Skip to contentSkip to conversation

ext4: mballoc: Use raw_cpu_ptr instead of this_cpu_ptr

PATCH In review Ritesh Harjani 2 participants 5 messages Jun 2, 2020 – Jun 9, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v5 1/1 ext4: mballoc: Use raw_cpu_ptr instead of this_cpu_ptr

It doesn't really matter in ext4_mb_new_blocks() about whether the code is rescheduled on any other cpu due to preemption. Because we care about discard_pa_seq only when the block allocation fails and then too we add the seq counter of all the cpus against the initial sampled one to check if anyone has freed any blocks while we were doing allocation.

So just use raw_cpu_ptr instead of this_cpu_ptr to avoid this BUG.

BUG: using smp_processor_id() in preemptible [00000000] code: syz-fuzzer/6927
caller is ext4_mb_new_blocks+0xa4d/0x3b70 fs/ext4/mballoc.c:4711
CPU: 1 PID: 6927 Comm: syz-fuzzer Not tainted 5.7.0-next-20200602-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x18f/0x20d lib/dump_stack.c:118
 check_preemption_disabled+0x20d/0x220 lib/smp_processor_id.c:48
 ext4_mb_new_blocks+0xa4d/0x3b70 fs/ext4/mballoc.c:4711
 ext4_ext_map_blocks+0x201b/0x33e0 fs/ext4/extents.c:4244
 ext4_map_blocks+0x4cb/0x1640 fs/ext4/inode.c:626
 ext4_getblk+0xad/0x520 fs/ext4/inode.c:833
 ext4_bread+0x7c/0x380 fs/ext4/inode.c:883
 ext4_append+0x153/0x360 fs/ext4/namei.c:67
 ext4_init_new_dir fs/ext4/namei.c:2757 [inline]
 ext4_mkdir+0x5e0/0xdf0 fs/ext4/namei.c:2802
 vfs_mkdir+0x419/0x690 fs/namei.c:3632
 do_mkdirat+0x21e/0x280 fs/namei.c:3655
 do_syscall_64+0x60/0xe0 arch/x86/entry/common.c:359
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
---
 fs/ext4/mballoc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/ext4/mballoc.c +1 -1
index a9083113a8c0..b79b32dbe3ea 100644
@@ -4708,7 +4708,7 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
 	}
 
 	ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
-	seq = *this_cpu_ptr(&discard_pa_seq);
+	seq = *raw_cpu_ptr(&discard_pa_seq);
 	if (!ext4_mb_use_preallocated(ac)) {
 		ac->ac_op = EXT4_MB_HISTORY_ALLOC;
 		ext4_mb_normalize_request(ac, ar);
Signature · 1 line
2.21.3
#1 lore raw

PATCH 1/1 ext4: mballoc: Use raw_cpu_ptr instead of this_cpu_ptr

It doesn't really matter in ext4_mb_new_blocks() about whether the code is rescheduled on any other cpu due to preemption. Because we care about discard_pa_seq only when the block allocation fails and then too we add the seq counter of all the cpus against the initial sampled one to check if anyone has freed any blocks while we were doing allocation.

So just use raw_cpu_ptr instead of this_cpu_ptr to avoid this BUG.

BUG: using smp_processor_id() in preemptible [00000000] code: syz-fuzzer/6927
caller is ext4_mb_new_blocks+0xa4d/0x3b70 fs/ext4/mballoc.c:4711
CPU: 1 PID: 6927 Comm: syz-fuzzer Not tainted 5.7.0-next-20200602-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x18f/0x20d lib/dump_stack.c:118
 check_preemption_disabled+0x20d/0x220 lib/smp_processor_id.c:48
 ext4_mb_new_blocks+0xa4d/0x3b70 fs/ext4/mballoc.c:4711
 ext4_ext_map_blocks+0x201b/0x33e0 fs/ext4/extents.c:4244
 ext4_map_blocks+0x4cb/0x1640 fs/ext4/inode.c:626
 ext4_getblk+0xad/0x520 fs/ext4/inode.c:833
 ext4_bread+0x7c/0x380 fs/ext4/inode.c:883
 ext4_append+0x153/0x360 fs/ext4/namei.c:67
 ext4_init_new_dir fs/ext4/namei.c:2757 [inline]
 ext4_mkdir+0x5e0/0xdf0 fs/ext4/namei.c:2802
 vfs_mkdir+0x419/0x690 fs/namei.c:3632
 do_mkdirat+0x21e/0x280 fs/namei.c:3655
 do_syscall_64+0x60/0xe0 arch/x86/entry/common.c:359
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
Fixes 42f56b7a4a7d ("ext4: mballoc: introduce pcpu seqcnt for freeing PA

to improve ENOSPC handling")

---
 fs/ext4/mballoc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/ext4/mballoc.c +1 -1
index a9083113a8c0..b79b32dbe3ea 100644
@@ -4708,7 +4708,7 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
 	}
 
 	ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
-	seq = *this_cpu_ptr(&discard_pa_seq);
+	seq = *raw_cpu_ptr(&discard_pa_seq);
 	if (!ext4_mb_use_preallocated(ac)) {
 		ac->ac_op = EXT4_MB_HISTORY_ALLOC;
 		ext4_mb_normalize_request(ac, ar);
Signature · 1 line
2.21.3
Tested-by on v5 1/1
#2 lore raw
In reply to Ritesh Harjani

Hi Ritesh,

Quoted 29 lines · Ritesh Harjani

It doesn't really matter in ext4_mb_new_blocks() about whether the code is rescheduled on any other cpu due to preemption. Because we care about discard_pa_seq only when the block allocation fails and then too we add the seq counter of all the cpus against the initial sampled one to check if anyone has freed any blocks while we were doing allocation.

So just use raw_cpu_ptr instead of this_cpu_ptr to avoid this BUG.

BUG: using smp_processor_id() in preemptible [00000000] code: syz-fuzzer/6927
caller is ext4_mb_new_blocks+0xa4d/0x3b70 fs/ext4/mballoc.c:4711
CPU: 1 PID: 6927 Comm: syz-fuzzer Not tainted 5.7.0-next-20200602-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
  __dump_stack lib/dump_stack.c:77 [inline]
  dump_stack+0x18f/0x20d lib/dump_stack.c:118
  check_preemption_disabled+0x20d/0x220 lib/smp_processor_id.c:48
  ext4_mb_new_blocks+0xa4d/0x3b70 fs/ext4/mballoc.c:4711
  ext4_ext_map_blocks+0x201b/0x33e0 fs/ext4/extents.c:4244
  ext4_map_blocks+0x4cb/0x1640 fs/ext4/inode.c:626
  ext4_getblk+0xad/0x520 fs/ext4/inode.c:833
  ext4_bread+0x7c/0x380 fs/ext4/inode.c:883
  ext4_append+0x153/0x360 fs/ext4/namei.c:67
  ext4_init_new_dir fs/ext4/namei.c:2757 [inline]
  ext4_mkdir+0x5e0/0xdf0 fs/ext4/namei.c:2802
  vfs_mkdir+0x419/0x690 fs/namei.c:3632
  do_mkdirat+0x21e/0x280 fs/namei.c:3655
  do_syscall_64+0x60/0xe0 arch/x86/entry/common.c:359
  entry_SYSCALL_64_after_hwframe+0x44/0xa9
Quoted 2 lines

This fixes the warning observed on various Samsung Exynos SoC based boards with linux-next 20200602.

Quoted 9 lines
---
  fs/ext4/mballoc.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
fs/ext4/mballoc.c +0 -0
index a9083113a8c0..b79b32dbe3ea 100644

fs/ext4/mballoc.c ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle, · line 4714

Quoted 8 lines
fs/ext4/mballoc.c +1 -1
@@ -4708,7 +4708,7 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
  	}
 
  	ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
-	seq = *this_cpu_ptr(&discard_pa_seq);
+	seq = *raw_cpu_ptr(&discard_pa_seq);
  	if (!ext4_mb_use_preallocated(ac)) {
  		ac->ac_op = EXT4_MB_HISTORY_ALLOC;
  		ext4_mb_normalize_request(ac, ar);

Best regards

Signature · 2 lines
Marek Szyprowski, PhD
Samsung R&D Institute Poland