nspawn: work around kernel bug with partition table probing on loopback devices

When we set up a loopback device with partition probing, the udev
"change" event about the configured device is first passed on to
userspace, only the the in-kernel partition prober is started. Since
partition probing fails with EBUSY when somebody has the device open,
the probing frequently fails since udev starts probing/opening the
device as soon as it gets the notification about it, and it might do so
earlier than the kernel probing.

This patch adds a (hopefully temporary) work-around for this, that
compares the number of probed partitions of the kernel with those of
blkid and synchronously asks for reprobing until the numebrs are in
sync.

This really deserves a proper kernel fix.

diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c
index 588a8ae8ac8c4ff230d704eb25a7ef0342a35184..d0e4ee785ce6740ee45e5f9ba129219aad808232 100644 (file)
--- a/src/nspawn/nspawn.c
+++ b/src/nspawn/nspawn.c
@@ -2653,63 +2653,6 @@ static int setup_image(char **device_path, int *loop_nr) {
         return r;
 }
 
-static int wait_for_block_device(struct udev *udev, dev_t devnum, struct udev_device **ret) {
-        _cleanup_udev_monitor_unref_ struct udev_monitor *monitor = NULL;
-        int r;
-
-        assert(udev);
-        assert(ret);
-
-        for (;;) {
-                _cleanup_udev_device_unref_ struct udev_device *d = NULL;
-                struct pollfd pfd = {
-                        .events = POLLIN
-                };
-
-                d = udev_device_new_from_devnum(udev, 'b', devnum);
-                if (!d)
-                        return log_oom();
-
-                r = udev_device_get_is_initialized(d);
-                if (r < 0)
-                        return log_error_errno(r, "Failed to check if device is initialized: %m");
-                if (r > 0) {
-                        *ret = d;
-                        d = NULL;
-                        return 0;
-                }
-                d = udev_device_unref(d);
-
-                if (!monitor) {
-                        monitor = udev_monitor_new_from_netlink(udev, "udev");
-                        if (!monitor)
-                                return log_oom();
-
-                        r = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block", NULL);
-                        if (r < 0)
-                                return log_error_errno(r, "Failed to add block match: %m");
-
-                        r = udev_monitor_enable_receiving(monitor);
-                        if (r < 0)
-                                return log_error_errno(r, "Failed to turn on monitor: %m");
-
-                        continue;
-                }
-
-                pfd.fd = udev_monitor_get_fd(monitor);
-                if (pfd.fd < 0)
-                        return log_error_errno(r, "Failed to get udev monitor fd: %m");
-
-                r = poll(&pfd, 1, -1);
-                if (r < 0)
-                        return log_error_errno(errno, "Failed to wait for device initialization: %m");
-
-                d = udev_monitor_receive_device(monitor);
-        }
-
-        return 0;
-}
-
 #define PARTITION_TABLE_BLURB \
         "Note that the disk image needs to either contain only a single MBR partition of\n" \
         "type 0x83 that is marked bootable, or a sinlge GPT partition of type" \
@@ -2739,11 +2682,12 @@ static int dissect_image(
         _cleanup_udev_unref_ struct udev *udev = NULL;
         struct udev_list_entry *first, *item;
         bool home_rw = true, root_rw = true, secondary_root_rw = true, srv_rw = true, generic_rw = true;
+        bool is_gpt, is_mbr, multiple_generic = false;
         const char *pttype = NULL;
         blkid_partlist pl;
         struct stat st;
+        unsigned i;
         int r;
-        bool is_gpt, is_mbr, multiple_generic = false;
 
         assert(fd >= 0);
         assert(root_device);
@@ -2812,21 +2756,81 @@ static int dissect_image(
         if (fstat(fd, &st) < 0)
                 return log_error_errno(errno, "Failed to stat block device: %m");
 
-        r = wait_for_block_device(udev, st.st_rdev, &d);
-        if (r < 0)
-                return r;
-
-        e = udev_enumerate_new(udev);
-        if (!e)
+        d = udev_device_new_from_devnum(udev, 'b', st.st_rdev);
+        if (!d)
                 return log_oom();
 
-        r = udev_enumerate_add_match_parent(e, d);
-        if (r < 0)
-                return log_oom();
+        for (i = 0;; i++) {
+                int n, m;
 
-        r = udev_enumerate_scan_devices(e);
-        if (r < 0)
-                return log_error_errno(r, "Failed to scan for partition devices of %s: %m", arg_image);
+                if (i >= 10) {
+                        log_error("Kernel partitions never appeared.");
+                        return -ENXIO;
+                }
+
+                e = udev_enumerate_new(udev);
+                if (!e)
+                        return log_oom();
+
+                r = udev_enumerate_add_match_parent(e, d);
+                if (r < 0)
+                        return log_oom();
+
+                r = udev_enumerate_scan_devices(e);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to scan for partition devices of %s: %m", arg_image);
+
+                /* Count the partitions enumerated by the kernel */
+                n = 0;
+                first = udev_enumerate_get_list_entry(e);
+                udev_list_entry_foreach(item, first)
+                        n++;
+
+                /* Count the partitions enumerated by blkid */
+                m = blkid_partlist_numof_partitions(pl);
+                if (n == m + 1)
+                        break;
+                if (n > m + 1) {
+                        log_error("blkid and kernel partition list do not match.");
+                        return -EIO;
+                }
+                if (n < m + 1) {
+                        unsigned j;
+
+                        /* The kernel has probed fewer partitions than
+                         * blkid? Maybe the kernel prober is still
+                         * running or it got EBUSY because udev
+                         * already opened the device. Let's reprobe
+                         * the device, which is a synchronous call
+                         * that waits until probing is complete. */
+
+                        for (j = 0; j < 20; j++) {
+
+                                r = ioctl(fd, BLKRRPART, 0);
+                                if (r < 0)
+                                        r = -errno;
+                                if (r >= 0 || r != -EBUSY)
+                                        break;
+
+                                /* If something else has the device
+                                 * open, such as an udev rule, the
+                                 * ioctl will return EBUSY. Since
+                                 * there's no way to wait until it
+                                 * isn't busy anymore, let's just wait
+                                 * a bit, and try again.
+                                 *
+                                 * This is really something they
+                                 * should fix in the kernel! */
+
+                                usleep(50 * USEC_PER_MSEC);
+                        }
+
+                        if (r < 0)
+                                return log_error_errno(r, "Failed to reread partition table: %m");
+                }
+
+                e = udev_enumerate_unref(e);
+        }
 
         first = udev_enumerate_get_list_entry(e);
         udev_list_entry_foreach(item, first) {