Skip to contentSkip to conversation

Drop CAP_FSETID if client needs to kill setuid/setgid bits

PATCH In review Vivek Goyal 2 participants 13 messages Aug 13, 2019 – Aug 21, 2019 Arm/actions semi architecture

lore
Share & tools
lore
#0 lore raw

cover Drop CAP_FSETID if client needs to kill setuid/setgid bits

If a file has setuid/setuid bit set and a writer writes to file without having CAP_FSETID capability, kernel clears setuid/setgid bit on file.

pjdfstest test chmod/12.t tests for this. With moving to 5.3 kernel and cache=none this test fails.

Now Miklos has introducd a commit where if client thinks that setuid/setgid bit should be cleared, it sets FUSE_KILL_PRIV flag in fuse_write_in->write_flags. This is an indication to daemon to clear setuid/setgid bit atomically.

So drop CAP_FSETID capability and then proceed with write and that should automatically clear setuid bit.

Vivek Goyal (4):
virtiofsd: Fix number of padding bits in fuse_file_info
virtiofsd: Use macros for write_flag parsing
virtiofsd: Parse flag FUSE_WRITE_KILL_PRIV
virtiofsd: Drop CAP_FSETID if client asked for it

 contrib/virtiofsd/Makefile.objs    |   2 +
 contrib/virtiofsd/fuse_common.h    |   5 +-
 contrib/virtiofsd/fuse_kernel.h    |   1 +
 contrib/virtiofsd/fuse_lowlevel.c  |   6 +-
 contrib/virtiofsd/passthrough_ll.c | 127 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.c        |   2 +
 6 files changed, 140 insertions(+), 3 deletions(-)
Signature · 1 line
2.17.2
#1 lore raw

PATCH 1/4 virtiofsd: Fix number of padding bits in fuse_file_info

Currently we have 27 padding bits while there are 6 bit fields. I suspect this is wrong. We probably are trying to aling to 32 bits and hence padding bits should be 26 instead.

---
 contrib/virtiofsd/fuse_common.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/virtiofsd/fuse_common.h +1 -1
index 6419f66470..4d95f6f28b 100644
@@ -65,7 +65,7 @@ struct fuse_file_info {
 	unsigned int flock_release : 1;
 
 	/** Padding.  Do not use*/
-	unsigned int padding : 27;
+	unsigned int padding : 26;
 
 	/** File handle.  May be filled in by filesystem in open().
 	    Available in all other file operations */
Signature · 1 line
2.17.2
Reviewed-by on v1 1/4
#2 lore raw

Re: [Virtio-fs] [PATCH 1/4] virtiofsd: Fix number of padding bits in fuse_file_info

Quoted 5 lines · * Vivek Goyal ([email protected]) wrote:

Currently we have 27 padding bits while there are 6 bit fields. I suspect this is wrong. We probably are trying to aling to 32 bits and hence padding bits should be 26 instead.

Yeh, I reckon this was broken upstream by 19accdf ~6 years ago; whether it's fixable upstream I don't know because they have API requierments since it's a library.

Still, it's definitely wrong so:

Quoted 20 lines
---
 contrib/virtiofsd/fuse_common.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/virtiofsd/fuse_common.h +1 -1
index 6419f66470..4d95f6f28b 100644
@@ -65,7 +65,7 @@ struct fuse_file_info {
 	unsigned int flock_release : 1;
 
 	/** Padding.  Do not use*/
-	unsigned int padding : 27;
+	unsigned int padding : 26;
 
 	/** File handle.  May be filled in by filesystem in open().
 	    Available in all other file operations */
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#3 lore raw

Re: [Virtio-fs] [PATCH 1/4] virtiofsd: Fix number of padding bits in fuse_file_info

Quoted 7 lines · * Dr. David Alan Gilbert ([email protected]) wrote:
Quoted 5 lines · * Vivek Goyal ([email protected]) wrote:

Currently we have 27 padding bits while there are 6 bit fields. I suspect this is wrong. We probably are trying to aling to 32 bits and hence padding bits should be 26 instead.

Quoted 7 lines

Yeh, I reckon this was broken upstream by 19accdf ~6 years ago; whether it's fixable upstream I don't know because they have API requierments since it's a library.

Still, it's definitely wrong so:

Actually I found this upstream bug:
https://github.com/libfuse/libfuse/issues/396

which is a 'wontfix'; I think the padding only really exists because of the ABI requirements; so there's not much point in us changing it. We could just drop padding.

Dave

Quoted 15 lines
Quoted 15 lines
---
 contrib/virtiofsd/fuse_common.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
contrib/virtiofsd/fuse_common.h +1 -1
index 6419f66470..4d95f6f28b 100644
@@ -65,7 +65,7 @@ struct fuse_file_info {
 	unsigned int flock_release : 1;
 
 	/** Padding.  Do not use*/
-	unsigned int padding : 27;
+	unsigned int padding : 26;
 
Quoted 7 lines
Quoted 5 lines
contrib/virtiofsd/fuse_common.h +0 -0
 	/** File handle.  May be filled in by filesystem in open().
 	    Available in all other file operations */
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#4 lore raw

PATCH 2/4 virtiofsd: Use macros for write_flag parsing

In reply to Vivek Goyal

Use macros instead of hard coded bit positions.

---
 contrib/virtiofsd/fuse_lowlevel.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
contrib/virtiofsd/fuse_lowlevel.c +2 -2
index 8d3f141d23..417f99e8dc 100644
@@ -1161,7 +1161,7 @@ static void do_write(fuse_req_t req, fuse_ino_t nodeid,
 
 	memset(&fi, 0, sizeof(fi));
 	fi.fh = arg->fh;
-	fi.writepage = (arg->write_flags & 1) != 0;
+	fi.writepage = (arg->write_flags & FUSE_WRITE_CACHE) != 0;
 
 	if (!compat) {
 		fi.lock_owner = arg->lock_owner;
@@ -1208,7 +1208,7 @@ static void do_write_buf(fuse_req_t req, fuse_ino_t nodeid,
 	}
 
 	fi.fh = arg->fh;
-	fi.writepage = arg->write_flags & 1;
+	fi.writepage = !!(arg->write_flags & FUSE_WRITE_CACHE);
 
         if (ibufv->count == 1) {
 		assert(!(tmpbufv.buf[0].flags & FUSE_BUF_IS_FD));
Signature · 1 line
2.17.2
Reviewed-by on v1 2/4
#5 lore raw

Re: [Virtio-fs] [PATCH 2/4] virtiofsd: Use macros for write_flag parsing

Quoted 3 lines · * Vivek Goyal ([email protected]) wrote:

Use macros instead of hard coded bit positions.

OK, this is part of upstream 1b7d2b8

Quoted 29 lines
---
 contrib/virtiofsd/fuse_lowlevel.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
contrib/virtiofsd/fuse_lowlevel.c +2 -2
index 8d3f141d23..417f99e8dc 100644
@@ -1161,7 +1161,7 @@ static void do_write(fuse_req_t req, fuse_ino_t nodeid,
 
 	memset(&fi, 0, sizeof(fi));
 	fi.fh = arg->fh;
-	fi.writepage = (arg->write_flags & 1) != 0;
+	fi.writepage = (arg->write_flags & FUSE_WRITE_CACHE) != 0;
 
 	if (!compat) {
 		fi.lock_owner = arg->lock_owner;
@@ -1208,7 +1208,7 @@ static void do_write_buf(fuse_req_t req, fuse_ino_t nodeid,
 	}
 
 	fi.fh = arg->fh;
-	fi.writepage = arg->write_flags & 1;
+	fi.writepage = !!(arg->write_flags & FUSE_WRITE_CACHE);
 
         if (ibufv->count == 1) {
 		assert(!(tmpbufv.buf[0].flags & FUSE_BUF_IS_FD));
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#6 lore raw

PATCH 3/4 virtiofsd: Parse flag FUSE_WRITE_KILL_PRIV

In reply to Vivek Goyal

Caller can set FUSE_WRITE_KILL_PRIV in write_flags. Parse it and pass it to the filesystem.

---
 contrib/virtiofsd/fuse_common.h   | 5 ++++-
 contrib/virtiofsd/fuse_kernel.h   | 1 +
 contrib/virtiofsd/fuse_lowlevel.c | 2 ++
 3 files changed, 7 insertions(+), 1 deletion(-)
contrib/virtiofsd/fuse_common.h +4 -1
index 4d95f6f28b..5f4af98999 100644
@@ -64,8 +64,11 @@ struct fuse_file_info {
 	   May only be set in ->release(). */
 	unsigned int flock_release : 1;
 
+	/* Indicates that suid/sgid bits should be removed upon write */
+	unsigned int kill_priv : 1;
+
 	/** Padding.  Do not use*/
-	unsigned int padding : 26;
+	unsigned int padding : 25;
 
 	/** File handle.  May be filled in by filesystem in open().
 	    Available in all other file operations */
contrib/virtiofsd/fuse_kernel.h +1 -0
index d477c70028..d2b7ccf96b 100644
@@ -325,6 +325,7 @@ struct fuse_file_lock {
  */
 #define FUSE_WRITE_CACHE	(1 << 0)
 #define FUSE_WRITE_LOCKOWNER	(1 << 1)
+#define FUSE_WRITE_KILL_PRIV    (1 << 2)
 
 /**
  * Read flags
contrib/virtiofsd/fuse_lowlevel.c +2 -0
index 417f99e8dc..ce27a6dd16 100644
@@ -1162,6 +1162,7 @@ static void do_write(fuse_req_t req, fuse_ino_t nodeid,
 	memset(&fi, 0, sizeof(fi));
 	fi.fh = arg->fh;
 	fi.writepage = (arg->write_flags & FUSE_WRITE_CACHE) != 0;
+	fi.kill_priv = !!(arg->write_flags & FUSE_WRITE_KILL_PRIV);
 
 	if (!compat) {
 		fi.lock_owner = arg->lock_owner;
@@ -1209,6 +1210,7 @@ static void do_write_buf(fuse_req_t req, fuse_ino_t nodeid,
 
 	fi.fh = arg->fh;
 	fi.writepage = !!(arg->write_flags & FUSE_WRITE_CACHE);
+	fi.kill_priv = !!(arg->write_flags & FUSE_WRITE_KILL_PRIV);
 
         if (ibufv->count == 1) {
 		assert(!(tmpbufv.buf[0].flags & FUSE_BUF_IS_FD));
Signature · 1 line
2.17.2
Reviewed-by on v1 3/4
#7 lore raw

Re: [Virtio-fs] [PATCH 3/4] virtiofsd: Parse flag FUSE_WRITE_KILL_PRIV

Quoted 4 lines · * Vivek Goyal ([email protected]) wrote:

Caller can set FUSE_WRITE_KILL_PRIV in write_flags. Parse it and pass it to the filesystem.

Quoted 58 lines
---
 contrib/virtiofsd/fuse_common.h   | 5 ++++-
 contrib/virtiofsd/fuse_kernel.h   | 1 +
 contrib/virtiofsd/fuse_lowlevel.c | 2 ++
 3 files changed, 7 insertions(+), 1 deletion(-)
contrib/virtiofsd/fuse_common.h +4 -1
index 4d95f6f28b..5f4af98999 100644
@@ -64,8 +64,11 @@ struct fuse_file_info {
 	   May only be set in ->release(). */
 	unsigned int flock_release : 1;
 
+	/* Indicates that suid/sgid bits should be removed upon write */
+	unsigned int kill_priv : 1;
+
 	/** Padding.  Do not use*/
-	unsigned int padding : 26;
+	unsigned int padding : 25;
 
 	/** File handle.  May be filled in by filesystem in open().
 	    Available in all other file operations */
contrib/virtiofsd/fuse_kernel.h +1 -0
index d477c70028..d2b7ccf96b 100644
@@ -325,6 +325,7 @@ struct fuse_file_lock {
  */
 #define FUSE_WRITE_CACHE	(1 << 0)
 #define FUSE_WRITE_LOCKOWNER	(1 << 1)
+#define FUSE_WRITE_KILL_PRIV    (1 << 2)
 
 /**
  * Read flags
contrib/virtiofsd/fuse_lowlevel.c +2 -0
index 417f99e8dc..ce27a6dd16 100644
@@ -1162,6 +1162,7 @@ static void do_write(fuse_req_t req, fuse_ino_t nodeid,
 	memset(&fi, 0, sizeof(fi));
 	fi.fh = arg->fh;
 	fi.writepage = (arg->write_flags & FUSE_WRITE_CACHE) != 0;
+	fi.kill_priv = !!(arg->write_flags & FUSE_WRITE_KILL_PRIV);
 
 	if (!compat) {
 		fi.lock_owner = arg->lock_owner;
@@ -1209,6 +1210,7 @@ static void do_write_buf(fuse_req_t req, fuse_ino_t nodeid,
 
 	fi.fh = arg->fh;
 	fi.writepage = !!(arg->write_flags & FUSE_WRITE_CACHE);
+	fi.kill_priv = !!(arg->write_flags & FUSE_WRITE_KILL_PRIV);
 
         if (ibufv->count == 1) {
 		assert(!(tmpbufv.buf[0].flags & FUSE_BUF_IS_FD));
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#8 lore raw

PATCH 4/4 virtiofsd: Drop CAP_FSETID if client asked for it

In reply to Vivek Goyal

If client requested killing setuid/setgid bits on file being written, drop CAP_FSETID capability so that setuid/setgid bits are cleared upon write automatically.

pjdfstest chown/12.t needs this.

---
 contrib/virtiofsd/Makefile.objs    |   2 +
 contrib/virtiofsd/passthrough_ll.c | 127 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.c        |   2 +
 3 files changed, 131 insertions(+)
contrib/virtiofsd/Makefile.objs +2 -0
index 941b19f18e..25f1e8dd73 100644
@@ -11,3 +11,5 @@ virtiofsd-obj-y = buffer.o \
 
 seccomp.o-cflags := $(SECCOMP_CFLAGS)
 seccomp.o-libs := $(SECCOMP_LIBS)
+
+passthrough_ll.o-libs += -lcap
contrib/virtiofsd/passthrough_ll.c +127 -0

157 lines not on this page · raw message

contrib/virtiofsd/seccomp.c +2 -0
index 5f1c873b82..7384ebee0a 100644
@@ -78,6 +78,8 @@ static const int syscall_whitelist[] = {
 	SCMP_SYS(unlinkat),
 	SCMP_SYS(utimensat),
 	SCMP_SYS(write),
+	SCMP_SYS(capget),
+	SCMP_SYS(capset),
 };
 
 /* Syscalls used when --syslog is enabled */
Signature · 1 line
2.17.2
#9 lore raw

Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Drop CAP_FSETID if client asked for it

Quoted 7 lines · * Vivek Goyal ([email protected]) wrote:

If client requested killing setuid/setgid bits on file being written, drop CAP_FSETID capability so that setuid/setgid bits are cleared upon write automatically.

pjdfstest chown/12.t needs this.

Some high level questions:
a) This is going to be *really* expensive isn't it - that's a lot of
syscalls for each write

b) and isn't this the normal case, i.e. a non-root guest writing to a
normal file?

c) I guess an unpriv virtiofsd still works OK because the cap is
already gone.

d) How is this thread safe? My reading is cap_set_proc is process
wide; you'd have other writesin other threads doing non-kill-priv
writes; also you'd have multiple threads doing (clear) write (set) -
that feels like they'd interact badly.

e) I think you could share some stuff between the drop and the
restore; for example you only ever need to do cap_from_name once
in the entire run. You can probably also do teh cap_get_proc when
you drop it, keep that 'caps' around and then set it again.

Dave

Quoted 198 lines
---
 contrib/virtiofsd/Makefile.objs    |   2 +
 contrib/virtiofsd/passthrough_ll.c | 127 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.c        |   2 +
 3 files changed, 131 insertions(+)
contrib/virtiofsd/Makefile.objs +2 -0
index 941b19f18e..25f1e8dd73 100644
@@ -11,3 +11,5 @@ virtiofsd-obj-y = buffer.o \
 
 seccomp.o-cflags := $(SECCOMP_CFLAGS)
 seccomp.o-libs := $(SECCOMP_LIBS)
+
+passthrough_ll.o-libs += -lcap
contrib/virtiofsd/passthrough_ll.c +127 -0
index 321bbb20be..412663653a 100644
@@ -56,6 +56,7 @@
 #include <sys/mman.h>
 #include <sys/socket.h>
 #include <sys/un.h>
+#include <sys/capability.h>
 
 #include "ireg.h"
 #include <sys/mount.h>
@@ -230,6 +231,115 @@ static struct lo_data *lo_data(fuse_req_t req)
 	return (struct lo_data *) fuse_req_userdata(req);
 }
 
+/* Helpers for dropping and regaining effective capabilities. Returns 0
+ * on success, error otherwise  */
+static int drop_effective_cap(const char *cap_name, bool *cap_dropped)
+{
+	cap_t caps;
+	cap_value_t cap;
+	cap_flag_value_t cap_value;
+	int ret = 0;
+
+	ret = cap_from_name(cap_name, &cap);
+	if (ret == -1) {
+		ret = errno;
+		fuse_err("cap_from_name(%s) failed:%s\n", cap_name,
+			 strerror(errno));
+		goto out;
+	}
+
+	if (!CAP_IS_SUPPORTED(cap)) {
+		fuse_err("capability(%s) is not supported\n", cap_name);
+		return EINVAL;
+	}
+
+	caps = cap_get_proc();
+	if (caps == NULL) {
+		ret = errno;
+		fuse_err("cap_get_proc() failed\n");
+		goto out;
+	}
+
+	if (cap_get_flag(caps, cap, CAP_EFFECTIVE, &cap_value) == -1) {
+		ret = errno;
+		fuse_err("cap_get_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	/* We dont have this capability in effective set already. */
+	if (cap_value == CAP_CLEAR) {
+		ret = 0;
+		goto out_cap_free;
+	}
+
+	if (cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap, CAP_CLEAR) == -1) {
+		ret = errno;
+		fuse_err("cap_set_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	if (cap_set_proc(caps) == -1) {
+		ret = errno;
+		fuse_err("cap_set_procs() failed\n");
+		goto out_cap_free;
+	}
+
+	ret = 0;
+	if (cap_dropped)
+		*cap_dropped = true;
+
+out_cap_free:
+	cap_free(caps);
+out:
+	return ret;
+}
+
+static int gain_effective_cap(const char *cap_name)
+{
+	cap_t caps;
+	cap_value_t cap;
+	int ret = 0;
+
+	ret = cap_from_name(cap_name, &cap);
+	if (ret == -1) {
+		ret = errno;
+		fuse_err("cap_from_name(%s) failed:%s\n", cap_name,
+			 strerror(errno));
+		goto out;
+	}
+
+	if (!CAP_IS_SUPPORTED(cap)) {
+		fuse_err("capability(%s) is not supported\n", cap_name);
+		return EINVAL;
+	}
+
+	caps = cap_get_proc();
+	if (caps == NULL) {
+		ret = errno;
+		fuse_err("cap_get_proc() failed\n");
+		goto out;
+	}
+
+
+	if (cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap, CAP_SET) == -1) {
+		ret = errno;
+		fuse_err("cap_set_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	if (cap_set_proc(caps) == -1) {
+		ret = errno;
+		fuse_err("cap_set_procs() failed\n");
+		goto out_cap_free;
+	}
+	ret = 0;
+
+out_cap_free:
+	cap_free(caps);
+out:
+	return ret;
+}
+
 static void lo_map_init(struct lo_map *map)
 {
 	map->elems = NULL;
@@ -2014,6 +2124,7 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 	ssize_t res;
 	struct fuse_bufvec out_buf = FUSE_BUFVEC_INIT(fuse_buf_size(in_buf));
 	struct lo_data *lo = lo_data(req);
+	bool cap_fsetid_dropped = false;
 
 	out_buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK;
 	out_buf.buf[0].fd = lo_fi_fd(req, fi);
@@ -2023,6 +2134,16 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 		fuse_debug("lo_write_buf(ino=%" PRIu64 ", size=%zd, off=%lu)\n",
 			   ino, out_buf.buf[0].size, (unsigned long) off);
 
+	/*
+	 * If kill_priv is set, drop CAP_FSETID which should lead to kernel
+	 * clearing setuid/setgid on file.
+	 */
+	if (fi->kill_priv) {
+		res = drop_effective_cap("cap_fsetid", &cap_fsetid_dropped);
+		if (res != 0)
+			fuse_reply_err(req, res);
+	}
+
 	res = fuse_buf_copy(req, &out_buf, in_buf, 0);
 	if(res < 0) {
 		fuse_reply_err(req, -res);
@@ -2037,6 +2158,12 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 
 		fuse_reply_write(req, (size_t) res);
 	}
+
+	if (cap_fsetid_dropped) {
+		res = gain_effective_cap("cap_fsetid");
+		if (res)
+			fuse_err("Failed to gain CAP_FSETID\n");
+	}
 }
 
 static void lo_statfs(fuse_req_t req, fuse_ino_t ino)
contrib/virtiofsd/seccomp.c +2 -0
index 5f1c873b82..7384ebee0a 100644
@@ -78,6 +78,8 @@ static const int syscall_whitelist[] = {
 	SCMP_SYS(unlinkat),
 	SCMP_SYS(utimensat),
 	SCMP_SYS(write),
+	SCMP_SYS(capget),
+	SCMP_SYS(capset),
 };
 
 /* Syscalls used when --syslog is enabled */
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#10 lore raw

Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Drop CAP_FSETID if client asked for it

Quoted 7 lines · Dr. David Alan Gilbert
Quoted 6 lines · * Vivek Goyal ([email protected]) wrote:

If client requested killing setuid/setgid bits on file being written, drop CAP_FSETID capability so that setuid/setgid bits are cleared upon write automatically.

pjdfstest chown/12.t needs this.

Quoted 8 lines
Quoted 1 line

Some high level questions:
a) This is going to be *really* expensive isn't it - that's a lot of
syscalls for each write

b) and isn't this the normal case, i.e. a non-root guest writing to a
normal file?

Hi David,

FUSE_WRITE_KILL_PRIV flag will not be set frequently. It is set only if a write happens to a file which has setuid/setgid bit set. Very few files in filesystem typically have this bit set and for those also it will be set only for first write and not subsequent writes.

So I will not be worried about performance impact of it.

Quoted 8 lines

c) I guess an unpriv virtiofsd still works OK because the cap is
already gone.

d) How is this thread safe? My reading is cap_set_proc is process
wide; you'd have other writesin other threads doing non-kill-priv
writes; also you'd have multiple threads doing (clear) write (set) -
that feels like they'd interact badly.

man page of capabilities says that its a per thread property.

**************************
Starting with kernel 2.2, Linux divides the privileges traditionally associated with superuser into distinct units, known as capabilities,
which can be independently enabled and disabled. Capabilities are a
per-thread attribute.
**************************

So dropping capability per thread and enabling it back should be fine.

Quoted 5 lines

e) I think you could share some stuff between the drop and the
restore; for example you only ever need to do cap_from_name once
in the entire run. You can probably also do teh cap_get_proc when
you drop it, keep that 'caps' around and then set it again.

Given its not a very frequent operation, I am not too worried about it. It will just make code structure little more complicated.

But if you have strong opinion about it, I am open to change it.

Thanks
Vivek

Quoted 204 lines

Dave

Quoted 198 lines
---
 contrib/virtiofsd/Makefile.objs    |   2 +
 contrib/virtiofsd/passthrough_ll.c | 127 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.c        |   2 +
 3 files changed, 131 insertions(+)
contrib/virtiofsd/Makefile.objs +2 -0
index 941b19f18e..25f1e8dd73 100644
@@ -11,3 +11,5 @@ virtiofsd-obj-y = buffer.o \
 
 seccomp.o-cflags := $(SECCOMP_CFLAGS)
 seccomp.o-libs := $(SECCOMP_LIBS)
+
+passthrough_ll.o-libs += -lcap
contrib/virtiofsd/passthrough_ll.c +127 -0
index 321bbb20be..412663653a 100644
@@ -56,6 +56,7 @@
 #include <sys/mman.h>
 #include <sys/socket.h>
 #include <sys/un.h>
+#include <sys/capability.h>
 
 #include "ireg.h"
 #include <sys/mount.h>
@@ -230,6 +231,115 @@ static struct lo_data *lo_data(fuse_req_t req)
 	return (struct lo_data *) fuse_req_userdata(req);
 }
 
+/* Helpers for dropping and regaining effective capabilities. Returns 0
+ * on success, error otherwise  */
+static int drop_effective_cap(const char *cap_name, bool *cap_dropped)
+{
+	cap_t caps;
+	cap_value_t cap;
+	cap_flag_value_t cap_value;
+	int ret = 0;
+
+	ret = cap_from_name(cap_name, &cap);
+	if (ret == -1) {
+		ret = errno;
+		fuse_err("cap_from_name(%s) failed:%s\n", cap_name,
+			 strerror(errno));
+		goto out;
+	}
+
+	if (!CAP_IS_SUPPORTED(cap)) {
+		fuse_err("capability(%s) is not supported\n", cap_name);
+		return EINVAL;
+	}
+
+	caps = cap_get_proc();
+	if (caps == NULL) {
+		ret = errno;
+		fuse_err("cap_get_proc() failed\n");
+		goto out;
+	}
+
+	if (cap_get_flag(caps, cap, CAP_EFFECTIVE, &cap_value) == -1) {
+		ret = errno;
+		fuse_err("cap_get_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	/* We dont have this capability in effective set already. */
+	if (cap_value == CAP_CLEAR) {
+		ret = 0;
+		goto out_cap_free;
+	}
+
+	if (cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap, CAP_CLEAR) == -1) {
+		ret = errno;
+		fuse_err("cap_set_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	if (cap_set_proc(caps) == -1) {
+		ret = errno;
+		fuse_err("cap_set_procs() failed\n");
+		goto out_cap_free;
+	}
+
+	ret = 0;
+	if (cap_dropped)
+		*cap_dropped = true;
+
+out_cap_free:
+	cap_free(caps);
+out:
+	return ret;
+}
+
+static int gain_effective_cap(const char *cap_name)
+{
+	cap_t caps;
+	cap_value_t cap;
+	int ret = 0;
+
+	ret = cap_from_name(cap_name, &cap);
+	if (ret == -1) {
+		ret = errno;
+		fuse_err("cap_from_name(%s) failed:%s\n", cap_name,
+			 strerror(errno));
+		goto out;
+	}
+
+	if (!CAP_IS_SUPPORTED(cap)) {
+		fuse_err("capability(%s) is not supported\n", cap_name);
+		return EINVAL;
+	}
+
+	caps = cap_get_proc();
+	if (caps == NULL) {
+		ret = errno;
+		fuse_err("cap_get_proc() failed\n");
+		goto out;
+	}
+
+
+	if (cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap, CAP_SET) == -1) {
+		ret = errno;
+		fuse_err("cap_set_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	if (cap_set_proc(caps) == -1) {
+		ret = errno;
+		fuse_err("cap_set_procs() failed\n");
+		goto out_cap_free;
+	}
+	ret = 0;
+
+out_cap_free:
+	cap_free(caps);
+out:
+	return ret;
+}
+
 static void lo_map_init(struct lo_map *map)
 {
 	map->elems = NULL;
@@ -2014,6 +2124,7 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 	ssize_t res;
 	struct fuse_bufvec out_buf = FUSE_BUFVEC_INIT(fuse_buf_size(in_buf));
 	struct lo_data *lo = lo_data(req);
+	bool cap_fsetid_dropped = false;
 
 	out_buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK;
 	out_buf.buf[0].fd = lo_fi_fd(req, fi);
@@ -2023,6 +2134,16 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 		fuse_debug("lo_write_buf(ino=%" PRIu64 ", size=%zd, off=%lu)\n",
 			   ino, out_buf.buf[0].size, (unsigned long) off);
 
+	/*
+	 * If kill_priv is set, drop CAP_FSETID which should lead to kernel
+	 * clearing setuid/setgid on file.
+	 */
+	if (fi->kill_priv) {
+		res = drop_effective_cap("cap_fsetid", &cap_fsetid_dropped);
+		if (res != 0)
+			fuse_reply_err(req, res);
+	}
+
 	res = fuse_buf_copy(req, &out_buf, in_buf, 0);
 	if(res < 0) {
 		fuse_reply_err(req, -res);
@@ -2037,6 +2158,12 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 
 		fuse_reply_write(req, (size_t) res);
 	}
+
+	if (cap_fsetid_dropped) {
+		res = gain_effective_cap("cap_fsetid");
+		if (res)
+			fuse_err("Failed to gain CAP_FSETID\n");
+	}
 }
 
 static void lo_statfs(fuse_req_t req, fuse_ino_t ino)
contrib/virtiofsd/seccomp.c +2 -0
index 5f1c873b82..7384ebee0a 100644
@@ -78,6 +78,8 @@ static const int syscall_whitelist[] = {
 	SCMP_SYS(unlinkat),
 	SCMP_SYS(utimensat),
 	SCMP_SYS(write),
+	SCMP_SYS(capget),
+	SCMP_SYS(capset),
 };
 
 /* Syscalls used when --syslog is enabled */
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#11 lore raw

Re: [Virtio-fs] [PATCH 4/4] virtiofsd: Drop CAP_FSETID if client asked for it

Quoted 17 lines · * Vivek Goyal ([email protected]) wrote:
Quoted 7 lines · Dr. David Alan Gilbert
Quoted 6 lines · * Vivek Goyal ([email protected]) wrote:

If client requested killing setuid/setgid bits on file being written, drop CAP_FSETID capability so that setuid/setgid bits are cleared upon write automatically.

pjdfstest chown/12.t needs this.

Quoted 8 lines
Quoted 1 line

Some high level questions:
a) This is going to be *really* expensive isn't it - that's a lot of
syscalls for each write

b) and isn't this the normal case, i.e. a non-root guest writing to a
normal file?

Quoted 8 lines

Hi David,

FUSE_WRITE_KILL_PRIV flag will not be set frequently. It is set only if a write happens to a file which has setuid/setgid bit set. Very few files in filesystem typically have this bit set and for those also it will be set only for first write and not subsequent writes.

So I will not be worried about performance impact of it.

OK, that makes me a lot less worried.

Quoted 12 lines
Quoted 9 lines

c) I guess an unpriv virtiofsd still works OK because the cap is
already gone.

d) How is this thread safe? My reading is cap_set_proc is process
wide; you'd have other writesin other threads doing non-kill-priv
writes; also you'd have multiple threads doing (clear) write (set) -
that feels like they'd interact badly.

man page of capabilities says that its a per thread property.

Quoted 8 lines

**************************
Starting with kernel 2.2, Linux divides the privileges traditionally associated with superuser into distinct units, known as capabilities,
which can be independently enabled and disabled. Capabilities are a
per-thread attribute.
**************************

So dropping capability per thread and enabling it back should be fine.

Huh OK; the man page of cap_get_proc/cap_set_proc say process everywhere.

Quoted 11 lines
Quoted 6 lines

e) I think you could share some stuff between the drop and the
restore; for example you only ever need to do cap_from_name once
in the entire run. You can probably also do teh cap_get_proc when
you drop it, keep that 'caps' around and then set it again.

Given its not a very frequent operation, I am not too worried about it. It will just make code structure little more complicated.

But if you have strong opinion about it, I am open to change it.

No that's fine given your explanation above.

Dave

Quoted 207 lines

Thanks
Vivek

Quoted 204 lines

Dave

Quoted 198 lines
---
 contrib/virtiofsd/Makefile.objs    |   2 +
 contrib/virtiofsd/passthrough_ll.c | 127 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.c        |   2 +
 3 files changed, 131 insertions(+)
contrib/virtiofsd/Makefile.objs +2 -0
index 941b19f18e..25f1e8dd73 100644
@@ -11,3 +11,5 @@ virtiofsd-obj-y = buffer.o \
 
 seccomp.o-cflags := $(SECCOMP_CFLAGS)
 seccomp.o-libs := $(SECCOMP_LIBS)
+
+passthrough_ll.o-libs += -lcap
contrib/virtiofsd/passthrough_ll.c +127 -0
index 321bbb20be..412663653a 100644
@@ -56,6 +56,7 @@
 #include <sys/mman.h>
 #include <sys/socket.h>
 #include <sys/un.h>
+#include <sys/capability.h>
 
 #include "ireg.h"
 #include <sys/mount.h>
@@ -230,6 +231,115 @@ static struct lo_data *lo_data(fuse_req_t req)
 	return (struct lo_data *) fuse_req_userdata(req);
 }
 
+/* Helpers for dropping and regaining effective capabilities. Returns 0
+ * on success, error otherwise  */
+static int drop_effective_cap(const char *cap_name, bool *cap_dropped)
+{
+	cap_t caps;
+	cap_value_t cap;
+	cap_flag_value_t cap_value;
+	int ret = 0;
+
+	ret = cap_from_name(cap_name, &cap);
+	if (ret == -1) {
+		ret = errno;
+		fuse_err("cap_from_name(%s) failed:%s\n", cap_name,
+			 strerror(errno));
+		goto out;
+	}
+
+	if (!CAP_IS_SUPPORTED(cap)) {
+		fuse_err("capability(%s) is not supported\n", cap_name);
+		return EINVAL;
+	}
+
+	caps = cap_get_proc();
+	if (caps == NULL) {
+		ret = errno;
+		fuse_err("cap_get_proc() failed\n");
+		goto out;
+	}
+
+	if (cap_get_flag(caps, cap, CAP_EFFECTIVE, &cap_value) == -1) {
+		ret = errno;
+		fuse_err("cap_get_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	/* We dont have this capability in effective set already. */
+	if (cap_value == CAP_CLEAR) {
+		ret = 0;
+		goto out_cap_free;
+	}
+
+	if (cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap, CAP_CLEAR) == -1) {
+		ret = errno;
+		fuse_err("cap_set_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	if (cap_set_proc(caps) == -1) {
+		ret = errno;
+		fuse_err("cap_set_procs() failed\n");
+		goto out_cap_free;
+	}
+
+	ret = 0;
+	if (cap_dropped)
+		*cap_dropped = true;
+
+out_cap_free:
+	cap_free(caps);
+out:
+	return ret;
+}
+
+static int gain_effective_cap(const char *cap_name)
+{
+	cap_t caps;
+	cap_value_t cap;
+	int ret = 0;
+
+	ret = cap_from_name(cap_name, &cap);
+	if (ret == -1) {
+		ret = errno;
+		fuse_err("cap_from_name(%s) failed:%s\n", cap_name,
+			 strerror(errno));
+		goto out;
+	}
+
+	if (!CAP_IS_SUPPORTED(cap)) {
+		fuse_err("capability(%s) is not supported\n", cap_name);
+		return EINVAL;
+	}
+
+	caps = cap_get_proc();
+	if (caps == NULL) {
+		ret = errno;
+		fuse_err("cap_get_proc() failed\n");
+		goto out;
+	}
+
+
+	if (cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap, CAP_SET) == -1) {
+		ret = errno;
+		fuse_err("cap_set_flag() failed\n");
+		goto out_cap_free;
+	}
+
+	if (cap_set_proc(caps) == -1) {
+		ret = errno;
+		fuse_err("cap_set_procs() failed\n");
+		goto out_cap_free;
+	}
+	ret = 0;
+
+out_cap_free:
+	cap_free(caps);
+out:
+	return ret;
+}
+
 static void lo_map_init(struct lo_map *map)
 {
 	map->elems = NULL;
@@ -2014,6 +2124,7 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 	ssize_t res;
 	struct fuse_bufvec out_buf = FUSE_BUFVEC_INIT(fuse_buf_size(in_buf));
 	struct lo_data *lo = lo_data(req);
+	bool cap_fsetid_dropped = false;
 
 	out_buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK;
 	out_buf.buf[0].fd = lo_fi_fd(req, fi);
@@ -2023,6 +2134,16 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 		fuse_debug("lo_write_buf(ino=%" PRIu64 ", size=%zd, off=%lu)\n",
 			   ino, out_buf.buf[0].size, (unsigned long) off);
 
+	/*
+	 * If kill_priv is set, drop CAP_FSETID which should lead to kernel
+	 * clearing setuid/setgid on file.
+	 */
+	if (fi->kill_priv) {
+		res = drop_effective_cap("cap_fsetid", &cap_fsetid_dropped);
+		if (res != 0)
+			fuse_reply_err(req, res);
+	}
+
 	res = fuse_buf_copy(req, &out_buf, in_buf, 0);
 	if(res < 0) {
 		fuse_reply_err(req, -res);
@@ -2037,6 +2158,12 @@ static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 
 		fuse_reply_write(req, (size_t) res);
 	}
+
+	if (cap_fsetid_dropped) {
+		res = gain_effective_cap("cap_fsetid");
+		if (res)
+			fuse_err("Failed to gain CAP_FSETID\n");
+	}
 }
 
 static void lo_statfs(fuse_req_t req, fuse_ino_t ino)
contrib/virtiofsd/seccomp.c +2 -0
index 5f1c873b82..7384ebee0a 100644
@@ -78,6 +78,8 @@ static const int syscall_whitelist[] = {
 	SCMP_SYS(unlinkat),
 	SCMP_SYS(utimensat),
 	SCMP_SYS(write),
+	SCMP_SYS(capget),
+	SCMP_SYS(capset),
 };
 
 /* Syscalls used when --syslog is enabled */
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK

#12 lore raw
In reply to Vivek Goyal
Quoted 6 lines · * Vivek Goyal ([email protected]) wrote:

If a file has setuid/setuid bit set and a writer writes to file without having CAP_FSETID capability, kernel clears setuid/setgid bit on file.

pjdfstest test chmod/12.t tests for this. With moving to 5.3 kernel and cache=none this test fails.

Quoted 7 lines

Now Miklos has introducd a commit where if client thinks that setuid/setgid bit should be cleared, it sets FUSE_KILL_PRIV flag in fuse_write_in->write_flags. This is an indication to daemon to clear setuid/setgid bit atomically.

So drop CAP_FSETID capability and then proceed with write and that should automatically clear setuid bit.

1,2,3 added to my world.
4 still to be discussed

Quoted 6 lines

Vivek Goyal (4):
virtiofsd: Fix number of padding bits in fuse_file_info
virtiofsd: Use macros for write_flag parsing
virtiofsd: Parse flag FUSE_WRITE_KILL_PRIV
virtiofsd: Drop CAP_FSETID if client asked for it

Quoted 11 lines
 contrib/virtiofsd/Makefile.objs    |   2 +
 contrib/virtiofsd/fuse_common.h    |   5 +-
 contrib/virtiofsd/fuse_kernel.h    |   1 +
 contrib/virtiofsd/fuse_lowlevel.c  |   6 +-
 contrib/virtiofsd/passthrough_ll.c | 127 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.c        |   2 +
 6 files changed, 140 insertions(+), 3 deletions(-)
Signature · 1 line
2.17.2

--
Dr. David Alan Gilbert / [email protected] / Manchester, UK