Add port of high performance pipe implementation written by John S. Dyson
for FreeBSD project. Besides huge speed boost compared with socketpair-based pipes, this implementation also uses pagable kernel memory instead of mbufs. Significant differences to FreeBSD version: * uses uvm_loan() facility for direct write * async/SIGIO handling correct also for sync writer, async reader * limits settable via sysctl, amountpipekva and nbigpipes available via sysctl * pipes are unidirectional - this is enforced on file descriptor level for now only, the code would be updated to take advantage of it eventually * uses lockmgr(9)-based locks instead of home brew variant * scatter-gather write is handled correctly for direct write case, data is transferred by PIPE_DIRECT_CHUNK bytes maximum, to avoid running out of kva All FreeBSD/NetBSD specific code is within appropriate #ifdef, in preparation to feed changes back to FreeBSD tree. This pipe implementation is optional for now, add 'options NEW_PIPE' to your kernel config to use it.
This commit is contained in:
jdolecek
parent
02d39f046b
commit
ee882e3a09
|
|
@ -1,4 +1,4 @@
|
|||
.\" $NetBSD: sysctl.8,v 1.66 2001/03/27 02:24:39 itojun Exp $
|
||||
.\" $NetBSD: sysctl.8,v 1.67 2001/06/16 12:00:04 jdolecek Exp $
|
||||
.\"
|
||||
.\" Copyright (c) 1993
|
||||
.\" The Regents of the University of California. All rights reserved.
|
||||
|
|
@ -198,7 +198,11 @@ privilege can change the value.
|
|||
.It kern.osrevision integer no
|
||||
.It kern.ostype string no
|
||||
.It kern.path_max integer no
|
||||
.It kern.pipe_buf integer no
|
||||
.It kern.pipe.maxkvasz integer yes
|
||||
.It kern.pipe.maxloankvasz integer yes
|
||||
.It kern.pipe.maxbigpipes integer yes
|
||||
.It kern.pipe.nbigpipes integer no
|
||||
.It kern.pipe.kvasize integer no
|
||||
.It kern.posix1version integer no
|
||||
.It kern.rawpartition integer no
|
||||
.It kern.saved_ids integer no
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
/* $NetBSD: sysctl.c,v 1.46 2001/03/09 01:02:11 chs Exp $ */
|
||||
/* $NetBSD: sysctl.c,v 1.47 2001/06/16 12:00:04 jdolecek Exp $ */
|
||||
|
||||
/*
|
||||
* Copyright (c) 1993
|
||||
|
|
@ -44,7 +44,7 @@ __COPYRIGHT(
|
|||
#if 0
|
||||
static char sccsid[] = "@(#)sysctl.c 8.1 (Berkeley) 6/6/93";
|
||||
#else
|
||||
__RCSID("$NetBSD: sysctl.c,v 1.46 2001/03/09 01:02:11 chs Exp $");
|
||||
__RCSID("$NetBSD: sysctl.c,v 1.47 2001/06/16 12:00:04 jdolecek Exp $");
|
||||
#endif
|
||||
#endif /* not lint */
|
||||
|
||||
|
|
@ -181,6 +181,7 @@ static int sysctl_key __P((char *, char **, int[], int, int *));
|
|||
static int sysctl_vfs __P((char *, char **, int[], int, int *));
|
||||
static int sysctl_vfsgen __P((char *, char **, int[], int, int *));
|
||||
static int sysctl_mbuf __P((char *, char **, int[], int, int *));
|
||||
static int sysctl_pipe __P((char *, char **, int[], int, int *));
|
||||
static int sysctl_proc __P((char *, char **, int[], int, int *));
|
||||
static int findname __P((char *, char *, char **, struct list *));
|
||||
static void usage __P((void));
|
||||
|
|
@ -397,6 +398,11 @@ parse(string, flags)
|
|||
case KERN_CONSDEV:
|
||||
special |= CONSDEV;
|
||||
break;
|
||||
case KERN_PIPE:
|
||||
len = sysctl_pipe(string, &bufp, mib, flags, &type);
|
||||
if (len < 0)
|
||||
return;
|
||||
break;
|
||||
}
|
||||
break;
|
||||
|
||||
|
|
@ -1119,6 +1125,33 @@ sysctl_mbuf(string, bufpp, mib, flags, typep)
|
|||
return (3);
|
||||
}
|
||||
|
||||
struct ctlname pipenames[] = CTL_PIPE_NAMES;
|
||||
struct list pipevars = { pipenames, KERN_PIPE_MAXID };
|
||||
/*
|
||||
* handle kern.mbuf requests
|
||||
*/
|
||||
static int
|
||||
sysctl_pipe(string, bufpp, mib, flags, typep)
|
||||
char *string;
|
||||
char **bufpp;
|
||||
int mib[];
|
||||
int flags;
|
||||
int *typep;
|
||||
{
|
||||
struct list *lp = &pipevars;
|
||||
int indx;
|
||||
|
||||
if (*bufpp == NULL) {
|
||||
listall(string, lp);
|
||||
return (-1);
|
||||
}
|
||||
if ((indx = findname(string, "third", bufpp, lp)) == -1)
|
||||
return (-1);
|
||||
mib[2] = indx;
|
||||
*typep = lp->list[indx].ctl_type;
|
||||
return (3);
|
||||
}
|
||||
|
||||
/*
|
||||
* Scan a list of names searching for a particular name.
|
||||
*/
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
# $NetBSD: files,v 1.446 2001/06/13 06:01:44 simonb Exp $
|
||||
# $NetBSD: files,v 1.447 2001/06/16 12:00:03 jdolecek Exp $
|
||||
|
||||
# @(#)files.newconf 7.5 (Berkeley) 5/10/93
|
||||
|
||||
|
|
@ -11,6 +11,7 @@ defopt MAXUPRC
|
|||
defopt RTC_OFFSET
|
||||
defopt DEFCORENAME
|
||||
defopt UCONSOLE
|
||||
defopt NEW_PIPE
|
||||
|
||||
defopt MULTIPROCESSOR
|
||||
|
||||
|
|
@ -863,6 +864,7 @@ file kern/subr_prof.c
|
|||
file kern/subr_rmap.c
|
||||
file kern/subr_xxx.c
|
||||
file kern/sys_generic.c
|
||||
file kern/sys_pipe.c new_pipe
|
||||
file kern/sys_process.c
|
||||
file kern/sys_socket.c
|
||||
file kern/syscalls.c syscall_debug
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
/* $NetBSD: init_main.c,v 1.191 2001/06/08 12:53:30 mrg Exp $ */
|
||||
/* $NetBSD: init_main.c,v 1.192 2001/06/16 12:00:02 jdolecek Exp $ */
|
||||
|
||||
/*
|
||||
* Copyright (c) 1995 Christopher G. Demetriou. All rights reserved.
|
||||
|
|
@ -46,6 +46,7 @@
|
|||
#include "opt_sysv.h"
|
||||
#include "opt_maxuprc.h"
|
||||
#include "opt_multiprocessor.h"
|
||||
#include "opt_new_pipe.h"
|
||||
#include "opt_syscall_debug.h"
|
||||
|
||||
#include "rnd.h"
|
||||
|
|
@ -91,6 +92,9 @@
|
|||
#if NRND > 0
|
||||
#include <sys/rnd.h>
|
||||
#endif
|
||||
#ifdef NEW_PIPE
|
||||
#include <sys/pipe.h>
|
||||
#endif
|
||||
|
||||
#include <sys/syscall.h>
|
||||
#include <sys/syscallargs.h>
|
||||
|
|
@ -489,6 +493,11 @@ main(void)
|
|||
/* Initialize exec structures */
|
||||
exec_init(1);
|
||||
|
||||
#ifdef NEW_PIPE
|
||||
/* Initialize pipe structures */
|
||||
pipe_init();
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Okay, now we can let init(8) exec! It's off to userland!
|
||||
*/
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
/* $NetBSD: kern_sysctl.c,v 1.88 2001/04/26 06:07:13 enami Exp $ */
|
||||
/* $NetBSD: kern_sysctl.c,v 1.89 2001/06/16 12:00:02 jdolecek Exp $ */
|
||||
|
||||
/*-
|
||||
* Copyright (c) 1982, 1986, 1989, 1993
|
||||
|
|
@ -45,6 +45,7 @@
|
|||
#include "opt_ddb.h"
|
||||
#include "opt_insecure.h"
|
||||
#include "opt_defcorename.h"
|
||||
#include "opt_new_pipe.h"
|
||||
#include "opt_sysv.h"
|
||||
#include "pty.h"
|
||||
|
||||
|
|
@ -334,6 +335,7 @@ kern_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
|
|||
case KERN_MBUF:
|
||||
case KERN_PROC_ARGS:
|
||||
case KERN_SYSVIPC_INFO:
|
||||
case KERN_PIPE:
|
||||
/* Not terminal. */
|
||||
break;
|
||||
default:
|
||||
|
|
@ -534,6 +536,11 @@ kern_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
|
|||
#if NPTY > 0
|
||||
case KERN_MAXPTYS:
|
||||
return sysctl_pty(oldp, oldlenp, newp, newlen);
|
||||
#endif
|
||||
#ifdef NEW_PIPE
|
||||
case KERN_PIPE:
|
||||
return (sysctl_dopipe(name + 1, namelen - 1, oldp, oldlenp,
|
||||
newp, newlen));
|
||||
#endif
|
||||
default:
|
||||
return (EOPNOTSUPP);
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load Diff
|
|
@ -1,4 +1,4 @@
|
|||
/* $NetBSD: uipc_syscalls.c,v 1.59 2001/06/14 20:32:47 thorpej Exp $ */
|
||||
/* $NetBSD: uipc_syscalls.c,v 1.60 2001/06/16 12:00:02 jdolecek Exp $ */
|
||||
|
||||
/*
|
||||
* Copyright (c) 1982, 1986, 1989, 1990, 1993
|
||||
|
|
@ -36,6 +36,7 @@
|
|||
*/
|
||||
|
||||
#include "opt_ktrace.h"
|
||||
#include "opt_new_pipe.h"
|
||||
|
||||
/*
|
||||
* Though COMPAT_OLDSOCK is needed only for COMPAT_43, SunOS, Linux,
|
||||
|
|
@ -912,6 +913,7 @@ sys_getsockopt(struct proc *p, void *v, register_t *retval)
|
|||
return (error);
|
||||
}
|
||||
|
||||
#ifndef NEW_PIPE
|
||||
/* ARGSUSED */
|
||||
int
|
||||
sys_pipe(struct proc *p, void *v, register_t *retval)
|
||||
|
|
@ -965,6 +967,7 @@ sys_pipe(struct proc *p, void *v, register_t *retval)
|
|||
(void)soclose(rso);
|
||||
return (error);
|
||||
}
|
||||
#endif /* !NEW_PIPE */
|
||||
|
||||
/*
|
||||
* Get socket name.
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
/* $NetBSD: malloc.h,v 1.62 2001/05/30 12:07:05 mrg Exp $ */
|
||||
/* $NetBSD: malloc.h,v 1.63 2001/06/16 12:00:03 jdolecek Exp $ */
|
||||
|
||||
/*
|
||||
* Copyright (c) 1987, 1993
|
||||
|
|
@ -171,7 +171,8 @@
|
|||
#define M_EMULDATA 115 /* Per-process emulation data */
|
||||
#define M_1394CTL 116 /* IEEE 1394 control structures */
|
||||
#define M_1394DATA 117 /* IEEE 1394 data buffers */
|
||||
#define M_LAST 118 /* Must be last type + 1 */
|
||||
#define M_PIPE 118 /* Pipe structures */
|
||||
#define M_LAST 119 /* Must be last type + 1 */
|
||||
|
||||
#define INITKMEMNAMES { \
|
||||
"free", /* 0 M_FREE */ \
|
||||
|
|
@ -292,7 +293,8 @@
|
|||
"emuldata", /* 115 M_EMULDATA */ \
|
||||
"1394ctl", /* 116 M_1394CTL */ \
|
||||
"1394data", /* 117 M_1394DATA */ \
|
||||
NULL, /* 118 */ \
|
||||
"pipe", /* 118 M_PIPE */ \
|
||||
NULL, /* 119 */ \
|
||||
}
|
||||
|
||||
struct kmemstats {
|
||||
|
|
|
|||
|
|
@ -25,11 +25,13 @@
|
|||
#define _SYS_PIPE_H_
|
||||
|
||||
#ifndef _KERNEL
|
||||
#include <sys/time.h> /* for struct timespec */
|
||||
#ifdef __FreeBSD__
|
||||
#include <sys/time.h> /* for struct timeval */
|
||||
#include <sys/selinfo.h> /* for struct selinfo */
|
||||
#include <vm/vm.h> /* for vm_page_t */
|
||||
#include <machine/param.h> /* for PAGE_SIZE */
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Pipe buffer size, keep moderate in value, pipes take kva space.
|
||||
|
|
@ -42,15 +44,27 @@
|
|||
#define BIG_PIPE_SIZE (64*1024)
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Maximum size of kva for direct write transfer. If the amount
|
||||
* of data in buffer is larger, it would be transferred in chunks of this
|
||||
* size. This kva memory is freed after use if amount of pipe kva memory
|
||||
* is bigger than limitpipekva.
|
||||
*/
|
||||
#ifndef PIPE_DIRECT_CHUNK
|
||||
#define PIPE_DIRECT_CHUNK (1*1024*1024)
|
||||
#endif
|
||||
|
||||
/*
|
||||
* PIPE_MINDIRECT MUST be smaller than PIPE_SIZE and MUST be bigger
|
||||
* than PIPE_BUF.
|
||||
*/
|
||||
#ifndef PIPE_MINDIRECT
|
||||
#if defined(__FreeBSD__)
|
||||
#define PIPE_MINDIRECT 8192
|
||||
#elif defined(__NetBSD__)
|
||||
#define PIPE_MINDIRECT PAGE_SIZE
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#define PIPENPAGES (BIG_PIPE_SIZE / PAGE_SIZE + 1)
|
||||
|
||||
/*
|
||||
* Pipe buffer information.
|
||||
|
|
@ -58,17 +72,21 @@
|
|||
* Buffered write is active when the buffer.cnt field is set.
|
||||
*/
|
||||
struct pipebuf {
|
||||
u_int cnt; /* number of chars currently in buffer */
|
||||
size_t cnt; /* number of chars currently in buffer */
|
||||
u_int in; /* in pointer */
|
||||
u_int out; /* out pointer */
|
||||
u_int size; /* size of buffer */
|
||||
size_t size; /* size of buffer */
|
||||
caddr_t buffer; /* kva of buffer */
|
||||
#ifdef __FreeBSD__
|
||||
struct vm_object *object; /* VM object containing buffer */
|
||||
#endif
|
||||
};
|
||||
|
||||
/*
|
||||
* Information to support direct transfers between processes for pipes.
|
||||
*/
|
||||
#if defined(__FreeBSD__)
|
||||
#define PIPENPAGES (BIG_PIPE_SIZE / PAGE_SIZE + 1)
|
||||
struct pipemapping {
|
||||
vm_offset_t kva; /* kernel virtual address */
|
||||
vm_size_t cnt; /* number of chars in buffer */
|
||||
|
|
@ -76,6 +94,15 @@ struct pipemapping {
|
|||
int npages; /* number of pages */
|
||||
vm_page_t ms[PIPENPAGES]; /* pages in source process */
|
||||
};
|
||||
#elif defined(__NetBSD__)
|
||||
struct pipemapping {
|
||||
vaddr_t kva; /* kernel virtual address */
|
||||
vsize_t cnt; /* number of chars in buffer */
|
||||
voff_t pos; /* current position within page */
|
||||
int npages; /* how many pages allocated */
|
||||
struct vm_page **ms;
|
||||
};
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Bits in pipe_state.
|
||||
|
|
@ -83,13 +110,17 @@ struct pipemapping {
|
|||
#define PIPE_ASYNC 0x004 /* Async? I/O. */
|
||||
#define PIPE_WANTR 0x008 /* Reader wants some characters. */
|
||||
#define PIPE_WANTW 0x010 /* Writer wants space to put characters. */
|
||||
#define PIPE_WANT 0x020 /* Pipe is wanted to be run-down. */
|
||||
#define PIPE_WANTCLOSE 0x020 /* Pipe is wanted to be run-down. */
|
||||
#define PIPE_SEL 0x040 /* Pipe has a select active. */
|
||||
#define PIPE_EOF 0x080 /* Pipe is in EOF condition. */
|
||||
#define PIPE_LOCK 0x100 /* Process has exclusive access to pointers/data. */
|
||||
#define PIPE_LWANT 0x200 /* Process wants exclusive access to pointers/data. */
|
||||
#define PIPE_DIRECTW 0x400 /* Pipe direct write active. */
|
||||
#define PIPE_DIRECTOK 0x800 /* Direct mode ok. */
|
||||
#define PIPE_SIGNALR 0x800 /* Do selwakeup() on read(2) */
|
||||
|
||||
#ifdef __NetBSD__
|
||||
#define PIPE_MOREW 0x2000 /* Writer has more data to write. */
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Per-pipe data structure.
|
||||
|
|
@ -99,13 +130,25 @@ struct pipe {
|
|||
struct pipebuf pipe_buffer; /* data storage */
|
||||
struct pipemapping pipe_map; /* pipe mapping for direct I/O */
|
||||
struct selinfo pipe_sel; /* for compat with select */
|
||||
#ifdef __FreeBSD__
|
||||
struct timespec pipe_atime; /* time of last access */
|
||||
struct timespec pipe_mtime; /* time of last modify */
|
||||
struct timespec pipe_ctime; /* time of status change */
|
||||
struct sigio *pipe_sigio; /* information for async I/O */
|
||||
#elif defined(__NetBSD__)
|
||||
struct timeval pipe_atime; /* time of last access */
|
||||
struct timeval pipe_mtime; /* time of last modify */
|
||||
struct timeval pipe_ctime; /* time of status change */
|
||||
gid_t pipe_pgid; /* process group for sigio */
|
||||
struct lock pipe_lock; /* pipe lock */
|
||||
#endif
|
||||
struct pipe *pipe_peer; /* link with other direction */
|
||||
u_int pipe_state; /* pipe status info */
|
||||
int pipe_busy; /* busy flag, mostly to handle rundown sanely */
|
||||
};
|
||||
|
||||
#ifdef __NetBSD__
|
||||
void pipe_init __P((void));
|
||||
#endif
|
||||
|
||||
#endif /* !_SYS_PIPE_H_ */
|
||||
|
|
|
|||
|
|
@ -1,4 +1,4 @@
|
|||
/* $NetBSD: sysctl.h,v 1.61 2001/05/04 15:50:13 simonb Exp $ */
|
||||
/* $NetBSD: sysctl.h,v 1.62 2001/06/16 12:00:03 jdolecek Exp $ */
|
||||
|
||||
/*
|
||||
* Copyright (c) 1989, 1993
|
||||
|
|
@ -173,7 +173,8 @@ struct ctlname {
|
|||
#define KERN_MSGBUF 53 /* kernel message buffer */
|
||||
#define KERN_CONSDEV 54 /* dev_t: console terminal device */
|
||||
#define KERN_MAXPTYS 55 /* int: maximum number of ptys */
|
||||
#define KERN_MAXID 56 /* number of valid kern ids */
|
||||
#define KERN_PIPE 56 /* node: pipe limits */
|
||||
#define KERN_MAXID 57 /* number of valid kern ids */
|
||||
|
||||
#define CTL_KERN_NAMES { \
|
||||
{ 0, 0 }, \
|
||||
|
|
@ -232,6 +233,7 @@ struct ctlname {
|
|||
{ "msgbuf", CTLTYPE_STRUCT }, \
|
||||
{ "consdev", CTLTYPE_STRUCT }, \
|
||||
{ "maxptys", CTLTYPE_INT }, \
|
||||
{ "pipe", CTLTYPE_NODE }, \
|
||||
}
|
||||
|
||||
/*
|
||||
|
|
@ -428,6 +430,25 @@ struct kinfo_proc2 {
|
|||
#define KERN_SYSVIPC_SEM_INFO 2 /* seminfo and semid_ds */
|
||||
#define KERN_SYSVIPC_SHM_INFO 3 /* shminfo and shmid_ds */
|
||||
|
||||
/*
|
||||
* KERN_PIPE subtypes
|
||||
*/
|
||||
#define KERN_PIPE_MAXKVASZ 1 /* maximum kva size */
|
||||
#define KERN_PIPE_LIMITKVA 2 /* */
|
||||
#define KERN_PIPE_MAXBIGPIPES 3 /* maximum # of "big" pipes */
|
||||
#define KERN_PIPE_NBIGPIPES 4 /* current number of "big" p. */
|
||||
#define KERN_PIPE_KVASIZE 5 /* current pipe kva size */
|
||||
#define KERN_PIPE_MAXID 6
|
||||
|
||||
#define CTL_PIPE_NAMES { \
|
||||
{ 0, 0 }, \
|
||||
{ "maxkvasz", CTLTYPE_INT }, \
|
||||
{ "maxloankvasz", CTLTYPE_INT }, \
|
||||
{ "maxbigpipes", CTLTYPE_INT }, \
|
||||
{ "nbigpipes", CTLTYPE_INT }, \
|
||||
{ "kvasize", CTLTYPE_INT }, \
|
||||
}
|
||||
|
||||
/*
|
||||
* CTL_HW identifiers
|
||||
*/
|
||||
|
|
@ -652,6 +673,9 @@ int sysctl_ntptime __P((void *, size_t *));
|
|||
int sysctl_doprof __P((int *, u_int, void *, size_t *, void *, size_t));
|
||||
#endif
|
||||
int sysctl_dombuf __P((int *, u_int, void *, size_t *, void *, size_t));
|
||||
#ifdef NEW_PIPE
|
||||
int sysctl_dopipe __P((int *, u_int, void *, size_t *, void *, size_t));
|
||||
#endif
|
||||
|
||||
void fill_eproc __P((struct proc *, struct eproc *));
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue