linux-ext4 - Re: [PATCH -next RFC 01/14] block: add some bdev apis

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <d195aba8-7b89-698f-b7a0-06b87ae01c21@huaweicloud.com>
Date: Thu, 7 Dec 2023 10:45:13 +0800
From: Yu Kuai <yukuai1@...weicloud.com>
To: Matthew Wilcox <willy@...radead.org>, Yu Kuai <yukuai1@...weicloud.com>
Cc: axboe@...nel.dk, roger.pau@...rix.com, colyli@...e.de,
 kent.overstreet@...il.com, joern@...ybastard.org, miquel.raynal@...tlin.com,
 richard@....at, vigneshr@...com, sth@...ux.ibm.com, hoeppner@...ux.ibm.com,
 hca@...ux.ibm.com, gor@...ux.ibm.com, agordeev@...ux.ibm.com,
 jejb@...ux.ibm.com, martin.petersen@...cle.com, clm@...com,
 josef@...icpanda.com, dsterba@...e.com, nico@...xnic.net, xiang@...nel.org,
 chao@...nel.org, tytso@....edu, adilger.kernel@...ger.ca,
 agruenba@...hat.com, jack@...e.com, konishi.ryusuke@...il.com,
 akpm@...ux-foundation.org, hare@...e.de, p.raghav@...sung.com,
 linux-block@...r.kernel.org, linux-kernel@...r.kernel.org,
 xen-devel@...ts.xenproject.org, linux-bcache@...r.kernel.org,
 linux-mtd@...ts.infradead.org, linux-s390@...r.kernel.org,
 linux-scsi@...r.kernel.org, linux-bcachefs@...r.kernel.org,
 linux-btrfs@...r.kernel.org, linux-erofs@...ts.ozlabs.org,
 linux-ext4@...r.kernel.org, gfs2@...ts.linux.dev,
 linux-nilfs@...r.kernel.org, yi.zhang@...wei.com, yangerkun@...wei.com,
 "yukuai (C)" <yukuai3@...wei.com>
Subject: Re: [PATCH -next RFC 01/14] block: add some bdev apis

Hi,

在 2023/12/06 22:58, Matthew Wilcox 写道:
> On Tue, Dec 05, 2023 at 08:37:15PM +0800, Yu Kuai wrote:
>> +struct folio *bdev_read_folio(struct block_device *bdev, pgoff_t index)
>> +{
>> +	return read_mapping_folio(bdev->bd_inode->i_mapping, index, NULL);
>> +}
>> +EXPORT_SYMBOL_GPL(bdev_read_folio);
> 
> I'm coming to the opinion that 'index' is the wrong parameter here.
> Looking through all the callers of bdev_read_folio() in this patchset,
> they all have a position in bytes, and they all convert it to
> index for this call.  The API should probably be:
> 
> struct folio *bdev_read_folio(struct block_device *bdev, loff_t pos)
> {
> 	return read_mapping_folio(bdev->bd_inode->i_mapping,
> 			pos / PAGE_SIZE, NULL);
> }

Thanks for reviewing this patchset! Okay, I'll convert to pass in "pos"
in v2.
> 
> ... and at some point, we'll get round to converting read_mapping_folio()
> to take its argument in loff_t.
> 
> Similiarly for these two APIs:
> 
>> +struct folio *bdev_read_folio_gfp(struct block_device *bdev, pgoff_t index,
>> +				  gfp_t gfp)
>> +struct folio *bdev_get_folio(struct block_device *bdev, pgoff_t index)
> 
>> +struct folio *bdev_find_or_create_folio(struct block_device *bdev,
>> +					pgoff_t index, gfp_t gfp)
>> +{
>> +	return __filemap_get_folio(bdev->bd_inode->i_mapping, index,
>> +				   FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
>> +}
>> +EXPORT_SYMBOL_GPL(bdev_find_or_create_folio);
> 
> This one probably shouldn't exist.  I've been converting callers of
> find_or_create_page() to call __filemap_get_folio; I suspect we
> should expose a __bdev_get_folio and have the callers use the FGP
> arguments directly, but I'm open to other opinions here.

If nobody against this, I will expose single __bdev_get_folio() to use
in v2.
> 
>> +void bdev_sync_readahead(struct block_device *bdev, struct file_ra_state *ra,
>> +			 struct file *file, pgoff_t index,
>> +			 unsigned long req_count)
>> +{
>> +	struct file_ra_state tmp_ra = {};
>> +
>> +	if (!ra) {
>> +		ra = &tmp_ra;
>> +		file_ra_state_init(ra, bdev->bd_inode->i_mapping);
>> +	}
>> +	page_cache_sync_readahead(bdev->bd_inode->i_mapping, ra, file, index,
>> +				  req_count);
>> +}
> 
> I think the caller should always be passing in a valid file_ra_state.
> It's only cramfs that doesn't have one, and it really should!
> Not entirely sure about the arguments here; part of me says "bytes",
> but this is weird enough to maybe take arguments in pages.

In fact, bdev_sync_readahead() is only called for cramfs and ext4.

For ext4 it's used in ext4_readdir() so there is valid file_ra_state.

Hoever, for cramfs it's used in cramfs_read(), and cramfs_read() is used
for:

1) cramfs_read_folio
2) cramfs_readdir
3) cramfs_lookup
4) cramfs_read_super

Looks like it's easy to pass in valid file_ra_state() for 1) and 2),
however, I don't see an easy way to do this for 3) and 4).

Thanks,
Kuai

> 
> .
>