m = m_prev) != NULL && m->pindex >= pindex &&
m->valid == VM_PAGE_BITS_ALL) {
m_prev = vm_page_prev(m);
if (m->busy != 0 || (m->oflags & VPO_BUSY) != 0)
continue;
vm_page_lock(m);
if (m->hold_count == 0 && m->wire_count == 0) {
pmap_remove_all(m);
vm_page_aflag_clear(m, PGA_REFERENCED);
if (m->dirty != 0)
vm_page_deactivate(m);
else
vm_page_cache(m);
}
vm_page_unlock(m);
}
}
if (first_object != object)
VM_OBJECT_WUNLOCK(first_object);
}
/*
* Speed up the reclamation of up to "distance" pages that precede the
* faulting pindex within the first object of the shadow chain.
*/
static void
vm_fault_cache_behind(const struct faultstate *fs, int distance)
{
vm_object_t first_object, object;
vm_page_t m, m_prev;
vm_pindex_t pindex;
object = fs->object;
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
first_object = fs->first_object;
if (first_object != object) {
if (!VM_OBJECT_TRYLOCK(first_object)) {
VM_OBJECT_UNLOCK(object);
VM_OBJECT_LOCK(first_object);
VM_OBJECT_LOCK(object);
}
}
if (first_object->type != OBJT_DEVICE &&
first_object->type != OBJT_PHYS && first_object->type != OBJT_SG) {
if (fs->first_pindex < distance)
pindex = 0;
else
pindex = fs->first_pindex - distance;
if (pindex < OFF_TO_IDX(fs->entry->offset))
pindex = OFF_TO_IDX(fs->entry->offset);
m = first_object != object ? fs->first_m : fs->m;
KASSERT((m->oflags & VPO_BUSY) != 0,
("vm_fault_cache_behind: page %p is not busy", m));
m_prev = vm_page_prev(m);
while ((m = m_prev) != NULL && m->pindex >= pindex &&
m->valid == VM_PAGE_BITS_ALL) {
m_prev = vm_page_prev(m);
if (m->busy != 0 || (m->oflags & VPO_BUSY) != 0)
continue;
vm_page_lock(m);
if (m->hold_count == 0 && m->wire_count == 0) {
pmap_remove_all(m);
vm_page_aflag_clear(m, PGA_REFERENCED);
if (m->dirty != 0)
vm_page_deactivate(m);
else
vm_page_cache(m);
}
vm_page_unlock(m);
}
}
if (first_object != object)
VM_OBJECT_UNLOCK(first_object);
}
/*
* Given a user pointer to a page of user memory, return an sf_buf for the
* page. Because we may be requesting quite a few sf_bufs, prefer failure to
* deadlock and use SFB_NOWAIT.
*/
static struct sf_buf *
zbuf_sfbuf_get(struct vm_map *map, vm_offset_t uaddr)
{
struct sf_buf *sf;
vm_page_t pp;
if (vm_fault_quick_hold_pages(map, uaddr, PAGE_SIZE, VM_PROT_READ |
VM_PROT_WRITE, &pp, 1) < 0)
return (NULL);
vm_page_lock(pp);
vm_page_wire(pp);
vm_page_unhold(pp);
vm_page_unlock(pp);
sf = sf_buf_alloc(pp, SFB_NOWAIT);
if (sf == NULL) {
zbuf_page_free(pp);
return (NULL);
}
return (sf);
}
static void
unlock_and_deallocate(struct faultstate *fs)
{
vm_object_pip_wakeup(fs->object);
VM_OBJECT_WUNLOCK(fs->object);
if (fs->object != fs->first_object) {
VM_OBJECT_WLOCK(fs->first_object);
vm_page_lock(fs->first_m);
vm_page_free(fs->first_m);
vm_page_unlock(fs->first_m);
vm_object_pip_wakeup(fs->first_object);
VM_OBJECT_WUNLOCK(fs->first_object);
fs->first_m = NULL;
}
vm_object_deallocate(fs->first_object);
unlock_map(fs);
if (fs->vp != NULL) {
vput(fs->vp);
fs->vp = NULL;
}
}
static int
privcmd_pg_fault(vm_object_t object, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
struct privcmd_map *map = object->handle;
vm_pindex_t pidx;
vm_page_t page, oldm;
if (map->mapped != true)
return (VM_PAGER_FAIL);
pidx = OFF_TO_IDX(offset);
if (pidx >= map->size || BIT_ISSET(map->size, pidx, map->err))
return (VM_PAGER_FAIL);
page = PHYS_TO_VM_PAGE(map->phys_base_addr + offset);
if (page == NULL)
return (VM_PAGER_FAIL);
KASSERT((page->flags & PG_FICTITIOUS) != 0,
("not fictitious %p", page));
KASSERT(page->wire_count == 1, ("wire_count not 1 %p", page));
KASSERT(vm_page_busied(page) == 0, ("page %p is busy", page));
if (*mres != NULL) {
oldm = *mres;
vm_page_lock(oldm);
vm_page_free(oldm);
vm_page_unlock(oldm);
*mres = NULL;
}
vm_page_insert(page, object, pidx);
page->valid = VM_PAGE_BITS_ALL;
vm_page_xbusy(page);
*mres = page;
return (VM_PAGER_OK);
}
static void
socow_iodone(void *addr, void *args)
{
struct sf_buf *sf;
vm_page_t pp;
sf = args;
pp = sf_buf_page(sf);
sf_buf_free(sf);
/* remove COW mapping */
vm_page_lock(pp);
vm_page_cowclear(pp);
vm_page_unwire(pp, 0);
/*
* Check for the object going away on us. This can
* happen since we don't hold a reference to it.
* If so, we're responsible for freeing the page.
*/
if (pp->wire_count == 0 && pp->object == NULL)
vm_page_free(pp);
vm_page_unlock(pp);
socow_stats.iodone++;
}
/* See: old_dev_pager_fault() in device_pager.c as an example. */
static int
cheri_compositor_cfb_pg_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
vm_page_t *mres)
{
vm_pindex_t pidx;
vm_paddr_t paddr;
vm_page_t page;
struct cfb_vm_object *cfb_vm_obj;
struct cdev *dev;
struct cheri_compositor_softc *sc;
struct cdevsw *csw;
vm_memattr_t memattr;
int ref;
int retval;
pidx = OFF_TO_IDX(offset);
VM_OBJECT_WUNLOCK(vm_obj);
cfb_vm_obj = vm_obj->handle;
dev = cfb_vm_obj->dev;
sc = dev->si_drv1;
retval = VM_PAGER_OK;
CHERI_COMPOSITOR_DEBUG(sc, "vm_obj: %p, offset: %lu, prot: %i", vm_obj,
offset, prot);
csw = dev_refthread(dev, &ref);
if (csw == NULL) {
retval = VM_PAGER_FAIL;
goto done_unlocked;
}
/* Traditional d_mmap() call. */
CHERI_COMPOSITOR_DEBUG(sc, "offset: %lu, nprot: %i", offset, prot);
if (validate_prot_and_offset(sc, cfb_vm_obj->pool->mapped_fd,
prot, offset) != 0) {
retval = VM_PAGER_FAIL;
goto done_unlocked;
}
paddr = calculate_physical_address(sc, cfb_vm_obj->pool, offset);
memattr = VM_MEMATTR_UNCACHEABLE;
CHERI_COMPOSITOR_DEBUG(sc, "paddr: %p, memattr: %i",
(void *) paddr, memattr);
dev_relthread(dev, ref);
/* Sanity checks. */
KASSERT((((*mres)->flags & PG_FICTITIOUS) == 0),
("Expected non-fictitious page."));
/*
* Replace the passed in reqpage page with our own fake page and
* free up the all of the original pages.
*/
page = vm_page_getfake(paddr, memattr);
VM_OBJECT_WLOCK(vm_obj);
vm_page_lock(*mres);
vm_page_free(*mres);
vm_page_unlock(*mres);
*mres = page;
vm_page_insert(page, vm_obj, pidx);
page->valid = VM_PAGE_BITS_ALL;
/* Success! */
retval = VM_PAGER_OK;
goto done;
done_unlocked:
VM_OBJECT_WLOCK(vm_obj);
done:
CHERI_COMPOSITOR_DEBUG(sc, "Finished with mres: %p (retval: %i)", *mres,
retval);
return (retval);
}
/*
* Routine:
* vm_fault_copy_entry
* Function:
* Create new shadow object backing dst_entry with private copy of
* all underlying pages. When src_entry is equal to dst_entry,
* function implements COW for wired-down map entry. Otherwise,
* it forks wired entry into dst_map.
*
* In/out conditions:
* The source and destination maps must be locked for write.
* The source map entry must be wired down (or be a sharing map
* entry corresponding to a main map entry that is wired down).
*/
void
vm_fault_copy_entry(vm_map_t dst_map, vm_map_t src_map,
vm_map_entry_t dst_entry, vm_map_entry_t src_entry,
vm_ooffset_t *fork_charge)
{
vm_object_t backing_object, dst_object, object, src_object;
vm_pindex_t dst_pindex, pindex, src_pindex;
vm_prot_t access, prot;
vm_offset_t vaddr;
vm_page_t dst_m;
vm_page_t src_m;
boolean_t src_readonly, upgrade;
#ifdef lint
src_map++;
#endif /* lint */
upgrade = src_entry == dst_entry;
src_object = src_entry->object.vm_object;
src_pindex = OFF_TO_IDX(src_entry->offset);
src_readonly = (src_entry->protection & VM_PROT_WRITE) == 0;
/*
* Create the top-level object for the destination entry. (Doesn't
* actually shadow anything - we copy the pages directly.)
*/
dst_object = vm_object_allocate(OBJT_DEFAULT,
OFF_TO_IDX(dst_entry->end - dst_entry->start));
#if VM_NRESERVLEVEL > 0
dst_object->flags |= OBJ_COLORED;
dst_object->pg_color = atop(dst_entry->start);
#endif
VM_OBJECT_WLOCK(dst_object);
KASSERT(upgrade || dst_entry->object.vm_object == NULL,
("vm_fault_copy_entry: vm_object not NULL"));
dst_entry->object.vm_object = dst_object;
dst_entry->offset = 0;
dst_object->charge = dst_entry->end - dst_entry->start;
if (fork_charge != NULL) {
KASSERT(dst_entry->cred == NULL,
("vm_fault_copy_entry: leaked swp charge"));
dst_object->cred = curthread->td_ucred;
crhold(dst_object->cred);
*fork_charge += dst_object->charge;
} else {
dst_object->cred = dst_entry->cred;
dst_entry->cred = NULL;
}
access = prot = dst_entry->protection;
/*
* If not an upgrade, then enter the mappings in the pmap as
* read and/or execute accesses. Otherwise, enter them as
* write accesses.
*
* A writeable large page mapping is only created if all of
* the constituent small page mappings are modified. Marking
* PTEs as modified on inception allows promotion to happen
* without taking potentially large number of soft faults.
*/
if (!upgrade)
access &= ~VM_PROT_WRITE;
/*
* Loop through all of the virtual pages within the entry's
* range, copying each page from the source object to the
* destination object. Since the source is wired, those pages
* must exist. In contrast, the destination is pageable.
* Since the destination object does share any backing storage
* with the source object, all of its pages must be dirtied,
* regardless of whether they can be written.
*/
for (vaddr = dst_entry->start, dst_pindex = 0;
vaddr < dst_entry->end;
vaddr += PAGE_SIZE, dst_pindex++) {
/*
* Allocate a page in the destination object.
*/
do {
dst_m = vm_page_alloc(dst_object, dst_pindex,
VM_ALLOC_NORMAL);
if (dst_m == NULL) {
VM_OBJECT_WUNLOCK(dst_object);
VM_WAIT;
VM_OBJECT_WLOCK(dst_object);
}
} while (dst_m == NULL);
/*
* Find the page in the source object, and copy it in.
* (Because the source is wired down, the page will be in
* memory.)
*/
VM_OBJECT_WLOCK(src_object);
object = src_object;
pindex = src_pindex + dst_pindex;
while ((src_m = vm_page_lookup(object, pindex)) == NULL &&
src_readonly &&
(backing_object = object->backing_object) != NULL) {
/*
* Allow fallback to backing objects if we are reading.
*/
VM_OBJECT_WLOCK(backing_object);
pindex += OFF_TO_IDX(object->backing_object_offset);
VM_OBJECT_WUNLOCK(object);
object = backing_object;
}
if (src_m == NULL)
panic("vm_fault_copy_wired: page missing");
pmap_copy_page(src_m, dst_m);
VM_OBJECT_WUNLOCK(object);
dst_m->valid = VM_PAGE_BITS_ALL;
dst_m->dirty = VM_PAGE_BITS_ALL;
VM_OBJECT_WUNLOCK(dst_object);
/*
* Enter it in the pmap. If a wired, copy-on-write
* mapping is being replaced by a write-enabled
* mapping, then wire that new mapping.
*/
pmap_enter(dst_map->pmap, vaddr, access, dst_m, prot, upgrade);
/*
* Mark it no longer busy, and put it on the active list.
*/
VM_OBJECT_WLOCK(dst_object);
if (upgrade) {
vm_page_lock(src_m);
vm_page_unwire(src_m, 0);
vm_page_unlock(src_m);
vm_page_lock(dst_m);
vm_page_wire(dst_m);
vm_page_unlock(dst_m);
} else {
vm_page_lock(dst_m);
vm_page_activate(dst_m);
vm_page_unlock(dst_m);
}
vm_page_wakeup(dst_m);
}
VM_OBJECT_WUNLOCK(dst_object);
if (upgrade) {
dst_entry->eflags &= ~(MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY);
vm_object_deallocate(src_object);
}
}
int
vm_fault_hold(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
int fault_flags, vm_page_t *m_hold)
{
vm_prot_t prot;
long ahead, behind;
int alloc_req, era, faultcount, nera, reqpage, result;
boolean_t growstack, is_first_object_locked, wired;
int map_generation;
vm_object_t next_object;
vm_page_t marray[VM_FAULT_READ_MAX];
int hardfault;
struct faultstate fs;
struct vnode *vp;
int locked, error;
hardfault = 0;
growstack = TRUE;
PCPU_INC(cnt.v_vm_faults);
fs.vp = NULL;
faultcount = reqpage = 0;
RetryFault:;
/*
* Find the backing store object and offset into it to begin the
* search.
*/
fs.map = map;
result = vm_map_lookup(&fs.map, vaddr, fault_type, &fs.entry,
&fs.first_object, &fs.first_pindex, &prot, &wired);
if (result != KERN_SUCCESS) {
if (growstack && result == KERN_INVALID_ADDRESS &&
map != kernel_map) {
result = vm_map_growstack(curproc, vaddr);
if (result != KERN_SUCCESS)
return (KERN_FAILURE);
growstack = FALSE;
goto RetryFault;
}
return (result);
}
map_generation = fs.map->timestamp;
if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
panic("vm_fault: fault on nofault entry, addr: %lx",
(u_long)vaddr);
}
/*
* Make a reference to this object to prevent its disposal while we
* are messing with it. Once we have the reference, the map is free
* to be diddled. Since objects reference their shadows (and copies),
* they will stay around as well.
*
* Bump the paging-in-progress count to prevent size changes (e.g.
* truncation operations) during I/O. This must be done after
* obtaining the vnode lock in order to avoid possible deadlocks.
*/
VM_OBJECT_WLOCK(fs.first_object);
vm_object_reference_locked(fs.first_object);
vm_object_pip_add(fs.first_object, 1);
fs.lookup_still_valid = TRUE;
if (wired)
fault_type = prot | (fault_type & VM_PROT_COPY);
fs.first_m = NULL;
/*
* Search for the page at object/offset.
*/
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
while (TRUE) {
/*
* If the object is dead, we stop here
*/
if (fs.object->flags & OBJ_DEAD) {
unlock_and_deallocate(&fs);
return (KERN_PROTECTION_FAILURE);
}
/*
* See if page is resident
*/
fs.m = vm_page_lookup(fs.object, fs.pindex);
if (fs.m != NULL) {
/*
* check for page-based copy on write.
* We check fs.object == fs.first_object so
* as to ensure the legacy COW mechanism is
* used when the page in question is part of
* a shadow object. Otherwise, vm_page_cowfault()
* removes the page from the backing object,
* which is not what we want.
*/
vm_page_lock(fs.m);
if ((fs.m->cow) &&
(fault_type & VM_PROT_WRITE) &&
(fs.object == fs.first_object)) {
vm_page_cowfault(fs.m);
unlock_and_deallocate(&fs);
goto RetryFault;
}
/*
* Wait/Retry if the page is busy. We have to do this
* if the page is busy via either VPO_BUSY or
* vm_page_t->busy because the vm_pager may be using
* vm_page_t->busy for pageouts ( and even pageins if
* it is the vnode pager ), and we could end up trying
* to pagein and pageout the same page simultaneously.
*
* We can theoretically allow the busy case on a read
* fault if the page is marked valid, but since such
* pages are typically already pmap'd, putting that
* special case in might be more effort then it is
* worth. We cannot under any circumstances mess
* around with a vm_page_t->busy page except, perhaps,
* to pmap it.
*/
if ((fs.m->oflags & VPO_BUSY) || fs.m->busy) {
/*
* Reference the page before unlocking and
* sleeping so that the page daemon is less
* likely to reclaim it.
*/
vm_page_aflag_set(fs.m, PGA_REFERENCED);
vm_page_unlock(fs.m);
if (fs.object != fs.first_object) {
if (!VM_OBJECT_TRYWLOCK(
fs.first_object)) {
VM_OBJECT_WUNLOCK(fs.object);
VM_OBJECT_WLOCK(fs.first_object);
VM_OBJECT_WLOCK(fs.object);
}
vm_page_lock(fs.first_m);
vm_page_free(fs.first_m);
vm_page_unlock(fs.first_m);
vm_object_pip_wakeup(fs.first_object);
VM_OBJECT_WUNLOCK(fs.first_object);
fs.first_m = NULL;
}
unlock_map(&fs);
if (fs.m == vm_page_lookup(fs.object,
fs.pindex)) {
vm_page_sleep_if_busy(fs.m, TRUE,
"vmpfw");
}
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
PCPU_INC(cnt.v_intrans);
vm_object_deallocate(fs.first_object);
goto RetryFault;
}
vm_page_remque(fs.m);
vm_page_unlock(fs.m);
/*
* Mark page busy for other processes, and the
* pagedaemon. If it still isn't completely valid
* (readable), jump to readrest, else break-out ( we
* found the page ).
*/
vm_page_busy(fs.m);
if (fs.m->valid != VM_PAGE_BITS_ALL)
goto readrest;
break;
}
/*
* Page is not resident, If this is the search termination
* or the pager might contain the page, allocate a new page.
*/
if (TRYPAGER || fs.object == fs.first_object) {
if (fs.pindex >= fs.object->size) {
unlock_and_deallocate(&fs);
return (KERN_PROTECTION_FAILURE);
}
/*
* Allocate a new page for this object/offset pair.
*
* Unlocked read of the p_flag is harmless. At
* worst, the P_KILLED might be not observed
* there, and allocation can fail, causing
* restart and new reading of the p_flag.
*/
fs.m = NULL;
if (!vm_page_count_severe() || P_KILLED(curproc)) {
#if VM_NRESERVLEVEL > 0
if ((fs.object->flags & OBJ_COLORED) == 0) {
fs.object->flags |= OBJ_COLORED;
fs.object->pg_color = atop(vaddr) -
fs.pindex;
}
#endif
alloc_req = P_KILLED(curproc) ?
VM_ALLOC_SYSTEM : VM_ALLOC_NORMAL;
if (fs.object->type != OBJT_VNODE &&
fs.object->backing_object == NULL)
alloc_req |= VM_ALLOC_ZERO;
fs.m = vm_page_alloc(fs.object, fs.pindex,
alloc_req);
}
if (fs.m == NULL) {
unlock_and_deallocate(&fs);
VM_WAITPFAULT;
goto RetryFault;
} else if (fs.m->valid == VM_PAGE_BITS_ALL)
break;
}
readrest:
/*
* We have found a valid page or we have allocated a new page.
* The page thus may not be valid or may not be entirely
* valid.
*
* Attempt to fault-in the page if there is a chance that the
* pager has it, and potentially fault in additional pages
* at the same time.
*/
if (TRYPAGER) {
int rv;
u_char behavior = vm_map_entry_behavior(fs.entry);
if (behavior == MAP_ENTRY_BEHAV_RANDOM ||
P_KILLED(curproc)) {
behind = 0;
ahead = 0;
} else if (behavior == MAP_ENTRY_BEHAV_SEQUENTIAL) {
behind = 0;
ahead = atop(fs.entry->end - vaddr) - 1;
if (ahead > VM_FAULT_READ_AHEAD_MAX)
ahead = VM_FAULT_READ_AHEAD_MAX;
if (fs.pindex == fs.entry->next_read)
vm_fault_cache_behind(&fs,
VM_FAULT_READ_MAX);
} else {
/*
* If this is a sequential page fault, then
* arithmetically increase the number of pages
* in the read-ahead window. Otherwise, reset
* the read-ahead window to its smallest size.
*/
behind = atop(vaddr - fs.entry->start);
if (behind > VM_FAULT_READ_BEHIND)
behind = VM_FAULT_READ_BEHIND;
ahead = atop(fs.entry->end - vaddr) - 1;
era = fs.entry->read_ahead;
if (fs.pindex == fs.entry->next_read) {
nera = era + behind;
if (nera > VM_FAULT_READ_AHEAD_MAX)
nera = VM_FAULT_READ_AHEAD_MAX;
behind = 0;
if (ahead > nera)
ahead = nera;
if (era == VM_FAULT_READ_AHEAD_MAX)
vm_fault_cache_behind(&fs,
VM_FAULT_CACHE_BEHIND);
} else if (ahead > VM_FAULT_READ_AHEAD_MIN)
ahead = VM_FAULT_READ_AHEAD_MIN;
if (era != ahead)
fs.entry->read_ahead = ahead;
}
/*
* Call the pager to retrieve the data, if any, after
* releasing the lock on the map. We hold a ref on
* fs.object and the pages are VPO_BUSY'd.
*/
unlock_map(&fs);
if (fs.object->type == OBJT_VNODE) {
vp = fs.object->handle;
if (vp == fs.vp)
goto vnode_locked;
else if (fs.vp != NULL) {
vput(fs.vp);
fs.vp = NULL;
}
locked = VOP_ISLOCKED(vp);
if (locked != LK_EXCLUSIVE)
locked = LK_SHARED;
/* Do not sleep for vnode lock while fs.m is busy */
error = vget(vp, locked | LK_CANRECURSE |
LK_NOWAIT, curthread);
if (error != 0) {
vhold(vp);
release_page(&fs);
unlock_and_deallocate(&fs);
error = vget(vp, locked | LK_RETRY |
LK_CANRECURSE, curthread);
vdrop(vp);
fs.vp = vp;
KASSERT(error == 0,
("vm_fault: vget failed"));
goto RetryFault;
}
fs.vp = vp;
}
vnode_locked:
KASSERT(fs.vp == NULL || !fs.map->system_map,
("vm_fault: vnode-backed object mapped by system map"));
/*
* now we find out if any other pages should be paged
* in at this time this routine checks to see if the
* pages surrounding this fault reside in the same
* object as the page for this fault. If they do,
* then they are faulted in also into the object. The
* array "marray" returned contains an array of
* vm_page_t structs where one of them is the
* vm_page_t passed to the routine. The reqpage
* return value is the index into the marray for the
* vm_page_t passed to the routine.
*
* fs.m plus the additional pages are VPO_BUSY'd.
*/
faultcount = vm_fault_additional_pages(
fs.m, behind, ahead, marray, &reqpage);
rv = faultcount ?
vm_pager_get_pages(fs.object, marray, faultcount,
reqpage) : VM_PAGER_FAIL;
if (rv == VM_PAGER_OK) {
/*
* Found the page. Leave it busy while we play
* with it.
*/
/*
* Relookup in case pager changed page. Pager
* is responsible for disposition of old page
* if moved.
*/
fs.m = vm_page_lookup(fs.object, fs.pindex);
if (!fs.m) {
unlock_and_deallocate(&fs);
goto RetryFault;
}
hardfault++;
break; /* break to PAGE HAS BEEN FOUND */
}
/*
* Remove the bogus page (which does not exist at this
* object/offset); before doing so, we must get back
* our object lock to preserve our invariant.
*
* Also wake up any other process that may want to bring
* in this page.
*
* If this is the top-level object, we must leave the
* busy page to prevent another process from rushing
* past us, and inserting the page in that object at
* the same time that we are.
*/
if (rv == VM_PAGER_ERROR)
printf("vm_fault: pager read error, pid %d (%s)\n",
curproc->p_pid, curproc->p_comm);
/*
* Data outside the range of the pager or an I/O error
*/
/*
* XXX - the check for kernel_map is a kludge to work
* around having the machine panic on a kernel space
* fault w/ I/O error.
*/
if (((fs.map != kernel_map) && (rv == VM_PAGER_ERROR)) ||
(rv == VM_PAGER_BAD)) {
vm_page_lock(fs.m);
vm_page_free(fs.m);
vm_page_unlock(fs.m);
fs.m = NULL;
unlock_and_deallocate(&fs);
return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE);
}
if (fs.object != fs.first_object) {
vm_page_lock(fs.m);
vm_page_free(fs.m);
vm_page_unlock(fs.m);
fs.m = NULL;
/*
* XXX - we cannot just fall out at this
* point, m has been freed and is invalid!
*/
}
}
/*
* We get here if the object has default pager (or unwiring)
* or the pager doesn't have the page.
*/
if (fs.object == fs.first_object)
fs.first_m = fs.m;
/*
* Move on to the next object. Lock the next object before
* unlocking the current one.
*/
fs.pindex += OFF_TO_IDX(fs.object->backing_object_offset);
next_object = fs.object->backing_object;
if (next_object == NULL) {
/*
* If there's no object left, fill the page in the top
* object with zeros.
*/
if (fs.object != fs.first_object) {
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
fs.m = fs.first_m;
VM_OBJECT_WLOCK(fs.object);
}
fs.first_m = NULL;
/*
* Zero the page if necessary and mark it valid.
*/
if ((fs.m->flags & PG_ZERO) == 0) {
pmap_zero_page(fs.m);
} else {
PCPU_INC(cnt.v_ozfod);
}
PCPU_INC(cnt.v_zfod);
fs.m->valid = VM_PAGE_BITS_ALL;
break; /* break to PAGE HAS BEEN FOUND */
} else {
KASSERT(fs.object != next_object,
("object loop %p", next_object));
VM_OBJECT_WLOCK(next_object);
vm_object_pip_add(next_object, 1);
if (fs.object != fs.first_object)
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
fs.object = next_object;
}
}
KASSERT((fs.m->oflags & VPO_BUSY) != 0,
("vm_fault: not busy after main loop"));
/*
* PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
* is held.]
*/
/*
* If the page is being written, but isn't already owned by the
* top-level object, we have to copy it into a new page owned by the
* top-level object.
*/
if (fs.object != fs.first_object) {
/*
* We only really need to copy if we want to write it.
*/
if ((fault_type & (VM_PROT_COPY | VM_PROT_WRITE)) != 0) {
/*
* This allows pages to be virtually copied from a
* backing_object into the first_object, where the
* backing object has no other refs to it, and cannot
* gain any more refs. Instead of a bcopy, we just
* move the page from the backing object to the
* first object. Note that we must mark the page
* dirty in the first object so that it will go out
* to swap when needed.
*/
is_first_object_locked = FALSE;
if (
/*
* Only one shadow object
*/
(fs.object->shadow_count == 1) &&
/*
* No COW refs, except us
*/
(fs.object->ref_count == 1) &&
/*
* No one else can look this object up
*/
(fs.object->handle == NULL) &&
/*
* No other ways to look the object up
*/
((fs.object->type == OBJT_DEFAULT) ||
(fs.object->type == OBJT_SWAP)) &&
(is_first_object_locked = VM_OBJECT_TRYWLOCK(fs.first_object)) &&
/*
* We don't chase down the shadow chain
*/
fs.object == fs.first_object->backing_object) {
/*
* get rid of the unnecessary page
*/
vm_page_lock(fs.first_m);
vm_page_free(fs.first_m);
vm_page_unlock(fs.first_m);
/*
* grab the page and put it into the
* process'es object. The page is
* automatically made dirty.
*/
vm_page_lock(fs.m);
vm_page_rename(fs.m, fs.first_object, fs.first_pindex);
vm_page_unlock(fs.m);
vm_page_busy(fs.m);
fs.first_m = fs.m;
fs.m = NULL;
PCPU_INC(cnt.v_cow_optim);
} else {
/*
* Oh, well, lets copy it.
*/
pmap_copy_page(fs.m, fs.first_m);
fs.first_m->valid = VM_PAGE_BITS_ALL;
if (wired && (fault_flags &
VM_FAULT_CHANGE_WIRING) == 0) {
vm_page_lock(fs.first_m);
vm_page_wire(fs.first_m);
vm_page_unlock(fs.first_m);
vm_page_lock(fs.m);
vm_page_unwire(fs.m, FALSE);
vm_page_unlock(fs.m);
}
/*
* We no longer need the old page or object.
*/
release_page(&fs);
}
/*
* fs.object != fs.first_object due to above
* conditional
*/
vm_object_pip_wakeup(fs.object);
VM_OBJECT_WUNLOCK(fs.object);
/*
* Only use the new page below...
*/
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
fs.m = fs.first_m;
if (!is_first_object_locked)
VM_OBJECT_WLOCK(fs.object);
PCPU_INC(cnt.v_cow_faults);
curthread->td_cow++;
} else {
prot &= ~VM_PROT_WRITE;
}
}
/*
* We must verify that the maps have not changed since our last
* lookup.
*/
if (!fs.lookup_still_valid) {
vm_object_t retry_object;
vm_pindex_t retry_pindex;
vm_prot_t retry_prot;
if (!vm_map_trylock_read(fs.map)) {
release_page(&fs);
unlock_and_deallocate(&fs);
goto RetryFault;
}
fs.lookup_still_valid = TRUE;
if (fs.map->timestamp != map_generation) {
result = vm_map_lookup_locked(&fs.map, vaddr, fault_type,
&fs.entry, &retry_object, &retry_pindex, &retry_prot, &wired);
/*
* If we don't need the page any longer, put it on the inactive
* list (the easiest thing to do here). If no one needs it,
* pageout will grab it eventually.
*/
if (result != KERN_SUCCESS) {
release_page(&fs);
unlock_and_deallocate(&fs);
/*
* If retry of map lookup would have blocked then
* retry fault from start.
*/
if (result == KERN_FAILURE)
goto RetryFault;
return (result);
}
if ((retry_object != fs.first_object) ||
(retry_pindex != fs.first_pindex)) {
release_page(&fs);
unlock_and_deallocate(&fs);
goto RetryFault;
}
/*
* Check whether the protection has changed or the object has
* been copied while we left the map unlocked. Changing from
* read to write permission is OK - we leave the page
* write-protected, and catch the write fault. Changing from
* write to read permission means that we can't mark the page
* write-enabled after all.
*/
prot &= retry_prot;
}
}
/*
* If the page was filled by a pager, update the map entry's
* last read offset. Since the pager does not return the
* actual set of pages that it read, this update is based on
* the requested set. Typically, the requested and actual
* sets are the same.
*
* XXX The following assignment modifies the map
* without holding a write lock on it.
*/
if (hardfault)
fs.entry->next_read = fs.pindex + faultcount - reqpage;
if ((prot & VM_PROT_WRITE) != 0 ||
(fault_flags & VM_FAULT_DIRTY) != 0) {
vm_object_set_writeable_dirty(fs.object);
/*
* If this is a NOSYNC mmap we do not want to set VPO_NOSYNC
* if the page is already dirty to prevent data written with
* the expectation of being synced from not being synced.
* Likewise if this entry does not request NOSYNC then make
* sure the page isn't marked NOSYNC. Applications sharing
* data should use the same flags to avoid ping ponging.
*/
if (fs.entry->eflags & MAP_ENTRY_NOSYNC) {
if (fs.m->dirty == 0)
fs.m->oflags |= VPO_NOSYNC;
} else {
fs.m->oflags &= ~VPO_NOSYNC;
}
/*
* If the fault is a write, we know that this page is being
* written NOW so dirty it explicitly to save on
* pmap_is_modified() calls later.
*
* Also tell the backing pager, if any, that it should remove
* any swap backing since the page is now dirty.
*/
if (((fault_type & VM_PROT_WRITE) != 0 &&
(fault_flags & VM_FAULT_CHANGE_WIRING) == 0) ||
(fault_flags & VM_FAULT_DIRTY) != 0) {
vm_page_dirty(fs.m);
vm_pager_page_unswapped(fs.m);
}
}
/*
* Page had better still be busy
*/
KASSERT(fs.m->oflags & VPO_BUSY,
("vm_fault: page %p not busy!", fs.m));
/*
* Page must be completely valid or it is not fit to
* map into user space. vm_pager_get_pages() ensures this.
*/
KASSERT(fs.m->valid == VM_PAGE_BITS_ALL,
("vm_fault: page %p partially invalid", fs.m));
VM_OBJECT_WUNLOCK(fs.object);
/*
* Put this page into the physical map. We had to do the unlock above
* because pmap_enter() may sleep. We don't put the page
* back on the active queue until later so that the pageout daemon
* won't find it (yet).
*/
pmap_enter(fs.map->pmap, vaddr, fault_type, fs.m, prot, wired);
if ((fault_flags & VM_FAULT_CHANGE_WIRING) == 0 && wired == 0)
vm_fault_prefault(fs.map->pmap, vaddr, fs.entry);
VM_OBJECT_WLOCK(fs.object);
vm_page_lock(fs.m);
/*
* If the page is not wired down, then put it where the pageout daemon
* can find it.
*/
if (fault_flags & VM_FAULT_CHANGE_WIRING) {
if (wired)
vm_page_wire(fs.m);
else
vm_page_unwire(fs.m, 1);
} else
vm_page_activate(fs.m);
if (m_hold != NULL) {
*m_hold = fs.m;
vm_page_hold(fs.m);
}
vm_page_unlock(fs.m);
vm_page_wakeup(fs.m);
/*
* Unlock everything, and return
*/
unlock_and_deallocate(&fs);
if (hardfault) {
PCPU_INC(cnt.v_io_faults);
curthread->td_ru.ru_majflt++;
} else
curthread->td_ru.ru_minflt++;
return (KERN_SUCCESS);
}
/*
* Hold each of the physical pages that are mapped by the specified range of
* virtual addresses, ["addr", "addr" + "len"), if those mappings are valid
* and allow the specified types of access, "prot". If all of the implied
* pages are successfully held, then the number of held pages is returned
* together with pointers to those pages in the array "ma". However, if any
* of the pages cannot be held, -1 is returned.
*/
int
vm_fault_quick_hold_pages(vm_map_t map, vm_offset_t addr, vm_size_t len,
vm_prot_t prot, vm_page_t *ma, int max_count)
{
vm_offset_t end, va;
vm_page_t *mp;
int count;
boolean_t pmap_failed;
if (len == 0)
return (0);
end = round_page(addr + len);
addr = trunc_page(addr);
/*
* Check for illegal addresses.
*/
if (addr < vm_map_min(map) || addr > end || end > vm_map_max(map))
return (-1);
count = howmany(end - addr, PAGE_SIZE);
if (count > max_count)
panic("vm_fault_quick_hold_pages: count > max_count");
/*
* Most likely, the physical pages are resident in the pmap, so it is
* faster to try pmap_extract_and_hold() first.
*/
pmap_failed = FALSE;
for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE) {
*mp = pmap_extract_and_hold(map->pmap, va, prot);
if (*mp == NULL)
pmap_failed = TRUE;
else if ((prot & VM_PROT_WRITE) != 0 &&
(*mp)->dirty != VM_PAGE_BITS_ALL) {
/*
* Explicitly dirty the physical page. Otherwise, the
* caller's changes may go unnoticed because they are
* performed through an unmanaged mapping or by a DMA
* operation.
*
* The object lock is not held here.
* See vm_page_clear_dirty_mask().
*/
vm_page_dirty(*mp);
}
}
if (pmap_failed) {
/*
* One or more pages could not be held by the pmap. Either no
* page was mapped at the specified virtual address or that
* mapping had insufficient permissions. Attempt to fault in
* and hold these pages.
*/
for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE)
if (*mp == NULL && vm_fault_hold(map, va, prot,
VM_FAULT_NORMAL, mp) != KERN_SUCCESS)
goto error;
}
return (count);
error:
for (mp = ma; mp < ma + count; mp++)
if (*mp != NULL) {
vm_page_lock(*mp);
vm_page_unhold(*mp);
vm_page_unlock(*mp);
}
return (-1);
}
static int
sg_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
{
struct sglist *sg;
vm_page_t m_paddr, page;
vm_pindex_t offset;
vm_paddr_t paddr;
vm_memattr_t memattr;
size_t space;
int i;
VM_OBJECT_ASSERT_WLOCKED(object);
sg = object->handle;
memattr = object->memattr;
VM_OBJECT_WUNLOCK(object);
offset = m[reqpage]->pindex;
/*
* Lookup the physical address of the requested page. An initial
* value of '1' instead of '0' is used so we can assert that the
* page is found since '0' can be a valid page-aligned physical
* address.
*/
space = 0;
paddr = 1;
for (i = 0; i < sg->sg_nseg; i++) {
if (space + sg->sg_segs[i].ss_len <= (offset * PAGE_SIZE)) {
space += sg->sg_segs[i].ss_len;
continue;
}
paddr = sg->sg_segs[i].ss_paddr + offset * PAGE_SIZE - space;
break;
}
KASSERT(paddr != 1, ("invalid SG page index"));
/* If "paddr" is a real page, perform a sanity check on "memattr". */
if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
pmap_page_get_memattr(m_paddr) != memattr) {
memattr = pmap_page_get_memattr(m_paddr);
printf(
"WARNING: A device driver has set \"memattr\" inconsistently.\n");
}
/* Return a fake page for the requested page. */
KASSERT(!(m[reqpage]->flags & PG_FICTITIOUS),
("backing page for SG is fake"));
/* Construct a new fake page. */
page = vm_page_getfake(paddr, memattr);
VM_OBJECT_WLOCK(object);
TAILQ_INSERT_TAIL(&object->un_pager.sgp.sgp_pglist, page, plinks.q);
/* Free the original pages and insert this fake page into the object. */
for (i = 0; i < count; i++) {
if (i == reqpage &&
vm_page_replace(page, object, offset) != m[i])
panic("sg_pager_getpages: invalid place replacement");
vm_page_lock(m[i]);
vm_page_free(m[i]);
vm_page_unlock(m[i]);
}
m[reqpage] = page;
page->valid = VM_PAGE_BITS_ALL;
return (VM_PAGER_OK);
}
int
socow_setup(struct mbuf *m0, struct uio *uio)
{
struct sf_buf *sf;
vm_page_t pp;
struct iovec *iov;
struct vmspace *vmspace;
struct vm_map *map;
vm_offset_t offset, uva;
socow_stats.attempted++;
vmspace = curproc->p_vmspace;
map = &vmspace->vm_map;
uva = (vm_offset_t) uio->uio_iov->iov_base;
offset = uva & PAGE_MASK;
/*
* Verify that access to the given address is allowed from user-space.
*/
if (vm_fault_quick((caddr_t)uva, VM_PROT_READ) < 0)
return (0);
/*
* verify page is mapped & not already wired for i/o
*/
pp = pmap_extract_and_hold(map->pmap, uva, VM_PROT_READ);
if (pp == NULL) {
socow_stats.fail_not_mapped++;
return(0);
}
/*
* set up COW
*/
vm_page_lock(pp);
if (vm_page_cowsetup(pp) != 0) {
vm_page_unhold(pp);
vm_page_unlock(pp);
return (0);
}
/*
* wire the page for I/O
*/
vm_page_wire(pp);
vm_page_unhold(pp);
vm_page_unlock(pp);
/*
* Allocate an sf buf
*/
sf = sf_buf_alloc(pp, SFB_CATCH);
if (sf == NULL) {
vm_page_lock(pp);
vm_page_cowclear(pp);
vm_page_unwire(pp, 0);
/*
* Check for the object going away on us. This can
* happen since we don't hold a reference to it.
* If so, we're responsible for freeing the page.
*/
if (pp->wire_count == 0 && pp->object == NULL)
vm_page_free(pp);
vm_page_unlock(pp);
socow_stats.fail_sf_buf++;
return(0);
}
/*
* attach to mbuf
*/
MEXTADD(m0, sf_buf_kva(sf), PAGE_SIZE, socow_iodone,
(void*)sf_buf_kva(sf), sf, M_RDONLY, EXT_SFBUF);
m0->m_len = PAGE_SIZE - offset;
m0->m_data = (caddr_t)sf_buf_kva(sf) + offset;
socow_stats.success++;
iov = uio->uio_iov;
iov->iov_base = (char *)iov->iov_base + m0->m_len;
iov->iov_len -= m0->m_len;
uio->uio_resid -= m0->m_len;
uio->uio_offset += m0->m_len;
if (iov->iov_len == 0) {
uio->uio_iov++;
uio->uio_iovcnt--;
}
return(m0->m_len);
}
