Upstream commit:
Date: Thu, 28 May 2020 21:31:45 +1000
redir: Retry open64 on EINTR
It is possible for open64 to block on named pipes, and therefore
it can be interrupted by signals and return EINTR. We should only
let it fail with EINTR if real signals are pending (i.e., it should
not fail on SIGCHLD if SIGCHLD has not been trapped).
This patch adds a new helper sh_open to retry the open64 call if
necessary. It also calls sh_error when appropriate.
Fixes: 3800d4934391 ("[JOBS] Fix dowait signal race")
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Upstream commit:
Date: Sun, 3 Mar 2019 21:57:50 +0800
eval: Reset handler when entering a subshell
As it is a subshell can execute code that is only meant for the
parent shell when it executes a longjmp that is caught by something
like evalcommand. This patch fixes it by resetting the handler
when entering a subshell.
function old new delta
evalsubshell 169 183 +14
evalpipe 342 356 +14
argstr 1406 1416 +10
ash_main 1236 1226 -10
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 4/1 up/down: 65/-10) Total: 28 bytes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Upstream commit:
Date: Wed, 6 Jan 2021 15:45:12 +1100
jobs: Block signals during tcsetpgrp
Harald van Dijk <harald@gigawatt.nl> wrote:
> On 19/12/2020 22:21, Steffen Nurpmeso wrote:
>> Steffen Nurpmeso wrote in
>> <20201219172838.1B-WB%steffen@sdaoden.eu>:
>> |Long story short, after falsely accusing BSD make of not working
>>
>> After dinner i shortened it a bit more, and attach it again, ok?
>> It is terrible, but now less redundant than before.
>> Sorry for being so terse, that problem crosses my head for about
>> a week, and i was totally mislead and if you bang your head
>> against the wall so many hours bugs or misbehaviours in a handful
>> of other programs is not the expected outcome.
>
> I think a minimal test case is simply
>
> all:
> $(SHELL) -c 'trap "echo TTOU" TTOU; set -m; echo all good'
>
> unless I accidentally oversimplified.
>
> The SIGTTOU is caused by setjobctl's xtcsetpgrp(fd, pgrp) call to make
> its newly started process group the foreground process group when job
> control is enabled, where xtcsetpgrp is a wrapper for tcsetpgrp. (That's
> in dash, the other variants may have some small differences.) tcsetpgrp
> has this little bit in its specification:
>
> Attempts to use tcsetpgrp() from a process which is a member of
> a background process group on a fildes associated with its con‐
> trolling terminal shall cause the process group to be sent a
> SIGTTOU signal. If the calling thread is blocking SIGTTOU sig‐
> nals or the process is ignoring SIGTTOU signals, the process
> shall be allowed to perform the operation, and no signal is
> sent.
>
> Ordinarily, when job control is enabled, SIGTTOU is ignored. However,
> when a trap action is specified for SIGTTOU, the signal is not ignored,
> and there is no blocking in place either, so the tcsetpgrp() call is not
> allowed.
>
> The lowest impact change to make here, the one that otherwise preserves
> the existing shell behaviour, is to block signals before calling
> tcsetpgrp and unblocking them afterwards. This ensures SIGTTOU does not
> get raised here, but also ensures that if SIGTTOU is sent to the shell
> for another reason, there is no window where it gets silently ignored.
>
> Another way to fix this is by not trying to make the shell start a new
> process group, or at least not make it the foreground process group.
> Most other shells appear to not try to do this.
This patch implements the blocking of SIGTTOU (and everything else)
while we call tcsetpgrp.
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Upstream commit:
Date: Sat, 19 May 2018 02:39:56 +0800
eval: Add vfork support
This patch adds basic vfork support for the case of a simple command.
Upstream commit:
Date: Tue, 12 Jan 2021 17:11:19 +1100
jobs: Always reset SIGINT/SIGQUIT handlers
On Fri, Jan 08, 2021 at 08:55:41PM +0000, Harald van Dijk wrote:
> On 18/05/2018 19:39, Herbert Xu wrote:
> > This patch adds basic vfork support for the case of a simple command.
> > ... @@ -879,17 +892,30 @@ forkchild(struct job *jp, union node *n, int
> > mode)
> > }
> > }
> > if (!oldlvl && iflag) {
> > - setsignal(SIGINT);
> > - setsignal(SIGQUIT);
> > + if (mode != FORK_BG) {
> > + setsignal(SIGINT);
> > + setsignal(SIGQUIT);
> > + }
> > setsignal(SIGTERM);
> > }
> > +
> > + if (lvforked)
> > + return;
> > +
> > for (jp = curjob; jp; jp = jp->prev_job)
> > freejob(jp);
> > }
>
> This leaves SIGQUIT ignored in background jobs in interactive shells.
>
> ENV= dash -ic 'dash -c "kill -QUIT \$\$; echo huh" & wait'
>
> As of dash 0.5.11, this prints "huh". Before, the subprocess process killed
> itself before it could print anything. Other shells do not leave SIGQUIT
> ignored.
>
> (In a few other shells, this also prints "huh", but in those other shells,
> that is because the inner shell chooses to ignore SIGQUIT, not because the
> outer shell leaves it ignored.)
Thanks for catching this. I have no idea how that got in there
and it makes no sense whatsoever. This patch removes the if
conditional.
Fixes: e94a964e7dd0 ("eval: Add vfork support")
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Comparing code with dash is more difficult with these differences.
(We didn't know back then that dash will be revived...)
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Upstream commit:
Date: Tue, 6 Dec 2022 16:49:14 +0800
eval: Always set exitstatus in evaltree
There is no harm in setting exitstatus unconditionally in evaltree.
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Upstream commit:
Date: Mon Feb 25 12:49:20 2019 +0800
options: Do not set commandname in procargs
We set commandname in procargs when we don't have to. This results
in a duplicated output of arg0 when an error occurs.
function old new delta
ash_main 1256 1236 -20
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
...and discover that we can do away with 1-3 initialization syscalls.
function old new delta
hush_main 1152 1146 -6
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Also, make second calls to install_special_sighandlers() less haphazard.
It's not a fix for anything, but looking more organized now.
function old new delta
fflush_restore_ttypgrp_and__exit - 20 +20
restore_ttypgrp_and__exit 21 - -21
hush_main 1181 1152 -29
------------------------------------------------------------------------------
(add/remove: 1/1 grow/shrink: 0/1 up/down: 20/-50) Total: -30 bytes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
The rationale here is that unsetting HISTFILE in /etc/profile
does not "stick": if it's unset, the default one is set later
(after /etc/profile is executed) by the shell.
But setting (and exporting, so it is inherited by all
(grand)child shells) an empty one works.
function old new delta
save_history 296 316 +20
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Make the read built-in more compatible with bash:
- Return an exit code of 142 on timeout.
- When the timeout expires before a newline is detected in the
input bash captures the partial input. This behaviour is new
since bash version 4.4. BusyBox shells had the pre-4.4 behaviour
where the input was lost.
Update the tests to suit and fix a couple of compiler errors in
the testsuite.
function old new delta
builtin_read 154 174 +20
readcmd 213 228 +15
shell_builtin_read 1364 1370 +6
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 3/0 up/down: 41/0) Total: 41 bytes
Signed-off-by: Ron Yorston <rmy@pobox.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
When the busybox is used as /sbin/init and the inittab file contains
below:
::respawn:-/bin/sh
/sbin/init spawns hush for the first time with the argv[0] contains '-',
and hush treats it as login shell. Then it reads /etc/profile and if
the file contains the command execution like below, it invokes hush as
login shell because the argv[0] argument is still '-/bin/sh' and reads
/etc/profile again. This will last until some failure (e.g., memory
failure) happens.
[ "$(id -u)" -eq 0 ] && PS1="${PS1}# " || PS1="${PS1}\$ "
This commit fixes this issues by adding an offset (+1) to the
G.argv0_for_re_execing variable.
This issue happens on our out-of-tree UML (use mode linux) with nommu
configuration.
Link: https://lore.kernel.org/all/cover.1731290567.git.thehajime@gmail.com/
Signed-off-by: Hajime Tazaki <thehajime@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Upstream commit:
Date: Fri, 5 Apr 2024 17:55:46 +0800
exec: Check executable bit when searching path
Andrej Shadura <andrew.shadura@collabora.co.uk> wrote:
...
> https://bugs.debian.org/874264
> -------- Forwarded Message --------
> Subject: dash: 'command -v' mistakenly returns a shell script whose
> executable is not set
> Date: Mon, 04 Sep 2017 10:45:48 -0400
> From: Norman Ramsey <nr@cs.tufts.edu>
> To: Debian Bug Tracking System <submit@bugs.debian.org>
...
> I tracked a build bug in s-nail to a problem with dash. Symptom:
> building s-nail tries to run /home/nr/bin/clang, a script whose
> executable bit is not set. We tracked the problem to the result of
> running `command -v clang` with /bin/sh:
...
This is inherited from NetBSD. There is even a commented-out
block of code that tried to fix this.
Anyway, we now have faccessat so we can simply use it.
function old new delta
test_exec - 125 +125
find_command 911 918 +7
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 1/0 up/down: 132/0) Total: 132 bytes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Commit 64f70cc755 (Add --login support) added code in options()
to handle the bash-compatible '--login' option. In doing so it
committed BusyBox ash to silently accepting all other long
options.
Restore compatibility with other ash variants by rejecting unknown
long options.
function old new delta
options 589 624 +35
Signed-off-by: Ron Yorston <rmy@pobox.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
dash has accepted a patch to remove the first argument of findvar().
It's commit e85e972 (var: move hashvar() calls into findvar()).
Apply the same change to BusyBox ash.
function old new delta
findvar 35 40 +5
mklocal 268 261 -7
exportcmd 164 157 -7
setvareq 319 310 -9
lookupvar 150 141 -9
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 1/4 up/down: 5/-32) Total: -27 bytes
Signed-off-by: Ron Yorston <rmy@pobox.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
"echo text >&0000000000002" works as you would expect,
"echo text >&9999999999" properly fails instead of creating a file
named "9999999999".
function old new delta
expredir 219 232 +13
readtoken1 3045 3053 +8
parsefname 204 201 -3
isdigit_str9 45 - -45
------------------------------------------------------------------------------
(add/remove: 0/1 grow/shrink: 2/1 up/down: 21/-48) Total: -27 bytes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
With very large fd#, the error code path is different
from one for closed but small fd#.
Make it not abort if we are interactive:
$ echo text >&99 # this wasn't buggy
ash: dup2(9,1): Bad file descriptor
$ echo text >&9999 # this was
ash: fcntl(1,F_DUPFD,10000): Invalid argument
function old new delta
.rodata 105637 105661 +24
dup2_or_raise 35 38 +3
redirect 1084 1044 -40
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 2/1 up/down: 27/-40) Total: -13 bytes
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
