squashfs: avoid bio_alloc() failure with 1Mbyte blocks
authorPhillip Lougher <phillip@squashfs.org.uk>
Fri, 21 Aug 2020 00:42:21 +0000 (17:42 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 21 Aug 2020 16:52:53 +0000 (09:52 -0700)
This is a regression introduced by the patch "migrate from ll_rw_block
usage to BIO".

Bio_alloc() is limited to 256 pages (1 Mbyte).  This can cause a failure
when reading 1 Mbyte block filesystems.  The problem is a datablock can be
fully (or almost uncompressed), requiring 256 pages, but, because blocks
are not aligned to page boundaries, it may require 257 pages to read.

Bio_kmalloc() can handle 1024 pages, and so use this for the edge
condition.

Fixes: 93e72b3c612a ("squashfs: migrate from ll_rw_block usage to BIO")
Reported-by: Nicolas Prochazka <nicolas.prochazka@gmail.com>
Reported-by: Tomoatsu Shimada <shimada@walbrix.com>
Signed-off-by: Phillip Lougher <phillip@squashfs.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Cc: Philippe Liard <pliard@google.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Adrien Schildknecht <adrien+dev@schischi.me>
Cc: Daniel Rosenberg <drosen@google.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200815035637.15319-1-phillip@squashfs.org.uk
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/squashfs/block.c

index 76bb1c846845e33737d428dcc26e4840fd4aeebe..8a19773b5a0b7f14b86f16e97643e7b4c76f941b 100644 (file)
@@ -87,7 +87,11 @@ static int squashfs_bio_read(struct super_block *sb, u64 index, int length,
        int error, i;
        struct bio *bio;
 
-       bio = bio_alloc(GFP_NOIO, page_count);
+       if (page_count <= BIO_MAX_PAGES)
+               bio = bio_alloc(GFP_NOIO, page_count);
+       else
+               bio = bio_kmalloc(GFP_NOIO, page_count);
+
        if (!bio)
                return -ENOMEM;