display_header() code to parse meminfo as is was buggy:
- uninitialized variables were used if meminfo was not as expected
- meminfo parsing failed on new kernels (3.14+) as new field 'MemAvailable'
was introduced between MemFree and Buffers
- shared memory was handled only for ancient kernels (2.4.x and earlier)
as result Buffers and shared memory fields were shown with bogus values
on current kernels.
The new code does not try to parse the old style summary header, as the
separated fields are always present (it saves code size). Additionally,
both Shmem (2.6+) and MemShared (2.4 and earlier) fields are now parsed
and summed for shared memory usage; as only one of them exists depending
on kernel version.
display_topmem_header() parses also meminfo so this makes it use the
same code for code shrink.
function old new delta
display_header - 681 +681
display_topmem_process_list 465 684 +219
parse_meminfo - 189 +189
static.fields - 106 +106
static.match 132 - -132
.rodata 120254 120117 -137
display_topmem_header 513 - -513
display_process_list 1705 667 -1038
------------------------------------------------------------------------------
(add/remove: 3/2 grow/shrink: 1/2 up/down: 1195/-1820) Total: -625 bytes
Signed-off-by: Timo Teräs <timo.teras@iki.fi>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Busybox lsof used the d_type field of a 'struct dirent' to verify whether the
entry is a symbolic link. This field, however, is not portable. On at least
one board [1] I have seen, that field is 0, and the check fails even though
the entry is a link.
The explicit check for a symbolic link is really only needed to skip the
default directory entries '.' and '..'. The directory /proc/<pid>/fd/
should not contain anything else but these two and symbolic links.
With these assumptions, this patch replaces the explicit link check with a
basic check for '.' and '..' (and any hidden file). In the unlikely case that
there are other file types, xmalloc_readlink() will return NULL, and we can
skip the entry.
[1] A MIPS-based board with glibc 2.9, Linux 2.6.32.27.
Signed-off-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
This also fixes a minor buffer overflow when displaying threads as
add_proc() only expects COMM_LEN bytes, but we give it one more than
that.
Reported-by: Dag Wieers <dag@wieers.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Because when -x is used (exact match), then we cannot compile the
regular expression with REG_NOSUB. The manual page regcomp(3) states
in section "Byte offsets":
Unless REG_NOSUB was set for the compilation of the pattern
buffer, it is possible to obtain substring match addressing
information.
The problem was detected on an ARM system with glibc 2.16.
Signed-off-by: Bernhard Walle <bernhard@bwalle.de>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Application tries to use timer_stats module instead of cpufreq_stats.
Error message is printed if opening of the file /proc/timer_stats fails.
Signed-off-by: Lauri Hintsala <lauri.hintsala@bluegiga.com>
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Create get_uptime() function in procps/ps.c, and use it rather than
relying on linux-specific sysinfo() inline.
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
was: 12:59:35 up 1:57, 4 users, load average: 0.11, 0.20, 0.53
is: 12:59:37 up 1:57, 4 users, load average: 0.11, 0.20, 0.53
While at it, switched code to unsigned division.
Based on patch by Pere Orga <gotrunks@gmail.com>.
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Done by enabling the REG_EXTENDED and REG_NOSUB regex flags,
which also increases compatibility with the procps versions.
Signed-off-by: Wade Berrier <wade_berrier@appsig.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
