Re: [PATCH v2] ext4: Annotate struct ext4_xattr_inode_array with __counted_by()

[Thorsten Blum]

On 30. Jul 2024, at 20:12, Gustavo A. R. Silva <gustavo@xxxxxxxxxxxxxx> wrote:
> On 30/07/24 11:23, Thorsten Blum wrote:
>> Add the __counted_by compiler attribute to the flexible array member
>> inodes to improve access bounds-checking via CONFIG_UBSAN_BOUNDS and
>> CONFIG_FORTIFY_SOURCE.
>> Remove the now obsolete comment on the count field.
>> Refactor ext4_expand_inode_array() by assigning count before copying any
>> data using memcpy(). Copy only the inodes array instead of the whole
>> struct because count has been set explicitly.
>> Use struct_size() and struct_size_t() instead of offsetof().
>> Change the data type of the local variable count to unsigned int to
>> match the struct's count data type.
>> Compile-tested only.
>> Signed-off-by: Thorsten Blum <thorsten.blum@xxxxxxxxxx>
>> ---
>> Changes in v2:
>> - Adjust ext4_expand_inode_array() as suggested by Gustavo A. R. Silva
>> - Use struct_size() and struct_size_t() instead of offsetof()
>> - Link to v1: https://lore.kernel.org/linux-kernel/20240729110454.346918-3-thorsten.blum@xxxxxxxxxx/
>> ---
>>  fs/ext4/xattr.c | 20 +++++++++-----------
>>  fs/ext4/xattr.h |  4 ++--
>>  2 files changed, 11 insertions(+), 13 deletions(-)
>> diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
>> index 46ce2f21fef9..b27543587103 100644
>> --- a/fs/ext4/xattr.c
>> +++ b/fs/ext4/xattr.c
>> @@ -2879,11 +2879,10 @@ ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
>>   if (*ea_inode_array == NULL) {
>>   /*
>>   * Start with 15 inodes, so it fits into a power-of-two size.
>> - * If *ea_inode_array is NULL, this is essentially offsetof()
>>   */
>>   (*ea_inode_array) =
>> - kmalloc(offsetof(struct ext4_xattr_inode_array,
>> - inodes[EIA_MASK]),
>> + kmalloc(struct_size_t(struct ext4_xattr_inode_array,
>> +      inodes, EIA_MASK),
> 
> As Kees previously commented, you can use struct_size() here.
> 
>>   GFP_NOFS);
>>   if (*ea_inode_array == NULL)
>>   return -ENOMEM;
>> @@ -2891,17 +2890,16 @@ ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
>>   } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
>>   /* expand the array once all 15 + n * 16 slots are full */
>>   struct ext4_xattr_inode_array *new_array = NULL;
>> - int count = (*ea_inode_array)->count;
>> + unsigned int count = (*ea_inode_array)->count;
> 
> It seems `count` is not actually needed anymore.
> 
> If you remove it and directly use `(*ea_inode_array)->count` in the following
> call to `kmalloc()`, you could use `struct_size()` in the call to `memcpy()`
> below, and copy the whole thing in one line. See below.
> 
>>  - /* if new_array is NULL, this is essentially offsetof() */
>> - new_array = kmalloc(
>> - offsetof(struct ext4_xattr_inode_array,
>> - inodes[count + EIA_INCR]),
>> - GFP_NOFS);
>> + new_array = kmalloc(struct_size(*ea_inode_array, inodes,
>> + count + EIA_INCR),
>> +    GFP_NOFS);
>>   if (new_array == NULL)
>>   return -ENOMEM;
>> - memcpy(new_array, *ea_inode_array,
>> -       offsetof(struct ext4_xattr_inode_array, inodes[count]));
>> + new_array->count = count;
>> + memcpy(new_array->inodes, (*ea_inode_array)->inodes,
>> +       count * sizeof(struct inode *));
> 
> memcpy(new_array, *ea_inode_array, struct_size(new_array, inodes, (*ea_inode_array)->count));
> 
>>   kfree(*ea_inode_array);
>>   *ea_inode_array = new_array;
>>   }
> 
> Also, you are missing one more like just below this one, where `(*ea_inode_array)->count`
> is currently being used to directly index `inodes`:
> 
> (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;

Thanks, I missed this one.

Thorsten