int
imageboot_setup()
{
dev_t dev;
int error = 0;
char *root_path = NULL;
DBG_TRACE("%s: entry\n", __FUNCTION__);
MALLOC_ZONE(root_path, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
if (root_path == NULL)
return (ENOMEM);
if(PE_parse_boot_argn("rp", root_path, MAXPATHLEN) == FALSE) {
error = ENOENT;
goto done;
}
printf("%s: root image url is %s\n", __FUNCTION__, root_path);
error = di_root_image(root_path, rootdevice, &dev);
if(error) {
printf("%s: di_root_image failed: %d\n", __FUNCTION__, error);
goto done;
}
rootdev = dev;
mountroot = NULL;
printf("%s: root device 0x%x\n", __FUNCTION__, rootdev);
error = vfs_mountroot();
if (error == 0 && rootvnode != NULL) {
struct vnode *tvp;
struct vnode *newdp;
/*
* Get the vnode for '/'.
* Set fdp->fd_fd.fd_cdir to reference it.
*/
if (VFS_ROOT(TAILQ_LAST(&mountlist,mntlist), &newdp, vfs_context_kernel()))
panic("%s: cannot find root vnode", __FUNCTION__);
vnode_ref(newdp);
vnode_put(newdp);
tvp = rootvnode;
vnode_rele(tvp);
filedesc0.fd_cdir = newdp;
rootvnode = newdp;
mount_list_lock();
TAILQ_REMOVE(&mountlist, TAILQ_FIRST(&mountlist), mnt_list);
mount_list_unlock();
mountlist.tqh_first->mnt_flag |= MNT_ROOTFS;
DBG_TRACE("%s: root switched\n", __FUNCTION__);
}
done:
FREE_ZONE(root_path, MAXPATHLEN, M_NAMEI);
DBG_TRACE("%s: exit\n", __FUNCTION__);
return (error);
}
static int
cp_lock_vfs_callback(mount_t mp, void *arg)
{
VFS_IOCTL(mp, FIODEVICELOCKED, arg, 0, vfs_context_kernel());
return 0;
}
int
vm_record_file_write(vnode_t vp, uint64_t offset, char *buf, int size)
{
int error = 0;
vfs_context_t ctx;
ctx = vfs_context_kernel();
error = vn_rdwr(UIO_WRITE, vp, (caddr_t)buf, size, offset,
UIO_SYSSPACE, IO_NODELOCKED, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
return (error);
}
u_int32_t vnode_trim_list (vnode_t vp, struct trim_list *tl, boolean_t route_only)
{
int error = 0;
int trim_index = 0;
u_int32_t blocksize = 0;
struct vnode *devvp;
dk_extent_t *extents;
dk_unmap_t unmap;
_dk_cs_unmap_t cs_unmap;
if ( !(vp->v_mount->mnt_ioflags & MNT_IOFLAGS_UNMAP_SUPPORTED))
return (ENOTSUP);
if (tl == NULL)
return (0);
/*
* Get the underlying device vnode and physical block size
*/
devvp = vp->v_mount->mnt_devvp;
blocksize = vp->v_mount->mnt_devblocksize;
extents = kalloc(sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
memset (&cs_unmap, 0, sizeof(_dk_cs_unmap_t));
cs_unmap.extents = extents;
if (route_only == TRUE)
cs_unmap.options = ROUTE_ONLY;
} else {
memset (&unmap, 0, sizeof(dk_unmap_t));
unmap.extents = extents;
}
while (tl) {
daddr64_t io_blockno; /* Block number corresponding to the start of the extent */
size_t io_bytecount; /* Number of bytes in current extent for the specified range */
size_t trimmed;
size_t remaining_length;
off_t current_offset;
current_offset = tl->tl_offset;
remaining_length = tl->tl_length;
trimmed = 0;
/*
* We may not get the entire range from tl_offset -> tl_offset+tl_length in a single
* extent from the blockmap call. Keep looping/going until we are sure we've hit
* the whole range or if we encounter an error.
*/
while (trimmed < tl->tl_length) {
/*
* VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the
* specified offset. It returns blocks in contiguous chunks, so if the logical range is
* broken into multiple extents, it must be called multiple times, increasing the offset
* in each call to ensure that the entire range is covered.
*/
error = VNOP_BLOCKMAP (vp, current_offset, remaining_length,
&io_blockno, &io_bytecount, NULL, VNODE_READ, NULL);
if (error) {
goto trim_exit;
}
extents[trim_index].offset = (uint64_t) io_blockno * (u_int64_t) blocksize;
extents[trim_index].length = io_bytecount;
trim_index++;
if (trim_index == MAX_BATCH_TO_TRIM) {
if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
cs_unmap.extentsCount = trim_index;
error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
} else {
unmap.extentsCount = trim_index;
error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
}
if (error) {
goto trim_exit;
}
trim_index = 0;
}
trimmed += io_bytecount;
current_offset += io_bytecount;
remaining_length -= io_bytecount;
}
tl = tl->tl_next;
}
if (trim_index) {
if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
cs_unmap.extentsCount = trim_index;
error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
} else {
unmap.extentsCount = trim_index;
error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
}
}
trim_exit:
kfree(extents, sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
return error;
}
int
vm_swapfile_io(vnode_t vp, uint64_t offset, uint64_t start, int npages, int flags)
{
int error = 0;
uint64_t io_size = npages * PAGE_SIZE_64;
#if 1
kern_return_t kr = KERN_SUCCESS;
upl_t upl = NULL;
unsigned int count = 0;
upl_control_flags_t upl_create_flags = 0;
int upl_control_flags = 0;
upl_size_t upl_size = 0;
upl_create_flags = UPL_SET_INTERNAL | UPL_SET_LITE
| UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_OSFMK);
#if ENCRYPTED_SWAP
upl_control_flags = UPL_IOSYNC | UPL_PAGING_ENCRYPTED;
#else
upl_control_flags = UPL_IOSYNC;
#endif
if ((flags & SWAP_READ) == FALSE) {
upl_create_flags |= UPL_COPYOUT_FROM;
}
upl_size = io_size;
kr = vm_map_create_upl( kernel_map,
start,
&upl_size,
&upl,
NULL,
&count,
&upl_create_flags);
if (kr != KERN_SUCCESS || (upl_size != io_size)) {
panic("vm_map_create_upl failed with %d\n", kr);
}
if (flags & SWAP_READ) {
vnode_pagein(vp,
upl,
0,
offset,
io_size,
upl_control_flags | UPL_IGNORE_VALID_PAGE_CHECK,
&error);
if (error) {
#if DEBUG
printf("vm_swapfile_io: vnode_pagein failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
#else /* DEBUG */
printf("vm_swapfile_io: vnode_pagein failed with %d.\n", error);
#endif /* DEBUG */
}
} else {
vnode_pageout(vp,
upl,
0,
offset,
io_size,
upl_control_flags,
&error);
if (error) {
#if DEBUG
printf("vm_swapfile_io: vnode_pageout failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
#else /* DEBUG */
printf("vm_swapfile_io: vnode_pageout failed with %d.\n", error);
#endif /* DEBUG */
}
}
return error;
#else /* 1 */
vfs_context_t ctx;
ctx = vfs_context_kernel();
error = vn_rdwr((flags & SWAP_READ) ? UIO_READ : UIO_WRITE, vp, (caddr_t)start, io_size, offset,
UIO_SYSSPACE, IO_SYNC | IO_NODELOCKED | IO_UNIT | IO_NOCACHE | IO_SWAP_DISPATCH, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
if (error) {
printf("vn_rdwr: Swap I/O failed with %d\n", error);
}
return error;
#endif /* 1 */
}
/*
* Function: devfs_kernel_mount
* Purpose:
* Mount devfs at the given mount point from within the kernel.
*/
int
devfs_kernel_mount(char * mntname)
{
struct mount *mp;
int error;
struct nameidata nd;
struct vnode * vp;
vfs_context_t ctx = vfs_context_kernel();
struct vfstable *vfsp;
/* Find our vfstable entry */
for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
if (!strncmp(vfsp->vfc_name, "devfs", sizeof(vfsp->vfc_name)))
break;
if (!vfsp) {
panic("Could not find entry in vfsconf for devfs.\n");
}
/*
* Get vnode to be covered
*/
NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
CAST_USER_ADDR_T(mntname), ctx);
if ((error = namei(&nd))) {
printf("devfs_kernel_mount: failed to find directory '%s', %d",
mntname, error);
return (error);
}
nameidone(&nd);
vp = nd.ni_vp;
if ((error = VNOP_FSYNC(vp, MNT_WAIT, ctx))) {
printf("devfs_kernel_mount: vnop_fsync failed: %d\n", error);
vnode_put(vp);
return (error);
}
if ((error = buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0))) {
printf("devfs_kernel_mount: buf_invalidateblks failed: %d\n", error);
vnode_put(vp);
return (error);
}
if (vnode_isdir(vp) == 0) {
printf("devfs_kernel_mount: '%s' is not a directory\n", mntname);
vnode_put(vp);
return (ENOTDIR);
}
if ((vnode_mountedhere(vp))) {
vnode_put(vp);
return (EBUSY);
}
/*
* Allocate and initialize the filesystem.
*/
MALLOC_ZONE(mp, struct mount *, sizeof(struct mount),
M_MOUNT, M_WAITOK);
bzero((char *)mp, sizeof(struct mount));
/* Initialize the default IO constraints */
mp->mnt_maxreadcnt = mp->mnt_maxwritecnt = MAXPHYS;
mp->mnt_segreadcnt = mp->mnt_segwritecnt = 32;
mp->mnt_ioflags = 0;
mp->mnt_realrootvp = NULLVP;
mp->mnt_authcache_ttl = CACHED_LOOKUP_RIGHT_TTL;
mount_lock_init(mp);
TAILQ_INIT(&mp->mnt_vnodelist);
TAILQ_INIT(&mp->mnt_workerqueue);
TAILQ_INIT(&mp->mnt_newvnodes);
(void)vfs_busy(mp, LK_NOWAIT);
mp->mnt_op = &devfs_vfsops;
mp->mnt_vtable = vfsp;
mp->mnt_flag = 0;
mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
strlcpy(mp->mnt_vfsstat.f_fstypename, vfsp->vfc_name, MFSTYPENAMELEN);
vp->v_mountedhere = mp;
mp->mnt_vnodecovered = vp;
mp->mnt_vfsstat.f_owner = kauth_cred_getuid(kauth_cred_get());
(void) copystr(mntname, mp->mnt_vfsstat.f_mntonname, MAXPATHLEN - 1, 0);
#if CONFIG_MACF
mac_mount_label_init(mp);
mac_mount_label_associate(ctx, mp);
#endif
error = devfs_mount(mp, NULL, USER_ADDR_NULL, ctx);
if (error) {
printf("devfs_kernel_mount: mount %s failed: %d", mntname, error);
mp->mnt_vtable->vfc_refcount--;
vfs_unbusy(mp);
mount_lock_destroy(mp);
#if CONFIG_MACF
mac_mount_label_destroy(mp);
#endif
FREE_ZONE(mp, sizeof (struct mount), M_MOUNT);
vnode_put(vp);
return (error);
}
vnode_ref(vp);
vnode_put(vp);
vfs_unbusy(mp);
mount_list_add(mp);
return (0);
}
vfs_context_t spl_vfs_context_kernel(void)
{
return vfs_context_kernel();
}
struct kern_direct_file_io_ref_t *
kern_open_file_for_direct_io(const char * name,
uint32_t iflags,
kern_get_file_extents_callback_t callback,
void * callback_ref,
off_t set_file_size,
off_t fs_free_size,
off_t write_file_offset,
void * write_file_addr,
size_t write_file_len,
dev_t * partition_device_result,
dev_t * image_device_result,
uint64_t * partitionbase_result,
uint64_t * maxiocount_result,
uint32_t * oflags)
{
struct kern_direct_file_io_ref_t * ref;
proc_t p;
struct vnode_attr va;
dk_apfs_wbc_range_t wbc_range;
int error;
off_t f_offset;
uint64_t fileblk;
size_t filechunk;
uint64_t physoffset, minoffset;
dev_t device;
dev_t target = 0;
int isssd = 0;
uint32_t flags = 0;
uint32_t blksize;
off_t maxiocount, count, segcount, wbctotal;
boolean_t locked = FALSE;
int fmode, cmode;
struct nameidata nd;
u_int32_t ndflags;
off_t mpFree;
int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
void * p1 = NULL;
void * p2 = NULL;
error = EFAULT;
ref = (struct kern_direct_file_io_ref_t *) kalloc(sizeof(struct kern_direct_file_io_ref_t));
if (!ref)
{
error = EFAULT;
goto out;
}
bzero(ref, sizeof(*ref));
p = kernproc;
ref->ctx = vfs_context_kernel();
fmode = (kIOPolledFileCreate & iflags) ? (O_CREAT | FWRITE) : FWRITE;
cmode = S_IRUSR | S_IWUSR;
ndflags = NOFOLLOW;
NDINIT(&nd, LOOKUP, OP_OPEN, ndflags, UIO_SYSSPACE, CAST_USER_ADDR_T(name), ref->ctx);
VATTR_INIT(&va);
VATTR_SET(&va, va_mode, cmode);
VATTR_SET(&va, va_dataprotect_flags, VA_DP_RAWENCRYPTED);
VATTR_SET(&va, va_dataprotect_class, PROTECTION_CLASS_D);
if ((error = vn_open_auth(&nd, &fmode, &va))) {
kprintf("vn_open_auth(fmode: %d, cmode: %d) failed with error: %d\n", fmode, cmode, error);
goto out;
}
ref->vp = nd.ni_vp;
if (ref->vp->v_type == VREG)
{
vnode_lock_spin(ref->vp);
SET(ref->vp->v_flag, VSWAP);
vnode_unlock(ref->vp);
}
if (write_file_addr && write_file_len)
{
if ((error = kern_write_file(ref, write_file_offset, write_file_addr, write_file_len, IO_SKIP_ENCRYPTION))) {
kprintf("kern_write_file() failed with error: %d\n", error);
goto out;
}
}
VATTR_INIT(&va);
VATTR_WANTED(&va, va_rdev);
VATTR_WANTED(&va, va_fsid);
VATTR_WANTED(&va, va_devid);
VATTR_WANTED(&va, va_data_size);
VATTR_WANTED(&va, va_data_alloc);
VATTR_WANTED(&va, va_nlink);
error = EFAULT;
if (vnode_getattr(ref->vp, &va, ref->ctx)) goto out;
wbctotal = 0;
mpFree = freespace_mb(ref->vp);
mpFree <<= 20;
kprintf("kern_direct_file(%s): vp size %qd, alloc %qd, mp free %qd, keep free %qd\n",
name, va.va_data_size, va.va_data_alloc, mpFree, fs_free_size);
if (ref->vp->v_type == VREG)
{
/* Don't dump files with links. */
if (va.va_nlink != 1) goto out;
device = (VATTR_IS_SUPPORTED(&va, va_devid)) ? va.va_devid : va.va_fsid;
ref->filelength = va.va_data_size;
p1 = &device;
p2 = p;
do_ioctl = &file_ioctl;
if (kIOPolledFileHibernate & iflags)
{
error = do_ioctl(p1, p2, DKIOCAPFSGETWBCRANGE, (caddr_t) &wbc_range);
ref->wbcranged = (error == 0);
}
if (ref->wbcranged)
{
uint32_t idx;
assert(wbc_range.count <= (sizeof(wbc_range.extents) / sizeof(wbc_range.extents[0])));
for (idx = 0; idx < wbc_range.count; idx++) wbctotal += wbc_range.extents[idx].length;
kprintf("kern_direct_file(%s): wbc %qd\n", name, wbctotal);
if (wbctotal) target = wbc_range.dev;
}
if (set_file_size)
{
if (wbctotal)
{
if (wbctotal >= set_file_size) set_file_size = HIBERNATE_MIN_FILE_SIZE;
else
{
set_file_size -= wbctotal;
if (set_file_size < HIBERNATE_MIN_FILE_SIZE) set_file_size = HIBERNATE_MIN_FILE_SIZE;
}
}
if (fs_free_size)
{
mpFree += va.va_data_alloc;
if ((mpFree < set_file_size) || ((mpFree - set_file_size) < fs_free_size))
{
error = ENOSPC;
goto out;
}
}
error = vnode_setsize(ref->vp, set_file_size, IO_NOZEROFILL | IO_NOAUTH, ref->ctx);
if (error) goto out;
ref->filelength = set_file_size;
}
}
else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR))
{
/* Partition. */
device = va.va_rdev;
p1 = ref->vp;
p2 = ref->ctx;
do_ioctl = &device_ioctl;
}
else
{
/* Don't dump to non-regular files. */
error = EFAULT;
goto out;
}
ref->device = device;
// probe for CF
dk_corestorage_info_t cs_info;
memset(&cs_info, 0, sizeof(dk_corestorage_info_t));
error = do_ioctl(p1, p2, DKIOCCORESTORAGE, (caddr_t)&cs_info);
ref->cf = (error == 0) && (cs_info.flags & DK_CORESTORAGE_ENABLE_HOTFILES);
// get block size
error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &ref->blksize);
if (error)
goto out;
minoffset = HIBERNATE_MIN_PHYSICAL_LBA * ref->blksize;
if (ref->vp->v_type != VREG)
{
error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &fileblk);
if (error) goto out;
ref->filelength = fileblk * ref->blksize;
}
// pin logical extents, CS version
error = kern_ioctl_file_extents(ref, _DKIOCCSPINEXTENT, 0, ref->filelength);
if (error && (ENOTTY != error)) goto out;
ref->pinned = (error == 0);
// pin logical extents, apfs version
error = VNOP_IOCTL(ref->vp, FSCTL_FREEZE_EXTENTS, NULL, 0, ref->ctx);
if (error && (ENOTTY != error)) goto out;
ref->frozen = (error == 0);
// generate the block list
error = do_ioctl(p1, p2, DKIOCLOCKPHYSICALEXTENTS, NULL);
if (error) goto out;
locked = TRUE;
f_offset = 0;
for (; f_offset < ref->filelength; f_offset += filechunk)
{
if (ref->vp->v_type == VREG)
{
filechunk = 1*1024*1024*1024;
daddr64_t blkno;
error = VNOP_BLOCKMAP(ref->vp, f_offset, filechunk, &blkno,
&filechunk, NULL, VNODE_WRITE | VNODE_BLOCKMAP_NO_TRACK, NULL);
if (error) goto out;
if (-1LL == blkno) continue;
fileblk = blkno * ref->blksize;
}
else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR))
{
fileblk = f_offset;
filechunk = f_offset ? 0 : ref->filelength;
}
physoffset = 0;
while (physoffset < filechunk)
{
dk_physical_extent_t getphysreq;
bzero(&getphysreq, sizeof(getphysreq));
getphysreq.offset = fileblk + physoffset;
getphysreq.length = (filechunk - physoffset);
error = do_ioctl(p1, p2, DKIOCGETPHYSICALEXTENT, (caddr_t) &getphysreq);
if (error) goto out;
if (!target)
{
target = getphysreq.dev;
}
else if (target != getphysreq.dev)
{
error = ENOTSUP;
goto out;
}
assert(getphysreq.offset >= minoffset);
#if HIBFRAGMENT
uint64_t rev;
for (rev = 4096; rev <= getphysreq.length; rev += 4096)
{
callback(callback_ref, getphysreq.offset + getphysreq.length - rev, 4096);
}
#else
callback(callback_ref, getphysreq.offset, getphysreq.length);
#endif
physoffset += getphysreq.length;
}
}
if (ref->wbcranged)
{
uint32_t idx;
for (idx = 0; idx < wbc_range.count; idx++)
{
assert(wbc_range.extents[idx].offset >= minoffset);
callback(callback_ref, wbc_range.extents[idx].offset, wbc_range.extents[idx].length);
}
}
callback(callback_ref, 0ULL, 0ULL);
if (ref->vp->v_type == VREG) p1 = ⌖
else
{
p1 = ⌖
p2 = p;
do_ioctl = &file_ioctl;
}
// get partition base
if (partitionbase_result)
{
error = do_ioctl(p1, p2, DKIOCGETBASE, (caddr_t) partitionbase_result);
if (error)
goto out;
}
// get block size & constraints
error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &blksize);
if (error)
goto out;
maxiocount = 1*1024*1024*1024;
error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTREAD, (caddr_t) &count);
if (error)
count = 0;
count *= blksize;
if (count && (count < maxiocount))
maxiocount = count;
error = do_ioctl(p1, p2, DKIOCGETMAXBLOCKCOUNTWRITE, (caddr_t) &count);
if (error)
count = 0;
count *= blksize;
if (count && (count < maxiocount))
maxiocount = count;
error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTREAD, (caddr_t) &count);
if (error)
count = 0;
if (count && (count < maxiocount))
maxiocount = count;
error = do_ioctl(p1, p2, DKIOCGETMAXBYTECOUNTWRITE, (caddr_t) &count);
if (error)
count = 0;
if (count && (count < maxiocount))
maxiocount = count;
error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTREAD, (caddr_t) &count);
if (!error)
error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTCOUNTREAD, (caddr_t) &segcount);
if (error)
count = segcount = 0;
count *= segcount;
if (count && (count < maxiocount))
maxiocount = count;
error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTWRITE, (caddr_t) &count);
if (!error)
error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTCOUNTWRITE, (caddr_t) &segcount);
if (error)
count = segcount = 0;
count *= segcount;
if (count && (count < maxiocount))
maxiocount = count;
kprintf("max io 0x%qx bytes\n", maxiocount);
if (maxiocount_result)
*maxiocount_result = maxiocount;
error = do_ioctl(p1, p2, DKIOCISSOLIDSTATE, (caddr_t)&isssd);
if (!error && isssd)
flags |= kIOPolledFileSSD;
if (partition_device_result)
*partition_device_result = device;
if (image_device_result)
*image_device_result = target;
if (oflags)
*oflags = flags;
if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR))
{
vnode_close(ref->vp, FWRITE, ref->ctx);
ref->vp = NULLVP;
ref->ctx = NULL;
}
out:
printf("kern_open_file_for_direct_io(%p, %d)\n", ref, error);
if (error && locked)
{
p1 = &device;
(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
}
if (error && ref)
{
if (ref->vp)
{
(void) kern_ioctl_file_extents(ref, _DKIOCCSUNPINEXTENT, 0, (ref->pinned && ref->cf) ? ref->filelength : 0);
if (ref->frozen)
{
(void) VNOP_IOCTL(ref->vp, FSCTL_THAW_EXTENTS, NULL, 0, ref->ctx);
}
if (ref->wbcranged)
{
(void) do_ioctl(p1, p2, DKIOCAPFSRELEASEWBCRANGE, (caddr_t) NULL);
}
vnode_close(ref->vp, FWRITE, ref->ctx);
ref->vp = NULLVP;
}
ref->ctx = NULL;
kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
ref = NULL;
}
return(ref);
}
/*
* This routine exists to support the load_dylinker().
*
* This routine has its own, separate, understanding of the FAT file format,
* which is terrifically unfortunate.
*/
static
load_return_t
get_macho_vnode(
char *path,
integer_t archbits,
struct mach_header *mach_header,
off_t *file_offset,
off_t *macho_size,
struct vnode **vpp
)
{
struct vnode *vp;
vfs_context_t ctx = vfs_context_current();
proc_t p = vfs_context_proc(ctx);
kauth_cred_t kerncred;
struct nameidata nid, *ndp;
boolean_t is_fat;
struct fat_arch fat_arch;
int error = LOAD_SUCCESS;
int resid;
union {
struct mach_header mach_header;
struct fat_header fat_header;
char pad[512];
} header;
off_t fsize = (off_t)0;
int err2;
/*
* Capture the kernel credential for use in the actual read of the
* file, since the user doing the execution may have execute rights
* but not read rights, but to exec something, we have to either map
* or read it into the new process address space, which requires
* read rights. This is to deal with lack of common credential
* serialization code which would treat NOCRED as "serialize 'root'".
*/
kerncred = vfs_context_ucred(vfs_context_kernel());
ndp = &nid;
/* init the namei data to point the file user's program name */
NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE32, CAST_USER_ADDR_T(path), ctx);
if ((error = namei(ndp)) != 0) {
if (error == ENOENT) {
error = LOAD_ENOENT;
} else {
error = LOAD_FAILURE;
}
return(error);
}
nameidone(ndp);
vp = ndp->ni_vp;
/* check for regular file */
if (vp->v_type != VREG) {
error = LOAD_PROTECT;
goto bad1;
}
/* get size */
if ((error = vnode_size(vp, &fsize, ctx)) != 0) {
error = LOAD_FAILURE;
goto bad1;
}
/* Check mount point */
if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
error = LOAD_PROTECT;
goto bad1;
}
/* check access */
if ((error = vnode_authorize(vp, NULL, KAUTH_VNODE_EXECUTE, ctx)) != 0) {
error = LOAD_PROTECT;
goto bad1;
}
/* try to open it */
if ((error = VNOP_OPEN(vp, FREAD, ctx)) != 0) {
error = LOAD_PROTECT;
goto bad1;
}
if ((error = vn_rdwr(UIO_READ, vp, (caddr_t)&header, sizeof(header), 0,
UIO_SYSSPACE32, IO_NODELOCKED, kerncred, &resid, p)) != 0) {
error = LOAD_IOERROR;
goto bad2;
}
if (header.mach_header.magic == MH_MAGIC ||
header.mach_header.magic == MH_MAGIC_64)
is_fat = FALSE;
else if (header.fat_header.magic == FAT_MAGIC ||
header.fat_header.magic == FAT_CIGAM)
is_fat = TRUE;
else {
error = LOAD_BADMACHO;
goto bad2;
}
if (is_fat) {
/* Look up our architecture in the fat file. */
error = fatfile_getarch_with_bits(vp, archbits, (vm_offset_t)(&header.fat_header), &fat_arch);
if (error != LOAD_SUCCESS)
goto bad2;
/* Read the Mach-O header out of it */
error = vn_rdwr(UIO_READ, vp, (caddr_t)&header.mach_header,
sizeof(header.mach_header), fat_arch.offset,
UIO_SYSSPACE32, IO_NODELOCKED, kerncred, &resid, p);
if (error) {
error = LOAD_IOERROR;
goto bad2;
}
/* Is this really a Mach-O? */
if (header.mach_header.magic != MH_MAGIC &&
header.mach_header.magic != MH_MAGIC_64) {
error = LOAD_BADMACHO;
goto bad2;
}
*file_offset = fat_arch.offset;
*macho_size = fat_arch.size;
} else {
/*
* Force get_macho_vnode() to fail if the architecture bits
* do not match the expected architecture bits. This in
* turn causes load_dylinker() to fail for the same reason,
* so it ensures the dynamic linker and the binary are in
* lock-step. This is potentially bad, if we ever add to
* the CPU_ARCH_* bits any bits that are desirable but not
* required, since the dynamic linker might work, but we will
* refuse to load it because of this check.
*/
if ((cpu_type_t)(header.mach_header.cputype & CPU_ARCH_MASK) != archbits)
return(LOAD_BADARCH);
*file_offset = 0;
*macho_size = fsize;
}
*mach_header = header.mach_header;
*vpp = vp;
ubc_setsize(vp, fsize);
return (error);
bad2:
err2 = VNOP_CLOSE(vp, FREAD, ctx);
vnode_put(vp);
return (error);
bad1:
vnode_put(vp);
return(error);
}
