| Message ID | 20201006161520.975-1-nate.karstens@garmin.com |
|---|---|
| State | New |
| Headers | show |
| Series | system.3: Indicate MT-Unsafe | expand |
On 06/10/2020 13:15, Nate Karstens via Libc-alpha wrote: > The fact that system(3) does not support pthread_atfork(3) also means > that it is not thread safe. See the discussion for the proposal of a > close-on-fork flag in the 2020 April and May timeframe, especially: > > https://lkml.org/lkml/2020/5/15/1067 > > Signed-off-by: Nate Karstens <nate.karstens@garmin.com> Not sure if man pages characterizes file descriptor leak as mt-unsafe, at least we don't have this concept on glibc manual. In fact, I think adding a MT-Unsafe mark to this potentially make any libc call that is not atomic potentially MT-Unsafe, either when they do not concurrent trigger race issues regarding memory semantic. At least I think it should add a 'race' mark to indicate what exactly is MT-unsafe (as for other implementations). > --- > man3/system.3 | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/man3/system.3 b/man3/system.3 > index aef40417a..8730fabd3 100644 > --- a/man3/system.3 > +++ b/man3/system.3 > @@ -127,7 +127,7 @@ l l l. > Interface Attribute Value > T{ > .BR system () > -T} Thread safety MT-Safe > +T} Thread safety MT-Unsafe > .TE > .SH CONFORMING TO > POSIX.1-2001, POSIX.1-2008, C89, C99. >
I'm fine with adding a "race" qualifier. Do you have any ideas on the type of race? I didn't see anything in the other man-pages that jumped out as being correct. Thanks, Nate -----Original Message----- From: Adhemerval Zanella <adhemerval.zanella@linaro.org> Sent: Tuesday, October 06, 2020 12:26 To: Karstens, Nate <Nate.Karstens@garmin.com>; mtk.manpages@gmail.com Cc: linux-man@vger.kernel.org; libc-alpha@sourceware.org; willy@infradead.org Subject: Re: [PATCH] system.3: Indicate MT-Unsafe CAUTION - EXTERNAL EMAIL: Do not click any links or open any attachments unless you trust the sender and know the content is safe. On 06/10/2020 13:15, Nate Karstens via Libc-alpha wrote: > The fact that system(3) does not support pthread_atfork(3) also means > that it is not thread safe. See the discussion for the proposal of a > close-on-fork flag in the 2020 April and May timeframe, especially: > > https://urldefense.com/v3/__https://lkml.org/lkml/2020/5/15/1067__;!!E > Jc4YC3iFmQ!D9YVAE760hT-YFoOT14KmIu4y2cjQb8ZflVgpX-3rxgBF2WvxyATUeQogZF > Ffv2sIQ$ > > Signed-off-by: Nate Karstens <nate.karstens@garmin.com> Not sure if man pages characterizes file descriptor leak as mt-unsafe, at least we don't have this concept on glibc manual. In fact, I think adding a MT-Unsafe mark to this potentially make any libc call that is not atomic potentially MT-Unsafe, either when they do not concurrent trigger race issues regarding memory semantic. At least I think it should add a 'race' mark to indicate what exactly is MT-unsafe (as for other implementations). > --- > man3/system.3 | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/man3/system.3 b/man3/system.3 index aef40417a..8730fabd3 > 100644 > --- a/man3/system.3 > +++ b/man3/system.3 > @@ -127,7 +127,7 @@ l l l. > Interface Attribute Value > T{ > .BR system () > -T} Thread safety MT-Safe > +T} Thread safety MT-Unsafe > .TE > .SH CONFORMING TO > POSIX.1-2001, POSIX.1-2008, C89, C99. >
But calling system does not really incur in a race w.r.t the resources of the caller itself and least glibc does handle concurrent sigactions calls and thread cancellation (by reaping the created process). I am not sure about other libc implementation though. The file leakage will be only for the spawn program itself and although it is might characterize as unsafe to call 'system' in multithread environment I don't think this characterize as MT-unsafe (and with FD_CLOEXEC/O_CLOEXEC system is indeed safe in this regarding). So maybe document using a different markup to make it explicit? On 07/10/2020 11:35, Karstens, Nate wrote: > I'm fine with adding a "race" qualifier. Do you have any ideas on the type of race? I didn't see anything in the other man-pages that jumped out as being correct. > > Thanks, > > Nate > > -----Original Message----- > From: Adhemerval Zanella <adhemerval.zanella@linaro.org> > Sent: Tuesday, October 06, 2020 12:26 > To: Karstens, Nate <Nate.Karstens@garmin.com>; mtk.manpages@gmail.com > Cc: linux-man@vger.kernel.org; libc-alpha@sourceware.org; willy@infradead.org > Subject: Re: [PATCH] system.3: Indicate MT-Unsafe > > CAUTION - EXTERNAL EMAIL: Do not click any links or open any attachments unless you trust the sender and know the content is safe. > > > On 06/10/2020 13:15, Nate Karstens via Libc-alpha wrote: >> The fact that system(3) does not support pthread_atfork(3) also means >> that it is not thread safe. See the discussion for the proposal of a >> close-on-fork flag in the 2020 April and May timeframe, especially: >> >> https://urldefense.com/v3/__https://lkml.org/lkml/2020/5/15/1067__;!!E >> Jc4YC3iFmQ!D9YVAE760hT-YFoOT14KmIu4y2cjQb8ZflVgpX-3rxgBF2WvxyATUeQogZF >> Ffv2sIQ$ >> >> Signed-off-by: Nate Karstens <nate.karstens@garmin.com> > > Not sure if man pages characterizes file descriptor leak as mt-unsafe, at least we don't have this concept on glibc manual. In fact, I think adding a MT-Unsafe mark to this potentially make any libc call that is not atomic potentially MT-Unsafe, either when they do not concurrent trigger race issues regarding memory semantic. At least I think it should add a 'race' > mark to indicate what exactly is MT-unsafe (as for other implementations). > >> --- >> man3/system.3 | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/man3/system.3 b/man3/system.3 index aef40417a..8730fabd3 >> 100644 >> --- a/man3/system.3 >> +++ b/man3/system.3 >> @@ -127,7 +127,7 @@ l l l. >> Interface Attribute Value >> T{ >> .BR system () >> -T} Thread safety MT-Safe >> +T} Thread safety MT-Unsafe >> .TE >> .SH CONFORMING TO >> POSIX.1-2001, POSIX.1-2008, C89, C99. >> > > ________________________________ > > CONFIDENTIALITY NOTICE: This email and any attachments are for the sole use of the intended recipient(s) and contain information that may be Garmin confidential and/or Garmin legally privileged. If you have received this email in error, please notify the sender by reply email and delete the message. Any disclosure, copying, distribution or use of this communication (including attachments) by someone other than the intended recipient is prohibited. Thank you. >
On 10/7/20 8:06 PM, Adhemerval Zanella wrote: > But calling system does not really incur in a race w.r.t the resources > of the caller itself and least glibc does handle concurrent sigactions > calls and thread cancellation (by reaping the created process). I am > not sure about other libc implementation though. > > The file leakage will be only for the spawn program itself and although > it is might characterize as unsafe to call 'system' in multithread > environment I don't think this characterize as MT-unsafe (and with > FD_CLOEXEC/O_CLOEXEC system is indeed safe in this regarding). > > So maybe document using a different markup to make it explicit? Any further conclusions out of this? Thanks, Michael > On 07/10/2020 11:35, Karstens, Nate wrote: >> I'm fine with adding a "race" qualifier. Do you have any ideas on the type of race? I didn't see anything in the other man-pages that jumped out as being correct. >> >> Thanks, >> >> Nate >> >> -----Original Message----- >> From: Adhemerval Zanella <adhemerval.zanella@linaro.org> >> Sent: Tuesday, October 06, 2020 12:26 >> To: Karstens, Nate <Nate.Karstens@garmin.com>; mtk.manpages@gmail.com >> Cc: linux-man@vger.kernel.org; libc-alpha@sourceware.org; willy@infradead.org >> Subject: Re: [PATCH] system.3: Indicate MT-Unsafe >> >> CAUTION - EXTERNAL EMAIL: Do not click any links or open any attachments unless you trust the sender and know the content is safe. >> >> >> On 06/10/2020 13:15, Nate Karstens via Libc-alpha wrote: >>> The fact that system(3) does not support pthread_atfork(3) also means >>> that it is not thread safe. See the discussion for the proposal of a >>> close-on-fork flag in the 2020 April and May timeframe, especially: >>> >>> https://urldefense.com/v3/__https://lkml.org/lkml/2020/5/15/1067__;!!E >>> Jc4YC3iFmQ!D9YVAE760hT-YFoOT14KmIu4y2cjQb8ZflVgpX-3rxgBF2WvxyATUeQogZF >>> Ffv2sIQ$ >>> >>> Signed-off-by: Nate Karstens <nate.karstens@garmin.com> >> >> Not sure if man pages characterizes file descriptor leak as mt-unsafe, at least we don't have this concept on glibc manual. In fact, I think adding a MT-Unsafe mark to this potentially make any libc call that is not atomic potentially MT-Unsafe, either when they do not concurrent trigger race issues regarding memory semantic. At least I think it should add a 'race' >> mark to indicate what exactly is MT-unsafe (as for other implementations). >> >>> --- >>> man3/system.3 | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/man3/system.3 b/man3/system.3 index aef40417a..8730fabd3 >>> 100644 >>> --- a/man3/system.3 >>> +++ b/man3/system.3 >>> @@ -127,7 +127,7 @@ l l l. >>> Interface Attribute Value >>> T{ >>> .BR system () >>> -T} Thread safety MT-Safe >>> +T} Thread safety MT-Unsafe >>> .TE >>> .SH CONFORMING TO >>> POSIX.1-2001, POSIX.1-2008, C89, C99. >>> >> >> ________________________________ >> >> CONFIDENTIALITY NOTICE: This email and any attachments are for the sole use of the intended recipient(s) and contain information that may be Garmin confidential and/or Garmin legally privileged. If you have received this email in error, please notify the sender by reply email and delete the message. Any disclosure, copying, distribution or use of this communication (including attachments) by someone other than the intended recipient is prohibited. Thank you. >>
I don't know, my interpretation of what happens is that the race is related to the resources of the caller. Inside of system() the program essentially forked, increasing the OS's reference count of those file descriptors. At the same time, another thread in the caller attempts to close & reopen those resources, but fails because it cannot get exclusive access. In a single-threaded process this would not happen because system() doesn't return until the child process is complete. I think most people will check the thread safety attribute and so changing this will be the most obvious way to indicate a potential problem. But maybe there is some subtlety I'm not picking up on, I'm open to other ideas on how best to communicate this. Thanks, Nate -----Original Message----- From: Michael Kerrisk (man-pages) <mtk.manpages@gmail.com> Sent: Tuesday, October 27, 2020 02:08 To: Adhemerval Zanella <adhemerval.zanella@linaro.org>; Karstens, Nate <Nate.Karstens@garmin.com> Cc: mtk.manpages@gmail.com; linux-man@vger.kernel.org; libc-alpha@sourceware.org; willy@infradead.org Subject: Re: [PATCH] system.3: Indicate MT-Unsafe On 10/7/20 8:06 PM, Adhemerval Zanella wrote: > But calling system does not really incur in a race w.r.t the resources > of the caller itself and least glibc does handle concurrent sigactions > calls and thread cancellation (by reaping the created process). I am > not sure about other libc implementation though. > > The file leakage will be only for the spawn program itself and > although it is might characterize as unsafe to call 'system' in > multithread environment I don't think this characterize as MT-unsafe > (and with FD_CLOEXEC/O_CLOEXEC system is indeed safe in this regarding). > > So maybe document using a different markup to make it explicit? Any further conclusions out of this? Thanks, Michael > On 07/10/2020 11:35, Karstens, Nate wrote: >> I'm fine with adding a "race" qualifier. Do you have any ideas on the type of race? I didn't see anything in the other man-pages that jumped out as being correct. >> >> Thanks, >> >> Nate >> >> -----Original Message----- >> From: Adhemerval Zanella <adhemerval.zanella@linaro.org> >> Sent: Tuesday, October 06, 2020 12:26 >> To: Karstens, Nate <Nate.Karstens@garmin.com>; mtk.manpages@gmail.com >> Cc: linux-man@vger.kernel.org; libc-alpha@sourceware.org; >> willy@infradead.org >> Subject: Re: [PATCH] system.3: Indicate MT-Unsafe >> >> CAUTION - EXTERNAL EMAIL: Do not click any links or open any attachments unless you trust the sender and know the content is safe. >> >> >> On 06/10/2020 13:15, Nate Karstens via Libc-alpha wrote: >>> The fact that system(3) does not support pthread_atfork(3) also >>> means that it is not thread safe. See the discussion for the >>> proposal of a close-on-fork flag in the 2020 April and May timeframe, especially: >>> >>> https://urldefense.com/v3/__https://lkml.org/lkml/2020/5/15/1067__;! >>> !E >>> Jc4YC3iFmQ!D9YVAE760hT-YFoOT14KmIu4y2cjQb8ZflVgpX-3rxgBF2WvxyATUeQog >>> ZF >>> Ffv2sIQ$ >>> >>> Signed-off-by: Nate Karstens <nate.karstens@garmin.com> >> >> Not sure if man pages characterizes file descriptor leak as mt-unsafe, at least we don't have this concept on glibc manual. In fact, I think adding a MT-Unsafe mark to this potentially make any libc call that is not atomic potentially MT-Unsafe, either when they do not concurrent trigger race issues regarding memory semantic. At least I think it should add a 'race' >> mark to indicate what exactly is MT-unsafe (as for other implementations). >> >>> --- >>> man3/system.3 | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/man3/system.3 b/man3/system.3 index >>> aef40417a..8730fabd3 >>> 100644 >>> --- a/man3/system.3 >>> +++ b/man3/system.3 >>> @@ -127,7 +127,7 @@ l l l. >>> Interface Attribute Value >>> T{ >>> .BR system () >>> -T} Thread safety MT-Safe >>> +T} Thread safety MT-Unsafe >>> .TE >>> .SH CONFORMING TO >>> POSIX.1-2001, POSIX.1-2008, C89, C99. >>> >> >> ________________________________ >> >> CONFIDENTIALITY NOTICE: This email and any attachments are for the sole use of the intended recipient(s) and contain information that may be Garmin confidential and/or Garmin legally privileged. If you have received this email in error, please notify the sender by reply email and delete the message. Any disclosure, copying, distribution or use of this communication (including attachments) by someone other than the intended recipient is prohibited. Thank you. >> -- Michael Kerrisk Linux man-pages maintainer; https://urldefense.com/v3/__http://www.kernel.org/doc/man-pages/__;!!EJc4YC3iFmQ!FgWJQcfb_nvuQ86nksCLxQpy18-8nBQ5-8PypqUkGVoLBW9dHZQStNHhiy-HhtQLxw$ Linux/UNIX System Programming Training: https://urldefense.com/v3/__http://man7.org/training/__;!!EJc4YC3iFmQ!FgWJQcfb_nvuQ86nksCLxQpy18-8nBQ5-8PypqUkGVoLBW9dHZQStNHhiy-9TDzqeA$
diff --git a/man3/system.3 b/man3/system.3 index aef40417a..8730fabd3 100644 --- a/man3/system.3 +++ b/man3/system.3 @@ -127,7 +127,7 @@ l l l. Interface Attribute Value T{ .BR system () -T} Thread safety MT-Safe +T} Thread safety MT-Unsafe .TE .SH CONFORMING TO POSIX.1-2001, POSIX.1-2008, C89, C99.
The fact that system(3) does not support pthread_atfork(3) also means that it is not thread safe. See the discussion for the proposal of a close-on-fork flag in the 2020 April and May timeframe, especially: https://lkml.org/lkml/2020/5/15/1067 Signed-off-by: Nate Karstens <nate.karstens@garmin.com> --- man3/system.3 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)