#include <sys/stat.h>
#include <sys/ioctl.h>
#include <sys/fcntl.h>
+#include <sys/fsctl.h>
#include <sys/malloc.h>
#include <sys/mman.h>
#include <sys/syslog.h>
#include <mach/mach_port.h>
#include <stdbool.h>
+#if CONFIG_MACF
+#include <security/mac_framework.h>
+#endif
+
+#define IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND 0x1
kern_return_t ipc_object_copyin(ipc_space_t, mach_port_name_t,
- mach_msg_type_name_t, ipc_port_t *);
+ mach_msg_type_name_t, ipc_port_t *, mach_port_context_t, mach_msg_guard_flags_t *, uint32_t);
void ipc_port_release_send(ipc_port_t);
struct psemnode;
extern void file_lock_init(void);
-extern kauth_scope_t kauth_scope_fileop;
+extern kauth_scope_t kauth_scope_fileop;
/* Conflict wait queue for when selects collide (opaque type) */
extern struct waitq select_conflict_queue;
+#ifndef HFS_GET_BOOT_INFO
+#define HFS_GET_BOOT_INFO (FCNTL_FS_SPECIFIC_BASE + 0x00004)
+#endif
+
+#ifndef HFS_SET_BOOT_INFO
+#define HFS_SET_BOOT_INFO (FCNTL_FS_SPECIFIC_BASE + 0x00005)
+#endif
+
+#ifndef APFSIOC_REVERT_TO_SNAPSHOT
+#define APFSIOC_REVERT_TO_SNAPSHOT _IOW('J', 1, u_int64_t)
+#endif
+
#define f_flag f_fglob->fg_flag
#define f_type f_fglob->fg_ops->fo_type
#define f_msgcount f_fglob->fg_msgcount
#define f_offset f_fglob->fg_offset
#define f_data f_fglob->fg_data
#define CHECK_ADD_OVERFLOW_INT64L(x, y) \
- (((((x) > 0) && ((y) > 0) && ((x) > LLONG_MAX - (y))) || \
- (((x) < 0) && ((y) < 0) && ((x) < LLONG_MIN - (y)))) \
- ? 1 : 0)
+ (((((x) > 0) && ((y) > 0) && ((x) > LLONG_MAX - (y))) || \
+ (((x) < 0) && ((y) < 0) && ((x) < LLONG_MIN - (y)))) \
+ ? 1 : 0)
/*
* Descriptor management.
*/
-struct fmsglist fmsghead; /* head of list of open files */
-struct fmsglist fmsg_ithead; /* head of list of open files */
-int nfiles; /* actual number of open files */
+struct fmsglist fmsghead; /* head of list of open files */
+struct fmsglist fmsg_ithead; /* head of list of open files */
+int nfiles; /* actual number of open files */
lck_grp_attr_t * file_lck_grp_attr;
* Parameters: fl Flock structure.
* cur_file_offset Current offset in the file.
*
- * Returns: 0 on Success.
+ * Returns: 0 on Success.
* EOVERFLOW on overflow.
- * EINVAL on offset less than zero.
+ * EINVAL on offset less than zero.
*/
static int
}
/* Check if end marker is beyond LLONG_MAX. */
if ((fl->l_len > 0) && (CHECK_ADD_OVERFLOW_INT64L(fl->l_start +
- cur_file_offset, fl->l_len - 1))) {
+ cur_file_offset, fl->l_len - 1))) {
return EOVERFLOW;
}
/* Check if the end marker is negative. */
if ((fl->l_len <= 0) && (fl->l_start + cur_file_offset +
- fl->l_len < 0)) {
+ fl->l_len < 0)) {
return EINVAL;
}
} else if (fl->l_whence == SEEK_SET) {
return EOVERFLOW;
}
/* Check if the end marker is negative. */
- if ((fl->l_len < 0) && fl->l_start + fl->l_len < 0) {
+ if ((fl->l_len < 0) && fl->l_start + fl->l_len < 0) {
return EINVAL;
}
}
file_lock_init(void)
{
/* allocate file lock group attribute and group */
- file_lck_grp_attr= lck_grp_attr_alloc_init();
+ file_lck_grp_attr = lck_grp_attr_alloc_init();
- file_lck_grp = lck_grp_alloc_init("file", file_lck_grp_attr);
+ file_lck_grp = lck_grp_alloc_init("file", file_lck_grp_attr);
/* Allocate file lock attribute */
file_lck_attr = lck_attr_alloc_init();
*retval = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
proc_fdunlock(p);
- return (0);
+ return 0;
}
procfdtbl_reservefd(struct proc * p, int fd)
{
p->p_fd->fd_ofiles[fd] = NULL;
- p->p_fd->fd_ofileflags[fd] |= UF_RESERVED;
+ p->p_fd->fd_ofileflags[fd] |= UF_RESERVED;
}
void
procfdtbl_markclosefd(struct proc * p, int fd)
{
- p->p_fd->fd_ofileflags[fd] |= (UF_RESERVED | UF_CLOSING);
+ p->p_fd->fd_ofileflags[fd] |= (UF_RESERVED | UF_CLOSING);
}
void
procfdtbl_releasefd(struct proc * p, int fd, struct fileproc * fp)
{
- if (fp != NULL)
- p->p_fd->fd_ofiles[fd] = fp;
- p->p_fd->fd_ofileflags[fd] &= ~UF_RESERVED;
+ if (fp != NULL) {
+ p->p_fd->fd_ofiles[fd] = fp;
+ }
+ p->p_fd->fd_ofileflags[fd] &= ~UF_RESERVED;
if ((p->p_fd->fd_ofileflags[fd] & UF_RESVWAIT) == UF_RESVWAIT) {
p->p_fd->fd_ofileflags[fd] &= ~UF_RESVWAIT;
wakeup(&p->p_fd);
void
procfdtbl_waitfd(struct proc * p, int fd)
{
- p->p_fd->fd_ofileflags[fd] |= UF_RESVWAIT;
+ p->p_fd->fd_ofileflags[fd] |= UF_RESVWAIT;
msleep(&p->p_fd, &p->p_fdmlock, PRIBIO, "ftbl_waitfd", NULL);
}
waiting = (p->p_fd->fd_ofileflags[fd] & UF_RESVWAIT);
p->p_fd->fd_ofiles[fd] = NULL;
p->p_fd->fd_ofileflags[fd] = 0;
- if ( waiting == UF_RESVWAIT) {
+ if (waiting == UF_RESVWAIT) {
wakeup(&p->p_fd);
}
}
struct filedesc *fdp = p->p_fd;
int nfd = 0;
- if (fd < fdp->fd_freefile)
+ if (fd < fdp->fd_freefile) {
fdp->fd_freefile = fd;
+ }
#if DIAGNOSTIC
- if (fd > fdp->fd_lastfile)
- panic("fdrelse: fd_lastfile inconsistent");
+ if (fd > fdp->fd_lastfile) {
+ panic("fdrelse: fd_lastfile inconsistent");
+ }
#endif
procfdtbl_clearfd(p, fd);
while ((nfd = fdp->fd_lastfile) > 0 &&
- fdp->fd_ofiles[nfd] == NULL &&
- !(fdp->fd_ofileflags[nfd] & UF_RESERVED))
+ fdp->fd_ofiles[nfd] == NULL &&
+ !(fdp->fd_ofileflags[nfd] & UF_RESERVED)) {
+ /* JMM - What about files with lingering EV_VANISHED knotes? */
fdp->fd_lastfile--;
+ }
}
uint64_t base,
int64_t len,
enum uio_seg segflg,
- off_t offset,
- int io_flg,
+ off_t offset,
+ int io_flg,
int64_t *aresid)
{
- struct fileproc *fp;
- proc_t p;
- int error = 0;
+ struct fileproc *fp;
+ proc_t p;
+ int error = 0;
int flags = 0;
int spacetype;
uio_t auio = NULL;
- char uio_buf[ UIO_SIZEOF(1) ];
+ char uio_buf[UIO_SIZEOF(1)];
struct vfs_context context = *(vfs_context_current());
bool wrote_some = false;
p = current_proc();
- error = fp_lookup(p, fd, &fp, 0);
- if (error)
- return(error);
+ error = fp_lookup(p, fd, &fp, 0);
+ if (error) {
+ return error;
+ }
if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_PIPE && fp->f_type != DTYPE_SOCKET) {
error = EINVAL;
goto out;
}
if (rw == UIO_WRITE && !(fp->f_flag & FWRITE)) {
- error = EBADF;
+ error = EBADF;
goto out;
}
if (rw == UIO_READ && !(fp->f_flag & FREAD)) {
- error = EBADF;
- goto out;
+ error = EBADF;
+ goto out;
}
context.vc_ucred = fp->f_fglob->fg_cred;
- if (UIO_SEG_IS_USER_SPACE(segflg))
+ if (UIO_SEG_IS_USER_SPACE(segflg)) {
spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
- else
+ } else {
spacetype = UIO_SYSSPACE;
+ }
auio = uio_createwithbuffer(1, offset, spacetype, rw, &uio_buf[0], sizeof(uio_buf));
uio_addiov(auio, base, len);
- if ( !(io_flg & IO_APPEND))
+ if (!(io_flg & IO_APPEND)) {
flags = FOF_OFFSET;
+ }
if (rw == UIO_WRITE) {
user_ssize_t orig_resid = uio_resid(auio);
error = fo_write(fp, auio, flags, &context);
wrote_some = uio_resid(auio) < orig_resid;
- } else
+ } else {
error = fo_read(fp, auio, flags, &context);
+ }
- if (aresid)
+ if (aresid) {
*aresid = uio_resid(auio);
- else {
- if (uio_resid(auio) && error == 0)
+ } else {
+ if (uio_resid(auio) && error == 0) {
error = EIO;
+ }
}
out:
- if (wrote_some)
- fp_drop_written(p, fd, fp);
- else
- fp_drop(p, fd, fp, 0);
+ if (wrote_some) {
+ fp_drop_written(p, fd, fp);
+ } else {
+ fp_drop(p, fd, fp, 0);
+ }
return error;
}
struct fileproc *fp;
proc_fdlock(p);
- if ( (error = fp_lookup(p, old, &fp, 1)) ) {
+ if ((error = fp_lookup(p, old, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
if (FP_ISGUARDED(fp, GUARD_DUP)) {
error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP);
(void) fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
- if ( (error = fdalloc(p, 0, &new)) ) {
+ if ((error = fdalloc(p, 0, &new))) {
fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
error = finishdup(p, fdp, old, new, 0, retval);
fp_drop(p, old, fp, 1);
new, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
}
- return (error);
+ return error;
}
/*
proc_fdlock(p);
startover:
- if ( (error = fp_lookup(p, old, &fp, 1)) ) {
+ if ((error = fp_lookup(p, old, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
if (FP_ISGUARDED(fp, GUARD_DUP)) {
error = fp_guard_exception(p, old, fp, kGUARD_EXC_DUP);
(void) fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
if (new < 0 ||
- (rlim_t)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
+ (rlim_t)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
new >= maxfiles) {
fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (old == new) {
fp_drop(p, old, fp, 1);
*retval = new;
proc_fdunlock(p);
- return (0);
+ return 0;
}
if (new < 0 || new >= fdp->fd_nfiles) {
- if ( (error = fdalloc(p, new, &i)) ) {
+ if ((error = fdalloc(p, new, &i))) {
fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
if (new != i) {
fdrelse(p, i);
}
} else {
closeit:
- while ((fdp->fd_ofileflags[new] & UF_RESERVED) == UF_RESERVED) {
- fp_drop(p, old, fp, 1);
- procfdtbl_waitfd(p, new);
+ while ((fdp->fd_ofileflags[new] & UF_RESERVED) == UF_RESERVED) {
+ fp_drop(p, old, fp, 1);
+ procfdtbl_waitfd(p, new);
#if DIAGNOSTIC
- proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
+ proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
#endif
- goto startover;
+ goto startover;
}
if ((fdp->fd_ofiles[new] != NULL) &&
new, nfp, kGUARD_EXC_CLOSE);
(void) fp_drop(p, new, nfp, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
(void)close_internal_locked(p, new, nfp, FD_DUP2RESV);
#if DIAGNOSTIC
#endif
procfdtbl_clearfd(p, new);
goto startover;
- } else {
+ } else {
#if DIAGNOSTIC
- if (fdp->fd_ofiles[new] != NULL)
+ if (fdp->fd_ofiles[new] != NULL) {
panic("dup2: no ref on fileproc %d", new);
+ }
#endif
procfdtbl_reservefd(p, new);
}
#if DIAGNOSTIC
proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
#endif
-
}
#if DIAGNOSTIC
- if (fdp->fd_ofiles[new] != 0)
+ if (fdp->fd_ofiles[new] != 0) {
panic("dup2: overwriting fd_ofiles with new %d", new);
- if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0)
+ }
+ if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0) {
panic("dup2: unreserved fileflags with new %d", new);
+ }
#endif
error = finishdup(p, fdp, old, new, 0, retval);
fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- return(error);
+ return error;
}
fcntl(proc_t p, struct fcntl_args *uap, int32_t *retval)
{
__pthread_testcancel(1);
- return(fcntl_nocancel(p, (struct fcntl_nocancel_args *)uap, retval));
+ return fcntl_nocancel(p, (struct fcntl_nocancel_args *)uap, retval);
}
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
char *pop;
- struct vnode *vp = NULLVP; /* for AUDIT_ARG() at end */
+ struct vnode *vp = NULLVP; /* for AUDIT_ARG() at end */
int i, tmp, error, error2, flg = 0;
- struct flock fl;
+ struct flock fl = {};
struct flocktimeout fltimeout;
struct timespec *timeout = NULL;
struct vfs_context context;
AUDIT_ARG(cmd, uap->cmd);
proc_fdlock(p);
- if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
context.vc_thread = current_thread();
context.vc_ucred = fp->f_cred;
is64bit = proc_is64bit(p);
if (is64bit) {
argp = uap->arg;
- }
- else {
+ } else {
/*
* Since the arg parameter is defined as a long but may be
* either a long or a pointer we must take care to handle
#if CONFIG_MACF
error = mac_file_check_fcntl(proc_ucred(p), fp->f_fglob, uap->cmd,
uap->arg);
- if (error)
+ if (error) {
goto out;
+ }
#endif
switch (uap->cmd) {
-
case F_DUPFD:
case F_DUPFD_CLOEXEC:
if (FP_ISGUARDED(fp, GUARD_DUP)) {
error = EINVAL;
goto out;
}
- if ( (error = fdalloc(p, newmin, &i)) )
+ if ((error = fdalloc(p, newmin, &i))) {
goto out;
+ }
error = finishdup(p, fdp, fd, i,
uap->cmd == F_DUPFD_CLOEXEC ? UF_EXCLOSE : 0, retval);
goto out;
case F_SETFD:
AUDIT_ARG(value32, uap->arg);
- if (uap->arg & FD_CLOEXEC)
+ if (uap->arg & FD_CLOEXEC) {
*pop |= UF_EXCLOSE;
- else {
+ } else {
if (FILEPROC_TYPE(fp) == FTYPE_GUARDED) {
error = fp_guard_exception(p,
fd, fp, kGUARD_EXC_NOCLOEXEC);
fp->f_flag |= FFLAGS(tmp) & FCNTLFLAGS;
tmp = fp->f_flag & FNONBLOCK;
error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
- if (error)
+ if (error) {
goto out;
+ }
tmp = fp->f_flag & FASYNC;
error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, &context);
- if (!error)
+ if (!error) {
goto out;
+ }
fp->f_flag &= ~FNONBLOCK;
tmp = 0;
(void)fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
error = 0;
goto out;
}
- error = fo_ioctl(fp, (int)TIOCGPGRP, (caddr_t)retval, &context);
+ error = fo_ioctl(fp, TIOCGPGRP, (caddr_t)retval, &context);
*retval = -*retval;
goto out;
AUDIT_ARG(value32, tmp);
if (fp->f_type == DTYPE_SOCKET) {
((struct socket *)fp->f_data)->so_pgid = tmp;
- error =0;
+ error = 0;
goto out;
}
if (fp->f_type == DTYPE_PIPE) {
tmp = (int)p1->p_pgrpid;
proc_rele(p1);
}
- error = fo_ioctl(fp, (int)TIOCSPGRP, (caddr_t)&tmp, &context);
+ error = fo_ioctl(fp, TIOCSPGRP, (caddr_t)&tmp, &context);
goto out;
case F_SETNOSIGPIPE:
if (fp->f_type == DTYPE_SOCKET) {
#if SOCKETS
error = sock_setsockopt((struct socket *)fp->f_data,
- SOL_SOCKET, SO_NOSIGPIPE, &tmp, sizeof (tmp));
+ SOL_SOCKET, SO_NOSIGPIPE, &tmp, sizeof(tmp));
#else
error = EINVAL;
#endif
struct fileglob *fg = fp->f_fglob;
lck_mtx_lock_spin(&fg->fg_lock);
- if (tmp)
+ if (tmp) {
fg->fg_lflags |= FG_NOSIGPIPE;
- else
- fg->fg_lflags &= FG_NOSIGPIPE;
+ } else {
+ fg->fg_lflags &= ~FG_NOSIGPIPE;
+ }
lck_mtx_unlock(&fg->fg_lock);
error = 0;
}
case F_GETNOSIGPIPE:
if (fp->f_type == DTYPE_SOCKET) {
#if SOCKETS
- int retsize = sizeof (*retval);
+ int retsize = sizeof(*retval);
error = sock_getsockopt((struct socket *)fp->f_data,
SOL_SOCKET, SO_NOSIGPIPE, retval, &retsize);
#else
#endif
} else {
*retval = (fp->f_fglob->fg_lflags & FG_NOSIGPIPE) ?
- 1 : 0;
+ 1 : 0;
error = 0;
}
goto out;
struct fileglob *fg = fp->f_fglob;
lck_mtx_lock_spin(&fg->fg_lock);
- if (fg->fg_lflags & FG_CONFINED)
+ if (fg->fg_lflags & FG_CONFINED) {
error = 0;
- else if (1 != fg->fg_count)
- error = EAGAIN; /* go close the dup .. */
- else if (UF_FORKCLOSE == (*pop & UF_FORKCLOSE)) {
+ } else if (1 != fg->fg_count) {
+ error = EAGAIN; /* go close the dup .. */
+ } else if (UF_FORKCLOSE == (*pop & UF_FORKCLOSE)) {
fg->fg_lflags |= FG_CONFINED;
error = 0;
- } else
- error = EBADF; /* open without O_CLOFORK? */
+ } else {
+ error = EBADF; /* open without O_CLOFORK? */
+ }
lck_mtx_unlock(&fg->fg_lock);
} else {
/*
* Other subsystems may have built on the immutability
* of FG_CONFINED; clearing it may be tricky.
*/
- error = EPERM; /* immutable */
+ error = EPERM; /* immutable */
}
goto out;
case F_OFD_SETLKWTIMEOUT:
case F_OFD_SETLKW:
flg |= F_WAIT;
- /* Fall into F_SETLK */
+ /* Fall into F_SETLK */
case F_SETLK:
case F_OFD_SETLK:
goto outdrop;
}
- if ( (error = vnode_getwithref(vp)) ) {
+ if ((error = vnode_getwithref(vp))) {
goto outdrop;
}
- if (fl.l_whence == SEEK_CUR)
+ if (fl.l_whence == SEEK_CUR) {
fl.l_start += offset;
+ }
#if CONFIG_MACF
error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
case F_GETLK:
case F_OFD_GETLK:
+ case F_GETLKPID:
+ case F_OFD_GETLKPID:
if (fp->f_type != DTYPE_VNODE) {
error = EBADF;
goto out;
/* Copy in the lock structure */
error = copyin(argp, (caddr_t)&fl, sizeof(fl));
- if (error)
+ if (error) {
goto outdrop;
+ }
/* Check starting byte and ending byte for EOVERFLOW in SEEK_CUR */
/* and ending byte for EOVERFLOW in SEEK_SET */
goto outdrop;
}
- if ( (error = vnode_getwithref(vp)) == 0 ) {
- if (fl.l_whence == SEEK_CUR)
- fl.l_start += offset;
+ if ((error = vnode_getwithref(vp)) == 0) {
+ if (fl.l_whence == SEEK_CUR) {
+ fl.l_start += offset;
+ }
#if CONFIG_MACF
error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
(void)vnode_put(vp);
- if (error == 0)
+ if (error == 0) {
error = copyout((caddr_t)&fl, argp, sizeof(fl));
+ }
}
goto outdrop;
}
error = copyin(argp, (caddr_t)&alloc_struct, sizeof(alloc_struct));
- if (error)
+ if (error) {
goto outdrop;
+ }
/* now set the space allocated to 0 */
alloc_struct.fst_bytesalloc = 0;
alloc_flags |= PREALLOCATE;
- if (alloc_struct.fst_flags & F_ALLOCATECONTIG)
+ if (alloc_struct.fst_flags & F_ALLOCATECONTIG) {
alloc_flags |= ALLOCATECONTIG;
+ }
- if (alloc_struct.fst_flags & F_ALLOCATEALL)
+ if (alloc_struct.fst_flags & F_ALLOCATEALL) {
alloc_flags |= ALLOCATEALL;
+ }
/*
* Do any position mode specific stuff. The only
*/
switch (alloc_struct.fst_posmode) {
-
case F_PEOFPOSMODE:
if (alloc_struct.fst_offset != 0) {
error = EINVAL;
default: {
error = EINVAL;
goto outdrop;
- }
}
- if ( (error = vnode_getwithref(vp)) == 0 ) {
- /*
+ }
+ if ((error = vnode_getwithref(vp)) == 0) {
+ /*
* call allocate to get the space
*/
- error = VNOP_ALLOCATE(vp,alloc_struct.fst_length,alloc_flags,
- &alloc_struct.fst_bytesalloc, alloc_struct.fst_offset,
- &context);
+ error = VNOP_ALLOCATE(vp, alloc_struct.fst_length, alloc_flags,
+ &alloc_struct.fst_bytesalloc, alloc_struct.fst_offset,
+ &context);
(void)vnode_put(vp);
error2 = copyout((caddr_t)&alloc_struct, argp, sizeof(alloc_struct));
- if (error == 0)
+ if (error == 0) {
error = error2;
+ }
}
goto outdrop;
}
goto outdrop;
}
+ case F_SPECULATIVE_READ: {
+ fspecread_t args;
+
+ if (fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
+ goto out;
+ }
+
+ vp = (struct vnode *)fp->f_data;
+ proc_fdunlock(p);
+
+ if ((error = copyin(argp, (caddr_t)&args, sizeof(args)))) {
+ goto outdrop;
+ }
+
+ /* Discard invalid offsets or lengths */
+ if ((args.fsr_offset < 0) || (args.fsr_length < 0)) {
+ error = EINVAL;
+ goto outdrop;
+ }
+
+ /*
+ * Round the file offset down to a page-size boundary (or to 0).
+ * The filesystem will need to round the length up to the end of the page boundary
+ * or to the EOF of the file.
+ */
+ uint64_t foff = (((uint64_t)args.fsr_offset) & ~((uint64_t)PAGE_MASK));
+ uint64_t foff_delta = args.fsr_offset - foff;
+ args.fsr_offset = (off_t) foff;
+
+ /*
+ * Now add in the delta to the supplied length. Since we may have adjusted the
+ * offset, increase it by the amount that we adjusted.
+ */
+ args.fsr_length += foff_delta;
+
+ if ((error = vnode_getwithref(vp))) {
+ goto outdrop;
+ }
+ error = VNOP_IOCTL(vp, F_SPECULATIVE_READ, (caddr_t)&args, 0, &context);
+ (void)vnode_put(vp);
+
+ goto outdrop;
+ }
case F_SETSIZE:
if (fp->f_type != DTYPE_VNODE) {
error = EBADF;
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- error = copyin(argp, (caddr_t)&offset, sizeof (off_t));
- if (error)
+ error = copyin(argp, (caddr_t)&offset, sizeof(off_t));
+ if (error) {
goto outdrop;
+ }
AUDIT_ARG(value64, offset);
error = vnode_getwithref(vp);
- if (error)
+ if (error) {
goto outdrop;
+ }
#if CONFIG_MACF
error = mac_vnode_check_truncate(&context,
&context);
#if CONFIG_MACF
- if (error == 0)
+ if (error == 0) {
mac_vnode_notify_truncate(&context, fp->f_fglob->fg_cred, vp);
+ }
#endif
}
error = EBADF;
goto out;
}
- if (uap->arg)
- fp->f_fglob->fg_flag &= ~FNORDAHEAD;
- else
- fp->f_fglob->fg_flag |= FNORDAHEAD;
+ if (uap->arg) {
+ fp->f_fglob->fg_flag &= ~FNORDAHEAD;
+ } else {
+ fp->f_fglob->fg_flag |= FNORDAHEAD;
+ }
goto out;
error = EBADF;
goto out;
}
- if (uap->arg)
- fp->f_fglob->fg_flag |= FNOCACHE;
- else
- fp->f_fglob->fg_flag &= ~FNOCACHE;
+ if (uap->arg) {
+ fp->f_fglob->fg_flag |= FNOCACHE;
+ } else {
+ fp->f_fglob->fg_flag &= ~FNOCACHE;
+ }
goto out;
error = EBADF;
goto out;
}
- if (uap->arg)
- fp->f_fglob->fg_flag |= FNODIRECT;
- else
- fp->f_fglob->fg_flag &= ~FNODIRECT;
+ if (uap->arg) {
+ fp->f_fglob->fg_flag |= FNODIRECT;
+ } else {
+ fp->f_fglob->fg_flag &= ~FNODIRECT;
+ }
goto out;
error = EBADF;
goto out;
}
- if (uap->arg)
- fp->f_fglob->fg_flag |= FSINGLE_WRITER;
- else
- fp->f_fglob->fg_flag &= ~FSINGLE_WRITER;
+ if (uap->arg) {
+ fp->f_fglob->fg_flag |= FSINGLE_WRITER;
+ } else {
+ fp->f_fglob->fg_flag &= ~FSINGLE_WRITER;
+ }
goto out;
case F_GLOBAL_NOCACHE:
- if (fp->f_type != DTYPE_VNODE) {
- error = EBADF;
+ if (fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
goto out;
}
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- if ( (error = vnode_getwithref(vp)) == 0 ) {
-
- *retval = vnode_isnocache(vp);
+ if ((error = vnode_getwithref(vp)) == 0) {
+ *retval = vnode_isnocache(vp);
- if (uap->arg)
- vnode_setnocache(vp);
- else
- vnode_clearnocache(vp);
+ if (uap->arg) {
+ vnode_setnocache(vp);
+ } else {
+ vnode_clearnocache(vp);
+ }
(void)vnode_put(vp);
}
goto outdrop;
case F_CHECK_OPENEVT:
- if (fp->f_type != DTYPE_VNODE) {
- error = EBADF;
+ if (fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
goto out;
}
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- if ( (error = vnode_getwithref(vp)) == 0 ) {
+ if ((error = vnode_getwithref(vp)) == 0) {
+ *retval = vnode_is_openevt(vp);
- *retval = vnode_is_openevt(vp);
-
- if (uap->arg)
- vnode_set_openevt(vp);
- else
- vnode_clear_openevt(vp);
+ if (uap->arg) {
+ vnode_set_openevt(vp);
+ } else {
+ vnode_clear_openevt(vp);
+ }
(void)vnode_put(vp);
}
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- if ( (error = copyin(argp, (caddr_t)&ra_struct, sizeof(ra_struct))) )
+ if ((error = copyin(argp, (caddr_t)&ra_struct, sizeof(ra_struct)))) {
goto outdrop;
- if ( (error = vnode_getwithref(vp)) == 0 ) {
- error = VNOP_IOCTL(vp, F_RDADVISE, (caddr_t)&ra_struct, 0, &context);
+ }
+ if ((error = vnode_getwithref(vp)) == 0) {
+ error = VNOP_IOCTL(vp, F_RDADVISE, (caddr_t)&ra_struct, 0, &context);
(void)vnode_put(vp);
}
goto outdrop;
- }
+ }
- case F_FLUSH_DATA:
+ case F_FLUSH_DATA:
- if (fp->f_type != DTYPE_VNODE) {
- error = EBADF;
- goto out;
- }
- vp = (struct vnode *)fp->f_data;
- proc_fdunlock(p);
+ if (fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
+ goto out;
+ }
+ vp = (struct vnode *)fp->f_data;
+ proc_fdunlock(p);
- if ( (error = vnode_getwithref(vp)) == 0 ) {
- error = VNOP_FSYNC(vp, MNT_NOWAIT, &context);
+ if ((error = vnode_getwithref(vp)) == 0) {
+ error = VNOP_FSYNC(vp, MNT_NOWAIT, &context);
- (void)vnode_put(vp);
- }
- goto outdrop;
+ (void)vnode_put(vp);
+ }
+ goto outdrop;
case F_LOG2PHYS:
case F_LOG2PHYS_EXT: {
- struct log2phys l2p_struct; /* structure for allocate command */
+ struct log2phys l2p_struct = {}; /* structure for allocate command */
int devBlockSize;
off_t file_offset = 0;
if (uap->cmd == F_LOG2PHYS_EXT) {
error = copyin(argp, (caddr_t)&l2p_struct, sizeof(l2p_struct));
- if (error)
+ if (error) {
goto out;
+ }
file_offset = l2p_struct.l2p_devoffset;
} else {
file_offset = fp->f_offset;
}
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- if ( (error = vnode_getwithref(vp)) ) {
+ if ((error = vnode_getwithref(vp))) {
goto outdrop;
}
error = VNOP_OFFTOBLK(vp, file_offset, &lbn);
(void)vnode_put(vp);
if (!error) {
- l2p_struct.l2p_flags = 0; /* for now */
+ l2p_struct.l2p_flags = 0; /* for now */
if (uap->cmd == F_LOG2PHYS_EXT) {
l2p_struct.l2p_contigbytes = run - (file_offset - offset);
} else {
- l2p_struct.l2p_contigbytes = 0; /* for now */
+ l2p_struct.l2p_contigbytes = 0; /* for now */
}
/*
if (bn == -1) {
/* Don't multiply it by the block size */
l2p_struct.l2p_devoffset = bn;
- }
- else {
+ } else {
l2p_struct.l2p_devoffset = bn * devBlockSize;
l2p_struct.l2p_devoffset += file_offset - offset;
}
error = copyout((caddr_t)&l2p_struct, argp, sizeof(l2p_struct));
}
goto outdrop;
- }
- case F_GETPATH: {
+ }
+ case F_GETPATH:
+ case F_GETPATH_NOFIRMLINK: {
char *pathbufp;
int pathlen;
error = ENOMEM;
goto outdrop;
}
- if ( (error = vnode_getwithref(vp)) == 0 ) {
- error = vn_getpath(vp, pathbufp, &pathlen);
- (void)vnode_put(vp);
+ if ((error = vnode_getwithref(vp)) == 0) {
+ if (uap->cmd == F_GETPATH_NOFIRMLINK) {
+ error = vn_getpath_ext(vp, NULL, pathbufp, &pathlen, VN_GETPATH_NO_FIRMLINK);
+ } else {
+ error = vn_getpath(vp, pathbufp, &pathlen);
+ }
+ (void)vnode_put(vp);
- if (error == 0)
- error = copyout((caddr_t)pathbufp, argp, pathlen);
+ if (error == 0) {
+ error = copyout((caddr_t)pathbufp, argp, pathlen);
+ }
}
FREE(pathbufp, M_TEMP);
goto outdrop;
size_t pathlen;
if (fp->f_type != DTYPE_VNODE) {
- error = EBADF;
+ error = EBADF;
goto out;
}
vp = (struct vnode *)fp->f_data;
pathlen = MAXPATHLEN;
pathbufp = kalloc(MAXPATHLEN);
- if ( (error = copyinstr(argp, pathbufp, MAXPATHLEN, &pathlen)) == 0 ) {
- if ( (error = vnode_getwithref(vp)) == 0 ) {
+ if ((error = copyinstr(argp, pathbufp, MAXPATHLEN, &pathlen)) == 0) {
+ if ((error = vnode_getwithref(vp)) == 0) {
AUDIT_ARG(text, pathbufp);
- error = vn_path_package_check(vp, pathbufp, pathlen, retval);
+ error = vn_path_package_check(vp, pathbufp, pathlen, retval);
(void)vnode_put(vp);
}
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- if ( (error = vnode_getwithref(vp)) == 0 ) {
- error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)NULL, 0, &context);
+ if ((error = vnode_getwithref(vp)) == 0) {
+ error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)NULL, 0, &context);
(void)vnode_put(vp);
}
AUDIT_ARG(mode, fopen.o_mode);
VATTR_INIT(&va);
/* Mask off all but regular access permissions */
- cmode = ((fopen.o_mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
+ cmode = ((fopen.o_mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
VATTR_SET(&va, va_mode, cmode & ACCESSPERMS);
/* Start the lookup relative to the file descriptor's vnode. */
NDINIT(&nd, LOOKUP, OP_OPEN, USEDVP | FOLLOW | AUDITVNPATH1, UIO_USERSPACE,
- fopen.o_pathname, &context);
+ fopen.o_pathname, &context);
nd.ni_dvp = vp;
error = open1(&context, &nd, fopen.o_flags, &va,
- fileproc_alloc_init, NULL, retval);
+ fileproc_alloc_init, NULL, retval);
vnode_put(vp);
break;
vnode_put(vp);
break;
-
}
case F_ADDSIGS:
}
error = vnode_getwithref(vp);
- if (error)
+ if (error) {
goto outdrop;
+ }
if (IS_64BIT_PROCESS(p)) {
- error = copyin(argp, &fs, sizeof (fs));
+ error = copyin(argp, &fs, sizeof(fs));
} else {
struct user32_fsignatures fs32;
- error = copyin(argp, &fs32, sizeof (fs32));
+ error = copyin(argp, &fs32, sizeof(fs32));
fs.fs_file_start = fs32.fs_file_start;
fs.fs_blob_start = CAST_USER_ADDR_T(fs32.fs_blob_start);
fs.fs_blob_size = fs32.fs_blob_size;
* First check if we have something loaded a this offset
*/
blob = ubc_cs_blob_get(vp, CPU_TYPE_ANY, fs.fs_file_start);
- if (blob != NULL)
- {
+ if (blob != NULL) {
/* If this is for dyld_sim revalidate the blob */
if (uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
error = ubc_cs_blob_revalidate(vp, blob, NULL, blob_add_flags);
if (error) {
- vnode_put(vp);
- goto outdrop;
+ blob = NULL;
+ if (error != EAGAIN) {
+ vnode_put(vp);
+ goto outdrop;
+ }
}
}
+ }
- } else {
+ if (blob == NULL) {
/*
* An arbitrary limit, to prevent someone from mapping in a 20GB blob. This should cover
* our use cases for the immediate future, but note that at the time of this commit, some
goto outdrop;
}
- if(uap->cmd == F_ADDSIGS) {
+ if (uap->cmd == F_ADDSIGS) {
error = copyin(fs.fs_blob_start,
- (void *) kernel_blob_addr,
- kernel_blob_size);
- } else /* F_ADDFILESIGS || F_ADDFILESIGS_RETURN || F_ADDFILESIGS_FOR_DYLD_SIM */ {
+ (void *) kernel_blob_addr,
+ kernel_blob_size);
+ } else { /* F_ADDFILESIGS || F_ADDFILESIGS_RETURN || F_ADDFILESIGS_FOR_DYLD_SIM */
int resid;
error = vn_rdwr(UIO_READ,
- vp,
- (caddr_t) kernel_blob_addr,
- kernel_blob_size,
- fs.fs_file_start + fs.fs_blob_start,
- UIO_SYSSPACE,
- 0,
- kauth_cred_get(),
- &resid,
- p);
+ vp,
+ (caddr_t) kernel_blob_addr,
+ kernel_blob_size,
+ fs.fs_file_start + fs.fs_blob_start,
+ UIO_SYSSPACE,
+ 0,
+ kauth_cred_get(),
+ &resid,
+ p);
if ((error == 0) && resid) {
/* kernel_blob_size rounded to a page size, but signature may be at end of file */
memset((void *)(kernel_blob_addr + (kernel_blob_size - resid)), 0x0, resid);
if (error) {
ubc_cs_blob_deallocate(kernel_blob_addr,
- kernel_blob_size);
+ kernel_blob_size);
vnode_put(vp);
goto outdrop;
}
blob = NULL;
error = ubc_cs_blob_add(vp,
- CPU_TYPE_ANY, /* not for a specific architecture */
- fs.fs_file_start,
- &kernel_blob_addr,
- kernel_blob_size,
- NULL,
- blob_add_flags,
- &blob);
+ CPU_TYPE_ANY, /* not for a specific architecture */
+ fs.fs_file_start,
+ &kernel_blob_addr,
+ kernel_blob_size,
+ NULL,
+ blob_add_flags,
+ &blob);
/* ubc_blob_add() has consumed "kernel_blob_addr" if it is zeroed */
if (error) {
if (kernel_blob_addr) {
ubc_cs_blob_deallocate(kernel_blob_addr,
- kernel_blob_size);
+ kernel_blob_size);
}
vnode_put(vp);
goto outdrop;
* all archs. Lets overwrite that.
*/
off_t end_offset = 0;
- if (blob)
+ if (blob) {
end_offset = blob->csb_end_offset;
- error = copyout(&end_offset, argp, sizeof (end_offset));
+ }
+ error = copyout(&end_offset, argp, sizeof(end_offset));
}
(void) vnode_put(vp);
}
case F_CHECK_LV: {
struct fileglob *fg;
- fchecklv_t lv;
+ fchecklv_t lv = {};
if (fp->f_type != DTYPE_VNODE) {
error = EBADF;
proc_fdunlock(p);
if (IS_64BIT_PROCESS(p)) {
- error = copyin(argp, &lv, sizeof (lv));
+ error = copyin(argp, &lv, sizeof(lv));
} else {
- struct user32_fchecklv lv32;
+ struct user32_fchecklv lv32 = {};
- error = copyin(argp, &lv32, sizeof (lv32));
+ error = copyin(argp, &lv32, sizeof(lv32));
lv.lv_file_start = lv32.lv_file_start;
- lv.lv_error_message = CAST_USER_ADDR_T(lv32.lv_error_message);
- lv.lv_error_message_size = lv32.lv_error_message;
+ lv.lv_error_message = (void *)(uintptr_t)lv32.lv_error_message;
+ lv.lv_error_message_size = lv32.lv_error_message_size;
}
- if (error)
+ if (error) {
goto outdrop;
+ }
#if CONFIG_MACF
error = mac_file_check_library_validation(p, fg, lv.lv_file_start,
- lv.lv_error_message, lv.lv_error_message_size);
+ (user_long_t)lv.lv_error_message, lv.lv_error_message_size);
#endif
break;
VATTR_WANTED(&va, va_dataprotect_class);
error = VNOP_GETATTR(vp, &va, &context);
if (!error) {
- if (VATTR_IS_SUPPORTED(&va, va_dataprotect_class))
+ if (VATTR_IS_SUPPORTED(&va, va_dataprotect_class)) {
*retval = va.va_dataprotect_class;
- else
+ } else {
error = ENOTSUP;
+ }
}
vnode_put(vp);
/* Only go forward if you have write access */
vfs_context_t ctx = vfs_context_current();
- if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
+ if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
vnode_put(vp);
error = EBADF;
goto outdrop;
.len = CP_MAX_WRAPPEDKEYSIZE,
};
- MALLOC(k.key, char *, k.len, M_TEMP, M_WAITOK);
+ MALLOC(k.key, char *, k.len, M_TEMP, M_WAITOK | M_ZERO);
error = VNOP_IOCTL(vp, F_TRANSCODEKEY, (caddr_t)&k, 1, &context);
}
vp = (struct vnode*) fp->f_data;
- proc_fdunlock (p);
+ proc_fdunlock(p);
if (vnode_getwithref(vp)) {
error = ENOENT;
error = VNOP_IOCTL(vp, F_GETPROTECTIONLEVEL, (caddr_t)retval, 0, &context);
- vnode_put (vp);
+ vnode_put(vp);
break;
}
}
vp = (struct vnode*) fp->f_data;
- proc_fdunlock (p);
+ proc_fdunlock(p);
if (vnode_getwithref(vp)) {
error = ENOENT;
error = VNOP_IOCTL(vp, F_GETDEFAULTPROTLEVEL, (caddr_t)retval, 0, &context);
- vnode_put (vp);
+ vnode_put(vp);
break;
}
int fd2 = CAST_DOWN_EXPLICIT(int32_t, uap->arg);
error = priv_check_cred(kauth_cred_get(), PRIV_VFS_MOVE_DATA_EXTENTS, 0);
- if (error)
+ if (error) {
goto out;
+ }
if (fp->f_type != DTYPE_VNODE) {
error = EBADF;
goto out;
}
- /* For now, special case HFS+ only, since this is SPI. */
+ /*
+ * For now, special case HFS+ and APFS only, since this
+ * is SPI.
+ */
src_vp = (struct vnode *)fp->f_data;
- if (src_vp->v_tag != VT_HFS) {
- error = EINVAL;
+ if (src_vp->v_tag != VT_HFS && src_vp->v_tag != VT_APFS) {
+ error = ENOTSUP;
goto out;
}
* Get the references before we start acquiring iocounts on the vnodes,
* while we still hold the proc fd lock
*/
- if ( (error = fp_lookup(p, fd2, &fp2, 1)) ) {
+ if ((error = fp_lookup(p, fd2, &fp2, 1))) {
error = EBADF;
goto out;
}
goto out;
}
dst_vp = (struct vnode *)fp2->f_data;
- if (dst_vp->v_tag != VT_HFS) {
+ if (dst_vp->v_tag != VT_HFS && dst_vp->v_tag != VT_APFS) {
fp_drop(p, fd2, fp2, 1);
- error = EINVAL;
+ error = ENOTSUP;
goto out;
}
goto outdrop;
}
if (vnode_getwithref(dst_vp)) {
- vnode_put (src_vp);
+ vnode_put(src_vp);
fp_drop(p, fd2, fp2, 0);
error = ENOENT;
goto outdrop;
* both live on the same filesystem.
*/
if (dst_vp == src_vp) {
- vnode_put (src_vp);
- vnode_put (dst_vp);
- fp_drop (p, fd2, fp2, 0);
+ vnode_put(src_vp);
+ vnode_put(dst_vp);
+ fp_drop(p, fd2, fp2, 0);
error = EINVAL;
goto outdrop;
}
if (dst_vp->v_mount != src_vp->v_mount) {
- vnode_put (src_vp);
- vnode_put (dst_vp);
- fp_drop (p, fd2, fp2, 0);
+ vnode_put(src_vp);
+ vnode_put(dst_vp);
+ fp_drop(p, fd2, fp2, 0);
error = EXDEV;
goto outdrop;
}
/* Now we have a legit pair of FDs. Go to work */
/* Now check for write access to the target files */
- if(vnode_authorize(src_vp, NULLVP,
- (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
+ if (vnode_authorize(src_vp, NULLVP,
+ (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
vnode_put(src_vp);
vnode_put(dst_vp);
fp_drop(p, fd2, fp2, 0);
goto outdrop;
}
- if(vnode_authorize(dst_vp, NULLVP,
- (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
+ if (vnode_authorize(dst_vp, NULLVP,
+ (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), &context) != 0) {
vnode_put(src_vp);
vnode_put(dst_vp);
fp_drop(p, fd2, fp2, 0);
}
/* Verify that both vps point to files and not directories */
- if ( !vnode_isreg(src_vp) || !vnode_isreg(dst_vp)) {
+ if (!vnode_isreg(src_vp) || !vnode_isreg(dst_vp)) {
error = EINVAL;
- vnode_put (src_vp);
- vnode_put (dst_vp);
- fp_drop (p, fd2, fp2, 0);
+ vnode_put(src_vp);
+ vnode_put(dst_vp);
+ fp_drop(p, fd2, fp2, 0);
goto outdrop;
}
error = VNOP_EXCHANGE(src_vp, dst_vp, FSOPT_EXCHANGE_DATA_ONLY, &context);
- vnode_put (src_vp);
- vnode_put (dst_vp);
+ vnode_put(src_vp);
+ vnode_put(dst_vp);
fp_drop(p, fd2, fp2, 0);
break;
}
}
vp = (struct vnode*) fp->f_data;
- proc_fdunlock (p);
+ proc_fdunlock(p);
/* get the vnode */
if (vnode_getwithref(vp)) {
/* invoke ioctl to pass off to FS */
/* Only go forward if you have write access */
vfs_context_t ctx = vfs_context_current();
- if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
+ if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
vnode_put(vp);
error = EBADF;
goto outdrop;
error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)&gcounter, 0, &context);
- vnode_put (vp);
+ vnode_put(vp);
break;
}
* the open FD will written to the Fastflow.
*/
case F_SET_GREEDY_MODE:
- /* intentionally drop through to the same handler as F_SETSTATIC.
- * both fcntls should pass the argument and their selector into VNOP_IOCTL.
- */
+ /* intentionally drop through to the same handler as F_SETSTATIC.
+ * both fcntls should pass the argument and their selector into VNOP_IOCTL.
+ */
/*
* SPI (private) for indicating to a filesystem that subsequent writes to
/* Only go forward if you have write access */
vfs_context_t ctx = vfs_context_current();
- if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
+ if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
vnode_put(vp);
error = EBADF;
goto outdrop;
/* extract 32 bits of flags from userland */
param_ptr = (caddr_t) uap->arg;
param = (uint32_t) param_ptr;
- }
- else {
+ } else {
/* If no argument is specified, error out */
error = EINVAL;
goto out;
* all of them are mutually exclusive for now.
*/
switch (param) {
- case F_IOTYPE_ISOCHRONOUS:
- break;
+ case F_IOTYPE_ISOCHRONOUS:
+ break;
- default:
- error = EINVAL;
- goto out;
+ default:
+ error = EINVAL;
+ goto out;
}
/* Only go forward if you have write access */
vfs_context_t ctx = vfs_context_current();
- if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
+ if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
vnode_put(vp);
error = EBADF;
goto outdrop;
/* only proceed if you have write access */
vfs_context_t ctx = vfs_context_current();
- if(vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
+ if (vnode_authorize(vp, NULLVP, (KAUTH_VNODE_ACCESS | KAUTH_VNODE_WRITE_DATA), ctx) != 0) {
vnode_put(vp);
error = EBADF;
goto outdrop;
/* If arg != 0, set, otherwise unset */
if (uap->arg) {
- error = VNOP_IOCTL (vp, uap->cmd, (caddr_t)1, 0, &context);
- }
- else {
- error = VNOP_IOCTL (vp, uap->cmd, (caddr_t)NULL, 0, &context);
+ error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)1, 0, &context);
+ } else {
+ error = VNOP_IOCTL(vp, uap->cmd, (caddr_t)NULL, 0, &context);
}
vnode_put(vp);
error = ENOMEM;
goto outdrop;
}
- if ( (error = vnode_getwithref(vp)) == 0 ) {
+ if ((error = vnode_getwithref(vp)) == 0) {
int backingstore = 0;
/* Check for error from vn_getpath before moving on */
if ((error = vn_getpath(vp, pathbufp, &pathlen)) == 0) {
if (vp->v_tag == VT_HFS) {
- error = VNOP_IOCTL (vp, uap->cmd, (caddr_t) &backingstore, 0, &context);
+ error = VNOP_IOCTL(vp, uap->cmd, (caddr_t) &backingstore, 0, &context);
}
(void)vnode_put(vp);
error = EBUSY;
}
}
- } else
+ } else {
(void)vnode_put(vp);
+ }
}
FREE(pathbufp, M_TEMP);
goto outdrop;
* for this vnode; this can include special devices, and will
* effectively overload fcntl() to send ioctl()'s.
*/
- if((uap->cmd & IOC_VOID) && (uap->cmd & IOC_INOUT)){
- error = EINVAL;
+ if ((uap->cmd & IOC_VOID) && (uap->cmd & IOC_INOUT)) {
+ error = EINVAL;
goto out;
}
/* Catch any now-invalid fcntl() selectors */
switch (uap->cmd) {
- case F_MARKDEPENDENCY:
- error = EINVAL;
- goto out;
- default:
- break;
+ case (int)APFSIOC_REVERT_TO_SNAPSHOT:
+ case (int)FSIOC_FIOSEEKHOLE:
+ case (int)FSIOC_FIOSEEKDATA:
+ case (int)FSIOC_CAS_BSDFLAGS:
+ case HFS_GET_BOOT_INFO:
+ case HFS_SET_BOOT_INFO:
+ case FIOPINSWAP:
+ case F_MARKDEPENDENCY:
+ case TIOCREVOKE:
+ error = EINVAL;
+ goto out;
+ default:
+ break;
}
if (fp->f_type != DTYPE_VNODE) {
vp = (struct vnode *)fp->f_data;
proc_fdunlock(p);
- if ( (error = vnode_getwithref(vp)) == 0 ) {
+ if ((error = vnode_getwithref(vp)) == 0) {
#define STK_PARAMS 128
char stkbuf[STK_PARAMS] = {0};
unsigned int size;
}
memp = NULL;
- if (size > sizeof (stkbuf)) {
+ if (size > sizeof(stkbuf)) {
if ((memp = (caddr_t)kalloc(size)) == 0) {
(void)vnode_put(vp);
error = ENOMEM;
error = copyin(argp, data, size);
if (error) {
(void)vnode_put(vp);
- if (memp)
+ if (memp) {
kfree(memp, size);
+ }
goto outdrop;
}
/* Bzero the section beyond that which was needed */
if (size <= sizeof(stkbuf)) {
- bzero ( (((uint8_t*)data) + size), (sizeof(stkbuf) - size));
+ bzero((((uint8_t*)data) + size), (sizeof(stkbuf) - size));
}
} else {
/* int */
bzero(data, size);
} else if (uap->cmd & IOC_VOID) {
if (is64bit) {
- *(user_addr_t *)data = argp;
+ *(user_addr_t *)data = argp;
} else {
- *(uint32_t *)data = (uint32_t)argp;
+ *(uint32_t *)data = (uint32_t)argp;
}
}
(void)vnode_put(vp);
/* Copy any output data to user */
- if (error == 0 && (uap->cmd & IOC_OUT) && size)
+ if (error == 0 && (uap->cmd & IOC_OUT) && size) {
error = copyout(data, argp, size);
- if (memp)
+ }
+ if (memp) {
kfree(memp, size);
+ }
}
break;
}
outdrop:
AUDIT_ARG(vnpath_withref, vp, ARG_VNODE1);
fp_drop(p, fd, fp, 0);
- return(error);
+ return error;
out:
fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return(error);
+ return error;
}
if ((ofp = fdp->fd_ofiles[old]) == NULL ||
(fdp->fd_ofileflags[old] & UF_RESERVED)) {
fdrelse(p, new);
- return (EBADF);
+ return EBADF;
}
fg_ref(ofp);
if (error) {
fg_drop(ofp);
fdrelse(p, new);
- return (error);
+ return error;
}
#endif
if (nfp == NULL) {
fg_drop(ofp);
fdrelse(p, new);
- return (ENOMEM);
+ return ENOMEM;
}
nfp->f_fglob = ofp->f_fglob;
#if DIAGNOSTIC
- if (fdp->fd_ofiles[new] != 0)
+ if (fdp->fd_ofiles[new] != 0) {
panic("finishdup: overwriting fd_ofiles with new %d", new);
- if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0)
+ }
+ if ((fdp->fd_ofileflags[new] & UF_RESERVED) == 0) {
panic("finishdup: unreserved fileflags with new %d", new);
+ }
#endif
- if (new > fdp->fd_lastfile)
+ if (new > fdp->fd_lastfile) {
fdp->fd_lastfile = new;
+ }
*fdflags(p, new) |= fd_flags;
procfdtbl_releasefd(p, new, nfp);
*retval = new;
- return (0);
+ return 0;
}
* fp_lookup:EBADF Bad file descriptor
* fp_guard_exception:??? Guarded file descriptor
* close_internal:EBADF
- * close_internal:??? Anything returnable by a per-fileops
+ * close_internal:??? Anything returnable by a per-fileops
* close function
*/
int
close(proc_t p, struct close_args *uap, int32_t *retval)
{
__pthread_testcancel(1);
- return(close_nocancel(p, (struct close_nocancel_args *)uap, retval));
+ return close_nocancel(p, (struct close_nocancel_args *)uap, retval);
}
proc_fdlock(p);
- if ( (error = fp_lookup(p,fd,&fp, 1)) ) {
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
if (FP_ISGUARDED(fp, GUARD_CLOSE)) {
error = fp_guard_exception(p, fd, fp, kGUARD_EXC_CLOSE);
(void) fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
error = close_internal_locked(p, fd, fp, 0);
proc_fdunlock(p);
- return (error);
+ return error;
}
*
* Returns: 0 Success
* EBADF fd already in close wait state
- * closef_locked:??? Anything returnable by a per-fileops
+ * closef_locked:??? Anything returnable by a per-fileops
* close function
*
* Locks: Assumes proc_fdlock for process is held by the caller and returns
close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags)
{
struct filedesc *fdp = p->p_fd;
- int error =0;
+ int error = 0;
int resvfd = flags & FD_DUP2RESV;
#if DIAGNOSTIC
- if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0)
+ if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) {
panic("close_internal: unreserved fileflags with fd %d", fd);
+ }
#endif
fp->f_flags |= FP_CLOSING;
- if ( (fp->f_flags & FP_AIOISSUED) || kauth_authorize_fileop_has_listeners() ) {
-
- proc_fdunlock(p);
+ if ((fp->f_flags & FP_AIOISSUED) || kauth_authorize_fileop_has_listeners()) {
+ proc_fdunlock(p);
- if ( (fp->f_type == DTYPE_VNODE) && kauth_authorize_fileop_has_listeners() ) {
- /*
+ if ((fp->f_type == DTYPE_VNODE) && kauth_authorize_fileop_has_listeners()) {
+ /*
* call out to allow 3rd party notification of close.
* Ignore result of kauth_authorize_fileop call.
*/
- if (vnode_getwithref((vnode_t)fp->f_data) == 0) {
- u_int fileop_flags = 0;
- if ((fp->f_flags & FP_WRITTEN) != 0)
- fileop_flags |= KAUTH_FILEOP_CLOSE_MODIFIED;
- kauth_authorize_fileop(fp->f_fglob->fg_cred, KAUTH_FILEOP_CLOSE,
- (uintptr_t)fp->f_data, (uintptr_t)fileop_flags);
+ if (vnode_getwithref((vnode_t)fp->f_data) == 0) {
+ u_int fileop_flags = 0;
+ if ((fp->f_flags & FP_WRITTEN) != 0) {
+ fileop_flags |= KAUTH_FILEOP_CLOSE_MODIFIED;
+ }
+ kauth_authorize_fileop(fp->f_fglob->fg_cred, KAUTH_FILEOP_CLOSE,
+ (uintptr_t)fp->f_data, (uintptr_t)fileop_flags);
vnode_put((vnode_t)fp->f_data);
}
}
- if (fp->f_flags & FP_AIOISSUED)
- /*
+ if (fp->f_flags & FP_AIOISSUED) {
+ /*
* cancel all async IO requests that can be cancelled.
*/
- _aio_close( p, fd );
+ _aio_close( p, fd );
+ }
proc_fdlock(p);
}
- if (fd < fdp->fd_knlistsize)
- knote_fdclose(p, fd, FALSE);
-
- if (fp->f_flags & FP_WAITEVENT)
- (void)waitevent_close(p, fp);
+ if (fd < fdp->fd_knlistsize) {
+ knote_fdclose(p, fd);
+ }
+ /* release the ref returned from fp_lookup before calling drain */
+ (void) os_ref_release_locked(&fp->f_iocount);
fileproc_drain(p, fp);
+ if (fp->f_flags & FP_WAITEVENT) {
+ (void)waitevent_close(p, fp);
+ }
+
if (resvfd == 0) {
_fdrelse(p, fd);
} else {
procfdtbl_reservefd(p, fd);
}
- if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET)
+ if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET) {
KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_END,
fd, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
+ }
error = closef_locked(fp, fp->f_fglob, p);
- if ((fp->f_flags & FP_WAITCLOSE) == FP_WAITCLOSE)
+ if ((fp->f_flags & FP_WAITCLOSE) == FP_WAITCLOSE) {
wakeup(&fp->f_flags);
+ }
fp->f_flags &= ~(FP_WAITCLOSE | FP_CLOSING);
proc_fdunlock(p);
#if DIAGNOSTIC
if (resvfd != 0) {
- if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0)
+ if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) {
panic("close with reserved fd returns with freed fd:%d: proc: %p", fd, p);
+ }
}
#endif
- return(error);
+ return error;
}
AUDIT_ARG(fd, fd);
if ((error = fp_lookup(p, fd, &fp, 0)) != 0) {
- return(error);
+ return error;
}
type = fp->f_type;
data = fp->f_data;
sbptr = (void *)&source;
switch (type) {
-
case DTYPE_VNODE:
if ((error = vnode_getwithref((vnode_t)data)) == 0) {
/*
* going to let them get the basic stat information.
*/
if (xsecurity == USER_ADDR_NULL) {
- error = vn_stat_noauth((vnode_t)data, sbptr, NULL, isstat64, ctx,
- fp->f_fglob->fg_cred);
+ error = vn_stat_noauth((vnode_t)data, sbptr, NULL, isstat64, 0, ctx,
+ fp->f_fglob->fg_cred);
} else {
- error = vn_stat((vnode_t)data, sbptr, &fsec, isstat64, ctx);
+ error = vn_stat((vnode_t)data, sbptr, &fsec, isstat64, 0, ctx);
}
AUDIT_ARG(vnpath, (struct vnode *)data, ARG_VNODE1);
/* caller wants extended security information? */
if (xsecurity != USER_ADDR_NULL) {
-
/* did we get any? */
- if (fsec == KAUTH_FILESEC_NONE) {
+ if (fsec == KAUTH_FILESEC_NONE) {
if (susize(xsecurity_size, 0) != 0) {
error = EFAULT;
goto out;
}
/* if the caller supplied enough room, copy out to it */
- if (xsecurity_bufsize >= KAUTH_FILESEC_COPYSIZE(fsec))
+ if (xsecurity_bufsize >= KAUTH_FILESEC_COPYSIZE(fsec)) {
error = copyout(fsec, xsecurity, KAUTH_FILESEC_COPYSIZE(fsec));
+ }
}
}
out:
fp_drop(p, fd, fp, 0);
- if (fsec != NULL)
+ if (fsec != NULL) {
kauth_filesec_free(fsec);
- return (error);
+ }
+ return error;
}
int
fstat_extended(proc_t p, struct fstat_extended_args *uap, __unused int32_t *retval)
{
- return(fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 0));
+ return fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 0);
}
int
fstat(proc_t p, struct fstat_args *uap, __unused int32_t *retval)
{
- return(fstat1(p, uap->fd, uap->ub, 0, 0, 0));
+ return fstat1(p, uap->fd, uap->ub, 0, 0, 0);
}
int
fstat64_extended(proc_t p, struct fstat64_extended_args *uap, __unused int32_t *retval)
{
- return(fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 1));
+ return fstat1(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 1);
}
int
fstat64(proc_t p, struct fstat64_args *uap, __unused int32_t *retval)
{
- return(fstat1(p, uap->fd, uap->ub, 0, 0, 1));
+ return fstat1(p, uap->fd, uap->ub, 0, 0, 1);
}
AUDIT_ARG(fd, uap->fd);
- if ( (error = fp_lookup(p, fd, &fp, 0)) )
- return(error);
+ if ((error = fp_lookup(p, fd, &fp, 0))) {
+ return error;
+ }
type = fp->f_type;
data = fp->f_data;
switch (type) {
-
case DTYPE_SOCKET:
if (uap->name != _PC_PIPE_BUF) {
error = EINVAL;
case DTYPE_VNODE:
vp = (struct vnode *)data;
- if ( (error = vnode_getwithref(vp)) == 0) {
- AUDIT_ARG(vnpath, vp, ARG_VNODE1);
+ if ((error = vnode_getwithref(vp)) == 0) {
+ AUDIT_ARG(vnpath, vp, ARG_VNODE1);
error = vn_pathconf(vp, uap->name, retval, vfs_context_current());
default:
error = EINVAL;
goto out;
-
}
/*NOTREACHED*/
out:
fp_drop(p, fd, fp, 0);
- return(error);
+ return error;
}
/*
lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
for (;;) {
last = min(fdp->fd_nfiles, lim);
- if ((i = want) < fdp->fd_freefile)
+ if ((i = want) < fdp->fd_freefile) {
i = fdp->fd_freefile;
+ }
for (; i < last; i++) {
if (fdp->fd_ofiles[i] == NULL && !(fdp->fd_ofileflags[i] & UF_RESERVED)) {
procfdtbl_reservefd(p, i);
- if (i > fdp->fd_lastfile)
+ if (i > fdp->fd_lastfile) {
fdp->fd_lastfile = i;
- if (want <= fdp->fd_freefile)
+ }
+ if (want <= fdp->fd_freefile) {
fdp->fd_freefile = i;
+ }
*result = i;
- return (0);
+ return 0;
}
}
/*
* No space in current array. Expand?
*/
- if (fdp->fd_nfiles >= lim)
- return (EMFILE);
- if (fdp->fd_nfiles < NDEXTENT)
+ if (fdp->fd_nfiles >= lim) {
+ return EMFILE;
+ }
+ if (fdp->fd_nfiles < NDEXTENT) {
numfiles = NDEXTENT;
- else
+ } else {
numfiles = 2 * fdp->fd_nfiles;
+ }
/* Enforce lim */
- if (numfiles > lim)
+ if (numfiles > lim) {
numfiles = lim;
+ }
proc_fdunlock(p);
MALLOC_ZONE(newofiles, struct fileproc **,
- numfiles * OFILESIZE, M_OFILETABL, M_WAITOK);
+ numfiles * OFILESIZE, M_OFILETABL, M_WAITOK);
proc_fdlock(p);
if (newofiles == NULL) {
- return (ENOMEM);
+ return ENOMEM;
}
if (fdp->fd_nfiles >= numfiles) {
FREE_ZONE(newofiles, numfiles * OFILESIZE, M_OFILETABL);
*/
oldnfiles = fdp->fd_nfiles;
(void) memcpy(newofiles, fdp->fd_ofiles,
- oldnfiles * sizeof(*fdp->fd_ofiles));
+ oldnfiles * sizeof(*fdp->fd_ofiles));
(void) memset(&newofiles[oldnfiles], 0,
- (numfiles - oldnfiles) * sizeof(*fdp->fd_ofiles));
+ (numfiles - oldnfiles) * sizeof(*fdp->fd_ofiles));
(void) memcpy(newofileflags, fdp->fd_ofileflags,
- oldnfiles * sizeof(*fdp->fd_ofileflags));
+ oldnfiles * sizeof(*fdp->fd_ofileflags));
(void) memset(&newofileflags[oldnfiles], 0,
- (numfiles - oldnfiles) *
- sizeof(*fdp->fd_ofileflags));
+ (numfiles - oldnfiles) *
+ sizeof(*fdp->fd_ofileflags));
ofiles = fdp->fd_ofiles;
fdp->fd_ofiles = newofiles;
fdp->fd_ofileflags = newofileflags;
int i, lim;
lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
- if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0)
- return (1);
+ if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
+ return 1;
+ }
fpp = &fdp->fd_ofiles[fdp->fd_freefile];
flags = &fdp->fd_ofileflags[fdp->fd_freefile];
- for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++, flags++)
- if (*fpp == NULL && !(*flags & UF_RESERVED) && --n <= 0)
- return (1);
- return (0);
+ for (i = fdp->fd_nfiles - fdp->fd_freefile; --i >= 0; fpp++, flags++) {
+ if (*fpp == NULL && !(*flags & UF_RESERVED) && --n <= 0) {
+ return 1;
+ }
+ }
+ return 0;
}
struct fileproc *fp;
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
- return (EBADF);
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ return EBADF;
}
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- return (0);
+ }
+ return 0;
}
proc_fdlock_spin(p);
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_VNODE) {
proc_fdunlock(p);
- return(ENOTSUP);
+ return ENOTSUP;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultvp)
+ }
+ if (resultvp) {
*resultvp = (struct vnode *)fp->f_data;
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
*/
int
fp_getfvpandvid(proc_t p, int fd, struct fileproc **resultfp,
- struct vnode **resultvp, uint32_t *vidp)
+ struct vnode **resultvp, uint32_t *vidp)
{
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
proc_fdlock_spin(p);
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_VNODE) {
proc_fdunlock(p);
- return(ENOTSUP);
+ return ENOTSUP;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultvp)
+ }
+ if (resultvp) {
*resultvp = (struct vnode *)fp->f_data;
- if (vidp)
+ }
+ if (vidp) {
*vidp = (uint32_t)vnode_vid((struct vnode *)fp->f_data);
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
*/
int
fp_getfsock(proc_t p, int fd, struct fileproc **resultfp,
- struct socket **results)
+ struct socket **results)
{
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
proc_fdlock_spin(p);
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_SOCKET) {
proc_fdunlock(p);
- return(EOPNOTSUPP);
+ return EOPNOTSUPP;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (results)
+ }
+ if (results) {
*results = (struct socket *)fp->f_data;
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
*/
int
fp_getfkq(proc_t p, int fd, struct fileproc **resultfp,
- struct kqueue **resultkq)
+ struct kqueue **resultkq)
{
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
proc_fdlock_spin(p);
- if ( fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ if (fd < 0 || fd >= fdp->fd_nfiles ||
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_KQUEUE) {
proc_fdunlock(p);
- return(EBADF);
+ return EBADF;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultkq)
+ }
+ if (resultkq) {
*resultkq = (struct kqueue *)fp->f_data;
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
*/
int
fp_getfpshm(proc_t p, int fd, struct fileproc **resultfp,
- struct pshmnode **resultpshm)
+ struct pshmnode **resultpshm)
{
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
proc_fdlock_spin(p);
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_PSXSHM) {
-
proc_fdunlock(p);
- return(EBADF);
+ return EBADF;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultpshm)
+ }
+ if (resultpshm) {
*resultpshm = (struct pshmnode *)fp->f_data;
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
*/
int
fp_getfpsem(proc_t p, int fd, struct fileproc **resultfp,
- struct psemnode **resultpsem)
+ struct psemnode **resultpsem)
{
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
proc_fdlock_spin(p);
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_PSXSEM) {
proc_fdunlock(p);
- return(EBADF);
+ return EBADF;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultpsem)
+ }
+ if (resultpsem) {
*resultpsem = (struct psemnode *)fp->f_data;
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
*/
int
fp_getfpipe(proc_t p, int fd, struct fileproc **resultfp,
- struct pipe **resultpipe)
+ struct pipe **resultpipe)
{
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
proc_fdlock_spin(p);
if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
if (fp->f_type != DTYPE_PIPE) {
proc_fdunlock(p);
- return(EBADF);
+ return EBADF;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultpipe)
+ }
+ if (resultpipe) {
*resultpipe = (struct pipe *)fp->f_data;
+ }
proc_fdunlock(p);
- return (0);
+ return 0;
}
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
- if (!locked)
+ if (!locked) {
proc_fdlock_spin(p);
+ }
if (fd < 0 || fdp == NULL || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
- if (!locked)
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ if (!locked) {
proc_fdunlock(p);
- return (EBADF);
+ }
+ return EBADF;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->f_iocount);
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (!locked)
+ }
+ if (!locked) {
proc_fdunlock(p);
+ }
- return (0);
+ return 0;
}
* Description: Swap the fileproc pointer for a given fd with a new
* fileproc pointer in the per-process open file table of
* the specified process. The fdlock must be held at entry.
+ * Iff the swap is successful, the old fileproc pointer is freed.
*
* Parameters: p Process containing the fd
* fd The fd of interest
* Returns: 0 Success
* EBADF Bad file descriptor
* EINTR Interrupted
- * EKEEPLOOKING f_iocount changed while lock was dropped.
+ * EKEEPLOOKING Other references were active, try again.
*/
int
fp_tryswap(proc_t p, int fd, struct fileproc *nfp)
proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
- if (0 != (error = fp_lookup(p, fd, &fp, 1)))
- return (error);
+ if (0 != (error = fp_lookup(p, fd, &fp, 1))) {
+ return error;
+ }
/*
* At this point, our caller (change_guardedfd_np) has
* one f_iocount reference, and we just took another
* one to begin the replacement.
+ * fp and nfp have a +1 reference from allocation.
+ * Thus if no-one else is looking, f_iocount should be 3.
*/
- if (fp->f_iocount < 2) {
- panic("f_iocount too small %d", fp->f_iocount);
- } else if (2 == fp->f_iocount) {
-
+ if (os_ref_get_count(&fp->f_iocount) < 3 ||
+ 1 != os_ref_get_count(&nfp->f_iocount)) {
+ panic("%s: f_iocount", __func__);
+ } else if (3 == os_ref_get_count(&fp->f_iocount)) {
/* Copy the contents of *fp, preserving the "type" of *nfp */
nfp->f_flags = (nfp->f_flags & FP_TYPEMASK) |
- (fp->f_flags & ~FP_TYPEMASK);
- nfp->f_iocount = fp->f_iocount;
+ (fp->f_flags & ~FP_TYPEMASK);
+ os_ref_retain_locked(&nfp->f_iocount);
+ os_ref_retain_locked(&nfp->f_iocount);
nfp->f_fglob = fp->f_fglob;
nfp->f_wset = fp->f_wset;
p->p_fd->fd_ofiles[fd] = nfp;
- (void) fp_drop(p, fd, nfp, 1);
+ fp_drop(p, fd, nfp, 1);
+
+ os_ref_release_live(&fp->f_iocount);
+ os_ref_release_live(&fp->f_iocount);
+ fileproc_free(fp);
} else {
/*
* Wait for all other active references to evaporate.
* reevaluation of the change-guard attempt.
*/
error = EKEEPLOOKING;
- printf("%s: lookup collision fd %d\n", __func__, fd);
}
(void) fp_drop(p, fd, fp, 1);
}
- return (error);
+ return error;
}
int
fp_drop_written(proc_t p, int fd, struct fileproc *fp)
{
- int error;
+ int error;
proc_fdlock_spin(p);
proc_fdunlock(p);
- return (error);
+ return error;
}
int
fp_drop_event(proc_t p, int fd, struct fileproc *fp)
{
- int error;
+ int error;
proc_fdlock_spin(p);
proc_fdunlock(p);
- return (error);
+ return error;
}
fp_drop(proc_t p, int fd, struct fileproc *fp, int locked)
{
struct filedesc *fdp = p->p_fd;
- int needwakeup = 0;
+ int needwakeup = 0;
- if (!locked)
+ if (!locked) {
proc_fdlock_spin(p);
- if ((fp == FILEPROC_NULL) && (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- ((fdp->fd_ofileflags[fd] & UF_RESERVED) &&
- !(fdp->fd_ofileflags[fd] & UF_CLOSING)))) {
- if (!locked)
+ }
+ if ((fp == FILEPROC_NULL) && (fd < 0 || fd >= fdp->fd_nfiles ||
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ ((fdp->fd_ofileflags[fd] & UF_RESERVED) &&
+ !(fdp->fd_ofileflags[fd] & UF_CLOSING)))) {
+ if (!locked) {
proc_fdunlock(p);
- return (EBADF);
+ }
+ return EBADF;
}
- fp->f_iocount--;
- if (fp->f_iocount == 0) {
- if (fp->f_flags & FP_SELCONFLICT)
+ if (1 == os_ref_release_locked(&fp->f_iocount)) {
+ if (fp->f_flags & FP_SELCONFLICT) {
fp->f_flags &= ~FP_SELCONFLICT;
+ }
if (p->p_fpdrainwait) {
p->p_fpdrainwait = 0;
needwakeup = 1;
}
}
- if (!locked)
+ if (!locked) {
proc_fdunlock(p);
- if (needwakeup)
- wakeup(&p->p_fpdrainwait);
+ }
+ if (needwakeup) {
+ wakeup(&p->p_fpdrainwait);
+ }
- return (0);
+ return 0;
}
int error;
proc_fdlock_spin(p);
- if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
if (fp->f_type != DTYPE_VNODE) {
- fp_drop(p, fd, fp,1);
+ fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return(EINVAL);
+ return EINVAL;
}
- if (vpp != NULL)
+ if (vpp != NULL) {
*vpp = (struct vnode *)fp->f_data;
+ }
proc_fdunlock(p);
- return(0);
+ return 0;
}
int error;
proc_fdlock_spin(p);
- if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
if (fp->f_type != DTYPE_VNODE) {
- fp_drop(p, fd, fp,1);
+ fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return(EINVAL);
+ return EINVAL;
}
vp = (struct vnode *)fp->f_data;
- if (vpp != NULL)
+ if (vpp != NULL) {
*vpp = vp;
+ }
- if ((vidp != NULL) && (vp != NULLVP))
+ if ((vidp != NULL) && (vp != NULLVP)) {
*vidp = (uint32_t)vp->v_id;
+ }
proc_fdunlock(p);
- return(0);
+ return 0;
}
int error;
proc_fdlock_spin(p);
- if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
if (fp->f_type != DTYPE_SOCKET) {
- fp_drop(p, fd, fp,1);
+ fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return(ENOTSOCK);
+ return ENOTSOCK;
}
*sp = (struct socket *)fp->f_data;
proc_fdunlock(p);
- return(0);
+ return 0;
}
int
file_flags(int fd, int *flags)
{
-
proc_t p = current_proc();
struct fileproc *fp;
int error;
proc_fdlock_spin(p);
- if ( (error = fp_lookup(p, fd, &fp, 1)) ) {
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return(error);
+ return error;
}
*flags = (int)fp->f_flag;
- fp_drop(p, fd, fp,1);
+ fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return(0);
+ return 0;
}
{
struct fileproc *fp;
proc_t p = current_proc();
- int needwakeup = 0;
+ int needwakeup = 0;
proc_fdlock_spin(p);
if (fd < 0 || fd >= p->p_fd->fd_nfiles ||
- (fp = p->p_fd->fd_ofiles[fd]) == NULL ||
- ((p->p_fd->fd_ofileflags[fd] & UF_RESERVED) &&
- !(p->p_fd->fd_ofileflags[fd] & UF_CLOSING))) {
+ (fp = p->p_fd->fd_ofiles[fd]) == NULL ||
+ ((p->p_fd->fd_ofileflags[fd] & UF_RESERVED) &&
+ !(p->p_fd->fd_ofileflags[fd] & UF_CLOSING))) {
proc_fdunlock(p);
- return (EBADF);
+ return EBADF;
}
- fp->f_iocount --;
- if (fp->f_iocount == 0) {
- if (fp->f_flags & FP_SELCONFLICT)
+ if (1 == os_ref_release_locked(&fp->f_iocount)) {
+ if (fp->f_flags & FP_SELCONFLICT) {
fp->f_flags &= ~FP_SELCONFLICT;
+ }
if (p->p_fpdrainwait) {
p->p_fpdrainwait = 0;
}
proc_fdunlock(p);
- if (needwakeup)
- wakeup(&p->p_fpdrainwait);
- return(0);
+ if (needwakeup) {
+ wakeup(&p->p_fpdrainwait);
+ }
+ return 0;
}
int
falloc(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx)
{
- return (falloc_withalloc(p, resultfp, resultfd, ctx,
- fileproc_alloc_init, NULL));
+ return falloc_withalloc(p, resultfp, resultfd, ctx,
+ fileproc_alloc_init, NULL);
}
/*
resultfp, resultfd, ctx, fp_zalloc, arg, 1);
proc_fdunlock(p);
- return (error);
+ return error;
}
/*
*/
int
falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
- vfs_context_t ctx, int locked)
+ vfs_context_t ctx, int locked)
{
- return (falloc_withalloc_locked(p, resultfp, resultfd, ctx,
- fileproc_alloc_init, NULL, locked));
+ return falloc_withalloc_locked(p, resultfp, resultfd, ctx,
+ fileproc_alloc_init, NULL, locked);
}
static int
struct fileglob *fg;
int error, nfd;
- if (!locked)
+ if (nfiles >= maxfiles) {
+ tablefull("file");
+ return ENFILE;
+ }
+
+ if (!locked) {
proc_fdlock(p);
- if ( (error = fdalloc(p, 0, &nfd)) ) {
- if (!locked)
- proc_fdunlock(p);
- return (error);
}
- if (nfiles >= maxfiles) {
- if (!locked)
+
+ if ((error = fdalloc(p, 0, &nfd))) {
+ if (!locked) {
proc_fdunlock(p);
- tablefull("file");
- return (ENFILE);
+ }
+ return error;
}
+
#if CONFIG_MACF
error = mac_file_check_create(proc_ucred(p));
if (error) {
- if (!locked)
+ if (!locked) {
proc_fdunlock(p);
- return (error);
+ }
+ return error;
}
#endif
fp = (*fp_zalloc)(crarg);
if (fp == NULL) {
- if (locked)
+ if (locked) {
proc_fdlock(p);
- return (ENOMEM);
+ }
+ return ENOMEM;
}
MALLOC_ZONE(fg, struct fileglob *, sizeof(struct fileglob), M_FILEGLOB, M_WAITOK);
if (fg == NULL) {
fileproc_free(fp);
- if (locked)
+ if (locked) {
proc_fdlock(p);
- return (ENOMEM);
+ }
+ return ENOMEM;
}
bzero(fg, sizeof(struct fileglob));
lck_mtx_init(&fg->fg_lock, file_lck_grp, file_lck_attr);
- fp->f_iocount = 1;
+ os_ref_retain_locked(&fp->f_iocount);
fg->fg_count = 1;
fg->fg_ops = &uninitops;
fp->f_fglob = fg;
p->p_fd->fd_ofiles[nfd] = fp;
- if (!locked)
+ if (!locked) {
proc_fdunlock(p);
+ }
- if (resultfp)
+ if (resultfp) {
*resultfp = fp;
- if (resultfd)
+ }
+ if (resultfd) {
*resultfd = nfd;
+ }
- return (0);
+ return 0;
}
}
+/*
+ * fg_get_vnode
+ *
+ * Description: Return vnode associated with the file structure, if
+ * any. The lifetime of the returned vnode is bound to
+ * the lifetime of the file structure.
+ *
+ * Parameters: fg Pointer to fileglob to
+ * inspect
+ *
+ * Returns: vnode_t
+ */
+vnode_t
+fg_get_vnode(struct fileglob *fg)
+{
+ if (FILEGLOB_DTYPE(fg) == DTYPE_VNODE) {
+ return (vnode_t)fg->fg_data;
+ } else {
+ return NULL;
+ }
+}
+
/*
* fdexec
*
* Returns: void
*
* Locks: This function internally takes and drops proc_fdlock()
+ * But assumes tables don't grow/change while unlocked.
*
*/
void
-fdexec(proc_t p, short flags)
+fdexec(proc_t p, short flags, int self_exec)
{
struct filedesc *fdp = p->p_fd;
int i;
boolean_t cloexec_default = (flags & POSIX_SPAWN_CLOEXEC_DEFAULT) != 0;
+ thread_t self = current_thread();
+ struct uthread *ut = get_bsdthread_info(self);
+ struct kqworkq *dealloc_kqwq = NULL;
+
+ /*
+ * If the current thread is bound as a workq/workloop
+ * servicing thread, we need to unbind it first.
+ */
+ if (ut->uu_kqr_bound && self_exec) {
+ kqueue_threadreq_unbind(p, ut->uu_kqr_bound);
+ }
proc_fdlock(p);
- for (i = fdp->fd_lastfile; i >= 0; i--) {
+ /*
+ * Deallocate the knotes for this process
+ * and mark the tables non-existent so
+ * subsequent kqueue closes go faster.
+ */
+ knotes_dealloc(p);
+ assert(fdp->fd_knlistsize == 0);
+ assert(fdp->fd_knhashmask == 0);
+
+ for (i = fdp->fd_lastfile; i >= 0; i--) {
struct fileproc *fp = fdp->fd_ofiles[i];
char *flagp = &fdp->fd_ofileflags[i];
* except files marked explicitly as "inherit" and
* not marked close-on-exec.
*/
- if ((*flagp & (UF_EXCLOSE|UF_INHERIT)) != UF_INHERIT)
+ if ((*flagp & (UF_EXCLOSE | UF_INHERIT)) != UF_INHERIT) {
*flagp |= UF_EXCLOSE;
+ }
*flagp &= ~UF_INHERIT;
}
if (
- ((*flagp & (UF_RESERVED|UF_EXCLOSE)) == UF_EXCLOSE)
+ ((*flagp & (UF_RESERVED | UF_EXCLOSE)) == UF_EXCLOSE)
#if CONFIG_MACF
- || (fp && mac_file_check_inherit(proc_ucred(p), fp->f_fglob))
+ || (fp && mac_file_check_inherit(proc_ucred(p), fp->f_fglob))
#endif
- ) {
- if (i < fdp->fd_knlistsize)
- knote_fdclose(p, i, TRUE);
+ ) {
procfdtbl_clearfd(p, i);
- if (i == fdp->fd_lastfile && i > 0)
+ if (i == fdp->fd_lastfile && i > 0) {
fdp->fd_lastfile--;
- if (i < fdp->fd_freefile)
+ }
+ if (i < fdp->fd_freefile) {
fdp->fd_freefile = i;
+ }
/*
* Wait for any third party viewers (e.g., lsof)
* to release their references to this fileproc.
*/
- while (fp->f_iocount > 0) {
+ while (os_ref_get_count(&fp->f_iocount) > 1) {
p->p_fpdrainwait = 1;
msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO,
"fpdrain", NULL);
}
-
+ if (fp->f_flags & FP_WAITEVENT) {
+ (void)waitevent_close(p, fp);
+ }
closef_locked(fp, fp->f_fglob, p);
fileproc_free(fp);
}
}
+
+ /* release the per-process workq kq */
+ if (fdp->fd_wqkqueue) {
+ dealloc_kqwq = fdp->fd_wqkqueue;
+ fdp->fd_wqkqueue = NULL;
+ }
+
proc_fdunlock(p);
+
+ /* Anything to free? */
+ if (dealloc_kqwq) {
+ kqworkq_dealloc(dealloc_kqwq);
+ }
}
struct filedesc *newfdp, *fdp = p->p_fd;
int i;
struct fileproc *ofp, *fp;
- vnode_t v_dir;
+ vnode_t v_dir;
MALLOC_ZONE(newfdp, struct filedesc *,
- sizeof(*newfdp), M_FILEDESC, M_WAITOK);
- if (newfdp == NULL)
- return(NULL);
+ sizeof(*newfdp), M_FILEDESC, M_WAITOK);
+ if (newfdp == NULL) {
+ return NULL;
+ }
proc_fdlock(p);
* inherit the new current working directory from the current thread
* instead, before we take our references.
*/
- if (uth_cdir != NULLVP)
+ if (uth_cdir != NULLVP) {
newfdp->fd_cdir = uth_cdir;
+ }
/*
* For both fd_cdir and fd_rdir make sure we get
* and allows us to do the vnode_rele only on
* a properly referenced vp
*/
- if ( (v_dir = newfdp->fd_cdir) ) {
- if (vnode_getwithref(v_dir) == 0) {
- if ( (vnode_ref(v_dir)) )
- newfdp->fd_cdir = NULL;
+ if ((v_dir = newfdp->fd_cdir)) {
+ if (vnode_getwithref(v_dir) == 0) {
+ if ((vnode_ref(v_dir))) {
+ newfdp->fd_cdir = NULL;
+ }
vnode_put(v_dir);
- } else
- newfdp->fd_cdir = NULL;
+ } else {
+ newfdp->fd_cdir = NULL;
+ }
}
if (newfdp->fd_cdir == NULL && fdp->fd_cdir) {
- /*
+ /*
* we couldn't get a new reference on
* the current working directory being
* inherited... we might as well drop
* it useless... by dropping it we'll
* be that much closer to recycling it
*/
- vnode_rele(fdp->fd_cdir);
+ vnode_rele(fdp->fd_cdir);
fdp->fd_cdir = NULL;
}
- if ( (v_dir = newfdp->fd_rdir) ) {
+ if ((v_dir = newfdp->fd_rdir)) {
if (vnode_getwithref(v_dir) == 0) {
- if ( (vnode_ref(v_dir)) )
- newfdp->fd_rdir = NULL;
+ if ((vnode_ref(v_dir))) {
+ newfdp->fd_rdir = NULL;
+ }
vnode_put(v_dir);
} else {
- newfdp->fd_rdir = NULL;
+ newfdp->fd_rdir = NULL;
}
}
/* Coming from a chroot environment and unable to get a reference... */
if (newfdp->fd_rdir == NULL && fdp->fd_rdir) {
- /*
+ /*
* We couldn't get a new reference on
* the chroot directory being
* inherited... this is fatal, since
* escape from a chroot environment by
* the new process.
*/
- if (newfdp->fd_cdir)
- vnode_rele(newfdp->fd_cdir);
+ if (newfdp->fd_cdir) {
+ vnode_rele(newfdp->fd_cdir);
+ }
FREE_ZONE(newfdp, sizeof *newfdp, M_FILEDESC);
- return(NULL);
+ return NULL;
}
/*
* additional memory for the number of descriptors currently
* in use.
*/
- if (newfdp->fd_lastfile < NDFILE)
+ if (newfdp->fd_lastfile < NDFILE) {
i = NDFILE;
- else {
+ } else {
/*
* Compute the smallest multiple of NDEXTENT needed
* for the file descriptors currently in use,
* allowing the table to shrink.
*/
i = newfdp->fd_nfiles;
- while (i > 1 + 2 * NDEXTENT && i > 1 + newfdp->fd_lastfile * 2)
+ while (i > 1 + 2 * NDEXTENT && i > 1 + newfdp->fd_lastfile * 2) {
i /= 2;
+ }
}
proc_fdunlock(p);
MALLOC_ZONE(newfdp->fd_ofiles, struct fileproc **,
- i * OFILESIZE, M_OFILETABL, M_WAITOK);
+ i * OFILESIZE, M_OFILETABL, M_WAITOK);
if (newfdp->fd_ofiles == NULL) {
- if (newfdp->fd_cdir)
- vnode_rele(newfdp->fd_cdir);
- if (newfdp->fd_rdir)
+ if (newfdp->fd_cdir) {
+ vnode_rele(newfdp->fd_cdir);
+ }
+ if (newfdp->fd_rdir) {
vnode_rele(newfdp->fd_rdir);
+ }
FREE_ZONE(newfdp, sizeof(*newfdp), M_FILEDESC);
- return(NULL);
+ return NULL;
}
(void) memset(newfdp->fd_ofiles, 0, i * OFILESIZE);
proc_fdlock(p);
char *flags;
(void) memcpy(newfdp->fd_ofiles, fdp->fd_ofiles,
- (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofiles));
+ (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofiles));
(void) memcpy(newfdp->fd_ofileflags, fdp->fd_ofileflags,
- (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofileflags));
-
- /*
- * kq descriptors cannot be copied.
- */
- if (newfdp->fd_knlistsize != -1) {
- fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
- flags = &newfdp->fd_ofileflags[newfdp->fd_lastfile];
- for (i = newfdp->fd_lastfile;
- i >= 0; i--, fpp--, flags--) {
- if (*flags & UF_RESERVED)
- continue; /* (removed below) */
- if (*fpp != NULL && (*fpp)->f_type == DTYPE_KQUEUE) {
- *fpp = NULL;
- *flags = 0;
- if (i < newfdp->fd_freefile)
- newfdp->fd_freefile = i;
- }
- if (*fpp == NULL && i == newfdp->fd_lastfile && i > 0)
- newfdp->fd_lastfile--;
- }
- newfdp->fd_knlist = NULL;
- newfdp->fd_knlistsize = -1;
- newfdp->fd_knhash = NULL;
- newfdp->fd_knhashmask = 0;
- }
- fpp = newfdp->fd_ofiles;
- flags = newfdp->fd_ofileflags;
+ (newfdp->fd_lastfile + 1) * sizeof(*fdp->fd_ofileflags));
- for (i = newfdp->fd_lastfile + 1; --i >= 0; fpp++, flags++)
+ fpp = &newfdp->fd_ofiles[newfdp->fd_lastfile];
+ flags = &newfdp->fd_ofileflags[newfdp->fd_lastfile];
+ for (i = newfdp->fd_lastfile; i >= 0; i--, fpp--, flags--) {
if ((ofp = *fpp) != NULL &&
0 == (ofp->f_fglob->fg_lflags & FG_CONFINED) &&
- 0 == (*flags & (UF_FORKCLOSE|UF_RESERVED))) {
+ 0 == (*flags & (UF_FORKCLOSE | UF_RESERVED))) {
#if DEBUG
- if (FILEPROC_TYPE(ofp) != FTYPE_SIMPLE)
+ if (FILEPROC_TYPE(ofp) != FTYPE_SIMPLE) {
panic("complex fileproc");
+ }
#endif
fp = fileproc_alloc_init(NULL);
if (fp == NULL) {
*fpp = fp;
}
} else {
- if (i < newfdp->fd_freefile)
- newfdp->fd_freefile = i;
*fpp = NULL;
*flags = 0;
}
+ if (*fpp == NULL) {
+ if (i == newfdp->fd_lastfile && i > 0) {
+ newfdp->fd_lastfile--;
+ }
+ if (i < newfdp->fd_freefile) {
+ newfdp->fd_freefile = i;
+ }
+ }
+ }
}
proc_fdunlock(p);
- return (newfdp);
+
+ /*
+ * Initialize knote and kqueue tracking structs
+ */
+ newfdp->fd_knlist = NULL;
+ newfdp->fd_knlistsize = 0;
+ newfdp->fd_knhash = NULL;
+ newfdp->fd_knhashmask = 0;
+ newfdp->fd_kqhash = NULL;
+ newfdp->fd_kqhashmask = 0;
+ newfdp->fd_wqkqueue = NULL;
+ lck_mtx_init(&newfdp->fd_kqhashlock, proc_kqhashlock_grp, proc_lck_attr);
+ lck_mtx_init(&newfdp->fd_knhashlock, proc_knhashlock_grp, proc_lck_attr);
+
+ return newfdp;
}
{
struct filedesc *fdp;
struct fileproc *fp;
+ struct kqworkq *dealloc_kqwq = NULL;
int i;
proc_fdlock(p);
if (p == kernproc || NULL == (fdp = p->p_fd)) {
- proc_fdunlock(p);
+ proc_fdunlock(p);
return;
}
extern struct filedesc filedesc0;
- if (&filedesc0 == fdp)
+ if (&filedesc0 == fdp) {
panic("filedesc0");
+ }
- if (fdp->fd_nfiles > 0 && fdp->fd_ofiles) {
- for (i = fdp->fd_lastfile; i >= 0; i--) {
+ /*
+ * deallocate all the knotes up front and claim empty
+ * tables to make any subsequent kqueue closes faster.
+ */
+ knotes_dealloc(p);
+ assert(fdp->fd_knlistsize == 0);
+ assert(fdp->fd_knhashmask == 0);
- /* May still have knotes for fd without open file */
- if (i < fdp->fd_knlistsize)
- knote_fdclose(p, i, TRUE);
+ /*
+ * dealloc all workloops that have outstanding retains
+ * when created with scheduling parameters.
+ */
+ kqworkloops_dealloc(p);
+ /* close file descriptors */
+ if (fdp->fd_nfiles > 0 && fdp->fd_ofiles) {
+ for (i = fdp->fd_lastfile; i >= 0; i--) {
if ((fp = fdp->fd_ofiles[i]) != NULL) {
+ if (fdp->fd_ofileflags[i] & UF_RESERVED) {
+ panic("fdfree: found fp with UF_RESERVED");
+ }
- if (fdp->fd_ofileflags[i] & UF_RESERVED)
- panic("fdfree: found fp with UF_RESERVED");
-
+ fileproc_drain(p, fp);
procfdtbl_reservefd(p, i);
- if (fp->f_flags & FP_WAITEVENT)
+ if (fp->f_flags & FP_WAITEVENT) {
(void)waitevent_close(p, fp);
+ }
(void) closef_locked(fp, fp->f_fglob, p);
fileproc_free(fp);
}
fdp->fd_nfiles = 0;
}
+ if (fdp->fd_wqkqueue) {
+ dealloc_kqwq = fdp->fd_wqkqueue;
+ fdp->fd_wqkqueue = NULL;
+ }
+
proc_fdunlock(p);
- if (fdp->fd_cdir)
- vnode_rele(fdp->fd_cdir);
- if (fdp->fd_rdir)
+ if (dealloc_kqwq) {
+ kqworkq_dealloc(dealloc_kqwq);
+ }
+ if (fdp->fd_cdir) {
+ vnode_rele(fdp->fd_cdir);
+ }
+ if (fdp->fd_rdir) {
vnode_rele(fdp->fd_rdir);
+ }
proc_fdlock_spin(p);
p->p_fd = NULL;
proc_fdunlock(p);
- if (fdp->fd_knlist)
- FREE(fdp->fd_knlist, M_KQUEUE);
- if (fdp->fd_knhash)
- FREE(fdp->fd_knhash, M_KQUEUE);
+ if (fdp->fd_kqhash) {
+ for (uint32_t j = 0; j <= fdp->fd_kqhashmask; j++) {
+ assert(LIST_EMPTY(&fdp->fd_kqhash[j]));
+ }
+ FREE(fdp->fd_kqhash, M_KQUEUE);
+ }
+
+ lck_mtx_destroy(&fdp->fd_kqhashlock, proc_kqhashlock_grp);
+ lck_mtx_destroy(&fdp->fd_knhashlock, proc_knhashlock_grp);
FREE_ZONE(fdp, sizeof(*fdp), M_FILEDESC);
}
* p Pointer to proc structure
*
* Returns: 0 Success
- * closef_finish:??? Anything returnable by a per-fileops
+ * closef_finish:??? Anything returnable by a per-fileops
* close function
*
* Note: Decrements reference count on file structure; if this was the
int error;
if (fg == NULL) {
- return (0);
+ return 0;
}
/* Set up context with cred stashed in fg */
- if (p == current_proc())
+ if (p == current_proc()) {
context.vc_thread = current_thread();
- else
+ } else {
context.vc_thread = NULL;
+ }
context.vc_ucred = fg->fg_cred;
/*
lf.l_type = F_UNLCK;
vp = (struct vnode *)fg->fg_data;
- if ( (error = vnode_getwithref(vp)) == 0 ) {
+ if ((error = vnode_getwithref(vp)) == 0) {
(void) VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX, &context, NULL);
(void)vnode_put(vp);
}
if (fg->fg_count > 0) {
lck_mtx_unlock(&fg->fg_lock);
- return (0);
+ return 0;
}
#if DIAGNOSTIC
- if (fg->fg_count != 0)
+ if (fg->fg_count != 0) {
panic("fg %p: being freed with bad fg_count (%d)", fg, fg->fg_count);
+ }
#endif
- if (fp && (fp->f_flags & FP_WRITTEN))
- fg->fg_flag |= FWASWRITTEN;
+ if (fp && (fp->f_flags & FP_WRITTEN)) {
+ fg->fg_flag |= FWASWRITTEN;
+ }
fg->fg_lflags |= FG_TERM;
lck_mtx_unlock(&fg->fg_lock);
- if (p)
+ if (p) {
proc_fdunlock(p);
+ }
/* Since we ensure that fg->fg_ops is always initialized,
* it is safe to invoke fo_close on the fg */
fg_free(fg);
- if (p)
+ if (p) {
proc_fdlock(p);
+ }
- return(error);
+ return error;
}
* close to prevent fd's from being closed out from under
* operations currently in progress and blocked
*
- * See Also: file_vnode(), file_socket(), file_drop(), and the cautions
+ * See Also: file_vnode(), file_socket(), file_drop(), and the cautions
* regarding their use and interaction with this function.
*/
void
{
struct vfs_context context;
- context.vc_thread = proc_thread(p); /* XXX */
+ context.vc_thread = proc_thread(p); /* XXX */
context.vc_ucred = fp->f_fglob->fg_cred;
- fp->f_iocount-- ; /* (the one the close holds) */
-
- while (fp->f_iocount) {
+ /* Set the vflag for drain */
+ fileproc_modify_vflags(fp, FPV_DRAIN, FALSE);
- lck_mtx_convert_spin(&p->p_fdmlock);
+ while (os_ref_get_count(&fp->f_iocount) > 1) {
+ lck_mtx_convert_spin(&p->p_fdmlock);
- if (fp->f_fglob->fg_ops->fo_drain) {
- (*fp->f_fglob->fg_ops->fo_drain)(fp, &context);
- }
+ fo_drain(fp, &context);
if ((fp->f_flags & FP_INSELECT) == FP_INSELECT) {
if (waitq_wakeup64_all((struct waitq *)fp->f_wset, NO_EVENT64,
- THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT)
+ THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT) {
panic("bad wait queue for waitq_wakeup64_all %p (fp:%p)", fp->f_wset, fp);
+ }
}
if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
if (waitq_wakeup64_all(&select_conflict_queue, NO_EVENT64,
- THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT)
+ THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT) {
panic("bad select_conflict_queue");
+ }
}
p->p_fpdrainwait = 1;
msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO, "fpdrain", NULL);
-
}
#if DIAGNOSTIC
- if ((fp->f_flags & FP_INSELECT) != 0)
+ if ((fp->f_flags & FP_INSELECT) != 0) {
panic("FP_INSELECT set on drained fp");
+ }
#endif
- if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT)
+ if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
fp->f_flags &= ~FP_SELCONFLICT;
+ }
}
* Parameters: p Process containing fd
* fd fd to be released
* fp fileproc to be freed
- *
- * Returns: 0 Success
- *
- * Notes: XXX function should be void - no one interprets the returns
- * XXX code
*/
-int
+void
fp_free(proc_t p, int fd, struct fileproc * fp)
{
- proc_fdlock_spin(p);
+ proc_fdlock_spin(p);
fdrelse(p, fd);
- proc_fdunlock(p);
+ proc_fdunlock(p);
fg_free(fp->f_fglob);
+ os_ref_release_live(&fp->f_iocount);
fileproc_free(fp);
- return(0);
}
struct vnode *vp;
struct flock lf;
vfs_context_t ctx = vfs_context_current();
- int error=0;
+ int error = 0;
AUDIT_ARG(fd, uap->fd);
- if ( (error = fp_getfvp(p, fd, &fp, &vp)) ) {
- return(error);
+ if ((error = fp_getfvp(p, fd, &fp, &vp))) {
+ return error;
}
- if ( (error = vnode_getwithref(vp)) ) {
+ if ((error = vnode_getwithref(vp))) {
goto out1;
}
AUDIT_ARG(vnpath, vp, ARG_VNODE1);
error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
goto out;
}
- if (how & LOCK_EX)
+ if (how & LOCK_EX) {
lf.l_type = F_WRLCK;
- else if (how & LOCK_SH)
+ } else if (how & LOCK_SH) {
lf.l_type = F_RDLCK;
- else {
- error = EBADF;
+ } else {
+ error = EBADF;
goto out;
}
#if CONFIG_MACF
error = mac_file_check_lock(proc_ucred(p), fp->f_fglob, F_SETLK, &lf);
- if (error)
+ if (error) {
goto out;
+ }
#endif
error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf,
(how & LOCK_NB ? F_FLOCK : F_FLOCK | F_WAIT),
ctx, NULL);
- if (!error)
+ if (!error) {
fp->f_flag |= FHASLOCK;
+ }
out:
(void)vnode_put(vp);
out1:
fp_drop(p, fd, fp, 0);
- return(error);
-
+ return error;
}
/*
* Description: Obtain a Mach send right for a given file descriptor.
*
* Parameters: p Process calling fileport
- * uap->fd The fd to reference
- * uap->portnamep User address at which to place port name.
+ * uap->fd The fd to reference
+ * uap->portnamep User address at which to place port name.
*
* Returns: 0 Success.
- * EBADF Bad file descriptor.
- * EINVAL File descriptor had type that cannot be sent, misc. other errors.
- * EFAULT Address at which to store port name is not valid.
- * EAGAIN Resource shortage.
+ * EBADF Bad file descriptor.
+ * EINVAL File descriptor had type that cannot be sent, misc. other errors.
+ * EFAULT Address at which to store port name is not valid.
+ * EAGAIN Resource shortage.
*
* Implicit returns:
* On success, name of send right is stored at user-specified address.
return;
}
-
/*
- * fileport_makefd
+ * fileport_makefd_internal
*
* Description: Obtain the file descriptor for a given Mach send right.
*
- * Parameters: p Process calling fileport
- * uap->port Name of send right to file port.
- *
* Returns: 0 Success
* EINVAL Invalid Mach port name, or port is not for a file.
* fdalloc:EMFILE
* *retval (modified) The new descriptor
*/
int
-fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval)
+fileport_makefd_internal(proc_t p, ipc_port_t port, int uf_flags, int *retval)
{
struct fileglob *fg;
- struct fileproc *fp = FILEPROC_NULL;
- ipc_port_t port = IPC_PORT_NULL;
- mach_port_name_t send = uap->port;
- kern_return_t res;
+ struct fileproc *fp = FILEPROC_NULL;
int fd;
int err;
- res = ipc_object_copyin(get_task_ipcspace(p->task),
- send, MACH_MSG_TYPE_COPY_SEND, &port);
-
- if (res != KERN_SUCCESS) {
- err = EINVAL;
- goto out;
- }
-
fg = fileport_port_to_fileglob(port);
if (fg == NULL) {
err = EINVAL;
fp->f_fglob = fg;
fg_ref(fp);
- proc_fdlock(p);
+ proc_fdlock(p);
err = fdalloc(p, 0, &fd);
if (err != 0) {
proc_fdunlock(p);
+ fg_drop(fp);
goto out;
}
- *fdflags(p, fd) |= UF_EXCLOSE;
+ if (uf_flags) {
+ *fdflags(p, fd) |= uf_flags;
+ }
procfdtbl_releasefd(p, fd, fp);
proc_fdunlock(p);
fileproc_free(fp);
}
+ return err;
+}
+
+/*
+ * fileport_makefd
+ *
+ * Description: Obtain the file descriptor for a given Mach send right.
+ *
+ * Parameters: p Process calling fileport
+ * uap->port Name of send right to file port.
+ *
+ * Returns: 0 Success
+ * EINVAL Invalid Mach port name, or port is not for a file.
+ * fdalloc:EMFILE
+ * fdalloc:ENOMEM Unable to allocate fileproc or extend file table.
+ *
+ * Implicit returns:
+ * *retval (modified) The new descriptor
+ */
+int
+fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval)
+{
+ ipc_port_t port = IPC_PORT_NULL;
+ mach_port_name_t send = uap->port;
+ kern_return_t res;
+ int err;
+
+ res = ipc_object_copyin(get_task_ipcspace(p->task),
+ send, MACH_MSG_TYPE_COPY_SEND, &port, 0, NULL, IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND);
+
+ if (res == KERN_SUCCESS) {
+ err = fileport_makefd_internal(p, port, UF_EXCLOSE, retval);
+ } else {
+ err = EINVAL;
+ }
+
if (IPC_PORT_NULL != port) {
ipc_port_release_send(port);
}
fp = fdp->fd_ofiles[indx];
if (dfd < 0 || dfd >= fdp->fd_nfiles ||
- (wfp = fdp->fd_ofiles[dfd]) == NULL || wfp == fp ||
- (fdp->fd_ofileflags[dfd] & UF_RESERVED)) {
-
- proc_fdunlock(p);
- return (EBADF);
+ (wfp = fdp->fd_ofiles[dfd]) == NULL || wfp == fp ||
+ (fdp->fd_ofileflags[dfd] & UF_RESERVED)) {
+ proc_fdunlock(p);
+ return EBADF;
}
#if CONFIG_MACF
myerror = mac_file_check_dup(proc_ucred(p), wfp->f_fglob, dfd);
if (myerror) {
proc_fdunlock(p);
- return (myerror);
+ return myerror;
}
#endif
/*
case ENODEV:
if (FP_ISGUARDED(wfp, GUARD_DUP)) {
proc_fdunlock(p);
- return (EPERM);
+ return EPERM;
}
/*
* Check that the mode the file is being opened for is a
* subset of the mode of the existing descriptor.
*/
- if (((flags & (FREAD|FWRITE)) | wfp->f_flag) != wfp->f_flag) {
- proc_fdunlock(p);
- return (EACCES);
+ if (((flags & (FREAD | FWRITE)) | wfp->f_flag) != wfp->f_flag) {
+ proc_fdunlock(p);
+ return EACCES;
}
- if (indx > fdp->fd_lastfile)
+ if (indx > fdp->fd_lastfile) {
fdp->fd_lastfile = indx;
+ }
(void)fg_ref(wfp);
- if (fp->f_fglob)
- fg_free(fp->f_fglob);
+ if (fp->f_fglob) {
+ fg_free(fp->f_fglob);
+ }
fp->f_fglob = wfp->f_fglob;
fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd] |
- (flags & O_CLOEXEC) ? UF_EXCLOSE : 0;
+ (flags & O_CLOEXEC) ? UF_EXCLOSE : 0;
- proc_fdunlock(p);
- return (0);
+ proc_fdunlock(p);
+ return 0;
default:
- proc_fdunlock(p);
- return (error);
+ proc_fdunlock(p);
+ return error;
}
/* NOTREACHED */
}
lck_mtx_lock_spin(&fg->fg_lock);
#if DIAGNOSTIC
- if ((fp->f_flags & ~((unsigned int)FP_VALID_FLAGS)) != 0)
+ if ((fp->f_flags & ~((unsigned int)FP_VALID_FLAGS)) != 0) {
panic("fg_ref: invalid bits on fp %p", fp);
+ }
- if (fg->fg_count == 0)
+ if (fg->fg_count == 0) {
panic("fg_ref: adding fgcount to zeroed fg: fp %p fg %p",
fp, fg);
+ }
#endif
fg->fg_count++;
lck_mtx_unlock(&fg->fg_lock);
insert = TRUE;
}
lck_mtx_unlock(&fg->fg_lock);
- return (insert);
+ return insert;
}
/*
remove = TRUE;
}
lck_mtx_unlock(&fg->fg_lock);
- return (remove);
+ return remove;
}
/*
int
fo_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
{
- return ((*fp->f_ops->fo_read)(fp, uio, flags, ctx));
+ return (*fp->f_ops->fo_read)(fp, uio, flags, ctx);
+}
+
+int
+fo_no_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
+{
+#pragma unused(fp, uio, flags, ctx)
+ return ENXIO;
}
int
fo_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
{
- return((*fp->f_ops->fo_write)(fp, uio, flags, ctx));
+ return (*fp->f_ops->fo_write)(fp, uio, flags, ctx);
+}
+
+int
+fo_no_write(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
+{
+#pragma unused(fp, uio, flags, ctx)
+ return ENXIO;
}
proc_fdunlock(vfs_context_proc(ctx));
error = (*fp->f_ops->fo_ioctl)(fp, com, data, ctx);
proc_fdlock(vfs_context_proc(ctx));
- return(error);
+ return error;
+}
+
+int
+fo_no_ioctl(struct fileproc *fp, u_long com, caddr_t data, vfs_context_t ctx)
+{
+#pragma unused(fp, com, data, ctx)
+ return ENOTTY;
}
int
fo_select(struct fileproc *fp, int which, void *wql, vfs_context_t ctx)
{
- return((*fp->f_ops->fo_select)(fp, which, wql, ctx));
+ return (*fp->f_ops->fo_select)(fp, which, wql, ctx);
+}
+
+int
+fo_no_select(struct fileproc *fp, int which, void *wql, vfs_context_t ctx)
+{
+#pragma unused(fp, which, wql, ctx)
+ return ENOTSUP;
}
int
fo_close(struct fileglob *fg, vfs_context_t ctx)
{
- return((*fg->fg_ops->fo_close)(fg, ctx));
+ return (*fg->fg_ops->fo_close)(fg, ctx);
+}
+
+
+/*
+ * fo_drain
+ *
+ * Description: Generic fileops kqueue filter indirected through the fileops
+ * pointer in the fileproc structure
+ *
+ * Parameters: fp fileproc structure pointer
+ * ctx VFS context for operation
+ *
+ * Returns: 0 Success
+ * !0 errno from drain
+ */
+int
+fo_drain(struct fileproc *fp, vfs_context_t ctx)
+{
+ return (*fp->f_ops->fo_drain)(fp, ctx);
+}
+
+int
+fo_no_drain(struct fileproc *fp, vfs_context_t ctx)
+{
+#pragma unused(fp, ctx)
+ return ENOTSUP;
}
*
* Parameters: fp fileproc structure pointer
* kn pointer to knote to filter on
- * ctx VFS context for operation
*
* Returns: (kn->kn_flags & EV_ERROR) error in kn->kn_data
* 0 Filter is not active
* !0 Filter is active
*/
int
-fo_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx)
+fo_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
{
- return ((*fp->f_ops->fo_kqfilter)(fp, kn, ctx));
+ return (*fp->f_ops->fo_kqfilter)(fp, kn, kev);
}
+int
+fo_no_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
+{
+#pragma unused(fp, kev)
+ knote_set_error(kn, ENOTSUP);
+ return 0;
+}
+
+
/*
* The ability to send a file descriptor to another
* process is opt-in by file type.
case DTYPE_SOCKET:
case DTYPE_PIPE:
case DTYPE_PSXSHM:
- return (0 == (fp->f_fglob->fg_lflags & FG_CONFINED));
+ case DTYPE_NETPOLICY:
+ return 0 == (fp->f_fglob->fg_lflags & FG_CONFINED);
default:
/* DTYPE_KQUEUE, DTYPE_FSEVENTS, DTYPE_PSXSEM */
return FALSE;
}
}
+os_refgrp_decl(, f_iocount_refgrp, "f_iocount", NULL);
struct fileproc *
fileproc_alloc_init(__unused void *arg)
{
struct fileproc *fp;
- MALLOC_ZONE(fp, struct fileproc *, sizeof (*fp), M_FILEPROC, M_WAITOK);
- if (fp)
- bzero(fp, sizeof (*fp));
+ MALLOC_ZONE(fp, struct fileproc *, sizeof(*fp), M_FILEPROC, M_WAITOK);
+ if (fp) {
+ bzero(fp, sizeof(*fp));
+ os_ref_init(&fp->f_iocount, &f_iocount_refgrp);
+ }
- return (fp);
+ return fp;
}
+
void
fileproc_free(struct fileproc *fp)
{
+ os_ref_count_t __unused refc = os_ref_release(&fp->f_iocount);
+#if DEVELOPMENT || DEBUG
+ if (0 != refc) {
+ panic("%s: pid %d refc: %u != 0",
+ __func__, proc_pid(current_proc()), refc);
+ }
+#endif
switch (FILEPROC_TYPE(fp)) {
case FTYPE_SIMPLE:
- FREE_ZONE(fp, sizeof (*fp), M_FILEPROC);
+ FREE_ZONE(fp, sizeof(*fp), M_FILEPROC);
break;
case FTYPE_GUARDED:
guarded_fileproc_free(fp);
panic("%s: corrupt fp %p flags %x", __func__, fp, fp->f_flags);
}
}
+
+void
+fileproc_modify_vflags(struct fileproc *fp, fileproc_vflags_t vflags, boolean_t clearflags)
+{
+ if (clearflags) {
+ os_atomic_andnot(&fp->f_vflags, vflags, relaxed);
+ } else {
+ os_atomic_or(&fp->f_vflags, vflags, relaxed);
+ }
+}
+
+fileproc_vflags_t
+fileproc_get_vflags(struct fileproc *fp)
+{
+ return os_atomic_load(&fp->f_vflags, relaxed);
+}