xfs_buf.c

linux-2.6.15.6

/*
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */
#include <linux/stddef.h>
#include <linux/errno.h>
#include <linux/slab.h>
#include <linux/pagemap.h>
#include <linux/init.h>
#include <linux/vmalloc.h>
#include <linux/bio.h>
#include <linux/sysctl.h>
#include <linux/proc_fs.h>
#include <linux/workqueue.h>
#include <linux/percpu.h>
#include <linux/blkdev.h>
#include <linux/hash.h>
#include <linux/kthread.h>
#include "xfs_linux.h"

STATIC kmem_cache_t *pagebuf_zone;
STATIC kmem_shaker_t pagebuf_shake;
STATIC int xfsbufd_wakeup(int, gfp_t);
STATIC void pagebuf_delwri_queue(xfs_buf_t *, int);

STATIC struct workqueue_struct *xfslogd_workqueue;
struct workqueue_struct *xfsdatad_workqueue;

#ifdef PAGEBUF_TRACE
void
pagebuf_trace(
	xfs_buf_t	*pb,
	char		*id,
	void		*data,
	void		*ra)
{
	ktrace_enter(pagebuf_trace_buf,
		pb, id,
		(void *)(unsigned long)pb->pb_flags,
		(void *)(unsigned long)pb->pb_hold.counter,
		(void *)(unsigned long)pb->pb_sema.count.counter,
		(void *)current,
		data, ra,
		(void *)(unsigned long)((pb->pb_file_offset>>32) & 0xffffffff),
		(void *)(unsigned long)(pb->pb_file_offset & 0xffffffff),
		(void *)(unsigned long)pb->pb_buffer_length,
		NULL, NULL, NULL, NULL, NULL);
}
ktrace_t *pagebuf_trace_buf;
#define PAGEBUF_TRACE_SIZE	4096
#define PB_TRACE(pb, id, data)	\
	pagebuf_trace(pb, id, (void *)data, (void *)__builtin_return_address(0))
#else
#define PB_TRACE(pb, id, data)	do { } while (0)
#endif

#ifdef PAGEBUF_LOCK_TRACKING
# define PB_SET_OWNER(pb)	((pb)->pb_last_holder = current->pid)
# define PB_CLEAR_OWNER(pb)	((pb)->pb_last_holder = -1)
# define PB_GET_OWNER(pb)	((pb)->pb_last_holder)
#else
# define PB_SET_OWNER(pb)	do { } while (0)
# define PB_CLEAR_OWNER(pb)	do { } while (0)
# define PB_GET_OWNER(pb)	do { } while (0)
#endif

#define pb_to_gfp(flags) \
	((((flags) & PBF_READ_AHEAD) ? __GFP_NORETRY : \
	  ((flags) & PBF_DONT_BLOCK) ? GFP_NOFS : GFP_KERNEL) | __GFP_NOWARN)

#define pb_to_km(flags) \
	 (((flags) & PBF_DONT_BLOCK) ? KM_NOFS : KM_SLEEP)

#define pagebuf_allocate(flags) \
	kmem_zone_alloc(pagebuf_zone, pb_to_km(flags))
#define pagebuf_deallocate(pb) \
	kmem_zone_free(pagebuf_zone, (pb));

/*
 * Page Region interfaces.
 *
 * For pages in filesystems where the blocksize is smaller than the
 * pagesize, we use the page->private field (long) to hold a bitmap
 * of uptodate regions within the page.
 *
 * Each such region is "bytes per page / bits per long" bytes long.
 *
 * NBPPR == number-of-bytes-per-page-region
 * BTOPR == bytes-to-page-region (rounded up)
 * BTOPRT == bytes-to-page-region-truncated (rounded down)
 */
#if (BITS_PER_LONG == 32)
#define PRSHIFT		(PAGE_CACHE_SHIFT - 5)	/* (32 == 1<<5) */
#elif (BITS_PER_LONG == 64)
#define PRSHIFT		(PAGE_CACHE_SHIFT - 6)	/* (64 == 1<<6) */
#else
#error BITS_PER_LONG must be 32 or 64
#endif
#define NBPPR		(PAGE_CACHE_SIZE/BITS_PER_LONG)
#define BTOPR(b)	(((unsigned int)(b) + (NBPPR - 1)) >> PRSHIFT)
#define BTOPRT(b)	(((unsigned int)(b) >> PRSHIFT))

STATIC unsigned long
page_region_mask(
	size_t		offset,
	size_t		length)
{
	unsigned long	mask;
	int		first, final;

	first = BTOPR(offset);
	final = BTOPRT(offset + length - 1);
	first = min(first, final);

	mask = ~0UL;
	mask <<= BITS_PER_LONG - (final - first);
	mask >>= BITS_PER_LONG - (final);

	ASSERT(offset + length <= PAGE_CACHE_SIZE);
	ASSERT((final - first) < BITS_PER_LONG && (final - first) >= 0);

	return mask;
}

STATIC inline void
set_page_region(
	struct page	*page,
	size_t		offset,
	size_t		length)
{
	set_page_private(page,
		page_private(page) | page_region_mask(offset, length));
	if (page_private(page) == ~0UL)
		SetPageUptodate(page);
}

STATIC inline int
test_page_region(
	struct page	*page,
	size_t		offset,
	size_t		length)
{
	unsigned long	mask = page_region_mask(offset, length);

	return (mask && (page_private(page) & mask) == mask);
}

/*
 * Mapping of multi-page buffers into contiguous virtual space
 */

typedef struct a_list {
	void		*vm_addr;
	struct a_list	*next;
} a_list_t;

STATIC a_list_t		*as_free_head;
STATIC int		as_list_len;
STATIC DEFINE_SPINLOCK(as_lock);

/*
 * Try to batch vunmaps because they are costly.
 */
STATIC void
free_address(
	void		*addr)
{
	a_list_t	*aentry;

	aentry = kmalloc(sizeof(a_list_t), GFP_ATOMIC & ~__GFP_HIGH);
	if (likely(aentry)) {
		spin_lock(&as_lock);
		aentry->next = as_free_head;
		aentry->vm_addr = addr;
		as_free_head = aentry;
		as_list_len++;
		spin_unlock(&as_lock);
	} else {
		vunmap(addr);
	}
}

STATIC void
purge_addresses(void)
{
	a_list_t	*aentry, *old;

	if (as_free_head == NULL)
		return;

	spin_lock(&as_lock);
	aentry = as_free_head;
	as_free_head = NULL;
	as_list_len = 0;
	spin_unlock(&as_lock);

	while ((old = aentry) != NULL) {
		vunmap(aentry->vm_addr);
		aentry = aentry->next;
		kfree(old);
	}
}

/*
 *	Internal pagebuf object manipulation
 */

STATIC void
_pagebuf_initialize(
	xfs_buf_t		*pb,
	xfs_buftarg_t		*target,
	loff_t			range_base,
	size_t			range_length,
	page_buf_flags_t	flags)
{
	/*
	 * We don't want certain flags to appear in pb->pb_flags.
	 */
	flags &= ~(PBF_LOCK|PBF_MAPPED|PBF_DONT_BLOCK|PBF_READ_AHEAD);

	memset(pb, 0, sizeof(xfs_buf_t));
	atomic_set(&pb->pb_hold, 1);
	init_MUTEX_LOCKED(&pb->pb_iodonesema);
	INIT_LIST_HEAD(&pb->pb_list);
	INIT_LIST_HEAD(&pb->pb_hash_list);
	init_MUTEX_LOCKED(&pb->pb_sema); /* held, no waiters */
	PB_SET_OWNER(pb);
	pb->pb_target = target;
	pb->pb_file_offset = range_base;
	/*
	 * Set buffer_length and count_desired to the same value initially.
	 * I/O routines should use count_desired, which will be the same in
	 * most cases but may be reset (e.g. XFS recovery).
	 */
	pb->pb_buffer_length = pb->pb_count_desired = range_length;
	pb->pb_flags = flags;
	pb->pb_bn = XFS_BUF_DADDR_NULL;
	atomic_set(&pb->pb_pin_count, 0);
	init_waitqueue_head(&pb->pb_waiters);

	XFS_STATS_INC(pb_create);
	PB_TRACE(pb, "initialize", target);
}

/*
 * Allocate a page array capable of holding a specified number
 * of pages, and point the page buf at it.
 */
STATIC int
_pagebuf_get_pages(
	xfs_buf_t		*pb,
	int			page_count,
	page_buf_flags_t	flags)
{
	/* Make sure that we have a page list */
	if (pb->pb_pages == NULL) {
		pb->pb_offset = page_buf_poff(pb->pb_file_offset);
		pb->pb_page_count = page_count;
		if (page_count <= PB_PAGES) {
			pb->pb_pages = pb->pb_page_array;
		} else {
			pb->pb_pages = kmem_alloc(sizeof(struct page *) *
					page_count, pb_to_km(flags));
			if (pb->pb_pages == NULL)
				return -ENOMEM;
		}
		memset(pb->pb_pages, 0, sizeof(struct page *) * page_count);
	}
	return 0;
}

/*
 *	Frees pb_pages if it was malloced.
 */
STATIC void
_pagebuf_free_pages(
	xfs_buf_t	*bp)
{
	if (bp->pb_pages != bp->pb_page_array) {
		kmem_free(bp->pb_pages,
			  bp->pb_page_count * sizeof(struct page *));
	}
}

/*
 *	Releases the specified buffer.
 *
 * 	The modification state of any associated pages is left unchanged.
 * 	The buffer most not be on any hash - use pagebuf_rele instead for
 * 	hashed and refcounted buffers
 */
void
pagebuf_free(
	xfs_buf_t		*bp)
{
	PB_TRACE(bp, "free", 0);

	ASSERT(list_empty(&bp->pb_hash_list));

	if (bp->pb_flags & _PBF_PAGE_CACHE) {
		uint		i;

		if ((bp->pb_flags & PBF_MAPPED) && (bp->pb_page_count > 1))
			free_address(bp->pb_addr - bp->pb_offset);

		for (i = 0; i < bp->pb_page_count; i++)
			page_cache_release(bp->pb_pages[i]);
		_pagebuf_free_pages(bp);
	} else if (bp->pb_flags & _PBF_KMEM_ALLOC) {
		 /*
		  * XXX(hch): bp->pb_count_desired might be incorrect (see
		  * pagebuf_associate_memory for details), but fortunately
		  * the Linux version of kmem_free ignores the len argument..
		  */
		kmem_free(bp->pb_addr, bp->pb_count_desired);
		_pagebuf_free_pages(bp);
	}

	pagebuf_deallocate(bp);
}

/*
 *	Finds all pages for buffer in question and builds it's page list.
 */
STATIC int
_pagebuf_lookup_pages(
	xfs_buf_t		*bp,
	uint			flags)
{
	struct address_space	*mapping = bp->pb_target->pbr_mapping;
	size_t			blocksize = bp->pb_target->pbr_bsize;
	size_t			size = bp->pb_count_desired;
	size_t			nbytes, offset;
	gfp_t			gfp_mask = pb_to_gfp(flags);
	unsigned short		page_count, i;
	pgoff_t			first;
	loff_t			end;
	int			error;

	end = bp->pb_file_offset + bp->pb_buffer_length;
	page_count = page_buf_btoc(end) - page_buf_btoct(bp->pb_file_offset);

	error = _pagebuf_get_pages(bp, page_count, flags);
	if (unlikely(error))
		return error;
	bp->pb_flags |= _PBF_PAGE_CACHE;

	offset = bp->pb_offset;
	first = bp->pb_file_offset >> PAGE_CACHE_SHIFT;

	for (i = 0; i < bp->pb_page_count; i++) {
		struct page	*page;
		uint		retries = 0;

	      retry:
		page = find_or_create_page(mapping, first + i, gfp_mask);
		if (unlikely(page == NULL)) {
			if (flags & PBF_READ_AHEAD) {
				bp->pb_page_count = i;
				for (i = 0; i < bp->pb_page_count; i++)
					unlock_page(bp->pb_pages[i]);
				return -ENOMEM;
			}

			/*
			 * This could deadlock.
			 *
			 * But until all the XFS lowlevel code is revamped to
			 * handle buffer allocation failures we can't do much.
			 */
			if (!(++retries % 100))
				printk(KERN_ERR
					"XFS: possible memory allocation "
					"deadlock in %s (mode:0x%x)\n",
					__FUNCTION__, gfp_mask);

			XFS_STATS_INC(pb_page_retries);
			xfsbufd_wakeup(0, gfp_mask);
			blk_congestion_wait(WRITE, HZ/50);
			goto retry;
		}

		XFS_STATS_INC(pb_page_found);

		nbytes = min_t(size_t, size, PAGE_CACHE_SIZE - offset);
		size -= nbytes;

		if (!PageUptodate(page)) {
			page_count--;
			if (blocksize >= PAGE_CACHE_SIZE) {
				if (flags & PBF_READ)
					bp->pb_locked = 1;
			} else if (!PagePrivate(page)) {
				if (test_page_region(page, offset, nbytes))
					page_count++;
			}
		}

		bp->pb_pages[i] = page;
		offset = 0;
	}

	if (!bp->pb_locked) {
		for (i = 0; i < bp->pb_page_count; i++)
			unlock_page(bp->pb_pages[i]);
	}

	if (page_count == bp->pb_page_count)
		bp->pb_flags |= PBF_DONE;

	PB_TRACE(bp, "lookup_pages", (long)page_count);
	return error;
}

/*
 *	Map buffer into kernel address-space if nessecary.
 */
STATIC int
_pagebuf_map_pages(
	xfs_buf_t		*bp,
	uint			flags)
{
	/* A single page buffer is always mappable */
	if (bp->pb_page_count == 1) {
		bp->pb_addr = page_address(bp->pb_pages[0]) + bp->pb_offset;
		bp->pb_flags |= PBF_MAPPED;
	} else if (flags & PBF_MAPPED) {
		if (as_list_len > 64)
			purge_addresses();
		bp->pb_addr = vmap(bp->pb_pages, bp->pb_page_count,
				VM_MAP, PAGE_KERNEL);
		if (unlikely(bp->pb_addr == NULL))
			return -ENOMEM;
		bp->pb_addr += bp->pb_offset;
		bp->pb_flags |= PBF_MAPPED;
	}

	return 0;
}

/*
 *	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(
	xfs_buftarg_t		*btp,	/* block device target		*/
	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;
	xfs_bufhash_t		*hash;
	xfs_buf_t		*pb, *n;

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

	/* Check for IOs smaller than the sector size / not sector aligned */
	ASSERT(!(range_length < (1 << btp->pbr_sshift)));
	ASSERT(!(range_base & (loff_t)btp->pbr_smask));

	hash = &btp->bt_hash[hash_long((unsigned long)ioff, btp->bt_hashshift)];

	spin_lock(&hash->bh_lock);

	list_for_each_entry_safe(pb, n, &hash->bh_list, pb_hash_list) {
		ASSERT(btp == pb->pb_target);
		if (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, &hash->bh_list);
			goto found;
		}
	}

	/* No match found */
	if (new_pb) {
		_pagebuf_initialize(new_pb, btp, range_base,
				range_length, flags);
		new_pb->pb_hash = hash;
		list_add(&new_pb->pb_hash_list, &hash->bh_list);
	} else {
		XFS_STATS_INC(pb_miss_locked);
	}

	spin_unlock(&hash->bh_lock);
	return new_pb;

found:
	spin_unlock(&hash->bh_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.
	 */
	if (down_trylock(&pb->pb_sema)) {
		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) {
		ASSERT((pb->pb_flags & _PBF_DELWRI_Q) == 0);
		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) {