? arch/i386/compile/GENERIC ? arch/i386/compile/KAZOO ? arch/i386/conf/KAZOO Index: arch/i386/conf/GENERIC =================================================================== RCS file: /cvsroot/src/sys/arch/i386/conf/GENERIC,v retrieving revision 1.594 diff -u -r1.594 GENERIC --- arch/i386/conf/GENERIC 25 Feb 2004 18:56:26 -0000 1.594 +++ arch/i386/conf/GENERIC 18 Jun 2004 16:57:03 -0000 @@ -690,7 +690,7 @@ # PCI network interfaces an* at pci? dev ? function ? # Aironet PC4500/PC4800 (802.11) -ath* at pci? dev ? function ? # Atheros 5210/5211/5212 802.11 +#ath* at pci? dev ? function ? # Atheros 5210/5211/5212 802.11 atw* at pci? dev ? function ? # ADMtek ADM8211 (802.11) bce* at pci? dev ? function ? # Broadcom 4401 10/100 Ethernet bge* at pci? dev ? function ? # Broadcom 570x gigabit Ethernet Index: conf/files =================================================================== RCS file: /cvsroot/src/sys/conf/files,v retrieving revision 1.657 diff -u -r1.657 files --- conf/files 17 Feb 2004 05:03:15 -0000 1.657 +++ conf/files 18 Jun 2004 16:57:12 -0000 @@ -1125,6 +1125,7 @@ file kern/init_sysent.c file kern/kern_acct.c file kern/kern_clock.c +file kern/kern_cpt.c file kern/kern_descrip.c file kern/kern_event.c file kern/kern_exec.c Index: kern/init_main.c =================================================================== RCS file: /cvsroot/src/sys/kern/init_main.c,v retrieving revision 1.233 diff -u -r1.233 init_main.c --- kern/init_main.c 9 Mar 2004 02:35:45 -0000 1.233 +++ kern/init_main.c 18 Jun 2004 16:57:15 -0000 @@ -147,6 +147,7 @@ #ifdef LKM #include #endif +#include #include #include @@ -431,6 +432,8 @@ ipsec_attach(); #endif + cpt_init(); + /* * Initialize protocols. Block reception of incoming packets * until everything is ready. Index: kern/init_sysent.c =================================================================== RCS file: /cvsroot/src/sys/kern/init_sysent.c,v retrieving revision 1.153 diff -u -r1.153 init_sysent.c --- kern/init_sysent.c 29 Jan 2004 02:00:03 -0000 1.153 +++ kern/init_sysent.c 18 Jun 2004 16:57:16 -0000 @@ -1,14 +1,14 @@ -/* $NetBSD: init_sysent.c,v 1.153 2004/01/29 02:00:03 tsarna Exp $ */ +/* $NetBSD$ */ /* * System call switch table. * * DO NOT EDIT-- this file is automatically generated. - * created from NetBSD: syscalls.master,v 1.135 2004/01/02 18:52:17 cl Exp + * created from NetBSD: syscalls.master,v 1.136 2004/01/29 02:00:03 tsarna Exp */ #include -__KERNEL_RCSID(0, "$NetBSD: init_sysent.c,v 1.153 2004/01/29 02:00:03 tsarna Exp $"); +__KERNEL_RCSID(0, "$NetBSD$"); #include "opt_ktrace.h" #include "opt_nfsserver.h" @@ -406,9 +406,9 @@ { 3, s(struct compat_43_sys_getsockname_args), 0, compat_43(sys_getsockname) }, /* 150 = compat_43 ogetsockname */ { 0, 0, 0, - sys_nosys }, /* 151 = unimplemented */ - { 0, 0, 0, - sys_nosys }, /* 152 = unimplemented */ + sys_cptfork }, /* 151 = cptfork */ + { 3, s(struct sys_cptctl_args), 0, + sys_cptctl }, /* 152 = cptctl */ { 0, 0, 0, sys_nosys }, /* 153 = unimplemented */ { 0, 0, 0, Index: kern/kern_cpt.c =================================================================== RCS file: kern/kern_cpt.c diff -N kern/kern_cpt.c --- /dev/null 1 Jan 1970 00:00:00 -0000 +++ kern/kern_cpt.c 18 Jun 2004 16:57:17 -0000 @@ -0,0 +1,509 @@ +/* $Id$ */ + +/*- + * Copyright (c) 2004 Cubical Solutions Ltd. + * 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. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 AUTHOR OR CONTRIBUTORS 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 +__KERNEL_RCSID(0, "$Id$"); + +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include + +#include +#include + +#ifdef DEBUG +#define CPT_DEBUG 0 +#endif + +#ifdef CPT_DEBUG +#define DPRINTF(x) if (cptdebug) printf x; +static int cptdebug = CPT_DEBUG; +#else +#define DPRINTF(x) +#endif + +MALLOC_DEFINE(M_CPT, "cpt", "kernel checkpointing support"); + +static int cpt_pages(struct vm_map *, struct __cpt_range *); +static int cpt_query(struct __cpt_range *, struct cpt_range *, size_t); + +static int cpt_install(struct proc *, struct cpt_range *, size_t); +static int cpt_purge(struct proc *, struct cpt_range *, size_t); +static void cpt_purgeall(struct proc *); + +static struct __cpt_range * cpt_range_get(void); +static void cpt_range_put(struct __cpt_range *); + +static struct pool cptr_pool; + +void +cpt_init() +{ + + pool_init(&cptr_pool, sizeof(struct __cpt_range),0,0,0, "cptpl", + &pool_allocator_nointr); +} + +static struct __cpt_range * +cpt_range_get() +{ + struct __cpt_range *cptr; + + cptr = pool_get(&cptr_pool, PR_WAITOK); + memset(cptr, 0, sizeof(struct __cpt_range)); + + return cptr; +} +static void +cpt_range_put(struct __cpt_range *cptr) +{ + + if (cptr->page) + free(cptr->page, M_CPT); + pool_put(&cptr_pool, cptr); +} + +#define CPTDIRTY(a) \ +do { \ + if (!dirtystart) \ + dirtystart = (vaddr_t)(a); \ + runlen++; \ +} while (/*CONSTCOND*/0) +#define CPTCLEAN() \ +do { \ + if (dirtystart) { \ + DPRINTF(("CPTCLEAN: adding range (0x%lx,%d)\n", \ + (unsigned long)dirtystart, (int)runlen*PAGE_SIZE)); \ + struct __cpt_range *cp_tmp; \ + cp_tmp = cpt_range_get(); \ + cp_tmp->cptr.addr = (caddr_t)dirtystart; \ + cp_tmp->cptr.len = runlen * PAGE_SIZE; \ + LIST_INSERT_HEAD(&p2->p_cptr_dirty, \ + cp_tmp, entries); \ + numdirty++; \ + } \ + runlen = 0; \ + dirtystart = 0; \ +} while (/*CONSTCOND*/0) +/* + * Scan dirty ranges from the pre-defined range. This is called from + * fork1(), and requires both the parent and child struct proc. Uses + * parent to figure out ranges and child to store dirty ranges. + */ +void +cpt_dirtys(struct proc *p1, struct proc *p2) +{ + struct __cpt_range *cptr; + vaddr_t dirtystart; + size_t numdirty, runlen; + int i, error; + + dirtystart = 0; + numdirty = 0; + runlen = 0; + + LIST_INIT(&p2->p_cptr_dirty); + + DPRINTF(("cpt_dirtys: scanning for dirty pages\n")); + LIST_FOREACH(cptr, &p1->p_cptr, entries) { + error = cpt_pages(&p1->p_vmspace->vm_map, cptr); + + /* + * XXX: just ignoring the error is *very* wrong, we + * should signal the luser program that it should + * expect a screwup (sautéed). + */ + if (error) + continue; + for (i = 0; i < cptr->cptr.len / PAGE_SIZE; i++) { + if (pmap_is_modified(cptr->page[i])) { +#ifdef CPT_DEBUG + if (cptdebug > 1) + DPRINTF(("vm_page at %p dirty\n", + cptr->page[i])); +#endif + CPTDIRTY((caddr_t)cptr->cptr.addr+i*PAGE_SIZE); + pmap_clear_modify(cptr->page[i]); + } else { + CPTCLEAN(); + } + } + CPTCLEAN(); + } + p2->p_cptr_numdirty = numdirty; +} +#undef CPTDIRTY +#undef CPTCLEAN + +int +sys_cptfork(struct lwp *l, void *v, register_t *retval) +{ + + DPRINTF(("cptfork for process at %p\n", l->l_proc)); + return fork1(l, FORK_CPT, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL); +} + +ssize_t +sys_cptctl(struct lwp *l, void *v, register_t *retval) +{ + struct sys_cptctl_args /* { + syscallarg(struct cpt_range *) ranges; + syscallarg(size_t) nranges; + syscallarg(int) op; + } */ *uap = v; + int error; + + switch (SCARG(uap, op)) { + case CPT_INSTALL: + DPRINTF(("sys_cptctl: installing %lu range(s)\n", + (unsigned long)SCARG(uap, nranges))); + return cpt_install(l->l_proc, + SCARG(uap, ranges), SCARG(uap, nranges)); + + case CPT_PURGE: + DPRINTF(("sys_cptctl: purging %lu range(s)\n", + (unsigned long)SCARG(uap, nranges))); + return cpt_purge(l->l_proc, + SCARG(uap, ranges), SCARG(uap, nranges)); + + case CPT_PURGE_ALL: + DPRINTF(("sys_cptctl: purging all ranges\n")); + cpt_purgeall(l->l_proc); + return 0; + + case CPT_QUERY: + /* + * If there's not enough space available, simply + * return the amount of space required. + */ + if (SCARG(uap, nranges) < l->l_proc->p_cptr_numdirty) { + *retval = l->l_proc->p_cptr_numdirty; + return 0; + } + + /* + * Return number of ranges copied. + */ + error = cpt_query(LIST_FIRST(&l->l_proc->p_cptr_dirty), + SCARG(uap, ranges), l->l_proc->p_cptr_numdirty); + if (!error) + *retval = l->l_proc->p_cptr_numdirty; + return error; + + case CPT_CLEAN: + case CPT_CLEAN_ALL: + case CPT_SOIL: + case CPT_SOIL_ALL: + printf("cptctl: not implemented\n"); + return 0; /* XXX */ + + default: + return EINVAL; + } +} + +static int +cpt_install(struct proc *p, struct cpt_range *cptr_user, size_t nranges) +{ + struct __cpt_range *cptr; + int error; + + /* + * No point in installing zero ranges. + */ + if (nranges == 0) + return EINVAL; + + /* + * Loop through ranges: + * + fill out __cpt_range structures, i.e. copy range list from + * userspace and cache struct vmpage's. + * + call uvm code to modify map entry flags. + */ + for (; nranges > 0; nranges--, cptr_user++) { + cptr = cpt_range_get(); + if (cptr == NULL) + return ENOMEM; + error = copyin(cptr_user,&cptr->cptr,sizeof(struct cpt_range)); + if (error) + goto fail; + + /* adjust to page ranges */ + /* XXX: casts? */ + cptr->cptr.addr = (void *)trunc_page((vaddr_t)cptr->cptr.addr); + cptr->cptr.len = round_page(cptr->cptr.len); + + DPRINTF(("cpt_install: %p/%zu\n", + cptr->cptr.addr, cptr->cptr.len)); + error = uvm_map_cpt(&p->p_vmspace->vm_map, + (vaddr_t)cptr->cptr.addr, + (vaddr_t)cptr->cptr.addr + cptr->cptr.len, + TRUE); + if (error) + goto fail; + + /* + * If we're installing ranges for the first time (or after + * removing all in between), initialize. + */ + if ((p->p_vmspace->vm_map.flags & UVM_MAP_CPT) == 0) { + p->p_vmspace->vm_map.flags |= UVM_MAP_CPT; + LIST_INIT(&p->p_cptr); + } + + LIST_INSERT_HEAD(&p->p_cptr, cptr, entries); + } + return 0; + +fail: + cpt_range_put(cptr); + return error; +} + +#define CPTR_COMPARE(a,b) ((a).addr == (b).addr && (a).len == (b).len) +static int +cpt_purge(struct proc *p, struct cpt_range *cptr_user, size_t nranges) +{ + struct __cpt_range *cptr; + struct cpt_range cptr_tmp; + int error, found; + + if (nranges == 0) + return EINVAL; + + /* + * Work through ranges one-by-one, locate "offending" entry + * from list and remove it. Call uvm code to modify map entry + * flags back to normal. + */ + for (; nranges > 0; nranges--, cptr_user++) { + error = copyin(cptr_user, &cptr_tmp, sizeof(struct cpt_range)); + if (error) + return error; + found = 0; + LIST_FOREACH(cptr, &p->p_cptr, entries) { + if (CPTR_COMPARE(cptr->cptr, cptr_tmp)) { + LIST_REMOVE(cptr, entries); + error = uvm_map_cpt(&p->p_vmspace->vm_map, + (vaddr_t)cptr->cptr.addr, + (vaddr_t)cptr->cptr.addr + cptr->cptr.len, + FALSE); + cpt_range_put(cptr); + if (error) + return error; + found = 1; + break; + } + } + if (!found) + return EINVAL; + } + + /* + * Remove flag that map has checkpointing entries if we remove + * the last entry. + */ + if (LIST_EMPTY(&p->p_cptr)) + p->p_vmspace->vm_map.flags &= ~UVM_MAP_CPT; + + return 0; +} + +static void +cpt_purgeall(struct proc *p) +{ + struct __cpt_range *cptr; + + while (!LIST_EMPTY(&p->p_cptr)) { + cptr = LIST_FIRST(&p->p_cptr); + LIST_REMOVE(cptr, entries); + cpt_range_put(cptr); + } + uvm_map_cpt_purgeall(&p->p_vmspace->vm_map); + p->p_vmspace->vm_map.flags &= ~UVM_MAP_CPT; +} + +static int +cpt_query(struct __cpt_range *cptr_src, struct cpt_range *cptr_to, size_t num) +{ + int error; + + for (; num > 0; num--, cptr_to++) { + KASSERT(cptr_src != NULL); + +#ifdef CPT_DEBUG + if (cptdebug > 1) + DPRINTF(("cpt_query: copying from %p to %p\n", + &cptr_src->cptr, cptr_to)); +#endif + error = copyout(&cptr_src->cptr, cptr_to, + sizeof(struct cpt_range)); + if (error) + return error; + cptr_src = LIST_NEXT(cptr_src, entries); + } + + return 0; +} + +/* + * Locate vm_page's. This means digging the mappings of each virtual + * address belonging to the range from the bowels of UVM. + */ +static int +cpt_pages(struct vm_map *map, struct __cpt_range *cptr) +{ + struct vm_map_entry *entry; + struct vm_amap *amap; + struct vm_anon *anon; + caddr_t startaddr; + vaddr_t currva; + size_t npages; + int error, i; + + i = 0; + /* + * If we have already reserved memory for the pages, don't + * reserve again. The amount of pages per in-kernel __cpt_range + * doesn't change. + */ + if (cptr->page == NULL) { + npages = cptr->cptr.len / PAGE_SIZE; + cptr->page = malloc(npages * sizeof(struct vm_page), + M_CPT, M_WAITOK); + } + + DPRINTF(("cpt_pages: start: %p, len %x\n", + cptr->cptr.addr, cptr->cptr.len)); + + /* lock vm_map and dig out vm_page's */ + vm_map_lock_read(map); + for (startaddr = (caddr_t)cptr->cptr.addr; + startaddr < (caddr_t)cptr->cptr.addr + cptr->cptr.len; + startaddr += PAGE_SIZE) { + + if (uvm_map_lookup_entry(map, (vaddr_t)startaddr, &entry) + == FALSE) { + DPRINTF(("cpt_pages: couldn't find entry for " + "address %p\n", startaddr)); + error = EFAULT; + goto fail; + } + +#if 0 + /* + * Check that entry is wired. We do it here, because we + * don't have the vm_map_entry earlier. + */ + if (!VM_MAPENT_ISWIRED(entry)) { + DPRINTF(("cpt_pages: entry at %p not wired\n", + startaddr)); + error = EFAULT; + goto fail; + } +#endif + + /* Get amap */ + amap = entry->aref.ar_amap; + if (!amap) { + DPRINTF(("cpt_pages: no amap for %p\n", startaddr)); + error = EFAULT; /* XXX? */ + goto fail; + } + + amap_lock(amap); + for (currva = (vaddr_t)startaddr; + currva < entry->end + && startaddr < (caddr_t)cptr->cptr.addr + cptr->cptr.len; + currva += PAGE_SIZE, startaddr += PAGE_SIZE) { + /* + * Lookup anon, easier to do it lookup call/anon + * instead of using amap_lookups. This is not + * along the critical path anyways. + */ + anon = amap_lookup(&entry->aref, currva - entry->start); + if (!anon) { + DPRINTF(("cpt_pages: can't find amap " + "for 0x%lx\n", currva)); + error = EFAULT; + goto fail_amap; + } + cptr->page[i++] = anon->u.an_page; +#ifdef CPT_DEBUG + if (cptdebug > 2) + DPRINTF(("cpt_pages: page %lx: vm_page at %p\n", + currva+entry->start, anon->u.an_page)); +#endif + } + amap_unlock(amap); + } + vm_map_unlock(map); + return 0; + +fail_amap: + amap_unlock(amap); +fail: + vm_map_unlock(map); + free(cptr->page, M_CPT); + return error; +} + +/* + * Free appropriate structures. + */ +void +cpt_proc_exit(struct proc *p) +{ + struct __cpt_range *cptr; + + /* + * If we are the parent, we'll have a range-list. + */ + cpt_purgeall(p); + + /* + * For the cptfork() child, we need to free the dirty-list. + */ + while (!LIST_EMPTY(&p->p_cptr_dirty)) { + KASSERT(p->p_cptr_numdirty > 0); + + cptr = LIST_FIRST(&p->p_cptr_dirty); + LIST_REMOVE(cptr, entries); + cpt_range_put(cptr); + + p->p_cptr_numdirty--; + } +} Index: kern/kern_exit.c =================================================================== RCS file: /cvsroot/src/sys/kern/kern_exit.c,v retrieving revision 1.138 diff -u -r1.138 kern_exit.c --- kern/kern_exit.c 5 Mar 2004 07:27:22 -0000 1.138 +++ kern/kern_exit.c 18 Jun 2004 16:57:17 -0000 @@ -113,6 +113,7 @@ #include #include #include +#include #include @@ -259,6 +260,11 @@ #endif /* + * Free all relevant checkpointing information. + */ + cpt_proc_exit(p); + + /* * Close open files and release open-file table. * This may block! */ Index: kern/kern_fork.c =================================================================== RCS file: /cvsroot/src/sys/kern/kern_fork.c,v retrieving revision 1.114 diff -u -r1.114 kern_fork.c --- kern/kern_fork.c 12 Feb 2004 23:47:21 -0000 1.114 +++ kern/kern_fork.c 18 Jun 2004 16:57:18 -0000 @@ -98,6 +98,7 @@ #include #include #include +#include #include #include @@ -203,7 +204,7 @@ struct proc *p1, *p2, *parent; uid_t uid; struct lwp *l2; - int count, s; + int count, s, uvmflags; vaddr_t uaddr; boolean_t inmem; @@ -317,6 +318,10 @@ else p2->p_cwdi = cwdinit(p1); + /* If we're doing cptfork(), record dirty pages into child. */ + if (flags & FORK_CPT) + cpt_dirtys(p1, p2); + /* * If p_limit is still copy-on-write, bump refcnt, * otherwise get a copy that won't be modified. @@ -389,7 +394,12 @@ */ PHOLD(l1); - uvm_proc_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE); + uvmflags = 0; + if (flags & FORK_SHAREVM) + uvmflags |= UVM_FORK_SHARE; + if (flags & FORK_CPT) + uvmflags |= UVM_FORK_CPT; + uvm_proc_fork(p1, p2, uvmflags); /* * Finish creating the child process. Index: kern/syscalls.c =================================================================== RCS file: /cvsroot/src/sys/kern/syscalls.c,v retrieving revision 1.148 diff -u -r1.148 syscalls.c --- kern/syscalls.c 29 Jan 2004 02:00:03 -0000 1.148 +++ kern/syscalls.c 18 Jun 2004 16:57:18 -0000 @@ -1,14 +1,14 @@ -/* $NetBSD: syscalls.c,v 1.148 2004/01/29 02:00:03 tsarna Exp $ */ +/* $NetBSD$ */ /* * System call names. * * DO NOT EDIT-- this file is automatically generated. - * created from NetBSD: syscalls.master,v 1.135 2004/01/02 18:52:17 cl Exp + * created from NetBSD: syscalls.master,v 1.136 2004/01/29 02:00:03 tsarna Exp */ #include -__KERNEL_RCSID(0, "$NetBSD: syscalls.c,v 1.148 2004/01/29 02:00:03 tsarna Exp $"); +__KERNEL_RCSID(0, "$NetBSD$"); #if defined(_KERNEL_OPT) #include "opt_ktrace.h" @@ -196,8 +196,8 @@ "quotactl", /* 148 = quotactl */ "compat_43_oquota", /* 149 = compat_43 oquota */ "compat_43_ogetsockname", /* 150 = compat_43 ogetsockname */ - "#151 (unimplemented)", /* 151 = unimplemented */ - "#152 (unimplemented)", /* 152 = unimplemented */ + "cptfork", /* 151 = cptfork */ + "cptctl", /* 152 = cptctl */ "#153 (unimplemented)", /* 153 = unimplemented */ "#154 (unimplemented)", /* 154 = unimplemented */ #if defined(NFS) || defined(NFSSERVER) || !defined(_KERNEL) Index: kern/syscalls.master =================================================================== RCS file: /cvsroot/src/sys/kern/syscalls.master,v retrieving revision 1.136 diff -u -r1.136 syscalls.master --- kern/syscalls.master 29 Jan 2004 02:00:03 -0000 1.136 +++ kern/syscalls.master 18 Jun 2004 16:57:19 -0000 @@ -307,8 +307,9 @@ ; system calls. (This includes various calls added for compatibity ; with other Unix variants.) ; Some of these calls are now supported by BSD... -151 UNIMPL -152 UNIMPL +151 STD { int sys_cptfork(void); } +152 STD { int sys_cptctl(struct cpt_range *ranges, \ + size_t nranges, int op); } 153 UNIMPL 154 UNIMPL #if defined(NFS) || defined(NFSSERVER) || !defined(_KERNEL) Index: sys/cpt.h =================================================================== RCS file: sys/cpt.h diff -N sys/cpt.h --- /dev/null 1 Jan 1970 00:00:00 -0000 +++ sys/cpt.h 18 Jun 2004 16:57:20 -0000 @@ -0,0 +1,80 @@ +/* $Id$ */ + +/*- + * Copyright (c) 2004 Cubical Solutions Ltd. + * 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. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``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 AUTHOR OR CONTRIBUTORS 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 +#include +#include +#include +#include +#include + +#include + +/* + * operation types + */ +#define CPT_INSTALL 1 +#define CPT_PURGE 2 +#define CPT_PURGE_ALL 3 +#define CPT_QUERY 4 +#define CPT_CLEAN 5 +#define CPT_CLEAN_ALL 6 +#define CPT_SOIL 7 +#define CPT_SOIL_ALL 8 + +/* + * Checkpoint memory range descriptors. + */ +struct cpt_range { + void *addr; + size_t len; + +}; + +#ifdef _KERNEL +/* + * We want to use the same struct in the kernel in a list-like fashion + * as opposed to the array-oriented approach in userland. + */ +struct __cpt_range { + struct cpt_range cptr; /* range for this struct */ + struct vm_page **page; /* cached vm_page array for range */ + + LIST_ENTRY(__cpt_range) entries; +}; +#endif /* _KERNEL */ + +__BEGIN_DECLS +void cpt_init(void); +void cpt_proc_exit(struct proc *); +void cpt_dirtys(struct proc *, struct proc *); + +/* syscalls */ +pid_t cptfork(size_t *); +ssize_t cptctl(struct cpt_range *, size_t, int); +__END_DECLS Index: sys/proc.h =================================================================== RCS file: /cvsroot/src/sys/sys/proc.h,v retrieving revision 1.190 diff -u -r1.190 proc.h --- sys/proc.h 13 Feb 2004 11:36:23 -0000 1.190 +++ sys/proc.h 18 Jun 2004 16:57:21 -0000 @@ -231,6 +231,10 @@ const struct execsw *p_execsw; /* Exec package information */ struct klist p_klist; /* Knotes attached to this process */ + LIST_HEAD(,__cpt_range) p_cptr; /* cpt areas currently installed */ + LIST_HEAD(,__cpt_range) p_cptr_dirty; /* changed memory ranges list */ + size_t p_cptr_numdirty; /* number of elements on dirty-list */ + /* * End area that is zeroed on creation */ @@ -371,6 +375,7 @@ #define FORK_SHARESIGS 0x10 /* Share signal actions */ #define FORK_NOWAIT 0x20 /* Make init the parent of the child */ #define FORK_CLEANFILES 0x40 /* Start with a clean descriptor set */ +#define FORK_CPT 0x80 /* cpt semantics for uvmspace_fork() */ /* * Allow machine-dependent code to override curproc in for Index: sys/syscall.h =================================================================== RCS file: /cvsroot/src/sys/sys/syscall.h,v retrieving revision 1.146 diff -u -r1.146 syscall.h --- sys/syscall.h 29 Jan 2004 02:00:03 -0000 1.146 +++ sys/syscall.h 18 Jun 2004 16:57:22 -0000 @@ -1,10 +1,10 @@ -/* $NetBSD: syscall.h,v 1.146 2004/01/29 02:00:03 tsarna Exp $ */ +/* $NetBSD$ */ /* * System call numbers. * * DO NOT EDIT-- this file is automatically generated. - * created from NetBSD: syscalls.master,v 1.135 2004/01/02 18:52:17 cl Exp + * created from NetBSD: syscalls.master,v 1.136 2004/01/29 02:00:03 tsarna Exp */ /* syscall: "syscall" ret: "int" args: "int" "..." */ @@ -458,6 +458,12 @@ /* syscall: "compat_43_ogetsockname" ret: "int" args: "int" "caddr_t" "int *" */ #define SYS_compat_43_ogetsockname 150 +/* syscall: "cptfork" ret: "int" args: */ +#define SYS_cptfork 151 + +/* syscall: "cptctl" ret: "int" args: "struct cpt_range *" "size_t" "int" */ +#define SYS_cptctl 152 + #if defined(NFS) || defined(NFSSERVER) || !defined(_KERNEL) /* syscall: "nfssvc" ret: "int" args: "int" "void *" */ #define SYS_nfssvc 155 Index: sys/syscallargs.h =================================================================== RCS file: /cvsroot/src/sys/sys/syscallargs.h,v retrieving revision 1.128 diff -u -r1.128 syscallargs.h --- sys/syscallargs.h 29 Jan 2004 02:00:03 -0000 1.128 +++ sys/syscallargs.h 18 Jun 2004 16:57:23 -0000 @@ -1,10 +1,10 @@ -/* $NetBSD: syscallargs.h,v 1.128 2004/01/29 02:00:03 tsarna Exp $ */ +/* $NetBSD$ */ /* * System call argument lists. * * DO NOT EDIT-- this file is automatically generated. - * created from NetBSD: syscalls.master,v 1.135 2004/01/02 18:52:17 cl Exp + * created from NetBSD: syscalls.master,v 1.136 2004/01/29 02:00:03 tsarna Exp */ #ifndef _SYS__SYSCALLARGS_H_ @@ -697,6 +697,12 @@ syscallarg(caddr_t) asa; syscallarg(int *) alen; }; + +struct sys_cptctl_args { + syscallarg(struct cpt_range *) ranges; + syscallarg(size_t) nranges; + syscallarg(int) op; +}; #if defined(NFS) || defined(NFSSERVER) || !defined(_KERNEL) struct sys_nfssvc_args { @@ -1766,6 +1772,10 @@ int compat_43_sys_getsockname(struct lwp *, void *, register_t *); +int sys_cptfork(struct lwp *, void *, register_t *); + +int sys_cptctl(struct lwp *, void *, register_t *); + #if defined(NFS) || defined(NFSSERVER) || !defined(_KERNEL) int sys_nfssvc(struct lwp *, void *, register_t *); Index: uvm/uvm_extern.h =================================================================== RCS file: /cvsroot/src/sys/uvm/uvm_extern.h,v retrieving revision 1.89 diff -u -r1.89 uvm_extern.h --- uvm/uvm_extern.h 13 Feb 2004 13:47:16 -0000 1.89 +++ uvm/uvm_extern.h 18 Jun 2004 16:57:24 -0000 @@ -201,6 +201,12 @@ #define UVM_LK_ENTER 0x00000001 /* map locked on entry */ #define UVM_LK_EXIT 0x00000002 /* leave map locked on exit */ + /* + * flags for forking + */ +#define UVM_FORK_SHARE 0x01 /* share vm space */ +#define UVM_FORK_CPT 0x02 /* cpt semantics */ + /* * structures */ @@ -561,7 +567,7 @@ #if defined(KGDB) void uvm_chgkprot __P((caddr_t, size_t, int)); #endif -void uvm_proc_fork __P((struct proc *, struct proc *, boolean_t)); +void uvm_proc_fork __P((struct proc *, struct proc *, int)); void uvm_lwp_fork __P((struct lwp *, struct lwp *, void *, size_t, void (*)(void *), void *)); int uvm_coredump_walkmap __P((struct proc *, @@ -664,7 +670,7 @@ void uvmspace_init __P((struct vmspace *, struct pmap *, vaddr_t, vaddr_t)); void uvmspace_exec __P((struct lwp *, vaddr_t, vaddr_t)); -struct vmspace *uvmspace_fork __P((struct vmspace *)); +struct vmspace *uvmspace_fork __P((struct vmspace *, int)); void uvmspace_free __P((struct vmspace *)); void uvmspace_share __P((struct proc *, struct proc *)); void uvmspace_unshare __P((struct lwp *)); Index: uvm/uvm_glue.c =================================================================== RCS file: /cvsroot/src/sys/uvm/uvm_glue.c,v retrieving revision 1.77 diff -u -r1.77 uvm_glue.c --- uvm/uvm_glue.c 9 Feb 2004 13:11:21 -0000 1.77 +++ uvm/uvm_glue.c 18 Jun 2004 16:57:24 -0000 @@ -215,16 +215,16 @@ * - the address space is copied as per parent map's inherit values */ void -uvm_proc_fork(p1, p2, shared) +uvm_proc_fork(p1, p2, flags) struct proc *p1, *p2; - boolean_t shared; + int flags; { - if (shared == TRUE) { + if (flags & UVM_FORK_SHARE) { p2->p_vmspace = NULL; uvmspace_share(p1, p2); } else { - p2->p_vmspace = uvmspace_fork(p1->p_vmspace); + p2->p_vmspace = uvmspace_fork(p1->p_vmspace, flags); } cpu_proc_fork(p1, p2); Index: uvm/uvm_map.c =================================================================== RCS file: /cvsroot/src/sys/uvm/uvm_map.c,v retrieving revision 1.164 diff -u -r1.164 uvm_map.c --- uvm/uvm_map.c 24 Mar 2004 07:47:33 -0000 1.164 +++ uvm/uvm_map.c 18 Jun 2004 16:57:27 -0000 @@ -2690,6 +2690,89 @@ } /* + * uvm_map_cpt: control map entry checkpointability features + * + * => map must be unlocked + */ + +int +uvm_map_cpt(struct vm_map *map, vaddr_t start, vaddr_t end, boolean_t cpt) +{ + struct vm_map_entry *entry; + UVMHIST_FUNC("uvm_map_cpt"); UVMHIST_CALLED(maphist); + UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,cpt=0x%x)", + map, start, end, cpt); + + vm_map_lock(map); + VM_MAP_RANGE_CHECK(map, start, end); + + /* see if the range is mapped */ + if (uvm_map_lookup_entry(map, start, &entry) == FALSE) { + vm_map_unlock(map); + return EFAULT; + } + + while ((entry != &map->header) && (entry->start < end)) { + /* + * Check if entry is wired as we require. + */ + if (VM_MAPENT_ISWIRED(entry) == 0) { + vm_map_unlock(map); + return EFAULT; + } + + /* + * Technically we need to clip only the first and last + * entries (if even those), but the macros handle the + * situation, so let's just invoke them always. + */ + UVM_MAP_CLIP_START(map, entry, start); + UVM_MAP_CLIP_END(map, entry, end); + + if (cpt) { + if (entry->flags & UVM_MAP_CPT) { + vm_map_unlock(map); + return EINVAL; + } + entry->flags |= UVM_MAP_CPT; + } else { + if ((entry->flags & UVM_MAP_CPT) == 0) { + vm_map_unlock(map); + return EINVAL; + } + entry->flags &= ~UVM_MAP_CPT; + } + + entry = entry->next; + } + + vm_map_unlock(map); + UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); + return 0; +} + +/* + * Remove all checkpointing entry flags from the map. + */ + +void +uvm_map_cpt_purgeall(struct vm_map *map) +{ + struct vm_map_entry *entry; + UVMHIST_FUNC("uvm_map_cpt_purgeall"); UVMHIST_CALLED(maphist); + UVMHIST_LOG(maphist,"(map=0x%x)", map, 0,0,0); + + vm_map_lock(map); + for (entry = map->header.next; entry != &map->header; + entry = entry->next) { + entry->flags &= ~UVM_MAP_CPT; + } + + vm_map_unlock(map); + UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0); +} + +/* * uvm_map_pageable: sets the pageability of a range in a map. * * => wires map entries. should not be used for transient page locking. @@ -3466,7 +3549,7 @@ return; /* make a new vmspace, still holding old one */ - nvm = uvmspace_fork(ovm); + nvm = uvmspace_fork(ovm, 0); pmap_deactivate(l); /* unbind old vmspace */ p->p_vmspace = nvm; @@ -3611,13 +3694,14 @@ */ struct vmspace * -uvmspace_fork(struct vmspace *vm1) +uvmspace_fork(struct vmspace *vm1, int flags) { struct vmspace *vm2; struct vm_map *old_map = &vm1->vm_map; struct vm_map *new_map; struct vm_map_entry *old_entry; struct vm_map_entry *new_entry; + vm_inherit_t inherit; UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist); vm_map_lock(old_map); @@ -3644,7 +3728,38 @@ KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) || !UVM_ET_ISNEEDSCOPY(old_entry)); - switch (old_entry->inheritance) { + /* + * If we're doing cptfork(), special inheritance treatment. + */ + if (flags & UVM_FORK_CPT) { + if (old_entry->flags & UVM_MAP_CPT) { + inherit = MAP_INHERIT_COPY; + } else { + /* + * The only map entrys that can have cow + * inheritance are the ones marked CPT. + * If COW inheritance is attempted for + * non-CPT map entrys, switch that to + * shared inheritance. + */ + if (old_entry->inheritance == MAP_INHERIT_COPY) + inherit = MAP_INHERIT_SHARE; + else + inherit = old_entry->inheritance; + + /* + * However, we need a COW stack to avoid + * processes mangling each other. + */ + if (old_entry->start>=(vaddr_t)vm1->vm_maxsaddr + && old_entry->end<=(vaddr_t)vm1->vm_minsaddr) + inherit = MAP_INHERIT_COPY; + } + } else { + inherit = old_entry->inheritance; + } + + switch (inherit) { case MAP_INHERIT_NONE: /* @@ -3784,7 +3899,8 @@ * allocated any needed amap (above). */ - if (VM_MAPENT_ISWIRED(old_entry)) { + if (VM_MAPENT_ISWIRED(old_entry) + && !(flags & UVM_FORK_CPT)) { /* * resolve all copy-on-write faults now Index: uvm/uvm_map.h =================================================================== RCS file: /cvsroot/src/sys/uvm/uvm_map.h,v retrieving revision 1.39 diff -u -r1.39 uvm_map.h --- uvm/uvm_map.h 10 Feb 2004 01:30:49 -0000 1.39 +++ uvm/uvm_map.h 18 Jun 2004 16:57:28 -0000 @@ -145,6 +145,7 @@ #define UVM_MAP_STATIC 0x01 /* static map entry */ #define UVM_MAP_KMEM 0x02 /* from kmem entry pool */ #define UVM_MAP_NOMERGE 0x10 /* this entry is not mergable */ +#define UVM_MAP_CPT 0x20 /* entry has cpt-semantics */ }; @@ -238,6 +239,7 @@ #define VM_MAP_WANTLOCK 0x10 /* rw: want to write-lock */ #define VM_MAP_DYING 0x20 /* rw: map is being destroyed */ #define VM_MAP_TOPDOWN 0x40 /* ro: arrange map top-down */ +#define VM_MAP_CPT 0x80 /* rw: map has cpt entries */ /* XXX: number of kernel maps and entries to statically allocate */ @@ -300,6 +302,8 @@ int uvm_map_inherit(struct vm_map *, vaddr_t, vaddr_t, vm_inherit_t); int uvm_map_advice(struct vm_map *, vaddr_t, vaddr_t, int); +int uvm_map_cpt(struct vm_map *, vaddr_t, vaddr_t, int); +void uvm_map_cpt_purgeall(struct vm_map *); void uvm_map_init(void); boolean_t uvm_map_lookup_entry(struct vm_map *, vaddr_t, struct vm_map_entry **); Index: uvm/uvm_stat.c =================================================================== RCS file: /cvsroot/src/sys/uvm/uvm_stat.c,v retrieving revision 1.22 diff -u -r1.22 uvm_stat.c --- uvm/uvm_stat.c 9 Dec 2001 03:07:20 -0000 1.22 +++ uvm/uvm_stat.c 18 Jun 2004 16:57:28 -0000 @@ -61,7 +61,7 @@ #endif #ifdef UVMHIST_PRINT -int uvmhist_print_enabled = 1; +int uvmhist_print_enabled = 0; #endif #ifdef DDB