search for: pollout

...layout.qual", 0xFFBEFAE0) = 0 lstat64("toolbox/vnet", 0xFFBEFAE0) = 0 lstat64("toolbox/vnet", 0xFFBEF1D8) = 0 lstat64("toolbox/vnet.VNet.attr", 0xFFBEFAE0) = 0 lstat64("toolbox/vnet/vnet.qual", 0xFFBEFAE0) = 0 poll(0xFFBEE7E0, 2, 60000) = 1 fd=1 ev=POLLOUT rev=POLLOUT fd=8 ev=POLLRDNORM rev=0 write(1, "04\0\007FFFFFFFF", 8) = 8 poll(0xFFBEF4D0, 2, 60000) = 1 fd=6 ev=POLLRDNORM rev=POLLRDNORM fd=8 ev=POLLRDNORM rev=0 read(6, "FFFFFFFF", 4) = 4 poll(0xFFBEE850, 2, 60000) = 1 fd=1 ev=POLLOUT rev=POLLOUT fd=8 ev=POL...

...d following) there's now a queue of commands waiting to be issued stored in the handle, and there's no need to issue commands only from the ready state. We can therefore remove the want_to_send logic, queue as many commands as possible, and don't need to test if the socket is ready for POLLOUT. This commit also removes some misleading comments, improves the documentation, and increases the limit on commands in flight (since this limit is effectively in place to stop memory exhaustion, not because of server limits). --- docs/libnbd.pod | 9 ++-- examples/threaded-r...

...eturn the good reply (552). Thank in advance for your help, Nathan => Postfix trace INET Call ************************** connect(21, {sa_family=AF_INET, sin_port=htons(12340), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress) poll([{fd=21, events=POLLOUT}], 1, 100000) = 1 ([{fd=21, revents=POLLOUT}]) getsockopt(21, SOL_SOCKET, SO_ERROR, [0], [4]) = 0 fcntl(21, F_GETFL) = 0x802 (flags O_RDWR|O_NONBLOCK) fcntl(21, F_SETFL, O_RDWR) = 0 fcntl(21, F_GETFD) = 0 fcntl(21, F_SETFD, FD_CLOEXEC)...

...p when the response is complete). The ideal division of labor that would permit reads to be in a separate thread from writes would require that the nbd_aio_pread() call (in the nbdkit thread) run until the entire request is out the door, even if it temporarily blocks on write() and has to wait for POLLOUT. Meanwhile, the reader thread would spin waiting for just POLLIN events, rather than having to service any of the POLLOUT of the writer thread(s) running in parallel. > > The reasons why libnbd can't do this are: > > Problem (a) : Only one thread of control can hold the libnbd h...

...F_GETFL) = 0x2 (flags O_RDWR) fcntl(4, F_SETFL, O_RDWR|O_NONBLOCK) = 0 fcntl(4, F_GETFD) = 0 fcntl(4, F_SETFD, FD_CLOEXEC) = 0 connect(4, {sa_family=AF_FILE, path="/var/run/samba/winbindd/pipe"}, 110) = 0 poll([{fd=4, events=POLLIN|POLLOUT|POLLHUP}], 1, -1) = 1 ([{fd=4, revents=POLLOUT}]) write(4, "0\10\0\0\0\0\0\0\0\0\0\0\17\34\0\0\0\10\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 2096) = 2096 poll([{fd=4, events=POLLIN|POLLHUP}], 1, 5000) = 1 ([{fd=4, revents=POLLIN}]) read(4, "\250\r\0\0\2\0\0\0\33\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0...

On Wed, Jun 19, 2019 at 09:11:52PM -0500, Eric Blake wrote: > When we are blocked waiting for POLLOUT during a request, and happen > to receive notice of POLLIN instead, we know that the work done in > response to POLLIN will be non-blocking (it returns to %.READY as soon > as it would block, which in turn jumps right back into ISSUE_COMMAND > because we have a pending request not fully...

...;;) { /* <-- If you need to issue more commands, do that here. */ dir = nbd_aio_get_direction (nbd); pollfd[0].fd = fd; pollfd[0].events = 0; if (dir & LIBNBD_AIO_DIRECTION_READ) pollfd[0].events |= POLLIN; if (dir & LIBNBD_AIO_DIRECTION_WRITE) pollfd[0].events |= POLLOUT; poll (pollfd, 1, -1); if (pollfd[0].revents & LIBNBD_AIO_DIRECTION_READ) nbd_aio_notify_read (); else if (pollfd[0].revents & LIBNBD_AIO_DIRECTION_WRITE) nbd_aio_notify_write (); } ---------------------------------------------------------------------- The above...

...1697 attached - interrupt to quit gettimeofday({1238146840, 724484}, {0, 0}) = 0 gettimeofday({1238146840, 724542}, NULL) = 0 poll([{fd=4, events=POLLIN|POLLPRI|POLLERR|POLLHUP|POLLNVAL}, {fd=2, events=POLLERR|POLLHUP|POLLNVAL}, {fd=0, events=POLLIN|POLLPRI|POLLERR|POLLHUP|POLLNVAL}, {fd=1, events=POLLOUT|POLLERR|POLLHUP|POLLNVAL}], 4, 3123) = 0 gettimeofday({1238146843, 847540}, {0, 0}) = 0 gettimeofday({1238146843, 847590}, NULL) = 0 poll([{fd=4, events=POLLIN|POLLPRI|POLLERR|POLLHUP|POLLNVAL}, {fd=2, events=POLLERR|POLLHUP|POLLNVAL}, {fd=0, events=POLLIN|POLLPRI|POLLERR|POLLHUP|POLLNVAL}, {fd=1,...

...tv_sec * 1000) * 1000; + timeo = &_timeo; + } + + FD_ZERO(&rfds); + FD_ZERO(&wfds); + FD_ZERO(&efds); + + for (i = 0; i < _nfds; i++) { + fd = _pfd[i].fd; + /* map POLL* into readfds and writefds */ + /* POLLIN -> readfds + * POLLOUT -> writefds + * POLL* -> none + */ + if (_pfd[i].events & POLLIN) + FD_SET(fd, &rfds); + if (_pfd[i].events & POLLOUT) + FD_SET(fd, &wfds); + /* always set exceptfds */ + FD_SET(fd, &efds); + _pfd[i...

...calls like these: [pid 1690] poll([{fd=3, events=POLLIN}, {fd=6, events=POLLIN}, {fd=12, events=POLLIN|POLLERR|POLLHUP}, {fd=11, events=POLLIN|POLLERR|POLLHUP}, {fd=10, events=POLLIN|POLLERR|POLLHUP}, {fd=9, events=POLLIN|POLLERR|POLLHUP}, {fd=24, events=POLLIN|POLLERR|POLLHUP}, {fd=21, events=POLLOUT}, {fd=14, events=POLLIN}, {fd=15, events=POLLIN}, {fd=16, events=POLLIN}, {fd=17, events=POLLIN}, {fd=21, events=POLLIN|POLLERR|POLLHUP}, {fd=20, events=POLLIN|POLLERR|POLLHUP}], 14, -1) = 1 ([{fd=21, revents=POLLOUT}]) It seems that because 1 is returned each time, libvirtd just goes crazy deal...

...ENOENT (No such file or >> directory) >> socket(AF_UNIX, SOCK_STREAM, 0) = 5 >> connect(5, {sa_family=AF_UNIX, >> sun_path="/var/run/libvirt/libvirt-sock"}, 110) = 0 >> getsockname(5, {sa_family=AF_UNIX}, [128->2]) = 0 >> poll([{fd=5, events=POLLOUT}, {fd=6, events=POLLIN}], 2, -1) = 1 >> ([{fd=5, revents=POLLOUT}]) >> poll([{fd=5, events=POLLIN}, {fd=6, events=POLLIN}], 2, -1) = ? >> ERESTART_RESTARTBLOCK (Interrupted by signal) >> --- SIGINT {si_signo=SIGINT, si_code=SI_KERNEL} --- >> +++ killed by SIGINT +++ &g...

It is necessary to kick the state machine any time another thread blocks due to write() encountering a full buffer, to get the reader thread to learn that it must now poll on POLLOUT rather than just POLLIN. POSIX only provides two ways to do this: either poll() on a second fd (our pipe-to-self trick), or interrupt the polling with EINTR due to delivery of a signal. So, I played with the latter approach by running the reader thread with SIGUSR1 blocked except during the poll,...

...src.aio_get_fd () + dfd = dst.aio_get_fd () + + sevents = 0 + devents = 0 + if src.aio_get_direction () & nbd.AIO_DIRECTION_READ: + sevents += select.POLLIN + if src.aio_get_direction () & nbd.AIO_DIRECTION_WRITE: + sevents += select.POLLOUT + if dst.aio_get_direction () & nbd.AIO_DIRECTION_READ: + devents += select.POLLIN + if dst.aio_get_direction () & nbd.AIO_DIRECTION_WRITE: + devents += select.POLLOUT + poll.register (sfd, sevents) + poll.register (dfd, devents) + fo...

...ags O_RDWR) fcntl(5, F_SETFL, O_RDWR|O_NONBLOCK) = 0 fcntl(5, F_GETFD) = 0 fcntl(5, F_SETFD, FD_CLOEXEC) = 0 connect(5, {sa_family=AF_FILE, path="/var/lib/sss/pipes/nss"}, 110) = 0 fstat(5, {st_mode=S_IFSOCK|0777, st_size=0, ...}) = 0 poll([{fd=5, events=POLLOUT}], 1, 300000) = 1 ([{fd=5, revents=POLLOUT}]) sendto(5, "\24\0\0\0\1\0\0\0\0\0\0\0\0\0\0\0", 16, MSG_NOSIGNAL, NULL, 0) = 16 poll([{fd=5, events=POLLOUT}], 1, 300000) = 1 ([{fd=5, revents=POLLOUT}]) sendto(5, "\1\0\0\0", 4, MSG_NOSIGNAL, NULL, 0) = 4 poll([{fd=5, events=POLLIN}]...

When we are blocked waiting for POLLOUT during a request, and happen to receive notice of POLLIN instead, we know that the work done in response to POLLIN will be non-blocking (it returns to %.READY as soon as it would block, which in turn jumps right back into ISSUE_COMMAND because we have a pending request not fully sent yet). Since t...

...xt to write to the server. If using L<poll(2)> @@ -1727,13 +1739,12 @@ you would then call C<nbd_aio_notify_write>. =item C<LIBNBD_AIO_DIRECTION_BOTH> = 3 -We are expected next to either read or write to the server. -If using L<poll(2)> you would set C<events = POLLIN|POLLOUT>. -If one of C<POLLIN> or C<POLLOUT> is returned, then see above. -However note that you shouldn't call C<nbd_aio_notify_read> -and C<nbd_aio_notify_write> because calling the first one will -change the state of the connection, possibly invalidating the second -notifi...

On 6/25/19 4:06 AM, Richard W.M. Jones wrote: > On Wed, Jun 19, 2019 at 09:11:52PM -0500, Eric Blake wrote: >> When we are blocked waiting for POLLOUT during a request, and happen >> to receive notice of POLLIN instead, we know that the work done in >> response to POLLIN will be non-blocking (it returns to %.READY as soon >> as it would block, which in turn jumps right back into ISSUE_COMMAND >> because we have a pending r...

...s is busy: clock_gettime(CLOCK_MONOTONIC, {8158, 708518222}) = 0 timerfd_settime(11, TFD_TIMER_ABSTIME, {it_interval={0, 0}, it_value={8159, 708518000}}, NULL) = 0 ioctl(12, USBDEVFS_SUBMITURB, 0x40810) = 0 poll([{fd=7, events=POLLIN}, {fd=9, events=POLLIN}, {fd=11, events=POLLIN}, {fd=12, events=POLLOUT}], 4, 60000) = 1 ([{fd=12, revents=POLLOUT}]) ioctl(12, USBDEVFS_REAPURBNDELAY, 0xbecf8f24) = 0 timerfd_settime(11, 0, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0 ioctl(12, USBDEVFS_REAPURBNDELAY, 0xbecf8f24) = -1 EAGAIN (Resource temporarily unavailable) clock_gettime(CLOCK_MONOTONIC, {8158,...

...647) = 0 futex(0x7fa716a67378, FUTEX_WAKE_PRIVATE, 2147483647) = 0 write(4, "\1\0\0\0\0\0\0\0", 8) = 8 futex(0x7fa70c001cb0, FUTEX_WAKE_PRIVATE, 1) = 1 write(6, "\1\0\0\0\0\0\0\0", 8) = 8 rt_sigprocmask(SIG_BLOCK, [PIPE CHLD WINCH], [], 8) = 0 poll([{fd=5, events=POLLOUT}, {fd=6, events=POLLIN}], 2, -1) = 2 ([{fd=5, revents=POLLOUT}, {fd=6, revents=POLLIN}]) read(6, "\2\0\0\0\0\0\0\0", 16) = 8 write(6, "\1\0\0\0\0\0\0\0", 8) = 8 write(6, "\1\0\0\0\0\0\0\0", 8) = 8 futex(0x5628ce6e9710, FUTEX_WAKE_PRIVATE, 214748...