]> git.ipfire.org Git - thirdparty/mdadm.git/commitdiff
Manage: Block unsafe member failing
authorMateusz Kusiak <mateusz.kusiak@intel.com>
Thu, 18 Aug 2022 09:47:21 +0000 (11:47 +0200)
committerJes Sorensen <jsorensen@fb.com>
Thu, 8 Sep 2022 16:52:25 +0000 (12:52 -0400)
Kernel may or may not block mdadm from removing member device if it
will cause arrays failed state. It depends on raid personality
implementation in kernel.
Add verification on requested removal path (#mdadm --set-faulty
command).

Signed-off-by: Mateusz Kusiak <mateusz.kusiak@intel.com>
Signed-off-by: Jes Sorensen <jsorensen@fb.com>
Manage.c

index a142f8bd8944459612296a5049b461c0765ac412..b1d0e630b70707e5739413fd51af45fa6b53a869 100644 (file)
--- a/Manage.c
+++ b/Manage.c
@@ -1285,6 +1285,50 @@ int Manage_with(struct supertype *tst, int fd, struct mddev_dev *dv,
        return -1;
 }
 
+/**
+ * is_remove_safe() - Check if remove is safe.
+ * @array: Array info.
+ * @fd: Array file descriptor.
+ * @devname: Name of device to remove.
+ * @verbose: Verbose.
+ *
+ * The function determines if array will be operational
+ * after removing &devname.
+ *
+ * Return: True if array will be operational, false otherwise.
+ */
+bool is_remove_safe(mdu_array_info_t *array, const int fd, char *devname, const int verbose)
+{
+       dev_t devid = devnm2devid(devname + 5);
+       struct mdinfo *mdi = sysfs_read(fd, NULL, GET_DEVS | GET_DISKS | GET_STATE);
+
+       if (!mdi) {
+               if (verbose)
+                       pr_err("Failed to read sysfs attributes for %s\n", devname);
+               return false;
+       }
+
+       char *avail = xcalloc(array->raid_disks, sizeof(char));
+
+       for (mdi = mdi->devs; mdi; mdi = mdi->next) {
+               if (mdi->disk.raid_disk < 0)
+                       continue;
+               if (!(mdi->disk.state & (1 << MD_DISK_SYNC)))
+                       continue;
+               if (makedev(mdi->disk.major, mdi->disk.minor) == devid)
+                       continue;
+               avail[mdi->disk.raid_disk] = 1;
+       }
+       sysfs_free(mdi);
+
+       bool is_enough = enough(array->level, array->raid_disks,
+                               array->layout, (array->state & 1),
+                               avail);
+
+       free(avail);
+       return is_enough;
+}
+
 int Manage_subdevs(char *devname, int fd,
                   struct mddev_dev *devlist, int verbose, int test,
                   char *update, int force)
@@ -1598,7 +1642,14 @@ int Manage_subdevs(char *devname, int fd,
                        break;
 
                case 'f': /* set faulty */
-                       /* FIXME check current member */
+                       if (!is_remove_safe(&array, fd, dv->devname, verbose)) {
+                               pr_err("Cannot remove %s from %s, array will be failed.\n",
+                                      dv->devname, devname);
+                               if (sysfd >= 0)
+                                       close(sysfd);
+                               goto abort;
+                       }
+
                        if ((sysfd >= 0 && write(sysfd, "faulty", 6) != 6) ||
                            (sysfd < 0 && ioctl(fd, SET_DISK_FAULTY,
                                                rdev))) {