On 1/5/2026 11:30 PM, Nanzhe Zhao wrote:
f2fs_read_data_large_folio() relies on f2fs_map_blocks() to decide whether
a subpage should be zero-filled or queued to a read bio.
However, f2fs_map_blocks() can set F2FS_MAP_MAPPED for NULL_ADDR and
NEW_ADDR in the non-DIO, no-create path. The large folio read code then
Nanzhe,
IIUC, f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT) won't map hole
space, or am I missing something?
Thanks,
treats such hole blocks as mapped blocks, and may account them
in read_pages_pending and attempt to build bios for them, which can
leave tasks stuck in readahead for heavily fragmented files.
Add a helper, f2fs_block_needs_zeroing(), which detects NULL_ADDR and
NEW_ADDR from struct f2fs_map_blocks. Use it to prioritize the zeroing
path by checking f2fs_block_needs_zeroing() before
(map.m_flags & F2FS_MAP_MAPPED) under got_it: label.
Signed-off-by: Nanzhe Zhao <[email protected]>
---
fs/f2fs/data.c | 21 +++++++++++++--------
1 file changed, 13 insertions(+), 8 deletions(-)
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 4bef04560924..66ab7a43a56f 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -2413,6 +2413,11 @@ static void ffs_detach_free(struct folio *folio)
kmem_cache_free(ffs_entry_slab, ffs);
}
+static inline bool f2fs_block_needs_zeroing(const struct f2fs_map_blocks *map)
+{
+ return map->m_pblk == NULL_ADDR || map->m_pblk == NEW_ADDR;
+}
+
static int f2fs_read_data_large_folio(struct inode *inode,
struct readahead_control *rac, struct folio *folio)
{
@@ -2468,14 +2473,7 @@ static int f2fs_read_data_large_folio(struct inode
*inode,
if (ret)
goto err_out;
got_it:
- if ((map.m_flags & F2FS_MAP_MAPPED)) {
- block_nr = map.m_pblk + index - map.m_lblk;
- if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
- DATA_GENERIC_ENHANCE_READ)) {
- ret = -EFSCORRUPTED;
- goto err_out;
- }
- } else {
+ if ((f2fs_block_needs_zeroing(&map))) {
folio_zero_range(folio, offset << PAGE_SHIFT,
PAGE_SIZE);
if (f2fs_need_verity(inode, index) &&
!fsverity_verify_page(folio_file_page(folio,
@@ -2484,6 +2482,13 @@ static int f2fs_read_data_large_folio(struct inode
*inode,
goto err_out;
}
continue;
+ } else if((map.m_flags & F2FS_MAP_MAPPED)) {
+ block_nr = map.m_pblk + index - map.m_lblk;
+ if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
+ DATA_GENERIC_ENHANCE_READ)) {
+ ret = -EFSCORRUPTED;
+ goto err_out;
+ }
}
/* We must increment read_pages_pending before possible BIOs
submitting
--
2.34.1
_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
