From: Daeho Jeong <daehojeong@google.com>

In age-based victim selection (ATGC, AT_SSR, or GC_CB), f2fs_get_victim
can encounter sections with zero valid blocks. This situation often
arises when checkpoint is disabled or due to race conditions between
SIT updates and dirty list management.

In such cases, f2fs_get_section_mtime() returns INVALID_MTIME, which
subsequently triggers a fatal f2fs_bug_on(sbi, mtime == INVALID_MTIME)
in add_victim_entry() or get_cb_cost().

This patch adds a check in f2fs_get_victim's selection loop to skip
sections with no valid blocks. This prevents unnecessary age
calculations for empty sections and avoids the associated kernel panic.
This change also allows removing redundant checks in add_victim_entry().

Signed-off-by: Daeho Jeong <daehojeong@google.com>
---
v2: changed the check position.
---
 fs/f2fs/gc.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index 2e0f67946914..246496fbe5e5 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -907,6 +907,9 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
 				if (!f2fs_segment_has_free_slot(sbi, segno))
 					goto next;
 			}
+
+			if (!get_valid_blocks(sbi, segno, true))
+				goto next;
 		}
 
 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
-- 
2.53.0.851.ga537e3e6e9-goog

On 2026/3/17 02:59, Daeho Jeong wrote:
> From: Daeho Jeong <daehojeong@google.com>
> 
> In age-based victim selection (ATGC, AT_SSR, or GC_CB), f2fs_get_victim
> can encounter sections with zero valid blocks. This situation often
> arises when checkpoint is disabled or due to race conditions between
> SIT updates and dirty list management.
> 
> In such cases, f2fs_get_section_mtime() returns INVALID_MTIME, which
> subsequently triggers a fatal f2fs_bug_on(sbi, mtime == INVALID_MTIME)
> in add_victim_entry() or get_cb_cost().
> 
> This patch adds a check in f2fs_get_victim's selection loop to skip
> sections with no valid blocks. This prevents unnecessary age
> calculations for empty sections and avoids the associated kernel panic.
> This change also allows removing redundant checks in add_victim_entry().
> 
> Signed-off-by: Daeho Jeong <daehojeong@google.com>

Reviewed-by: Chao Yu <chao@kernel.org>

Thanks,