/*
* It appears on export/reboot, iokit can hold a lock, then call our
* termination handler, and we end up locking-against-ourselves inside
* IOKit. We are then forced to make the vnode_close() call be async.
*/
static void vdev_disk_close_thread(void *arg)
{
struct vnode *vp = arg;
(void) vnode_close(vp, 0,
spl_vfs_context_kernel());
thread_exit();
}
int
kern_file_close(struct cfs_kern_file *fp)
{
vnode_close(fp->f_vp, fp->f_flags, fp->f_ctxt);
vfs_context_rele(fp->f_ctxt);
_FREE(fp, M_TEMP);
return 0;
}
void VNodeDiskDeviceClass::closeVNode()
{
if (m_vnode != NULL) {
IOLog("Closing the file node\n");
vfs_context_t vfsContext = vfs_context_create((vfs_context_t) 0);
vnode_close(m_vnode, 0, vfsContext);
vfs_context_rele(vfsContext);
m_vnode = NULL;
}
}
int
VOP_CLOSE(struct vnode *vp, int flag, int count, offset_t off, void *cr, void *k)
{
vfs_context_t vctx;
int error;
vctx = vfs_context_create((vfs_context_t)0);
error = vnode_close(vp, flag & FWRITE, vctx);
(void) vfs_context_rele(vctx);
return (error);
}
static int
sd_closelog(vfs_context_t ctx)
{
int error = 0;
if (sd_logvp != NULLVP) {
VNOP_FSYNC(sd_logvp, MNT_WAIT, ctx);
error = vnode_close(sd_logvp, FWRITE, ctx);
}
return error;
}
void
kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref,
off_t write_offset, caddr_t addr, vm_size_t write_length,
off_t discard_offset, off_t discard_end)
{
int error;
kprintf("kern_close_file_for_direct_io\n");
if (!ref) return;
if (ref->vp)
{
int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
void * p1;
void * p2;
if (ref->vp->v_type == VREG)
{
p1 = &ref->device;
p2 = kernproc;
do_ioctl = &file_ioctl;
}
else
{
/* Partition. */
p1 = ref->vp;
p2 = ref->ctx;
do_ioctl = &device_ioctl;
}
(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
if (addr && write_length)
{
(void) kern_write_file(ref, write_offset, addr, write_length);
}
if (discard_offset && discard_end && !ref->pinned)
{
(void) kern_ioctl_file_extents(ref, DKIOCUNMAP, discard_offset, discard_end);
}
error = vnode_close(ref->vp, FWRITE, ref->ctx);
ref->vp = NULLVP;
kprintf("vnode_close(%d)\n", error);
}
vfs_context_rele(ref->ctx);
ref->ctx = NULL;
kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
}
void
vm_swapfile_close(uint64_t path_addr, vnode_t vp)
{
struct nameidata nd;
vfs_context_t context = vfs_context_current();
int error = 0;
vnode_getwithref(vp);
vnode_close(vp, 0, context);
NDINIT(&nd, DELETE, OP_UNLINK, AUDITVNPATH1, UIO_SYSSPACE,
path_addr, context);
error = unlink1(context, &nd, 0);
}
int
osi_UFSClose(struct osi_file *afile)
{
AFS_STATCNT(osi_Close);
if (afile->vnode) {
#ifdef AFS_DARWIN80_ENV
vnode_close(afile->vnode, O_RDWR, afs_osi_ctxtp);
#else
AFS_RELE(afile->vnode);
#endif
}
osi_FreeSmallSpace(afile);
return 0;
}
bool VNodeDiskDeviceClass::setupVNode() {
int vapError = -1;
struct vnode_attr vap;
if (m_vnode != NULL)
return true;
vfs_context_t vfsContext = vfs_context_create((vfs_context_t) 0);
int vnodeError = vnode_open(m_filePath->getCStringNoCopy(), (FREAD | FWRITE), 0, 0,
&m_vnode, vfsContext);
if (vnodeError || m_vnode == NULL) {
IOLog("Error when opening file %s: error %d\n",
m_filePath->getCStringNoCopy(), vnodeError);
goto failure;
}
if (!vnode_isreg(m_vnode)) {
IOLog("Error when opening file %s: not a regular file\n",
m_filePath->getCStringNoCopy());
vnode_close(m_vnode, (FREAD | FWRITE), vfsContext);
goto failure;
}
VATTR_INIT(&vap);
VATTR_WANTED(&vap, va_data_size);
vapError = vnode_getattr(m_vnode, &vap, vfsContext);
if (vapError) {
IOLog("Error when retrieving vnode's attributes with error code %d\n", vapError);
goto failure;
}
if (vap.va_data_size < m_blockSize * m_blockNum) {
IOLog("Error file %s is too small, actual size is %llu\n",
m_filePath->getCStringNoCopy(), vap.va_data_size);
goto failure;
}
vfs_context_rele(vfsContext);
return true;
failure:
vfs_context_rele(vfsContext);
return false;
}
void
vm_swapfile_close(uint64_t path_addr, vnode_t vp)
{
vfs_context_t context = vfs_context_current();
int error;
vnode_getwithref(vp);
vnode_close(vp, 0, context);
error = unlink1(context, NULLVP, CAST_USER_ADDR_T(path_addr),
UIO_SYSSPACE, 0);
#if DEVELOPMENT || DEBUG
if (error)
printf("%s : unlink of %s failed with error %d", __FUNCTION__,
(char *)path_addr, error);
#endif
}
IOReturn FileNVRAM::write_buffer(char* aBuffer, vfs_context_t aCtx)
{
IOReturn error = 0;
int length = (int)strlen(aBuffer);
struct vnode * vp;
if (aCtx)
{
if ((error = vnode_open(FILE_NVRAM_PATH, (O_TRUNC | O_CREAT | FWRITE | O_NOFOLLOW), S_IRUSR | S_IWUSR, VNODE_LOOKUP_NOFOLLOW, &vp, aCtx)))
{
printf("FileNVRAM.kext: Error, vnode_open(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
return error;
}
else
{
if ((error = vnode_isreg(vp)) == VREG)
{
if ((error = vn_rdwr(UIO_WRITE, vp, aBuffer, length, 0, UIO_SYSSPACE, IO_NOCACHE|IO_NODELOCKED|IO_UNIT, vfs_context_ucred(aCtx), (int *) 0, vfs_context_proc(aCtx))))
{
printf("FileNVRAM.kext: Error, vn_rdwr(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
}
if ((error = vnode_close(vp, FWASWRITTEN, aCtx)))
{
printf("FileNVRAM.kext: Error, vnode_close(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
}
}
else
{
printf("FileNVRAM.kext: Error, vnode_isreg(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
}
}
}
else
{
printf("FileNVRAM.kext: aCtx == NULL!\n");
error = 0xFFFF; // EINVAL;
}
return error;
}
static void
vdev_disk_close(vdev_t *vd)
{
vdev_disk_t *dvd = vd->vdev_tsd;
if (dvd == NULL)
return;
if (dvd->vd_devvp != NULL) {
vfs_context_t context;
context = vfs_context_create((vfs_context_t)0);
(void) vnode_close(dvd->vd_devvp, spa_mode(vd->vdev_spa), context);
(void) vfs_context_rele(context);
}
kmem_free(dvd, sizeof (vdev_disk_t));
vd->vdev_tsd = NULL;
}
void DldIOShadowFile::free()
{
assert( preemption_enabled() );
if( NULLVP != this->vnode ){
//
// write a terminator
//
UInt64 terminator = DLD_SHADOW_FILE_TERMINATOR;
this->write( &terminator, sizeof( terminator ), DLD_IGNR_FSIZE );
//
// TO DO - ponder the vfs context retrieval as it seems vfs_context_current might be
// an arbitrary one which differs from the open context
//
vfs_context_t vfs_context;
vfs_context = vfs_context_create( NULL );
assert( vfs_context );
if( vfs_context ){
VNOP_FSYNC( this->vnode, MNT_WAIT, vfs_context );
vnode_close( this->vnode, ( this->bytesWritten )? FWASWRITTEN: 0x0, vfs_context );
vfs_context_rele( vfs_context );
}// end if( vfs_context )
}
if( this->path ){
assert( this->pathLength );
IOFree( this->path, this->pathLength );
}
if( this->rwLock )
IORWLockFree( this->rwLock );
}
void
kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref,
off_t write_offset, void * addr, size_t write_length,
off_t discard_offset, off_t discard_end)
{
int error;
printf("kern_close_file_for_direct_io(%p)\n", ref);
if (!ref) return;
if (ref->vp)
{
int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result);
void * p1;
void * p2;
discard_offset = ((discard_offset + ref->blksize - 1) & ~(((off_t) ref->blksize) - 1));
discard_end = ((discard_end) & ~(((off_t) ref->blksize) - 1));
if (ref->vp->v_type == VREG)
{
p1 = &ref->device;
p2 = kernproc;
do_ioctl = &file_ioctl;
}
else
{
/* Partition. */
p1 = ref->vp;
p2 = ref->ctx;
do_ioctl = &device_ioctl;
}
(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
//XXX If unmapping extents then don't also need to unpin; except ...
//XXX if file unaligned (HFS 4k / Fusion 128k) then pin is superset and
//XXX unmap is subset, so save extra walk over file extents (and the risk
//XXX that CF drain starts) vs leaving partial units pinned to SSD
//XXX (until whatever was sharing also unmaps). Err on cleaning up fully.
boolean_t will_unmap = (!ref->pinned || ref->cf) && (discard_end > discard_offset);
boolean_t will_unpin = (ref->pinned && ref->cf /* && !will_unmap */);
(void) kern_ioctl_file_extents(ref, _DKIOCCSUNPINEXTENT, 0, (will_unpin) ? ref->filelength : 0);
if (will_unmap)
{
(void) kern_ioctl_file_extents(ref, DKIOCUNMAP, discard_offset, (ref->cf) ? ref->filelength : discard_end);
}
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);
}
if (addr && write_length)
{
(void) kern_write_file(ref, write_offset, addr, write_length, IO_SKIP_ENCRYPTION);
}
error = vnode_close(ref->vp, FWRITE, ref->ctx);
ref->vp = NULLVP;
kprintf("vnode_close(%d)\n", error);
}
ref->ctx = NULL;
kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
}
static int
vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
{
vdev_file_t *vf;
#ifdef __APPLE__
struct vnode *vp, *rootdir;
struct vnode_attr vattr;
vfs_context_t context;
#else
vnode_t *vp;
vattr_t vattr;
#endif
int error;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
/*
* We always open the files from the root of the global zone, even if
* we're in a local zone. If the user has gotten to this point, the
* administrator has already decided that the pool should be available
* to local zone users, so the underlying devices should be as well.
*/
ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
#ifdef __APPLE__
rootdir = getrootdir();
#endif
error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE, spa_mode | FOFFMAX,
0, &vp, 0, 0, rootdir);
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (error);
}
vf->vf_vnode = vp;
#ifdef _KERNEL
/*
* Make sure it's a regular file.
*/
#ifdef __APPLE__
if (!vnode_isreg(vp)) {
#else
if (vp->v_type != VREG) {
#endif
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (ENODEV);
}
#endif
/*
* Determine the physical size of the file.
*/
#ifdef __APPLE__
VATTR_INIT(&vattr);
VATTR_WANTED(&vattr, va_data_size);
context = vfs_context_create((vfs_context_t)0);
error = vnode_getattr(vp, &vattr, context);
(void) vfs_context_rele(context);
if (error || !VATTR_IS_SUPPORTED(&vattr, va_data_size)) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (error);
}
*psize = vattr.va_data_size;
#else
vattr.va_mask = AT_SIZE;
error = VOP_GETATTR(vp, &vattr, 0, kcred);
if (error) {
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
return (error);
}
*psize = vattr.va_size;
#endif
*ashift = SPA_MINBLOCKSHIFT;
return (0);
}
static void
vdev_file_close(vdev_t *vd)
{
vdev_file_t *vf = vd->vdev_tsd;
if (vf == NULL)
return;
if (vf->vf_vnode != NULL) {
#ifdef __APPLE__
vfs_context_t context;
context = vfs_context_create((vfs_context_t)0);
/* ### APPLE TODO #### */
// (void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred);
(void) vnode_close(vf->vf_vnode, spa_mode, context);
(void) vfs_context_rele(context);
#else
(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred);
(void) VOP_CLOSE(vf->vf_vnode, spa_mode, 1, 0, kcred);
VN_RELE(vf->vf_vnode);
#endif
}
kmem_free(vf, sizeof (vdev_file_t));
vd->vdev_tsd = NULL;
}
struct kern_direct_file_io_ref_t *
kern_open_file_for_direct_io(const char * name,
kern_get_file_extents_callback_t callback,
void * callback_ref,
dev_t * partition_device_result,
dev_t * image_device_result,
uint64_t * partitionbase_result,
uint64_t * maxiocount_result,
uint32_t * oflags,
off_t offset,
caddr_t addr,
vm_size_t len)
{
struct kern_direct_file_io_ref_t * ref;
proc_t p;
struct vnode_attr va;
int error;
off_t f_offset;
uint64_t fileblk;
size_t filechunk;
uint64_t physoffset;
dev_t device;
dev_t target = 0;
int isssd = 0;
uint32_t flags = 0;
uint32_t blksize;
off_t maxiocount, count;
boolean_t locked = FALSE;
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_create(vfs_context_current());
if ((error = vnode_open(name, (O_CREAT | FWRITE), (0), 0, &ref->vp, ref->ctx)))
goto out;
if (addr && len)
{
if ((error = kern_write_file(ref, offset, addr, len)))
goto out;
}
VATTR_INIT(&va);
VATTR_WANTED(&va, va_rdev);
VATTR_WANTED(&va, va_fsid);
VATTR_WANTED(&va, va_data_size);
VATTR_WANTED(&va, va_nlink);
error = EFAULT;
if (vnode_getattr(ref->vp, &va, ref->ctx))
goto out;
kprintf("vp va_rdev major %d minor %d\n", major(va.va_rdev), minor(va.va_rdev));
kprintf("vp va_fsid major %d minor %d\n", major(va.va_fsid), minor(va.va_fsid));
kprintf("vp size %qd\n", va.va_data_size);
if (ref->vp->v_type == VREG)
{
/* Don't dump files with links. */
if (va.va_nlink != 1)
goto out;
device = va.va_fsid;
p1 = &device;
p2 = p;
do_ioctl = &file_ioctl;
}
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;
// get block size
error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &ref->blksize);
if (error)
goto out;
if (ref->vp->v_type == VREG)
ref->filelength = va.va_data_size;
else
{
error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &fileblk);
if (error)
goto out;
ref->filelength = fileblk * ref->blksize;
}
// pin logical extents
error = kern_ioctl_file_extents(ref, _DKIOCCSPINEXTENT, 0, ref->filelength);
if (error && (ENOTTY != error)) goto out;
ref->pinned = (error == 0);
// generate the block list
error = do_ioctl(p1, p2, DKIOCLOCKPHYSICALEXTENTS, NULL);
if (error)
goto out;
locked = TRUE;
f_offset = 0;
while (f_offset < ref->filelength)
{
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, 0, NULL);
if (error)
goto out;
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;
}
callback(callback_ref, getphysreq.offset, getphysreq.length);
physoffset += getphysreq.length;
}
f_offset += filechunk;
}
callback(callback_ref, 0ULL, 0ULL);
if (ref->vp->v_type == VREG)
p1 = ⌖
// get partition base
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)
count = 0;
if (count && (count < maxiocount))
maxiocount = count;
error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTWRITE, (caddr_t) &count);
if (error)
count = 0;
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 |= kIOHibernateOptionSSD;
if (partition_device_result)
*partition_device_result = device;
if (image_device_result)
*image_device_result = target;
if (flags)
*oflags = flags;
out:
kprintf("kern_open_file_for_direct_io(%d)\n", error);
if (error && locked)
{
p1 = &device;
(void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL);
}
if (error && ref)
{
if (ref->vp)
{
vnode_close(ref->vp, FWRITE, ref->ctx);
ref->vp = NULLVP;
}
vfs_context_rele(ref->ctx);
kfree(ref, sizeof(struct kern_direct_file_io_ref_t));
ref = NULL;
}
return(ref);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
spa_t *spa = vd->vdev_spa;
vdev_disk_t *dvd = vd->vdev_tsd;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
int isssd;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
/*
* Reopen the device if it's not currently open. Otherwise,
* just update the physical size of the device.
*/
if (dvd != NULL) {
if (dvd->vd_offline) {
/*
* If we are opening a device in its offline notify
* context, the LDI handle was just closed. Clean
* up the LDI event callbacks and free vd->vdev_tsd.
*/
vdev_disk_free(vd);
} else {
ASSERT(vd->vdev_reopening);
devvp = dvd->vd_devvp;
goto skip_open;
}
}
/*
* Create vd->vdev_tsd.
*/
vdev_disk_alloc(vd);
dvd = vd->vdev_tsd;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the same device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*/
/* ### APPLE TODO ### */
#ifdef illumos
if (vd->vdev_devid != NULL) {
if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
&dvd->vd_minor) != 0) {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (SET_ERROR(EINVAL));
}
}
#endif
error = EINVAL; /* presume failure */
if (vd->vdev_path != NULL) {
context = vfs_context_create( spl_vfs_context_kernel() );
/* Obtain an opened/referenced vnode for the device. */
if ((error = vnode_open(vd->vdev_path, spa_mode(spa), 0, 0,
&devvp, context))) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/*
* ### APPLE TODO ###
* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
} else {
goto out;
}
int len = MAXPATHLEN;
if (vn_getpath(devvp, dvd->vd_readlinkname, &len) == 0) {
dprintf("ZFS: '%s' resolved name is '%s'\n",
vd->vdev_path, dvd->vd_readlinkname);
} else {
dvd->vd_readlinkname[0] = 0;
}
skip_open:
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0,
context) != 0 ||
VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt, 0,
context) != 0) {
error = EINVAL;
goto out;
}
*psize = blkcnt * (uint64_t)blksize;
*max_psize = *psize;
dvd->vd_ashift = highbit(blksize) - 1;
dprintf("vdev_disk: Device %p ashift set to %d\n", devvp,
dvd->vd_ashift);
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* ### APPLE TODO ###
*/
#ifdef illumos
if (vd->vdev_wholedisk == 1) {
int wce = 1;
if (error == 0) {
/*
* If we have the capability to expand, we'd have
* found out via success from DKIOCGMEDIAINFO{,EXT}.
* Adjust max_psize upward accordingly since we know
* we own the whole disk now.
*/
*max_psize = capacity * blksz;
}
/*
* Since we own the whole disk, try to enable disk write
* caching. We ignore errors because it's OK if we can't do it.
*/
(void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
FKIOCTL, kcred, NULL);
}
#endif
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
/* Inform the ZIO pipeline that we are non-rotational */
vd->vdev_nonrot = B_FALSE;
if (VNOP_IOCTL(devvp, DKIOCISSOLIDSTATE, (caddr_t)&isssd, 0,
context) == 0) {
if (isssd)
vd->vdev_nonrot = B_TRUE;
}
dprintf("ZFS: vdev_disk(%s) isSSD %d\n", vd->vdev_path ? vd->vdev_path : "",
isssd);
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp) {
vnode_close(devvp, fmode, context);
dvd->vd_devvp = NULL;
}
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context)
(void) vfs_context_rele(context);
if (error) printf("ZFS: vdev_disk_open('%s') failed error %d\n",
vd->vdev_path ? vd->vdev_path : "", error);
return (error);
}
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);
}
static int
vdev_disk_open(vdev_t *vd, uint64_t *size, uint64_t *max_size, uint64_t *ashift)
{
vdev_disk_t *dvd = NULL;
vnode_t *devvp = NULLVP;
vfs_context_t context = NULL;
uint64_t blkcnt;
uint32_t blksize;
int fmode = 0;
int error = 0;
/*
* We must have a pathname, and it must be absolute.
*/
if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
return (EINVAL);
}
dvd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
if (dvd == NULL)
return ENOMEM;
/*
* When opening a disk device, we want to preserve the user's original
* intent. We always want to open the device by the path the user gave
* us, even if it is one of multiple paths to the save device. But we
* also want to be able to survive disks being removed/recabled.
* Therefore the sequence of opening devices is:
*
* 1. Try opening the device by path. For legacy pools without the
* 'whole_disk' property, attempt to fix the path by appending 's0'.
*
* 2. If the devid of the device matches the stored value, return
* success.
*
* 3. Otherwise, the device may have moved. Try opening the device
* by the devid instead.
*
*/
/* ### APPLE TODO ### */
/* ddi_devid_str_decode */
context = vfs_context_create((vfs_context_t)0);
/* Obtain an opened/referenced vnode for the device. */
error = vnode_open(vd->vdev_path, spa_mode(vd->vdev_spa), 0, 0, &devvp, context);
if (error) {
goto out;
}
if (!vnode_isblk(devvp)) {
error = ENOTBLK;
goto out;
}
/* ### APPLE TODO ### */
/* vnode_authorize devvp for KAUTH_VNODE_READ_DATA and
* KAUTH_VNODE_WRITE_DATA
*/
/*
* Disallow opening of a device that is currently in use.
* Flush out any old buffers remaining from a previous use.
*/
if ((error = vfs_mountedon(devvp))) {
goto out;
}
if (VNOP_FSYNC(devvp, MNT_WAIT, context) != 0) {
error = ENOTBLK;
goto out;
}
if ((error = buf_invalidateblks(devvp, BUF_WRITE_DATA, 0, 0))) {
goto out;
}
/*
* Determine the actual size of the device.
*/
if (VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0, context)
!= 0 || VNOP_IOCTL(devvp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt,
0, context) != 0) {
error = EINVAL;
goto out;
}
*size = blkcnt * (uint64_t)blksize;
/*
* ### APPLE TODO ###
* If we own the whole disk, try to enable disk write caching.
*/
/*
* Take the device's minimum transfer size into account.
*/
*ashift = highbit(MAX(blksize, SPA_MINBLOCKSIZE)) - 1;
/*
* Setting the vdev_ashift did in fact break the pool for import
* on ZEVO. This puts the logic into question. It appears that vdev_top
* will also then change. It then panics in space_map from metaslab_alloc
*/
//vd->vdev_ashift = *ashift;
dvd->vd_ashift = *ashift;
/*
* Clear the nowritecache bit, so that on a vdev_reopen() we will
* try again.
*/
vd->vdev_nowritecache = B_FALSE;
vd->vdev_tsd = dvd;
dvd->vd_devvp = devvp;
out:
if (error) {
if (devvp)
vnode_close(devvp, fmode, context);
if (dvd)
kmem_free(dvd, sizeof (vdev_disk_t));
/*
* Since the open has failed, vd->vdev_tsd should
* be NULL when we get here, signaling to the
* rest of the spa not to try and reopen or close this device
*/
vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
}
if (context) {
(void) vfs_context_rele(context);
}
return (error);
}
IOReturn FileNVRAM::read_buffer(char** aBuffer, uint64_t* aLength, vfs_context_t aCtx)
{
IOReturn error = 0;
struct vnode * vp;
struct vnode_attr va;
if (aCtx)
{
if ((error = vnode_open(FILE_NVRAM_PATH, (O_RDONLY | FREAD | O_NOFOLLOW), S_IRUSR, VNODE_LOOKUP_NOFOLLOW, &vp, aCtx)))
{
printf("failed opening vnode at path %s, errno %d\n", FILE_NVRAM_PATH, error);
return error;
}
else
{
if ((error = vnode_isreg(vp)) == VREG)
{
VATTR_INIT(&va);
VATTR_WANTED(&va, va_data_size); /* size in bytes of the fork managed by current vnode */
// Determine size of vnode
if ((error = vnode_getattr(vp, &va, aCtx)))
{
printf("FileNVRAM.kext: Error, failed to determine file size of %s, errno %d.\n", FILE_NVRAM_PATH, error);
}
else
{
if (aLength)
{
*aLength = va.va_data_size;
}
*aBuffer = (char *)IOMalloc((size_t)va.va_data_size);
int len = (int)va.va_data_size;
if ((error = vn_rdwr(UIO_READ, vp, *aBuffer, len, 0, UIO_SYSSPACE, IO_NOCACHE|IO_NODELOCKED|IO_UNIT, vfs_context_ucred(aCtx), (int *) 0, vfs_context_proc(aCtx))))
{
printf("FileNVRAM.kext: Error, writing to vnode(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
}
}
if ((error = vnode_close(vp, 0, aCtx)))
{
printf("FileNVRAM.kext: Error, vnode_close(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
}
}
else
{
printf("FileNVRAM.kext: Error, vnode_isreg(%s) failed with error %d!\n", FILE_NVRAM_PATH, error);
}
}
}
else
{
printf("FileNVRAM.kext: aCtx == NULL!\n");
error = 0xFFFF; // EINVAL;
}
return error;
}
