[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20180718155416.11c9ad26@xeon-e3>
Date: Wed, 18 Jul 2018 15:54:16 -0700
From: Stephen Hemminger <stephen@...workplumber.org>
To: Serhey Popovych <serhe.popovych@...il.com>
Cc: netdev@...r.kernel.org, vincent@...nat.im, dsahern@...il.com
Subject: Re: [PATCH iproute2] ipaddress: Fix and make consistent label match
handling
On Sun, 15 Jul 2018 00:36:34 +0300
Serhey Popovych <serhe.popovych@...il.com> wrote:
> Since commit 9516823051ce ("ipaddress: Improve print_linkinfo()") we
> return -1 instead of 0 when ip-address(8) label does not match network
> device name as we did before change. This causes regression when trying
> to output ip address matching label:
>
> # ip addr add 192.168.192.1/24 dev lo label lo:1
> # ip addr show label lo:1
> <no output>
>
> This is special case and return 0 from print_linkinfo() earlier to match
> only filter.ifindex and filter.up if given, but not rest fields in
> @filter. Then call print_selected_addrinfo() without calling
> print_link_stats() in ipaddr_list_flush_or_save().
>
> Later print_selected_addrinfo() calls print_addrinfo() that finally
> matches IFA_LABEL attribute in netlink buffer with filter.label using
> ifa_label_match_rta().
>
> On the other hand there is three conditions checked in print_linkinfo()
> to determine label special case:
>
> 1) filter.label != NULL
> 2) filter.family == AF_UNSPEC || filter.family == AF_PACKET
> 3) fnmatch(filter.label, name, 0)
>
> With 1) it is ok to check if filtering by label is on by given pattern
> in @filter.label.
>
> Since label is IPv4 specific and AF_PACKET is for printing ip-link(8)
> information (see ipaddr_link_list()::ipaddress.c as example) checking
> for AF_PACKET in 2) doesn't take much sense: better to defer these
> checks to print_addrinfo() determine valid combinations before calling
> ifa_label_match_rta() to finally match IFA_LABEL to pattern in
> filter.label.
>
> For 3) we have following call for test case:
>
> fnmatch(pattern, string, flags) ->
> fnmatch(filter.label, name, 0) ->
> fnmatch("lo:1", "lo", 0) == FNM_NOMATCH (1) or non-zero on error
>
> To support special case in print_linkinfo() for filtering by label we
> only need to check if label pattern is given in filter.label and return
> 0 to skip print_link_stats() in ipaddr_list_flush_or_save(): actual
> filtering will be done in print_addrinfo().
>
> Before commit 9516823051ce ("ipaddress: Improve print_linkinfo()"):
> -------------------------------------------------------------------
>
> $ ip addr sh label lo
> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN \
> group default qlen 1000
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> fnmatch("lo", "lo", 0) == 0
> link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
> inet 127.0.0.1/8 scope host lo
> valid_lft forever preferred_lft forever
> inet6 ::1/128 scope host
> valid_lft forever preferred_lft forever
> $ ip addr show label 'lo:*'
> inet 192.168.192.1/24 scope global lo:1
> valid_lft forever preferred_lft forever
> $ ip addr sh label lo:1
> inet 192.168.192.1/24 scope global lo:1
> valid_lft forever preferred_lft forever
> $ ip -4 addr sh label lo:1
> 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN \
> group default qlen 1000
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> filter.family == AF_INET
> inet 192.168.192.1/24 scope global lo:1
> valid_lft forever preferred_lft forever
>
> After this change applied:
> --------------------------
>
> $ ip/ip addr show label lo
> inet 127.0.0.1/8 scope host lo
> valid_lft forever preferred_lft forever
> inet6 ::1/128 scope host
> valid_lft forever preferred_lft forever
> $ ip/ip addr show label 'lo:*'
> inet 192.168.192.1/24 scope global lo:1
> valid_lft forever preferred_lft forever
> $ ip/ip addr show label lo:1
> inet 192.168.192.1/24 scope global lo:1
> valid_lft forever preferred_lft forever
> $ ip/ip -4 addr show label lo:1
> inet 192.168.192.1/24 scope global lo:1
> valid_lft forever preferred_lft forever
>
> Note that we no longer show link information as we did previously:
> we are filtering by "label" pattern, not showing by "dev".
>
> Fixes: commit 9516823051ce ("ipaddress: Improve print_linkinfo()")
> Reported-by: Vincent Bernat <vincent@...nat.im>
> Signed-off-by: Serhey Popovych <serhe.popovych@...il.com>
Makes sense applied. Thanks for following through on this.
Powered by blists - more mailing lists