linux/fs/nfs/blocklayout/extent_tree.c

606 lines
14 KiB
C
Raw Normal View History

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>
2014-09-10 23:23:34 +08:00
/*
* Copyright (c) 2014 Christoph Hellwig.
*/
#include <linux/vmalloc.h>
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>
2014-09-10 23:23:34 +08:00
#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_device != be2->be_device)
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>
2014-09-10 23:23:34 +08:00
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);
nfs4_put_deviceid_node(be->be_device);
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>
2014-09-10 23:23:34 +08:00
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);
nfs4_put_deviceid_node(right->be_device);
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>
2014-09-10 23:23:34 +08:00
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);
goto free_new;
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>
2014-09-10 23:23:34 +08:00
}
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);
goto free_new;
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>
2014-09-10 23:23:34 +08:00
}
p = &(*p)->rb_right;
} else {
BUG();
}
}
rb_link_node(&new->be_node, parent, p);
rb_insert_color(&new->be_node, root);
return;
free_new:
nfs4_put_deviceid_node(new->be_device);
kfree(new);
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>
2014-09-10 23:23:34 +08:00
}
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_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_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_device = nfs4_get_deviceid(be->be_device);
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>
2014-09-10 23:23:34 +08:00
__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);
nfs4_put_deviceid_node(be->be_device);
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>
2014-09-10 23:23:34 +08:00
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)) {
nfs4_put_deviceid_node(new->be_device);
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>
2014-09-10 23:23:34 +08:00
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;
split->be_device = nfs4_get_deviceid(new->be_device);
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>
2014-09-10 23:23:34 +08:00
__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;
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_device = nfs4_get_deviceid(be->be_device);
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>
2014-09-10 23:23:34 +08:00
__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;
}
static size_t ext_tree_layoutupdate_size(size_t count)
{
return sizeof(__be32) /* number of entries */ +
PNFS_BLOCK_EXTENT_SIZE * count;
}
static void ext_tree_free_commitdata(struct nfs4_layoutcommit_args *arg,
size_t buffer_size)
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>
2014-09-10 23:23:34 +08:00
{
if (arg->layoutupdate_pages != &arg->layoutupdate_page) {
int nr_pages = DIV_ROUND_UP(buffer_size, PAGE_SIZE), i;
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>
2014-09-10 23:23:34 +08:00
for (i = 0; i < nr_pages; i++)
put_page(arg->layoutupdate_pages[i]);
kfree(arg->layoutupdate_pages);
} else {
put_page(arg->layoutupdate_page);
}
}
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>
2014-09-10 23:23:34 +08:00
static int ext_tree_encode_commit(struct pnfs_block_layout *bl, __be32 *p,
size_t buffer_size, size_t *count)
{
struct pnfs_block_extent *be;
int ret = 0;
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>
2014-09-10 23:23:34 +08:00
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;
(*count)++;
if (ext_tree_layoutupdate_size(*count) > buffer_size) {
/* keep counting.. */
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>
2014-09-10 23:23:34 +08:00
ret = -ENOSPC;
continue;
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>
2014-09-10 23:23:34 +08:00
}
p = xdr_encode_opaque_fixed(p, be->be_device->deviceid.data,
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>
2014-09-10 23:23:34 +08:00
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;
}
spin_unlock(&bl->bl_ext_lock);
return ret;
}
int
ext_tree_prepare_commit(struct nfs4_layoutcommit_args *arg)
{
struct pnfs_block_layout *bl = BLK_LO2EXT(NFS_I(arg->inode)->layout);
size_t count = 0, buffer_size = PAGE_SIZE;
__be32 *start_p;
int ret;
dprintk("%s enter\n", __func__);
arg->layoutupdate_page = alloc_page(GFP_NOFS);
if (!arg->layoutupdate_page)
return -ENOMEM;
start_p = page_address(arg->layoutupdate_page);
arg->layoutupdate_pages = &arg->layoutupdate_page;
retry:
ret = ext_tree_encode_commit(bl, start_p + 1, buffer_size, &count);
if (unlikely(ret)) {
ext_tree_free_commitdata(arg, buffer_size);
buffer_size = ext_tree_layoutupdate_size(count);
count = 0;
arg->layoutupdate_pages =
kcalloc(DIV_ROUND_UP(buffer_size, PAGE_SIZE),
sizeof(struct page *), GFP_NOFS);
if (!arg->layoutupdate_pages)
return -ENOMEM;
start_p = __vmalloc(buffer_size, GFP_NOFS, PAGE_KERNEL);
if (!start_p) {
kfree(arg->layoutupdate_pages);
return -ENOMEM;
}
goto retry;
}
*start_p = cpu_to_be32(count);
arg->layoutupdate_len = ext_tree_layoutupdate_size(count);
if (unlikely(arg->layoutupdate_pages != &arg->layoutupdate_page)) {
void *p = start_p, *end = p + arg->layoutupdate_len;
int i = 0;
for ( ; p < end; p += PAGE_SIZE)
arg->layoutupdate_pages[i++] = vmalloc_to_page(p);
}
dprintk("%s found %zu ranges\n", __func__, count);
return 0;
}
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>
2014-09-10 23:23:34 +08:00
void
ext_tree_mark_committed(struct nfs4_layoutcommit_args *arg, int status)
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>
2014-09-10 23:23:34 +08:00
{
struct pnfs_block_layout *bl = BLK_LO2EXT(NFS_I(arg->inode)->layout);
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>
2014-09-10 23:23:34 +08:00
struct rb_root *root = &bl->bl_ext_rw;
struct pnfs_block_extent *be;
dprintk("%s status %d\n", __func__, status);
ext_tree_free_commitdata(arg, arg->layoutupdate_len);
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>
2014-09-10 23:23:34 +08:00
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);
}