#include <sys/uio_internal.h>
#include <sys/codesign.h>
#include <sys/codedir_internal.h>
-
-#include <security/audit/audit.h>
-
#include <sys/mount_internal.h>
#include <sys/kdebug.h>
#include <sys/sysproto.h>
#include <sys/pipe.h>
#include <sys/spawn.h>
#include <sys/cprotect.h>
+#include <sys/ubc_internal.h>
+
#include <kern/kern_types.h>
#include <kern/kalloc.h>
#include <kern/waitq.h>
-#include <libkern/OSAtomic.h>
-
-#include <sys/ubc_internal.h>
-
#include <kern/ipc_misc.h>
-#include <vm/vm_protos.h>
+#include <vm/vm_protos.h>
#include <mach/mach_port.h>
-#include <stdbool.h>
+#include <security/audit/audit.h>
#if CONFIG_MACF
#include <security/mac_framework.h>
#endif
+#include <stdbool.h>
+#include <os/atomic_private.h>
+#include <IOKit/IOBSD.h>
+
+#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;
-struct pshmnode;
-
+static void fileproc_drain(proc_t, struct fileproc *);
static int finishdup(proc_t p,
struct filedesc *fdp, int old, int new, int flags, int32_t *retval);
-int falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx, int locked);
-void fg_drop(struct fileproc * fp);
-void fg_free(struct fileglob *fg);
-void fg_ref(struct fileproc * fp);
void fileport_releasefg(struct fileglob *fg);
-/* flags for close_internal_locked */
+/* flags for fp_close_and_unlock */
#define FD_DUP2RESV 1
/* We don't want these exported */
__private_extern__
int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
-static void _fdrelse(struct proc * p, int fd);
-
+static void fdrelse(struct proc * p, int fd);
extern void file_lock_init(void);
#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_cred f_fglob->fg_cred
-#define f_ops f_fglob->fg_ops
-#define f_offset f_fglob->fg_offset
-#define f_data f_fglob->fg_data
+#define f_flag fp_glob->fg_flag
+#define f_type fp_glob->fg_ops->fo_type
+#define f_cred fp_glob->fg_cred
+#define f_ops fp_glob->fg_ops
+#define f_offset fp_glob->fg_offset
+#define f_data fp_glob->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)
+
+ZONE_DECLARE(fg_zone, "fileglob",
+ sizeof(struct fileglob), ZC_NOENCRYPT | ZC_ZFREE_CLEARMEM);
+ZONE_DECLARE(fp_zone, "fileproc",
+ sizeof(struct fileproc), ZC_NOENCRYPT | ZC_ZFREE_CLEARMEM);
+ZONE_DECLARE(fdp_zone, "filedesc",
+ sizeof(struct filedesc), ZC_NOENCRYPT | ZC_ZFREE_CLEARMEM);
+
/*
* 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 */
+/*
+ * "uninitialized" ops -- ensure FILEGLOB_DTYPE(fg) always exists
+ */
+static const struct fileops uninitops;
-
+os_refgrp_decl(, f_refgrp, "files refcounts", NULL);
lck_grp_attr_t * file_lck_grp_attr;
lck_grp_t * file_lck_grp;
lck_attr_t * file_lck_attr;
-lck_mtx_t * uipc_lock;
+#pragma mark fileglobs
+
+/*!
+ * @function fg_free
+ *
+ * @brief
+ * Free a file structure.
+ */
+static void
+fg_free(struct fileglob *fg)
+{
+ os_atomic_dec(&nfiles, relaxed);
+
+ if (fg->fg_vn_data) {
+ fg_vn_data_free(fg->fg_vn_data);
+ fg->fg_vn_data = NULL;
+ }
+
+ if (IS_VALID_CRED(fg->fg_cred)) {
+ kauth_cred_unref(&fg->fg_cred);
+ }
+ lck_mtx_destroy(&fg->fg_lock, file_lck_grp);
+
+#if CONFIG_MACF
+ mac_file_label_destroy(fg);
+#endif
+ zfree(fg_zone, fg);
+}
+
+OS_ALWAYS_INLINE
+void
+fg_ref(proc_t p, struct fileglob *fg)
+{
+#if DEBUG || DEVELOPMENT
+ proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
+#else
+ (void)p;
+#endif
+ os_ref_retain_raw(&fg->fg_count, &f_refgrp);
+}
+
+void
+fg_drop_live(struct fileglob *fg)
+{
+ os_ref_release_live_raw(&fg->fg_count, &f_refgrp);
+}
+
+int
+fg_drop(proc_t p, struct fileglob *fg)
+{
+ struct vnode *vp;
+ struct vfs_context context;
+ int error = 0;
+
+ if (fg == NULL) {
+ return 0;
+ }
+
+ /* Set up context with cred stashed in fg */
+ if (p == current_proc()) {
+ context.vc_thread = current_thread();
+ } else {
+ context.vc_thread = NULL;
+ }
+ context.vc_ucred = fg->fg_cred;
+
+ /*
+ * POSIX record locking dictates that any close releases ALL
+ * locks owned by this process. This is handled by setting
+ * a flag in the unlock to free ONLY locks obeying POSIX
+ * semantics, and not to free BSD-style file locks.
+ * If the descriptor was in a message, POSIX-style locks
+ * aren't passed with the descriptor.
+ */
+ if (p && DTYPE_VNODE == FILEGLOB_DTYPE(fg) &&
+ (p->p_ladvflag & P_LADVLOCK)) {
+ struct flock lf = {
+ .l_whence = SEEK_SET,
+ .l_type = F_UNLCK,
+ };
+
+ vp = (struct vnode *)fg->fg_data;
+ 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 (os_ref_release_raw(&fg->fg_count, &f_refgrp) == 0) {
+ /*
+ * Since we ensure that fg->fg_ops is always initialized,
+ * it is safe to invoke fo_close on the fg
+ */
+ error = fo_close(fg, &context);
+
+ fg_free(fg);
+ }
+
+ return error;
+}
+
+/*
+ * 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;
+ }
+}
+
+bool
+fg_sendable(struct fileglob *fg)
+{
+ switch (FILEGLOB_DTYPE(fg)) {
+ case DTYPE_VNODE:
+ case DTYPE_SOCKET:
+ case DTYPE_PIPE:
+ case DTYPE_PSXSHM:
+ case DTYPE_NETPOLICY:
+ return (fg->fg_lflags & FG_CONFINED) == 0;
+
+ default:
+ return false;
+ }
+}
+
+#pragma mark fileprocs
/*
* check_file_seek_range
/* Allocate file lock attribute */
file_lck_attr = lck_attr_alloc_init();
+}
+
+
+void
+proc_dirs_lock_shared(proc_t p)
+{
+ lck_rw_lock_shared(&p->p_dirs_lock);
+}
+
+void
+proc_dirs_unlock_shared(proc_t p)
+{
+ lck_rw_unlock_shared(&p->p_dirs_lock);
+}
- uipc_lock = lck_mtx_alloc_init(file_lck_grp, file_lck_attr);
+void
+proc_dirs_lock_exclusive(proc_t p)
+{
+ lck_rw_lock_exclusive(&p->p_dirs_lock);
}
+void
+proc_dirs_unlock_exclusive(proc_t p)
+{
+ lck_rw_unlock_exclusive(&p->p_dirs_lock);
+}
/*
* proc_fdlock, proc_fdlock_spin
lck_mtx_unlock(&p->p_fdmlock);
}
+struct fdt_iterator
+fdt_next(proc_t p, int fd, bool only_settled)
+{
+ struct fdt_iterator it;
+ struct filedesc *fdp = p->p_fd;
+ struct fileproc *fp;
+ int nfds = min(fdp->fd_lastfile + 1, fdp->fd_nfiles);
+
+ while (++fd < nfds) {
+ fp = fdp->fd_ofiles[fd];
+ if (fp == NULL || fp->fp_glob == NULL) {
+ continue;
+ }
+ if (only_settled && (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ continue;
+ }
+ it.fdti_fd = fd;
+ it.fdti_fp = fp;
+ return it;
+ }
+
+ it.fdti_fd = nfds;
+ it.fdti_fp = NULL;
+ return it;
+}
+
+struct fdt_iterator
+fdt_prev(proc_t p, int fd, bool only_settled)
+{
+ struct fdt_iterator it;
+ struct filedesc *fdp = p->p_fd;
+ struct fileproc *fp;
+
+ while (--fd >= 0) {
+ fp = fdp->fd_ofiles[fd];
+ if (fp == NULL || fp->fp_glob == NULL) {
+ continue;
+ }
+ if (only_settled && (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ continue;
+ }
+ it.fdti_fd = fd;
+ it.fdti_fp = fp;
+ return it;
+ }
+
+ it.fdti_fd = -1;
+ it.fdti_fp = NULL;
+ return it;
+}
/*
* System calls on descriptors.
/*
- * getdtablesize
+ * sys_getdtablesize
*
* Description: Returns the per process maximum size of the descriptor table
*
* *retval (modified) Size of dtable
*/
int
-getdtablesize(proc_t p, __unused struct getdtablesize_args *uap, int32_t *retval)
+sys_getdtablesize(proc_t p, __unused struct getdtablesize_args *uap, int32_t *retval)
{
- proc_fdlock_spin(p);
- *retval = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
- proc_fdunlock(p);
+ *retval = (int32_t)MIN(proc_limitgetcur(p, RLIMIT_NOFILE, TRUE), maxfilesperproc);
return 0;
}
-void
+static void
procfdtbl_reservefd(struct proc * p, int fd)
{
p->p_fd->fd_ofiles[fd] = NULL;
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);
-}
-
void
procfdtbl_releasefd(struct proc * p, int fd, struct fileproc * fp)
{
}
}
-void
+static void
procfdtbl_waitfd(struct proc * p, int fd)
{
p->p_fd->fd_ofileflags[fd] |= UF_RESVWAIT;
msleep(&p->p_fd, &p->p_fdmlock, PRIBIO, "ftbl_waitfd", NULL);
}
-
-void
+static void
procfdtbl_clearfd(struct proc * p, int fd)
{
int waiting;
}
/*
- * _fdrelse
+ * fdrelse
*
* Description: Inline utility function to free an fd in a filedesc
*
* the caller
*/
static void
-_fdrelse(struct proc * p, int fd)
+fdrelse(struct proc * p, int fd)
{
struct filedesc *fdp = p->p_fd;
int nfd = 0;
uio_t auio = NULL;
char uio_buf[UIO_SIZEOF(1)];
struct vfs_context context = *(vfs_context_current());
- bool wrote_some = false;
p = current_proc();
return error;
}
- if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_PIPE && fp->f_type != DTYPE_SOCKET) {
+ switch (FILEGLOB_DTYPE(fp->fp_glob)) {
+ case DTYPE_VNODE:
+ case DTYPE_PIPE:
+ case DTYPE_SOCKET:
+ break;
+ default:
error = EINVAL;
goto out;
}
goto out;
}
- context.vc_ucred = fp->f_fglob->fg_cred;
+ context.vc_ucred = fp->fp_glob->fg_cred;
if (UIO_SEG_IS_USER_SPACE(segflg)) {
spacetype = proc_is64bit(p) ? UIO_USERSPACE64 : UIO_USERSPACE32;
auio = uio_createwithbuffer(1, offset, spacetype, rw, &uio_buf[0], sizeof(uio_buf));
- uio_addiov(auio, base, len);
+ uio_addiov(auio, (user_addr_t)base, (user_size_t)len);
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;
+ if (uio_resid(auio) < orig_resid) {
+ os_atomic_or(&fp->fp_glob->fg_flag, FWASWRITTEN, relaxed);
+ }
} else {
error = fo_read(fp, auio, flags, &context);
}
if (aresid) {
*aresid = uio_resid(auio);
- } else {
- if (uio_resid(auio) && error == 0) {
- error = EIO;
- }
+ } 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);
- }
-
+ fp_drop(p, fd, fp, 0);
return error;
}
/*
- * dup
+ * sys_dup
*
* Description: Duplicate a file descriptor.
*
* *retval (modified) The new descriptor
*/
int
-dup(proc_t p, struct dup_args *uap, int32_t *retval)
+sys_dup(proc_t p, struct dup_args *uap, int32_t *retval)
{
struct filedesc *fdp = p->p_fd;
int old = uap->fd;
fp_drop(p, old, fp, 1);
proc_fdunlock(p);
- if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET) {
+ if (ENTR_SHOULDTRACE && FILEGLOB_DTYPE(fp->fp_glob) == DTYPE_SOCKET) {
KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_START,
new, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
}
}
/*
- * dup2
+ * sys_dup2
*
* Description: Duplicate a file descriptor to a particular value.
*
* *retval (modified) The new descriptor
*/
int
-dup2(proc_t p, struct dup2_args *uap, int32_t *retval)
+sys_dup2(proc_t p, struct dup2_args *uap, int32_t *retval)
+{
+ return dup2(p, uap->from, uap->to, retval);
+}
+
+int
+dup2(proc_t p, int old, int new, int *retval)
{
struct filedesc *fdp = p->p_fd;
- int old = uap->from, new = uap->to;
- int i, error;
struct fileproc *fp, *nfp;
+ int i, error;
+ rlim_t nofile = proc_limitgetcur(p, RLIMIT_NOFILE, TRUE);
proc_fdlock(p);
return error;
}
if (new < 0 ||
- (rlim_t)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
- new >= maxfiles) {
+ (rlim_t)new >= nofile ||
+ new >= maxfilesperproc) {
fp_drop(p, old, fp, 1);
proc_fdunlock(p);
return EBADF;
}
} else {
closeit:
- while ((fdp->fd_ofileflags[new] & UF_RESERVED) == UF_RESERVED) {
+ if ((fdp->fd_ofileflags[new] & UF_RESERVED) == UF_RESERVED) {
fp_drop(p, old, fp, 1);
procfdtbl_waitfd(p, new);
#if DIAGNOSTIC
goto startover;
}
- if ((fdp->fd_ofiles[new] != NULL) &&
- ((error = fp_lookup(p, new, &nfp, 1)) == 0)) {
- fp_drop(p, old, fp, 1);
+ if ((nfp = fdp->fd_ofiles[new]) != NULL) {
if (FP_ISGUARDED(nfp, GUARD_CLOSE)) {
+ fp_drop(p, old, fp, 1);
error = fp_guard_exception(p,
new, nfp, kGUARD_EXC_CLOSE);
- (void) fp_drop(p, new, nfp, 1);
proc_fdunlock(p);
return error;
}
- (void)close_internal_locked(p, new, nfp, FD_DUP2RESV);
-#if DIAGNOSTIC
- proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
-#endif
- procfdtbl_clearfd(p, new);
- goto startover;
+ (void)fp_close_and_unlock(p, new, nfp, FD_DUP2RESV);
+ proc_fdlock(p);
+ assert(fdp->fd_ofileflags[new] & UF_RESERVED);
} else {
#if DIAGNOSTIC
if (fdp->fd_ofiles[new] != NULL) {
#endif
procfdtbl_reservefd(p, new);
}
-
-#if DIAGNOSTIC
- proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
-#endif
}
#if DIAGNOSTIC
if (fdp->fd_ofiles[new] != 0) {
* blocking operation.
*/
int
-fcntl(proc_t p, struct fcntl_args *uap, int32_t *retval)
+sys_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 sys_fcntl_nocancel(p, (struct fcntl_nocancel_args *)uap, retval);
}
+#define ACCOUNT_OPENFROM_ENTITLEMENT \
+ "com.apple.private.vfs.role-account-openfrom"
/*
- * fcntl_nocancel
+ * sys_fcntl_nocancel
*
* Description: A non-cancel-testing file control system call.
*
* VNOP_ALLOCATE:???
* [F_SETSIZE,F_RDADVISE]
* EBADF
+ * EINVAL
* copyin:EFAULT
* vnode_getwithref:???
* [F_RDAHEAD,F_NOCACHE]
* *retval (modified) fcntl return value (if any)
*/
int
-fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval)
+sys_fcntl_nocancel(proc_t p, struct fcntl_nocancel_args *uap, int32_t *retval)
{
int fd = uap->fd;
struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
char *pop;
struct vnode *vp = NULLVP; /* for AUDIT_ARG() at end */
+ unsigned int oflags, nflags;
int i, tmp, error, error2, flg = 0;
struct flock fl = {};
struct flocktimeout fltimeout;
unsigned int fflag;
user_addr_t argp;
boolean_t is64bit;
+ rlim_t nofile;
+ int has_entitlement = 0;
AUDIT_ARG(fd, uap->fd);
AUDIT_ARG(cmd, uap->cmd);
+ nofile = proc_limitgetcur(p, RLIMIT_NOFILE, TRUE);
+
proc_fdlock(p);
if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
argp = CAST_USER_ADDR_T((uint32_t)uap->arg);
}
- pop = &fdp->fd_ofileflags[fd];
-
#if CONFIG_MACF
- error = mac_file_check_fcntl(proc_ucred(p), fp->f_fglob, uap->cmd,
+ error = mac_file_check_fcntl(proc_ucred(p), fp->fp_glob, uap->cmd,
uap->arg);
if (error) {
goto out;
}
#endif
+ pop = &fdp->fd_ofileflags[fd];
+
switch (uap->cmd) {
case F_DUPFD:
case F_DUPFD_CLOEXEC:
}
newmin = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */
AUDIT_ARG(value32, newmin);
- if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
- newmin >= maxfiles) {
+ if ((rlim_t)newmin >= nofile ||
+ newmin >= maxfilesperproc) {
error = EINVAL;
goto out;
}
goto out;
case F_SETFD:
- AUDIT_ARG(value32, uap->arg);
+ AUDIT_ARG(value32, (uint32_t)uap->arg);
if (uap->arg & FD_CLOEXEC) {
*pop |= UF_EXCLOSE;
} else {
goto out;
case F_SETFL:
- fp->f_flag &= ~FCNTLFLAGS;
+ // FIXME (rdar://54898652)
+ //
+ // this code is broken if fnctl(F_SETFL), ioctl() are
+ // called concurrently for the same fileglob.
+
tmp = CAST_DOWN_EXPLICIT(int, uap->arg); /* arg is an int, so we won't lose bits */
AUDIT_ARG(value32, tmp);
- fp->f_flag |= FFLAGS(tmp) & FCNTLFLAGS;
- tmp = fp->f_flag & FNONBLOCK;
+
+ os_atomic_rmw_loop(&fp->f_flag, oflags, nflags, relaxed, {
+ nflags = oflags & ~FCNTLFLAGS;
+ nflags |= FFLAGS(tmp) & FCNTLFLAGS;
+ });
+ tmp = nflags & FNONBLOCK;
error = fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
if (error) {
goto out;
}
- tmp = fp->f_flag & FASYNC;
+ tmp = nflags & FASYNC;
error = fo_ioctl(fp, FIOASYNC, (caddr_t)&tmp, &context);
if (!error) {
goto out;
}
- fp->f_flag &= ~FNONBLOCK;
+ os_atomic_andnot(&fp->f_flag, FNONBLOCK, relaxed);
tmp = 0;
(void)fo_ioctl(fp, FIONBIO, (caddr_t)&tmp, &context);
goto out;
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;
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:
error = EINVAL;
#endif
} else {
- struct fileglob *fg = fp->f_fglob;
+ struct fileglob *fg = fp->fp_glob;
lck_mtx_lock_spin(&fg->fg_lock);
if (tmp) {
error = EINVAL;
#endif
} else {
- *retval = (fp->f_fglob->fg_lflags & FG_NOSIGPIPE) ?
+ *retval = (fp->fp_glob->fg_lflags & FG_NOSIGPIPE) ?
1 : 0;
error = 0;
}
* mechanism to move the descriptor elsewhere will fail.
*/
if (CAST_DOWN_EXPLICIT(int, uap->arg)) {
- struct fileglob *fg = fp->f_fglob;
+ struct fileglob *fg = fp->fp_glob;
lck_mtx_lock_spin(&fg->fg_lock);
if (fg->fg_lflags & FG_CONFINED) {
error = 0;
- } else if (1 != fg->fg_count) {
+ } else if (1 != os_ref_get_count_raw(&fg->fg_count)) {
error = EAGAIN; /* go close the dup .. */
} else if (UF_FORKCLOSE == (*pop & UF_FORKCLOSE)) {
fg->fg_lflags |= FG_CONFINED;
goto out;
case F_GETCONFINED:
- *retval = (fp->f_fglob->fg_lflags & FG_CONFINED) ? 1 : 0;
+ *retval = (fp->fp_glob->fg_lflags & FG_CONFINED) ? 1 : 0;
error = 0;
goto out;
case F_OFD_SETLKWTIMEOUT:
case F_OFD_SETLKW:
flg |= F_WAIT;
- /* Fall into F_SETLK */
+ OS_FALLTHROUGH;
case F_SETLK:
case F_OFD_SETLK:
}
#if CONFIG_MACF
- error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
+ error = mac_file_check_lock(proc_ucred(p), fp->fp_glob,
F_SETLK, &fl);
if (error) {
(void)vnode_put(vp);
error = EBADF;
break;
}
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob,
F_SETLK, &fl, flg, &context, timeout);
break;
case F_WRLCK:
error = EBADF;
break;
}
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob,
F_SETLK, &fl, flg, &context, timeout);
break;
case F_UNLCK:
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob,
F_UNLCK, &fl, F_OFD_LOCK, &context,
timeout);
break;
}
if (0 == error &&
(F_RDLCK == fl.l_type || F_WRLCK == fl.l_type)) {
- struct fileglob *fg = fp->f_fglob;
+ struct fileglob *fg = fp->fp_glob;
/*
* arrange F_UNLCK on last close (once
break;
}
// XXX UInt32 unsafe for LP64 kernel
- OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag);
+ os_atomic_or(&p->p_ladvflag, P_LADVLOCK, relaxed);
error = VNOP_ADVLOCK(vp, (caddr_t)p,
F_SETLK, &fl, flg, &context, timeout);
break;
break;
}
// XXX UInt32 unsafe for LP64 kernel
- OSBitOrAtomic(P_LADVLOCK, &p->p_ladvflag);
+ os_atomic_or(&p->p_ladvflag, P_LADVLOCK, relaxed);
error = VNOP_ADVLOCK(vp, (caddr_t)p,
F_SETLK, &fl, flg, &context, timeout);
break;
}
#if CONFIG_MACF
- error = mac_file_check_lock(proc_ucred(p), fp->f_fglob,
+ error = mac_file_check_lock(proc_ucred(p), fp->fp_glob,
uap->cmd, &fl);
if (error == 0)
#endif
switch (uap->cmd) {
case F_OFD_GETLK:
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob,
F_GETLK, &fl, F_OFD_LOCK, &context, NULL);
break;
case F_OFD_GETLKPID:
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob,
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob,
F_GETLKPID, &fl, F_OFD_LOCK, &context, NULL);
break;
default:
}
#if CONFIG_MACF
- if ((error = mac_vnode_check_write(&context, fp->f_fglob->fg_cred, vp))) {
+ if ((error = mac_vnode_check_write(&context, fp->fp_glob->fg_cred, vp))) {
(void)vnode_put(vp);
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;
#if CONFIG_MACF
error = mac_vnode_check_truncate(&context,
- fp->f_fglob->fg_cred, vp);
+ fp->fp_glob->fg_cred, vp);
if (error) {
(void)vnode_put(vp);
goto outdrop;
#endif
/*
* Make sure that we are root. Growing a file
- * without zero filling the data is a security hole
- * root would have access anyway so we'll allow it
+ * without zero filling the data is a security hole.
*/
if (!kauth_cred_issuser(kauth_cred_get())) {
error = EACCES;
} else {
/*
- * set the file size
+ * Require privilege to change file size without zerofill,
+ * else will change the file size and zerofill it.
*/
- error = vnode_setsize(vp, offset, IO_NOZEROFILL,
- &context);
+ error = priv_check_cred(kauth_cred_get(), PRIV_VFS_SETSIZE, 0);
+ if (error == 0) {
+ error = vnode_setsize(vp, offset, IO_NOZEROFILL, &context);
+ } else {
+ error = vnode_setsize(vp, offset, 0, &context);
+ }
#if CONFIG_MACF
if (error == 0) {
- mac_vnode_notify_truncate(&context, fp->f_fglob->fg_cred, vp);
+ mac_vnode_notify_truncate(&context, fp->fp_glob->fg_cred, vp);
}
#endif
}
goto out;
}
if (uap->arg) {
- fp->f_fglob->fg_flag &= ~FNORDAHEAD;
+ os_atomic_andnot(&fp->fp_glob->fg_flag, FNORDAHEAD, relaxed);
} else {
- fp->f_fglob->fg_flag |= FNORDAHEAD;
+ os_atomic_or(&fp->fp_glob->fg_flag, FNORDAHEAD, relaxed);
}
-
goto out;
case F_NOCACHE:
goto out;
}
if (uap->arg) {
- fp->f_fglob->fg_flag |= FNOCACHE;
+ os_atomic_or(&fp->fp_glob->fg_flag, FNOCACHE, relaxed);
} else {
- fp->f_fglob->fg_flag &= ~FNOCACHE;
+ os_atomic_andnot(&fp->fp_glob->fg_flag, FNOCACHE, relaxed);
}
-
goto out;
case F_NODIRECT:
goto out;
}
if (uap->arg) {
- fp->f_fglob->fg_flag |= FNODIRECT;
+ os_atomic_or(&fp->fp_glob->fg_flag, FNODIRECT, relaxed);
} else {
- fp->f_fglob->fg_flag &= ~FNODIRECT;
+ os_atomic_andnot(&fp->fp_glob->fg_flag, FNODIRECT, relaxed);
}
-
goto out;
case F_SINGLE_WRITER:
goto out;
}
if (uap->arg) {
- fp->f_fglob->fg_flag |= FSINGLE_WRITER;
+ os_atomic_or(&fp->fp_glob->fg_flag, FSINGLE_WRITER, relaxed);
} else {
- fp->f_fglob->fg_flag &= ~FSINGLE_WRITER;
+ os_atomic_andnot(&fp->fp_glob->fg_flag, FSINGLE_WRITER, relaxed);
}
-
goto out;
case F_GLOBAL_NOCACHE:
if ((error = copyin(argp, (caddr_t)&ra_struct, sizeof(ra_struct)))) {
goto outdrop;
}
+ if (ra_struct.ra_offset < 0 || ra_struct.ra_count < 0) {
+ error = EINVAL;
+ goto outdrop;
+ }
if ((error = vnode_getwithref(vp)) == 0) {
error = VNOP_IOCTL(vp, F_RDADVISE, (caddr_t)&ra_struct, 0, &context);
goto outdrop;
}
- a_size = MIN((uint64_t)l2p_struct.l2p_contigbytes, SIZE_MAX);
+ a_size = (size_t)MIN((uint64_t)l2p_struct.l2p_contigbytes, SIZE_MAX);
} else {
a_size = devBlockSize;
}
}
goto outdrop;
}
- case F_GETPATH: {
+ case F_GETPATH:
+ case F_GETPATH_NOFIRMLINK: {
char *pathbufp;
int pathlen;
goto outdrop;
}
if ((error = vnode_getwithref(vp)) == 0) {
- error = vn_getpath(vp, pathbufp, &pathlen);
+ 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) {
proc_fdunlock(p);
pathlen = MAXPATHLEN;
- pathbufp = kalloc(MAXPATHLEN);
+ pathbufp = zalloc(ZV_NAMEI);
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, (int)pathlen, retval);
(void)vnode_put(vp);
}
}
- kfree(pathbufp, MAXPATHLEN);
+ zfree(ZV_NAMEI, pathbufp);
goto outdrop;
}
goto outdrop;
}
+ /*
+ * Only entitled apps may use the credentials of the thread
+ * that opened the file descriptor.
+ * Non-entitled threads will use their own context.
+ */
+ if (IOTaskHasEntitlement(current_task(), ACCOUNT_OPENFROM_ENTITLEMENT)) {
+ has_entitlement = 1;
+ }
+
/* Get flags, mode and pathname arguments. */
if (IS_64BIT_PROCESS(p)) {
error = copyin(argp, &fopen, sizeof(fopen));
/* 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, has_entitlement ? &context : vfs_context_current());
nd.ni_dvp = vp;
- error = open1(&context, &nd, fopen.o_flags, &va,
- fileproc_alloc_init, NULL, retval);
+ error = open1(has_entitlement ? &context : vfs_context_current(),
+ &nd, fopen.o_flags, &va, fileproc_alloc_init, NULL, retval);
vnode_put(vp);
break;
goto outdrop;
}
+ /*
+ * Only entitled apps may use the credentials of the thread
+ * that opened the file descriptor.
+ * Non-entitled threads will use their own context.
+ */
+ if (IOTaskHasEntitlement(current_task(), ACCOUNT_OPENFROM_ENTITLEMENT)) {
+ has_entitlement = 1;
+ }
+
/* Get flags, mode and pathname arguments. */
if (IS_64BIT_PROCESS(p)) {
pathname = (user_addr_t)argp;
}
/* Start the lookup relative to the file descriptor's vnode. */
- error = unlink1(&context, vp, pathname, UIO_USERSPACE, 0);
+ error = unlink1(has_entitlement ? &context : vfs_context_current(),
+ vp, pathname, UIO_USERSPACE, 0);
vnode_put(vp);
break;
case F_ADDFILESIGS:
case F_ADDFILESIGS_FOR_DYLD_SIM:
case F_ADDFILESIGS_RETURN:
+ case F_ADDFILESIGS_INFO:
{
struct cs_blob *blob = NULL;
struct user_fsignatures fs;
vm_offset_t kernel_blob_addr;
vm_size_t kernel_blob_size;
int blob_add_flags = 0;
+ const size_t sizeof_fs = (uap->cmd == F_ADDFILESIGS_INFO ?
+ offsetof(struct user_fsignatures, fs_cdhash /* first output element */) :
+ offsetof(struct user_fsignatures, fs_fsignatures_size /* compat */));
if (fp->f_type != DTYPE_VNODE) {
error = EBADF;
}
if (IS_64BIT_PROCESS(p)) {
- error = copyin(argp, &fs, sizeof(fs));
+ error = copyin(argp, &fs, sizeof_fs);
} else {
+ if (uap->cmd == F_ADDFILESIGS_INFO) {
+ error = EINVAL;
+ vnode_put(vp);
+ goto outdrop;
+ }
+
struct user32_fsignatures fs32;
error = copyin(argp, &fs32, sizeof(fs32));
/*
* First check if we have something loaded a this offset
*/
- blob = ubc_cs_blob_get(vp, CPU_TYPE_ANY, fs.fs_file_start);
+ blob = ubc_cs_blob_get(vp, CPU_TYPE_ANY, CPU_SUBTYPE_ANY, fs.fs_file_start);
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);
+ error = ubc_cs_blob_revalidate(vp, blob, NULL, blob_add_flags, proc_platform(p));
if (error) {
blob = NULL;
if (error != EAGAIN) {
kernel_blob_size = CAST_DOWN(vm_size_t, fs.fs_blob_size);
kr = ubc_cs_blob_allocate(&kernel_blob_addr, &kernel_blob_size);
- if (kr != KERN_SUCCESS) {
+ if (kr != KERN_SUCCESS || kernel_blob_size < fs.fs_blob_size) {
error = ENOMEM;
vnode_put(vp);
goto outdrop;
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 */
+ fs.fs_blob_size);
+ } else { /* F_ADDFILESIGS || F_ADDFILESIGS_RETURN || F_ADDFILESIGS_FOR_DYLD_SIM || F_ADDFILESIGS_INFO */
int resid;
error = vn_rdwr(UIO_READ,
vp,
(caddr_t) kernel_blob_addr,
- kernel_blob_size,
+ (int)kernel_blob_size,
fs.fs_file_start + fs.fs_blob_start,
UIO_SYSSPACE,
0,
blob = NULL;
error = ubc_cs_blob_add(vp,
+ proc_platform(p),
CPU_TYPE_ANY, /* not for a specific architecture */
+ CPU_SUBTYPE_ANY,
fs.fs_file_start,
&kernel_blob_addr,
kernel_blob_size,
}
}
- if (uap->cmd == F_ADDFILESIGS_RETURN || uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM) {
+ if (uap->cmd == F_ADDFILESIGS_RETURN || uap->cmd == F_ADDFILESIGS_FOR_DYLD_SIM ||
+ uap->cmd == F_ADDFILESIGS_INFO) {
/*
* The first element of the structure is a
* off_t that happen to have the same size for
end_offset = blob->csb_end_offset;
}
error = copyout(&end_offset, argp, sizeof(end_offset));
+
+ if (error) {
+ vnode_put(vp);
+ goto outdrop;
+ }
+ }
+
+ if (uap->cmd == F_ADDFILESIGS_INFO) {
+ /* Return information. What we copy out depends on the size of the
+ * passed in structure, to keep binary compatibility. */
+
+ if (fs.fs_fsignatures_size >= sizeof(struct user_fsignatures)) {
+ // enough room for fs_cdhash[20]+fs_hash_type
+
+ if (blob != NULL) {
+ error = copyout(blob->csb_cdhash,
+ (vm_address_t)argp + offsetof(struct user_fsignatures, fs_cdhash),
+ USER_FSIGNATURES_CDHASH_LEN);
+ if (error) {
+ vnode_put(vp);
+ goto outdrop;
+ }
+ int hashtype = cs_hash_type(blob->csb_hashtype);
+ error = copyout(&hashtype,
+ (vm_address_t)argp + offsetof(struct user_fsignatures, fs_hash_type),
+ sizeof(int));
+ if (error) {
+ vnode_put(vp);
+ goto outdrop;
+ }
+ }
+ }
}
(void) vnode_put(vp);
break;
}
+#if CONFIG_SUPPLEMENTAL_SIGNATURES
+ case F_ADDFILESUPPL:
+ {
+ struct vnode *ivp;
+ struct cs_blob *blob = NULL;
+ struct user_fsupplement fs;
+ int orig_fd;
+ struct fileproc* orig_fp = NULL;
+ kern_return_t kr;
+ vm_offset_t kernel_blob_addr;
+ vm_size_t kernel_blob_size;
+
+ if (!IS_64BIT_PROCESS(p)) {
+ error = EINVAL;
+ goto out; // drop fp and unlock fds
+ }
+
+ if (fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
+ goto out;
+ }
+
+ error = copyin(argp, &fs, sizeof(fs));
+ if (error) {
+ goto out;
+ }
+
+ orig_fd = fs.fs_orig_fd;
+ if ((error = fp_lookup(p, orig_fd, &orig_fp, 1))) {
+ printf("CODE SIGNING: Failed to find original file for supplemental signature attachment\n");
+ goto out;
+ }
+
+ if (orig_fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
+ fp_drop(p, orig_fd, orig_fp, 1);
+ goto out;
+ }
+
+ ivp = (struct vnode *)orig_fp->f_data;
+
+ vp = (struct vnode *)fp->f_data;
+
+ proc_fdunlock(p);
+
+ error = vnode_getwithref(ivp);
+ if (error) {
+ fp_drop(p, orig_fd, orig_fp, 0);
+ goto outdrop; //drop fp
+ }
+
+ error = vnode_getwithref(vp);
+ if (error) {
+ vnode_put(ivp);
+ fp_drop(p, orig_fd, orig_fp, 0);
+ goto outdrop;
+ }
+
+ if (fs.fs_blob_size > CS_MAX_BLOB_SIZE) {
+ error = E2BIG;
+ goto dropboth; // drop iocounts on vp and ivp, drop orig_fp then drop fp via outdrop
+ }
+
+ kernel_blob_size = CAST_DOWN(vm_size_t, fs.fs_blob_size);
+ kr = ubc_cs_blob_allocate(&kernel_blob_addr, &kernel_blob_size);
+ if (kr != KERN_SUCCESS) {
+ error = ENOMEM;
+ goto dropboth;
+ }
+
+ int resid;
+ error = vn_rdwr(UIO_READ, vp,
+ (caddr_t)kernel_blob_addr, (int)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);
+ goto dropboth;
+ }
+
+ error = ubc_cs_blob_add_supplement(vp, ivp, fs.fs_file_start,
+ &kernel_blob_addr, kernel_blob_size, &blob);
+
+ /* ubc_blob_add_supplement() 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);
+ }
+ goto dropboth;
+ }
+ vnode_put(ivp);
+ vnode_put(vp);
+ fp_drop(p, orig_fd, orig_fp, 0);
+ break;
+
+dropboth:
+ vnode_put(ivp);
+ vnode_put(vp);
+ fp_drop(p, orig_fd, orig_fp, 0);
+ goto outdrop;
+ }
+#endif
case F_GETCODEDIR:
case F_FINDSIGS: {
error = ENOTSUP;
error = EBADF;
goto out;
}
- fg = fp->f_fglob;
+ fg = fp->fp_glob;
proc_fdunlock(p);
if (IS_64BIT_PROCESS(p)) {
break;
}
+ case F_GETSIGSINFO: {
+ struct cs_blob *blob = NULL;
+ fgetsigsinfo_t sigsinfo = {};
+
+ if (fp->f_type != DTYPE_VNODE) {
+ error = EBADF;
+ goto out;
+ }
+ vp = (struct vnode *)fp->f_data;
+ proc_fdunlock(p);
+
+ error = vnode_getwithref(vp);
+ if (error) {
+ goto outdrop;
+ }
+
+ error = copyin(argp, &sigsinfo, sizeof(sigsinfo));
+ if (error) {
+ vnode_put(vp);
+ goto outdrop;
+ }
+
+ blob = ubc_cs_blob_get(vp, CPU_TYPE_ANY, CPU_SUBTYPE_ANY, sigsinfo.fg_file_start);
+ if (blob == NULL) {
+ error = ENOENT;
+ vnode_put(vp);
+ goto outdrop;
+ }
+ switch (sigsinfo.fg_info_request) {
+ case GETSIGSINFO_PLATFORM_BINARY:
+ sigsinfo.fg_sig_is_platform = blob->csb_platform_binary;
+ error = copyout(&sigsinfo.fg_sig_is_platform,
+ (vm_address_t)argp + offsetof(struct fgetsigsinfo, fg_sig_is_platform),
+ sizeof(sigsinfo.fg_sig_is_platform));
+ if (error) {
+ vnode_put(vp);
+ goto outdrop;
+ }
+ break;
+ default:
+ error = EINVAL;
+ vnode_put(vp);
+ goto outdrop;
+ }
+ vnode_put(vp);
+ break;
+ }
#if CONFIG_PROTECT
case F_GETPROTECTIONCLASS: {
if (fp->f_type != DTYPE_VNODE) {
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) {
+ if (src_vp->v_tag != VT_HFS && src_vp->v_tag != VT_APFS) {
error = ENOTSUP;
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 = ENOTSUP;
goto out;
#if CONFIG_MACF
/* Re-do MAC checks against the new FD, pass in a fake argument */
- error = mac_file_check_fcntl(proc_ucred(p), fp2->f_fglob, uap->cmd, 0);
+ error = mac_file_check_fcntl(proc_ucred(p), fp2->fp_glob, uap->cmd, 0);
if (error) {
fp_drop(p, fd2, fp2, 1);
goto out;
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:
+ case TIOCREVOKECLEAR:
error = EINVAL;
goto out;
default:
memp = NULL;
if (size > sizeof(stkbuf)) {
- if ((memp = (caddr_t)kalloc(size)) == 0) {
+ memp = (caddr_t)kheap_alloc(KHEAP_TEMP, size, Z_WAITOK);
+ if (memp == 0) {
(void)vnode_put(vp);
error = ENOMEM;
goto outdrop;
if (error) {
(void)vnode_put(vp);
if (memp) {
- kfree(memp, size);
+ kheap_free(KHEAP_TEMP, memp, size);
}
goto outdrop;
}
error = copyout(data, argp, size);
}
if (memp) {
- kfree(memp, size);
+ kheap_free(KHEAP_TEMP, memp, size);
}
}
break;
fdrelse(p, new);
return EBADF;
}
- fg_ref(ofp);
#if CONFIG_MACF
- error = mac_file_check_dup(proc_ucred(p), ofp->f_fglob, new);
+ error = mac_file_check_dup(proc_ucred(p), ofp->fp_glob, new);
if (error) {
- fg_drop(ofp);
fdrelse(p, new);
return error;
}
proc_fdlock(p);
if (nfp == NULL) {
- fg_drop(ofp);
fdrelse(p, new);
return ENOMEM;
}
- nfp->f_fglob = ofp->f_fglob;
+ fg_ref(p, ofp->fp_glob);
+ nfp->fp_glob = ofp->fp_glob;
#if DIAGNOSTIC
if (fdp->fd_ofiles[new] != 0) {
/*
- * close
+ * sys_close
*
* Description: The implementation of the close(2) system call
*
* close function
*/
int
-close(proc_t p, struct close_args *uap, int32_t *retval)
+sys_close(proc_t p, struct close_args *uap, __unused int32_t *retval)
{
__pthread_testcancel(1);
- return close_nocancel(p, (struct close_nocancel_args *)uap, retval);
+ return close_nocancel(p, uap->fd);
}
+int
+sys_close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retval)
+{
+ return close_nocancel(p, uap->fd);
+}
int
-close_nocancel(proc_t p, struct close_nocancel_args *uap, __unused int32_t *retval)
+close_nocancel(proc_t p, int fd)
{
struct fileproc *fp;
- int fd = uap->fd;
- int error;
AUDIT_SYSCLOSE(p, fd);
proc_fdlock(p);
-
- if ((error = fp_lookup(p, fd, &fp, 1))) {
+ if ((fp = fp_get_noref_locked(p, fd)) == NULL) {
proc_fdunlock(p);
- return error;
+ return EBADF;
}
if (FP_ISGUARDED(fp, GUARD_CLOSE)) {
- error = fp_guard_exception(p, fd, fp, kGUARD_EXC_CLOSE);
- (void) fp_drop(p, fd, fp, 1);
+ int error = fp_guard_exception(p, fd, fp, kGUARD_EXC_CLOSE);
proc_fdunlock(p);
return error;
}
- error = close_internal_locked(p, fd, fp, 0);
-
- proc_fdunlock(p);
-
- return error;
+ return fp_close_and_unlock(p, fd, fp, 0);
}
-/*
- * close_internal_locked
- *
- * Close a file descriptor.
- *
- * Parameters: p Process in whose per process file table
- * the close is to occur
- * fd fd to be closed
- * fp fileproc associated with the fd
- *
- * Returns: 0 Success
- * EBADF fd already in close wait state
- * closef_locked:??? Anything returnable by a per-fileops
- * close function
- *
- * Locks: Assumes proc_fdlock for process is held by the caller and returns
- * with lock held
- *
- * Notes: This function may drop and reacquire this lock; it is unsafe
- * for a caller to assume that other state protected by the lock
- * has not been subsequently changed out from under it.
- */
int
-close_internal_locked(proc_t p, int fd, struct fileproc *fp, int flags)
+fp_close_and_unlock(proc_t p, int fd, struct fileproc *fp, int flags)
{
struct filedesc *fdp = p->p_fd;
- int error = 0;
- int resvfd = flags & FD_DUP2RESV;
-
+ struct fileglob *fg = fp->fp_glob;
#if DIAGNOSTIC
proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
#endif
- /* Keep people from using the filedesc while we are closing it */
- procfdtbl_markclosefd(p, fd);
-
-
- if ((fp->f_flags & FP_CLOSING) == FP_CLOSING) {
- panic("close_internal_locked: being called on already closing fd");
- }
-
-
-#if DIAGNOSTIC
- if ((fdp->fd_ofileflags[fd] & UF_RESERVED) == 0) {
- panic("close_internal: unreserved fileflags with fd %d", fd);
+ /*
+ * Keep most people from finding the filedesc while we are closing it.
+ *
+ * Callers are:
+ *
+ * - dup2() which always waits for UF_RESERVED to clear
+ *
+ * - close/guarded_close/... who will fail the fileproc lookup if
+ * UF_RESERVED is set,
+ *
+ * - fdexec()/fdfree() who only run once all threads in the proc
+ * are properly canceled, hence no fileproc in this proc should
+ * be in flux.
+ *
+ * Which means that neither UF_RESERVED nor UF_CLOSING should be set.
+ *
+ * Callers of fp_get_noref_locked_with_iocount() can still find
+ * this entry so that they can drop their I/O reference despite
+ * not having remembered the fileproc pointer (namely select() and
+ * file_drop()).
+ */
+ if (p->p_fd->fd_ofileflags[fd] & (UF_RESERVED | UF_CLOSING)) {
+ panic("%s: called with fileproc in flux (%d/:%p)",
+ __func__, fd, fp);
}
-#endif
-
- fp->f_flags |= FP_CLOSING;
+ p->p_fd->fd_ofileflags[fd] |= (UF_RESERVED | UF_CLOSING);
- if ((fp->f_flags & FP_AIOISSUED) || kauth_authorize_fileop_has_listeners()) {
+ if ((fp->fp_flags & FP_AIOISSUED) || kauth_authorize_fileop_has_listeners()) {
proc_fdunlock(p);
- if ((fp->f_type == DTYPE_VNODE) && kauth_authorize_fileop_has_listeners()) {
+ if ((FILEGLOB_DTYPE(fg) == 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) {
+ if (vnode_getwithref((vnode_t)fg->fg_data) == 0) {
u_int fileop_flags = 0;
- if ((fp->f_flags & FP_WRITTEN) != 0) {
+ if (fg->fg_flag & FWASWRITTEN) {
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);
+ kauth_authorize_fileop(fg->fg_cred, KAUTH_FILEOP_CLOSE,
+ (uintptr_t)fg->fg_data, (uintptr_t)fileop_flags);
+#if CONFIG_MACF
+ mac_file_notify_close(proc_ucred(p), fp->fp_glob);
+#endif
+ vnode_put((vnode_t)fg->fg_data);
}
}
- if (fp->f_flags & FP_AIOISSUED) {
+ if (fp->fp_flags & FP_AIOISSUED) {
/*
* cancel all async IO requests that can be cancelled.
*/
fileproc_drain(p, fp);
- if (fp->f_flags & FP_WAITEVENT) {
- (void)waitevent_close(p, fp);
- }
-
- if (resvfd == 0) {
- _fdrelse(p, fd);
+ if (flags & FD_DUP2RESV) {
+ fdp->fd_ofiles[fd] = NULL;
+ fdp->fd_ofileflags[fd] &= ~(UF_CLOSING | UF_EXCLOSE | UF_FORKCLOSE);
} else {
- procfdtbl_reservefd(p, fd);
+ fdrelse(p, fd);
}
- if (ENTR_SHOULDTRACE && fp->f_type == DTYPE_SOCKET) {
- KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_END,
- fd, 0, (int64_t)VM_KERNEL_ADDRPERM(fp->f_data));
- }
+ proc_fdunlock(p);
- error = closef_locked(fp, fp->f_fglob, p);
- if ((fp->f_flags & FP_WAITCLOSE) == FP_WAITCLOSE) {
- wakeup(&fp->f_flags);
+ if (ENTR_SHOULDTRACE && FILEGLOB_DTYPE(fg) == DTYPE_SOCKET) {
+ KERNEL_ENERGYTRACE(kEnTrActKernSocket, DBG_FUNC_END,
+ fd, 0, (int64_t)VM_KERNEL_ADDRPERM(fg->fg_data));
}
- fp->f_flags &= ~(FP_WAITCLOSE | FP_CLOSING);
-
- proc_fdunlock(p);
fileproc_free(fp);
- proc_fdlock(p);
-
-#if DIAGNOSTIC
- if (resvfd != 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 fg_drop(p, fg);
}
/*
- * fstat1
+ * fstat
*
* Description: Return status information about a file descriptor.
*
* XXX fileops instead.
*/
static int
-fstat1(proc_t p, int fd, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
+fstat(proc_t p, int fd, user_addr_t ub, user_addr_t xsecurity, user_addr_t xsecurity_size, int isstat64)
{
struct fileproc *fp;
union {
* 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->fp_glob->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);
/*
- * fstat_extended
+ * sys_fstat_extended
*
* Description: Extended version of fstat supporting returning extended
* security information
* uap->xsecurity_size The size of xsecurity, or 0
*
* Returns: 0 Success
- * !0 Errno (see fstat1)
+ * !0 Errno (see fstat)
*/
int
-fstat_extended(proc_t p, struct fstat_extended_args *uap, __unused int32_t *retval)
+sys_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 fstat(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 0);
}
/*
- * fstat
+ * sys_fstat
*
* Description: Get file status for the file associated with fd
*
* uap->ub The user stat buffer
*
* Returns: 0 Success
- * !0 Errno (see fstat1)
+ * !0 Errno (see fstat)
*/
int
-fstat(proc_t p, struct fstat_args *uap, __unused int32_t *retval)
+sys_fstat(proc_t p, struct fstat_args *uap, __unused int32_t *retval)
{
- return fstat1(p, uap->fd, uap->ub, 0, 0, 0);
+ return fstat(p, uap->fd, uap->ub, 0, 0, 0);
}
/*
- * fstat64_extended
+ * sys_fstat64_extended
*
* Description: Extended version of fstat64 supporting returning extended
* security information
* uap->xsecurity_size The size of xsecurity, or 0
*
* Returns: 0 Success
- * !0 Errno (see fstat1)
+ * !0 Errno (see fstat)
*/
int
-fstat64_extended(proc_t p, struct fstat64_extended_args *uap, __unused int32_t *retval)
+sys_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 fstat(p, uap->fd, uap->ub, uap->xsecurity, uap->xsecurity_size, 1);
}
/*
- * fstat64
+ * sys_fstat64
*
* Description: Get 64 bit version of the file status for the file associated
* with fd
* uap->ub The user stat buffer
*
* Returns: 0 Success
- * !0 Errno (see fstat1)
+ * !0 Errno (see fstat)
*/
int
-fstat64(proc_t p, struct fstat64_args *uap, __unused int32_t *retval)
+sys_fstat64(proc_t p, struct fstat64_args *uap, __unused int32_t *retval)
{
- return fstat1(p, uap->fd, uap->ub, 0, 0, 1);
+ return fstat(p, uap->fd, uap->ub, 0, 0, 1);
}
/*
- * fpathconf
+ * sys_fpathconf
*
* Description: Return pathconf information about a file descriptor.
*
* *retval (modified) Returned information (numeric)
*/
int
-fpathconf(proc_t p, struct fpathconf_args *uap, int32_t *retval)
+sys_fpathconf(proc_t p, struct fpathconf_args *uap, int32_t *retval)
{
int fd = uap->fd;
struct fileproc *fp;
{
struct filedesc *fdp = p->p_fd;
int i;
- int lim, last, numfiles, oldnfiles;
+ int last, numfiles, oldnfiles;
struct fileproc **newofiles, **ofiles;
char *newofileflags;
+ rlim_t lim;
+ rlim_t nofile = proc_limitgetcur(p, RLIMIT_NOFILE, TRUE);
+
+ nofile = MIN(nofile, INT_MAX);
/*
* Search for a free descriptor starting at the higher
proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
#endif
- lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
+ lim = MIN(nofile, maxfilesperproc);
for (;;) {
- last = min(fdp->fd_nfiles, lim);
+ last = (int)MIN((unsigned int)fdp->fd_nfiles, (unsigned int)lim);
if ((i = want) < fdp->fd_freefile) {
i = fdp->fd_freefile;
}
/*
* No space in current array. Expand?
*/
- if (fdp->fd_nfiles >= lim) {
+ if ((rlim_t)fdp->fd_nfiles >= lim) {
return EMFILE;
}
if (fdp->fd_nfiles < NDEXTENT) {
numfiles = 2 * fdp->fd_nfiles;
}
/* Enforce lim */
- if (numfiles > lim) {
- numfiles = lim;
+ if ((rlim_t)numfiles > lim) {
+ numfiles = (int)lim;
}
proc_fdunlock(p);
- MALLOC_ZONE(newofiles, struct fileproc **,
+ MALLOC(newofiles, struct fileproc **,
numfiles * OFILESIZE, M_OFILETABL, M_WAITOK);
proc_fdlock(p);
if (newofiles == NULL) {
return ENOMEM;
}
if (fdp->fd_nfiles >= numfiles) {
- FREE_ZONE(newofiles, numfiles * OFILESIZE, M_OFILETABL);
+ FREE(newofiles, M_OFILETABL);
continue;
}
newofileflags = (char *) &newofiles[numfiles];
fdp->fd_ofiles = newofiles;
fdp->fd_ofileflags = newofileflags;
fdp->fd_nfiles = numfiles;
- FREE_ZONE(ofiles, oldnfiles * OFILESIZE, M_OFILETABL);
+ FREE(ofiles, M_OFILETABL);
fdexpand++;
}
}
struct filedesc *fdp = p->p_fd;
struct fileproc **fpp;
char *flags;
- int i, lim;
+ int i;
+ int lim;
+ rlim_t nofile = proc_limitgetcur(p, RLIMIT_NOFILE, TRUE);
- lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
+ lim = (int)MIN(nofile, maxfilesperproc);
if ((i = lim - fdp->fd_nfiles) > 0 && (n -= i) <= 0) {
return 1;
}
}
-/*
- * fdrelse
- *
- * Description: Legacy KPI wrapper function for _fdrelse
- *
- * Parameters: p Process in which fd lives
- * fd fd to free
- *
- * Returns: void
- *
- * Locks: Assumes proc_fdlock for process is held by the caller
- */
-void
-fdrelse(proc_t p, int fd)
+struct fileproc *
+fp_get_noref_locked(proc_t p, int fd)
{
- _fdrelse(p, fd);
+ struct filedesc *fdp = p->p_fd;
+ struct fileproc *fp;
+
+ if (fd < 0 || fd >= fdp->fd_nfiles ||
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
+ return NULL;
+ }
+ return fp;
}
+struct fileproc *
+fp_get_noref_locked_with_iocount(proc_t p, int fd)
+{
+ struct filedesc *fdp = p->p_fd;
+ struct fileproc *fp = NULL;
+
+ if (fd < 0 || fd >= fdp->fd_nfiles ||
+ (fp = fdp->fd_ofiles[fd]) == NULL ||
+ os_ref_get_count(&fp->fp_iocount) <= 1 ||
+ ((fdp->fd_ofileflags[fd] & UF_RESERVED) &&
+ !(fdp->fd_ofileflags[fd] & UF_CLOSING))) {
+ panic("%s: caller without an ioccount on fileproc (%d/:%p)",
+ __func__, fd, fp);
+ }
+
+ return fp;
+}
-/*
- * fdgetf_noref
- *
- * Description: Get the fileproc pointer for the given fd from the per process
- * open file table without taking an explicit reference on it.
- *
- * Parameters: p Process containing fd
- * fd fd to obtain fileproc for
- * resultfp Pointer to pointer return area
- *
- * Returns: 0 Success
- * EBADF
- *
- * Implicit returns:
- * *resultfp (modified) Pointer to fileproc pointer
- *
- * Locks: Assumes proc_fdlock for process is held by the caller
- *
- * Notes: Because there is no reference explicitly taken, the returned
- * fileproc pointer is only valid so long as the proc_fdlock
- * remains held by the caller.
- */
int
-fdgetf_noref(proc_t p, int fd, struct fileproc **resultfp)
+fp_get_ftype(proc_t p, int fd, file_type_t ftype, int err, struct fileproc **fpp)
{
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)) {
+ proc_fdunlock(p);
return EBADF;
}
- if (resultfp) {
- *resultfp = fp;
+
+ if (fp->f_type != ftype) {
+ proc_fdunlock(p);
+ return err;
}
+
+ os_ref_retain_locked(&fp->fp_iocount);
+ proc_fdunlock(p);
+
+ *fpp = fp;
return 0;
}
*
* Description: Get fileproc and vnode pointer for a given fd from the per
* process open file table of the specified process, and if
- * successful, increment the f_iocount
+ * successful, increment the fp_iocount
*
* Parameters: p Process in which fd lives
* fd fd to get information for
int
fp_getfvp(proc_t p, int fd, struct fileproc **resultfp, struct vnode **resultvp)
{
- struct filedesc *fdp = p->p_fd;
struct fileproc *fp;
+ int error;
- proc_fdlock_spin(p);
- if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_VNODE) {
- proc_fdunlock(p);
- return ENOTSUP;
- }
- fp->f_iocount++;
-
- if (resultfp) {
- *resultfp = fp;
- }
- if (resultvp) {
- *resultvp = (struct vnode *)fp->f_data;
+ error = fp_get_ftype(p, fd, DTYPE_VNODE, ENOTSUP, &fp);
+ if (error == 0) {
+ if (resultfp) {
+ *resultfp = fp;
+ }
+ if (resultvp) {
+ *resultvp = (struct vnode *)fp->f_data;
+ }
}
- proc_fdunlock(p);
- return 0;
+ return error;
}
/*
- * fp_getfvpandvid
+ * fp_get_pipe_id
*
- * Description: Get fileproc, vnode pointer, and vid for a given fd from the
- * per process open file table of the specified process, and if
- * successful, increment the f_iocount
+ * Description: Get pipe id for a given fd from the per process open file table
+ * of the specified process.
*
* Parameters: p Process in which fd lives
* fd fd to get information for
- * resultfp Pointer to result fileproc
- * pointer area, or 0 if none
- * resultvp Pointer to result vnode pointer
- * area, or 0 if none
- * vidp Pointer to resuld vid area
+ * result_pipe_id Pointer to result pipe id
*
* Returns: 0 Success
- * EBADF Bad file descriptor
- * ENOTSUP fd does not refer to a vnode
+ * EIVAL NULL pointer arguments passed
+ * fp_lookup:EBADF Bad file descriptor
+ * ENOTSUP fd does not refer to a pipe
*
* Implicit returns:
- * *resultfp (modified) Fileproc pointer
- * *resultvp (modified) vnode pointer
- * *vidp vid value
- *
- * Notes: The resultfp and resultvp fields are optional, and may be
- * independently specified as NULL to skip returning information
+ * *result_pipe_id (modified) pipe id
*
* Locks: Internally takes and releases proc_fdlock
*/
int
-fp_getfvpandvid(proc_t p, int fd, struct fileproc **resultfp,
- struct vnode **resultvp, uint32_t *vidp)
+fp_get_pipe_id(proc_t p, int fd, uint64_t *result_pipe_id)
{
- 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)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_VNODE) {
- proc_fdunlock(p);
- return ENOTSUP;
- }
- fp->f_iocount++;
-
- if (resultfp) {
- *resultfp = fp;
- }
- if (resultvp) {
- *resultvp = (struct vnode *)fp->f_data;
- }
- if (vidp) {
- *vidp = (uint32_t)vnode_vid((struct vnode *)fp->f_data);
- }
- proc_fdunlock(p);
-
- return 0;
-}
-
-
-/*
- * fp_getfsock
- *
- * Description: Get fileproc and socket pointer for a given fd from the
- * per process open file table of the specified process, and if
- * successful, increment the f_iocount
- *
- * Parameters: p Process in which fd lives
- * fd fd to get information for
- * resultfp Pointer to result fileproc
- * pointer area, or 0 if none
- * results Pointer to result socket
- * pointer area, or 0 if none
- *
- * Returns: EBADF The file descriptor is invalid
- * EOPNOTSUPP The file descriptor is not a socket
- * 0 Success
- *
- * Implicit returns:
- * *resultfp (modified) Fileproc pointer
- * *results (modified) socket pointer
- *
- * Notes: EOPNOTSUPP should probably be ENOTSOCK; this function is only
- * ever called from accept1().
- */
-int
-fp_getfsock(proc_t p, int fd, struct fileproc **resultfp,
- 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)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_SOCKET) {
- proc_fdunlock(p);
- return EOPNOTSUPP;
- }
- fp->f_iocount++;
-
- if (resultfp) {
- *resultfp = fp;
- }
- if (results) {
- *results = (struct socket *)fp->f_data;
- }
- proc_fdunlock(p);
-
- return 0;
-}
-
-
-/*
- * fp_getfkq
- *
- * Description: Get fileproc and kqueue pointer for a given fd from the
- * per process open file table of the specified process, and if
- * successful, increment the f_iocount
- *
- * Parameters: p Process in which fd lives
- * fd fd to get information for
- * resultfp Pointer to result fileproc
- * pointer area, or 0 if none
- * resultkq Pointer to result kqueue
- * pointer area, or 0 if none
- *
- * Returns: EBADF The file descriptor is invalid
- * EBADF The file descriptor is not a socket
- * 0 Success
- *
- * Implicit returns:
- * *resultfp (modified) Fileproc pointer
- * *resultkq (modified) kqueue pointer
- *
- * Notes: The second EBADF should probably be something else to make
- * the error condition distinct.
- */
-int
-fp_getfkq(proc_t p, int fd, struct fileproc **resultfp,
- 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)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_KQUEUE) {
- proc_fdunlock(p);
- return EBADF;
- }
- fp->f_iocount++;
-
- if (resultfp) {
- *resultfp = fp;
- }
- if (resultkq) {
- *resultkq = (struct kqueue *)fp->f_data;
- }
- proc_fdunlock(p);
-
- return 0;
-}
-
-
-/*
- * fp_getfpshm
- *
- * Description: Get fileproc and POSIX shared memory pointer for a given fd
- * from the per process open file table of the specified process
- * and if successful, increment the f_iocount
- *
- * Parameters: p Process in which fd lives
- * fd fd to get information for
- * resultfp Pointer to result fileproc
- * pointer area, or 0 if none
- * resultpshm Pointer to result POSIX
- * shared memory pointer
- * pointer area, or 0 if none
- *
- * Returns: EBADF The file descriptor is invalid
- * EBADF The file descriptor is not a POSIX
- * shared memory area
- * 0 Success
- *
- * Implicit returns:
- * *resultfp (modified) Fileproc pointer
- * *resultpshm (modified) POSIX shared memory pointer
- *
- * Notes: The second EBADF should probably be something else to make
- * the error condition distinct.
- */
-int
-fp_getfpshm(proc_t p, int fd, struct fileproc **resultfp,
- struct pshmnode **resultpshm)
-{
- struct filedesc *fdp = p->p_fd;
- struct fileproc *fp;
+ struct fileproc *fp = FILEPROC_NULL;
+ struct fileglob *fg = NULL;
+ int error = 0;
- proc_fdlock_spin(p);
- if (fd < 0 || fd >= fdp->fd_nfiles ||
- (fp = fdp->fd_ofiles[fd]) == NULL ||
- (fdp->fd_ofileflags[fd] & UF_RESERVED)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_PSXSHM) {
- proc_fdunlock(p);
- return EBADF;
+ if (p == NULL || result_pipe_id == NULL) {
+ return EINVAL;
}
- fp->f_iocount++;
- if (resultfp) {
- *resultfp = fp;
- }
- if (resultpshm) {
- *resultpshm = (struct pshmnode *)fp->f_data;
- }
- proc_fdunlock(p);
-
- return 0;
-}
-
-
-/*
- * fp_getfsem
- *
- * Description: Get fileproc and POSIX semaphore pointer for a given fd from
- * the per process open file table of the specified process
- * and if successful, increment the f_iocount
- *
- * Parameters: p Process in which fd lives
- * fd fd to get information for
- * resultfp Pointer to result fileproc
- * pointer area, or 0 if none
- * resultpsem Pointer to result POSIX
- * semaphore pointer area, or
- * 0 if none
- *
- * Returns: EBADF The file descriptor is invalid
- * EBADF The file descriptor is not a POSIX
- * semaphore
- * 0 Success
- *
- * Implicit returns:
- * *resultfp (modified) Fileproc pointer
- * *resultpsem (modified) POSIX semaphore pointer
- *
- * Notes: The second EBADF should probably be something else to make
- * the error condition distinct.
- *
- * In order to support unnamed POSIX semaphores, the named
- * POSIX semaphores will have to move out of the per-process
- * open filetable, and into a global table that is shared with
- * unnamed POSIX semaphores, since unnamed POSIX semaphores
- * are typically used by declaring instances in shared memory,
- * and there's no other way to do this without changing the
- * underlying type, which would introduce binary compatibility
- * issues.
- */
-int
-fp_getfpsem(proc_t p, int fd, struct fileproc **resultfp,
- 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)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_PSXSEM) {
+ proc_fdlock(p);
+ if ((error = fp_lookup(p, fd, &fp, 1))) {
proc_fdunlock(p);
- return EBADF;
- }
- fp->f_iocount++;
-
- if (resultfp) {
- *resultfp = fp;
- }
- if (resultpsem) {
- *resultpsem = (struct psemnode *)fp->f_data;
+ return error;
}
- proc_fdunlock(p);
-
- return 0;
-}
+ fg = fp->fp_glob;
-
-/*
- * fp_getfpipe
- *
- * Description: Get fileproc and pipe pointer for a given fd from the
- * per process open file table of the specified process
- * and if successful, increment the f_iocount
- *
- * Parameters: p Process in which fd lives
- * fd fd to get information for
- * resultfp Pointer to result fileproc
- * pointer area, or 0 if none
- * resultpipe Pointer to result pipe
- * pointer area, or 0 if none
- *
- * Returns: EBADF The file descriptor is invalid
- * EBADF The file descriptor is not a socket
- * 0 Success
- *
- * Implicit returns:
- * *resultfp (modified) Fileproc pointer
- * *resultpipe (modified) pipe pointer
- *
- * Notes: The second EBADF should probably be something else to make
- * the error condition distinct.
- */
-int
-fp_getfpipe(proc_t p, int fd, struct fileproc **resultfp,
- 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)) {
- proc_fdunlock(p);
- return EBADF;
- }
- if (fp->f_type != DTYPE_PIPE) {
- proc_fdunlock(p);
- return EBADF;
+ if (FILEGLOB_DTYPE(fg) == DTYPE_PIPE) {
+ *result_pipe_id = pipe_id((struct pipe*)fg->fg_data);
+ } else {
+ error = ENOTSUP;
}
- fp->f_iocount++;
- if (resultfp) {
- *resultfp = fp;
- }
- if (resultpipe) {
- *resultpipe = (struct pipe *)fp->f_data;
- }
+ fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
-
- return 0;
+ return error;
}
*
* Description: Get fileproc pointer for a given fd from the per process
* open file table of the specified process and if successful,
- * increment the f_iocount
+ * increment the fp_iocount
*
* Parameters: p Process in which fd lives
* fd fd to get information for
}
return EBADF;
}
- fp->f_iocount++;
+ os_ref_retain_locked(&fp->fp_iocount);
if (resultfp) {
*resultfp = fp;
* 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)
}
/*
* At this point, our caller (change_guardedfd_np) has
- * one f_iocount reference, and we just took another
+ * one fp_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, fp_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->fp_iocount) < 3 ||
+ 1 != os_ref_get_count(&nfp->fp_iocount)) {
+ panic("%s: fp_iocount", __func__);
+ } else if (3 == os_ref_get_count(&fp->fp_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;
- nfp->f_fglob = fp->f_fglob;
- nfp->f_wset = fp->f_wset;
+ nfp->fp_flags = (nfp->fp_flags & FP_TYPEMASK) |
+ (fp->fp_flags & ~FP_TYPEMASK);
+ os_ref_retain_locked(&nfp->fp_iocount);
+ os_ref_retain_locked(&nfp->fp_iocount);
+ nfp->fp_glob = fp->fp_glob;
+ nfp->fp_wset = fp->fp_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->fp_iocount);
+ os_ref_release_live(&fp->fp_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);
}
}
-/*
- * fp_drop_written
- *
- * Description: Set the FP_WRITTEN flag on the fileproc and drop the I/O
- * reference previously taken by calling fp_lookup et. al.
- *
- * Parameters: p Process in which the fd lives
- * fd fd associated with the fileproc
- * fp fileproc on which to set the
- * flag and drop the reference
- *
- * Returns: 0 Success
- * fp_drop:EBADF Bad file descriptor
- *
- * Locks: This function internally takes and drops the proc_fdlock for
- * the supplied process
- *
- * Notes: The fileproc must correspond to the fd in the supplied proc
- */
-int
-fp_drop_written(proc_t p, int fd, struct fileproc *fp)
-{
- int error;
-
- proc_fdlock_spin(p);
-
- fp->f_flags |= FP_WRITTEN;
-
- error = fp_drop(p, fd, fp, 1);
-
- proc_fdunlock(p);
-
- return error;
-}
-
-
-/*
- * fp_drop_event
- *
- * Description: Set the FP_WAITEVENT flag on the fileproc and drop the I/O
- * reference previously taken by calling fp_lookup et. al.
- *
- * Parameters: p Process in which the fd lives
- * fd fd associated with the fileproc
- * fp fileproc on which to set the
- * flag and drop the reference
- *
- * Returns: 0 Success
- * fp_drop:EBADF Bad file descriptor
- *
- * Locks: This function internally takes and drops the proc_fdlock for
- * the supplied process
- *
- * Notes: The fileproc must correspond to the fd in the supplied proc
- */
-int
-fp_drop_event(proc_t p, int fd, struct fileproc *fp)
-{
- int error;
-
- proc_fdlock_spin(p);
-
- fp->f_flags |= FP_WAITEVENT;
-
- error = fp_drop(p, fd, fp, 1);
-
- proc_fdunlock(p);
-
- return error;
-}
-
-
/*
* fp_drop
*
}
return EBADF;
}
- fp->f_iocount--;
- if (fp->f_iocount == 0) {
- if (fp->f_flags & FP_SELCONFLICT) {
- fp->f_flags &= ~FP_SELCONFLICT;
+ if (1 == os_ref_release_locked(&fp->fp_iocount)) {
+ if (fp->fp_flags & FP_SELCONFLICT) {
+ fp->fp_flags &= ~FP_SELCONFLICT;
}
if (p->p_fpdrainwait) {
* Locks: This function internally takes and drops the proc_fdlock for
* the current process
*
- * Notes: If successful, this function increments the f_iocount on the
+ * Notes: If successful, this function increments the fp_iocount on the
* fd's corresponding fileproc.
*
* The fileproc referenced is not returned; because of this, care
* not be recoverable from the vnode, if there is a subsequent
* close that destroys the associate fileproc. The caller should
* therefore retain their own reference on the fileproc so that
- * the f_iocount can be dropped subsequently. Failure to do this
+ * the fp_iocount can be dropped subsequently. Failure to do this
* can result in the returned pointer immediately becoming invalid
* following the call.
*
int
file_vnode(int fd, struct vnode **vpp)
{
- proc_t p = current_proc();
- struct fileproc *fp;
- int error;
-
- proc_fdlock_spin(p);
- if ((error = fp_lookup(p, fd, &fp, 1))) {
- proc_fdunlock(p);
- return error;
- }
- if (fp->f_type != DTYPE_VNODE) {
- fp_drop(p, fd, fp, 1);
- proc_fdunlock(p);
- return EINVAL;
- }
- if (vpp != NULL) {
- *vpp = (struct vnode *)fp->f_data;
- }
- proc_fdunlock(p);
-
- return 0;
+ return file_vnode_withvid(fd, vpp, NULL);
}
-
/*
* file_vnode_withvid
*
* Locks: This function internally takes and drops the proc_fdlock for
* the current process
*
- * Notes: If successful, this function increments the f_iocount on the
+ * Notes: If successful, this function increments the fp_iocount on the
* fd's corresponding fileproc.
*
* The fileproc referenced is not returned; because of this, care
* not be recoverable from the vnode, if there is a subsequent
* close that destroys the associate fileproc. The caller should
* therefore retain their own reference on the fileproc so that
- * the f_iocount can be dropped subsequently. Failure to do this
+ * the fp_iocount can be dropped subsequently. Failure to do this
* can result in the returned pointer immediately becoming invalid
* following the call.
*
* Use of this function is discouraged.
*/
int
-file_vnode_withvid(int fd, struct vnode **vpp, uint32_t * vidp)
+file_vnode_withvid(int fd, struct vnode **vpp, uint32_t *vidp)
{
- proc_t p = current_proc();
struct fileproc *fp;
- vnode_t vp;
- int error;
-
- proc_fdlock_spin(p);
- if ((error = fp_lookup(p, fd, &fp, 1))) {
- proc_fdunlock(p);
- return error;
- }
- if (fp->f_type != DTYPE_VNODE) {
- fp_drop(p, fd, fp, 1);
- proc_fdunlock(p);
- return EINVAL;
- }
- vp = (struct vnode *)fp->f_data;
- if (vpp != NULL) {
- *vpp = vp;
- }
+ int error;
- if ((vidp != NULL) && (vp != NULLVP)) {
- *vidp = (uint32_t)vp->v_id;
+ error = fp_get_ftype(current_proc(), fd, DTYPE_VNODE, EINVAL, &fp);
+ if (error == 0) {
+ if (vpp) {
+ *vpp = fp->f_data;
+ }
+ if (vidp) {
+ *vidp = vnode_vid(fp->f_data);
+ }
}
-
- proc_fdunlock(p);
-
- return 0;
+ return error;
}
-
/*
* file_socket
*
* Locks: This function internally takes and drops the proc_fdlock for
* the current process
*
- * Notes: If successful, this function increments the f_iocount on the
+ * Notes: If successful, this function increments the fp_iocount on the
* fd's corresponding fileproc.
*
* The fileproc referenced is not returned; because of this, care
* not be recoverable from the socket, if there is a subsequent
* close that destroys the associate fileproc. The caller should
* therefore retain their own reference on the fileproc so that
- * the f_iocount can be dropped subsequently. Failure to do this
+ * the fp_iocount can be dropped subsequently. Failure to do this
* can result in the returned pointer immediately becoming invalid
* following the call.
*
int
file_socket(int fd, struct socket **sp)
{
- proc_t p = current_proc();
struct fileproc *fp;
int error;
- proc_fdlock_spin(p);
- if ((error = fp_lookup(p, fd, &fp, 1))) {
- proc_fdunlock(p);
- return error;
- }
- if (fp->f_type != DTYPE_SOCKET) {
- fp_drop(p, fd, fp, 1);
- proc_fdunlock(p);
- return ENOTSOCK;
+ error = fp_get_ftype(current_proc(), fd, DTYPE_SOCKET, ENOTSOCK, &fp);
+ if (error == 0) {
+ if (sp) {
+ *sp = (struct socket *)fp->f_data;
+ }
}
- *sp = (struct socket *)fp->f_data;
- proc_fdunlock(p);
-
- return 0;
+ return error;
}
*
* Locks: This function internally takes and drops the proc_fdlock for
* the current process
- *
- * Notes: This function will internally increment and decrement the
- * f_iocount of the fileproc as part of its operation.
*/
int
file_flags(int fd, int *flags)
{
proc_t p = current_proc();
struct fileproc *fp;
- int error;
+ int error = EBADF;
proc_fdlock_spin(p);
- if ((error = fp_lookup(p, fd, &fp, 1))) {
- proc_fdunlock(p);
- return error;
+ fp = fp_get_noref_locked(p, fd);
+ if (fp) {
+ *flags = (int)fp->f_flag;
+ error = 0;
}
- *flags = (int)fp->f_flag;
- fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
- return 0;
+ return error;
}
* to be dropped
*
* Returns: 0 Success
- * EBADF Bad file descriptor
*
* Description: Given an fd, look it up in the current process's per process
- * open file table, and drop it's fileproc's f_iocount by one
+ * open file table, and drop it's fileproc's fp_iocount by one
*
* Notes: This is intended as a corresponding operation to the functions
* file_vnode() and file_socket() operations.
*
- * Technically, the close reference is supposed to be protected
- * by a fileproc_drain(), however, a drain will only block if
- * the fd refers to a character device, and that device has had
- * preparefileread() called on it. If it refers to something
- * other than a character device, then the drain will occur and
- * block each close attempt, rather than merely the last close.
- *
- * Since it's possible for an fd that refers to a character
- * device to have an intermediate close followed by an open to
- * cause a different file to correspond to that descriptor,
- * unless there was a cautionary reference taken on the fileproc,
- * this is an inherently unsafe function. This happens in the
- * case where multiple fd's in a process refer to the same
- * character device (e.g. stdin/out/err pointing to a tty, etc.).
+ * If the caller can't possibly hold an I/O reference,
+ * this function will panic the kernel rather than allowing
+ * for memory corruption. Callers should always call this
+ * because they acquired an I/O reference on this file before.
*
* Use of this function is discouraged.
*/
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))) {
- proc_fdunlock(p);
- return EBADF;
- }
- fp->f_iocount--;
+ fp = fp_get_noref_locked_with_iocount(p, fd);
- if (fp->f_iocount == 0) {
- if (fp->f_flags & FP_SELCONFLICT) {
- fp->f_flags &= ~FP_SELCONFLICT;
+ if (1 == os_ref_release_locked(&fp->fp_iocount)) {
+ if (fp->fp_flags & FP_SELCONFLICT) {
+ fp->fp_flags &= ~FP_SELCONFLICT;
}
if (p->p_fpdrainwait) {
}
-static int falloc_withalloc_locked(proc_t, struct fileproc **, int *,
- vfs_context_t, struct fileproc * (*)(void *), void *, int);
-
-/*
- * falloc
- *
- * Description: Allocate an entry in the per process open file table and
- * return the corresponding fileproc and fd.
- *
- * Parameters: p The process in whose open file
- * table the fd is to be allocated
- * resultfp Pointer to fileproc pointer
- * return area
- * resultfd Pointer to fd return area
- * ctx VFS context
- *
- * Returns: 0 Success
- * falloc:ENFILE Too many open files in system
- * falloc:EMFILE Too many open files in process
- * falloc:ENOMEM M_FILEPROC or M_FILEGLOB zone
- * exhausted
- *
- * Implicit returns:
- * *resultfd (modified) Returned fileproc pointer
- * *resultfd (modified) Returned fd
- *
- * Locks: This function takes and drops the proc_fdlock; if this lock
- * is already held, use falloc_locked() instead.
- *
- * Notes: This function takes separate process and context arguments
- * solely to support kern_exec.c; otherwise, it would take
- * neither, and expect falloc_locked() to use the
- * vfs_context_current() routine internally.
- */
-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);
-}
-
-/*
- * Like falloc, but including the fileproc allocator and create-args
- */
-int
-falloc_withalloc(proc_t p, struct fileproc **resultfp, int *resultfd,
- vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *arg)
-{
- int error;
-
- proc_fdlock(p);
- error = falloc_withalloc_locked(p,
- resultfp, resultfd, ctx, fp_zalloc, arg, 1);
- proc_fdunlock(p);
-
- return error;
-}
-
-/*
- * "uninitialized" ops -- ensure fg->fg_ops->fo_type always exists
- */
-static const struct fileops uninitops;
/*
- * falloc_locked
+ * falloc_withalloc
*
* Create a new open file structure and allocate
* a file descriptor for the process that refers to it.
* return area
* resultfd Pointer to fd return area
* ctx VFS context
- * locked Flag to indicate whether the
- * caller holds proc_fdlock
+ * fp_zalloc fileproc allocator to use
+ * crarg allocator args
*
* Returns: 0 Success
* ENFILE Too many open files in system
* fdalloc:EMFILE Too many open files in process
- * ENOMEM M_FILEPROC or M_FILEGLOB zone
+ * fdalloc:ENOMEM M_OFILETABL zone exhausted
+ * ENOMEM fp_zone or fg_zone zone
* exhausted
- * fdalloc:ENOMEM
*
* Implicit returns:
* *resultfd (modified) Returned fileproc pointer
* *resultfd (modified) Returned fd
*
- * Locks: If the parameter 'locked' is zero, this function takes and
- * drops the proc_fdlock; if non-zero, the caller must hold the
- * lock.
- *
- * Notes: If you intend to use a non-zero 'locked' parameter, use the
- * utility function falloc() instead.
- *
- * This function takes separate process and context arguments
+ * Notes: This function takes separate process and context arguments
* solely to support kern_exec.c; otherwise, it would take
* neither, and use the vfs_context_current() routine internally.
*/
int
-falloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
- vfs_context_t ctx, int locked)
-{
- return falloc_withalloc_locked(p, resultfp, resultfd, ctx,
- fileproc_alloc_init, NULL, locked);
-}
-
-static int
-falloc_withalloc_locked(proc_t p, struct fileproc **resultfp, int *resultfd,
- vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *crarg,
- int locked)
+falloc_withalloc(proc_t p, struct fileproc **resultfp, int *resultfd,
+ vfs_context_t ctx, fp_allocfn_t fp_zalloc, void *crarg)
{
struct fileproc *fp;
struct fileglob *fg;
int error, nfd;
- if (!locked) {
- proc_fdlock(p);
- }
- if ((error = fdalloc(p, 0, &nfd))) {
- if (!locked) {
- proc_fdunlock(p);
- }
- return error;
- }
+ /* Make sure we don't go beyond the system-wide limit */
if (nfiles >= maxfiles) {
- if (!locked) {
- proc_fdunlock(p);
- }
tablefull("file");
return ENFILE;
}
+
+ proc_fdlock(p);
+
+ /* fdalloc will make sure the process stays below per-process limit */
+ if ((error = fdalloc(p, 0, &nfd))) {
+ proc_fdunlock(p);
+ return error;
+ }
+
#if CONFIG_MACF
error = mac_file_check_create(proc_ucred(p));
if (error) {
- if (!locked) {
- proc_fdunlock(p);
- }
+ proc_fdunlock(p);
return error;
}
#endif
fp = (*fp_zalloc)(crarg);
if (fp == NULL) {
- if (locked) {
- proc_fdlock(p);
- }
- return ENOMEM;
- }
- MALLOC_ZONE(fg, struct fileglob *, sizeof(struct fileglob), M_FILEGLOB, M_WAITOK);
- if (fg == NULL) {
- fileproc_free(fp);
- if (locked) {
- proc_fdlock(p);
- }
return ENOMEM;
}
- bzero(fg, sizeof(struct fileglob));
+ fg = zalloc_flags(fg_zone, Z_WAITOK | Z_ZERO);
lck_mtx_init(&fg->fg_lock, file_lck_grp, file_lck_attr);
- fp->f_iocount = 1;
- fg->fg_count = 1;
+ os_ref_retain_locked(&fp->fp_iocount);
+ os_ref_init_raw(&fg->fg_count, &f_refgrp);
fg->fg_ops = &uninitops;
- fp->f_fglob = fg;
+ fp->fp_glob = fg;
#if CONFIG_MACF
mac_file_label_init(fg);
#endif
kauth_cred_ref(ctx->vc_ucred);
- proc_fdlock(p);
-
fp->f_cred = ctx->vc_ucred;
#if CONFIG_MACF
mac_file_label_associate(fp->f_cred, fg);
#endif
- OSAddAtomic(1, &nfiles);
+ os_atomic_inc(&nfiles, relaxed);
+
+ proc_fdlock(p);
p->p_fd->fd_ofiles[nfd] = fp;
- if (!locked) {
- proc_fdunlock(p);
- }
+ proc_fdunlock(p);
if (resultfp) {
*resultfp = fp;
return 0;
}
-
-/*
- * fg_free
- *
- * Description: Free a file structure; drop the global open file count, and
- * drop the credential reference, if the fileglob has one, and
- * destroy the instance mutex before freeing
- *
- * Parameters: fg Pointer to fileglob to be
- * freed
- *
- * Returns: void
- */
-void
-fg_free(struct fileglob *fg)
+int
+falloc(proc_t p, struct fileproc **resultfp, int *resultfd, vfs_context_t ctx)
{
- OSAddAtomic(-1, &nfiles);
-
- if (fg->fg_vn_data) {
- fg_vn_data_free(fg->fg_vn_data);
- fg->fg_vn_data = NULL;
- }
-
- if (IS_VALID_CRED(fg->fg_cred)) {
- kauth_cred_unref(&fg->fg_cred);
- }
- lck_mtx_destroy(&fg->fg_lock, file_lck_grp);
-
-#if CONFIG_MACF
- mac_file_label_destroy(fg);
-#endif
- FREE_ZONE(fg, sizeof *fg, M_FILEGLOB);
+ return falloc_withalloc(p, resultfp, resultfd, ctx,
+ fileproc_alloc_init, NULL);
}
-
-
/*
* fdexec
*
boolean_t cloexec_default = (flags & POSIX_SPAWN_CLOEXEC_DEFAULT) != 0;
thread_t self = current_thread();
struct uthread *ut = get_bsdthread_info(self);
- struct kqueue *dealloc_kq = NULL;
+ struct kqworkq *dealloc_kqwq = NULL;
/*
* If the current thread is bound as a workq/workloop
* subsequent kqueue closes go faster.
*/
knotes_dealloc(p);
- assert(fdp->fd_knlistsize == -1);
+ assert(fdp->fd_knlistsize == 0);
assert(fdp->fd_knhashmask == 0);
for (i = fdp->fd_lastfile; i >= 0; i--) {
if (
((*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->fp_glob))
#endif
) {
- procfdtbl_clearfd(p, i);
- if (i == fdp->fd_lastfile && i > 0) {
- fdp->fd_lastfile--;
- }
- 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) {
- 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);
+ fp_close_and_unlock(p, i, fp, 0);
+ proc_fdlock(p);
}
}
/* release the per-process workq kq */
if (fdp->fd_wqkqueue) {
- dealloc_kq = fdp->fd_wqkqueue;
+ dealloc_kqwq = fdp->fd_wqkqueue;
fdp->fd_wqkqueue = NULL;
}
proc_fdunlock(p);
/* Anything to free? */
- if (dealloc_kq) {
- kqueue_dealloc(dealloc_kq);
+ if (dealloc_kqwq) {
+ kqworkq_dealloc(dealloc_kqwq);
}
}
struct fileproc *ofp, *fp;
vnode_t v_dir;
- MALLOC_ZONE(newfdp, struct filedesc *,
- sizeof(*newfdp), M_FILEDESC, M_WAITOK);
- if (newfdp == NULL) {
- return NULL;
- }
+ newfdp = zalloc(fdp_zone);
proc_fdlock(p);
}
/* Coming from a chroot environment and unable to get a reference... */
if (newfdp->fd_rdir == NULL && fdp->fd_rdir) {
+ proc_fdunlock(p);
/*
* We couldn't get a new reference on
* the chroot directory being
if (newfdp->fd_cdir) {
vnode_rele(newfdp->fd_cdir);
}
- FREE_ZONE(newfdp, sizeof *newfdp, M_FILEDESC);
+ zfree(fdp_zone, newfdp);
return NULL;
}
}
proc_fdunlock(p);
- MALLOC_ZONE(newfdp->fd_ofiles, struct fileproc **,
+ MALLOC(newfdp->fd_ofiles, struct fileproc **,
i * OFILESIZE, M_OFILETABL, M_WAITOK);
if (newfdp->fd_ofiles == NULL) {
if (newfdp->fd_cdir) {
vnode_rele(newfdp->fd_rdir);
}
- FREE_ZONE(newfdp, sizeof(*newfdp), M_FILEDESC);
+ zfree(fdp_zone, newfdp);
return NULL;
}
(void) memset(newfdp->fd_ofiles, 0, i * OFILESIZE);
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 == (ofp->fp_glob->fg_lflags & FG_CONFINED) &&
0 == (*flags & (UF_FORKCLOSE | UF_RESERVED))) {
#if DEBUG
if (FILEPROC_TYPE(ofp) != FTYPE_SIMPLE) {
*/
*fpp = NULL;
} else {
- fp->f_flags |=
- (ofp->f_flags & ~FP_TYPEMASK);
- fp->f_fglob = ofp->f_fglob;
- (void)fg_ref(fp);
+ fp->fp_flags |=
+ (ofp->fp_flags & ~FP_TYPEMASK);
+ fp->fp_glob = ofp->fp_glob;
+ fg_ref(p, fp->fp_glob);
*fpp = fp;
}
} else {
* Initialize knote and kqueue tracking structs
*/
newfdp->fd_knlist = NULL;
- newfdp->fd_knlistsize = -1;
+ newfdp->fd_knlistsize = 0;
newfdp->fd_knhash = NULL;
newfdp->fd_knhashmask = 0;
newfdp->fd_kqhash = NULL;
{
struct filedesc *fdp;
struct fileproc *fp;
- struct kqueue *dealloc_kq = NULL;
+ struct kqworkq *dealloc_kqwq = NULL;
int i;
proc_fdlock(p);
* tables to make any subsequent kqueue closes faster.
*/
knotes_dealloc(p);
- assert(fdp->fd_knlistsize == -1);
+ assert(fdp->fd_knlistsize == 0);
assert(fdp->fd_knhashmask == 0);
/*
if (fdp->fd_ofileflags[i] & UF_RESERVED) {
panic("fdfree: found fp with UF_RESERVED");
}
-
- procfdtbl_reservefd(p, i);
-
- if (fp->f_flags & FP_WAITEVENT) {
- (void)waitevent_close(p, fp);
- }
- (void) closef_locked(fp, fp->f_fglob, p);
- fileproc_free(fp);
- }
- }
- FREE_ZONE(fdp->fd_ofiles, fdp->fd_nfiles * OFILESIZE, M_OFILETABL);
- fdp->fd_ofiles = NULL;
- fdp->fd_nfiles = 0;
- }
-
- if (fdp->fd_wqkqueue) {
- dealloc_kq = fdp->fd_wqkqueue;
- fdp->fd_wqkqueue = NULL;
- }
-
- proc_fdunlock(p);
-
- if (dealloc_kq) {
- kqueue_dealloc(dealloc_kq);
- }
-
- 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_kqhash) {
- for (uint32_t j = 0; j <= fdp->fd_kqhashmask; j++) {
- assert(SLIST_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);
-}
-
-/*
- * closef_locked
- *
- * Description: Internal form of closef; called with proc_fdlock held
- *
- * Parameters: fp Pointer to fileproc for fd
- * fg Pointer to fileglob for fd
- * p Pointer to proc structure
- *
- * Returns: 0 Success
- * closef_finish:??? Anything returnable by a per-fileops
- * close function
- *
- * Note: Decrements reference count on file structure; if this was the
- * last reference, then closef_finish() is called
- *
- * p and fp are allowed to be NULL when closing a file that was
- * being passed in a message (but only if we are called when this
- * is NOT the last reference).
- */
-int
-closef_locked(struct fileproc *fp, struct fileglob *fg, proc_t p)
-{
- struct vnode *vp;
- struct flock lf;
- struct vfs_context context;
- int error;
-
- if (fg == NULL) {
- return 0;
- }
-
- /* Set up context with cred stashed in fg */
- if (p == current_proc()) {
- context.vc_thread = current_thread();
- } else {
- context.vc_thread = NULL;
- }
- context.vc_ucred = fg->fg_cred;
-
- /*
- * POSIX record locking dictates that any close releases ALL
- * locks owned by this process. This is handled by setting
- * a flag in the unlock to free ONLY locks obeying POSIX
- * semantics, and not to free BSD-style file locks.
- * If the descriptor was in a message, POSIX-style locks
- * aren't passed with the descriptor.
- */
- if (p && (p->p_ladvflag & P_LADVLOCK) &&
- DTYPE_VNODE == FILEGLOB_DTYPE(fg)) {
- proc_fdunlock(p);
-
- lf.l_whence = SEEK_SET;
- lf.l_start = 0;
- lf.l_len = 0;
- lf.l_type = F_UNLCK;
- vp = (struct vnode *)fg->fg_data;
-
- if ((error = vnode_getwithref(vp)) == 0) {
- (void) VNOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_POSIX, &context, NULL);
- (void)vnode_put(vp);
+ fp_close_and_unlock(p, i, fp, 0);
+ proc_fdlock(p);
+ }
}
- proc_fdlock(p);
- }
- lck_mtx_lock_spin(&fg->fg_lock);
- fg->fg_count--;
-
- if (fg->fg_count > 0) {
- lck_mtx_unlock(&fg->fg_lock);
- return 0;
- }
-#if DIAGNOSTIC
- if (fg->fg_count != 0) {
- panic("fg %p: being freed with bad fg_count (%d)", fg, fg->fg_count);
+ FREE(fdp->fd_ofiles, M_OFILETABL);
+ fdp->fd_ofiles = NULL;
+ fdp->fd_nfiles = 0;
}
-#endif
- if (fp && (fp->f_flags & FP_WRITTEN)) {
- fg->fg_flag |= FWASWRITTEN;
+ if (fdp->fd_wqkqueue) {
+ dealloc_kqwq = fdp->fd_wqkqueue;
+ fdp->fd_wqkqueue = NULL;
}
- fg->fg_lflags |= FG_TERM;
- lck_mtx_unlock(&fg->fg_lock);
+ proc_fdunlock(p);
- if (p) {
- proc_fdunlock(p);
+ 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);
}
- /* Since we ensure that fg->fg_ops is always initialized,
- * it is safe to invoke fo_close on the fg */
- error = fo_close(fg, &context);
-
- fg_free(fg);
+ proc_fdlock_spin(p);
+ p->p_fd = NULL;
+ proc_fdunlock(p);
- if (p) {
- proc_fdlock(p);
+ if (fdp->fd_kqhash) {
+ for (uint32_t j = 0; j <= fdp->fd_kqhashmask; j++) {
+ assert(LIST_EMPTY(&fdp->fd_kqhash[j]));
+ }
+ hashdestroy(fdp->fd_kqhash, M_KQUEUE, fdp->fd_kqhashmask);
}
- return error;
-}
+ lck_mtx_destroy(&fdp->fd_kqhashlock, proc_kqhashlock_grp);
+ lck_mtx_destroy(&fdp->fd_knhashlock, proc_knhashlock_grp);
+ zfree(fdp_zone, fdp);
+}
/*
* fileproc_drain
fileproc_drain(proc_t p, struct fileproc * fp)
{
struct vfs_context context;
+ thread_t thread;
+ bool is_current_proc;
+
+ is_current_proc = (p == current_proc());
- context.vc_thread = proc_thread(p); /* XXX */
- context.vc_ucred = fp->f_fglob->fg_cred;
+ if (!is_current_proc) {
+ proc_lock(p);
+ thread = proc_thread(p); /* XXX */
+ thread_reference(thread);
+ proc_unlock(p);
+ } else {
+ thread = current_thread();
+ }
- fp->f_iocount--; /* (the one the close holds) */
+ context.vc_thread = thread;
+ context.vc_ucred = fp->fp_glob->fg_cred;
- while (fp->f_iocount) {
+ /* Set the vflag for drain */
+ fileproc_modify_vflags(fp, FPV_DRAIN, FALSE);
+
+ while (os_ref_get_count(&fp->fp_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);
- }
- if ((fp->f_flags & FP_INSELECT) == FP_INSELECT) {
- if (waitq_wakeup64_all((struct waitq *)fp->f_wset, NO_EVENT64,
+ fo_drain(fp, &context);
+ if ((fp->fp_flags & FP_INSELECT) == FP_INSELECT) {
+ if (waitq_wakeup64_all((struct waitq *)fp->fp_wset, NO_EVENT64,
THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT) {
- panic("bad wait queue for waitq_wakeup64_all %p (fp:%p)", fp->f_wset, fp);
+ panic("bad wait queue for waitq_wakeup64_all %p (fp:%p)", fp->fp_wset, fp);
}
}
- if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
+ if ((fp->fp_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
if (waitq_wakeup64_all(&select_conflict_queue, NO_EVENT64,
THREAD_INTERRUPTED, WAITQ_ALL_PRIORITIES) == KERN_INVALID_ARGUMENT) {
panic("bad select_conflict_queue");
msleep(&p->p_fpdrainwait, &p->p_fdmlock, PRIBIO, "fpdrain", NULL);
}
#if DIAGNOSTIC
- if ((fp->f_flags & FP_INSELECT) != 0) {
+ if ((fp->fp_flags & FP_INSELECT) != 0) {
panic("FP_INSELECT set on drained fp");
}
#endif
- if ((fp->f_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
- fp->f_flags &= ~FP_SELCONFLICT;
+ if ((fp->fp_flags & FP_SELCONFLICT) == FP_SELCONFLICT) {
+ fp->fp_flags &= ~FP_SELCONFLICT;
+ }
+
+ if (!is_current_proc) {
+ thread_deallocate(thread);
}
}
* 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);
fdrelse(p, fd);
proc_fdunlock(p);
- fg_free(fp->f_fglob);
+ fg_free(fp->fp_glob);
+ os_ref_release_live(&fp->fp_iocount);
fileproc_free(fp);
- return 0;
}
/*
- * flock
+ * sys_flock
*
* Description: Apply an advisory lock on a file descriptor.
*
* the entire file (l_whence = SEEK_SET, l_start = 0, l_len = 0).
*/
int
-flock(proc_t p, struct flock_args *uap, __unused int32_t *retval)
+sys_flock(proc_t p, struct flock_args *uap, __unused int32_t *retval)
{
int fd = uap->fd;
int how = uap->how;
lf.l_len = 0;
if (how & LOCK_UN) {
lf.l_type = F_UNLCK;
- fp->f_flag &= ~FHASLOCK;
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
goto out;
}
if (how & LOCK_EX) {
goto out;
}
#if CONFIG_MACF
- error = mac_file_check_lock(proc_ucred(p), fp->f_fglob, F_SETLK, &lf);
+ error = mac_file_check_lock(proc_ucred(p), fp->fp_glob, F_SETLK, &lf);
if (error) {
goto out;
}
#endif
- error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf,
+ error = VNOP_ADVLOCK(vp, (caddr_t)fp->fp_glob, F_SETLK, &lf,
(how & LOCK_NB ? F_FLOCK : F_FLOCK | F_WAIT),
ctx, NULL);
if (!error) {
- fp->f_flag |= FHASLOCK;
+ os_atomic_or(&fp->fp_glob->fg_flag, FWASLOCKED, relaxed);
}
out:
(void)vnode_put(vp);
}
/*
- * fileport_makeport
+ * sys_fileport_makeport
*
* Description: Obtain a Mach send right for a given file descriptor.
*
* On success, name of send right is stored at user-specified address.
*/
int
-fileport_makeport(proc_t p, struct fileport_makeport_args *uap,
+sys_fileport_makeport(proc_t p, struct fileport_makeport_args *uap,
__unused int *retval)
{
int err;
goto out_unlock;
}
- if (!file_issendable(p, fp)) {
+ fg = fp->fp_glob;
+ if (!fg_sendable(fg)) {
err = EINVAL;
goto out_unlock;
}
}
/* Dropped when port is deallocated */
- fg = fp->f_fglob;
- fg_ref(fp);
+ fg_ref(p, fg);
proc_fdunlock(p);
/* Allocate and initialize a port */
fileport = fileport_alloc(fg);
if (fileport == IPC_PORT_NULL) {
+ fg_drop_live(fg);
err = EAGAIN;
- fg_drop(fp);
goto out;
}
void
fileport_releasefg(struct fileglob *fg)
{
- (void)closef_locked(NULL, fg, PROC_NULL);
-
- return;
+ (void)fg_drop(PROC_NULL, fg);
}
-
/*
* 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
* *retval (modified) The new descriptor
*/
int
-fileport_makefd(proc_t p, struct fileport_makefd_args *uap, int32_t *retval)
+fileport_makefd(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;
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;
goto out;
}
- fp->f_fglob = fg;
- fg_ref(fp);
-
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;
+ }
+
+ fp->fp_glob = fg;
+ fg_ref(p, fg);
procfdtbl_releasefd(p, fd, fp);
proc_fdunlock(p);
fileproc_free(fp);
}
+ return err;
+}
+
+/*
+ * sys_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
+sys_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(p, port, UF_EXCLOSE, retval);
+ } else {
+ err = EINVAL;
+ }
+
if (IPC_PORT_NULL != port) {
ipc_port_release_send(port);
}
return EBADF;
}
#if CONFIG_MACF
- myerror = mac_file_check_dup(proc_ucred(p), wfp->f_fglob, dfd);
+ myerror = mac_file_check_dup(proc_ucred(p), wfp->fp_glob, dfd);
if (myerror) {
proc_fdunlock(p);
return myerror;
if (indx > fdp->fd_lastfile) {
fdp->fd_lastfile = indx;
}
- (void)fg_ref(wfp);
- if (fp->f_fglob) {
- fg_free(fp->f_fglob);
+ if (fp->fp_glob) {
+ fg_free(fp->fp_glob);
}
- fp->f_fglob = wfp->f_fglob;
+ fg_ref(p, wfp->fp_glob);
+ fp->fp_glob = wfp->fp_glob;
fdp->fd_ofileflags[indx] = fdp->fd_ofileflags[dfd] |
(flags & O_CLOEXEC) ? UF_EXCLOSE : 0;
}
-/*
- * fg_ref
- *
- * Description: Add a reference to a fileglob by fileproc
- *
- * Parameters: fp fileproc containing fileglob
- * pointer
- *
- * Returns: void
- *
- * Notes: XXX Should use OSAddAtomic?
- */
-void
-fg_ref(struct fileproc * fp)
-{
- struct fileglob *fg;
-
- fg = fp->f_fglob;
-
- lck_mtx_lock_spin(&fg->fg_lock);
-
-#if DIAGNOSTIC
- if ((fp->f_flags & ~((unsigned int)FP_VALID_FLAGS)) != 0) {
- panic("fg_ref: invalid bits on fp %p", fp);
- }
-
- 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);
-}
-
-
-/*
- * fg_drop
- *
- * Description: Remove a reference to a fileglob by fileproc
- *
- * Parameters: fp fileproc containing fileglob
- * pointer
- *
- * Returns: void
- *
- * Notes: XXX Should use OSAddAtomic?
- */
-void
-fg_drop(struct fileproc * fp)
-{
- struct fileglob *fg;
-
- fg = fp->f_fglob;
- lck_mtx_lock_spin(&fg->fg_lock);
- fg->fg_count--;
- lck_mtx_unlock(&fg->fg_lock);
-}
-
-#if SOCKETS
-/*
- * fg_insertuipc_mark
- *
- * Description: Mark fileglob for insertion onto message queue if needed
- * Also takes fileglob reference
- *
- * Parameters: fg Fileglob pointer to insert
- *
- * Returns: true, if the fileglob needs to be inserted onto msg queue
- *
- * Locks: Takes and drops fg_lock, potentially many times
- */
-boolean_t
-fg_insertuipc_mark(struct fileglob * fg)
-{
- boolean_t insert = FALSE;
-
- lck_mtx_lock_spin(&fg->fg_lock);
- while (fg->fg_lflags & FG_RMMSGQ) {
- lck_mtx_convert_spin(&fg->fg_lock);
-
- fg->fg_lflags |= FG_WRMMSGQ;
- msleep(&fg->fg_lflags, &fg->fg_lock, 0, "fg_insertuipc", NULL);
- }
-
- fg->fg_count++;
- fg->fg_msgcount++;
- if (fg->fg_msgcount == 1) {
- fg->fg_lflags |= FG_INSMSGQ;
- insert = TRUE;
- }
- lck_mtx_unlock(&fg->fg_lock);
- return insert;
-}
-
-/*
- * fg_insertuipc
- *
- * Description: Insert marked fileglob onto message queue
- *
- * Parameters: fg Fileglob pointer to insert
- *
- * Returns: void
- *
- * Locks: Takes and drops fg_lock & uipc_lock
- * DO NOT call this function with proc_fdlock held as unp_gc()
- * can potentially try to acquire proc_fdlock, which can result
- * in a deadlock if this function is in unp_gc_wait().
- */
-void
-fg_insertuipc(struct fileglob * fg)
-{
- if (fg->fg_lflags & FG_INSMSGQ) {
- lck_mtx_lock_spin(uipc_lock);
- unp_gc_wait();
- LIST_INSERT_HEAD(&fmsghead, fg, f_msglist);
- lck_mtx_unlock(uipc_lock);
- lck_mtx_lock(&fg->fg_lock);
- fg->fg_lflags &= ~FG_INSMSGQ;
- if (fg->fg_lflags & FG_WINSMSGQ) {
- fg->fg_lflags &= ~FG_WINSMSGQ;
- wakeup(&fg->fg_lflags);
- }
- lck_mtx_unlock(&fg->fg_lock);
- }
-}
-
-/*
- * fg_removeuipc_mark
- *
- * Description: Mark the fileglob for removal from message queue if needed
- * Also releases fileglob message queue reference
- *
- * Parameters: fg Fileglob pointer to remove
- *
- * Returns: true, if the fileglob needs to be removed from msg queue
- *
- * Locks: Takes and drops fg_lock, potentially many times
- */
-boolean_t
-fg_removeuipc_mark(struct fileglob * fg)
-{
- boolean_t remove = FALSE;
-
- lck_mtx_lock_spin(&fg->fg_lock);
- while (fg->fg_lflags & FG_INSMSGQ) {
- lck_mtx_convert_spin(&fg->fg_lock);
-
- fg->fg_lflags |= FG_WINSMSGQ;
- msleep(&fg->fg_lflags, &fg->fg_lock, 0, "fg_removeuipc", NULL);
- }
- fg->fg_msgcount--;
- if (fg->fg_msgcount == 0) {
- fg->fg_lflags |= FG_RMMSGQ;
- remove = TRUE;
- }
- lck_mtx_unlock(&fg->fg_lock);
- return remove;
-}
-
-/*
- * fg_removeuipc
- *
- * Description: Remove marked fileglob from message queue
- *
- * Parameters: fg Fileglob pointer to remove
- *
- * Returns: void
- *
- * Locks: Takes and drops fg_lock & uipc_lock
- * DO NOT call this function with proc_fdlock held as unp_gc()
- * can potentially try to acquire proc_fdlock, which can result
- * in a deadlock if this function is in unp_gc_wait().
- */
-void
-fg_removeuipc(struct fileglob * fg)
-{
- if (fg->fg_lflags & FG_RMMSGQ) {
- lck_mtx_lock_spin(uipc_lock);
- unp_gc_wait();
- LIST_REMOVE(fg, f_msglist);
- lck_mtx_unlock(uipc_lock);
- lck_mtx_lock(&fg->fg_lock);
- fg->fg_lflags &= ~FG_RMMSGQ;
- if (fg->fg_lflags & FG_WRMMSGQ) {
- fg->fg_lflags &= ~FG_WRMMSGQ;
- wakeup(&fg->fg_lflags);
- }
- lck_mtx_unlock(&fg->fg_lock);
- }
-}
-#endif /* SOCKETS */
-
/*
* fo_read
*
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;
+}
+
/*
* fo_write
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;
+}
+
/*
* fo_ioctl
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;
+}
+
/*
* fo_select
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;
+}
+
/*
* fo_close
}
+/*
+ * 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;
+}
+
+
/*
* fo_kqfilter
*
*
* 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,
- struct kevent_internal_s *kev, 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, kev, ctx);
+ return (*fp->f_ops->fo_kqfilter)(fp, kn, kev);
}
-/*
- * The ability to send a file descriptor to another
- * process is opt-in by file type.
- */
-boolean_t
-file_issendable(proc_t p, struct fileproc *fp)
+int
+fo_no_kqfilter(struct fileproc *fp, struct knote *kn, struct kevent_qos_s *kev)
{
- proc_fdlock_assert(p, LCK_MTX_ASSERT_OWNED);
-
- switch (fp->f_type) {
- case DTYPE_VNODE:
- case DTYPE_SOCKET:
- case DTYPE_PIPE:
- case DTYPE_PSXSHM:
- case DTYPE_NETPOLICY:
- return 0 == (fp->f_fglob->fg_lflags & FG_CONFINED);
- default:
- /* DTYPE_KQUEUE, DTYPE_FSEVENTS, DTYPE_PSXSEM */
- return FALSE;
- }
+#pragma unused(fp, kev)
+ knote_set_error(kn, ENOTSUP);
+ return 0;
}
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));
- }
+ struct fileproc *fp = zalloc_flags(fp_zone, Z_WAITOK | Z_ZERO);
+ os_ref_init(&fp->fp_iocount, &f_refgrp);
return fp;
}
+
void
fileproc_free(struct fileproc *fp)
{
+ os_ref_count_t __unused refc = os_ref_release(&fp->fp_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);
+ zfree(fp_zone, fp);
break;
case FTYPE_GUARDED:
guarded_fileproc_free(fp);
break;
default:
- panic("%s: corrupt fp %p flags %x", __func__, fp, fp->f_flags);
+ panic("%s: corrupt fp %p flags %x", __func__, fp, fp->fp_flags);
+ }
+}
+
+void
+fileproc_modify_vflags(struct fileproc *fp, fileproc_vflags_t vflags, boolean_t clearflags)
+{
+ if (clearflags) {
+ os_atomic_andnot(&fp->fp_vflags, vflags, relaxed);
+ } else {
+ os_atomic_or(&fp->fp_vflags, vflags, relaxed);
}
}
+
+fileproc_vflags_t
+fileproc_get_vflags(struct fileproc *fp)
+{
+ return os_atomic_load(&fp->fp_vflags, relaxed);
+}
projects / apple / xnu.git / blobdiff