From: "zhaoyifan (H)" <zhaoyifan28@huawei.com>
To: Zhiguo Niu <niuzhiguo84@gmail.com>
Cc: <linux-erofs@lists.ozlabs.org>, <hsiangkao@linux.alibaba.com>,
<zhukeqian1@huawei.com>
Subject: Re: [PATCH] erofs-utils: lib: fix ztailpacking fallback across lclusters
Date: Tue, 21 Jul 2026 14:55:21 +0800 [thread overview]
Message-ID: <5700f79d-2cfb-4408-80d3-509ac154bef0@huawei.com> (raw)
In-Reply-To: <CAHJ8P3JK5gKS=RH2b=waY7g1AJ7S6ZHzDV0asg46pkzHsWzMHw@mail.gmail.com>
On 2026/7/20 19:42, Zhiguo Niu wrote:
> Yifan Zhao <zhaoyifan28@huawei.com> 于2026年7月9日周四 20:36写道:
>> With ztailpacking, the final compressed pcluster is first stored as
>> inline data. If the inode metadata area cannot hold it, mkfs falls back
>> to a normal tail block and drops the inline pcluster marker.
>>
>> The current fallback path assumes that the inline tail pcluster belongs
>> to the EOF lcluster. That is not always true: the tail pcluster can
>> start in the previous lcluster and end at EOF, while its raw size still
>> fits in one block. In that case, patching the EOF lcluster is
>> semantically wrong.
>>
>> Let's keep raw tail data whenever it fits in one block, and convert the
>> corresponding lcluster index to PLAIN during fallback.
>>
>> Reported-by: Alberto Salvia Novella <es20490446e@gmail.com>
>> Closes: https://github.com/erofs/erofs-utils/issues/51
>> Assisted-by: Codex:GPT-5.5
>> Signed-off-by: Yifan Zhao <zhaoyifan28@huawei.com>
>> ---
> Hi Yifan,
> I tested this patch to focus on the issue fixed by commit
> 277a42502a7a, and it passed.
> But I have some questions:
>> include/erofs/internal.h | 5 +-
>> lib/compress.c | 109 ++++++++++++++++++++++++++++-----------
>> 2 files changed, 83 insertions(+), 31 deletions(-)
>>
>> diff --git a/include/erofs/internal.h b/include/erofs/internal.h
>> index 2cc9cc8..bdde41f 100644
>> --- a/include/erofs/internal.h
>> +++ b/include/erofs/internal.h
>> @@ -212,8 +212,11 @@ struct erofs_diskbuf;
>>
>> enum erofs_idata_type {
>> EROFS_IDATA_TYPE_RAW,
>> - EROFS_IDATA_TYPE_COMPRESSED_DEFAULT,
>> + EROFS_IDATA_TYPE_COMPRESSED,
>> + /* compressed idata follows a final 2B compacted index pack */
>> EROFS_IDATA_TYPE_COMPRESSED_END_OF_2B,
>> + /* compressed idata follows a final single-entry 4B pack after a 2B pack */
>> + EROFS_IDATA_TYPE_COMPRESSED_4B1_PREV2B,
>> };
>>
>> #define EROFS_I_BLKADDR_DEV_ID_BIT 48
>> diff --git a/lib/compress.c b/lib/compress.c
>> index f7ad5a1..ec90f65 100644
>> --- a/lib/compress.c
>> +++ b/lib/compress.c
>> @@ -483,7 +483,7 @@ static int z_erofs_fill_inline_data(struct erofs_inode *inode, void *data,
>> {
>> inode->z_advise |= Z_EROFS_ADVISE_INLINE_PCLUSTER;
>> inode->idata_size = len;
>> - inode->idata_type = EROFS_IDATA_TYPE_COMPRESSED_DEFAULT;
>> + inode->idata_type = EROFS_IDATA_TYPE_COMPRESSED;
>>
>> inode->idata = malloc(inode->idata_size);
>> if (!inode->idata)
>> @@ -664,7 +664,7 @@ frag_packing:
>> ictx->fragemitted = true;
>> /* tailpcluster should be less than 1 block */
>> } else if (may_inline && len == e->length && compressedsize < blksz) {
>> - if (ctx->clusterofs + len <= blksz) {
> Shouldn't this condition restrict the `tail pcluster` so that it
> corresponds to only eof`lcluster`?
> and we drop the inline pcluster just when eof_tailraw is not null.
> Thanks!
Hi Zhiguo,
A tail pcluster may start in the previous lcluster and end in the EOF
lcluster while its whole raw payload still fits in one physical block.
In that case `ctx->clusterofs + len > blksz`, but `len <= blksz`, but it
should still be converted into a single PLAIN pcluster.
This patch targets this edge case so I think change the if statement
here is necessary?
Thanks,
Yifan
>> + if (len <= blksz) {
>> inode->eof_tailraw = malloc(len);
>> if (!inode->eof_tailraw)
>> return -ENOMEM;
>> @@ -962,6 +962,14 @@ int z_erofs_convert_to_compacted_format(struct erofs_inode *inode,
>> dummy_head = true;
>> }
>>
>> + if (inode->idata_size) {
>> + if (compacted_2b && !compacted_4b_end)
>> + inode->idata_type = EROFS_IDATA_TYPE_COMPRESSED_END_OF_2B;
>> + else if (compacted_2b && compacted_4b_end == 1)
>> + inode->idata_type =
>> + EROFS_IDATA_TYPE_COMPRESSED_4B1_PREV2B;
>> + }
>> +
>> /* generate compacted_4b_initial */
>> while (compacted_4b_initial) {
>> in = parse_legacy_indexes(cv, 2, in);
>> @@ -974,8 +982,6 @@ int z_erofs_convert_to_compacted_format(struct erofs_inode *inode,
>>
>> /* generate compacted_2b */
>> if (compacted_2b) {
>> - if (!compacted_4b_end && inode->idata_size)
>> - inode->idata_type = EROFS_IDATA_TYPE_COMPRESSED_END_OF_2B;
>> do {
>> in = parse_legacy_indexes(cv, 16, in);
>> out = write_compacted_indexes(out, cv, &blkaddr,
>> @@ -1205,11 +1211,65 @@ out:
>> return metabuf;
>> }
>>
>> +static void z_erofs_patch_tail_compacted_index(struct erofs_inode *inode,
>> + bool previous,
>> + unsigned int type)
>> +{
>> + const unsigned int totalidx = BLK_ROUND_UP(inode->sbi, inode->i_size);
>> + const unsigned int lobits = max_t(unsigned int, inode->z_lclusterbits,
>> + ilog2(Z_EROFS_LI_D0_CBLKCNT) + 1U);
>> + u8 *base = inode->compressmeta;
>> + u8 *pack = base + inode->extent_isize;
>> + unsigned int bitpos, bitoff;
>> + u8 *out;
>> + u32 v;
>> +
>> + DBG_BUGON(!totalidx);
>> + DBG_BUGON(inode->z_lclusterbits > 14);
>> +
>> + if (inode->idata_type == EROFS_IDATA_TYPE_COMPRESSED_END_OF_2B) {
>> + pack -= 32;
>> + bitpos = 14 * (previous ? 14 : 15);
>> + goto out;
>> + }
>> +
>> + /* Last compacted index pack is 4B */
>> + pack -= 8;
>> + if (!(totalidx & 1)) {
>> + bitpos = previous ? 0 : 16;
>> + goto out;
>> + }
>> +
>> + if (!previous) {
>> + bitpos = 0;
>> + goto out;
>> + }
>> +
>> + /* Second to last compacted index pack is 2B */
>> + if (inode->idata_type == EROFS_IDATA_TYPE_COMPRESSED_4B1_PREV2B) {
>> + pack -= 32;
>> + bitpos = 14 * 15;
>> + } else {
>> + pack -= 8;
>> + bitpos = 16;
>> + }
>> +out:
>> + DBG_BUGON(pack < base);
>> + bitoff = bitpos & 7;
>> + out = pack + bitpos / 8;
>> + v = get_unaligned_le32(out);
>> + v &= ~(Z_EROFS_LI_LCLUSTER_TYPE_MASK << (lobits + bitoff));
>> + v |= type << (lobits + bitoff);
>> + put_unaligned_le32(v, out);
>> +}
>> +
>> void z_erofs_drop_inline_pcluster(struct erofs_inode *inode)
>> {
>> struct erofs_sb_info *sbi = inode->sbi;
>> const unsigned int type = Z_EROFS_LCLUSTER_TYPE_PLAIN;
>> struct z_erofs_map_header *h = inode->compressmeta;
>> + erofs_off_t rawstart;
>> + erofs_blk_t head_lcn, eof_lcn;
>>
>> h->h_advise = cpu_to_le16(le16_to_cpu(h->h_advise) &
>> ~Z_EROFS_ADVISE_INLINE_PCLUSTER);
>> @@ -1218,38 +1278,27 @@ void z_erofs_drop_inline_pcluster(struct erofs_inode *inode)
>> if (!inode->eof_tailraw)
>> return;
>> DBG_BUGON(inode->idata_type == EROFS_IDATA_TYPE_RAW);
>> + DBG_BUGON(!inode->i_size);
>> + DBG_BUGON(inode->eof_tailrawsize > erofs_blksiz(sbi));
>> + DBG_BUGON(inode->eof_tailrawsize > inode->i_size);
>> +
>> + rawstart = inode->i_size - inode->eof_tailrawsize;
>> + head_lcn = rawstart >> sbi->blkszbits;
>> + eof_lcn = (inode->i_size - 1) >> sbi->blkszbits;
>> + DBG_BUGON(head_lcn != eof_lcn && head_lcn + 1 != eof_lcn);
>>
>> - /* patch the EOF lcluster to uncompressed type first */
>> + /* patch the tail pcluster head to uncompressed type first */
>> if (inode->datalayout == EROFS_INODE_COMPRESSED_FULL) {
>> struct z_erofs_lcluster_index *di =
>> - (inode->compressmeta + inode->extent_isize) -
>> - sizeof(struct z_erofs_lcluster_index);
>> + (void *)((u8 *)inode->compressmeta +
>> + Z_EROFS_LEGACY_MAP_HEADER_SIZE +
>> + head_lcn *
>> + sizeof(struct z_erofs_lcluster_index));
>>
>> di->di_advise = cpu_to_le16(type);
>> } else if (inode->datalayout == EROFS_INODE_COMPRESSED_COMPACT) {
>> - /* handle the last compacted 4B/2B pack */
>> - unsigned int lclusterbits = inode->z_lclusterbits;
>> - unsigned int lobits, eofs, base, pos, v;
>> - u8 *out;
>> -
>> - lobits = max(lclusterbits, ilog2(Z_EROFS_LI_D0_CBLKCNT) + 1U);
>> -
>> - if (inode->idata_type == EROFS_IDATA_TYPE_COMPRESSED_DEFAULT) {
>> - eofs = inode->extent_isize -
>> - (4 << (BLK_ROUND_UP(sbi, inode->i_size) & 1));
>> - base = round_down(eofs, 8);
>> - pos = 16 /* encodebits */ * ((eofs - base) / 4);
>> - out = inode->compressmeta + base + pos / 8;
>> - } else {
>> - out = inode->compressmeta + inode->extent_isize -
>> - sizeof(__le32) - sizeof(__le16);
>> - lobits = 16 - 14 /* encodebits */ + lobits;
>> - }
>> -
>> - v = (get_unaligned_le16(out) & (BIT(lobits) - 1)) |
>> - (type << lobits);
>> - *out = v & 0xff;
>> - *(out + 1) = v >> 8;
>> + z_erofs_patch_tail_compacted_index(inode, head_lcn != eof_lcn,
>> + type);
>> } else {
>> DBG_BUGON(1);
>> return;
>> --
>> 2.47.3
>>
>>
next prev parent reply other threads:[~2026-07-21 6:55 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-07-09 12:34 [PATCH] erofs-utils: lib: fix ztailpacking fallback across lclusters Yifan Zhao
2026-07-20 11:42 ` Zhiguo Niu
2026-07-21 6:55 ` zhaoyifan (H) [this message]
2026-07-21 6:57 ` Gao Xiang
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=5700f79d-2cfb-4408-80d3-509ac154bef0@huawei.com \
--to=zhaoyifan28@huawei.com \
--cc=hsiangkao@linux.alibaba.com \
--cc=linux-erofs@lists.ozlabs.org \
--cc=niuzhiguo84@gmail.com \
--cc=zhukeqian1@huawei.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