* [PATCH v3 1/7] um: Move faultinfo extraction into userspace routine
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 2/7] um: Add stub side of SECCOMP/futex based process handling Benjamin Berg
` (5 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg
The segv handler is called slightly differently depending on whether
PTRACE_FULL_FAULTINFO is set or not (32bit vs. 64bit). The only
difference is that we don't try to pass the registers and instruction
pointer to the segv handler.
It would be good to either document or remove the difference, but I do
not know why this difference exists. And, passing NULL can even result
in a crash.
Signed-off-by: Benjamin Berg <benjamin@sipsolutions.net>
---
arch/um/os-Linux/skas/process.c | 17 ++++++-----------
1 file changed, 6 insertions(+), 11 deletions(-)
diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
index ae2aea062f06..97c2f964f5fc 100644
--- a/arch/um/os-Linux/skas/process.c
+++ b/arch/um/os-Linux/skas/process.c
@@ -163,12 +163,6 @@ static void get_skas_faultinfo(int pid, struct faultinfo *fi)
memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
}
-static void handle_segv(int pid, struct uml_pt_regs *regs)
-{
- get_skas_faultinfo(pid, ®s->faultinfo);
- segv(regs->faultinfo, 0, 1, NULL, NULL);
-}
-
static void handle_trap(int pid, struct uml_pt_regs *regs)
{
if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
@@ -521,13 +515,14 @@ void userspace(struct uml_pt_regs *regs)
switch (sig) {
case SIGSEGV:
- if (PTRACE_FULL_FAULTINFO) {
- get_skas_faultinfo(pid,
- ®s->faultinfo);
+ get_skas_faultinfo(pid, ®s->faultinfo);
+
+ if (PTRACE_FULL_FAULTINFO)
(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
regs, NULL);
- }
- else handle_segv(pid, regs);
+ else
+ segv(regs->faultinfo, 0, 1, NULL, NULL);
+
break;
case SIGTRAP + 0x80:
handle_trap(pid, regs);
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH v3 2/7] um: Add stub side of SECCOMP/futex based process handling
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 1/7] um: Move faultinfo extraction into userspace routine Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 3/7] um: Add helper functions to get/set state for SECCOMP Benjamin Berg
` (4 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg, Johannes Berg, Benjamin Berg
This adds the stub side for the new seccomp process management code. In
this case we do register save/restore through the signal handler
mcontext.
Add special code for handling TLS, which for x86_64 means setting the
FS_BASE/GS_BASE registers while for i386 it means calling the
set_thread_area syscall.
Co-authored-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
v1:
- Cleanup futex EINTR/EAGAIN handling
RFCv2:
- Add include guards into new architecture specific header file
---
arch/um/include/shared/common-offsets.h | 2 +
arch/um/include/shared/skas/stub-data.h | 14 +++++++
arch/um/kernel/skas/stub.c | 49 +++++++++++++++++++++++++
arch/x86/um/shared/sysdep/stub-data.h | 23 ++++++++++++
arch/x86/um/shared/sysdep/stub.h | 2 +
arch/x86/um/shared/sysdep/stub_32.h | 13 +++++++
arch/x86/um/shared/sysdep/stub_64.h | 17 +++++++++
7 files changed, 120 insertions(+)
create mode 100644 arch/x86/um/shared/sysdep/stub-data.h
diff --git a/arch/um/include/shared/common-offsets.h b/arch/um/include/shared/common-offsets.h
index 73f3a4792ed8..93e7097a2922 100644
--- a/arch/um/include/shared/common-offsets.h
+++ b/arch/um/include/shared/common-offsets.h
@@ -14,3 +14,5 @@ DEFINE(UM_THREAD_SIZE, THREAD_SIZE);
DEFINE(UM_NSEC_PER_SEC, NSEC_PER_SEC);
DEFINE(UM_NSEC_PER_USEC, NSEC_PER_USEC);
+
+DEFINE(UM_KERN_GDT_ENTRY_TLS_ENTRIES, GDT_ENTRY_TLS_ENTRIES);
diff --git a/arch/um/include/shared/skas/stub-data.h b/arch/um/include/shared/skas/stub-data.h
index 81a4cace032c..81ac2cd12112 100644
--- a/arch/um/include/shared/skas/stub-data.h
+++ b/arch/um/include/shared/skas/stub-data.h
@@ -11,6 +11,10 @@
#include <linux/compiler_types.h>
#include <as-layout.h>
#include <sysdep/tls.h>
+#include <sysdep/stub-data.h>
+
+#define FUTEX_IN_CHILD 0
+#define FUTEX_IN_KERN 1
struct stub_init_data {
unsigned long stub_start;
@@ -52,6 +56,16 @@ struct stub_data {
/* 128 leaves enough room for additional fields in the struct */
struct stub_syscall syscall_data[(UM_KERN_PAGE_SIZE - 128) / sizeof(struct stub_syscall)] __aligned(16);
+ /* data shared with signal handler (only used in seccomp mode) */
+ short restart_wait;
+ unsigned int futex;
+ int signal;
+ unsigned short si_offset;
+ unsigned short mctx_offset;
+
+ /* seccomp architecture specific state restore */
+ struct stub_data_arch arch_data;
+
/* Stack for our signal handlers and for calling into . */
unsigned char sigstack[UM_KERN_PAGE_SIZE] __aligned(UM_KERN_PAGE_SIZE);
};
diff --git a/arch/um/kernel/skas/stub.c b/arch/um/kernel/skas/stub.c
index 796fc266d3bb..9041f6b6e28b 100644
--- a/arch/um/kernel/skas/stub.c
+++ b/arch/um/kernel/skas/stub.c
@@ -5,6 +5,9 @@
#include <sysdep/stub.h>
+#include <linux/futex.h>
+#include <errno.h>
+
static __always_inline int syscall_handler(struct stub_data *d)
{
int i;
@@ -57,3 +60,49 @@ stub_syscall_handler(void)
trap_myself();
}
+
+void __section(".__syscall_stub")
+stub_signal_interrupt(int sig, siginfo_t *info, void *p)
+{
+ struct stub_data *d = get_stub_data();
+ ucontext_t *uc = p;
+ long res;
+
+ d->signal = sig;
+ d->si_offset = (unsigned long)info - (unsigned long)&d->sigstack[0];
+ d->mctx_offset = (unsigned long)&uc->uc_mcontext - (unsigned long)&d->sigstack[0];
+
+restart_wait:
+ d->futex = FUTEX_IN_KERN;
+ do {
+ res = stub_syscall3(__NR_futex, (unsigned long)&d->futex,
+ FUTEX_WAKE, 1);
+ } while (res == -EINTR);
+ do {
+ res = stub_syscall4(__NR_futex, (unsigned long)&d->futex,
+ FUTEX_WAIT, FUTEX_IN_KERN, 0);
+ } while (res == -EINTR || d->futex == FUTEX_IN_KERN);
+
+ if (res < 0 && res != -EAGAIN)
+ stub_syscall1(__NR_exit_group, 1);
+
+ /* Try running queued syscalls. */
+ if (syscall_handler(d) < 0 || d->restart_wait) {
+ /* Report SIGSYS if we restart. */
+ d->signal = SIGSYS;
+ d->restart_wait = 0;
+ goto restart_wait;
+ }
+
+ /* Restore arch dependent state that is not part of the mcontext */
+ stub_seccomp_restore_state(&d->arch_data);
+
+ /* Return so that the host modified mcontext is restored. */
+}
+
+void __section(".__syscall_stub")
+stub_signal_restorer(void)
+{
+ /* We must not have anything on the stack when doing rt_sigreturn */
+ stub_syscall0(__NR_rt_sigreturn);
+}
diff --git a/arch/x86/um/shared/sysdep/stub-data.h b/arch/x86/um/shared/sysdep/stub-data.h
new file mode 100644
index 000000000000..82b1b7f8ac3d
--- /dev/null
+++ b/arch/x86/um/shared/sysdep/stub-data.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __ARCH_STUB_DATA_H
+#define __ARCH_STUB_DATA_H
+
+#ifdef __i386__
+#include <generated/asm-offsets.h>
+#include <asm/ldt.h>
+
+struct stub_data_arch {
+ int sync;
+ struct user_desc tls[UM_KERN_GDT_ENTRY_TLS_ENTRIES];
+};
+#else
+#define STUB_SYNC_FS_BASE (1 << 0)
+#define STUB_SYNC_GS_BASE (1 << 1)
+struct stub_data_arch {
+ int sync;
+ unsigned long fs_base;
+ unsigned long gs_base;
+};
+#endif
+
+#endif /* __ARCH_STUB_DATA_H */
diff --git a/arch/x86/um/shared/sysdep/stub.h b/arch/x86/um/shared/sysdep/stub.h
index dc89f4423454..4fa58f5b4fca 100644
--- a/arch/x86/um/shared/sysdep/stub.h
+++ b/arch/x86/um/shared/sysdep/stub.h
@@ -13,3 +13,5 @@
extern void stub_segv_handler(int, siginfo_t *, void *);
extern void stub_syscall_handler(void);
+extern void stub_signal_interrupt(int, siginfo_t *, void *);
+extern void stub_signal_restorer(void);
diff --git a/arch/x86/um/shared/sysdep/stub_32.h b/arch/x86/um/shared/sysdep/stub_32.h
index 390988132c0a..df568fc3ceb4 100644
--- a/arch/x86/um/shared/sysdep/stub_32.h
+++ b/arch/x86/um/shared/sysdep/stub_32.h
@@ -131,4 +131,17 @@ static __always_inline void *get_stub_data(void)
"call *%%eax ;" \
:: "i" ((1 + STUB_DATA_PAGES) * UM_KERN_PAGE_SIZE), \
"i" (&fn))
+
+static __always_inline void
+stub_seccomp_restore_state(struct stub_data_arch *arch)
+{
+ for (int i = 0; i < sizeof(arch->tls) / sizeof(arch->tls[0]); i++) {
+ if (arch->sync & (1 << i))
+ stub_syscall1(__NR_set_thread_area,
+ (unsigned long) &arch->tls[i]);
+ }
+
+ arch->sync = 0;
+}
+
#endif
diff --git a/arch/x86/um/shared/sysdep/stub_64.h b/arch/x86/um/shared/sysdep/stub_64.h
index 294affbec742..9cfd31afa769 100644
--- a/arch/x86/um/shared/sysdep/stub_64.h
+++ b/arch/x86/um/shared/sysdep/stub_64.h
@@ -10,6 +10,7 @@
#include <sysdep/ptrace_user.h>
#include <generated/asm-offsets.h>
#include <linux/stddef.h>
+#include <asm/prctl.h>
#define STUB_MMAP_NR __NR_mmap
#define MMAP_OFFSET(o) (o)
@@ -134,4 +135,20 @@ static __always_inline void *get_stub_data(void)
"call *%%rax ;" \
:: "i" ((1 + STUB_DATA_PAGES) * UM_KERN_PAGE_SIZE), \
"i" (&fn))
+
+static __always_inline void
+stub_seccomp_restore_state(struct stub_data_arch *arch)
+{
+ /*
+ * We could use _writefsbase_u64/_writegsbase_u64 if the host reports
+ * support in the hwcaps (HWCAP2_FSGSBASE).
+ */
+ if (arch->sync & STUB_SYNC_FS_BASE)
+ stub_syscall2(__NR_arch_prctl, ARCH_SET_FS, arch->fs_base);
+ if (arch->sync & STUB_SYNC_GS_BASE)
+ stub_syscall2(__NR_arch_prctl, ARCH_SET_GS, arch->gs_base);
+
+ arch->sync = 0;
+}
+
#endif
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH v3 3/7] um: Add helper functions to get/set state for SECCOMP
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 1/7] um: Move faultinfo extraction into userspace routine Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 2/7] um: Add stub side of SECCOMP/futex based process handling Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 4/7] um: Track userspace children dying in SECCOMP mode Benjamin Berg
` (3 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg, Benjamin Berg
When not using ptrace, we need to both save and restore registers
through the mcontext as provided by the host kernel to our signal
handlers.
Add corresponding functions to store the state to an mcontext and
helpers to access the mcontext of the subprocess through the stub data.
Signed-off-by: Benjamin Berg <benjamin@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
RFCv2:
- Proper FP register handling
---
arch/x86/um/os-Linux/mcontext.c | 218 ++++++++++++++++++++++++++-
arch/x86/um/ptrace.c | 76 +++++++---
arch/x86/um/shared/sysdep/mcontext.h | 9 ++
3 files changed, 283 insertions(+), 20 deletions(-)
diff --git a/arch/x86/um/os-Linux/mcontext.c b/arch/x86/um/os-Linux/mcontext.c
index 37decaa74761..e661fdc44db9 100644
--- a/arch/x86/um/os-Linux/mcontext.c
+++ b/arch/x86/um/os-Linux/mcontext.c
@@ -1,7 +1,10 @@
// SPDX-License-Identifier: GPL-2.0
-#include <sys/ucontext.h>
#define __FRAME_OFFSETS
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <sys/ucontext.h>
#include <asm/ptrace.h>
+#include <asm/sigcontext.h>
#include <sysdep/ptrace.h>
#include <sysdep/mcontext.h>
#include <arch.h>
@@ -18,6 +21,10 @@ void get_regs_from_mc(struct uml_pt_regs *regs, mcontext_t *mc)
COPY2(UESP, ESP); /* sic */
COPY(EBX); COPY(EDX); COPY(ECX); COPY(EAX);
COPY(EIP); COPY_SEG_CPL3(CS); COPY(EFL); COPY_SEG_CPL3(SS);
+#undef COPY2
+#undef COPY
+#undef COPY_SEG
+#undef COPY_SEG_CPL3
#else
#define COPY2(X,Y) regs->gp[X/sizeof(unsigned long)] = mc->gregs[REG_##Y]
#define COPY(X) regs->gp[X/sizeof(unsigned long)] = mc->gregs[REG_##X]
@@ -29,6 +36,8 @@ void get_regs_from_mc(struct uml_pt_regs *regs, mcontext_t *mc)
COPY2(EFLAGS, EFL);
COPY2(CS, CSGSFS);
regs->gp[SS / sizeof(unsigned long)] = mc->gregs[REG_CSGSFS] >> 48;
+#undef COPY2
+#undef COPY
#endif
}
@@ -42,3 +51,210 @@ void mc_set_rip(void *_mc, void *target)
mc->gregs[REG_RIP] = (unsigned long)target;
#endif
}
+
+/* Same thing, but the copy macros are turned around. */
+void get_mc_from_regs(struct uml_pt_regs *regs, mcontext_t *mc, int single_stepping)
+{
+#ifdef __i386__
+#define COPY2(X,Y) mc->gregs[REG_##Y] = regs->gp[X]
+#define COPY(X) mc->gregs[REG_##X] = regs->gp[X]
+#define COPY_SEG(X) mc->gregs[REG_##X] = regs->gp[X] & 0xffff;
+#define COPY_SEG_CPL3(X) mc->gregs[REG_##X] = (regs->gp[X] & 0xffff) | 3;
+ COPY_SEG(GS); COPY_SEG(FS); COPY_SEG(ES); COPY_SEG(DS);
+ COPY(EDI); COPY(ESI); COPY(EBP);
+ COPY2(UESP, ESP); /* sic */
+ COPY(EBX); COPY(EDX); COPY(ECX); COPY(EAX);
+ COPY(EIP); COPY_SEG_CPL3(CS); COPY(EFL); COPY_SEG_CPL3(SS);
+#else
+#define COPY2(X,Y) mc->gregs[REG_##Y] = regs->gp[X/sizeof(unsigned long)]
+#define COPY(X) mc->gregs[REG_##X] = regs->gp[X/sizeof(unsigned long)]
+ COPY(R8); COPY(R9); COPY(R10); COPY(R11);
+ COPY(R12); COPY(R13); COPY(R14); COPY(R15);
+ COPY(RDI); COPY(RSI); COPY(RBP); COPY(RBX);
+ COPY(RDX); COPY(RAX); COPY(RCX); COPY(RSP);
+ COPY(RIP);
+ COPY2(EFLAGS, EFL);
+ mc->gregs[REG_CSGSFS] = mc->gregs[REG_CSGSFS] & 0xffffffffffffl;
+ mc->gregs[REG_CSGSFS] |= (regs->gp[SS / sizeof(unsigned long)] & 0xffff) << 48;
+#endif
+
+ if (single_stepping)
+ mc->gregs[REG_EFL] |= X86_EFLAGS_TF;
+ else
+ mc->gregs[REG_EFL] &= ~X86_EFLAGS_TF;
+}
+
+#ifdef CONFIG_X86_32
+struct _xstate_64 {
+ struct _fpstate_64 fpstate;
+ struct _header xstate_hdr;
+ struct _ymmh_state ymmh;
+ /* New processor state extensions go here: */
+};
+
+/* Not quite the right structures as these contain more information */
+int um_i387_from_fxsr(struct _fpstate_32 *i387,
+ const struct _fpstate_64 *fxsave);
+int um_fxsr_from_i387(struct _fpstate_64 *fxsave,
+ const struct _fpstate_32 *from);
+#else
+#define _xstate_64 _xstate
+#endif
+
+static struct _fpstate *get_fpstate(struct stub_data *data,
+ mcontext_t *mcontext,
+ int *fp_size)
+{
+ struct _fpstate *res;
+
+ /* Assume floating point registers are on the same page */
+ res = (void *)(((unsigned long)mcontext->fpregs &
+ (UM_KERN_PAGE_SIZE - 1)) +
+ (unsigned long)&data->sigstack[0]);
+
+ if ((void *)res + sizeof(struct _fpstate) >
+ (void *)data->sigstack + sizeof(data->sigstack))
+ return NULL;
+
+ if (res->sw_reserved.magic1 != FP_XSTATE_MAGIC1) {
+ *fp_size = sizeof(struct _fpstate);
+ } else {
+ char *magic2_addr;
+
+ magic2_addr = (void *)res;
+ magic2_addr += res->sw_reserved.extended_size;
+ magic2_addr -= FP_XSTATE_MAGIC2_SIZE;
+
+ /* We still need to be within our stack */
+ if ((void *)magic2_addr >
+ (void *)data->sigstack + sizeof(data->sigstack))
+ return NULL;
+
+ /* If we do not read MAGIC2, then we did something wrong */
+ if (*(__u32 *)magic2_addr != FP_XSTATE_MAGIC2)
+ return NULL;
+
+ /* Remove MAGIC2 from the size, we do not save/restore it */
+ *fp_size = res->sw_reserved.extended_size -
+ FP_XSTATE_MAGIC2_SIZE;
+ }
+
+ return res;
+}
+
+int get_stub_state(struct uml_pt_regs *regs, struct stub_data *data,
+ unsigned long *fp_size_out)
+{
+ mcontext_t *mcontext;
+ struct _fpstate *fpstate_stub;
+ struct _xstate_64 *xstate_stub;
+ int fp_size, xstate_size;
+
+ /* mctx_offset is verified by wait_stub_done_seccomp */
+ mcontext = (void *)&data->sigstack[data->mctx_offset];
+
+ get_regs_from_mc(regs, mcontext);
+
+ fpstate_stub = get_fpstate(data, mcontext, &fp_size);
+ if (!fpstate_stub)
+ return -EINVAL;
+
+#ifdef CONFIG_X86_32
+ xstate_stub = (void *)&fpstate_stub->_fxsr_env;
+ xstate_size = fp_size - offsetof(struct _fpstate_32, _fxsr_env);
+#else
+ xstate_stub = (void *)fpstate_stub;
+ xstate_size = fp_size;
+#endif
+
+ if (fp_size_out)
+ *fp_size_out = xstate_size;
+
+ if (xstate_size > host_fp_size)
+ return -ENOSPC;
+
+ memcpy(®s->fp, xstate_stub, xstate_size);
+
+ /* We do not need to read the x86_64 FS_BASE/GS_BASE registers as
+ * we do not permit userspace to set them directly.
+ */
+
+#ifdef CONFIG_X86_32
+ /* Read the i387 legacy FP registers */
+ if (um_fxsr_from_i387((void *)®s->fp, fpstate_stub))
+ return -EINVAL;
+#endif
+
+ return 0;
+}
+
+/* Copied because we cannot include regset.h here. */
+struct task_struct;
+struct user_regset;
+struct membuf {
+ void *p;
+ size_t left;
+};
+
+int fpregs_legacy_get(struct task_struct *target,
+ const struct user_regset *regset,
+ struct membuf to);
+
+int set_stub_state(struct uml_pt_regs *regs, struct stub_data *data,
+ int single_stepping)
+{
+ mcontext_t *mcontext;
+ struct _fpstate *fpstate_stub;
+ struct _xstate_64 *xstate_stub;
+ int fp_size, xstate_size;
+
+ /* mctx_offset is verified by wait_stub_done_seccomp */
+ mcontext = (void *)&data->sigstack[data->mctx_offset];
+
+ if ((unsigned long)mcontext < (unsigned long)data->sigstack ||
+ (unsigned long)mcontext >
+ (unsigned long) data->sigstack +
+ sizeof(data->sigstack) - sizeof(*mcontext))
+ return -EINVAL;
+
+ get_mc_from_regs(regs, mcontext, single_stepping);
+
+ fpstate_stub = get_fpstate(data, mcontext, &fp_size);
+ if (!fpstate_stub)
+ return -EINVAL;
+
+#ifdef CONFIG_X86_32
+ xstate_stub = (void *)&fpstate_stub->_fxsr_env;
+ xstate_size = fp_size - offsetof(struct _fpstate_32, _fxsr_env);
+#else
+ xstate_stub = (void *)fpstate_stub;
+ xstate_size = fp_size;
+#endif
+
+ memcpy(fpstate_stub, ®s->fp, fp_size);
+
+#ifdef __i386__
+ /*
+ * On x86, the GDT entries are updated by arch_set_tls.
+ */
+
+ /* Store the i387 legacy FP registers which the host will use */
+ if (um_i387_from_fxsr(fpstate_stub, (void *)®s->fp))
+ return -EINVAL;
+#else
+ /*
+ * On x86_64, we need to sync the FS_BASE/GS_BASE registers using the
+ * arch specific data.
+ */
+ if (data->arch_data.fs_base != regs->gp[FS_BASE / sizeof(unsigned long)]) {
+ data->arch_data.fs_base = regs->gp[FS_BASE / sizeof(unsigned long)];
+ data->arch_data.sync |= STUB_SYNC_FS_BASE;
+ }
+ if (data->arch_data.gs_base != regs->gp[GS_BASE / sizeof(unsigned long)]) {
+ data->arch_data.gs_base = regs->gp[GS_BASE / sizeof(unsigned long)];
+ data->arch_data.sync |= STUB_SYNC_GS_BASE;
+ }
+#endif
+
+ return 0;
+}
diff --git a/arch/x86/um/ptrace.c b/arch/x86/um/ptrace.c
index 57c504fd5626..3275870330fe 100644
--- a/arch/x86/um/ptrace.c
+++ b/arch/x86/um/ptrace.c
@@ -25,7 +25,8 @@ static inline unsigned short twd_i387_to_fxsr(unsigned short twd)
return tmp;
}
-static inline unsigned long twd_fxsr_to_i387(struct user_fxsr_struct *fxsave)
+static inline unsigned long
+twd_fxsr_to_i387(const struct user_fxsr_struct *fxsave)
{
struct _fpxreg *st = NULL;
unsigned long twd = (unsigned long) fxsave->twd;
@@ -69,12 +70,16 @@ static inline unsigned long twd_fxsr_to_i387(struct user_fxsr_struct *fxsave)
return ret;
}
-/* Get/set the old 32bit i387 registers (pre-FPX) */
-static int fpregs_legacy_get(struct task_struct *target,
- const struct user_regset *regset,
- struct membuf to)
+/*
+ * Get/set the old 32bit i387 registers (pre-FPX)
+ *
+ * We provide simple wrappers for mcontext.c, they are only defined locally
+ * because mcontext.c is userspace facing and needs to a different definition
+ * of the structures.
+ */
+static int _um_i387_from_fxsr(struct membuf to,
+ const struct user_fxsr_struct *fxsave)
{
- struct user_fxsr_struct *fxsave = (void *)target->thread.regs.regs.fp;
int i;
membuf_store(&to, (unsigned long)fxsave->cwd | 0xffff0000ul);
@@ -91,23 +96,36 @@ static int fpregs_legacy_get(struct task_struct *target,
return 0;
}
-static int fpregs_legacy_set(struct task_struct *target,
+int um_i387_from_fxsr(struct user_i387_struct *i387,
+ const struct user_fxsr_struct *fxsave);
+
+int um_i387_from_fxsr(struct user_i387_struct *i387,
+ const struct user_fxsr_struct *fxsave)
+{
+ struct membuf to = {
+ .p = i387,
+ .left = sizeof(*i387),
+ };
+
+ return _um_i387_from_fxsr(to, fxsave);
+}
+
+static int fpregs_legacy_get(struct task_struct *target,
const struct user_regset *regset,
- unsigned int pos, unsigned int count,
- const void *kbuf, const void __user *ubuf)
+ struct membuf to)
{
struct user_fxsr_struct *fxsave = (void *)target->thread.regs.regs.fp;
- const struct user_i387_struct *from;
- struct user_i387_struct buf;
- int i;
- if (ubuf) {
- if (copy_from_user(&buf, ubuf, sizeof(buf)))
- return -EFAULT;
- from = &buf;
- } else {
- from = kbuf;
- }
+ return _um_i387_from_fxsr(to, fxsave);
+}
+
+int um_fxsr_from_i387(struct user_fxsr_struct *fxsave,
+ const struct user_i387_struct *from);
+
+int um_fxsr_from_i387(struct user_fxsr_struct *fxsave,
+ const struct user_i387_struct *from)
+{
+ int i;
fxsave->cwd = (unsigned short)(from->cwd & 0xffff);
fxsave->swd = (unsigned short)(from->swd & 0xffff);
@@ -125,6 +143,26 @@ static int fpregs_legacy_set(struct task_struct *target,
return 0;
}
+
+static int fpregs_legacy_set(struct task_struct *target,
+ const struct user_regset *regset,
+ unsigned int pos, unsigned int count,
+ const void *kbuf, const void __user *ubuf)
+{
+ struct user_fxsr_struct *fxsave = (void *)target->thread.regs.regs.fp;
+ const struct user_i387_struct *from;
+ struct user_i387_struct buf;
+
+ if (ubuf) {
+ if (copy_from_user(&buf, ubuf, sizeof(buf)))
+ return -EFAULT;
+ from = &buf;
+ } else {
+ from = kbuf;
+ }
+
+ return um_fxsr_from_i387(fxsave, &buf);
+}
#endif
static int genregs_get(struct task_struct *target,
diff --git a/arch/x86/um/shared/sysdep/mcontext.h b/arch/x86/um/shared/sysdep/mcontext.h
index b724c54da316..6fe490cc5b98 100644
--- a/arch/x86/um/shared/sysdep/mcontext.h
+++ b/arch/x86/um/shared/sysdep/mcontext.h
@@ -6,7 +6,16 @@
#ifndef __SYS_SIGCONTEXT_X86_H
#define __SYS_SIGCONTEXT_X86_H
+#include <stub-data.h>
+
extern void get_regs_from_mc(struct uml_pt_regs *, mcontext_t *);
+extern void get_mc_from_regs(struct uml_pt_regs *regs, mcontext_t *mc,
+ int single_stepping);
+
+extern int get_stub_state(struct uml_pt_regs *regs, struct stub_data *data,
+ unsigned long *fp_size_out);
+extern int set_stub_state(struct uml_pt_regs *regs, struct stub_data *data,
+ int single_stepping);
#ifdef __i386__
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH v3 4/7] um: Track userspace children dying in SECCOMP mode
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
` (2 preceding siblings ...)
2025-06-02 13:00 ` [PATCH v3 3/7] um: Add helper functions to get/set state for SECCOMP Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 5/7] um: Implement kernel side of SECCOMP based process handling Benjamin Berg
` (2 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg, Benjamin Berg
When in seccomp mode, we would hang forever on the futex if a child has
died unexpectedly. In contrast, ptrace mode will notice it and kill the
corresponding thread when it fails to run it.
Fix this issue using a new IRQ that is fired after a SIGCHLD and keeping
an (internal) list of all MMs. In the IRQ handler, find the affected MM
and set its PID to -1 as well as the futex variable to FUTEX_IN_KERN.
This, together with futex returning -EINTR after the signal is
sufficient to implement a race-free detection of a child dying.
Note that this also enables IRQ handling while starting a userspace
process. This should be safe and SECCOMP requires the IRQ in case the
process does not come up properly.
Signed-off-by: Benjamin Berg <benjamin@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
v3:
- Move using_seccomp definition into this patch
v2:
- Fix whitespace issue
v1:
- Permit IRQs during startup to enable detection there
RFCv2:
- Use "struct list_head" for the list by placing it into the mm_context
---
arch/um/include/asm/irq.h | 5 +-
arch/um/include/asm/mmu.h | 3 ++
arch/um/include/shared/irq_user.h | 2 +
arch/um/include/shared/os.h | 1 +
arch/um/include/shared/skas/mm_id.h | 2 +
arch/um/include/shared/skas/skas.h | 1 +
arch/um/kernel/irq.c | 6 +++
arch/um/kernel/skas/mmu.c | 82 +++++++++++++++++++++++++++--
arch/um/os-Linux/process.c | 31 +++++++++++
arch/um/os-Linux/signal.c | 19 ++++++-
arch/um/os-Linux/skas/process.c | 1 +
11 files changed, 145 insertions(+), 8 deletions(-)
diff --git a/arch/um/include/asm/irq.h b/arch/um/include/asm/irq.h
index 749dfe8512e8..36dbedd1af48 100644
--- a/arch/um/include/asm/irq.h
+++ b/arch/um/include/asm/irq.h
@@ -13,17 +13,18 @@
#define TELNETD_IRQ 8
#define XTERM_IRQ 9
#define RANDOM_IRQ 10
+#define SIGCHLD_IRQ 11
#ifdef CONFIG_UML_NET_VECTOR
-#define VECTOR_BASE_IRQ (RANDOM_IRQ + 1)
+#define VECTOR_BASE_IRQ (SIGCHLD_IRQ + 1)
#define VECTOR_IRQ_SPACE 8
#define UM_FIRST_DYN_IRQ (VECTOR_IRQ_SPACE + VECTOR_BASE_IRQ)
#else
-#define UM_FIRST_DYN_IRQ (RANDOM_IRQ + 1)
+#define UM_FIRST_DYN_IRQ (SIGCHLD_IRQ + 1)
#endif
diff --git a/arch/um/include/asm/mmu.h b/arch/um/include/asm/mmu.h
index a3eaca41ff61..4d0e4239f3cc 100644
--- a/arch/um/include/asm/mmu.h
+++ b/arch/um/include/asm/mmu.h
@@ -6,11 +6,14 @@
#ifndef __ARCH_UM_MMU_H
#define __ARCH_UM_MMU_H
+#include "linux/types.h"
#include <mm_id.h>
typedef struct mm_context {
struct mm_id id;
+ struct list_head list;
+
/* Address range in need of a TLB sync */
unsigned long sync_tlb_range_from;
unsigned long sync_tlb_range_to;
diff --git a/arch/um/include/shared/irq_user.h b/arch/um/include/shared/irq_user.h
index 88835b52ae2b..746abc24a5d5 100644
--- a/arch/um/include/shared/irq_user.h
+++ b/arch/um/include/shared/irq_user.h
@@ -17,6 +17,8 @@ enum um_irq_type {
struct siginfo;
extern void sigio_handler(int sig, struct siginfo *unused_si,
struct uml_pt_regs *regs, void *mc);
+extern void sigchld_handler(int sig, struct siginfo *unused_si,
+ struct uml_pt_regs *regs, void *mc);
void sigio_run_timetravel_handlers(void);
extern void free_irq_by_fd(int fd);
extern void deactivate_fd(int fd, int irqnum);
diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h
index e63a74a5ff19..3046728ec42e 100644
--- a/arch/um/include/shared/os.h
+++ b/arch/um/include/shared/os.h
@@ -197,6 +197,7 @@ extern int create_mem_file(unsigned long long len);
extern void report_enomem(void);
/* process.c */
+pid_t os_reap_child(void);
extern void os_alarm_process(int pid);
extern void os_kill_process(int pid, int reap_child);
extern void os_kill_ptraced_process(int pid, int reap_child);
diff --git a/arch/um/include/shared/skas/mm_id.h b/arch/um/include/shared/skas/mm_id.h
index 140388c282f6..0654c57bb28e 100644
--- a/arch/um/include/shared/skas/mm_id.h
+++ b/arch/um/include/shared/skas/mm_id.h
@@ -14,4 +14,6 @@ struct mm_id {
void __switch_mm(struct mm_id *mm_idp);
+void notify_mm_kill(int pid);
+
#endif
diff --git a/arch/um/include/shared/skas/skas.h b/arch/um/include/shared/skas/skas.h
index 85c50122ab98..7d1de4cab551 100644
--- a/arch/um/include/shared/skas/skas.h
+++ b/arch/um/include/shared/skas/skas.h
@@ -8,6 +8,7 @@
#include <sysdep/ptrace.h>
+extern int using_seccomp;
extern int userspace_pid[];
extern void new_thread_handler(void);
diff --git a/arch/um/kernel/irq.c b/arch/um/kernel/irq.c
index abe8f30a521c..f1787be3983c 100644
--- a/arch/um/kernel/irq.c
+++ b/arch/um/kernel/irq.c
@@ -690,3 +690,9 @@ void __init init_IRQ(void)
/* Initialize EPOLL Loop */
os_setup_epoll();
}
+
+extern void sigchld_handler(int sig, struct siginfo *unused_si,
+ struct uml_pt_regs *regs, void *mc)
+{
+ do_IRQ(SIGCHLD_IRQ, regs);
+}
diff --git a/arch/um/kernel/skas/mmu.c b/arch/um/kernel/skas/mmu.c
index 0eb5a1d3ba70..1e146a0f9549 100644
--- a/arch/um/kernel/skas/mmu.c
+++ b/arch/um/kernel/skas/mmu.c
@@ -8,6 +8,7 @@
#include <linux/sched/signal.h>
#include <linux/slab.h>
+#include <shared/irq_kern.h>
#include <asm/pgalloc.h>
#include <asm/sections.h>
#include <asm/mmu_context.h>
@@ -19,6 +20,9 @@
/* Ensure the stub_data struct covers the allocated area */
static_assert(sizeof(struct stub_data) == STUB_DATA_PAGES * UM_KERN_PAGE_SIZE);
+spinlock_t mm_list_lock;
+struct list_head mm_list;
+
int init_new_context(struct task_struct *task, struct mm_struct *mm)
{
struct mm_id *new_id = &mm->context.id;
@@ -31,10 +35,12 @@ int init_new_context(struct task_struct *task, struct mm_struct *mm)
new_id->stack = stack;
- block_signals_trace();
- new_id->pid = start_userspace(stack);
- unblock_signals_trace();
+ scoped_guard(spinlock_irqsave, &mm_list_lock) {
+ /* Insert into list, used for lookups when the child dies */
+ list_add(&mm->context.list, &mm_list);
+ }
+ new_id->pid = start_userspace(stack);
if (new_id->pid < 0) {
ret = new_id->pid;
goto out_free;
@@ -60,13 +66,79 @@ void destroy_context(struct mm_struct *mm)
* zero, resulting in a kill(0), which will result in the
* whole UML suddenly dying. Also, cover negative and
* 1 cases, since they shouldn't happen either.
+ *
+ * Negative cases happen if the child died unexpectedly.
*/
- if (mmu->id.pid < 2) {
+ if (mmu->id.pid >= 0 && mmu->id.pid < 2) {
printk(KERN_ERR "corrupt mm_context - pid = %d\n",
mmu->id.pid);
return;
}
- os_kill_ptraced_process(mmu->id.pid, 1);
+
+ if (mmu->id.pid > 0) {
+ os_kill_ptraced_process(mmu->id.pid, 1);
+ mmu->id.pid = -1;
+ }
free_pages(mmu->id.stack, ilog2(STUB_DATA_PAGES));
+
+ guard(spinlock_irqsave)(&mm_list_lock);
+
+ list_del(&mm->context.list);
+}
+
+static irqreturn_t mm_sigchld_irq(int irq, void* dev)
+{
+ struct mm_context *mm_context;
+ pid_t pid;
+
+ guard(spinlock)(&mm_list_lock);
+
+ while ((pid = os_reap_child()) > 0) {
+ /*
+ * A child died, check if we have an MM with the PID. This is
+ * only relevant in SECCOMP mode (as ptrace will fail anyway).
+ *
+ * See wait_stub_done_seccomp for more details.
+ */
+ list_for_each_entry(mm_context, &mm_list, list) {
+ if (mm_context->id.pid == pid) {
+ struct stub_data *stub_data;
+ printk("Unexpectedly lost MM child! Affected tasks will segfault.");
+
+ /* Marks the MM as dead */
+ mm_context->id.pid = -1;
+
+ /*
+ * NOTE: If SMP is implemented, a futex_wake
+ * needs to be added here.
+ */
+ stub_data = (void *)mm_context->id.stack;
+ stub_data->futex = FUTEX_IN_KERN;
+
+ /*
+ * NOTE: Currently executing syscalls by
+ * affected tasks may finish normally.
+ */
+ break;
+ }
+ }
+ }
+
+ return IRQ_HANDLED;
+}
+
+static int __init init_child_tracking(void)
+{
+ int err;
+
+ spin_lock_init(&mm_list_lock);
+ INIT_LIST_HEAD(&mm_list);
+
+ err = request_irq(SIGCHLD_IRQ, mm_sigchld_irq, 0, "SIGCHLD", NULL);
+ if (err < 0)
+ panic("Failed to register SIGCHLD IRQ: %d", err);
+
+ return 0;
}
+early_initcall(init_child_tracking)
diff --git a/arch/um/os-Linux/process.c b/arch/um/os-Linux/process.c
index 184566edeee9..00b49e90d05f 100644
--- a/arch/um/os-Linux/process.c
+++ b/arch/um/os-Linux/process.c
@@ -18,17 +18,29 @@
#include <init.h>
#include <longjmp.h>
#include <os.h>
+#include <skas/skas.h>
void os_alarm_process(int pid)
{
+ if (pid <= 0)
+ return;
+
kill(pid, SIGALRM);
}
void os_kill_process(int pid, int reap_child)
{
+ if (pid <= 0)
+ return;
+
+ /* Block signals until child is reaped */
+ block_signals();
+
kill(pid, SIGKILL);
if (reap_child)
CATCH_EINTR(waitpid(pid, NULL, __WALL));
+
+ unblock_signals();
}
/* Kill off a ptraced child by all means available. kill it normally first,
@@ -38,11 +50,27 @@ void os_kill_process(int pid, int reap_child)
void os_kill_ptraced_process(int pid, int reap_child)
{
+ if (pid <= 0)
+ return;
+
+ /* Block signals until child is reaped */
+ block_signals();
+
kill(pid, SIGKILL);
ptrace(PTRACE_KILL, pid);
ptrace(PTRACE_CONT, pid);
if (reap_child)
CATCH_EINTR(waitpid(pid, NULL, __WALL));
+
+ unblock_signals();
+}
+
+pid_t os_reap_child(void)
+{
+ int status;
+
+ /* Try to reap a child */
+ return waitpid(-1, &status, WNOHANG);
}
/* Don't use the glibc version, which caches the result in TLS. It misses some
@@ -151,6 +179,9 @@ void init_new_thread_signals(void)
set_handler(SIGBUS);
signal(SIGHUP, SIG_IGN);
set_handler(SIGIO);
+ /* We (currently) only use the child reaper IRQ in seccomp mode */
+ if (using_seccomp)
+ set_handler(SIGCHLD);
signal(SIGWINCH, SIG_IGN);
}
diff --git a/arch/um/os-Linux/signal.c b/arch/um/os-Linux/signal.c
index e71e5b4878d1..11f07f498270 100644
--- a/arch/um/os-Linux/signal.c
+++ b/arch/um/os-Linux/signal.c
@@ -29,6 +29,7 @@ void (*sig_info[NSIG])(int, struct siginfo *, struct uml_pt_regs *, void *mc) =
[SIGBUS] = relay_signal,
[SIGSEGV] = segv_handler,
[SIGIO] = sigio_handler,
+ [SIGCHLD] = sigchld_handler,
};
static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
@@ -44,7 +45,7 @@ static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
}
/* enable signals if sig isn't IRQ signal */
- if ((sig != SIGIO) && (sig != SIGWINCH))
+ if ((sig != SIGIO) && (sig != SIGWINCH) && (sig != SIGCHLD))
unblock_signals_trace();
(*sig_info[sig])(sig, si, &r, mc);
@@ -64,6 +65,9 @@ static void sig_handler_common(int sig, struct siginfo *si, mcontext_t *mc)
#define SIGALRM_BIT 1
#define SIGALRM_MASK (1 << SIGALRM_BIT)
+#define SIGCHLD_BIT 2
+#define SIGCHLD_MASK (1 << SIGCHLD_BIT)
+
int signals_enabled;
#if IS_ENABLED(CONFIG_UML_TIME_TRAVEL_SUPPORT)
static int signals_blocked, signals_blocked_pending;
@@ -102,6 +106,11 @@ static void sig_handler(int sig, struct siginfo *si, mcontext_t *mc)
return;
}
+ if (!enabled && (sig == SIGCHLD)) {
+ signals_pending |= SIGCHLD_MASK;
+ return;
+ }
+
block_signals_trace();
sig_handler_common(sig, si, mc);
@@ -181,6 +190,8 @@ static void (*handlers[_NSIG])(int sig, struct siginfo *si, mcontext_t *mc) = {
[SIGIO] = sig_handler,
[SIGWINCH] = sig_handler,
+ /* SIGCHLD is only actually registered in seccomp mode. */
+ [SIGCHLD] = sig_handler,
[SIGALRM] = timer_alarm_handler,
[SIGUSR1] = sigusr1_handler,
@@ -309,6 +320,12 @@ void unblock_signals(void)
if (save_pending & SIGIO_MASK)
sig_handler_common(SIGIO, NULL, NULL);
+ if (save_pending & SIGCHLD_MASK) {
+ struct uml_pt_regs regs = {};
+
+ sigchld_handler(SIGCHLD, NULL, ®s, NULL);
+ }
+
/* Do not reenter the handler */
if ((save_pending & SIGALRM_MASK) && (!(signals_active & SIGALRM_MASK)))
diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
index 97c2f964f5fc..150e4f6ba633 100644
--- a/arch/um/os-Linux/skas/process.c
+++ b/arch/um/os-Linux/skas/process.c
@@ -309,6 +309,7 @@ static int __init init_stub_exe_fd(void)
}
__initcall(init_stub_exe_fd);
+int using_seccomp;
int userspace_pid[NR_CPUS];
/**
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH v3 5/7] um: Implement kernel side of SECCOMP based process handling
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
` (3 preceding siblings ...)
2025-06-02 13:00 ` [PATCH v3 4/7] um: Track userspace children dying in SECCOMP mode Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 6/7] um: Add SECCOMP support detection and initialization Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 7/7] um: pass FD for memory operations when needed Benjamin Berg
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg, Johannes Berg, Benjamin Berg
This adds the kernel side of the seccomp based process handling.
Co-authored-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
v1:
- Fix FUTEX_WAIT EINTR handling
- Don't send fatal_sigsegv when waiting during child startup
---
arch/um/include/shared/common-offsets.h | 2 +
arch/um/include/shared/os.h | 2 +-
arch/um/include/shared/skas/stub-data.h | 5 +-
arch/um/kernel/skas/mmu.c | 6 +-
arch/um/kernel/skas/stub_exe.c | 141 +++++++-
arch/um/os-Linux/internal.h | 5 +-
arch/um/os-Linux/skas/mem.c | 37 +-
arch/um/os-Linux/skas/process.c | 374 +++++++++++++++------
arch/x86/um/shared/sysdep/kernel-offsets.h | 2 +
arch/x86/um/tls_32.c | 23 +-
10 files changed, 459 insertions(+), 138 deletions(-)
diff --git a/arch/um/include/shared/common-offsets.h b/arch/um/include/shared/common-offsets.h
index 93e7097a2922..8ca66a1918c3 100644
--- a/arch/um/include/shared/common-offsets.h
+++ b/arch/um/include/shared/common-offsets.h
@@ -16,3 +16,5 @@ DEFINE(UM_NSEC_PER_SEC, NSEC_PER_SEC);
DEFINE(UM_NSEC_PER_USEC, NSEC_PER_USEC);
DEFINE(UM_KERN_GDT_ENTRY_TLS_ENTRIES, GDT_ENTRY_TLS_ENTRIES);
+
+DEFINE(UM_SECCOMP_ARCH_NATIVE, SECCOMP_ARCH_NATIVE);
diff --git a/arch/um/include/shared/os.h b/arch/um/include/shared/os.h
index 3046728ec42e..b35cc8ce333b 100644
--- a/arch/um/include/shared/os.h
+++ b/arch/um/include/shared/os.h
@@ -286,7 +286,7 @@ int unmap(struct mm_id *mm_idp, unsigned long addr, unsigned long len);
/* skas/process.c */
extern int is_skas_winch(int pid, int fd, void *data);
-extern int start_userspace(unsigned long stub_stack);
+extern int start_userspace(struct mm_id *mm_id);
extern void userspace(struct uml_pt_regs *regs);
extern void new_thread(void *stack, jmp_buf *buf, void (*handler)(void));
extern void switch_threads(jmp_buf *me, jmp_buf *you);
diff --git a/arch/um/include/shared/skas/stub-data.h b/arch/um/include/shared/skas/stub-data.h
index 81ac2cd12112..675f1a0a1390 100644
--- a/arch/um/include/shared/skas/stub-data.h
+++ b/arch/um/include/shared/skas/stub-data.h
@@ -17,6 +17,8 @@
#define FUTEX_IN_KERN 1
struct stub_init_data {
+ int seccomp;
+
unsigned long stub_start;
int stub_code_fd;
@@ -24,7 +26,8 @@ struct stub_init_data {
int stub_data_fd;
unsigned long stub_data_offset;
- unsigned long segv_handler;
+ unsigned long signal_handler;
+ unsigned long signal_restorer;
};
#define STUB_NEXT_SYSCALL(s) \
diff --git a/arch/um/kernel/skas/mmu.c b/arch/um/kernel/skas/mmu.c
index 1e146a0f9549..87a18ae4da19 100644
--- a/arch/um/kernel/skas/mmu.c
+++ b/arch/um/kernel/skas/mmu.c
@@ -40,11 +40,9 @@ int init_new_context(struct task_struct *task, struct mm_struct *mm)
list_add(&mm->context.list, &mm_list);
}
- new_id->pid = start_userspace(stack);
- if (new_id->pid < 0) {
- ret = new_id->pid;
+ ret = start_userspace(new_id);
+ if (ret < 0)
goto out_free;
- }
/* Ensure the new MM is clean and nothing unwanted is mapped */
unmap(new_id, 0, STUB_START);
diff --git a/arch/um/kernel/skas/stub_exe.c b/arch/um/kernel/skas/stub_exe.c
index 23c99b285e82..f40f2332b676 100644
--- a/arch/um/kernel/skas/stub_exe.c
+++ b/arch/um/kernel/skas/stub_exe.c
@@ -3,6 +3,9 @@
#include <asm/unistd.h>
#include <sysdep/stub.h>
#include <stub-data.h>
+#include <linux/filter.h>
+#include <linux/seccomp.h>
+#include <generated/asm-offsets.h>
void _start(void);
@@ -25,8 +28,6 @@ noinline static void real_init(void)
} sa = {
/* Need to set SA_RESTORER (but the handler never returns) */
.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO | 0x04000000,
- /* no need to mask any signals */
- .sa_mask = 0,
};
/* set a nice name */
@@ -35,6 +36,9 @@ noinline static void real_init(void)
/* Make sure this process dies if the kernel dies */
stub_syscall2(__NR_prctl, PR_SET_PDEATHSIG, SIGKILL);
+ /* Needed in SECCOMP mode (and safe to do anyway) */
+ stub_syscall5(__NR_prctl, PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
+
/* read information from STDIN and close it */
res = stub_syscall3(__NR_read, 0,
(unsigned long)&init_data, sizeof(init_data));
@@ -63,18 +67,133 @@ noinline static void real_init(void)
stack.ss_sp = (void *)init_data.stub_start + UM_KERN_PAGE_SIZE;
stub_syscall2(__NR_sigaltstack, (unsigned long)&stack, 0);
- /* register SIGSEGV handler */
- sa.sa_handler_ = (void *) init_data.segv_handler;
- res = stub_syscall4(__NR_rt_sigaction, SIGSEGV, (unsigned long)&sa, 0,
- sizeof(sa.sa_mask));
- if (res != 0)
- stub_syscall1(__NR_exit, 13);
+ /* register signal handlers */
+ sa.sa_handler_ = (void *) init_data.signal_handler;
+ sa.sa_restorer = (void *) init_data.signal_restorer;
+ if (!init_data.seccomp) {
+ /* In ptrace mode, the SIGSEGV handler never returns */
+ sa.sa_mask = 0;
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGSEGV,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 13);
+ } else {
+ /* SECCOMP mode uses rt_sigreturn, need to mask all signals */
+ sa.sa_mask = ~0ULL;
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGSEGV,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 14);
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGSYS,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 15);
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGALRM,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 16);
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGTRAP,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 17);
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGILL,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 18);
+
+ res = stub_syscall4(__NR_rt_sigaction, SIGFPE,
+ (unsigned long)&sa, 0, sizeof(sa.sa_mask));
+ if (res != 0)
+ stub_syscall1(__NR_exit, 19);
+ }
+
+ /*
+ * If in seccomp mode, install the SECCOMP filter and trigger a syscall.
+ * Otherwise set PTRACE_TRACEME and do a SIGSTOP.
+ */
+ if (init_data.seccomp) {
+ struct sock_filter filter[] = {
+#if __BITS_PER_LONG > 32
+ /* [0] Load upper 32bit of instruction pointer from seccomp_data */
+ BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
+ (offsetof(struct seccomp_data, instruction_pointer) + 4)),
+
+ /* [1] Jump forward 3 instructions if the upper address is not identical */
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, (init_data.stub_start) >> 32, 0, 3),
+#endif
+ /* [2] Load lower 32bit of instruction pointer from seccomp_data */
+ BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
+ (offsetof(struct seccomp_data, instruction_pointer))),
+
+ /* [3] Mask out lower bits */
+ BPF_STMT(BPF_ALU | BPF_AND | BPF_K, 0xfffff000),
+
+ /* [4] Jump to [6] if the lower bits are not on the expected page */
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, (init_data.stub_start) & 0xfffff000, 1, 0),
+
+ /* [5] Trap call, allow */
+ BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_TRAP),
+
+ /* [6,7] Check architecture */
+ BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
+ offsetof(struct seccomp_data, arch)),
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K,
+ UM_SECCOMP_ARCH_NATIVE, 1, 0),
+
+ /* [8] Kill (for architecture check) */
+ BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS),
+
+ /* [9] Load syscall number */
+ BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
+ offsetof(struct seccomp_data, nr)),
+
+ /* [10-14] Check against permitted syscalls */
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_futex,
+ 5, 0),
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, STUB_MMAP_NR,
+ 4, 0),
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_munmap,
+ 3, 0),
+#ifdef __i386__
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_set_thread_area,
+ 2, 0),
+#else
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_arch_prctl,
+ 2, 0),
+#endif
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_rt_sigreturn,
+ 1, 0),
+
+ /* [15] Not one of the permitted syscalls */
+ BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS),
+
+ /* [16] Permitted call for the stub */
+ BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
+ };
+ struct sock_fprog prog = {
+ .len = sizeof(filter) / sizeof(filter[0]),
+ .filter = filter,
+ };
+
+ if (stub_syscall3(__NR_seccomp, SECCOMP_SET_MODE_FILTER,
+ SECCOMP_FILTER_FLAG_TSYNC,
+ (unsigned long)&prog) != 0)
+ stub_syscall1(__NR_exit, 20);
- stub_syscall4(__NR_ptrace, PTRACE_TRACEME, 0, 0, 0);
+ /* Fall through, the exit syscall will cause SIGSYS */
+ } else {
+ stub_syscall4(__NR_ptrace, PTRACE_TRACEME, 0, 0, 0);
- stub_syscall2(__NR_kill, stub_syscall0(__NR_getpid), SIGSTOP);
+ stub_syscall2(__NR_kill, stub_syscall0(__NR_getpid), SIGSTOP);
+ }
- stub_syscall1(__NR_exit, 14);
+ stub_syscall1(__NR_exit, 30);
__builtin_unreachable();
}
diff --git a/arch/um/os-Linux/internal.h b/arch/um/os-Linux/internal.h
index 317fca190c2b..5d8d3b0817a9 100644
--- a/arch/um/os-Linux/internal.h
+++ b/arch/um/os-Linux/internal.h
@@ -2,6 +2,9 @@
#ifndef __UM_OS_LINUX_INTERNAL_H
#define __UM_OS_LINUX_INTERNAL_H
+#include <mm_id.h>
+#include <stub-data.h>
+
/*
* elf_aux.c
*/
@@ -16,5 +19,5 @@ void check_tmpexec(void);
* skas/process.c
*/
void wait_stub_done(int pid);
-
+void wait_stub_done_seccomp(struct mm_id *mm_idp, int running, int wait_sigsys);
#endif /* __UM_OS_LINUX_INTERNAL_H */
diff --git a/arch/um/os-Linux/skas/mem.c b/arch/um/os-Linux/skas/mem.c
index d7f1814b0e5a..31bf3a52047a 100644
--- a/arch/um/os-Linux/skas/mem.c
+++ b/arch/um/os-Linux/skas/mem.c
@@ -80,27 +80,32 @@ static inline long do_syscall_stub(struct mm_id *mm_idp)
int n, i;
int err, pid = mm_idp->pid;
- n = ptrace_setregs(pid, syscall_regs);
- if (n < 0) {
- printk(UM_KERN_ERR "Registers - \n");
- for (i = 0; i < MAX_REG_NR; i++)
- printk(UM_KERN_ERR "\t%d\t0x%lx\n", i, syscall_regs[i]);
- panic("%s : PTRACE_SETREGS failed, errno = %d\n",
- __func__, -n);
- }
-
/* Inform process how much we have filled in. */
proc_data->syscall_data_len = mm_idp->syscall_data_len;
- err = ptrace(PTRACE_CONT, pid, 0, 0);
- if (err)
- panic("Failed to continue stub, pid = %d, errno = %d\n", pid,
- errno);
-
- wait_stub_done(pid);
+ if (using_seccomp) {
+ proc_data->restart_wait = 1;
+ wait_stub_done_seccomp(mm_idp, 0, 1);
+ } else {
+ n = ptrace_setregs(pid, syscall_regs);
+ if (n < 0) {
+ printk(UM_KERN_ERR "Registers -\n");
+ for (i = 0; i < MAX_REG_NR; i++)
+ printk(UM_KERN_ERR "\t%d\t0x%lx\n", i, syscall_regs[i]);
+ panic("%s : PTRACE_SETREGS failed, errno = %d\n",
+ __func__, -n);
+ }
+
+ err = ptrace(PTRACE_CONT, pid, 0, 0);
+ if (err)
+ panic("Failed to continue stub, pid = %d, errno = %d\n",
+ pid, errno);
+
+ wait_stub_done(pid);
+ }
/*
- * proc_data->err will be non-zero if there was an (unexpected) error.
+ * proc_data->err will be negative if there was an (unexpected) error.
* In that case, syscall_data_len points to the last executed syscall,
* otherwise it will be zero (but we do not need to rely on that).
*/
diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
index 150e4f6ba633..e1aaa144e273 100644
--- a/arch/um/os-Linux/skas/process.c
+++ b/arch/um/os-Linux/skas/process.c
@@ -1,5 +1,6 @@
// SPDX-License-Identifier: GPL-2.0
/*
+ * Copyright (C) 2021 Benjamin Berg <benjamin@sipsolutions.net>
* Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
* Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
*/
@@ -25,8 +26,11 @@
#include <registers.h>
#include <skas.h>
#include <sysdep/stub.h>
+#include <sysdep/mcontext.h>
+#include <linux/futex.h>
#include <linux/threads.h>
#include <timetravel.h>
+#include <asm-generic/rwonce.h>
#include "../internal.h"
int is_skas_winch(int pid, int fd, void *data)
@@ -142,6 +146,73 @@ void wait_stub_done(int pid)
fatal_sigsegv();
}
+void wait_stub_done_seccomp(struct mm_id *mm_idp, int running, int wait_sigsys)
+{
+ struct stub_data *data = (void *)mm_idp->stack;
+ int ret;
+
+ do {
+ if (!running) {
+ data->signal = 0;
+ data->futex = FUTEX_IN_CHILD;
+ CATCH_EINTR(syscall(__NR_futex, &data->futex,
+ FUTEX_WAKE, 1, NULL, NULL, 0));
+ }
+
+ do {
+ /*
+ * We need to check whether the child is still alive
+ * before and after the FUTEX_WAIT call. Before, in
+ * case it just died but we still updated data->futex
+ * to FUTEX_IN_CHILD. And after, in case it died while
+ * we were waiting (and SIGCHLD woke us up, see the
+ * IRQ handler in mmu.c).
+ *
+ * Either way, if PID is negative, then we have no
+ * choice but to kill the task.
+ */
+ if (__READ_ONCE(mm_idp->pid) < 0)
+ goto out_kill;
+
+ ret = syscall(__NR_futex, &data->futex,
+ FUTEX_WAIT, FUTEX_IN_CHILD,
+ NULL, NULL, 0);
+ if (ret < 0 && errno != EINTR && errno != EAGAIN) {
+ printk(UM_KERN_ERR "%s : FUTEX_WAIT failed, errno = %d\n",
+ __func__, errno);
+ goto out_kill;
+ }
+ } while (data->futex == FUTEX_IN_CHILD);
+
+ if (__READ_ONCE(mm_idp->pid) < 0)
+ goto out_kill;
+
+ running = 0;
+
+ /* We may receive a SIGALRM before SIGSYS, iterate again. */
+ } while (wait_sigsys && data->signal == SIGALRM);
+
+ if (data->mctx_offset > sizeof(data->sigstack) - sizeof(mcontext_t)) {
+ printk(UM_KERN_ERR "%s : invalid mcontext offset", __func__);
+ goto out_kill;
+ }
+
+ if (wait_sigsys && data->signal != SIGSYS) {
+ printk(UM_KERN_ERR "%s : expected SIGSYS but got %d",
+ __func__, data->signal);
+ goto out_kill;
+ }
+
+ return;
+
+out_kill:
+ printk(UM_KERN_ERR "%s : failed to wait for stub, pid = %d, errno = %d\n",
+ __func__, mm_idp->pid, errno);
+ /* This is not true inside start_userspace */
+ if (current_mm_id() == mm_idp)
+ fatal_sigsegv();
+}
+
extern unsigned long current_stub_stack(void);
static void get_skas_faultinfo(int pid, struct faultinfo *fi)
@@ -185,14 +256,26 @@ static int userspace_tramp(void *stack)
int pipe_fds[2];
unsigned long long offset;
struct stub_init_data init_data = {
+ .seccomp = using_seccomp,
.stub_start = STUB_START,
- .segv_handler = STUB_CODE +
- (unsigned long) stub_segv_handler -
- (unsigned long) __syscall_stub_start,
};
struct iomem_region *iomem;
int ret;
+ if (using_seccomp) {
+ init_data.signal_handler = STUB_CODE +
+ (unsigned long) stub_signal_interrupt -
+ (unsigned long) __syscall_stub_start;
+ init_data.signal_restorer = STUB_CODE +
+ (unsigned long) stub_signal_restorer -
+ (unsigned long) __syscall_stub_start;
+ } else {
+ init_data.signal_handler = STUB_CODE +
+ (unsigned long) stub_segv_handler -
+ (unsigned long) __syscall_stub_start;
+ init_data.signal_restorer = 0;
+ }
+
init_data.stub_code_fd = phys_mapping(uml_to_phys(__syscall_stub_start),
&offset);
init_data.stub_code_offset = MMAP_OFFSET(offset);
@@ -323,8 +406,9 @@ int userspace_pid[NR_CPUS];
* when negative: an error number.
* FIXME: can PIDs become negative?!
*/
-int start_userspace(unsigned long stub_stack)
+int start_userspace(struct mm_id *mm_id)
{
+ struct stub_data *proc_data = (void *)mm_id->stack;
void *stack;
unsigned long sp;
int pid, status, n, err;
@@ -343,10 +427,13 @@ int start_userspace(unsigned long stub_stack)
/* set stack pointer to the end of the stack page, so it can grow downwards */
sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
+ if (using_seccomp)
+ proc_data->futex = FUTEX_IN_CHILD;
+
/* clone into new userspace process */
pid = clone(userspace_tramp, (void *) sp,
CLONE_VFORK | CLONE_VM | SIGCHLD,
- (void *)stub_stack);
+ (void *)mm_id->stack);
if (pid < 0) {
err = -errno;
printk(UM_KERN_ERR "%s : clone failed, errno = %d\n",
@@ -354,29 +441,34 @@ int start_userspace(unsigned long stub_stack)
return err;
}
- do {
- CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
- if (n < 0) {
+ if (using_seccomp) {
+ wait_stub_done_seccomp(mm_id, 1, 1);
+ } else {
+ do {
+ CATCH_EINTR(n = waitpid(pid, &status,
+ WUNTRACED | __WALL));
+ if (n < 0) {
+ err = -errno;
+ printk(UM_KERN_ERR "%s : wait failed, errno = %d\n",
+ __func__, errno);
+ goto out_kill;
+ }
+ } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
+
+ if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
+ err = -EINVAL;
+ printk(UM_KERN_ERR "%s : expected SIGSTOP, got status = %d\n",
+ __func__, status);
+ goto out_kill;
+ }
+
+ if (ptrace(PTRACE_SETOPTIONS, pid, NULL,
+ (void *) PTRACE_O_TRACESYSGOOD) < 0) {
err = -errno;
- printk(UM_KERN_ERR "%s : wait failed, errno = %d\n",
+ printk(UM_KERN_ERR "%s : PTRACE_SETOPTIONS failed, errno = %d\n",
__func__, errno);
goto out_kill;
}
- } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
-
- if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
- err = -EINVAL;
- printk(UM_KERN_ERR "%s : expected SIGSTOP, got status = %d\n",
- __func__, status);
- goto out_kill;
- }
-
- if (ptrace(PTRACE_SETOPTIONS, pid, NULL,
- (void *) PTRACE_O_TRACESYSGOOD) < 0) {
- err = -errno;
- printk(UM_KERN_ERR "%s : PTRACE_SETOPTIONS failed, errno = %d\n",
- __func__, errno);
- goto out_kill;
}
if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
@@ -386,6 +478,8 @@ int start_userspace(unsigned long stub_stack)
goto out_kill;
}
+ mm_id->pid = pid;
+
return pid;
out_kill:
@@ -399,7 +493,9 @@ extern unsigned long tt_extra_sched_jiffies;
void userspace(struct uml_pt_regs *regs)
{
int err, status, op, pid = userspace_pid[0];
- siginfo_t si;
+ siginfo_t si_ptrace;
+ siginfo_t *si;
+ int sig;
/* Handle any immediate reschedules or signals */
interrupt_end();
@@ -432,104 +528,182 @@ void userspace(struct uml_pt_regs *regs)
current_mm_sync();
- /* Flush out any pending syscalls */
- err = syscall_stub_flush(current_mm_id());
- if (err) {
- if (err == -ENOMEM)
- report_enomem();
+ if (using_seccomp) {
+ struct mm_id *mm_id = current_mm_id();
+ struct stub_data *proc_data = (void *) mm_id->stack;
+ int ret;
- printk(UM_KERN_ERR "%s - Error flushing stub syscalls: %d",
- __func__, -err);
- fatal_sigsegv();
- }
+ ret = set_stub_state(regs, proc_data, singlestepping());
+ if (ret) {
+ printk(UM_KERN_ERR "%s - failed to set regs: %d",
+ __func__, ret);
+ fatal_sigsegv();
+ }
- /*
- * This can legitimately fail if the process loads a
- * bogus value into a segment register. It will
- * segfault and PTRACE_GETREGS will read that value
- * out of the process. However, PTRACE_SETREGS will
- * fail. In this case, there is nothing to do but
- * just kill the process.
- */
- if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
- printk(UM_KERN_ERR "%s - ptrace set regs failed, errno = %d\n",
- __func__, errno);
- fatal_sigsegv();
- }
+ /* Must have been reset by the syscall caller */
+ if (proc_data->restart_wait != 0)
+ panic("Programming error: Flag to only run syscalls in child was not cleared!");
+
+ /* Mark pending syscalls for flushing */
+ proc_data->syscall_data_len = mm_id->syscall_data_len;
+ mm_id->syscall_data_len = 0;
+
+ proc_data->signal = 0;
+ proc_data->futex = FUTEX_IN_CHILD;
+ CATCH_EINTR(syscall(__NR_futex, &proc_data->futex,
+ FUTEX_WAKE, 1, NULL, NULL, 0));
+ do {
+ ret = syscall(__NR_futex, &proc_data->futex,
+ FUTEX_WAIT, FUTEX_IN_CHILD, NULL, NULL, 0);
+ } while ((ret == -1 && errno == EINTR) ||
+ proc_data->futex == FUTEX_IN_CHILD);
+
+ sig = proc_data->signal;
+
+ if (sig == SIGTRAP && proc_data->err != 0) {
+ printk(UM_KERN_ERR "%s - Error flushing stub syscalls",
+ __func__);
+ syscall_stub_dump_error(mm_id);
+ fatal_sigsegv();
+ }
- if (put_fp_registers(pid, regs->fp)) {
- printk(UM_KERN_ERR "%s - ptrace set fp regs failed, errno = %d\n",
- __func__, errno);
- fatal_sigsegv();
- }
+ ret = get_stub_state(regs, proc_data, NULL);
+ if (ret) {
+ printk(UM_KERN_ERR "%s - failed to get regs: %d",
+ __func__, ret);
+ fatal_sigsegv();
+ }
- if (singlestepping())
- op = PTRACE_SYSEMU_SINGLESTEP;
- else
- op = PTRACE_SYSEMU;
+ if (proc_data->si_offset > sizeof(proc_data->sigstack) - sizeof(*si))
+ panic("%s - Invalid siginfo offset from child",
+ __func__);
+ si = (void *)&proc_data->sigstack[proc_data->si_offset];
- if (ptrace(op, pid, 0, 0)) {
- printk(UM_KERN_ERR "%s - ptrace continue failed, op = %d, errno = %d\n",
- __func__, op, errno);
- fatal_sigsegv();
- }
+ regs->is_user = 1;
- CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
- if (err < 0) {
- printk(UM_KERN_ERR "%s - wait failed, errno = %d\n",
- __func__, errno);
- fatal_sigsegv();
- }
+ /* Fill in ORIG_RAX and extract fault information */
+ PT_SYSCALL_NR(regs->gp) = si->si_syscall;
+ if (sig == SIGSEGV) {
+ mcontext_t *mcontext = (void *)&proc_data->sigstack[proc_data->mctx_offset];
- regs->is_user = 1;
- if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
- printk(UM_KERN_ERR "%s - PTRACE_GETREGS failed, errno = %d\n",
- __func__, errno);
- fatal_sigsegv();
- }
+ GET_FAULTINFO_FROM_MC(regs->faultinfo, mcontext);
+ }
+ } else {
+ /* Flush out any pending syscalls */
+ err = syscall_stub_flush(current_mm_id());
+ if (err) {
+ if (err == -ENOMEM)
+ report_enomem();
+
+ printk(UM_KERN_ERR "%s - Error flushing stub syscalls: %d",
+ __func__, -err);
+ fatal_sigsegv();
+ }
- if (get_fp_registers(pid, regs->fp)) {
- printk(UM_KERN_ERR "%s - get_fp_registers failed, errno = %d\n",
- __func__, errno);
- fatal_sigsegv();
- }
+ /*
+ * This can legitimately fail if the process loads a
+ * bogus value into a segment register. It will
+ * segfault and PTRACE_GETREGS will read that value
+ * out of the process. However, PTRACE_SETREGS will
+ * fail. In this case, there is nothing to do but
+ * just kill the process.
+ */
+ if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
+ printk(UM_KERN_ERR "%s - ptrace set regs failed, errno = %d\n",
+ __func__, errno);
+ fatal_sigsegv();
+ }
- UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
+ if (put_fp_registers(pid, regs->fp)) {
+ printk(UM_KERN_ERR "%s - ptrace set fp regs failed, errno = %d\n",
+ __func__, errno);
+ fatal_sigsegv();
+ }
- if (WIFSTOPPED(status)) {
- int sig = WSTOPSIG(status);
+ if (singlestepping())
+ op = PTRACE_SYSEMU_SINGLESTEP;
+ else
+ op = PTRACE_SYSEMU;
- /* These signal handlers need the si argument.
- * The SIGIO and SIGALARM handlers which constitute the
- * majority of invocations, do not use it.
- */
- switch (sig) {
- case SIGSEGV:
- case SIGTRAP:
- case SIGILL:
- case SIGBUS:
- case SIGFPE:
- case SIGWINCH:
- ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
- break;
+ if (ptrace(op, pid, 0, 0)) {
+ printk(UM_KERN_ERR "%s - ptrace continue failed, op = %d, errno = %d\n",
+ __func__, op, errno);
+ fatal_sigsegv();
+ }
+
+ CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
+ if (err < 0) {
+ printk(UM_KERN_ERR "%s - wait failed, errno = %d\n",
+ __func__, errno);
+ fatal_sigsegv();
}
+ regs->is_user = 1;
+ if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
+ printk(UM_KERN_ERR "%s - PTRACE_GETREGS failed, errno = %d\n",
+ __func__, errno);
+ fatal_sigsegv();
+ }
+
+ if (get_fp_registers(pid, regs->fp)) {
+ printk(UM_KERN_ERR "%s - get_fp_registers failed, errno = %d\n",
+ __func__, errno);
+ fatal_sigsegv();
+ }
+
+ if (WIFSTOPPED(status)) {
+ sig = WSTOPSIG(status);
+
+ /*
+ * These signal handlers need the si argument
+ * and SIGSEGV needs the faultinfo.
+ * The SIGIO and SIGALARM handlers which constitute
+ * the majority of invocations, do not use it.
+ */
+ switch (sig) {
+ case SIGSEGV:
+ get_skas_faultinfo(pid,
+ ®s->faultinfo);
+ fallthrough;
+ case SIGTRAP:
+ case SIGILL:
+ case SIGBUS:
+ case SIGFPE:
+ case SIGWINCH:
+ ptrace(PTRACE_GETSIGINFO, pid, 0,
+ (struct siginfo *)&si_ptrace);
+ si = &si_ptrace;
+ break;
+ default:
+ si = NULL;
+ break;
+ }
+ } else {
+ sig = 0;
+ }
+ }
+
+ UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
+
+ if (sig) {
switch (sig) {
case SIGSEGV:
- get_skas_faultinfo(pid, ®s->faultinfo);
-
- if (PTRACE_FULL_FAULTINFO)
- (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
+ if (using_seccomp || PTRACE_FULL_FAULTINFO)
+ (*sig_info[SIGSEGV])(SIGSEGV,
+ (struct siginfo *)si,
regs, NULL);
else
segv(regs->faultinfo, 0, 1, NULL, NULL);
+ break;
+ case SIGSYS:
+ handle_syscall(regs);
break;
case SIGTRAP + 0x80:
handle_trap(pid, regs);
break;
case SIGTRAP:
- relay_signal(SIGTRAP, (struct siginfo *)&si, regs, NULL);
+ relay_signal(SIGTRAP, (struct siginfo *)si, regs, NULL);
break;
case SIGALRM:
break;
@@ -539,7 +713,7 @@ void userspace(struct uml_pt_regs *regs)
case SIGFPE:
case SIGWINCH:
block_signals_trace();
- (*sig_info[sig])(sig, (struct siginfo *)&si, regs, NULL);
+ (*sig_info[sig])(sig, (struct siginfo *)si, regs, NULL);
unblock_signals_trace();
break;
default:
diff --git a/arch/x86/um/shared/sysdep/kernel-offsets.h b/arch/x86/um/shared/sysdep/kernel-offsets.h
index 48de3a71f845..6fd1ed400399 100644
--- a/arch/x86/um/shared/sysdep/kernel-offsets.h
+++ b/arch/x86/um/shared/sysdep/kernel-offsets.h
@@ -4,7 +4,9 @@
#include <linux/elf.h>
#include <linux/crypto.h>
#include <linux/kbuild.h>
+#include <linux/audit.h>
#include <asm/mman.h>
+#include <asm/seccomp.h>
/* workaround for a warning with -Wmissing-prototypes */
void foo(void);
diff --git a/arch/x86/um/tls_32.c b/arch/x86/um/tls_32.c
index fbb129023080..21cbb70cf771 100644
--- a/arch/x86/um/tls_32.c
+++ b/arch/x86/um/tls_32.c
@@ -12,6 +12,7 @@
#include <skas.h>
#include <sysdep/tls.h>
#include <asm/desc.h>
+#include <stub-data.h>
/*
* If needed we can detect when it's uninitialized.
@@ -21,13 +22,27 @@
static int host_supports_tls = -1;
int host_gdt_entry_tls_min;
-static int do_set_thread_area(struct user_desc *info)
+static int do_set_thread_area(struct task_struct* task, struct user_desc *info)
{
int ret;
u32 cpu;
+ if (info->entry_number < host_gdt_entry_tls_min ||
+ info->entry_number >= host_gdt_entry_tls_min + GDT_ENTRY_TLS_ENTRIES)
+ return -EINVAL;
+
+ if (using_seccomp) {
+ int idx = info->entry_number - host_gdt_entry_tls_min;
+ struct stub_data *data = (void *)task->mm->context.id.stack;
+
+ data->arch_data.tls[idx] = *info;
+ data->arch_data.sync |= BIT(idx);
+
+ return 0;
+ }
+
cpu = get_cpu();
- ret = os_set_thread_area(info, userspace_pid[cpu]);
+ ret = os_set_thread_area(info, task->mm->context.id.pid);
put_cpu();
if (ret)
@@ -97,7 +112,7 @@ static int load_TLS(int flags, struct task_struct *to)
if (!(flags & O_FORCE) && curr->flushed)
continue;
- ret = do_set_thread_area(&curr->tls);
+ ret = do_set_thread_area(current, &curr->tls);
if (ret)
goto out;
@@ -275,7 +290,7 @@ SYSCALL_DEFINE1(set_thread_area, struct user_desc __user *, user_desc)
return -EFAULT;
}
- ret = do_set_thread_area(&info);
+ ret = do_set_thread_area(current, &info);
if (ret)
return ret;
return set_tls_entry(current, &info, idx, 1);
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH v3 6/7] um: Add SECCOMP support detection and initialization
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
` (4 preceding siblings ...)
2025-06-02 13:00 ` [PATCH v3 5/7] um: Implement kernel side of SECCOMP based process handling Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
2025-06-02 13:00 ` [PATCH v3 7/7] um: pass FD for memory operations when needed Benjamin Berg
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg, Benjamin Berg
This detects seccomp support, sets the global using_seccomp variable and
initilizes the exec registers. The support is only enabled if the
seccomp= kernel parameter is set to either "on" or "auto". With "auto" a
fallback to ptrace mode will happen if initialization failed.
Signed-off-by: Benjamin Berg <benjamin@sipsolutions.net>
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
v3:
- Fix FP size detection on some hosts. The MAP_SHARED | MAP_ANON mapping
would not always be shared after fork(), resulting in the detection to
fail.
- Add runtime option to select seccomp mode
---
arch/um/os-Linux/registers.c | 4 +-
arch/um/os-Linux/start_up.c | 177 ++++++++++++++++++++++++++++++++++-
2 files changed, 177 insertions(+), 4 deletions(-)
diff --git a/arch/um/os-Linux/registers.c b/arch/um/os-Linux/registers.c
index d7ca148807b2..bfba2cbc9478 100644
--- a/arch/um/os-Linux/registers.c
+++ b/arch/um/os-Linux/registers.c
@@ -14,8 +14,8 @@
/* This is set once at boot time and not changed thereafter */
-static unsigned long exec_regs[MAX_REG_NR];
-static unsigned long *exec_fp_regs;
+unsigned long exec_regs[MAX_REG_NR];
+unsigned long *exec_fp_regs;
int init_pid_registers(int pid)
{
diff --git a/arch/um/os-Linux/start_up.c b/arch/um/os-Linux/start_up.c
index 93fc82c01aba..5d2e604ab02d 100644
--- a/arch/um/os-Linux/start_up.c
+++ b/arch/um/os-Linux/start_up.c
@@ -1,5 +1,6 @@
// SPDX-License-Identifier: GPL-2.0
/*
+ * Copyright (C) 2021 Benjamin Berg <benjamin@sipsolutions.net>
* Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
*/
@@ -24,6 +25,13 @@
#include <kern_util.h>
#include <mem_user.h>
#include <ptrace_user.h>
+#include <stdbool.h>
+#include <stub-data.h>
+#include <sys/prctl.h>
+#include <linux/seccomp.h>
+#include <linux/filter.h>
+#include <sysdep/mcontext.h>
+#include <sysdep/stub.h>
#include <registers.h>
#include <skas.h>
#include "internal.h"
@@ -224,6 +232,136 @@ static void __init check_ptrace(void)
check_sysemu();
}
+extern unsigned long host_fp_size;
+extern unsigned long exec_regs[MAX_REG_NR];
+extern unsigned long *exec_fp_regs;
+
+__initdata static struct stub_data *seccomp_test_stub_data;
+
+static void __init sigsys_handler(int sig, siginfo_t *info, void *p)
+{
+ ucontext_t *uc = p;
+
+ /* Stow away the location of the mcontext in the stack */
+ seccomp_test_stub_data->mctx_offset = (unsigned long)&uc->uc_mcontext -
+ (unsigned long)&seccomp_test_stub_data->sigstack[0];
+ _exit(0);
+}
+
+static int __init seccomp_helper(void *data)
+{
+ static struct sock_filter filter[] = {
+ BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
+ offsetof(struct seccomp_data, nr)),
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_clock_nanosleep, 1, 0),
+ BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
+ BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_TRAP),
+ };
+ static struct sock_fprog prog = {
+ .len = ARRAY_SIZE(filter),
+ .filter = filter,
+ };
+ struct sigaction sa;
+
+ seccomp_test_stub_data->mctx_offset = 1;
+
+ set_sigstack(seccomp_test_stub_data->sigstack,
+ sizeof(seccomp_test_stub_data->sigstack));
+
+ sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
+ sa.sa_sigaction = (void *) sigsys_handler;
+ sa.sa_restorer = NULL;
+ if (sigaction(SIGSYS, &sa, NULL) < 0)
+ exit(1);
+
+ prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
+ if (syscall(__NR_seccomp, SECCOMP_SET_MODE_FILTER,
+ SECCOMP_FILTER_FLAG_TSYNC, &prog) != 0)
+ exit(2);
+
+ sleep(0);
+
+ /* Never reached. */
+ _exit(3);
+}
+
+static bool __init init_seccomp(void)
+{
+ int pid;
+ int status;
+ int n;
+ unsigned long sp;
+
+ /*
+ * We check that we can install a seccomp filter and then exit(0)
+ * from a trapped syscall.
+ *
+ * Note that we cannot verify that no seccomp filter already exists
+ * for a syscall that results in the process/thread to be killed.
+ */
+
+ os_info("Checking that seccomp filters can be installed...");
+
+ seccomp_test_stub_data = mmap(0, sizeof(*seccomp_test_stub_data),
+ PROT_READ | PROT_WRITE,
+ MAP_SHARED | MAP_ANON, 0, 0);
+
+ /* Use the syscall data area as stack, we just need something */
+ sp = (unsigned long)&seccomp_test_stub_data->syscall_data +
+ sizeof(seccomp_test_stub_data->syscall_data) -
+ sizeof(void *);
+ pid = clone(seccomp_helper, (void *)sp, CLONE_VFORK | CLONE_VM, NULL);
+
+ if (pid < 0)
+ fatal_perror("check_seccomp : clone failed");
+
+ CATCH_EINTR(n = waitpid(pid, &status, __WCLONE));
+ if (n < 0)
+ fatal_perror("check_seccomp : waitpid failed");
+
+ if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
+ struct uml_pt_regs *regs;
+ unsigned long fp_size;
+ int r;
+
+ /* Fill in the host_fp_size from the mcontext. */
+ regs = calloc(1, sizeof(struct uml_pt_regs));
+ get_stub_state(regs, seccomp_test_stub_data, &fp_size);
+ host_fp_size = fp_size;
+ free(regs);
+
+ /* Repeat with the correct size */
+ regs = calloc(1, sizeof(struct uml_pt_regs) + host_fp_size);
+ r = get_stub_state(regs, seccomp_test_stub_data, NULL);
+
+ /* Store as the default startup registers */
+ exec_fp_regs = malloc(host_fp_size);
+ memcpy(exec_regs, regs->gp, sizeof(exec_regs));
+ memcpy(exec_fp_regs, regs->fp, host_fp_size);
+
+ munmap(seccomp_test_stub_data, sizeof(*seccomp_test_stub_data));
+
+ free(regs);
+
+ if (r) {
+ os_info("failed to fetch registers: %d\n", r);
+ return false;
+ }
+
+ os_info("OK\n");
+ return true;
+ }
+
+ if (WIFEXITED(status) && WEXITSTATUS(status) == 2)
+ os_info("missing\n");
+ else
+ os_info("error\n");
+
+ munmap(seccomp_test_stub_data, sizeof(*seccomp_test_stub_data));
+ return false;
+}
+
+
static void __init check_coredump_limit(void)
{
struct rlimit lim;
@@ -278,6 +416,30 @@ void __init get_host_cpu_features(
}
}
+static int seccomp_config __initdata;
+
+static int __init uml_seccomp_config(char *line, int *add)
+{
+ *add = 0;
+
+ if (strcmp(line, "off") == 0)
+ seccomp_config = 0;
+ else if (strcmp(line, "auto") == 0)
+ seccomp_config = 1;
+ else if (strcmp(line, "on") == 0)
+ seccomp_config = 2;
+ else
+ fatal("Invalid seccomp option '%s', expected on/auto/off\n",
+ line);
+
+ return 0;
+}
+
+__uml_setup("seccomp=", uml_seccomp_config,
+"seccomp=<on/auto/off>\n"
+" Whether to run userspace using SECCOMP.\n"
+" This is insecure and should only be used with a trusted userspace\n\n"
+);
void __init os_early_checks(void)
{
@@ -286,13 +448,24 @@ void __init os_early_checks(void)
/* Print out the core dump limits early */
check_coredump_limit();
- check_ptrace();
-
/* Need to check this early because mmapping happens before the
* kernel is running.
*/
check_tmpexec();
+ if (seccomp_config) {
+ if (init_seccomp()) {
+ using_seccomp = 1;
+ return;
+ }
+
+ if (seccomp_config == 2)
+ fatal("SECCOMP userspace requested but not functional!");
+ }
+
+ using_seccomp = 0;
+ check_ptrace();
+
pid = start_ptraced_child();
if (init_pid_registers(pid))
fatal("Failed to initialize default registers");
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH v3 7/7] um: pass FD for memory operations when needed
2025-06-02 13:00 [PATCH v3 0/7] SECCOMP based userspace for UML Benjamin Berg
` (5 preceding siblings ...)
2025-06-02 13:00 ` [PATCH v3 6/7] um: Add SECCOMP support detection and initialization Benjamin Berg
@ 2025-06-02 13:00 ` Benjamin Berg
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Berg @ 2025-06-02 13:00 UTC (permalink / raw)
To: linux-um; +Cc: Benjamin Berg
From: Benjamin Berg <benjamin.berg@intel.com>
Instead of always sharing the FDs with the userspace process, only hand
over the FDs needed for mmap when required. The idea is that userspace
might be able to force the stub into executing an mmap syscall, however,
it will not be able to manipulate the control flow sufficiently to have
access to an FD that would allow mapping arbitrary memory.
Security wise, we need to be sure that only the expected syscalls are
executed after the kernel sends FDs through the socket. This is
currently not the case, as userspace can trivially jump to the
rt_sigreturn syscall instruction to execute any syscall that the stub is
permitted to do. With this, it can trick the kernel to send the FD,
which in turn allows userspace to freely map any physical memory.
As such, this is currently *not* secure. However, in principle the
approach should be fine with a more strict SECCOMP filter and a careful
review of the stub control flow (as userspace can prepare a stack). With
some care, it is likely possible to extend the security model to SMP if
desired.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
---
v1:
* Add startup and runtime checks for close_range
---
arch/um/include/shared/skas/mm_id.h | 7 ++
arch/um/include/shared/skas/stub-data.h | 1 +
arch/um/kernel/skas/mmu.c | 3 +
arch/um/kernel/skas/stub.c | 87 ++++++++++++++--
arch/um/kernel/skas/stub_exe.c | 40 +++++---
arch/um/os-Linux/skas/mem.c | 66 ++++++++++++-
arch/um/os-Linux/skas/process.c | 126 +++++++++++++++++-------
arch/um/os-Linux/start_up.c | 10 +-
8 files changed, 280 insertions(+), 60 deletions(-)
diff --git a/arch/um/include/shared/skas/mm_id.h b/arch/um/include/shared/skas/mm_id.h
index 0654c57bb28e..89df9a55fbea 100644
--- a/arch/um/include/shared/skas/mm_id.h
+++ b/arch/um/include/shared/skas/mm_id.h
@@ -6,10 +6,17 @@
#ifndef __MM_ID_H
#define __MM_ID_H
+#define STUB_MAX_FDS 4
+
struct mm_id {
int pid;
unsigned long stack;
int syscall_data_len;
+
+ /* Only used with SECCOMP mode */
+ int sock;
+ int syscall_fd_num;
+ int syscall_fd_map[STUB_MAX_FDS];
};
void __switch_mm(struct mm_id *mm_idp);
diff --git a/arch/um/include/shared/skas/stub-data.h b/arch/um/include/shared/skas/stub-data.h
index 675f1a0a1390..c261a77a32f6 100644
--- a/arch/um/include/shared/skas/stub-data.h
+++ b/arch/um/include/shared/skas/stub-data.h
@@ -12,6 +12,7 @@
#include <as-layout.h>
#include <sysdep/tls.h>
#include <sysdep/stub-data.h>
+#include <mm_id.h>
#define FUTEX_IN_CHILD 0
#define FUTEX_IN_KERN 1
diff --git a/arch/um/kernel/skas/mmu.c b/arch/um/kernel/skas/mmu.c
index 87a18ae4da19..849fafa4b54f 100644
--- a/arch/um/kernel/skas/mmu.c
+++ b/arch/um/kernel/skas/mmu.c
@@ -78,6 +78,9 @@ void destroy_context(struct mm_struct *mm)
mmu->id.pid = -1;
}
+ if (using_seccomp && mmu->id.sock)
+ os_close_file(mmu->id.sock);
+
free_pages(mmu->id.stack, ilog2(STUB_DATA_PAGES));
guard(spinlock_irqsave)(&mm_list_lock);
diff --git a/arch/um/kernel/skas/stub.c b/arch/um/kernel/skas/stub.c
index 9041f6b6e28b..67cab46a602c 100644
--- a/arch/um/kernel/skas/stub.c
+++ b/arch/um/kernel/skas/stub.c
@@ -6,23 +6,53 @@
#include <sysdep/stub.h>
#include <linux/futex.h>
+#include <sys/socket.h>
#include <errno.h>
-static __always_inline int syscall_handler(struct stub_data *d)
+/*
+ * Known security issues
+ *
+ * Userspace can jump to this address to execute *any* syscall that is
+ * permitted by the stub. As we will return afterwards, it can do
+ * whatever it likes, including:
+ * - Tricking the kernel into handing out the memory FD
+ * - Using this memory FD to read/write all physical memory
+ * - Running in parallel to the kernel processing a syscall
+ * (possibly creating data races?)
+ * - Blocking e.g. SIGALRM to avoid time based scheduling
+ *
+ * To avoid this, the permitted location for each syscall needs to be
+ * checked for in the SECCOMP filter (which is reasonably simple). Also,
+ * more care will need to go into considerations how the code might be
+ * tricked by using a prepared stack (or even modifying the stack from
+ * another thread in case SMP support is added).
+ *
+ * As for the SIGALRM, the best counter measure will be to check in the
+ * kernel that the process is reporting back the SIGALRM in a timely
+ * fashion.
+ */
+static __always_inline int syscall_handler(int fd_map[STUB_MAX_FDS])
{
+ struct stub_data *d = get_stub_data();
int i;
unsigned long res;
+ int fd;
for (i = 0; i < d->syscall_data_len; i++) {
struct stub_syscall *sc = &d->syscall_data[i];
switch (sc->syscall) {
case STUB_SYSCALL_MMAP:
+ if (fd_map)
+ fd = fd_map[sc->mem.fd];
+ else
+ fd = sc->mem.fd;
+
res = stub_syscall6(STUB_MMAP_NR,
sc->mem.addr, sc->mem.length,
sc->mem.prot,
MAP_SHARED | MAP_FIXED,
- sc->mem.fd, sc->mem.offset);
+ fd, sc->mem.offset);
if (res != sc->mem.addr) {
d->err = res;
d->syscall_data_len = i;
@@ -54,9 +84,7 @@ static __always_inline int syscall_handler(struct stub_data *d)
void __section(".__syscall_stub")
stub_syscall_handler(void)
{
- struct stub_data *d = get_stub_data();
-
- syscall_handler(d);
+ syscall_handler(NULL);
trap_myself();
}
@@ -65,7 +93,25 @@ void __section(".__syscall_stub")
stub_signal_interrupt(int sig, siginfo_t *info, void *p)
{
struct stub_data *d = get_stub_data();
+ char rcv_data;
+ union {
+ char data[CMSG_SPACE(sizeof(int) * STUB_MAX_FDS)];
+ struct cmsghdr align;
+ } ctrl = {};
+ struct iovec iov = {
+ .iov_base = &rcv_data,
+ .iov_len = 1,
+ };
+ struct msghdr msghdr = {
+ .msg_iov = &iov,
+ .msg_iovlen = 1,
+ .msg_control = &ctrl,
+ .msg_controllen = sizeof(ctrl),
+ };
ucontext_t *uc = p;
+ struct cmsghdr *fd_msg;
+ int *fd_map;
+ int num_fds;
long res;
d->signal = sig;
@@ -78,6 +124,7 @@ stub_signal_interrupt(int sig, siginfo_t *info, void *p)
res = stub_syscall3(__NR_futex, (unsigned long)&d->futex,
FUTEX_WAKE, 1);
} while (res == -EINTR);
+
do {
res = stub_syscall4(__NR_futex, (unsigned long)&d->futex,
FUTEX_WAIT, FUTEX_IN_KERN, 0);
@@ -86,11 +133,37 @@ stub_signal_interrupt(int sig, siginfo_t *info, void *p)
if (res < 0 && res != -EAGAIN)
stub_syscall1(__NR_exit_group, 1);
- /* Try running queued syscalls. */
- if (syscall_handler(d) < 0 || d->restart_wait) {
+ if (d->syscall_data_len) {
+ /* Read passed FDs (if any) */
+ do {
+ res = stub_syscall3(__NR_recvmsg, 0, (unsigned long)&msghdr, 0);
+ } while (res == -EINTR);
+
+ /* We should never have a receive error (other than -EAGAIN) */
+ if (res < 0 && res != -EAGAIN)
+ stub_syscall1(__NR_exit_group, 1);
+
+ /* Receive the FDs */
+ num_fds = 0;
+ fd_msg = msghdr.msg_control;
+ fd_map = (void *)&CMSG_DATA(fd_msg);
+ if (res == iov.iov_len && msghdr.msg_controllen > sizeof(struct cmsghdr))
+ num_fds = (fd_msg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
+
+ /* Try running queued syscalls. */
+ res = syscall_handler(fd_map);
+
+ while (num_fds)
+ stub_syscall2(__NR_close, fd_map[--num_fds], 0);
+ } else {
+ res = 0;
+ }
+
+ if (res < 0 || d->restart_wait) {
/* Report SIGSYS if we restart. */
d->signal = SIGSYS;
d->restart_wait = 0;
+
goto restart_wait;
}
diff --git a/arch/um/kernel/skas/stub_exe.c b/arch/um/kernel/skas/stub_exe.c
index f40f2332b676..cbafaa684e66 100644
--- a/arch/um/kernel/skas/stub_exe.c
+++ b/arch/um/kernel/skas/stub_exe.c
@@ -1,5 +1,6 @@
#include <sys/ptrace.h>
#include <sys/prctl.h>
+#include <sys/fcntl.h>
#include <asm/unistd.h>
#include <sysdep/stub.h>
#include <stub-data.h>
@@ -45,7 +46,11 @@ noinline static void real_init(void)
if (res != sizeof(init_data))
stub_syscall1(__NR_exit, 10);
- stub_syscall1(__NR_close, 0);
+ /* In SECCOMP mode, FD 0 is a socket and is later used for FD passing */
+ if (!init_data.seccomp)
+ stub_syscall1(__NR_close, 0);
+ else
+ stub_syscall3(__NR_fcntl, 0, F_SETFL, O_NONBLOCK);
/* map stub code + data */
res = stub_syscall6(STUB_MMAP_NR,
@@ -63,6 +68,13 @@ noinline static void real_init(void)
if (res != init_data.stub_start + UM_KERN_PAGE_SIZE)
stub_syscall1(__NR_exit, 12);
+ /* In SECCOMP mode, we only need the signalling FD from now on */
+ if (init_data.seccomp) {
+ res = stub_syscall3(__NR_close_range, 1, ~0U, 0);
+ if (res != 0)
+ stub_syscall1(__NR_exit, 13);
+ }
+
/* setup signal stack inside stub data */
stack.ss_sp = (void *)init_data.stub_start + UM_KERN_PAGE_SIZE;
stub_syscall2(__NR_sigaltstack, (unsigned long)&stack, 0);
@@ -77,7 +89,7 @@ noinline static void real_init(void)
res = stub_syscall4(__NR_rt_sigaction, SIGSEGV,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 13);
+ stub_syscall1(__NR_exit, 14);
} else {
/* SECCOMP mode uses rt_sigreturn, need to mask all signals */
sa.sa_mask = ~0ULL;
@@ -85,32 +97,32 @@ noinline static void real_init(void)
res = stub_syscall4(__NR_rt_sigaction, SIGSEGV,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 14);
+ stub_syscall1(__NR_exit, 15);
res = stub_syscall4(__NR_rt_sigaction, SIGSYS,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 15);
+ stub_syscall1(__NR_exit, 16);
res = stub_syscall4(__NR_rt_sigaction, SIGALRM,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 16);
+ stub_syscall1(__NR_exit, 17);
res = stub_syscall4(__NR_rt_sigaction, SIGTRAP,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 17);
+ stub_syscall1(__NR_exit, 18);
res = stub_syscall4(__NR_rt_sigaction, SIGILL,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 18);
+ stub_syscall1(__NR_exit, 19);
res = stub_syscall4(__NR_rt_sigaction, SIGFPE,
(unsigned long)&sa, 0, sizeof(sa.sa_mask));
if (res != 0)
- stub_syscall1(__NR_exit, 19);
+ stub_syscall1(__NR_exit, 20);
}
/*
@@ -153,8 +165,12 @@ noinline static void real_init(void)
BPF_STMT(BPF_LD | BPF_W | BPF_ABS,
offsetof(struct seccomp_data, nr)),
- /* [10-14] Check against permitted syscalls */
+ /* [10-16] Check against permitted syscalls */
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_futex,
+ 7, 0),
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K,__NR_recvmsg,
+ 6, 0),
+ BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K,__NR_close,
5, 0),
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, STUB_MMAP_NR,
4, 0),
@@ -170,10 +186,10 @@ noinline static void real_init(void)
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_rt_sigreturn,
1, 0),
- /* [15] Not one of the permitted syscalls */
+ /* [17] Not one of the permitted syscalls */
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS),
- /* [16] Permitted call for the stub */
+ /* [18] Permitted call for the stub */
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
};
struct sock_fprog prog = {
@@ -184,7 +200,7 @@ noinline static void real_init(void)
if (stub_syscall3(__NR_seccomp, SECCOMP_SET_MODE_FILTER,
SECCOMP_FILTER_FLAG_TSYNC,
(unsigned long)&prog) != 0)
- stub_syscall1(__NR_exit, 20);
+ stub_syscall1(__NR_exit, 21);
/* Fall through, the exit syscall will cause SIGSYS */
} else {
diff --git a/arch/um/os-Linux/skas/mem.c b/arch/um/os-Linux/skas/mem.c
index 31bf3a52047a..8b9921ac3ef8 100644
--- a/arch/um/os-Linux/skas/mem.c
+++ b/arch/um/os-Linux/skas/mem.c
@@ -43,6 +43,16 @@ void syscall_stub_dump_error(struct mm_id *mm_idp)
print_hex_dump(UM_KERN_ERR, " syscall data: ", 0,
16, 4, sc, sizeof(*sc), 0);
+
+ if (using_seccomp) {
+ printk(UM_KERN_ERR "%s: FD map num: %d", __func__,
+ mm_idp->syscall_fd_num);
+ print_hex_dump(UM_KERN_ERR,
+ " FD map: ", 0, 16,
+ sizeof(mm_idp->syscall_fd_map[0]),
+ mm_idp->syscall_fd_map,
+ sizeof(mm_idp->syscall_fd_map), 0);
+ }
}
static inline unsigned long *check_init_stack(struct mm_id * mm_idp,
@@ -118,6 +128,9 @@ static inline long do_syscall_stub(struct mm_id *mm_idp)
mm_idp->syscall_data_len = 0;
}
+ if (using_seccomp)
+ mm_idp->syscall_fd_num = 0;
+
return mm_idp->syscall_data_len;
}
@@ -180,6 +193,44 @@ static struct stub_syscall *syscall_stub_get_previous(struct mm_id *mm_idp,
return NULL;
}
+static int get_stub_fd(struct mm_id *mm_idp, int fd)
+{
+ int i;
+
+ /* Find an FD slot (or flush and use first) */
+ if (!using_seccomp)
+ return fd;
+
+ /* Already crashed, value does not matter */
+ if (mm_idp->syscall_data_len < 0)
+ return 0;
+
+ /* Find existing FD in map if we can allocate another syscall */
+ if (mm_idp->syscall_data_len <
+ ARRAY_SIZE(((struct stub_data *)NULL)->syscall_data)) {
+ for (i = 0; i < mm_idp->syscall_fd_num; i++) {
+ if (mm_idp->syscall_fd_map[i] == fd)
+ return i;
+ }
+
+ if (mm_idp->syscall_fd_num < STUB_MAX_FDS) {
+ i = mm_idp->syscall_fd_num;
+ mm_idp->syscall_fd_map[i] = fd;
+
+ mm_idp->syscall_fd_num++;
+
+ return i;
+ }
+ }
+
+ /* FD map full or no syscall space available, continue after flush */
+ do_syscall_stub(mm_idp);
+ mm_idp->syscall_fd_map[0] = fd;
+ mm_idp->syscall_fd_num = 1;
+
+ return 0;
+}
+
int map(struct mm_id *mm_idp, unsigned long virt, unsigned long len, int prot,
int phys_fd, unsigned long long offset)
{
@@ -187,12 +238,21 @@ int map(struct mm_id *mm_idp, unsigned long virt, unsigned long len, int prot,
/* Compress with previous syscall if that is possible */
sc = syscall_stub_get_previous(mm_idp, STUB_SYSCALL_MMAP, virt);
- if (sc && sc->mem.prot == prot && sc->mem.fd == phys_fd &&
+ if (sc && sc->mem.prot == prot &&
sc->mem.offset == MMAP_OFFSET(offset - sc->mem.length)) {
- sc->mem.length += len;
- return 0;
+ int prev_fd = sc->mem.fd;
+
+ if (using_seccomp)
+ prev_fd = mm_idp->syscall_fd_map[sc->mem.fd];
+
+ if (phys_fd == prev_fd) {
+ sc->mem.length += len;
+ return 0;
+ }
}
+ phys_fd = get_stub_fd(mm_idp, phys_fd);
+
sc = syscall_stub_alloc(mm_idp);
sc->syscall = STUB_SYSCALL_MMAP;
sc->mem.addr = virt;
diff --git a/arch/um/os-Linux/skas/process.c b/arch/um/os-Linux/skas/process.c
index e1aaa144e273..e42ffac23e3c 100644
--- a/arch/um/os-Linux/skas/process.c
+++ b/arch/um/os-Linux/skas/process.c
@@ -16,6 +16,7 @@
#include <sys/mman.h>
#include <sys/wait.h>
#include <sys/stat.h>
+#include <sys/socket.h>
#include <asm/unistd.h>
#include <as-layout.h>
#include <init.h>
@@ -152,7 +153,39 @@ void wait_stub_done_seccomp(struct mm_id *mm_idp, int running, int wait_sigsys)
int ret;
do {
+ const char byte = 0;
+ struct iovec iov = {
+ .iov_base = (void *)&byte,
+ .iov_len = sizeof(byte),
+ };
+ union {
+ char data[CMSG_SPACE(sizeof(mm_idp->syscall_fd_map))];
+ struct cmsghdr align;
+ } ctrl;
+ struct msghdr msgh = {
+ .msg_iov = &iov,
+ .msg_iovlen = 1,
+ };
+
if (!running) {
+ if (mm_idp->syscall_fd_num) {
+ unsigned int fds_size =
+ sizeof(int) * mm_idp->syscall_fd_num;
+ struct cmsghdr *cmsg;
+
+ msgh.msg_control = ctrl.data;
+ msgh.msg_controllen = CMSG_SPACE(fds_size);
+ cmsg = CMSG_FIRSTHDR(&msgh);
+ cmsg->cmsg_level = SOL_SOCKET;
+ cmsg->cmsg_type = SCM_RIGHTS;
+ cmsg->cmsg_len = CMSG_LEN(fds_size);
+ memcpy(CMSG_DATA(cmsg), mm_idp->syscall_fd_map,
+ fds_size);
+
+ CATCH_EINTR(syscall(__NR_sendmsg, mm_idp->sock,
+ &msgh, 0));
+ }
+
data->signal = 0;
data->futex = FUTEX_IN_CHILD;
CATCH_EINTR(syscall(__NR_futex, &data->futex,
@@ -246,14 +279,20 @@ extern char __syscall_stub_start[];
static int stub_exe_fd;
+struct tramp_data {
+ struct stub_data *stub_data;
+ /* 0 is inherited, 1 is the kernel side */
+ int sockpair[2];
+};
+
#ifndef CLOSE_RANGE_CLOEXEC
#define CLOSE_RANGE_CLOEXEC (1U << 2)
#endif
-static int userspace_tramp(void *stack)
+static int userspace_tramp(void *data)
{
+ struct tramp_data *tramp_data = data;
char *const argv[] = { "uml-userspace", NULL };
- int pipe_fds[2];
unsigned long long offset;
struct stub_init_data init_data = {
.seccomp = using_seccomp,
@@ -280,7 +319,8 @@ static int userspace_tramp(void *stack)
&offset);
init_data.stub_code_offset = MMAP_OFFSET(offset);
- init_data.stub_data_fd = phys_mapping(uml_to_phys(stack), &offset);
+ init_data.stub_data_fd = phys_mapping(uml_to_phys(tramp_data->stub_data),
+ &offset);
init_data.stub_data_offset = MMAP_OFFSET(offset);
/*
@@ -291,20 +331,21 @@ static int userspace_tramp(void *stack)
syscall(__NR_close_range, 0, ~0U, CLOSE_RANGE_CLOEXEC);
fcntl(init_data.stub_data_fd, F_SETFD, 0);
- for (iomem = iomem_regions; iomem; iomem = iomem->next)
- fcntl(iomem->fd, F_SETFD, 0);
- /* Create a pipe for init_data (no CLOEXEC) and dup2 to STDIN */
- if (pipe(pipe_fds))
- exit(2);
+ /* In SECCOMP mode, these FDs are passed when needed */
+ if (!using_seccomp) {
+ for (iomem = iomem_regions; iomem; iomem = iomem->next)
+ fcntl(iomem->fd, F_SETFD, 0);
+ }
- if (dup2(pipe_fds[0], 0) < 0)
+ /* dup2 signaling FD/socket to STDIN */
+ if (dup2(tramp_data->sockpair[0], 0) < 0)
exit(3);
- close(pipe_fds[0]);
+ close(tramp_data->sockpair[0]);
/* Write init_data and close write side */
- ret = write(pipe_fds[1], &init_data, sizeof(init_data));
- close(pipe_fds[1]);
+ ret = write(tramp_data->sockpair[1], &init_data, sizeof(init_data));
+ close(tramp_data->sockpair[1]);
if (ret != sizeof(init_data))
exit(4);
@@ -397,7 +438,7 @@ int userspace_pid[NR_CPUS];
/**
* start_userspace() - prepare a new userspace process
- * @stub_stack: pointer to the stub stack.
+ * @mm_id: The corresponding struct mm_id
*
* Setups a new temporary stack page that is used while userspace_tramp() runs
* Clones the kernel process into a new userspace process, with FDs only.
@@ -409,9 +450,12 @@ int userspace_pid[NR_CPUS];
int start_userspace(struct mm_id *mm_id)
{
struct stub_data *proc_data = (void *)mm_id->stack;
+ struct tramp_data tramp_data = {
+ .stub_data = proc_data,
+ };
void *stack;
unsigned long sp;
- int pid, status, n, err;
+ int status, n, err;
/* setup a temporary stack page */
stack = mmap(NULL, UM_KERN_PAGE_SIZE,
@@ -427,25 +471,32 @@ int start_userspace(struct mm_id *mm_id)
/* set stack pointer to the end of the stack page, so it can grow downwards */
sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
+ /* socket pair for init data and SECCOMP FD passing (no CLOEXEC here) */
+ if (socketpair(AF_UNIX, SOCK_STREAM, 0, tramp_data.sockpair)) {
+ err = -errno;
+ printk(UM_KERN_ERR "%s : socketpair failed, errno = %d\n",
+ __func__, errno);
+ return err;
+ }
+
if (using_seccomp)
proc_data->futex = FUTEX_IN_CHILD;
- /* clone into new userspace process */
- pid = clone(userspace_tramp, (void *) sp,
+ mm_id->pid = clone(userspace_tramp, (void *) sp,
CLONE_VFORK | CLONE_VM | SIGCHLD,
- (void *)mm_id->stack);
- if (pid < 0) {
+ (void *)&tramp_data);
+ if (mm_id->pid < 0) {
err = -errno;
printk(UM_KERN_ERR "%s : clone failed, errno = %d\n",
__func__, errno);
- return err;
+ goto out_close;
}
if (using_seccomp) {
wait_stub_done_seccomp(mm_id, 1, 1);
} else {
do {
- CATCH_EINTR(n = waitpid(pid, &status,
+ CATCH_EINTR(n = waitpid(mm_id->pid, &status,
WUNTRACED | __WALL));
if (n < 0) {
err = -errno;
@@ -462,7 +513,7 @@ int start_userspace(struct mm_id *mm_id)
goto out_kill;
}
- if (ptrace(PTRACE_SETOPTIONS, pid, NULL,
+ if (ptrace(PTRACE_SETOPTIONS, mm_id->pid, NULL,
(void *) PTRACE_O_TRACESYSGOOD) < 0) {
err = -errno;
printk(UM_KERN_ERR "%s : PTRACE_SETOPTIONS failed, errno = %d\n",
@@ -478,12 +529,22 @@ int start_userspace(struct mm_id *mm_id)
goto out_kill;
}
- mm_id->pid = pid;
+ close(tramp_data.sockpair[0]);
+ if (using_seccomp)
+ mm_id->sock = tramp_data.sockpair[1];
+ else
+ close(tramp_data.sockpair[1]);
+
+ return 0;
- return pid;
+out_kill:
+ os_kill_ptraced_process(mm_id->pid, 1);
+out_close:
+ close(tramp_data.sockpair[0]);
+ close(tramp_data.sockpair[1]);
+
+ mm_id->pid = -1;
- out_kill:
- os_kill_ptraced_process(pid, 1);
return err;
}
@@ -546,17 +607,8 @@ void userspace(struct uml_pt_regs *regs)
/* Mark pending syscalls for flushing */
proc_data->syscall_data_len = mm_id->syscall_data_len;
- mm_id->syscall_data_len = 0;
- proc_data->signal = 0;
- proc_data->futex = FUTEX_IN_CHILD;
- CATCH_EINTR(syscall(__NR_futex, &proc_data->futex,
- FUTEX_WAKE, 1, NULL, NULL, 0));
- do {
- ret = syscall(__NR_futex, &proc_data->futex,
- FUTEX_WAIT, FUTEX_IN_CHILD, NULL, NULL, 0);
- } while ((ret == -1 && errno == EINTR) ||
- proc_data->futex == FUTEX_IN_CHILD);
+ wait_stub_done_seccomp(mm_id, 0, 0);
sig = proc_data->signal;
@@ -564,9 +616,13 @@ void userspace(struct uml_pt_regs *regs)
printk(UM_KERN_ERR "%s - Error flushing stub syscalls",
__func__);
syscall_stub_dump_error(mm_id);
+ mm_id->syscall_data_len = proc_data->err;
fatal_sigsegv();
}
+ mm_id->syscall_data_len = 0;
+ mm_id->syscall_fd_num = 0;
+
ret = get_stub_state(regs, proc_data, NULL);
if (ret) {
printk(UM_KERN_ERR "%s - failed to get regs: %d",
diff --git a/arch/um/os-Linux/start_up.c b/arch/um/os-Linux/start_up.c
index 5d2e604ab02d..3cbec7627fe1 100644
--- a/arch/um/os-Linux/start_up.c
+++ b/arch/um/os-Linux/start_up.c
@@ -263,6 +263,10 @@ static int __init seccomp_helper(void *data)
};
struct sigaction sa;
+ /* close_range is needed for the stub */
+ if (stub_syscall3(__NR_close_range, 1, ~0U, 0))
+ exit(1);
+
seccomp_test_stub_data->mctx_offset = 1;
set_sigstack(seccomp_test_stub_data->sigstack,
@@ -272,17 +276,17 @@ static int __init seccomp_helper(void *data)
sa.sa_sigaction = (void *) sigsys_handler;
sa.sa_restorer = NULL;
if (sigaction(SIGSYS, &sa, NULL) < 0)
- exit(1);
+ exit(2);
prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
if (syscall(__NR_seccomp, SECCOMP_SET_MODE_FILTER,
SECCOMP_FILTER_FLAG_TSYNC, &prog) != 0)
- exit(2);
+ exit(3);
sleep(0);
/* Never reached. */
- _exit(3);
+ _exit(4);
}
static bool __init init_seccomp(void)
--
2.49.0
^ permalink raw reply related [flat|nested] 8+ messages in thread