Libguestfs - Jan 2016 - [supermin] [PATCH] ext2: check for needed block size

Check early that there are enough free blocks to store each file,
erroring out with ENOSPC if not; this avoids slightly more obscure
errors later on.
---
 src/ext2fs-c.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

diff --git a/src/ext2fs-c.c b/src/ext2fs-c.c
index f01ca9d..e45980a 100644
--- a/src/ext2fs-c.c
+++ b/src/ext2fs-c.c
@@ -52,6 +52,9 @@
 /* fts.h in glibc is broken, forcing us to use the GNUlib alternative. */
 #include "fts_.h"
 
+/* How many blocks of size S are needed for storing N bytes. */
+#define ROUND_UP(N, S) (((N) + (S) - 1) / (S))
+
 struct ext2_data
 {
   ext2_filsys fs;
@@ -629,6 +632,7 @@ ext2_copy_file (struct ext2_data *data, const char *src,
const char *dest)
   errcode_t err;
   struct stat statbuf;
   struct statvfs statvfsbuf;
+  size_t blocks;
 
   if (data->debug >= 3)
     printf ("supermin: ext2: copy_file %s -> %s\n", src, dest);
@@ -649,6 +653,20 @@ ext2_copy_file (struct ext2_data *data, const char *src,
const char *dest)
                   caml_copy_string (data->fs->device_name));
   }
 
+  /* Check that we have enough free blocks to store the resulting blocks
+   * for this file.  The file might need more than that in the filesystem,
+   * but at least this provides a quick check to avoid failing later on.
+   */
+  blocks = ROUND_UP (statbuf.st_size, data->fs->blocksize);
+  if (blocks > ext2fs_free_blocks_count (data->fs->super)) {
+    fprintf (stderr, "supermin: %s: needed %lu blocks (%d each) for "
+                     "%lu bytes, available only %llu\n",
+             src, blocks, data->fs->blocksize, statbuf.st_size,
+             ext2fs_free_blocks_count (data->fs->super));
+    unix_error (ENOSPC, (char *) "block size",
+                data->fs->device_name ? caml_copy_string
(data->fs->device_name) : Val_none);
+  }
+
   /* Sanity check the path.  These rules are always true for the paths
    * passed to us here from the appliance layer.  The assertions just
    * verify that the rules haven't changed.
-- 
2.5.0

On Fri, Jan 22, 2016 at 05:37:22PM +0100, Pino Toscano
wrote:
> Check early that there are enough free blocks to store each file,
> erroring out with ENOSPC if not; this avoids slightly more obscure
> errors later on.
> ---
>  src/ext2fs-c.c | 18 ++++++++++++++++++
>  1 file changed, 18 insertions(+)
> 
> diff --git a/src/ext2fs-c.c b/src/ext2fs-c.c
> index f01ca9d..e45980a 100644
> --- a/src/ext2fs-c.c
> +++ b/src/ext2fs-c.c
> @@ -52,6 +52,9 @@
>  /* fts.h in glibc is broken, forcing us to use the GNUlib alternative. */
>  #include "fts_.h"
>  
> +/* How many blocks of size S are needed for storing N bytes. */
> +#define ROUND_UP(N, S) (((N) + (S) - 1) / (S))
> +
>  struct ext2_data
>  {
>    ext2_filsys fs;
> @@ -629,6 +632,7 @@ ext2_copy_file (struct ext2_data *data, const char
*src, const char *dest)
>    errcode_t err;
>    struct stat statbuf;
>    struct statvfs statvfsbuf;
> +  size_t blocks;
>  
>    if (data->debug >= 3)
>      printf ("supermin: ext2: copy_file %s -> %s\n", src,
dest);
> @@ -649,6 +653,20 @@ ext2_copy_file (struct ext2_data *data, const char
*src, const char *dest)
>                    caml_copy_string (data->fs->device_name));
>    }
>  
> +  /* Check that we have enough free blocks to store the resulting blocks
> +   * for this file.  The file might need more than that in the filesystem,
> +   * but at least this provides a quick check to avoid failing later on.
> +   */
> +  blocks = ROUND_UP (statbuf.st_size, data->fs->blocksize);
> +  if (blocks > ext2fs_free_blocks_count (data->fs->super)) {
> +    fprintf (stderr, "supermin: %s: needed %lu blocks (%d each) for
"
> +                     "%lu bytes, available only %llu\n",
> +             src, blocks, data->fs->blocksize, statbuf.st_size,
> +             ext2fs_free_blocks_count (data->fs->super));
> +    unix_error (ENOSPC, (char *) "block size",
> +                data->fs->device_name ? caml_copy_string
(data->fs->device_name) : Val_none);
> +  }
> +
>    /* Sanity check the path.  These rules are always true for the paths
>     * passed to us here from the appliance layer.  The assertions just
>     * verify that the rules haven't changed.
> -- 
> 2.5.0
ACK.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.
http://fedoraproject.org/wiki/MinGW