pnfs/blocklayout: rewrite extent tracking

Currently the block layout driver tracks extents in three separate
data structures:

 - the two list of pnfs_block_extent structures returned by the server
 - the list of sectors that were in invalid state but have been written to
 - a list of pnfs_block_short_extent structures for LAYOUTCOMMIT

All of these share the property that they are not only highly inefficient
data structures, but also that operations on them are even more inefficient
than nessecary.

In addition there are various implementation defects like:

 - using an int to track sectors, causing corruption for large offsets
 - incorrect normalization of page or block granularity ranges
 - insufficient error handling
 - incorrect synchronization as extents can be modified while they are in
   use

This patch replace all three data with a single unified rbtree structure
tracking all extents, as well as their in-memory state, although we still
need to instance for read-only and read-write extent due to the arcane
client side COW feature in the block layouts spec.

To fix the problem of extent possibly being modified while in use we make
sure to return a copy of the extent for use in the write path - the
extent can only be invalidated by a layout recall or return which has
to wait until the I/O operations finished due to refcounts on the layout
segment.

The new extent tree work similar to the schemes used by block based
filesystems like XFS or ext4.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

6 files changed, 651 insertions, 1212 deletions

diff --git a/fs/nfs/blocklayout/Makefile b/fs/nfs/blocklayout/Makefile
index d5815505c020..3fa5ec780a8e 100644
--- a/fs/nfs/blocklayout/Makefile
+++ b/fs/nfs/blocklayout/Makefile
@@ -2,4 +2,5 @@
 # Makefile for the pNFS block layout driver kernel module
 #
 obj-$(CONFIG_PNFS_BLOCK) += blocklayoutdriver.o
-blocklayoutdriver-objs := blocklayout.o extents.o blocklayoutdev.o blocklayoutdm.o
+blocklayoutdriver-objs := blocklayout.o blocklayoutdev.o blocklayoutdm.o \
+	extent_tree.o
diff --git a/fs/nfs/blocklayout/blocklayout.c b/fs/nfs/blocklayout/blocklayout.c
index 5aa23750a149..8502e620f644 100644
--- a/fs/nfs/blocklayout/blocklayout.c
+++ b/fs/nfs/blocklayout/blocklayout.c
@@ -49,26 +49,16 @@ MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
 
-/* Given the be associated with isect, determine if page data needs to be
- * initialized.
- */
-static int is_hole(struct pnfs_block_extent *be, sector_t isect)
-{
-	if (be->be_state == PNFS_BLOCK_NONE_DATA)
-		return 1;
-	else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
-		return 0;
-	else
-		return !bl_is_sector_init(be->be_inval, isect);
-}
-
-/* Given the be associated with isect, determine if page data can be
- * written to disk.
- */
-static int is_writable(struct pnfs_block_extent *be, sector_t isect)
+static bool is_hole(struct pnfs_block_extent *be)
 {
-	return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
-		be->be_state == PNFS_BLOCK_INVALID_DATA);
+	switch (be->be_state) {
+	case PNFS_BLOCK_NONE_DATA:
+		return true;
+	case PNFS_BLOCK_INVALID_DATA:
+		return be->be_tag ? false : true;
+	default:
+		return false;
+	}
 }
 
 /* The data we are handed might be spread across several bios.  We need
@@ -76,9 +66,8 @@ static int is_writable(struct pnfs_block_extent *be, sector_t isect)
  */
 struct parallel_io {
 	struct kref refcnt;
-	void (*pnfs_callback) (void *data, int num_se);
+	void (*pnfs_callback) (void *data);
 	void *data;
-	int bse_count;
 };
 
 static inline struct parallel_io *alloc_parallel(void *data)
@@ -89,7 +78,6 @@ static inline struct parallel_io *alloc_parallel(void *data)
 	if (rv) {
 		rv->data = data;
 		kref_init(&rv->refcnt);
-		rv->bse_count = 0;
 	}
 	return rv;
 }
@@ -104,7 +92,7 @@ static void destroy_parallel(struct kref *kref)
 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
 
 	dprintk("%s enter\n", __func__);
-	p->pnfs_callback(p->data, p->bse_count);
+	p->pnfs_callback(p->data);
 	kfree(p);
 }
 
@@ -200,7 +188,7 @@ static void bl_read_cleanup(struct work_struct *work)
 }
 
 static void
-bl_end_par_io_read(void *data, int unused)
+bl_end_par_io_read(void *data)
 {
 	struct nfs_pgio_header *hdr = data;
 
@@ -210,56 +198,46 @@ bl_end_par_io_read(void *data, int unused)
 }
 
 static enum pnfs_try_status
-bl_read_pagelist(struct nfs_pgio_header *hdr)
+bl_read_pagelist(struct nfs_pgio_header *header)
 {
-	struct nfs_pgio_header *header = hdr;
-	int i, hole;
+	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
 	struct bio *bio = NULL;
-	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
+	struct pnfs_block_extent be;
 	sector_t isect, extent_length = 0;
 	struct parallel_io *par;
-	loff_t f_offset = hdr->args.offset;
-	size_t bytes_left = hdr->args.count;
+	loff_t f_offset = header->args.offset;
+	size_t bytes_left = header->args.count;
 	unsigned int pg_offset, pg_len;
-	struct page **pages = hdr->args.pages;
-	int pg_index = hdr->args.pgbase >> PAGE_CACHE_SHIFT;
+	struct page **pages = header->args.pages;
+	int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
 	const bool is_dio = (header->dreq != NULL);
 	struct blk_plug plug;
+	int i;
 
 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
-		hdr->page_array.npages, f_offset,
-		(unsigned int)hdr->args.count);
+		header->page_array.npages, f_offset,
+		(unsigned int)header->args.count);
 
-	par = alloc_parallel(hdr);
+	par = alloc_parallel(header);
 	if (!par)
-		goto use_mds;
+		return PNFS_NOT_ATTEMPTED;
 	par->pnfs_callback = bl_end_par_io_read;
-	/* At this point, we can no longer jump to use_mds */
 
 	blk_start_plug(&plug);
 
 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
 	/* Code assumes extents are page-aligned */
-	for (i = pg_index; i < hdr->page_array.npages; i++) {
+	for (i = pg_index; i < header->page_array.npages; i++) {
 		if (extent_length <= 0) {
 			/* We've used up the previous extent */
-			bl_put_extent(be);
-			bl_put_extent(cow_read);
 			bio = bl_submit_bio(READ, bio);
+
 			/* Get the next one */
-			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
-					     isect, &cow_read);
-			if (!be) {
+			if (!ext_tree_lookup(bl, isect, &be, false)) {
 				header->pnfs_error = -EIO;
 				goto out;
 			}
-			extent_length = be->be_length -
-				(isect - be->be_f_offset);
-			if (cow_read) {
-				sector_t cow_length = cow_read->be_length -
-					(isect - cow_read->be_f_offset);
-				extent_length = min(extent_length, cow_length);
-			}
+			extent_length = be.be_length - (isect - be.be_f_offset);
 		}
 
 		pg_offset = f_offset & ~PAGE_CACHE_MASK;
@@ -278,20 +256,16 @@ bl_read_pagelist(struct nfs_pgio_header *hdr)
 			pg_len = PAGE_CACHE_SIZE;
 		}
 
-		hole = is_hole(be, isect);
-		if (hole && !cow_read) {
+		if (is_hole(&be)) {
 			bio = bl_submit_bio(READ, bio);
 			/* Fill hole w/ zeroes w/o accessing device */
 			dprintk("%s Zeroing page for hole\n", __func__);
 			zero_user_segment(pages[i], pg_offset, pg_len);
 		} else {
-			struct pnfs_block_extent *be_read;
-
-			be_read = (hole && cow_read) ? cow_read : be;
 			bio = do_add_page_to_bio(bio,
-						 hdr->page_array.npages - i,
+						 header->page_array.npages - i,
 						 READ,
-						 isect, pages[i], be_read,
+						 isect, pages[i], &be,
 						 bl_end_io_read, par,
 						 pg_offset, pg_len);
 			if (IS_ERR(bio)) {
@@ -304,50 +278,16 @@ bl_read_pagelist(struct nfs_pgio_header *hdr)
 		extent_length -= (pg_len >> SECTOR_SHIFT);
 	}
 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
-		hdr->res.eof = 1;
-		hdr->res.count = header->inode->i_size - hdr->args.offset;
+		header->res.eof = 1;
+		header->res.count = header->inode->i_size - header->args.offset;
 	} else {
-		hdr->res.count = (isect << SECTOR_SHIFT) - hdr->args.offset;
+		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
 	}
 out:
-	bl_put_extent(be);
-	bl_put_extent(cow_read);
 	bl_submit_bio(READ, bio);
 	blk_finish_plug(&plug);
 	put_parallel(par);
 	return PNFS_ATTEMPTED;
-
- use_mds:
-	dprintk("Giving up and using normal NFS\n");
-	return PNFS_NOT_ATTEMPTED;
-}
-
-static void mark_extents_written(struct pnfs_block_layout *bl,
-				 __u64 offset, __u32 count)
-{
-	sector_t isect, end;
-	struct pnfs_block_extent *be;
-	struct pnfs_block_short_extent *se;
-
-	dprintk("%s(%llu, %u)\n", __func__, offset, count);
-	if (count == 0)
-		return;
-	isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
-	end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
-	end >>= SECTOR_SHIFT;
-	while (isect < end) {
-		sector_t len;
-		be = bl_find_get_extent(bl, isect, NULL);
-		BUG_ON(!be); /* FIXME */
-		len = min(end, be->be_f_offset + be->be_length) - isect;
-		if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
-			se = bl_pop_one_short_extent(be->be_inval);
-			BUG_ON(!se);
-			bl_mark_for_commit(be, isect, len, se);
-		}
-		isect += len;
-		bl_put_extent(be);
-	}
 }
 
 static void bl_end_io_write(struct bio *bio, int err)
@@ -370,29 +310,30 @@ static void bl_end_io_write(struct bio *bio, int err)
  */
 static void bl_write_cleanup(struct work_struct *work)
 {
-	struct rpc_task *task;
-	struct nfs_pgio_header *hdr;
+	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
+	struct nfs_pgio_header *hdr =
+			container_of(task, struct nfs_pgio_header, task);
+
 	dprintk("%s enter\n", __func__);
-	task = container_of(work, struct rpc_task, u.tk_work);
-	hdr = container_of(task, struct nfs_pgio_header, task);
+
 	if (likely(!hdr->pnfs_error)) {
-		/* Marks for LAYOUTCOMMIT */
-		mark_extents_written(BLK_LSEG2EXT(hdr->lseg),
-				     hdr->args.offset, hdr->args.count);
+		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
+		u64 start = hdr->args.offset & (loff_t)PAGE_CACHE_MASK;
+		u64 end = (hdr->args.offset + hdr->args.count +
+			PAGE_CACHE_SIZE - 1) & (loff_t)PAGE_CACHE_MASK;
+
+		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
+					(end - start) >> SECTOR_SHIFT);
 	}
+
 	pnfs_ld_write_done(hdr);
 }
 
 /* Called when last of bios associated with a bl_write_pagelist call finishes */
-static void bl_end_par_io_write(void *data, int num_se)
+static void bl_end_par_io_write(void *data)
 {
 	struct nfs_pgio_header *hdr = data;
 
-	if (unlikely(hdr->pnfs_error)) {
-		bl_free_short_extents(&BLK_LSEG2EXT(hdr->lseg)->bl_inval,
-					num_se);
-	}
-
 	hdr->task.tk_status = hdr->pnfs_error;
 	hdr->verf.committed = NFS_FILE_SYNC;
 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
@@ -402,9 +343,9 @@ static void bl_end_par_io_write(void *data, int num_se)
 static enum pnfs_try_status
 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 {
-	int i, ret;
+	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
 	struct bio *bio = NULL;
-	struct pnfs_block_extent *be = NULL;
+	struct pnfs_block_extent be;
 	sector_t isect, extent_length = 0;
 	struct parallel_io *par = NULL;
 	loff_t offset = header->args.offset;
@@ -412,6 +353,7 @@ bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 	struct page **pages = header->args.pages;
 	int pg_index = pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
 	struct blk_plug plug;
+	int i;
 
 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
 
@@ -421,9 +363,8 @@ bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 	 */
 	par = alloc_parallel(header);
 	if (!par)
-		goto out_mds;
+		return PNFS_NOT_ATTEMPTED;
 	par->pnfs_callback = bl_end_par_io_write;
-	/* At this point, have to be more careful with error handling */
 
 	blk_start_plug(&plug);
 
@@ -434,44 +375,18 @@ bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 	for (i = pg_index; i < header->page_array.npages; i++) {
 		if (extent_length <= 0) {
 			/* We've used up the previous extent */
-			bl_put_extent(be);
 			bio = bl_submit_bio(WRITE, bio);
 			/* Get the next one */
-			be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
-					     isect, NULL);
-			if (!be || !is_writable(be, isect)) {
+			if (!ext_tree_lookup(bl, isect, &be, true)) {
 				header->pnfs_error = -EINVAL;
 				goto out;
 			}
-			if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
-				if (likely(!bl_push_one_short_extent(
-								be->be_inval)))
-					par->bse_count++;
-				else {
-					header->pnfs_error = -ENOMEM;
-					goto out;
-				}
-			}
-			extent_length = be->be_length -
-			    (isect - be->be_f_offset);
-		}
 
-		BUG_ON(offset & ~PAGE_CACHE_MASK);
-
-		if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
-		    !bl_is_sector_init(be->be_inval, isect)) {
-			ret = bl_mark_sectors_init(be->be_inval, isect,
-						       PAGE_CACHE_SECTORS);
-			if (unlikely(ret)) {
-				dprintk("%s bl_mark_sectors_init fail %d\n",
-					__func__, ret);
-				header->pnfs_error = ret;
-				goto out;
-			}
+			extent_length = be.be_length - (isect - be.be_f_offset);
 		}
 
 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
-					 WRITE, isect, pages[i], be,
+					 WRITE, isect, pages[i], &be,
 					 bl_end_io_write, par,
 					 0, PAGE_CACHE_SIZE);
 		if (IS_ERR(bio)) {
@@ -487,60 +402,22 @@ bl_write_pagelist(struct nfs_pgio_header *header, int sync)
 
 	header->res.count = header->args.count;
 out:
-	bl_put_extent(be);
 	bl_submit_bio(WRITE, bio);
 	blk_finish_plug(&plug);
 	put_parallel(par);
 	return PNFS_ATTEMPTED;
-out_mds:
-	return PNFS_NOT_ATTEMPTED;
-}
-
-/* FIXME - range ignored */
-static void
-release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
-{
-	int i;
-	struct pnfs_block_extent *be;
-
-	spin_lock(&bl->bl_ext_lock);
-	for (i = 0; i < EXTENT_LISTS; i++) {
-		while (!list_empty(&bl->bl_extents[i])) {
-			be = list_first_entry(&bl->bl_extents[i],
-					      struct pnfs_block_extent,
-					      be_node);
-			list_del(&be->be_node);
-			bl_put_extent(be);
-		}
-	}
-	spin_unlock(&bl->bl_ext_lock);
-}
-
-static void
-release_inval_marks(struct pnfs_inval_markings *marks)
-{
-	struct pnfs_inval_tracking *pos, *temp;
-	struct pnfs_block_short_extent *se, *stemp;
-
-	list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
-		list_del(&pos->it_link);
-		kfree(pos);
-	}
-
-	list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
-		list_del(&se->bse_node);
-		kfree(se);
-	}
-	return;
 }
 
 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
 {
 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
+	int err;
 
 	dprintk("%s enter\n", __func__);
-	release_extents(bl, NULL);
-	release_inval_marks(&bl->bl_inval);
+
+	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
+	WARN_ON(err);
+
 	kfree(bl);
 }
 
@@ -553,14 +430,11 @@ static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
 	bl = kzalloc(sizeof(*bl), gfp_flags);
 	if (!bl)
 		return NULL;
+
+	bl->bl_ext_rw = RB_ROOT;
+	bl->bl_ext_ro = RB_ROOT;
 	spin_lock_init(&bl->bl_ext_lock);
-	INIT_LIST_HEAD(&bl->bl_extents[0]);
-	INIT_LIST_HEAD(&bl->bl_extents[1]);
-	INIT_LIST_HEAD(&bl->bl_commit);
-	INIT_LIST_HEAD(&bl->bl_committing);
-	bl->bl_count = 0;
-	bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
-	BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
+
 	return &bl->bl_layout;
 }
 
@@ -600,7 +474,7 @@ bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
 		       const struct nfs4_layoutcommit_args *arg)
 {
 	dprintk("%s enter\n", __func__);
-	encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
+	ext_tree_encode_commit(BLK_LO2EXT(lo), xdr);
 }
 
 static void
@@ -609,7 +483,7 @@ bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
 	struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
 
 	dprintk("%s enter\n", __func__);
-	clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
+	ext_tree_mark_committed(BLK_LO2EXT(lo), lcdata->res.status);
 }
 
 static void free_blk_mountid(struct block_mount_id *mid)
diff --git a/fs/nfs/blocklayout/blocklayout.h b/fs/nfs/blocklayout/blocklayout.h
index 9838fb020473..b4f66d875f12 100644
--- a/fs/nfs/blocklayout/blocklayout.h
+++ b/fs/nfs/blocklayout/blocklayout.h
@@ -63,82 +63,28 @@ enum exstate4 {
 	PNFS_BLOCK_NONE_DATA		= 3  /* unmapped, it's a hole */
 };
 
-#define MY_MAX_TAGS (15) /* tag bitnums used must be less than this */
-
-struct my_tree {
-	sector_t		mtt_step_size;	/* Internal sector alignment */
-	struct list_head	mtt_stub; /* Should be a radix tree */
-};
-
-struct pnfs_inval_markings {
-	spinlock_t	im_lock;
-	struct my_tree	im_tree;	/* Sectors that need LAYOUTCOMMIT */
-	sector_t	im_block_size;	/* Server blocksize in sectors */
-	struct list_head im_extents;	/* Short extents for INVAL->RW conversion */
-};
-
-struct pnfs_inval_tracking {
-	struct list_head it_link;
-	int		 it_sector;
-	int		 it_tags;
-};
-
 /* sector_t fields are all in 512-byte sectors */
 struct pnfs_block_extent {
-	struct kref	be_refcnt;
-	struct list_head be_node;	/* link into lseg list */
-	struct nfs4_deviceid be_devid;  /* FIXME: could use device cache instead */
+	union {
+		struct rb_node	be_node;
+		struct list_head be_list;
+	};
+	struct nfs4_deviceid be_devid;	/* FIXME: could use device cache instead */
 	struct block_device *be_mdev;
 	sector_t	be_f_offset;	/* the starting offset in the file */
 	sector_t	be_length;	/* the size of the extent */
 	sector_t	be_v_offset;	/* the starting offset in the volume */
 	enum exstate4	be_state;	/* the state of this extent */
-	struct pnfs_inval_markings *be_inval; /* tracks INVAL->RW transition */
-};
-
-/* Shortened extent used by LAYOUTCOMMIT */
-struct pnfs_block_short_extent {
-	struct list_head bse_node;
-	struct nfs4_deviceid bse_devid;
-	struct block_device *bse_mdev;
-	sector_t	bse_f_offset;	/* the starting offset in the file */
-	sector_t	bse_length;	/* the size of the extent */
+#define EXTENT_WRITTEN		1
+#define EXTENT_COMMITTING	2
+	unsigned int	be_tag;
 };
 
-static inline void
-BL_INIT_INVAL_MARKS(struct pnfs_inval_markings *marks, sector_t blocksize)
-{
-	spin_lock_init(&marks->im_lock);
-	INIT_LIST_HEAD(&marks->im_tree.mtt_stub);
-	INIT_LIST_HEAD(&marks->im_extents);
-	marks->im_block_size = blocksize;
-	marks->im_tree.mtt_step_size = min((sector_t)PAGE_CACHE_SECTORS,
-					   blocksize);
-}
-
-enum extentclass4 {
-	RW_EXTENT       = 0, /* READWRTE and INVAL */
-	RO_EXTENT       = 1, /* READ and NONE */
-	EXTENT_LISTS    = 2,
-};
-
-static inline int bl_choose_list(enum exstate4 state)
-{
-	if (state == PNFS_BLOCK_READ_DATA || state == PNFS_BLOCK_NONE_DATA)
-		return RO_EXTENT;
-	else
-		return RW_EXTENT;
-}
-
 struct pnfs_block_layout {
-	struct pnfs_layout_hdr bl_layout;
-	struct pnfs_inval_markings bl_inval; /* tracks INVAL->RW transition */
+	struct pnfs_layout_hdr	bl_layout;
+	struct rb_root		bl_ext_rw;
+	struct rb_root		bl_ext_ro;
 	spinlock_t		bl_ext_lock;   /* Protects list manipulation */
-	struct list_head	bl_extents[EXTENT_LISTS]; /* R and RW extents */
-	struct list_head	bl_commit;	/* Needs layout commit */
-	struct list_head	bl_committing;	/* Layout committing */
-	unsigned int		bl_count;	/* entries in bl_commit */
-	sector_t		bl_blocksize;  /* Server blocksize in sectors */
 };
 
 #define BLK_ID(lo) ((struct block_mount_id *)(NFS_SERVER(lo->plh_inode)->pnfs_ld_data))
@@ -183,29 +129,17 @@ int nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
 /* blocklayoutdm.c */
 void bl_free_block_dev(struct pnfs_block_dev *bdev);
 
-/* extents.c */
-struct pnfs_block_extent *
-bl_find_get_extent(struct pnfs_block_layout *bl, sector_t isect,
-		struct pnfs_block_extent **cow_read);
-int bl_mark_sectors_init(struct pnfs_inval_markings *marks,
-			     sector_t offset, sector_t length);
-void bl_put_extent(struct pnfs_block_extent *be);
-struct pnfs_block_extent *bl_alloc_extent(void);
-int bl_is_sector_init(struct pnfs_inval_markings *marks, sector_t isect);
-int encode_pnfs_block_layoutupdate(struct pnfs_block_layout *bl,
-				   struct xdr_stream *xdr,
-				   const struct nfs4_layoutcommit_args *arg);
-void clean_pnfs_block_layoutupdate(struct pnfs_block_layout *bl,
-				   const struct nfs4_layoutcommit_args *arg,
-				   int status);
-int bl_add_merge_extent(struct pnfs_block_layout *bl,
-			 struct pnfs_block_extent *new);
-int bl_mark_for_commit(struct pnfs_block_extent *be,
-			sector_t offset, sector_t length,
-			struct pnfs_block_short_extent *new);
-int bl_push_one_short_extent(struct pnfs_inval_markings *marks);
-struct pnfs_block_short_extent *
-bl_pop_one_short_extent(struct pnfs_inval_markings *marks);
-void bl_free_short_extents(struct pnfs_inval_markings *marks, int num_to_free);
+/* extent_tree.c */
+int ext_tree_insert(struct pnfs_block_layout *bl,
+		struct pnfs_block_extent *new);
+int ext_tree_remove(struct pnfs_block_layout *bl, bool rw, sector_t start,
+		sector_t end);
+int ext_tree_mark_written(struct pnfs_block_layout *bl, sector_t start,
+		sector_t len);
+bool ext_tree_lookup(struct pnfs_block_layout *bl, sector_t isect,
+		struct pnfs_block_extent *ret, bool rw);
+int ext_tree_encode_commit(struct pnfs_block_layout *bl,
+		struct xdr_stream *xdr);
+void ext_tree_mark_committed(struct pnfs_block_layout *bl, int status);
 
 #endif /* FS_NFS_NFS4BLOCKLAYOUT_H */
diff --git a/fs/nfs/blocklayout/blocklayoutdev.c b/fs/nfs/blocklayout/blocklayoutdev.c
index 63f77925aa87..cd71b5e231ec 100644
--- a/fs/nfs/blocklayout/blocklayoutdev.c
+++ b/fs/nfs/blocklayout/blocklayoutdev.c
@@ -309,7 +309,7 @@ nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
 	 * recovery easier.
 	 */
 	for (i = 0; i < count; i++) {
-		be = bl_alloc_extent();
+		be = kzalloc(sizeof(struct pnfs_block_extent), GFP_NOFS);
 		if (!be) {
 			status = -ENOMEM;
 			goto out_err;
@@ -330,13 +330,11 @@ nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
 		if (decode_sector_number(&p, &be->be_v_offset) < 0)
 			goto out_err;
 		be->be_state = be32_to_cpup(p++);
-		if (be->be_state == PNFS_BLOCK_INVALID_DATA)
-			be->be_inval = &bl->bl_inval;
 		if (verify_extent(be, &lv)) {
 			dprintk("%s verify failed\n", __func__);
 			goto out_err;
 		}
-		list_add_tail(&be->be_node, &extents);
+		list_add_tail(&be->be_list, &extents);
 	}
 	if (lgr->range.offset + lgr->range.length !=
 			lv.start << SECTOR_SHIFT) {
@@ -352,21 +350,13 @@ nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
 	/* Extents decoded properly, now try to merge them in to
 	 * existing layout extents.
 	 */
-	spin_lock(&bl->bl_ext_lock);
-	list_for_each_entry_safe(be, save, &extents, be_node) {
-		list_del(&be->be_node);
-		status = bl_add_merge_extent(bl, be);
-		if (status) {
-			spin_unlock(&bl->bl_ext_lock);
-			/* This is a fairly catastrophic error, as the
-			 * entire layout extent lists are now corrupted.
-			 * We should have some way to distinguish this.
-			 */
-			be = NULL;
-			goto out_err;
-		}
+	list_for_each_entry_safe(be, save, &extents, be_list) {
+		list_del(&be->be_list);
+
+		status = ext_tree_insert(bl, be);
+		if (status)
+			goto out_free_list;
 	}
-	spin_unlock(&bl->bl_ext_lock);
 	status = 0;
  out:
 	__free_page(scratch);
@@ -374,12 +364,13 @@ nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
 	return status;
 
  out_err:
-	bl_put_extent(be);
+	kfree(be);
+ out_free_list:
 	while (!list_empty(&extents)) {
 		be = list_first_entry(&extents, struct pnfs_block_extent,
-				      be_node);
-		list_del(&be->be_node);
-		bl_put_extent(be);
+				      be_list);
+		list_del(&be->be_list);
+		kfree(be);
 	}
 	goto out;
 }
diff --git a/fs/nfs/blocklayout/extent_tree.c b/fs/nfs/blocklayout/extent_tree.c
new file mode 100644
index 000000000000..c8c59a5b1a8f
--- /dev/null
+++ b/fs/nfs/blocklayout/extent_tree.c
@@ -0,0 +1,547 @@
+/*
+ * Copyright (c) 2014 Christoph Hellwig.
+ */
+
+#include "blocklayout.h"
+
+#define NFSDBG_FACILITY		NFSDBG_PNFS_LD
+
+static inline struct pnfs_block_extent *
+ext_node(struct rb_node *node)
+{
+	return rb_entry(node, struct pnfs_block_extent, be_node);
+}
+
+static struct pnfs_block_extent *
+ext_tree_first(struct rb_root *root)
+{
+	struct rb_node *node = rb_first(root);
+	return node ? ext_node(node) : NULL;
+}
+
+static struct pnfs_block_extent *
+ext_tree_prev(struct pnfs_block_extent *be)
+{
+	struct rb_node *node = rb_prev(&be->be_node);
+	return node ? ext_node(node) : NULL;
+}
+
+static struct pnfs_block_extent *
+ext_tree_next(struct pnfs_block_extent *be)
+{
+	struct rb_node *node = rb_next(&be->be_node);
+	return node ? ext_node(node) : NULL;
+}
+
+static inline sector_t
+ext_f_end(struct pnfs_block_extent *be)
+{
+	return be->be_f_offset + be->be_length;
+}
+
+static struct pnfs_block_extent *
+__ext_tree_search(struct rb_root *root, sector_t start)
+{
+	struct rb_node *node = root->rb_node;
+	struct pnfs_block_extent *be = NULL;
+
+	while (node) {
+		be = ext_node(node);
+		if (start < be->be_f_offset)
+			node = node->rb_left;
+		else if (start >= ext_f_end(be))
+			node = node->rb_right;
+		else
+			return be;
+	}
+
+	if (be) {
+		if (start < be->be_f_offset)
+			return be;
+
+		if (start >= ext_f_end(be))
+			return ext_tree_next(be);
+	}
+
+	return NULL;
+}
+
+static bool
+ext_can_merge(struct pnfs_block_extent *be1, struct pnfs_block_extent *be2)
+{
+	if (be1->be_state != be2->be_state)
+		return false;
+	if (be1->be_mdev != be2->be_mdev)
+		return false;
+
+	if (be1->be_f_offset + be1->be_length != be2->be_f_offset)
+		return false;
+
+	if (be1->be_state != PNFS_BLOCK_NONE_DATA &&
+	    (be1->be_v_offset + be1->be_length != be2->be_v_offset))
+		return false;
+
+	if (be1->be_state == PNFS_BLOCK_INVALID_DATA &&
+	    be1->be_tag != be2->be_tag)
+		return false;
+
+	return true;
+}
+
+static struct pnfs_block_extent *
+ext_try_to_merge_left(struct rb_root *root, struct pnfs_block_extent *be)
+{
+	struct pnfs_block_extent *left = ext_tree_prev(be);
+
+	if (left && ext_can_merge(left, be)) {
+		left->be_length += be->be_length;
+		rb_erase(&be->be_node, root);
+		kfree(be);
+		return left;
+	}
+
+	return be;
+}
+
+static struct pnfs_block_extent *
+ext_try_to_merge_right(struct rb_root *root, struct pnfs_block_extent *be)
+{
+	struct pnfs_block_extent *right = ext_tree_next(be);
+
+	if (right && ext_can_merge(be, right)) {
+		be->be_length += right->be_length;
+		rb_erase(&right->be_node, root);
+		kfree(right);
+	}
+
+	return be;
+}
+
+static void
+__ext_tree_insert(struct rb_root *root,
+		struct pnfs_block_extent *new, bool merge_ok)
+{
+	struct rb_node **p = &root->rb_node, *parent = NULL;
+	struct pnfs_block_extent *be;
+
+	while (*p) {
+		parent = *p;
+		be = ext_node(parent);
+
+		if (new->be_f_offset < be->be_f_offset) {
+			if (merge_ok && ext_can_merge(new, be)) {
+				be->be_f_offset = new->be_f_offset;
+				if (be->be_state != PNFS_BLOCK_NONE_DATA)
+					be->be_v_offset = new->be_v_offset;
+				be->be_length += new->be_length;
+				be = ext_try_to_merge_left(root, be);
+				kfree(new);
+				return;
+			}
+			p = &(*p)->rb_left;
+		} else if (new->be_f_offset >= ext_f_end(be)) {
+			if (merge_ok && ext_can_merge(be, new)) {
+				be->be_length += new->be_length;
+				be = ext_try_to_merge_right(root, be);
+				kfree(new);
+				return;
+			}
+			p = &(*p)->rb_right;
+		} else {
+			BUG();
+		}
+	}
+
+	rb_link_node(&new->be_node, parent, p);
+	rb_insert_color(&new->be_node, root);
+}
+
+static int
+__ext_tree_remove(struct rb_root *root, sector_t start, sector_t end)
+{
+	struct pnfs_block_extent *be;
+	sector_t len1 = 0, len2 = 0;
+	sector_t orig_f_offset;
+	sector_t orig_v_offset;
+	sector_t orig_len;
+
+	be = __ext_tree_search(root, start);
+	if (!be)
+		return 0;
+	if (be->be_f_offset >= end)
+		return 0;
+
+	orig_f_offset = be->be_f_offset;
+	orig_v_offset = be->be_v_offset;
+	orig_len = be->be_length;
+
+	if (start > be->be_f_offset)
+		len1 = start - be->be_f_offset;
+	if (ext_f_end(be) > end)
+		len2 = ext_f_end(be) - end;
+
+	if (len2 > 0) {
+		if (len1 > 0) {
+			struct pnfs_block_extent *new;
+
+			new = kzalloc(sizeof(*new), GFP_ATOMIC);
+			if (!new)
+				return -ENOMEM;
+
+			be->be_length = len1;
+
+			new->be_f_offset = end;
+			if (be->be_state != PNFS_BLOCK_NONE_DATA) {
+				new->be_v_offset =
+					orig_v_offset + orig_len - len2;
+			}
+			new->be_length = len2;
+			new->be_state = be->be_state;
+			new->be_tag = be->be_tag;
+			new->be_mdev = be->be_mdev;
+			memcpy(&new->be_devid, &be->be_devid,
+				sizeof(struct nfs4_deviceid));
+
+			__ext_tree_insert(root, new, true);
+		} else {
+			be->be_f_offset = end;
+			if (be->be_state != PNFS_BLOCK_NONE_DATA) {
+				be->be_v_offset =
+					orig_v_offset + orig_len - len2;
+			}
+			be->be_length = len2;
+		}
+	} else {
+		if (len1 > 0) {
+			be->be_length = len1;
+			be = ext_tree_next(be);
+		}
+
+		while (be && ext_f_end(be) <= end) {
+			struct pnfs_block_extent *next = ext_tree_next(be);
+
+			rb_erase(&be->be_node, root);
+			kfree(be);
+			be = next;
+		}
+
+		if (be && be->be_f_offset < end) {
+			len1 = ext_f_end(be) - end;
+			be->be_f_offset = end;
+			if (be->be_state != PNFS_BLOCK_NONE_DATA)
+				be->be_v_offset += be->be_length - len1;
+			be->be_length = len1;
+		}
+	}
+
+	return 0;
+}
+
+int
+ext_tree_insert(struct pnfs_block_layout *bl, struct pnfs_block_extent *new)
+{
+	struct pnfs_block_extent *be;
+	struct rb_root *root;
+	int err = 0;
+
+	switch (new->be_state) {
+	case PNFS_BLOCK_READWRITE_DATA:
+	case PNFS_BLOCK_INVALID_DATA:
+		root = &bl->bl_ext_rw;
+		break;
+	case PNFS_BLOCK_READ_DATA:
+	case PNFS_BLOCK_NONE_DATA:
+		root = &bl->bl_ext_ro;
+		break;
+	default:
+		dprintk("invalid extent type\n");
+		return -EINVAL;
+	}
+
+	spin_lock(&bl->bl_ext_lock);
+retry:
+	be = __ext_tree_search(root, new->be_f_offset);
+	if (!be || be->be_f_offset >= ext_f_end(new)) {
+		__ext_tree_insert(root, new, true);
+	} else if (new->be_f_offset >= be->be_f_offset) {
+		if (ext_f_end(new) <= ext_f_end(be)) {
+			kfree(new);
+		} else {
+			sector_t new_len = ext_f_end(new) - ext_f_end(be);
+			sector_t diff = new->be_length - new_len;
+
+			new->be_f_offset += diff;
+			new->be_v_offset += diff;
+			new->be_length = new_len;
+			goto retry;
+		}
+	} else if (ext_f_end(new) <= ext_f_end(be)) {
+		new->be_length = be->be_f_offset - new->be_f_offset;
+		__ext_tree_insert(root, new, true);
+	} else {
+		struct pnfs_block_extent *split;
+		sector_t new_len = ext_f_end(new) - ext_f_end(be);
+		sector_t diff = new->be_length - new_len;
+
+		split = kmemdup(new, sizeof(*new), GFP_ATOMIC);
+		if (!split) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		split->be_length = be->be_f_offset - split->be_f_offset;
+		__ext_tree_insert(root, split, true);
+
+		new->be_f_offset += diff;
+		new->be_v_offset += diff;
+		new->be_length = new_len;
+		goto retry;
+	}
+out:
+	spin_unlock(&bl->bl_ext_lock);
+	return err;
+}
+
+static bool
+__ext_tree_lookup(struct rb_root *root, sector_t isect,
+		struct pnfs_block_extent *ret)
+{
+	struct rb_node *node;
+	struct pnfs_block_extent *be;
+
+	node = root->rb_node;
+	while (node) {
+		be = ext_node(node);
+		if (isect < be->be_f_offset)
+			node = node->rb_left;
+		else if (isect >= ext_f_end(be))
+			node = node->rb_right;
+		else {
+			*ret = *be;
+			return true;
+		}
+	}
+
+	return false;
+}
+
+bool
+ext_tree_lookup(struct pnfs_block_layout *bl, sector_t isect,
+	    struct pnfs_block_extent *ret, bool rw)
+{
+	bool found = false;
+
+	spin_lock(&bl->bl_ext_lock);
+	if (!rw)
+		found = __ext_tree_lookup(&bl->bl_ext_ro, isect, ret);
+	if (!found)
+		found = __ext_tree_lookup(&bl->bl_ext_rw, isect, ret);
+	spin_unlock(&bl->bl_ext_lock);
+
+	return found;
+}
+
+int ext_tree_remove(struct pnfs_block_layout *bl, bool rw,
+		sector_t start, sector_t end)
+{
+	int err, err2;
+
+	spin_lock(&bl->bl_ext_lock);
+	err = __ext_tree_remove(&bl->bl_ext_ro, start, end);
+	if (rw) {
+		err2 = __ext_tree_remove(&bl->bl_ext_rw, start, end);
+		if (!err)
+			err = err2;
+	}
+	spin_unlock(&bl->bl_ext_lock);
+
+	return err;
+}
+
+static int
+ext_tree