[PATCH v2 1/2] rtnl_link: decode ifinfomsg netlink attributes
JingPiao Chen
chenjingpiao at gmail.com
Sun Aug 20 12:15:47 UTC 2017
On Sun, Aug 20, 2017 at 12:47:56PM +0300, Dmitry V. Levin wrote:
> On Sun, Aug 20, 2017 at 04:06:25PM +0800, JingPiao Chen wrote:
> > On Sun, Aug 20, 2017 at 01:33:53AM +0300, Dmitry V. Levin wrote:
> > > On Sat, Aug 19, 2017 at 09:50:10AM +0800, JingPiao Chen wrote:
> > > > * configure.ac (AC_CHECK_HEADERS): Add linux/if_link.h.
> > > > (AC_CHECK_TYPES): Check for struct rtnl_link_stats64 in linux/if_link.h.
> > > > (AC_CHECK_MEMBERS): Check for rx_nohandler field
> > > > in struct rtnl_link_stats and struct rtnl_link_stats64.
> > > > * rtnl_link.c: Include <arpa/inet.h>, <linux/if_arp.h>,
> > > > <linux/if_link.h> and <linux/netdevice.h>.
> > > > (min_ifla_address_len, ifla_address_default_decoder,
> > > > ifla_address_type_specific_decoder,
> > > > decode_ifla_address, decode_rtnl_link_stats,
> > > > decode_rtnl_link_ifmap, decode_rtnl_link_stats64,
> > > > print_item_id, decode_ifla_phys_item_id): New functions.
> > > > (decode_ifla_phys_item_id): New array.
> > >
> > > How could decode_ifla_phys_item_id be both a function and an array?
> > >
> > > > (decode_ifinfomsg): Use it.
> > > > ---
> > [...]
> > > > +static bool
> > > > +decode_rtnl_link_stats(struct tcb *const tcp,
> > > > + const kernel_ulong_t addr,
> > > > + const unsigned int len,
> > > > + const void *const opaque_data)
> > > > +{
> > > > + struct rtnl_link_stats st;
> > > > +
> > > > + if (len < sizeof(st))
> > > > + return false;
> > >
> > > The kernel may not transfer struct rtnl_link_stats.rx_nohandler despite
> > > the latter being defined by linux/if_link.h, e.g. if the kernel uses
> > > an older version of struct rtnl_link_stats.
> > > The minimal size is therefore not sizeof(struct rtnl_link_stats)
> > > but offsetofend(struct rtnl_link_stats, tx_compressed).
> > >
> > > Likewise, with struct rtnl_link_stats64.rx_nohandler.
> >
> > You means:
> >
> > struct rtnl_link_stats st;
> > const unsigned int sizeof_stat =
> > offsetofend(struct rtnl_link_stats, tx_compressed);
> >
> > if (len < sizeof_stat)
> > return false;
> > else if (!umoven_or_printaddr(tcp, addr, sizeof_stat, &st)) {
> > ...
> >
> > I can not understand where I can know kernel not transfer
> > struct rtnl_link_stats.rx_nohandler.
>
> This can happen, for example, when kernel headers used to build strace
> are newer when the kernel itself.
static bool
decode_rtnl_link_stats(struct tcb *const tcp,
const kernel_ulong_t addr,
const unsigned int len,
const void *const opaque_data)
{
struct rtnl_link_stats st;
const unsigned int sizeof_stat =
offsetofend(struct rtnl_link_stats, tx_compressed);
const unsigned int size = len < sizeof(st) ? sizeof_stat : sizeof(st);
if (len < sizeof(st) && len < sizeof_stat)
return false;
else if (!umoven_or_printaddr(tcp, addr, size, &st)) {
...
#ifdef HAVE_STRUCT_RTNL_LINK_STATS_RX_NOHANDLER
if (len > sizeof_stat)
PRINT_FIELD_U(", ", st, rx_nohandler);
#endif
tprints("}");
}
return true;
}
This is ok?
> > > > +static bool
> > > > +print_item_id(struct tcb *const tcp, void *const elem_buf,
> > > > + const size_t elem_size, void *const opaque_data)
> > > > +{
> > > > + unsigned int *const count = opaque_data;
> > > > +
> > > > + /* MAX_PHYS_ITEM_ID_LEN = 32 */
> > > > + if ((*count)++ >= 32) {
> > > > + tprints("...");
> > > > + return false;
> > > > + }
> > > > +
> > > > + tprintf("%" PRIu8, *(uint8_t *) elem_buf);
> >
> > I changed:
> > tprintf("%02x", *(uint8_t *) elem_buf);
> >
> > > > +
> > > > + return true;
> > > > +}
> > > > +
> > > > +static bool
> > > > +decode_ifla_phys_item_id(struct tcb *const tcp,
> > > > + const kernel_ulong_t addr,
> > > > + const unsigned int len,
> > > > + const void *const opaque_data)
> > > > +{
> > > > + uint8_t id;
> > > > + unsigned int count = 0;
> > > > +
> > > > + print_array(tcp, addr, len, &id, sizeof(id),
> > > > + umoven_or_printaddr, print_item_id, &count);
> > >
> > > I'm not sure it's the best way to decode struct netdev_phys_item_id.id.
> >
> > Or I can reference iproute2/lib/utils.c: hexstring_n2a
> > Now: [xx, xx, xx, xx, xx]
> > hexstring_n2a: xx xx xx xx xx xx
>
> AFAICT, hexstring_n2a outputs xxxxxxxxxxxx.
>
> > strace print xx xx xx xx xx xx seems strange, here space as delimiter,
> > easy to misunderstand it is in memory.
> > what do think?
>
> I wonder wouldn't it be better just to use
> printstr_ex(tcp, addr, len, QUOTE_FORCE_HEX)?
>
> > > > +static const nla_decoder_t ifinfomsg_nla_decoders[] = {
> > > > + [IFLA_ADDRESS] = decode_ifla_address,
> > > > + [IFLA_BROADCAST] = decode_ifla_address,
> > > > + [IFLA_IFNAME] = decode_nla_str,
> > > > + [IFLA_MTU] = decode_nla_u32,
> > > > + [IFLA_LINK] = decode_nla_u32,
> > > > + [IFLA_QDISC] = decode_nla_str,
> > > > + [IFLA_STATS] = decode_rtnl_link_stats,
> > > > + [IFLA_COST] = NULL,
> > > > + [IFLA_PRIORITY] = NULL,
> > > > + [IFLA_MASTER] = decode_nla_u32,
> > > > + [IFLA_WIRELESS] = NULL,
> > >
> > > Here a parser of struct iw_event is expected.
> > >
> > > > + [IFLA_PROTINFO] = NULL,
> > >
> > > This one seems to be used in the kernel.
> > >
> > > > + [IFLA_TXQLEN] = decode_nla_u32,
> > > > + [IFLA_MAP] = decode_rtnl_link_ifmap,
> > > > + [IFLA_WEIGHT] = decode_nla_u32,
> > > > + [IFLA_OPERSTATE] = decode_nla_u8,
> > > > + [IFLA_LINKMODE] = decode_nla_u8,
> > > > + [IFLA_LINKINFO] = NULL,
> > >
> > > This one also seems to be used in the kernel.
> > >
> > > > + [IFLA_NET_NS_PID] = decode_nla_u32,
> > > > + [IFLA_IFALIAS] = decode_nla_str,
> > > > + [IFLA_NUM_VF] = decode_nla_u32,
> > > > + [IFLA_VFINFO_LIST] = NULL,
> > >
> > > Likewise.
> > >
> > > > + [IFLA_STATS64] = decode_rtnl_link_stats64,
> > > > + [IFLA_VF_PORTS] = NULL,
> > >
> > > Likewise.
> > >
> > > > + [IFLA_PORT_SELF] = NULL,
> > >
> > > Likewise.
> > >
> > > > + [IFLA_AF_SPEC] = NULL,
> > >
> > > Likewise.
> > >
> > > > + [IFLA_GROUP] = decode_nla_u32,
> > > > + [IFLA_NET_NS_FD] = decode_nla_u32,
> > > > + [IFLA_EXT_MASK] = decode_nla_u32,
> > > > + [IFLA_PROMISCUITY] = decode_nla_u32,
> > > > + [IFLA_NUM_TX_QUEUES] = decode_nla_u32,
> > > > + [IFLA_NUM_RX_QUEUES] = decode_nla_u32,
> > > > + [IFLA_CARRIER] = decode_nla_u8,
> > > > + [IFLA_PHYS_PORT_ID] = decode_ifla_phys_item_id,
> > > > + [IFLA_CARRIER_CHANGES] = decode_nla_u32,
> > > > + [IFLA_PHYS_SWITCH_ID] = decode_ifla_phys_item_id,
> > > > + [IFLA_LINK_NETNSID] = decode_nla_s32,
> > > > + [IFLA_PHYS_PORT_NAME] = decode_nla_str,
> > > > + [IFLA_PROTO_DOWN] = decode_nla_u8,
> > > > + [IFLA_GSO_MAX_SEGS] = decode_nla_u32,
> > > > + [IFLA_GSO_MAX_SIZE] = decode_nla_u32,
> > > > + [IFLA_PAD] = NULL,
> > > > + [IFLA_XDP] = NULL,
> > >
> > > Likewise.
> >
> > You means that I should decode all the attributes?
>
> There are few attributes that are not used in the kernel,
> for those NULL is fine but a comment, e.g. /* unused */,
> would be nice to have added, too.
>
> > Decode all the attributes in one commit the patch will very big,
>
> There is definitely no need to do it all in a single commit.
>
> > and I'm not familiar with some attribute data, e.g. struct iw_event.
> > So need a lot of time.
>
> How much is a lot? ;)
Ok, I will do my best. But first I will send other patch to review.
--
JingPiao Chen
More information about the Strace-devel
mailing list