vdso: Escape double dumping of rt-vdso if proxy present

In case if we have created vdso proxy the rt-vdso should
not be dumped because it will be re-created on next restore
anyway. Thus with help of parasite service routine find
the rt-vdso and tear it off from VMAs list.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>

cr-dump.c

@@ -61,6 +61,7 @@
 #include "kerndat.h"
 #include "stats.h"
 #include "mem.h"
+#include "vdso.h"
 #include "page-pipe.h"
 #include "vdso.h"
 
@@ -1484,6 +1485,12 @@ static int dump_one_task(struct pstree_item *item)
 		}
 	}
 
+	ret = parasite_fixup_vdso(parasite_ctl, pid, &vmas);
+	if (ret) {
+		pr_err("Can't fixup vdso VMAs (pid: %d)\n", pid);
+		goto err_cure_fdset;
+	}
+
 	ret = parasite_dump_misc_seized(parasite_ctl, &misc);
 	if (ret) {
 		pr_err("Can't dump misc (pid: %d)\n", pid);

include/parasite-syscall.h

@@ -84,4 +84,7 @@ int syscall_seized(struct parasite_ctl *ctl, int nr, unsigned long *ret,
 extern int __parasite_execute(struct parasite_ctl *ctl, pid_t pid, user_regs_struct_t *regs);
 extern bool arch_can_dump_task(pid_t pid);
 
+extern int parasite_fixup_vdso(struct parasite_ctl *ctl, pid_t pid,
+			       struct vm_area_list *vma_area_list);
+
 #endif /* __CR_PARASITE_SYSCALL_H__ */

include/parasite.h

@@ -35,6 +35,7 @@ enum {
 	PARASITE_CMD_DRAIN_FDS,
 	PARASITE_CMD_GET_PROC_FD,
 	PARASITE_CMD_DUMP_TTY,
+	PARASITE_CMD_CHECK_VDSO_MARK,
 
 	PARASITE_CMD_MAX,
 };
@@ -60,6 +61,13 @@ struct parasite_vma_entry
 	int		prot;
 };
 
+struct parasite_vdso_vma_entry {
+	unsigned long	start;
+	unsigned long	len;
+	unsigned long	proxy_addr;
+	int		is_marked;
+};
+
 struct parasite_dump_pages_args {
 	unsigned int	nr_vmas;
 	unsigned int	add_prot;

include/vdso.h

@@ -99,6 +99,13 @@ struct vdso_mark {
 #define INIT_VDSO_MARK(m)					\
 	*(m) = (struct vdso_mark)VDSO_MARK_INIT
 
+static inline bool is_vdso_mark(void *addr)
+{
+	struct vdso_mark *m = addr;
+
+	return m->signature == VDSO_MARK_SIGNATURE &&
+		m->proxy_addr != VDSO_BAD_ADDR;
+}
 
 extern struct vdso_symtable vdso_sym_rt;
 extern u64 vdso_pfn;

parasite-syscall.c

@@ -20,9 +20,11 @@
 #include "parasite.h"
 #include "crtools.h"
 #include "namespaces.h"
+#include "kerndat.h"
 #include "pstree.h"
 #include "net.h"
 #include "mem.h"
+#include "vdso.h"
 
 #include <string.h>
 #include <stdlib.h>
@@ -511,6 +513,125 @@ err:
 	return ret;
 }
 
+/*
+ * Find out proxy vdso vma and drop it from the list. Also
+ * fix vdso status on vmas if wrong status found.
+ */
+int parasite_fixup_vdso(struct parasite_ctl *ctl, pid_t pid,
+			struct vm_area_list *vma_area_list)
+{
+	unsigned long proxy_addr = VDSO_BAD_ADDR;
+	struct parasite_vdso_vma_entry *args;
+	struct vma_area *marked = NULL;
+	struct vma_area *vma;
+	int fd, ret = -1;
+	off_t off;
+	u64 pfn;
+
+	args = parasite_args(ctl, struct parasite_vdso_vma_entry);
+	fd = open_proc(pid, "pagemap");
+	if (fd < 0)
+		return -1;
+
+	list_for_each_entry(vma, &vma_area_list->h, list) {
+		if (!vma_area_is(vma, VMA_AREA_REGULAR))
+			continue;
+
+		if ((vma->vma.prot & VDSO_PROT) != VDSO_PROT)
+			continue;
+
+		/*
+		 * I need to poke every potentially marked vma,
+		 * otherwise if task never called for vdso functions
+		 * page frame number won't be reported.
+		 */
+		args->start = vma->vma.start;
+		args->len = vma_area_len(vma);
+
+		if (parasite_execute(PARASITE_CMD_CHECK_VDSO_MARK, ctl)) {
+			pr_err("vdso: Parasite failed to poke for mark\n");
+			ret = -1;
+			goto err;
+		}
+
+		off = (vma->vma.start / PAGE_SIZE) * sizeof(u64);
+		if (lseek(fd, off, SEEK_SET) != off) {
+			pr_perror("Failed to seek address %lx\n", vma->vma.start);
+			ret = -1;
+			goto err;
+		}
+
+		ret = read(fd, &pfn, sizeof(pfn));
+		if (ret < 0 || ret != sizeof(pfn)) {
+			pr_perror("Can't read pme for pid %d", pid);
+			ret = -1;
+			goto err;
+		}
+
+		pfn = PME_PFRAME(pfn);
+		BUG_ON(!pfn);
+
+		/*
+		 * Defer handling marked vdso.
+		 */
+		if (unlikely(args->is_marked)) {
+			BUG_ON(args->proxy_addr == VDSO_BAD_ADDR);
+			BUG_ON(marked);
+			marked = vma;
+			proxy_addr = args->proxy_addr;
+			continue;
+		}
+
+		/*
+		 * Set proper VMA statuses.
+		 */
+		if (pfn == vdso_pfn) {
+			if (!vma_area_is(vma, VMA_AREA_VDSO)) {
+				pr_debug("vdso: Restore status by pfn at %lx\n",
+					 (long)vma->vma.start);
+				vma->vma.status |= VMA_AREA_VDSO;
+			}
+		} else {
+			if (vma_area_is(vma, VMA_AREA_VDSO)) {
+				pr_debug("vdso: Drop mishinted status at %lx\n",
+					 (long)vma->vma.start);
+				vma->vma.status &= ~VMA_AREA_VDSO;
+			}
+		}
+	}
+
+	/*
+	 * There is marked vdso, it means such vdso is autogenerated
+	 * and must be dropped from vma list.
+	 */
+	if (marked) {
+		pr_debug("vdso: Found marked at %lx (proxy at %lx)\n",
+			 (long)marked->vma.start, (long)proxy_addr);
+
+		/*
+		 * Don't forget to restore the proxy vdso status, since
+		 * it's being not recognized by the kernel as vdso.
+		 */
+		list_for_each_entry(vma, &vma_area_list->h, list) {
+			if (vma->vma.start == proxy_addr) {
+				vma->vma.status |= VMA_AREA_REGULAR | VMA_AREA_VDSO;
+				pr_debug("vdso: Restore proxy status at %lx\n",
+					 (long)vma->vma.start);
+				break;
+			}
+		}
+
+		pr_debug("vdso: Droppping marked vdso at %lx\n",
+			 (long)vma->vma.start);
+		list_del(&marked->list);
+		xfree(marked);
+	}
+	ret = 0;
+err:
+	close(fd);
+	return ret;
+}
+
 int parasite_get_proc_fd_seized(struct parasite_ctl *ctl)
 {
 	int ret = -1, fd;

pie/parasite.c

@@ -8,6 +8,7 @@
 
 #include "syscall.h"
 #include "parasite.h"
+#include "vdso.h"
 #include "log.h"
 
 #include <string.h>
@@ -415,6 +416,21 @@ static int parasite_cfg_log(struct parasite_log_args *args)
 	return ret;
 }
 
+static int parasite_check_vdso_mark(struct parasite_vdso_vma_entry *args)
+{
+	struct vdso_mark *m = (void *)args->start;
+
+	if (is_vdso_mark(m)) {
+		args->is_marked = 1;
+		args->proxy_addr = m->proxy_addr;
+	} else {
+		args->is_marked = 0;
+		args->proxy_addr = VDSO_BAD_ADDR;
+	}
+
+	return 0;
+}
+
 static int fini(void)
 {
 	int ret;
@@ -463,6 +479,8 @@ int __used parasite_service(unsigned int cmd, void *args)
 		return parasite_get_proc_fd();
 	case PARASITE_CMD_DUMP_TTY:
 		return parasite_dump_tty(args);
+	case PARASITE_CMD_CHECK_VDSO_MARK:
+		return parasite_check_vdso_mark(args);
 	}
 
 	pr_err("Unknown command to parasite: %d\n", cmd);