Skip to contentSkip to conversation

btrfs: fix memory leaks after failure to lookup checksums during inode logging

PATCH In review [email protected] 3 participants 5 messages Jul 29, 2020 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 btrfs: fix memory leaks after failure to lookup checksums during inode logging

From: Filipe Manana <[email protected]>

While logging an inode, at copy_items(), if we fail to lookup the checksums for an extent we release the destination path, free the ins_data array and then return immediately. However a previous iteration of the for loop may have added checksums to the ordered_sums list, in which case we leak the memory used by them.

So fix this by making sure we iterate the ordered_sums list and free all its checksums before returning.

Fixes 3650860b90cc2a ("Btrfs: remove almost all of the BUG()'s from tree-log.c")
---
 fs/btrfs/tree-log.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)
fs/btrfs/tree-log.c +2 -6
index 20334bebcaf2..f1812f5baec4 100644
@@ -4035,11 +4035,8 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 						fs_info->csum_root,
 						ds + cs, ds + cs + cl - 1,
 						&ordered_sums, 0);
-				if (ret) {
-					btrfs_release_path(dst_path);
-					kfree(ins_data);
-					return ret;
-				}
+				if (ret)
+					break;
 			}
 		}
 	}
@@ -4052,7 +4049,6 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	 * we have to do this after the loop above to avoid changing the
 	 * log tree while trying to change the log tree.
 	 */
-	ret = 0;
 	while (!list_empty(&ordered_sums)) {
 		struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
 						   struct btrfs_ordered_sum,
Signature · 1 line
2.26.2
#1 lore raw
Quoted 19 lines · On 29/07/2020 11:18, [email protected] wrote:
fs/btrfs/tree-log.c +2 -5
index 20334bebcaf2..f1812f5baec4 100644
@@ -4035,11 +4035,8 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 						fs_info->csum_root,
 						ds + cs, ds + cs + cl - 1,
 						&ordered_sums, 0);
-				if (ret) {
-					btrfs_release_path(dst_path);
-					kfree(ins_data);
-					return ret;
-				}
+				if (ret)
+					break;
 			}
 		}
 	}

fs/btrfs/tree-log.c static noinline int copy_items(struct btrfs_trans_handle *trans, · line 4053

Quoted 6 lines
fs/btrfs/tree-log.c +0 -1
@@ -4052,7 +4049,6 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
 	 * we have to do this after the loop above to avoid changing the
 	 * log tree while trying to change the log tree.
 	 */
-	ret = 0;
 	while (!list_empty(&ordered_sums)) {
 		struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,

Isn't there a potential that ret get overridden by this hunk:

       while (!list_empty(&ordered_sums)) {
                struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
                                                   struct btrfs_ordered_sum,
                                                   list);
                if (!ret)
                        ret = log_csums(trans, inode, log, sums);
                list_del(&sums->list);
                kfree(sums);
        }

        return ret;

and we're potentially returning 0 instead of the -ENOMEM from 'btrfs_lookup_csums_range'?

Maybe we should do something like this:

       while (!list_empty(&ordered_sums)) {
                struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
                                                   struct btrfs_ordered_sum,
                                                   list);
                if (!ret)
                        ret2 = log_csums(trans, inode, log, sums);
                list_del(&sums->list);
                kfree(sums);
        }

        return ret2 ? ret2 : ret;

Thanks,
Johannes

#2 lore raw
Quoted 30 lines · Johannes Thumshirn
Quoted 19 lines · On 29/07/2020 11:18, [email protected] wrote:
fs/btrfs/tree-log.c +2 -5
index 20334bebcaf2..f1812f5baec4 100644
@@ -4035,11 +4035,8 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
                                              fs_info->csum_root,
                                              ds + cs, ds + cs + cl - 1,
                                              &ordered_sums, 0);
-                             if (ret) {
-                                     btrfs_release_path(dst_path);
-                                     kfree(ins_data);
-                                     return ret;
-                             }
+                             if (ret)
+                                     break;
                      }
              }
      }

fs/btrfs/tree-log.c static noinline int copy_items(struct btrfs_trans_handle *trans, · line 4053

Quoted 6 lines
fs/btrfs/tree-log.c +0 -1
@@ -4052,7 +4049,6 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
       * we have to do this after the loop above to avoid changing the
       * log tree while trying to change the log tree.
       */
-     ret = 0;
      while (!list_empty(&ordered_sums)) {
              struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,

Isn't there a potential that ret get overridden by this hunk:

Quoted 9 lines
       while (!list_empty(&ordered_sums)) {
                struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
                                                   struct btrfs_ordered_sum,
                                                   list);
                if (!ret)
                        ret = log_csums(trans, inode, log, sums);
                list_del(&sums->list);
                kfree(sums);
        }

Nop, when ret is non-zero it never gets overwritten.

Quoted 4 lines
        return ret;

and we're potentially returning 0 instead of the -ENOMEM from 'btrfs_lookup_csums_range'?

Nop, for the same reason mentioned before.

Thanks.

Quoted 17 lines

Maybe we should do something like this:

       while (!list_empty(&ordered_sums)) {
                struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
                                                   struct btrfs_ordered_sum,
                                                   list);
                if (!ret)
                        ret2 = log_csums(trans, inode, log, sums);
                list_del(&sums->list);
                kfree(sums);
        }

        return ret2 ? ret2 : ret;

Thanks,
Johannes

Reviewed-by
#3 lore raw
Quoted 42 lines · Filipe Manana
Quoted 30 lines · Johannes Thumshirn
Quoted 19 lines · On 29/07/2020 11:18, [email protected] wrote:
fs/btrfs/tree-log.c +2 -5
index 20334bebcaf2..f1812f5baec4 100644
@@ -4035,11 +4035,8 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
                                              fs_info->csum_root,
                                              ds + cs, ds + cs + cl - 1,
                                              &ordered_sums, 0);
-                             if (ret) {
-                                     btrfs_release_path(dst_path);
-                                     kfree(ins_data);
-                                     return ret;
-                             }
+                             if (ret)
+                                     break;
                      }
              }
      }

fs/btrfs/tree-log.c static noinline int copy_items(struct btrfs_trans_handle *trans, · line 4053

Quoted 6 lines
fs/btrfs/tree-log.c +0 -1
@@ -4052,7 +4049,6 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
       * we have to do this after the loop above to avoid changing the
       * log tree while trying to change the log tree.
       */
-     ret = 0;
      while (!list_empty(&ordered_sums)) {
              struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,

Isn't there a potential that ret get overridden by this hunk:

Quoted 9 lines
       while (!list_empty(&ordered_sums)) {
                struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
                                                   struct btrfs_ordered_sum,
                                                   list);
                if (!ret)
                        ret = log_csums(trans, inode, log, sums);
                list_del(&sums->list);
                kfree(sums);
        }
Quoted 1 line

Nop, when ret is non-zero it never gets overwritten.

You're absolutely right, I'm sorry.

#4 lore raw
In reply to [email protected]
Quoted 8 lines · On Wed, Jul 29, 2020 at 10:17:50AM +0100, [email protected] wrote:

From: Filipe Manana <[email protected]>

While logging an inode, at copy_items(), if we fail to lookup the checksums for an extent we release the destination path, free the ins_data array and then return immediately. However a previous iteration of the for loop may have added checksums to the ordered_sums list, in which case we leak the memory used by them.

Quoted 5 lines

So fix this by making sure we iterate the ordered_sums list and free all its checksums before returning.

Fixes 3650860b90cc2a ("Btrfs: remove almost all of the BUG()'s from tree-log.c")

Added to misc-next, thanks.