xfs_buf.c

linux-2.4.29操作系统的源码

	if (! (free = pb_resv_bh_cnt >= NR_RESERVED_BH)) {
		spin_lock_irqsave(&pb_resv_bh_lock, flags);

		if (! (free = pb_resv_bh_cnt >= NR_RESERVED_BH)) {
			bh->b_pprev = &pb_resv_bh;
			bh->b_next = pb_resv_bh;
			pb_resv_bh = bh;
			pb_resv_bh_cnt++;

			if (waitqueue_active(&pb_resv_bh_wait)) {
				wake_up(&pb_resv_bh_wait);
			}
		}

		spin_unlock_irqrestore(&pb_resv_bh_lock, flags);
	}
	if (free) {
		kmem_cache_free(bh_cachep, bh);
	}
}

/*
 *	Finding and Reading Buffers
 */

/*
 *	_pagebuf_find
 *
 *	Looks up, and creates if absent, a lockable buffer for
 *	a given range of an inode.  The buffer is returned
 *	locked.	 If other overlapping buffers exist, they are
 *	released before the new buffer is created and locked,
 *	which may imply that this call will block until those buffers
 *	are unlocked.  No I/O is implied by this call.
 */
xfs_buf_t *
_pagebuf_find(				/* find buffer for block	*/
	xfs_buftarg_t		*target,/* target for block		*/
	loff_t			ioff,	/* starting offset of range	*/
	size_t			isize,	/* length of range		*/
	page_buf_flags_t	flags,	/* PBF_TRYLOCK			*/
	xfs_buf_t		*new_pb)/* newly allocated buffer	*/
{
	loff_t			range_base;
	size_t			range_length;
	int			hval;
	pb_hash_t		*h;
	xfs_buf_t		*pb, *n;
	int			not_locked;

	range_base = (ioff << BBSHIFT);
	range_length = (isize << BBSHIFT);

	/* Ensure we never do IOs smaller than the sector size */
	BUG_ON(range_length < (1 << target->pbr_sshift));

	/* Ensure we never do IOs that are not sector aligned */
	BUG_ON(range_base & (loff_t)target->pbr_smask);

	hval = _bhash(target->pbr_bdev, range_base);
	h = &pbhash[hval];

	spin_lock(&h->pb_hash_lock);

	list_for_each_entry_safe(pb, n, &h->pb_hash, pb_hash_list) {
		if (pb->pb_target == target &&
		    pb->pb_file_offset == range_base &&
		    pb->pb_buffer_length == range_length) {
			/* If we look at something bring it to the
			 * front of the list for next time
			 */
			atomic_inc(&pb->pb_hold);
			list_move(&pb->pb_hash_list, &h->pb_hash);
			goto found;
		}
	}

	/* No match found */
	if (new_pb) {
		_pagebuf_initialize(new_pb, target, range_base,
				range_length, flags);
		new_pb->pb_hash_index = hval;
		list_add(&new_pb->pb_hash_list, &h->pb_hash);
	} else {
		XFS_STATS_INC(pb_miss_locked);
	}

	spin_unlock(&h->pb_hash_lock);
	return (new_pb);

found:
	spin_unlock(&h->pb_hash_lock);

	/* Attempt to get the semaphore without sleeping,
	 * if this does not work then we need to drop the
	 * spinlock and do a hard attempt on the semaphore.
	 */
	not_locked = down_trylock(&pb->pb_sema);
	if (not_locked) {
		if (!(flags & PBF_TRYLOCK)) {
			/* wait for buffer ownership */
			PB_TRACE(pb, "get_lock", 0);
			pagebuf_lock(pb);
			XFS_STATS_INC(pb_get_locked_waited);
		} else {
			/* We asked for a trylock and failed, no need
			 * to look at file offset and length here, we
			 * know that this pagebuf at least overlaps our
			 * pagebuf and is locked, therefore our buffer
			 * either does not exist, or is this buffer
			 */

			pagebuf_rele(pb);
			XFS_STATS_INC(pb_busy_locked);
			return (NULL);
		}
	} else {
		/* trylock worked */
		PB_SET_OWNER(pb);
	}

	if (pb->pb_flags & PBF_STALE)
		pb->pb_flags &= PBF_MAPPED;
	PB_TRACE(pb, "got_lock", 0);
	XFS_STATS_INC(pb_get_locked);
	return (pb);
}

/*
 *	xfs_buf_get_flags assembles a buffer covering the specified range.
 *
 *	Storage in memory for all portions of the buffer will be allocated,
 *	although backing storage may not be.
 */
xfs_buf_t *
xfs_buf_get_flags(			/* allocate a buffer		*/
	xfs_buftarg_t		*target,/* target for buffer		*/
	loff_t			ioff,	/* starting offset of range	*/
	size_t			isize,	/* length of range		*/
	page_buf_flags_t	flags)	/* PBF_TRYLOCK			*/
{
	xfs_buf_t		*pb, *new_pb;
	int			error = 0, i;

	new_pb = pagebuf_allocate(flags);
	if (unlikely(!new_pb))
		return NULL;

	pb = _pagebuf_find(target, ioff, isize, flags, new_pb);
	if (pb == new_pb) {
		error = _pagebuf_lookup_pages(pb, flags);
		if (error)
			goto no_buffer;
	} else {
		pagebuf_deallocate(new_pb);
		if (unlikely(pb == NULL))
			return NULL;
	}

	for (i = 0; i < pb->pb_page_count; i++)
		mark_page_accessed(pb->pb_pages[i]);

	if (!(pb->pb_flags & PBF_MAPPED)) {
		error = _pagebuf_map_pages(pb, flags);
		if (unlikely(error)) {
			printk(KERN_WARNING "%s: failed to map pages\n",
					__FUNCTION__);
			goto no_buffer;
		}
	}

	XFS_STATS_INC(pb_get);

	/*
	 * Always fill in the block number now, the mapped cases can do
	 * their own overlay of this later.
	 */
	pb->pb_bn = ioff;
	pb->pb_count_desired = pb->pb_buffer_length;

	PB_TRACE(pb, "get", (unsigned long)flags);
	return pb;

 no_buffer:
	if (flags & (PBF_LOCK | PBF_TRYLOCK))
		pagebuf_unlock(pb);
	pagebuf_rele(pb);
	return NULL;
}

xfs_buf_t *
xfs_buf_read_flags(
	xfs_buftarg_t		*target,
	loff_t			ioff,
	size_t			isize,
	page_buf_flags_t	flags)
{
	xfs_buf_t		*pb;

	flags |= PBF_READ;

	pb = xfs_buf_get_flags(target, ioff, isize, flags);
	if (pb) {
		if (PBF_NOT_DONE(pb)) {
			PB_TRACE(pb, "read", (unsigned long)flags);
			XFS_STATS_INC(pb_get_read);
			pagebuf_iostart(pb, flags);
		} else if (flags & PBF_ASYNC) {
			PB_TRACE(pb, "read_async", (unsigned long)flags);
			/*
			 * Read ahead call which is already satisfied,
			 * drop the buffer
			 */
			goto no_buffer;
		} else {
			PB_TRACE(pb, "read_done", (unsigned long)flags);
			/* We do not want read in the flags */
			pb->pb_flags &= ~PBF_READ;
		}
	}

	return pb;

 no_buffer:
	if (flags & (PBF_LOCK | PBF_TRYLOCK))
		pagebuf_unlock(pb);
	pagebuf_rele(pb);
	return NULL;
}

/*
 * Create a skeletal pagebuf (no pages associated with it).
 */
xfs_buf_t *
pagebuf_lookup(
	xfs_buftarg_t		*target,
	loff_t			ioff,
	size_t			isize,
	page_buf_flags_t	flags)
{
	xfs_buf_t		*pb;

	flags |= _PBF_PRIVATE_BH;
	pb = pagebuf_allocate(flags);
	if (pb) {
		_pagebuf_initialize(pb, target, ioff, isize, flags);
	}
	return pb;
}

/*
 * If we are not low on memory then do the readahead in a deadlock
 * safe manner.
 */
void
pagebuf_readahead(
	xfs_buftarg_t		*target,
	loff_t			ioff,
	size_t			isize,
	page_buf_flags_t	flags)
{
	flags |= (PBF_TRYLOCK|PBF_ASYNC|PBF_READ_AHEAD);
	xfs_buf_read_flags(target, ioff, isize, flags);
}

xfs_buf_t *
pagebuf_get_empty(
	size_t			len,
	xfs_buftarg_t		*target)
{
	xfs_buf_t		*pb;

	pb = pagebuf_allocate(0);
	if (pb)
		_pagebuf_initialize(pb, target, 0, len, 0);
	return pb;
}

static inline struct page *
mem_to_page(
	void			*addr)
{
	if (((unsigned long)addr < VMALLOC_START) ||
	    ((unsigned long)addr >= VMALLOC_END)) {
		return virt_to_page(addr);
	} else {
		return vmalloc_to_page(addr);
	}
}

int
pagebuf_associate_memory(
	xfs_buf_t		*pb,
	void			*mem,
	size_t			len)
{
	int			rval;
	int			i = 0;
	size_t			ptr;
	size_t			end, end_cur;
	off_t			offset;
	int			page_count;

	page_count = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
	offset = (off_t) mem - ((off_t)mem & PAGE_CACHE_MASK);
	if (offset && (len > PAGE_CACHE_SIZE))
		page_count++;

	/* Free any previous set of page pointers */
	if (pb->pb_pages)
		_pagebuf_free_pages(pb);

	pb->pb_pages = NULL;
	pb->pb_addr = mem;

	rval = _pagebuf_get_pages(pb, page_count, 0);
	if (rval)
		return rval;

	pb->pb_offset = offset;
	ptr = (size_t) mem & PAGE_CACHE_MASK;
	end = PAGE_CACHE_ALIGN((size_t) mem + len);
	end_cur = end;
	/* set up first page */
	pb->pb_pages[0] = mem_to_page(mem);

	ptr += PAGE_CACHE_SIZE;
	pb->pb_page_count = ++i;
	while (ptr < end) {
		pb->pb_pages[i] = mem_to_page((void *)ptr);
		pb->pb_page_count = ++i;
		ptr += PAGE_CACHE_SIZE;
	}
	pb->pb_locked = 0;

	pb->pb_count_desired = pb->pb_buffer_length = len;
	pb->pb_flags |= PBF_MAPPED | _PBF_PRIVATE_BH;

	return 0;
}

xfs_buf_t *
pagebuf_get_no_daddr(
	size_t			len,
	xfs_buftarg_t		*target)
{
	size_t			malloc_len = len;
	xfs_buf_t		*bp;
	void			*data;
	int			error;

	bp = pagebuf_allocate(0);
	if (unlikely(bp == NULL))
		goto fail;
	_pagebuf_initialize(bp, target, 0, len, PBF_FORCEIO);

 try_again:
	data = kmem_alloc(malloc_len, KM_SLEEP | KM_MAYFAIL);
	if (unlikely(data == NULL))
		goto fail_free_buf;

	/* check whether alignment matches.. */
	if ((__psunsigned_t)data !=
	    ((__psunsigned_t)data & ~target->pbr_smask)) {
		/* .. else double the size and try again */
		kmem_free(data, malloc_len);
		malloc_len <<= 1;
		goto try_again;
	}

	error = pagebuf_associate_memory(bp, data, len);
	if (error)
		goto fail_free_mem;
	bp->pb_flags |= _PBF_KMEM_ALLOC;

	pagebuf_unlock(bp);

	PB_TRACE(bp, "no_daddr", data);
	return bp;
 fail_free_mem:
	kmem_free(data, malloc_len);
 fail_free_buf:
	pagebuf_free(bp);
 fail:
	return NULL;
}

/*
 *	pagebuf_hold
 *
 *	Increment reference count on buffer, to hold the buffer concurrently
 *	with another thread which may release (free) the buffer asynchronously.
 *
 *	Must hold the buffer already to call this function.
 */
void
pagebuf_hold(
	xfs_buf_t		*pb)
{
	atomic_inc(&pb->pb_hold);
	PB_TRACE(pb, "hold", 0);
}

/*
 *	pagebuf_rele
 *
 *	pagebuf_rele releases a hold on the specified buffer.  If the
 *	the hold count is 1, pagebuf_rele calls pagebuf_free.
 */
void
pagebuf_rele(
	xfs_buf_t		*pb)
{
	pb_hash_t		*hash = pb_hash(pb);

	PB_TRACE(pb, "rele", pb->pb_relse);

	if (atomic_dec_and_lock(&pb->pb_hold, &hash->pb_hash_lock)) {
		int		do_free = 1;

		if (pb->pb_relse) {
			atomic_inc(&pb->pb_hold);
			spin_unlock(&hash->pb_hash_lock);
			(*(pb->pb_relse)) (pb);
			spin_lock(&hash->pb_hash_lock);
			do_free = 0;
		}

		if (pb->pb_flags & PBF_DELWRI) {
			pb->pb_flags |= PBF_ASYNC;
			atomic_inc(&pb->pb_hold);
			pagebuf_delwri_queue(pb, 0);
			do_free = 0;
		} else if (pb->pb_flags & PBF_FS_MANAGED) {
			do_free = 0;
		}

		if (do_free) {
			list_del_init(&pb->pb_hash_list);
			spin_unlock(&hash->pb_hash_lock);
			xfs_buf_free(pb);
		} else {
			spin_unlock(&hash->pb_hash_lock);
		}
	}
}


/*
 *	Mutual exclusion on buffers.  Locking model:
 *
 *	Buffers associated with inodes for which buffer locking
 *	is not enabled are not protected by semaphores, and are
 *	assumed to be exclusively owned by the caller.  There is a
 *	spinlock in the buffer, used by the caller when concurrent
 *	access is possible.
 */

/*
 *	pagebuf_cond_lock
 *
 *	pagebuf_cond_lock locks a buffer object, if it is not already locked.
 *	Note that this in no way
 *	locks the underlying pages, so it is only useful for synchronizing
 *	concurrent use of page buffer objects, not for synchronizing independent
 *	access to the underlying pages.
 */
int
pagebuf_cond_lock(			/* lock buffer, if not locked	*/
					/* returns -EBUSY if locked)	*/
	xfs_buf_t		*pb)
{
	int			locked;

	locked = down_trylock(&pb->pb_sema) == 0;
	if (locked) {
		PB_SET_OWNER(pb);
	}
	PB_TRACE(pb, "cond_lock", (long)locked);
	return(locked ? 0 : -EBUSY);
}

#ifdef DEBUG
/*
 *	pagebuf_lock_value
 *
 *	Return lock value for a pagebuf
 */
int
pagebuf_lock_value(
	xfs_buf_t		*pb)
{
	return(atomic_read(&pb->pb_sema.count));
}
#endif

/*
 *	pagebuf_lock
 *
 *	pagebuf_lock locks a buffer object.  Note that this in no way
 *	locks the underlying pages, so it is only useful for synchronizing
 *	concurrent use of page buffer objects, not for synchronizing independent
 *	access to the underlying pages.
 */
int
pagebuf_lock(
	xfs_buf_t		*pb)
{
	PB_TRACE(pb, "lock", 0);
	if (atomic_read(&pb->pb_io_remaining))
		run_task_queue(&tq_disk);
	down(&pb->pb_sema);
	PB_SET_OWNER(pb);
	PB_TRACE(pb, "locked", 0);
	return 0;
}

/*
 *	pagebuf_unlock
 *
 *	pagebuf_unlock releases the lock on the buffer object created by
 *	pagebuf_lock or pagebuf_cond_lock (not any
 *	pinning of underlying pages created by pagebuf_pin).
 */
void
pagebuf_unlock(				/* unlock buffer		*/
	xfs_buf_t		*pb)	/* buffer to unlock		*/
{
	PB_CLEAR_OWNER(pb);
	up(&pb->pb_sema);
	PB_TRACE(pb, "unlock", 0);
}


/*
 *	Pinning Buffer Storage in Memory
 */

/*
 *	pagebuf_pin
 *
 *	pagebuf_pin locks all of the memory represented by a buffer in
 *	memory.  Multiple calls to pagebuf_pin and pagebuf_unpin, for
 *	the same or different buffers affecting a given page, will
 *	properly count the number of outstanding "pin" requests.  The
 *	buffer may be released after the pagebuf_pin and a different
 *	buffer used when calling pagebuf_unpin, if desired.
 *	pagebuf_pin should be used by the file system when it wants be
 *	assured that no attempt will be made to force the affected
 *	memory to disk.	 It does not assure that a given logical page
 *	will not be moved to a different physical page.
 */
void
pagebuf_pin(
	xfs_buf_t		*pb)
{
	atomic_inc(&pb->pb_pin_count);
	PB_TRACE(pb, "pin", (long)pb->pb_pin_count.counter);
}

/*
 *	pagebuf_unpin
 *
 *	pagebuf_unpin reverses the locking of memory performed by
 *	pagebuf_pin.  Note that both functions affected the logical
 *	pages associated with the buffer, not the buffer itself.
 */
void
pagebuf_unpin(
	xfs_buf_t		*pb)
{
	if (atomic_dec_and_test(&pb->pb_pin_count)) {
		wake_up_all(&pb->pb_waiters);
	}
	PB_TRACE(pb, "unpin", (long)pb->pb_pin_count.counter);
}

int
pagebuf_ispin(
	xfs_buf_t		*pb)
{
	return atomic_read(&pb->pb_pin_count);
}

/*
 *	pagebuf_wait_unpin
 *
 *	pagebuf_wait_unpin waits until all of the memory associated
 *	with the buffer is not longer locked in memory.  It returns
 *	immediately if none of the affected pages are locked.
 */
static inline void
_pagebuf_wait_unpin(