Skip to contentSkip to conversation

linux-user: fix guards for the fsmount(2) syscall series

PATCH Superseded Matt Turner 2 participants 4 messages Jul 26

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 linux-user: fix guards for the fsmount(2) syscall series

The fsopen(), fsconfig(), fsmount() and fspick() implementations are guarded by defined(NR_fsopen) rather than defined(__NR_fsopen). No such macro exists, so the guard is never true and the entire series compiles out. Guests calling any of the four get -ENOSYS, which for example makes systemd's credential setup fail with EXIT_CREDENTIALS for most units.

The strace bits for fsconfig() have the same typo.

Fixes 767c32fe6983 ("linux-user: implement fsmount(2) series of syscalls")
Fixes 6e0aa9f6c731 ("linux-user/strace: add fsmount series of syscalls")
---
 linux-user/strace.c    | 2 +-
 linux-user/strace.list | 2 +-
 linux-user/syscall.c   | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)
linux-user/strace.c +1 -1
index 3a81cc95f4..d861f311f6 100644
@@ -4344,7 +4344,7 @@ print_statx(CPUArchState *cpu_env, const struct syscallname *name,
 }
 #endif
 
-#if defined(TARGET_NR_fsconfig) && defined(NR_fsconfig)
+#ifdef TARGET_NR_fsconfig
 static void
 print_fsconfig_cmd_name(int cmd)
 {
linux-user/strace.list +1 -1
index e363892e0a..3a366b8cac 100644
@@ -1725,7 +1725,7 @@
 #ifdef TARGET_NR_fsopen
 { TARGET_NR_fsopen, "fsopen", "%s(%s,%d)", NULL, NULL },
 #endif
-#if defined(TARGET_NR_fsconfig) && defined(NR_fsconfig)
+#ifdef TARGET_NR_fsconfig
 { TARGET_NR_fsconfig, "fsconfig", NULL, print_fsconfig, NULL },
 #endif
 #ifdef TARGET_NR_fsmount
linux-user/syscall.c +2 -2
index 3da5530d42..740142825d 100644
@@ -9740,7 +9740,7 @@ _syscall5(int, sys_move_mount, int, __from_dfd, const char *, __from_pathname,
            int, __to_dfd, const char *, __to_pathname, unsigned int, flag)
 #endif
 
-#if defined(TARGET_NR_fsopen) && defined(NR_fsopen)
+#if defined(TARGET_NR_fsopen) && defined(__NR_fsopen)
 #define __NR_sys_fsopen __NR_fsopen
 _syscall2(int, sys_fsopen, const char *, fs_name, unsigned int, flags);
 #define __NR_sys_fsconfig __NR_fsconfig
@@ -14485,7 +14485,7 @@ static abi_long do_syscall1(CPUArchState *cpu_env, int num, abi_long arg1,
         return do_map_shadow_stack(cpu_env, arg1, arg2, arg3);
 #endif
 
-#if defined(TARGET_NR_fsopen) && defined(NR_fsopen)
+#if defined(TARGET_NR_fsopen) && defined(__NR_fsopen)
     case TARGET_NR_fsopen:
         {
             p = lock_user_string(arg1);
Signature · 1 line
2.54.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Matt Turner

The fsopen(), fsconfig(), fsmount() and fspick() implementations are guarded by defined(NR_fsopen) rather than defined(__NR_fsopen). No such macro exists, so the guard is never true and the entire series compiles out. Guests calling any of the four get -ENOSYS, which for example makes systemd's credential setup fail with EXIT_CREDENTIALS for most units.

The strace bits for fsconfig() have the same typo.

Quoted 8 lines
Fixes 767c32fe6983 ("linux-user: implement fsmount(2) series of syscalls")
Fixes 6e0aa9f6c731 ("linux-user/strace: add fsmount series of syscalls")
---
  linux-user/strace.c    | 2 +-
  linux-user/strace.list | 2 +-
  linux-user/syscall.c   | 4 ++--
  3 files changed, 4 insertions(+), 4 deletions(-)

Matt, thanks for noticing and fixing!

I'll include it in the next linux-user pull request...

Helge

#2 lore raw
In reply to Matt Turner
Quoted 22 lines · Matt Turner

The fsopen(), fsconfig(), fsmount() and fspick() implementations are guarded by defined(NR_fsopen) rather than defined(__NR_fsopen). No such macro exists, so the guard is never true and the entire series compiles out. Guests calling any of the four get -ENOSYS, which for example makes systemd's credential setup fail with EXIT_CREDENTIALS for most units.

The strace bits for fsconfig() have the same typo.

Fixes 767c32fe6983 ("linux-user: implement fsmount(2) series of syscalls")
Fixes 6e0aa9f6c731 ("linux-user/strace: add fsmount series of syscalls")
---
  linux-user/strace.c    | 2 +-
  linux-user/strace.list | 2 +-
  linux-user/syscall.c   | 4 ++--
  3 files changed, 4 insertions(+), 4 deletions(-)
linux-user/strace.c +0 -0
index 3a81cc95f4..d861f311f6 100644

linux-user/strace.c print_statx(CPUArchState *cpu_env, const struct syscallname *name, · line 4347

Quoted 5 lines
linux-user/strace.c +1 -1
@@ -4344,7 +4344,7 @@ print_statx(CPUArchState *cpu_env, const struct syscallname *name,
  }
  #endif
 
-#if defined(TARGET_NR_fsconfig) && defined(NR_fsconfig)
+#ifdef TARGET_NR_fsconfig

Your patch is basically correct.
But I think instead of removing the (wrong) NR_fsconfig, we should instead depend on the (possibly existing) __NR_fsconfig.

Reason is, that if __NR_fsconfig doesn't exist on the host machine, the constants used in print_fsconfig() like FSCONFIG_SET_FLAG aren't known either, leading to build errors if we don't check against __NR_fsconfig.

I change the patch accordingly.

Helge

Quoted 38 lines
linux-user/strace.c +0 -0
  static void
  print_fsconfig_cmd_name(int cmd)
  {
linux-user/strace.list +1 -1
index e363892e0a..3a366b8cac 100644
@@ -1725,7 +1725,7 @@
  #ifdef TARGET_NR_fsopen
  { TARGET_NR_fsopen, "fsopen", "%s(%s,%d)", NULL, NULL },
  #endif
-#if defined(TARGET_NR_fsconfig) && defined(NR_fsconfig)
+#ifdef TARGET_NR_fsconfig
  { TARGET_NR_fsconfig, "fsconfig", NULL, print_fsconfig, NULL },
  #endif
  #ifdef TARGET_NR_fsmount
linux-user/syscall.c +2 -2
index 3da5530d42..740142825d 100644
@@ -9740,7 +9740,7 @@ _syscall5(int, sys_move_mount, int, __from_dfd, const char *, __from_pathname,
             int, __to_dfd, const char *, __to_pathname, unsigned int, flag)
  #endif
 
-#if defined(TARGET_NR_fsopen) && defined(NR_fsopen)
+#if defined(TARGET_NR_fsopen) && defined(__NR_fsopen)
  #define __NR_sys_fsopen __NR_fsopen
  _syscall2(int, sys_fsopen, const char *, fs_name, unsigned int, flags);
  #define __NR_sys_fsconfig __NR_fsconfig
@@ -14485,7 +14485,7 @@ static abi_long do_syscall1(CPUArchState *cpu_env, int num, abi_long arg1,
          return do_map_shadow_stack(cpu_env, arg1, arg2, arg3);
  #endif
 
-#if defined(TARGET_NR_fsopen) && defined(NR_fsopen)
+#if defined(TARGET_NR_fsopen) && defined(__NR_fsopen)
      case TARGET_NR_fsopen:
          {
              p = lock_user_string(arg1);
#3 lore raw
Quoted 32 lines · Helge Deller
Quoted 22 lines · Matt Turner

The fsopen(), fsconfig(), fsmount() and fspick() implementations are guarded by defined(NR_fsopen) rather than defined(__NR_fsopen). No such macro exists, so the guard is never true and the entire series compiles out. Guests calling any of the four get -ENOSYS, which for example makes systemd's credential setup fail with EXIT_CREDENTIALS for most units.

The strace bits for fsconfig() have the same typo.

Fixes 767c32fe6983 ("linux-user: implement fsmount(2) series of syscalls")
Fixes 6e0aa9f6c731 ("linux-user/strace: add fsmount series of syscalls")

---
  linux-user/strace.c    | 2 +-
  linux-user/strace.list | 2 +-
  linux-user/syscall.c   | 4 ++--

  3 files changed, 4 insertions(+), 4 deletions(-)
linux-user/strace.c +0 -0
index 3a81cc95f4..d861f311f6 100644

linux-user/strace.c print_statx(CPUArchState *cpu_env, const struct syscallname *name, · line 4346

Quoted 4 lines
linux-user/strace.c +1 -1
@@ -4344,7 +4344,7 @@ print_statx(CPUArchState *cpu_env, const struct syscallname *name,
   }
   #endif
-#if defined(TARGET_NR_fsconfig) && defined(NR_fsconfig)
+#ifdef TARGET_NR_fsconfig

Your patch is basically correct.
But I think instead of removing the (wrong) NR_fsconfig, we should instead depend on the (possibly existing) __NR_fsconfig.

Quoted 5 lines

Reason is, that if __NR_fsconfig doesn't exist on the host machine, the constants used in print_fsconfig() like FSCONFIG_SET_FLAG aren't known either, leading to build errors if we don't check against __NR_fsconfig.

I change the patch accordingly.

and pushed to my linux-user git tree:
https://github.com/hdeller/qemu-hppa/commit/4fdc24696cbf1ac6d97075d877cafc9fead4c475

Helge