| Message ID | 1454343665-1706-4-git-send-email-adhemerval.zanella@linaro.org |
|---|---|
| State | New |
| Headers | show |
On 02/01/2016 05:21 PM, Adhemerval Zanella wrote: > + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), > + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); Does this set up new per-thread variables? Otherwise, errno in the parent and child will be same and the code still has races. Florian
On 02-02-2016 11:05, Florian Weimer wrote: > On 02/01/2016 05:21 PM, Adhemerval Zanella wrote: > >> + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), >> + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); > > Does this set up new per-thread variables? Otherwise, errno in the > parent and child will be same and the code still has races. > > Florian > Could you elaborate? In my understanding there is no requirement of using CLONE_SETTLS to avoid races: CLONE_VFORK will suspend the calling process and even with child using the same TLS namespace as the parent there will be no concurrent access between them.
On Tue, 2016-02-02 at 11:31 -0200, Adhemerval Zanella wrote: > > On 02-02-2016 11:05, Florian Weimer wrote: > > On 02/01/2016 05:21 PM, Adhemerval Zanella wrote: > > > >> + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), > >> + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); > > > > Does this set up new per-thread variables? Otherwise, errno in the > > parent and child will be same and the code still has races. > > > > Florian > > > > Could you elaborate? In my understanding there is no requirement of > using CLONE_SETTLS to avoid races: CLONE_VFORK will suspend the > calling process and even with child using the same TLS namespace as > the parent there will be no concurrent access between them. Whatever you agree on eventually, it sounds as if this should be documented as part of the concurrency notes for this function.
On 03-02-2016 09:06, Torvald Riegel wrote: > On Tue, 2016-02-02 at 11:31 -0200, Adhemerval Zanella wrote: >> >> On 02-02-2016 11:05, Florian Weimer wrote: >>> On 02/01/2016 05:21 PM, Adhemerval Zanella wrote: >>> >>>> + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), >>>> + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); >>> >>> Does this set up new per-thread variables? Otherwise, errno in the >>> parent and child will be same and the code still has races. >>> >>> Florian >>> >> >> Could you elaborate? In my understanding there is no requirement of >> using CLONE_SETTLS to avoid races: CLONE_VFORK will suspend the >> calling process and even with child using the same TLS namespace as >> the parent there will be no concurrent access between them. > > Whatever you agree on eventually, it sounds as if this should be > documented as part of the concurrency notes for this function. > I think it is worth to comment internally on the function implementation about the CLONE_VFORK synchronization, but since it is transparent to user (the function will either spawn a new process or fails and it can be either through a syscall, {v}fork/exec or any other mechanism), I do not see why expose these implementation details explicit.
On Wed, 2016-02-03 at 10:05 -0200, Adhemerval Zanella wrote: > > On 03-02-2016 09:06, Torvald Riegel wrote: > > On Tue, 2016-02-02 at 11:31 -0200, Adhemerval Zanella wrote: > >> > >> On 02-02-2016 11:05, Florian Weimer wrote: > >>> On 02/01/2016 05:21 PM, Adhemerval Zanella wrote: > >>> > >>>> + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), > >>>> + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); > >>> > >>> Does this set up new per-thread variables? Otherwise, errno in the > >>> parent and child will be same and the code still has races. > >>> > >>> Florian > >>> > >> > >> Could you elaborate? In my understanding there is no requirement of > >> using CLONE_SETTLS to avoid races: CLONE_VFORK will suspend the > >> calling process and even with child using the same TLS namespace as > >> the parent there will be no concurrent access between them. > > > > Whatever you agree on eventually, it sounds as if this should be > > documented as part of the concurrency notes for this function. > > > > I think it is worth to comment internally on the function implementation > about the CLONE_VFORK synchronization, but since it is transparent > to user (the function will either spawn a new process or fails and it can > be either through a syscall, {v}fork/exec or any other mechanism), > I do not see why expose these implementation details explicit. I meant the internal documentation; IOW, comments in the code.
On 03-02-2016 10:06, Torvald Riegel wrote: > On Wed, 2016-02-03 at 10:05 -0200, Adhemerval Zanella wrote: >> >> On 03-02-2016 09:06, Torvald Riegel wrote: >>> On Tue, 2016-02-02 at 11:31 -0200, Adhemerval Zanella wrote: >>>> >>>> On 02-02-2016 11:05, Florian Weimer wrote: >>>>> On 02/01/2016 05:21 PM, Adhemerval Zanella wrote: >>>>> >>>>>> + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), >>>>>> + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); >>>>> >>>>> Does this set up new per-thread variables? Otherwise, errno in the >>>>> parent and child will be same and the code still has races. >>>>> >>>>> Florian >>>>> >>>> >>>> Could you elaborate? In my understanding there is no requirement of >>>> using CLONE_SETTLS to avoid races: CLONE_VFORK will suspend the >>>> calling process and even with child using the same TLS namespace as >>>> the parent there will be no concurrent access between them. >>> >>> Whatever you agree on eventually, it sounds as if this should be >>> documented as part of the concurrency notes for this function. >>> >> >> I think it is worth to comment internally on the function implementation >> about the CLONE_VFORK synchronization, but since it is transparent >> to user (the function will either spawn a new process or fails and it can >> be either through a syscall, {v}fork/exec or any other mechanism), >> I do not see why expose these implementation details explicit. > > I meant the internal documentation; IOW, comments in the code. > Fair enough, I will add them.
On Mon, Feb 01 2016, Adhemerval Zanella <adhemerval.zanella@linaro.org> wrote: > + for (cnt = 0; cnt < file_actions->__used; ++cnt) > + { > + struct __spawn_action *action = &file_actions->__actions[cnt]; > + > + /* Dup the pipe fd onto an unoccupied one to avoid any file > + operation to clobber it. */ > + if ((action->action.close_action.fd == p) > + || (action->action.open_action.fd == p) > + || (action->action.dup2_action.fd == p)) > + { > + if ((ret = __dup (p)) < 0) > + goto fail; > + p = ret; > + } > + Rather late to the party, but I think there's a few bugs here. Most importantly, dup() doesn't preserve the CLOEXEC flag, so if we do move the write end around like this, the fd will not automatically be closed during the exec, and hence the parent won't receive EOF and will block in read() call until the child finally exits. That's easily fixable with fcntl(p, F_DUPFD_CLOEXEC, 0). Pretty annoying to add a test for, though. Then there's the condition. First of all I think it's a little ugly reading the fd member from the "irrelevant" union members, even if gcc probably optimizes this to a single comparison. But more importantly, for the purpose of preventing clobbering our back channel to our parent, in the case of dup2_action the target aka newfd is much more relevant. That doesn't mean we don't also have to do some checking of the source of a dup2 action: If we allow dup2'ing it, we'd end up with the same problem as above; some copy of the write end of the pipe not being closed by exec(). Arguably, from the application's POV, p is not an open file descriptor, so we could fail the action (and hence the entire spawn) with EBADF. In fact, I can't really see what else we could do. For the close action, we can get by with just dupping the fd away and letting the action close the original (effectively making the action a no-op), but I also think that's an application bug. So I think the pipe clobber checking should be done in each action individually, since there's different logic that needs to be applied in each case. > + switch (action->tag) > + { > + case spawn_do_close: > + if ((ret = > + close_not_cancel (action->action.close_action.fd)) != 0) > + { > + if (!have_fdlimit) > + { > + __getrlimit64 (RLIMIT_NOFILE, &fdlimit); > + have_fdlimit = true; > + } > + > + /* Signal errors only for file descriptors out of range. */ > + if (action->action.close_action.fd < 0 > + || action->action.close_action.fd >= fdlimit.rlim_cur) > + goto fail; > + } I may have missed it in the original discussion, but what is the rationale for this? POSIX says If the file_actions argument is not NULL, and specifies any close, dup2, or open actions to be performed, and if posix_spawn() or posix_spawnp() fails for any of the reasons that would cause close(), dup2(), or open() to fail, an error value shall be returned as described by close(), dup2(), and open(), respectively > + break; > + > + case spawn_do_open: > + { > + ret = open_not_cancel (action->action.open_action.path, > + action->action. > + open_action.oflag | O_LARGEFILE, > + action->action.open_action.mode); > + > + if (ret == -1) > + goto fail; > + > + int new_fd = ret; > + > + /* Make sure the desired file descriptor is used. */ > + if (ret != action->action.open_action.fd) > + { > + if ((ret = __dup2 (new_fd, action->action.open_action.fd)) > + != action->action.open_action.fd) > + goto fail; > + > + if ((ret = close_not_cancel (new_fd)) != 0) > + goto fail; > + } > + } This is also how I'd have implemented it, but POSIX explicitly says If fildes was already an open file descriptor, it shall be closed before the new file is opened. Some, if slightly pathological, examples of how the difference could be observed: * ENFILE/EMFILE * Some single-open special device; following POSIX, 'action_open("/dev/watchdog", 47); action_open("/dev/watchdog", 47);' should both succeed if the first does, whereas the second will fail with the current code. > + break; > + > + case spawn_do_dup2: > + if ((ret = __dup2 (action->action.dup2_action.fd, > + action->action.dup2_action.newfd)) > + != action->action.dup2_action.newfd) > + goto fail; > + break; > + } > + } > + } > + > + /* Set the initial signal mask of the child if POSIX_SPAWN_SETSIGMASK > + is set, otherwise restore the previous one. */ > + __sigprocmask (SIG_SETMASK, (attr->__flags & POSIX_SPAWN_SETSIGMASK) > + ? &attr->__ss : &args->oldmask, 0); > + > + args->exec (args->file, args->argv, args->envp); > + > + /* This is compatibility function required to enable posix_spawn run > + script without shebang definition for older posix_spawn versions > + (2.15). */ > + maybe_script_execute (args); > + > + ret = -errno; > + > +fail: > + /* Since sizeof errno < PIPE_BUF, the write is atomic. */ > + ret = -ret; > + if (ret) > + while (write_not_cancel (p, &ret, sizeof ret) < 0) > + continue; This also seems wrong, at least if the intention is to write the proper errno value back. That only happens if we reach fail: after failing to exec - in most or all other cases, ret has the value -1, so we'd be writing EPERM to our parent. So why not just pull the value from errno in all cases? Rasmus
On Wed, 31 Aug 2016, Rasmus Villemoes wrote: > Rather late to the party, but I think there's a few bugs here. Most > importantly, dup() doesn't preserve the CLOEXEC flag, so if we do move > the write end around like this, the fd will not automatically be closed > during the exec, and hence the parent won't receive EOF and will block > in read() call until the child finally exits. That's easily fixable with > fcntl(p, F_DUPFD_CLOEXEC, 0). Pretty annoying to add a test for, though. In Linux-specific code we can assume the presence of dup3 (which needs to be called by the name __dup3 in implementations of POSIX functions).
One more thing: The signal blocking also seems to be flawed; sigprocmask silently removes SIGCANCEL and SIGSETXID before doing the syscall. Rasmus
On 31/08/2016 18:13, Rasmus Villemoes wrote: >> + switch (action->tag) >> + { >> + case spawn_do_close: >> + if ((ret = >> + close_not_cancel (action->action.close_action.fd)) != 0) >> + { >> + if (!have_fdlimit) >> + { >> + __getrlimit64 (RLIMIT_NOFILE, &fdlimit); >> + have_fdlimit = true; >> + } >> + >> + /* Signal errors only for file descriptors out of range. */ >> + if (action->action.close_action.fd < 0 >> + || action->action.close_action.fd >= fdlimit.rlim_cur) >> + goto fail; >> + } > > I may have missed it in the original discussion, but what is the > rationale for this? POSIX says > > If the file_actions argument is not NULL, and specifies any close, > dup2, or open actions to be performed, and if posix_spawn() or > posix_spawnp() fails for any of the reasons that would cause close(), > dup2(), or open() to fail, an error value shall be returned as > described by close(), dup2(), and open(), respectively I haven't joined the original discussion also (if any) for old implementation behaviour, but I think this conforms to austin group issue #370 [1] where it changed: fails for any of the reasons that would cause close( ), dup2( ), or open( ) to fail, an error value shall be returned to: fails for any of the reasons that would cause close( ), dup2( ), or open( ) to fail, other than attempting a close( ) on a file descriptor that is in range but already closed, an error value shall be returned The documentation at [2] seems to not have this updates issues description. [1] http://austingroupbugs.net/view.php?id=370 [2] http://pubs.opengroup.org/onlinepubs/9699919799/functions/posix_spawn.html > >> + break; >> + >> + case spawn_do_open: >> + { >> + ret = open_not_cancel (action->action.open_action.path, >> + action->action. >> + open_action.oflag | O_LARGEFILE, >> + action->action.open_action.mode); >> + >> + if (ret == -1) >> + goto fail; >> + >> + int new_fd = ret; >> + >> + /* Make sure the desired file descriptor is used. */ >> + if (ret != action->action.open_action.fd) >> + { >> + if ((ret = __dup2 (new_fd, action->action.open_action.fd)) >> + != action->action.open_action.fd) >> + goto fail; >> + >> + if ((ret = close_not_cancel (new_fd)) != 0) >> + goto fail; >> + } >> + } > > This is also how I'd have implemented it, but POSIX explicitly says > > If fildes was already an open file descriptor, it shall be closed > before the new file is opened. > > Some, if slightly pathological, examples of how the difference could be > observed: > > * ENFILE/EMFILE I think this should be an issue iff the program tries call posix_spawn with a total number of file descriptors equal to RLIMIT_NOFILE. > > * Some single-open special device; following POSIX, > 'action_open("/dev/watchdog", 47); action_open("/dev/watchdog", 47);' > should both succeed if the first does, whereas the second will fail > with the current code. Right, this should more problematic to handle. I think an option is to add a 'fcntl(fd, F_GETFD) != -1 || errno != EBADF' check before the open syscall and close the file descriptor in this case.
diff --git a/include/sched.h b/include/sched.h index 4f59397..b4d7406 100644 --- a/include/sched.h +++ b/include/sched.h @@ -19,7 +19,9 @@ extern int __sched_rr_get_interval (__pid_t __pid, struct timespec *__t); /* These are Linux specific. */ extern int __clone (int (*__fn) (void *__arg), void *__child_stack, int __flags, void *__arg, ...); +libc_hidden_proto (__clone) extern int __clone2 (int (*__fn) (void *__arg), void *__child_stack_base, size_t __child_stack_size, int __flags, void *__arg, ...); +libc_hidden_proto (__clone2) #endif #endif diff --git a/include/unistd.h b/include/unistd.h index 5152f64..d2802b2 100644 --- a/include/unistd.h +++ b/include/unistd.h @@ -81,6 +81,7 @@ char *__canonicalize_directory_name_internal (const char *__thisdir, size_t __size) attribute_hidden; extern int __dup (int __fd); +libc_hidden_proto (__dup) extern int __dup2 (int __fd, int __fd2); libc_hidden_proto (__dup2) extern int __dup3 (int __fd, int __fd2, int flags); diff --git a/posix/Makefile b/posix/Makefile index e251363..906817e 100644 --- a/posix/Makefile +++ b/posix/Makefile @@ -93,7 +93,7 @@ tests := tstgetopt testfnm runtests runptests \ xtests := bug-ga2 ifeq (yes,$(build-shared)) test-srcs := globtest -tests += wordexp-test tst-exec tst-spawn +tests += wordexp-test tst-exec tst-spawn tst-spawn2 endif tests-static = tst-exec-static tst-spawn-static tests += $(tests-static) diff --git a/posix/tst-spawn2.c b/posix/tst-spawn2.c new file mode 100644 index 0000000..41bfde8 --- /dev/null +++ b/posix/tst-spawn2.c @@ -0,0 +1,73 @@ +/* Further tests for spawn in case of invalid binary paths. + Copyright (C) 2016 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <spawn.h> +#include <error.h> +#include <errno.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#include <sys/wait.h> + +#include <stdio.h> + +int +posix_spawn_test (void) +{ + /* Check if posix_spawn correctly returns an error and an invalid pid + by trying to spawn an invalid binary. */ + + const char *program = "/path/to/invalid/binary"; + char * const args[] = { 0 }; + pid_t pid = -1; + + int ret = posix_spawn (&pid, program, 0, 0, args, environ); + if (ret != ENOENT) + error (EXIT_FAILURE, errno, "posix_spawn"); + + /* POSIX states the value returned on pid variable in case of an error + is not specified. GLIBC will update the value iff the child + execution is successful. */ + if (pid != -1) + error (EXIT_FAILURE, errno, "posix_spawn returned pid != -1"); + + /* Check if no child is actually created. */ + ret = waitpid (-1, NULL, 0); + if (ret != -1 || errno != ECHILD) + error (EXIT_FAILURE, errno, "waitpid"); + + /* Same as before, but with posix_spawnp. */ + char *args2[] = { (char*) program, 0 }; + + ret = posix_spawnp (&pid, args2[0], 0, 0, args2, environ); + if (ret != ENOENT) + error (EXIT_FAILURE, errno, "posix_spawnp"); + + if (pid != -1) + error (EXIT_FAILURE, errno, "posix_spawnp returned pid != -1"); + + ret = waitpid (-1, NULL, 0); + if (ret != -1 || errno != ECHILD) + error (EXIT_FAILURE, errno, "waitpid"); + + return 0; +} + +#define TEST_FUNCTION posix_spawn_test () +#include "../test-skeleton.c" + diff --git a/sysdeps/posix/dup.c b/sysdeps/posix/dup.c index 9525e76..abf3ff5 100644 --- a/sysdeps/posix/dup.c +++ b/sysdeps/posix/dup.c @@ -26,5 +26,5 @@ __dup (int fd) { return fcntl (fd, F_DUPFD, 0); } - +libc_hidden_def (__dup) weak_alias (__dup, dup) diff --git a/sysdeps/unix/sysv/linux/aarch64/clone.S b/sysdeps/unix/sysv/linux/aarch64/clone.S index 596fb9c..8f3ab40 100644 --- a/sysdeps/unix/sysv/linux/aarch64/clone.S +++ b/sysdeps/unix/sysv/linux/aarch64/clone.S @@ -93,4 +93,5 @@ thread_start: cfi_endproc .size thread_start, .-thread_start +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/alpha/clone.S b/sysdeps/unix/sysv/linux/alpha/clone.S index ec9c06d..556aa63 100644 --- a/sysdeps/unix/sysv/linux/alpha/clone.S +++ b/sysdeps/unix/sysv/linux/alpha/clone.S @@ -137,4 +137,5 @@ thread_start: cfi_endproc .end thread_start +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/arm/clone.S b/sysdeps/unix/sysv/linux/arm/clone.S index 460a8f5..c103123 100644 --- a/sysdeps/unix/sysv/linux/arm/clone.S +++ b/sysdeps/unix/sysv/linux/arm/clone.S @@ -93,4 +93,5 @@ PSEUDO_END (__clone) .fnend +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/hppa/clone.S b/sysdeps/unix/sysv/linux/hppa/clone.S index e80fd8d..0a18137 100644 --- a/sysdeps/unix/sysv/linux/hppa/clone.S +++ b/sysdeps/unix/sysv/linux/hppa/clone.S @@ -175,4 +175,5 @@ ENTRY(__clone) PSEUDO_END(__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/i386/clone.S b/sysdeps/unix/sysv/linux/i386/clone.S index 7d818c1..ef447d1 100644 --- a/sysdeps/unix/sysv/linux/i386/clone.S +++ b/sysdeps/unix/sysv/linux/i386/clone.S @@ -139,4 +139,5 @@ L(nomoregetpid): cfi_startproc PSEUDO_END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/ia64/clone2.S b/sysdeps/unix/sysv/linux/ia64/clone2.S index 9b257b3..fc046eb 100644 --- a/sysdeps/unix/sysv/linux/ia64/clone2.S +++ b/sysdeps/unix/sysv/linux/ia64/clone2.S @@ -96,6 +96,8 @@ ENTRY(__clone2) ret /* Not reached. */ PSEUDO_END(__clone2) +libc_hidden_def (__clone2) + /* For now we leave __clone undefined. This is unlikely to be a */ /* problem, since at least the i386 __clone in glibc always failed */ /* with a 0 sp (eventhough the kernel explicitly handled it). */ diff --git a/sysdeps/unix/sysv/linux/m68k/clone.S b/sysdeps/unix/sysv/linux/m68k/clone.S index 8b40df2..33474cc 100644 --- a/sysdeps/unix/sysv/linux/m68k/clone.S +++ b/sysdeps/unix/sysv/linux/m68k/clone.S @@ -127,4 +127,5 @@ donepid: cfi_startproc PSEUDO_END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/microblaze/clone.S b/sysdeps/unix/sysv/linux/microblaze/clone.S index 035d88b..cbc95ce 100644 --- a/sysdeps/unix/sysv/linux/microblaze/clone.S +++ b/sysdeps/unix/sysv/linux/microblaze/clone.S @@ -69,4 +69,5 @@ L(thread_start): nop PSEUDO_END(__clone) +libc_hidden_def (__clone) weak_alias (__clone,clone) diff --git a/sysdeps/unix/sysv/linux/mips/clone.S b/sysdeps/unix/sysv/linux/mips/clone.S index 755e8cc..feb8250 100644 --- a/sysdeps/unix/sysv/linux/mips/clone.S +++ b/sysdeps/unix/sysv/linux/mips/clone.S @@ -166,4 +166,5 @@ L(gotpid): END(__thread_start) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/nios2/clone.S b/sysdeps/unix/sysv/linux/nios2/clone.S index 4da5c19..e4d3970 100644 --- a/sysdeps/unix/sysv/linux/nios2/clone.S +++ b/sysdeps/unix/sysv/linux/nios2/clone.S @@ -104,4 +104,5 @@ thread_start: cfi_startproc PSEUDO_END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/nptl-signals.h b/sysdeps/unix/sysv/linux/nptl-signals.h index 7560a21..01f34c2 100644 --- a/sysdeps/unix/sysv/linux/nptl-signals.h +++ b/sysdeps/unix/sysv/linux/nptl-signals.h @@ -16,6 +16,8 @@ License along with the GNU C Library; if not, see <http://www.gnu.org/licenses/>. */ +#include <signal.h> + /* The signal used for asynchronous cancelation. */ #define SIGCANCEL __SIGRTMIN @@ -29,5 +31,13 @@ /* Signal used to implement the setuid et.al. functions. */ #define SIGSETXID (__SIGRTMIN + 1) + +/* Return is sig is used internally. */ +static inline int +__nptl_is_internal_signal (int sig) +{ + return (sig == SIGCANCEL) || (sig == SIGTIMER) || (sig == SIGSETXID); +} + /* Used to communicate with signal handler. */ extern struct xid_command *__xidcmd attribute_hidden; diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S b/sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S index 28948ea..eb973db 100644 --- a/sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S @@ -108,4 +108,5 @@ L(badargs): cfi_startproc END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S b/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S index c8c6de8..959fdb7 100644 --- a/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S +++ b/sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S @@ -126,4 +126,5 @@ L(parent): END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/s390/s390-32/clone.S b/sysdeps/unix/sysv/linux/s390/s390-32/clone.S index cb2afbb..f774e90 100644 --- a/sysdeps/unix/sysv/linux/s390/s390-32/clone.S +++ b/sysdeps/unix/sysv/linux/s390/s390-32/clone.S @@ -70,4 +70,6 @@ thread_start: xc 0(4,%r15),0(%r15) basr %r14,%r1 /* jump to fn */ DO_CALL (exit, 1) + +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/s390/s390-64/clone.S b/sysdeps/unix/sysv/linux/s390/s390-64/clone.S index eddab35..928a881 100644 --- a/sysdeps/unix/sysv/linux/s390/s390-64/clone.S +++ b/sysdeps/unix/sysv/linux/s390/s390-64/clone.S @@ -73,4 +73,6 @@ thread_start: xc 0(8,%r15),0(%r15) basr %r14,%r1 /* jump to fn */ DO_CALL (exit, 1) + +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/sh/clone.S b/sysdeps/unix/sysv/linux/sh/clone.S index 53cc08b..a73dd7d 100644 --- a/sysdeps/unix/sysv/linux/sh/clone.S +++ b/sysdeps/unix/sysv/linux/sh/clone.S @@ -123,4 +123,5 @@ ENTRY(__clone) .word TID - TLS_PRE_TCB_SIZE PSEUDO_END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/sparc/sparc32/clone.S b/sysdeps/unix/sysv/linux/sparc/sparc32/clone.S index 6f41f0f..68f8202 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc32/clone.S +++ b/sysdeps/unix/sysv/linux/sparc/sparc32/clone.S @@ -100,4 +100,5 @@ __thread_start: .size __thread_start, .-__thread_start +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/sparc/sparc64/clone.S b/sysdeps/unix/sysv/linux/sparc/sparc64/clone.S index e2d3914..cecffa7 100644 --- a/sysdeps/unix/sysv/linux/sparc/sparc64/clone.S +++ b/sysdeps/unix/sysv/linux/sparc/sparc64/clone.S @@ -96,4 +96,5 @@ __thread_start: .size __thread_start, .-__thread_start +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/spawni.c b/sysdeps/unix/sysv/linux/spawni.c new file mode 100644 index 0000000..407d696 --- /dev/null +++ b/sysdeps/unix/sysv/linux/spawni.c @@ -0,0 +1,424 @@ +/* POSIX spawn interface. Linux version. + Copyright (C) 2016 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <spawn.h> +#include <fcntl.h> +#include <paths.h> +#include <string.h> +#include <sys/resource.h> +#include <sys/wait.h> +#include <sys/param.h> +#include <sys/mman.h> +#include <not-cancel.h> +#include <local-setxid.h> +#include <shlib-compat.h> +#include <nptl/pthreadP.h> +#include <dl-sysdep.h> +#include <ldsodefs.h> +#include "spawn_int.h" + +/* The Linux implementation of posix_spawn{p} uses the clone syscall directly + with CLONE_VM and CLONE_VFORK flags and an allocated stack. The new stack + and start function solves most the vfork limitation (possible parent + clobber due stack spilling). The remaning issue are: + + 1. That no signal handlers must run in child context, to avoid corrupt + parent's state. + 2. The parent must ensure child's stack freeing. + 3. Child must synchronize with parent to enforce 2. and to possible + return execv issues. + + The first issue is solved by blocking all signals in child, even the + NPTL-internal ones (SIGCANCEL and SIGSETXID). The second and third issue + is done by a stack allocation in parent and a synchronization with using + a pipe or waitpid (in case or error). The pipe has the advantage of + allowing the child the signal an exec error. */ + + +/* The Unix standard contains a long explanation of the way to signal + an error after the fork() was successful. Since no new wait status + was wanted there is no way to signal an error using one of the + available methods. The committee chose to signal an error by a + normal program exit with the exit code 127. */ +#define SPAWN_ERROR 127 + +/* We need to block both SIGCANCEL and SIGSETXID. */ +#define SIGALL_SET \ + ((__sigset_t) { .__val = {[0 ... _SIGSET_NWORDS-1 ] = -1 } }) + +#ifdef __ia64__ +# define CLONE(__fn, __stack, __flags, __args) \ + __clone2 (__fn, __stack, __flags, __args, 0, 0, 0) +#else +# define CLONE(__fn, __stack, __flags, __args) \ + __clone (__fn, __stack, __flags, __args) +#endif + +#if _STACK_GROWS_DOWN +# define STACK(__stack, __stack_size) (__stack + __stack_size) +#elif _STACK_GROWS_UP +# define STACK(__stack, __stack_size) (__stack) +#endif + +/* Issues a 'pipe2' when the architecture supports or a 'pipe' plus O_CLOEXEC + otherwise. When pipe2 syscall is not available the pipe plus fcntl have a + inherent race condition (pipe2 is the right solution to avoid it). */ + +#ifdef __ASSUME_PIPE2 +# define __have_pipe2 1 +#endif + +static int +__do_pipe2_cloexec (int pipe_fds[2]) +{ + /* Try pipe2 even if __ASSUME_PIPE2 is not define and returning an error + iff the call returns ENOSYS. */ + int r = -1; + if (__have_pipe2 >= 0) + { + r = __pipe2 (pipe_fds, O_CLOEXEC); +#ifndef __ASSUME_PIPE2 + if (__have_pipe2 == 0) + __have_pipe2 = r != -1 || errno != ENOSYS ? 1 : -1; +#endif + + if (__have_pipe2 > 0) + return r; + } + if (__have_pipe2 < 0) + if (__pipe (pipe_fds) < 0) + return -1; + + /* Then apply FD_CLOCEXEC if it is supported in the pipe call case. */ + if (__have_pipe2 < 0) + { + if ((r = __fcntl (pipe_fds[0], F_SETFD, FD_CLOEXEC)) == -1 + || (r = __fcntl (pipe_fds[1], F_SETFD, FD_CLOEXEC)) == -1) + { + close_not_cancel (pipe_fds[0]); + close_not_cancel (pipe_fds[1]); + return r; + } + } + + return r; +} + +struct posix_spawn_args +{ + int pipe[2]; + sigset_t oldmask; + const char *file; + int (*exec) (const char *, char *const *, char *const *); + const posix_spawn_file_actions_t *fa; + const posix_spawnattr_t *restrict attr; + char *const *argv; + char *const *envp; + int xflags; +}; + +/* Older version requires that shell script without shebang definition + to be called explicity using /bin/sh (_PATH_BSHELL). */ +static void +maybe_script_execute (struct posix_spawn_args *args) +{ + if (SHLIB_COMPAT (libc, GLIBC_2_2, GLIBC_2_15) + && (args->xflags & SPAWN_XFLAGS_TRY_SHELL) && errno == ENOEXEC) + { + char *const *argv = args->argv; + int argc = 0; + while (argv[argc++] != NULL) + continue; + + /* Construct an argument list for the shell. */ + char *new_argv[argc]; + new_argv[0] = (char *) _PATH_BSHELL; + new_argv[1] = (char *) args->file; + while (argc > 1) + { + new_argv[argc] = argv[argc - 1]; + --argc; + } + + /* Execute the shell. */ + args->exec (new_argv[0], new_argv, args->envp); + } +} + +/* Function used in the clone call to setup the signals mask, posix_spawn + attributes, and file actions. It run on its own stack (provided by the + posix_spawn call). */ +static int +__spawni_child (void *arguments) +{ + struct posix_spawn_args *args = arguments; + const posix_spawnattr_t *restrict attr = args->attr; + const posix_spawn_file_actions_t *file_actions = args->fa; + int p = args->pipe[1]; + int ret; + + close_not_cancel (args->pipe[0]); + + /* The child must ensure that no signal handler are enabled because it shared + memory with parent, so the signal disposition must be either SIG_DFL or + SIG_IGN. It does by iterating over all signals and although it could + possibly be more optimized (by tracking which signal potentially have a + signal handler), it might requires system specific solutions (since the + sigset_t data type can be very different on different architectures). */ + struct sigaction sa; + memset (&sa, '\0', sizeof (sa)); + + sigset_t hset; + __sigprocmask (SIG_BLOCK, 0, &hset); + for (int sig = 1; sig < _NSIG; ++sig) + { + if ((attr->__flags & POSIX_SPAWN_SETSIGDEF) + && sigismember (&attr->__sd, sig)) + { + sa.sa_handler = SIG_DFL; + } + else if (sigismember (&hset, sig)) + { + if (__nptl_is_internal_signal (sig)) + sa.sa_handler = SIG_IGN; + else + { + __libc_sigaction (sig, 0, &sa); + if (sa.sa_handler == SIG_IGN) + continue; + sa.sa_handler = SIG_DFL; + } + } + else + continue; + + __libc_sigaction (sig, &sa, 0); + } + +#ifdef _POSIX_PRIORITY_SCHEDULING + /* Set the scheduling algorithm and parameters. */ + if ((attr->__flags & (POSIX_SPAWN_SETSCHEDPARAM | POSIX_SPAWN_SETSCHEDULER)) + == POSIX_SPAWN_SETSCHEDPARAM) + { + if ((ret = __sched_setparam (0, &attr->__sp)) == -1) + goto fail; + } + else if ((attr->__flags & POSIX_SPAWN_SETSCHEDULER) != 0) + { + if ((ret = __sched_setscheduler (0, attr->__policy, &attr->__sp)) == -1) + goto fail; + } +#endif + + /* Set the process group ID. */ + if ((attr->__flags & POSIX_SPAWN_SETPGROUP) != 0 + && (ret = __setpgid (0, attr->__pgrp)) != 0) + goto fail; + + /* Set the effective user and group IDs. */ + if ((attr->__flags & POSIX_SPAWN_RESETIDS) != 0 + && ((ret = local_seteuid (__getuid ())) != 0 + || (ret = local_setegid (__getgid ())) != 0)) + goto fail; + + /* Execute the file actions. */ + if (file_actions != 0) + { + int cnt; + struct rlimit64 fdlimit; + bool have_fdlimit = false; + + for (cnt = 0; cnt < file_actions->__used; ++cnt) + { + struct __spawn_action *action = &file_actions->__actions[cnt]; + + /* Dup the pipe fd onto an unoccupied one to avoid any file + operation to clobber it. */ + if ((action->action.close_action.fd == p) + || (action->action.open_action.fd == p) + || (action->action.dup2_action.fd == p)) + { + if ((ret = __dup (p)) < 0) + goto fail; + p = ret; + } + + switch (action->tag) + { + case spawn_do_close: + if ((ret = + close_not_cancel (action->action.close_action.fd)) != 0) + { + if (!have_fdlimit) + { + __getrlimit64 (RLIMIT_NOFILE, &fdlimit); + have_fdlimit = true; + } + + /* Signal errors only for file descriptors out of range. */ + if (action->action.close_action.fd < 0 + || action->action.close_action.fd >= fdlimit.rlim_cur) + goto fail; + } + break; + + case spawn_do_open: + { + ret = open_not_cancel (action->action.open_action.path, + action->action. + open_action.oflag | O_LARGEFILE, + action->action.open_action.mode); + + if (ret == -1) + goto fail; + + int new_fd = ret; + + /* Make sure the desired file descriptor is used. */ + if (ret != action->action.open_action.fd) + { + if ((ret = __dup2 (new_fd, action->action.open_action.fd)) + != action->action.open_action.fd) + goto fail; + + if ((ret = close_not_cancel (new_fd)) != 0) + goto fail; + } + } + break; + + case spawn_do_dup2: + if ((ret = __dup2 (action->action.dup2_action.fd, + action->action.dup2_action.newfd)) + != action->action.dup2_action.newfd) + goto fail; + break; + } + } + } + + /* Set the initial signal mask of the child if POSIX_SPAWN_SETSIGMASK + is set, otherwise restore the previous one. */ + __sigprocmask (SIG_SETMASK, (attr->__flags & POSIX_SPAWN_SETSIGMASK) + ? &attr->__ss : &args->oldmask, 0); + + args->exec (args->file, args->argv, args->envp); + + /* This is compatibility function required to enable posix_spawn run + script without shebang definition for older posix_spawn versions + (2.15). */ + maybe_script_execute (args); + + ret = -errno; + +fail: + /* Since sizeof errno < PIPE_BUF, the write is atomic. */ + ret = -ret; + if (ret) + while (write_not_cancel (p, &ret, sizeof ret) < 0) + continue; + exit (SPAWN_ERROR); +} + +/* Spawn a new process executing PATH with the attributes describes in *ATTRP. + Before running the process perform the actions described in FILE-ACTIONS. */ +static int +__spawnix (pid_t * pid, const char *file, + const posix_spawn_file_actions_t * file_actions, + const posix_spawnattr_t * attrp, char *const argv[], + char *const envp[], int xflags, + int (*exec) (const char *, char *const *, char *const *)) +{ + pid_t new_pid; + struct posix_spawn_args args; + int ec; + + if (__do_pipe2_cloexec (args.pipe)) + return errno; + + /* Allocates a child stack with sufficient size to handle a large argument + set in both the 'maybe_script_execute' compatibility code and on the + execvpe (posix_spawnp) where it might allocate stack size when path + does not contain a slash. */ + const int prot = (PROT_READ | PROT_WRITE + | ((GL (dl_stack_flags) & PF_X) ? PROT_EXEC : 0)); + + const int stack_size = ARCH_STACK_DEFAULT_SIZE; + void *stack = __mmap (NULL, stack_size, prot, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, -1, 0); + if (__glibc_unlikely (stack == MAP_FAILED)) + { + close_not_cancel (args.pipe[0]); + close_not_cancel (args.pipe[1]); + return errno; + } + + /* Disable asynchronous cancellation. */ + int cs = LIBC_CANCEL_ASYNC (); + + args.file = file; + args.exec = exec; + args.fa = file_actions; + args.attr = attrp ? attrp : &(const posix_spawnattr_t) { 0 }; + args.argv = argv; + args.envp = envp; + args.xflags = xflags; + + __sigprocmask (SIG_BLOCK, &SIGALL_SET, &args.oldmask); + + new_pid = CLONE (__spawni_child, STACK (stack, stack_size), + CLONE_VM | CLONE_VFORK | SIGCHLD, &args); + + close_not_cancel (args.pipe[1]); + + if (new_pid > 0) + { + if (__read (args.pipe[0], &ec, sizeof ec) != sizeof ec) + ec = 0; + else + __waitpid (new_pid, NULL, 0); + } + else + ec = -new_pid; + + __munmap (stack, stack_size); + + close_not_cancel (args.pipe[0]); + + if (!ec && new_pid) + *pid = new_pid; + + __sigprocmask (SIG_SETMASK, &args.oldmask, 0); + + LIBC_CANCEL_RESET (cs); + + return ec; +} + +/* Spawn a new process executing PATH with the attributes describes in *ATTRP. + Before running the process perform the actions described in FILE-ACTIONS. */ +int +__spawni (pid_t * pid, const char *file, + const posix_spawn_file_actions_t * acts, + const posix_spawnattr_t * attrp, char *const argv[], + char *const envp[], int xflags) +{ + if (xflags & SPAWN_XFLAGS_USE_PATH) + return __spawnix (pid, file, acts, attrp, argv, envp, xflags, __execvpe); + return __spawnix (pid, file, acts, attrp, argv, envp, xflags, __execve); +} diff --git a/sysdeps/unix/sysv/linux/tile/clone.S b/sysdeps/unix/sysv/linux/tile/clone.S index 02fe3a8..a300eb5 100644 --- a/sysdeps/unix/sysv/linux/tile/clone.S +++ b/sysdeps/unix/sysv/linux/tile/clone.S @@ -219,4 +219,5 @@ ENTRY (__clone) } PSEUDO_END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone) diff --git a/sysdeps/unix/sysv/linux/x86_64/clone.S b/sysdeps/unix/sysv/linux/x86_64/clone.S index 382568a..1a50957 100644 --- a/sysdeps/unix/sysv/linux/x86_64/clone.S +++ b/sysdeps/unix/sysv/linux/x86_64/clone.S @@ -115,4 +115,5 @@ L(thread_start): cfi_startproc; PSEUDO_END (__clone) +libc_hidden_def (__clone) weak_alias (__clone, clone)
This patch implements a new posix_spawn{p} implementation for Linux. The main difference is it uses the clone syscall directly with CLONE_VM and CLONE_VFORK flags and a direct allocated stack. The new stack and start function solves most the vfork limitation (possible parent clobber due stack spilling). The remaning issue are related to signal handling: 1. That no signal handlers must run in child context, to avoid corrupt parent's state. 2. Child must synchronize with parent to enforce stack deallocation and to possible return execv issues. The first one is solved by blocking all signals in child, even NPTL-internal ones (SIGCANCEL and SIGSETXID). The second issue is done by a stack allocation in parent and a synchronization with using a pipe or waitpid (in case or error). The pipe has the advantage of allowing the child signal an exec error (checked with new tst-spawn2 test). There is an inherent race condition in pipe2 usage for architectures that do not support the syscall directly. In such cases the a pipe plus fctnl is used instead and it may lead to file descriptor leak in parent (as decribed by fcntl documentation). The child process stack is allocate with a mmap with MAP_STACK flag using default architecture stack size. Although it is slower than use a stack buffer from parent, it allows some slack for the compatibility code to run scripts with no shebang (which may use a buffer with size depending of argument list count). Performance should be similar to the vfork default posix implementation and way faster than fork path (vfork on mostly linux ports are basically clone with CLONE_VM plus CLONE_VFORK). The only difference is the syscalls required for the stack allocation/deallocation. It fixes BZ#10354, BZ#14750, and BZ#18433. Changes from previous version: - Remove O_CLOCEXEC define (since one can assume it always defined within glibc/linux). Tested on i386, x86_64, powerpc64le, and aarch64. [BZ #14750] [BZ #10354] [BZ #18433] * include/sched.h (__clone): Add hidden prototype. (__clone2): Likewise. * include/unistd.h (__dup): Likewise. * posix/Makefile (tests): Add tst-spawn2. * posix/tst-spawn2.c: New file. * sysdeps/posix/dup.c (__dup): Add hidden definition. * sysdeps/unix/sysv/linux/aarch64/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/alpha/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/arm/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/hppa/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/i386/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/ia64/clone2.S (__clone): Likewise. * sysdeps/unix/sysv/linux/m68k/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/microblaze/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/mips/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/nios2/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/s390/s390-32/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/s390/s390-64/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/sh/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/sparc/sparc32/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/sparc/sparc64/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/tile/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/x86_64/clone.S (__clone): Likewise. * sysdeps/unix/sysv/linux/nptl-signals.h (____nptl_is_internal_signal): New function. * sysdeps/unix/sysv/linux/spawni.c: New file. --- include/sched.h | 2 + include/unistd.h | 1 + posix/Makefile | 2 +- posix/tst-spawn2.c | 73 ++++ sysdeps/posix/dup.c | 2 +- sysdeps/unix/sysv/linux/aarch64/clone.S | 1 + sysdeps/unix/sysv/linux/alpha/clone.S | 1 + sysdeps/unix/sysv/linux/arm/clone.S | 1 + sysdeps/unix/sysv/linux/hppa/clone.S | 1 + sysdeps/unix/sysv/linux/i386/clone.S | 1 + sysdeps/unix/sysv/linux/ia64/clone2.S | 2 + sysdeps/unix/sysv/linux/m68k/clone.S | 1 + sysdeps/unix/sysv/linux/microblaze/clone.S | 1 + sysdeps/unix/sysv/linux/mips/clone.S | 1 + sysdeps/unix/sysv/linux/nios2/clone.S | 1 + sysdeps/unix/sysv/linux/nptl-signals.h | 10 + sysdeps/unix/sysv/linux/powerpc/powerpc32/clone.S | 1 + sysdeps/unix/sysv/linux/powerpc/powerpc64/clone.S | 1 + sysdeps/unix/sysv/linux/s390/s390-32/clone.S | 2 + sysdeps/unix/sysv/linux/s390/s390-64/clone.S | 2 + sysdeps/unix/sysv/linux/sh/clone.S | 1 + sysdeps/unix/sysv/linux/sparc/sparc32/clone.S | 1 + sysdeps/unix/sysv/linux/sparc/sparc64/clone.S | 1 + sysdeps/unix/sysv/linux/spawni.c | 424 ++++++++++++++++++++++ sysdeps/unix/sysv/linux/tile/clone.S | 1 + sysdeps/unix/sysv/linux/x86_64/clone.S | 1 + 27 files changed, 568 insertions(+), 2 deletions(-) create mode 100644 posix/tst-spawn2.c create mode 100644 sysdeps/unix/sysv/linux/spawni.c