Re: [f2fs-dev] [PATCH] f2fs: compress: support chksum

From: Jaegeuk Kim
Date: Mon Nov 09 2020 - 23:23:59 EST

Next message: Rayagonda Kokatanur: "Re: [PATCH v3 5/6] i2c: iproc: handle master read request"
Previous message: patchwork-bot+netdevbpf: "Re: [PATCH v2 net-next] net: skb_vlan_untag(): don't reset transport offset if set by GRO layer"
In reply to: Chao Yu: "Re: [f2fs-dev] [PATCH] f2fs: compress: support chksum"
Next in thread: Chao Yu: "Re: [f2fs-dev] [PATCH] f2fs: compress: support chksum"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 11/10, Chao Yu wrote:
> On 2020/11/10 1:06, Jaegeuk Kim wrote:
> > On 11/09, Chao Yu wrote:
> > > On 2020/11/7 5:12, Jaegeuk Kim wrote:
> > > > On 11/03, Chao Yu wrote:
> > > > > On 2020/11/3 10:02, Chao Yu wrote:
> > > > > > On 2020/11/3 0:31, Jaegeuk Kim wrote:
> > > > > > > On 11/02, Chao Yu wrote:
> > > > > > > > This patch supports to store chksum value with compressed
> > > > > > > > data, and verify the integrality of compressed data while
> > > > > > > > reading the data.
> > > > > > > >
> > > > > > > > The feature can be enabled through specifying mount option
> > > > > > > > 'compress_chksum'.
> > > > > > > >
> > > > > > > > Signed-off-by: Chao Yu <yuchao0@xxxxxxxxxx>
> > > > > > > > ---
> > > > > > > > Documentation/filesystems/f2fs.rst | 1 +
> > > > > > > > fs/f2fs/compress.c | 20 ++++++++++++++++++++
> > > > > > > > fs/f2fs/f2fs.h | 13 ++++++++++++-
> > > > > > > > fs/f2fs/inode.c | 3 +++
> > > > > > > > fs/f2fs/super.c | 9 +++++++++
> > > > > > > > include/linux/f2fs_fs.h | 2 +-
> > > > > > > > 6 files changed, 46 insertions(+), 2 deletions(-)
> > > > > > > >
> > > > > > > > diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
> > > > > > > > index b8ee761c9922..985ae7d35066 100644
> > > > > > > > --- a/Documentation/filesystems/f2fs.rst
> > > > > > > > +++ b/Documentation/filesystems/f2fs.rst
> > > > > > > > @@ -260,6 +260,7 @@ compress_extension=%s Support adding specified extension, so that f2fs can enab
> > > > > > > > For other files, we can still enable compression via ioctl.
> > > > > > > > Note that, there is one reserved special extension '*', it
> > > > > > > > can be set to enable compression for all files.
> > > > > > > > +compress_chksum Support verifying chksum of raw data in compressed cluster.
> > > > > > > > inlinecrypt When possible, encrypt/decrypt the contents of encrypted
> > > > > > > > files using the blk-crypto framework rather than
> > > > > > > > filesystem-layer encryption. This allows the use of
> > > > > > > > diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> > > > > > > > index 14262e0f1cd6..a4e0d2c745b6 100644
> > > > > > > > --- a/fs/f2fs/compress.c
> > > > > > > > +++ b/fs/f2fs/compress.c
> > > > > > > > @@ -602,6 +602,7 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
> > > > > > > > f2fs_cops[fi->i_compress_algorithm];
> > > > > > > > unsigned int max_len, new_nr_cpages;
> > > > > > > > struct page **new_cpages;
> > > > > > > > + u32 chksum = 0;
> > > > > > > > int i, ret;
> > > > > > > > trace_f2fs_compress_pages_start(cc->inode, cc->cluster_idx,
> > > > > > > > @@ -655,6 +656,11 @@ static int f2fs_compress_pages(struct compress_ctx *cc)
> > > > > > > > cc->cbuf->clen = cpu_to_le32(cc->clen);
> > > > > > > > + if (fi->i_compress_flag & 1 << COMPRESS_CHKSUM)
> > > > > > > > + chksum = f2fs_crc32(F2FS_I_SB(cc->inode),
> > > > > > > > + cc->cbuf->cdata, cc->clen);
> > > > > > > > + cc->cbuf->chksum = cpu_to_le32(chksum);
> > > > > > > > +
> > > > > > > > for (i = 0; i < COMPRESS_DATA_RESERVED_SIZE; i++)
> > > > > > > > cc->cbuf->reserved[i] = cpu_to_le32(0);
> > > > > > > > @@ -721,6 +727,7 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
> > > > > > > > (struct decompress_io_ctx *)page_private(page);
> > > > > > > > struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
> > > > > > > > struct f2fs_inode_info *fi= F2FS_I(dic->inode);
> > > > > > > > + struct f2fs_sb_info *sbi = F2FS_I_SB(dic->inode);
> > > > > > > > const struct f2fs_compress_ops *cops =
> > > > > > > > f2fs_cops[fi->i_compress_algorithm];
> > > > > > > > int ret;
> > > > > > > > @@ -790,6 +797,19 @@ void f2fs_decompress_pages(struct bio *bio, struct page *page, bool verity)
> > > > > > > > ret = cops->decompress_pages(dic);
> > > > > > > > + if (!ret && fi->i_compress_flag & 1 << COMPRESS_CHKSUM) {
> > > > > > > > + u32 provided = le32_to_cpu(dic->cbuf->chksum);
> > > > > > > > + u32 calculated = f2fs_crc32(sbi, dic->cbuf->cdata, dic->clen);
> > > > > > > > +
> > > > > > > > + if (provided != calculated) {
> > > > > > > > + printk_ratelimited(
> > > > > > > > + "%sF2FS-fs (%s): checksum invalid, nid = %lu, %x vs %x",
> > > > > > > > + KERN_INFO, sbi->sb->s_id, dic->inode->i_ino,
> > > > > > > > + provided, calculated);
> > > > > > > > + ret = -EFSCORRUPTED;
> > > > > > >
> > > > > > > Do we need to change fsck.f2fs to recover this?
> > > > >
> > > > > However, we don't know which one is correct, compressed data or chksum value?
> > > > > if compressed data was corrupted, repairing chksum value doesn't help.
> > > > >
> > > > > Or how about adding chksum values for both raw data and compressed data.
> > > > >
> > > > > #define COMPRESS_DATA_RESERVED_SIZE 3
> > > > > struct compress_data {
> > > > > __le32 clen; /* compressed data size */
> > > > > + __le32 raw_chksum; /* raw data chksum */
> > > > > + __le32 compress_chksum; /* compressed data chksum */
> > > > > __le32 reserved[COMPRESS_DATA_RESERVED_SIZE]; /* reserved */
> > > > > u8 cdata[]; /* compressed data */
> > > > > };
> > > > >
> > > > > raw_chksum compress_chksum
> > > > > match match -> data is verified, pass
> > > > > not match match -> repair raw_chksum
> > > > > matcth not match -> repair compress_chksum
> > > >
> > > > I think only compress_chksum would be enough. BTW, can we give WARN_ON and
> > > > marking a FSCK flag without returning EFSCORRUPTED, since we don't really
> > > > know who was corrupted. If data was corrupted, we should be able to see app
> > > > corruption. In that case, we can check the kernel log. If checksum was simply
> > >
> > > I don't think that app will always corrupt once data was corrupted, I doubt its
> > > behavior could be slightly abnormal in some cases, e.g. it can just cause apps to
> > > show wrong number in interaction interface.
> >
> > I didn't say we can always get it. But, it's likely to happen with something
> > like that.
> >
> > >
> > > In this case, if we fix chksum in fsck, the wrong data will never be found due to
> > > data's chksum matches data itself after repair.
> >
> > At least, we should see that log as a hint.
> >
> > >
> > > IMO, the chksum and data was a whole dataset, once they are mismatch, we can
> > > not trust either of them, fsck should do nothing on them unless we store parity
> > > bits or replica.
> >
> > Yeah, the point is those are not written as a transaction. Three concerns to me:
>
> Just to confirm:
>
> > 1) agreed to see compressed data corruption by checksum
>
> return EFSBADCRC always like we did as inode chksum feature.

inode chksum is a bit different, since 4KB write is atomic. If chksum is
corrupted, inode should be broken.

>
> 2) I don't want to see the error messages all the time
>
> Something like below?
>
> if (provided != calculated) {
> if (!is_inode_flag_set(,FI_DATA_CORRUPTED)) {
> printk_ratelimited();
> WARN_ON()
> set_inode_flag(, FI_DATA_CORRUPTED);
> }
> return EFSBADCRC;

No, what if only checksum was wrong? I mean giving WARN_ON() and setting
FSCK_FLAG to stop endless WARN_ON().

> }
>
> 3) don't touch the data, even if it was corrupted, since there's no way to fix it.
>
> It seems that we don't need to change fsck.f2fs finally...
>
> Thanks,
>
> >
> > >
> > > Thanks,
> > >
> > > > corrupted, next fsck will fix the checksum. So, in general, I hope to keep
> > > > the data as is and raise a flag by the checksum.
> > > >
> > > > > not match not match -> corrupted, can not repair
> > > > >
> > > > > Thanks,
> > > > >
> > > > > >
> > > > > > Yes, prepared to update inode layout in fsck.f2fs w/ kernel side change. >
> > > > > > Thanks,
> > > > > >
> > > > > > >
> > > > > > > > + }
> > > > > > > > + }
> > > > > > > > +
> > > > > > > > out_vunmap_cbuf:
> > > > > > > > vm_unmap_ram(dic->cbuf, dic->nr_cpages);
> > > > > > > > out_vunmap_rbuf:
> > > > > > > > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > > > > > > > index 99bcf4b44a9c..2ae254ab7b7d 100644
> > > > > > > > --- a/fs/f2fs/f2fs.h
> > > > > > > > +++ b/fs/f2fs/f2fs.h
> > > > > > > > @@ -147,7 +147,8 @@ struct f2fs_mount_info {
> > > > > > > > /* For compression */
> > > > > > > > unsigned char compress_algorithm; /* algorithm type */
> > > > > > > > - unsigned compress_log_size; /* cluster log size */
> > > > > > > > + unsigned char compress_log_size; /* cluster log size */
> > > > > > > > + bool compress_chksum; /* compressed data chksum */
> > > > > > > > unsigned char compress_ext_cnt; /* extension count */
> > > > > > > > unsigned char extensions[COMPRESS_EXT_NUM][F2FS_EXTENSION_LEN]; /* extensions */
> > > > > > > > };
> > > > > > > > @@ -731,6 +732,7 @@ struct f2fs_inode_info {
> > > > > > > > atomic_t i_compr_blocks; /* # of compressed blocks */
> > > > > > > > unsigned char i_compress_algorithm; /* algorithm type */
> > > > > > > > unsigned char i_log_cluster_size; /* log of cluster size */
> > > > > > > > + unsigned short i_compress_flag; /* compress flag */
> > > > > > > > unsigned int i_cluster_size; /* cluster size */
> > > > > > > > };
> > > > > > > > @@ -1270,9 +1272,15 @@ enum compress_algorithm_type {
> > > > > > > > COMPRESS_MAX,
> > > > > > > > };
> > > > > > > > +enum compress_flag {
> > > > > > > > + COMPRESS_CHKSUM,
> > > > > > > > + COMPRESS_MAX_FLAG,
> > > > > > > > +};
> > > > > > > > +
> > > > > > > > #define COMPRESS_DATA_RESERVED_SIZE 5
> > > > > > > > struct compress_data {
> > > > > > > > __le32 clen; /* compressed data size */
> > > > > > > > + __le32 chksum; /* compressed data chksum */
> > > > > > > > __le32 reserved[COMPRESS_DATA_RESERVED_SIZE]; /* reserved */
> > > > > > > > u8 cdata[]; /* compressed data */
> > > > > > > > };
> > > > > > > > @@ -3882,6 +3890,9 @@ static inline void set_compress_context(struct inode *inode)
> > > > > > > > F2FS_OPTION(sbi).compress_algorithm;
> > > > > > > > F2FS_I(inode)->i_log_cluster_size =
> > > > > > > > F2FS_OPTION(sbi).compress_log_size;
> > > > > > > > + F2FS_I(inode)->i_compress_flag =
> > > > > > > > + F2FS_OPTION(sbi).compress_chksum ?
> > > > > > > > + 1 << COMPRESS_CHKSUM : 0;
> > > > > > > > F2FS_I(inode)->i_cluster_size =
> > > > > > > > 1 << F2FS_I(inode)->i_log_cluster_size;
> > > > > > > > F2FS_I(inode)->i_flags |= F2FS_COMPR_FL;
> > > > > > > > diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> > > > > > > > index 657db2fb6739..de8f7fc89efa 100644
> > > > > > > > --- a/fs/f2fs/inode.c
> > > > > > > > +++ b/fs/f2fs/inode.c
> > > > > > > > @@ -456,6 +456,7 @@ static int do_read_inode(struct inode *inode)
> > > > > > > > le64_to_cpu(ri->i_compr_blocks));
> > > > > > > > fi->i_compress_algorithm = ri->i_compress_algorithm;
> > > > > > > > fi->i_log_cluster_size = ri->i_log_cluster_size;
> > > > > > > > + fi->i_compress_flag = ri->i_compress_flag;
> > > > > > > > fi->i_cluster_size = 1 << fi->i_log_cluster_size;
> > > > > > > > set_inode_flag(inode, FI_COMPRESSED_FILE);
> > > > > > > > }
> > > > > > > > @@ -634,6 +635,8 @@ void f2fs_update_inode(struct inode *inode, struct page *node_page)
> > > > > > > > &F2FS_I(inode)->i_compr_blocks));
> > > > > > > > ri->i_compress_algorithm =
> > > > > > > > F2FS_I(inode)->i_compress_algorithm;
> > > > > > > > + ri->i_compress_flag =
> > > > > > > > + cpu_to_le16(F2FS_I(inode)->i_compress_flag);
> > > > > > > > ri->i_log_cluster_size =
> > > > > > > > F2FS_I(inode)->i_log_cluster_size;
> > > > > > > > }
> > > > > > > > diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> > > > > > > > index 00eff2f51807..f8de4d83a5be 100644
> > > > > > > > --- a/fs/f2fs/super.c
> > > > > > > > +++ b/fs/f2fs/super.c
> > > > > > > > @@ -146,6 +146,7 @@ enum {
> > > > > > > > Opt_compress_algorithm,
> > > > > > > > Opt_compress_log_size,
> > > > > > > > Opt_compress_extension,
> > > > > > > > + Opt_compress_chksum,
> > > > > > > > Opt_atgc,
> > > > > > > > Opt_err,
> > > > > > > > };
> > > > > > > > @@ -214,6 +215,7 @@ static match_table_t f2fs_tokens = {
> > > > > > > > {Opt_compress_algorithm, "compress_algorithm=%s"},
> > > > > > > > {Opt_compress_log_size, "compress_log_size=%u"},
> > > > > > > > {Opt_compress_extension, "compress_extension=%s"},
> > > > > > > > + {Opt_compress_chksum, "compress_chksum"},
> > > > > > > > {Opt_atgc, "atgc"},
> > > > > > > > {Opt_err, NULL},
> > > > > > > > };
> > > > > > > > @@ -934,10 +936,14 @@ static int parse_options(struct super_block *sb, char *options, bool is_remount)
> > > > > > > > F2FS_OPTION(sbi).compress_ext_cnt++;
> > > > > > > > kfree(name);
> > > > > > > > break;
> > > > > > > > + case Opt_compress_chksum:
> > > > > > > > + F2FS_OPTION(sbi).compress_chksum = true;
> > > > > > > > + break;
> > > > > > > > #else
> > > > > > > > case Opt_compress_algorithm:
> > > > > > > > case Opt_compress_log_size:
> > > > > > > > case Opt_compress_extension:
> > > > > > > > + case Opt_compress_chksum:
> > > > > > > > f2fs_info(sbi, "compression options not supported");
> > > > > > > > break;
> > > > > > > > #endif
> > > > > > > > @@ -1523,6 +1529,9 @@ static inline void f2fs_show_compress_options(struct seq_file *seq,
> > > > > > > > seq_printf(seq, ",compress_extension=%s",
> > > > > > > > F2FS_OPTION(sbi).extensions[i]);
> > > > > > > > }
> > > > > > > > +
> > > > > > > > + if (F2FS_OPTION(sbi).compress_chksum)
> > > > > > > > + seq_puts(seq, ",compress_chksum");
> > > > > > > > }
> > > > > > > > static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
> > > > > > > > diff --git a/include/linux/f2fs_fs.h b/include/linux/f2fs_fs.h
> > > > > > > > index a5dbb57a687f..7dc2a06cf19a 100644
> > > > > > > > --- a/include/linux/f2fs_fs.h
> > > > > > > > +++ b/include/linux/f2fs_fs.h
> > > > > > > > @@ -273,7 +273,7 @@ struct f2fs_inode {
> > > > > > > > __le64 i_compr_blocks; /* # of compressed blocks */
> > > > > > > > __u8 i_compress_algorithm; /* compress algorithm */
> > > > > > > > __u8 i_log_cluster_size; /* log of cluster size */
> > > > > > > > - __le16 i_padding; /* padding */
> > > > > > > > + __le16 i_compress_flag; /* compress flag */
> > > > > > > > __le32 i_extra_end[0]; /* for attribute size calculation */
> > > > > > > > } __packed;
> > > > > > > > __le32 i_addr[DEF_ADDRS_PER_INODE]; /* Pointers to data blocks */
> > > > > > > > --
> > > > > > > > 2.26.2
> > > > > > > .
> > > > > > >
> > > > > >
> > > > > >
> > > > > > _______________________________________________
> > > > > > Linux-f2fs-devel mailing list
> > > > > > Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx
> > > > > > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> > > > > > .
> > > > > >
> > > > .
> > > >
> > .
> >

Next message: Rayagonda Kokatanur: "Re: [PATCH v3 5/6] i2c: iproc: handle master read request"
Previous message: patchwork-bot+netdevbpf: "Re: [PATCH v2 net-next] net: skb_vlan_untag(): don't reset transport offset if set by GRO layer"
In reply to: Chao Yu: "Re: [f2fs-dev] [PATCH] f2fs: compress: support chksum"
Next in thread: Chao Yu: "Re: [f2fs-dev] [PATCH] f2fs: compress: support chksum"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]