Stateless and I stumbled upon this issue while discussing the
semantics of read, accepting a size_t but only being able to return
ssize_t, effectively lacking the ability to report successful
reads > SSIZE_MAX.
The discussion went along and we came to the topic of input-based
memory allocations. Basically, it was possible for the argument
to a memory-allocation-function to overflow, leading to a segfault
later.
The OpenBSD-guys came up with the ingenious reallocarray-function,
and I implemented it as ereallocarray, which automatically returns
on error.
Read more about it here[0].
A simple testcase is this (courtesy to stateless):
$ sbase-strings -n (2^(32|64) / 4)
This will segfault before this patch and properly return an OOM-
situation afterwards (thanks to the overflow-check in reallocarray).
[0]: http://www.openbsd.org/cgi-bin/man.cgi/OpenBSD-current/man3/calloc.3
Allows dropping a local variable if the explicit PID is not needed
and it makes it clearer what happens.
Also, one should always strive for consistency for cases like these.
Quoting POSIX[0]:
"Care should be taken, also, to call _exit() rather than exit() if exec cannot be used, since
exit() flushes and closes standard I/O channels, thereby damaging the parent process' standard
I/O data structures. (Even with fork(), it is wrong to call exit(), since buffered data would
then be flushed twice.)"
[0]: http://pubs.opengroup.org/onlinepubs/009695399/functions/vfork.html
Similar to the chgrp(1)-audit:
1) Refactor manpage so it's actually fun to read
2) BUGFIX: Call (l)chown properly when the H-flag is specified
(only when depth > 0)
3) BUGFIX: Call (l)chown properly when the h-flag is specified
(only when depth = 0).
4) BUGFIX: Only recurse() in chgrp() when the initial chownf()
succeeds.
5) Style fixes, argv-basing.
6) Rename status to ret for consistency.
7) Add blank line before return.
1) Refactor manpage so it's actually fun to read.
2) BUGFIX: Call (l)chown properly when the H-flag is specified
(only when depth > 0).
3) BUGFIX: Call (l)chown properly when the h-flag is specified
(only when depth = 0).
4) BUGFIX: Only recurse() in chgrp() when the initial chownf()
succeeds.
5) Style fixes, argv-basing.
6) Rename status to ret for consistency.
7) Add blank line before return.
1) Update manpage with the num-syntax.
2) Use size_t for years and derivatives.
3) Use putchar instead of printf wherever possible.
4) Update usage().
5) Style changes.
1) Refactor manpage.
2) De-globalize local values.
3) update usage().
4) sort local variable declarations.
5) fix wrong argument in strtonum (3 -> 1).
6) argc-argv style, boolean style.
7) check bytes > 0 before accessing b.lines[i][bytes - 1]
relying on len only makes sense but let's not push it.
7) don't break on maxlen > (chars - 1) / 2. This didn't even
make sense.
8) _correctly_ calculate cols and rows in a readable way.
9) Rewrite loop over rows and cols in a readable way and
using putchar in a loop instead of printf-magic or fputs
where not necessary.
- Also fix a few typos, style and section order.
- Changed the text "800 characters per line" to "800 bytes per line" as col
doesn't seem to support UTF-8 right now.
1) don't mix declarations and code (leave recursion alone for now as I
plan on changing/using recurse)
2) change **argv to *argv[]
3) check for error on fork()
We'll probably develop this outside of sbase. A simple script that
parses /etc/magic and generates magic.h would be sufficient.
The table can be huge and we do not want to bloat up binary size
only for file(1).
1) Clarify behaviour when the f-flag is given and a target is in its
own way.
2) Fix usage()-style.
3) Group local variable declarations.
4) reorder args
5) argc style, other boolean style changes
6) improve error messages
7) set argv[argc - 1] to NULL to allow argv-centric loop later
8) BUGFIX: POSIX specifies that when with the f-flag there's a
situation where a file stands in its own way for linking it
should be ignored.
9) Add weprintf() where possible, so we don't pussy out when there's
a small issue. This is sbase ffs!
1) Update manpage, refactor the HLP-section and other wordings.
2) BUGFIX: If chmod() fails, don't recurse.
3) Rewrite the arg-loop, fixing several issues:
BUGFIX: Handle multi-flags (e.g. -RH)
BUGFIX: Properly handle the termination flag --, error on e.g. --x
BUGFIX: Error out on an empty flag -.
4) Refactor logic after the arg-loop, which is now simpler thanks
to argv-incremention.
1) No need for strchr() in mkdirp or a while-loop. Rewrite it in
a sane and readable way.
2) fix usage according to the manpage.
3) order includes, don't align local variables.
4) argc-style-fix.
5) BUGFIX: Don't try to chmod() *argv when mkdir() / mkdirp() failed.
6) Add newline before return in two places.
1) use arg.h
2) !strcmp
3) **argv to *argv[]
4) fix test to check if basename(argv0) == "[" but avoid
basename(3p) as it may change the contents of the string
passed to it and I didn't want to make a copy.
1) Use (s)size_t in head().
2) BUGFIX: only check buf[len - 1] when len > 0, else there would
be an overflow when getline returns 0 (which can happen) and a
very potential segmentation fault.
3) fix error-messages.
4) update usage().
5) argv-argc-style.
6) clear up the main loop with if (newline).
7) add newline before return.
1) fix usage().
2) sort includes and comment properly. rename rbeg and rend to r0 and r1.
3) argc style and usage fixes.
4) make error-messages clearer.
5) BUGFIX: It was ignored when fork() failed.
6) Don't call enprintf() after execvp and use _exit instead.
1) Make argument-naming consistent with other tools (cp(1), ...)
2) style fixes
3) usage() fix
4) BUGFIX: Probably from the old non-arg.h days, the directory-
check was only done when argc > 3, but with arg.h, this ignores
the case when 3 arguments were given.
This is actually a pretty serious issue and I'm glad it's fixed.
5) Moreover, be more verbose when stat() fails and make it clearer
what the hell is going on at this checkpoint.
1) "duplicate" implies that you can only specify two outputs,
"multiply" is a better word describing the functionality.
2) fix other wording in the manpage
3) fix usage()
4) reorder local variables
5) fix sizeof() style
6) we need argv later, don't increment argv and rather iterate
over argc.
7) Improve error messages, print the filename which the write
failed to instead of printing the buffer itself (how much
sense does that make, printing 1024 Bytes of garbage?).
Also, give the name of the function which failed.
1) no need to include sys/stat.h
2) remove the enum which just added a layer too thick on this simple
program
3) argc-style, other style
4) weprintf instead of enprintf, then save the error-message of
execvp before and return the proper status.
5) write consistent "not reached" comment.
