criu/{mem, vdso, cr-restore}: introduce VMA_AREA_GUARD fake VMAs

Introduce a new kind of VMA - VMA_AREA_GUARD. In fact, it is not
a real VMA as it is not represented as struct vm_area_struct in
the kernel.

We want to reuse an existing vma infrastructure in CRIU to dump
an information about MADV_GUARD_INSTALL-covered address space
ranges as VMAs. Then, on restore, we need to carefully skip
those fake VMAs everywhere we expect a normal VMAs to be processed.
And only in restorer we use these VMAs to get an information about
where to call MADV_GUARD_INSTALL.

Suggested-by: Andrei Vagin <[email protected]>
Signed-off-by: Alexander Mikhalitsyn <[email protected]>

Author: mihalicyn

criu/cr-restore.c

@@ -2447,7 +2447,8 @@ static long restorer_get_vma_hint(struct list_head *tgt_vma_list, struct list_he
 
 	while (1) {
 		if (prev_vma_end + vma_len > s_vma->e->start) {
-			if (s_vma->list.next == self_vma_list) {
+			if ((s_vma->list.next == self_vma_list) ||
+			    vma_area_is(vma_next(s_vma), VMA_AREA_GUARD)) {
 				s_vma = &end_vma;
 				continue;
 			}
@@ -2460,7 +2461,8 @@ static long restorer_get_vma_hint(struct list_head *tgt_vma_list, struct list_he
 		}
 
 		if (prev_vma_end + vma_len > t_vma->e->start) {
-			if (t_vma->list.next == tgt_vma_list) {
+			if ((t_vma->list.next == tgt_vma_list) ||
+			    vma_area_is(vma_next(t_vma), VMA_AREA_GUARD)) {
 				t_vma = &end_vma;
 				continue;
 			}

criu/include/image.h

@@ -68,6 +68,12 @@
  *  	processing exiting with error; while the rest of bits
  *  	are part of image ABI, this particular one must never
  *  	be used in image.
+ *  - guard
+ *  	stands for a fake VMA (not represented in the kernel
+ *  	by a struct vm_area_struct). Used to keep an information
+ *  	about virtual address space ranges covered by
+ *  	MADV_GUARD_INSTALL guards. These ones must be always at
+ *  	the end of the vma_area_list and properly skipped a.e.
  */
 #define VMA_AREA_NONE	  (0 << 0)
 #define VMA_AREA_REGULAR  (1 << 0)
@@ -87,6 +93,7 @@
 #define VMA_AREA_AIORING (1 << 13)
 #define VMA_AREA_MEMFD	 (1 << 14)
 #define VMA_AREA_SHSTK	 (1 << 15)
+#define VMA_AREA_GUARD	 (1 << 16)
 
 #define VMA_EXT_PLUGIN	  (1 << 27)
 #define VMA_CLOSE	  (1 << 28)

criu/mem.c

@@ -599,6 +599,9 @@ static int __parasite_dump_pages_seized(struct pstree_item *item, struct parasit
 		parent_predump_mode = mdc->parent_ie->pre_dump_mode;
 
 	list_for_each_entry(vma_area, &vma_area_list->h, list) {
+		if (vma_area_is(vma_area, VMA_AREA_GUARD))
+			continue;
+
 		ret = generate_vma_iovs(item, vma_area, pp, &xfer, args, ctl, &pmc, has_parent, mdc->pre_dump,
 					parent_predump_mode);
 		if (ret < 0)
@@ -861,14 +864,14 @@ static void prepare_cow_vmas_for(struct vm_area_list *vmas, struct vm_area_list
 		/* <= here to shift from matching VMAs and ... */
 		while (vma->e->start <= pvma->e->start) {
 			vma = vma_next(vma);
-			if (&vma->list == &vmas->h)
+			if ((&vma->list == &vmas->h) || vma_area_is(vma, VMA_AREA_GUARD))
 				return;
 		}
 
 		/* ... no == here since we must stop on matching pair */
 		while (pvma->e->start < vma->e->start) {
 			pvma = vma_next(pvma);
-			if (&pvma->list == &pvmas->h)
+			if ((&pvma->list == &pvmas->h) || vma_area_is(pvma, VMA_AREA_GUARD))
 				return;
 		}
 	}
@@ -1069,6 +1072,9 @@ static int premap_priv_vmas(struct pstree_item *t, struct vm_area_list *vmas, vo
 	filemap_ctx_init(true);
 
 	list_for_each_entry(vma, &vmas->h, list) {
+		if (vma_area_is(vma, VMA_AREA_GUARD))
+			continue;
+
 		if (task_size_check(vpid(t), vma->e)) {
 			ret = -1;
 			break;
@@ -1276,6 +1282,9 @@ static int restore_priv_vma_content(struct pstree_item *t, struct page_read *pr)
 		unsigned long size, i = 0;
 		void *addr = decode_pointer(vma->premmaped_addr);
 
+		if (vma_area_is(vma, VMA_AREA_GUARD))
+			continue;
+
 		if (!vma_inherited(vma))
 			continue;
 

criu/vdso.c

@@ -145,6 +145,9 @@ static void drop_rt_vdso(struct vm_area_list *vma_area_list, struct vdso_quarter
 	 * Also BTW search for rt-vvar to remove it later.
 	 */
 	list_for_each_entry(vma, &vma_area_list->h, list) {
+		if (vma_area_is(vma, VMA_AREA_GUARD))
+			continue;
+
 		if (vma->e->start == addr->orig_vdso) {
 			vma->e->status |= VMA_AREA_REGULAR | VMA_AREA_VDSO;
 			pr_debug("vdso: Restore orig vDSO status at %lx\n", (long)vma->e->start);
@@ -276,6 +279,9 @@ int parasite_fixup_vdso(struct parasite_ctl *ctl, pid_t pid, struct vm_area_list
 	}
 
 	list_for_each_entry(vma, &vma_area_list->h, list) {
+		if (vma_area_is(vma, VMA_AREA_GUARD))
+			continue;
+
 		/*
 		 * Defer handling marked vdso until we walked over
 		 * all vmas and restore potentially remapped vDSO