static void
physio_done(struct work *wk, void *dummy)
{
struct buf *bp = (void *)wk;
size_t todo = bp->b_bufsize;
size_t done = bp->b_bcount - bp->b_resid;
struct physio_stat *ps = bp->b_private;
bool is_iobuf;
KASSERT(&bp->b_work == wk);
KASSERT(bp->b_bcount <= todo);
KASSERT(bp->b_resid <= bp->b_bcount);
KASSERT((bp->b_flags & B_PHYS) != 0);
KASSERT(dummy == NULL);
vunmapbuf(bp, todo);
uvm_vsunlock(bp->b_proc->p_vmspace, bp->b_data, todo);
mutex_enter(&ps->ps_lock);
is_iobuf = (bp != ps->ps_orig_bp);
if (__predict_false(done != todo)) {
off_t endoffset = dbtob(bp->b_blkno) + done;
/*
* we got an error or hit EOM.
*
* we only care about the first one.
* ie. the one at the lowest offset.
*/
KASSERT(ps->ps_endoffset != endoffset);
DPRINTF(("%s: error=%d at %" PRIu64 " - %" PRIu64
", blkno=%" PRIu64 ", bcount=%d, flags=0x%x\n",
__func__, bp->b_error, dbtob(bp->b_blkno), endoffset,
bp->b_blkno, bp->b_bcount, bp->b_flags));
if (ps->ps_endoffset == -1 || endoffset < ps->ps_endoffset) {
DPRINTF(("%s: ps=%p, error %d -> %d, endoff %" PRIu64
" -> %" PRIu64 "\n",
__func__, ps,
ps->ps_error, bp->b_error,
ps->ps_endoffset, endoffset));
ps->ps_endoffset = endoffset;
ps->ps_error = bp->b_error;
}
ps->ps_failed++;
} else {
KASSERT(bp->b_error == 0);
}
ps->ps_running--;
cv_signal(&ps->ps_cv);
mutex_exit(&ps->ps_lock);
if (is_iobuf)
putiobuf(bp);
}
/* ARGSUSED */
int
sys_mincore(struct lwp *l, const struct sys_mincore_args *uap,
register_t *retval)
{
/* {
syscallarg(void *) addr;
syscallarg(size_t) len;
syscallarg(char *) vec;
} */
struct proc *p = l->l_proc;
struct vm_page *pg;
char *vec, pgi;
struct uvm_object *uobj;
struct vm_amap *amap;
struct vm_anon *anon;
struct vm_map_entry *entry;
vaddr_t start, end, lim;
struct vm_map *map;
vsize_t len;
int error = 0, npgs;
map = &p->p_vmspace->vm_map;
start = (vaddr_t)SCARG(uap, addr);
len = SCARG(uap, len);
vec = SCARG(uap, vec);
if (start & PAGE_MASK)
return EINVAL;
len = round_page(len);
end = start + len;
if (end <= start)
return EINVAL;
/*
* Lock down vec, so our returned status isn't outdated by
* storing the status byte for a page.
*/
npgs = len >> PAGE_SHIFT;
error = uvm_vslock(p->p_vmspace, vec, npgs, VM_PROT_WRITE);
if (error) {
return error;
}
vm_map_lock_read(map);
if (uvm_map_lookup_entry(map, start, &entry) == false) {
error = ENOMEM;
goto out;
}
for (/* nothing */;
entry != &map->header && entry->start < end;
entry = entry->next) {
KASSERT(!UVM_ET_ISSUBMAP(entry));
KASSERT(start >= entry->start);
/* Make sure there are no holes. */
if (entry->end < end &&
(entry->next == &map->header ||
entry->next->start > entry->end)) {
error = ENOMEM;
goto out;
}
lim = end < entry->end ? end : entry->end;
/*
* Special case for objects with no "real" pages. Those
* are always considered resident (mapped devices).
*/
if (UVM_ET_ISOBJ(entry)) {
KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
if (UVM_OBJ_IS_DEVICE(entry->object.uvm_obj)) {
for (/* nothing */; start < lim;
start += PAGE_SIZE, vec++)
subyte(vec, 1);
continue;
}
}
amap = entry->aref.ar_amap; /* upper layer */
uobj = entry->object.uvm_obj; /* lower layer */
if (amap != NULL)
amap_lock(amap);
if (uobj != NULL)
mutex_enter(uobj->vmobjlock);
for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
pgi = 0;
if (amap != NULL) {
/* Check the upper layer first. */
anon = amap_lookup(&entry->aref,
start - entry->start);
/* Don't need to lock anon here. */
if (anon != NULL && anon->an_page != NULL) {
/*
* Anon has the page for this entry
* offset.
*/
pgi = 1;
}
}
if (uobj != NULL && pgi == 0) {
/* Check the lower layer. */
pg = uvm_pagelookup(uobj,
entry->offset + (start - entry->start));
if (pg != NULL) {
/*
* Object has the page for this entry
* offset.
*/
pgi = 1;
}
}
(void) subyte(vec, pgi);
}
if (uobj != NULL)
mutex_exit(uobj->vmobjlock);
if (amap != NULL)
amap_unlock(amap);
}
out:
vm_map_unlock_read(map);
uvm_vsunlock(p->p_vmspace, SCARG(uap, vec), npgs);
return error;
}
/*
* Do "physical I/O" on behalf of a user. "Physical I/O" is I/O directly
* from the raw device to user buffers, and bypasses the buffer cache.
*
* Comments in brackets are from Leffler, et al.'s pseudo-code implementation.
*/
int
physio(void (*strategy)(struct buf *), struct buf *bp, dev_t dev, int flags,
void (*minphys)(struct buf *), struct uio *uio)
{
struct iovec *iovp;
struct proc *p = curproc;
int error, done, i, nobuf, s, todo;
error = 0;
flags &= B_READ | B_WRITE;
/* Make sure we have a buffer, creating one if necessary. */
if ((nobuf = (bp == NULL)) != 0)
bp = getphysbuf();
/* [raise the processor priority level to splbio;] */
s = splbio();
/* [while the buffer is marked busy] */
while (bp->b_flags & B_BUSY) {
/* [mark the buffer wanted] */
bp->b_flags |= B_WANTED;
/* [wait until the buffer is available] */
tsleep(bp, PRIBIO+1, "physbuf", 0);
}
/* Mark it busy, so nobody else will use it. */
bp->b_flags |= B_BUSY;
/* [lower the priority level] */
splx(s);
/* [set up the fixed part of the buffer for a transfer] */
bp->b_dev = dev;
bp->b_error = 0;
bp->b_proc = p;
LIST_INIT(&bp->b_dep);
/*
* [while there are data to transfer and no I/O error]
* Note that I/O errors are handled with a 'goto' at the bottom
* of the 'while' loop.
*/
for (i = 0; i < uio->uio_iovcnt; i++) {
iovp = &uio->uio_iov[i];
while (iovp->iov_len > 0) {
/*
* [mark the buffer busy for physical I/O]
* (i.e. set B_PHYS (because it's an I/O to user
* memory), and B_RAW, because B_RAW is to be
* "Set by physio for raw transfers.", in addition
* to the "busy" and read/write flag.)
*/
bp->b_flags = B_BUSY | B_PHYS | B_RAW | flags;
/* [set up the buffer for a maximum-sized transfer] */
bp->b_blkno = btodb(uio->uio_offset);
bp->b_data = iovp->iov_base;
/*
* Because iov_len is unsigned but b_bcount is signed,
* an overflow is possible. Therefore bound to MAXPHYS
* before calling minphys.
*/
if (iovp->iov_len > MAXPHYS)
bp->b_bcount = MAXPHYS;
else
bp->b_bcount = iovp->iov_len;
/*
* [call minphys to bound the transfer size]
* and remember the amount of data to transfer,
* for later comparison.
*/
(*minphys)(bp);
todo = bp->b_bcount;
#ifdef DIAGNOSTIC
if (todo < 0)
panic("todo < 0; minphys broken");
if (todo > MAXPHYS)
panic("todo > MAXPHYS; minphys broken");
#endif
/*
* [lock the part of the user address space involved
* in the transfer]
* Beware vmapbuf(); it clobbers b_data and
* saves it in b_saveaddr. However, vunmapbuf()
* restores it.
*/
error = uvm_vslock(p, bp->b_data, todo,
(flags & B_READ) ?
VM_PROT_READ | VM_PROT_WRITE : VM_PROT_READ);
if (error) {
bp->b_flags |= B_ERROR;
bp->b_error = error;
goto after_unlock;
}
vmapbuf(bp, todo);
/* [call strategy to start the transfer] */
(*strategy)(bp);
/*
* Note that the raise/wait/lower/get error
* steps below would be done by biowait(), but
* we want to unlock the address space before
* we lower the priority.
*
* [raise the priority level to splbio]
*/
s = splbio();
/* [wait for the transfer to complete] */
while ((bp->b_flags & B_DONE) == 0)
tsleep(bp, PRIBIO + 1, "physio", 0);
/* Mark it busy again, so nobody else will use it. */
bp->b_flags |= B_BUSY;
/* [lower the priority level] */
splx(s);
/*
* [unlock the part of the address space previously
* locked]
*/
vunmapbuf(bp, todo);
uvm_vsunlock(p, bp->b_data, todo);
after_unlock:
/* remember error value (save a splbio/splx pair) */
if (bp->b_flags & B_ERROR)
error = (bp->b_error ? bp->b_error : EIO);
/*
* [deduct the transfer size from the total number
* of data to transfer]
*/
done = bp->b_bcount - bp->b_resid;
#ifdef DIAGNOSTIC
if (done < 0)
panic("done < 0; strategy broken");
if (done > todo)
panic("done > todo; strategy broken");
#endif
iovp->iov_len -= done;
iovp->iov_base = (caddr_t)iovp->iov_base + done;
uio->uio_offset += done;
uio->uio_resid -= done;
/*
* Now, check for an error.
* Also, handle weird end-of-disk semantics.
*/
if (error || done < todo)
goto done;
}
}
done:
/*
* [clean up the state of the buffer]
* Remember if somebody wants it, so we can wake them up below.
* Also, if we had to steal it, give it back.
*/
s = splbio();
bp->b_flags &= ~(B_BUSY | B_PHYS | B_RAW);
if (nobuf)
putphysbuf(bp);
else {
/*
* [if another process is waiting for the raw I/O buffer,
* wake up processes waiting to do physical I/O]
*/
if (bp->b_flags & B_WANTED) {
bp->b_flags &= ~B_WANTED;
wakeup(bp);
}
}
splx(s);
return (error);
}
/* ARGSUSED */
int
sys_mincore(struct proc *p, void *v, register_t *retval)
{
struct sys_mincore_args /* {
syscallarg(void *) addr;
syscallarg(size_t) len;
syscallarg(char *) vec;
} */ *uap = v;
vm_page_t m;
char *vec, pgi;
struct uvm_object *uobj;
struct vm_amap *amap;
struct vm_anon *anon;
vm_map_entry_t entry;
vaddr_t start, end, lim;
vm_map_t map;
vsize_t len, npgs;
int error = 0;
map = &p->p_vmspace->vm_map;
start = (vaddr_t)SCARG(uap, addr);
len = SCARG(uap, len);
vec = SCARG(uap, vec);
if (start & PAGE_MASK)
return (EINVAL);
len = round_page(len);
end = start + len;
if (end <= start)
return (EINVAL);
npgs = len >> PAGE_SHIFT;
/*
* Lock down vec, so our returned status isn't outdated by
* storing the status byte for a page.
*/
if ((error = uvm_vslock(p, vec, npgs, VM_PROT_WRITE)) != 0)
return (error);
vm_map_lock_read(map);
if (uvm_map_lookup_entry(map, start, &entry) == FALSE) {
error = ENOMEM;
goto out;
}
for (/* nothing */;
entry != &map->header && entry->start < end;
entry = entry->next) {
KASSERT(!UVM_ET_ISSUBMAP(entry));
KASSERT(start >= entry->start);
/* Make sure there are no holes. */
if (entry->end < end &&
(entry->next == &map->header ||
entry->next->start > entry->end)) {
error = ENOMEM;
goto out;
}
lim = end < entry->end ? end : entry->end;
/*
* Special case for objects with no "real" pages. Those
* are always considered resident (mapped devices).
*/
if (UVM_ET_ISOBJ(entry)) {
KASSERT(!UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj));
if (entry->object.uvm_obj->pgops->pgo_releasepg
== NULL) {
pgi = 1;
for (/* nothing */; start < lim;
start += PAGE_SIZE, vec++)
copyout(&pgi, vec, sizeof(char));
continue;
}
}
amap = entry->aref.ar_amap; /* top layer */
uobj = entry->object.uvm_obj; /* bottom layer */
if (uobj != NULL)
simple_lock(&uobj->vmobjlock);
for (/* nothing */; start < lim; start += PAGE_SIZE, vec++) {
pgi = 0;
if (amap != NULL) {
/* Check the top layer first. */
anon = amap_lookup(&entry->aref,
start - entry->start);
/* Don't need to lock anon here. */
if (anon != NULL && anon->an_page != NULL) {
/*
* Anon has the page for this entry
* offset.
*/
pgi = 1;
}
}
if (uobj != NULL && pgi == 0) {
/* Check the bottom layer. */
m = uvm_pagelookup(uobj,
entry->offset + (start - entry->start));
if (m != NULL) {
/*
* Object has the page for this entry
* offset.
*/
pgi = 1;
}
}
copyout(&pgi, vec, sizeof(char));
}
if (uobj != NULL)
simple_unlock(&uobj->vmobjlock);
}
out:
vm_map_unlock_read(map);
uvm_vsunlock(p, SCARG(uap, vec), npgs);
return (error);
}
