Karel Zak [Thu, 20 Aug 2020 10:28:52 +0000 (12:28 +0200)]
lscpu: use cache ID, keep caches independent on CPU type
The cache is identified by Type, Level and ID, the ID is unique cache
instance identifier (of the type).
This changes forces lscpu allocate more lscpu_cache instances (than
old version), but now we're ready for arbitrary scenario where
different CPU types share caches and the same cache type uses
different size in different instances, etc.
Karel Zak [Wed, 19 Aug 2020 10:43:48 +0000 (12:43 +0200)]
lscpu: remove obsolete code
This is the first step in conversion from old lscpu to the new code.
The patch removes obsolete code from lscpu.c and lscpu.h. The old
output code in lscpu.c is temporary disabled by #ifdef due to
incompatibility between old and new internal APIs -- this will be
changed later by small steps to make all all the changes review-able.
Karel Zak [Thu, 16 Jul 2020 08:51:14 +0000 (10:51 +0200)]
lscpu: calculate threads number from type specific values
Don't use global CPU masks (like "online" or "present") to
calculate type specific number of threads due systems with
mixed CPU types.
It's also necessary to check all thread_siblings maps to get the
highest number, because some threads (CPUs) may be disables, for
example old lscpu calculates number of threads from the cpu0 and
if you disable cpu0's sibling (cpu4):
CPU(s): 8
On-line CPU(s) list: 0-7
Thread(s) per core: 2 <---
Core(s) per socket: 4
Socket(s): 1
Karel Zak [Tue, 17 Mar 2020 12:12:12 +0000 (13:12 +0100)]
lscpu: add very basic cputype code
The current lscpu assumes that all CPUs in the system are the same.
Unfortunately this is not true. We need to split all internal CPUs
descriptions to CPU-type and CPU.
This patch add lscpu-cputype.c where will be CPU-type description --
mostly based on /proc/cpuinfo.
Samanta Navarro [Sun, 8 Nov 2020 11:46:56 +0000 (11:46 +0000)]
libblkid: fix time_t handling
The time_t data type is a signed integer. A signed integer overflow is
not defined in C programming language.
A signed overflow occurs on 32 bit systems with 32 time_t for loop back
devices, e.g. when calling "blkid /dev/loop0". This happens because
bid_time is set to INT_MIN and the diff calculation cannot store the
result in time_t (positive int - INT_MIN > INT_MAX).
This fix changes the code to use an unsigned integer calculation. It
pretty much means that the code works as before, but well defined in C.
Checking diff to be positive protects the code against system setups
with dates before 1970 as well.
The time_t data type on modern Linux systems is 64 bit even for 32 bit
systems. Since long is 32 bit on these systems, long long is a better
data type for 64 bit output.
Karel Zak [Mon, 9 Nov 2020 10:06:27 +0000 (11:06 +0100)]
Merge branch 'vipw-shortwrite' of https://github.com/DankRank/util-linux
* 'vipw-shortwrite' of https://github.com/DankRank/util-linux:
ul_copy_file: make defines for return values
read_all: return 0 when EOF occurs after 0 bytes
ul_copy_file: add test program
ul_copy_file: handle EAGAIN and EINTR
ul_copy_file: use all_read/all_write
ul_copy_file: use BUFSSIZ for buffer size
nologin: use ul_copy_file
login: use ul_copy_file
configure.ac: check for sendfile
ul_copy_file: use sendfile
vipw: move copyfile to the lib
vipw: fix short write handling in copyfile
Egor Chelak [Fri, 6 Nov 2020 10:15:36 +0000 (12:15 +0200)]
configure.ac: check for sendfile
Do note that according to man sendfile, "Other UNIX systems implement
sendfile() with different semantics and prototypes."
If this is something we care about, a better check is needed.
Suggested-by: Karel Zak <kzak@redhat.com> Reviewed-by: Sami Kerola <kerolasa@iki.fi> Signed-off-by: Egor Chelak <egor.chelak@gmail.com>
Samanta Navarro [Sun, 8 Nov 2020 11:47:18 +0000 (11:47 +0000)]
libuuid: check quality of random bytes
If a libuuid application is unable to access /dev/random or /dev/urandom
then uuid generation by uuid_generate falls back to uuid_generate_time.
This could happen in chroot or container environments.
The function ul_random_get_bytes from lib/randutils.c uses getrandom if
it is available. This could either mean that the libuuid application
skips good random bytes because the character special files do not exist
or the application trusts in good random bytes just because these files
are accessible but not necessarily usable, e.g. limit of open file
descriptors reached, lack of data, kernel without getrandom, etc.
This commit modifies ul_random_get_bytes to return an integer which
indicates if random bytes are of good quality (0) or not (1). Callers
can decide based on this information if they want to discard the random
bytes. Only libuuid checks the return value. I decided to return 1
instead of -1 because -1 feels more like an error, but weak random bytes
can be totally fine.
Another issue is that getrandom sets errno to specific values only in
case of an error, i.e. with return value -1. Set errno to 0 explicitly
if getrandom succeeds so we do not enter the fallback routine for
ENOSYS by mistake. I do not think that this is likely to happen, but it
really depends on possible wrapper function supplied by a C library.
Samanta Navarro [Sun, 8 Nov 2020 11:45:18 +0000 (11:45 +0000)]
libblkid: allow a lot of mac partitions
If the map count is set to INT_MAX then the for loop does not stop
because its check is never false.
I have not found a correct upper limit. The other partition logics have
a maximum amount (exception is atari.c).
The loop itself wouldn't be endless. If the iteration reaches block 0
then the signature will be wrong. This means that map count = INT_MAX
case would fail even if such a setup would be correct on disk.
Karel Zak [Fri, 6 Nov 2020 08:44:52 +0000 (09:44 +0100)]
Merge branch 'whereis' of https://github.com/ferivoz/util-linux
* 'whereis' of https://github.com/ferivoz/util-linux:
whereis: extend test case
whereis: filter bin, man and src differently
whereis: do not strip suffixes
whereis: do not ignore trailing numbers
whereis: add --disable-whereis to configure
whereis: add lib32 directories
whereis: support zst compressed man pages
whereis: fix out of boundary read
`--json` implies `--dump`, thus `--json --dump` must be
allowed. `--list-free` is incompatible with `--dump`, and thus also
with `--json`. Currently `--json --dump` is prohibited, even though
`--list-free` is not specified at all.
Samanta Navarro [Wed, 4 Nov 2020 11:38:00 +0000 (11:38 +0000)]
whereis: filter bin, man and src differently
Consider "s." prefixes for source code files only (even though I do not
know which VCS does that), compression suffixes for manual pages and
strict matching for executables.
Calling "whereis python3" is kind of okay to return python3.8 next to
python3, but python3.8-config is not the same tool as python3.
Samanta Navarro [Wed, 4 Nov 2020 11:37:00 +0000 (11:37 +0000)]
whereis: do not strip suffixes
The whereis implementations of FreeBSD, macOS, NetBSD, and OpenBSD do
not strip suffixes. Although whereis is not a POSIX tool and has no
shared standard, even its manual page indicates that the supplied names
are command names.
Commands do not have a suffix on Linux systems.
Stripping suffixes actually leads to issues with tools like fsck.ext4,
since fsck.ext4 is not the same tool as fsck and definitely not the same
tool as fsck.minix.
Samanta Navarro [Wed, 4 Nov 2020 11:36:00 +0000 (11:36 +0000)]
whereis: do not ignore trailing numbers
The commands diff and diff3 are so distinct that their manual pages
should not be mixed in whereis output.
Theoretically this works for commands and binaries with links to each
other, e.g. gpg and gpg2, but if gpg is version 1 and gpg2 is version 2
then manual pages do not match either.
Also the while loop does not decrement "i" while incrementing "dp". The
effect of this is that the output of whereis depends on manual pages
being compressed or not.
The easiest solution is to not ignore trailing numbers.
projects / thirdparty / util-linux.git / log
