May 2011 - Darwin-dev - lists.apple.com

Re: waitpid() does not wait for process exit
by Anatol Pomozov 04 May '11

04 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:date :message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=ghHZ0ZqqJwNTCgECjlK1y2XShRvWnYdbC9vUWoSkaWA=; b=QDLN0L+q8XqBI6F3GnYr9ubSTeqrwR61Q5eGYqn5bET4y5B4S+F/IvSRlugnt5auz4 uNVUbOUwLEcivD29pY6P0+3qBSgsfbaT37AglTBkPVfMM3XQEidJ7pYUNrqasgyyu3ws /bGxHtpYBui9YaUYJ8ZflVQL/1JWQFCvuc3tg= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type:content-transfer-encoding; b=jh8ixsnkyGHXHtCxGwPaly/AQ9odpRHfgpcN0cHXjtzdVv6GOxFbeIEnvdd4yO6uLr hu3P0iSNlshWpwght6mDRN+N58TgRjItQkKn/Ia43wvLQrN/8e48QrIuKMDVXfpCcZPM lkrnN3jU8Q9zgD8Cr/v88F853QesjKO2N/r1M= Hi On Tue, May 3, 2011 at 9:49 PM, Dave Keck <davekeck(a)gmail.com> wrote: > A quick guess: perhaps your parent process has elected for the kernel > to automatically reap child processes, via a call similar to: > > signal(SIGCHLD, SIG_IGN); > > I would expect the behavior you describe from waitpid() if the above > functionality is enabled. > > See http://en.wikipedia.org/wiki/SIGCHLD. Thanks everyone for your help. If I remove signal(SIGCHLD, SIG_IGN) from the macfuse code my waitpid() starts working again as I expected. What macfuse does is some kind of "process demonization" and should be reimplemented without ignoring SIGCHLD in the parent process. Thanks. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: waitpid() does not wait for process exit
by Bayard Bell 04 May '11

04 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:sender:subject:mime-version:content-type:from :in-reply-to:date:cc:message-id:references:to :content-transfer-encoding:x-pgp-agent:x-mailer; bh=GqKooD2Xx7B/DQ9siNJzoZ+pMfNTTk0Vc08AN2MtIZY=; b=GcYwSG/rWjR9qgkCU/e0icuBhOHmzCrWGsdp3KZWyenfLNv9zR1+DYl9P/QqPyJnzY Po6Zc+UWRrwW8zPGpdGNU3lal4hLBpY1b2ZytYcd8KSfCkDyu2ao/1FMslJqIJeMIR0u lgly/5OOn1dlaQ4HobJZgwGC5NJtLmttTYfmw= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=sender:subject:mime-version:content-type:from:in-reply-to:date:cc :message-id:references:to:content-transfer-encoding:x-pgp-agent :x-mailer; b=Mho9JfDO+5V6+3qAH4gbWtDDdpVTCWUnYZZKd/cgeaVTX4FPALicQe3X5BM+SyyyxG 5G3rq3myyJuOu6bXZYKSo18nZN2oX5L6bRpVCh92TIkP5LetyOEXsGM+wglkIUilRfWZ 22trJEVXGPI3Vw9infwld608nLkNpjJP7A9Uo= It may not be as explicit as you'd like in referencing SIGCHLD, but it does, via the last paragraph below: NOTES See sigaction(2) for a list of termination signals. A status of 0 indicates normal termination. If a parent process terminates without waiting for all of its child processes to terminate, the remain- ing child processes are assigned the parent process 1 ID (the init process ID). If a signal is caught while any of the wait() calls is pending, the call may be interrupted or restarted when the signal-catching routine returns, depending on the options in effect for the signal; see intro(2), System call restart. On 4 May 2011, at 16:33, Anatol Pomozov wrote: > Hi > > On Tue, May 3, 2011 at 9:49 PM, Dave Keck <davekeck(a)gmail.com> wrote: >> A quick guess: perhaps your parent process has elected for the kernel >> to automatically reap child processes, via a call similar to: >> >> signal(SIGCHLD, SIG_IGN); >> >> I would expect the behavior you describe from waitpid() if the above >> functionality is enabled. >> >> See http://en.wikipedia.org/wiki/SIGCHLD. > > Aha, Thanks. > > Although tup does not use SIGCHLD, the macfuse library does. In the > function fuse_mount_core() from > http://macfuse.googlecode.com/svn/trunk/core/10.5/libfuse/fuse-2.7.3-macosx… > > it has > signal(SIGCHLD, SIG_DFL); > and later > signal(SIGCHLD, SIG_IGN); > > Now I am trying to understand why macfuse does it. > > PS "man waitpid" on my macosx does not mention SIGCHLD > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/darwin-dev/buffer.g.overflow%40gmail… > > This email sent to buffer.g.overflow(a)gmail.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com smime.p7s PGP.sig

1 0

Re: waitpid() does not wait for process exit
by Anatol Pomozov 04 May '11

04 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:date :message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=PGEg760b3eqm2UvkHo7g5YploBTlWFU5YKG2Z3PED/E=; b=D2mb3jg8sg7inpM0SvFMlWTE1XBpZUJYuSQUH0/vDM2p5wAp5owHgyUQJhA4gGzHe/ TxrKR1XadTz6DjlB7WGw7adGhiaSym0p/kzBvgXyp2fw9wypU6m/x6/BKrEwWchV4qgI Vkb8aBk+dB+KRHU94hn93gofZqz56xqTLFoMA= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type:content-transfer-encoding; b=Vc2tfVT0tNfD3z8w1s30L14flEc+1lrh+VSnK5jKz0/0OZpTF/vgHpTk4MtVmuvhE2 siwTrQek84ivGPz8NSdsX3Hs/GD3fMmwtKE05JbAeTi5CquntNpKTkW2+swToIARSc2e HGD8i7wK/dSg+PCG+8jlCtHUUJxMntz+sMqP0= Hi On Tue, May 3, 2011 at 9:49 PM, Dave Keck <davekeck(a)gmail.com> wrote: > A quick guess: perhaps your parent process has elected for the kernel > to automatically reap child processes, via a call similar to: > > signal(SIGCHLD, SIG_IGN); > > I would expect the behavior you describe from waitpid() if the above > functionality is enabled. > > See http://en.wikipedia.org/wiki/SIGCHLD. Aha, Thanks. Although tup does not use SIGCHLD, the macfuse library does. In the function fuse_mount_core() from http://macfuse.googlecode.com/svn/trunk/core/10.5/libfuse/fuse-2.7.3-macosx… it has signal(SIGCHLD, SIG_DFL); and later signal(SIGCHLD, SIG_IGN); Now I am trying to understand why macfuse does it. PS "man waitpid" on my macosx does not mention SIGCHLD _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: waitpid() does not wait for process exit
by Jonas Maebe 04 May '11

04 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 04 May 2011, at 06:13, Anatol Pomozov wrote: if (waitpid(pid, &status, 0) < 0) { perror("waitpid); printf("fail waitpid = %d\n", pid); } The code looks pretty much typical, everything seems fine. But the code fails on macosx - waitpid() returns ECHILD error. waitpid() thinks that there is no any child with such pid. Hmm, why? int res; do { res = waitpid(pid, &status, 0); } while ((res < 0) && (errno == EINTR)); if (res < 0) { perror("waitpid); printf("fail waitpid = %d\n", pid); } Jonas _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… Although probably unrelated to the error you are getting, the above code is not safe. You must use something like this instead: There are many low level libc routines that directly map to syscalls and which can return EINTR. In every such case you must use a loop like the above, or at least correctly deal with the special case of EINTR. This email sent to site_archiver(a)lists.apple.com

1 0

Re: waitpid() does not wait for process exit
by Dave Keck 04 May '11

04 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:from:date :message-id:subject:to:cc:content-type; bh=4QtPOCCSLrwmYMa7X6vfId8y9c5h0DNqwvzc9GWH/Sk=; b=ItE+G47ORWJSF/0J5pnge7g6CqI4wKhzuOqb/jlRc/LSGwIq9lAnaaL890n3NyM7o/ cI4WlVHQf5iVHHNWKDvmLczxPxFeSPFsFVCT5XjyRvtoeYI2FOqp45p6FCTaVPK9zM7H 1jwAcZW1GQKO5N2u+JwREh8ZkLLqPkJeI/3qI= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:from:date:message-id:subject:to :cc:content-type; b=LDrPbxxRjwXDAS41RxM5yhFRIO3K8WOFKCCBDdOV0ukxKMvHjpEW4PbsdvdFSGUM7j G7B4C/+uvy5+ihB3td6sOD3GOm2Zg9JsBDj7+tC5/4SLudjBBZhk9jBXYDFM/HomZJSw 6tyFs8gsw7b5wyyTIdn8Vyqdf1PrLvRStfbkg= A quick guess: perhaps your parent process has elected for the kernel to automatically reap child processes, via a call similar to: signal(SIGCHLD, SIG_IGN); I would expect the behavior you describe from waitpid() if the above functionality is enabled. See http://en.wikipedia.org/wiki/SIGCHLD. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

waitpid() does not wait for process exit
by Anatol Pomozov 04 May '11

04 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:date:message-id:subject:from:to :content-type; bh=VXSplRMultyBGtE3VJGcyaPBk+wq/KbxbzlJXwKSLPM=; b=DgOCn7TGeIhslsR/HegW7QJy+z/OCJCSpe9Q1eRcRrxX5Nf7ddjYbQFi+tpt0aHqyB +gFX73Z4bNz39rzACVNJOXH+Gc4qSqQDZr6njVU/BWvuuHaeNJBgA0PATfPonCJwcBpL M3s/mhonRv0EqQppJKA5StgF/H5WAc52/+pho= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:date:message-id:subject:from:to:content-type; b=BB1KtziJ7cXhg2GMxidsZGgPmSReTS48Dl2It+Ofv7rWOehE5727ONEKf6YIsuXpok Wlfww9ElM4XVvf04v8U5dfbpNlk3SBty3SoUxanGNwfVtqL4FkNoGIdzMiFP44UAmnOu mV9s0hf5HL6KRPIaTIIV0akyb5E3WiGv81avE= Hi, I am working on porting tup build tool to Macosx https://github.com/anatol/tup/tree/fuse There is a piece of code that works fine on Linux but fails on osx for some reason. I am trying to understand why. The exact failure point is the waitpid() function at https://github.com/anatol/tup/blob/fuse/src/tup/server/fuse_server.c line 231 Here is a simplified version of the code: pid_t pid = fork(); if (pid == 0) { // fuse fs is served by a thread in the parent process chdir(PATH_TO_FUSE_FS); execl("gcc compile foo.c"); } if (waitpid(pid, &status, 0) < 0) { perror("waitpid); printf("fail waitpid = %d\n", pid); } The code looks pretty much typical, everything seems fine. But the code fails on macosx - waitpid() returns ECHILD error. waitpid() thinks that there is no any child with such pid. Hmm, why? Here is the output of "sudo dtruss ./tup upd -j" maybe it can be useful https://gist.github.com/273ef2e45ceb11824241 The most interested part is near line 5595 where it prints "fail waitpid" 58952/0x19dcc: wait4(0xE64F, 0x100780D9C, 0x0) = -1 Err#10 it fails on waiting of the process 58959 (in hex it is 0xE64F), but the thing is that the process exists - check the logs below at lines 6110. And it makes me puzzled. I carefully read "man waitpid" and did not find anything that can shed any light on this. I am suspecting that either child or parent process (macfuse??) does something that makes kernel think that there is no relation between these 2 processes anymore. RESUME: waitpid() for some reason returns ECHILD for a process that was just recently forked. It does not conform to man page. Is there reason why waitpid() fails? _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Equivalent of openat() on osx?
by Anatol Pomozov 03 May '11

03 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:date :message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=GPvc4d0NX04sGWBu617KZ/cRn/sHmYQ3LcirkAXTTKs=; b=Xn0iXy51thaiujDsItVo17RV/0zQbFDMCCVUVCXX4xem25ZMYxRpybCciHXCPJ5ZDa y6byz1nNp+qDO3nLimbaCp4rUEuBmOC+0kKnMIBj3QsMObEns+eFjp+FPJZ0LEhKEkoG hEaYA2AVlHqRJGCPbJgrb26spstEZHzZ2vrfc= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type:content-transfer-encoding; b=UoWPJQIiQmvB+EvRAxlh2/oxuF7/xMqZFYR1ISsRIld66nbNrfWE0fSeK1EbzALDSq hPI+aNtPjuE2lMXNBSFjzNf1DyxVbaV2sHIP2wa+DPAnv/jIEXxsSFPeybK3ofQSVlrM az/EJwhqdWVFHgKIxWPgTnwdYOAKGp47frGYY= Hi On Tue, May 3, 2011 at 11:34 AM, Paul Nelson <nelson(a)thursby.com> wrote: > > On May 3, 2011, at 12:45 PM, Tilghman Lesher wrote: > >> On Tuesday 03 May 2011 11:13:09 Anatol Pomozov wrote: >>> I am porting a Linux program to macosx (target is 10.6 version) and >>> the program uses openat/unlinkat/... functions. Although these >>> functions are not in POSIX, they are widely used on *nixes and the >>> part proposal the next POSIX. >>> >>> My question what is the best equivalent for the opentat() functions >>> that can be used in multithreaded application. Currently I use >>> something like this: >>> >>> >>> int openat(int dirfd, const char *pathname, int flags, ...) >>> { >>> int fd; >>> >>> if(fchdir(dirfd) < 0) { >>> perror("fchdir"); >>> return -1; >>> } >>> fd = open(pathname, flags, mode); >>> // chdir to the original CWD >>> return fd; >>> } >>> >>> But this is not a perfect solution - current working directory is >>> shared among all threads, it means that openat() may affect other >>> threads. Is there an equivalent of this implementation without >>> changing the current directory? >> >> If you aren't using chroot(2) (or moving directories) in your application, >> you can do this. Wrap calls to opendir(2) and closedir(2), mapping >> descriptors to directory paths (and disposing of them), then in your *at() >> functions, remap the relative pathname back to an absolute pathname. >> You're right that this isn't a perfect compromise, but I think it should >> work, with the caveats above. > > This is exactly what openat is supposed to avoid. You don't want to use a path that is stale in the open call. If you are using openat, you probably have a reason for it. The directory fd passed in might get renamed before the open can be called. > I think the poster was on the right track with fchdir, assuming they also use open(".", O_RDONLY,0) to save the current wd. This will guarantee that the folder does not disappear before the open call happens. I would use pthread mutexes to protect the fchdir stuff, and you might consider something more sophisticated when you receive a blocking open call so the mutex doesn't get held too long. Also, avoid using getcwd, as it is a performance pig. Thanks Paul for the open&fchdir suggestion, initially I was using getcwd&chdir. Here is the new version https://github.com/anatol/tup/blob/fix_openat_macosx/src/compat/openat.c _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Equivalent of openat() on osx?
by Paul Nelson 03 May '11

03 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On May 3, 2011, at 12:45 PM, Tilghman Lesher wrote: > On Tuesday 03 May 2011 11:13:09 Anatol Pomozov wrote: >> I am porting a Linux program to macosx (target is 10.6 version) and >> the program uses openat/unlinkat/... functions. Although these >> functions are not in POSIX, they are widely used on *nixes and the >> part proposal the next POSIX. >> >> My question what is the best equivalent for the opentat() functions >> that can be used in multithreaded application. Currently I use >> something like this: >> >> >> int openat(int dirfd, const char *pathname, int flags, ...) >> { >> int fd; >> >> if(fchdir(dirfd) < 0) { >> perror("fchdir"); >> return -1; >> } >> fd = open(pathname, flags, mode); >> // chdir to the original CWD >> return fd; >> } >> >> But this is not a perfect solution - current working directory is >> shared among all threads, it means that openat() may affect other >> threads. Is there an equivalent of this implementation without >> changing the current directory? > > If you aren't using chroot(2) (or moving directories) in your application, > you can do this. Wrap calls to opendir(2) and closedir(2), mapping > descriptors to directory paths (and disposing of them), then in your *at() > functions, remap the relative pathname back to an absolute pathname. > You're right that this isn't a perfect compromise, but I think it should > work, with the caveats above. This is exactly what openat is supposed to avoid. You don't want to use a path that is stale in the open call. If you are using openat, you probably have a reason for it. The directory fd passed in might get renamed before the open can be called. I think the poster was on the right track with fchdir, assuming they also use open(".", O_RDONLY,0) to save the current wd. This will guarantee that the folder does not disappear before the open call happens. I would use pthread mutexes to protect the fchdir stuff, and you might consider something more sophisticated when you receive a blocking open call so the mutex doesn't get held too long. Also, avoid using getcwd, as it is a performance pig. This way, the open should not fail because the path argument is invalid, which is what openat is designed to do. To avoid having an init function for your mutex, use pthread_once from inside your openat when you find that the mutex is uninitialized. > > Alternatively, you can use a global mutex in your *at functions, ensuring > that only one thread is accessing the environment path, but that > potentially creates a major bottleneck. > > -- > Tilghman _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Equivalent of openat() on osx?
by Anatol Pomozov 03 May '11

03 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:date :message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=ICZwadwDHhEIpGI8Z7Ow9S1cBR6Ib9EnWq6KLwhtnZ4=; b=LxGLuXF7ys/su17lpumTHR+EUMPgZ0XaywpXjbJQqx39Yy2u/2wyw9R5CgVepfJ/3S Y8k4kslC+uPGQwwK6l0mNuiwgJyxW36RYS15UPp2xKV+Jd4WQ2wCNKUPdOJF9vmPwVCR Dbo4m+sDgbtNaDrZ/HrET4/8TpDdJI8sYnWaI= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type:content-transfer-encoding; b=IwH6Q1O16Jz8sNcfazj6KYQdJTm5V2PI/3N3Yvv+H2MeOdBbHQ7GhhS3lvzBmxx4ka 1KCWGYuJbJyAF3Y+gNegmxEBOW7LOQZTWIg1BAT1ORzE6vYRw1LlfOelby4ffarfPo3d QyuEb1IEMks5GYIHohhpUTEFVHXodErFhjAsA= Hi On Tue, May 3, 2011 at 10:45 AM, Tilghman Lesher <tilghman(a)meg.abyt.es> wrote: > On Tuesday 03 May 2011 11:13:09 Anatol Pomozov wrote: >> I am porting a Linux program to macosx (target is 10.6 version) and >> the program uses openat/unlinkat/... functions. Although these >> functions are not in POSIX, they are widely used on *nixes and the >> part proposal the next POSIX. >> >> My question what is the best equivalent for the opentat() functions >> that can be used in multithreaded application. Currently I use >> something like this: >> >> >> int openat(int dirfd, const char *pathname, int flags, ...) >> { >> int fd; >> >> if(fchdir(dirfd) < 0) { >> perror("fchdir"); >> return -1; >> } >> fd = open(pathname, flags, mode); >> // chdir to the original CWD >> return fd; >> } >> >> But this is not a perfect solution - current working directory is >> shared among all threads, it means that openat() may affect other >> threads. Is there an equivalent of this implementation without >> changing the current directory? > > If you aren't using chroot(2) (or moving directories) in your application, > you can do this. Wrap calls to opendir(2) and closedir(2), mapping > descriptors to directory paths (and disposing of them), then in your *at() > functions, remap the relative pathname back to an absolute pathname. > You're right that this isn't a perfect compromise, but I think it should > work, with the caveats above. > > Alternatively, you can use a global mutex in your *at functions, ensuring > that only one thread is accessing the environment path, but that > potentially creates a major bottleneck. Yeah, currently I use a global mutex in my *at() functions. It works fine for me but theoretically we need to guard with the mutex all chdir() calls and file operations that use relative path. Here is the code that I mentioned above https://github.com/anatol/tup/blob/fix_openat_macosx/src/compat/openat.c BTW openat() mentioned in the Open Group specifications http://pubs.opengroup.org/onlinepubs/9699919799/ is there any chance that darwin team implement openat() functions family in the next macosx version? _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Equivalent of openat() on osx?
by Tilghman Lesher 03 May '11

03 May '11

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=meg.abyt.es; s=google; h=domainkey-signature:from:to:subject:date:user-agent:references :in-reply-to:reply-by:x-message:mime-version:content-type :content-transfer-encoding:message-id; bh=iyQphFGee16SsXL6CdCzHqzgpEZaQ7HQVU3DhSKpjc8=; b=MR8VRDF69FIfsa2VRydEPhrLK2IpT4CuW3mP9saFp50Q6oGA0B3p6NokEJ3XCrSg+G 7uXhgUmyCjcPzMkQFKAaN8crQhDlXv5BwzAeZ3AbwFjCNN3a3yfIdjsBNroTn8+t7IAf wUpT6TPC8C5kotmae/1LPMWMhzlj1pcmIQg4I= Domainkey-signature: a=rsa-sha1; c=nofws; d=meg.abyt.es; s=google; h=from:to:subject:date:user-agent:references:in-reply-to:reply-by :x-message:mime-version:content-type:content-transfer-encoding :message-id; b=XcS2F05xZ4RCMfsHXwW69zlKB0N+Mo73nH3Vi9F7KD93hyadHp8zD61WgPuk5rFvCF nhpaEOjD4Zij1AnYWj/+adTURtkKu4x8SGd9sng33c585x5OeltVxyIzdIcXeGzj58oV VcdRIjF3PNQYaeGRMkMdElBsqKJ2o1K/oAOoY= Reply-by: Wed, 19 Jan 2011 16:20:00 -0600 User-agent: KMail/1.13.5 (Linux/2.6.32-31-server; KDE/4.4.5; x86_64; ; ) On Tuesday 03 May 2011 11:13:09 Anatol Pomozov wrote: > I am porting a Linux program to macosx (target is 10.6 version) and > the program uses openat/unlinkat/... functions. Although these > functions are not in POSIX, they are widely used on *nixes and the > part proposal the next POSIX. > > My question what is the best equivalent for the opentat() functions > that can be used in multithreaded application. Currently I use > something like this: > > > int openat(int dirfd, const char *pathname, int flags, ...) > { > int fd; > > if(fchdir(dirfd) < 0) { > perror("fchdir"); > return -1; > } > fd = open(pathname, flags, mode); > // chdir to the original CWD > return fd; > } > > But this is not a perfect solution - current working directory is > shared among all threads, it means that openat() may affect other > threads. Is there an equivalent of this implementation without > changing the current directory? If you aren't using chroot(2) (or moving directories) in your application, you can do this. Wrap calls to opendir(2) and closedir(2), mapping descriptors to directory paths (and disposing of them), then in your *at() functions, remap the relative pathname back to an absolute pathname. You're right that this isn't a perfect compromise, but I think it should work, with the caveats above. Alternatively, you can use a global mutex in your *at functions, ensuring that only one thread is accessing the environment path, but that potentially creates a major bottleneck. -- Tilghman _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0