Date: Sun, 31 Jan 2021 12:53:22 +0300 From: Rozhuk Ivan <rozhuk.im@gmail.com> To: freebsd-current@freebsd.org Subject: tcsh hang on exit Message-ID: <20210131125322.34d6c52a@rimwks.local>
index | next in thread | raw e-mail
Hi! http://bsd.pw/scale_slides.pdf p 28 recomends add to ~/.tcshrc: set savehist = (99999999 merge lock) I use this for a while, but some time ago found that after term close csh hangs. Term close - close xfce4-terminal by X bytton. Exit in terminal - does not trigger issue. Reproduce: Add/set in ~/.cshrc: set savehist = (99999999 merge lock) Open and close terminal with tsch shell. Run top -aSCI -m io to ensure that csh writes to disk. (13+ only) ~/.history.lock exist, csh process continue running. Truss show infinite loop: ... 78009: link("/home/rim/.rimwks.local.90dc6","/home/rim/.history.lock") ERR#17 'File exists' 78009: unlink("/home/rim/.rimwks.local.90dc6") = 0 (0x0) 78009: sigprocmask(SIG_SETMASK,{ },0x0) = 0 (0x0) 78009: nanosleep({ 0.100000000 }) = 0 (0x0) 78009: sigprocmask(SIG_BLOCK,{ SIGHUP|SIGINT|SIGQUIT|SIGTERM|SIGTSTP|SIGCHLD|SIGTTIN|SIGTTOU },{ }) = 0 (0x0) 78009: __sysctl("kern.hostname",2,0x7fffffffb8a0,0x7fffffffb348,0x0,0) = 0 (0x0) 78009: getpid() = 78009 (0x130b9) 78009: openat(AT_FDCWD,"/home/rim/.rimwks.local.b5e4e",O_WRONLY|O_FSYNC|O_CREAT|O_TRUNC|O_EXCL,00) = 0 (0x0) 78009: close(0) = 0 (0x0) 78009: link("/home/rim/.rimwks.local.b5e4e","/home/rim/.history.lock") ERR#17 'File exists' 78009: unlink("/home/rim/.rimwks.local.b5e4e") = 0 (0x0) 78009: sigprocmask(SIG_SETMASK,{ },0x0) = 0 (0x0) 78009: nanosleep({ 0.100000000 }) = 0 (0x0) 78009: sigprocmask(SIG_BLOCK,{ SIGHUP|SIGINT|SIGQUIT|SIGTERM|SIGTSTP|SIGCHLD|SIGTTIN|SIGTTOU },{ }) = 0 (0x0) 78009: __sysctl("kern.hostname",2,0x7fffffffb8a0,0x7fffffffb348,0x0,0) = 0 (0x0) 78009: getpid() = 78009 (0x130b9) 78009: openat(AT_FDCWD,"/home/rim/.rimwks.local.d0437",O_WRONLY|O_FSYNC|O_CREAT|O_TRUNC|O_EXCL,00) = 0 (0x0) 78009: close(0) = 0 (0x0) 78009: link("/home/rim/.rimwks.local.d0437","/home/rim/.history.lock") ERR#17 'File exists' 78009: unlink("/home/rim/.rimwks.local.d0437") = 0 (0x0) 78009: sigprocmask(SIG_SETMASK,{ },0x0) = 0 (0x0) ... This stop only after I remove empty .history.lock or kill process. Reprodused on 12.2 and 13. amd64. On 13 it causes disk writes with ~2mb/sec per csh process, on 12.2 no disk writes. I have try use "truss -fp PID" to understand what goes wrong, but every time I connect truss before exit/term close - issue does not happen. contrib/tcsh/dotlock.c - code looks simple, and I suspect that some race condition in kernel exist that causes fail at this code: if (st.st_nlink != 2) { then it run first time on process exit, in next time code fails on link - file already exist. I try to reproduce this in small test app, by copy-paste code from contrib/tcsh/dotlock.c but no errors happen and st_nlink == 2. My second version that tcsh do something unusual with signal handling and truss revert it back. 1. Does some one expect same issue/can reproduse it? 2. I have mounted ".eli on /home/rim (ufs, local, noatime, nosuid, soft-updates)", is it OK that FreeBSD 13 write/flush to often/on every file create/delete and FreeBSD 12.2 does not? 3. If I do not use lock option then no chs hang on exit, but if terminal closed without exit file .history.1583640 (diff nums) created. Is any work around for this?home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20210131125322.34d6c52a>