All of lore.kernel.org
 help / color / mirror / Atom feed
From: Boris Burkov <boris@bur.io>
To: linux-btrfs@vger.kernel.org, kernel-team@fb.com, fdmanana@kernel.org
Subject: [PATCH v4 1/2] btrfs: warn on invalid slot in tree mod log rewind
Date: Thu,  1 Jun 2023 11:55:13 -0700	[thread overview]
Message-ID: <742a0f5e01de22aae0ada9504834b0e5e3177349.1685645613.git.boris@bur.io> (raw)
In-Reply-To: <cover.1685645613.git.boris@bur.io>

The way that tree mod log tracks the ultimate length of the eb, the
variable 'n', eventually turns up the correct value, but at intermediate
steps during the rewind, n can be inaccurate as a representation of the
end of the eb. For example, it doesn't get updated on move rewinds, and
it does get updated for add/remove in the middle of the eb.

To detect cases with invalid moves, introduce a separate variable called
max_slot which tries to track the maximum valid slot in the rewind eb.
We can then warn if we do a move whose src range goes beyond the max
valid slot.

There is a commented caveat that it is possible to have this value be an
overestimate due to the challenge of properly handling 'add' operations
in the middle of the eb, but in practice it doesn't cause enough of a
problem to throw out the max idea in favor of tracking every valid slot.

Signed-off-by: Boris Burkov <boris@bur.io>
---
 fs/btrfs/tree-mod-log.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)

diff --git a/fs/btrfs/tree-mod-log.c b/fs/btrfs/tree-mod-log.c
index a555baa0143a..157e1a0efab8 100644
--- a/fs/btrfs/tree-mod-log.c
+++ b/fs/btrfs/tree-mod-log.c
@@ -664,10 +664,27 @@ static void tree_mod_log_rewind(struct btrfs_fs_info *fs_info,
 	unsigned long o_dst;
 	unsigned long o_src;
 	unsigned long p_size = sizeof(struct btrfs_key_ptr);
+	/*
+	 * max_slot tracks the maximum valid slot of the rewind eb at every
+	 * step of the rewind. This is in contrast with 'n' which eventually
+	 * matches the number of items, but can be wrong during moves or if
+	 * removes overlap on already valid slots (which is probably separately
+	 * a bug). We do this to validate the offsets of memmoves for rewinding
+	 * moves and detect invalid memmoves.
+	 *
+	 * Since a rewind eb can start empty, max_slot is a signed integer with
+	 * a special meaning for -1, which is that no slot is valid to move out
+	 * of. Any other negative value is invalid.
+	 */
+	int max_slot;
+	int move_src_end_slot;
+	int move_dst_end_slot;
 
 	n = btrfs_header_nritems(eb);
+	max_slot = n - 1;
 	read_lock(&fs_info->tree_mod_log_lock);
 	while (tm && tm->seq >= time_seq) {
+		ASSERT(max_slot >= -1);
 		/*
 		 * All the operations are recorded with the operator used for
 		 * the modification. As we're going backwards, we do the
@@ -684,6 +701,8 @@ static void tree_mod_log_rewind(struct btrfs_fs_info *fs_info,
 			btrfs_set_node_ptr_generation(eb, tm->slot,
 						      tm->generation);
 			n++;
+			if (tm->slot > max_slot)
+				max_slot = tm->slot;
 			break;
 		case BTRFS_MOD_LOG_KEY_REPLACE:
 			BUG_ON(tm->slot >= n);
@@ -693,14 +712,36 @@ static void tree_mod_log_rewind(struct btrfs_fs_info *fs_info,
 						      tm->generation);
 			break;
 		case BTRFS_MOD_LOG_KEY_ADD:
+			/*
+			 * It is possible we could have already removed keys behind the known
+			 * max slot, so this will be an overestimate. In practice, the copy
+			 * operation inserts them in increasing order, and overestimating just
+			 * means we miss some warnings, so it's OK. It isn't worth carefully
+			 * tracking the full array of valid slots to check against when moving.
+			 */
+			if (tm->slot == max_slot)
+				max_slot--;
 			/* if a move operation is needed it's in the log */
 			n--;
 			break;
 		case BTRFS_MOD_LOG_MOVE_KEYS:
+			ASSERT(tm->move.nr_items > 0);
+			move_src_end_slot = tm->move.dst_slot + tm->move.nr_items - 1;
+			move_dst_end_slot = tm->slot + tm->move.nr_items - 1;
 			o_dst = btrfs_node_key_ptr_offset(eb, tm->slot);
 			o_src = btrfs_node_key_ptr_offset(eb, tm->move.dst_slot);
+			if (WARN_ON(move_src_end_slot > max_slot ||
+				    tm->move.nr_items <= 0)) {
+				btrfs_warn(fs_info,
+					   "Move from invalid tree mod log slot eb %llu slot %d dst_slot %d nr_items %d seq %llu n %u max_slot %d\n",
+					   eb->start, tm->slot,
+					   tm->move.dst_slot, tm->move.nr_items,
+					   tm->seq, n, max_slot);
+
+			}
 			memmove_extent_buffer(eb, o_dst, o_src,
 					      tm->move.nr_items * p_size);
+			max_slot = move_dst_end_slot;
 			break;
 		case BTRFS_MOD_LOG_ROOT_REPLACE:
 			/*
-- 
2.40.1


  reply	other threads:[~2023-06-01 18:55 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-06-01 18:55 [PATCH v4 0/2] btrfs: fix logical_to_ino panic in btrfs_map_bio Boris Burkov
2023-06-01 18:55 ` Boris Burkov [this message]
2023-06-02 10:09   ` [PATCH v4 1/2] btrfs: warn on invalid slot in tree mod log rewind Filipe Manana
2023-06-01 18:55 ` [PATCH v4 2/2] btrfs: insert tree mod log move in push_node_left Boris Burkov
2023-06-02 10:10   ` Filipe Manana
2023-06-02 20:07 ` [PATCH v4 0/2] btrfs: fix logical_to_ino panic in btrfs_map_bio David Sterba

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=742a0f5e01de22aae0ada9504834b0e5e3177349.1685645613.git.boris@bur.io \
    --to=boris@bur.io \
    --cc=fdmanana@kernel.org \
    --cc=kernel-team@fb.com \
    --cc=linux-btrfs@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.