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>

1 files changed, 547 insertions, 0 deletions

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_split(struct rb_root *root, struct pnfs_block_extent *be,
+		sector_t split)
+{
+	struct pnfs_block_extent *new;
+	sector_t orig_len = be->be_length;
+
+	dprintk("%s: need split for 0x%lx:0x%lx at 0x%lx\n",
+		__func__, be->be_f_offset, ext_f_end(be), split);
+
+	new = kzalloc(sizeof(*new), GFP_ATOMIC);
+	if (!new)
+		return -ENOMEM;
+
+	be->be_length = split - be->be_f_offset;
+
+	new->be_f_offset = split;
+	if (be->be_state != PNFS_BLOCK_NONE_DATA)
+		new->be_v_offset = be->be_v_offset + be->be_length;
+	new->be_length = orig_len - be->be_length;
+	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));
+
+	dprintk("%s: got 0x%lx:0x%lx!\n",
+		__func__, be->be_f_offset, ext_f_end(be));
+	dprintk("%s: got 0x%lx:0x%lx!\n",
+		__func__, new->be_f_offset, ext_f_end(new));
+
+	__ext_tree_insert(root, new, false);
+	return 0;
+}
+
+int
+ext_tree_mark_written(struct pnfs_block_layout *bl, sector_t start,
+		sector_t len)
+{
+	struct rb_root *root = &bl->bl_ext_rw;
+	sector_t end = start + len;
+	struct pnfs_block_extent *be;
+	int err = 0;
+
+	spin_lock(&bl->bl_ext_lock);
+	/*
+	 * First remove all COW extents or holes from written to range.
+	 */
+	err = __ext_tree_remove(&bl->bl_ext_ro, start, end);
+	if (err)
+		goto out;
+
+	/*
+	 * Then mark all invalid extents in the range as written to.
+	 */
+	for (be = __ext_tree_search(root, start); be; be = ext_tree_next(be)) {
+		if (be->be_f_offset >= end)
+			break;
+
+		if (be->be_state != PNFS_BLOCK_INVALID_DATA || be->be_tag)
+			continue;
+
+		if (be->be_f_offset < start) {
+			struct pnfs_block_extent *left = ext_tree_prev(be);
+
+			if (left && ext_can_merge(left, be)) {
+				sector_t diff = start - be->be_f_offset;
+
+				left->be_length += diff;
+
+				be->be_f_offset += diff;
+				be->be_v_offset += diff;
+				be->be_length -= diff;
+			} else {
+				err = ext_tree_split(root, be, start);
+				if (err)
+					goto out;
+			}
+		}
+
+		if (ext_f_end(be) > end) {
+			struct pnfs_block_extent *right = ext_tree_next(be);
+
+			if (right && ext_can_merge(be, right)) {
+				sector_t diff = end - be->be_f_offset;
+
+				be->be_length -= diff;
+
+				right->be_f_offset -= diff;
+				right->be_v_offset -= diff;
+				right->be_length += diff;
+			} else {
+				err = ext_tree_split(root, be, end);
+				if (err)
+					goto out;
+			}
+		}
+
+		if (be->be_f_offset >= start && ext_f_end(be) <= end) {
+			be->be_tag = EXTENT_WRITTEN;
+			be = ext_try_to_merge_left(root, be);
+			be = ext_try_to_merge_right(root, be);
+		}
+	}
+out:
+	spin_unlock(&bl->bl_ext_lock);
+	return err;
+}
+
+int
+ext_tree_encode_commit(struct pnfs_block_layout *bl, struct xdr_stream *xdr)
+{
+	struct pnfs_block_extent *be;
+	unsigned int count = 0;
+	__be32 *p, *xdr_start;
+	int ret = 0;
+
+	dprintk("%s enter\n", __func__);
+
+	xdr_start = xdr_reserve_space(xdr, 8);
+	if (!xdr_start)
+		return -ENOSPC;
+
+	spin_lock(&bl->bl_ext_lock);
+	for (be = ext_tree_first(&bl->bl_ext_rw); be; be = ext_tree_next(be)) {
+		if (be->be_state != PNFS_BLOCK_INVALID_DATA ||
+		    be->be_tag != EXTENT_WRITTEN)
+			continue;
+
+		p = xdr_reserve_space(xdr, 7 * sizeof(__be32) +
+					NFS4_DEVICEID4_SIZE);
+		if (!p) {
+			printk("%s: out of space for extent list\n", __func__);
+			ret = -ENOSPC;
+			break;
+		}
+
+		p = xdr_encode_opaque_fixed(p, be->be_devid.data,
+				NFS4_DEVICEID4_SIZE);
+		p = xdr_encode_hyper(p, be->be_f_offset << SECTOR_SHIFT);
+		p = xdr_encode_hyper(p, be->be_length << SECTOR_SHIFT);
+		p = xdr_encode_hyper(p, 0LL);
+		*p++ = cpu_to_be32(PNFS_BLOCK_READWRITE_DATA);
+
+		be->be_tag = EXTENT_COMMITTING;
+		count++;
+	}
+	spin_unlock(&bl->bl_ext_lock);
+
+	xdr_start[0] = cpu_to_be32((xdr->p - xdr_start - 1) * 4);
+	xdr_start[1] = cpu_to_be32(count);
+
+	dprintk("%s found %i ranges\n", __func__, count);
+	return ret;
+}
+
+void
+ext_tree_mark_committed(struct pnfs_block_layout *bl, int status)
+{
+	struct rb_root *root = &bl->bl_ext_rw;
+	struct pnfs_block_extent *be;
+
+	dprintk("%s status %d\n", __func__, status);
+
+	spin_lock(&bl->bl_ext_lock);
+	for (be = ext_tree_first(root); be; be = ext_tree_next(be)) {
+		if (be->be_state != PNFS_BLOCK_INVALID_DATA ||
+		    be->be_tag != EXTENT_COMMITTING)
+			continue;
+
+		if (status) {
+			/*
+			 * Mark as written and try again.
+			 *
+			 * XXX: some real error handling here wouldn't hurt..
+			 */
+			be->be_tag = EXTENT_WRITTEN;
+		} else {
+			be->be_state = PNFS_BLOCK_READWRITE_DATA;
+			be->be_tag = 0;
+		}
+
+		be = ext_try_to_merge_left(root, be);
+		be = ext_try_to_merge_right(root, be);
+	}
+	spin_unlock(&bl->bl_ext_lock);
+}