maybe_escape_markup: Make function memory-safe

[orestisfl]

This fixes #492 and an additional buffer overflow that can happen when pango markup is enabled.

Using config

general {
        output_format = "none"
        markup = "pango"
}

order += "wireless _first_"

wireless _first_ {
  format_up = "W: (%quality at %essid, %bitrate) %ip"
}

and renaming my phone's hotspot to Hello world &<<<<<<hello world>> i3status will throw an AddressSanitizer error:

==1373240==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7411d720923e at pc 0x7411daa7cee9 bp 0x7ffdae6ce070 sp 0x7ffdae6cd800
WRITE of size 5 at 0x7411d720923e thread T0
    #0 0x7411daa7cee8 in __interceptor_vsprintf /usr/src/debug/gcc/gcc/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc:1765
    #1 0x7411daa7d0ff in __interceptor_sprintf /usr/src/debug/gcc/gcc/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc:1808
    #2 0x653b2764cdaf in maybe_escape_markup ../src/output.c:102
    #3 0x653b27677df9 in print_wireless_info ../src/print_wireless_info.c:607
    #4 0x653b27640bf1 in main ../i3status.c:709
    #5 0x7411da641ccf  (/usr/lib/libc.so.6+0x25ccf) (BuildId: 6542915cee3354fbcf2b3ac5542201faec43b5c9)
    #6 0x7411da641d89 in __libc_start_main (/usr/lib/libc.so.6+0x25d89) (BuildId: 6542915cee3354fbcf2b3ac5542201faec43b5c9)
    #7 0x653b27633f24 in _start (/tmp/xx/i3status/build/i3status+0x4ff24) (BuildId: c737ce6288265fa02a7617c66f51ddd16b5a8275)

Address 0x7411d720923e is located in stack of thread T0 at offset 574 in frame
    #0 0x653b276750ed in print_wireless_info ../src/print_wireless_info.c:513

  This frame has 10 object(s):
    [48, 56) 'tmp' (line 604)
    [80, 168) 'info' (line 516)
    [208, 320) 'placeholders' (line 623)
    [352, 382) 'string_quality' (line 569)
    [416, 446) 'string_signal' (line 570)
    [480, 510) 'string_noise' (line 571)
    [544, 574) 'string_essid' (line 572) <== Memory access at offset 574 overflows this variable
    [608, 638) 'string_frequency' (line 573)
    [672, 702) 'string_ip' (line 574)
    [736, 766) 'string_bitrate' (line 575)

With pango disabled, the error is thrown elsewhere (#492):

==1366779==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7bab43a0923e at pc 0x7bab4727cee9 bp 0x7ffc289d2540 sp 0x7ffc289d1cd0
WRITE of size 33 at 0x7bab43a0923e thread T0
    #0 0x7bab4727cee8 in __interceptor_vsprintf /usr/src/debug/gcc/gcc/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc:1765
    #1 0x7bab4727d0ff in __interceptor_sprintf /usr/src/debug/gcc/gcc/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc:1808
    #2 0x5dd180858aa4 in maybe_escape_markup ../src/output.c:93
    #3 0x5dd180883df9 in print_wireless_info ../src/print_wireless_info.c:607
    #4 0x5dd18084cbf1 in main ../i3status.c:709
    #5 0x7bab46843ccf  (/usr/lib/libc.so.6+0x25ccf) (BuildId: 6542915cee3354fbcf2b3ac5542201faec43b5c9)
    #6 0x7bab46843d89 in __libc_start_main (/usr/lib/libc.so.6+0x25d89) (BuildId: 6542915cee3354fbcf2b3ac5542201faec43b5c9)
    #7 0x5dd18083ff24 in _start (/tmp/xx/i3status/build/i3status+0x4ff24) (BuildId: c737ce6288265fa02a7617c66f51ddd16b5a8275)

Address 0x7bab43a0923e is located in stack of thread T0 at offset 574 in frame
    #0 0x5dd1808810ed in print_wireless_info ../src/print_wireless_info.c:513

  This frame has 10 object(s):
    [48, 56) 'tmp' (line 604)
    [80, 168) 'info' (line 516)
    [208, 320) 'placeholders' (line 623)
    [352, 382) 'string_quality' (line 569)
    [416, 446) 'string_signal' (line 570)
    [480, 510) 'string_noise' (line 571)
    [544, 574) 'string_essid' (line 572) <== Memory access at offset 574 overflows this variable
    [608, 638) 'string_frequency' (line 573)
    [672, 702) 'string_ip' (line 574)
    [736, 766) 'string_bitrate' (line 575)

With the patch output is correct:

W: ( 72% at Hello world &<<<<<<hello world>>, 1,2009 Gb/s) 192.168.26.237

and

W: ( 73% at Hello world &<<<<<<hello world>>, 1,1342 Gb/s) 192.168.26.237

The patch changes the maybe_escape_markup function to use dynamic allocation instead of a static buffer. Confusing pointer arithmetic is replaced with index-based memory access. The buffer pointer does not move around except for reallocations.

Fixes #492
Closes #525 (alternative PR)
CC @stapelberg @xstefank @raefko @mahrud

[stapelberg]

I have a slight preference for avoiding heap allocations.

Could we instead pass the size of the buffer and use snprintf to truncate? If you absolutely want to stick with allocations, I think we should just use asprintf instead of malloc/realloc/sprintf.

[orestisfl]

Avoid allocations for performance reasons? I feel it will be a drop in the bucket with this amount of memory. I prefer to keep dynamic allocations to actually allocate the correct amount of memory.

Using asprintf will mean one allocation per character which seems excessive to me.

[stapelberg]

Avoid allocations for performance reasons?

Yes. We rarely allocate in i3status (e.g. print_ipv6_addr dynamically allocates based on resp->ai_addrlen, though most of the remaining allocations could be optimized away, too).

Using asprintf will mean one allocation per character which seems excessive to me.

No, it will be one allocation per required escape, which seems reasonable. The advantage is that it gets rid of any manual allocation, so the code is easier to understand/verify overall.

[orestisfl]

WDYT about doubling the size of the essid buffer and using snprintf? This should make it memory safe and avoid any trimming except for specifically crafted essids with special symbols

[stapelberg]

Sounds good

[orestisfl]

Made the change, tested with a STRING_SIZE of 5, 30, 60.