aboutsummaryrefslogtreecommitdiff
path: root/linux-user/qemu.h
diff options
context:
space:
mode:
authorTimothy E Baldwin <T.E.Baldwin99@members.leeds.ac.uk>2016-05-12 18:47:46 +0100
committerRiku Voipio <riku.voipio@linaro.org>2016-05-27 14:49:51 +0300
commit4d330cee37a21aabfc619a1948953559e66951a4 (patch)
tree28df41a9f6e04a3332db1206df24e683a8782f0a /linux-user/qemu.h
parent71a8f7fece3e42dc55e865e081866f62f5c8c07e (diff)
downloadqemu-4d330cee37a21aabfc619a1948953559e66951a4.zip
qemu-4d330cee37a21aabfc619a1948953559e66951a4.tar.gz
qemu-4d330cee37a21aabfc619a1948953559e66951a4.tar.bz2
linux-user: Provide safe_syscall for fixing races between signals and syscalls
If a signal is delivered immediately before a blocking system call the handler will only be called after the system call returns, which may be a long time later or never. This is fixed by using a function (safe_syscall) that checks if a guest signal is pending prior to making a system call, and if so does not call the system call and returns -TARGET_ERESTARTSYS. If a signal is received between the check and the system call host_signal_handler() rewinds execution to before the check. This rewinding has the effect of closing the race window so that safe_syscall will reliably either (a) go into the host syscall with no unprocessed guest signals pending or or (b) return -TARGET_ERESTARTSYS so that the caller can deal with the signals. Implementing this requires a per-host-architecture assembly language fragment. This will also resolve the mishandling of the SA_RESTART flag where we would restart a host system call and not call the guest signal handler until the syscall finally completed -- syscall restarting now always happens at the guest syscall level so the guest signal handler will run. (The host syscall will never be restarted because if the host kernel rewinds the PC to point at the syscall insn for a restart then our host_signal_handler() will see this and arrange the guest PC rewind.) This commit contains the infrastructure for implementing safe_syscall and the assembly language fragment for x86-64, but does not change any syscalls to use it. Signed-off-by: Timothy Edward Baldwin <T.E.Baldwin99@members.leeds.ac.uk> Message-id: 1441497448-32489-14-git-send-email-T.E.Baldwin99@members.leeds.ac.uk [PMM: * Avoid having an architecture if-ladder in configure by putting linux-user/host/$(ARCH) on the include path and including safe-syscall.inc.S from it * Avoid ifdef ladder in signal.c by creating new hostdep.h to hold host-architecture-specific things * Added copyright/license header to safe-syscall.inc.S * Rewrote commit message * Added comments to safe-syscall.inc.S * Changed calling convention of safe_syscall() to match syscall() (returns -1 and host error in errno on failure) * Added a long comment in qemu.h about how to use safe_syscall() to implement guest syscalls. ] RV: squashed Peters "fixup! linux-user: compile on non-x86-64 hosts" patch Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Diffstat (limited to 'linux-user/qemu.h')
-rw-r--r--linux-user/qemu.h127
1 files changed, 126 insertions, 1 deletions
diff --git a/linux-user/qemu.h b/linux-user/qemu.h
index 208c63e..f09b750 100644
--- a/linux-user/qemu.h
+++ b/linux-user/qemu.h
@@ -1,7 +1,7 @@
#ifndef QEMU_H
#define QEMU_H
-
+#include "hostdep.h"
#include "cpu.h"
#include "exec/exec-all.h"
#include "exec/cpu_ldst.h"
@@ -205,6 +205,131 @@ unsigned long init_guest_space(unsigned long host_start,
#include "qemu/log.h"
+/* safe_syscall.S */
+
+/**
+ * safe_syscall:
+ * @int number: number of system call to make
+ * ...: arguments to the system call
+ *
+ * Call a system call if guest signal not pending.
+ * This has the same API as the libc syscall() function, except that it
+ * may return -1 with errno == TARGET_ERESTARTSYS if a signal was pending.
+ *
+ * Returns: the system call result, or -1 with an error code in errno
+ * (Errnos are host errnos; we rely on TARGET_ERESTARTSYS not clashing
+ * with any of the host errno values.)
+ */
+
+/* A guide to using safe_syscall() to handle interactions between guest
+ * syscalls and guest signals:
+ *
+ * Guest syscalls come in two flavours:
+ *
+ * (1) Non-interruptible syscalls
+ *
+ * These are guest syscalls that never get interrupted by signals and
+ * so never return EINTR. They can be implemented straightforwardly in
+ * QEMU: just make sure that if the implementation code has to make any
+ * blocking calls that those calls are retried if they return EINTR.
+ * It's also OK to implement these with safe_syscall, though it will be
+ * a little less efficient if a signal is delivered at the 'wrong' moment.
+ *
+ * (2) Interruptible syscalls
+ *
+ * These are guest syscalls that can be interrupted by signals and
+ * for which we need to either return EINTR or arrange for the guest
+ * syscall to be restarted. This category includes both syscalls which
+ * always restart (and in the kernel return -ERESTARTNOINTR), ones
+ * which only restart if there is no handler (kernel returns -ERESTARTNOHAND
+ * or -ERESTART_RESTARTBLOCK), and the most common kind which restart
+ * if the handler was registered with SA_RESTART (kernel returns
+ * -ERESTARTSYS). System calls which are only interruptible in some
+ * situations (like 'open') also need to be handled this way.
+ *
+ * Here it is important that the host syscall is made
+ * via this safe_syscall() function, and *not* via the host libc.
+ * If the host libc is used then the implementation will appear to work
+ * most of the time, but there will be a race condition where a
+ * signal could arrive just before we make the host syscall inside libc,
+ * and then then guest syscall will not correctly be interrupted.
+ * Instead the implementation of the guest syscall can use the safe_syscall
+ * function but otherwise just return the result or errno in the usual
+ * way; the main loop code will take care of restarting the syscall
+ * if appropriate.
+ *
+ * (If the implementation needs to make multiple host syscalls this is
+ * OK; any which might really block must be via safe_syscall(); for those
+ * which are only technically blocking (ie which we know in practice won't
+ * stay in the host kernel indefinitely) it's OK to use libc if necessary.
+ * You must be able to cope with backing out correctly if some safe_syscall
+ * you make in the implementation returns either -TARGET_ERESTARTSYS or
+ * EINTR though.)
+ *
+ *
+ * How and why the safe_syscall implementation works:
+ *
+ * The basic setup is that we make the host syscall via a known
+ * section of host native assembly. If a signal occurs, our signal
+ * handler checks the interrupted host PC against the addresse of that
+ * known section. If the PC is before or at the address of the syscall
+ * instruction then we change the PC to point at a "return
+ * -TARGET_ERESTARTSYS" code path instead, and then exit the signal handler
+ * (causing the safe_syscall() call to immediately return that value).
+ * Then in the main.c loop if we see this magic return value we adjust
+ * the guest PC to wind it back to before the system call, and invoke
+ * the guest signal handler as usual.
+ *
+ * This winding-back will happen in two cases:
+ * (1) signal came in just before we took the host syscall (a race);
+ * in this case we'll take the guest signal and have another go
+ * at the syscall afterwards, and this is indistinguishable for the
+ * guest from the timing having been different such that the guest
+ * signal really did win the race
+ * (2) signal came in while the host syscall was blocking, and the
+ * host kernel decided the syscall should be restarted;
+ * in this case we want to restart the guest syscall also, and so
+ * rewinding is the right thing. (Note that "restart" semantics mean
+ * "first call the signal handler, then reattempt the syscall".)
+ * The other situation to consider is when a signal came in while the
+ * host syscall was blocking, and the host kernel decided that the syscall
+ * should not be restarted; in this case QEMU's host signal handler will
+ * be invoked with the PC pointing just after the syscall instruction,
+ * with registers indicating an EINTR return; the special code in the
+ * handler will not kick in, and we will return EINTR to the guest as
+ * we should.
+ *
+ * Notice that we can leave the host kernel to make the decision for
+ * us about whether to do a restart of the syscall or not; we do not
+ * need to check SA_RESTART flags in QEMU or distinguish the various
+ * kinds of restartability.
+ */
+#ifdef HAVE_SAFE_SYSCALL
+/* The core part of this function is implemented in assembly */
+extern long safe_syscall_base(int *pending, long number, ...);
+
+#define safe_syscall(...) \
+ ({ \
+ long ret_; \
+ int *psp_ = &((TaskState *)thread_cpu->opaque)->signal_pending; \
+ ret_ = safe_syscall_base(psp_, __VA_ARGS__); \
+ if (is_error(ret_)) { \
+ errno = -ret_; \
+ ret_ = -1; \
+ } \
+ ret_; \
+ })
+
+#else
+
+/* Fallback for architectures which don't yet provide a safe-syscall assembly
+ * fragment; note that this is racy!
+ * This should go away when all host architectures have been updated.
+ */
+#define safe_syscall syscall
+
+#endif
+
/* syscall.c */
int host_to_target_waitstatus(int status);