Skip to contentSkip to conversation

btrfs: fix and simplify the inline extent decompression path for subpage

PATCH Superseded Qu Wenruo 4 participants 12 messages Jan 8, 2024 – Jan 10, 2024

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

cover btrfs: fix and simplify the inline extent decompression path for subpage

There is a long existing bug in subpage inline extent reflinking to another location.

The bug is caused by an existing bad code, which is from the beginning of btrfs.
The bad code is never properly explained and got further copied into new compression code.

The bad condition never got properly triggered by different reasons for different platforms:

- On 4K page sized system, the @start_byte is always 0 Thus the existing checks are all dead code, thus never triggered.

- For subpage (4K sectorsize 64K page size) cases, inline extent creation is disable for a different reason
Since no inline extent can be created, there is no way to reflink any inlined extent thus no way to trigger it.

The fixes are mostly going to rework the decompression loop, making sure the input and output buffer are always large enough for inline extent. Thus no need for any loop, but a single decompression call.

But the difficulty lies in how to properly test the bug. For now I'm only doing cross-platform tests, using image created on x86_64, and do the reflink on aarch64.
Not sure if it's possible to upload a binary image for fstests, or I don't have any good way to test the bug.

Qu Wenruo (3):
btrfs: zlib: fix and simplify the inline extent decompression
btrfs: lzo: fix and simplify the inline extent decompression
btrfs: zstd: fix and simplify the inline extent decompression

 fs/btrfs/compression.c | 23 +++++++++----
 fs/btrfs/compression.h |  6 ++--
 fs/btrfs/lzo.c         | 34 +++++--------------
 fs/btrfs/super.h       |  3 ++
 fs/btrfs/zlib.c        | 73 +++++++++++------------------------------
 fs/btrfs/zstd.c        | 74 +++++++++++++-----------------------------
 6 files changed, 72 insertions(+), 141 deletions(-)
Signature · 1 line
2.43.0
author
#1 lore raw

PATCH 1/3 btrfs: zlib: fix and simplify the inline extent decompression

[BUG]

If we have a filesystem with 4k sectorsize, and an inlined compressed extent created like this:

item 4 key (257 INODE_ITEM 0) itemoff 15863 itemsize 160
generation 8 transid 8 size 4096 nbytes 4096 block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0 sequence 1 flags 0x0(none)
item 5 key (257 INODE_REF 256) itemoff 15839 itemsize 24
index 2 namelen 14 name: source_inlined
item 6 key (257 EXTENT_DATA 0) itemoff 15770 itemsize 69
generation 8 type 0 (inline)
inline extent data size 48 ram_bytes 4096 compression 1 (zlib)

Which has an inline compressed extent at file offset 0, and its decompressed size is 4K, allowing us to reflink that 4K range to another location (which will not be compressed).

If we do such reflink on a subpage system, it would fail like this:

 # xfs_io -f -c "reflink $mnt/source_inlined 0 60k 4k" $mnt/dest
 XFS_IOC_CLONE_RANGE: Input/output error

[CAUSE]
In zlib_decompress(), we didn't treat @start_byte as just a page offset, but also use it as an indicator on whether we should switch our output buffer.

In reality, for subpage cases, although @start_byte can be non-zero, we should never switch input/output buffer, since the whole input/output buffer should never exceed one sector.
(The above assumption is only not true if we're going to support multi-page sectorsize)

Thus the current code using @start_byte as a condition to switch input/output buffer or finish the decompression is completely incorrect.

[FIX]
The fix involves several modification:

- Rename @start_byte to @dest_pgoff to properly express its meaning

- Add extra ASSERT() inside btrfs_decompress() to make sure the input/output size never exceed one sector.

- Use Z_FINISH flag to make sure the decompression happens in one go

- Remove all the loop needed to switch input/output buffer

- Use correct destination offset inside the destination page

- Consider early end as an error

After the fix, even on 64K page sized aarch64, above reflink now works as expected:

 # xfs_io -f -c "reflink $mnt/source_inlined 0 60k 4k" $mnt/dest
 linked 4096/4096 bytes at offset 61440

And resulted a correct file layout:

	item 9 key (258 INODE_ITEM 0) itemoff 15542 itemsize 160
		generation 10 transid 10 size 65536 nbytes 4096
		block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0
		sequence 1 flags 0x0(none)
	item 10 key (258 INODE_REF 256) itemoff 15528 itemsize 14
		index 3 namelen 4 name: dest
	item 11 key (258 XATTR_ITEM 3817753667) itemoff 15445 itemsize 83
		location key (0 UNKNOWN.0 0) type XATTR
		transid 10 data_len 37 name_len 16
		name: security.selinux
		data unconfined_u:object_r:unlabeled_t:s0
	item 12 key (258 EXTENT_DATA 61440) itemoff 15392 itemsize 53
		generation 10 type 1 (regular)
		extent data disk byte 13631488 nr 4096
		extent data offset 0 nr 4096 ram 4096
		extent compression 0 (none)
---
 fs/btrfs/compression.c | 23 +++++++++----
 fs/btrfs/compression.h |  2 +-
 fs/btrfs/super.h       |  3 ++
 fs/btrfs/zlib.c        | 73 +++++++++++-------------------------------
 4 files changed, 39 insertions(+), 62 deletions(-)
fs/btrfs/compression.c +16 -7
index 3d8fc2ad0f42..9cf7d38dc66c 100644
@@ -141,16 +141,16 @@ static int compression_decompress_bio(struct list_head *ws,
 }
 
 static int compression_decompress(int type, struct list_head *ws,
-               const u8 *data_in, struct page *dest_page,
-               unsigned long start_byte, size_t srclen, size_t destlen)
+		const u8 *data_in, struct page *dest_page,
+		unsigned long dest_pgoff, size_t srclen, size_t destlen)
 {
 	switch (type) {
 	case BTRFS_COMPRESS_ZLIB: return zlib_decompress(ws, data_in, dest_page,
-						start_byte, srclen, destlen);
+						dest_pgoff, srclen, destlen);
 	case BTRFS_COMPRESS_LZO:  return lzo_decompress(ws, data_in, dest_page,
-						start_byte, srclen, destlen);
+						dest_pgoff, srclen, destlen);
 	case BTRFS_COMPRESS_ZSTD: return zstd_decompress(ws, data_in, dest_page,
-						start_byte, srclen, destlen);
+						dest_pgoff, srclen, destlen);
 	case BTRFS_COMPRESS_NONE:
 	default:
 		/*
@@ -1037,14 +1037,23 @@ static int btrfs_decompress_bio(struct compressed_bio *cb)
  * start_byte tells us the offset into the compressed data we're interested in
  */
 int btrfs_decompress(int type, const u8 *data_in, struct page *dest_page,
-		     unsigned long start_byte, size_t srclen, size_t destlen)
+		     unsigned long dest_pgoff, size_t srclen, size_t destlen)
 {
+	struct btrfs_fs_info *fs_info = btrfs_sb(dest_page->mapping->host->i_sb);
 	struct list_head *workspace;
+	const u32 sectorsize = fs_info->sectorsize;
 	int ret;
 
+	/*
+	 * The full destination page range should not exceed the page size.
+	 * And the @destlen should not exceed sectorsize, as this is only called for
+	 * inline file extents, which should not exceed sectorsize.
+	 */
+	ASSERT(dest_pgoff + destlen <= PAGE_SIZE && destlen <= sectorsize);
+
 	workspace = get_workspace(type, 0);
 	ret = compression_decompress(type, workspace, data_in, dest_page,
-				     start_byte, srclen, destlen);
+				     dest_pgoff, srclen, destlen);
 	put_workspace(type, workspace);
 
 	return ret;
fs/btrfs/compression.h +1 -1
index 93cc92974dee..2b4dfb1b010c 100644
@@ -148,7 +148,7 @@ int zlib_compress_pages(struct list_head *ws, struct address_space *mapping,
 		unsigned long *total_in, unsigned long *total_out);
 int zlib_decompress_bio(struct list_head *ws, struct compressed_bio *cb);
 int zlib_decompress(struct list_head *ws, const u8 *data_in,
-		struct page *dest_page, unsigned long start_byte, size_t srclen,
+		struct page *dest_page, unsigned long dest_pgoff, size_t srclen,
 		size_t destlen);
 struct list_head *zlib_alloc_workspace(unsigned int level);
 void zlib_free_workspace(struct list_head *ws);
fs/btrfs/super.h +3 -0
index f18253ca280d..0dc3dd713cbb 100644
@@ -3,6 +3,9 @@
 #ifndef BTRFS_SUPER_H
 #define BTRFS_SUPER_H
 
+#include <linux/fs.h>
+#include "fs.h"
+
 bool btrfs_check_options(struct btrfs_fs_info *info, unsigned long *mount_opt,
 			 unsigned long flags);
 int btrfs_sync_fs(struct super_block *sb, int wait);
fs/btrfs/zlib.c +19 -54

97 lines not on this page · raw message

Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression

kernel test robot noticed the following build warnings:

Hi Qu,

kernel test robot noticed the following build warnings:

[auto build test WARNING on kdave/for-next]
[also build test WARNING on next-20240108]
[cannot apply to linus/master v6.7]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qu-Wenruo/btrfs-zlib-fix-and-simplify-the-inline-extent-decompression/20240108-171206
base:   https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git for-next
patch link:    https://lore.kernel.org/r/29b7793e53e1cdd559ad212ee69cec211a3b4db2.1704704328.git.wqu%40suse.com
patch subject: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression
config: i386-allmodconfig (https://download.01.org/0day-ci/archive/20240109/[email protected]/config)
compiler: ClangBuiltLinux clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240109/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All warnings (new ones prefixed by >>):

>> fs/btrfs/zlib.c:402:15: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]
     401 |                 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",
         |                                                                                       ~~~
         |                                                                                       %zu
     402 |                                         to_copy, destlen);
         |                                                  ^~~~~~~
   include/linux/printk.h:665:49: note: expanded from macro 'pr_warn_ratelimited'
     665 |         printk_ratelimited(KERN_WARNING pr_fmt(fmt), ##__VA_ARGS__)
         |                                                ~~~     ^~~~~~~~~~~
   include/linux/printk.h:649:17: note: expanded from macro 'printk_ratelimited'
     649 |                 printk(fmt, ##__VA_ARGS__);                             \
         |                        ~~~    ^~~~~~~~~~~
   include/linux/printk.h:455:60: note: expanded from macro 'printk'
     455 | #define printk(fmt, ...) printk_index_wrap(_printk, fmt, ##__VA_ARGS__)
         |                                                     ~~~    ^~~~~~~~~~~
   include/linux/printk.h:427:19: note: expanded from macro 'printk_index_wrap'
     427 |                 _p_func(_fmt, ##__VA_ARGS__);                           \
         |                         ~~~~    ^~~~~~~~~~~

1 warning generated.

vim +402 fs/btrfs/zlib.c

355
356 int zlib_decompress(struct list_head *ws, const u8 *data_in,
357 struct page *dest_page, unsigned long dest_pgoff, size_t srclen,
358 size_t destlen)
359 {
360 struct workspace *workspace = list_entry(ws, struct workspace, list);
361 int ret = 0;
362 int wbits = MAX_WBITS;
363 unsigned long to_copy;
364
365 workspace->strm.next_in = data_in;
366 workspace->strm.avail_in = srclen;
367 workspace->strm.total_in = 0;
368
369 workspace->strm.next_out = workspace->buf;
370 workspace->strm.avail_out = workspace->buf_size;
371 workspace->strm.total_out = 0;
372 /* If it's deflate, and it's got no preset dictionary, then
373 we can tell zlib to skip the adler32 check. */
374 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
375 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
376 !(((data_in[0]<<8) + data_in[1]) % 31)) {
377
378 wbits = -((data_in[0] >> 4) + 8);
379 workspace->strm.next_in += 2;
380 workspace->strm.avail_in -= 2;
381 }
382
383 if (Z_OK != zlib_inflateInit2(&workspace->strm, wbits)) {
384 pr_warn("BTRFS: inflateInit failed\n");
385 return -EIO;
386 }
387
388 /*
389 * Everything (in/out buf) should be at most one sector, there should
390 * be no need to switch any input/output buffer.
391 */
392 ret = zlib_inflate(&workspace->strm, Z_FINISH);
393 to_copy = min(workspace->strm.total_out, destlen);
394 if (ret != Z_STREAM_END)
395 goto out;
396
397 memcpy_to_page(dest_page, dest_pgoff, workspace->buf, to_copy);
398
399 out:
400 if (unlikely(to_copy != destlen)) {
401 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",

 > 402						to_copy, destlen);

403 ret = -EIO;
404 } else {
405 ret = 0;
406 }
407
408 zlib_inflateEnd(&workspace->strm);
409
410 if (unlikely(to_copy < destlen))
411 memzero_page(dest_page, dest_pgoff + to_copy, destlen - to_copy);
412 return ret;
413 }
414

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
#3 lore raw

Re: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression

Quoted 24 lines · kernel test robot

Hi Qu,

kernel test robot noticed the following build warnings:

[auto build test WARNING on kdave/for-next]
[also build test WARNING on next-20240108]
[cannot apply to linus/master v6.7]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qu-Wenruo/btrfs-zlib-fix-and-simplify-the-inline-extent-decompression/20240108-171206
base:   https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git for-next
patch link:    https://lore.kernel.org/r/29b7793e53e1cdd559ad212ee69cec211a3b4db2.1704704328.git.wqu%40suse.com
patch subject: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression
config: i386-allmodconfig (https://download.01.org/0day-ci/archive/20240109/[email protected]/config)
compiler: ClangBuiltLinux clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240109/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

Quoted 8 lines

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

fs/btrfs/zlib.c:402:15: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]

     401 |                 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",
         |                                                                                       ~~~
         |                                                                                       %zu
     402 |                                         to_copy, destlen);
         |                                                  ^~~~~~~

Valid report but I can't reproduce it. Built with clang 17 and explicitly enabled -Wformat. We have additional warnings enabled per directory fs/btrfs/ so we can add -Wformat, I'd like to know what I'm missing, we've had fixups for the size_t printk format so it would make sense to catch it early.

author
#4 lore raw

Re: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression

Quoted 34 lines · David Sterba
Quoted 24 lines · kernel test robot

Hi Qu,

kernel test robot noticed the following build warnings:

[auto build test WARNING on kdave/for-next]
[also build test WARNING on next-20240108]
[cannot apply to linus/master v6.7]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qu-Wenruo/btrfs-zlib-fix-and-simplify-the-inline-extent-decompression/20240108-171206
base:   https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git for-next
patch link:    https://lore.kernel.org/r/29b7793e53e1cdd559ad212ee69cec211a3b4db2.1704704328.git.wqu%40suse.com
patch subject: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression
config: i386-allmodconfig (https://download.01.org/0day-ci/archive/20240109/[email protected]/config)
compiler: ClangBuiltLinux clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240109/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

Quoted 8 lines

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

fs/btrfs/zlib.c:402:15: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]

      401 |                 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",
          |                                                                                       ~~~
          |                                                                                       %zu
      402 |                                         to_copy, destlen);
          |                                                  ^~~~~~~
Quoted 5 lines

Valid report but I can't reproduce it. Built with clang 17 and explicitly enabled -Wformat. We have additional warnings enabled per directory fs/btrfs/ so we can add -Wformat, I'd like to know what I'm missing, we've had fixups for the size_t printk format so it would make sense to catch it early.

I guess it's due to the platform? (The report is from 32bit system).

Otherwise it's indeed my bad, for now I don't even have a 32bit VM to verify, thus my LSP doesn't warn me about the format.

Thanks,
Qu

#5 lore raw

Re: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression

Quoted 37 lines · Qu Wenruo
Quoted 34 lines · David Sterba
Quoted 32 lines · kernel test robot

Hi Qu,

kernel test robot noticed the following build warnings:

[auto build test WARNING on kdave/for-next]
[also build test WARNING on next-20240108]
[cannot apply to linus/master v6.7]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qu-Wenruo/btrfs-zlib-fix-and-simplify-the-inline-extent-decompression/20240108-171206
base:   https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git for-next
patch link:    https://lore.kernel.org/r/29b7793e53e1cdd559ad212ee69cec211a3b4db2.1704704328.git.wqu%40suse.com
patch subject: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression
config: i386-allmodconfig (https://download.01.org/0day-ci/archive/20240109/[email protected]/config)
compiler: ClangBuiltLinux clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240109/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

fs/btrfs/zlib.c:402:15: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]

      401 |                 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",
          |                                                                                       ~~~
          |                                                                                       %zu
      402 |                                         to_copy, destlen);
          |                                                  ^~~~~~~
Quoted 7 lines
Quoted 5 lines

Valid report but I can't reproduce it. Built with clang 17 and explicitly enabled -Wformat. We have additional warnings enabled per directory fs/btrfs/ so we can add -Wformat, I'd like to know what I'm missing, we've had fixups for the size_t printk format so it would make sense to catch it early.

I guess it's due to the platform? (The report is from 32bit system).

Ah I see, I build on 64bit platform but should the Wformat warning also point out mismatch there? The size_t type is an alias of unsigned long so it is not an error but when size_t and %zu don't match could be a platform-independent error, no? This would save us reports and followup fixups roundtrips.

Quoted 2 lines

Otherwise it's indeed my bad, for now I don't even have a 32bit VM to verify, thus my LSP doesn't warn me about the format.

Yeah, it could/should though.

author
#6 lore raw

Re: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression

Quoted 44 lines · David Sterba
Quoted 43 lines · Qu Wenruo
Quoted 39 lines · David Sterba
Quoted 24 lines · kernel test robot

Hi Qu,

kernel test robot noticed the following build warnings:

[auto build test WARNING on kdave/for-next]
[also build test WARNING on next-20240108]
[cannot apply to linus/master v6.7]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Qu-Wenruo/btrfs-zlib-fix-and-simplify-the-inline-extent-decompression/20240108-171206
base:   https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux.git for-next
patch link:    https://lore.kernel.org/r/29b7793e53e1cdd559ad212ee69cec211a3b4db2.1704704328.git.wqu%40suse.com
patch subject: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression
config: i386-allmodconfig (https://download.01.org/0day-ci/archive/20240109/[email protected]/config)
compiler: ClangBuiltLinux clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240109/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

Quoted 8 lines

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

fs/btrfs/zlib.c:402:15: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]

       401 |                 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",
           |                                                                                       ~~~
           |                                                                                       %zu
       402 |                                         to_copy, destlen);
           |                                                  ^~~~~~~

Valid report but I can't reproduce it. Built with clang 17 and explicitly enabled -Wformat. We have additional warnings enabled per directory fs/btrfs/ so we can add -Wformat, I'd like to know what I'm missing, we've had fixups for the size_t printk format so it would make sense to catch it early.

Quoted 7 lines
Quoted 1 line

I guess it's due to the platform? (The report is from 32bit system).

Ah I see, I build on 64bit platform but should the Wformat warning also point out mismatch there? The size_t type is an alias of unsigned long so it is not an error but when size_t and %zu don't match could be a platform-independent error, no? This would save us reports and followup fixups roundtrips.

size_t is defined differently, in include/uapi/asm-generic/posix_types.h:

/*
* Most 32 bit architectures use "unsigned int" size_t,
* and all 64 bit architectures use "unsigned long" size_t. */
#ifndef __kernel_size_t
#if __BITS_PER_LONG != 64

typedef unsigned int    __kernel_size_t;
typedef int             __kernel_ssize_t;
typedef int             __kernel_ptrdiff_t;

#else
typedef __kernel_ulong_t __kernel_size_t;
typedef __kernel_long_t __kernel_ssize_t;
typedef __kernel_long_t __kernel_ptrdiff_t;
#endif
#endif

Thus this is the reason why we need @zu for size_t to handle the difference, and since for 64bit it's just unsigned long, thus compiler won't give any warning.

(That's also why I tend to not use size_t at all, and why I like rust's explicit sized type, and we may want to go that path to prefer u8/u16/u32/u64 when possible)

Thanks,
Qu

Quoted 5 lines
Quoted 3 lines

Otherwise it's indeed my bad, for now I don't even have a 32bit VM to verify, thus my LSP doesn't warn me about the format.

Yeah, it could/should though.

#7 lore raw

Re: [PATCH 1/3] btrfs: zlib: fix and simplify the inline extent decompression

Quoted 47 lines · Qu Wenruo
Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 6 lines

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

Quoted 8 lines

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

fs/btrfs/zlib.c:402:15: warning: format specifies type 'unsigned long' but the argument has type 'size_t' (aka 'unsigned int') [-Wformat]

       401 |                 pr_warn_ratelimited("BTRFS: infalte failed, decompressed=%lu expected=%lu\n",
           |                                                                                       ~~~
           |                                                                                       %zu
       402 |                                         to_copy, destlen);
           |                                                  ^~~~~~~

Valid report but I can't reproduce it. Built with clang 17 and explicitly enabled -Wformat. We have additional warnings enabled per directory fs/btrfs/ so we can add -Wformat, I'd like to know what I'm missing, we've had fixups for the size_t printk format so it would make sense to catch it early.

Quoted 7 lines
Quoted 1 line

I guess it's due to the platform? (The report is from 32bit system).

Ah I see, I build on 64bit platform but should the Wformat warning also point out mismatch there? The size_t type is an alias of unsigned long so it is not an error but when size_t and %zu don't match could be a platform-independent error, no? This would save us reports and followup fixups roundtrips.

size_t is defined differently, in include/uapi/asm-generic/posix_types.h:

/*
* Most 32 bit architectures use "unsigned int" size_t,
* and all 64 bit architectures use "unsigned long" size_t. */
#ifndef __kernel_size_t
#if __BITS_PER_LONG != 64

typedef unsigned int    __kernel_size_t;
typedef int             __kernel_ssize_t;
typedef int             __kernel_ptrdiff_t;

#else
typedef __kernel_ulong_t __kernel_size_t;
typedef __kernel_long_t __kernel_ssize_t;
typedef __kernel_long_t __kernel_ptrdiff_t;
#endif
#endif

Quoted 3 lines

Thus this is the reason why we need @zu for size_t to handle the difference, and since for 64bit it's just unsigned long, thus compiler won't give any warning.

So it's the int/long difference and kernel does it in a special way due to absence of the standard libc headers.

Quoted 3 lines

(That's also why I tend to not use size_t at all, and why I like rust's explicit sized type, and we may want to go that path to prefer u8/u16/u32/u64 when possible)

Yeah, from what I've seen so far is that size_t brings us only problems, I would not mind conversions to explicit types like u32 or u64. We do that in many places, I think it's namely on the interface boundaries where we provide a callback with a size_t type, but from that down it could be u64.

Quick grep for size_t returns 300+ lines and 100+ of that is ssize_t, that's still a lot so the conversions should be targeted.

author
#8 lore raw

PATCH 2/3 btrfs: lzo: fix and simplify the inline extent decompression

In reply to Qu Wenruo

[BUG]
If we have a filesystem with 4k sectorsize, and an inlined compressed extent created like this:

item 4 key (257 INODE_ITEM 0) itemoff 15863 itemsize 160
generation 8 transid 8 size 4096 nbytes 4096 block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0 sequence 1 flags 0x0(none)
item 5 key (257 INODE_REF 256) itemoff 15839 itemsize 24
index 2 namelen 14 name: source_inlined
item 6 key (257 EXTENT_DATA 0) itemoff 15770 itemsize 69
generation 8 type 0 (inline)
inline extent data size 48 ram_bytes 4096 compression 2 (lzo)

Then trying to reflink that extent in an aarch64 system with 64K page size, the reflink would just fail:

  # xfs_io -f -c "reflink $mnt/source_inlined 0 60k 4k" $mnt/dest
  XFS_IOC_CLONE_RANGE: Input/output error

[CAUSE]
In zlib_decompress(), we didn't treat @start_byte as just a page offset, but also use it as an indicator on whether we should error out, without any proper explanation (this is from the very beginning of btrfs).

In reality, for subpage cases, although @start_byte can be non-zero, we should never switch input/output buffer nor error out, since the whole input/output buffer should never exceed one sector.
(The above assumption is only not true if we're going to support multi-page sectorsize)

Thus the current code using @start_byte as a condition to switch input/output buffer or finish the decompression is completely incorrect.

[FIX]
The fix involves several modification:

- Rename @start_byte to @dest_pgoff to properly express its meaning

- Use @sectorsize other than PAGE_SIZE to properly initialize the output buffer size

- Use correct destination offset inside the destination page

- Use memcpy_to_page() to copy the contents to the destination page

- Use memzero_page() to zero out the tailing part

- Consider early end as an error

After the fix, even on 64K page sized aarch64, above reflink now works as expected:

  # xfs_io -f -c "reflink $mnt/source_inlined 0 60k 4k" $mnt/dest
  linked 4096/4096 bytes at offset 61440

And results the correct file layout:

	item 9 key (258 INODE_ITEM 0) itemoff 15542 itemsize 160
		generation 10 transid 10 size 65536 nbytes 4096
		block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0
		sequence 1 flags 0x0(none)
	item 10 key (258 INODE_REF 256) itemoff 15528 itemsize 14
		index 3 namelen 4 name: dest
	item 11 key (258 XATTR_ITEM 3817753667) itemoff 15445 itemsize 83
		location key (0 UNKNOWN.0 0) type XATTR
		transid 10 data_len 37 name_len 16
		name: security.selinux
		data unconfined_u:object_r:unlabeled_t:s0
	item 12 key (258 EXTENT_DATA 61440) itemoff 15392 itemsize 53
		generation 10 type 1 (regular)
		extent data disk byte 13631488 nr 4096
		extent data offset 0 nr 4096 ram 4096
		extent compression 0 (none)
---
 fs/btrfs/compression.h |  2 +-
 fs/btrfs/lzo.c         | 34 +++++++++-------------------------
 2 files changed, 10 insertions(+), 26 deletions(-)
fs/btrfs/compression.h +1 -1
index 2b4dfb1b010c..afd7e50d073d 100644
@@ -159,7 +159,7 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
 		unsigned long *total_in, unsigned long *total_out);
 int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb);
 int lzo_decompress(struct list_head *ws, const u8 *data_in,
-		struct page *dest_page, unsigned long start_byte, size_t srclen,
+		struct page *dest_page, unsigned long dest_pgoff, size_t srclen,
 		size_t destlen);
 struct list_head *lzo_alloc_workspace(unsigned int level);
 void lzo_free_workspace(struct list_head *ws);
fs/btrfs/lzo.c +9 -25
index 1131d5a29d61..e43bc0fdc74e 100644
@@ -425,16 +425,16 @@ int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
 }
 
 int lzo_decompress(struct list_head *ws, const u8 *data_in,
-		struct page *dest_page, unsigned long start_byte, size_t srclen,
+		struct page *dest_page, unsigned long dest_pgoff, size_t srclen,
 		size_t destlen)
 {
 	struct workspace *workspace = list_entry(ws, struct workspace, list);
+	struct btrfs_fs_info *fs_info = btrfs_sb(dest_page->mapping->host->i_sb);
+	const u32 sectorsize = fs_info->sectorsize;
 	size_t in_len;
 	size_t out_len;
 	size_t max_segment_len = WORKSPACE_BUF_LENGTH;
 	int ret = 0;
-	char *kaddr;
-	unsigned long bytes;
 
 	if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2)
 		return -EUCLEAN;
@@ -451,7 +451,7 @@ int lzo_decompress(struct list_head *ws, const u8 *data_in,
 	}
 	data_in += LZO_LEN;
 
-	out_len = PAGE_SIZE;
+	out_len = sectorsize;
 	ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len);
 	if (ret != LZO_E_OK) {
 		pr_warn("BTRFS: decompress failed!\n");
@@ -459,29 +459,13 @@ int lzo_decompress(struct list_head *ws, const u8 *data_in,
 		goto out;
 	}
 
-	if (out_len < start_byte) {
+	ASSERT(out_len <= sectorsize);
+	memcpy_to_page(dest_page, dest_pgoff, workspace->buf, out_len);
+	/* Early end, considered as an error. */
+	if (unlikely(out_len < destlen)) {
 		ret = -EIO;
-		goto out;
+		memzero_page(dest_page, dest_pgoff + out_len, destlen - out_len);
 	}
-
-	/*
-	 * the caller is already checking against PAGE_SIZE, but lets
-	 * move this check closer to the memcpy/memset
-	 */
-	destlen = min_t(unsigned long, destlen, PAGE_SIZE);
-	bytes = min_t(unsigned long, destlen, out_len - start_byte);
-
-	kaddr = kmap_local_page(dest_page);
-	memcpy(kaddr, workspace->buf + start_byte, bytes);
-
-	/*
-	 * btrfs_getblock is doing a zero on the tail of the page too,
-	 * but this will cover anything missing from the decompressed
-	 * data.
-	 */
-	if (bytes < destlen)
-		memset(kaddr+bytes, 0, destlen-bytes);
-	kunmap_local(kaddr);
 out:
 	return ret;
 }
Signature · 1 line
2.43.0
author
#9 lore raw

PATCH 3/3 btrfs: zstd: fix and simplify the inline extent decompression

In reply to Qu Wenruo

[BUG]
If we have a filesystem with 4k sectorsize, and an inlined compressed extent created like this:

item 4 key (257 INODE_ITEM 0) itemoff 15863 itemsize 160
generation 8 transid 8 size 4096 nbytes 4096 block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0 sequence 1 flags 0x0(none)
item 5 key (257 INODE_REF 256) itemoff 15839 itemsize 24
index 2 namelen 14 name: source_inlined
item 6 key (257 EXTENT_DATA 0) itemoff 15770 itemsize 69
generation 8 type 0 (inline)
inline extent data size 48 ram_bytes 4096 compression 3 (zstd)

Then trying to reflink that extent in an aarch64 system with 64K page size, the reflink would just fail:

  # xfs_io -f -c "reflink $mnt/source_inlined 0 60k 4k" $mnt/dest
  XFS_IOC_CLONE_RANGE: Input/output error

[CAUSE]
In zstd_decompress(), we didn't treat @start_byte as just a page offset, but also use it as an indicator on whether we should error out, without any proper explanation (this is copied from other decompression code).

In reality, for subpage cases, although @start_byte can be non-zero, we should never switch input/output buffer nor error out, since the whole input/output buffer should never exceed one sector, thus we should not need to do any buffer switch.

Thus the current code using @start_byte as a condition to switch input/output buffer or finish the decompression is completely incorrect.

[FIX]
The fix involves several modification:

- Rename @start_byte to @dest_pgoff to properly express its meaning

- Use @sectorsize other than PAGE_SIZE to properly initialize the output buffer size

- Use correct destination offset inside the destination page

- Simplify the main loop
Since the input/output buffer should never switch, we only need one zstd_decompress_stream() call.

- Consider early end as an error

After the fix, even on 64K page sized aarch64, above reflink now works as expected:

  # xfs_io -f -c "reflink $mnt/source_inlined 0 60k 4k" $mnt/dest
  linked 4096/4096 bytes at offset 61440

And results the correct file layout:

	item 9 key (258 INODE_ITEM 0) itemoff 15542 itemsize 160
		generation 10 transid 10 size 65536 nbytes 4096
		block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0
		sequence 1 flags 0x0(none)
	item 10 key (258 INODE_REF 256) itemoff 15528 itemsize 14
		index 3 namelen 4 name: dest
	item 11 key (258 XATTR_ITEM 3817753667) itemoff 15445 itemsize 83
		location key (0 UNKNOWN.0 0) type XATTR
		transid 10 data_len 37 name_len 16
		name: security.selinux
		data unconfined_u:object_r:unlabeled_t:s0
	item 12 key (258 EXTENT_DATA 61440) itemoff 15392 itemsize 53
		generation 10 type 1 (regular)
		extent data disk byte 13631488 nr 4096
		extent data offset 0 nr 4096 ram 4096
		extent compression 0 (none)
---
 fs/btrfs/compression.h |  2 +-
 fs/btrfs/zstd.c        | 74 +++++++++++++-----------------------------
 2 files changed, 23 insertions(+), 53 deletions(-)
fs/btrfs/compression.h +1 -1
index afd7e50d073d..97fe3ebf11a2 100644
@@ -169,7 +169,7 @@ int zstd_compress_pages(struct list_head *ws, struct address_space *mapping,
 		unsigned long *total_in, unsigned long *total_out);
 int zstd_decompress_bio(struct list_head *ws, struct compressed_bio *cb);
 int zstd_decompress(struct list_head *ws, const u8 *data_in,
-		struct page *dest_page, unsigned long start_byte, size_t srclen,
+		struct page *dest_page, unsigned long dest_pgoff, size_t srclen,
 		size_t destlen);
 void zstd_init_workspace_manager(void);
 void zstd_cleanup_workspace_manager(void);
fs/btrfs/zstd.c +22 -52

108 lines not on this page · raw message

Signature · 1 line
2.43.0
#10 lore raw
In reply to Qu Wenruo
Quoted 7 lines · Qu Wenruo

There is a long existing bug in subpage inline extent reflinking to another location.

The bug is caused by an existing bad code, which is from the beginning of btrfs.
The bad code is never properly explained and got further copied into new compression code.

I think there was an idea to let an inline compressed extent to span more than one page, but it never materialized. There could be code handling it but due to the invariants (like that inline extent is always smaller than a sector) it's never executed.

Quoted 6 lines

The bad condition never got properly triggered by different reasons for different platforms:

- On 4K page sized system, the @start_byte is always 0 Thus the existing checks are all dead code, thus never triggered.

Quoted 8 lines

- For subpage (4K sectorsize 64K page size) cases, inline extent creation is disable for a different reason
Since no inline extent can be created, there is no way to reflink any inlined extent thus no way to trigger it.

The fixes are mostly going to rework the decompression loop, making sure the input and output buffer are always large enough for inline extent. Thus no need for any loop, but a single decompression call.

Yeah, as long as we have the page == sectorsize it's not necessary.

Quoted 5 lines

But the difficulty lies in how to properly test the bug. For now I'm only doing cross-platform tests, using image created on x86_64, and do the reflink on aarch64.
Not sure if it's possible to upload a binary image for fstests, or I don't have any good way to test the bug.

We had this discussion year ago (no crafted images) but the maintainer of fstests changed meanwhile so you can try again. You can always add that to progs but it won't get the same level of testing due to its primary purpose to test the userspace code.

Quoted 4 lines

Qu Wenruo (3):
btrfs: zlib: fix and simplify the inline extent decompression
btrfs: lzo: fix and simplify the inline extent decompression
btrfs: zstd: fix and simplify the inline extent decompression

Added to misc-next, thanks.

There are new error messages when the decompressed length is not expected, this could be still improved (and made more consistent). The name of the file or inode number and filesystem identification are missing, but that's how the current code works.

#11 lore raw
Quoted 9 lines · David Sterba
Quoted 7 lines · Qu Wenruo

There is a long existing bug in subpage inline extent reflinking to another location.

The bug is caused by an existing bad code, which is from the beginning of btrfs.
The bad code is never properly explained and got further copied into new compression code.

Quoted 4 lines

I think there was an idea to let an inline compressed extent to span more than one page, but it never materialized. There could be code handling it but due to the invariants (like that inline extent is always smaller than a sector) it's never executed.

Unfortunately it doesn't look like this.

For multiple page cases, we should not call offset_in_page() for that @start_byte, and even in that case, we still don't need that @start_byte parameter.

To me, that parameter is just for subpage, but not properly documented and since it's mostly dead code, we copy them around without properly knowing the reason.

Quoted 18 lines
Quoted 7 lines

The bad condition never got properly triggered by different reasons for different platforms:

- On 4K page sized system, the @start_byte is always 0 Thus the existing checks are all dead code, thus never triggered.

Quoted 8 lines

- For subpage (4K sectorsize 64K page size) cases, inline extent creation is disable for a different reason
Since no inline extent can be created, there is no way to reflink any inlined extent thus no way to trigger it.

The fixes are mostly going to rework the decompression loop, making sure the input and output buffer are always large enough for inline extent. Thus no need for any loop, but a single decompression call.

Yeah, as long as we have the page == sectorsize it's not necessary.

Quoted 8 lines
Quoted 5 lines

But the difficulty lies in how to properly test the bug. For now I'm only doing cross-platform tests, using image created on x86_64, and do the reflink on aarch64.
Not sure if it's possible to upload a binary image for fstests, or I don't have any good way to test the bug.

We had this discussion year ago (no crafted images) but the maintainer of fstests changed meanwhile so you can try again.

It looks like that's the best way to go for now.

Thanks,
Qu

Quoted 16 lines

You can always add
that to progs but it won't get the same level of testing due to its primary purpose to test the userspace code.

Quoted 4 lines

Qu Wenruo (3):
btrfs: zlib: fix and simplify the inline extent decompression
btrfs: lzo: fix and simplify the inline extent decompression
btrfs: zstd: fix and simplify the inline extent decompression

Added to misc-next, thanks.

There are new error messages when the decompressed length is not expected, this could be still improved (and made more consistent). The name of the file or inode number and filesystem identification are missing, but that's how the current code works.