Linux Audio

Check our new training course

Embedded Linux training

Mar 31-Apr 8, 2025
Register
Loading...
v3.15
 
 
  1#include <linux/err.h>
  2#include <linux/slab.h>
  3#include <linux/spinlock.h>
  4#include <linux/hardirq.h>
  5#include "ctree.h"
  6#include "extent_map.h"
 
  7
  8
  9static struct kmem_cache *extent_map_cache;
 10
 11int __init extent_map_init(void)
 12{
 13	extent_map_cache = kmem_cache_create("btrfs_extent_map",
 14			sizeof(struct extent_map), 0,
 15			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
 16	if (!extent_map_cache)
 17		return -ENOMEM;
 18	return 0;
 19}
 20
 21void extent_map_exit(void)
 22{
 23	if (extent_map_cache)
 24		kmem_cache_destroy(extent_map_cache);
 25}
 26
 27/**
 28 * extent_map_tree_init - initialize extent map tree
 29 * @tree:		tree to initialize
 30 *
 31 * Initialize the extent tree @tree.  Should be called for each new inode
 32 * or other user of the extent_map interface.
 33 */
 34void extent_map_tree_init(struct extent_map_tree *tree)
 35{
 36	tree->map = RB_ROOT;
 37	INIT_LIST_HEAD(&tree->modified_extents);
 38	rwlock_init(&tree->lock);
 39}
 40
 41/**
 42 * alloc_extent_map - allocate new extent map structure
 43 *
 44 * Allocate a new extent_map structure.  The new structure is
 45 * returned with a reference count of one and needs to be
 46 * freed using free_extent_map()
 47 */
 48struct extent_map *alloc_extent_map(void)
 49{
 50	struct extent_map *em;
 51	em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
 52	if (!em)
 53		return NULL;
 54	RB_CLEAR_NODE(&em->rb_node);
 55	em->flags = 0;
 56	em->compress_type = BTRFS_COMPRESS_NONE;
 57	em->generation = 0;
 58	atomic_set(&em->refs, 1);
 59	INIT_LIST_HEAD(&em->list);
 60	return em;
 61}
 62
 63/**
 64 * free_extent_map - drop reference count of an extent_map
 65 * @em:		extent map beeing releasead
 66 *
 67 * Drops the reference out on @em by one and free the structure
 68 * if the reference count hits zero.
 69 */
 70void free_extent_map(struct extent_map *em)
 71{
 72	if (!em)
 73		return;
 74	WARN_ON(atomic_read(&em->refs) == 0);
 75	if (atomic_dec_and_test(&em->refs)) {
 76		WARN_ON(extent_map_in_tree(em));
 77		WARN_ON(!list_empty(&em->list));
 
 
 78		kmem_cache_free(extent_map_cache, em);
 79	}
 80}
 81
 82/* simple helper to do math around the end of an extent, handling wrap */
 83static u64 range_end(u64 start, u64 len)
 84{
 85	if (start + len < start)
 86		return (u64)-1;
 87	return start + len;
 88}
 89
 90static int tree_insert(struct rb_root *root, struct extent_map *em)
 91{
 92	struct rb_node **p = &root->rb_node;
 93	struct rb_node *parent = NULL;
 94	struct extent_map *entry = NULL;
 95	struct rb_node *orig_parent = NULL;
 96	u64 end = range_end(em->start, em->len);
 97
 98	while (*p) {
 99		parent = *p;
100		entry = rb_entry(parent, struct extent_map, rb_node);
101
102		if (em->start < entry->start)
103			p = &(*p)->rb_left;
104		else if (em->start >= extent_map_end(entry))
105			p = &(*p)->rb_right;
106		else
107			return -EEXIST;
108	}
109
110	orig_parent = parent;
111	while (parent && em->start >= extent_map_end(entry)) {
112		parent = rb_next(parent);
113		entry = rb_entry(parent, struct extent_map, rb_node);
114	}
115	if (parent)
116		if (end > entry->start && em->start < extent_map_end(entry))
117			return -EEXIST;
118
119	parent = orig_parent;
120	entry = rb_entry(parent, struct extent_map, rb_node);
121	while (parent && em->start < entry->start) {
122		parent = rb_prev(parent);
123		entry = rb_entry(parent, struct extent_map, rb_node);
124	}
125	if (parent)
126		if (end > entry->start && em->start < extent_map_end(entry))
127			return -EEXIST;
128
129	rb_link_node(&em->rb_node, orig_parent, p);
130	rb_insert_color(&em->rb_node, root);
131	return 0;
132}
133
134/*
135 * search through the tree for an extent_map with a given offset.  If
136 * it can't be found, try to find some neighboring extents
137 */
138static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
139				     struct rb_node **prev_ret,
140				     struct rb_node **next_ret)
141{
142	struct rb_node *n = root->rb_node;
143	struct rb_node *prev = NULL;
144	struct rb_node *orig_prev = NULL;
145	struct extent_map *entry;
146	struct extent_map *prev_entry = NULL;
147
148	while (n) {
149		entry = rb_entry(n, struct extent_map, rb_node);
150		prev = n;
151		prev_entry = entry;
152
153		if (offset < entry->start)
154			n = n->rb_left;
155		else if (offset >= extent_map_end(entry))
156			n = n->rb_right;
157		else
158			return n;
159	}
160
161	if (prev_ret) {
162		orig_prev = prev;
163		while (prev && offset >= extent_map_end(prev_entry)) {
164			prev = rb_next(prev);
165			prev_entry = rb_entry(prev, struct extent_map, rb_node);
166		}
167		*prev_ret = prev;
168		prev = orig_prev;
169	}
170
171	if (next_ret) {
172		prev_entry = rb_entry(prev, struct extent_map, rb_node);
173		while (prev && offset < prev_entry->start) {
174			prev = rb_prev(prev);
175			prev_entry = rb_entry(prev, struct extent_map, rb_node);
176		}
177		*next_ret = prev;
178	}
179	return NULL;
180}
181
182/* check to see if two extent_map structs are adjacent and safe to merge */
183static int mergable_maps(struct extent_map *prev, struct extent_map *next)
184{
185	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
186		return 0;
187
188	/*
189	 * don't merge compressed extents, we need to know their
190	 * actual size
191	 */
192	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
193		return 0;
194
195	if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
196	    test_bit(EXTENT_FLAG_LOGGING, &next->flags))
197		return 0;
198
199	/*
200	 * We don't want to merge stuff that hasn't been written to the log yet
201	 * since it may not reflect exactly what is on disk, and that would be
202	 * bad.
203	 */
204	if (!list_empty(&prev->list) || !list_empty(&next->list))
205		return 0;
206
207	if (extent_map_end(prev) == next->start &&
208	    prev->flags == next->flags &&
209	    prev->bdev == next->bdev &&
210	    ((next->block_start == EXTENT_MAP_HOLE &&
211	      prev->block_start == EXTENT_MAP_HOLE) ||
212	     (next->block_start == EXTENT_MAP_INLINE &&
213	      prev->block_start == EXTENT_MAP_INLINE) ||
214	     (next->block_start == EXTENT_MAP_DELALLOC &&
215	      prev->block_start == EXTENT_MAP_DELALLOC) ||
216	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
217	      next->block_start == extent_map_block_end(prev)))) {
218		return 1;
219	}
220	return 0;
221}
222
223static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
224{
225	struct extent_map *merge = NULL;
226	struct rb_node *rb;
227
228	if (em->start != 0) {
229		rb = rb_prev(&em->rb_node);
230		if (rb)
231			merge = rb_entry(rb, struct extent_map, rb_node);
232		if (rb && mergable_maps(merge, em)) {
233			em->start = merge->start;
234			em->orig_start = merge->orig_start;
235			em->len += merge->len;
236			em->block_len += merge->block_len;
237			em->block_start = merge->block_start;
238			em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
239			em->mod_start = merge->mod_start;
240			em->generation = max(em->generation, merge->generation);
241
242			rb_erase(&merge->rb_node, &tree->map);
243			RB_CLEAR_NODE(&merge->rb_node);
244			free_extent_map(merge);
245		}
246	}
247
248	rb = rb_next(&em->rb_node);
249	if (rb)
250		merge = rb_entry(rb, struct extent_map, rb_node);
251	if (rb && mergable_maps(em, merge)) {
252		em->len += merge->len;
253		em->block_len += merge->block_len;
254		rb_erase(&merge->rb_node, &tree->map);
255		RB_CLEAR_NODE(&merge->rb_node);
256		em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
257		em->generation = max(em->generation, merge->generation);
258		free_extent_map(merge);
259	}
260}
261
262/**
263 * unpin_extent_cache - unpin an extent from the cache
264 * @tree:	tree to unpin the extent in
265 * @start:	logical offset in the file
266 * @len:	length of the extent
267 * @gen:	generation that this extent has been modified in
268 *
269 * Called after an extent has been written to disk properly.  Set the generation
270 * to the generation that actually added the file item to the inode so we know
271 * we need to sync this extent when we call fsync().
272 */
273int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
274		       u64 gen)
275{
276	int ret = 0;
277	struct extent_map *em;
278	bool prealloc = false;
279
280	write_lock(&tree->lock);
281	em = lookup_extent_mapping(tree, start, len);
282
283	WARN_ON(!em || em->start != start);
284
285	if (!em)
286		goto out;
287
288	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
289		list_move(&em->list, &tree->modified_extents);
290	em->generation = gen;
291	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
292	em->mod_start = em->start;
293	em->mod_len = em->len;
294
295	if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
296		prealloc = true;
297		clear_bit(EXTENT_FLAG_FILLING, &em->flags);
298	}
299
300	try_merge_map(tree, em);
301
302	if (prealloc) {
303		em->mod_start = em->start;
304		em->mod_len = em->len;
305	}
306
307	free_extent_map(em);
308out:
309	write_unlock(&tree->lock);
310	return ret;
311
312}
313
314void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
315{
316	clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
317	if (extent_map_in_tree(em))
318		try_merge_map(tree, em);
319}
320
321static inline void setup_extent_mapping(struct extent_map_tree *tree,
322					struct extent_map *em,
323					int modified)
324{
325	atomic_inc(&em->refs);
326	em->mod_start = em->start;
327	em->mod_len = em->len;
328
329	if (modified)
330		list_move(&em->list, &tree->modified_extents);
331	else
332		try_merge_map(tree, em);
333}
334
335/**
336 * add_extent_mapping - add new extent map to the extent tree
337 * @tree:	tree to insert new map in
338 * @em:		map to insert
339 *
340 * Insert @em into @tree or perform a simple forward/backward merge with
341 * existing mappings.  The extent_map struct passed in will be inserted
342 * into the tree directly, with an additional reference taken, or a
343 * reference dropped if the merge attempt was successful.
344 */
345int add_extent_mapping(struct extent_map_tree *tree,
346		       struct extent_map *em, int modified)
347{
348	int ret = 0;
349
350	ret = tree_insert(&tree->map, em);
351	if (ret)
352		goto out;
353
354	setup_extent_mapping(tree, em, modified);
355out:
356	return ret;
357}
358
359static struct extent_map *
360__lookup_extent_mapping(struct extent_map_tree *tree,
361			u64 start, u64 len, int strict)
362{
363	struct extent_map *em;
364	struct rb_node *rb_node;
365	struct rb_node *prev = NULL;
366	struct rb_node *next = NULL;
367	u64 end = range_end(start, len);
368
369	rb_node = __tree_search(&tree->map, start, &prev, &next);
370	if (!rb_node) {
371		if (prev)
372			rb_node = prev;
373		else if (next)
374			rb_node = next;
375		else
376			return NULL;
377	}
378
379	em = rb_entry(rb_node, struct extent_map, rb_node);
380
381	if (strict && !(end > em->start && start < extent_map_end(em)))
382		return NULL;
383
384	atomic_inc(&em->refs);
385	return em;
386}
387
388/**
389 * lookup_extent_mapping - lookup extent_map
390 * @tree:	tree to lookup in
391 * @start:	byte offset to start the search
392 * @len:	length of the lookup range
393 *
394 * Find and return the first extent_map struct in @tree that intersects the
395 * [start, len] range.  There may be additional objects in the tree that
396 * intersect, so check the object returned carefully to make sure that no
397 * additional lookups are needed.
398 */
399struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
400					 u64 start, u64 len)
401{
402	return __lookup_extent_mapping(tree, start, len, 1);
403}
404
405/**
406 * search_extent_mapping - find a nearby extent map
407 * @tree:	tree to lookup in
408 * @start:	byte offset to start the search
409 * @len:	length of the lookup range
410 *
411 * Find and return the first extent_map struct in @tree that intersects the
412 * [start, len] range.
413 *
414 * If one can't be found, any nearby extent may be returned
415 */
416struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
417					 u64 start, u64 len)
418{
419	return __lookup_extent_mapping(tree, start, len, 0);
420}
421
422/**
423 * remove_extent_mapping - removes an extent_map from the extent tree
424 * @tree:	extent tree to remove from
425 * @em:		extent map beeing removed
426 *
427 * Removes @em from @tree.  No reference counts are dropped, and no checks
428 * are done to see if the range is in use
429 */
430int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
431{
432	int ret = 0;
433
434	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
435	rb_erase(&em->rb_node, &tree->map);
436	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
437		list_del_init(&em->list);
438	RB_CLEAR_NODE(&em->rb_node);
439	return ret;
440}
441
442void replace_extent_mapping(struct extent_map_tree *tree,
443			    struct extent_map *cur,
444			    struct extent_map *new,
445			    int modified)
446{
447	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
448	ASSERT(extent_map_in_tree(cur));
449	if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
450		list_del_init(&cur->list);
451	rb_replace_node(&cur->rb_node, &new->rb_node, &tree->map);
452	RB_CLEAR_NODE(&cur->rb_node);
453
454	setup_extent_mapping(tree, new, modified);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
455}
v4.17
  1// SPDX-License-Identifier: GPL-2.0
  2
  3#include <linux/err.h>
  4#include <linux/slab.h>
  5#include <linux/spinlock.h>
 
  6#include "ctree.h"
  7#include "extent_map.h"
  8#include "compression.h"
  9
 10
 11static struct kmem_cache *extent_map_cache;
 12
 13int __init extent_map_init(void)
 14{
 15	extent_map_cache = kmem_cache_create("btrfs_extent_map",
 16			sizeof(struct extent_map), 0,
 17			SLAB_MEM_SPREAD, NULL);
 18	if (!extent_map_cache)
 19		return -ENOMEM;
 20	return 0;
 21}
 22
 23void __cold extent_map_exit(void)
 24{
 25	kmem_cache_destroy(extent_map_cache);
 
 26}
 27
 28/**
 29 * extent_map_tree_init - initialize extent map tree
 30 * @tree:		tree to initialize
 31 *
 32 * Initialize the extent tree @tree.  Should be called for each new inode
 33 * or other user of the extent_map interface.
 34 */
 35void extent_map_tree_init(struct extent_map_tree *tree)
 36{
 37	tree->map = RB_ROOT;
 38	INIT_LIST_HEAD(&tree->modified_extents);
 39	rwlock_init(&tree->lock);
 40}
 41
 42/**
 43 * alloc_extent_map - allocate new extent map structure
 44 *
 45 * Allocate a new extent_map structure.  The new structure is
 46 * returned with a reference count of one and needs to be
 47 * freed using free_extent_map()
 48 */
 49struct extent_map *alloc_extent_map(void)
 50{
 51	struct extent_map *em;
 52	em = kmem_cache_zalloc(extent_map_cache, GFP_NOFS);
 53	if (!em)
 54		return NULL;
 55	RB_CLEAR_NODE(&em->rb_node);
 56	em->flags = 0;
 57	em->compress_type = BTRFS_COMPRESS_NONE;
 58	em->generation = 0;
 59	refcount_set(&em->refs, 1);
 60	INIT_LIST_HEAD(&em->list);
 61	return em;
 62}
 63
 64/**
 65 * free_extent_map - drop reference count of an extent_map
 66 * @em:		extent map being released
 67 *
 68 * Drops the reference out on @em by one and free the structure
 69 * if the reference count hits zero.
 70 */
 71void free_extent_map(struct extent_map *em)
 72{
 73	if (!em)
 74		return;
 75	WARN_ON(refcount_read(&em->refs) == 0);
 76	if (refcount_dec_and_test(&em->refs)) {
 77		WARN_ON(extent_map_in_tree(em));
 78		WARN_ON(!list_empty(&em->list));
 79		if (test_bit(EXTENT_FLAG_FS_MAPPING, &em->flags))
 80			kfree(em->map_lookup);
 81		kmem_cache_free(extent_map_cache, em);
 82	}
 83}
 84
 85/* simple helper to do math around the end of an extent, handling wrap */
 86static u64 range_end(u64 start, u64 len)
 87{
 88	if (start + len < start)
 89		return (u64)-1;
 90	return start + len;
 91}
 92
 93static int tree_insert(struct rb_root *root, struct extent_map *em)
 94{
 95	struct rb_node **p = &root->rb_node;
 96	struct rb_node *parent = NULL;
 97	struct extent_map *entry = NULL;
 98	struct rb_node *orig_parent = NULL;
 99	u64 end = range_end(em->start, em->len);
100
101	while (*p) {
102		parent = *p;
103		entry = rb_entry(parent, struct extent_map, rb_node);
104
105		if (em->start < entry->start)
106			p = &(*p)->rb_left;
107		else if (em->start >= extent_map_end(entry))
108			p = &(*p)->rb_right;
109		else
110			return -EEXIST;
111	}
112
113	orig_parent = parent;
114	while (parent && em->start >= extent_map_end(entry)) {
115		parent = rb_next(parent);
116		entry = rb_entry(parent, struct extent_map, rb_node);
117	}
118	if (parent)
119		if (end > entry->start && em->start < extent_map_end(entry))
120			return -EEXIST;
121
122	parent = orig_parent;
123	entry = rb_entry(parent, struct extent_map, rb_node);
124	while (parent && em->start < entry->start) {
125		parent = rb_prev(parent);
126		entry = rb_entry(parent, struct extent_map, rb_node);
127	}
128	if (parent)
129		if (end > entry->start && em->start < extent_map_end(entry))
130			return -EEXIST;
131
132	rb_link_node(&em->rb_node, orig_parent, p);
133	rb_insert_color(&em->rb_node, root);
134	return 0;
135}
136
137/*
138 * search through the tree for an extent_map with a given offset.  If
139 * it can't be found, try to find some neighboring extents
140 */
141static struct rb_node *__tree_search(struct rb_root *root, u64 offset,
142				     struct rb_node **prev_ret,
143				     struct rb_node **next_ret)
144{
145	struct rb_node *n = root->rb_node;
146	struct rb_node *prev = NULL;
147	struct rb_node *orig_prev = NULL;
148	struct extent_map *entry;
149	struct extent_map *prev_entry = NULL;
150
151	while (n) {
152		entry = rb_entry(n, struct extent_map, rb_node);
153		prev = n;
154		prev_entry = entry;
155
156		if (offset < entry->start)
157			n = n->rb_left;
158		else if (offset >= extent_map_end(entry))
159			n = n->rb_right;
160		else
161			return n;
162	}
163
164	if (prev_ret) {
165		orig_prev = prev;
166		while (prev && offset >= extent_map_end(prev_entry)) {
167			prev = rb_next(prev);
168			prev_entry = rb_entry(prev, struct extent_map, rb_node);
169		}
170		*prev_ret = prev;
171		prev = orig_prev;
172	}
173
174	if (next_ret) {
175		prev_entry = rb_entry(prev, struct extent_map, rb_node);
176		while (prev && offset < prev_entry->start) {
177			prev = rb_prev(prev);
178			prev_entry = rb_entry(prev, struct extent_map, rb_node);
179		}
180		*next_ret = prev;
181	}
182	return NULL;
183}
184
185/* check to see if two extent_map structs are adjacent and safe to merge */
186static int mergable_maps(struct extent_map *prev, struct extent_map *next)
187{
188	if (test_bit(EXTENT_FLAG_PINNED, &prev->flags))
189		return 0;
190
191	/*
192	 * don't merge compressed extents, we need to know their
193	 * actual size
194	 */
195	if (test_bit(EXTENT_FLAG_COMPRESSED, &prev->flags))
196		return 0;
197
198	if (test_bit(EXTENT_FLAG_LOGGING, &prev->flags) ||
199	    test_bit(EXTENT_FLAG_LOGGING, &next->flags))
200		return 0;
201
202	/*
203	 * We don't want to merge stuff that hasn't been written to the log yet
204	 * since it may not reflect exactly what is on disk, and that would be
205	 * bad.
206	 */
207	if (!list_empty(&prev->list) || !list_empty(&next->list))
208		return 0;
209
210	if (extent_map_end(prev) == next->start &&
211	    prev->flags == next->flags &&
212	    prev->bdev == next->bdev &&
213	    ((next->block_start == EXTENT_MAP_HOLE &&
214	      prev->block_start == EXTENT_MAP_HOLE) ||
215	     (next->block_start == EXTENT_MAP_INLINE &&
216	      prev->block_start == EXTENT_MAP_INLINE) ||
217	     (next->block_start == EXTENT_MAP_DELALLOC &&
218	      prev->block_start == EXTENT_MAP_DELALLOC) ||
219	     (next->block_start < EXTENT_MAP_LAST_BYTE - 1 &&
220	      next->block_start == extent_map_block_end(prev)))) {
221		return 1;
222	}
223	return 0;
224}
225
226static void try_merge_map(struct extent_map_tree *tree, struct extent_map *em)
227{
228	struct extent_map *merge = NULL;
229	struct rb_node *rb;
230
231	if (em->start != 0) {
232		rb = rb_prev(&em->rb_node);
233		if (rb)
234			merge = rb_entry(rb, struct extent_map, rb_node);
235		if (rb && mergable_maps(merge, em)) {
236			em->start = merge->start;
237			em->orig_start = merge->orig_start;
238			em->len += merge->len;
239			em->block_len += merge->block_len;
240			em->block_start = merge->block_start;
241			em->mod_len = (em->mod_len + em->mod_start) - merge->mod_start;
242			em->mod_start = merge->mod_start;
243			em->generation = max(em->generation, merge->generation);
244
245			rb_erase(&merge->rb_node, &tree->map);
246			RB_CLEAR_NODE(&merge->rb_node);
247			free_extent_map(merge);
248		}
249	}
250
251	rb = rb_next(&em->rb_node);
252	if (rb)
253		merge = rb_entry(rb, struct extent_map, rb_node);
254	if (rb && mergable_maps(em, merge)) {
255		em->len += merge->len;
256		em->block_len += merge->block_len;
257		rb_erase(&merge->rb_node, &tree->map);
258		RB_CLEAR_NODE(&merge->rb_node);
259		em->mod_len = (merge->mod_start + merge->mod_len) - em->mod_start;
260		em->generation = max(em->generation, merge->generation);
261		free_extent_map(merge);
262	}
263}
264
265/**
266 * unpin_extent_cache - unpin an extent from the cache
267 * @tree:	tree to unpin the extent in
268 * @start:	logical offset in the file
269 * @len:	length of the extent
270 * @gen:	generation that this extent has been modified in
271 *
272 * Called after an extent has been written to disk properly.  Set the generation
273 * to the generation that actually added the file item to the inode so we know
274 * we need to sync this extent when we call fsync().
275 */
276int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len,
277		       u64 gen)
278{
279	int ret = 0;
280	struct extent_map *em;
281	bool prealloc = false;
282
283	write_lock(&tree->lock);
284	em = lookup_extent_mapping(tree, start, len);
285
286	WARN_ON(!em || em->start != start);
287
288	if (!em)
289		goto out;
290
 
 
291	em->generation = gen;
292	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
293	em->mod_start = em->start;
294	em->mod_len = em->len;
295
296	if (test_bit(EXTENT_FLAG_FILLING, &em->flags)) {
297		prealloc = true;
298		clear_bit(EXTENT_FLAG_FILLING, &em->flags);
299	}
300
301	try_merge_map(tree, em);
302
303	if (prealloc) {
304		em->mod_start = em->start;
305		em->mod_len = em->len;
306	}
307
308	free_extent_map(em);
309out:
310	write_unlock(&tree->lock);
311	return ret;
312
313}
314
315void clear_em_logging(struct extent_map_tree *tree, struct extent_map *em)
316{
317	clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
318	if (extent_map_in_tree(em))
319		try_merge_map(tree, em);
320}
321
322static inline void setup_extent_mapping(struct extent_map_tree *tree,
323					struct extent_map *em,
324					int modified)
325{
326	refcount_inc(&em->refs);
327	em->mod_start = em->start;
328	em->mod_len = em->len;
329
330	if (modified)
331		list_move(&em->list, &tree->modified_extents);
332	else
333		try_merge_map(tree, em);
334}
335
336/**
337 * add_extent_mapping - add new extent map to the extent tree
338 * @tree:	tree to insert new map in
339 * @em:		map to insert
340 *
341 * Insert @em into @tree or perform a simple forward/backward merge with
342 * existing mappings.  The extent_map struct passed in will be inserted
343 * into the tree directly, with an additional reference taken, or a
344 * reference dropped if the merge attempt was successful.
345 */
346int add_extent_mapping(struct extent_map_tree *tree,
347		       struct extent_map *em, int modified)
348{
349	int ret = 0;
350
351	ret = tree_insert(&tree->map, em);
352	if (ret)
353		goto out;
354
355	setup_extent_mapping(tree, em, modified);
356out:
357	return ret;
358}
359
360static struct extent_map *
361__lookup_extent_mapping(struct extent_map_tree *tree,
362			u64 start, u64 len, int strict)
363{
364	struct extent_map *em;
365	struct rb_node *rb_node;
366	struct rb_node *prev = NULL;
367	struct rb_node *next = NULL;
368	u64 end = range_end(start, len);
369
370	rb_node = __tree_search(&tree->map, start, &prev, &next);
371	if (!rb_node) {
372		if (prev)
373			rb_node = prev;
374		else if (next)
375			rb_node = next;
376		else
377			return NULL;
378	}
379
380	em = rb_entry(rb_node, struct extent_map, rb_node);
381
382	if (strict && !(end > em->start && start < extent_map_end(em)))
383		return NULL;
384
385	refcount_inc(&em->refs);
386	return em;
387}
388
389/**
390 * lookup_extent_mapping - lookup extent_map
391 * @tree:	tree to lookup in
392 * @start:	byte offset to start the search
393 * @len:	length of the lookup range
394 *
395 * Find and return the first extent_map struct in @tree that intersects the
396 * [start, len] range.  There may be additional objects in the tree that
397 * intersect, so check the object returned carefully to make sure that no
398 * additional lookups are needed.
399 */
400struct extent_map *lookup_extent_mapping(struct extent_map_tree *tree,
401					 u64 start, u64 len)
402{
403	return __lookup_extent_mapping(tree, start, len, 1);
404}
405
406/**
407 * search_extent_mapping - find a nearby extent map
408 * @tree:	tree to lookup in
409 * @start:	byte offset to start the search
410 * @len:	length of the lookup range
411 *
412 * Find and return the first extent_map struct in @tree that intersects the
413 * [start, len] range.
414 *
415 * If one can't be found, any nearby extent may be returned
416 */
417struct extent_map *search_extent_mapping(struct extent_map_tree *tree,
418					 u64 start, u64 len)
419{
420	return __lookup_extent_mapping(tree, start, len, 0);
421}
422
423/**
424 * remove_extent_mapping - removes an extent_map from the extent tree
425 * @tree:	extent tree to remove from
426 * @em:		extent map being removed
427 *
428 * Removes @em from @tree.  No reference counts are dropped, and no checks
429 * are done to see if the range is in use
430 */
431int remove_extent_mapping(struct extent_map_tree *tree, struct extent_map *em)
432{
433	int ret = 0;
434
435	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
436	rb_erase(&em->rb_node, &tree->map);
437	if (!test_bit(EXTENT_FLAG_LOGGING, &em->flags))
438		list_del_init(&em->list);
439	RB_CLEAR_NODE(&em->rb_node);
440	return ret;
441}
442
443void replace_extent_mapping(struct extent_map_tree *tree,
444			    struct extent_map *cur,
445			    struct extent_map *new,
446			    int modified)
447{
448	WARN_ON(test_bit(EXTENT_FLAG_PINNED, &cur->flags));
449	ASSERT(extent_map_in_tree(cur));
450	if (!test_bit(EXTENT_FLAG_LOGGING, &cur->flags))
451		list_del_init(&cur->list);
452	rb_replace_node(&cur->rb_node, &new->rb_node, &tree->map);
453	RB_CLEAR_NODE(&cur->rb_node);
454
455	setup_extent_mapping(tree, new, modified);
456}
457
458static struct extent_map *next_extent_map(struct extent_map *em)
459{
460	struct rb_node *next;
461
462	next = rb_next(&em->rb_node);
463	if (!next)
464		return NULL;
465	return container_of(next, struct extent_map, rb_node);
466}
467
468static struct extent_map *prev_extent_map(struct extent_map *em)
469{
470	struct rb_node *prev;
471
472	prev = rb_prev(&em->rb_node);
473	if (!prev)
474		return NULL;
475	return container_of(prev, struct extent_map, rb_node);
476}
477
478/* helper for btfs_get_extent.  Given an existing extent in the tree,
479 * the existing extent is the nearest extent to map_start,
480 * and an extent that you want to insert, deal with overlap and insert
481 * the best fitted new extent into the tree.
482 */
483static noinline int merge_extent_mapping(struct extent_map_tree *em_tree,
484					 struct extent_map *existing,
485					 struct extent_map *em,
486					 u64 map_start)
487{
488	struct extent_map *prev;
489	struct extent_map *next;
490	u64 start;
491	u64 end;
492	u64 start_diff;
493
494	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
495
496	if (existing->start > map_start) {
497		next = existing;
498		prev = prev_extent_map(next);
499	} else {
500		prev = existing;
501		next = next_extent_map(prev);
502	}
503
504	start = prev ? extent_map_end(prev) : em->start;
505	start = max_t(u64, start, em->start);
506	end = next ? next->start : extent_map_end(em);
507	end = min_t(u64, end, extent_map_end(em));
508	start_diff = start - em->start;
509	em->start = start;
510	em->len = end - start;
511	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
512	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
513		em->block_start += start_diff;
514		em->block_len = em->len;
515	}
516	return add_extent_mapping(em_tree, em, 0);
517}
518
519/**
520 * btrfs_add_extent_mapping - add extent mapping into em_tree
521 * @em_tree - the extent tree into which we want to insert the extent mapping
522 * @em_in   - extent we are inserting
523 * @start   - start of the logical range btrfs_get_extent() is requesting
524 * @len     - length of the logical range btrfs_get_extent() is requesting
525 *
526 * Note that @em_in's range may be different from [start, start+len),
527 * but they must be overlapped.
528 *
529 * Insert @em_in into @em_tree. In case there is an overlapping range, handle
530 * the -EEXIST by either:
531 * a) Returning the existing extent in @em_in if @start is within the
532 *    existing em.
533 * b) Merge the existing extent with @em_in passed in.
534 *
535 * Return 0 on success, otherwise -EEXIST.
536 *
537 */
538int btrfs_add_extent_mapping(struct extent_map_tree *em_tree,
539			     struct extent_map **em_in, u64 start, u64 len)
540{
541	int ret;
542	struct extent_map *em = *em_in;
543
544	ret = add_extent_mapping(em_tree, em, 0);
545	/* it is possible that someone inserted the extent into the tree
546	 * while we had the lock dropped.  It is also possible that
547	 * an overlapping map exists in the tree
548	 */
549	if (ret == -EEXIST) {
550		struct extent_map *existing;
551
552		ret = 0;
553
554		existing = search_extent_mapping(em_tree, start, len);
555
556		trace_btrfs_handle_em_exist(existing, em, start, len);
557
558		/*
559		 * existing will always be non-NULL, since there must be
560		 * extent causing the -EEXIST.
561		 */
562		if (start >= existing->start &&
563		    start < extent_map_end(existing)) {
564			free_extent_map(em);
565			*em_in = existing;
566			ret = 0;
567		} else {
568			u64 orig_start = em->start;
569			u64 orig_len = em->len;
570
571			/*
572			 * The existing extent map is the one nearest to
573			 * the [start, start + len) range which overlaps
574			 */
575			ret = merge_extent_mapping(em_tree, existing,
576						   em, start);
577			if (ret) {
578				free_extent_map(em);
579				*em_in = NULL;
580				WARN_ONCE(ret,
581"unexpected error %d: merge existing(start %llu len %llu) with em(start %llu len %llu)\n",
582					  ret, existing->start, existing->len,
583					  orig_start, orig_len);
584			}
585			free_extent_map(existing);
586		}
587	}
588
589	ASSERT(ret == 0 || ret == -EEXIST);
590	return ret;
591}