Skip to contentSkip to conversation

LVM2: some code cleanups and bugfixs checked by codeDEX tools

PATCH New Wu Guanghao 2 participants 8 messages Sep 11, 2020 – Sep 14, 2020

lore
Share & tools
lore
#0 lore raw

cover LVM2: some code cleanups and bugfixs checked by codeDEX tools

When I learn the LVM2 source code and test it, I find some bugs and fix them. At the same time, static scanning tools were used to analyze the code, and several possible problems were found.

repo: https://sourceware.org/git/lvm2.git
repo link: https://sourceware.org/git/lvm2.git
branch: master

wuguanghao (4):
pvmove: check return value of top_level_lv_name() in _set_up_pvmove_lv
lvconvert_poll: check whether lv is a snapshot type in poll_thin_merge_progress
lvconvert: _lvconvert_splitsnapshot check if lv is snapshot type fix core dump when thinpool size more than the remaining space of vg

 tools/lvconvert.c      | 3 +++
 tools/lvconvert_poll.c | 2 +-
 tools/lvcreate.c       | 7 +++++++
 tools/pvmove.c         | 7 ++++++-
 4 files changed, 17 insertions(+), 2 deletions(-)
#1 lore raw

PATCH 1/4 pvmove: check return value of top_level_lv_name() in _set_up_pvmove_lv

The return value of top_level_lv_name may be NULL, so we should check return value of top_level_lv_name before calling strcmp(lv->name, top_level_lv_name(vg, lv_name)).

---
 tools/pvmove.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
tools/pvmove.c +6 -1
index cc512b9de..3101de1df 100644
@@ -326,6 +326,7 @@ static struct logical_volume *_set_up_pvmove_lv(struct cmd_context *cmd,
 	int lv_skipped = 0;
 	int needs_exclusive = *exclusive;
 	const struct logical_volume *holder;
+	const char *new_lv_name;
 
 	/* FIXME Cope with non-contiguous => splitting existing segments */
 	if (!(lv_mirr = lv_create_empty("pvmove%d", NULL,
@@ -363,7 +364,11 @@ static struct logical_volume *_set_up_pvmove_lv(struct cmd_context *cmd,
 		if (lv == lv_mirr)
 			continue;
 
-		if (lv_name && strcmp(lv->name, top_level_lv_name(vg, lv_name)))
+		new_lv_name = top_level_lv_name(vg, lv_name);
+		if (!new_lv_name)
+			return NULL;
+
+		if (lv_name && strcmp(lv->name, new_lv_name))
 			continue;
 
 		if (!lv_is_on_pvs(lv, source_pvl))
#2 lore raw

PATCH 2/4 lvconvert_poll: check whether lv is a snapshot type in poll_thin_merge_progress

In reply to Wu Guanghao

lv may not be a snapshot type, the return value of find_snapshot(lv) may be NULL. Here, we will call stack if lv is not a snapshot type.

---
 tools/lvconvert_poll.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/lvconvert_poll.c +1 -1
index b9ec316e1..d0831cfd8 100644
@@ -176,7 +176,7 @@ progress_t poll_thin_merge_progress(struct cmd_context *cmd,
 {
 	uint32_t device_id;
 
-	if (!lv_thin_device_id(lv, &device_id)) {
+	if (!lv_thin_device_id(lv, &device_id) || !lv_is_snapshot(lv)) {
 		stack;
 		return PROGRESS_CHECK_FAILED;
 	}
#3 lore raw

PATCH 3/4 lvconvert: _lvconvert_splitsnapshot check if lv is snapshot type

In reply to Wu Guanghao

cow may not be a snapshot type, the return value of origin_from_cow(cow) may be NULL

---
 tools/lvconvert.c | 3 +++
 1 file changed, 3 insertions(+)
tools/lvconvert.c +3 -0
index 0adc61e36..f6a5572c7 100644
@@ -1825,6 +1825,9 @@ static int _lvconvert_splitsnapshot(struct cmd_context *cmd, struct logical_volu
 	struct volume_group *vg = cow->vg;
 	const char *cow_name = display_lvname(cow);
 
+	if (!lv_is_snapshot(cow))
+		return 0;
+
 	if (lv_is_virtual_origin(origin_from_cow(cow))) {
 		log_error("Unable to split off snapshot %s with virtual origin.", cow_name);
 		return 0;
#4 lore raw

PATCH 4/4 fix core dump when thinpool size more than the remaining, space of vg

In reply to Wu Guanghao

Create thinpool through the lvcreate command,
specify the data device size of the thinpool as x,
the metadata device size as y, and the remaining space of the volume group as z. When x + y < z and x + 2y > z, then it must enter the core dump.

Example:
"
sdc 8:32 0 40G 0 disk
# pvcreate /dev/sdc
Physical volume "/dev/sdc" successfully created.
# vgcreate vg /dev/sdc
Volume group "vg" successfully created
# lvcreate -L 30G --poolmetadatasize 5G -T vg/thinpool
Thin pool volume with chunk size 64.00 KiB can address at most 15.81 TiB of data.
lvcreate: metadata/pv_map.c:198: consume_pv_area: Assertion `to_go <= pva->count' failed. Aborted (core dumped)
"

When the volume group allocates space, it will surround the assertion when it finds that its own space is insufficient, leading to coredump.

So before actually creating the thinpool, we will check if the remaining space of the volume group is sufficient

---
 tools/lvcreate.c | 7 +++++++
 1 file changed, 7 insertions(+)
tools/lvcreate.c +7 -0
index e384291ae..5e86ecba6 100644
@@ -1721,6 +1721,13 @@ static int _lvcreate_single(struct cmd_context *cmd, const char *vg_name,
 		if (!handle_pool_metadata_spare(vg, lp->pool_metadata_extents,
 						lp->pvh, lp->pool_metadata_spare))
 			goto_out;
+		
+		if (vg->free_count < (uint64_t)lp->extents + lp->pool_metadata_extents) {
+			log_error("%s doesn't have enough space, create thinpool failed. %s",
+				vg->name, lp->pool_metadata_spare ?
+				"metadata back device occupies part of the VG space" : "");
+			goto_out;
+		}
 
 		log_verbose("Making pool %s in VG %s using segtype %s",
 			    lp->pool_name ? : "with generated name", lp->vg_name, lp->segtype->name);
#5 lore raw

cover LVM2: some code cleanups and bugfixs checked by codeDEX tools

In reply to Wu Guanghao

Dne 11. 09. 20 v 11:38 Wu Guanghao napsal(a):

Quoted 9 lines

When I learn the LVM2 source code and test it, I find some bugs and fix them. At the same time, static scanning tools were used to analyze the code, and several possible problems were found.

repo: https://sourceware.org/git/lvm2.git
repo link: https://sourceware.org/git/lvm2.git
branch: master

Quoted 6 lines

wuguanghao (4):
pvmove: check return value of top_level_lv_name() in _set_up_pvmove_lv
lvconvert_poll: check whether lv is a snapshot type in poll_thin_merge_progress
lvconvert: _lvconvert_splitsnapshot check if lv is snapshot type fix core dump when thinpool size more than the remaining space of vg

Hi

Great findings - I'll check how they could be applied. In some cases you are rather fixed internal error so it's always great if some test case which show how the error can be hit is provided (like in patch 4)

Regards

Zdenek

#6 lore raw

cover LVM2: some code cleanups and bugfixs checked by codeDEX tools

Hi

Thanks for your review. If you have a better solution for patch 4, please reply by email.

Regards

Wu Guanghao

? 2020/9/11 19:07, Zdenek Kabelac ??:

Quoted 31 lines

Dne 11. 09. 20 v 11:38 Wu Guanghao napsal(a):

Quoted 9 lines

When I learn the LVM2 source code and test it, I find some bugs and fix them. At the same time, static scanning tools were used to analyze the code, and several possible problems were found.

repo: https://sourceware.org/git/lvm2.git
repo link: https://sourceware.org/git/lvm2.git
branch: master

Quoted 6 lines

wuguanghao (4):
?? pvmove: check return value of top_level_lv_name() in _set_up_pvmove_lv ?? lvconvert_poll: check whether lv is a snapshot type in poll_thin_merge_progress ?? lvconvert: _lvconvert_splitsnapshot check if lv is snapshot type ?? fix core dump when thinpool size more than the remaining space of vg

Hi

Great findings - I'll check how they could be applied. In some cases you are rather fixed internal error so it's always great if some test case which show how the error can be hit is provided (like in patch 4)

Regards

Zdenek

.

#7 lore raw

cover LVM2: some code cleanups and bugfixs checked by codeDEX tools

In reply to Zdenek Kabelac

Dne 11. 09. 20 v 13:07 Zdenek Kabelac napsal(a):

Quoted 19 lines

Dne 11. 09. 20 v 11:38 Wu Guanghao napsal(a):

Quoted 9 lines

When I learn the LVM2 source code and test it, I find some bugs and fix them. At the same time, static scanning tools were used to analyze the code, and several possible problems were found.

repo: https://sourceware.org/git/lvm2.git
repo link: https://sourceware.org/git/lvm2.git
branch: master

Quoted 7 lines

wuguanghao (4):
?? pvmove: check return value of top_level_lv_name() in _set_up_pvmove_lv ?? lvconvert_poll: check whether lv is a snapshot type in poll_thin_merge_progress
?? lvconvert: _lvconvert_splitsnapshot check if lv is snapshot type ?? fix core dump when thinpool size more than the remaining space of vg

Quoted 6 lines

Hi

Great findings - I'll check how they could be applied. In some cases you are rather fixed internal error so it's always great if some test case which show how the error can be hit is provided (like in patch 4)

Hi

So the patches are upstream - although they need some extra care:

1 ->
https://www.redhat.com/archives/lvm-devel/2020-September/msg00059.html
Added handling of lv_name

2 ->
https://www.redhat.com/archives/lvm-devel/2020-September/msg00063.html
Switched || to &&

3 ->
https://www.redhat.com/archives/lvm-devel/2020-September/msg00069.html
We are actually splitting 'cow' so lv_is_cow()

4 ->
https://www.redhat.com/archives/lvm-devel/2020-September/msg00062.html
Fixed differently by correcting evaluation of needed extents for allocation.

Once again thanks for reporting mainly the issue from patch 4.

Regards

Zdenek