[PATCH] fcntl: Add 32bit filesystem mode

Subsystems: file locking (flock() and fcntl()/lockf()), filesystems (vfs and infrastructure), performance events subsystem, the rest

STALE2297d

6 messages, 4 authors, 2020-04-20 · open the first message on its own page

[PATCH] fcntl: Add 32bit filesystem mode

From: Linus Walleij <hidden>
Date: 2020-03-31 13:37:50

It was brought to my attention that this bug from 2018 was
still unresolved: 32 bit emulators like QEMU were given
64 bit hashes when running 32 bit emulation on 64 bit systems.

This adds a fcntl() operation to set the underlying filesystem
into 32bit mode even if the file hanle was opened using 64bit
mode without the compat syscalls.

Programs that need the 32 bit file system behavior need to
issue a fcntl() system call such as in this example:

  #define F_SET_FILE_32BIT_FS (1024 + 15)

  int main(int argc, char** argv) {
    DIR* dir;
    int err;
    int fd;

    dir = opendir("/boot");
    fd = dirfd(dir);
    err = fcntl(fd, F_SET_FILE_32BIT_FS);
    if (err) {
      printf("fcntl() failed! err=%d\n", err);
      return 1;
    }
    printf("dir=%p\n", dir);
    printf("readdir(dir)=%p\n", readdir(dir));
    printf("errno=%d: %s\n", errno, strerror(errno));
    return 0;
  }

This can be pretty hard to test since C libraries and linux
userspace security extensions aggressively filter the parameters
that are passed down and allowed to commit into actual system
calls.

Cc: Florian Weimer <redacted>
Cc: Peter Maydell <redacted>
Cc: Andy Lutomirski <luto@kernel.org>
Suggested-by: Theodore Ts'o <tytso@mit.edu>
Link: https://bugs.launchpad.net/qemu/+bug/1805913
Link: https://lore.kernel.org/lkml/87bm56vqg4.fsf@mid.deneb.enyo.de/
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=205957
Signed-off-by: Linus Walleij <redacted>
---
 fs/fcntl.c                       | 4 ++++
 include/uapi/linux/fcntl.h       | 9 +++++++++
 tools/include/uapi/linux/fcntl.h | 9 +++++++++
 tools/perf/trace/beauty/fcntl.c  | 3 ++-
 4 files changed, 24 insertions(+), 1 deletion(-)
diff --git a/fs/fcntl.c b/fs/fcntl.c
index 2e4c0fa2074b..d194b1265bd4 100644
--- a/fs/fcntl.c
+++ b/fs/fcntl.c
@@ -426,6 +426,10 @@ static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
 	case F_SET_FILE_RW_HINT:
 		err = fcntl_rw_hint(filp, cmd, arg);
 		break;
+	case F_SET_FILE_32BIT_FS:
+		filp->f_mode |= FMODE_32BITHASH;
+		err = 0;
+		break;
 	default:
 		break;
 	}
diff --git a/include/uapi/linux/fcntl.h b/include/uapi/linux/fcntl.h
index ca88b7bce553..b9ad934147e8 100644
--- a/include/uapi/linux/fcntl.h
+++ b/include/uapi/linux/fcntl.h
@@ -73,6 +73,15 @@
  */
 #define RWF_WRITE_LIFE_NOT_SET	RWH_WRITE_LIFE_NOT_SET
 
+/*
+ * This instructs the kernel to provide 32bit semantics (such as hashes) from
+ * the file system layer, when running a userland that depend on 32bit
+ * semantics on a kernel that supports 64bit userland, but does not use the
+ * compat ioctl() for e.g. open(), so that the kernel would otherwise assume
+ * that the userland process is capable of dealing with 64bit semantics.
+ */
+#define F_SET_FILE_32BIT_FS	(F_LINUX_SPECIFIC_BASE + 15)
+
 /*
  * Types of directory notifications that may be requested.
  */
diff --git a/tools/include/uapi/linux/fcntl.h b/tools/include/uapi/linux/fcntl.h
index ca88b7bce553..b9ad934147e8 100644
--- a/tools/include/uapi/linux/fcntl.h
+++ b/tools/include/uapi/linux/fcntl.h
@@ -73,6 +73,15 @@
  */
 #define RWF_WRITE_LIFE_NOT_SET	RWH_WRITE_LIFE_NOT_SET
 
+/*
+ * This instructs the kernel to provide 32bit semantics (such as hashes) from
+ * the file system layer, when running a userland that depend on 32bit
+ * semantics on a kernel that supports 64bit userland, but does not use the
+ * compat ioctl() for e.g. open(), so that the kernel would otherwise assume
+ * that the userland process is capable of dealing with 64bit semantics.
+ */
+#define F_SET_FILE_32BIT_FS	(F_LINUX_SPECIFIC_BASE + 15)
+
 /*
  * Types of directory notifications that may be requested.
  */
diff --git a/tools/perf/trace/beauty/fcntl.c b/tools/perf/trace/beauty/fcntl.c
index 56ef83b3d130..da80264678cb 100644
--- a/tools/perf/trace/beauty/fcntl.c
+++ b/tools/perf/trace/beauty/fcntl.c
@@ -94,7 +94,8 @@ size_t syscall_arg__scnprintf_fcntl_arg(char *bf, size_t size, struct syscall_ar
 	    cmd == F_OFD_SETLK || cmd == F_OFD_SETLKW || cmd == F_OFD_GETLK ||
 	    cmd == F_GETOWN_EX || cmd == F_SETOWN_EX ||
 	    cmd == F_GET_RW_HINT || cmd == F_SET_RW_HINT ||
-	    cmd == F_GET_FILE_RW_HINT || cmd == F_SET_FILE_RW_HINT)
+	    cmd == F_GET_FILE_RW_HINT || cmd == F_SET_FILE_RW_HINT ||
+	    cmd == F_SET_FILE_32BIT_FS)
 		return syscall_arg__scnprintf_hex(bf, size, arg);
 
 	return syscall_arg__scnprintf_long(bf, size, arg);
-- 
2.25.1

Re: [PATCH] fcntl: Add 32bit filesystem mode

From: Peter Maydell <hidden>
Date: 2020-04-20 11:19:32

On Tue, 31 Mar 2020 at 14:37, Linus Walleij [off-list ref] wrote:
It was brought to my attention that this bug from 2018 was
still unresolved: 32 bit emulators like QEMU were given
64 bit hashes when running 32 bit emulation on 64 bit systems.

This adds a fcntl() operation to set the underlying filesystem
into 32bit mode even if the file hanle was opened using 64bit
mode without the compat syscalls.

Programs that need the 32 bit file system behavior need to
issue a fcntl() system call such as in this example:

  #define F_SET_FILE_32BIT_FS (1024 + 15)

  int main(int argc, char** argv) {
    DIR* dir;
    int err;
    int fd;

    dir = opendir("/boot");
    fd = dirfd(dir);
    err = fcntl(fd, F_SET_FILE_32BIT_FS);
    if (err) {
      printf("fcntl() failed! err=%d\n", err);
      return 1;
    }
    printf("dir=%p\n", dir);
    printf("readdir(dir)=%p\n", readdir(dir));
    printf("errno=%d: %s\n", errno, strerror(errno));
    return 0;
  }
I gave this a try with a modified QEMU, but it doesn't seem
to fix the problem. Here's the relevant chunk of the strace
output from stracing a QEMU that's running a 32-bit guest
binary that issues a getdents64 and fails (it's the 'readdir-bug'
test case from the launchpad bug):

openat(AT_FDCWD, ".", O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_DIRECTORY) = 3
fcntl(3, 0x40f /* F_??? */, 0x3)        = 0
fstat(3, {st_dev=makedev(0, 16), st_ino=4637237, st_mode=S_IFDIR|0755,
st_nlink=12, st_uid=1000, st_gid=1000, st_blksize=8192, st_blocks=8,
st_size=4096, st_atime=1587380917 /*
2020-04-20T11:08:37.756174607+0000 */, st_atime_nsec=756174607,
st_mtime=1587380910 /* 2020-04-20T11:08:30.356230179+0000 */,
st_mtime_nsec=356230179, st_ctime=1587380910 /*
2020-04-20T11:08:30.356230179+0000 */, st_ctime_nsec=356230179}) = 0
fstat(1, {st_dev=makedev(0, 2), st_ino=9017, st_mode=S_IFCHR|0600,
st_nlink=1, st_uid=0, st_gid=0, st_blksize=4096, st_blocks=0,
st_rdev=makedev(5, 1), st_atime=1587381196 /* 2020-04-20T11:13:16+0000
*/, st_atime_nsec=0, st_mtime=1587381196 /* 2020-04-20T11:13:16+0000
*/, st_mtime_nsec=0, st_ctime=1587381042 /*
2020-04-20T11:10:42.484981152+0000 */, st_ctime_nsec=484981152}) = 0
ioctl(1, TCGETS, {c_iflags=0x2502, c_oflags=0x5, c_cflags=0xcbd,
c_lflags=0x8a3b, c_line=0,
c_cc="\x03\x1c\x7f\x15\x04\x00\x01\x00\x11\x13\x1a\x00\x12\x0f\x17\x16\x00\x00\x00"})
= 0
write(1, "dir=0x76128\n", 12)           = 12
getdents64(3, [{d_ino=1, d_off=273341893525646730, d_reclen=24,
d_type=DT_DIR, d_name=".."}, {d_ino=4637239, d_off=849308795555391993,
d_reclen=24, d_type=DT_DIR, d_name="etc"}, {d_ino=4587984,
d_off=1620709961571101518, d_reclen=24, d_type=DT_LNK, d_name="usr"},
{d_ino=4637238, d_off=2787937917159437645, d_reclen=24, d_type=DT_DIR,
d_name="dev"}, {d_ino=4637244, d_off=3015508490233103491, d_reclen=24,
d_type=DT_DIR, d_name="sys"}, {d_ino=4587608,
d_off=3551089360661460833, d_reclen=24, d_type=DT_LNK, d_name="lib"},
{d_ino=4637246, d_off=3857320197951442970, d_reclen=24, d_type=DT_DIR,
d_name="var"}, {d_ino=4637242, d_off=4103122318823701457, d_reclen=24,
d_type=DT_DIR, d_name="proc"}, {d_ino=4587541,
d_off=4252201186220906002, d_reclen=24, d_type=DT_LNK, d_name="bin"},
{d_ino=4637245, d_off=4386533378951587638, d_reclen=24, d_type=DT_DIR,
d_name="tmp"}, {d_ino=4637241, d_off=4883206313583644962, d_reclen=24,
d_type=DT_DIR, d_name="host"}, {d_ino=4637237,
d_off=4941119754928488586, d_reclen=24, d_type=DT_DIR, d_name="."},
{d_ino=4637243, d_off=5301154723342888169, d_reclen=24, d_type=DT_DIR,
d_name="root"}, {d_ino=4587838, d_off=6989908915879243400,
d_reclen=32, d_type=DT_LNK, d_name="lib64"}, {d_ino=4587679,
d_off=7356513223657690979, d_reclen=32, d_type=DT_REG,
d_name="strace.log"}, {d_ino=4587847, d_off=7810090083157553519,
d_reclen=24, d_type=DT_LNK, d_name="sbin"}, {d_ino=4637240,
d_off=8254997891991845677, d_reclen=24, d_type=DT_DIR, d_name="home"},
{d_ino=4637248, d_off=9223372036854775807, d_reclen=24, d_type=DT_DIR,
d_name="virt"}], 32768) = 448
write(1, "readdir(dir)=(nil)\n", 19)    = 19
write(1, "errno=75: Value too large for de"..., 48) = 48
exit_group(0)                           = ?


We open fd 3 to read '.'; we issue the new fcntl, which
succeeds. Then there's some unrelated stuff operating on
stdout. Then we do a getdents64(), but the d_off values
we get back are still 64 bits. The guest binary doesn't
like those, so it fails. My expectation was that we would
get back d_off values here that were in the 32 bit range.

(To be clear, the guest binary here is doing a getdents64(),
which QEMU translates into a host getdents64().)

thanks
-- PMM

Re: [PATCH] fcntl: Add 32bit filesystem mode

From: "Theodore Y. Ts'o" <tytso@mit.edu>
Date: 2020-04-20 15:14:01

On Tue, Mar 31, 2020 at 03:35:36PM +0200, Linus Walleij wrote:
It was brought to my attention that this bug from 2018 was
still unresolved: 32 bit emulators like QEMU were given
64 bit hashes when running 32 bit emulation on 64 bit systems.

This adds a fcntl() operation to set the underlying filesystem
into 32bit mode even if the file hanle was opened using 64bit
mode without the compat syscalls.
s/hanle/handle/

The API that you've proposed as a way to set the 32-bit mode, but
there is no way to clear the 32-bit mode, nor there is a way to get
the current status mode.

My suggestion is to add a flag bit for F_GETFD and F_SETFD (set and
get file descriptor flags).  Currently the only file descriptor flag
is FD_CLOEXEC, so why not add a FD_32BIT_MODE bit?

Cheers,

						- Ted

Re: [PATCH] fcntl: Add 32bit filesystem mode

From: Eric Blake <hidden>
Date: 2020-04-20 15:24:08

On 4/20/20 10:13 AM, Theodore Y. Ts'o wrote:
On Tue, Mar 31, 2020 at 03:35:36PM +0200, Linus Walleij wrote:
quoted
It was brought to my attention that this bug from 2018 was
still unresolved: 32 bit emulators like QEMU were given
64 bit hashes when running 32 bit emulation on 64 bit systems.

This adds a fcntl() operation to set the underlying filesystem
into 32bit mode even if the file hanle was opened using 64bit
mode without the compat syscalls.
s/hanle/handle/

The API that you've proposed as a way to set the 32-bit mode, but
there is no way to clear the 32-bit mode, nor there is a way to get
the current status mode.

My suggestion is to add a flag bit for F_GETFD and F_SETFD (set and
get file descriptor flags).  Currently the only file descriptor flag
is FD_CLOEXEC, so why not add a FD_32BIT_MODE bit?
Also, POSIX is proposing standardizing FD_CLOFORK, which would be 
another file descriptor flag worth considering in Linux (Solaris and BSD 
already have it):

https://www.austingroupbugs.net/view.php?id=1318

It will be interesting to find how much code (wrongly) assumes it can 
use a blind assignment of fcntl(fd, F_SETFD, 1) and thereby accidentally 
wipes out other existing flags, when it should have instead been doing a 
read-modify-write to protect flags other than FD_CLOEXEC.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org

Re: [PATCH] fcntl: Add 32bit filesystem mode

From: Peter Maydell <hidden>
Date: 2020-04-20 15:29:45

On Mon, 20 Apr 2020 at 16:24, Eric Blake [off-list ref] wrote:
It will be interesting to find how much code (wrongly) assumes it can
use a blind assignment of fcntl(fd, F_SETFD, 1) and thereby accidentally
wipes out other existing flags, when it should have instead been doing a
read-modify-write to protect flags other than FD_CLOEXEC.
For instance, a quick grep shows 4 instances of this in QEMU :-)

thanks
-- PMM

Re: [PATCH] fcntl: Add 32bit filesystem mode

From: "Theodore Y. Ts'o" <tytso@mit.edu>
Date: 2020-04-20 17:01:48

On Mon, Apr 20, 2020 at 04:29:32PM +0100, Peter Maydell wrote:
On Mon, 20 Apr 2020 at 16:24, Eric Blake [off-list ref] wrote:
quoted
It will be interesting to find how much code (wrongly) assumes it can
use a blind assignment of fcntl(fd, F_SETFD, 1) and thereby accidentally
wipes out other existing flags, when it should have instead been doing a
read-modify-write to protect flags other than FD_CLOEXEC.
For instance, a quick grep shows 4 instances of this in QEMU :-)
Fortunately, most applications aren't going to be interested in
forcing 32-bit mode for 64-bit applications, QEMU being the notable
exception.  We do need to make sure that for 32-bit applications, we
either make FD_32BIT_MODE a no-op (don't set the bit, and ignore the
bit).  We could allow the bit to be visible for 32-bit applications,
but we would want to disallow clearing the the bit for 32-bit
applications if it was visible.

If we did that, then blind assignments of fcntl(fd, F_SETFD, 1) should
be mostly harmless with respect to the FD_32BIT_MODE bit.

   	      	 	  	       - Ted
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help