From 8453e704305b92f043e436d6f90a0c5f068b09eb Mon Sep 17 00:00:00 2001
From: NeilBrown <neilb@suse.de>
Date: Mon, 22 Nov 2010 19:35:25 +1100
Subject: [PATCH] Manage:  be more careful about --add attempts.

If an --add is requested and a re-add looks promising but fails or
cannot possibly succeed, then don't try the add.  This avoids
inadvertently turning devices into spares when an array is failed but
the devices seem to actually work.

Signed-off-by: NeilBrown <neilb@suse.de>
---
 Manage.c       | 36 +++++++++++++++++++++++++++---------
 mdadm.h        |  1 +
 tests/01r1fail |  1 +
 util.c         | 30 ++++++++++++++++++++++++++++++
 4 files changed, 59 insertions(+), 9 deletions(-)

diff --git a/Manage.c b/Manage.c
index 6e9d4a03..acfec750 100644
--- a/Manage.c
+++ b/Manage.c
@@ -383,6 +383,7 @@ int Manage_subdevs(char *devname, int fd,
 		char *dnprintable = dv->devname;
 		char *add_dev = dv->devname;
 		int err;
+		int re_add_failed = 0;
 
 		next = dv->next;
 		jnext = 0;
@@ -662,14 +663,20 @@ int Manage_subdevs(char *devname, int fd,
 				    get_linux_version() <= 2006018)
 					;
 				else if (st->sb) {
+					struct mdinfo mdi;
+					st->ss->getinfo_super(st, &mdi);
 					st->ss->uuid_from_super(st, ouuid);
-					if (memcmp(duuid, ouuid, sizeof(ouuid))==0) {
-						/* looks close enough for now.  Kernel
-						 * will worry about whether a bitmap
-						 * based reconstruction is possible.
+					if ((mdi.disk.state & (1<<MD_DISK_ACTIVE)) &&
+					    !(mdi.disk.state & (1<<MD_DISK_FAULTY)) &&
+					    memcmp(duuid, ouuid, sizeof(ouuid))==0) {
+						/* look like it is worth a try.  Need to
+						 * make sure kernel will accept it though.
 						 */
-						struct mdinfo mdi;
-						st->ss->getinfo_super(st, &mdi);
+						disc.number = mdi.disk.number;
+						if (ioctl(fd, GET_DISK_INFO, &disc) != 0
+						    || disc.major != 0 || disc.minor != 0
+						    || !enough_fd(fd))
+							goto skip_re_add;
 						disc.major = major(stb.st_rdev);
 						disc.minor = minor(stb.st_rdev);
 						disc.number = mdi.disk.number;
@@ -684,8 +691,7 @@ int Manage_subdevs(char *devname, int fd,
 						tfd = -1;
 						/* don't even try if disk is marked as faulty */
 						errno = 0;
-						if ((disc.state & 1) == 0 &&
-						    ioctl(fd, ADD_NEW_DISK, &disc) == 0) {
+						if (ioctl(fd, ADD_NEW_DISK, &disc) == 0) {
 							if (verbose >= 0)
 								fprintf(stderr, Name ": re-added %s\n", add_dev);
 							count++;
@@ -698,7 +704,8 @@ int Manage_subdevs(char *devname, int fd,
 								continue;
 							return 1;
 						}
-						/* fall back on normal-add */
+					skip_re_add:
+						re_add_failed = 1;
 					}
 				}
 				if (add_dev != dv->devname) {
@@ -720,6 +727,17 @@ int Manage_subdevs(char *devname, int fd,
 						dv->devname, devname);
 					return 1;
 				}
+				if (re_add_failed) {
+					fprintf(stderr, Name ": %s reports being an active member for %s, but a --re-add fails.\n",
+						dv->devname, devname);
+					fprintf(stderr, Name ": not performing --add as that would convert %s in to a spare.\n",
+						dv->devname);
+					fprintf(stderr, Name ": To make this a spare, use \"mdadm --zero-superblock %s\" first.\n",	
+						dv->devname);
+					if (tfd >= 0)
+						close(tfd);
+					return 1;
+				}
 			} else {
 				/* non-persistent. Must ensure that new drive
 				 * is at least array.size big.
diff --git a/mdadm.h b/mdadm.h
index 9ad99f04..3e229103 100644
--- a/mdadm.h
+++ b/mdadm.h
@@ -986,6 +986,7 @@ extern char *fname_from_uuid(struct supertype *st,
 extern unsigned long calc_csum(void *super, int bytes);
 extern int enough(int level, int raid_disks, int layout, int clean,
 		   char *avail, int avail_disks);
+extern int enough_fd(int fd);
 extern int ask(char *mesg);
 extern unsigned long long get_component_size(int fd);
 extern void remove_partitions(int fd);
diff --git a/tests/01r1fail b/tests/01r1fail
index c3786630..9f556321 100644
--- a/tests/01r1fail
+++ b/tests/01r1fail
@@ -20,6 +20,7 @@ mdadm $md0 --remove $dev2 $dev1
 check nosync
 check state UUU_
 
+mdadm --zero-superblock $dev2
 mdadm $md0 -a $dev2 
 check recovery
 check wait
diff --git a/util.c b/util.c
index 0cb251c3..c9976218 100644
--- a/util.c
+++ b/util.c
@@ -320,6 +320,36 @@ int enough(int level, int raid_disks, int layout, int clean,
 	}
 }
 
+int enough_fd(int fd)
+{
+	struct mdu_array_info_s array;
+	struct mdu_disk_info_s disk;
+	int avail_disks = 0;
+	int i;
+	char *avail;
+
+	if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 ||
+	    array.raid_disks <= 0)
+		return 0;
+	avail = calloc(array.raid_disks, 1);
+	for (i=0; i<array.raid_disks + array.nr_disks; i++) {
+		disk.number = i;
+		if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
+			continue;
+		if (! (disk.state & (1<<MD_DISK_SYNC)))
+			continue;
+		if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks)
+			continue;
+		avail_disks++;
+		avail[disk.raid_disk] = 1;
+	}
+	/* This is used on an active array, so assume it is clean */
+	return enough(array.level, array.raid_disks, array.layout,
+		      1,
+		      avail, avail_disks);
+}
+
+
 const int uuid_match_any[4] = { ~0, ~0, ~0, ~0 };
 int same_uuid(int a[4], int b[4], int swapuuid)
 {
-- 
2.39.2