Re: [PATCH] erofs: fix two loop issues when read page beyond EOF

[Chunhai Guo]

On 2023/7/10 11:37, Gao Xiang wrote:
> On 2023/7/10 11:32, Chunhai Guo wrote:
> > Hi Xiang,
> >
> > On 2023/7/8 17:00, Gao Xiang wrote:
> > > Hi Chunhai,
> > >
> > > On 2023/7/8 14:24, Chunhai Guo wrote:
> > > > When z_erofs_read_folio() reads a page with an offset far beyond EOF, two
> > > > issues may occur:
> > > > - z_erofs_pcluster_readmore() may take a long time to loop when the offset
> > > >      is big enough, which is unnecessary.
> > > >        - For example, it will loop 4691368 times and take about 27 seconds
> > > >          with following case.
> > > >            - offset = 19217289215
> > > >            - inode_size = 1442672
> > > > - z_erofs_do_read_page() may loop infinitely due to the inappropriate
> > > >      truncation in the below statement. Since the offset is 64 bits and
> > > > min_t() truncates the result to 32 bits. The solution is to replace
> > > > unsigned int with another 64-bit type, such as erofs_off_t.
> > > >        cur = end - min_t(unsigned int, offset + end - map->m_la, end);
> > > >        - For example:
> > > >            - offset = 0x400160000
> > > >            - end = 0x370
> > > >            - map->m_la = 0x160370
> > > >            - offset + end - map->m_la = 0x400000000
> > > >            - offset + end - map->m_la = 0x00000000 (truncated as unsigned int)
> > >
> > > Thanks for the catch!
> > >
> > > Could you split these two into two patches?
> > >
> > > how about using:
> > > cur = end - min_t(erofs_off_t, offend + end - map->m_la, end)
> > > for this?
> > >
> > > since cur and end are all [0, PAGE_SIZE - 1] for now, and
> > > folio_size() later.
> >
> > OK. I will split the patch.
> >
> > Sorry that I can not understand what is 'offend' refer to and what do you mean. Could you please describe it more clearly?
>
> Sorry, there is a typo here, I meant 'offset'.
>
> `cur` and `end` both are not exceed 4096 if your page_size
> is 4096.
>
> Does
> cur = end - min_t(erofs_off_t, offset + end - map->m_la, end)
>
> fix your issue?

Yes. I think this will fix this issue. Do you mean the below change is 
unncessary?
>>>> -    unsigned int cur, end, spiltted;
>>>> +    erofs_off_t cur, end;
>>>> +    unsigned int spiltted;

> > > >        - Expected result:
> > > >            - cur = 0
> > > >        - Actual result:
> > > >            - cur = 0x370
> > > >
> > > > Signed-off-by: Chunhai Guo <guochunhai@xxxxxxxx>
> > > > ---
> > > >     fs/erofs/zdata.c | 13 ++++++++++---
> > > >     1 file changed, 10 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/fs/erofs/zdata.c b/fs/erofs/zdata.c
> > > > index 5f1890e309c6..6abbd4510076 100644
> > > > --- a/fs/erofs/zdata.c
> > > > +++ b/fs/erofs/zdata.c
> > > > @@ -972,7 +972,8 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
> > > >         struct erofs_map_blocks *const map = &fe->map;
> > > >         const loff_t offset = page_offset(page);
> > > >         bool tight = true, exclusive;
> > > > -    unsigned int cur, end, spiltted;
> > > > +    erofs_off_t cur, end;
> > > > +    unsigned int spiltted;
> > > >         int err = 0;
> > > >         /* register locked file pages as online pages in pack */
> > > > @@ -1035,7 +1036,7 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
> > > >          */
> > > >         tight &= (fe->mode > Z_EROFS_PCLUSTER_FOLLOWED_NOINPLACE);
> > > > -    cur = end - min_t(unsigned int, offset + end - map->m_la, end);
> > > > +    cur = end - min_t(erofs_off_t, offset + end - map->m_la, end);
> > > >         if (!(map->m_flags & EROFS_MAP_MAPPED)) {
> > > >             zero_user_segment(page, cur, end);
> > > >             goto next_part;
> > > > @@ -1841,7 +1842,7 @@ static void z_erofs_pcluster_readmore(struct z_erofs_decompress_frontend *f,
> > > >         }
> > > >         cur = map->m_la + map->m_llen - 1;
> > > > -    while (cur >= end) {
> > > > +    while ((cur >= end) && (cur < i_size_read(inode))) {
> > > >             pgoff_t index = cur >> PAGE_SHIFT;
> > > >             struct page *page;
> > > > @@ -1876,6 +1877,12 @@ static int z_erofs_read_folio(struct file *file, struct folio *folio)
> > > >         trace_erofs_readpage(page, false);
> > > >         f.headoffset = (erofs_off_t)page->index << PAGE_SHIFT;
> > > > +    /* when trying to read beyond EOF, return zero page directly */
> > > > +    if (f.headoffset >= i_size_read(inode)) {
> > > > +        zero_user_segment(page, 0, PAGE_SIZE);
> > > > +        return 0;
> > > > +    }
> > > Do we really need to optimize this rare case?
> > > I guess the follow readmore fix is enough, thoughts? >
> > > Thanks,
> > > Gao Xiang
> >
> > Since the code is constantly being updated and someone may encounter this bug again, I think we had better fix it if possible.
>
> z_erofs_do_read_page() already covers this case so I'm
> not sure why we need to add another logic here.

You are right. I have removed this logic and sent the patch.

> Thanks,
> Gao Xiang
>
> > Thanks.
> > Guo Chunhai