On Fri, May 19, 2017 at 10:35 AM, Catalin Marinas <catalin.mari...@arm.com> wrote: > On Fri, May 19, 2017 at 05:40:16PM +0200, Luis R. Rodriguez wrote: >> If the following is a legit forced way to get query the kernel to ask it >> who owns a page then perhaps this technique can be used in the future to >> figure out who the hell caused this. Catalin, can you confirm? In this >> case this is perhaps not a leaked page but I am trying to abuse the >> kmemleak debugfs API to query who allocated the page. Is that fine? >> >> [ 0.916771] WARNING: CPU: 0 PID: 1 at arch/x86/mm/dump_pagetables.c:235 >> note_page+0x63c/0x7e0 >> [ 0.917636] x86/mm: Found insecure W+X mapping at address >> ffffffffc03d5000/0xffffffffc03d5000 >> [ 0.918502] Modules linked in: >> [ 0.918819] CPU: 0 PID: 1 Comm: swapper/0 Not tainted >> 4.11.0-mcgrof-force-config #340 >> [ 0.919631] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS >> rel-1.10.2-0-g5f4c7b1-prebuilt.qemu-project.org 04/01/2014 >> [ 0.920011] Call Trace: >> [ 0.920011] dump_stack+0x63/0x81 >> [ 0.920011] __warn+0xcb/0xf0 >> [ 0.920011] warn_slowpath_fmt+0x5a/0x80 >> [ 0.920011] note_page+0x63c/0x7e0 >> [ 0.920011] ptdump_walk_pgd_level_core+0x3b1/0x460 >> [ 0.920011] ? 0xffffffff86c00000 >> [ 0.920011] ptdump_walk_pgd_level_checkwx+0x17/0x20 >> [ 0.920011] mark_rodata_ro+0xf4/0x100 >> [ 0.920011] ? rest_init+0x80/0x80 >> [ 0.920011] kernel_init+0x2a/0x100 >> [ 0.920011] ret_from_fork+0x2c/0x40 >> [ 0.925474] ---[ end trace dca00cd779490a2b ]--- >> [ 0.925959] x86/mm: Checked W+X mappings: FAILED, 1 W+X pages found. >> >> echo dump=0xffffffffc03d5000 > /sys/kernel/debug/kmemleak >> dmesg | tail >> >> [ 49.209565] kmemleak: Object 0xffffffffc03d5000 (size 335): >> [ 49.210814] kmemleak: comm "swapper/0", pid 1, jiffies 4294892440 >> [ 49.212148] kmemleak: min_count = 2 >> [ 49.212852] kmemleak: count = 0 >> [ 49.213363] kmemleak: flags = 0x1 >> [ 49.213363] kmemleak: checksum = 0 >> [ 49.213363] kmemleak: backtrace: >> [ 49.213363] kmemleak_alloc+0x4a/0xa0 >> [ 49.213363] __vmalloc_node_range+0x20a/0x2b0 >> [ 49.213363] module_alloc+0x67/0xc0 >> [ 49.213363] arch_ftrace_update_trampoline+0xba/0x260 >> [ 49.213363] ftrace_startup+0x90/0x210 >> [ 49.213363] register_ftrace_function+0x4b/0x60 >> [ 49.213363] arm_kprobe+0x84/0xe0 >> [ 49.213363] register_kprobe+0x56e/0x5b0 >> [ 49.213363] init_test_probes+0x61/0x560 >> [ 49.213363] init_kprobes+0x1e3/0x206 >> [ 49.213363] do_one_initcall+0x52/0x1a0 >> [ 49.213363] kernel_init_freeable+0x178/0x200 >> [ 49.213363] kernel_init+0xe/0x100 >> [ 49.213363] ret_from_fork+0x2c/0x40 >> [ 49.213363] 0xffffffffffffffff > > You could as well use kmemleak this way since it tracks the memory > allocations. However, it doesn't track alloc_pages and also doesn't > track mapping existing pages (vmap etc.)
One thing I've pondered: can we make some debugging mode (kmemleak, perhaps?) check that freed memory is RW at the time it's freed? I once wrote some buggy code that freed an R page and caused an OOPS much later, and this bug here seems likely to be some code that frees RWX memory. --Andy
