Skip to contentSkip to conversation

linux-user: Use memfd for open syscall emulation

PATCH Superseded Rainer Müller 2 participants 6 messages Jul 25, 2022 – Jul 29, 2022

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 linux-user: Use memfd for open syscall emulation

For certain paths in /proc, the open syscall is intercepted and the returned file descriptor points to a temporary file with emulated contents.

If TMPDIR is not accessible or writable for the current user (for example in a read-only mounted chroot or container) tools such as ps from procps may fail unexpectedly. Trying to read one of these paths such as /proc/self/stat would return an error such as ENOENT or EROFS.

To relax the requirement on a writable TMPDIR, use memfd_create() instead to create an anonymous file and return its file descriptor.

---
 linux-user/syscall.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
linux-user/syscall.c +9 -1
index 991b85e6b4..3e4af930ad 100644
@@ -8265,9 +8265,11 @@ static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int
     }
 
     if (fake_open->filename) {
+        int fd, r;
+
+#ifndef CONFIG_MEMFD
         const char *tmpdir;
         char filename[PATH_MAX];
-        int fd, r;
 
         /* create temporary file to map stat to */
         tmpdir = getenv("TMPDIR");
@@ -8279,6 +8281,12 @@ static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int
             return fd;
         }
         unlink(filename);
+#else
+        fd = memfd_create("qemu-open", 0);
+        if (fd < 0) {
+            return fd;
+        }
+#endif
 
         if ((r = fake_open->fill(cpu_env, fd))) {
             int e = errno;
Signature · 1 line
2.25.1
#1 lore raw
Quoted 35 lines · Rainer Müller

For certain paths in /proc, the open syscall is intercepted and the returned file descriptor points to a temporary file with emulated contents.

If TMPDIR is not accessible or writable for the current user (for example in a read-only mounted chroot or container) tools such as ps from procps may fail unexpectedly. Trying to read one of these paths such as /proc/self/stat would return an error such as ENOENT or EROFS.

To relax the requirement on a writable TMPDIR, use memfd_create() instead to create an anonymous file and return its file descriptor.

---
  linux-user/syscall.c | 10 +++++++++-
  1 file changed, 9 insertions(+), 1 deletion(-)
linux-user/syscall.c +3 -1
index 991b85e6b4..3e4af930ad 100644
@@ -8265,9 +8265,11 @@ static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int
      }
 
      if (fake_open->filename) {
+        int fd, r;
+
+#ifndef CONFIG_MEMFD
          const char *tmpdir;
          char filename[PATH_MAX];
-        int fd, r;
 
          /* create temporary file to map stat to */
          tmpdir = getenv("TMPDIR");

linux-user/syscall.c static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int · line 8289

Quoted 9 lines
linux-user/syscall.c +6 -0
@@ -8279,6 +8281,12 @@ static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int
              return fd;
          }
          unlink(filename);
+#else
+        fd = memfd_create("qemu-open", 0);
+        if (fd < 0) {
+            return fd;
+        }
+#endif

Even without CONFIG_MEMFD, we will have the memfd_create function available in util/. I think you should drop the ifdefs like so:

#include "qemu/memfd.h"

    fd = memfd_create(...);
    if (fd < 0) {
        if (errno != ENOSYS) {
            return fd;
        }
        // tmpdir fallback
    }

r~

#2 lore raw

PATCH v2 1/1 linux-user: Use memfd for open syscall emulation

In reply to Rainer Müller

For certain paths in /proc, the open syscall is intercepted and the returned file descriptor points to a temporary file with emulated contents.

If TMPDIR is not accessible or writable for the current user (for example in a read-only mounted chroot or container) tools such as ps from procps may fail unexpectedly. Trying to read one of these paths such as /proc/self/stat would return an error such as ENOENT or EROFS.

To relax the requirement on a writable TMPDIR, use memfd_create() instead to create an anonymous file and return its file descriptor.

---
v2: no more #ifdefs, use stub from util/memfd.c with ENOSYS fallback,
tested with 'strace -e fault=memfd_create'

---
 linux-user/syscall.c | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)
linux-user/syscall.c +14 -8
index 991b85e6b4..7b55726f25 100644
@@ -8269,16 +8269,22 @@ static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int
         char filename[PATH_MAX];
         int fd, r;
 
-        /* create temporary file to map stat to */
-        tmpdir = getenv("TMPDIR");
-        if (!tmpdir)
-            tmpdir = "/tmp";
-        snprintf(filename, sizeof(filename), "%s/qemu-open.XXXXXX", tmpdir);
-        fd = mkstemp(filename);
+        fd = memfd_create("qemu-open", 0);
         if (fd < 0) {
-            return fd;
+            if (errno != ENOSYS) {
+                return fd;
+            }
+            /* create temporary file to map stat to */
+            tmpdir = getenv("TMPDIR");
+            if (!tmpdir)
+                tmpdir = "/tmp";
+            snprintf(filename, sizeof(filename), "%s/qemu-open.XXXXXX", tmpdir);
+            fd = mkstemp(filename);
+            if (fd < 0) {
+                return fd;
+            }
+            unlink(filename);
         }
-        unlink(filename);
 
         if ((r = fake_open->fill(cpu_env, fd))) {
             int e = errno;
Signature · 1 line
2.25.1
#3 lore raw

linux-user/syscall.c static int do_openat(CPUArchState *cpu_env, int dirfd, const char *pathname, int · line 8285

Quoted 9 lines · Rainer Müller
linux-user/syscall.c +9 -0
+            /* create temporary file to map stat to */
+            tmpdir = getenv("TMPDIR");
+            if (!tmpdir)
+                tmpdir = "/tmp";
+            snprintf(filename, sizeof(filename), "%s/qemu-open.XXXXXX", tmpdir);
+            fd = mkstemp(filename);
+            if (fd < 0) {
+                return fd;
+            }

We've been using g_file_open_tmp elsewhere; probably good to follow suit here.

r~

#4 lore raw
Quoted 12 lines · Richard Henderson
Quoted 10 lines · Rainer Müller

+            /* create temporary file to map stat to */ +            tmpdir = getenv("TMPDIR");
+            if (!tmpdir)
+                tmpdir = "/tmp";
+            snprintf(filename, sizeof(filename), "%s/qemu-open.XXXXXX", tmpdir);
+            fd = mkstemp(filename);
+            if (fd < 0) {
+                return fd;
+            }

Quoted 2 lines

We've been using g_file_open_tmp elsewhere; probably good to follow suit here.

That seemed reasonable at first, but with regards to error handling it gets a bit complicated.

The suggested g_file_open_tmp() would leave us with a GError only, but to return something meaningful to the caller we must set errno in this context. As far as I can see, there is no way to convert back to an errno from GError.

With g_file_open_tmp() we could always set the same generic errno, but that would hide the real cause completely. I debugged this problem with this message that was confusing, but at least it gave away a hint:
cat: can't open '/proc/self/stat': Read-only file system

The other option would be to g_assert_true(fd >= 0) and kill the process in case opening the temporary file failed. This also feels wrong, as the caller could still recover from this state and continue.

Rainer

Reviewed-by
#5 lore raw
Quoted 24 lines · Rainer Müller
Quoted 12 lines · Richard Henderson
Quoted 10 lines · Rainer Müller

+            /* create temporary file to map stat to */ +            tmpdir = getenv("TMPDIR");
+            if (!tmpdir)
+                tmpdir = "/tmp";
+            snprintf(filename, sizeof(filename), "%s/qemu-open.XXXXXX", tmpdir);
+            fd = mkstemp(filename);
+            if (fd < 0) {
+                return fd;
+            }

Quoted 2 lines

We've been using g_file_open_tmp elsewhere; probably good to follow suit here.

That seemed reasonable at first, but with regards to error handling it gets a bit complicated.

The suggested g_file_open_tmp() would leave us with a GError only, but to return something meaningful to the caller we must set errno in this context. As far as I can see, there is no way to convert back to an errno from GError.

Quoted 4 lines

With g_file_open_tmp() we could always set the same generic errno, but that would hide the real cause completely. I debugged this problem with this message that was confusing, but at least it gave away a hint:
cat: can't open '/proc/self/stat': Read-only file system

That's a good reason. Let's leave this alone then.

r~