freebsd-dev/sys/kern/sysv_shm.c

891 lines
21 KiB
C
Raw Normal View History

1999-08-28 01:08:13 +00:00
/* $FreeBSD$ */
/* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */
/*
* Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Adam Glass and Charles
* Hannum.
* 4. The names of the authors may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "opt_compat.h"
#include "opt_sysvipc.h"
#include <sys/param.h>
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/sysctl.h>
#include <sys/shm.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/mman.h>
2001-05-21 18:52:02 +00:00
#include <sys/mutex.h>
#include <sys/stat.h>
#include <sys/syscall.h>
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
#include <sys/sysent.h>
#include <sys/sysproto.h>
#include <sys/jail.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_object.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
#include <vm/vm_pager.h>
static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
struct oshmctl_args;
2002-03-19 21:25:46 +00:00
static int oshmctl(struct thread *td, struct oshmctl_args *uap);
2002-03-19 21:25:46 +00:00
static int shmget_allocate_segment(struct thread *td,
struct shmget_args *uap, int mode);
static int shmget_existing(struct thread *td, struct shmget_args *uap,
int mode, int segnum);
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
/* XXX casting to (sy_call_t *) is bogus, as usual. */
1998-02-09 06:11:36 +00:00
static sy_call_t *shmcalls[] = {
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
(sy_call_t *)shmat, (sy_call_t *)oshmctl,
(sy_call_t *)shmdt, (sy_call_t *)shmget,
(sy_call_t *)shmctl
};
1995-05-30 08:16:23 +00:00
#define SHMSEG_FREE 0x0200
#define SHMSEG_REMOVED 0x0400
#define SHMSEG_ALLOCATED 0x0800
#define SHMSEG_WANTED 0x1000
static int shm_last_free, shm_nused, shm_committed, shmalloced;
static struct shmid_ds *shmsegs;
struct shm_handle {
/* vm_offset_t kva; */
vm_object_t shm_object;
};
struct shmmap_state {
vm_offset_t va;
int shmid;
};
2002-03-19 21:25:46 +00:00
static void shm_deallocate_segment(struct shmid_ds *);
static int shm_find_segment_by_key(key_t);
static struct shmid_ds *shm_find_segment_by_shmid(int);
static struct shmid_ds *shm_find_segment_by_shmidx(int);
static int shm_delete_mapping(struct proc *p, struct shmmap_state *);
static void shmrealloc(void);
static void shminit(void);
static int sysvshm_modload(struct module *, int, void *);
static int shmunload(void);
static void shmexit_myhook(struct proc *p);
static void shmfork_myhook(struct proc *p1, struct proc *p2);
static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS);
/*
* Tuneable values.
*/
#ifndef SHMMAXPGS
#define SHMMAXPGS 8192 /* Note: sysv shared memory is swap backed. */
#endif
#ifndef SHMMAX
#define SHMMAX (SHMMAXPGS*PAGE_SIZE)
#endif
#ifndef SHMMIN
#define SHMMIN 1
#endif
#ifndef SHMMNI
#define SHMMNI 192
#endif
#ifndef SHMSEG
#define SHMSEG 128
#endif
#ifndef SHMALL
#define SHMALL (SHMMAXPGS)
#endif
struct shminfo shminfo = {
SHMMAX,
SHMMIN,
SHMMNI,
SHMSEG,
SHMALL
};
static int shm_use_phys;
SYSCTL_DECL(_kern_ipc);
SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RD, &shminfo.shmseg, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, "");
SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW,
&shm_use_phys, 0, "");
SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD,
NULL, 0, sysctl_shmsegs, "", "");
static int
shm_find_segment_by_key(key)
key_t key;
{
int i;
for (i = 0; i < shmalloced; i++)
if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
shmsegs[i].shm_perm.key == key)
2002-08-15 02:10:12 +00:00
return (i);
return (-1);
}
static struct shmid_ds *
shm_find_segment_by_shmid(shmid)
int shmid;
{
int segnum;
struct shmid_ds *shmseg;
segnum = IPCID_TO_IX(shmid);
if (segnum < 0 || segnum >= shmalloced)
2002-08-15 02:10:12 +00:00
return (NULL);
shmseg = &shmsegs[segnum];
if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
!= SHMSEG_ALLOCATED ||
shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid))
2002-08-15 02:10:12 +00:00
return (NULL);
return (shmseg);
}
static struct shmid_ds *
shm_find_segment_by_shmidx(int segnum)
{
struct shmid_ds *shmseg;
if (segnum < 0 || segnum >= shmalloced)
2002-08-15 02:10:12 +00:00
return (NULL);
shmseg = &shmsegs[segnum];
if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
!= SHMSEG_ALLOCATED )
2002-08-15 02:10:12 +00:00
return (NULL);
return (shmseg);
}
static void
shm_deallocate_segment(shmseg)
struct shmid_ds *shmseg;
{
struct shm_handle *shm_handle;
size_t size;
GIANT_REQUIRED;
shm_handle = shmseg->shm_internal;
vm_object_deallocate(shm_handle->shm_object);
2002-07-22 16:12:55 +00:00
free(shm_handle, M_SHM);
shmseg->shm_internal = NULL;
size = round_page(shmseg->shm_segsz);
shm_committed -= btoc(size);
shm_nused--;
shmseg->shm_perm.mode = SHMSEG_FREE;
}
static int
shm_delete_mapping(p, shmmap_s)
struct proc *p;
struct shmmap_state *shmmap_s;
{
struct shmid_ds *shmseg;
int segnum, result;
size_t size;
1995-05-30 08:16:23 +00:00
GIANT_REQUIRED;
segnum = IPCID_TO_IX(shmmap_s->shmid);
shmseg = &shmsegs[segnum];
size = round_page(shmseg->shm_segsz);
result = vm_map_remove(&p->p_vmspace->vm_map, shmmap_s->va,
shmmap_s->va + size);
if (result != KERN_SUCCESS)
2002-08-15 02:10:12 +00:00
return (EINVAL);
shmmap_s->shmid = -1;
shmseg->shm_dtime = time_second;
if ((--shmseg->shm_nattch <= 0) &&
(shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
shm_deallocate_segment(shmseg);
shm_last_free = segnum;
}
2002-08-15 02:10:12 +00:00
return (0);
}
#ifndef _SYS_SYSPROTO_H_
struct shmdt_args {
void *shmaddr;
};
#endif
/*
* MPSAFE
*/
int
shmdt(td, uap)
struct thread *td;
struct shmdt_args *uap;
{
struct proc *p = td->td_proc;
struct shmmap_state *shmmap_s;
int i;
int error = 0;
if (!jail_sysvipc_allowed && jailed(td->td_ucred))
return (ENOSYS);
mtx_lock(&Giant);
shmmap_s = p->p_vmspace->vm_shm;
if (shmmap_s == NULL) {
error = EINVAL;
goto done2;
}
for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
if (shmmap_s->shmid != -1 &&
shmmap_s->va == (vm_offset_t)uap->shmaddr) {
break;
}
}
if (i == shminfo.shmseg) {
error = EINVAL;
goto done2;
}
error = shm_delete_mapping(p, shmmap_s);
done2:
mtx_unlock(&Giant);
return (error);
}
#ifndef _SYS_SYSPROTO_H_
struct shmat_args {
int shmid;
void *shmaddr;
int shmflg;
};
#endif
/*
* MPSAFE
*/
int
shmat(td, uap)
struct thread *td;
struct shmat_args *uap;
{
struct proc *p = td->td_proc;
int i, flags;
struct shmid_ds *shmseg;
struct shmmap_state *shmmap_s = NULL;
struct shm_handle *shm_handle;
vm_offset_t attach_va;
vm_prot_t prot;
vm_size_t size;
int rv;
int error = 0;
if (!jail_sysvipc_allowed && jailed(td->td_ucred))
return (ENOSYS);
mtx_lock(&Giant);
shmmap_s = p->p_vmspace->vm_shm;
if (shmmap_s == NULL) {
size = shminfo.shmseg * sizeof(struct shmmap_state);
shmmap_s = malloc(size, M_SHM, M_WAITOK);
for (i = 0; i < shminfo.shmseg; i++)
shmmap_s[i].shmid = -1;
2002-07-22 16:12:55 +00:00
p->p_vmspace->vm_shm = shmmap_s;
}
shmseg = shm_find_segment_by_shmid(uap->shmid);
if (shmseg == NULL) {
error = EINVAL;
goto done2;
}
error = ipcperm(td, &shmseg->shm_perm,
(uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
if (error)
goto done2;
for (i = 0; i < shminfo.shmseg; i++) {
if (shmmap_s->shmid == -1)
break;
shmmap_s++;
}
if (i >= shminfo.shmseg) {
error = EMFILE;
goto done2;
}
size = round_page(shmseg->shm_segsz);
#ifdef VM_PROT_READ_IS_EXEC
prot = VM_PROT_READ | VM_PROT_EXECUTE;
#else
prot = VM_PROT_READ;
#endif
if ((uap->shmflg & SHM_RDONLY) == 0)
prot |= VM_PROT_WRITE;
flags = MAP_ANON | MAP_SHARED;
if (uap->shmaddr) {
flags |= MAP_FIXED;
if (uap->shmflg & SHM_RND) {
attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
} else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
attach_va = (vm_offset_t)uap->shmaddr;
} else {
error = EINVAL;
goto done2;
}
} else {
/*
* This is just a hint to vm_map_find() about where to
* put it.
*/
attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr
+ maxtsiz + maxdsiz);
}
shm_handle = shmseg->shm_internal;
vm_object_reference(shm_handle->shm_object);
rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object,
0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0);
if (rv != KERN_SUCCESS) {
error = ENOMEM;
goto done2;
}
vm_map_inherit(&p->p_vmspace->vm_map,
attach_va, attach_va + size, VM_INHERIT_SHARE);
shmmap_s->va = attach_va;
shmmap_s->shmid = uap->shmid;
shmseg->shm_lpid = p->p_pid;
shmseg->shm_atime = time_second;
shmseg->shm_nattch++;
td->td_retval[0] = attach_va;
done2:
mtx_unlock(&Giant);
return (error);
}
struct oshmid_ds {
struct ipc_perm shm_perm; /* operation perms */
int shm_segsz; /* size of segment (bytes) */
ushort shm_cpid; /* pid, creator */
ushort shm_lpid; /* pid, last operation */
short shm_nattch; /* no. of current attaches */
time_t shm_atime; /* last attach time */
time_t shm_dtime; /* last detach time */
time_t shm_ctime; /* last change time */
void *shm_handle; /* internal handle for shm segment */
};
struct oshmctl_args {
int shmid;
int cmd;
struct oshmid_ds *ubuf;
};
/*
* MPSAFE
*/
static int
oshmctl(td, uap)
struct thread *td;
struct oshmctl_args *uap;
{
#ifdef COMPAT_43
int error = 0;
struct shmid_ds *shmseg;
struct oshmid_ds outbuf;
if (!jail_sysvipc_allowed && jailed(td->td_ucred))
return (ENOSYS);
mtx_lock(&Giant);
shmseg = shm_find_segment_by_shmid(uap->shmid);
if (shmseg == NULL) {
error = EINVAL;
goto done2;
}
switch (uap->cmd) {
case IPC_STAT:
error = ipcperm(td, &shmseg->shm_perm, IPC_R);
if (error)
goto done2;
outbuf.shm_perm = shmseg->shm_perm;
outbuf.shm_segsz = shmseg->shm_segsz;
outbuf.shm_cpid = shmseg->shm_cpid;
outbuf.shm_lpid = shmseg->shm_lpid;
outbuf.shm_nattch = shmseg->shm_nattch;
outbuf.shm_atime = shmseg->shm_atime;
outbuf.shm_dtime = shmseg->shm_dtime;
outbuf.shm_ctime = shmseg->shm_ctime;
outbuf.shm_handle = shmseg->shm_internal;
2002-07-22 16:12:55 +00:00
error = copyout(&outbuf, uap->ubuf, sizeof(outbuf));
if (error)
goto done2;
break;
default:
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
/* XXX casting to (sy_call_t *) is bogus, as usual. */
error = ((sy_call_t *)shmctl)(td, uap);
break;
}
done2:
mtx_unlock(&Giant);
return (error);
#else
2002-08-15 02:10:12 +00:00
return (EINVAL);
#endif
1995-05-30 08:16:23 +00:00
}
#ifndef _SYS_SYSPROTO_H_
struct shmctl_args {
int shmid;
int cmd;
struct shmid_ds *buf;
};
#endif
/*
* MPSAFE
*/
int
shmctl(td, uap)
struct thread *td;
struct shmctl_args *uap;
{
int error = 0;
struct shmid_ds inbuf;
struct shmid_ds *shmseg;
if (!jail_sysvipc_allowed && jailed(td->td_ucred))
return (ENOSYS);
mtx_lock(&Giant);
switch (uap->cmd) {
case IPC_INFO:
2002-07-22 16:12:55 +00:00
error = copyout(&shminfo, uap->buf, sizeof(shminfo));
if (error)
goto done2;
td->td_retval[0] = shmalloced;
goto done2;
case SHM_INFO: {
struct shm_info shm_info;
shm_info.used_ids = shm_nused;
shm_info.shm_rss = 0; /*XXX where to get from ? */
shm_info.shm_tot = 0; /*XXX where to get from ? */
shm_info.shm_swp = 0; /*XXX where to get from ? */
shm_info.swap_attempts = 0; /*XXX where to get from ? */
shm_info.swap_successes = 0; /*XXX where to get from ? */
2002-07-22 16:12:55 +00:00
error = copyout(&shm_info, uap->buf, sizeof(shm_info));
if (error)
goto done2;
td->td_retval[0] = shmalloced;
goto done2;
}
}
if( (uap->cmd) == SHM_STAT )
shmseg = shm_find_segment_by_shmidx(uap->shmid);
else
shmseg = shm_find_segment_by_shmid(uap->shmid);
if (shmseg == NULL) {
error = EINVAL;
goto done2;
}
switch (uap->cmd) {
case SHM_STAT:
case IPC_STAT:
error = ipcperm(td, &shmseg->shm_perm, IPC_R);
if (error)
goto done2;
2002-07-22 16:12:55 +00:00
error = copyout(shmseg, uap->buf, sizeof(inbuf));
if (error)
goto done2;
else if( (uap->cmd) == SHM_STAT )
td->td_retval[0] = IXSEQ_TO_IPCID( uap->shmid, shmseg->shm_perm );
break;
case IPC_SET:
error = ipcperm(td, &shmseg->shm_perm, IPC_M);
if (error)
goto done2;
2002-07-22 16:12:55 +00:00
error = copyin(uap->buf, &inbuf, sizeof(inbuf));
if (error)
goto done2;
shmseg->shm_perm.uid = inbuf.shm_perm.uid;
shmseg->shm_perm.gid = inbuf.shm_perm.gid;
shmseg->shm_perm.mode =
(shmseg->shm_perm.mode & ~ACCESSPERMS) |
(inbuf.shm_perm.mode & ACCESSPERMS);
shmseg->shm_ctime = time_second;
break;
case IPC_RMID:
error = ipcperm(td, &shmseg->shm_perm, IPC_M);
if (error)
goto done2;
shmseg->shm_perm.key = IPC_PRIVATE;
shmseg->shm_perm.mode |= SHMSEG_REMOVED;
if (shmseg->shm_nattch <= 0) {
shm_deallocate_segment(shmseg);
shm_last_free = IPCID_TO_IX(uap->shmid);
}
break;
#if 0
case SHM_LOCK:
case SHM_UNLOCK:
#endif
default:
error = EINVAL;
break;
}
done2:
mtx_unlock(&Giant);
return (error);
}
#ifndef _SYS_SYSPROTO_H_
struct shmget_args {
key_t key;
size_t size;
int shmflg;
};
#endif
static int
shmget_existing(td, uap, mode, segnum)
struct thread *td;
struct shmget_args *uap;
int mode;
int segnum;
{
struct shmid_ds *shmseg;
int error;
shmseg = &shmsegs[segnum];
if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
/*
* This segment is in the process of being allocated. Wait
* until it's done, and look the key up again (in case the
* allocation failed or it was freed).
*/
shmseg->shm_perm.mode |= SHMSEG_WANTED;
2002-07-22 16:12:55 +00:00
error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0);
if (error)
2002-08-15 02:10:12 +00:00
return (error);
return (EAGAIN);
}
if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
2002-08-15 02:10:12 +00:00
return (EEXIST);
error = ipcperm(td, &shmseg->shm_perm, mode);
if (error)
2002-08-15 02:10:12 +00:00
return (error);
if (uap->size && uap->size > shmseg->shm_segsz)
2002-08-15 02:10:12 +00:00
return (EINVAL);
td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
2002-08-15 02:10:12 +00:00
return (0);
}
static int
shmget_allocate_segment(td, uap, mode)
struct thread *td;
struct shmget_args *uap;
int mode;
{
int i, segnum, shmid, size;
struct ucred *cred = td->td_ucred;
struct shmid_ds *shmseg;
struct shm_handle *shm_handle;
1995-05-30 08:16:23 +00:00
GIANT_REQUIRED;
if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
2002-08-15 02:10:12 +00:00
return (EINVAL);
if (shm_nused >= shminfo.shmmni) /* Any shmids left? */
2002-08-15 02:10:12 +00:00
return (ENOSPC);
size = round_page(uap->size);
if (shm_committed + btoc(size) > shminfo.shmall)
2002-08-15 02:10:12 +00:00
return (ENOMEM);
if (shm_last_free < 0) {
shmrealloc(); /* Maybe expand the shmsegs[] array. */
for (i = 0; i < shmalloced; i++)
if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
break;
if (i == shmalloced)
2002-08-15 02:10:12 +00:00
return (ENOSPC);
segnum = i;
} else {
segnum = shm_last_free;
shm_last_free = -1;
}
shmseg = &shmsegs[segnum];
/*
* In case we sleep in malloc(), mark the segment present but deleted
* so that noone else tries to create the same key.
*/
shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
shmseg->shm_perm.key = uap->key;
shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
shm_handle = (struct shm_handle *)
malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
/*
* We make sure that we have allocated a pager before we need
* to.
*/
if (shm_use_phys) {
shm_handle->shm_object =
vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0);
} else {
shm_handle->shm_object =
vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0);
}
vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
shmseg->shm_internal = shm_handle;
shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
(mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
shmseg->shm_segsz = uap->size;
shmseg->shm_cpid = td->td_proc->p_pid;
shmseg->shm_lpid = shmseg->shm_nattch = 0;
shmseg->shm_atime = shmseg->shm_dtime = 0;
shmseg->shm_ctime = time_second;
shm_committed += btoc(size);
shm_nused++;
if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
/*
* Somebody else wanted this key while we were asleep. Wake
* them up now.
*/
shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
2002-07-22 16:12:55 +00:00
wakeup(shmseg);
}
td->td_retval[0] = shmid;
2002-08-15 02:10:12 +00:00
return (0);
}
/*
* MPSAFE
*/
int
shmget(td, uap)
struct thread *td;
struct shmget_args *uap;
{
int segnum, mode;
int error;
if (!jail_sysvipc_allowed && jailed(td->td_ucred))
return (ENOSYS);
mtx_lock(&Giant);
mode = uap->shmflg & ACCESSPERMS;
if (uap->key != IPC_PRIVATE) {
again:
segnum = shm_find_segment_by_key(uap->key);
if (segnum >= 0) {
error = shmget_existing(td, uap, mode, segnum);
if (error == EAGAIN)
goto again;
goto done2;
}
if ((uap->shmflg & IPC_CREAT) == 0) {
error = ENOENT;
goto done2;
}
}
error = shmget_allocate_segment(td, uap, mode);
done2:
mtx_unlock(&Giant);
return (error);
}
/*
* MPSAFE
*/
int
shmsys(td, uap)
struct thread *td;
Start including <sys/sysproto.h> to get the correct args structs and prototypes for all syscalls. The args structs are still declared in comments as in VOP implementation functions. I don't like the duplication for this, but several more layers of changes are required to get it right. First we need to catch up with 4.4lite2, which uses macros to handle struct padding. Then we need to catch up with NetBSD, which passes the args correctly (as void *). Then we need to handle varargs functions and struct padding better. I think all the details can be hidden in machine-generated functions so that the args structs and verbose macros to reference them don't have to appear in the core sources. Add prototypes. Add bogus casts to hide the evil type puns exposed by the previous steps. &uap[1] was used to get at the args after the first. This worked because only the first arg in *uap was declared. This broke when the machine- genenerated args struct declared all the args (actually it declares extra args in some cases and depends on the user stack having some accessible junk after the last arg, not to mention the user args being on the stack. It isn't possible to declare a correct args struct for a varargs syscall). The msgsys(), semsys() and shmsys() syscall interfaces are BAD because they multiplex several syscalls that have different types of args. There was no reason to duplicate this sysv braindamage but now we're stuck with it. NetBSD has reimplemented the syscalls properly as separate syscalls #220-231. Declare static functions as static in both their prototype and their implementation (the latter is optional, and this misfeature was used). Remove gratuitous #includes. Continue cleaning up new init stuff.
1995-10-21 19:50:00 +00:00
/* XXX actually varargs. */
struct shmsys_args /* {
u_int which;
int a2;
int a3;
int a4;
} */ *uap;
{
int error;
if (!jail_sysvipc_allowed && jailed(td->td_ucred))
return (ENOSYS);
if (uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0]))
return (EINVAL);
mtx_lock(&Giant);
error = (*shmcalls[uap->which])(td, &uap->a2);
mtx_unlock(&Giant);
return (error);
}
static void
shmfork_myhook(p1, p2)
struct proc *p1, *p2;
{
struct shmmap_state *shmmap_s;
size_t size;
int i;
size = shminfo.shmseg * sizeof(struct shmmap_state);
shmmap_s = malloc(size, M_SHM, M_WAITOK);
2002-07-22 16:12:55 +00:00
bcopy(p1->p_vmspace->vm_shm, shmmap_s, size);
p2->p_vmspace->vm_shm = shmmap_s;
for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
if (shmmap_s->shmid != -1)
shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
}
static void
shmexit_myhook(p)
struct proc *p;
{
struct shmmap_state *shmmap_s;
int i;
GIANT_REQUIRED;
shmmap_s = p->p_vmspace->vm_shm;
for (i = 0; i < shminfo.shmseg; i++, shmmap_s++)
if (shmmap_s->shmid != -1)
shm_delete_mapping(p, shmmap_s);
2002-07-22 16:12:55 +00:00
free(p->p_vmspace->vm_shm, M_SHM);
p->p_vmspace->vm_shm = NULL;
}
static void
shmrealloc(void)
{
int i;
struct shmid_ds *newsegs;
if (shmalloced >= shminfo.shmmni)
return;
newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
if (newsegs == NULL)
return;
for (i = 0; i < shmalloced; i++)
bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
for (; i < shminfo.shmmni; i++) {
shmsegs[i].shm_perm.mode = SHMSEG_FREE;
shmsegs[i].shm_perm.seq = 0;
}
free(shmsegs, M_SHM);
shmsegs = newsegs;
shmalloced = shminfo.shmmni;
}
static void
shminit()
{
int i;
TUNABLE_INT_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall);
shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
TUNABLE_INT_FETCH("kern.ipc.shmmin", &shminfo.shmmin);
TUNABLE_INT_FETCH("kern.ipc.shmmni", &shminfo.shmmni);
TUNABLE_INT_FETCH("kern.ipc.shmseg", &shminfo.shmseg);
TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys);
shmalloced = shminfo.shmmni;
shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
if (shmsegs == NULL)
panic("cannot allocate initial memory for sysvshm");
for (i = 0; i < shmalloced; i++) {
shmsegs[i].shm_perm.mode = SHMSEG_FREE;
shmsegs[i].shm_perm.seq = 0;
}
shm_last_free = 0;
shm_nused = 0;
shm_committed = 0;
shmexit_hook = &shmexit_myhook;
shmfork_hook = &shmfork_myhook;
}
static int
shmunload()
{
if (shm_nused > 0)
return (EBUSY);
free(shmsegs, M_SHM);
shmexit_hook = NULL;
shmfork_hook = NULL;
return (0);
}
static int
sysctl_shmsegs(SYSCTL_HANDLER_ARGS)
{
return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0])));
}
static int
sysvshm_modload(struct module *module, int cmd, void *arg)
{
int error = 0;
switch (cmd) {
case MOD_LOAD:
shminit();
break;
case MOD_UNLOAD:
error = shmunload();
break;
case MOD_SHUTDOWN:
break;
default:
error = EINVAL;
break;
}
return (error);
}
static moduledata_t sysvshm_mod = {
"sysvshm",
&sysvshm_modload,
NULL
};
SYSCALL_MODULE_HELPER(shmsys);
SYSCALL_MODULE_HELPER(shmat);
SYSCALL_MODULE_HELPER(shmctl);
SYSCALL_MODULE_HELPER(shmdt);
SYSCALL_MODULE_HELPER(shmget);
DECLARE_MODULE(sysvshm, sysvshm_mod,
SI_SUB_SYSV_SHM, SI_ORDER_FIRST);
MODULE_VERSION(sysvshm, 1);