vgcfgbackup: do not produce corrupted metadata if PVs are missing
PATCH 1/1 vgcfgbackup: do not produce corrupted metadata if PVs are missing
Use PV UUID in hash for device name when exporting metadata.
Currently code uses pv_dev_name() for hash when getting internal "pvX" name.
This produce corrupted metadata if PVs are missing, pv->dev is NULL and all these missing devices returns one name (using "unknown device" for all missing devices as hash key).
--- lib/format_text/export.c | 29 +++++++++++++++++++++-------- test/t-vgcfgbackup-usage.sh | 22 ++++++++++++++++++++-- 2 files changed, 41 insertions(+), 10 deletions(-)
lib/format_text/export.c +21 -8
@@ -378,10 +378,19 @@ static int _print_vg(struct formatter *f, struct volume_group *vg)* Get the pv%d name from the formatters hash * table. */ +static const char *_get_pv_name_from_uuid (struct formatter *f,char *uuid) +{ +return dm_hash_lookup (f->pv_names, uuid); +} +static const char *_get_pv_name (struct formatter *f,struct physical_volume *pv) { -return (pv) ? (const char *) -dm_hash_lookup (f->pv_names,pv_dev_name (pv)) :"Missing" ; +char uuid[40 ]; + +if (!pv || !id_write_format (&pv->id, uuid,sizeof (uuid))) + return_NULL; + +return _get_pv_name_from_uuid (f, uuid); }static int _print_pvs (struct formatter *f,struct volume_group *vg)
@@ -398,16 +407,16 @@ static int _print_pvs(struct formatter *f, struct volume_group *vg) dm_list_iterate_items(pvl, &vg->pvs) { pv = pvl->pv; -if (!(name =_get_pv_name (f, pv))) +if (!id_write_format (&pv->id, buffer,sizeof (buffer))) + return_0; + +if (!(name =_get_pv_name_from_uuid (f, buffer))) return_0;outnl (f);outf (f,"%s {" , name);_inc_indent (f); -if (!id_write_format (&pv->id, buffer,sizeof (buffer))) - return_0; -outf (f,"id = \"%s\"" , buffer);if (!(buf =alloca (escaped_len (pv_dev_name (pv))))) {
@@ -621,7 +630,7 @@ static int _build_pv_names(struct formatter *f, struct volume_group *vg)int count =0 ;struct pv_list *pvl;struct physical_volume *pv; -char buffer[32 ], *name; +char buffer[32 ], *uuid, *name;if (!(f->mem =dm_pool_create ("text pv_names" ,512 ))) return_0;
@@ -639,7 +648,11 @@ static int _build_pv_names(struct formatter *f, struct volume_group *vg)if (!(name =dm_pool_strdup (f->mem, buffer))) return_0; -if (!dm_hash_insert (f->pv_names,pv_dev_name (pv), name)) +if (!(uuid =dm_pool_zalloc (f->mem,40 )) || + !id_write_format (&pv->id, uuid,40 )) + return_0; + +if (!dm_hash_insert (f->pv_names, uuid, name)) return_0; }
test/t-vgcfgbackup-usage.sh +20 -2
@@ -11,14 +11,32 @@ . ./test-utils.sh -aux prepare_pvs 2 +aux prepare_pvs 4 # vgcfgbackup handles similar VG names (bz458941) vg1=${PREFIX}vg00 -vg1=${PREFIX}vg01 +vg2=${PREFIX}vg01 vgcreate $vg1 $dev1 vgcreate $vg2 $dev2 vgcfgbackup -f /tmp/bak-%s >out grep "Volume group \"$vg1\" successfully backed up." out grep "Volume group \"$vg2\" successfully backed up." out +vgremove -ff $vg1 +vgremove -ff $vg2 +# vgcfgbackup correctly stores metadata with missing PVs +# and vgcfgrestore able to restore them when device reappears +pv1_uuid=$(pvs --noheadings -o pv_uuid $dev1) +pv2_uuid=$(pvs --noheadings -o pv_uuid $dev2) +vgcreate $vg $devs +lvcreate -l1 -n $lv1 $vg $dev1 +lvcreate -l1 -n $lv2 $vg $dev2 +lvcreate -l1 -n $lv3 $vg $dev3 +vgchange -a n $vg +pvcreate -ff -y $dev1 +pvcreate -ff -y $dev2 +vgcfgbackup -f "$(pwd)/backup.$$" $vg +sed 's/flags = \[\"MISSING\"\]/flags = \[\]/' "$(pwd)/backup.$$" > "$(pwd)/backup.$$1" +pvcreate -ff -y -u $pv1_uuid $dev1 +pvcreate -ff -y -u $pv2_uuid $dev2 +vgcfgrestore -f "$(pwd)/backup.$$1" $vg
PATCH 1/1 vgcfgbackup: do not produce corrupted metadata if PVs are missing
Quoted 6 lines · Milan Broz
Currently code uses
pv_dev_name()for hash when getting internal "pvX" name.This produce corrupted metadata if PVs are missing,
pv->devis NULL and all these missing devices returns one name (using "unknown device" for all missing devices as hash key).
Good catch. Patch looks good (and comes with a test).
Quoted 1 line
Signed-off-by Milan Broz <[email protected]>
Signature · 6 lines
Peter Rockai | me()mornfall!net | prockai()redhat!com
http://blog.mornfall.net | http://web.mornfall.net
"In My Egotistical Opinion, most people's C programs should be
indented six feet downward and covered with dirt."
-- Blair P. Houghton on the subject of C program indentation