Bug 1164234

When /etc/lvm/lvm.conf contains "use_lvmetad = 1" and the lvmetad daemon is running, a long running process linked against liblvm2app will connect to a socket to talk with lvmetad, but doesn't close it after use. This leads the process to fail with "too many open files" at some point.

This problem seems to also occur on RHEL6, as well as other distros such as debian. But it's aggravated in RHEL7 by the fact that lvmetad is now enabled by default.

Digging a bit further, it seeems that the socket gets created when "lvm_list_vg_names()" (among other functions, I imagine) is called. And not released (as one would expect) when "lvm_quit()" is called.

The example code added with this patch reproduces the problem:
http://www.redhat.com/archives/lvm-devel/2010-September/msg00025.html

strace running this example code:

[...]
socket(PF_LOCAL, SOCK_STREAM, 0)        = 3
connect(3, {sa_family=AF_LOCAL, sun_path="/run/lvm/lvmetad.socket"}, 110) = 0
write(3, "request = \"hello\"\n", 18)   = 18
write(3, "\n##\n", 4)                   = 4
read(3, "response = \"OK\"\nprotocol = \"lvme", 32) = 32
read(3, "tad\"\nversion = 1\n\n##\n", 1024) = 21
write(3, "request=\"vg_list\"\ntoken =\"filter"..., 36) = 36
write(3, "\n##\n", 4)                   = 4
read(3, "response=\"OK\"\nvolume_groups {\n\tn", 32) = 32
read(3, "f56N6-M9UE-e4rR-jBjK-KqA4-sNme-1"..., 1024) = 64
write(3, "request=\"vg_lookup\"\nuuid =\"nf56N"..., 85) = 85
write(3, "\n##\n", 4)                   = 4
read(3, "response=\"OK\"\nname=\"vg0\"\nmetadat", 32) = 32
read(3, "a {\n\tid=\"nf56N6-M9UE-e4rR-jBjK-K"..., 1024) = 1024
read(3, "10782889\n\t\t\tsegment_count=1\n\t\t\ts"..., 1056) = 1056
read(3, "e=1410782891\n\t\t\tsegment_count=1\n"..., 2112) = 1426
stat("/sys/subsystem", 0x7fffc5060f60)  = -1 ENOENT (No such file or directory)
openat(AT_FDCWD, "/sys/bus", O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC) = 4
getdents(4, /* 22 entries */, 32768)    = 608
open("/usr/lib64/gconv/gconv-modules.cache", O_RDONLY) = 5
fstat(5, {st_mode=S_IFREG|0644, st_size=26254, ...}) = 0
mmap(NULL, 26254, PROT_READ, MAP_SHARED, 5, 0) = 0x7f3318b80000
close(5)                                = 0
futex(0x7f3318436a70, FUTEX_WAKE_PRIVATE, 2147483647) = 0
getdents(4, /* 0 entries */, 32768)     = 0
close(4)                                = 0
[...]

close(3) never happens.


This issue is also discussed here: https://github.com/collectd/collectd/issues/753
Finally, I wonder if this might be related to https://bugzilla.redhat.com/show_bug.cgi?id=1095463

Thanks !