linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v3 1/2] f2fs: add lookup_mode mount option
@ 2025-08-05  6:52 Daniel Lee
  2025-08-05  6:52 ` [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode Daniel Lee
                   ` (2 more replies)
  0 siblings, 3 replies; 5+ messages in thread
From: Daniel Lee @ 2025-08-05  6:52 UTC (permalink / raw)
  To: Jaegeuk Kim, Chao Yu; +Cc: linux-f2fs-devel, linux-kernel, Daniel Lee

For casefolded directories, f2fs may fall back to a linear search if
a hash-based lookup fails. This can cause severe performance
regressions.

While this behavior can be controlled by userspace tools (e.g. mkfs,
fsck) by setting an on-disk flag, a kernel-level solution is needed
to guarantee the lookup behavior regardless of the on-disk state.

This commit introduces the 'lookup_mode' mount option to provide this
kernel-side control.

The option accepts three values:
- perf: (Default) Enforces a hash-only lookup. The linear fallback
  is always disabled.
- compat: Enables the linear search fallback for compatibility with
  directory entries from older kernels.
- auto: Determines the mode based on the on-disk flag, preserving the
  userspace-based behavior.

Signed-off-by: Daniel Lee <chullee@google.com>
---
v3:
- show option lookup_mode=perf.
v2:
- rework mount option parsing to use the new mount API.
- add lookup_mode field to struct f2fs_mount_info.
- add show_options support for the new option.
 Documentation/filesystems/f2fs.rst | 19 +++++++++++++++++++
 fs/f2fs/dir.c                      | 17 ++++++++++++++++-
 fs/f2fs/f2fs.h                     |  7 +++++++
 fs/f2fs/super.c                    | 25 +++++++++++++++++++++++++
 4 files changed, 67 insertions(+), 1 deletion(-)

diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
index 03b1efa6d3b2..a80ed82a547a 100644
--- a/Documentation/filesystems/f2fs.rst
+++ b/Documentation/filesystems/f2fs.rst
@@ -370,6 +370,25 @@ errors=%s		 Specify f2fs behavior on critical errors. This supports modes:
 			 ====================== =============== =============== ========
 nat_bits		 Enable nat_bits feature to enhance full/empty nat blocks access,
 			 by default it's disabled.
+lookup_mode=%s		 Control the directory lookup behavior for casefolded
+			 directories. This option has no effect on directories
+			 that do not have the casefold feature enabled.
+
+			 ================== ========================================
+			 Value		    Description
+			 ================== ========================================
+			 perf		    (Default) Enforces a hash-only lookup.
+					    The linear search fallback is always
+					    disabled, ignoring the on-disk flag.
+			 compat		    Enables the linear search fallback for
+					    compatibility with directory entries
+					    created by older kernel that used a
+					    different case-folding algorithm.
+					    This mode ignores the on-disk flag.
+			 auto		    F2FS determines the mode based on the
+					    on-disk `SB_ENC_NO_COMPAT_FALLBACK_FL`
+					    flag.
+			 ================== ========================================
 ======================== ============================================================
 
 Debugfs Entries
diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c
index fffd7749d6d1..48f4f98afb01 100644
--- a/fs/f2fs/dir.c
+++ b/fs/f2fs/dir.c
@@ -16,6 +16,21 @@
 #include "xattr.h"
 #include <trace/events/f2fs.h>
 
+static inline bool f2fs_should_fallback_to_linear(struct inode *dir)
+{
+	struct f2fs_sb_info *sbi = F2FS_I_SB(dir);
+
+	switch (F2FS_OPTION(sbi).lookup_mode) {
+	case LOOKUP_PERF:
+		return false;
+	case LOOKUP_COMPAT:
+		return true;
+	case LOOKUP_AUTO:
+		return !sb_no_casefold_compat_fallback(sbi->sb);
+	}
+	return false;
+}
+
 #if IS_ENABLED(CONFIG_UNICODE)
 extern struct kmem_cache *f2fs_cf_name_slab;
 #endif
@@ -366,7 +381,7 @@ struct f2fs_dir_entry *__f2fs_find_entry(struct inode *dir,
 
 out:
 #if IS_ENABLED(CONFIG_UNICODE)
-	if (!sb_no_casefold_compat_fallback(dir->i_sb) &&
+	if (f2fs_should_fallback_to_linear(dir) &&
 		IS_CASEFOLDED(dir) && !de && use_hash) {
 		use_hash = false;
 		goto start_find_entry;
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 7029aa8b430e..1c0edb8a7134 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -212,6 +212,7 @@ struct f2fs_mount_info {
 	int compress_mode;			/* compression mode */
 	unsigned char extensions[COMPRESS_EXT_NUM][F2FS_EXTENSION_LEN];	/* extensions */
 	unsigned char noextensions[COMPRESS_EXT_NUM][F2FS_EXTENSION_LEN]; /* extensions */
+	unsigned int lookup_mode;
 };
 
 #define F2FS_FEATURE_ENCRYPT			0x00000001
@@ -1448,6 +1449,12 @@ enum {
 	TOTAL_CALL = FOREGROUND,
 };
 
+enum f2fs_lookup_mode {
+	LOOKUP_PERF,
+	LOOKUP_COMPAT,
+	LOOKUP_AUTO,
+};
+
 static inline int f2fs_test_bit(unsigned int nr, char *addr);
 static inline void f2fs_set_bit(unsigned int nr, char *addr);
 static inline void f2fs_clear_bit(unsigned int nr, char *addr);
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index e16c4e2830c2..0638ae9816ac 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -181,6 +181,7 @@ enum {
 	Opt_nat_bits,
 	Opt_jqfmt,
 	Opt_checkpoint,
+	Opt_lookup_mode,
 	Opt_err,
 };
 
@@ -244,6 +245,13 @@ static const struct constant_table f2fs_param_errors[] = {
 	{}
 };
 
+static const struct constant_table f2fs_param_lookup_mode[] = {
+	{"perf",	LOOKUP_PERF},
+	{"compat",	LOOKUP_COMPAT},
+	{"auto",	LOOKUP_AUTO},
+	{}
+};
+
 static const struct fs_parameter_spec f2fs_param_specs[] = {
 	fsparam_enum("background_gc", Opt_gc_background, f2fs_param_background_gc),
 	fsparam_flag("disable_roll_forward", Opt_disable_roll_forward),
@@ -300,6 +308,7 @@ static const struct fs_parameter_spec f2fs_param_specs[] = {
 	fsparam_enum("memory", Opt_memory_mode, f2fs_param_memory_mode),
 	fsparam_flag("age_extent_cache", Opt_age_extent_cache),
 	fsparam_enum("errors", Opt_errors, f2fs_param_errors),
+	fsparam_enum("lookup_mode", Opt_lookup_mode, f2fs_param_lookup_mode),
 	{}
 };
 
@@ -336,6 +345,7 @@ static match_table_t f2fs_checkpoint_tokens = {
 #define F2FS_SPEC_discard_unit			(1 << 21)
 #define F2FS_SPEC_memory_mode			(1 << 22)
 #define F2FS_SPEC_errors			(1 << 23)
+#define F2FS_SPEC_lookup_mode			(1 << 24)
 
 struct f2fs_fs_context {
 	struct f2fs_mount_info info;
@@ -1143,6 +1153,10 @@ static int f2fs_parse_param(struct fs_context *fc, struct fs_parameter *param)
 	case Opt_nat_bits:
 		ctx_set_opt(ctx, F2FS_MOUNT_NAT_BITS);
 		break;
+	case Opt_lookup_mode:
+		F2FS_CTX_INFO(ctx).lookup_mode = result.uint_32;
+		ctx->spec_mask |= F2FS_SPEC_lookup_mode;
+		break;
 	}
 	return 0;
 }
@@ -1652,6 +1666,8 @@ static void f2fs_apply_options(struct fs_context *fc, struct super_block *sb)
 		F2FS_OPTION(sbi).memory_mode = F2FS_CTX_INFO(ctx).memory_mode;
 	if (ctx->spec_mask & F2FS_SPEC_errors)
 		F2FS_OPTION(sbi).errors = F2FS_CTX_INFO(ctx).errors;
+	if (ctx->spec_mask & F2FS_SPEC_lookup_mode)
+		F2FS_OPTION(sbi).lookup_mode = F2FS_CTX_INFO(ctx).lookup_mode;
 
 	f2fs_apply_compression(fc, sb);
 	f2fs_apply_test_dummy_encryption(fc, sb);
@@ -2416,6 +2432,13 @@ static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
 	if (test_opt(sbi, NAT_BITS))
 		seq_puts(seq, ",nat_bits");
 
+	if (F2FS_OPTION(sbi).lookup_mode == LOOKUP_PERF)
+		seq_show_option(seq, "lookup_mode", "perf");
+	else if (F2FS_OPTION(sbi).lookup_mode == LOOKUP_COMPAT)
+		seq_show_option(seq, "lookup_mode", "compat");
+	else if (F2FS_OPTION(sbi).lookup_mode == LOOKUP_AUTO)
+		seq_show_option(seq, "lookup_mode", "auto");
+
 	return 0;
 }
 
@@ -2480,6 +2503,8 @@ static void default_options(struct f2fs_sb_info *sbi, bool remount)
 #endif
 
 	f2fs_build_fault_attr(sbi, 0, 0, FAULT_ALL);
+
+	F2FS_OPTION(sbi).lookup_mode = LOOKUP_PERF;
 }
 
 #ifdef CONFIG_QUOTA
-- 
2.50.1.565.gc32cd1483b-goog


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode
  2025-08-05  6:52 [PATCH v3 1/2] f2fs: add lookup_mode mount option Daniel Lee
@ 2025-08-05  6:52 ` Daniel Lee
  2025-08-05  7:00   ` Chao Yu
  2025-08-05  6:59 ` [PATCH v3 1/2] f2fs: add lookup_mode mount option Chao Yu
  2025-08-15 16:40 ` [f2fs-dev] " patchwork-bot+f2fs
  2 siblings, 1 reply; 5+ messages in thread
From: Daniel Lee @ 2025-08-05  6:52 UTC (permalink / raw)
  To: Jaegeuk Kim, Chao Yu; +Cc: linux-f2fs-devel, linux-kernel, Daniel Lee

This commit introduces a new read-only sysfs entry at
/sys/fs/f2fs/<device>/effective_lookup_mode.

This entry displays the actual directory lookup mode F2FS is
currently using. This is needed for debugging and verification,
as the behavior is determined by both on-disk flags and mount
options.

Signed-off-by: Daniel Lee <chullee@google.com>
---
v2:
- show "auto:perf" or "auto:compat" when lookup_mode is 'auto'.
 Documentation/ABI/testing/sysfs-fs-f2fs | 15 +++++++++++++++
 fs/f2fs/sysfs.c                         | 18 ++++++++++++++++++
 2 files changed, 33 insertions(+)

diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
index bc0e7fefc39d..ce189acd1908 100644
--- a/Documentation/ABI/testing/sysfs-fs-f2fs
+++ b/Documentation/ABI/testing/sysfs-fs-f2fs
@@ -883,3 +883,18 @@ Date:		June 2025
 Contact:	"Daeho Jeong" <daehojeong@google.com>
 Description:	Control GC algorithm for boost GC. 0: cost benefit, 1: greedy
 		Default: 1
+
+What:		/sys/fs/f2fs/<disk>/effective_lookup_mode
+Date:		August 2025
+Contact:	"Daniel Lee" <chullee@google.com>
+Description:
+		This is a read-only entry to show the effective directory lookup mode
+		F2FS is currently using for casefolded directories.
+		This considers both the "lookup_mode" mount option and the on-disk
+		encoding flag, SB_ENC_NO_COMPAT_FALLBACK_FL.
+
+		Possible values are:
+		- "perf": Hash-only lookup.
+		- "compat": Hash-based lookup with a linear search fallback enabled
+		- "auto:perf": lookup_mode is auto and fallback is disabled on-disk
+		- "auto:compat": lookup_mode is auto and fallback is enabled on-disk
diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
index f736052dea50..82489c78aeda 100644
--- a/fs/f2fs/sysfs.c
+++ b/fs/f2fs/sysfs.c
@@ -281,6 +281,22 @@ static ssize_t encoding_flags_show(struct f2fs_attr *a,
 		le16_to_cpu(F2FS_RAW_SUPER(sbi)->s_encoding_flags));
 }
 
+static ssize_t effective_lookup_mode_show(struct f2fs_attr *a,
+		struct f2fs_sb_info *sbi, char *buf)
+{
+	switch (F2FS_OPTION(sbi).lookup_mode) {
+	case LOOKUP_PERF:
+		return sysfs_emit(buf, "perf\n");
+	case LOOKUP_COMPAT:
+		return sysfs_emit(buf, "compat\n");
+	case LOOKUP_AUTO:
+		if (sb_no_casefold_compat_fallback(sbi->sb))
+			return sysfs_emit(buf, "auto:perf\n");
+		return sysfs_emit(buf, "auto:compat\n");
+	}
+	return 0;
+}
+
 static ssize_t mounted_time_sec_show(struct f2fs_attr *a,
 		struct f2fs_sb_info *sbi, char *buf)
 {
@@ -1211,6 +1227,7 @@ F2FS_GENERAL_RO_ATTR(current_reserved_blocks);
 F2FS_GENERAL_RO_ATTR(unusable);
 F2FS_GENERAL_RO_ATTR(encoding);
 F2FS_GENERAL_RO_ATTR(encoding_flags);
+F2FS_GENERAL_RO_ATTR(effective_lookup_mode);
 F2FS_GENERAL_RO_ATTR(mounted_time_sec);
 F2FS_GENERAL_RO_ATTR(main_blkaddr);
 F2FS_GENERAL_RO_ATTR(pending_discard);
@@ -1329,6 +1346,7 @@ static struct attribute *f2fs_attrs[] = {
 	ATTR_LIST(current_reserved_blocks),
 	ATTR_LIST(encoding),
 	ATTR_LIST(encoding_flags),
+	ATTR_LIST(effective_lookup_mode),
 	ATTR_LIST(mounted_time_sec),
 #ifdef CONFIG_F2FS_STAT_FS
 	ATTR_LIST(cp_foreground_calls),
-- 
2.50.1.565.gc32cd1483b-goog


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* Re: [PATCH v3 1/2] f2fs: add lookup_mode mount option
  2025-08-05  6:52 [PATCH v3 1/2] f2fs: add lookup_mode mount option Daniel Lee
  2025-08-05  6:52 ` [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode Daniel Lee
@ 2025-08-05  6:59 ` Chao Yu
  2025-08-15 16:40 ` [f2fs-dev] " patchwork-bot+f2fs
  2 siblings, 0 replies; 5+ messages in thread
From: Chao Yu @ 2025-08-05  6:59 UTC (permalink / raw)
  To: Daniel Lee, Jaegeuk Kim; +Cc: chao, linux-f2fs-devel, linux-kernel

On 8/5/25 14:52, Daniel Lee wrote:
> For casefolded directories, f2fs may fall back to a linear search if
> a hash-based lookup fails. This can cause severe performance
> regressions.
> 
> While this behavior can be controlled by userspace tools (e.g. mkfs,
> fsck) by setting an on-disk flag, a kernel-level solution is needed
> to guarantee the lookup behavior regardless of the on-disk state.
> 
> This commit introduces the 'lookup_mode' mount option to provide this
> kernel-side control.
> 
> The option accepts three values:
> - perf: (Default) Enforces a hash-only lookup. The linear fallback
>   is always disabled.
> - compat: Enables the linear search fallback for compatibility with
>   directory entries from older kernels.
> - auto: Determines the mode based on the on-disk flag, preserving the
>   userspace-based behavior.
> 
> Signed-off-by: Daniel Lee <chullee@google.com>

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

Thanks,

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode
  2025-08-05  6:52 ` [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode Daniel Lee
@ 2025-08-05  7:00   ` Chao Yu
  0 siblings, 0 replies; 5+ messages in thread
From: Chao Yu @ 2025-08-05  7:00 UTC (permalink / raw)
  To: Daniel Lee, Jaegeuk Kim; +Cc: chao, linux-f2fs-devel, linux-kernel

On 8/5/25 14:52, Daniel Lee wrote:
> This commit introduces a new read-only sysfs entry at
> /sys/fs/f2fs/<device>/effective_lookup_mode.
> 
> This entry displays the actual directory lookup mode F2FS is
> currently using. This is needed for debugging and verification,
> as the behavior is determined by both on-disk flags and mount
> options.
> 
> Signed-off-by: Daniel Lee <chullee@google.com>

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

Thanks,

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [f2fs-dev] [PATCH v3 1/2] f2fs: add lookup_mode mount option
  2025-08-05  6:52 [PATCH v3 1/2] f2fs: add lookup_mode mount option Daniel Lee
  2025-08-05  6:52 ` [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode Daniel Lee
  2025-08-05  6:59 ` [PATCH v3 1/2] f2fs: add lookup_mode mount option Chao Yu
@ 2025-08-15 16:40 ` patchwork-bot+f2fs
  2 siblings, 0 replies; 5+ messages in thread
From: patchwork-bot+f2fs @ 2025-08-15 16:40 UTC (permalink / raw)
  To: Daniel Lee; +Cc: jaegeuk, chao, linux-kernel, linux-f2fs-devel

Hello:

This series was applied to jaegeuk/f2fs.git (dev)
by Jaegeuk Kim <jaegeuk@kernel.org>:

On Mon,  4 Aug 2025 23:52:26 -0700 you wrote:
> For casefolded directories, f2fs may fall back to a linear search if
> a hash-based lookup fails. This can cause severe performance
> regressions.
> 
> While this behavior can be controlled by userspace tools (e.g. mkfs,
> fsck) by setting an on-disk flag, a kernel-level solution is needed
> to guarantee the lookup behavior regardless of the on-disk state.
> 
> [...]

Here is the summary with links:
  - [f2fs-dev,v3,1/2] f2fs: add lookup_mode mount option
    https://git.kernel.org/jaegeuk/f2fs/c/632f0b6c3e32
  - [f2fs-dev,v3,2/2] f2fs: add sysfs entry for effective lookup mode
    https://git.kernel.org/jaegeuk/f2fs/c/1bd119da0b93

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html



^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2025-08-15 16:40 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-08-05  6:52 [PATCH v3 1/2] f2fs: add lookup_mode mount option Daniel Lee
2025-08-05  6:52 ` [PATCH v3 2/2] f2fs: add sysfs entry for effective lookup mode Daniel Lee
2025-08-05  7:00   ` Chao Yu
2025-08-05  6:59 ` [PATCH v3 1/2] f2fs: add lookup_mode mount option Chao Yu
2025-08-15 16:40 ` [f2fs-dev] " patchwork-bot+f2fs

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).