Commit bd5fe6c5 authored by Christoph Hellwig's avatar Christoph Hellwig Committed by Al Viro
Browse files

fs: kill i_alloc_sem 



i_alloc_sem is a rather special rw_semaphore.  It's the last one that may
be released by a non-owner, and it's write side is always mirrored by
real exclusion.  It's intended use it to wait for all pending direct I/O
requests to finish before starting a truncate.

Replace it with a hand-grown construct:

 - exclusion for truncates is already guaranteed by i_mutex, so it can
   simply fall way
 - the reader side is replaced by an i_dio_count member in struct inode
   that counts the number of pending direct I/O requests.  Truncate can't
   proceed as long as it's non-zero
 - when i_dio_count reaches non-zero we wake up a pending truncate using
   wake_up_bit on a new bit in i_flags
 - new references to i_dio_count can't appear while we are waiting for
   it to read zero because the direct I/O count always needs i_mutex
   (or an equivalent like XFS's i_iolock) for starting a new operation.

This scheme is much simpler, and saves the space of a spinlock_t and a
struct list_head in struct inode (typically 160 bits on a non-debug 64-bit
system).

Signed-off-by: default avatarChristoph Hellwig <hch@lst.de>
Signed-off-by: default avatarAl Viro <viro@zeniv.linux.org.uk>
parent f9b5570d

fs/attr.c

+1 −4
Original line number Diff line number Diff line
@@ -233,16 +233,13 @@ int notify_change(struct dentry * dentry, struct iattr * attr) 
		return error;



	if (ia_valid & ATTR_SIZE)

		down_write(&dentry->d_inode->i_alloc_sem);

		inode_dio_wait(inode);



	if (inode->i_op->setattr)

		error = inode->i_op->setattr(dentry, attr);

	else

		error = simple_setattr(dentry, attr);



	if (ia_valid & ATTR_SIZE)

		up_write(&dentry->d_inode->i_alloc_sem);



	if (!error)

		fsnotify_change(dentry, ia_valid);

fs/direct-io.c

+51 −14
Original line number Diff line number Diff line
@@ -135,6 +135,50 @@ struct dio { 
	struct page *pages[DIO_PAGES];	/* page buffer */

};



static void __inode_dio_wait(struct inode *inode)

{

	wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);

	DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);



	do {

		prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);

		if (atomic_read(&inode->i_dio_count))

			schedule();

	} while (atomic_read(&inode->i_dio_count));

	finish_wait(wq, &q.wait);

}



/**

 * inode_dio_wait - wait for outstanding DIO requests to finish

 * @inode: inode to wait for

 *

 * Waits for all pending direct I/O requests to finish so that we can

 * proceed with a truncate or equivalent operation.

 *

 * Must be called under a lock that serializes taking new references

 * to i_dio_count, usually by inode->i_mutex.

 */

void inode_dio_wait(struct inode *inode)

{

	if (atomic_read(&inode->i_dio_count))

		__inode_dio_wait(inode);

}

EXPORT_SYMBOL_GPL(inode_dio_wait);



/*

 * inode_dio_done - signal finish of a direct I/O requests

 * @inode: inode the direct I/O happens on

 *

 * This is called once we've finished processing a direct I/O request,

 * and is used to wake up callers waiting for direct I/O to be quiesced.

 */

void inode_dio_done(struct inode *inode)

{

	if (atomic_dec_and_test(&inode->i_dio_count))

		wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);

}

EXPORT_SYMBOL_GPL(inode_dio_done);



/*

 * How many pages are in the queue?

 */
@@ -254,9 +298,7 @@ static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is 
	}



	if (dio->flags & DIO_LOCKING)

		/* lockdep: non-owner release */

		up_read_non_owner(&dio->inode->i_alloc_sem);



		inode_dio_done(dio->inode);

	return ret;

}
@@ -980,9 +1022,6 @@ out: 
	return ret;

}



/*

 * Releases both i_mutex and i_alloc_sem

 */

static ssize_t

direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode, 

	const struct iovec *iov, loff_t offset, unsigned long nr_segs,
@@ -1146,15 +1185,14 @@ direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode, 
 *    For writes this function is called under i_mutex and returns with

 *    i_mutex held, for reads, i_mutex is not held on entry, but it is

 *    taken and dropped again before returning.

 *    For reads and writes i_alloc_sem is taken in shared mode and released

 *    on I/O completion (which may happen asynchronously after returning to

 *    the caller).

 *    The i_dio_count counter keeps track of the number of outstanding

 *    direct I/O requests, and truncate waits for it to reach zero.

 *    New references to i_dio_count must only be grabbed with i_mutex

 *    held.

 *

 *  - if the flags value does NOT contain DIO_LOCKING we don't use any

 *    internal locking but rather rely on the filesystem to synchronize

 *    direct I/O reads/writes versus each other and truncate.

 *    For reads and writes both i_mutex and i_alloc_sem are not held on

 *    entry and are never taken.

 */

ssize_t

__blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
@@ -1234,10 +1272,9 @@ __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode, 
		}



		/*

		 * Will be released at I/O completion, possibly in a

		 * different thread.

		 * Will be decremented at I/O completion time.

		 */

		down_read_non_owner(&inode->i_alloc_sem);

		atomic_inc(&inode->i_dio_count);

	}



	/*

fs/inode.c

+1 −2
Original line number Diff line number Diff line
@@ -168,8 +168,7 @@ int inode_init_always(struct super_block *sb, struct inode *inode) 
	mutex_init(&inode->i_mutex);

	lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);



	init_rwsem(&inode->i_alloc_sem);

	lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key);

	atomic_set(&inode->i_dio_count, 0);



	mapping->a_ops = &empty_aops;

	mapping->host = inode;

fs/ntfs/file.c

+1 −2
Original line number Diff line number Diff line
@@ -1832,9 +1832,8 @@ static ssize_t ntfs_file_buffered_write(struct kiocb *iocb, 
	 * fails again.

	 */

	if (unlikely(NInoTruncateFailed(ni))) {

		down_write(&vi->i_alloc_sem);

		inode_dio_wait(vi);

		err = ntfs_truncate(vi);

		up_write(&vi->i_alloc_sem);

		if (err || NInoTruncateFailed(ni)) {

			if (!err)

				err = -EIO;

fs/ntfs/inode.c

+2 −8
Original line number Diff line number Diff line
@@ -2357,12 +2357,7 @@ static const char *es = " Leaving inconsistent metadata. Unmount and run " 
 *

 * Returns 0 on success or -errno on error.

 *

 * Called with ->i_mutex held.  In all but one case ->i_alloc_sem is held for

 * writing.  The only case in the kernel where ->i_alloc_sem is not held is

 * mm/filemap.c::generic_file_buffered_write() where vmtruncate() is called

 * with the current i_size as the offset.  The analogous place in NTFS is in

 * fs/ntfs/file.c::ntfs_file_buffered_write() where we call vmtruncate() again

 * without holding ->i_alloc_sem.

 * Called with ->i_mutex held.

 */

int ntfs_truncate(struct inode *vi)

{
@@ -2887,8 +2882,7 @@ void ntfs_truncate_vfs(struct inode *vi) { 
 * We also abort all changes of user, group, and mode as we do not implement

 * the NTFS ACLs yet.

 *

 * Called with ->i_mutex held.  For the ATTR_SIZE (i.e. ->truncate) case, also

 * called with ->i_alloc_sem held for writing.

 * Called with ->i_mutex held.

 */

int ntfs_setattr(struct dentry *dentry, struct iattr *attr)

{

CRA Git | Maintained and supported by SUSTech CRA and CCSE