[PATCH] erofs-utils: code for superblock checksum calculation.
Gao Xiang
hsiangkao at aol.com
Thu Oct 10 01:27:04 AEDT 2019
Hi Pratik,
On Wed, Oct 09, 2019 at 07:44:08PM +0530, Pratik Shinde wrote:
> Hello Gao,
>
> I have few questions regarding the checksum feature:
> 1) Are we still keeping it as a feature. (i.e enabled by default . but can
> be disabled during mkfs)
Yes, I think it should be a new compatible feature suggested by Richard
and enabled by default (this feature should be marked in on-disk
super_block...)
> 2) For small images we are going to cksum the entire image. what could be
> the maximum size of 'small' images.
I think we can checksum the first 4k for now in mkfs by default
(or only checksum 4k as the first step), and then it can be
configured to a larger number n*4k. (But for the kernel side,
it has to be fully implemented due to forward compatibility.)
>
> also, I think we don't have routines to read buffers back into the memory.
> I will write them first and send a patch.
Yes, I remembered that you once wrote a similar function, that is ok.
And it's no rush, we have more than a month to develop pending features
for linux-5.5. Just a reminder because we can prepare them for now :)
Thanks,
Gao Xiang
>
> -Pratik
>
> On Wed, Oct 9, 2019 at 2:14 PM Gao Xiang <gaoxiang25 at huawei.com> wrote:
>
> > Hi Pratik,
> >
> > On Wed, Oct 09, 2019 at 01:54:40PM +0530, Pratik Shinde wrote:
> > > Hello Gao,
> > >
> > > Yes I would like to work on pending features.
> > > Also please let us know the new development items.
> >
> > Thanks, I will post later after gathering the first round
> > suggestions to mailing lists. We have time to improve this
> > filesystem. :)
> >
> > Thanks,
> > Gao Xiang
> >
> > >
> > > --Pratik
> > >
> > > On Wed, 9 Oct, 2019, 12:24 PM Gao Xiang, <gaoxiang25 at huawei.com> wrote:
> > >
> > > > Hi Pratik,
> > > >
> > > > On Wed, Oct 09, 2019 at 11:59:01AM +0530, Pratik Shinde wrote:
> > > > > Hi Gao,
> > > > >
> > > > > Yes I can work on it. Sorry I missed this mail. I think your
> > approach is
> > > > > good. Let me go through it one more time and reply back.
> > > >
> > > > Thanks for your reply and interest :)
> > > > I think we can complete all pending features together
> > > > if you have some time on this stuff. (fsdebug utility is
> > > > on the way as well...)
> > > >
> > > > BTW, I'm now investigating new high CR algorithm (very likely
> > > > XZ) as well, it will be likely a RFC version in this round for
> > > > wider scenarios and later decompression subsystem.
> > > >
> > > > Preliminary TODO lists will be discussed in this year China
> > > > Linux Storage & Filesystem workshop (next week) and will be
> > > > posted to mailing lists for further wider discussion (if more
> > > > folks have interest in developing it) as well. :)
> > > >
> > > > Thanks,
> > > > Gao Xiang
> > > >
> > > > >
> > > > > --Pratik
> > > > >
> > > > > On Sun, 6 Oct, 2019, 11:09 AM Gao Xiang, <hsiangkao at aol.com> wrote:
> > > > >
> > > > > > Hi Pratik,
> > > > > >
> > > > > > On Sat, Aug 24, 2019 at 08:26:28PM +0530, Pratik Shinde wrote:
> > > > > > > Hi Gao,
> > > > > > >
> > > > > > > I completely understand your concern.You can drop this patch for
> > now.
> > > > > > > Once erofs makes it 'fs/' please do reconsider implementing it.
> > > > > >
> > > > > > I think we can work on this pending feature for v5.5 now.
> > > > > > My idea is to add an extra field in erofs_super_block to
> > > > > > indicate the number of blocks (4k) for checking.
> > > > > >
> > > > > > So for small images, this feature can checksum the whole image at
> > mount
> > > > > > time,
> > > > > > and for large images, this feature can be used to checksum the
> > super
> > > > block
> > > > > > only
> > > > > > (and then use verficiation subsystem to verify on-the-fly in the
> > > > future...)
> > > > > >
> > > > > > The following workflow is for erofs-utils, I think it's
> > > > > > 1) erofs_mkfs_update_super_block with checksum = 0
> > > > > > 2) erofs_bflush(NULL)
> > > > > > 3) reread the corresponding blocks and calculate checksum;
> > > > > > 4) write checksum to erofs_super_block;
> > > > > >
> > > > > > Does it sound reasonable? and do you still have interest and
> > > > > > time for this? Looking forword to your reply...
> > > > > >
> > > > > > Thanks,
> > > > > > Gao Xiang
> > > > > >
> > > > > > >
> > > > > > > One more thing, can we still send non feature patches?
> > > > > > >
> > > > > > > --Pratik
> > > > > > >
> > > > > > >
> > > > > > > On Sat, 24 Aug, 2019, 7:30 PM Gao Xiang, <hsiangkao at aol.com>
> > wrote:
> > > > > > >
> > > > > > > > Hi Pratik,
> > > > > > > >
> > > > > > > > On Sat, Aug 24, 2019 at 06:08:03PM +0530, Pratik Shinde wrote:
> > > > > > > > > Adding code for superblock checksum calculation.
> > > > > > > > >
> > > > > > > > > incorporated the changes suggested in previous patch.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Pratik Shinde <pratikshinde320 at gmail.com>
> > > > > > > >
> > > > > > > > Thanks for your v2 patch.
> > > > > > > >
> > > > > > > > Actually, I have some concern about the length of checksum,
> > > > > > > > sizeof(struct erofs_super_block) could be changed in the
> > > > > > > > later version, it's bad for EROFS future scalablity.
> > > > > > > >
> > > > > > > > And I tend not to add another on-disk field to record
> > > > > > > > the size of erofs_super_block as well, because the old
> > > > > > > > Linux kernel cannot handle more about the new size,
> > > > > > > > so it has little use except for checksum calculation.
> > > > > > > >
> > > > > > > > Few hours ago, I discussed with Chao about this concern,
> > > > > > > > I think this feature can be changed to do multiple-block
> > > > > > > > checksum at the mount time, e.g:
> > > > > > > > - for small images, we can check the whole image once
> > > > > > > > at the mount time;
> > > > > > > > - for the large image, we can check the superblock
> > > > > > > > at the mount time, the rest can be handled by
> > > > > > > > block-based verification layer.
> > > > > > > >
> > > > > > > > But we agreed that don't add this for this round
> > > > > > > > since it's quite a new feature.
> > > > > > > >
> > > > > > > > All in all, it's a new feature since we are addressing moving
> > > > > > > > out of staging for this round. I tend to postpone this feature
> > > > > > > > for now. I understand that you are very interested in EROFS.
> > > > > > > > Considering EROFS current staging status, it's not such a place
> > > > > > > > to add new features at all! I have marked your patch down and
> > > > > > > > I will work with you later. Hope to get your understanding...
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > Gao Xiang
> > > > > > > >
> > > > > > > > > ---
> > > > > > > > > include/erofs/config.h | 1 +
> > > > > > > > > include/erofs_fs.h | 10 ++++++++
> > > > > > > > > mkfs/main.c | 64
> > > > > > > > +++++++++++++++++++++++++++++++++++++++++++++++++-
> > > > > > > > > 3 files changed, 74 insertions(+), 1 deletion(-)
> > > > > > > > >
> > > > > > > > > diff --git a/include/erofs/config.h b/include/erofs/config.h
> > > > > > > > > index 05fe6b2..40cd466 100644
> > > > > > > > > --- a/include/erofs/config.h
> > > > > > > > > +++ b/include/erofs/config.h
> > > > > > > > > @@ -22,6 +22,7 @@ struct erofs_configure {
> > > > > > > > > char *c_src_path;
> > > > > > > > > char *c_compr_alg_master;
> > > > > > > > > int c_compr_level_master;
> > > > > > > > > + int c_feature_flags;
> > > > > > > > > };
> > > > > > > > >
> > > > > > > > > extern struct erofs_configure cfg;
> > > > > > > > > diff --git a/include/erofs_fs.h b/include/erofs_fs.h
> > > > > > > > > index 601b477..9ac2635 100644
> > > > > > > > > --- a/include/erofs_fs.h
> > > > > > > > > +++ b/include/erofs_fs.h
> > > > > > > > > @@ -20,6 +20,16 @@
> > > > > > > > > #define EROFS_REQUIREMENT_LZ4_0PADDING 0x00000001
> > > > > > > > > #define EROFS_ALL_REQUIREMENTS
> > > > > > > > EROFS_REQUIREMENT_LZ4_0PADDING
> > > > > > > > >
> > > > > > > > > +/*
> > > > > > > > > + * feature definations.
> > > > > > > > > + */
> > > > > > > > > +#define EROFS_DEFAULT_FEATURES
> > > > EROFS_FEATURE_SB_CHKSUM
> > > > > > > > > +#define EROFS_FEATURE_SB_CHKSUM 0x0001
> > > > > > > > > +
> > > > > > > > > +
> > > > > > > > > +#define EROFS_HAS_COMPAT_FEATURE(super,mask) \
> > > > > > > > > + ( le32_to_cpu((super)->features) & (mask) )
> > > > > > > > > +
> > > > > > > > > struct erofs_super_block {
> > > > > > > > > /* 0 */__le32 magic; /* in the little endian */
> > > > > > > > > /* 4 */__le32 checksum; /* crc32c(super_block) */
> > > > > > > > > diff --git a/mkfs/main.c b/mkfs/main.c
> > > > > > > > > index f127fe1..355fd2c 100644
> > > > > > > > > --- a/mkfs/main.c
> > > > > > > > > +++ b/mkfs/main.c
> > > > > > > > > @@ -31,6 +31,45 @@ static void usage(void)
> > > > > > > > > fprintf(stderr, " -EX[,...] X=extended options\n");
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > +#define CRCPOLY 0x82F63B78
> > > > > > > > > +static inline u32 crc32c(u32 seed, unsigned char const *in,
> > > > size_t
> > > > > > len)
> > > > > > > > > +{
> > > > > > > > > + int i;
> > > > > > > > > + u32 crc = seed;
> > > > > > > > > +
> > > > > > > > > + while (len--) {
> > > > > > > > > + crc ^= *in++;
> > > > > > > > > + for (i = 0; i < 8; i++) {
> > > > > > > > > + crc = (crc >> 1) ^ ((crc & 1) ?
> > CRCPOLY :
> > > > 0);
> > > > > > > > > + }
> > > > > > > > > + }
> > > > > > > > > + erofs_dump("calculated crc: 0x%x\n", crc);
> > > > > > > > > + return crc;
> > > > > > > > > +}
> > > > > > > > > +
> > > > > > > > > +char *feature_opts[] = {
> > > > > > > > > + "nosbcrc", NULL
> > > > > > > > > +};
> > > > > > > > > +#define O_SB_CKSUM 0
> > > > > > > > > +
> > > > > > > > > +static int parse_feature_subopts(char *opts)
> > > > > > > > > +{
> > > > > > > > > + char *arg;
> > > > > > > > > +
> > > > > > > > > + cfg.c_feature_flags = EROFS_DEFAULT_FEATURES;
> > > > > > > > > + while (*opts != '\0') {
> > > > > > > > > + switch(getsubopt(&opts, feature_opts, &arg)) {
> > > > > > > > > + case O_SB_CKSUM:
> > > > > > > > > + cfg.c_feature_flags |=
> > > > > > (~EROFS_FEATURE_SB_CHKSUM);
> > > > > > > > > + break;
> > > > > > > > > + default:
> > > > > > > > > + erofs_err("incorrect suboption");
> > > > > > > > > + return -EINVAL;
> > > > > > > > > + }
> > > > > > > > > + }
> > > > > > > > > + return 0;
> > > > > > > > > +}
> > > > > > > > > +
> > > > > > > > > static int parse_extended_opts(const char *opts)
> > > > > > > > > {
> > > > > > > > > #define MATCH_EXTENTED_OPT(opt, token, keylen) \
> > > > > > > > > @@ -79,7 +118,8 @@ static int mkfs_parse_options_cfg(int
> > argc,
> > > > char
> > > > > > > > *argv[])
> > > > > > > > > {
> > > > > > > > > int opt, i;
> > > > > > > > >
> > > > > > > > > - while ((opt = getopt(argc, argv, "d:z:E:")) != -1) {
> > > > > > > > > + cfg.c_feature_flags = EROFS_DEFAULT_FEATURES;
> > > > > > > > > + while ((opt = getopt(argc, argv, "d:z:E:o:")) != -1) {
> > > > > > > > > switch (opt) {
> > > > > > > > > case 'z':
> > > > > > > > > if (!optarg) {
> > > > > > > > > @@ -113,6 +153,12 @@ static int mkfs_parse_options_cfg(int
> > argc,
> > > > char
> > > > > > > > *argv[])
> > > > > > > > > return opt;
> > > > > > > > > break;
> > > > > > > > >
> > > > > > > > > + case 'O':
> > > > > > > > > + opt = parse_feature_subopts(optarg);
> > > > > > > > > + if (opt)
> > > > > > > > > + return opt;
> > > > > > > > > + break;
> > > > > > > > > +
> > > > > > > > > default: /* '?' */
> > > > > > > > > return -EINVAL;
> > > > > > > > > }
> > > > > > > > > @@ -144,6 +190,15 @@ static int mkfs_parse_options_cfg(int
> > argc,
> > > > char
> > > > > > > > *argv[])
> > > > > > > > > return 0;
> > > > > > > > > }
> > > > > > > > >
> > > > > > > > > +u32 erofs_superblock_checksum(struct erofs_super_block *sb)
> > > > > > > > > +{
> > > > > > > > > + u32 crc;
> > > > > > > > > + crc = crc32c(~0, (const unsigned char *)sb,
> > > > > > > > > + sizeof(struct erofs_super_block));
> > > > > > > > > + erofs_dump("superblock checksum: 0x%x\n", crc);
> > > > > > > > > + return crc;
> > > > > > > > > +}
> > > > > > > > > +
> > > > > > > > > int erofs_mkfs_update_super_block(struct erofs_buffer_head
> > *bh,
> > > > > > > > > erofs_nid_t root_nid)
> > > > > > > > > {
> > > > > > > > > @@ -155,6 +210,7 @@ int erofs_mkfs_update_super_block(struct
> > > > > > > > erofs_buffer_head *bh,
> > > > > > > > > .meta_blkaddr = sbi.meta_blkaddr,
> > > > > > > > > .xattr_blkaddr = 0,
> > > > > > > > > .requirements = cpu_to_le32(sbi.requirements),
> > > > > > > > > + .features = cpu_to_le32(cfg.c_feature_flags),
> > > > > > > > > };
> > > > > > > > > const unsigned int sb_blksize =
> > > > > > > > > round_up(EROFS_SUPER_END, EROFS_BLKSIZ);
> > > > > > > > > @@ -169,6 +225,12 @@ int erofs_mkfs_update_super_block(struct
> > > > > > > > erofs_buffer_head *bh,
> > > > > > > > > sb.blocks = cpu_to_le32(erofs_mapbh(NULL, true));
> > > > > > > > > sb.root_nid = cpu_to_le16(root_nid);
> > > > > > > > >
> > > > > > > > > + if (EROFS_HAS_COMPAT_FEATURE(&sb,
> > > > EROFS_FEATURE_SB_CHKSUM)) {
> > > > > > > > > + sb.checksum = 0;
> > > > > > > > > + u32 crc = erofs_superblock_checksum(&sb);
> > > > > > > > > + sb.checksum = cpu_to_le32(crc);
> > > > > > > > > + }
> > > > > > > > > +
> > > > > > > > > buf = calloc(sb_blksize, 1);
> > > > > > > > > if (!buf) {
> > > > > > > > > erofs_err("Failed to allocate memory for sb:
> > %s",
> > > > > > > > > --
> > > > > > > > > 2.9.3
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > >
> >
More information about the Linux-erofs
mailing list