From c6a564ffadc9105880329710164ee493f0de103c Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@lst.de>
Date: Wed, 25 Mar 2020 16:48:42 +0100
Subject: block: move the part_stat* helpers from genhd.h to a new header

These macros are just used by a few files.  Move them out of genhd.h,
which is included everywhere into a new standalone header.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/ext4/sysfs.c | 1 +
 1 file changed, 1 insertion(+)

(limited to 'fs/ext4/sysfs.c')

diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index d218ebdafa4a..04bfaf63752c 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -13,6 +13,7 @@
 #include <linux/seq_file.h>
 #include <linux/slab.h>
 #include <linux/proc_fs.h>
+#include <linux/part_stat.h>
 
 #include "ext4.h"
 #include "ext4_jbd2.h"
-- 
cgit v1.2.3


From ed318a6cc0b620440e65f48eb527dc3df7269ce4 Mon Sep 17 00:00:00 2001
From: Eric Biggers <ebiggers@google.com>
Date: Tue, 12 May 2020 16:32:50 -0700
Subject: fscrypt: support test_dummy_encryption=v2

v1 encryption policies are deprecated in favor of v2, and some new
features (e.g. encryption+casefolding) are only being added for v2.

Therefore, the "test_dummy_encryption" mount option (which is used for
encryption I/O testing with xfstests) needs to support v2 policies.

To do this, extend its syntax to be "test_dummy_encryption=v1" or
"test_dummy_encryption=v2".  The existing "test_dummy_encryption" (no
argument) also continues to be accepted, to specify the default setting
-- currently v1, but the next patch changes it to v2.

To cleanly support both v1 and v2 while also making it easy to support
specifying other encryption settings in the future (say, accepting
"$contents_mode:$filenames_mode:v2"), make ext4 and f2fs maintain a
pointer to the dummy fscrypt_context rather than using mount flags.

To avoid concurrency issues, don't allow test_dummy_encryption to be set
or changed during a remount.  (The former restriction is new, but
xfstests doesn't run into it, so no one should notice.)

Tested with 'gce-xfstests -c {ext4,f2fs}/encrypt -g auto'.  On ext4,
there are two regressions, both of which are test bugs: ext4/023 and
ext4/028 fail because they set an xattr and expect it to be stored
inline, but the increase in size of the fscrypt_context from
24 to 40 bytes causes this xattr to be spilled into an external block.

Link: https://lore.kernel.org/r/20200512233251.118314-4-ebiggers@kernel.org
Acked-by: Jaegeuk Kim <jaegeuk@kernel.org>
Reviewed-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 Documentation/filesystems/f2fs.rst |   6 +-
 fs/crypto/keysetup.c               |  15 ++---
 fs/crypto/policy.c                 | 125 +++++++++++++++++++++++++++++++++++++
 fs/ext4/ext4.h                     |   7 ++-
 fs/ext4/super.c                    |  68 +++++++++++++++-----
 fs/ext4/sysfs.c                    |   2 +
 fs/f2fs/f2fs.h                     |   4 +-
 fs/f2fs/super.c                    |  85 ++++++++++++++++++-------
 fs/f2fs/sysfs.c                    |   4 ++
 include/linux/fscrypt.h            |  51 ++++++++++++---
 10 files changed, 307 insertions(+), 60 deletions(-)

(limited to 'fs/ext4/sysfs.c')

diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
index 87d794bc75a4..4218ac658629 100644
--- a/Documentation/filesystems/f2fs.rst
+++ b/Documentation/filesystems/f2fs.rst
@@ -225,8 +225,12 @@ fsync_mode=%s          Control the policy of fsync. Currently supports "posix",
                        pass, but the performance will regress. "nobarrier" is
                        based on "posix", but doesn't issue flush command for
                        non-atomic files likewise "nobarrier" mount option.
-test_dummy_encryption  Enable dummy encryption, which provides a fake fscrypt
+test_dummy_encryption
+test_dummy_encryption=%s
+                       Enable dummy encryption, which provides a fake fscrypt
                        context. The fake fscrypt context is used by xfstests.
+                       The argument may be either "v1" or "v2", in order to
+                       select the corresponding fscrypt policy version.
 checkpoint=%s[:%u[%]]  Set to "disable" to turn off checkpointing. Set to "enable"
                        to reenable checkpointing. Is enabled by default. While
                        disabled, any unmounting or unexpected shutdowns will cause
diff --git a/fs/crypto/keysetup.c b/fs/crypto/keysetup.c
index edc4590c69f0..675479f8e6f3 100644
--- a/fs/crypto/keysetup.c
+++ b/fs/crypto/keysetup.c
@@ -395,21 +395,18 @@ int fscrypt_get_encryption_info(struct inode *inode)
 
 	res = inode->i_sb->s_cop->get_context(inode, &ctx, sizeof(ctx));
 	if (res < 0) {
-		if (!fscrypt_dummy_context_enabled(inode) ||
-		    IS_ENCRYPTED(inode)) {
+		const union fscrypt_context *dummy_ctx =
+			fscrypt_get_dummy_context(inode->i_sb);
+
+		if (IS_ENCRYPTED(inode) || !dummy_ctx) {
 			fscrypt_warn(inode,
 				     "Error %d getting encryption context",
 				     res);
 			return res;
 		}
 		/* Fake up a context for an unencrypted directory */
-		memset(&ctx, 0, sizeof(ctx));
-		ctx.version = FSCRYPT_CONTEXT_V1;
-		ctx.v1.contents_encryption_mode = FSCRYPT_MODE_AES_256_XTS;
-		ctx.v1.filenames_encryption_mode = FSCRYPT_MODE_AES_256_CTS;
-		memset(ctx.v1.master_key_descriptor, 0x42,
-		       FSCRYPT_KEY_DESCRIPTOR_SIZE);
-		res = sizeof(ctx.v1);
+		res = fscrypt_context_size(dummy_ctx);
+		memcpy(&ctx, dummy_ctx, res);
 	}
 
 	crypt_info = kmem_cache_zalloc(fscrypt_info_cachep, GFP_NOFS);
diff --git a/fs/crypto/policy.c b/fs/crypto/policy.c
index 0a95537fcef9..bcab94b29d10 100644
--- a/fs/crypto/policy.c
+++ b/fs/crypto/policy.c
@@ -11,6 +11,7 @@
  */
 
 #include <linux/random.h>
+#include <linux/seq_file.h>
 #include <linux/string.h>
 #include <linux/mount.h>
 #include "fscrypt_private.h"
@@ -616,3 +617,127 @@ int fscrypt_inherit_context(struct inode *parent, struct inode *child,
 	return preload ? fscrypt_get_encryption_info(child): 0;
 }
 EXPORT_SYMBOL(fscrypt_inherit_context);
+
+/**
+ * fscrypt_set_test_dummy_encryption() - handle '-o test_dummy_encryption'
+ * @sb: the filesystem on which test_dummy_encryption is being specified
+ * @arg: the argument to the test_dummy_encryption option.
+ *	 If no argument was specified, then @arg->from == NULL.
+ * @dummy_ctx: the filesystem's current dummy context (input/output, see below)
+ *
+ * Handle the test_dummy_encryption mount option by creating a dummy encryption
+ * context, saving it in @dummy_ctx, and adding the corresponding dummy
+ * encryption key to the filesystem.  If the @dummy_ctx is already set, then
+ * instead validate that it matches @arg.  Don't support changing it via
+ * remount, as that is difficult to do safely.
+ *
+ * The reason we use an fscrypt_context rather than an fscrypt_policy is because
+ * we mustn't generate a new nonce each time we access a dummy-encrypted
+ * directory, as that would change the way filenames are encrypted.
+ *
+ * Return: 0 on success (dummy context set, or the same context is already set);
+ *         -EEXIST if a different dummy context is already set;
+ *         or another -errno value.
+ */
+int fscrypt_set_test_dummy_encryption(struct super_block *sb,
+				      const substring_t *arg,
+				      struct fscrypt_dummy_context *dummy_ctx)
+{
+	const char *argstr = "v1";
+	const char *argstr_to_free = NULL;
+	struct fscrypt_key_specifier key_spec = { 0 };
+	int version;
+	union fscrypt_context *ctx = NULL;
+	int err;
+
+	if (arg->from) {
+		argstr = argstr_to_free = match_strdup(arg);
+		if (!argstr)
+			return -ENOMEM;
+	}
+
+	if (!strcmp(argstr, "v1")) {
+		version = FSCRYPT_CONTEXT_V1;
+		key_spec.type = FSCRYPT_KEY_SPEC_TYPE_DESCRIPTOR;
+		memset(key_spec.u.descriptor, 0x42,
+		       FSCRYPT_KEY_DESCRIPTOR_SIZE);
+	} else if (!strcmp(argstr, "v2")) {
+		version = FSCRYPT_CONTEXT_V2;
+		key_spec.type = FSCRYPT_KEY_SPEC_TYPE_IDENTIFIER;
+		/* key_spec.u.identifier gets filled in when adding the key */
+	} else {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (dummy_ctx->ctx) {
+		/*
+		 * Note: if we ever make test_dummy_encryption support
+		 * specifying other encryption settings, such as the encryption
+		 * modes, we'll need to compare those settings here.
+		 */
+		if (dummy_ctx->ctx->version == version)
+			err = 0;
+		else
+			err = -EEXIST;
+		goto out;
+	}
+
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	err = fscrypt_add_test_dummy_key(sb, &key_spec);
+	if (err)
+		goto out;
+
+	ctx->version = version;
+	switch (ctx->version) {
+	case FSCRYPT_CONTEXT_V1:
+		ctx->v1.contents_encryption_mode = FSCRYPT_MODE_AES_256_XTS;
+		ctx->v1.filenames_encryption_mode = FSCRYPT_MODE_AES_256_CTS;
+		memcpy(ctx->v1.master_key_descriptor, key_spec.u.descriptor,
+		       FSCRYPT_KEY_DESCRIPTOR_SIZE);
+		break;
+	case FSCRYPT_CONTEXT_V2:
+		ctx->v2.contents_encryption_mode = FSCRYPT_MODE_AES_256_XTS;
+		ctx->v2.filenames_encryption_mode = FSCRYPT_MODE_AES_256_CTS;
+		memcpy(ctx->v2.master_key_identifier, key_spec.u.identifier,
+		       FSCRYPT_KEY_IDENTIFIER_SIZE);
+		break;
+	default:
+		WARN_ON(1);
+		err = -EINVAL;
+		goto out;
+	}
+	dummy_ctx->ctx = ctx;
+	ctx = NULL;
+	err = 0;
+out:
+	kfree(ctx);
+	kfree(argstr_to_free);
+	return err;
+}
+EXPORT_SYMBOL_GPL(fscrypt_set_test_dummy_encryption);
+
+/**
+ * fscrypt_show_test_dummy_encryption() - show '-o test_dummy_encryption'
+ * @seq: the seq_file to print the option to
+ * @sep: the separator character to use
+ * @sb: the filesystem whose options are being shown
+ *
+ * Show the test_dummy_encryption mount option, if it was specified.
+ * This is mainly used for /proc/mounts.
+ */
+void fscrypt_show_test_dummy_encryption(struct seq_file *seq, char sep,
+					struct super_block *sb)
+{
+	const union fscrypt_context *ctx = fscrypt_get_dummy_context(sb);
+
+	if (!ctx)
+		return;
+	seq_printf(seq, "%ctest_dummy_encryption=v%d", sep, ctx->version);
+}
+EXPORT_SYMBOL_GPL(fscrypt_show_test_dummy_encryption);
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 91eb4381cae5..546504cba842 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1357,11 +1357,9 @@ struct ext4_super_block {
  */
 #define EXT4_MF_MNTDIR_SAMPLED		0x0001
 #define EXT4_MF_FS_ABORTED		0x0002	/* Fatal error detected */
-#define EXT4_MF_TEST_DUMMY_ENCRYPTION	0x0004
 
 #ifdef CONFIG_FS_ENCRYPTION
-#define DUMMY_ENCRYPTION_ENABLED(sbi) (unlikely((sbi)->s_mount_flags & \
-						EXT4_MF_TEST_DUMMY_ENCRYPTION))
+#define DUMMY_ENCRYPTION_ENABLED(sbi) ((sbi)->s_dummy_enc_ctx.ctx != NULL)
 #else
 #define DUMMY_ENCRYPTION_ENABLED(sbi) (0)
 #endif
@@ -1551,6 +1549,9 @@ struct ext4_sb_info {
 	struct ratelimit_state s_warning_ratelimit_state;
 	struct ratelimit_state s_msg_ratelimit_state;
 
+	/* Encryption context for '-o test_dummy_encryption' */
+	struct fscrypt_dummy_context s_dummy_enc_ctx;
+
 	/*
 	 * Barrier between writepages ops and changing any inode's JOURNAL_DATA
 	 * or EXTENTS flag.
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index bf5fcb477f66..4a3d21972011 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -1106,6 +1106,7 @@ static void ext4_put_super(struct super_block *sb)
 		crypto_free_shash(sbi->s_chksum_driver);
 	kfree(sbi->s_blockgroup_lock);
 	fs_put_dax(sbi->s_daxdev);
+	fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
 #ifdef CONFIG_UNICODE
 	utf8_unload(sbi->s_encoding);
 #endif
@@ -1389,9 +1390,10 @@ retry:
 	return res;
 }
 
-static bool ext4_dummy_context(struct inode *inode)
+static const union fscrypt_context *
+ext4_get_dummy_context(struct super_block *sb)
 {
-	return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
+	return EXT4_SB(sb)->s_dummy_enc_ctx.ctx;
 }
 
 static bool ext4_has_stable_inodes(struct super_block *sb)
@@ -1410,7 +1412,7 @@ static const struct fscrypt_operations ext4_cryptops = {
 	.key_prefix		= "ext4:",
 	.get_context		= ext4_get_context,
 	.set_context		= ext4_set_context,
-	.dummy_context		= ext4_dummy_context,
+	.get_dummy_context	= ext4_get_dummy_context,
 	.empty_dir		= ext4_empty_dir,
 	.max_namelen		= EXT4_NAME_LEN,
 	.has_stable_inodes	= ext4_has_stable_inodes,
@@ -1605,6 +1607,7 @@ static const match_table_t tokens = {
 	{Opt_init_itable, "init_itable"},
 	{Opt_noinit_itable, "noinit_itable"},
 	{Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
+	{Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
 	{Opt_test_dummy_encryption, "test_dummy_encryption"},
 	{Opt_nombcache, "nombcache"},
 	{Opt_nombcache, "no_mbcache"},	/* for backward compatibility */
@@ -1816,7 +1819,7 @@ static const struct mount_opts {
 	{Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
 	{Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
 	{Opt_max_dir_size_kb, 0, MOPT_GTE0},
-	{Opt_test_dummy_encryption, 0, MOPT_GTE0},
+	{Opt_test_dummy_encryption, 0, MOPT_STRING},
 	{Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 	{Opt_err, 0, 0}
 };
@@ -1851,6 +1854,48 @@ static int ext4_sb_read_encoding(const struct ext4_super_block *es,
 }
 #endif
 
+static int ext4_set_test_dummy_encryption(struct super_block *sb,
+					  const char *opt,
+					  const substring_t *arg,
+					  bool is_remount)
+{
+#ifdef CONFIG_FS_ENCRYPTION
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	int err;
+
+	/*
+	 * This mount option is just for testing, and it's not worthwhile to
+	 * implement the extra complexity (e.g. RCU protection) that would be
+	 * needed to allow it to be set or changed during remount.  We do allow
+	 * it to be specified during remount, but only if there is no change.
+	 */
+	if (is_remount && !sbi->s_dummy_enc_ctx.ctx) {
+		ext4_msg(sb, KERN_WARNING,
+			 "Can't set test_dummy_encryption on remount");
+		return -1;
+	}
+	err = fscrypt_set_test_dummy_encryption(sb, arg, &sbi->s_dummy_enc_ctx);
+	if (err) {
+		if (err == -EEXIST)
+			ext4_msg(sb, KERN_WARNING,
+				 "Can't change test_dummy_encryption on remount");
+		else if (err == -EINVAL)
+			ext4_msg(sb, KERN_WARNING,
+				 "Value of option \"%s\" is unrecognized", opt);
+		else
+			ext4_msg(sb, KERN_WARNING,
+				 "Error processing option \"%s\" [%d]",
+				 opt, err);
+		return -1;
+	}
+	ext4_msg(sb, KERN_WARNING, "Test dummy encryption mode enabled");
+#else
+	ext4_msg(sb, KERN_WARNING,
+		 "Test dummy encryption mount option ignored");
+#endif
+	return 1;
+}
+
 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
 			    substring_t *args, unsigned long *journal_devnum,
 			    unsigned int *journal_ioprio, int is_remount)
@@ -2047,14 +2092,8 @@ static int handle_mount_opt(struct super_block *sb, char *opt, int token,
 		*journal_ioprio =
 			IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
 	} else if (token == Opt_test_dummy_encryption) {
-#ifdef CONFIG_FS_ENCRYPTION
-		sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
-		ext4_msg(sb, KERN_WARNING,
-			 "Test dummy encryption mode enabled");
-#else
-		ext4_msg(sb, KERN_WARNING,
-			 "Test dummy encryption mount option ignored");
-#endif
+		return ext4_set_test_dummy_encryption(sb, opt, &args[0],
+						      is_remount);
 	} else if (m->flags & MOPT_DATAJ) {
 		if (is_remount) {
 			if (!sbi->s_journal)
@@ -2311,8 +2350,8 @@ static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
 		SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
 	if (test_opt(sb, DATA_ERR_ABORT))
 		SEQ_OPTS_PUTS("data_err=abort");
-	if (DUMMY_ENCRYPTION_ENABLED(sbi))
-		SEQ_OPTS_PUTS("test_dummy_encryption");
+
+	fscrypt_show_test_dummy_encryption(seq, sep, sb);
 
 	ext4_show_quota_options(seq, sb);
 	return 0;
@@ -4780,6 +4819,7 @@ failed_mount:
 	for (i = 0; i < EXT4_MAXQUOTAS; i++)
 		kfree(get_qf_name(sb, sbi, i));
 #endif
+	fscrypt_free_dummy_context(&sbi->s_dummy_enc_ctx);
 	ext4_blkdev_remove(sbi);
 	brelse(bh);
 out_fail:
diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index 04bfaf63752c..6c9fc9e21c13 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -293,6 +293,7 @@ EXT4_ATTR_FEATURE(batched_discard);
 EXT4_ATTR_FEATURE(meta_bg_resize);
 #ifdef CONFIG_FS_ENCRYPTION
 EXT4_ATTR_FEATURE(encryption);
+EXT4_ATTR_FEATURE(test_dummy_encryption_v2);
 #endif
 #ifdef CONFIG_UNICODE
 EXT4_ATTR_FEATURE(casefold);
@@ -308,6 +309,7 @@ static struct attribute *ext4_feat_attrs[] = {
 	ATTR_LIST(meta_bg_resize),
 #ifdef CONFIG_FS_ENCRYPTION
 	ATTR_LIST(encryption),
+	ATTR_LIST(test_dummy_encryption_v2),
 #endif
 #ifdef CONFIG_UNICODE
 	ATTR_LIST(casefold),
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index ba470d5687fe..157eec348970 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -138,7 +138,7 @@ struct f2fs_mount_info {
 	int fsync_mode;			/* fsync policy */
 	int fs_mode;			/* fs mode: LFS or ADAPTIVE */
 	int bggc_mode;			/* bggc mode: off, on or sync */
-	bool test_dummy_encryption;	/* test dummy encryption */
+	struct fscrypt_dummy_context dummy_enc_ctx; /* test dummy encryption */
 	block_t unusable_cap;		/* Amount of space allowed to be
 					 * unusable when disabling checkpoint
 					 */
@@ -1259,7 +1259,7 @@ enum fsync_mode {
 
 #ifdef CONFIG_FS_ENCRYPTION
 #define DUMMY_ENCRYPTION_ENABLED(sbi) \
-			(unlikely(F2FS_OPTION(sbi).test_dummy_encryption))
+	(unlikely(F2FS_OPTION(sbi).dummy_enc_ctx.ctx != NULL))
 #else
 #define DUMMY_ENCRYPTION_ENABLED(sbi) (0)
 #endif
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index f2dfc21c6abb..8a9955902d84 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -202,6 +202,7 @@ static match_table_t f2fs_tokens = {
 	{Opt_whint, "whint_mode=%s"},
 	{Opt_alloc, "alloc_mode=%s"},
 	{Opt_fsync, "fsync_mode=%s"},
+	{Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
 	{Opt_test_dummy_encryption, "test_dummy_encryption"},
 	{Opt_checkpoint_disable, "checkpoint=disable"},
 	{Opt_checkpoint_disable_cap, "checkpoint=disable:%u"},
@@ -394,7 +395,52 @@ static int f2fs_check_quota_options(struct f2fs_sb_info *sbi)
 }
 #endif
 
-static int parse_options(struct super_block *sb, char *options)
+static int f2fs_set_test_dummy_encryption(struct super_block *sb,
+					  const char *opt,
+					  const substring_t *arg,
+					  bool is_remount)
+{
+	struct f2fs_sb_info *sbi = F2FS_SB(sb);
+#ifdef CONFIG_FS_ENCRYPTION
+	int err;
+
+	if (!f2fs_sb_has_encrypt(sbi)) {
+		f2fs_err(sbi, "Encrypt feature is off");
+		return -EINVAL;
+	}
+
+	/*
+	 * This mount option is just for testing, and it's not worthwhile to
+	 * implement the extra complexity (e.g. RCU protection) that would be
+	 * needed to allow it to be set or changed during remount.  We do allow
+	 * it to be specified during remount, but only if there is no change.
+	 */
+	if (is_remount && !F2FS_OPTION(sbi).dummy_enc_ctx.ctx) {
+		f2fs_warn(sbi, "Can't set test_dummy_encryption on remount");
+		return -EINVAL;
+	}
+	err = fscrypt_set_test_dummy_encryption(
+		sb, arg, &F2FS_OPTION(sbi).dummy_enc_ctx);
+	if (err) {
+		if (err == -EEXIST)
+			f2fs_warn(sbi,
+				  "Can't change test_dummy_encryption on remount");
+		else if (err == -EINVAL)
+			f2fs_warn(sbi, "Value of option \"%s\" is unrecognized",
+				  opt);
+		else
+			f2fs_warn(sbi, "Error processing option \"%s\" [%d]",
+				  opt, err);
+		return -EINVAL;
+	}
+	f2fs_warn(sbi, "Test dummy encryption mode enabled");
+#else
+	f2fs_warn(sbi, "Test dummy encryption mount option ignored");
+#endif
+	return 0;
+}
+
+static int parse_options(struct super_block *sb, char *options, bool is_remount)
 {
 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
 	substring_t args[MAX_OPT_ARGS];
@@ -403,9 +449,7 @@ static int parse_options(struct super_block *sb, char *options)
 	int arg = 0, ext_cnt;
 	kuid_t uid;
 	kgid_t gid;
-#ifdef CONFIG_QUOTA
 	int ret;
-#endif
 
 	if (!options)
 		return 0;
@@ -778,17 +822,10 @@ static int parse_options(struct super_block *sb, char *options)
 			kvfree(name);
 			break;
 		case Opt_test_dummy_encryption:
-#ifdef CONFIG_FS_ENCRYPTION
-			if (!f2fs_sb_has_encrypt(sbi)) {
-				f2fs_err(sbi, "Encrypt feature is off");
-				return -EINVAL;
-			}
-
-			F2FS_OPTION(sbi).test_dummy_encryption = true;
-			f2fs_info(sbi, "Test dummy encryption mode enabled");
-#else
-			f2fs_info(sbi, "Test dummy encryption mount option ignored");
-#endif
+			ret = f2fs_set_test_dummy_encryption(sb, p, &args[0],
+							     is_remount);
+			if (ret)
+				return ret;
 			break;
 		case Opt_checkpoint_disable_cap_perc:
 			if (args->from && match_int(args, &arg))
@@ -1213,6 +1250,7 @@ static void f2fs_put_super(struct super_block *sb)
 	for (i = 0; i < MAXQUOTAS; i++)
 		kvfree(F2FS_OPTION(sbi).s_qf_names[i]);
 #endif
+	fscrypt_free_dummy_context(&F2FS_OPTION(sbi).dummy_enc_ctx);
 	destroy_percpu_info(sbi);
 	for (i = 0; i < NR_PAGE_TYPE; i++)
 		kvfree(sbi->write_io[i]);
@@ -1543,10 +1581,8 @@ static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
 		seq_printf(seq, ",whint_mode=%s", "user-based");
 	else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS)
 		seq_printf(seq, ",whint_mode=%s", "fs-based");
-#ifdef CONFIG_FS_ENCRYPTION
-	if (F2FS_OPTION(sbi).test_dummy_encryption)
-		seq_puts(seq, ",test_dummy_encryption");
-#endif
+
+	fscrypt_show_test_dummy_encryption(seq, ',', sbi->sb);
 
 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_DEFAULT)
 		seq_printf(seq, ",alloc_mode=%s", "default");
@@ -1575,7 +1611,6 @@ static void default_options(struct f2fs_sb_info *sbi)
 	F2FS_OPTION(sbi).whint_mode = WHINT_MODE_OFF;
 	F2FS_OPTION(sbi).alloc_mode = ALLOC_MODE_DEFAULT;
 	F2FS_OPTION(sbi).fsync_mode = FSYNC_MODE_POSIX;
-	F2FS_OPTION(sbi).test_dummy_encryption = false;
 	F2FS_OPTION(sbi).s_resuid = make_kuid(&init_user_ns, F2FS_DEF_RESUID);
 	F2FS_OPTION(sbi).s_resgid = make_kgid(&init_user_ns, F2FS_DEF_RESGID);
 	F2FS_OPTION(sbi).compress_algorithm = COMPRESS_LZ4;
@@ -1734,7 +1769,7 @@ static int f2fs_remount(struct super_block *sb, int *flags, char *data)
 	default_options(sbi);
 
 	/* parse mount options */
-	err = parse_options(sb, data);
+	err = parse_options(sb, data, true);
 	if (err)
 		goto restore_opts;
 	checkpoint_changed =
@@ -2410,9 +2445,10 @@ static int f2fs_set_context(struct inode *inode, const void *ctx, size_t len,
 				ctx, len, fs_data, XATTR_CREATE);
 }
 
-static bool f2fs_dummy_context(struct inode *inode)
+static const union fscrypt_context *
+f2fs_get_dummy_context(struct super_block *sb)
 {
-	return DUMMY_ENCRYPTION_ENABLED(F2FS_I_SB(inode));
+	return F2FS_OPTION(F2FS_SB(sb)).dummy_enc_ctx.ctx;
 }
 
 static bool f2fs_has_stable_inodes(struct super_block *sb)
@@ -2431,7 +2467,7 @@ static const struct fscrypt_operations f2fs_cryptops = {
 	.key_prefix		= "f2fs:",
 	.get_context		= f2fs_get_context,
 	.set_context		= f2fs_set_context,
-	.dummy_context		= f2fs_dummy_context,
+	.get_dummy_context	= f2fs_get_dummy_context,
 	.empty_dir		= f2fs_empty_dir,
 	.max_namelen		= F2FS_NAME_LEN,
 	.has_stable_inodes	= f2fs_has_stable_inodes,
@@ -3366,7 +3402,7 @@ try_onemore:
 		goto free_sb_buf;
 	}
 
-	err = parse_options(sb, options);
+	err = parse_options(sb, options, false);
 	if (err)
 		goto free_options;
 
@@ -3769,6 +3805,7 @@ free_options:
 	for (i = 0; i < MAXQUOTAS; i++)
 		kvfree(F2FS_OPTION(sbi).s_qf_names[i]);
 #endif
+	fscrypt_free_dummy_context(&F2FS_OPTION(sbi).dummy_enc_ctx);
 	kvfree(options);
 free_sb_buf:
 	kvfree(raw_super);
diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
index e3bbbef9b4f0..3162f46b3c9b 100644
--- a/fs/f2fs/sysfs.c
+++ b/fs/f2fs/sysfs.c
@@ -446,6 +446,7 @@ enum feat_id {
 	FEAT_SB_CHECKSUM,
 	FEAT_CASEFOLD,
 	FEAT_COMPRESSION,
+	FEAT_TEST_DUMMY_ENCRYPTION_V2,
 };
 
 static ssize_t f2fs_feature_show(struct f2fs_attr *a,
@@ -466,6 +467,7 @@ static ssize_t f2fs_feature_show(struct f2fs_attr *a,
 	case FEAT_SB_CHECKSUM:
 	case FEAT_CASEFOLD:
 	case FEAT_COMPRESSION:
+	case FEAT_TEST_DUMMY_ENCRYPTION_V2:
 		return sprintf(buf, "supported\n");
 	}
 	return 0;
@@ -563,6 +565,7 @@ F2FS_GENERAL_RO_ATTR(avg_vblocks);
 
 #ifdef CONFIG_FS_ENCRYPTION
 F2FS_FEATURE_RO_ATTR(encryption, FEAT_CRYPTO);
+F2FS_FEATURE_RO_ATTR(test_dummy_encryption_v2, FEAT_TEST_DUMMY_ENCRYPTION_V2);
 #endif
 #ifdef CONFIG_BLK_DEV_ZONED
 F2FS_FEATURE_RO_ATTR(block_zoned, FEAT_BLKZONED);
@@ -647,6 +650,7 @@ ATTRIBUTE_GROUPS(f2fs);
 static struct attribute *f2fs_feat_attrs[] = {
 #ifdef CONFIG_FS_ENCRYPTION
 	ATTR_LIST(encryption),
+	ATTR_LIST(test_dummy_encryption_v2),
 #endif
 #ifdef CONFIG_BLK_DEV_ZONED
 	ATTR_LIST(block_zoned),
diff --git a/include/linux/fscrypt.h b/include/linux/fscrypt.h
index 0e0c7ad19383..2862ca5fea33 100644
--- a/include/linux/fscrypt.h
+++ b/include/linux/fscrypt.h
@@ -15,12 +15,15 @@
 
 #include <linux/fs.h>
 #include <linux/mm.h>
+#include <linux/parser.h>
 #include <linux/slab.h>
 #include <uapi/linux/fscrypt.h>
 
 #define FS_CRYPTO_BLOCK_SIZE		16
 
+union fscrypt_context;
 struct fscrypt_info;
+struct seq_file;
 
 struct fscrypt_str {
 	unsigned char *name;
@@ -59,7 +62,8 @@ struct fscrypt_operations {
 	int (*get_context)(struct inode *inode, void *ctx, size_t len);
 	int (*set_context)(struct inode *inode, const void *ctx, size_t len,
 			   void *fs_data);
-	bool (*dummy_context)(struct inode *inode);
+	const union fscrypt_context *(*get_dummy_context)(
+		struct super_block *sb);
 	bool (*empty_dir)(struct inode *inode);
 	unsigned int max_namelen;
 	bool (*has_stable_inodes)(struct super_block *sb);
@@ -89,10 +93,12 @@ static inline bool fscrypt_needs_contents_encryption(const struct inode *inode)
 	return IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode);
 }
 
-static inline bool fscrypt_dummy_context_enabled(struct inode *inode)
+static inline const union fscrypt_context *
+fscrypt_get_dummy_context(struct super_block *sb)
 {
-	return inode->i_sb->s_cop->dummy_context &&
-		inode->i_sb->s_cop->dummy_context(inode);
+	if (!sb->s_cop->get_dummy_context)
+		return NULL;
+	return sb->s_cop->get_dummy_context(sb);
 }
 
 /*
@@ -145,6 +151,22 @@ int fscrypt_has_permitted_context(struct inode *parent, struct inode *child);
 int fscrypt_inherit_context(struct inode *parent, struct inode *child,
 			    void *fs_data, bool preload);
 
+struct fscrypt_dummy_context {
+	const union fscrypt_context *ctx;
+};
+
+int fscrypt_set_test_dummy_encryption(struct super_block *sb,
+				      const substring_t *arg,
+				      struct fscrypt_dummy_context *dummy_ctx);
+void fscrypt_show_test_dummy_encryption(struct seq_file *seq, char sep,
+					struct super_block *sb);
+static inline void
+fscrypt_free_dummy_context(struct fscrypt_dummy_context *dummy_ctx)
+{
+	kfree(dummy_ctx->ctx);
+	dummy_ctx->ctx = NULL;
+}
+
 /* keyring.c */
 void fscrypt_sb_free(struct super_block *sb);
 int fscrypt_ioctl_add_key(struct file *filp, void __user *arg);
@@ -219,9 +241,10 @@ static inline bool fscrypt_needs_contents_encryption(const struct inode *inode)
 	return false;
 }
 
-static inline bool fscrypt_dummy_context_enabled(struct inode *inode)
+static inline const union fscrypt_context *
+fscrypt_get_dummy_context(struct super_block *sb)
 {
-	return false;
+	return NULL;
 }
 
 static inline void fscrypt_handle_d_move(struct dentry *dentry)
@@ -316,6 +339,20 @@ static inline int fscrypt_inherit_context(struct inode *parent,
 	return -EOPNOTSUPP;
 }
 
+struct fscrypt_dummy_context {
+};
+
+static inline void fscrypt_show_test_dummy_encryption(struct seq_file *seq,
+						      char sep,
+						      struct super_block *sb)
+{
+}
+
+static inline void
+fscrypt_free_dummy_context(struct fscrypt_dummy_context *dummy_ctx)
+{
+}
+
 /* keyring.c */
 static inline void fscrypt_sb_free(struct super_block *sb)
 {
@@ -677,7 +714,7 @@ static inline int fscrypt_prepare_symlink(struct inode *dir,
 					  unsigned int max_len,
 					  struct fscrypt_str *disk_link)
 {
-	if (IS_ENCRYPTED(dir) || fscrypt_dummy_context_enabled(dir))
+	if (IS_ENCRYPTED(dir) || fscrypt_get_dummy_context(dir->i_sb) != NULL)
 		return __fscrypt_prepare_symlink(dir, len, max_len, disk_link);
 
 	disk_link->name = (unsigned char *)target;
-- 
cgit v1.2.3


From cfd73237722135807967f389bcbda558a60a30d6 Mon Sep 17 00:00:00 2001
From: Alex Zhuravlev <bzzz@whamcloud.com>
Date: Tue, 21 Apr 2020 10:54:07 +0300
Subject: ext4: add prefetching for block allocation bitmaps

This should significantly improve bitmap loading, especially for flex
groups as it tries to load all bitmaps within a flex.group instead of
one by one synchronously.

Prefetching is done in 8 * flex_bg groups, so it should be 8
read-ahead reads for a single allocating thread. At the end of
allocation the thread waits for read-ahead completion and initializes
buddy information so that read-aheads are not lost in case of memory
pressure.

At cr=0 the number of prefetching IOs is limited per allocation
context to prevent a situation when mballoc loads thousands of bitmaps
looking for a perfect group and ignoring groups with good chunks.

Together with the patch "ext4: limit scanning of uninitialized groups"
the mount time (which includes few tiny allocations) of a 1PB
filesystem is reduced significantly:

               0% full    50%-full unpatched    patched
  mount time       33s                9279s       563s

[ Restructured by tytso; removed the state flags in the allocation
  context, so it can be used to lazily prefetch the allocation bitmaps
  immediately after the file system is mounted.  Skip prefetching
  block groups which are uninitialized.  Finally pass in the
  REQ_RAHEAD flag to the block layer while prefetching. ]

Signed-off-by: Alex Zhuravlev <bzzz@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
---
 fs/ext4/balloc.c  |  14 ++++--
 fs/ext4/ext4.h    |   8 +++-
 fs/ext4/mballoc.c | 133 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 fs/ext4/sysfs.c   |   4 ++
 4 files changed, 153 insertions(+), 6 deletions(-)

(limited to 'fs/ext4/sysfs.c')

diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index 1ba46d87cdf1..1e2b1b4093aa 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -413,7 +413,8 @@ verified:
  * Return buffer_head on success or an ERR_PTR in case of failure.
  */
 struct buffer_head *
-ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
+ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
+			      bool ignore_locked)
 {
 	struct ext4_group_desc *desc;
 	struct ext4_sb_info *sbi = EXT4_SB(sb);
@@ -441,6 +442,12 @@ ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 		return ERR_PTR(-ENOMEM);
 	}
 
+	if (ignore_locked && buffer_locked(bh)) {
+		/* buffer under IO already, return if called for prefetching */
+		put_bh(bh);
+		return NULL;
+	}
+
 	if (bitmap_uptodate(bh))
 		goto verify;
 
@@ -490,7 +497,8 @@ ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
 	trace_ext4_read_block_bitmap_load(sb, block_group);
 	bh->b_end_io = ext4_end_bitmap_read;
 	get_bh(bh);
-	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
+	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO |
+		  (ignore_locked ? REQ_RAHEAD : 0), bh);
 	return bh;
 verify:
 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
@@ -534,7 +542,7 @@ ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
 	struct buffer_head *bh;
 	int err;
 
-	bh = ext4_read_block_bitmap_nowait(sb, block_group);
+	bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
 	if (IS_ERR(bh))
 		return bh;
 	err = ext4_wait_block_bitmap(sb, block_group, bh);
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index dd5e6c645442..4fba138b8722 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1521,6 +1521,8 @@ struct ext4_sb_info {
 	/* where last allocation was done - for stream allocation */
 	unsigned long s_mb_last_group;
 	unsigned long s_mb_last_start;
+	unsigned int s_mb_prefetch;
+	unsigned int s_mb_prefetch_limit;
 
 	/* stats for buddy allocator */
 	atomic_t s_bal_reqs;	/* number of reqs with len > 1 */
@@ -2462,7 +2464,8 @@ extern struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb,
 extern int ext4_should_retry_alloc(struct super_block *sb, int *retries);
 
 extern struct buffer_head *ext4_read_block_bitmap_nowait(struct super_block *sb,
-						ext4_group_t block_group);
+						ext4_group_t block_group,
+						bool ignore_locked);
 extern int ext4_wait_block_bitmap(struct super_block *sb,
 				  ext4_group_t block_group,
 				  struct buffer_head *bh);
@@ -3161,6 +3164,7 @@ struct ext4_group_info {
 	(1 << EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT)
 #define EXT4_GROUP_INFO_IBITMAP_CORRUPT		\
 	(1 << EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT)
+#define EXT4_GROUP_INFO_BBITMAP_READ_BIT	4
 
 #define EXT4_MB_GRP_NEED_INIT(grp)	\
 	(test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state)))
@@ -3175,6 +3179,8 @@ struct ext4_group_info {
 	(set_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, &((grp)->bb_state)))
 #define EXT4_MB_GRP_CLEAR_TRIMMED(grp)	\
 	(clear_bit(EXT4_GROUP_INFO_WAS_TRIMMED_BIT, &((grp)->bb_state)))
+#define EXT4_MB_GRP_TEST_AND_SET_READ(grp)	\
+	(test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_READ_BIT, &((grp)->bb_state)))
 
 #define EXT4_MAX_CONTENTION		8
 #define EXT4_CONTENTION_THRESHOLD	2
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 6dc2c6c535ef..c350ae9cb0d9 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -922,7 +922,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp)
 			bh[i] = NULL;
 			continue;
 		}
-		bh[i] = ext4_read_block_bitmap_nowait(sb, group);
+		bh[i] = ext4_read_block_bitmap_nowait(sb, group, false);
 		if (IS_ERR(bh[i])) {
 			err = PTR_ERR(bh[i]);
 			bh[i] = NULL;
@@ -2209,12 +2209,93 @@ out:
 	return ret;
 }
 
+/*
+ * Start prefetching @nr block bitmaps starting at @group.
+ * Return the next group which needs to be prefetched.
+ */
+static ext4_group_t
+ext4_mb_prefetch(struct super_block *sb, ext4_group_t group,
+		 unsigned int nr, int *cnt)
+{
+	ext4_group_t ngroups = ext4_get_groups_count(sb);
+	struct buffer_head *bh;
+	struct blk_plug plug;
+
+	blk_start_plug(&plug);
+	while (nr-- > 0) {
+		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group,
+								  NULL);
+		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+
+		/*
+		 * Prefetch block groups with free blocks; but don't
+		 * bother if it is marked uninitialized on disk, since
+		 * it won't require I/O to read.  Also only try to
+		 * prefetch once, so we avoid getblk() call, which can
+		 * be expensive.
+		 */
+		if (!EXT4_MB_GRP_TEST_AND_SET_READ(grp) &&
+		    EXT4_MB_GRP_NEED_INIT(grp) &&
+		    ext4_free_group_clusters(sb, gdp) > 0 &&
+		    !(ext4_has_group_desc_csum(sb) &&
+		      (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)))) {
+			bh = ext4_read_block_bitmap_nowait(sb, group, true);
+			if (bh && !IS_ERR(bh)) {
+				if (!buffer_uptodate(bh) && cnt)
+					(*cnt)++;
+				brelse(bh);
+			}
+		}
+		if (++group >= ngroups)
+			group = 0;
+	}
+	blk_finish_plug(&plug);
+	return group;
+}
+
+/*
+ * Prefetching reads the block bitmap into the buffer cache; but we
+ * need to make sure that the buddy bitmap in the page cache has been
+ * initialized.  Note that ext4_mb_init_group() will block if the I/O
+ * is not yet completed, or indeed if it was not initiated by
+ * ext4_mb_prefetch did not start the I/O.
+ *
+ * TODO: We should actually kick off the buddy bitmap setup in a work
+ * queue when the buffer I/O is completed, so that we don't block
+ * waiting for the block allocation bitmap read to finish when
+ * ext4_mb_prefetch_fini is called from ext4_mb_regular_allocator().
+ */
+static void
+ext4_mb_prefetch_fini(struct super_block *sb, ext4_group_t group,
+		      unsigned int nr)
+{
+	while (nr-- > 0) {
+		struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group,
+								  NULL);
+		struct ext4_group_info *grp = ext4_get_group_info(sb, group);
+
+		if (!group)
+			group = ext4_get_groups_count(sb);
+		group--;
+		grp = ext4_get_group_info(sb, group);
+
+		if (EXT4_MB_GRP_NEED_INIT(grp) &&
+		    ext4_free_group_clusters(sb, gdp) > 0 &&
+		    !(ext4_has_group_desc_csum(sb) &&
+		      (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)))) {
+			if (ext4_mb_init_group(sb, group, GFP_NOFS))
+				break;
+		}
+	}
+}
+
 static noinline_for_stack int
 ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
 {
-	ext4_group_t ngroups, group, i;
+	ext4_group_t prefetch_grp = 0, ngroups, group, i;
 	int cr = -1;
 	int err = 0, first_err = 0;
+	unsigned int nr = 0, prefetch_ios = 0;
 	struct ext4_sb_info *sbi;
 	struct super_block *sb;
 	struct ext4_buddy e4b;
@@ -2282,6 +2363,7 @@ repeat:
 		 * from the goal value specified
 		 */
 		group = ac->ac_g_ex.fe_group;
+		prefetch_grp = group;
 
 		for (i = 0; i < ngroups; group++, i++) {
 			int ret = 0;
@@ -2293,6 +2375,29 @@ repeat:
 			if (group >= ngroups)
 				group = 0;
 
+			/*
+			 * Batch reads of the block allocation bitmaps
+			 * to get multiple READs in flight; limit
+			 * prefetching at cr=0/1, otherwise mballoc can
+			 * spend a lot of time loading imperfect groups
+			 */
+			if ((prefetch_grp == group) &&
+			    (cr > 1 ||
+			     prefetch_ios < sbi->s_mb_prefetch_limit)) {
+				unsigned int curr_ios = prefetch_ios;
+
+				nr = sbi->s_mb_prefetch;
+				if (ext4_has_feature_flex_bg(sb)) {
+					nr = (group / sbi->s_mb_prefetch) *
+						sbi->s_mb_prefetch;
+					nr = nr + sbi->s_mb_prefetch - group;
+				}
+				prefetch_grp = ext4_mb_prefetch(sb, group,
+							nr, &prefetch_ios);
+				if (prefetch_ios == curr_ios)
+					nr = 0;
+			}
+
 			/* This now checks without needing the buddy page */
 			ret = ext4_mb_good_group_nolock(ac, group, cr);
 			if (ret <= 0) {
@@ -2367,6 +2472,10 @@ out:
 	mb_debug(sb, "Best len %d, origin len %d, ac_status %u, ac_flags 0x%x, cr %d ret %d\n",
 		 ac->ac_b_ex.fe_len, ac->ac_o_ex.fe_len, ac->ac_status,
 		 ac->ac_flags, cr, err);
+
+	if (nr)
+		ext4_mb_prefetch_fini(sb, prefetch_grp, nr);
+
 	return err;
 }
 
@@ -2613,6 +2722,26 @@ static int ext4_mb_init_backend(struct super_block *sb)
 			goto err_freebuddy;
 	}
 
+	if (ext4_has_feature_flex_bg(sb)) {
+		/* a single flex group is supposed to be read by a single IO */
+		sbi->s_mb_prefetch = 1 << sbi->s_es->s_log_groups_per_flex;
+		sbi->s_mb_prefetch *= 8; /* 8 prefetch IOs in flight at most */
+	} else {
+		sbi->s_mb_prefetch = 32;
+	}
+	if (sbi->s_mb_prefetch > ext4_get_groups_count(sb))
+		sbi->s_mb_prefetch = ext4_get_groups_count(sb);
+	/* now many real IOs to prefetch within a single allocation at cr=0
+	 * given cr=0 is an CPU-related optimization we shouldn't try to
+	 * load too many groups, at some point we should start to use what
+	 * we've got in memory.
+	 * with an average random access time 5ms, it'd take a second to get
+	 * 200 groups (* N with flex_bg), so let's make this limit 4
+	 */
+	sbi->s_mb_prefetch_limit = sbi->s_mb_prefetch * 4;
+	if (sbi->s_mb_prefetch_limit > ext4_get_groups_count(sb))
+		sbi->s_mb_prefetch_limit = ext4_get_groups_count(sb);
+
 	return 0;
 
 err_freebuddy:
diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index 6c9fc9e21c13..31e0db726d21 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -240,6 +240,8 @@ EXT4_RO_ATTR_ES_STRING(last_error_func, s_last_error_func, 32);
 EXT4_ATTR(first_error_time, 0444, first_error_time);
 EXT4_ATTR(last_error_time, 0444, last_error_time);
 EXT4_ATTR(journal_task, 0444, journal_task);
+EXT4_RW_ATTR_SBI_UI(mb_prefetch, s_mb_prefetch);
+EXT4_RW_ATTR_SBI_UI(mb_prefetch_limit, s_mb_prefetch_limit);
 
 static unsigned int old_bump_val = 128;
 EXT4_ATTR_PTR(max_writeback_mb_bump, 0444, pointer_ui, &old_bump_val);
@@ -283,6 +285,8 @@ static struct attribute *ext4_attrs[] = {
 #ifdef CONFIG_EXT4_DEBUG
 	ATTR_LIST(simulate_fail),
 #endif
+	ATTR_LIST(mb_prefetch),
+	ATTR_LIST(mb_prefetch_limit),
 	NULL,
 };
 ATTRIBUTE_GROUPS(ext4);
-- 
cgit v1.2.3


From 1cf006ed19a887c22e085189c8b4a3cbf60d2246 Mon Sep 17 00:00:00 2001
From: Dmitry Monakhov <dmtrmonakhov@yandex-team.ru>
Date: Sat, 25 Jul 2020 12:33:13 +0000
Subject: ext4: export msg_count and warning_count via sysfs

This numbers can be analized by system automation similar to errors_count.
In ideal world it would be nice to have separate counters for different
log-levels, but this makes this patch too intrusive.

Signed-off-by: Dmitry Monakhov <dmtrmonakhov@yandex-team.ru>
Link: https://lore.kernel.org/r/20200725123313.4467-1-dmtrmonakhov@yandex-team.ru
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
---
 fs/ext4/ext4.h  |  2 ++
 fs/ext4/super.c | 12 +++++++++---
 fs/ext4/sysfs.c |  7 +++++++
 3 files changed, 18 insertions(+), 3 deletions(-)

(limited to 'fs/ext4/sysfs.c')

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 32260ceb9cb5..c73b9a1959a9 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1591,6 +1591,8 @@ struct ext4_sb_info {
 	struct ratelimit_state s_err_ratelimit_state;
 	struct ratelimit_state s_warning_ratelimit_state;
 	struct ratelimit_state s_msg_ratelimit_state;
+	atomic_t s_warning_count;
+	atomic_t s_msg_count;
 
 	/* Encryption context for '-o test_dummy_encryption' */
 	struct fscrypt_dummy_context s_dummy_enc_ctx;
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index a6de88fc7060..7deb2b75fcd0 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -744,6 +744,7 @@ void __ext4_msg(struct super_block *sb,
 	struct va_format vaf;
 	va_list args;
 
+	atomic_inc(&EXT4_SB(sb)->s_msg_count);
 	if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 		return;
 
@@ -754,9 +755,12 @@ void __ext4_msg(struct super_block *sb,
 	va_end(args);
 }
 
-#define ext4_warning_ratelimit(sb)					\
-		___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),	\
-			     "EXT4-fs warning")
+static int ext4_warning_ratelimit(struct super_block *sb)
+{
+	atomic_inc(&EXT4_SB(sb)->s_warning_count);
+	return ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state),
+			    "EXT4-fs warning");
+}
 
 void __ext4_warning(struct super_block *sb, const char *function,
 		    unsigned int line, const char *fmt, ...)
@@ -4840,6 +4844,8 @@ no_journal:
 	ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
 	ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
 	ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
+	atomic_set(&sbi->s_warning_count, 0);
+	atomic_set(&sbi->s_msg_count, 0);
 
 	kfree(orig_data);
 	return 0;
diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index 31e0db726d21..7fee11cc30e7 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -189,6 +189,9 @@ static struct ext4_attr ext4_attr_##_name = {			\
 #define EXT4_RW_ATTR_SBI_UL(_name,_elname)	\
 	EXT4_ATTR_OFFSET(_name, 0644, pointer_ul, ext4_sb_info, _elname)
 
+#define EXT4_RO_ATTR_SBI_ATOMIC(_name,_elname)	\
+	EXT4_ATTR_OFFSET(_name, 0444, pointer_atomic, ext4_sb_info, _elname)
+
 #define EXT4_ATTR_PTR(_name,_mode,_id,_ptr) \
 static struct ext4_attr ext4_attr_##_name = {			\
 	.attr = {.name = __stringify(_name), .mode = _mode },	\
@@ -226,6 +229,8 @@ EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
 #ifdef CONFIG_EXT4_DEBUG
 EXT4_RW_ATTR_SBI_UL(simulate_fail, s_simulate_fail);
 #endif
+EXT4_RO_ATTR_SBI_ATOMIC(warning_count, s_warning_count);
+EXT4_RO_ATTR_SBI_ATOMIC(msg_count, s_msg_count);
 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
 EXT4_RO_ATTR_ES_U8(first_error_errcode, s_first_error_errcode);
 EXT4_RO_ATTR_ES_U8(last_error_errcode, s_last_error_errcode);
@@ -269,6 +274,8 @@ static struct attribute *ext4_attrs[] = {
 	ATTR_LIST(msg_ratelimit_interval_ms),
 	ATTR_LIST(msg_ratelimit_burst),
 	ATTR_LIST(errors_count),
+	ATTR_LIST(warning_count),
+	ATTR_LIST(msg_count),
 	ATTR_LIST(first_error_ino),
 	ATTR_LIST(last_error_ino),
 	ATTR_LIST(first_error_block),
-- 
cgit v1.2.3


From 27bc446e2def38db3244a6eb4bb1d6312936610a Mon Sep 17 00:00:00 2001
From: brookxu <brookxu.cn@gmail.com>
Date: Mon, 17 Aug 2020 15:36:15 +0800
Subject: ext4: limit the length of per-inode prealloc list
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

In the scenario of writing sparse files, the per-inode prealloc list may
be very long, resulting in high overhead for ext4_mb_use_preallocated().
To circumvent this problem, we limit the maximum length of per-inode
prealloc list to 512 and allow users to modify it.

After patching, we observed that the sys ratio of cpu has dropped, and
the system throughput has increased significantly. We created a process
to write the sparse file, and the running time of the process on the
fixed kernel was significantly reduced, as follows:

Running time on unfixed kernel：
[root@TENCENT64 ~]# time taskset 0x01 ./sparse /data1/sparce.dat
real    0m2.051s
user    0m0.008s
sys     0m2.026s

Running time on fixed kernel：
[root@TENCENT64 ~]# time taskset 0x01 ./sparse /data1/sparce.dat
real    0m0.471s
user    0m0.004s
sys     0m0.395s

Signed-off-by: Chunguang Xu <brookxu@tencent.com>
Link: https://lore.kernel.org/r/d7a98178-056b-6db5-6bce-4ead23f4a257@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
---
 Documentation/admin-guide/ext4.rst |  3 ++
 fs/ext4/ext4.h                     |  4 ++-
 fs/ext4/extents.c                  | 10 +++---
 fs/ext4/file.c                     |  2 +-
 fs/ext4/indirect.c                 |  2 +-
 fs/ext4/inode.c                    |  6 ++--
 fs/ext4/ioctl.c                    |  2 +-
 fs/ext4/mballoc.c                  | 74 +++++++++++++++++++++++++++++++++-----
 fs/ext4/mballoc.h                  |  4 +++
 fs/ext4/move_extent.c              |  4 +--
 fs/ext4/super.c                    |  3 +-
 fs/ext4/sysfs.c                    |  2 ++
 include/trace/events/ext4.h        | 17 +++++----
 13 files changed, 104 insertions(+), 29 deletions(-)

(limited to 'fs/ext4/sysfs.c')

diff --git a/Documentation/admin-guide/ext4.rst b/Documentation/admin-guide/ext4.rst
index 7fc6a72920c9..59bcc4a92602 100644
--- a/Documentation/admin-guide/ext4.rst
+++ b/Documentation/admin-guide/ext4.rst
@@ -482,6 +482,9 @@ Files in /sys/fs/ext4/<devname>:
         multiple of this tuning parameter if the stripe size is not set in the
         ext4 superblock
 
+  mb_max_inode_prealloc
+        The maximum length of per-inode ext4_prealloc_space list.
+
   mb_max_to_scan
         The maximum number of extents the multiblock allocator will search to
         find the best extent.
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 71b4370a3f91..523e00d7b392 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1070,6 +1070,7 @@ struct ext4_inode_info {
 	struct timespec64 i_crtime;
 
 	/* mballoc */
+	atomic_t i_prealloc_active;
 	struct list_head i_prealloc_list;
 	spinlock_t i_prealloc_lock;
 
@@ -1518,6 +1519,7 @@ struct ext4_sb_info {
 	unsigned int s_mb_stats;
 	unsigned int s_mb_order2_reqs;
 	unsigned int s_mb_group_prealloc;
+	unsigned int s_mb_max_inode_prealloc;
 	unsigned int s_max_dir_size_kb;
 	/* where last allocation was done - for stream allocation */
 	unsigned long s_mb_last_group;
@@ -2682,7 +2684,7 @@ extern int ext4_mb_release(struct super_block *);
 extern ext4_fsblk_t ext4_mb_new_blocks(handle_t *,
 				struct ext4_allocation_request *, int *);
 extern int ext4_mb_reserve_blocks(struct super_block *, int);
-extern void ext4_discard_preallocations(struct inode *);
+extern void ext4_discard_preallocations(struct inode *, unsigned int);
 extern int __init ext4_init_mballoc(void);
 extern void ext4_exit_mballoc(void);
 extern ext4_group_t ext4_mb_prefetch(struct super_block *sb,
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 0eea09aa0f26..a0481582187a 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -100,7 +100,7 @@ static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
 	 * i_mutex. So we can safely drop the i_data_sem here.
 	 */
 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 	up_write(&EXT4_I(inode)->i_data_sem);
 	*dropped = 1;
 	return 0;
@@ -4266,7 +4266,7 @@ got_allocated_blocks:
 			 * not a good idea to call discard here directly,
 			 * but otherwise we'd need to call it every free().
 			 */
-			ext4_discard_preallocations(inode);
+			ext4_discard_preallocations(inode, 0);
 			if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
 				fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
 			ext4_free_blocks(handle, inode, NULL, newblock,
@@ -5293,7 +5293,7 @@ static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
 	}
 
 	down_write(&EXT4_I(inode)->i_data_sem);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	ret = ext4_es_remove_extent(inode, punch_start,
 				    EXT_MAX_BLOCKS - punch_start);
@@ -5307,7 +5307,7 @@ static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
 		up_write(&EXT4_I(inode)->i_data_sem);
 		goto out_stop;
 	}
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
 				     punch_stop - punch_start, SHIFT_LEFT);
@@ -5439,7 +5439,7 @@ static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
 		goto out_stop;
 
 	down_write(&EXT4_I(inode)->i_data_sem);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
 	if (IS_ERR(path)) {
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 7a2720517bbb..e608ce3fb535 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -147,7 +147,7 @@ static int ext4_release_file(struct inode *inode, struct file *filp)
 			(atomic_read(&inode->i_writecount) == 1) &&
 			!EXT4_I(inode)->i_reserved_data_blocks) {
 		down_write(&EXT4_I(inode)->i_data_sem);
-		ext4_discard_preallocations(inode);
+		ext4_discard_preallocations(inode, 0);
 		up_write(&EXT4_I(inode)->i_data_sem);
 	}
 	if (is_dx(inode) && filp->private_data)
diff --git a/fs/ext4/indirect.c b/fs/ext4/indirect.c
index 433ca8415c5a..80c9f33800be 100644
--- a/fs/ext4/indirect.c
+++ b/fs/ext4/indirect.c
@@ -696,7 +696,7 @@ static int ext4_ind_trunc_restart_fn(handle_t *handle, struct inode *inode,
 	 * i_mutex. So we can safely drop the i_data_sem here.
 	 */
 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 	up_write(&EXT4_I(inode)->i_data_sem);
 	*dropped = 1;
 	return 0;
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 0b07576af3bf..77543f988258 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -383,7 +383,7 @@ void ext4_da_update_reserve_space(struct inode *inode,
 	 */
 	if ((ei->i_reserved_data_blocks == 0) &&
 	    !inode_is_open_for_write(inode))
-		ext4_discard_preallocations(inode);
+		ext4_discard_preallocations(inode, 0);
 }
 
 static int __check_block_validity(struct inode *inode, const char *func,
@@ -4055,7 +4055,7 @@ int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
 	if (stop_block > first_block) {
 
 		down_write(&EXT4_I(inode)->i_data_sem);
-		ext4_discard_preallocations(inode);
+		ext4_discard_preallocations(inode, 0);
 
 		ret = ext4_es_remove_extent(inode, first_block,
 					    stop_block - first_block);
@@ -4210,7 +4210,7 @@ int ext4_truncate(struct inode *inode)
 
 	down_write(&EXT4_I(inode)->i_data_sem);
 
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
 		err = ext4_ext_truncate(handle, inode);
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index 6e70a63dcca7..36eca3bc036a 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -202,7 +202,7 @@ static long swap_inode_boot_loader(struct super_block *sb,
 	reset_inode_seed(inode);
 	reset_inode_seed(inode_bl);
 
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 
 	err = ext4_mark_inode_dirty(handle, inode);
 	if (err < 0) {
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 45ac6088b4ac..132c118d12e1 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -2878,6 +2878,7 @@ int ext4_mb_init(struct super_block *sb)
 	sbi->s_mb_stats = MB_DEFAULT_STATS;
 	sbi->s_mb_stream_request = MB_DEFAULT_STREAM_THRESHOLD;
 	sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
+	sbi->s_mb_max_inode_prealloc = MB_DEFAULT_MAX_INODE_PREALLOC;
 	/*
 	 * The default group preallocation is 512, which for 4k block
 	 * sizes translates to 2 megabytes.  However for bigalloc file
@@ -3816,6 +3817,26 @@ void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
 	mb_debug(sb, "preallocated %d for group %u\n", preallocated, group);
 }
 
+static void ext4_mb_mark_pa_deleted(struct super_block *sb,
+				    struct ext4_prealloc_space *pa)
+{
+	struct ext4_inode_info *ei;
+
+	if (pa->pa_deleted) {
+		ext4_warning(sb, "deleted pa, type:%d, pblk:%llu, lblk:%u, len:%d\n",
+			     pa->pa_type, pa->pa_pstart, pa->pa_lstart,
+			     pa->pa_len);
+		return;
+	}
+
+	pa->pa_deleted = 1;
+
+	if (pa->pa_type == MB_INODE_PA) {
+		ei = EXT4_I(pa->pa_inode);
+		atomic_dec(&ei->i_prealloc_active);
+	}
+}
+
 static void ext4_mb_pa_callback(struct rcu_head *head)
 {
 	struct ext4_prealloc_space *pa;
@@ -3848,7 +3869,7 @@ static void ext4_mb_put_pa(struct ext4_allocation_context *ac,
 		return;
 	}
 
-	pa->pa_deleted = 1;
+	ext4_mb_mark_pa_deleted(sb, pa);
 	spin_unlock(&pa->pa_lock);
 
 	grp_blk = pa->pa_pstart;
@@ -3972,6 +3993,7 @@ ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
 	spin_lock(pa->pa_obj_lock);
 	list_add_rcu(&pa->pa_inode_list, &ei->i_prealloc_list);
 	spin_unlock(pa->pa_obj_lock);
+	atomic_inc(&ei->i_prealloc_active);
 }
 
 /*
@@ -4182,7 +4204,7 @@ repeat:
 		}
 
 		/* seems this one can be freed ... */
-		pa->pa_deleted = 1;
+		ext4_mb_mark_pa_deleted(sb, pa);
 
 		/* we can trust pa_free ... */
 		free += pa->pa_free;
@@ -4245,7 +4267,7 @@ out_dbg:
  *
  * FIXME!! Make sure it is valid at all the call sites
  */
-void ext4_discard_preallocations(struct inode *inode)
+void ext4_discard_preallocations(struct inode *inode, unsigned int needed)
 {
 	struct ext4_inode_info *ei = EXT4_I(inode);
 	struct super_block *sb = inode->i_sb;
@@ -4263,15 +4285,19 @@ void ext4_discard_preallocations(struct inode *inode)
 
 	mb_debug(sb, "discard preallocation for inode %lu\n",
 		 inode->i_ino);
-	trace_ext4_discard_preallocations(inode);
+	trace_ext4_discard_preallocations(inode,
+			atomic_read(&ei->i_prealloc_active), needed);
 
 	INIT_LIST_HEAD(&list);
 
+	if (needed == 0)
+		needed = UINT_MAX;
+
 repeat:
 	/* first, collect all pa's in the inode */
 	spin_lock(&ei->i_prealloc_lock);
-	while (!list_empty(&ei->i_prealloc_list)) {
-		pa = list_entry(ei->i_prealloc_list.next,
+	while (!list_empty(&ei->i_prealloc_list) && needed) {
+		pa = list_entry(ei->i_prealloc_list.prev,
 				struct ext4_prealloc_space, pa_inode_list);
 		BUG_ON(pa->pa_obj_lock != &ei->i_prealloc_lock);
 		spin_lock(&pa->pa_lock);
@@ -4288,10 +4314,11 @@ repeat:
 
 		}
 		if (pa->pa_deleted == 0) {
-			pa->pa_deleted = 1;
+			ext4_mb_mark_pa_deleted(sb, pa);
 			spin_unlock(&pa->pa_lock);
 			list_del_rcu(&pa->pa_inode_list);
 			list_add(&pa->u.pa_tmp_list, &list);
+			needed--;
 			continue;
 		}
 
@@ -4592,7 +4619,7 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
 		BUG_ON(pa->pa_type != MB_GROUP_PA);
 
 		/* seems this one can be freed ... */
-		pa->pa_deleted = 1;
+		ext4_mb_mark_pa_deleted(sb, pa);
 		spin_unlock(&pa->pa_lock);
 
 		list_del_rcu(&pa->pa_inode_list);
@@ -4690,11 +4717,30 @@ static void ext4_mb_add_n_trim(struct ext4_allocation_context *ac)
 	return ;
 }
 
+/*
+ * if per-inode prealloc list is too long, trim some PA
+ */
+static void ext4_mb_trim_inode_pa(struct inode *inode)
+{
+	struct ext4_inode_info *ei = EXT4_I(inode);
+	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+	int count, delta;
+
+	count = atomic_read(&ei->i_prealloc_active);
+	delta = (sbi->s_mb_max_inode_prealloc >> 2) + 1;
+	if (count > sbi->s_mb_max_inode_prealloc + delta) {
+		count -= sbi->s_mb_max_inode_prealloc;
+		ext4_discard_preallocations(inode, count);
+	}
+}
+
 /*
  * release all resource we used in allocation
  */
 static int ext4_mb_release_context(struct ext4_allocation_context *ac)
 {
+	struct inode *inode = ac->ac_inode;
+	struct ext4_inode_info *ei = EXT4_I(inode);
 	struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
 	struct ext4_prealloc_space *pa = ac->ac_pa;
 	if (pa) {
@@ -4720,6 +4766,17 @@ static int ext4_mb_release_context(struct ext4_allocation_context *ac)
 				ext4_mb_add_n_trim(ac);
 			}
 		}
+
+		if (pa->pa_type == MB_INODE_PA) {
+			/*
+			 * treat per-inode prealloc list as a lru list, then try
+			 * to trim the least recently used PA.
+			 */
+			spin_lock(pa->pa_obj_lock);
+			list_move(&pa->pa_inode_list, &ei->i_prealloc_list);
+			spin_unlock(pa->pa_obj_lock);
+		}
+
 		ext4_mb_put_pa(ac, ac->ac_sb, pa);
 	}
 	if (ac->ac_bitmap_page)
@@ -4729,6 +4786,7 @@ static int ext4_mb_release_context(struct ext4_allocation_context *ac)
 	if (ac->ac_flags & EXT4_MB_HINT_GROUP_ALLOC)
 		mutex_unlock(&ac->ac_lg->lg_mutex);
 	ext4_mb_collect_stats(ac);
+	ext4_mb_trim_inode_pa(inode);
 	return 0;
 }
 
diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
index 6b4d17c2935d..e75b4749aa1c 100644
--- a/fs/ext4/mballoc.h
+++ b/fs/ext4/mballoc.h
@@ -73,6 +73,10 @@
  */
 #define MB_DEFAULT_GROUP_PREALLOC	512
 
+/*
+ * maximum length of inode prealloc list
+ */
+#define MB_DEFAULT_MAX_INODE_PREALLOC	512
 
 struct ext4_free_data {
 	/* this links the free block information from sb_info */
diff --git a/fs/ext4/move_extent.c b/fs/ext4/move_extent.c
index 1ed86fb6c302..0d601b822875 100644
--- a/fs/ext4/move_extent.c
+++ b/fs/ext4/move_extent.c
@@ -686,8 +686,8 @@ ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
 
 out:
 	if (*moved_len) {
-		ext4_discard_preallocations(orig_inode);
-		ext4_discard_preallocations(donor_inode);
+		ext4_discard_preallocations(orig_inode, 0);
+		ext4_discard_preallocations(donor_inode, 0);
 	}
 
 	ext4_ext_drop_refs(path);
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index daa94c7f7271..13bdddc081e0 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -1127,6 +1127,7 @@ static struct inode *ext4_alloc_inode(struct super_block *sb)
 	inode_set_iversion(&ei->vfs_inode, 1);
 	spin_lock_init(&ei->i_raw_lock);
 	INIT_LIST_HEAD(&ei->i_prealloc_list);
+	atomic_set(&ei->i_prealloc_active, 0);
 	spin_lock_init(&ei->i_prealloc_lock);
 	ext4_es_init_tree(&ei->i_es_tree);
 	rwlock_init(&ei->i_es_lock);
@@ -1220,7 +1221,7 @@ void ext4_clear_inode(struct inode *inode)
 {
 	invalidate_inode_buffers(inode);
 	clear_inode(inode);
-	ext4_discard_preallocations(inode);
+	ext4_discard_preallocations(inode, 0);
 	ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 	dquot_drop(inode);
 	if (EXT4_I(inode)->jinode) {
diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index 7fee11cc30e7..bfabb799fa45 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -218,6 +218,7 @@ EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
+EXT4_RW_ATTR_SBI_UI(mb_max_inode_prealloc, s_mb_max_inode_prealloc);
 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
 EXT4_ATTR(trigger_fs_error, 0200, trigger_test_error);
 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
@@ -264,6 +265,7 @@ static struct attribute *ext4_attrs[] = {
 	ATTR_LIST(mb_order2_req),
 	ATTR_LIST(mb_stream_req),
 	ATTR_LIST(mb_group_prealloc),
+	ATTR_LIST(mb_max_inode_prealloc),
 	ATTR_LIST(max_writeback_mb_bump),
 	ATTR_LIST(extent_max_zeroout_kb),
 	ATTR_LIST(trigger_fs_error),
diff --git a/include/trace/events/ext4.h b/include/trace/events/ext4.h
index 8008d2e116b9..4c8b99ec8606 100644
--- a/include/trace/events/ext4.h
+++ b/include/trace/events/ext4.h
@@ -746,24 +746,29 @@ TRACE_EVENT(ext4_mb_release_group_pa,
 );
 
 TRACE_EVENT(ext4_discard_preallocations,
-	TP_PROTO(struct inode *inode),
+	TP_PROTO(struct inode *inode, unsigned int len, unsigned int needed),
 
-	TP_ARGS(inode),
+	TP_ARGS(inode, len, needed),
 
 	TP_STRUCT__entry(
-		__field(	dev_t,	dev			)
-		__field(	ino_t,	ino			)
+		__field(	dev_t,		dev		)
+		__field(	ino_t,		ino		)
+		__field(	unsigned int,	len		)
+		__field(	unsigned int,	needed		)
 
 	),
 
 	TP_fast_assign(
 		__entry->dev	= inode->i_sb->s_dev;
 		__entry->ino	= inode->i_ino;
+		__entry->len	= len;
+		__entry->needed	= needed;
 	),
 
-	TP_printk("dev %d,%d ino %lu",
+	TP_printk("dev %d,%d ino %lu len: %u needed %u",
 		  MAJOR(__entry->dev), MINOR(__entry->dev),
-		  (unsigned long) __entry->ino)
+		  (unsigned long) __entry->ino, __entry->len,
+		  __entry->needed)
 );
 
 TRACE_EVENT(ext4_mb_discard_preallocations,
-- 
cgit v1.2.3