execve: Add guard pages around execve KVA buffers

This helps ensure that overflows will trigger a panic instead of
silently corrupting adjacent buffers, as happened in SA-26:13.exec.

Extend kmap_alloc_wait() to support allocation of guard pages on both
sides of a KVA allocation.  Modify the exec_map setup accordingly.  Add
the "vm.exec_map_guard_pages" tunable to provide control over the guard
page allocations.

Reviewed by:	kib
MFC after:	2 weeks
Differential Revision:	https://reviews.freebsd.org/D56711

5 files changed, 42 insertions, 11 deletions

diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c
index 8e3b41170cab..958ec559fd8d 100644
--- a/sys/kern/kern_exec.c
+++ b/sys/kern/kern_exec.c
@@ -1435,7 +1435,8 @@ exec_prealloc_args_kva(void *arg __unused)
 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
 	for (i = 0; i < exec_map_entries; i++) {
 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
-		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
+		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size,
+		    ptoa(exec_map_guard_pages));
 		argkva->gen = exec_args_gen;
 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
 	}
diff --git a/sys/vm/vm_extern.h b/sys/vm/vm_extern.h
index 207eb721d129..799b89ce2fc8 100644
--- a/sys/vm/vm_extern.h
+++ b/sys/vm/vm_extern.h
@@ -51,7 +51,7 @@ void *kva_alloc_aligned(vm_size_t, vm_size_t);
 void kva_free(void *, vm_size_t);
 
 /* These operate on pageable virtual addresses. */
-void *kmap_alloc_wait(vm_map_t, vm_size_t);
+void *kmap_alloc_wait(vm_map_t, vm_size_t, vm_size_t);
 void kmap_free_wakeup(vm_map_t, void *, vm_size_t);
 
 /* These operate on virtual addresses backed by memory. */
diff --git a/sys/vm/vm_init.c b/sys/vm/vm_init.c
index 40a001afb300..316b43c1c240 100644
--- a/sys/vm/vm_init.c
+++ b/sys/vm/vm_init.c
@@ -274,8 +274,12 @@ again:
 	exec_map_entries = 2 * mp_ncpus + 4;
 #endif
 	exec_map_entry_size = round_page(PATH_MAX + ARG_MAX);
-	kmem_subinit(exec_map, kernel_map, &minaddr, &maxaddr,
-	    exec_map_entries * exec_map_entry_size + 64 * PAGE_SIZE, false);
+	exec_map_guard_pages = 1;
+	TUNABLE_INT_FETCH("vm.exec_map_guard_pages", &exec_map_guard_pages);
+	size = exec_map_entries *
+	    (exec_map_entry_size + 2 * ptoa(exec_map_guard_pages)) +
+	    64 * PAGE_SIZE;
+	kmem_subinit(exec_map, kernel_map, &minaddr, &maxaddr, size, false);
 	kmem_subinit(pipe_map, kernel_map, &minaddr, &maxaddr, maxpipekva,
 	    false);
 	TSEXIT();
diff --git a/sys/vm/vm_kern.c b/sys/vm/vm_kern.c
index 3a6fed6cb3e1..fc5d0de424bd 100644
--- a/sys/vm/vm_kern.c
+++ b/sys/vm/vm_kern.c
@@ -107,6 +107,7 @@ CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
 const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
 
 u_int exec_map_entry_size;
+u_int exec_map_guard_pages;
 u_int exec_map_entries;
 
 SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
@@ -706,34 +707,52 @@ kmem_free(void *addr, vm_size_t size)
 		vmem_free(arena, (uintptr_t)addr, size);
 }
 
+static void
+kmap_alloc_map(vm_map_t map, vm_offset_t addr, vm_size_t size,
+    vm_prot_t prot, int flags)
+{
+	int error __diagused;
+
+	error = vm_map_insert(map, NULL, 0,
+	    addr, addr + size, prot, prot, flags);
+	KASSERT(error == KERN_SUCCESS,
+	    ("%s: unexpected error %d", __func__, error));
+}
+
 /*
  *	kmap_alloc_wait:
  *
  *	Allocates pageable memory from a sub-map of the kernel.  If the submap
  *	has no room, the caller sleeps waiting for more memory in the submap.
+ *	If "guard_size" is non-zero, then unmapped KVA is left at the beginning
+ *	and end of the allocated range.
  *
  *	This routine may block.
  */
 void *
-kmap_alloc_wait(vm_map_t map, vm_size_t size)
+kmap_alloc_wait(vm_map_t map, vm_size_t size, vm_size_t guard_size)
 {
 	vm_offset_t addr;
+	vm_size_t total_size;
+
+	KASSERT(size % PAGE_SIZE == 0 && guard_size % PAGE_SIZE == 0,
+	    ("%s: size %zu guard_size %zu", __func__, size, guard_size));
 
-	size = round_page(size);
 	if (!swap_reserve(size))
 		return (NULL);
 
+	total_size = size + 2 * guard_size;
 	for (;;) {
 		/*
 		 * To make this work for more than one map, use the map's lock
 		 * to lock out sleepers/wakers.
 		 */
 		vm_map_lock(map);
-		addr = vm_map_findspace(map, vm_map_min(map), size);
-		if (addr + size <= vm_map_max(map))
+		addr = vm_map_findspace(map, vm_map_min(map), total_size);
+		if (addr + total_size <= vm_map_max(map))
 			break;
 		/* no space now; see if we can ever get space */
-		if (vm_map_max(map) - vm_map_min(map) < size) {
+		if (vm_map_max(map) - vm_map_min(map) < total_size) {
 			vm_map_unlock(map);
 			swap_release(size);
 			return (0);
@@ -741,10 +760,16 @@ kmap_alloc_wait(vm_map_t map, vm_size_t size)
 		vm_map_modflags(map, MAP_NEEDS_WAKEUP, 0);
 		vm_map_unlock_and_wait(map, 0);
 	}
-	vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW,
+	if (guard_size != 0) {
+		kmap_alloc_map(map, addr, guard_size,
+		    VM_PROT_NONE, MAP_CREATE_GUARD);
+		kmap_alloc_map(map, addr + guard_size + size, guard_size,
+		    VM_PROT_NONE, MAP_CREATE_GUARD);
+	}
+	kmap_alloc_map(map, addr + guard_size, size, VM_PROT_RW,
 	    MAP_ACC_CHARGED);
 	vm_map_unlock(map);
-	return ((void *)addr);
+	return ((void *)(addr + guard_size));
 }
 
 /*
diff --git a/sys/vm/vm_kern.h b/sys/vm/vm_kern.h
index 942c03480364..296a50ae0058 100644
--- a/sys/vm/vm_kern.h
+++ b/sys/vm/vm_kern.h
@@ -75,4 +75,5 @@ extern struct vmem *memguard_arena;
 extern u_long vm_kmem_size;
 extern u_int exec_map_entries;
 extern u_int exec_map_entry_size;
+extern u_int exec_map_guard_pages;
 #endif /* _VM_VM_KERN_H_ */