btrfs: don't fallback to buffered read if we don't need to
authorJohannes Thumshirn <johannes.thumshirn@wdc.com>
Thu, 22 Oct 2020 14:05:05 +0000 (23:05 +0900)
committerDavid Sterba <dsterba@suse.com>
Tue, 27 Oct 2020 14:11:37 +0000 (15:11 +0100)
commit0425e7badbdcb6e47109bfdfe494d86217e29009
tree9c5ed26d947ad44825a694c38db746687abe35b4
parent49d11bead7d596e031fbd34051d8765587cd645b
btrfs: don't fallback to buffered read if we don't need to

Since we switched to the iomap infrastructure in b5ff9f1a96e8f ("btrfs:
switch to iomap for direct IO") we're calling generic_file_buffered_read()
directly and not via generic_file_read_iter() anymore.

If the read could read everything there is no need to bother calling
generic_file_buffered_read(), like it is handled in
generic_file_read_iter().

If we call generic_file_buffered_read() in this case we can hit a
situation where we do an invalid readahead and cause this UBSAN splat
in fstest generic/091:

  run fstests generic/091 at 2020-10-21 10:52:32
  ================================================================================
  UBSAN: shift-out-of-bounds in ./include/linux/log2.h:57:13
  shift exponent 64 is too large for 64-bit type 'long unsigned int'
  CPU: 0 PID: 656 Comm: fsx Not tainted 5.9.0-rc7+ #821
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4-rebuilt.opensuse.org 04/01/2014
  Call Trace:
   __dump_stack lib/dump_stack.c:77
   dump_stack+0x57/0x70 lib/dump_stack.c:118
   ubsan_epilogue+0x5/0x40 lib/ubsan.c:148
   __ubsan_handle_shift_out_of_bounds.cold+0x61/0xe9 lib/ubsan.c:395
   __roundup_pow_of_two ./include/linux/log2.h:57
   get_init_ra_size mm/readahead.c:318
   ondemand_readahead.cold+0x16/0x2c mm/readahead.c:530
   generic_file_buffered_read+0x3ac/0x840 mm/filemap.c:2199
   call_read_iter ./include/linux/fs.h:1876
   new_sync_read+0x102/0x180 fs/read_write.c:415
   vfs_read+0x11c/0x1a0 fs/read_write.c:481
   ksys_read+0x4f/0xc0 fs/read_write.c:615
   do_syscall_64+0x33/0x40 arch/x86/entry/common.c:46
   entry_SYSCALL_64_after_hwframe+0x44/0xa9 arch/x86/entry/entry_64.S:118
  RIP: 0033:0x7fe87fee992e
  RSP: 002b:00007ffe01605278 EFLAGS: 00000246 ORIG_RAX: 0000000000000000
  RAX: ffffffffffffffda RBX: 000000000004f000 RCX: 00007fe87fee992e
  RDX: 0000000000004000 RSI: 0000000001677000 RDI: 0000000000000003
  RBP: 000000000004f000 R08: 0000000000004000 R09: 000000000004f000
  R10: 0000000000053000 R11: 0000000000000246 R12: 0000000000004000
  R13: 0000000000000000 R14: 000000000007a120 R15: 0000000000000000
  ================================================================================
  BTRFS info (device nullb0): has skinny extents
  BTRFS info (device nullb0): ZONED mode enabled, zone size 268435456 B
  BTRFS info (device nullb0): enabling ssd optimizations

Fixes: f85781fb505e ("btrfs: switch to iomap for direct IO")
Reviewed-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/file.c