dup2 w/flags
Subsystems (4)
PATCH 1/1 dup2 w/flags
This patch adds dup3 which is dup2 with an additional parameter. The parameter takes the usual O_* flags, with only O_CLOEXEC handled so far (it's possible to handle O_NONBLOCK in future, if this is wanted).
I've read some people mentioning that we simply should use a new fcntl() extension similar to F_DUPFD_CLOEXEC. I think that's a terrible idea and in any case not better than sys_indirect. A new syscall is the cleanest solution and it's really only a little bit of effort.
b/arch/x86/ia32/ia32entry.S | 1 + b/arch/x86/kernel/syscall_table_32.S | 1 + b/include/asm-x86/unistd_32.h | 1 + b/include/asm-x86/unistd_64.h | 2 ++ b/include/linux/syscalls.h | 1 + b/fs/fcntl.c | 15 +++++++++++++-- 6 files changed, 19 insertions(+), 2 deletions(-)
arch/x86/ia32/ia32entry.S +1 -0
@@ -736,2 +736,3 @@ .quad sys_epoll_create2 + .quad sys_dup3 /* 330 */ ia32_syscall_end:
arch/x86/kernel/syscall_table_32.S +1 -0
@@ -331,0 +332 @@ + .long sys_dup3 /* 330 */
include/asm-x86/unistd_32.h +1 -0
@@ -335,6 +335,7 @@#define __NR_signalfd4 327#define __NR_eventfd2 328#define __NR_epoll_create2 329 +#define __NR_dup3 330#ifdef __KERNEL__
include/asm-x86/unistd_64.h +2 -0
@@ -647,6 +647,8 @@ __SYSCALL(__NR_eventfd2, sys_eventfd2)#define __NR_epoll_create2 291 __SYSCALL(__NR_epoll_create2, sys_epoll_create2) +#define __NR_dup3 292 +__SYSCALL(__NR_dup3, sys_dup3)#ifndef __NO_STUBS
include/linux/syscalls.h +1 -0
@@ -303,6 +303,7 @@#endif asmlinkage long sys_dup (unsigned int fildes);asmlinkage long sys_dup2 (unsigned int oldfd,unsigned int newfd); +asmlinkage long sys_dup3 (unsigned int oldfd,unsigned int newfd,int flags);asmlinkage long sys_ioperm (unsigned long from,unsigned long num,int on);asmlinkage long sys_ioctl (unsigned int fd,unsigned int cmd,unsigned long arg);
fs/fcntl.c +13 -2
@@ -125,13 +125,16 @@ static int dupfd(struct file *file, unsigned int start, int cloexec)return fd; } -asmlinkage long sys_dup2 (unsigned int oldfd,unsigned int newfd) +asmlinkage long sys_dup3 (unsigned int oldfd,unsigned int newfd,int flags) {int err = -EBADF ;struct file * file, *tofree;struct files_struct * files = current->files;struct fdtable *fdt; +if ((flags & ~O_CLOEXEC ) !=0 ) +return -EINVAL ; +spin_lock (&files->file_lock);if (!(file =fcheck (oldfd))) goto out_unlock;
@@ -163,7 +166,10 @@ asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)rcu_assign_pointer (fdt->fd[newfd], file);FD_SET (newfd, fdt->open_fds); -FD_CLR (newfd, fdt->close_on_exec); +if (flags &O_CLOEXEC ) +FD_SET (newfd, fdt->close_on_exec); +else +FD_CLR (newfd, fdt->close_on_exec);spin_unlock (&files->file_lock);if (tofree)
@@ -181,6 +187,11 @@ out_fput: goto out; } +asmlinkage long sys_dup2 (unsigned int oldfd,unsigned int newfd) +{ +return sys_dup3 (oldfd, newfd,0 ); +} +asmlinkage long sys_dup (unsigned int fildes) {int ret = -EBADF ;
Quoted 8 lines · Ulrich Drepper
This patch adds dup3 which is dup2 with an additional parameter. The
parameter takes the usual O_* flags, with onlyO_CLOEXEChandled so far (it's possible to handleO_NONBLOCKin future, if this is wanted).I've read some people mentioning that we simply should use a new
fcntl()extension similar toF_DUPFD_CLOEXEC. I think that's a terrible idea and in any case not better thansys_indirect. A new syscall is the cleanest solution and it's really only a little bit of effort.
Ulrich,
To repeat my earlier request, and as requested in Documentation/HOWTO, could you please CC me on API changes. This alerts me to what may eventually need documenting in man-pages.
Cheers,
Michael
H. Peter Anvin
reviewerQuoted 7 lines · Michael Kerrisk
Ulrich,
To repeat my earlier request, and as requested in
Documentation/HOWTO, could you please CC me on API changes. This alerts me to what may eventually need documenting in man-pages.
I think we may need a mailing list for this; you're not the only person who would be affected by ABI-visible changes.
-hpa
Quoted 5 lines · H. Peter Anvin
Quoted 3 lines · Michael Kerrisk
Ulrich,
Quoted 8 lines
Quoted 5 lines
To repeat my earlier request, and as requested in
Documentation/HOWTO, could you please CC me on API changes. This alerts me to what may eventually need documenting in man-pages.I think we may need a mailing list for this; you're not the only person who
would be affected by ABI-visible changes.
I've been thinking the same thing.
Perhaps linux-api@vger?