File 0001-Use-mountinfo-to-be-able-to-use-the-mount-... of Package psmisc

Overview Repositories Revisions Requests Users Attributes Meta

File 0001-Use-mountinfo-to-be-able-to-use-the-mount-identity.patch of Package psmisc (Revision ce46f845016955ada4da1af181138587)

Currently displaying revision ce46f845016955ada4da1af181138587 , Show latest

From 1066ab018235e9981d8770f6c94b294f8b16c7fe Mon Sep 17 00:00:00 2001
From: Werner Fink <werner@suse.de>
Date: Wed, 23 Feb 2022 13:58:32 +0100
Subject: [PATCH] Use mountinfo to be able to use the mount identity

which allows to distinguish different mounts with the
same device number as it happens with NFS shares.
Smaller cleanup as support of chroot environments
and older systems.

Add support for name_to_handle_at() system call to
get the real mount ID for each file

Support also BtrFS with its various subvolumes
On BtrFS stat(2) on binary does not see subvol dev

Allow not unique mounts as well as not unique mountpoint

Fuser does not show open kvm storage image files
such as qcow2 files. Patch from Ali Abdallah <ali.abdallah@suse.com>

Correct last change by using our own namespace to
determine the appropriate mount ID for e.g. qcow2 files.

Determine the namespace of a process only once to speed
up the parsing of fdinfo.

Add a fallback if the system call name_to_handle_at() is
not supported by the used file system.

Signed-off-by: Werner Fink <werner@suse.de>
---
 configure.ac  |  15 +-
 src/fuser.c   | 734 +++++++++++++++++++++++++++++++++++++++-----------
 src/fuser.h   |  25 +-
 src/timeout.c |   5 +-
 4 files changed, 618 insertions(+), 161 deletions(-)

diff --git configure.ac configure.ac
index 182a0df..d4dee51 100644
--- configure.ac
+++ configure.ac
@@ -60,6 +60,19 @@ if test "$enable_timeout_stat" = "static"; then
 fi
 AM_CONDITIONAL([WANT_TIMEOUT_STAT], [test "$enable_timeout_stat" = "static"])
 
+# Use /proc/self/mountinfo if available
+if test -e /proc/self/mountinfo ; then
+  AC_DEFINE([HAS_MOUNTINFO], [1], [System has /proc/self/mountinfo which can used instead /proc(/self)/mounts])
+fi
+# Use /proc/self/fdinfo if available
+if test -e /proc/self/fdinfo ; then
+  AC_DEFINE([HAS_FDINFO], [1], [System has /proc/self/fdinfo for informations on file descriptors])
+fi
+
+# Check for Linux specific name_to_handle_at(2) system call for getting mount IDs
+AC_CHECK_FUNC([name_to_handle_at],[
+  AC_DEFINE([HAS_NAME_TO_HANDLE_AT], [1], [System has name_to_handle_at(2) system call])])
+
 # Use string search for network based file systems but only if the system
 # has /proc/self/mountinfo
 AC_SUBST([WITH_MOUNTINFO_LIST])
@@ -101,7 +114,7 @@ dnl Checks for header files.
 AC_HEADER_DIRENT
 AC_HEADER_STDC
 AC_HEADER_SYS_WAIT
-AC_CHECK_HEADERS([arpa/inet.h fcntl.h langinfo.h libintl.h limits.h locale.h mntent.h netdb.h netinet/in.h stdlib.h string.h sys/ioctl.h sys/socket.h termios.h unistd.h])
+AC_CHECK_HEADERS([arpa/inet.h fcntl.h langinfo.h libintl.h limits.h locale.h mntent.h netdb.h netinet/in.h stdlib.h string.h sys/ioctl.h sys/mount.h sys/param.h sys/stat.h sys/socket.h sys/types.h termios.h unistd.h])
 
 dnl Checks for typedefs, structures, and compiler characteristics.
 AC_C_CONST
diff --git src/fuser.c src/fuser.c
index 03e6237..ee2ba15 100644
--- src/fuser.c
+++ src/fuser.c
@@ -32,6 +32,8 @@
 #include <stdlib.h>
 #include <string.h>
 #include <errno.h>
+#include <fcntl.h>
+#include <sys/mount.h>
 #include <sys/param.h>
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -87,7 +89,7 @@ static void check_map(const pid_t pid, const char *filename,
 		      struct device_list *dev_head,
 		      struct inode_list *ino_head, const uid_t uid,
 		      const char access);
-static struct stat *get_pidstat(const pid_t pid, const char *filename);
+static struct stat *get_pidstat(const pid_t pid, const char *filename, int *id);
 static uid_t getpiduid(const pid_t pid);
 static int print_matches(struct names *names_head, const opt_type opts,
 			 const int sig_number);
@@ -96,9 +98,9 @@ static int kill_matched_proc(struct procs *pptr, const opt_type opts,
 
 /*int parse_mount(struct names *this_name, struct device_list **dev_list);*/
 static void add_device(struct device_list **dev_list,
-		       struct names *this_name, dev_t device);
-void fill_unix_cache(struct unixsocket_list **unixsocket_head);
-void clear_unix_cache(struct unixsocket_list **unixsocket_head);
+		       struct names *this_name, dev_t device, dev_t subvol);
+static void fill_unix_cache(struct unixsocket_list **unixsocket_head);
+static void clear_unix_cache(struct unixsocket_list **unixsocket_head);
 static void atexit_clear_unix_cache();
 static dev_t find_net_dev(void);
 static void scan_procs(struct names *names_head, struct inode_list *ino_head,
@@ -117,9 +119,16 @@ static void debug_match_lists(struct names *names_head,
 			      struct device_list *dev_head);
 #endif
 
-#if defined(WITH_MOUNTINFO_LIST)
+static list_t mntinfo = { &mntinfo, &mntinfo };
 static void clear_mntinfo(void) __attribute__ ((__destructor__));
 static void init_mntinfo(void) __attribute__ ((__constructor__));
+static int get_fdinfo(const pid_t pid, const char *fd, struct fdinfo *info, const ino_t ns);
+#if defined(HAS_NAME_TO_HANDLE_AT)
+static ino_t get_namespace(const pid_t pid);
+static int get_mountid(const char *path);
+#endif
+static int find_mountpoint(const char *path, mntinfo_t **mountinfo);
+#if defined(WITH_MOUNTINFO_LIST)
 static int mntstat(const char *path, struct stat *buf);
 #endif
 static stat_t thestat = stat;
@@ -210,6 +219,7 @@ scan_procs(struct names *names_head, struct inode_list *ino_head,
 		struct stat *cwd_stat = NULL;
 		struct stat *exe_stat = NULL;
 		struct stat *root_stat = NULL;
+		int cwd_id, exe_id, root_id;
 
 		if (topproc_dent->d_name[0] < '0' || topproc_dent->d_name[0] > '9')	/* Not a process */
 			continue;
@@ -219,9 +229,9 @@ scan_procs(struct names *names_head, struct inode_list *ino_head,
 			continue;
 		uid = getpiduid(pid);
 
-		cwd_stat = get_pidstat(pid, "cwd");
-		exe_stat = get_pidstat(pid, "exe");
-		root_stat = get_pidstat(pid, "root");
+		cwd_stat = get_pidstat(pid, "cwd", &cwd_id);
+		exe_stat = get_pidstat(pid, "exe", &exe_id);
+		root_stat = get_pidstat(pid, "root", &root_id);
 		cwd_dev = cwd_stat ? cwd_stat->st_dev : 0;
 		exe_dev = exe_stat ? exe_stat->st_dev : 0;
 		root_dev = root_stat ? root_stat->st_dev : 0;
@@ -229,39 +239,51 @@ scan_procs(struct names *names_head, struct inode_list *ino_head,
 		/* Scan the devices */
 		for (dev_tmp = dev_head; dev_tmp != NULL;
 		     dev_tmp = dev_tmp->next) {
-			if (exe_dev == dev_tmp->device)
-				add_matched_proc(dev_tmp->name, pid, uid,
-						 ACCESS_EXE);
-			if (root_dev == dev_tmp->device)
-				add_matched_proc(dev_tmp->name, pid, uid,
-						 ACCESS_ROOT);
-			if (cwd_dev == dev_tmp->device)
-				add_matched_proc(dev_tmp->name, pid, uid,
-						 ACCESS_CWD);
+			struct subvol *vol_tmp;
+			char access = 0;
+			if (exe_dev == dev_tmp->device && dev_tmp->mnt_id == exe_id)
+				access |= ACCESS_EXE;
+			if (root_dev == dev_tmp->device && dev_tmp->mnt_id == root_id)
+				access |= ACCESS_ROOT;
+			if (cwd_dev == dev_tmp->device && dev_tmp->mnt_id == cwd_id)
+				access |= ACCESS_CWD;
+			for (vol_tmp = dev_tmp->vol; vol_tmp != NULL;
+				vol_tmp = vol_tmp->next) {
+				/* For BtrFS sub volumes there are different
+				   mount ids for the same device */
+				if (exe_dev == vol_tmp->device && vol_tmp->mnt_id == exe_id)
+					access |= ACCESS_EXE;
+				if (root_dev == vol_tmp->device && vol_tmp->mnt_id == root_id)
+					access |= ACCESS_ROOT;
+				if (cwd_dev == vol_tmp->device && vol_tmp->mnt_id == cwd_id)
+					access |= ACCESS_CWD;
+			}
+			if (access)
+				add_matched_proc(dev_tmp->name, pid, uid, access);
 		}
 		for (ino_tmp = ino_head; ino_tmp != NULL;
 		     ino_tmp = ino_tmp->next) {
-			if (exe_dev == ino_tmp->device) {
+			if (exe_dev == ino_tmp->device && ino_tmp->mnt_id == exe_id) {
 				if (!exe_stat)
-					exe_stat = get_pidstat(pid, "exe");
+					exe_stat = get_pidstat(pid, "exe", NULL);
 				if (exe_stat
 				    && exe_stat->st_dev == ino_tmp->device
 				    && exe_stat->st_ino == ino_tmp->inode)
 					add_matched_proc(ino_tmp->name, pid,
 							 uid, ACCESS_EXE);
 			}
-			if (root_dev == ino_tmp->device) {
+			if (root_dev == ino_tmp->device && ino_tmp->mnt_id == root_id) {
 				if (!root_stat)
-					root_stat = get_pidstat(pid, "root");
+					root_stat = get_pidstat(pid, "root", NULL);
 				if (root_stat
 				    && root_stat->st_dev == ino_tmp->device
 				    && root_stat->st_ino == ino_tmp->inode)
 					add_matched_proc(ino_tmp->name, pid,
 							 uid, ACCESS_ROOT);
 			}
-			if (cwd_dev == ino_tmp->device) {
+			if (cwd_dev == ino_tmp->device && ino_tmp->mnt_id == cwd_id) {
 				if (!cwd_stat)
-					cwd_stat = get_pidstat(pid, "cwd");
+					cwd_stat = get_pidstat(pid, "cwd", NULL);
 				if (cwd_stat
 				    && cwd_stat->st_dev == ino_tmp->device
 				    && cwd_stat->st_ino == ino_tmp->inode)
@@ -302,18 +324,42 @@ add_inode(struct inode_list **ino_list, struct names *this_name,
 	ino_tmp->name = this_name;
 	ino_tmp->device = device;
 	ino_tmp->inode = inode;
+	ino_tmp->mnt_id = this_name->mnt_id;
 	ino_tmp->next = ino_head;
 	*ino_list = ino_tmp;
 }
+	
+static void
+add_subvol(struct subvol **head, dev_t device, int mnt_id)
+{
+	struct subvol *vol_tmp, *vol_head;
+
+	if ((vol_tmp =
+	     (struct subvol *)malloc(sizeof(struct subvol))) == NULL)
+		return;
+	vol_head = *head;
+	vol_tmp->device = device;
+	vol_tmp->mnt_id = mnt_id;
+	vol_tmp->next = vol_head;
+	*head = vol_tmp;
+}
 
 static void
-add_device(struct device_list **dev_list, struct names *this_name, dev_t device)
+add_device(struct device_list **dev_list, struct names *this_name, dev_t device, dev_t subvol)
 {
 	struct device_list *dev_tmp, *dev_head;
 #ifdef DEBUG
 	fprintf(stderr, "add_device(%s %u\n", this_name->filename,
 		(unsigned int)device);
 #endif				/* DEBUG */
+	/* For BtrFS there are many sub volumes all on the same block device */
+	for (dev_tmp = *dev_list; dev_tmp != NULL; dev_tmp = dev_tmp->next)
+		if (dev_tmp->device == device) {
+			if (dev_tmp->device != subvol)
+				add_subvol(&dev_tmp->vol, subvol, this_name->mnt_id);
+			*dev_list = dev_tmp;
+			return;
+		}
 
 	if ((dev_tmp =
 	     (struct device_list *)malloc(sizeof(struct device_list))) == NULL)
@@ -321,6 +367,10 @@ add_device(struct device_list **dev_list, struct names *this_name, dev_t device)
 	dev_head = *dev_list;
 	dev_tmp->name = this_name;
 	dev_tmp->device = device;
+	dev_tmp->mnt_id = this_name->mnt_id;
+	dev_tmp->vol = NULL;
+	if (dev_tmp->device != subvol)
+		add_subvol(&dev_tmp->vol, subvol, this_name->mnt_id);
 	dev_tmp->next = dev_head;
 	*dev_list = dev_tmp;
 }
@@ -462,13 +512,15 @@ add_special_proc(struct names *name_list, const char ptype, const uid_t uid,
 int parse_file(struct names *this_name, struct inode_list **ino_list,
 	       const opt_type opts)
 {
+	mntinfo_t *mountinfo;
 	char *new = expandpath(this_name->filename);
 	if (new) {
 		if (this_name->filename)
 			free(this_name->filename);
 		this_name->filename = strdup(new);
 	}
-	if (timeout(thestat, this_name->filename, &(this_name->st), 5) != 0) {
+	if (timeout(thestat, this_name->filename, &(this_name->st), 5) != 0 ||
+	    find_mountpoint(this_name->filename, &mountinfo) != 0) {
 		if (errno == ENOENT)
 			fprintf(stderr,
 				_("Specified filename %s does not exist.\n"),
@@ -478,10 +530,12 @@ int parse_file(struct names *this_name, struct inode_list **ino_list,
 				this_name->filename, strerror(errno));
 		return -1;
 	}
+	this_name->mnt_id = mountinfo->id;
 #ifdef DEBUG
-	printf("adding file %s %lX %lX\n", this_name->filename,
+	printf("adding file %s %lX %lX %d nfs=%s\n", this_name->filename,
 	       (unsigned long)this_name->st.st_dev,
-	       (unsigned long)this_name->st.st_ino);
+	       (unsigned long)this_name->st.st_ino,
+	       mountinfo->id, mountinfo->isremote ? "yes" : "no");
 #endif				/* DEBUG */
 	add_inode(ino_list, this_name, this_name->st.st_dev,
 		  this_name->st.st_ino);
@@ -513,12 +567,43 @@ parse_mounts(struct names *this_name, struct device_list **dev_list,
 	     const opt_type opts)
 {
 	dev_t match_device;
+	list_t *ptr;
+	int count;
 
 	if (S_ISBLK(this_name->st.st_mode))
 		match_device = this_name->st.st_rdev;
 	else
 		match_device = this_name->st.st_dev;
-	add_device(dev_list, this_name, match_device);
+
+	count = 0;
+	list_for_each(ptr, &mntinfo) {
+		mntinfo_t *mnt = list_entry(ptr, mntinfo_t);
+		if (match_device != mnt->dev)
+			continue;
+		if (S_ISBLK(this_name->st.st_mode)) {
+			/* Correct mount IDs check if a block device
+			 * was specified */
+			this_name->mnt_id = mnt->id;
+			add_device(dev_list, this_name, match_device, mnt->dev);
+			if (mnt->dev == mnt->vol)
+				count++;
+			else	count = 1;
+			continue;
+		}
+		if (this_name->mnt_id != mnt->id)
+			continue;
+		count++;
+	}
+	if (count == 0) {
+		errno = ENOENT;
+		fprintf(stderr,
+			_("Specified filename %s has no mountpoint.\n"),
+			this_name->filename);
+		return -1;
+	}
+	if (S_ISBLK(this_name->st.st_mode))
+		return 0;
+	add_device(dev_list, this_name, match_device, match_device);
 	return 0;
 }
 
@@ -695,10 +780,12 @@ int parse_inet(struct names *this_name, struct ip_connections **ip_list)
 					break;
 #endif
 				}
-			}	/*while */
-		    freeaddrinfo(res);
+			}	/* for (;;) */
+
+			freeaddrinfo(res);
 			return 0;
-		}
+
+		}	/* if getaddrinfo */
 	}
 	return 1;
 }
@@ -965,6 +1052,21 @@ free_inodes(struct inode_list **match_inodes)
 /*
  * Free up structures allocated in add_device
  */
+
+static void
+free_subvol(struct subvol **volumes)
+{
+	struct subvol *vol_tmp, *vol_next;
+
+	vol_tmp = *volumes;
+	while (vol_tmp != NULL) {
+		vol_next = vol_tmp->next;
+		free(vol_tmp);
+		vol_tmp = vol_next;
+	}
+	*volumes =NULL;
+}
+
 static void
 free_devices(struct device_list **match_devices)
 {
@@ -972,6 +1074,8 @@ free_devices(struct device_list **match_devices)
 
 	device_tmp = *match_devices;
 	while(device_tmp != NULL) {
+		if (device_tmp->vol)
+			free_subvol(&device_tmp->vol);
 		device_next = device_tmp->next;
 		free(device_tmp);
 		device_tmp = device_next;
@@ -1174,16 +1278,11 @@ int main(int argc, char *argv[])
 					skip_argv = 1;
 					//while(option != '\0') option++;
 					if (strcmp(argv[argc_cnt], "tcp") == 0)
-						default_namespace =
-						    NAMESPACE_TCP;
-					else if (strcmp(argv[argc_cnt], "udp")
-						 == 0)
-						default_namespace =
-						    NAMESPACE_UDP;
-					else if (strcmp(argv[argc_cnt], "file")
-						 == 0)
-						default_namespace =
-						    NAMESPACE_FILE;
+						default_namespace = NAMESPACE_TCP;
+					else if (strcmp(argv[argc_cnt], "udp") == 0)
+						default_namespace = NAMESPACE_UDP;
+					else if (strcmp(argv[argc_cnt], "file") == 0)
+						default_namespace = NAMESPACE_FILE;
 					else
 						usage(_
 						      ("Invalid namespace name"));
@@ -1223,7 +1322,7 @@ int main(int argc, char *argv[])
 		}
 
 #if defined(WITH_MOUNTINFO_LIST)
-		if ((opts & OPT_ALWAYSSTAT) == 0)
+		if ((opts & (OPT_MOUNTS|OPT_ALWAYSSTAT)) == OPT_MOUNTS)
 			thestat = mntstat;
 #endif
 		/* an option */
@@ -1537,7 +1636,7 @@ print_matches(struct names *names_head, const opt_type opts,
 
 }
 
-static struct stat *get_pidstat(const pid_t pid, const char *filename)
+static struct stat *get_pidstat(const pid_t pid, const char *filename, int *id)
 {
 	char pathname[PATH_MAX];
 	struct stat *st;
@@ -1549,6 +1648,15 @@ static struct stat *get_pidstat(const pid_t pid, const char *filename)
 		free(st);
 		return NULL;
 	}
+
+	if (id) {
+		mntinfo_t *info;
+		char *new = expandpath(pathname);
+		if (new && find_mountpoint(new, &info) == 0)
+			*id = info->id;
+		else	*id = -1;
+	}
+
 	return st;
 }
 
@@ -1563,10 +1671,11 @@ check_dir(const pid_t pid, const char *dirname, struct device_list *dev_head,
 	struct inode_list *ino_tmp;
 	struct device_list *dev_tmp;
 	struct unixsocket_list *sock_tmp;
-	struct stat st, lst;
+	struct stat st;
 	char *dirpath;
 	char filepath[PATH_MAX];
 	char real_filepath[PATH_MAX];
+	ino_t ns;
 
 	if (asprintf(&dirpath, "/proc/%d/%s", pid, dirname) < 0)
         return;
@@ -1576,6 +1685,12 @@ check_dir(const pid_t pid, const char *dirname, struct device_list *dev_head,
     }
     free(dirpath);
 
+#if defined(HAS_NAME_TO_HANDLE_AT)
+	ns = get_namespace(pid);
+#else
+	ns = -1;
+#endif
+
 	while ((direntry = readdir(dirp)) != NULL) {
 		if (direntry->d_name[0] < '0' || direntry->d_name[0] > '9')
 			continue;
@@ -1589,6 +1704,9 @@ check_dir(const pid_t pid, const char *dirname, struct device_list *dev_head,
 					filepath, strerror(errno));
 			}
 		} else {
+			struct fdinfo fd;
+			int fdret;
+
 			thedev = st.st_dev;
 			if (thedev == netdev) {
 				for (sock_tmp = sockets; sock_tmp != NULL;
@@ -1601,10 +1719,29 @@ check_dir(const pid_t pid, const char *dirname, struct device_list *dev_head,
 					}
 				}
 			}
+
+			memset(&fd, 0, sizeof(struct fdinfo));
+			fdret = get_fdinfo(pid, direntry->d_name, &fd, ns);
+
 			for (dev_tmp = dev_head; dev_tmp != NULL;
 			     dev_tmp = dev_tmp->next) {
-				if (thedev != dev_tmp->device)
-					continue;
+
+				if (thedev != dev_tmp->device) {
+					struct subvol *vol_tmp;
+					int subvol_found = 0;
+
+					for (vol_tmp = dev_tmp->vol; vol_tmp != NULL;
+					     vol_tmp = vol_tmp->next) {
+						/* Check for BtrFS sub volumes as well */
+						if (thedev == vol_tmp->device) {
+							subvol_found++;
+							break;
+						}
+					}
+
+					if (!subvol_found)
+						continue;
+				}
 
 				/* check the paths match if it is not a block device */
 				if (! S_ISBLK(dev_tmp->name->st.st_mode)) {
@@ -1616,9 +1753,14 @@ check_dir(const pid_t pid, const char *dirname, struct device_list *dev_head,
 					    continue;
 				    }
 				}
+
+				if (fdret != 0)
+					continue;
+				if (fd.mnt_id != dev_tmp->mnt_id)
+					continue;
+
 				if (access == ACCESS_FILE
-				    && (lstat(filepath, &lst) == 0)
-				    && (lst.st_mode & S_IWUSR)) {
+				    && (fd.flags & (O_WRONLY|O_RDWR))) {
 					add_matched_proc(dev_tmp->name,
 							 pid, uid,
 							 ACCESS_FILEWR |
@@ -1640,9 +1782,10 @@ check_dir(const pid_t pid, const char *dirname, struct device_list *dev_head,
 					continue;
 				}
 				if (st.st_ino == ino_tmp->inode) {
+					if (fdret != 0)
+						continue;
 					if (access == ACCESS_FILE
-					    && (lstat(filepath, &lst) == 0)
-					    && (lst.st_mode & S_IWUSR)) {
+					    && (fd.flags & (O_WRONLY|O_RDWR))) {
 						add_matched_proc(ino_tmp->name,
 								 pid, uid,
 								 ACCESS_FILEWR |
@@ -1671,31 +1814,54 @@ check_map(const pid_t pid, const char *filename,
 	FILE *fp;
 	unsigned long long tmp_inode;
 	unsigned int tmp_maj, tmp_min;
-	dev_t tmp_device;
 
 	if (asprintf(&pathname, "/proc/%d/%s", pid, filename) < 0)
         return;
 	if ((fp = fopen(pathname, "r")) == NULL) {
-        free(pathname);
+		free(pathname);
 		return;
-    }
-    free(pathname);
+	}
+	free(pathname);
 	while (fgets(line, BUFSIZ, fp)) {
-		if (sscanf(line, "%*s %*s %*s %x:%x %lld",
-			   &tmp_maj, &tmp_min, &tmp_inode) == 3) {
-			tmp_device = tmp_maj * 256 + tmp_min;
+		char *scanned_path = NULL;
+		if (sscanf(line, "%*s %*s %*s %x:%x %lld %ms",
+			   &tmp_maj, &tmp_min, &tmp_inode, &scanned_path) == 4) {
+			dev_t tmp_device = makedev(tmp_maj, tmp_min);
+			int mnt_id = -1;
+
+			if (scanned_path) {
+				if (*scanned_path == '/') {
+					mntinfo_t *mountinfo;
+					if (find_mountpoint(scanned_path, &mountinfo) == 0)
+						mnt_id = mountinfo->id;
+				}
+				free(scanned_path);
+			}
+
 			for (dev_tmp = dev_head; dev_tmp != NULL;
-			     dev_tmp = dev_tmp->next)
-				if (dev_tmp->device == tmp_device)
+			     dev_tmp = dev_tmp->next) {
+				struct subvol *vol_tmp;
+				if (dev_tmp->device == tmp_device &&
+				    mnt_id == dev_tmp->mnt_id)
 					add_matched_proc(dev_tmp->name, pid,
 							 uid, access);
+				for (vol_tmp = dev_tmp->vol; vol_tmp != NULL;
+				     vol_tmp = vol_tmp->next) {
+					/* For BtrFS sub volumes there are different
+					   mount ids for the same device */
+					if (vol_tmp->device == tmp_device && mnt_id == vol_tmp->mnt_id)
+						add_matched_proc(dev_tmp->name, pid,
+								 uid, access);
+				}
+			}
 			for (ino_tmp = ino_head; ino_tmp != NULL;
 			     ino_tmp = ino_tmp->next)
 				if (ino_tmp->device == tmp_device
 				    && ino_tmp->inode == tmp_inode)
 					add_matched_proc(ino_tmp->name, pid,
 							 uid, access);
-		}
+		} else if (scanned_path)
+			free(scanned_path);
 	}
 	fclose(fp);
 }
@@ -1719,6 +1885,7 @@ static uid_t getpiduid(const pid_t pid)
  * fill_unix_cache : Create a list of Unix sockets
  *   This list is used later for matching purposes
  */
+static
 void fill_unix_cache(struct unixsocket_list **unixsocket_head)
 {
 	FILE *fp;
@@ -1735,6 +1902,8 @@ void fill_unix_cache(struct unixsocket_list **unixsocket_head)
 	while (fgets(line, BUFSIZ, fp) != NULL) {
 		char *path;
 		char *scanned_path = NULL;
+		int mnt_id = -1;
+		mntinfo_t *mountinfo;
 		if (sscanf(line, "%*x: %*x %*x %*x %*x %*d %llu %ms",
 			   &scanned_inode, &scanned_path) != 2) {
 			if (scanned_path)
@@ -1750,6 +1919,8 @@ void fill_unix_cache(struct unixsocket_list **unixsocket_head)
 			free(path);
 			continue;
 		}
+		if (find_mountpoint(scanned_path, &mountinfo) == 0)
+			mnt_id = mountinfo->id;
 		if ((newsocket = (struct unixsocket_list *)
 		     malloc(sizeof(struct unixsocket_list))) == NULL) {
 			free(path);
@@ -1758,6 +1929,7 @@ void fill_unix_cache(struct unixsocket_list **unixsocket_head)
 		newsocket->sun_name = strdup(scanned_path);
 		newsocket->inode = st.st_ino;
 		newsocket->dev = st.st_dev;
+		newsocket->mnt_id = mnt_id;
 		newsocket->net_inode = scanned_inode;
 		newsocket->next = *unixsocket_head;
 		*unixsocket_head = newsocket;
@@ -1770,6 +1942,7 @@ void fill_unix_cache(struct unixsocket_list **unixsocket_head)
 /*
  * Free up the list of Unix sockets
  */
+static
 void clear_unix_cache(struct unixsocket_list **unixsocket_head)
 {
 	while(*unixsocket_head != NULL) {
@@ -1941,34 +2114,21 @@ scan_mounts(struct names *names_head, struct inode_list *ino_head,
 {
 	struct device_list *dev_tmp;
 	struct inode_list *ino_tmp;
-	FILE *fp;
-	char line[BUFSIZ];
-	char *find_mountp;
-	char *find_space;
 	struct stat st;
+	list_t *ptr;
 
-	if ( (ino_head == NULL) && (dev_head == NULL) )
-		return;
-
+	list_for_each(ptr, &mntinfo) {
+		mntinfo_t *mnt = list_entry(ptr, mntinfo_t);
+		const char *find_mountp = mnt->mpoint;
 
-	if ((fp = fopen(PROC_MOUNTS, "r")) == NULL) {
-		fprintf(stderr, "Cannot open %s\n", PROC_MOUNTS);
-		return;
-	}
-	while (fgets(line, BUFSIZ, fp) != NULL) {
-		if ((find_mountp = strchr(line, ' ')) == NULL)
-			continue;
-		find_mountp++;
-		if ((find_space = strchr(find_mountp, ' ')) == NULL)
+		if (timeout(thestat, find_mountp, &st, 5) != 0)
 			continue;
-		*find_space = '\0';
-		if (timeout(thestat, find_mountp, &st, 5) != 0) {
-			continue;
-		}
+
 		/* Scan the devices */
 		for (dev_tmp = dev_head; dev_tmp != NULL;
 		     dev_tmp = dev_tmp->next) {
-			if (st.st_dev == dev_tmp->device)
+			if (st.st_dev == dev_tmp->device &&
+			    mnt->id == dev_tmp->mnt_id)
 				add_special_proc(dev_tmp->name, PTYPE_MOUNT, 0,
 						 find_mountp);
 		}
@@ -1980,7 +2140,6 @@ scan_mounts(struct names *names_head, struct inode_list *ino_head,
 						 find_mountp);
 		}
 	}
-	fclose(fp);
 }
 
 static void
@@ -2033,16 +2192,44 @@ scan_swaps(struct names *names_head, struct inode_list *ino_head,
 	fclose(fp);
 }
 
-#if defined(WITH_MOUNTINFO_LIST)
 /*
  * Use /proc/self/mountinfo of modern linux system to determine
  * the device numbers of the mount points. Use this to avoid the
- * stat(2) system call wherever possible.
+ * stat(2) system call for remote file system.
  */
 
-static list_t mntinfo = { &mntinfo, &mntinfo };
+static mntinfo_t*
+add_mntinfo(const char *mpoint, const char *type, int mid, int parid, dev_t dev)
+{
+	const size_t nlen = strlen(mpoint);
+	mntinfo_t *restrict mnt;
+	if (posix_memalign
+	    ((void *)&mnt, sizeof(void *),
+	     alignof(mntinfo_t) + (nlen + 1)) != 0) {
+		fprintf(stderr,
+			_
+			("Cannot allocate memory for matched proc: %s\n"),
+			strerror(errno));
+		exit(1);
+	}
+	append(mnt, mntinfo);
+	mnt->mpoint = ((char *)mnt) + alignof(mntinfo_t);
+	strcpy(mnt->mpoint, mpoint);
+	mnt->nlen = nlen;
+	mnt->parid = parid;
+	mnt->dev = dev;
+	mnt->id = mid;
+	if (strncmp("nfs", type, 3) == 0 || strncmp("afs", type, 3) == 0 || strncmp("autofs", type, 6))
+		mnt->isremote = 1;
+	else	mnt->isremote = 0;
+	/* E.g. sub volumes of BtrFS do not show main device number in
+	   /proc/self/mountinfo, for now just map the device to this */
+	mnt->vol = dev;
+	return mnt;
+}
 
-static void clear_mntinfo(void)
+static void
+clear_mntinfo(void)
 {
 	list_t *ptr, *tmp;
 
@@ -2053,72 +2240,319 @@ static void clear_mntinfo(void)
 	}
 }
 
-static void init_mntinfo(void)
+static void
+init_mntinfo(void)
 {
+	char type[256];
 	char mpoint[PATH_MAX *4 + 1]; // octal escaping takes 4 chars per 1 char
-	int mid, parid, max = 0;
+	char devname[PATH_MAX];
+	int mid, parid;
+	mntinfo_t *mntinf;
+#if defined(HAS_MOUNTINFO)
 	uint maj, min;
-	list_t sort;
+#endif
 	FILE *mnt;
 
 	if (!list_empty(&mntinfo))
 		return;
-	if ((mnt = fopen("/proc/self/mountinfo", "r")) == (FILE *) 0)
+#if defined(HAS_MOUNTINFO)
+	/*
+	 * We could also loop over all mount namespaces, that is not only
+	 * over /proc/self/ns/mnt.  But this expands this list as well as
+	 * we can use name_to_handle_at(2) in our get_mountid() function.
+	 */
+	if ((mnt = fopen(PROC_MOUNTINFO, "r")) == (FILE *) 0)
 		return;
 	while (fscanf
-	       (mnt, "%i %i %u:%u %*s %s %*[^\n]", &mid, &parid, &maj, &min,
-		&mpoint[0]) == 5) {
-		const size_t nlen = strlen(mpoint);
-		mntinfo_t *restrict mnt;
-		if (posix_memalign
-		    ((void *)&mnt, sizeof(void *),
-		     alignof(mntinfo_t) + (nlen + 1)) != 0) {
-			fprintf(stderr,
-				_
-				("Cannot allocate memory for matched proc: %s\n"),
-				strerror(errno));
-			exit(1);
+	       (mnt, "%i %i %u:%u %*s %s %*[^-] - %s %s %*[^\n]",
+		&mid, &parid, &maj, &min, &mpoint[0], &type[0], &devname[0]) == 7) {
+		struct stat st;
+		mntinf = add_mntinfo(mpoint, type, mid, parid, makedev(maj, min));
+		if (mntinf && strncmp(devname, "/dev/", 5) == 0 && stat(devname, &st) == 0) {
+			if (st.st_rdev != 0 && mntinf->dev != st.st_rdev) {
+				mntinf->vol = st.st_rdev;
+				stat(mpoint, &st);
+				mntinf->dev = st.st_dev;	/* stat(2) on binary does not see subvol dev */
+			}
 		}
-		append(mnt, mntinfo);
-		mnt->mpoint = ((char *)mnt) + alignof(mntinfo_t);
-		strcpy(mnt->mpoint, mpoint);
-		mnt->nlen = nlen;
-		mnt->parid = parid;
-		mnt->dev = makedev(maj, min);
-		mnt->id = mid;
-		if (mid > max)
-			max = mid;
 	}
+#else
+	if ((mnt = fopen(PROC_MOUNTS, "r")) == (FILE *) 0)
+		return;
+	mid = 1;
+	parid = -1;
+	while (fscanf (mnt, "%s %s %s %*[^\n]", &devname[0], &mpoint[0], &type[0]) == 3) {
+		struct stat st;
+		if (stat(mpoint, &st) != 0) {
+			if (errno != EACCES) {
+				fprintf(stderr, _("Cannot stat %s: %s\n"),
+					mpoint, strerror(errno));
+				exit(1);
+			}
+			st.st_dev = (dev_t)-1;
+		}
+		mntinf = add_mntinfo(mpoint, type, mid++, parid, st.st_dev);
+		if (mntinf && strncmp(devname, "/dev/", 5) == 0 && stat(devname, &st) == 0) {
+			if (st.st_rdev != 0 && mntinf->dev != st.st_rdev)
+				mntinf->vol = st.st_rdev;
+		}
+	}
+#endif
 	fclose(mnt);
+}
+
+static int
+get_fdinfo(const pid_t pid, const char *fd, struct fdinfo *info, const ino_t ns)
+{
+	int ret = 0;
+	char pathname[512];
+	int mnt_id = 0, flags = 0;
+#if defined(HAS_FDINFO)					/* Here we get the private namespace as well */
+	const static char delimiters[] = ": \t\n";
+	char line[BUFSIZ];
+	FILE *fp;
+# if defined(HAS_NAME_TO_HANDLE_AT)
+	static ino_t mynamespace;
+	struct stat st;
+ 
+	if (!mynamespace) {
+		if (stat("/proc/self/ns/mnt", &st) != 0) {
+			fprintf(stderr, _("Cannot stat %s: %s\n"),
+				"/proc/self/ns/mnt", strerror(errno));
+			exit(1);
+		}
+		mynamespace = st.st_ino;
+	}
+# endif
+	snprintf(pathname, sizeof(pathname)-1, "/proc/%d/fdinfo/%s", pid, fd);
+	if ((fp = fopen(pathname, "r")) == NULL)
+               goto out;				/* forbidden namesspace, try our own namespace */
+	while (fgets(line, BUFSIZ, fp) && ret < 2) {
+		char *xp, *vp, *ep;
+		unsigned long ul;
+		xp = strtok(&line[0], delimiters);
+		if (!xp || *xp == 0)
+			continue;
+		vp = strtok(NULL, delimiters);
+		if (!vp || *vp == 0)
+			continue;
+		if (strcmp(xp, "flags")  == 0 && (ul = strtoul(vp, &ep, 0)) != ULONG_MAX && ep && *ep == 0) {
+			info->flags = (mode_t)ul;
+			flags++;
+			ret++;
+		}
+		if (strcmp(xp, "mnt_id") == 0 && (ul = strtoul(vp, &ep, 0)) != ULONG_MAX && ep && *ep == 0) {
+			info->mnt_id = (int)ul;
+			mnt_id++;
+			ret++;
+		}
+	}
+	fclose(fp);
+out:
+# if defined(HAS_NAME_TO_HANDLE_AT)
+	if (mynamespace != ns) {
+		char *realname;
+		snprintf(pathname, sizeof(pathname)-1, "/proc/%d/fd/%s", pid, fd);
+		realname = expandpath(pathname);
+		if (realname) {				/* Use our namespace for mount ID <sigh> */
+			info->mnt_id = get_mountid(realname);
+			mnt_id++;
+		}
+	}
+# endif
+#endif
+	if (!flags || !mnt_id) {
+		struct stat lst;
+
+		snprintf(pathname, sizeof(pathname)-1, "/proc/%d/fd/%s", pid, fd);
+		if (!flags && lstat(pathname, &lst) == 0) {
+			if (lst.st_mode & S_IWUSR)
+				info->flags |= O_WRONLY;
+			ret++;
+		}
+
+		if (!mnt_id) {
+			char *realname = expandpath(pathname);
+			if (realname) {
+				mntinfo_t *mountinfo;
+				if (find_mountpoint(realname, &mountinfo) == 0) {
+					info->mnt_id = mountinfo->id;
+					ret++;
+				}
+			}
+		}
+	}
+	return ret == 2 ? 0 : -1;
+}
 
-	/* Sort mount points accordingly to the reverse mount order */
-	initial(&sort);
-	for (mid = 1; mid <= max; mid++) {
-		list_t *ptr, *tmp;
-		list_for_each_safe(ptr, tmp, &mntinfo) {
+#if defined(HAS_NAME_TO_HANDLE_AT)
+static ino_t
+get_namespace(const pid_t pid)
+{
+	char pathname[512];
+	struct stat st;
+	snprintf(pathname, sizeof(pathname)-1, "/proc/%d/ns/mnt", pid);
+	if (stat(pathname, &st) == 0)
+		return st.st_ino;
+	return -1;
+}
+
+static int
+get_mountid(const char *path)
+{
+	union fh_u {
+	    struct file_handle handle;
+	    char buffer[sizeof(struct file_handle) + MAX_HANDLE_SZ];
+	} fh = { .handle.handle_bytes = MAX_HANDLE_SZ };
+	int mnt_id = -1;
+
+	errno = 0;
+	if (name_to_handle_at(0, path, &fh.handle, &mnt_id, 0) == -1) {
+		const static char delimiters[] = ": \t\n";
+		char fdinfo[PATH_MAX+1];
+		char line[BUFSIZ];
+		FILE* fp;
+		int fd;
+
+		if (errno != EOPNOTSUPP)
+			return -1;
+		fd = open(path, O_PATH);
+		if (fd < 0)
+			return -1;
+
+		snprintf(fdinfo, PATH_MAX, "/proc/self/fdinfo/%d", fd);
+		if ((fp = fopen(fdinfo, "r")) == NULL) {
+			close(fd);
+			return -1;
+		}
+		while (fgets(line, BUFSIZ, fp)) {
+			char *xp, *vp, *ep;
+			unsigned long ul;
+			xp = strtok(&line[0], delimiters);
+			if (!xp || *xp == 0)
+				continue;
+			vp = strtok(NULL, delimiters);
+			if (!vp || *vp == 0)
+				continue;
+			if (strcmp(xp, "mnt_id") == 0 && (ul = strtoul(vp, &ep, 0)) != ULONG_MAX && ep && *ep == 0) {
+				mnt_id = (int)ul;
+				break;
+			}
+		}
+		fclose(fp);
+		close(fd);
+	}
+	return mnt_id;
+}
+#endif
+
+static int
+find_mountpoint(const char *path, mntinfo_t **mountinfo)
+{
+	char *use, *end;
+	ssize_t nlen;
+#if defined(HAS_NAME_TO_HANDLE_AT)
+	int mnt_id = get_mountid(path);
+#endif
+	int ret = -1;
+
+	*mountinfo = NULL;
+
+#if defined(HAS_NAME_TO_HANDLE_AT)
+	if (mnt_id >= 0) {
+		list_t *ptr;
+
+		errno = ENOENT;
+		list_for_each(ptr, &mntinfo) {
 			mntinfo_t *mnt = list_entry(ptr, mntinfo_t);
-			if (mid != mnt->id)
+
+			if (mnt_id != mnt->id)
 				continue;
-			move_head(ptr, &sort);
+
+			ret = 0;
+			errno = 0;
+			*mountinfo = mnt;
 			break;
 		}
-		list_for_each_safe(ptr, tmp, &mntinfo) {
+		if (*mountinfo)
+			goto out;
+
+		if (strlen(path) == 1 && path[0] == '/') {
+			struct stat st;
+
+			/* could be a chroot or a container */
+
+			if (stat(path, &st) != 0) {
+				if (errno != EACCES) {
+					fprintf(stderr, _("Cannot stat %s: %s\n"),
+						path, strerror(errno));
+					exit(1);
+				}
+				st.st_dev = (dev_t)-1;
+			}
+			ret = 0;
+			errno = 0;
+			*mountinfo = add_mntinfo(path, "unkown", mnt_id, -1, st.st_dev);
+			goto out;
+		}
+	}
+#endif
+	use = strdup(path);
+	if (!use)
+		goto out;
+
+	nlen = strlen(use);
+	end = use+nlen;
+	errno = ENOENT;
+	do {
+		list_t *ptr;
+		if (*end == '/') {
+			if (end == use)	{	/* root file system */
+				end++;
+				if (nlen == 1) {
+					struct stat st;
+
+					/* could be a chroot or a container */
+
+					if (stat(use, &st) != 0) {
+						if (errno != EACCES) {
+							fprintf(stderr, _("Cannot stat %s: %s\n"),
+								use, strerror(errno));
+							exit(1);
+						}
+						st.st_dev = (dev_t)-1;
+					}
+					ret = 0;
+					errno = 0;
+					*mountinfo = add_mntinfo(use, "unkown", 0, -1, st.st_dev);
+					break;
+				}
+			}
+		}
+		*end = '\0';
+		nlen = end-use;
+		list_for_each(ptr, &mntinfo) {
 			mntinfo_t *mnt = list_entry(ptr, mntinfo_t);
-			if (mid != mnt->parid)
+
+			if (nlen != mnt->nlen)
 				continue;
-			move_head(ptr, &sort);
+
+			if (strcmp(use, mnt->mpoint))
+				continue;
+
+			ret = 0;
+			errno = 0;
+			*mountinfo = mnt;
+			break;
 		}
-	}
-	if (!list_empty(&mntinfo)) {
-#ifdef EBADE
-		errno = EBADE;
-#else
-		errno = ENOENT;
-#endif				/* EBADE */
-	}
-	join(&sort, &mntinfo);
+
+	} while (!*mountinfo && (end = strrchr(use, '/')));
+	free(use);
+out:
+	return ret;
 }
 
+#if defined(WITH_MOUNTINFO_LIST)
 /*
  * Determine device of links below /proc/
  */
@@ -2126,8 +2560,7 @@ static int mntstat(const char *path, struct stat *buf)
 {
 	char name[PATH_MAX + 1];
 	const char *use;
-	ssize_t nlen;
-	list_t *ptr;
+	mntinfo_t *mnt;
 
 	if ((use = realpath(path, name)) == NULL || *use != '/')
 	{
@@ -2139,27 +2572,26 @@ static int mntstat(const char *path, struct stat *buf)
 		errno = 0;
 		return stat(path, buf);
 	}
+	if (strncmp("/dev/", use, 5) == 0) {
+		/*
+		 * Could be a special file (socket, pipe, inotify)
+		 */
+		errno = 0;
+		return stat(path, buf);
+	}
 
-	nlen = strlen(use);
-	list_for_each(ptr, &mntinfo) {
-		mntinfo_t *mnt = list_entry(ptr, mntinfo_t);
-		if (nlen < mnt->nlen)
-			continue;
-		if (mnt->nlen == 1) {	/* root fs is the last entry */
-			buf->st_dev = mnt->dev;
-			buf->st_ino = 0;
-			return 0;
-		}
-		if (use[mnt->nlen] != '\0' && use[mnt->nlen] != '/')
-			continue;
-		if (strncmp(use, mnt->mpoint, mnt->nlen) == 0) {
-			buf->st_dev = mnt->dev;
-			buf->st_ino = 0;
-			return 0;
-		}
+	if (find_mountpoint(use, &mnt) < 0) {
+		errno = ENOENT;
+		return -1;
 	}
-	errno = ENOENT;
-	return -1;
+
+	if (mnt->isremote) {
+		buf->st_dev = mnt->dev;
+		buf->st_ino = mnt->id;	    /* inode substitute */
+		return 0;		    /* found on NFS */
+	}
+
+	return stat(path, buf);
 }
 #endif				/* WITH_MOUNTINFO_LIST */
 
diff --git src/fuser.h src/fuser.h
index 4500ec5..16abc55 100644
--- src/fuser.h
+++ src/fuser.h
@@ -37,10 +37,16 @@ struct procs {
 #define PTYPE_KNFSD 2
 #define PTYPE_SWAP 3
 
+struct fdinfo {
+	mode_t flags;
+	int mnt_id;
+};
+
 struct names {
 	char *filename;
 	unsigned char name_space;
 	struct stat st;
+	int mnt_id;
 	struct procs *matched_procs;
 	struct names *next;
 };
@@ -65,12 +71,21 @@ struct inode_list {
 	struct names *name;
 	dev_t	device;
 	ino_t	inode;
+	int	mnt_id;
 	struct inode_list *next;
 };
 
+struct subvol {
+	dev_t	device;
+	int	mnt_id;
+	struct subvol *next;
+};
+
 struct device_list {
 	struct names *name;
+	struct subvol *vol;
 	dev_t	device;
+	int	mnt_id;
 	struct device_list *next;
 };
 
@@ -79,6 +94,7 @@ struct unixsocket_list {
 	ino_t	inode;
 	ino_t	net_inode;
 	dev_t	dev;
+	int	mnt_id;
 	struct unixsocket_list *next;
 };
 
@@ -87,18 +103,16 @@ struct mount_list {
 	struct mount_list *next;
 };
 
-#if defined (__GNUC__) && defined(WITH_MOUNTINFO_LIST)
-# include "lists.h"
+#include "lists.h"
 typedef struct mntinfo_s {
     list_t   this;
     int id, parid;
+    char isremote;
     dev_t     dev;
     size_t   nlen;
+    dev_t     vol;
     char  *mpoint;
 } mntinfo_t;
-#else
-# undef WITH_MOUNTINFO_LIST
-#endif
 
 #define NAMESPACE_FILE 0
 #define NAMESPACE_TCP 1
@@ -109,5 +123,6 @@ typedef struct mntinfo_s {
 #endif /* PATH_MAX */
 
 #define KNFSD_EXPORTS "/proc/fs/nfs/exports"
+#define PROC_MOUNTINFO "/proc/self/mountinfo"
 #define PROC_MOUNTS "/proc/mounts"
 #define PROC_SWAPS "/proc/swaps"
diff --git src/timeout.c src/timeout.c
index ca4a7cd..a3c7055 100644
--- src/timeout.c
+++ src/timeout.c
@@ -67,9 +67,6 @@
 # ifndef  constructor
 #  define constructor		__constructor__
 # endif
-# ifndef  packed
-#  define packed		__packed__
-# endif
 # ifndef  inline
 #  define inline		__inline__
 # endif
@@ -107,7 +104,7 @@ typedef struct _handle {
 	stat_t function;
 	size_t len;
 	char path[0];
-} attribute((packed)) handle_t;
+} handle_t;
 
 /*
  * Using a forked process for doing e.g. stat(2) system call as this
-- 
2.26.2

Places

File 0001-Use-mountinfo-to-be-able-to-use-the-mount-identity.patch of Package psmisc (Revision ce46f845016955ada4da1af181138587)

Places