Add kern_vm_mmap2(), kern_vm_mprotect(), kern_vm_msync(), kern_vm_munlock(),

kern_vm_munmap(), and kern_vm_madvise(), and use them in various compats
instead of their sys_*() counterparts.

Reviewed by:	ed, dchagin, kib
MFC after:	2 weeks
Sponsored by:	DARPA, AFRL
Differential Revision:	https://reviews.freebsd.org/D9378

4 files changed, 70 insertions, 131 deletions

diff --git a/sys/compat/cloudabi/cloudabi_mem.c b/sys/compat/cloudabi/cloudabi_mem.c
index 55685284fa9e..42234f9f99c5 100644
--- a/sys/compat/cloudabi/cloudabi_mem.c
+++ b/sys/compat/cloudabi/cloudabi_mem.c
@@ -28,7 +28,9 @@ __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
 #include <sys/mman.h>
-#include <sys/sysproto.h>
+#include <sys/proc.h>
+
+#include <vm/vm_extern.h>
 
 #include <contrib/cloudabi/cloudabi_types_common.h>
 
@@ -62,137 +64,115 @@ int
 cloudabi_sys_mem_advise(struct thread *td,
     struct cloudabi_sys_mem_advise_args *uap)
 {
-	struct madvise_args madvise_args = {
-		.addr	= uap->mapping,
-		.len	= uap->mapping_len
-	};
+	int behav;
 
 	switch (uap->advice) {
 	case CLOUDABI_ADVICE_DONTNEED:
-		madvise_args.behav = MADV_DONTNEED;
+		behav = MADV_DONTNEED;
 		break;
 	case CLOUDABI_ADVICE_NORMAL:
-		madvise_args.behav = MADV_NORMAL;
+		behav = MADV_NORMAL;
 		break;
 	case CLOUDABI_ADVICE_RANDOM:
-		madvise_args.behav = MADV_RANDOM;
+		behav = MADV_RANDOM;
 		break;
 	case CLOUDABI_ADVICE_SEQUENTIAL:
-		madvise_args.behav = MADV_SEQUENTIAL;
+		behav = MADV_SEQUENTIAL;
 		break;
 	case CLOUDABI_ADVICE_WILLNEED:
-		madvise_args.behav = MADV_WILLNEED;
+		behav = MADV_WILLNEED;
 		break;
 	default:
 		return (EINVAL);
 	}
 
-	return (sys_madvise(td, &madvise_args));
+	return (kern_vm_madvise(td, (vm_offset_t)uap->mapping,
+	    uap->mapping_len, behav));
 }
 
 int
 cloudabi_sys_mem_lock(struct thread *td, struct cloudabi_sys_mem_lock_args *uap)
 {
-	struct mlock_args mlock_args = {
-		.addr	= uap->mapping,
-		.len	= uap->mapping_len
-	};
 
-	return (sys_mlock(td, &mlock_args));
+	return (vm_mlock(td->td_proc, td->td_ucred, uap->mapping,
+	    uap->mapping_len));
 }
 
 int
 cloudabi_sys_mem_map(struct thread *td, struct cloudabi_sys_mem_map_args *uap)
 {
-	struct mmap_args mmap_args = {
-		.addr	= uap->addr,
-		.len	= uap->len,
-		.fd	= uap->fd,
-		.pos	= uap->off
-	};
-	int error;
+	int error, flags, prot;
 
 	/* Translate flags. */
+	flags = 0;
 	if (uap->flags & CLOUDABI_MAP_ANON)
-		mmap_args.flags |= MAP_ANON;
+		flags |= MAP_ANON;
 	if (uap->flags & CLOUDABI_MAP_FIXED)
-		mmap_args.flags |= MAP_FIXED;
+		flags |= MAP_FIXED;
 	if (uap->flags & CLOUDABI_MAP_PRIVATE)
-		mmap_args.flags |= MAP_PRIVATE;
+		flags |= MAP_PRIVATE;
 	if (uap->flags & CLOUDABI_MAP_SHARED)
-		mmap_args.flags |= MAP_SHARED;
+		flags |= MAP_SHARED;
 
 	/* Translate protection. */
-	error = convert_mprot(uap->prot, &mmap_args.prot);
+	error = convert_mprot(uap->prot, &prot);
 	if (error != 0)
 		return (error);
 
-	return (sys_mmap(td, &mmap_args));
+	return (kern_vm_mmap(td, (vm_offset_t)uap->addr, uap->len, prot,
+	    flags, uap->fd, uap->off));
 }
 
 int
 cloudabi_sys_mem_protect(struct thread *td,
     struct cloudabi_sys_mem_protect_args *uap)
 {
-	struct mprotect_args mprotect_args = {
-		.addr	= uap->mapping,
-		.len	= uap->mapping_len,
-	};
-	int error;
+	int error, prot;
 
 	/* Translate protection. */
-	error = convert_mprot(uap->prot, &mprotect_args.prot);
+	error = convert_mprot(uap->prot, &prot);
 	if (error != 0)
 		return (error);
 
-	return (sys_mprotect(td, &mprotect_args));
+	return (kern_vm_mprotect(td, (vm_offset_t)uap->mapping,
+	    uap->mapping_len, prot));
 }
 
 int
 cloudabi_sys_mem_sync(struct thread *td, struct cloudabi_sys_mem_sync_args *uap)
 {
-	struct msync_args msync_args = {
-		.addr	= uap->mapping,
-		.len	= uap->mapping_len,
-	};
+	int flags;
 
 	/* Convert flags. */
 	switch (uap->flags & (CLOUDABI_MS_ASYNC | CLOUDABI_MS_SYNC)) {
 	case CLOUDABI_MS_ASYNC:
-		msync_args.flags |= MS_ASYNC;
+		flags = MS_ASYNC;
 		break;
 	case CLOUDABI_MS_SYNC:
-		msync_args.flags |= MS_SYNC;
+		flags = MS_SYNC;
 		break;
 	default:
 		return (EINVAL);
 	}
 	if ((uap->flags & CLOUDABI_MS_INVALIDATE) != 0)
-		msync_args.flags |= MS_INVALIDATE;
+		flags |= MS_INVALIDATE;
 
-	return (sys_msync(td, &msync_args));
+	return (kern_vm_msync(td, (vm_offset_t)uap->mapping,
+	    uap->mapping_len, flags));
 }
 
 int
 cloudabi_sys_mem_unlock(struct thread *td,
     struct cloudabi_sys_mem_unlock_args *uap)
 {
-	struct munlock_args munlock_args = {
-		.addr	= uap->mapping,
-		.len	= uap->mapping_len
-	};
 
-	return (sys_munlock(td, &munlock_args));
+	return (kern_vm_munlock(td, (vm_offset_t)uap->mapping, uap->mapping_len));
 }
 
 int
 cloudabi_sys_mem_unmap(struct thread *td,
     struct cloudabi_sys_mem_unmap_args *uap)
 {
-	struct munmap_args munmap_args = {
-		.addr	= uap->mapping,
-		.len	= uap->mapping_len
-	};
 
-	return (sys_munmap(td, &munmap_args));
+	return (kern_vm_munmap(td, (vm_offset_t)uap->mapping, uap->mapping_len));
 }
diff --git a/sys/compat/freebsd32/freebsd32_misc.c b/sys/compat/freebsd32/freebsd32_misc.c
index b105d03b44dd..d30c520b0e49 100644
--- a/sys/compat/freebsd32/freebsd32_misc.c
+++ b/sys/compat/freebsd32/freebsd32_misc.c
@@ -449,42 +449,30 @@ freebsd32_fexecve(struct thread *td, struct freebsd32_fexecve_args *uap)
 int
 freebsd32_mprotect(struct thread *td, struct freebsd32_mprotect_args *uap)
 {
-	struct mprotect_args ap;
+	int prot;
 
-	ap.addr = PTRIN(uap->addr);
-	ap.len = uap->len;
-	ap.prot = uap->prot;
+	prot = uap->prot;
 #if defined(__amd64__)
-	if (i386_read_exec && (ap.prot & PROT_READ) != 0)
-		ap.prot |= PROT_EXEC;
+	if (i386_read_exec && (prot & PROT_READ) != 0)
+		prot |= PROT_EXEC;
 #endif
-	return (sys_mprotect(td, &ap));
+	return (kern_vm_mprotect(td, (vm_offset_t)PTRIN(uap->addr),
+	    uap->len, prot));
 }
 
 int
 freebsd32_mmap(struct thread *td, struct freebsd32_mmap_args *uap)
 {
-	struct mmap_args ap;
-	vm_offset_t addr = (vm_offset_t) uap->addr;
-	vm_size_t len	 = uap->len;
-	int prot	 = uap->prot;
-	int flags	 = uap->flags;
-	int fd		 = uap->fd;
-	off_t pos	 = PAIR32TO64(off_t,uap->pos);
+	int prot;
 
+	prot = uap->prot;
 #if defined(__amd64__)
 	if (i386_read_exec && (prot & PROT_READ))
 		prot |= PROT_EXEC;
 #endif
 
-	ap.addr = (void *) addr;
-	ap.len = len;
-	ap.prot = prot;
-	ap.flags = flags;
-	ap.fd = fd;
-	ap.pos = pos;
-
-	return (sys_mmap(td, &ap));
+	return (kern_vm_mmap(td, (vm_offset_t)uap->addr, uap->len, prot,
+	    uap->flags, uap->fd, PAIR32TO64(off_t, uap->pos)));
 }
 
 #ifdef COMPAT_FREEBSD6
diff --git a/sys/compat/linux/linux_misc.c b/sys/compat/linux/linux_misc.c
index 139b902b20d7..b73e5d9146f8 100644
--- a/sys/compat/linux/linux_misc.c
+++ b/sys/compat/linux/linux_misc.c
@@ -585,10 +585,8 @@ select_out:
 int
 linux_mremap(struct thread *td, struct linux_mremap_args *args)
 {
-	struct munmap_args /* {
-		void *addr;
-		size_t len;
-	} */ bsd_args;
+	uintptr_t addr;
+	size_t len;
 	int error = 0;
 
 #ifdef DEBUG
@@ -623,10 +621,9 @@ linux_mremap(struct thread *td, struct linux_mremap_args *args)
 	}
 
 	if (args->new_len < args->old_len) {
-		bsd_args.addr =
-		    (caddr_t)((uintptr_t)args->addr + args->new_len);
-		bsd_args.len = args->old_len - args->new_len;
-		error = sys_munmap(td, &bsd_args);
+		addr = args->addr + args->new_len;
+		len = args->old_len - args->new_len;
+		error = kern_vm_munmap(td, addr, len);
 	}
 
 	td->td_retval[0] = error ? 0 : (uintptr_t)args->addr;
@@ -640,13 +637,9 @@ linux_mremap(struct thread *td, struct linux_mremap_args *args)
 int
 linux_msync(struct thread *td, struct linux_msync_args *args)
 {
-	struct msync_args bsd_args;
 
-	bsd_args.addr = (caddr_t)(uintptr_t)args->addr;
-	bsd_args.len = (uintptr_t)args->len;
-	bsd_args.flags = args->fl & ~LINUX_MS_SYNC;
-
-	return (sys_msync(td, &bsd_args));
+	return (kern_vm_msync(td, args->addr, args->len,
+	    args->fl & ~LINUX_MS_SYNC));
 }
 
 int
diff --git a/sys/compat/linux/linux_mmap.c b/sys/compat/linux/linux_mmap.c
index 870f0be418f6..0c3f7688db59 100644
--- a/sys/compat/linux/linux_mmap.c
+++ b/sys/compat/linux/linux_mmap.c
@@ -45,6 +45,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/sysproto.h>
 
 #include <vm/pmap.h>
+#include <vm/vm_extern.h>
 #include <vm/vm_map.h>
 
 #include <compat/linux/linux_emul.h>
@@ -67,15 +68,7 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 {
 	struct proc *p = td->td_proc;
 	struct vmspace *vms = td->td_proc->p_vmspace;
-	struct mmap_args /* {
-		caddr_t addr;
-		size_t len;
-		int prot;
-		int flags;
-		int fd;
-		off_t pos;
-	} */ bsd_args;
-	int error;
+	int bsd_flags, error;
 	struct file *fp;
 
 	cap_rights_t rights;
@@ -83,7 +76,7 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 	    addr, len, prot, flags, fd, pos);
 
 	error = 0;
-	bsd_args.flags = 0;
+	bsd_flags = 0;
 	fp = NULL;
 
 	/*
@@ -94,21 +87,21 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 		return (EINVAL);
 
 	if (flags & LINUX_MAP_SHARED)
-		bsd_args.flags |= MAP_SHARED;
+		bsd_flags |= MAP_SHARED;
 	if (flags & LINUX_MAP_PRIVATE)
-		bsd_args.flags |= MAP_PRIVATE;
+		bsd_flags |= MAP_PRIVATE;
 	if (flags & LINUX_MAP_FIXED)
-		bsd_args.flags |= MAP_FIXED;
+		bsd_flags |= MAP_FIXED;
 	if (flags & LINUX_MAP_ANON) {
 		/* Enforce pos to be on page boundary, then ignore. */
 		if ((pos & PAGE_MASK) != 0)
 			return (EINVAL);
 		pos = 0;
-		bsd_args.flags |= MAP_ANON;
+		bsd_flags |= MAP_ANON;
 	} else
-		bsd_args.flags |= MAP_NOSYNC;
+		bsd_flags |= MAP_NOSYNC;
 	if (flags & LINUX_MAP_GROWSDOWN)
-		bsd_args.flags |= MAP_STACK;
+		bsd_flags |= MAP_STACK;
 
 	/*
 	 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC
@@ -118,14 +111,13 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 	 *
 	 * XXX. Linux checks that the file system is not mounted with noexec.
 	 */
-	bsd_args.prot = prot;
 #if defined(__amd64__)
-	linux_fixup_prot(td, &bsd_args.prot);
+	linux_fixup_prot(td, &prot);
 #endif
 
 	/* Linux does not check file descriptor when MAP_ANONYMOUS is set. */
-	bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd;
-	if (bsd_args.fd != -1) {
+	fd = (bsd_flags & MAP_ANON) ? -1 : fd;
+	if (fd != -1) {
 		/*
 		 * Linux follows Solaris mmap(2) description:
 		 * The file descriptor fildes is opened with
@@ -133,8 +125,7 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 		 * protection options specified.
 		 */
 
-		error = fget(td, bsd_args.fd,
-		    cap_rights_init(&rights, CAP_MMAP), &fp);
+		error = fget(td, fd, cap_rights_init(&rights, CAP_MMAP), &fp);
 		if (error != 0)
 			return (error);
 		if (fp->f_type != DTYPE_VNODE) {
@@ -205,21 +196,13 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 		 * we map the full stack, since we don't have a way
 		 * to autogrow it.
 		 */
-		if (len > STACK_SIZE - GUARD_SIZE) {
-			bsd_args.addr = (caddr_t)addr;
-			bsd_args.len = len;
-		} else {
-			bsd_args.addr = (caddr_t)addr -
-			    (STACK_SIZE - GUARD_SIZE - len);
-			bsd_args.len = STACK_SIZE - GUARD_SIZE;
+		if (len <= STACK_SIZE - GUARD_SIZE) {
+			addr = addr - (STACK_SIZE - GUARD_SIZE - len);
+			len = STACK_SIZE - GUARD_SIZE;
 		}
-	} else {
-		bsd_args.addr = (caddr_t)addr;
-		bsd_args.len  = len;
 	}
-	bsd_args.pos = pos;
 
-	error = sys_mmap(td, &bsd_args);
+	error = kern_vm_mmap(td, addr, len, prot, bsd_flags, fd, pos);
 
 	LINUX_CTR2(mmap2, "return: %d (%p)", error, td->td_retval[0]);
 
@@ -229,16 +212,11 @@ linux_mmap_common(struct thread *td, uintptr_t addr, size_t len, int prot,
 int
 linux_mprotect_common(struct thread *td, uintptr_t addr, size_t len, int prot)
 {
-	struct mprotect_args bsd_args;
-
-	bsd_args.addr = (void *)addr;
-	bsd_args.len = len;
-	bsd_args.prot = prot;
 
 #if defined(__amd64__)
-	linux_fixup_prot(td, &bsd_args.prot);
+	linux_fixup_prot(td, &prot);
 #endif
-	return (sys_mprotect(td, &bsd_args));
+	return (kern_vm_mprotect(td, addr, len, prot));
 }
 
 #if defined(__amd64__)