CVE-2022-48689

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> tcp: TX zerocopy should not sense pfmemalloc status<br /> <br /> We got a recent syzbot report [1] showing a possible misuse<br /> of pfmemalloc page status in TCP zerocopy paths.<br /> <br /> Indeed, for pages coming from user space or other layers,<br /> using page_is_pfmemalloc() is moot, and possibly could give<br /> false positives.<br /> <br /> There has been attempts to make page_is_pfmemalloc() more robust,<br /> but not using it in the first place in this context is probably better,<br /> removing cpu cycles.<br /> <br /> Note to stable teams :<br /> <br /> You need to backport 84ce071e38a6 ("net: introduce<br /> __skb_fill_page_desc_noacc") as a prereq.<br /> <br /> Race is more probable after commit c07aea3ef4d4<br /> ("mm: add a signature in struct page") because page_is_pfmemalloc()<br /> is now using low order bit from page-&gt;lru.next, which can change<br /> more often than page-&gt;index.<br /> <br /> Low order bit should never be set for lru.next (when used as an anchor<br /> in LRU list), so KCSAN report is mostly a false positive.<br /> <br /> Backporting to older kernel versions seems not necessary.<br /> <br /> [1]<br /> BUG: KCSAN: data-race in lru_add_fn / tcp_build_frag<br /> <br /> write to 0xffffea0004a1d2c8 of 8 bytes by task 18600 on cpu 0:<br /> __list_add include/linux/list.h:73 [inline]<br /> list_add include/linux/list.h:88 [inline]<br /> lruvec_add_folio include/linux/mm_inline.h:105 [inline]<br /> lru_add_fn+0x440/0x520 mm/swap.c:228<br /> folio_batch_move_lru+0x1e1/0x2a0 mm/swap.c:246<br /> folio_batch_add_and_move mm/swap.c:263 [inline]<br /> folio_add_lru+0xf1/0x140 mm/swap.c:490<br /> filemap_add_folio+0xf8/0x150 mm/filemap.c:948<br /> __filemap_get_folio+0x510/0x6d0 mm/filemap.c:1981<br /> pagecache_get_page+0x26/0x190 mm/folio-compat.c:104<br /> grab_cache_page_write_begin+0x2a/0x30 mm/folio-compat.c:116<br /> ext4_da_write_begin+0x2dd/0x5f0 fs/ext4/inode.c:2988<br /> generic_perform_write+0x1d4/0x3f0 mm/filemap.c:3738<br /> ext4_buffered_write_iter+0x235/0x3e0 fs/ext4/file.c:270<br /> ext4_file_write_iter+0x2e3/0x1210<br /> call_write_iter include/linux/fs.h:2187 [inline]<br /> new_sync_write fs/read_write.c:491 [inline]<br /> vfs_write+0x468/0x760 fs/read_write.c:578<br /> ksys_write+0xe8/0x1a0 fs/read_write.c:631<br /> __do_sys_write fs/read_write.c:643 [inline]<br /> __se_sys_write fs/read_write.c:640 [inline]<br /> __x64_sys_write+0x3e/0x50 fs/read_write.c:640<br /> do_syscall_x64 arch/x86/entry/common.c:50 [inline]<br /> do_syscall_64+0x2b/0x70 arch/x86/entry/common.c:80<br /> entry_SYSCALL_64_after_hwframe+0x63/0xcd<br /> <br /> read to 0xffffea0004a1d2c8 of 8 bytes by task 18611 on cpu 1:<br /> page_is_pfmemalloc include/linux/mm.h:1740 [inline]<br /> __skb_fill_page_desc include/linux/skbuff.h:2422 [inline]<br /> skb_fill_page_desc include/linux/skbuff.h:2443 [inline]<br /> tcp_build_frag+0x613/0xb20 net/ipv4/tcp.c:1018<br /> do_tcp_sendpages+0x3e8/0xaf0 net/ipv4/tcp.c:1075<br /> tcp_sendpage_locked net/ipv4/tcp.c:1140 [inline]<br /> tcp_sendpage+0x89/0xb0 net/ipv4/tcp.c:1150<br /> inet_sendpage+0x7f/0xc0 net/ipv4/af_inet.c:833<br /> kernel_sendpage+0x184/0x300 net/socket.c:3561<br /> sock_sendpage+0x5a/0x70 net/socket.c:1054<br /> pipe_to_sendpage+0x128/0x160 fs/splice.c:361<br /> splice_from_pipe_feed fs/splice.c:415 [inline]<br /> __splice_from_pipe+0x222/0x4d0 fs/splice.c:559<br /> splice_from_pipe fs/splice.c:594 [inline]<br /> generic_splice_sendpage+0x89/0xc0 fs/splice.c:743<br /> do_splice_from fs/splice.c:764 [inline]<br /> direct_splice_actor+0x80/0xa0 fs/splice.c:931<br /> splice_direct_to_actor+0x305/0x620 fs/splice.c:886<br /> do_splice_direct+0xfb/0x180 fs/splice.c:974<br /> do_sendfile+0x3bf/0x910 fs/read_write.c:1249<br /> __do_sys_sendfile64 fs/read_write.c:1317 [inline]<br /> __se_sys_sendfile64 fs/read_write.c:1303 [inline]<br /> __x64_sys_sendfile64+0x10c/0x150 fs/read_write.c:1303<br /> do_syscall_x64 arch/x86/entry/common.c:50 [inline]<br /> do_syscall_64+0x2b/0x70 arch/x86/entry/common.c:80<br /> entry_SYSCALL_64_after_hwframe+0x63/0xcd<br /> <br /> value changed: 0x0000000000000000 -&gt; 0xffffea0004a1d288<br /> <br /> Reported by Kernel Concurrency Sanitizer on:<br /> CPU: 1 PID: 18611 Comm: syz-executor.4 Not tainted 6.0.0-rc2-syzkaller-00248-ge022620b5d05-dirty #0<br /> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 07/22/2022