【OLK-5.10】文件系统回写和fsync竞争导致死锁触发Hung task

【标题描述】能够简要描述问题：说明什么场景下，做了什么操作，出现什么问题（尽量使用正向表达方式）
往硬件队列深度都为1的U盘vfat拷贝大文件并同时执行fsync卡主
![输入图片说明](https://images.gitee.com/uploads/images/2022/0429/101224_a183839f_8908684.png "问题截图.PNG")
【环境信息】
硬件信息：
1）	arm32
软件信息：
1）	OLK5.10
【问题复现步骤】
具体操作步骤
1. 打入内核延时补丁，打开CONFIG_VFAT_FS，CONFIG_FAT_FS
2. 编译执行b.c

出现概率(是否必现，概率性错误)必现
【预期结果】
fsync进程卡主，内核报hungtask
【实际结果】
```
[   45.714127] kworker/u8:4 cannot lock for inode ffff8881058003f8 bh ffff88810fd05958
[   45.874154] kworker/u8:4 cannot lock for inode ffff8881058003f8 bh ffff88810fd05958
[   46.034180] kworker/u8:4 cannot lock for inode ffff8881058003f8 bh ffff88810fd05958
[   46.194204] kworker/u8:4 cannot lock for inode ffff8881058003f8 bh ffff88810fd05958
[   46.281892] INFO: task bb:2694 blocked for more than 15 seconds.
[   46.286758]       Not tainted 5.10.0-13176-ga304d9b8bdc9-dirty #321
[   46.287606] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[   46.289334] task:bb              state:D stack:    0 pid: 2694 ppid:  2412 flags:0x00004000
[   46.289351] Call Trace:
[   46.289380]  __schedule+0x31a/0xb90
[   46.289387]  schedule+0x7e/0x1a0
[   46.289392]  io_schedule+0x22/0x50
[   46.289400]  blk_mq_get_tag+0x21a/0x3d0
[   46.289412]  ? woken_wake_function+0x30/0x30
[   46.289418]  __blk_mq_alloc_request+0x123/0x1b0
[   46.289422]  blk_mq_submit_bio+0x163/0x9f0
[   46.289428]  submit_bio_noacct+0x4dd/0x5f0
[   46.289435]  ? vprintk_func+0x79/0x180
[   46.289439]  submit_bio+0x3f/0x1b0
[   46.289444]  submit_bh_wbc+0x1c8/0x2f0
[   46.289449]  __sync_dirty_buffer+0xc4/0x360
[   46.289453]  sync_dirty_buffer+0x17/0x20
[   46.289458]  __fat_write_inode+0x294/0x320
[   46.289463]  fat_write_inode+0x2a/0xa0
[   46.289469]  ? filemap_fdatawait_range+0x26/0x30
[   46.289474]  __writeback_single_inode+0x435/0x560
[   46.289478]  writeback_single_inode+0x154/0x220
[   46.289482]  sync_inode_metadata+0x45/0x70
[   46.289487]  __generic_file_fsync+0xa3/0x150
[   46.289491]  fat_file_fsync+0x1d/0x90
[   46.289496]  vfs_fsync_range+0x40/0xb0
[   46.289501]  ? exit_to_user_mode_prepare+0x2a/0x210
[   46.289505]  do_fsync+0x48/0xa0
[   46.289509]  __x64_sys_fsync+0x18/0x30
[   46.289514]  do_syscall_64+0x45/0x70
[   46.289518]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
```
【附件信息】
延时补丁：
```
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 873eeb97fde4..fdcb5adb63cb 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -507,6 +507,11 @@ static void __blk_mq_free_request(struct request *rq)
 	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
 	const int sched_tag = rq->internal_tag;
 
+	if (rq->rq_disk && rq->rq_disk->first_minor == 16) {
+		global_sync_cond2 = 0;
+		smp_wmb();
+	}
+
 	blk_crypto_free_request(rq);
 	blk_pm_mark_last_busy(rq);
 	rq->mq_hctx = NULL;
@@ -2158,6 +2163,7 @@ static inline unsigned short blk_plug_max_rq_count(struct blk_plug *plug)
  *
  * Returns: Request queue cookie.
  */
+int global_sync_cond2 = 0;
 blk_qc_t blk_mq_submit_bio(struct bio *bio)
 {
 	struct request_queue *q = bio->bi_disk->queue;
@@ -2236,6 +2242,11 @@ blk_qc_t blk_mq_submit_bio(struct bio *bio)
 		else
 			last = list_entry_rq(plug->mq_list.prev);
 
+		if (rq->rq_disk->first_minor == 16) {
+			global_sync_cond2 = 1;
+			smp_wmb();
+		}
+
 		if (request_count >= blk_plug_max_rq_count(plug) || (last &&
 		    blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE)) {
 			blk_flush_plug_list(plug, false);
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 9352a309fe55..019b89b3becb 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1908,8 +1908,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost)
 		tag_set->ops = &scsi_mq_ops;
 	else
 		tag_set->ops = &scsi_mq_ops_no_commit;
-	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
-	tag_set->queue_depth = shost->can_queue;
+	tag_set->nr_hw_queues = 1;
+	tag_set->queue_depth = 1;
 	tag_set->cmd_size = cmd_size;
 	tag_set->numa_node = NUMA_NO_NODE;
 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
index fabbad5cda60..863d8cae4e71 100644
--- a/drivers/scsi/virtio_scsi.c
+++ b/drivers/scsi/virtio_scsi.c
@@ -877,10 +877,8 @@ static int virtscsi_probe(struct virtio_device *vdev)
 	if (err)
 		goto virtscsi_init_failed;
 
-	shost->can_queue = virtqueue_get_vring_size(vscsi->req_vqs[0].vq);
-
-	cmd_per_lun = virtscsi_config_get(vdev, cmd_per_lun) ?: 1;
-	shost->cmd_per_lun = min_t(u32, cmd_per_lun, shost->can_queue);
+	shost->can_queue = 1;
+	shost->cmd_per_lun = 1;
 	shost->max_sectors = virtscsi_config_get(vdev, max_sectors) ?: 0xFFFF;
 
 	/* LUNs > 256 are reported with format 1, so they go in the range
diff --git a/fs/buffer.c b/fs/buffer.c
index 23f645657488..32de2592134a 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -1709,6 +1709,7 @@ static struct buffer_head *create_page_buffers(struct page *page, struct inode *
  * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
  * causes the writes to be flagged as synchronous writes.
  */
+#include <linux/delay.h>
 int __block_write_full_page(struct inode *inode, struct page *page,
 			get_block_t *get_block, struct writeback_control *wbc,
 			bh_end_io_t *handler)
@@ -1784,9 +1785,14 @@ int __block_write_full_page(struct inode *inode, struct page *page,
 		 * and kswapd activity, but those code paths have their own
 		 * higher-level throttling.
 		 */
+		smp_rmb();
+		if (global_sync_cond && global_sync_cond2 && global_sync_cond3) {
+			mdelay(20);
+		}
 		if (wbc->sync_mode != WB_SYNC_NONE) {
 			lock_buffer(bh);
 		} else if (!trylock_buffer(bh)) {
+			pr_info("%s cannot lock for inode %px bh %px\n", current->comm, inode, bh);
 			redirty_page_for_writepage(wbc, page);
 			continue;
 		}
@@ -3119,13 +3125,19 @@ EXPORT_SYMBOL(ll_rw_block);
 void write_dirty_buffer(struct buffer_head *bh, int op_flags)
 {
 	lock_buffer(bh);
+	if (!strcmp(bh->b_page->mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(bh->b_page->mapping->host)->bd_dev) == 16 && bh->b_assoc_map && !strcmp(bh->b_assoc_map->host->i_sb->s_type->name, "vfat"))
+		pr_err("%s:%s lock for inode %px bh %px\n", current->comm, __func__, bh->b_page->mapping->host, bh);
 	if (!test_clear_buffer_dirty(bh)) {
+	if (!strcmp(bh->b_page->mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(bh->b_page->mapping->host)->bd_dev) == 16 && bh->b_assoc_map && !strcmp(bh->b_assoc_map->host->i_sb->s_type->name, "vfat"))
+		pr_err("%s:%s unlock for inode %px bh %px\n", current->comm, __func__, bh->b_page->mapping->host, bh);
 		unlock_buffer(bh);
 		return;
 	}
 	bh->b_end_io = end_buffer_write_sync;
 	get_bh(bh);
 	submit_bh(REQ_OP_WRITE, op_flags, bh);
+	if (!strcmp(bh->b_page->mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(bh->b_page->mapping->host)->bd_dev) == 16 && bh->b_assoc_map && !strcmp(bh->b_assoc_map->host->i_sb->s_type->name, "vfat"))
+		pr_err("%s:%s unlock2 for inode %px bh %px\n", current->comm, __func__, bh->b_page->mapping->host, bh);
 }
 EXPORT_SYMBOL(write_dirty_buffer);
 
@@ -3140,6 +3152,8 @@ int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
 
 	WARN_ON(atomic_read(&bh->b_count) < 1);
 	lock_buffer(bh);
+	if (!strcmp(bh->b_page->mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(bh->b_page->mapping->host)->bd_dev) == 16 && bh->b_assoc_map && !strcmp(bh->b_assoc_map->host->i_sb->s_type->name, "vfat"))
+		pr_err("%s:%s lock for inode %px bh %px\n", current->comm, __func__, bh->b_page->mapping->host, bh);
 	if (test_clear_buffer_dirty(bh)) {
 		/*
 		 * The bh should be mapped, but it might not be if the
@@ -3147,6 +3161,8 @@ int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
 		 */
 		if (!buffer_mapped(bh)) {
 			unlock_buffer(bh);
+			if (!strcmp(bh->b_page->mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(bh->b_page->mapping->host)->bd_dev) == 16 && bh->b_assoc_map && !strcmp(bh->b_assoc_map->host->i_sb->s_type->name, "vfat"))
+				pr_err("%s:%s unlock for inode %px bh %px\n", current->comm, __func__, bh->b_page->mapping->host, bh);
 			return -EIO;
 		}
 
@@ -3159,6 +3175,8 @@ int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
 	} else {
 		unlock_buffer(bh);
 	}
+	if (!strcmp(bh->b_page->mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(bh->b_page->mapping->host)->bd_dev) == 16 && bh->b_assoc_map && !strcmp(bh->b_assoc_map->host->i_sb->s_type->name, "vfat"))
+		pr_err("%s:%s unlock2 for inode %px bh %px\n", current->comm, __func__, bh->b_page->mapping->host, bh);
 	return ret;
 }
 EXPORT_SYMBOL(__sync_dirty_buffer);
diff --git a/fs/fat/file.c b/fs/fat/file.c
index f9ee27cf4d7c..56a9aefc6451 100644
--- a/fs/fat/file.c
+++ b/fs/fat/file.c
@@ -182,6 +182,7 @@ static int fat_file_release(struct inode *inode, struct file *filp)
 	return 0;
 }
 
+#include <linux/delay.h>
 int fat_file_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 {
 	struct inode *inode = filp->f_mapping->host;
diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
index 050d40c465bc..cbed73becf9e 100644
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c
@@ -1408,6 +1408,8 @@ static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
 		 * writeback is not making progress due to locked
 		 * buffers. Skip this inode for now.
 		 */
+//		if (!strcmp("bdev", inode->i_sb->s_type->name))
+//			pr_info("requeue inode %px\n", inode);
 		redirty_tail_locked(inode, wb);
 		return;
 	}
@@ -1514,8 +1516,11 @@ __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
 	 */
 	smp_mb();
 
-	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
+	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
+//		if (!strcmp("bdev", inode->i_sb->s_type->name))
+//			pr_info("tag dirty inode %px\n", inode);
 		inode->i_state |= I_DIRTY_PAGES;
+	}
 
 	spin_unlock(&inode->i_lock);
 
@@ -1776,11 +1781,13 @@ static long writeback_sb_inodes(struct super_block *sb,
 	return wrote;
 }
 
+int global_sync_cond = 0;
 static long __writeback_inodes_wb(struct bdi_writeback *wb,
 				  struct wb_writeback_work *work)
 {
 	unsigned long start_time = jiffies;
 	long wrote = 0;
+	int has_vfat = 0;
 
 	while (!list_empty(&wb->b_io)) {
 		struct inode *inode = wb_inode(wb->b_io.prev);
@@ -1795,7 +1802,15 @@ static long __writeback_inodes_wb(struct bdi_writeback *wb,
 			redirty_tail(inode, wb);
 			continue;
 		}
+
+		if (!strcmp(sb->s_type->name, "bdev") && MINOR(I_BDEV(inode)->bd_dev) == 16 && has_vfat && wb_inode(wb->b_io.prev) == wb_inode(wb->b_io.next)) {
+			global_sync_cond = 1;
+			smp_wmb();
+		}
+
 		wrote += writeback_sb_inodes(sb, wb, work);
+		if (!strcmp(sb->s_type->name, "vfat"))
+			has_vfat++;
 		up_read(&sb->s_umount);
 
 		/* refer to the same tests at the end of writeback_sb_inodes */
@@ -1873,8 +1888,12 @@ static long wb_writeback(struct bdi_writeback *wb,
 		 * after the other works are all done.
 		 */
 		if ((work->for_background || work->for_kupdate) &&
-		    !list_empty(&wb->work_list))
+		    !list_empty(&wb->work_list)) {
+			if (!list_empty(&wb->b_dirty)) {
+				pr_err("exit %s\n", current->comm);
+			}
 			break;
+		}
 
 		/*
 		 * For background writeout, stop when we are below the
@@ -1936,6 +1955,8 @@ static long wb_writeback(struct bdi_writeback *wb,
 	}
 	spin_unlock(&wb->list_lock);
 	blk_finish_plug(&plug);
+	global_sync_cond = 0;
+	smp_wmb();
 
 	return nr_pages - work->nr_pages;
 }
diff --git a/fs/sync.c b/fs/sync.c
index 1373a610dc78..2a888f53fb28 100644
--- a/fs/sync.c
+++ b/fs/sync.c
@@ -112,10 +112,10 @@ void ksys_sync(void)
 
 	wakeup_flusher_threads(WB_REASON_SYNC);
 	iterate_supers(sync_inodes_one_sb, NULL);
-	iterate_supers(sync_fs_one_sb, &nowait);
+	/*iterate_supers(sync_fs_one_sb, &nowait);
 	iterate_supers(sync_fs_one_sb, &wait);
 	iterate_bdevs(fdatawrite_one_bdev, NULL);
-	iterate_bdevs(fdatawait_one_bdev, NULL);
+	iterate_bdevs(fdatawait_one_bdev, NULL);*/
 	if (unlikely(laptop_mode))
 		laptop_sync_completion();
 }
diff --git a/include/linux/fs.h b/include/linux/fs.h
index b7f42d3dce26..1725352a8c14 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -44,6 +44,10 @@
 #include <asm/byteorder.h>
 #include <uapi/linux/fs.h>
 
+extern int global_sync_cond;	// There are bdev inode and fat inode in a same work, and bdev inode is the last inode to writeback
+extern int global_sync_cond2;	// The only one tag is gotten by one request, and the request is still in plug->mq_list
+extern int global_sync_cond3;	// wb_writeback is writing last page of bdev inode
+
 struct backing_dev_info;
 struct bdi_writeback;
 struct bio;
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index b72da123f242..ad560bb8550d 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -2160,6 +2160,7 @@ EXPORT_SYMBOL(tag_pages_for_writeback);
  *
  * Return: %0 on success, negative error code otherwise
  */
+int global_sync_cond3 = 0;
 int write_cache_pages(struct address_space *mapping,
 		      struct writeback_control *wbc, writepage_t writepage,
 		      void *data)
@@ -2175,6 +2176,8 @@ int write_cache_pages(struct address_space *mapping,
 	int range_whole = 0;
 	xa_mark_t tag;
 
+	global_sync_cond3 = 0;
+	smp_wmb();
 	pagevec_init(&pvec);
 	if (wbc->range_cyclic) {
 		index = mapping->writeback_index; /* prev offset */
@@ -2237,6 +2240,11 @@ int write_cache_pages(struct address_space *mapping,
 			if (!clear_page_dirty_for_io(page))
 				goto continue_unlock;
 
+			if (i == nr_pages - 1 && !strcmp(mapping->host->i_sb->s_type->name, "bdev") && MINOR(I_BDEV(mapping->host)->bd_dev) == 16) {
+				global_sync_cond3 = 1;
+				smp_wmb();
+			}
+
 			trace_wbc_writepage(wbc, inode_to_bdi(mapping->host));
 			error = (*writepage)(page, wbc, data);
 			if (unlikely(error)) {
-- 
2.31.1
```
b.c
```
#define _GNU_SOURCE

#include <dirent.h>
#include <endian.h>
#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/prctl.h>
#include <sys/stat.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>

#define BUFF_SIZE (1024 * 600)
#define DEV "sdb"

int main(void) {
	system("umount temp");
	system("mkfs.fat /dev/" DEV);
	system("sysctl -w vm.dirty_background_bytes=20");
	system("echo none > /sys/block/" DEV "/queue/scheduler");
	system("mount /dev/" DEV " /root/temp");
	int fd = open("/root/temp/file", O_RDWR | O_CREAT);
	int i;
	char buf[BUFF_SIZE];
	for (i = 0; i < 1000; ++i) {
		write(fd, buf, BUFF_SIZE);
		printf("fsync %d\n", i);
		fsync(fd);
		printf("done %d\n", i);
	}
	return 0;
}
```

GVP openEuler/kernel

内容风险标识

评论 (1)

GVPopenEuler/kernel

内容风险标识