Re: [PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
On Wed, 12 Sep 2007 13:28:42 -0700 Rick Jones [EMAIL PROTECTED] wrote: The api in netlink should be in milliseconds rather than compensating in the application (iproute2). My understanding of the in-kernel rtnetlink code is far from complete, but it doesn't seem to have much in the way of provisions for unit conversion, which would suggest no nice suffix-based ui as in tc, and ip is already doing some massaging of units on the display side for a couple of the other parameters, so I'm at something of a loss. So, I used the source and looked and saw that tc seems to convert everything to nanoseconds and passes that to the kernel. The user can give it seconds, milliseconds, microseconds or nanoseconds by using a suffix. It then does something ostensibly intelligent to display those to the user. Ip converts nothing when passing things to the kernel (rtt rttvar or rto - when/if at least the intial rto changes are included - were they?), but when they come-out of the kernel ip converts them to milliseconds. So the units in != the units out. Tc seems much more friendly and less prone to user error. I'm still not sure how easily rtnetlink can do conversions itself - feedback there would be _very_ welcome - but at the very least, having ip provide at least the illusion of what tc does would seem to be a good thing. rick jones Your observations are correct. rtnetlink can't/shouldn't be doing conversions itself. The 'ip' command should use a consistent unit for all values and do conversions if necessary. - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
Your observations are correct. rtnetlink can't/shouldn't be doing conversions itself. The 'ip' command should use a consistent unit for all values and do conversions if necessary. That being the case I'll start looking to see what is involved in leveraging the time conversion stuff in tc for use in ip. rick jones - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
The api in netlink should be in milliseconds rather than compensating in the application (iproute2). My understanding of the in-kernel rtnetlink code is far from complete, but it doesn't seem to have much in the way of provisions for unit conversion, which would suggest no nice suffix-based ui as in tc, and ip is already doing some massaging of units on the display side for a couple of the other parameters, so I'm at something of a loss. So, I used the source and looked and saw that tc seems to convert everything to nanoseconds and passes that to the kernel. The user can give it seconds, milliseconds, microseconds or nanoseconds by using a suffix. It then does something ostensibly intelligent to display those to the user. Ip converts nothing when passing things to the kernel (rtt rttvar or rto - when/if at least the intial rto changes are included - were they?), but when they come-out of the kernel ip converts them to milliseconds. So the units in != the units out. Tc seems much more friendly and less prone to user error. I'm still not sure how easily rtnetlink can do conversions itself - feedback there would be _very_ welcome - but at the very least, having ip provide at least the illusion of what tc does would seem to be a good thing. rick jones - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
Stephen Hemminger wrote: On Tue, 4 Sep 2007 13:20:47 -0700 (PDT) Rick Jones [EMAIL PROTECTED] wrote: Build upon David Miller's initial patches to set the per-route rto_min so users can specify the rto_min in the same units (milliseconds) in which they are displayed. This is desirable because asking users to convert to and from jiffies themselves, when there can be different values of HZ from system to system would be error prone. Signed-off-by: Rick Jones [EMAIL PROTECTED] --- The api in netlink should be in milliseconds rather than compensating in the application (iproute2). My understanding of the in-kernel rtnetlink code is far from complete, but it doesn't seem to have much in the way of provisions for unit conversion, which would suggest no nice suffix-based ui as in tc, and ip is already doing some massaging of units on the display side for a couple of the other parameters, so I'm at something of a loss. rick jones - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
On Tue, 4 Sep 2007 13:20:47 -0700 (PDT) Rick Jones [EMAIL PROTECTED] wrote: Build upon David Miller's initial patches to set the per-route rto_min so users can specify the rto_min in the same units (milliseconds) in which they are displayed. This is desirable because asking users to convert to and from jiffies themselves, when there can be different values of HZ from system to system would be error prone. Signed-off-by: Rick Jones [EMAIL PROTECTED] --- The api in netlink should be in milliseconds rather than compensating in the application (iproute2). - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
Stephen Hemminger wrote: On Tue, 4 Sep 2007 13:20:47 -0700 (PDT) Rick Jones [EMAIL PROTECTED] wrote: Build upon David Miller's initial patches to set the per-route rto_min so users can specify the rto_min in the same units (milliseconds) in which they are displayed. This is desirable because asking users to convert to and from jiffies themselves, when there can be different values of HZ from system to system would be error prone. Signed-off-by: Rick Jones [EMAIL PROTECTED] --- The api in netlink should be in milliseconds rather than compensating in the application (iproute2). In unrelated work I found myself using tc to setup some WAN emulation. There I noticed that one can specify times using a units suffix. So, one can say 100ms, 100us, or 100s. That seems like a nice, friendly interface. I've not looked yet to see if it was converting up in tc and then passing a consistent unit to the kernel, or if all the conversion was happening in the kernel. rick jones - To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] [RFC] allow admin/users to specify rto_min in milliseconds rather than jiffies
Build upon David Miller's initial patches to set the per-route rto_min so users can specify the rto_min in the same units (milliseconds) in which they are displayed. This is desirable because asking users to convert to and from jiffies themselves, when there can be different values of HZ from system to system would be error prone. Signed-off-by: Rick Jones [EMAIL PROTECTED] --- *** ./lib/utils.c.orig 2007-09-04 13:11:01.0 -0700 --- ./lib/utils.c 2007-09-04 13:06:11.0 -0700 *** *** 61,66 --- 61,141 return 0; } + /* liberally lifted from kernel/time.c. raj */ + unsigned int jiffies_to_msecs(const unsigned long j) + { + if (__iproute2_hz_internal == 0) + __iproute2_hz_internal = __get_hz(); + + if ((__iproute2_hz_internal = 1000) + !(1000 % __iproute2_hz_internal)) + return (1000 / __iproute2_hz_internal) * j; + else if (__iproute2_hz_internal 1000 + !(__iproute2_hz_internal % 1000)) + return (j + (__iproute2_hz_internal / 1000) - 1) / + (__iproute2_hz_internal / 1000); + else + return (j * 1000) / __iproute2_hz_internal; + } + + /* liberally lifted from kernel/time.c raj */ + unsigned long msecs_to_jiffies(const unsigned int m) + { + + if (__iproute2_hz_internal == 0) + __iproute2_hz_internal = __get_hz(); + + if (__iproute2_hz_internal = 1000 + !(1000 % __iproute2_hz_internal)) + /* +* HZ is equal to or smaller than 1000, and 1000 is a nice +* round multiple of HZ, divide with the factor between them, +* but round upwards: +*/ + return (m + (1000 / __iproute2_hz_internal) - 1) / + (1000 / __iproute2_hz_internal); + else if (__iproute2_hz_internal 1000 + !(__iproute2_hz_internal % 1000)) { + /* +* HZ is larger than 1000, and HZ is a nice round multiple of +* 1000 - simply multiply with the factor between them. +* +* But first make sure the multiplication result cannot +* overflow: +*/ + if (m jiffies_to_msecs(MAX_JIFFY_OFFSET)) + return MAX_JIFFY_OFFSET; + + return m * (__iproute2_hz_internal / 1000); + } + else { + /* +* Generic case - multiply, round and divide. But first +* check that if we are doing a net multiplication, that +* we wouldnt overflow: +*/ + if (__iproute2_hz_internal 1000 + m jiffies_to_msecs(MAX_JIFFY_OFFSET)) + return MAX_JIFFY_OFFSET; + + return (m * __iproute2_hz_internal + 1000 - 1) / 1000; + } + } + + int get_unsigned_ms_to_jiffies(unsigned *val, const char *arg, int base) + { + int ret; + unsigned myval = *val; + + if (__iproute2_hz_internal == 0) + __iproute2_hz_internal = __get_hz(); + ret = get_unsigned(myval,arg,base); + if (ret == -1 || myval (UINT_MAX / __iproute2_hz_internal)) + return -1; + *val = msecs_to_jiffies(myval); + return 0; + } + int get_u64(__u64 *val, const char *arg, int base) { unsigned long long res; *** ./include/utils.h.orig 2007-09-04 12:56:09.0 -0700 --- ./include/utils.h 2007-09-04 12:55:56.0 -0700 *** *** 69,74 --- 69,78 u_int8_t ipx_node[IPX_NODE_LEN]; }; + #ifndef MAX_JIFFY_OFFSET + #define MAX_JIFFY_OFFSET ((LONG_MAX 1)-1) + #endif + extern __u32 get_addr32(const char *name); extern int get_addr_1(inet_prefix *dst, const char *arg, int family); extern int get_prefix_1(inet_prefix *dst, char *arg, int family); *** *** 77,82 --- 81,87 extern int get_integer(int *val, const char *arg, int base); extern int get_unsigned(unsigned *val, const char *arg, int base); + extern int get_unsigned_ms_to_jiffies(unsigned *val, const char *arg, int base); #define get_byte get_u8 #define get_ushort get_u16 #define get_short get_s16 *** ./include/linux/rtnetlink.h.orig2007-03-13 14:50:56.0 -0700 --- ./include/linux/rtnetlink.h 2007-08-31 13:27:36.0 -0700 *** *** 352,357 --- 352,359 #define RTAX_INITCWND RTAX_INITCWND RTAX_FEATURES, #define RTAX_FEATURES RTAX_FEATURES + RTAX_RTO_MIN, + #define RTAX_RTO_MIN RTAX_RTO_MIN __RTAX_MAX }; *** ./ip/iproute.c.orig 2007-08-31 13:15:45.0 -0700 --- ./ip/iproute.c 2007-09-04 13:02:42.0 -0700 *** *** 51,56 --- 51,57 [RTAX_HOPLIMIT] = hoplimit, [RTAX_INITCWND] = initcwnd, [RTAX_FEATURES] = features, +