On 5/13/15, Bart Van Assche <bvanass...@acm.org> wrote: > On 05/13/15 13:59, Dixon Xavier wrote: >> Hi, >> >> This is regarding net-snmp 5.7.3 in HPUX IA64. >> We are using aCC to compile netsnmp libraries. >> >> While running snmptrapd, it is not receiving traps. >> Further analysis shows that, fd_set returned by netsnmp_select_info is >> not correct (incorrect bits are set in it). >> >> Issue seems to be related to follwing LFD_* macros defined in >> snmplib/large_fd_set.c >> >> #define LFD_SET(n, p) ((p)->fds_bits[(n)/NFDBITS] |= (1 << >> ((n) % NFDBITS))) >> #define LFD_CLR(n, p) ((p)->fds_bits[(n)/NFDBITS] &= ~(1 << >> ((n) % NFDBITS))) >> #define LFD_ISSET(n, p) ((p)->fds_bits[(n)/NFDBITS] & (1 << >> ((n) % NFDBITS))) >> >> In HP-UX 11.23 ia64 ( /usr/include/sys/_fd_macros.h) __fd_mask is >> defined as int32_t and FD_* macros operate fds_bits as an array of 32 >> bit items. >> But, the LFD_* macros treat them as 64bit items on IA64 when compiled >> in 64 bit mode. This leads to incorrect bits being set/cleared for >> corresponding fd's >> >> So, the LFD_* macros need to be modified as below so that it is >> handled correctly on all platforms. >> >> #define LFD_SET(n, p) (((__fd_mask >> *)((p)->fds_bits))[(n)/NFDBITS] |= (1ULL << ((n) % NFDBITS))) >> #define LFD_CLR(n, p) (((__fd_mask >> *)((p)->fds_bits))[(n)/NFDBITS] &= ~(1ULL << ((n) % NFDBITS))) >> #define LFD_ISSET(n, p) (((__fd_mask >> *)((p)->fds_bits))[(n)/NFDBITS] & (1ULL << ((n) % NFDBITS))) >> >> I modified the code as above and verified that the traps are now >> received by 'snmptrapd' correctly. >> >> Kindly let us know if you consider this as an 'issue' in NetSNMP and >> if yes when can we expect a patch for the issue. > > Hello Dixon, > > Sorry but I'm afraid that the proposed changes would not compile on > non-HP-UX systems. Could you test the patch below ? > > Thanks, > > Bart. > > [PATCH] Large fd set bug fix for ia64 HP-UX systems > > --- > snmplib/large_fd_set.c | 9 +++++++++ > 1 file changed, 9 insertions(+) > > diff --git a/snmplib/large_fd_set.c b/snmplib/large_fd_set.c > index a90c06b..e591ff2 100644 > --- a/snmplib/large_fd_set.c > +++ b/snmplib/large_fd_set.c > @@ -83,6 +83,7 @@ netsnmp_large_fd_is_set(SOCKET fd, netsnmp_large_fd_set * > fdset) > > #else > > +#if defined(linux) > /* > * Recent versions of glibc trigger abort() if FD_SET(), FD_CLR() or > * FD_ISSET() is invoked with n >= FD_SETSIZE. Hence these replacement > macros. > @@ -90,6 +91,14 @@ netsnmp_large_fd_is_set(SOCKET fd, netsnmp_large_fd_set * > fdset) > #define LFD_SET(n, p) ((p)->fds_bits[(n)/NFDBITS] |= (1ULL << ((n) % > NFDBITS))) > #define LFD_CLR(n, p) ((p)->fds_bits[(n)/NFDBITS] &= ~(1ULL << ((n) % > NFDBITS))) > #define LFD_ISSET(n, p) ((p)->fds_bits[(n)/NFDBITS] & (1ULL << ((n) % > NFDBITS))) > +#else > +/* > + * The above macros do not work correctly on HP-UX ia64. > + */ > +#define LFD_SET(n, p) FD_SET(n, p) > +#define LFD_CLR(n, p) FD_CLR(n, p) > +#define LFD_ISSET(n, p) FD_ISSET(n, p) > +#endif > > void > netsnmp_large_fd_setfd(int fd, netsnmp_large_fd_set * fdset) > -- > 2.1.4 > >
Hello Bart, Thanks for the patch. It worked fine too. We are not using netsnmp's large_fd_set feature in our application and it seems there won't be any issues. However, we are unsure of whether large_fd_set feature works properly in hp-ux. Should we test further ? Also, when will this patch get merged to main branch ? Regards, Dixon Xavier ------------------------------------------------------------------------------ One dashboard for servers and applications across Physical-Virtual-Cloud Widest out-of-the-box monitoring support with 50+ applications Performance metrics, stats and reports that give you Actionable Insights Deep dive visibility with transaction tracing using APM Insight. http://ad.doubleclick.net/ddm/clk/290420510;117567292;y _______________________________________________ Net-snmp-coders mailing list Net-snmp-coders@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/net-snmp-coders
