linux-f2fs-devel.lists.sourceforge.net archive mirror
 help / color / mirror / Atom feed
From: Chao Yu <yuchao0@huawei.com>
To: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>,
	Jaegeuk Kim <jaegeuk@kernel.org>,
	<linux-f2fs-devel@lists.sourceforge.net>
Cc: Damien Le Moal <Damien.LeMoal@wdc.com>
Subject: Re: [f2fs-dev] [PATCH v4 2/2] fsck.f2fs: Check write pointer consistency with current segments
Date: Mon, 2 Sep 2019 15:02:26 +0800	[thread overview]
Message-ID: <486550b2-bb47-d725-79a9-4fb3a4ba28e3@huawei.com> (raw)
In-Reply-To: <20190830101936.32265-3-shinichiro.kawasaki@wdc.com>

On 2019/8/30 18:19, Shin'ichiro Kawasaki wrote:
> On sudden f2fs shutdown, zoned block device status and f2fs current
> segment positions in meta data can be inconsistent. When f2fs shutdown
> happens before write operations completes, write pointers of zoned block
> devices can go further but f2fs meta data keeps current segments at
> positions before the write operations. After remounting the f2fs, the
> inconsistency causes write operations not at write pointers and
> "Unaligned write command" error is reported. This error was observed when
> xfstests test case generic/388 was run with f2fs on a zoned block device.
> 
> To avoid the error, have f2fs.fsck check consistency between each current
> segment's position and the write pointer of the zone the current segment
> points to. If the write pointer goes advance from the current segment,
> fix the current segment position setting at same as the write pointer
> position. If the write pointer goes to the zone end, find a new zone and
> set the current segment position at the new zone start. In case the write
> pointer is behind the current segment, write zero data at the write
> pointer position to make write pointer position at same as the current
> segment.
> 
> When inconsistencies are found, turn on c.bug_on flag in fsck_verify() to
> ask users to fix them or not. When inconsistencies get fixed, turn on
> 'force' flag in fsck_verify() to enforce fixes in following checks. This
> position fix is done at the beginning of do_fsck() function so that other
> checks reflect the current segment modification.
> 
> Also add GET_SEC_FROM_SEG and GET_SEG_FROM_SEC macros in fsck/fsck.h to
> simplify the code.
> 
> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> ---
>  fsck/f2fs.h |   5 ++
>  fsck/fsck.c | 198 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>  fsck/fsck.h |   3 +
>  fsck/main.c |   2 +
>  4 files changed, 208 insertions(+)
> 
> diff --git a/fsck/f2fs.h b/fsck/f2fs.h
> index 4dc6698..2c1c2b3 100644
> --- a/fsck/f2fs.h
> +++ b/fsck/f2fs.h
> @@ -337,6 +337,11 @@ static inline block_t __end_block_addr(struct f2fs_sb_info *sbi)
>  #define GET_BLKOFF_FROM_SEG0(sbi, blk_addr)				\
>  	(GET_SEGOFF_FROM_SEG0(sbi, blk_addr) & (sbi->blocks_per_seg - 1))
>  
> +#define GET_SEC_FROM_SEG(sbi, segno)					\
> +	((segno) / (sbi)->segs_per_sec)
> +#define GET_SEG_FROM_SEC(sbi, secno)					\
> +	((secno) * (sbi)->segs_per_sec)
> +
>  #define FREE_I_START_SEGNO(sbi)						\
>  	GET_SEGNO_FROM_SEG0(sbi, SM_I(sbi)->main_blkaddr)
>  #define GET_R2L_SEGNO(sbi, segno)	(segno + FREE_I_START_SEGNO(sbi))
> diff --git a/fsck/fsck.c b/fsck/fsck.c
> index 8953ca1..a0f6849 100644
> --- a/fsck/fsck.c
> +++ b/fsck/fsck.c
> @@ -2574,6 +2574,190 @@ out:
>  	return cnt;
>  }
>  
> +/*
> + * Search a free section in main area. Start search from the section specified
> + * with segno argument toward main area end. Return first segment of the found
> + * section in segno argument.
> + */
> +static int find_next_free_section(struct f2fs_sb_info *sbi,
> +				  unsigned int *segno)
> +{
> +	unsigned int i, sec, section_valid_blocks;
> +	unsigned int end_segno = GET_SEGNO(sbi, SM_I(sbi)->main_blkaddr)
> +		+ SM_I(sbi)->main_segments;
> +	unsigned int end_sec = GET_SEC_FROM_SEG(sbi, end_segno);
> +	struct seg_entry *se;
> +	struct curseg_info *cs;
> +
> +	for (sec = GET_SEC_FROM_SEG(sbi, *segno); sec < end_sec; sec++) {
> +		/* find a section without valid blocks */
> +		section_valid_blocks = 0;
> +		for (i = 0; i < sbi->segs_per_sec; i++) {
> +			se = get_seg_entry(sbi, GET_SEG_FROM_SEC(sbi, sec) + i);
> +			section_valid_blocks += se->valid_blocks;

If we want to find a 'real' free section, we'd better to use
se->ckpt_valid_blocks (wrapped with get_seg_vblocks()) in where we has recorded
fsynced data count.

> +		}
> +		if (section_valid_blocks)
> +			continue;
> +
> +		/* check the cursegs do not use the section */
> +		for (i = 0; i < NO_CHECK_TYPE; i++) {
> +			cs = &SM_I(sbi)->curseg_array[i];
> +			if (GET_SEC_FROM_SEG(sbi, cs->segno) == sec)
> +				break;
> +		}
> +		if (i >= NR_CURSEG_TYPE) {
> +			*segno = GET_SEG_FROM_SEC(sbi, sec);
> +			return 0;
> +		}
> +	}
> +
> +	return -1;
> +}
> +
> +struct write_pointer_check_data {
> +	struct f2fs_sb_info *sbi;
> +	struct device_info *dev;
> +};
> +
> +static int fsck_chk_write_pointer(int i, struct blk_zone *blkz, void *opaque)
> +{
> +	struct write_pointer_check_data *wpd = opaque;
> +	struct f2fs_sb_info *sbi = wpd->sbi;
> +	struct device_info *dev = wpd->dev;
> +	struct f2fs_fsck *fsck = F2FS_FSCK(sbi);
> +	block_t zone_block, wp_block, wp_blkoff, cs_block, b;
> +	unsigned int zone_segno, wp_segno, new_segno;
> +	struct seg_entry *se;
> +	struct curseg_info *cs;
> +	int cs_index, ret;
> +	int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
> +	unsigned int segs_per_zone = sbi->segs_per_sec * sbi->secs_per_zone;
> +	void *zero_blk;
> +
> +	if (blk_zone_conv(blkz))
> +		return 0;
> +
> +	zone_block = dev->start_blkaddr
> +		+ (blk_zone_sector(blkz) >> log_sectors_per_block);
> +	zone_segno = GET_SEGNO(sbi, zone_block);
> +	wp_block = dev->start_blkaddr
> +		+ (blk_zone_wp_sector(blkz) >> log_sectors_per_block);
> +	wp_segno = GET_SEGNO(sbi, wp_block);
> +	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
> +
> +	/* find the curseg which points to the zone */
> +	for (cs_index = 0; cs_index < NO_CHECK_TYPE; cs_index++) {
> +		cs = &SM_I(sbi)->curseg_array[cs_index];
> +		if (zone_segno <= cs->segno &&
> +		    cs->segno < zone_segno + segs_per_zone)
> +			break;
> +	}
> +
> +	if (cs_index >= NR_CURSEG_TYPE)
> +		return 0;
> +
> +	/* check write pointer consistency with the curseg in the zone */
> +	cs_block = START_BLOCK(sbi, cs->segno) + cs->next_blkoff;
> +	if (wp_block == cs_block)
> +		return 0;
> +
> +	if (!c.fix_on) {
> +		MSG(0, "Inconsistent write pointer: "
> +		    "curseg %d[0x%x,0x%x] wp[0x%x,0x%x]\n",
> +		    cs_index, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
> +		fsck->chk.wp_inconsistent_zones++;
> +		return 0;
> +	}
> +
> +	/*
> +	 * If the curseg is in advance from the write pointer, write zero to
> +	 * move the write pointer forward to the same position as the curseg.
> +	 */
> +	if (wp_block < cs_block) {
> +		ret = 0;
> +		zero_blk = calloc(BLOCK_SZ, 1);
> +		if (!zero_blk)
> +			return -EINVAL;
> +
> +		FIX_MSG("Advance write pointer to match with curseg %d: "
> +			"[0x%x,0x%x]->[0x%x,0x%x]",
> +			cs_index, wp_segno, wp_blkoff,
> +			cs->segno, cs->next_blkoff);
> +		for (b = wp_block; b < cs_block && !ret; b++)
> +			ret = dev_write_block(zero_blk, b);
> +
> +		fsck->chk.wp_fixed_zones++;
> +		free(zero_blk);
> +		return ret;
> +	}
> +
> +	if (wp_segno == zone_segno + segs_per_zone) {
> +		/*
> +		 * If the write pointer is in advance from the curseg and at
> +		 * the zone end (section end), search a new free zone (section)
> +		 * between the curseg and main area end.
> +		 */
> +		new_segno = wp_segno;
> +		ret = find_next_free_section(sbi, &new_segno);
> +		if (ret) {
> +			/* search again from main area start */
> +			new_segno = GET_SEGNO(sbi, SM_I(sbi)->main_blkaddr);
> +			ret = find_next_free_section(sbi, &new_segno);
> +		}

If curseg's type is warm node, relocating curseg would ruin warm node chain,
result in losing fsynced data for ever as well.

So I guess we should migrate all dnode blocks chained with cs->next_blkoff in
current section.

> +		if (ret) {
> +			MSG(0, "Free section not found\n");
> +			return ret;
> +		}
> +		FIX_MSG("New section for curseg %d: [0x%x,0x%x]->[0x%x,0x%x]",
> +			cs_index, cs->segno, cs->next_blkoff, new_segno, 0);
> +		cs->segno = new_segno;
> +		cs->next_blkoff = 0;
> +	} else {
> +		/*
> +		 * If the write pointer is in advance from the curseg within
> +		 * the zone, modify the curseg position to be same as the
> +		 * write pointer.
> +		 */
> +		ASSERT(wp_segno < zone_segno + segs_per_zone);
> +		FIX_MSG("Advance curseg %d: [0x%x,0x%x]->[0x%x,0x%x]",
> +			cs_index, cs->segno, cs->next_blkoff,
> +			wp_segno, wp_blkoff);
> +		cs->segno = wp_segno;
> +		cs->next_blkoff = wp_blkoff;

The same data lose problem here, I guess we'd better handle it with the same way
as above.

Thoughts?

Thanks,

> +	}
> +
> +	se = get_seg_entry(sbi, cs->segno);
> +	se->type = cs_index;
> +	fsck->chk.wp_fixed_zones++;
> +
> +	return 0;
> +}
> +
> +void fsck_chk_write_pointers(struct f2fs_sb_info *sbi)
> +{
> +	unsigned int i;
> +	struct f2fs_fsck *fsck = F2FS_FSCK(sbi);
> +	struct write_pointer_check_data wpd = {	sbi, NULL };
> +
> +	if (c.zoned_model != F2FS_ZONED_HM)
> +		return;
> +
> +	for (i = 0; i < MAX_DEVICES; i++) {
> +		if (!c.devices[i].path)
> +			break;
> +
> +		wpd.dev = c.devices + i;
> +		if (f2fs_report_zones(i, fsck_chk_write_pointer, &wpd)) {
> +			printf("[FSCK] Write pointer check failed: %s\n",
> +			       c.devices[i].path);
> +			return;
> +		}
> +	}
> +
> +	if (fsck->chk.wp_fixed_zones && c.fix_on)
> +		write_curseg_info(sbi);
> +}
> +
>  int fsck_chk_curseg_info(struct f2fs_sb_info *sbi)
>  {
>  	struct curseg_info *curseg;
> @@ -2624,6 +2808,20 @@ int fsck_verify(struct f2fs_sb_info *sbi)
>  
>  	printf("\n");
>  
> +	if (c.zoned_model == F2FS_ZONED_HM) {
> +		printf("[FSCK] Write pointers consistency                    ");
> +		if (fsck->chk.wp_inconsistent_zones == 0x0) {
> +			printf(" [Ok..]\n");
> +		} else {
> +			printf(" [Fail] [0x%x]\n",
> +			       fsck->chk.wp_inconsistent_zones);
> +			c.bug_on = 1;
> +		}
> +
> +		if (fsck->chk.wp_fixed_zones && c.fix_on)
> +			force = 1;
> +	}
> +
>  	if (c.feature & cpu_to_le32(F2FS_FEATURE_LOST_FOUND)) {
>  		for (i = 0; i < fsck->nr_nat_entries; i++)
>  			if (f2fs_test_bit(i, fsck->nat_area_bitmap) != 0)
> diff --git a/fsck/fsck.h b/fsck/fsck.h
> index d38e8de..aa3dbe7 100644
> --- a/fsck/fsck.h
> +++ b/fsck/fsck.h
> @@ -80,6 +80,8 @@ struct f2fs_fsck {
>  		u32 multi_hard_link_files;
>  		u64 sit_valid_blocks;
>  		u32 sit_free_segs;
> +		u32 wp_fixed_zones;
> +		u32 wp_inconsistent_zones;
>  	} chk;
>  
>  	struct hard_link_node *hard_link_list_head;
> @@ -156,6 +158,7 @@ int fsck_chk_inline_dentries(struct f2fs_sb_info *, struct f2fs_node *,
>  		struct child_info *);
>  void fsck_chk_checkpoint(struct f2fs_sb_info *sbi);
>  int fsck_chk_meta(struct f2fs_sb_info *sbi);
> +void fsck_chk_write_pointers(struct f2fs_sb_info *);
>  int fsck_chk_curseg_info(struct f2fs_sb_info *);
>  void pretty_print_filename(const u8 *raw_name, u32 len,
>  			   char out[F2FS_PRINT_NAMELEN], int enc_name);
> diff --git a/fsck/main.c b/fsck/main.c
> index 9aca024..4b4a789 100644
> --- a/fsck/main.c
> +++ b/fsck/main.c
> @@ -584,6 +584,8 @@ static void do_fsck(struct f2fs_sb_info *sbi)
>  
>  	print_cp_state(flag);
>  
> +	fsck_chk_write_pointers(sbi);
> +
>  	fsck_chk_curseg_info(sbi);
>  
>  	if (!c.fix_on && !c.bug_on) {
> 


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

  reply	other threads:[~2019-09-02  7:02 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-08-30 10:19 [f2fs-dev] [PATCH v4 0/2] fsck: Check write pointers of zoned block devices Shin'ichiro Kawasaki
2019-08-30 10:19 ` [f2fs-dev] [PATCH v4 1/2] libf2fs_zoned: Introduce f2fs_report_zones() helper function Shin'ichiro Kawasaki
2019-08-30 10:19 ` [f2fs-dev] [PATCH v4 2/2] fsck.f2fs: Check write pointer consistency with current segments Shin'ichiro Kawasaki
2019-09-02  7:02   ` Chao Yu [this message]
2019-09-03  8:37     ` Shinichiro Kawasaki
2019-09-05  9:58       ` Chao Yu
2019-09-06  8:31         ` Shinichiro Kawasaki
2019-09-09  7:14           ` Chao Yu
2019-09-10  8:10             ` Shinichiro Kawasaki
2019-09-10  9:12               ` Chao Yu
2019-09-12  8:16                 ` Shinichiro Kawasaki
2019-09-16  1:37                   ` Chao Yu
2019-09-18  3:07                     ` Shinichiro Kawasaki
2019-09-21  9:42                       ` Chao Yu
2019-09-25  8:05                         ` Shinichiro Kawasaki
2019-09-29  2:09                           ` Chao Yu
2019-10-02  5:30                             ` Shinichiro Kawasaki
2019-10-18  6:55                               ` Shinichiro Kawasaki

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=486550b2-bb47-d725-79a9-4fb3a4ba28e3@huawei.com \
    --to=yuchao0@huawei.com \
    --cc=Damien.LeMoal@wdc.com \
    --cc=jaegeuk@kernel.org \
    --cc=linux-f2fs-devel@lists.sourceforge.net \
    --cc=shinichiro.kawasaki@wdc.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).