git.librecmc.org Git - oweals/openwrt.git/blob

   1 From b72c51a58e6d63ef673ac96b8ab5bc98799c5f7b Mon Sep 17 00:00:00 2001
   2 From: Lyude Paul <lyude@redhat.com>
   3 Date: Sat, 24 Nov 2018 17:57:05 -0500
   4 Subject: [PATCH] brcmfmac: Fix out of bounds memory access during fw load
   5
   6 I ended up tracking down some rather nasty issues with f2fs (and other
   7 filesystem modules) constantly crashing on my kernel down to a
   8 combination of out of bounds memory accesses, one of which was coming
   9 from brcmfmac during module load:
  10
  11 [   30.891382] brcmfmac: brcmf_fw_alloc_request: using brcm/brcmfmac4356-sdio for chip BCM4356/2
  12 [   30.894437] ==================================================================
  13 [   30.901581] BUG: KASAN: global-out-of-bounds in brcmf_fw_alloc_request+0x42c/0x480 [brcmfmac]
  14 [   30.909935] Read of size 1 at addr ffff2000024865df by task kworker/6:2/387
  15 [   30.916805]
  16 [   30.918261] CPU: 6 PID: 387 Comm: kworker/6:2 Tainted: G           O      4.20.0-rc3Lyude-Test+ #19
  17 [   30.927251] Hardware name: amlogic khadas-vim2/khadas-vim2, BIOS 2018.07-rc2-armbian 09/11/2018
  18 [   30.935964] Workqueue: events brcmf_driver_register [brcmfmac]
  19 [   30.941641] Call trace:
  20 [   30.944058]  dump_backtrace+0x0/0x3e8
  21 [   30.947676]  show_stack+0x14/0x20
  22 [   30.950968]  dump_stack+0x130/0x1c4
  23 [   30.954406]  print_address_description+0x60/0x25c
  24 [   30.959066]  kasan_report+0x1b4/0x368
  25 [   30.962683]  __asan_report_load1_noabort+0x18/0x20
  26 [   30.967547]  brcmf_fw_alloc_request+0x42c/0x480 [brcmfmac]
  27 [   30.967639]  brcmf_sdio_probe+0x163c/0x2050 [brcmfmac]
  28 [   30.978035]  brcmf_ops_sdio_probe+0x598/0xa08 [brcmfmac]
  29 [   30.983254]  sdio_bus_probe+0x190/0x398
  30 [   30.983270]  really_probe+0x2a0/0xa70
  31 [   30.983296]  driver_probe_device+0x1b4/0x2d8
  32 [   30.994901]  __driver_attach+0x200/0x280
  33 [   30.994914]  bus_for_each_dev+0x10c/0x1a8
  34 [   30.994925]  driver_attach+0x38/0x50
  35 [   30.994935]  bus_add_driver+0x330/0x608
  36 [   30.994953]  driver_register+0x140/0x388
  37 [   31.013965]  sdio_register_driver+0x74/0xa0
  38 [   31.014076]  brcmf_sdio_register+0x14/0x60 [brcmfmac]
  39 [   31.023177]  brcmf_driver_register+0xc/0x18 [brcmfmac]
  40 [   31.023209]  process_one_work+0x654/0x1080
  41 [   31.032266]  worker_thread+0x4f0/0x1308
  42 [   31.032286]  kthread+0x2a8/0x320
  43 [   31.039254]  ret_from_fork+0x10/0x1c
  44 [   31.039269]
  45 [   31.044226] The buggy address belongs to the variable:
  46 [   31.044351]  brcmf_firmware_path+0x11f/0xfffffffffffd3b40 [brcmfmac]
  47 [   31.055601]
  48 [   31.057031] Memory state around the buggy address:
  49 [   31.061800]  ffff200002486480: 04 fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00
  50 [   31.068983]  ffff200002486500: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  51 [   31.068993] >ffff200002486580: 00 00 00 00 00 00 00 00 fa fa fa fa 00 00 00 00
  52 [   31.068999]                                                     ^
  53 [   31.069017]  ffff200002486600: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  54 [   31.096521]  ffff200002486680: 00 00 00 00 00 00 00 00 00 00 00 00 fa fa fa fa
  55 [   31.096528] ==================================================================
  56 [   31.096533] Disabling lock debugging due to kernel taint
  57
  58 It appears that when trying to determine the length of the string in the
  59 alternate firmware path, we make the mistake of not handling the case
  60 where the firmware path is empty correctly. Since strlen(mp_path) can
  61 return 0, we'll end up accessing mp_path[-1] when the firmware_path
  62 isn't provided through the module arguments.
  63
  64 So, fix this by just setting the end char to '\0' by default, and only
  65 changing it if we have a non-zero length. Additionally, use strnlen()
  66 with BRCMF_FW_ALTPATH_LEN instead of strlen() just to be extra safe.
  67
  68 Fixes: 2baa3aaee27f ("brcmfmac: introduce brcmf_fw_alloc_request() function")
  69 Cc: Hante Meuleman <hante.meuleman@broadcom.com>
  70 Cc: Pieter-Paul Giesberts <pieter-paul.giesberts@broadcom.com>
  71 Cc: Franky Lin <franky.lin@broadcom.com>
  72 Cc: Arend van Spriel <arend.vanspriel@broadcom.com>
  73 Cc: Kalle Valo <kvalo@codeaurora.org>
  74 Cc: Arend Van Spriel <arend.vanspriel@broadcom.com>
  75 Cc: Himanshu Jha <himanshujha199640@gmail.com>
  76 Cc: Dan Haab <dhaab@luxul.com>
  77 Cc: Jia-Shyr Chuang <saint.chuang@cypress.com>
  78 Cc: Ian Molton <ian@mnementh.co.uk>
  79 Cc: <stable@vger.kernel.org> # v4.17+
  80 Signed-off-by: Lyude Paul <lyude@redhat.com>
  81 Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
  82 ---
  83  .../net/wireless/broadcom/brcm80211/brcmfmac/firmware.c   | 8 ++++++--
  84  1 file changed, 6 insertions(+), 2 deletions(-)
  85
  86 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/firmware.c
  87 +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/firmware.c
  88 @@ -708,8 +708,9 @@ brcmf_fw_alloc_request(u32 chip, u32 chi
  89         struct brcmf_fw_request *fwreq;
  90         char chipname[12];
  91         const char *mp_path;
  92 +       size_t mp_path_len;
  93         u32 i, j;
  94 -       char end;
  95 +       char end = '\0';
  96         size_t reqsz;
  97
  98         for (i = 0; i < table_size; i++) {
  99 @@ -734,7 +735,10 @@ brcmf_fw_alloc_request(u32 chip, u32 chi
 100                    mapping_table[i].fw_base, chipname);
 101
 102         mp_path = brcmf_mp_global.firmware_path;
 103 -       end = mp_path[strlen(mp_path) - 1];
 104 +       mp_path_len = strnlen(mp_path, BRCMF_FW_ALTPATH_LEN);
 105 +       if (mp_path_len)
 106 +               end = mp_path[mp_path_len - 1];
 107 +
 108         fwreq->n_items = n_fwnames;
 109
 110         for (j = 0; j < n_fwnames; j++) {