[xen/f17] limit the size of guest kernels and ramdisks to avoid running out of memory on dom0 during guest boo
myoung
myoung at fedoraproject.org
Sun Oct 28 22:49:00 UTC 2012
commit 55d363215b495e547a1770806678b07088a5a0ca
Author: Michael Young <m.a.young at durham.ac.uk>
Date: Sun Oct 28 22:47:54 2012 +0000
limit the size of guest kernels and ramdisks to avoid running out
of memory on dom0 during guest boot [XSA-25, CVE-2012-4544] (#870414)
pygrub.size.limits.patch | 116 -----------
xen-4.1-testing.23385.patch | 463 +++++++++++++++++++++++++++++++++++++++++++
xen.spec | 11 +-
3 files changed, 471 insertions(+), 119 deletions(-)
---
diff --git a/xen-4.1-testing.23385.patch b/xen-4.1-testing.23385.patch
new file mode 100644
index 0000000..581f7a8
--- /dev/null
+++ b/xen-4.1-testing.23385.patch
@@ -0,0 +1,463 @@
+
+# HG changeset patch
+# User Ian Jackson <Ian.Jackson at eu.citrix.com>
+# Date 1351264204 -3600
+# Node ID 69d1cc78a5bd066ca84019512c6e3d60c1a3fef4
+# Parent a15596a619ed738c163c08d3ec275c8880890c52
+libxc: builder: limit maximum size of kernel/ramdisk.
+
+Allowing user supplied kernels of arbitrary sizes, especially during
+decompression, can swallow up dom0 memory leading to either virtual
+address space exhaustion in the builder process or allocation
+failures/OOM killing of both toolstack and unrelated processes.
+
+We disable these checks when building in a stub domain for pvgrub
+since this uses the guest's own memory and is isolated.
+
+Decompression of gzip compressed kernels and ramdisks has been safe
+since 14954:58205257517d (Xen 3.1.0 onwards).
+
+This is XSA-25 / CVE-2012-4544.
+
+Also make explicit checks for buffer overflows in various
+decompression routines. These were already ruled out due to other
+properties of the code but check them as a belt-and-braces measure.
+
+Signed-off-by: Ian Campbell <ian.campbell at citrix.com>
+Acked-by: Ian Jackson <ian.jackson at eu.citrix.com>
+[ Includes 25589:60f09d1ab1fe for CVE-2012-2625 ]
+
+diff -r a15596a619ed -r 69d1cc78a5bd stubdom/grub/kexec.c
+--- a/stubdom/grub/kexec.c Thu Oct 04 10:44:43 2012 +0200
++++ b/stubdom/grub/kexec.c Fri Oct 26 16:10:04 2012 +0100
+@@ -137,6 +137,10 @@ void kexec(void *kernel, long kernel_siz
+ dom = xc_dom_allocate(xc_handle, cmdline, features);
+ dom->allocate = kexec_allocate;
+
++ /* We are using guest owned memory, therefore no limits. */
++ xc_dom_kernel_max_size(dom, 0);
++ xc_dom_ramdisk_max_size(dom, 0);
++
+ dom->kernel_blob = kernel;
+ dom->kernel_size = kernel_size;
+
+diff -r a15596a619ed -r 69d1cc78a5bd tools/libxc/xc_dom.h
+--- a/tools/libxc/xc_dom.h Thu Oct 04 10:44:43 2012 +0200
++++ b/tools/libxc/xc_dom.h Fri Oct 26 16:10:04 2012 +0100
+@@ -52,6 +52,9 @@ struct xc_dom_image {
+ void *ramdisk_blob;
+ size_t ramdisk_size;
+
++ size_t max_kernel_size;
++ size_t max_ramdisk_size;
++
+ /* arguments and parameters */
+ char *cmdline;
+ uint32_t f_requested[XENFEAT_NR_SUBMAPS];
+@@ -175,6 +178,23 @@ void xc_dom_release_phys(struct xc_dom_i
+ void xc_dom_release(struct xc_dom_image *dom);
+ int xc_dom_mem_init(struct xc_dom_image *dom, unsigned int mem_mb);
+
++/* Set this larger if you have enormous ramdisks/kernels. Note that
++ * you should trust all kernels not to be maliciously large (e.g. to
++ * exhaust all dom0 memory) if you do this (see CVE-2012-4544 /
++ * XSA-25). You can also set the default independently for
++ * ramdisks/kernels in xc_dom_allocate() or call
++ * xc_dom_{kernel,ramdisk}_max_size.
++ */
++#ifndef XC_DOM_DECOMPRESS_MAX
++#define XC_DOM_DECOMPRESS_MAX (1024*1024*1024) /* 1GB */
++#endif
++
++int xc_dom_kernel_check_size(struct xc_dom_image *dom, size_t sz);
++int xc_dom_kernel_max_size(struct xc_dom_image *dom, size_t sz);
++
++int xc_dom_ramdisk_check_size(struct xc_dom_image *dom, size_t sz);
++int xc_dom_ramdisk_max_size(struct xc_dom_image *dom, size_t sz);
++
+ size_t xc_dom_check_gzip(xc_interface *xch,
+ void *blob, size_t ziplen);
+ int xc_dom_do_gunzip(xc_interface *xch,
+@@ -224,7 +244,8 @@ void xc_dom_log_memory_footprint(struct
+ void *xc_dom_malloc(struct xc_dom_image *dom, size_t size);
+ void *xc_dom_malloc_page_aligned(struct xc_dom_image *dom, size_t size);
+ void *xc_dom_malloc_filemap(struct xc_dom_image *dom,
+- const char *filename, size_t * size);
++ const char *filename, size_t * size,
++ const size_t max_size);
+ char *xc_dom_strdup(struct xc_dom_image *dom, const char *str);
+
+ /* --- alloc memory pool ------------------------------------------- */
+diff -r a15596a619ed -r 69d1cc78a5bd tools/libxc/xc_dom_bzimageloader.c
+--- a/tools/libxc/xc_dom_bzimageloader.c Thu Oct 04 10:44:43 2012 +0200
++++ b/tools/libxc/xc_dom_bzimageloader.c Fri Oct 26 16:10:04 2012 +0100
+@@ -47,13 +47,19 @@ static int xc_try_bzip2_decode(
+ char *out_buf;
+ char *tmp_buf;
+ int retval = -1;
+- int outsize;
++ unsigned int outsize;
+ uint64_t total;
+
+ stream.bzalloc = NULL;
+ stream.bzfree = NULL;
+ stream.opaque = NULL;
+
++ if ( dom->kernel_size == 0)
++ {
++ DOMPRINTF("BZIP2: Input is 0 size");
++ return -1;
++ }
++
+ ret = BZ2_bzDecompressInit(&stream, 0, 0);
+ if ( ret != BZ_OK )
+ {
+@@ -66,6 +72,17 @@ static int xc_try_bzip2_decode(
+ * the input buffer to start, and we'll realloc as needed.
+ */
+ outsize = dom->kernel_size;
++
++ /*
++ * stream.avail_in and outsize are unsigned int, while kernel_size
++ * is a size_t. Check we aren't overflowing.
++ */
++ if ( outsize != dom->kernel_size )
++ {
++ DOMPRINTF("BZIP2: Input too large");
++ goto bzip2_cleanup;
++ }
++
+ out_buf = malloc(outsize);
+ if ( out_buf == NULL )
+ {
+@@ -98,13 +115,20 @@ static int xc_try_bzip2_decode(
+ if ( stream.avail_out == 0 )
+ {
+ /* Protect against output buffer overflow */
+- if ( outsize > INT_MAX / 2 )
++ if ( outsize > UINT_MAX / 2 )
+ {
+ DOMPRINTF("BZIP2: output buffer overflow");
+ free(out_buf);
+ goto bzip2_cleanup;
+ }
+
++ if ( xc_dom_kernel_check_size(dom, outsize * 2) )
++ {
++ DOMPRINTF("BZIP2: output too large");
++ free(out_buf);
++ goto bzip2_cleanup;
++ }
++
+ tmp_buf = realloc(out_buf, outsize * 2);
+ if ( tmp_buf == NULL )
+ {
+@@ -172,9 +196,15 @@ static int xc_try_lzma_decode(
+ unsigned char *out_buf;
+ unsigned char *tmp_buf;
+ int retval = -1;
+- int outsize;
++ size_t outsize;
+ const char *msg;
+
++ if ( dom->kernel_size == 0)
++ {
++ DOMPRINTF("LZMA: Input is 0 size");
++ return -1;
++ }
++
+ ret = lzma_alone_decoder(&stream, 128*1024*1024);
+ if ( ret != LZMA_OK )
+ {
+@@ -251,13 +281,20 @@ static int xc_try_lzma_decode(
+ if ( stream.avail_out == 0 )
+ {
+ /* Protect against output buffer overflow */
+- if ( outsize > INT_MAX / 2 )
++ if ( outsize > SIZE_MAX / 2 )
+ {
+ DOMPRINTF("LZMA: output buffer overflow");
+ free(out_buf);
+ goto lzma_cleanup;
+ }
+
++ if ( xc_dom_kernel_check_size(dom, outsize * 2) )
++ {
++ DOMPRINTF("LZMA: output too large");
++ free(out_buf);
++ goto lzma_cleanup;
++ }
++
+ tmp_buf = realloc(out_buf, outsize * 2);
+ if ( tmp_buf == NULL )
+ {
+@@ -327,6 +364,12 @@ static int xc_try_lzo1x_decode(
+ 0x89, 0x4c, 0x5a, 0x4f, 0x00, 0x0d, 0x0a, 0x1a, 0x0a
+ };
+
++ /*
++ * lzo_uint should match size_t. Check that this is the case to be
++ * sure we won't overflow various lzo_uint fields.
++ */
++ XC_BUILD_BUG_ON(sizeof(lzo_uint) != sizeof(size_t));
++
+ ret = lzo_init();
+ if ( ret != LZO_E_OK )
+ {
+@@ -406,6 +449,14 @@ static int xc_try_lzo1x_decode(
+ if ( src_len <= 0 || src_len > dst_len || src_len > left )
+ break;
+
++ msg = "Output buffer overflow";
++ if ( *size > SIZE_MAX - dst_len )
++ break;
++
++ msg = "Decompressed image too large";
++ if ( xc_dom_kernel_check_size(dom, *size + dst_len) )
++ break;
++
+ msg = "Failed to (re)alloc memory";
+ tmp_buf = realloc(out_buf, *size + dst_len);
+ if ( tmp_buf == NULL )
+diff -r a15596a619ed -r 69d1cc78a5bd tools/libxc/xc_dom_core.c
+--- a/tools/libxc/xc_dom_core.c Thu Oct 04 10:44:43 2012 +0200
++++ b/tools/libxc/xc_dom_core.c Fri Oct 26 16:10:04 2012 +0100
+@@ -159,7 +159,8 @@ void *xc_dom_malloc_page_aligned(struct
+ }
+
+ void *xc_dom_malloc_filemap(struct xc_dom_image *dom,
+- const char *filename, size_t * size)
++ const char *filename, size_t * size,
++ const size_t max_size)
+ {
+ struct xc_dom_mem *block = NULL;
+ int fd = -1;
+@@ -171,6 +172,13 @@ void *xc_dom_malloc_filemap(struct xc_do
+ lseek(fd, 0, SEEK_SET);
+ *size = lseek(fd, 0, SEEK_END);
+
++ if ( max_size && *size > max_size )
++ {
++ xc_dom_panic(dom->xch, XC_OUT_OF_MEMORY,
++ "tried to map file which is too large");
++ goto err;
++ }
++
+ block = malloc(sizeof(*block));
+ if ( block == NULL )
+ goto err;
+@@ -222,6 +230,40 @@ char *xc_dom_strdup(struct xc_dom_image
+ }
+
+ /* ------------------------------------------------------------------------ */
++/* decompression buffer sizing */
++int xc_dom_kernel_check_size(struct xc_dom_image *dom, size_t sz)
++{
++ /* No limit */
++ if ( !dom->max_kernel_size )
++ return 0;
++
++ if ( sz > dom->max_kernel_size )
++ {
++ xc_dom_panic(dom->xch, XC_INVALID_KERNEL,
++ "kernel image too large");
++ return 1;
++ }
++
++ return 0;
++}
++
++int xc_dom_ramdisk_check_size(struct xc_dom_image *dom, size_t sz)
++{
++ /* No limit */
++ if ( !dom->max_ramdisk_size )
++ return 0;
++
++ if ( sz > dom->max_ramdisk_size )
++ {
++ xc_dom_panic(dom->xch, XC_INVALID_KERNEL,
++ "ramdisk image too large");
++ return 1;
++ }
++
++ return 0;
++}
++
++/* ------------------------------------------------------------------------ */
+ /* read files, copy memory blocks, with transparent gunzip */
+
+ size_t xc_dom_check_gzip(xc_interface *xch, void *blob, size_t ziplen)
+@@ -235,7 +277,7 @@ size_t xc_dom_check_gzip(xc_interface *x
+
+ gzlen = blob + ziplen - 4;
+ unziplen = gzlen[3] << 24 | gzlen[2] << 16 | gzlen[1] << 8 | gzlen[0];
+- if ( (unziplen < 0) || (unziplen > (1024*1024*1024)) ) /* 1GB limit */
++ if ( (unziplen < 0) || (unziplen > XC_DOM_DECOMPRESS_MAX) )
+ {
+ xc_dom_printf
+ (xch,
+@@ -288,6 +330,9 @@ int xc_dom_try_gunzip(struct xc_dom_imag
+ if ( unziplen == 0 )
+ return 0;
+
++ if ( xc_dom_kernel_check_size(dom, unziplen) )
++ return 0;
++
+ unzip = xc_dom_malloc(dom, unziplen);
+ if ( unzip == NULL )
+ return -1;
+@@ -588,6 +633,9 @@ struct xc_dom_image *xc_dom_allocate(xc_
+ memset(dom, 0, sizeof(*dom));
+ dom->xch = xch;
+
++ dom->max_kernel_size = XC_DOM_DECOMPRESS_MAX;
++ dom->max_ramdisk_size = XC_DOM_DECOMPRESS_MAX;
++
+ if ( cmdline )
+ dom->cmdline = xc_dom_strdup(dom, cmdline);
+ if ( features )
+@@ -608,10 +656,25 @@ struct xc_dom_image *xc_dom_allocate(xc_
+ return NULL;
+ }
+
++int xc_dom_kernel_max_size(struct xc_dom_image *dom, size_t sz)
++{
++ DOMPRINTF("%s: kernel_max_size=%zx", __FUNCTION__, sz);
++ dom->max_kernel_size = sz;
++ return 0;
++}
++
++int xc_dom_ramdisk_max_size(struct xc_dom_image *dom, size_t sz)
++{
++ DOMPRINTF("%s: ramdisk_max_size=%zx", __FUNCTION__, sz);
++ dom->max_ramdisk_size = sz;
++ return 0;
++}
++
+ int xc_dom_kernel_file(struct xc_dom_image *dom, const char *filename)
+ {
+ DOMPRINTF("%s: filename=\"%s\"", __FUNCTION__, filename);
+- dom->kernel_blob = xc_dom_malloc_filemap(dom, filename, &dom->kernel_size);
++ dom->kernel_blob = xc_dom_malloc_filemap(dom, filename, &dom->kernel_size,
++ dom->max_kernel_size);
+ if ( dom->kernel_blob == NULL )
+ return -1;
+ return xc_dom_try_gunzip(dom, &dom->kernel_blob, &dom->kernel_size);
+@@ -621,7 +684,9 @@ int xc_dom_ramdisk_file(struct xc_dom_im
+ {
+ DOMPRINTF("%s: filename=\"%s\"", __FUNCTION__, filename);
+ dom->ramdisk_blob =
+- xc_dom_malloc_filemap(dom, filename, &dom->ramdisk_size);
++ xc_dom_malloc_filemap(dom, filename, &dom->ramdisk_size,
++ dom->max_ramdisk_size);
++
+ if ( dom->ramdisk_blob == NULL )
+ return -1;
+ // return xc_dom_try_gunzip(dom, &dom->ramdisk_blob, &dom->ramdisk_size);
+@@ -781,7 +846,11 @@ int xc_dom_build_image(struct xc_dom_ima
+ void *ramdiskmap;
+
+ unziplen = xc_dom_check_gzip(dom->xch, dom->ramdisk_blob, dom->ramdisk_size);
++ if ( xc_dom_ramdisk_check_size(dom, unziplen) != 0 )
++ unziplen = 0;
++
+ ramdisklen = unziplen ? unziplen : dom->ramdisk_size;
++
+ if ( xc_dom_alloc_segment(dom, &dom->ramdisk_seg, "ramdisk", 0,
+ ramdisklen) != 0 )
+ goto err;
+diff -r a15596a619ed -r 69d1cc78a5bd tools/pygrub/src/pygrub
+--- a/tools/pygrub/src/pygrub Thu Oct 04 10:44:43 2012 +0200
++++ b/tools/pygrub/src/pygrub Fri Oct 26 16:10:04 2012 +0100
+@@ -28,6 +28,7 @@ import grub.LiloConf
+ import grub.ExtLinuxConf
+
+ PYGRUB_VER = 0.6
++FS_READ_MAX = 1024 * 1024
+
+ def enable_cursor(ison):
+ if ison:
+@@ -421,7 +422,8 @@ class Grub:
+ if self.__dict__.get('cf', None) is None:
+ raise RuntimeError, "couldn't find bootloader config file in the image provided."
+ f = fs.open_file(self.cf.filename)
+- buf = f.read()
++ # limit read size to avoid pathological cases
++ buf = f.read(FS_READ_MAX)
+ del f
+ self.cf.parse(buf)
+
+@@ -670,6 +672,37 @@ if __name__ == "__main__":
+ def usage():
+ print >> sys.stderr, "Usage: %s [-q|--quiet] [-i|--interactive] [-n|--not-really] [--output=] [--kernel=] [--ramdisk=] [--args=] [--entry=] [--output-directory=] [--output-format=sxp|simple|simple0] <image>" %(sys.argv[0],)
+
++ def copy_from_image(fs, file_to_read, file_type, output_directory,
++ not_really):
++ if not_really:
++ if fs.file_exists(file_to_read):
++ return "<%s:%s>" % (file_type, file_to_read)
++ else:
++ sys.exit("The requested %s file does not exist" % file_type)
++ try:
++ datafile = fs.open_file(file_to_read)
++ except Exception, e:
++ print >>sys.stderr, e
++ sys.exit("Error opening %s in guest" % file_to_read)
++ (tfd, ret) = tempfile.mkstemp(prefix="boot_"+file_type+".",
++ dir=output_directory)
++ dataoff = 0
++ while True:
++ data = datafile.read(FS_READ_MAX, dataoff)
++ if len(data) == 0:
++ os.close(tfd)
++ del datafile
++ return ret
++ try:
++ os.write(tfd, data)
++ except Exception, e:
++ print >>sys.stderr, e
++ os.close(tfd)
++ os.unlink(ret)
++ del datafile
++ sys.exit("Error writing temporary copy of "+file_type)
++ dataoff += len(data)
++
+ try:
+ opts, args = getopt.gnu_getopt(sys.argv[1:], 'qinh::',
+ ["quiet", "interactive", "not-really", "help",
+@@ -786,24 +819,18 @@ if __name__ == "__main__":
+ if not fs:
+ raise RuntimeError, "Unable to find partition containing kernel"
+
+- if not_really:
+- bootcfg["kernel"] = "<kernel:%s>" % chosencfg["kernel"]
+- else:
+- data = fs.open_file(chosencfg["kernel"]).read()
+- (tfd, bootcfg["kernel"]) = tempfile.mkstemp(prefix="boot_kernel.",
+- dir=output_directory)
+- os.write(tfd, data)
+- os.close(tfd)
++ bootcfg["kernel"] = copy_from_image(fs, chosencfg["kernel"], "kernel",
++ output_directory, not_really)
+
+ if chosencfg["ramdisk"]:
+- if not_really:
+- bootcfg["ramdisk"] = "<ramdisk:%s>" % chosencfg["ramdisk"]
+- else:
+- data = fs.open_file(chosencfg["ramdisk"],).read()
+- (tfd, bootcfg["ramdisk"]) = tempfile.mkstemp(
+- prefix="boot_ramdisk.", dir=output_directory)
+- os.write(tfd, data)
+- os.close(tfd)
++ try:
++ bootcfg["ramdisk"] = copy_from_image(fs, chosencfg["ramdisk"],
++ "ramdisk", output_directory,
++ not_really)
++ except:
++ if not not_really:
++ os.unlink(bootcfg["kernel"])
++ raise
+ else:
+ initrd = None
+
+
diff --git a/xen.spec b/xen.spec
index fb61d40..edcf80d 100644
--- a/xen.spec
+++ b/xen.spec
@@ -20,7 +20,7 @@
Summary: Xen is a virtual machine monitor
Name: xen
Version: 4.1.3
-Release: 4%{?dist}
+Release: 5%{?dist}
Group: Development/Libraries
License: GPLv2+ and LGPLv2+ and BSD
URL: http://xen.org/
@@ -64,7 +64,6 @@ Patch34: xend.catchbt.patch
Patch35: xend-pci-loop.patch
Patch38: xen-backend.rules.patch
Patch39: xend.selinux.fixes.patch
-Patch40: pygrub.size.limits.patch
Patch45: xen-no-pyxml.patch
Patch50: upstream-23936:cdb34816a40a-rework
@@ -79,6 +78,7 @@ Patch62: xen-4.1-testing.23351.patch
Patch63: xen-4.1-testing.23352.patch
Patch64: qemu-xen-4.1-testing.git-3220480734832a148d26f7a81f90af61c2ecfdd9.patch
Patch65: qemu-xen-4.1-testing.git-d7d453f51459b591faa96d1c123b5bfff7c5b6b6.patch
+Patch66: xen-4.1-testing.23385.patch
Patch100: xen-configure-xend.patch
@@ -228,7 +228,6 @@ manage Xen virtual machines.
%patch35 -p1
%patch38 -p1
%patch39 -p1
-%patch40 -p1
%patch45 -p1
%patch50 -p1
@@ -243,6 +242,7 @@ manage Xen virtual machines.
%patch63 -p1
%patch64 -p1
%patch65 -p1
+%patch66 -p1
%patch100 -p1
@@ -701,6 +701,11 @@ rm -rf %{buildroot}
%endif
%changelog
+* Sun Oct 28 2012 Michael Young <m.a.young at durham.ac.uk> - 4.1.3-5
+- limit the size of guest kernels and ramdisks to avoid running out
+ of memory on dom0 during guest boot [XSA-25, CVE-2012-4544] (#870414)
+- remove pygrub.size.limits.patch patch included in new patch
+
* Thu Sep 06 2012 Michael Young <m.a.young at durham.ac.uk> - 4.1.3-4
- disable qemu monitor by default [XSA-19, CVE-2012-4411] (#855141)
More information about the scm-commits
mailing list