| Message ID | 4a0fcf72130d3ef5c4ca91b518f66ac6449cf57f.1622565590.git.gnault@redhat.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Stephen Hemminger |
| Headers | show |
| Series | [iproute2] utils: bump max args number to 256 for batch files | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Not a local patch |
On Tue, 1 Jun 2021 19:09:31 +0200 Guillaume Nault <gnault@redhat.com> wrote: > Large tc filters can have many arguments. For example the following > filter matches the first 7 MPLS LSEs, pops all of them, then updates > the Ethernet header and redirects the resulting packet to eth1. > > filter add dev eth0 ingress handle 44 priority 100 \ > protocol mpls_uc flower mpls \ > lse depth 1 label 1040076 tc 4 bos 0 ttl 175 \ > lse depth 2 label 89648 tc 2 bos 0 ttl 9 \ > lse depth 3 label 63417 tc 5 bos 0 ttl 185 \ > lse depth 4 label 593135 tc 5 bos 0 ttl 67 \ > lse depth 5 label 857021 tc 0 bos 0 ttl 181 \ > lse depth 6 label 239239 tc 1 bos 0 ttl 254 \ > lse depth 7 label 30 tc 7 bos 1 ttl 237 \ > action mpls pop protocol mpls_uc pipe \ > action mpls pop protocol mpls_uc pipe \ > action mpls pop protocol mpls_uc pipe \ > action mpls pop protocol mpls_uc pipe \ > action mpls pop protocol mpls_uc pipe \ > action mpls pop protocol mpls_uc pipe \ > action mpls pop protocol ipv6 pipe \ > action vlan pop_eth pipe \ > action vlan push_eth \ > dst_mac 00:00:5e:00:53:7e \ > src_mac 00:00:5e:00:53:03 pipe \ > action mirred egress redirect dev eth1 > > This filter has 149 arguments, so it can't be used with tc -batch > which is limited to a 100. > > Let's bump the limit to the next power of 2. That should leave a lot of > room for big batch commands. > > Signed-off-by: Guillaume Nault <gnault@redhat.com> Good idea, but we should probably go further up to 512. Also, rather than keeping magic constant. Why not add value to utils.h. I considered using sysconf(_SC_ARG_MAX) gut that is huge on modern machines (2M). And we don't need to allocate for all possible args. diff --git a/include/utils.h b/include/utils.h index 187444d52b41..6c4c403fe6c2 100644 --- a/include/utils.h +++ b/include/utils.h @@ -50,6 +50,9 @@ void incomplete_command(void) __attribute__((noreturn)); #define NEXT_ARG_FWD() do { argv++; argc--; } while(0) #define PREV_ARG() do { argv--; argc++; } while(0) +/* upper limit for batch mode */ +#define MAX_ARGS 512 + #define TIME_UNITS_PER_SEC 1000000 #define NSEC_PER_USEC 1000 #define NSEC_PER_MSEC 1000000 diff --git a/lib/utils.c b/lib/utils.c index 93ae0c55063a..0559923beced 100644 --- a/lib/utils.c +++ b/lib/utils.c @@ -1714,10 +1714,10 @@ int do_batch(const char *name, bool force, cmdlineno = 0; while (getcmdline(&line, &len, stdin) != -1) { - char *largv[100]; + char *largv[MAX_ARGS]; int largc; - largc = makeargs(line, largv, 100); + largc = makeargs(line, largv, MAX_ARGS); if (!largc) continue; /* blank line */
On Wed, Jun 09, 2021 at 04:59:49PM -0700, Stephen Hemminger wrote: > On Tue, 1 Jun 2021 19:09:31 +0200 > Guillaume Nault <gnault@redhat.com> wrote: > > > Large tc filters can have many arguments. For example the following > > filter matches the first 7 MPLS LSEs, pops all of them, then updates > > the Ethernet header and redirects the resulting packet to eth1. > > > > filter add dev eth0 ingress handle 44 priority 100 \ > > protocol mpls_uc flower mpls \ > > lse depth 1 label 1040076 tc 4 bos 0 ttl 175 \ > > lse depth 2 label 89648 tc 2 bos 0 ttl 9 \ > > lse depth 3 label 63417 tc 5 bos 0 ttl 185 \ > > lse depth 4 label 593135 tc 5 bos 0 ttl 67 \ > > lse depth 5 label 857021 tc 0 bos 0 ttl 181 \ > > lse depth 6 label 239239 tc 1 bos 0 ttl 254 \ > > lse depth 7 label 30 tc 7 bos 1 ttl 237 \ > > action mpls pop protocol mpls_uc pipe \ > > action mpls pop protocol mpls_uc pipe \ > > action mpls pop protocol mpls_uc pipe \ > > action mpls pop protocol mpls_uc pipe \ > > action mpls pop protocol mpls_uc pipe \ > > action mpls pop protocol mpls_uc pipe \ > > action mpls pop protocol ipv6 pipe \ > > action vlan pop_eth pipe \ > > action vlan push_eth \ > > dst_mac 00:00:5e:00:53:7e \ > > src_mac 00:00:5e:00:53:03 pipe \ > > action mirred egress redirect dev eth1 > > > > This filter has 149 arguments, so it can't be used with tc -batch > > which is limited to a 100. > > > > Let's bump the limit to the next power of 2. That should leave a lot of > > room for big batch commands. > > > > Signed-off-by: Guillaume Nault <gnault@redhat.com> > > Good idea, but we should probably go further up to 512. > Also, rather than keeping magic constant. Why not add value to > utils.h. Yes, right. > I considered using sysconf(_SC_ARG_MAX) gut that is huge on modern > machines (2M). And we don't need to allocate for all possible args. Yes, 2M is probably overkill (and too much to allocate on the stack). > diff --git a/include/utils.h b/include/utils.h > index 187444d52b41..6c4c403fe6c2 100644 > --- a/include/utils.h > +++ b/include/utils.h > @@ -50,6 +50,9 @@ void incomplete_command(void) __attribute__((noreturn)); > #define NEXT_ARG_FWD() do { argv++; argc--; } while(0) > #define PREV_ARG() do { argv--; argc++; } while(0) > > +/* upper limit for batch mode */ > +#define MAX_ARGS 512 > + > #define TIME_UNITS_PER_SEC 1000000 > #define NSEC_PER_USEC 1000 > #define NSEC_PER_MSEC 1000000 > diff --git a/lib/utils.c b/lib/utils.c > index 93ae0c55063a..0559923beced 100644 > --- a/lib/utils.c > +++ b/lib/utils.c > @@ -1714,10 +1714,10 @@ int do_batch(const char *name, bool force, > > cmdlineno = 0; > while (getcmdline(&line, &len, stdin) != -1) { > - char *largv[100]; > + char *largv[MAX_ARGS]; > int largc; > > - largc = makeargs(line, largv, 100); > + largc = makeargs(line, largv, MAX_ARGS); > if (!largc) > continue; /* blank line */ > > Is this a patch you're going to apply, or should I repost it formally?
On Thu, 10 Jun 2021 09:58:57 +0200 Guillaume Nault <gnault@redhat.com> wrote: > On Wed, Jun 09, 2021 at 04:59:49PM -0700, Stephen Hemminger wrote: > > On Tue, 1 Jun 2021 19:09:31 +0200 > > Guillaume Nault <gnault@redhat.com> wrote: > > > > > Large tc filters can have many arguments. For example the following > > > filter matches the first 7 MPLS LSEs, pops all of them, then updates > > > the Ethernet header and redirects the resulting packet to eth1. > > > > > > filter add dev eth0 ingress handle 44 priority 100 \ > > > protocol mpls_uc flower mpls \ > > > lse depth 1 label 1040076 tc 4 bos 0 ttl 175 \ > > > lse depth 2 label 89648 tc 2 bos 0 ttl 9 \ > > > lse depth 3 label 63417 tc 5 bos 0 ttl 185 \ > > > lse depth 4 label 593135 tc 5 bos 0 ttl 67 \ > > > lse depth 5 label 857021 tc 0 bos 0 ttl 181 \ > > > lse depth 6 label 239239 tc 1 bos 0 ttl 254 \ > > > lse depth 7 label 30 tc 7 bos 1 ttl 237 \ > > > action mpls pop protocol mpls_uc pipe \ > > > action mpls pop protocol mpls_uc pipe \ > > > action mpls pop protocol mpls_uc pipe \ > > > action mpls pop protocol mpls_uc pipe \ > > > action mpls pop protocol mpls_uc pipe \ > > > action mpls pop protocol mpls_uc pipe \ > > > action mpls pop protocol ipv6 pipe \ > > > action vlan pop_eth pipe \ > > > action vlan push_eth \ > > > dst_mac 00:00:5e:00:53:7e \ > > > src_mac 00:00:5e:00:53:03 pipe \ > > > action mirred egress redirect dev eth1 > > > > > > This filter has 149 arguments, so it can't be used with tc -batch > > > which is limited to a 100. > > > > > > Let's bump the limit to the next power of 2. That should leave a lot of > > > room for big batch commands. > > > > > > Signed-off-by: Guillaume Nault <gnault@redhat.com> > > > > Good idea, but we should probably go further up to 512. > > Also, rather than keeping magic constant. Why not add value to > > utils.h. > > Yes, right. > > > I considered using sysconf(_SC_ARG_MAX) gut that is huge on modern > > machines (2M). And we don't need to allocate for all possible args. > > Yes, 2M is probably overkill (and too much to allocate on the stack). > > > diff --git a/include/utils.h b/include/utils.h > > index 187444d52b41..6c4c403fe6c2 100644 > > --- a/include/utils.h > > +++ b/include/utils.h > > @@ -50,6 +50,9 @@ void incomplete_command(void) __attribute__((noreturn)); > > #define NEXT_ARG_FWD() do { argv++; argc--; } while(0) > > #define PREV_ARG() do { argv--; argc++; } while(0) > > > > +/* upper limit for batch mode */ > > +#define MAX_ARGS 512 > > + > > #define TIME_UNITS_PER_SEC 1000000 > > #define NSEC_PER_USEC 1000 > > #define NSEC_PER_MSEC 1000000 > > diff --git a/lib/utils.c b/lib/utils.c > > index 93ae0c55063a..0559923beced 100644 > > --- a/lib/utils.c > > +++ b/lib/utils.c > > @@ -1714,10 +1714,10 @@ int do_batch(const char *name, bool force, > > > > cmdlineno = 0; > > while (getcmdline(&line, &len, stdin) != -1) { > > - char *largv[100]; > > + char *largv[MAX_ARGS]; > > int largc; > > > > - largc = makeargs(line, largv, 100); > > + largc = makeargs(line, largv, MAX_ARGS); > > if (!largc) > > continue; /* blank line */ > > > > > > Is this a patch you're going to apply, or should I repost it formally? > Either way, you get credit
On Thu, Jun 10, 2021 at 04:17:42PM -0700, Stephen Hemminger wrote: > On Thu, 10 Jun 2021 09:58:57 +0200 > Guillaume Nault <gnault@redhat.com> wrote: > > > diff --git a/include/utils.h b/include/utils.h > > > index 187444d52b41..6c4c403fe6c2 100644 > > > --- a/include/utils.h > > > +++ b/include/utils.h > > > @@ -50,6 +50,9 @@ void incomplete_command(void) __attribute__((noreturn)); > > > #define NEXT_ARG_FWD() do { argv++; argc--; } while(0) > > > #define PREV_ARG() do { argv--; argc++; } while(0) > > > > > > +/* upper limit for batch mode */ > > > +#define MAX_ARGS 512 > > > + > > > #define TIME_UNITS_PER_SEC 1000000 > > > #define NSEC_PER_USEC 1000 > > > #define NSEC_PER_MSEC 1000000 > > > diff --git a/lib/utils.c b/lib/utils.c > > > index 93ae0c55063a..0559923beced 100644 > > > --- a/lib/utils.c > > > +++ b/lib/utils.c > > > @@ -1714,10 +1714,10 @@ int do_batch(const char *name, bool force, > > > > > > cmdlineno = 0; > > > while (getcmdline(&line, &len, stdin) != -1) { > > > - char *largv[100]; > > > + char *largv[MAX_ARGS]; > > > int largc; > > > > > > - largc = makeargs(line, largv, 100); > > > + largc = makeargs(line, largv, MAX_ARGS); > > > if (!largc) > > > continue; /* blank line */ > > > > > > > > > > Is this a patch you're going to apply, or should I repost it formally? > > > > Either way, you get credit I've sent v2. Thanks.
diff --git a/lib/utils.c b/lib/utils.c index 93ae0c55..d5496e45 100644 --- a/lib/utils.c +++ b/lib/utils.c @@ -1714,10 +1714,10 @@ int do_batch(const char *name, bool force, cmdlineno = 0; while (getcmdline(&line, &len, stdin) != -1) { - char *largv[100]; + char *largv[256]; int largc; - largc = makeargs(line, largv, 100); + largc = makeargs(line, largv, 256); if (!largc) continue; /* blank line */
Large tc filters can have many arguments. For example the following filter matches the first 7 MPLS LSEs, pops all of them, then updates the Ethernet header and redirects the resulting packet to eth1. filter add dev eth0 ingress handle 44 priority 100 \ protocol mpls_uc flower mpls \ lse depth 1 label 1040076 tc 4 bos 0 ttl 175 \ lse depth 2 label 89648 tc 2 bos 0 ttl 9 \ lse depth 3 label 63417 tc 5 bos 0 ttl 185 \ lse depth 4 label 593135 tc 5 bos 0 ttl 67 \ lse depth 5 label 857021 tc 0 bos 0 ttl 181 \ lse depth 6 label 239239 tc 1 bos 0 ttl 254 \ lse depth 7 label 30 tc 7 bos 1 ttl 237 \ action mpls pop protocol mpls_uc pipe \ action mpls pop protocol mpls_uc pipe \ action mpls pop protocol mpls_uc pipe \ action mpls pop protocol mpls_uc pipe \ action mpls pop protocol mpls_uc pipe \ action mpls pop protocol mpls_uc pipe \ action mpls pop protocol ipv6 pipe \ action vlan pop_eth pipe \ action vlan push_eth \ dst_mac 00:00:5e:00:53:7e \ src_mac 00:00:5e:00:53:03 pipe \ action mirred egress redirect dev eth1 This filter has 149 arguments, so it can't be used with tc -batch which is limited to a 100. Let's bump the limit to the next power of 2. That should leave a lot of room for big batch commands. Signed-off-by: Guillaume Nault <gnault@redhat.com> --- Note: I have no production use case for MPLS stacks with 7 LSEs at the moment, but 7 is the maximum depth the flow dissector can handle, and having the possibility to express such rules with tc -batch would help testing the kernel API (writing scripts that generate filters, without worrying about the 100 parameters limit). lib/utils.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)