#include "xfs_inum.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
-#include "xfs_dir.h"
#include "xfs_dir2.h"
#include "xfs_trans.h"
#include "xfs_dmapi.h"
#include "xfs_bmap_btree.h"
#include "xfs_alloc_btree.h"
#include "xfs_ialloc_btree.h"
-#include "xfs_dir_sf.h"
#include "xfs_dir2_sf.h"
#include "xfs_attr_sf.h"
#include "xfs_dinode.h"
#include "xfs_error.h"
#include "xfs_rw.h"
#include "xfs_iomap.h"
+#include "xfs_vnodeops.h"
#include <linux/mpage.h>
+#include <linux/pagevec.h>
#include <linux/writeback.h>
-STATIC void xfs_count_page_state(struct page *, int *, int *, int *);
-STATIC void xfs_convert_page(struct inode *, struct page *, xfs_iomap_t *,
- struct writeback_control *wbc, void *, int, int);
+
+/*
+ * Prime number of hash buckets since address is used as the key.
+ */
+#define NVSYNC 37
+#define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
+static wait_queue_head_t xfs_ioend_wq[NVSYNC];
+
+void __init
+xfs_ioend_init(void)
+{
+ int i;
+
+ for (i = 0; i < NVSYNC; i++)
+ init_waitqueue_head(&xfs_ioend_wq[i]);
+}
+
+void
+xfs_ioend_wait(
+ xfs_inode_t *ip)
+{
+ wait_queue_head_t *wq = to_ioend_wq(ip);
+
+ wait_event(*wq, (atomic_read(&ip->i_iocount) == 0));
+}
+
+STATIC void
+xfs_ioend_wake(
+ xfs_inode_t *ip)
+{
+ if (atomic_dec_and_test(&ip->i_iocount))
+ wake_up(to_ioend_wq(ip));
+}
+
+STATIC void
+xfs_count_page_state(
+ struct page *page,
+ int *delalloc,
+ int *unmapped,
+ int *unwritten)
+{
+ struct buffer_head *bh, *head;
+
+ *delalloc = *unmapped = *unwritten = 0;
+
+ bh = head = page_buffers(page);
+ do {
+ if (buffer_uptodate(bh) && !buffer_mapped(bh))
+ (*unmapped) = 1;
+ else if (buffer_unwritten(bh))
+ (*unwritten) = 1;
+ else if (buffer_delay(bh))
+ (*delalloc) = 1;
+ } while ((bh = bh->b_this_page) != head);
+}
#if defined(XFS_RW_TRACE)
void
int tag,
struct inode *inode,
struct page *page,
- int mask)
+ unsigned long pgoff)
{
xfs_inode_t *ip;
- bhv_desc_t *bdp;
- vnode_t *vp = LINVFS_GET_VP(inode);
loff_t isize = i_size_read(inode);
- loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
+ loff_t offset = page_offset(page);
int delalloc = -1, unmapped = -1, unwritten = -1;
if (page_has_buffers(page))
xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
- bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops);
- ip = XFS_BHVTOI(bdp);
+ ip = XFS_I(inode);
if (!ip->i_rwtrace)
return;
(void *)ip,
(void *)inode,
(void *)page,
- (void *)((unsigned long)mask),
+ (void *)pgoff,
(void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
(void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
(void *)((unsigned long)((isize >> 32) & 0xffffffff)),
(void *)((unsigned long)delalloc),
(void *)((unsigned long)unmapped),
(void *)((unsigned long)unwritten),
- (void *)NULL,
+ (void *)((unsigned long)current_pid()),
(void *)NULL);
}
#else
-#define xfs_page_trace(tag, inode, page, mask)
+#define xfs_page_trace(tag, inode, page, pgoff)
#endif
+STATIC struct block_device *
+xfs_find_bdev_for_inode(
+ struct xfs_inode *ip)
+{
+ struct xfs_mount *mp = ip->i_mount;
+
+ if (XFS_IS_REALTIME_INODE(ip))
+ return mp->m_rtdev_targp->bt_bdev;
+ else
+ return mp->m_ddev_targp->bt_bdev;
+}
+
/*
- * Schedule IO completion handling on a xfsdatad if this was
- * the final hold on this ioend.
+ * We're now finished for good with this ioend structure.
+ * Update the page state via the associated buffer_heads,
+ * release holds on the inode and bio, and finally free
+ * up memory. Do not use the ioend after this.
*/
STATIC void
-xfs_finish_ioend(
+xfs_destroy_ioend(
+ xfs_ioend_t *ioend)
+{
+ struct buffer_head *bh, *next;
+ struct xfs_inode *ip = XFS_I(ioend->io_inode);
+
+ for (bh = ioend->io_buffer_head; bh; bh = next) {
+ next = bh->b_private;
+ bh->b_end_io(bh, !ioend->io_error);
+ }
+
+ /*
+ * Volume managers supporting multiple paths can send back ENODEV
+ * when the final path disappears. In this case continuing to fill
+ * the page cache with dirty data which cannot be written out is
+ * evil, so prevent that.
+ */
+ if (unlikely(ioend->io_error == -ENODEV)) {
+ xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ,
+ __FILE__, __LINE__);
+ }
+
+ xfs_ioend_wake(ip);
+ mempool_free(ioend, xfs_ioend_pool);
+}
+
+/*
+ * If the end of the current ioend is beyond the current EOF,
+ * return the new EOF value, otherwise zero.
+ */
+STATIC xfs_fsize_t
+xfs_ioend_new_eof(
xfs_ioend_t *ioend)
{
- if (atomic_dec_and_test(&ioend->io_remaining))
- queue_work(xfsdatad_workqueue, &ioend->io_work);
+ xfs_inode_t *ip = XFS_I(ioend->io_inode);
+ xfs_fsize_t isize;
+ xfs_fsize_t bsize;
+
+ bsize = ioend->io_offset + ioend->io_size;
+ isize = MAX(ip->i_size, ip->i_new_size);
+ isize = MIN(isize, bsize);
+ return isize > ip->i_d.di_size ? isize : 0;
}
+/*
+ * Update on-disk file size now that data has been written to disk.
+ * The current in-memory file size is i_size. If a write is beyond
+ * eof i_new_size will be the intended file size until i_size is
+ * updated. If this write does not extend all the way to the valid
+ * file size then restrict this update to the end of the write.
+ */
+
STATIC void
-xfs_destroy_ioend(
+xfs_setfilesize(
xfs_ioend_t *ioend)
{
- vn_iowake(ioend->io_vnode);
- mempool_free(ioend, xfs_ioend_pool);
+ xfs_inode_t *ip = XFS_I(ioend->io_inode);
+ xfs_fsize_t isize;
+
+ ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
+ ASSERT(ioend->io_type != IOMAP_READ);
+
+ if (unlikely(ioend->io_error))
+ return;
+
+ xfs_ilock(ip, XFS_ILOCK_EXCL);
+ isize = xfs_ioend_new_eof(ioend);
+ if (isize) {
+ ip->i_d.di_size = isize;
+ xfs_mark_inode_dirty_sync(ip);
+ }
+
+ xfs_iunlock(ip, XFS_ILOCK_EXCL);
+}
+
+/*
+ * Buffered IO write completion for delayed allocate extents.
+ */
+STATIC void
+xfs_end_bio_delalloc(
+ struct work_struct *work)
+{
+ xfs_ioend_t *ioend =
+ container_of(work, xfs_ioend_t, io_work);
+
+ xfs_setfilesize(ioend);
+ xfs_destroy_ioend(ioend);
+}
+
+/*
+ * Buffered IO write completion for regular, written extents.
+ */
+STATIC void
+xfs_end_bio_written(
+ struct work_struct *work)
+{
+ xfs_ioend_t *ioend =
+ container_of(work, xfs_ioend_t, io_work);
+
+ xfs_setfilesize(ioend);
+ xfs_destroy_ioend(ioend);
}
/*
+ * IO write completion for unwritten extents.
+ *
* Issue transactions to convert a buffer range from unwritten
* to written extents.
*/
STATIC void
xfs_end_bio_unwritten(
- void *data)
+ struct work_struct *work)
{
- xfs_ioend_t *ioend = data;
- vnode_t *vp = ioend->io_vnode;
+ xfs_ioend_t *ioend =
+ container_of(work, xfs_ioend_t, io_work);
+ struct xfs_inode *ip = XFS_I(ioend->io_inode);
xfs_off_t offset = ioend->io_offset;
size_t size = ioend->io_size;
- struct buffer_head *bh, *next;
- int error;
-
- if (ioend->io_uptodate)
- VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
-
- /* ioend->io_buffer_head is only non-NULL for buffered I/O */
- for (bh = ioend->io_buffer_head; bh; bh = next) {
- next = bh->b_private;
- bh->b_end_io = NULL;
- clear_buffer_unwritten(bh);
- end_buffer_async_write(bh, ioend->io_uptodate);
+ if (likely(!ioend->io_error)) {
+ if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
+ int error;
+ error = xfs_iomap_write_unwritten(ip, offset, size);
+ if (error)
+ ioend->io_error = error;
+ }
+ xfs_setfilesize(ioend);
}
+ xfs_destroy_ioend(ioend);
+}
+
+/*
+ * IO read completion for regular, written extents.
+ */
+STATIC void
+xfs_end_bio_read(
+ struct work_struct *work)
+{
+ xfs_ioend_t *ioend =
+ container_of(work, xfs_ioend_t, io_work);
xfs_destroy_ioend(ioend);
}
/*
+ * Schedule IO completion handling on a xfsdatad if this was
+ * the final hold on this ioend. If we are asked to wait,
+ * flush the workqueue.
+ */
+STATIC void
+xfs_finish_ioend(
+ xfs_ioend_t *ioend,
+ int wait)
+{
+ if (atomic_dec_and_test(&ioend->io_remaining)) {
+ struct workqueue_struct *wq = xfsdatad_workqueue;
+ if (ioend->io_work.func == xfs_end_bio_unwritten)
+ wq = xfsconvertd_workqueue;
+
+ queue_work(wq, &ioend->io_work);
+ if (wait)
+ flush_workqueue(wq);
+ }
+}
+
+/*
* Allocate and initialise an IO completion structure.
* We need to track unwritten extent write completion here initially.
* We'll need to extend this for updating the ondisk inode size later
*/
STATIC xfs_ioend_t *
xfs_alloc_ioend(
- struct inode *inode)
+ struct inode *inode,
+ unsigned int type)
{
xfs_ioend_t *ioend;
* all the I/O from calling the completion routine too early.
*/
atomic_set(&ioend->io_remaining, 1);
- ioend->io_uptodate = 1; /* cleared if any I/O fails */
- ioend->io_vnode = LINVFS_GET_VP(inode);
+ ioend->io_error = 0;
+ ioend->io_list = NULL;
+ ioend->io_type = type;
+ ioend->io_inode = inode;
ioend->io_buffer_head = NULL;
- atomic_inc(&ioend->io_vnode->v_iocount);
+ ioend->io_buffer_tail = NULL;
+ atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
ioend->io_offset = 0;
ioend->io_size = 0;
- INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend);
+ if (type == IOMAP_UNWRITTEN)
+ INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
+ else if (type == IOMAP_DELAY)
+ INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
+ else if (type == IOMAP_READ)
+ INIT_WORK(&ioend->io_work, xfs_end_bio_read);
+ else
+ INIT_WORK(&ioend->io_work, xfs_end_bio_written);
return ioend;
}
-void
-linvfs_unwritten_done(
- struct buffer_head *bh,
- int uptodate)
-{
- xfs_ioend_t *ioend = bh->b_private;
- static spinlock_t unwritten_done_lock = SPIN_LOCK_UNLOCKED;
- unsigned long flags;
-
- ASSERT(buffer_unwritten(bh));
- bh->b_end_io = NULL;
-
- if (!uptodate)
- ioend->io_uptodate = 0;
-
- /*
- * Deep magic here. We reuse b_private in the buffer_heads to build
- * a chain for completing the I/O from user context after we've issued
- * a transaction to convert the unwritten extent.
- */
- spin_lock_irqsave(&unwritten_done_lock, flags);
- bh->b_private = ioend->io_buffer_head;
- ioend->io_buffer_head = bh;
- spin_unlock_irqrestore(&unwritten_done_lock, flags);
-
- xfs_finish_ioend(ioend);
-}
-
STATIC int
xfs_map_blocks(
struct inode *inode,
xfs_iomap_t *mapp,
int flags)
{
- vnode_t *vp = LINVFS_GET_VP(inode);
- int error, nmaps = 1;
+ int nmaps = 1;
+
+ return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps);
+}
- VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error);
- if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
- VMODIFY(vp);
- return -error;
+STATIC_INLINE int
+xfs_iomap_valid(
+ xfs_iomap_t *iomapp,
+ loff_t offset)
+{
+ return offset >= iomapp->iomap_offset &&
+ offset < iomapp->iomap_offset + iomapp->iomap_bsize;
}
/*
- * Finds the corresponding mapping in block @map array of the
- * given @offset within a @page.
+ * BIO completion handler for buffered IO.
*/
-STATIC xfs_iomap_t *
-xfs_offset_to_map(
- struct page *page,
- xfs_iomap_t *iomapp,
- unsigned long offset)
+STATIC void
+xfs_end_bio(
+ struct bio *bio,
+ int error)
{
- loff_t full_offset; /* offset from start of file */
+ xfs_ioend_t *ioend = bio->bi_private;
- ASSERT(offset < PAGE_CACHE_SIZE);
+ ASSERT(atomic_read(&bio->bi_cnt) >= 1);
+ ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
- full_offset = page->index; /* NB: using 64bit number */
- full_offset <<= PAGE_CACHE_SHIFT; /* offset from file start */
- full_offset += offset; /* offset from page start */
+ /* Toss bio and pass work off to an xfsdatad thread */
+ bio->bi_private = NULL;
+ bio->bi_end_io = NULL;
+ bio_put(bio);
- if (full_offset < iomapp->iomap_offset)
- return NULL;
- if (iomapp->iomap_offset + (iomapp->iomap_bsize -1) >= full_offset)
- return iomapp;
- return NULL;
+ xfs_finish_ioend(ioend, 0);
}
STATIC void
-xfs_map_at_offset(
- struct page *page,
- struct buffer_head *bh,
- unsigned long offset,
- int block_bits,
- xfs_iomap_t *iomapp)
+xfs_submit_ioend_bio(
+ xfs_ioend_t *ioend,
+ struct bio *bio)
{
- xfs_daddr_t bn;
- loff_t delta;
- int sector_shift;
+ atomic_inc(&ioend->io_remaining);
+ bio->bi_private = ioend;
+ bio->bi_end_io = xfs_end_bio;
- ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
- ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
- ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL);
+ /*
+ * If the I/O is beyond EOF we mark the inode dirty immediately
+ * but don't update the inode size until I/O completion.
+ */
+ if (xfs_ioend_new_eof(ioend))
+ xfs_mark_inode_dirty_sync(XFS_I(ioend->io_inode));
+
+ submit_bio(WRITE, bio);
+ ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
+ bio_put(bio);
+}
- delta = page->index;
- delta <<= PAGE_CACHE_SHIFT;
- delta += offset;
- delta -= iomapp->iomap_offset;
- delta >>= block_bits;
+STATIC struct bio *
+xfs_alloc_ioend_bio(
+ struct buffer_head *bh)
+{
+ struct bio *bio;
+ int nvecs = bio_get_nr_vecs(bh->b_bdev);
- sector_shift = block_bits - BBSHIFT;
- bn = iomapp->iomap_bn >> sector_shift;
- bn += delta;
- BUG_ON(!bn && !(iomapp->iomap_flags & IOMAP_REALTIME));
- ASSERT((bn << sector_shift) >= iomapp->iomap_bn);
+ do {
+ bio = bio_alloc(GFP_NOIO, nvecs);
+ nvecs >>= 1;
+ } while (!bio);
+
+ ASSERT(bio->bi_private == NULL);
+ bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
+ bio->bi_bdev = bh->b_bdev;
+ bio_get(bio);
+ return bio;
+}
- lock_buffer(bh);
- bh->b_blocknr = bn;
- bh->b_bdev = iomapp->iomap_target->pbr_bdev;
- set_buffer_mapped(bh);
- clear_buffer_delay(bh);
+STATIC void
+xfs_start_buffer_writeback(
+ struct buffer_head *bh)
+{
+ ASSERT(buffer_mapped(bh));
+ ASSERT(buffer_locked(bh));
+ ASSERT(!buffer_delay(bh));
+ ASSERT(!buffer_unwritten(bh));
+
+ mark_buffer_async_write(bh);
+ set_buffer_uptodate(bh);
+ clear_buffer_dirty(bh);
+}
+
+STATIC void
+xfs_start_page_writeback(
+ struct page *page,
+ int clear_dirty,
+ int buffers)
+{
+ ASSERT(PageLocked(page));
+ ASSERT(!PageWriteback(page));
+ if (clear_dirty)
+ clear_page_dirty_for_io(page);
+ set_page_writeback(page);
+ unlock_page(page);
+ /* If no buffers on the page are to be written, finish it here */
+ if (!buffers)
+ end_page_writeback(page);
+}
+
+static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
+{
+ return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
}
/*
- * Look for a page at index which is unlocked and contains our
- * unwritten extent flagged buffers at its head. Returns page
- * locked and with an extra reference count, and length of the
- * unwritten extent component on this page that we can write,
- * in units of filesystem blocks.
+ * Submit all of the bios for all of the ioends we have saved up, covering the
+ * initial writepage page and also any probed pages.
+ *
+ * Because we may have multiple ioends spanning a page, we need to start
+ * writeback on all the buffers before we submit them for I/O. If we mark the
+ * buffers as we got, then we can end up with a page that only has buffers
+ * marked async write and I/O complete on can occur before we mark the other
+ * buffers async write.
+ *
+ * The end result of this is that we trip a bug in end_page_writeback() because
+ * we call it twice for the one page as the code in end_buffer_async_write()
+ * assumes that all buffers on the page are started at the same time.
+ *
+ * The fix is two passes across the ioend list - one to start writeback on the
+ * buffer_heads, and then submit them for I/O on the second pass.
*/
-STATIC struct page *
-xfs_probe_unwritten_page(
- struct address_space *mapping,
- pgoff_t index,
- xfs_iomap_t *iomapp,
- xfs_ioend_t *ioend,
- unsigned long max_offset,
- unsigned long *fsbs,
- unsigned int bbits)
+STATIC void
+xfs_submit_ioend(
+ xfs_ioend_t *ioend)
{
- struct page *page;
+ xfs_ioend_t *head = ioend;
+ xfs_ioend_t *next;
+ struct buffer_head *bh;
+ struct bio *bio;
+ sector_t lastblock = 0;
- page = find_trylock_page(mapping, index);
- if (!page)
- return NULL;
- if (PageWriteback(page))
- goto out;
+ /* Pass 1 - start writeback */
+ do {
+ next = ioend->io_list;
+ for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
+ xfs_start_buffer_writeback(bh);
+ }
+ } while ((ioend = next) != NULL);
- if (page->mapping && page_has_buffers(page)) {
- struct buffer_head *bh, *head;
- unsigned long p_offset = 0;
+ /* Pass 2 - submit I/O */
+ ioend = head;
+ do {
+ next = ioend->io_list;
+ bio = NULL;
- *fsbs = 0;
- bh = head = page_buffers(page);
+ for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
+
+ if (!bio) {
+ retry:
+ bio = xfs_alloc_ioend_bio(bh);
+ } else if (bh->b_blocknr != lastblock + 1) {
+ xfs_submit_ioend_bio(ioend, bio);
+ goto retry;
+ }
+
+ if (bio_add_buffer(bio, bh) != bh->b_size) {
+ xfs_submit_ioend_bio(ioend, bio);
+ goto retry;
+ }
+
+ lastblock = bh->b_blocknr;
+ }
+ if (bio)
+ xfs_submit_ioend_bio(ioend, bio);
+ xfs_finish_ioend(ioend, 0);
+ } while ((ioend = next) != NULL);
+}
+
+/*
+ * Cancel submission of all buffer_heads so far in this endio.
+ * Toss the endio too. Only ever called for the initial page
+ * in a writepage request, so only ever one page.
+ */
+STATIC void
+xfs_cancel_ioend(
+ xfs_ioend_t *ioend)
+{
+ xfs_ioend_t *next;
+ struct buffer_head *bh, *next_bh;
+
+ do {
+ next = ioend->io_list;
+ bh = ioend->io_buffer_head;
do {
- if (!buffer_unwritten(bh) || !buffer_uptodate(bh))
- break;
- if (!xfs_offset_to_map(page, iomapp, p_offset))
- break;
- if (p_offset >= max_offset)
- break;
- xfs_map_at_offset(page, bh, p_offset, bbits, iomapp);
- set_buffer_unwritten_io(bh);
- bh->b_private = ioend;
- p_offset += bh->b_size;
- (*fsbs)++;
- } while ((bh = bh->b_this_page) != head);
+ next_bh = bh->b_private;
+ clear_buffer_async_write(bh);
+ unlock_buffer(bh);
+ } while ((bh = next_bh) != NULL);
+
+ xfs_ioend_wake(XFS_I(ioend->io_inode));
+ mempool_free(ioend, xfs_ioend_pool);
+ } while ((ioend = next) != NULL);
+}
+
+/*
+ * Test to see if we've been building up a completion structure for
+ * earlier buffers -- if so, we try to append to this ioend if we
+ * can, otherwise we finish off any current ioend and start another.
+ * Return true if we've finished the given ioend.
+ */
+STATIC void
+xfs_add_to_ioend(
+ struct inode *inode,
+ struct buffer_head *bh,
+ xfs_off_t offset,
+ unsigned int type,
+ xfs_ioend_t **result,
+ int need_ioend)
+{
+ xfs_ioend_t *ioend = *result;
+
+ if (!ioend || need_ioend || type != ioend->io_type) {
+ xfs_ioend_t *previous = *result;
- if (p_offset)
- return page;
+ ioend = xfs_alloc_ioend(inode, type);
+ ioend->io_offset = offset;
+ ioend->io_buffer_head = bh;
+ ioend->io_buffer_tail = bh;
+ if (previous)
+ previous->io_list = ioend;
+ *result = ioend;
+ } else {
+ ioend->io_buffer_tail->b_private = bh;
+ ioend->io_buffer_tail = bh;
}
-out:
- unlock_page(page);
- return NULL;
+ bh->b_private = NULL;
+ ioend->io_size += bh->b_size;
+}
+
+STATIC void
+xfs_map_buffer(
+ struct buffer_head *bh,
+ xfs_iomap_t *mp,
+ xfs_off_t offset,
+ uint block_bits)
+{
+ sector_t bn;
+
+ ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
+
+ bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
+ ((offset - mp->iomap_offset) >> block_bits);
+
+ ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
+
+ bh->b_blocknr = bn;
+ set_buffer_mapped(bh);
+}
+
+STATIC void
+xfs_map_at_offset(
+ struct buffer_head *bh,
+ loff_t offset,
+ int block_bits,
+ xfs_iomap_t *iomapp)
+{
+ ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
+ ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
+
+ lock_buffer(bh);
+ xfs_map_buffer(bh, iomapp, offset, block_bits);
+ bh->b_bdev = iomapp->iomap_target->bt_bdev;
+ set_buffer_mapped(bh);
+ clear_buffer_delay(bh);
+ clear_buffer_unwritten(bh);
}
/*
- * Look for a page at index which is unlocked and not mapped
- * yet - clustering for mmap write case.
+ * Look for a page at index that is suitable for clustering.
*/
STATIC unsigned int
-xfs_probe_unmapped_page(
- struct address_space *mapping,
- pgoff_t index,
- unsigned int pg_offset)
+xfs_probe_page(
+ struct page *page,
+ unsigned int pg_offset,
+ int mapped)
{
- struct page *page;
int ret = 0;
- page = find_trylock_page(mapping, index);
- if (!page)
- return 0;
if (PageWriteback(page))
- goto out;
+ return 0;
if (page->mapping && PageDirty(page)) {
if (page_has_buffers(page)) {
bh = head = page_buffers(page);
do {
- if (buffer_mapped(bh) || !buffer_uptodate(bh))
+ if (!buffer_uptodate(bh))
+ break;
+ if (mapped != buffer_mapped(bh))
break;
ret += bh->b_size;
if (ret >= pg_offset)
break;
} while ((bh = bh->b_this_page) != head);
} else
- ret = PAGE_CACHE_SIZE;
+ ret = mapped ? 0 : PAGE_CACHE_SIZE;
}
-out:
- unlock_page(page);
return ret;
}
-STATIC unsigned int
-xfs_probe_unmapped_cluster(
+STATIC size_t
+xfs_probe_cluster(
struct inode *inode,
struct page *startpage,
struct buffer_head *bh,
- struct buffer_head *head)
+ struct buffer_head *head,
+ int mapped)
{
+ struct pagevec pvec;
pgoff_t tindex, tlast, tloff;
- unsigned int pg_offset, len, total = 0;
- struct address_space *mapping = inode->i_mapping;
+ size_t total = 0;
+ int done = 0, i;
/* First sum forwards in this page */
do {
- if (buffer_mapped(bh))
- break;
+ if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
+ return total;
total += bh->b_size;
} while ((bh = bh->b_this_page) != head);
- /* If we reached the end of the page, sum forwards in
- * following pages.
- */
- if (bh == head) {
- tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
- /* Prune this back to avoid pathological behavior */
- tloff = min(tlast, startpage->index + 64);
- for (tindex = startpage->index + 1; tindex < tloff; tindex++) {
- len = xfs_probe_unmapped_page(mapping, tindex,
- PAGE_CACHE_SIZE);
- if (!len)
- return total;
- total += len;
- }
- if (tindex == tlast &&
- (pg_offset = i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
- total += xfs_probe_unmapped_page(mapping,
- tindex, pg_offset);
- }
- }
- return total;
-}
+ /* if we reached the end of the page, sum forwards in following pages */
+ tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
+ tindex = startpage->index + 1;
-/*
- * Probe for a given page (index) in the inode and test if it is delayed
- * and without unwritten buffers. Returns page locked and with an extra
- * reference count.
- */
-STATIC struct page *
-xfs_probe_delalloc_page(
- struct inode *inode,
- pgoff_t index)
-{
- struct page *page;
+ /* Prune this back to avoid pathological behavior */
+ tloff = min(tlast, startpage->index + 64);
- page = find_trylock_page(inode->i_mapping, index);
- if (!page)
- return NULL;
- if (PageWriteback(page))
- goto out;
-
- if (page->mapping && page_has_buffers(page)) {
- struct buffer_head *bh, *head;
- int acceptable = 0;
-
- bh = head = page_buffers(page);
- do {
- if (buffer_unwritten(bh)) {
- acceptable = 0;
- break;
- } else if (buffer_delay(bh)) {
- acceptable = 1;
- }
- } while ((bh = bh->b_this_page) != head);
-
- if (acceptable)
- return page;
- }
+ pagevec_init(&pvec, 0);
+ while (!done && tindex <= tloff) {
+ unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
-out:
- unlock_page(page);
- return NULL;
-}
-
-STATIC int
-xfs_map_unwritten(
- struct inode *inode,
- struct page *start_page,
- struct buffer_head *head,
- struct buffer_head *curr,
- unsigned long p_offset,
- int block_bits,
- xfs_iomap_t *iomapp,
- struct writeback_control *wbc,
- int startio,
- int all_bh)
-{
- struct buffer_head *bh = curr;
- xfs_iomap_t *tmp;
- xfs_ioend_t *ioend;
- loff_t offset;
- unsigned long nblocks = 0;
-
- offset = start_page->index;
- offset <<= PAGE_CACHE_SHIFT;
- offset += p_offset;
+ if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
+ break;
- ioend = xfs_alloc_ioend(inode);
+ for (i = 0; i < pagevec_count(&pvec); i++) {
+ struct page *page = pvec.pages[i];
+ size_t pg_offset, pg_len = 0;
- /* First map forwards in the page consecutive buffers
- * covering this unwritten extent
- */
- do {
- if (!buffer_unwritten(bh))
- break;
- tmp = xfs_offset_to_map(start_page, iomapp, p_offset);
- if (!tmp)
- break;
- xfs_map_at_offset(start_page, bh, p_offset, block_bits, iomapp);
- set_buffer_unwritten_io(bh);
- bh->b_private = ioend;
- p_offset += bh->b_size;
- nblocks++;
- } while ((bh = bh->b_this_page) != head);
+ if (tindex == tlast) {
+ pg_offset =
+ i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
+ if (!pg_offset) {
+ done = 1;
+ break;
+ }
+ } else
+ pg_offset = PAGE_CACHE_SIZE;
- atomic_add(nblocks, &ioend->io_remaining);
+ if (page->index == tindex && trylock_page(page)) {
+ pg_len = xfs_probe_page(page, pg_offset, mapped);
+ unlock_page(page);
+ }
- /* If we reached the end of the page, map forwards in any
- * following pages which are also covered by this extent.
- */
- if (bh == head) {
- struct address_space *mapping = inode->i_mapping;
- pgoff_t tindex, tloff, tlast;
- unsigned long bs;
- unsigned int pg_offset, bbits = inode->i_blkbits;
- struct page *page;
-
- tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
- tloff = (iomapp->iomap_offset + iomapp->iomap_bsize) >> PAGE_CACHE_SHIFT;
- tloff = min(tlast, tloff);
- for (tindex = start_page->index + 1; tindex < tloff; tindex++) {
- page = xfs_probe_unwritten_page(mapping,
- tindex, iomapp, ioend,
- PAGE_CACHE_SIZE, &bs, bbits);
- if (!page)
+ if (!pg_len) {
+ done = 1;
break;
- nblocks += bs;
- atomic_add(bs, &ioend->io_remaining);
- xfs_convert_page(inode, page, iomapp, wbc, ioend,
- startio, all_bh);
- /* stop if converting the next page might add
- * enough blocks that the corresponding byte
- * count won't fit in our ulong page buf length */
- if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
- goto enough;
- }
-
- if (tindex == tlast &&
- (pg_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1)))) {
- page = xfs_probe_unwritten_page(mapping,
- tindex, iomapp, ioend,
- pg_offset, &bs, bbits);
- if (page) {
- nblocks += bs;
- atomic_add(bs, &ioend->io_remaining);
- xfs_convert_page(inode, page, iomapp, wbc, ioend,
- startio, all_bh);
- if (nblocks >= ((ULONG_MAX - PAGE_SIZE) >> block_bits))
- goto enough;
}
+
+ total += pg_len;
+ tindex++;
}
+
+ pagevec_release(&pvec);
+ cond_resched();
}
-enough:
- ioend->io_size = (xfs_off_t)nblocks << block_bits;
- ioend->io_offset = offset;
- xfs_finish_ioend(ioend);
- return 0;
+ return total;
}
-STATIC void
-xfs_submit_page(
+/*
+ * Test if a given page is suitable for writing as part of an unwritten
+ * or delayed allocate extent.
+ */
+STATIC int
+xfs_is_delayed_page(
struct page *page,
- struct writeback_control *wbc,
- struct buffer_head *bh_arr[],
- int bh_count,
- int probed_page,
- int clear_dirty)
+ unsigned int type)
{
- struct buffer_head *bh;
- int i;
+ if (PageWriteback(page))
+ return 0;
- BUG_ON(PageWriteback(page));
- if (bh_count)
- set_page_writeback(page);
- if (clear_dirty)
- clear_page_dirty(page);
- unlock_page(page);
+ if (page->mapping && page_has_buffers(page)) {
+ struct buffer_head *bh, *head;
+ int acceptable = 0;
- if (bh_count) {
- for (i = 0; i < bh_count; i++) {
- bh = bh_arr[i];
- mark_buffer_async_write(bh);
+ bh = head = page_buffers(page);
+ do {
if (buffer_unwritten(bh))
- set_buffer_unwritten_io(bh);
- set_buffer_uptodate(bh);
- clear_buffer_dirty(bh);
- }
-
- for (i = 0; i < bh_count; i++)
- submit_bh(WRITE, bh_arr[i]);
+ acceptable = (type == IOMAP_UNWRITTEN);
+ else if (buffer_delay(bh))
+ acceptable = (type == IOMAP_DELAY);
+ else if (buffer_dirty(bh) && buffer_mapped(bh))
+ acceptable = (type == IOMAP_NEW);
+ else
+ break;
+ } while ((bh = bh->b_this_page) != head);
- if (probed_page && clear_dirty)
- wbc->nr_to_write--; /* Wrote an "extra" page */
+ if (acceptable)
+ return 1;
}
+
+ return 0;
}
/*
* delalloc/unwritten pages only, for the original page it is possible
* that the page has no mapping at all.
*/
-STATIC void
+STATIC int
xfs_convert_page(
struct inode *inode,
struct page *page,
- xfs_iomap_t *iomapp,
+ loff_t tindex,
+ xfs_iomap_t *mp,
+ xfs_ioend_t **ioendp,
struct writeback_control *wbc,
- void *private,
int startio,
int all_bh)
{
- struct buffer_head *bh_arr[MAX_BUF_PER_PAGE], *bh, *head;
- xfs_iomap_t *mp = iomapp, *tmp;
- unsigned long offset, end_offset;
- int index = 0;
+ struct buffer_head *bh, *head;
+ xfs_off_t end_offset;
+ unsigned long p_offset;
+ unsigned int type;
int bbits = inode->i_blkbits;
int len, page_dirty;
+ int count = 0, done = 0, uptodate = 1;
+ xfs_off_t offset = page_offset(page);
- end_offset = (i_size_read(inode) & (PAGE_CACHE_SIZE - 1));
+ if (page->index != tindex)
+ goto fail;
+ if (!trylock_page(page))
+ goto fail;
+ if (PageWriteback(page))
+ goto fail_unlock_page;
+ if (page->mapping != inode->i_mapping)
+ goto fail_unlock_page;
+ if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
+ goto fail_unlock_page;
/*
* page_dirty is initially a count of buffers on the page before
- * EOF and is decrememted as we move each into a cleanable state.
+ * EOF and is decremented as we move each into a cleanable state.
+ *
+ * Derivation:
+ *
+ * End offset is the highest offset that this page should represent.
+ * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
+ * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
+ * hence give us the correct page_dirty count. On any other page,
+ * it will be zero and in that case we need page_dirty to be the
+ * count of buffers on the page.
*/
+ end_offset = min_t(unsigned long long,
+ (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
+ i_size_read(inode));
+
len = 1 << inode->i_blkbits;
- end_offset = max(end_offset, PAGE_CACHE_SIZE);
- end_offset = roundup(end_offset, len);
- page_dirty = end_offset / len;
+ p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
+ PAGE_CACHE_SIZE);
+ p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
+ page_dirty = p_offset / len;
- offset = 0;
bh = head = page_buffers(page);
do {
if (offset >= end_offset)
break;
- if (!(PageUptodate(page) || buffer_uptodate(bh)))
+ if (!buffer_uptodate(bh))
+ uptodate = 0;
+ if (!(PageUptodate(page) || buffer_uptodate(bh))) {
+ done = 1;
continue;
- if (buffer_mapped(bh) && all_bh &&
- !(buffer_unwritten(bh) || buffer_delay(bh))) {
+ }
+
+ if (buffer_unwritten(bh) || buffer_delay(bh)) {
+ if (buffer_unwritten(bh))
+ type = IOMAP_UNWRITTEN;
+ else
+ type = IOMAP_DELAY;
+
+ if (!xfs_iomap_valid(mp, offset)) {
+ done = 1;
+ continue;
+ }
+
+ ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
+ ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
+
+ xfs_map_at_offset(bh, offset, bbits, mp);
if (startio) {
+ xfs_add_to_ioend(inode, bh, offset,
+ type, ioendp, done);
+ } else {
+ set_buffer_dirty(bh);
+ unlock_buffer(bh);
+ mark_buffer_dirty(bh);
+ }
+ page_dirty--;
+ count++;
+ } else {
+ type = IOMAP_NEW;
+ if (buffer_mapped(bh) && all_bh && startio) {
lock_buffer(bh);
- bh_arr[index++] = bh;
+ xfs_add_to_ioend(inode, bh, offset,
+ type, ioendp, done);
+ count++;
page_dirty--;
+ } else {
+ done = 1;
}
- continue;
}
- tmp = xfs_offset_to_map(page, mp, offset);
- if (!tmp)
- continue;
- ASSERT(!(tmp->iomap_flags & IOMAP_HOLE));
- ASSERT(!(tmp->iomap_flags & IOMAP_DELAY));
+ } while (offset += len, (bh = bh->b_this_page) != head);
- /* If this is a new unwritten extent buffer (i.e. one
- * that we haven't passed in private data for, we must
- * now map this buffer too.
- */
- if (buffer_unwritten(bh) && !bh->b_end_io) {
- ASSERT(tmp->iomap_flags & IOMAP_UNWRITTEN);
- xfs_map_unwritten(inode, page, head, bh, offset,
- bbits, tmp, wbc, startio, all_bh);
- } else if (! (buffer_unwritten(bh) && buffer_locked(bh))) {
- xfs_map_at_offset(page, bh, offset, bbits, tmp);
- if (buffer_unwritten(bh)) {
- set_buffer_unwritten_io(bh);
- bh->b_private = private;
- ASSERT(private);
+ if (uptodate && bh == head)
+ SetPageUptodate(page);
+
+ if (startio) {
+ if (count) {
+ struct backing_dev_info *bdi;
+
+ bdi = inode->i_mapping->backing_dev_info;
+ wbc->nr_to_write--;
+ if (bdi_write_congested(bdi)) {
+ wbc->encountered_congestion = 1;
+ done = 1;
+ } else if (wbc->nr_to_write <= 0) {
+ done = 1;
}
}
- if (startio) {
- bh_arr[index++] = bh;
- } else {
- set_buffer_dirty(bh);
- unlock_buffer(bh);
- mark_buffer_dirty(bh);
- }
- page_dirty--;
- } while (offset += len, (bh = bh->b_this_page) != head);
-
- if (startio && index) {
- xfs_submit_page(page, wbc, bh_arr, index, 1, !page_dirty);
- } else {
- unlock_page(page);
+ xfs_start_page_writeback(page, !page_dirty, count);
}
+
+ return done;
+ fail_unlock_page:
+ unlock_page(page);
+ fail:
+ return 1;
}
/*
struct inode *inode,
pgoff_t tindex,
xfs_iomap_t *iomapp,
+ xfs_ioend_t **ioendp,
struct writeback_control *wbc,
int startio,
int all_bh,
pgoff_t tlast)
{
- struct page *page;
+ struct pagevec pvec;
+ int done = 0, i;
+
+ pagevec_init(&pvec, 0);
+ while (!done && tindex <= tlast) {
+ unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
- for (; tindex <= tlast; tindex++) {
- page = xfs_probe_delalloc_page(inode, tindex);
- if (!page)
+ if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
break;
- xfs_convert_page(inode, page, iomapp, wbc, NULL,
- startio, all_bh);
+
+ for (i = 0; i < pagevec_count(&pvec); i++) {
+ done = xfs_convert_page(inode, pvec.pages[i], tindex++,
+ iomapp, ioendp, wbc, startio, all_bh);
+ if (done)
+ break;
+ }
+
+ pagevec_release(&pvec);
+ cond_resched();
}
}
* page if possible.
* The bh->b_state's cannot know if any of the blocks or which block for
* that matter are dirty due to mmap writes, and therefore bh uptodate is
- * only vaild if the page itself isn't completely uptodate. Some layers
+ * only valid if the page itself isn't completely uptodate. Some layers
* may clear the page dirty flag prior to calling write page, under the
* assumption the entire page will be written out; by not writing out the
* whole page the page can be reused before all valid dirty data is
int startio,
int unmapped) /* also implies page uptodate */
{
- struct buffer_head *bh_arr[MAX_BUF_PER_PAGE], *bh, *head;
- xfs_iomap_t *iomp, iomap;
+ struct buffer_head *bh, *head;
+ xfs_iomap_t iomap;
+ xfs_ioend_t *ioend = NULL, *iohead = NULL;
loff_t offset;
unsigned long p_offset = 0;
+ unsigned int type;
__uint64_t end_offset;
pgoff_t end_index, last_index, tlast;
- int len, err, i, cnt = 0, uptodate = 1;
- int flags;
- int page_dirty;
+ ssize_t size, len;
+ int flags, err, iomap_valid = 0, uptodate = 1;
+ int page_dirty, count = 0;
+ int trylock = 0;
+ int all_bh = unmapped;
- /* wait for other IO threads? */
- flags = (startio && wbc->sync_mode != WB_SYNC_NONE) ? 0 : BMAPI_TRYLOCK;
+ if (startio) {
+ if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
+ trylock |= BMAPI_TRYLOCK;
+ }
/* Is this page beyond the end of the file? */
offset = i_size_read(inode);
if (page->index >= end_index) {
if ((page->index >= end_index + 1) ||
!(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
- err = -EIO;
- goto error;
+ if (startio)
+ unlock_page(page);
+ return 0;
}
}
- end_offset = min_t(unsigned long long,
- (loff_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
- offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
-
/*
* page_dirty is initially a count of buffers on the page before
- * EOF and is decrememted as we move each into a cleanable state.
- */
+ * EOF and is decremented as we move each into a cleanable state.
+ *
+ * Derivation:
+ *
+ * End offset is the highest offset that this page should represent.
+ * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
+ * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
+ * hence give us the correct page_dirty count. On any other page,
+ * it will be zero and in that case we need page_dirty to be the
+ * count of buffers on the page.
+ */
+ end_offset = min_t(unsigned long long,
+ (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
len = 1 << inode->i_blkbits;
- p_offset = max(p_offset, PAGE_CACHE_SIZE);
- p_offset = roundup(p_offset, len);
+ p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
+ PAGE_CACHE_SIZE);
+ p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
page_dirty = p_offset / len;
- iomp = NULL;
- p_offset = 0;
bh = head = page_buffers(page);
+ offset = page_offset(page);
+ flags = BMAPI_READ;
+ type = IOMAP_NEW;
+
+ /* TODO: cleanup count and page_dirty */
do {
if (offset >= end_offset)
break;
if (!buffer_uptodate(bh))
uptodate = 0;
- if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio)
+ if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
+ /*
+ * the iomap is actually still valid, but the ioend
+ * isn't. shouldn't happen too often.
+ */
+ iomap_valid = 0;
continue;
-
- if (iomp) {
- iomp = xfs_offset_to_map(page, &iomap, p_offset);
}
+ if (iomap_valid)
+ iomap_valid = xfs_iomap_valid(&iomap, offset);
+
/*
* First case, map an unwritten extent and prepare for
* extent state conversion transaction on completion.
+ *
+ * Second case, allocate space for a delalloc buffer.
+ * We can return EAGAIN here in the release page case.
+ *
+ * Third case, an unmapped buffer was found, and we are
+ * in a path where we need to write the whole page out.
*/
- if (buffer_unwritten(bh)) {
- if (!startio)
- continue;
- if (!iomp) {
- err = xfs_map_blocks(inode, offset, len, &iomap,
- BMAPI_WRITE|BMAPI_IGNSTATE);
- if (err) {
- goto error;
- }
- iomp = xfs_offset_to_map(page, &iomap,
- p_offset);
+ if (buffer_unwritten(bh) || buffer_delay(bh) ||
+ ((buffer_uptodate(bh) || PageUptodate(page)) &&
+ !buffer_mapped(bh) && (unmapped || startio))) {
+ int new_ioend = 0;
+
+ /*
+ * Make sure we don't use a read-only iomap
+ */
+ if (flags == BMAPI_READ)
+ iomap_valid = 0;
+
+ if (buffer_unwritten(bh)) {
+ type = IOMAP_UNWRITTEN;
+ flags = BMAPI_WRITE | BMAPI_IGNSTATE;
+ } else if (buffer_delay(bh)) {
+ type = IOMAP_DELAY;
+ flags = BMAPI_ALLOCATE | trylock;
+ } else {
+ type = IOMAP_NEW;
+ flags = BMAPI_WRITE | BMAPI_MMAP;
}
- if (iomp) {
- if (!bh->b_end_io) {
- err = xfs_map_unwritten(inode, page,
- head, bh, p_offset,
- inode->i_blkbits, iomp,
- wbc, startio, unmapped);
- if (err) {
- goto error;
- }
+
+ if (!iomap_valid) {
+ /*
+ * if we didn't have a valid mapping then we
+ * need to ensure that we put the new mapping
+ * in a new ioend structure. This needs to be
+ * done to ensure that the ioends correctly
+ * reflect the block mappings at io completion
+ * for unwritten extent conversion.
+ */
+ new_ioend = 1;
+ if (type == IOMAP_NEW) {
+ size = xfs_probe_cluster(inode,
+ page, bh, head, 0);
} else {
- set_bit(BH_Lock, &bh->b_state);
+ size = len;
}
- BUG_ON(!buffer_locked(bh));
- bh_arr[cnt++] = bh;
- page_dirty--;
- }
- /*
- * Second case, allocate space for a delalloc buffer.
- * We can return EAGAIN here in the release page case.
- */
- } else if (buffer_delay(bh)) {
- if (!iomp) {
- err = xfs_map_blocks(inode, offset, len, &iomap,
- BMAPI_ALLOCATE | flags);
- if (err) {
+
+ err = xfs_map_blocks(inode, offset, size,
+ &iomap, flags);
+ if (err)
goto error;
- }
- iomp = xfs_offset_to_map(page, &iomap,
- p_offset);
+ iomap_valid = xfs_iomap_valid(&iomap, offset);
}
- if (iomp) {
- xfs_map_at_offset(page, bh, p_offset,
- inode->i_blkbits, iomp);
+ if (iomap_valid) {
+ xfs_map_at_offset(bh, offset,
+ inode->i_blkbits, &iomap);
if (startio) {
- bh_arr[cnt++] = bh;
+ xfs_add_to_ioend(inode, bh, offset,
+ type, &ioend,
+ new_ioend);
} else {
set_buffer_dirty(bh);
unlock_buffer(bh);
mark_buffer_dirty(bh);
}
page_dirty--;
+ count++;
+ }
+ } else if (buffer_uptodate(bh) && startio) {
+ /*
+ * we got here because the buffer is already mapped.
+ * That means it must already have extents allocated
+ * underneath it. Map the extent by reading it.
+ */
+ if (!iomap_valid || flags != BMAPI_READ) {
+ flags = BMAPI_READ;
+ size = xfs_probe_cluster(inode, page, bh,
+ head, 1);
+ err = xfs_map_blocks(inode, offset, size,
+ &iomap, flags);
+ if (err)
+ goto error;
+ iomap_valid = xfs_iomap_valid(&iomap, offset);
+ }
+
+ /*
+ * We set the type to IOMAP_NEW in case we are doing a
+ * small write at EOF that is extending the file but
+ * without needing an allocation. We need to update the
+ * file size on I/O completion in this case so it is
+ * the same case as having just allocated a new extent
+ * that we are writing into for the first time.
+ */
+ type = IOMAP_NEW;
+ if (trylock_buffer(bh)) {
+ ASSERT(buffer_mapped(bh));
+ if (iomap_valid)
+ all_bh = 1;
+ xfs_add_to_ioend(inode, bh, offset, type,
+ &ioend, !iomap_valid);
+ page_dirty--;
+ count++;
+ } else {
+ iomap_valid = 0;
}
} else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
(unmapped || startio)) {
+ iomap_valid = 0;
+ }
- if (!buffer_mapped(bh)) {
- int size;
+ if (!iohead)
+ iohead = ioend;
- /*
- * Getting here implies an unmapped buffer
- * was found, and we are in a path where we
- * need to write the whole page out.
- */
- if (!iomp) {
- size = xfs_probe_unmapped_cluster(
- inode, page, bh, head);
- err = xfs_map_blocks(inode, offset,
- size, &iomap,
- BMAPI_WRITE|BMAPI_MMAP);
- if (err) {
- goto error;
- }
- iomp = xfs_offset_to_map(page, &iomap,
- p_offset);
- }
- if (iomp) {
- xfs_map_at_offset(page,
- bh, p_offset,
- inode->i_blkbits, iomp);
- if (startio) {
- bh_arr[cnt++] = bh;
- } else {
- set_buffer_dirty(bh);
- unlock_buffer(bh);
- mark_buffer_dirty(bh);
- }
- page_dirty--;
- }
- } else if (startio) {
- if (buffer_uptodate(bh) &&
- !test_and_set_bit(BH_Lock, &bh->b_state)) {
- bh_arr[cnt++] = bh;
- page_dirty--;
- }
- }
- }
- } while (offset += len, p_offset += len,
- ((bh = bh->b_this_page) != head));
+ } while (offset += len, ((bh = bh->b_this_page) != head));
if (uptodate && bh == head)
SetPageUptodate(page);
- if (startio) {
- xfs_submit_page(page, wbc, bh_arr, cnt, 0, !page_dirty);
- }
+ if (startio)
+ xfs_start_page_writeback(page, 1, count);
- if (iomp) {
- offset = (iomp->iomap_offset + iomp->iomap_bsize - 1) >>
+ if (ioend && iomap_valid) {
+ offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
PAGE_CACHE_SHIFT;
tlast = min_t(pgoff_t, offset, last_index);
- xfs_cluster_write(inode, page->index + 1, iomp, wbc,
- startio, unmapped, tlast);
+ xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
+ wbc, startio, all_bh, tlast);
}
+ if (iohead)
+ xfs_submit_ioend(iohead);
+
return page_dirty;
error:
- for (i = 0; i < cnt; i++) {
- unlock_buffer(bh_arr[i]);
- }
+ if (iohead)
+ xfs_cancel_ioend(iohead);
/*
* If it's delalloc and we have nowhere to put it,
* us to try again.
*/
if (err != -EAGAIN) {
- if (!unmapped) {
+ if (!unmapped)
block_invalidatepage(page, 0);
- }
ClearPageUptodate(page);
}
return err;
}
-STATIC int
-__linvfs_get_block(
- struct inode *inode,
- sector_t iblock,
- unsigned long blocks,
- struct buffer_head *bh_result,
- int create,
- int direct,
- bmapi_flags_t flags)
-{
- vnode_t *vp = LINVFS_GET_VP(inode);
- xfs_iomap_t iomap;
- xfs_off_t offset;
- ssize_t size;
- int retpbbm = 1;
- int error;
-
- if (blocks) {
- offset = blocks << inode->i_blkbits; /* 64 bit goodness */
- size = (ssize_t) min_t(xfs_off_t, offset, LONG_MAX);
- } else {
- size = 1 << inode->i_blkbits;
- }
- offset = (xfs_off_t)iblock << inode->i_blkbits;
-
- VOP_BMAP(vp, offset, size,
- create ? flags : BMAPI_READ, &iomap, &retpbbm, error);
- if (error)
- return -error;
-
- if (retpbbm == 0)
- return 0;
-
- if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
- xfs_daddr_t bn;
- xfs_off_t delta;
-
- /* For unwritten extents do not report a disk address on
- * the read case (treat as if we're reading into a hole).
- */
- if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
- delta = offset - iomap.iomap_offset;
- delta >>= inode->i_blkbits;
-
- bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT);
- bn += delta;
- BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME));
- bh_result->b_blocknr = bn;
- set_buffer_mapped(bh_result);
- }
- if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
- if (direct)
- bh_result->b_private = inode;
- set_buffer_unwritten(bh_result);
- set_buffer_delay(bh_result);
- }
- }
-
- /* If this is a realtime file, data might be on a new device */
- bh_result->b_bdev = iomap.iomap_target->pbr_bdev;
-
- /* If we previously allocated a block out beyond eof and
- * we are now coming back to use it then we will need to
- * flag it as new even if it has a disk address.
- */
- if (create &&
- ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
- (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW)))
- set_buffer_new(bh_result);
-
- if (iomap.iomap_flags & IOMAP_DELAY) {
- BUG_ON(direct);
- if (create) {
- set_buffer_uptodate(bh_result);
- set_buffer_mapped(bh_result);
- set_buffer_delay(bh_result);
- }
- }
-
- if (blocks) {
- ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
- offset = min_t(xfs_off_t,
- iomap.iomap_bsize - iomap.iomap_delta,
- blocks << inode->i_blkbits);
- bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset);
- }
-
- return 0;
-}
-
-int
-linvfs_get_block(
- struct inode *inode,
- sector_t iblock,
- struct buffer_head *bh_result,
- int create)
-{
- return __linvfs_get_block(inode, iblock, 0, bh_result,
- create, 0, BMAPI_WRITE);
-}
-
-STATIC int
-linvfs_get_blocks_direct(
- struct inode *inode,
- sector_t iblock,
- unsigned long max_blocks,
- struct buffer_head *bh_result,
- int create)
-{
- return __linvfs_get_block(inode, iblock, max_blocks, bh_result,
- create, 1, BMAPI_WRITE|BMAPI_DIRECT);
-}
-
-STATIC void
-linvfs_end_io_direct(
- struct kiocb *iocb,
- loff_t offset,
- ssize_t size,
- void *private)
-{
- xfs_ioend_t *ioend = iocb->private;
-
- /*
- * Non-NULL private data means we need to issue a transaction to
- * convert a range from unwritten to written extents. This needs
- * to happen from process contect but aio+dio I/O completion
- * happens from irq context so we need to defer it to a workqueue.
- * This is not nessecary for synchronous direct I/O, but we do
- * it anyway to keep the code uniform and simpler.
- *
- * The core direct I/O code might be changed to always call the
- * completion handler in the future, in which case all this can
- * go away.
- */
- if (private && size > 0) {
- ioend->io_offset = offset;
- ioend->io_size = size;
- xfs_finish_ioend(ioend);
- } else {
- ASSERT(size >= 0);
- xfs_destroy_ioend(ioend);
- }
-
- /*
- * blockdev_direct_IO can return an error even afer the I/O
- * completion handler was called. Thus we need to protect
- * against double-freeing.
- */
- iocb->private = NULL;
-}
-
-STATIC ssize_t
-linvfs_direct_IO(
- int rw,
- struct kiocb *iocb,
- const struct iovec *iov,
- loff_t offset,
- unsigned long nr_segs)
-{
- struct file *file = iocb->ki_filp;
- struct inode *inode = file->f_mapping->host;
- vnode_t *vp = LINVFS_GET_VP(inode);
- xfs_iomap_t iomap;
- int maps = 1;
- int error;
- ssize_t ret;
-
- VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error);
- if (error)
- return -error;
-
- iocb->private = xfs_alloc_ioend(inode);
-
- ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
- iomap.iomap_target->pbr_bdev,
- iov, offset, nr_segs,
- linvfs_get_blocks_direct,
- linvfs_end_io_direct);
-
- if (unlikely(ret <= 0 && iocb->private))
- xfs_destroy_ioend(iocb->private);
- return ret;
-}
-
-
-STATIC sector_t
-linvfs_bmap(
- struct address_space *mapping,
- sector_t block)
-{
- struct inode *inode = (struct inode *)mapping->host;
- vnode_t *vp = LINVFS_GET_VP(inode);
- int error;
-
- vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address);
-
- VOP_RWLOCK(vp, VRWLOCK_READ);
- VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
- VOP_RWUNLOCK(vp, VRWLOCK_READ);
- return generic_block_bmap(mapping, block, linvfs_get_block);
-}
-
-STATIC int
-linvfs_readpage(
- struct file *unused,
- struct page *page)
-{
- return mpage_readpage(page, linvfs_get_block);
-}
-
-STATIC int
-linvfs_readpages(
- struct file *unused,
- struct address_space *mapping,
- struct list_head *pages,
- unsigned nr_pages)
-{
- return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block);
-}
-
-STATIC void
-xfs_count_page_state(
- struct page *page,
- int *delalloc,
- int *unmapped,
- int *unwritten)
-{
- struct buffer_head *bh, *head;
-
- *delalloc = *unmapped = *unwritten = 0;
-
- bh = head = page_buffers(page);
- do {
- if (buffer_uptodate(bh) && !buffer_mapped(bh))
- (*unmapped) = 1;
- else if (buffer_unwritten(bh) && !buffer_delay(bh))
- clear_buffer_unwritten(bh);
- else if (buffer_unwritten(bh))
- (*unwritten) = 1;
- else if (buffer_delay(bh))
- (*delalloc) = 1;
- } while ((bh = bh->b_this_page) != head);
-}
-
-
/*
* writepage: Called from one of two places:
*
*/
STATIC int
-linvfs_writepage(
+xfs_vm_writepage(
struct page *page,
struct writeback_control *wbc)
{
* then mark the page dirty again and leave the page
* as is.
*/
- if (PFLAGS_TEST_FSTRANS() && need_trans)
+ if (current_test_flags(PF_FSTRANS) && need_trans)
goto out_fail;
/*
if (!page_has_buffers(page))
create_empty_buffers(page, 1 << inode->i_blkbits, 0);
+
+ /*
+ * VM calculation for nr_to_write seems off. Bump it way
+ * up, this gets simple streaming writes zippy again.
+ * To be reviewed again after Jens' writeback changes.
+ */
+ wbc->nr_to_write *= 4;
+
/*
* Convert delayed allocate, unwritten or unmapped space
* to real space and flush out to disk.
}
STATIC int
-linvfs_invalidate_page(
- struct page *page,
- unsigned long offset)
+xfs_vm_writepages(
+ struct address_space *mapping,
+ struct writeback_control *wbc)
{
- xfs_page_trace(XFS_INVALIDPAGE_ENTER,
- page->mapping->host, page, offset);
- return block_invalidatepage(page, offset);
+ xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
+ return generic_writepages(mapping, wbc);
}
/*
* free them and we should come back later via writepage.
*/
STATIC int
-linvfs_release_page(
+xfs_vm_releasepage(
struct page *page,
gfp_t gfp_mask)
{
.nr_to_write = 1,
};
- xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask);
+ xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
+
+ if (!page_has_buffers(page))
+ return 0;
xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
if (!delalloc && !unwritten)
/* If we are already inside a transaction or the thread cannot
* do I/O, we cannot release this page.
*/
- if (PFLAGS_TEST_FSTRANS())
+ if (current_test_flags(PF_FSTRANS))
return 0;
/*
}
STATIC int
-linvfs_prepare_write(
+__xfs_get_blocks(
+ struct inode *inode,
+ sector_t iblock,
+ struct buffer_head *bh_result,
+ int create,
+ int direct,
+ bmapi_flags_t flags)
+{
+ xfs_iomap_t iomap;
+ xfs_off_t offset;
+ ssize_t size;
+ int niomap = 1;
+ int error;
+
+ offset = (xfs_off_t)iblock << inode->i_blkbits;
+ ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
+ size = bh_result->b_size;
+
+ if (!create && direct && offset >= i_size_read(inode))
+ return 0;
+
+ error = xfs_iomap(XFS_I(inode), offset, size,
+ create ? flags : BMAPI_READ, &iomap, &niomap);
+ if (error)
+ return -error;
+ if (niomap == 0)
+ return 0;
+
+ if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
+ /*
+ * For unwritten extents do not report a disk address on
+ * the read case (treat as if we're reading into a hole).
+ */
+ if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
+ xfs_map_buffer(bh_result, &iomap, offset,
+ inode->i_blkbits);
+ }
+ if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
+ if (direct)
+ bh_result->b_private = inode;
+ set_buffer_unwritten(bh_result);
+ }
+ }
+
+ /*
+ * If this is a realtime file, data may be on a different device.
+ * to that pointed to from the buffer_head b_bdev currently.
+ */
+ bh_result->b_bdev = iomap.iomap_target->bt_bdev;
+
+ /*
+ * If we previously allocated a block out beyond eof and we are now
+ * coming back to use it then we will need to flag it as new even if it
+ * has a disk address.
+ *
+ * With sub-block writes into unwritten extents we also need to mark
+ * the buffer as new so that the unwritten parts of the buffer gets
+ * correctly zeroed.
+ */
+ if (create &&
+ ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
+ (offset >= i_size_read(inode)) ||
+ (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
+ set_buffer_new(bh_result);
+
+ if (iomap.iomap_flags & IOMAP_DELAY) {
+ BUG_ON(direct);
+ if (create) {
+ set_buffer_uptodate(bh_result);
+ set_buffer_mapped(bh_result);
+ set_buffer_delay(bh_result);
+ }
+ }
+
+ if (direct || size > (1 << inode->i_blkbits)) {
+ ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
+ offset = min_t(xfs_off_t,
+ iomap.iomap_bsize - iomap.iomap_delta, size);
+ bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
+ }
+
+ return 0;
+}
+
+int
+xfs_get_blocks(
+ struct inode *inode,
+ sector_t iblock,
+ struct buffer_head *bh_result,
+ int create)
+{
+ return __xfs_get_blocks(inode, iblock,
+ bh_result, create, 0, BMAPI_WRITE);
+}
+
+STATIC int
+xfs_get_blocks_direct(
+ struct inode *inode,
+ sector_t iblock,
+ struct buffer_head *bh_result,
+ int create)
+{
+ return __xfs_get_blocks(inode, iblock,
+ bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
+}
+
+STATIC void
+xfs_end_io_direct(
+ struct kiocb *iocb,
+ loff_t offset,
+ ssize_t size,
+ void *private)
+{
+ xfs_ioend_t *ioend = iocb->private;
+
+ /*
+ * Non-NULL private data means we need to issue a transaction to
+ * convert a range from unwritten to written extents. This needs
+ * to happen from process context but aio+dio I/O completion
+ * happens from irq context so we need to defer it to a workqueue.
+ * This is not necessary for synchronous direct I/O, but we do
+ * it anyway to keep the code uniform and simpler.
+ *
+ * Well, if only it were that simple. Because synchronous direct I/O
+ * requires extent conversion to occur *before* we return to userspace,
+ * we have to wait for extent conversion to complete. Look at the
+ * iocb that has been passed to us to determine if this is AIO or
+ * not. If it is synchronous, tell xfs_finish_ioend() to kick the
+ * workqueue and wait for it to complete.
+ *
+ * The core direct I/O code might be changed to always call the
+ * completion handler in the future, in which case all this can
+ * go away.
+ */
+ ioend->io_offset = offset;
+ ioend->io_size = size;
+ if (ioend->io_type == IOMAP_READ) {
+ xfs_finish_ioend(ioend, 0);
+ } else if (private && size > 0) {
+ xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
+ } else {
+ /*
+ * A direct I/O write ioend starts it's life in unwritten
+ * state in case they map an unwritten extent. This write
+ * didn't map an unwritten extent so switch it's completion
+ * handler.
+ */
+ INIT_WORK(&ioend->io_work, xfs_end_bio_written);
+ xfs_finish_ioend(ioend, 0);
+ }
+
+ /*
+ * blockdev_direct_IO can return an error even after the I/O
+ * completion handler was called. Thus we need to protect
+ * against double-freeing.
+ */
+ iocb->private = NULL;
+}
+
+STATIC ssize_t
+xfs_vm_direct_IO(
+ int rw,
+ struct kiocb *iocb,
+ const struct iovec *iov,
+ loff_t offset,
+ unsigned long nr_segs)
+{
+ struct file *file = iocb->ki_filp;
+ struct inode *inode = file->f_mapping->host;
+ struct block_device *bdev;
+ ssize_t ret;
+
+ bdev = xfs_find_bdev_for_inode(XFS_I(inode));
+
+ if (rw == WRITE) {
+ iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
+ ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
+ bdev, iov, offset, nr_segs,
+ xfs_get_blocks_direct,
+ xfs_end_io_direct);
+ } else {
+ iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
+ ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
+ bdev, iov, offset, nr_segs,
+ xfs_get_blocks_direct,
+ xfs_end_io_direct);
+ }
+
+ if (unlikely(ret != -EIOCBQUEUED && iocb->private))
+ xfs_destroy_ioend(iocb->private);
+ return ret;
+}
+
+STATIC int
+xfs_vm_write_begin(
struct file *file,
+ struct address_space *mapping,
+ loff_t pos,
+ unsigned len,
+ unsigned flags,
+ struct page **pagep,
+ void **fsdata)
+{
+ *pagep = NULL;
+ return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
+ xfs_get_blocks);
+}
+
+STATIC sector_t
+xfs_vm_bmap(
+ struct address_space *mapping,
+ sector_t block)
+{
+ struct inode *inode = (struct inode *)mapping->host;
+ struct xfs_inode *ip = XFS_I(inode);
+
+ xfs_itrace_entry(XFS_I(inode));
+ xfs_ilock(ip, XFS_IOLOCK_SHARED);
+ xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
+ xfs_iunlock(ip, XFS_IOLOCK_SHARED);
+ return generic_block_bmap(mapping, block, xfs_get_blocks);
+}
+
+STATIC int
+xfs_vm_readpage(
+ struct file *unused,
+ struct page *page)
+{
+ return mpage_readpage(page, xfs_get_blocks);
+}
+
+STATIC int
+xfs_vm_readpages(
+ struct file *unused,
+ struct address_space *mapping,
+ struct list_head *pages,
+ unsigned nr_pages)
+{
+ return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
+}
+
+STATIC void
+xfs_vm_invalidatepage(
struct page *page,
- unsigned int from,
- unsigned int to)
+ unsigned long offset)
{
- return block_prepare_write(page, from, to, linvfs_get_block);
+ xfs_page_trace(XFS_INVALIDPAGE_ENTER,
+ page->mapping->host, page, offset);
+ block_invalidatepage(page, offset);
}
-struct address_space_operations linvfs_aops = {
- .readpage = linvfs_readpage,
- .readpages = linvfs_readpages,
- .writepage = linvfs_writepage,
+const struct address_space_operations xfs_address_space_operations = {
+ .readpage = xfs_vm_readpage,
+ .readpages = xfs_vm_readpages,
+ .writepage = xfs_vm_writepage,
+ .writepages = xfs_vm_writepages,
.sync_page = block_sync_page,
- .releasepage = linvfs_release_page,
- .invalidatepage = linvfs_invalidate_page,
- .prepare_write = linvfs_prepare_write,
- .commit_write = generic_commit_write,
- .bmap = linvfs_bmap,
- .direct_IO = linvfs_direct_IO,
+ .releasepage = xfs_vm_releasepage,
+ .invalidatepage = xfs_vm_invalidatepage,
+ .write_begin = xfs_vm_write_begin,
+ .write_end = generic_write_end,
+ .bmap = xfs_vm_bmap,
+ .direct_IO = xfs_vm_direct_IO,
+ .migratepage = buffer_migrate_page,
+ .is_partially_uptodate = block_is_partially_uptodate,
+ .error_remove_page = generic_error_remove_page,
};
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff