diff mbox

[RFC,V2] ext3: add ioctl to force 32-bit hashes from indexed dirs

Message ID 5154AAB4.2000701@redhat.com
State Not Applicable, archived
Headers show

Commit Message

Eric Sandeen March 28, 2013, 8:40 p.m. UTC
This adds a new ioctl, EXT3_IOC_32BITHASH, which allows a
userspace application to request 32-bit rather than 64-bit
hashes from readdir on an indexed / dx / htree directory.

Gluster had been relying on the top bits of the d_off being
free; there are some reports that filling all 64 bits breaks
Samba as well.  The infrastructure to return 32-bit hashes
already exists; NFS can turn it on, and it's turned on for
32-bit processes as well.  So it's just a matter of flipping
on the f_mode flag before readdir starts.

Care needs to be taken that we don't change the FMODE flag
after readdir has been started, so we make sure that
filp->private_data has not yet been set before we set the flag.
(Thanks Zach!).

Pre-submission-fixes-by: Zach Brown <zab@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
---

V2:
fix "readir" typo
rename goto target to *_out like others
remove parameter; we can't really ever turn this back off once it's used.
closing and reopening is the only way to get back to 64 bit hashes.



--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Jan Kara March 29, 2013, 11:43 a.m. UTC | #1
On Thu 28-03-13 15:40:20, Eric Sandeen wrote:
> This adds a new ioctl, EXT3_IOC_32BITHASH, which allows a
> userspace application to request 32-bit rather than 64-bit
> hashes from readdir on an indexed / dx / htree directory.
> 
> Gluster had been relying on the top bits of the d_off being
> free; there are some reports that filling all 64 bits breaks
> Samba as well.  The infrastructure to return 32-bit hashes
> already exists; NFS can turn it on, and it's turned on for
> 32-bit processes as well.  So it's just a matter of flipping
> on the f_mode flag before readdir starts.
> 
> Care needs to be taken that we don't change the FMODE flag
> after readdir has been started, so we make sure that
> filp->private_data has not yet been set before we set the flag.
> (Thanks Zach!).
> 
> Pre-submission-fixes-by: Zach Brown <zab@redhat.com>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> ---
> 
> V2:
> fix "readir" typo
> rename goto target to *_out like others
> remove parameter; we can't really ever turn this back off once it's used.
> closing and reopening is the only way to get back to 64 bit hashes.
  Looks good. Just one nit below:

> diff --git a/fs/ext3/ext3.h b/fs/ext3/ext3.h
> +	case EXT3_IOC_32BITHASH: {
> +		int err = 0;
> +
> +		/* Serialize with readdir */
> +		if ((err = mutex_lock_killable(&inode->i_mutex)))
> +			return err;
> +
> +		/* protect f_mode */
> +		spin_lock(&filp->f_lock);
> +
> +		/* Only valid for htree directories */
> +		if (!S_ISDIR(inode->i_mode) || !is_dx_dir(inode)) {
  Won't it be better to return ENOTDIR in !S_ISDIR case?

> +			err = -EINVAL;
> +			goto hash32bits_out;
> +		}
> +
> +		/* Have we already started readdir on this dx dir? */
> +		if (filp->private_data) {
> +			err = -EINVAL;
  And here maybe EBUSY, but in this case I'm really undecided.

> +			goto hash32bits_out;
> +		}
> +
> +		filp->f_mode |= FMODE_32BITHASH;
> +hash32bits_out:
> +		spin_unlock(&filp->f_lock);
> +		mutex_unlock(&inode->i_mutex);
> +		return err;
> +	}
>  	case FITRIM: {
>  
>  		struct super_block *sb = inode->i_sb;

								Honza
Eric Sandeen April 1, 2013, 3:13 p.m. UTC | #2
On 3/29/13 6:43 AM, Jan Kara wrote:
> On Thu 28-03-13 15:40:20, Eric Sandeen wrote:
>> This adds a new ioctl, EXT3_IOC_32BITHASH, which allows a
>> userspace application to request 32-bit rather than 64-bit
>> hashes from readdir on an indexed / dx / htree directory.
>>
>> Gluster had been relying on the top bits of the d_off being
>> free; there are some reports that filling all 64 bits breaks
>> Samba as well.  The infrastructure to return 32-bit hashes
>> already exists; NFS can turn it on, and it's turned on for
>> 32-bit processes as well.  So it's just a matter of flipping
>> on the f_mode flag before readdir starts.
>>
>> Care needs to be taken that we don't change the FMODE flag
>> after readdir has been started, so we make sure that
>> filp->private_data has not yet been set before we set the flag.
>> (Thanks Zach!).
>>
>> Pre-submission-fixes-by: Zach Brown <zab@redhat.com>
>> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
>> ---
>>
>> V2:
>> fix "readir" typo
>> rename goto target to *_out like others
>> remove parameter; we can't really ever turn this back off once it's used.
>> closing and reopening is the only way to get back to 64 bit hashes.
>   Looks good. Just one nit below:

Oh, of course, duh!  I'll send V3, thanks.

-Eric

>> diff --git a/fs/ext3/ext3.h b/fs/ext3/ext3.h
>> +	case EXT3_IOC_32BITHASH: {
>> +		int err = 0;
>> +
>> +		/* Serialize with readdir */
>> +		if ((err = mutex_lock_killable(&inode->i_mutex)))
>> +			return err;
>> +
>> +		/* protect f_mode */
>> +		spin_lock(&filp->f_lock);
>> +
>> +		/* Only valid for htree directories */
>> +		if (!S_ISDIR(inode->i_mode) || !is_dx_dir(inode)) {
>   Won't it be better to return ENOTDIR in !S_ISDIR case?
> 
>> +			err = -EINVAL;
>> +			goto hash32bits_out;
>> +		}
>> +
>> +		/* Have we already started readdir on this dx dir? */
>> +		if (filp->private_data) {
>> +			err = -EINVAL;
>   And here maybe EBUSY, but in this case I'm really undecided.
> 
>> +			goto hash32bits_out;
>> +		}
>> +
>> +		filp->f_mode |= FMODE_32BITHASH;
>> +hash32bits_out:
>> +		spin_unlock(&filp->f_lock);
>> +		mutex_unlock(&inode->i_mutex);
>> +		return err;
>> +	}
>>  	case FITRIM: {
>>  
>>  		struct super_block *sb = inode->i_sb;
> 
> 								Honza
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/fs/ext3/dir.c b/fs/ext3/dir.c
index 87eccbb..83df29f 100644
--- a/fs/ext3/dir.c
+++ b/fs/ext3/dir.c
@@ -47,7 +47,7 @@  static unsigned char get_dtype(struct super_block *sb, int filetype)
  *
  * Return 1 if it is a dx dir, 0 if not
  */
-static int is_dx_dir(struct inode *inode)
+int is_dx_dir(struct inode *inode)
 {
 	struct super_block *sb = inode->i_sb;
 
diff --git a/fs/ext3/ext3.h b/fs/ext3/ext3.h
index e85ff15..f3018f4 100644
--- a/fs/ext3/ext3.h
+++ b/fs/ext3/ext3.h
@@ -220,6 +220,7 @@  struct ext3_new_group_data {
 #endif
 #define EXT3_IOC_GETRSVSZ		_IOR('f', 5, long)
 #define EXT3_IOC_SETRSVSZ		_IOW('f', 6, long)
+#define EXT3_IOC_32BITHASH		_IO('f', 13)
 
 /*
  * ioctl commands in 32 bit emulation
@@ -1010,6 +1011,7 @@  extern void ext3_rsv_window_add(struct super_block *sb, struct ext3_reserve_wind
 extern int ext3_trim_fs(struct super_block *sb, struct fstrim_range *range);
 
 /* dir.c */
+extern int is_dx_dir(struct inode *inode);
 extern int ext3_check_dir_entry(const char *, struct inode *,
 				struct ext3_dir_entry_2 *,
 				struct buffer_head *, unsigned long);
diff --git a/fs/ext3/ioctl.c b/fs/ext3/ioctl.c
index 4d96e9a..3811664 100644
--- a/fs/ext3/ioctl.c
+++ b/fs/ext3/ioctl.c
@@ -251,6 +251,34 @@  group_add_out:
 		mnt_drop_write_file(filp);
 		return err;
 	}
+	case EXT3_IOC_32BITHASH: {
+		int err = 0;
+
+		/* Serialize with readdir */
+		if ((err = mutex_lock_killable(&inode->i_mutex)))
+			return err;
+
+		/* protect f_mode */
+		spin_lock(&filp->f_lock);
+
+		/* Only valid for htree directories */
+		if (!S_ISDIR(inode->i_mode) || !is_dx_dir(inode)) {
+			err = -EINVAL;
+			goto hash32bits_out;
+		}
+
+		/* Have we already started readdir on this dx dir? */
+		if (filp->private_data) {
+			err = -EINVAL;
+			goto hash32bits_out;
+		}
+
+		filp->f_mode |= FMODE_32BITHASH;
+hash32bits_out:
+		spin_unlock(&filp->f_lock);
+		mutex_unlock(&inode->i_mutex);
+		return err;
+	}
 	case FITRIM: {
 
 		struct super_block *sb = inode->i_sb;