Fix squashfs failing to load sparse files

SquashFS supports sprase blocks in files - that is, if a given block is
composed only of zeros, it's not written to the output file to save
space and it's on-disk length field is set to zero to indicate that.

Previously the squashfs driver did not recognise that, and would attempt
to read and decompress a zero-sized block, which obviously failed.

The following command may be used to create a file for testing:

cat <(dd if=/dev/urandom of=/dev/stdout bs=1M count=1) \
	<(dd if=/dev/zero of=/dev/stdout bs=1M count=1) \
	<(dd if=/dev/urandom of=/dev/stdout bs=1k count=200) >test_file

Signed-off-by: Campbell Suter <campbell@snapit.group>
This commit is contained in:
Campbell Suter 2020-11-23 15:40:03 +13:00 committed by Tom Rini
parent 3b72612ad1
commit 9dba07f143

View File

@ -1310,7 +1310,7 @@ int sqfs_read(const char *filename, void *buf, loff_t offset, loff_t len,
{
char *dir = NULL, *fragment_block, *datablock = NULL, *data_buffer = NULL;
char *fragment = NULL, *file = NULL, *resolved, *data;
u64 start, n_blks, table_size, data_offset, table_offset;
u64 start, n_blks, table_size, data_offset, table_offset, sparse_size;
int ret, j, i_number, datablk_count = 0;
struct squashfs_super_block *sblk = ctxt.sblk;
struct squashfs_fragment_block_entry frag_entry;
@ -1444,28 +1444,43 @@ int sqfs_read(const char *filename, void *buf, loff_t offset, loff_t len,
n_blks = DIV_ROUND_UP(table_size + table_offset,
ctxt.cur_dev->blksz);
data_buffer = malloc_cache_aligned(n_blks * ctxt.cur_dev->blksz);
/* Don't load any data for sparse blocks */
if (finfo.blk_sizes[j] == 0) {
n_blks = 0;
table_offset = 0;
data_buffer = NULL;
data = NULL;
} else {
data_buffer = malloc_cache_aligned(n_blks * ctxt.cur_dev->blksz);
if (!data_buffer) {
ret = -ENOMEM;
goto out;
if (!data_buffer) {
ret = -ENOMEM;
goto out;
}
ret = sqfs_disk_read(start, n_blks, data_buffer);
if (ret < 0) {
/*
* Possible causes: too many data blocks or too large
* SquashFS block size. Tip: re-compile the SquashFS
* image with mksquashfs's -b <block_size> option.
*/
printf("Error: too many data blocks to be read.\n");
goto out;
}
data = data_buffer + table_offset;
}
ret = sqfs_disk_read(start, n_blks, data_buffer);
if (ret < 0) {
/*
* Possible causes: too many data blocks or too large
* SquashFS block size. Tip: re-compile the SquashFS
* image with mksquashfs's -b <block_size> option.
*/
printf("Error: too many data blocks to be read.\n");
goto out;
}
data = data_buffer + table_offset;
/* Load the data */
if (SQFS_COMPRESSED_BLOCK(finfo.blk_sizes[j])) {
if (finfo.blk_sizes[j] == 0) {
/* This is a sparse block */
sparse_size = get_unaligned_le32(&sblk->block_size);
if ((*actread + sparse_size) > len)
sparse_size = len - *actread;
memset(buf + *actread, 0, sparse_size);
*actread += sparse_size;
} else if (SQFS_COMPRESSED_BLOCK(finfo.blk_sizes[j])) {
dest_len = get_unaligned_le32(&sblk->block_size);
ret = sqfs_decompress(&ctxt, datablock, &dest_len,
data, table_size);
@ -1484,7 +1499,8 @@ int sqfs_read(const char *filename, void *buf, loff_t offset, loff_t len,
}
data_offset += table_size;
free(data_buffer);
if (data_buffer)
free(data_buffer);
data_buffer = NULL;
if (*actread >= len)
break;