I killed a terminal with a running bc command. The terminal is gone, but bc(1) runs in an endless loop now: top(1) output: 40325 wosch 1 135 0 13M 3332K CPU0 0 1:14 99.86% bc ktrace -p 40325 kdump 40325 bc CALL read(0,0x217008d92da8,0x1) 40325 bc GIO fd 0 read 0 bytes "" 40325 bc RET read 0 40325 bc CALL ioctl(0,TIOCSETAW,0x3821a3c5b238) 40325 bc RET ioctl -1 errno 25 Inappropriate ioctl for device 40325 bc CALL sigprocmask(SIG_BLOCK,0x217008d92e58,0x217008d92e68) 40325 bc RET sigprocmask 0 40325 bc CALL ioctl(0,TIOCGWINSZ,0x217008d92e10) 40325 bc RET ioctl -1 errno 25 Inappropriate ioctl for device 40325 bc CALL sigprocmask(SIG_SETMASK,0x217008d92e68,0) 40325 bc RET sigprocmask 0 40325 bc CALL read(0,0x217008d92da8,0x1) 40325 bc GIO fd 0 read 0 bytes "" 40325 bc RET read 0 40325 bc CALL ioctl(0,TIOCSETAW,0x3821a3c5b238) 40325 bc RET ioctl -1 errno 25 Inappropriate ioctl for device 40325 bc CALL sigprocmask(SIG_BLOCK,0x217008d92e58,0x217008d92e68) 40325 bc RET sigprocmask 0 40325 bc CALL ioctl(0,TIOCGWINSZ,0x217008d92e10) 40325 bc RET ioctl -1 errno 25 Inappropriate ioctl for device 40325 bc CALL sigprocmask(SIG_SETMASK,0x217008d92e68,0) 40325 bc RET sigprocmask 0 40325 bc CALL read(0,0x217008d92da8,0x1) 40325 bc GIO fd 0 read 0 bytes I tried to kill the process, but it worked only with SIGKILL and not SIGTERM/SIGHUP kill -9 40325
I see this on stable/13, stable14 and current, both amd64 and arm. /usr/bin/bc --version bc 7.0.1
(In reply to Wolfram Schneider from comment #1) the previous version is not affected (as on freefall right now) /usr/bin/bc --version bc 6.7.6
There is no obvious change to bc that would be responsible, but commit 8e022d3cdea10ee1039a632f670c27fd93f65625 changed the way yacc-generated programs process input.
I was looking in the wrong place for bc when I commented on the change history. Source directory usr.bin/bc is not the one that is installed. (That one also behaves oddly in response to ^D, probably due to the "error QUIT" rule in the grammar not causing it to quit.) Here is what truss says when I type ^D to an older bc, version 5.2.2 or 6.7.6: 28787: ioctl(0,TIOCSETAW,0x75c7d3a5b264) = 0 (0x0) 28787: read(0,"\^D",1) = 1 (0x1) 28787: fstat(1,{ mode=crw--w---- ,inode=96,size=0,blksize=4096 }) = 0 (0x0) 28787: ioctl(1,TIOCGETA,0x4c2f87f0125c) = 0 (0x0) 28787: write(1,"^D",2) = 2 (0x2) 28787: ioctl(0,TIOCSETAW,0x75c7d3a5b238) = 0 (0x0) 28787: write(1,"\n",1) = 1 (0x1) 28787: ioctl(0,TIOCSETAF,0x75c7d3a5b20c) = 0 (0x0) 28787: exit(0x0) All is well despite the odd behavior of a literal ^D being read. If I have 7.0.1 then it does not quit and truss says 37822: ioctl(0,TIOCSETAW,0x88d4efe5b264) = 0 (0x0) 37822: read(0,"\^D",1) = 1 (0x1) 37822: fstat(1,{ mode=crw------- ,inode=45,size=0,blksize=4096 }) = 0 (0x0) 37822: ioctl(1,TIOCGETA,0x7ddad4152e7c) = 0 (0x0) 37822: write(1,"^D",2) = 2 (0x2) 37822: ioctl(0,TIOCSETAW,0x88d4efe5b238) = 0 (0x0) 37822: sigprocmask(SIG_BLOCK,{ SIGWINCH },{ }) = 0 (0x0) 37822: ioctl(0,TIOCGWINSZ,0x7ddad4153100) = 0 (0x0) 37822: sigprocmask(SIG_SETMASK,{ },0x0) = 0 (0x0) 37822: ioctl(0,TIOCGETA,0x88d4efe5b290) = 0 (0x0) 37822: ioctl(0,TIOCSETAW,0x88d4efe5b264)