linux-kernel - Re: [PATCH v7 4/6] dax: add DAX_RECOVERY flag and .recovery_write dev_pgmap

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

Date:   Tue, 5 Apr 2022 22:19:58 -0700
From:   Christoph Hellwig <hch@...radead.org>
To:     Jane Chu <jane.chu@...cle.com>
Cc:     david@...morbit.com, djwong@...nel.org, dan.j.williams@...el.com,
        hch@...radead.org, vishal.l.verma@...el.com, dave.jiang@...el.com,
        agk@...hat.com, snitzer@...hat.com, dm-devel@...hat.com,
        ira.weiny@...el.com, willy@...radead.org, vgoyal@...hat.com,
        linux-fsdevel@...r.kernel.org, nvdimm@...ts.linux.dev,
        linux-kernel@...r.kernel.org, linux-xfs@...r.kernel.org,
        x86@...nel.org
Subject: Re: [PATCH v7 4/6] dax: add DAX_RECOVERY flag and .recovery_write
 dev_pgmap_ops

On Tue, Apr 05, 2022 at 01:47:45PM -0600, Jane Chu wrote:
> Introduce DAX_RECOVERY flag to dax_direct_access(). The flag is
> not set by default in dax_direct_access() such that the helper
> does not translate a pmem range to kernel virtual address if the
> range contains uncorrectable errors.  When the flag is set,
> the helper ignores the UEs and return kernel virtual adderss so
> that the caller may get on with data recovery via write.
> 
> Also introduce a new dev_pagemap_ops .recovery_write function.
> The function is applicable to FSDAX device only. The device
> page backend driver provides .recovery_write function if the
> device has underlying mechanism to clear the uncorrectable
> errors on the fly.

I know Dan suggested it, but I still think dev_pagemap_ops is the very
wrong choice here.  It is about VM callbacks to ZONE_DEVICE owners
independent of what pagemap type they are.  .recovery_write on the
other hand is completely specific to the DAX write path and has no
MM interactions at all.

>  /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
>  __weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
> -		long nr_pages, void **kaddr, pfn_t *pfn)
> +		long nr_pages, int flags, void **kaddr, pfn_t *pfn)
>  {
>  	resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
> +	sector_t sector = PFN_PHYS(pgoff) >> SECTOR_SHIFT;
> +	unsigned int num = PFN_PHYS(nr_pages) >> SECTOR_SHIFT;
> +	struct badblocks *bb = &pmem->bb;
> +	sector_t first_bad;
> +	int num_bad;
> +	bool bad_in_range;
> +	long actual_nr;
> +
> +	if (!bb->count)
> +		bad_in_range = false;
> +	else
> +		bad_in_range = !!badblocks_check(bb, sector, num, &first_bad, &num_bad);
>  
> -	if (unlikely(is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) / 512,
> -					PFN_PHYS(nr_pages))))
> +	if (bad_in_range && !(flags & DAX_RECOVERY))
>  		return -EIO;

The use of bad_in_range here seems a litle convoluted.  See the attached
patch on how I would structure the function to avoid the variable and
have the reocvery code in a self-contained chunk.

> -		map_len = dax_direct_access(dax_dev, pgoff, PHYS_PFN(size),
> -				&kaddr, NULL);
> +		nrpg = PHYS_PFN(size);
> +		map_len = dax_direct_access(dax_dev, pgoff, nrpg, 0, &kaddr, NULL);

Overly long line here.

View attachment "diff" of type "text/plain" (2352 bytes)