md/raid: only permit hot-add of compatible integrity profiles
authorDan Williams <dan.j.williams@intel.com>
Thu, 14 Jan 2016 00:00:07 +0000 (16:00 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 17 Feb 2016 20:30:57 +0000 (12:30 -0800)
commit 1501efadc524a0c99494b576923091589a52d2a4 upstream.

It is not safe for an integrity profile to be changed while i/o is
in-flight in the queue.  Prevent adding new disks or otherwise online
spares to an array if the device has an incompatible integrity profile.

The original change to the blk_integrity_unregister implementation in
md, commmit c7bfced9a671 "md: suspend i/o during runtime
blk_integrity_unregister" introduced an immediate hang regression.

This policy of disallowing changes the integrity profile once one has
been established is shared with DM.

Here is an abbreviated log from a test run that:
1/ Creates a degraded raid1 with an integrity-enabled device (pmem0s) [   59.076127]
2/ Tries to add an integrity-disabled device (pmem1m) [   90.489209]
3/ Retries with an integrity-enabled device (pmem1s) [  205.671277]

[   59.076127] md/raid1:md0: active with 1 out of 2 mirrors
[   59.078302] md: data integrity enabled on md0
[..]
[   90.489209] md0: incompatible integrity profile for pmem1m
[..]
[  205.671277] md: super_written gets error=-5
[  205.677386] md/raid1:md0: Disk failure on pmem1m, disabling device.
[  205.677386] md/raid1:md0: Operation continuing on 1 devices.
[  205.683037] RAID1 conf printout:
[  205.684699]  --- wd:1 rd:2
[  205.685972]  disk 0, wo:0, o:1, dev:pmem0s
[  205.687562]  disk 1, wo:1, o:1, dev:pmem1s
[  205.691717] md: recovery of RAID array md0

Fixes: c7bfced9a671 ("md: suspend i/o during runtime blk_integrity_unregister")
Cc: Mike Snitzer <snitzer@redhat.com>
Reported-by: NeilBrown <neilb@suse.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/md/md.c
drivers/md/md.h
drivers/md/multipath.c
drivers/md/raid1.c
drivers/md/raid10.c

index 61aacab424cf1860dbb5651f5a4fc6c9f7188027..b1e1f6b9578246c3a834a4c1e75ef8f6972045e6 100644 (file)
@@ -2017,28 +2017,32 @@ int md_integrity_register(struct mddev *mddev)
 }
 EXPORT_SYMBOL(md_integrity_register);
 
-/* Disable data integrity if non-capable/non-matching disk is being added */
-void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
+/*
+ * Attempt to add an rdev, but only if it is consistent with the current
+ * integrity profile
+ */
+int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
 {
        struct blk_integrity *bi_rdev;
        struct blk_integrity *bi_mddev;
+       char name[BDEVNAME_SIZE];
 
        if (!mddev->gendisk)
-               return;
+               return 0;
 
        bi_rdev = bdev_get_integrity(rdev->bdev);
        bi_mddev = blk_get_integrity(mddev->gendisk);
 
        if (!bi_mddev) /* nothing to do */
-               return;
-       if (rdev->raid_disk < 0) /* skip spares */
-               return;
-       if (bi_rdev && blk_integrity_compare(mddev->gendisk,
-                                            rdev->bdev->bd_disk) >= 0)
-               return;
-       WARN_ON_ONCE(!mddev->suspended);
-       printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
-       blk_integrity_unregister(mddev->gendisk);
+               return 0;
+
+       if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
+               printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
+                               mdname(mddev), bdevname(rdev->bdev, name));
+               return -ENXIO;
+       }
+
+       return 0;
 }
 EXPORT_SYMBOL(md_integrity_add_rdev);
 
index ca0b643fe3c18070b417a66daedb571457b440ca..dfa57b41541b1bf2a48e3e710d09022856fb6f03 100644 (file)
@@ -657,7 +657,7 @@ extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
 extern int md_check_no_bitmap(struct mddev *mddev);
 extern int md_integrity_register(struct mddev *mddev);
-extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
+extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
 
 extern void mddev_init(struct mddev *mddev);
index 7331a80d89f1987a42d22a2ae7510210519e6bae..0a72ab6e6c204a6a1a56d7dff922753bb6294c76 100644 (file)
@@ -257,6 +257,9 @@ static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
                        disk_stack_limits(mddev->gendisk, rdev->bdev,
                                          rdev->data_offset << 9);
 
+                       err = md_integrity_add_rdev(rdev, mddev);
+                       if (err)
+                               break;
                        spin_lock_irq(&conf->device_lock);
                        mddev->degraded--;
                        rdev->raid_disk = path;
@@ -264,9 +267,6 @@ static int multipath_add_disk(struct mddev *mddev, struct md_rdev *rdev)
                        spin_unlock_irq(&conf->device_lock);
                        rcu_assign_pointer(p->rdev, rdev);
                        err = 0;
-                       mddev_suspend(mddev);
-                       md_integrity_add_rdev(rdev, mddev);
-                       mddev_resume(mddev);
                        break;
                }
 
index e2169ff6e0f00ca7b23d533ffa2ccee9c60ad822..c4b9134092260f22939027a0a82945b146a4ebf9 100644 (file)
@@ -1589,6 +1589,9 @@ static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
        if (mddev->recovery_disabled == conf->recovery_disabled)
                return -EBUSY;
 
+       if (md_integrity_add_rdev(rdev, mddev))
+               return -ENXIO;
+
        if (rdev->raid_disk >= 0)
                first = last = rdev->raid_disk;
 
@@ -1632,9 +1635,6 @@ static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
                        break;
                }
        }
-       mddev_suspend(mddev);
-       md_integrity_add_rdev(rdev, mddev);
-       mddev_resume(mddev);
        if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
                queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
        print_conf(conf);
index 84e597e1c4890cd911243c511d1a65e39232e98e..ce959b4ae4dfd9a09d33828f5706d3a6aadd7f71 100644 (file)
@@ -1698,6 +1698,9 @@ static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
        if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1))
                return -EINVAL;
 
+       if (md_integrity_add_rdev(rdev, mddev))
+               return -ENXIO;
+
        if (rdev->raid_disk >= 0)
                first = last = rdev->raid_disk;
 
@@ -1739,9 +1742,6 @@ static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev)
                rcu_assign_pointer(p->rdev, rdev);
                break;
        }
-       mddev_suspend(mddev);
-       md_integrity_add_rdev(rdev, mddev);
-       mddev_resume(mddev);
        if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
                queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);