int
ttm_bo_mmap_single(struct ttm_bo_device *bdev, vm_ooffset_t *offset, vm_size_t size,
struct vm_object **obj_res, int nprot)
{
struct ttm_bo_driver *driver;
struct ttm_buffer_object *bo;
struct vm_object *vm_obj;
int ret;
rw_wlock(&bdev->vm_lock);
bo = ttm_bo_vm_lookup_rb(bdev, OFF_TO_IDX(*offset), OFF_TO_IDX(size));
if (likely(bo != NULL))
refcount_acquire(&bo->kref);
rw_wunlock(&bdev->vm_lock);
if (unlikely(bo == NULL)) {
printf("[TTM] Could not find buffer object to map\n");
return (EINVAL);
}
driver = bo->bdev->driver;
if (unlikely(!driver->verify_access)) {
ret = EPERM;
goto out_unref;
}
ret = -driver->verify_access(bo);
if (unlikely(ret != 0))
goto out_unref;
vm_obj = cdev_pager_allocate(bo, OBJT_MGTDEVICE, &ttm_pager_ops,
size, nprot, 0, curthread->td_ucred);
if (vm_obj == NULL) {
ret = EINVAL;
goto out_unref;
}
/*
* Note: We're transferring the bo reference to vm_obj->handle here.
*/
*offset = 0;
*obj_res = vm_obj;
return 0;
out_unref:
ttm_bo_unref(&bo);
return ret;
}
/*
* MPSAFE
*/
static vm_object_t
phys_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t foff, struct ucred *cred)
{
vm_object_t object, object1;
vm_pindex_t pindex;
/*
* Offset should be page aligned.
*/
if (foff & PAGE_MASK)
return (NULL);
pindex = OFF_TO_IDX(foff + PAGE_MASK + size);
if (handle != NULL) {
mtx_lock(&phys_pager_mtx);
/*
* Look up pager, creating as necessary.
*/
object1 = NULL;
object = vm_pager_object_lookup(&phys_pager_object_list, handle);
if (object == NULL) {
/*
* Allocate object and associate it with the pager.
*/
mtx_unlock(&phys_pager_mtx);
object1 = vm_object_allocate(OBJT_PHYS, pindex);
mtx_lock(&phys_pager_mtx);
object = vm_pager_object_lookup(&phys_pager_object_list,
handle);
if (object != NULL) {
/*
* We raced with other thread while
* allocating object.
*/
if (pindex > object->size)
object->size = pindex;
} else {
object = object1;
object1 = NULL;
object->handle = handle;
TAILQ_INSERT_TAIL(&phys_pager_object_list, object,
pager_object_list);
}
} else {
if (pindex > object->size)
object->size = pindex;
}
mtx_unlock(&phys_pager_mtx);
vm_object_deallocate(object1);
} else {
object = vm_object_allocate(OBJT_PHYS, pindex);
}
return (object);
}
/*
* no_pager_alloc just returns an initialized object.
*/
static vm_object_t
default_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t offset)
{
if (handle != NULL)
panic("default_pager_alloc: handle specified");
return vm_object_allocate(OBJT_DEFAULT, OFF_TO_IDX(round_page(offset + size)));
}
int drm_legacy_sg_alloc(struct drm_device *dev, void *data,
struct drm_file *file_priv)
{
struct drm_scatter_gather *request = data;
struct drm_sg_mem *entry;
vm_size_t size;
vm_pindex_t pindex;
if (dev->sg)
return -EINVAL;
DRM_DEBUG("request size=%ld\n", request->size);
entry = kmalloc(sizeof(*entry), M_DRM, M_WAITOK | M_ZERO);
size = round_page(request->size);
entry->pages = OFF_TO_IDX(size);
entry->busaddr = kmalloc(entry->pages * sizeof(*entry->busaddr),
M_DRM, M_WAITOK | M_ZERO);
entry->vaddr = kmem_alloc_attr(&kernel_map, size,
VM_SUBSYS_DRM_SCAT, M_WAITOK | M_ZERO,
0, BUS_SPACE_MAXADDR_32BIT,
VM_MEMATTR_WRITE_COMBINING);
if (entry->vaddr == 0) {
drm_sg_cleanup(entry);
return (-ENOMEM);
}
for(pindex = 0; pindex < entry->pages; pindex++) {
entry->busaddr[pindex] =
vtophys(entry->vaddr + IDX_TO_OFF(pindex));
}
DRM_LOCK(dev);
if (dev->sg) {
DRM_UNLOCK(dev);
drm_sg_cleanup(entry);
return (-EINVAL);
}
dev->sg = entry;
DRM_UNLOCK(dev);
request->handle = entry->vaddr;
DRM_DEBUG("allocated %ju pages @ 0x%08jx, contents=%08lx\n",
entry->pages, (uintmax_t)entry->vaddr,
*(unsigned long *)entry->vaddr);
return (0);
}
int drm_sg_alloc(struct drm_device *dev, struct drm_scatter_gather * request)
{
struct drm_sg_mem *entry;
vm_size_t size;
vm_pindex_t pindex;
DRM_DEBUG("\n");
if (!drm_core_check_feature(dev, DRIVER_SG))
return -EINVAL;
if (dev->sg)
return -EINVAL;
entry = malloc(sizeof(*entry), DRM_MEM_DRIVER, M_NOWAIT | M_ZERO);
if (!entry)
return -ENOMEM;
DRM_DEBUG("request size=%ld\n", request->size);
size = round_page(request->size);
entry->pages = OFF_TO_IDX(size);
entry->busaddr = malloc(entry->pages * sizeof(*entry->busaddr),
DRM_MEM_SGLISTS, M_NOWAIT | M_ZERO);
if (!entry->busaddr) {
free(entry, DRM_MEM_DRIVER);
return -ENOMEM;
}
entry->vaddr = drm_vmalloc_dma(size);
if (entry->vaddr == 0) {
free(entry->busaddr, DRM_MEM_DRIVER);
free(entry, DRM_MEM_DRIVER);
return -ENOMEM;
}
for (pindex = 0; pindex < entry->pages; pindex++) {
entry->busaddr[pindex] =
vtophys(entry->vaddr + IDX_TO_OFF(pindex));
}
request->handle = entry->vaddr;
dev->sg = entry;
DRM_DEBUG("allocated %ju pages @ 0x%08zx, contents=%08lx\n",
entry->pages, entry->vaddr, *(unsigned long *)entry->vaddr);
return 0;
}
static vm_object_t
sg_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t foff, struct ucred *cred)
{
struct sglist *sg;
vm_object_t object;
vm_pindex_t npages, pindex;
int i;
/*
* Offset should be page aligned.
*/
if (foff & PAGE_MASK)
return (NULL);
/*
* The scatter/gather list must only include page-aligned
* ranges.
*/
npages = 0;
sg = handle;
for (i = 0; i < sg->sg_nseg; i++) {
if ((sg->sg_segs[i].ss_paddr % PAGE_SIZE) != 0 ||
(sg->sg_segs[i].ss_len % PAGE_SIZE) != 0)
return (NULL);
npages += sg->sg_segs[i].ss_len / PAGE_SIZE;
}
/*
* The scatter/gather list has a fixed size. Refuse requests
* to map beyond that.
*/
size = round_page(size);
pindex = OFF_TO_IDX(foff + size);
if (pindex > npages)
return (NULL);
/*
* Allocate a new object and associate it with the
* scatter/gather list. It is ok for our purposes to have
* multiple VM objects associated with the same scatter/gather
* list because scatter/gather lists are static. This is also
* simpler than ensuring a unique object per scatter/gather
* list.
*/
object = vm_object_allocate(OBJT_SG, npages);
object->handle = sglist_hold(sg);
TAILQ_INIT(&object->un_pager.sgp.sgp_pglist);
return (object);
}
static int
netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
struct netmap_vm_handle_t *vmh = object->handle;
struct netmap_priv_d *priv = vmh->priv;
vm_paddr_t paddr;
vm_page_t page;
vm_memattr_t memattr;
vm_pindex_t pidx;
ND("object %p offset %jd prot %d mres %p",
object, (intmax_t)offset, prot, mres);
memattr = object->memattr;
pidx = OFF_TO_IDX(offset);
paddr = netmap_mem_ofstophys(priv->np_mref, offset);
if (paddr == 0)
return VM_PAGER_FAIL;
if (((*mres)->flags & PG_FICTITIOUS) != 0) {
/*
* If the passed in result page is a fake page, update it with
* the new physical address.
*/
page = *mres;
vm_page_updatefake(page, paddr, memattr);
} else {
/*
* Replace the passed in reqpage page with our own fake page and
* free up the all of the original pages.
*/
#ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */
#define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK
#define VM_OBJECT_WLOCK VM_OBJECT_LOCK
#endif /* VM_OBJECT_WUNLOCK */
VM_OBJECT_WUNLOCK(object);
page = vm_page_getfake(paddr, memattr);
VM_OBJECT_WLOCK(object);
vm_page_free(*mres);
*mres = page;
vm_page_insert(page, object, pidx);
}
page->valid = VM_PAGE_BITS_ALL;
return (VM_PAGER_OK);
}
static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
vm_paddr_t paddr;
vm_page_t page;
vm_offset_t pidx = OFF_TO_IDX(offset);
cdev_t dev;
page = *mres;
dev = object->handle;
paddr = pmap_phys_address(
dev_dmmap(dev, offset, prot, NULL));
KASSERT(paddr != -1,("dev_pager_getpage: map function returns error"));
KKASSERT(object->type == OBJT_DEVICE);
if (page->flags & PG_FICTITIOUS) {
/*
* If the passed in reqpage page is already a fake page,
* update it with the new physical address.
*/
page->phys_addr = paddr;
page->valid = VM_PAGE_BITS_ALL;
} else {
/*
* Replace the passed in reqpage page with our own fake page
* and free up all the original pages.
*/
page = dev_pager_getfake(paddr, object->memattr);
TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
page, pageq);
vm_object_hold(object);
vm_page_free(*mres);
if (vm_page_insert(page, object, pidx) == FALSE) {
panic("dev_pager_getpage: page (%p,%016jx) exists",
object, (uintmax_t)pidx);
}
vm_object_drop(object);
}
return (VM_PAGER_OK);
}
static void
pscnv_gem_pager_dtor(void *handle)
{
struct drm_gem_object *gem_obj = handle;
struct pscnv_bo *bo = gem_obj->driver_private;
struct drm_device *dev = gem_obj->dev;
vm_object_t devobj;
DRM_LOCK(dev);
devobj = cdev_pager_lookup(handle);
if (devobj != NULL) {
vm_size_t page_count = OFF_TO_IDX(bo->size);
vm_page_t m;
int i;
VM_OBJECT_LOCK(devobj);
for (i = 0; i < page_count; i++) {
m = vm_page_lookup(devobj, i);
if (!m)
continue;
if (pscnv_mem_debug > 0)
NV_WARN(dev, "Freeing %010llx + %08llx (%p\n", bo->start, i * PAGE_SIZE, m);
cdev_pager_free_page(devobj, m);
}
VM_OBJECT_UNLOCK(devobj);
vm_object_deallocate(devobj);
}
else {
DRM_UNLOCK(dev);
NV_ERROR(dev, "Could not find handle %p bo %p\n", handle, bo);
return;
}
if (pscnv_mem_debug > 0)
NV_WARN(dev, "Freed %010llx (%p)\n", bo->start, bo);
//kfree(bo->fake_pages);
if (bo->chan)
pscnv_chan_unref(bo->chan);
else
drm_gem_object_unreference_unlocked(gem_obj);
DRM_UNLOCK(dev);
}
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);
}
/*
* no_pager_alloc just returns an initialized object.
*/
static vm_object_t
default_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t offset, struct ucred *cred)
{
vm_object_t object;
if (handle != NULL)
panic("default_pager_alloc: handle specified");
if (cred != NULL) {
if (!swap_reserve_by_cred(size, cred))
return (NULL);
crhold(cred);
}
object = vm_object_allocate(OBJT_DEFAULT,
OFF_TO_IDX(round_page(offset + size)));
if (cred != NULL) {
VM_OBJECT_WLOCK(object);
object->cred = cred;
object->charge = size;
VM_OBJECT_WUNLOCK(object);
}
return (object);
}
static int
uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
{
vm_page_t m;
vm_pindex_t idx;
size_t tlen;
int error, offset, rv;
idx = OFF_TO_IDX(uio->uio_offset);
offset = uio->uio_offset & PAGE_MASK;
tlen = MIN(PAGE_SIZE - offset, len);
VM_OBJECT_WLOCK(obj);
/*
* Read I/O without either a corresponding resident page or swap
* page: use zero_region. This is intended to avoid instantiating
* pages on read from a sparse region.
*/
if (uio->uio_rw == UIO_READ && vm_page_lookup(obj, idx) == NULL &&
!vm_pager_has_page(obj, idx, NULL, NULL)) {
VM_OBJECT_WUNLOCK(obj);
return (uiomove(__DECONST(void *, zero_region), tlen, uio));
}
static int
old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t foff, struct ucred *cred, u_short *color)
{
unsigned int npages;
vm_offset_t off;
cdev_t dev;
dev = handle;
/*
* Check that the specified range of the device allows the desired
* protection.
*
* XXX assumes VM_PROT_* == PROT_*
*/
npages = OFF_TO_IDX(size);
for (off = foff; npages--; off += PAGE_SIZE) {
if (dev_dmmap(dev, off, (int)prot, NULL) == -1)
return (EINVAL);
}
return (0);
}
/*
* Allocate (or lookup) pager for a vnode.
* Handle is a vnode pointer.
*
* MPSAFE
*/
vm_object_t
vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
vm_ooffset_t offset, struct ucred *cred)
{
vm_object_t object;
struct vnode *vp;
/*
* Pageout to vnode, no can do yet.
*/
if (handle == NULL)
return (NULL);
vp = (struct vnode *) handle;
/*
* If the object is being terminated, wait for it to
* go away.
*/
retry:
while ((object = vp->v_object) != NULL) {
VM_OBJECT_WLOCK(object);
if ((object->flags & OBJ_DEAD) == 0)
break;
vm_object_set_flag(object, OBJ_DISCONNECTWNT);
VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0);
}
KASSERT(vp->v_usecount != 0, ("vnode_pager_alloc: no vnode reference"));
if (object == NULL) {
/*
* Add an object of the appropriate size
*/
object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size)));
object->un_pager.vnp.vnp_size = size;
object->un_pager.vnp.writemappings = 0;
object->handle = handle;
VI_LOCK(vp);
if (vp->v_object != NULL) {
/*
* Object has been created while we were sleeping
*/
VI_UNLOCK(vp);
VM_OBJECT_WLOCK(object);
KASSERT(object->ref_count == 1,
("leaked ref %p %d", object, object->ref_count));
object->type = OBJT_DEAD;
object->ref_count = 0;
VM_OBJECT_WUNLOCK(object);
vm_object_destroy(object);
goto retry;
}
vp->v_object = object;
VI_UNLOCK(vp);
} else {
object->ref_count++;
#if VM_NRESERVLEVEL > 0
vm_object_color(object, 0);
#endif
VM_OBJECT_WUNLOCK(object);
}
vref(vp);
return (object);
}
/*
* Lets the VM system know about a change in size for a file.
* We adjust our own internal size and flush any cached pages in
* the associated object that are affected by the size change.
*
* NOTE: This routine may be invoked as a result of a pager put
* operation (possibly at object termination time), so we must be careful.
*
* NOTE: vp->v_filesize is initialized to NOOFFSET (-1), be sure that
* we do not blow up on the case. nsize will always be >= 0, however.
*/
void
vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize)
{
vm_pindex_t nobjsize;
vm_pindex_t oobjsize;
vm_object_t object;
object = vp->v_object;
if (object == NULL)
return;
vm_object_hold(object);
KKASSERT(vp->v_object == object);
/*
* Hasn't changed size
*/
if (nsize == vp->v_filesize) {
vm_object_drop(object);
return;
}
/*
* Has changed size. Adjust the VM object's size and v_filesize
* before we start scanning pages to prevent new pages from being
* allocated during the scan.
*/
nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
oobjsize = object->size;
object->size = nobjsize;
/*
* File has shrunk. Toss any cached pages beyond the new EOF.
*/
if (nsize < vp->v_filesize) {
vp->v_filesize = nsize;
if (nobjsize < oobjsize) {
vm_object_page_remove(object, nobjsize, oobjsize,
FALSE);
}
/*
* This gets rid of garbage at the end of a page that is now
* only partially backed by the vnode. Since we are setting
* the entire page valid & clean after we are done we have
* to be sure that the portion of the page within the file
* bounds is already valid. If it isn't then making it
* valid would create a corrupt block.
*/
if (nsize & PAGE_MASK) {
vm_offset_t kva;
vm_page_t m;
m = vm_page_lookup_busy_wait(object, OFF_TO_IDX(nsize),
TRUE, "vsetsz");
if (m && m->valid) {
int base = (int)nsize & PAGE_MASK;
int size = PAGE_SIZE - base;
struct lwbuf *lwb;
struct lwbuf lwb_cache;
/*
* Clear out partial-page garbage in case
* the page has been mapped.
*
* This is byte aligned.
*/
lwb = lwbuf_alloc(m, &lwb_cache);
kva = lwbuf_kva(lwb);
bzero((caddr_t)kva + base, size);
lwbuf_free(lwb);
/*
* XXX work around SMP data integrity race
* by unmapping the page from user processes.
* The garbage we just cleared may be mapped
* to a user process running on another cpu
* and this code is not running through normal
* I/O channels which handle SMP issues for
* us, so unmap page to synchronize all cpus.
*
* XXX should vm_pager_unmap_page() have
* dealt with this?
*/
vm_page_protect(m, VM_PROT_NONE);
/*
* Clear out partial-page dirty bits. This
* has the side effect of setting the valid
* bits, but that is ok. There are a bunch
* of places in the VM system where we expected
* m->dirty == VM_PAGE_BITS_ALL. The file EOF
* case is one of them. If the page is still
* partially dirty, make it fully dirty.
*
* NOTE: We do not clear out the valid
* bits. This would prevent bogus_page
* replacement from working properly.
*
* NOTE: We do not want to clear the dirty
* bit for a partial DEV_BSIZE'd truncation!
* This is DEV_BSIZE aligned!
*/
vm_page_clear_dirty_beg_nonincl(m, base, size);
if (m->dirty != 0)
m->dirty = VM_PAGE_BITS_ALL;
vm_page_wakeup(m);
} else if (m) {
vm_page_wakeup(m);
}
}
} else {
vp->v_filesize = nsize;
}
vm_object_drop(object);
}
static void
os_balloonobject_create(void)
{
global_state.vmobject = vm_object_allocate(OBJT_DEFAULT,
OFF_TO_IDX(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS));
}
static int
uiomove_object_page(vm_object_t obj, size_t len, struct uio *uio)
{
vm_page_t m;
vm_pindex_t idx;
size_t tlen;
int error, offset, rv;
idx = OFF_TO_IDX(uio->uio_offset);
offset = uio->uio_offset & PAGE_MASK;
tlen = MIN(PAGE_SIZE - offset, len);
VM_OBJECT_WLOCK(obj);
/*
* Parallel reads of the page content from disk are prevented
* by exclusive busy.
*
* Although the tmpfs vnode lock is held here, it is
* nonetheless safe to sleep waiting for a free page. The
* pageout daemon does not need to acquire the tmpfs vnode
* lock to page out tobj's pages because tobj is a OBJT_SWAP
* type object.
*/
m = vm_page_grab(obj, idx, VM_ALLOC_NORMAL);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(obj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(obj, &m, 1, 0);
m = vm_page_lookup(obj, idx);
if (m == NULL) {
printf(
"uiomove_object: vm_obj %p idx %jd null lookup rv %d\n",
obj, idx, rv);
VM_OBJECT_WUNLOCK(obj);
return (EIO);
}
if (rv != VM_PAGER_OK) {
printf(
"uiomove_object: vm_obj %p idx %jd valid %x pager error %d\n",
obj, idx, m->valid, rv);
vm_page_lock(m);
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_WUNLOCK(obj);
return (EIO);
}
} else
vm_page_zero_invalid(m, TRUE);
}
vm_page_xunbusy(m);
vm_page_lock(m);
vm_page_hold(m);
vm_page_unlock(m);
VM_OBJECT_WUNLOCK(obj);
error = uiomove_fromphys(&m, offset, tlen, uio);
if (uio->uio_rw == UIO_WRITE && error == 0) {
VM_OBJECT_WLOCK(obj);
vm_page_dirty(m);
VM_OBJECT_WUNLOCK(obj);
}
vm_page_lock(m);
vm_page_unhold(m);
if (m->queue == PQ_NONE) {
vm_page_deactivate(m);
} else {
/* Requeue to maintain LRU ordering. */
vm_page_requeue(m);
}
vm_page_unlock(m);
return (error);
}
static int
tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
{
struct sf_buf *sf;
vm_pindex_t idx;
vm_page_t m;
vm_offset_t offset;
off_t addr;
size_t tlen;
char *ma;
int error;
addr = uio->uio_offset;
idx = OFF_TO_IDX(addr);
offset = addr & PAGE_MASK;
tlen = MIN(PAGE_SIZE - offset, len);
if ((vobj == NULL) ||
(vobj->resident_page_count == 0 && vobj->cache == NULL))
goto nocache;
VM_OBJECT_LOCK(vobj);
lookupvpg:
if (((m = vm_page_lookup(vobj, idx)) != NULL) &&
vm_page_is_valid(m, offset, tlen)) {
if ((m->oflags & VPO_BUSY) != 0) {
/*
* Reference the page before unlocking and sleeping so
* that the page daemon is less likely to reclaim it.
*/
vm_page_reference(m);
vm_page_sleep(m, "tmfsmr");
goto lookupvpg;
}
vm_page_busy(m);
VM_OBJECT_UNLOCK(vobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(vobj);
vm_page_wakeup(m);
VM_OBJECT_UNLOCK(vobj);
return (error);
} else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) {
KASSERT(offset == 0,
("unexpected offset in tmpfs_mappedread for sendfile"));
if ((m->oflags & VPO_BUSY) != 0) {
/*
* Reference the page before unlocking and sleeping so
* that the page daemon is less likely to reclaim it.
*/
vm_page_reference(m);
vm_page_sleep(m, "tmfsmr");
goto lookupvpg;
}
vm_page_busy(m);
VM_OBJECT_UNLOCK(vobj);
sched_pin();
sf = sf_buf_alloc(m, SFB_CPUPRIVATE);
ma = (char *)sf_buf_kva(sf);
error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma);
if (error == 0) {
if (tlen != PAGE_SIZE)
bzero(ma + tlen, PAGE_SIZE - tlen);
uio->uio_offset += tlen;
uio->uio_resid -= tlen;
}
sf_buf_free(sf);
sched_unpin();
VM_OBJECT_LOCK(vobj);
if (error == 0)
m->valid = VM_PAGE_BITS_ALL;
vm_page_wakeup(m);
VM_OBJECT_UNLOCK(vobj);
return (error);
}
VM_OBJECT_UNLOCK(vobj);
nocache:
error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio);
return (error);
}
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);
}
static int
shm_dotruncate(struct shmfd *shmfd, off_t length)
{
vm_object_t object;
vm_page_t m;
vm_pindex_t nobjsize;
vm_ooffset_t delta;
object = shmfd->shm_object;
VM_OBJECT_LOCK(object);
if (length == shmfd->shm_size) {
VM_OBJECT_UNLOCK(object);
return (0);
}
nobjsize = OFF_TO_IDX(length + PAGE_MASK);
/* Are we shrinking? If so, trim the end. */
if (length < shmfd->shm_size) {
delta = ptoa(object->size - nobjsize);
/* Toss in memory pages. */
if (nobjsize < object->size)
vm_object_page_remove(object, nobjsize, object->size,
FALSE);
/* Toss pages from swap. */
if (object->type == OBJT_SWAP)
swap_pager_freespace(object, nobjsize, delta);
/* Free the swap accounted for shm */
swap_release_by_uid(delta, object->uip);
object->charge -= delta;
/*
* If the last page is partially mapped, then zero out
* the garbage at the end of the page. See comments
* in vnode_pager_setsize() for more details.
*
* XXXJHB: This handles in memory pages, but what about
* a page swapped out to disk?
*/
if ((length & PAGE_MASK) &&
(m = vm_page_lookup(object, OFF_TO_IDX(length))) != NULL &&
m->valid != 0) {
int base = (int)length & PAGE_MASK;
int size = PAGE_SIZE - base;
pmap_zero_page_area(m, base, size);
/*
* Update the valid bits to reflect the blocks that
* have been zeroed. Some of these valid bits may
* have already been set.
*/
vm_page_set_valid(m, base, size);
/*
* Round "base" to the next block boundary so that the
* dirty bit for a partially zeroed block is not
* cleared.
*/
base = roundup2(base, DEV_BSIZE);
vm_page_lock_queues();
vm_page_clear_dirty(m, base, PAGE_SIZE - base);
vm_page_unlock_queues();
} else if ((length & PAGE_MASK) &&
__predict_false(object->cache != NULL)) {
vm_page_cache_free(object, OFF_TO_IDX(length),
nobjsize);
}
} else {
/* Attempt to reserve the swap */
delta = ptoa(nobjsize - object->size);
if (!swap_reserve_by_uid(delta, object->uip)) {
VM_OBJECT_UNLOCK(object);
return (ENOMEM);
}
object->charge += delta;
}
shmfd->shm_size = length;
mtx_lock(&shm_timestamp_lock);
vfs_timestamp(&shmfd->shm_ctime);
shmfd->shm_mtime = shmfd->shm_ctime;
mtx_unlock(&shm_timestamp_lock);
object->size = nobjsize;
VM_OBJECT_UNLOCK(object);
return (0);
}
/* 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);
}
vm_object_t
cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
{
vm_object_t object, object1;
vm_pindex_t pindex;
u_short color;
if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
return (NULL);
/*
* Offset should be page aligned.
*/
if (foff & PAGE_MASK)
return (NULL);
size = round_page(size);
pindex = OFF_TO_IDX(foff + size);
if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
return (NULL);
mtx_lock(&dev_pager_mtx);
/*
* Look up pager, creating as necessary.
*/
object1 = NULL;
object = vm_pager_object_lookup(&dev_pager_object_list, handle);
if (object == NULL) {
/*
* Allocate object and associate it with the pager. Initialize
* the object's pg_color based upon the physical address of the
* device's memory.
*/
mtx_unlock(&dev_pager_mtx);
object1 = vm_object_allocate(tp, pindex);
object1->flags |= OBJ_COLORED;
object1->pg_color = color;
object1->handle = handle;
object1->un_pager.devp.ops = ops;
object1->un_pager.devp.dev = handle;
TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
mtx_lock(&dev_pager_mtx);
object = vm_pager_object_lookup(&dev_pager_object_list, handle);
if (object != NULL) {
/*
* We raced with other thread while allocating object.
*/
if (pindex > object->size)
object->size = pindex;
} else {
object = object1;
object1 = NULL;
object->handle = handle;
TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
pager_object_list);
KASSERT(object->type == tp,
("Inconsistent device pager type %p %d", object, tp));
}
} else {
if (pindex > object->size)
object->size = pindex;
}
mtx_unlock(&dev_pager_mtx);
if (object1 != NULL) {
object1->handle = object1;
mtx_lock(&dev_pager_mtx);
TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
pager_object_list);
mtx_unlock(&dev_pager_mtx);
vm_object_deallocate(object1);
}
return (object);
}
/*
* Truncate the inode ip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(vnode *vp, off_t length, int flags, Ucred *cred)
{
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
ufs2_daddr_t bn, lbn, lastblock, lastiblock[UFS_NIADDR];
ufs2_daddr_t indir_lbn[UFS_NIADDR], oldblks[UFS_NDADDR + UFS_NIADDR];
ufs2_daddr_t newblks[UFS_NDADDR + UFS_NIADDR];
ufs2_daddr_t count, blocksreleased = 0, datablocks, blkno;
struct bufobj *bo;
struct fs *fs;
struct buf *bp;
struct ufsmount *ump;
int softdeptrunc, journaltrunc;
int needextclean, extblocks;
int offset, size, level, nblocks;
int i, error, allerror, indiroff, waitforupdate;
off_t osize;
ip = VTOI(vp);
ump = VFSTOUFS(vp->v_mount);
fs = ump->um_fs;
bo = &vp->v_bufobj;
ASSERT_VOP_LOCKED(vp, "ffs_truncate");
if (length < 0)
return (EINVAL);
if (length > fs->fs_maxfilesize)
return (EFBIG);
#ifdef QUOTA
error = getinoquota(ip);
if (error)
return (error);
#endif
/*
* Historically clients did not have to specify which data
* they were truncating. So, if not specified, we assume
* traditional behavior, e.g., just the normal data.
*/
if ((flags & (IO_EXT | IO_NORMAL)) == 0)
flags |= IO_NORMAL;
if (!DOINGSOFTDEP(vp) && !DOINGASYNC(vp))
flags |= IO_SYNC;
waitforupdate = (flags & IO_SYNC) != 0 || !DOINGASYNC(vp);
/*
* If we are truncating the extended-attributes, and cannot
* do it with soft updates, then do it slowly here. If we are
* truncating both the extended attributes and the file contents
* (e.g., the file is being unlinked), then pick it off with
* soft updates below.
*/
allerror = 0;
needextclean = 0;
softdeptrunc = 0;
journaltrunc = DOINGSUJ(vp);
if (journaltrunc == 0 && DOINGSOFTDEP(vp) && length == 0)
softdeptrunc = !softdep_slowdown(vp);
extblocks = 0;
datablocks = DIP(ip, i_blocks);
if (fs->fs_magic == FS_UFS2_MAGIC && ip->i_din2->di_extsize > 0) {
extblocks = btodb(fragroundup(fs, ip->i_din2->di_extsize));
datablocks -= extblocks;
}
if ((flags & IO_EXT) && extblocks > 0) {
if (length != 0)
panic("ffs_truncate: partial trunc of extdata");
if (softdeptrunc || journaltrunc) {
if ((flags & IO_NORMAL) == 0)
goto extclean;
needextclean = 1;
} else {
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
#ifdef QUOTA
(void) chkdq(ip, -extblocks, NOCRED, 0);
#endif
vinvalbuf(vp, V_ALT, 0, 0);
vn_pages_remove(vp,
OFF_TO_IDX(lblktosize(fs, -extblocks)), 0);
osize = ip->i_din2->di_extsize;
ip->i_din2->di_blocks -= extblocks;
ip->i_din2->di_extsize = 0;
for (i = 0; i < UFS_NXADDR; i++) {
oldblks[i] = ip->i_din2->di_extb[i];
ip->i_din2->di_extb[i] = 0;
}
ip->i_flag |= IN_CHANGE;
if ((error = ffs_update(vp, waitforupdate)))
return (error);
for (i = 0; i < UFS_NXADDR; i++) {
if (oldblks[i] == 0)
continue;
ffs_blkfree(ump, fs, ITODEVVP(ip), oldblks[i],
sblksize(fs, osize, i), ip->i_number,
vp->v_type, nil);
}
}
}
if ((flags & IO_NORMAL) == 0)
return (0);
if (vp->v_type == VLNK &&
(ip->i_size < vp->v_mount->mnt_maxsymlinklen ||
datablocks == 0)) {
#ifdef INVARIANTS
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif
bzero(SHORTLINK(ip), (uint)ip->i_size);
ip->i_size = 0;
DIP_SET(ip, i_size, 0);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (needextclean)
goto extclean;
return (ffs_update(vp, waitforupdate));
}
if (ip->i_size == length) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (needextclean)
goto extclean;
return (ffs_update(vp, 0));
}
if (fs->fs_ronly)
panic("ffs_truncate: read-only filesystem");
if (IS_SNAPSHOT(ip))
ffs_snapremove(vp);
vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
osize = ip->i_size;
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
vnode_pager_setsize(vp, length);
flags |= BA_CLRBUF;
error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
ip->i_size = length;
DIP_SET(ip, i_size, length);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(vp))
bdwrite(bp);
else
bawrite(bp);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(vp, waitforupdate));
}
/*
* Lookup block number for a given offset. Zero length files
* have no blocks, so return a blkno of -1.
*/
lbn = lblkno(fs, length - 1);
if (length == 0) {
blkno = -1;
} else if (lbn < UFS_NDADDR) {
blkno = DIP(ip, i_db[lbn]);
} else {
error = UFS_BALLOC(vp, lblktosize(fs, (off_t)lbn), fs->fs_bsize,
cred, BA_METAONLY, &bp);
if (error)
return (error);
indiroff = (lbn - UFS_NDADDR) % NINDIR(fs);
if (I_IS_UFS1(ip))
blkno = ((ufs1_daddr_t *)(bp->b_data))[indiroff];
else
blkno = ((ufs2_daddr_t *)(bp->b_data))[indiroff];
/*
* If the block number is non-zero, then the indirect block
* must have been previously allocated and need not be written.
* If the block number is zero, then we may have allocated
* the indirect block and hence need to write it out.
*/
if (blkno != 0)
brelse(bp);
else if (flags & IO_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
/*
* If the block number at the new end of the file is zero,
* then we must allocate it to ensure that the last block of
* the file is allocated. Soft updates does not handle this
* case, so here we have to clean up the soft updates data
* structures describing the allocation past the truncation
* point. Finding and deallocating those structures is a lot of
* work. Since partial truncation with a hole at the end occurs
* rarely, we solve the problem by syncing the file so that it
* will have no soft updates data structures left.
*/
if (blkno == 0 && (error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
if (blkno != 0 && DOINGSOFTDEP(vp)) {
if (softdeptrunc == 0 && journaltrunc == 0) {
/*
* If soft updates cannot handle this truncation,
* clean up soft dependency data structures and
* fall through to the synchronous truncation.
*/
if ((error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
} else {
flags = IO_NORMAL | (needextclean ? IO_EXT: 0);
if (journaltrunc)
softdep_journal_freeblocks(ip, cred, length,
flags);
else
softdep_setup_freeblocks(ip, length, flags);
ASSERT_VOP_LOCKED(vp, "ffs_truncate1");
if (journaltrunc == 0) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
error = ffs_update(vp, 0);
}
return (error);
}
}
/*
* Shorten the size of the file. If the last block of the
* shortened file is unallocated, we must allocate it.
* Additionally, if the file is not being truncated to a
* block boundary, the contents of the partial block
* following the end of the file must be zero'ed in
* case it ever becomes accessible again because of
* subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = blkoff(fs, length);
if (blkno != 0 && offset == 0) {
ip->i_size = length;
DIP_SET(ip, i_size, length);
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = UFS_BALLOC(vp, length - 1, 1, cred, flags, &bp);
if (error)
return (error);
/*
* When we are doing soft updates and the UFS_BALLOC
* above fills in a direct block hole with a full sized
* block that will be truncated down to a fragment below,
* we must flush out the block dependency with an FSYNC
* so that we do not get a soft updates inconsistency
* when we create the fragment below.
*/
if (DOINGSOFTDEP(vp) && lbn < UFS_NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
(error = ffs_syncvnode(vp, MNT_WAIT, 0)) != 0)
return (error);
ip->i_size = length;
DIP_SET(ip, i_size, length);
size = blksize(fs, ip, lbn);
if (vp->v_type != VDIR && offset != 0)
bzero((char *)bp->b_data + offset,
(uint)(size - offset));
/* Kirk's code has reallocbuf(bp, size, 1) here */
allocbuf(bp, size);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(vp))
bdwrite(bp);
else
bawrite(bp);
}
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - UFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ffs_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[UFS_NDADDR + level] = DIP(ip, i_ib[level]);
if (lastiblock[level] < 0) {
DIP_SET(ip, i_ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < UFS_NDADDR; i++) {
oldblks[i] = DIP(ip, i_db[i]);
if (i > lastblock)
DIP_SET(ip, i_db[i], 0);
}
ip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ffs_update(vp, waitforupdate);
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < UFS_NDADDR; i++) {
newblks[i] = DIP(ip, i_db[i]);
DIP_SET(ip, i_db[i], oldblks[i]);
}
for (i = 0; i < UFS_NIADDR; i++) {
newblks[UFS_NDADDR + i] = DIP(ip, i_ib[i]);
DIP_SET(ip, i_ib[i], oldblks[UFS_NDADDR + i]);
}
ip->i_size = osize;
DIP_SET(ip, i_size, osize);
error = vtruncbuf(vp, cred, length, fs->fs_bsize);
if (error && (allerror == 0))
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -UFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = DIP(ip, i_ib[level]);
if (bn != 0) {
error = ffs_indirtrunc(ip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_SET(ip, i_ib[level], 0);
ffs_blkfree(ump, fs, ump->um_devvp, bn,
fs->fs_bsize, ip->i_number,
vp->v_type, nil);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = UFS_NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = DIP(ip, i_db[i]);
if (bn == 0)
continue;
DIP_SET(ip, i_db[i], 0);
bsize = blksize(fs, ip, i);
ffs_blkfree(ump, fs, ump->um_devvp, bn, bsize, ip->i_number,
vp->v_type, nil);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = DIP(ip, i_db[lastblock]);
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, ip, lastblock);
ip->i_size = length;
DIP_SET(ip, i_size, length);
newspace = blksize(fs, ip, lastblock);
if (newspace == 0)
panic("ffs_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ffs_blkfree(ump, fs, ump->um_devvp, bn,
oldspace - newspace, ip->i_number, vp->v_type, nil);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[UFS_NDADDR + level] != DIP(ip, i_ib[level]))
panic("ffs_truncate1");
for (i = 0; i < UFS_NDADDR; i++)
if (newblks[i] != DIP(ip, i_db[i]))
panic("ffs_truncate2");
BO_LOCK(bo);
if (length == 0 &&
(fs->fs_magic != FS_UFS2_MAGIC || ip->i_din2->di_extsize == 0) &&
(bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
panic("ffs_truncate3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
ip->i_size = length;
DIP_SET(ip, i_size, length);
if (DIP(ip, i_blocks) >= blocksreleased)
DIP_SET(ip, i_blocks, DIP(ip, i_blocks) - blocksreleased);
else /* sanity */
DIP_SET(ip, i_blocks, 0);
ip->i_flag |= IN_CHANGE;
#ifdef QUOTA
(void) chkdq(ip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
extclean:
if (journaltrunc)
softdep_journal_freeblocks(ip, cred, length, IO_EXT);
else
softdep_setup_freeblocks(ip, length, IO_EXT);
return (ffs_update(vp, waitforupdate));
#endif // 0
return 0;
}
static int
ttm_bo_vm_fault(vm_object_t vm_obj, vm_ooffset_t offset,
int prot, vm_page_t *mres)
{
struct ttm_buffer_object *bo = vm_obj->handle;
struct ttm_bo_device *bdev = bo->bdev;
struct ttm_tt *ttm = NULL;
vm_page_t m, m1, oldm;
int ret;
int retval = VM_PAGER_OK;
struct ttm_mem_type_manager *man =
&bdev->man[bo->mem.mem_type];
vm_object_pip_add(vm_obj, 1);
oldm = *mres;
if (oldm != NULL) {
vm_page_lock(oldm);
vm_page_remove(oldm);
vm_page_unlock(oldm);
*mres = NULL;
} else
oldm = NULL;
retry:
VM_OBJECT_WUNLOCK(vm_obj);
m = NULL;
reserve:
ret = ttm_bo_reserve(bo, false, false, false, 0);
if (unlikely(ret != 0)) {
if (ret == -EBUSY) {
kern_yield(0);
goto reserve;
}
}
if (bdev->driver->fault_reserve_notify) {
ret = bdev->driver->fault_reserve_notify(bo);
switch (ret) {
case 0:
break;
case -EBUSY:
case -ERESTART:
case -EINTR:
kern_yield(0);
goto reserve;
default:
retval = VM_PAGER_ERROR;
goto out_unlock;
}
}
/*
* Wait for buffer data in transit, due to a pipelined
* move.
*/
mtx_lock(&bdev->fence_lock);
if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) {
/*
* Here, the behavior differs between Linux and FreeBSD.
*
* On Linux, the wait is interruptible (3rd argument to
* ttm_bo_wait). There must be some mechanism to resume
* page fault handling, once the signal is processed.
*
* On FreeBSD, the wait is uninteruptible. This is not a
* problem as we can't end up with an unkillable process
* here, because the wait will eventually time out.
*
* An example of this situation is the Xorg process
* which uses SIGALRM internally. The signal could
* interrupt the wait, causing the page fault to fail
* and the process to receive SIGSEGV.
*/
ret = ttm_bo_wait(bo, false, false, false);
mtx_unlock(&bdev->fence_lock);
if (unlikely(ret != 0)) {
retval = VM_PAGER_ERROR;
goto out_unlock;
}
} else
mtx_unlock(&bdev->fence_lock);
ret = ttm_mem_io_lock(man, true);
if (unlikely(ret != 0)) {
retval = VM_PAGER_ERROR;
goto out_unlock;
}
ret = ttm_mem_io_reserve_vm(bo);
if (unlikely(ret != 0)) {
retval = VM_PAGER_ERROR;
goto out_io_unlock;
}
/*
* Strictly, we're not allowed to modify vma->vm_page_prot here,
* since the mmap_sem is only held in read mode. However, we
* modify only the caching bits of vma->vm_page_prot and
* consider those bits protected by
* the bo->mutex, as we should be the only writers.
* There shouldn't really be any readers of these bits except
* within vm_insert_mixed()? fork?
*
* TODO: Add a list of vmas to the bo, and change the
* vma->vm_page_prot when the object changes caching policy, with
* the correct locks held.
*/
if (!bo->mem.bus.is_iomem) {
/* Allocate all page at once, most common usage */
ttm = bo->ttm;
if (ttm->bdev->driver->ttm_tt_populate(ttm)) {
retval = VM_PAGER_ERROR;
goto out_io_unlock;
}
}
if (bo->mem.bus.is_iomem) {
m = PHYS_TO_VM_PAGE(bo->mem.bus.base + bo->mem.bus.offset +
offset);
KASSERT((m->flags & PG_FICTITIOUS) != 0,
("physical address %#jx not fictitious",
(uintmax_t)(bo->mem.bus.base + bo->mem.bus.offset
+ offset)));
pmap_page_set_memattr(m, ttm_io_prot(bo->mem.placement));
} else {
ttm = bo->ttm;
m = ttm->pages[OFF_TO_IDX(offset)];
if (unlikely(!m)) {
retval = VM_PAGER_ERROR;
goto out_io_unlock;
}
pmap_page_set_memattr(m,
(bo->mem.placement & TTM_PL_FLAG_CACHED) ?
VM_MEMATTR_WRITE_BACK : ttm_io_prot(bo->mem.placement));
}
VM_OBJECT_WLOCK(vm_obj);
if (vm_page_busied(m)) {
vm_page_lock(m);
VM_OBJECT_WUNLOCK(vm_obj);
vm_page_busy_sleep(m, "ttmpbs");
VM_OBJECT_WLOCK(vm_obj);
ttm_mem_io_unlock(man);
ttm_bo_unreserve(bo);
goto retry;
}
m1 = vm_page_lookup(vm_obj, OFF_TO_IDX(offset));
if (m1 == NULL) {
if (vm_page_insert(m, vm_obj, OFF_TO_IDX(offset))) {
VM_OBJECT_WUNLOCK(vm_obj);
VM_WAIT;
VM_OBJECT_WLOCK(vm_obj);
ttm_mem_io_unlock(man);
ttm_bo_unreserve(bo);
goto retry;
}
} else {
KASSERT(m == m1,
("inconsistent insert bo %p m %p m1 %p offset %jx",
bo, m, m1, (uintmax_t)offset));
}
m->valid = VM_PAGE_BITS_ALL;
*mres = m;
vm_page_xbusy(m);
if (oldm != NULL) {
vm_page_lock(oldm);
vm_page_free(oldm);
vm_page_unlock(oldm);
}
out_io_unlock1:
ttm_mem_io_unlock(man);
out_unlock1:
ttm_bo_unreserve(bo);
vm_object_pip_wakeup(vm_obj);
return (retval);
out_io_unlock:
VM_OBJECT_WLOCK(vm_obj);
goto out_io_unlock1;
out_unlock:
VM_OBJECT_WLOCK(vm_obj);
goto out_unlock1;
}
/*
* mincore system call handler
*
* mincore_args(const void *addr, size_t len, char *vec)
*
* No requirements
*/
int
sys_mincore(struct mincore_args *uap)
{
struct proc *p = curproc;
vm_offset_t addr, first_addr;
vm_offset_t end, cend;
pmap_t pmap;
vm_map_t map;
char *vec;
int error;
int vecindex, lastvecindex;
vm_map_entry_t current;
vm_map_entry_t entry;
int mincoreinfo;
unsigned int timestamp;
/*
* Make sure that the addresses presented are valid for user
* mode.
*/
first_addr = addr = trunc_page((vm_offset_t) uap->addr);
end = addr + (vm_size_t)round_page(uap->len);
if (end < addr)
return (EINVAL);
if (VM_MAX_USER_ADDRESS > 0 && end > VM_MAX_USER_ADDRESS)
return (EINVAL);
/*
* Address of byte vector
*/
vec = uap->vec;
map = &p->p_vmspace->vm_map;
pmap = vmspace_pmap(p->p_vmspace);
lwkt_gettoken(&map->token);
vm_map_lock_read(map);
RestartScan:
timestamp = map->timestamp;
if (!vm_map_lookup_entry(map, addr, &entry))
entry = entry->next;
/*
* Do this on a map entry basis so that if the pages are not
* in the current processes address space, we can easily look
* up the pages elsewhere.
*/
lastvecindex = -1;
for(current = entry;
(current != &map->header) && (current->start < end);
current = current->next) {
/*
* ignore submaps (for now) or null objects
*/
if (current->maptype != VM_MAPTYPE_NORMAL &&
current->maptype != VM_MAPTYPE_VPAGETABLE) {
continue;
}
if (current->object.vm_object == NULL)
continue;
/*
* limit this scan to the current map entry and the
* limits for the mincore call
*/
if (addr < current->start)
addr = current->start;
cend = current->end;
if (cend > end)
cend = end;
/*
* scan this entry one page at a time
*/
while (addr < cend) {
/*
* Check pmap first, it is likely faster, also
* it can provide info as to whether we are the
* one referencing or modifying the page.
*
* If we have to check the VM object, only mess
* around with normal maps. Do not mess around
* with virtual page tables (XXX).
*/
mincoreinfo = pmap_mincore(pmap, addr);
if (mincoreinfo == 0 &&
current->maptype == VM_MAPTYPE_NORMAL) {
vm_pindex_t pindex;
vm_ooffset_t offset;
vm_page_t m;
/*
* calculate the page index into the object
*/
offset = current->offset + (addr - current->start);
pindex = OFF_TO_IDX(offset);
/*
* if the page is resident, then gather
* information about it. spl protection is
* required to maintain the object
* association. And XXX what if the page is
* busy? What's the deal with that?
*
* XXX vm_token - legacy for pmap_ts_referenced
* in i386 and vkernel pmap code.
*/
lwkt_gettoken(&vm_token);
vm_object_hold(current->object.vm_object);
m = vm_page_lookup(current->object.vm_object,
pindex);
if (m && m->valid) {
mincoreinfo = MINCORE_INCORE;
if (m->dirty ||
pmap_is_modified(m))
mincoreinfo |= MINCORE_MODIFIED_OTHER;
if ((m->flags & PG_REFERENCED) ||
pmap_ts_referenced(m)) {
vm_page_flag_set(m, PG_REFERENCED);
mincoreinfo |= MINCORE_REFERENCED_OTHER;
}
}
vm_object_drop(current->object.vm_object);
lwkt_reltoken(&vm_token);
}
/*
* subyte may page fault. In case it needs to modify
* the map, we release the lock.
*/
vm_map_unlock_read(map);
/*
* calculate index into user supplied byte vector
*/
vecindex = OFF_TO_IDX(addr - first_addr);
/*
* If we have skipped map entries, we need to make sure that
* the byte vector is zeroed for those skipped entries.
*/
while((lastvecindex + 1) < vecindex) {
error = subyte( vec + lastvecindex, 0);
if (error) {
error = EFAULT;
goto done;
}
++lastvecindex;
}
/*
* Pass the page information to the user
*/
error = subyte( vec + vecindex, mincoreinfo);
if (error) {
error = EFAULT;
goto done;
}
/*
* If the map has changed, due to the subyte, the previous
* output may be invalid.
*/
vm_map_lock_read(map);
if (timestamp != map->timestamp)
goto RestartScan;
lastvecindex = vecindex;
addr += PAGE_SIZE;
}
}
/*
* subyte may page fault. In case it needs to modify
* the map, we release the lock.
*/
vm_map_unlock_read(map);
/*
* Zero the last entries in the byte vector.
*/
vecindex = OFF_TO_IDX(end - first_addr);
while((lastvecindex + 1) < vecindex) {
error = subyte( vec + lastvecindex, 0);
if (error) {
error = EFAULT;
goto done;
}
++lastvecindex;
}
/*
* If the map has changed, due to the subyte, the previous
* output may be invalid.
*/
vm_map_lock_read(map);
if (timestamp != map->timestamp)
goto RestartScan;
vm_map_unlock_read(map);
error = 0;
done:
lwkt_reltoken(&map->token);
return (error);
}
/*
* Lets the VM system know about a change in size for a file.
* We adjust our own internal size and flush any cached pages in
* the associated object that are affected by the size change.
*
* Note: this routine may be invoked as a result of a pager put
* operation (possibly at object termination time), so we must be careful.
*/
void
vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize)
{
vm_object_t object;
vm_page_t m;
vm_pindex_t nobjsize;
if ((object = vp->v_object) == NULL)
return;
/* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */
VM_OBJECT_WLOCK(object);
if (object->type == OBJT_DEAD) {
VM_OBJECT_WUNLOCK(object);
return;
}
KASSERT(object->type == OBJT_VNODE,
("not vnode-backed object %p", object));
if (nsize == object->un_pager.vnp.vnp_size) {
/*
* Hasn't changed size
*/
VM_OBJECT_WUNLOCK(object);
return;
}
nobjsize = OFF_TO_IDX(nsize + PAGE_MASK);
if (nsize < object->un_pager.vnp.vnp_size) {
/*
* File has shrunk. Toss any cached pages beyond the new EOF.
*/
if (nobjsize < object->size)
vm_object_page_remove(object, nobjsize, object->size,
0);
/*
* this gets rid of garbage at the end of a page that is now
* only partially backed by the vnode.
*
* XXX for some reason (I don't know yet), if we take a
* completely invalid page and mark it partially valid
* it can screw up NFS reads, so we don't allow the case.
*/
if ((nsize & PAGE_MASK) &&
(m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL &&
m->valid != 0) {
int base = (int)nsize & PAGE_MASK;
int size = PAGE_SIZE - base;
/*
* Clear out partial-page garbage in case
* the page has been mapped.
*/
pmap_zero_page_area(m, base, size);
/*
* Update the valid bits to reflect the blocks that
* have been zeroed. Some of these valid bits may
* have already been set.
*/
vm_page_set_valid_range(m, base, size);
/*
* Round "base" to the next block boundary so that the
* dirty bit for a partially zeroed block is not
* cleared.
*/
base = roundup2(base, DEV_BSIZE);
/*
* Clear out partial-page dirty bits.
*
* note that we do not clear out the valid
* bits. This would prevent bogus_page
* replacement from working properly.
*/
vm_page_clear_dirty(m, base, PAGE_SIZE - base);
}
}
object->un_pager.vnp.vnp_size = nsize;
object->size = nobjsize;
VM_OBJECT_WUNLOCK(object);
}
/*
* 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);
}
}
vm_object_t
cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
{
cdev_t dev;
vm_object_t object;
u_short color;
/*
* Offset should be page aligned.
*/
if (foff & PAGE_MASK)
return (NULL);
size = round_page64(size);
if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
return (NULL);
/*
* Look up pager, creating as necessary.
*/
mtx_lock(&dev_pager_mtx);
object = vm_pager_object_lookup(&dev_pager_object_list, handle);
if (object == NULL) {
/*
* Allocate object and associate it with the pager.
*/
object = vm_object_allocate_hold(tp,
OFF_TO_IDX(foff + size));
object->handle = handle;
object->un_pager.devp.ops = ops;
object->un_pager.devp.dev = handle;
TAILQ_INIT(&object->un_pager.devp.devp_pglist);
/*
* handle is only a device for old_dev_pager_ctor.
*/
if (ops->cdev_pg_ctor == old_dev_pager_ctor) {
dev = handle;
dev->si_object = object;
}
TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
pager_object_list);
vm_object_drop(object);
} else {
/*
* Gain a reference to the object.
*/
vm_object_hold(object);
vm_object_reference_locked(object);
if (OFF_TO_IDX(foff + size) > object->size)
object->size = OFF_TO_IDX(foff + size);
vm_object_drop(object);
}
mtx_unlock(&dev_pager_mtx);
return (object);
}
static int
shm_dotruncate(struct shmfd *shmfd, off_t length)
{
vm_object_t object;
vm_page_t m, ma[1];
vm_pindex_t idx, nobjsize;
vm_ooffset_t delta;
int base, rv;
object = shmfd->shm_object;
VM_OBJECT_LOCK(object);
if (length == shmfd->shm_size) {
VM_OBJECT_UNLOCK(object);
return (0);
}
nobjsize = OFF_TO_IDX(length + PAGE_MASK);
/* Are we shrinking? If so, trim the end. */
if (length < shmfd->shm_size) {
/*
* Disallow any requests to shrink the size if this
* object is mapped into the kernel.
*/
if (shmfd->shm_kmappings > 0) {
VM_OBJECT_UNLOCK(object);
return (EBUSY);
}
/*
* Zero the truncated part of the last page.
*/
base = length & PAGE_MASK;
if (base != 0) {
idx = OFF_TO_IDX(length);
retry:
m = vm_page_lookup(object, idx);
if (m != NULL) {
if ((m->oflags & VPO_BUSY) != 0 ||
m->busy != 0) {
vm_page_sleep(m, "shmtrc");
goto retry;
}
} else if (vm_pager_has_page(object, idx, NULL, NULL)) {
m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL);
if (m == NULL) {
VM_OBJECT_UNLOCK(object);
VM_WAIT;
VM_OBJECT_LOCK(object);
goto retry;
} else if (m->valid != VM_PAGE_BITS_ALL) {
ma[0] = m;
rv = vm_pager_get_pages(object, ma, 1,
0);
m = vm_page_lookup(object, idx);
} else
/* A cached page was reactivated. */
rv = VM_PAGER_OK;
vm_page_lock(m);
if (rv == VM_PAGER_OK) {
vm_page_deactivate(m);
vm_page_unlock(m);
vm_page_wakeup(m);
} else {
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_UNLOCK(object);
return (EIO);
}
}
if (m != NULL) {
pmap_zero_page_area(m, base, PAGE_SIZE - base);
KASSERT(m->valid == VM_PAGE_BITS_ALL,
("shm_dotruncate: page %p is invalid", m));
vm_page_dirty(m);
vm_pager_page_unswapped(m);
}
}
delta = ptoa(object->size - nobjsize);
/* Toss in memory pages. */
if (nobjsize < object->size)
vm_object_page_remove(object, nobjsize, object->size,
0);
/* Toss pages from swap. */
if (object->type == OBJT_SWAP)
swap_pager_freespace(object, nobjsize, delta);
/* Free the swap accounted for shm */
swap_release_by_cred(delta, object->cred);
object->charge -= delta;
} else {
/* Attempt to reserve the swap */
delta = ptoa(nobjsize - object->size);
if (!swap_reserve_by_cred(delta, object->cred)) {
VM_OBJECT_UNLOCK(object);
return (ENOMEM);
}
object->charge += delta;
}
shmfd->shm_size = length;
mtx_lock(&shm_timestamp_lock);
vfs_timestamp(&shmfd->shm_ctime);
shmfd->shm_mtime = shmfd->shm_ctime;
mtx_unlock(&shm_timestamp_lock);
object->size = nobjsize;
VM_OBJECT_UNLOCK(object);
return (0);
}
static int
tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio)
{
vm_pindex_t idx;
vm_page_t vpg, tpg;
vm_offset_t offset;
off_t addr;
size_t tlen;
int error, rv;
error = 0;
addr = uio->uio_offset;
idx = OFF_TO_IDX(addr);
offset = addr & PAGE_MASK;
tlen = MIN(PAGE_SIZE - offset, len);
if ((vobj == NULL) ||
(vobj->resident_page_count == 0 && vobj->cache == NULL)) {
vpg = NULL;
goto nocache;
}
VM_OBJECT_LOCK(vobj);
lookupvpg:
if (((vpg = vm_page_lookup(vobj, idx)) != NULL) &&
vm_page_is_valid(vpg, offset, tlen)) {
if ((vpg->oflags & VPO_BUSY) != 0) {
/*
* Reference the page before unlocking and sleeping so
* that the page daemon is less likely to reclaim it.
*/
vm_page_reference(vpg);
vm_page_sleep(vpg, "tmfsmw");
goto lookupvpg;
}
vm_page_busy(vpg);
vm_page_undirty(vpg);
VM_OBJECT_UNLOCK(vobj);
error = uiomove_fromphys(&vpg, offset, tlen, uio);
} else {
if (__predict_false(vobj->cache != NULL))
vm_page_cache_free(vobj, idx, idx + 1);
VM_OBJECT_UNLOCK(vobj);
vpg = NULL;
}
nocache:
VM_OBJECT_LOCK(tobj);
tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (tpg->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(tobj, &tpg, 1, 0);
if (rv != VM_PAGER_OK) {
vm_page_lock(tpg);
vm_page_free(tpg);
vm_page_unlock(tpg);
error = EIO;
goto out;
}
} else
vm_page_zero_invalid(tpg, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
if (vpg == NULL)
error = uiomove_fromphys(&tpg, offset, tlen, uio);
else {
KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid"));
pmap_copy_page(vpg, tpg);
}
VM_OBJECT_LOCK(tobj);
if (error == 0) {
KASSERT(tpg->valid == VM_PAGE_BITS_ALL,
("parts of tpg invalid"));
vm_page_dirty(tpg);
}
vm_page_lock(tpg);
vm_page_unwire(tpg, TRUE);
vm_page_unlock(tpg);
vm_page_wakeup(tpg);
out:
VM_OBJECT_UNLOCK(tobj);
if (vpg != NULL) {
VM_OBJECT_LOCK(vobj);
vm_page_wakeup(vpg);
VM_OBJECT_UNLOCK(vobj);
}
return (error);
}
