/* --------------------------------------------------------------------- */
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_WLOCK(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_WUNLOCK(tobj);
return (EIO);
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_WUNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_WLOCK(tobj);
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
VM_OBJECT_WUNLOCK(tobj);
return (error);
}
/*
* vm_fault_prefault provides a quick way of clustering
* pagefaults into a processes address space. It is a "cousin"
* of vm_map_pmap_enter, except it runs at page fault time instead
* of mmap time.
*/
static void
vm_fault_prefault(pmap_t pmap, vm_offset_t addra, vm_map_entry_t entry)
{
int i;
vm_offset_t addr, starta;
vm_pindex_t pindex;
vm_page_t m;
vm_object_t object;
if (pmap != vmspace_pmap(curthread->td_proc->p_vmspace))
return;
object = entry->object.vm_object;
starta = addra - PFBAK * PAGE_SIZE;
if (starta < entry->start) {
starta = entry->start;
} else if (starta > addra) {
starta = 0;
}
for (i = 0; i < PAGEORDER_SIZE; i++) {
vm_object_t backing_object, lobject;
addr = addra + prefault_pageorder[i];
if (addr > addra + (PFFOR * PAGE_SIZE))
addr = 0;
if (addr < starta || addr >= entry->end)
continue;
if (!pmap_is_prefaultable(pmap, addr))
continue;
pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
lobject = object;
VM_OBJECT_WLOCK(lobject);
while ((m = vm_page_lookup(lobject, pindex)) == NULL &&
lobject->type == OBJT_DEFAULT &&
(backing_object = lobject->backing_object) != NULL) {
KASSERT((lobject->backing_object_offset & PAGE_MASK) ==
0, ("vm_fault_prefault: unaligned object offset"));
pindex += lobject->backing_object_offset >> PAGE_SHIFT;
VM_OBJECT_WLOCK(backing_object);
VM_OBJECT_WUNLOCK(lobject);
lobject = backing_object;
}
/*
* give-up when a page is not in memory
*/
if (m == NULL) {
VM_OBJECT_WUNLOCK(lobject);
break;
}
if (m->valid == VM_PAGE_BITS_ALL &&
(m->flags & PG_FICTITIOUS) == 0)
pmap_enter_quick(pmap, addr, m, entry->protection);
VM_OBJECT_WUNLOCK(lobject);
}
}
/*
* Allocates a region from the kernel address map and physically
* contiguous pages within the specified address range to the kernel
* object. Creates a wired mapping from this region to these pages, and
* returns the region's starting virtual address. If M_ZERO is specified
* through the given flags, then the pages are zeroed before they are
* mapped.
*/
vm_offset_t
kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr)
{
vmem_t *vmem;
vm_object_t object = kernel_object;
vm_offset_t addr, offset, tmp;
vm_page_t end_m, m;
u_long npages;
int pflags, tries;
size = round_page(size);
vmem = vm_dom[domain].vmd_kernel_arena;
if (vmem_alloc(vmem, size, flags | M_BESTFIT, &addr))
return (0);
offset = addr - VM_MIN_KERNEL_ADDRESS;
pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
pflags |= VM_ALLOC_NOWAIT;
npages = atop(size);
VM_OBJECT_WLOCK(object);
tries = 0;
retry:
m = vm_page_alloc_contig_domain(object, atop(offset), domain, pflags,
npages, low, high, alignment, boundary, memattr);
if (m == NULL) {
VM_OBJECT_WUNLOCK(object);
if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
if (!vm_page_reclaim_contig_domain(domain, pflags,
npages, low, high, alignment, boundary) &&
(flags & M_WAITOK) != 0)
vm_wait_domain(domain);
VM_OBJECT_WLOCK(object);
tries++;
goto retry;
}
vmem_free(vmem, addr, size);
return (0);
}
KASSERT(vm_phys_domain(m) == domain,
("kmem_alloc_contig_domain: Domain mismatch %d != %d",
vm_phys_domain(m), domain));
end_m = m + npages;
tmp = addr;
for (; m < end_m; m++) {
if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
m->valid = VM_PAGE_BITS_ALL;
pmap_enter(kernel_pmap, tmp, m, VM_PROT_ALL,
VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
tmp += PAGE_SIZE;
}
VM_OBJECT_WUNLOCK(object);
return (addr);
}
/*
* Helper routines to allow the backing object of a shared memory file
* descriptor to be mapped in the kernel.
*/
int
shm_map(struct file *fp, size_t size, off_t offset, void **memp)
{
struct shmfd *shmfd;
vm_offset_t kva, ofs;
vm_object_t obj;
int rv;
if (fp->f_type != DTYPE_SHM)
return (EINVAL);
shmfd = fp->f_data;
obj = shmfd->shm_object;
VM_OBJECT_WLOCK(obj);
/*
* XXXRW: This validation is probably insufficient, and subject to
* sign errors. It should be fixed.
*/
if (offset >= shmfd->shm_size ||
offset + size > round_page(shmfd->shm_size)) {
VM_OBJECT_WUNLOCK(obj);
return (EINVAL);
}
shmfd->shm_kmappings++;
vm_object_reference_locked(obj);
VM_OBJECT_WUNLOCK(obj);
/* Map the object into the kernel_map and wire it. */
kva = vm_map_min(kernel_map);
ofs = offset & PAGE_MASK;
offset = trunc_page(offset);
size = round_page(size + ofs);
rv = vm_map_find(kernel_map, obj, offset, &kva, size, 0,
VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE,
VM_PROT_READ | VM_PROT_WRITE, 0);
if (rv == KERN_SUCCESS) {
rv = vm_map_wire(kernel_map, kva, kva + size,
VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
if (rv == KERN_SUCCESS) {
*memp = (void *)(kva + ofs);
return (0);
}
vm_map_remove(kernel_map, kva, kva + size);
} else
vm_object_deallocate(obj);
/* On failure, drop our mapping reference. */
VM_OBJECT_WLOCK(obj);
shmfd->shm_kmappings--;
VM_OBJECT_WUNLOCK(obj);
return (vm_mmap_to_errno(rv));
}
/*
* Allocates a region from the kernel address map and physical pages
* within the specified address range to the kernel object. Creates a
* wired mapping from this region to these pages, and returns the
* region's starting virtual address. The allocated pages are not
* necessarily physically contiguous. If M_ZERO is specified through the
* given flags, then the pages are zeroed before they are mapped.
*/
vm_offset_t
kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
vm_paddr_t high, vm_memattr_t memattr)
{
vmem_t *vmem;
vm_object_t object = kernel_object;
vm_offset_t addr, i, offset;
vm_page_t m;
int pflags, tries;
size = round_page(size);
vmem = vm_dom[domain].vmd_kernel_arena;
if (vmem_alloc(vmem, size, M_BESTFIT | flags, &addr))
return (0);
offset = addr - VM_MIN_KERNEL_ADDRESS;
pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
pflags |= VM_ALLOC_NOWAIT;
VM_OBJECT_WLOCK(object);
for (i = 0; i < size; i += PAGE_SIZE) {
tries = 0;
retry:
m = vm_page_alloc_contig_domain(object, atop(offset + i),
domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr);
if (m == NULL) {
VM_OBJECT_WUNLOCK(object);
if (tries < ((flags & M_NOWAIT) != 0 ? 1 : 3)) {
if (!vm_page_reclaim_contig_domain(domain,
pflags, 1, low, high, PAGE_SIZE, 0) &&
(flags & M_WAITOK) != 0)
vm_wait_domain(domain);
VM_OBJECT_WLOCK(object);
tries++;
goto retry;
}
kmem_unback(object, addr, i);
vmem_free(vmem, addr, size);
return (0);
}
KASSERT(vm_phys_domain(m) == domain,
("kmem_alloc_attr_domain: Domain mismatch %d != %d",
vm_phys_domain(m), domain));
if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
m->valid = VM_PAGE_BITS_ALL;
pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
}
VM_OBJECT_WUNLOCK(object);
return (addr);
}
/*
* 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);
}
/*
* The object must be locked.
*/
static void
vnode_pager_dealloc(vm_object_t object)
{
struct vnode *vp;
int refs;
vp = object->handle;
if (vp == NULL)
panic("vnode_pager_dealloc: pager already dealloced");
VM_OBJECT_ASSERT_WLOCKED(object);
vm_object_pip_wait(object, "vnpdea");
refs = object->ref_count;
object->handle = NULL;
object->type = OBJT_DEAD;
if (object->flags & OBJ_DISCONNECTWNT) {
vm_object_clear_flag(object, OBJ_DISCONNECTWNT);
wakeup(object);
}
ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc");
if (object->un_pager.vnp.writemappings > 0) {
object->un_pager.vnp.writemappings = 0;
VOP_ADD_WRITECOUNT(vp, -1);
CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
__func__, vp, vp->v_writecount);
}
vp->v_object = NULL;
VOP_UNSET_TEXT(vp);
VM_OBJECT_WUNLOCK(object);
while (refs-- > 0)
vunref(vp);
VM_OBJECT_WLOCK(object);
}
/*
* kmem_unback:
*
* Unmap and free the physical pages underlying the specified virtual
* address range.
*
* A physical page must exist within the specified object at each index
* that is being unmapped.
*/
static int
_kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
{
vm_page_t m, next;
vm_offset_t end, offset;
int domain;
KASSERT(object == kernel_object,
("kmem_unback: only supports kernel object."));
if (size == 0)
return (0);
pmap_remove(kernel_pmap, addr, addr + size);
offset = addr - VM_MIN_KERNEL_ADDRESS;
end = offset + size;
VM_OBJECT_WLOCK(object);
m = vm_page_lookup(object, atop(offset));
domain = vm_phys_domain(m);
for (; offset < end; offset += PAGE_SIZE, m = next) {
next = vm_page_next(m);
vm_page_unwire(m, PQ_NONE);
vm_page_free(m);
}
VM_OBJECT_WUNLOCK(object);
return (domain);
}
void
vnode_destroy_vobject(struct vnode *vp)
{
struct vm_object *obj;
obj = vp->v_object;
if (obj == NULL)
return;
ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject");
VM_OBJECT_WLOCK(obj);
if (obj->ref_count == 0) {
/*
* don't double-terminate the object
*/
if ((obj->flags & OBJ_DEAD) == 0)
vm_object_terminate(obj);
else
VM_OBJECT_WUNLOCK(obj);
} else {
/*
* Woe to the process that tries to page now :-).
*/
vm_pager_deallocate(obj);
VM_OBJECT_WUNLOCK(obj);
}
vp->v_object = NULL;
}
/*
* shmfd object management including creation and reference counting
* routines.
*/
static struct shmfd *
shm_alloc(struct ucred *ucred, mode_t mode)
{
struct shmfd *shmfd;
shmfd = malloc(sizeof(*shmfd), M_SHMFD, M_WAITOK | M_ZERO);
shmfd->shm_size = 0;
shmfd->shm_uid = ucred->cr_uid;
shmfd->shm_gid = ucred->cr_gid;
shmfd->shm_mode = mode;
shmfd->shm_object = vm_pager_allocate(OBJT_DEFAULT, NULL,
shmfd->shm_size, VM_PROT_DEFAULT, 0, ucred);
KASSERT(shmfd->shm_object != NULL, ("shm_create: vm_pager_allocate"));
VM_OBJECT_WLOCK(shmfd->shm_object);
vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
VM_OBJECT_WUNLOCK(shmfd->shm_object);
vfs_timestamp(&shmfd->shm_birthtime);
shmfd->shm_atime = shmfd->shm_mtime = shmfd->shm_ctime =
shmfd->shm_birthtime;
refcount_init(&shmfd->shm_refs, 1);
mtx_init(&shmfd->shm_mtx, "shmrl", NULL, MTX_DEF);
rangelock_init(&shmfd->shm_rl);
#ifdef MAC
mac_posixshm_init(shmfd);
mac_posixshm_create(ucred, shmfd);
#endif
return (shmfd);
}
void
vnode_pager_update_writecount(vm_object_t object, vm_offset_t start,
vm_offset_t end)
{
struct vnode *vp;
vm_ooffset_t old_wm;
VM_OBJECT_WLOCK(object);
if (object->type != OBJT_VNODE) {
VM_OBJECT_WUNLOCK(object);
return;
}
old_wm = object->un_pager.vnp.writemappings;
object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start;
vp = object->handle;
if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) {
ASSERT_VOP_ELOCKED(vp, "v_writecount inc");
VOP_ADD_WRITECOUNT(vp, 1);
CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
__func__, vp, vp->v_writecount);
} else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) {
ASSERT_VOP_ELOCKED(vp, "v_writecount dec");
VOP_ADD_WRITECOUNT(vp, -1);
CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
__func__, vp, vp->v_writecount);
}
VM_OBJECT_WUNLOCK(object);
}
/*
* old style vnode pager input routine
*/
static int
vnode_pager_input_old(vm_object_t object, vm_page_t m)
{
struct uio auio;
struct iovec aiov;
int error;
int size;
struct sf_buf *sf;
struct vnode *vp;
VM_OBJECT_ASSERT_WLOCKED(object);
error = 0;
/*
* Return failure if beyond current EOF
*/
if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) {
return VM_PAGER_BAD;
} else {
size = PAGE_SIZE;
if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size)
size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex);
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
/*
* Allocate a kernel virtual address and initialize so that
* we can use VOP_READ/WRITE routines.
*/
sf = sf_buf_alloc(m, 0);
aiov.iov_base = (caddr_t)sf_buf_kva(sf);
aiov.iov_len = size;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = IDX_TO_OFF(m->pindex);
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_READ;
auio.uio_resid = size;
auio.uio_td = curthread;
error = VOP_READ(vp, &auio, 0, curthread->td_ucred);
if (!error) {
int count = size - auio.uio_resid;
if (count == 0)
error = EINVAL;
else if (count != PAGE_SIZE)
bzero((caddr_t)sf_buf_kva(sf) + count,
PAGE_SIZE - count);
}
sf_buf_free(sf);
VM_OBJECT_WLOCK(object);
}
KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m));
if (!error)
m->valid = VM_PAGE_BITS_ALL;
return error ? VM_PAGER_ERROR : VM_PAGER_OK;
}
/*
* EOPNOTSUPP is no longer legal. For local media VFS's that do not
* implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to
* vnode_pager_generic_putpages() to implement the previous behaviour.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_PUTPAGES.
*/
static void
vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count,
int flags, int *rtvals)
{
int rtval;
struct vnode *vp;
int bytes = count * PAGE_SIZE;
/*
* Force synchronous operation if we are extremely low on memory
* to prevent a low-memory deadlock. VOP operations often need to
* allocate more memory to initiate the I/O ( i.e. do a BMAP
* operation ). The swapper handles the case by limiting the amount
* of asynchronous I/O, but that sort of solution doesn't scale well
* for the vnode pager without a lot of work.
*
* Also, the backing vnode's iodone routine may not wake the pageout
* daemon up. This should be probably be addressed XXX.
*/
if (vm_cnt.v_free_count < vm_cnt.v_pageout_free_min)
flags |= VM_PAGER_PUT_SYNC;
/*
* Call device-specific putpages function
*/
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
rtval = VOP_PUTPAGES(vp, m, bytes, flags, rtvals);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: stale FS putpages\n"));
VM_OBJECT_WLOCK(object);
}
void
ttm_bo_release_mmap(struct ttm_buffer_object *bo)
{
vm_object_t vm_obj;
vm_page_t m;
int i;
vm_obj = cdev_pager_lookup(bo);
if (vm_obj == NULL)
return;
VM_OBJECT_WLOCK(vm_obj);
retry:
for (i = 0; i < bo->num_pages; i++) {
m = vm_page_lookup(vm_obj, i);
if (m == NULL)
continue;
if (vm_page_sleep_if_busy(m, "ttm_unm"))
goto retry;
cdev_pager_free_page(vm_obj, m);
}
VM_OBJECT_WUNLOCK(vm_obj);
vm_object_deallocate(vm_obj);
}
/* Create the VM system backing object for this vnode */
int
vnode_create_vobject(struct vnode *vp, off_t isize, struct thread *td)
{
vm_object_t object;
vm_ooffset_t size = isize;
struct vattr va;
if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE)
return (0);
while ((object = vp->v_object) != NULL) {
VM_OBJECT_WLOCK(object);
if (!(object->flags & OBJ_DEAD)) {
VM_OBJECT_WUNLOCK(object);
return (0);
}
VOP_UNLOCK(vp, 0);
vm_object_set_flag(object, OBJ_DISCONNECTWNT);
VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vodead", 0);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
}
if (size == 0) {
if (vn_isdisk(vp, NULL)) {
size = IDX_TO_OFF(INT_MAX);
} else {
if (VOP_GETATTR(vp, &va, td->td_ucred))
return (0);
size = va.va_size;
}
}
object = vnode_pager_alloc(vp, size, 0, 0, td->td_ucred);
/*
* Dereference the reference we just created. This assumes
* that the object is associated with the vp.
*/
VM_OBJECT_WLOCK(object);
object->ref_count--;
VM_OBJECT_WUNLOCK(object);
vrele(vp);
KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object"));
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_destroy(object);
goto retry;
}
vp->v_object = object;
VI_UNLOCK(vp);
} else {
object->ref_count++;
VM_OBJECT_WUNLOCK(object);
}
vref(vp);
return (object);
}
/*
* Flush and invalidate all dirty buffers. If another process is already
* doing the flush, just wait for completion.
*/
int
fuse_io_invalbuf(struct vnode *vp, struct thread *td)
{
struct fuse_vnode_data *fvdat = VTOFUD(vp);
int error = 0;
if (vp->v_iflag & VI_DOOMED)
return 0;
ASSERT_VOP_ELOCKED(vp, "fuse_io_invalbuf");
while (fvdat->flag & FN_FLUSHINPROG) {
struct proc *p = td->td_proc;
if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)
return EIO;
fvdat->flag |= FN_FLUSHWANT;
tsleep(&fvdat->flag, PRIBIO + 2, "fusevinv", 2 * hz);
error = 0;
if (p != NULL) {
PROC_LOCK(p);
if (SIGNOTEMPTY(p->p_siglist) ||
SIGNOTEMPTY(td->td_siglist))
error = EINTR;
PROC_UNLOCK(p);
}
if (error == EINTR)
return EINTR;
}
fvdat->flag |= FN_FLUSHINPROG;
if (vp->v_bufobj.bo_object != NULL) {
VM_OBJECT_WLOCK(vp->v_bufobj.bo_object);
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_WUNLOCK(vp->v_bufobj.bo_object);
}
error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
while (error) {
if (error == ERESTART || error == EINTR) {
fvdat->flag &= ~FN_FLUSHINPROG;
if (fvdat->flag & FN_FLUSHWANT) {
fvdat->flag &= ~FN_FLUSHWANT;
wakeup(&fvdat->flag);
}
return EINTR;
}
error = vinvalbuf(vp, V_SAVE, PCATCH, 0);
}
fvdat->flag &= ~FN_FLUSHINPROG;
if (fvdat->flag & FN_FLUSHWANT) {
fvdat->flag &= ~FN_FLUSHWANT;
wakeup(&fvdat->flag);
}
return (error);
}
/*
* kmem_back:
*
* Allocate physical pages for the specified virtual address range.
*/
int
kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
vm_size_t size, int flags)
{
vm_offset_t offset, i;
vm_page_t m, mpred;
int pflags;
KASSERT(object == kernel_object,
("kmem_back_domain: only supports kernel object."));
offset = addr - VM_MIN_KERNEL_ADDRESS;
pflags = malloc2vm_flags(flags) | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED;
pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
if (flags & M_WAITOK)
pflags |= VM_ALLOC_WAITFAIL;
i = 0;
VM_OBJECT_WLOCK(object);
retry:
mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
for (; i < size; i += PAGE_SIZE, mpred = m) {
m = vm_page_alloc_domain_after(object, atop(offset + i),
domain, pflags, mpred);
/*
* Ran out of space, free everything up and return. Don't need
* to lock page queues here as we know that the pages we got
* aren't on any queues.
*/
if (m == NULL) {
if ((flags & M_NOWAIT) == 0)
goto retry;
VM_OBJECT_WUNLOCK(object);
kmem_unback(object, addr, i);
return (KERN_NO_SPACE);
}
KASSERT(vm_phys_domain(m) == domain,
("kmem_back_domain: Domain mismatch %d != %d",
vm_phys_domain(m), domain));
if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
KASSERT((m->oflags & VPO_UNMANAGED) != 0,
("kmem_malloc: page %p is managed", m));
m->valid = VM_PAGE_BITS_ALL;
pmap_enter(kernel_pmap, addr + i, m, VM_PROT_ALL,
VM_PROT_ALL | PMAP_ENTER_WIRED, 0);
}
VM_OBJECT_WUNLOCK(object);
return (KERN_SUCCESS);
}
static int rtR0MemObjFreeBSDAllocPhysPages(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
size_t cb,
RTHCPHYS PhysHighest, size_t uAlignment,
bool fContiguous, int rcNoMem)
{
uint32_t cPages = atop(cb);
vm_paddr_t VmPhysAddrHigh;
/* create the object. */
PRTR0MEMOBJFREEBSD pMemFreeBSD = (PRTR0MEMOBJFREEBSD)rtR0MemObjNew(sizeof(*pMemFreeBSD),
enmType, NULL, cb);
if (!pMemFreeBSD)
return VERR_NO_MEMORY;
pMemFreeBSD->pObject = vm_object_allocate(OBJT_PHYS, atop(cb));
if (PhysHighest != NIL_RTHCPHYS)
VmPhysAddrHigh = PhysHighest;
else
VmPhysAddrHigh = ~(vm_paddr_t)0;
int rc = rtR0MemObjFreeBSDPhysAllocHelper(pMemFreeBSD->pObject, cPages, VmPhysAddrHigh,
uAlignment, fContiguous, true, rcNoMem);
if (RT_SUCCESS(rc))
{
if (fContiguous)
{
Assert(enmType == RTR0MEMOBJTYPE_PHYS);
#if __FreeBSD_version >= 1000030
VM_OBJECT_WLOCK(pMemFreeBSD->pObject);
#else
VM_OBJECT_LOCK(pMemFreeBSD->pObject);
#endif
pMemFreeBSD->Core.u.Phys.PhysBase = VM_PAGE_TO_PHYS(vm_page_find_least(pMemFreeBSD->pObject, 0));
#if __FreeBSD_version >= 1000030
VM_OBJECT_WUNLOCK(pMemFreeBSD->pObject);
#else
VM_OBJECT_UNLOCK(pMemFreeBSD->pObject);
#endif
pMemFreeBSD->Core.u.Phys.fAllocated = true;
}
*ppMem = &pMemFreeBSD->Core;
}
else
{
vm_object_deallocate(pMemFreeBSD->pObject);
rtR0MemObjDelete(&pMemFreeBSD->Core);
}
return rc;
}
int
ncl_inactive(struct vop_inactive_args *ap)
{
struct nfsnode *np;
struct sillyrename *sp;
struct vnode *vp = ap->a_vp;
boolean_t retv;
np = VTONFS(vp);
if (NFS_ISV4(vp) && vp->v_type == VREG) {
/*
* Since mmap()'d files do I/O after VOP_CLOSE(), the NFSv4
* Close operations are delayed until now. Any dirty
* buffers/pages must be flushed before the close, so that the
* stateid is available for the writes.
*/
if (vp->v_object != NULL) {
VM_OBJECT_WLOCK(vp->v_object);
retv = vm_object_page_clean(vp->v_object, 0, 0,
OBJPC_SYNC);
VM_OBJECT_WUNLOCK(vp->v_object);
} else
retv = TRUE;
if (retv == TRUE) {
(void)ncl_flush(vp, MNT_WAIT, NULL, ap->a_td, 1, 0);
(void)nfsrpc_close(vp, 1, ap->a_td);
}
}
mtx_lock(&np->n_mtx);
if (vp->v_type != VDIR) {
sp = np->n_sillyrename;
np->n_sillyrename = NULL;
} else
sp = NULL;
if (sp) {
mtx_unlock(&np->n_mtx);
(void) ncl_vinvalbuf(vp, 0, ap->a_td, 1);
/*
* Remove the silly file that was rename'd earlier
*/
ncl_removeit(sp, vp);
crfree(sp->s_cred);
TASK_INIT(&sp->s_task, 0, nfs_freesillyrename, sp);
taskqueue_enqueue(taskqueue_thread, &sp->s_task);
mtx_lock(&np->n_mtx);
}
np->n_flag &= NMODIFIED;
mtx_unlock(&np->n_mtx);
return (0);
}
vm_object_t
vmm_mmio_alloc(struct vmspace *vmspace, vm_paddr_t gpa, size_t len,
vm_paddr_t hpa)
{
int error;
vm_object_t obj;
struct sglist *sg;
sg = sglist_alloc(1, M_WAITOK);
error = sglist_append_phys(sg, hpa, len);
KASSERT(error == 0, ("error %d appending physaddr to sglist", error));
obj = vm_pager_allocate(OBJT_SG, sg, len, VM_PROT_RW, 0, NULL);
if (obj != NULL) {
/*
* VT-x ignores the MTRR settings when figuring out the
* memory type for translations obtained through EPT.
*
* Therefore we explicitly force the pages provided by
* this object to be mapped as uncacheable.
*/
VM_OBJECT_WLOCK(obj);
error = vm_object_set_memattr(obj, VM_MEMATTR_UNCACHEABLE);
VM_OBJECT_WUNLOCK(obj);
if (error != KERN_SUCCESS) {
panic("vmm_mmio_alloc: vm_object_set_memattr error %d",
error);
}
error = vm_map_find(&vmspace->vm_map, obj, 0, &gpa, len, 0,
VMFS_NO_SPACE, VM_PROT_RW, VM_PROT_RW, 0);
if (error != KERN_SUCCESS) {
vm_object_deallocate(obj);
obj = NULL;
}
}
/*
* Drop the reference on the sglist.
*
* If the scatter/gather object was successfully allocated then it
* has incremented the reference count on the sglist. Dropping the
* initial reference count ensures that the sglist will be freed
* when the object is deallocated.
*
* If the object could not be allocated then we end up freeing the
* sglist.
*/
sglist_free(sg);
return (obj);
}
/*
* MPSAFE
*/
static void
phys_pager_dealloc(vm_object_t object)
{
if (object->handle != NULL) {
VM_OBJECT_WUNLOCK(object);
mtx_lock(&phys_pager_mtx);
TAILQ_REMOVE(&phys_pager_object_list, object, pager_object_list);
mtx_unlock(&phys_pager_mtx);
VM_OBJECT_WLOCK(object);
}
object->handle = NULL;
object->type = OBJT_DEAD;
}
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);
}
void
vnode_pager_release_writecount(vm_object_t object, vm_offset_t start,
vm_offset_t end)
{
struct vnode *vp;
struct mount *mp;
vm_offset_t inc;
VM_OBJECT_WLOCK(object);
/*
* First, recheck the object type to account for the race when
* the vnode is reclaimed.
*/
if (object->type != OBJT_VNODE) {
VM_OBJECT_WUNLOCK(object);
return;
}
/*
* Optimize for the case when writemappings is not going to
* zero.
*/
inc = end - start;
if (object->un_pager.vnp.writemappings != inc) {
object->un_pager.vnp.writemappings -= inc;
VM_OBJECT_WUNLOCK(object);
return;
}
vp = object->handle;
vhold(vp);
VM_OBJECT_WUNLOCK(object);
mp = NULL;
vn_start_write(vp, &mp, V_WAIT);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
/*
* Decrement the object's writemappings, by swapping the start
* and end arguments for vnode_pager_update_writecount(). If
* there was not a race with vnode reclaimation, then the
* vnode's v_writecount is decremented.
*/
vnode_pager_update_writecount(object, end, start);
VOP_UNLOCK(vp, 0);
vdrop(vp);
if (mp != NULL)
vn_finished_write(mp);
}
/*
* Local media VFS's that do not implement their own VOP_GETPAGES
* should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
* to implement the previous behaviour.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_GETPAGES.
*/
static int
vnode_pager_getpages(vm_object_t object, vm_page_t *m, int count, int reqpage)
{
int rtval;
struct vnode *vp;
int bytes = count * PAGE_SIZE;
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
rtval = VOP_GETPAGES(vp, m, bytes, reqpage);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: FS getpages not implemented\n"));
VM_OBJECT_WLOCK(object);
return rtval;
}
static int
vnode_pager_getpages_async(vm_object_t object, vm_page_t *m, int count,
int *rbehind, int *rahead, vop_getpages_iodone_t iodone, void *arg)
{
struct vnode *vp;
int rtval;
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
rtval = VOP_GETPAGES_ASYNC(vp, m, count, rbehind, rahead, iodone, arg);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: FS getpages_async not implemented\n"));
VM_OBJECT_WLOCK(object);
return (rtval);
}
/*
* Local media VFS's that do not implement their own VOP_GETPAGES
* should have their VOP_GETPAGES call to vnode_pager_generic_getpages()
* to implement the previous behaviour.
*
* All other FS's should use the bypass to get to the local media
* backing vp's VOP_GETPAGES.
*/
static int
vnode_pager_getpages(vm_object_t object, vm_page_t *m, int count, int *rbehind,
int *rahead)
{
struct vnode *vp;
int rtval;
vp = object->handle;
VM_OBJECT_WUNLOCK(object);
rtval = VOP_GETPAGES(vp, m, count, rbehind, rahead);
KASSERT(rtval != EOPNOTSUPP,
("vnode_pager: FS getpages not implemented\n"));
VM_OBJECT_WLOCK(object);
return rtval;
}
static int rtR0MemObjFreeBSDPhysAllocHelper(vm_object_t pObject, u_long cPages,
vm_paddr_t VmPhysAddrHigh, u_long uAlignment,
bool fContiguous, bool fWire, int rcNoMem)
{
if (fContiguous)
{
if (rtR0MemObjFreeBSDContigPhysAllocHelper(pObject, 0, cPages, VmPhysAddrHigh,
uAlignment, fWire) != NULL)
return VINF_SUCCESS;
return rcNoMem;
}
for (vm_pindex_t iPage = 0; iPage < cPages; iPage++)
{
vm_page_t pPage = rtR0MemObjFreeBSDContigPhysAllocHelper(pObject, iPage, 1, VmPhysAddrHigh,
uAlignment, fWire);
if (!pPage)
{
/* Free all allocated pages */
#if __FreeBSD_version >= 1000030
VM_OBJECT_WLOCK(pObject);
#else
VM_OBJECT_LOCK(pObject);
#endif
while (iPage-- > 0)
{
pPage = vm_page_lookup(pObject, iPage);
#if __FreeBSD_version < 1000000
vm_page_lock_queues();
#endif
if (fWire)
vm_page_unwire(pPage, 0);
vm_page_free(pPage);
#if __FreeBSD_version < 1000000
vm_page_unlock_queues();
#endif
}
#if __FreeBSD_version >= 1000030
VM_OBJECT_WUNLOCK(pObject);
#else
VM_OBJECT_UNLOCK(pObject);
#endif
return rcNoMem;
}
}
return VINF_SUCCESS;
}
static int
spigen_mmap_single(struct cdev *cdev, vm_ooffset_t *offset,
vm_size_t size, struct vm_object **object, int nprot)
{
device_t dev = cdev->si_drv1;
struct spigen_softc *sc = device_get_softc(dev);
vm_page_t *m;
size_t n, pages;
if (size == 0 ||
(nprot & (PROT_EXEC | PROT_READ | PROT_WRITE))
!= (PROT_READ | PROT_WRITE))
return (EINVAL);
size = roundup2(size, PAGE_SIZE);
pages = size / PAGE_SIZE;
mtx_lock(&sc->sc_mtx);
if (sc->sc_mmap_buffer != NULL) {
mtx_unlock(&sc->sc_mtx);
return (EBUSY);
} else if (size > sc->sc_command_length_max + sc->sc_data_length_max) {
mtx_unlock(&sc->sc_mtx);
return (E2BIG);
}
sc->sc_mmap_buffer_size = size;
*offset = 0;
sc->sc_mmap_buffer = *object = vm_pager_allocate(OBJT_PHYS, 0, size,
nprot, *offset, curthread->td_ucred);
m = malloc(sizeof(*m) * pages, M_TEMP, M_WAITOK);
VM_OBJECT_WLOCK(*object);
vm_object_reference_locked(*object); // kernel and userland both
for (n = 0; n < pages; n++) {
m[n] = vm_page_grab(*object, n,
VM_ALLOC_NOBUSY | VM_ALLOC_ZERO | VM_ALLOC_WIRED);
m[n]->valid = VM_PAGE_BITS_ALL;
}
VM_OBJECT_WUNLOCK(*object);
sc->sc_mmap_kvaddr = kva_alloc(size);
pmap_qenter(sc->sc_mmap_kvaddr, m, pages);
free(m, M_TEMP);
mtx_unlock(&sc->sc_mtx);
if (*object == NULL)
return (EINVAL);
return (0);
}
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);
}
