> AFAICS, any threads blocked on f_pos_lock are not holding anything else and > cannot impede the rest. What am I missing here?
As far as I understand it is true only for files on regular filesystem like ext4. Let's to see how XEN guys run into trouble with that f_pos_lock: ---------- Forwarded message ---------- From: Marek Marczykowski-Górecki <[email protected]> Date: Thu, Mar 19, 2015 at 3:19 AM Subject: [Xen-devel] Deadlock in /proc/xen/xenbus watch+read on 3.17+ (maybe earlier) To: xen-devel <[email protected]> Cc: Boris Ostrovsky <[email protected]>, David Vrabel <[email protected]> Hi, I've hit some deadlock in kernel xenstore client exposed via /proc/xen/xenbus. Steps to reproduce are simple: int main() { struct xs_handle *xs; xs = xs_open(0); xs_watch(xs, "domid", "token"); xs_read(xs, 0, "name", NULL); return 0; } xs_watch internally creates new thread, which uses read to wait for the watch. And in the same time, the program tries to read some value, but actually it hangs at sending the command (before even sending a path to be read). Strace gives this (simplified for readability): [pid 2494] write(3, "\4\0\0\0\0\0\0\0\0\0\0\0\f\0\0\0", 160 = 16 [pid 2494] write(3, "domid\0", 6) = 6 [pid 2494] write(3, "token\0", 6) = 6 [pid 2495] read(3, <unfinished ...> [pid 2494] futex(0x71c0d4, FUTEX_WAIT_PRIVATE, 1, NULL <unfinished ...> [pid 2495] <... read resumed> "\17\0\0\0\377\377\377\377\220~\255\27\f\0\0\0", 16) = 16 [pid 2495] read(3, "domid\0token\0", 12) = 12 [pid 2495] read(3, "\4\0\0\0\0\0\0\0\0\0\0\0\3\0\0\0", 16) = 16 [pid 2495] read(3, "OK\0", 3) = 3 [pid 2495] futex(0x71c0d4, FUTEX_WAKE_OP_PRIVATE, 1, 1, 0x71c0d0, {FUTEX_OP_SET, 0, FUTEX_OP_CMP_GT, 1} <unfinished ...> [pid 2494] <... futex resumed> ) = 0 [pid 2495] <... futex resumed> ) = 1 [pid 2494] futex(0x71c0a8, FUTEX_WAIT_PRIVATE, 2, NULL <unfinished ...> [pid 2495] futex(0x71c0a8, FUTEX_WAKE_PRIVATE, 1 <unfinished ...> [pid 2494] <... futex resumed> ) = -1 EAGAIN (Resource temporarily unavailable) [pid 2495] <... futex resumed> ) = 0 [pid 2494] futex(0x71c0a8, FUTEX_WAKE_PRIVATE, 1 <unfinished ...> [pid 2495] read(3, <unfinished ...> [pid 2494] <... futex resumed> ) = 0 [pid 2494] rt_sigaction(SIGPIPE, {SIG_DFL, [], SA_RESTORER, 0x7fc78c1488f0}, NULL, 8) = 0 [pid 2494] rt_sigaction(SIGPIPE, {SIG_IGN, [], SA_RESTORER, 0x7fc78c1488f0}, {SIG_DFL, [], SA_RESTORER, 0x7fc78c1488f0}, 8) = 0 [pid 2494] write(3, "\2\0\0\0\0\0\0\0\0\0\0\0\5\0\0\0", 16 And thats all - 2494 is waiting on write, 2495 is waiting on read. On 3.12.x it is working. On 3.17.0 and 3.18.7 it is broken. I haven't checked versions in the middle. Any ideas? -- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab And Iurii Konovalenko has debugged that the root of their troubles is that Commit 9c225f2655e36a470c4f58dbbc99244c5fc7f2d4. With best regards, -- Vitaly Chernooky | Senior Developer - Product Engineering and Development GlobalLogic P +380.44.4929695 ext.1136 M +380.63.6011802 S cvv_2k www.globallogic.com http://www.globallogic.com/email_disclaimer.txt -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
