Skip to contentSkip to conversation

ext4: Prevent memory reclaim from re-entering the filesystem and deadlocking

PATCH Applied Naoto Kobayashi 3 participants 13 messages Dec 27, 2019 – Jan 17, 2020 ext4 file system

lore
Share & tools
lore
#0 lore raw

v2 cover ext4: Prevent memory reclaim from re-entering the filesystem and deadlocking

This series aims to prevent ext4_kvmalloc() from re-entering the filesystem and deadlocking. Although __vmalloc() doesn't support GFP_NOFS[1], all of the user of ext4_kvmalloc are using GFP_NOFS (e.g. fs/ext4/resize.c::add_new_gdb), causing the memory reclaim to re-enter the filesystem.

To fix this issue, use memalloc_nfs_save/restore() that get __vmalloc() to drop __GFP_FS.

[1] linux-tree/Documentation/core-api/gfp-mask-fs-io.rst

Changes since v1:
- Add Patch 1 to delete ext4_kvzvalloc() that is not used anywhere.
- Add Patch 2 to rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument since all the users of ext4_kvmalloc() are using GFP_NOFS

Thanks,

Naoto Kobayashi (3):
ext4: Delete ext4_kvzvalloc()
ext4: Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its
flags argument
ext4: Prevent ext4_kvmalloc_nofs() from re-entering the filesystem and
deadlocking

 fs/ext4/ext4.h   |  3 +--
 fs/ext4/resize.c | 10 ++++------
 fs/ext4/super.c  | 21 +++++++--------------
 fs/ext4/xattr.c  |  2 +-
 4 files changed, 13 insertions(+), 23 deletions(-)

--
2.16.6

#1 lore raw

PATCH v2 1/3 ext4: Delete ext4_kvzvalloc()

Since we're not using ext4_kvzalloc(), delete this function.

---
 fs/ext4/ext4.h  |  1 -
 fs/ext4/super.c | 10 ----------
 2 files changed, 11 deletions(-)
fs/ext4/ext4.h +0 -1
index 61987c106511..b25089e3896d 100644
@@ -2678,7 +2678,6 @@ extern int ext4_seq_options_show(struct seq_file *seq, void *offset);
 extern int ext4_calculate_overhead(struct super_block *sb);
 extern void ext4_superblock_csum_set(struct super_block *sb);
 extern void *ext4_kvmalloc(size_t size, gfp_t flags);
-extern void *ext4_kvzalloc(size_t size, gfp_t flags);
 extern int ext4_alloc_flex_bg_array(struct super_block *sb,
 				    ext4_group_t ngroup);
 extern const char *ext4_decode_error(struct super_block *sb, int errno,
fs/ext4/super.c +0 -10
index b205112ca051..83a231dedcbf 100644
@@ -214,16 +214,6 @@ void *ext4_kvmalloc(size_t size, gfp_t flags)
 	return ret;
 }
 
-void *ext4_kvzalloc(size_t size, gfp_t flags)
-{
-	void *ret;
-
-	ret = kzalloc(size, flags | __GFP_NOWARN);
-	if (!ret)
-		ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
-	return ret;
-}
-
 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 			       struct ext4_group_desc *bg)
 {

--
2.16.6

reviewer Reviewed-by on v2 1/3
#2 lore raw

Re: [PATCH v2 1/3] ext4: Delete ext4_kvzvalloc()

Quoted 3 lines · Naoto Kobayashi

Since we're not using ext4_kvzalloc(), delete this function.

Looks good to me. You can add:

Honza

Quoted 41 lines
---
 fs/ext4/ext4.h  |  1 -
 fs/ext4/super.c | 10 ----------
 2 files changed, 11 deletions(-)
fs/ext4/ext4.h +0 -1
index 61987c106511..b25089e3896d 100644
@@ -2678,7 +2678,6 @@ extern int ext4_seq_options_show(struct seq_file *seq, void *offset);
 extern int ext4_calculate_overhead(struct super_block *sb);
 extern void ext4_superblock_csum_set(struct super_block *sb);
 extern void *ext4_kvmalloc(size_t size, gfp_t flags);
-extern void *ext4_kvzalloc(size_t size, gfp_t flags);
 extern int ext4_alloc_flex_bg_array(struct super_block *sb,
 				    ext4_group_t ngroup);
 extern const char *ext4_decode_error(struct super_block *sb, int errno,
fs/ext4/super.c +0 -11
index b205112ca051..83a231dedcbf 100644
@@ -214,16 +214,6 @@ void *ext4_kvmalloc(size_t size, gfp_t flags)
 	return ret;
 }
 
-void *ext4_kvzalloc(size_t size, gfp_t flags)
-{
-	void *ret;
-
-	ret = kzalloc(size, flags | __GFP_NOWARN);
-	if (!ret)
-		ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
-	return ret;
-}
-
 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 			       struct ext4_group_desc *bg)
 {
--

2.16.6

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#4 lore raw

PATCH v2 2/3 ext4: Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument

In reply to Naoto Kobayashi

Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument, because ext4_kvmalloc() callers must be under GFP_NOFS (otherwise, they should use generic kvmalloc() helper function).

---
 fs/ext4/ext4.h   |  2 +-
 fs/ext4/resize.c | 10 ++++------
 fs/ext4/super.c  |  6 +++---
 fs/ext4/xattr.c  |  2 +-
 4 files changed, 9 insertions(+), 11 deletions(-)
fs/ext4/ext4.h +1 -1
index b25089e3896d..e1bdeffca0ad 100644
@@ -2677,7 +2677,7 @@ extern struct buffer_head *ext4_sb_bread(struct super_block *sb,
 extern int ext4_seq_options_show(struct seq_file *seq, void *offset);
 extern int ext4_calculate_overhead(struct super_block *sb);
 extern void ext4_superblock_csum_set(struct super_block *sb);
-extern void *ext4_kvmalloc(size_t size, gfp_t flags);
+extern void *ext4_kvmalloc_nofs(size_t size);
 extern int ext4_alloc_flex_bg_array(struct super_block *sb,
 				    ext4_group_t ngroup);
 extern const char *ext4_decode_error(struct super_block *sb, int errno,
fs/ext4/resize.c +4 -6
index a8c0f2b5b6e1..7998bbe66eed 100644
@@ -824,9 +824,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
 	if (unlikely(err))
 		goto errout;
 
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = ext4_kvmalloc_nofs((gdb_num + 1) *
+				     sizeof(struct buffer_head *));
 	if (!n_group_desc) {
 		err = -ENOMEM;
 		ext4_warning(sb, "not enough memory for %lu groups",
@@ -900,9 +899,8 @@ static int add_new_gdb_meta_bg(struct super_block *sb,
 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 	if (IS_ERR(gdb_bh))
 		return PTR_ERR(gdb_bh);
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = ext4_kvmalloc_nofs((gdb_num + 1) *
+				     sizeof(struct buffer_head *));
 	if (!n_group_desc) {
 		brelse(gdb_bh);
 		err = -ENOMEM;
fs/ext4/super.c +3 -3
index 83a231dedcbf..e8965aa6ecce 100644
@@ -204,13 +204,13 @@ void ext4_superblock_csum_set(struct super_block *sb)
 	es->s_checksum = ext4_superblock_csum(sb, es);
 }
 
-void *ext4_kvmalloc(size_t size, gfp_t flags)
+void *ext4_kvmalloc_nofs(size_t size)
 {
 	void *ret;
 
-	ret = kmalloc(size, flags | __GFP_NOWARN);
+	ret = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
 	if (!ret)
-		ret = __vmalloc(size, flags, PAGE_KERNEL);
+		ret = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
 	return ret;
 }
 
fs/ext4/xattr.c +1 -1
index 8966a5439a22..d5bc970ef331 100644
@@ -1456,7 +1456,7 @@ ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
 	if (!ce)
 		return NULL;
 
-	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
+	ea_data = ext4_kvmalloc_nofs(value_len);
 	if (!ea_data) {
 		mb_cache_entry_put(ea_inode_cache, ce);
 		return NULL;

--
2.16.6

reviewer
#5 lore raw

Re: [PATCH v2 2/3] ext4: Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument

Quoted 6 lines · Naoto Kobayashi

Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument, because ext4_kvmalloc() callers must be under GFP_NOFS (otherwise, they should use generic kvmalloc() helper function).

Thanks for the patch but I don't think this or the following patch is actually needed. Did you ever see the deadlock with reclaim you mention in the initial message with a recent kernel? The reason is that in all three call sites of ext4_kvmalloc() in ext4 we have a transaction started (which is the reason for GFP_NOFS there after all) but since commit 81378da64de "jbd2: mark the transaction context with the scope GFP_NOFS context" the transaction machinery takes care of updating reclaim context as needed... So I'd be almost inclined to just drop 'flags' argument from ext4_kvmalloc() instead and if we ever create a callsite for which current memalloc_nofs machinery won't be enough, I'd rather extend that than randomly sprinkle GFP_NOFS flags in the code.

Honza

Quoted 85 lines
---
 fs/ext4/ext4.h   |  2 +-
 fs/ext4/resize.c | 10 ++++------
 fs/ext4/super.c  |  6 +++---
 fs/ext4/xattr.c  |  2 +-
 4 files changed, 9 insertions(+), 11 deletions(-)
fs/ext4/ext4.h +1 -1
index b25089e3896d..e1bdeffca0ad 100644
@@ -2677,7 +2677,7 @@ extern struct buffer_head *ext4_sb_bread(struct super_block *sb,
 extern int ext4_seq_options_show(struct seq_file *seq, void *offset);
 extern int ext4_calculate_overhead(struct super_block *sb);
 extern void ext4_superblock_csum_set(struct super_block *sb);
-extern void *ext4_kvmalloc(size_t size, gfp_t flags);
+extern void *ext4_kvmalloc_nofs(size_t size);
 extern int ext4_alloc_flex_bg_array(struct super_block *sb,
 				    ext4_group_t ngroup);
 extern const char *ext4_decode_error(struct super_block *sb, int errno,
fs/ext4/resize.c +4 -6
index a8c0f2b5b6e1..7998bbe66eed 100644
@@ -824,9 +824,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
 	if (unlikely(err))
 		goto errout;
 
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = ext4_kvmalloc_nofs((gdb_num + 1) *
+				     sizeof(struct buffer_head *));
 	if (!n_group_desc) {
 		err = -ENOMEM;
 		ext4_warning(sb, "not enough memory for %lu groups",
@@ -900,9 +899,8 @@ static int add_new_gdb_meta_bg(struct super_block *sb,
 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 	if (IS_ERR(gdb_bh))
 		return PTR_ERR(gdb_bh);
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = ext4_kvmalloc_nofs((gdb_num + 1) *
+				     sizeof(struct buffer_head *));
 	if (!n_group_desc) {
 		brelse(gdb_bh);
 		err = -ENOMEM;
fs/ext4/super.c +3 -3
index 83a231dedcbf..e8965aa6ecce 100644
@@ -204,13 +204,13 @@ void ext4_superblock_csum_set(struct super_block *sb)
 	es->s_checksum = ext4_superblock_csum(sb, es);
 }
 
-void *ext4_kvmalloc(size_t size, gfp_t flags)
+void *ext4_kvmalloc_nofs(size_t size)
 {
 	void *ret;
 
-	ret = kmalloc(size, flags | __GFP_NOWARN);
+	ret = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
 	if (!ret)
-		ret = __vmalloc(size, flags, PAGE_KERNEL);
+		ret = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
 	return ret;
 }
 
fs/ext4/xattr.c +1 -2
index 8966a5439a22..d5bc970ef331 100644
@@ -1456,7 +1456,7 @@ ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
 	if (!ce)
 		return NULL;
 
-	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
+	ea_data = ext4_kvmalloc_nofs(value_len);
 	if (!ea_data) {
 		mb_cache_entry_put(ea_inode_cache, ce);
 		return NULL;
--

2.16.6

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
maintainer Applied
#6 lore raw

Re: [PATCH v2 2/3] ext4: Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument

In reply to Naoto Kobayashi
Quoted 6 lines · Naoto Kobayashi

Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument, because ext4_kvmalloc() callers must be under GFP_NOFS (otherwise, they should use generic kvmalloc() helper function).

Thanks, applied.

- Ted

reviewer
#7 lore raw

Re: [PATCH v2 2/3] ext4: Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument

Quoted 9 lines · Theodore Y. Ts'o
Quoted 6 lines · Naoto Kobayashi

Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument, because ext4_kvmalloc() callers must be under GFP_NOFS (otherwise, they should use generic kvmalloc() helper function).

Thanks, applied.

Ted, I don't think this patch is needed at all - see my email [1]. Sadly Naoto didn't reply to my question whether he really saw any deadlock / lockdep splat or whether it was just a theoretical concern he had...

Honza

[1] https://lore.kernel.org/linux-ext4/[email protected]

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
maintainer
#8 lore raw

Re: [PATCH v2 2/3] ext4: Rename ext4_kvmalloc() to ext4_kvmalloc_nofs() and drop its flags argument

Quoted 4 lines · Jan Kara

Ted, I don't think this patch is needed at all - see my email [1]. Sadly Naoto didn't reply to my question whether he really saw any deadlock / lockdep splat or whether it was just a theoretical concern he had...

Thanks, good point. So what we should do instead is just drop ext4_kvmalloc() entirely.

- Ted

maintainer
#9 lore raw

PATCH 1/1 ext4: drop ext4_kvmalloc()

As Jan pointed out[1], as of commit 81378da64de ("jbd2: mark the transaction context with the scope GFP_NOFS context") we use memalloc_nofs_{save,restore}() while a jbd2 handle is active. So ext4_kvmalloc() so we can call allocate using GFP_NOFS is no longer necessary.

[1] https://lore.kernel.org/r/[email protected]

---
 fs/ext4/ext4.h   |  1 -
 fs/ext4/resize.c | 10 ++++------
 fs/ext4/super.c  | 10 ----------
 fs/ext4/xattr.c  |  2 +-
 4 files changed, 5 insertions(+), 18 deletions(-)
fs/ext4/ext4.h +0 -1
index 5e621b0da4da..9a2ee2428ecc 100644
@@ -2740,7 +2740,6 @@ extern struct buffer_head *ext4_sb_bread(struct super_block *sb,
 extern int ext4_seq_options_show(struct seq_file *seq, void *offset);
 extern int ext4_calculate_overhead(struct super_block *sb);
 extern void ext4_superblock_csum_set(struct super_block *sb);
-extern void *ext4_kvmalloc(size_t size, gfp_t flags);
 extern int ext4_alloc_flex_bg_array(struct super_block *sb,
 				    ext4_group_t ngroup);
 extern const char *ext4_decode_error(struct super_block *sb, int errno,
fs/ext4/resize.c +4 -6
index a8c0f2b5b6e1..86a2500ed292 100644
@@ -824,9 +824,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
 	if (unlikely(err))
 		goto errout;
 
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
+				GFP_KERNEL);
 	if (!n_group_desc) {
 		err = -ENOMEM;
 		ext4_warning(sb, "not enough memory for %lu groups",
@@ -900,9 +899,8 @@ static int add_new_gdb_meta_bg(struct super_block *sb,
 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 	if (IS_ERR(gdb_bh))
 		return PTR_ERR(gdb_bh);
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
+				GFP_KERNEL);
 	if (!n_group_desc) {
 		brelse(gdb_bh);
 		err = -ENOMEM;
fs/ext4/super.c +0 -10
index 84a86d9b790f..ecf36a23e0c4 100644
@@ -204,16 +204,6 @@ void ext4_superblock_csum_set(struct super_block *sb)
 	es->s_checksum = ext4_superblock_csum(sb, es);
 }
 
-void *ext4_kvmalloc(size_t size, gfp_t flags)
-{
-	void *ret;
-
-	ret = kmalloc(size, flags | __GFP_NOWARN);
-	if (!ret)
-		ret = __vmalloc(size, flags, PAGE_KERNEL);
-	return ret;
-}
-
 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 			       struct ext4_group_desc *bg)
 {
fs/ext4/xattr.c +1 -1
index 246fbeeb6366..8cac7d95c3ad 100644
@@ -1456,7 +1456,7 @@ ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
 	if (!ce)
 		return NULL;
 
-	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
+	ea_data = kvmalloc(value_len, GFP_KERNEL);
 	if (!ea_data) {
 		mb_cache_entry_put(ea_inode_cache, ce);
 		return NULL;
Signature · 1 line
2.24.1
reviewer Reviewed-by on v1 1/1
#10 lore raw

Re: [PATCH] ext4: drop ext4_kvmalloc()

Quoted 7 lines · Theodore Ts'o

As Jan pointed out[1], as of commit 81378da64de ("jbd2: mark the transaction context with the scope GFP_NOFS context") we use memalloc_nofs_{save,restore}() while a jbd2 handle is active. So ext4_kvmalloc() so we can call allocate using GFP_NOFS is no longer necessary.

[1] https://lore.kernel.org/r/[email protected]

Your signed-off-by is missing but otherwise the patch looks good to me. You can add:

Honza

Quoted 84 lines
---
 fs/ext4/ext4.h   |  1 -
 fs/ext4/resize.c | 10 ++++------
 fs/ext4/super.c  | 10 ----------
 fs/ext4/xattr.c  |  2 +-
 4 files changed, 5 insertions(+), 18 deletions(-)
fs/ext4/ext4.h +0 -1
index 5e621b0da4da..9a2ee2428ecc 100644
@@ -2740,7 +2740,6 @@ extern struct buffer_head *ext4_sb_bread(struct super_block *sb,
 extern int ext4_seq_options_show(struct seq_file *seq, void *offset);
 extern int ext4_calculate_overhead(struct super_block *sb);
 extern void ext4_superblock_csum_set(struct super_block *sb);
-extern void *ext4_kvmalloc(size_t size, gfp_t flags);
 extern int ext4_alloc_flex_bg_array(struct super_block *sb,
 				    ext4_group_t ngroup);
 extern const char *ext4_decode_error(struct super_block *sb, int errno,
fs/ext4/resize.c +4 -6
index a8c0f2b5b6e1..86a2500ed292 100644
@@ -824,9 +824,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
 	if (unlikely(err))
 		goto errout;
 
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
+				GFP_KERNEL);
 	if (!n_group_desc) {
 		err = -ENOMEM;
 		ext4_warning(sb, "not enough memory for %lu groups",
@@ -900,9 +899,8 @@ static int add_new_gdb_meta_bg(struct super_block *sb,
 	gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 	if (IS_ERR(gdb_bh))
 		return PTR_ERR(gdb_bh);
-	n_group_desc = ext4_kvmalloc((gdb_num + 1) *
-				     sizeof(struct buffer_head *),
-				     GFP_NOFS);
+	n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
+				GFP_KERNEL);
 	if (!n_group_desc) {
 		brelse(gdb_bh);
 		err = -ENOMEM;
fs/ext4/super.c +0 -10
index 84a86d9b790f..ecf36a23e0c4 100644
@@ -204,16 +204,6 @@ void ext4_superblock_csum_set(struct super_block *sb)
 	es->s_checksum = ext4_superblock_csum(sb, es);
 }
 
-void *ext4_kvmalloc(size_t size, gfp_t flags)
-{
-	void *ret;
-
-	ret = kmalloc(size, flags | __GFP_NOWARN);
-	if (!ret)
-		ret = __vmalloc(size, flags, PAGE_KERNEL);
-	return ret;
-}
-
 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 			       struct ext4_group_desc *bg)
 {
fs/ext4/xattr.c +1 -1
index 246fbeeb6366..8cac7d95c3ad 100644
@@ -1456,7 +1456,7 @@ ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
 	if (!ce)
 		return NULL;
 
-	ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
+	ea_data = kvmalloc(value_len, GFP_KERNEL);
 	if (!ea_data) {
 		mb_cache_entry_put(ea_inode_cache, ce);
 		return NULL;
Signature · 1 line
2.24.1
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
maintainer Applied
#11 lore raw

Re: [PATCH] ext4: drop ext4_kvmalloc()

Quoted 7 lines · Jan Kara
Quoted 6 lines · Theodore Ts'o

As Jan pointed out[1], as of commit 81378da64de ("jbd2: mark the transaction context with the scope GFP_NOFS context") we use memalloc_nofs_{save,restore}() while a jbd2 handle is active. So ext4_kvmalloc() so we can call allocate using GFP_NOFS is no longer necessary.

Quoted 6 lines
Quoted 1 line

[1] https://lore.kernel.org/r/[email protected]

Your signed-off-by is missing but otherwise the patch looks good to me. You can add:

Thanks, applied with my signed-off-by and a Link: trailer.

- Ted

#12 lore raw

PATCH v2 3/3 ext4: Prevent ext4_kvmalloc_nofs() from re-entering the filesystem and deadlocking

In reply to Naoto Kobayashi

Even if __vmalloc() receives GFP_NOFS, this function allocates data pages and auxiliary structures (e.g. pagetables) with __GFP_FS[1]. To prevent memory reclaim from re-entering the filesystem here and potentially deadlocking, use memalloc_nofs_save() that gets __vmalloc() to drop __GFP_FS.

[1] linux-tree/Documentation/core-api/gfp-mask-fs-io.rst

---
 fs/ext4/super.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
fs/ext4/super.c +6 -3
index e8965aa6ecce..7f4c9a43a3f3 100644
@@ -43,6 +43,7 @@
 #include <linux/uaccess.h>
 #include <linux/iversion.h>
 #include <linux/unicode.h>
+#include <linux/sched/mm.h>
 
 #include <linux/kthread.h>
 #include <linux/freezer.h>
@@ -206,11 +207,13 @@ void ext4_superblock_csum_set(struct super_block *sb)
 
 void *ext4_kvmalloc_nofs(size_t size)
 {
+	unsigned int nofs_flag;
 	void *ret;
 
-	ret = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
-	if (!ret)
-		ret = __vmalloc(size, GFP_NOFS, PAGE_KERNEL);
+	/* kvmalloc() does not support GFP_NOFS */
+	nofs_flag = memalloc_nofs_save();
+	ret = kvmalloc(size, GFP_KERNEL);
+	memalloc_nofs_restore(nofs_flag);
 	return ret;
 }
 

--
2.16.6