Added AUFS support.

The AUFS support code handles the "bad" information that we get from
the kernel in /proc/<pid>/map_files and /proc/<pid>/mountinfo files.
For details see comments in sysfs_parse.c.

The main motivation for this work was dumping and restoring Docker
containers which by default use the AUFS graph driver.  For dump,
--aufs-root <container_root> should be added to the command line options.
For restore, there is no need for AUFS-specific command line options
but the container's AUFS filesystem should already be set up before
calling criu restore.

[ xemul: With AUFS files sometimes, in particular -- in case of a
  mapping of an executable file (likekely the one created at elf load),
  in the /proc/pid/map_files/xxx link target we see not the path
  by which the file is seen in AUFS, but the path by which AUFS
  accesses this file from one of its "branches". In order to fix
  the path we get the info about branches from sysfs and when we
  meet such a file, we cut the branch part of the path. ]

Signed-off-by: Saied Kazemi <saied@google.com>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>

12 files changed, 428 insertions, 5 deletions

diff --git a/Makefile.crtools b/Makefile.crtools
index 6033b2ce0..ef30b05b3 100644
--- a/Makefile.crtools
+++ b/Makefile.crtools
@@ -9,6 +9,7 @@ obj-y	+= image-desc.o
 obj-y	+= net.o
 obj-y	+= tun.o
 obj-y	+= proc_parse.o
+obj-y	+= sysfs_parse.o
 obj-y	+= cr-dump.o
 obj-y	+= cr-show.o
 obj-y	+= cr-check.o
diff --git a/cr-dump.c b/cr-dump.c
index 110dc624b..cec579ead 100644
--- a/cr-dump.c
+++ b/cr-dump.c
@@ -71,6 +71,7 @@
 #include "cr-service.h"
 #include "plugin.h"
 #include "irmap.h"
+#include "sysfs_parse.h"
 
 #include "asm/dump.h"
 
@@ -370,6 +371,22 @@ static int dump_filemap(pid_t pid, struct vma_area *vma_area,
 	BUG_ON(!vma_area->st);
 	p.stat = *vma_area->st;
 
+	/*
+	 * AUFS support to compensate for the kernel bug
+	 * exposing branch pathnames in map_files.
+	 *
+	 * If the link found in vma_get_mapfile() pointed
+	 * inside a branch, we should use the pathname
+	 * from root that was saved in vma_area->aufs_rpath.
+	 */
+	if (vma_area->aufs_rpath) {
+		struct fd_link aufs_link;
+
+		strcpy(aufs_link.name, vma_area->aufs_rpath);
+		aufs_link.len = strlen(aufs_link.name);
+		p.link = &aufs_link;
+	}
+
 	if (get_fd_mntid(vma_area->vm_file_fd, &p.mnt_id))
 		return -1;
 
@@ -1927,6 +1944,7 @@ err:
 	free_pstree(root_item);
 	free_file_locks();
 	free_link_remaps();
+	free_aufs_branches();
 
 	close_service_fd(CR_PROC_FD_OFF);
 
diff --git a/crtools.c b/crtools.c
index da870690d..27164c4f5 100644
--- a/crtools.c
+++ b/crtools.c
@@ -172,6 +172,7 @@ int main(int argc, char *argv[])
 		{ "exec-cmd", no_argument, 0, 59},
 		{ "manage-cgroups", no_argument, 0, 60},
 		{ "cgroup-root", required_argument, 0, 61},
+		{ "aufs-root", required_argument, 0, 62},
 		{ },
 	};
 
@@ -397,6 +398,9 @@ int main(int argc, char *argv[])
 		case 'h':
 			usage_error = false;
 			goto usage;
+		case 62:
+			opts.aufs_root = optarg;
+			break;
 		default:
 			goto usage;
 		}
@@ -549,6 +553,9 @@ usage:
 "     --exec-cmd         execute the command specified after '--' on successful\n"
 "                        restore making it the parent of the restored process\n"
 "\n"
+"* AUFS support (applicable to dump only):\n"
+"     --aufs-root        root pathname of the AUFS filesystem\n"
+"\n"
 "* Special resources support:\n"
 "  -x|--" USK_EXT_PARAM "      allow external unix connections\n"
 "     --" SK_EST_PARAM "  checkpoint/restore established TCP connections\n"
diff --git a/include/cr_options.h b/include/cr_options.h
index 73ec50c37..48e4c93cd 100644
--- a/include/cr_options.h
+++ b/include/cr_options.h
@@ -60,6 +60,8 @@ struct cr_options {
 	bool			manage_cgroups;
 	char			*new_global_cg_root;
 	struct list_head	new_cgroup_roots;
+	bool			aufs;		/* auto-deteced, not via cli */
+	char			*aufs_root;
 };
 
 extern struct cr_options opts;
diff --git a/include/fs-magic.h b/include/fs-magic.h
index 12bb982f5..13c4961fb 100644
--- a/include/fs-magic.h
+++ b/include/fs-magic.h
@@ -37,4 +37,8 @@
 #define BTRFS_SUPER_MAGIC	0x9123683E
 #endif
 
+#ifndef AUFS_SUPER_MAGIC
+#define AUFS_SUPER_MAGIC	0x61756673
+#endif
+
 #endif /* __CR_FS_MAGIC_H__ */
diff --git a/include/proc_parse.h b/include/proc_parse.h
index 35d429224..3ac25b260 100644
--- a/include/proc_parse.h
+++ b/include/proc_parse.h
@@ -214,4 +214,7 @@ extern void put_ctls(struct list_head *);
 
 int parse_cgroups(struct list_head *cgroups, unsigned int *n_cgroups);
 
+/* callback for AUFS support */
+extern int aufs_parse(struct mount_info *mi);
+
 #endif /* __CR_PROC_PARSE_H__ */
diff --git a/include/sysfs_parse.h b/include/sysfs_parse.h
new file mode 100644
index 000000000..4d74c4ee5
--- /dev/null
+++ b/include/sysfs_parse.h
@@ -0,0 +1,14 @@
+#ifndef __CR_SYSFS_PARSE_H__
+#define __CR_SYSFS_PARSE_H__
+
+#define SYSFS_AUFS	"/sys/fs/aufs/"
+#define SBINFO_LEN	(3 + 16 + 1)			/* si_%lx */
+#define SBINFO_PATH_LEN	(sizeof SYSFS_AUFS + SBINFO_LEN) /* /sys/fs/aufs/<sbinfo> */
+#define AUFSBR_PATH_LEN	(SBINFO_PATH_LEN + 6 + 1)	/* /sys/fs/aufs/<sbinfo>/br%3d */
+
+extern int parse_aufs_branches(struct mount_info *mi);
+extern int fixup_aufs_vma_fd(struct vma_area *vma);
+extern void free_aufs_branches(void);
+
+#endif /* __CR_SYSFS_PARSE_H__ */
+
diff --git a/include/vma.h b/include/vma.h
index c0bd80bc0..80d228ac8 100644
--- a/include/vma.h
+++ b/include/vma.h
@@ -32,8 +32,14 @@ struct vma_area {
 		int		vm_socket_id;
 		struct file_desc *fd;
 	};
-	unsigned long		*page_bitmap;  /* existent pages */
-	unsigned long		*ppage_bitmap; /* parent's existent pages */
+	union {
+		unsigned long	*page_bitmap;	/* existent pages, restore only */
+		char		*aufs_rpath;	/* path from aufs root, dump only */
+	};
+	union {
+		unsigned long	*ppage_bitmap;	/* parent's existent pages */
+		char		*aufs_fpath;	/* full path from global root, dump only */
+	};
 
 	unsigned long		premmaped_addr;
 
diff --git a/mount.c b/mount.c
index 9a147abe4..459dfee35 100644
--- a/mount.c
+++ b/mount.c
@@ -26,6 +26,7 @@
 #include "protobuf.h"
 #include "kerndat.h"
 #include "fs-magic.h"
+#include "sysfs_parse.h"
 
 #include "protobuf/mnt.pb-c.h"
 
@@ -861,7 +862,11 @@ static struct fstype fstypes[] = {
 	}, {
 		.name = "cgroup",
 		.code = FSTYPE__CGROUP,
-	}
+	}, {
+		.name = "aufs",
+		.code = FSTYPE__AUFS,
+		.parse = aufs_parse,
+	},
 };
 
 struct fstype *find_fstype_by_name(char *fst)
diff --git a/proc_parse.c b/proc_parse.c
index 271d46bfb..768d4f584 100644
--- a/proc_parse.c
+++ b/proc_parse.c
@@ -7,6 +7,7 @@
 #include <errno.h>
 #include <sys/stat.h>
 #include <string.h>
+#include <ctype.h>
 #include <linux/fs.h>
 
 #include "asm/types.h"
@@ -24,8 +25,11 @@
 #include "vma.h"
 
 #include "proc_parse.h"
+#include "cr_options.h"
+#include "sysfs_parse.h"
 #include "protobuf.h"
 #include "protobuf/fdinfo.pb-c.h"
+#include "protobuf/mnt.pb-c.h"
 
 #include <stdlib.h>
 
@@ -240,7 +244,8 @@ static int vma_get_mapfile(struct vma_area *vma, DIR *mfd,
 			vma->vm_socket_id = buf.st_ino;
 		} else if (errno != ENOENT)
 			return -1;
-	}
+	} else if (opts.aufs && fixup_aufs_vma_fd(vma) < 0)
+		return -1;
 
 	return 0;
 }
@@ -455,7 +460,19 @@ int parse_smaps(pid_t pid, struct vm_area_list *vma_area_list, bool use_map_file
 			if (!st_buf)
 				goto err;
 
-			if (fstat(vma_area->vm_file_fd, st_buf) < 0) {
+			/*
+			 * For AUFS support, we cannot fstat() a file descriptor that
+			 * is a symbolic link to a branch (it would return different
+			 * dev/ino than the real file).  Instead, we stat() using the
+			 * full pathname that we saved before.
+			 */
+			if (vma_area->aufs_fpath) {
+				if (stat(vma_area->aufs_fpath, st_buf) < 0) {
+					pr_perror("Failed stat on %d's map %lu (%s)",
+						pid, start, vma_area->aufs_fpath);
+					goto err;
+				}
+			} else if (fstat(vma_area->vm_file_fd, st_buf) < 0) {
 				pr_perror("Failed fstat on %d's map %lu", pid, start);
 				goto err;
 			}
@@ -1718,3 +1735,32 @@ out:
 	fclose(f);
 	return ret;
 }
+
+/*
+ * AUFS callback function to "fix up" the root pathname.
+ * See sysfs_parse.c for details.
+ */
+int aufs_parse(struct mount_info *new)
+{
+	char *cp;
+
+	opts.aufs = true;
+
+	if (!opts.aufs_root || strcmp(new->mountpoint, "./"))
+		return 0;
+
+	cp = malloc(strlen(opts.aufs_root) + 1);
+	if (!cp) {
+		pr_err("Cannot allocate memory for %s\n", opts.aufs_root);
+		return -1;
+	}
+	strcpy(cp, opts.aufs_root);
+
+	pr_debug("Replacing %s with %s\n", new->root, opts.aufs_root);
+	free(new->root);
+	new->root = cp;
+
+	parse_aufs_branches(new);
+
+	return 0;
+}
diff --git a/protobuf/mnt.proto b/protobuf/mnt.proto
index 603bb37d0..2e399762f 100644
--- a/protobuf/mnt.proto
+++ b/protobuf/mnt.proto
@@ -12,6 +12,7 @@ enum fstype {
 	FUSECTL			= 10;
 	DEBUGFS			= 11;
 	CGROUP			= 12;
+	AUFS			= 13;
 };
 
 message mnt_entry {
diff --git a/sysfs_parse.c b/sysfs_parse.c
new file mode 100644
index 000000000..dcfaa4f18
--- /dev/null
+++ b/sysfs_parse.c
@@ -0,0 +1,316 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+#include <ctype.h>
+#include <errno.h>
+#include <sys/types.h>
+#include <dirent.h>
+
+#include "cr_options.h"
+#include "criu-log.h"
+#include "xmalloc.h"
+#include "files.h"
+#include "proc_parse.h"
+#include "util.h"
+#include "sysfs_parse.h"
+#include "namespaces.h"
+
+/*
+ * Currently, there are two kernel problems dealing with AUFS
+ * filesystems.  Until these problems are fixed in the kernel,
+ * we have AUFS support in CRIU to handle the following issues:
+ *
+ * 1) /proc/<pid>/mountinfo: The problem is that for AUFS the root field
+ * of the root entry is missing the pathname (it's only /).  For example:
+ *
+ * 90 61 0:33 / / rw,relatime - aufs none rw,si=4476a910a24617e6
+ *
+ * To handle this issue, the user has to specify the root of the AUFS
+ * filesystem with the --aufs-root command line option.
+ *
+ * 2) /proc/<pid>/map_files: The symlinks are absolute pathnames of the
+ * corresponding *physical* files in the branch they exist.  For example,
+ * for a Docker container using AUFS, a symlink would look like:
+ * 400000-489000 -> /var/lib/docker/aufs/diff/<LAYER_ID>/bin/<cmd>
+ *
+ * Therefore, when we use the link file descriptor vm_file_fd in
+ * dump_one_reg_file() to read the link, we get the file's physical
+ * absolute pathname which does not exist relative to the root of the
+ * mount namespace and even if we used its relative pathname, the dev:ino
+ * values would be different from the physical file's dev:ino causing the
+ * dump to fail.
+ *
+ * To handle this issue, we figure out the "correct" paths when parsing
+ * map_files and save it for later use.  See fixup_aufs_vma_fd() for
+ * details.
+ */
+
+struct ns_id *aufs_nsid;
+static char **aufs_branches;
+
+/*
+ * Parse out and save the AUFS superblock info in the
+ * given buffer.
+ */
+static int parse_aufs_sbinfo(struct mount_info *mi, char *sbinfo, int len)
+{
+	char *cp;
+	int n;
+
+	cp = strstr(mi->options, "si=");
+	if (!cp) {
+		pr_err("Cannot find sbinfo in option string %s\n", mi->options);
+		return -1;
+	}
+
+	/* all ok, copy */
+	if (len < 4) {		/* 4 for "si_" */
+		pr_err("Buffer of %d bytes too small for sbinfo\n", len);
+		return -1;
+	}
+	strcpy(sbinfo, "si_");
+	n = 3;
+	sbinfo += n;
+	cp += n;
+	while (isxdigit(*cp) && n < len) {
+		*sbinfo++ = *cp++;
+		n++;
+	}
+	if (n >= len) {
+		pr_err("Sbinfo in options string %s too long\n", mi->options);
+		return -1;
+	}
+	*sbinfo = '\0';
+	return 0;
+}
+
+/*
+ * If the specified path is in a branch, replace it
+ * with pathname from root.
+ */
+static int fixup_aufs_path(char *path, int size)
+{
+	char rpath[PATH_MAX];
+	int n;
+	int blen;
+
+	if (aufs_branches == NULL) {
+		pr_err("No aufs branches to search for %s\n", path);
+		return -1;
+	}
+
+	for (n = 0; aufs_branches[n] != NULL; n++) {
+		blen = strlen(aufs_branches[n]);
+		if (!strncmp(path, aufs_branches[n], blen))
+			break;
+	}
+
+	if (aufs_branches[n] == NULL)
+		return 0;	/* not in a branch */
+
+	n = snprintf(rpath, PATH_MAX, "%s", &path[blen]);
+	if (n >= min(PATH_MAX, size)) {
+		pr_err("Not enough space to replace %s\n", path);
+		return -1;
+	}
+
+	pr_debug("Replacing %s with %s\n", path, rpath);
+	strcpy(path, rpath);
+	return n;
+}
+
+/*
+ * Kernel stores patchnames to AUFS branches in the br<n> files in
+ * the /sys/fs/aufs/si_<sbinfo> directory where <n> denotes a branch
+ * number and <sbinfo> is a hexadecimal number in %lx format. For
+ * example:
+ *
+ *     $ cat /sys/fs/aufs/si_f598876b087ed883/br0
+ *     /path/to/branch0/directory=rw
+ *
+ * This function sets up an array of pointers to branch pathnames.
+ */
+int parse_aufs_branches(struct mount_info *mi)
+{
+	char path[AUFSBR_PATH_LEN];
+	char *cp;
+	int n;
+	int ret;
+	unsigned int br_num;
+	unsigned int br_max;
+	DIR *dp;
+	FILE *fp;
+	struct dirent *de;
+
+	pr_info("Collecting AUFS branch pathnames ...\n");
+
+	if (mi->nsid == 0) {
+		pr_err("No nsid to parse its aufs branches\n");
+		return -1;
+	}
+
+	if (mi->nsid == aufs_nsid) {
+		pr_debug("Using cached aufs branch paths for nsid %p\n", aufs_nsid);
+		return 0;
+	}
+
+	if (aufs_nsid)
+		free_aufs_branches();
+
+	strcpy(path, SYSFS_AUFS);	/* /sys/fs/aufs/ */
+	if (parse_aufs_sbinfo(mi, &path[sizeof SYSFS_AUFS - 1], SBINFO_LEN) < 0)
+		return -1;
+	if ((dp = opendir(path)) == NULL) {
+		pr_perror("Cannot opendir %s", path);
+		return -1;
+	}
+
+	/*
+	 * Find out how many branches we have.
+	 */
+	br_max = 0;
+	ret = 0;
+	while (1) {
+		errno = 0;
+		if ((de = readdir(dp)) == NULL) {
+			if (errno) {
+				pr_perror("Cannot readdir %s", path);
+				ret = -1;
+			}
+			break;
+		}
+
+		ret = sscanf(de->d_name, "br%d", &br_num);
+		if (ret == 1 && br_num > br_max)
+			br_max = br_num;
+	}
+	closedir(dp);
+	if (ret == -1)
+		return -1;
+
+	/*
+	 * Default AUFS maximum is 127, so 1000 should be plenty.
+	 * If you increase the maximum to more than 3 digits,
+	 * make sure to change AUFSBR_PATH_LEN accordingly.
+	 */
+	if (br_max > 999) {
+		pr_err("Too many branches %d\n", br_max);
+		return -1;
+	}
+
+	/*
+	 * Allocate an array of pointers to branch pathnames to be read.
+	 * Branches are indexed from 0 and we need a NULL pointer at the end.
+	 */
+	aufs_branches = xzalloc((br_max + 2) * sizeof (char *));
+	if (!aufs_branches)
+		return -1;
+
+	/*
+	 * Now read branch pathnames from the branch files.
+	 */
+	n = strlen(path);
+	fp = NULL;
+	for (br_num = 0; br_num <= br_max; br_num++) {
+		ret = snprintf(&path[n], sizeof path - n, "/br%d", br_num);
+		if (ret >= sizeof path - n) {
+			pr_err("Buffer overrun creating path for branch %d\n", br_num);
+			goto err;
+		}
+
+		if ((fp = fopen(path, "r")) == NULL) {
+			pr_perror("Cannot fopen %s", path);
+			goto err;
+		}
+
+		if (fscanf(fp, "%ms=", &aufs_branches[br_num]) != 1 ||
+		    aufs_branches[br_num] == NULL) {
+			pr_perror("Parse error reading %s", path);
+			goto err;
+		}
+
+		/* chop off the trailing "=..." stuff */
+		if ((cp = strchr(aufs_branches[br_num], '=')) == NULL) {
+			pr_err("Bad format in branch pathname %s\n", aufs_branches[br_num]);
+			goto err;
+		}
+		*cp = '\0';
+
+		fclose(fp);
+		/*
+		 * Log branch information for extenal utitilies that
+		 * want to recreate the process's AUFS filesystem
+		 * before calling criu restore.
+		 *
+		 * DO NOT CHANGE this format!
+		 */
+		pr_info("%s : %s\n", path, aufs_branches[br_num]);
+	}
+
+	aufs_nsid = mi->nsid;
+	return 0;
+
+err:
+	if (fp)
+		fclose(fp);
+	free_aufs_branches();
+	return -1;
+}
+
+/*
+ * AUFS support to compensate for the kernel bug
+ * exposing branch pathnames in map_files.
+ *
+ * If the link points inside a branch, save the
+ * relative pathname from the root of the mount
+ * namespace as well as the full pathname from
+ * globl root (/) for later use in dump_filemap()
+ * and parse_smaps().
+ */
+int fixup_aufs_vma_fd(struct vma_area *vma)
+{
+	char path[PATH_MAX];
+	int len;
+
+	path[0] = '.';
+	len = read_fd_link(vma->vm_file_fd, &path[1], sizeof path - 1);
+	if (len < 0)
+		return -1;
+
+	len = fixup_aufs_path(&path[1], sizeof path - 1);
+	if (len < 0)
+		return -1;
+
+	if (len > 0) {
+		vma->aufs_rpath = xmalloc(len + 2);
+		if (!vma->aufs_rpath)
+			return -1;
+		strcpy(vma->aufs_rpath, path);
+		if (opts.aufs_root) {
+			vma->aufs_fpath = xmalloc(strlen(opts.aufs_root) + 1 + len + 1);
+			if (!vma->aufs_fpath)
+				return -1;
+			/* skip ./ in path */
+			sprintf(vma->aufs_fpath, "%s/%s", opts.aufs_root, &path[2]);
+		}
+		pr_debug("Saved AUFS paths %s and %s\n", vma->aufs_rpath, vma->aufs_fpath);
+	}
+
+	return 0;
+}
+
+void free_aufs_branches(void)
+{
+	int n;
+
+	if (aufs_branches) {
+		for (n = 0; aufs_branches[n] != NULL; n++)
+			xfree(aufs_branches[n]);
+
+		xfree(aufs_branches);
+		aufs_branches = NULL;
+	}
+
+	aufs_nsid = NULL;
+}