/* --------------------------------------------------------------------- */
static int
tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
vm_offset_t offset, size_t tlen, struct uio *uio)
{
vm_page_t m;
int error, rv;
VM_OBJECT_LOCK(tobj);
m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
rv = vm_pager_get_pages(tobj, &m, 1, 0);
if (rv != VM_PAGER_OK) {
vm_page_lock(m);
vm_page_free(m);
vm_page_unlock(m);
VM_OBJECT_UNLOCK(tobj);
return (EIO);
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(tobj);
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
/* --------------------------------------------------------------------- */
static int
tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx,
vm_offset_t offset, size_t tlen, struct uio *uio)
{
vm_page_t m;
int error;
VM_OBJECT_LOCK(tobj);
vm_object_pip_add(tobj, 1);
m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED |
VM_ALLOC_ZERO | VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
if (m->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(tobj, idx, NULL, NULL)) {
error = vm_pager_get_pages(tobj, &m, 1, 0);
if (error != 0) {
printf("tmpfs get pages from pager error [read]\n");
goto out;
}
} else
vm_page_zero_invalid(m, TRUE);
}
VM_OBJECT_UNLOCK(tobj);
error = uiomove_fromphys(&m, offset, tlen, uio);
VM_OBJECT_LOCK(tobj);
out:
vm_page_lock(m);
vm_page_unwire(m, TRUE);
vm_page_unlock(m);
vm_page_wakeup(m);
vm_object_pip_subtract(tobj, 1);
VM_OBJECT_UNLOCK(tobj);
return (error);
}
/*
* 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_LOCK(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_LOCK(backing_object);
VM_OBJECT_UNLOCK(lobject);
lobject = backing_object;
}
/*
* give-up when a page is not in memory
*/
if (m == NULL) {
VM_OBJECT_UNLOCK(lobject);
break;
}
if (m->valid == VM_PAGE_BITS_ALL &&
(m->flags & PG_FICTITIOUS) == 0)
pmap_enter_quick(pmap, addr, m, entry->protection);
VM_OBJECT_UNLOCK(lobject);
}
}
/*
* 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_LOCK(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_UNLOCK(obj);
return (EINVAL);
}
shmfd->shm_kmappings++;
vm_object_reference_locked(obj);
VM_OBJECT_UNLOCK(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,
VMFS_ALIGNED_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_LOCK(obj);
shmfd->shm_kmappings--;
VM_OBJECT_UNLOCK(obj);
return (vm_mmap_to_errno(rv));
}
static vm_page_t rtR0MemObjFreeBSDContigPhysAllocHelper(vm_object_t pObject, vm_pindex_t iPIndex,
u_long cPages, vm_paddr_t VmPhysAddrHigh,
u_long uAlignment, bool fWire)
{
vm_page_t pPages;
int cTries = 0;
#if __FreeBSD_version > 1000000
int fFlags = VM_ALLOC_INTERRUPT | VM_ALLOC_NOBUSY;
if (fWire)
fFlags |= VM_ALLOC_WIRED;
while (cTries <= 1)
{
VM_OBJECT_LOCK(pObject);
pPages = vm_page_alloc_contig(pObject, iPIndex, fFlags, cPages, 0,
VmPhysAddrHigh, uAlignment, 0, VM_MEMATTR_DEFAULT);
VM_OBJECT_UNLOCK(pObject);
if (pPages)
break;
vm_pageout_grow_cache(cTries, 0, VmPhysAddrHigh);
cTries++;
}
return pPages;
#else
while (cTries <= 1)
{
pPages = vm_phys_alloc_contig(cPages, 0, VmPhysAddrHigh, uAlignment, 0);
if (pPages)
break;
vm_contig_grow_cache(cTries, 0, VmPhysAddrHigh);
cTries++;
}
if (!pPages)
return pPages;
VM_OBJECT_LOCK(pObject);
for (vm_pindex_t iPage = 0; iPage < cPages; iPage++)
{
vm_page_t pPage = pPages + iPage;
vm_page_insert(pPage, pObject, iPIndex + iPage);
pPage->valid = VM_PAGE_BITS_ALL;
if (fWire)
{
pPage->wire_count = 1;
atomic_add_int(&cnt.v_wire_count, 1);
}
}
VM_OBJECT_UNLOCK(pObject);
return pPages;
#endif
}
/*
* Speed up the reclamation of up to "distance" pages that precede the
* faulting pindex within the first object of the shadow chain.
*/
static void
vm_fault_cache_behind(const struct faultstate *fs, int distance)
{
vm_object_t first_object, object;
vm_page_t m, m_prev;
vm_pindex_t pindex;
object = fs->object;
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
first_object = fs->first_object;
if (first_object != object) {
if (!VM_OBJECT_TRYLOCK(first_object)) {
VM_OBJECT_UNLOCK(object);
VM_OBJECT_LOCK(first_object);
VM_OBJECT_LOCK(object);
}
}
if (first_object->type != OBJT_DEVICE &&
first_object->type != OBJT_PHYS && first_object->type != OBJT_SG) {
if (fs->first_pindex < distance)
pindex = 0;
else
pindex = fs->first_pindex - distance;
if (pindex < OFF_TO_IDX(fs->entry->offset))
pindex = OFF_TO_IDX(fs->entry->offset);
m = first_object != object ? fs->first_m : fs->m;
KASSERT((m->oflags & VPO_BUSY) != 0,
("vm_fault_cache_behind: page %p is not busy", m));
m_prev = vm_page_prev(m);
while ((m = m_prev) != NULL && m->pindex >= pindex &&
m->valid == VM_PAGE_BITS_ALL) {
m_prev = vm_page_prev(m);
if (m->busy != 0 || (m->oflags & VPO_BUSY) != 0)
continue;
vm_page_lock(m);
if (m->hold_count == 0 && m->wire_count == 0) {
pmap_remove_all(m);
vm_page_aflag_clear(m, PGA_REFERENCED);
if (m->dirty != 0)
vm_page_deactivate(m);
else
vm_page_cache(m);
}
vm_page_unlock(m);
}
}
if (first_object != object)
VM_OBJECT_UNLOCK(first_object);
}
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_LOCK(obj);
if (obj->ref_count == 0) {
/*
* vclean() may be called twice. The first time
* removes the primary reference to the object,
* the second time goes one further and is a
* special-case to terminate the object.
*
* don't double-terminate the object
*/
if ((obj->flags & OBJ_DEAD) == 0)
vm_object_terminate(obj);
else
VM_OBJECT_UNLOCK(obj);
} else {
/*
* Woe to the process that tries to page now :-).
*/
vm_pager_deallocate(obj);
VM_OBJECT_UNLOCK(obj);
}
vp->v_object = NULL;
}
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 */
VM_OBJECT_LOCK(pObject);
while (iPage-- > 0)
{
pPage = vm_page_lookup(pObject, iPage);
vm_page_lock_queues();
if (fWire)
vm_page_unwire(pPage, 0);
vm_page_free(pPage);
vm_page_unlock_queues();
}
VM_OBJECT_UNLOCK(pObject);
return rcNoMem;
}
}
return VINF_SUCCESS;
}
/*
* 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_LOCK(shmfd->shm_object);
vm_object_clear_flag(shmfd->shm_object, OBJ_ONEMAPPING);
vm_object_set_flag(shmfd->shm_object, OBJ_NOSPLIT);
VM_OBJECT_UNLOCK(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);
#ifdef MAC
mac_posixshm_init(shmfd);
mac_posixshm_create(ucred, shmfd);
#endif
return (shmfd);
}
/*
* 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_LOCK(object);
if ((object->flags & OBJ_DEAD) == 0)
break;
vm_object_set_flag(object, OBJ_DISCONNECTWNT);
msleep(object, VM_OBJECT_MTX(object), PDROP | PVM, "vadead", 0);
}
if (vp->v_usecount == 0)
panic("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_UNLOCK(object);
}
vref(vp);
return (object);
}
/* 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_LOCK(object);
if (!(object->flags & OBJ_DEAD)) {
VM_OBJECT_UNLOCK(object);
return (0);
}
VOP_UNLOCK(vp, 0);
vm_object_set_flag(object, OBJ_DISCONNECTWNT);
msleep(object, VM_OBJECT_MTX(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_LOCK(object);
object->ref_count--;
VM_OBJECT_UNLOCK(object);
vrele(vp);
KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object"));
return (0);
}
/* Try to invalidate pages, for "fs flush" or "fs flushv"; or
* try to free pages, when deleting a file.
*
* Locking: the vcache entry's lock is held. It may be dropped and
* re-obtained.
*
* Since we drop and re-obtain the lock, we can't guarantee that there won't
* be some pages around when we return, newly created by concurrent activity.
*/
void
osi_VM_TryToSmush(struct vcache *avc, afs_ucred_t *acred, int sync)
{
struct vnode *vp;
int tries, code;
int islocked;
vp = AFSTOV(avc);
VI_LOCK(vp);
if (vp->v_iflag & VI_DOOMED) {
VI_UNLOCK(vp);
return;
}
VI_UNLOCK(vp);
islocked = islocked_vnode(vp);
if (islocked == LK_EXCLOTHER)
panic("Trying to Smush over someone else's lock");
else if (islocked == LK_SHARED) {
afs_warn("Trying to Smush with a shared lock");
lock_vnode(vp, LK_UPGRADE);
} else if (!islocked)
lock_vnode(vp, LK_EXCLUSIVE);
if (vp->v_bufobj.bo_object != NULL) {
VM_OBJECT_LOCK(vp->v_bufobj.bo_object);
/*
* Do we really want OBJPC_SYNC? OBJPC_INVAL would be
* faster, if invalidation is really what we are being
* asked to do. (It would make more sense, too, since
* otherwise this function is practically identical to
* osi_VM_StoreAllSegments().) -GAW
*/
/*
* Dunno. We no longer resemble osi_VM_StoreAllSegments,
* though maybe that's wrong, now. And OBJPC_SYNC is the
* common thing in 70 file systems, it seems. Matt.
*/
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_UNLOCK(vp->v_bufobj.bo_object);
}
tries = 5;
code = osi_vinvalbuf(vp, V_SAVE, PCATCH, 0);
while (code && (tries > 0)) {
afs_warn("TryToSmush retrying vinvalbuf");
code = osi_vinvalbuf(vp, V_SAVE, PCATCH, 0);
--tries;
}
if (islocked == LK_SHARED)
lock_vnode(vp, LK_DOWNGRADE);
else if (!islocked)
unlock_vnode(vp);
}
/*
* 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_LOCK(vp->v_bufobj.bo_object);
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_UNLOCK(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);
}
/*
* MPSAFE
*/
static void
phys_pager_dealloc(vm_object_t object)
{
if (object->handle != NULL) {
VM_OBJECT_UNLOCK(object);
mtx_lock(&phys_pager_mtx);
TAILQ_REMOVE(&phys_pager_object_list, object, pager_object_list);
mtx_unlock(&phys_pager_mtx);
VM_OBJECT_LOCK(object);
}
}
/*
* Identify the physical page mapped at the given kernel virtual
* address. Insert this physical page into the given address space at
* the given virtual address, replacing the physical page, if any,
* that already exists there.
*/
static int
vm_pgmoveco(vm_map_t mapa, vm_offset_t kaddr, vm_offset_t uaddr)
{
vm_map_t map = mapa;
vm_page_t kern_pg, user_pg;
vm_object_t uobject;
vm_map_entry_t entry;
vm_pindex_t upindex;
vm_prot_t prot;
boolean_t wired;
KASSERT((uaddr & PAGE_MASK) == 0,
("vm_pgmoveco: uaddr is not page aligned"));
/*
* Herein the physical page is validated and dirtied. It is
* unwired in sf_buf_mext().
*/
kern_pg = PHYS_TO_VM_PAGE(vtophys(kaddr));
kern_pg->valid = VM_PAGE_BITS_ALL;
KASSERT(kern_pg->queue == PQ_NONE && kern_pg->wire_count == 1,
("vm_pgmoveco: kern_pg is not correctly wired"));
if ((vm_map_lookup(&map, uaddr,
VM_PROT_WRITE, &entry, &uobject,
&upindex, &prot, &wired)) != KERN_SUCCESS) {
return(EFAULT);
}
VM_OBJECT_LOCK(uobject);
retry:
if ((user_pg = vm_page_lookup(uobject, upindex)) != NULL) {
if (vm_page_sleep_if_busy(user_pg, TRUE, "vm_pgmoveco"))
goto retry;
vm_page_lock_queues();
pmap_remove_all(user_pg);
vm_page_free(user_pg);
} else {
/*
* Even if a physical page does not exist in the
* object chain's first object, a physical page from a
* backing object may be mapped read only.
*/
if (uobject->backing_object != NULL)
pmap_remove(map->pmap, uaddr, uaddr + PAGE_SIZE);
vm_page_lock_queues();
}
vm_page_insert(kern_pg, uobject, upindex);
vm_page_dirty(kern_pg);
vm_page_unlock_queues();
VM_OBJECT_UNLOCK(uobject);
vm_map_lookup_done(map, entry);
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_LOCK(vp->v_object);
retv = vm_object_page_clean(vp->v_object, 0, 0,
OBJPC_SYNC);
VM_OBJECT_UNLOCK(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);
}
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_LOCK(obj);
vm_object_pip_add(obj, 1);
for (i = 0; i < ttm->num_pages; ++i) {
from_page = vm_page_grab(obj, i, VM_ALLOC_NORMAL |
VM_ALLOC_RETRY);
if (from_page->valid != VM_PAGE_BITS_ALL) {
if (vm_pager_has_page(obj, i)) {
rv = vm_pager_get_page(obj, &from_page, 1);
if (rv != VM_PAGER_OK) {
vm_page_free(from_page);
ret = -EIO;
goto err_ret;
}
} else {
vm_page_zero_invalid(from_page, TRUE);
}
}
to_page = ttm->pages[i];
if (unlikely(to_page == NULL)) {
ret = -ENOMEM;
vm_page_wakeup(from_page);
goto err_ret;
}
pmap_copy_page(VM_PAGE_TO_PHYS(from_page),
VM_PAGE_TO_PHYS(to_page));
vm_page_wakeup(from_page);
}
vm_object_pip_wakeup(obj);
VM_OBJECT_UNLOCK(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_UNLOCK(obj);
return (ret);
}
/* Purge VM for a file when its callback is revoked.
*
* Locking: No lock is held, not even the global lock.
*/
void
osi_VM_FlushPages(struct vcache *avc, afs_ucred_t *credp)
{
struct vnode *vp;
struct vm_object *obj;
vp = AFSTOV(avc);
ASSERT_VOP_LOCKED(vp, __func__);
if (VOP_GETVOBJECT(vp, &obj) == 0) {
VM_OBJECT_LOCK(obj);
vm_object_page_remove(obj, 0, 0, FALSE);
VM_OBJECT_UNLOCK(obj);
}
osi_vinvalbuf(vp, 0, 0, 0);
}
int ttm_tt_swapout(struct ttm_tt *ttm, vm_object_t persistent_swap_storage)
{
vm_object_t obj;
vm_page_t from_page, to_page;
int i;
BUG_ON(ttm->state != tt_unbound && ttm->state != tt_unpopulated);
BUG_ON(ttm->caching_state != tt_cached);
if (!persistent_swap_storage) {
obj = swap_pager_alloc(NULL,
IDX_TO_OFF(ttm->num_pages), VM_PROT_DEFAULT, 0);
if (obj == NULL) {
pr_err("Failed allocating swap storage\n");
return (-ENOMEM);
}
} else
obj = persistent_swap_storage;
VM_OBJECT_LOCK(obj);
vm_object_pip_add(obj, 1);
for (i = 0; i < ttm->num_pages; ++i) {
from_page = ttm->pages[i];
if (unlikely(from_page == NULL))
continue;
to_page = vm_page_grab(obj, i, VM_ALLOC_NORMAL |
VM_ALLOC_RETRY);
pmap_copy_page(VM_PAGE_TO_PHYS(from_page),
VM_PAGE_TO_PHYS(to_page));
to_page->valid = VM_PAGE_BITS_ALL;
vm_page_dirty(to_page);
vm_page_wakeup(to_page);
}
vm_object_pip_wakeup(obj);
VM_OBJECT_UNLOCK(obj);
ttm->bdev->driver->ttm_tt_unpopulate(ttm);
ttm->swap_storage = obj;
ttm->page_flags |= TTM_PAGE_FLAG_SWAPPED;
if (persistent_swap_storage)
ttm->page_flags |= TTM_PAGE_FLAG_PERSISTENT_SWAP;
return 0;
}
/* Try to invalidate pages, for "fs flush" or "fs flushv"; or
* try to free pages, when deleting a file.
*
* Locking: the vcache entry's lock is held. It may be dropped and
* re-obtained.
*
* Since we drop and re-obtain the lock, we can't guarantee that there won't
* be some pages around when we return, newly created by concurrent activity.
*/
void
osi_VM_TryToSmush(struct vcache *avc, afs_ucred_t *acred, int sync)
{
struct vnode *vp;
int tries, code;
SPLVAR;
vp = AFSTOV(avc);
if (vp->v_iflag & VI_DOOMED) {
USERPRI;
return;
}
if (vp->v_bufobj.bo_object != NULL) {
VM_OBJECT_LOCK(vp->v_bufobj.bo_object);
/*
* Do we really want OBJPC_SYNC? OBJPC_INVAL would be
* faster, if invalidation is really what we are being
* asked to do. (It would make more sense, too, since
* otherwise this function is practically identical to
* osi_VM_StoreAllSegments().) -GAW
*/
/*
* Dunno. We no longer resemble osi_VM_StoreAllSegments,
* though maybe that's wrong, now. And OBJPC_SYNC is the
* common thing in 70 file systems, it seems. Matt.
*/
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
VM_OBJECT_UNLOCK(vp->v_bufobj.bo_object);
}
tries = 5;
code = osi_vinvalbuf(vp, V_SAVE, PCATCH, 0);
while (code && (tries > 0)) {
code = osi_vinvalbuf(vp, V_SAVE, PCATCH, 0);
--tries;
}
USERPRI;
}
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);
}
/* Try to discard pages, in order to recycle a vcache entry.
*
* We also make some sanity checks: ref count, open count, held locks.
*
* We also do some non-VM-related chores, such as releasing the cred pointer
* (for AIX and Solaris) and releasing the gnode (for AIX).
*
* Locking: afs_xvcache lock is held. If it is dropped and re-acquired,
* *slept should be set to warn the caller.
*
* Formerly, afs_xvcache was dropped and re-acquired for Solaris, but now it
* is not dropped and re-acquired for any platform. It may be that *slept is
* therefore obsolescent.
*
* OSF/1 Locking: VN_LOCK has been called.
* We do not lock the vnode here, but instead require that it be exclusive
* locked by code calling osi_VM_StoreAllSegments directly, or scheduling it
* from the bqueue - Matt
* Maybe better to just call vnode_pager_setsize()?
*/
int
osi_VM_FlushVCache(struct vcache *avc, int *slept)
{
struct vm_object *obj;
struct vnode *vp;
if (VREFCOUNT(avc) > 1)
return EBUSY;
if (avc->opens)
return EBUSY;
/* if a lock is held, give up */
if (CheckLock(&avc->lock))
return EBUSY;
return(0);
AFS_GUNLOCK();
vp = AFSTOV(avc);
#ifndef AFS_FBSD70_ENV
lock_vnode(vp);
#endif
if (VOP_GETVOBJECT(vp, &obj) == 0) {
VM_OBJECT_LOCK(obj);
vm_object_page_remove(obj, 0, 0, FALSE);
#if 1
if (obj->ref_count == 0) {
simple_lock(&vp->v_interlock);
vgonel(vp, curthread);
vp->v_tag = VT_AFS;
SetAfsVnode(vp);
}
#endif
VM_OBJECT_UNLOCK(obj);
}
#ifndef AFS_FBSD70_ENV
unlock_vnode(vp);
#endif
AFS_GLOCK();
return 0;
}
/* Try to store pages to cache, in order to store a file back to the server.
*
* Locking: the vcache entry's lock is held. It will usually be dropped and
* re-obtained.
*/
void
osi_VM_StoreAllSegments(struct vcache *avc)
{
struct vnode *vp;
struct vm_object *obj;
int anyio, tries;
ReleaseWriteLock(&avc->lock);
AFS_GUNLOCK();
tries = 5;
vp = AFSTOV(avc);
/*
* I don't understand this. Why not just call vm_object_page_clean()
* and be done with it? I particularly don't understand why we're calling
* vget() here. Is there some reason to believe that the vnode might
* be being recycled at this point? I don't think there's any need for
* this loop, either -- if we keep the vnode locked all the time,
* that and the object lock will prevent any new pages from appearing.
* The loop is what causes the race condition. -GAW
*/
do {
anyio = 0;
obj = vp->v_object;
if (obj != NULL && obj->flags & OBJ_MIGHTBEDIRTY) {
if (!vget(vp, LK_EXCLUSIVE | LK_RETRY, curthread)) {
obj = vp->v_object;
if (obj != NULL) {
VM_OBJECT_LOCK(obj);
vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
VM_OBJECT_UNLOCK(obj);
anyio = 1;
}
vput(vp);
}
}
} while (anyio && (--tries > 0));
AFS_GLOCK();
ObtainWriteLock(&avc->lock, 94);
}
/*
* We require the caller to unmap the entire entry. This allows us to
* safely decrement shm_kmappings when a mapping is removed.
*/
int
shm_unmap(struct file *fp, void *mem, size_t size)
{
struct shmfd *shmfd;
vm_map_entry_t entry;
vm_offset_t kva, ofs;
vm_object_t obj;
vm_pindex_t pindex;
vm_prot_t prot;
boolean_t wired;
vm_map_t map;
int rv;
if (fp->f_type != DTYPE_SHM)
return (EINVAL);
shmfd = fp->f_data;
kva = (vm_offset_t)mem;
ofs = kva & PAGE_MASK;
kva = trunc_page(kva);
size = round_page(size + ofs);
map = kernel_map;
rv = vm_map_lookup(&map, kva, VM_PROT_READ | VM_PROT_WRITE, &entry,
&obj, &pindex, &prot, &wired);
if (rv != KERN_SUCCESS)
return (EINVAL);
if (entry->start != kva || entry->end != kva + size) {
vm_map_lookup_done(map, entry);
return (EINVAL);
}
vm_map_lookup_done(map, entry);
if (obj != shmfd->shm_object)
return (EINVAL);
vm_map_remove(map, kva, kva + size);
VM_OBJECT_LOCK(obj);
KASSERT(shmfd->shm_kmappings > 0, ("shm_unmap: object not mapped"));
shmfd->shm_kmappings--;
VM_OBJECT_UNLOCK(obj);
return (0);
}
static void
unlock_and_deallocate(struct faultstate *fs)
{
vm_object_pip_wakeup(fs->object);
VM_OBJECT_UNLOCK(fs->object);
if (fs->object != fs->first_object) {
VM_OBJECT_LOCK(fs->first_object);
vm_page_lock(fs->first_m);
vm_page_free(fs->first_m);
vm_page_unlock(fs->first_m);
vm_object_pip_wakeup(fs->first_object);
VM_OBJECT_UNLOCK(fs->first_object);
fs->first_m = NULL;
}
vm_object_deallocate(fs->first_object);
unlock_map(fs);
if (fs->vp != NULL) {
vput(fs->vp);
fs->vp = NULL;
}
VFS_UNLOCK_GIANT(fs->vfslocked);
fs->vfslocked = 0;
}
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;
fs.vfslocked = 0;
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_LOCK(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_TRYLOCK(
fs.first_object)) {
VM_OBJECT_UNLOCK(fs.object);
VM_OBJECT_LOCK(fs.first_object);
VM_OBJECT_LOCK(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_UNLOCK(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_UNLOCK(fs.object);
PCPU_INC(cnt.v_intrans);
vm_object_deallocate(fs.first_object);
goto RetryFault;
}
vm_pageq_remove(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);
vnode_lock:
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 (VFS_NEEDSGIANT(vp->v_mount) && !fs.vfslocked) {
fs.vfslocked = 1;
if (!mtx_trylock(&Giant)) {
VM_OBJECT_UNLOCK(fs.object);
mtx_lock(&Giant);
VM_OBJECT_LOCK(fs.object);
goto vnode_lock;
}
}
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) {
int vfslocked;
vfslocked = fs.vfslocked;
fs.vfslocked = 0; /* Keep Giant */
vhold(vp);
release_page(&fs);
unlock_and_deallocate(&fs);
error = vget(vp, locked | LK_RETRY |
LK_CANRECURSE, curthread);
vdrop(vp);
fs.vp = vp;
fs.vfslocked = vfslocked;
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_UNLOCK(fs.object);
fs.object = fs.first_object;
fs.pindex = fs.first_pindex;
fs.m = fs.first_m;
VM_OBJECT_LOCK(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_LOCK(next_object);
vm_object_pip_add(next_object, 1);
if (fs.object != fs.first_object)
vm_object_pip_wakeup(fs.object);
VM_OBJECT_UNLOCK(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_TRYLOCK(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_UNLOCK(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_LOCK(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_UNLOCK(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_LOCK(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)
curthread->td_ru.ru_majflt++;
else
curthread->td_ru.ru_minflt++;
return (KERN_SUCCESS);
}
/*
* 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_LOCK(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 pages in the entry's range, copying each
* one from the source object (it should be there) to the destination
* object.
*/
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_UNLOCK(dst_object);
VM_WAIT;
VM_OBJECT_LOCK(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_LOCK(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_LOCK(backing_object);
pindex += OFF_TO_IDX(object->backing_object_offset);
VM_OBJECT_UNLOCK(object);
object = backing_object;
}
if (src_m == NULL)
panic("vm_fault_copy_wired: page missing");
pmap_copy_page(src_m, dst_m);
VM_OBJECT_UNLOCK(object);
dst_m->valid = VM_PAGE_BITS_ALL;
VM_OBJECT_UNLOCK(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_LOCK(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_UNLOCK(dst_object);
if (upgrade) {
dst_entry->eflags &= ~(MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY);
vm_object_deallocate(src_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);
}
/*
* If blocks are contiguous on disk, use this to provide clustered
* read ahead. We will read as many blocks as possible sequentially
* and then parcel them up into logical blocks in the buffer hash table.
*/
static struct buf *
cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn,
daddr_t blkno, long size, int run, int gbflags, struct buf *fbp)
{
struct bufobj *bo;
struct buf *bp, *tbp;
daddr_t bn;
off_t off;
long tinc, tsize;
int i, inc, j, toff;
KASSERT(size == vp->v_mount->mnt_stat.f_iosize,
("cluster_rbuild: size %ld != filesize %jd\n",
size, (intmax_t)vp->v_mount->mnt_stat.f_iosize));
/*
* avoid a division
*/
while ((u_quad_t) size * (lbn + run) > filesize) {
--run;
}
if (fbp) {
tbp = fbp;
tbp->b_iocmd = BIO_READ;
} else {
tbp = getblk(vp, lbn, size, 0, 0, gbflags);
if (tbp->b_flags & B_CACHE)
return tbp;
tbp->b_flags |= B_ASYNC | B_RAM;
tbp->b_iocmd = BIO_READ;
}
tbp->b_blkno = blkno;
if( (tbp->b_flags & B_MALLOC) ||
((tbp->b_flags & B_VMIO) == 0) || (run <= 1) )
return tbp;
bp = trypbuf(&cluster_pbuf_freecnt);
if (bp == 0)
return tbp;
/*
* We are synthesizing a buffer out of vm_page_t's, but
* if the block size is not page aligned then the starting
* address may not be either. Inherit the b_data offset
* from the original buffer.
*/
bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO;
if ((gbflags & GB_UNMAPPED) != 0) {
bp->b_flags |= B_UNMAPPED;
bp->b_data = unmapped_buf;
} else {
bp->b_data = (char *)((vm_offset_t)bp->b_data |
((vm_offset_t)tbp->b_data & PAGE_MASK));
}
bp->b_iocmd = BIO_READ;
bp->b_iodone = cluster_callback;
bp->b_blkno = blkno;
bp->b_lblkno = lbn;
bp->b_offset = tbp->b_offset;
KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset"));
pbgetvp(vp, bp);
TAILQ_INIT(&bp->b_cluster.cluster_head);
bp->b_bcount = 0;
bp->b_bufsize = 0;
bp->b_npages = 0;
inc = btodb(size);
bo = &vp->v_bufobj;
for (bn = blkno, i = 0; i < run; ++i, bn += inc) {
if (i != 0) {
if ((bp->b_npages * PAGE_SIZE) +
round_page(size) > vp->v_mount->mnt_iosize_max) {
break;
}
tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT |
(gbflags & GB_UNMAPPED));
/* Don't wait around for locked bufs. */
if (tbp == NULL)
break;
/*
* Stop scanning if the buffer is fully valid
* (marked B_CACHE), or locked (may be doing a
* background write), or if the buffer is not
* VMIO backed. The clustering code can only deal
* with VMIO-backed buffers.
*/
BO_LOCK(bo);
if ((tbp->b_vflags & BV_BKGRDINPROG) ||
(tbp->b_flags & B_CACHE) ||
(tbp->b_flags & B_VMIO) == 0) {
BO_UNLOCK(bo);
bqrelse(tbp);
break;
}
BO_UNLOCK(bo);
/*
* The buffer must be completely invalid in order to
* take part in the cluster. If it is partially valid
* then we stop.
*/
off = tbp->b_offset;
tsize = size;
VM_OBJECT_LOCK(tbp->b_bufobj->bo_object);
for (j = 0; tsize > 0; j++) {
toff = off & PAGE_MASK;
tinc = tsize;
if (toff + tinc > PAGE_SIZE)
tinc = PAGE_SIZE - toff;
VM_OBJECT_LOCK_ASSERT(tbp->b_pages[j]->object,
MA_OWNED);
if ((tbp->b_pages[j]->valid &
vm_page_bits(toff, tinc)) != 0)
break;
off += tinc;
tsize -= tinc;
}
VM_OBJECT_UNLOCK(tbp->b_bufobj->bo_object);
if (tsize > 0) {
bqrelse(tbp);
break;
}
/*
* Set a read-ahead mark as appropriate
*/
if ((fbp && (i == 1)) || (i == (run - 1)))
tbp->b_flags |= B_RAM;
/*
* Set the buffer up for an async read (XXX should
* we do this only if we do not wind up brelse()ing?).
* Set the block number if it isn't set, otherwise
* if it is make sure it matches the block number we
* expect.
*/
tbp->b_flags |= B_ASYNC;
tbp->b_iocmd = BIO_READ;
if (tbp->b_blkno == tbp->b_lblkno) {
tbp->b_blkno = bn;
} else if (tbp->b_blkno != bn) {
brelse(tbp);
break;
}
}
/*
* XXX fbp from caller may not be B_ASYNC, but we are going
* to biodone() it in cluster_callback() anyway
*/
BUF_KERNPROC(tbp);
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
tbp, b_cluster.cluster_entry);
VM_OBJECT_LOCK(tbp->b_bufobj->bo_object);
for (j = 0; j < tbp->b_npages; j += 1) {
vm_page_t m;
m = tbp->b_pages[j];
vm_page_io_start(m);
vm_object_pip_add(m->object, 1);
if ((bp->b_npages == 0) ||
(bp->b_pages[bp->b_npages-1] != m)) {
bp->b_pages[bp->b_npages] = m;
bp->b_npages++;
}
if (m->valid == VM_PAGE_BITS_ALL)
tbp->b_pages[j] = bogus_page;
}
VM_OBJECT_UNLOCK(tbp->b_bufobj->bo_object);
/*
* Don't inherit tbp->b_bufsize as it may be larger due to
* a non-page-aligned size. Instead just aggregate using
* 'size'.
*/
if (tbp->b_bcount != size)
printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size);
if (tbp->b_bufsize != size)
printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size);
bp->b_bcount += size;
bp->b_bufsize += size;
}
/*
* Fully valid pages in the cluster are already good and do not need
* to be re-read from disk. Replace the page with bogus_page
*/
VM_OBJECT_LOCK(bp->b_bufobj->bo_object);
for (j = 0; j < bp->b_npages; j++) {
VM_OBJECT_LOCK_ASSERT(bp->b_pages[j]->object, MA_OWNED);
if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL)
bp->b_pages[j] = bogus_page;
}
VM_OBJECT_UNLOCK(bp->b_bufobj->bo_object);
if (bp->b_bufsize > bp->b_kvasize)
panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
bp->b_bufsize, bp->b_kvasize);
bp->b_kvasize = bp->b_bufsize;
if ((bp->b_flags & B_UNMAPPED) == 0) {
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
(vm_page_t *)bp->b_pages, bp->b_npages);
}
return (bp);
}