mirror of git://sourceware.org/git/glibc.git
The benchtests/inet_ntop_ipv4 and benchtests/inet_ntop_ipv6 profile shows that most of time is spent in costly sprint operations: $ perf record ./benchtests/bench-inet_ntop_ipv4 && perf report --stdio [...] 38.53% bench-inet_ntop libc.so [.] __printf_buffer 18.69% bench-inet_ntop libc.so [.] __printf_buffer_write 11.01% bench-inet_ntop libc.so [.] _itoa_word 8.02% bench-inet_ntop bench-inet_ntop_ipv4 [.] bench_start 6.99% bench-inet_ntop libc.so [.] __memmove_avx_unaligned_erms 3.86% bench-inet_ntop libc.so [.] __strchrnul_avx2 2.82% bench-inet_ntop libc.so [.] __strcpy_avx2 1.90% bench-inet_ntop libc.so [.] inet_ntop4 1.78% bench-inet_ntop libc.so [.] __vsprintf_internal 1.55% bench-inet_ntop libc.so [.] __sprintf_chk 1.18% bench-inet_ntop libc.so [.] __GI___inet_ntop $ perf record ./benchtests/bench-inet_ntop_ipv6 && perf report --stdio 35.44% bench-inet_ntop libc.so [.] __printf_buffer 14.35% bench-inet_ntop libc.so [.] __printf_buffer_write 10.27% bench-inet_ntop libc.so [.] __GI___inet_ntop 7.93% bench-inet_ntop libc.so [.] _itoa_word 7.00% bench-inet_ntop libc.so [.] __sprintf_chk 6.20% bench-inet_ntop libc.so [.] __vsprintf_internal 5.26% bench-inet_ntop libc.so [.] __strchrnul_avx2 5.05% bench-inet_ntop bench-inet_ntop_ipv6 [.] bench_start 3.70% bench-inet_ntop libc.so [.] __memmove_avx_unaligned_erms 2.11% bench-inet_ntop libc.so [.] __printf_buffer_done A new implementation is used instead: * The printf usage is replaced with an expanded function that prints either an IPv4 octet or an IPv6 quartet; * The strcpy is replaced with a memcpy (since ABIs usually tends to optimize the latter); * For IPv6, the '::' shorthanding is done in-place instead of using a temporary buffer. * An temporary buffer is used iff the size if larger than INET_ADDRSTRLEN/INET6_ADDRSTRLEN. * Inline is used for both inet_ntop4 and inet_ntop6, The code is significand rewrote, so I take this requires a new license. The performance results on aarch64 Neoverse1 with gcc 14.2.1: * master aarch64-linux-gnu-master$ ./benchtests/bench-inet_ntop_ipv4 "inet_ntop_ipv4": { "workload-ipv4-random": { "duration": 1.43067e+09, "iterations": 8e+06, "reciprocal-throughput": 178.572, "latency": 179.096, "max-throughput": 5.59997e+06, "min-throughput": 5.58359e+06 } aarch64-linux-gnu-master$ ./benchtests/bench-inet_ntop_ipv6 "inet_ntop_ipv6": { "workload-ipv6-random": { "duration": 1.68539e+09, "iterations": 4e+06, "reciprocal-throughput": 421.307, "latency": 421.388, "max-throughput": 2.37357e+06, "min-throughput": 2.37311e+06 } } * patched aarch64-linux-gnu$ ./benchtests/bench-inet_ntop_ipv4 "inet_ntop_ipv4": { "workload-ipv4-random": { "duration": 1.06133e+09, "iterations": 5.6e+07, "reciprocal-throughput": 18.8482, "latency": 19.0565, "max-throughput": 5.30555e+07, "min-throughput": 5.24755e+07 } } aarch64-linux-gnu$ ./benchtests/bench-inet_ntop_ipv6 "inet_ntop_ipv6": { "workload-ipv6-random": { "duration": 1.01246e+09, "iterations": 2.4e+07, "reciprocal-throughput": 42.5576, "latency": 41.8139, "max-throughput": 2.34976e+07, "min-throughput": 2.39155e+07 } } Checked on aarch64-linux-gnu and x86_64-linux-gnu. Reviewed-by: DJ Delorie <dj@redhat.com> |
||
---|---|---|
.. | ||
arpa | ||
bits/types | ||
nss_dns | ||
sys | ||
tst-leaks2.root/etc | ||
Depend | ||
Makefile | ||
README | ||
Versions | ||
base64.c | ||
compat-gethnamaddr.c | ||
compat-hooks.c | ||
dn_comp.c | ||
dn_expand.c | ||
dn_skipname.c | ||
gai_cancel.c | ||
gai_error.c | ||
gai_misc.c | ||
gai_misc.h | ||
gai_notify.c | ||
gai_sigqueue.c | ||
gai_suspend.c | ||
getaddrinfo_a.c | ||
herror.c | ||
inet_addr.c | ||
inet_net_ntop.c | ||
inet_net_pton.c | ||
inet_neta.c | ||
inet_ntop.c | ||
inet_pton.c | ||
inet_pton_length.c | ||
libanl-compat.c | ||
mapv4v6addr.h | ||
netdb.h | ||
ns_date.c | ||
ns_makecanon.c | ||
ns_name.c | ||
ns_name_compress.c | ||
ns_name_length_uncompressed.c | ||
ns_name_ntop.c | ||
ns_name_pack.c | ||
ns_name_pton.c | ||
ns_name_skip.c | ||
ns_name_uncompress.c | ||
ns_name_unpack.c | ||
ns_netint.c | ||
ns_parse.c | ||
ns_print.c | ||
ns_rr_cursor_init.c | ||
ns_rr_cursor_next.c | ||
ns_samebinaryname.c | ||
ns_samedomain.c | ||
ns_samename.c | ||
ns_ttl.c | ||
nsap_addr.c | ||
nss_dns_functions.c | ||
res-close.c | ||
res-name-checking.c | ||
res-noaaaa.c | ||
res-putget.c | ||
res-state.c | ||
res_context_hostalias.c | ||
res_data.c | ||
res_debug.c | ||
res_enable_icmp.c | ||
res_get_nsaddr.c | ||
res_hconf.c | ||
res_hconf.h | ||
res_hostalias.c | ||
res_init.c | ||
res_isourserver.c | ||
res_libc.c | ||
res_mkquery.c | ||
res_nameinquery.c | ||
res_queriesmatch.c | ||
res_query.c | ||
res_randomid.c | ||
res_send.c | ||
resolv-deprecated.c | ||
resolv-internal.h | ||
resolv.h | ||
resolv_conf.c | ||
resolv_conf.h | ||
resolv_context.c | ||
resolv_context.h | ||
tst-aton.c | ||
tst-bug18665-tcp.c | ||
tst-bug18665.c | ||
tst-inet_addr-binary.c | ||
tst-inet_aton_exact.c | ||
tst-inet_ntop.c | ||
tst-inet_pton.c | ||
tst-leaks.c | ||
tst-leaks2.c | ||
tst-no-libidn2.c | ||
tst-ns_name.c | ||
tst-ns_name.data | ||
tst-ns_name_compress.c | ||
tst-ns_name_length_uncompressed.c | ||
tst-ns_name_pton.c | ||
tst-ns_rr_cursor.c | ||
tst-ns_samebinaryname.c | ||
tst-p_secstodate.c | ||
tst-res_hconf_reorder.c | ||
tst-res_hnok.c | ||
tst-resolv-ai_idn-common.c | ||
tst-resolv-ai_idn-latin1.c | ||
tst-resolv-ai_idn-nolibidn2.c | ||
tst-resolv-ai_idn.c | ||
tst-resolv-aliases.c | ||
tst-resolv-basic.c | ||
tst-resolv-binary.c | ||
tst-resolv-byaddr.c | ||
tst-resolv-canonname.c | ||
tst-resolv-edns.c | ||
tst-resolv-getaddrinfo-fqdn.c | ||
tst-resolv-invalid-cname.c | ||
tst-resolv-maybe_insert_sig.h | ||
tst-resolv-network.c | ||
tst-resolv-noaaaa-vc.c | ||
tst-resolv-noaaaa.c | ||
tst-resolv-nondecimal.c | ||
tst-resolv-qtypes.c | ||
tst-resolv-res_init-multi.c | ||
tst-resolv-res_init-skeleton.c | ||
tst-resolv-res_init-thread.c | ||
tst-resolv-res_init.c | ||
tst-resolv-res_ninit.c | ||
tst-resolv-rotate.c | ||
tst-resolv-search.c | ||
tst-resolv-semi-failure.c | ||
tst-resolv-short-response.c | ||
tst-resolv-threads.c | ||
tst-resolv-trailing.c | ||
tst-resolv-trustad.c | ||
tst-resolv-txnid-collision.c |
README
The resolver in the GNU C Library ********************************* Starting with version 2.2, the resolver in the GNU C Library comes from BIND 8. Only a subset of the src/lib/resolv part of libbind is included here; basically the parts that are needed to provide the functionality present in the resolver from BIND 4.9.7 that was included in the previous release of the GNU C Library, augmented by the parts needed to provide thread-safety. This means that support for things as dynamic DNS updates and TSIG keys isn't included. If you need those facilities, please take a look at the full BIND distribution. Differences =========== The resolver in the GNU C Library still differs from what's in BIND 8.2.3-T5B: * The RES_DEBUG option (`options debug' in /etc/resolv.conf) has been disabled. * The resolver in glibc allows underscores in domain names. * The <resolv.h> header in glibc includes <netinet/in.h> and <arpa/nameser.h> to make it self-contained. * The `res_close' function in glibc only tries to close open files referenced through `_res' if the RES_INIT bit is set in `_res.options'. This fixes a potential security bug with programs that bogusly call `res_close' without initialising the resolver state first. Note that the thread-safe `res_nclose' still doesn't check the RES_INIT bit. By the way, you're not really supposed to call `res_close/res_nclose' directly. * The resolver in glibc can connect to a nameserver over IPv6. Just specify the IPv6 address in /etc/resolv.conf. You cannot change the address of an IPv6 nameserver dynamically in your program though. Using the resolver in multi-threaded code ========================================= The traditional resolver interfaces `res_query', `res_search', `res_mkquery', `res_send' and `res_init', used a static (global) resolver state stored in the `_res' structure. Therefore, these interfaces are not thread-safe. Therefore, BIND 8.2 introduced a set of "new" interfaces `res_nquery', `res_nsearch', `res_nmkquery', `res_nsend' and `res_ninit' that take a `res_state' as their first argument, so you can use a per-thread resolver state. In glibc, when you link with -lpthread, such a per-thread resolver state is already present. It can be accessed using `_res', which has been redefined as a macro, in a similar way to what has been done for the `errno' and `h_errno' variables. This per-thread resolver state is also used for the `gethostby*' family of functions, which means that for example `gethostbyname_r' is now fully thread-safe and re-entrant. The traditional resolver interfaces however, continue to use a single resolver state and are therefore still thread-unsafe. The resolver state is the same resolver state that is used for the initial ("main") thread. This has the following consequences for existing binaries and source code: * Single-threaded programs will continue to work. There should be no user-visible changes when you recompile them. * Multi-threaded programs that use the traditional resolver interfaces in the "main" thread should continue to work, except that they no longer see any changes in the global resolver state caused by calls to, for example, `gethostbyname' in other threads. Again there should be no user-visible changes when you recompile these programs. * Multi-threaded programs that use the traditional resolver interfaces in more than one thread should be just as buggy as before (there are no problems if you use proper locking of course). If you recompile these programs, manipulating the _res structure in threads other than the "main" thread will seem to have no effect though. * In Multi-threaded that manipulate the _res structure, calls to functions like `gethostbyname' in threads other than the "main" thread won't be influenced by the those changes anymore. We recommend to use the new thread-safe interfaces in new code, since the traditional interfaces have been deprecated by the BIND folks. For compatibility with other (older) systems you might want to continue to use those interfaces though. Using the resolver in C++ code ============================== There resolver contains some hooks which will allow the user to install some callback functions that make it possible to filter DNS requests and responses. Although we do not encourage you to make use of this facility at all, C++ developers should realise that it isn't safe to throw exceptions from such callback functions. Source code =========== The following files come from the BIND distribution (currently version 8.2.3-T5B): src/include/ arpa/nameser.h arpa/nameser_compat.h resolv.h src/lib/resolv/ herror.c res_comp.c res_data.c res_debug.c res_init.c res_mkquery.c res_query.c res_send.c src/lib/nameser/ ns_name.c ns_netint.c ns_parse.c ns_print.c ns_samedomain.c ns_ttl.c src/lib/inet/ inet_addr.c inet_net_ntop.c inet_net_pton.c inet_neta.c inet_ntop.c inet_pton.c nsap_addr.c src/lib/isc/ base64.c Some of these files have been optimised a bit, and adaptations have been made to make them fit in with the rest of glibc. res_libc.c is home-brewn, although parts of it are taken from res_data.c. res_hconf.c and res_hconf.h were contributed by David Mosberger, and do not come from BIND.