Re: [PATCH v6 5/5] libfs: Use d_children list to iterate simple_offset directories

[yangerkun <yangerkun@huaweicloud.com>]

在 2024/12/23 22:44, Chuck Lever 写道:
> On 12/23/24 9:21 AM, yangerkun wrote:
>>
>>
>> 在 2024/12/20 23:33, cel@kernel.org 写道:
>>> From: Chuck Lever <chuck.lever@oracle.com>
>>>
>>> The mtree mechanism has been effective at creating directory offsets
>>> that are stable over multiple opendir instances. However, it has not
>>> been able to handle the subtleties of renames that are concurrent
>>> with readdir.
>>>
>>> Instead of using the mtree to emit entries in the order of their
>>> offset values, use it only to map incoming ctx->pos to a starting
>>> entry. Then use the directory's d_children list, which is already
>>> maintained properly by the dcache, to find the next child to emit.
>>>
>>> One of the sneaky things about this is that when the mtree-allocated
>>> offset value wraps (which is very rare), looking up ctx->pos++ is
>>> not going to find the next entry; it will return NULL. Instead, by
>>> following the d_children list, the offset values can appear in any
>>> order but all of the entries in the directory will be visited
>>> eventually.
>>>
>>> Note also that the readdir() is guaranteed to reach the tail of this
>>> list. Entries are added only at the head of d_children, and readdir
>>> walks from its current position in that list towards its tail.
>>>
>>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>>> ---
>>>   fs/libfs.c | 84 +++++++++++++++++++++++++++++++++++++-----------------
>>>   1 file changed, 58 insertions(+), 26 deletions(-)
>>>
>>> diff --git a/fs/libfs.c b/fs/libfs.c
>>> index 5c56783c03a5..f7ead02062ad 100644
>>> --- a/fs/libfs.c
>>> +++ b/fs/libfs.c
>>> @@ -247,12 +247,13 @@ EXPORT_SYMBOL(simple_dir_inode_operations);
>>>   /* simple_offset_add() allocation range */
>>>   enum {
>>> -    DIR_OFFSET_MIN        = 2,
>>> +    DIR_OFFSET_MIN        = 3,
>>>       DIR_OFFSET_MAX        = LONG_MAX - 1,
>>>   };
>>>   /* simple_offset_add() never assigns these to a dentry */
>>>   enum {
>>> +    DIR_OFFSET_FIRST    = 2,        /* Find first real entry */
>>>       DIR_OFFSET_EOD        = LONG_MAX,    /* Marks EOD */
>>>   };
>>> @@ -458,51 +459,82 @@ static loff_t offset_dir_llseek(struct file 
>>> *file, loff_t offset, int whence)
>>>       return vfs_setpos(file, offset, LONG_MAX);
>>>   }
>>> -static struct dentry *offset_find_next(struct offset_ctx *octx, 
>>> loff_t offset)
>>> +static struct dentry *find_positive_dentry(struct dentry *parent,
>>> +                       struct dentry *dentry,
>>> +                       bool next)
>>>   {
>>> -    MA_STATE(mas, &octx->mt, offset, offset);
>>> +    struct dentry *found = NULL;
>>> +
>>> +    spin_lock(&parent->d_lock);
>>> +    if (next)
>>> +        dentry = d_next_sibling(dentry);
>>> +    else if (!dentry)
>>> +        dentry = d_first_child(parent);
>>> +    hlist_for_each_entry_from(dentry, d_sib) {
>>> +        if (!simple_positive(dentry))
>>> +            continue;
>>> +        spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
>>> +        if (simple_positive(dentry))
>>> +            found = dget_dlock(dentry);
>>> +        spin_unlock(&dentry->d_lock);
>>> +        if (likely(found))
>>> +            break;
>>> +    }
>>> +    spin_unlock(&parent->d_lock);
>>> +    return found;
>>> +}
>>> +
>>> +static noinline_for_stack struct dentry *
>>> +offset_dir_lookup(struct dentry *parent, loff_t offset)
>>> +{
>>> +    struct inode *inode = d_inode(parent);
>>> +    struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
>>>       struct dentry *child, *found = NULL;
>>> -    rcu_read_lock();
>>> -    child = mas_find(&mas, DIR_OFFSET_MAX);
>>> -    if (!child)
>>> -        goto out;
>>> -    spin_lock(&child->d_lock);
>>> -    if (simple_positive(child))
>>> -        found = dget_dlock(child);
>>> -    spin_unlock(&child->d_lock);
>>> -out:
>>> -    rcu_read_unlock();
>>> +    MA_STATE(mas, &octx->mt, offset, offset);
>>> +
>>> +    if (offset == DIR_OFFSET_FIRST)
>>> +        found = find_positive_dentry(parent, NULL, false);
>>> +    else {
>>> +        rcu_read_lock();
>>> +        child = mas_find(&mas, DIR_OFFSET_MAX);
>>
>> Can this child be NULL?
> 
> Yes, this mas_find() call can return NULL. find_positive_dentry() should
> then return NULL. Kind of subtle.
> 
> 
>> Like we delete some file after first readdir, maybe we should break 
>> here, or we may rescan all dentry and return them to userspace again?
> 
> You mean to deal with the case where the "next" entry has an offset
> that is lower than @offset? mas_find() will return the entry in the
> tree that is "at or after" mas->index.
> 
> I'm not sure either "break" or returning repeats is safe. But, now that
> you point it out, this function probably does need additional logic to
> deal with the offset wrap case.
> 
> But since this logic already exists here, IMO it is reasonable to leave
> that to be addressed by a subsequent patch. So far there aren't any
> regression test failures that warn of a user-visible problem the way it
> is now.

Sorry for the confusing, the case I am talking is something like below:

mkdir /tmp/dir && cd /tmp/dir
touch file1 # offset is 3
touch file2 # offset is 4
touch file3 # offset is 5
touch file4 # offset is 6
touch file5 # offset is 7
first readdir and get file5 file4 file3 file2 #ctx->pos is 3, which
means we will get file1 for second readdir

unlink file1 # can not get entry for ctx->pos == 3

second readdir # offset_dir_lookup will use mas_find but return NULL,
and we will get file5 file4 file3 file2 again?


And for the offset wrap case, I prefer it's safe with your patch if we 
won't unlink file between two readdir. The second readdir will use an
active ctx->pos which means there is a active dentry attach to this
ctx->pos. find_positive_dentry will stop once we meet the last child.


I am not sure if I understand correctly, if not, please point out!

Thanks!

> 
> 
>>> +        found = find_positive_dentry(parent, child, false);
>>> +        rcu_read_unlock();
>>> +    }
>>>       return found;
>>>   }
>>>   static bool offset_dir_emit(struct dir_context *ctx, struct dentry 
>>> *dentry)
>>>   {
>>>       struct inode *inode = d_inode(dentry);
>>> -    long offset = dentry2offset(dentry);
>>> -    return ctx->actor(ctx, dentry->d_name.name, dentry->d_name.len, 
>>> offset,
>>> -              inode->i_ino, fs_umode_to_dtype(inode->i_mode));
>>> +    return dir_emit(ctx, dentry->d_name.name, dentry->d_name.len,
>>> +            inode->i_ino, fs_umode_to_dtype(inode->i_mode));
>>>   }
>>> -static void offset_iterate_dir(struct inode *inode, struct 
>>> dir_context *ctx)
>>> +static void offset_iterate_dir(struct file *file, struct dir_context 
>>> *ctx)
>>>   {
>>> -    struct offset_ctx *octx = inode->i_op->get_offset_ctx(inode);
>>> +    struct dentry *dir = file->f_path.dentry;
>>>       struct dentry *dentry;
>>> +    dentry = offset_dir_lookup(dir, ctx->pos);
>>> +    if (!dentry)
>>> +        goto out_eod;
>>>       while (true) {
>>> -        dentry = offset_find_next(octx, ctx->pos);
>>> -        if (!dentry)
>>> -            goto out_eod;
>>> +        struct dentry *next;
>>> -        if (!offset_dir_emit(ctx, dentry)) {
>>> -            dput(dentry);
>>> +        ctx->pos = dentry2offset(dentry);
>>> +        if (!offset_dir_emit(ctx, dentry))
>>>               break;
>>> -        }
>>> -        ctx->pos = dentry2offset(dentry) + 1;
>>> +        next = find_positive_dentry(dir, dentry, true);
>>>           dput(dentry);
>>> +
>>> +        if (!next)
>>> +            goto out_eod;
>>> +        dentry = next;
>>>       }
>>> +    dput(dentry);
>>>       return;
>>>   out_eod:
>>> @@ -541,7 +573,7 @@ static int offset_readdir(struct file *file, 
>>> struct dir_context *ctx)
>>>       if (!dir_emit_dots(file, ctx))
>>>           return 0;
>>>       if (ctx->pos != DIR_OFFSET_EOD)
>>> -        offset_iterate_dir(d_inode(dir), ctx);
>>> +        offset_iterate_dir(file, ctx);
>>>       return 0;
>>>   }
>>
> 
>