Fix readding of a readwrite drive into a writemostly array
PATCH 1/1 Fix readding of a readwrite drive into a writemostly array
If you create a two drive raid1 array with one device writemostly, then fail the readwrite drive, when you add a new device, it will get the writemostly bit copied out of the remaining device's superblock into it's own. You can then remove the new drive and readd it as readwrite, which will work for the readd, but it leaves the stale WriteMostly1 bit in devflags resulting in the device going back to writemostly on the next assembly.
The fix is to make sure that A) when we readd a device and we might have filled the st->sb info from a running device instead of the device being readded, then clear/set the WriteMostly1 bit in the super1 struct in addition to setting the disk state (ditto for super0, but slightly different mechanism) and B) when adding a clean device to an array (when we most certainly did copy the superblock info from an existing device), then clear any writemostly bits.
Manage.c renamed from Manage.c.writemostly +15 -2
@@ -741,11 +741,24 @@ int Manage_subdevs(char *devname, int fdremove_partitions (tfd);close (tfd); tfd =-1 ; -if (update) { +if (update || dv->writemostly >0 ) {int rv =-1 ; tfd =dev_open (dv->devname,O_RDWR ); +if (tfd <0 ) { +fprintf (stderr, Name": failed to open %s for" +" superblock update during re-add\n" , dv->devname); +return 1 ; + } -if (tfd >=0 ) +if (dv->writemostly ==1 ) + rv = st->ss->update_super ( + st,NULL ,"writemostly" , + devname, verbose,0 ,NULL ); +if (dv->writemostly ==2 ) + rv = st->ss->update_super ( + st,NULL ,"readwrite" , + devname, verbose,0 ,NULL ); +if (update) rv = st->ss->update_super ( st,NULL , update, devname, verbose,0 ,NULL );
mdadm.h renamed from mdadm.h.writemostly +2 -0
@@ -646,6 +646,8 @@ extern struct superswitch {* linear-grow-new - add a new device to a linear array, but don't * change the size: so superblock still matches * linear-grow-update - now change the size of the array. +* writemostly - set the WriteMostly1 bit in the superblock devflags +* readwrite - clear the WriteMostly1 bit in the superblock devflags */ int (*update_super)(struct supertype *st, struct mdinfo *info, char *update,
super0.c renamed from super0.c.writemostly +6 -0
@@ -570,6 +570,10 @@ static int update_super0(struct supertyp sb->state &= ~(1 <<MD_SB_BITMAP_PRESENT ); }else if (strcmp (update,"_reshape_progress" )==0 ) sb->reshape_position = info->reshape_progress; +else if (strcmp (update,"writemostly" )==0 ) + sb->state |= (1 <<MD_DISK_WRITEMOSTLY ); +else if (strcmp (update,"readwrite" )==0 ) + sb->state &= ~(1 <<MD_DISK_WRITEMOSTLY );else rv =-1 ;
@@ -688,6 +692,8 @@ static int add_to_super0(struct supertyp dk->minor = dinfo->minor; dk->raid_disk = dinfo->raid_disk; dk->state = dinfo->state; +/* In case our source disk was writemostly, don't copy that bit */ + dk->state &= ~(1 <<MD_DISK_WRITEMOSTLY ); sb->this_disk = sb->disks[dinfo->number]; sb->sb_csum =calc_sb0_csum (sb);
super1.c renamed from super1.c.writemostly +5 -0
@@ -803,6 +803,10 @@ static int update_super1(struct supertyp __le64_to_cpu(sb->data_size)); }else if (strcmp (update,"_reshape_progress" )==0 ) sb->reshape_position =__cpu_to_le64 (info->reshape_progress); +else if (strcmp (update,"writemostly" )==0 ) + sb->devflags |= WriteMostly1; +else if (strcmp (update,"readwrite" )==0 ) + sb->devflags &= ~WriteMostly1;else rv =-1 ;
@@ -923,6 +927,7 @@ static int add_to_super1(struct supertyp sb->max_dev =__cpu_to_le32 (dk->number+1 ); sb->dev_number =__cpu_to_le32 (dk->number); + sb->devflags =0 ;/* don't copy another disks flags */ sb->sb_csum =calc_sb_1_csum (sb); dip = (struct devinfo **)&st->info;
Quoted 16 lines · On Wed, 27 Jul 2011 14:34:37 -0400 Doug Ledford <[email protected]> wrote:
If you create a two drive raid1 array with one device writemostly, then fail the readwrite drive, when you add a new device, it will get the writemostly bit copied out of the remaining device's superblock into it's own. You can then remove the new drive and readd it as readwrite, which will work for the readd, but it leaves the stale WriteMostly1 bit in devflags resulting in the device going back to writemostly on the next assembly.
The fix is to make sure that A) when we readd a device and we might have filled the
st->sbinfo from a running device instead of the device being readded, then clear/set the WriteMostly1 bit in the super1 struct in addition to setting the disk state (ditto for super0, but slightly different mechanism) and B) when adding a clean device to an array (when we most certainly did copy the superblock info from an existing device), then clear any writemostly bits.
Quoted 1 line
Signed-off-by Doug Ledford <[email protected]>
Applied - thanks.
This function really needs to be refactored though - some of that indenting is WAY too deep!!
Thanks,
NeilBrown
Manage.c renamed from Manage.c.writemostly +15 -2
@@ -741,11 +741,24 @@ int Manage_subdevs(char *devname, int fdremove_partitions (tfd);close (tfd); tfd =-1 ; -if (update) { +if (update || dv->writemostly >0 ) {int rv =-1 ; tfd =dev_open (dv->devname,O_RDWR ); +if (tfd <0 ) { +fprintf (stderr, Name": failed to open %s for" +" superblock update during re-add\n" , dv->devname); +return 1 ; + } -if (tfd >=0 ) +if (dv->writemostly ==1 ) + rv = st->ss->update_super ( + st,NULL ,"writemostly" , + devname, verbose,0 ,NULL ); +if (dv->writemostly ==2 ) + rv = st->ss->update_super ( + st,NULL ,"readwrite" , + devname, verbose,0 ,NULL ); +if (update) rv = st->ss->update_super ( st,NULL , update, devname, verbose,0 ,NULL );
mdadm.h renamed from mdadm.h.writemostly +2 -0
@@ -646,6 +646,8 @@ extern struct superswitch {* linear-grow-new - add a new device to a linear array, but don't * change the size: so superblock still matches * linear-grow-update - now change the size of the array. +* writemostly - set the WriteMostly1 bit in the superblock devflags +* readwrite - clear the WriteMostly1 bit in the superblock devflags */ int (*update_super)(struct supertype *st, struct mdinfo *info, char *update,
super0.c renamed from super0.c.writemostly +6 -0
@@ -570,6 +570,10 @@ static int update_super0(struct supertyp sb->state &= ~(1 <<MD_SB_BITMAP_PRESENT ); }else if (strcmp (update,"_reshape_progress" )==0 ) sb->reshape_position = info->reshape_progress; +else if (strcmp (update,"writemostly" )==0 ) + sb->state |= (1 <<MD_DISK_WRITEMOSTLY ); +else if (strcmp (update,"readwrite" )==0 ) + sb->state &= ~(1 <<MD_DISK_WRITEMOSTLY );else rv =-1 ;
@@ -688,6 +692,8 @@ static int add_to_super0(struct supertyp dk->minor = dinfo->minor; dk->raid_disk = dinfo->raid_disk; dk->state = dinfo->state; +/* In case our source disk was writemostly, don't copy that bit */ + dk->state &= ~(1 <<MD_DISK_WRITEMOSTLY ); sb->this_disk = sb->disks[dinfo->number]; sb->sb_csum =calc_sb0_csum (sb);
super1.c renamed from super1.c.writemostly +5 -0
@@ -803,6 +803,10 @@ static int update_super1(struct supertyp __le64_to_cpu(sb->data_size)); }else if (strcmp (update,"_reshape_progress" )==0 ) sb->reshape_position =__cpu_to_le64 (info->reshape_progress); +else if (strcmp (update,"writemostly" )==0 ) + sb->devflags |= WriteMostly1; +else if (strcmp (update,"readwrite" )==0 ) + sb->devflags &= ~WriteMostly1;else rv =-1 ;
@@ -923,6 +927,7 @@ static int add_to_super1(struct supertyp sb->max_dev =__cpu_to_le32 (dk->number+1 ); sb->dev_number =__cpu_to_le32 (dk->number); + sb->devflags =0 ;/* don't copy another disks flags */ sb->sb_csum =calc_sb_1_csum (sb); dip = (struct devinfo **)&st->info;
I don't deny the need to refactor, but at the time I wrote it I was in "fix the problem" mode, not "clean up the code" mode ;-)
----- Original Message -----
Quoted 138 lines
Quoted 27 lines · On Wed, 27 Jul 2011 14:34:37 -0400 Doug Ledford <[email protected]> wrote:
If you create a two drive raid1 array with one device writemostly, then
fail the readwrite drive, when you add a new device, it will get the
writemostly bit copied out of the remaining device's superblock into
it's own. You can then remove the new drive and readd it as readwrite,
which will work for the readd, but it leaves the stale WriteMostly1 bit
in devflags resulting in the device going back to writemostly on the
next assembly.The fix is to make sure that A) when we readd a device and we might have
filled thest->sbinfo from a running device instead of the device being
readded, then clear/set the WriteMostly1 bit in the super1 struct in
addition to setting the disk state (ditto for super0, but slightly different mechanism) and B) when adding a clean device to an array (when
we most certainly did copy the superblock info from an existing device),
then clear any writemostly bits.Quoted 1 line
Signed-off-by Doug Ledford <[email protected]>Applied - thanks.
This function really needs to be refactored though - some of that indenting
is WAY too deep!!Thanks,
NeilBrowndiff -up mdadm-3.2.2/Manage.c.writemostly
mdadm-3.2.2/Manage.c
--- mdadm-3.2.2/Manage.c.writemostly 2011-06-13 22:50:01.000000000
-0400
+++mdadm-3.2.2/Manage.c2011-07-27 14:12:18.629889841 -0400Manage.c renamed from Manage.c.writemostly +15
-2@@ -741,11 +741,24 @@ int Manage_subdevs(char *devname, int fdremove_partitions (tfd);close (tfd); tfd =-1 ; -if (update) { +if (update || dv->writemostly >0 ) {int rv =-1 ; tfd =dev_open (dv->devname,O_RDWR ); +if (tfd <0 ) { +fprintf (stderr, Name": failed to open %s for" +" superblock update during re-add\n" , dv->devname); +return 1 ; + } -if (tfd >=0 ) +if (dv->writemostly ==1 ) + rv = st->ss->update_super ( + st,NULL ,"writemostly" , + devname, verbose,0 ,NULL ); +if (dv->writemostly ==2 ) + rv = st->ss->update_super ( + st,NULL ,"readwrite" , + devname, verbose,0 ,NULL ); +if (update) rv = st->ss->update_super ( st,NULL , update, devname, verbose,0 ,NULL );diff -up mdadm-3.2.2/mdadm.h.writemostly
mdadm-3.2.2/mdadm.h
--- mdadm-3.2.2/mdadm.h.writemostly 2011-07-27 14:12:28.800779575
-0400
+++mdadm-3.2.2/mdadm.h2011-07-27 14:04:34.669932148 -0400(quoted diff) +1
-0@@ -646,6 +646,8 @@ extern struct superswitch { * linear-grow-new - add a new device to a linear array, but don't * change the size: so superblock still matches * linear-grow-update - now change the size of the array. + * writemostly - set the WriteMostly1 bit in the superblockdevflags
(quoted diff) +1
-0+ * readwrite - clear the WriteMostly1 bit in the superblockdevflags
mdadm.h renamed from mdadm.h.writemostly +0
-0*/ int (*update_super)(struct supertype *st, struct mdinfo *info, char *update,diff -up mdadm-3.2.2/super0.c.writemostly
mdadm-3.2.2/super0.c
--- mdadm-3.2.2/super0.c.writemostly 2011-06-17 01:15:50.000000000
-0400
+++mdadm-3.2.2/super0.c2011-07-27 14:12:18.655889559 -0400super0.c renamed from super0.c.writemostly +6
-0@@ -570,6 +570,10 @@ static int update_super0(struct supertyp sb->state &= ~(1 <<MD_SB_BITMAP_PRESENT ); }else if (strcmp (update,"_reshape_progress" )==0 ) sb->reshape_position = info->reshape_progress; +else if (strcmp (update,"writemostly" )==0 ) + sb->state |= (1 <<MD_DISK_WRITEMOSTLY ); +else if (strcmp (update,"readwrite" )==0 ) + sb->state &= ~(1 <<MD_DISK_WRITEMOSTLY );else rv =-1 ;@@ -688,6 +692,8 @@ static int add_to_super0(struct supertyp dk->minor = dinfo->minor; dk->raid_disk = dinfo->raid_disk; dk->state = dinfo->state; +/* In case our source disk was writemostly, don't copy that bit */ + dk->state &= ~(1 <<MD_DISK_WRITEMOSTLY ); sb->this_disk = sb->disks[dinfo->number]; sb->sb_csum =calc_sb0_csum (sb);diff -up mdadm-3.2.2/super1.c.writemostly
mdadm-3.2.2/super1.c
--- mdadm-3.2.2/super1.c.writemostly 2011-06-17 01:15:50.000000000
-0400
+++mdadm-3.2.2/super1.c2011-07-27 14:12:18.656889548 -0400super1.c renamed from super1.c.writemostly +5
-0@@ -803,6 +803,10 @@ static int update_super1(struct supertyp __le64_to_cpu(sb->data_size)); }else if (strcmp (update,"_reshape_progress" )==0 ) sb->reshape_position =__cpu_to_le64 (info->reshape_progress); +else if (strcmp (update,"writemostly" )==0 ) + sb->devflags |= WriteMostly1; +else if (strcmp (update,"readwrite" )==0 ) + sb->devflags &= ~WriteMostly1;else rv =-1 ;@@ -923,6 +927,7 @@ static int add_to_super1(struct supertyp sb->max_dev =__cpu_to_le32 (dk->number+1 ); sb->dev_number =__cpu_to_le32 (dk->number); + sb->devflags =0 ;/* don't copy another disks flags */ sb->sb_csum =calc_sb_1_csum (sb); dip = (struct devinfo **)&st->info;