#include <sys/param.h>
#include <sys/proc_internal.h>
#include <sys/kauth.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/conf.h>
#include <sys/buf_internal.h>
#include <sys/mount_internal.h>
#include <sys/vnode_internal.h>
#include <sys/file_internal.h>
#include <sys/namei.h>
#include <sys/stat.h>
#include <sys/errno.h>
#include <sys/ioctl.h>
#include <sys/file.h>
#include <sys/user.h>
#include <sys/malloc.h>
#include <sys/disk.h>
#include <sys/uio_internal.h>
#include <sys/resource.h>
#include <miscfs/specfs/specdev.h>
#include <vfs/vfs_support.h>
#include <kern/assert.h>
#include <kern/task.h>
#include <sys/kdebug.h>
extern dev_t chrtoblk(dev_t dev);
extern int iskmemdev(dev_t dev);
extern int bpfkqfilter(dev_t dev, struct knote *kn);
extern int ptsd_kqfilter(dev_t dev, struct knote *kn);
struct vnode *speclisth[SPECHSZ];
char devopn[] = "devopn";
char devio[] = "devio";
char devwait[] = "devwait";
char devin[] = "devin";
char devout[] = "devout";
char devioc[] = "devioc";
char devcls[] = "devcls";
#define VOPFUNC int (*)(void *)
int (**spec_vnodeop_p)(void *);
struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
{ &vnop_default_desc, (VOPFUNC)vn_default_error },
{ &vnop_lookup_desc, (VOPFUNC)spec_lookup },
{ &vnop_create_desc, (VOPFUNC)err_create },
{ &vnop_mknod_desc, (VOPFUNC)err_mknod },
{ &vnop_open_desc, (VOPFUNC)spec_open },
{ &vnop_close_desc, (VOPFUNC)spec_close },
{ &vnop_access_desc, (VOPFUNC)spec_access },
{ &vnop_getattr_desc, (VOPFUNC)spec_getattr },
{ &vnop_setattr_desc, (VOPFUNC)spec_setattr },
{ &vnop_read_desc, (VOPFUNC)spec_read },
{ &vnop_write_desc, (VOPFUNC)spec_write },
{ &vnop_ioctl_desc, (VOPFUNC)spec_ioctl },
{ &vnop_select_desc, (VOPFUNC)spec_select },
{ &vnop_revoke_desc, (VOPFUNC)nop_revoke },
{ &vnop_mmap_desc, (VOPFUNC)err_mmap },
{ &vnop_fsync_desc, (VOPFUNC)spec_fsync },
{ &vnop_remove_desc, (VOPFUNC)err_remove },
{ &vnop_link_desc, (VOPFUNC)err_link },
{ &vnop_rename_desc, (VOPFUNC)err_rename },
{ &vnop_mkdir_desc, (VOPFUNC)err_mkdir },
{ &vnop_rmdir_desc, (VOPFUNC)err_rmdir },
{ &vnop_symlink_desc, (VOPFUNC)err_symlink },
{ &vnop_readdir_desc, (VOPFUNC)err_readdir },
{ &vnop_readlink_desc, (VOPFUNC)err_readlink },
{ &vnop_inactive_desc, (VOPFUNC)nop_inactive },
{ &vnop_reclaim_desc, (VOPFUNC)nop_reclaim },
{ &vnop_strategy_desc, (VOPFUNC)spec_strategy },
{ &vnop_pathconf_desc, (VOPFUNC)spec_pathconf },
{ &vnop_advlock_desc, (VOPFUNC)err_advlock },
{ &vnop_bwrite_desc, (VOPFUNC)spec_bwrite },
{ &vnop_pagein_desc, (VOPFUNC)err_pagein },
{ &vnop_pageout_desc, (VOPFUNC)err_pageout },
{ &vnop_copyfile_desc, (VOPFUNC)err_copyfile },
{ &vnop_blktooff_desc, (VOPFUNC)spec_blktooff },
{ &vnop_offtoblk_desc, (VOPFUNC)spec_offtoblk },
{ &vnop_blockmap_desc, (VOPFUNC)spec_blockmap },
{ (struct vnodeop_desc*)NULL, (int(*)())NULL }
};
struct vnodeopv_desc spec_vnodeop_opv_desc =
{ &spec_vnodeop_p, spec_vnodeop_entries };
static void set_blocksize(vnode_t, dev_t);
struct _throttle_io_info_t {
struct timeval last_normal_IO_timestamp;
struct timeval last_IO_timestamp;
SInt32 numthreads_throttling;
SInt32 refcnt;
SInt32 alloc;
};
struct _throttle_io_info_t _throttle_io_info[LOWPRI_MAX_NUM_DEV];
static void throttle_info_update_internal(void *throttle_info, int flags, boolean_t isssd);
int
spec_lookup(struct vnop_lookup_args *ap)
{
*ap->a_vpp = NULL;
return (ENOTDIR);
}
static void
set_blocksize(struct vnode *vp, dev_t dev)
{
int (*size)(dev_t);
int rsize;
if ((major(dev) < nblkdev) && (size = bdevsw[major(dev)].d_psize)) {
rsize = (*size)(dev);
if (rsize <= 0)
vp->v_specsize = DEV_BSIZE;
else
vp->v_specsize = rsize;
}
else
vp->v_specsize = DEV_BSIZE;
}
void
set_fsblocksize(struct vnode *vp)
{
if (vp->v_type == VBLK) {
dev_t dev = (dev_t)vp->v_rdev;
int maj = major(dev);
if ((u_int)maj >= (u_int)nblkdev)
return;
vnode_lock(vp);
set_blocksize(vp, dev);
vnode_unlock(vp);
}
}
int
spec_open(struct vnop_open_args *ap)
{
struct proc *p = vfs_context_proc(ap->a_context);
kauth_cred_t cred = vfs_context_ucred(ap->a_context);
struct vnode *vp = ap->a_vp;
dev_t bdev, dev = (dev_t)vp->v_rdev;
int maj = major(dev);
int error;
if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
return (ENXIO);
switch (vp->v_type) {
case VCHR:
if ((u_int)maj >= (u_int)nchrdev)
return (ENXIO);
if (cred != FSCRED && (ap->a_mode & FWRITE)) {
if (securelevel >= 2 && isdisk(dev, VCHR))
return (EPERM);
if (securelevel >= 1) {
if ((bdev = chrtoblk(dev)) != NODEV && check_mountedon(bdev, VBLK, &error))
return (error);
if (iskmemdev(dev))
return (EPERM);
}
}
if (cdevsw[maj].d_type == D_TTY) {
vnode_lock(vp);
vp->v_flag |= VISTTY;
vnode_unlock(vp);
}
devsw_lock(dev, S_IFCHR);
error = (*cdevsw[maj].d_open)(dev, ap->a_mode, S_IFCHR, p);
if (error == 0) {
vp->v_specinfo->si_opencount++;
}
devsw_unlock(dev, S_IFCHR);
if (error == 0 && cdevsw[maj].d_type == D_DISK && !vp->v_un.vu_specinfo->si_initted) {
int isssd = 0;
uint64_t throttle_mask = 0;
uint32_t devbsdunit = 0;
if (VNOP_IOCTL(vp, DKIOCGETTHROTTLEMASK, (caddr_t)&throttle_mask, 0, NULL) == 0) {
if (VNOP_IOCTL(vp, DKIOCISSOLIDSTATE, (caddr_t)&isssd, 0, ap->a_context) == 0) {
devbsdunit = num_trailing_0(throttle_mask);
vnode_lock(vp);
vp->v_un.vu_specinfo->si_isssd = isssd;
vp->v_un.vu_specinfo->si_devbsdunit = devbsdunit;
vp->v_un.vu_specinfo->si_throttle_mask = throttle_mask;
vp->v_un.vu_specinfo->si_throttleable = 1;
vp->v_un.vu_specinfo->si_initted = 1;
vnode_unlock(vp);
}
}
if (vp->v_un.vu_specinfo->si_initted == 0) {
vnode_lock(vp);
vp->v_un.vu_specinfo->si_initted = 1;
vnode_unlock(vp);
}
}
return (error);
case VBLK:
if ((u_int)maj >= (u_int)nblkdev)
return (ENXIO);
if (securelevel >= 2 && cred != FSCRED &&
(ap->a_mode & FWRITE) && bdevsw[maj].d_type == D_DISK)
return (EPERM);
if ( (error = vfs_mountedon(vp)) )
return (error);
devsw_lock(dev, S_IFBLK);
error = (*bdevsw[maj].d_open)(dev, ap->a_mode, S_IFBLK, p);
if (!error) {
vp->v_specinfo->si_opencount++;
}
devsw_unlock(dev, S_IFBLK);
if (!error) {
u_int64_t blkcnt;
u_int32_t blksize;
int setsize = 0;
u_int32_t size512 = 512;
if (!VNOP_IOCTL(vp, DKIOCGETBLOCKSIZE, (caddr_t)&blksize, 0, ap->a_context)) {
if (!VNOP_IOCTL(vp, DKIOCSETBLOCKSIZE, (caddr_t)&size512, FWRITE, ap->a_context)) {
if (!VNOP_IOCTL(vp, DKIOCGETBLOCKCOUNT, (caddr_t)&blkcnt, 0, ap->a_context)) {
setsize = 1;
}
}
if (VNOP_IOCTL(vp, DKIOCSETBLOCKSIZE, (caddr_t)&blksize, FWRITE, ap->a_context))
error = ENXIO;
}
vnode_lock(vp);
set_blocksize(vp, dev);
if (setsize)
vp->v_specdevsize = blkcnt * (u_int64_t)size512;
else
vp->v_specdevsize = (u_int64_t)0;
vnode_unlock(vp);
}
return(error);
default:
panic("spec_open type");
}
return (0);
}
int
spec_read(struct vnop_read_args *ap)
{
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct buf *bp;
daddr64_t bn, nextbn;
long bsize, bscale;
int devBlockSize=0;
int n, on;
int error = 0;
dev_t dev;
#if DIAGNOSTIC
if (uio->uio_rw != UIO_READ)
panic("spec_read mode");
if (UIO_SEG_IS_USER_SPACE(uio->uio_segflg))
panic("spec_read proc");
#endif
if (uio_resid(uio) == 0)
return (0);
switch (vp->v_type) {
case VCHR:
if (cdevsw[major(vp->v_rdev)].d_type == D_DISK && vp->v_un.vu_specinfo->si_throttleable) {
struct _throttle_io_info_t *throttle_info;
throttle_info = &_throttle_io_info[vp->v_un.vu_specinfo->si_devbsdunit];
throttle_info_update_internal(throttle_info, 0, vp->v_un.vu_specinfo->si_isssd);
}
error = (*cdevsw[major(vp->v_rdev)].d_read)
(vp->v_rdev, uio, ap->a_ioflag);
return (error);
case VBLK:
if (uio->uio_offset < 0)
return (EINVAL);
dev = vp->v_rdev;
devBlockSize = vp->v_specsize;
if (devBlockSize > PAGE_SIZE)
return (EINVAL);
bscale = PAGE_SIZE / devBlockSize;
bsize = bscale * devBlockSize;
do {
on = uio->uio_offset % bsize;
bn = (daddr64_t)((uio->uio_offset / devBlockSize) &~ (bscale - 1));
if (vp->v_speclastr + bscale == bn) {
nextbn = bn + bscale;
error = buf_breadn(vp, bn, (int)bsize, &nextbn,
(int *)&bsize, 1, NOCRED, &bp);
} else
error = buf_bread(vp, bn, (int)bsize, NOCRED, &bp);
vnode_lock(vp);
vp->v_speclastr = bn;
vnode_unlock(vp);
n = bsize - buf_resid(bp);
if ((on > n) || error) {
if (!error)
error = EINVAL;
buf_brelse(bp);
return (error);
}
n = min((unsigned)(n - on), uio_resid(uio));
error = uiomove((char *)buf_dataptr(bp) + on, n, uio);
if (n + on == bsize)
buf_markaged(bp);
buf_brelse(bp);
} while (error == 0 && uio_resid(uio) > 0 && n != 0);
return (error);
default:
panic("spec_read type");
}
return (0);
}
int
spec_write(struct vnop_write_args *ap)
{
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct buf *bp;
daddr64_t bn;
int bsize, blkmask, bscale;
int io_sync;
int devBlockSize=0;
int n, on;
int error = 0;
dev_t dev;
#if DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("spec_write mode");
if (UIO_SEG_IS_USER_SPACE(uio->uio_segflg))
panic("spec_write proc");
#endif
switch (vp->v_type) {
case VCHR:
if (cdevsw[major(vp->v_rdev)].d_type == D_DISK && vp->v_un.vu_specinfo->si_throttleable) {
struct _throttle_io_info_t *throttle_info;
throttle_info = &_throttle_io_info[vp->v_un.vu_specinfo->si_devbsdunit];
throttle_info_update_internal(throttle_info, 0, vp->v_un.vu_specinfo->si_isssd);
microuptime(&throttle_info->last_IO_timestamp);
}
error = (*cdevsw[major(vp->v_rdev)].d_write)
(vp->v_rdev, uio, ap->a_ioflag);
return (error);
case VBLK:
if (uio_resid(uio) == 0)
return (0);
if (uio->uio_offset < 0)
return (EINVAL);
io_sync = (ap->a_ioflag & IO_SYNC);
dev = (vp->v_rdev);
devBlockSize = vp->v_specsize;
if (devBlockSize > PAGE_SIZE)
return(EINVAL);
bscale = PAGE_SIZE / devBlockSize;
blkmask = bscale - 1;
bsize = bscale * devBlockSize;
do {
bn = (daddr64_t)((uio->uio_offset / devBlockSize) &~ blkmask);
on = uio->uio_offset % bsize;
n = min((unsigned)(bsize - on), uio_resid(uio));
if (n == bsize &&
vp->v_specdevsize != (u_int64_t)0 &&
(uio->uio_offset + (u_int64_t)n) > vp->v_specdevsize) {
n = (uio->uio_offset + (u_int64_t)n) - vp->v_specdevsize;
}
if (n == bsize)
bp = buf_getblk(vp, bn, bsize, 0, 0, BLK_WRITE);
else
error = (int)buf_bread(vp, bn, bsize, NOCRED, &bp);
if (!error)
error = (int)buf_error(bp);
if (error) {
buf_brelse(bp);
return (error);
}
n = min(n, bsize - buf_resid(bp));
error = uiomove((char *)buf_dataptr(bp) + on, n, uio);
if (error) {
buf_brelse(bp);
return (error);
}
buf_markaged(bp);
if (io_sync)
error = buf_bwrite(bp);
else {
if ((n + on) == bsize)
error = buf_bawrite(bp);
else
error = buf_bdwrite(bp);
}
} while (error == 0 && uio_resid(uio) > 0 && n != 0);
return (error);
default:
panic("spec_write type");
}
return (0);
}
int
spec_ioctl(struct vnop_ioctl_args *ap)
{
proc_t p = vfs_context_proc(ap->a_context);
dev_t dev = ap->a_vp->v_rdev;
int retval = 0;
KERNEL_DEBUG_CONSTANT(FSDBG_CODE(DBG_IOCTL, 0) | DBG_FUNC_START,
(unsigned int)dev, (unsigned int)ap->a_command, (unsigned int)ap->a_fflag, (unsigned int)ap->a_vp->v_type, 0);
switch (ap->a_vp->v_type) {
case VCHR:
retval = (*cdevsw[major(dev)].d_ioctl)(dev, ap->a_command, ap->a_data,
ap->a_fflag, p);
break;
case VBLK:
retval = (*bdevsw[major(dev)].d_ioctl)(dev, ap->a_command, ap->a_data,
ap->a_fflag, p);
break;
default:
panic("spec_ioctl");
}
KERNEL_DEBUG_CONSTANT(FSDBG_CODE(DBG_IOCTL, 0) | DBG_FUNC_END,
(unsigned int)dev, (unsigned int)ap->a_command, (unsigned int)ap->a_fflag, retval, 0);
return (retval);
}
int
spec_select(struct vnop_select_args *ap)
{
proc_t p = vfs_context_proc(ap->a_context);
dev_t dev;
switch (ap->a_vp->v_type) {
default:
return (1);
case VCHR:
dev = ap->a_vp->v_rdev;
return (*cdevsw[major(dev)].d_select)(dev, ap->a_which, ap->a_wql, p);
}
}
static int filt_specattach(struct knote *kn);
int
spec_kqfilter(vnode_t vp, struct knote *kn)
{
dev_t dev;
int err = EINVAL;
dev = vnode_specrdev(vp);
if (vnode_istty(vp)) {
err = filt_specattach(kn);
} else {
err = bpfkqfilter(dev, kn);
}
return err;
}
int
spec_fsync_internal(vnode_t vp, int waitfor, __unused vfs_context_t context)
{
if (vp->v_type == VCHR)
return (0);
buf_flushdirtyblks(vp, (waitfor == MNT_WAIT || waitfor == MNT_DWAIT), 0, "spec_fsync");
return (0);
}
int
spec_fsync(struct vnop_fsync_args *ap)
{
return spec_fsync_internal(ap->a_vp, ap->a_waitfor, ap->a_context);
}
extern int hard_throttle_on_root;
void IOSleep(int);
#define LOWPRI_INITIAL_WINDOW_MSECS 100
#define LOWPRI_WINDOW_MSECS_INC 50
#define LOWPRI_MAX_WINDOW_MSECS 200
#define LOWPRI_MAX_WAITING_MSECS 200
#if CONFIG_EMBEDDED
#define LOWPRI_SLEEP_INTERVAL 5
#else
#define LOWPRI_SLEEP_INTERVAL 2
#endif
int lowpri_IO_initial_window_msecs = LOWPRI_INITIAL_WINDOW_MSECS;
int lowpri_IO_window_msecs_inc = LOWPRI_WINDOW_MSECS_INC;
int lowpri_max_window_msecs = LOWPRI_MAX_WINDOW_MSECS;
int lowpri_max_waiting_msecs = LOWPRI_MAX_WAITING_MSECS;
#if 0
#define DEBUG_ALLOC_THROTTLE_INFO(format, debug_info, args...) \
do { \
if ((debug_info)->alloc) \
printf("%s: "format, __FUNCTION__, ## args); \
} while(0)
#else
#define DEBUG_ALLOC_THROTTLE_INFO(format, debug_info, args...)
#endif
SYSCTL_INT(_debug, OID_AUTO, lowpri_IO_initial_window_msecs, CTLFLAG_RW | CTLFLAG_LOCKED, &lowpri_IO_initial_window_msecs, LOWPRI_INITIAL_WINDOW_MSECS, "");
SYSCTL_INT(_debug, OID_AUTO, lowpri_IO_window_inc, CTLFLAG_RW | CTLFLAG_LOCKED, &lowpri_IO_window_msecs_inc, LOWPRI_INITIAL_WINDOW_MSECS, "");
SYSCTL_INT(_debug, OID_AUTO, lowpri_max_window_msecs, CTLFLAG_RW | CTLFLAG_LOCKED, &lowpri_max_window_msecs, LOWPRI_INITIAL_WINDOW_MSECS, "");
SYSCTL_INT(_debug, OID_AUTO, lowpri_max_waiting_msecs, CTLFLAG_RW | CTLFLAG_LOCKED, &lowpri_max_waiting_msecs, LOWPRI_INITIAL_WINDOW_MSECS, "");
int
num_trailing_0(uint64_t n)
{
if (n == 0)
return sizeof(n) * 8;
int count = 0;
while (!ISSET(n, 1)) {
n >>= 1;
++count;
}
return count;
}
static int
throttle_info_rel(struct _throttle_io_info_t *info)
{
SInt32 oldValue = OSDecrementAtomic(&info->refcnt);
DEBUG_ALLOC_THROTTLE_INFO("refcnt = %d info = %p\n",
info, (int)(oldValue -1), info );
if (oldValue == 0)
panic("throttle info ref cnt went negative!");
if ((info->refcnt == 0) && (info->alloc)) {
DEBUG_ALLOC_THROTTLE_INFO("Freeing info = %p\n", info, info );
FREE(info, M_TEMP);
}
return oldValue;
}
static SInt32
throttle_info_ref(struct _throttle_io_info_t *info)
{
SInt32 oldValue = OSIncrementAtomic(&info->refcnt);
DEBUG_ALLOC_THROTTLE_INFO("refcnt = %d info = %p\n",
info, (int)(oldValue -1), info );
if (info->alloc && (oldValue == 0))
panic("Taking a reference without calling create throttle info!\n");
return oldValue;
}
void *
throttle_info_create(void)
{
struct _throttle_io_info_t *info;
MALLOC(info, struct _throttle_io_info_t *, sizeof(*info), M_TEMP, M_ZERO | M_WAITOK);
if (info == NULL)
return NULL;
DEBUG_ALLOC_THROTTLE_INFO("Creating info = %p\n", info, info );
info->alloc = TRUE;
OSIncrementAtomic(&info->refcnt);
return info;
}
void
throttle_info_release(void *throttle_info)
{
DEBUG_ALLOC_THROTTLE_INFO("Releaseing info = %p\n",
(struct _throttle_io_info_t *)throttle_info,
(struct _throttle_io_info_t *)throttle_info);
if (throttle_info)
throttle_info_rel(throttle_info);
}
void
throttle_info_mount_ref(mount_t mp, void *throttle_info)
{
if ((throttle_info == NULL) || (mp == NULL))
return;
throttle_info_ref(throttle_info);
if (mp->mnt_throttle_info)
throttle_info_rel(mp->mnt_throttle_info);
mp->mnt_throttle_info = throttle_info;
}
int
throttle_info_ref_by_mask(uint64_t throttle_mask,
throttle_info_handle_t *throttle_info_handle)
{
int dev_index;
struct _throttle_io_info_t *info;
if (throttle_info_handle == NULL)
return EINVAL;
dev_index = num_trailing_0(throttle_mask);
info = &_throttle_io_info[dev_index];
throttle_info_ref(info);
*(struct _throttle_io_info_t**)throttle_info_handle = info;
return 0;
}
void
throttle_info_rel_by_mask(throttle_info_handle_t throttle_info_handle)
{
throttle_info_rel((struct _throttle_io_info_t*)throttle_info_handle);
}
void
throttle_info_mount_rel(mount_t mp)
{
if (mp->mnt_throttle_info)
throttle_info_rel(mp->mnt_throttle_info);
mp->mnt_throttle_info = NULL;
}
void
throttle_info_get_last_io_time(mount_t mp, struct timeval *tv)
{
struct _throttle_io_info_t *info;
if (mp == NULL)
info = &_throttle_io_info[LOWPRI_MAX_NUM_DEV - 1];
else if (mp->mnt_throttle_info == NULL)
info = &_throttle_io_info[mp->mnt_devbsdunit];
else
info = mp->mnt_throttle_info;
*tv = info->last_IO_timestamp;
}
void
update_last_io_time(mount_t mp)
{
struct _throttle_io_info_t *info;
if (mp == NULL)
info = &_throttle_io_info[LOWPRI_MAX_NUM_DEV - 1];
else if (mp->mnt_throttle_info == NULL)
info = &_throttle_io_info[mp->mnt_devbsdunit];
else
info = mp->mnt_throttle_info;
microuptime(&info->last_IO_timestamp);
}
#if CONFIG_EMBEDDED
int throttle_get_io_policy(struct uthread **ut)
{
int policy = IOPOL_DEFAULT;
proc_t p = current_proc();
*ut = get_bsdthread_info(current_thread());
if (p != NULL)
policy = p->p_iopol_disk;
if (*ut != NULL) {
if ((*ut)->uu_iopol_disk != IOPOL_DEFAULT)
policy = (*ut)->uu_iopol_disk;
}
return policy;
}
#else
int throttle_get_io_policy(__unused struct uthread **ut)
{
*ut = get_bsdthread_info(current_thread());
return (proc_get_task_selfdiskacc());
}
#endif
static int
throttle_io_will_be_throttled_internal(int lowpri_window_msecs, void * throttle_info)
{
struct _throttle_io_info_t *info = throttle_info;
struct timeval elapsed;
int elapsed_msecs;
int policy;
struct uthread *ut;
policy = throttle_get_io_policy(&ut);
if (ut->uu_throttle_bc == FALSE && policy != IOPOL_THROTTLE)
return (0);
microuptime(&elapsed);
timevalsub(&elapsed, &info->last_normal_IO_timestamp);
elapsed_msecs = elapsed.tv_sec * 1000 + elapsed.tv_usec / 1000;
if (lowpri_window_msecs == -1) lowpri_window_msecs = lowpri_max_waiting_msecs;
return elapsed_msecs < lowpri_window_msecs;
}
int
throttle_io_will_be_throttled(int lowpri_window_msecs, mount_t mp)
{
void *info;
if (mp == NULL)
info = &_throttle_io_info[LOWPRI_MAX_NUM_DEV - 1];
else if (mp->mnt_throttle_info == NULL)
info = &_throttle_io_info[mp->mnt_devbsdunit];
else
info = mp->mnt_throttle_info;
return throttle_io_will_be_throttled_internal(lowpri_window_msecs, info);
}
uint32_t
throttle_lowpri_io(int sleep_amount)
{
int sleep_cnt = 0;
int numthreads_throttling;
int max_try_num;
struct uthread *ut;
struct _throttle_io_info_t *info;
int max_waiting_msecs;
ut = get_bsdthread_info(current_thread());
if ((ut->uu_lowpri_window == 0) || (ut->uu_throttle_info == NULL))
goto done;
info = ut->uu_throttle_info;
if (sleep_amount != 0) {
#if CONFIG_EMBEDDED
max_waiting_msecs = lowpri_max_waiting_msecs;
#else
if (ut->uu_throttle_isssd == TRUE)
max_waiting_msecs = lowpri_max_waiting_msecs / 100;
else
max_waiting_msecs = lowpri_max_waiting_msecs;
#endif
if (max_waiting_msecs < LOWPRI_SLEEP_INTERVAL)
max_waiting_msecs = LOWPRI_SLEEP_INTERVAL;
numthreads_throttling = info->numthreads_throttling + MIN(10, MAX(1, sleep_amount)) - 1;
max_try_num = max_waiting_msecs / LOWPRI_SLEEP_INTERVAL * MAX(1, numthreads_throttling);
for (sleep_cnt = 0; sleep_cnt < max_try_num; sleep_cnt++) {
if (throttle_io_will_be_throttled_internal(ut->uu_lowpri_window, info)) {
if (sleep_cnt == 0) {
KERNEL_DEBUG_CONSTANT((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_START,
ut->uu_lowpri_window, max_try_num, numthreads_throttling, 0, 0);
}
IOSleep(LOWPRI_SLEEP_INTERVAL);
DEBUG_ALLOC_THROTTLE_INFO("sleeping because of info = %p\n", info, info );
} else {
break;
}
}
if (sleep_cnt) {
KERNEL_DEBUG_CONSTANT((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_END,
ut->uu_lowpri_window, sleep_cnt, 0, 0, 0);
}
}
SInt32 oldValue;
oldValue = OSDecrementAtomic(&info->numthreads_throttling);
if (oldValue <= 0) {
panic("%s: numthreads negative", __func__);
}
done:
ut->uu_lowpri_window = 0;
if (ut->uu_throttle_info)
throttle_info_rel(ut->uu_throttle_info);
ut->uu_throttle_info = NULL;
ut->uu_throttle_bc = FALSE;
return (sleep_cnt * LOWPRI_SLEEP_INTERVAL);
}
void throttle_set_thread_io_policy(int policy)
{
#if !CONFIG_EMBEDDED
proc_apply_thread_selfdiskacc(policy);
#else
struct uthread *ut;
ut = get_bsdthread_info(current_thread());
ut->uu_iopol_disk = policy;
#endif
}
static
void throttle_info_reset_window(struct uthread *ut)
{
struct _throttle_io_info_t *info;
info = ut->uu_throttle_info;
OSDecrementAtomic(&info->numthreads_throttling);
throttle_info_rel(info);
ut->uu_throttle_info = NULL;
ut->uu_lowpri_window = 0;
}
static
void throttle_info_set_initial_window(struct uthread *ut, struct _throttle_io_info_t *info, boolean_t isssd, boolean_t BC_throttle)
{
SInt32 oldValue;
ut->uu_throttle_info = info;
throttle_info_ref(info);
DEBUG_ALLOC_THROTTLE_INFO("updating info = %p\n", info, info );
oldValue = OSIncrementAtomic(&info->numthreads_throttling);
if (oldValue < 0) {
panic("%s: numthreads negative", __func__);
}
ut->uu_lowpri_window = lowpri_IO_initial_window_msecs;
ut->uu_lowpri_window += lowpri_IO_window_msecs_inc * oldValue;
ut->uu_throttle_isssd = isssd;
ut->uu_throttle_bc = BC_throttle;
}
static
void throttle_info_update_internal(void *throttle_info, int flags, boolean_t isssd)
{
struct _throttle_io_info_t *info = throttle_info;
struct uthread *ut;
int policy;
int is_throttleable_io = 0;
int is_passive_io = 0;
if (!lowpri_IO_initial_window_msecs || (info == NULL))
return;
policy = throttle_get_io_policy(&ut);
switch (policy) {
case IOPOL_DEFAULT:
case IOPOL_NORMAL:
break;
case IOPOL_THROTTLE:
is_throttleable_io = 1;
break;
case IOPOL_PASSIVE:
is_passive_io = 1;
break;
default:
printf("unknown I/O policy %d", policy);
break;
}
if (!is_throttleable_io && ISSET(flags, B_PASSIVE))
is_passive_io |= 1;
if (!is_throttleable_io) {
if (!is_passive_io){
microuptime(&info->last_normal_IO_timestamp);
}
} else if (ut) {
if (ut->uu_lowpri_window == 0)
throttle_info_set_initial_window(ut, info, isssd, FALSE);
else {
if (ut->uu_throttle_info != info) {
struct _throttle_io_info_t *old_info = ut->uu_throttle_info;
OSDecrementAtomic(&old_info->numthreads_throttling);
OSIncrementAtomic(&info->numthreads_throttling);
DEBUG_ALLOC_THROTTLE_INFO("switching from info = %p\n", old_info, old_info );
DEBUG_ALLOC_THROTTLE_INFO("switching to info = %p\n", info, info );
throttle_info_rel(ut->uu_throttle_info);
ut->uu_throttle_info = info;
throttle_info_ref(ut->uu_throttle_info);
}
int numthreads = MAX(1, info->numthreads_throttling);
ut->uu_lowpri_window += lowpri_IO_window_msecs_inc * numthreads;
if (ut->uu_lowpri_window > lowpri_max_window_msecs * numthreads)
ut->uu_lowpri_window = lowpri_max_window_msecs * numthreads;
if (isssd == FALSE) {
ut->uu_throttle_isssd = FALSE;
}
}
}
}
void throttle_info_update(void *throttle_info, int flags)
{
throttle_info_update_internal(throttle_info, flags, FALSE);
}
void throttle_info_update_by_mask(void *throttle_info_handle, int flags)
{
void *throttle_info = throttle_info_handle;
throttle_info_update(throttle_info, flags);
}
extern int ignore_is_ssd;
int
spec_strategy(struct vnop_strategy_args *ap)
{
buf_t bp;
int bflags;
int policy;
dev_t bdev;
uthread_t ut;
mount_t mp;
int strategy_ret;
struct _throttle_io_info_t *throttle_info;
boolean_t isssd = FALSE;
bp = ap->a_bp;
bdev = buf_device(bp);
mp = buf_vnode(bp)->v_mount;
policy = throttle_get_io_policy(&ut);
if (policy == IOPOL_THROTTLE) {
bp->b_flags |= B_THROTTLED_IO;
bp->b_attr.ba_flags |= BA_THROTTLED_IO;
bp->b_flags &= ~B_PASSIVE;
} else if (policy == IOPOL_PASSIVE)
bp->b_flags |= B_PASSIVE;
bflags = bp->b_flags;
if (kdebug_enable) {
int code = 0;
if (bflags & B_READ)
code |= DKIO_READ;
if (bflags & B_ASYNC)
code |= DKIO_ASYNC;
if (bflags & B_META)
code |= DKIO_META;
else if (bflags & B_PAGEIO)
code |= DKIO_PAGING;
if (bflags & B_THROTTLED_IO)
code |= DKIO_THROTTLE;
else if (bflags & B_PASSIVE)
code |= DKIO_PASSIVE;
KERNEL_DEBUG_CONSTANT(FSDBG_CODE(DBG_DKRW, code) | DBG_FUNC_NONE,
bp, bdev, (int)buf_blkno(bp), buf_count(bp), 0);
}
if (((bflags & (B_IOSTREAMING | B_PAGEIO | B_READ)) == (B_PAGEIO | B_READ)) &&
mp && (mp->mnt_kern_flag & MNTK_ROOTDEV))
hard_throttle_on_root = 1;
if (mp != NULL) {
if ((mp->mnt_kern_flag & MNTK_SSD) && !ignore_is_ssd)
isssd = TRUE;
throttle_info = &_throttle_io_info[mp->mnt_devbsdunit];
} else
throttle_info = &_throttle_io_info[LOWPRI_MAX_NUM_DEV - 1];
throttle_info_update_internal(throttle_info, bflags, isssd);
if ((bflags & B_READ) == 0) {
microuptime(&throttle_info->last_IO_timestamp);
if (mp) {
INCR_PENDING_IO(buf_count(bp), mp->mnt_pending_write_size);
}
} else if (mp) {
INCR_PENDING_IO(buf_count(bp), mp->mnt_pending_read_size);
}
#define IO_SATISFIED_BY_CACHE ((int)0xcafefeed)
#define IO_SHOULD_BE_THROTTLED ((int)0xcafebeef)
typedef int strategy_fcn_ret_t(struct buf *bp);
strategy_ret = (*(strategy_fcn_ret_t*)bdevsw[major(bdev)].d_strategy)(bp);
if ((IO_SATISFIED_BY_CACHE == strategy_ret) && (ut->uu_lowpri_window != 0) && (ut->uu_throttle_info != NULL)) {
throttle_info_reset_window(ut);
} else if ((IO_SHOULD_BE_THROTTLED == strategy_ret) && (ut->uu_lowpri_window == 0) && (ut->uu_throttle_info == NULL)) {
throttle_info_set_initial_window(ut, throttle_info, isssd, TRUE);
}
return (0);
}
int
spec_blockmap(__unused struct vnop_blockmap_args *ap)
{
return (ENOTSUP);
}
int
spec_close(struct vnop_close_args *ap)
{
struct vnode *vp = ap->a_vp;
dev_t dev = vp->v_rdev;
int error = 0;
int flags = ap->a_fflag;
struct proc *p = vfs_context_proc(ap->a_context);
struct session *sessp;
int do_rele = 0;
switch (vp->v_type) {
case VCHR:
sessp = proc_session(p);
if (sessp != SESSION_NULL) {
if ((vcount(vp) == 1) &&
(vp == sessp->s_ttyvp)) {
session_lock(sessp);
if (vp == sessp->s_ttyvp) {
sessp->s_ttyvp = NULL;
sessp->s_ttyvid = 0;
sessp->s_ttyp = TTY_NULL;
sessp->s_ttypgrpid = NO_PID;
do_rele = 1;
}
session_unlock(sessp);
if (do_rele) {
vnode_rele(vp);
}
}
session_rele(sessp);
}
devsw_lock(dev, S_IFCHR);
vp->v_specinfo->si_opencount--;
if (vp->v_specinfo->si_opencount < 0) {
panic("Negative open count?");
}
if ((vcount(vp) > 0) && ((flags & IO_REVOKE) == 0)) {
devsw_unlock(dev, S_IFCHR);
return (0);
}
error = cdevsw[major(dev)].d_close(dev, flags, S_IFCHR, p);
devsw_unlock(dev, S_IFCHR);
break;
case VBLK:
devsw_lock(dev, S_IFBLK);
if (vcount(vp) > 1) {
vp->v_specinfo->si_opencount--;
devsw_unlock(dev, S_IFBLK);
return (0);
}
devsw_unlock(dev, S_IFBLK);
if ((error = spec_fsync_internal(vp, MNT_WAIT, ap->a_context)))
return (error);
error = buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0);
if (error)
return (error);
devsw_lock(dev, S_IFBLK);
vp->v_specinfo->si_opencount--;
if (vp->v_specinfo->si_opencount < 0) {
panic("Negative open count?");
}
if (vcount(vp) > 0) {
devsw_unlock(dev, S_IFBLK);
return (0);
}
error = bdevsw[major(dev)].d_close(dev, flags, S_IFBLK, p);
devsw_unlock(dev, S_IFBLK);
break;
default:
panic("spec_close: not special");
return(EBADF);
}
return error;
}
int
spec_pathconf(struct vnop_pathconf_args *ap)
{
switch (ap->a_name) {
case _PC_LINK_MAX:
*ap->a_retval = LINK_MAX;
return (0);
case _PC_MAX_CANON:
*ap->a_retval = MAX_CANON;
return (0);
case _PC_MAX_INPUT:
*ap->a_retval = MAX_INPUT;
return (0);
case _PC_PIPE_BUF:
*ap->a_retval = PIPE_BUF;
return (0);
case _PC_CHOWN_RESTRICTED:
*ap->a_retval = 200112;
return (0);
case _PC_VDISABLE:
*ap->a_retval = _POSIX_VDISABLE;
return (0);
default:
return (EINVAL);
}
}
int
spec_ebadf(__unused void *dummy)
{
return (EBADF);
}
int
spec_blktooff(struct vnop_blktooff_args *ap)
{
struct vnode *vp = ap->a_vp;
switch (vp->v_type) {
case VCHR:
*ap->a_offset = (off_t)-1;
return (ENOTSUP);
case VBLK:
printf("spec_blktooff: not implemented for VBLK\n");
*ap->a_offset = (off_t)-1;
return (ENOTSUP);
default:
panic("spec_blktooff type");
}
return (0);
}
int
spec_offtoblk(struct vnop_offtoblk_args *ap)
{
struct vnode *vp = ap->a_vp;
switch (vp->v_type) {
case VCHR:
*ap->a_lblkno = (daddr64_t)-1;
return (ENOTSUP);
case VBLK:
printf("spec_offtoblk: not implemented for VBLK\n");
*ap->a_lblkno = (daddr64_t)-1;
return (ENOTSUP);
default:
panic("spec_offtoblk type");
}
return (0);
}
static void filt_specdetach(struct knote *kn);
static int filt_spec(struct knote *kn, long hint);
static unsigned filt_specpeek(struct knote *kn);
struct filterops spec_filtops = {
.f_isfd = 1,
.f_attach = filt_specattach,
.f_detach = filt_specdetach,
.f_event = filt_spec,
.f_peek = filt_specpeek
};
static int
filter_to_seltype(int16_t filter)
{
switch (filter) {
case EVFILT_READ:
return FREAD;
case EVFILT_WRITE:
return FWRITE;
break;
default:
panic("filt_to_seltype(): invalid filter %d\n", filter);
return 0;
}
}
static int
filt_specattach(struct knote *kn)
{
vnode_t vp;
dev_t dev;
vp = (vnode_t)kn->kn_fp->f_fglob->fg_data;
assert(vnode_ischr(vp));
dev = vnode_specrdev(vp);
if (major(dev) > nchrdev) {
return ENXIO;
}
if ((cdevsw_flags[major(dev)] & CDEVSW_SELECT_KQUEUE) == 0) {
return EINVAL;
}
kn->kn_hook = wait_queue_link_allocate();
if (kn->kn_hook == NULL) {
return EAGAIN;
}
kn->kn_fop = &spec_filtops;
kn->kn_hookid = vnode_vid(vp);
knote_markstayqueued(kn);
return 0;
}
static void
filt_specdetach(struct knote *kn)
{
kern_return_t ret;
ret = wait_queue_set_unlink_one(kn->kn_kq->kq_wqs, kn->kn_hook);
if (ret != KERN_SUCCESS) {
panic("filt_specdetach(): failed to unlink wait queue link.");
}
(void)wait_queue_link_free(kn->kn_hook);
kn->kn_hook = NULL;
kn->kn_status &= ~KN_STAYQUEUED;
}
static int
filt_spec(struct knote *kn, long hint)
{
vnode_t vp;
uthread_t uth;
wait_queue_set_t old_wqs;
vfs_context_t ctx;
int selres;
int error;
int use_offset;
dev_t dev;
uint64_t flags;
assert(kn->kn_hook != NULL);
if (hint != 0) {
panic("filt_spec(): nonzero hint?");
}
uth = get_bsdthread_info(current_thread());
ctx = vfs_context_current();
vp = (vnode_t)kn->kn_fp->f_fglob->fg_data;
error = vnode_getwithvid(vp, kn->kn_hookid);
if (error != 0) {
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
return 1;
}
dev = vnode_specrdev(vp);
flags = cdevsw_flags[major(dev)];
use_offset = ((flags & CDEVSW_USE_OFFSET) != 0);
assert((flags & CDEVSW_SELECT_KQUEUE) != 0);
old_wqs = uth->uu_wqset;
uth->uu_wqset = kn->kn_kq->kq_wqs;
selres = VNOP_SELECT(vp, filter_to_seltype(kn->kn_filter), 0, kn->kn_hook, ctx);
uth->uu_wqset = old_wqs;
if (use_offset) {
if (kn->kn_fp->f_fglob->fg_offset >= (uint32_t)selres) {
kn->kn_data = 0;
} else {
kn->kn_data = ((uint32_t)selres) - kn->kn_fp->f_fglob->fg_offset;
}
} else {
kn->kn_data = selres;
}
vnode_put(vp);
return (kn->kn_data != 0);
}
static unsigned
filt_specpeek(struct knote *kn)
{
vnode_t vp;
uthread_t uth;
wait_queue_set_t old_wqs;
vfs_context_t ctx;
int error, selres;
uth = get_bsdthread_info(current_thread());
ctx = vfs_context_current();
vp = (vnode_t)kn->kn_fp->f_fglob->fg_data;
error = vnode_getwithvid(vp, kn->kn_hookid);
if (error != 0) {
return 1;
}
old_wqs = uth->uu_wqset;
uth->uu_wqset = kn->kn_kq->kq_wqs;
selres = VNOP_SELECT(vp, filter_to_seltype(kn->kn_filter), 0, kn->kn_hook, ctx);
uth->uu_wqset = old_wqs;
vnode_put(vp);
return selres;
}