Skip to contentSkip to conversation

libext2fs: fix block-mapped file punch

PATCH Applied Andreas Dilger 2 participants 4 messages Nov 19, 2015 – Nov 30, 2015

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 libext2fs: fix block-mapped file punch

If ext2fs_punch() was called with "end = ~0ULL" to indicate truncate to the end of file it tried to compute "count" for ext2fs_punch_ind() based on "start" and "end", but incorrectly passed "count = ~0U" even when "start" was non-zero, causing an overflow in some cases.

The calling convention for ext2fs_punch_ind() was also gratuitously different from ext2fs_punch() and ext2fs_punch_extent(), passing "count" instead of "end" as the last parameter. Fix this by passing it "end" like the other functions, and handle "count" internally.

Add checks to ext2fs_punch_ind() if "end" is at or beyond the 2^32 indirect block limit so the 32-bit internal variables don't overflow.

---
 lib/ext2fs/punch.c |   48 +++++++++++++++++++++++++++---------------------
 1 files changed, 27 insertions(+), 21 deletions(-)
lib/ext2fs/punch.c +21 -12
index 2a2cf10..62ddd50 100644
@@ -47,7 +47,7 @@ static int check_zero_block(char *buf, int blocksize)
  */
 static errcode_t ind_punch(ext2_filsys fs, struct ext2_inode *inode,
 			   char *block_buf, blk_t *p, int level,
-			   blk_t start, blk_t count, int max)
+			   blk64_t start, blk64_t count, int max)
 {
 	errcode_t	retval;
 	blk_t		b;
@@ -56,11 +56,11 @@ static errcode_t ind_punch(ext2_filsys fs, struct ext2_inode *inode,
 	int		freed = 0;
 
 #ifdef PUNCH_DEBUG
-	printf("Entering ind_punch, level %d, start %u, count %u, "
+	printf("Entering ind_punch, level %d, start %llu, count %llu, "
 	       "max %d\n", level, start, count, max);
 #endif
-	incr = 1ULL << ((EXT2_BLOCK_SIZE_BITS(fs->super)-2)*level);
-	for (i=0, offset=0; i < max; i++, p++, offset += incr) {
+	incr = 1ULL << ((EXT2_BLOCK_SIZE_BITS(fs->super) - 2) * level);
+	for (i = 0, offset = 0; i < max; i++, p++, offset += incr) {
 		if (offset >= start + count)
 			break;
 		if (*p == 0 || (offset+incr) <= start)
@@ -100,8 +100,9 @@ static errcode_t ind_punch(ext2_filsys fs, struct ext2_inode *inode,
 	return ext2fs_iblk_sub_blocks(fs, inode, freed);
 }
 
+#define BLK_T_MAX ((blk_t)~0ULL)
 static errcode_t ext2fs_punch_ind(ext2_filsys fs, struct ext2_inode *inode,
-				  char *block_buf, blk_t start, blk_t count)
+				  char *block_buf, blk64_t start, blk64_t end)
 {
 	errcode_t		retval;
 	char			*buf = 0;
@@ -110,6 +111,15 @@ static errcode_t ext2fs_punch_ind(ext2_filsys fs, struct ext2_inode *inode,
 	blk_t			*bp = inode->i_block;
 	blk_t			addr_per_block;
 	blk64_t			max = EXT2_NDIR_BLOCKS;
+	blk_t			count;
+
+	/* Check start/end don't overflow the 2^32-1 indirect block limit */
+	if (start > BLK_T_MAX)
+		return 0;
+	if (end >= BLK_T_MAX || end - start + 1 >= BLK_T_MAX)
+		count = BLK_T_MAX - start;
+	else
+		count = end - start + 1;
 
 	if (!block_buf) {
 		retval = ext2fs_get_array(3, fs->blocksize, &buf);
@@ -118,11 +128,11 @@ static errcode_t ext2fs_punch_ind(ext2_filsys fs, struct ext2_inode *inode,
 		block_buf = buf;
 	}
 
-	addr_per_block = (blk_t) fs->blocksize >> 2;
+	addr_per_block = (blk_t)fs->blocksize >> 2;
 
 	for (level = 0; level < 4; level++, max *= (blk64_t)addr_per_block) {
 #ifdef PUNCH_DEBUG
-		printf("Main loop level %d, start %u count %u "
+		printf("Main loop level %d, start %llu count %u "
 		       "max %llu num %d\n", level, start, count, max, num);
 #endif
 		if (start < max) {
@@ -149,6 +159,7 @@ errout:
 		ext2fs_free_mem(&buf);
 	return retval;
 }
+#undef BLK_T_MAX
 
 #ifdef PUNCH_DEBUG
 
@@ -428,10 +439,10 @@ errout:
 	ext2fs_extent_free(handle);
 	return retval;
 }
-	
+
 /*
- * Deallocate all logical blocks starting at start to end, inclusive.
- * If end is ~0, then this is effectively truncate.
+ * Deallocate all logical _blocks_ starting at start to end, inclusive.
+ * If end is ~0ULL, then this is effectively truncate.
  */
 errcode_t ext2fs_punch(ext2_filsys fs, ext2_ino_t ino,
 		       struct ext2_inode *inode,
@@ -453,19 +464,14 @@ errcode_t ext2fs_punch(ext2_filsys fs, ext2_ino_t ino,
 	}
 	if (inode->i_flags & EXT4_EXTENTS_FL)
 		retval = ext2fs_punch_extent(fs, ino, inode, start, end);
-	else {
-		blk_t	count;
#1 lore raw

PATCH 2/2 e2fsck: fix e2fsck -fD directory truncation

When an extent-mapped directory is compacted by "e2fsck -fD" and frees enough leaf blocks that it loses an extent tree index block, the old e2fsck_rehash_dir->ext2fs_block_iterate3->write_dir_block() code would not free the extent block, which would result in the extent tree becoming corrupted when it is written out.

Pass 1: Checking inodes, blocks, and sizes
Inode 17825800, end of extent exceeds allowed value
(logical block 710, physical block 570459684, len 1019)

This results in loss of a whole index block of directory leaf blocks and maybe thousands or millions of files in lost+found.

Fix e2fsck_rehash_dir() to call ext2fs_punch() to free the blocks at the end of the directory instead of trying to handle this itself while writing out the directory. That properly handles all of the cases of updating the extent tree as well as accounting for blocks that are released (both leaf blocks and index blocks).

Add a test case for compacting the directory to be smaller than the index block that originally caused the corruption.

---
 e2fsck/rehash.c               |   72 ++++++++++++++++++++++-------------------
 tests/f_extent_htree/expect.1 |   29 ++++++++++++++++
 tests/f_extent_htree/expect.2 |    7 ++++
 tests/f_extent_htree/image.gz |  Bin 0 -> 10101 bytes
 tests/f_extent_htree/name     |    1 +
 tests/f_extent_htree/script   |   69 +++++++++++++++++++++++++++++++++++++++
 tests/f_h_badnode/expect.1    |    2 +-
 tests/f_h_badnode/expect.2    |    2 +-
 8 files changed, 147 insertions(+), 35 deletions(-)
 create mode 100644 tests/f_extent_htree/expect.1
 create mode 100644 tests/f_extent_htree/expect.2
 create mode 100644 tests/f_extent_htree/image.gz
 create mode 100644 tests/f_extent_htree/name
 create mode 100644 tests/f_extent_htree/script
e2fsck/rehash.c +39 -33
index 8ff4883..52d99a3 100644
@@ -52,10 +52,13 @@
 #include "e2fsck.h"
 #include "problem.h"
 
+#undef REHASH_DEBUG
+
 struct fill_dir_struct {
 	char *buf;
 	struct ext2_inode *inode;
 	errcode_t err;
+	ext2_ino_t ino;
 	e2fsck_t ctx;
 	struct hash_entry *harray;
 	int max_array, num_array;
@@ -625,8 +628,8 @@ static errcode_t calculate_tree(ext2_filsys fs,
 struct write_dir_struct {
 	struct out_dir *outdir;
 	errcode_t	err;
+	ext2_ino_t	ino;
 	e2fsck_t	ctx;
-	blk64_t		cleared;
 };
 
 /*
@@ -643,28 +646,35 @@ static int write_dir_block(ext2_filsys fs,
 	blk64_t	blk;
 	char	*dir;
 
-	if (*block_nr == 0)
+#ifdef REHASH_DEBUG
+	printf("%u: write_dir_block %lld:%lld", wd->ino, blockcnt, *block_nr);
+#endif
+	if (*block_nr == 0) {
+#ifdef REHASH_DEBUG
+		printf(" - skip\n");
+#endif
 		return 0;
+	}
+	/* Don't free blocks at the end of the directory, they will be
+	 * truncated by the caller. */
 	if (blockcnt >= wd->outdir->num) {
-		e2fsck_read_bitmaps(wd->ctx);
-		blk = *block_nr;
-		/*
-		 * In theory, we only release blocks from the end of the
-		 * directory file, so it's fine to clobber a whole cluster at
-		 * once.
-		 */
-		if (blk % EXT2FS_CLUSTER_RATIO(fs) == 0) {
-			ext2fs_block_alloc_stats2(fs, blk, -1);
-			wd->cleared++;
-		}
-		*block_nr = 0;
-		return BLOCK_CHANGED;
+#ifdef REHASH_DEBUG
+		printf(" - not freed\n");
+#endif
+		return 0;
 	}
-	if (blockcnt < 0)
+	if (blockcnt < 0) {
+#ifdef REHASH_DEBUG
+		printf(" - skip\n");
+#endif
 		return 0;
+	}
 
 	dir = wd->outdir->buf + (blockcnt * fs->blocksize);
 	wd->err = ext2fs_write_dir_block3(fs, *block_nr, dir, 0);
+#ifdef REHASH_DEBUG
+	printf(" - write (%d)\n", wd->err);
+#endif
 	if (wd->err)
 		return BLOCK_ABORT;
 	return 0;
@@ -684,10 +694,10 @@ static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 
 	wd.outdir = outdir;
 	wd.err = 0;
+	wd.ino = ino;
 	wd.ctx = ctx;
-	wd.cleared = 0;
 
-	retval = ext2fs_block_iterate3(fs, ino, 0, 0,
+	retval = ext2fs_block_iterate3(fs, ino, 0, NULL,
 				       write_dir_block, &wd);
 	if (retval)
 		return retval;
@@ -699,14 +709,17 @@ static errcode_t write_directory(e2fsck_t ctx, ext2_filsys fs,
 		inode.i_flags &= ~EXT2_INDEX_FL;
 	else
 		inode.i_flags |= EXT2_INDEX_FL;
-	retval = ext2fs_inode_size_set(fs, &inode,
-				       outdir->num * fs->blocksize);
+#ifdef REHASH_DEBUG
+	printf("%u: set inode size to %u blocks = %u bytes\n",
+	       ino, outdir->num, outdir->num * fs->blocksize);
+#endif
+	retval = ext2fs_inode_size_set(fs, &inode, (ext2_off64_t)outdir->num *
+						   fs->blocksize);
 	if (retval)
 		return retval;
-	ext2fs_iblk_sub_blocks(fs, &inode, wd.cleared);
-	e2fsck_write_inode(ctx, ino, &inode, "rehash_dir");
 
-	return 0;
+	/* ext2fs_punch() calls ext2fs_write_inode() which writes the size */
+	return ext2fs_punch(fs, ino, &inode, NULL, outdir->num, ~0ULL);
 }
 
 errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
@@ -715,32 +728,25 @@ errcode_t e2fsck_rehash_dir(e2fsck_t ctx, ext2_ino_t ino)
 	errcode_t		retval;
 	struct ext2_inode 	inode;
 	char			*dir_buf = 0;
-	struct fill_dir_struct	fd;
-	struct out_dir		outdir;
+	struct fill_dir_struct	fd = { NULL };
+	struct out_dir		outdir = { 0 };
 
-	outdir.max = outdir.num = 0;
-	outdir.buf = 0;
-	outdir.hashes = 0;
 	e2fsck_read_inode(ctx, ino, &inode, "rehash_dir");
 
 	retval = ENOMEM;
-	fd.harray = 0;
 	dir_buf = malloc(inode.i_size);
 	if (!dir_buf)
 		goto errout;
 
 	fd.max_array = inode.i_size / 32;
-	fd.num_array = 0;
 	fd.harray = malloc(fd.max_array * sizeof(struct hash_entry));
 	if (!fd.harray)
 		goto errout;
 
+	fd.ino = ino;
 	fd.ctx = ctx;
 	fd.buf = dir_buf;
 	fd.inode = &inode;
-	fd.err = 0;
-	fd.dir_size = 0;
-	fd.compress = 0;
 	if (!(fs->super->s_feature_compat & EXT2_FEATURE_COMPAT_DIR_INDEX) ||
 	    (inode.i_size / fs->blocksize) < 2)
 		fd.compress = 1;
tests/f_extent_htree/expect.1 new file +29 -0
new file mode 100644
index 0000000..223ca69
@@ -0,0 +1,29 @@
+Pass 1: Checking inodes, blocks, and sizes
+Pass 2: Checking directory structure
+Pass 3: Checking directory connectivity
+Pass 3A: Optimizing directories
+Pass 4: Checking reference counts
+Pass 5: Checking group summary information
+
+test_filesys: ***** FILE SYSTEM WAS MODIFIED *****
+
+         352 inodes used (41.12%, out of 856)
+           0 non-contiguous files (0.0%)
+           1 non-contiguous directory (0.3%)
+             # of inodes with ind/dind/tind blocks: 0/0/0
+             Extent depth histogram: 342/1
+         586 blocks used (68.94%, out of 850)
+           0 bad blocks
+           0 large files
+
+         340 regular files
+           3 directories
+           0 character device files
+           0 block device files
+           0 fifos
+           0 links
+           0 symbolic links (0 fast symbolic links)
+           0 sockets
+------------
+         343 files
+Exit status is 1
tests/f_extent_htree/expect.2 new file +7 -0
new file mode 100644
index 0000000..860b491
@@ -0,0 +1,7 @@
+Pass 1: Checking inodes, blocks, and sizes
+Pass 2: Checking directory structure
+Pass 3: Checking directory connectivity
+Pass 4: Checking reference counts
+Pass 5: Checking group summary information
+test_filesys: 352/856 files (0.3% non-contiguous), 586/850 blocks
+Exit status is 0
tests/f_extent_htree/image.gz new file binary +0 -0
new file mode 100644
index 0000000000000000000000000000000000000000..284207efb12a04136c9f26c33f2197fed97cbd99
GIT binary patch
tests/f_extent_htree/name new file +1 -0
new file mode 100644
index 0000000..fc3812d
@@ -0,0 +1 @@
+htree extent compression
tests/f_extent_htree/script new file +69 -0

69 lines not on this page · raw message

tests/f_h_badnode/expect.1 +1 -1
index ce2adb3..95b1cee 100644
@@ -14,5 +14,5 @@ Pass 4: Checking reference counts
 Pass 5: Checking group summary information
 
 test_filesys: ***** FILE SYSTEM WAS MODIFIED *****
-test_filesys: 47730/100192 files (0.0% non-contiguous), 13551/31745 blocks
+test_filesys: 47730/100192 files (0.0% non-contiguous), 13550/31745 blocks
 Exit status is 1
tests/f_h_badnode/expect.2 +1 -1
index b9dadb7..65985d1 100644
@@ -3,5 +3,5 @@ Pass 2: Checking directory structure
 Pass 3: Checking directory connectivity
 Pass 4: Checking reference counts
 Pass 5: Checking group summary information
-test_filesys: 47730/100192 files (0.0% non-contiguous), 13551/31745 blocks
+test_filesys: 47730/100192 files (0.0% non-contiguous), 13550/31745 blocks
 Exit status is 0
Signature · 1 line
1.7.3.4
#2 lore raw

Re: [PATCH 2/2] e2fsck: fix e2fsck -fD directory truncation

Quoted 19 lines · Andreas Dilger

When an extent-mapped directory is compacted by "e2fsck -fD" and frees enough leaf blocks that it loses an extent tree index block, the old e2fsck_rehash_dir->ext2fs_block_iterate3->write_dir_block() code would not free the extent block, which would result in the extent tree becoming corrupted when it is written out.

Pass 1: Checking inodes, blocks, and sizes
Inode 17825800, end of extent exceeds allowed value
(logical block 710, physical block 570459684, len 1019)

This results in loss of a whole index block of directory leaf blocks and maybe thousands or millions of files in lost+found.

Fix e2fsck_rehash_dir() to call ext2fs_punch() to free the blocks at the end of the directory instead of trying to handle this itself while writing out the directory. That properly handles all of the cases of updating the extent tree as well as accounting for blocks that are released (both leaf blocks and index blocks).

Quoted 4 lines

Add a test case for compacting the directory to be smaller than the index block that originally caused the corruption.

Thanks, applied.

- Ted

#3 lore raw
In reply to Andreas Dilger
Quoted 10 lines · Andreas Dilger

If ext2fs_punch() was called with "end = ~0ULL" to indicate truncate to the end of file it tried to compute "count" for ext2fs_punch_ind() based on "start" and "end", but incorrectly passed "count = ~0U" even when "start" was non-zero, causing an overflow in some cases.

The calling convention for ext2fs_punch_ind() was also gratuitously different from ext2fs_punch() and ext2fs_punch_extent(), passing "count" instead of "end" as the last parameter. Fix this by passing it "end" like the other functions, and handle "count" internally.

Quoted 4 lines

Add checks to ext2fs_punch_ind() if "end" is at or beyond the 2^32 indirect block limit so the 32-bit internal variables don't overflow.

Thanks, applied.

- Ted