* Re: [PATCH] swap: fix swapfile read/write offset
2021-03-03 16:24 ` Jens Axboe
@ 2021-03-04 3:19 ` Anthony Iliopoulos
0 siblings, 0 replies; 3+ messages in thread
From: Anthony Iliopoulos @ 2021-03-04 3:19 UTC (permalink / raw)
To: Jens Axboe
Cc: Linux Memory Management List, linux-block, Christoph Hellwig, akpm
On Wed, Mar 03, 2021 at 09:24:13AM -0700, Jens Axboe wrote:
> On 3/3/21 8:46 AM, Anthony Iliopoulos wrote:
> > On Tue, Mar 02, 2021 at 03:36:19PM -0700, Jens Axboe wrote:
> >> We're not factoring in the start of the file for where to write and
> >> read the swapfile, which leads to very unfortunate side effects of
> >> writing where we should not be...
> >>
> >> Fixes: 48d15436fde6 ("mm: remove get_swap_bio")
> >
> > Presumably the usage of swap_page_sector was already affecting swap on
> > blockdevs that implement rw_page (currently brd, zram, btt, pmem), so it
> > may worth adding:
> >
> > Fixes: dd6bd0d9c7db ("swap: use bdev_read_page() / bdev_write_page()")
> > Cc: <stable@vger.kernel.org> # v3.16+
> >
> > for backporting, since it also affects stable.
>
> yes indeed, in fact that is the source of the original issue (copy/paste
> from that broken path).
>
> Fix is already upstream, but would be nice if someone would turn it into
> something that could be applied to stable.
Sure, I have the following for v5.10+ and can post this later today to
stable for review along with backports for the rest of the longterm
kernels:
From f0d75f9a18e5184670ea11b87bf513b0225b6826 Mon Sep 17 00:00:00 2001
From: Anthony Iliopoulos <ailiop@suse.com>
Date: Wed, 3 Mar 2021 20:07:05 +0100
Subject: [PATCH] swap: fix swapfile page offset mapping
Fix block device sector offset calculation for swap page io on top of
blockdevs that provide a rw_page operation and do page-sized io directly
(without the block layer).
Currently swap_page_sector() maps a swap page into a blockdev sector by
obtaining the swap page offset (swap map slot), but ignores the swapfile
starting offset into the blockdev.
In setups where swapfiles are sitting on top of a filesystem, this
results into swapping out activity potentially overwriting filesystem
blocks that fall outside the swapfile region.
[This issue only affects swapfiles on filesystems on top of blockdevs
that implement rw_page ops (brd, zram, btt, pmem), and not on top of any
other regular block devices.]
Fixes: dd6bd0d9c7db ("swap: use bdev_read_page() / bdev_write_page()")
Cc: <stable@vger.kernel.org> # v5.10+
Signed-off-by: Anthony Iliopoulos <ailiop@suse.com>
---
mm/page_io.c | 12 ++++--------
mm/swapfile.c | 2 +-
2 files changed, 5 insertions(+), 9 deletions(-)
diff --git a/mm/page_io.c b/mm/page_io.c
index 9bca17ecc4df..d2d4d1b3db10 100644
--- a/mm/page_io.c
+++ b/mm/page_io.c
@@ -37,7 +37,6 @@ static struct bio *get_swap_bio(gfp_t gfp_flags,
bio->bi_iter.bi_sector = map_swap_page(page, &bdev);
bio_set_dev(bio, bdev);
- bio->bi_iter.bi_sector <<= PAGE_SHIFT - 9;
bio->bi_end_io = end_io;
bio_add_page(bio, page, thp_size(page), 0);
@@ -273,11 +272,6 @@ int swap_writepage(struct page *page, struct writeback_control *wbc)
return ret;
}
-static sector_t swap_page_sector(struct page *page)
-{
- return (sector_t)__page_file_index(page) << (PAGE_SHIFT - 9);
-}
-
static inline void count_swpout_vm_event(struct page *page)
{
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
@@ -355,7 +349,8 @@ int __swap_writepage(struct page *page, struct writeback_control *wbc,
return ret;
}
- ret = bdev_write_page(sis->bdev, swap_page_sector(page), page, wbc);
+ ret = bdev_write_page(sis->bdev, map_swap_page(page, &sis->bdev),
+ page, wbc);
if (!ret) {
count_swpout_vm_event(page);
return 0;
@@ -414,7 +409,8 @@ int swap_readpage(struct page *page, bool synchronous)
}
if (sis->flags & SWP_SYNCHRONOUS_IO) {
- ret = bdev_read_page(sis->bdev, swap_page_sector(page), page);
+ ret = bdev_read_page(sis->bdev, map_swap_page(page, &sis->bdev),
+ page);
if (!ret) {
if (trylock_page(page)) {
swap_slot_free_notify(page);
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 9fffc5af29d1..47524a4d5e90 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -2308,7 +2308,7 @@ sector_t map_swap_page(struct page *page, struct block_device **bdev)
{
swp_entry_t entry;
entry.val = page_private(page);
- return map_swap_entry(entry, bdev);
+ return map_swap_entry(entry, bdev) << (PAGE_SHIFT - 9);
}
/*
--
2.30.1
^ permalink raw reply related [flat|nested] 3+ messages in thread